[hrtimer] Enforce resolution as lower limit of intervals
[sfrench/cifs-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/capability.h>
32 #include <linux/cpu.h>
33 #include <linux/cpuset.h>
34 #include <linux/security.h>
35 #include <linux/swap.h>
36 #include <linux/syscalls.h>
37 #include <linux/jiffies.h>
38 #include <linux/futex.h>
39 #include <linux/rcupdate.h>
40 #include <linux/ptrace.h>
41 #include <linux/mount.h>
42 #include <linux/audit.h>
43 #include <linux/profile.h>
44 #include <linux/rmap.h>
45 #include <linux/acct.h>
46 #include <linux/cn_proc.h>
47
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/uaccess.h>
51 #include <asm/mmu_context.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
54
55 /*
56  * Protected counters by write_lock_irq(&tasklist_lock)
57  */
58 unsigned long total_forks;      /* Handle normal Linux uptimes. */
59 int nr_threads;                 /* The idle threads do not count.. */
60
61 int max_threads;                /* tunable limit on nr_threads */
62
63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
64
65  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
66
67 EXPORT_SYMBOL(tasklist_lock);
68
69 int nr_processes(void)
70 {
71         int cpu;
72         int total = 0;
73
74         for_each_online_cpu(cpu)
75                 total += per_cpu(process_counts, cpu);
76
77         return total;
78 }
79
80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
81 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
82 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
83 static kmem_cache_t *task_struct_cachep;
84 #endif
85
86 /* SLAB cache for signal_struct structures (tsk->signal) */
87 kmem_cache_t *signal_cachep;
88
89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
90 kmem_cache_t *sighand_cachep;
91
92 /* SLAB cache for files_struct structures (tsk->files) */
93 kmem_cache_t *files_cachep;
94
95 /* SLAB cache for fs_struct structures (tsk->fs) */
96 kmem_cache_t *fs_cachep;
97
98 /* SLAB cache for vm_area_struct structures */
99 kmem_cache_t *vm_area_cachep;
100
101 /* SLAB cache for mm_struct structures (tsk->mm) */
102 static kmem_cache_t *mm_cachep;
103
104 void free_task(struct task_struct *tsk)
105 {
106         free_thread_info(tsk->thread_info);
107         free_task_struct(tsk);
108 }
109 EXPORT_SYMBOL(free_task);
110
111 void __put_task_struct(struct task_struct *tsk)
112 {
113         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
114         WARN_ON(atomic_read(&tsk->usage));
115         WARN_ON(tsk == current);
116
117         if (unlikely(tsk->audit_context))
118                 audit_free(tsk);
119         security_task_free(tsk);
120         free_uid(tsk->user);
121         put_group_info(tsk->group_info);
122
123         if (!profile_handoff_task(tsk))
124                 free_task(tsk);
125 }
126
127 void __init fork_init(unsigned long mempages)
128 {
129 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
130 #ifndef ARCH_MIN_TASKALIGN
131 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
132 #endif
133         /* create a slab on which task_structs can be allocated */
134         task_struct_cachep =
135                 kmem_cache_create("task_struct", sizeof(struct task_struct),
136                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
137 #endif
138
139         /*
140          * The default maximum number of threads is set to a safe
141          * value: the thread structures can take up at most half
142          * of memory.
143          */
144         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
145
146         /*
147          * we need to allow at least 20 threads to boot a system
148          */
149         if(max_threads < 20)
150                 max_threads = 20;
151
152         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
153         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
154         init_task.signal->rlim[RLIMIT_SIGPENDING] =
155                 init_task.signal->rlim[RLIMIT_NPROC];
156 }
157
158 static struct task_struct *dup_task_struct(struct task_struct *orig)
159 {
160         struct task_struct *tsk;
161         struct thread_info *ti;
162
163         prepare_to_copy(orig);
164
165         tsk = alloc_task_struct();
166         if (!tsk)
167                 return NULL;
168
169         ti = alloc_thread_info(tsk);
170         if (!ti) {
171                 free_task_struct(tsk);
172                 return NULL;
173         }
174
175         *tsk = *orig;
176         tsk->thread_info = ti;
177         setup_thread_stack(tsk, orig);
178
179         /* One for us, one for whoever does the "release_task()" (usually parent) */
180         atomic_set(&tsk->usage,2);
181         atomic_set(&tsk->fs_excl, 0);
182         return tsk;
183 }
184
185 #ifdef CONFIG_MMU
186 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
187 {
188         struct vm_area_struct *mpnt, *tmp, **pprev;
189         struct rb_node **rb_link, *rb_parent;
190         int retval;
191         unsigned long charge;
192         struct mempolicy *pol;
193
194         down_write(&oldmm->mmap_sem);
195         flush_cache_mm(oldmm);
196         down_write(&mm->mmap_sem);
197
198         mm->locked_vm = 0;
