projects/linux·blob v3.19 · refs

2001 lines · 47.7 KiB

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/mm.h>
32#include <linux/vmacache.h>
33#include <linux/nsproxy.h>
34#include <linux/capability.h>
35#include <linux/cpu.h>
36#include <linux/cgroup.h>
37#include <linux/security.h>
38#include <linux/hugetlb.h>
39#include <linux/seccomp.h>
40#include <linux/swap.h>
41#include <linux/syscalls.h>
42#include <linux/jiffies.h>
43#include <linux/futex.h>
44#include <linux/compat.h>
45#include <linux/kthread.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/rcupdate.h>
48#include <linux/ptrace.h>
49#include <linux/mount.h>
50#include <linux/audit.h>
51#include <linux/memcontrol.h>
52#include <linux/ftrace.h>
53#include <linux/proc_fs.h>
54#include <linux/profile.h>
55#include <linux/rmap.h>
56#include <linux/ksm.h>
57#include <linux/acct.h>
58#include <linux/tsacct_kern.h>
59#include <linux/cn_proc.h>
60#include <linux/freezer.h>
61#include <linux/delayacct.h>
62#include <linux/taskstats_kern.h>
63#include <linux/random.h>
64#include <linux/tty.h>
65#include <linux/blkdev.h>
66#include <linux/fs_struct.h>
67#include <linux/magic.h>
68#include <linux/perf_event.h>
69#include <linux/posix-timers.h>
70#include <linux/user-return-notifier.h>
71#include <linux/oom.h>
72#include <linux/khugepaged.h>
73#include <linux/signalfd.h>
74#include <linux/uprobes.h>
75#include <linux/aio.h>
76#include <linux/compiler.h>
77
78#include <asm/pgtable.h>
79#include <asm/pgalloc.h>
80#include <asm/uaccess.h>
81#include <asm/mmu_context.h>
82#include <asm/cacheflush.h>
83#include <asm/tlbflush.h>
84
85#include <trace/events/sched.h>
86
87#define CREATE_TRACE_POINTS
88#include <trace/events/task.h>
89
90/*
91 * Protected counters by write_lock_irq(&tasklist_lock)
92 */
93unsigned long total_forks; /* Handle normal Linux uptimes. */
94int nr_threads; /* The idle threads do not count.. */
95
96int max_threads; /* tunable limit on nr_threads */
97
98DEFINE_PER_CPU(unsigned long, process_counts) = 0;
99
100__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
101
102#ifdef CONFIG_PROVE_RCU
103int lockdep_tasklist_lock_is_held(void)
104{
105 return lockdep_is_held(&tasklist_lock);
106}
107EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
108#endif /* #ifdef CONFIG_PROVE_RCU */
109
110int nr_processes(void)
111{
112 int cpu;
113 int total = 0;
114
115 for_each_possible_cpu(cpu)
116 total += per_cpu(process_counts, cpu);
117
118 return total;
119}
120
121void __weak arch_release_task_struct(struct task_struct *tsk)
122{
123}
124
125#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
126static struct kmem_cache *task_struct_cachep;
127
128static inline struct task_struct *alloc_task_struct_node(int node)
129{
130 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
131}
132
133static inline void free_task_struct(struct task_struct *tsk)
134{
135 kmem_cache_free(task_struct_cachep, tsk);
136}
137#endif
138
139void __weak arch_release_thread_info(struct thread_info *ti)
140{
141}
142
143#ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
144
145/*
146 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
147 * kmemcache based allocator.
148 */
149# if THREAD_SIZE >= PAGE_SIZE
150static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
151 int node)
152{
153 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
154 THREAD_SIZE_ORDER);
155
156 return page ? page_address(page) : NULL;
157}
158
159static inline void free_thread_info(struct thread_info *ti)
160{
161 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
162}
163# else
164static struct kmem_cache *thread_info_cache;
165
166static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
167 int node)
168{
169 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
170}
171
172static void free_thread_info(struct thread_info *ti)
173{
174 kmem_cache_free(thread_info_cache, ti);
175}
176
177void thread_info_cache_init(void)
178{
179 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
180 THREAD_SIZE, 0, NULL);
181 BUG_ON(thread_info_cache == NULL);
182}
183# endif
184#endif
185
186/* SLAB cache for signal_struct structures (tsk->signal) */
187static struct kmem_cache *signal_cachep;
188
189/* SLAB cache for sighand_struct structures (tsk->sighand) */
190struct kmem_cache *sighand_cachep;
191
192/* SLAB cache for files_struct structures (tsk->files) */
193struct kmem_cache *files_cachep;
194
195/* SLAB cache for fs_struct structures (tsk->fs) */
196struct kmem_cache *fs_cachep;
197
198/* SLAB cache for vm_area_struct structures */
199struct kmem_cache *vm_area_cachep;
200
201/* SLAB cache for mm_struct structures (tsk->mm) */
202static struct kmem_cache *mm_cachep;
203
204static void account_kernel_stack(struct thread_info *ti, int account)
205{
206 struct zone *zone = page_zone(virt_to_page(ti));
207
208 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
209}
210
211void free_task(struct task_struct *tsk)
212{
213 account_kernel_stack(tsk->stack, -1);
214 arch_release_thread_info(tsk->stack);
215 free_thread_info(tsk->stack);
216 rt_mutex_debug_task_free(tsk);
217 ftrace_graph_exit_task(tsk);
218 put_seccomp_filter(tsk);
219 arch_release_task_struct(tsk);
220 free_task_struct(tsk);
221}
222EXPORT_SYMBOL(free_task);
223
224static inline void free_signal_struct(struct signal_struct *sig)
225{
226 taskstats_tgid_free(sig);
227 sched_autogroup_exit(sig);
228 kmem_cache_free(signal_cachep, sig);
229}
230
231static inline void put_signal_struct(struct signal_struct *sig)
232{
233 if (atomic_dec_and_test(&sig->sigcnt))
234 free_signal_struct(sig);
235}
236
237void __put_task_struct(struct task_struct *tsk)
238{
239 WARN_ON(!tsk->exit_state);
240 WARN_ON(atomic_read(&tsk->usage));
241 WARN_ON(tsk == current);
242
243 task_numa_free(tsk);
244 security_task_free(tsk);
245 exit_creds(tsk);
246 delayacct_tsk_free(tsk);
247 put_signal_struct(tsk->signal);
248
249 if (!profile_handoff_task(tsk))
250 free_task(tsk);
251}
252EXPORT_SYMBOL_GPL(__put_task_struct);
253
254void __init __weak arch_task_cache_init(void) { }
255
256void __init fork_init(unsigned long mempages)
257{
258#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
259#ifndef ARCH_MIN_TASKALIGN
260#define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
261#endif
262 /* create a slab on which task_structs can be allocated */
263 task_struct_cachep =
264 kmem_cache_create("task_struct", sizeof(struct task_struct),
265 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
266#endif
267
268 /* do the arch specific task caches init */
269 arch_task_cache_init();
270
271 /*
272 * The default maximum number of threads is set to a safe
273 * value: the thread structures can take up at most half
274 * of memory.
