ia64/xen-unstable

view linux-2.6-xen-sparse/kernel/fork.c @ 12327:eea9247ad5a0

[XENOPROF] Oprofile user level samples for passive domains are being lost.
The number of of lost samples is most significant when dom0 is idle.

From: joserenato.santos@hp.com
Signed-off-by: Keir Fraser <keir@xensource.com>
author kfraser@localhost.localdomain
date Thu Nov 09 11:47:42 2006 +0000 (2006-11-09)
parents 2fea03842f40
children 4fad820a2233
line source
1 /*
2 * linux/kernel/fork.c
3 *
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()'
12 */
14 #include <linux/config.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/unistd.h>
18 #include <linux/smp_lock.h>
19 #include <linux/module.h>
20 #include <linux/vmalloc.h>
21 #include <linux/completion.h>
22 #include <linux/namespace.h>
23 #include <linux/personality.h>
24 #include <linux/mempolicy.h>
25 #include <linux/sem.h>
26 #include <linux/file.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/fs.h>
31 #include <linux/capability.h>
32 #include <linux/cpu.h>
33 #include <linux/cpuset.h>
34 #include <linux/security.h>
35 #include <linux/swap.h>
36 #include <linux/syscalls.h>
37 #include <linux/jiffies.h>
38 #include <linux/futex.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/cn_proc.h>
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/uaccess.h>
51 #include <asm/mmu_context.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
55 /*
56 * Protected counters by write_lock_irq(&tasklist_lock)
57 */
58 unsigned long total_forks; /* Handle normal Linux uptimes. */
59 int nr_threads; /* The idle threads do not count.. */
61 int max_threads; /* tunable limit on nr_threads */
63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
65 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
67 EXPORT_SYMBOL(tasklist_lock);
69 int nr_processes(void)
70 {
71 int cpu;
72 int total = 0;
74 for_each_online_cpu(cpu)
75 total += per_cpu(process_counts, cpu);
77 return total;
78 }
80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
81 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
82 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
83 static kmem_cache_t *task_struct_cachep;
84 #endif
86 /* SLAB cache for signal_struct structures (tsk->signal) */
87 kmem_cache_t *signal_cachep;
89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
90 kmem_cache_t *sighand_cachep;
92 /* SLAB cache for files_struct structures (tsk->files) */
93 kmem_cache_t *files_cachep;
95 /* SLAB cache for fs_struct structures (tsk->fs) */
96 kmem_cache_t *fs_cachep;
98 /* SLAB cache for vm_area_struct structures */
99 kmem_cache_t *vm_area_cachep;
101 /* SLAB cache for mm_struct structures (tsk->mm) */
102 static kmem_cache_t *mm_cachep;
104 void free_task(struct task_struct *tsk)
105 {
106 free_thread_info(tsk->thread_info);
107 free_task_struct(tsk);
108 }
109 EXPORT_SYMBOL(free_task);
111 void __put_task_struct_cb(struct rcu_head *rhp)
112 {
113 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
115 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
116 WARN_ON(atomic_read(&tsk->usage));
117 WARN_ON(tsk == current);
119 if (unlikely(tsk->audit_context))
120 audit_free(tsk);
121 security_task_free(tsk);
122 free_uid(tsk->user);
123 put_group_info(tsk->group_info);
125 if (!profile_handoff_task(tsk))
126 free_task(tsk);
127 }
129 void __init fork_init(unsigned long mempages)
130 {
131 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
132 #ifndef ARCH_MIN_TASKALIGN
133 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
134 #endif
135 /* create a slab on which task_structs can be allocated */
136 task_struct_cachep =
137 kmem_cache_create("task_struct", sizeof(struct task_struct),
138 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
139 #endif
141 /*
142 * The default maximum number of threads is set to a safe
143 * value: the thread structures can take up at most half
144 * of memory.
145 */
146 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
148 /*
149 * we need to allow at least 20 threads to boot a system
150 */
151 if(max_threads < 20)
152 max_threads = 20;
154 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
155 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
156 init_task.signal->rlim[RLIMIT_SIGPENDING] =
157 init_task.signal->rlim[RLIMIT_NPROC];
158 }
160 static struct task_struct *dup_task_struct(struct task_struct *orig)
161 {
162 struct task_struct *tsk;
163 struct thread_info *ti;
165 prepare_to_copy(orig);
167 tsk = alloc_task_struct();
168 if (!tsk)
169 return NULL;
171 ti = alloc_thread_info(tsk);
172 if (!ti) {
173 free_task_struct(tsk);
174 return NULL;
175 }
177 *tsk = *orig;
178 tsk->thread_info = ti;
179 setup_thread_stack(tsk, orig);
181 /* One for us, one for whoever does the "release_task()" (usually parent) */
182 atomic_set(&tsk->usage,2);
183 atomic_set(&tsk->fs_excl, 0);
184 return tsk;
185 }
187 #ifdef CONFIG_MMU
188 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
189 {
190 struct vm_area_struct *mpnt, *tmp, **pprev;
191 struct rb_node **rb_link, *rb_parent;
192 int retval;
193 unsigned long charge;
194 struct mempolicy *pol;
196 down_write(&oldmm->mmap_sem);
197 flush_cache_mm(oldmm);
198 down_write(&mm->mmap_sem);
200 mm->locked_vm = 0;
201 mm->mmap = NULL;
202 mm->mmap_cache = NULL;
203 mm->free_area_cache = oldmm->mmap_base;
204 mm->cached_hole_size = ~0UL;
205 mm->map_count = 0;
206 cpus_clear(mm->cpu_vm_mask);
207 mm->mm_rb = RB_ROOT;
208 rb_link = &mm->mm_rb.rb_node;
209 rb_parent = NULL;
210 pprev = &mm->mmap;
212 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
213 struct file *file;
215 if (mpnt->vm_flags & VM_DONTCOPY) {
216 long pages = vma_pages(mpnt);
217 mm->total_vm -= pages;
218 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
219 -pages);
220 continue;
221 }
222 charge = 0;
223 if (mpnt->vm_flags & VM_ACCOUNT) {
224 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
225 if (security_vm_enough_memory(len))
226 goto fail_nomem;
227 charge = len;
228 }
229 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
230 if (!tmp)
231 goto fail_nomem;
232 *tmp = *mpnt;
233 pol = mpol_copy(vma_policy(mpnt));
234 retval = PTR_ERR(pol);
235 if (IS_ERR(pol))
236 goto fail_nomem_policy;
237 vma_set_policy(tmp, pol);
238 tmp->vm_flags &= ~VM_LOCKED;
239 tmp->vm_mm = mm;
240 tmp->vm_next = NULL;
241 anon_vma_link(tmp);
242 file = tmp->vm_file;
243 if (file) {
244 struct inode *inode = file->f_dentry->d_inode;
245 get_file(file);
246 if (tmp->vm_flags & VM_DENYWRITE)
247 atomic_dec(&inode->i_writecount);
249 /* insert tmp into the share list, just after mpnt */
250 spin_lock(&file->f_mapping->i_mmap_lock);
251 tmp->vm_truncate_count = mpnt->vm_truncate_count;
252 flush_dcache_mmap_lock(file->f_mapping);
253 vma_prio_tree_add(tmp, mpnt);
254 flush_dcache_mmap_unlock(file->f_mapping);
255 spin_unlock(&file->f_mapping->i_mmap_lock);
256 }
258 /*
259 * Link in the new vma and copy the page table entries.
