EDAC, AMD: decode load store MCEs
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / fork.c
blobe6c04d462ab250b806cd2e7d76d84fd1729d121e
1 /*
2 * linux/kernel/fork.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
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/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_counter.h>
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
73 #include <trace/events/sched.h>
76 * Protected counters by write_lock_irq(&tasklist_lock)
78 unsigned long total_forks; /* Handle normal Linux uptimes. */
79 int nr_threads; /* The idle threads do not count.. */
81 int max_threads; /* tunable limit on nr_threads */
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
87 int nr_processes(void)
89 int cpu;
90 int total = 0;
92 for_each_online_cpu(cpu)
93 total += per_cpu(process_counts, cpu);
95 return total;
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
102 #endif
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
109 #else
110 gfp_t mask = GFP_KERNEL;
111 #endif
112 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
115 static inline void free_thread_info(struct thread_info *ti)
117 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
119 #endif
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
139 void free_task(struct task_struct *tsk)
141 prop_local_destroy_single(&tsk->dirties);
142 free_thread_info(tsk->stack);
143 rt_mutex_debug_task_free(tsk);
144 ftrace_graph_exit_task(tsk);
145 free_task_struct(tsk);
147 EXPORT_SYMBOL(free_task);
149 void __put_task_struct(struct task_struct *tsk)
151 WARN_ON(!tsk->exit_state);
152 WARN_ON(atomic_read(&tsk->usage));
153 WARN_ON(tsk == current);
155 put_cred(tsk->real_cred);
156 put_cred(tsk->cred);
157 delayacct_tsk_free(tsk);
159 if (!profile_handoff_task(tsk))
160 free_task(tsk);
164 * macro override instead of weak attribute alias, to workaround
165 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
167 #ifndef arch_task_cache_init
168 #define arch_task_cache_init()
169 #endif
171 void __init fork_init(unsigned long mempages)
173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
174 #ifndef ARCH_MIN_TASKALIGN
175 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
176 #endif
177 /* create a slab on which task_structs can be allocated */
178 task_struct_cachep =
179 kmem_cache_create("task_struct", sizeof(struct task_struct),
180 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
181 #endif
183 /* do the arch specific task caches init */
184 arch_task_cache_init();
187 * The default maximum number of threads is set to a safe
188 * value: the thread structures can take up at most half
189 * of memory.
191 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
194 * we need to allow at least 20 threads to boot a system
196 if(max_threads < 20)
197 max_threads = 20;
199 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
200 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
201 init_task.signal->rlim[RLIMIT_SIGPENDING] =
202 init_task.signal->rlim[RLIMIT_NPROC];
205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
206 struct task_struct *src)
208 *dst = *src;
209 return 0;
212 static struct task_struct *dup_task_struct(struct task_struct *orig)
214 struct task_struct *tsk;
215 struct thread_info *ti;
216 unsigned long *stackend;
218 int err;
220 prepare_to_copy(orig);
222 tsk = alloc_task_struct();
223 if (!tsk)
224 return NULL;
226 ti = alloc_thread_info(tsk);
227 if (!ti) {
228 free_task_struct(tsk);
229 return NULL;
232 err = arch_dup_task_struct(tsk, orig);
233 if (err)
234 goto out;
236 tsk->stack = ti;
238 err = prop_local_init_single(&tsk->dirties);
239 if (err)
240 goto out;
242 setup_thread_stack(tsk, orig);
243 stackend = end_of_stack(tsk);
244 *stackend = STACK_END_MAGIC; /* for overflow detection */
246 #ifdef CONFIG_CC_STACKPROTECTOR
247 tsk->stack_canary = get_random_int();
248 #endif
250 /* One for us, one for whoever does the "release_task()" (usually parent) */
251 atomic_set(&tsk->usage,2);
252 atomic_set(&tsk->fs_excl, 0);
253 #ifdef CONFIG_BLK_DEV_IO_TRACE
254 tsk->btrace_seq = 0;
255 #endif
256 tsk->splice_pipe = NULL;
257 return tsk;
259 out:
260 free_thread_info(ti);
261 free_task_struct(tsk);
262 return NULL;
265 #ifdef CONFIG_MMU
266 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
268 struct vm_area_struct *mpnt, *tmp, **pprev;
269 struct rb_node **rb_link, *rb_parent;
270 int retval;
271 unsigned long charge;
272 struct mempolicy *pol;
274 down_write(&oldmm->mmap_sem);
275 flush_cache_dup_mm(oldmm);
277 * Not linked in yet - no deadlock potential:
279 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
281 mm->locked_vm = 0;
282 mm->mmap = NULL;
283 mm->mmap_cache = NULL;
284 mm->free_area_cache = oldmm->mmap_base;
285 mm->cached_hole_size = ~0UL;
286 mm->map_count = 0;
287 cpumask_clear(mm_cpumask(mm));
288 mm->mm_rb = RB_ROOT;
289 rb_link = &mm->mm_rb.rb_node;
290 rb_parent = NULL;
291 pprev = &mm->mmap;
293 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
294 struct file *file;
296 if (mpnt->vm_flags & VM_DONTCOPY) {
297 long pages = vma_pages(mpnt);
298 mm->total_vm -= pages;
299 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
300 -pages);
301 continue;
303 charge = 0;
304 if (mpnt->vm_flags & VM_ACCOUNT) {
305 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
306 if (security_vm_enough_memory(len))
307 goto fail_nomem;
308 charge = len;
310 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
311 if (!tmp)
312 goto fail_nomem;
313 *tmp = *mpnt;
314 pol = mpol_dup(vma_policy(mpnt));
315 retval = PTR_ERR(pol);
316 if (IS_ERR(pol))
317 goto fail_nomem_policy;
318 vma_set_policy(tmp, pol);
319 tmp->vm_flags &= ~VM_LOCKED;
320 tmp->vm_mm = mm;
321 tmp->vm_next = NULL;
322 anon_vma_link(tmp);
323 file = tmp->vm_file;
324 if (file) {
325 struct inode *inode = file->f_path.dentry->d_inode;
326 struct address_space *mapping = file->f_mapping;
328 get_file(file);
329 if (tmp->vm_flags & VM_DENYWRITE)
330 atomic_dec(&inode->i_writecount);
331 spin_lock(&mapping->i_mmap_lock);
332 if (tmp->vm_flags & VM_SHARED)
333 mapping->i_mmap_writable++;
334 tmp->vm_truncate_count = mpnt->vm_truncate_count;
335 flush_dcache_mmap_lock(mapping);
336 /* insert tmp into the share list, just after mpnt */
337 vma_prio_tree_add(tmp, mpnt);
338 flush_dcache_mmap_unlock(mapping);
339 spin_unlock(&mapping->i_mmap_lock);
343 * Clear hugetlb-related page reserves for children. This only
344 * affects MAP_PRIVATE mappings. Faults generated by the child
345 * are not guaranteed to succeed, even if read-only
347 if (is_vm_hugetlb_page(tmp))
348 reset_vma_resv_huge_pages(tmp);
351 * Link in the new vma and copy the page table entries.
