ecc694debb50ca1db8dcdea2cddb407d9f8ecead
[linux-3.10.git] / kernel / fork.c
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/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  */
55 unsigned long total_forks;      /* Handle normal Linux uptimes. */
56 int nr_threads;                 /* The idle threads do not count.. */
57
58 int max_threads;                /* tunable limit on nr_threads */
59
60 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
61
62  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
63
64 EXPORT_SYMBOL(tasklist_lock);
65
66 int 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))
80 static kmem_cache_t *task_struct_cachep;
81 #endif
82
83 /* SLAB cache for signal_struct structures (tsk->signal) */
84 kmem_cache_t *signal_cachep;
85
86 /* SLAB cache for sighand_struct structures (tsk->sighand) */
87 kmem_cache_t *sighand_cachep;
88
89 /* SLAB cache for files_struct structures (tsk->files) */
90 kmem_cache_t *files_cachep;
91
92 /* SLAB cache for fs_struct structures (tsk->fs) */
93 kmem_cache_t *fs_cachep;
94
95 /* SLAB cache for vm_area_struct structures */
96 kmem_cache_t *vm_area_cachep;
97
98 /* SLAB cache for mm_struct structures (tsk->mm) */
99 static kmem_cache_t *mm_cachep;
100
101 void free_task(struct task_struct *tsk)
102 {
103         free_thread_info(tsk->thread_info);
104         free_task_struct(tsk);
105 }
106 EXPORT_SYMBOL(free_task);
107
108 void __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
124 void __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
155 static 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         atomic_set(&tsk->fs_excl, 0);
180         return tsk;
181 }
182
183 #ifdef CONFIG_MMU
184 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
185 {
186         struct vm_area_struct * mpnt, *tmp, **pprev;
187         struct rb_node **rb_link, *rb_parent;
188         int retval;
189         unsigned long charge;
190         struct mempolicy *pol;
191
192         down_write(&oldmm->mmap_sem);
193         flush_cache_mm(current->mm);
194         mm->locked_vm = 0;
195         mm->mmap = NULL;
196         mm->mmap_cache = NULL;
197         mm->free_area_cache = oldmm->mmap_base;
198         mm->cached_hole_size = ~0UL;
199         mm->map_count = 0;
200         set_mm_counter(mm, rss, 0);
201         set_mm_counter(mm, anon_rss, 0);
202         cpus_clear(mm->cpu_vm_mask);
203         mm->mm_rb = RB_ROOT;
204         rb_link = &mm->mm_rb.rb_node;
205         rb_parent = NULL;
206         pprev = &mm->mmap;
207
208         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
209                 struct file *file;
210
211                 if (mpnt->vm_flags & VM_DONTCOPY) {
212                         long pages = vma_pages(mpnt);
213                         mm->total_vm -= pages;
214                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
215                                                                 -pages);
216                         continue;
217                 }
218                 charge = 0;
219                 if (mpnt->vm_flags & VM_ACCOUNT) {
220                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
221                         if (security_vm_enough_memory(len))
222                                 goto fail_nomem;
223                         charge = len;
224                 }
225                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
226                 if (!tmp)
227                         goto fail_nomem;
228                 *tmp = *mpnt;
229                 pol = mpol_copy(vma_policy(mpnt));
230                 retval = PTR_ERR(pol);
231                 if (IS_ERR(pol))
232                         goto fail_nomem_policy;
233                 vma_set_policy(tmp, pol);
234                 tmp->vm_flags &= ~VM_LOCKED;
235                 tmp->vm_mm = mm;
236                 tmp->vm_next = NULL;
237                 anon_vma_link(tmp);
238                 file = tmp->vm_file;
239                 if (file) {
240                         struct inode *inode = file->f_dentry->d_inode;
241                         get_file(file);
242                         if (tmp->vm_flags & VM_DENYWRITE)
243                                 atomic_dec(&inode->i_writecount);
244       
245                         /* insert tmp into the share list, just after mpnt */
246                         spin_lock(&file->f_mapping->i_mmap_lock);
247                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
248                         flush_dcache_mmap_lock(file->f_mapping);
249                         vma_prio_tree_add(tmp, mpnt);
250                         flush_dcache_mmap_unlock(file->f_mapping);
251                         spin_unlock(&file->f_mapping->i_mmap_lock);
252                 }
253
254                 /*
255                  * Link in the new vma and copy the page table entries:
256                  * link in first so that swapoff can see swap entries.
257                  * Note that, exceptionally, here the vma is inserted
258                  * without holding mm->mmap_sem.
