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_stack(unsigned long *stack)
155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
158 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159 * kmemcache based allocator.
161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
163 #ifdef CONFIG_VMAP_STACK
165 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
166 * flush. Try to minimize the number of calls by caching stacks.
168 #define NR_CACHED_STACKS 2
169 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
172 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
174 #ifdef CONFIG_VMAP_STACK
179 for (i = 0; i < NR_CACHED_STACKS; i++) {
180 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
184 this_cpu_write(cached_stacks[i], NULL);
186 tsk->stack_vm_area = s;
192 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
193 VMALLOC_START, VMALLOC_END,
194 THREADINFO_GFP | __GFP_HIGHMEM,
196 0, node, __builtin_return_address(0));
199 * We can't call find_vm_area() in interrupt context, and
200 * free_thread_stack() can be called in interrupt context,
201 * so cache the vm_struct.
204 tsk->stack_vm_area = find_vm_area(stack);
207 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
210 return page ? page_address(page) : NULL;
214 static inline void free_thread_stack(struct task_struct *tsk)
216 #ifdef CONFIG_VMAP_STACK
217 if (task_stack_vm_area(tsk)) {
221 local_irq_save(flags);
222 for (i = 0; i < NR_CACHED_STACKS; i++) {
223 if (this_cpu_read(cached_stacks[i]))
226 this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
227 local_irq_restore(flags);
230 local_irq_restore(flags);
237 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
240 static struct kmem_cache *thread_stack_cache;
242 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
245 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
248 static void free_thread_stack(struct task_struct *tsk)
250 kmem_cache_free(thread_stack_cache, tsk->stack);
253 void thread_stack_cache_init(void)
255 thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
256 THREAD_SIZE, 0, NULL);
257 BUG_ON(thread_stack_cache == NULL);
262 /* SLAB cache for signal_struct structures (tsk->signal) */
263 static struct kmem_cache *signal_cachep;
265 /* SLAB cache for sighand_struct structures (tsk->sighand) */
266 struct kmem_cache *sighand_cachep;
268 /* SLAB cache for files_struct structures (tsk->files) */
269 struct kmem_cache *files_cachep;
271 /* SLAB cache for fs_struct structures (tsk->fs) */
272 struct kmem_cache *fs_cachep;
274 /* SLAB cache for vm_area_struct structures */
275 struct kmem_cache *vm_area_cachep;
277 /* SLAB cache for mm_struct structures (tsk->mm) */
278 static struct kmem_cache *mm_cachep;
280 static void account_kernel_stack(struct task_struct *tsk, int account)
282 void *stack = task_stack_page(tsk);
283 struct vm_struct *vm = task_stack_vm_area(tsk);
285 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
290 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
292 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
293 mod_zone_page_state(page_zone(vm->pages[i]),
295 PAGE_SIZE / 1024 * account);
298 /* All stack pages belong to the same memcg. */
299 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
300 account * (THREAD_SIZE / 1024));
303 * All stack pages are in the same zone and belong to the
306 struct page *first_page = virt_to_page(stack);
308 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
309 THREAD_SIZE / 1024 * account);
311 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
312 account * (THREAD_SIZE / 1024));
316 static void release_task_stack(struct task_struct *tsk)
318 account_kernel_stack(tsk, -1);
319 arch_release_thread_stack(tsk->stack);
320 free_thread_stack(tsk);
322 #ifdef CONFIG_VMAP_STACK
323 tsk->stack_vm_area = NULL;
327 #ifdef CONFIG_THREAD_INFO_IN_TASK
328 void put_task_stack(struct task_struct *tsk)
330 if (atomic_dec_and_test(&tsk->stack_refcount))
331 release_task_stack(tsk);
335 void free_task(struct task_struct *tsk)
337 #ifndef CONFIG_THREAD_INFO_IN_TASK
339 * The task is finally done with both the stack and thread_info,
342 release_task_stack(tsk);
345 * If the task had a separate stack allocation, it should be gone
348 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
350 rt_mutex_debug_task_free(tsk);
351 ftrace_graph_exit_task(tsk);
352 put_seccomp_filter(tsk);
353 arch_release_task_struct(tsk);
354 free_task_struct(tsk);
356 EXPORT_SYMBOL(free_task);
358 static inline void free_signal_struct(struct signal_struct *sig)
360 taskstats_tgid_free(sig);
361 sched_autogroup_exit(sig);
363 * __mmdrop is not safe to call from softirq context on x86 due to
364 * pgd_dtor so postpone it to the async context
367 mmdrop_async(sig->oom_mm);
368 kmem_cache_free(signal_cachep, sig);
371 static inline void put_signal_struct(struct signal_struct *sig)
373 if (atomic_dec_and_test(&sig->sigcnt))
374 free_signal_struct(sig);
377 void __put_task_struct(struct task_struct *tsk)
379 WARN_ON(!tsk->exit_state);
380 WARN_ON(atomic_read(&tsk->usage));
381 WARN_ON(tsk == current);
385 security_task_free(tsk);
387 delayacct_tsk_free(tsk);
388 put_signal_struct(tsk->signal);
390 if (!profile_handoff_task(tsk))
393 EXPORT_SYMBOL_GPL(__put_task_struct);
395 void __init __weak arch_task_cache_init(void) { }
400 static void set_max_threads(unsigned int max_threads_suggested)
405 * The number of threads shall be limited such that the thread
406 * structures may only consume a small part of the available memory.
