4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/key.h>
26 #include <linux/binfmts.h>
27 #include <linux/mman.h>
29 #include <linux/nsproxy.h>
30 #include <linux/capability.h>
31 #include <linux/cpu.h>
32 #include <linux/cpuset.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/task_io_accounting_ops.h>
39 #include <linux/rcupdate.h>
40 #include <linux/ptrace.h>
41 #include <linux/mount.h>
42 #include <linux/audit.h>
43 #include <linux/profile.h>
44 #include <linux/rmap.h>
45 #include <linux/acct.h>
46 #include <linux/tsacct_kern.h>
47 #include <linux/cn_proc.h>
48 #include <linux/freezer.h>
49 #include <linux/delayacct.h>
50 #include <linux/taskstats_kern.h>
51 #include <linux/random.h>
52 #include <linux/tty.h>
54 #include <asm/pgtable.h>
55 #include <asm/pgalloc.h>
56 #include <asm/uaccess.h>
57 #include <asm/mmu_context.h>
58 #include <asm/cacheflush.h>
59 #include <asm/tlbflush.h>
62 * Protected counters by write_lock_irq(&tasklist_lock)
64 unsigned long total_forks; /* Handle normal Linux uptimes. */
65 int nr_threads; /* The idle threads do not count.. */
67 int max_threads; /* tunable limit on nr_threads */
69 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
71 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
73 int nr_processes(void)
78 for_each_online_cpu(cpu)
79 total += per_cpu(process_counts, cpu);
84 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
85 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
86 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
87 static struct kmem_cache *task_struct_cachep;
90 /* SLAB cache for signal_struct structures (tsk->signal) */
91 static struct kmem_cache *signal_cachep;
93 /* SLAB cache for sighand_struct structures (tsk->sighand) */
94 struct kmem_cache *sighand_cachep;
96 /* SLAB cache for files_struct structures (tsk->files) */
97 struct kmem_cache *files_cachep;
99 /* SLAB cache for fs_struct structures (tsk->fs) */
100 struct kmem_cache *fs_cachep;
102 /* SLAB cache for vm_area_struct structures */
103 struct kmem_cache *vm_area_cachep;
105 /* SLAB cache for mm_struct structures (tsk->mm) */
106 static struct kmem_cache *mm_cachep;
108 void free_task(struct task_struct *tsk)
110 free_thread_info(tsk->stack);
111 rt_mutex_debug_task_free(tsk);
112 free_task_struct(tsk);
114 EXPORT_SYMBOL(free_task);
116 void __put_task_struct(struct task_struct *tsk)
118 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
119 WARN_ON(atomic_read(&tsk->usage));
120 WARN_ON(tsk == current);
122 security_task_free(tsk);
124 put_group_info(tsk->group_info);
125 delayacct_tsk_free(tsk);
127 if (!profile_handoff_task(tsk))
130 EXPORT_SYMBOL_GPL(__put_task_struct);
132 void __init fork_init(unsigned long mempages)
134 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
135 #ifndef ARCH_MIN_TASKALIGN
136 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
138 /* create a slab on which task_structs can be allocated */
140 kmem_cache_create("task_struct", sizeof(struct task_struct),
141 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
145 * The default maximum number of threads is set to a safe
146 * value: the thread structures can take up at most half
149 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
152 * we need to allow at least 20 threads to boot a system
157 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
158 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
159 init_task.signal->rlim[RLIMIT_SIGPENDING] =
160 init_task.signal->rlim[RLIMIT_NPROC];
163 static struct task_struct *dup_task_struct(struct task_struct *orig)
165 struct task_struct *tsk;
166 struct thread_info *ti;
168 prepare_to_copy(orig);
170 tsk = alloc_task_struct();
174 ti = alloc_thread_info(tsk);
176 free_task_struct(tsk);
182 setup_thread_stack(tsk, orig);
184 #ifdef CONFIG_CC_STACKPROTECTOR
185 tsk->stack_canary = get_random_int();
188 /* One for us, one for whoever does the "release_task()" (usually parent) */
189 atomic_set(&tsk->usage,2);
190 atomic_set(&tsk->fs_excl, 0);
191 #ifdef CONFIG_BLK_DEV_IO_TRACE
194 tsk->splice_pipe = NULL;
199 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
201 struct vm_area_struct *mpnt, *tmp, **pprev;
202 struct rb_node **rb_link, *rb_parent;
204 unsigned long charge;
205 struct mempolicy *pol;
207 down_write(&oldmm->mmap_sem);
208 flush_cache_dup_mm(oldmm);
210 * Not linked in yet - no deadlock potential:
212 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
216 mm->mmap_cache = NULL;
217 mm->free_area_cache = oldmm->mmap_base;
218 mm->cached_hole_size = ~0UL;
220 cpus_clear(mm->cpu_vm_mask);
222 rb_link = &mm->mm_rb.rb_node;
226 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
229 if (mpnt->vm_flags & VM_DONTCOPY) {
230 long pages = vma_pages(mpnt);
231 mm->total_vm -= pages;
232 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
237 if (mpnt->vm_flags & VM_ACCOUNT) {
238 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
239 if (security_vm_enough_memory(len))
243 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
247 pol = mpol_copy(vma_policy(mpnt));
248 retval = PTR_ERR(pol);
250 goto fail_nomem_policy;
251 vma_set_policy(tmp, pol);
252 tmp->vm_flags &= ~VM_LOCKED;
258 struct inode *inode = file->f_path.dentry->d_inode;
260 if (tmp->vm_flags & VM_DENYWRITE)
261 atomic_dec(&inode->i_writecount);
263 /* insert tmp into the share list, just after mpnt */
264 spin_lock(&file->f_mapping->i_mmap_lock);
265 tmp->vm_truncate_count = mpnt->vm_truncate_count;
266 flush_dcache_mmap_lock(file->f_mapping);
267 vma_prio_tree_add(tmp, mpnt);
268 flush_dcache_mmap_unlock(file->f_mapping);
269 spin_unlock(&file->f_mapping->i_mmap_lock);
273 * Link in the new vma and copy the page table entries.
