Merge ../linux-2.6
[linux-2.6] / 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/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
47 #include <asm/pgtable.h>
48 #include <asm/pgalloc.h>
49 #include <asm/uaccess.h>
50 #include <asm/mmu_context.h>
51 #include <asm/cacheflush.h>
52 #include <asm/tlbflush.h>
53
54 /*
55  * Protected counters by write_lock_irq(&tasklist_lock)
56  */
57 unsigned long total_forks;      /* Handle normal Linux uptimes. */
58 int nr_threads;                 /* The idle threads do not count.. */
59
60 int max_threads;                /* tunable limit on nr_threads */
61
62 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
63
64  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
65
66 EXPORT_SYMBOL(tasklist_lock);
67
68 int nr_processes(void)
69 {
70         int cpu;
71         int total = 0;
72
73         for_each_online_cpu(cpu)
74                 total += per_cpu(process_counts, cpu);
75
76         return total;
77 }
78
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
84
85 /* SLAB cache for signal_struct structures (tsk->signal) */
86 kmem_cache_t *signal_cachep;
87
88 /* SLAB cache for sighand_struct structures (tsk->sighand) */
89 kmem_cache_t *sighand_cachep;
90
91 /* SLAB cache for files_struct structures (tsk->files) */
92 kmem_cache_t *files_cachep;
93
94 /* SLAB cache for fs_struct structures (tsk->fs) */
95 kmem_cache_t *fs_cachep;
96
97 /* SLAB cache for vm_area_struct structures */
98 kmem_cache_t *vm_area_cachep;
99
100 /* SLAB cache for mm_struct structures (tsk->mm) */
101 static kmem_cache_t *mm_cachep;
102
103 void free_task(struct task_struct *tsk)
104 {
105         free_thread_info(tsk->thread_info);
106         free_task_struct(tsk);
107 }
108 EXPORT_SYMBOL(free_task);
109
110 void __put_task_struct(struct task_struct *tsk)
111 {
112         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
113         WARN_ON(atomic_read(&tsk->usage));
114         WARN_ON(tsk == current);
115
116         if (unlikely(tsk->audit_context))
117                 audit_free(tsk);
118         security_task_free(tsk);
119         free_uid(tsk->user);
120         put_group_info(tsk->group_info);
121
122         if (!profile_handoff_task(tsk))
123                 free_task(tsk);
124 }
125
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
137
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);
144
145         /*
146          * we need to allow at least 20 threads to boot a system
147          */
148         if(max_threads < 20)
149                 max_threads = 20;
150
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 }
156
157 static struct task_struct *dup_task_struct(struct task_struct *orig)
158 {
159         struct task_struct *tsk;
160         struct thread_info *ti;
161
162         prepare_to_copy(orig);
163
164         tsk = alloc_task_struct();
165         if (!tsk)
166                 return NULL;
167
168         ti = alloc_thread_info(tsk);
169         if (!ti) {
170                 free_task_struct(tsk);
171                 return NULL;
172         }
173
174         *ti = *orig->thread_info;
175         *tsk = *orig;
176         tsk->thread_info = ti;
177         ti->task = tsk;
178
179         /* One for us, one for whoever does the "release_task()" (usually parent) */
180         atomic_set(&tsk->usage,2);
181         atomic_set(&tsk->fs_excl, 0);
182         return tsk;
183 }
184
185 #ifdef CONFIG_MMU
186 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
187 {
188         struct vm_area_struct *mpnt, *tmp, **pprev;
189         struct rb_node **rb_link, *rb_parent;
190         int retval;
191         unsigned long charge;
192         struct mempolicy *pol;
193
194         down_write(&oldmm->mmap_sem);
195         flush_cache_mm(oldmm);
196         down_write(&mm->mmap_sem);
197
198         mm->locked_vm = 0;
199         mm->mmap = NULL;
200         mm->mmap_cache = NULL;
201         mm->free_area_cache = oldmm->mmap_base;
202         mm->cached_hole_size = ~0UL;
203         mm->map_count = 0;
204         cpus_clear(mm->cpu_vm_mask);
205         mm->mm_rb = RB_ROOT;
206         rb_link = &mm->mm_rb.rb_node;
207         rb_parent = NULL;
208         pprev = &mm->mmap;
209
210         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
211                 struct file *file;
212
213                 if (mpnt->vm_flags & VM_DONTCOPY) {
214                         long pages = vma_pages(mpnt);
215                         mm->total_vm -= pages;
216                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
217                                                                 -pages);
218                         continue;
219                 }
220                 charge = 0;
221                 if (mpnt->vm_flags & VM_ACCOUNT) {
222                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
223                         if (security_vm_enough_memory(len))
224                                 goto fail_nomem;
225                         charge = len;
226                 }
227                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
228                 if (!tmp)
229                         goto fail_nomem;
230                 *tmp = *mpnt;
231                 pol = mpol_copy(vma_policy(mpnt));
232                 retval = PTR_ERR(pol);
233                 if (IS_ERR(pol))
234                         goto fail_nomem_policy;
235                 vma_set_policy(tmp, pol);
236                 tmp->vm_flags &= ~VM_LOCKED;
237                 tmp->vm_mm = mm;
238                 tmp->vm_next = NULL;
239                 anon_vma_link(tmp);
240                 file = tmp->vm_file;
241                 if (file) {
242                         struct inode *inode = file->f_dentry->d_inode;
243                         get_file(file);
244                         if (tmp->vm_flags & VM_DENYWRITE)
245                                 atomic_dec(&inode->i_writecount);
246       
247                         /* insert tmp into the share list, just after mpnt */
248                         spin_lock(&file->f_mapping->i_mmap_lock);
249                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
250                         flush_dcache_mmap_lock(file->f_mapping);
251                         vma_prio_tree_add(tmp, mpnt);
252                         flush_dcache_mmap_unlock(file->f_mapping);
253                         spin_unlock(&file->f_mapping->i_mmap_lock);
254                 }
255
256                 /*
257                  * Link in the new vma and copy the page table entries.
