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/kaiser.h>
62 #include <linux/delayacct.h>
63 #include <linux/taskstats_kern.h>
64 #include <linux/random.h>
65 #include <linux/tty.h>
66 #include <linux/blkdev.h>
67 #include <linux/fs_struct.h>
68 #include <linux/magic.h>
69 #include <linux/perf_event.h>
70 #include <linux/posix-timers.h>
71 #include <linux/user-return-notifier.h>
72 #include <linux/oom.h>
73 #include <linux/khugepaged.h>
74 #include <linux/signalfd.h>
75 #include <linux/uprobes.h>
76 #include <linux/aio.h>
77 #include <linux/compiler.h>
78 #include <linux/sysctl.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,
168 return page ? page_address(page) : NULL;
171 static inline void free_thread_info(struct thread_info *ti)
173 kaiser_unmap_thread_stack(ti);
174 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
177 static struct kmem_cache *thread_info_cache;
179 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
182 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
185 static void free_thread_info(struct thread_info *ti)
187 kmem_cache_free(thread_info_cache, ti);
190 void thread_info_cache_init(void)
192 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
193 THREAD_SIZE, 0, NULL);
194 BUG_ON(thread_info_cache == NULL);
199 /* SLAB cache for signal_struct structures (tsk->signal) */
200 static struct kmem_cache *signal_cachep;
202 /* SLAB cache for sighand_struct structures (tsk->sighand) */
203 struct kmem_cache *sighand_cachep;
205 /* SLAB cache for files_struct structures (tsk->files) */
206 struct kmem_cache *files_cachep;
208 /* SLAB cache for fs_struct structures (tsk->fs) */
209 struct kmem_cache *fs_cachep;
211 /* SLAB cache for vm_area_struct structures */
212 struct kmem_cache *vm_area_cachep;
214 /* SLAB cache for mm_struct structures (tsk->mm) */
215 static struct kmem_cache *mm_cachep;
217 static void account_kernel_stack(struct thread_info *ti, int account)
219 struct zone *zone = page_zone(virt_to_page(ti));
221 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
224 void free_task(struct task_struct *tsk)
226 account_kernel_stack(tsk->stack, -1);
227 arch_release_thread_info(tsk->stack);
228 free_thread_info(tsk->stack);
229 rt_mutex_debug_task_free(tsk);
230 ftrace_graph_exit_task(tsk);
231 put_seccomp_filter(tsk);
232 arch_release_task_struct(tsk);
233 free_task_struct(tsk);
235 EXPORT_SYMBOL(free_task);
237 static inline void free_signal_struct(struct signal_struct *sig)
239 taskstats_tgid_free(sig);
240 sched_autogroup_exit(sig);
241 kmem_cache_free(signal_cachep, sig);
244 static inline void put_signal_struct(struct signal_struct *sig)
246 if (atomic_dec_and_test(&sig->sigcnt))
247 free_signal_struct(sig);
250 void __put_task_struct(struct task_struct *tsk)
252 WARN_ON(!tsk->exit_state);
253 WARN_ON(atomic_read(&tsk->usage));
254 WARN_ON(tsk == current);
257 task_numa_free(tsk, true);
258 security_task_free(tsk);
260 delayacct_tsk_free(tsk);
261 put_signal_struct(tsk->signal);
263 if (!profile_handoff_task(tsk))
266 EXPORT_SYMBOL_GPL(__put_task_struct);
268 void __init __weak arch_task_cache_init(void) { }
273 static void set_max_threads(unsigned int max_threads_suggested)
278 * The number of threads shall be limited such that the thread
279 * structures may only consume a small part of the available memory.
281 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
282 threads = MAX_THREADS;
284 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
285 (u64) THREAD_SIZE * 8UL);
287 if (threads > max_threads_suggested)
288 threads = max_threads_suggested;
290 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
293 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
294 /* Initialized by the architecture: */
295 int arch_task_struct_size __read_mostly;
298 void __init fork_init(void)
300 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
301 #ifndef ARCH_MIN_TASKALIGN
302 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
304 /* create a slab on which task_structs can be allocated */
306 kmem_cache_create("task_struct", arch_task_struct_size,
307 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
310 /* do the arch specific task caches init */
311 arch_task_cache_init();
313 set_max_threads(MAX_THREADS);
315 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
316 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
317 init_task.signal->rlim[RLIMIT_SIGPENDING] =
318 init_task.signal->rlim[RLIMIT_NPROC];
321 int __weak arch_dup_task_struct(struct task_struct *dst,
322 struct task_struct *src)
328 void set_task_stack_end_magic(struct task_struct *tsk)
330 unsigned long *stackend;
332 stackend = end_of_stack(tsk);
333 *stackend = STACK_END_MAGIC; /* for overflow detection */
336 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
338 struct task_struct *tsk;
339 struct thread_info *ti;
342 if (node == NUMA_NO_NODE)
343 node = tsk_fork_get_node(orig);
344 tsk = alloc_task_struct_node(node);
348 ti = alloc_thread_info_node(tsk, node);
352 err = arch_dup_task_struct(tsk, orig);
358 err = kaiser_map_thread_stack(tsk->stack);
361 #ifdef CONFIG_SECCOMP
363 * We must handle setting up seccomp filters once we're under
364 * the sighand lock in case orig has changed between now and
365 * then. Until then, filter must be NULL to avoid messing up
366 * the usage counts on the error path calling free_task.