276 pprev = &tmp->vm_next;
278 __vma_link_rb(mm, tmp, rb_link, rb_parent);
279 rb_link = &tmp->vm_rb.rb_right;
280 rb_parent = &tmp->vm_rb;
283 retval = copy_page_range(mm, oldmm, mpnt);
285 if (tmp->vm_ops && tmp->vm_ops->open)
286 tmp->vm_ops->open(tmp);
291 /* a new mm has just been created */
292 arch_dup_mmap(oldmm, mm);
295 up_write(&mm->mmap_sem);
297 up_write(&oldmm->mmap_sem);
300 kmem_cache_free(vm_area_cachep, tmp);
303 vm_unacct_memory(charge);
307 static inline int mm_alloc_pgd(struct mm_struct * mm)
309 mm->pgd = pgd_alloc(mm);
310 if (unlikely(!mm->pgd))
315 static inline void mm_free_pgd(struct mm_struct * mm)
320 #define dup_mmap(mm, oldmm) (0)
321 #define mm_alloc_pgd(mm) (0)
322 #define mm_free_pgd(mm)
323 #endif /* CONFIG_MMU */
325 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
327 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
328 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
330 #include <linux/init_task.h>
332 static struct mm_struct * mm_init(struct mm_struct * mm)
334 atomic_set(&mm->mm_users, 1);
335 atomic_set(&mm->mm_count, 1);
336 init_rwsem(&mm->mmap_sem);
337 INIT_LIST_HEAD(&mm->mmlist);
338 mm->flags = (current->mm) ? current->mm->flags
339 : MMF_DUMP_FILTER_DEFAULT;
340 mm->core_waiters = 0;
342 set_mm_counter(mm, file_rss, 0);
343 set_mm_counter(mm, anon_rss, 0);
344 spin_lock_init(&mm->page_table_lock);
345 rwlock_init(&mm->ioctx_list_lock);
346 mm->ioctx_list = NULL;
347 mm->free_area_cache = TASK_UNMAPPED_BASE;
348 mm->cached_hole_size = ~0UL;
350 if (likely(!mm_alloc_pgd(mm))) {
359 * Allocate and initialize an mm_struct.
361 struct mm_struct * mm_alloc(void)
363 struct mm_struct * mm;
367 memset(mm, 0, sizeof(*mm));
374 * Called when the last reference to the mm
375 * is dropped: either by a lazy thread or by
376 * mmput. Free the page directory and the mm.
378 void fastcall __mmdrop(struct mm_struct *mm)
380 BUG_ON(mm == &init_mm);
387 * Decrement the use count and release all resources for an mm.
389 void mmput(struct mm_struct *mm)
393 if (atomic_dec_and_test(&mm->mm_users)) {
396 if (!list_empty(&mm->mmlist)) {
397 spin_lock(&mmlist_lock);
398 list_del(&mm->mmlist);
399 spin_unlock(&mmlist_lock);
405 EXPORT_SYMBOL_GPL(mmput);
408 * get_task_mm - acquire a reference to the task's mm
410 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
411 * this kernel workthread has transiently adopted a user mm with use_mm,
412 * to do its AIO) is not set and if so returns a reference to it, after
413 * bumping up the use count. User must release the mm via mmput()
414 * after use. Typically used by /proc and ptrace.
416 struct mm_struct *get_task_mm(struct task_struct *task)
418 struct mm_struct *mm;
423 if (task->flags & PF_BORROWED_MM)
426 atomic_inc(&mm->mm_users);
431 EXPORT_SYMBOL_GPL(get_task_mm);
433 /* Please note the differences between mmput and mm_release.
434 * mmput is called whenever we stop holding onto a mm_struct,
435 * error success whatever.
437 * mm_release is called after a mm_struct has been removed
438 * from the current process.
440 * This difference is important for error handling, when we
441 * only half set up a mm_struct for a new process and need to restore
442 * the old one. Because we mmput the new mm_struct before
443 * restoring the old one. . .
