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(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
);
368 void vm_area_free(struct vm_area_struct
*vma
)
370 kmem_cache_free(vm_area_cachep
, vma
);
373 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
375 void *stack
= task_stack_page(tsk
);
376 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
378 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
383 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
385 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
386 mod_zone_page_state(page_zone(vm
->pages
[i
]),
388 PAGE_SIZE
/ 1024 * account
);
392 * All stack pages are in the same zone and belong to the
395 struct page
*first_page
= virt_to_page(stack
);
397 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
398 THREAD_SIZE
/ 1024 * account
);
400 mod_memcg_page_state(first_page
, MEMCG_KERNEL_STACK_KB
,
401 account
* (THREAD_SIZE
/ 1024));
405 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
407 #ifdef CONFIG_VMAP_STACK
408 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
414 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
416 * If memcg_kmem_charge() fails, page->mem_cgroup
417 * pointer is NULL, and both memcg_kmem_uncharge()
418 * and mod_memcg_page_state() in free_thread_stack()
419 * will ignore this page. So it's safe.
421 ret
= memcg_kmem_charge(vm
->pages
[i
], GFP_KERNEL
, 0);
425 mod_memcg_page_state(vm
->pages
[i
],
426 MEMCG_KERNEL_STACK_KB
,
434 static void release_task_stack(struct task_struct
*tsk
)
436 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
437 return; /* Better to leak the stack than to free prematurely */
439 account_kernel_stack(tsk
, -1);
440 free_thread_stack(tsk
);
442 #ifdef CONFIG_VMAP_STACK
443 tsk
->stack_vm_area
= NULL
;
447 #ifdef CONFIG_THREAD_INFO_IN_TASK
448 void put_task_stack(struct task_struct
*tsk
)
450 if (refcount_dec_and_test(&tsk
->stack_refcount
))
451 release_task_stack(tsk
);
455 void free_task(struct task_struct
*tsk
)
457 #ifndef CONFIG_THREAD_INFO_IN_TASK
459 * The task is finally done with both the stack and thread_info,
462 release_task_stack(tsk
);
465 * If the task had a separate stack allocation, it should be gone
468 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
470 rt_mutex_debug_task_free(tsk
);
471 ftrace_graph_exit_task(tsk
);
472 put_seccomp_filter(tsk
);
473 arch_release_task_struct(tsk
);
474 if (tsk
->flags
& PF_KTHREAD
)
475 free_kthread_struct(tsk
);
476 free_task_struct(tsk
);
478 EXPORT_SYMBOL(free_task
);
481 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
482 struct mm_struct
*oldmm
)
484 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
485 struct rb_node
**rb_link
, *rb_parent
;
487 unsigned long charge
;
490 uprobe_start_dup_mmap();
491 if (down_write_killable(&oldmm
->mmap_sem
)) {
493 goto fail_uprobe_end
;
495 flush_cache_dup_mm(oldmm
);
496 uprobe_dup_mmap(oldmm
, mm
);
498 * Not linked in yet - no deadlock potential:
500 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
502 /* No ordering required: file already has been exposed. */
503 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
505 mm
->total_vm
= oldmm
->total_vm
;
506 mm
->data_vm
= oldmm
->data_vm
;
507 mm
->exec_vm
= oldmm
->exec_vm
;
508 mm
->stack_vm
= oldmm
->stack_vm
;
510 rb_link
= &mm
->mm_rb
.rb_node
;
513 retval
= ksm_fork(mm
, oldmm
);
516 retval
= khugepaged_fork(mm
, oldmm
);
521 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
524 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
525 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
530 * Don't duplicate many vmas if we've been oom-killed (for
533 if (fatal_signal_pending(current
)) {
537 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
538 unsigned long len
= vma_pages(mpnt
);
540 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
544 tmp
= vm_area_dup(mpnt
);
547 retval
= vma_dup_policy(mpnt
, tmp
);
549 goto fail_nomem_policy
;
551 retval
= dup_userfaultfd(tmp
, &uf
);
553 goto fail_nomem_anon_vma_fork
;
554 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
555 /* VM_WIPEONFORK gets a clean slate in the child. */
556 tmp
->anon_vma
= NULL
;
557 if (anon_vma_prepare(tmp
))
558 goto fail_nomem_anon_vma_fork
;
559 } else if (anon_vma_fork(tmp
, mpnt
))
560 goto fail_nomem_anon_vma_fork
;
561 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
562 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
565 struct inode
*inode
= file_inode(file
);
566 struct address_space
*mapping
= file
->f_mapping
;
569 if (tmp
->vm_flags
& VM_DENYWRITE
)
570 atomic_dec(&inode
->i_writecount
);
571 i_mmap_lock_write(mapping
);
572 if (tmp
->vm_flags
& VM_SHARED
)
573 atomic_inc(&mapping
->i_mmap_writable
);
574 flush_dcache_mmap_lock(mapping
);
575 /* insert tmp into the share list, just after mpnt */
576 vma_interval_tree_insert_after(tmp
, mpnt
,
578 flush_dcache_mmap_unlock(mapping
);
579 i_mmap_unlock_write(mapping
);
583 * Clear hugetlb-related page reserves for children. This only
584 * affects MAP_PRIVATE mappings. Faults generated by the child
585 * are not guaranteed to succeed, even if read-only
587 if (is_vm_hugetlb_page(tmp
))
588 reset_vma_resv_huge_pages(tmp
);
591 * Link in the new vma and copy the page table entries.
