Commit | Line | Data |
---|---|---|
1da177e4 LT |
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> | |
c59ede7b | 31 | #include <linux/capability.h> |
1da177e4 LT |
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> | |
ab2af1f5 | 39 | #include <linux/rcupdate.h> |
1da177e4 LT |
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> | |
9f46080c | 46 | #include <linux/cn_proc.h> |
1da177e4 LT |
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 | ||
1da177e4 LT |
175 | *tsk = *orig; |
176 | tsk->thread_info = ti; | |
10ebffde | 177 | setup_thread_stack(tsk, orig); |
1da177e4 LT |
178 | |
179 | /* One for us, one for whoever does the "release_task()" (usually parent) */ | |
180 | atomic_set(&tsk->usage,2); | |
4b5d37ac | 181 | atomic_set(&tsk->fs_excl, 0); |
1da177e4 LT |
182 | return tsk; |
183 | } | |
184 | ||
185 | #ifdef CONFIG_MMU | |
fd3e42fc | 186 | static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm) |
1da177e4 | 187 | { |
fd3e42fc | 188 | struct vm_area_struct *mpnt, *tmp, **pprev; |
1da177e4 LT |
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); | |
fd3e42fc | 195 | flush_cache_mm(oldmm); |
7ee78232 HD |
196 | down_write(&mm->mmap_sem); |
197 | ||
1da177e4 LT |
198 | mm->locked_vm = 0; |
199 | mm->mmap = NULL; | |
200 | mm->mmap_cache = NULL; | |
201 | mm->free_area_cache = oldmm->mmap_base; | |
1363c3cd | 202 | mm->cached_hole_size = ~0UL; |
1da177e4 | 203 | mm->map_count = 0; |
1da177e4 LT |
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 | ||
fd3e42fc | 210 | for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) { |
1da177e4 LT |
211 | struct file *file; |
212 | ||
213 | if (mpnt->vm_flags & VM_DONTCOPY) { | |
3b6bfcdb HD |
214 | long pages = vma_pages(mpnt); |
215 | mm->total_vm -= pages; | |
ab50b8ed | 216 | vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file, |
3b6bfcdb | 217 | -pages); |
1da177e4 LT |
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 | /* | |
7ee78232 | 257 | * Link in the new vma and copy the page table entries. |
1da177e4 | 258 | */ |
1da177e4 LT |
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++; | |
0b0db14c | 267 | retval = copy_page_range(mm, oldmm, mpnt); |
1da177e4 LT |
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; | |
1da177e4 | 276 | out: |
7ee78232 | 277 | up_write(&mm->mmap_sem); |
fd3e42fc | 278 | flush_tlb_mm(oldmm); |
1da177e4 LT |
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; | |
4294621f | 322 | set_mm_counter(mm, file_rss, 0); |
404351e6 | 323 | set_mm_counter(mm, anon_rss, 0); |
1da177e4 LT |
324 | spin_lock_init(&mm->page_table_lock); |
325 | rwlock_init(&mm->ioctx_list_lock); | |
326 | mm->ioctx_list = NULL; | |
1da177e4 | 327 | mm->free_area_cache = TASK_UNMAPPED_BASE; |
1363c3cd | 328 | mm->cached_hole_size = ~0UL; |
1da177e4 LT |
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 | ||
a0a7ec30 JD |
449 | /* |
450 | * Allocate a new mm structure and copy contents from the | |
451 | * mm structure of the passed in task structure. | |
452 | */ | |
453 | static struct mm_struct *dup_mm(struct task_struct *tsk) | |
454 | { | |
455 | struct mm_struct *mm, *oldmm = current->mm; | |
456 | int err; | |
457 | ||
458 | if (!oldmm) | |
459 | return NULL; | |
460 | ||
461 | mm = allocate_mm(); | |
462 | if (!mm) | |
463 | goto fail_nomem; | |
464 | ||
465 | memcpy(mm, oldmm, sizeof(*mm)); | |
466 | ||
467 | if (!mm_init(mm)) | |
468 | goto fail_nomem; | |
469 | ||
470 | if (init_new_context(tsk, mm)) | |
471 | goto fail_nocontext; | |
472 | ||
473 | err = dup_mmap(mm, oldmm); | |
474 | if (err) | |
475 | goto free_pt; | |
476 | ||
477 | mm->hiwater_rss = get_mm_rss(mm); | |
478 | mm->hiwater_vm = mm->total_vm; | |
479 | ||
480 | return mm; | |
481 | ||
482 | free_pt: | |
483 | mmput(mm); | |
484 | ||
485 | fail_nomem: | |
486 | return NULL; | |
487 | ||
488 | fail_nocontext: | |
489 | /* | |
490 | * If init_new_context() failed, we cannot use mmput() to free the mm | |
491 | * because it calls destroy_context() | |
492 | */ | |
493 | mm_free_pgd(mm); | |
494 | free_mm(mm); | |
495 | return NULL; | |
496 | } | |
497 | ||
1da177e4 LT |
498 | static int copy_mm(unsigned long clone_flags, struct task_struct * tsk) |
499 | { | |
500 | struct mm_struct * mm, *oldmm; | |
501 | int retval; | |
502 | ||
503 | tsk->min_flt = tsk->maj_flt = 0; | |
504 | tsk->nvcsw = tsk->nivcsw = 0; | |
505 | ||
506 | tsk->mm = NULL; | |
507 | tsk->active_mm = NULL; | |
508 | ||
509 | /* | |
510 | * Are we cloning a kernel thread? | |
511 | * | |
512 | * We need to steal a active VM for that.. | |
513 | */ | |
514 | oldmm = current->mm; | |
515 | if (!oldmm) | |
516 | return 0; | |
517 | ||
518 | if (clone_flags & CLONE_VM) { | |
519 | atomic_inc(&oldmm->mm_users); | |
520 | mm = oldmm; | |
1da177e4 LT |
521 | goto good_mm; |
522 | } | |
523 | ||
524 | retval = -ENOMEM; | |
a0a7ec30 | 525 | mm = dup_mm(tsk); |
1da177e4 LT |
526 | if (!mm) |
527 | goto fail_nomem; | |
528 | ||
1da177e4 LT |
529 | good_mm: |
530 | tsk->mm = mm; | |
531 | tsk->active_mm = mm; | |
532 | return 0; | |
533 | ||
1da177e4 LT |
534 | fail_nomem: |
535 | return retval; | |
1da177e4 LT |
536 | } |
537 | ||
538 | static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old) | |
539 | { | |
