[PATCH] files: files struct with RCU
[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/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
46 #include <asm/pgtable.h>
47 #include <asm/pgalloc.h>
48 #include <asm/uaccess.h>
49 #include <asm/mmu_context.h>
50 #include <asm/cacheflush.h>
51 #include <asm/tlbflush.h>
52
53 /*
54  * Protected counters by write_lock_irq(&tasklist_lock)
55  */
56 unsigned long total_forks;      /* Handle normal Linux uptimes. */
57 int nr_threads;                 /* The idle threads do not count.. */
58
59 int max_threads;                /* tunable limit on nr_threads */
60
61 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
62
63  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
64
65 EXPORT_SYMBOL(tasklist_lock);
66
67 int nr_processes(void)
68 {
69         int cpu;
70         int total = 0;
71
72         for_each_online_cpu(cpu)
73                 total += per_cpu(process_counts, cpu);
74
75         return total;
76 }
77
78 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
79 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
80 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
81 static kmem_cache_t *task_struct_cachep;
82 #endif
83
84 /* SLAB cache for signal_struct structures (tsk->signal) */
85 kmem_cache_t *signal_cachep;
86
87 /* SLAB cache for sighand_struct structures (tsk->sighand) */
88 kmem_cache_t *sighand_cachep;
89
90 /* SLAB cache for files_struct structures (tsk->files) */
91 kmem_cache_t *files_cachep;
92
93 /* SLAB cache for fs_struct structures (tsk->fs) */
94 kmem_cache_t *fs_cachep;
95
96 /* SLAB cache for vm_area_struct structures */
97 kmem_cache_t *vm_area_cachep;
98
99 /* SLAB cache for mm_struct structures (tsk->mm) */
100 static kmem_cache_t *mm_cachep;
101
102 void free_task(struct task_struct *tsk)
103 {
104         free_thread_info(tsk->thread_info);
105         free_task_struct(tsk);
106 }
107 EXPORT_SYMBOL(free_task);
108
109 void __put_task_struct(struct task_struct *tsk)
110 {
111         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
112         WARN_ON(atomic_read(&tsk->usage));
113         WARN_ON(tsk == current);
114
115         if (unlikely(tsk->audit_context))
116                 audit_free(tsk);
117         security_task_free(tsk);
118         free_uid(tsk->user);
119         put_group_info(tsk->group_info);
120
121         if (!profile_handoff_task(tsk))
122                 free_task(tsk);
123 }
124
125 void __init fork_init(unsigned long mempages)
126 {
127 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
128 #ifndef ARCH_MIN_TASKALIGN
129 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
130 #endif
131         /* create a slab on which task_structs can be allocated */
132         task_struct_cachep =
133                 kmem_cache_create("task_struct", sizeof(struct task_struct),
134                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
135 #endif
136
137         /*
138          * The default maximum number of threads is set to a safe
139          * value: the thread structures can take up at most half
140          * of memory.
141          */
142         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
143
144         /*
145          * we need to allow at least 20 threads to boot a system
146          */
147         if(max_threads < 20)
148                 max_threads = 20;
149
150         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
151         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
152         init_task.signal->rlim[RLIMIT_SIGPENDING] =
153                 init_task.signal->rlim[RLIMIT_NPROC];
154 }
155
156 static struct task_struct *dup_task_struct(struct task_struct *orig)
157 {
158         struct task_struct *tsk;
159         struct thread_info *ti;
160
161         prepare_to_copy(orig);
162
163         tsk = alloc_task_struct();
164         if (!tsk)
165                 return NULL;
166
167         ti = alloc_thread_info(tsk);
168         if (!ti) {
169                 free_task_struct(tsk);
170                 return NULL;
171         }
172
173         *ti = *orig->thread_info;
174         *tsk = *orig;
175         tsk->thread_info = ti;
176         ti->task = tsk;
177
178         /* One for us, one for whoever does the "release_task()" (usually parent) */
179         atomic_set(&tsk->usage,2);
180         atomic_set(&tsk->fs_excl, 0);
181         return tsk;
182 }
183
184 #ifdef CONFIG_MMU
185 static inline int dup_mmap(struct mm_struct * mm, struct mm_struct * oldmm)
186 {
187         struct vm_area_struct * mpnt, *tmp, **pprev;
188         struct rb_node **rb_link, *rb_parent;
189         int retval;
190         unsigned long charge;
191         struct mempolicy *pol;
192
193         down_write(&oldmm->mmap_sem);
194         flush_cache_mm(current->mm);
195         mm->locked_vm = 0;
196         mm->mmap = NULL;
197         mm->mmap_cache = NULL;
198         mm->free_area_cache = oldmm->mmap_base;
199         mm->cached_hole_size = ~0UL;
200         mm->map_count = 0;
201         set_mm_counter(mm, rss, 0);
202         set_mm_counter(mm, anon_rss, 0);
203         cpus_clear(mm->cpu_vm_mask);
204         mm->mm_rb = RB_ROOT;
205         rb_link = &mm->mm_rb.rb_node;
206         rb_parent = NULL;
207         pprev = &mm->mmap;
208
209         for (mpnt = current->mm->mmap ; mpnt ; mpnt = mpnt->vm_next) {
210                 struct file *file;
211