258                  */
259                 *pprev = tmp;
260                 pprev = &tmp->vm_next;
261
262                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
263                 rb_link = &tmp->vm_rb.rb_right;
264                 rb_parent = &tmp->vm_rb;
265
266                 mm->map_count++;
267                 retval = copy_page_range(mm, oldmm, tmp);
268
269                 if (tmp->vm_ops && tmp->vm_ops->open)
270                         tmp->vm_ops->open(tmp);
271
272                 if (retval)
273                         goto out;
274         }
275         retval = 0;
276 out:
277         up_write(&mm->mmap_sem);
278         flush_tlb_mm(oldmm);
279         up_write(&oldmm->mmap_sem);
280         return retval;
281 fail_nomem_policy:
282         kmem_cache_free(vm_area_cachep, tmp);
283 fail_nomem:
284         retval = -ENOMEM;
285         vm_unacct_memory(charge);
286         goto out;
287 }
288
289 static inline int mm_alloc_pgd(struct mm_struct * mm)
290 {
291         mm->pgd = pgd_alloc(mm);
292         if (unlikely(!mm->pgd))
293                 return -ENOMEM;
294         return 0;
295 }
296
297 static inline void mm_free_pgd(struct mm_struct * mm)
298 {
299         pgd_free(mm->pgd);
300 }
301 #else
302 #define dup_mmap(mm, oldmm)     (0)
303 #define mm_alloc_pgd(mm)        (0)
304 #define mm_free_pgd(mm)
305 #endif /* CONFIG_MMU */
306
307  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
308
309 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
310 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
311
312 #include <linux/init_task.h>
313
314 static struct mm_struct * mm_init(struct mm_struct * mm)
315 {
316         atomic_set(&mm->mm_users, 1);
317         atomic_set(&mm->mm_count, 1);
318         init_rwsem(&mm->mmap_sem);
319         INIT_LIST_HEAD(&mm->mmlist);
320         mm->core_waiters = 0;
321         mm->nr_ptes = 0;
322         set_mm_counter(mm, file_rss, 0);
323         set_mm_counter(mm, anon_rss, 0);
324         spin_lock_init(&mm->page_table_lock);
325         rwlock_init(&mm->ioctx_list_lock);
326         mm->ioctx_list = NULL;
327         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
328         mm->free_area_cache = TASK_UNMAPPED_BASE;
329         mm->cached_hole_size = ~0UL;
330
331         if (likely(!mm_alloc_pgd(mm))) {
332                 mm->def_flags = 0;
333                 return mm;
334         }
335         free_mm(mm);
336         return NULL;
337 }
338
339 /*
340  * Allocate and initialize an mm_struct.
341  */
342 struct mm_struct * mm_alloc(void)
343 {
344         struct mm_struct * mm;
345
346         mm = allocate_mm();
347         if (mm) {
348                 memset(mm, 0, sizeof(*mm));
349                 mm = mm_init(mm);
350         }
351         return mm;
352 }
353
354 /*
355  * Called when the last reference to the mm
356  * is dropped: either by a lazy thread or by
357  * mmput. Free the page directory and the mm.
358  */
359 void fastcall __mmdrop(struct mm_struct *mm)
360 {
361         BUG_ON(mm == &init_mm);
362         mm_free_pgd(mm);
363         destroy_context(mm);
364         free_mm(mm);
365 }
366
367 /*
368  * Decrement the use count and release all resources for an mm.
369  */
370 void mmput(struct mm_struct *mm)
371 {
372         if (atomic_dec_and_test(&mm->mm_users)) {
373                 exit_aio(mm);
374                 exit_mmap(mm);
375                 if (!list_empty(&mm->mmlist)) {
376                         spin_lock(&mmlist_lock);
377                         list_del(&mm->mmlist);
378                         spin_unlock(&mmlist_lock);
379                 }
380                 put_swap_token(mm);
381                 mmdrop(mm);
382         }
383 }
384 EXPORT_SYMBOL_GPL(mmput);
385
386 /**
387  * get_task_mm - acquire a reference to the task's mm
388  *
389  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
390  * this kernel workthread has transiently adopted a user mm with use_mm,
391  * to do its AIO) is not set and if so returns a reference to it, after
392  * bumping up the use count.  User must release the mm via mmput()
393  * after use.  Typically used by /proc and ptrace.