408 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
409 threads = MAX_THREADS;
411 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
412 (u64) THREAD_SIZE * 8UL);
414 if (threads > max_threads_suggested)
415 threads = max_threads_suggested;
417 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
420 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
421 /* Initialized by the architecture: */
422 int arch_task_struct_size __read_mostly;
425 void __init fork_init(void)
428 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
429 #ifndef ARCH_MIN_TASKALIGN
430 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
432 /* create a slab on which task_structs can be allocated */
433 task_struct_cachep = kmem_cache_create("task_struct",
434 arch_task_struct_size, ARCH_MIN_TASKALIGN,
435 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
438 /* do the arch specific task caches init */
439 arch_task_cache_init();
441 set_max_threads(MAX_THREADS);
443 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
444 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
445 init_task.signal->rlim[RLIMIT_SIGPENDING] =
446 init_task.signal->rlim[RLIMIT_NPROC];
448 for (i = 0; i < UCOUNT_COUNTS; i++) {
449 init_user_ns.ucount_max[i] = max_threads/2;
453 int __weak arch_dup_task_struct(struct task_struct *dst,
454 struct task_struct *src)
460 void set_task_stack_end_magic(struct task_struct *tsk)
462 unsigned long *stackend;
464 stackend = end_of_stack(tsk);
465 *stackend = STACK_END_MAGIC; /* for overflow detection */
468 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
470 struct task_struct *tsk;
471 unsigned long *stack;
472 struct vm_struct *stack_vm_area;
475 if (node == NUMA_NO_NODE)
476 node = tsk_fork_get_node(orig);
477 tsk = alloc_task_struct_node(node);
481 stack = alloc_thread_stack_node(tsk, node);
485 stack_vm_area = task_stack_vm_area(tsk);
487 err = arch_dup_task_struct(tsk, orig);
490 * arch_dup_task_struct() clobbers the stack-related fields. Make
491 * sure they're properly initialized before using any stack-related
495 #ifdef CONFIG_VMAP_STACK
496 tsk->stack_vm_area = stack_vm_area;
498 #ifdef CONFIG_THREAD_INFO_IN_TASK
499 atomic_set(&tsk->stack_refcount, 1);
505 #ifdef CONFIG_SECCOMP
507 * We must handle setting up seccomp filters once we're under
508 * the sighand lock in case orig has changed between now and
509 * then. Until then, filter must be NULL to avoid messing up
510 * the usage counts on the error path calling free_task.
512 tsk->seccomp.filter = NULL;
515 setup_thread_stack(tsk, orig);
516 clear_user_return_notifier(tsk);
517 clear_tsk_need_resched(tsk);
518 set_task_stack_end_magic(tsk);
520 #ifdef CONFIG_CC_STACKPROTECTOR
521 tsk->stack_canary = get_random_int();
525 * One for us, one for whoever does the "release_task()" (usually
528 atomic_set(&tsk->usage, 2);
529 #ifdef CONFIG_BLK_DEV_IO_TRACE
532 tsk->splice_pipe = NULL;
533 tsk->task_frag.page = NULL;
534 tsk->wake_q.next = NULL;
536 account_kernel_stack(tsk, 1);
543 free_thread_stack(tsk);
545 free_task_struct(tsk);
550 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
552 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
553 struct rb_node **rb_link, *rb_parent;
555 unsigned long charge;
557 uprobe_start_dup_mmap();
558 if (down_write_killable(&oldmm->mmap_sem)) {
560 goto fail_uprobe_end;
562 flush_cache_dup_mm(oldmm);
563 uprobe_dup_mmap(oldmm, mm);
565 * Not linked in yet - no deadlock potential:
567 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
569 /* No ordering required: file already has been exposed. */
570 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
572 mm->total_vm = oldmm->total_vm;
573 mm->data_vm = oldmm->data_vm;
574 mm->exec_vm = oldmm->exec_vm;
575 mm->stack_vm = oldmm->stack_vm;
577 rb_link = &mm->mm_rb.rb_node;
580 retval = ksm_fork(mm, oldmm);
583 retval = khugepaged_fork(mm, oldmm);
588 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
591 if (mpnt->vm_flags & VM_DONTCOPY) {
592 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
596 if (mpnt->vm_flags & VM_ACCOUNT) {
597 unsigned long len = vma_pages(mpnt);
599 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
603 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
607 INIT_LIST_HEAD(&tmp->anon_vma_chain);
608 retval = vma_dup_policy(mpnt, tmp);
610 goto fail_nomem_policy;
612 if (anon_vma_fork(tmp, mpnt))
613 goto fail_nomem_anon_vma_fork;
615 ~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
616 tmp->vm_next = tmp->vm_prev = NULL;
617 tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
620 struct inode *inode = file_inode(file);
621 struct address_space *mapping = file->f_mapping;
624 if (tmp->vm_flags & VM_DENYWRITE)
625 atomic_dec(&inode->i_writecount);
626 i_mmap_lock_write(mapping);
627 if (tmp->vm_flags & VM_SHARED)
628 atomic_inc(&mapping->i_mmap_writable);
629 flush_dcache_mmap_lock(mapping);
630 /* insert tmp into the share list, just after mpnt */
631 vma_interval_tree_insert_after(tmp, mpnt,
633 flush_dcache_mmap_unlock(mapping);
634 i_mmap_unlock_write(mapping);
638 * Clear hugetlb-related page reserves for children. This only
639 * affects MAP_PRIVATE mappings. Faults generated by the child
640 * are not guaranteed to succeed, even if read-only
642 if (is_vm_hugetlb_page(tmp))
643 reset_vma_resv_huge_pages(tmp);
646 * Link in the new vma and copy the page table entries.
