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>
78 #include <linux/kcov.h>
80 #include <asm/pgtable.h>
81 #include <asm/pgalloc.h>
82 #include <asm/uaccess.h>
83 #include <asm/mmu_context.h>
84 #include <asm/cacheflush.h>
85 #include <asm/tlbflush.h>
87 #include <trace/events/sched.h>
89 #define CREATE_TRACE_POINTS
90 #include <trace/events/task.h>
93 * Minimum number of threads to boot the kernel
95 #define MIN_THREADS 20
98 * Maximum number of threads
100 #define MAX_THREADS FUTEX_TID_MASK
103 * Protected counters by write_lock_irq(&tasklist_lock)
105 unsigned long total_forks; /* Handle normal Linux uptimes. */
106 int nr_threads; /* The idle threads do not count.. */
108 int max_threads; /* tunable limit on nr_threads */
110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
114 #ifdef CONFIG_PROVE_RCU
115 int lockdep_tasklist_lock_is_held(void)
117 return lockdep_is_held(&tasklist_lock);
119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
120 #endif /* #ifdef CONFIG_PROVE_RCU */
122 int nr_processes(void)
127 for_each_possible_cpu(cpu)
128 total += per_cpu(process_counts, cpu);
133 void __weak arch_release_task_struct(struct task_struct *tsk)
137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
138 static struct kmem_cache *task_struct_cachep;
140 static inline struct task_struct *alloc_task_struct_node(int node)
142 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
145 static inline void free_task_struct(struct task_struct *tsk)
147 kmem_cache_free(task_struct_cachep, tsk);
151 void __weak arch_release_thread_info(struct thread_info *ti)
155 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
158 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159 * kmemcache based allocator.
161 # if THREAD_SIZE >= PAGE_SIZE
162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
165 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
169 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
170 1 << THREAD_SIZE_ORDER);
172 return page ? page_address(page) : NULL;
175 static inline void free_thread_info(struct thread_info *ti)
177 struct page *page = virt_to_page(ti);
179 memcg_kmem_update_page_stat(page, MEMCG_KERNEL_STACK,
180 -(1 << THREAD_SIZE_ORDER));
181 __free_kmem_pages(page, THREAD_SIZE_ORDER);
184 static struct kmem_cache *thread_info_cache;
186 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
189 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
192 static void free_thread_info(struct thread_info *ti)
194 kmem_cache_free(thread_info_cache, ti);
197 void thread_info_cache_init(void)
199 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
200 THREAD_SIZE, 0, NULL);
201 BUG_ON(thread_info_cache == NULL);
206 /* SLAB cache for signal_struct structures (tsk->signal) */
207 static struct kmem_cache *signal_cachep;
209 /* SLAB cache for sighand_struct structures (tsk->sighand) */
210 struct kmem_cache *sighand_cachep;
212 /* SLAB cache for files_struct structures (tsk->files) */
213 struct kmem_cache *files_cachep;
215 /* SLAB cache for fs_struct structures (tsk->fs) */
216 struct kmem_cache *fs_cachep;
218 /* SLAB cache for vm_area_struct structures */
219 struct kmem_cache *vm_area_cachep;
221 /* SLAB cache for mm_struct structures (tsk->mm) */
222 static struct kmem_cache *mm_cachep;
224 static void account_kernel_stack(struct thread_info *ti, int account)
226 struct zone *zone = page_zone(virt_to_page(ti));
228 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
231 void free_task(struct task_struct *tsk)
233 account_kernel_stack(tsk->stack, -1);
234 arch_release_thread_info(tsk->stack);
235 free_thread_info(tsk->stack);
236 rt_mutex_debug_task_free(tsk);
237 ftrace_graph_exit_task(tsk);
238 put_seccomp_filter(tsk);
239 arch_release_task_struct(tsk);
240 free_task_struct(tsk);
242 EXPORT_SYMBOL(free_task);
244 static inline void free_signal_struct(struct signal_struct *sig)
246 taskstats_tgid_free(sig);
247 sched_autogroup_exit(sig);
248 kmem_cache_free(signal_cachep, sig);
251 static inline void put_signal_struct(struct signal_struct *sig)
253 if (atomic_dec_and_test(&sig->sigcnt))
254 free_signal_struct(sig);
257 void __put_task_struct(struct task_struct *tsk)
259 WARN_ON(!tsk->exit_state);
260 WARN_ON(atomic_read(&tsk->usage));
261 WARN_ON(tsk == current);
265 security_task_free(tsk);
267 delayacct_tsk_free(tsk);
268 put_signal_struct(tsk->signal);
270 if (!profile_handoff_task(tsk))
273 EXPORT_SYMBOL_GPL(__put_task_struct);
275 void __init __weak arch_task_cache_init(void) { }
280 static void set_max_threads(unsigned int max_threads_suggested)
285 * The number of threads shall be limited such that the thread
286 * structures may only consume a small part of the available memory.
288 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
289 threads = MAX_THREADS;
291 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
292 (u64) THREAD_SIZE * 8UL);
294 if (threads > max_threads_suggested)
295 threads = max_threads_suggested;
297 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
300 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
301 /* Initialized by the architecture: */
302 int arch_task_struct_size __read_mostly;
305 void __init fork_init(void)
307 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
308 #ifndef ARCH_MIN_TASKALIGN
309 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
311 /* create a slab on which task_structs can be allocated */
312 task_struct_cachep = kmem_cache_create("task_struct",
313 arch_task_struct_size, ARCH_MIN_TASKALIGN,
314 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
317 /* do the arch specific task caches init */
318 arch_task_cache_init();
320 set_max_threads(MAX_THREADS);
322 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
323 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
324 init_task.signal->rlim[RLIMIT_SIGPENDING] =
325 init_task.signal->rlim[RLIMIT_NPROC];
328 int __weak arch_dup_task_struct(struct task_struct *dst,
329 struct task_struct *src)
335 void set_task_stack_end_magic(struct task_struct *tsk)
337 unsigned long *stackend;
339 stackend = end_of_stack(tsk);
340 *stackend = STACK_END_MAGIC; /* for overflow detection */
343 static struct task_struct *dup_task_struct(struct task_struct *orig)
345 struct task_struct *tsk;
346 struct thread_info *ti;
347 int node = tsk_fork_get_node(orig);
350 tsk = alloc_task_struct_node(node);
354 ti = alloc_thread_info_node(tsk, node);
358 err = arch_dup_task_struct(tsk, orig);
363 #ifdef CONFIG_SECCOMP
365 * We must handle setting up seccomp filters once we're under
366 * the sighand lock in case orig has changed between now and
367 * then. Until then, filter must be NULL to avoid messing up
368 * the usage counts on the error path calling free_task.
