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/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_counter.h>
65
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
72
73 #include <trace/events/sched.h>
74
75 /*
76 * Protected counters by write_lock_irq(&tasklist_lock)
77 */
78 unsigned long total_forks; /* Handle normal Linux uptimes. */
79 int nr_threads; /* The idle threads do not count.. */
80
81 int max_threads; /* tunable limit on nr_threads */
82
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
84
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
86
87 int nr_processes(void)
88 {
89 int cpu;
90 int total = 0;
91
92 for_each_online_cpu(cpu)
93 total += per_cpu(process_counts, cpu);
94
95 return total;
96 }
97
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct() kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk) kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
102 #endif
103
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
106 {
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108 gfp_t mask = GFP_KERNEL | __GFP_ZERO;
109 #else
110 gfp_t mask = GFP_KERNEL;
111 #endif
112 return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
113 }
114
115 static inline void free_thread_info(struct thread_info *ti)
116 {
117 free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
118 }
119 #endif
120
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
123
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
126
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
129
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
132
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
135
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
138
139 void free_task(struct task_struct *tsk)
140 {
141 prop_local_destroy_single(&tsk->dirties);
142 free_thread_info(tsk->stack);
143 rt_mutex_debug_task_free(tsk);
144 ftrace_graph_exit_task(tsk);
145 free_task_struct(tsk);
146 }
147 EXPORT_SYMBOL(free_task);
148
149 void __put_task_struct(struct task_struct *tsk)
150 {
151 WARN_ON(!tsk->exit_state);
152 WARN_ON(atomic_read(&tsk->usage));
153 WARN_ON(tsk == current);
154
155 put_cred(tsk->real_cred);
156 put_cred(tsk->cred);
157 delayacct_tsk_free(tsk);
158
159 if (!profile_handoff_task(tsk))
160 free_task(tsk);
161 }
162
163 /*
164 * macro override instead of weak attribute alias, to workaround
165 * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
166 */
167 #ifndef arch_task_cache_init
168 #define arch_task_cache_init()
169 #endif
170
171 void __init fork_init(unsigned long mempages)
172 {
173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
174 #ifndef ARCH_MIN_TASKALIGN
175 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
176 #endif
177 /* create a slab on which task_structs can be allocated */
178 task_struct_cachep =
179 kmem_cache_create("task_struct", sizeof(struct task_struct),
180 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
181 #endif
182
183 /* do the arch specific task caches init */
184 arch_task_cache_init();
185
186 /*
187 * The default maximum number of threads is set to a safe
188 * value: the thread structures can take up at most half
189 * of memory.
190 */
191 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
192
193 /*
194 * we need to allow at least 20 threads to boot a system
195 */
196 if(max_threads < 20)
197 max_threads = 20;
198
199 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
200 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
201 init_task.signal->rlim[RLIMIT_SIGPENDING] =
202 init_task.signal->rlim[RLIMIT_NPROC];
203 }
204
205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
206 struct task_struct *src)
207 {
208 *dst = *src;
209 return 0;
210 }
211
212 static struct task_struct *dup_task_struct(struct task_struct *orig)
213 {
214 struct task_struct *tsk;
215 struct thread_info *ti;
216 unsigned long *stackend;
217
218 int err;
219
220 prepare_to_copy(orig);
221
222 tsk = alloc_task_struct();
223 if (!tsk)
224 return NULL;
225
226 ti = alloc_thread_info(tsk);
227 if (!ti) {
228 free_task_struct(tsk);
229 return NULL;
230 }
231
232 err = arch_dup_task_struct(tsk, orig);
233 if (err)
234 goto out;
235
236 tsk->stack = ti;
237
238 err = prop_local_init_single(&tsk->dirties);
239 if (err)
240 goto out;
241
242 setup_thread_stack(tsk, orig);
243 stackend = end_of_stack(tsk);
244 *stackend = STACK_END_MAGIC; /* for overflow detection */
245
246 #ifdef CONFIG_CC_STACKPROTECTOR
247 tsk->stack_canary = get_random_int();
248 #endif
249
250 /* One for us, one for whoever does the "release_task()" (usually parent) */
251 atomic_set(&tsk->usage,2);
252 atomic_set(&tsk->fs_excl, 0);
253 #ifdef CONFIG_BLK_DEV_IO_TRACE
254 tsk->btrace_seq = 0;
255 #endif
256 tsk->splice_pipe = NULL;
257 return tsk;
258
259 out:
260 free_thread_info(ti);
261 free_task_struct(tsk);
262 return NULL;
263 }
264
265 #ifdef CONFIG_MMU
266 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
267 {
268 struct vm_area_struct *mpnt, *tmp, **pprev;
269 struct rb_node **rb_link, *rb_parent;
270 int retval;
271 unsigned long charge;
272 struct mempolicy *pol;
273
274 down_write(&oldmm->mmap_sem);
275 flush_cache_dup_mm(oldmm);
276 /*
277 * Not linked in yet - no deadlock potential:
278 */
279 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
280
281 mm->locked_vm = 0;
282 mm->mmap = NULL;
283 mm->mmap_cache = NULL;
284 mm->free_area_cache = oldmm->mmap_base;
285 mm->cached_hole_size = ~0UL;
286 mm->map_count = 0;
287 cpumask_clear(mm_cpumask(mm));
288 mm->mm_rb = RB_ROOT;
289 rb_link = &mm->mm_rb.rb_node;
290 rb_parent = NULL;
291 pprev = &mm->mmap;
292
293 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
294 struct file *file;
295
296 if (mpnt->vm_flags & VM_DONTCOPY) {
297 long pages = vma_pages(mpnt);
298 mm->total_vm -= pages;
299 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
300 -pages);
301 continue;
302 }
303 charge = 0;
304 if (mpnt->vm_flags & VM_ACCOUNT) {
305 unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
306 if (security_vm_enough_memory(len))
307 goto fail_nomem;
308 charge = len;
309 }
310 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
311 if (!tmp)
312 goto fail_nomem;
313 *tmp = *mpnt;
314 pol = mpol_dup(vma_policy(mpnt));
315 retval = PTR_ERR(pol);
316 if (IS_ERR(pol))
317 goto fail_nomem_policy;
318 vma_set_policy(tmp, pol);
319 tmp->vm_flags &= ~VM_LOCKED;
320 tmp->vm_mm = mm;
321 tmp->vm_next = NULL;
322 anon_vma_link(tmp);
323 file = tmp->vm_file;
324 if (file) {
325 struct inode *inode = file->f_path.dentry->d_inode;
326 struct address_space *mapping = file->f_mapping;
327
328 get_file(file);
329 if (tmp->vm_flags & VM_DENYWRITE)
330 atomic_dec(&inode->i_writecount);
331 spin_lock(&mapping->i_mmap_lock);
332 if (tmp->vm_flags & VM_SHARED)
333 mapping->i_mmap_writable++;
334 tmp->vm_truncate_count = mpnt->vm_truncate_count;
335 flush_dcache_mmap_lock(mapping);
336 /* insert tmp into the share list, just after mpnt */
337 vma_prio_tree_add(tmp, mpnt);
338 flush_dcache_mmap_unlock(mapping);
339 spin_unlock(&mapping->i_mmap_lock);
340 }
341
342 /*
343 * Clear hugetlb-related page reserves for children. This only
344 * affects MAP_PRIVATE mappings. Faults generated by the child
345 * are not guaranteed to succeed, even if read-only
346 */
347 if (is_vm_hugetlb_page(tmp))
348 reset_vma_resv_huge_pages(tmp);
349
350 /*
351 * Link in the new vma and copy the page table entries.