212                 if (mpnt->vm_flags & VM_DONTCOPY) {
213                         long pages = vma_pages(mpnt);
214                         mm->total_vm -= pages;
215                         __vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
216                                                                 -pages);
217                         continue;
218                 }
219                 charge = 0;
220                 if (mpnt->vm_flags & VM_ACCOUNT) {
221                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
222                         if (security_vm_enough_memory(len))
223                                 goto fail_nomem;
224                         charge = len;
225                 }
226                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
227                 if (!tmp)
228                         goto fail_nomem;
229                 *tmp = *mpnt;
230                 pol = mpol_copy(vma_policy(mpnt));
231                 retval = PTR_ERR(pol);
232                 if (IS_ERR(pol))
233                         goto fail_nomem_policy;
234                 vma_set_policy(tmp, pol);
235                 tmp->vm_flags &= ~VM_LOCKED;
236                 tmp->vm_mm = mm;
237                 tmp->vm_next = NULL;
238                 anon_vma_link(tmp);
239                 file = tmp->vm_file;
240                 if (file) {
241                         struct inode *inode = file->f_dentry->d_inode;
242                         get_file(file);
243                         if (tmp->vm_flags & VM_DENYWRITE)
244                                 atomic_dec(&inode->i_writecount);
245       
246                         /* insert tmp into the share list, just after mpnt */
247                         spin_lock(&file->f_mapping->i_mmap_lock);
248                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
249                         flush_dcache_mmap_lock(file->f_mapping);
250                         vma_prio_tree_add(tmp, mpnt);
251                         flush_dcache_mmap_unlock(file->f_mapping);
252                         spin_unlock(&file->f_mapping->i_mmap_lock);
253                 }
254
255                 /*
256                  * Link in the new vma and copy the page table entries:
257                  * link in first so that swapoff can see swap entries.
258                  * Note that, exceptionally, here the vma is inserted
259                  * without holding mm->mmap_sem.
260                  */
261                 spin_lock(&mm->page_table_lock);
262                 *pprev = tmp;
263                 pprev = &tmp->vm_next;
264
265                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
266                 rb_link = &tmp->vm_rb.rb_right;
267                 rb_parent = &tmp->vm_rb;
268
269                 mm->map_count++;
270                 retval = copy_page_range(mm, current->mm, tmp);
271                 spin_unlock(&mm->page_table_lock);
272
273                 if (tmp->vm_ops && tmp->vm_ops->open)
274                         tmp->vm_ops->open(tmp);
275
276                 if (retval)
277                         goto out;
278         }
279         retval = 0;
280
281 out:
282         flush_tlb_mm(current->mm);
283         up_write(&oldmm->mmap_sem);
284         return retval;
285 fail_nomem_policy:
286         kmem_cache_free(vm_area_cachep, tmp);
287 fail_nomem:
288         retval = -ENOMEM;
289         vm_unacct_memory(charge);
290         goto out;
291 }
292
293 static inline int mm_alloc_pgd(struct mm_struct * mm)
294 {
295         mm->pgd = pgd_alloc(mm);
296         if (unlikely(!mm->pgd))
297                 return -ENOMEM;
298         return 0;
299 }
300
301 static inline void mm_free_pgd(struct mm_struct * mm)
302 {
303         pgd_free(mm->pgd);
304 }
305 #else
306 #define dup_mmap(mm, oldmm)     (0)
307 #define mm_alloc_pgd(mm)        (0)
308 #define mm_free_pgd(mm)
309 #endif /* CONFIG_MMU */
310
311  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
312
313 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
314 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
315
316 #include <linux/init_task.h>
317
318 static struct mm_struct * mm_init(struct mm_struct * mm)
319 {
320         atomic_set(&mm->mm_users, 1);
321         atomic_set(&mm->mm_count, 1);
322         init_rwsem(&mm->mmap_sem);
323         INIT_LIST_HEAD(&mm->mmlist);
324         mm->core_waiters = 0;
325         mm->nr_ptes = 0;
326         spin_lock_init(&mm->page_table_lock);
327         rwlock_init(&mm->ioctx_list_lock);
328         mm->ioctx_list = NULL;
329         mm->default_kioctx = (struct kioctx)INIT_KIOCTX(mm->default_kioctx, *mm);
330         mm->free_area_cache = TASK_UNMAPPED_BASE;
331         mm->cached_hole_size = ~0UL;
332
333         if (likely(!mm_alloc_pgd(mm))) {
334                 mm->def_flags = 0;
335                 return mm;
336         }
337         free_mm(mm);
338         return NULL;
339 }
340
341 /*
342  * Allocate and initialize an mm_struct.
343  */
344 struct mm_struct * mm_alloc(void)
345 {
346         struct mm_struct * mm;
347
348         mm = allocate_mm();
349         if (mm) {
350                 memset(mm, 0, sizeof(*mm));
351                 mm = mm_init(mm);
352         }
353         return mm;
354 }
355
356 /*
357  * Called when the last reference to the mm
358  * is dropped: either by a lazy thread or by
359  * mmput. Free the page directory and the mm.