199         mm->mmap = NULL;
200         mm->mmap_cache = NULL;
201         mm->free_area_cache = oldmm->mmap_base;
202         mm->cached_hole_size = ~0UL;
203         mm->map_count = 0;
204         cpus_clear(mm->cpu_vm_mask);
205         mm->mm_rb = RB_ROOT;
206         rb_link = &mm->mm_rb.rb_node;
207         rb_parent = NULL;
208         pprev = &mm->mmap;
209
210         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
211                 struct file *file;
212
213                 if (mpnt->vm_flags & VM_DONTCOPY) {
214                         long pages = vma_pages(mpnt);
215                         mm->total_vm -= pages;
216                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
217                                                                 -pages);
218                         continue;
219                 }
220                 charge = 0;
221                 if (mpnt->vm_flags & VM_ACCOUNT) {
222                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
223                         if (security_vm_enough_memory(len))
224                                 goto fail_nomem;
225                         charge = len;
226                 }
227                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
228                 if (!tmp)
229                         goto fail_nomem;
230                 *tmp = *mpnt;
231                 pol = mpol_copy(vma_policy(mpnt));
232                 retval = PTR_ERR(pol);
233                 if (IS_ERR(pol))
234                         goto fail_nomem_policy;
235                 vma_set_policy(tmp, pol);
236                 tmp->vm_flags &= ~VM_LOCKED;
237                 tmp->vm_mm = mm;
238                 tmp->vm_next = NULL;
239                 anon_vma_link(tmp);
240                 file = tmp->vm_file;
241                 if (file) {
242                         struct inode *inode = file->f_dentry->d_inode;
243                         get_file(file);
244                         if (tmp->vm_flags & VM_DENYWRITE)
245                                 atomic_dec(&inode->i_writecount);
246       
247                         /* insert tmp into the share list, just after mpnt */
248                         spin_lock(&file->f_mapping->i_mmap_lock);
249                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
250                         flush_dcache_mmap_lock(file->f_mapping);
251                         vma_prio_tree_add(tmp, mpnt);
252                         flush_dcache_mmap_unlock(file->f_mapping);
253                         spin_unlock(&file->f_mapping->i_mmap_lock);
254                 }
255
256                 /*
257                  * Link in the new vma and copy the page table entries.
258                  */
259                 *pprev = tmp;
260                 pprev = &tmp->vm_next;
261
262                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
263                 rb_link = &tmp->vm_rb.rb_right;
264                 rb_parent = &tmp->vm_rb;
265
266                 mm->map_count++;
267                 retval = copy_page_range(mm, oldmm, mpnt);
268
269                 if (tmp->vm_ops && tmp->vm_ops->open)
270                         tmp->vm_ops->open(tmp);
271
272                 if (retval)
273                         goto out;
274         }
275         retval = 0;
276 out:
277         up_write(&mm->mmap_sem);
278         flush_tlb_mm(oldmm);
279         up_write(&oldmm->mmap_sem);
280         return retval;
281 fail_nomem_policy:
282         kmem_cache_free(vm_area_cachep, tmp);
283 fail_nomem:
284         retval = -ENOMEM;
285         vm_unacct_memory(charge);
286         goto out;
287 }
288
289 static inline int mm_alloc_pgd(struct mm_struct * mm)
290 {
291         mm->pgd = pgd_alloc(mm);
292         if (unlikely(!mm->pgd))
293                 return -ENOMEM;
294         return 0;
295 }
296
297 static inline void mm_free_pgd(struct mm_struct * mm)
298 {
299         pgd_free(mm->pgd);
300 }
301 #else
302 #define dup_mmap(mm, oldmm)     (0)
303 #define mm_alloc_pgd(mm)        (0)
304 #define mm_free_pgd(mm)
305 #endif /* CONFIG_MMU */
306
307  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
308
309 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
310 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
311
312 #include <linux/init_task.h>
313
314 static struct mm_struct * mm_init(struct mm_struct * mm)
315 {
316         atomic_set(&mm->mm_users, 1);
317         atomic_set(&mm->mm_count, 1);
318         init_rwsem(&mm->mmap_sem);
319         INIT_LIST_HEAD(&mm->mmlist);
320         mm->core_waiters = 0;
321         mm->nr_ptes = 0;
322         set_mm_counter(mm, file_rss, 0);
323         set_mm_counter(mm, anon_rss, 0);
324         spin_lock_init(&mm->page_table_lock);
325         rwlock_init(&mm->ioctx_list_lock);
326         mm->ioctx_list = NULL;
327         mm->free_area_cache = TASK_UNMAPPED_BASE;
328         mm->cached_hole_size = ~0UL;
329
330         if (likely(!mm_alloc_pgd(mm))) {
331                 mm->def_flags = 0;
332                 return mm;
333         }
334         free_mm(mm);
335         return NULL;
336 }
337
338 /*
339  * Allocate and initialize an mm_struct.