594 pprev
= &tmp
->vm_next
;
598 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
599 rb_link
= &tmp
->vm_rb
.rb_right
;
600 rb_parent
= &tmp
->vm_rb
;
603 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
604 retval
= copy_page_range(mm
, oldmm
, mpnt
);
606 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
607 tmp
->vm_ops
->open(tmp
);
612 /* a new mm has just been created */
613 retval
= arch_dup_mmap(oldmm
, mm
);
615 up_write(&mm
->mmap_sem
);
617 up_write(&oldmm
->mmap_sem
);
618 dup_userfaultfd_complete(&uf
);
620 uprobe_end_dup_mmap();
622 fail_nomem_anon_vma_fork
:
623 mpol_put(vma_policy(tmp
));
628 vm_unacct_memory(charge
);
632 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
634 mm
->pgd
= pgd_alloc(mm
);
635 if (unlikely(!mm
->pgd
))
640 static inline void mm_free_pgd(struct mm_struct
*mm
)
642 pgd_free(mm
, mm
->pgd
);
645 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
647 down_write(&oldmm
->mmap_sem
);
648 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
649 up_write(&oldmm
->mmap_sem
);
652 #define mm_alloc_pgd(mm) (0)
653 #define mm_free_pgd(mm)
654 #endif /* CONFIG_MMU */
656 static void check_mm(struct mm_struct
*mm
)
660 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
661 "Please make sure 'struct resident_page_types[]' is updated as well");
663 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
664 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
667 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
668 mm
, resident_page_types
[i
], x
);
671 if (mm_pgtables_bytes(mm
))
672 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
673 mm_pgtables_bytes(mm
));
675 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
676 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
680 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
681 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
684 * Called when the last reference to the mm
685 * is dropped: either by a lazy thread or by
686 * mmput. Free the page directory and the mm.
688 void __mmdrop(struct mm_struct
*mm
)
690 BUG_ON(mm
== &init_mm
);
691 WARN_ON_ONCE(mm
== current
->mm
);
692 WARN_ON_ONCE(mm
== current
->active_mm
);
695 mmu_notifier_mm_destroy(mm
);
697 put_user_ns(mm
->user_ns
);
700 EXPORT_SYMBOL_GPL(__mmdrop
);
702 static void mmdrop_async_fn(struct work_struct
*work
)
704 struct mm_struct
*mm
;
706 mm
= container_of(work
, struct mm_struct
, async_put_work
);
710 static void mmdrop_async(struct mm_struct
*mm
)
712 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
713 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
714 schedule_work(&mm
->async_put_work
);
718 static inline void free_signal_struct(struct signal_struct
*sig
)
720 taskstats_tgid_free(sig
);
721 sched_autogroup_exit(sig
);
723 * __mmdrop is not safe to call from softirq context on x86 due to
724 * pgd_dtor so postpone it to the async context
727 mmdrop_async(sig
->oom_mm
);
728 kmem_cache_free(signal_cachep
, sig
);
731 static inline void put_signal_struct(struct signal_struct
*sig
)
733 if (refcount_dec_and_test(&sig
->sigcnt
))
734 free_signal_struct(sig
);
737 void __put_task_struct(struct task_struct
*tsk
)
739 WARN_ON(!tsk
->exit_state
);
740 WARN_ON(refcount_read(&tsk
->usage
));
741 WARN_ON(tsk
== current
);
744 task_numa_free(tsk
, true);
745 security_task_free(tsk
);
747 delayacct_tsk_free(tsk
);
748 put_signal_struct(tsk
->signal
);
750 if (!profile_handoff_task(tsk
))
753 EXPORT_SYMBOL_GPL(__put_task_struct
);
755 void __init __weak
arch_task_cache_init(void) { }
760 static void set_max_threads(unsigned int max_threads_suggested
)
763 unsigned long nr_pages
= totalram_pages();
766 * The number of threads shall be limited such that the thread
767 * structures may only consume a small part of the available memory.