394  */
395 struct mm_struct *get_task_mm(struct task_struct *task)
396 {
397         struct mm_struct *mm;
398
399         task_lock(task);
400         mm = task->mm;
401         if (mm) {
402                 if (task->flags & PF_BORROWED_MM)
403                         mm = NULL;
404                 else
405                         atomic_inc(&mm->mm_users);
406         }
407         task_unlock(task);
408         return mm;
409 }
410 EXPORT_SYMBOL_GPL(get_task_mm);
411
412 /* Please note the differences between mmput and mm_release.
413  * mmput is called whenever we stop holding onto a mm_struct,
414  * error success whatever.
415  *
416  * mm_release is called after a mm_struct has been removed
417  * from the current process.
418  *
419  * This difference is important for error handling, when we
420  * only half set up a mm_struct for a new process and need to restore
421  * the old one.  Because we mmput the new mm_struct before
422  * restoring the old one. . .
423  * Eric Biederman 10 January 1998
424  */
425 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
426 {
427         struct completion *vfork_done = tsk->vfork_done;
428
429         /* Get rid of any cached register state */
430         deactivate_mm(tsk, mm);
431
432         /* notify parent sleeping on vfork() */
433         if (vfork_done) {
434                 tsk->vfork_done = NULL;
435                 complete(vfork_done);
436         }
437         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
438                 u32 __user * tidptr = tsk->clear_child_tid;
439                 tsk->clear_child_tid = NULL;
440
441                 /*
442                  * We don't check the error code - if userspace has
443                  * not set up a proper pointer then tough luck.
444                  */
445                 put_user(0, tidptr);
446                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
447         }
448 }
449
450 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
451 {
452         struct mm_struct * mm, *oldmm;
453         int retval;
454
455         tsk->min_flt = tsk->maj_flt = 0;
456         tsk->nvcsw = tsk->nivcsw = 0;
457
458         tsk->mm = NULL;
459         tsk->active_mm = NULL;
460
461         /*
462          * Are we cloning a kernel thread?
463          *
464          * We need to steal a active VM for that..
465          */
466         oldmm = current->mm;
467         if (!oldmm)
468                 return 0;
469
470         if (clone_flags & CLONE_VM) {
471                 atomic_inc(&oldmm->mm_users);
472                 mm = oldmm;
473                 goto good_mm;
474         }
475
476         retval = -ENOMEM;
477         mm = allocate_mm();
478         if (!mm)
479                 goto fail_nomem;
480
481         /* Copy the current MM stuff.. */
482         memcpy(mm, oldmm, sizeof(*mm));
483         if (!mm_init(mm))
484                 goto fail_nomem;
485
486         if (init_new_context(tsk,mm))
487                 goto fail_nocontext;
488
489         retval = dup_mmap(mm, oldmm);
490         if (retval)
491                 goto free_pt;
492
493         mm->hiwater_rss = get_mm_rss(mm);
494         mm->hiwater_vm = mm->total_vm;
495
496 good_mm:
497         tsk->mm = mm;
498         tsk->active_mm = mm;
499         return 0;
500
501 free_pt:
502         mmput(mm);
503 fail_nomem:
504         return retval;
505
506 fail_nocontext:
507         /*
508          * If init_new_context() failed, we cannot use mmput() to free the mm
509          * because it calls destroy_context()
510          */
511         mm_free_pgd(mm);
512         free_mm(mm);
513         return retval;
514 }
515
516 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
517 {
518         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
519         /* We don't need to lock fs - think why ;-) */
520         if (fs) {
521                 atomic_set(&fs->count, 1);
522                 rwlock_init(&fs->lock);
523                 fs->umask = old->umask;
524                 read_lock(&old->lock);
525                 fs->rootmnt = mntget(old->rootmnt);
526                 fs->root = dget(old->root);
527                 fs->pwdmnt = mntget(old->pwdmnt);
528                 fs->pwd = dget(old->pwd);
529                 if (old->altroot) {
530                         fs->altrootmnt = mntget(old->altrootmnt);
531                         fs->altroot = dget(old->altroot);
532                 } else {
533                         fs->altrootmnt = NULL;
534                         fs->altroot = NULL;
535                 }
536                 read_unlock(&old->lock);
537         }
538         return fs;
539 }
540
541 struct fs_struct *copy_fs_struct(struct fs_struct *old)
542 {
543         return __copy_fs_struct(old);
544 }
545
546 EXPORT_SYMBOL_GPL(copy_fs_struct);
547
548 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
549 {
550         if (clone_flags & CLONE_FS) {
551                 atomic_inc(&current->fs->count);
552                 return 0;
553         }
554         tsk->fs = __copy_fs_struct(current->fs);
555         if (!tsk->fs)
556                 return -ENOMEM;
557         return 0;
558 }
559
560 static int count_open_files(struct fdtable *fdt)
561 {
562         int size = fdt->max_fdset;
563         int i;
564
565         /* Find the last open fd */
566         for (i = size/(8*sizeof(long)); i > 0; ) {
567                 if (fdt->open_fds->fds_bits[--i])
