4 * Copyright (C) 1991, 1992 Linus Torvalds
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()'
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>
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 #include <linux/sysctl.h>
79 #include <asm/pgtable.h>
80 #include <asm/pgalloc.h>
81 #include <asm/uaccess.h>
82 #include <asm/mmu_context.h>
83 #include <asm/cacheflush.h>
84 #include <asm/tlbflush.h>
86 #include <trace/events/sched.h>
88 #define CREATE_TRACE_POINTS
89 #include <trace/events/task.h>
92 * Minimum number of threads to boot the kernel
94 #define MIN_THREADS 20
97 * Maximum number of threads
99 #define MAX_THREADS FUTEX_TID_MASK
102 * Protected counters by write_lock_irq(&tasklist_lock)
104 unsigned long total_forks; /* Handle normal Linux uptimes. */
105 int nr_threads; /* The idle threads do not count.. */
107 int max_threads; /* tunable limit on nr_threads */
109 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
111 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
113 #ifdef CONFIG_PROVE_RCU
114 int lockdep_tasklist_lock_is_held(void)
116 return lockdep_is_held(&tasklist_lock);
118 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
119 #endif /* #ifdef CONFIG_PROVE_RCU */
121 int nr_processes(void)
126 for_each_possible_cpu(cpu)
127 total += per_cpu(process_counts, cpu);
132 void __weak arch_release_task_struct(struct task_struct *tsk)
136 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
137 static struct kmem_cache *task_struct_cachep;
139 static inline struct task_struct *alloc_task_struct_node(int node)
141 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
144 static inline void free_task_struct(struct task_struct *tsk)
146 kmem_cache_free(task_struct_cachep, tsk);
150 void __weak arch_release_thread_info(struct thread_info *ti)
154 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
157 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
158 * kmemcache based allocator.
160 # if THREAD_SIZE >= PAGE_SIZE
161 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
164 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
167 return page ? page_address(page) : NULL;
170 static inline void free_thread_info(struct thread_info *ti)
172 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
175 static struct kmem_cache *thread_info_cache;
177 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
180 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
183 static void free_thread_info(struct thread_info *ti)
185 kmem_cache_free(thread_info_cache, ti);
188 void thread_info_cache_init(void)
190 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
191 THREAD_SIZE, 0, NULL);
192 BUG_ON(thread_info_cache == NULL);
197 /* SLAB cache for signal_struct structures (tsk->signal) */
198 static struct kmem_cache *signal_cachep;
200 /* SLAB cache for sighand_struct structures (tsk->sighand) */
201 struct kmem_cache *sighand_cachep;
203 /* SLAB cache for files_struct structures (tsk->files) */
204 struct kmem_cache *files_cachep;
206 /* SLAB cache for fs_struct structures (tsk->fs) */
207 struct kmem_cache *fs_cachep;
209 /* SLAB cache for vm_area_struct structures */
210 struct kmem_cache *vm_area_cachep;
212 /* SLAB cache for mm_struct structures (tsk->mm) */
213 static struct kmem_cache *mm_cachep;
215 static void account_kernel_stack(struct thread_info *ti, int account)
217 struct zone *zone = page_zone(virt_to_page(ti));
219 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
222 void free_task(struct task_struct *tsk)
224 account_kernel_stack(tsk->stack, -1);
225 arch_release_thread_info(tsk->stack);
226 free_thread_info(tsk->stack);
227 rt_mutex_debug_task_free(tsk);
228 ftrace_graph_exit_task(tsk);
229 put_seccomp_filter(tsk);
230 arch_release_task_struct(tsk);
231 free_task_struct(tsk);
233 EXPORT_SYMBOL(free_task);
235 static inline void free_signal_struct(struct signal_struct *sig)
237 taskstats_tgid_free(sig);
238 sched_autogroup_exit(sig);
239 kmem_cache_free(signal_cachep, sig);
242 static inline void put_signal_struct(struct signal_struct *sig)
244 if (atomic_dec_and_test(&sig->sigcnt))
245 free_signal_struct(sig);
248 void __put_task_struct(struct task_struct *tsk)
250 WARN_ON(!tsk->exit_state);
251 WARN_ON(atomic_read(&tsk->usage));
252 WARN_ON(tsk == current);
256 security_task_free(tsk);
258 delayacct_tsk_free(tsk);
259 put_signal_struct(tsk->signal);
261 if (!profile_handoff_task(tsk))
264 EXPORT_SYMBOL_GPL(__put_task_struct);
266 void __init __weak arch_task_cache_init(void) { }
271 static void set_max_threads(unsigned int max_threads_suggested)
276 * The number of threads shall be limited such that the thread
277 * structures may only consume a small part of the available memory.
279 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
280 threads = MAX_THREADS;
282 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
283 (u64) THREAD_SIZE * 8UL);
285 if (threads > max_threads_suggested)
286 threads = max_threads_suggested;
288 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
291 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
292 /* Initialized by the architecture: */
293 int arch_task_struct_size __read_mostly;
296 void __init fork_init(void)
298 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
299 #ifndef ARCH_MIN_TASKALIGN
300 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
302 /* create a slab on which task_structs can be allocated */
304 kmem_cache_create("task_struct", arch_task_struct_size,
305 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
308 /* do the arch specific task caches init */
309 arch_task_cache_init();
311 set_max_threads(MAX_THREADS);
313 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
314 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
315 init_task.signal->rlim[RLIMIT_SIGPENDING] =
316 init_task.signal->rlim[RLIMIT_NPROC];
319 int __weak arch_dup_task_struct(struct task_struct *dst,
320 struct task_struct *src)
326 void set_task_stack_end_magic(struct task_struct *tsk)
328 unsigned long *stackend;
330 stackend = end_of_stack(tsk);
331 *stackend = STACK_END_MAGIC; /* for overflow detection */
334 static struct task_struct *dup_task_struct(struct task_struct *orig)
336 struct task_struct *tsk;
337 struct thread_info *ti;
338 int node = tsk_fork_get_node(orig);
341 tsk = alloc_task_struct_node(node);
345 ti = alloc_thread_info_node(tsk, node);
349 err = arch_dup_task_struct(tsk, orig);
354 #ifdef CONFIG_SECCOMP
356 * We must handle setting up seccomp filters once we're under
357 * the sighand lock in case orig has changed between now and
358 * then. Until then, filter must be NULL to avoid messing up
359 * the usage counts on the error path calling free_task.
