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/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/numa_balancing.h>
20 #include <linux/sched/stat.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/sched/cputime.h>
24 #include <linux/rtmutex.h>
25 #include <linux/init.h>
26 #include <linux/unistd.h>
27 #include <linux/module.h>
28 #include <linux/vmalloc.h>
29 #include <linux/completion.h>
30 #include <linux/personality.h>
31 #include <linux/mempolicy.h>
32 #include <linux/sem.h>
33 #include <linux/file.h>
34 #include <linux/fdtable.h>
35 #include <linux/iocontext.h>
36 #include <linux/key.h>
37 #include <linux/binfmts.h>
38 #include <linux/mman.h>
39 #include <linux/mmu_notifier.h>
42 #include <linux/vmacache.h>
43 #include <linux/nsproxy.h>
44 #include <linux/capability.h>
45 #include <linux/cpu.h>
46 #include <linux/cgroup.h>
47 #include <linux/security.h>
48 #include <linux/hugetlb.h>
49 #include <linux/seccomp.h>
50 #include <linux/swap.h>
51 #include <linux/syscalls.h>
52 #include <linux/jiffies.h>
53 #include <linux/futex.h>
54 #include <linux/compat.h>
55 #include <linux/kthread.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/rcupdate.h>
58 #include <linux/ptrace.h>
59 #include <linux/mount.h>
60 #include <linux/audit.h>
61 #include <linux/memcontrol.h>
62 #include <linux/ftrace.h>
63 #include <linux/proc_fs.h>
64 #include <linux/profile.h>
65 #include <linux/rmap.h>
66 #include <linux/ksm.h>
67 #include <linux/acct.h>
68 #include <linux/userfaultfd_k.h>
69 #include <linux/tsacct_kern.h>
70 #include <linux/cn_proc.h>
71 #include <linux/freezer.h>
72 #include <linux/delayacct.h>
73 #include <linux/taskstats_kern.h>
74 #include <linux/random.h>
75 #include <linux/tty.h>
76 #include <linux/blkdev.h>
77 #include <linux/fs_struct.h>
78 #include <linux/magic.h>
79 #include <linux/perf_event.h>
80 #include <linux/posix-timers.h>
81 #include <linux/user-return-notifier.h>
82 #include <linux/oom.h>
83 #include <linux/khugepaged.h>
84 #include <linux/signalfd.h>
85 #include <linux/uprobes.h>
86 #include <linux/aio.h>
87 #include <linux/compiler.h>
88 #include <linux/sysctl.h>
89 #include <linux/kcov.h>
90 #include <linux/livepatch.h>
92 #include <asm/pgtable.h>
93 #include <asm/pgalloc.h>
94 #include <linux/uaccess.h>
95 #include <asm/mmu_context.h>
96 #include <asm/cacheflush.h>
97 #include <asm/tlbflush.h>
99 #include <trace/events/sched.h>
101 #define CREATE_TRACE_POINTS
102 #include <trace/events/task.h>
105 * Minimum number of threads to boot the kernel
107 #define MIN_THREADS 20
110 * Maximum number of threads
112 #define MAX_THREADS FUTEX_TID_MASK
115 * Protected counters by write_lock_irq(&tasklist_lock)
117 unsigned long total_forks; /* Handle normal Linux uptimes. */
118 int nr_threads; /* The idle threads do not count.. */
120 int max_threads; /* tunable limit on nr_threads */
122 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
124 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
126 #ifdef CONFIG_PROVE_RCU
127 int lockdep_tasklist_lock_is_held(void)
129 return lockdep_is_held(&tasklist_lock);
131 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
132 #endif /* #ifdef CONFIG_PROVE_RCU */
134 int nr_processes(void)
139 for_each_possible_cpu(cpu)
140 total += per_cpu(process_counts, cpu);
145 void __weak arch_release_task_struct(struct task_struct *tsk)
149 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
150 static struct kmem_cache *task_struct_cachep;
152 static inline struct task_struct *alloc_task_struct_node(int node)
154 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
157 static inline void free_task_struct(struct task_struct *tsk)
159 kmem_cache_free(task_struct_cachep, tsk);
163 void __weak arch_release_thread_stack(unsigned long *stack)
167 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
170 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
171 * kmemcache based allocator.
173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
175 #ifdef CONFIG_VMAP_STACK
177 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
178 * flush. Try to minimize the number of calls by caching stacks.
180 #define NR_CACHED_STACKS 2
181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
183 static int free_vm_stack_cache(unsigned int cpu)
185 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
188 for (i = 0; i < NR_CACHED_STACKS; i++) {
189 struct vm_struct *vm_stack = cached_vm_stacks[i];
194 vfree(vm_stack->addr);
195 cached_vm_stacks[i] = NULL;
202 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
204 #ifdef CONFIG_VMAP_STACK
209 for (i = 0; i < NR_CACHED_STACKS; i++) {
210 struct vm_struct *s = this_cpu_read(cached_stacks[i]);
214 this_cpu_write(cached_stacks[i], NULL);
216 tsk->stack_vm_area = s;
222 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
223 VMALLOC_START, VMALLOC_END,
226 0, node, __builtin_return_address(0));
229 * We can't call find_vm_area() in interrupt context, and
230 * free_thread_stack() can be called in interrupt context,
231 * so cache the vm_struct.