568                         break;
569         }
570         i = (i+1) * 8 * sizeof(long);
571         return i;
572 }
573
574 static struct files_struct *alloc_files(void)
575 {
576         struct files_struct *newf;
577         struct fdtable *fdt;
578
579         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
580         if (!newf)
581                 goto out;
582
583         atomic_set(&newf->count, 1);
584
585         spin_lock_init(&newf->file_lock);
586         fdt = &newf->fdtab;
587         fdt->next_fd = 0;
588         fdt->max_fds = NR_OPEN_DEFAULT;
589         fdt->max_fdset = __FD_SETSIZE;
590         fdt->close_on_exec = &newf->close_on_exec_init;
591         fdt->open_fds = &newf->open_fds_init;
592         fdt->fd = &newf->fd_array[0];
593         INIT_RCU_HEAD(&fdt->rcu);
594         fdt->free_files = NULL;
595         fdt->next = NULL;
596         rcu_assign_pointer(newf->fdt, fdt);
597 out:
598         return newf;
599 }
600
601 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
602 {
603         struct files_struct *oldf, *newf;
604         struct file **old_fds, **new_fds;
605         int open_files, size, i, error = 0, expand;
606         struct fdtable *old_fdt, *new_fdt;
607
608         /*
609          * A background process may not have any files ...
610          */
611         oldf = current->files;
612         if (!oldf)
613                 goto out;
614
615         if (clone_flags & CLONE_FILES) {
616                 atomic_inc(&oldf->count);
617                 goto out;
618         }
619
620         /*
621          * Note: we may be using current for both targets (See exec.c)
622          * This works because we cache current->files (old) as oldf. Don't
623          * break this.
624          */
625         tsk->files = NULL;
626         error = -ENOMEM;
627         newf = alloc_files();
628         if (!newf)
629                 goto out;
630
631         spin_lock(&oldf->file_lock);
632         old_fdt = files_fdtable(oldf);
633         new_fdt = files_fdtable(newf);
634         size = old_fdt->max_fdset;
635         open_files = count_open_files(old_fdt);
636         expand = 0;
637
638         /*
639          * Check whether we need to allocate a larger fd array or fd set.
640          * Note: we're not a clone task, so the open count won't  change.
641          */
642         if (open_files > new_fdt->max_fdset) {
643                 new_fdt->max_fdset = 0;
644                 expand = 1;
645         }
646         if (open_files > new_fdt->max_fds) {
647                 new_fdt->max_fds = 0;
648                 expand = 1;
649         }
650
651         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
652         if (expand) {
653                 spin_unlock(&oldf->file_lock);
654                 spin_lock(&newf->file_lock);
655                 error = expand_files(newf, open_files-1);
656                 spin_unlock(&newf->file_lock);
657                 if (error < 0)
658                         goto out_release;
659                 new_fdt = files_fdtable(newf);
660                 /*
661                  * Reacquire the oldf lock and a pointer to its fd table
662                  * who knows it may have a new bigger fd table. We need
663                  * the latest pointer.
664                  */
665                 spin_lock(&oldf->file_lock);
666                 old_fdt = files_fdtable(oldf);
667         }
668
669         old_fds = old_fdt->fd;
670         new_fds = new_fdt->fd;
671
672         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
673         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
674
675         for (i = open_files; i != 0; i--) {
676                 struct file *f = *old_fds++;
677                 if (f) {
678                         get_file(f);
679                 } else {
680                         /*
681                          * The fd may be claimed in the fd bitmap but not yet
682                          * instantiated in the files array if a sibling thread
683                          * is partway through open().  So make sure that this
684                          * fd is available to the new process.
685                          */
686                         FD_CLR(open_files - i, new_fdt->open_fds);
687                 }
688                 rcu_assign_pointer(*new_fds++, f);
689         }
690         spin_unlock(&oldf->file_lock);
691
692         /* compute the remainder to be cleared */
693         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
694
695         /* This is long word aligned thus could use a optimized version */ 
696         memset(new_fds, 0, size); 
697
698         if (new_fdt->max_fdset > open_files) {
699                 int left = (new_fdt->max_fdset-open_files)/8;
700                 int start = open_files / (8 * sizeof(unsigned long));
701
702                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
703                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
704         }
705
706         tsk->files = newf;
707         error = 0;
708 out:
709         return error;
710
711 out_release:
712         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
713         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
714         free_fd_array(new_fdt->fd, new_fdt->max_fds);
715         kmem_cache_free(files_cachep, newf);
716         goto out;
717 }
718
719 /*
720  *      Helper to unshare the files of the current task.