361 tsk->seccomp.filter = NULL;
364 setup_thread_stack(tsk, orig);
365 clear_user_return_notifier(tsk);
366 clear_tsk_need_resched(tsk);
367 set_task_stack_end_magic(tsk);
369 #ifdef CONFIG_CC_STACKPROTECTOR
370 tsk->stack_canary = get_random_int();
374 * One for us, one for whoever does the "release_task()" (usually
377 atomic_set(&tsk->usage, 2);
378 #ifdef CONFIG_BLK_DEV_IO_TRACE
381 tsk->splice_pipe = NULL;
382 tsk->task_frag.page = NULL;
383 tsk->wake_q.next = NULL;
385 account_kernel_stack(ti, 1);
390 free_thread_info(ti);
392 free_task_struct(tsk);
397 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
399 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
400 struct rb_node **rb_link, *rb_parent;
402 unsigned long charge;
404 uprobe_start_dup_mmap();
405 down_write(&oldmm->mmap_sem);
406 flush_cache_dup_mm(oldmm);
407 uprobe_dup_mmap(oldmm, mm);
409 * Not linked in yet - no deadlock potential:
411 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
413 /* No ordering required: file already has been exposed. */
414 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
416 mm->total_vm = oldmm->total_vm;
417 mm->shared_vm = oldmm->shared_vm;
418 mm->exec_vm = oldmm->exec_vm;
419 mm->stack_vm = oldmm->stack_vm;
421 rb_link = &mm->mm_rb.rb_node;
424 retval = ksm_fork(mm, oldmm);
427 retval = khugepaged_fork(mm, oldmm);
432 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
435 if (mpnt->vm_flags & VM_DONTCOPY) {
436 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
441 if (mpnt->vm_flags & VM_ACCOUNT) {
442 unsigned long len = vma_pages(mpnt);
444 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
448 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
452 INIT_LIST_HEAD(&tmp->anon_vma_chain);
453 retval = vma_dup_policy(mpnt, tmp);
455 goto fail_nomem_policy;
457 if (anon_vma_fork(tmp, mpnt))
458 goto fail_nomem_anon_vma_fork;
460 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
461 tmp->vm_next = tmp->vm_prev = NULL;
462 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
465 struct inode *inode = file_inode(file);
466 struct address_space *mapping = file->f_mapping;
469 if (tmp->vm_flags & VM_DENYWRITE)
470 atomic_dec(&inode->i_writecount);
471 i_mmap_lock_write(mapping);
472 if (tmp->vm_flags & VM_SHARED)
473 atomic_inc(&mapping->i_mmap_writable);
474 flush_dcache_mmap_lock(mapping);
475 /* insert tmp into the share list, just after mpnt */
476 vma_interval_tree_insert_after(tmp, mpnt,
478 flush_dcache_mmap_unlock(mapping);
479 i_mmap_unlock_write(mapping);
483 * Clear hugetlb-related page reserves for children. This only
484 * affects MAP_PRIVATE mappings. Faults generated by the child
485 * are not guaranteed to succeed, even if read-only
487 if (is_vm_hugetlb_page(tmp))
488 reset_vma_resv_huge_pages(tmp);
491 * Link in the new vma and copy the page table entries.
494 pprev = &tmp->vm_next;
498 __vma_link_rb(mm, tmp, rb_link, rb_parent);
499 rb_link = &tmp->vm_rb.rb_right;
500 rb_parent = &tmp->vm_rb;
503 retval = copy_page_range(mm, oldmm, mpnt);
505 if (tmp->vm_ops && tmp->vm_ops->open)
506 tmp->vm_ops->open(tmp);
511 /* a new mm has just been created */
512 arch_dup_mmap(oldmm, mm);
515 up_write(&mm->mmap_sem);
517 up_write(&oldmm->mmap_sem);
518 uprobe_end_dup_mmap();
520 fail_nomem_anon_vma_fork:
521 mpol_put(vma_policy(tmp));
523 kmem_cache_free(vm_area_cachep, tmp);
526 vm_unacct_memory(charge);
530 static inline int mm_alloc_pgd(struct mm_struct *mm)
532 mm->pgd = pgd_alloc(mm);
533 if (unlikely(!mm->pgd))
538 static inline void mm_free_pgd(struct mm_struct *mm)
540 pgd_free(mm, mm->pgd);
543 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
545 down_write(&oldmm->mmap_sem);
546 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
547 up_write(&oldmm->mmap_sem);
550 #define mm_alloc_pgd(mm) (0)
551 #define mm_free_pgd(mm)
552 #endif /* CONFIG_MMU */
554 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
556 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
557 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
559 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
561 static int __init coredump_filter_setup(char *s)
563 default_dump_filter =
564 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
565 MMF_DUMP_FILTER_MASK;
569 __setup("coredump_filter=", coredump_filter_setup);
571 #include <linux/init_task.h>
573 static void mm_init_aio(struct mm_struct *mm)
576 spin_lock_init(&mm->ioctx_lock);
577 mm->ioctx_table = NULL;
581 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
588 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
592 mm->vmacache_seqnum = 0;
593 atomic_set(&mm->mm_users, 1);
594 atomic_set(&mm->mm_count, 1);
595 init_rwsem(&mm->mmap_sem);
596 INIT_LIST_HEAD(&mm->mmlist);
597 mm->core_state = NULL;
598 atomic_long_set(&mm->nr_ptes, 0);
603 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
604 spin_lock_init(&mm->page_table_lock);
607 mm_init_owner(mm, p);
608 mmu_notifier_mm_init(mm);
609 clear_tlb_flush_pending(mm);
610 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
611 mm->pmd_huge_pte = NULL;
615 mm->flags = current->mm->flags & MMF_INIT_MASK;
616 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
618 mm->flags = default_dump_filter;
622 if (mm_alloc_pgd(mm))
625 if (init_new_context(p, mm))
637 static void check_mm(struct mm_struct *mm)
641 for (i = 0; i < NR_MM_COUNTERS; i++) {
642 long x = atomic_long_read(&mm->rss_stat.count[i]);
645 printk(KERN_ALERT "BUG: Bad rss-counter state "
646 "mm:%p idx:%d val:%ld\n", mm, i, x);
649 if (atomic_long_read(&mm->nr_ptes))
650 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
651 atomic_long_read(&mm->nr_ptes));
653 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
656 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
657 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
662 * Allocate and initialize an mm_struct.