370 tsk->seccomp.filter = NULL;
373 setup_thread_stack(tsk, orig);
374 clear_user_return_notifier(tsk);
375 clear_tsk_need_resched(tsk);
376 set_task_stack_end_magic(tsk);
378 #ifdef CONFIG_CC_STACKPROTECTOR
379 tsk->stack_canary = get_random_int();
383 * One for us, one for whoever does the "release_task()" (usually
386 atomic_set(&tsk->usage, 2);
387 #ifdef CONFIG_BLK_DEV_IO_TRACE
390 tsk->splice_pipe = NULL;
391 tsk->task_frag.page = NULL;
392 tsk->wake_q.next = NULL;
394 account_kernel_stack(ti, 1);
401 free_thread_info(ti);
403 free_task_struct(tsk);
408 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
410 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
411 struct rb_node **rb_link, *rb_parent;
413 unsigned long charge;
415 uprobe_start_dup_mmap();
416 down_write(&oldmm->mmap_sem);
417 flush_cache_dup_mm(oldmm);
418 uprobe_dup_mmap(oldmm, mm);
420 * Not linked in yet - no deadlock potential:
422 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
424 /* No ordering required: file already has been exposed. */
425 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
427 mm->total_vm = oldmm->total_vm;
428 mm->data_vm = oldmm->data_vm;
429 mm->exec_vm = oldmm->exec_vm;
430 mm->stack_vm = oldmm->stack_vm;
432 rb_link = &mm->mm_rb.rb_node;
435 retval = ksm_fork(mm, oldmm);
438 retval = khugepaged_fork(mm, oldmm);
443 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
446 if (mpnt->vm_flags & VM_DONTCOPY) {
447 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
451 if (mpnt->vm_flags & VM_ACCOUNT) {
452 unsigned long len = vma_pages(mpnt);
454 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
458 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
462 INIT_LIST_HEAD(&tmp->anon_vma_chain);
463 retval = vma_dup_policy(mpnt, tmp);
465 goto fail_nomem_policy;
467 if (anon_vma_fork(tmp, mpnt))
468 goto fail_nomem_anon_vma_fork;
470 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
471 tmp->vm_next = tmp->vm_prev = NULL;
472 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
475 struct inode *inode = file_inode(file);
476 struct address_space *mapping = file->f_mapping;
479 if (tmp->vm_flags & VM_DENYWRITE)
480 atomic_dec(&inode->i_writecount);
481 i_mmap_lock_write(mapping);
482 if (tmp->vm_flags & VM_SHARED)
483 atomic_inc(&mapping->i_mmap_writable);
484 flush_dcache_mmap_lock(mapping);
485 /* insert tmp into the share list, just after mpnt */
486 vma_interval_tree_insert_after(tmp, mpnt,
488 flush_dcache_mmap_unlock(mapping);
489 i_mmap_unlock_write(mapping);
493 * Clear hugetlb-related page reserves for children. This only
494 * affects MAP_PRIVATE mappings. Faults generated by the child
495 * are not guaranteed to succeed, even if read-only
497 if (is_vm_hugetlb_page(tmp))
498 reset_vma_resv_huge_pages(tmp);
501 * Link in the new vma and copy the page table entries.
504 pprev = &tmp->vm_next;
508 __vma_link_rb(mm, tmp, rb_link, rb_parent);
509 rb_link = &tmp->vm_rb.rb_right;
510 rb_parent = &tmp->vm_rb;
513 retval = copy_page_range(mm, oldmm, mpnt);
515 if (tmp->vm_ops && tmp->vm_ops->open)
516 tmp->vm_ops->open(tmp);
521 /* a new mm has just been created */
522 arch_dup_mmap(oldmm, mm);
525 up_write(&mm->mmap_sem);
527 up_write(&oldmm->mmap_sem);
528 uprobe_end_dup_mmap();
530 fail_nomem_anon_vma_fork:
531 mpol_put(vma_policy(tmp));
533 kmem_cache_free(vm_area_cachep, tmp);
536 vm_unacct_memory(charge);
540 static inline int mm_alloc_pgd(struct mm_struct *mm)
542 mm->pgd = pgd_alloc(mm);
543 if (unlikely(!mm->pgd))
548 static inline void mm_free_pgd(struct mm_struct *mm)
550 pgd_free(mm, mm->pgd);
553 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
555 down_write(&oldmm->mmap_sem);
556 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
557 up_write(&oldmm->mmap_sem);
560 #define mm_alloc_pgd(mm) (0)
561 #define mm_free_pgd(mm)
562 #endif /* CONFIG_MMU */
564 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
566 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
567 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
569 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
571 static int __init coredump_filter_setup(char *s)
573 default_dump_filter =
574 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
575 MMF_DUMP_FILTER_MASK;
579 __setup("coredump_filter=", coredump_filter_setup);
581 #include <linux/init_task.h>
583 static void mm_init_aio(struct mm_struct *mm)
586 spin_lock_init(&mm->ioctx_lock);
587 mm->ioctx_table = NULL;
591 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
598 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
602 mm->vmacache_seqnum = 0;
603 atomic_set(&mm->mm_users, 1);
604 atomic_set(&mm->mm_count, 1);
605 init_rwsem(&mm->mmap_sem);
606 INIT_LIST_HEAD(&mm->mmlist);
607 mm->core_state = NULL;
608 atomic_long_set(&mm->nr_ptes, 0);
613 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
614 spin_lock_init(&mm->page_table_lock);
617 mm_init_owner(mm, p);
618 mmu_notifier_mm_init(mm);
619 clear_tlb_flush_pending(mm);
620 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
621 mm->pmd_huge_pte = NULL;
625 mm->flags = current->mm->flags & MMF_INIT_MASK;
626 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
628 mm->flags = default_dump_filter;
632 if (mm_alloc_pgd(mm))
635 if (init_new_context(p, mm))
647 static void check_mm(struct mm_struct *mm)
651 for (i = 0; i < NR_MM_COUNTERS; i++) {
652 long x = atomic_long_read(&mm->rss_stat.count[i]);
655 printk(KERN_ALERT "BUG: Bad rss-counter state "
656 "mm:%p idx:%d val:%ld\n", mm, i, x);
659 if (atomic_long_read(&mm->nr_ptes))
660 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
661 atomic_long_read(&mm->nr_ptes));
663 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
666 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
667 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
672 * Allocate and initialize an mm_struct.