260 */
261 *pprev = tmp;
262 pprev = &tmp->vm_next;
264 __vma_link_rb(mm, tmp, rb_link, rb_parent);
265 rb_link = &tmp->vm_rb.rb_right;
266 rb_parent = &tmp->vm_rb;
268 mm->map_count++;
269 retval = copy_page_range(mm, oldmm, mpnt);
271 if (tmp->vm_ops && tmp->vm_ops->open)
272 tmp->vm_ops->open(tmp);
274 if (retval)
275 goto out;
276 }
277 #ifdef arch_dup_mmap
278 arch_dup_mmap(mm, oldmm);
279 #endif
280 retval = 0;
281 out:
282 up_write(&mm->mmap_sem);
283 flush_tlb_mm(oldmm);
284 up_write(&oldmm->mmap_sem);
285 return retval;
286 fail_nomem_policy:
287 kmem_cache_free(vm_area_cachep, tmp);
288 fail_nomem:
289 retval = -ENOMEM;
290 vm_unacct_memory(charge);
291 goto out;
292 }
294 static inline int mm_alloc_pgd(struct mm_struct * mm)
295 {
296 mm->pgd = pgd_alloc(mm);
297 if (unlikely(!mm->pgd))
298 return -ENOMEM;
299 return 0;
300 }
302 static inline void mm_free_pgd(struct mm_struct * mm)
303 {
304 pgd_free(mm->pgd);
305 }
306 #else
307 #define dup_mmap(mm, oldmm) (0)
308 #define mm_alloc_pgd(mm) (0)
309 #define mm_free_pgd(mm)
310 #endif /* CONFIG_MMU */
312 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
314 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
315 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
317 #include <linux/init_task.h>
319 static struct mm_struct * mm_init(struct mm_struct * mm)
320 {
321 atomic_set(&mm->mm_users, 1);
322 atomic_set(&mm->mm_count, 1);
323 init_rwsem(&mm->mmap_sem);
324 INIT_LIST_HEAD(&mm->mmlist);
325 mm->core_waiters = 0;
326 mm->nr_ptes = 0;
327 set_mm_counter(mm, file_rss, 0);
328 set_mm_counter(mm, anon_rss, 0);
329 spin_lock_init(&mm->page_table_lock);
330 rwlock_init(&mm->ioctx_list_lock);
331 mm->ioctx_list = NULL;
332 mm->free_area_cache = TASK_UNMAPPED_BASE;
333 mm->cached_hole_size = ~0UL;
335 if (likely(!mm_alloc_pgd(mm))) {
336 mm->def_flags = 0;
337 return mm;
338 }
339 free_mm(mm);
340 return NULL;
341 }
343 /*
344 * Allocate and initialize an mm_struct.
345 */
346 struct mm_struct * mm_alloc(void)
347 {
348 struct mm_struct * mm;
350 mm = allocate_mm();
351 if (mm) {
352 memset(mm, 0, sizeof(*mm));
353 mm = mm_init(mm);
354 }
355 return mm;
356 }
358 /*
359 * Called when the last reference to the mm
360 * is dropped: either by a lazy thread or by
361 * mmput. Free the page directory and the mm.
362 */
363 void fastcall __mmdrop(struct mm_struct *mm)
364 {
365 BUG_ON(mm == &init_mm);
366 mm_free_pgd(mm);
367 destroy_context(mm);
368 free_mm(mm);
369 }
371 /*
372 * Decrement the use count and release all resources for an mm.
373 */
374 void mmput(struct mm_struct *mm)
375 {
376 if (atomic_dec_and_test(&mm->mm_users)) {
377 exit_aio(mm);
378 exit_mmap(mm);
379 if (!list_empty(&mm->mmlist)) {
380 spin_lock(&mmlist_lock);
381 list_del(&mm->mmlist);
382 spin_unlock(&mmlist_lock);
383 }
384 put_swap_token(mm);
385 mmdrop(mm);
386 }
387 }
388 EXPORT_SYMBOL_GPL(mmput);
390 /**
391 * get_task_mm - acquire a reference to the task's mm
392 *
393 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
394 * this kernel workthread has transiently adopted a user mm with use_mm,
395 * to do its AIO) is not set and if so returns a reference to it, after
396 * bumping up the use count. User must release the mm via mmput()
397 * after use. Typically used by /proc and ptrace.
