nortti
/
sortix-mirror
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
sortix-mirror/kernel/process.cpp

1658 lines
44 KiB
C++
Raw Normal View History

Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
/*******************************************************************************
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Replace system call array types with pointers.
2015-01-22 23:29:04 +00:00
Copyright(C) Jonas 'Sortie' Termansen 2011, 2012, 2013, 2014, 2015.
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update copyright headers of old files to the current format.
2013-07-10 13:26:01 +00:00
This file is part of Sortix.
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update copyright headers of old files to the current format.
2013-07-10 13:26:01 +00:00
Sortix is free software: you can redistribute it and/or modify it under the
terms of the GNU General Public License as published by the Free Software
Foundation, either version 3 of the License, or (at your option) any later
version.
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update copyright headers of old files to the current format.
2013-07-10 13:26:01 +00:00
Sortix is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
details.
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update copyright headers of old files to the current format.
2013-07-10 13:26:01 +00:00
You should have received a copy of the GNU General Public License along with
Sortix. If not, see <http://www.gnu.org/licenses/>.
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update copyright headers of old files to the current format.
2013-07-10 13:26:01 +00:00
process.cpp
A named collection of threads.
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
*******************************************************************************/
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <sys/wait.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <assert.h>
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
#include <ctype.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <errno.h>
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
#include <limits.h>
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#include <msr.h>
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
#include <signal.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <stdint.h>
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
#include <stdlib.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <string.h>
#include <sortix/clock.h>
#include <sortix/fcntl.h>
#include <sortix/fork.h>
#include <sortix/mman.h>
Add prlimit(2).
2013-08-30 21:55:09 +00:00
#include <sortix/resource.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/signal.h>
#include <sortix/stat.h>
#include <sortix/unistd.h>
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#include <sortix/uthread.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/wait.h>
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
#include <sortix/kernel/addralloc.h>
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
#include <sortix/kernel/copy.h>
#include <sortix/kernel/descriptor.h>
#include <sortix/kernel/dtable.h>
Rewrite program loader.
2014-06-25 23:05:07 +00:00
#include <sortix/kernel/elf.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/ioctx.h>
Remove <sortix/kernel/platform.h>.
2013-10-27 00:42:10 +00:00
#include <sortix/kernel/kernel.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/kthread.h>
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
#include <sortix/kernel/memorymanagement.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/mtable.h>
#include <sortix/kernel/process.h>
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
#include <sortix/kernel/ptable.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/refcount.h>
Refactor scheduler API.
2013-01-09 09:47:22 +00:00
#include <sortix/kernel/scheduler.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/sortedlist.h>
#include <sortix/kernel/string.h>
Load program symbol tables.
2013-05-22 20:11:29 +00:00
#include <sortix/kernel/symbol.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/syscall.h>
Refactor kernel process.h and thread.h headers.
2013-05-12 22:41:30 +00:00
#include <sortix/kernel/thread.h>
Maintain process execution and system time.
2013-05-14 22:45:04 +00:00
#include <sortix/kernel/time.h>
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
#include <sortix/kernel/worker.h>
Refactor system call API.
2013-01-08 23:41:35 +00:00
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#if defined(__i386__) || defined(__x86_64__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
#include "x86-family/float.h"
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#include "x86-family/gdt.h"
#endif
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
namespace Sortix {
Ported kernel to new syscall API and started cleaning up the old one.
2011-10-26 22:05:20 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Process::Process()
{
string_table = NULL;
string_table_length = 0;
symbol_table = NULL;
symbol_table_length = 0;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
program_image_path = NULL;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
addrspace = 0;
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
pid = 0;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
nicelock = KTHREAD_MUTEX_INITIALIZER;
nice = 0;
idlock = KTHREAD_MUTEX_INITIALIZER;
uid = euid = 0;
gid = egid = 0;
umask = 0022;
ptrlock = KTHREAD_MUTEX_INITIALIZER;
// root set to null reference in the member constructor.
// cwd set to null reference in the member constructor.
// mtable set to null reference in the member constructor.
// dtable set to null reference in the member constructor.
// ptable set to null reference in the member constructor.
resource_limits_lock = KTHREAD_MUTEX_INITIALIZER;
for ( size_t i = 0; i < RLIMIT_NUM_DECLARED; i++ )
{
resource_limits[i].rlim_cur = RLIM_INFINITY;
resource_limits[i].rlim_max = RLIM_INFINITY;
}
signal_lock = KTHREAD_MUTEX_INITIALIZER;
memset(&signal_actions, 0, sizeof(signal_actions));
for ( int i = 0; i < SIG_MAX_NUM; i++ )
{
sigemptyset(&signal_actions[i].sa_mask);
signal_actions[i].sa_handler = SIG_DFL;
signal_actions[i].sa_cookie = NULL;
signal_actions[i].sa_flags = 0;
}
sigemptyset(&signal_pending);
sigreturn = NULL;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
parent = NULL;
prevsibling = NULL;
nextsibling = NULL;
firstchild = NULL;
zombiechild = NULL;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
childlock = KTHREAD_MUTEX_INITIALIZER;
parentlock = KTHREAD_MUTEX_INITIALIZER;
zombiecond = KTHREAD_COND_INITIALIZER;
zombiewaiting = 0;
iszombie = false;
nozombify = false;
exit_code = -1;
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
group = NULL;
groupprev = NULL;
groupnext = NULL;
groupfirst = NULL;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
groupparentlock = KTHREAD_MUTEX_INITIALIZER;
groupchildlock = KTHREAD_MUTEX_INITIALIZER;
groupchildleft = KTHREAD_COND_INITIALIZER;
grouplimbo = false;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
firstthread = NULL;
threadlock = KTHREAD_MUTEX_INITIALIZER;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
threads_exiting = false;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
segments = NULL;
segments_used = 0;
segments_length = 0;
Fix read-only mmap with backing store.
2015-05-14 13:19:23 +00:00
segment_write_lock = KTHREAD_MUTEX_INITIALIZER;
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
segment_lock = KTHREAD_MUTEX_INITIALIZER;
Fix class process constructor forgetting some members.
2014-09-15 19:09:30 +00:00
user_timers_lock = KTHREAD_MUTEX_INITIALIZER;
memset(&user_timers, 0, sizeof(user_timers));
// alarm_timer initialized in member constructor.
// execute_clock initialized in member constructor.
// system_clock initialized in member constructor.
// execute_clock initialized in member constructor.
// child_execute_clock initialized in member constructor.
// child_system_clock initialized in member constructor.
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Time::InitializeProcessClocks(this);
alarm_timer.Attach(Time::GetClock(CLOCK_MONOTONIC));
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Process::~Process()
{
delete[] string_table;
delete[] symbol_table;
if ( alarm_timer.IsAttached() )
alarm_timer.Detach();
Fix memory leak of processes that are not process group leaders.
2014-05-14 17:11:14 +00:00
delete[] program_image_path;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
assert(!zombiechild);
assert(!firstchild);
assert(!addrspace);
assert(!segments);
assert(!dtable);
assert(!mtable);
assert(!cwd);
assert(!root);
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
assert(ptable);
ptable->Free(pid);
ptable.Reset();
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::BootstrapTables(Ref<DescriptorTable> dtable, Ref<MountTable> mtable)
{
ScopedLock lock(&ptrlock);
assert(!this->dtable);
assert(!this->mtable);
this->dtable = dtable;
this->mtable = mtable;
}
Refactored all the sortix headers into a include directory. Also got rid of trailing white space. That corrupted .git/. Big ass-commit because of recovered .git directory.
2012-03-21 23:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::BootstrapDirectories(Ref<Descriptor> root)
{
ScopedLock lock(&ptrlock);
assert(!this->root);
assert(!this->cwd);
this->root = root;
this->cwd = root;
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process__OnLastThreadExit(void* user);
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::OnThreadDestruction(Thread* thread)
{
assert(thread->process == this);
kthread_mutex_lock(&threadlock);
if ( thread->prevsibling )
thread->prevsibling->nextsibling = thread->nextsibling;
if ( thread->nextsibling )
thread->nextsibling->prevsibling = thread->prevsibling;
if ( thread == firstthread )
firstthread = thread->nextsibling;
if ( firstthread )
firstthread->prevsibling = NULL;
thread->prevsibling = thread->nextsibling = NULL;
bool threadsleft = firstthread;
kthread_mutex_unlock(&threadlock);
// We are called from the threads destructor, let it finish before we
// we handle the situation by killing ourselves.
if ( !threadsleft )
ScheduleDeath();
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::ScheduleDeath()
{
// All our threads must have exited at this point.
assert(!firstthread);
Worker::Schedule(Process__OnLastThreadExit, this);
}
Add alarm(2) and alarmns(2).