769 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
770 threads
= MAX_THREADS
;
772 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
773 (u64
) THREAD_SIZE
* 8UL);
775 if (threads
> max_threads_suggested
)
776 threads
= max_threads_suggested
;
778 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
781 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
782 /* Initialized by the architecture: */
783 int arch_task_struct_size __read_mostly
;
786 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
787 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
789 /* Fetch thread_struct whitelist for the architecture. */
790 arch_thread_struct_whitelist(offset
, size
);
793 * Handle zero-sized whitelist or empty thread_struct, otherwise
794 * adjust offset to position of thread_struct in task_struct.
796 if (unlikely(*size
== 0))
799 *offset
+= offsetof(struct task_struct
, thread
);
801 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
803 void __init
fork_init(void)
806 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
807 #ifndef ARCH_MIN_TASKALIGN
808 #define ARCH_MIN_TASKALIGN 0
810 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
811 unsigned long useroffset
, usersize
;
813 /* create a slab on which task_structs can be allocated */
814 task_struct_whitelist(&useroffset
, &usersize
);
815 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
816 arch_task_struct_size
, align
,
817 SLAB_PANIC
|SLAB_ACCOUNT
,
818 useroffset
, usersize
, NULL
);
821 /* do the arch specific task caches init */
822 arch_task_cache_init();
824 set_max_threads(MAX_THREADS
);
826 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
827 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
828 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
829 init_task
.signal
->rlim
[RLIMIT_NPROC
];
831 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
832 init_user_ns
.ucount_max
[i
] = max_threads
/2;
835 #ifdef CONFIG_VMAP_STACK
836 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
837 NULL
, free_vm_stack_cache
);
840 lockdep_init_task(&init_task
);
844 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
845 struct task_struct
*src
)
851 void set_task_stack_end_magic(struct task_struct
*tsk
)
853 unsigned long *stackend
;
855 stackend
= end_of_stack(tsk
);
856 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
859 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
861 struct task_struct
*tsk
;
862 unsigned long *stack
;
863 struct vm_struct
*stack_vm_area __maybe_unused
;
866 if (node
== NUMA_NO_NODE
)
867 node
= tsk_fork_get_node(orig
);
868 tsk
= alloc_task_struct_node(node
);
872 stack
= alloc_thread_stack_node(tsk
, node
);
876 if (memcg_charge_kernel_stack(tsk
))
879 stack_vm_area
= task_stack_vm_area(tsk
);
881 err
= arch_dup_task_struct(tsk
, orig
);
884 * arch_dup_task_struct() clobbers the stack-related fields. Make
885 * sure they're properly initialized before using any stack-related
889 #ifdef CONFIG_VMAP_STACK
890 tsk
->stack_vm_area
= stack_vm_area
;
892 #ifdef CONFIG_THREAD_INFO_IN_TASK
893 refcount_set(&tsk
->stack_refcount
, 1);
899 #ifdef CONFIG_SECCOMP
901 * We must handle setting up seccomp filters once we're under
902 * the sighand lock in case orig has changed between now and
903 * then. Until then, filter must be NULL to avoid messing up
904 * the usage counts on the error path calling free_task.
906 tsk
->seccomp
.filter
= NULL
;
909 setup_thread_stack(tsk
, orig
);
910 clear_user_return_notifier(tsk
);
911 clear_tsk_need_resched(tsk
);
912 set_task_stack_end_magic(tsk
);
914 #ifdef CONFIG_STACKPROTECTOR
915 tsk
->stack_canary
= get_random_canary();
917 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
918 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
921 * One for the user space visible state that goes away when reaped.
922 * One for the scheduler.
