blob: 876b31cd822d2669b5831b31707aabfdba676304 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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/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/ptrace.h>
39#include <linux/mount.h>
40#include <linux/audit.h>
41#include <linux/profile.h>
42#include <linux/rmap.h>
43#include <linux/acct.h>
44
45#include <asm/pgtable.h>
46#include <asm/pgalloc.h>
47#include <asm/uaccess.h>
48#include <asm/mmu_context.h>
49#include <asm/cacheflush.h>
50#include <asm/tlbflush.h>
51
52/*
53 * Protected counters by write_lock_irq(&tasklist_lock)
54 */
55unsigned long total_forks; /* Handle normal Linux uptimes. */
56int nr_threads; /* The idle threads do not count.. */
57
58int max_threads; /* tunable limit on nr_threads */
59
60DEFINE_PER_CPU(unsigned long, process_counts) = 0;
61
62 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
63
64EXPORT_SYMBOL(tasklist_lock);
65
66int nr_processes(void)
67{
68 int cpu;
69 int total = 0;
70
71 for_each_online_cpu(cpu)
72 total += per_cpu(process_counts, cpu);
73
74 return total;
75}
76
77#ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
78# define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
79# define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
80static kmem_cache_t *task_struct_cachep;
81#endif
82
83/* SLAB cache for signal_struct structures (tsk->signal) */
84kmem_cache_t *signal_cachep;
85
86/* SLAB cache for sighand_struct structures (tsk->sighand) */
87kmem_cache_t *sighand_cachep;
88
89/* SLAB cache for files_struct structures (tsk->files) */
90kmem_cache_t *files_cachep;
91
92/* SLAB cache for fs_struct structures (tsk->fs) */
93kmem_cache_t *fs_cachep;
94
95/* SLAB cache for vm_area_struct structures */
96kmem_cache_t *vm_area_cachep;
97
98/* SLAB cache for mm_struct structures (tsk->mm) */
99static kmem_cache_t *mm_cachep;
100
101void free_task(struct task_struct *tsk)
102{
103 free_thread_info(tsk->thread_info);
104 free_task_struct(tsk);
105}
106EXPORT_SYMBOL(free_task);
107
108void __put_task_struct(struct task_struct *tsk)
109{
110 WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
111 WARN_ON(atomic_read(&tsk->usage));
112 WARN_ON(tsk == current);
113
114 if (unlikely(tsk->audit_context))
115 audit_free(tsk);
116 security_task_free(tsk);
117 free_uid(tsk->user);
118 put_group_info(tsk->group_info);
119
120 if (!profile_handoff_task(tsk))
121 free_task(tsk);
122}
123
124void __init fork_init(unsigned long mempages)
125{
126#ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
127#ifndef ARCH_MIN_TASKALIGN
128#define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
129#endif
130 /* create a slab on which task_structs can be allocated */
131 task_struct_cachep =
132 kmem_cache_create("task_struct", sizeof(struct task_struct),
133 ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
134#endif
135
136 /*
137 * The default maximum number of threads is set to a safe
138 * value: the thread structures can take up at most half
139 * of memory.
140 */
141 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
142
143 /*
144 * we need to allow at least 20 threads to boot a system
145 */
146 if(max_threads < 20)
147 max_threads = 20;
148
149 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
150 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
151 init_task.signal->rlim[RLIMIT_SIGPENDING] =
152 init_task.signal->rlim[RLIMIT_NPROC];
153}
154
155static struct task_struct *dup_task_struct(struct task_struct *orig)
156{
157 struct task_struct *tsk;
158 struct thread_info *ti;
159
160 prepare_to_copy(orig);
161
162 tsk = alloc_task_struct();
163 if (!tsk)
164 return NULL;
165
166 ti = alloc_thread_info(tsk);
167 if (!ti) {
168 free_task_struct(tsk);
169 return NULL;
170 }
171
172 *ti = *orig->thread_info;
173 *tsk = *orig;
174 tsk->thread_info = ti;
175 ti->task = tsk;
176
177 /* One for us, one for whoever does the "release_task()" (usually parent) */
178 atomic_set(&tsk->usage,2);
179 return tsk;
180}
181
182#ifdef CONFIG_MMU
183static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
184{
185 struct vm_area_struct * mpnt, *tmp, **pprev;
186 struct rb_node **rb_link, *rb_parent;
187 int retval;
188 unsigned long charge;
189 struct mempolicy *pol;
190
191 down_write(&oldmm->mmap_sem);
192 flush_cache_mm(current->mm);
193 mm->locked_vm = 0;
194 mm->mmap = NULL;