2013-05-11 23:24:42 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Useful for killing a partially constructed process without waiting for
// it to die and garbage collect its zombie. It is not safe to access this
// process after this call as another thread may garbage collect it.
void Process::AbortConstruction()
{
nozombify = true;
ScheduleDeath();
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process__OnLastThreadExit(void* user)
{
return ((Process*) user)->OnLastThreadExit();
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::OnLastThreadExit()
{
LastPrayer();
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::DeleteTimers()
{
for ( timer_t i = 0; i < PROCESS_TIMER_NUM_MAX; i++ )
{
if ( user_timers[i].timer.IsAttached() )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
user_timers[i].timer.Cancel();
user_timers[i].timer.Detach();
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::LastPrayer()
{
assert(this);
// This must never be called twice.
assert(!iszombie);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// This must be called from a thread using another address space as the
// address space of this process is about to be destroyed.
Thread* curthread = CurrentThread();
assert(curthread->process != this);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// This can't be called if the process is still alive.
assert(!firstthread);
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Disarm and detach all the timers in the process.
DeleteTimers();
if ( alarm_timer.IsAttached() )
Processes now remember the memory segments it has loaded.
2011-08-27 14:46:00 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
alarm_timer.Cancel();
alarm_timer.Detach();
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// We need to temporarily reload the correct addrese space of the dying
// process such that we can unmap and free its memory.
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
addr_t prevaddrspace = Memory::SwitchAddressSpace(addrspace);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
ResetAddressSpace();
if ( dtable ) dtable.Reset();
if ( cwd ) cwd.Reset();
if ( root ) root.Reset();
if ( mtable ) mtable.Reset();
// Destroy the address space and safely switch to the replacement
// address space before things get dangerous.
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
Memory::DestroyAddressSpace(prevaddrspace);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
addrspace = 0;
// Unload the process symbol and string tables.
delete[] symbol_table; symbol_table = NULL;
delete[] string_table; string_table = NULL;
// Init is nice and will gladly raise our orphaned children and zombies.
Process* init = Scheduler::GetInitProcess();
assert(init);
kthread_mutex_lock(&childlock);
while ( firstchild )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
ScopedLock firstchildlock(&firstchild->parentlock);
ScopedLock initlock(&init->childlock);
Process* process = firstchild;
firstchild = process->nextsibling;
process->parent = init;
process->prevsibling = NULL;
process->nextsibling = init->firstchild;
if ( init->firstchild )
init->firstchild->prevsibling = process;
init->firstchild = process;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Since we have no more children (they are with init now), we don't
// have to worry about new zombie processes showing up, so just collect
// those that are left. Then we satisfiy the invariant !zombiechild that
// applies on process termination.
bool hadzombies = zombiechild;
while ( zombiechild )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
ScopedLock zombiechildlock(&zombiechild->parentlock);
ScopedLock initlock(&init->childlock);
Process* zombie = zombiechild;
zombiechild = zombie->nextsibling;
zombie->prevsibling = NULL;
zombie->nextsibling = init->zombiechild;
if ( init->zombiechild )
init->zombiechild->prevsibling = zombie;
init->zombiechild = zombie;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
kthread_mutex_unlock(&childlock);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( hadzombies )
init->NotifyNewZombies();
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
iszombie = true;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
bool zombify = !nozombify;
// Remove ourself from our process group.
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
kthread_mutex_lock(&groupchildlock);
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
if ( group )
group->NotifyMemberExit(this);
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
kthread_mutex_unlock(&groupchildlock);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// This class instance will be destroyed by our parent process when it
// has received and acknowledged our death.
kthread_mutex_lock(&parentlock);
if ( parent )
parent->NotifyChildExit(this, zombify);
kthread_mutex_unlock(&parentlock);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// If nobody is waiting for us, then simply commit suicide.
if ( !zombify )
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
{
kthread_mutex_lock(&groupparentlock);
Fix memory leak of processes that are not process group leaders.
2014-05-14 17:11:14 +00:00
bool in_limbo = groupfirst && (grouplimbo = true);
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
kthread_mutex_unlock(&groupparentlock);
if ( !in_limbo )
delete this;
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
void Process::ResetAddressSpace()
{
Fix read-only mmap with backing store.
2015-05-14 13:19:23 +00:00
ScopedLock lock1(&segment_write_lock);
ScopedLock lock2(&segment_lock);
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
assert(Memory::GetAddressSpace() == addrspace);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
for ( size_t i = 0; i < segments_used; i++ )
Maintain counts of physical frames used for particular purposes.
2014-05-15 17:55:23 +00:00
Memory::UnmapRange(segments[i].addr, segments[i].size, PAGE_USAGE_USER_SPACE);
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
Memory::Flush();
segments_used = segments_length = 0;
free(segments);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
segments = NULL;
}
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
void Process::NotifyMemberExit(Process* child)
{
assert(child->group == this);
kthread_mutex_lock(&groupparentlock);
if ( child->groupprev )
child->groupprev->groupnext = child->groupnext;
else
groupfirst = child->groupnext;
if ( child->groupnext )
child->groupnext->groupprev = child->groupprev;
kthread_cond_signal(&groupchildleft);
kthread_mutex_unlock(&groupparentlock);
child->group = NULL;
NotifyLeftProcessGroup();
}
void Process::NotifyLeftProcessGroup()
{
ScopedLock parentlock(&groupparentlock);
if ( !grouplimbo || groupfirst )
return;
grouplimbo = false;
delete this;
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::NotifyChildExit(Process* child, bool zombify)
{
kthread_mutex_lock(&childlock);
if ( child->prevsibling )
child->prevsibling->nextsibling = child->nextsibling;
if ( child->nextsibling )
child->nextsibling->prevsibling = child->prevsibling;
if ( firstchild == child )
firstchild = child->nextsibling;
if ( firstchild )
firstchild->prevsibling = NULL;
if ( zombify )
{
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
if ( zombiechild )
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
zombiechild->prevsibling = child;
child->prevsibling = NULL;
child->nextsibling = zombiechild;
zombiechild = child;
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
kthread_mutex_unlock(&childlock);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( zombify )
NotifyNewZombies();
}
Maintain recursive process execution and system time.
2013-05-14 22:58:54 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::NotifyNewZombies()
{
ScopedLock lock(&childlock);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
DeliverSignal(SIGCHLD);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( zombiewaiting )
kthread_cond_broadcast(&zombiecond);
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Fix waitpid status copying to user-space.
2014-10-01 17:17:49 +00:00
pid_t Process::Wait(pid_t thepid, int* status_ptr, int options)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
// TODO: Process groups are not supported yet.
Fix waitpid status copying to user-space.
2014-10-01 17:17:49 +00:00
if ( thepid < -1 || thepid == 0 )
return errno = ENOSYS, -1;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
ScopedLock lock(&childlock);
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// A process can only wait if it has children.
if ( !firstchild && !zombiechild )
return errno = ECHILD, -1;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Processes can only wait for their own children to exit.
if ( 0 < thepid )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// TODO: This is a slow but multithread safe way to verify that the
// target process has the correct parent.
bool found = false;
for ( Process* p = firstchild; !found && p; p = p->nextsibling )
if ( p->pid == thepid )
found = true;
for ( Process* p = zombiechild; !found && p; p = p->nextsibling )
if ( p->pid == thepid )
found = true;
if ( !found )
return errno = ECHILD, -1;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Process* zombie = NULL;
while ( !zombie )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
for ( zombie = zombiechild; zombie; zombie = zombie->nextsibling )
if ( thepid == -1 || thepid == zombie->pid )
break;
if ( zombie )
break;
if ( options & WNOHANG )
return 0;
zombiewaiting++;
unsigned long r = kthread_cond_wait_signal(&zombiecond, &childlock);
zombiewaiting--;
if ( !r )
return errno = EINTR, -1;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
// Remove from the list of zombies.