674 struct mm_struct *mm_alloc(void)
676 struct mm_struct *mm;
682 memset(mm, 0, sizeof(*mm));
683 return mm_init(mm, current);
687 * Called when the last reference to the mm
688 * is dropped: either by a lazy thread or by
689 * mmput. Free the page directory and the mm.
691 void __mmdrop(struct mm_struct *mm)
693 BUG_ON(mm == &init_mm);
696 mmu_notifier_mm_destroy(mm);
700 EXPORT_SYMBOL_GPL(__mmdrop);
703 * Decrement the use count and release all resources for an mm.
705 void mmput(struct mm_struct *mm)
709 if (atomic_dec_and_test(&mm->mm_users)) {
710 uprobe_clear_state(mm);
713 khugepaged_exit(mm); /* must run before exit_mmap */
715 set_mm_exe_file(mm, NULL);
716 if (!list_empty(&mm->mmlist)) {
717 spin_lock(&mmlist_lock);
718 list_del(&mm->mmlist);
719 spin_unlock(&mmlist_lock);
722 module_put(mm->binfmt->module);
726 EXPORT_SYMBOL_GPL(mmput);
729 * set_mm_exe_file - change a reference to the mm's executable file
731 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
733 * Main users are mmput() and sys_execve(). Callers prevent concurrent
734 * invocations: in mmput() nobody alive left, in execve task is single
735 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
736 * mm->exe_file, but does so without using set_mm_exe_file() in order
737 * to do avoid the need for any locks.
739 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
741 struct file *old_exe_file;
744 * It is safe to dereference the exe_file without RCU as
745 * this function is only called if nobody else can access
746 * this mm -- see comment above for justification.
748 old_exe_file = rcu_dereference_raw(mm->exe_file);
751 get_file(new_exe_file);
752 rcu_assign_pointer(mm->exe_file, new_exe_file);
758 * get_mm_exe_file - acquire a reference to the mm's executable file
760 * Returns %NULL if mm has no associated executable file.
761 * User must release file via fput().
763 struct file *get_mm_exe_file(struct mm_struct *mm)
765 struct file *exe_file;
768 exe_file = rcu_dereference(mm->exe_file);
769 if (exe_file && !get_file_rcu(exe_file))
774 EXPORT_SYMBOL(get_mm_exe_file);
777 * get_task_exe_file - acquire a reference to the task's executable file
779 * Returns %NULL if task's mm (if any) has no associated executable file or
780 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
781 * User must release file via fput().
783 struct file *get_task_exe_file(struct task_struct *task)
785 struct file *exe_file = NULL;
786 struct mm_struct *mm;
791 if (!(task->flags & PF_KTHREAD))
792 exe_file = get_mm_exe_file(mm);
797 EXPORT_SYMBOL(get_task_exe_file);
800 * get_task_mm - acquire a reference to the task's mm
802 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
803 * this kernel workthread has transiently adopted a user mm with use_mm,
804 * to do its AIO) is not set and if so returns a reference to it, after
805 * bumping up the use count. User must release the mm via mmput()
806 * after use. Typically used by /proc and ptrace.
808 struct mm_struct *get_task_mm(struct task_struct *task)
810 struct mm_struct *mm;
815 if (task->flags & PF_KTHREAD)
818 atomic_inc(&mm->mm_users);
823 EXPORT_SYMBOL_GPL(get_task_mm);
825 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
827 struct mm_struct *mm;
830 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
834 mm = get_task_mm(task);
835 if (mm && mm != current->mm &&
836 !ptrace_may_access(task, mode)) {
838 mm = ERR_PTR(-EACCES);
840 mutex_unlock(&task->signal->cred_guard_mutex);
845 static void complete_vfork_done(struct task_struct *tsk)
847 struct completion *vfork;
850 vfork = tsk->vfork_done;
852 tsk->vfork_done = NULL;
858 static int wait_for_vfork_done(struct task_struct *child,
859 struct completion *vfork)
863 freezer_do_not_count();
864 killed = wait_for_completion_killable(vfork);
869 child->vfork_done = NULL;
873 put_task_struct(child);
877 /* Please note the differences between mmput and mm_release.
878 * mmput is called whenever we stop holding onto a mm_struct,
879 * error success whatever.
881 * mm_release is called after a mm_struct has been removed
882 * from the current process.
884 * This difference is important for error handling, when we
885 * only half set up a mm_struct for a new process and need to restore
886 * the old one. Because we mmput the new mm_struct before
887 * restoring the old one. . .