195 mm->mmap_cache = NULL;
196 mm->free_area_cache = oldmm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700197 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 mm->map_count = 0;
199 set_mm_counter(mm, rss, 0);
200 set_mm_counter(mm, anon_rss, 0);
201 cpus_clear(mm->cpu_vm_mask);
202 mm->mm_rb = RB_ROOT;
203 rb_link = &mm->mm_rb.rb_node;
204 rb_parent = NULL;
205 pprev = &mm->mmap;
206
207 for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
208 struct file *file;
209
210 if (mpnt->vm_flags & VM_DONTCOPY) {
211 __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
212 -vma_pages(mpnt));
213 continue;
214 }
215 charge = 0;
216 if (mpnt->vm_flags & VM_ACCOUNT) {
217 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
218 if (security_vm_enough_memory(len))
219 goto fail_nomem;
220 charge = len;
221 }
222 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
223 if (!tmp)
224 goto fail_nomem;
225 *tmp = *mpnt;
226 pol = mpol_copy(vma_policy(mpnt));
227 retval = PTR_ERR(pol);
228 if (IS_ERR(pol))
229 goto fail_nomem_policy;
230 vma_set_policy(tmp, pol);
231 tmp->vm_flags &= ~VM_LOCKED;
232 tmp->vm_mm = mm;
233 tmp->vm_next = NULL;
234 anon_vma_link(tmp);
235 file = tmp->vm_file;
236 if (file) {
237 struct inode *inode = file->f_dentry->d_inode;
238 get_file(file);
239 if (tmp->vm_flags & VM_DENYWRITE)
240 atomic_dec(&inode->i_writecount);
241
242 /* insert tmp into the share list, just after mpnt */
243 spin_lock(&file->f_mapping->i_mmap_lock);
244 tmp->vm_truncate_count = mpnt->vm_truncate_count;
245 flush_dcache_mmap_lock(file->f_mapping);
246 vma_prio_tree_add(tmp, mpnt);
247 flush_dcache_mmap_unlock(file->f_mapping);
248 spin_unlock(&file->f_mapping->i_mmap_lock);
249 }
250
251 /*
252 * Link in the new vma and copy the page table entries:
253 * link in first so that swapoff can see swap entries,
254 * and try_to_unmap_one's find_vma find the new vma.
255 */
256 spin_lock(&mm->page_table_lock);
257 *pprev = tmp;
258 pprev = &tmp->vm_next;
259
260 __vma_link_rb(mm, tmp, rb_link, rb_parent);
261 rb_link = &tmp->vm_rb.rb_right;
262 rb_parent = &tmp->vm_rb;
263
264 mm->map_count++;
265 retval = copy_page_range(mm, current->mm, tmp);
266 spin_unlock(&mm->page_table_lock);
267
268 if (tmp->vm_ops && tmp->vm_ops->open)
269 tmp->vm_ops->open(tmp);
270
271 if (retval)
272 goto out;
273 }
274 retval = 0;
275
276out:
277 flush_tlb_mm(current->mm);
278 up_write(&oldmm->mmap_sem);
279 return retval;
280fail_nomem_policy:
281 kmem_cache_free(vm_area_cachep, tmp);
282fail_nomem:
283 retval = -ENOMEM;
284 vm_unacct_memory(charge);
285 goto out;
286}
287
288static inline int mm_alloc_pgd(struct mm_struct * mm)
289{
290 mm->pgd = pgd_alloc(mm);
291 if (unlikely(!mm->pgd))
292 return -ENOMEM;
293 return 0;
294}
295
296static inline void mm_free_pgd(struct mm_struct * mm)
297{
298 pgd_free(mm->pgd);
299}
300#else
301#define dup_mmap(mm, oldmm) (0)
302#define mm_alloc_pgd(mm) (0)
303#define mm_free_pgd(mm)
304#endif /* CONFIG_MMU */
305
306 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
307
308#define allocate_mm() (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
309#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
310
311#include <linux/init_task.h>
312
313static struct mm_struct * mm_init(struct mm_struct * mm)
314{
315 atomic_set(&mm->mm_users, 1);
316 atomic_set(&mm->mm_count, 1);
317 init_rwsem(&mm->mmap_sem);
318 INIT_LIST_HEAD(&mm->mmlist);
319 mm->core_waiters = 0;
320 mm->nr_ptes = 0;
321 spin_lock_init(&mm->page_table_lock);
322 rwlock_init(&mm->ioctx_list_lock);
323 mm->ioctx_list = NULL;
324 mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
325 mm->free_area_cache = TASK_UNMAPPED_BASE;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700326 mm->cached_hole_size = ~0UL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 if (likely(!mm_alloc_pgd(mm))) {
329 mm->def_flags = 0;
330 return mm;
331 }
332 free_mm(mm);
333 return NULL;
334}
335
336/*
337 * Allocate and initialize an mm_struct.
338 */
339struct mm_struct * mm_alloc(void)
340{
341 struct mm_struct * mm;
342
343 mm = allocate_mm();
344 if (mm) {
345 memset(mm, 0, sizeof(*mm));
346 mm = mm_init(mm);
347 }
348 return mm;
349}
350
351/*
352 * Called when the last reference to the mm
353 * is dropped: either by a lazy thread or by
354 * mmput. Free the page directory and the mm.