368 tsk->seccomp.filter = NULL;
371 setup_thread_stack(tsk, orig);
372 clear_user_return_notifier(tsk);
373 clear_tsk_need_resched(tsk);
374 set_task_stack_end_magic(tsk);
376 #ifdef CONFIG_CC_STACKPROTECTOR
377 tsk->stack_canary = get_random_long();
381 * One for us, one for whoever does the "release_task()" (usually
384 atomic_set(&tsk->usage, 2);
385 #ifdef CONFIG_BLK_DEV_IO_TRACE
388 tsk->splice_pipe = NULL;
389 tsk->task_frag.page = NULL;
390 tsk->wake_q.next = NULL;
392 account_kernel_stack(ti, 1);
397 free_thread_info(ti);
399 free_task_struct(tsk);
404 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
406 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
407 struct rb_node **rb_link, *rb_parent;
409 unsigned long charge;
411 uprobe_start_dup_mmap();
412 down_write(&oldmm->mmap_sem);
413 flush_cache_dup_mm(oldmm);
414 uprobe_dup_mmap(oldmm, mm);
416 * Not linked in yet - no deadlock potential:
418 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
420 /* No ordering required: file already has been exposed. */
421 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
423 mm->total_vm = oldmm->total_vm;
424 mm->shared_vm = oldmm->shared_vm;
425 mm->exec_vm = oldmm->exec_vm;
426 mm->stack_vm = oldmm->stack_vm;
428 rb_link = &mm->mm_rb.rb_node;
431 retval = ksm_fork(mm, oldmm);
434 retval = khugepaged_fork(mm, oldmm);
439 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
442 if (mpnt->vm_flags & VM_DONTCOPY) {
443 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
448 if (mpnt->vm_flags & VM_ACCOUNT) {
449 unsigned long len = vma_pages(mpnt);
451 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
455 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
459 INIT_LIST_HEAD(&tmp->anon_vma_chain);
460 retval = vma_dup_policy(mpnt, tmp);
462 goto fail_nomem_policy;
464 if (anon_vma_fork(tmp, mpnt))
465 goto fail_nomem_anon_vma_fork;
467 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
468 tmp->vm_next = tmp->vm_prev = NULL;
469 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
472 struct inode *inode = file_inode(file);
473 struct address_space *mapping = file->f_mapping;
476 if (tmp->vm_flags & VM_DENYWRITE)
477 atomic_dec(&inode->i_writecount);
478 i_mmap_lock_write(mapping);
479 if (tmp->vm_flags & VM_SHARED)
480 atomic_inc(&mapping->i_mmap_writable);
481 flush_dcache_mmap_lock(mapping);
482 /* insert tmp into the share list, just after mpnt */
483 vma_interval_tree_insert_after(tmp, mpnt,
485 flush_dcache_mmap_unlock(mapping);
486 i_mmap_unlock_write(mapping);
490 * Clear hugetlb-related page reserves for children. This only
491 * affects MAP_PRIVATE mappings. Faults generated by the child
492 * are not guaranteed to succeed, even if read-only
494 if (is_vm_hugetlb_page(tmp))
495 reset_vma_resv_huge_pages(tmp);
498 * Link in the new vma and copy the page table entries.
501 pprev = &tmp->vm_next;
505 __vma_link_rb(mm, tmp, rb_link, rb_parent);
506 rb_link = &tmp->vm_rb.rb_right;
507 rb_parent = &tmp->vm_rb;
510 retval = copy_page_range(mm, oldmm, mpnt);
512 if (tmp->vm_ops && tmp->vm_ops->open)
513 tmp->vm_ops->open(tmp);
518 /* a new mm has just been created */
519 arch_dup_mmap(oldmm, mm);
522 up_write(&mm->mmap_sem);
524 up_write(&oldmm->mmap_sem);
525 uprobe_end_dup_mmap();
527 fail_nomem_anon_vma_fork:
528 mpol_put(vma_policy(tmp));
530 kmem_cache_free(vm_area_cachep, tmp);
533 vm_unacct_memory(charge);
537 static inline int mm_alloc_pgd(struct mm_struct *mm)
539 mm->pgd = pgd_alloc(mm);
540 if (unlikely(!mm->pgd))
545 static inline void mm_free_pgd(struct mm_struct *mm)
547 pgd_free(mm, mm->pgd);
550 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
552 down_write(&oldmm->mmap_sem);
553 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
554 up_write(&oldmm->mmap_sem);
557 #define mm_alloc_pgd(mm) (0)
558 #define mm_free_pgd(mm)
559 #endif /* CONFIG_MMU */
561 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
563 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
564 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
566 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
568 static int __init coredump_filter_setup(char *s)
570 default_dump_filter =
571 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
572 MMF_DUMP_FILTER_MASK;
576 __setup("coredump_filter=", coredump_filter_setup);
578 #include <linux/init_task.h>
580 static void mm_init_aio(struct mm_struct *mm)
583 spin_lock_init(&mm->ioctx_lock);
584 mm->ioctx_table = NULL;
588 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
595 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
596 struct user_namespace *user_ns)
600 mm->vmacache_seqnum = 0;
601 atomic_set(&mm->mm_users, 1);
602 atomic_set(&mm->mm_count, 1);
603 init_rwsem(&mm->mmap_sem);
604 INIT_LIST_HEAD(&mm->mmlist);
605 mm->core_state = NULL;
606 atomic_long_set(&mm->nr_ptes, 0);
611 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
612 spin_lock_init(&mm->page_table_lock);
615 mm_init_owner(mm, p);
616 mmu_notifier_mm_init(mm);
617 clear_tlb_flush_pending(mm);
618 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
619 mm->pmd_huge_pte = NULL;
623 mm->flags = current->mm->flags & MMF_INIT_MASK;
624 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
626 mm->flags = default_dump_filter;
630 if (mm_alloc_pgd(mm))
633 if (init_new_context(p, mm))
636 mm->user_ns = get_user_ns(user_ns);
646 static void check_mm(struct mm_struct *mm)
650 for (i = 0; i < NR_MM_COUNTERS; i++) {
651 long x = atomic_long_read(&mm->rss_stat.count[i]);
654 printk(KERN_ALERT "BUG: Bad rss-counter state "
655 "mm:%p idx:%d val:%ld\n", mm, i, x);
658 if (atomic_long_read(&mm->nr_ptes))
659 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
660 atomic_long_read(&mm->nr_ptes));
662 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
665 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
666 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
671 * Allocate and initialize an mm_struct.