924 refcount_set(&tsk
->rcu_users
, 2);
925 /* One for the rcu users */
926 refcount_set(&tsk
->usage
, 1);
927 #ifdef CONFIG_BLK_DEV_IO_TRACE
930 tsk
->splice_pipe
= NULL
;
931 tsk
->task_frag
.page
= NULL
;
932 tsk
->wake_q
.next
= NULL
;
934 account_kernel_stack(tsk
, 1);
938 #ifdef CONFIG_FAULT_INJECTION
942 #ifdef CONFIG_BLK_CGROUP
943 tsk
->throttle_queue
= NULL
;
944 tsk
->use_memdelay
= 0;
948 tsk
->active_memcg
= NULL
;
953 free_thread_stack(tsk
);
955 free_task_struct(tsk
);
959 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
961 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
963 static int __init
coredump_filter_setup(char *s
)
965 default_dump_filter
=
966 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
967 MMF_DUMP_FILTER_MASK
;
971 __setup("coredump_filter=", coredump_filter_setup
);
973 #include <linux/init_task.h>
975 static void mm_init_aio(struct mm_struct
*mm
)
978 spin_lock_init(&mm
->ioctx_lock
);
979 mm
->ioctx_table
= NULL
;
983 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
984 struct task_struct
*p
)
988 WRITE_ONCE(mm
->owner
, NULL
);
992 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
999 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1001 #ifdef CONFIG_UPROBES
1002 mm
->uprobes_state
.xol_area
= NULL
;
1006 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1007 struct user_namespace
*user_ns
)
1010 mm
->mm_rb
= RB_ROOT
;
1011 mm
->vmacache_seqnum
= 0;
1012 atomic_set(&mm
->mm_users
, 1);
1013 atomic_set(&mm
->mm_count
, 1);
1014 init_rwsem(&mm
->mmap_sem
);
1015 INIT_LIST_HEAD(&mm
->mmlist
);
1016 mm
->core_state
= NULL
;
1017 mm_pgtables_bytes_init(mm
);
1020 atomic64_set(&mm
->pinned_vm
, 0);
1021 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1022 spin_lock_init(&mm
->page_table_lock
);
1023 spin_lock_init(&mm
->arg_lock
);
1024 mm_init_cpumask(mm
);
1026 mm_init_owner(mm
, p
);
1027 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1028 mmu_notifier_mm_init(mm
);
1029 init_tlb_flush_pending(mm
);
1030 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1031 mm
->pmd_huge_pte
= NULL
;
1033 mm_init_uprobes_state(mm
);
1036 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1037 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1039 mm
->flags
= default_dump_filter
;
1043 if (mm_alloc_pgd(mm
))
1046 if (init_new_context(p
, mm
))
1047 goto fail_nocontext
;
1049 mm
->user_ns
= get_user_ns(user_ns
);
1060 * Allocate and initialize an mm_struct.
1062 struct mm_struct
*mm_alloc(void)
1064 struct mm_struct
*mm
;
1070 memset(mm
, 0, sizeof(*mm
));
1071 return mm_init(mm
, current
, current_user_ns());
1074 static inline void __mmput(struct mm_struct
*mm
)
1076 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1078 uprobe_clear_state(mm
);
1081 khugepaged_exit(mm
); /* must run before exit_mmap */
1083 mm_put_huge_zero_page(mm
);
1084 set_mm_exe_file(mm
, NULL
);
1085 if (!list_empty(&mm
->mmlist
)) {
1086 spin_lock(&mmlist_lock
);
1087 list_del(&mm
->mmlist
);
1088 spin_unlock(&mmlist_lock
);
1091 module_put(mm
->binfmt
->module
);
1096 * Decrement the use count and release all resources for an mm.
1098 void mmput(struct mm_struct
*mm
)
1102 if (atomic_dec_and_test(&mm
->mm_users
))
1105 EXPORT_SYMBOL_GPL(mmput
);
1108 static void mmput_async_fn(struct work_struct
*work
)
1110 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1116 void mmput_async(struct mm_struct
*mm
)
1118 if (atomic_dec_and_test(&mm
->mm_users
)) {
1119 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1120 schedule_work(&mm
->async_put_work
);
1126 * set_mm_exe_file - change a reference to the mm's executable file
1128 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1130 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1131 * invocations: in mmput() nobody alive left, in execve task is single
1132 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1133 * mm->exe_file, but does so without using set_mm_exe_file() in order
1134 * to do avoid the need for any locks.
1136 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1138 struct file
*old_exe_file
;
1141 * It is safe to dereference the exe_file without RCU as
1142 * this function is only called if nobody else can access
1143 * this mm -- see comment above for justification.
1145 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1148 get_file(new_exe_file
);
1149 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1155 * get_mm_exe_file - acquire a reference to the mm's executable file
1157 * Returns %NULL if mm has no associated executable file.
1158 * User must release file via fput().
1160 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1162 struct file
*exe_file
;
1165 exe_file
= rcu_dereference(mm
->exe_file
);
1166 if (exe_file
&& !get_file_rcu(exe_file
))
1171 EXPORT_SYMBOL(get_mm_exe_file
);
1174 * get_task_exe_file - acquire a reference to the task's executable file
1176 * Returns %NULL if task's mm (if any) has no associated executable file or
1177 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1178 * User must release file via fput().
1180 struct file
*get_task_exe_file(struct task_struct
*task
)
1182 struct file
*exe_file
= NULL
;
1183 struct mm_struct
*mm
;
1188 if (!(task
->flags
& PF_KTHREAD
))
1189 exe_file
= get_mm_exe_file(mm
);
1194 EXPORT_SYMBOL(get_task_exe_file
);
1197 * get_task_mm - acquire a reference to the task's mm
1199 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1200 * this kernel workthread has transiently adopted a user mm with use_mm,
1201 * to do its AIO) is not set and if so returns a reference to it, after
1202 * bumping up the use count. User must release the mm via mmput()
1203 * after use. Typically used by /proc and ptrace.