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( zombie->prevsibling )
zombie->prevsibling->nextsibling = zombie->nextsibling;
if ( zombie->nextsibling )
zombie->nextsibling->prevsibling = zombie->prevsibling;
if ( zombiechild == zombie )
zombiechild = zombie->nextsibling;
if ( zombiechild )
zombiechild->prevsibling = NULL;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
thepid = zombie->pid;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// It is safe to access these clocks directly as the child process is no
// longer running at this point and the values are nicely frozen.
child_execute_clock.Advance(zombie->child_execute_clock.current_time);
child_system_clock.Advance(zombie->child_system_clock.current_time);
Added a process execute API.
2011-08-27 18:57:39 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
int status = zombie->exit_code;
Fix insecure user-space pointer dereferences in sys_waitpid.
2014-01-19 17:16:19 +00:00
if ( status < 0 )
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
status = WCONSTRUCT(WNATURE_SIGNALED, 128 + SIGKILL, SIGKILL);
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
kthread_mutex_lock(&zombie->groupparentlock);
Fix memory leak of processes that are not process group leaders.
2014-05-14 17:11:14 +00:00
bool in_limbo = zombie->groupfirst && (zombie->grouplimbo = true);
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
kthread_mutex_unlock(&zombie->groupparentlock);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// And so, the process was fully deleted.
Implement sending signals to process groups.
2013-06-11 23:35:31 +00:00
if ( !in_limbo )
delete zombie;
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
Fix waitpid status copying to user-space.
2014-10-01 17:17:49 +00:00
if ( status_ptr )
*status_ptr = status;
Fix insecure user-space pointer dereferences in sys_waitpid.
2014-01-19 17:16:19 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return thepid;
}
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
pid_t sys_waitpid(pid_t pid, int* user_status, int options)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
Fix waitpid status copying to user-space.
2014-10-01 17:17:49 +00:00
int status = 0;
pid_t ret = CurrentProcess()->Wait(pid, &status, options);
if ( 0 < ret && !CopyToUser(user_status, &status, sizeof(status)) )
return -1;
return ret;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Add chroot(2), fchroot(2) and fchrootat(2).
2013-05-29 22:32:22 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
void Process::ExitThroughSignal(int signal)
{
ExitWithCode(WCONSTRUCT(WNATURE_SIGNALED, 128 + signal, signal));
}
void Process::ExitWithCode(int requested_exit_code)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
ScopedLock lock(&threadlock);
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( exit_code == -1 )
exit_code = requested_exit_code;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Broadcast SIGKILL to all our threads which will begin our long path
// of process termination. We simply can't stop the threads as they may
// be running in kernel mode doing dangerous stuff. This thread will be
// destroyed by SIGKILL once the system call returns.
for ( Thread* t = firstthread; t; t = t->nextsibling )
t->DeliverSignal(SIGKILL);
}
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::AddChildProcess(Process* child)
{
ScopedLock mylock(&childlock);
ScopedLock itslock(&child->parentlock);
assert(!child->parent);
assert(!child->nextsibling);
assert(!child->prevsibling);
child->parent = this;
child->nextsibling = firstchild;
child->prevsibling = NULL;
if ( firstchild )
firstchild->prevsibling = child;
firstchild = child;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Ref<MountTable> Process::GetMTable()
{
ScopedLock lock(&ptrlock);
assert(mtable);
return mtable;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Ref<DescriptorTable> Process::GetDTable()
{
ScopedLock lock(&ptrlock);
assert(dtable);
return dtable;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
Ref<ProcessTable> Process::GetPTable()
{
ScopedLock lock(&ptrlock);
assert(ptable);
return ptable;
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Ref<Descriptor> Process::GetRoot()
{
ScopedLock lock(&ptrlock);
assert(root);
return root;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Ref<Descriptor> Process::GetCWD()
{
ScopedLock lock(&ptrlock);
assert(cwd);
return cwd;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::SetRoot(Ref<Descriptor> newroot)
{
ScopedLock lock(&ptrlock);
assert(newroot);
root = newroot;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::SetCWD(Ref<Descriptor> newcwd)
{
ScopedLock lock(&ptrlock);
assert(newcwd);
cwd = newcwd;
}
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Ref<Descriptor> Process::GetDescriptor(int fd)
{
ScopedLock lock(&ptrlock);
assert(dtable);
return dtable->Get(fd);
}
Refactor kernel VFS. Note: This is an incompatible ABI change.
2012-08-07 22:19:44 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Process* Process::Fork()
{
assert(CurrentProcess() == this);
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
// TODO: This adds the new process to the process table, but it's not ready
// and functions that access this new process will be surprised that
// it's not fully constructed and really bad things will happen.
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Process* clone = new Process;
if ( !clone )
return NULL;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
if ( (clone->pid = (clone->ptable = ptable)->Allocate(clone)) < 0 )
{
delete clone;
return NULL;
}
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
struct segment* clone_segments = NULL;
Add {g,s}et{e,}{g,u}id(2).
2013-01-13 01:37:14 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Fork the segment list.
if ( segments )
{
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
size_t segments_size = sizeof(struct segment) * segments_used;
if ( !(clone_segments = (struct segment*) malloc(segments_size)) )
{
delete clone;
return NULL;
}
memcpy(clone_segments, segments, segments_size);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Keep track of program image path.
2013-03-19 22:17:36 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Fork address-space here and copy memory.
clone->addrspace = Memory::Fork();
if ( !clone->addrspace )
{
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
free(clone_segments);
delete clone;
return NULL;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Load program symbol tables.
2013-05-22 20:11:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Now it's too late to clean up here, if anything goes wrong, we simply
// ask the process to commit suicide before it goes live.
Refactor kernel segment bookkeeping.
2013-08-20 00:23:53 +00:00
clone->segments = clone_segments;
clone->segments_used = segments_used;
clone->segments_length = segments_used;
Load program symbol tables.
2013-05-22 20:11:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Remember the relation to the child process.
AddChildProcess(clone);
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
// Add the new process to the current process group.
kthread_mutex_lock(&groupchildlock);
kthread_mutex_lock(&group->groupparentlock);
clone->group = group;
clone->groupprev = NULL;
if ( (clone->groupnext = group->groupfirst) )
group->groupfirst->groupprev = clone;
group->groupfirst = clone;
kthread_mutex_unlock(&group->groupparentlock);
kthread_mutex_unlock(&groupchildlock);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Initialize everything that is safe and can't fail.
Add prlimit(2).
2013-08-30 21:55:09 +00:00
kthread_mutex_lock(&resource_limits_lock);
for ( size_t i = 0; i < RLIMIT_NUM_DECLARED; i++ )
clone->resource_limits[i] = resource_limits[i];
kthread_mutex_unlock(&resource_limits_lock);
Add getpriority(2) and setpriority(2).
2013-08-30 15:35:30 +00:00
kthread_mutex_lock(&nicelock);
clone->nice = nice;
kthread_mutex_unlock(&nicelock);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
kthread_mutex_lock(&ptrlock);
clone->root = root;
clone->cwd = cwd;
kthread_mutex_unlock(&ptrlock);
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Fix tfork(2) not copying signal properties.
2014-09-15 19:05:14 +00:00
kthread_mutex_lock(&idlock);
clone->uid = uid;
clone->gid = gid;
clone->euid = euid;
clone->egid = egid;
clone->umask = umask;
kthread_mutex_unlock(&idlock);
kthread_mutex_lock(&signal_lock);
memcpy(&clone->signal_actions, &signal_actions, sizeof(signal_actions));
sigemptyset(&clone->signal_pending);
clone->sigreturn = sigreturn;
kthread_mutex_unlock(&signal_lock);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// Initialize things that can fail and abort if needed.
bool failure = false;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
kthread_mutex_lock(&ptrlock);
if ( !(clone->dtable = dtable->Fork()) )
failure = true;
//if ( !(clone->mtable = mtable->Fork()) )
// failure = true;
clone->mtable = mtable;
kthread_mutex_unlock(&ptrlock);
Improved JSVM compabillity.
