projects/linux·blob v2.6.17-rc6 · refs

1650 lines · 39.7 KiB

1/*
2 * linux/kernel/fork.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
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 */
13
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>
47
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>
54
55/*
56 * Protected counters by write_lock_irq(&tasklist_lock)
57 */
58unsigned long total_forks; /* Handle normal Linux uptimes. */
59int nr_threads; /* The idle threads do not count.. */
60
61int max_threads; /* tunable limit on nr_threads */
62
63DEFINE_PER_CPU(unsigned long, process_counts) = 0;
64
65 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
66
67EXPORT_SYMBOL(tasklist_lock);
68
69int nr_processes(void)
70{
71 int cpu;
72 int total = 0;
73
74 for_each_online_cpu(cpu)
75 total += per_cpu(process_counts, cpu);
76
77 return total;
78}
79
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))
83static kmem_cache_t *task_struct_cachep;
84#endif
85
86/* SLAB cache for signal_struct structures (tsk->signal) */
87static kmem_cache_t *signal_cachep;
88
89/* SLAB cache for sighand_struct structures (tsk->sighand) */
90kmem_cache_t *sighand_cachep;
91
92/* SLAB cache for files_struct structures (tsk->files) */
93kmem_cache_t *files_cachep;
94
95/* SLAB cache for fs_struct structures (tsk->fs) */
96kmem_cache_t *fs_cachep;
97
98/* SLAB cache for vm_area_struct structures */
99kmem_cache_t *vm_area_cachep;
100
101/* SLAB cache for mm_struct structures (tsk->mm) */
102static kmem_cache_t *mm_cachep;
103
104void free_task(struct task_struct *tsk)
105{
106 free_thread_info(tsk->thread_info);
107 free_task_struct(tsk);
108}
109EXPORT_SYMBOL(free_task);
110
111void __put_task_struct(struct task_struct *tsk)
112{
113 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
114 WARN_ON(atomic_read(&tsk->usage));
115 WARN_ON(tsk == current);
116
117 security_task_free(tsk);
118 free_uid(tsk->user);
119 put_group_info(tsk->group_info);
120
121 if (!profile_handoff_task(tsk))
122 free_task(tsk);
123}
124
125void __init fork_init(unsigned long mempages)
126{
127#ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
128#ifndef ARCH_MIN_TASKALIGN
129#define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
130#endif
131 /* create a slab on which task_structs can be allocated */
132 task_struct_cachep =
133 kmem_cache_create("task_struct", sizeof(struct task_struct),
134 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
135#endif
136
137 /*
138 * The default maximum number of threads is set to a safe
139 * value: the thread structures can take up at most half
140 * of memory.
141 */
142 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
143
144 /*
145 * we need to allow at least 20 threads to boot a system
146 */
147 if(max_threads < 20)
148 max_threads = 20;
149
150 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
151 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
152 init_task.signal->rlim[RLIMIT_SIGPENDING] =
153 init_task.signal->rlim[RLIMIT_NPROC];
154}
155
156static struct task_struct *dup_task_struct(struct task_struct *orig)
157{
158 struct task_struct *tsk;
159 struct thread_info *ti;
160
161 prepare_to_copy(orig);
162
163 tsk = alloc_task_struct();
164 if (!tsk)
165 return NULL;
166
167 ti = alloc_thread_info(tsk);
168 if (!ti) {
169 free_task_struct(tsk);
170 return NULL;
171 }
172
173 *tsk = *orig;
174 tsk->thread_info = ti;
175 setup_thread_stack(tsk, orig);
176
177 /* One for us, one for whoever does the "release_task()" (usually parent) */
178 atomic_set(&tsk->usage,2);
179 atomic_set(&tsk->fs_excl, 0);
180 tsk->btrace_seq = 0;
181 tsk->splice_pipe = NULL;
182 return tsk;
183}
184
185#ifdef CONFIG_MMU
186static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
187{
188 struct vm_area_struct *mpnt, *tmp, **pprev;
189 struct rb_node **rb_link, *rb_parent;
190 int retval;
191 unsigned long charge;
192 struct mempolicy *pol;
193
194 down_write(&oldmm->mmap_sem);
195 flush_cache_mm(oldmm);
196 down_write(&mm->mmap_sem);
197
198 mm->locked_vm = 0;
199 mm->mmap = NULL;
200 mm->mmap_cache = NULL;
201 mm->free_area_cache = oldmm->mmap_base;
202 mm->cached_hole_size = ~0UL;
203 mm->map_count = 0;
204 cpus_clear(mm->cpu_vm_mask);
205 mm->mm_rb = RB_ROOT;
206 rb_link = &mm->mm_rb.rb_node;
207 rb_parent = NULL;
208 pprev = &mm->mmap;
209
210 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
211 struct file *file;
212
213 if (mpnt->vm_flags & VM_DONTCOPY) {
214 long pages = vma_pages(mpnt);
215 mm->total_vm -= pages;
216 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
217 -pages);
218 continue;
219 }
220 charge = 0;
221 if (mpnt->vm_flags & VM_ACCOUNT) {
222 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
223 if (security_vm_enough_memory(len))
224 goto fail_nomem;
225 charge = len;
226 }
227 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
228 if (!tmp)
229 goto fail_nomem;
230 *tmp = *mpnt;
231 pol = mpol_copy(vma_policy(mpnt));
232 retval = PTR_ERR(pol);
233 if (IS_ERR(pol))
234 goto fail_nomem_policy;
235 vma_set_policy(tmp, pol);
236 tmp->vm_flags &= ~VM_LOCKED;
237 tmp->vm_mm = mm;
238 tmp->vm_next = NULL;
239 anon_vma_link(tmp);
240 file = tmp->vm_file;
241 if (file) {
242 struct inode *inode = file->f_dentry->d_inode;
243 get_file(file);
244 if (tmp->vm_flags & VM_DENYWRITE)
245 atomic_dec(&inode->i_writecount);
246
247 /* insert tmp into the share list, just after mpnt */
248 spin_lock(&file->f_mapping->i_mmap_lock);
249 tmp->vm_truncate_count = mpnt->vm_truncate_count;
250 flush_dcache_mmap_lock(file->f_mapping);
251 vma_prio_tree_add(tmp, mpnt);
252 flush_dcache_mmap_unlock(file->f_mapping);
253 spin_unlock(&file->f_mapping->i_mmap_lock);
254 }
255
256 /*
257 * Link in the new vma and copy the page table entries.