259                  */
260                 spin_lock(&mm->page_table_lock);
261                 *pprev = tmp;
262                 pprev = &tmp->vm_next;
263
264                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
265                 rb_link = &tmp->vm_rb.rb_right;
266                 rb_parent = &tmp->vm_rb;
267
268                 mm->map_count++;
269                 retval = copy_page_range(mm, current->mm, tmp);
270                 spin_unlock(&mm->page_table_lock);
271
272                 if (tmp->vm_ops && tmp->vm_ops->open)
273                         tmp->vm_ops->open(tmp);
274
275                 if (retval)
276                         goto out;
277         }
278         retval = 0;
279
280 out:
281         flush_tlb_mm(current->mm);
282         up_write(&oldmm->mmap_sem);
283         return retval;
284 fail_nomem_policy:
285         kmem_cache_free(vm_area_cachep, tmp);
286 fail_nomem:
287         retval = -ENOMEM;
288         vm_unacct_memory(charge);
289         goto out;
290 }
291
292 static inline int mm_alloc_pgd(struct mm_struct * mm)
293 {
294         mm->pgd = pgd_alloc(mm);
295         if (unlikely(!mm->pgd))
296                 return -ENOMEM;
297         return 0;
298 }
299
300 static inline void mm_free_pgd(struct mm_struct * mm)
301 {
302         pgd_free(mm->pgd);
303 }
304 #else
305 #define dup_mmap(mm, oldmm)     (0)
306 #define mm_alloc_pgd(mm)        (0)
307 #define mm_free_pgd(mm)
308 #endif /* CONFIG_MMU */
309
310  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
311
312 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
313 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
314
315 #include <linux/init_task.h>
316
317 static struct mm_struct * mm_init(struct mm_struct * mm)
318 {
319         atomic_set(&mm->mm_users, 1);
320         atomic_set(&mm->mm_count, 1);
321         init_rwsem(&mm->mmap_sem);
322         INIT_LIST_HEAD(&mm->mmlist);
323         mm->core_waiters = 0;
324         mm->nr_ptes = 0;
325         spin_lock_init(&mm->page_table_lock);
326         rwlock_init(&mm->ioctx_list_lock);
327         mm->ioctx_list = NULL;
328         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
329         mm->free_area_cache = TASK_UNMAPPED_BASE;
330         mm->cached_hole_size = ~0UL;
331
332         if (likely(!mm_alloc_pgd(mm))) {
333                 mm->def_flags = 0;
334                 return mm;
335         }
336         free_mm(mm);
337         return NULL;
338 }
339
340 /*
341  * Allocate and initialize an mm_struct.
342  */
343 struct mm_struct * mm_alloc(void)
344 {
345         struct mm_struct * mm;
346
347         mm = allocate_mm();
348         if (mm) {
349                 memset(mm, 0, sizeof(*mm));
350                 mm = mm_init(mm);
351         }
352         return mm;
353 }
354
355 /*
356  * Called when the last reference to the mm
357  * is dropped: either by a lazy thread or by
358  * mmput. Free the page directory and the mm.
359  */
360 void fastcall __mmdrop(struct mm_struct *mm)
361 {
362         BUG_ON(mm == &init_mm);
363         mm_free_pgd(mm);
364         destroy_context(mm);
365         free_mm(mm);
366 }
367
368 /*
369  * Decrement the use count and release all resources for an mm.
370  */
371 void mmput(struct mm_struct *mm)
372 {
373         if (atomic_dec_and_test(&mm->mm_users)) {
374                 exit_aio(mm);
375                 exit_mmap(mm);
376                 if (!list_empty(&mm->mmlist)) {
377                         spin_lock(&mmlist_lock);
378                         list_del(&mm->mmlist);
379                         spin_unlock(&mmlist_lock);
380                 }
381                 put_swap_token(mm);
382                 mmdrop(mm);
383         }
384 }
385 EXPORT_SYMBOL_GPL(mmput);
386
387 /**
388  * get_task_mm - acquire a reference to the task's mm
389  *
390  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
391  * this kernel workthread has transiently adopted a user mm with use_mm,
392  * to do its AIO) is not set and if so returns a reference to it, after
393  * bumping up the use count.  User must release the mm via mmput()
394  * after use.  Typically used by /proc and ptrace.
395  */
396 struct mm_struct *get_task_mm(struct task_struct *task)
397 {
398         struct mm_struct *mm;
399
400         task_lock(task);
401         mm = task->mm;
402         if (mm) {
403                 if (task->flags & PF_BORROWED_MM)
404                         mm = NULL;
405                 else
406                         atomic_inc(&mm->mm_users);
407         }
408         task_unlock(task);
409         return mm;
410 }
411 EXPORT_SYMBOL_GPL(get_task_mm);
412
413 /* Please note the differences between mmput and mm_release.
414  * mmput is called whenever we stop holding onto a mm_struct,
415  * error success whatever.
416  *
417  * mm_release is called after a mm_struct has been removed
418  * from the current process.