340  */
341 struct mm_struct * mm_alloc(void)
342 {
343         struct mm_struct * mm;
344
345         mm = allocate_mm();
346         if (mm) {
347                 memset(mm, 0, sizeof(*mm));
348                 mm = mm_init(mm);
349         }
350         return mm;
351 }
352
353 /*
354  * Called when the last reference to the mm
355  * is dropped: either by a lazy thread or by
356  * mmput. Free the page directory and the mm.
357  */
358 void fastcall __mmdrop(struct mm_struct *mm)
359 {
360         BUG_ON(mm == &init_mm);
361         mm_free_pgd(mm);
362         destroy_context(mm);
363         free_mm(mm);
364 }
365
366 /*
367  * Decrement the use count and release all resources for an mm.
368  */
369 void mmput(struct mm_struct *mm)
370 {
371         if (atomic_dec_and_test(&mm->mm_users)) {
372                 exit_aio(mm);
373                 exit_mmap(mm);
374                 if (!list_empty(&mm->mmlist)) {
375                         spin_lock(&mmlist_lock);
376                         list_del(&mm->mmlist);
377                         spin_unlock(&mmlist_lock);
378                 }
379                 put_swap_token(mm);
380                 mmdrop(mm);
381         }
382 }
383 EXPORT_SYMBOL_GPL(mmput);
384
385 /**
386  * get_task_mm - acquire a reference to the task's mm
387  *
388  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
389  * this kernel workthread has transiently adopted a user mm with use_mm,
390  * to do its AIO) is not set and if so returns a reference to it, after
391  * bumping up the use count.  User must release the mm via mmput()
392  * after use.  Typically used by /proc and ptrace.
393  */
394 struct mm_struct *get_task_mm(struct task_struct *task)
395 {
396         struct mm_struct *mm;
397
398         task_lock(task);
399         mm = task->mm;
400         if (mm) {
401                 if (task->flags & PF_BORROWED_MM)
402                         mm = NULL;
403                 else
404                         atomic_inc(&mm->mm_users);
405         }
406         task_unlock(task);
407         return mm;
408 }
409 EXPORT_SYMBOL_GPL(get_task_mm);
410
411 /* Please note the differences between mmput and mm_release.
412  * mmput is called whenever we stop holding onto a mm_struct,
413  * error success whatever.
414  *
415  * mm_release is called after a mm_struct has been removed
416  * from the current process.
417  *
418  * This difference is important for error handling, when we
419  * only half set up a mm_struct for a new process and need to restore
420  * the old one.  Because we mmput the new mm_struct before
421  * restoring the old one. . .
422  * Eric Biederman 10 January 1998
423  */
424 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
425 {
426         struct completion *vfork_done = tsk->vfork_done;
427
428         /* Get rid of any cached register state */
429         deactivate_mm(tsk, mm);
430
431         /* notify parent sleeping on vfork() */
432         if (vfork_done) {
433                 tsk->vfork_done = NULL;
434                 complete(vfork_done);
435         }
436         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
437                 u32 __user * tidptr = tsk->clear_child_tid;
438                 tsk->clear_child_tid = NULL;
439
440                 /*
441                  * We don't check the error code - if userspace has
442                  * not set up a proper pointer then tough luck.
443                  */
444                 put_user(0, tidptr);
445                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
446         }
447 }
448
449 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
450 {
451         struct mm_struct * mm, *oldmm;
452         int retval;
453
454         tsk->min_flt = tsk->maj_flt = 0;
455         tsk->nvcsw = tsk->nivcsw = 0;
456
457         tsk->mm = NULL;
458         tsk->active_mm = NULL;
459
460         /*
461          * Are we cloning a kernel thread?
462          *
463          * We need to steal a active VM for that..