258 */
259 *pprev = tmp;
260 pprev = &tmp->vm_next;
261
262 __vma_link_rb(mm, tmp, rb_link, rb_parent);
263 rb_link = &tmp->vm_rb.rb_right;
264 rb_parent = &tmp->vm_rb;
265
266 mm->map_count++;
267 retval = copy_page_range(mm, oldmm, mpnt);
268
269 if (tmp->vm_ops && tmp->vm_ops->open)
270 tmp->vm_ops->open(tmp);
271
272 if (retval)
273 goto out;
274 }
275 retval = 0;
276out:
277 up_write(&mm->mmap_sem);
278 flush_tlb_mm(oldmm);
279 up_write(&oldmm->mmap_sem);
280 return retval;
281fail_nomem_policy:
282 kmem_cache_free(vm_area_cachep, tmp);
283fail_nomem:
284 retval = -ENOMEM;
285 vm_unacct_memory(charge);
286 goto out;
287}
288
289static inline int mm_alloc_pgd(struct mm_struct * mm)
290{
291 mm->pgd = pgd_alloc(mm);
292 if (unlikely(!mm->pgd))
293 return -ENOMEM;
294 return 0;
295}
296
297static inline void mm_free_pgd(struct mm_struct * mm)
298{
299 pgd_free(mm->pgd);
300}
301#else
302#define dup_mmap(mm, oldmm) (0)
303#define mm_alloc_pgd(mm) (0)
304#define mm_free_pgd(mm)
305#endif /* CONFIG_MMU */
306
307 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
308
309#define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
310#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
311
312#include <linux/init_task.h>
313
314static struct mm_struct * mm_init(struct mm_struct * mm)
315{
316 atomic_set(&mm->mm_users, 1);
317 atomic_set(&mm->mm_count, 1);
318 init_rwsem(&mm->mmap_sem);
319 INIT_LIST_HEAD(&mm->mmlist);
320 mm->core_waiters = 0;
321 mm->nr_ptes = 0;
322 set_mm_counter(mm, file_rss, 0);
323 set_mm_counter(mm, anon_rss, 0);
324 spin_lock_init(&mm->page_table_lock);
325 rwlock_init(&mm->ioctx_list_lock);
326 mm->ioctx_list = NULL;
327 mm->free_area_cache = TASK_UNMAPPED_BASE;
328 mm->cached_hole_size = ~0UL;
329
330 if (likely(!mm_alloc_pgd(mm))) {
331 mm->def_flags = 0;
332 return mm;
333 }
334 free_mm(mm);
335 return NULL;
336}
337
338/*
339 * Allocate and initialize an mm_struct.
340 */
341struct mm_struct * mm_alloc(void)
342{
343 struct mm_struct * mm;
344
345 mm = allocate_mm();
346 if (mm) {
347 memset(mm, 0, sizeof(*mm));
348 mm = mm_init(mm);
349 }
350 return mm;
351}
352
353/*
354 * Called when the last reference to the mm
355 * is dropped: either by a lazy thread or by
356 * mmput. Free the page directory and the mm.
357 */
358void fastcall __mmdrop(struct mm_struct *mm)
359{
360 BUG_ON(mm == &init_mm);
361 mm_free_pgd(mm);
362 destroy_context(mm);
363 free_mm(mm);
364}
365
366/*
367 * Decrement the use count and release all resources for an mm.
368 */
369void mmput(struct mm_struct *mm)
370{
371 if (atomic_dec_and_test(&mm->mm_users)) {
372 exit_aio(mm);
373 exit_mmap(mm);
374 if (!list_empty(&mm->mmlist)) {
375 spin_lock(&mmlist_lock);
376 list_del(&mm->mmlist);
377 spin_unlock(&mmlist_lock);
378 }
379 put_swap_token(mm);
380 mmdrop(mm);
381 }
382}
383EXPORT_SYMBOL_GPL(mmput);
384
385/**
386 * get_task_mm - acquire a reference to the task's mm
387 *
388 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
389 * this kernel workthread has transiently adopted a user mm with use_mm,
390 * to do its AIO) is not set and if so returns a reference to it, after
391 * bumping up the use count. User must release the mm via mmput()
392 * after use. Typically used by /proc and ptrace.
393 */
394struct mm_struct *get_task_mm(struct task_struct *task)
395{
396 struct mm_struct *mm;
397
398 task_lock(task);
399 mm = task->mm;
400 if (mm) {
401 if (task->flags & PF_BORROWED_MM)
402 mm = NULL;
403 else
404 atomic_inc(&mm->mm_users);
405 }
406 task_unlock(task);
407 return mm;
408}
409EXPORT_SYMBOL_GPL(get_task_mm);
410
411/* Please note the differences between mmput and mm_release.
412 * mmput is called whenever we stop holding onto a mm_struct,
413 * error success whatever.
414 *
415 * mm_release is called after a mm_struct has been removed
416 * from the current process.
417 *
418 * This difference is important for error handling, when we
419 * only half set up a mm_struct for a new process and need to restore
420 * the old one. Because we mmput the new mm_struct before
421 * restoring the old one. . .
422 * Eric Biederman 10 January 1998
423 */
424void mm_release(struct task_struct *tsk, struct mm_struct *mm)
425{
426 struct completion *vfork_done = tsk->vfork_done;
427
428 /* Get rid of any cached register state */
429 deactivate_mm(tsk, mm);
430
431 /* notify parent sleeping on vfork() */
432 if (vfork_done) {
433 tsk->vfork_done = NULL;
434 complete(vfork_done);
435 }
436 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
437 u32 __user * tidptr = tsk->clear_child_tid;
438 tsk->clear_child_tid = NULL;
439
440 /*
441 * We don't check the error code - if userspace has
442 * not set up a proper pointer then tough luck.
443 */
444 put_user(0, tidptr);
445 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
446 }
447}
448
449/*
450 * Allocate a new mm structure and copy contents from the
451 * mm structure of the passed in task structure.
452 */
453static struct mm_struct *dup_mm(struct task_struct *tsk)
454{
455 struct mm_struct *mm, *oldmm = current->mm;
456 int err;
457
458 if (!oldmm)
459 return NULL;
460
461 mm = allocate_mm();
462 if (!mm)
463 goto fail_nomem;
464
465 memcpy(mm, oldmm, sizeof(*mm));
466
467 if (!mm_init(mm))
468 goto fail_nomem;
469
470 if (init_new_context(tsk, mm))
471 goto fail_nocontext;
472
473 err = dup_mmap(mm, oldmm);
474 if (err)
475 goto free_pt;
476
477 mm->hiwater_rss = get_mm_rss(mm);
478 mm->hiwater_vm = mm->total_vm;
479
480 return mm;
481
482free_pt:
483 mmput(mm);
484
485fail_nomem:
486 return NULL;
487
488fail_nocontext:
489 /*
490 * If init_new_context() failed, we cannot use mmput() to free the mm
491 * because it calls destroy_context()
492 */
493 mm_free_pgd(mm);
494 free_mm(mm);
495 return NULL;
496}
497
498static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
499{
500 struct mm_struct * mm, *oldmm;
501 int retval;
502
503 tsk->min_flt = tsk->maj_flt = 0;
504 tsk->nvcsw = tsk->nivcsw = 0;
505
506 tsk->mm = NULL;
507 tsk->active_mm = NULL;
508
509 /*
510 * Are we cloning a kernel thread?
511 *
512 * We need to steal a active VM for that..
513 */
514 oldmm = current->mm;
515 if (!oldmm)
516 return 0;
517
518 if (clone_flags & CLONE_VM) {
519 atomic_inc(&oldmm->mm_users);
520 mm = oldmm;
521 goto good_mm;
522 }
523
524 retval = -ENOMEM;
525 mm = dup_mm(tsk);
526 if (!mm)
527 goto fail_nomem;
528
529good_mm:
530 tsk->mm = mm;
531 tsk->active_mm = mm;
532 return 0;
533
534fail_nomem:
535 return retval;
536}
537
538static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
539{
540 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
541 /* We don't need to lock fs - think why ;-) */
542 if (fs) {
543 atomic_set(&fs->count, 1);
544 rwlock_init(&fs->lock);
545 fs->umask = old->umask;
546 read_lock(&old->lock);
547 fs->rootmnt = mntget(old->rootmnt);
548 fs->root = dget(old->root);
549 fs->pwdmnt = mntget(old->pwdmnt);
550 fs->pwd = dget(old->pwd);
551 if (old->altroot) {
552 fs->altrootmnt = mntget(old->altrootmnt);
553 fs->altroot = dget(old->altroot);
554 } else {
555 fs->altrootmnt = NULL;
556 fs->altroot = NULL;
557 }
558 read_unlock(&old->lock);
559 }
560 return fs;
561}
562
563struct fs_struct *copy_fs_struct(struct fs_struct *old)
564{
565 return __copy_fs_struct(old);
566}
567
568EXPORT_SYMBOL_GPL(copy_fs_struct);
569
570static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
571{
572 if (clone_flags & CLONE_FS) {
573 atomic_inc(&current->fs->count);
574 return 0;
575 }
576 tsk->fs = __copy_fs_struct(current->fs);
577 if (!tsk->fs)
578 return -ENOMEM;
579 return 0;
580}
581
582static int count_open_files(struct fdtable *fdt)
583{
584 int size = fdt->max_fdset;
585 int i;
586
587 /* Find the last open fd */
588 for (i = size/(8*sizeof(long)); i > 0; ) {
589 if (fdt->open_fds->fds_bits[--i])
590 break;
591 }
592 i = (i+1) * 8 * sizeof(long);
593 return i;
594}
595
596static struct files_struct *alloc_files(void)
597{
598 struct files_struct *newf;
599 struct fdtable *fdt;
600
601 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
602 if (!newf)
603 goto out;
604
605 atomic_set(&newf->count, 1);
606
607 spin_lock_init(&newf->file_lock);
608 newf->next_fd = 0;
609 fdt = &newf->fdtab;
610 fdt->max_fds = NR_OPEN_DEFAULT;
611 fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
612 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
613 fdt->open_fds = (fd_set *)&newf->open_fds_init;
614 fdt->fd = &newf->fd_array[0];
615 INIT_RCU_HEAD(&fdt->rcu);
616 fdt->free_files = NULL;
617 fdt->next = NULL;
618 rcu_assign_pointer(newf->fdt, fdt);
619out:
620 return newf;
621}
622
623/*
624 * Allocate a new files structure and copy contents from the
625 * passed in files structure.
626 */
627static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
628{
629 struct files_struct *newf;
630 struct file **old_fds, **new_fds;
631 int open_files, size, i, expand;
632 struct fdtable *old_fdt, *new_fdt;
633
634 newf = alloc_files();
635 if (!newf)
636 goto out;
637
638 spin_lock(&oldf->file_lock);
639 old_fdt = files_fdtable(oldf);
640 new_fdt = files_fdtable(newf);
641 size = old_fdt->max_fdset;
642 open_files = count_open_files(old_fdt);
643 expand = 0;
644
645 /*
646 * Check whether we need to allocate a larger fd array or fd set.
647 * Note: we're not a clone task, so the open count won't change.
648 */
649 if (open_files > new_fdt->max_fdset) {
650 new_fdt->max_fdset = 0;
651 expand = 1;
652 }
653 if (open_files > new_fdt->max_fds) {
654 new_fdt->max_fds = 0;
655 expand = 1;
656 }
657
658 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
659 if (expand) {
660 spin_unlock(&oldf->file_lock);
661 spin_lock(&newf->file_lock);
662 *errorp = expand_files(newf, open_files-1);
663 spin_unlock(&newf->file_lock);
664 if (*errorp < 0)
665 goto out_release;
666 new_fdt = files_fdtable(newf);
667 /*
668 * Reacquire the oldf lock and a pointer to its fd table
669 * who knows it may have a new bigger fd table. We need
670 * the latest pointer.
671 */
672 spin_lock(&oldf->file_lock);
673 old_fdt = files_fdtable(oldf);
674 }
675
676 old_fds = old_fdt->fd;
677 new_fds = new_fdt->fd;
678
679 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
680 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
681
682 for (i = open_files; i != 0; i--) {
683 struct file *f = *old_fds++;
684 if (f) {
685 get_file(f);
686 } else {
687 /*
688 * The fd may be claimed in the fd bitmap but not yet
689 * instantiated in the files array if a sibling thread
690 * is partway through open(). So make sure that this
691 * fd is available to the new process.
692 */
693 FD_CLR(open_files - i, new_fdt->open_fds);
694 }
695 rcu_assign_pointer(*new_fds++, f);
696 }
697 spin_unlock(&oldf->file_lock);
698
699 /* compute the remainder to be cleared */
700 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
701
702 /* This is long word aligned thus could use a optimized version */
703 memset(new_fds, 0, size);
704
705 if (new_fdt->max_fdset > open_files) {
706 int left = (new_fdt->max_fdset-open_files)/8;
707 int start = open_files / (8 * sizeof(unsigned long));
708
709 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
710 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
711 }
712
713out:
714 return newf;
715
716out_release:
717 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
718 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
719 free_fd_array(new_fdt->fd, new_fdt->max_fds);
720 kmem_cache_free(files_cachep, newf);
721 return NULL;
722}
723
724static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
725{
726 struct files_struct *oldf, *newf;
727 int error = 0;
728
729 /*
730 * A background process may not have any files ...
731 */
732 oldf = current->files;
733 if (!oldf)
734 goto out;
735
736 if (clone_flags & CLONE_FILES) {
737 atomic_inc(&oldf->count);
738 goto out;
739 }
740
741 /*
742 * Note: we may be using current for both targets (See exec.c)
743 * This works because we cache current->files (old) as oldf. Don't
744 * break this.
745 */
746 tsk->files = NULL;
747 error = -ENOMEM;
748 newf = dup_fd(oldf, &error);
749 if (!newf)
750 goto out;
751
752 tsk->files = newf;
753 error = 0;
754out:
755 return error;
756}
757
758/*
759 * Helper to unshare the files of the current task.
760 * We don't want to expose copy_files internals to
761 * the exec layer of the kernel.
762 */
763
764int unshare_files(void)
765{
766 struct files_struct *files = current->files;
767 int rc;
768
769 BUG_ON(!files);
770
771 /* This can race but the race causes us to copy when we don't
772 need to and drop the copy */
773 if(atomic_read(&files->count) == 1)
774 {
775 atomic_inc(&files->count);
776 return 0;
777 }
778 rc = copy_files(0, current);
779 if(rc)
780 current->files = files;
781 return rc;
782}
783
784EXPORT_SYMBOL(unshare_files);
785
786static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
787{
788 struct sighand_struct *sig;
789
790 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
791 atomic_inc(&current->sighand->count);
792 return 0;
793 }
794 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
795 rcu_assign_pointer(tsk->sighand, sig);
796 if (!sig)
797 return -ENOMEM;
798 atomic_set(&sig->count, 1);
799 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
800 return 0;
801}
802
803void __cleanup_sighand(struct sighand_struct *sighand)
804{
805 if (atomic_dec_and_test(&sighand->count))
806 kmem_cache_free(sighand_cachep, sighand);
807}
808
809static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
810{
811 struct signal_struct *sig;
812 int ret;
813
814 if (clone_flags & CLONE_THREAD) {
815 atomic_inc(&current->signal->count);
816 atomic_inc(&current->signal->live);
817 return 0;
818 }
819 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
820 tsk->signal = sig;
821 if (!sig)
822 return -ENOMEM;
823
824 ret = copy_thread_group_keys(tsk);
825 if (ret < 0) {
826 kmem_cache_free(signal_cachep, sig);
827 return ret;
828 }
829
830 atomic_set(&sig->count, 1);
831 atomic_set(&sig->live, 1);
832 init_waitqueue_head(&sig->wait_chldexit);
833 sig->flags = 0;
834 sig->group_exit_code = 0;
835 sig->group_exit_task = NULL;
836 sig->group_stop_count = 0;
837 sig->curr_target = NULL;
838 init_sigpending(&sig->shared_pending);
839 INIT_LIST_HEAD(&sig->posix_timers);
840
841 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
842 sig->it_real_incr.tv64 = 0;
843 sig->real_timer.function = it_real_fn;
844 sig->tsk = tsk;
845
846 sig->it_virt_expires = cputime_zero;
847 sig->it_virt_incr = cputime_zero;
848 sig->it_prof_expires = cputime_zero;
849 sig->it_prof_incr = cputime_zero;
850
851 sig->leader = 0; /* session leadership doesn't inherit */
852 sig->tty_old_pgrp = 0;
853
854 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
855 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
856 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
857 sig->sched_time = 0;
858 INIT_LIST_HEAD(&sig->cpu_timers[0]);
859 INIT_LIST_HEAD(&sig->cpu_timers[1]);
860 INIT_LIST_HEAD(&sig->cpu_timers[2]);
861
862 task_lock(current->group_leader);
863 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
864 task_unlock(current->group_leader);
865
866 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
867 /*
868 * New sole thread in the process gets an expiry time
869 * of the whole CPU time limit.
870 */
871 tsk->it_prof_expires =
872 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
873 }
874
875 return 0;
876}
877
878void __cleanup_signal(struct signal_struct *sig)
879{
880 exit_thread_group_keys(sig);
881 kmem_cache_free(signal_cachep, sig);
882}
883
884static inline void cleanup_signal(struct task_struct *tsk)
885{
886 struct signal_struct *sig = tsk->signal;
887
888 atomic_dec(&sig->live);
889
890 if (atomic_dec_and_test(&sig->count))
891 __cleanup_signal(sig);
892}
893
894static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
895{
896 unsigned long new_flags = p->flags;
897
898 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
899 new_flags |= PF_FORKNOEXEC;
900 if (!(clone_flags & CLONE_PTRACE))
901 p->ptrace = 0;
902 p->flags = new_flags;
903}
904
905asmlinkage long sys_set_tid_address(int __user *tidptr)
906{
907 current->clear_child_tid = tidptr;
908
909 return current->pid;
910}
911
912/*
913 * This creates a new process as a copy of the old one,
914 * but does not actually start it yet.
915 *
916 * It copies the registers, and all the appropriate
917 * parts of the process environment (as per the clone
918 * flags). The actual kick-off is left to the caller.
919 */
920static task_t *copy_process(unsigned long clone_flags,
921 unsigned long stack_start,
922 struct pt_regs *regs,
923 unsigned long stack_size,
924 int __user *parent_tidptr,
925 int __user *child_tidptr,
926 int pid)
927{
928 int retval;
929 struct task_struct *p = NULL;
930
931 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
932 return ERR_PTR(-EINVAL);
933
934 /*
935 * Thread groups must share signals as well, and detached threads
936 * can only be started up within the thread group.
937 */
938 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
939 return ERR_PTR(-EINVAL);
940
941 /*
942 * Shared signal handlers imply shared VM. By way of the above,
943 * thread groups also imply shared VM. Blocking this case allows
944 * for various simplifications in other code.
945 */
946 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
947 return ERR_PTR(-EINVAL);
948
949 retval = security_task_create(clone_flags);
950 if (retval)
951 goto fork_out;
952
953 retval = -ENOMEM;
954 p = dup_task_struct(current);
955 if (!p)
956 goto fork_out;
957
958 retval = -EAGAIN;
959 if (atomic_read(&p->user->processes) >=
960 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
961 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
962 p->user != &root_user)
963 goto bad_fork_free;
964 }
965
966 atomic_inc(&p->user->__count);
967 atomic_inc(&p->user->processes);
968 get_group_info(p->group_info);
969
970 /*
971 * If multiple threads are within copy_process(), then this check
972 * triggers too late. This doesn't hurt, the check is only there
973 * to stop root fork bombs.
974 */
975 if (nr_threads >= max_threads)
976 goto bad_fork_cleanup_count;
977
978 if (!try_module_get(task_thread_info(p)->exec_domain->module))
979 goto bad_fork_cleanup_count;
980
981 if (p->binfmt && !try_module_get(p->binfmt->module))
982 goto bad_fork_cleanup_put_domain;
983
984 p->did_exec = 0;
985 copy_flags(clone_flags, p);
986 p->pid = pid;
987 retval = -EFAULT;
988 if (clone_flags & CLONE_PARENT_SETTID)
989 if (put_user(p->pid, parent_tidptr))
990 goto bad_fork_cleanup;
991
992 p->proc_dentry = NULL;
993
994 INIT_LIST_HEAD(&p->children);
995 INIT_LIST_HEAD(&p->sibling);
996 p->vfork_done = NULL;
997 spin_lock_init(&p->alloc_lock);
998 spin_lock_init(&p->proc_lock);
999
1000 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1001 init_sigpending(&p->pending);
1002
1003 p->utime = cputime_zero;
1004 p->stime = cputime_zero;
1005 p->sched_time = 0;
1006 p->rchar = 0; /* I/O counter: bytes read */
1007 p->wchar = 0; /* I/O counter: bytes written */
1008 p->syscr = 0; /* I/O counter: read syscalls */
1009 p->syscw = 0; /* I/O counter: write syscalls */
1010 acct_clear_integrals(p);
1011
1012 p->it_virt_expires = cputime_zero;
1013 p->it_prof_expires = cputime_zero;
1014 p->it_sched_expires = 0;
1015 INIT_LIST_HEAD(&p->cpu_timers[0]);
1016 INIT_LIST_HEAD(&p->cpu_timers[1]);
1017 INIT_LIST_HEAD(&p->cpu_timers[2]);
1018
1019 p->lock_depth = -1; /* -1 = no lock */
1020 do_posix_clock_monotonic_gettime(&p->start_time);
1021 p->security = NULL;
1022 p->io_context = NULL;
1023 p->io_wait = NULL;
1024 p->audit_context = NULL;
1025 cpuset_fork(p);
1026#ifdef CONFIG_NUMA
1027 p->mempolicy = mpol_copy(p->mempolicy);
1028 if (IS_ERR(p->mempolicy)) {
1029 retval = PTR_ERR(p->mempolicy);
1030 p->mempolicy = NULL;
1031 goto bad_fork_cleanup_cpuset;
1032 }
1033 mpol_fix_fork_child_flag(p);
1034#endif
1035
1036#ifdef CONFIG_DEBUG_MUTEXES
1037 p->blocked_on = NULL; /* not blocked yet */
1038#endif
1039
1040 p->tgid = p->pid;
1041 if (clone_flags & CLONE_THREAD)
1042 p->tgid = current->tgid;
1043
1044 if ((retval = security_task_alloc(p)))
1045 goto bad_fork_cleanup_policy;
1046 if ((retval = audit_alloc(p)))
1047 goto bad_fork_cleanup_security;
1048 /* copy all the process information */
1049 if ((retval = copy_semundo(clone_flags, p)))
1050 goto bad_fork_cleanup_audit;
1051 if ((retval = copy_files(clone_flags, p)))
1052 goto bad_fork_cleanup_semundo;
1053 if ((retval = copy_fs(clone_flags, p)))
1054 goto bad_fork_cleanup_files;
1055 if ((retval = copy_sighand(clone_flags, p)))
1056 goto bad_fork_cleanup_fs;
1057 if ((retval = copy_signal(clone_flags, p)))
1058 goto bad_fork_cleanup_sighand;
1059 if ((retval = copy_mm(clone_flags, p)))
1060 goto bad_fork_cleanup_signal;
1061 if ((retval = copy_keys(clone_flags, p)))
1062 goto bad_fork_cleanup_mm;
1063 if ((retval = copy_namespace(clone_flags, p)))
1064 goto bad_fork_cleanup_keys;
1065 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1066 if (retval)
1067 goto bad_fork_cleanup_namespace;
1068
1069 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1070 /*
1071 * Clear TID on mm_release()?
1072 */
1073 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1074 p->robust_list = NULL;
1075#ifdef CONFIG_COMPAT
1076 p->compat_robust_list = NULL;
1077#endif
1078 /*
1079 * sigaltstack should be cleared when sharing the same VM
1080 */
1081 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1082 p->sas_ss_sp = p->sas_ss_size = 0;
1083
1084 /*
1085 * Syscall tracing should be turned off in the child regardless
1086 * of CLONE_PTRACE.
1087 */
1088 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1089#ifdef TIF_SYSCALL_EMU
1090 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1091#endif
1092
1093 /* Our parent execution domain becomes current domain
1094 These must match for thread signalling to apply */
1095
1096 p->parent_exec_id = p->self_exec_id;
1097
1098 /* ok, now we should be set up.. */
1099 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1100 p->pdeath_signal = 0;
1101 p->exit_state = 0;
1102
1103 /*
1104 * Ok, make it visible to the rest of the system.
1105 * We dont wake it up yet.
1106 */
1107 p->group_leader = p;
1108 INIT_LIST_HEAD(&p->thread_group);
1109 INIT_LIST_HEAD(&p->ptrace_children);
1110 INIT_LIST_HEAD(&p->ptrace_list);
1111
1112 /* Perform scheduler related setup. Assign this task to a CPU. */
1113 sched_fork(p, clone_flags);
1114
1115 /* Need tasklist lock for parent etc handling! */
1116 write_lock_irq(&tasklist_lock);
1117
1118 /*
1119 * The task hasn't been attached yet, so its cpus_allowed mask will
1120 * not be changed, nor will its assigned CPU.
1121 *
1122 * The cpus_allowed mask of the parent may have changed after it was
1123 * copied first time - so re-copy it here, then check the child's CPU
1124 * to ensure it is on a valid CPU (and if not, just force it back to
1125 * parent's CPU). This avoids alot of nasty races.
1126 */
1127 p->cpus_allowed = current->cpus_allowed;
1128 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1129 !cpu_online(task_cpu(p))))
1130 set_task_cpu(p, smp_processor_id());
1131
1132 /* CLONE_PARENT re-uses the old parent */
1133 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1134 p->real_parent = current->real_parent;
1135 else
1136 p->real_parent = current;
1137 p->parent = p->real_parent;
1138
1139 spin_lock(&current->sighand->siglock);
1140
1141 /*
1142 * Process group and session signals need to be delivered to just the
1143 * parent before the fork or both the parent and the child after the
1144 * fork. Restart if a signal comes in before we add the new process to
1145 * it's process group.
1146 * A fatal signal pending means that current will exit, so the new
1147 * thread can't slip out of an OOM kill (or normal SIGKILL).
1148 */
1149 recalc_sigpending();
1150 if (signal_pending(current)) {
1151 spin_unlock(&current->sighand->siglock);
1152 write_unlock_irq(&tasklist_lock);
1153 retval = -ERESTARTNOINTR;
1154 goto bad_fork_cleanup_namespace;
1155 }
1156
1157 if (clone_flags & CLONE_THREAD) {
1158 /*
1159 * Important: if an exit-all has been started then
1160 * do not create this new thread - the whole thread
1161 * group is supposed to exit anyway.
1162 */
1163 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1164 spin_unlock(&current->sighand->siglock);
1165 write_unlock_irq(&tasklist_lock);
1166 retval = -EAGAIN;
1167 goto bad_fork_cleanup_namespace;
1168 }
1169
1170 p->group_leader = current->group_leader;
1171 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1172
1173 if (!cputime_eq(current->signal->it_virt_expires,
1174 cputime_zero) ||
1175 !cputime_eq(current->signal->it_prof_expires,
1176 cputime_zero) ||
1177 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1178 !list_empty(&current->signal->cpu_timers[0]) ||
1179 !list_empty(&current->signal->cpu_timers[1]) ||
1180 !list_empty(&current->signal->cpu_timers[2])) {
1181 /*
1182 * Have child wake up on its first tick to check
1183 * for process CPU timers.
1184 */
1185 p->it_prof_expires = jiffies_to_cputime(1);
1186 }
1187 }
1188
1189 /*
1190 * inherit ioprio
1191 */
1192 p->ioprio = current->ioprio;
1193
1194 if (likely(p->pid)) {
1195 add_parent(p);
1196 if (unlikely(p->ptrace & PT_PTRACED))
1197 __ptrace_link(p, current->parent);
1198
1199 if (thread_group_leader(p)) {
1200 p->signal->tty = current->signal->tty;
1201 p->signal->pgrp = process_group(current);
1202 p->signal->session = current->signal->session;
1203 attach_pid(p, PIDTYPE_PGID, process_group(p));
1204 attach_pid(p, PIDTYPE_SID, p->signal->session);
1205
1206 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1207 __get_cpu_var(process_counts)++;
1208 }
1209 attach_pid(p, PIDTYPE_PID, p->pid);
1210 nr_threads++;
1211 }
1212
1213 total_forks++;
1214 spin_unlock(&current->sighand->siglock);
1215 write_unlock_irq(&tasklist_lock);
1216 proc_fork_connector(p);
1217 return p;
1218
1219bad_fork_cleanup_namespace:
1220 exit_namespace(p);
1221bad_fork_cleanup_keys:
1222 exit_keys(p);
1223bad_fork_cleanup_mm:
1224 if (p->mm)
1225 mmput(p->mm);
1226bad_fork_cleanup_signal:
1227 cleanup_signal(p);
1228bad_fork_cleanup_sighand:
1229 __cleanup_sighand(p->sighand);
1230bad_fork_cleanup_fs:
1231 exit_fs(p); /* blocking */
1232bad_fork_cleanup_files:
1233 exit_files(p); /* blocking */
1234bad_fork_cleanup_semundo:
1235 exit_sem(p);
1236bad_fork_cleanup_audit:
1237 audit_free(p);
1238bad_fork_cleanup_security:
1239 security_task_free(p);
1240bad_fork_cleanup_policy:
1241#ifdef CONFIG_NUMA
1242 mpol_free(p->mempolicy);
1243bad_fork_cleanup_cpuset:
1244#endif
1245 cpuset_exit(p);
1246bad_fork_cleanup:
1247 if (p->binfmt)
1248 module_put(p->binfmt->module);
1249bad_fork_cleanup_put_domain:
1250 module_put(task_thread_info(p)->exec_domain->module);
1251bad_fork_cleanup_count:
1252 put_group_info(p->group_info);
1253 atomic_dec(&p->user->processes);
1254 free_uid(p->user);
1255bad_fork_free:
1256 free_task(p);
1257fork_out:
1258 return ERR_PTR(retval);
1259}
1260
1261struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1262{
1263 memset(regs, 0, sizeof(struct pt_regs));
1264 return regs;
1265}
1266
1267task_t * __devinit fork_idle(int cpu)
1268{
1269 task_t *task;
1270 struct pt_regs regs;
1271
1272 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1273 if (!task)
1274 return ERR_PTR(-ENOMEM);
1275 init_idle(task, cpu);
1276
1277 return task;
1278}
1279
1280static inline int fork_traceflag (unsigned clone_flags)
1281{
1282 if (clone_flags & CLONE_UNTRACED)
1283 return 0;
1284 else if (clone_flags & CLONE_VFORK) {
1285 if (current->ptrace & PT_TRACE_VFORK)
1286 return PTRACE_EVENT_VFORK;
1287 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1288 if (current->ptrace & PT_TRACE_CLONE)
1289 return PTRACE_EVENT_CLONE;
1290 } else if (current->ptrace & PT_TRACE_FORK)
1291 return PTRACE_EVENT_FORK;
1292
1293 return 0;
1294}
1295
1296/*
1297 * Ok, this is the main fork-routine.
1298 *
1299 * It copies the process, and if successful kick-starts
1300 * it and waits for it to finish using the VM if required.
1301 */
1302long do_fork(unsigned long clone_flags,
1303 unsigned long stack_start,
1304 struct pt_regs *regs,
1305 unsigned long stack_size,
1306 int __user *parent_tidptr,
1307 int __user *child_tidptr)
1308{
1309 struct task_struct *p;
1310 int trace = 0;
1311 struct pid *pid = alloc_pid();
1312 long nr;
1313
1314 if (!pid)
1315 return -EAGAIN;
1316 nr = pid->nr;
1317 if (unlikely(current->ptrace)) {
1318 trace = fork_traceflag (clone_flags);
1319 if (trace)
1320 clone_flags |= CLONE_PTRACE;
1321 }
1322
1323 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1324 /*
1325 * Do this prior waking up the new thread - the thread pointer
1326 * might get invalid after that point, if the thread exits quickly.
1327 */
1328 if (!IS_ERR(p)) {
1329 struct completion vfork;
1330
1331 if (clone_flags & CLONE_VFORK) {
1332 p->vfork_done = &vfork;
1333 init_completion(&vfork);
1334 }
1335
1336 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1337 /*
1338 * We'll start up with an immediate SIGSTOP.
1339 */
1340 sigaddset(&p->pending.signal, SIGSTOP);
1341 set_tsk_thread_flag(p, TIF_SIGPENDING);
1342 }
1343
1344 if (!(clone_flags & CLONE_STOPPED))
1345 wake_up_new_task(p, clone_flags);
1346 else
1347 p->state = TASK_STOPPED;
1348
1349 if (unlikely (trace)) {
1350 current->ptrace_message = nr;
1351 ptrace_notify ((trace << 8) | SIGTRAP);
1352 }
1353
1354 if (clone_flags & CLONE_VFORK) {
1355 wait_for_completion(&vfork);
1356 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1357 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1358 }
1359 } else {
1360 free_pid(pid);
1361 nr = PTR_ERR(p);
1362 }
1363 return nr;
1364}
1365
1366#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1367#define ARCH_MIN_MMSTRUCT_ALIGN 0
1368#endif
1369
1370static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1371{
1372 struct sighand_struct *sighand = data;
1373
1374 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1375 SLAB_CTOR_CONSTRUCTOR)
1376 spin_lock_init(&sighand->siglock);
1377}
1378
1379void __init proc_caches_init(void)
1380{
1381 sighand_cachep = kmem_cache_create("sighand_cache",
1382 sizeof(struct sighand_struct), 0,
1383 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1384 sighand_ctor, NULL);
1385 signal_cachep = kmem_cache_create("signal_cache",
1386 sizeof(struct signal_struct), 0,
1387 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1388 files_cachep = kmem_cache_create("files_cache",
1389 sizeof(struct files_struct), 0,
1390 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1391 fs_cachep = kmem_cache_create("fs_cache",
1392 sizeof(struct fs_struct), 0,
1393 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1394 vm_area_cachep = kmem_cache_create("vm_area_struct",
1395 sizeof(struct vm_area_struct), 0,
1396 SLAB_PANIC, NULL, NULL);
1397 mm_cachep = kmem_cache_create("mm_struct",
1398 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1399 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1400}
1401
1402
1403/*
1404 * Check constraints on flags passed to the unshare system call and
1405 * force unsharing of additional process context as appropriate.
1406 */
1407static inline void check_unshare_flags(unsigned long *flags_ptr)
1408{
1409 /*
1410 * If unsharing a thread from a thread group, must also
1411 * unshare vm.
1412 */
1413 if (*flags_ptr & CLONE_THREAD)
1414 *flags_ptr |= CLONE_VM;
1415
1416 /*
1417 * If unsharing vm, must also unshare signal handlers.
1418 */
1419 if (*flags_ptr & CLONE_VM)
1420 *flags_ptr |= CLONE_SIGHAND;
1421
1422 /*
1423 * If unsharing signal handlers and the task was created
1424 * using CLONE_THREAD, then must unshare the thread
1425 */
1426 if ((*flags_ptr & CLONE_SIGHAND) &&
1427 (atomic_read(&current->signal->count) > 1))
1428 *flags_ptr |= CLONE_THREAD;
1429
1430 /*
1431 * If unsharing namespace, must also unshare filesystem information.
1432 */
1433 if (*flags_ptr & CLONE_NEWNS)
1434 *flags_ptr |= CLONE_FS;
1435}
1436
1437/*
1438 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1439 */
1440static int unshare_thread(unsigned long unshare_flags)
1441{
1442 if (unshare_flags & CLONE_THREAD)
1443 return -EINVAL;
1444
1445 return 0;
1446}
1447
1448/*
1449 * Unshare the filesystem structure if it is being shared
1450 */
1451static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1452{
1453 struct fs_struct *fs = current->fs;
1454
1455 if ((unshare_flags & CLONE_FS) &&
1456 (fs && atomic_read(&fs->count) > 1)) {
1457 *new_fsp = __copy_fs_struct(current->fs);
1458 if (!*new_fsp)
1459 return -ENOMEM;
1460 }
1461
1462 return 0;
1463}
1464
1465/*
1466 * Unshare the namespace structure if it is being shared
1467 */
1468static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1469{
1470 struct namespace *ns = current->namespace;
1471
1472 if ((unshare_flags & CLONE_NEWNS) &&
1473 (ns && atomic_read(&ns->count) > 1)) {
1474 if (!capable(CAP_SYS_ADMIN))
1475 return -EPERM;
1476
1477 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1478 if (!*new_nsp)
1479 return -ENOMEM;
1480 }
1481
1482 return 0;
1483}
1484
1485/*
1486 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1487 * supported yet
1488 */
1489static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1490{
1491 struct sighand_struct *sigh = current->sighand;
1492
1493 if ((unshare_flags & CLONE_SIGHAND) &&
1494 (sigh && atomic_read(&sigh->count) > 1))
1495 return -EINVAL;
1496 else
1497 return 0;
1498}
1499
1500/*
1501 * Unshare vm if it is being shared
1502 */
1503static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1504{
1505 struct mm_struct *mm = current->mm;
1506
1507 if ((unshare_flags & CLONE_VM) &&
1508 (mm && atomic_read(&mm->mm_users) > 1)) {
1509 return -EINVAL;
1510 }
1511
1512 return 0;
1513}
1514
1515/*
1516 * Unshare file descriptor table if it is being shared
1517 */
1518static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1519{
1520 struct files_struct *fd = current->files;
1521 int error = 0;
1522
1523 if ((unshare_flags & CLONE_FILES) &&
1524 (fd && atomic_read(&fd->count) > 1)) {
1525 *new_fdp = dup_fd(fd, &error);
1526 if (!*new_fdp)
1527 return error;
1528 }
1529
1530 return 0;
1531}
1532
1533/*
1534 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1535 * supported yet
1536 */
1537static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1538{
1539 if (unshare_flags & CLONE_SYSVSEM)
1540 return -EINVAL;
1541
1542 return 0;
1543}
1544
1545/*
1546 * unshare allows a process to 'unshare' part of the process
1547 * context which was originally shared using clone. copy_*
1548 * functions used by do_fork() cannot be used here directly
1549 * because they modify an inactive task_struct that is being
1550 * constructed. Here we are modifying the current, active,
1551 * task_struct.
1552 */
1553asmlinkage long sys_unshare(unsigned long unshare_flags)
1554{
1555 int err = 0;
1556 struct fs_struct *fs, *new_fs = NULL;
1557 struct namespace *ns, *new_ns = NULL;
1558 struct sighand_struct *sigh, *new_sigh = NULL;
1559 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1560 struct files_struct *fd, *new_fd = NULL;
1561 struct sem_undo_list *new_ulist = NULL;
1562
1563 check_unshare_flags(&unshare_flags);
1564
1565 /* Return -EINVAL for all unsupported flags */
1566 err = -EINVAL;
1567 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1568 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1569 goto bad_unshare_out;
1570
1571 if ((err = unshare_thread(unshare_flags)))
1572 goto bad_unshare_out;
1573 if ((err = unshare_fs(unshare_flags, &new_fs)))
1574 goto bad_unshare_cleanup_thread;
1575 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1576 goto bad_unshare_cleanup_fs;
1577 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1578 goto bad_unshare_cleanup_ns;
1579 if ((err = unshare_vm(unshare_flags, &new_mm)))
1580 goto bad_unshare_cleanup_sigh;
1581 if ((err = unshare_fd(unshare_flags, &new_fd)))
1582 goto bad_unshare_cleanup_vm;
1583 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1584 goto bad_unshare_cleanup_fd;
1585
1586 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1587
1588 task_lock(current);
1589
1590 if (new_fs) {
1591 fs = current->fs;
1592 current->fs = new_fs;
1593 new_fs = fs;
1594 }
1595
1596 if (new_ns) {
1597 ns = current->namespace;
1598 current->namespace = new_ns;
1599 new_ns = ns;
1600 }
1601
1602 if (new_sigh) {
1603 sigh = current->sighand;
1604 rcu_assign_pointer(current->sighand, new_sigh);
1605 new_sigh = sigh;
1606 }
1607
1608 if (new_mm) {
1609 mm = current->mm;
1610 active_mm = current->active_mm;
1611 current->mm = new_mm;
1612 current->active_mm = new_mm;
1613 activate_mm(active_mm, new_mm);
1614 new_mm = mm;
1615 }
1616
1617 if (new_fd) {
1618 fd = current->files;
1619 current->files = new_fd;
1620 new_fd = fd;
1621 }
1622
1623 task_unlock(current);
1624 }
1625
1626bad_unshare_cleanup_fd:
1627 if (new_fd)
1628 put_files_struct(new_fd);
1629
1630bad_unshare_cleanup_vm:
1631 if (new_mm)
1632 mmput(new_mm);
1633
1634bad_unshare_cleanup_sigh:
1635 if (new_sigh)
1636 if (atomic_dec_and_test(&new_sigh->count))
1637 kmem_cache_free(sighand_cachep, new_sigh);
1638
1639bad_unshare_cleanup_ns:
1640 if (new_ns)
1641 put_namespace(new_ns);
1642
1643bad_unshare_cleanup_fs:
1644 if (new_fs)
1645 put_fs_struct(new_fs);
1646
1647bad_unshare_cleanup_thread:
1648bad_unshare_out:
1649 return err;
1650}