664 struct mm_struct *mm_alloc(void)
666 struct mm_struct *mm;
672 memset(mm, 0, sizeof(*mm));
673 return mm_init(mm, current);
677 * Called when the last reference to the mm
678 * is dropped: either by a lazy thread or by
679 * mmput. Free the page directory and the mm.
681 void __mmdrop(struct mm_struct *mm)
683 BUG_ON(mm == &init_mm);
686 mmu_notifier_mm_destroy(mm);
690 EXPORT_SYMBOL_GPL(__mmdrop);
693 * Decrement the use count and release all resources for an mm.
695 void mmput(struct mm_struct *mm)
699 if (atomic_dec_and_test(&mm->mm_users)) {
700 uprobe_clear_state(mm);
703 khugepaged_exit(mm); /* must run before exit_mmap */
705 set_mm_exe_file(mm, NULL);
706 if (!list_empty(&mm->mmlist)) {
707 spin_lock(&mmlist_lock);
708 list_del(&mm->mmlist);
709 spin_unlock(&mmlist_lock);
712 module_put(mm->binfmt->module);
716 EXPORT_SYMBOL_GPL(mmput);
719 * set_mm_exe_file - change a reference to the mm's executable file
721 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
723 * Main users are mmput() and sys_execve(). Callers prevent concurrent
724 * invocations: in mmput() nobody alive left, in execve task is single
725 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
726 * mm->exe_file, but does so without using set_mm_exe_file() in order
727 * to do avoid the need for any locks.
729 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
731 struct file *old_exe_file;
734 * It is safe to dereference the exe_file without RCU as
735 * this function is only called if nobody else can access
736 * this mm -- see comment above for justification.
738 old_exe_file = rcu_dereference_raw(mm->exe_file);
741 get_file(new_exe_file);
742 rcu_assign_pointer(mm->exe_file, new_exe_file);
748 * get_mm_exe_file - acquire a reference to the mm's executable file
750 * Returns %NULL if mm has no associated executable file.
751 * User must release file via fput().
753 struct file *get_mm_exe_file(struct mm_struct *mm)
755 struct file *exe_file;
758 exe_file = rcu_dereference(mm->exe_file);
759 if (exe_file && !get_file_rcu(exe_file))
764 EXPORT_SYMBOL(get_mm_exe_file);
767 * get_task_exe_file - acquire a reference to the task's executable file
769 * Returns %NULL if task's mm (if any) has no associated executable file or
770 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
771 * User must release file via fput().
773 struct file *get_task_exe_file(struct task_struct *task)
775 struct file *exe_file = NULL;
776 struct mm_struct *mm;
781 if (!(task->flags & PF_KTHREAD))
782 exe_file = get_mm_exe_file(mm);
787 EXPORT_SYMBOL(get_task_exe_file);
790 * get_task_mm - acquire a reference to the task's mm
792 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
793 * this kernel workthread has transiently adopted a user mm with use_mm,
794 * to do its AIO) is not set and if so returns a reference to it, after
795 * bumping up the use count. User must release the mm via mmput()
796 * after use. Typically used by /proc and ptrace.
798 struct mm_struct *get_task_mm(struct task_struct *task)
800 struct mm_struct *mm;
805 if (task->flags & PF_KTHREAD)
808 atomic_inc(&mm->mm_users);
813 EXPORT_SYMBOL_GPL(get_task_mm);
815 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
817 struct mm_struct *mm;
820 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
824 mm = get_task_mm(task);
825 if (mm && mm != current->mm &&
826 !ptrace_may_access(task, mode) &&
827 !capable(CAP_SYS_RESOURCE)) {
829 mm = ERR_PTR(-EACCES);
831 mutex_unlock(&task->signal->cred_guard_mutex);
836 static void complete_vfork_done(struct task_struct *tsk)
838 struct completion *vfork;
841 vfork = tsk->vfork_done;
843 tsk->vfork_done = NULL;
849 static int wait_for_vfork_done(struct task_struct *child,
850 struct completion *vfork)
854 freezer_do_not_count();
855 killed = wait_for_completion_killable(vfork);
860 child->vfork_done = NULL;
864 put_task_struct(child);
868 /* Please note the differences between mmput and mm_release.
869 * mmput is called whenever we stop holding onto a mm_struct,
870 * error success whatever.
872 * mm_release is called after a mm_struct has been removed
873 * from the current process.
875 * This difference is important for error handling, when we
876 * only half set up a mm_struct for a new process and need to restore
877 * the old one. Because we mmput the new mm_struct before
878 * restoring the old one. . .