888 * Eric Biederman 10 January 1998
890 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
892 /* Get rid of any futexes when releasing the mm */
894 if (unlikely(tsk->robust_list)) {
895 exit_robust_list(tsk);
896 tsk->robust_list = NULL;
899 if (unlikely(tsk->compat_robust_list)) {
900 compat_exit_robust_list(tsk);
901 tsk->compat_robust_list = NULL;
904 if (unlikely(!list_empty(&tsk->pi_state_list)))
905 exit_pi_state_list(tsk);
908 uprobe_free_utask(tsk);
910 /* Get rid of any cached register state */
911 deactivate_mm(tsk, mm);
914 * If we're exiting normally, clear a user-space tid field if
915 * requested. We leave this alone when dying by signal, to leave
916 * the value intact in a core dump, and to save the unnecessary
917 * trouble, say, a killed vfork parent shouldn't touch this mm.
918 * Userland only wants this done for a sys_exit.
920 if (tsk->clear_child_tid) {
921 if (!(tsk->flags & PF_SIGNALED) &&
922 atomic_read(&mm->mm_users) > 1) {
924 * We don't check the error code - if userspace has
925 * not set up a proper pointer then tough luck.
927 put_user(0, tsk->clear_child_tid);
928 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
931 tsk->clear_child_tid = NULL;
935 * All done, finally we can wake up parent and return this mm to him.
936 * Also kthread_stop() uses this completion for synchronization.
939 complete_vfork_done(tsk);
943 * Allocate a new mm structure and copy contents from the
944 * mm structure of the passed in task structure.
946 static struct mm_struct *dup_mm(struct task_struct *tsk)
948 struct mm_struct *mm, *oldmm = current->mm;
955 memcpy(mm, oldmm, sizeof(*mm));
957 if (!mm_init(mm, tsk))
960 err = dup_mmap(mm, oldmm);
964 mm->hiwater_rss = get_mm_rss(mm);
965 mm->hiwater_vm = mm->total_vm;
967 if (mm->binfmt && !try_module_get(mm->binfmt->module))
973 /* don't put binfmt in mmput, we haven't got module yet */
981 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
983 struct mm_struct *mm, *oldmm;
986 tsk->min_flt = tsk->maj_flt = 0;
987 tsk->nvcsw = tsk->nivcsw = 0;
988 #ifdef CONFIG_DETECT_HUNG_TASK
989 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
993 tsk->active_mm = NULL;
996 * Are we cloning a kernel thread?
998 * We need to steal a active VM for that..
1000 oldmm = current->mm;
1004 /* initialize the new vmacache entries */
1005 vmacache_flush(tsk);
1007 if (clone_flags & CLONE_VM) {
1008 atomic_inc(&oldmm->mm_users);
1020 tsk->active_mm = mm;
1027 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1029 struct fs_struct *fs = current->fs;
1030 if (clone_flags & CLONE_FS) {
1031 /* tsk->fs is already what we want */
1032 spin_lock(&fs->lock);
1034 spin_unlock(&fs->lock);
1038 spin_unlock(&fs->lock);
1041 tsk->fs = copy_fs_struct(fs);
1047 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1049 struct files_struct *oldf, *newf;
1053 * A background process may not have any files ...
1055 oldf = current->files;
1059 if (clone_flags & CLONE_FILES) {
1060 atomic_inc(&oldf->count);
1064 newf = dup_fd(oldf, &error);
1074 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1077 struct io_context *ioc = current->io_context;
1078 struct io_context *new_ioc;
1083 * Share io context with parent, if CLONE_IO is set
1085 if (clone_flags & CLONE_IO) {
1087 tsk->io_context = ioc;
1088 } else if (ioprio_valid(ioc->ioprio)) {
1089 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1090 if (unlikely(!new_ioc))
1093 new_ioc->ioprio = ioc->ioprio;
1094 put_io_context(new_ioc);
1100 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1102 struct sighand_struct *sig;
1104 if (clone_flags & CLONE_SIGHAND) {
1105 atomic_inc(¤t->sighand->count);
1108 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1109 rcu_assign_pointer(tsk->sighand, sig);
1113 atomic_set(&sig->count, 1);
1114 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1118 void __cleanup_sighand(struct sighand_struct *sighand)
1120 if (atomic_dec_and_test(&sighand->count)) {
1121 signalfd_cleanup(sighand);
1123 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1124 * without an RCU grace period, see __lock_task_sighand().
1126 kmem_cache_free(sighand_cachep, sighand);
1131 * Initialize POSIX timer handling for a thread group.
1133 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1135 unsigned long cpu_limit;
1137 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1138 if (cpu_limit != RLIM_INFINITY) {
1139 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1140 sig->cputimer.running = true;
1143 /* The timer lists. */
1144 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1145 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1146 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1149 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1151 struct signal_struct *sig;
1153 if (clone_flags & CLONE_THREAD)
1156 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1161 sig->nr_threads = 1;
1162 atomic_set(&sig->live, 1);
1163 atomic_set(&sig->sigcnt, 1);
1165 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1166 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1167 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1169 init_waitqueue_head(&sig->wait_chldexit);
1170 sig->curr_target = tsk;
1171 init_sigpending(&sig->shared_pending);
1172 INIT_LIST_HEAD(&sig->posix_timers);
1173 seqlock_init(&sig->stats_lock);
1174 prev_cputime_init(&sig->prev_cputime);
1176 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1177 sig->real_timer.function = it_real_fn;
1179 task_lock(current->group_leader);
1180 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1181 task_unlock(current->group_leader);
1183 posix_cpu_timers_init_group(sig);
1185 tty_audit_fork(sig);
1186 sched_autogroup_fork(sig);
1188 sig->oom_score_adj = current->signal->oom_score_adj;
1189 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1191 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1192 current->signal->is_child_subreaper;
1194 mutex_init(&sig->cred_guard_mutex);
1199 static void copy_seccomp(struct task_struct *p)
1201 #ifdef CONFIG_SECCOMP
1203 * Must be called with sighand->lock held, which is common to
1204 * all threads in the group. Holding cred_guard_mutex is not
1205 * needed because this new task is not yet running and cannot
1208 assert_spin_locked(¤t->sighand->siglock);
1210 /* Ref-count the new filter user, and assign it. */
1211 get_seccomp_filter(current);
1212 p->seccomp = current->seccomp;
1215 * Explicitly enable no_new_privs here in case it got set
1216 * between the task_struct being duplicated and holding the
1217 * sighand lock. The seccomp state and nnp must be in sync.