360  */
361 void fastcall __mmdrop(struct mm_struct *mm)
362 {
363         BUG_ON(mm == &init_mm);
364         mm_free_pgd(mm);
365         destroy_context(mm);
366         free_mm(mm);
367 }
368
369 /*
370  * Decrement the use count and release all resources for an mm.
371  */
372 void mmput(struct mm_struct *mm)
373 {
374         if (atomic_dec_and_test(&mm->mm_users)) {
375                 exit_aio(mm);
376                 exit_mmap(mm);
377                 if (!list_empty(&mm->mmlist)) {
378                         spin_lock(&mmlist_lock);
379                         list_del(&mm->mmlist);
380                         spin_unlock(&mmlist_lock);
381                 }
382                 put_swap_token(mm);
383                 mmdrop(mm);
384         }
385 }
386 EXPORT_SYMBOL_GPL(mmput);
387
388 /**
389  * get_task_mm - acquire a reference to the task's mm
390  *
391  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
392  * this kernel workthread has transiently adopted a user mm with use_mm,
393  * to do its AIO) is not set and if so returns a reference to it, after
394  * bumping up the use count.  User must release the mm via mmput()
395  * after use.  Typically used by /proc and ptrace.
396  */
397 struct mm_struct *get_task_mm(struct task_struct *task)
398 {
399         struct mm_struct *mm;
400
401         task_lock(task);
402         mm = task->mm;
403         if (mm) {
404                 if (task->flags & PF_BORROWED_MM)
405                         mm = NULL;
406                 else
407                         atomic_inc(&mm->mm_users);
408         }
409         task_unlock(task);
410         return mm;
411 }
412 EXPORT_SYMBOL_GPL(get_task_mm);
413
414 /* Please note the differences between mmput and mm_release.
415  * mmput is called whenever we stop holding onto a mm_struct,
416  * error success whatever.
417  *
418  * mm_release is called after a mm_struct has been removed
419  * from the current process.
420  *
421  * This difference is important for error handling, when we
422  * only half set up a mm_struct for a new process and need to restore
423  * the old one.  Because we mmput the new mm_struct before
424  * restoring the old one. . .
425  * Eric Biederman 10 January 1998
426  */
427 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
428 {
429         struct completion *vfork_done = tsk->vfork_done;
430
431         /* Get rid of any cached register state */
432         deactivate_mm(tsk, mm);
433
434         /* notify parent sleeping on vfork() */
435         if (vfork_done) {
436                 tsk->vfork_done = NULL;
437                 complete(vfork_done);
438         }
439         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
440                 u32 __user * tidptr = tsk->clear_child_tid;
441                 tsk->clear_child_tid = NULL;
442
443                 /*
444                  * We don't check the error code - if userspace has
445                  * not set up a proper pointer then tough luck.
446                  */
447                 put_user(0, tidptr);
448                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
449         }
450 }
451
452 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
453 {
454         struct mm_struct * mm, *oldmm;
455         int retval;
456
457         tsk->min_flt = tsk->maj_flt = 0;
458         tsk->nvcsw = tsk->nivcsw = 0;
459
460         tsk->mm = NULL;
461         tsk->active_mm = NULL;
462
463         /*
464          * Are we cloning a kernel thread?
465          *
466          * We need to steal a active VM for that..
467          */
468         oldmm = current->mm;
469         if (!oldmm)
470                 return 0;
471
472         if (clone_flags & CLONE_VM) {
473                 atomic_inc(&oldmm->mm_users);
474                 mm = oldmm;
475                 /*
476                  * There are cases where the PTL is held to ensure no
477                  * new threads start up in user mode using an mm, which
478                  * allows optimizing out ipis; the tlb_gather_mmu code
479                  * is an example.