649 pprev = &tmp->vm_next;
653 __vma_link_rb(mm, tmp, rb_link, rb_parent);
654 rb_link = &tmp->vm_rb.rb_right;
655 rb_parent = &tmp->vm_rb;
658 retval = copy_page_range(mm, oldmm, mpnt);
660 if (tmp->vm_ops && tmp->vm_ops->open)
661 tmp->vm_ops->open(tmp);
666 /* a new mm has just been created */
667 arch_dup_mmap(oldmm, mm);
670 up_write(&mm->mmap_sem);
672 up_write(&oldmm->mmap_sem);
674 uprobe_end_dup_mmap();
676 fail_nomem_anon_vma_fork:
677 mpol_put(vma_policy(tmp));
679 kmem_cache_free(vm_area_cachep, tmp);
682 vm_unacct_memory(charge);
686 static inline int mm_alloc_pgd(struct mm_struct *mm)
688 mm->pgd = pgd_alloc(mm);
689 if (unlikely(!mm->pgd))
694 static inline void mm_free_pgd(struct mm_struct *mm)
696 pgd_free(mm, mm->pgd);
699 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
701 down_write(&oldmm->mmap_sem);
702 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
703 up_write(&oldmm->mmap_sem);
706 #define mm_alloc_pgd(mm) (0)
707 #define mm_free_pgd(mm)
708 #endif /* CONFIG_MMU */
710 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
712 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
713 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
715 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
717 static int __init coredump_filter_setup(char *s)
719 default_dump_filter =
720 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
721 MMF_DUMP_FILTER_MASK;
725 __setup("coredump_filter=", coredump_filter_setup);
727 #include <linux/init_task.h>
729 static void mm_init_aio(struct mm_struct *mm)
732 spin_lock_init(&mm->ioctx_lock);
733 mm->ioctx_table = NULL;
737 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
744 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
748 mm->vmacache_seqnum = 0;
749 atomic_set(&mm->mm_users, 1);
750 atomic_set(&mm->mm_count, 1);
751 init_rwsem(&mm->mmap_sem);
752 INIT_LIST_HEAD(&mm->mmlist);
753 mm->core_state = NULL;
754 atomic_long_set(&mm->nr_ptes, 0);
759 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
760 spin_lock_init(&mm->page_table_lock);
763 mm_init_owner(mm, p);
764 mmu_notifier_mm_init(mm);
765 clear_tlb_flush_pending(mm);
766 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
767 mm->pmd_huge_pte = NULL;
771 mm->flags = current->mm->flags & MMF_INIT_MASK;
772 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
774 mm->flags = default_dump_filter;
778 if (mm_alloc_pgd(mm))
781 if (init_new_context(p, mm))
793 static void check_mm(struct mm_struct *mm)
797 for (i = 0; i < NR_MM_COUNTERS; i++) {
798 long x = atomic_long_read(&mm->rss_stat.count[i]);
801 printk(KERN_ALERT "BUG: Bad rss-counter state "
802 "mm:%p idx:%d val:%ld\n", mm, i, x);
805 if (atomic_long_read(&mm->nr_ptes))
806 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
807 atomic_long_read(&mm->nr_ptes));
809 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
812 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
813 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
818 * Allocate and initialize an mm_struct.
820 struct mm_struct *mm_alloc(void)
822 struct mm_struct *mm;
828 memset(mm, 0, sizeof(*mm));
829 return mm_init(mm, current);
833 * Called when the last reference to the mm
834 * is dropped: either by a lazy thread or by
835 * mmput. Free the page directory and the mm.
837 void __mmdrop(struct mm_struct *mm)
839 BUG_ON(mm == &init_mm);
842 mmu_notifier_mm_destroy(mm);
846 EXPORT_SYMBOL_GPL(__mmdrop);
848 static inline void __mmput(struct mm_struct *mm)
850 VM_BUG_ON(atomic_read(&mm->mm_users));
852 uprobe_clear_state(mm);
855 khugepaged_exit(mm); /* must run before exit_mmap */
857 set_mm_exe_file(mm, NULL);
858 if (!list_empty(&mm->mmlist)) {
859 spin_lock(&mmlist_lock);
860 list_del(&mm->mmlist);
861 spin_unlock(&mmlist_lock);
864 module_put(mm->binfmt->module);
869 * Decrement the use count and release all resources for an mm.
871 void mmput(struct mm_struct *mm)
875 if (atomic_dec_and_test(&mm->mm_users))
878 EXPORT_SYMBOL_GPL(mmput);
881 static void mmput_async_fn(struct work_struct *work)
883 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
887 void mmput_async(struct mm_struct *mm)
889 if (atomic_dec_and_test(&mm->mm_users)) {
890 INIT_WORK(&mm->async_put_work, mmput_async_fn);
891 schedule_work(&mm->async_put_work);
897 * set_mm_exe_file - change a reference to the mm's executable file
899 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
901 * Main users are mmput() and sys_execve(). Callers prevent concurrent
902 * invocations: in mmput() nobody alive left, in execve task is single
903 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
904 * mm->exe_file, but does so without using set_mm_exe_file() in order
905 * to do avoid the need for any locks.
907 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
909 struct file *old_exe_file;
912 * It is safe to dereference the exe_file without RCU as
913 * this function is only called if nobody else can access
914 * this mm -- see comment above for justification.
916 old_exe_file = rcu_dereference_raw(mm->exe_file);
919 get_file(new_exe_file);
920 rcu_assign_pointer(mm->exe_file, new_exe_file);
926 * get_mm_exe_file - acquire a reference to the mm's executable file
928 * Returns %NULL if mm has no associated executable file.
929 * User must release file via fput().
931 struct file *get_mm_exe_file(struct mm_struct *mm)
933 struct file *exe_file;
936 exe_file = rcu_dereference(mm->exe_file);
937 if (exe_file && !get_file_rcu(exe_file))
942 EXPORT_SYMBOL(get_mm_exe_file);
945 * get_task_exe_file - acquire a reference to the task's executable file
947 * Returns %NULL if task's mm (if any) has no associated executable file or
948 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
949 * User must release file via fput().
951 struct file *get_task_exe_file(struct task_struct *task)
953 struct file *exe_file = NULL;
954 struct mm_struct *mm;
959 if (!(task->flags & PF_KTHREAD))
960 exe_file = get_mm_exe_file(mm);
965 EXPORT_SYMBOL(get_task_exe_file);
968 * get_task_mm - acquire a reference to the task's mm
970 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
971 * this kernel workthread has transiently adopted a user mm with use_mm,
972 * to do its AIO) is not set and if so returns a reference to it, after
973 * bumping up the use count. User must release the mm via mmput()
974 * after use. Typically used by /proc and ptrace.