540 | struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL); | |
541 | /* We don't need to lock fs - think why ;-) */ | |
542 | if (fs) { | |
543 | atomic_set(&fs->count, 1); | |
544 | rwlock_init(&fs->lock); | |
545 | fs->umask = old->umask; | |
546 | read_lock(&old->lock); | |
547 | fs->rootmnt = mntget(old->rootmnt); | |
548 | fs->root = dget(old->root); | |
549 | fs->pwdmnt = mntget(old->pwdmnt); | |
550 | fs->pwd = dget(old->pwd); | |
551 | if (old->altroot) { | |
552 | fs->altrootmnt = mntget(old->altrootmnt); | |
553 | fs->altroot = dget(old->altroot); | |
554 | } else { | |
555 | fs->altrootmnt = NULL; | |
556 | fs->altroot = NULL; | |
557 | } | |
558 | read_unlock(&old->lock); | |
559 | } | |
560 | return fs; | |
561 | } | |
562 | ||
563 | struct fs_struct *copy_fs_struct(struct fs_struct *old) | |
564 | { | |
565 | return __copy_fs_struct(old); | |
566 | } | |
567 | ||
568 | EXPORT_SYMBOL_GPL(copy_fs_struct); | |
569 | ||
570 | static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk) | |
571 | { | |
572 | if (clone_flags & CLONE_FS) { | |
573 | atomic_inc(¤t->fs->count); | |
574 | return 0; | |
575 | } | |
576 | tsk->fs = __copy_fs_struct(current->fs); | |
577 | if (!tsk->fs) | |
578 | return -ENOMEM; | |
579 | return 0; | |
580 | } | |
581 | ||
ab2af1f5 | 582 | static int count_open_files(struct fdtable *fdt) |
1da177e4 | 583 | { |
ab2af1f5 | 584 | int size = fdt->max_fdset; |
1da177e4 LT |
585 | int i; |
586 | ||
587 | /* Find the last open fd */ | |
588 | for (i = size/(8*sizeof(long)); i > 0; ) { | |
badf1662 | 589 | if (fdt->open_fds->fds_bits[--i]) |
1da177e4 LT |
590 | break; |
591 | } | |
592 | i = (i+1) * 8 * sizeof(long); | |
593 | return i; | |
594 | } | |
595 | ||
badf1662 DS |
596 | static struct files_struct *alloc_files(void) |
597 | { | |
598 | struct files_struct *newf; | |
599 | struct fdtable *fdt; | |
600 | ||
601 | newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL); | |
602 | if (!newf) | |
603 | goto out; | |
604 | ||
605 | atomic_set(&newf->count, 1); | |
606 | ||
607 | spin_lock_init(&newf->file_lock); | |
ab2af1f5 | 608 | fdt = &newf->fdtab; |
badf1662 DS |
609 | fdt->next_fd = 0; |
610 | fdt->max_fds = NR_OPEN_DEFAULT; | |
611 | fdt->max_fdset = __FD_SETSIZE; | |
612 | fdt->close_on_exec = &newf->close_on_exec_init; | |
613 | fdt->open_fds = &newf->open_fds_init; | |
614 | fdt->fd = &newf->fd_array[0]; | |
ab2af1f5 DS |
615 | INIT_RCU_HEAD(&fdt->rcu); |
616 | fdt->free_files = NULL; | |
617 | fdt->next = NULL; | |
618 | rcu_assign_pointer(newf->fdt, fdt); | |
badf1662 DS |
619 | out: |
620 | return newf; | |
621 | } | |
622 | ||
1da177e4 LT |
623 | static int copy_files(unsigned long clone_flags, struct task_struct * tsk) |
624 | { | |
625 | struct files_struct *oldf, *newf; | |
626 | struct file **old_fds, **new_fds; | |
627 | int open_files, size, i, error = 0, expand; | |
badf1662 | 628 | struct fdtable *old_fdt, *new_fdt; |
1da177e4 LT |
629 | |
630 | /* | |
631 | * A background process may not have any files ... | |
632 | */ | |
633 | oldf = current->files; | |
634 | if (!oldf) | |
635 | goto out; | |
636 | ||
637 | if (clone_flags & CLONE_FILES) { | |
638 | atomic_inc(&oldf->count); | |
639 | goto out; | |
640 | } | |
641 | ||
642 | /* | |
643 | * Note: we may be using current for both targets (See exec.c) | |
644 | * This works because we cache current->files (old) as oldf. Don't | |
645 | * break this. | |
646 | */ | |
647 | tsk->files = NULL; | |
648 | error = -ENOMEM; | |
badf1662 DS |
649 | newf = alloc_files(); |
650 | if (!newf) | |
1da177e4 LT |
651 | goto out; |
652 | ||
1da177e4 | 653 | spin_lock(&oldf->file_lock); |
badf1662 DS |
654 | old_fdt = files_fdtable(oldf); |
655 | new_fdt = files_fdtable(newf); | |
656 | size = old_fdt->max_fdset; | |
ab2af1f5 | 657 | open_files = count_open_files(old_fdt); |
1da177e4 LT |
658 | expand = 0; |
659 | ||
660 | /* | |
661 | * Check whether we need to allocate a larger fd array or fd set. | |
662 | * Note: we're not a clone task, so the open count won't change. | |
663 | */ | |
badf1662 DS |
664 | if (open_files > new_fdt->max_fdset) { |
665 | new_fdt->max_fdset = 0; | |
1da177e4 LT |
666 | expand = 1; |
667 | } | |
badf1662 DS |
668 | if (open_files > new_fdt->max_fds) { |
669 | new_fdt->max_fds = 0; | |
1da177e4 LT |
670 | expand = 1; |
671 | } | |
672 | ||
673 | /* if the old fdset gets grown now, we'll only copy up to "size" fds */ | |
674 | if (expand) { | |
675 | spin_unlock(&oldf->file_lock); | |
676 | spin_lock(&newf->file_lock); | |
677 | error = expand_files(newf, open_files-1); | |
678 | spin_unlock(&newf->file_lock); | |
679 | if (error < 0) | |
680 | goto out_release; | |
ab2af1f5 DS |
681 | new_fdt = files_fdtable(newf); |
682 | /* | |
683 | * Reacquire the oldf lock and a pointer to its fd table | |
684 | * who knows it may have a new bigger fd table. We need | |
685 | * the latest pointer. | |
686 | */ | |
1da177e4 | 687 | spin_lock(&oldf->file_lock); |
ab2af1f5 | 688 | old_fdt = files_fdtable(oldf); |
1da177e4 LT |
689 | } |
690 | ||
badf1662 DS |
691 | old_fds = old_fdt->fd; |
692 | new_fds = new_fdt->fd; | |
1da177e4 | 693 | |
badf1662 DS |
694 | memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8); |
695 | memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8); | |