234 tsk->stack_vm_area = find_vm_area(stack);
237 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
240 return page ? page_address(page) : NULL;
244 static inline void free_thread_stack(struct task_struct *tsk)
246 #ifdef CONFIG_VMAP_STACK
247 if (task_stack_vm_area(tsk)) {
251 local_irq_save(flags);
252 for (i = 0; i < NR_CACHED_STACKS; i++) {
253 if (this_cpu_read(cached_stacks[i]))
256 this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
257 local_irq_restore(flags);
260 local_irq_restore(flags);
262 vfree_atomic(tsk->stack);
267 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
270 static struct kmem_cache *thread_stack_cache;
272 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
275 return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
278 static void free_thread_stack(struct task_struct *tsk)
280 kmem_cache_free(thread_stack_cache, tsk->stack);
283 void thread_stack_cache_init(void)
285 thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
286 THREAD_SIZE, 0, NULL);
287 BUG_ON(thread_stack_cache == NULL);
292 /* SLAB cache for signal_struct structures (tsk->signal) */
293 static struct kmem_cache *signal_cachep;
295 /* SLAB cache for sighand_struct structures (tsk->sighand) */
296 struct kmem_cache *sighand_cachep;
298 /* SLAB cache for files_struct structures (tsk->files) */
299 struct kmem_cache *files_cachep;
301 /* SLAB cache for fs_struct structures (tsk->fs) */
302 struct kmem_cache *fs_cachep;
304 /* SLAB cache for vm_area_struct structures */
305 struct kmem_cache *vm_area_cachep;
307 /* SLAB cache for mm_struct structures (tsk->mm) */
308 static struct kmem_cache *mm_cachep;
310 static void account_kernel_stack(struct task_struct *tsk, int account)
312 void *stack = task_stack_page(tsk);
313 struct vm_struct *vm = task_stack_vm_area(tsk);
315 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
320 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
322 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
323 mod_zone_page_state(page_zone(vm->pages[i]),
325 PAGE_SIZE / 1024 * account);
328 /* All stack pages belong to the same memcg. */
329 memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
330 account * (THREAD_SIZE / 1024));
333 * All stack pages are in the same zone and belong to the
336 struct page *first_page = virt_to_page(stack);
338 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
339 THREAD_SIZE / 1024 * account);
341 memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
342 account * (THREAD_SIZE / 1024));
346 static void release_task_stack(struct task_struct *tsk)
348 if (WARN_ON(tsk->state != TASK_DEAD))
349 return; /* Better to leak the stack than to free prematurely */
351 account_kernel_stack(tsk, -1);
352 arch_release_thread_stack(tsk->stack);
353 free_thread_stack(tsk);
355 #ifdef CONFIG_VMAP_STACK
356 tsk->stack_vm_area = NULL;
360 #ifdef CONFIG_THREAD_INFO_IN_TASK
361 void put_task_stack(struct task_struct *tsk)
363 if (atomic_dec_and_test(&tsk->stack_refcount))
364 release_task_stack(tsk);
368 void free_task(struct task_struct *tsk)
370 #ifndef CONFIG_THREAD_INFO_IN_TASK
372 * The task is finally done with both the stack and thread_info,
375 release_task_stack(tsk);
378 * If the task had a separate stack allocation, it should be gone
381 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
383 rt_mutex_debug_task_free(tsk);
384 ftrace_graph_exit_task(tsk);
385 put_seccomp_filter(tsk);
386 arch_release_task_struct(tsk);
387 if (tsk->flags & PF_KTHREAD)
388 free_kthread_struct(tsk);
389 free_task_struct(tsk);
391 EXPORT_SYMBOL(free_task);
393 static inline void free_signal_struct(struct signal_struct *sig)
395 taskstats_tgid_free(sig);
396 sched_autogroup_exit(sig);
398 * __mmdrop is not safe to call from softirq context on x86 due to
399 * pgd_dtor so postpone it to the async context
402 mmdrop_async(sig->oom_mm);
403 kmem_cache_free(signal_cachep, sig);
406 static inline void put_signal_struct(struct signal_struct *sig)
408 if (atomic_dec_and_test(&sig->sigcnt))
409 free_signal_struct(sig);
412 void __put_task_struct(struct task_struct *tsk)
414 WARN_ON(!tsk->exit_state);
415 WARN_ON(atomic_read(&tsk->usage));
416 WARN_ON(tsk == current);
420 security_task_free(tsk);
422 delayacct_tsk_free(tsk);
423 put_signal_struct(tsk->signal);
425 if (!profile_handoff_task(tsk))
428 EXPORT_SYMBOL_GPL(__put_task_struct);
430 void __init __weak arch_task_cache_init(void) { }
435 static void set_max_threads(unsigned int max_threads_suggested)
440 * The number of threads shall be limited such that the thread
441 * structures may only consume a small part of the available memory.
443 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
444 threads = MAX_THREADS;
446 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
447 (u64) THREAD_SIZE * 8UL);
449 if (threads > max_threads_suggested)
450 threads = max_threads_suggested;
452 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
455 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
456 /* Initialized by the architecture: */
457 int arch_task_struct_size __read_mostly;
460 void __init fork_init(void)
463 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
464 #ifndef ARCH_MIN_TASKALIGN
465 #define ARCH_MIN_TASKALIGN 0
467 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
469 /* create a slab on which task_structs can be allocated */
470 task_struct_cachep = kmem_cache_create("task_struct",
471 arch_task_struct_size, align,
472 SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
475 /* do the arch specific task caches init */
476 arch_task_cache_init();
478 set_max_threads(MAX_THREADS);
480 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
481 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
482 init_task.signal->rlim[RLIMIT_SIGPENDING] =
483 init_task.signal->rlim[RLIMIT_NPROC];
485 for (i = 0; i < UCOUNT_COUNTS; i++) {
486 init_user_ns.ucount_max[i] = max_threads/2;
489 #ifdef CONFIG_VMAP_STACK
490 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
491 NULL, free_vm_stack_cache);
495 int __weak arch_dup_task_struct(struct task_struct *dst,
496 struct task_struct *src)
502 void set_task_stack_end_magic(struct task_struct *tsk)
504 unsigned long *stackend;
506 stackend = end_of_stack(tsk);
507 *stackend = STACK_END_MAGIC; /* for overflow detection */
510 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
512 struct task_struct *tsk;
513 unsigned long *stack;
514 struct vm_struct *stack_vm_area;
517 if (node == NUMA_NO_NODE)
518 node = tsk_fork_get_node(orig);
519 tsk = alloc_task_struct_node(node);
523 stack = alloc_thread_stack_node(tsk, node);
527 stack_vm_area = task_stack_vm_area(tsk);
529 err = arch_dup_task_struct(tsk, orig);
532 * arch_dup_task_struct() clobbers the stack-related fields. Make
533 * sure they're properly initialized before using any stack-related
537 #ifdef CONFIG_VMAP_STACK
538 tsk->stack_vm_area = stack_vm_area;
540 #ifdef CONFIG_THREAD_INFO_IN_TASK
541 atomic_set(&tsk->stack_refcount, 1);
547 #ifdef CONFIG_SECCOMP
549 * We must handle setting up seccomp filters once we're under
550 * the sighand lock in case orig has changed between now and
551 * then. Until then, filter must be NULL to avoid messing up
552 * the usage counts on the error path calling free_task.