1205 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1207 struct mm_struct
*mm
;
1212 if (task
->flags
& PF_KTHREAD
)
1220 EXPORT_SYMBOL_GPL(get_task_mm
);
1222 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1224 struct mm_struct
*mm
;
1227 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1229 return ERR_PTR(err
);
1231 mm
= get_task_mm(task
);
1232 if (mm
&& mm
!= current
->mm
&&
1233 !ptrace_may_access(task
, mode
)) {
1235 mm
= ERR_PTR(-EACCES
);
1237 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1242 static void complete_vfork_done(struct task_struct
*tsk
)
1244 struct completion
*vfork
;
1247 vfork
= tsk
->vfork_done
;
1248 if (likely(vfork
)) {
1249 tsk
->vfork_done
= NULL
;
1255 static int wait_for_vfork_done(struct task_struct
*child
,
1256 struct completion
*vfork
)
1260 freezer_do_not_count();
1261 cgroup_enter_frozen();
1262 killed
= wait_for_completion_killable(vfork
);
1263 cgroup_leave_frozen(false);
1268 child
->vfork_done
= NULL
;
1272 put_task_struct(child
);
1276 /* Please note the differences between mmput and mm_release.
1277 * mmput is called whenever we stop holding onto a mm_struct,
1278 * error success whatever.
1280 * mm_release is called after a mm_struct has been removed
1281 * from the current process.
1283 * This difference is important for error handling, when we
1284 * only half set up a mm_struct for a new process and need to restore
1285 * the old one. Because we mmput the new mm_struct before
1286 * restoring the old one. . .
1287 * Eric Biederman 10 January 1998
1289 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1291 uprobe_free_utask(tsk
);
1293 /* Get rid of any cached register state */
1294 deactivate_mm(tsk
, mm
);
1297 * Signal userspace if we're not exiting with a core dump
1298 * because we want to leave the value intact for debugging
1301 if (tsk
->clear_child_tid
) {
1302 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1303 atomic_read(&mm
->mm_users
) > 1) {
1305 * We don't check the error code - if userspace has
1306 * not set up a proper pointer then tough luck.
1308 put_user(0, tsk
->clear_child_tid
);
1309 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1310 1, NULL
, NULL
, 0, 0);
1312 tsk
->clear_child_tid
= NULL
;
1316 * All done, finally we can wake up parent and return this mm to him.
1317 * Also kthread_stop() uses this completion for synchronization.
1319 if (tsk
->vfork_done
)
1320 complete_vfork_done(tsk
);
1323 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1325 futex_exit_release(tsk
);
1326 mm_release(tsk
, mm
);
1329 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1331 futex_exec_release(tsk
);
1332 mm_release(tsk
, mm
);
1336 * dup_mm() - duplicates an existing mm structure
1337 * @tsk: the task_struct with which the new mm will be associated.
1338 * @oldmm: the mm to duplicate.
1340 * Allocates a new mm structure and duplicates the provided @oldmm structure
1343 * Return: the duplicated mm or NULL on failure.
1345 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1346 struct mm_struct
*oldmm
)
1348 struct mm_struct
*mm
;
1355 memcpy(mm
, oldmm
, sizeof(*mm
));
1357 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1360 err
= dup_mmap(mm
, oldmm
);
1364 mm
->hiwater_rss
= get_mm_rss(mm
);
1365 mm
->hiwater_vm
= mm
->total_vm
;
1367 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1373 /* don't put binfmt in mmput, we haven't got module yet */
1375 mm_init_owner(mm
, NULL
);
1382 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1384 struct mm_struct
*mm
, *oldmm
;
1387 tsk
->min_flt
= tsk
->maj_flt
= 0;
1388 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1389 #ifdef CONFIG_DETECT_HUNG_TASK
1390 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1391 tsk
->last_switch_time
= 0;
1395 tsk
->active_mm
= NULL
;
1398 * Are we cloning a kernel thread?
1400 * We need to steal a active VM for that..
1402 oldmm
= current
->mm
;
1406 /* initialize the new vmacache entries */
1407 vmacache_flush(tsk
);
1409 if (clone_flags
& CLONE_VM
) {
1416 mm
= dup_mm(tsk
, current
->mm
);
1422 tsk
->active_mm
= mm
;
1429 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1431 struct fs_struct
*fs
= current
->fs
;
1432 if (clone_flags
& CLONE_FS
) {
1433 /* tsk->fs is already what we want */
1434 spin_lock(&fs
->lock
);
1436 spin_unlock(&fs
->lock
);
1440 spin_unlock(&fs
->lock
);
1443 tsk
->fs
= copy_fs_struct(fs
);
1449 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1451 struct files_struct
*oldf
, *newf
;
1455 * A background process may not have any files ...