1da177e4 LT |
696 | |
697 | for (i = open_files; i != 0; i--) { | |
698 | struct file *f = *old_fds++; | |
699 | if (f) { | |
700 | get_file(f); | |
701 | } else { | |
702 | /* | |
703 | * The fd may be claimed in the fd bitmap but not yet | |
704 | * instantiated in the files array if a sibling thread | |
705 | * is partway through open(). So make sure that this | |
706 | * fd is available to the new process. | |
707 | */ | |
badf1662 | 708 | FD_CLR(open_files - i, new_fdt->open_fds); |
1da177e4 | 709 | } |
ab2af1f5 | 710 | rcu_assign_pointer(*new_fds++, f); |
1da177e4 LT |
711 | } |
712 | spin_unlock(&oldf->file_lock); | |
713 | ||
714 | /* compute the remainder to be cleared */ | |
badf1662 | 715 | size = (new_fdt->max_fds - open_files) * sizeof(struct file *); |
1da177e4 LT |
716 | |
717 | /* This is long word aligned thus could use a optimized version */ | |
718 | memset(new_fds, 0, size); | |
719 | ||
badf1662 DS |
720 | if (new_fdt->max_fdset > open_files) { |
721 | int left = (new_fdt->max_fdset-open_files)/8; | |
1da177e4 LT |
722 | int start = open_files / (8 * sizeof(unsigned long)); |
723 | ||
badf1662 DS |
724 | memset(&new_fdt->open_fds->fds_bits[start], 0, left); |
725 | memset(&new_fdt->close_on_exec->fds_bits[start], 0, left); | |
1da177e4 LT |
726 | } |
727 | ||
728 | tsk->files = newf; | |
729 | error = 0; | |
730 | out: | |
731 | return error; | |
732 | ||
733 | out_release: | |
badf1662 DS |
734 | free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset); |
735 | free_fdset (new_fdt->open_fds, new_fdt->max_fdset); | |
736 | free_fd_array(new_fdt->fd, new_fdt->max_fds); | |
1da177e4 LT |
737 | kmem_cache_free(files_cachep, newf); |
738 | goto out; | |
739 | } | |
740 | ||
741 | /* | |
742 | * Helper to unshare the files of the current task. | |
743 | * We don't want to expose copy_files internals to | |
744 | * the exec layer of the kernel. | |
745 | */ | |
746 | ||
747 | int unshare_files(void) | |
748 | { | |
749 | struct files_struct *files = current->files; | |
750 | int rc; | |
751 | ||
752 | if(!files) | |
753 | BUG(); | |
754 | ||
755 | /* This can race but the race causes us to copy when we don't | |
756 | need to and drop the copy */ | |
757 | if(atomic_read(&files->count) == 1) | |
758 | { | |
759 | atomic_inc(&files->count); | |
760 | return 0; | |
761 | } | |
762 | rc = copy_files(0, current); | |
763 | if(rc) | |
764 | current->files = files; | |
765 | return rc; | |
766 | } | |
767 | ||
768 | EXPORT_SYMBOL(unshare_files); | |
769 | ||
e56d0903 IM |
770 | void sighand_free_cb(struct rcu_head *rhp) |
771 | { | |
772 | struct sighand_struct *sp; | |
773 | ||
774 | sp = container_of(rhp, struct sighand_struct, rcu); | |
775 | kmem_cache_free(sighand_cachep, sp); | |
776 | } | |
777 | ||
1da177e4 LT |
778 | static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk) |
779 | { | |
780 | struct sighand_struct *sig; | |
781 | ||
782 | if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) { | |
783 | atomic_inc(¤t->sighand->count); | |
784 | return 0; | |
785 | } | |
786 | sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); | |
e56d0903 | 787 | rcu_assign_pointer(tsk->sighand, sig); |
1da177e4 LT |
788 | if (!sig) |
789 | return -ENOMEM; | |
790 | spin_lock_init(&sig->siglock); | |
791 | atomic_set(&sig->count, 1); | |
792 | memcpy(sig->action, current->sighand->action, sizeof(sig->action)); | |
793 | return 0; | |
794 | } | |
795 | ||
796 | static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk) | |
797 | { | |
798 | struct signal_struct *sig; | |
799 | int ret; | |
800 | ||
801 | if (clone_flags & CLONE_THREAD) { | |
802 | atomic_inc(¤t->signal->count); | |
803 | atomic_inc(¤t->signal->live); | |
804 | return 0; | |
805 | } | |
806 | sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL); | |
807 | tsk->signal = sig; | |
808 | if (!sig) | |
809 | return -ENOMEM; | |
810 | ||
811 | ret = copy_thread_group_keys(tsk); | |
812 | if (ret < 0) { | |
813 | kmem_cache_free(signal_cachep, sig); | |
814 | return ret; | |
815 | } | |
816 | ||
817 | atomic_set(&sig->count, 1); | |
818 | atomic_set(&sig->live, 1); | |
819 | init_waitqueue_head(&sig->wait_chldexit); | |
820 | sig->flags = 0; | |
821 | sig->group_exit_code = 0; | |
822 | sig->group_exit_task = NULL; | |
823 | sig->group_stop_count = 0; | |
824 | sig->curr_target = NULL; | |
825 | init_sigpending(&sig->shared_pending); | |
826 | INIT_LIST_HEAD(&sig->posix_timers); | |
827 | ||
7978672c | 828 | hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL); |
2ff678b8 | 829 | sig->it_real_incr.tv64 = 0; |
1da177e4 | 830 | sig->real_timer.function = it_real_fn; |
2ff678b8 | 831 | sig->real_timer.data = tsk; |
1da177e4 LT |
832 | |
833 | sig->it_virt_expires = cputime_zero; | |
834 | sig->it_virt_incr = cputime_zero; | |
835 | sig->it_prof_expires = cputime_zero; | |
836 | sig->it_prof_incr = cputime_zero; | |
837 | ||
1da177e4 LT |
838 | sig->leader = 0; /* session leadership doesn't inherit */ |
839 | sig->tty_old_pgrp = 0; | |
840 | ||
841 | sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero; | |
842 | sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0; | |
843 | sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0; | |
844 | sig->sched_time = 0; | |
845 | INIT_LIST_HEAD(&sig->cpu_timers[0]); | |
