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