ia64/linux-2.6.18-xen.hg

view kernel/fork.c @ 562:66faefe721eb

pvSCSI backend driver

Signed-off-by: Tomonari Horikoshi <t.horikoshi@jp.fujitsu.com>
Signed-off-by: Jun Kamada <kama@jp.fujitsu.com>
author Keir Fraser <keir.fraser@citrix.com>
date Mon Jun 02 09:58:27 2008 +0100 (2008-06-02)
parents a533be77c572
children
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/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/smp_lock.h>
18 #include <linux/module.h>
19 #include <linux/vmalloc.h>
20 #include <linux/completion.h>
21 #include <linux/namespace.h>
22 #include <linux/personality.h>
23 #include <linux/mempolicy.h>
24 #include <linux/sem.h>
25 #include <linux/file.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/fs.h>
30 #include <linux/capability.h>
31 #include <linux/cpu.h>
32 #include <linux/cpuset.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/rcupdate.h>
39 #include <linux/ptrace.h>
40 #include <linux/mount.h>
41 #include <linux/audit.h>
42 #include <linux/profile.h>
43 #include <linux/rmap.h>
44 #include <linux/acct.h>
45 #include <linux/cn_proc.h>
46 #include <linux/delayacct.h>
47 #include <linux/taskstats_kern.h>
49 #include <asm/pgtable.h>
50 #include <asm/pgalloc.h>
51 #include <asm/uaccess.h>
52 #include <asm/mmu_context.h>
53 #include <asm/cacheflush.h>
54 #include <asm/tlbflush.h>
56 /*
57 * Protected counters by write_lock_irq(&tasklist_lock)
58 */
59 unsigned long total_forks; /* Handle normal Linux uptimes. */
60 int nr_threads; /* The idle threads do not count.. */
62 int max_threads; /* tunable limit on nr_threads */
64 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
66 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
68 int nr_processes(void)
69 {
70 int cpu;
71 int total = 0;
73 for_each_online_cpu(cpu)
74 total += per_cpu(process_counts, cpu);
76 return total;
77 }
79 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
80 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
81 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
82 static kmem_cache_t *task_struct_cachep;
83 #endif
85 /* SLAB cache for signal_struct structures (tsk->signal) */
86 static kmem_cache_t *signal_cachep;
88 /* SLAB cache for sighand_struct structures (tsk->sighand) */
89 kmem_cache_t *sighand_cachep;
91 /* SLAB cache for files_struct structures (tsk->files) */
92 kmem_cache_t *files_cachep;
94 /* SLAB cache for fs_struct structures (tsk->fs) */
95 kmem_cache_t *fs_cachep;
97 /* SLAB cache for vm_area_struct structures */
98 kmem_cache_t *vm_area_cachep;
100 /* SLAB cache for mm_struct structures (tsk->mm) */
101 static kmem_cache_t *mm_cachep;
103 void free_task(struct task_struct *tsk)
104 {
105 free_thread_info(tsk->thread_info);
106 rt_mutex_debug_task_free(tsk);
107 free_task_struct(tsk);
108 }
109 EXPORT_SYMBOL(free_task);
111 void __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);
117 security_task_free(tsk);
118 free_uid(tsk->user);
119 put_group_info(tsk->group_info);
120 delayacct_tsk_free(tsk);
122 if (!profile_handoff_task(tsk))
123 free_task(tsk);
124 }
126 void __init fork_init(unsigned long mempages)
127 {
128 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
129 #ifndef ARCH_MIN_TASKALIGN
130 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
131 #endif
132 /* create a slab on which task_structs can be allocated */
133 task_struct_cachep =
134 kmem_cache_create("task_struct", sizeof(struct task_struct),
135 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
136 #endif
138 /*
139 * The default maximum number of threads is set to a safe
140 * value: the thread structures can take up at most half
141 * of memory.
142 */
143 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
145 /*
146 * we need to allow at least 20 threads to boot a system
147 */
148 if(max_threads < 20)
149 max_threads = 20;
151 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
152 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
153 init_task.signal->rlim[RLIMIT_SIGPENDING] =
154 init_task.signal->rlim[RLIMIT_NPROC];
155 }
157 static struct task_struct *dup_task_struct(struct task_struct *orig)
158 {
159 struct task_struct *tsk;
160 struct thread_info *ti;
162 prepare_to_copy(orig);
164 tsk = alloc_task_struct();
165 if (!tsk)
166 return NULL;
168 ti = alloc_thread_info(tsk);
169 if (!ti) {
170 free_task_struct(tsk);
171 return NULL;
172 }
174 *tsk = *orig;
175 tsk->thread_info = ti;
176 setup_thread_stack(tsk, orig);
178 /* One for us, one for whoever does the "release_task()" (usually parent) */
179 atomic_set(&tsk->usage,2);
180 atomic_set(&tsk->fs_excl, 0);
181 tsk->btrace_seq = 0;
182 tsk->splice_pipe = NULL;
183 return tsk;
184 }
186 #ifdef CONFIG_MMU
187 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
188 {
189 struct vm_area_struct *mpnt, *tmp, **pprev;
190 struct rb_node **rb_link, *rb_parent;
191 int retval;
192 unsigned long charge;
193 struct mempolicy *pol;
195 down_write(&oldmm->mmap_sem);
196 flush_cache_mm(oldmm);
197 /*
198 * Not linked in yet - no deadlock potential:
199 */
200 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
202 mm->locked_vm = 0;
203 mm->mmap = NULL;
204 mm->mmap_cache = NULL;
205 mm->free_area_cache = oldmm->mmap_base;
206 mm->cached_hole_size = ~0UL;
207 mm->map_count = 0;
208 cpus_clear(mm->cpu_vm_mask);
209 mm->mm_rb = RB_ROOT;
210 rb_link = &mm->mm_rb.rb_node;
211 rb_parent = NULL;
212 pprev = &mm->mmap;
214 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
215 struct file *file;
217 if (mpnt->vm_flags & VM_DONTCOPY) {
218 long pages = vma_pages(mpnt);
219 mm->total_vm -= pages;
220 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
221 -pages);
222 continue;
223 }
224 charge = 0;
225 if (mpnt->vm_flags & VM_ACCOUNT) {
226 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
227 if (security_vm_enough_memory(len))
228 goto fail_nomem;
229 charge = len;
230 }
231 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
232 if (!tmp)
233 goto fail_nomem;
234 *tmp = *mpnt;
235 pol = mpol_copy(vma_policy(mpnt));
236 retval = PTR_ERR(pol);
237 if (IS_ERR(pol))
238 goto fail_nomem_policy;
239 vma_set_policy(tmp, pol);
240 tmp->vm_flags &= ~VM_LOCKED;
241 tmp->vm_mm = mm;
242 tmp->vm_next = NULL;
243 anon_vma_link(tmp);
244 file = tmp->vm_file;
245 if (file) {
246 struct inode *inode = file->f_dentry->d_inode;
247 get_file(file);
248 if (tmp->vm_flags & VM_DENYWRITE)
249 atomic_dec(&inode->i_writecount);
251 /* insert tmp into the share list, just after mpnt */
252 spin_lock(&file->f_mapping->i_mmap_lock);
253 tmp->vm_truncate_count = mpnt->vm_truncate_count;
254 flush_dcache_mmap_lock(file->f_mapping);
255 vma_prio_tree_add(tmp, mpnt);
256 flush_dcache_mmap_unlock(file->f_mapping);
257 spin_unlock(&file->f_mapping->i_mmap_lock);
258 }
260 /*
261 * Link in the new vma and copy the page table entries.