444 * Eric Biederman 10 January 1998
446 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
448 struct completion *vfork_done = tsk->vfork_done;
450 /* Get rid of any cached register state */
451 deactivate_mm(tsk, mm);
453 /* notify parent sleeping on vfork() */
455 tsk->vfork_done = NULL;
456 complete(vfork_done);
460 * If we're exiting normally, clear a user-space tid field if
461 * requested. We leave this alone when dying by signal, to leave
462 * the value intact in a core dump, and to save the unnecessary
463 * trouble otherwise. Userland only wants this done for a sys_exit.
465 if (tsk->clear_child_tid
466 && !(tsk->flags & PF_SIGNALED)
467 && atomic_read(&mm->mm_users) > 1) {
468 u32 __user * tidptr = tsk->clear_child_tid;
469 tsk->clear_child_tid = NULL;
472 * We don't check the error code - if userspace has
473 * not set up a proper pointer then tough luck.
476 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
481 * Allocate a new mm structure and copy contents from the
482 * mm structure of the passed in task structure.
484 static struct mm_struct *dup_mm(struct task_struct *tsk)
486 struct mm_struct *mm, *oldmm = current->mm;
496 memcpy(mm, oldmm, sizeof(*mm));
498 /* Initializing for Swap token stuff */
499 mm->token_priority = 0;
500 mm->last_interval = 0;
505 if (init_new_context(tsk, mm))
508 err = dup_mmap(mm, oldmm);
512 mm->hiwater_rss = get_mm_rss(mm);
513 mm->hiwater_vm = mm->total_vm;
525 * If init_new_context() failed, we cannot use mmput() to free the mm
526 * because it calls destroy_context()
533 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
535 struct mm_struct * mm, *oldmm;
538 tsk->min_flt = tsk->maj_flt = 0;
539 tsk->nvcsw = tsk->nivcsw = 0;
542 tsk->active_mm = NULL;
545 * Are we cloning a kernel thread?
547 * We need to steal a active VM for that..
553 if (clone_flags & CLONE_VM) {
554 atomic_inc(&oldmm->mm_users);
565 /* Initializing for Swap token stuff */
566 mm->token_priority = 0;
567 mm->last_interval = 0;
577 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
579 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
580 /* We don't need to lock fs - think why ;-) */
582 atomic_set(&fs->count, 1);
583 rwlock_init(&fs->lock);
584 fs->umask = old->umask;
585 read_lock(&old->lock);
586 fs->rootmnt = mntget(old->rootmnt);
587 fs->root = dget(old->root);
588 fs->pwdmnt = mntget(old->pwdmnt);
589 fs->pwd = dget(old->pwd);
591 fs->altrootmnt = mntget(old->altrootmnt);
592 fs->altroot = dget(old->altroot);
594 fs->altrootmnt = NULL;
597 read_unlock(&old->lock);
602 struct fs_struct *copy_fs_struct(struct fs_struct *old)
604 return __copy_fs_struct(old);
607 EXPORT_SYMBOL_GPL(copy_fs_struct);
609 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
611 if (clone_flags & CLONE_FS) {
612 atomic_inc(¤t->fs->count);
615 tsk->fs = __copy_fs_struct(current->fs);
621 static int count_open_files(struct fdtable *fdt)
623 int size = fdt->max_fds;
626 /* Find the last open fd */
627 for (i = size/(8*sizeof(long)); i > 0; ) {
628 if (fdt->open_fds->fds_bits[--i])
631 i = (i+1) * 8 * sizeof(long);
635 static struct files_struct *alloc_files(void)
637 struct files_struct *newf;
640 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
644 atomic_set(&newf->count, 1);
646 spin_lock_init(&newf->file_lock);
649 fdt->max_fds = NR_OPEN_DEFAULT;
650 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
651 fdt->open_fds = (fd_set *)&newf->open_fds_init;
652 fdt->fd = &newf->fd_array[0];
653 INIT_RCU_HEAD(&fdt->rcu);
655 rcu_assign_pointer(newf->fdt, fdt);
661 * Allocate a new files structure and copy contents from the
662 * passed in files structure.
663 * errorp will be valid only when the returned files_struct is NULL.
665 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
667 struct files_struct *newf;
668 struct file **old_fds, **new_fds;
669 int open_files, size, i;
670 struct fdtable *old_fdt, *new_fdt;
673 newf = alloc_files();
677 spin_lock(&oldf->file_lock);
678 old_fdt = files_fdtable(oldf);
679 new_fdt = files_fdtable(newf);
680 open_files = count_open_files(old_fdt);
683 * Check whether we need to allocate a larger fd array and fd set.
684 * Note: we're not a clone task, so the open count won't change.
686 if (open_files > new_fdt->max_fds) {
687 new_fdt->max_fds = 0;
688 spin_unlock(&oldf->file_lock);
689 spin_lock(&newf->file_lock);
690 *errorp = expand_files(newf, open_files-1);
691 spin_unlock(&newf->file_lock);
694 new_fdt = files_fdtable(newf);
696 * Reacquire the oldf lock and a pointer to its fd table
697 * who knows it may have a new bigger fd table. We need
698 * the latest pointer.