976 struct mm_struct *get_task_mm(struct task_struct *task)
978 struct mm_struct *mm;
983 if (task->flags & PF_KTHREAD)
986 atomic_inc(&mm->mm_users);
991 EXPORT_SYMBOL_GPL(get_task_mm);
993 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
995 struct mm_struct *mm;
998 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1000 return ERR_PTR(err);
1002 mm = get_task_mm(task);
1003 if (mm && mm != current->mm &&
1004 !ptrace_may_access(task, mode)) {
1006 mm = ERR_PTR(-EACCES);
1008 mutex_unlock(&task->signal->cred_guard_mutex);
1013 static void complete_vfork_done(struct task_struct *tsk)
1015 struct completion *vfork;
1018 vfork = tsk->vfork_done;
1019 if (likely(vfork)) {
1020 tsk->vfork_done = NULL;
1026 static int wait_for_vfork_done(struct task_struct *child,
1027 struct completion *vfork)
1031 freezer_do_not_count();
1032 killed = wait_for_completion_killable(vfork);
1037 child->vfork_done = NULL;
1041 put_task_struct(child);
1045 /* Please note the differences between mmput and mm_release.
1046 * mmput is called whenever we stop holding onto a mm_struct,
1047 * error success whatever.
1049 * mm_release is called after a mm_struct has been removed
1050 * from the current process.
1052 * This difference is important for error handling, when we
1053 * only half set up a mm_struct for a new process and need to restore
1054 * the old one. Because we mmput the new mm_struct before
1055 * restoring the old one. . .
1056 * Eric Biederman 10 January 1998
1058 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1060 /* Get rid of any futexes when releasing the mm */
1062 if (unlikely(tsk->robust_list)) {
1063 exit_robust_list(tsk);
1064 tsk->robust_list = NULL;
1066 #ifdef CONFIG_COMPAT
1067 if (unlikely(tsk->compat_robust_list)) {
1068 compat_exit_robust_list(tsk);
1069 tsk->compat_robust_list = NULL;
1072 if (unlikely(!list_empty(&tsk->pi_state_list)))
1073 exit_pi_state_list(tsk);
1076 uprobe_free_utask(tsk);
1078 /* Get rid of any cached register state */
1079 deactivate_mm(tsk, mm);
1082 * Signal userspace if we're not exiting with a core dump
1083 * because we want to leave the value intact for debugging
1086 if (tsk->clear_child_tid) {
1087 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1088 atomic_read(&mm->mm_users) > 1) {
1090 * We don't check the error code - if userspace has
1091 * not set up a proper pointer then tough luck.
1093 put_user(0, tsk->clear_child_tid);
1094 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1097 tsk->clear_child_tid = NULL;
1101 * All done, finally we can wake up parent and return this mm to him.
1102 * Also kthread_stop() uses this completion for synchronization.
1104 if (tsk->vfork_done)
1105 complete_vfork_done(tsk);
1109 * Allocate a new mm structure and copy contents from the
1110 * mm structure of the passed in task structure.
1112 static struct mm_struct *dup_mm(struct task_struct *tsk)
1114 struct mm_struct *mm, *oldmm = current->mm;
1121 memcpy(mm, oldmm, sizeof(*mm));
1123 if (!mm_init(mm, tsk))
1126 err = dup_mmap(mm, oldmm);
1130 mm->hiwater_rss = get_mm_rss(mm);
1131 mm->hiwater_vm = mm->total_vm;
1133 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1139 /* don't put binfmt in mmput, we haven't got module yet */
1147 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1149 struct mm_struct *mm, *oldmm;
1152 tsk->min_flt = tsk->maj_flt = 0;
1153 tsk->nvcsw = tsk->nivcsw = 0;
1154 #ifdef CONFIG_DETECT_HUNG_TASK
1155 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1159 tsk->active_mm = NULL;
1162 * Are we cloning a kernel thread?
1164 * We need to steal a active VM for that..
1166 oldmm = current->mm;
1170 /* initialize the new vmacache entries */
1171 vmacache_flush(tsk);
1173 if (clone_flags & CLONE_VM) {
1174 atomic_inc(&oldmm->mm_users);
1186 tsk->active_mm = mm;
1193 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1195 struct fs_struct *fs = current->fs;
1196 if (clone_flags & CLONE_FS) {
1197 /* tsk->fs is already what we want */
1198 spin_lock(&fs->lock);
1200 spin_unlock(&fs->lock);
1204 spin_unlock(&fs->lock);
1207 tsk->fs = copy_fs_struct(fs);
1213 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1215 struct files_struct *oldf, *newf;
1219 * A background process may not have any files ...
1221 oldf = current->files;
1225 if (clone_flags & CLONE_FILES) {
1226 atomic_inc(&oldf->count);
1230 newf = dup_fd(oldf, &error);
1240 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1243 struct io_context *ioc = current->io_context;
1244 struct io_context *new_ioc;
1249 * Share io context with parent, if CLONE_IO is set
1251 if (clone_flags & CLONE_IO) {
1253 tsk->io_context = ioc;
1254 } else if (ioprio_valid(ioc->ioprio)) {
1255 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1256 if (unlikely(!new_ioc))
1259 new_ioc->ioprio = ioc->ioprio;
1260 put_io_context(new_ioc);
1266 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1268 struct sighand_struct *sig;
1270 if (clone_flags & CLONE_SIGHAND) {
1271 atomic_inc(¤t->sighand->count);
1274 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1275 rcu_assign_pointer(tsk->sighand, sig);
1279 atomic_set(&sig->count, 1);
1280 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1284 void __cleanup_sighand(struct sighand_struct *sighand)
1286 if (atomic_dec_and_test(&sighand->count)) {
1287 signalfd_cleanup(sighand);
1289 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1290 * without an RCU grace period, see __lock_task_sighand().
1292 kmem_cache_free(sighand_cachep, sighand);
1297 * Initialize POSIX timer handling for a thread group.
