1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
98 #include <asm/pgtable.h>
99 #include <asm/pgalloc.h>
100 #include <linux/uaccess.h>
101 #include <asm/mmu_context.h>
102 #include <asm/cacheflush.h>
103 #include <asm/tlbflush.h>
105 #include <trace/events/sched.h>
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/task.h>
111 * Minimum number of threads to boot the kernel
113 #define MIN_THREADS 20
116 * Maximum number of threads
118 #define MAX_THREADS FUTEX_TID_MASK
121 * Protected counters by write_lock_irq(&tasklist_lock)
123 unsigned long total_forks
; /* Handle normal Linux uptimes. */
124 int nr_threads
; /* The idle threads do not count.. */
126 static int max_threads
; /* tunable limit on nr_threads */
128 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 static const char * const resident_page_types
[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
137 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
139 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
141 #ifdef CONFIG_PROVE_RCU
142 int lockdep_tasklist_lock_is_held(void)
144 return lockdep_is_held(&tasklist_lock
);
146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
147 #endif /* #ifdef CONFIG_PROVE_RCU */
149 int nr_processes(void)
154 for_each_possible_cpu(cpu
)
155 total
+= per_cpu(process_counts
, cpu
);
160 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165 static struct kmem_cache
*task_struct_cachep
;
167 static inline struct task_struct
*alloc_task_struct_node(int node
)
169 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
172 static inline void free_task_struct(struct task_struct
*tsk
)
174 kmem_cache_free(task_struct_cachep
, tsk
);
178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 #ifdef CONFIG_VMAP_STACK
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
191 #define NR_CACHED_STACKS 2
192 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
194 static int free_vm_stack_cache(unsigned int cpu
)
196 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
199 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
200 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
205 vfree(vm_stack
->addr
);
206 cached_vm_stacks
[i
] = NULL
;
213 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
215 #ifdef CONFIG_VMAP_STACK
219 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
222 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
227 /* Clear the KASAN shadow of the stack. */
228 kasan_unpoison_shadow(s
->addr
, THREAD_SIZE
);
230 /* Clear stale pointers from reused stack. */
231 memset(s
->addr
, 0, THREAD_SIZE
);
233 tsk
->stack_vm_area
= s
;
234 tsk
->stack
= s
->addr
;
239 * Allocated stacks are cached and later reused by new threads,
240 * so memcg accounting is performed manually on assigning/releasing
241 * stacks to tasks. Drop __GFP_ACCOUNT.
243 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
244 VMALLOC_START
, VMALLOC_END
,
245 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
247 0, node
, __builtin_return_address(0));
250 * We can't call find_vm_area() in interrupt context, and
251 * free_thread_stack() can be called in interrupt context,
252 * so cache the vm_struct.
255 tsk
->stack_vm_area
= find_vm_area(stack
);
260 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
264 tsk
->stack
= page_address(page
);
271 static inline void free_thread_stack(struct task_struct
*tsk
)
273 #ifdef CONFIG_VMAP_STACK
274 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
279 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
280 mod_memcg_page_state(vm
->pages
[i
],
281 MEMCG_KERNEL_STACK_KB
,
282 -(int)(PAGE_SIZE
/ 1024));
284 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
287 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
288 if (this_cpu_cmpxchg(cached_stacks
[i
],
289 NULL
, tsk
->stack_vm_area
) != NULL
)
295 vfree_atomic(tsk
->stack
);
300 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
303 static struct kmem_cache
*thread_stack_cache
;
305 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
308 unsigned long *stack
;
309 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
314 static void free_thread_stack(struct task_struct
*tsk
)
316 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
319 void thread_stack_cache_init(void)
321 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
322 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
324 BUG_ON(thread_stack_cache
== NULL
);
329 /* SLAB cache for signal_struct structures (tsk->signal) */
330 static struct kmem_cache
*signal_cachep
;
332 /* SLAB cache for sighand_struct structures (tsk->sighand) */
333 struct kmem_cache
*sighand_cachep
;
335 /* SLAB cache for files_struct structures (tsk->files) */
336 struct kmem_cache
*files_cachep
;
338 /* SLAB cache for fs_struct structures (tsk->fs) */
339 struct kmem_cache
*fs_cachep
;
341 /* SLAB cache for vm_area_struct structures */
342 static struct kmem_cache
*vm_area_cachep
;
344 /* SLAB cache for mm_struct structures (tsk->mm) */
345 static struct kmem_cache
*mm_cachep
;
347 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
349 struct vm_area_struct
*vma
;
351 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
357 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
359 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
363 INIT_LIST_HEAD(&new->anon_vma_chain
);
364 new->vm_next
= new->vm_prev
= NULL
;
369 void vm_area_free(struct vm_area_struct
*vma
)
371 kmem_cache_free(vm_area_cachep
, vma
);
374 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
376 void *stack
= task_stack_page(tsk
);
377 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
379 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
384 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
386 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
387 mod_zone_page_state(page_zone(vm
->pages
[i
]),
389 PAGE_SIZE
/ 1024 * account
);
393 * All stack pages are in the same zone and belong to the
396 struct page
*first_page
= virt_to_page(stack
);
398 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
399 THREAD_SIZE
/ 1024 * account
);
401 mod_memcg_obj_state(stack
, MEMCG_KERNEL_STACK_KB
,
402 account
* (THREAD_SIZE
/ 1024));
406 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
408 #ifdef CONFIG_VMAP_STACK
409 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
415 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
417 * If memcg_kmem_charge_page() fails, page->mem_cgroup
418 * pointer is NULL, and both memcg_kmem_uncharge_page()
419 * and mod_memcg_page_state() in free_thread_stack()
420 * will ignore this page. So it's safe.
422 ret
= memcg_kmem_charge_page(vm
->pages
[i
], GFP_KERNEL
,
427 mod_memcg_page_state(vm
->pages
[i
],
428 MEMCG_KERNEL_STACK_KB
,
436 static void release_task_stack(struct task_struct
*tsk
)
438 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
439 return; /* Better to leak the stack than to free prematurely */
441 account_kernel_stack(tsk
, -1);
442 free_thread_stack(tsk
);
444 #ifdef CONFIG_VMAP_STACK
445 tsk
->stack_vm_area
= NULL
;
449 #ifdef CONFIG_THREAD_INFO_IN_TASK
450 void put_task_stack(struct task_struct
*tsk
)
452 if (refcount_dec_and_test(&tsk
->stack_refcount
))
453 release_task_stack(tsk
);
457 void free_task(struct task_struct
*tsk
)
459 #ifndef CONFIG_THREAD_INFO_IN_TASK
461 * The task is finally done with both the stack and thread_info,
464 release_task_stack(tsk
);
467 * If the task had a separate stack allocation, it should be gone
470 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
472 rt_mutex_debug_task_free(tsk
);
473 ftrace_graph_exit_task(tsk
);
474 put_seccomp_filter(tsk
);
475 arch_release_task_struct(tsk
);
476 if (tsk
->flags
& PF_KTHREAD
)
477 free_kthread_struct(tsk
);
478 free_task_struct(tsk
);
480 EXPORT_SYMBOL(free_task
);
483 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
484 struct mm_struct
*oldmm
)
486 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
487 struct rb_node
**rb_link
, *rb_parent
;
489 unsigned long charge
;
492 uprobe_start_dup_mmap();
493 if (down_write_killable(&oldmm
->mmap_sem
)) {
495 goto fail_uprobe_end
;
497 flush_cache_dup_mm(oldmm
);
498 uprobe_dup_mmap(oldmm
, mm
);
500 * Not linked in yet - no deadlock potential:
502 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
504 /* No ordering required: file already has been exposed. */
505 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
507 mm
->total_vm
= oldmm
->total_vm
;
508 mm
->data_vm
= oldmm
->data_vm
;
509 mm
->exec_vm
= oldmm
->exec_vm
;
510 mm
->stack_vm
= oldmm
->stack_vm
;
512 rb_link
= &mm
->mm_rb
.rb_node
;
515 retval
= ksm_fork(mm
, oldmm
);
518 retval
= khugepaged_fork(mm
, oldmm
);
523 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
526 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
527 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
532 * Don't duplicate many vmas if we've been oom-killed (for
535 if (fatal_signal_pending(current
)) {
539 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
540 unsigned long len
= vma_pages(mpnt
);
542 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
546 tmp
= vm_area_dup(mpnt
);
549 retval
= vma_dup_policy(mpnt
, tmp
);
551 goto fail_nomem_policy
;
553 retval
= dup_userfaultfd(tmp
, &uf
);
555 goto fail_nomem_anon_vma_fork
;
556 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
558 * VM_WIPEONFORK gets a clean slate in the child.
559 * Don't prepare anon_vma until fault since we don't
560 * copy page for current vma.
562 tmp
->anon_vma
= NULL
;
563 } else if (anon_vma_fork(tmp
, mpnt
))
564 goto fail_nomem_anon_vma_fork
;
565 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
568 struct inode
*inode
= file_inode(file
);
569 struct address_space
*mapping
= file
->f_mapping
;
572 if (tmp
->vm_flags
& VM_DENYWRITE
)
573 atomic_dec(&inode
->i_writecount
);
574 i_mmap_lock_write(mapping
);
575 if (tmp
->vm_flags
& VM_SHARED
)
576 atomic_inc(&mapping
->i_mmap_writable
);
577 flush_dcache_mmap_lock(mapping
);
578 /* insert tmp into the share list, just after mpnt */
579 vma_interval_tree_insert_after(tmp
, mpnt
,
581 flush_dcache_mmap_unlock(mapping
);
582 i_mmap_unlock_write(mapping
);
586 * Clear hugetlb-related page reserves for children. This only
587 * affects MAP_PRIVATE mappings. Faults generated by the child
588 * are not guaranteed to succeed, even if read-only
590 if (is_vm_hugetlb_page(tmp
))
591 reset_vma_resv_huge_pages(tmp
);
594 * Link in the new vma and copy the page table entries.
597 pprev
= &tmp
->vm_next
;
601 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
602 rb_link
= &tmp
->vm_rb
.rb_right
;
603 rb_parent
= &tmp
->vm_rb
;
606 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
607 retval
= copy_page_range(mm
, oldmm
, mpnt
);
609 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
610 tmp
->vm_ops
->open(tmp
);
615 /* a new mm has just been created */
616 retval
= arch_dup_mmap(oldmm
, mm
);
618 up_write(&mm
->mmap_sem
);
620 up_write(&oldmm
->mmap_sem
);
621 dup_userfaultfd_complete(&uf
);
623 uprobe_end_dup_mmap();
625 fail_nomem_anon_vma_fork
:
626 mpol_put(vma_policy(tmp
));
631 vm_unacct_memory(charge
);
635 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
637 mm
->pgd
= pgd_alloc(mm
);
638 if (unlikely(!mm
->pgd
))
643 static inline void mm_free_pgd(struct mm_struct
*mm
)
645 pgd_free(mm
, mm
->pgd
);
648 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
650 down_write(&oldmm
->mmap_sem
);
651 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
652 up_write(&oldmm
->mmap_sem
);
655 #define mm_alloc_pgd(mm) (0)
656 #define mm_free_pgd(mm)
657 #endif /* CONFIG_MMU */
659 static void check_mm(struct mm_struct
*mm
)
663 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
664 "Please make sure 'struct resident_page_types[]' is updated as well");
666 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
667 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
670 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
671 mm
, resident_page_types
[i
], x
);
674 if (mm_pgtables_bytes(mm
))
675 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
676 mm_pgtables_bytes(mm
));
678 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
679 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
683 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
684 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
687 * Called when the last reference to the mm
688 * is dropped: either by a lazy thread or by
689 * mmput. Free the page directory and the mm.
691 void __mmdrop(struct mm_struct
*mm
)
693 BUG_ON(mm
== &init_mm
);
694 WARN_ON_ONCE(mm
== current
->mm
);
695 WARN_ON_ONCE(mm
== current
->active_mm
);
698 mmu_notifier_subscriptions_destroy(mm
);
700 put_user_ns(mm
->user_ns
);
703 EXPORT_SYMBOL_GPL(__mmdrop
);
705 static void mmdrop_async_fn(struct work_struct
*work
)
707 struct mm_struct
*mm
;
709 mm
= container_of(work
, struct mm_struct
, async_put_work
);
713 static void mmdrop_async(struct mm_struct
*mm
)
715 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
716 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
717 schedule_work(&mm
->async_put_work
);
721 static inline void free_signal_struct(struct signal_struct
*sig
)
723 taskstats_tgid_free(sig
);
724 sched_autogroup_exit(sig
);
726 * __mmdrop is not safe to call from softirq context on x86 due to
727 * pgd_dtor so postpone it to the async context
730 mmdrop_async(sig
->oom_mm
);
731 kmem_cache_free(signal_cachep
, sig
);
734 static inline void put_signal_struct(struct signal_struct
*sig
)
736 if (refcount_dec_and_test(&sig
->sigcnt
))
737 free_signal_struct(sig
);
740 void __put_task_struct(struct task_struct
*tsk
)
742 WARN_ON(!tsk
->exit_state
);
743 WARN_ON(refcount_read(&tsk
->usage
));
744 WARN_ON(tsk
== current
);
747 task_numa_free(tsk
, true);
748 security_task_free(tsk
);
750 delayacct_tsk_free(tsk
);
751 put_signal_struct(tsk
->signal
);
753 if (!profile_handoff_task(tsk
))
756 EXPORT_SYMBOL_GPL(__put_task_struct
);
758 void __init __weak
arch_task_cache_init(void) { }
763 static void set_max_threads(unsigned int max_threads_suggested
)
766 unsigned long nr_pages
= totalram_pages();
769 * The number of threads shall be limited such that the thread
770 * structures may only consume a small part of the available memory.
772 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
773 threads
= MAX_THREADS
;
775 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
776 (u64
) THREAD_SIZE
* 8UL);
778 if (threads
> max_threads_suggested
)
779 threads
= max_threads_suggested
;
781 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
784 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
785 /* Initialized by the architecture: */
786 int arch_task_struct_size __read_mostly
;
789 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
790 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
792 /* Fetch thread_struct whitelist for the architecture. */
793 arch_thread_struct_whitelist(offset
, size
);
796 * Handle zero-sized whitelist or empty thread_struct, otherwise
797 * adjust offset to position of thread_struct in task_struct.
799 if (unlikely(*size
== 0))
802 *offset
+= offsetof(struct task_struct
, thread
);
804 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
806 void __init
fork_init(void)
809 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
810 #ifndef ARCH_MIN_TASKALIGN
811 #define ARCH_MIN_TASKALIGN 0
813 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
814 unsigned long useroffset
, usersize
;
816 /* create a slab on which task_structs can be allocated */
817 task_struct_whitelist(&useroffset
, &usersize
);
818 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
819 arch_task_struct_size
, align
,
820 SLAB_PANIC
|SLAB_ACCOUNT
,
821 useroffset
, usersize
, NULL
);
824 /* do the arch specific task caches init */
825 arch_task_cache_init();
827 set_max_threads(MAX_THREADS
);
829 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
830 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
831 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
832 init_task
.signal
->rlim
[RLIMIT_NPROC
];
834 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
835 init_user_ns
.ucount_max
[i
] = max_threads
/2;
838 #ifdef CONFIG_VMAP_STACK
839 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
840 NULL
, free_vm_stack_cache
);
843 lockdep_init_task(&init_task
);
847 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
848 struct task_struct
*src
)
854 void set_task_stack_end_magic(struct task_struct
*tsk
)
856 unsigned long *stackend
;
858 stackend
= end_of_stack(tsk
);
859 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
862 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
864 struct task_struct
*tsk
;
865 unsigned long *stack
;
866 struct vm_struct
*stack_vm_area __maybe_unused
;
869 if (node
== NUMA_NO_NODE
)
870 node
= tsk_fork_get_node(orig
);
871 tsk
= alloc_task_struct_node(node
);
875 stack
= alloc_thread_stack_node(tsk
, node
);
879 if (memcg_charge_kernel_stack(tsk
))
882 stack_vm_area
= task_stack_vm_area(tsk
);
884 err
= arch_dup_task_struct(tsk
, orig
);
887 * arch_dup_task_struct() clobbers the stack-related fields. Make
888 * sure they're properly initialized before using any stack-related
892 #ifdef CONFIG_VMAP_STACK
893 tsk
->stack_vm_area
= stack_vm_area
;
895 #ifdef CONFIG_THREAD_INFO_IN_TASK
896 refcount_set(&tsk
->stack_refcount
, 1);
902 #ifdef CONFIG_SECCOMP
904 * We must handle setting up seccomp filters once we're under
905 * the sighand lock in case orig has changed between now and
906 * then. Until then, filter must be NULL to avoid messing up
907 * the usage counts on the error path calling free_task.
909 tsk
->seccomp
.filter
= NULL
;
912 setup_thread_stack(tsk
, orig
);
913 clear_user_return_notifier(tsk
);
914 clear_tsk_need_resched(tsk
);
915 set_task_stack_end_magic(tsk
);
917 #ifdef CONFIG_STACKPROTECTOR
918 tsk
->stack_canary
= get_random_canary();
920 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
921 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
924 * One for the user space visible state that goes away when reaped.
925 * One for the scheduler.
927 refcount_set(&tsk
->rcu_users
, 2);
928 /* One for the rcu users */
929 refcount_set(&tsk
->usage
, 1);
930 #ifdef CONFIG_BLK_DEV_IO_TRACE
933 tsk
->splice_pipe
= NULL
;
934 tsk
->task_frag
.page
= NULL
;
935 tsk
->wake_q
.next
= NULL
;
937 account_kernel_stack(tsk
, 1);
941 #ifdef CONFIG_FAULT_INJECTION
945 #ifdef CONFIG_BLK_CGROUP
946 tsk
->throttle_queue
= NULL
;
947 tsk
->use_memdelay
= 0;
951 tsk
->active_memcg
= NULL
;
956 free_thread_stack(tsk
);
958 free_task_struct(tsk
);
962 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
964 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
966 static int __init
coredump_filter_setup(char *s
)
968 default_dump_filter
=
969 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
970 MMF_DUMP_FILTER_MASK
;
974 __setup("coredump_filter=", coredump_filter_setup
);
976 #include <linux/init_task.h>
978 static void mm_init_aio(struct mm_struct
*mm
)
981 spin_lock_init(&mm
->ioctx_lock
);
982 mm
->ioctx_table
= NULL
;
986 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
987 struct task_struct
*p
)
991 WRITE_ONCE(mm
->owner
, NULL
);
995 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
1002 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1004 #ifdef CONFIG_UPROBES
1005 mm
->uprobes_state
.xol_area
= NULL
;
1009 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1010 struct user_namespace
*user_ns
)
1013 mm
->mm_rb
= RB_ROOT
;
1014 mm
->vmacache_seqnum
= 0;
1015 atomic_set(&mm
->mm_users
, 1);
1016 atomic_set(&mm
->mm_count
, 1);
1017 init_rwsem(&mm
->mmap_sem
);
1018 INIT_LIST_HEAD(&mm
->mmlist
);
1019 mm
->core_state
= NULL
;
1020 mm_pgtables_bytes_init(mm
);
1023 atomic64_set(&mm
->pinned_vm
, 0);
1024 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1025 spin_lock_init(&mm
->page_table_lock
);
1026 spin_lock_init(&mm
->arg_lock
);
1027 mm_init_cpumask(mm
);
1029 mm_init_owner(mm
, p
);
1030 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1031 mmu_notifier_subscriptions_init(mm
);
1032 init_tlb_flush_pending(mm
);
1033 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1034 mm
->pmd_huge_pte
= NULL
;
1036 mm_init_uprobes_state(mm
);
1039 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1040 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1042 mm
->flags
= default_dump_filter
;
1046 if (mm_alloc_pgd(mm
))
1049 if (init_new_context(p
, mm
))
1050 goto fail_nocontext
;
1052 mm
->user_ns
= get_user_ns(user_ns
);
1063 * Allocate and initialize an mm_struct.
1065 struct mm_struct
*mm_alloc(void)
1067 struct mm_struct
*mm
;
1073 memset(mm
, 0, sizeof(*mm
));
1074 return mm_init(mm
, current
, current_user_ns());
1077 static inline void __mmput(struct mm_struct
*mm
)
1079 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1081 uprobe_clear_state(mm
);
1084 khugepaged_exit(mm
); /* must run before exit_mmap */
1086 mm_put_huge_zero_page(mm
);
1087 set_mm_exe_file(mm
, NULL
);
1088 if (!list_empty(&mm
->mmlist
)) {
1089 spin_lock(&mmlist_lock
);
1090 list_del(&mm
->mmlist
);
1091 spin_unlock(&mmlist_lock
);
1094 module_put(mm
->binfmt
->module
);
1099 * Decrement the use count and release all resources for an mm.
1101 void mmput(struct mm_struct
*mm
)
1105 if (atomic_dec_and_test(&mm
->mm_users
))
1108 EXPORT_SYMBOL_GPL(mmput
);
1111 static void mmput_async_fn(struct work_struct
*work
)
1113 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1119 void mmput_async(struct mm_struct
*mm
)
1121 if (atomic_dec_and_test(&mm
->mm_users
)) {
1122 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1123 schedule_work(&mm
->async_put_work
);
1129 * set_mm_exe_file - change a reference to the mm's executable file
1131 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1133 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1134 * invocations: in mmput() nobody alive left, in execve task is single
1135 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1136 * mm->exe_file, but does so without using set_mm_exe_file() in order
1137 * to do avoid the need for any locks.
1139 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1141 struct file
*old_exe_file
;
1144 * It is safe to dereference the exe_file without RCU as
1145 * this function is only called if nobody else can access
1146 * this mm -- see comment above for justification.
1148 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1151 get_file(new_exe_file
);
1152 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1158 * get_mm_exe_file - acquire a reference to the mm's executable file
1160 * Returns %NULL if mm has no associated executable file.
1161 * User must release file via fput().
1163 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1165 struct file
*exe_file
;
1168 exe_file
= rcu_dereference(mm
->exe_file
);
1169 if (exe_file
&& !get_file_rcu(exe_file
))
1174 EXPORT_SYMBOL(get_mm_exe_file
);
1177 * get_task_exe_file - acquire a reference to the task's executable file
1179 * Returns %NULL if task's mm (if any) has no associated executable file or
1180 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1181 * User must release file via fput().
1183 struct file
*get_task_exe_file(struct task_struct
*task
)
1185 struct file
*exe_file
= NULL
;
1186 struct mm_struct
*mm
;
1191 if (!(task
->flags
& PF_KTHREAD
))
1192 exe_file
= get_mm_exe_file(mm
);
1197 EXPORT_SYMBOL(get_task_exe_file
);
1200 * get_task_mm - acquire a reference to the task's mm
1202 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1203 * this kernel workthread has transiently adopted a user mm with use_mm,
1204 * to do its AIO) is not set and if so returns a reference to it, after
1205 * bumping up the use count. User must release the mm via mmput()
1206 * after use. Typically used by /proc and ptrace.
1208 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1210 struct mm_struct
*mm
;
1215 if (task
->flags
& PF_KTHREAD
)
1223 EXPORT_SYMBOL_GPL(get_task_mm
);
1225 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1227 struct mm_struct
*mm
;
1230 err
= mutex_lock_killable(&task
->signal
->exec_update_mutex
);
1232 return ERR_PTR(err
);
1234 mm
= get_task_mm(task
);
1235 if (mm
&& mm
!= current
->mm
&&
1236 !ptrace_may_access(task
, mode
)) {
1238 mm
= ERR_PTR(-EACCES
);
1240 mutex_unlock(&task
->signal
->exec_update_mutex
);
1245 static void complete_vfork_done(struct task_struct
*tsk
)
1247 struct completion
*vfork
;
1250 vfork
= tsk
->vfork_done
;
1251 if (likely(vfork
)) {
1252 tsk
->vfork_done
= NULL
;
1258 static int wait_for_vfork_done(struct task_struct
*child
,
1259 struct completion
*vfork
)
1263 freezer_do_not_count();
1264 cgroup_enter_frozen();
1265 killed
= wait_for_completion_killable(vfork
);
1266 cgroup_leave_frozen(false);
1271 child
->vfork_done
= NULL
;
1275 put_task_struct(child
);
1279 /* Please note the differences between mmput and mm_release.
1280 * mmput is called whenever we stop holding onto a mm_struct,
1281 * error success whatever.
1283 * mm_release is called after a mm_struct has been removed
1284 * from the current process.
1286 * This difference is important for error handling, when we
1287 * only half set up a mm_struct for a new process and need to restore
1288 * the old one. Because we mmput the new mm_struct before
1289 * restoring the old one. . .
1290 * Eric Biederman 10 January 1998
1292 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1294 uprobe_free_utask(tsk
);
1296 /* Get rid of any cached register state */
1297 deactivate_mm(tsk
, mm
);
1300 * Signal userspace if we're not exiting with a core dump
1301 * because we want to leave the value intact for debugging
1304 if (tsk
->clear_child_tid
) {
1305 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1306 atomic_read(&mm
->mm_users
) > 1) {
1308 * We don't check the error code - if userspace has
1309 * not set up a proper pointer then tough luck.
1311 put_user(0, tsk
->clear_child_tid
);
1312 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1313 1, NULL
, NULL
, 0, 0);
1315 tsk
->clear_child_tid
= NULL
;
1319 * All done, finally we can wake up parent and return this mm to him.
1320 * Also kthread_stop() uses this completion for synchronization.
1322 if (tsk
->vfork_done
)
1323 complete_vfork_done(tsk
);
1326 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1328 futex_exit_release(tsk
);
1329 mm_release(tsk
, mm
);
1332 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1334 futex_exec_release(tsk
);
1335 mm_release(tsk
, mm
);
1339 * dup_mm() - duplicates an existing mm structure
1340 * @tsk: the task_struct with which the new mm will be associated.
1341 * @oldmm: the mm to duplicate.
1343 * Allocates a new mm structure and duplicates the provided @oldmm structure
1346 * Return: the duplicated mm or NULL on failure.
1348 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1349 struct mm_struct
*oldmm
)
1351 struct mm_struct
*mm
;
1358 memcpy(mm
, oldmm
, sizeof(*mm
));
1360 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1363 err
= dup_mmap(mm
, oldmm
);
1367 mm
->hiwater_rss
= get_mm_rss(mm
);
1368 mm
->hiwater_vm
= mm
->total_vm
;
1370 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1376 /* don't put binfmt in mmput, we haven't got module yet */
1378 mm_init_owner(mm
, NULL
);
1385 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1387 struct mm_struct
*mm
, *oldmm
;
1390 tsk
->min_flt
= tsk
->maj_flt
= 0;
1391 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1392 #ifdef CONFIG_DETECT_HUNG_TASK
1393 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1394 tsk
->last_switch_time
= 0;
1398 tsk
->active_mm
= NULL
;
1401 * Are we cloning a kernel thread?
1403 * We need to steal a active VM for that..
1405 oldmm
= current
->mm
;
1409 /* initialize the new vmacache entries */
1410 vmacache_flush(tsk
);
1412 if (clone_flags
& CLONE_VM
) {
1419 mm
= dup_mm(tsk
, current
->mm
);
1425 tsk
->active_mm
= mm
;
1432 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1434 struct fs_struct
*fs
= current
->fs
;
1435 if (clone_flags
& CLONE_FS
) {
1436 /* tsk->fs is already what we want */
1437 spin_lock(&fs
->lock
);
1439 spin_unlock(&fs
->lock
);
1443 spin_unlock(&fs
->lock
);
1446 tsk
->fs
= copy_fs_struct(fs
);
1452 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1454 struct files_struct
*oldf
, *newf
;
1458 * A background process may not have any files ...
1460 oldf
= current
->files
;
1464 if (clone_flags
& CLONE_FILES
) {
1465 atomic_inc(&oldf
->count
);
1469 newf
= dup_fd(oldf
, &error
);
1479 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1482 struct io_context
*ioc
= current
->io_context
;
1483 struct io_context
*new_ioc
;
1488 * Share io context with parent, if CLONE_IO is set
1490 if (clone_flags
& CLONE_IO
) {
1492 tsk
->io_context
= ioc
;
1493 } else if (ioprio_valid(ioc
->ioprio
)) {
1494 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1495 if (unlikely(!new_ioc
))
1498 new_ioc
->ioprio
= ioc
->ioprio
;
1499 put_io_context(new_ioc
);
1505 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1507 struct sighand_struct
*sig
;
1509 if (clone_flags
& CLONE_SIGHAND
) {
1510 refcount_inc(¤t
->sighand
->count
);
1513 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1514 RCU_INIT_POINTER(tsk
->sighand
, sig
);
1518 refcount_set(&sig
->count
, 1);
1519 spin_lock_irq(¤t
->sighand
->siglock
);
1520 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1521 spin_unlock_irq(¤t
->sighand
->siglock
);
1523 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1524 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1525 flush_signal_handlers(tsk
, 0);
1530 void __cleanup_sighand(struct sighand_struct
*sighand
)
1532 if (refcount_dec_and_test(&sighand
->count
)) {
1533 signalfd_cleanup(sighand
);
1535 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1536 * without an RCU grace period, see __lock_task_sighand().
1538 kmem_cache_free(sighand_cachep
, sighand
);
1543 * Initialize POSIX timer handling for a thread group.
1545 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1547 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1548 unsigned long cpu_limit
;
1550 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1551 posix_cputimers_group_init(pct
, cpu_limit
);
1554 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1556 struct signal_struct
*sig
;
1558 if (clone_flags
& CLONE_THREAD
)
1561 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1566 sig
->nr_threads
= 1;
1567 atomic_set(&sig
->live
, 1);
1568 refcount_set(&sig
->sigcnt
, 1);
1570 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1571 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1572 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1574 init_waitqueue_head(&sig
->wait_chldexit
);
1575 sig
->curr_target
= tsk
;
1576 init_sigpending(&sig
->shared_pending
);
1577 INIT_HLIST_HEAD(&sig
->multiprocess
);
1578 seqlock_init(&sig
->stats_lock
);
1579 prev_cputime_init(&sig
->prev_cputime
);
1581 #ifdef CONFIG_POSIX_TIMERS
1582 INIT_LIST_HEAD(&sig
->posix_timers
);
1583 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1584 sig
->real_timer
.function
= it_real_fn
;
1587 task_lock(current
->group_leader
);
1588 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1589 task_unlock(current
->group_leader
);
1591 posix_cpu_timers_init_group(sig
);
1593 tty_audit_fork(sig
);
1594 sched_autogroup_fork(sig
);
1596 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1597 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1599 mutex_init(&sig
->cred_guard_mutex
);
1600 mutex_init(&sig
->exec_update_mutex
);
1605 static void copy_seccomp(struct task_struct
*p
)
1607 #ifdef CONFIG_SECCOMP
1609 * Must be called with sighand->lock held, which is common to
1610 * all threads in the group. Holding cred_guard_mutex is not
1611 * needed because this new task is not yet running and cannot
1614 assert_spin_locked(¤t
->sighand
->siglock
);
1616 /* Ref-count the new filter user, and assign it. */
1617 get_seccomp_filter(current
);
1618 p
->seccomp
= current
->seccomp
;
1621 * Explicitly enable no_new_privs here in case it got set
1622 * between the task_struct being duplicated and holding the
1623 * sighand lock. The seccomp state and nnp must be in sync.
1625 if (task_no_new_privs(current
))
1626 task_set_no_new_privs(p
);
1629 * If the parent gained a seccomp mode after copying thread
1630 * flags and between before we held the sighand lock, we have
1631 * to manually enable the seccomp thread flag here.
1633 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1634 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1638 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1640 current
->clear_child_tid
= tidptr
;
1642 return task_pid_vnr(current
);
1645 static void rt_mutex_init_task(struct task_struct
*p
)
1647 raw_spin_lock_init(&p
->pi_lock
);
1648 #ifdef CONFIG_RT_MUTEXES
1649 p
->pi_waiters
= RB_ROOT_CACHED
;
1650 p
->pi_top_task
= NULL
;
1651 p
->pi_blocked_on
= NULL
;
1655 static inline void init_task_pid_links(struct task_struct
*task
)
1659 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1660 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1665 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1667 if (type
== PIDTYPE_PID
)
1668 task
->thread_pid
= pid
;
1670 task
->signal
->pids
[type
] = pid
;
1673 static inline void rcu_copy_process(struct task_struct
*p
)
1675 #ifdef CONFIG_PREEMPT_RCU
1676 p
->rcu_read_lock_nesting
= 0;
1677 p
->rcu_read_unlock_special
.s
= 0;
1678 p
->rcu_blocked_node
= NULL
;
1679 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1680 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1681 #ifdef CONFIG_TASKS_RCU
1682 p
->rcu_tasks_holdout
= false;
1683 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1684 p
->rcu_tasks_idle_cpu
= -1;
1685 #endif /* #ifdef CONFIG_TASKS_RCU */
1688 struct pid
*pidfd_pid(const struct file
*file
)
1690 if (file
->f_op
== &pidfd_fops
)
1691 return file
->private_data
;
1693 return ERR_PTR(-EBADF
);
1696 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1698 struct pid
*pid
= file
->private_data
;
1700 file
->private_data
= NULL
;
1705 #ifdef CONFIG_PROC_FS
1707 * pidfd_show_fdinfo - print information about a pidfd
1708 * @m: proc fdinfo file
1709 * @f: file referencing a pidfd
1712 * This function will print the pid that a given pidfd refers to in the
1713 * pid namespace of the procfs instance.
1714 * If the pid namespace of the process is not a descendant of the pid
1715 * namespace of the procfs instance 0 will be shown as its pid. This is
1716 * similar to calling getppid() on a process whose parent is outside of
1717 * its pid namespace.
1720 * If pid namespaces are supported then this function will also print
1721 * the pid of a given pidfd refers to for all descendant pid namespaces
1722 * starting from the current pid namespace of the instance, i.e. the
1723 * Pid field and the first entry in the NSpid field will be identical.
1724 * If the pid namespace of the process is not a descendant of the pid
1725 * namespace of the procfs instance 0 will be shown as its first NSpid
1726 * entry and no others will be shown.
1727 * Note that this differs from the Pid and NSpid fields in
1728 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1729 * the pid namespace of the procfs instance. The difference becomes
1730 * obvious when sending around a pidfd between pid namespaces from a
1731 * different branch of the tree, i.e. where no ancestoral relation is
1732 * present between the pid namespaces:
1733 * - create two new pid namespaces ns1 and ns2 in the initial pid
1734 * namespace (also take care to create new mount namespaces in the
1735 * new pid namespace and mount procfs)
1736 * - create a process with a pidfd in ns1
1737 * - send pidfd from ns1 to ns2
1738 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1739 * have exactly one entry, which is 0
1741 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1743 struct pid
*pid
= f
->private_data
;
1744 struct pid_namespace
*ns
;
1747 if (likely(pid_has_task(pid
, PIDTYPE_PID
))) {
1748 ns
= proc_pid_ns(file_inode(m
->file
));
1749 nr
= pid_nr_ns(pid
, ns
);
1752 seq_put_decimal_ll(m
, "Pid:\t", nr
);
1754 #ifdef CONFIG_PID_NS
1755 seq_put_decimal_ll(m
, "\nNSpid:\t", nr
);
1759 /* If nr is non-zero it means that 'pid' is valid and that
1760 * ns, i.e. the pid namespace associated with the procfs
1761 * instance, is in the pid namespace hierarchy of pid.
1762 * Start at one below the already printed level.
1764 for (i
= ns
->level
+ 1; i
<= pid
->level
; i
++)
1765 seq_put_decimal_ll(m
, "\t", pid
->numbers
[i
].nr
);
1773 * Poll support for process exit notification.
1775 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1777 struct task_struct
*task
;
1778 struct pid
*pid
= file
->private_data
;
1779 __poll_t poll_flags
= 0;
1781 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1784 task
= pid_task(pid
, PIDTYPE_PID
);
1786 * Inform pollers only when the whole thread group exits.
1787 * If the thread group leader exits before all other threads in the
1788 * group, then poll(2) should block, similar to the wait(2) family.
1790 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1791 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1797 const struct file_operations pidfd_fops
= {
1798 .release
= pidfd_release
,
1800 #ifdef CONFIG_PROC_FS
1801 .show_fdinfo
= pidfd_show_fdinfo
,
1805 static void __delayed_free_task(struct rcu_head
*rhp
)
1807 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1812 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1814 if (IS_ENABLED(CONFIG_MEMCG
))
1815 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1821 * This creates a new process as a copy of the old one,
1822 * but does not actually start it yet.
1824 * It copies the registers, and all the appropriate
1825 * parts of the process environment (as per the clone
1826 * flags). The actual kick-off is left to the caller.
1828 static __latent_entropy
struct task_struct
*copy_process(
1832 struct kernel_clone_args
*args
)
1834 int pidfd
= -1, retval
;
1835 struct task_struct
*p
;
1836 struct multiprocess_signals delayed
;
1837 struct file
*pidfile
= NULL
;
1838 u64 clone_flags
= args
->flags
;
1839 struct nsproxy
*nsp
= current
->nsproxy
;
1842 * Don't allow sharing the root directory with processes in a different
1845 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1846 return ERR_PTR(-EINVAL
);
1848 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1849 return ERR_PTR(-EINVAL
);
1852 * Thread groups must share signals as well, and detached threads
1853 * can only be started up within the thread group.
1855 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1856 return ERR_PTR(-EINVAL
);
1859 * Shared signal handlers imply shared VM. By way of the above,
1860 * thread groups also imply shared VM. Blocking this case allows
1861 * for various simplifications in other code.
1863 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1864 return ERR_PTR(-EINVAL
);
1867 * Siblings of global init remain as zombies on exit since they are
1868 * not reaped by their parent (swapper). To solve this and to avoid
1869 * multi-rooted process trees, prevent global and container-inits
1870 * from creating siblings.
1872 if ((clone_flags
& CLONE_PARENT
) &&
1873 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1874 return ERR_PTR(-EINVAL
);
1877 * If the new process will be in a different pid or user namespace
1878 * do not allow it to share a thread group with the forking task.
1880 if (clone_flags
& CLONE_THREAD
) {
1881 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1882 (task_active_pid_ns(current
) != nsp
->pid_ns_for_children
))
1883 return ERR_PTR(-EINVAL
);
1887 * If the new process will be in a different time namespace
1888 * do not allow it to share VM or a thread group with the forking task.
1890 if (clone_flags
& (CLONE_THREAD
| CLONE_VM
)) {
1891 if (nsp
->time_ns
!= nsp
->time_ns_for_children
)
1892 return ERR_PTR(-EINVAL
);
1895 if (clone_flags
& CLONE_PIDFD
) {
1897 * - CLONE_DETACHED is blocked so that we can potentially
1898 * reuse it later for CLONE_PIDFD.
1899 * - CLONE_THREAD is blocked until someone really needs it.
1901 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1902 return ERR_PTR(-EINVAL
);
1906 * Force any signals received before this point to be delivered
1907 * before the fork happens. Collect up signals sent to multiple
1908 * processes that happen during the fork and delay them so that
1909 * they appear to happen after the fork.
1911 sigemptyset(&delayed
.signal
);
1912 INIT_HLIST_NODE(&delayed
.node
);
1914 spin_lock_irq(¤t
->sighand
->siglock
);
1915 if (!(clone_flags
& CLONE_THREAD
))
1916 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1917 recalc_sigpending();
1918 spin_unlock_irq(¤t
->sighand
->siglock
);
1919 retval
= -ERESTARTNOINTR
;
1920 if (signal_pending(current
))
1924 p
= dup_task_struct(current
, node
);
1929 * This _must_ happen before we call free_task(), i.e. before we jump
1930 * to any of the bad_fork_* labels. This is to avoid freeing
1931 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1932 * kernel threads (PF_KTHREAD).
1934 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1936 * Clear TID on mm_release()?
1938 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1940 ftrace_graph_init_task(p
);
1942 rt_mutex_init_task(p
);
1944 #ifdef CONFIG_PROVE_LOCKING
1945 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1946 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1949 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1950 task_rlimit(p
, RLIMIT_NPROC
)) {
1951 if (p
->real_cred
->user
!= INIT_USER
&&
1952 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1955 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1957 retval
= copy_creds(p
, clone_flags
);
1962 * If multiple threads are within copy_process(), then this check
1963 * triggers too late. This doesn't hurt, the check is only there
1964 * to stop root fork bombs.
1967 if (nr_threads
>= max_threads
)
1968 goto bad_fork_cleanup_count
;
1970 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1971 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1972 p
->flags
|= PF_FORKNOEXEC
;
1973 INIT_LIST_HEAD(&p
->children
);
1974 INIT_LIST_HEAD(&p
->sibling
);
1975 rcu_copy_process(p
);
1976 p
->vfork_done
= NULL
;
1977 spin_lock_init(&p
->alloc_lock
);
1979 init_sigpending(&p
->pending
);
1981 p
->utime
= p
->stime
= p
->gtime
= 0;
1982 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1983 p
->utimescaled
= p
->stimescaled
= 0;
1985 prev_cputime_init(&p
->prev_cputime
);
1987 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1988 seqcount_init(&p
->vtime
.seqcount
);
1989 p
->vtime
.starttime
= 0;
1990 p
->vtime
.state
= VTIME_INACTIVE
;
1993 #if defined(SPLIT_RSS_COUNTING)
1994 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1997 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
2003 task_io_accounting_init(&p
->ioac
);
2004 acct_clear_integrals(p
);
2006 posix_cputimers_init(&p
->posix_cputimers
);
2008 p
->io_context
= NULL
;
2009 audit_set_context(p
, NULL
);
2012 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2013 if (IS_ERR(p
->mempolicy
)) {
2014 retval
= PTR_ERR(p
->mempolicy
);
2015 p
->mempolicy
= NULL
;
2016 goto bad_fork_cleanup_threadgroup_lock
;
2019 #ifdef CONFIG_CPUSETS
2020 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2021 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2022 seqcount_init(&p
->mems_allowed_seq
);
2024 #ifdef CONFIG_TRACE_IRQFLAGS
2026 p
->hardirqs_enabled
= 0;
2027 p
->hardirq_enable_ip
= 0;
2028 p
->hardirq_enable_event
= 0;
2029 p
->hardirq_disable_ip
= _THIS_IP_
;
2030 p
->hardirq_disable_event
= 0;
2031 p
->softirqs_enabled
= 1;
2032 p
->softirq_enable_ip
= _THIS_IP_
;
2033 p
->softirq_enable_event
= 0;
2034 p
->softirq_disable_ip
= 0;
2035 p
->softirq_disable_event
= 0;
2036 p
->hardirq_context
= 0;
2037 p
->softirq_context
= 0;
2040 p
->pagefault_disabled
= 0;
2042 #ifdef CONFIG_LOCKDEP
2043 lockdep_init_task(p
);
2046 #ifdef CONFIG_DEBUG_MUTEXES
2047 p
->blocked_on
= NULL
; /* not blocked yet */
2049 #ifdef CONFIG_BCACHE
2050 p
->sequential_io
= 0;
2051 p
->sequential_io_avg
= 0;
2054 /* Perform scheduler related setup. Assign this task to a CPU. */
2055 retval
= sched_fork(clone_flags
, p
);
2057 goto bad_fork_cleanup_policy
;
2059 retval
= perf_event_init_task(p
);
2061 goto bad_fork_cleanup_policy
;
2062 retval
= audit_alloc(p
);
2064 goto bad_fork_cleanup_perf
;
2065 /* copy all the process information */
2067 retval
= security_task_alloc(p
, clone_flags
);
2069 goto bad_fork_cleanup_audit
;
2070 retval
= copy_semundo(clone_flags
, p
);
2072 goto bad_fork_cleanup_security
;
2073 retval
= copy_files(clone_flags
, p
);
2075 goto bad_fork_cleanup_semundo
;
2076 retval
= copy_fs(clone_flags
, p
);
2078 goto bad_fork_cleanup_files
;
2079 retval
= copy_sighand(clone_flags
, p
);
2081 goto bad_fork_cleanup_fs
;
2082 retval
= copy_signal(clone_flags
, p
);
2084 goto bad_fork_cleanup_sighand
;
2085 retval
= copy_mm(clone_flags
, p
);
2087 goto bad_fork_cleanup_signal
;
2088 retval
= copy_namespaces(clone_flags
, p
);
2090 goto bad_fork_cleanup_mm
;
2091 retval
= copy_io(clone_flags
, p
);
2093 goto bad_fork_cleanup_namespaces
;
2094 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2097 goto bad_fork_cleanup_io
;
2099 stackleak_task_init(p
);
2101 if (pid
!= &init_struct_pid
) {
2102 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2103 args
->set_tid_size
);
2105 retval
= PTR_ERR(pid
);
2106 goto bad_fork_cleanup_thread
;
2111 * This has to happen after we've potentially unshared the file
2112 * descriptor table (so that the pidfd doesn't leak into the child
2113 * if the fd table isn't shared).
2115 if (clone_flags
& CLONE_PIDFD
) {
2116 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2118 goto bad_fork_free_pid
;
2122 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2123 O_RDWR
| O_CLOEXEC
);
2124 if (IS_ERR(pidfile
)) {
2125 put_unused_fd(pidfd
);
2126 retval
= PTR_ERR(pidfile
);
2127 goto bad_fork_free_pid
;
2129 get_pid(pid
); /* held by pidfile now */
2131 retval
= put_user(pidfd
, args
->pidfd
);
2133 goto bad_fork_put_pidfd
;
2142 * sigaltstack should be cleared when sharing the same VM
2144 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2148 * Syscall tracing and stepping should be turned off in the
2149 * child regardless of CLONE_PTRACE.
2151 user_disable_single_step(p
);
2152 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2153 #ifdef TIF_SYSCALL_EMU
2154 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2156 clear_tsk_latency_tracing(p
);
2158 /* ok, now we should be set up.. */
2159 p
->pid
= pid_nr(pid
);
2160 if (clone_flags
& CLONE_THREAD
) {
2161 p
->exit_signal
= -1;
2162 p
->group_leader
= current
->group_leader
;
2163 p
->tgid
= current
->tgid
;
2165 if (clone_flags
& CLONE_PARENT
)
2166 p
->exit_signal
= current
->group_leader
->exit_signal
;
2168 p
->exit_signal
= args
->exit_signal
;
2169 p
->group_leader
= p
;
2174 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2175 p
->dirty_paused_when
= 0;
2177 p
->pdeath_signal
= 0;
2178 INIT_LIST_HEAD(&p
->thread_group
);
2179 p
->task_works
= NULL
;
2182 * Ensure that the cgroup subsystem policies allow the new process to be
2183 * forked. It should be noted the the new process's css_set can be changed
2184 * between here and cgroup_post_fork() if an organisation operation is in
2187 retval
= cgroup_can_fork(p
, args
);
2189 goto bad_fork_put_pidfd
;
2192 * From this point on we must avoid any synchronous user-space
2193 * communication until we take the tasklist-lock. In particular, we do
2194 * not want user-space to be able to predict the process start-time by
2195 * stalling fork(2) after we recorded the start_time but before it is
2196 * visible to the system.
2199 p
->start_time
= ktime_get_ns();
2200 p
->start_boottime
= ktime_get_boottime_ns();
2203 * Make it visible to the rest of the system, but dont wake it up yet.
2204 * Need tasklist lock for parent etc handling!
2206 write_lock_irq(&tasklist_lock
);
2208 /* CLONE_PARENT re-uses the old parent */
2209 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2210 p
->real_parent
= current
->real_parent
;
2211 p
->parent_exec_id
= current
->parent_exec_id
;
2213 p
->real_parent
= current
;
2214 p
->parent_exec_id
= current
->self_exec_id
;
2217 klp_copy_process(p
);
2219 spin_lock(¤t
->sighand
->siglock
);
2222 * Copy seccomp details explicitly here, in case they were changed
2223 * before holding sighand lock.
2227 rseq_fork(p
, clone_flags
);
2229 /* Don't start children in a dying pid namespace */
2230 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2232 goto bad_fork_cancel_cgroup
;
2235 /* Let kill terminate clone/fork in the middle */
2236 if (fatal_signal_pending(current
)) {
2238 goto bad_fork_cancel_cgroup
;
2241 /* past the last point of failure */
2243 fd_install(pidfd
, pidfile
);
2245 init_task_pid_links(p
);
2246 if (likely(p
->pid
)) {
2247 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2249 init_task_pid(p
, PIDTYPE_PID
, pid
);
2250 if (thread_group_leader(p
)) {
2251 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2252 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2253 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2255 if (is_child_reaper(pid
)) {
2256 ns_of_pid(pid
)->child_reaper
= p
;
2257 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2259 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2260 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2262 * Inherit has_child_subreaper flag under the same
2263 * tasklist_lock with adding child to the process tree
2264 * for propagate_has_child_subreaper optimization.
2266 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2267 p
->real_parent
->signal
->is_child_subreaper
;
2268 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2269 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2270 attach_pid(p
, PIDTYPE_TGID
);
2271 attach_pid(p
, PIDTYPE_PGID
);
2272 attach_pid(p
, PIDTYPE_SID
);
2273 __this_cpu_inc(process_counts
);
2275 current
->signal
->nr_threads
++;
2276 atomic_inc(¤t
->signal
->live
);
2277 refcount_inc(¤t
->signal
->sigcnt
);
2278 task_join_group_stop(p
);
2279 list_add_tail_rcu(&p
->thread_group
,
2280 &p
->group_leader
->thread_group
);
2281 list_add_tail_rcu(&p
->thread_node
,
2282 &p
->signal
->thread_head
);
2284 attach_pid(p
, PIDTYPE_PID
);
2288 hlist_del_init(&delayed
.node
);
2289 spin_unlock(¤t
->sighand
->siglock
);
2290 syscall_tracepoint_update(p
);
2291 write_unlock_irq(&tasklist_lock
);
2293 proc_fork_connector(p
);
2294 cgroup_post_fork(p
, args
);
2297 trace_task_newtask(p
, clone_flags
);
2298 uprobe_copy_process(p
, clone_flags
);
2302 bad_fork_cancel_cgroup
:
2303 spin_unlock(¤t
->sighand
->siglock
);
2304 write_unlock_irq(&tasklist_lock
);
2305 cgroup_cancel_fork(p
, args
);
2307 if (clone_flags
& CLONE_PIDFD
) {
2309 put_unused_fd(pidfd
);
2312 if (pid
!= &init_struct_pid
)
2314 bad_fork_cleanup_thread
:
2316 bad_fork_cleanup_io
:
2319 bad_fork_cleanup_namespaces
:
2320 exit_task_namespaces(p
);
2321 bad_fork_cleanup_mm
:
2323 mm_clear_owner(p
->mm
, p
);
2326 bad_fork_cleanup_signal
:
2327 if (!(clone_flags
& CLONE_THREAD
))
2328 free_signal_struct(p
->signal
);
2329 bad_fork_cleanup_sighand
:
2330 __cleanup_sighand(p
->sighand
);
2331 bad_fork_cleanup_fs
:
2332 exit_fs(p
); /* blocking */
2333 bad_fork_cleanup_files
:
2334 exit_files(p
); /* blocking */
2335 bad_fork_cleanup_semundo
:
2337 bad_fork_cleanup_security
:
2338 security_task_free(p
);
2339 bad_fork_cleanup_audit
:
2341 bad_fork_cleanup_perf
:
2342 perf_event_free_task(p
);
2343 bad_fork_cleanup_policy
:
2344 lockdep_free_task(p
);
2346 mpol_put(p
->mempolicy
);
2347 bad_fork_cleanup_threadgroup_lock
:
2349 delayacct_tsk_free(p
);
2350 bad_fork_cleanup_count
:
2351 atomic_dec(&p
->cred
->user
->processes
);
2354 p
->state
= TASK_DEAD
;
2356 delayed_free_task(p
);
2358 spin_lock_irq(¤t
->sighand
->siglock
);
2359 hlist_del_init(&delayed
.node
);
2360 spin_unlock_irq(¤t
->sighand
->siglock
);
2361 return ERR_PTR(retval
);
2364 static inline void init_idle_pids(struct task_struct
*idle
)
2368 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2369 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2370 init_task_pid(idle
, type
, &init_struct_pid
);
2374 struct task_struct
*fork_idle(int cpu
)
2376 struct task_struct
*task
;
2377 struct kernel_clone_args args
= {
2381 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2382 if (!IS_ERR(task
)) {
2383 init_idle_pids(task
);
2384 init_idle(task
, cpu
);
2390 struct mm_struct
*copy_init_mm(void)
2392 return dup_mm(NULL
, &init_mm
);
2396 * Ok, this is the main fork-routine.
2398 * It copies the process, and if successful kick-starts
2399 * it and waits for it to finish using the VM if required.
2401 * args->exit_signal is expected to be checked for sanity by the caller.
2403 long _do_fork(struct kernel_clone_args
*args
)
2405 u64 clone_flags
= args
->flags
;
2406 struct completion vfork
;
2408 struct task_struct
*p
;
2413 * Determine whether and which event to report to ptracer. When
2414 * called from kernel_thread or CLONE_UNTRACED is explicitly
2415 * requested, no event is reported; otherwise, report if the event
2416 * for the type of forking is enabled.
2418 if (!(clone_flags
& CLONE_UNTRACED
)) {
2419 if (clone_flags
& CLONE_VFORK
)
2420 trace
= PTRACE_EVENT_VFORK
;
2421 else if (args
->exit_signal
!= SIGCHLD
)
2422 trace
= PTRACE_EVENT_CLONE
;
2424 trace
= PTRACE_EVENT_FORK
;
2426 if (likely(!ptrace_event_enabled(current
, trace
)))
2430 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2431 add_latent_entropy();
2437 * Do this prior waking up the new thread - the thread pointer
2438 * might get invalid after that point, if the thread exits quickly.
2440 trace_sched_process_fork(current
, p
);
2442 pid
= get_task_pid(p
, PIDTYPE_PID
);
2445 if (clone_flags
& CLONE_PARENT_SETTID
)
2446 put_user(nr
, args
->parent_tid
);
2448 if (clone_flags
& CLONE_VFORK
) {
2449 p
->vfork_done
= &vfork
;
2450 init_completion(&vfork
);
2454 wake_up_new_task(p
);
2456 /* forking complete and child started to run, tell ptracer */
2457 if (unlikely(trace
))
2458 ptrace_event_pid(trace
, pid
);
2460 if (clone_flags
& CLONE_VFORK
) {
2461 if (!wait_for_vfork_done(p
, &vfork
))
2462 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2469 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2471 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2472 if ((kargs
->flags
& CLONE_PIDFD
) &&
2473 (kargs
->flags
& CLONE_PARENT_SETTID
))
2479 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2480 /* For compatibility with architectures that call do_fork directly rather than
2481 * using the syscall entry points below. */
2482 long do_fork(unsigned long clone_flags
,
2483 unsigned long stack_start
,
2484 unsigned long stack_size
,
2485 int __user
*parent_tidptr
,
2486 int __user
*child_tidptr
)
2488 struct kernel_clone_args args
= {
2489 .flags
= (clone_flags
& ~CSIGNAL
),
2490 .pidfd
= parent_tidptr
,
2491 .child_tid
= child_tidptr
,
2492 .parent_tid
= parent_tidptr
,
2493 .exit_signal
= (clone_flags
& CSIGNAL
),
2494 .stack
= stack_start
,
2495 .stack_size
= stack_size
,
2498 if (!legacy_clone_args_valid(&args
))
2501 return _do_fork(&args
);
2506 * Create a kernel thread.
2508 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2510 struct kernel_clone_args args
= {
2511 .flags
= ((flags
| CLONE_VM
| CLONE_UNTRACED
) & ~CSIGNAL
),
2512 .exit_signal
= (flags
& CSIGNAL
),
2513 .stack
= (unsigned long)fn
,
2514 .stack_size
= (unsigned long)arg
,
2517 return _do_fork(&args
);
2520 #ifdef __ARCH_WANT_SYS_FORK
2521 SYSCALL_DEFINE0(fork
)
2524 struct kernel_clone_args args
= {
2525 .exit_signal
= SIGCHLD
,
2528 return _do_fork(&args
);
2530 /* can not support in nommu mode */
2536 #ifdef __ARCH_WANT_SYS_VFORK
2537 SYSCALL_DEFINE0(vfork
)
2539 struct kernel_clone_args args
= {
2540 .flags
= CLONE_VFORK
| CLONE_VM
,
2541 .exit_signal
= SIGCHLD
,
2544 return _do_fork(&args
);
2548 #ifdef __ARCH_WANT_SYS_CLONE
2549 #ifdef CONFIG_CLONE_BACKWARDS
2550 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2551 int __user
*, parent_tidptr
,
2553 int __user
*, child_tidptr
)
2554 #elif defined(CONFIG_CLONE_BACKWARDS2)
2555 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2556 int __user
*, parent_tidptr
,
2557 int __user
*, child_tidptr
,
2559 #elif defined(CONFIG_CLONE_BACKWARDS3)
2560 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2562 int __user
*, parent_tidptr
,
2563 int __user
*, child_tidptr
,
2566 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2567 int __user
*, parent_tidptr
,
2568 int __user
*, child_tidptr
,
2572 struct kernel_clone_args args
= {
2573 .flags
= (clone_flags
& ~CSIGNAL
),
2574 .pidfd
= parent_tidptr
,
2575 .child_tid
= child_tidptr
,
2576 .parent_tid
= parent_tidptr
,
2577 .exit_signal
= (clone_flags
& CSIGNAL
),
2582 if (!legacy_clone_args_valid(&args
))
2585 return _do_fork(&args
);
2589 #ifdef __ARCH_WANT_SYS_CLONE3
2592 * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2593 * the registers containing the syscall arguments for clone. This doesn't work
2594 * with clone3 since the TLS value is passed in clone_args instead.
2596 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2597 #error clone3 requires copy_thread_tls support in arch
2600 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2601 struct clone_args __user
*uargs
,
2605 struct clone_args args
;
2606 pid_t
*kset_tid
= kargs
->set_tid
;
2608 if (unlikely(usize
> PAGE_SIZE
))
2610 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2613 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2617 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2620 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2623 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2627 * Verify that higher 32bits of exit_signal are unset and that
2628 * it is a valid signal
2630 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2631 !valid_signal(args
.exit_signal
)))
2634 if ((args
.flags
& CLONE_INTO_CGROUP
) && args
.cgroup
< 0)
2637 *kargs
= (struct kernel_clone_args
){
2638 .flags
= args
.flags
,
2639 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2640 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2641 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2642 .exit_signal
= args
.exit_signal
,
2643 .stack
= args
.stack
,
2644 .stack_size
= args
.stack_size
,
2646 .set_tid_size
= args
.set_tid_size
,
2647 .cgroup
= args
.cgroup
,
2651 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
2652 (kargs
->set_tid_size
* sizeof(pid_t
))))
2655 kargs
->set_tid
= kset_tid
;
2661 * clone3_stack_valid - check and prepare stack
2662 * @kargs: kernel clone args
2664 * Verify that the stack arguments userspace gave us are sane.
2665 * In addition, set the stack direction for userspace since it's easy for us to
2668 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2670 if (kargs
->stack
== 0) {
2671 if (kargs
->stack_size
> 0)
2674 if (kargs
->stack_size
== 0)
2677 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2680 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2681 kargs
->stack
+= kargs
->stack_size
;
2688 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2690 /* Verify that no unknown flags are passed along. */
2692 ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
| CLONE_INTO_CGROUP
))
2696 * - make the CLONE_DETACHED bit reuseable for clone3
2697 * - make the CSIGNAL bits reuseable for clone3
2699 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2702 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
2703 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
2706 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2710 if (!clone3_stack_valid(kargs
))
2717 * clone3 - create a new process with specific properties
2718 * @uargs: argument structure
2719 * @size: size of @uargs
2721 * clone3() is the extensible successor to clone()/clone2().
2722 * It takes a struct as argument that is versioned by its size.
2724 * Return: On success, a positive PID for the child process.
2725 * On error, a negative errno number.
2727 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2731 struct kernel_clone_args kargs
;
2732 pid_t set_tid
[MAX_PID_NS_LEVEL
];
2734 kargs
.set_tid
= set_tid
;
2736 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2740 if (!clone3_args_valid(&kargs
))
2743 return _do_fork(&kargs
);
2747 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2749 struct task_struct
*leader
, *parent
, *child
;
2752 read_lock(&tasklist_lock
);
2753 leader
= top
= top
->group_leader
;
2755 for_each_thread(leader
, parent
) {
2756 list_for_each_entry(child
, &parent
->children
, sibling
) {
2757 res
= visitor(child
, data
);
2769 if (leader
!= top
) {
2771 parent
= child
->real_parent
;
2772 leader
= parent
->group_leader
;
2776 read_unlock(&tasklist_lock
);
2779 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2780 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2783 static void sighand_ctor(void *data
)
2785 struct sighand_struct
*sighand
= data
;
2787 spin_lock_init(&sighand
->siglock
);
2788 init_waitqueue_head(&sighand
->signalfd_wqh
);
2791 void __init
proc_caches_init(void)
2793 unsigned int mm_size
;
2795 sighand_cachep
= kmem_cache_create("sighand_cache",
2796 sizeof(struct sighand_struct
), 0,
2797 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2798 SLAB_ACCOUNT
, sighand_ctor
);
2799 signal_cachep
= kmem_cache_create("signal_cache",
2800 sizeof(struct signal_struct
), 0,
2801 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2803 files_cachep
= kmem_cache_create("files_cache",
2804 sizeof(struct files_struct
), 0,
2805 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2807 fs_cachep
= kmem_cache_create("fs_cache",
2808 sizeof(struct fs_struct
), 0,
2809 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2813 * The mm_cpumask is located at the end of mm_struct, and is
2814 * dynamically sized based on the maximum CPU number this system
2815 * can have, taking hotplug into account (nr_cpu_ids).
2817 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2819 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2820 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2821 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2822 offsetof(struct mm_struct
, saved_auxv
),
2823 sizeof_field(struct mm_struct
, saved_auxv
),
2825 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2827 nsproxy_cache_init();
2831 * Check constraints on flags passed to the unshare system call.
2833 static int check_unshare_flags(unsigned long unshare_flags
)
2835 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2836 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2837 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2838 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
|
2842 * Not implemented, but pretend it works if there is nothing
2843 * to unshare. Note that unsharing the address space or the
2844 * signal handlers also need to unshare the signal queues (aka
2847 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2848 if (!thread_group_empty(current
))
2851 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2852 if (refcount_read(¤t
->sighand
->count
) > 1)
2855 if (unshare_flags
& CLONE_VM
) {
2856 if (!current_is_single_threaded())
2864 * Unshare the filesystem structure if it is being shared
2866 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2868 struct fs_struct
*fs
= current
->fs
;
2870 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2873 /* don't need lock here; in the worst case we'll do useless copy */
2877 *new_fsp
= copy_fs_struct(fs
);
2885 * Unshare file descriptor table if it is being shared
2887 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2889 struct files_struct
*fd
= current
->files
;
2892 if ((unshare_flags
& CLONE_FILES
) &&
2893 (fd
&& atomic_read(&fd
->count
) > 1)) {
2894 *new_fdp
= dup_fd(fd
, &error
);
2903 * unshare allows a process to 'unshare' part of the process
2904 * context which was originally shared using clone. copy_*
2905 * functions used by do_fork() cannot be used here directly
2906 * because they modify an inactive task_struct that is being
2907 * constructed. Here we are modifying the current, active,
2910 int ksys_unshare(unsigned long unshare_flags
)
2912 struct fs_struct
*fs
, *new_fs
= NULL
;
2913 struct files_struct
*fd
, *new_fd
= NULL
;
2914 struct cred
*new_cred
= NULL
;
2915 struct nsproxy
*new_nsproxy
= NULL
;
2920 * If unsharing a user namespace must also unshare the thread group
2921 * and unshare the filesystem root and working directories.
2923 if (unshare_flags
& CLONE_NEWUSER
)
2924 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2926 * If unsharing vm, must also unshare signal handlers.
2928 if (unshare_flags
& CLONE_VM
)
2929 unshare_flags
|= CLONE_SIGHAND
;
2931 * If unsharing a signal handlers, must also unshare the signal queues.
2933 if (unshare_flags
& CLONE_SIGHAND
)
2934 unshare_flags
|= CLONE_THREAD
;
2936 * If unsharing namespace, must also unshare filesystem information.
2938 if (unshare_flags
& CLONE_NEWNS
)
2939 unshare_flags
|= CLONE_FS
;
2941 err
= check_unshare_flags(unshare_flags
);
2943 goto bad_unshare_out
;
2945 * CLONE_NEWIPC must also detach from the undolist: after switching
2946 * to a new ipc namespace, the semaphore arrays from the old
2947 * namespace are unreachable.
2949 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2951 err
= unshare_fs(unshare_flags
, &new_fs
);
2953 goto bad_unshare_out
;
2954 err
= unshare_fd(unshare_flags
, &new_fd
);
2956 goto bad_unshare_cleanup_fs
;
2957 err
= unshare_userns(unshare_flags
, &new_cred
);
2959 goto bad_unshare_cleanup_fd
;
2960 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2963 goto bad_unshare_cleanup_cred
;
2965 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2968 * CLONE_SYSVSEM is equivalent to sys_exit().
2972 if (unshare_flags
& CLONE_NEWIPC
) {
2973 /* Orphan segments in old ns (see sem above). */
2975 shm_init_task(current
);
2979 switch_task_namespaces(current
, new_nsproxy
);
2985 spin_lock(&fs
->lock
);
2986 current
->fs
= new_fs
;
2991 spin_unlock(&fs
->lock
);
2995 fd
= current
->files
;
2996 current
->files
= new_fd
;
3000 task_unlock(current
);
3003 /* Install the new user namespace */
3004 commit_creds(new_cred
);
3009 perf_event_namespaces(current
);
3011 bad_unshare_cleanup_cred
:
3014 bad_unshare_cleanup_fd
:
3016 put_files_struct(new_fd
);
3018 bad_unshare_cleanup_fs
:
3020 free_fs_struct(new_fs
);
3026 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3028 return ksys_unshare(unshare_flags
);
3032 * Helper to unshare the files of the current task.
3033 * We don't want to expose copy_files internals to
3034 * the exec layer of the kernel.
3037 int unshare_files(struct files_struct
**displaced
)
3039 struct task_struct
*task
= current
;
3040 struct files_struct
*copy
= NULL
;
3043 error
= unshare_fd(CLONE_FILES
, ©
);
3044 if (error
|| !copy
) {
3048 *displaced
= task
->files
;
3055 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3056 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
3060 int threads
= max_threads
;
3062 int max
= MAX_THREADS
;
3069 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3073 max_threads
= threads
;