700 spin_lock(&oldf->file_lock);
701 old_fdt = files_fdtable(oldf);
704 old_fds = old_fdt->fd;
705 new_fds = new_fdt->fd;
707 memcpy(new_fdt->open_fds->fds_bits,
708 old_fdt->open_fds->fds_bits, open_files/8);
709 memcpy(new_fdt->close_on_exec->fds_bits,
710 old_fdt->close_on_exec->fds_bits, open_files/8);
712 for (i = open_files; i != 0; i--) {
713 struct file *f = *old_fds++;
718 * The fd may be claimed in the fd bitmap but not yet
719 * instantiated in the files array if a sibling thread
720 * is partway through open(). So make sure that this
721 * fd is available to the new process.
723 FD_CLR(open_files - i, new_fdt->open_fds);
725 rcu_assign_pointer(*new_fds++, f);
727 spin_unlock(&oldf->file_lock);
729 /* compute the remainder to be cleared */
730 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
732 /* This is long word aligned thus could use a optimized version */
733 memset(new_fds, 0, size);
735 if (new_fdt->max_fds > open_files) {
736 int left = (new_fdt->max_fds-open_files)/8;
737 int start = open_files / (8 * sizeof(unsigned long));
739 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
740 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
746 kmem_cache_free(files_cachep, newf);
751 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
753 struct files_struct *oldf, *newf;
757 * A background process may not have any files ...
759 oldf = current->files;
763 if (clone_flags & CLONE_FILES) {
764 atomic_inc(&oldf->count);
769 * Note: we may be using current for both targets (See exec.c)
770 * This works because we cache current->files (old) as oldf. Don't
774 newf = dup_fd(oldf, &error);
785 * Helper to unshare the files of the current task.
786 * We don't want to expose copy_files internals to
787 * the exec layer of the kernel.
790 int unshare_files(void)
792 struct files_struct *files = current->files;
797 /* This can race but the race causes us to copy when we don't
798 need to and drop the copy */
799 if(atomic_read(&files->count) == 1)
801 atomic_inc(&files->count);
804 rc = copy_files(0, current);
806 current->files = files;
810 EXPORT_SYMBOL(unshare_files);
812 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
814 struct sighand_struct *sig;
816 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
817 atomic_inc(¤t->sighand->count);
820 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
821 rcu_assign_pointer(tsk->sighand, sig);
824 atomic_set(&sig->count, 1);
825 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
829 void __cleanup_sighand(struct sighand_struct *sighand)
831 if (atomic_dec_and_test(&sighand->count))
832 kmem_cache_free(sighand_cachep, sighand);
835 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
837 struct signal_struct *sig;
840 if (clone_flags & CLONE_THREAD) {
841 atomic_inc(¤t->signal->count);
842 atomic_inc(¤t->signal->live);
845 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
850 ret = copy_thread_group_keys(tsk);
852 kmem_cache_free(signal_cachep, sig);
856 atomic_set(&sig->count, 1);
857 atomic_set(&sig->live, 1);
858 init_waitqueue_head(&sig->wait_chldexit);
860 sig->group_exit_code = 0;
861 sig->group_exit_task = NULL;
862 sig->group_stop_count = 0;
863 sig->curr_target = NULL;
864 init_sigpending(&sig->shared_pending);
865 INIT_LIST_HEAD(&sig->posix_timers);
867 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
868 sig->it_real_incr.tv64 = 0;
869 sig->real_timer.function = it_real_fn;
872 sig->it_virt_expires = cputime_zero;
873 sig->it_virt_incr = cputime_zero;
874 sig->it_prof_expires = cputime_zero;
875 sig->it_prof_incr = cputime_zero;
877 sig->leader = 0; /* session leadership doesn't inherit */
878 sig->tty_old_pgrp = NULL;
880 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
881 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
882 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
883 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
884 sig->sum_sched_runtime = 0;
885 INIT_LIST_HEAD(&sig->cpu_timers[0]);
886 INIT_LIST_HEAD(&sig->cpu_timers[1]);
887 INIT_LIST_HEAD(&sig->cpu_timers[2]);
888 taskstats_tgid_init(sig);
890 task_lock(current->group_leader);
891 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
892 task_unlock(current->group_leader);
894 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
896 * New sole thread in the process gets an expiry time
897 * of the whole CPU time limit.
899 tsk->it_prof_expires =
900 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
902 acct_init_pacct(&sig->pacct);
909 void __cleanup_signal(struct signal_struct *sig)
911 exit_thread_group_keys(sig);
912 kmem_cache_free(signal_cachep, sig);
915 static inline void cleanup_signal(struct task_struct *tsk)
917 struct signal_struct *sig = tsk->signal;
919 atomic_dec(&sig->live);
921 if (atomic_dec_and_test(&sig->count))
922 __cleanup_signal(sig);
925 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
927 unsigned long new_flags = p->flags;
929 new_flags &= ~PF_SUPERPRIV;
930 new_flags |= PF_FORKNOEXEC;
931 if (!(clone_flags & CLONE_PTRACE))
933 p->flags = new_flags;
936 asmlinkage long sys_set_tid_address(int __user *tidptr)
938 current->clear_child_tid = tidptr;
943 static inline void rt_mutex_init_task(struct task_struct *p)
945 spin_lock_init(&p->pi_lock);
946 #ifdef CONFIG_RT_MUTEXES
947 plist_head_init(&p->pi_waiters, &p->pi_lock);
948 p->pi_blocked_on = NULL;
953 * This creates a new process as a copy of the old one,
954 * but does not actually start it yet.