554 tsk->seccomp.filter = NULL;
557 setup_thread_stack(tsk, orig);
558 clear_user_return_notifier(tsk);
559 clear_tsk_need_resched(tsk);
560 set_task_stack_end_magic(tsk);
562 #ifdef CONFIG_CC_STACKPROTECTOR
563 tsk->stack_canary = get_random_int();
567 * One for us, one for whoever does the "release_task()" (usually
570 atomic_set(&tsk->usage, 2);
571 #ifdef CONFIG_BLK_DEV_IO_TRACE
574 tsk->splice_pipe = NULL;
575 tsk->task_frag.page = NULL;
576 tsk->wake_q.next = NULL;
578 account_kernel_stack(tsk, 1);
585 free_thread_stack(tsk);
587 free_task_struct(tsk);
592 static __latent_entropy int dup_mmap(struct mm_struct *mm,
593 struct mm_struct *oldmm)
595 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
596 struct rb_node **rb_link, *rb_parent;
598 unsigned long charge;
601 uprobe_start_dup_mmap();
602 if (down_write_killable(&oldmm->mmap_sem)) {
604 goto fail_uprobe_end;
606 flush_cache_dup_mm(oldmm);
607 uprobe_dup_mmap(oldmm, mm);
609 * Not linked in yet - no deadlock potential:
611 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
613 /* No ordering required: file already has been exposed. */
614 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
616 mm->total_vm = oldmm->total_vm;
617 mm->data_vm = oldmm->data_vm;
618 mm->exec_vm = oldmm->exec_vm;
619 mm->stack_vm = oldmm->stack_vm;
621 rb_link = &mm->mm_rb.rb_node;
624 retval = ksm_fork(mm, oldmm);
627 retval = khugepaged_fork(mm, oldmm);
632 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
635 if (mpnt->vm_flags & VM_DONTCOPY) {
636 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
640 if (mpnt->vm_flags & VM_ACCOUNT) {
641 unsigned long len = vma_pages(mpnt);
643 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
647 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
651 INIT_LIST_HEAD(&tmp->anon_vma_chain);
652 retval = vma_dup_policy(mpnt, tmp);
654 goto fail_nomem_policy;
656 retval = dup_userfaultfd(tmp, &uf);
658 goto fail_nomem_anon_vma_fork;
659 if (anon_vma_fork(tmp, mpnt))
660 goto fail_nomem_anon_vma_fork;
661 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
662 tmp->vm_next = tmp->vm_prev = NULL;
665 struct inode *inode = file_inode(file);
666 struct address_space *mapping = file->f_mapping;
669 if (tmp->vm_flags & VM_DENYWRITE)
670 atomic_dec(&inode->i_writecount);
671 i_mmap_lock_write(mapping);
672 if (tmp->vm_flags & VM_SHARED)
673 atomic_inc(&mapping->i_mmap_writable);
674 flush_dcache_mmap_lock(mapping);
675 /* insert tmp into the share list, just after mpnt */
676 vma_interval_tree_insert_after(tmp, mpnt,
678 flush_dcache_mmap_unlock(mapping);
679 i_mmap_unlock_write(mapping);
683 * Clear hugetlb-related page reserves for children. This only
684 * affects MAP_PRIVATE mappings. Faults generated by the child
685 * are not guaranteed to succeed, even if read-only
687 if (is_vm_hugetlb_page(tmp))
688 reset_vma_resv_huge_pages(tmp);
691 * Link in the new vma and copy the page table entries.
694 pprev = &tmp->vm_next;
698 __vma_link_rb(mm, tmp, rb_link, rb_parent);
699 rb_link = &tmp->vm_rb.rb_right;
700 rb_parent = &tmp->vm_rb;
703 retval = copy_page_range(mm, oldmm, mpnt);
705 if (tmp->vm_ops && tmp->vm_ops->open)
706 tmp->vm_ops->open(tmp);
711 /* a new mm has just been created */
712 arch_dup_mmap(oldmm, mm);
715 up_write(&mm->mmap_sem);
717 up_write(&oldmm->mmap_sem);
718 dup_userfaultfd_complete(&uf);
720 uprobe_end_dup_mmap();
722 fail_nomem_anon_vma_fork:
723 mpol_put(vma_policy(tmp));
725 kmem_cache_free(vm_area_cachep, tmp);
728 vm_unacct_memory(charge);
732 static inline int mm_alloc_pgd(struct mm_struct *mm)
734 mm->pgd = pgd_alloc(mm);
735 if (unlikely(!mm->pgd))
740 static inline void mm_free_pgd(struct mm_struct *mm)
742 pgd_free(mm, mm->pgd);
745 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
747 down_write(&oldmm->mmap_sem);
748 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
749 up_write(&oldmm->mmap_sem);
752 #define mm_alloc_pgd(mm) (0)
753 #define mm_free_pgd(mm)
754 #endif /* CONFIG_MMU */
756 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
758 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
759 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
761 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
763 static int __init coredump_filter_setup(char *s)
765 default_dump_filter =
766 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
767 MMF_DUMP_FILTER_MASK;
771 __setup("coredump_filter=", coredump_filter_setup);
773 #include <linux/init_task.h>
775 static void mm_init_aio(struct mm_struct *mm)
778 spin_lock_init(&mm->ioctx_lock);
779 mm->ioctx_table = NULL;
783 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
790 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
791 struct user_namespace *user_ns)
795 mm->vmacache_seqnum = 0;
796 atomic_set(&mm->mm_users, 1);
797 atomic_set(&mm->mm_count, 1);
798 init_rwsem(&mm->mmap_sem);
799 INIT_LIST_HEAD(&mm->mmlist);
800 mm->core_state = NULL;
801 atomic_long_set(&mm->nr_ptes, 0);
806 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
807 spin_lock_init(&mm->page_table_lock);
810 mm_init_owner(mm, p);
811 mmu_notifier_mm_init(mm);
812 clear_tlb_flush_pending(mm);
813 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
814 mm->pmd_huge_pte = NULL;
818 mm->flags = current->mm->flags & MMF_INIT_MASK;
819 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
821 mm->flags = default_dump_filter;
825 if (mm_alloc_pgd(mm))
828 if (init_new_context(p, mm))
831 mm->user_ns = get_user_ns(user_ns);
841 static void check_mm(struct mm_struct *mm)
845 for (i = 0; i < NR_MM_COUNTERS; i++) {
846 long x = atomic_long_read(&mm->rss_stat.count[i]);
849 printk(KERN_ALERT "BUG: Bad rss-counter state "
850 "mm:%p idx:%d val:%ld\n", mm, i, x);
853 if (atomic_long_read(&mm->nr_ptes))
854 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
855 atomic_long_read(&mm->nr_ptes));
857 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
860 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
861 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
866 * Allocate and initialize an mm_struct.
868 struct mm_struct *mm_alloc(void)
870 struct mm_struct *mm;
876 memset(mm, 0, sizeof(*mm));
877 return mm_init(mm, current, current_user_ns());
881 * Called when the last reference to the mm
882 * is dropped: either by a lazy thread or by
883 * mmput. Free the page directory and the mm.
885 void __mmdrop(struct mm_struct *mm)
887 BUG_ON(mm == &init_mm);
890 mmu_notifier_mm_destroy(mm);
892 put_user_ns(mm->user_ns);
895 EXPORT_SYMBOL_GPL(__mmdrop);
897 static inline void __mmput(struct mm_struct *mm)
899 VM_BUG_ON(atomic_read(&mm->mm_users));
901 uprobe_clear_state(mm);
904 khugepaged_exit(mm); /* must run before exit_mmap */
906 mm_put_huge_zero_page(mm);
907 set_mm_exe_file(mm, NULL);
908 if (!list_empty(&mm->mmlist)) {
909 spin_lock(&mmlist_lock);
910 list_del(&mm->mmlist);
911 spin_unlock(&mmlist_lock);
914 module_put(mm->binfmt->module);
915 set_bit(MMF_OOM_SKIP, &mm->flags);
920 * Decrement the use count and release all resources for an mm.