355 */
356void fastcall __mmdrop(struct mm_struct *mm)
357{
358 BUG_ON(mm == &init_mm);
359 mm_free_pgd(mm);
360 destroy_context(mm);
361 free_mm(mm);
362}
363
364/*
365 * Decrement the use count and release all resources for an mm.
366 */
367void mmput(struct mm_struct *mm)
368{
369 if (atomic_dec_and_test(&mm->mm_users)) {
370 exit_aio(mm);
371 exit_mmap(mm);
372 if (!list_empty(&mm->mmlist)) {
373 spin_lock(&mmlist_lock);
374 list_del(&mm->mmlist);
375 spin_unlock(&mmlist_lock);
376 }
377 put_swap_token(mm);
378 mmdrop(mm);
379 }
380}
381EXPORT_SYMBOL_GPL(mmput);
382
383/**
384 * get_task_mm - acquire a reference to the task's mm
385 *
386 * Returns %NULL if the task has no mm. Checks PF_BORROWED_MM (meaning
387 * this kernel workthread has transiently adopted a user mm with use_mm,
388 * to do its AIO) is not set and if so returns a reference to it, after
389 * bumping up the use count. User must release the mm via mmput()
390 * after use. Typically used by /proc and ptrace.
391 */
392struct mm_struct *get_task_mm(struct task_struct *task)
393{
394 struct mm_struct *mm;
395
396 task_lock(task);
397 mm = task->mm;
398 if (mm) {
399 if (task->flags & PF_BORROWED_MM)
400 mm = NULL;
401 else
402 atomic_inc(&mm->mm_users);
403 }
404 task_unlock(task);
405 return mm;
406}
407EXPORT_SYMBOL_GPL(get_task_mm);
408
409/* Please note the differences between mmput and mm_release.
410 * mmput is called whenever we stop holding onto a mm_struct,
411 * error success whatever.
412 *
413 * mm_release is called after a mm_struct has been removed
414 * from the current process.
415 *
416 * This difference is important for error handling, when we
417 * only half set up a mm_struct for a new process and need to restore
418 * the old one. Because we mmput the new mm_struct before
419 * restoring the old one. . .
420 * Eric Biederman 10 January 1998
421 */
422void mm_release(struct task_struct *tsk, struct mm_struct *mm)
423{
424 struct completion *vfork_done = tsk->vfork_done;
425
426 /* Get rid of any cached register state */
427 deactivate_mm(tsk, mm);
428
429 /* notify parent sleeping on vfork() */
430 if (vfork_done) {
431 tsk->vfork_done = NULL;
432 complete(vfork_done);
433 }
434 if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
435 u32 __user * tidptr = tsk->clear_child_tid;
436 tsk->clear_child_tid = NULL;
437
438 /*
439 * We don't check the error code - if userspace has
440 * not set up a proper pointer then tough luck.
441 */
442 put_user(0, tidptr);
443 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
444 }
445}
446
447static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
448{
449 struct mm_struct * mm, *oldmm;
450 int retval;
451
452 tsk->min_flt = tsk->maj_flt = 0;
453 tsk->nvcsw = tsk->nivcsw = 0;
454
455 tsk->mm = NULL;
456 tsk->active_mm = NULL;
457
458 /*
459 * Are we cloning a kernel thread?
460 *
461 * We need to steal a active VM for that..
462 */
463 oldmm = current->mm;
464 if (!oldmm)
465 return 0;
466
467 if (clone_flags & CLONE_VM) {
468 atomic_inc(&oldmm->mm_users);
469 mm = oldmm;
470 /*
471 * There are cases where the PTL is held to ensure no
472 * new threads start up in user mode using an mm, which
473 * allows optimizing out ipis; the tlb_gather_mmu code
474 * is an example.