398 */
399 struct mm_struct *get_task_mm(struct task_struct *task)
400 {
401 struct mm_struct *mm;
403 task_lock(task);
404 mm = task->mm;
405 if (mm) {
406 if (task->flags & PF_BORROWED_MM)
407 mm = NULL;
408 else
409 atomic_inc(&mm->mm_users);
410 }
411 task_unlock(task);
412 return mm;
413 }
414 EXPORT_SYMBOL_GPL(get_task_mm);
416 /* Please note the differences between mmput and mm_release.
417 * mmput is called whenever we stop holding onto a mm_struct,
418 * error success whatever.
419 *
420 * mm_release is called after a mm_struct has been removed
421 * from the current process.
422 *
423 * This difference is important for error handling, when we
424 * only half set up a mm_struct for a new process and need to restore
425 * the old one. Because we mmput the new mm_struct before
426 * restoring the old one. . .
427 * Eric Biederman 10 January 1998
428 */
429 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
430 {
431 struct completion *vfork_done = tsk->vfork_done;
433 /* Get rid of any cached register state */
434 deactivate_mm(tsk, mm);
436 /* notify parent sleeping on vfork() */
437 if (vfork_done) {
438 tsk->vfork_done = NULL;
439 complete(vfork_done);
440 }
441 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
442 u32 __user * tidptr = tsk->clear_child_tid;
443 tsk->clear_child_tid = NULL;
445 /*
446 * We don't check the error code - if userspace has
447 * not set up a proper pointer then tough luck.
448 */
449 put_user(0, tidptr);
450 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
451 }
452 }
454 /*
455 * Allocate a new mm structure and copy contents from the
456 * mm structure of the passed in task structure.
457 */
458 static struct mm_struct *dup_mm(struct task_struct *tsk)
459 {
460 struct mm_struct *mm, *oldmm = current->mm;
461 int err;
463 if (!oldmm)
464 return NULL;
466 mm = allocate_mm();
467 if (!mm)
468 goto fail_nomem;
470 memcpy(mm, oldmm, sizeof(*mm));
472 if (!mm_init(mm))
473 goto fail_nomem;
475 if (init_new_context(tsk, mm))
476 goto fail_nocontext;
478 err = dup_mmap(mm, oldmm);
479 if (err)
480 goto free_pt;
482 mm->hiwater_rss = get_mm_rss(mm);
483 mm->hiwater_vm = mm->total_vm;
485 return mm;
487 free_pt:
488 mmput(mm);
490 fail_nomem:
491 return NULL;
493 fail_nocontext:
494 /*
495 * If init_new_context() failed, we cannot use mmput() to free the mm
496 * because it calls destroy_context()
497 */
498 mm_free_pgd(mm);
499 free_mm(mm);
500 return NULL;
501 }
503 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
504 {
505 struct mm_struct * mm, *oldmm;
506 int retval;
508 tsk->min_flt = tsk->maj_flt = 0;
509 tsk->nvcsw = tsk->nivcsw = 0;
511 tsk->mm = NULL;
512 tsk->active_mm = NULL;
514 /*
515 * Are we cloning a kernel thread?
516 *
517 * We need to steal a active VM for that..
518 */
519 oldmm = current->mm;
520 if (!oldmm)
521 return 0;
523 if (clone_flags & CLONE_VM) {
524 atomic_inc(&oldmm->mm_users);
525 mm = oldmm;
526 goto good_mm;
527 }
529 retval = -ENOMEM;
530 mm = dup_mm(tsk);
531 if (!mm)
532 goto fail_nomem;
534 good_mm:
535 tsk->mm = mm;
536 tsk->active_mm = mm;
537 return 0;
539 fail_nomem:
540 return retval;
541 }
543 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
544 {
545 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
546 /* We don't need to lock fs - think why ;-) */
547 if (fs) {
548 atomic_set(&fs->count, 1);
549 rwlock_init(&fs->lock);
550 fs->umask = old->umask;
551 read_lock(&old->lock);
552 fs->rootmnt = mntget(old->rootmnt);
553 fs->root = dget(old->root);
554 fs->pwdmnt = mntget(old->pwdmnt);
555 fs->pwd = dget(old->pwd);
556 if (old->altroot) {
557 fs->altrootmnt = mntget(old->altrootmnt);
558 fs->altroot = dget(old->altroot);
559 } else {
560 fs->altrootmnt = NULL;
561 fs->altroot = NULL;
562 }
563 read_unlock(&old->lock);
564 }
565 return fs;
566 }
568 struct fs_struct *copy_fs_struct(struct fs_struct *old)
569 {
570 return __copy_fs_struct(old);
571 }
573 EXPORT_SYMBOL_GPL(copy_fs_struct);
575 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
576 {
577 if (clone_flags & CLONE_FS) {
578 atomic_inc(&current->fs->count);
579 return 0;
580 }
581 tsk->fs = __copy_fs_struct(current->fs);
582 if (!tsk->fs)
583 return -ENOMEM;
584 return 0;
585 }
587 static int count_open_files(struct fdtable *fdt)
588 {
589 int size = fdt->max_fdset;
590 int i;
592 /* Find the last open fd */
593 for (i = size/(8*sizeof(long)); i > 0; ) {
594 if (fdt->open_fds->fds_bits[--i])
595 break;
596 }
597 i = (i+1) * 8 * sizeof(long);
598 return i;
599 }
601 static struct files_struct *alloc_files(void)
602 {
603 struct files_struct *newf;
604 struct fdtable *fdt;
606 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
607 if (!newf)
608 goto out;
610 atomic_set(&newf->count, 1);
612 spin_lock_init(&newf->file_lock);
613 fdt = &newf->fdtab;
614 fdt->next_fd = 0;
615 fdt->max_fds = NR_OPEN_DEFAULT;
616 fdt->max_fdset = __FD_SETSIZE;
617 fdt->close_on_exec = &newf->close_on_exec_init;
618 fdt->open_fds = &newf->open_fds_init;
619 fdt->fd = &newf->fd_array[0];
620 INIT_RCU_HEAD(&fdt->rcu);
621 fdt->free_files = NULL;
622 fdt->next = NULL;
623 rcu_assign_pointer(newf->fdt, fdt);
624 out:
625 return newf;
626 }
628 /*
629 * Allocate a new files structure and copy contents from the
630 * passed in files structure.