352 */
353 *pprev = tmp;
354 pprev = &tmp->vm_next;
355
356 __vma_link_rb(mm, tmp, rb_link, rb_parent);
357 rb_link = &tmp->vm_rb.rb_right;
358 rb_parent = &tmp->vm_rb;
359
360 mm->map_count++;
361 retval = copy_page_range(mm, oldmm, mpnt);
362
363 if (tmp->vm_ops && tmp->vm_ops->open)
364 tmp->vm_ops->open(tmp);
365
366 if (retval)
367 goto out;
368 }
369 /* a new mm has just been created */
370 arch_dup_mmap(oldmm, mm);
371 retval = 0;
372 out:
373 up_write(&mm->mmap_sem);
374 flush_tlb_mm(oldmm);
375 up_write(&oldmm->mmap_sem);
376 return retval;
377 fail_nomem_policy:
378 kmem_cache_free(vm_area_cachep, tmp);
379 fail_nomem:
380 retval = -ENOMEM;
381 vm_unacct_memory(charge);
382 goto out;
383 }
384
385 static inline int mm_alloc_pgd(struct mm_struct * mm)
386 {
387 mm->pgd = pgd_alloc(mm);
388 if (unlikely(!mm->pgd))
389 return -ENOMEM;
390 return 0;
391 }
392
393 static inline void mm_free_pgd(struct mm_struct * mm)
394 {
395 pgd_free(mm, mm->pgd);
396 }
397 #else
398 #define dup_mmap(mm, oldmm) (0)
399 #define mm_alloc_pgd(mm) (0)
400 #define mm_free_pgd(mm)
401 #endif /* CONFIG_MMU */
402
403 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
404
405 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
406 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
407
408 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
409
410 static int __init coredump_filter_setup(char *s)
411 {
412 default_dump_filter =
413 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
414 MMF_DUMP_FILTER_MASK;
415 return 1;
416 }
417
418 __setup("coredump_filter=", coredump_filter_setup);
419
420 #include <linux/init_task.h>
421
422 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
423 {
424 atomic_set(&mm->mm_users, 1);
425 atomic_set(&mm->mm_count, 1);
426 init_rwsem(&mm->mmap_sem);
427 INIT_LIST_HEAD(&mm->mmlist);
428 mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
429 mm->core_state = NULL;
430 mm->nr_ptes = 0;
431 set_mm_counter(mm, file_rss, 0);
432 set_mm_counter(mm, anon_rss, 0);
433 spin_lock_init(&mm->page_table_lock);
434 spin_lock_init(&mm->ioctx_lock);
435 INIT_HLIST_HEAD(&mm->ioctx_list);
436 mm->free_area_cache = TASK_UNMAPPED_BASE;
437 mm->cached_hole_size = ~0UL;
438 mm_init_owner(mm, p);
439
440 if (likely(!mm_alloc_pgd(mm))) {
441 mm->def_flags = 0;
442 mmu_notifier_mm_init(mm);
443 return mm;
444 }
445
446 free_mm(mm);
447 return NULL;
448 }
449
450 /*
451 * Allocate and initialize an mm_struct.
452 */
453 struct mm_struct * mm_alloc(void)
454 {
455 struct mm_struct * mm;
456
457 mm = allocate_mm();
458 if (mm) {
459 memset(mm, 0, sizeof(*mm));
460 mm = mm_init(mm, current);
461 }
462 return mm;
463 }
464
465 /*
466 * Called when the last reference to the mm
467 * is dropped: either by a lazy thread or by
468 * mmput. Free the page directory and the mm.
469 */
470 void __mmdrop(struct mm_struct *mm)
471 {
472 BUG_ON(mm == &init_mm);
473 mm_free_pgd(mm);
474 destroy_context(mm);
475 mmu_notifier_mm_destroy(mm);
476 free_mm(mm);
477 }
478 EXPORT_SYMBOL_GPL(__mmdrop);
479
480 /*
481 * Decrement the use count and release all resources for an mm.
482 */
483 void mmput(struct mm_struct *mm)
484 {
485 might_sleep();
486
487 if (atomic_dec_and_test(&mm->mm_users)) {
488 exit_aio(mm);
489 exit_mmap(mm);
490 set_mm_exe_file(mm, NULL);
491 if (!list_empty(&mm->mmlist)) {
492 spin_lock(&mmlist_lock);
493 list_del(&mm->mmlist);
494 spin_unlock(&mmlist_lock);
495 }
496 put_swap_token(mm);
497 mmdrop(mm);
498 }
499 }
500 EXPORT_SYMBOL_GPL(mmput);
501
502 /**
503 * get_task_mm - acquire a reference to the task's mm
504 *
505 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
506 * this kernel workthread has transiently adopted a user mm with use_mm,
507 * to do its AIO) is not set and if so returns a reference to it, after
508 * bumping up the use count. User must release the mm via mmput()
509 * after use. Typically used by /proc and ptrace.
510 */
511 struct mm_struct *get_task_mm(struct task_struct *task)
512 {
513 struct mm_struct *mm;
514
515 task_lock(task);
516 mm = task->mm;
517 if (mm) {
518 if (task->flags & PF_KTHREAD)
519 mm = NULL;
520 else
521 atomic_inc(&mm->mm_users);
522 }
523 task_unlock(task);
524 return mm;
525 }
526 EXPORT_SYMBOL_GPL(get_task_mm);
527
528 /* Please note the differences between mmput and mm_release.