879 * Eric Biederman 10 January 1998
881 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
883 /* Get rid of any futexes when releasing the mm */
885 if (unlikely(tsk->robust_list)) {
886 exit_robust_list(tsk);
887 tsk->robust_list = NULL;
890 if (unlikely(tsk->compat_robust_list)) {
891 compat_exit_robust_list(tsk);
892 tsk->compat_robust_list = NULL;
895 if (unlikely(!list_empty(&tsk->pi_state_list)))
896 exit_pi_state_list(tsk);
899 uprobe_free_utask(tsk);
901 /* Get rid of any cached register state */
902 deactivate_mm(tsk, mm);
905 * Signal userspace if we're not exiting with a core dump
906 * because we want to leave the value intact for debugging
909 if (tsk->clear_child_tid) {
910 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
911 atomic_read(&mm->mm_users) > 1) {
913 * We don't check the error code - if userspace has
914 * not set up a proper pointer then tough luck.
916 put_user(0, tsk->clear_child_tid);
917 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
920 tsk->clear_child_tid = NULL;
924 * All done, finally we can wake up parent and return this mm to him.
925 * Also kthread_stop() uses this completion for synchronization.
928 complete_vfork_done(tsk);
932 * Allocate a new mm structure and copy contents from the
933 * mm structure of the passed in task structure.
935 static struct mm_struct *dup_mm(struct task_struct *tsk)
937 struct mm_struct *mm, *oldmm = current->mm;
944 memcpy(mm, oldmm, sizeof(*mm));
946 if (!mm_init(mm, tsk))
949 err = dup_mmap(mm, oldmm);
953 mm->hiwater_rss = get_mm_rss(mm);
954 mm->hiwater_vm = mm->total_vm;
956 if (mm->binfmt && !try_module_get(mm->binfmt->module))
962 /* don't put binfmt in mmput, we haven't got module yet */
970 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
972 struct mm_struct *mm, *oldmm;
975 tsk->min_flt = tsk->maj_flt = 0;
976 tsk->nvcsw = tsk->nivcsw = 0;
977 #ifdef CONFIG_DETECT_HUNG_TASK
978 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
982 tsk->active_mm = NULL;
985 * Are we cloning a kernel thread?
987 * We need to steal a active VM for that..
993 /* initialize the new vmacache entries */
996 if (clone_flags & CLONE_VM) {
997 atomic_inc(&oldmm->mm_users);
1009 tsk->active_mm = mm;
1016 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1018 struct fs_struct *fs = current->fs;
1019 if (clone_flags & CLONE_FS) {
1020 /* tsk->fs is already what we want */
1021 spin_lock(&fs->lock);
1023 spin_unlock(&fs->lock);
1027 spin_unlock(&fs->lock);
1030 tsk->fs = copy_fs_struct(fs);
1036 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1038 struct files_struct *oldf, *newf;
1042 * A background process may not have any files ...
1044 oldf = current->files;
1048 if (clone_flags & CLONE_FILES) {
1049 atomic_inc(&oldf->count);
1053 newf = dup_fd(oldf, &error);
1063 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1066 struct io_context *ioc = current->io_context;
1067 struct io_context *new_ioc;
1072 * Share io context with parent, if CLONE_IO is set
1074 if (clone_flags & CLONE_IO) {
1076 tsk->io_context = ioc;
1077 } else if (ioprio_valid(ioc->ioprio)) {
1078 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1079 if (unlikely(!new_ioc))
1082 new_ioc->ioprio = ioc->ioprio;
1083 put_io_context(new_ioc);
1089 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1091 struct sighand_struct *sig;
1093 if (clone_flags & CLONE_SIGHAND) {
1094 atomic_inc(¤t->sighand->count);
1097 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1098 rcu_assign_pointer(tsk->sighand, sig);
1102 atomic_set(&sig->count, 1);
1103 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1107 void __cleanup_sighand(struct sighand_struct *sighand)
1109 if (atomic_dec_and_test(&sighand->count)) {
1110 signalfd_cleanup(sighand);
1112 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1113 * without an RCU grace period, see __lock_task_sighand().
1115 kmem_cache_free(sighand_cachep, sighand);
1120 * Initialize POSIX timer handling for a thread group.
1122 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1124 unsigned long cpu_limit;
1126 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1127 if (cpu_limit != RLIM_INFINITY) {
1128 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1129 sig->cputimer.running = true;
1132 /* The timer lists. */
1133 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1134 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1135 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1138 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1140 struct signal_struct *sig;
1142 if (clone_flags & CLONE_THREAD)
1145 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1150 sig->nr_threads = 1;
1151 atomic_set(&sig->live, 1);
1152 atomic_set(&sig->sigcnt, 1);
1154 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1155 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1156 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1158 init_waitqueue_head(&sig->wait_chldexit);
1159 sig->curr_target = tsk;
1160 init_sigpending(&sig->shared_pending);
1161 INIT_LIST_HEAD(&sig->posix_timers);
1162 seqlock_init(&sig->stats_lock);
1163 prev_cputime_init(&sig->prev_cputime);
1165 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1166 sig->real_timer.function = it_real_fn;
1168 task_lock(current->group_leader);
1169 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1170 task_unlock(current->group_leader);
1172 posix_cpu_timers_init_group(sig);
1174 tty_audit_fork(sig);
1175 sched_autogroup_fork(sig);
1177 sig->oom_score_adj = current->signal->oom_score_adj;
1178 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1180 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1181 current->signal->is_child_subreaper;
1183 mutex_init(&sig->cred_guard_mutex);
1188 static void copy_seccomp(struct task_struct *p)
1190 #ifdef CONFIG_SECCOMP
1192 * Must be called with sighand->lock held, which is common to
1193 * all threads in the group. Holding cred_guard_mutex is not
1194 * needed because this new task is not yet running and cannot
1197 assert_spin_locked(¤t->sighand->siglock);
1199 /* Ref-count the new filter user, and assign it. */
1200 get_seccomp_filter(current);
1201 p->seccomp = current->seccomp;
1204 * Explicitly enable no_new_privs here in case it got set
1205 * between the task_struct being duplicated and holding the
1206 * sighand lock. The seccomp state and nnp must be in sync.