353 *pprev = tmp;
354 pprev = &tmp->vm_next;
356 __vma_link_rb(mm, tmp, rb_link, rb_parent);
357 rb_link = &tmp->vm_rb.rb_right;
358 rb_parent = &tmp->vm_rb;
360 mm->map_count++;
361 retval = copy_page_range(mm, oldmm, mpnt);
363 if (tmp->vm_ops && tmp->vm_ops->open)
364 tmp->vm_ops->open(tmp);
366 if (retval)
367 goto out;
369 /* a new mm has just been created */
370 arch_dup_mmap(oldmm, mm);
371 retval = 0;
372 out:
373 up_write(&mm->mmap_sem);
374 flush_tlb_mm(oldmm);
375 up_write(&oldmm->mmap_sem);
376 return retval;
377 fail_nomem_policy:
378 kmem_cache_free(vm_area_cachep, tmp);
379 fail_nomem:
380 retval = -ENOMEM;
381 vm_unacct_memory(charge);
382 goto out;
385 static inline int mm_alloc_pgd(struct mm_struct * mm)
387 mm->pgd = pgd_alloc(mm);
388 if (unlikely(!mm->pgd))
389 return -ENOMEM;
390 return 0;
393 static inline void mm_free_pgd(struct mm_struct * mm)
395 pgd_free(mm, mm->pgd);
397 #else
398 #define dup_mmap(mm, oldmm) (0)
399 #define mm_alloc_pgd(mm) (0)
400 #define mm_free_pgd(mm)
401 #endif /* CONFIG_MMU */
403 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
405 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
406 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
408 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
410 static int __init coredump_filter_setup(char *s)
412 default_dump_filter =
413 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
414 MMF_DUMP_FILTER_MASK;
415 return 1;
418 __setup("coredump_filter=", coredump_filter_setup);
420 #include <linux/init_task.h>
422 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
424 atomic_set(&mm->mm_users, 1);
425 atomic_set(&mm->mm_count, 1);
426 init_rwsem(&mm->mmap_sem);
427 INIT_LIST_HEAD(&mm->mmlist);
428 mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
429 mm->core_state = NULL;
430 mm->nr_ptes = 0;
431 set_mm_counter(mm, file_rss, 0);
432 set_mm_counter(mm, anon_rss, 0);
433 spin_lock_init(&mm->page_table_lock);
434 spin_lock_init(&mm->ioctx_lock);
435 INIT_HLIST_HEAD(&mm->ioctx_list);
436 mm->free_area_cache = TASK_UNMAPPED_BASE;
437 mm->cached_hole_size = ~0UL;
438 mm_init_owner(mm, p);
440 if (likely(!mm_alloc_pgd(mm))) {
441 mm->def_flags = 0;
442 mmu_notifier_mm_init(mm);
443 return mm;
446 free_mm(mm);
447 return NULL;
451 * Allocate and initialize an mm_struct.
453 struct mm_struct * mm_alloc(void)
455 struct mm_struct * mm;
457 mm = allocate_mm();
458 if (mm) {
459 memset(mm, 0, sizeof(*mm));
460 mm = mm_init(mm, current);
462 return mm;
466 * Called when the last reference to the mm
467 * is dropped: either by a lazy thread or by
468 * mmput. Free the page directory and the mm.
470 void __mmdrop(struct mm_struct *mm)
472 BUG_ON(mm == &init_mm);
473 mm_free_pgd(mm);
474 destroy_context(mm);
475 mmu_notifier_mm_destroy(mm);
476 free_mm(mm);
478 EXPORT_SYMBOL_GPL(__mmdrop);
481 * Decrement the use count and release all resources for an mm.
483 void mmput(struct mm_struct *mm)
485 might_sleep();
487 if (atomic_dec_and_test(&mm->mm_users)) {
488 exit_aio(mm);
489 exit_mmap(mm);
490 set_mm_exe_file(mm, NULL);
491 if (!list_empty(&mm->mmlist)) {
492 spin_lock(&mmlist_lock);
493 list_del(&mm->mmlist);
494 spin_unlock(&mmlist_lock);
496 put_swap_token(mm);
497 mmdrop(mm);
500 EXPORT_SYMBOL_GPL(mmput);
503 * get_task_mm - acquire a reference to the task's mm
505 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
506 * this kernel workthread has transiently adopted a user mm with use_mm,
507 * to do its AIO) is not set and if so returns a reference to it, after
508 * bumping up the use count. User must release the mm via mmput()
509 * after use. Typically used by /proc and ptrace.
511 struct mm_struct *get_task_mm(struct task_struct *task)
513 struct mm_struct *mm;
515 task_lock(task);
516 mm = task->mm;
517 if (mm) {
518 if (task->flags & PF_KTHREAD)
519 mm = NULL;
520 else
521 atomic_inc(&mm->mm_users);
523 task_unlock(task);
524 return mm;
526 EXPORT_SYMBOL_GPL(get_task_mm);
528 /* Please note the differences between mmput and mm_release.
529 * mmput is called whenever we stop holding onto a mm_struct,
530 * error success whatever.
532 * mm_release is called after a mm_struct has been removed
533 * from the current process.
535 * This difference is important for error handling, when we
536 * only half set up a mm_struct for a new process and need to restore
537 * the old one. Because we mmput the new mm_struct before
538 * restoring the old one. . .
539 * Eric Biederman 10 January 1998
541 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
543 struct completion *vfork_done = tsk->vfork_done;
545 /* Get rid of any futexes when releasing the mm */
546 #ifdef CONFIG_FUTEX
547 if (unlikely(tsk->robust_list))
548 exit_robust_list(tsk);
549 #ifdef CONFIG_COMPAT
550 if (unlikely(tsk->compat_robust_list))
551 compat_exit_robust_list(tsk);
552 #endif
553 #endif
555 /* Get rid of any cached register state */
556 deactivate_mm(tsk, mm);
558 /* notify parent sleeping on vfork() */
559 if (vfork_done) {
560 tsk->vfork_done = NULL;
561 complete(vfork_done);
565 * If we're exiting normally, clear a user-space tid field if
566 * requested. We leave this alone when dying by signal, to leave
567 * the value intact in a core dump, and to save the unnecessary
568 * trouble otherwise. Userland only wants this done for a sys_exit.
570 if (tsk->clear_child_tid) {
571 if (!(tsk->flags & PF_SIGNALED) &&
572 atomic_read(&mm->mm_users) > 1) {
574 * We don't check the error code - if userspace has
575 * not set up a proper pointer then tough luck.
577 put_user(0, tsk->clear_child_tid);
578 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
579 1, NULL, NULL, 0);
581 tsk->clear_child_tid = NULL;
586 * Allocate a new mm structure and copy contents from the
587 * mm structure of the passed in task structure.
589 struct mm_struct *dup_mm(struct task_struct *tsk)
591 struct mm_struct *mm, *oldmm = current->mm;
592 int err;
594 if (!oldmm)
595 return NULL;
597 mm = allocate_mm();
598 if (!mm)
599 goto fail_nomem;
601 memcpy(mm, oldmm, sizeof(*mm));
603 /* Initializing for Swap token stuff */
604 mm->token_priority = 0;
605 mm->last_interval = 0;
607 if (!mm_init(mm, tsk))
608 goto fail_nomem;
610 if (init_new_context(tsk, mm))
611 goto fail_nocontext;
613 dup_mm_exe_file(oldmm, mm);
615 err = dup_mmap(mm, oldmm);
616 if (err)
617 goto free_pt;
619 mm->hiwater_rss = get_mm_rss(mm);
620 mm->hiwater_vm = mm->total_vm;
622 return mm;
624 free_pt:
625 mmput(mm);
627 fail_nomem:
628 return NULL;
630 fail_nocontext:
632 * If init_new_context() failed, we cannot use mmput() to free the mm
633 * because it calls destroy_context()
635 mm_free_pgd(mm);
636 free_mm(mm);
637 return NULL;
640 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
642 struct mm_struct * mm, *oldmm;
643 int retval;
645 tsk->min_flt = tsk->maj_flt = 0;
646 tsk->nvcsw = tsk->nivcsw = 0;
647 #ifdef CONFIG_DETECT_HUNG_TASK
648 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
649 #endif
651 tsk->mm = NULL;
652 tsk->active_mm = NULL;
655 * Are we cloning a kernel thread?
657 * We need to steal a active VM for that..
659 oldmm = current->mm;
660 if (!oldmm)
661 return 0;
663 if (clone_flags & CLONE_VM) {
664 atomic_inc(&oldmm->mm_users);
665 mm = oldmm;
666 goto good_mm;
669 retval = -ENOMEM;
670 mm = dup_mm(tsk);
671 if (!mm)
672 goto fail_nomem;
674 good_mm:
675 /* Initializing for Swap token stuff */
676 mm->token_priority = 0;
677 mm->last_interval = 0;
679 tsk->mm = mm;
680 tsk->active_mm = mm;
681 return 0;
683 fail_nomem:
684 return retval;
687 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
689 struct fs_struct *fs = current->fs;
690 if (clone_flags & CLONE_FS) {
691 /* tsk->fs is already what we want */
692 write_lock(&fs->lock);
693 if (fs->in_exec) {
694 write_unlock(&fs->lock);
695 return -EAGAIN;
697 fs->users++;
698 write_unlock(&fs->lock);
699 return 0;
701 tsk->fs = copy_fs_struct(fs);
702 if (!tsk->fs)
703 return -ENOMEM;
704 return 0;
707 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
709 struct files_struct *oldf, *newf;
710 int error = 0;
713 * A background process may not have any files ...
715 oldf = current->files;
716 if (!oldf)
717 goto out;
719 if (clone_flags & CLONE_FILES) {
720 atomic_inc(&oldf->count);
721 goto out;
724 newf = dup_fd(oldf, &error);
725 if (!newf)
726 goto out;
728 tsk->files = newf;
729 error = 0;
730 out:
731 return error;
734 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
736 #ifdef CONFIG_BLOCK
737 struct io_context *ioc = current->io_context;
739 if (!ioc)
740 return 0;
742 * Share io context with parent, if CLONE_IO is set
744 if (clone_flags & CLONE_IO) {
745 tsk->io_context = ioc_task_link(ioc);
746 if (unlikely(!tsk->io_context))
747 return -ENOMEM;
748 } else if (ioprio_valid(ioc->ioprio)) {
749 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
750 if (unlikely(!tsk->io_context))
751 return -ENOMEM;
753 tsk->io_context->ioprio = ioc->ioprio;
755 #endif
756 return 0;
759 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
761 struct sighand_struct *sig;
763 if (clone_flags & CLONE_SIGHAND) {
764 atomic_inc(&current->sighand->count);
765 return 0;
767 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
768 rcu_assign_pointer(tsk->sighand, sig);
769 if (!sig)
770 return -ENOMEM;
771 atomic_set(&sig->count, 1);
772 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
773 return 0;
776 void __cleanup_sighand(struct sighand_struct *sighand)
778 if (atomic_dec_and_test(&sighand->count))
779 kmem_cache_free(sighand_cachep, sighand);
784 * Initialize POSIX timer handling for a thread group.
786 static void posix_cpu_timers_init_group(struct signal_struct *sig)
788 /* Thread group counters. */
789 thread_group_cputime_init(sig);
791 /* Expiration times and increments. */
792 sig->it_virt_expires = cputime_zero;
793 sig->it_virt_incr = cputime_zero;
794 sig->it_prof_expires = cputime_zero;
795 sig->it_prof_incr = cputime_zero;
797 /* Cached expiration times. */
798 sig->cputime_expires.prof_exp = cputime_zero;
799 sig->cputime_expires.virt_exp = cputime_zero;
800 sig->cputime_expires.sched_exp = 0;
802 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
803 sig->cputime_expires.prof_exp =
804 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
805 sig->cputimer.running = 1;
808 /* The timer lists. */
809 INIT_LIST_HEAD(&sig->cpu_timers[0]);
810 INIT_LIST_HEAD(&sig->cpu_timers[1]);
811 INIT_LIST_HEAD(&sig->cpu_timers[2]);
814 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
816 struct signal_struct *sig;
818 if (clone_flags & CLONE_THREAD)
819 return 0;
821 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
822 tsk->signal = sig;
823 if (!sig)
824 return -ENOMEM;
826 atomic_set(&sig->count, 1);
827 atomic_set(&sig->live, 1);
828 init_waitqueue_head(&sig->wait_chldexit);
829 sig->flags = 0;
830 if (clone_flags & CLONE_NEWPID)
831 sig->flags |= SIGNAL_UNKILLABLE;
832 sig->group_exit_code = 0;
833 sig->group_exit_task = NULL;
834 sig->group_stop_count = 0;
835 sig->curr_target = tsk;
836 init_sigpending(&sig->shared_pending);
837 INIT_LIST_HEAD(&sig->posix_timers);
839 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
840 sig->it_real_incr.tv64 = 0;
841 sig->real_timer.function = it_real_fn;
843 sig->leader = 0; /* session leadership doesn't inherit */
844 sig->tty_old_pgrp = NULL;
845 sig->tty = NULL;
847 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
848 sig->gtime = cputime_zero;
849 sig->cgtime = cputime_zero;
850 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
851 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
852 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
853 task_io_accounting_init(&sig->ioac);
854 sig->sum_sched_runtime = 0;
855 taskstats_tgid_init(sig);
857 task_lock(current->group_leader);
858 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
859 task_unlock(current->group_leader);
861 posix_cpu_timers_init_group(sig);
863 acct_init_pacct(&sig->pacct);
865 tty_audit_fork(sig);
867 return 0;
870 void __cleanup_signal(struct signal_struct *sig)
872 thread_group_cputime_free(sig);
873 tty_kref_put(sig->tty);
874 kmem_cache_free(signal_cachep, sig);
877 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
879 unsigned long new_flags = p->flags;
881 new_flags &= ~PF_SUPERPRIV;
882 new_flags |= PF_FORKNOEXEC;
883 new_flags |= PF_STARTING;
884 p->flags = new_flags;
885 clear_freeze_flag(p);
888 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
890 current->clear_child_tid = tidptr;
892 return task_pid_vnr(current);
895 static void rt_mutex_init_task(struct task_struct *p)
897 spin_lock_init(&p->pi_lock);
898 #ifdef CONFIG_RT_MUTEXES
899 plist_head_init(&p->pi_waiters, &p->pi_lock);
900 p->pi_blocked_on = NULL;
901 #endif
904 #ifdef CONFIG_MM_OWNER
905 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
907 mm->owner = p;
909 #endif /* CONFIG_MM_OWNER */
912 * Initialize POSIX timer handling for a single task.
914 static void posix_cpu_timers_init(struct task_struct *tsk)
916 tsk->cputime_expires.prof_exp = cputime_zero;
917 tsk->cputime_expires.virt_exp = cputime_zero;
918 tsk->cputime_expires.sched_exp = 0;
919 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
920 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
921 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
925 * This creates a new process as a copy of the old one,
926 * but does not actually start it yet.
928 * It copies the registers, and all the appropriate
929 * parts of the process environment (as per the clone
930 * flags). The actual kick-off is left to the caller.
932 static struct task_struct *copy_process(unsigned long clone_flags,
933 unsigned long stack_start,
934 struct pt_regs *regs,
935 unsigned long stack_size,
936 int __user *child_tidptr,
937 struct pid *pid,
938 int trace)
940 int retval;
941 struct task_struct *p;
942 int cgroup_callbacks_done = 0;
944 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
945 return ERR_PTR(-EINVAL);
948 * Thread groups must share signals as well, and detached threads
949 * can only be started up within the thread group.
951 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
952 return ERR_PTR(-EINVAL);
955 * Shared signal handlers imply shared VM. By way of the above,
956 * thread groups also imply shared VM. Blocking this case allows
957 * for various simplifications in other code.