529 * mmput is called whenever we stop holding onto a mm_struct,
530 * error success whatever.
531 *
532 * mm_release is called after a mm_struct has been removed
533 * from the current process.
534 *
535 * This difference is important for error handling, when we
536 * only half set up a mm_struct for a new process and need to restore
537 * the old one. Because we mmput the new mm_struct before
538 * restoring the old one. . .
539 * Eric Biederman 10 January 1998
540 */
541 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
542 {
543 struct completion *vfork_done = tsk->vfork_done;
544
545 /* Get rid of any futexes when releasing the mm */
546 #ifdef CONFIG_FUTEX
547 if (unlikely(tsk->robust_list)) {
548 exit_robust_list(tsk);
549 tsk->robust_list = NULL;
550 }
551 #ifdef CONFIG_COMPAT
552 if (unlikely(tsk->compat_robust_list)) {
553 compat_exit_robust_list(tsk);
554 tsk->compat_robust_list = NULL;
555 }
556 #endif
557 if (unlikely(!list_empty(&tsk->pi_state_list)))
558 exit_pi_state_list(tsk);
559 #endif
560
561 /* Get rid of any cached register state */
562 deactivate_mm(tsk, mm);
563
564 /* notify parent sleeping on vfork() */
565 if (vfork_done) {
566 tsk->vfork_done = NULL;
567 complete(vfork_done);
568 }
569
570 /*
571 * If we're exiting normally, clear a user-space tid field if
572 * requested. We leave this alone when dying by signal, to leave
573 * the value intact in a core dump, and to save the unnecessary
574 * trouble otherwise. Userland only wants this done for a sys_exit.
575 */
576 if (tsk->clear_child_tid) {
577 if (!(tsk->flags & PF_SIGNALED) &&
578 atomic_read(&mm->mm_users) > 1) {
579 /*
580 * We don't check the error code - if userspace has
581 * not set up a proper pointer then tough luck.
582 */
583 put_user(0, tsk->clear_child_tid);
584 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
585 1, NULL, NULL, 0);
586 }
587 tsk->clear_child_tid = NULL;
588 }
589 }
590
591 /*
592 * Allocate a new mm structure and copy contents from the
593 * mm structure of the passed in task structure.
594 */
595 struct mm_struct *dup_mm(struct task_struct *tsk)
596 {
597 struct mm_struct *mm, *oldmm = current->mm;
598 int err;
599
600 if (!oldmm)
601 return NULL;
602
603 mm = allocate_mm();
604 if (!mm)
605 goto fail_nomem;
606
607 memcpy(mm, oldmm, sizeof(*mm));
608
609 /* Initializing for Swap token stuff */
610 mm->token_priority = 0;
611 mm->last_interval = 0;
612
613 if (!mm_init(mm, tsk))
614 goto fail_nomem;
615
616 if (init_new_context(tsk, mm))
617 goto fail_nocontext;
618
619 dup_mm_exe_file(oldmm, mm);
620
621 err = dup_mmap(mm, oldmm);
622 if (err)
623 goto free_pt;
624
625 mm->hiwater_rss = get_mm_rss(mm);
626 mm->hiwater_vm = mm->total_vm;
627
628 return mm;
629
630 free_pt:
631 mmput(mm);
632
633 fail_nomem:
634 return NULL;
635
636 fail_nocontext:
637 /*
638 * If init_new_context() failed, we cannot use mmput() to free the mm
639 * because it calls destroy_context()
640 */
641 mm_free_pgd(mm);
642 free_mm(mm);
643 return NULL;
644 }
645
646 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
647 {
648 struct mm_struct * mm, *oldmm;
649 int retval;
650
651 tsk->min_flt = tsk->maj_flt = 0;
652 tsk->nvcsw = tsk->nivcsw = 0;
653 #ifdef CONFIG_DETECT_HUNG_TASK
654 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
655 #endif
656
657 tsk->mm = NULL;
658 tsk->active_mm = NULL;
659
660 /*
661 * Are we cloning a kernel thread?
662 *
663 * We need to steal a active VM for that..
664 */
665 oldmm = current->mm;
666 if (!oldmm)
667 return 0;
668
669 if (clone_flags & CLONE_VM) {
670 atomic_inc(&oldmm->mm_users);
671 mm = oldmm;
672 goto good_mm;
673 }
674
675 retval = -ENOMEM;
676 mm = dup_mm(tsk);
677 if (!mm)
678 goto fail_nomem;
679
680 good_mm:
681 /* Initializing for Swap token stuff */
682 mm->token_priority = 0;
683 mm->last_interval = 0;
684
685 tsk->mm = mm;
686 tsk->active_mm = mm;
687 return 0;
688
689 fail_nomem:
690 return retval;
691 }
692
693 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
694 {
695 struct fs_struct *fs = current->fs;
696 if (clone_flags & CLONE_FS) {
697 /* tsk->fs is already what we want */
698 write_lock(&fs->lock);
699 if (fs->in_exec) {
700 write_unlock(&fs->lock);
701 return -EAGAIN;
702 }
703 fs->users++;
704 write_unlock(&fs->lock);
705 return 0;
706 }
707 tsk->fs = copy_fs_struct(fs);
708 if (!tsk->fs)
709 return -ENOMEM;
710 return 0;
711 }
712
713 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
714 {
715 struct files_struct *oldf, *newf;
716 int error = 0;
717
718 /*
719 * A background process may not have any files ...