846 | INIT_LIST_HEAD(&sig->cpu_timers[1]); | |
847 | INIT_LIST_HEAD(&sig->cpu_timers[2]); | |
848 | ||
849 | task_lock(current->group_leader); | |
850 | memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim); | |
851 | task_unlock(current->group_leader); | |
852 | ||
853 | if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) { | |
854 | /* | |
855 | * New sole thread in the process gets an expiry time | |
856 | * of the whole CPU time limit. | |
857 | */ | |
858 | tsk->it_prof_expires = | |
859 | secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur); | |
860 | } | |
861 | ||
862 | return 0; | |
863 | } | |
864 | ||
865 | static inline void copy_flags(unsigned long clone_flags, struct task_struct *p) | |
866 | { | |
867 | unsigned long new_flags = p->flags; | |
868 | ||
d1209d04 | 869 | new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE); |
1da177e4 LT |
870 | new_flags |= PF_FORKNOEXEC; |
871 | if (!(clone_flags & CLONE_PTRACE)) | |
872 | p->ptrace = 0; | |
873 | p->flags = new_flags; | |
874 | } | |
875 | ||
876 | asmlinkage long sys_set_tid_address(int __user *tidptr) | |
877 | { | |
878 | current->clear_child_tid = tidptr; | |
879 | ||
880 | return current->pid; | |
881 | } | |
882 | ||
883 | /* | |
884 | * This creates a new process as a copy of the old one, | |
885 | * but does not actually start it yet. | |
886 | * | |
887 | * It copies the registers, and all the appropriate | |
888 | * parts of the process environment (as per the clone | |
889 | * flags). The actual kick-off is left to the caller. | |
890 | */ | |
891 | static task_t *copy_process(unsigned long clone_flags, | |
892 | unsigned long stack_start, | |
893 | struct pt_regs *regs, | |
894 | unsigned long stack_size, | |
895 | int __user *parent_tidptr, | |
896 | int __user *child_tidptr, | |
897 | int pid) | |
898 | { | |
899 | int retval; | |
900 | struct task_struct *p = NULL; | |
901 | ||
902 | if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS)) | |
903 | return ERR_PTR(-EINVAL); | |
904 | ||
905 | /* | |
906 | * Thread groups must share signals as well, and detached threads | |
907 | * can only be started up within the thread group. | |
908 | */ | |
909 | if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND)) | |
910 | return ERR_PTR(-EINVAL); | |
911 | ||
912 | /* | |
913 | * Shared signal handlers imply shared VM. By way of the above, | |
914 | * thread groups also imply shared VM. Blocking this case allows | |
915 | * for various simplifications in other code. | |
916 | */ | |
917 | if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM)) | |
918 | return ERR_PTR(-EINVAL); | |
919 | ||
920 | retval = security_task_create(clone_flags); | |
921 | if (retval) | |
922 | goto fork_out; | |
923 | ||
924 | retval = -ENOMEM; | |
925 | p = dup_task_struct(current); | |
926 | if (!p) | |
927 | goto fork_out; | |
928 | ||
929 | retval = -EAGAIN; | |
930 | if (atomic_read(&p->user->processes) >= | |
931 | p->signal->rlim[RLIMIT_NPROC].rlim_cur) { | |
932 | if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) && | |
933 | p->user != &root_user) | |
934 | goto bad_fork_free; | |
935 | } | |
936 | ||
937 | atomic_inc(&p->user->__count); | |
938 | atomic_inc(&p->user->processes); | |
939 | get_group_info(p->group_info); | |
940 | ||
941 | /* | |
942 | * If multiple threads are within copy_process(), then this check | |
943 | * triggers too late. This doesn't hurt, the check is only there | |
944 | * to stop root fork bombs. | |
945 | */ | |
946 | if (nr_threads >= max_threads) | |
947 | goto bad_fork_cleanup_count; | |
948 | ||
a1261f54 | 949 | if (!try_module_get(task_thread_info(p)->exec_domain->module)) |
1da177e4 LT |
950 | goto bad_fork_cleanup_count; |
951 | ||
952 | if (p->binfmt && !try_module_get(p->binfmt->module)) | |
953 | goto bad_fork_cleanup_put_domain; | |
954 | ||
955 | p->did_exec = 0; | |
956 | copy_flags(clone_flags, p); | |
957 | p->pid = pid; | |
958 | retval = -EFAULT; | |
959 | if (clone_flags & CLONE_PARENT_SETTID) | |
960 | if (put_user(p->pid, parent_tidptr)) | |
961 | goto bad_fork_cleanup; | |
962 | ||
963 | p->proc_dentry = NULL; | |
964 | ||
965 | INIT_LIST_HEAD(&p->children); | |
966 | INIT_LIST_HEAD(&p->sibling); | |
967 | p->vfork_done = NULL; | |
968 | spin_lock_init(&p->alloc_lock); | |
969 | spin_lock_init(&p->proc_lock); | |
970 | ||
971 | clear_tsk_thread_flag(p, TIF_SIGPENDING); | |
972 | init_sigpending(&p->pending); | |
973 | ||
974 | p->utime = cputime_zero; | |
975 | p->stime = cputime_zero; | |
976 | p->sched_time = 0; | |
977 | p->rchar = 0; /* I/O counter: bytes read */ | |
978 | p->wchar = 0; /* I/O counter: bytes written */ | |
979 | p->syscr = 0; /* I/O counter: read syscalls */ | |
980 | p->syscw = 0; /* I/O counter: write syscalls */ | |
981 | acct_clear_integrals(p); | |
982 | ||
983 | p->it_virt_expires = cputime_zero; | |
984 | p->it_prof_expires = cputime_zero; | |
985 | p->it_sched_expires = 0; | |
986 | INIT_LIST_HEAD(&p->cpu_timers[0]); | |
987 | INIT_LIST_HEAD(&p->cpu_timers[1]); | |
988 | INIT_LIST_HEAD(&p->cpu_timers[2]); | |
989 | ||
990 | p->lock_depth = -1; /* -1 = no lock */ | |
991 | do_posix_clock_monotonic_gettime(&p->start_time); | |
992 | p->security = NULL; | |
993 | p->io_context = NULL; | |