673 struct mm_struct *mm_alloc(void)
675 struct mm_struct *mm;
681 memset(mm, 0, sizeof(*mm));
682 return mm_init(mm, current, current_user_ns());
686 * Called when the last reference to the mm
687 * is dropped: either by a lazy thread or by
688 * mmput. Free the page directory and the mm.
690 void __mmdrop(struct mm_struct *mm)
692 BUG_ON(mm == &init_mm);
695 mmu_notifier_mm_destroy(mm);
697 put_user_ns(mm->user_ns);
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 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
892 uprobe_free_utask(tsk);
894 /* Get rid of any cached register state */
895 deactivate_mm(tsk, mm);
898 * Signal userspace if we're not exiting with a core dump
899 * because we want to leave the value intact for debugging
902 if (tsk->clear_child_tid) {
903 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
904 atomic_read(&mm->mm_users) > 1) {
906 * We don't check the error code - if userspace has
907 * not set up a proper pointer then tough luck.
909 put_user(0, tsk->clear_child_tid);
910 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
913 tsk->clear_child_tid = NULL;
917 * All done, finally we can wake up parent and return this mm to him.
918 * Also kthread_stop() uses this completion for synchronization.
921 complete_vfork_done(tsk);
924 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
926 futex_exit_release(tsk);
930 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
932 futex_exec_release(tsk);
937 * Allocate a new mm structure and copy contents from the
938 * mm structure of the passed in task structure.
940 static struct mm_struct *dup_mm(struct task_struct *tsk)
942 struct mm_struct *mm, *oldmm = current->mm;
949 memcpy(mm, oldmm, sizeof(*mm));
951 if (!mm_init(mm, tsk, mm->user_ns))
954 err = dup_mmap(mm, oldmm);
958 mm->hiwater_rss = get_mm_rss(mm);
959 mm->hiwater_vm = mm->total_vm;
961 if (mm->binfmt && !try_module_get(mm->binfmt->module))
967 /* don't put binfmt in mmput, we haven't got module yet */
975 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
977 struct mm_struct *mm, *oldmm;
980 tsk->min_flt = tsk->maj_flt = 0;
981 tsk->nvcsw = tsk->nivcsw = 0;
982 #ifdef CONFIG_DETECT_HUNG_TASK
983 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
987 tsk->active_mm = NULL;
990 * Are we cloning a kernel thread?
992 * We need to steal a active VM for that..
998 /* initialize the new vmacache entries */
1001 if (clone_flags & CLONE_VM) {
1002 atomic_inc(&oldmm->mm_users);
1014 tsk->active_mm = mm;
1021 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1023 struct fs_struct *fs = current->fs;
1024 if (clone_flags & CLONE_FS) {
1025 /* tsk->fs is already what we want */
1026 spin_lock(&fs->lock);
1028 spin_unlock(&fs->lock);
1032 spin_unlock(&fs->lock);
1035 tsk->fs = copy_fs_struct(fs);
1041 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1043 struct files_struct *oldf, *newf;
1047 * A background process may not have any files ...
1049 oldf = current->files;
1053 if (clone_flags & CLONE_FILES) {
1054 atomic_inc(&oldf->count);
1058 newf = dup_fd(oldf, &error);
1068 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1071 struct io_context *ioc = current->io_context;
1072 struct io_context *new_ioc;
1077 * Share io context with parent, if CLONE_IO is set
1079 if (clone_flags & CLONE_IO) {
1081 tsk->io_context = ioc;
1082 } else if (ioprio_valid(ioc->ioprio)) {
1083 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1084 if (unlikely(!new_ioc))
1087 new_ioc->ioprio = ioc->ioprio;
1088 put_io_context(new_ioc);
1094 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1096 struct sighand_struct *sig;
1098 if (clone_flags & CLONE_SIGHAND) {
1099 atomic_inc(¤t->sighand->count);
1102 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1103 rcu_assign_pointer(tsk->sighand, sig);
1107 atomic_set(&sig->count, 1);
1108 spin_lock_irq(¤t->sighand->siglock);
1109 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1110 spin_unlock_irq(¤t->sighand->siglock);
1114 void __cleanup_sighand(struct sighand_struct *sighand)
1116 if (atomic_dec_and_test(&sighand->count)) {
1117 signalfd_cleanup(sighand);
1119 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1120 * without an RCU grace period, see __lock_task_sighand().