720 */
721 oldf = current->files;
722 if (!oldf)
723 goto out;
724
725 if (clone_flags & CLONE_FILES) {
726 atomic_inc(&oldf->count);
727 goto out;
728 }
729
730 newf = dup_fd(oldf, &error);
731 if (!newf)
732 goto out;
733
734 tsk->files = newf;
735 error = 0;
736 out:
737 return error;
738 }
739
740 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
741 {
742 #ifdef CONFIG_BLOCK
743 struct io_context *ioc = current->io_context;
744
745 if (!ioc)
746 return 0;
747 /*
748 * Share io context with parent, if CLONE_IO is set
749 */
750 if (clone_flags & CLONE_IO) {
751 tsk->io_context = ioc_task_link(ioc);
752 if (unlikely(!tsk->io_context))
753 return -ENOMEM;
754 } else if (ioprio_valid(ioc->ioprio)) {
755 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
756 if (unlikely(!tsk->io_context))
757 return -ENOMEM;
758
759 tsk->io_context->ioprio = ioc->ioprio;
760 }
761 #endif
762 return 0;
763 }
764
765 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
766 {
767 struct sighand_struct *sig;
768
769 if (clone_flags & CLONE_SIGHAND) {
770 atomic_inc(¤t->sighand->count);
771 return 0;
772 }
773 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
774 rcu_assign_pointer(tsk->sighand, sig);
775 if (!sig)
776 return -ENOMEM;
777 atomic_set(&sig->count, 1);
778 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
779 return 0;
780 }
781
782 void __cleanup_sighand(struct sighand_struct *sighand)
783 {
784 if (atomic_dec_and_test(&sighand->count))
785 kmem_cache_free(sighand_cachep, sighand);
786 }
787
788
789 /*
790 * Initialize POSIX timer handling for a thread group.
791 */
792 static void posix_cpu_timers_init_group(struct signal_struct *sig)
793 {
794 /* Thread group counters. */
795 thread_group_cputime_init(sig);
796
797 /* Expiration times and increments. */
798 sig->it_virt_expires = cputime_zero;
799 sig->it_virt_incr = cputime_zero;
800 sig->it_prof_expires = cputime_zero;
801 sig->it_prof_incr = cputime_zero;
802
803 /* Cached expiration times. */
804 sig->cputime_expires.prof_exp = cputime_zero;
805 sig->cputime_expires.virt_exp = cputime_zero;
806 sig->cputime_expires.sched_exp = 0;
807
808 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
809 sig->cputime_expires.prof_exp =
810 secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
811 sig->cputimer.running = 1;
812 }
813
814 /* The timer lists. */
815 INIT_LIST_HEAD(&sig->cpu_timers[0]);
816 INIT_LIST_HEAD(&sig->cpu_timers[1]);
817 INIT_LIST_HEAD(&sig->cpu_timers[2]);
818 }
819
820 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
821 {
822 struct signal_struct *sig;
823
824 if (clone_flags & CLONE_THREAD)
825 return 0;
826
827 sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
828 tsk->signal = sig;
829 if (!sig)
830 return -ENOMEM;
831
832 atomic_set(&sig->count, 1);
833 atomic_set(&sig->live, 1);
834 init_waitqueue_head(&sig->wait_chldexit);
835 sig->flags = 0;
836 if (clone_flags & CLONE_NEWPID)
837 sig->flags |= SIGNAL_UNKILLABLE;
838 sig->group_exit_code = 0;
839 sig->group_exit_task = NULL;
840 sig->group_stop_count = 0;
841 sig->curr_target = tsk;
842 init_sigpending(&sig->shared_pending);
843 INIT_LIST_HEAD(&sig->posix_timers);
844
845 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
846 sig->it_real_incr.tv64 = 0;
847 sig->real_timer.function = it_real_fn;
848
849 sig->leader = 0; /* session leadership doesn't inherit */
850 sig->tty_old_pgrp = NULL;
851 sig->tty = NULL;
852
853 sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
854 sig->gtime = cputime_zero;
855 sig->cgtime = cputime_zero;
856 sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
857 sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
858 sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
859 task_io_accounting_init(&sig->ioac);
860 sig->sum_sched_runtime = 0;
861 taskstats_tgid_init(sig);
862
863 task_lock(current->group_leader);
864 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
865 task_unlock(current->group_leader);
866
867 posix_cpu_timers_init_group(sig);
868
869 acct_init_pacct(&sig->pacct);
870
871 tty_audit_fork(sig);
872
873 return 0;
874 }
875
876 void __cleanup_signal(struct signal_struct *sig)
877 {
878 thread_group_cputime_free(sig);
879 tty_kref_put(sig->tty);
880 kmem_cache_free(signal_cachep, sig);
881 }
882
883 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
884 {
885 unsigned long new_flags = p->flags;
886
887 new_flags &= ~PF_SUPERPRIV;
888 new_flags |= PF_FORKNOEXEC;
889 new_flags |= PF_STARTING;
890 p->flags = new_flags;
891 clear_freeze_flag(p);
892 }
893
894 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
895 {
896 current->clear_child_tid = tidptr;
897
898 return task_pid_vnr(current);
899 }
900
901 static void rt_mutex_init_task(struct task_struct *p)
902 {
903 spin_lock_init(&p->pi_lock);
904 #ifdef CONFIG_RT_MUTEXES
905 plist_head_init(&p->pi_waiters, &p->pi_lock);
906 p->pi_blocked_on = NULL;
907 #endif
908 }
909
910 #ifdef CONFIG_MM_OWNER
911 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
912 {
913 mm->owner = p;
914 }
915 #endif /* CONFIG_MM_OWNER */
916
917 /*
918 * Initialize POSIX timer handling for a single task.
919 */
920 static void posix_cpu_timers_init(struct task_struct *tsk)
921 {
922 tsk->cputime_expires.prof_exp = cputime_zero;
923 tsk->cputime_expires.virt_exp = cputime_zero;
924 tsk->cputime_expires.sched_exp = 0;
925 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
926 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
927 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
928 }
929
930 /*
931 * This creates a new process as a copy of the old one,
932 * but does not actually start it yet.
933 *
934 * It copies the registers, and all the appropriate
935 * parts of the process environment (as per the clone
936 * flags). The actual kick-off is left to the caller.
937 */
938 static struct task_struct *copy_process(unsigned long clone_flags,
939 unsigned long stack_start,
940 struct pt_regs *regs,
941 unsigned long stack_size,
942 int __user *child_tidptr,
943 struct pid *pid,
944 int trace)
945 {
946 int retval;
947 struct task_struct *p;
948 int cgroup_callbacks_done = 0;
949
950 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
951 return ERR_PTR(-EINVAL);
952
953 /*
954 * Thread groups must share signals as well, and detached threads
955 * can only be started up within the thread group.
956 */
957 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
958 return ERR_PTR(-EINVAL);
959
960 /*
961 * Shared signal handlers imply shared VM. By way of the above,
962 * thread groups also imply shared VM. Blocking this case allows
963 * for various simplifications in other code.