275 */
276 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
277
278 /*
279 * we need to allow at least 20 threads to boot a system
280 */
281 if (max_threads < 20)
282 max_threads = 20;
283
284 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
285 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
286 init_task.signal->rlim[RLIMIT_SIGPENDING] =
287 init_task.signal->rlim[RLIMIT_NPROC];
288}
289
290int __weak arch_dup_task_struct(struct task_struct *dst,
291 struct task_struct *src)
292{
293 *dst = *src;
294 return 0;
295}
296
297void set_task_stack_end_magic(struct task_struct *tsk)
298{
299 unsigned long *stackend;
300
301 stackend = end_of_stack(tsk);
302 *stackend = STACK_END_MAGIC; /* for overflow detection */
303}
304
305static struct task_struct *dup_task_struct(struct task_struct *orig)
306{
307 struct task_struct *tsk;
308 struct thread_info *ti;
309 int node = tsk_fork_get_node(orig);
310 int err;
311
312 tsk = alloc_task_struct_node(node);
313 if (!tsk)
314 return NULL;
315
316 ti = alloc_thread_info_node(tsk, node);
317 if (!ti)
318 goto free_tsk;
319
320 err = arch_dup_task_struct(tsk, orig);
321 if (err)
322 goto free_ti;
323
324 tsk->stack = ti;
325#ifdef CONFIG_SECCOMP
326 /*
327 * We must handle setting up seccomp filters once we're under
328 * the sighand lock in case orig has changed between now and
329 * then. Until then, filter must be NULL to avoid messing up
330 * the usage counts on the error path calling free_task.
331 */
332 tsk->seccomp.filter = NULL;
333#endif
334
335 setup_thread_stack(tsk, orig);
336 clear_user_return_notifier(tsk);
337 clear_tsk_need_resched(tsk);
338 set_task_stack_end_magic(tsk);
339
340#ifdef CONFIG_CC_STACKPROTECTOR
341 tsk->stack_canary = get_random_int();
342#endif
343
344 /*
345 * One for us, one for whoever does the "release_task()" (usually
346 * parent)
347 */
348 atomic_set(&tsk->usage, 2);
349#ifdef CONFIG_BLK_DEV_IO_TRACE
350 tsk->btrace_seq = 0;
351#endif
352 tsk->splice_pipe = NULL;
353 tsk->task_frag.page = NULL;
354
355 account_kernel_stack(ti, 1);
356
357 return tsk;
358
359free_ti:
360 free_thread_info(ti);
361free_tsk:
362 free_task_struct(tsk);
363 return NULL;
364}
365
366#ifdef CONFIG_MMU
367static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
368{
369 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
370 struct rb_node **rb_link, *rb_parent;
371 int retval;
372 unsigned long charge;
373
374 uprobe_start_dup_mmap();
375 down_write(&oldmm->mmap_sem);
376 flush_cache_dup_mm(oldmm);
377 uprobe_dup_mmap(oldmm, mm);
378 /*
379 * Not linked in yet - no deadlock potential:
380 */
381 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
382
383 mm->total_vm = oldmm->total_vm;
384 mm->shared_vm = oldmm->shared_vm;
385 mm->exec_vm = oldmm->exec_vm;
386 mm->stack_vm = oldmm->stack_vm;
387
388 rb_link = &mm->mm_rb.rb_node;
389 rb_parent = NULL;
390 pprev = &mm->mmap;
391 retval = ksm_fork(mm, oldmm);
392 if (retval)
393 goto out;
394 retval = khugepaged_fork(mm, oldmm);
395 if (retval)
396 goto out;
397
398 prev = NULL;
399 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
400 struct file *file;
401
402 if (mpnt->vm_flags & VM_DONTCOPY) {
403 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
404 -vma_pages(mpnt));
405 continue;
406 }
407 charge = 0;
408 if (mpnt->vm_flags & VM_ACCOUNT) {
409 unsigned long len = vma_pages(mpnt);
410
411 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
412 goto fail_nomem;
413 charge = len;
414 }
415 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
416 if (!tmp)
417 goto fail_nomem;
418 *tmp = *mpnt;
419 INIT_LIST_HEAD(&tmp->anon_vma_chain);
420 retval = vma_dup_policy(mpnt, tmp);
421 if (retval)
422 goto fail_nomem_policy;
423 tmp->vm_mm = mm;
424 if (anon_vma_fork(tmp, mpnt))
425 goto fail_nomem_anon_vma_fork;
426 tmp->vm_flags &= ~VM_LOCKED;
427 tmp->vm_next = tmp->vm_prev = NULL;
428 file = tmp->vm_file;
429 if (file) {
430 struct inode *inode = file_inode(file);
431 struct address_space *mapping = file->f_mapping;
432
433 get_file(file);
434 if (tmp->vm_flags & VM_DENYWRITE)
435 atomic_dec(&inode->i_writecount);
436 i_mmap_lock_write(mapping);
437 if (tmp->vm_flags & VM_SHARED)
438 atomic_inc(&mapping->i_mmap_writable);
439 flush_dcache_mmap_lock(mapping);
440 /* insert tmp into the share list, just after mpnt */
441 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
442 vma_nonlinear_insert(tmp,
443 &mapping->i_mmap_nonlinear);
444 else
445 vma_interval_tree_insert_after(tmp, mpnt,
446 &mapping->i_mmap);
447 flush_dcache_mmap_unlock(mapping);
448 i_mmap_unlock_write(mapping);
449 }
450
451 /*
452 * Clear hugetlb-related page reserves for children. This only
453 * affects MAP_PRIVATE mappings. Faults generated by the child
454 * are not guaranteed to succeed, even if read-only
455 */
456 if (is_vm_hugetlb_page(tmp))
457 reset_vma_resv_huge_pages(tmp);
458
459 /*
460 * Link in the new vma and copy the page table entries.