1122 kmem_cache_free(sighand_cachep, sighand);
1127 * Initialize POSIX timer handling for a thread group.
1129 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1131 unsigned long cpu_limit;
1133 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1134 if (cpu_limit != RLIM_INFINITY) {
1135 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1136 sig->cputimer.running = true;
1139 /* The timer lists. */
1140 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1141 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1142 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1145 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1147 struct signal_struct *sig;
1149 if (clone_flags & CLONE_THREAD)
1152 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1157 sig->nr_threads = 1;
1158 atomic_set(&sig->live, 1);
1159 atomic_set(&sig->sigcnt, 1);
1161 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1162 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1163 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1165 init_waitqueue_head(&sig->wait_chldexit);
1166 sig->curr_target = tsk;
1167 init_sigpending(&sig->shared_pending);
1168 INIT_LIST_HEAD(&sig->posix_timers);
1169 seqlock_init(&sig->stats_lock);
1170 prev_cputime_init(&sig->prev_cputime);
1172 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1173 sig->real_timer.function = it_real_fn;
1175 task_lock(current->group_leader);
1176 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1177 task_unlock(current->group_leader);
1179 posix_cpu_timers_init_group(sig);
1181 tty_audit_fork(sig);
1182 sched_autogroup_fork(sig);
1184 sig->oom_score_adj = current->signal->oom_score_adj;
1185 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1187 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1188 current->signal->is_child_subreaper;
1190 mutex_init(&sig->cred_guard_mutex);
1195 static void copy_seccomp(struct task_struct *p)
1197 #ifdef CONFIG_SECCOMP
1199 * Must be called with sighand->lock held, which is common to
1200 * all threads in the group. Holding cred_guard_mutex is not
1201 * needed because this new task is not yet running and cannot
1204 assert_spin_locked(¤t->sighand->siglock);
1206 /* Ref-count the new filter user, and assign it. */
1207 get_seccomp_filter(current);
1208 p->seccomp = current->seccomp;
1211 * Explicitly enable no_new_privs here in case it got set
1212 * between the task_struct being duplicated and holding the
1213 * sighand lock. The seccomp state and nnp must be in sync.
1215 if (task_no_new_privs(current))
1216 task_set_no_new_privs(p);
1219 * If the parent gained a seccomp mode after copying thread
1220 * flags and between before we held the sighand lock, we have
1221 * to manually enable the seccomp thread flag here.
1223 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1224 set_tsk_thread_flag(p, TIF_SECCOMP);
1228 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1230 current->clear_child_tid = tidptr;
1232 return task_pid_vnr(current);
1235 static void rt_mutex_init_task(struct task_struct *p)
1237 raw_spin_lock_init(&p->pi_lock);
1238 #ifdef CONFIG_RT_MUTEXES
1239 p->pi_waiters = RB_ROOT;
1240 p->pi_waiters_leftmost = NULL;
1241 p->pi_blocked_on = NULL;
1246 * Initialize POSIX timer handling for a single task.
1248 static void posix_cpu_timers_init(struct task_struct *tsk)
1250 tsk->cputime_expires.prof_exp = 0;
1251 tsk->cputime_expires.virt_exp = 0;
1252 tsk->cputime_expires.sched_exp = 0;
1253 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1254 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1255 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1259 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1261 task->pids[type].pid = pid;
1265 * This creates a new process as a copy of the old one,
1266 * but does not actually start it yet.
1268 * It copies the registers, and all the appropriate
1269 * parts of the process environment (as per the clone
1270 * flags). The actual kick-off is left to the caller.
1272 static struct task_struct *copy_process(unsigned long clone_flags,
1273 unsigned long stack_start,
1274 unsigned long stack_size,
1275 int __user *child_tidptr,
1282 struct task_struct *p;
1283 void *cgrp_ss_priv[CGROUP_CANFORK_COUNT] = {};
1285 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1286 return ERR_PTR(-EINVAL);
1288 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1289 return ERR_PTR(-EINVAL);
1292 * Thread groups must share signals as well, and detached threads
1293 * can only be started up within the thread group.
1295 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1296 return ERR_PTR(-EINVAL);
1299 * Shared signal handlers imply shared VM. By way of the above,
1300 * thread groups also imply shared VM. Blocking this case allows
1301 * for various simplifications in other code.
1303 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1304 return ERR_PTR(-EINVAL);
1307 * Siblings of global init remain as zombies on exit since they are
1308 * not reaped by their parent (swapper). To solve this and to avoid
1309 * multi-rooted process trees, prevent global and container-inits
1310 * from creating siblings.
1312 if ((clone_flags & CLONE_PARENT) &&
1313 current->signal->flags & SIGNAL_UNKILLABLE)
1314 return ERR_PTR(-EINVAL);
1317 * If the new process will be in a different pid or user namespace
1318 * do not allow it to share a thread group with the forking task.