1457 oldf
= current
->files
;
1461 if (clone_flags
& CLONE_FILES
) {
1462 atomic_inc(&oldf
->count
);
1466 newf
= dup_fd(oldf
, &error
);
1476 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1479 struct io_context
*ioc
= current
->io_context
;
1480 struct io_context
*new_ioc
;
1485 * Share io context with parent, if CLONE_IO is set
1487 if (clone_flags
& CLONE_IO
) {
1489 tsk
->io_context
= ioc
;
1490 } else if (ioprio_valid(ioc
->ioprio
)) {
1491 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1492 if (unlikely(!new_ioc
))
1495 new_ioc
->ioprio
= ioc
->ioprio
;
1496 put_io_context(new_ioc
);
1502 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1504 struct sighand_struct
*sig
;
1506 if (clone_flags
& CLONE_SIGHAND
) {
1507 refcount_inc(¤t
->sighand
->count
);
1510 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1511 rcu_assign_pointer(tsk
->sighand
, sig
);
1515 refcount_set(&sig
->count
, 1);
1516 spin_lock_irq(¤t
->sighand
->siglock
);
1517 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1518 spin_unlock_irq(¤t
->sighand
->siglock
);
1520 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1521 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1522 flush_signal_handlers(tsk
, 0);
1527 void __cleanup_sighand(struct sighand_struct
*sighand
)
1529 if (refcount_dec_and_test(&sighand
->count
)) {
1530 signalfd_cleanup(sighand
);
1532 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1533 * without an RCU grace period, see __lock_task_sighand().
1535 kmem_cache_free(sighand_cachep
, sighand
);
1540 * Initialize POSIX timer handling for a thread group.
1542 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1544 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1545 unsigned long cpu_limit
;
1547 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1548 posix_cputimers_group_init(pct
, cpu_limit
);
1551 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1553 struct signal_struct
*sig
;
1555 if (clone_flags
& CLONE_THREAD
)
1558 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1563 sig
->nr_threads
= 1;
1564 atomic_set(&sig
->live
, 1);
1565 refcount_set(&sig
->sigcnt
, 1);
1567 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1568 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1569 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1571 init_waitqueue_head(&sig
->wait_chldexit
);
1572 sig
->curr_target
= tsk
;
1573 init_sigpending(&sig
->shared_pending
);
1574 INIT_HLIST_HEAD(&sig
->multiprocess
);
1575 seqlock_init(&sig
->stats_lock
);
1576 prev_cputime_init(&sig
->prev_cputime
);
1578 #ifdef CONFIG_POSIX_TIMERS
1579 INIT_LIST_HEAD(&sig
->posix_timers
);
1580 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1581 sig
->real_timer
.function
= it_real_fn
;
1584 task_lock(current
->group_leader
);
1585 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1586 task_unlock(current
->group_leader
);
1588 posix_cpu_timers_init_group(sig
);
1590 tty_audit_fork(sig
);
1591 sched_autogroup_fork(sig
);
1593 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1594 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1596 mutex_init(&sig
->cred_guard_mutex
);
1601 static void copy_seccomp(struct task_struct
*p
)
1603 #ifdef CONFIG_SECCOMP
1605 * Must be called with sighand->lock held, which is common to
1606 * all threads in the group. Holding cred_guard_mutex is not
1607 * needed because this new task is not yet running and cannot
1610 assert_spin_locked(¤t
->sighand
->siglock
);
1612 /* Ref-count the new filter user, and assign it. */
1613 get_seccomp_filter(current
);
1614 p
->seccomp
= current
->seccomp
;
1617 * Explicitly enable no_new_privs here in case it got set
1618 * between the task_struct being duplicated and holding the
1619 * sighand lock. The seccomp state and nnp must be in sync.
1621 if (task_no_new_privs(current
))
1622 task_set_no_new_privs(p
);
1625 * If the parent gained a seccomp mode after copying thread
1626 * flags and between before we held the sighand lock, we have
1627 * to manually enable the seccomp thread flag here.
1629 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1630 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1634 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1636 current
->clear_child_tid
= tidptr
;
1638 return task_pid_vnr(current
);
1641 static void rt_mutex_init_task(struct task_struct
*p
)
1643 raw_spin_lock_init(&p
->pi_lock
);
1644 #ifdef CONFIG_RT_MUTEXES
1645 p
->pi_waiters
= RB_ROOT_CACHED
;
1646 p
->pi_top_task
= NULL
;
1647 p
->pi_blocked_on
= NULL
;
1651 static inline void init_task_pid_links(struct task_struct
*task
)
1655 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1656 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1661 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1663 if (type
== PIDTYPE_PID
)
1664 task
->thread_pid
= pid
;
1666 task
->signal
->pids
[type
] = pid
;
1669 static inline void rcu_copy_process(struct task_struct
*p
)
1671 #ifdef CONFIG_PREEMPT_RCU
1672 p
->rcu_read_lock_nesting
= 0;
1673 p
->rcu_read_unlock_special
.s
= 0;
1674 p
->rcu_blocked_node
= NULL
;
1675 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1676 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1677 #ifdef CONFIG_TASKS_RCU
1678 p
->rcu_tasks_holdout
= false;
1679 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1680 p
->rcu_tasks_idle_cpu
= -1;
1681 #endif /* #ifdef CONFIG_TASKS_RCU */
1684 struct pid
*pidfd_pid(const struct file
*file
)
1686 if (file
->f_op
== &pidfd_fops
)
1687 return file
->private_data
;
1689 return ERR_PTR(-EBADF
);
1692 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1694 struct pid
*pid
= file
->private_data
;
1696 file
->private_data
= NULL
;
1701 #ifdef CONFIG_PROC_FS
1703 * pidfd_show_fdinfo - print information about a pidfd
1704 * @m: proc fdinfo file
1705 * @f: file referencing a pidfd
1708 * This function will print the pid that a given pidfd refers to in the
1709 * pid namespace of the procfs instance.
1710 * If the pid namespace of the process is not a descendant of the pid
1711 * namespace of the procfs instance 0 will be shown as its pid. This is
1712 * similar to calling getppid() on a process whose parent is outside of
1713 * its pid namespace.
1716 * If pid namespaces are supported then this function will also print
1717 * the pid of a given pidfd refers to for all descendant pid namespaces
1718 * starting from the current pid namespace of the instance, i.e. the
1719 * Pid field and the first entry in the NSpid field will be identical.
1720 * If the pid namespace of the process is not a descendant of the pid
1721 * namespace of the procfs instance 0 will be shown as its first NSpid
1722 * entry and no others will be shown.
1723 * Note that this differs from the Pid and NSpid fields in
1724 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1725 * the pid namespace of the procfs instance. The difference becomes
1726 * obvious when sending around a pidfd between pid namespaces from a
1727 * different branch of the tree, i.e. where no ancestoral relation is
1728 * present between the pid namespaces:
1729 * - create two new pid namespaces ns1 and ns2 in the initial pid
1730 * namespace (also take care to create new mount namespaces in the
1731 * new pid namespace and mount procfs)
1732 * - create a process with a pidfd in ns1
1733 * - send pidfd from ns1 to ns2
1734 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1735 * have exactly one entry, which is 0
1737 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1739 struct pid
*pid
= f
->private_data
;
1740 struct pid_namespace
*ns
;
1743 if (likely(pid_has_task(pid
, PIDTYPE_PID
))) {
1744 ns
= proc_pid_ns(file_inode(m
->file
));
1745 nr
= pid_nr_ns(pid
, ns
);
1748 seq_put_decimal_ll(m
, "Pid:\t", nr
);
1750 #ifdef CONFIG_PID_NS
1751 seq_put_decimal_ll(m
, "\nNSpid:\t", nr
);
1755 /* If nr is non-zero it means that 'pid' is valid and that
1756 * ns, i.e. the pid namespace associated with the procfs
1757 * instance, is in the pid namespace hierarchy of pid.
1758 * Start at one below the already printed level.
1760 for (i
= ns
->level
+ 1; i
<= pid
->level
; i
++)
1761 seq_put_decimal_ll(m
, "\t", pid
->numbers
[i
].nr
);
1769 * Poll support for process exit notification.
1771 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1773 struct task_struct
*task
;
1774 struct pid
*pid
= file
->private_data
;
1775 __poll_t poll_flags
= 0;
1777 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1780 task
= pid_task(pid
, PIDTYPE_PID
);
1782 * Inform pollers only when the whole thread group exits.
1783 * If the thread group leader exits before all other threads in the
1784 * group, then poll(2) should block, similar to the wait(2) family.
1786 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1787 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1793 const struct file_operations pidfd_fops
= {
1794 .release
= pidfd_release
,
1796 #ifdef CONFIG_PROC_FS
1797 .show_fdinfo
= pidfd_show_fdinfo
,
1801 static void __delayed_free_task(struct rcu_head
*rhp
)
1803 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1808 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1810 if (IS_ENABLED(CONFIG_MEMCG
))
1811 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1817 * This creates a new process as a copy of the old one,
1818 * but does not actually start it yet.
1820 * It copies the registers, and all the appropriate
1821 * parts of the process environment (as per the clone
1822 * flags). The actual kick-off is left to the caller.