922 void mmput(struct mm_struct *mm)
926 if (atomic_dec_and_test(&mm->mm_users))
929 EXPORT_SYMBOL_GPL(mmput);
932 static void mmput_async_fn(struct work_struct *work)
934 struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
938 void mmput_async(struct mm_struct *mm)
940 if (atomic_dec_and_test(&mm->mm_users)) {
941 INIT_WORK(&mm->async_put_work, mmput_async_fn);
942 schedule_work(&mm->async_put_work);
948 * set_mm_exe_file - change a reference to the mm's executable file
950 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
952 * Main users are mmput() and sys_execve(). Callers prevent concurrent
953 * invocations: in mmput() nobody alive left, in execve task is single
954 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
955 * mm->exe_file, but does so without using set_mm_exe_file() in order
956 * to do avoid the need for any locks.
958 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
960 struct file *old_exe_file;
963 * It is safe to dereference the exe_file without RCU as
964 * this function is only called if nobody else can access
965 * this mm -- see comment above for justification.
967 old_exe_file = rcu_dereference_raw(mm->exe_file);
970 get_file(new_exe_file);
971 rcu_assign_pointer(mm->exe_file, new_exe_file);
977 * get_mm_exe_file - acquire a reference to the mm's executable file
979 * Returns %NULL if mm has no associated executable file.
980 * User must release file via fput().
982 struct file *get_mm_exe_file(struct mm_struct *mm)
984 struct file *exe_file;
987 exe_file = rcu_dereference(mm->exe_file);
988 if (exe_file && !get_file_rcu(exe_file))
993 EXPORT_SYMBOL(get_mm_exe_file);
996 * get_task_exe_file - acquire a reference to the task's executable file
998 * Returns %NULL if task's mm (if any) has no associated executable file or
999 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1000 * User must release file via fput().
1002 struct file *get_task_exe_file(struct task_struct *task)
1004 struct file *exe_file = NULL;
1005 struct mm_struct *mm;
1010 if (!(task->flags & PF_KTHREAD))
1011 exe_file = get_mm_exe_file(mm);
1016 EXPORT_SYMBOL(get_task_exe_file);
1019 * get_task_mm - acquire a reference to the task's mm
1021 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1022 * this kernel workthread has transiently adopted a user mm with use_mm,
1023 * to do its AIO) is not set and if so returns a reference to it, after
1024 * bumping up the use count. User must release the mm via mmput()
1025 * after use. Typically used by /proc and ptrace.
1027 struct mm_struct *get_task_mm(struct task_struct *task)
1029 struct mm_struct *mm;
1034 if (task->flags & PF_KTHREAD)
1042 EXPORT_SYMBOL_GPL(get_task_mm);
1044 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1046 struct mm_struct *mm;
1049 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1051 return ERR_PTR(err);
1053 mm = get_task_mm(task);
1054 if (mm && mm != current->mm &&
1055 !ptrace_may_access(task, mode)) {
1057 mm = ERR_PTR(-EACCES);
1059 mutex_unlock(&task->signal->cred_guard_mutex);
1064 static void complete_vfork_done(struct task_struct *tsk)
1066 struct completion *vfork;
1069 vfork = tsk->vfork_done;
1070 if (likely(vfork)) {
1071 tsk->vfork_done = NULL;
1077 static int wait_for_vfork_done(struct task_struct *child,
1078 struct completion *vfork)
1082 freezer_do_not_count();
1083 killed = wait_for_completion_killable(vfork);
1088 child->vfork_done = NULL;
1092 put_task_struct(child);
1096 /* Please note the differences between mmput and mm_release.
1097 * mmput is called whenever we stop holding onto a mm_struct,
1098 * error success whatever.
1100 * mm_release is called after a mm_struct has been removed
1101 * from the current process.
1103 * This difference is important for error handling, when we
1104 * only half set up a mm_struct for a new process and need to restore
1105 * the old one. Because we mmput the new mm_struct before
1106 * restoring the old one. . .
1107 * Eric Biederman 10 January 1998
1109 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1111 /* Get rid of any futexes when releasing the mm */
1113 if (unlikely(tsk->robust_list)) {
1114 exit_robust_list(tsk);
1115 tsk->robust_list = NULL;
1117 #ifdef CONFIG_COMPAT
1118 if (unlikely(tsk->compat_robust_list)) {
1119 compat_exit_robust_list(tsk);
1120 tsk->compat_robust_list = NULL;
1123 if (unlikely(!list_empty(&tsk->pi_state_list)))
1124 exit_pi_state_list(tsk);
1127 uprobe_free_utask(tsk);
1129 /* Get rid of any cached register state */
1130 deactivate_mm(tsk, mm);
1133 * Signal userspace if we're not exiting with a core dump
1134 * because we want to leave the value intact for debugging
1137 if (tsk->clear_child_tid) {
1138 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1139 atomic_read(&mm->mm_users) > 1) {
1141 * We don't check the error code - if userspace has
1142 * not set up a proper pointer then tough luck.
1144 put_user(0, tsk->clear_child_tid);
1145 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1148 tsk->clear_child_tid = NULL;
1152 * All done, finally we can wake up parent and return this mm to him.
1153 * Also kthread_stop() uses this completion for synchronization.
1155 if (tsk->vfork_done)
1156 complete_vfork_done(tsk);
1160 * Allocate a new mm structure and copy contents from the
1161 * mm structure of the passed in task structure.
1163 static struct mm_struct *dup_mm(struct task_struct *tsk)
1165 struct mm_struct *mm, *oldmm = current->mm;
1172 memcpy(mm, oldmm, sizeof(*mm));
1174 if (!mm_init(mm, tsk, mm->user_ns))
1177 err = dup_mmap(mm, oldmm);
1181 mm->hiwater_rss = get_mm_rss(mm);
1182 mm->hiwater_vm = mm->total_vm;
1184 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1190 /* don't put binfmt in mmput, we haven't got module yet */
1198 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1200 struct mm_struct *mm, *oldmm;
1203 tsk->min_flt = tsk->maj_flt = 0;
1204 tsk->nvcsw = tsk->nivcsw = 0;
1205 #ifdef CONFIG_DETECT_HUNG_TASK
1206 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1210 tsk->active_mm = NULL;
1213 * Are we cloning a kernel thread?
1215 * We need to steal a active VM for that..
1217 oldmm = current->mm;
1221 /* initialize the new vmacache entries */
1222 vmacache_flush(tsk);
1224 if (clone_flags & CLONE_VM) {
1237 tsk->active_mm = mm;
1244 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1246 struct fs_struct *fs = current->fs;
1247 if (clone_flags & CLONE_FS) {
1248 /* tsk->fs is already what we want */
1249 spin_lock(&fs->lock);
1251 spin_unlock(&fs->lock);
1255 spin_unlock(&fs->lock);
1258 tsk->fs = copy_fs_struct(fs);
1264 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1266 struct files_struct *oldf, *newf;
1270 * A background process may not have any files ...