1320 if (clone_flags & CLONE_THREAD) {
1321 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1322 (task_active_pid_ns(current) !=
1323 current->nsproxy->pid_ns_for_children))
1324 return ERR_PTR(-EINVAL);
1327 retval = security_task_create(clone_flags);
1332 p = dup_task_struct(current, node);
1337 * This _must_ happen before we call free_task(), i.e. before we jump
1338 * to any of the bad_fork_* labels. This is to avoid freeing
1339 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1340 * kernel threads (PF_KTHREAD).
1342 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1344 * Clear TID on mm_release()?
1346 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1348 ftrace_graph_init_task(p);
1350 rt_mutex_init_task(p);
1352 #ifdef CONFIG_PROVE_LOCKING
1353 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1354 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1357 if (atomic_read(&p->real_cred->user->processes) >=
1358 task_rlimit(p, RLIMIT_NPROC)) {
1359 if (p->real_cred->user != INIT_USER &&
1360 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1363 current->flags &= ~PF_NPROC_EXCEEDED;
1365 retval = copy_creds(p, clone_flags);
1370 * If multiple threads are within copy_process(), then this check
1371 * triggers too late. This doesn't hurt, the check is only there
1372 * to stop root fork bombs.
1375 if (nr_threads >= max_threads)
1376 goto bad_fork_cleanup_count;
1378 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1379 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1380 p->flags |= PF_FORKNOEXEC;
1381 INIT_LIST_HEAD(&p->children);
1382 INIT_LIST_HEAD(&p->sibling);
1383 rcu_copy_process(p);
1384 p->vfork_done = NULL;
1385 spin_lock_init(&p->alloc_lock);
1387 init_sigpending(&p->pending);
1389 p->utime = p->stime = p->gtime = 0;
1390 p->utimescaled = p->stimescaled = 0;
1391 prev_cputime_init(&p->prev_cputime);
1393 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1394 seqlock_init(&p->vtime_seqlock);
1396 p->vtime_snap_whence = VTIME_SLEEPING;
1399 #if defined(SPLIT_RSS_COUNTING)
1400 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1403 p->default_timer_slack_ns = current->timer_slack_ns;
1405 task_io_accounting_init(&p->ioac);
1406 acct_clear_integrals(p);
1408 posix_cpu_timers_init(p);
1410 p->io_context = NULL;
1411 p->audit_context = NULL;
1414 p->mempolicy = mpol_dup(p->mempolicy);
1415 if (IS_ERR(p->mempolicy)) {
1416 retval = PTR_ERR(p->mempolicy);
1417 p->mempolicy = NULL;
1418 goto bad_fork_cleanup_threadgroup_lock;
1421 #ifdef CONFIG_CPUSETS
1422 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1423 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1424 seqcount_init(&p->mems_allowed_seq);
1426 #ifdef CONFIG_TRACE_IRQFLAGS
1428 p->hardirqs_enabled = 0;
1429 p->hardirq_enable_ip = 0;
1430 p->hardirq_enable_event = 0;
1431 p->hardirq_disable_ip = _THIS_IP_;
1432 p->hardirq_disable_event = 0;
1433 p->softirqs_enabled = 1;
1434 p->softirq_enable_ip = _THIS_IP_;
1435 p->softirq_enable_event = 0;
1436 p->softirq_disable_ip = 0;
1437 p->softirq_disable_event = 0;
1438 p->hardirq_context = 0;
1439 p->softirq_context = 0;
1442 p->pagefault_disabled = 0;
1444 #ifdef CONFIG_LOCKDEP
1445 p->lockdep_depth = 0; /* no locks held yet */
1446 p->curr_chain_key = 0;
1447 p->lockdep_recursion = 0;
1450 #ifdef CONFIG_DEBUG_MUTEXES
1451 p->blocked_on = NULL; /* not blocked yet */
1453 #ifdef CONFIG_BCACHE
1454 p->sequential_io = 0;
1455 p->sequential_io_avg = 0;
1458 /* Perform scheduler related setup. Assign this task to a CPU. */
1459 retval = sched_fork(clone_flags, p);
1461 goto bad_fork_cleanup_policy;
1463 retval = perf_event_init_task(p);
1465 goto bad_fork_cleanup_policy;
1466 retval = audit_alloc(p);
1468 goto bad_fork_cleanup_perf;
1469 /* copy all the process information */
1471 retval = copy_semundo(clone_flags, p);
1473 goto bad_fork_cleanup_audit;
1474 retval = copy_files(clone_flags, p);
1476 goto bad_fork_cleanup_semundo;
1477 retval = copy_fs(clone_flags, p);
1479 goto bad_fork_cleanup_files;
1480 retval = copy_sighand(clone_flags, p);
1482 goto bad_fork_cleanup_fs;
1483 retval = copy_signal(clone_flags, p);
1485 goto bad_fork_cleanup_sighand;
1486 retval = copy_mm(clone_flags, p);
1488 goto bad_fork_cleanup_signal;
1489 retval = copy_namespaces(clone_flags, p);
1491 goto bad_fork_cleanup_mm;
1492 retval = copy_io(clone_flags, p);
1494 goto bad_fork_cleanup_namespaces;
1495 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1497 goto bad_fork_cleanup_io;
1499 if (pid != &init_struct_pid) {
1500 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1502 retval = PTR_ERR(pid);
1503 goto bad_fork_cleanup_io;
1513 * sigaltstack should be cleared when sharing the same VM
1515 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1516 p->sas_ss_sp = p->sas_ss_size = 0;
1519 * Syscall tracing and stepping should be turned off in the
1520 * child regardless of CLONE_PTRACE.