1299 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1301 unsigned long cpu_limit;
1303 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1304 if (cpu_limit != RLIM_INFINITY) {
1305 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1306 sig->cputimer.running = true;
1309 /* The timer lists. */
1310 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1311 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1312 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1315 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1317 struct signal_struct *sig;
1319 if (clone_flags & CLONE_THREAD)
1322 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1327 sig->nr_threads = 1;
1328 atomic_set(&sig->live, 1);
1329 atomic_set(&sig->sigcnt, 1);
1331 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1332 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1333 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1335 init_waitqueue_head(&sig->wait_chldexit);
1336 sig->curr_target = tsk;
1337 init_sigpending(&sig->shared_pending);
1338 INIT_LIST_HEAD(&sig->posix_timers);
1339 seqlock_init(&sig->stats_lock);
1340 prev_cputime_init(&sig->prev_cputime);
1342 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1343 sig->real_timer.function = it_real_fn;
1345 task_lock(current->group_leader);
1346 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1347 task_unlock(current->group_leader);
1349 posix_cpu_timers_init_group(sig);
1351 tty_audit_fork(sig);
1352 sched_autogroup_fork(sig);
1354 sig->oom_score_adj = current->signal->oom_score_adj;
1355 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1357 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1358 current->signal->is_child_subreaper;
1360 mutex_init(&sig->cred_guard_mutex);
1365 static void copy_seccomp(struct task_struct *p)
1367 #ifdef CONFIG_SECCOMP
1369 * Must be called with sighand->lock held, which is common to
1370 * all threads in the group. Holding cred_guard_mutex is not
1371 * needed because this new task is not yet running and cannot
1374 assert_spin_locked(¤t->sighand->siglock);
1376 /* Ref-count the new filter user, and assign it. */
1377 get_seccomp_filter(current);
1378 p->seccomp = current->seccomp;
1381 * Explicitly enable no_new_privs here in case it got set
1382 * between the task_struct being duplicated and holding the
1383 * sighand lock. The seccomp state and nnp must be in sync.
1385 if (task_no_new_privs(current))
1386 task_set_no_new_privs(p);
1389 * If the parent gained a seccomp mode after copying thread
1390 * flags and between before we held the sighand lock, we have
1391 * to manually enable the seccomp thread flag here.
1393 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1394 set_tsk_thread_flag(p, TIF_SECCOMP);
1398 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1400 current->clear_child_tid = tidptr;
1402 return task_pid_vnr(current);
1405 static void rt_mutex_init_task(struct task_struct *p)
1407 raw_spin_lock_init(&p->pi_lock);
1408 #ifdef CONFIG_RT_MUTEXES
1409 p->pi_waiters = RB_ROOT;
1410 p->pi_waiters_leftmost = NULL;
1411 p->pi_blocked_on = NULL;
1416 * Initialize POSIX timer handling for a single task.
1418 static void posix_cpu_timers_init(struct task_struct *tsk)
1420 tsk->cputime_expires.prof_exp = 0;
1421 tsk->cputime_expires.virt_exp = 0;
1422 tsk->cputime_expires.sched_exp = 0;
1423 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1424 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1425 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1429 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1431 task->pids[type].pid = pid;
1435 * This creates a new process as a copy of the old one,
1436 * but does not actually start it yet.
1438 * It copies the registers, and all the appropriate
1439 * parts of the process environment (as per the clone
1440 * flags). The actual kick-off is left to the caller.
1442 static struct task_struct *copy_process(unsigned long clone_flags,
1443 unsigned long stack_start,
1444 unsigned long stack_size,
1445 int __user *child_tidptr,
1452 struct task_struct *p;
1454 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1455 return ERR_PTR(-EINVAL);
1457 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1458 return ERR_PTR(-EINVAL);
1461 * Thread groups must share signals as well, and detached threads
1462 * can only be started up within the thread group.
1464 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1465 return ERR_PTR(-EINVAL);
1468 * Shared signal handlers imply shared VM. By way of the above,
1469 * thread groups also imply shared VM. Blocking this case allows
1470 * for various simplifications in other code.
1472 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1473 return ERR_PTR(-EINVAL);
1476 * Siblings of global init remain as zombies on exit since they are
1477 * not reaped by their parent (swapper). To solve this and to avoid
1478 * multi-rooted process trees, prevent global and container-inits
1479 * from creating siblings.
1481 if ((clone_flags & CLONE_PARENT) &&
1482 current->signal->flags & SIGNAL_UNKILLABLE)
1483 return ERR_PTR(-EINVAL);
1486 * If the new process will be in a different pid or user namespace
1487 * do not allow it to share a thread group with the forking task.
1489 if (clone_flags & CLONE_THREAD) {
1490 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1491 (task_active_pid_ns(current) !=
1492 current->nsproxy->pid_ns_for_children))
1493 return ERR_PTR(-EINVAL);
1496 retval = security_task_create(clone_flags);
1501 p = dup_task_struct(current, node);
1505 ftrace_graph_init_task(p);
1507 rt_mutex_init_task(p);
1509 #ifdef CONFIG_PROVE_LOCKING
1510 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1511 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1514 if (atomic_read(&p->real_cred->user->processes) >=
1515 task_rlimit(p, RLIMIT_NPROC)) {
1516 if (p->real_cred->user != INIT_USER &&
1517 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1520 current->flags &= ~PF_NPROC_EXCEEDED;
1522 retval = copy_creds(p, clone_flags);
1527 * If multiple threads are within copy_process(), then this check
1528 * triggers too late. This doesn't hurt, the check is only there
1529 * to stop root fork bombs.