1272 oldf = current->files;
1276 if (clone_flags & CLONE_FILES) {
1277 atomic_inc(&oldf->count);
1281 newf = dup_fd(oldf, &error);
1291 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1294 struct io_context *ioc = current->io_context;
1295 struct io_context *new_ioc;
1300 * Share io context with parent, if CLONE_IO is set
1302 if (clone_flags & CLONE_IO) {
1304 tsk->io_context = ioc;
1305 } else if (ioprio_valid(ioc->ioprio)) {
1306 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1307 if (unlikely(!new_ioc))
1310 new_ioc->ioprio = ioc->ioprio;
1311 put_io_context(new_ioc);
1317 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1319 struct sighand_struct *sig;
1321 if (clone_flags & CLONE_SIGHAND) {
1322 atomic_inc(¤t->sighand->count);
1325 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1326 rcu_assign_pointer(tsk->sighand, sig);
1330 atomic_set(&sig->count, 1);
1331 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1335 void __cleanup_sighand(struct sighand_struct *sighand)
1337 if (atomic_dec_and_test(&sighand->count)) {
1338 signalfd_cleanup(sighand);
1340 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1341 * without an RCU grace period, see __lock_task_sighand().
1343 kmem_cache_free(sighand_cachep, sighand);
1347 #ifdef CONFIG_POSIX_TIMERS
1349 * Initialize POSIX timer handling for a thread group.
1351 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1353 unsigned long cpu_limit;
1355 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1356 if (cpu_limit != RLIM_INFINITY) {
1357 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1358 sig->cputimer.running = true;
1361 /* The timer lists. */
1362 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1363 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1364 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1367 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1370 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1372 struct signal_struct *sig;
1374 if (clone_flags & CLONE_THREAD)
1377 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1382 sig->nr_threads = 1;
1383 atomic_set(&sig->live, 1);
1384 atomic_set(&sig->sigcnt, 1);
1386 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1387 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1388 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1390 init_waitqueue_head(&sig->wait_chldexit);
1391 sig->curr_target = tsk;
1392 init_sigpending(&sig->shared_pending);
1393 seqlock_init(&sig->stats_lock);
1394 prev_cputime_init(&sig->prev_cputime);
1396 #ifdef CONFIG_POSIX_TIMERS
1397 INIT_LIST_HEAD(&sig->posix_timers);
1398 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1399 sig->real_timer.function = it_real_fn;
1402 task_lock(current->group_leader);
1403 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1404 task_unlock(current->group_leader);
1406 posix_cpu_timers_init_group(sig);
1408 tty_audit_fork(sig);
1409 sched_autogroup_fork(sig);
1411 sig->oom_score_adj = current->signal->oom_score_adj;
1412 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1414 mutex_init(&sig->cred_guard_mutex);
1419 static void copy_seccomp(struct task_struct *p)
1421 #ifdef CONFIG_SECCOMP
1423 * Must be called with sighand->lock held, which is common to
1424 * all threads in the group. Holding cred_guard_mutex is not
1425 * needed because this new task is not yet running and cannot
1428 assert_spin_locked(¤t->sighand->siglock);
1430 /* Ref-count the new filter user, and assign it. */
1431 get_seccomp_filter(current);
1432 p->seccomp = current->seccomp;
1435 * Explicitly enable no_new_privs here in case it got set
1436 * between the task_struct being duplicated and holding the
1437 * sighand lock. The seccomp state and nnp must be in sync.
1439 if (task_no_new_privs(current))
1440 task_set_no_new_privs(p);
1443 * If the parent gained a seccomp mode after copying thread
1444 * flags and between before we held the sighand lock, we have
1445 * to manually enable the seccomp thread flag here.
1447 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1448 set_tsk_thread_flag(p, TIF_SECCOMP);
1452 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1454 current->clear_child_tid = tidptr;
1456 return task_pid_vnr(current);
1459 static void rt_mutex_init_task(struct task_struct *p)
1461 raw_spin_lock_init(&p->pi_lock);
1462 #ifdef CONFIG_RT_MUTEXES
1463 p->pi_waiters = RB_ROOT;
1464 p->pi_waiters_leftmost = NULL;
1465 p->pi_top_task = NULL;
1466 p->pi_blocked_on = NULL;
1470 #ifdef CONFIG_POSIX_TIMERS
1472 * Initialize POSIX timer handling for a single task.
1474 static void posix_cpu_timers_init(struct task_struct *tsk)
1476 tsk->cputime_expires.prof_exp = 0;
1477 tsk->cputime_expires.virt_exp = 0;
1478 tsk->cputime_expires.sched_exp = 0;
1479 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1480 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1481 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1484 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1488 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1490 task->pids[type].pid = pid;
1493 static inline void rcu_copy_process(struct task_struct *p)
1495 #ifdef CONFIG_PREEMPT_RCU
1496 p->rcu_read_lock_nesting = 0;
1497 p->rcu_read_unlock_special.s = 0;
1498 p->rcu_blocked_node = NULL;
1499 INIT_LIST_HEAD(&p->rcu_node_entry);
1500 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1501 #ifdef CONFIG_TASKS_RCU
1502 p->rcu_tasks_holdout = false;
1503 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1504 p->rcu_tasks_idle_cpu = -1;
1505 #endif /* #ifdef CONFIG_TASKS_RCU */
1509 * This creates a new process as a copy of the old one,
1510 * but does not actually start it yet.
1512 * It copies the registers, and all the appropriate
1513 * parts of the process environment (as per the clone
1514 * flags). The actual kick-off is left to the caller.
1516 static __latent_entropy struct task_struct *copy_process(
1517 unsigned long clone_flags,
1518 unsigned long stack_start,
1519 unsigned long stack_size,
1520 int __user *child_tidptr,
1527 struct task_struct *p;
1529 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1530 return ERR_PTR(-EINVAL);
1532 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1533 return ERR_PTR(-EINVAL);
1536 * Thread groups must share signals as well, and detached threads
1537 * can only be started up within the thread group.
1539 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1540 return ERR_PTR(-EINVAL);
1543 * Shared signal handlers imply shared VM. By way of the above,
1544 * thread groups also imply shared VM. Blocking this case allows
1545 * for various simplifications in other code.
1547 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1548 return ERR_PTR(-EINVAL);
1551 * Siblings of global init remain as zombies on exit since they are
1552 * not reaped by their parent (swapper). To solve this and to avoid
1553 * multi-rooted process trees, prevent global and container-inits
1554 * from creating siblings.
1556 if ((clone_flags & CLONE_PARENT) &&
1557 current->signal->flags & SIGNAL_UNKILLABLE)
1558 return ERR_PTR(-EINVAL);
1561 * If the new process will be in a different pid or user namespace
1562 * do not allow it to share a thread group with the forking task.
1564 if (clone_flags & CLONE_THREAD) {
1565 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1566 (task_active_pid_ns(current) !=
1567 current->nsproxy->pid_ns_for_children))
1568 return ERR_PTR(-EINVAL);
1571 retval = security_task_create(clone_flags);
1576 p = dup_task_struct(current, node);
1580 ftrace_graph_init_task(p);
1582 rt_mutex_init_task(p);
1584 #ifdef CONFIG_PROVE_LOCKING
1585 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1586 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1589 if (atomic_read(&p->real_cred->user->processes) >=
1590 task_rlimit(p, RLIMIT_NPROC)) {
1591 if (p->real_cred->user != INIT_USER &&
1592 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1595 current->flags &= ~PF_NPROC_EXCEEDED;
1597 retval = copy_creds(p, clone_flags);
1602 * If multiple threads are within copy_process(), then this check
1603 * triggers too late. This doesn't hurt, the check is only there
1604 * to stop root fork bombs.