964 */
965 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
966 return ERR_PTR(-EINVAL);
967
968 retval = security_task_create(clone_flags);
969 if (retval)
970 goto fork_out;
971
972 retval = -ENOMEM;
973 p = dup_task_struct(current);
974 if (!p)
975 goto fork_out;
976
977 ftrace_graph_init_task(p);
978
979 rt_mutex_init_task(p);
980
981 #ifdef CONFIG_PROVE_LOCKING
982 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
983 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
984 #endif
985 retval = -EAGAIN;
986 if (atomic_read(&p->real_cred->user->processes) >=
987 p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
988 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
989 p->real_cred->user != INIT_USER)
990 goto bad_fork_free;
991 }
992
993 retval = copy_creds(p, clone_flags);
994 if (retval < 0)
995 goto bad_fork_free;
996
997 /*
998 * If multiple threads are within copy_process(), then this check
999 * triggers too late. This doesn't hurt, the check is only there
1000 * to stop root fork bombs.
1001 */
1002 retval = -EAGAIN;
1003 if (nr_threads >= max_threads)
1004 goto bad_fork_cleanup_count;
1005
1006 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1007 goto bad_fork_cleanup_count;
1008
1009 if (p->binfmt && !try_module_get(p->binfmt->module))
1010 goto bad_fork_cleanup_put_domain;
1011
1012 p->did_exec = 0;
1013 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1014 copy_flags(clone_flags, p);
1015 INIT_LIST_HEAD(&p->children);
1016 INIT_LIST_HEAD(&p->sibling);
1017 #ifdef CONFIG_PREEMPT_RCU
1018 p->rcu_read_lock_nesting = 0;
1019 p->rcu_flipctr_idx = 0;
1020 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1021 p->vfork_done = NULL;
1022 spin_lock_init(&p->alloc_lock);
1023
1024 init_sigpending(&p->pending);
1025
1026 p->utime = cputime_zero;
1027 p->stime = cputime_zero;
1028 p->gtime = cputime_zero;
1029 p->utimescaled = cputime_zero;
1030 p->stimescaled = cputime_zero;
1031 p->prev_utime = cputime_zero;
1032 p->prev_stime = cputime_zero;
1033
1034 p->default_timer_slack_ns = current->timer_slack_ns;
1035
1036 task_io_accounting_init(&p->ioac);
1037 acct_clear_integrals(p);
1038
1039 posix_cpu_timers_init(p);
1040
1041 p->lock_depth = -1; /* -1 = no lock */
1042 do_posix_clock_monotonic_gettime(&p->start_time);
1043 p->real_start_time = p->start_time;
1044 monotonic_to_bootbased(&p->real_start_time);
1045 p->io_context = NULL;
1046 p->audit_context = NULL;
1047 cgroup_fork(p);
1048 #ifdef CONFIG_NUMA
1049 p->mempolicy = mpol_dup(p->mempolicy);
1050 if (IS_ERR(p->mempolicy)) {
1051 retval = PTR_ERR(p->mempolicy);
1052 p->mempolicy = NULL;
1053 goto bad_fork_cleanup_cgroup;
1054 }
1055 mpol_fix_fork_child_flag(p);
1056 #endif
1057 #ifdef CONFIG_TRACE_IRQFLAGS
1058 p->irq_events = 0;
1059 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1060 p->hardirqs_enabled = 1;
1061 #else
1062 p->hardirqs_enabled = 0;
1063 #endif
1064 p->hardirq_enable_ip = 0;
1065 p->hardirq_enable_event = 0;
1066 p->hardirq_disable_ip = _THIS_IP_;
1067 p->hardirq_disable_event = 0;
1068 p->softirqs_enabled = 1;
1069 p->softirq_enable_ip = _THIS_IP_;
1070 p->softirq_enable_event = 0;
1071 p->softirq_disable_ip = 0;
1072 p->softirq_disable_event = 0;
1073 p->hardirq_context = 0;
1074 p->softirq_context = 0;
1075 #endif
1076 #ifdef CONFIG_LOCKDEP
1077 p->lockdep_depth = 0; /* no locks held yet */
1078 p->curr_chain_key = 0;
1079 p->lockdep_recursion = 0;
1080 #endif
1081
1082 #ifdef CONFIG_DEBUG_MUTEXES
1083 p->blocked_on = NULL; /* not blocked yet */
1084 #endif
1085
1086 p->bts = NULL;
1087
1088 /* Perform scheduler related setup. Assign this task to a CPU. */
1089 sched_fork(p, clone_flags);
1090
1091 retval = perf_counter_init_task(p);
1092 if (retval)
1093 goto bad_fork_cleanup_policy;
1094
1095 if ((retval = audit_alloc(p)))
1096 goto bad_fork_cleanup_policy;
1097 /* copy all the process information */
1098 if ((retval = copy_semundo(clone_flags, p)))
1099 goto bad_fork_cleanup_audit;
1100 if ((retval = copy_files(clone_flags, p)))
1101 goto bad_fork_cleanup_semundo;
1102 if ((retval = copy_fs(clone_flags, p)))
1103 goto bad_fork_cleanup_files;
1104 if ((retval = copy_sighand(clone_flags, p)))
1105 goto bad_fork_cleanup_fs;
1106 if ((retval = copy_signal(clone_flags, p)))
1107 goto bad_fork_cleanup_sighand;
1108 if ((retval = copy_mm(clone_flags, p)))
1109 goto bad_fork_cleanup_signal;
1110 if ((retval = copy_namespaces(clone_flags, p)))
1111 goto bad_fork_cleanup_mm;
1112 if ((retval = copy_io(clone_flags, p)))
1113 goto bad_fork_cleanup_namespaces;
1114 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1115 if (retval)
1116 goto bad_fork_cleanup_io;
1117
1118 if (pid != &init_struct_pid) {
1119 retval = -ENOMEM;
1120 pid = alloc_pid(p->nsproxy->pid_ns);
1121 if (!pid)
1122 goto bad_fork_cleanup_io;
1123
1124 if (clone_flags & CLONE_NEWPID) {
1125 retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1126 if (retval < 0)
1127 goto bad_fork_free_pid;
1128 }
1129 }
1130
1131 p->pid = pid_nr(pid);
1132 p->tgid = p->pid;
1133 if (clone_flags & CLONE_THREAD)
1134 p->tgid = current->tgid;
1135
1136 if (current->nsproxy != p->nsproxy) {
1137 retval = ns_cgroup_clone(p, pid);
1138 if (retval)
1139 goto bad_fork_free_pid;
1140 }
1141
1142 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1143 /*
1144 * Clear TID on mm_release()?