464          */
465         oldmm = current->mm;
466         if (!oldmm)
467                 return 0;
468
469         if (clone_flags & CLONE_VM) {
470                 atomic_inc(&oldmm->mm_users);
471                 mm = oldmm;
472                 goto good_mm;
473         }
474
475         retval = -ENOMEM;
476         mm = allocate_mm();
477         if (!mm)
478                 goto fail_nomem;
479
480         /* Copy the current MM stuff.. */
481         memcpy(mm, oldmm, sizeof(*mm));
482         if (!mm_init(mm))
483                 goto fail_nomem;
484
485         if (init_new_context(tsk,mm))
486                 goto fail_nocontext;
487
488         retval = dup_mmap(mm, oldmm);
489         if (retval)
490                 goto free_pt;
491
492         mm->hiwater_rss = get_mm_rss(mm);
493         mm->hiwater_vm = mm->total_vm;
494
495 good_mm:
496         tsk->mm = mm;
497         tsk->active_mm = mm;
498         return 0;
499
500 free_pt:
501         mmput(mm);
502 fail_nomem:
503         return retval;
504
505 fail_nocontext:
506         /*
507          * If init_new_context() failed, we cannot use mmput() to free the mm
508          * because it calls destroy_context()
509          */
510         mm_free_pgd(mm);
511         free_mm(mm);
512         return retval;
513 }
514
515 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
516 {
517         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
518         /* We don't need to lock fs - think why ;-) */
519         if (fs) {
520                 atomic_set(&fs->count, 1);
521                 rwlock_init(&fs->lock);
522                 fs->umask = old->umask;
523                 read_lock(&old->lock);
524                 fs->rootmnt = mntget(old->rootmnt);
525                 fs->root = dget(old->root);
526                 fs->pwdmnt = mntget(old->pwdmnt);
527                 fs->pwd = dget(old->pwd);
528                 if (old->altroot) {
529                         fs->altrootmnt = mntget(old->altrootmnt);
530                         fs->altroot = dget(old->altroot);
531                 } else {
532                         fs->altrootmnt = NULL;
533                         fs->altroot = NULL;
534                 }
535                 read_unlock(&old->lock);
536         }
537         return fs;
538 }
539
540 struct fs_struct *copy_fs_struct(struct fs_struct *old)
541 {
542         return __copy_fs_struct(old);
543 }
544
545 EXPORT_SYMBOL_GPL(copy_fs_struct);
546
547 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
548 {
549         if (clone_flags & CLONE_FS) {
550                 atomic_inc(&current->fs->count);
551                 return 0;
552         }
553         tsk->fs = __copy_fs_struct(current->fs);
554         if (!tsk->fs)
555                 return -ENOMEM;
556         return 0;
557 }
558
559 static int count_open_files(struct fdtable *fdt)
560 {
561         int size = fdt->max_fdset;
562         int i;
563
564         /* Find the last open fd */
565         for (i = size/(8*sizeof(long)); i > 0; ) {
566                 if (fdt->open_fds->fds_bits[--i])
567                         break;
568         }
569         i = (i+1) * 8 * sizeof(long);
570         return i;
571 }
572
573 static struct files_struct *alloc_files(void)
574 {
575         struct files_struct *newf;
576         struct fdtable *fdt;
577
578         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
579         if (!newf)
580                 goto out;
581
582         atomic_set(&newf->count, 1);
583
584         spin_lock_init(&newf->file_lock);
585         fdt = &newf->fdtab;
586         fdt->next_fd = 0;
587         fdt->max_fds = NR_OPEN_DEFAULT;
588         fdt->max_fdset = __FD_SETSIZE;
589         fdt->close_on_exec = &newf->close_on_exec_init;
590         fdt->open_fds = &newf->open_fds_init;
591         fdt->fd = &newf->fd_array[0];
592         INIT_RCU_HEAD(&fdt->rcu);
593         fdt->free_files = NULL;
594         fdt->next = NULL;
595         rcu_assign_pointer(newf->fdt, fdt);
596 out:
597         return newf;
598 }
599
600 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
601 {
602         struct files_struct *oldf, *newf;
603         struct file **old_fds, **new_fds;
604         int open_files, size, i, error = 0, expand;
605         struct fdtable *old_fdt, *new_fdt;
606
607         /*
608          * A background process may not have any files ...
609          */
610         oldf = current->files;
611         if (!oldf)
612                 goto out;
613
614         if (clone_flags & CLONE_FILES) {
615                 atomic_inc(&oldf->count);
616                 goto out;
617         }
618
619         /*
620          * Note: we may be using current for both targets (See exec.c)
621          * This works because we cache current->files (old) as oldf. Don't
622          * break this.
623          */
624         tsk->files = NULL;
625         error = -ENOMEM;
626         newf = alloc_files();
627         if (!newf)
628                 goto out;
629
630         spin_lock(&oldf->file_lock);
631         old_fdt = files_fdtable(oldf);
632         new_fdt = files_fdtable(newf);
633         size = old_fdt->max_fdset;
634         open_files = count_open_files(old_fdt);
635         expand = 0;
636
637         /*
638          * Check whether we need to allocate a larger fd array or fd set.
639          * Note: we're not a clone task, so the open count won't  change.
640          */
641         if (open_files > new_fdt->max_fdset) {
642                 new_fdt->max_fdset = 0;
643                 expand = 1;
644         }
645         if (open_files > new_fdt->max_fds) {
646                 new_fdt->max_fds = 0;
647                 expand = 1;
648         }
649
650         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
651         if (expand) {
652                 spin_unlock(&oldf->file_lock);
653                 spin_lock(&newf->file_lock);
654                 error = expand_files(newf, open_files-1);
655                 spin_unlock(&newf->file_lock);
656                 if (error < 0)
657                         goto out_release;
658                 new_fdt = files_fdtable(newf);
659                 /*
660                  * Reacquire the oldf lock and a pointer to its fd table
661                  * who knows it may have a new bigger fd table. We need
662                  * the latest pointer.