721  *      We don't want to expose copy_files internals to
722  *      the exec layer of the kernel.
723  */
724
725 int unshare_files(void)
726 {
727         struct files_struct *files  = current->files;
728         int rc;
729
730         if(!files)
731                 BUG();
732
733         /* This can race but the race causes us to copy when we don't
734            need to and drop the copy */
735         if(atomic_read(&files->count) == 1)
736         {
737                 atomic_inc(&files->count);
738                 return 0;
739         }
740         rc = copy_files(0, current);
741         if(rc)
742                 current->files = files;
743         return rc;
744 }
745
746 EXPORT_SYMBOL(unshare_files);
747
748 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
749 {
750         struct sighand_struct *sig;
751
752         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
753                 atomic_inc(&current->sighand->count);
754                 return 0;
755         }
756         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
757         tsk->sighand = sig;
758         if (!sig)
759                 return -ENOMEM;
760         spin_lock_init(&sig->siglock);
761         atomic_set(&sig->count, 1);
762         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
763         return 0;
764 }
765
766 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
767 {
768         struct signal_struct *sig;
769         int ret;
770
771         if (clone_flags & CLONE_THREAD) {
772                 atomic_inc(&current->signal->count);
773                 atomic_inc(&current->signal->live);
774                 return 0;
775         }
776         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
777         tsk->signal = sig;
778         if (!sig)
779                 return -ENOMEM;
780
781         ret = copy_thread_group_keys(tsk);
782         if (ret < 0) {
783                 kmem_cache_free(signal_cachep, sig);
784                 return ret;
785         }
786
787         atomic_set(&sig->count, 1);
788         atomic_set(&sig->live, 1);
789         init_waitqueue_head(&sig->wait_chldexit);
790         sig->flags = 0;
791         sig->group_exit_code = 0;
792         sig->group_exit_task = NULL;
793         sig->group_stop_count = 0;
794         sig->curr_target = NULL;
795         init_sigpending(&sig->shared_pending);
796         INIT_LIST_HEAD(&sig->posix_timers);
797
798         sig->it_real_value = sig->it_real_incr = 0;
799         sig->real_timer.function = it_real_fn;
800         sig->real_timer.data = (unsigned long) tsk;
801         init_timer(&sig->real_timer);
802
803         sig->it_virt_expires = cputime_zero;
804         sig->it_virt_incr = cputime_zero;
805         sig->it_prof_expires = cputime_zero;
806         sig->it_prof_incr = cputime_zero;
807
808         sig->tty = current->signal->tty;
809         sig->pgrp = process_group(current);
810         sig->session = current->signal->session;
811         sig->leader = 0;        /* session leadership doesn't inherit */
812         sig->tty_old_pgrp = 0;
813
814         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
815         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
816         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
817         sig->sched_time = 0;
818         INIT_LIST_HEAD(&sig->cpu_timers[0]);
819         INIT_LIST_HEAD(&sig->cpu_timers[1]);
820         INIT_LIST_HEAD(&sig->cpu_timers[2]);
821
822         task_lock(current->group_leader);
823         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
824         task_unlock(current->group_leader);
825
826         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
827                 /*
828                  * New sole thread in the process gets an expiry time
829                  * of the whole CPU time limit.
830                  */
831                 tsk->it_prof_expires =
832                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
833         }
834
835         return 0;
836 }
837
838 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
839 {
840         unsigned long new_flags = p->flags;
841
842         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
843         new_flags |= PF_FORKNOEXEC;
844         if (!(clone_flags & CLONE_PTRACE))
845                 p->ptrace = 0;
846         p->flags = new_flags;
847 }
848
849 asmlinkage long sys_set_tid_address(int __user *tidptr)
850 {
851         current->clear_child_tid = tidptr;
852
853         return current->pid;
854 }
855
856 /*
857  * This creates a new process as a copy of the old one,
858  * but does not actually start it yet.
859  *
860  * It copies the registers, and all the appropriate
861  * parts of the process environment (as per the clone
862  * flags). The actual kick-off is left to the caller.
863  */
864 static task_t *copy_process(unsigned long clone_flags,
865                                  unsigned long stack_start,
866                                  struct pt_regs *regs,
867                                  unsigned long stack_size,
868                                  int __user *parent_tidptr,
869                                  int __user *child_tidptr,
870                                  int pid)
871 {
872         int retval;
873         struct task_struct *p = NULL;
874
875         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
876                 return ERR_PTR(-EINVAL);
877
878         /*
879          * Thread groups must share signals as well, and detached threads
880          * can only be started up within the thread group.
881          */
882         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
883                 return ERR_PTR(-EINVAL);
884
885         /*
886          * Shared signal handlers imply shared VM. By way of the above,
887          * thread groups also imply shared VM. Blocking this case allows
888          * for various simplifications in other code.
889          */
890         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
891                 return ERR_PTR(-EINVAL);
892
893         retval = security_task_create(clone_flags);
894         if (retval)
895                 goto fork_out;
896
897         retval = -ENOMEM;
898         p = dup_task_struct(current);
899         if (!p)
900                 goto fork_out;
901
902         retval = -EAGAIN;
903         if (atomic_read(&p->user->processes) >=
904                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
905                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
906                                 p->user != &root_user)
907                         goto bad_fork_free;
908         }
909
910         atomic_inc(&p->user->__count);
911         atomic_inc(&p->user->processes);
912         get_group_info(p->group_info);
913
914         /*
915          * If multiple threads are within copy_process(), then this check
916          * triggers too late. This doesn't hurt, the check is only there
917          * to stop root fork bombs.
918          */
919         if (nr_threads >= max_threads)
920                 goto bad_fork_cleanup_count;
921
922         if (!try_module_get(p->thread_info->exec_domain->module))
923                 goto bad_fork_cleanup_count;
924
925         if (p->binfmt && !try_module_get(p->binfmt->module))
926                 goto bad_fork_cleanup_put_domain;
927
928         p->did_exec = 0;
929         copy_flags(clone_flags, p);
930         p->pid = pid;
931         retval = -EFAULT;
932         if (clone_flags & CLONE_PARENT_SETTID)
933                 if (put_user(p->pid, parent_tidptr))
934                         goto bad_fork_cleanup;
935
936         p->proc_dentry = NULL;
937
938         INIT_LIST_HEAD(&p->children);
939         INIT_LIST_HEAD(&p->sibling);
940         p->vfork_done = NULL;
941         spin_lock_init(&p->alloc_lock);
942         spin_lock_init(&p->proc_lock);
943
944         clear_tsk_thread_flag(p, TIF_SIGPENDING);
945         init_sigpending(&p->pending);
946
947         p->utime = cputime_zero;
948         p->stime = cputime_zero;
949         p->sched_time = 0;
950         p->rchar = 0;           /* I/O counter: bytes read */
951         p->wchar = 0;           /* I/O counter: bytes written */
952         p->syscr = 0;           /* I/O counter: read syscalls */
953         p->syscw = 0;           /* I/O counter: write syscalls */
954         acct_clear_integrals(p);
955
956         p->it_virt_expires = cputime_zero;
957         p->it_prof_expires = cputime_zero;
958         p->it_sched_expires = 0;
959         INIT_LIST_HEAD(&p->cpu_timers[0]);
960         INIT_LIST_HEAD(&p->cpu_timers[1]);
961         INIT_LIST_HEAD(&p->cpu_timers[2]);
962
963         p->lock_depth = -1;             /* -1 = no lock */
964         do_posix_clock_monotonic_gettime(&p->start_time);
965         p->security = NULL;
966         p->io_context = NULL;
967         p->io_wait = NULL;
968         p->audit_context = NULL;
969 #ifdef CONFIG_NUMA
970         p->mempolicy = mpol_copy(p->mempolicy);
971         if (IS_ERR(p->mempolicy)) {
972                 retval = PTR_ERR(p->mempolicy);
973                 p->mempolicy = NULL;
974                 goto bad_fork_cleanup;
975         }
976 #endif
977
978         p->tgid = p->pid;
979         if (clone_flags & CLONE_THREAD)
980                 p->tgid = current->tgid;
981
982         if ((retval = security_task_alloc(p)))
983                 goto bad_fork_cleanup_policy;
984         if ((retval = audit_alloc(p)))
985                 goto bad_fork_cleanup_security;
986         /* copy all the process information */
987         if ((retval = copy_semundo(clone_flags, p)))
988                 goto bad_fork_cleanup_audit;
989         if ((retval = copy_files(clone_flags, p)))
990                 goto bad_fork_cleanup_semundo;
991         if ((retval = copy_fs(clone_flags, p)))
992                 goto bad_fork_cleanup_files;
993         if ((retval = copy_sighand(clone_flags, p)))
994                 goto bad_fork_cleanup_fs;
995         if ((retval = copy_signal(clone_flags, p)))
996                 goto bad_fork_cleanup_sighand;
997         if ((retval = copy_mm(clone_flags, p)))
998                 goto bad_fork_cleanup_signal;
999         if ((retval = copy_keys(clone_flags, p)))
1000                 goto bad_fork_cleanup_mm;
1001         if ((retval = copy_namespace(clone_flags, p)))
1002                 goto bad_fork_cleanup_keys;
1003         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1004         if (retval)
1005                 goto bad_fork_cleanup_namespace;
1006
1007         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1008         /*
1009          * Clear TID on mm_release()?
1010          */
1011         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1012
1013         /*
1014          * Syscall tracing should be turned off in the child regardless
1015          * of CLONE_PTRACE.
1016          */
1017         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1018 #ifdef TIF_SYSCALL_EMU
1019         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1020 #endif
1021
1022         /* Our parent execution domain becomes current domain
1023            These must match for thread signalling to apply */
1024            
1025         p->parent_exec_id = p->self_exec_id;
1026
1027         /* ok, now we should be set up.. */
1028         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1029         p->pdeath_signal = 0;
1030         p->exit_state = 0;
1031
1032         /*
1033          * Ok, make it visible to the rest of the system.
1034          * We dont wake it up yet.
1035          */
1036         p->group_leader = p;
1037         INIT_LIST_HEAD(&p->ptrace_children);
1038         INIT_LIST_HEAD(&p->ptrace_list);
1039
1040         /* Perform scheduler related setup. Assign this task to a CPU. */
1041         sched_fork(p, clone_flags);
1042
1043         /* Need tasklist lock for parent etc handling! */
1044         write_lock_irq(&tasklist_lock);
1045
1046         /*
1047          * The task hasn't been attached yet, so its cpus_allowed mask will
1048          * not be changed, nor will its assigned CPU.
1049          *
1050          * The cpus_allowed mask of the parent may have changed after it was
1051          * copied first time - so re-copy it here, then check the child's CPU
1052          * to ensure it is on a valid CPU (and if not, just force it back to
1053          * parent's CPU). This avoids alot of nasty races.
1054          */
1055         p->cpus_allowed = current->cpus_allowed;
1056         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1057                         !cpu_online(task_cpu(p))))
1058                 set_task_cpu(p, smp_processor_id());
1059
1060         /*
1061          * Check for pending SIGKILL! The new thread should not be allowed
1062          * to slip out of an OOM kill. (or normal SIGKILL.)
1063          */
1064         if (sigismember(&current->pending.signal, SIGKILL)) {
1065                 write_unlock_irq(&tasklist_lock);
1066                 retval = -EINTR;
1067                 goto bad_fork_cleanup_namespace;
1068         }
1069
1070         /* CLONE_PARENT re-uses the old parent */
1071         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1072                 p->real_parent = current->real_parent;
1073         else
1074                 p->real_parent = current;
1075         p->parent = p->real_parent;
1076
1077         if (clone_flags & CLONE_THREAD) {
1078                 spin_lock(&current->sighand->siglock);
1079                 /*
1080                  * Important: if an exit-all has been started then
1081                  * do not create this new thread - the whole thread
1082                  * group is supposed to exit anyway.
1083                  */
1084                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1085                         spin_unlock(&current->sighand->siglock);
1086                         write_unlock_irq(&tasklist_lock);
1087                         retval = -EAGAIN;
1088                         goto bad_fork_cleanup_namespace;
1089                 }
1090                 p->group_leader = current->group_leader;
1091
1092                 if (current->signal->group_stop_count > 0) {
1093                         /*
1094                          * There is an all-stop in progress for the group.
1095                          * We ourselves will stop as soon as we check signals.
1096                          * Make the new thread part of that group stop too.
1097                          */
1098                         current->signal->group_stop_count++;
1099                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1100                 }
1101
1102                 if (!cputime_eq(current->signal->it_virt_expires,
1103                                 cputime_zero) ||
1104                     !cputime_eq(current->signal->it_prof_expires,
1105                                 cputime_zero) ||
1106                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1107                     !list_empty(&current->signal->cpu_timers[0]) ||
1108                     !list_empty(&current->signal->cpu_timers[1]) ||
1109                     !list_empty(&current->signal->cpu_timers[2])) {
1110                         /*
1111                          * Have child wake up on its first tick to check
1112                          * for process CPU timers.
1113                          */
1114                         p->it_prof_expires = jiffies_to_cputime(1);
1115                 }
1116
1117                 spin_unlock(&current->sighand->siglock);
1118         }
1119
1120         /*
1121          * inherit ioprio
1122          */
1123         p->ioprio = current->ioprio;
1124
1125         SET_LINKS(p);
1126         if (unlikely(p->ptrace & PT_PTRACED))
1127                 __ptrace_link(p, current->parent);
1128
1129         cpuset_fork(p);
1130
1131         attach_pid(p, PIDTYPE_PID, p->pid);
1132         attach_pid(p, PIDTYPE_TGID, p->tgid);
1133         if (thread_group_leader(p)) {
1134                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1135                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1136                 if (p->pid)
1137                         __get_cpu_var(process_counts)++;
1138         }
1139
1140         proc_fork_connector(p);
1141         if (!current->signal->tty && p->signal->tty)
1142                 p->signal->tty = NULL;
1143
1144         nr_threads++;
1145         total_forks++;
1146         write_unlock_irq(&tasklist_lock);
1147         retval = 0;
1148
1149 fork_out:
1150         if (retval)
1151                 return ERR_PTR(retval);
1152         return p;
1153
1154 bad_fork_cleanup_namespace:
1155         exit_namespace(p);
1156 bad_fork_cleanup_keys:
1157         exit_keys(p);
1158 bad_fork_cleanup_mm:
1159         if (p->mm)
1160                 mmput(p->mm);
1161 bad_fork_cleanup_signal:
1162         exit_signal(p);
1163 bad_fork_cleanup_sighand:
1164         exit_sighand(p);
1165 bad_fork_cleanup_fs:
1166         exit_fs(p); /* blocking */
1167 bad_fork_cleanup_files:
1168         exit_files(p); /* blocking */
1169 bad_fork_cleanup_semundo:
1170         exit_sem(p);
1171 bad_fork_cleanup_audit:
1172         audit_free(p);
1173 bad_fork_cleanup_security:
1174         security_task_free(p);
1175 bad_fork_cleanup_policy:
1176 #ifdef CONFIG_NUMA
1177         mpol_free(p->mempolicy);
1178 #endif
1179 bad_fork_cleanup:
1180         if (p->binfmt)
1181                 module_put(p->binfmt->module);
1182 bad_fork_cleanup_put_domain:
1183         module_put(p->thread_info->exec_domain->module);
1184 bad_fork_cleanup_count:
1185         put_group_info(p->group_info);
1186         atomic_dec(&p->user->processes);
1187         free_uid(p->user);
1188 bad_fork_free:
1189         free_task(p);
1190         goto fork_out;
1191 }
1192
1193 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1194 {
1195         memset(regs, 0, sizeof(struct pt_regs));
1196         return regs;
1197 }
1198
1199 task_t * __devinit fork_idle(int cpu)
1200 {
1201         task_t *task;
1202         struct pt_regs regs;
1203
1204         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1205         if (!task)
1206                 return ERR_PTR(-ENOMEM);
1207         init_idle(task, cpu);
1208         unhash_process(task);
1209         return task;
1210 }
1211
1212 static inline int fork_traceflag (unsigned clone_flags)
1213 {
1214         if (clone_flags & CLONE_UNTRACED)
1215                 return 0;
1216         else if (clone_flags & CLONE_VFORK) {
1217                 if (current->ptrace & PT_TRACE_VFORK)
1218                         return PTRACE_EVENT_VFORK;
1219         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1220                 if (current->ptrace & PT_TRACE_CLONE)
1221                         return PTRACE_EVENT_CLONE;
1222         } else if (current->ptrace & PT_TRACE_FORK)
1223                 return PTRACE_EVENT_FORK;
1224
1225         return 0;
1226 }
1227
1228 /*
1229  *  Ok, this is the main fork-routine.
1230  *
1231  * It copies the process, and if successful kick-starts
1232  * it and waits for it to finish using the VM if required.
1233  */
1234 long do_fork(unsigned long clone_flags,
1235               unsigned long stack_start,
1236               struct pt_regs *regs,
1237               unsigned long stack_size,
1238               int __user *parent_tidptr,
1239               int __user *child_tidptr)
1240 {
1241         struct task_struct *p;
1242         int trace = 0;
1243         long pid = alloc_pidmap();
1244
1245         if (pid < 0)
1246                 return -EAGAIN;
1247         if (unlikely(current->ptrace)) {
1248                 trace = fork_traceflag (clone_flags);
1249                 if (trace)
1250                         clone_flags |= CLONE_PTRACE;
1251         }
1252
1253         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1254         /*
1255          * Do this prior waking up the new thread - the thread pointer
1256          * might get invalid after that point, if the thread exits quickly.
1257          */
1258         if (!IS_ERR(p)) {
1259                 struct completion vfork;
1260
1261                 if (clone_flags & CLONE_VFORK) {
1262                         p->vfork_done = &vfork;
1263                         init_completion(&vfork);
1264                 }
1265
1266                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1267                         /*
1268                          * We'll start up with an immediate SIGSTOP.
1269                          */
1270                         sigaddset(&p->pending.signal, SIGSTOP);
1271                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1272                 }
1273
1274                 if (!(clone_flags & CLONE_STOPPED))
1275                         wake_up_new_task(p, clone_flags);
1276                 else
1277                         p->state = TASK_STOPPED;
1278
1279                 if (unlikely (trace)) {
1280                         current->ptrace_message = pid;
1281                         ptrace_notify ((trace << 8) | SIGTRAP);
1282                 }
1283
1284                 if (clone_flags & CLONE_VFORK) {
1285                         wait_for_completion(&vfork);
1286                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1287                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1288                 }
1289         } else {
1290                 free_pidmap(pid);
1291                 pid = PTR_ERR(p);
1292         }
1293         return pid;
1294 }
1295
1296 void __init proc_caches_init(void)
1297 {
1298         sighand_cachep = kmem_cache_create("sighand_cache",
1299                         sizeof(struct sighand_struct), 0,
1300                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1301         signal_cachep = kmem_cache_create("signal_cache",
1302                         sizeof(struct signal_struct), 0,
1303                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1304         files_cachep = kmem_cache_create("files_cache", 
1305                         sizeof(struct files_struct), 0,
1306                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1307         fs_cachep = kmem_cache_create("fs_cache", 
1308                         sizeof(struct fs_struct), 0,
1309                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1310         vm_area_cachep = kmem_cache_create("vm_area_struct",
1311                         sizeof(struct vm_area_struct), 0,
1312                         SLAB_PANIC, NULL, NULL);
1313         mm_cachep = kmem_cache_create("mm_struct",
1314                         sizeof(struct mm_struct), 0,
1315                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1316 }