956 * It copies the registers, and all the appropriate
957 * parts of the process environment (as per the clone
958 * flags). The actual kick-off is left to the caller.
960 static struct task_struct *copy_process(unsigned long clone_flags,
961 unsigned long stack_start,
962 struct pt_regs *regs,
963 unsigned long stack_size,
964 int __user *parent_tidptr,
965 int __user *child_tidptr,
969 struct task_struct *p = NULL;
971 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
972 return ERR_PTR(-EINVAL);
975 * Thread groups must share signals as well, and detached threads
976 * can only be started up within the thread group.
978 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
979 return ERR_PTR(-EINVAL);
982 * Shared signal handlers imply shared VM. By way of the above,
983 * thread groups also imply shared VM. Blocking this case allows
984 * for various simplifications in other code.
986 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
987 return ERR_PTR(-EINVAL);
989 retval = security_task_create(clone_flags);
994 p = dup_task_struct(current);
998 rt_mutex_init_task(p);
1000 #ifdef CONFIG_TRACE_IRQFLAGS
1001 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1002 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1005 if (atomic_read(&p->user->processes) >=
1006 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1007 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1008 p->user != current->nsproxy->user_ns->root_user)
1012 atomic_inc(&p->user->__count);
1013 atomic_inc(&p->user->processes);
1014 get_group_info(p->group_info);
1017 * If multiple threads are within copy_process(), then this check
1018 * triggers too late. This doesn't hurt, the check is only there
1019 * to stop root fork bombs.
1021 if (nr_threads >= max_threads)
1022 goto bad_fork_cleanup_count;
1024 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1025 goto bad_fork_cleanup_count;
1027 if (p->binfmt && !try_module_get(p->binfmt->module))
1028 goto bad_fork_cleanup_put_domain;
1031 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1032 copy_flags(clone_flags, p);
1033 p->pid = pid_nr(pid);
1035 if (clone_flags & CLONE_PARENT_SETTID)
1036 if (put_user(p->pid, parent_tidptr))
1037 goto bad_fork_cleanup_delays_binfmt;
1039 INIT_LIST_HEAD(&p->children);
1040 INIT_LIST_HEAD(&p->sibling);
1041 p->vfork_done = NULL;
1042 spin_lock_init(&p->alloc_lock);
1044 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1045 init_sigpending(&p->pending);
1047 p->utime = cputime_zero;
1048 p->stime = cputime_zero;
1050 #ifdef CONFIG_TASK_XACCT
1051 p->rchar = 0; /* I/O counter: bytes read */
1052 p->wchar = 0; /* I/O counter: bytes written */
1053 p->syscr = 0; /* I/O counter: read syscalls */
1054 p->syscw = 0; /* I/O counter: write syscalls */
1056 task_io_accounting_init(p);
1057 acct_clear_integrals(p);
1059 p->it_virt_expires = cputime_zero;
1060 p->it_prof_expires = cputime_zero;
1061 p->it_sched_expires = 0;
1062 INIT_LIST_HEAD(&p->cpu_timers[0]);
1063 INIT_LIST_HEAD(&p->cpu_timers[1]);
1064 INIT_LIST_HEAD(&p->cpu_timers[2]);
1066 p->lock_depth = -1; /* -1 = no lock */
1067 do_posix_clock_monotonic_gettime(&p->start_time);
1068 p->real_start_time = p->start_time;
1069 monotonic_to_bootbased(&p->real_start_time);
1071 p->io_context = NULL;
1073 p->audit_context = NULL;
1076 p->mempolicy = mpol_copy(p->mempolicy);
1077 if (IS_ERR(p->mempolicy)) {
1078 retval = PTR_ERR(p->mempolicy);
1079 p->mempolicy = NULL;
1080 goto bad_fork_cleanup_cpuset;
1082 mpol_fix_fork_child_flag(p);
1084 #ifdef CONFIG_TRACE_IRQFLAGS
1086 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1087 p->hardirqs_enabled = 1;
1089 p->hardirqs_enabled = 0;
1091 p->hardirq_enable_ip = 0;
1092 p->hardirq_enable_event = 0;
1093 p->hardirq_disable_ip = _THIS_IP_;
1094 p->hardirq_disable_event = 0;
1095 p->softirqs_enabled = 1;
1096 p->softirq_enable_ip = _THIS_IP_;
1097 p->softirq_enable_event = 0;
1098 p->softirq_disable_ip = 0;
1099 p->softirq_disable_event = 0;
1100 p->hardirq_context = 0;
1101 p->softirq_context = 0;
1103 #ifdef CONFIG_LOCKDEP
1104 p->lockdep_depth = 0; /* no locks held yet */
1105 p->curr_chain_key = 0;
1106 p->lockdep_recursion = 0;
1109 #ifdef CONFIG_DEBUG_MUTEXES
1110 p->blocked_on = NULL; /* not blocked yet */
1114 if (clone_flags & CLONE_THREAD)
1115 p->tgid = current->tgid;
1117 if ((retval = security_task_alloc(p)))