1219 if (task_no_new_privs(current))
1220 task_set_no_new_privs(p);
1223 * If the parent gained a seccomp mode after copying thread
1224 * flags and between before we held the sighand lock, we have
1225 * to manually enable the seccomp thread flag here.
1227 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1228 set_tsk_thread_flag(p, TIF_SECCOMP);
1232 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1234 current->clear_child_tid = tidptr;
1236 return task_pid_vnr(current);
1239 static void rt_mutex_init_task(struct task_struct *p)
1241 raw_spin_lock_init(&p->pi_lock);
1242 #ifdef CONFIG_RT_MUTEXES
1243 p->pi_waiters = RB_ROOT;
1244 p->pi_waiters_leftmost = NULL;
1245 p->pi_blocked_on = NULL;
1250 * Initialize POSIX timer handling for a single task.
1252 static void posix_cpu_timers_init(struct task_struct *tsk)
1254 tsk->cputime_expires.prof_exp = 0;
1255 tsk->cputime_expires.virt_exp = 0;
1256 tsk->cputime_expires.sched_exp = 0;
1257 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1258 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1259 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1263 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1265 task->pids[type].pid = pid;
1269 * This creates a new process as a copy of the old one,
1270 * but does not actually start it yet.
1272 * It copies the registers, and all the appropriate
1273 * parts of the process environment (as per the clone
1274 * flags). The actual kick-off is left to the caller.
1276 static struct task_struct *copy_process(unsigned long clone_flags,
1277 unsigned long stack_start,
1278 unsigned long stack_size,
1279 int __user *child_tidptr,
1285 struct task_struct *p;
1287 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1288 return ERR_PTR(-EINVAL);
1290 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1291 return ERR_PTR(-EINVAL);
1294 * Thread groups must share signals as well, and detached threads
1295 * can only be started up within the thread group.
1297 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1298 return ERR_PTR(-EINVAL);
1301 * Shared signal handlers imply shared VM. By way of the above,
1302 * thread groups also imply shared VM. Blocking this case allows
1303 * for various simplifications in other code.
1305 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1306 return ERR_PTR(-EINVAL);
1309 * Siblings of global init remain as zombies on exit since they are
1310 * not reaped by their parent (swapper). To solve this and to avoid
1311 * multi-rooted process trees, prevent global and container-inits
1312 * from creating siblings.
1314 if ((clone_flags & CLONE_PARENT) &&
1315 current->signal->flags & SIGNAL_UNKILLABLE)
1316 return ERR_PTR(-EINVAL);
1319 * If the new process will be in a different pid or user namespace
1320 * do not allow it to share a thread group with the forking task.
1322 if (clone_flags & CLONE_THREAD) {
1323 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1324 (task_active_pid_ns(current) !=
1325 current->nsproxy->pid_ns_for_children))
1326 return ERR_PTR(-EINVAL);
1329 retval = security_task_create(clone_flags);
1334 p = dup_task_struct(current);
1338 ftrace_graph_init_task(p);
1340 rt_mutex_init_task(p);
1342 #ifdef CONFIG_PROVE_LOCKING
1343 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1344 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1347 if (atomic_read(&p->real_cred->user->processes) >=
1348 task_rlimit(p, RLIMIT_NPROC)) {
1349 if (p->real_cred->user != INIT_USER &&
1350 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1353 current->flags &= ~PF_NPROC_EXCEEDED;
1355 retval = copy_creds(p, clone_flags);
1360 * If multiple threads are within copy_process(), then this check
1361 * triggers too late. This doesn't hurt, the check is only there
1362 * to stop root fork bombs.
1365 if (nr_threads >= max_threads)
1366 goto bad_fork_cleanup_count;
1368 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1369 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1370 p->flags |= PF_FORKNOEXEC;
1371 INIT_LIST_HEAD(&p->children);
1372 INIT_LIST_HEAD(&p->sibling);
1373 rcu_copy_process(p);
1374 p->vfork_done = NULL;
1375 spin_lock_init(&p->alloc_lock);
1377 init_sigpending(&p->pending);
1379 p->utime = p->stime = p->gtime = 0;
1380 p->utimescaled = p->stimescaled = 0;
1381 prev_cputime_init(&p->prev_cputime);
1383 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1384 seqcount_init(&p->vtime_seqcount);
1386 p->vtime_snap_whence = VTIME_INACTIVE;
1389 #if defined(SPLIT_RSS_COUNTING)
1390 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1393 p->default_timer_slack_ns = current->timer_slack_ns;
1395 task_io_accounting_init(&p->ioac);
1396 acct_clear_integrals(p);
1398 posix_cpu_timers_init(p);
1400 p->start_time = ktime_get_ns();
1401 p->real_start_time = ktime_get_boot_ns();
1402 p->io_context = NULL;
1403 p->audit_context = NULL;
1404 threadgroup_change_begin(current);