475 */
476 spin_unlock_wait(&oldmm->page_table_lock);
477 goto good_mm;
478 }
479
480 retval = -ENOMEM;
481 mm = allocate_mm();
482 if (!mm)
483 goto fail_nomem;
484
485 /* Copy the current MM stuff.. */
486 memcpy(mm, oldmm, sizeof(*mm));
487 if (!mm_init(mm))
488 goto fail_nomem;
489
490 if (init_new_context(tsk,mm))
491 goto fail_nocontext;
492
493 retval = dup_mmap(mm, oldmm);
494 if (retval)
495 goto free_pt;
496
497 mm->hiwater_rss = get_mm_counter(mm,rss);
498 mm->hiwater_vm = mm->total_vm;
499
500good_mm:
501 tsk->mm = mm;
502 tsk->active_mm = mm;
503 return 0;
504
505free_pt:
506 mmput(mm);
507fail_nomem:
508 return retval;
509
510fail_nocontext:
511 /*
512 * If init_new_context() failed, we cannot use mmput() to free the mm
513 * because it calls destroy_context()
514 */
515 mm_free_pgd(mm);
516 free_mm(mm);
517 return retval;
518}
519
520static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
521{
522 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
523 /* We don't need to lock fs - think why ;-) */
524 if (fs) {
525 atomic_set(&fs->count, 1);
526 rwlock_init(&fs->lock);
527 fs->umask = old->umask;
528 read_lock(&old->lock);
529 fs->rootmnt = mntget(old->rootmnt);
530 fs->root = dget(old->root);
531 fs->pwdmnt = mntget(old->pwdmnt);
532 fs->pwd = dget(old->pwd);
533 if (old->altroot) {
534 fs->altrootmnt = mntget(old->altrootmnt);
535 fs->altroot = dget(old->altroot);
536 } else {
537 fs->altrootmnt = NULL;
538 fs->altroot = NULL;
539 }
540 read_unlock(&old->lock);
541 }
542 return fs;
543}
544
545struct fs_struct *copy_fs_struct(struct fs_struct *old)
546{
547 return __copy_fs_struct(old);
548}
549
550EXPORT_SYMBOL_GPL(copy_fs_struct);
551
552static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
553{
554 if (clone_flags & CLONE_FS) {
555 atomic_inc(&current->fs->count);
556 return 0;
557 }
558 tsk->fs = __copy_fs_struct(current->fs);
559 if (!tsk->fs)
560 return -ENOMEM;
561 return 0;
562}
563
564static int count_open_files(struct files_struct *files, int size)
565{
566 int i;
567
568 /* Find the last open fd */
569 for (i = size/(8*sizeof(long)); i > 0; ) {
570 if (files->open_fds->fds_bits[--i])
571 break;
572 }
573 i = (i+1) * 8 * sizeof(long);
574 return i;
575}
576
577static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
578{
579 struct files_struct *oldf, *newf;
580 struct file **old_fds, **new_fds;
581 int open_files, size, i, error = 0, expand;
582
583 /*
584 * A background process may not have any files ...
585 */
586 oldf = current->files;
587 if (!oldf)
588 goto out;
589
590 if (clone_flags & CLONE_FILES) {
591 atomic_inc(&oldf->count);
592 goto out;
593 }
594
595 /*
596 * Note: we may be using current for both targets (See exec.c)
597 * This works because we cache current->files (old) as oldf. Don't
598 * break this.
599 */
600 tsk->files = NULL;
601 error = -ENOMEM;
602 newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
603 if (!newf)
604 goto out;
605
606 atomic_set(&newf->count, 1);
607
608 spin_lock_init(&newf->file_lock);
609 newf->next_fd = 0;
610 newf->max_fds = NR_OPEN_DEFAULT;
611 newf->max_fdset = __FD_SETSIZE;
612 newf->close_on_exec = &newf->close_on_exec_init;
613 newf->open_fds = &newf->open_fds_init;
614 newf->fd = &newf->fd_array[0];
615
616 spin_lock(&oldf->file_lock);
617
618 open_files = count_open_files(oldf, oldf->max_fdset);
619 expand = 0;
620
621 /*
622 * Check whether we need to allocate a larger fd array or fd set.
623 * Note: we're not a clone task, so the open count won't change.
624 */
625 if (open_files > newf->max_fdset) {
626 newf->max_fdset = 0;
627 expand = 1;
628 }
629 if (open_files > newf->max_fds) {
630 newf->max_fds = 0;
631 expand = 1;
632 }
633
634 /* if the old fdset gets grown now, we'll only copy up to "size" fds */
635 if (expand) {
636 spin_unlock(&oldf->file_lock);
637 spin_lock(&newf->file_lock);
638 error = expand_files(newf, open_files-1);
639 spin_unlock(&newf->file_lock);
640 if (error < 0)
641 goto out_release;
642 spin_lock(&oldf->file_lock);
643 }
644
645 old_fds = oldf->fd;
646 new_fds = newf->fd;
647
648 memcpy(newf->open_fds->fds_bits, oldf->open_fds->fds_bits, open_files/8);
649 memcpy(newf->close_on_exec->fds_bits, oldf->close_on_exec->fds_bits, open_files/8);
650
651 for (i = open_files; i != 0; i--) {
652 struct file *f = *old_fds++;
653 if (f) {
654 get_file(f);
655 } else {
656 /*
657 * The fd may be claimed in the fd bitmap but not yet
658 * instantiated in the files array if a sibling thread
659 * is partway through open(). So make sure that this
660 * fd is available to the new process.
661 */
662 FD_CLR(open_files - i, newf->open_fds);
663 }
664 *new_fds++ = f;
665 }
666 spin_unlock(&oldf->file_lock);
667
668 /* compute the remainder to be cleared */
669 size = (newf->max_fds - open_files) * sizeof(struct file *);
670
671 /* This is long word aligned thus could use a optimized version */
672 memset(new_fds, 0, size);
673
674 if (newf->max_fdset > open_files) {
675 int left = (newf->max_fdset-open_files)/8;
676 int start = open_files / (8 * sizeof(unsigned long));
677
678 memset(&newf->open_fds->fds_bits[start], 0, left);
679 memset(&newf->close_on_exec->fds_bits[start], 0, left);
680 }
681
682 tsk->files = newf;
683 error = 0;
684out:
685 return error;
686
687out_release:
688 free_fdset (newf->close_on_exec, newf->max_fdset);
689 free_fdset (newf->open_fds, newf->max_fdset);
690 free_fd_array(newf->fd, newf->max_fds);
691 kmem_cache_free(files_cachep, newf);
692 goto out;
693}
694
695/*
696 * Helper to unshare the files of the current task.