1118 goto bad_fork_cleanup_policy;
1119 if ((retval = audit_alloc(p)))
1120 goto bad_fork_cleanup_security;
1121 /* copy all the process information */
1122 if ((retval = copy_semundo(clone_flags, p)))
1123 goto bad_fork_cleanup_audit;
1124 if ((retval = copy_files(clone_flags, p)))
1125 goto bad_fork_cleanup_semundo;
1126 if ((retval = copy_fs(clone_flags, p)))
1127 goto bad_fork_cleanup_files;
1128 if ((retval = copy_sighand(clone_flags, p)))
1129 goto bad_fork_cleanup_fs;
1130 if ((retval = copy_signal(clone_flags, p)))
1131 goto bad_fork_cleanup_sighand;
1132 if ((retval = copy_mm(clone_flags, p)))
1133 goto bad_fork_cleanup_signal;
1134 if ((retval = copy_keys(clone_flags, p)))
1135 goto bad_fork_cleanup_mm;
1136 if ((retval = copy_namespaces(clone_flags, p)))
1137 goto bad_fork_cleanup_keys;
1138 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1140 goto bad_fork_cleanup_namespaces;
1142 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1144 * Clear TID on mm_release()?
1146 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1147 p->robust_list = NULL;
1148 #ifdef CONFIG_COMPAT
1149 p->compat_robust_list = NULL;
1151 INIT_LIST_HEAD(&p->pi_state_list);
1152 p->pi_state_cache = NULL;
1155 * sigaltstack should be cleared when sharing the same VM
1157 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1158 p->sas_ss_sp = p->sas_ss_size = 0;
1161 * Syscall tracing should be turned off in the child regardless
1164 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1165 #ifdef TIF_SYSCALL_EMU
1166 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1169 /* Our parent execution domain becomes current domain
1170 These must match for thread signalling to apply */
1171 p->parent_exec_id = p->self_exec_id;
1173 /* ok, now we should be set up.. */
1174 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1175 p->pdeath_signal = 0;
1179 * Ok, make it visible to the rest of the system.
1180 * We dont wake it up yet.
1182 p->group_leader = p;
1183 INIT_LIST_HEAD(&p->thread_group);
1184 INIT_LIST_HEAD(&p->ptrace_children);
1185 INIT_LIST_HEAD(&p->ptrace_list);
1187 /* Perform scheduler related setup. Assign this task to a CPU. */
1188 sched_fork(p, clone_flags);
1190 /* Need tasklist lock for parent etc handling! */
1191 write_lock_irq(&tasklist_lock);
1193 /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1194 p->ioprio = current->ioprio;
1197 * The task hasn't been attached yet, so its cpus_allowed mask will
1198 * not be changed, nor will its assigned CPU.
1200 * The cpus_allowed mask of the parent may have changed after it was
1201 * copied first time - so re-copy it here, then check the child's CPU
1202 * to ensure it is on a valid CPU (and if not, just force it back to
1203 * parent's CPU). This avoids alot of nasty races.
1205 p->cpus_allowed = current->cpus_allowed;
1206 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1207 !cpu_online(task_cpu(p))))
1208 set_task_cpu(p, smp_processor_id());
1210 /* CLONE_PARENT re-uses the old parent */
1211 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1212 p->real_parent = current->real_parent;
1214 p->real_parent = current;
1215 p->parent = p->real_parent;
1217 spin_lock(¤t->sighand->siglock);
1220 * Process group and session signals need to be delivered to just the
1221 * parent before the fork or both the parent and the child after the
1222 * fork. Restart if a signal comes in before we add the new process to
1223 * it's process group.
1224 * A fatal signal pending means that current will exit, so the new
1225 * thread can't slip out of an OOM kill (or normal SIGKILL).
1227 recalc_sigpending();
1228 if (signal_pending(current)) {
1229 spin_unlock(¤t->sighand->siglock);
1230 write_unlock_irq(&tasklist_lock);
1231 retval = -ERESTARTNOINTR;
1232 goto bad_fork_cleanup_namespaces;
1235 if (clone_flags & CLONE_THREAD) {
1236 p->group_leader = current->group_leader;
1237 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1239 if (!cputime_eq(current->signal->it_virt_expires,
1241 !cputime_eq(current->signal->it_prof_expires,
1243 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1244 !list_empty(¤t->signal->cpu_timers[0]) ||
1245 !list_empty(¤t->signal->cpu_timers[1]) ||
1246 !list_empty(¤t->signal->cpu_timers[2])) {
1248 * Have child wake up on its first tick to check
1249 * for process CPU timers.