2011-11-20 14:58:42 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( !(clone->program_image_path = String::Clone(program_image_path)) )
Fix error case in tfork(2).
2014-09-15 18:58:12 +00:00
failure = true;
Add alarm(2) and alarmns(2).
2013-05-11 23:24:42 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( string_table && (clone->string_table = new char[string_table_length]) )
{
memcpy(clone->string_table, string_table, string_table_length);
clone->string_table_length = string_table_length;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( clone->string_table && symbol_table &&
(clone->symbol_table = new Symbol[symbol_table_length]) )
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
{
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
for ( size_t i = 0; i < symbol_table_length; i++ )
{
clone->symbol_table[i].address = symbol_table[i].address;
clone->symbol_table[i].size = symbol_table[i].size;
clone->symbol_table[i].name =
(const char*)((uintptr_t) symbol_table[i].name -
(uintptr_t) string_table +
(uintptr_t) clone->string_table);
}
clone->symbol_table_length = symbol_table_length;
}
Keep track of program image path.
2013-03-19 22:17:36 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// If the proces creation failed, ask the process to commit suicide and
// not become a zombie, as we don't wait for it to exit. It will clean
// up all the above resources and delete itself.
if ( failure )
{
clone->AbortConstruction();
return NULL;
}
Argv now works in main.
2011-11-09 22:18:26 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return clone;
}
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
void Process::ResetForExecute()
{
string_table_length = 0;
symbol_table_length = 0;
delete[] string_table; string_table = NULL;
delete[] symbol_table; symbol_table = NULL;
Argv now works in main.
2011-11-09 22:18:26 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
DeleteTimers();
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
for ( int i = 0; i < SIG_MAX_NUM; i++ )
{
signal_actions[i].sa_flags = 0;
if ( signal_actions[i].sa_handler == SIG_DFL )
continue;
if ( signal_actions[i].sa_handler == SIG_IGN )
continue;
signal_actions[i].sa_handler = SIG_DFL;
}
sigreturn = NULL;
stack_t* signal_stack = &CurrentThread()->signal_stack;
memset(signal_stack, 0, sizeof(*signal_stack));
signal_stack->ss_flags = SS_DISABLE;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
ResetAddressSpace();
}
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
bool Process::MapSegment(struct segment* result, void* hint, size_t size,
int flags, int prot)
{
Fix read-only mmap with backing store.
2015-05-14 13:19:23 +00:00
// process->segment_write_lock is held at this point.
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
// process->segment_lock is held at this point.
if ( !size )
{
result->addr = 0x0;
result->size = 0x0;
result->prot = prot;
return true;
}
if ( !PlaceSegment(result, this, hint, size, flags) )
return false;
if ( !Memory::MapMemory(this, result->addr, result->size, result->prot = prot) )
{
// The caller is expected to self-destruct in this case, so the
// segment just created is not removed.
return false;
}
Memory::Flush();
return true;
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
int Process::Execute(const char* programname, const uint8_t* program,
size_t programsize, int argc, const char* const* argv,
int envc, const char* const* envp,
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
struct thread_registers* regs)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
assert(argc != INT_MAX);
assert(envc != INT_MAX);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
assert(CurrentProcess() == this);
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
char* programname_clone = String::Clone(programname);
if ( !programname_clone )
return -1;
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Handle ELF notes during program load.
2014-02-17 19:56:19 +00:00
ELF::Auxiliary aux;
Rewrite program loader.
2014-06-25 23:05:07 +00:00
addr_t entry = ELF::Load(program, programsize, &aux);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( !entry ) { delete[] programname_clone; return -1; }
Argv now works in main.
2011-11-09 22:18:26 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
delete[] program_image_path;
program_image_path = programname_clone; programname_clone = NULL;
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Refactor user-space stack creation.
2013-08-22 14:08:08 +00:00
uintptr_t userspace_addr;
size_t userspace_size;
Memory::GetUserVirtualArea(&userspace_addr, &userspace_size);
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
const size_t stack_size = 512UL * 1024UL;
void* stack_hint = (void*) (userspace_addr + userspace_size - stack_size);
const int stack_prot = PROT_READ | PROT_WRITE | PROT_KREAD | PROT_KWRITE | PROT_FORK;
if ( !aux.tls_mem_align )
aux.tls_mem_align = 1;
if ( Page::Size() < aux.tls_mem_align )
return errno = EINVAL, -1;
if ( !aux.uthread_align )
aux.uthread_align = 1;
if ( Page::Size() < aux.uthread_align )
return errno = EINVAL, -1;
if ( aux.uthread_size < sizeof(struct uthread) )
aux.uthread_size = sizeof(struct uthread);
size_t raw_tls_size = aux.tls_mem_size;
size_t raw_tls_size_aligned = -(-raw_tls_size & ~(aux.tls_mem_align-1));
if ( raw_tls_size && raw_tls_size_aligned == 0 /* overflow */ )
return errno = EINVAL, -1;
int raw_tls_prot = PROT_READ | PROT_KREAD | PROT_KWRITE | PROT_FORK;
void* raw_tls_hint = stack_hint;
size_t tls_size = raw_tls_size_aligned + aux.uthread_size;
size_t tls_offset_tls = 0;
size_t tls_offset_uthread = raw_tls_size_aligned;
if ( aux.tls_mem_align < aux.uthread_align )
{
size_t more_aligned = -(-raw_tls_size_aligned & ~(aux.uthread_align-1));
if ( raw_tls_size_aligned && more_aligned == 0 /* overflow */ )
return errno = EINVAL, -1;
size_t difference = more_aligned - raw_tls_size_aligned;
tls_size += difference;
tls_offset_tls += difference;
tls_offset_uthread += difference;
}
assert((tls_offset_tls & (aux.tls_mem_align-1)) == 0);
assert((tls_offset_uthread & (aux.uthread_align-1)) == 0);
int tls_prot = PROT_READ | PROT_WRITE | PROT_KREAD | PROT_KWRITE | PROT_FORK;
void* tls_hint = stack_hint;
Refactor user-space stack creation.
2013-08-22 14:08:08 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
size_t auxcode_size = Page::Size();
int auxcode_prot = PROT_EXEC | PROT_READ | PROT_KREAD | PROT_KWRITE | PROT_FORK;
void* auxcode_hint = stack_hint;
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
size_t arg_size = 0;
size_t argv_size = sizeof(char*) * (argc + 1);
size_t envp_size = sizeof(char*) * (envc + 1);
arg_size += argv_size;
arg_size += envp_size;
for ( int i = 0; i < argc; i++ )
arg_size += strlen(argv[i]) + 1;
for ( int i = 0; i < envc; i++ )
arg_size += strlen(envp[i]) + 1;
struct segment arg_segment;
Refactor user-space stack creation.
2013-08-22 14:08:08 +00:00
struct segment stack_segment;
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
struct segment raw_tls_segment;
struct segment tls_segment;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
struct segment auxcode_segment;
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
Fix read-only mmap with backing store.
2015-05-14 13:19:23 +00:00
kthread_mutex_lock(&segment_write_lock);
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
kthread_mutex_lock(&segment_lock);
if ( !(MapSegment(&arg_segment, stack_hint, arg_size, 0, stack_prot) &&
MapSegment(&stack_segment, stack_hint, stack_size, 0, stack_prot) &&
MapSegment(&raw_tls_segment, raw_tls_hint, raw_tls_size, 0, raw_tls_prot) &&
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
MapSegment(&tls_segment, tls_hint, tls_size, 0, tls_prot) &&
MapSegment(&auxcode_segment, auxcode_hint, auxcode_size, 0, auxcode_prot)) )
Refactor user-space stack creation.
2013-08-22 14:08:08 +00:00
{
kthread_mutex_unlock(&segment_lock);
Fix read-only mmap with backing store.
2015-05-14 13:19:23 +00:00
kthread_mutex_unlock(&segment_write_lock);
Refactor user-space stack creation.
2013-08-22 14:08:08 +00:00
ResetForExecute();
return errno = ENOMEM, -1;
}
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
kthread_mutex_unlock(&segment_lock);
Fix read-only mmap with backing store.