1607 if (nr_threads >= max_threads)
1608 goto bad_fork_cleanup_count;
1610 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1611 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1612 p->flags |= PF_FORKNOEXEC;
1613 INIT_LIST_HEAD(&p->children);
1614 INIT_LIST_HEAD(&p->sibling);
1615 rcu_copy_process(p);
1616 p->vfork_done = NULL;
1617 spin_lock_init(&p->alloc_lock);
1619 init_sigpending(&p->pending);
1621 p->utime = p->stime = p->gtime = 0;
1622 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1623 p->utimescaled = p->stimescaled = 0;
1625 prev_cputime_init(&p->prev_cputime);
1627 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1628 seqcount_init(&p->vtime_seqcount);
1630 p->vtime_snap_whence = VTIME_INACTIVE;
1633 #if defined(SPLIT_RSS_COUNTING)
1634 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1637 p->default_timer_slack_ns = current->timer_slack_ns;
1639 task_io_accounting_init(&p->ioac);
1640 acct_clear_integrals(p);
1642 posix_cpu_timers_init(p);
1644 p->start_time = ktime_get_ns();
1645 p->real_start_time = ktime_get_boot_ns();
1646 p->io_context = NULL;
1647 p->audit_context = NULL;
1650 p->mempolicy = mpol_dup(p->mempolicy);
1651 if (IS_ERR(p->mempolicy)) {
1652 retval = PTR_ERR(p->mempolicy);
1653 p->mempolicy = NULL;
1654 goto bad_fork_cleanup_threadgroup_lock;
1657 #ifdef CONFIG_CPUSETS
1658 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1659 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1660 seqcount_init(&p->mems_allowed_seq);
1662 #ifdef CONFIG_TRACE_IRQFLAGS
1664 p->hardirqs_enabled = 0;
1665 p->hardirq_enable_ip = 0;
1666 p->hardirq_enable_event = 0;
1667 p->hardirq_disable_ip = _THIS_IP_;
1668 p->hardirq_disable_event = 0;
1669 p->softirqs_enabled = 1;
1670 p->softirq_enable_ip = _THIS_IP_;
1671 p->softirq_enable_event = 0;
1672 p->softirq_disable_ip = 0;
1673 p->softirq_disable_event = 0;
1674 p->hardirq_context = 0;
1675 p->softirq_context = 0;
1678 p->pagefault_disabled = 0;
1680 #ifdef CONFIG_LOCKDEP
1681 p->lockdep_depth = 0; /* no locks held yet */
1682 p->curr_chain_key = 0;
1683 p->lockdep_recursion = 0;
1686 #ifdef CONFIG_DEBUG_MUTEXES
1687 p->blocked_on = NULL; /* not blocked yet */
1689 #ifdef CONFIG_BCACHE
1690 p->sequential_io = 0;
1691 p->sequential_io_avg = 0;
1694 /* Perform scheduler related setup. Assign this task to a CPU. */
1695 retval = sched_fork(clone_flags, p);
1697 goto bad_fork_cleanup_policy;
1699 retval = perf_event_init_task(p);
1701 goto bad_fork_cleanup_policy;
1702 retval = audit_alloc(p);
1704 goto bad_fork_cleanup_perf;
1705 /* copy all the process information */
1707 retval = security_task_alloc(p, clone_flags);
1709 goto bad_fork_cleanup_audit;
1710 retval = copy_semundo(clone_flags, p);
1712 goto bad_fork_cleanup_security;
1713 retval = copy_files(clone_flags, p);
1715 goto bad_fork_cleanup_semundo;
1716 retval = copy_fs(clone_flags, p);
1718 goto bad_fork_cleanup_files;
1719 retval = copy_sighand(clone_flags, p);
1721 goto bad_fork_cleanup_fs;
1722 retval = copy_signal(clone_flags, p);
1724 goto bad_fork_cleanup_sighand;
1725 retval = copy_mm(clone_flags, p);
1727 goto bad_fork_cleanup_signal;
1728 retval = copy_namespaces(clone_flags, p);
1730 goto bad_fork_cleanup_mm;
1731 retval = copy_io(clone_flags, p);
1733 goto bad_fork_cleanup_namespaces;
1734 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1736 goto bad_fork_cleanup_io;
1738 if (pid != &init_struct_pid) {
1739 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1741 retval = PTR_ERR(pid);
1742 goto bad_fork_cleanup_thread;
1746 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1748 * Clear TID on mm_release()?
1750 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1755 p->robust_list = NULL;
1756 #ifdef CONFIG_COMPAT
1757 p->compat_robust_list = NULL;
1759 INIT_LIST_HEAD(&p->pi_state_list);
1760 p->pi_state_cache = NULL;
1763 * sigaltstack should be cleared when sharing the same VM
1765 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1769 * Syscall tracing and stepping should be turned off in the
1770 * child regardless of CLONE_PTRACE.
1772 user_disable_single_step(p);
1773 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1774 #ifdef TIF_SYSCALL_EMU
1775 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1777 clear_all_latency_tracing(p);
1779 /* ok, now we should be set up.. */
1780 p->pid = pid_nr(pid);
1781 if (clone_flags & CLONE_THREAD) {
1782 p->exit_signal = -1;
1783 p->group_leader = current->group_leader;
1784 p->tgid = current->tgid;
1786 if (clone_flags & CLONE_PARENT)
1787 p->exit_signal = current->group_leader->exit_signal;
1789 p->exit_signal = (clone_flags & CSIGNAL);
1790 p->group_leader = p;
1795 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1796 p->dirty_paused_when = 0;
1798 p->pdeath_signal = 0;
1799 INIT_LIST_HEAD(&p->thread_group);
1800 p->task_works = NULL;
1802 cgroup_threadgroup_change_begin(current);
1804 * Ensure that the cgroup subsystem policies allow the new process to be
1805 * forked. It should be noted the the new process's css_set can be changed
1806 * between here and cgroup_post_fork() if an organisation operation is in
1809 retval = cgroup_can_fork(p);
1811 goto bad_fork_free_pid;
1814 * Make it visible to the rest of the system, but dont wake it up yet.
1815 * Need tasklist lock for parent etc handling!
1817 write_lock_irq(&tasklist_lock);
1819 /* CLONE_PARENT re-uses the old parent */
1820 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1821 p->real_parent = current->real_parent;
1822 p->parent_exec_id = current->parent_exec_id;
1824 p->real_parent = current;
1825 p->parent_exec_id = current->self_exec_id;
1828 klp_copy_process(p);
1830 spin_lock(¤t->sighand->siglock);
1833 * Copy seccomp details explicitly here, in case they were changed
1834 * before holding sighand lock.