480                  */
481                 spin_unlock_wait(&oldmm->page_table_lock);
482                 goto good_mm;
483         }
484
485         retval = -ENOMEM;
486         mm = allocate_mm();
487         if (!mm)
488                 goto fail_nomem;
489
490         /* Copy the current MM stuff.. */
491         memcpy(mm, oldmm, sizeof(*mm));
492         if (!mm_init(mm))
493                 goto fail_nomem;
494
495         if (init_new_context(tsk,mm))
496                 goto fail_nocontext;
497
498         retval = dup_mmap(mm, oldmm);
499         if (retval)
500                 goto free_pt;
501
502         mm->hiwater_rss = get_mm_counter(mm,rss);
503         mm->hiwater_vm = mm->total_vm;
504
505 good_mm:
506         tsk->mm = mm;
507         tsk->active_mm = mm;
508         return 0;
509
510 free_pt:
511         mmput(mm);
512 fail_nomem:
513         return retval;
514
515 fail_nocontext:
516         /*
517          * If init_new_context() failed, we cannot use mmput() to free the mm
518          * because it calls destroy_context()
519          */
520         mm_free_pgd(mm);
521         free_mm(mm);
522         return retval;
523 }
524
525 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
526 {
527         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
528         /* We don't need to lock fs - think why ;-) */
529         if (fs) {
530                 atomic_set(&fs->count, 1);
531                 rwlock_init(&fs->lock);
532                 fs->umask = old->umask;
533                 read_lock(&old->lock);
534                 fs->rootmnt = mntget(old->rootmnt);
535                 fs->root = dget(old->root);
536                 fs->pwdmnt = mntget(old->pwdmnt);
537                 fs->pwd = dget(old->pwd);
538                 if (old->altroot) {
539                         fs->altrootmnt = mntget(old->altrootmnt);
540                         fs->altroot = dget(old->altroot);
541                 } else {
542                         fs->altrootmnt = NULL;
543                         fs->altroot = NULL;
544                 }
545                 read_unlock(&old->lock);
546         }
547         return fs;
548 }
549
550 struct fs_struct *copy_fs_struct(struct fs_struct *old)
551 {
552         return __copy_fs_struct(old);
553 }
554
555 EXPORT_SYMBOL_GPL(copy_fs_struct);
556
557 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
558 {
559         if (clone_flags & CLONE_FS) {
560                 atomic_inc(&current->fs->count);
561                 return 0;
562         }
563         tsk->fs = __copy_fs_struct(current->fs);
564         if (!tsk->fs)
565                 return -ENOMEM;
566         return 0;
567 }
568
569 static int count_open_files(struct fdtable *fdt)
570 {
571         int size = fdt->max_fdset;
572         int i;
573
574         /* Find the last open fd */
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 = &newf->fdtab;
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         INIT_RCU_HEAD(&fdt->rcu);
603         fdt->free_files = NULL;
604         fdt->next = NULL;
605         rcu_assign_pointer(newf->fdt, fdt);
606 out:
607         return newf;
608 }
609
610 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
611 {
612         struct files_struct *oldf, *newf;
613         struct file **old_fds, **new_fds;
614         int open_files, size, i, error = 0, expand;
615         struct fdtable *old_fdt, *new_fdt;
616
617         /*
618          * A background process may not have any files ...
619          */
620         oldf = current->files;
621         if (!oldf)
622                 goto out;
623
624         if (clone_flags & CLONE_FILES) {
625                 atomic_inc(&oldf->count);
626                 goto out;
627         }
628
629         /*
630          * Note: we may be using current for both targets (See exec.c)
631          * This works because we cache current->files (old) as oldf. Don't
632          * break this.
633          */
634         tsk->files = NULL;
635         error = -ENOMEM;
636         newf = alloc_files();
637         if (!newf)
638                 goto out;
639
640         spin_lock(&oldf->file_lock);
641         old_fdt = files_fdtable(oldf);
642         new_fdt = files_fdtable(newf);
643         size = old_fdt->max_fdset;
644         open_files = count_open_files(old_fdt);
645         expand = 0;
646
647         /*
648          * Check whether we need to allocate a larger fd array or fd set.
649          * Note: we're not a clone task, so the open count won't  change.
650          */
651         if (open_files > new_fdt->max_fdset) {
652                 new_fdt->max_fdset = 0;
653                 expand = 1;
654         }
655         if (open_files > new_fdt->max_fds) {
656                 new_fdt->max_fds = 0;
657                 expand = 1;
658         }
659
660         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
661         if (expand) {
662                 spin_unlock(&oldf->file_lock);
663                 spin_lock(&newf->file_lock);
664                 error = expand_files(newf, open_files-1);
665                 spin_unlock(&newf->file_lock);
666                 if (error < 0)
667                         goto out_release;
668                 new_fdt = files_fdtable(newf);
669                 /*
670                  * Reacquire the oldf lock and a pointer to its fd table
671                  * who knows it may have a new bigger fd table. We need
672                  * the latest pointer.
673                  */
674                 spin_lock(&oldf->file_lock);
675                 old_fdt = files_fdtable(oldf);
676         }
677
678         old_fds = old_fdt->fd;
679         new_fds = new_fdt->fd;
680
681         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
682         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
683
684         for (i = open_files; i != 0; i--) {
685                 struct file *f = *old_fds++;
686                 if (f) {
687                         get_file(f);
688                 } else {
689                         /*
690                          * The fd may be claimed in the fd bitmap but not yet
691                          * instantiated in the files array if a sibling thread
692                          * is partway through open().  So make sure that this
693                          * fd is available to the new process.
694                          */
695                         FD_CLR(open_files - i, new_fdt->open_fds);
696                 }
697                 rcu_assign_pointer(*new_fds++, f);
698         }
699         spin_unlock(&oldf->file_lock);
700
701         /* compute the remainder to be cleared */
702         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
703
704         /* This is long word aligned thus could use a optimized version */ 
705         memset(new_fds, 0, size); 
706
707         if (new_fdt->max_fdset > open_files) {
708                 int left = (new_fdt->max_fdset-open_files)/8;
709                 int start = open_files / (8 * sizeof(unsigned long));
710
711                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
712                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
713         }
714
715         tsk->files = newf;
716         error = 0;
717 out:
718         return error;
719
720 out_release:
721         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
722         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
723         free_fd_array(new_fdt->fd, new_fdt->max_fds);
724         kmem_cache_free(files_cachep, newf);
725         goto out;
726 }
727
728 /*
729  *      Helper to unshare the files of the current task.