1532 if (nr_threads >= max_threads)
1533 goto bad_fork_cleanup_count;
1535 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1536 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1537 p->flags |= PF_FORKNOEXEC;
1538 INIT_LIST_HEAD(&p->children);
1539 INIT_LIST_HEAD(&p->sibling);
1540 rcu_copy_process(p);
1541 p->vfork_done = NULL;
1542 spin_lock_init(&p->alloc_lock);
1544 init_sigpending(&p->pending);
1546 p->utime = p->stime = p->gtime = 0;
1547 p->utimescaled = p->stimescaled = 0;
1548 prev_cputime_init(&p->prev_cputime);
1550 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1551 seqcount_init(&p->vtime_seqcount);
1553 p->vtime_snap_whence = VTIME_INACTIVE;
1556 #if defined(SPLIT_RSS_COUNTING)
1557 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1560 p->default_timer_slack_ns = current->timer_slack_ns;
1562 task_io_accounting_init(&p->ioac);
1563 acct_clear_integrals(p);
1565 posix_cpu_timers_init(p);
1567 p->start_time = ktime_get_ns();
1568 p->real_start_time = ktime_get_boot_ns();
1569 p->io_context = NULL;
1570 p->audit_context = NULL;
1573 p->mempolicy = mpol_dup(p->mempolicy);
1574 if (IS_ERR(p->mempolicy)) {
1575 retval = PTR_ERR(p->mempolicy);
1576 p->mempolicy = NULL;
1577 goto bad_fork_cleanup_threadgroup_lock;
1580 #ifdef CONFIG_CPUSETS
1581 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1582 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1583 seqcount_init(&p->mems_allowed_seq);
1585 #ifdef CONFIG_TRACE_IRQFLAGS
1587 p->hardirqs_enabled = 0;
1588 p->hardirq_enable_ip = 0;
1589 p->hardirq_enable_event = 0;
1590 p->hardirq_disable_ip = _THIS_IP_;
1591 p->hardirq_disable_event = 0;
1592 p->softirqs_enabled = 1;
1593 p->softirq_enable_ip = _THIS_IP_;
1594 p->softirq_enable_event = 0;
1595 p->softirq_disable_ip = 0;
1596 p->softirq_disable_event = 0;
1597 p->hardirq_context = 0;
1598 p->softirq_context = 0;
1601 p->pagefault_disabled = 0;
1603 #ifdef CONFIG_LOCKDEP
1604 p->lockdep_depth = 0; /* no locks held yet */
1605 p->curr_chain_key = 0;
1606 p->lockdep_recursion = 0;
1609 #ifdef CONFIG_DEBUG_MUTEXES
1610 p->blocked_on = NULL; /* not blocked yet */
1612 #ifdef CONFIG_BCACHE
1613 p->sequential_io = 0;
1614 p->sequential_io_avg = 0;
1617 /* Perform scheduler related setup. Assign this task to a CPU. */
1618 retval = sched_fork(clone_flags, p);
1620 goto bad_fork_cleanup_policy;
1622 retval = perf_event_init_task(p);
1624 goto bad_fork_cleanup_policy;
1625 retval = audit_alloc(p);
1627 goto bad_fork_cleanup_perf;
1628 /* copy all the process information */
1630 retval = copy_semundo(clone_flags, p);
1632 goto bad_fork_cleanup_audit;
1633 retval = copy_files(clone_flags, p);
1635 goto bad_fork_cleanup_semundo;
1636 retval = copy_fs(clone_flags, p);
1638 goto bad_fork_cleanup_files;
1639 retval = copy_sighand(clone_flags, p);
1641 goto bad_fork_cleanup_fs;
1642 retval = copy_signal(clone_flags, p);
1644 goto bad_fork_cleanup_sighand;
1645 retval = copy_mm(clone_flags, p);
1647 goto bad_fork_cleanup_signal;
1648 retval = copy_namespaces(clone_flags, p);
1650 goto bad_fork_cleanup_mm;
1651 retval = copy_io(clone_flags, p);
1653 goto bad_fork_cleanup_namespaces;
1654 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1656 goto bad_fork_cleanup_io;
1658 if (pid != &init_struct_pid) {
1659 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1661 retval = PTR_ERR(pid);
1662 goto bad_fork_cleanup_thread;
1666 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1668 * Clear TID on mm_release()?
1670 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1675 p->robust_list = NULL;
1676 #ifdef CONFIG_COMPAT
1677 p->compat_robust_list = NULL;
1679 INIT_LIST_HEAD(&p->pi_state_list);
1680 p->pi_state_cache = NULL;
1683 * sigaltstack should be cleared when sharing the same VM
1685 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1689 * Syscall tracing and stepping should be turned off in the
1690 * child regardless of CLONE_PTRACE.
1692 user_disable_single_step(p);
1693 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1694 #ifdef TIF_SYSCALL_EMU
1695 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1697 clear_all_latency_tracing(p);
1699 /* ok, now we should be set up.. */
1700 p->pid = pid_nr(pid);
1701 if (clone_flags & CLONE_THREAD) {
1702 p->exit_signal = -1;
1703 p->group_leader = current->group_leader;
1704 p->tgid = current->tgid;
1706 if (clone_flags & CLONE_PARENT)
1707 p->exit_signal = current->group_leader->exit_signal;
1709 p->exit_signal = (clone_flags & CSIGNAL);
1710 p->group_leader = p;
1715 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1716 p->dirty_paused_when = 0;
1718 p->pdeath_signal = 0;
1719 INIT_LIST_HEAD(&p->thread_group);
1720 p->task_works = NULL;
1722 threadgroup_change_begin(current);
1724 * Ensure that the cgroup subsystem policies allow the new process to be
1725 * forked. It should be noted the the new process's css_set can be changed
1726 * between here and cgroup_post_fork() if an organisation operation is in
1729 retval = cgroup_can_fork(p);
1731 goto bad_fork_free_pid;
1734 * Make it visible to the rest of the system, but dont wake it up yet.
1735 * Need tasklist lock for parent etc handling!
1737 write_lock_irq(&tasklist_lock);
1739 /* CLONE_PARENT re-uses the old parent */
1740 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1741 p->real_parent = current->real_parent;
1742 p->parent_exec_id = current->parent_exec_id;
1744 p->real_parent = current;
1745 p->parent_exec_id = current->self_exec_id;
1748 spin_lock(¤t->sighand->siglock);
1751 * Copy seccomp details explicitly here, in case they were changed
1752 * before holding sighand lock.
1757 * Process group and session signals need to be delivered to just the
1758 * parent before the fork or both the parent and the child after the
1759 * fork. Restart if a signal comes in before we add the new process to
1760 * it's process group.
1761 * A fatal signal pending means that current will exit, so the new
1762 * thread can't slip out of an OOM kill (or normal SIGKILL).