419  *
420  * This difference is important for error handling, when we
421  * only half set up a mm_struct for a new process and need to restore
422  * the old one.  Because we mmput the new mm_struct before
423  * restoring the old one. . .
424  * Eric Biederman 10 January 1998
425  */
426 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
427 {
428         struct completion *vfork_done = tsk->vfork_done;
429
430         /* Get rid of any cached register state */
431         deactivate_mm(tsk, mm);
432
433         /* notify parent sleeping on vfork() */
434         if (vfork_done) {
435                 tsk->vfork_done = NULL;
436                 complete(vfork_done);
437         }
438         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
439                 u32 __user * tidptr = tsk->clear_child_tid;
440                 tsk->clear_child_tid = NULL;
441
442                 /*
443                  * We don't check the error code - if userspace has
444                  * not set up a proper pointer then tough luck.
445                  */
446                 put_user(0, tidptr);
447                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
448         }
449 }
450
451 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
452 {
453         struct mm_struct * mm, *oldmm;
454         int retval;
455
456         tsk->min_flt = tsk->maj_flt = 0;
457         tsk->nvcsw = tsk->nivcsw = 0;
458
459         tsk->mm = NULL;
460         tsk->active_mm = NULL;
461
462         /*
463          * Are we cloning a kernel thread?
464          *
465          * We need to steal a active VM for that..
466          */
467         oldmm = current->mm;
468         if (!oldmm)
469                 return 0;
470
471         if (clone_flags & CLONE_VM) {
472                 atomic_inc(&oldmm->mm_users);
473                 mm = oldmm;
474                 /*
475                  * There are cases where the PTL is held to ensure no
476                  * new threads start up in user mode using an mm, which
477                  * allows optimizing out ipis; the tlb_gather_mmu code
478                  * is an example.
479                  */
480                 spin_unlock_wait(&oldmm->page_table_lock);
481                 goto good_mm;
482         }
483
484         retval = -ENOMEM;
485         mm = allocate_mm();
486         if (!mm)
487                 goto fail_nomem;
488
489         /* Copy the current MM stuff.. */
490         memcpy(mm, oldmm, sizeof(*mm));
491         if (!mm_init(mm))
492                 goto fail_nomem;
493
494         if (init_new_context(tsk,mm))
495                 goto fail_nocontext;
496
497         retval = dup_mmap(mm, oldmm);
498         if (retval)
499                 goto free_pt;
500
501         mm->hiwater_rss = get_mm_counter(mm,rss);
502         mm->hiwater_vm = mm->total_vm;
503
504 good_mm:
505         tsk->mm = mm;
506         tsk->active_mm = mm;
507         return 0;
508
509 free_pt:
510         mmput(mm);
511 fail_nomem:
512         return retval;
513
514 fail_nocontext:
515         /*
516          * If init_new_context() failed, we cannot use mmput() to free the mm
517          * because it calls destroy_context()
518          */
519         mm_free_pgd(mm);
520         free_mm(mm);
521         return retval;
522 }
523
524 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
525 {
526         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
527         /* We don't need to lock fs - think why ;-) */
528         if (fs) {
529                 atomic_set(&fs->count, 1);
530                 rwlock_init(&fs->lock);
531                 fs->umask = old->umask;
532                 read_lock(&old->lock);
533                 fs->rootmnt = mntget(old->rootmnt);
534                 fs->root = dget(old->root);
535                 fs->pwdmnt = mntget(old->pwdmnt);
536                 fs->pwd = dget(old->pwd);
537                 if (old->altroot) {
538                         fs->altrootmnt = mntget(old->altrootmnt);
539                         fs->altroot = dget(old->altroot);
540                 } else {
541                         fs->altrootmnt = NULL;
542                         fs->altroot = NULL;
543                 }
544                 read_unlock(&old->lock);
545         }
546         return fs;
547 }
548
549 struct fs_struct *copy_fs_struct(struct fs_struct *old)
550 {
551         return __copy_fs_struct(old);
552 }
553
554 EXPORT_SYMBOL_GPL(copy_fs_struct);
555
556 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
557 {
558         if (clone_flags & CLONE_FS) {
559                 atomic_inc(&current->fs->count);
560                 return 0;
561         }
562         tsk->fs = __copy_fs_struct(current->fs);
563         if (!tsk->fs)
564                 return -ENOMEM;
565         return 0;
566 }
567
568 static int count_open_files(struct files_struct *files, int size)
569 {
570         int i;
571         struct fdtable *fdt;
572
573         /* Find the last open fd */
574         fdt = files_fdtable(files);
575         for (i = size/(8*sizeof(long)); i > 0; ) {
576                 if (fdt->open_fds->fds_bits[--i])
577                         break;
578         }
579         i = (i+1) * 8 * sizeof(long);
580         return i;
581 }
582
583 static struct files_struct *alloc_files(void)
584 {
585         struct files_struct *newf;
586         struct fdtable *fdt;
587
588         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
589         if (!newf)
590                 goto out;
591
592         atomic_set(&newf->count, 1);
593
594         spin_lock_init(&newf->file_lock);
595         fdt = files_fdtable(newf);
596         fdt->next_fd = 0;
597         fdt->max_fds = NR_OPEN_DEFAULT;
598         fdt->max_fdset = __FD_SETSIZE;
599         fdt->close_on_exec = &newf->close_on_exec_init;
600         fdt->open_fds = &newf->open_fds_init;
601         fdt->fd = &newf->fd_array[0];
602 out:
603         return newf;
604 }
605
606 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
607 {
608         struct files_struct *oldf, *newf;
609         struct file **old_fds, **new_fds;
610         int open_files, size, i, error = 0, expand;
611         struct fdtable *old_fdt, *new_fdt;
612
613         /*
614          * A background process may not have any files ...