663                  */
664                 spin_lock(&oldf->file_lock);
665                 old_fdt = files_fdtable(oldf);
666         }
667
668         old_fds = old_fdt->fd;
669         new_fds = new_fdt->fd;
670
671         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
672         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
673
674         for (i = open_files; i != 0; i--) {
675                 struct file *f = *old_fds++;
676                 if (f) {
677                         get_file(f);
678                 } else {
679                         /*
680                          * The fd may be claimed in the fd bitmap but not yet
681                          * instantiated in the files array if a sibling thread
682                          * is partway through open().  So make sure that this
683                          * fd is available to the new process.
684                          */
685                         FD_CLR(open_files - i, new_fdt->open_fds);
686                 }
687                 rcu_assign_pointer(*new_fds++, f);
688         }
689         spin_unlock(&oldf->file_lock);
690
691         /* compute the remainder to be cleared */
692         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
693
694         /* This is long word aligned thus could use a optimized version */ 
695         memset(new_fds, 0, size); 
696
697         if (new_fdt->max_fdset > open_files) {
698                 int left = (new_fdt->max_fdset-open_files)/8;
699                 int start = open_files / (8 * sizeof(unsigned long));
700
701                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
702                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
703         }
704
705         tsk->files = newf;
706         error = 0;
707 out:
708         return error;
709
710 out_release:
711         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
712         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
713         free_fd_array(new_fdt->fd, new_fdt->max_fds);
714         kmem_cache_free(files_cachep, newf);
715         goto out;
716 }
717
718 /*
719  *      Helper to unshare the files of the current task.
720  *      We don't want to expose copy_files internals to
721  *      the exec layer of the kernel.
722  */
723
724 int unshare_files(void)
725 {
726         struct files_struct *files  = current->files;
727         int rc;
728
729         if(!files)
730                 BUG();
731
732         /* This can race but the race causes us to copy when we don't
733            need to and drop the copy */
734         if(atomic_read(&files->count) == 1)
735         {
736                 atomic_inc(&files->count);
737                 return 0;
738         }
739         rc = copy_files(0, current);
740         if(rc)
741                 current->files = files;
742         return rc;
743 }
744
745 EXPORT_SYMBOL(unshare_files);
746
747 void sighand_free_cb(struct rcu_head *rhp)
748 {
749         struct sighand_struct *sp;
750
751         sp = container_of(rhp, struct sighand_struct, rcu);
752         kmem_cache_free(sighand_cachep, sp);
753 }
754
755 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
756 {
757         struct sighand_struct *sig;
758
759         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
760                 atomic_inc(&current->sighand->count);
761                 return 0;
762         }
763         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
764         rcu_assign_pointer(tsk->sighand, sig);
765         if (!sig)
766                 return -ENOMEM;
767         spin_lock_init(&sig->siglock);
768         atomic_set(&sig->count, 1);
769         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
770         return 0;
771 }
772
773 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
774 {
775         struct signal_struct *sig;
776         int ret;
777
778         if (clone_flags & CLONE_THREAD) {
779                 atomic_inc(&current->signal->count);
780                 atomic_inc(&current->signal->live);
781                 return 0;
782         }
783         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
784         tsk->signal = sig;
785         if (!sig)
786                 return -ENOMEM;
787
788         ret = copy_thread_group_keys(tsk);
789         if (ret < 0) {
790                 kmem_cache_free(signal_cachep, sig);
791                 return ret;
792         }
793
794         atomic_set(&sig->count, 1);
795         atomic_set(&sig->live, 1);
796         init_waitqueue_head(&sig->wait_chldexit);
797         sig->flags = 0;
798         sig->group_exit_code = 0;
799         sig->group_exit_task = NULL;
800         sig->group_stop_count = 0;
801         sig->curr_target = NULL;
802         init_sigpending(&sig->shared_pending);
803         INIT_LIST_HEAD(&sig->posix_timers);
804
805         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC);
806         sig->it_real_incr.tv64 = 0;
807         sig->real_timer.function = it_real_fn;
808         sig->real_timer.data = tsk;
809
810         sig->it_virt_expires = cputime_zero;
811         sig->it_virt_incr = cputime_zero;
812         sig->it_prof_expires = cputime_zero;
813         sig->it_prof_incr = cputime_zero;
814
815         sig->leader = 0;        /* session leadership doesn't inherit */
816         sig->tty_old_pgrp = 0;
817
818         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
819         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
820         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
821         sig->sched_time = 0;
822         INIT_LIST_HEAD(&sig->cpu_timers[0]);
823         INIT_LIST_HEAD(&sig->cpu_timers[1]);
824         INIT_LIST_HEAD(&sig->cpu_timers[2]);
825
826         task_lock(current->group_leader);
827         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
828         task_unlock(current->group_leader);
829
830         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
831                 /*
832                  * New sole thread in the process gets an expiry time
833                  * of the whole CPU time limit.