1839 * Process group and session signals need to be delivered to just the
1840 * parent before the fork or both the parent and the child after the
1841 * fork. Restart if a signal comes in before we add the new process to
1842 * it's process group.
1843 * A fatal signal pending means that current will exit, so the new
1844 * thread can't slip out of an OOM kill (or normal SIGKILL).
1846 recalc_sigpending();
1847 if (signal_pending(current)) {
1848 spin_unlock(¤t->sighand->siglock);
1849 write_unlock_irq(&tasklist_lock);
1850 retval = -ERESTARTNOINTR;
1851 goto bad_fork_cancel_cgroup;
1854 if (likely(p->pid)) {
1855 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1857 init_task_pid(p, PIDTYPE_PID, pid);
1858 if (thread_group_leader(p)) {
1859 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1860 init_task_pid(p, PIDTYPE_SID, task_session(current));
1862 if (is_child_reaper(pid)) {
1863 ns_of_pid(pid)->child_reaper = p;
1864 p->signal->flags |= SIGNAL_UNKILLABLE;
1867 p->signal->leader_pid = pid;
1868 p->signal->tty = tty_kref_get(current->signal->tty);
1870 * Inherit has_child_subreaper flag under the same
1871 * tasklist_lock with adding child to the process tree
1872 * for propagate_has_child_subreaper optimization.
1874 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1875 p->real_parent->signal->is_child_subreaper;
1876 list_add_tail(&p->sibling, &p->real_parent->children);
1877 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1878 attach_pid(p, PIDTYPE_PGID);
1879 attach_pid(p, PIDTYPE_SID);
1880 __this_cpu_inc(process_counts);
1882 current->signal->nr_threads++;
1883 atomic_inc(¤t->signal->live);
1884 atomic_inc(¤t->signal->sigcnt);
1885 list_add_tail_rcu(&p->thread_group,
1886 &p->group_leader->thread_group);
1887 list_add_tail_rcu(&p->thread_node,
1888 &p->signal->thread_head);
1890 attach_pid(p, PIDTYPE_PID);
1895 spin_unlock(¤t->sighand->siglock);
1896 syscall_tracepoint_update(p);
1897 write_unlock_irq(&tasklist_lock);
1899 proc_fork_connector(p);
1900 cgroup_post_fork(p);
1901 cgroup_threadgroup_change_end(current);
1904 trace_task_newtask(p, clone_flags);
1905 uprobe_copy_process(p, clone_flags);
1909 bad_fork_cancel_cgroup:
1910 cgroup_cancel_fork(p);
1912 cgroup_threadgroup_change_end(current);
1913 if (pid != &init_struct_pid)
1915 bad_fork_cleanup_thread:
1917 bad_fork_cleanup_io:
1920 bad_fork_cleanup_namespaces:
1921 exit_task_namespaces(p);
1922 bad_fork_cleanup_mm:
1925 bad_fork_cleanup_signal:
1926 if (!(clone_flags & CLONE_THREAD))
1927 free_signal_struct(p->signal);
1928 bad_fork_cleanup_sighand:
1929 __cleanup_sighand(p->sighand);
1930 bad_fork_cleanup_fs:
1931 exit_fs(p); /* blocking */
1932 bad_fork_cleanup_files:
1933 exit_files(p); /* blocking */
1934 bad_fork_cleanup_semundo:
1936 bad_fork_cleanup_security:
1937 security_task_free(p);
1938 bad_fork_cleanup_audit:
1940 bad_fork_cleanup_perf:
1941 perf_event_free_task(p);
1942 bad_fork_cleanup_policy:
1944 mpol_put(p->mempolicy);
1945 bad_fork_cleanup_threadgroup_lock:
1947 delayacct_tsk_free(p);
1948 bad_fork_cleanup_count:
1949 atomic_dec(&p->cred->user->processes);
1952 p->state = TASK_DEAD;
1956 return ERR_PTR(retval);
1959 static inline void init_idle_pids(struct pid_link *links)
1963 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1964 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1965 links[type].pid = &init_struct_pid;
1969 struct task_struct *fork_idle(int cpu)
1971 struct task_struct *task;
1972 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1974 if (!IS_ERR(task)) {
1975 init_idle_pids(task->pids);
1976 init_idle(task, cpu);
1983 * Ok, this is the main fork-routine.
1985 * It copies the process, and if successful kick-starts
1986 * it and waits for it to finish using the VM if required.
1988 long _do_fork(unsigned long clone_flags,
1989 unsigned long stack_start,
1990 unsigned long stack_size,
1991 int __user *parent_tidptr,
1992 int __user *child_tidptr,
1995 struct task_struct *p;
2000 * Determine whether and which event to report to ptracer. When
2001 * called from kernel_thread or CLONE_UNTRACED is explicitly
2002 * requested, no event is reported; otherwise, report if the event
2003 * for the type of forking is enabled.
2005 if (!(clone_flags & CLONE_UNTRACED)) {
2006 if (clone_flags & CLONE_VFORK)
2007 trace = PTRACE_EVENT_VFORK;
2008 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2009 trace = PTRACE_EVENT_CLONE;
2011 trace = PTRACE_EVENT_FORK;
2013 if (likely(!ptrace_event_enabled(current, trace)))
2017 p = copy_process(clone_flags, stack_start, stack_size,
2018 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2019 add_latent_entropy();
2021 * Do this prior waking up the new thread - the thread pointer
2022 * might get invalid after that point, if the thread exits quickly.
2025 struct completion vfork;
2028 trace_sched_process_fork(current, p);
2030 pid = get_task_pid(p, PIDTYPE_PID);
2033 if (clone_flags & CLONE_PARENT_SETTID)
2034 put_user(nr, parent_tidptr);
2036 if (clone_flags & CLONE_VFORK) {
2037 p->vfork_done = &vfork;
2038 init_completion(&vfork);
2042 wake_up_new_task(p);
2044 /* forking complete and child started to run, tell ptracer */
2045 if (unlikely(trace))
2046 ptrace_event_pid(trace, pid);
2048 if (clone_flags & CLONE_VFORK) {
2049 if (!wait_for_vfork_done(p, &vfork))
2050 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2060 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2061 /* For compatibility with architectures that call do_fork directly rather than
2062 * using the syscall entry points below. */
2063 long do_fork(unsigned long clone_flags,
2064 unsigned long stack_start,
2065 unsigned long stack_size,
2066 int __user *parent_tidptr,
2067 int __user *child_tidptr)
2069 return _do_fork(clone_flags, stack_start, stack_size,
2070 parent_tidptr, child_tidptr, 0);
2075 * Create a kernel thread.