730  *      We don't want to expose copy_files internals to
731  *      the exec layer of the kernel.
732  */
733
734 int unshare_files(void)
735 {
736         struct files_struct *files  = current->files;
737         int rc;
738
739         if(!files)
740                 BUG();
741
742         /* This can race but the race causes us to copy when we don't
743            need to and drop the copy */
744         if(atomic_read(&files->count) == 1)
745         {
746                 atomic_inc(&files->count);
747                 return 0;
748         }
749         rc = copy_files(0, current);
750         if(rc)
751                 current->files = files;
752         return rc;
753 }
754
755 EXPORT_SYMBOL(unshare_files);
756
757 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
758 {
759         struct sighand_struct *sig;
760
761         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
762                 atomic_inc(&current->sighand->count);
763                 return 0;
764         }
765         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
766         tsk->sighand = sig;
767         if (!sig)
768                 return -ENOMEM;
769         spin_lock_init(&sig->siglock);
770         atomic_set(&sig->count, 1);
771         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
772         return 0;
773 }
774
775 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
776 {
777         struct signal_struct *sig;
778         int ret;
779
780         if (clone_flags & CLONE_THREAD) {
781                 atomic_inc(&current->signal->count);
782                 atomic_inc(&current->signal->live);
783                 return 0;
784         }
785         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
786         tsk->signal = sig;
787         if (!sig)
788                 return -ENOMEM;
789
790         ret = copy_thread_group_keys(tsk);
791         if (ret < 0) {
792                 kmem_cache_free(signal_cachep, sig);
793                 return ret;
794         }
795
796         atomic_set(&sig->count, 1);
797         atomic_set(&sig->live, 1);
798         init_waitqueue_head(&sig->wait_chldexit);
799         sig->flags = 0;
800         sig->group_exit_code = 0;
801         sig->group_exit_task = NULL;
802         sig->group_stop_count = 0;
803         sig->curr_target = NULL;
804         init_sigpending(&sig->shared_pending);
805         INIT_LIST_HEAD(&sig->posix_timers);
806
807         sig->it_real_value = sig->it_real_incr = 0;
808         sig->real_timer.function = it_real_fn;
809         sig->real_timer.data = (unsigned long) tsk;
810         init_timer(&sig->real_timer);
811
812         sig->it_virt_expires = cputime_zero;
813         sig->it_virt_incr = cputime_zero;
814         sig->it_prof_expires = cputime_zero;
815         sig->it_prof_incr = cputime_zero;
816
817         sig->tty = current->signal->tty;
818         sig->pgrp = process_group(current);
819         sig->session = current->signal->session;
820         sig->leader = 0;        /* session leadership doesn't inherit */
821         sig->tty_old_pgrp = 0;
822
823         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
824         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
825         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
826         sig->sched_time = 0;
827         INIT_LIST_HEAD(&sig->cpu_timers[0]);
828         INIT_LIST_HEAD(&sig->cpu_timers[1]);
829         INIT_LIST_HEAD(&sig->cpu_timers[2]);
830
831         task_lock(current->group_leader);
832         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
833         task_unlock(current->group_leader);
834
835         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
836                 /*
837                  * New sole thread in the process gets an expiry time
838                  * of the whole CPU time limit.
839                  */
840                 tsk->it_prof_expires =
841                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
842         }
843
844         return 0;
845 }
846
847 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
848 {
849         unsigned long new_flags = p->flags;
850
851         new_flags &= ~PF_SUPERPRIV;
852         new_flags |= PF_FORKNOEXEC;
853         if (!(clone_flags & CLONE_PTRACE))
854                 p->ptrace = 0;
855         p->flags = new_flags;
856 }
857
858 asmlinkage long sys_set_tid_address(int __user *tidptr)
859 {
860         current->clear_child_tid = tidptr;
861
862         return current->pid;
863 }
864
865 /*
866  * This creates a new process as a copy of the old one,
867  * but does not actually start it yet.
868  *
869  * It copies the registers, and all the appropriate
870  * parts of the process environment (as per the clone
871  * flags). The actual kick-off is left to the caller.
872  */
873 static task_t *copy_process(unsigned long clone_flags,
874                                  unsigned long stack_start,
875                                  struct pt_regs *regs,
876                                  unsigned long stack_size,
877                                  int __user *parent_tidptr,
878                                  int __user *child_tidptr,
879                                  int pid)
880 {
881         int retval;
882         struct task_struct *p = NULL;
883
884         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
885                 return ERR_PTR(-EINVAL);
886
887         /*
888          * Thread groups must share signals as well, and detached threads
889          * can only be started up within the thread group.
890          */
891         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
892                 return ERR_PTR(-EINVAL);
893
894         /*
895          * Shared signal handlers imply shared VM. By way of the above,
896          * thread groups also imply shared VM. Blocking this case allows
897          * for various simplifications in other code.