697 * We don't want to expose copy_files internals to
698 * the exec layer of the kernel.
699 */
700
701int unshare_files(void)
702{
703 struct files_struct *files = current->files;
704 int rc;
705
706 if(!files)
707 BUG();
708
709 /* This can race but the race causes us to copy when we don't
710 need to and drop the copy */
711 if(atomic_read(&files->count) == 1)
712 {
713 atomic_inc(&files->count);
714 return 0;
715 }
716 rc = copy_files(0, current);
717 if(rc)
718 current->files = files;
719 return rc;
720}
721
722EXPORT_SYMBOL(unshare_files);
723
724static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
725{
726 struct sighand_struct *sig;
727
728 if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
729 atomic_inc(&current->sighand->count);
730 return 0;
731 }
732 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
733 tsk->sighand = sig;
734 if (!sig)
735 return -ENOMEM;
736 spin_lock_init(&sig->siglock);
737 atomic_set(&sig->count, 1);
738 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
739 return 0;
740}
741
742static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
743{
744 struct signal_struct *sig;
745 int ret;
746
747 if (clone_flags & CLONE_THREAD) {
748 atomic_inc(&current->signal->count);
749 atomic_inc(&current->signal->live);
750 return 0;
751 }
752 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
753 tsk->signal = sig;
754 if (!sig)
755 return -ENOMEM;
756
757 ret = copy_thread_group_keys(tsk);
758 if (ret < 0) {
759 kmem_cache_free(signal_cachep, sig);
760 return ret;
761 }
762
763 atomic_set(&sig->count, 1);
764 atomic_set(&sig->live, 1);
765 init_waitqueue_head(&sig->wait_chldexit);
766 sig->flags = 0;
767 sig->group_exit_code = 0;
768 sig->group_exit_task = NULL;
769 sig->group_stop_count = 0;
770 sig->curr_target = NULL;
771 init_sigpending(&sig->shared_pending);
772 INIT_LIST_HEAD(&sig->posix_timers);
773
774 sig->it_real_value = sig->it_real_incr = 0;
775 sig->real_timer.function = it_real_fn;
776 sig->real_timer.data = (unsigned long) tsk;
777 init_timer(&sig->real_timer);
778
779 sig->it_virt_expires = cputime_zero;
780 sig->it_virt_incr = cputime_zero;
781 sig->it_prof_expires = cputime_zero;
782 sig->it_prof_incr = cputime_zero;
783
784 sig->tty = current->signal->tty;
785 sig->pgrp = process_group(current);
786 sig->session = current->signal->session;
787 sig->leader = 0; /* session leadership doesn't inherit */
788 sig->tty_old_pgrp = 0;
789
790 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
791 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
792 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
793 sig->sched_time = 0;
794 INIT_LIST_HEAD(&sig->cpu_timers[0]);
795 INIT_LIST_HEAD(&sig->cpu_timers[1]);
796 INIT_LIST_HEAD(&sig->cpu_timers[2]);
797
798 task_lock(current->group_leader);
799 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
800 task_unlock(current->group_leader);
801
802 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
803 /*
804 * New sole thread in the process gets an expiry time
805 * of the whole CPU time limit.
806 */
807 tsk->it_prof_expires =
808 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
809 }
810
811 return 0;
812}
813
814static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
815{
816 unsigned long new_flags = p->flags;
817
818 new_flags &= ~PF_SUPERPRIV;
819 new_flags |= PF_FORKNOEXEC;
820 if (!(clone_flags & CLONE_PTRACE))
821 p->ptrace = 0;
822 p->flags = new_flags;
823}
824
825asmlinkage long sys_set_tid_address(int __user *tidptr)
826{
827 current->clear_child_tid = tidptr;
828
829 return current->pid;
830}
831
832/*
833 * This creates a new process as a copy of the old one,
834 * but does not actually start it yet.
835 *
836 * It copies the registers, and all the appropriate
837 * parts of the process environment (as per the clone
838 * flags). The actual kick-off is left to the caller.
839 */
840static task_t *copy_process(unsigned long clone_flags,
841 unsigned long stack_start,
842 struct pt_regs *regs,
843 unsigned long stack_size,
844 int __user *parent_tidptr,
845 int __user *child_tidptr,
846 int pid)
847{
848 int retval;
849 struct task_struct *p = NULL;
850
851 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
852 return ERR_PTR(-EINVAL);
853
854 /*
855 * Thread groups must share signals as well, and detached threads
856 * can only be started up within the thread group.