1522 user_disable_single_step(p);
1523 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1524 #ifdef TIF_SYSCALL_EMU
1525 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1527 clear_all_latency_tracing(p);
1529 /* ok, now we should be set up.. */
1530 p->pid = pid_nr(pid);
1531 if (clone_flags & CLONE_THREAD) {
1532 p->group_leader = current->group_leader;
1533 p->tgid = current->tgid;
1535 p->group_leader = p;
1540 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1541 p->dirty_paused_when = 0;
1543 p->pdeath_signal = 0;
1544 INIT_LIST_HEAD(&p->thread_group);
1545 p->task_works = NULL;
1547 threadgroup_change_begin(current);
1549 * Ensure that the cgroup subsystem policies allow the new process to be
1550 * forked. It should be noted the the new process's css_set can be changed
1551 * between here and cgroup_post_fork() if an organisation operation is in
1554 retval = cgroup_can_fork(p, cgrp_ss_priv);
1556 goto bad_fork_free_pid;
1559 * From this point on we must avoid any synchronous user-space
1560 * communication until we take the tasklist-lock. In particular, we do
1561 * not want user-space to be able to predict the process start-time by
1562 * stalling fork(2) after we recorded the start_time but before it is
1563 * visible to the system.
1566 p->start_time = ktime_get_ns();
1567 p->real_start_time = ktime_get_boot_ns();
1570 * Make it visible to the rest of the system, but dont wake it up yet.
1571 * Need tasklist lock for parent etc handling!
1573 write_lock_irq(&tasklist_lock);
1575 /* CLONE_PARENT re-uses the old parent */
1576 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1577 p->real_parent = current->real_parent;
1578 p->parent_exec_id = current->parent_exec_id;
1579 if (clone_flags & CLONE_THREAD)
1580 p->exit_signal = -1;
1582 p->exit_signal = current->group_leader->exit_signal;
1584 p->real_parent = current;
1585 p->parent_exec_id = current->self_exec_id;
1586 p->exit_signal = (clone_flags & CSIGNAL);
1589 spin_lock(¤t->sighand->siglock);
1592 * Copy seccomp details explicitly here, in case they were changed
1593 * before holding sighand lock.
1598 * Process group and session signals need to be delivered to just the
1599 * parent before the fork or both the parent and the child after the
1600 * fork. Restart if a signal comes in before we add the new process to
1601 * it's process group.
1602 * A fatal signal pending means that current will exit, so the new
1603 * thread can't slip out of an OOM kill (or normal SIGKILL).
1605 recalc_sigpending();
1606 if (signal_pending(current)) {
1607 retval = -ERESTARTNOINTR;
1608 goto bad_fork_cancel_cgroup;
1610 if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1612 goto bad_fork_cancel_cgroup;
1615 if (likely(p->pid)) {
1616 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1618 init_task_pid(p, PIDTYPE_PID, pid);
1619 if (thread_group_leader(p)) {
1620 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1621 init_task_pid(p, PIDTYPE_SID, task_session(current));
1623 if (is_child_reaper(pid)) {
1624 ns_of_pid(pid)->child_reaper = p;
1625 p->signal->flags |= SIGNAL_UNKILLABLE;
1628 p->signal->leader_pid = pid;
1629 p->signal->tty = tty_kref_get(current->signal->tty);
1630 list_add_tail(&p->sibling, &p->real_parent->children);
1631 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1632 attach_pid(p, PIDTYPE_PGID);
1633 attach_pid(p, PIDTYPE_SID);
1634 __this_cpu_inc(process_counts);
1636 current->signal->nr_threads++;
1637 atomic_inc(¤t->signal->live);
1638 atomic_inc(¤t->signal->sigcnt);
1639 list_add_tail_rcu(&p->thread_group,
1640 &p->group_leader->thread_group);
1641 list_add_tail_rcu(&p->thread_node,
1642 &p->signal->thread_head);
1644 attach_pid(p, PIDTYPE_PID);
1649 spin_unlock(¤t->sighand->siglock);
1650 syscall_tracepoint_update(p);
1651 write_unlock_irq(&tasklist_lock);
1653 proc_fork_connector(p);
1654 cgroup_post_fork(p, cgrp_ss_priv);
1655 threadgroup_change_end(current);
1658 trace_task_newtask(p, clone_flags);
1659 uprobe_copy_process(p, clone_flags);
1663 bad_fork_cancel_cgroup:
1664 spin_unlock(¤t->sighand->siglock);
1665 write_unlock_irq(&tasklist_lock);
1666 cgroup_cancel_fork(p, cgrp_ss_priv);
1668 threadgroup_change_end(current);
1669 if (pid != &init_struct_pid)
1671 bad_fork_cleanup_io:
1674 bad_fork_cleanup_namespaces:
1675 exit_task_namespaces(p);
1676 bad_fork_cleanup_mm:
1679 bad_fork_cleanup_signal:
1680 if (!(clone_flags & CLONE_THREAD))
1681 free_signal_struct(p->signal);
1682 bad_fork_cleanup_sighand:
1683 __cleanup_sighand(p->sighand);
1684 bad_fork_cleanup_fs:
1685 exit_fs(p); /* blocking */
1686 bad_fork_cleanup_files:
1687 exit_files(p); /* blocking */
1688 bad_fork_cleanup_semundo:
1690 bad_fork_cleanup_audit:
1692 bad_fork_cleanup_perf:
1693 perf_event_free_task(p);
1694 bad_fork_cleanup_policy:
1696 mpol_put(p->mempolicy);
1697 bad_fork_cleanup_threadgroup_lock:
1699 delayacct_tsk_free(p);
1700 bad_fork_cleanup_count:
1701 atomic_dec(&p->cred->user->processes);
1706 return ERR_PTR(retval);
1709 static inline void init_idle_pids(struct pid_link *links)
1713 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1714 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1715 links[type].pid = &init_struct_pid;
1719 struct task_struct *fork_idle(int cpu)
1721 struct task_struct *task;
1722 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1724 if (!IS_ERR(task)) {
1725 init_idle_pids(task->pids);
1726 init_idle(task, cpu);
1733 * Ok, this is the main fork-routine.