461 */
462 *pprev = tmp;
463 pprev = &tmp->vm_next;
464 tmp->vm_prev = prev;
465 prev = tmp;
466
467 __vma_link_rb(mm, tmp, rb_link, rb_parent);
468 rb_link = &tmp->vm_rb.rb_right;
469 rb_parent = &tmp->vm_rb;
470
471 mm->map_count++;
472 retval = copy_page_range(mm, oldmm, mpnt);
473
474 if (tmp->vm_ops && tmp->vm_ops->open)
475 tmp->vm_ops->open(tmp);
476
477 if (retval)
478 goto out;
479 }
480 /* a new mm has just been created */
481 arch_dup_mmap(oldmm, mm);
482 retval = 0;
483out:
484 up_write(&mm->mmap_sem);
485 flush_tlb_mm(oldmm);
486 up_write(&oldmm->mmap_sem);
487 uprobe_end_dup_mmap();
488 return retval;
489fail_nomem_anon_vma_fork:
490 mpol_put(vma_policy(tmp));
491fail_nomem_policy:
492 kmem_cache_free(vm_area_cachep, tmp);
493fail_nomem:
494 retval = -ENOMEM;
495 vm_unacct_memory(charge);
496 goto out;
497}
498
499static inline int mm_alloc_pgd(struct mm_struct *mm)
500{
501 mm->pgd = pgd_alloc(mm);
502 if (unlikely(!mm->pgd))
503 return -ENOMEM;
504 return 0;
505}
506
507static inline void mm_free_pgd(struct mm_struct *mm)
508{
509 pgd_free(mm, mm->pgd);
510}
511#else
512#define dup_mmap(mm, oldmm) (0)
513#define mm_alloc_pgd(mm) (0)
514#define mm_free_pgd(mm)
515#endif /* CONFIG_MMU */
516
517__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
518
519#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
520#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
521
522static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
523
524static int __init coredump_filter_setup(char *s)
525{
526 default_dump_filter =
527 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
528 MMF_DUMP_FILTER_MASK;
529 return 1;
530}
531
532__setup("coredump_filter=", coredump_filter_setup);
533
534#include <linux/init_task.h>
535
536static void mm_init_aio(struct mm_struct *mm)
537{
538#ifdef CONFIG_AIO
539 spin_lock_init(&mm->ioctx_lock);
540 mm->ioctx_table = NULL;
541#endif
542}
543
544static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
545{
546#ifdef CONFIG_MEMCG
547 mm->owner = p;
548#endif
549}
550
551static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
552{
553 mm->mmap = NULL;
554 mm->mm_rb = RB_ROOT;
555 mm->vmacache_seqnum = 0;
556 atomic_set(&mm->mm_users, 1);
557 atomic_set(&mm->mm_count, 1);
558 init_rwsem(&mm->mmap_sem);
559 INIT_LIST_HEAD(&mm->mmlist);
560 mm->core_state = NULL;
561 atomic_long_set(&mm->nr_ptes, 0);
562 mm->map_count = 0;
563 mm->locked_vm = 0;
564 mm->pinned_vm = 0;
565 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
566 spin_lock_init(&mm->page_table_lock);
567 mm_init_cpumask(mm);
568 mm_init_aio(mm);
569 mm_init_owner(mm, p);
570 mmu_notifier_mm_init(mm);
571 clear_tlb_flush_pending(mm);
572#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
573 mm->pmd_huge_pte = NULL;
574#endif
575
576 if (current->mm) {
577 mm->flags = current->mm->flags & MMF_INIT_MASK;
578 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
579 } else {
580 mm->flags = default_dump_filter;
581 mm->def_flags = 0;
582 }
583
584 if (mm_alloc_pgd(mm))
585 goto fail_nopgd;
586
587 if (init_new_context(p, mm))
588 goto fail_nocontext;
589
590 return mm;
591
592fail_nocontext:
593 mm_free_pgd(mm);
594fail_nopgd:
595 free_mm(mm);
596 return NULL;
597}
598
599static void check_mm(struct mm_struct *mm)
600{
601 int i;
602
603 for (i = 0; i < NR_MM_COUNTERS; i++) {
604 long x = atomic_long_read(&mm->rss_stat.count[i]);
605
606 if (unlikely(x))
607 printk(KERN_ALERT "BUG: Bad rss-counter state "
608 "mm:%p idx:%d val:%ld\n", mm, i, x);
609 }
610#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
611 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
612#endif
613}
614
615/*
616 * Allocate and initialize an mm_struct.
617 */
618struct mm_struct *mm_alloc(void)
619{
620 struct mm_struct *mm;
621
622 mm = allocate_mm();
623 if (!mm)
624 return NULL;
625
626 memset(mm, 0, sizeof(*mm));
627 return mm_init(mm, current);
628}
629
630/*
631 * Called when the last reference to the mm
632 * is dropped: either by a lazy thread or by
633 * mmput. Free the page directory and the mm.
634 */
635void __mmdrop(struct mm_struct *mm)
636{
637 BUG_ON(mm == &init_mm);
638 mm_free_pgd(mm);
639 destroy_context(mm);
640 mmu_notifier_mm_destroy(mm);
641 check_mm(mm);
642 free_mm(mm);
643}
644EXPORT_SYMBOL_GPL(__mmdrop);
645
646/*
647 * Decrement the use count and release all resources for an mm.
648 */
649void mmput(struct mm_struct *mm)
650{
651 might_sleep();
652
653 if (atomic_dec_and_test(&mm->mm_users)) {
654 uprobe_clear_state(mm);
655 exit_aio(mm);
656 ksm_exit(mm);
657 khugepaged_exit(mm); /* must run before exit_mmap */
658 exit_mmap(mm);
659 set_mm_exe_file(mm, NULL);
660 if (!list_empty(&mm->mmlist)) {
661 spin_lock(&mmlist_lock);
662 list_del(&mm->mmlist);
663 spin_unlock(&mmlist_lock);
664 }
665 if (mm->binfmt)
666 module_put(mm->binfmt->module);
667 mmdrop(mm);
668 }
669}
670EXPORT_SYMBOL_GPL(mmput);
671
672void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
673{
674 if (new_exe_file)
675 get_file(new_exe_file);
676 if (mm->exe_file)
677 fput(mm->exe_file);
678 mm->exe_file = new_exe_file;
679}
680
681struct file *get_mm_exe_file(struct mm_struct *mm)
682{
683 struct file *exe_file;
684
685 /* We need mmap_sem to protect against races with removal of exe_file */
686 down_read(&mm->mmap_sem);
687 exe_file = mm->exe_file;
688 if (exe_file)
689 get_file(exe_file);
690 up_read(&mm->mmap_sem);
691 return exe_file;
692}
693
694static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
695{
696 /* It's safe to write the exe_file pointer without exe_file_lock because
697 * this is called during fork when the task is not yet in /proc */
698 newmm->exe_file = get_mm_exe_file(oldmm);
699}
700
701/**
702 * get_task_mm - acquire a reference to the task's mm
703 *
704 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
705 * this kernel workthread has transiently adopted a user mm with use_mm,
706 * to do its AIO) is not set and if so returns a reference to it, after
707 * bumping up the use count. User must release the mm via mmput()
708 * after use. Typically used by /proc and ptrace.
709 */
710struct mm_struct *get_task_mm(struct task_struct *task)
711{
712 struct mm_struct *mm;
713
714 task_lock(task);
715 mm = task->mm;
716 if (mm) {
717 if (task->flags & PF_KTHREAD)
718 mm = NULL;
719 else
720 atomic_inc(&mm->mm_users);
721 }
722 task_unlock(task);
723 return mm;
724}
725EXPORT_SYMBOL_GPL(get_task_mm);
726
727struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
728{
729 struct mm_struct *mm;
730 int err;
731
732 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
733 if (err)
734 return ERR_PTR(err);
735
736 mm = get_task_mm(task);
737 if (mm && mm != current->mm &&
738 !ptrace_may_access(task, mode)) {
739 mmput(mm);
740 mm = ERR_PTR(-EACCES);
741 }
742 mutex_unlock(&task->signal->cred_guard_mutex);
743
744 return mm;
745}
746
747static void complete_vfork_done(struct task_struct *tsk)
748{
749 struct completion *vfork;
750
751 task_lock(tsk);
752 vfork = tsk->vfork_done;
753 if (likely(vfork)) {
754 tsk->vfork_done = NULL;
755 complete(vfork);
756 }
757 task_unlock(tsk);
758}
759
760static int wait_for_vfork_done(struct task_struct *child,
761 struct completion *vfork)
762{
763 int killed;
764
765 freezer_do_not_count();
766 killed = wait_for_completion_killable(vfork);
767 freezer_count();
768
769 if (killed) {
770 task_lock(child);
771 child->vfork_done = NULL;
772 task_unlock(child);
773 }
774
775 put_task_struct(child);
776 return killed;
777}
778
779/* Please note the differences between mmput and mm_release.
780 * mmput is called whenever we stop holding onto a mm_struct,
781 * error success whatever.
782 *
783 * mm_release is called after a mm_struct has been removed
784 * from the current process.
785 *
786 * This difference is important for error handling, when we
787 * only half set up a mm_struct for a new process and need to restore
788 * the old one. Because we mmput the new mm_struct before
789 * restoring the old one. . .
790 * Eric Biederman 10 January 1998
791 */
792void mm_release(struct task_struct *tsk, struct mm_struct *mm)
793{
794 /* Get rid of any futexes when releasing the mm */
795#ifdef CONFIG_FUTEX
796 if (unlikely(tsk->robust_list)) {
797 exit_robust_list(tsk);
798 tsk->robust_list = NULL;
799 }
800#ifdef CONFIG_COMPAT
801 if (unlikely(tsk->compat_robust_list)) {
802 compat_exit_robust_list(tsk);
803 tsk->compat_robust_list = NULL;
804 }
805#endif
806 if (unlikely(!list_empty(&tsk->pi_state_list)))
807 exit_pi_state_list(tsk);
808#endif
809
810 uprobe_free_utask(tsk);
811
812 /* Get rid of any cached register state */
813 deactivate_mm(tsk, mm);
814
815 /*
816 * If we're exiting normally, clear a user-space tid field if
817 * requested. We leave this alone when dying by signal, to leave
818 * the value intact in a core dump, and to save the unnecessary
819 * trouble, say, a killed vfork parent shouldn't touch this mm.
820 * Userland only wants this done for a sys_exit.
821 */
822 if (tsk->clear_child_tid) {
823 if (!(tsk->flags & PF_SIGNALED) &&
824 atomic_read(&mm->mm_users) > 1) {
825 /*
826 * We don't check the error code - if userspace has
827 * not set up a proper pointer then tough luck.
828 */
829 put_user(0, tsk->clear_child_tid);
830 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
831 1, NULL, NULL, 0);
832 }
833 tsk->clear_child_tid = NULL;
834 }
835
836 /*
837 * All done, finally we can wake up parent and return this mm to him.
838 * Also kthread_stop() uses this completion for synchronization.
839 */
840 if (tsk->vfork_done)
841 complete_vfork_done(tsk);
842}
843
844/*
845 * Allocate a new mm structure and copy contents from the
846 * mm structure of the passed in task structure.
847 */
848static struct mm_struct *dup_mm(struct task_struct *tsk)
849{
850 struct mm_struct *mm, *oldmm = current->mm;
851 int err;
852
853 mm = allocate_mm();
854 if (!mm)
855 goto fail_nomem;
856
857 memcpy(mm, oldmm, sizeof(*mm));
858
859 if (!mm_init(mm, tsk))
860 goto fail_nomem;
861
862 dup_mm_exe_file(oldmm, mm);
863
864 err = dup_mmap(mm, oldmm);
865 if (err)
866 goto free_pt;
867
868 mm->hiwater_rss = get_mm_rss(mm);
869 mm->hiwater_vm = mm->total_vm;
870
871 if (mm->binfmt && !try_module_get(mm->binfmt->module))
872 goto free_pt;
873
874 return mm;
875
876free_pt:
877 /* don't put binfmt in mmput, we haven't got module yet */
878 mm->binfmt = NULL;
879 mmput(mm);
880
881fail_nomem:
882 return NULL;
883}
884
885static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
886{
887 struct mm_struct *mm, *oldmm;
888 int retval;
889
890 tsk->min_flt = tsk->maj_flt = 0;
891 tsk->nvcsw = tsk->nivcsw = 0;
892#ifdef CONFIG_DETECT_HUNG_TASK
893 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
894#endif
895
896 tsk->mm = NULL;
897 tsk->active_mm = NULL;
898
899 /*
900 * Are we cloning a kernel thread?
901 *
902 * We need to steal a active VM for that..
903 */
904 oldmm = current->mm;
905 if (!oldmm)
906 return 0;
907
908 /* initialize the new vmacache entries */
909 vmacache_flush(tsk);
910
911 if (clone_flags & CLONE_VM) {
912 atomic_inc(&oldmm->mm_users);
913 mm = oldmm;
914 goto good_mm;
915 }
916
917 retval = -ENOMEM;
918 mm = dup_mm(tsk);
919 if (!mm)
920 goto fail_nomem;
921
922good_mm:
923 tsk->mm = mm;
924 tsk->active_mm = mm;
925 return 0;
926
927fail_nomem:
928 return retval;
929}
930
931static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
932{
933 struct fs_struct *fs = current->fs;
934 if (clone_flags & CLONE_FS) {
935 /* tsk->fs is already what we want */
936 spin_lock(&fs->lock);
937 if (fs->in_exec) {
938 spin_unlock(&fs->lock);
939 return -EAGAIN;
940 }
941 fs->users++;
942 spin_unlock(&fs->lock);
943 return 0;
944 }
945 tsk->fs = copy_fs_struct(fs);
946 if (!tsk->fs)
947 return -ENOMEM;
948 return 0;
949}
950
951static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
952{
953 struct files_struct *oldf, *newf;
954 int error = 0;
955
956 /*
957 * A background process may not have any files ...
958 */
959 oldf = current->files;
960 if (!oldf)
961 goto out;
962
963 if (clone_flags & CLONE_FILES) {
964 atomic_inc(&oldf->count);
965 goto out;
966 }
967
968 newf = dup_fd(oldf, &error);
969 if (!newf)
970 goto out;
971
972 tsk->files = newf;
973 error = 0;
974out:
975 return error;
976}
977
978static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
979{
980#ifdef CONFIG_BLOCK
981 struct io_context *ioc = current->io_context;
982 struct io_context *new_ioc;
983
984 if (!ioc)
985 return 0;
986 /*
987 * Share io context with parent, if CLONE_IO is set
988 */
989 if (clone_flags & CLONE_IO) {
990 ioc_task_link(ioc);
991 tsk->io_context = ioc;
992 } else if (ioprio_valid(ioc->ioprio)) {
993 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
994 if (unlikely(!new_ioc))
995 return -ENOMEM;
996
997 new_ioc->ioprio = ioc->ioprio;
998 put_io_context(new_ioc);
999 }
1000#endif
1001 return 0;
1002}
1003
1004static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1005{
1006 struct sighand_struct *sig;
1007
1008 if (clone_flags & CLONE_SIGHAND) {
1009 atomic_inc(&current->sighand->count);
1010 return 0;
1011 }
1012 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1013 rcu_assign_pointer(tsk->sighand, sig);
1014 if (!sig)
1015 return -ENOMEM;
1016 atomic_set(&sig->count, 1);
1017 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1018 return 0;
1019}
1020
1021void __cleanup_sighand(struct sighand_struct *sighand)
1022{
1023 if (atomic_dec_and_test(&sighand->count)) {
1024 signalfd_cleanup(sighand);
1025 /*
1026 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1027 * without an RCU grace period, see __lock_task_sighand().
1028 */
1029 kmem_cache_free(sighand_cachep, sighand);
1030 }
1031}
1032
1033/*
1034 * Initialize POSIX timer handling for a thread group.
1035 */
1036static void posix_cpu_timers_init_group(struct signal_struct *sig)
1037{
1038 unsigned long cpu_limit;
1039
1040 /* Thread group counters. */
1041 thread_group_cputime_init(sig);
1042
1043 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1044 if (cpu_limit != RLIM_INFINITY) {
1045 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1046 sig->cputimer.running = 1;
1047 }
1048
1049 /* The timer lists. */
1050 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1051 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1052 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1053}
1054
1055static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1056{
1057 struct signal_struct *sig;
1058
1059 if (clone_flags & CLONE_THREAD)
1060 return 0;
1061
1062 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1063 tsk->signal = sig;
1064 if (!sig)
1065 return -ENOMEM;
1066
1067 sig->nr_threads = 1;
1068 atomic_set(&sig->live, 1);
1069 atomic_set(&sig->sigcnt, 1);
1070
1071 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1072 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1073 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1074
1075 init_waitqueue_head(&sig->wait_chldexit);
1076 sig->curr_target = tsk;
1077 init_sigpending(&sig->shared_pending);
1078 INIT_LIST_HEAD(&sig->posix_timers);
1079 seqlock_init(&sig->stats_lock);
1080
1081 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1082 sig->real_timer.function = it_real_fn;
1083
1084 task_lock(current->group_leader);
1085 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1086 task_unlock(current->group_leader);
1087
1088 posix_cpu_timers_init_group(sig);
1089
1090 tty_audit_fork(sig);
1091 sched_autogroup_fork(sig);
1092
1093#ifdef CONFIG_CGROUPS
1094 init_rwsem(&sig->group_rwsem);
1095#endif
1096
1097 sig->oom_score_adj = current->signal->oom_score_adj;
1098 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1099
1100 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1101 current->signal->is_child_subreaper;
1102
1103 mutex_init(&sig->cred_guard_mutex);
1104
1105 return 0;
1106}
1107
1108static void copy_seccomp(struct task_struct *p)
1109{
1110#ifdef CONFIG_SECCOMP
1111 /*
1112 * Must be called with sighand->lock held, which is common to
1113 * all threads in the group. Holding cred_guard_mutex is not
1114 * needed because this new task is not yet running and cannot
1115 * be racing exec.
1116 */
1117 assert_spin_locked(&current->sighand->siglock);
1118
1119 /* Ref-count the new filter user, and assign it. */
1120 get_seccomp_filter(current);
1121 p->seccomp = current->seccomp;
1122
1123 /*
1124 * Explicitly enable no_new_privs here in case it got set
1125 * between the task_struct being duplicated and holding the
1126 * sighand lock. The seccomp state and nnp must be in sync.
1127 */
1128 if (task_no_new_privs(current))
1129 task_set_no_new_privs(p);
1130
1131 /*
1132 * If the parent gained a seccomp mode after copying thread
1133 * flags and between before we held the sighand lock, we have
1134 * to manually enable the seccomp thread flag here.
1135 */
1136 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1137 set_tsk_thread_flag(p, TIF_SECCOMP);
1138#endif
1139}
1140
1141SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1142{
1143 current->clear_child_tid = tidptr;
1144
1145 return task_pid_vnr(current);
1146}
1147
1148static void rt_mutex_init_task(struct task_struct *p)
1149{
1150 raw_spin_lock_init(&p->pi_lock);
1151#ifdef CONFIG_RT_MUTEXES
1152 p->pi_waiters = RB_ROOT;
1153 p->pi_waiters_leftmost = NULL;
1154 p->pi_blocked_on = NULL;
1155#endif
1156}
1157
1158/*
1159 * Initialize POSIX timer handling for a single task.
1160 */
1161static void posix_cpu_timers_init(struct task_struct *tsk)
1162{
1163 tsk->cputime_expires.prof_exp = 0;
1164 tsk->cputime_expires.virt_exp = 0;
1165 tsk->cputime_expires.sched_exp = 0;
1166 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1167 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1168 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1169}
1170
1171static inline void
1172init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1173{
1174 task->pids[type].pid = pid;
1175}
1176
1177/*
1178 * This creates a new process as a copy of the old one,
1179 * but does not actually start it yet.
1180 *
1181 * It copies the registers, and all the appropriate
1182 * parts of the process environment (as per the clone
1183 * flags). The actual kick-off is left to the caller.
1184 */
1185static struct task_struct *copy_process(unsigned long clone_flags,
1186 unsigned long stack_start,
1187 unsigned long stack_size,
1188 int __user *child_tidptr,
1189 struct pid *pid,
1190 int trace)
1191{
1192 int retval;
1193 struct task_struct *p;
1194
1195 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1196 return ERR_PTR(-EINVAL);
1197
1198 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1199 return ERR_PTR(-EINVAL);
1200
1201 /*
1202 * Thread groups must share signals as well, and detached threads
1203 * can only be started up within the thread group.
1204 */
1205 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1206 return ERR_PTR(-EINVAL);
1207
1208 /*
1209 * Shared signal handlers imply shared VM. By way of the above,
1210 * thread groups also imply shared VM. Blocking this case allows
1211 * for various simplifications in other code.
1212 */
1213 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1214 return ERR_PTR(-EINVAL);
1215
1216 /*
1217 * Siblings of global init remain as zombies on exit since they are
1218 * not reaped by their parent (swapper). To solve this and to avoid
1219 * multi-rooted process trees, prevent global and container-inits
1220 * from creating siblings.
1221 */
1222 if ((clone_flags & CLONE_PARENT) &&
1223 current->signal->flags & SIGNAL_UNKILLABLE)
1224 return ERR_PTR(-EINVAL);
1225
1226 /*
1227 * If the new process will be in a different pid or user namespace
1228 * do not allow it to share a thread group or signal handlers or
1229 * parent with the forking task.
1230 */
1231 if (clone_flags & CLONE_SIGHAND) {
1232 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1233 (task_active_pid_ns(current) !=
1234 current->nsproxy->pid_ns_for_children))
1235 return ERR_PTR(-EINVAL);
1236 }
1237
1238 retval = security_task_create(clone_flags);
1239 if (retval)
1240 goto fork_out;
1241
1242 retval = -ENOMEM;
1243 p = dup_task_struct(current);
1244 if (!p)
1245 goto fork_out;
1246
1247 ftrace_graph_init_task(p);
1248
1249 rt_mutex_init_task(p);
1250
1251#ifdef CONFIG_PROVE_LOCKING
1252 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1253 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1254#endif
1255 retval = -EAGAIN;
1256 if (atomic_read(&p->real_cred->user->processes) >=
1257 task_rlimit(p, RLIMIT_NPROC)) {
1258 if (p->real_cred->user != INIT_USER &&
1259 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1260 goto bad_fork_free;
1261 }
1262 current->flags &= ~PF_NPROC_EXCEEDED;
1263
1264 retval = copy_creds(p, clone_flags);
1265 if (retval < 0)
1266 goto bad_fork_free;
1267
1268 /*
1269 * If multiple threads are within copy_process(), then this check
1270 * triggers too late. This doesn't hurt, the check is only there
1271 * to stop root fork bombs.
1272 */
1273 retval = -EAGAIN;
1274 if (nr_threads >= max_threads)
1275 goto bad_fork_cleanup_count;
1276
1277 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1278 goto bad_fork_cleanup_count;
1279
1280 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1281 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1282 p->flags |= PF_FORKNOEXEC;
1283 INIT_LIST_HEAD(&p->children);
1284 INIT_LIST_HEAD(&p->sibling);
1285 rcu_copy_process(p);
1286 p->vfork_done = NULL;
1287 spin_lock_init(&p->alloc_lock);
1288
1289 init_sigpending(&p->pending);
1290
1291 p->utime = p->stime = p->gtime = 0;
1292 p->utimescaled = p->stimescaled = 0;
1293#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1294 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1295#endif
1296#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1297 seqlock_init(&p->vtime_seqlock);
1298 p->vtime_snap = 0;
1299 p->vtime_snap_whence = VTIME_SLEEPING;
1300#endif
1301
1302#if defined(SPLIT_RSS_COUNTING)
1303 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1304#endif
1305
1306 p->default_timer_slack_ns = current->timer_slack_ns;
1307
1308 task_io_accounting_init(&p->ioac);
1309 acct_clear_integrals(p);
1310
1311 posix_cpu_timers_init(p);
1312
1313 p->start_time = ktime_get_ns();
1314 p->real_start_time = ktime_get_boot_ns();
1315 p->io_context = NULL;
1316 p->audit_context = NULL;
1317 if (clone_flags & CLONE_THREAD)
1318 threadgroup_change_begin(current);
1319 cgroup_fork(p);
1320#ifdef CONFIG_NUMA
1321 p->mempolicy = mpol_dup(p->mempolicy);
1322 if (IS_ERR(p->mempolicy)) {
1323 retval = PTR_ERR(p->mempolicy);
1324 p->mempolicy = NULL;
1325 goto bad_fork_cleanup_threadgroup_lock;
1326 }
1327#endif
1328#ifdef CONFIG_CPUSETS
1329 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1330 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1331 seqcount_init(&p->mems_allowed_seq);
1332#endif
1333#ifdef CONFIG_TRACE_IRQFLAGS
1334 p->irq_events = 0;
1335 p->hardirqs_enabled = 0;
1336 p->hardirq_enable_ip = 0;
1337 p->hardirq_enable_event = 0;
1338 p->hardirq_disable_ip = _THIS_IP_;
1339 p->hardirq_disable_event = 0;
1340 p->softirqs_enabled = 1;
1341 p->softirq_enable_ip = _THIS_IP_;
1342 p->softirq_enable_event = 0;
1343 p->softirq_disable_ip = 0;
1344 p->softirq_disable_event = 0;
1345 p->hardirq_context = 0;
1346 p->softirq_context = 0;
1347#endif
1348#ifdef CONFIG_LOCKDEP
1349 p->lockdep_depth = 0; /* no locks held yet */
1350 p->curr_chain_key = 0;
1351 p->lockdep_recursion = 0;
1352#endif
1353
1354#ifdef CONFIG_DEBUG_MUTEXES
1355 p->blocked_on = NULL; /* not blocked yet */
1356#endif
1357#ifdef CONFIG_BCACHE
1358 p->sequential_io = 0;
1359 p->sequential_io_avg = 0;
1360#endif
1361
1362 /* Perform scheduler related setup. Assign this task to a CPU. */
1363 retval = sched_fork(clone_flags, p);
1364 if (retval)
1365 goto bad_fork_cleanup_policy;
1366
1367 retval = perf_event_init_task(p);
1368 if (retval)
1369 goto bad_fork_cleanup_policy;
1370 retval = audit_alloc(p);
1371 if (retval)
1372 goto bad_fork_cleanup_perf;
1373 /* copy all the process information */
1374 shm_init_task(p);
1375 retval = copy_semundo(clone_flags, p);
1376 if (retval)
1377 goto bad_fork_cleanup_audit;
1378 retval = copy_files(clone_flags, p);
1379 if (retval)
1380 goto bad_fork_cleanup_semundo;
1381 retval = copy_fs(clone_flags, p);
1382 if (retval)
1383 goto bad_fork_cleanup_files;
1384 retval = copy_sighand(clone_flags, p);
1385 if (retval)
1386 goto bad_fork_cleanup_fs;
1387 retval = copy_signal(clone_flags, p);
1388 if (retval)
1389 goto bad_fork_cleanup_sighand;
1390 retval = copy_mm(clone_flags, p);
1391 if (retval)
1392 goto bad_fork_cleanup_signal;
1393 retval = copy_namespaces(clone_flags, p);
1394 if (retval)
1395 goto bad_fork_cleanup_mm;
1396 retval = copy_io(clone_flags, p);
1397 if (retval)
1398 goto bad_fork_cleanup_namespaces;
1399 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1400 if (retval)
1401 goto bad_fork_cleanup_io;
1402
1403 if (pid != &init_struct_pid) {
1404 retval = -ENOMEM;
1405 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1406 if (!pid)
1407 goto bad_fork_cleanup_io;
1408 }
1409
1410 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1411 /*
1412 * Clear TID on mm_release()?
1413 */
1414 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1415#ifdef CONFIG_BLOCK
1416 p->plug = NULL;
1417#endif
1418#ifdef CONFIG_FUTEX
1419 p->robust_list = NULL;
1420#ifdef CONFIG_COMPAT
1421 p->compat_robust_list = NULL;
1422#endif
1423 INIT_LIST_HEAD(&p->pi_state_list);
1424 p->pi_state_cache = NULL;
1425#endif
1426 /*
1427 * sigaltstack should be cleared when sharing the same VM
1428 */
1429 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1430 p->sas_ss_sp = p->sas_ss_size = 0;
1431
1432 /*
1433 * Syscall tracing and stepping should be turned off in the
1434 * child regardless of CLONE_PTRACE.
1435 */
1436 user_disable_single_step(p);
1437 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1438#ifdef TIF_SYSCALL_EMU
1439 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1440#endif
1441 clear_all_latency_tracing(p);
1442
1443 /* ok, now we should be set up.. */
1444 p->pid = pid_nr(pid);
1445 if (clone_flags & CLONE_THREAD) {
1446 p->exit_signal = -1;
1447 p->group_leader = current->group_leader;
1448 p->tgid = current->tgid;
1449 } else {
1450 if (clone_flags & CLONE_PARENT)
1451 p->exit_signal = current->group_leader->exit_signal;
1452 else
1453 p->exit_signal = (clone_flags & CSIGNAL);
1454 p->group_leader = p;
1455 p->tgid = p->pid;
1456 }
1457
1458 p->nr_dirtied = 0;
1459 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1460 p->dirty_paused_when = 0;
1461
1462 p->pdeath_signal = 0;
1463 INIT_LIST_HEAD(&p->thread_group);
1464 p->task_works = NULL;
1465
1466 /*
1467 * Make it visible to the rest of the system, but dont wake it up yet.
1468 * Need tasklist lock for parent etc handling!
1469 */
1470 write_lock_irq(&tasklist_lock);
1471
1472 /* CLONE_PARENT re-uses the old parent */
1473 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1474 p->real_parent = current->real_parent;
1475 p->parent_exec_id = current->parent_exec_id;
1476 } else {
1477 p->real_parent = current;
1478 p->parent_exec_id = current->self_exec_id;
1479 }
1480
1481 spin_lock(&current->sighand->siglock);
1482
1483 /*
1484 * Copy seccomp details explicitly here, in case they were changed
1485 * before holding sighand lock.
1486 */
1487 copy_seccomp(p);
1488
1489 /*
1490 * Process group and session signals need to be delivered to just the
1491 * parent before the fork or both the parent and the child after the
1492 * fork. Restart if a signal comes in before we add the new process to
1493 * it's process group.
1494 * A fatal signal pending means that current will exit, so the new
1495 * thread can't slip out of an OOM kill (or normal SIGKILL).
1496 */
1497 recalc_sigpending();
1498 if (signal_pending(current)) {
1499 spin_unlock(&current->sighand->siglock);
1500 write_unlock_irq(&tasklist_lock);
1501 retval = -ERESTARTNOINTR;
1502 goto bad_fork_free_pid;
1503 }
1504
1505 if (likely(p->pid)) {
1506 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1507
1508 init_task_pid(p, PIDTYPE_PID, pid);
1509 if (thread_group_leader(p)) {
1510 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1511 init_task_pid(p, PIDTYPE_SID, task_session(current));
1512
1513 if (is_child_reaper(pid)) {
1514 ns_of_pid(pid)->child_reaper = p;
1515 p->signal->flags |= SIGNAL_UNKILLABLE;
1516 }
1517
1518 p->signal->leader_pid = pid;
1519 p->signal->tty = tty_kref_get(current->signal->tty);
1520 list_add_tail(&p->sibling, &p->real_parent->children);
1521 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1522 attach_pid(p, PIDTYPE_PGID);
1523 attach_pid(p, PIDTYPE_SID);
1524 __this_cpu_inc(process_counts);
1525 } else {
1526 current->signal->nr_threads++;
1527 atomic_inc(&current->signal->live);
1528 atomic_inc(&current->signal->sigcnt);
1529 list_add_tail_rcu(&p->thread_group,
1530 &p->group_leader->thread_group);
1531 list_add_tail_rcu(&p->thread_node,
1532 &p->signal->thread_head);
1533 }
1534 attach_pid(p, PIDTYPE_PID);
1535 nr_threads++;
1536 }
1537
1538 total_forks++;
1539 spin_unlock(&current->sighand->siglock);
1540 syscall_tracepoint_update(p);
1541 write_unlock_irq(&tasklist_lock);
1542
1543 proc_fork_connector(p);
1544 cgroup_post_fork(p);
1545 if (clone_flags & CLONE_THREAD)
1546 threadgroup_change_end(current);
1547 perf_event_fork(p);
1548
1549 trace_task_newtask(p, clone_flags);
1550 uprobe_copy_process(p, clone_flags);
1551
1552 return p;
1553
1554bad_fork_free_pid:
1555 if (pid != &init_struct_pid)
1556 free_pid(pid);
1557bad_fork_cleanup_io:
1558 if (p->io_context)
1559 exit_io_context(p);
1560bad_fork_cleanup_namespaces:
1561 exit_task_namespaces(p);
1562bad_fork_cleanup_mm:
1563 if (p->mm)
1564 mmput(p->mm);
1565bad_fork_cleanup_signal:
1566 if (!(clone_flags & CLONE_THREAD))
1567 free_signal_struct(p->signal);
1568bad_fork_cleanup_sighand:
1569 __cleanup_sighand(p->sighand);
1570bad_fork_cleanup_fs:
1571 exit_fs(p); /* blocking */
1572bad_fork_cleanup_files:
1573 exit_files(p); /* blocking */
1574bad_fork_cleanup_semundo:
1575 exit_sem(p);
1576bad_fork_cleanup_audit:
1577 audit_free(p);
1578bad_fork_cleanup_perf:
1579 perf_event_free_task(p);
1580bad_fork_cleanup_policy:
1581#ifdef CONFIG_NUMA
1582 mpol_put(p->mempolicy);
1583bad_fork_cleanup_threadgroup_lock:
1584#endif
1585 if (clone_flags & CLONE_THREAD)
1586 threadgroup_change_end(current);
1587 delayacct_tsk_free(p);
1588 module_put(task_thread_info(p)->exec_domain->module);
1589bad_fork_cleanup_count:
1590 atomic_dec(&p->cred->user->processes);
1591 exit_creds(p);
1592bad_fork_free:
1593 free_task(p);
1594fork_out:
1595 return ERR_PTR(retval);
1596}
1597
1598static inline void init_idle_pids(struct pid_link *links)
1599{
1600 enum pid_type type;
1601
1602 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1603 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1604 links[type].pid = &init_struct_pid;
1605 }
1606}
1607
1608struct task_struct *fork_idle(int cpu)
1609{
1610 struct task_struct *task;
1611 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1612 if (!IS_ERR(task)) {
1613 init_idle_pids(task->pids);
1614 init_idle(task, cpu);
1615 }
1616
1617 return task;
1618}
1619
1620/*
1621 * Ok, this is the main fork-routine.
1622 *
1623 * It copies the process, and if successful kick-starts
1624 * it and waits for it to finish using the VM if required.
1625 */
1626long do_fork(unsigned long clone_flags,
1627 unsigned long stack_start,
1628 unsigned long stack_size,
1629 int __user *parent_tidptr,
1630 int __user *child_tidptr)
1631{
1632 struct task_struct *p;
1633 int trace = 0;
1634 long nr;
1635
1636 /*
1637 * Determine whether and which event to report to ptracer. When
1638 * called from kernel_thread or CLONE_UNTRACED is explicitly
1639 * requested, no event is reported; otherwise, report if the event
1640 * for the type of forking is enabled.
1641 */
1642 if (!(clone_flags & CLONE_UNTRACED)) {
1643 if (clone_flags & CLONE_VFORK)
1644 trace = PTRACE_EVENT_VFORK;
1645 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1646 trace = PTRACE_EVENT_CLONE;
1647 else
1648 trace = PTRACE_EVENT_FORK;
1649
1650 if (likely(!ptrace_event_enabled(current, trace)))
1651 trace = 0;
1652 }
1653
1654 p = copy_process(clone_flags, stack_start, stack_size,
1655 child_tidptr, NULL, trace);
1656 /*
1657 * Do this prior waking up the new thread - the thread pointer
1658 * might get invalid after that point, if the thread exits quickly.
1659 */
1660 if (!IS_ERR(p)) {
1661 struct completion vfork;
1662 struct pid *pid;
1663
1664 trace_sched_process_fork(current, p);
1665
1666 pid = get_task_pid(p, PIDTYPE_PID);
1667 nr = pid_vnr(pid);
1668
1669 if (clone_flags & CLONE_PARENT_SETTID)
1670 put_user(nr, parent_tidptr);
1671
1672 if (clone_flags & CLONE_VFORK) {
1673 p->vfork_done = &vfork;
1674 init_completion(&vfork);
1675 get_task_struct(p);
1676 }
1677
1678 wake_up_new_task(p);
1679
1680 /* forking complete and child started to run, tell ptracer */
1681 if (unlikely(trace))
1682 ptrace_event_pid(trace, pid);
1683
1684 if (clone_flags & CLONE_VFORK) {
1685 if (!wait_for_vfork_done(p, &vfork))
1686 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1687 }
1688
1689 put_pid(pid);
1690 } else {
1691 nr = PTR_ERR(p);
1692 }
1693 return nr;
1694}
1695
1696/*
1697 * Create a kernel thread.
1698 */
1699pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1700{
1701 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1702 (unsigned long)arg, NULL, NULL);
1703}
1704
1705#ifdef __ARCH_WANT_SYS_FORK
1706SYSCALL_DEFINE0(fork)
1707{
1708#ifdef CONFIG_MMU
1709 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1710#else
1711 /* can not support in nommu mode */
1712 return -EINVAL;
1713#endif
1714}
1715#endif
1716
1717#ifdef __ARCH_WANT_SYS_VFORK
1718SYSCALL_DEFINE0(vfork)
1719{
1720 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1721 0, NULL, NULL);
1722}
1723#endif
1724
1725#ifdef __ARCH_WANT_SYS_CLONE
1726#ifdef CONFIG_CLONE_BACKWARDS
1727SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1728 int __user *, parent_tidptr,
1729 int, tls_val,
1730 int __user *, child_tidptr)
1731#elif defined(CONFIG_CLONE_BACKWARDS2)
1732SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1733 int __user *, parent_tidptr,
1734 int __user *, child_tidptr,
1735 int, tls_val)
1736#elif defined(CONFIG_CLONE_BACKWARDS3)
1737SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1738 int, stack_size,
1739 int __user *, parent_tidptr,
1740 int __user *, child_tidptr,
1741 int, tls_val)
1742#else
1743SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1744 int __user *, parent_tidptr,
1745 int __user *, child_tidptr,
1746 int, tls_val)
1747#endif
1748{
1749 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1750}
1751#endif
1752
1753#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1754#define ARCH_MIN_MMSTRUCT_ALIGN 0
1755#endif
1756
1757static void sighand_ctor(void *data)
1758{
1759 struct sighand_struct *sighand = data;
1760
1761 spin_lock_init(&sighand->siglock);
1762 init_waitqueue_head(&sighand->signalfd_wqh);
1763}
1764
1765void __init proc_caches_init(void)
1766{
1767 sighand_cachep = kmem_cache_create("sighand_cache",
1768 sizeof(struct sighand_struct), 0,
1769 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1770 SLAB_NOTRACK, sighand_ctor);
1771 signal_cachep = kmem_cache_create("signal_cache",
1772 sizeof(struct signal_struct), 0,
1773 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1774 files_cachep = kmem_cache_create("files_cache",
1775 sizeof(struct files_struct), 0,
1776 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1777 fs_cachep = kmem_cache_create("fs_cache",
1778 sizeof(struct fs_struct), 0,
1779 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1780 /*
1781 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1782 * whole struct cpumask for the OFFSTACK case. We could change
1783 * this to *only* allocate as much of it as required by the
1784 * maximum number of CPU's we can ever have. The cpumask_allocation
1785 * is at the end of the structure, exactly for that reason.