615          */
616         oldf = current->files;
617         if (!oldf)
618                 goto out;
619
620         if (clone_flags & CLONE_FILES) {
621                 atomic_inc(&oldf->count);
622                 goto out;
623         }
624
625         /*
626          * Note: we may be using current for both targets (See exec.c)
627          * This works because we cache current->files (old) as oldf. Don't
628          * break this.
629          */
630         tsk->files = NULL;
631         error = -ENOMEM;
632         newf = alloc_files();
633         if (!newf)
634                 goto out;
635
636         spin_lock(&oldf->file_lock);
637         old_fdt = files_fdtable(oldf);
638         new_fdt = files_fdtable(newf);
639         size = old_fdt->max_fdset;
640         open_files = count_open_files(oldf, old_fdt->max_fdset);
641         expand = 0;
642
643         /*
644          * Check whether we need to allocate a larger fd array or fd set.
645          * Note: we're not a clone task, so the open count won't  change.
646          */
647         if (open_files > new_fdt->max_fdset) {
648                 new_fdt->max_fdset = 0;
649                 expand = 1;
650         }
651         if (open_files > new_fdt->max_fds) {
652                 new_fdt->max_fds = 0;
653                 expand = 1;
654         }
655
656         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
657         if (expand) {
658                 spin_unlock(&oldf->file_lock);
659                 spin_lock(&newf->file_lock);
660                 error = expand_files(newf, open_files-1);
661                 spin_unlock(&newf->file_lock);
662                 if (error < 0)
663                         goto out_release;
664                 spin_lock(&oldf->file_lock);
665         }
666
667         old_fds = old_fdt->fd;
668         new_fds = new_fdt->fd;
669
670         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
671         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
672
673         for (i = open_files; i != 0; i--) {
674                 struct file *f = *old_fds++;
675                 if (f) {
676                         get_file(f);
677                 } else {
678                         /*
679                          * The fd may be claimed in the fd bitmap but not yet
680                          * instantiated in the files array if a sibling thread
681                          * is partway through open().  So make sure that this
682                          * fd is available to the new process.
683                          */
684                         FD_CLR(open_files - i, new_fdt->open_fds);
685                 }
686                 *new_fds++ = f;
687         }
688         spin_unlock(&oldf->file_lock);
689
690         /* compute the remainder to be cleared */
691         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
692
693         /* This is long word aligned thus could use a optimized version */ 
694         memset(new_fds, 0, size); 
695
696         if (new_fdt->max_fdset > open_files) {
697                 int left = (new_fdt->max_fdset-open_files)/8;
698                 int start = open_files / (8 * sizeof(unsigned long));
699
700                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
701                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
702         }
703
704         tsk->files = newf;
705         error = 0;
706 out:
707         return error;
708
709 out_release:
710         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
711         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
712         free_fd_array(new_fdt->fd, new_fdt->max_fds);
713         kmem_cache_free(files_cachep, newf);
714         goto out;
715 }
716
717 /*
718  *      Helper to unshare the files of the current task.
719  *      We don't want to expose copy_files internals to
720  *      the exec layer of the kernel.