1208 if (task_no_new_privs(current))
1209 task_set_no_new_privs(p);
1212 * If the parent gained a seccomp mode after copying thread
1213 * flags and between before we held the sighand lock, we have
1214 * to manually enable the seccomp thread flag here.
1216 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1217 set_tsk_thread_flag(p, TIF_SECCOMP);
1221 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1223 current->clear_child_tid = tidptr;
1225 return task_pid_vnr(current);
1228 static void rt_mutex_init_task(struct task_struct *p)
1230 raw_spin_lock_init(&p->pi_lock);
1231 #ifdef CONFIG_RT_MUTEXES
1232 p->pi_waiters = RB_ROOT;
1233 p->pi_waiters_leftmost = NULL;
1234 p->pi_blocked_on = NULL;
1239 * Initialize POSIX timer handling for a single task.
1241 static void posix_cpu_timers_init(struct task_struct *tsk)
1243 tsk->cputime_expires.prof_exp = 0;
1244 tsk->cputime_expires.virt_exp = 0;
1245 tsk->cputime_expires.sched_exp = 0;
1246 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1247 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1248 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1252 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1254 task->pids[type].pid = pid;
1258 * This creates a new process as a copy of the old one,
1259 * but does not actually start it yet.
1261 * It copies the registers, and all the appropriate
1262 * parts of the process environment (as per the clone
1263 * flags). The actual kick-off is left to the caller.
1265 static struct task_struct *copy_process(unsigned long clone_flags,
1266 unsigned long stack_start,
1267 unsigned long stack_size,
1268 int __user *child_tidptr,
1274 struct task_struct *p;
1275 void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1277 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1278 return ERR_PTR(-EINVAL);
1280 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1281 return ERR_PTR(-EINVAL);
1284 * Thread groups must share signals as well, and detached threads
1285 * can only be started up within the thread group.
1287 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1288 return ERR_PTR(-EINVAL);
1291 * Shared signal handlers imply shared VM. By way of the above,
1292 * thread groups also imply shared VM. Blocking this case allows
1293 * for various simplifications in other code.
1295 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1296 return ERR_PTR(-EINVAL);
1299 * Siblings of global init remain as zombies on exit since they are
1300 * not reaped by their parent (swapper). To solve this and to avoid
1301 * multi-rooted process trees, prevent global and container-inits
1302 * from creating siblings.
1304 if ((clone_flags & CLONE_PARENT) &&
1305 current->signal->flags & SIGNAL_UNKILLABLE)
1306 return ERR_PTR(-EINVAL);
1309 * If the new process will be in a different pid or user namespace
1310 * do not allow it to share a thread group with the forking task.
1312 if (clone_flags & CLONE_THREAD) {
1313 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1314 (task_active_pid_ns(current) !=
1315 current->nsproxy->pid_ns_for_children))
1316 return ERR_PTR(-EINVAL);
1319 retval = security_task_create(clone_flags);
1324 p = dup_task_struct(current);
1328 ftrace_graph_init_task(p);
1330 rt_mutex_init_task(p);
1332 #ifdef CONFIG_PROVE_LOCKING
1333 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1334 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1337 if (atomic_read(&p->real_cred->user->processes) >=
1338 task_rlimit(p, RLIMIT_NPROC)) {
1339 if (p->real_cred->user != INIT_USER &&
1340 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1343 current->flags &= ~PF_NPROC_EXCEEDED;
1345 retval = copy_creds(p, clone_flags);
1350 * If multiple threads are within copy_process(), then this check
1351 * triggers too late. This doesn't hurt, the check is only there
1352 * to stop root fork bombs.
1355 if (nr_threads >= max_threads)
1356 goto bad_fork_cleanup_count;
1358 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1359 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1360 p->flags |= PF_FORKNOEXEC;
1361 INIT_LIST_HEAD(&p->children);
1362 INIT_LIST_HEAD(&p->sibling);
1363 rcu_copy_process(p);
1364 p->vfork_done = NULL;
1365 spin_lock_init(&p->alloc_lock);
1367 init_sigpending(&p->pending);
1369 p->utime = p->stime = p->gtime = 0;
1370 p->utimescaled = p->stimescaled = 0;
1371 prev_cputime_init(&p->prev_cputime);
1373 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1374 seqlock_init(&p->vtime_seqlock);
1376 p->vtime_snap_whence = VTIME_SLEEPING;
1379 #if defined(SPLIT_RSS_COUNTING)
1380 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1383 p->default_timer_slack_ns = current->timer_slack_ns;
1385 task_io_accounting_init(&p->ioac);
1386 acct_clear_integrals(p);
1388 posix_cpu_timers_init(p);
1390 p->start_time = ktime_get_ns();
1391 p->real_start_time = ktime_get_boot_ns();
1392 p->io_context = NULL;
1393 p->audit_context = NULL;
1396 p->mempolicy = mpol_dup(p->mempolicy);
1397 if (IS_ERR(p->mempolicy)) {
1398 retval = PTR_ERR(p->mempolicy);
1399 p->mempolicy = NULL;
1400 goto bad_fork_cleanup_threadgroup_lock;
1403 #ifdef CONFIG_CPUSETS
1404 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1405 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1406 seqcount_init(&p->mems_allowed_seq);
1408 #ifdef CONFIG_TRACE_IRQFLAGS
1410 p->hardirqs_enabled = 0;
1411 p->hardirq_enable_ip = 0;
1412 p->hardirq_enable_event = 0;
1413 p->hardirq_disable_ip = _THIS_IP_;
1414 p->hardirq_disable_event = 0;
1415 p->softirqs_enabled = 1;
1416 p->softirq_enable_ip = _THIS_IP_;
1417 p->softirq_enable_event = 0;
1418 p->softirq_disable_ip = 0;
1419 p->softirq_disable_event = 0;
1420 p->hardirq_context = 0;
1421 p->softirq_context = 0;
1424 p->pagefault_disabled = 0;
1426 #ifdef CONFIG_LOCKDEP
1427 p->lockdep_depth = 0; /* no locks held yet */
1428 p->curr_chain_key = 0;
1429 p->lockdep_recursion = 0;
1432 #ifdef CONFIG_DEBUG_MUTEXES
1433 p->blocked_on = NULL; /* not blocked yet */
1435 #ifdef CONFIG_BCACHE
1436 p->sequential_io = 0;
1437 p->sequential_io_avg = 0;
1440 /* Perform scheduler related setup. Assign this task to a CPU. */
1441 retval = sched_fork(clone_flags, p);
1443 goto bad_fork_cleanup_policy;
1445 retval = perf_event_init_task(p);
1447 goto bad_fork_cleanup_policy;
1448 retval = audit_alloc(p);
1450 goto bad_fork_cleanup_perf;
1451 /* copy all the process information */
1453 retval = copy_semundo(clone_flags, p);
1455 goto bad_fork_cleanup_audit;
1456 retval = copy_files(clone_flags, p);
1458 goto bad_fork_cleanup_semundo;
1459 retval = copy_fs(clone_flags, p);
1461 goto bad_fork_cleanup_files;
1462 retval = copy_sighand(clone_flags, p);
1464 goto bad_fork_cleanup_fs;
1465 retval = copy_signal(clone_flags, p);
1467 goto bad_fork_cleanup_sighand;
1468 retval = copy_mm(clone_flags, p);
1470 goto bad_fork_cleanup_signal;
1471 retval = copy_namespaces(clone_flags, p);
1473 goto bad_fork_cleanup_mm;
1474 retval = copy_io(clone_flags, p);
1476 goto bad_fork_cleanup_namespaces;
1477 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1479 goto bad_fork_cleanup_io;
1481 if (pid != &init_struct_pid) {
1482 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1484 retval = PTR_ERR(pid);
1485 goto bad_fork_cleanup_io;
1489 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1491 * Clear TID on mm_release()?
1493 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1498 p->robust_list = NULL;
1499 #ifdef CONFIG_COMPAT
1500 p->compat_robust_list = NULL;
1502 INIT_LIST_HEAD(&p->pi_state_list);
1503 p->pi_state_cache = NULL;
1506 * sigaltstack should be cleared when sharing the same VM
1508 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1509 p->sas_ss_sp = p->sas_ss_size = 0;
1512 * Syscall tracing and stepping should be turned off in the
1513 * child regardless of CLONE_PTRACE.
1515 user_disable_single_step(p);
1516 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1517 #ifdef TIF_SYSCALL_EMU
1518 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1520 clear_all_latency_tracing(p);
1522 /* ok, now we should be set up.. */
1523 p->pid = pid_nr(pid);
1524 if (clone_flags & CLONE_THREAD) {
1525 p->exit_signal = -1;
1526 p->group_leader = current->group_leader;
1527 p->tgid = current->tgid;
1529 if (clone_flags & CLONE_PARENT)
1530 p->exit_signal = current->group_leader->exit_signal;
1532 p->exit_signal = (clone_flags & CSIGNAL);
1533 p->group_leader = p;
1538 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1539 p->dirty_paused_when = 0;
1541 p->pdeath_signal = 0;
1542 INIT_LIST_HEAD(&p->thread_group);
1543 p->task_works = NULL;
1545 threadgroup_change_begin(current);
1547 * Ensure that the cgroup subsystem policies allow the new process to be
1548 * forked. It should be noted the the new process's css_set can be changed
1549 * between here and cgroup_post_fork() if an organisation operation is in
1552 retval = cgroup_can_fork(p, cgrp_ss_priv);
1554 goto bad_fork_free_pid;
1557 * Make it visible to the rest of the system, but dont wake it up yet.
1558 * Need tasklist lock for parent etc handling!
1560 write_lock_irq(&tasklist_lock);
1562 /* CLONE_PARENT re-uses the old parent */
1563 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1564 p->real_parent = current->real_parent;
1565 p->parent_exec_id = current->parent_exec_id;
1567 p->real_parent = current;
1568 p->parent_exec_id = current->self_exec_id;
1571 spin_lock(¤t->sighand->siglock);
1574 * Copy seccomp details explicitly here, in case they were changed
1575 * before holding sighand lock.
1580 * Process group and session signals need to be delivered to just the
1581 * parent before the fork or both the parent and the child after the
1582 * fork. Restart if a signal comes in before we add the new process to
1583 * it's process group.
1584 * A fatal signal pending means that current will exit, so the new
1585 * thread can't slip out of an OOM kill (or normal SIGKILL).
1587 recalc_sigpending();
1588 if (signal_pending(current)) {
1589 spin_unlock(¤t->sighand->siglock);
1590 write_unlock_irq(&tasklist_lock);
1591 retval = -ERESTARTNOINTR;
1592 goto bad_fork_cancel_cgroup;
1595 if (likely(p->pid)) {
1596 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1598 init_task_pid(p, PIDTYPE_PID, pid);
1599 if (thread_group_leader(p)) {
1600 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1601 init_task_pid(p, PIDTYPE_SID, task_session(current));
1603 if (is_child_reaper(pid)) {
1604 ns_of_pid(pid)->child_reaper = p;
1605 p->signal->flags |= SIGNAL_UNKILLABLE;
1608 p->signal->leader_pid = pid;
1609 p->signal->tty = tty_kref_get(current->signal->tty);
1610 list_add_tail(&p->sibling, &p->real_parent->children);
1611 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1612 attach_pid(p, PIDTYPE_PGID);
1613 attach_pid(p, PIDTYPE_SID);
1614 __this_cpu_inc(process_counts);
1616 current->signal->nr_threads++;
1617 atomic_inc(¤t->signal->live);
1618 atomic_inc(¤t->signal->sigcnt);
1619 list_add_tail_rcu(&p->thread_group,
1620 &p->group_leader->thread_group);
1621 list_add_tail_rcu(&p->thread_node,
1622 &p->signal->thread_head);
1624 attach_pid(p, PIDTYPE_PID);
1629 spin_unlock(¤t->sighand->siglock);
1630 syscall_tracepoint_update(p);
1631 write_unlock_irq(&tasklist_lock);
1633 proc_fork_connector(p);
1634 cgroup_post_fork(p, cgrp_ss_priv);
1635 threadgroup_change_end(current);
1638 trace_task_newtask(p, clone_flags);
1639 uprobe_copy_process(p, clone_flags);
1643 bad_fork_cancel_cgroup:
1644 cgroup_cancel_fork(p, cgrp_ss_priv);
1646 threadgroup_change_end(current);
1647 if (pid != &init_struct_pid)
1649 bad_fork_cleanup_io:
1652 bad_fork_cleanup_namespaces:
1653 exit_task_namespaces(p);
1654 bad_fork_cleanup_mm:
1657 bad_fork_cleanup_signal:
1658 if (!(clone_flags & CLONE_THREAD))
1659 free_signal_struct(p->signal);
1660 bad_fork_cleanup_sighand:
1661 __cleanup_sighand(p->sighand);
1662 bad_fork_cleanup_fs:
1663 exit_fs(p); /* blocking */
1664 bad_fork_cleanup_files:
1665 exit_files(p); /* blocking */
1666 bad_fork_cleanup_semundo:
1668 bad_fork_cleanup_audit:
1670 bad_fork_cleanup_perf:
1671 perf_event_free_task(p);
1672 bad_fork_cleanup_policy:
1674 mpol_put(p->mempolicy);
1675 bad_fork_cleanup_threadgroup_lock:
1677 delayacct_tsk_free(p);
1678 bad_fork_cleanup_count:
1679 atomic_dec(&p->cred->user->processes);
1684 return ERR_PTR(retval);
1687 static inline void init_idle_pids(struct pid_link *links)
1691 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1692 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1693 links[type].pid = &init_struct_pid;
1697 struct task_struct *fork_idle(int cpu)
1699 struct task_struct *task;
1700 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1701 if (!IS_ERR(task)) {
1702 init_idle_pids(task->pids);
1703 init_idle(task, cpu);
1710 * Ok, this is the main fork-routine.
1712 * It copies the process, and if successful kick-starts
1713 * it and waits for it to finish using the VM if required.
1715 long _do_fork(unsigned long clone_flags,
1716 unsigned long stack_start,
1717 unsigned long stack_size,
1718 int __user *parent_tidptr,
1719 int __user *child_tidptr,
1722 struct task_struct *p;
1727 * Determine whether and which event to report to ptracer. When
1728 * called from kernel_thread or CLONE_UNTRACED is explicitly
1729 * requested, no event is reported; otherwise, report if the event
1730 * for the type of forking is enabled.
1732 if (!(clone_flags & CLONE_UNTRACED)) {
1733 if (clone_flags & CLONE_VFORK)
1734 trace = PTRACE_EVENT_VFORK;
1735 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1736 trace = PTRACE_EVENT_CLONE;
1738 trace = PTRACE_EVENT_FORK;
1740 if (likely(!ptrace_event_enabled(current, trace)))
1744 p = copy_process(clone_flags, stack_start, stack_size,
1745 child_tidptr, NULL, trace, tls);
1747 * Do this prior waking up the new thread - the thread pointer
1748 * might get invalid after that point, if the thread exits quickly.
1751 struct completion vfork;
1754 trace_sched_process_fork(current, p);
1756 pid = get_task_pid(p, PIDTYPE_PID);
1759 if (clone_flags & CLONE_PARENT_SETTID)
1760 put_user(nr, parent_tidptr);
1762 if (clone_flags & CLONE_VFORK) {
1763 p->vfork_done = &vfork;
1764 init_completion(&vfork);
1768 wake_up_new_task(p);
1770 /* forking complete and child started to run, tell ptracer */
1771 if (unlikely(trace))
1772 ptrace_event_pid(trace, pid);
1774 if (clone_flags & CLONE_VFORK) {
1775 if (!wait_for_vfork_done(p, &vfork))
1776 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1786 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1787 /* For compatibility with architectures that call do_fork directly rather than
1788 * using the syscall entry points below. */
1789 long do_fork(unsigned long clone_flags,
1790 unsigned long stack_start,
1791 unsigned long stack_size,
1792 int __user *parent_tidptr,
1793 int __user *child_tidptr)
1795 return _do_fork(clone_flags, stack_start, stack_size,
1796 parent_tidptr, child_tidptr, 0);
1801 * Create a kernel thread.
1803 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1805 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1806 (unsigned long)arg, NULL, NULL, 0);
1809 #ifdef __ARCH_WANT_SYS_FORK
1810 SYSCALL_DEFINE0(fork)
1813 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1815 /* can not support in nommu mode */
1821 #ifdef __ARCH_WANT_SYS_VFORK
1822 SYSCALL_DEFINE0(vfork)
1824 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1829 #ifdef __ARCH_WANT_SYS_CLONE
1830 #ifdef CONFIG_CLONE_BACKWARDS
1831 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1832 int __user *, parent_tidptr,
1834 int __user *, child_tidptr)
1835 #elif defined(CONFIG_CLONE_BACKWARDS2)
1836 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1837 int __user *, parent_tidptr,
1838 int __user *, child_tidptr,
1840 #elif defined(CONFIG_CLONE_BACKWARDS3)
1841 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1843 int __user *, parent_tidptr,
1844 int __user *, child_tidptr,
1847 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1848 int __user *, parent_tidptr,
1849 int __user *, child_tidptr,
1853 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1857 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1858 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1861 static void sighand_ctor(void *data)
1863 struct sighand_struct *sighand = data;
1865 spin_lock_init(&sighand->siglock);
1866 init_waitqueue_head(&sighand->signalfd_wqh);
1869 void __init proc_caches_init(void)
1871 sighand_cachep = kmem_cache_create("sighand_cache",
1872 sizeof(struct sighand_struct), 0,
1873 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1874 SLAB_NOTRACK, sighand_ctor);
1875 signal_cachep = kmem_cache_create("signal_cache",
1876 sizeof(struct signal_struct), 0,
1877 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1878 files_cachep = kmem_cache_create("files_cache",
1879 sizeof(struct files_struct), 0,
1880 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1881 fs_cachep = kmem_cache_create("fs_cache",
1882 sizeof(struct fs_struct), 0,
1883 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1885 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1886 * whole struct cpumask for the OFFSTACK case. We could change
1887 * this to *only* allocate as much of it as required by the
1888 * maximum number of CPU's we can ever have. The cpumask_allocation
1889 * is at the end of the structure, exactly for that reason.