1407 p->mempolicy = mpol_dup(p->mempolicy);
1408 if (IS_ERR(p->mempolicy)) {
1409 retval = PTR_ERR(p->mempolicy);
1410 p->mempolicy = NULL;
1411 goto bad_fork_cleanup_threadgroup_lock;
1414 #ifdef CONFIG_CPUSETS
1415 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1416 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1417 seqcount_init(&p->mems_allowed_seq);
1419 #ifdef CONFIG_TRACE_IRQFLAGS
1421 p->hardirqs_enabled = 0;
1422 p->hardirq_enable_ip = 0;
1423 p->hardirq_enable_event = 0;
1424 p->hardirq_disable_ip = _THIS_IP_;
1425 p->hardirq_disable_event = 0;
1426 p->softirqs_enabled = 1;
1427 p->softirq_enable_ip = _THIS_IP_;
1428 p->softirq_enable_event = 0;
1429 p->softirq_disable_ip = 0;
1430 p->softirq_disable_event = 0;
1431 p->hardirq_context = 0;
1432 p->softirq_context = 0;
1435 p->pagefault_disabled = 0;
1437 #ifdef CONFIG_LOCKDEP
1438 p->lockdep_depth = 0; /* no locks held yet */
1439 p->curr_chain_key = 0;
1440 p->lockdep_recursion = 0;
1443 #ifdef CONFIG_DEBUG_MUTEXES
1444 p->blocked_on = NULL; /* not blocked yet */
1446 #ifdef CONFIG_BCACHE
1447 p->sequential_io = 0;
1448 p->sequential_io_avg = 0;
1451 /* Perform scheduler related setup. Assign this task to a CPU. */
1452 retval = sched_fork(clone_flags, p);
1454 goto bad_fork_cleanup_policy;
1456 retval = perf_event_init_task(p);
1458 goto bad_fork_cleanup_policy;
1459 retval = audit_alloc(p);
1461 goto bad_fork_cleanup_perf;
1462 /* copy all the process information */
1464 retval = copy_semundo(clone_flags, p);
1466 goto bad_fork_cleanup_audit;
1467 retval = copy_files(clone_flags, p);
1469 goto bad_fork_cleanup_semundo;
1470 retval = copy_fs(clone_flags, p);
1472 goto bad_fork_cleanup_files;
1473 retval = copy_sighand(clone_flags, p);
1475 goto bad_fork_cleanup_fs;
1476 retval = copy_signal(clone_flags, p);
1478 goto bad_fork_cleanup_sighand;
1479 retval = copy_mm(clone_flags, p);
1481 goto bad_fork_cleanup_signal;
1482 retval = copy_namespaces(clone_flags, p);
1484 goto bad_fork_cleanup_mm;
1485 retval = copy_io(clone_flags, p);
1487 goto bad_fork_cleanup_namespaces;
1488 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1490 goto bad_fork_cleanup_io;
1492 if (pid != &init_struct_pid) {
1493 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1495 retval = PTR_ERR(pid);
1496 goto bad_fork_cleanup_io;
1500 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1502 * Clear TID on mm_release()?
1504 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1509 p->robust_list = NULL;
1510 #ifdef CONFIG_COMPAT
1511 p->compat_robust_list = NULL;
1513 INIT_LIST_HEAD(&p->pi_state_list);
1514 p->pi_state_cache = NULL;
1517 * sigaltstack should be cleared when sharing the same VM
1519 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1520 p->sas_ss_sp = p->sas_ss_size = 0;
1523 * Syscall tracing and stepping should be turned off in the
1524 * child regardless of CLONE_PTRACE.
1526 user_disable_single_step(p);
1527 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1528 #ifdef TIF_SYSCALL_EMU
1529 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1531 clear_all_latency_tracing(p);
1533 /* ok, now we should be set up.. */
1534 p->pid = pid_nr(pid);
1535 if (clone_flags & CLONE_THREAD) {
1536 p->exit_signal = -1;
1537 p->group_leader = current->group_leader;
1538 p->tgid = current->tgid;
1540 if (clone_flags & CLONE_PARENT)
1541 p->exit_signal = current->group_leader->exit_signal;
1543 p->exit_signal = (clone_flags & CSIGNAL);
1544 p->group_leader = p;
1549 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1550 p->dirty_paused_when = 0;
1552 p->pdeath_signal = 0;
1553 INIT_LIST_HEAD(&p->thread_group);
1554 p->task_works = NULL;
1557 * Ensure that the cgroup subsystem policies allow the new process to be
1558 * forked. It should be noted the the new process's css_set can be changed
1559 * between here and cgroup_post_fork() if an organisation operation is in
1562 retval = cgroup_can_fork(p);
1564 goto bad_fork_free_pid;
1567 * Make it visible to the rest of the system, but dont wake it up yet.
1568 * Need tasklist lock for parent etc handling!
1570 write_lock_irq(&tasklist_lock);
1572 /* CLONE_PARENT re-uses the old parent */
1573 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1574 p->real_parent = current->real_parent;
1575 p->parent_exec_id = current->parent_exec_id;
1577 p->real_parent = current;
1578 p->parent_exec_id = current->self_exec_id;
1581 spin_lock(¤t->sighand->siglock);
1584 * Copy seccomp details explicitly here, in case they were changed
1585 * before holding sighand lock.
1590 * Process group and session signals need to be delivered to just the
1591 * parent before the fork or both the parent and the child after the
1592 * fork. Restart if a signal comes in before we add the new process to
1593 * it's process group.
1594 * A fatal signal pending means that current will exit, so the new
1595 * thread can't slip out of an OOM kill (or normal SIGKILL).