1251 p->it_prof_expires = jiffies_to_cputime(1);
1255 if (likely(p->pid)) {
1257 if (unlikely(p->ptrace & PT_PTRACED))
1258 __ptrace_link(p, current->parent);
1260 if (thread_group_leader(p)) {
1261 p->signal->tty = current->signal->tty;
1262 p->signal->pgrp = process_group(current);
1263 set_signal_session(p->signal, process_session(current));
1264 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1265 attach_pid(p, PIDTYPE_SID, task_session(current));
1267 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1268 __get_cpu_var(process_counts)++;
1270 attach_pid(p, PIDTYPE_PID, pid);
1275 spin_unlock(¤t->sighand->siglock);
1276 write_unlock_irq(&tasklist_lock);
1277 proc_fork_connector(p);
1280 bad_fork_cleanup_namespaces:
1281 exit_task_namespaces(p);
1282 bad_fork_cleanup_keys:
1284 bad_fork_cleanup_mm:
1287 bad_fork_cleanup_signal:
1289 bad_fork_cleanup_sighand:
1290 __cleanup_sighand(p->sighand);
1291 bad_fork_cleanup_fs:
1292 exit_fs(p); /* blocking */
1293 bad_fork_cleanup_files:
1294 exit_files(p); /* blocking */
1295 bad_fork_cleanup_semundo:
1297 bad_fork_cleanup_audit:
1299 bad_fork_cleanup_security:
1300 security_task_free(p);
1301 bad_fork_cleanup_policy:
1303 mpol_free(p->mempolicy);
1304 bad_fork_cleanup_cpuset:
1307 bad_fork_cleanup_delays_binfmt:
1308 delayacct_tsk_free(p);
1310 module_put(p->binfmt->module);
1311 bad_fork_cleanup_put_domain:
1312 module_put(task_thread_info(p)->exec_domain->module);
1313 bad_fork_cleanup_count:
1314 put_group_info(p->group_info);
1315 atomic_dec(&p->user->processes);
1320 return ERR_PTR(retval);
1323 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1325 memset(regs, 0, sizeof(struct pt_regs));
1329 struct task_struct * __cpuinit fork_idle(int cpu)
1331 struct task_struct *task;
1332 struct pt_regs regs;
1334 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL,
1337 init_idle(task, cpu);
1342 static inline int fork_traceflag (unsigned clone_flags)
1344 if (clone_flags & CLONE_UNTRACED)
1346 else if (clone_flags & CLONE_VFORK) {
1347 if (current->ptrace & PT_TRACE_VFORK)
1348 return PTRACE_EVENT_VFORK;
1349 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1350 if (current->ptrace & PT_TRACE_CLONE)
1351 return PTRACE_EVENT_CLONE;
1352 } else if (current->ptrace & PT_TRACE_FORK)
1353 return PTRACE_EVENT_FORK;
1359 * Ok, this is the main fork-routine.
1361 * It copies the process, and if successful kick-starts
1362 * it and waits for it to finish using the VM if required.
1364 long do_fork(unsigned long clone_flags,
1365 unsigned long stack_start,
1366 struct pt_regs *regs,
1367 unsigned long stack_size,
1368 int __user *parent_tidptr,
1369 int __user *child_tidptr)
1371 struct task_struct *p;
1373 struct pid *pid = alloc_pid();
1379 if (unlikely(current->ptrace)) {
1380 trace = fork_traceflag (clone_flags);
1382 clone_flags |= CLONE_PTRACE;
1385 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1387 * Do this prior waking up the new thread - the thread pointer
1388 * might get invalid after that point, if the thread exits quickly.
1391 struct completion vfork;
1393 if (clone_flags & CLONE_VFORK) {
1394 p->vfork_done = &vfork;
1395 init_completion(&vfork);
1398 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1400 * We'll start up with an immediate SIGSTOP.
1402 sigaddset(&p->pending.signal, SIGSTOP);
1403 set_tsk_thread_flag(p, TIF_SIGPENDING);
1406 if (!(clone_flags & CLONE_STOPPED))
1407 wake_up_new_task(p, clone_flags);
1409 p->state = TASK_STOPPED;
1411 if (unlikely (trace)) {
1412 current->ptrace_message = nr;
1413 ptrace_notify ((trace << 8) | SIGTRAP);
1416 if (clone_flags & CLONE_VFORK) {
1417 freezer_do_not_count();
1418 wait_for_completion(&vfork);
1420 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1421 current->ptrace_message = nr;
1422 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1432 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1433 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1436 static void sighand_ctor(void *data, struct kmem_cache *cachep,
1437 unsigned long flags)
1439 struct sighand_struct *sighand = data;
1441 spin_lock_init(&sighand->siglock);
1442 INIT_LIST_HEAD(&sighand->signalfd_list);
1445 void __init proc_caches_init(void)
1447 sighand_cachep = kmem_cache_create("sighand_cache",
1448 sizeof(struct sighand_struct), 0,
1449 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1450 sighand_ctor, NULL);
1451 signal_cachep = kmem_cache_create("signal_cache",
1452 sizeof(struct signal_struct), 0,
1453 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1454 files_cachep = kmem_cache_create("files_cache",
1455 sizeof(struct files_struct), 0,
1456 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1457 fs_cachep = kmem_cache_create("fs_cache",
1458 sizeof(struct fs_struct), 0,
1459 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1460 vm_area_cachep = kmem_cache_create("vm_area_struct",
1461 sizeof(struct vm_area_struct), 0,
1462 SLAB_PANIC, NULL, NULL);
1463 mm_cachep = kmem_cache_create("mm_struct",
1464 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1465 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1469 * Check constraints on flags passed to the unshare system call and
1470 * force unsharing of additional process context as appropriate.