262 */
263 *pprev = tmp;
264 pprev = &tmp->vm_next;
266 __vma_link_rb(mm, tmp, rb_link, rb_parent);
267 rb_link = &tmp->vm_rb.rb_right;
268 rb_parent = &tmp->vm_rb;
270 mm->map_count++;
271 retval = copy_page_range(mm, oldmm, mpnt);
273 if (tmp->vm_ops && tmp->vm_ops->open)
274 tmp->vm_ops->open(tmp);
276 if (retval)
277 goto out;
278 }
279 #ifdef arch_dup_mmap
280 arch_dup_mmap(mm, oldmm);
281 #endif
282 retval = 0;
283 out:
284 up_write(&mm->mmap_sem);
285 flush_tlb_mm(oldmm);
286 up_write(&oldmm->mmap_sem);
287 return retval;
288 fail_nomem_policy:
289 kmem_cache_free(vm_area_cachep, tmp);
290 fail_nomem:
291 retval = -ENOMEM;
292 vm_unacct_memory(charge);
293 goto out;
294 }
296 static inline int mm_alloc_pgd(struct mm_struct * mm)
297 {
298 mm->pgd = pgd_alloc(mm);
299 if (unlikely(!mm->pgd))
300 return -ENOMEM;
301 return 0;
302 }
304 static inline void mm_free_pgd(struct mm_struct * mm)
305 {
306 pgd_free(mm->pgd);
307 }
308 #else
309 #define dup_mmap(mm, oldmm) (0)
310 #define mm_alloc_pgd(mm) (0)
311 #define mm_free_pgd(mm)
312 #endif /* CONFIG_MMU */
314 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
316 #define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
317 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
319 #include <linux/init_task.h>
321 static struct mm_struct * mm_init(struct mm_struct * mm)
322 {
323 atomic_set(&mm->mm_users, 1);
324 atomic_set(&mm->mm_count, 1);
325 init_rwsem(&mm->mmap_sem);
326 INIT_LIST_HEAD(&mm->mmlist);
327 mm->core_waiters = 0;
328 mm->nr_ptes = 0;
329 set_mm_counter(mm, file_rss, 0);
330 set_mm_counter(mm, anon_rss, 0);
331 spin_lock_init(&mm->page_table_lock);
332 rwlock_init(&mm->ioctx_list_lock);
333 mm->ioctx_list = NULL;
334 mm->free_area_cache = TASK_UNMAPPED_BASE;
335 mm->cached_hole_size = ~0UL;
337 if (likely(!mm_alloc_pgd(mm))) {
338 mm->def_flags = 0;
339 return mm;
340 }
341 free_mm(mm);
342 return NULL;
343 }
345 /*
346 * Allocate and initialize an mm_struct.
347 */
348 struct mm_struct * mm_alloc(void)
349 {
350 struct mm_struct * mm;
352 mm = allocate_mm();
353 if (mm) {
354 memset(mm, 0, sizeof(*mm));
355 mm = mm_init(mm);
356 }
357 return mm;
358 }
360 /*
361 * Called when the last reference to the mm
362 * is dropped: either by a lazy thread or by
363 * mmput. Free the page directory and the mm.
364 */
365 void fastcall __mmdrop(struct mm_struct *mm)
366 {
367 BUG_ON(mm == &init_mm);
368 mm_free_pgd(mm);
369 destroy_context(mm);
370 free_mm(mm);
371 }
373 /*
374 * Decrement the use count and release all resources for an mm.
375 */
376 void mmput(struct mm_struct *mm)
377 {
378 might_sleep();
380 if (atomic_dec_and_test(&mm->mm_users)) {
381 exit_aio(mm);
382 exit_mmap(mm);
383 if (!list_empty(&mm->mmlist)) {
384 spin_lock(&mmlist_lock);
385 list_del(&mm->mmlist);
386 spin_unlock(&mmlist_lock);
387 }
388 put_swap_token(mm);
389 mmdrop(mm);
390 }
391 }
392 EXPORT_SYMBOL_GPL(mmput);
394 /**
395 * get_task_mm - acquire a reference to the task's mm
396 *
397 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
398 * this kernel workthread has transiently adopted a user mm with use_mm,
399 * to do its AIO) is not set and if so returns a reference to it, after
400 * bumping up the use count. User must release the mm via mmput()
401 * after use. Typically used by /proc and ptrace.
402 */
403 struct mm_struct *get_task_mm(struct task_struct *task)
404 {
405 struct mm_struct *mm;
407 task_lock(task);
408 mm = task->mm;
409 if (mm) {
410 if (task->flags & PF_BORROWED_MM)
411 mm = NULL;
412 else
413 atomic_inc(&mm->mm_users);
414 }
415 task_unlock(task);
416 return mm;
417 }
418 EXPORT_SYMBOL_GPL(get_task_mm);
420 /* Please note the differences between mmput and mm_release.
421 * mmput is called whenever we stop holding onto a mm_struct,
422 * error success whatever.
423 *
424 * mm_release is called after a mm_struct has been removed
425 * from the current process.
426 *
427 * This difference is important for error handling, when we
428 * only half set up a mm_struct for a new process and need to restore
429 * the old one. Because we mmput the new mm_struct before
430 * restoring the old one. . .
431 * Eric Biederman 10 January 1998
432 */
433 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
434 {
435 struct completion *vfork_done = tsk->vfork_done;
437 /* Get rid of any cached register state */
438 deactivate_mm(tsk, mm);
440 /* notify parent sleeping on vfork() */
441 if (vfork_done) {
442 tsk->vfork_done = NULL;
443 complete(vfork_done);
444 }
445 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
446 u32 __user * tidptr = tsk->clear_child_tid;
447 tsk->clear_child_tid = NULL;
449 /*
450 * We don't check the error code - if userspace has
451 * not set up a proper pointer then tough luck.
452 */
453 put_user(0, tidptr);
454 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
455 }
456 }
458 /*
459 * Allocate a new mm structure and copy contents from the
460 * mm structure of the passed in task structure.
461 */
462 static struct mm_struct *dup_mm(struct task_struct *tsk)
463 {
464 struct mm_struct *mm, *oldmm = current->mm;
465 int err;
467 if (!oldmm)
468 return NULL;
470 mm = allocate_mm();
471 if (!mm)
472 goto fail_nomem;
474 memcpy(mm, oldmm, sizeof(*mm));
476 if (!mm_init(mm))
477 goto fail_nomem;
479 if (init_new_context(tsk, mm))
480 goto fail_nocontext;
482 err = dup_mmap(mm, oldmm);
483 if (err)
484 goto free_pt;
486 mm->hiwater_rss = get_mm_rss(mm);
487 mm->hiwater_vm = mm->total_vm;
489 return mm;
491 free_pt:
492 mmput(mm);
494 fail_nomem:
495 return NULL;
497 fail_nocontext:
498 /*
499 * If init_new_context() failed, we cannot use mmput() to free the mm
500 * because it calls destroy_context()
501 */
502 mm_free_pgd(mm);
503 free_mm(mm);
504 return NULL;
505 }
507 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
508 {
509 struct mm_struct * mm, *oldmm;
510 int retval;
512 tsk->min_flt = tsk->maj_flt = 0;
513 tsk->nvcsw = tsk->nivcsw = 0;
515 tsk->mm = NULL;
516 tsk->active_mm = NULL;
518 /*
519 * Are we cloning a kernel thread?
520 *
521 * We need to steal a active VM for that..
522 */
523 oldmm = current->mm;
524 if (!oldmm)
525 return 0;
527 if (clone_flags & CLONE_VM) {
528 atomic_inc(&oldmm->mm_users);
529 mm = oldmm;
530 goto good_mm;
531 }
533 retval = -ENOMEM;
534 mm = dup_mm(tsk);
535 if (!mm)
536 goto fail_nomem;
538 good_mm:
539 tsk->mm = mm;
540 tsk->active_mm = mm;
541 return 0;
543 fail_nomem:
544 return retval;
545 }
547 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
548 {
549 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
550 /* We don't need to lock fs - think why ;-) */
551 if (fs) {
552 atomic_set(&fs->count, 1);
553 rwlock_init(&fs->lock);
554 fs->umask = old->umask;
555 read_lock(&old->lock);
556 fs->rootmnt = mntget(old->rootmnt);
557 fs->root = dget(old->root);
558 fs->pwdmnt = mntget(old->pwdmnt);
559 fs->pwd = dget(old->pwd);
560 if (old->altroot) {
561 fs->altrootmnt = mntget(old->altrootmnt);
562 fs->altroot = dget(old->altroot);
563 } else {
564 fs->altrootmnt = NULL;
565 fs->altroot = NULL;
566 }
567 read_unlock(&old->lock);
568 }
569 return fs;
570 }
572 struct fs_struct *copy_fs_struct(struct fs_struct *old)
573 {
574 return __copy_fs_struct(old);
575 }
577 EXPORT_SYMBOL_GPL(copy_fs_struct);
579 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
580 {
581 if (clone_flags & CLONE_FS) {
582 atomic_inc(&current->fs->count);
583 return 0;
584 }
585 tsk->fs = __copy_fs_struct(current->fs);
586 if (!tsk->fs)
587 return -ENOMEM;
588 return 0;
589 }
591 static int count_open_files(struct fdtable *fdt)
592 {
593 int size = fdt->max_fdset;
594 int i;
596 /* Find the last open fd */
597 for (i = size/(8*sizeof(long)); i > 0; ) {
598 if (fdt->open_fds->fds_bits[--i])
599 break;
600 }
601 i = (i+1) * 8 * sizeof(long);
602 return i;
603 }
605 static struct files_struct *alloc_files(void)
606 {
607 struct files_struct *newf;
608 struct fdtable *fdt;
610 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
611 if (!newf)
612 goto out;
614 atomic_set(&newf->count, 1);
616 spin_lock_init(&newf->file_lock);
617 newf->next_fd = 0;
618 fdt = &newf->fdtab;
619 fdt->max_fds = NR_OPEN_DEFAULT;
620 fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
621 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
622 fdt->open_fds = (fd_set *)&newf->open_fds_init;
623 fdt->fd = &newf->fd_array[0];
624 INIT_RCU_HEAD(&fdt->rcu);
625 fdt->free_files = NULL;
626 fdt->next = NULL;
627 rcu_assign_pointer(newf->fdt, fdt);
628 out:
629 return newf;
630 }
632 /*
633 * Allocate a new files structure and copy contents from the
634 * passed in files structure.
635 * errorp will be valid only when the returned files_struct is NULL.
636 */
637 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
638 {
639 struct files_struct *newf;
640 struct file **old_fds, **new_fds;
641 int open_files, size, i, expand;
642 struct fdtable *old_fdt, *new_fdt;
644 *errorp = -ENOMEM;
645 newf = alloc_files();
646 if (!newf)
647 goto out;
649 spin_lock(&oldf->file_lock);
650 old_fdt = files_fdtable(oldf);
651 new_fdt = files_fdtable(newf);
652 size = old_fdt->max_fdset;
653 open_files = count_open_files(old_fdt);
654 expand = 0;
656 /*
657 * Check whether we need to allocate a larger fd array or fd set.
658 * Note: we're not a clone task, so the open count won't change.
659 */
660 if (open_files > new_fdt->max_fdset) {
661 new_fdt->max_fdset = 0;
662 expand = 1;
663 }
664 if (open_files > new_fdt->max_fds) {
665 new_fdt->max_fds = 0;
666 expand = 1;
667 }
669 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
670 if (expand) {
671 spin_unlock(&oldf->file_lock);
672 spin_lock(&newf->file_lock);
673 *errorp = expand_files(newf, open_files-1);
674 spin_unlock(&newf->file_lock);
675 if (*errorp < 0)
676 goto out_release;
677 new_fdt = files_fdtable(newf);
678 /*
679 * Reacquire the oldf lock and a pointer to its fd table
680 * who knows it may have a new bigger fd table. We need
681 * the latest pointer.
682 */
683 spin_lock(&oldf->file_lock);
684 old_fdt = files_fdtable(oldf);
685 }
687 old_fds = old_fdt->fd;
688 new_fds = new_fdt->fd;
690 memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
691 memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
693 for (i = open_files; i != 0; i--) {
694 struct file *f = *old_fds++;
695 if (f) {
696 get_file(f);
697 } else {
698 /*
699 * The fd may be claimed in the fd bitmap but not yet
700 * instantiated in the files array if a sibling thread
701 * is partway through open(). So make sure that this
702 * fd is available to the new process.
703 */
704 FD_CLR(open_files - i, new_fdt->open_fds);
705 }
706 rcu_assign_pointer(*new_fds++, f);
707 }
708 spin_unlock(&oldf->file_lock);
710 /* compute the remainder to be cleared */
711 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
713 /* This is long word aligned thus could use a optimized version */
714 memset(new_fds, 0, size);
716 if (new_fdt->max_fdset > open_files) {
717 int left = (new_fdt->max_fdset-open_files)/8;
718 int start = open_files / (8 * sizeof(unsigned long));
720 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
721 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
722 }
724 out:
725 return newf;
727 out_release:
728 free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
729 free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
730 free_fd_array(new_fdt->fd, new_fdt->max_fds);
731 kmem_cache_free(files_cachep, newf);
732 return NULL;
733 }
735 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
736 {
737 struct files_struct *oldf, *newf;
738 int error = 0;
740 /*
741 * A background process may not have any files ...
742 */
743 oldf = current->files;
744 if (!oldf)
745 goto out;
747 if (clone_flags & CLONE_FILES) {
748 atomic_inc(&oldf->count);
749 goto out;
750 }
752 /*
753 * Note: we may be using current for both targets (See exec.c)
754 * This works because we cache current->files (old) as oldf. Don't
755 * break this.
756 */
757 tsk->files = NULL;
758 newf = dup_fd(oldf, &error);
759 if (!newf)
760 goto out;
762 tsk->files = newf;
763 error = 0;
764 out:
765 return error;
766 }
768 /*
769 * Helper to unshare the files of the current task.
770 * We don't want to expose copy_files internals to
771 * the exec layer of the kernel.
772 */
774 int unshare_files(void)
775 {
776 struct files_struct *files = current->files;
777 int rc;
779 BUG_ON(!files);
781 /* This can race but the race causes us to copy when we don't
782 need to and drop the copy */
783 if(atomic_read(&files->count) == 1)
784 {
785 atomic_inc(&files->count);
786 return 0;
787 }
788 rc = copy_files(0, current);
789 if(rc)
790 current->files = files;
791 return rc;
792 }
794 EXPORT_SYMBOL(unshare_files);
796 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
797 {
798 struct sighand_struct *sig;
800 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
801 atomic_inc(&current->sighand->count);
802 return 0;
803 }
804 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
805 rcu_assign_pointer(tsk->sighand, sig);
806 if (!sig)
807 return -ENOMEM;
808 atomic_set(&sig->count, 1);
809 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
810 return 0;
811 }
813 void __cleanup_sighand(struct sighand_struct *sighand)
814 {
815 if (atomic_dec_and_test(&sighand->count))
816 kmem_cache_free(sighand_cachep, sighand);
817 }
819 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
820 {
821 struct signal_struct *sig;
822 int ret;
824 if (clone_flags & CLONE_THREAD) {
825 atomic_inc(&current->signal->count);
826 atomic_inc(&current->signal->live);
827 taskstats_tgid_alloc(current->signal);
828 return 0;
829 }
830 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
831 tsk->signal = sig;
832 if (!sig)
833 return -ENOMEM;
835 ret = copy_thread_group_keys(tsk);
836 if (ret < 0) {
837 kmem_cache_free(signal_cachep, sig);
838 return ret;
839 }
841 atomic_set(&sig->count, 1);
842 atomic_set(&sig->live, 1);
843 init_waitqueue_head(&sig->wait_chldexit);
844 sig->flags = 0;
845 sig->group_exit_code = 0;
846 sig->group_exit_task = NULL;
847 sig->group_stop_count = 0;
848 sig->curr_target = NULL;
849 init_sigpending(&sig->shared_pending);
850 INIT_LIST_HEAD(&sig->posix_timers);
852 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
853 sig->it_real_incr.tv64 = 0;
854 sig->real_timer.function = it_real_fn;
855 sig->tsk = tsk;
857 sig->it_virt_expires = cputime_zero;
858 sig->it_virt_incr = cputime_zero;
859 sig->it_prof_expires = cputime_zero;
860 sig->it_prof_incr = cputime_zero;
862 sig->leader = 0; /* session leadership doesn't inherit */
863 sig->tty_old_pgrp = 0;
865 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
866 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
867 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
868 sig->sched_time = 0;
869 INIT_LIST_HEAD(&sig->cpu_timers[0]);
870 INIT_LIST_HEAD(&sig->cpu_timers[1]);
871 INIT_LIST_HEAD(&sig->cpu_timers[2]);
872 taskstats_tgid_init(sig);
874 task_lock(current->group_leader);
875 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
876 task_unlock(current->group_leader);
878 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
879 /*
880 * New sole thread in the process gets an expiry time
881 * of the whole CPU time limit.
882 */
883 tsk->it_prof_expires =
884 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
885 }
886 acct_init_pacct(&sig->pacct);
888 return 0;
889 }
891 void __cleanup_signal(struct signal_struct *sig)
892 {
893 exit_thread_group_keys(sig);
894 taskstats_tgid_free(sig);
895 kmem_cache_free(signal_cachep, sig);
896 }
898 static inline void cleanup_signal(struct task_struct *tsk)
899 {
900 struct signal_struct *sig = tsk->signal;
902 atomic_dec(&sig->live);
904 if (atomic_dec_and_test(&sig->count))
905 __cleanup_signal(sig);
906 }
908 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
909 {
910 unsigned long new_flags = p->flags;
912 new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
913 new_flags |= PF_FORKNOEXEC;
914 if (!(clone_flags & CLONE_PTRACE))
915 p->ptrace = 0;
916 p->flags = new_flags;
917 }
919 asmlinkage long sys_set_tid_address(int __user *tidptr)
920 {
921 current->clear_child_tid = tidptr;
923 return current->pid;
924 }
926 static inline void rt_mutex_init_task(struct task_struct *p)
927 {
928 #ifdef CONFIG_RT_MUTEXES
929 spin_lock_init(&p->pi_lock);
930 plist_head_init(&p->pi_waiters, &p->pi_lock);
931 p->pi_blocked_on = NULL;
932 #endif
933 }
935 /*
936 * This creates a new process as a copy of the old one,
937 * but does not actually start it yet.
938 *
939 * It copies the registers, and all the appropriate
940 * parts of the process environment (as per the clone
941 * flags). The actual kick-off is left to the caller.
942 */
943 static struct task_struct *copy_process(unsigned long clone_flags,
944 unsigned long stack_start,
945 struct pt_regs *regs,
946 unsigned long stack_size,
947 int __user *parent_tidptr,
948 int __user *child_tidptr,
949 int pid)
950 {
951 int retval;
952 struct task_struct *p = NULL;
954 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
955 return ERR_PTR(-EINVAL);
957 /*
958 * Thread groups must share signals as well, and detached threads
959 * can only be started up within the thread group.
960 */
961 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
962 return ERR_PTR(-EINVAL);
964 /*
965 * Shared signal handlers imply shared VM. By way of the above,
966 * thread groups also imply shared VM. Blocking this case allows
967 * for various simplifications in other code.
968 */
969 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
970 return ERR_PTR(-EINVAL);
972 retval = security_task_create(clone_flags);
973 if (retval)
974 goto fork_out;
976 retval = -ENOMEM;
977 p = dup_task_struct(current);
978 if (!p)
979 goto fork_out;
981 #ifdef CONFIG_TRACE_IRQFLAGS
982 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
983 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
984 #endif
985 retval = -EAGAIN;
986 if (atomic_read(&p->user->processes) >=
987 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
988 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
989 p->user != &root_user)
990 goto bad_fork_free;
991 }
993 atomic_inc(&p->user->__count);
994 atomic_inc(&p->user->processes);
995 get_group_info(p->group_info);
997 /*
998 * If multiple threads are within copy_process(), then this check
999 * triggers too late. This doesn't hurt, the check is only there
1000 * to stop root fork bombs.
1001 */
1002 if (nr_threads >= max_threads)
1003 goto bad_fork_cleanup_count;
1005 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1006 goto bad_fork_cleanup_count;
1008 if (p->binfmt && !try_module_get(p->binfmt->module))
1009 goto bad_fork_cleanup_put_domain;
1011 p->did_exec = 0;
1012 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1013 copy_flags(clone_flags, p);
1014 p->pid = pid;
1015 retval = -EFAULT;
1016 if (clone_flags & CLONE_PARENT_SETTID)
1017 if (put_user(p->pid, parent_tidptr))
1018 goto bad_fork_cleanup_delays_binfmt;
1020 INIT_LIST_HEAD(&p->children);
1021 INIT_LIST_HEAD(&p->sibling);
1022 p->vfork_done = NULL;
1023 spin_lock_init(&p->alloc_lock);
1025 clear_tsk_thread_flag(p, TIF_SIGPENDING);
1026 init_sigpending(&p->pending);
1028 p->utime = cputime_zero;
1029 p->stime = cputime_zero;
1030 p->sched_time = 0;
1031 p->rchar = 0; /* I/O counter: bytes read */
1032 p->wchar = 0; /* I/O counter: bytes written */
1033 p->syscr = 0; /* I/O counter: read syscalls */
1034 p->syscw = 0; /* I/O counter: write syscalls */
1035 acct_clear_integrals(p);
1037 p->it_virt_expires = cputime_zero;
1038 p->it_prof_expires = cputime_zero;
1039 p->it_sched_expires = 0;
1040 INIT_LIST_HEAD(&p->cpu_timers[0]);
1041 INIT_LIST_HEAD(&p->cpu_timers[1]);
1042 INIT_LIST_HEAD(&p->cpu_timers[2]);
1044 p->lock_depth = -1; /* -1 = no lock */
1045 do_posix_clock_monotonic_gettime(&p->start_time);
1046 p->security = NULL;
1047 p->io_context = NULL;
1048 p->io_wait = NULL;
1049 p->audit_context = NULL;
1050 cpuset_fork(p);
1051 #ifdef CONFIG_NUMA
1052 p->mempolicy = mpol_copy(p->mempolicy);
1053 if (IS_ERR(p->mempolicy)) {
1054 retval = PTR_ERR(p->mempolicy);
1055 p->mempolicy = NULL;
1056 goto bad_fork_cleanup_cpuset;
1058 mpol_fix_fork_child_flag(p);
1059 #endif
1060 #ifdef CONFIG_TRACE_IRQFLAGS
1061 p->irq_events = 0;
1062 p->hardirqs_enabled = 0;
1063 p->hardirq_enable_ip = 0;
1064 p->hardirq_enable_event = 0;
1065 p->hardirq_disable_ip = _THIS_IP_;
1066 p->hardirq_disable_event = 0;
1067 p->softirqs_enabled = 1;
1068 p->softirq_enable_ip = _THIS_IP_;
1069 p->softirq_enable_event = 0;
1070 p->softirq_disable_ip = 0;
1071 p->softirq_disable_event = 0;
1072 p->hardirq_context = 0;
1073 p->softirq_context = 0;
1074 #endif
1075 #ifdef CONFIG_LOCKDEP
1076 p->lockdep_depth = 0; /* no locks held yet */
1077 p->curr_chain_key = 0;
1078 p->lockdep_recursion = 0;
1079 #endif
1081 rt_mutex_init_task(p);
1083 #ifdef CONFIG_DEBUG_MUTEXES
1084 p->blocked_on = NULL; /* not blocked yet */
1085 #endif
1087 p->tgid = p->pid;
1088 if (clone_flags & CLONE_THREAD)
1089 p->tgid = current->tgid;
1091 if ((retval = security_task_alloc(p)))
1092 goto bad_fork_cleanup_policy;
1093 if ((retval = audit_alloc(p)))
1094 goto bad_fork_cleanup_security;
1095 /* copy all the process information */
1096 if ((retval = copy_semundo(clone_flags, p)))
1097 goto bad_fork_cleanup_audit;
1098 if ((retval = copy_files(clone_flags, p)))
1099 goto bad_fork_cleanup_semundo;
1100 if ((retval = copy_fs(clone_flags, p)))
1101 goto bad_fork_cleanup_files;
1102 if ((retval = copy_sighand(clone_flags, p)))
1103 goto bad_fork_cleanup_fs;
1104 if ((retval = copy_signal(clone_flags, p)))
1105 goto bad_fork_cleanup_sighand;
1106 if ((retval = copy_mm(clone_flags, p)))
1107 goto bad_fork_cleanup_signal;
1108 if ((retval = copy_keys(clone_flags, p)))
1109 goto bad_fork_cleanup_mm;
1110 if ((retval = copy_namespace(clone_flags, p)))
1111 goto bad_fork_cleanup_keys;
1112 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1113 if (retval)
1114 goto bad_fork_cleanup_namespace;
1116 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1117 /*
1118 * Clear TID on mm_release()?
1119 */
1120 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1121 p->robust_list = NULL;
1122 #ifdef CONFIG_COMPAT
1123 p->compat_robust_list = NULL;
1124 #endif
1125 INIT_LIST_HEAD(&p->pi_state_list);
1126 p->pi_state_cache = NULL;
1128 /*
1129 * sigaltstack should be cleared when sharing the same VM
1130 */
1131 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1132 p->sas_ss_sp = p->sas_ss_size = 0;
1134 /*
1135 * Syscall tracing should be turned off in the child regardless
1136 * of CLONE_PTRACE.
1137 */
1138 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1139 #ifdef TIF_SYSCALL_EMU
1140 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1141 #endif
1143 /* Our parent execution domain becomes current domain
1144 These must match for thread signalling to apply */
1146 p->parent_exec_id = p->self_exec_id;
1148 /* ok, now we should be set up.. */
1149 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1150 p->pdeath_signal = 0;
1151 p->exit_state = 0;
1153 /*
1154 * Ok, make it visible to the rest of the system.
1155 * We dont wake it up yet.
1156 */
1157 p->group_leader = p;
1158 INIT_LIST_HEAD(&p->thread_group);
1159 INIT_LIST_HEAD(&p->ptrace_children);
1160 INIT_LIST_HEAD(&p->ptrace_list);
1162 /* Perform scheduler related setup. Assign this task to a CPU. */
1163 sched_fork(p, clone_flags);
1165 /* Need tasklist lock for parent etc handling! */
1166 write_lock_irq(&tasklist_lock);
1168 /*
1169 * The task hasn't been attached yet, so its cpus_allowed mask will
1170 * not be changed, nor will its assigned CPU.
1172 * The cpus_allowed mask of the parent may have changed after it was
1173 * copied first time - so re-copy it here, then check the child's CPU
1174 * to ensure it is on a valid CPU (and if not, just force it back to
1175 * parent's CPU). This avoids alot of nasty races.
1176 */
1177 p->cpus_allowed = current->cpus_allowed;
1178 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1179 !cpu_online(task_cpu(p))))
1180 set_task_cpu(p, smp_processor_id());
1182 /* CLONE_PARENT re-uses the old parent */
1183 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1184 p->real_parent = current->real_parent;
1185 else
1186 p->real_parent = current;
1187 p->parent = p->real_parent;
1189 spin_lock(&current->sighand->siglock);
1191 /*
1192 * Process group and session signals need to be delivered to just the
1193 * parent before the fork or both the parent and the child after the
1194 * fork. Restart if a signal comes in before we add the new process to
1195 * it's process group.
1196 * A fatal signal pending means that current will exit, so the new
1197 * thread can't slip out of an OOM kill (or normal SIGKILL).
1198 */
1199 recalc_sigpending();
1200 if (signal_pending(current)) {
1201 spin_unlock(&current->sighand->siglock);
1202 write_unlock_irq(&tasklist_lock);
1203 retval = -ERESTARTNOINTR;
1204 goto bad_fork_cleanup_namespace;
1207 if (clone_flags & CLONE_THREAD) {
1208 p->group_leader = current->group_leader;
1209 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1211 if (!cputime_eq(current->signal->it_virt_expires,
1212 cputime_zero) ||
1213 !cputime_eq(current->signal->it_prof_expires,
1214 cputime_zero) ||
1215 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1216 !list_empty(&current->signal->cpu_timers[0]) ||
1217 !list_empty(&current->signal->cpu_timers[1]) ||
1218 !list_empty(&current->signal->cpu_timers[2])) {
1219 /*
1220 * Have child wake up on its first tick to check
1221 * for process CPU timers.
1222 */
1223 p->it_prof_expires = jiffies_to_cputime(1);
1227 /*
1228 * inherit ioprio
1229 */
1230 p->ioprio = current->ioprio;
1232 if (likely(p->pid)) {
1233 add_parent(p);
1234 if (unlikely(p->ptrace & PT_PTRACED))
1235 __ptrace_link(p, current->parent);
1237 if (thread_group_leader(p)) {
1238 p->signal->tty = current->signal->tty;
1239 p->signal->pgrp = process_group(current);
1240 p->signal->session = current->signal->session;
1241 attach_pid(p, PIDTYPE_PGID, process_group(p));
1242 attach_pid(p, PIDTYPE_SID, p->signal->session);
1244 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1245 __get_cpu_var(process_counts)++;
1247 attach_pid(p, PIDTYPE_PID, p->pid);
1248 nr_threads++;
1251 total_forks++;
1252 spin_unlock(&current->sighand->siglock);
1253 write_unlock_irq(&tasklist_lock);
1254 proc_fork_connector(p);
1255 return p;
1257 bad_fork_cleanup_namespace:
1258 exit_namespace(p);
1259 bad_fork_cleanup_keys:
1260 exit_keys(p);
1261 bad_fork_cleanup_mm:
1262 if (p->mm)
1263 mmput(p->mm);
1264 bad_fork_cleanup_signal:
1265 cleanup_signal(p);
1266 bad_fork_cleanup_sighand:
1267 __cleanup_sighand(p->sighand);
1268 bad_fork_cleanup_fs:
1269 exit_fs(p); /* blocking */
1270 bad_fork_cleanup_files:
1271 exit_files(p); /* blocking */
1272 bad_fork_cleanup_semundo:
1273 exit_sem(p);
1274 bad_fork_cleanup_audit:
1275 audit_free(p);
1276 bad_fork_cleanup_security:
1277 security_task_free(p);
1278 bad_fork_cleanup_policy:
1279 #ifdef CONFIG_NUMA
1280 mpol_free(p->mempolicy);
1281 bad_fork_cleanup_cpuset:
1282 #endif
1283 cpuset_exit(p);
1284 bad_fork_cleanup_delays_binfmt:
1285 delayacct_tsk_free(p);
1286 if (p->binfmt)
1287 module_put(p->binfmt->module);
1288 bad_fork_cleanup_put_domain:
1289 module_put(task_thread_info(p)->exec_domain->module);
1290 bad_fork_cleanup_count:
1291 put_group_info(p->group_info);
1292 atomic_dec(&p->user->processes);
1293 free_uid(p->user);
1294 bad_fork_free:
1295 free_task(p);
1296 fork_out:
1297 return ERR_PTR(retval);
1300 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1302 memset(regs, 0, sizeof(struct pt_regs));
1303 return regs;
1306 struct task_struct * __devinit fork_idle(int cpu)
1308 struct task_struct *task;
1309 struct pt_regs regs;
1311 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1312 if (!task)
1313 return ERR_PTR(-ENOMEM);
1314 init_idle(task, cpu);
1316 return task;
1319 static inline int fork_traceflag (unsigned clone_flags)
1321 if (clone_flags & CLONE_UNTRACED)
1322 return 0;
1323 else if (clone_flags & CLONE_VFORK) {
1324 if (current->ptrace & PT_TRACE_VFORK)
1325 return PTRACE_EVENT_VFORK;
1326 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1327 if (current->ptrace & PT_TRACE_CLONE)
1328 return PTRACE_EVENT_CLONE;
1329 } else if (current->ptrace & PT_TRACE_FORK)
1330 return PTRACE_EVENT_FORK;
1332 return 0;
1335 /*
1336 * Ok, this is the main fork-routine.
1338 * It copies the process, and if successful kick-starts
1339 * it and waits for it to finish using the VM if required.
1340 */
1341 long do_fork(unsigned long clone_flags,
1342 unsigned long stack_start,
1343 struct pt_regs *regs,
1344 unsigned long stack_size,
1345 int __user *parent_tidptr,
1346 int __user *child_tidptr)
1348 struct task_struct *p;
1349 int trace = 0;
1350 struct pid *pid = alloc_pid();
1351 long nr;
1353 if (!pid)
1354 return -EAGAIN;
1355 nr = pid->nr;
1356 if (unlikely(current->ptrace)) {
1357 trace = fork_traceflag (clone_flags);
1358 if (trace)
1359 clone_flags |= CLONE_PTRACE;
1362 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1363 /*
1364 * Do this prior waking up the new thread - the thread pointer
1365 * might get invalid after that point, if the thread exits quickly.
1366 */
1367 if (!IS_ERR(p)) {
1368 struct completion vfork;
1370 if (clone_flags & CLONE_VFORK) {
1371 p->vfork_done = &vfork;
1372 init_completion(&vfork);
1375 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1376 /*
1377 * We'll start up with an immediate SIGSTOP.
1378 */
1379 sigaddset(&p->pending.signal, SIGSTOP);
1380 set_tsk_thread_flag(p, TIF_SIGPENDING);
1383 if (!(clone_flags & CLONE_STOPPED))
1384 wake_up_new_task(p, clone_flags);
1385 else
1386 p->state = TASK_STOPPED;
1388 if (unlikely (trace)) {
1389 current->ptrace_message = nr;
1390 ptrace_notify ((trace << 8) | SIGTRAP);
1393 if (clone_flags & CLONE_VFORK) {
1394 wait_for_completion(&vfork);
1395 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1396 current->ptrace_message = nr;
1397 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1400 } else {
1401 free_pid(pid);
1402 nr = PTR_ERR(p);
1404 return nr;
1407 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1408 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1409 #endif
1411 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1413 struct sighand_struct *sighand = data;
1415 if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1416 SLAB_CTOR_CONSTRUCTOR)
1417 spin_lock_init(&sighand->siglock);
1420 void __init proc_caches_init(void)
1422 sighand_cachep = kmem_cache_create("sighand_cache",
1423 sizeof(struct sighand_struct), 0,
1424 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1425 sighand_ctor, NULL);
1426 signal_cachep = kmem_cache_create("signal_cache",
1427 sizeof(struct signal_struct), 0,
1428 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1429 files_cachep = kmem_cache_create("files_cache",
1430 sizeof(struct files_struct), 0,
1431 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1432 fs_cachep = kmem_cache_create("fs_cache",
1433 sizeof(struct fs_struct), 0,
1434 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1435 vm_area_cachep = kmem_cache_create("vm_area_struct",
1436 sizeof(struct vm_area_struct), 0,
1437 SLAB_PANIC, NULL, NULL);
1438 mm_cachep = kmem_cache_create("mm_struct",
1439 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1440 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1444 /*
1445 * Check constraints on flags passed to the unshare system call and
1446 * force unsharing of additional process context as appropriate.
1447 */
1448 static inline void check_unshare_flags(unsigned long *flags_ptr)
1450 /*
1451 * If unsharing a thread from a thread group, must also
1452 * unshare vm.
1453 */
1454 if (*flags_ptr & CLONE_THREAD)
1455 *flags_ptr |= CLONE_VM;
1457 /*
1458 * If unsharing vm, must also unshare signal handlers.
1459 */
1460 if (*flags_ptr & CLONE_VM)
1461 *flags_ptr |= CLONE_SIGHAND;
1463 /*
1464 * If unsharing signal handlers and the task was created
1465 * using CLONE_THREAD, then must unshare the thread
1466 */
1467 if ((*flags_ptr & CLONE_SIGHAND) &&
1468 (atomic_read(&current->signal->count) > 1))
1469 *flags_ptr |= CLONE_THREAD;
1471 /*
1472 * If unsharing namespace, must also unshare filesystem information.
1473 */
1474 if (*flags_ptr & CLONE_NEWNS)
1475 *flags_ptr |= CLONE_FS;
1478 /*
1479 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1480 */
1481 static int unshare_thread(unsigned long unshare_flags)
1483 if (unshare_flags & CLONE_THREAD)
1484 return -EINVAL;
1486 return 0;
1489 /*
1490 * Unshare the filesystem structure if it is being shared
1491 */
1492 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1494 struct fs_struct *fs = current->fs;
1496 if ((unshare_flags & CLONE_FS) &&
1497 (fs && atomic_read(&fs->count) > 1)) {
1498 *new_fsp = __copy_fs_struct(current->fs);
1499 if (!*new_fsp)
1500 return -ENOMEM;
1503 return 0;
1506 /*
1507 * Unshare the namespace structure if it is being shared
1508 */
1509 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1511 struct namespace *ns = current->namespace;
1513 if ((unshare_flags & CLONE_NEWNS) &&
1514 (ns && atomic_read(&ns->count) > 1)) {
1515 if (!capable(CAP_SYS_ADMIN))
1516 return -EPERM;
1518 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1519 if (!*new_nsp)
1520 return -ENOMEM;
1523 return 0;
1526 /*
1527 * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1528 * supported yet
1529 */
1530 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1532 struct sighand_struct *sigh = current->sighand;
1534 if ((unshare_flags & CLONE_SIGHAND) &&
1535 (sigh && atomic_read(&sigh->count) > 1))
1536 return -EINVAL;
1537 else
1538 return 0;
1541 /*
1542 * Unshare vm if it is being shared
1543 */
1544 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1546 struct mm_struct *mm = current->mm;
1548 if ((unshare_flags & CLONE_VM) &&
1549 (mm && atomic_read(&mm->mm_users) > 1)) {
1550 return -EINVAL;
1553 return 0;
1556 /*
1557 * Unshare file descriptor table if it is being shared
1558 */
1559 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1561 struct files_struct *fd = current->files;
1562 int error = 0;
1564 if ((unshare_flags & CLONE_FILES) &&
1565 (fd && atomic_read(&fd->count) > 1)) {
1566 *new_fdp = dup_fd(fd, &error);
1567 if (!*new_fdp)
1568 return error;
1571 return 0;
1574 /*
1575 * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1576 * supported yet
1577 */
1578 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1580 if (unshare_flags & CLONE_SYSVSEM)
1581 return -EINVAL;
1583 return 0;
1586 /*
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.
1593 */
1594 asmlinkage long sys_unshare(unsigned long unshare_flags)
1596 int err = 0;
1597 struct fs_struct *fs, *new_fs = NULL;
1598 struct namespace *ns, *new_ns = NULL;
1599 struct sighand_struct *sigh, *new_sigh = NULL;
1600 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1601 struct files_struct *fd, *new_fd = NULL;
1602 struct sem_undo_list *new_ulist = NULL;
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 goto bad_unshare_out;
1612 if ((err = unshare_thread(unshare_flags)))
1613 goto bad_unshare_out;
1614 if ((err = unshare_fs(unshare_flags, &new_fs)))
1615 goto bad_unshare_cleanup_thread;
1616 if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1617 goto bad_unshare_cleanup_fs;
1618 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1619 goto bad_unshare_cleanup_ns;
1620 if ((err = unshare_vm(unshare_flags, &new_mm)))
1621 goto bad_unshare_cleanup_sigh;
1622 if ((err = unshare_fd(unshare_flags, &new_fd)))
1623 goto bad_unshare_cleanup_vm;
1624 if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1625 goto bad_unshare_cleanup_fd;
1627 if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1629 task_lock(current);
1631 if (new_fs) {
1632 fs = current->fs;
1633 current->fs = new_fs;
1634 new_fs = fs;
1637 if (new_ns) {
1638 ns = current->namespace;
1639 current->namespace = new_ns;
1640 new_ns = ns;
1643 if (new_sigh) {
1644 sigh = current->sighand;
1645 rcu_assign_pointer(current->sighand, new_sigh);
1646 new_sigh = sigh;
1649 if (new_mm) {
1650 mm = current->mm;
1651 active_mm = current->active_mm;
1652 current->mm = new_mm;
1653 current->active_mm = new_mm;
1654 activate_mm(active_mm, new_mm);
1655 new_mm = mm;
1658 if (new_fd) {
1659 fd = current->files;
1660 current->files = new_fd;
1661 new_fd = fd;
1664 task_unlock(current);
1667 bad_unshare_cleanup_fd:
1668 if (new_fd)
1669 put_files_struct(new_fd);
1671 bad_unshare_cleanup_vm:
1672 if (new_mm)
1673 mmput(new_mm);
1675 bad_unshare_cleanup_sigh:
1676 if (new_sigh)
1677 if (atomic_dec_and_test(&new_sigh->count))
1678 kmem_cache_free(sighand_cachep, new_sigh);
1680 bad_unshare_cleanup_ns:
1681 if (new_ns)
1682 put_namespace(new_ns);
1684 bad_unshare_cleanup_fs:
1685 if (new_fs)
1686 put_fs_struct(new_fs);
1688 bad_unshare_cleanup_thread:
1689 bad_unshare_out:
1690 return err;