857 */
858 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
859 return ERR_PTR(-EINVAL);
860
861 /*
862 * Shared signal handlers imply shared VM. By way of the above,
863 * thread groups also imply shared VM. Blocking this case allows
864 * for various simplifications in other code.
865 */
866 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
867 return ERR_PTR(-EINVAL);
868
869 retval = security_task_create(clone_flags);
870 if (retval)
871 goto fork_out;
872
873 retval = -ENOMEM;
874 p = dup_task_struct(current);
875 if (!p)
876 goto fork_out;
877
878 retval = -EAGAIN;
879 if (atomic_read(&p->user->processes) >=
880 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
881 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
882 p->user != &root_user)
883 goto bad_fork_free;
884 }
885
886 atomic_inc(&p->user->__count);
887 atomic_inc(&p->user->processes);
888 get_group_info(p->group_info);
889
890 /*
891 * If multiple threads are within copy_process(), then this check
892 * triggers too late. This doesn't hurt, the check is only there
893 * to stop root fork bombs.
894 */
895 if (nr_threads >= max_threads)
896 goto bad_fork_cleanup_count;
897
898 if (!try_module_get(p->thread_info->exec_domain->module))
899 goto bad_fork_cleanup_count;
900
901 if (p->binfmt && !try_module_get(p->binfmt->module))
902 goto bad_fork_cleanup_put_domain;
903
904 p->did_exec = 0;
905 copy_flags(clone_flags, p);
906 p->pid = pid;
907 retval = -EFAULT;
908 if (clone_flags & CLONE_PARENT_SETTID)
909 if (put_user(p->pid, parent_tidptr))
910 goto bad_fork_cleanup;
911
912 p->proc_dentry = NULL;
913
914 INIT_LIST_HEAD(&p->children);
915 INIT_LIST_HEAD(&p->sibling);
916 p->vfork_done = NULL;
917 spin_lock_init(&p->alloc_lock);
918 spin_lock_init(&p->proc_lock);
919
920 clear_tsk_thread_flag(p, TIF_SIGPENDING);
921 init_sigpending(&p->pending);
922
923 p->utime = cputime_zero;
924 p->stime = cputime_zero;
925 p->sched_time = 0;
926 p->rchar = 0; /* I/O counter: bytes read */
927 p->wchar = 0; /* I/O counter: bytes written */
928 p->syscr = 0; /* I/O counter: read syscalls */
929 p->syscw = 0; /* I/O counter: write syscalls */
930 acct_clear_integrals(p);
931
932 p->it_virt_expires = cputime_zero;
933 p->it_prof_expires = cputime_zero;
934 p->it_sched_expires = 0;
935 INIT_LIST_HEAD(&p->cpu_timers[0]);
936 INIT_LIST_HEAD(&p->cpu_timers[1]);
937 INIT_LIST_HEAD(&p->cpu_timers[2]);
938
939 p->lock_depth = -1; /* -1 = no lock */
940 do_posix_clock_monotonic_gettime(&p->start_time);
941 p->security = NULL;
942 p->io_context = NULL;
943 p->io_wait = NULL;
944 p->audit_context = NULL;
945#ifdef CONFIG_NUMA
946 p->mempolicy = mpol_copy(p->mempolicy);
947 if (IS_ERR(p->mempolicy)) {
948 retval = PTR_ERR(p->mempolicy);
949 p->mempolicy = NULL;
950 goto bad_fork_cleanup;
951 }
952#endif
953
954 p->tgid = p->pid;
955 if (clone_flags & CLONE_THREAD)
956 p->tgid = current->tgid;
957
958 if ((retval = security_task_alloc(p)))
959 goto bad_fork_cleanup_policy;
960 if ((retval = audit_alloc(p)))
961 goto bad_fork_cleanup_security;
962 /* copy all the process information */
963 if ((retval = copy_semundo(clone_flags, p)))
964 goto bad_fork_cleanup_audit;
965 if ((retval = copy_files(clone_flags, p)))
966 goto bad_fork_cleanup_semundo;
967 if ((retval = copy_fs(clone_flags, p)))
968 goto bad_fork_cleanup_files;
969 if ((retval = copy_sighand(clone_flags, p)))
970 goto bad_fork_cleanup_fs;
971 if ((retval = copy_signal(clone_flags, p)))
972 goto bad_fork_cleanup_sighand;
973 if ((retval = copy_mm(clone_flags, p)))
974 goto bad_fork_cleanup_signal;
975 if ((retval = copy_keys(clone_flags, p)))
976 goto bad_fork_cleanup_mm;
977 if ((retval = copy_namespace(clone_flags, p)))
978 goto bad_fork_cleanup_keys;
979 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
980 if (retval)
981 goto bad_fork_cleanup_namespace;
982
983 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
984 /*
985 * Clear TID on mm_release()?
986 */
987 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
988
989 /*
990 * Syscall tracing should be turned off in the child regardless
991 * of CLONE_PTRACE.