1472 static inline void check_unshare_flags(unsigned long *flags_ptr)
1475 * If unsharing a thread from a thread group, must also
1478 if (*flags_ptr & CLONE_THREAD)
1479 *flags_ptr |= CLONE_VM;
1482 * If unsharing vm, must also unshare signal handlers.
1484 if (*flags_ptr & CLONE_VM)
1485 *flags_ptr |= CLONE_SIGHAND;
1488 * If unsharing signal handlers and the task was created
1489 * using CLONE_THREAD, then must unshare the thread
1491 if ((*flags_ptr & CLONE_SIGHAND) &&
1492 (atomic_read(¤t->signal->count) > 1))
1493 *flags_ptr |= CLONE_THREAD;
1496 * If unsharing namespace, must also unshare filesystem information.
1498 if (*flags_ptr & CLONE_NEWNS)
1499 *flags_ptr |= CLONE_FS;
1503 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1505 static int unshare_thread(unsigned long unshare_flags)
1507 if (unshare_flags & CLONE_THREAD)
1514 * Unshare the filesystem structure if it is being shared
1516 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1518 struct fs_struct *fs = current->fs;
1520 if ((unshare_flags & CLONE_FS) &&
1521 (fs && atomic_read(&fs->count) > 1)) {
1522 *new_fsp = __copy_fs_struct(current->fs);
1531 * Unsharing of sighand is not supported yet
1533 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1535 struct sighand_struct *sigh = current->sighand;
1537 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1544 * Unshare vm if it is being shared
1546 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1548 struct mm_struct *mm = current->mm;
1550 if ((unshare_flags & CLONE_VM) &&
1551 (mm && atomic_read(&mm->mm_users) > 1)) {
1559 * Unshare file descriptor table if it is being shared
1561 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1563 struct files_struct *fd = current->files;
1566 if ((unshare_flags & CLONE_FILES) &&
1567 (fd && atomic_read(&fd->count) > 1)) {
1568 *new_fdp = dup_fd(fd, &error);
1577 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1580 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1582 if (unshare_flags & CLONE_SYSVSEM)
1589 * unshare allows a process to 'unshare' part of the process
1590 * context which was originally shared using clone. copy_*
1591 * functions used by do_fork() cannot be used here directly
1592 * because they modify an inactive task_struct that is being
1593 * constructed. Here we are modifying the current, active,
1596 asmlinkage long sys_unshare(unsigned long unshare_flags)
1599 struct fs_struct *fs, *new_fs = NULL;
1600 struct sighand_struct *new_sigh = NULL;
1601 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1602 struct files_struct *fd, *new_fd = NULL;
1603 struct sem_undo_list *new_ulist = NULL;
1604 struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1606 check_unshare_flags(&unshare_flags);
1608 /* Return -EINVAL for all unsupported flags */
1610 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1611 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1612 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER))
1613 goto bad_unshare_out;
1615 if ((err = unshare_thread(unshare_flags)))
1616 goto bad_unshare_out;
1617 if ((err = unshare_fs(unshare_flags, &new_fs)))
1618 goto bad_unshare_cleanup_thread;
1619 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1620 goto bad_unshare_cleanup_fs;
1621 if ((err = unshare_vm(unshare_flags, &new_mm)))
1622 goto bad_unshare_cleanup_sigh;
1623 if ((err = unshare_fd(unshare_flags, &new_fd)))
1624 goto bad_unshare_cleanup_vm;
1625 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1626 goto bad_unshare_cleanup_fd;
1627 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1629 goto bad_unshare_cleanup_semundo;
1631 if (new_fs || new_mm || new_fd || new_ulist || new_nsproxy) {
1636 old_nsproxy = current->nsproxy;
1637 current->nsproxy = new_nsproxy;
1638 new_nsproxy = old_nsproxy;
1643 current->fs = new_fs;
1649 active_mm = current->active_mm;
1650 current->mm = new_mm;
1651 current->active_mm = new_mm;
1652 activate_mm(active_mm, new_mm);
1657 fd = current->files;
1658 current->files = new_fd;
1662 task_unlock(current);
1666 put_nsproxy(new_nsproxy);
1668 bad_unshare_cleanup_semundo:
1669 bad_unshare_cleanup_fd:
1671 put_files_struct(new_fd);
1673 bad_unshare_cleanup_vm:
1677 bad_unshare_cleanup_sigh:
1679 if (atomic_dec_and_test(&new_sigh->count))
1680 kmem_cache_free(sighand_cachep, new_sigh);
1682 bad_unshare_cleanup_fs:
1684 put_fs_struct(new_fs);
1686 bad_unshare_cleanup_thread: