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>
40 #include <linux/hmm.h>
43 #include <linux/vmacache.h>
44 #include <linux/nsproxy.h>
45 #include <linux/capability.h>
46 #include <linux/cpu.h>
47 #include <linux/cgroup.h>
48 #include <linux/security.h>
49 #include <linux/hugetlb.h>
50 #include <linux/seccomp.h>
51 #include <linux/swap.h>
52 #include <linux/syscalls.h>
53 #include <linux/jiffies.h>
54 #include <linux/futex.h>
55 #include <linux/compat.h>
56 #include <linux/kthread.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/rcupdate.h>
59 #include <linux/ptrace.h>
60 #include <linux/mount.h>
61 #include <linux/audit.h>
62 #include <linux/memcontrol.h>
63 #include <linux/ftrace.h>
64 #include <linux/proc_fs.h>
65 #include <linux/profile.h>
66 #include <linux/rmap.h>
67 #include <linux/ksm.h>
68 #include <linux/acct.h>
69 #include <linux/userfaultfd_k.h>
70 #include <linux/tsacct_kern.h>
71 #include <linux/cn_proc.h>
72 #include <linux/freezer.h>
73 #include <linux/delayacct.h>
74 #include <linux/taskstats_kern.h>
75 #include <linux/random.h>
76 #include <linux/tty.h>
77 #include <linux/blkdev.h>
78 #include <linux/fs_struct.h>
79 #include <linux/magic.h>
80 #include <linux/sched/mm.h>
81 #include <linux/perf_event.h>
82 #include <linux/posix-timers.h>
83 #include <linux/user-return-notifier.h>
84 #include <linux/oom.h>
85 #include <linux/khugepaged.h>
86 #include <linux/signalfd.h>
87 #include <linux/uprobes.h>
88 #include <linux/aio.h>
89 #include <linux/compiler.h>
90 #include <linux/sysctl.h>
91 #include <linux/kcov.h>
92 #include <linux/livepatch.h>
93 #include <linux/thread_info.h>
95 #include <asm/pgtable.h>
96 #include <asm/pgalloc.h>
97 #include <linux/uaccess.h>
98 #include <asm/mmu_context.h>
99 #include <asm/cacheflush.h>
100 #include <asm/tlbflush.h>
102 #include <trace/events/sched.h>
104 #define CREATE_TRACE_POINTS
105 #include <trace/events/task.h>
108 * Minimum number of threads to boot the kernel
110 #define MIN_THREADS 20
113 * Maximum number of threads
115 #define MAX_THREADS FUTEX_TID_MASK
118 * Protected counters by write_lock_irq(&tasklist_lock)
120 unsigned long total_forks; /* Handle normal Linux uptimes. */
121 int nr_threads; /* The idle threads do not count.. */
123 int max_threads; /* tunable limit on nr_threads */
125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
129 #ifdef CONFIG_PROVE_RCU
130 int lockdep_tasklist_lock_is_held(void)
132 return lockdep_is_held(&tasklist_lock);
134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
135 #endif /* #ifdef CONFIG_PROVE_RCU */
137 int nr_processes(void)
142 for_each_possible_cpu(cpu)
143 total += per_cpu(process_counts, cpu);
148 void __weak arch_release_task_struct(struct task_struct *tsk)
152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
153 static struct kmem_cache *task_struct_cachep;
155 static inline struct task_struct *alloc_task_struct_node(int node)
157 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
160 static inline void free_task_struct(struct task_struct *tsk)
162 kmem_cache_free(task_struct_cachep, tsk);
166 void __weak arch_release_thread_stack(unsigned long *stack)
170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
173 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
174 * kmemcache based allocator.
176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
178 #ifdef CONFIG_VMAP_STACK
180 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
181 * flush. Try to minimize the number of calls by caching stacks.
183 #define NR_CACHED_STACKS 2
184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
186 static int free_vm_stack_cache(unsigned int cpu)
188 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
191 for (i = 0; i < NR_CACHED_STACKS; i++) {
192 struct vm_struct *vm_stack = cached_vm_stacks[i];
197 vfree(vm_stack->addr);
198 cached_vm_stacks[i] = NULL;
205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
207 #ifdef CONFIG_VMAP_STACK
211 for (i = 0; i < NR_CACHED_STACKS; i++) {
214 s = this_cpu_xchg(cached_stacks[i], NULL);
219 /* Clear stale pointers from reused stack. */
220 memset(s->addr, 0, THREAD_SIZE);
222 tsk->stack_vm_area = s;
226 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
227 VMALLOC_START, VMALLOC_END,
230 0, node, __builtin_return_address(0));
233 * We can't call find_vm_area() in interrupt context, and
234 * free_thread_stack() can be called in interrupt context,
235 * so cache the vm_struct.
238 tsk->stack_vm_area = find_vm_area(stack);
241 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
244 return page ? page_address(page) : NULL;
248 static inline void free_thread_stack(struct task_struct *tsk)
250 #ifdef CONFIG_VMAP_STACK
251 if (task_stack_vm_area(tsk)) {
254 for (i = 0; i < NR_CACHED_STACKS; i++) {
255 if (this_cpu_cmpxchg(cached_stacks[i],
256 NULL, tsk->stack_vm_area) != NULL)
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_usercopy("thread_stack",
286 THREAD_SIZE, THREAD_SIZE, 0, 0,
288 BUG_ON(thread_stack_cache == NULL);
293 /* SLAB cache for signal_struct structures (tsk->signal) */
294 static struct kmem_cache *signal_cachep;
296 /* SLAB cache for sighand_struct structures (tsk->sighand) */
297 struct kmem_cache *sighand_cachep;
299 /* SLAB cache for files_struct structures (tsk->files) */
300 struct kmem_cache *files_cachep;
302 /* SLAB cache for fs_struct structures (tsk->fs) */
303 struct kmem_cache *fs_cachep;
305 /* SLAB cache for vm_area_struct structures */
306 static struct kmem_cache *vm_area_cachep;
308 /* SLAB cache for mm_struct structures (tsk->mm) */
309 static struct kmem_cache *mm_cachep;
311 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
313 struct vm_area_struct *vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
320 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
322 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
326 INIT_LIST_HEAD(&new->anon_vma_chain);
331 void vm_area_free(struct vm_area_struct *vma)
333 kmem_cache_free(vm_area_cachep, vma);
336 static void account_kernel_stack(struct task_struct *tsk, int account)
338 void *stack = task_stack_page(tsk);
339 struct vm_struct *vm = task_stack_vm_area(tsk);
341 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
346 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
348 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
349 mod_zone_page_state(page_zone(vm->pages[i]),
351 PAGE_SIZE / 1024 * account);
354 /* All stack pages belong to the same memcg. */
355 mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
356 account * (THREAD_SIZE / 1024));
359 * All stack pages are in the same zone and belong to the
362 struct page *first_page = virt_to_page(stack);
364 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
365 THREAD_SIZE / 1024 * account);
367 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
368 account * (THREAD_SIZE / 1024));
372 static void release_task_stack(struct task_struct *tsk)
374 if (WARN_ON(tsk->state != TASK_DEAD))
375 return; /* Better to leak the stack than to free prematurely */
377 account_kernel_stack(tsk, -1);
378 arch_release_thread_stack(tsk->stack);
379 free_thread_stack(tsk);
381 #ifdef CONFIG_VMAP_STACK
382 tsk->stack_vm_area = NULL;
386 #ifdef CONFIG_THREAD_INFO_IN_TASK
387 void put_task_stack(struct task_struct *tsk)
389 if (atomic_dec_and_test(&tsk->stack_refcount))
390 release_task_stack(tsk);
394 void free_task(struct task_struct *tsk)
396 #ifndef CONFIG_THREAD_INFO_IN_TASK
398 * The task is finally done with both the stack and thread_info,
401 release_task_stack(tsk);
404 * If the task had a separate stack allocation, it should be gone
407 WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
409 rt_mutex_debug_task_free(tsk);
410 ftrace_graph_exit_task(tsk);
411 put_seccomp_filter(tsk);
412 arch_release_task_struct(tsk);
413 if (tsk->flags & PF_KTHREAD)
414 free_kthread_struct(tsk);
415 free_task_struct(tsk);
417 EXPORT_SYMBOL(free_task);
420 static __latent_entropy int dup_mmap(struct mm_struct *mm,
421 struct mm_struct *oldmm)
423 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
424 struct rb_node **rb_link, *rb_parent;
426 unsigned long charge;
429 uprobe_start_dup_mmap();
430 if (down_write_killable(&oldmm->mmap_sem)) {
432 goto fail_uprobe_end;
434 flush_cache_dup_mm(oldmm);
435 uprobe_dup_mmap(oldmm, mm);
437 * Not linked in yet - no deadlock potential:
439 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
441 /* No ordering required: file already has been exposed. */
442 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
444 mm->total_vm = oldmm->total_vm;
445 mm->data_vm = oldmm->data_vm;
446 mm->exec_vm = oldmm->exec_vm;
447 mm->stack_vm = oldmm->stack_vm;
449 rb_link = &mm->mm_rb.rb_node;
452 retval = ksm_fork(mm, oldmm);
455 retval = khugepaged_fork(mm, oldmm);
460 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
463 if (mpnt->vm_flags & VM_DONTCOPY) {
464 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
469 * Don't duplicate many vmas if we've been oom-killed (for
472 if (fatal_signal_pending(current)) {
476 if (mpnt->vm_flags & VM_ACCOUNT) {
477 unsigned long len = vma_pages(mpnt);
479 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
483 tmp = vm_area_dup(mpnt);
486 retval = vma_dup_policy(mpnt, tmp);
488 goto fail_nomem_policy;
490 retval = dup_userfaultfd(tmp, &uf);
492 goto fail_nomem_anon_vma_fork;
493 if (tmp->vm_flags & VM_WIPEONFORK) {
494 /* VM_WIPEONFORK gets a clean slate in the child. */
495 tmp->anon_vma = NULL;
496 if (anon_vma_prepare(tmp))
497 goto fail_nomem_anon_vma_fork;
498 } else if (anon_vma_fork(tmp, mpnt))
499 goto fail_nomem_anon_vma_fork;
500 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
501 tmp->vm_next = tmp->vm_prev = NULL;
504 struct inode *inode = file_inode(file);
505 struct address_space *mapping = file->f_mapping;
508 if (tmp->vm_flags & VM_DENYWRITE)
509 atomic_dec(&inode->i_writecount);
510 i_mmap_lock_write(mapping);
511 if (tmp->vm_flags & VM_SHARED)
512 atomic_inc(&mapping->i_mmap_writable);
513 flush_dcache_mmap_lock(mapping);
514 /* insert tmp into the share list, just after mpnt */
515 vma_interval_tree_insert_after(tmp, mpnt,
517 flush_dcache_mmap_unlock(mapping);
518 i_mmap_unlock_write(mapping);
522 * Clear hugetlb-related page reserves for children. This only
523 * affects MAP_PRIVATE mappings. Faults generated by the child
524 * are not guaranteed to succeed, even if read-only
526 if (is_vm_hugetlb_page(tmp))
527 reset_vma_resv_huge_pages(tmp);
530 * Link in the new vma and copy the page table entries.
533 pprev = &tmp->vm_next;
537 __vma_link_rb(mm, tmp, rb_link, rb_parent);
538 rb_link = &tmp->vm_rb.rb_right;
539 rb_parent = &tmp->vm_rb;
542 if (!(tmp->vm_flags & VM_WIPEONFORK))
543 retval = copy_page_range(mm, oldmm, mpnt);
545 if (tmp->vm_ops && tmp->vm_ops->open)
546 tmp->vm_ops->open(tmp);
551 /* a new mm has just been created */
552 arch_dup_mmap(oldmm, mm);
555 up_write(&mm->mmap_sem);
557 up_write(&oldmm->mmap_sem);
558 dup_userfaultfd_complete(&uf);
560 uprobe_end_dup_mmap();
562 fail_nomem_anon_vma_fork:
563 mpol_put(vma_policy(tmp));
568 vm_unacct_memory(charge);
572 static inline int mm_alloc_pgd(struct mm_struct *mm)
574 mm->pgd = pgd_alloc(mm);
575 if (unlikely(!mm->pgd))
580 static inline void mm_free_pgd(struct mm_struct *mm)
582 pgd_free(mm, mm->pgd);
585 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
587 down_write(&oldmm->mmap_sem);
588 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
589 up_write(&oldmm->mmap_sem);
592 #define mm_alloc_pgd(mm) (0)
593 #define mm_free_pgd(mm)
594 #endif /* CONFIG_MMU */
596 static void check_mm(struct mm_struct *mm)
600 for (i = 0; i < NR_MM_COUNTERS; i++) {
601 long x = atomic_long_read(&mm->rss_stat.count[i]);
604 printk(KERN_ALERT "BUG: Bad rss-counter state "
605 "mm:%p idx:%d val:%ld\n", mm, i, x);
608 if (mm_pgtables_bytes(mm))
609 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
610 mm_pgtables_bytes(mm));
612 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
613 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
617 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
618 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
621 * Called when the last reference to the mm
622 * is dropped: either by a lazy thread or by
623 * mmput. Free the page directory and the mm.
625 void __mmdrop(struct mm_struct *mm)
627 BUG_ON(mm == &init_mm);
628 WARN_ON_ONCE(mm == current->mm);
629 WARN_ON_ONCE(mm == current->active_mm);
633 mmu_notifier_mm_destroy(mm);
635 put_user_ns(mm->user_ns);
638 EXPORT_SYMBOL_GPL(__mmdrop);
640 static void mmdrop_async_fn(struct work_struct *work)
642 struct mm_struct *mm;
644 mm = container_of(work, struct mm_struct, async_put_work);
648 static void mmdrop_async(struct mm_struct *mm)
650 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
651 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
652 schedule_work(&mm->async_put_work);
656 static inline void free_signal_struct(struct signal_struct *sig)
658 taskstats_tgid_free(sig);
659 sched_autogroup_exit(sig);
661 * __mmdrop is not safe to call from softirq context on x86 due to
662 * pgd_dtor so postpone it to the async context
665 mmdrop_async(sig->oom_mm);
666 kmem_cache_free(signal_cachep, sig);
669 static inline void put_signal_struct(struct signal_struct *sig)
671 if (atomic_dec_and_test(&sig->sigcnt))
672 free_signal_struct(sig);
675 void __put_task_struct(struct task_struct *tsk)
677 WARN_ON(!tsk->exit_state);
678 WARN_ON(atomic_read(&tsk->usage));
679 WARN_ON(tsk == current);
683 security_task_free(tsk);
685 delayacct_tsk_free(tsk);
686 put_signal_struct(tsk->signal);
688 if (!profile_handoff_task(tsk))
691 EXPORT_SYMBOL_GPL(__put_task_struct);
693 void __init __weak arch_task_cache_init(void) { }
698 static void set_max_threads(unsigned int max_threads_suggested)
703 * The number of threads shall be limited such that the thread
704 * structures may only consume a small part of the available memory.
706 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
707 threads = MAX_THREADS;
709 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
710 (u64) THREAD_SIZE * 8UL);
712 if (threads > max_threads_suggested)
713 threads = max_threads_suggested;
715 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
718 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
719 /* Initialized by the architecture: */
720 int arch_task_struct_size __read_mostly;
723 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
725 /* Fetch thread_struct whitelist for the architecture. */
726 arch_thread_struct_whitelist(offset, size);
729 * Handle zero-sized whitelist or empty thread_struct, otherwise
730 * adjust offset to position of thread_struct in task_struct.
732 if (unlikely(*size == 0))
735 *offset += offsetof(struct task_struct, thread);
738 void __init fork_init(void)
741 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
742 #ifndef ARCH_MIN_TASKALIGN
743 #define ARCH_MIN_TASKALIGN 0
745 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
746 unsigned long useroffset, usersize;
748 /* create a slab on which task_structs can be allocated */
749 task_struct_whitelist(&useroffset, &usersize);
750 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
751 arch_task_struct_size, align,
752 SLAB_PANIC|SLAB_ACCOUNT,
753 useroffset, usersize, NULL);
756 /* do the arch specific task caches init */
757 arch_task_cache_init();
759 set_max_threads(MAX_THREADS);
761 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
762 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
763 init_task.signal->rlim[RLIMIT_SIGPENDING] =
764 init_task.signal->rlim[RLIMIT_NPROC];
766 for (i = 0; i < UCOUNT_COUNTS; i++) {
767 init_user_ns.ucount_max[i] = max_threads/2;
770 #ifdef CONFIG_VMAP_STACK
771 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
772 NULL, free_vm_stack_cache);
775 lockdep_init_task(&init_task);
778 int __weak arch_dup_task_struct(struct task_struct *dst,
779 struct task_struct *src)
785 void set_task_stack_end_magic(struct task_struct *tsk)
787 unsigned long *stackend;
789 stackend = end_of_stack(tsk);
790 *stackend = STACK_END_MAGIC; /* for overflow detection */
793 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
795 struct task_struct *tsk;
796 unsigned long *stack;
797 struct vm_struct *stack_vm_area;
800 if (node == NUMA_NO_NODE)
801 node = tsk_fork_get_node(orig);
802 tsk = alloc_task_struct_node(node);
806 stack = alloc_thread_stack_node(tsk, node);
810 stack_vm_area = task_stack_vm_area(tsk);
812 err = arch_dup_task_struct(tsk, orig);
815 * arch_dup_task_struct() clobbers the stack-related fields. Make
816 * sure they're properly initialized before using any stack-related
820 #ifdef CONFIG_VMAP_STACK
821 tsk->stack_vm_area = stack_vm_area;
823 #ifdef CONFIG_THREAD_INFO_IN_TASK
824 atomic_set(&tsk->stack_refcount, 1);
830 #ifdef CONFIG_SECCOMP
832 * We must handle setting up seccomp filters once we're under
833 * the sighand lock in case orig has changed between now and
834 * then. Until then, filter must be NULL to avoid messing up
835 * the usage counts on the error path calling free_task.
837 tsk->seccomp.filter = NULL;
840 setup_thread_stack(tsk, orig);
841 clear_user_return_notifier(tsk);
842 clear_tsk_need_resched(tsk);
843 set_task_stack_end_magic(tsk);
845 #ifdef CONFIG_STACKPROTECTOR
846 tsk->stack_canary = get_random_canary();
850 * One for us, one for whoever does the "release_task()" (usually
853 atomic_set(&tsk->usage, 2);
854 #ifdef CONFIG_BLK_DEV_IO_TRACE
857 tsk->splice_pipe = NULL;
858 tsk->task_frag.page = NULL;
859 tsk->wake_q.next = NULL;
861 account_kernel_stack(tsk, 1);
865 #ifdef CONFIG_FAULT_INJECTION
869 #ifdef CONFIG_BLK_CGROUP
870 tsk->throttle_queue = NULL;
871 tsk->use_memdelay = 0;
875 tsk->active_memcg = NULL;
880 free_thread_stack(tsk);
882 free_task_struct(tsk);
886 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
888 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
890 static int __init coredump_filter_setup(char *s)
892 default_dump_filter =
893 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
894 MMF_DUMP_FILTER_MASK;
898 __setup("coredump_filter=", coredump_filter_setup);
900 #include <linux/init_task.h>
902 static void mm_init_aio(struct mm_struct *mm)
905 spin_lock_init(&mm->ioctx_lock);
906 mm->ioctx_table = NULL;
910 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
917 static void mm_init_uprobes_state(struct mm_struct *mm)
919 #ifdef CONFIG_UPROBES
920 mm->uprobes_state.xol_area = NULL;
924 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
925 struct user_namespace *user_ns)
929 mm->vmacache_seqnum = 0;
930 atomic_set(&mm->mm_users, 1);
931 atomic_set(&mm->mm_count, 1);
932 init_rwsem(&mm->mmap_sem);
933 INIT_LIST_HEAD(&mm->mmlist);
934 mm->core_state = NULL;
935 mm_pgtables_bytes_init(mm);
939 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
940 spin_lock_init(&mm->page_table_lock);
941 spin_lock_init(&mm->arg_lock);
944 mm_init_owner(mm, p);
945 RCU_INIT_POINTER(mm->exe_file, NULL);
946 mmu_notifier_mm_init(mm);
948 init_tlb_flush_pending(mm);
949 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
950 mm->pmd_huge_pte = NULL;
952 mm_init_uprobes_state(mm);
955 mm->flags = current->mm->flags & MMF_INIT_MASK;
956 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
958 mm->flags = default_dump_filter;
962 if (mm_alloc_pgd(mm))
965 if (init_new_context(p, mm))
968 mm->user_ns = get_user_ns(user_ns);
979 * Allocate and initialize an mm_struct.
981 struct mm_struct *mm_alloc(void)
983 struct mm_struct *mm;
989 memset(mm, 0, sizeof(*mm));
990 return mm_init(mm, current, current_user_ns());
993 static inline void __mmput(struct mm_struct *mm)
995 VM_BUG_ON(atomic_read(&mm->mm_users));
997 uprobe_clear_state(mm);
1000 khugepaged_exit(mm); /* must run before exit_mmap */
1002 mm_put_huge_zero_page(mm);
1003 set_mm_exe_file(mm, NULL);
1004 if (!list_empty(&mm->mmlist)) {
1005 spin_lock(&mmlist_lock);
1006 list_del(&mm->mmlist);
1007 spin_unlock(&mmlist_lock);
1010 module_put(mm->binfmt->module);
1015 * Decrement the use count and release all resources for an mm.
1017 void mmput(struct mm_struct *mm)
1021 if (atomic_dec_and_test(&mm->mm_users))
1024 EXPORT_SYMBOL_GPL(mmput);
1027 static void mmput_async_fn(struct work_struct *work)
1029 struct mm_struct *mm = container_of(work, struct mm_struct,
1035 void mmput_async(struct mm_struct *mm)
1037 if (atomic_dec_and_test(&mm->mm_users)) {
1038 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1039 schedule_work(&mm->async_put_work);
1045 * set_mm_exe_file - change a reference to the mm's executable file
1047 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1049 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1050 * invocations: in mmput() nobody alive left, in execve task is single
1051 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1052 * mm->exe_file, but does so without using set_mm_exe_file() in order
1053 * to do avoid the need for any locks.
1055 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1057 struct file *old_exe_file;
1060 * It is safe to dereference the exe_file without RCU as
1061 * this function is only called if nobody else can access
1062 * this mm -- see comment above for justification.
1064 old_exe_file = rcu_dereference_raw(mm->exe_file);
1067 get_file(new_exe_file);
1068 rcu_assign_pointer(mm->exe_file, new_exe_file);
1074 * get_mm_exe_file - acquire a reference to the mm's executable file
1076 * Returns %NULL if mm has no associated executable file.
1077 * User must release file via fput().
1079 struct file *get_mm_exe_file(struct mm_struct *mm)
1081 struct file *exe_file;
1084 exe_file = rcu_dereference(mm->exe_file);
1085 if (exe_file && !get_file_rcu(exe_file))
1090 EXPORT_SYMBOL(get_mm_exe_file);
1093 * get_task_exe_file - acquire a reference to the task's executable file
1095 * Returns %NULL if task's mm (if any) has no associated executable file or
1096 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1097 * User must release file via fput().
1099 struct file *get_task_exe_file(struct task_struct *task)
1101 struct file *exe_file = NULL;
1102 struct mm_struct *mm;
1107 if (!(task->flags & PF_KTHREAD))
1108 exe_file = get_mm_exe_file(mm);
1113 EXPORT_SYMBOL(get_task_exe_file);
1116 * get_task_mm - acquire a reference to the task's mm
1118 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1119 * this kernel workthread has transiently adopted a user mm with use_mm,
1120 * to do its AIO) is not set and if so returns a reference to it, after
1121 * bumping up the use count. User must release the mm via mmput()
1122 * after use. Typically used by /proc and ptrace.
1124 struct mm_struct *get_task_mm(struct task_struct *task)
1126 struct mm_struct *mm;
1131 if (task->flags & PF_KTHREAD)
1139 EXPORT_SYMBOL_GPL(get_task_mm);
1141 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1143 struct mm_struct *mm;
1146 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1148 return ERR_PTR(err);
1150 mm = get_task_mm(task);
1151 if (mm && mm != current->mm &&
1152 !ptrace_may_access(task, mode)) {
1154 mm = ERR_PTR(-EACCES);
1156 mutex_unlock(&task->signal->cred_guard_mutex);
1161 static void complete_vfork_done(struct task_struct *tsk)
1163 struct completion *vfork;
1166 vfork = tsk->vfork_done;
1167 if (likely(vfork)) {
1168 tsk->vfork_done = NULL;
1174 static int wait_for_vfork_done(struct task_struct *child,
1175 struct completion *vfork)
1179 freezer_do_not_count();
1180 killed = wait_for_completion_killable(vfork);
1185 child->vfork_done = NULL;
1189 put_task_struct(child);
1193 /* Please note the differences between mmput and mm_release.
1194 * mmput is called whenever we stop holding onto a mm_struct,
1195 * error success whatever.
1197 * mm_release is called after a mm_struct has been removed
1198 * from the current process.
1200 * This difference is important for error handling, when we
1201 * only half set up a mm_struct for a new process and need to restore
1202 * the old one. Because we mmput the new mm_struct before
1203 * restoring the old one. . .
1204 * Eric Biederman 10 January 1998
1206 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1208 /* Get rid of any futexes when releasing the mm */
1210 if (unlikely(tsk->robust_list)) {
1211 exit_robust_list(tsk);
1212 tsk->robust_list = NULL;
1214 #ifdef CONFIG_COMPAT
1215 if (unlikely(tsk->compat_robust_list)) {
1216 compat_exit_robust_list(tsk);
1217 tsk->compat_robust_list = NULL;
1220 if (unlikely(!list_empty(&tsk->pi_state_list)))
1221 exit_pi_state_list(tsk);
1224 uprobe_free_utask(tsk);
1226 /* Get rid of any cached register state */
1227 deactivate_mm(tsk, mm);
1230 * Signal userspace if we're not exiting with a core dump
1231 * because we want to leave the value intact for debugging
1234 if (tsk->clear_child_tid) {
1235 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1236 atomic_read(&mm->mm_users) > 1) {
1238 * We don't check the error code - if userspace has
1239 * not set up a proper pointer then tough luck.
1241 put_user(0, tsk->clear_child_tid);
1242 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1243 1, NULL, NULL, 0, 0);
1245 tsk->clear_child_tid = NULL;
1249 * All done, finally we can wake up parent and return this mm to him.
1250 * Also kthread_stop() uses this completion for synchronization.
1252 if (tsk->vfork_done)
1253 complete_vfork_done(tsk);
1257 * Allocate a new mm structure and copy contents from the
1258 * mm structure of the passed in task structure.
1260 static struct mm_struct *dup_mm(struct task_struct *tsk)
1262 struct mm_struct *mm, *oldmm = current->mm;
1269 memcpy(mm, oldmm, sizeof(*mm));
1271 if (!mm_init(mm, tsk, mm->user_ns))
1274 err = dup_mmap(mm, oldmm);
1278 mm->hiwater_rss = get_mm_rss(mm);
1279 mm->hiwater_vm = mm->total_vm;
1281 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1287 /* don't put binfmt in mmput, we haven't got module yet */
1295 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1297 struct mm_struct *mm, *oldmm;
1300 tsk->min_flt = tsk->maj_flt = 0;
1301 tsk->nvcsw = tsk->nivcsw = 0;
1302 #ifdef CONFIG_DETECT_HUNG_TASK
1303 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1307 tsk->active_mm = NULL;
1310 * Are we cloning a kernel thread?
1312 * We need to steal a active VM for that..
1314 oldmm = current->mm;
1318 /* initialize the new vmacache entries */
1319 vmacache_flush(tsk);
1321 if (clone_flags & CLONE_VM) {
1334 tsk->active_mm = mm;
1341 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1343 struct fs_struct *fs = current->fs;
1344 if (clone_flags & CLONE_FS) {
1345 /* tsk->fs is already what we want */
1346 spin_lock(&fs->lock);
1348 spin_unlock(&fs->lock);
1352 spin_unlock(&fs->lock);
1355 tsk->fs = copy_fs_struct(fs);
1361 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1363 struct files_struct *oldf, *newf;
1367 * A background process may not have any files ...
1369 oldf = current->files;
1373 if (clone_flags & CLONE_FILES) {
1374 atomic_inc(&oldf->count);
1378 newf = dup_fd(oldf, &error);
1388 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1391 struct io_context *ioc = current->io_context;
1392 struct io_context *new_ioc;
1397 * Share io context with parent, if CLONE_IO is set
1399 if (clone_flags & CLONE_IO) {
1401 tsk->io_context = ioc;
1402 } else if (ioprio_valid(ioc->ioprio)) {
1403 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1404 if (unlikely(!new_ioc))
1407 new_ioc->ioprio = ioc->ioprio;
1408 put_io_context(new_ioc);
1414 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1416 struct sighand_struct *sig;
1418 if (clone_flags & CLONE_SIGHAND) {
1419 atomic_inc(¤t->sighand->count);
1422 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1423 rcu_assign_pointer(tsk->sighand, sig);
1427 atomic_set(&sig->count, 1);
1428 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1432 void __cleanup_sighand(struct sighand_struct *sighand)
1434 if (atomic_dec_and_test(&sighand->count)) {
1435 signalfd_cleanup(sighand);
1437 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1438 * without an RCU grace period, see __lock_task_sighand().
1440 kmem_cache_free(sighand_cachep, sighand);
1444 #ifdef CONFIG_POSIX_TIMERS
1446 * Initialize POSIX timer handling for a thread group.
1448 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1450 unsigned long cpu_limit;
1452 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1453 if (cpu_limit != RLIM_INFINITY) {
1454 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1455 sig->cputimer.running = true;
1458 /* The timer lists. */
1459 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1460 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1461 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1464 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1467 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1469 struct signal_struct *sig;
1471 if (clone_flags & CLONE_THREAD)
1474 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1479 sig->nr_threads = 1;
1480 atomic_set(&sig->live, 1);
1481 atomic_set(&sig->sigcnt, 1);
1483 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1484 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1485 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1487 init_waitqueue_head(&sig->wait_chldexit);
1488 sig->curr_target = tsk;
1489 init_sigpending(&sig->shared_pending);
1490 seqlock_init(&sig->stats_lock);
1491 prev_cputime_init(&sig->prev_cputime);
1493 #ifdef CONFIG_POSIX_TIMERS
1494 INIT_LIST_HEAD(&sig->posix_timers);
1495 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1496 sig->real_timer.function = it_real_fn;
1499 task_lock(current->group_leader);
1500 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1501 task_unlock(current->group_leader);
1503 posix_cpu_timers_init_group(sig);
1505 tty_audit_fork(sig);
1506 sched_autogroup_fork(sig);
1508 sig->oom_score_adj = current->signal->oom_score_adj;
1509 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1511 mutex_init(&sig->cred_guard_mutex);
1516 static void copy_seccomp(struct task_struct *p)
1518 #ifdef CONFIG_SECCOMP
1520 * Must be called with sighand->lock held, which is common to
1521 * all threads in the group. Holding cred_guard_mutex is not
1522 * needed because this new task is not yet running and cannot
1525 assert_spin_locked(¤t->sighand->siglock);
1527 /* Ref-count the new filter user, and assign it. */
1528 get_seccomp_filter(current);
1529 p->seccomp = current->seccomp;
1532 * Explicitly enable no_new_privs here in case it got set
1533 * between the task_struct being duplicated and holding the
1534 * sighand lock. The seccomp state and nnp must be in sync.
1536 if (task_no_new_privs(current))
1537 task_set_no_new_privs(p);
1540 * If the parent gained a seccomp mode after copying thread
1541 * flags and between before we held the sighand lock, we have
1542 * to manually enable the seccomp thread flag here.
1544 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1545 set_tsk_thread_flag(p, TIF_SECCOMP);
1549 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1551 current->clear_child_tid = tidptr;
1553 return task_pid_vnr(current);
1556 static void rt_mutex_init_task(struct task_struct *p)
1558 raw_spin_lock_init(&p->pi_lock);
1559 #ifdef CONFIG_RT_MUTEXES
1560 p->pi_waiters = RB_ROOT_CACHED;
1561 p->pi_top_task = NULL;
1562 p->pi_blocked_on = NULL;
1566 #ifdef CONFIG_POSIX_TIMERS
1568 * Initialize POSIX timer handling for a single task.
1570 static void posix_cpu_timers_init(struct task_struct *tsk)
1572 tsk->cputime_expires.prof_exp = 0;
1573 tsk->cputime_expires.virt_exp = 0;
1574 tsk->cputime_expires.sched_exp = 0;
1575 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1576 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1577 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1580 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1584 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1586 task->pids[type].pid = pid;
1589 static inline void rcu_copy_process(struct task_struct *p)
1591 #ifdef CONFIG_PREEMPT_RCU
1592 p->rcu_read_lock_nesting = 0;
1593 p->rcu_read_unlock_special.s = 0;
1594 p->rcu_blocked_node = NULL;
1595 INIT_LIST_HEAD(&p->rcu_node_entry);
1596 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1597 #ifdef CONFIG_TASKS_RCU
1598 p->rcu_tasks_holdout = false;
1599 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1600 p->rcu_tasks_idle_cpu = -1;
1601 #endif /* #ifdef CONFIG_TASKS_RCU */
1605 * This creates a new process as a copy of the old one,
1606 * but does not actually start it yet.
1608 * It copies the registers, and all the appropriate
1609 * parts of the process environment (as per the clone
1610 * flags). The actual kick-off is left to the caller.
1612 static __latent_entropy struct task_struct *copy_process(
1613 unsigned long clone_flags,
1614 unsigned long stack_start,
1615 unsigned long stack_size,
1616 int __user *child_tidptr,
1623 struct task_struct *p;
1626 * Don't allow sharing the root directory with processes in a different
1629 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1630 return ERR_PTR(-EINVAL);
1632 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1633 return ERR_PTR(-EINVAL);
1636 * Thread groups must share signals as well, and detached threads
1637 * can only be started up within the thread group.
1639 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1640 return ERR_PTR(-EINVAL);
1643 * Shared signal handlers imply shared VM. By way of the above,
1644 * thread groups also imply shared VM. Blocking this case allows
1645 * for various simplifications in other code.
1647 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1648 return ERR_PTR(-EINVAL);
1651 * Siblings of global init remain as zombies on exit since they are
1652 * not reaped by their parent (swapper). To solve this and to avoid
1653 * multi-rooted process trees, prevent global and container-inits
1654 * from creating siblings.
1656 if ((clone_flags & CLONE_PARENT) &&
1657 current->signal->flags & SIGNAL_UNKILLABLE)
1658 return ERR_PTR(-EINVAL);
1661 * If the new process will be in a different pid or user namespace
1662 * do not allow it to share a thread group with the forking task.
1664 if (clone_flags & CLONE_THREAD) {
1665 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1666 (task_active_pid_ns(current) !=
1667 current->nsproxy->pid_ns_for_children))
1668 return ERR_PTR(-EINVAL);
1672 p = dup_task_struct(current, node);
1677 * This _must_ happen before we call free_task(), i.e. before we jump
1678 * to any of the bad_fork_* labels. This is to avoid freeing
1679 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1680 * kernel threads (PF_KTHREAD).
1682 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1684 * Clear TID on mm_release()?
1686 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1688 ftrace_graph_init_task(p);
1690 rt_mutex_init_task(p);
1692 #ifdef CONFIG_PROVE_LOCKING
1693 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1694 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1697 if (atomic_read(&p->real_cred->user->processes) >=
1698 task_rlimit(p, RLIMIT_NPROC)) {
1699 if (p->real_cred->user != INIT_USER &&
1700 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1703 current->flags &= ~PF_NPROC_EXCEEDED;
1705 retval = copy_creds(p, clone_flags);
1710 * If multiple threads are within copy_process(), then this check
1711 * triggers too late. This doesn't hurt, the check is only there
1712 * to stop root fork bombs.
1715 if (nr_threads >= max_threads)
1716 goto bad_fork_cleanup_count;
1718 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1719 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1720 p->flags |= PF_FORKNOEXEC;
1721 INIT_LIST_HEAD(&p->children);
1722 INIT_LIST_HEAD(&p->sibling);
1723 rcu_copy_process(p);
1724 p->vfork_done = NULL;
1725 spin_lock_init(&p->alloc_lock);
1727 init_sigpending(&p->pending);
1729 p->utime = p->stime = p->gtime = 0;
1730 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1731 p->utimescaled = p->stimescaled = 0;
1733 prev_cputime_init(&p->prev_cputime);
1735 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1736 seqcount_init(&p->vtime.seqcount);
1737 p->vtime.starttime = 0;
1738 p->vtime.state = VTIME_INACTIVE;
1741 #if defined(SPLIT_RSS_COUNTING)
1742 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1745 p->default_timer_slack_ns = current->timer_slack_ns;
1747 task_io_accounting_init(&p->ioac);
1748 acct_clear_integrals(p);
1750 posix_cpu_timers_init(p);
1752 p->start_time = ktime_get_ns();
1753 p->real_start_time = ktime_get_boot_ns();
1754 p->io_context = NULL;
1755 audit_set_context(p, NULL);
1758 p->mempolicy = mpol_dup(p->mempolicy);
1759 if (IS_ERR(p->mempolicy)) {
1760 retval = PTR_ERR(p->mempolicy);
1761 p->mempolicy = NULL;
1762 goto bad_fork_cleanup_threadgroup_lock;
1765 #ifdef CONFIG_CPUSETS
1766 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1767 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1768 seqcount_init(&p->mems_allowed_seq);
1770 #ifdef CONFIG_TRACE_IRQFLAGS
1772 p->hardirqs_enabled = 0;
1773 p->hardirq_enable_ip = 0;
1774 p->hardirq_enable_event = 0;
1775 p->hardirq_disable_ip = _THIS_IP_;
1776 p->hardirq_disable_event = 0;
1777 p->softirqs_enabled = 1;
1778 p->softirq_enable_ip = _THIS_IP_;
1779 p->softirq_enable_event = 0;
1780 p->softirq_disable_ip = 0;
1781 p->softirq_disable_event = 0;
1782 p->hardirq_context = 0;
1783 p->softirq_context = 0;
1786 p->pagefault_disabled = 0;
1788 #ifdef CONFIG_LOCKDEP
1789 p->lockdep_depth = 0; /* no locks held yet */
1790 p->curr_chain_key = 0;
1791 p->lockdep_recursion = 0;
1792 lockdep_init_task(p);
1795 #ifdef CONFIG_DEBUG_MUTEXES
1796 p->blocked_on = NULL; /* not blocked yet */
1798 #ifdef CONFIG_BCACHE
1799 p->sequential_io = 0;
1800 p->sequential_io_avg = 0;
1803 /* Perform scheduler related setup. Assign this task to a CPU. */
1804 retval = sched_fork(clone_flags, p);
1806 goto bad_fork_cleanup_policy;
1808 retval = perf_event_init_task(p);
1810 goto bad_fork_cleanup_policy;
1811 retval = audit_alloc(p);
1813 goto bad_fork_cleanup_perf;
1814 /* copy all the process information */
1816 retval = security_task_alloc(p, clone_flags);
1818 goto bad_fork_cleanup_audit;
1819 retval = copy_semundo(clone_flags, p);
1821 goto bad_fork_cleanup_security;
1822 retval = copy_files(clone_flags, p);
1824 goto bad_fork_cleanup_semundo;
1825 retval = copy_fs(clone_flags, p);
1827 goto bad_fork_cleanup_files;
1828 retval = copy_sighand(clone_flags, p);
1830 goto bad_fork_cleanup_fs;
1831 retval = copy_signal(clone_flags, p);
1833 goto bad_fork_cleanup_sighand;
1834 retval = copy_mm(clone_flags, p);
1836 goto bad_fork_cleanup_signal;
1837 retval = copy_namespaces(clone_flags, p);
1839 goto bad_fork_cleanup_mm;
1840 retval = copy_io(clone_flags, p);
1842 goto bad_fork_cleanup_namespaces;
1843 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1845 goto bad_fork_cleanup_io;
1847 if (pid != &init_struct_pid) {
1848 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1850 retval = PTR_ERR(pid);
1851 goto bad_fork_cleanup_thread;
1859 p->robust_list = NULL;
1860 #ifdef CONFIG_COMPAT
1861 p->compat_robust_list = NULL;
1863 INIT_LIST_HEAD(&p->pi_state_list);
1864 p->pi_state_cache = NULL;
1867 * sigaltstack should be cleared when sharing the same VM
1869 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1873 * Syscall tracing and stepping should be turned off in the
1874 * child regardless of CLONE_PTRACE.
1876 user_disable_single_step(p);
1877 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1878 #ifdef TIF_SYSCALL_EMU
1879 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1881 clear_all_latency_tracing(p);
1883 /* ok, now we should be set up.. */
1884 p->pid = pid_nr(pid);
1885 if (clone_flags & CLONE_THREAD) {
1886 p->exit_signal = -1;
1887 p->group_leader = current->group_leader;
1888 p->tgid = current->tgid;
1890 if (clone_flags & CLONE_PARENT)
1891 p->exit_signal = current->group_leader->exit_signal;
1893 p->exit_signal = (clone_flags & CSIGNAL);
1894 p->group_leader = p;
1899 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1900 p->dirty_paused_when = 0;
1902 p->pdeath_signal = 0;
1903 INIT_LIST_HEAD(&p->thread_group);
1904 p->task_works = NULL;
1906 cgroup_threadgroup_change_begin(current);
1908 * Ensure that the cgroup subsystem policies allow the new process to be
1909 * forked. It should be noted the the new process's css_set can be changed
1910 * between here and cgroup_post_fork() if an organisation operation is in
1913 retval = cgroup_can_fork(p);
1915 goto bad_fork_free_pid;
1918 * Make it visible to the rest of the system, but dont wake it up yet.
1919 * Need tasklist lock for parent etc handling!
1921 write_lock_irq(&tasklist_lock);
1923 /* CLONE_PARENT re-uses the old parent */
1924 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1925 p->real_parent = current->real_parent;
1926 p->parent_exec_id = current->parent_exec_id;
1928 p->real_parent = current;
1929 p->parent_exec_id = current->self_exec_id;
1932 klp_copy_process(p);
1934 spin_lock(¤t->sighand->siglock);
1937 * Copy seccomp details explicitly here, in case they were changed
1938 * before holding sighand lock.
1942 rseq_fork(p, clone_flags);
1945 * Process group and session signals need to be delivered to just the
1946 * parent before the fork or both the parent and the child after the
1947 * fork. Restart if a signal comes in before we add the new process to
1948 * it's process group.
1949 * A fatal signal pending means that current will exit, so the new
1950 * thread can't slip out of an OOM kill (or normal SIGKILL).
1952 recalc_sigpending();
1953 if (signal_pending(current)) {
1954 retval = -ERESTARTNOINTR;
1955 goto bad_fork_cancel_cgroup;
1957 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1959 goto bad_fork_cancel_cgroup;
1962 if (likely(p->pid)) {
1963 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1965 init_task_pid(p, PIDTYPE_PID, pid);
1966 if (thread_group_leader(p)) {
1967 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1968 init_task_pid(p, PIDTYPE_SID, task_session(current));
1970 if (is_child_reaper(pid)) {
1971 ns_of_pid(pid)->child_reaper = p;
1972 p->signal->flags |= SIGNAL_UNKILLABLE;
1975 p->signal->leader_pid = pid;
1976 p->signal->tty = tty_kref_get(current->signal->tty);
1978 * Inherit has_child_subreaper flag under the same
1979 * tasklist_lock with adding child to the process tree
1980 * for propagate_has_child_subreaper optimization.
1982 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1983 p->real_parent->signal->is_child_subreaper;
1984 list_add_tail(&p->sibling, &p->real_parent->children);
1985 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1986 attach_pid(p, PIDTYPE_PGID);
1987 attach_pid(p, PIDTYPE_SID);
1988 __this_cpu_inc(process_counts);
1990 current->signal->nr_threads++;
1991 atomic_inc(¤t->signal->live);
1992 atomic_inc(¤t->signal->sigcnt);
1993 list_add_tail_rcu(&p->thread_group,
1994 &p->group_leader->thread_group);
1995 list_add_tail_rcu(&p->thread_node,
1996 &p->signal->thread_head);
1998 attach_pid(p, PIDTYPE_PID);
2003 spin_unlock(¤t->sighand->siglock);
2004 syscall_tracepoint_update(p);
2005 write_unlock_irq(&tasklist_lock);
2007 proc_fork_connector(p);
2008 cgroup_post_fork(p);
2009 cgroup_threadgroup_change_end(current);
2012 trace_task_newtask(p, clone_flags);
2013 uprobe_copy_process(p, clone_flags);
2017 bad_fork_cancel_cgroup:
2018 spin_unlock(¤t->sighand->siglock);
2019 write_unlock_irq(&tasklist_lock);
2020 cgroup_cancel_fork(p);
2022 cgroup_threadgroup_change_end(current);
2023 if (pid != &init_struct_pid)
2025 bad_fork_cleanup_thread:
2027 bad_fork_cleanup_io:
2030 bad_fork_cleanup_namespaces:
2031 exit_task_namespaces(p);
2032 bad_fork_cleanup_mm:
2035 bad_fork_cleanup_signal:
2036 if (!(clone_flags & CLONE_THREAD))
2037 free_signal_struct(p->signal);
2038 bad_fork_cleanup_sighand:
2039 __cleanup_sighand(p->sighand);
2040 bad_fork_cleanup_fs:
2041 exit_fs(p); /* blocking */
2042 bad_fork_cleanup_files:
2043 exit_files(p); /* blocking */
2044 bad_fork_cleanup_semundo:
2046 bad_fork_cleanup_security:
2047 security_task_free(p);
2048 bad_fork_cleanup_audit:
2050 bad_fork_cleanup_perf:
2051 perf_event_free_task(p);
2052 bad_fork_cleanup_policy:
2053 lockdep_free_task(p);
2055 mpol_put(p->mempolicy);
2056 bad_fork_cleanup_threadgroup_lock:
2058 delayacct_tsk_free(p);
2059 bad_fork_cleanup_count:
2060 atomic_dec(&p->cred->user->processes);
2063 p->state = TASK_DEAD;
2067 return ERR_PTR(retval);
2070 static inline void init_idle_pids(struct pid_link *links)
2074 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2075 INIT_HLIST_NODE(&links[type].node); /* not really needed */
2076 links[type].pid = &init_struct_pid;
2080 struct task_struct *fork_idle(int cpu)
2082 struct task_struct *task;
2083 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2085 if (!IS_ERR(task)) {
2086 init_idle_pids(task->pids);
2087 init_idle(task, cpu);
2094 * Ok, this is the main fork-routine.
2096 * It copies the process, and if successful kick-starts
2097 * it and waits for it to finish using the VM if required.
2099 long _do_fork(unsigned long clone_flags,
2100 unsigned long stack_start,
2101 unsigned long stack_size,
2102 int __user *parent_tidptr,
2103 int __user *child_tidptr,
2106 struct completion vfork;
2108 struct task_struct *p;
2113 * Determine whether and which event to report to ptracer. When
2114 * called from kernel_thread or CLONE_UNTRACED is explicitly
2115 * requested, no event is reported; otherwise, report if the event
2116 * for the type of forking is enabled.
2118 if (!(clone_flags & CLONE_UNTRACED)) {
2119 if (clone_flags & CLONE_VFORK)
2120 trace = PTRACE_EVENT_VFORK;
2121 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2122 trace = PTRACE_EVENT_CLONE;
2124 trace = PTRACE_EVENT_FORK;
2126 if (likely(!ptrace_event_enabled(current, trace)))
2130 p = copy_process(clone_flags, stack_start, stack_size,
2131 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2132 add_latent_entropy();
2138 * Do this prior waking up the new thread - the thread pointer
2139 * might get invalid after that point, if the thread exits quickly.
2141 trace_sched_process_fork(current, p);
2143 pid = get_task_pid(p, PIDTYPE_PID);
2146 if (clone_flags & CLONE_PARENT_SETTID)
2147 put_user(nr, parent_tidptr);
2149 if (clone_flags & CLONE_VFORK) {
2150 p->vfork_done = &vfork;
2151 init_completion(&vfork);
2155 wake_up_new_task(p);
2157 /* forking complete and child started to run, tell ptracer */
2158 if (unlikely(trace))
2159 ptrace_event_pid(trace, pid);
2161 if (clone_flags & CLONE_VFORK) {
2162 if (!wait_for_vfork_done(p, &vfork))
2163 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2170 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2171 /* For compatibility with architectures that call do_fork directly rather than
2172 * using the syscall entry points below. */
2173 long do_fork(unsigned long clone_flags,
2174 unsigned long stack_start,
2175 unsigned long stack_size,
2176 int __user *parent_tidptr,
2177 int __user *child_tidptr)
2179 return _do_fork(clone_flags, stack_start, stack_size,
2180 parent_tidptr, child_tidptr, 0);
2185 * Create a kernel thread.
2187 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2189 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2190 (unsigned long)arg, NULL, NULL, 0);
2193 #ifdef __ARCH_WANT_SYS_FORK
2194 SYSCALL_DEFINE0(fork)
2197 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2199 /* can not support in nommu mode */
2205 #ifdef __ARCH_WANT_SYS_VFORK
2206 SYSCALL_DEFINE0(vfork)
2208 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2213 #ifdef __ARCH_WANT_SYS_CLONE
2214 #ifdef CONFIG_CLONE_BACKWARDS
2215 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2216 int __user *, parent_tidptr,
2218 int __user *, child_tidptr)
2219 #elif defined(CONFIG_CLONE_BACKWARDS2)
2220 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2221 int __user *, parent_tidptr,
2222 int __user *, child_tidptr,
2224 #elif defined(CONFIG_CLONE_BACKWARDS3)
2225 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2227 int __user *, parent_tidptr,
2228 int __user *, child_tidptr,
2231 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2232 int __user *, parent_tidptr,
2233 int __user *, child_tidptr,
2237 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2241 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2243 struct task_struct *leader, *parent, *child;
2246 read_lock(&tasklist_lock);
2247 leader = top = top->group_leader;
2249 for_each_thread(leader, parent) {
2250 list_for_each_entry(child, &parent->children, sibling) {
2251 res = visitor(child, data);
2263 if (leader != top) {
2265 parent = child->real_parent;
2266 leader = parent->group_leader;
2270 read_unlock(&tasklist_lock);
2273 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2274 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2277 static void sighand_ctor(void *data)
2279 struct sighand_struct *sighand = data;
2281 spin_lock_init(&sighand->siglock);
2282 init_waitqueue_head(&sighand->signalfd_wqh);
2285 void __init proc_caches_init(void)
2287 unsigned int mm_size;
2289 sighand_cachep = kmem_cache_create("sighand_cache",
2290 sizeof(struct sighand_struct), 0,
2291 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2292 SLAB_ACCOUNT, sighand_ctor);
2293 signal_cachep = kmem_cache_create("signal_cache",
2294 sizeof(struct signal_struct), 0,
2295 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2297 files_cachep = kmem_cache_create("files_cache",
2298 sizeof(struct files_struct), 0,
2299 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2301 fs_cachep = kmem_cache_create("fs_cache",
2302 sizeof(struct fs_struct), 0,
2303 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2307 * The mm_cpumask is located at the end of mm_struct, and is
2308 * dynamically sized based on the maximum CPU number this system
2309 * can have, taking hotplug into account (nr_cpu_ids).
2311 mm_size = sizeof(struct mm_struct) + cpumask_size();
2313 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2314 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2315 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2316 offsetof(struct mm_struct, saved_auxv),
2317 sizeof_field(struct mm_struct, saved_auxv),
2319 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2321 nsproxy_cache_init();
2325 * Check constraints on flags passed to the unshare system call.
2327 static int check_unshare_flags(unsigned long unshare_flags)
2329 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2330 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2331 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2332 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2335 * Not implemented, but pretend it works if there is nothing
2336 * to unshare. Note that unsharing the address space or the
2337 * signal handlers also need to unshare the signal queues (aka
2340 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2341 if (!thread_group_empty(current))
2344 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2345 if (atomic_read(¤t->sighand->count) > 1)
2348 if (unshare_flags & CLONE_VM) {
2349 if (!current_is_single_threaded())
2357 * Unshare the filesystem structure if it is being shared
2359 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2361 struct fs_struct *fs = current->fs;
2363 if (!(unshare_flags & CLONE_FS) || !fs)
2366 /* don't need lock here; in the worst case we'll do useless copy */
2370 *new_fsp = copy_fs_struct(fs);
2378 * Unshare file descriptor table if it is being shared
2380 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2382 struct files_struct *fd = current->files;
2385 if ((unshare_flags & CLONE_FILES) &&
2386 (fd && atomic_read(&fd->count) > 1)) {
2387 *new_fdp = dup_fd(fd, &error);
2396 * unshare allows a process to 'unshare' part of the process
2397 * context which was originally shared using clone. copy_*
2398 * functions used by do_fork() cannot be used here directly
2399 * because they modify an inactive task_struct that is being
2400 * constructed. Here we are modifying the current, active,
2403 int ksys_unshare(unsigned long unshare_flags)
2405 struct fs_struct *fs, *new_fs = NULL;
2406 struct files_struct *fd, *new_fd = NULL;
2407 struct cred *new_cred = NULL;
2408 struct nsproxy *new_nsproxy = NULL;
2413 * If unsharing a user namespace must also unshare the thread group
2414 * and unshare the filesystem root and working directories.
2416 if (unshare_flags & CLONE_NEWUSER)
2417 unshare_flags |= CLONE_THREAD | CLONE_FS;
2419 * If unsharing vm, must also unshare signal handlers.
2421 if (unshare_flags & CLONE_VM)
2422 unshare_flags |= CLONE_SIGHAND;
2424 * If unsharing a signal handlers, must also unshare the signal queues.
2426 if (unshare_flags & CLONE_SIGHAND)
2427 unshare_flags |= CLONE_THREAD;
2429 * If unsharing namespace, must also unshare filesystem information.
2431 if (unshare_flags & CLONE_NEWNS)
2432 unshare_flags |= CLONE_FS;
2434 err = check_unshare_flags(unshare_flags);
2436 goto bad_unshare_out;
2438 * CLONE_NEWIPC must also detach from the undolist: after switching
2439 * to a new ipc namespace, the semaphore arrays from the old
2440 * namespace are unreachable.
2442 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2444 err = unshare_fs(unshare_flags, &new_fs);
2446 goto bad_unshare_out;
2447 err = unshare_fd(unshare_flags, &new_fd);
2449 goto bad_unshare_cleanup_fs;
2450 err = unshare_userns(unshare_flags, &new_cred);
2452 goto bad_unshare_cleanup_fd;
2453 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2456 goto bad_unshare_cleanup_cred;
2458 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2461 * CLONE_SYSVSEM is equivalent to sys_exit().
2465 if (unshare_flags & CLONE_NEWIPC) {
2466 /* Orphan segments in old ns (see sem above). */
2468 shm_init_task(current);
2472 switch_task_namespaces(current, new_nsproxy);
2478 spin_lock(&fs->lock);
2479 current->fs = new_fs;
2484 spin_unlock(&fs->lock);
2488 fd = current->files;
2489 current->files = new_fd;
2493 task_unlock(current);
2496 /* Install the new user namespace */
2497 commit_creds(new_cred);
2502 perf_event_namespaces(current);
2504 bad_unshare_cleanup_cred:
2507 bad_unshare_cleanup_fd:
2509 put_files_struct(new_fd);
2511 bad_unshare_cleanup_fs:
2513 free_fs_struct(new_fs);
2519 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2521 return ksys_unshare(unshare_flags);
2525 * Helper to unshare the files of the current task.
2526 * We don't want to expose copy_files internals to
2527 * the exec layer of the kernel.
2530 int unshare_files(struct files_struct **displaced)
2532 struct task_struct *task = current;
2533 struct files_struct *copy = NULL;
2536 error = unshare_fd(CLONE_FILES, ©);
2537 if (error || !copy) {
2541 *displaced = task->files;
2548 int sysctl_max_threads(struct ctl_table *table, int write,
2549 void __user *buffer, size_t *lenp, loff_t *ppos)
2553 int threads = max_threads;
2554 int min = MIN_THREADS;
2555 int max = MAX_THREADS;
2562 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2566 set_max_threads(threads);