2015-05-14 13:19:23 +00:00
kthread_mutex_unlock(&segment_write_lock);
Refactored the descriptor table and added FD_CLOEXEC and FD_CLOFORK. open(2) now also respects O_CLOEXEC and O_CLOFORK.
2012-03-04 20:06:49 +00:00
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
char** target_argv = (char**) (arg_segment.addr + 0);
char** target_envp = (char**) (arg_segment.addr + argv_size);
char* target_strings = (char*) (arg_segment.addr + argv_size + envp_size);
size_t target_strings_offset = 0;
execve(2) can now load programs from the filesystem. Previously it was restricted to only the ramdisk.
2011-11-22 12:53:36 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
for ( int i = 0; i < argc; i++ )
{
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
const char* arg = argv[i];
size_t arg_len = strlen(arg);
char* target_arg = target_strings + target_strings_offset;
strcpy(target_arg, arg);
target_argv[i] = target_arg;
target_strings_offset += arg_len + 1;
execve(2) can now load programs from the filesystem. Previously it was restricted to only the ramdisk.
2011-11-22 12:53:36 +00:00
}
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
target_argv[argc] = (char*) NULL;
execve(2) can now load programs from the filesystem. Previously it was restricted to only the ramdisk.
2011-11-22 12:53:36 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
for ( int i = 0; i < envc; i++ )
execve(2) can now load programs from the filesystem. Previously it was restricted to only the ramdisk.
2011-11-22 12:53:36 +00:00
{
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
const char* env = envp[i];
size_t env_len = strlen(env);
char* target_env = target_strings + target_strings_offset;
strcpy(target_env, env);
target_envp[i] = target_env;
target_strings_offset += env_len + 1;
execve(2) can now load programs from the filesystem. Previously it was restricted to only the ramdisk.
2011-11-22 12:53:36 +00:00
}
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
target_envp[envc] = (char*) NULL;
const uint8_t* file_raw_tls = program + aux.tls_file_offset;
uint8_t* target_raw_tls = (uint8_t*) raw_tls_segment.addr;
memcpy(target_raw_tls, file_raw_tls, aux.tls_file_size);
memset(target_raw_tls + aux.tls_file_size, 0, aux.tls_mem_size - aux.tls_file_size);
uint8_t* target_tls = (uint8_t*) (tls_segment.addr + tls_offset_tls);
assert((((uintptr_t) target_tls) & (aux.tls_mem_align-1)) == 0);
memcpy(target_tls, file_raw_tls, aux.tls_file_size);
memset(target_tls + aux.tls_file_size, 0, aux.tls_mem_size - aux.tls_file_size);
struct uthread* uthread = (struct uthread*) (tls_segment.addr + tls_offset_uthread);
assert((((uintptr_t) uthread) & (aux.uthread_align-1)) == 0);
memset(uthread, 0, sizeof(*uthread));
uthread->uthread_pointer = uthread;
uthread->uthread_size = aux.uthread_size;
uthread->uthread_flags = UTHREAD_FLAG_INITIAL;
uthread->tls_master_mmap = (void*) raw_tls_segment.addr;
uthread->tls_master_size = aux.tls_mem_size;
uthread->tls_master_align = aux.tls_mem_align;
uthread->tls_mmap = (void*) tls_segment.addr;
uthread->tls_size = tls_size;
uthread->stack_mmap = (void*) stack_segment.addr;
uthread->stack_size = stack_segment.size;
uthread->arg_mmap = (void*) arg_segment.addr;
uthread->arg_size = arg_segment.size;
memset(uthread + 1, 0, aux.uthread_size - sizeof(struct uthread));
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
memset(regs, 0, sizeof(*regs));
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#if defined(__i386__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
regs->eax = argc;
regs->ebx = (size_t) target_argv;
regs->edx = envc;
regs->ecx = (size_t) target_envp;
regs->eip = entry;
regs->esp = (stack_segment.addr + stack_segment.size) & ~15UL;
regs->ebp = regs->esp;
regs->cs = UCS | URPL;
regs->ds = UDS | URPL;
regs->ss = UDS | URPL;
regs->eflags = FLAGS_RESERVED1 | FLAGS_INTERRUPT | FLAGS_ID;
regs->signal_pending = 0;
regs->gsbase = (uint32_t) uthread;
regs->cr3 = addrspace;
regs->kernel_stack = GDT::GetKernelStack();
memcpy(regs->fpuenv, Float::fpu_initialized_regs, 512);
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#elif defined(__x86_64__)
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
regs->rdi = argc;
regs->rsi = (size_t) target_argv;
regs->rdx = envc;
regs->rcx = (size_t) target_envp;
regs->rip = entry;
regs->rsp = (stack_segment.addr + stack_segment.size) & ~15UL;
regs->rbp = regs->rsp;
regs->cs = UCS | URPL;
regs->ds = UDS | URPL;
regs->ss = UDS | URPL;
regs->rflags = FLAGS_RESERVED1 | FLAGS_INTERRUPT | FLAGS_ID;
regs->signal_pending = 0;
regs->fsbase = (uint64_t) uthread;
regs->cr3 = addrspace;
regs->kernel_stack = GDT::GetKernelStack();
memcpy(regs->fpuenv, Float::fpu_initialized_regs, 512);
#else
#warning "You need to implement initializing the first thread after execute"
Add kernel uthread support.
2014-02-17 22:53:03 +00:00
#endif
execve(2) now pushes envp to the new stack and sets up registers. This fully implements environmental variables over exec.
2012-04-03 23:41:29 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
uint8_t* auxcode = (uint8_t*) auxcode_segment.addr;
#if defined(__i386__)
sigreturn = (void (*)(void)) &auxcode[0];
auxcode[0] = 0xCD; /* int .... */
auxcode[1] = 0x83; /* ... $131 */
#elif defined(__x86_64__)
sigreturn = (void (*)(void)) &auxcode[0];
auxcode[0] = 0xCD; /* int .... */
auxcode[1] = 0x83; /* ... $131 */
#else
(void) auxcode;
#warning "You need to initialize auxcode with a sigreturn routine"
#endif
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
dtable->OnExecute();
Added a shell, a few programs, and added support for SIGINT.
2011-08-27 21:03:39 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return 0;
}
Argv now works in main.
2011-11-09 22:18:26 +00:00
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
static
const char* shebang_lookup_environment(const char* name, char* const* envp)
{
size_t equalpos = strcspn(name, "=");
if ( name[equalpos] == '=' )
return NULL;
size_t namelen = equalpos;
for ( size_t i = 0; envp[i]; i++ )
{
if ( strncmp(name, envp[i], namelen) )
continue;
if ( envp[i][namelen] != '=' )
continue;
return envp[i] + namelen + 1;
}
return NULL;
}
static char* shebang_tokenize(char** saved)
{
char* data = *saved;
if ( !data )
return *saved = NULL;
Fix ctype invocations with wrong domain.
2015-08-11 13:37:31 +00:00
while ( data[0] && isspace((unsigned char) data[0]) )
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
data++;
if ( !data[0] )
return *saved = NULL;
size_t input = 0;
size_t output = 0;
bool singly = false;
bool doubly = false;
bool escaped = false;
for ( ; data[input]; input++ )
{
char c = data[input];
Fix ctype invocations with wrong domain.
2015-08-11 13:37:31 +00:00
if ( !escaped && !singly && !doubly && isspace((unsigned char) c) )
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
break;
if ( !escaped && !doubly && c == '\'' )
{
singly = !singly;
continue;
}
if ( !escaped && !singly && c == '"' )
{
doubly = !doubly;
continue;
}
if ( !singly && !escaped && c == '\\' )
{
escaped = true;
continue;
}
if ( escaped )
{
switch ( c )
{
case 'a': c = '\a'; break;
case 'b': c = '\b'; break;
case 'e': c = '\e'; break;
case 'f': c = '\f'; break;
case 'n': c = '\n'; break;
case 'r': c = '\r'; break;
case 't': c = '\t'; break;
case 'v': c = '\v'; break;
default: break;
};
}
escaped = false;
data[output++] = c;
}
if ( data[input] )
*saved = data + input + 1;
else
*saved = NULL;
data[output] = '\0';
return data;
}
static size_t shebang_count_arguments(char* line)
{
size_t result = 0;
while ( shebang_tokenize(&line) )
result++;
return result;
}
// NOTE: The PATH-searching logic is repeated multiple places. Until this logic
// can be shared somehow, you need to keep this comment in sync as well
// as the logic in these files:
// * kernel/process.cpp
// * libc/unistd/execvpe.cpp
// * utils/which.cpp
Fix NOTO typo.
2014-12-28 15:40:36 +00:00
// NOTE: See comments in execvpe() for algorithmic commentary.
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
static bool sys_execve_alloc(addralloc_t* alloc, size_t size)
{
if ( !AllocateKernelAddress(alloc, size) )
return false;
if ( !Memory::MapRange(alloc->from, alloc->size, PROT_KREAD | PROT_KWRITE, PAGE_USAGE_EXECUTE) )
return FreeKernelAddress(alloc), false;
Memory::Flush();
return true;
}
static void sys_execve_free(addralloc_t* alloc)
{
Memory::UnmapRange(alloc->from, alloc->size, PAGE_USAGE_EXECUTE);
Memory::Flush();
FreeKernelAddress(alloc);
}
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
static
int sys_execve_kernel(const char* filename,
int argc,
Replace system call array types with pointers.
2015-01-22 23:29:04 +00:00
char* const* argv,
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
int envc,
Replace system call array types with pointers.
2015-01-22 23:29:04 +00:00
char* const* envp,
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
struct thread_registers* regs)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
Process* process = CurrentProcess();
ioctx_t ctx;
SetupKernelIOCtx(&ctx);
Ref<Descriptor> from = filename[0] == '/' ? process->GetRoot() : process->GetCWD();
Ref<Descriptor> desc = from->open(&ctx, filename, O_EXEC | O_READ, 0);
if ( !desc )
return -1;
Fix execve(2) keeping program descriptors open.
2015-07-09 16:32:30 +00:00
from.Reset();
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
struct stat st;
if ( desc->stat(&ctx, &st) )
return -1;
if ( st.st_size < 0 )
return errno = EINVAL, -1;
if ( (uintmax_t) SIZE_MAX < (uintmax_t) st.st_size )
return errno = EFBIG, -1;
size_t filesize = (size_t) st.st_size;
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
addralloc_t buffer_alloc;
if ( !sys_execve_alloc(&buffer_alloc, filesize) )
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
return -1;
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
uint8_t* buffer = (uint8_t*) buffer_alloc.from;
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
for ( size_t sofar = 0; sofar < filesize; )
{
ssize_t amount = desc->read(&ctx, buffer + sofar, filesize - sofar);
if ( amount < 0 )
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
return sys_execve_free(&buffer_alloc), -1;
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
if ( amount == 0 )
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
return sys_execve_free(&buffer_alloc), errno = EEOF, -1;
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
sofar += amount;
}
Fix execve(2) keeping program descriptors open.
2015-07-09 16:32:30 +00:00
desc.Reset();
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
int result = process->Execute(filename, buffer, filesize, argc, argv, envc, envp, regs);
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
if ( result == 0 || errno != ENOEXEC ||
filesize < 2 || buffer[0] != '#' || buffer[1] != '!' )
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
return sys_execve_free(&buffer_alloc), result;
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
size_t line_length = 0;
while ( line_length < filesize && buffer[2 + line_length] != '\n' )
line_length++;
if ( line_length == filesize )
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
return sys_execve_free(&buffer_alloc), errno = ENOEXEC, -1;
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
char* line = new char[line_length+1];
if ( !line )
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
return sys_execve_free(&buffer_alloc), -1;
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
memcpy(line, buffer + 2, line_length);
line[line_length] = '\0';
Store loaded programs in special allocations rather than the kernel heap.
2014-05-14 12:41:51 +00:00
sys_execve_free(&buffer_alloc);
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
Add #! support to execve(2).
2014-05-10 19:34:29 +00:00
char* line_clone = String::Clone(line);
if ( !line_clone )
return delete[] line, -1;
size_t argument_count = shebang_count_arguments(line_clone);
delete[] line_clone;
if ( !argument_count || INT_MAX < argument_count )
return delete[] line, errno = ENOEXEC, -1;
int sb_argc = (int) argument_count;
char** sb_argv = new char*[sb_argc];
if ( !sb_argv )
return delete[] line, -1;
char* sb_saved = line;
for ( int i = 0; i < sb_argc; i++ )
sb_argv[i] = shebang_tokenize(&sb_saved);
if ( INT_MAX - argc <= sb_argc )
return delete[] sb_argv, delete[] line, errno = EOVERFLOW, -1;
if ( !sb_argv[0][0] )
return delete[] sb_argv, delete[] line, errno = ENOENT, -1;
int new_argc = sb_argc + argc;
char** new_argv = new char*[new_argc + 1];
if ( !new_argv )
return delete[] sb_argv, delete[] line, -1;
for ( int i = 0; i < sb_argc; i++ )
new_argv[i] = sb_argv[i];
new_argv[sb_argc + 0] = (char*) filename;
for ( int i = 1; i < argc; i++ )
new_argv[sb_argc + i] = argv[i];
new_argv[new_argc] = (char*) NULL;
result = -1;
// (See the above comment block before editing this searching logic)
const char* path = shebang_lookup_environment("PATH", envp);
bool search_path = !strchr(sb_argv[0], '/') && path;
bool any_tries = false;
bool any_eacces = false;
const char* new_argv0 = sb_argv[0];
while ( search_path && *path )
{
size_t len = strcspn(path, ":");
if ( !len )
{
path++;
continue;
}
any_tries = true;
char* dirpath = strndup(path, len);
if ( !dirpath )
return -1;
if ( (path += len)[0] == ':' )
path++;
while ( len && dirpath[len - 1] == '/' )
dirpath[--len] = '\0';
char* fullpath;
if ( asprintf(&fullpath, "%s/%s", dirpath, sb_argv[0]) < 0 )
return free(dirpath), -1;
result = sys_execve_kernel(fullpath, new_argc, new_argv, envc, envp, regs);
free(fullpath);
free(dirpath);
if ( result == 0 )
break;
if ( errno == ENOENT )
continue;
if ( errno == ELOOP ||
errno == EISDIR ||
errno == ENAMETOOLONG ||
errno == ENOTDIR )
continue;
if ( errno == EACCES )
{
any_eacces = true;
continue;
}
if ( errno == EACCES )
{
any_eacces = true;
continue;
}
break;
}
if ( !any_tries )
result = sys_execve_kernel(new_argv0, new_argc, new_argv, envc, envp, regs);
if ( result < 0 && any_eacces )
errno = EACCES;
delete[] new_argv;
delete[] sb_argv;
delete[] line;
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
return result;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
int sys_execve(const char* user_filename,
Replace system call array types with pointers.
2015-01-22 23:29:04 +00:00
char* const* user_argv,
char* const* user_envp)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
char* filename;
int argc;
int envc;
char** argv;
char** envp;
int result = -1;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
struct thread_registers regs;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
memset(&regs, 0, sizeof(regs));
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
if ( !user_filename || !user_argv || !user_envp )
return errno = EFAULT, -1;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
if ( !(filename = GetStringFromUser(user_filename)) )
goto cleanup_done;
argc = 0;
while ( true )
{
const char* user_arg;
if ( !CopyFromUser(&user_arg, user_argv + argc, sizeof(user_arg)) )
goto cleanup_filename;
if ( !user_arg )
break;
if ( ++argc == INT_MAX )
{
errno = E2BIG;
goto cleanup_filename;
}
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
argv = new char*[argc+1];
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
if ( !argv )
goto cleanup_filename;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
memset(argv, 0, sizeof(char*) * (argc+1));
for ( int i = 0; i < argc; i++ )
{
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
const char* user_arg;
if ( !CopyFromUser(&user_arg, user_argv + i, sizeof(user_arg)) )
goto cleanup_argv;
if ( !(argv[i] = GetStringFromUser(user_arg)) )
goto cleanup_argv;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
}
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
envc = 0;
while ( true )
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
{
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
const char* user_env;
if ( !CopyFromUser(&user_env, user_envp + envc, sizeof(user_env)) )
goto cleanup_argv;
if ( !user_env )
break;
if ( ++envc == INT_MAX )
{
errno = E2BIG;
goto cleanup_argv;
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
envp = new char*[envc+1];
if ( !envp )
goto cleanup_argv;
memset(envp, 0, sizeof(char*) * (envc+1));
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
for ( int i = 0; i < envc; i++ )
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
const char* user_env;
if ( !CopyFromUser(&user_env, user_envp + i, sizeof(user_env)) )
goto cleanup_envp;
if ( !(envp[i] = GetStringFromUser(user_envp[i])) )
goto cleanup_envp;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
result = sys_execve_kernel(filename, argc, argv, envc, envp, &regs);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
cleanup_envp:
for ( int i = 0; i < envc; i++)
delete[] envp[i];
delete[] envp;
cleanup_argv:
for ( int i = 0; i < argc; i++)
delete[] argv[i];
delete[] argv;
cleanup_filename:
delete[] filename;
cleanup_done:
Fix insecure user-space pointer dereferences in sys_execve.