994 | p->io_wait = NULL; | |
995 | p->audit_context = NULL; | |
b4b26418 | 996 | cpuset_fork(p); |
1da177e4 LT |
997 | #ifdef CONFIG_NUMA |
998 | p->mempolicy = mpol_copy(p->mempolicy); | |
999 | if (IS_ERR(p->mempolicy)) { | |
1000 | retval = PTR_ERR(p->mempolicy); | |
1001 | p->mempolicy = NULL; | |
b4b26418 | 1002 | goto bad_fork_cleanup_cpuset; |
1da177e4 LT |
1003 | } |
1004 | #endif | |
1005 | ||
408894ee IM |
1006 | #ifdef CONFIG_DEBUG_MUTEXES |
1007 | p->blocked_on = NULL; /* not blocked yet */ | |
1008 | #endif | |
1009 | ||
1da177e4 LT |
1010 | p->tgid = p->pid; |
1011 | if (clone_flags & CLONE_THREAD) | |
1012 | p->tgid = current->tgid; | |
1013 | ||
1014 | if ((retval = security_task_alloc(p))) | |
1015 | goto bad_fork_cleanup_policy; | |
1016 | if ((retval = audit_alloc(p))) | |
1017 | goto bad_fork_cleanup_security; | |
1018 | /* copy all the process information */ | |
1019 | if ((retval = copy_semundo(clone_flags, p))) | |
1020 | goto bad_fork_cleanup_audit; | |
1021 | if ((retval = copy_files(clone_flags, p))) | |
1022 | goto bad_fork_cleanup_semundo; | |
1023 | if ((retval = copy_fs(clone_flags, p))) | |
1024 | goto bad_fork_cleanup_files; | |
1025 | if ((retval = copy_sighand(clone_flags, p))) | |
1026 | goto bad_fork_cleanup_fs; | |
1027 | if ((retval = copy_signal(clone_flags, p))) | |
1028 | goto bad_fork_cleanup_sighand; | |
1029 | if ((retval = copy_mm(clone_flags, p))) | |
1030 | goto bad_fork_cleanup_signal; | |
1031 | if ((retval = copy_keys(clone_flags, p))) | |
1032 | goto bad_fork_cleanup_mm; | |
1033 | if ((retval = copy_namespace(clone_flags, p))) | |
1034 | goto bad_fork_cleanup_keys; | |
1035 | retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs); | |
1036 | if (retval) | |
1037 | goto bad_fork_cleanup_namespace; | |
1038 | ||
1039 | p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL; | |
1040 | /* | |
1041 | * Clear TID on mm_release()? | |
1042 | */ | |
1043 | p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL; | |
1044 | ||
1045 | /* | |
1046 | * Syscall tracing should be turned off in the child regardless | |
1047 | * of CLONE_PTRACE. | |
1048 | */ | |
1049 | clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE); | |
ed75e8d5 LV |
1050 | #ifdef TIF_SYSCALL_EMU |
1051 | clear_tsk_thread_flag(p, TIF_SYSCALL_EMU); | |
1052 | #endif | |
1da177e4 LT |
1053 | |
1054 | /* Our parent execution domain becomes current domain | |
1055 | These must match for thread signalling to apply */ | |
1056 | ||
1057 | p->parent_exec_id = p->self_exec_id; | |
1058 | ||
1059 | /* ok, now we should be set up.. */ | |
1060 | p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL); | |
1061 | p->pdeath_signal = 0; | |
1062 | p->exit_state = 0; | |
1063 | ||
1da177e4 LT |
1064 | /* |
1065 | * Ok, make it visible to the rest of the system. | |
1066 | * We dont wake it up yet. | |
1067 | */ | |
1068 | p->group_leader = p; | |
1069 | INIT_LIST_HEAD(&p->ptrace_children); | |
1070 | INIT_LIST_HEAD(&p->ptrace_list); | |
1071 | ||
476d139c NP |
1072 | /* Perform scheduler related setup. Assign this task to a CPU. */ |
1073 | sched_fork(p, clone_flags); | |
1074 | ||
1da177e4 LT |
1075 | /* Need tasklist lock for parent etc handling! */ |
1076 | write_lock_irq(&tasklist_lock); | |
1077 | ||
1078 | /* | |
476d139c NP |
1079 | * The task hasn't been attached yet, so its cpus_allowed mask will |
1080 | * not be changed, nor will its assigned CPU. | |
1081 | * | |
1082 | * The cpus_allowed mask of the parent may have changed after it was | |
1083 | * copied first time - so re-copy it here, then check the child's CPU | |
1084 | * to ensure it is on a valid CPU (and if not, just force it back to | |
1085 | * parent's CPU). This avoids alot of nasty races. | |
1da177e4 LT |
1086 | */ |
1087 | p->cpus_allowed = current->cpus_allowed; | |
26ff6ad9 SV |
1088 | if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) || |
1089 | !cpu_online(task_cpu(p)))) | |
476d139c | 1090 | set_task_cpu(p, smp_processor_id()); |
1da177e4 LT |
1091 | |
1092 | /* | |
1093 | * Check for pending SIGKILL! The new thread should not be allowed | |
1094 | * to slip out of an OOM kill. (or normal SIGKILL.) | |
1095 | */ | |
1096 | if (sigismember(¤t->pending.signal, SIGKILL)) { | |
1097 | write_unlock_irq(&tasklist_lock); | |
1098 | retval = -EINTR; | |
1099 | goto bad_fork_cleanup_namespace; | |
1100 | } | |
1101 | ||
1102 | /* CLONE_PARENT re-uses the old parent */ | |
1103 | if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) | |
1104 | p->real_parent = current->real_parent; | |
1105 | else | |
1106 | p->real_parent = current; | |
1107 | p->parent = p->real_parent; | |
1108 | ||
1109 | if (clone_flags & CLONE_THREAD) { | |
1110 | spin_lock(¤t->sighand->siglock); | |
1111 | /* | |
1112 | * Important: if an exit-all has been started then | |
1113 | * do not create this new thread - the whole thread | |
1114 | * group is supposed to exit anyway. | |
1115 | */ | |
1116 | if (current->signal->flags & SIGNAL_GROUP_EXIT) { | |
1117 | spin_unlock(¤t->sighand->siglock); | |
1118 | write_unlock_irq(&tasklist_lock); | |
1119 | retval = -EAGAIN; | |
1120 | goto bad_fork_cleanup_namespace; | |
1121 | } | |
1122 | p->group_leader = current->group_leader; | |
1123 | ||
1124 | if (current->signal->group_stop_count > 0) { | |