1735 * It copies the process, and if successful kick-starts
1736 * it and waits for it to finish using the VM if required.
1738 long _do_fork(unsigned long clone_flags,
1739 unsigned long stack_start,
1740 unsigned long stack_size,
1741 int __user *parent_tidptr,
1742 int __user *child_tidptr,
1745 struct task_struct *p;
1750 * Determine whether and which event to report to ptracer. When
1751 * called from kernel_thread or CLONE_UNTRACED is explicitly
1752 * requested, no event is reported; otherwise, report if the event
1753 * for the type of forking is enabled.
1755 if (!(clone_flags & CLONE_UNTRACED)) {
1756 if (clone_flags & CLONE_VFORK)
1757 trace = PTRACE_EVENT_VFORK;
1758 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1759 trace = PTRACE_EVENT_CLONE;
1761 trace = PTRACE_EVENT_FORK;
1763 if (likely(!ptrace_event_enabled(current, trace)))
1767 p = copy_process(clone_flags, stack_start, stack_size,
1768 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1770 * Do this prior waking up the new thread - the thread pointer
1771 * might get invalid after that point, if the thread exits quickly.
1774 struct completion vfork;
1777 trace_sched_process_fork(current, p);
1779 pid = get_task_pid(p, PIDTYPE_PID);
1782 if (clone_flags & CLONE_PARENT_SETTID)
1783 put_user(nr, parent_tidptr);
1785 if (clone_flags & CLONE_VFORK) {
1786 p->vfork_done = &vfork;
1787 init_completion(&vfork);
1791 wake_up_new_task(p);
1793 /* forking complete and child started to run, tell ptracer */
1794 if (unlikely(trace))
1795 ptrace_event_pid(trace, pid);
1797 if (clone_flags & CLONE_VFORK) {
1798 if (!wait_for_vfork_done(p, &vfork))
1799 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1809 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1810 /* For compatibility with architectures that call do_fork directly rather than
1811 * using the syscall entry points below. */
1812 long do_fork(unsigned long clone_flags,
1813 unsigned long stack_start,
1814 unsigned long stack_size,
1815 int __user *parent_tidptr,
1816 int __user *child_tidptr)
1818 return _do_fork(clone_flags, stack_start, stack_size,
1819 parent_tidptr, child_tidptr, 0);
1824 * Create a kernel thread.
1826 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1828 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1829 (unsigned long)arg, NULL, NULL, 0);
1832 #ifdef __ARCH_WANT_SYS_FORK
1833 SYSCALL_DEFINE0(fork)
1836 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1838 /* can not support in nommu mode */
1844 #ifdef __ARCH_WANT_SYS_VFORK
1845 SYSCALL_DEFINE0(vfork)
1847 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1852 #ifdef __ARCH_WANT_SYS_CLONE
1853 #ifdef CONFIG_CLONE_BACKWARDS
1854 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1855 int __user *, parent_tidptr,
1857 int __user *, child_tidptr)
1858 #elif defined(CONFIG_CLONE_BACKWARDS2)
1859 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1860 int __user *, parent_tidptr,
1861 int __user *, child_tidptr,
1863 #elif defined(CONFIG_CLONE_BACKWARDS3)
1864 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1866 int __user *, parent_tidptr,
1867 int __user *, child_tidptr,
1870 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1871 int __user *, parent_tidptr,
1872 int __user *, child_tidptr,
1876 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1880 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1881 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1884 static void sighand_ctor(void *data)
1886 struct sighand_struct *sighand = data;
1888 spin_lock_init(&sighand->siglock);
1889 init_waitqueue_head(&sighand->signalfd_wqh);
1892 void __init proc_caches_init(void)
1894 sighand_cachep = kmem_cache_create("sighand_cache",
1895 sizeof(struct sighand_struct), 0,
1896 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1897 SLAB_NOTRACK, sighand_ctor);
1898 signal_cachep = kmem_cache_create("signal_cache",
1899 sizeof(struct signal_struct), 0,
1900 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1901 files_cachep = kmem_cache_create("files_cache",
1902 sizeof(struct files_struct), 0,
1903 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1904 fs_cachep = kmem_cache_create("fs_cache",
1905 sizeof(struct fs_struct), 0,
1906 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1908 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1909 * whole struct cpumask for the OFFSTACK case. We could change
1910 * this to *only* allocate as much of it as required by the
1911 * maximum number of CPU's we can ever have. The cpumask_allocation
1912 * is at the end of the structure, exactly for that reason.