631 */
632 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
633 {
634 struct files_struct *newf;
635 struct file **old_fds, **new_fds;
636 int open_files, size, i, expand;
637 struct fdtable *old_fdt, *new_fdt;
639 newf = alloc_files();
640 if (!newf)
641 goto out;
643 spin_lock(&oldf->file_lock);
644 old_fdt = files_fdtable(oldf);
645 new_fdt = files_fdtable(newf);
646 size = old_fdt->max_fdset;
647 open_files = count_open_files(old_fdt);
648 expand = 0;
650 /*
651 * Check whether we need to allocate a larger fd array or fd set.
652 * Note: we're not a clone task, so the open count won't change.
653 */
654 if (open_files > new_fdt->max_fdset) {
655 new_fdt->max_fdset = 0;
656 expand = 1;
657 }
658 if (open_files > new_fdt->max_fds) {
659 new_fdt->max_fds = 0;
660 expand = 1;
661 }
663 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
664 if (expand) {
665 spin_unlock(&oldf->file_lock);
666 spin_lock(&newf->file_lock);
667 *errorp = expand_files(newf, open_files-1);
668 spin_unlock(&newf->file_lock);
669 if (*errorp < 0)
670 goto out_release;
671 new_fdt = files_fdtable(newf);
672 /*
673 * Reacquire the oldf lock and a pointer to its fd table
674 * who knows it may have a new bigger fd table. We need
675 * the latest pointer.
676 */
677 spin_lock(&oldf->file_lock);
678 old_fdt = files_fdtable(oldf);
679 }
681 old_fds = old_fdt->fd;
682 new_fds = new_fdt->fd;
684 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
685 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
687 for (i = open_files; i != 0; i--) {
688 struct file *f = *old_fds++;
689 if (f) {
690 get_file(f);
691 } else {
692 /*
693 * The fd may be claimed in the fd bitmap but not yet
694 * instantiated in the files array if a sibling thread
695 * is partway through open(). So make sure that this
696 * fd is available to the new process.
697 */
698 FD_CLR(open_files - i, new_fdt->open_fds);
699 }
700 rcu_assign_pointer(*new_fds++, f);
701 }
702 spin_unlock(&oldf->file_lock);
704 /* compute the remainder to be cleared */
705 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
707 /* This is long word aligned thus could use a optimized version */
708 memset(new_fds, 0, size);
710 if (new_fdt->max_fdset > open_files) {
711 int left = (new_fdt->max_fdset-open_files)/8;
712 int start = open_files / (8 * sizeof(unsigned long));
714 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
715 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
716 }
718 out:
719 return newf;
721 out_release:
722 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
723 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
724 free_fd_array(new_fdt->fd, new_fdt->max_fds);
725 kmem_cache_free(files_cachep, newf);
726 return NULL;
727 }
729 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
730 {
731 struct files_struct *oldf, *newf;
732 int error = 0;
734 /*
735 * A background process may not have any files ...
736 */
737 oldf = current->files;
738 if (!oldf)
739 goto out;
741 if (clone_flags & CLONE_FILES) {
742 atomic_inc(&oldf->count);
743 goto out;
744 }
746 /*
747 * Note: we may be using current for both targets (See exec.c)
748 * This works because we cache current->files (old) as oldf. Don't
749 * break this.
750 */
751 tsk->files = NULL;
752 error = -ENOMEM;
753 newf = dup_fd(oldf, &error);
754 if (!newf)
755 goto out;
757 tsk->files = newf;
758 error = 0;
759 out:
760 return error;
761 }
763 /*
764 * Helper to unshare the files of the current task.
765 * We don't want to expose copy_files internals to
766 * the exec layer of the kernel.
767 */
769 int unshare_files(void)
770 {
771 struct files_struct *files = current->files;
772 int rc;
774 if(!files)
775 BUG();
777 /* This can race but the race causes us to copy when we don't
778 need to and drop the copy */
779 if(atomic_read(&files->count) == 1)
780 {
781 atomic_inc(&files->count);
782 return 0;
783 }
784 rc = copy_files(0, current);
785 if(rc)
786 current->files = files;
787 return rc;
788 }
790 EXPORT_SYMBOL(unshare_files);
792 void sighand_free_cb(struct rcu_head *rhp)
793 {
794 struct sighand_struct *sp;
796 sp = container_of(rhp, struct sighand_struct, rcu);
797 kmem_cache_free(sighand_cachep, sp);
798 }
800 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
801 {
802 struct sighand_struct *sig;
804 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
805 atomic_inc(&current->sighand->count);
806 return 0;
807 }
808 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
809 rcu_assign_pointer(tsk->sighand, sig);
810 if (!sig)
811 return -ENOMEM;
812 spin_lock_init(&sig->siglock);
813 atomic_set(&sig->count, 1);
814 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
815 return 0;
816 }
818 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
819 {
820 struct signal_struct *sig;
821 int ret;
823 if (clone_flags & CLONE_THREAD) {
824 atomic_inc(&current->signal->count);
825 atomic_inc(&current->signal->live);
826 return 0;
827 }
828 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
829 tsk->signal = sig;
830 if (!sig)
831 return -ENOMEM;
833 ret = copy_thread_group_keys(tsk);
834 if (ret < 0) {
835 kmem_cache_free(signal_cachep, sig);
836 return ret;
837 }
839 atomic_set(&sig->count, 1);
840 atomic_set(&sig->live, 1);
841 init_waitqueue_head(&sig->wait_chldexit);
842 sig->flags = 0;
843 sig->group_exit_code = 0;
844 sig->group_exit_task = NULL;
845 sig->group_stop_count = 0;
846 sig->curr_target = NULL;
847 init_sigpending(&sig->shared_pending);
848 INIT_LIST_HEAD(&sig->posix_timers);
850 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
851 sig->it_real_incr.tv64 = 0;
852 sig->real_timer.function = it_real_fn;
853 sig->real_timer.data = tsk;
855 sig->it_virt_expires = cputime_zero;
856 sig->it_virt_incr = cputime_zero;
857 sig->it_prof_expires = cputime_zero;
858 sig->it_prof_incr = cputime_zero;
860 sig->leader = 0; /* session leadership doesn't inherit */
861 sig->tty_old_pgrp = 0;
863 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
864 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
865 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
866 sig->sched_time = 0;
867 INIT_LIST_HEAD(&sig->cpu_timers[0]);
868 INIT_LIST_HEAD(&sig->cpu_timers[1]);
869 INIT_LIST_HEAD(&sig->cpu_timers[2]);
871 task_lock(current->group_leader);
872 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
873 task_unlock(current->group_leader);
875 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
876 /*
877 * New sole thread in the process gets an expiry time
878 * of the whole CPU time limit.
879 */
880 tsk->it_prof_expires =
881 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
882 }
884 return 0;
885 }
887 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
888 {
889 unsigned long new_flags = p->flags;
891 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
892 new_flags |= PF_FORKNOEXEC;
893 if (!(clone_flags & CLONE_PTRACE))
894 p->ptrace = 0;
895 p->flags = new_flags;
896 }
898 asmlinkage long sys_set_tid_address(int __user *tidptr)
899 {
900 current->clear_child_tid = tidptr;
902 return current->pid;
903 }
905 /*
906 * This creates a new process as a copy of the old one,
907 * but does not actually start it yet.
908 *
909 * It copies the registers, and all the appropriate
910 * parts of the process environment (as per the clone
911 * flags). The actual kick-off is left to the caller.
912 */
913 static task_t *copy_process(unsigned long clone_flags,
914 unsigned long stack_start,
915 struct pt_regs *regs,
916 unsigned long stack_size,
917 int __user *parent_tidptr,
918 int __user *child_tidptr,
919 int pid)
920 {
921 int retval;
922 struct task_struct *p = NULL;
924 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
925 return ERR_PTR(-EINVAL);
927 /*
928 * Thread groups must share signals as well, and detached threads
929 * can only be started up within the thread group.
930 */
931 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
932 return ERR_PTR(-EINVAL);
934 /*
935 * Shared signal handlers imply shared VM. By way of the above,
936 * thread groups also imply shared VM. Blocking this case allows
937 * for various simplifications in other code.
938 */
939 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
940 return ERR_PTR(-EINVAL);
942 retval = security_task_create(clone_flags);
943 if (retval)
944 goto fork_out;
946 retval = -ENOMEM;
947 p = dup_task_struct(current);
948 if (!p)
949 goto fork_out;
951 retval = -EAGAIN;
952 if (atomic_read(&p->user->processes) >=
953 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
954 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
955 p->user != &root_user)
956 goto bad_fork_free;
957 }
959 atomic_inc(&p->user->__count);
960 atomic_inc(&p->user->processes);
961 get_group_info(p->group_info);
963 /*
964 * If multiple threads are within copy_process(), then this check
965 * triggers too late. This doesn't hurt, the check is only there
966 * to stop root fork bombs.
967 */
968 if (nr_threads >= max_threads)
969 goto bad_fork_cleanup_count;
971 if (!try_module_get(task_thread_info(p)->exec_domain->module))
972 goto bad_fork_cleanup_count;
974 if (p->binfmt && !try_module_get(p->binfmt->module))
975 goto bad_fork_cleanup_put_domain;
977 p->did_exec = 0;
978 copy_flags(clone_flags, p);
979 p->pid = pid;
980 retval = -EFAULT;
981 if (clone_flags & CLONE_PARENT_SETTID)
982 if (put_user(p->pid, parent_tidptr))
983 goto bad_fork_cleanup;
985 p->proc_dentry = NULL;
987 INIT_LIST_HEAD(&p->children);
988 INIT_LIST_HEAD(&p->sibling);
989 p->vfork_done = NULL;
990 spin_lock_init(&p->alloc_lock);
991 spin_lock_init(&p->proc_lock);
993 clear_tsk_thread_flag(p, TIF_SIGPENDING);
994 init_sigpending(&p->pending);
996 p->utime = cputime_zero;
997 p->stime = cputime_zero;
998 p->sched_time = 0;
999 p->rchar = 0; /* I/O counter: bytes read */
1000 p->wchar = 0; /* I/O counter: bytes written */
1001 p->syscr = 0; /* I/O counter: read syscalls */
1002 p->syscw = 0; /* I/O counter: write syscalls */
1003 acct_clear_integrals(p);
1005 p->it_virt_expires = cputime_zero;
1006 p->it_prof_expires = cputime_zero;
1007 p->it_sched_expires = 0;
1008 INIT_LIST_HEAD(&p->cpu_timers[0]);
1009 INIT_LIST_HEAD(&p->cpu_timers[1]);
1010 INIT_LIST_HEAD(&p->cpu_timers[2]);
1012 p->lock_depth = -1; /* -1 = no lock */
1013 do_posix_clock_monotonic_gettime(&p->start_time);
1014 p->security = NULL;
1015 p->io_context = NULL;
1016 p->io_wait = NULL;
1017 p->audit_context = NULL;
1018 cpuset_fork(p);
1019 #ifdef CONFIG_NUMA
1020 p->mempolicy = mpol_copy(p->mempolicy);
1021 if (IS_ERR(p->mempolicy)) {
1022 retval = PTR_ERR(p->mempolicy);
1023 p->mempolicy = NULL;
1024 goto bad_fork_cleanup_cpuset;
1026 #endif
1028 #ifdef CONFIG_DEBUG_MUTEXES
1029 p->blocked_on = NULL; /* not blocked yet */
1030 #endif
1032 p->tgid = p->pid;
1033 if (clone_flags & CLONE_THREAD)
1034 p->tgid = current->tgid;
1036 if ((retval = security_task_alloc(p)))
1037 goto bad_fork_cleanup_policy;
1038 if ((retval = audit_alloc(p)))
1039 goto bad_fork_cleanup_security;
1040 /* copy all the process information */
1041 if ((retval = copy_semundo(clone_flags, p)))
1042 goto bad_fork_cleanup_audit;
1043 if ((retval = copy_files(clone_flags, p)))
1044 goto bad_fork_cleanup_semundo;
1045 if ((retval = copy_fs(clone_flags, p)))
1046 goto bad_fork_cleanup_files;
1047 if ((retval = copy_sighand(clone_flags, p)))
1048 goto bad_fork_cleanup_fs;
1049 if ((retval = copy_signal(clone_flags, p)))
1050 goto bad_fork_cleanup_sighand;
1051 if ((retval = copy_mm(clone_flags, p)))
1052 goto bad_fork_cleanup_signal;
1053 if ((retval = copy_keys(clone_flags, p)))
1054 goto bad_fork_cleanup_mm;
1055 if ((retval = copy_namespace(clone_flags, p)))
1056 goto bad_fork_cleanup_keys;
1057 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1058 if (retval)
1059 goto bad_fork_cleanup_namespace;
1061 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1062 /*
1063 * Clear TID on mm_release()?
1064 */
1065 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1067 /*
1068 * sigaltstack should be cleared when sharing the same VM
1069 */
1070 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1071 p->sas_ss_sp = p->sas_ss_size = 0;
1073 /*
1074 * Syscall tracing should be turned off in the child regardless
1075 * of CLONE_PTRACE.
1076 */
1077 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1078 #ifdef TIF_SYSCALL_EMU
1079 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1080 #endif
1082 /* Our parent execution domain becomes current domain
1083 These must match for thread signalling to apply */
1085 p->parent_exec_id = p->self_exec_id;
1087 /* ok, now we should be set up.. */
1088 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1089 p->pdeath_signal = 0;
1090 p->exit_state = 0;
1092 /*
1093 * Ok, make it visible to the rest of the system.
1094 * We dont wake it up yet.
1095 */
1096 p->group_leader = p;
1097 INIT_LIST_HEAD(&p->ptrace_children);
1098 INIT_LIST_HEAD(&p->ptrace_list);
1100 /* Perform scheduler related setup. Assign this task to a CPU. */
1101 sched_fork(p, clone_flags);
1103 /* Need tasklist lock for parent etc handling! */
1104 write_lock_irq(&tasklist_lock);
1106 /*
1107 * The task hasn't been attached yet, so its cpus_allowed mask will
1108 * not be changed, nor will its assigned CPU.
1110 * The cpus_allowed mask of the parent may have changed after it was
1111 * copied first time - so re-copy it here, then check the child's CPU
1112 * to ensure it is on a valid CPU (and if not, just force it back to
1113 * parent's CPU). This avoids alot of nasty races.
1114 */
1115 p->cpus_allowed = current->cpus_allowed;
1116 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1117 !cpu_online(task_cpu(p))))
1118 set_task_cpu(p, smp_processor_id());
1120 /*
1121 * Check for pending SIGKILL! The new thread should not be allowed
1122 * to slip out of an OOM kill. (or normal SIGKILL.)
1123 */
1124 if (sigismember(&current->pending.signal, SIGKILL)) {
1125 write_unlock_irq(&tasklist_lock);
1126 retval = -EINTR;
1127 goto bad_fork_cleanup_namespace;
1130 /* CLONE_PARENT re-uses the old parent */
1131 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1132 p->real_parent = current->real_parent;
1133 else
1134 p->real_parent = current;
1135 p->parent = p->real_parent;
1137 spin_lock(&current->sighand->siglock);
1138 if (clone_flags & CLONE_THREAD) {
1139 /*
1140 * Important: if an exit-all has been started then
1141 * do not create this new thread - the whole thread
1142 * group is supposed to exit anyway.
1143 */
1144 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1145 spin_unlock(&current->sighand->siglock);
1146 write_unlock_irq(&tasklist_lock);
1147 retval = -EAGAIN;
1148 goto bad_fork_cleanup_namespace;
1150 p->group_leader = current->group_leader;
1152 if (current->signal->group_stop_count > 0) {
1153 /*
1154 * There is an all-stop in progress for the group.
1155 * We ourselves will stop as soon as we check signals.
1156 * Make the new thread part of that group stop too.
1157 */
1158 current->signal->group_stop_count++;
1159 set_tsk_thread_flag(p, TIF_SIGPENDING);
1162 if (!cputime_eq(current->signal->it_virt_expires,
1163 cputime_zero) ||
1164 !cputime_eq(current->signal->it_prof_expires,
1165 cputime_zero) ||
1166 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1167 !list_empty(&current->signal->cpu_timers[0]) ||
1168 !list_empty(&current->signal->cpu_timers[1]) ||
1169 !list_empty(&current->signal->cpu_timers[2])) {
1170 /*
1171 * Have child wake up on its first tick to check
1172 * for process CPU timers.
1173 */
1174 p->it_prof_expires = jiffies_to_cputime(1);
1178 /*
1179 * inherit ioprio
1180 */
1181 p->ioprio = current->ioprio;
1183 SET_LINKS(p);
1184 if (unlikely(p->ptrace & PT_PTRACED))
1185 __ptrace_link(p, current->parent);
1187 if (thread_group_leader(p)) {
1188 p->signal->tty = current->signal->tty;
1189 p->signal->pgrp = process_group(current);
1190 p->signal->session = current->signal->session;
1191 attach_pid(p, PIDTYPE_PGID, process_group(p));
1192 attach_pid(p, PIDTYPE_SID, p->signal->session);
1193 if (p->pid)
1194 __get_cpu_var(process_counts)++;
1196 attach_pid(p, PIDTYPE_TGID, p->tgid);
1197 attach_pid(p, PIDTYPE_PID, p->pid);
1199 nr_threads++;
1200 total_forks++;
1201 spin_unlock(&current->sighand->siglock);
1202 write_unlock_irq(&tasklist_lock);
1203 proc_fork_connector(p);
1204 return p;
1206 bad_fork_cleanup_namespace:
1207 exit_namespace(p);
1208 bad_fork_cleanup_keys:
1209 exit_keys(p);
1210 bad_fork_cleanup_mm:
1211 if (p->mm)
1212 mmput(p->mm);
1213 bad_fork_cleanup_signal:
1214 exit_signal(p);
1215 bad_fork_cleanup_sighand:
1216 exit_sighand(p);
1217 bad_fork_cleanup_fs:
1218 exit_fs(p); /* blocking */
1219 bad_fork_cleanup_files:
1220 exit_files(p); /* blocking */
1221 bad_fork_cleanup_semundo:
1222 exit_sem(p);
1223 bad_fork_cleanup_audit:
1224 audit_free(p);
1225 bad_fork_cleanup_security:
1226 security_task_free(p);
1227 bad_fork_cleanup_policy:
1228 #ifdef CONFIG_NUMA
1229 mpol_free(p->mempolicy);
1230 bad_fork_cleanup_cpuset:
1231 #endif
1232 cpuset_exit(p);
1233 bad_fork_cleanup:
1234 if (p->binfmt)
1235 module_put(p->binfmt->module);
1236 bad_fork_cleanup_put_domain:
1237 module_put(task_thread_info(p)->exec_domain->module);
1238 bad_fork_cleanup_count:
1239 put_group_info(p->group_info);
1240 atomic_dec(&p->user->processes);
1241 free_uid(p->user);
1242 bad_fork_free:
1243 free_task(p);
1244 fork_out:
1245 return ERR_PTR(retval);
1248 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1250 memset(regs, 0, sizeof(struct pt_regs));
1251 return regs;
1254 task_t * __devinit fork_idle(int cpu)
1256 task_t *task;
1257 struct pt_regs regs;
1259 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1260 if (!task)
1261 return ERR_PTR(-ENOMEM);
1262 init_idle(task, cpu);
1263 unhash_process(task);
1264 return task;
1267 static inline int fork_traceflag (unsigned clone_flags)
1269 if (clone_flags & CLONE_UNTRACED)
1270 return 0;
1271 else if (clone_flags & CLONE_VFORK) {
1272 if (current->ptrace & PT_TRACE_VFORK)
1273 return PTRACE_EVENT_VFORK;
1274 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1275 if (current->ptrace & PT_TRACE_CLONE)
1276 return PTRACE_EVENT_CLONE;
1277 } else if (current->ptrace & PT_TRACE_FORK)
1278 return PTRACE_EVENT_FORK;
1280 return 0;
1283 /*
1284 * Ok, this is the main fork-routine.
1286 * It copies the process, and if successful kick-starts
1287 * it and waits for it to finish using the VM if required.
1288 */
1289 long do_fork(unsigned long clone_flags,
1290 unsigned long stack_start,
1291 struct pt_regs *regs,
1292 unsigned long stack_size,
1293 int __user *parent_tidptr,
1294 int __user *child_tidptr)
1296 struct task_struct *p;
1297 int trace = 0;
1298 long pid = alloc_pidmap();
1300 if (pid < 0)
1301 return -EAGAIN;
1302 if (unlikely(current->ptrace)) {
1303 trace = fork_traceflag (clone_flags);
1304 if (trace)
1305 clone_flags |= CLONE_PTRACE;
1308 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1309 /*
1310 * Do this prior waking up the new thread - the thread pointer
1311 * might get invalid after that point, if the thread exits quickly.
1312 */
1313 if (!IS_ERR(p)) {
1314 struct completion vfork;
1316 if (clone_flags & CLONE_VFORK) {
1317 p->vfork_done = &vfork;
1318 init_completion(&vfork);
1321 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1322 /*
1323 * We'll start up with an immediate SIGSTOP.
1324 */
1325 sigaddset(&p->pending.signal, SIGSTOP);
1326 set_tsk_thread_flag(p, TIF_SIGPENDING);
1329 if (!(clone_flags & CLONE_STOPPED))
1330 wake_up_new_task(p, clone_flags);
1331 else
1332 p->state = TASK_STOPPED;
1334 if (unlikely (trace)) {
1335 current->ptrace_message = pid;
1336 ptrace_notify ((trace << 8) | SIGTRAP);
1339 if (clone_flags & CLONE_VFORK) {
1340 wait_for_completion(&vfork);
1341 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1342 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1344 } else {
1345 free_pidmap(pid);
1346 pid = PTR_ERR(p);
1348 return pid;
1351 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1352 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1353 #endif
1355 void __init proc_caches_init(void)
1357 sighand_cachep = kmem_cache_create("sighand_cache",
1358 sizeof(struct sighand_struct), 0,
1359 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1360 signal_cachep = kmem_cache_create("signal_cache",
1361 sizeof(struct signal_struct), 0,
1362 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1363 files_cachep = kmem_cache_create("files_cache",
1364 sizeof(struct files_struct), 0,
1365 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1366 fs_cachep = kmem_cache_create("fs_cache",
1367 sizeof(struct fs_struct), 0,
1368 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1369 vm_area_cachep = kmem_cache_create("vm_area_struct",
1370 sizeof(struct vm_area_struct), 0,
1371 SLAB_PANIC, NULL, NULL);
1372 mm_cachep = kmem_cache_create("mm_struct",
1373 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1374 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1378 /*
1379 * Check constraints on flags passed to the unshare system call and
1380 * force unsharing of additional process context as appropriate.
1381 */
1382 static inline void check_unshare_flags(unsigned long *flags_ptr)
1384 /*
1385 * If unsharing a thread from a thread group, must also
1386 * unshare vm.
1387 */
1388 if (*flags_ptr & CLONE_THREAD)
1389 *flags_ptr |= CLONE_VM;
1391 /*
1392 * If unsharing vm, must also unshare signal handlers.
1393 */
1394 if (*flags_ptr & CLONE_VM)
1395 *flags_ptr |= CLONE_SIGHAND;
1397 /*
1398 * If unsharing signal handlers and the task was created
1399 * using CLONE_THREAD, then must unshare the thread
1400 */
1401 if ((*flags_ptr & CLONE_SIGHAND) &&
1402 (atomic_read(&current->signal->count) > 1))
1403 *flags_ptr |= CLONE_THREAD;
1405 /*
1406 * If unsharing namespace, must also unshare filesystem information.
1407 */
1408 if (*flags_ptr & CLONE_NEWNS)
1409 *flags_ptr |= CLONE_FS;
1412 /*
1413 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1414 */
1415 static int unshare_thread(unsigned long unshare_flags)
1417 if (unshare_flags & CLONE_THREAD)
1418 return -EINVAL;
1420 return 0;
1423 /*
1424 * Unshare the filesystem structure if it is being shared
1425 */
1426 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1428 struct fs_struct *fs = current->fs;
1430 if ((unshare_flags & CLONE_FS) &&
1431 (fs && atomic_read(&fs->count) > 1)) {
1432 *new_fsp = __copy_fs_struct(current->fs);
1433 if (!*new_fsp)
1434 return -ENOMEM;
1437 return 0;
1440 /*
1441 * Unshare the namespace structure if it is being shared
1442 */
1443 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1445 struct namespace *ns = current->namespace;
1447 if ((unshare_flags & CLONE_NEWNS) &&
1448 (ns && atomic_read(&ns->count) > 1)) {
1449 if (!capable(CAP_SYS_ADMIN))
1450 return -EPERM;
1452 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1453 if (!*new_nsp)
1454 return -ENOMEM;
1457 return 0;
1460 /*
1461 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1462 * supported yet
1463 */
1464 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1466 struct sighand_struct *sigh = current->sighand;
1468 if ((unshare_flags & CLONE_SIGHAND) &&
1469 (sigh && atomic_read(&sigh->count) > 1))
1470 return -EINVAL;
1471 else
1472 return 0;
1475 /*
1476 * Unshare vm if it is being shared
1477 */
1478 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1480 struct mm_struct *mm = current->mm;
1482 if ((unshare_flags & CLONE_VM) &&
1483 (mm && atomic_read(&mm->mm_users) > 1)) {
1484 return -EINVAL;
1487 return 0;
1490 /*
1491 * Unshare file descriptor table if it is being shared
1492 */
1493 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1495 struct files_struct *fd = current->files;
1496 int error = 0;
1498 if ((unshare_flags & CLONE_FILES) &&
1499 (fd && atomic_read(&fd->count) > 1)) {
1500 *new_fdp = dup_fd(fd, &error);
1501 if (!*new_fdp)
1502 return error;
1505 return 0;
1508 /*
1509 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1510 * supported yet
1511 */
1512 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1514 if (unshare_flags & CLONE_SYSVSEM)
1515 return -EINVAL;
1517 return 0;
1520 /*
1521 * unshare allows a process to 'unshare' part of the process
1522 * context which was originally shared using clone. copy_*
1523 * functions used by do_fork() cannot be used here directly
1524 * because they modify an inactive task_struct that is being
1525 * constructed. Here we are modifying the current, active,
1526 * task_struct.
1527 */
1528 asmlinkage long sys_unshare(unsigned long unshare_flags)
1530 int err = 0;
1531 struct fs_struct *fs, *new_fs = NULL;
1532 struct namespace *ns, *new_ns = NULL;
1533 struct sighand_struct *sigh, *new_sigh = NULL;
1534 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1535 struct files_struct *fd, *new_fd = NULL;
1536 struct sem_undo_list *new_ulist = NULL;
1538 check_unshare_flags(&unshare_flags);
1540 if ((err = unshare_thread(unshare_flags)))
1541 goto bad_unshare_out;
1542 if ((err = unshare_fs(unshare_flags, &new_fs)))
1543 goto bad_unshare_cleanup_thread;
1544 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1545 goto bad_unshare_cleanup_fs;
1546 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1547 goto bad_unshare_cleanup_ns;
1548 if ((err = unshare_vm(unshare_flags, &new_mm)))
1549 goto bad_unshare_cleanup_sigh;
1550 if ((err = unshare_fd(unshare_flags, &new_fd)))
1551 goto bad_unshare_cleanup_vm;
1552 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1553 goto bad_unshare_cleanup_fd;
1555 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1557 task_lock(current);
1559 if (new_fs) {
1560 fs = current->fs;
1561 current->fs = new_fs;
1562 new_fs = fs;
1565 if (new_ns) {
1566 ns = current->namespace;
1567 current->namespace = new_ns;
1568 new_ns = ns;
1571 if (new_sigh) {
1572 sigh = current->sighand;
1573 rcu_assign_pointer(current->sighand, new_sigh);
1574 new_sigh = sigh;
1577 if (new_mm) {
1578 mm = current->mm;
1579 active_mm = current->active_mm;
1580 current->mm = new_mm;
1581 current->active_mm = new_mm;
1582 activate_mm(active_mm, new_mm);
1583 new_mm = mm;
1586 if (new_fd) {
1587 fd = current->files;
1588 current->files = new_fd;
1589 new_fd = fd;
1592 task_unlock(current);
1595 bad_unshare_cleanup_fd:
1596 if (new_fd)
1597 put_files_struct(new_fd);
1599 bad_unshare_cleanup_vm:
1600 if (new_mm)
1601 mmput(new_mm);
1603 bad_unshare_cleanup_sigh:
1604 if (new_sigh)
1605 if (atomic_dec_and_test(&new_sigh->count))
1606 kmem_cache_free(sighand_cachep, new_sigh);
1608 bad_unshare_cleanup_ns:
1609 if (new_ns)
1610 put_namespace(new_ns);
1612 bad_unshare_cleanup_fs:
1613 if (new_fs)
1614 put_fs_struct(new_fs);
1616 bad_unshare_cleanup_thread:
1617 bad_unshare_out:
1618 return err;