1764 recalc_sigpending();
1765 if (signal_pending(current)) {
1766 spin_unlock(¤t->sighand->siglock);
1767 write_unlock_irq(&tasklist_lock);
1768 retval = -ERESTARTNOINTR;
1769 goto bad_fork_cancel_cgroup;
1772 if (likely(p->pid)) {
1773 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1775 init_task_pid(p, PIDTYPE_PID, pid);
1776 if (thread_group_leader(p)) {
1777 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1778 init_task_pid(p, PIDTYPE_SID, task_session(current));
1780 if (is_child_reaper(pid)) {
1781 ns_of_pid(pid)->child_reaper = p;
1782 p->signal->flags |= SIGNAL_UNKILLABLE;
1785 p->signal->leader_pid = pid;
1786 p->signal->tty = tty_kref_get(current->signal->tty);
1787 list_add_tail(&p->sibling, &p->real_parent->children);
1788 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1789 attach_pid(p, PIDTYPE_PGID);
1790 attach_pid(p, PIDTYPE_SID);
1791 __this_cpu_inc(process_counts);
1793 current->signal->nr_threads++;
1794 atomic_inc(¤t->signal->live);
1795 atomic_inc(¤t->signal->sigcnt);
1796 list_add_tail_rcu(&p->thread_group,
1797 &p->group_leader->thread_group);
1798 list_add_tail_rcu(&p->thread_node,
1799 &p->signal->thread_head);
1801 attach_pid(p, PIDTYPE_PID);
1806 spin_unlock(¤t->sighand->siglock);
1807 syscall_tracepoint_update(p);
1808 write_unlock_irq(&tasklist_lock);
1810 proc_fork_connector(p);
1811 cgroup_post_fork(p);
1812 threadgroup_change_end(current);
1815 trace_task_newtask(p, clone_flags);
1816 uprobe_copy_process(p, clone_flags);
1820 bad_fork_cancel_cgroup:
1821 cgroup_cancel_fork(p);
1823 threadgroup_change_end(current);
1824 if (pid != &init_struct_pid)
1826 bad_fork_cleanup_thread:
1828 bad_fork_cleanup_io:
1831 bad_fork_cleanup_namespaces:
1832 exit_task_namespaces(p);
1833 bad_fork_cleanup_mm:
1836 bad_fork_cleanup_signal:
1837 if (!(clone_flags & CLONE_THREAD))
1838 free_signal_struct(p->signal);
1839 bad_fork_cleanup_sighand:
1840 __cleanup_sighand(p->sighand);
1841 bad_fork_cleanup_fs:
1842 exit_fs(p); /* blocking */
1843 bad_fork_cleanup_files:
1844 exit_files(p); /* blocking */
1845 bad_fork_cleanup_semundo:
1847 bad_fork_cleanup_audit:
1849 bad_fork_cleanup_perf:
1850 perf_event_free_task(p);
1851 bad_fork_cleanup_policy:
1853 mpol_put(p->mempolicy);
1854 bad_fork_cleanup_threadgroup_lock:
1856 delayacct_tsk_free(p);
1857 bad_fork_cleanup_count:
1858 atomic_dec(&p->cred->user->processes);
1864 return ERR_PTR(retval);
1867 static inline void init_idle_pids(struct pid_link *links)
1871 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1872 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1873 links[type].pid = &init_struct_pid;
1877 struct task_struct *fork_idle(int cpu)
1879 struct task_struct *task;
1880 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1882 if (!IS_ERR(task)) {
1883 init_idle_pids(task->pids);
1884 init_idle(task, cpu);
1891 * Ok, this is the main fork-routine.
1893 * It copies the process, and if successful kick-starts
1894 * it and waits for it to finish using the VM if required.
1896 long _do_fork(unsigned long clone_flags,
1897 unsigned long stack_start,
1898 unsigned long stack_size,
1899 int __user *parent_tidptr,
1900 int __user *child_tidptr,
1903 struct task_struct *p;
1908 * Determine whether and which event to report to ptracer. When
1909 * called from kernel_thread or CLONE_UNTRACED is explicitly
1910 * requested, no event is reported; otherwise, report if the event
1911 * for the type of forking is enabled.
1913 if (!(clone_flags & CLONE_UNTRACED)) {
1914 if (clone_flags & CLONE_VFORK)
1915 trace = PTRACE_EVENT_VFORK;
1916 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1917 trace = PTRACE_EVENT_CLONE;
1919 trace = PTRACE_EVENT_FORK;
1921 if (likely(!ptrace_event_enabled(current, trace)))
1925 p = copy_process(clone_flags, stack_start, stack_size,
1926 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1928 * Do this prior waking up the new thread - the thread pointer
1929 * might get invalid after that point, if the thread exits quickly.
1932 struct completion vfork;
1935 trace_sched_process_fork(current, p);
1937 pid = get_task_pid(p, PIDTYPE_PID);
1940 if (clone_flags & CLONE_PARENT_SETTID)
1941 put_user(nr, parent_tidptr);
1943 if (clone_flags & CLONE_VFORK) {
1944 p->vfork_done = &vfork;
1945 init_completion(&vfork);
1949 wake_up_new_task(p);
1951 /* forking complete and child started to run, tell ptracer */
1952 if (unlikely(trace))
1953 ptrace_event_pid(trace, pid);
1955 if (clone_flags & CLONE_VFORK) {
1956 if (!wait_for_vfork_done(p, &vfork))
1957 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1967 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1968 /* For compatibility with architectures that call do_fork directly rather than
1969 * using the syscall entry points below. */
1970 long do_fork(unsigned long clone_flags,
1971 unsigned long stack_start,
1972 unsigned long stack_size,
1973 int __user *parent_tidptr,
1974 int __user *child_tidptr)
1976 return _do_fork(clone_flags, stack_start, stack_size,
1977 parent_tidptr, child_tidptr, 0);
1982 * Create a kernel thread.