898          */
899         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
900                 return ERR_PTR(-EINVAL);
901
902         retval = security_task_create(clone_flags);
903         if (retval)
904                 goto fork_out;
905
906         retval = -ENOMEM;
907         p = dup_task_struct(current);
908         if (!p)
909                 goto fork_out;
910
911         retval = -EAGAIN;
912         if (atomic_read(&p->user->processes) >=
913                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
914                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
915                                 p->user != &root_user)
916                         goto bad_fork_free;
917         }
918
919         atomic_inc(&p->user->__count);
920         atomic_inc(&p->user->processes);
921         get_group_info(p->group_info);
922
923         /*
924          * If multiple threads are within copy_process(), then this check
925          * triggers too late. This doesn't hurt, the check is only there
926          * to stop root fork bombs.
927          */
928         if (nr_threads >= max_threads)
929                 goto bad_fork_cleanup_count;
930
931         if (!try_module_get(p->thread_info->exec_domain->module))
932                 goto bad_fork_cleanup_count;
933
934         if (p->binfmt && !try_module_get(p->binfmt->module))
935                 goto bad_fork_cleanup_put_domain;
936
937         p->did_exec = 0;
938         copy_flags(clone_flags, p);
939         p->pid = pid;
940         retval = -EFAULT;
941         if (clone_flags & CLONE_PARENT_SETTID)
942                 if (put_user(p->pid, parent_tidptr))
943                         goto bad_fork_cleanup;
944
945         p->proc_dentry = NULL;
946
947         INIT_LIST_HEAD(&p->children);
948         INIT_LIST_HEAD(&p->sibling);
949         p->vfork_done = NULL;
950         spin_lock_init(&p->alloc_lock);
951         spin_lock_init(&p->proc_lock);
952
953         clear_tsk_thread_flag(p, TIF_SIGPENDING);
954         init_sigpending(&p->pending);
955
956         p->utime = cputime_zero;
957         p->stime = cputime_zero;
958         p->sched_time = 0;
959         p->rchar = 0;           /* I/O counter: bytes read */
960         p->wchar = 0;           /* I/O counter: bytes written */
961         p->syscr = 0;           /* I/O counter: read syscalls */
962         p->syscw = 0;           /* I/O counter: write syscalls */
963         acct_clear_integrals(p);
964
965         p->it_virt_expires = cputime_zero;
966         p->it_prof_expires = cputime_zero;
967         p->it_sched_expires = 0;
968         INIT_LIST_HEAD(&p->cpu_timers[0]);
969         INIT_LIST_HEAD(&p->cpu_timers[1]);
970         INIT_LIST_HEAD(&p->cpu_timers[2]);
971
972         p->lock_depth = -1;             /* -1 = no lock */
973         do_posix_clock_monotonic_gettime(&p->start_time);
974         p->security = NULL;
975         p->io_context = NULL;
976         p->io_wait = NULL;
977         p->audit_context = NULL;
978 #ifdef CONFIG_NUMA
979         p->mempolicy = mpol_copy(p->mempolicy);
980         if (IS_ERR(p->mempolicy)) {
981                 retval = PTR_ERR(p->mempolicy);
982                 p->mempolicy = NULL;
983                 goto bad_fork_cleanup;
984         }
985 #endif
986
987         p->tgid = p->pid;
988         if (clone_flags & CLONE_THREAD)
989                 p->tgid = current->tgid;
990
991         if ((retval = security_task_alloc(p)))
992                 goto bad_fork_cleanup_policy;
993         if ((retval = audit_alloc(p)))
994                 goto bad_fork_cleanup_security;
995         /* copy all the process information */
996         if ((retval = copy_semundo(clone_flags, p)))
997                 goto bad_fork_cleanup_audit;
998         if ((retval = copy_files(clone_flags, p)))
999                 goto bad_fork_cleanup_semundo;
1000         if ((retval = copy_fs(clone_flags, p)))
1001                 goto bad_fork_cleanup_files;
1002         if ((retval = copy_sighand(clone_flags, p)))
1003                 goto bad_fork_cleanup_fs;
1004         if ((retval = copy_signal(clone_flags, p)))
1005                 goto bad_fork_cleanup_sighand;
1006         if ((retval = copy_mm(clone_flags, p)))
1007                 goto bad_fork_cleanup_signal;
1008         if ((retval = copy_keys(clone_flags, p)))
1009                 goto bad_fork_cleanup_mm;
1010         if ((retval = copy_namespace(clone_flags, p)))
1011                 goto bad_fork_cleanup_keys;
1012         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1013         if (retval)
1014                 goto bad_fork_cleanup_namespace;
1015
1016         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1017         /*
1018          * Clear TID on mm_release()?
1019          */
1020         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1021
1022         /*
1023          * Syscall tracing should be turned off in the child regardless
1024          * of CLONE_PTRACE.