1125 | /* | |
1126 | * There is an all-stop in progress for the group. | |
1127 | * We ourselves will stop as soon as we check signals. | |
1128 | * Make the new thread part of that group stop too. | |
1129 | */ | |
1130 | current->signal->group_stop_count++; | |
1131 | set_tsk_thread_flag(p, TIF_SIGPENDING); | |
1132 | } | |
1133 | ||
1134 | if (!cputime_eq(current->signal->it_virt_expires, | |
1135 | cputime_zero) || | |
1136 | !cputime_eq(current->signal->it_prof_expires, | |
1137 | cputime_zero) || | |
1138 | current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY || | |
1139 | !list_empty(¤t->signal->cpu_timers[0]) || | |
1140 | !list_empty(¤t->signal->cpu_timers[1]) || | |
1141 | !list_empty(¤t->signal->cpu_timers[2])) { | |
1142 | /* | |
1143 | * Have child wake up on its first tick to check | |
1144 | * for process CPU timers. | |
1145 | */ | |
1146 | p->it_prof_expires = jiffies_to_cputime(1); | |
1147 | } | |
1148 | ||
1149 | spin_unlock(¤t->sighand->siglock); | |
1150 | } | |
1151 | ||
22e2c507 JA |
1152 | /* |
1153 | * inherit ioprio | |
1154 | */ | |
1155 | p->ioprio = current->ioprio; | |
1156 | ||
1da177e4 LT |
1157 | SET_LINKS(p); |
1158 | if (unlikely(p->ptrace & PT_PTRACED)) | |
1159 | __ptrace_link(p, current->parent); | |
1160 | ||
1da177e4 LT |
1161 | attach_pid(p, PIDTYPE_PID, p->pid); |
1162 | attach_pid(p, PIDTYPE_TGID, p->tgid); | |
1163 | if (thread_group_leader(p)) { | |
9a5d3023 OL |
1164 | p->signal->tty = current->signal->tty; |
1165 | p->signal->pgrp = process_group(current); | |
1166 | p->signal->session = current->signal->session; | |
1da177e4 LT |
1167 | attach_pid(p, PIDTYPE_PGID, process_group(p)); |
1168 | attach_pid(p, PIDTYPE_SID, p->signal->session); | |
1169 | if (p->pid) | |
1170 | __get_cpu_var(process_counts)++; | |
1171 | } | |
1172 | ||
1173 | nr_threads++; | |
1174 | total_forks++; | |
1175 | write_unlock_irq(&tasklist_lock); | |
c13cf856 | 1176 | proc_fork_connector(p); |
1da177e4 LT |
1177 | return p; |
1178 | ||
1179 | bad_fork_cleanup_namespace: | |
1180 | exit_namespace(p); | |
1181 | bad_fork_cleanup_keys: | |
1182 | exit_keys(p); | |
1183 | bad_fork_cleanup_mm: | |
1184 | if (p->mm) | |
1185 | mmput(p->mm); | |
1186 | bad_fork_cleanup_signal: | |
1187 | exit_signal(p); | |
1188 | bad_fork_cleanup_sighand: | |
1189 | exit_sighand(p); | |
1190 | bad_fork_cleanup_fs: | |
1191 | exit_fs(p); /* blocking */ | |
1192 | bad_fork_cleanup_files: | |
1193 | exit_files(p); /* blocking */ | |
1194 | bad_fork_cleanup_semundo: | |
1195 | exit_sem(p); | |
1196 | bad_fork_cleanup_audit: | |
1197 | audit_free(p); | |
1198 | bad_fork_cleanup_security: | |
1199 | security_task_free(p); | |
1200 | bad_fork_cleanup_policy: | |
1201 | #ifdef CONFIG_NUMA | |
1202 | mpol_free(p->mempolicy); | |
b4b26418 | 1203 | bad_fork_cleanup_cpuset: |
1da177e4 | 1204 | #endif |
b4b26418 | 1205 | cpuset_exit(p); |
1da177e4 LT |
1206 | bad_fork_cleanup: |
1207 | if (p->binfmt) | |
1208 | module_put(p->binfmt->module); | |
1209 | bad_fork_cleanup_put_domain: | |
a1261f54 | 1210 | module_put(task_thread_info(p)->exec_domain->module); |
1da177e4 LT |
1211 | bad_fork_cleanup_count: |
1212 | put_group_info(p->group_info); | |
1213 | atomic_dec(&p->user->processes); | |
1214 | free_uid(p->user); | |
1215 | bad_fork_free: | |
1216 | free_task(p); | |
fe7d37d1 ON |
1217 | fork_out: |
1218 | return ERR_PTR(retval); | |
1da177e4 LT |
1219 | } |
1220 | ||
1221 | struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs) | |
1222 | { | |
1223 | memset(regs, 0, sizeof(struct pt_regs)); | |
1224 | return regs; | |
1225 | } | |
1226 | ||
1227 | task_t * __devinit fork_idle(int cpu) | |
1228 | { | |
1229 | task_t *task; | |
1230 | struct pt_regs regs; | |
1231 | ||
1232 | task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL, NULL, 0); | |
1233 | if (!task) | |
1234 | return ERR_PTR(-ENOMEM); | |
1235 | init_idle(task, cpu); | |
1236 | unhash_process(task); | |
1237 | return task; | |
1238 | } | |
1239 | ||
1240 | static inline int fork_traceflag (unsigned clone_flags) | |
1241 | { | |
1242 | if (clone_flags & CLONE_UNTRACED) | |
1243 | return 0; | |
1244 | else if (clone_flags & CLONE_VFORK) { | |
1245 | if (current->ptrace & PT_TRACE_VFORK) | |
1246 | return PTRACE_EVENT_VFORK; | |
1247 | } else if ((clone_flags & CSIGNAL) != SIGCHLD) { | |
1248 | if (current->ptrace & PT_TRACE_CLONE) | |
1249 | return PTRACE_EVENT_CLONE; | |
1250 | } else if (current->ptrace & PT_TRACE_FORK) | |
1251 | return PTRACE_EVENT_FORK; | |
1252 | ||
1253 | return 0; | |
1254 | } | |
1255 | ||
1256 | /* | |
1257 | * Ok, this is the main fork-routine. | |
1258 | * | |
1259 | * It copies the process, and if successful kick-starts | |
1260 | * it and waits for it to finish using the VM if required. | |
1261 | */ | |
1262 | long do_fork(unsigned long clone_flags, | |
1263 | unsigned long stack_start, | |
1264 | struct pt_regs *regs, | |
1265 | unsigned long stack_size, | |
1266 | int __user *parent_tidptr, | |
1267 | int __user *child_tidptr) | |
1268 | { | |
1269 | struct task_struct *p; | |
1270 | int trace = 0; | |
1271 | long pid = alloc_pidmap(); | |
1272 | ||
1273 | if (pid < 0) | |
1274 | return -EAGAIN; | |
1275 | if (unlikely(current->ptrace)) { | |