1984 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1986 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1987 (unsigned long)arg, NULL, NULL, 0);
1990 #ifdef __ARCH_WANT_SYS_FORK
1991 SYSCALL_DEFINE0(fork)
1994 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1996 /* can not support in nommu mode */
2002 #ifdef __ARCH_WANT_SYS_VFORK
2003 SYSCALL_DEFINE0(vfork)
2005 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2010 #ifdef __ARCH_WANT_SYS_CLONE
2011 #ifdef CONFIG_CLONE_BACKWARDS
2012 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2013 int __user *, parent_tidptr,
2015 int __user *, child_tidptr)
2016 #elif defined(CONFIG_CLONE_BACKWARDS2)
2017 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2018 int __user *, parent_tidptr,
2019 int __user *, child_tidptr,
2021 #elif defined(CONFIG_CLONE_BACKWARDS3)
2022 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2024 int __user *, parent_tidptr,
2025 int __user *, child_tidptr,
2028 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2029 int __user *, parent_tidptr,
2030 int __user *, child_tidptr,
2034 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2038 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2039 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2042 static void sighand_ctor(void *data)
2044 struct sighand_struct *sighand = data;
2046 spin_lock_init(&sighand->siglock);
2047 init_waitqueue_head(&sighand->signalfd_wqh);
2050 void __init proc_caches_init(void)
2052 sighand_cachep = kmem_cache_create("sighand_cache",
2053 sizeof(struct sighand_struct), 0,
2054 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2055 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2056 signal_cachep = kmem_cache_create("signal_cache",
2057 sizeof(struct signal_struct), 0,
2058 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2060 files_cachep = kmem_cache_create("files_cache",
2061 sizeof(struct files_struct), 0,
2062 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2064 fs_cachep = kmem_cache_create("fs_cache",
2065 sizeof(struct fs_struct), 0,
2066 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2069 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2070 * whole struct cpumask for the OFFSTACK case. We could change
2071 * this to *only* allocate as much of it as required by the
2072 * maximum number of CPU's we can ever have. The cpumask_allocation
2073 * is at the end of the structure, exactly for that reason.
2075 mm_cachep = kmem_cache_create("mm_struct",
2076 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2077 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2079 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2081 nsproxy_cache_init();
2085 * Check constraints on flags passed to the unshare system call.
2087 static int check_unshare_flags(unsigned long unshare_flags)
2089 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2090 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2091 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2092 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2095 * Not implemented, but pretend it works if there is nothing
2096 * to unshare. Note that unsharing the address space or the
2097 * signal handlers also need to unshare the signal queues (aka
2100 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2101 if (!thread_group_empty(current))
2104 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2105 if (atomic_read(¤t->sighand->count) > 1)
2108 if (unshare_flags & CLONE_VM) {
2109 if (!current_is_single_threaded())
2117 * Unshare the filesystem structure if it is being shared
2119 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2121 struct fs_struct *fs = current->fs;
2123 if (!(unshare_flags & CLONE_FS) || !fs)
2126 /* don't need lock here; in the worst case we'll do useless copy */
2130 *new_fsp = copy_fs_struct(fs);
2138 * Unshare file descriptor table if it is being shared
2140 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2142 struct files_struct *fd = current->files;
2145 if ((unshare_flags & CLONE_FILES) &&
2146 (fd && atomic_read(&fd->count) > 1)) {
2147 *new_fdp = dup_fd(fd, &error);
2156 * unshare allows a process to 'unshare' part of the process
2157 * context which was originally shared using clone. copy_*
2158 * functions used by do_fork() cannot be used here directly
2159 * because they modify an inactive task_struct that is being
2160 * constructed. Here we are modifying the current, active,
2163 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2165 struct fs_struct *fs, *new_fs = NULL;
2166 struct files_struct *fd, *new_fd = NULL;
2167 struct cred *new_cred = NULL;
2168 struct nsproxy *new_nsproxy = NULL;
2173 * If unsharing a user namespace must also unshare the thread group
2174 * and unshare the filesystem root and working directories.
2176 if (unshare_flags & CLONE_NEWUSER)
2177 unshare_flags |= CLONE_THREAD | CLONE_FS;
2179 * If unsharing vm, must also unshare signal handlers.
2181 if (unshare_flags & CLONE_VM)
2182 unshare_flags |= CLONE_SIGHAND;
2184 * If unsharing a signal handlers, must also unshare the signal queues.
2186 if (unshare_flags & CLONE_SIGHAND)
2187 unshare_flags |= CLONE_THREAD;
2189 * If unsharing namespace, must also unshare filesystem information.
2191 if (unshare_flags & CLONE_NEWNS)
2192 unshare_flags |= CLONE_FS;
2194 err = check_unshare_flags(unshare_flags);
2196 goto bad_unshare_out;
2198 * CLONE_NEWIPC must also detach from the undolist: after switching
2199 * to a new ipc namespace, the semaphore arrays from the old
2200 * namespace are unreachable.
2202 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2204 err = unshare_fs(unshare_flags, &new_fs);
2206 goto bad_unshare_out;
2207 err = unshare_fd(unshare_flags, &new_fd);
2209 goto bad_unshare_cleanup_fs;
2210 err = unshare_userns(unshare_flags, &new_cred);
2212 goto bad_unshare_cleanup_fd;
2213 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2216 goto bad_unshare_cleanup_cred;
2218 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2221 * CLONE_SYSVSEM is equivalent to sys_exit().
2225 if (unshare_flags & CLONE_NEWIPC) {
2226 /* Orphan segments in old ns (see sem above). */
2228 shm_init_task(current);
2232 switch_task_namespaces(current, new_nsproxy);
2238 spin_lock(&fs->lock);
2239 current->fs = new_fs;
2244 spin_unlock(&fs->lock);
2248 fd = current->files;
2249 current->files = new_fd;
2253 task_unlock(current);
2256 /* Install the new user namespace */
2257 commit_creds(new_cred);
2262 bad_unshare_cleanup_cred:
2265 bad_unshare_cleanup_fd:
2267 put_files_struct(new_fd);
2269 bad_unshare_cleanup_fs:
2271 free_fs_struct(new_fs);
2278 * Helper to unshare the files of the current task.
2279 * We don't want to expose copy_files internals to
2280 * the exec layer of the kernel.
2283 int unshare_files(struct files_struct **displaced)
2285 struct task_struct *task = current;
2286 struct files_struct *copy = NULL;
2289 error = unshare_fd(CLONE_FILES, ©);
2290 if (error || !copy) {
2294 *displaced = task->files;
2301 int sysctl_max_threads(struct ctl_table *table, int write,
2302 void __user *buffer, size_t *lenp, loff_t *ppos)
2306 int threads = max_threads;
2307 int min = MIN_THREADS;
2308 int max = MAX_THREADS;
2315 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2319 set_max_threads(threads);