1145 */
1146 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1147 #ifdef CONFIG_FUTEX
1148 p->robust_list = NULL;
1149 #ifdef CONFIG_COMPAT
1150 p->compat_robust_list = NULL;
1151 #endif
1152 INIT_LIST_HEAD(&p->pi_state_list);
1153 p->pi_state_cache = NULL;
1154 #endif
1155 /*
1156 * sigaltstack should be cleared when sharing the same VM
1157 */
1158 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1159 p->sas_ss_sp = p->sas_ss_size = 0;
1160
1161 /*
1162 * Syscall tracing should be turned off in the child regardless
1163 * of CLONE_PTRACE.
1164 */
1165 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1166 #ifdef TIF_SYSCALL_EMU
1167 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1168 #endif
1169 clear_all_latency_tracing(p);
1170
1171 /* ok, now we should be set up.. */
1172 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1173 p->pdeath_signal = 0;
1174 p->exit_state = 0;
1175
1176 /*
1177 * Ok, make it visible to the rest of the system.
1178 * We dont wake it up yet.
1179 */
1180 p->group_leader = p;
1181 INIT_LIST_HEAD(&p->thread_group);
1182
1183 /* Now that the task is set up, run cgroup callbacks if
1184 * necessary. We need to run them before the task is visible
1185 * on the tasklist. */
1186 cgroup_fork_callbacks(p);
1187 cgroup_callbacks_done = 1;
1188
1189 /* Need tasklist lock for parent etc handling! */
1190 write_lock_irq(&tasklist_lock);
1191
1192 /*
1193 * The task hasn't been attached yet, so its cpus_allowed mask will
1194 * not be changed, nor will its assigned CPU.
1195 *
1196 * The cpus_allowed mask of the parent may have changed after it was
1197 * copied first time - so re-copy it here, then check the child's CPU
1198 * to ensure it is on a valid CPU (and if not, just force it back to
1199 * parent's CPU). This avoids alot of nasty races.
1200 */
1201 p->cpus_allowed = current->cpus_allowed;
1202 p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1203 if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1204 !cpu_online(task_cpu(p))))
1205 set_task_cpu(p, smp_processor_id());
1206
1207 /* CLONE_PARENT re-uses the old parent */
1208 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1209 p->real_parent = current->real_parent;
1210 p->parent_exec_id = current->parent_exec_id;
1211 } else {
1212 p->real_parent = current;
1213 p->parent_exec_id = current->self_exec_id;
1214 }
1215
1216 spin_lock(¤t->sighand->siglock);
1217
1218 /*
1219 * Process group and session signals need to be delivered to just the
1220 * parent before the fork or both the parent and the child after the
1221 * fork. Restart if a signal comes in before we add the new process to
1222 * it's process group.
1223 * A fatal signal pending means that current will exit, so the new
1224 * thread can't slip out of an OOM kill (or normal SIGKILL).
1225 */
1226 recalc_sigpending();
1227 if (signal_pending(current)) {
1228 spin_unlock(¤t->sighand->siglock);
1229 write_unlock_irq(&tasklist_lock);
1230 retval = -ERESTARTNOINTR;
1231 goto bad_fork_free_pid;
1232 }
1233
1234 if (clone_flags & CLONE_THREAD) {
1235 atomic_inc(¤t->signal->count);
1236 atomic_inc(¤t->signal->live);
1237 p->group_leader = current->group_leader;
1238 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1239 }
1240
1241 if (likely(p->pid)) {
1242 list_add_tail(&p->sibling, &p->real_parent->children);
1243 tracehook_finish_clone(p, clone_flags, trace);
1244
1245 if (thread_group_leader(p)) {
1246 if (clone_flags & CLONE_NEWPID)
1247 p->nsproxy->pid_ns->child_reaper = p;
1248
1249 p->signal->leader_pid = pid;
1250 tty_kref_put(p->signal->tty);
1251 p->signal->tty = tty_kref_get(current->signal->tty);
1252 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1253 attach_pid(p, PIDTYPE_SID, task_session(current));
1254 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1255 __get_cpu_var(process_counts)++;
1256 }
1257 attach_pid(p, PIDTYPE_PID, pid);
1258 nr_threads++;
1259 }
1260
1261 total_forks++;
1262 spin_unlock(¤t->sighand->siglock);
1263 write_unlock_irq(&tasklist_lock);
1264 proc_fork_connector(p);
1265 cgroup_post_fork(p);
1266 perf_counter_fork(p);
1267 return p;
1268
1269 bad_fork_free_pid:
1270 if (pid != &init_struct_pid)
1271 free_pid(pid);
1272 bad_fork_cleanup_io:
1273 put_io_context(p->io_context);
1274 bad_fork_cleanup_namespaces:
1275 exit_task_namespaces(p);
1276 bad_fork_cleanup_mm:
1277 if (p->mm)
1278 mmput(p->mm);
1279 bad_fork_cleanup_signal:
1280 if (!(clone_flags & CLONE_THREAD))
1281 __cleanup_signal(p->signal);
1282 bad_fork_cleanup_sighand:
1283 __cleanup_sighand(p->sighand);
1284 bad_fork_cleanup_fs:
1285 exit_fs(p); /* blocking */
1286 bad_fork_cleanup_files:
1287 exit_files(p); /* blocking */
1288 bad_fork_cleanup_semundo:
1289 exit_sem(p);
1290 bad_fork_cleanup_audit:
1291 audit_free(p);
1292 bad_fork_cleanup_policy:
1293 perf_counter_free_task(p);
1294 #ifdef CONFIG_NUMA
1295 mpol_put(p->mempolicy);
1296 bad_fork_cleanup_cgroup:
1297 #endif
1298 cgroup_exit(p, cgroup_callbacks_done);
1299 delayacct_tsk_free(p);
1300 if (p->binfmt)
1301 module_put(p->binfmt->module);
1302 bad_fork_cleanup_put_domain:
1303 module_put(task_thread_info(p)->exec_domain->module);
1304 bad_fork_cleanup_count:
1305 atomic_dec(&p->cred->user->processes);
1306 put_cred(p->real_cred);
1307 put_cred(p->cred);
1308 bad_fork_free:
1309 free_task(p);
1310 fork_out:
1311 return ERR_PTR(retval);
1312 }
1313
1314 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1315 {
1316 memset(regs, 0, sizeof(struct pt_regs));
1317 return regs;
1318 }
1319
1320 struct task_struct * __cpuinit fork_idle(int cpu)
1321 {
1322 struct task_struct *task;
1323 struct pt_regs regs;
1324
1325 task = copy_process(CLONE_VM, 0, idle_regs(®s), 0, NULL,
1326 &init_struct_pid, 0);
1327 if (!IS_ERR(task))
1328 init_idle(task, cpu);
1329
1330 return task;
1331 }
1332
1333 /*
1334 * Ok, this is the main fork-routine.
1335 *
1336 * It copies the process, and if successful kick-starts
1337 * it and waits for it to finish using the VM if required.
1338 */
1339 long do_fork(unsigned long clone_flags,
1340 unsigned long stack_start,
1341 struct pt_regs *regs,
1342 unsigned long stack_size,
1343 int __user *parent_tidptr,
1344 int __user *child_tidptr)
1345 {
1346 struct task_struct *p;
1347 int trace = 0;
1348 long nr;
1349
1350 /*
1351 * Do some preliminary argument and permissions checking before we
1352 * actually start allocating stuff
1353 */
1354 if (clone_flags & CLONE_NEWUSER) {
1355 if (clone_flags & CLONE_THREAD)
1356 return -EINVAL;
1357 /* hopefully this check will go away when userns support is
1358 * complete
1359 */
1360 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1361 !capable(CAP_SETGID))
1362 return -EPERM;
1363 }
1364
1365 /*
1366 * We hope to recycle these flags after 2.6.26
1367 */
1368 if (unlikely(clone_flags & CLONE_STOPPED)) {
1369 static int __read_mostly count = 100;
1370
1371 if (count > 0 && printk_ratelimit()) {
1372 char comm[TASK_COMM_LEN];
1373
1374 count--;
1375 printk(KERN_INFO "fork(): process `%s' used deprecated "
1376 "clone flags 0x%lx\n",
1377 get_task_comm(comm, current),
1378 clone_flags & CLONE_STOPPED);
1379 }
1380 }
1381
1382 /*
1383 * When called from kernel_thread, don't do user tracing stuff.
1384 */
1385 if (likely(user_mode(regs)))
1386 trace = tracehook_prepare_clone(clone_flags);
1387
1388 p = copy_process(clone_flags, stack_start, regs, stack_size,
1389 child_tidptr, NULL, trace);
1390 /*
1391 * Do this prior waking up the new thread - the thread pointer
1392 * might get invalid after that point, if the thread exits quickly.
1393 */
1394 if (!IS_ERR(p)) {
1395 struct completion vfork;
1396
1397 trace_sched_process_fork(current, p);
1398
1399 nr = task_pid_vnr(p);
1400
1401 if (clone_flags & CLONE_PARENT_SETTID)
1402 put_user(nr, parent_tidptr);
1403
1404 if (clone_flags & CLONE_VFORK) {
1405 p->vfork_done = &vfork;
1406 init_completion(&vfork);
1407 }
1408
1409 audit_finish_fork(p);
1410 tracehook_report_clone(regs, clone_flags, nr, p);
1411
1412 /*
1413 * We set PF_STARTING at creation in case tracing wants to
1414 * use this to distinguish a fully live task from one that
1415 * hasn't gotten to tracehook_report_clone() yet. Now we
1416 * clear it and set the child going.
1417 */
1418 p->flags &= ~PF_STARTING;
1419
1420 if (unlikely(clone_flags & CLONE_STOPPED)) {
1421 /*
1422 * We'll start up with an immediate SIGSTOP.
1423 */
1424 sigaddset(&p->pending.signal, SIGSTOP);
1425 set_tsk_thread_flag(p, TIF_SIGPENDING);
1426 __set_task_state(p, TASK_STOPPED);
1427 } else {
1428 wake_up_new_task(p, clone_flags);
1429 }
1430
1431 tracehook_report_clone_complete(trace, regs,
1432 clone_flags, nr, p);
1433
1434 if (clone_flags & CLONE_VFORK) {
1435 freezer_do_not_count();
1436 wait_for_completion(&vfork);
1437 freezer_count();
1438 tracehook_report_vfork_done(p, nr);
1439 }
1440 } else {
1441 nr = PTR_ERR(p);
1442 }
1443 return nr;
1444 }
1445
1446 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1447 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1448 #endif
1449
1450 static void sighand_ctor(void *data)
1451 {
1452 struct sighand_struct *sighand = data;
1453
1454 spin_lock_init(&sighand->siglock);
1455 init_waitqueue_head(&sighand->signalfd_wqh);
1456 }
1457
1458 void __init proc_caches_init(void)
1459 {
1460 sighand_cachep = kmem_cache_create("sighand_cache",
1461 sizeof(struct sighand_struct), 0,
1462 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1463 SLAB_NOTRACK, sighand_ctor);
1464 signal_cachep = kmem_cache_create("signal_cache",
1465 sizeof(struct signal_struct), 0,
1466 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1467 files_cachep = kmem_cache_create("files_cache",
1468 sizeof(struct files_struct), 0,
1469 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1470 fs_cachep = kmem_cache_create("fs_cache",
1471 sizeof(struct fs_struct), 0,
1472 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1473 mm_cachep = kmem_cache_create("mm_struct",
1474 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1475 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1476 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1477 mmap_init();
1478 }
1479
1480 /*
1481 * Check constraints on flags passed to the unshare system call and
1482 * force unsharing of additional process context as appropriate.
1483 */
1484 static void check_unshare_flags(unsigned long *flags_ptr)
1485 {
1486 /*
1487 * If unsharing a thread from a thread group, must also
1488 * unshare vm.
1489 */
1490 if (*flags_ptr & CLONE_THREAD)
1491 *flags_ptr |= CLONE_VM;
1492
1493 /*
1494 * If unsharing vm, must also unshare signal handlers.
1495 */
1496 if (*flags_ptr & CLONE_VM)
1497 *flags_ptr |= CLONE_SIGHAND;
1498
1499 /*
1500 * If unsharing signal handlers and the task was created
1501 * using CLONE_THREAD, then must unshare the thread
1502 */
1503 if ((*flags_ptr & CLONE_SIGHAND) &&
1504 (atomic_read(¤t->signal->count) > 1))
1505 *flags_ptr |= CLONE_THREAD;
1506
1507 /*
1508 * If unsharing namespace, must also unshare filesystem information.