1276 | trace = fork_traceflag (clone_flags); | |
1277 | if (trace) | |
1278 | clone_flags |= CLONE_PTRACE; | |
1279 | } | |
1280 | ||
1281 | p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid); | |
1282 | /* | |
1283 | * Do this prior waking up the new thread - the thread pointer | |
1284 | * might get invalid after that point, if the thread exits quickly. | |
1285 | */ | |
1286 | if (!IS_ERR(p)) { | |
1287 | struct completion vfork; | |
1288 | ||
1289 | if (clone_flags & CLONE_VFORK) { | |
1290 | p->vfork_done = &vfork; | |
1291 | init_completion(&vfork); | |
1292 | } | |
1293 | ||
1294 | if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) { | |
1295 | /* | |
1296 | * We'll start up with an immediate SIGSTOP. | |
1297 | */ | |
1298 | sigaddset(&p->pending.signal, SIGSTOP); | |
1299 | set_tsk_thread_flag(p, TIF_SIGPENDING); | |
1300 | } | |
1301 | ||
1302 | if (!(clone_flags & CLONE_STOPPED)) | |
1303 | wake_up_new_task(p, clone_flags); | |
1304 | else | |
1305 | p->state = TASK_STOPPED; | |
1306 | ||
1307 | if (unlikely (trace)) { | |
1308 | current->ptrace_message = pid; | |
1309 | ptrace_notify ((trace << 8) | SIGTRAP); | |
1310 | } | |
1311 | ||
1312 | if (clone_flags & CLONE_VFORK) { | |
1313 | wait_for_completion(&vfork); | |
1314 | if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) | |
1315 | ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP); | |
1316 | } | |
1317 | } else { | |
1318 | free_pidmap(pid); | |
1319 | pid = PTR_ERR(p); | |
1320 | } | |
1321 | return pid; | |
1322 | } | |
1323 | ||
5fd63b30 RT |
1324 | #ifndef ARCH_MIN_MMSTRUCT_ALIGN |
1325 | #define ARCH_MIN_MMSTRUCT_ALIGN 0 | |
1326 | #endif | |
1327 | ||
1da177e4 LT |
1328 | void __init proc_caches_init(void) |
1329 | { | |
1330 | sighand_cachep = kmem_cache_create("sighand_cache", | |
1331 | sizeof(struct sighand_struct), 0, | |
1332 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1333 | signal_cachep = kmem_cache_create("signal_cache", | |
1334 | sizeof(struct signal_struct), 0, | |
1335 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1336 | files_cachep = kmem_cache_create("files_cache", | |
1337 | sizeof(struct files_struct), 0, | |
1338 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1339 | fs_cachep = kmem_cache_create("fs_cache", | |
1340 | sizeof(struct fs_struct), 0, | |
1341 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1342 | vm_area_cachep = kmem_cache_create("vm_area_struct", | |
1343 | sizeof(struct vm_area_struct), 0, | |
1344 | SLAB_PANIC, NULL, NULL); | |
1345 | mm_cachep = kmem_cache_create("mm_struct", | |
5fd63b30 | 1346 | sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, |
1da177e4 LT |
1347 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
1348 | } | |
cf2e340f JD |
1349 | |
1350 | ||
1351 | /* | |
1352 | * Check constraints on flags passed to the unshare system call and | |
1353 | * force unsharing of additional process context as appropriate. | |
1354 | */ | |
1355 | static inline void check_unshare_flags(unsigned long *flags_ptr) | |
1356 | { | |
1357 | /* | |
1358 | * If unsharing a thread from a thread group, must also | |
1359 | * unshare vm. | |
1360 | */ | |
1361 | if (*flags_ptr & CLONE_THREAD) | |
1362 | *flags_ptr |= CLONE_VM; | |
1363 | ||
1364 | /* | |
1365 | * If unsharing vm, must also unshare signal handlers. | |
1366 | */ | |
1367 | if (*flags_ptr & CLONE_VM) | |
1368 | *flags_ptr |= CLONE_SIGHAND; | |
1369 | ||
1370 | /* | |
1371 | * If unsharing signal handlers and the task was created | |
1372 | * using CLONE_THREAD, then must unshare the thread | |
1373 | */ | |
1374 | if ((*flags_ptr & CLONE_SIGHAND) && | |
1375 | (atomic_read(¤t->signal->count) > 1)) | |
1376 | *flags_ptr |= CLONE_THREAD; | |
1377 | ||
1378 | /* | |
1379 | * If unsharing namespace, must also unshare filesystem information. | |
1380 | */ | |
1381 | if (*flags_ptr & CLONE_NEWNS) | |
1382 | *flags_ptr |= CLONE_FS; | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * Unsharing of tasks created with CLONE_THREAD is not supported yet | |
1387 | */ | |
1388 | static int unshare_thread(unsigned long unshare_flags) | |
1389 | { | |
1390 | if (unshare_flags & CLONE_THREAD) | |
1391 | return -EINVAL; | |
1392 | ||
1393 | return 0; | |
1394 | } | |
1395 | ||
1396 | /* | |
99d1419d | 1397 | * Unshare the filesystem structure if it is being shared |
cf2e340f JD |
1398 | */ |
1399 | static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp) | |
1400 | { | |
1401 | struct fs_struct *fs = current->fs; | |
1402 | ||
1403 | if ((unshare_flags & CLONE_FS) && | |
99d1419d JD |
1404 | (fs && atomic_read(&fs->count) > 1)) { |
1405 | *new_fsp = __copy_fs_struct(current->fs); | |
1406 | if (!*new_fsp) | |
1407 | return -ENOMEM; | |
1408 | } | |
cf2e340f JD |
1409 | |
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | /* | |
741a2951 | 1414 | * Unshare the namespace structure if it is being shared |
cf2e340f | 1415 | */ |
741a2951 | 1416 | static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs) |
cf2e340f JD |
1417 | { |
1418 | struct namespace *ns = current->namespace; | |
1419 | ||
1420 | if ((unshare_flags & CLONE_NEWNS) && | |
741a2951 JD |
1421 | (ns && atomic_read(&ns->count) > 1)) { |
1422 | if (!capable(CAP_SYS_ADMIN)) | |
1423 | return -EPERM; | |
1424 | ||