1597 recalc_sigpending();
1598 if (signal_pending(current)) {
1599 spin_unlock(¤t->sighand->siglock);
1600 write_unlock_irq(&tasklist_lock);
1601 retval = -ERESTARTNOINTR;
1602 goto bad_fork_cancel_cgroup;
1605 if (likely(p->pid)) {
1606 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1608 init_task_pid(p, PIDTYPE_PID, pid);
1609 if (thread_group_leader(p)) {
1610 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1611 init_task_pid(p, PIDTYPE_SID, task_session(current));
1613 if (is_child_reaper(pid)) {
1614 ns_of_pid(pid)->child_reaper = p;
1615 p->signal->flags |= SIGNAL_UNKILLABLE;
1618 p->signal->leader_pid = pid;
1619 p->signal->tty = tty_kref_get(current->signal->tty);
1620 list_add_tail(&p->sibling, &p->real_parent->children);
1621 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1622 attach_pid(p, PIDTYPE_PGID);
1623 attach_pid(p, PIDTYPE_SID);
1624 __this_cpu_inc(process_counts);
1626 current->signal->nr_threads++;
1627 atomic_inc(¤t->signal->live);
1628 atomic_inc(¤t->signal->sigcnt);
1629 list_add_tail_rcu(&p->thread_group,
1630 &p->group_leader->thread_group);
1631 list_add_tail_rcu(&p->thread_node,
1632 &p->signal->thread_head);
1634 attach_pid(p, PIDTYPE_PID);
1639 spin_unlock(¤t->sighand->siglock);
1640 syscall_tracepoint_update(p);
1641 write_unlock_irq(&tasklist_lock);
1643 proc_fork_connector(p);
1644 cgroup_post_fork(p);
1645 threadgroup_change_end(current);
1648 trace_task_newtask(p, clone_flags);
1649 uprobe_copy_process(p, clone_flags);
1653 bad_fork_cancel_cgroup:
1654 cgroup_cancel_fork(p);
1656 if (pid != &init_struct_pid)
1658 bad_fork_cleanup_io:
1661 bad_fork_cleanup_namespaces:
1662 exit_task_namespaces(p);
1663 bad_fork_cleanup_mm:
1666 bad_fork_cleanup_signal:
1667 if (!(clone_flags & CLONE_THREAD))
1668 free_signal_struct(p->signal);
1669 bad_fork_cleanup_sighand:
1670 __cleanup_sighand(p->sighand);
1671 bad_fork_cleanup_fs:
1672 exit_fs(p); /* blocking */
1673 bad_fork_cleanup_files:
1674 exit_files(p); /* blocking */
1675 bad_fork_cleanup_semundo:
1677 bad_fork_cleanup_audit:
1679 bad_fork_cleanup_perf:
1680 perf_event_free_task(p);
1681 bad_fork_cleanup_policy:
1683 mpol_put(p->mempolicy);
1684 bad_fork_cleanup_threadgroup_lock:
1686 threadgroup_change_end(current);
1687 delayacct_tsk_free(p);
1688 bad_fork_cleanup_count:
1689 atomic_dec(&p->cred->user->processes);
1694 return ERR_PTR(retval);
1697 static inline void init_idle_pids(struct pid_link *links)
1701 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1702 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1703 links[type].pid = &init_struct_pid;
1707 struct task_struct *fork_idle(int cpu)
1709 struct task_struct *task;
1710 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0);
1711 if (!IS_ERR(task)) {
1712 init_idle_pids(task->pids);
1713 init_idle(task, cpu);
1720 * Ok, this is the main fork-routine.
1722 * It copies the process, and if successful kick-starts
1723 * it and waits for it to finish using the VM if required.
1725 long _do_fork(unsigned long clone_flags,
1726 unsigned long stack_start,
1727 unsigned long stack_size,
1728 int __user *parent_tidptr,
1729 int __user *child_tidptr,
1732 struct task_struct *p;
1737 * Determine whether and which event to report to ptracer. When
1738 * called from kernel_thread or CLONE_UNTRACED is explicitly
1739 * requested, no event is reported; otherwise, report if the event
1740 * for the type of forking is enabled.
1742 if (!(clone_flags & CLONE_UNTRACED)) {
1743 if (clone_flags & CLONE_VFORK)
1744 trace = PTRACE_EVENT_VFORK;
1745 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1746 trace = PTRACE_EVENT_CLONE;
1748 trace = PTRACE_EVENT_FORK;
1750 if (likely(!ptrace_event_enabled(current, trace)))
1754 p = copy_process(clone_flags, stack_start, stack_size,
1755 child_tidptr, NULL, trace, tls);
1757 * Do this prior waking up the new thread - the thread pointer
1758 * might get invalid after that point, if the thread exits quickly.
1761 struct completion vfork;
1764 trace_sched_process_fork(current, p);
1766 pid = get_task_pid(p, PIDTYPE_PID);
1769 if (clone_flags & CLONE_PARENT_SETTID)
1770 put_user(nr, parent_tidptr);
1772 if (clone_flags & CLONE_VFORK) {
1773 p->vfork_done = &vfork;
1774 init_completion(&vfork);
1778 wake_up_new_task(p);
1780 /* forking complete and child started to run, tell ptracer */
1781 if (unlikely(trace))
1782 ptrace_event_pid(trace, pid);
1784 if (clone_flags & CLONE_VFORK) {
1785 if (!wait_for_vfork_done(p, &vfork))
1786 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1796 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1797 /* For compatibility with architectures that call do_fork directly rather than
1798 * using the syscall entry points below. */
1799 long do_fork(unsigned long clone_flags,
1800 unsigned long stack_start,
1801 unsigned long stack_size,
1802 int __user *parent_tidptr,
1803 int __user *child_tidptr)
1805 return _do_fork(clone_flags, stack_start, stack_size,
1806 parent_tidptr, child_tidptr, 0);
1811 * Create a kernel thread.