1509 */
1510 if (*flags_ptr & CLONE_NEWNS)
1511 *flags_ptr |= CLONE_FS;
1512 }
1513
1514 /*
1515 * Unsharing of tasks created with CLONE_THREAD is not supported yet
1516 */
1517 static int unshare_thread(unsigned long unshare_flags)
1518 {
1519 if (unshare_flags & CLONE_THREAD)
1520 return -EINVAL;
1521
1522 return 0;
1523 }
1524
1525 /*
1526 * Unshare the filesystem structure if it is being shared
1527 */
1528 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1529 {
1530 struct fs_struct *fs = current->fs;
1531
1532 if (!(unshare_flags & CLONE_FS) || !fs)
1533 return 0;
1534
1535 /* don't need lock here; in the worst case we'll do useless copy */
1536 if (fs->users == 1)
1537 return 0;
1538
1539 *new_fsp = copy_fs_struct(fs);
1540 if (!*new_fsp)
1541 return -ENOMEM;
1542
1543 return 0;
1544 }
1545
1546 /*
1547 * Unsharing of sighand is not supported yet
1548 */
1549 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1550 {
1551 struct sighand_struct *sigh = current->sighand;
1552
1553 if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1554 return -EINVAL;
1555 else
1556 return 0;
1557 }
1558
1559 /*
1560 * Unshare vm if it is being shared
1561 */
1562 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1563 {
1564 struct mm_struct *mm = current->mm;
1565
1566 if ((unshare_flags & CLONE_VM) &&
1567 (mm && atomic_read(&mm->mm_users) > 1)) {
1568 return -EINVAL;
1569 }
1570
1571 return 0;
1572 }
1573
1574 /*
1575 * Unshare file descriptor table if it is being shared
1576 */
1577 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1578 {
1579 struct files_struct *fd = current->files;
1580 int error = 0;
1581
1582 if ((unshare_flags & CLONE_FILES) &&
1583 (fd && atomic_read(&fd->count) > 1)) {
1584 *new_fdp = dup_fd(fd, &error);
1585 if (!*new_fdp)
1586 return error;
1587 }
1588
1589 return 0;
1590 }
1591
1592 /*
1593 * unshare allows a process to 'unshare' part of the process
1594 * context which was originally shared using clone. copy_*
1595 * functions used by do_fork() cannot be used here directly
1596 * because they modify an inactive task_struct that is being
1597 * constructed. Here we are modifying the current, active,
1598 * task_struct.
1599 */
1600 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1601 {
1602 int err = 0;
1603 struct fs_struct *fs, *new_fs = NULL;
1604 struct sighand_struct *new_sigh = NULL;
1605 struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1606 struct files_struct *fd, *new_fd = NULL;
1607 struct nsproxy *new_nsproxy = NULL;
1608 int do_sysvsem = 0;
1609
1610 check_unshare_flags(&unshare_flags);
1611
1612 /* Return -EINVAL for all unsupported flags */
1613 err = -EINVAL;
1614 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1615 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1616 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1617 goto bad_unshare_out;
1618
1619 /*
1620 * CLONE_NEWIPC must also detach from the undolist: after switching
1621 * to a new ipc namespace, the semaphore arrays from the old
1622 * namespace are unreachable.
1623 */
1624 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1625 do_sysvsem = 1;
1626 if ((err = unshare_thread(unshare_flags)))
1627 goto bad_unshare_out;
1628 if ((err = unshare_fs(unshare_flags, &new_fs)))
1629 goto bad_unshare_cleanup_thread;
1630 if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1631 goto bad_unshare_cleanup_fs;
1632 if ((err = unshare_vm(unshare_flags, &new_mm)))
1633 goto bad_unshare_cleanup_sigh;
1634 if ((err = unshare_fd(unshare_flags, &new_fd)))
1635 goto bad_unshare_cleanup_vm;
1636 if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1637 new_fs)))
1638 goto bad_unshare_cleanup_fd;
1639
1640 if (new_fs || new_mm || new_fd || do_sysvsem || new_nsproxy) {
1641 if (do_sysvsem) {
1642 /*
1643 * CLONE_SYSVSEM is equivalent to sys_exit().
1644 */
1645 exit_sem(current);
1646 }
1647
1648 if (new_nsproxy) {
1649 switch_task_namespaces(current, new_nsproxy);
1650 new_nsproxy = NULL;
1651 }
1652
1653 task_lock(current);
1654
1655 if (new_fs) {
1656 fs = current->fs;
1657 write_lock(&fs->lock);
1658 current->fs = new_fs;
1659 if (--fs->users)
1660 new_fs = NULL;
1661 else
1662 new_fs = fs;
1663 write_unlock(&fs->lock);
1664 }
1665
1666 if (new_mm) {
1667 mm = current->mm;
1668 active_mm = current->active_mm;
1669 current->mm = new_mm;
1670 current->active_mm = new_mm;
1671 activate_mm(active_mm, new_mm);
1672 new_mm = mm;
1673 }
1674
1675 if (new_fd) {
1676 fd = current->files;
1677 current->files = new_fd;
1678 new_fd = fd;
1679 }
1680
1681 task_unlock(current);
1682 }
1683
1684 if (new_nsproxy)
1685 put_nsproxy(new_nsproxy);
1686
1687 bad_unshare_cleanup_fd:
1688 if (new_fd)
1689 put_files_struct(new_fd);
1690
1691 bad_unshare_cleanup_vm:
1692 if (new_mm)
1693 mmput(new_mm);
1694
1695 bad_unshare_cleanup_sigh:
1696 if (new_sigh)
1697 if (atomic_dec_and_test(&new_sigh->count))
1698 kmem_cache_free(sighand_cachep, new_sigh);
1699
1700 bad_unshare_cleanup_fs:
1701 if (new_fs)
1702 free_fs_struct(new_fs);
1703
1704 bad_unshare_cleanup_thread:
1705 bad_unshare_out:
1706 return err;
1707 }
1708
1709 /*
1710 * Helper to unshare the files of the current task.
1711 * We don't want to expose copy_files internals to
1712 * the exec layer of the kernel.
1713 */
1714
1715 int unshare_files(struct files_struct **displaced)
1716 {
1717 struct task_struct *task = current;
1718 struct files_struct *copy = NULL;
1719 int error;
1720
1721 error = unshare_fd(CLONE_FILES, ©);
1722 if (error || !copy) {
1723 *displaced = NULL;
1724 return error;
1725 }
1726 *displaced = task->files;
1727 task_lock(task);
1728 task->files = copy;
1729 task_unlock(task);
1730 return 0;
1731 }
1732
|
This page was automatically generated by the
LXR engine.
|