1891 mm_cachep = kmem_cache_create("mm_struct",
1892 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1893 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1894 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1896 nsproxy_cache_init();
1900 * Check constraints on flags passed to the unshare system call.
1902 static int check_unshare_flags(unsigned long unshare_flags)
1904 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1905 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1906 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1907 CLONE_NEWUSER|CLONE_NEWPID))
1910 * Not implemented, but pretend it works if there is nothing
1911 * to unshare. Note that unsharing the address space or the
1912 * signal handlers also need to unshare the signal queues (aka
1915 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1916 if (!thread_group_empty(current))
1919 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1920 if (atomic_read(¤t->sighand->count) > 1)
1923 if (unshare_flags & CLONE_VM) {
1924 if (!current_is_single_threaded())
1932 * Unshare the filesystem structure if it is being shared
1934 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1936 struct fs_struct *fs = current->fs;
1938 if (!(unshare_flags & CLONE_FS) || !fs)
1941 /* don't need lock here; in the worst case we'll do useless copy */
1945 *new_fsp = copy_fs_struct(fs);
1953 * Unshare file descriptor table if it is being shared
1955 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1957 struct files_struct *fd = current->files;
1960 if ((unshare_flags & CLONE_FILES) &&
1961 (fd && atomic_read(&fd->count) > 1)) {
1962 *new_fdp = dup_fd(fd, &error);
1971 * unshare allows a process to 'unshare' part of the process
1972 * context which was originally shared using clone. copy_*
1973 * functions used by do_fork() cannot be used here directly
1974 * because they modify an inactive task_struct that is being
1975 * constructed. Here we are modifying the current, active,
1978 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1980 struct fs_struct *fs, *new_fs = NULL;
1981 struct files_struct *fd, *new_fd = NULL;
1982 struct cred *new_cred = NULL;
1983 struct nsproxy *new_nsproxy = NULL;
1988 * If unsharing a user namespace must also unshare the thread group
1989 * and unshare the filesystem root and working directories.
1991 if (unshare_flags & CLONE_NEWUSER)
1992 unshare_flags |= CLONE_THREAD | CLONE_FS;
1994 * If unsharing vm, must also unshare signal handlers.
1996 if (unshare_flags & CLONE_VM)
1997 unshare_flags |= CLONE_SIGHAND;
1999 * If unsharing a signal handlers, must also unshare the signal queues.
2001 if (unshare_flags & CLONE_SIGHAND)
2002 unshare_flags |= CLONE_THREAD;
2004 * If unsharing namespace, must also unshare filesystem information.
2006 if (unshare_flags & CLONE_NEWNS)
2007 unshare_flags |= CLONE_FS;
2009 err = check_unshare_flags(unshare_flags);
2011 goto bad_unshare_out;
2013 * CLONE_NEWIPC must also detach from the undolist: after switching
2014 * to a new ipc namespace, the semaphore arrays from the old
2015 * namespace are unreachable.
2017 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2019 err = unshare_fs(unshare_flags, &new_fs);
2021 goto bad_unshare_out;
2022 err = unshare_fd(unshare_flags, &new_fd);
2024 goto bad_unshare_cleanup_fs;
2025 err = unshare_userns(unshare_flags, &new_cred);
2027 goto bad_unshare_cleanup_fd;
2028 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2031 goto bad_unshare_cleanup_cred;
2033 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2036 * CLONE_SYSVSEM is equivalent to sys_exit().
2040 if (unshare_flags & CLONE_NEWIPC) {
2041 /* Orphan segments in old ns (see sem above). */
2043 shm_init_task(current);
2047 switch_task_namespaces(current, new_nsproxy);
2053 spin_lock(&fs->lock);
2054 current->fs = new_fs;
2059 spin_unlock(&fs->lock);
2063 fd = current->files;
2064 current->files = new_fd;
2068 task_unlock(current);
2071 /* Install the new user namespace */
2072 commit_creds(new_cred);
2077 bad_unshare_cleanup_cred:
2080 bad_unshare_cleanup_fd:
2082 put_files_struct(new_fd);
2084 bad_unshare_cleanup_fs:
2086 free_fs_struct(new_fs);
2093 * Helper to unshare the files of the current task.
2094 * We don't want to expose copy_files internals to
2095 * the exec layer of the kernel.
2098 int unshare_files(struct files_struct **displaced)
2100 struct task_struct *task = current;
2101 struct files_struct *copy = NULL;
2104 error = unshare_fd(CLONE_FILES, ©);
2105 if (error || !copy) {
2109 *displaced = task->files;
2116 int sysctl_max_threads(struct ctl_table *table, int write,
2117 void __user *buffer, size_t *lenp, loff_t *ppos)
2121 int threads = max_threads;
2122 int min = MIN_THREADS;
2123 int max = MAX_THREADS;
2130 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2134 set_max_threads(threads);