1914 mm_cachep = kmem_cache_create("mm_struct",
1915 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1916 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1917 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1919 nsproxy_cache_init();
1923 * Check constraints on flags passed to the unshare system call.
1925 static int check_unshare_flags(unsigned long unshare_flags)
1927 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1928 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1929 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1930 CLONE_NEWUSER|CLONE_NEWPID))
1933 * Not implemented, but pretend it works if there is nothing
1934 * to unshare. Note that unsharing the address space or the
1935 * signal handlers also need to unshare the signal queues (aka
1938 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1939 if (!thread_group_empty(current))
1942 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1943 if (atomic_read(¤t->sighand->count) > 1)
1946 if (unshare_flags & CLONE_VM) {
1947 if (!current_is_single_threaded())
1955 * Unshare the filesystem structure if it is being shared
1957 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1959 struct fs_struct *fs = current->fs;
1961 if (!(unshare_flags & CLONE_FS) || !fs)
1964 /* don't need lock here; in the worst case we'll do useless copy */
1968 *new_fsp = copy_fs_struct(fs);
1976 * Unshare file descriptor table if it is being shared
1978 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1980 struct files_struct *fd = current->files;
1983 if ((unshare_flags & CLONE_FILES) &&
1984 (fd && atomic_read(&fd->count) > 1)) {
1985 *new_fdp = dup_fd(fd, &error);
1994 * unshare allows a process to 'unshare' part of the process
1995 * context which was originally shared using clone. copy_*
1996 * functions used by do_fork() cannot be used here directly
1997 * because they modify an inactive task_struct that is being
1998 * constructed. Here we are modifying the current, active,
2001 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2003 struct fs_struct *fs, *new_fs = NULL;
2004 struct files_struct *fd, *new_fd = NULL;
2005 struct cred *new_cred = NULL;
2006 struct nsproxy *new_nsproxy = NULL;
2011 * If unsharing a user namespace must also unshare the thread group
2012 * and unshare the filesystem root and working directories.
2014 if (unshare_flags & CLONE_NEWUSER)
2015 unshare_flags |= CLONE_THREAD | CLONE_FS;
2017 * If unsharing vm, must also unshare signal handlers.
2019 if (unshare_flags & CLONE_VM)
2020 unshare_flags |= CLONE_SIGHAND;
2022 * If unsharing a signal handlers, must also unshare the signal queues.
2024 if (unshare_flags & CLONE_SIGHAND)
2025 unshare_flags |= CLONE_THREAD;
2027 * If unsharing namespace, must also unshare filesystem information.
2029 if (unshare_flags & CLONE_NEWNS)
2030 unshare_flags |= CLONE_FS;
2032 err = check_unshare_flags(unshare_flags);
2034 goto bad_unshare_out;
2036 * CLONE_NEWIPC must also detach from the undolist: after switching
2037 * to a new ipc namespace, the semaphore arrays from the old
2038 * namespace are unreachable.
2040 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2042 err = unshare_fs(unshare_flags, &new_fs);
2044 goto bad_unshare_out;
2045 err = unshare_fd(unshare_flags, &new_fd);
2047 goto bad_unshare_cleanup_fs;
2048 err = unshare_userns(unshare_flags, &new_cred);
2050 goto bad_unshare_cleanup_fd;
2051 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2054 goto bad_unshare_cleanup_cred;
2056 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2059 * CLONE_SYSVSEM is equivalent to sys_exit().
2063 if (unshare_flags & CLONE_NEWIPC) {
2064 /* Orphan segments in old ns (see sem above). */
2066 shm_init_task(current);
2070 switch_task_namespaces(current, new_nsproxy);
2076 spin_lock(&fs->lock);
2077 current->fs = new_fs;
2082 spin_unlock(&fs->lock);
2086 fd = current->files;
2087 current->files = new_fd;
2091 task_unlock(current);
2094 /* Install the new user namespace */
2095 commit_creds(new_cred);
2100 bad_unshare_cleanup_cred:
2103 bad_unshare_cleanup_fd:
2105 put_files_struct(new_fd);
2107 bad_unshare_cleanup_fs:
2109 free_fs_struct(new_fs);
2116 * Helper to unshare the files of the current task.
2117 * We don't want to expose copy_files internals to
2118 * the exec layer of the kernel.
2121 int unshare_files(struct files_struct **displaced)
2123 struct task_struct *task = current;
2124 struct files_struct *copy = NULL;
2127 error = unshare_fd(CLONE_FILES, ©);
2128 if (error || !copy) {
2132 *displaced = task->files;
2139 int sysctl_max_threads(struct ctl_table *table, int write,
2140 void __user *buffer, size_t *lenp, loff_t *ppos)
2144 int threads = max_threads;
2146 int max = MAX_THREADS;
2153 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2157 max_threads = threads;