2077 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2079 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2080 (unsigned long)arg, NULL, NULL, 0);
2083 #ifdef __ARCH_WANT_SYS_FORK
2084 SYSCALL_DEFINE0(fork)
2087 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2089 /* can not support in nommu mode */
2095 #ifdef __ARCH_WANT_SYS_VFORK
2096 SYSCALL_DEFINE0(vfork)
2098 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2103 #ifdef __ARCH_WANT_SYS_CLONE
2104 #ifdef CONFIG_CLONE_BACKWARDS
2105 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2106 int __user *, parent_tidptr,
2108 int __user *, child_tidptr)
2109 #elif defined(CONFIG_CLONE_BACKWARDS2)
2110 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2111 int __user *, parent_tidptr,
2112 int __user *, child_tidptr,
2114 #elif defined(CONFIG_CLONE_BACKWARDS3)
2115 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2117 int __user *, parent_tidptr,
2118 int __user *, child_tidptr,
2121 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2122 int __user *, parent_tidptr,
2123 int __user *, child_tidptr,
2127 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2131 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2133 struct task_struct *leader, *parent, *child;
2136 read_lock(&tasklist_lock);
2137 leader = top = top->group_leader;
2139 for_each_thread(leader, parent) {
2140 list_for_each_entry(child, &parent->children, sibling) {
2141 res = visitor(child, data);
2153 if (leader != top) {
2155 parent = child->real_parent;
2156 leader = parent->group_leader;
2160 read_unlock(&tasklist_lock);
2163 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2164 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2167 static void sighand_ctor(void *data)
2169 struct sighand_struct *sighand = data;
2171 spin_lock_init(&sighand->siglock);
2172 init_waitqueue_head(&sighand->signalfd_wqh);
2175 void __init proc_caches_init(void)
2177 sighand_cachep = kmem_cache_create("sighand_cache",
2178 sizeof(struct sighand_struct), 0,
2179 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2180 SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2181 signal_cachep = kmem_cache_create("signal_cache",
2182 sizeof(struct signal_struct), 0,
2183 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2185 files_cachep = kmem_cache_create("files_cache",
2186 sizeof(struct files_struct), 0,
2187 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2189 fs_cachep = kmem_cache_create("fs_cache",
2190 sizeof(struct fs_struct), 0,
2191 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2194 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2195 * whole struct cpumask for the OFFSTACK case. We could change
2196 * this to *only* allocate as much of it as required by the
2197 * maximum number of CPU's we can ever have. The cpumask_allocation
2198 * is at the end of the structure, exactly for that reason.
2200 mm_cachep = kmem_cache_create("mm_struct",
2201 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2202 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2204 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2206 nsproxy_cache_init();
2210 * Check constraints on flags passed to the unshare system call.
2212 static int check_unshare_flags(unsigned long unshare_flags)
2214 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2215 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2216 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2217 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2220 * Not implemented, but pretend it works if there is nothing
2221 * to unshare. Note that unsharing the address space or the
2222 * signal handlers also need to unshare the signal queues (aka
2225 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2226 if (!thread_group_empty(current))
2229 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2230 if (atomic_read(¤t->sighand->count) > 1)
2233 if (unshare_flags & CLONE_VM) {
2234 if (!current_is_single_threaded())
2242 * Unshare the filesystem structure if it is being shared
2244 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2246 struct fs_struct *fs = current->fs;
2248 if (!(unshare_flags & CLONE_FS) || !fs)
2251 /* don't need lock here; in the worst case we'll do useless copy */
2255 *new_fsp = copy_fs_struct(fs);
2263 * Unshare file descriptor table if it is being shared
2265 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2267 struct files_struct *fd = current->files;
2270 if ((unshare_flags & CLONE_FILES) &&
2271 (fd && atomic_read(&fd->count) > 1)) {
2272 *new_fdp = dup_fd(fd, &error);
2281 * unshare allows a process to 'unshare' part of the process
2282 * context which was originally shared using clone. copy_*
2283 * functions used by do_fork() cannot be used here directly
2284 * because they modify an inactive task_struct that is being
2285 * constructed. Here we are modifying the current, active,
2288 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2290 struct fs_struct *fs, *new_fs = NULL;
2291 struct files_struct *fd, *new_fd = NULL;
2292 struct cred *new_cred = NULL;
2293 struct nsproxy *new_nsproxy = NULL;
2298 * If unsharing a user namespace must also unshare the thread group
2299 * and unshare the filesystem root and working directories.
2301 if (unshare_flags & CLONE_NEWUSER)
2302 unshare_flags |= CLONE_THREAD | CLONE_FS;
2304 * If unsharing vm, must also unshare signal handlers.
2306 if (unshare_flags & CLONE_VM)
2307 unshare_flags |= CLONE_SIGHAND;
2309 * If unsharing a signal handlers, must also unshare the signal queues.
2311 if (unshare_flags & CLONE_SIGHAND)
2312 unshare_flags |= CLONE_THREAD;
2314 * If unsharing namespace, must also unshare filesystem information.
2316 if (unshare_flags & CLONE_NEWNS)
2317 unshare_flags |= CLONE_FS;
2319 err = check_unshare_flags(unshare_flags);
2321 goto bad_unshare_out;
2323 * CLONE_NEWIPC must also detach from the undolist: after switching
2324 * to a new ipc namespace, the semaphore arrays from the old
2325 * namespace are unreachable.
2327 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2329 err = unshare_fs(unshare_flags, &new_fs);
2331 goto bad_unshare_out;
2332 err = unshare_fd(unshare_flags, &new_fd);
2334 goto bad_unshare_cleanup_fs;
2335 err = unshare_userns(unshare_flags, &new_cred);
2337 goto bad_unshare_cleanup_fd;
2338 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2341 goto bad_unshare_cleanup_cred;
2343 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2346 * CLONE_SYSVSEM is equivalent to sys_exit().
2350 if (unshare_flags & CLONE_NEWIPC) {
2351 /* Orphan segments in old ns (see sem above). */
2353 shm_init_task(current);
2357 switch_task_namespaces(current, new_nsproxy);
2363 spin_lock(&fs->lock);
2364 current->fs = new_fs;
2369 spin_unlock(&fs->lock);
2373 fd = current->files;
2374 current->files = new_fd;
2378 task_unlock(current);
2381 /* Install the new user namespace */
2382 commit_creds(new_cred);
2387 perf_event_namespaces(current);
2389 bad_unshare_cleanup_cred:
2392 bad_unshare_cleanup_fd:
2394 put_files_struct(new_fd);
2396 bad_unshare_cleanup_fs:
2398 free_fs_struct(new_fs);
2405 * Helper to unshare the files of the current task.
2406 * We don't want to expose copy_files internals to
2407 * the exec layer of the kernel.
2410 int unshare_files(struct files_struct **displaced)
2412 struct task_struct *task = current;
2413 struct files_struct *copy = NULL;
2416 error = unshare_fd(CLONE_FILES, ©);
2417 if (error || !copy) {
2421 *displaced = task->files;
2428 int sysctl_max_threads(struct ctl_table *table, int write,
2429 void __user *buffer, size_t *lenp, loff_t *ppos)
2433 int threads = max_threads;
2434 int min = MIN_THREADS;
2435 int max = MAX_THREADS;
2442 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2446 set_max_threads(threads);