1813 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1815 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1816 (unsigned long)arg, NULL, NULL, 0);
1819 #ifdef __ARCH_WANT_SYS_FORK
1820 SYSCALL_DEFINE0(fork)
1823 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1825 /* can not support in nommu mode */
1831 #ifdef __ARCH_WANT_SYS_VFORK
1832 SYSCALL_DEFINE0(vfork)
1834 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1839 #ifdef __ARCH_WANT_SYS_CLONE
1840 #ifdef CONFIG_CLONE_BACKWARDS
1841 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1842 int __user *, parent_tidptr,
1844 int __user *, child_tidptr)
1845 #elif defined(CONFIG_CLONE_BACKWARDS2)
1846 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1847 int __user *, parent_tidptr,
1848 int __user *, child_tidptr,
1850 #elif defined(CONFIG_CLONE_BACKWARDS3)
1851 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1853 int __user *, parent_tidptr,
1854 int __user *, child_tidptr,
1857 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1858 int __user *, parent_tidptr,
1859 int __user *, child_tidptr,
1863 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1867 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1868 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1871 static void sighand_ctor(void *data)
1873 struct sighand_struct *sighand = data;
1875 spin_lock_init(&sighand->siglock);
1876 init_waitqueue_head(&sighand->signalfd_wqh);
1879 void __init proc_caches_init(void)
1881 sighand_cachep = kmem_cache_create("sighand_cache",
1882 sizeof(struct sighand_struct), 0,
1883 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1884 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1885 signal_cachep = kmem_cache_create("signal_cache",
1886 sizeof(struct signal_struct), 0,
1887 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1889 files_cachep = kmem_cache_create("files_cache",
1890 sizeof(struct files_struct), 0,
1891 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1893 fs_cachep = kmem_cache_create("fs_cache",
1894 sizeof(struct fs_struct), 0,
1895 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1898 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1899 * whole struct cpumask for the OFFSTACK case. We could change
1900 * this to *only* allocate as much of it as required by the
1901 * maximum number of CPU's we can ever have. The cpumask_allocation
1902 * is at the end of the structure, exactly for that reason.
1904 mm_cachep = kmem_cache_create("mm_struct",
1905 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1906 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1908 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1910 nsproxy_cache_init();
1914 * Check constraints on flags passed to the unshare system call.
1916 static int check_unshare_flags(unsigned long unshare_flags)
1918 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1919 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1920 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1921 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1924 * Not implemented, but pretend it works if there is nothing
1925 * to unshare. Note that unsharing the address space or the
1926 * signal handlers also need to unshare the signal queues (aka
1929 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1930 if (!thread_group_empty(current))
1933 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1934 if (atomic_read(¤t->sighand->count) > 1)
1937 if (unshare_flags & CLONE_VM) {
1938 if (!current_is_single_threaded())
1946 * Unshare the filesystem structure if it is being shared
1948 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1950 struct fs_struct *fs = current->fs;
1952 if (!(unshare_flags & CLONE_FS) || !fs)
1955 /* don't need lock here; in the worst case we'll do useless copy */
1959 *new_fsp = copy_fs_struct(fs);
1967 * Unshare file descriptor table if it is being shared
1969 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1971 struct files_struct *fd = current->files;
1974 if ((unshare_flags & CLONE_FILES) &&
1975 (fd && atomic_read(&fd->count) > 1)) {
1976 *new_fdp = dup_fd(fd, &error);
1985 * unshare allows a process to 'unshare' part of the process
1986 * context which was originally shared using clone. copy_*
1987 * functions used by do_fork() cannot be used here directly
1988 * because they modify an inactive task_struct that is being
1989 * constructed. Here we are modifying the current, active,
1992 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1994 struct fs_struct *fs, *new_fs = NULL;
1995 struct files_struct *fd, *new_fd = NULL;
1996 struct cred *new_cred = NULL;
1997 struct nsproxy *new_nsproxy = NULL;
2002 * If unsharing a user namespace must also unshare the thread group
2003 * and unshare the filesystem root and working directories.
2005 if (unshare_flags & CLONE_NEWUSER)
2006 unshare_flags |= CLONE_THREAD | CLONE_FS;
2008 * If unsharing vm, must also unshare signal handlers.
2010 if (unshare_flags & CLONE_VM)
2011 unshare_flags |= CLONE_SIGHAND;
2013 * If unsharing a signal handlers, must also unshare the signal queues.
2015 if (unshare_flags & CLONE_SIGHAND)
2016 unshare_flags |= CLONE_THREAD;
2018 * If unsharing namespace, must also unshare filesystem information.
2020 if (unshare_flags & CLONE_NEWNS)
2021 unshare_flags |= CLONE_FS;
2023 err = check_unshare_flags(unshare_flags);
2025 goto bad_unshare_out;
2027 * CLONE_NEWIPC must also detach from the undolist: after switching
2028 * to a new ipc namespace, the semaphore arrays from the old
2029 * namespace are unreachable.
2031 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2033 err = unshare_fs(unshare_flags, &new_fs);
2035 goto bad_unshare_out;
2036 err = unshare_fd(unshare_flags, &new_fd);
2038 goto bad_unshare_cleanup_fs;
2039 err = unshare_userns(unshare_flags, &new_cred);
2041 goto bad_unshare_cleanup_fd;
2042 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2045 goto bad_unshare_cleanup_cred;
2047 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2050 * CLONE_SYSVSEM is equivalent to sys_exit().
2054 if (unshare_flags & CLONE_NEWIPC) {
2055 /* Orphan segments in old ns (see sem above). */
2057 shm_init_task(current);
2061 switch_task_namespaces(current, new_nsproxy);
2067 spin_lock(&fs->lock);
2068 current->fs = new_fs;
2073 spin_unlock(&fs->lock);
2077 fd = current->files;
2078 current->files = new_fd;
2082 task_unlock(current);
2085 /* Install the new user namespace */
2086 commit_creds(new_cred);
2091 bad_unshare_cleanup_cred:
2094 bad_unshare_cleanup_fd:
2096 put_files_struct(new_fd);
2098 bad_unshare_cleanup_fs:
2100 free_fs_struct(new_fs);
2107 * Helper to unshare the files of the current task.
2108 * We don't want to expose copy_files internals to
2109 * the exec layer of the kernel.
2112 int unshare_files(struct files_struct **displaced)
2114 struct task_struct *task = current;
2115 struct files_struct *copy = NULL;
2118 error = unshare_fd(CLONE_FILES, ©);
2119 if (error || !copy) {
2123 *displaced = task->files;
2130 int sysctl_max_threads(struct ctl_table *table, int write,
2131 void __user *buffer, size_t *lenp, loff_t *ppos)
2135 int threads = max_threads;
2136 int min = MIN_THREADS;
2137 int max = MAX_THREADS;
2144 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2148 set_max_threads(threads);