2014-05-10 15:21:27 +00:00
if ( result == 0 )
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
LoadRegisters(&regs);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return result;
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
pid_t sys_tfork(int flags, struct tfork* user_regs)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
struct tfork regs;
Fix insecure user-space pointer dereferences in sys_tfork.
2014-01-19 17:22:53 +00:00
if ( !CopyFromUser(&regs, user_regs, sizeof(regs)) )
return -1;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( Signal::IsPending() )
return errno = EINTR, -1;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Implement thread creation in tfork(2).
2013-08-31 16:02:53 +00:00
bool making_process = flags == SFFORK;
bool making_thread = (flags & (SFPROC | SFPID | SFFD | SFMEM | SFCWD | SFROOT)) == SFPROC;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// TODO: Properly support tfork(2).
Implement thread creation in tfork(2).
2013-08-31 16:02:53 +00:00
if ( !(making_thread || making_process) )
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return errno = ENOSYS, -1;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
if ( regs.altstack.ss_flags & ~__SS_SUPPORTED_FLAGS )
return errno = EINVAL, -1;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
size_t stack_alignment = 16;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// TODO: Is it a hack to create a new kernel stack here?
Thread* curthread = CurrentThread();
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
size_t newkernelstacksize = curthread->kernelstacksize;
uint8_t* newkernelstack = new uint8_t[newkernelstacksize + stack_alignment];
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( !newkernelstack )
return -1;
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
uintptr_t stack_aligned = (uintptr_t) newkernelstack;
size_t stack_aligned_size = newkernelstacksize;
if ( ((uintptr_t) stack_aligned) & (stack_alignment-1) )
stack_aligned = (stack_aligned + 16) & ~(stack_alignment-1);
stack_aligned_size &= 0xFFFFFFF0;
Implement thread creation in tfork(2).
2013-08-31 16:02:53 +00:00
Process* child_process;
if ( making_thread )
child_process = CurrentProcess();
else if ( !(child_process = CurrentProcess()->Fork()) )
{
delete[] newkernelstack;
return -1;
}
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
struct thread_registers cpuregs;
memset(&cpuregs, 0, sizeof(cpuregs));
#if defined(__i386__)
cpuregs.eip = regs.eip;
cpuregs.esp = regs.esp;
cpuregs.eax = regs.eax;
cpuregs.ebx = regs.ebx;
cpuregs.ecx = regs.ecx;
cpuregs.edx = regs.edx;
cpuregs.edi = regs.edi;
cpuregs.esi = regs.esi;
cpuregs.ebp = regs.ebp;
cpuregs.cs = UCS | URPL;
cpuregs.ds = UDS | URPL;
cpuregs.ss = UDS | URPL;
cpuregs.eflags = FLAGS_RESERVED1 | FLAGS_INTERRUPT | FLAGS_ID;
cpuregs.fsbase = regs.fsbase;
cpuregs.gsbase = regs.gsbase;
cpuregs.cr3 = child_process->addrspace;
cpuregs.kernel_stack = stack_aligned + stack_aligned_size;
#elif defined(__x86_64__)
cpuregs.rip = regs.rip;
cpuregs.rsp = regs.rsp;
cpuregs.rax = regs.rax;
cpuregs.rbx = regs.rbx;
cpuregs.rcx = regs.rcx;
cpuregs.rdx = regs.rdx;
cpuregs.rdi = regs.rdi;
cpuregs.rsi = regs.rsi;
cpuregs.rbp = regs.rbp;
cpuregs.r8 = regs.r8;
cpuregs.r9 = regs.r9;
cpuregs.r10 = regs.r10;
cpuregs.r11 = regs.r11;
cpuregs.r12 = regs.r12;
cpuregs.r13 = regs.r13;
cpuregs.r14 = regs.r14;
cpuregs.r15 = regs.r15;
cpuregs.cs = UCS | URPL;
cpuregs.ds = UDS | URPL;
cpuregs.ss = UDS | URPL;
cpuregs.rflags = FLAGS_RESERVED1 | FLAGS_INTERRUPT | FLAGS_ID;
cpuregs.fsbase = regs.fsbase;
cpuregs.gsbase = regs.gsbase;
cpuregs.cr3 = child_process->addrspace;
cpuregs.kernel_stack = stack_aligned + stack_aligned_size;
#else
#warning "You need to implement initializing the registers of the new thread"
#endif
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
// If the thread could not be created, make the process commit suicide
// in a manner such that we don't wait for its zombie.
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
Thread* thread = CreateKernelThread(child_process, &cpuregs);
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
if ( !thread )
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
{
Implement thread creation in tfork(2).
2013-08-31 16:02:53 +00:00
if ( making_process )
child_process->AbortConstruction();
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return -1;
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
thread->kernelstackpos = (addr_t) newkernelstack;
Refactor kernel interrupt and thread register support.
2014-03-03 23:11:13 +00:00
thread->kernelstacksize = newkernelstacksize;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
thread->kernelstackmalloced = true;
Implement signals. Note: This is an incompatible ABI change.
2013-08-04 18:24:59 +00:00
memcpy(&thread->signal_mask, &regs.sigmask, sizeof(sigset_t));
memcpy(&thread->signal_stack, &regs.altstack, sizeof(stack_t));
Implemented the fork() system call and what it needed to work properly. This commit got completely out of control. Added the fork(), getpid(), getppid(), sleep(), usleep() system calls, and aliases in the Maxsi:: namespace. Fixed a bug where zero-byte allocation would fail. Worked on the DescriptorTable class which now works and can fork. Got rid of some massive print-registers statements and replaced them with the portable InterruptRegisters::LogRegisters() function. Removed the SysExecuteOld function and replaced it with Process::Execute(). Rewrote the boot sequence in kernel.cpp such that it now loads the system idle process 'idle' as PID 0, and the initization process 'init' as PID 1. Rewrote the SIGINT hack. Processes now maintain a family-tree structure and keep track of their threads. PIDs are now allocated using a simple hack. Virtual memory per-process can now be allocated using a simple hack. Processes can now be forked. Fixed the Process::Execute function such that it now resets the stack pointer to where the stack actually is - not just a magic value. Removed the old and ugly Process::_endcodesection hack. Rewrote the scheduler into a much cleaner and faster version. Debug code is now moved to designated functions. The noop kernel-thread has been replaced by a simple user-space infinite-loop program 'idle'. The Thread class has been seperated from the Scheduler except in Scheduler- related code. Thread::{Save,Load}Registers has been improved and has been moved to $(CPU)/thread.cpp. Threads can now be forked. A new CreateThread function creates threads properly and portably. Added a MicrosecondsSinceBoot() function. Fixed a crucial bug in MemoryManagement::Fork(). Added an 'idle' user-space program that is a noop infinite loop, which is used by the scheduler when there is nothing to do. Rewrote the 'init' program such that it now forks off a shell, instead of becoming the shell. Added the $$ (current PID) and $PPID (parent PPID) variables to the shell.
2011-09-21 18:52:29 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
StartKernelThread(thread);
Ported kernel to new syscall API and started cleaning up the old one.
2011-10-26 22:05:20 +00:00
Implement thread creation in tfork(2).