992 */
993 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
994
995 /* Our parent execution domain becomes current domain
996 These must match for thread signalling to apply */
997
998 p->parent_exec_id = p->self_exec_id;
999
1000 /* ok, now we should be set up.. */
1001 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1002 p->pdeath_signal = 0;
1003 p->exit_state = 0;
1004
1005 /* Perform scheduler related setup */
1006 sched_fork(p);
1007
1008 /*
1009 * Ok, make it visible to the rest of the system.
1010 * We dont wake it up yet.
1011 */
1012 p->group_leader = p;
1013 INIT_LIST_HEAD(&p->ptrace_children);
1014 INIT_LIST_HEAD(&p->ptrace_list);
1015
1016 /* Need tasklist lock for parent etc handling! */
1017 write_lock_irq(&tasklist_lock);
1018
1019 /*
1020 * The task hasn't been attached yet, so cpus_allowed mask cannot
1021 * have changed. The cpus_allowed mask of the parent may have
1022 * changed after it was copied first time, and it may then move to
1023 * another CPU - so we re-copy it here and set the child's CPU to
1024 * the parent's CPU. This avoids alot of nasty races.
1025 */
1026 p->cpus_allowed = current->cpus_allowed;
1027 set_task_cpu(p, smp_processor_id());
1028
1029 /*
1030 * Check for pending SIGKILL! The new thread should not be allowed
1031 * to slip out of an OOM kill. (or normal SIGKILL.)
1032 */
1033 if (sigismember(&current->pending.signal, SIGKILL)) {
1034 write_unlock_irq(&tasklist_lock);
1035 retval = -EINTR;
1036 goto bad_fork_cleanup_namespace;
1037 }
1038
1039 /* CLONE_PARENT re-uses the old parent */
1040 if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1041 p->real_parent = current->real_parent;
1042 else
1043 p->real_parent = current;
1044 p->parent = p->real_parent;
1045
1046 if (clone_flags & CLONE_THREAD) {
1047 spin_lock(&current->sighand->siglock);
1048 /*
1049 * Important: if an exit-all has been started then
1050 * do not create this new thread - the whole thread
1051 * group is supposed to exit anyway.
1052 */
1053 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1054 spin_unlock(&current->sighand->siglock);
1055 write_unlock_irq(&tasklist_lock);
1056 retval = -EAGAIN;
1057 goto bad_fork_cleanup_namespace;
1058 }
1059 p->group_leader = current->group_leader;
1060
1061 if (current->signal->group_stop_count > 0) {
1062 /*
1063 * There is an all-stop in progress for the group.
1064 * We ourselves will stop as soon as we check signals.
1065 * Make the new thread part of that group stop too.
1066 */
1067 current->signal->group_stop_count++;
1068 set_tsk_thread_flag(p, TIF_SIGPENDING);
1069 }
1070
1071 if (!cputime_eq(current->signal->it_virt_expires,
1072 cputime_zero) ||
1073 !cputime_eq(current->signal->it_prof_expires,
1074 cputime_zero) ||
1075 current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1076 !list_empty(&current->signal->cpu_timers[0]) ||
1077 !list_empty(&current->signal->cpu_timers[1]) ||
1078 !list_empty(&current->signal->cpu_timers[2])) {
1079 /*
1080 * Have child wake up on its first tick to check
1081 * for process CPU timers.
1082 */
1083 p->it_prof_expires = jiffies_to_cputime(1);
1084 }
1085
1086 spin_unlock(&current->sighand->siglock);
1087 }
1088
1089 SET_LINKS(p);
1090 if (unlikely(p->ptrace & PT_PTRACED))
1091 __ptrace_link(p, current->parent);
1092
1093 cpuset_fork(p);
1094
1095 attach_pid(p, PIDTYPE_PID, p->pid);
1096 attach_pid(p, PIDTYPE_TGID, p->tgid);
1097 if (thread_group_leader(p)) {
1098 attach_pid(p, PIDTYPE_PGID, process_group(p));
1099 attach_pid(p, PIDTYPE_SID, p->signal->session);
1100 if (p->pid)
1101 __get_cpu_var(process_counts)++;
1102 }
1103
1104 nr_threads++;
1105 total_forks++;
1106 write_unlock_irq(&tasklist_lock);
1107 retval = 0;
1108
1109fork_out:
1110 if (retval)
1111 return ERR_PTR(retval);
1112 return p;
1113
1114bad_fork_cleanup_namespace:
1115 exit_namespace(p);
1116bad_fork_cleanup_keys:
1117 exit_keys(p);
1118bad_fork_cleanup_mm:
1119 if (p->mm)
1120 mmput(p->mm);
1121bad_fork_cleanup_signal:
1122 exit_signal(p);
1123bad_fork_cleanup_sighand:
1124 exit_sighand(p);
1125bad_fork_cleanup_fs:
1126 exit_fs(p); /* blocking */
1127bad_fork_cleanup_files:
1128 exit_files(p); /* blocking */
1129bad_fork_cleanup_semundo:
1130 exit_sem(p);
1131bad_fork_cleanup_audit:
1132 audit_free(p);
1133bad_fork_cleanup_security:
1134 security_task_free(p);
1135bad_fork_cleanup_policy:
1136#ifdef CONFIG_NUMA
1137 mpol_free(p->mempolicy);
1138#endif
1139bad_fork_cleanup:
1140 if (p->binfmt)
1141 module_put(p->binfmt->module);
1142bad_fork_cleanup_put_domain:
1143 module_put(p->thread_info->exec_domain->module);
1144bad_fork_cleanup_count:
1145 put_group_info(p->group_info);
1146 atomic_dec(&p->user->processes);
1147 free_uid(p->user);
1148bad_fork_free:
1149 free_task(p);
1150 goto fork_out;
1151}
1152
1153struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1154{
1155 memset(regs, 0, sizeof(struct pt_regs));
1156 return regs;
1157}
1158
1159task_t * __devinit fork_idle(int cpu)
1160{
1161 task_t *task;
1162 struct pt_regs regs;
1163
1164 task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1165 if (!task)
1166 return ERR_PTR(-ENOMEM);
1167 init_idle(task, cpu);
1168 unhash_process(task);
1169 return task;
1170}
1171
1172static inline int fork_traceflag (unsigned clone_flags)
1173{
1174 if (clone_flags & CLONE_UNTRACED)
1175 return 0;
1176 else if (clone_flags & CLONE_VFORK) {
1177 if (current->ptrace & PT_TRACE_VFORK)
1178 return PTRACE_EVENT_VFORK;
1179 } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1180 if (current->ptrace & PT_TRACE_CLONE)
1181 return PTRACE_EVENT_CLONE;
1182 } else if (current->ptrace & PT_TRACE_FORK)
1183 return PTRACE_EVENT_FORK;
1184
1185 return 0;
1186}
1187
1188/*
1189 * Ok, this is the main fork-routine.
1190 *
1191 * It copies the process, and if successful kick-starts
1192 * it and waits for it to finish using the VM if required.
1193 */
1194long do_fork(unsigned long clone_flags,
1195 unsigned long stack_start,
1196 struct pt_regs *regs,
1197 unsigned long stack_size,
1198 int __user *parent_tidptr,
1199 int __user *child_tidptr)
1200{
1201 struct task_struct *p;
1202 int trace = 0;
1203 long pid = alloc_pidmap();
1204
1205 if (pid < 0)
1206 return -EAGAIN;
1207 if (unlikely(current->ptrace)) {
1208 trace = fork_traceflag (clone_flags);
1209 if (trace)
1210 clone_flags |= CLONE_PTRACE;
1211 }
1212
1213 p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1214 /*
1215 * Do this prior waking up the new thread - the thread pointer
1216 * might get invalid after that point, if the thread exits quickly.
1217 */
1218 if (!IS_ERR(p)) {
1219 struct completion vfork;
1220
1221 if (clone_flags & CLONE_VFORK) {
1222 p->vfork_done = &vfork;
1223 init_completion(&vfork);
1224 }
1225
1226 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1227 /*
1228 * We'll start up with an immediate SIGSTOP.
1229 */
1230 sigaddset(&p->pending.signal, SIGSTOP);
1231 set_tsk_thread_flag(p, TIF_SIGPENDING);
1232 }
1233
1234 if (!(clone_flags & CLONE_STOPPED))
1235 wake_up_new_task(p, clone_flags);
1236 else
1237 p->state = TASK_STOPPED;
1238
1239 if (unlikely (trace)) {
1240 current->ptrace_message = pid;
1241 ptrace_notify ((trace << 8) | SIGTRAP);
1242 }
1243
1244 if (clone_flags & CLONE_VFORK) {
1245 wait_for_completion(&vfork);
1246 if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1247 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1248 }
1249 } else {
1250 free_pidmap(pid);
1251 pid = PTR_ERR(p);
1252 }
1253 return pid;
1254}
1255
1256void __init proc_caches_init(void)
1257{
1258 sighand_cachep = kmem_cache_create("sighand_cache",
1259 sizeof(struct sighand_struct), 0,
1260 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1261 signal_cachep = kmem_cache_create("signal_cache",
1262 sizeof(struct signal_struct), 0,
1263 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1264 files_cachep = kmem_cache_create("files_cache",
1265 sizeof(struct files_struct), 0,
1266 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1267 fs_cachep = kmem_cache_create("fs_cache",
1268 sizeof(struct fs_struct), 0,
1269 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1270 vm_area_cachep = kmem_cache_create("vm_area_struct",
1271 sizeof(struct vm_area_struct), 0,
1272 SLAB_PANIC, NULL, NULL);
1273 mm_cachep = kmem_cache_create("mm_struct",
1274 sizeof(struct mm_struct), 0,
1275 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1276}