834                  */
835                 tsk->it_prof_expires =
836                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
837         }
838
839         return 0;
840 }
841
842 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
843 {
844         unsigned long new_flags = p->flags;
845
846         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
847         new_flags |= PF_FORKNOEXEC;
848         if (!(clone_flags & CLONE_PTRACE))
849                 p->ptrace = 0;
850         p->flags = new_flags;
851 }
852
853 asmlinkage long sys_set_tid_address(int __user *tidptr)
854 {
855         current->clear_child_tid = tidptr;
856
857         return current->pid;
858 }
859
860 /*
861  * This creates a new process as a copy of the old one,
862  * but does not actually start it yet.
863  *
864  * It copies the registers, and all the appropriate
865  * parts of the process environment (as per the clone
866  * flags). The actual kick-off is left to the caller.
867  */
868 static task_t *copy_process(unsigned long clone_flags,
869                                  unsigned long stack_start,
870                                  struct pt_regs *regs,
871                                  unsigned long stack_size,
872                                  int __user *parent_tidptr,
873                                  int __user *child_tidptr,
874                                  int pid)
875 {
876         int retval;
877         struct task_struct *p = NULL;
878
879         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
880                 return ERR_PTR(-EINVAL);
881
882         /*
883          * Thread groups must share signals as well, and detached threads
884          * can only be started up within the thread group.
885          */
886         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
887                 return ERR_PTR(-EINVAL);
888
889         /*
890          * Shared signal handlers imply shared VM. By way of the above,
891          * thread groups also imply shared VM. Blocking this case allows
892          * for various simplifications in other code.
893          */
894         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
895                 return ERR_PTR(-EINVAL);
896
897         retval = security_task_create(clone_flags);
898         if (retval)
899                 goto fork_out;
900
901         retval = -ENOMEM;
902         p = dup_task_struct(current);
903         if (!p)
904                 goto fork_out;
905
906         retval = -EAGAIN;
907         if (atomic_read(&p->user->processes) >=
908                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
909                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
910                                 p->user != &root_user)
911                         goto bad_fork_free;
912         }
913
914         atomic_inc(&p->user->__count);
915         atomic_inc(&p->user->processes);
916         get_group_info(p->group_info);
917
918         /*
919          * If multiple threads are within copy_process(), then this check
920          * triggers too late. This doesn't hurt, the check is only there
921          * to stop root fork bombs.
922          */
923         if (nr_threads >= max_threads)
924                 goto bad_fork_cleanup_count;
925
926         if (!try_module_get(task_thread_info(p)->exec_domain->module))
927                 goto bad_fork_cleanup_count;
928
929         if (p->binfmt && !try_module_get(p->binfmt->module))
930                 goto bad_fork_cleanup_put_domain;
931
932         p->did_exec = 0;
933         copy_flags(clone_flags, p);
934         p->pid = pid;
935         retval = -EFAULT;
936         if (clone_flags & CLONE_PARENT_SETTID)
937                 if (put_user(p->pid, parent_tidptr))
938                         goto bad_fork_cleanup;
939
940         p->proc_dentry = NULL;
941
942         INIT_LIST_HEAD(&p->children);
943         INIT_LIST_HEAD(&p->sibling);
944         p->vfork_done = NULL;
945         spin_lock_init(&p->alloc_lock);
946         spin_lock_init(&p->proc_lock);
947
948         clear_tsk_thread_flag(p, TIF_SIGPENDING);
949         init_sigpending(&p->pending);
950
951         p->utime = cputime_zero;
952         p->stime = cputime_zero;
953         p->sched_time = 0;
954         p->rchar = 0;           /* I/O counter: bytes read */
955         p->wchar = 0;           /* I/O counter: bytes written */
956         p->syscr = 0;           /* I/O counter: read syscalls */
957         p->syscw = 0;           /* I/O counter: write syscalls */
958         acct_clear_integrals(p);
959
960         p->it_virt_expires = cputime_zero;
961         p->it_prof_expires = cputime_zero;
962         p->it_sched_expires = 0;
963         INIT_LIST_HEAD(&p->cpu_timers[0]);
964         INIT_LIST_HEAD(&p->cpu_timers[1]);
965         INIT_LIST_HEAD(&p->cpu_timers[2]);
966
967         p->lock_depth = -1;             /* -1 = no lock */
968         do_posix_clock_monotonic_gettime(&p->start_time);
969         p->security = NULL;
970         p->io_context = NULL;
971         p->io_wait = NULL;
972         p->audit_context = NULL;
973         cpuset_fork(p);
974 #ifdef CONFIG_NUMA
975         p->mempolicy = mpol_copy(p->mempolicy);
976         if (IS_ERR(p->mempolicy)) {
977                 retval = PTR_ERR(p->mempolicy);
978                 p->mempolicy = NULL;
979                 goto bad_fork_cleanup_cpuset;
980         }
981 #endif
982
983 #ifdef CONFIG_DEBUG_MUTEXES
984         p->blocked_on = NULL; /* not blocked yet */
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                         !cpu_online(task_cpu(p))))
1067                 set_task_cpu(p, smp_processor_id());
1068
1069         /*
1070          * Check for pending SIGKILL! The new thread should not be allowed
1071          * to slip out of an OOM kill. (or normal SIGKILL.)