721  */
722
723 int unshare_files(void)
724 {
725         struct files_struct *files  = current->files;
726         int rc;
727
728         if(!files)
729                 BUG();
730
731         /* This can race but the race causes us to copy when we don't
732            need to and drop the copy */
733         if(atomic_read(&files->count) == 1)
734         {
735                 atomic_inc(&files->count);
736                 return 0;
737         }
738         rc = copy_files(0, current);
739         if(rc)
740                 current->files = files;
741         return rc;
742 }
743
744 EXPORT_SYMBOL(unshare_files);
745
746 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
747 {
748         struct sighand_struct *sig;
749
750         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
751                 atomic_inc(&current->sighand->count);
752                 return 0;
753         }
754         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
755         tsk->sighand = sig;
756         if (!sig)
757                 return -ENOMEM;
758         spin_lock_init(&sig->siglock);
759         atomic_set(&sig->count, 1);
760         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
761         return 0;
762 }
763
764 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
765 {
766         struct signal_struct *sig;
767         int ret;
768
769         if (clone_flags & CLONE_THREAD) {
770                 atomic_inc(&current->signal->count);
771                 atomic_inc(&current->signal->live);
772                 return 0;
773         }
774         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
775         tsk->signal = sig;
776         if (!sig)
777                 return -ENOMEM;
778
779         ret = copy_thread_group_keys(tsk);
780         if (ret < 0) {
781                 kmem_cache_free(signal_cachep, sig);
782                 return ret;
783         }
784
785         atomic_set(&sig->count, 1);
786         atomic_set(&sig->live, 1);
787         init_waitqueue_head(&sig->wait_chldexit);
788         sig->flags = 0;
789         sig->group_exit_code = 0;
790         sig->group_exit_task = NULL;
791         sig->group_stop_count = 0;
792         sig->curr_target = NULL;
793         init_sigpending(&sig->shared_pending);
794         INIT_LIST_HEAD(&sig->posix_timers);
795
796         sig->it_real_value = sig->it_real_incr = 0;
797         sig->real_timer.function = it_real_fn;
798         sig->real_timer.data = (unsigned long) tsk;
799         init_timer(&sig->real_timer);
800
801         sig->it_virt_expires = cputime_zero;
802         sig->it_virt_incr = cputime_zero;
803         sig->it_prof_expires = cputime_zero;
804         sig->it_prof_incr = cputime_zero;
805
806         sig->tty = current->signal->tty;
807         sig->pgrp = process_group(current);
808         sig->session = current->signal->session;
809         sig->leader = 0;        /* session leadership doesn't inherit */
810         sig->tty_old_pgrp = 0;
811
812         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
813         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
814         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
815         sig->sched_time = 0;
816         INIT_LIST_HEAD(&sig->cpu_timers[0]);
817         INIT_LIST_HEAD(&sig->cpu_timers[1]);
818         INIT_LIST_HEAD(&sig->cpu_timers[2]);
819
820         task_lock(current->group_leader);
821         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
822         task_unlock(current->group_leader);
823
824         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
825                 /*
826                  * New sole thread in the process gets an expiry time
827                  * of the whole CPU time limit.
828                  */
829                 tsk->it_prof_expires =
830                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
831         }
832
833         return 0;
834 }
835
836 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
837 {
838         unsigned long new_flags = p->flags;
839
840         new_flags &= ~PF_SUPERPRIV;
841         new_flags |= PF_FORKNOEXEC;
842         if (!(clone_flags & CLONE_PTRACE))
843                 p->ptrace = 0;
844         p->flags = new_flags;
845 }
846
847 asmlinkage long sys_set_tid_address(int __user *tidptr)
848 {
849         current->clear_child_tid = tidptr;
850
851         return current->pid;
852 }
853
854 /*
855  * This creates a new process as a copy of the old one,
856  * but does not actually start it yet.
857  *
858  * It copies the registers, and all the appropriate
859  * parts of the process environment (as per the clone
860  * flags). The actual kick-off is left to the caller.
861  */
862 static task_t *copy_process(unsigned long clone_flags,
863                                  unsigned long stack_start,
864                                  struct pt_regs *regs,
865                                  unsigned long stack_size,
866                                  int __user *parent_tidptr,
867                                  int __user *child_tidptr,
868                                  int pid)
869 {
870         int retval;
871         struct task_struct *p = NULL;
872
873         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
874                 return ERR_PTR(-EINVAL);
875
876         /*
877          * Thread groups must share signals as well, and detached threads
878          * can only be started up within the thread group.
879          */
880         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
881                 return ERR_PTR(-EINVAL);
882
883         /*
884          * Shared signal handlers imply shared VM. By way of the above,
885          * thread groups also imply shared VM. Blocking this case allows
886          * for various simplifications in other code.
887          */
888         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
889                 return ERR_PTR(-EINVAL);
890
891         retval = security_task_create(clone_flags);
892         if (retval)
893                 goto fork_out;
894
895         retval = -ENOMEM;
896         p = dup_task_struct(current);
897         if (!p)
898                 goto fork_out;
899
900         retval = -EAGAIN;
901         if (atomic_read(&p->user->processes) >=
902                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
903                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
904                                 p->user != &root_user)
905                         goto bad_fork_free;
906         }
907
908         atomic_inc(&p->user->__count);
909         atomic_inc(&p->user->processes);
910         get_group_info(p->group_info);
911
912         /*
913          * If multiple threads are within copy_process(), then this check
914          * triggers too late. This doesn't hurt, the check is only there
915          * to stop root fork bombs.