959 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
960 return ERR_PTR(-EINVAL);
962 retval = security_task_create(clone_flags);
963 if (retval)
964 goto fork_out;
966 retval = -ENOMEM;
967 p = dup_task_struct(current);
968 if (!p)
969 goto fork_out;
971 ftrace_graph_init_task(p);
973 rt_mutex_init_task(p);
975 #ifdef CONFIG_PROVE_LOCKING
976 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
977 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
978 #endif
979 retval = -EAGAIN;
980 if (atomic_read(&p->real_cred->user->processes) >=
981 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
982 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
983 p->real_cred->user != INIT_USER)
984 goto bad_fork_free;
987 retval = copy_creds(p, clone_flags);
988 if (retval < 0)
989 goto bad_fork_free;
992 * If multiple threads are within copy_process(), then this check
993 * triggers too late. This doesn't hurt, the check is only there
994 * to stop root fork bombs.
996 retval = -EAGAIN;
997 if (nr_threads >= max_threads)
998 goto bad_fork_cleanup_count;
1000 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1001 goto bad_fork_cleanup_count;
1003 if (p->binfmt && !try_module_get(p->binfmt->module))
1004 goto bad_fork_cleanup_put_domain;
1006 p->did_exec = 0;
1007 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1008 copy_flags(clone_flags, p);
1009 INIT_LIST_HEAD(&p->children);
1010 INIT_LIST_HEAD(&p->sibling);
1011 #ifdef CONFIG_PREEMPT_RCU
1012 p->rcu_read_lock_nesting = 0;
1013 p->rcu_flipctr_idx = 0;
1014 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1015 p->vfork_done = NULL;
1016 spin_lock_init(&p->alloc_lock);
1018 init_sigpending(&p->pending);
1020 p->utime = cputime_zero;
1021 p->stime = cputime_zero;
1022 p->gtime = cputime_zero;
1023 p->utimescaled = cputime_zero;
1024 p->stimescaled = cputime_zero;
1025 p->prev_utime = cputime_zero;
1026 p->prev_stime = cputime_zero;
1028 p->default_timer_slack_ns = current->timer_slack_ns;
1030 task_io_accounting_init(&p->ioac);
1031 acct_clear_integrals(p);
1033 posix_cpu_timers_init(p);
1035 p->lock_depth = -1; /* -1 = no lock */
1036 do_posix_clock_monotonic_gettime(&p->start_time);
1037 p->real_start_time = p->start_time;
1038 monotonic_to_bootbased(&p->real_start_time);
1039 p->io_context = NULL;
1040 p->audit_context = NULL;
1041 cgroup_fork(p);
1042 #ifdef CONFIG_NUMA
1043 p->mempolicy = mpol_dup(p->mempolicy);
1044 if (IS_ERR(p->mempolicy)) {
1045 retval = PTR_ERR(p->mempolicy);
1046 p->mempolicy = NULL;
1047 goto bad_fork_cleanup_cgroup;
1049 mpol_fix_fork_child_flag(p);
1050 #endif
1051 #ifdef CONFIG_TRACE_IRQFLAGS
1052 p->irq_events = 0;
1053 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1054 p->hardirqs_enabled = 1;
1055 #else
1056 p->hardirqs_enabled = 0;
1057 #endif
1058 p->hardirq_enable_ip = 0;
1059 p->hardirq_enable_event = 0;
1060 p->hardirq_disable_ip = _THIS_IP_;
1061 p->hardirq_disable_event = 0;
1062 p->softirqs_enabled = 1;
1063 p->softirq_enable_ip = _THIS_IP_;
1064 p->softirq_enable_event = 0;
1065 p->softirq_disable_ip = 0;
1066 p->softirq_disable_event = 0;
1067 p->hardirq_context = 0;
1068 p->softirq_context = 0;
1069 #endif
1070 #ifdef CONFIG_LOCKDEP
1071 p->lockdep_depth = 0; /* no locks held yet */
1072 p->curr_chain_key = 0;
1073 p->lockdep_recursion = 0;
1074 #endif
1076 #ifdef CONFIG_DEBUG_MUTEXES
1077 p->blocked_on = NULL; /* not blocked yet */
1078 #endif
1080 p->bts = NULL;
1082 /* Perform scheduler related setup. Assign this task to a CPU. */
1083 sched_fork(p, clone_flags);
1085 retval = perf_counter_init_task(p);
1086 if (retval)
1087 goto bad_fork_cleanup_policy;
1089 if ((retval = audit_alloc(p)))
1090 goto bad_fork_cleanup_policy;
1091 /* copy all the process information */
1092 if ((retval = copy_semundo(clone_flags, p)))
1093 goto bad_fork_cleanup_audit;
1094 if ((retval = copy_files(clone_flags, p)))
1095 goto bad_fork_cleanup_semundo;
1096 if ((retval = copy_fs(clone_flags, p)))
1097 goto bad_fork_cleanup_files;
1098 if ((retval = copy_sighand(clone_flags, p)))
1099 goto bad_fork_cleanup_fs;
1100 if ((retval = copy_signal(clone_flags, p)))
1101 goto bad_fork_cleanup_sighand;
1102 if ((retval = copy_mm(clone_flags, p)))
1103 goto bad_fork_cleanup_signal;
1104 if ((retval = copy_namespaces(clone_flags, p)))
1105 goto bad_fork_cleanup_mm;
1106 if ((retval = copy_io(clone_flags, p)))
1107 goto bad_fork_cleanup_namespaces;
1108 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1109 if (retval)
1110 goto bad_fork_cleanup_io;
1112 if (pid != &init_struct_pid) {
1113 retval = -ENOMEM;
1114 pid = alloc_pid(p->nsproxy->pid_ns);
1115 if (!pid)
1116 goto bad_fork_cleanup_io;
1118 if (clone_flags & CLONE_NEWPID) {
1119 retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1120 if (retval < 0)
1121 goto bad_fork_free_pid;
1125 p->pid = pid_nr(pid);
1126 p->tgid = p->pid;
1127 if (clone_flags & CLONE_THREAD)
1128 p->tgid = current->tgid;
1130 if (current->nsproxy != p->nsproxy) {
1131 retval = ns_cgroup_clone(p, pid);
1132 if (retval)
1133 goto bad_fork_free_pid;
1136 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1138 * Clear TID on mm_release()?
1140 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1141 #ifdef CONFIG_FUTEX
1142 p->robust_list = NULL;
1143 #ifdef CONFIG_COMPAT
1144 p->compat_robust_list = NULL;
1145 #endif
1146 INIT_LIST_HEAD(&p->pi_state_list);
1147 p->pi_state_cache = NULL;
1148 #endif
1150 * sigaltstack should be cleared when sharing the same VM
1152 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1153 p->sas_ss_sp = p->sas_ss_size = 0;
1156 * Syscall tracing should be turned off in the child regardless
1157 * of CLONE_PTRACE.
1159 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1160 #ifdef TIF_SYSCALL_EMU
1161 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1162 #endif
1163 clear_all_latency_tracing(p);
1165 /* ok, now we should be set up.. */
1166 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1167 p->pdeath_signal = 0;
1168 p->exit_state = 0;
1171 * Ok, make it visible to the rest of the system.
1172 * We dont wake it up yet.
1174 p->group_leader = p;
1175 INIT_LIST_HEAD(&p->thread_group);
1177 /* Now that the task is set up, run cgroup callbacks if
1178 * necessary. We need to run them before the task is visible
1179 * on the tasklist. */
1180 cgroup_fork_callbacks(p);
1181 cgroup_callbacks_done = 1;
1183 /* Need tasklist lock for parent etc handling! */
1184 write_lock_irq(&tasklist_lock);
1187 * The task hasn't been attached yet, so its cpus_allowed mask will
1188 * not be changed, nor will its assigned CPU.
1190 * The cpus_allowed mask of the parent may have changed after it was
1191 * copied first time - so re-copy it here, then check the child's CPU
1192 * to ensure it is on a valid CPU (and if not, just force it back to
1193 * parent's CPU). This avoids alot of nasty races.
1195 p->cpus_allowed = current->cpus_allowed;
1196 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1197 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1198 !cpu_online(task_cpu(p))))
1199 set_task_cpu(p, smp_processor_id());
1201 /* CLONE_PARENT re-uses the old parent */
1202 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1203 p->real_parent = current->real_parent;
1204 p->parent_exec_id = current->parent_exec_id;
1205 } else {
1206 p->real_parent = current;
1207 p->parent_exec_id = current->self_exec_id;
1210 spin_lock(&current->sighand->siglock);
1213 * Process group and session signals need to be delivered to just the
1214 * parent before the fork or both the parent and the child after the
1215 * fork. Restart if a signal comes in before we add the new process to
1216 * it's process group.
1217 * A fatal signal pending means that current will exit, so the new
1218 * thread can't slip out of an OOM kill (or normal SIGKILL).
1220 recalc_sigpending();
1221 if (signal_pending(current)) {
1222 spin_unlock(&current->sighand->siglock);
1223 write_unlock_irq(&tasklist_lock);
1224 retval = -ERESTARTNOINTR;
1225 goto bad_fork_free_pid;
1228 if (clone_flags & CLONE_THREAD) {
1229 atomic_inc(&current->signal->count);
1230 atomic_inc(&current->signal->live);
1231 p->group_leader = current->group_leader;
1232 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1235 if (likely(p->pid)) {
1236 list_add_tail(&p->sibling, &p->real_parent->children);
1237 tracehook_finish_clone(p, clone_flags, trace);
1239 if (thread_group_leader(p)) {
1240 if (clone_flags & CLONE_NEWPID)
1241 p->nsproxy->pid_ns->child_reaper = p;
1243 p->signal->leader_pid = pid;
1244 tty_kref_put(p->signal->tty);
1245 p->signal->tty = tty_kref_get(current->signal->tty);
1246 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1247 attach_pid(p, PIDTYPE_SID, task_session(current));
1248 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1249 __get_cpu_var(process_counts)++;
1251 attach_pid(p, PIDTYPE_PID, pid);
1252 nr_threads++;
1255 total_forks++;
1256 spin_unlock(&current->sighand->siglock);
1257 write_unlock_irq(&tasklist_lock);
1258 proc_fork_connector(p);
1259 cgroup_post_fork(p);
1260 perf_counter_fork(p);
1261 return p;
1263 bad_fork_free_pid:
1264 if (pid != &init_struct_pid)
1265 free_pid(pid);
1266 bad_fork_cleanup_io:
1267 put_io_context(p->io_context);
1268 bad_fork_cleanup_namespaces:
1269 exit_task_namespaces(p);
1270 bad_fork_cleanup_mm:
1271 if (p->mm)
1272 mmput(p->mm);
1273 bad_fork_cleanup_signal:
1274 if (!(clone_flags & CLONE_THREAD))
1275 __cleanup_signal(p->signal);
1276 bad_fork_cleanup_sighand:
1277 __cleanup_sighand(p->sighand);
1278 bad_fork_cleanup_fs:
1279 exit_fs(p); /* blocking */
1280 bad_fork_cleanup_files:
1281 exit_files(p); /* blocking */
1282 bad_fork_cleanup_semundo:
1283 exit_sem(p);
1284 bad_fork_cleanup_audit:
1285 audit_free(p);
1286 bad_fork_cleanup_policy:
1287 perf_counter_free_task(p);
1288 #ifdef CONFIG_NUMA
1289 mpol_put(p->mempolicy);
1290 bad_fork_cleanup_cgroup:
1291 #endif
1292 cgroup_exit(p, cgroup_callbacks_done);
1293 delayacct_tsk_free(p);
1294 if (p->binfmt)
1295 module_put(p->binfmt->module);
1296 bad_fork_cleanup_put_domain:
1297 module_put(task_thread_info(p)->exec_domain->module);
1298 bad_fork_cleanup_count:
1299 atomic_dec(&p->cred->user->processes);
1300 put_cred(p->real_cred);
1301 put_cred(p->cred);
1302 bad_fork_free:
1303 free_task(p);
1304 fork_out:
1305 return ERR_PTR(retval);
1308 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1310 memset(regs, 0, sizeof(struct pt_regs));
1311 return regs;
1314 struct task_struct * __cpuinit fork_idle(int cpu)
1316 struct task_struct *task;
1317 struct pt_regs regs;
1319 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1320 &init_struct_pid, 0);
1321 if (!IS_ERR(task))
1322 init_idle(task, cpu);
1324 return task;
1328 * Ok, this is the main fork-routine.
1330 * It copies the process, and if successful kick-starts
1331 * it and waits for it to finish using the VM if required.
1333 long do_fork(unsigned long clone_flags,
1334 unsigned long stack_start,
1335 struct pt_regs *regs,
1336 unsigned long stack_size,
1337 int __user *parent_tidptr,
1338 int __user *child_tidptr)
1340 struct task_struct *p;
1341 int trace = 0;
1342 long nr;
1345 * Do some preliminary argument and permissions checking before we
1346 * actually start allocating stuff
1348 if (clone_flags & CLONE_NEWUSER) {
1349 if (clone_flags & CLONE_THREAD)
1350 return -EINVAL;
1351 /* hopefully this check will go away when userns support is
1352 * complete
1354 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1355 !capable(CAP_SETGID))
1356 return -EPERM;
1360 * We hope to recycle these flags after 2.6.26
1362 if (unlikely(clone_flags & CLONE_STOPPED)) {
1363 static int __read_mostly count = 100;
1365 if (count > 0 && printk_ratelimit()) {
1366 char comm[TASK_COMM_LEN];
1368 count--;
1369 printk(KERN_INFO "fork(): process `%s' used deprecated "
1370 "clone flags 0x%lx\n",
1371 get_task_comm(comm, current),
1372 clone_flags & CLONE_STOPPED);
1377 * When called from kernel_thread, don't do user tracing stuff.
1379 if (likely(user_mode(regs)))
1380 trace = tracehook_prepare_clone(clone_flags);
1382 p = copy_process(clone_flags, stack_start, regs, stack_size,
1383 child_tidptr, NULL, trace);
1385 * Do this prior waking up the new thread - the thread pointer
1386 * might get invalid after that point, if the thread exits quickly.
1388 if (!IS_ERR(p)) {
1389 struct completion vfork;
1391 trace_sched_process_fork(current, p);
1393 nr = task_pid_vnr(p);
1395 if (clone_flags & CLONE_PARENT_SETTID)
1396 put_user(nr, parent_tidptr);
1398 if (clone_flags & CLONE_VFORK) {
1399 p->vfork_done = &vfork;
1400 init_completion(&vfork);
1403 audit_finish_fork(p);
1404 tracehook_report_clone(regs, clone_flags, nr, p);
1407 * We set PF_STARTING at creation in case tracing wants to
1408 * use this to distinguish a fully live task from one that
1409 * hasn't gotten to tracehook_report_clone() yet. Now we
1410 * clear it and set the child going.
1412 p->flags &= ~PF_STARTING;
1414 if (unlikely(clone_flags & CLONE_STOPPED)) {
1416 * We'll start up with an immediate SIGSTOP.
1418 sigaddset(&p->pending.signal, SIGSTOP);
1419 set_tsk_thread_flag(p, TIF_SIGPENDING);
1420 __set_task_state(p, TASK_STOPPED);
1421 } else {
1422 wake_up_new_task(p, clone_flags);
1425 tracehook_report_clone_complete(trace, regs,
1426 clone_flags, nr, p);
1428 if (clone_flags & CLONE_VFORK) {
1429 freezer_do_not_count();
1430 wait_for_completion(&vfork);
1431 freezer_count();
1432 tracehook_report_vfork_done(p, nr);
1434 } else {
1435 nr = PTR_ERR(p);
1437 return nr;
1440 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1441 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1442 #endif
1444 static void sighand_ctor(void *data)
1446 struct sighand_struct *sighand = data;
1448 spin_lock_init(&sighand->siglock);
1449 init_waitqueue_head(&sighand->signalfd_wqh);
1452 void __init proc_caches_init(void)
1454 sighand_cachep = kmem_cache_create("sighand_cache",
1455 sizeof(struct sighand_struct), 0,
1456 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1457 SLAB_NOTRACK, sighand_ctor);
1458 signal_cachep = kmem_cache_create("signal_cache",
1459 sizeof(struct signal_struct), 0,
1460 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1461 files_cachep = kmem_cache_create("files_cache",
1462 sizeof(struct files_struct), 0,
1463 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1464 fs_cachep = kmem_cache_create("fs_cache",
1465 sizeof(struct fs_struct), 0,
1466 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1467 mm_cachep = kmem_cache_create("mm_struct",
1468 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1469 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1470 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1471 mmap_init();
1475 * Check constraints on flags passed to the unshare system call and
1476 * force unsharing of additional process context as appropriate.
1478 static void check_unshare_flags(unsigned long *flags_ptr)
1481 * If unsharing a thread from a thread group, must also
1482 * unshare vm.
1484 if (*flags_ptr & CLONE_THREAD)
1485 *flags_ptr |= CLONE_VM;
1488 * If unsharing vm, must also unshare signal handlers.
1490 if (*flags_ptr & CLONE_VM)
1491 *flags_ptr |= CLONE_SIGHAND;
1494 * If unsharing signal handlers and the task was created
1495 * using CLONE_THREAD, then must unshare the thread
1497 if ((*flags_ptr & CLONE_SIGHAND) &&
1498 (atomic_read(&current->signal->count) > 1))
1499 *flags_ptr |= CLONE_THREAD;
1502 * If unsharing namespace, must also unshare filesystem information.
1504 if (*flags_ptr & CLONE_NEWNS)
1505 *flags_ptr |= CLONE_FS;
1509 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1511 static int unshare_thread(unsigned long unshare_flags)
1513 if (unshare_flags & CLONE_THREAD)
1514 return -EINVAL;
1516 return 0;
1520 * Unshare the filesystem structure if it is being shared
1522 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1524 struct fs_struct *fs = current->fs;
1526 if (!(unshare_flags & CLONE_FS) || !fs)
1527 return 0;
1529 /* don't need lock here; in the worst case we'll do useless copy */
1530 if (fs->users == 1)
1531 return 0;
1533 *new_fsp = copy_fs_struct(fs);
1534 if (!*new_fsp)
1535 return -ENOMEM;
1537 return 0;
1541 * Unsharing of sighand is not supported yet
1543 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1545 struct sighand_struct *sigh = current->sighand;
1547 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1548 return -EINVAL;
1549 else
1550 return 0;
1554 * Unshare vm if it is being shared
1556 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1558 struct mm_struct *mm = current->mm;
1560 if ((unshare_flags & CLONE_VM) &&
1561 (mm && atomic_read(&mm->mm_users) > 1)) {
1562 return -EINVAL;
1565 return 0;
1569 * Unshare file descriptor table if it is being shared
1571 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1573 struct files_struct *fd = current->files;
1574 int error = 0;
1576 if ((unshare_flags & CLONE_FILES) &&
1577 (fd && atomic_read(&fd->count) > 1)) {
1578 *new_fdp = dup_fd(fd, &error);
1579 if (!*new_fdp)
1580 return error;
1583 return 0;
1587 * unshare allows a process to 'unshare' part of the process
1588 * context which was originally shared using clone. copy_*
1589 * functions used by do_fork() cannot be used here directly
1590 * because they modify an inactive task_struct that is being
1591 * constructed. Here we are modifying the current, active,
1592 * task_struct.
1594 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1596 int err = 0;
1597 struct fs_struct *fs, *new_fs = NULL;
1598 struct sighand_struct *new_sigh = NULL;
1599 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1600 struct files_struct *fd, *new_fd = NULL;
1601 struct nsproxy *new_nsproxy = NULL;
1602 int do_sysvsem = 0;
1604 check_unshare_flags(&unshare_flags);
1606 /* Return -EINVAL for all unsupported flags */
1607 err = -EINVAL;
1608 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1609 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1610 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1611 goto bad_unshare_out;
1614 * CLONE_NEWIPC must also detach from the undolist: after switching
1615 * to a new ipc namespace, the semaphore arrays from the old
1616 * namespace are unreachable.
1618 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1619 do_sysvsem = 1;
1620 if ((err = unshare_thread(unshare_flags)))
1621 goto bad_unshare_out;
1622 if ((err = unshare_fs(unshare_flags, &new_fs)))
1623 goto bad_unshare_cleanup_thread;
1624 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1625 goto bad_unshare_cleanup_fs;
1626 if ((err = unshare_vm(unshare_flags, &new_mm)))
1627 goto bad_unshare_cleanup_sigh;
1628 if ((err = unshare_fd(unshare_flags, &new_fd)))
1629 goto bad_unshare_cleanup_vm;
1630 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1631 new_fs)))
1632 goto bad_unshare_cleanup_fd;
1634 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1635 if (do_sysvsem) {
1637 * CLONE_SYSVSEM is equivalent to sys_exit().
1639 exit_sem(current);
1642 if (new_nsproxy) {
1643 switch_task_namespaces(current, new_nsproxy);
1644 new_nsproxy = NULL;
1647 task_lock(current);
1649 if (new_fs) {
1650 fs = current->fs;
1651 write_lock(&fs->lock);
1652 current->fs = new_fs;
1653 if (--fs->users)
1654 new_fs = NULL;
1655 else
1656 new_fs = fs;
1657 write_unlock(&fs->lock);
1660 if (new_mm) {
1661 mm = current->mm;
1662 active_mm = current->active_mm;
1663 current->mm = new_mm;
1664 current->active_mm = new_mm;
1665 activate_mm(active_mm, new_mm);
1666 new_mm = mm;
1669 if (new_fd) {
1670 fd = current->files;
1671 current->files = new_fd;
1672 new_fd = fd;
1675 task_unlock(current);
1678 if (new_nsproxy)
1679 put_nsproxy(new_nsproxy);
1681 bad_unshare_cleanup_fd:
1682 if (new_fd)
1683 put_files_struct(new_fd);
1685 bad_unshare_cleanup_vm:
1686 if (new_mm)
1687 mmput(new_mm);
1689 bad_unshare_cleanup_sigh:
1690 if (new_sigh)
1691 if (atomic_dec_and_test(&new_sigh->count))
1692 kmem_cache_free(sighand_cachep, new_sigh);
1694 bad_unshare_cleanup_fs:
1695 if (new_fs)
1696 free_fs_struct(new_fs);
1698 bad_unshare_cleanup_thread:
1699 bad_unshare_out:
1700 return err;
1704 * Helper to unshare the files of the current task.
1705 * We don't want to expose copy_files internals to
1706 * the exec layer of the kernel.
1709 int unshare_files(struct files_struct **displaced)
1711 struct task_struct *task = current;
1712 struct files_struct *copy = NULL;
1713 int error;
1715 error = unshare_fd(CLONE_FILES, &copy);
1716 if (error || !copy) {
1717 *displaced = NULL;
1718 return error;
1720 *displaced = task->files;
1721 task_lock(task);
1722 task->files = copy;
1723 task_unlock(task);
1724 return 0;