1786 */
1787 mm_cachep = kmem_cache_create("mm_struct",
1788 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1789 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1790 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1791 mmap_init();
1792 nsproxy_cache_init();
1793}
1794
1795/*
1796 * Check constraints on flags passed to the unshare system call.
1797 */
1798static int check_unshare_flags(unsigned long unshare_flags)
1799{
1800 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1801 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1802 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1803 CLONE_NEWUSER|CLONE_NEWPID))
1804 return -EINVAL;
1805 /*
1806 * Not implemented, but pretend it works if there is nothing to
1807 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1808 * needs to unshare vm.
1809 */
1810 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1811 /* FIXME: get_task_mm() increments ->mm_users */
1812 if (atomic_read(&current->mm->mm_users) > 1)
1813 return -EINVAL;
1814 }
1815
1816 return 0;
1817}
1818
1819/*
1820 * Unshare the filesystem structure if it is being shared
1821 */
1822static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1823{
1824 struct fs_struct *fs = current->fs;
1825
1826 if (!(unshare_flags & CLONE_FS) || !fs)
1827 return 0;
1828
1829 /* don't need lock here; in the worst case we'll do useless copy */
1830 if (fs->users == 1)
1831 return 0;
1832
1833 *new_fsp = copy_fs_struct(fs);
1834 if (!*new_fsp)
1835 return -ENOMEM;
1836
1837 return 0;
1838}
1839
1840/*
1841 * Unshare file descriptor table if it is being shared
1842 */
1843static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1844{
1845 struct files_struct *fd = current->files;
1846 int error = 0;
1847
1848 if ((unshare_flags & CLONE_FILES) &&
1849 (fd && atomic_read(&fd->count) > 1)) {
1850 *new_fdp = dup_fd(fd, &error);
1851 if (!*new_fdp)
1852 return error;
1853 }
1854
1855 return 0;
1856}
1857
1858/*
1859 * unshare allows a process to 'unshare' part of the process
1860 * context which was originally shared using clone. copy_*
1861 * functions used by do_fork() cannot be used here directly
1862 * because they modify an inactive task_struct that is being
1863 * constructed. Here we are modifying the current, active,
1864 * task_struct.
1865 */
1866SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1867{
1868 struct fs_struct *fs, *new_fs = NULL;
1869 struct files_struct *fd, *new_fd = NULL;
1870 struct cred *new_cred = NULL;
1871 struct nsproxy *new_nsproxy = NULL;
1872 int do_sysvsem = 0;
1873 int err;
1874
1875 /*
1876 * If unsharing a user namespace must also unshare the thread.
1877 */
1878 if (unshare_flags & CLONE_NEWUSER)
1879 unshare_flags |= CLONE_THREAD | CLONE_FS;
1880 /*
1881 * If unsharing a thread from a thread group, must also unshare vm.
1882 */
1883 if (unshare_flags & CLONE_THREAD)
1884 unshare_flags |= CLONE_VM;
1885 /*
1886 * If unsharing vm, must also unshare signal handlers.
1887 */
1888 if (unshare_flags & CLONE_VM)
1889 unshare_flags |= CLONE_SIGHAND;
1890 /*
1891 * If unsharing namespace, must also unshare filesystem information.
1892 */
1893 if (unshare_flags & CLONE_NEWNS)
1894 unshare_flags |= CLONE_FS;
1895
1896 err = check_unshare_flags(unshare_flags);
1897 if (err)
1898 goto bad_unshare_out;
1899 /*
1900 * CLONE_NEWIPC must also detach from the undolist: after switching
1901 * to a new ipc namespace, the semaphore arrays from the old
1902 * namespace are unreachable.
1903 */
1904 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1905 do_sysvsem = 1;
1906 err = unshare_fs(unshare_flags, &new_fs);
1907 if (err)
1908 goto bad_unshare_out;
1909 err = unshare_fd(unshare_flags, &new_fd);
1910 if (err)
1911 goto bad_unshare_cleanup_fs;
1912 err = unshare_userns(unshare_flags, &new_cred);
1913 if (err)
1914 goto bad_unshare_cleanup_fd;
1915 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1916 new_cred, new_fs);
1917 if (err)
1918 goto bad_unshare_cleanup_cred;
1919
1920 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1921 if (do_sysvsem) {
1922 /*
1923 * CLONE_SYSVSEM is equivalent to sys_exit().
1924 */
1925 exit_sem(current);
1926 }
1927 if (unshare_flags & CLONE_NEWIPC) {
1928 /* Orphan segments in old ns (see sem above). */
1929 exit_shm(current);
1930 shm_init_task(current);
1931 }
1932
1933 if (new_nsproxy)
1934 switch_task_namespaces(current, new_nsproxy);
1935
1936 task_lock(current);
1937
1938 if (new_fs) {
1939 fs = current->fs;
1940 spin_lock(&fs->lock);
1941 current->fs = new_fs;
1942 if (--fs->users)
1943 new_fs = NULL;
1944 else
1945 new_fs = fs;
1946 spin_unlock(&fs->lock);
1947 }
1948
1949 if (new_fd) {
1950 fd = current->files;
1951 current->files = new_fd;
1952 new_fd = fd;
1953 }
1954
1955 task_unlock(current);
1956
1957 if (new_cred) {
1958 /* Install the new user namespace */
1959 commit_creds(new_cred);
1960 new_cred = NULL;
1961 }
1962 }
1963
1964bad_unshare_cleanup_cred:
1965 if (new_cred)
1966 put_cred(new_cred);
1967bad_unshare_cleanup_fd:
1968 if (new_fd)
1969 put_files_struct(new_fd);
1970
1971bad_unshare_cleanup_fs:
1972 if (new_fs)
1973 free_fs_struct(new_fs);
1974
1975bad_unshare_out:
1976 return err;
1977}
1978
1979/*
1980 * Helper to unshare the files of the current task.
1981 * We don't want to expose copy_files internals to
1982 * the exec layer of the kernel.
1983 */
1984
1985int unshare_files(struct files_struct **displaced)
1986{
1987 struct task_struct *task = current;
1988 struct files_struct *copy = NULL;
1989 int error;
1990
1991 error = unshare_fd(CLONE_FILES, &copy);
1992 if (error || !copy) {
1993 *displaced = NULL;
1994 return error;
1995 }
1996 *displaced = task->files;
1997 task_lock(task);
1998 task->files = copy;
1999 task_unlock(task);
2000 return 0;
2001}