916          */
917         if (nr_threads >= max_threads)
918                 goto bad_fork_cleanup_count;
919
920         if (!try_module_get(p->thread_info->exec_domain->module))
921                 goto bad_fork_cleanup_count;
922
923         if (p->binfmt && !try_module_get(p->binfmt->module))
924                 goto bad_fork_cleanup_put_domain;
925
926         p->did_exec = 0;
927         copy_flags(clone_flags, p);
928         p->pid = pid;
929         retval = -EFAULT;
930         if (clone_flags & CLONE_PARENT_SETTID)
931                 if (put_user(p->pid, parent_tidptr))
932                         goto bad_fork_cleanup;
933
934         p->proc_dentry = NULL;
935
936         INIT_LIST_HEAD(&p->children);
937         INIT_LIST_HEAD(&p->sibling);
938         p->vfork_done = NULL;
939         spin_lock_init(&p->alloc_lock);
940         spin_lock_init(&p->proc_lock);
941
942         clear_tsk_thread_flag(p, TIF_SIGPENDING);
943         init_sigpending(&p->pending);
944
945         p->utime = cputime_zero;
946         p->stime = cputime_zero;
947         p->sched_time = 0;
948         p->rchar = 0;           /* I/O counter: bytes read */
949         p->wchar = 0;           /* I/O counter: bytes written */
950         p->syscr = 0;           /* I/O counter: read syscalls */
951         p->syscw = 0;           /* I/O counter: write syscalls */
952         acct_clear_integrals(p);
953
954         p->it_virt_expires = cputime_zero;
955         p->it_prof_expires = cputime_zero;
956         p->it_sched_expires = 0;
957         INIT_LIST_HEAD(&p->cpu_timers[0]);
958         INIT_LIST_HEAD(&p->cpu_timers[1]);
959         INIT_LIST_HEAD(&p->cpu_timers[2]);
960
961         p->lock_depth = -1;             /* -1 = no lock */
962         do_posix_clock_monotonic_gettime(&p->start_time);
963         p->security = NULL;
964         p->io_context = NULL;
965         p->io_wait = NULL;
966         p->audit_context = NULL;
967 #ifdef CONFIG_NUMA
968         p->mempolicy = mpol_copy(p->mempolicy);
969         if (IS_ERR(p->mempolicy)) {
970                 retval = PTR_ERR(p->mempolicy);
971                 p->mempolicy = NULL;
972                 goto bad_fork_cleanup;
973         }
974 #endif
975
976         p->tgid = p->pid;
977         if (clone_flags & CLONE_THREAD)
978                 p->tgid = current->tgid;
979
980         if ((retval = security_task_alloc(p)))
981                 goto bad_fork_cleanup_policy;
982         if ((retval = audit_alloc(p)))
983                 goto bad_fork_cleanup_security;
984         /* copy all the process information */
985         if ((retval = copy_semundo(clone_flags, p)))
986                 goto bad_fork_cleanup_audit;
987         if ((retval = copy_files(clone_flags, p)))
988                 goto bad_fork_cleanup_semundo;
989         if ((retval = copy_fs(clone_flags, p)))
990                 goto bad_fork_cleanup_files;
991         if ((retval = copy_sighand(clone_flags, p)))
992                 goto bad_fork_cleanup_fs;
993         if ((retval = copy_signal(clone_flags, p)))
994                 goto bad_fork_cleanup_sighand;
995         if ((retval = copy_mm(clone_flags, p)))
996                 goto bad_fork_cleanup_signal;
997         if ((retval = copy_keys(clone_flags, p)))
998                 goto bad_fork_cleanup_mm;
999         if ((retval = copy_namespace(clone_flags, p)))
1000                 goto bad_fork_cleanup_keys;
1001         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1002         if (retval)
1003                 goto bad_fork_cleanup_namespace;
1004
1005         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1006         /*
1007          * Clear TID on mm_release()?
1008          */
1009         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1010
1011         /*
1012          * Syscall tracing should be turned off in the child regardless
1013          * of CLONE_PTRACE.
1014          */
1015         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1016 #ifdef TIF_SYSCALL_EMU
1017         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1018 #endif
1019
1020         /* Our parent execution domain becomes current domain
1021            These must match for thread signalling to apply */
1022            
1023         p->parent_exec_id = p->self_exec_id;
1024
1025         /* ok, now we should be set up.. */
1026         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1027         p->pdeath_signal = 0;
1028         p->exit_state = 0;
1029
1030         /*
1031          * Ok, make it visible to the rest of the system.
1032          * We dont wake it up yet.
1033          */
1034         p->group_leader = p;
1035         INIT_LIST_HEAD(&p->ptrace_children);
1036         INIT_LIST_HEAD(&p->ptrace_list);
1037
1038         /* Perform scheduler related setup. Assign this task to a CPU. */
1039         sched_fork(p, clone_flags);
1040
1041         /* Need tasklist lock for parent etc handling! */
1042         write_lock_irq(&tasklist_lock);
1043
1044         /*
1045          * The task hasn't been attached yet, so its cpus_allowed mask will
1046          * not be changed, nor will its assigned CPU.
1047          *
1048          * The cpus_allowed mask of the parent may have changed after it was
1049          * copied first time - so re-copy it here, then check the child's CPU
1050          * to ensure it is on a valid CPU (and if not, just force it back to
1051          * parent's CPU). This avoids alot of nasty races.
1052          */
1053         p->cpus_allowed = current->cpus_allowed;
1054         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed)))
1055                 set_task_cpu(p, smp_processor_id());
1056
1057         /*
1058          * Check for pending SIGKILL! The new thread should not be allowed
1059          * to slip out of an OOM kill. (or normal SIGKILL.)
1060          */
1061         if (sigismember(&current->pending.signal, SIGKILL)) {
1062                 write_unlock_irq(&tasklist_lock);
1063                 retval = -EINTR;
1064                 goto bad_fork_cleanup_namespace;
1065         }
1066
1067         /* CLONE_PARENT re-uses the old parent */
1068         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1069                 p->real_parent = current->real_parent;
1070         else
1071                 p->real_parent = current;
1072         p->parent = p->real_parent;
1073
1074         if (clone_flags & CLONE_THREAD) {
1075                 spin_lock(&current->sighand->siglock);
1076                 /*
1077                  * Important: if an exit-all has been started then
1078                  * do not create this new thread - the whole thread
1079                  * group is supposed to exit anyway.
1080                  */
1081                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1082                         spin_unlock(&current->sighand->siglock);
1083                         write_unlock_irq(&tasklist_lock);
1084                         retval = -EAGAIN;
1085                         goto bad_fork_cleanup_namespace;
1086                 }
1087                 p->group_leader = current->group_leader;
1088
1089                 if (current->signal->group_stop_count > 0) {
1090                         /*
1091                          * There is an all-stop in progress for the group.
1092                          * We ourselves will stop as soon as we check signals.
1093                          * Make the new thread part of that group stop too.
1094                          */
1095                         current->signal->group_stop_count++;
1096                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1097                 }
1098
1099                 if (!cputime_eq(current->signal->it_virt_expires,
1100                                 cputime_zero) ||
1101                     !cputime_eq(current->signal->it_prof_expires,
1102                                 cputime_zero) ||
1103                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1104                     !list_empty(&current->signal->cpu_timers[0]) ||
1105                     !list_empty(&current->signal->cpu_timers[1]) ||
1106                     !list_empty(&current->signal->cpu_timers[2])) {
1107                         /*
1108                          * Have child wake up on its first tick to check
1109                          * for process CPU timers.
1110                          */
1111                         p->it_prof_expires = jiffies_to_cputime(1);
1112                 }
1113
1114                 spin_unlock(&current->sighand->siglock);
1115         }
1116
1117         /*
1118          * inherit ioprio
1119          */
1120         p->ioprio = current->ioprio;
1121
1122         SET_LINKS(p);
1123         if (unlikely(p->ptrace & PT_PTRACED))
1124                 __ptrace_link(p, current->parent);
1125
1126         cpuset_fork(p);
1127
1128         attach_pid(p, PIDTYPE_PID, p->pid);
1129         attach_pid(p, PIDTYPE_TGID, p->tgid);
1130         if (thread_group_leader(p)) {
1131                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1132                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1133                 if (p->pid)
1134                         __get_cpu_var(process_counts)++;
1135         }
1136
1137         if (!current->signal->tty && p->signal->tty)
1138                 p->signal->tty = NULL;
1139
1140         nr_threads++;
1141         total_forks++;
1142         write_unlock_irq(&tasklist_lock);
1143         retval = 0;
1144
1145 fork_out:
1146         if (retval)
1147                 return ERR_PTR(retval);
1148         return p;
1149
1150 bad_fork_cleanup_namespace:
1151         exit_namespace(p);
1152 bad_fork_cleanup_keys:
1153         exit_keys(p);
1154 bad_fork_cleanup_mm:
1155         if (p->mm)
1156                 mmput(p->mm);
1157 bad_fork_cleanup_signal:
1158         exit_signal(p);
1159 bad_fork_cleanup_sighand:
1160         exit_sighand(p);
1161 bad_fork_cleanup_fs:
1162         exit_fs(p); /* blocking */
1163 bad_fork_cleanup_files:
1164         exit_files(p); /* blocking */
1165 bad_fork_cleanup_semundo:
1166         exit_sem(p);
1167 bad_fork_cleanup_audit:
1168         audit_free(p);
1169 bad_fork_cleanup_security:
1170         security_task_free(p);
1171 bad_fork_cleanup_policy:
1172 #ifdef CONFIG_NUMA
1173         mpol_free(p->mempolicy);
1174 #endif
1175 bad_fork_cleanup:
1176         if (p->binfmt)
1177                 module_put(p->binfmt->module);
1178 bad_fork_cleanup_put_domain:
1179         module_put(p->thread_info->exec_domain->module);
1180 bad_fork_cleanup_count:
1181         put_group_info(p->group_info);
1182         atomic_dec(&p->user->processes);
1183         free_uid(p->user);
1184 bad_fork_free:
1185         free_task(p);
1186         goto fork_out;
1187 }
1188
1189 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1190 {
1191         memset(regs, 0, sizeof(struct pt_regs));
1192         return regs;
1193 }
1194
1195 task_t * __devinit fork_idle(int cpu)
1196 {
1197         task_t *task;
1198         struct pt_regs regs;
1199
1200         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1201         if (!task)
1202                 return ERR_PTR(-ENOMEM);
1203         init_idle(task, cpu);
1204         unhash_process(task);
1205         return task;
1206 }
1207
1208 static inline int fork_traceflag (unsigned clone_flags)
1209 {
1210         if (clone_flags & CLONE_UNTRACED)
1211                 return 0;
1212         else if (clone_flags & CLONE_VFORK) {
1213                 if (current->ptrace & PT_TRACE_VFORK)
1214                         return PTRACE_EVENT_VFORK;
1215         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1216                 if (current->ptrace & PT_TRACE_CLONE)
1217                         return PTRACE_EVENT_CLONE;
1218         } else if (current->ptrace & PT_TRACE_FORK)
1219                 return PTRACE_EVENT_FORK;
1220
1221         return 0;
1222 }
1223
1224 /*
1225  *  Ok, this is the main fork-routine.
1226  *
1227  * It copies the process, and if successful kick-starts
1228  * it and waits for it to finish using the VM if required.
1229  */
1230 long do_fork(unsigned long clone_flags,
1231               unsigned long stack_start,
1232               struct pt_regs *regs,
1233               unsigned long stack_size,
1234               int __user *parent_tidptr,
1235               int __user *child_tidptr)
1236 {
1237         struct task_struct *p;
1238         int trace = 0;
1239         long pid = alloc_pidmap();
1240
1241         if (pid < 0)
1242                 return -EAGAIN;
1243         if (unlikely(current->ptrace)) {
1244                 trace = fork_traceflag (clone_flags);
1245                 if (trace)
1246                         clone_flags |= CLONE_PTRACE;
1247         }
1248
1249         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1250         /*
1251          * Do this prior waking up the new thread - the thread pointer
1252          * might get invalid after that point, if the thread exits quickly.
1253          */
1254         if (!IS_ERR(p)) {
1255                 struct completion vfork;
1256
1257                 if (clone_flags & CLONE_VFORK) {
1258                         p->vfork_done = &vfork;
1259                         init_completion(&vfork);
1260                 }
1261
1262                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1263                         /*
1264                          * We'll start up with an immediate SIGSTOP.
1265                          */
1266                         sigaddset(&p->pending.signal, SIGSTOP);
1267                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1268                 }
1269
1270                 if (!(clone_flags & CLONE_STOPPED))
1271                         wake_up_new_task(p, clone_flags);
1272                 else
1273                         p->state = TASK_STOPPED;
1274
1275                 if (unlikely (trace)) {
1276                         current->ptrace_message = pid;
1277                         ptrace_notify ((trace << 8) | SIGTRAP);
1278                 }
1279
1280                 if (clone_flags & CLONE_VFORK) {
1281                         wait_for_completion(&vfork);
1282                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1283                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1284                 }
1285         } else {
1286                 free_pidmap(pid);
1287                 pid = PTR_ERR(p);
1288         }
1289         return pid;
1290 }
1291
1292 void __init proc_caches_init(void)
1293 {
1294         sighand_cachep = kmem_cache_create("sighand_cache",
1295                         sizeof(struct sighand_struct), 0,
1296                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1297         signal_cachep = kmem_cache_create("signal_cache",
1298                         sizeof(struct signal_struct), 0,
1299                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1300         files_cachep = kmem_cache_create("files_cache", 
1301                         sizeof(struct files_struct), 0,
1302                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1303         fs_cachep = kmem_cache_create("fs_cache", 
1304                         sizeof(struct fs_struct), 0,
1305                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1306         vm_area_cachep = kmem_cache_create("vm_area_struct",
1307                         sizeof(struct vm_area_struct), 0,
1308                         SLAB_PANIC, NULL, NULL);
1309         mm_cachep = kmem_cache_create("mm_struct",
1310                         sizeof(struct mm_struct), 0,
1311                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1312 }