1025          */
1026         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1027 #ifdef TIF_SYSCALL_EMU
1028         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1029 #endif
1030
1031         /* Our parent execution domain becomes current domain
1032            These must match for thread signalling to apply */
1033            
1034         p->parent_exec_id = p->self_exec_id;
1035
1036         /* ok, now we should be set up.. */
1037         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1038         p->pdeath_signal = 0;
1039         p->exit_state = 0;
1040
1041         /*
1042          * Ok, make it visible to the rest of the system.
1043          * We dont wake it up yet.
1044          */
1045         p->group_leader = p;
1046         INIT_LIST_HEAD(&p->ptrace_children);
1047         INIT_LIST_HEAD(&p->ptrace_list);
1048
1049         /* Perform scheduler related setup. Assign this task to a CPU. */
1050         sched_fork(p, clone_flags);
1051
1052         /* Need tasklist lock for parent etc handling! */
1053         write_lock_irq(&tasklist_lock);
1054
1055         /*
1056          * The task hasn't been attached yet, so its cpus_allowed mask will
1057          * not be changed, nor will its assigned CPU.
1058          *
1059          * The cpus_allowed mask of the parent may have changed after it was
1060          * copied first time - so re-copy it here, then check the child's CPU
1061          * to ensure it is on a valid CPU (and if not, just force it back to
1062          * parent's CPU). This avoids alot of nasty races.
1063          */
1064         p->cpus_allowed = current->cpus_allowed;
1065         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed)))
1066                 set_task_cpu(p, smp_processor_id());
1067
1068         /*
1069          * Check for pending SIGKILL! The new thread should not be allowed
1070          * to slip out of an OOM kill. (or normal SIGKILL.)
1071          */
1072         if (sigismember(&current->pending.signal, SIGKILL)) {
1073                 write_unlock_irq(&tasklist_lock);
1074                 retval = -EINTR;
1075                 goto bad_fork_cleanup_namespace;
1076         }
1077
1078         /* CLONE_PARENT re-uses the old parent */
1079         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1080                 p->real_parent = current->real_parent;
1081         else
1082                 p->real_parent = current;
1083         p->parent = p->real_parent;
1084
1085         if (clone_flags & CLONE_THREAD) {
1086                 spin_lock(&current->sighand->siglock);
1087                 /*
1088                  * Important: if an exit-all has been started then
1089                  * do not create this new thread - the whole thread
1090                  * group is supposed to exit anyway.
1091                  */
1092                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1093                         spin_unlock(&current->sighand->siglock);
1094                         write_unlock_irq(&tasklist_lock);
1095                         retval = -EAGAIN;
1096                         goto bad_fork_cleanup_namespace;
1097                 }
1098                 p->group_leader = current->group_leader;
1099
1100                 if (current->signal->group_stop_count > 0) {
1101                         /*
1102                          * There is an all-stop in progress for the group.
1103                          * We ourselves will stop as soon as we check signals.
1104                          * Make the new thread part of that group stop too.
1105                          */
1106                         current->signal->group_stop_count++;
1107                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1108                 }
1109
1110                 if (!cputime_eq(current->signal->it_virt_expires,
1111                                 cputime_zero) ||
1112                     !cputime_eq(current->signal->it_prof_expires,
1113                                 cputime_zero) ||
1114                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1115                     !list_empty(&current->signal->cpu_timers[0]) ||
1116                     !list_empty(&current->signal->cpu_timers[1]) ||
1117                     !list_empty(&current->signal->cpu_timers[2])) {
1118                         /*
1119                          * Have child wake up on its first tick to check
1120                          * for process CPU timers.
1121                          */
1122                         p->it_prof_expires = jiffies_to_cputime(1);
1123                 }
1124
1125                 spin_unlock(&current->sighand->siglock);
1126         }
1127
1128         /*
1129          * inherit ioprio
1130          */
1131         p->ioprio = current->ioprio;
1132
1133         SET_LINKS(p);
1134         if (unlikely(p->ptrace & PT_PTRACED))
1135                 __ptrace_link(p, current->parent);
1136
1137         cpuset_fork(p);
1138
1139         attach_pid(p, PIDTYPE_PID, p->pid);
1140         attach_pid(p, PIDTYPE_TGID, p->tgid);
1141         if (thread_group_leader(p)) {
1142                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1143                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1144                 if (p->pid)
1145                         __get_cpu_var(process_counts)++;
1146         }
1147
1148         if (!current->signal->tty && p->signal->tty)
1149                 p->signal->tty = NULL;
1150
1151         nr_threads++;
1152         total_forks++;
1153         write_unlock_irq(&tasklist_lock);
1154         retval = 0;
1155
1156 fork_out:
1157         if (retval)
1158                 return ERR_PTR(retval);
1159         return p;
1160
1161 bad_fork_cleanup_namespace:
1162         exit_namespace(p);
1163 bad_fork_cleanup_keys:
1164         exit_keys(p);
1165 bad_fork_cleanup_mm:
1166         if (p->mm)
1167                 mmput(p->mm);