1072          */
1073         if (sigismember(&current->pending.signal, SIGKILL)) {
1074                 write_unlock_irq(&tasklist_lock);
1075                 retval = -EINTR;
1076                 goto bad_fork_cleanup_namespace;
1077         }
1078
1079         /* CLONE_PARENT re-uses the old parent */
1080         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1081                 p->real_parent = current->real_parent;
1082         else
1083                 p->real_parent = current;
1084         p->parent = p->real_parent;
1085
1086         if (clone_flags & CLONE_THREAD) {
1087                 spin_lock(&current->sighand->siglock);
1088                 /*
1089                  * Important: if an exit-all has been started then
1090                  * do not create this new thread - the whole thread
1091                  * group is supposed to exit anyway.
1092                  */
1093                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1094                         spin_unlock(&current->sighand->siglock);
1095                         write_unlock_irq(&tasklist_lock);
1096                         retval = -EAGAIN;
1097                         goto bad_fork_cleanup_namespace;
1098                 }
1099                 p->group_leader = current->group_leader;
1100
1101                 if (current->signal->group_stop_count > 0) {
1102                         /*
1103                          * There is an all-stop in progress for the group.
1104                          * We ourselves will stop as soon as we check signals.
1105                          * Make the new thread part of that group stop too.
1106                          */
1107                         current->signal->group_stop_count++;
1108                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1109                 }
1110
1111                 if (!cputime_eq(current->signal->it_virt_expires,
1112                                 cputime_zero) ||
1113                     !cputime_eq(current->signal->it_prof_expires,
1114                                 cputime_zero) ||
1115                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1116                     !list_empty(&current->signal->cpu_timers[0]) ||
1117                     !list_empty(&current->signal->cpu_timers[1]) ||
1118                     !list_empty(&current->signal->cpu_timers[2])) {
1119                         /*
1120                          * Have child wake up on its first tick to check
1121                          * for process CPU timers.
1122                          */
1123                         p->it_prof_expires = jiffies_to_cputime(1);
1124                 }
1125
1126                 spin_unlock(&current->sighand->siglock);
1127         }
1128
1129         /*
1130          * inherit ioprio
1131          */
1132         p->ioprio = current->ioprio;
1133
1134         SET_LINKS(p);
1135         if (unlikely(p->ptrace & PT_PTRACED))
1136                 __ptrace_link(p, current->parent);
1137
1138         attach_pid(p, PIDTYPE_PID, p->pid);
1139         attach_pid(p, PIDTYPE_TGID, p->tgid);
1140         if (thread_group_leader(p)) {
1141                 p->signal->tty = current->signal->tty;
1142                 p->signal->pgrp = process_group(current);
1143                 p->signal->session = current->signal->session;
1144                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1145                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1146                 if (p->pid)
1147                         __get_cpu_var(process_counts)++;
1148         }
1149
1150         nr_threads++;
1151         total_forks++;
1152         write_unlock_irq(&tasklist_lock);
1153         proc_fork_connector(p);
1154         return p;
1155
1156 bad_fork_cleanup_namespace:
1157         exit_namespace(p);
1158 bad_fork_cleanup_keys:
1159         exit_keys(p);
1160 bad_fork_cleanup_mm:
1161         if (p->mm)
1162                 mmput(p->mm);
1163 bad_fork_cleanup_signal:
1164         exit_signal(p);
1165 bad_fork_cleanup_sighand:
1166         exit_sighand(p);
1167 bad_fork_cleanup_fs:
1168         exit_fs(p); /* blocking */
1169 bad_fork_cleanup_files:
1170         exit_files(p); /* blocking */
1171 bad_fork_cleanup_semundo:
1172         exit_sem(p);
1173 bad_fork_cleanup_audit:
1174         audit_free(p);
1175 bad_fork_cleanup_security:
1176         security_task_free(p);
1177 bad_fork_cleanup_policy:
1178 #ifdef CONFIG_NUMA
1179         mpol_free(p->mempolicy);
1180 bad_fork_cleanup_cpuset:
1181 #endif
1182         cpuset_exit(p);
1183 bad_fork_cleanup:
1184         if (p->binfmt)
1185                 module_put(p->binfmt->module);
1186 bad_fork_cleanup_put_domain:
1187         module_put(task_thread_info(p)->exec_domain->module);
1188 bad_fork_cleanup_count:
1189         put_group_info(p->group_info);
1190         atomic_dec(&p->user->processes);
1191         free_uid(p->user);
1192 bad_fork_free:
1193         free_task(p);
1194 fork_out:
1195         return ERR_PTR(retval);
1196 }
1197