2013-08-31 16:02:53 +00:00
return child_process->pid;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
}
Multithreaded kernel and improvement of signal handling. Pardon the big ass-commit, this took months to develop and debug and the refactoring got so far that a clean merge became impossible. The good news is that this commit does quite a bit of cleaning up and generally improves the kernel quality. This makes the kernel fully pre-emptive and multithreaded. This was done by rewriting the interrupt code, the scheduler, introducing new threading primitives, and rewriting large parts of the kernel. During the past few commits the kernel has had its device drivers thread secured; this commit thread secures large parts of the core kernel. There still remains some parts of the kernel that is _not_ thread secured, but this is not a problem at this point. Each user-space thread has an associated kernel stack that it uses when it goes into kernel mode. This stack is by default 8 KiB since that value works for me and is also used by Linux. Strange things tends to happen on x86 in case of a stack overflow - there is no ideal way to catch such a situation right now. The system call conventions were changed, too. The %edx register is now used to provide the errno value of the call, instead of the kernel writing it into a registered global variable. The system call code has also been updated to better reflect the native calling conventions: not all registers have to be preserved. This makes system calls faster and simplifies the assembly. In the kernel, there is no longer the event.h header or the hacky method of 'resuming system calls' that closely resembles cooperative multitasking. If a system call wants to block, it should just block. The signal handling was also improved significantly. At this point, signals cannot interrupt kernel threads (but can always interrupt user-space threads if enabled), which introduces some problems with how a SIGINT could interrupt a blocking read, for instance. This commit introduces and uses a number of new primitives such as kthread_lock_mutex_signal() that attempts to get the lock but fails if a signal is pending. In this manner, the kernel is safer as kernel threads cannot be shut down inconveniently, but in return for complexity as blocking operations must check they if they should fail. Process exiting has also been refactored significantly. The _exit(2) system call sets the exit code and sends SIGKILL to all the threads in the process. Once all the threads have cleaned themselves up and exited, a worker thread calls the process's LastPrayer() method that unmaps memory, deletes the address space, notifies the parent, etc. This provides a very robust way to terminate processes as even half-constructed processes (during a failing fork for instance) can be gracefully terminated. I have introduced a number of kernel threads to help avoid threading problems and simplify kernel design. For instance, there is now a functional generic kernel worker thread that any kernel thread can schedule jobs for. Interrupt handlers run with interrupts off (hence they cannot call kthread_ functions as it may deadlock the system if another thread holds the lock) therefore they cannot use the standard kernel worker threads. Instead, they use a special purpose interrupt worker thread that works much like the generic one expect that interrupt handlers can safely queue work with interrupts off. Note that this also means that interrupt handlers cannot allocate memory or print to the kernel log/screen as such mechanisms uses locks. I'll introduce a lock free algorithm for such cases later on. The boot process has also changed. The original kernel init thread in kernel.cpp creates a new bootstrap thread and becomes the system idle thread. Note that pid=0 now means the kernel, as there is no longer a system idle process. The bootstrap thread launches all the kernel worker threads and then creates a new process and loads /bin/init into it and then creates a thread in pid=1, which starts the system. The bootstrap thread then quietly waits for pid=1 to exit after which it shuts down/reboots/panics the system. In general, the introduction of race conditions and dead locks have forced me to revise a lot of the design and make sure it was thread secure. Since early parts of the kernel was quite hacky, I had to refactor such code. So it seems that the risk of dead locks forces me to write better code. Note that a real preemptive multithreaded kernel simplifies the construction of blocking system calls. My hope is that this will trigger a clean up of the filesystem code that current is almost beyond repair. Almost all of the kernel was modified during this refactoring. To the extent possible, these changes have been backported to older non-multithreaded kernel, but many changes were tightly coupled and went into this commit. Of interest is the implementation of the kthread_ api based on the design of pthreads; this library allows easy synchronization mechanisms and includes C++-style scoped locks. This commit also introduces new worker threads and tested mechanisms for interrupt handlers to schedule work in a kernel worker thread. A lot of code have been rewritten from scratch and has become a lot more stable and correct. Share and enjoy!
2012-08-01 15:30:34 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
pid_t sys_getpid(void)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
return CurrentProcess()->pid;
}
Processes are listed in a global array sorted after pid.
2011-11-02 14:41:35 +00:00
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
pid_t Process::GetParentProcessId()
{
ScopedLock lock(&parentlock);
if( !parent )
Processes are listed in a global array sorted after pid.
2011-11-02 14:41:35 +00:00
return 0;
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
return parent->pid;
}
Processes are listed in a global array sorted after pid.
2011-11-02 14:41:35 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
pid_t sys_getppid(void)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
return CurrentProcess()->GetParentProcessId();
}
Processes are listed in a global array sorted after pid.
2011-11-02 14:41:35 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
pid_t sys_getpgid(pid_t pid)
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
{
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
Process* process = !pid ? CurrentProcess() : CurrentProcess()->GetPTable()->Get(pid);
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
if ( !process )
return errno = ESRCH, -1;
// Prevent the process group from changing while we read it.
ScopedLock childlock(&process->groupchildlock);
assert(process->group);
return process->group->pid;
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
int sys_setpgid(pid_t pid, pid_t pgid)
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
{
// TODO: Prevent changing the process group of zombies and other volatile
// things that are about to implode.
// TODO: Either prevent changing the process group after an exec or provide
// a version of this system call with a flags parameter that lets you
// decide if you want this behavior. This will fix a race condition
// where the shell spawns a child and both parent and child sets the
// process group, but the child sets the process group and execve's
// and the new program image exploits this 'bug' and also changes the
// process group, and then the shell gets around to change the process
// group. This probably unlikely, but correctness over all!
// Find the processes in question.
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
Process* process = !pid ? CurrentProcess() : CurrentProcess()->GetPTable()->Get(pid);
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
if ( !process )
return errno = ESRCH, -1;
Refactor process id allocation and accounting.
2014-03-19 22:34:09 +00:00
Process* group = !pgid ? process : CurrentProcess()->GetPTable()->Get(pgid);
Add setpgid(2) and getpgid(2).
2013-06-11 23:02:01 +00:00
if ( !group )
return errno = ESRCH, -1;
// Prevent the current group from being changed while we also change it
ScopedLock childlock(&process->groupchildlock);
assert(process->group);
// Exit early if this is a noop.
if ( process->group == group )
return 0;
// Prevent changing the process group of a process group leader.
if ( process->group == process )
return errno = EPERM, -1;
// Remove the process from its current process group.
kthread_mutex_lock(&process->group->groupparentlock);
if ( process->groupprev )
process->groupprev->groupnext = process->groupnext;
else
process->group->groupfirst = process->groupnext;
if ( process->groupnext )
process->groupnext->groupprev = process->groupprev;
kthread_cond_signal(&process->group->groupchildleft);
kthread_mutex_unlock(&process->group->groupparentlock);
process->group->NotifyLeftProcessGroup();
process->group = NULL;
// TODO: Somehow prevent joining a zombie group, or worse yet, one that is
// currently being deleted by its parent!
// Insert the process into its new process group.
kthread_mutex_lock(&group->groupparentlock);
process->groupprev = NULL;
process->groupnext = group->groupfirst;
if ( group->groupfirst )
group->groupfirst->groupprev = process;
group->groupfirst = process;
process->group = group;
kthread_mutex_unlock(&group->groupparentlock);
return 0;
}
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
size_t sys_getpagesize(void)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
return Page::Size();
}
Added getpagesize(2) with the new return type size_t. This function is non-standard and size_t is better than int here.
2012-01-08 13:58:57 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
mode_t sys_umask(mode_t newmask)
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
{
Process* process = CurrentProcess();
ScopedLock lock(&process->idlock);
mode_t oldmask = process->umask;
process->umask = newmask & 0666;
return oldmask;
}
Add umask(2).
2013-05-16 20:03:15 +00:00
Initialize system call table at compile time.
2014-10-16 22:04:47 +00:00
mode_t sys_getumask(void)
Add getumask(2).
2014-01-19 21:45:49 +00:00
{
Process* process = CurrentProcess();
ScopedLock lock(&process->idlock);
return process->umask;
}
Update sortix/process.cpp coding style.
2013-06-11 23:01:49 +00:00
} // namespace Sortix