1168 bad_fork_cleanup_signal:
1169         exit_signal(p);
1170 bad_fork_cleanup_sighand:
1171         exit_sighand(p);
1172 bad_fork_cleanup_fs:
1173         exit_fs(p); /* blocking */
1174 bad_fork_cleanup_files:
1175         exit_files(p); /* blocking */
1176 bad_fork_cleanup_semundo:
1177         exit_sem(p);
1178 bad_fork_cleanup_audit:
1179         audit_free(p);
1180 bad_fork_cleanup_security:
1181         security_task_free(p);
1182 bad_fork_cleanup_policy:
1183 #ifdef CONFIG_NUMA
1184         mpol_free(p->mempolicy);
1185 #endif
1186 bad_fork_cleanup:
1187         if (p->binfmt)
1188                 module_put(p->binfmt->module);
1189 bad_fork_cleanup_put_domain:
1190         module_put(p->thread_info->exec_domain->module);
1191 bad_fork_cleanup_count:
1192         put_group_info(p->group_info);
1193         atomic_dec(&p->user->processes);
1194         free_uid(p->user);
1195 bad_fork_free:
1196         free_task(p);
1197         goto fork_out;
1198 }
1199
1200 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1201 {
1202         memset(regs, 0, sizeof(struct pt_regs));
1203         return regs;
1204 }
1205
1206 task_t * __devinit fork_idle(int cpu)
1207 {
1208         task_t *task;
1209         struct pt_regs regs;
1210
1211         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1212         if (!task)
1213                 return ERR_PTR(-ENOMEM);
1214         init_idle(task, cpu);
1215         unhash_process(task);
1216         return task;
1217 }
1218
1219 static inline int fork_traceflag (unsigned clone_flags)
1220 {
1221         if (clone_flags & CLONE_UNTRACED)
1222                 return 0;
1223         else if (clone_flags & CLONE_VFORK) {
1224                 if (current->ptrace & PT_TRACE_VFORK)
1225                         return PTRACE_EVENT_VFORK;
1226         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1227                 if (current->ptrace & PT_TRACE_CLONE)
1228                         return PTRACE_EVENT_CLONE;
1229         } else if (current->ptrace & PT_TRACE_FORK)
1230                 return PTRACE_EVENT_FORK;
1231
1232         return 0;
1233 }
1234
1235 /*
1236  *  Ok, this is the main fork-routine.
1237  *
1238  * It copies the process, and if successful kick-starts
1239  * it and waits for it to finish using the VM if required.
1240  */
1241 long do_fork(unsigned long clone_flags,
1242               unsigned long stack_start,
1243               struct pt_regs *regs,
1244               unsigned long stack_size,
1245               int __user *parent_tidptr,
1246               int __user *child_tidptr)
1247 {
1248         struct task_struct *p;
1249         int trace = 0;
1250         long pid = alloc_pidmap();
1251
1252         if (pid < 0)
1253                 return -EAGAIN;
1254         if (unlikely(current->ptrace)) {
1255                 trace = fork_traceflag (clone_flags);
1256                 if (trace)
1257                         clone_flags |= CLONE_PTRACE;
1258         }
1259
1260         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1261         /*
1262          * Do this prior waking up the new thread - the thread pointer
1263          * might get invalid after that point, if the thread exits quickly.
1264          */
1265         if (!IS_ERR(p)) {
1266                 struct completion vfork;
1267
1268                 if (clone_flags & CLONE_VFORK) {
1269                         p->vfork_done = &vfork;
1270                         init_completion(&vfork);
1271                 }
1272
1273                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1274                         /*
1275                          * We'll start up with an immediate SIGSTOP.
1276                          */
1277                         sigaddset(&p->pending.signal, SIGSTOP);
1278                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1279                 }
1280
1281                 if (!(clone_flags & CLONE_STOPPED))
1282                         wake_up_new_task(p, clone_flags);
1283                 else
1284                         p->state = TASK_STOPPED;
1285
1286                 if (unlikely (trace)) {
1287                         current->ptrace_message = pid;
1288                         ptrace_notify ((trace << 8) | SIGTRAP);
1289                 }
1290
1291                 if (clone_flags & CLONE_VFORK) {
1292                         wait_for_completion(&vfork);
1293                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1294                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1295                 }
1296         } else {
1297                 free_pidmap(pid);
1298                 pid = PTR_ERR(p);
1299         }
1300         return pid;
1301 }
1302
1303 void __init proc_caches_init(void)
1304 {
1305         sighand_cachep = kmem_cache_create("sighand_cache",
1306                         sizeof(struct sighand_struct), 0,
1307                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1308         signal_cachep = kmem_cache_create("signal_cache",
1309                         sizeof(struct signal_struct), 0,
1310                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1311         files_cachep = kmem_cache_create("files_cache", 
1312                         sizeof(struct files_struct), 0,
1313                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1314         fs_cachep = kmem_cache_create("fs_cache", 
1315                         sizeof(struct fs_struct), 0,
1316                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1317         vm_area_cachep = kmem_cache_create("vm_area_struct",
1318                         sizeof(struct vm_area_struct), 0,
1319                         SLAB_PANIC, NULL, NULL);
1320         mm_cachep = kmem_cache_create("mm_struct",
1321                         sizeof(struct mm_struct), 0,
1322                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1323 }