1425 | *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs); | |
1426 | if (!*new_nsp) | |
1427 | return -ENOMEM; | |
1428 | } | |
cf2e340f JD |
1429 | |
1430 | return 0; | |
1431 | } | |
1432 | ||
1433 | /* | |
1434 | * Unsharing of sighand for tasks created with CLONE_SIGHAND is not | |
1435 | * supported yet | |
1436 | */ | |
1437 | static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp) | |
1438 | { | |
1439 | struct sighand_struct *sigh = current->sighand; | |
1440 | ||
1441 | if ((unshare_flags & CLONE_SIGHAND) && | |
1442 | (sigh && atomic_read(&sigh->count) > 1)) | |
1443 | return -EINVAL; | |
1444 | else | |
1445 | return 0; | |
1446 | } | |
1447 | ||
1448 | /* | |
a0a7ec30 | 1449 | * Unshare vm if it is being shared |
cf2e340f JD |
1450 | */ |
1451 | static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp) | |
1452 | { | |
1453 | struct mm_struct *mm = current->mm; | |
1454 | ||
1455 | if ((unshare_flags & CLONE_VM) && | |
a0a7ec30 JD |
1456 | (mm && atomic_read(&mm->mm_users) > 1)) { |
1457 | *new_mmp = dup_mm(current); | |
1458 | if (!*new_mmp) | |
1459 | return -ENOMEM; | |
1460 | } | |
cf2e340f JD |
1461 | |
1462 | return 0; | |
cf2e340f JD |
1463 | } |
1464 | ||
1465 | /* | |
1466 | * Unsharing of files for tasks created with CLONE_FILES is not supported yet | |
1467 | */ | |
1468 | static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp) | |
1469 | { | |
1470 | struct files_struct *fd = current->files; | |
1471 | ||
1472 | if ((unshare_flags & CLONE_FILES) && | |
1473 | (fd && atomic_read(&fd->count) > 1)) | |
1474 | return -EINVAL; | |
1475 | ||
1476 | return 0; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not | |
1481 | * supported yet | |
1482 | */ | |
1483 | static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp) | |
1484 | { | |
1485 | if (unshare_flags & CLONE_SYSVSEM) | |
1486 | return -EINVAL; | |
1487 | ||
1488 | return 0; | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * unshare allows a process to 'unshare' part of the process | |
1493 | * context which was originally shared using clone. copy_* | |
1494 | * functions used by do_fork() cannot be used here directly | |
1495 | * because they modify an inactive task_struct that is being | |
1496 | * constructed. Here we are modifying the current, active, | |
1497 | * task_struct. | |
1498 | */ | |
1499 | asmlinkage long sys_unshare(unsigned long unshare_flags) | |
1500 | { | |
1501 | int err = 0; | |
1502 | struct fs_struct *fs, *new_fs = NULL; | |
1503 | struct namespace *ns, *new_ns = NULL; | |
1504 | struct sighand_struct *sigh, *new_sigh = NULL; | |
1505 | struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL; | |
1506 | struct files_struct *fd, *new_fd = NULL; | |
1507 | struct sem_undo_list *new_ulist = NULL; | |
1508 | ||
1509 | check_unshare_flags(&unshare_flags); | |
1510 | ||
1511 | if ((err = unshare_thread(unshare_flags))) | |
1512 | goto bad_unshare_out; | |
1513 | if ((err = unshare_fs(unshare_flags, &new_fs))) | |
1514 | goto bad_unshare_cleanup_thread; | |
741a2951 | 1515 | if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs))) |
cf2e340f JD |
1516 | goto bad_unshare_cleanup_fs; |
1517 | if ((err = unshare_sighand(unshare_flags, &new_sigh))) | |
1518 | goto bad_unshare_cleanup_ns; | |
1519 | if ((err = unshare_vm(unshare_flags, &new_mm))) | |
1520 | goto bad_unshare_cleanup_sigh; | |
1521 | if ((err = unshare_fd(unshare_flags, &new_fd))) | |
1522 | goto bad_unshare_cleanup_vm; | |
1523 | if ((err = unshare_semundo(unshare_flags, &new_ulist))) | |
1524 | goto bad_unshare_cleanup_fd; | |
1525 | ||
1526 | if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) { | |
1527 | ||
1528 | task_lock(current); | |
1529 | ||
1530 | if (new_fs) { | |
1531 | fs = current->fs; | |
1532 | current->fs = new_fs; | |
1533 | new_fs = fs; | |
1534 | } | |
1535 | ||
1536 | if (new_ns) { | |
1537 | ns = current->namespace; | |
1538 | current->namespace = new_ns; | |
1539 | new_ns = ns; | |
1540 | } | |
1541 | ||
1542 | if (new_sigh) { | |
1543 | sigh = current->sighand; | |
1544 | current->sighand = new_sigh; | |
1545 | new_sigh = sigh; | |
1546 | } | |
1547 | ||
1548 | if (new_mm) { | |
1549 | mm = current->mm; | |
1550 | active_mm = current->active_mm; | |
1551 | current->mm = new_mm; | |
1552 | current->active_mm = new_mm; | |
1553 | activate_mm(active_mm, new_mm); | |
1554 | new_mm = mm; | |
1555 | } | |
1556 | ||
1557 | if (new_fd) { | |
1558 | fd = current->files; | |
1559 | current->files = new_fd; | |
1560 | new_fd = fd; | |
1561 | } | |
1562 | ||
1563 | task_unlock(current); | |
1564 | } | |
1565 | ||
1566 | bad_unshare_cleanup_fd: | |
1567 | if (new_fd) | |
1568 | put_files_struct(new_fd); | |
1569 | ||
1570 | bad_unshare_cleanup_vm: | |
1571 | if (new_mm) | |
1572 | mmput(new_mm); | |
1573 | ||
1574 | bad_unshare_cleanup_sigh: | |
1575 | if (new_sigh) | |
1576 | if (atomic_dec_and_test(&new_sigh->count)) | |
1577 | kmem_cache_free(sighand_cachep, new_sigh); | |
1578 | ||
1579 | bad_unshare_cleanup_ns: | |
1580 | if (new_ns) | |
1581 | put_namespace(new_ns); | |
1582 | ||
1583 | bad_unshare_cleanup_fs: | |
1584 | if (new_fs) | |
1585 | put_fs_struct(new_fs); | |
1586 | ||
1587 | bad_unshare_cleanup_thread: | |
1588 | bad_unshare_out: | |
1589 | return err; | |
1590 | } |