1198 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1199 {
1200         memset(regs, 0, sizeof(struct pt_regs));
1201         return regs;
1202 }
1203
1204 task_t * __devinit fork_idle(int cpu)
1205 {
1206         task_t *task;
1207         struct pt_regs regs;
1208
1209         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1210         if (!task)
1211                 return ERR_PTR(-ENOMEM);
1212         init_idle(task, cpu);
1213         unhash_process(task);
1214         return task;
1215 }
1216
1217 static inline int fork_traceflag (unsigned clone_flags)
1218 {
1219         if (clone_flags & CLONE_UNTRACED)
1220                 return 0;
1221         else if (clone_flags & CLONE_VFORK) {
1222                 if (current->ptrace & PT_TRACE_VFORK)
1223                         return PTRACE_EVENT_VFORK;
1224         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1225                 if (current->ptrace & PT_TRACE_CLONE)
1226                         return PTRACE_EVENT_CLONE;
1227         } else if (current->ptrace & PT_TRACE_FORK)
1228                 return PTRACE_EVENT_FORK;
1229
1230         return 0;
1231 }
1232
1233 /*
1234  *  Ok, this is the main fork-routine.
1235  *
1236  * It copies the process, and if successful kick-starts
1237  * it and waits for it to finish using the VM if required.
1238  */
1239 long do_fork(unsigned long clone_flags,
1240               unsigned long stack_start,
1241               struct pt_regs *regs,
1242               unsigned long stack_size,
1243               int __user *parent_tidptr,
1244               int __user *child_tidptr)
1245 {
1246         struct task_struct *p;
1247         int trace = 0;
1248         long pid = alloc_pidmap();
1249
1250         if (pid < 0)
1251                 return -EAGAIN;
1252         if (unlikely(current->ptrace)) {
1253                 trace = fork_traceflag (clone_flags);
1254                 if (trace)
1255                         clone_flags |= CLONE_PTRACE;
1256         }
1257
1258         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1259         /*
1260          * Do this prior waking up the new thread - the thread pointer
1261          * might get invalid after that point, if the thread exits quickly.
1262          */
1263         if (!IS_ERR(p)) {
1264                 struct completion vfork;
1265
1266                 if (clone_flags & CLONE_VFORK) {
1267                         p->vfork_done = &vfork;
1268                         init_completion(&vfork);
1269                 }
1270
1271                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1272                         /*
1273                          * We'll start up with an immediate SIGSTOP.
1274                          */
1275                         sigaddset(&p->pending.signal, SIGSTOP);
1276                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1277                 }
1278
1279                 if (!(clone_flags & CLONE_STOPPED))
1280                         wake_up_new_task(p, clone_flags);
1281                 else
1282                         p->state = TASK_STOPPED;
1283
1284                 if (unlikely (trace)) {
1285                         current->ptrace_message = pid;
1286                         ptrace_notify ((trace << 8) | SIGTRAP);
1287                 }
1288
1289                 if (clone_flags & CLONE_VFORK) {
1290                         wait_for_completion(&vfork);
1291                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1292                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1293                 }
1294         } else {
1295                 free_pidmap(pid);
1296                 pid = PTR_ERR(p);
1297         }
1298         return pid;
1299 }
1300
1301 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1302 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1303 #endif
1304
1305 void __init proc_caches_init(void)
1306 {
1307         sighand_cachep = kmem_cache_create("sighand_cache",
1308                         sizeof(struct sighand_struct), 0,
1309                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1310         signal_cachep = kmem_cache_create("signal_cache",
1311                         sizeof(struct signal_struct), 0,
1312                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1313         files_cachep = kmem_cache_create("files_cache", 
1314                         sizeof(struct files_struct), 0,
1315                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1316         fs_cachep = kmem_cache_create("fs_cache", 
1317                         sizeof(struct fs_struct), 0,
1318                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1319         vm_area_cachep = kmem_cache_create("vm_area_struct",
1320                         sizeof(struct vm_area_struct), 0,
1321                         SLAB_PANIC, NULL, NULL);
1322         mm_cachep = kmem_cache_create("mm_struct",
1323                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1324                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1325 }