1824 static __latent_entropy
struct task_struct
*copy_process(
1828 struct kernel_clone_args
*args
)
1830 int pidfd
= -1, retval
;
1831 struct task_struct
*p
;
1832 struct multiprocess_signals delayed
;
1833 struct file
*pidfile
= NULL
;
1834 u64 clone_flags
= args
->flags
;
1837 * Don't allow sharing the root directory with processes in a different
1840 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1841 return ERR_PTR(-EINVAL
);
1843 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1844 return ERR_PTR(-EINVAL
);
1847 * Thread groups must share signals as well, and detached threads
1848 * can only be started up within the thread group.
1850 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1851 return ERR_PTR(-EINVAL
);
1854 * Shared signal handlers imply shared VM. By way of the above,
1855 * thread groups also imply shared VM. Blocking this case allows
1856 * for various simplifications in other code.
1858 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1859 return ERR_PTR(-EINVAL
);
1862 * Siblings of global init remain as zombies on exit since they are
1863 * not reaped by their parent (swapper). To solve this and to avoid
1864 * multi-rooted process trees, prevent global and container-inits
1865 * from creating siblings.
1867 if ((clone_flags
& CLONE_PARENT
) &&
1868 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1869 return ERR_PTR(-EINVAL
);
1872 * If the new process will be in a different pid or user namespace
1873 * do not allow it to share a thread group with the forking task.
1875 if (clone_flags
& CLONE_THREAD
) {
1876 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1877 (task_active_pid_ns(current
) !=
1878 current
->nsproxy
->pid_ns_for_children
))
1879 return ERR_PTR(-EINVAL
);
1882 if (clone_flags
& CLONE_PIDFD
) {
1884 * - CLONE_DETACHED is blocked so that we can potentially
1885 * reuse it later for CLONE_PIDFD.
1886 * - CLONE_THREAD is blocked until someone really needs it.
1888 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1889 return ERR_PTR(-EINVAL
);
1893 * Force any signals received before this point to be delivered
1894 * before the fork happens. Collect up signals sent to multiple
1895 * processes that happen during the fork and delay them so that
1896 * they appear to happen after the fork.
1898 sigemptyset(&delayed
.signal
);
1899 INIT_HLIST_NODE(&delayed
.node
);
1901 spin_lock_irq(¤t
->sighand
->siglock
);
1902 if (!(clone_flags
& CLONE_THREAD
))
1903 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1904 recalc_sigpending();
1905 spin_unlock_irq(¤t
->sighand
->siglock
);
1906 retval
= -ERESTARTNOINTR
;
1907 if (signal_pending(current
))
1911 p
= dup_task_struct(current
, node
);
1916 * This _must_ happen before we call free_task(), i.e. before we jump
1917 * to any of the bad_fork_* labels. This is to avoid freeing
1918 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1919 * kernel threads (PF_KTHREAD).
1921 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1923 * Clear TID on mm_release()?
1925 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1927 ftrace_graph_init_task(p
);
1929 rt_mutex_init_task(p
);
1931 #ifdef CONFIG_PROVE_LOCKING
1932 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1933 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1936 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1937 task_rlimit(p
, RLIMIT_NPROC
)) {
1938 if (p
->real_cred
->user
!= INIT_USER
&&
1939 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1942 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1944 retval
= copy_creds(p
, clone_flags
);
1949 * If multiple threads are within copy_process(), then this check
1950 * triggers too late. This doesn't hurt, the check is only there
1951 * to stop root fork bombs.
1954 if (nr_threads
>= max_threads
)
1955 goto bad_fork_cleanup_count
;
1957 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1958 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1959 p
->flags
|= PF_FORKNOEXEC
;
1960 INIT_LIST_HEAD(&p
->children
);
1961 INIT_LIST_HEAD(&p
->sibling
);
1962 rcu_copy_process(p
);
1963 p
->vfork_done
= NULL
;
1964 spin_lock_init(&p
->alloc_lock
);
1966 init_sigpending(&p
->pending
);
1968 p
->utime
= p
->stime
= p
->gtime
= 0;
1969 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1970 p
->utimescaled
= p
->stimescaled
= 0;
1972 prev_cputime_init(&p
->prev_cputime
);
1974 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1975 seqcount_init(&p
->vtime
.seqcount
);
1976 p
->vtime
.starttime
= 0;
1977 p
->vtime
.state
= VTIME_INACTIVE
;
1980 #if defined(SPLIT_RSS_COUNTING)
1981 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1984 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1990 task_io_accounting_init(&p
->ioac
);
1991 acct_clear_integrals(p
);
1993 posix_cputimers_init(&p
->posix_cputimers
);
1995 p
->io_context
= NULL
;
1996 audit_set_context(p
, NULL
);
1999 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2000 if (IS_ERR(p
->mempolicy
)) {
2001 retval
= PTR_ERR(p
->mempolicy
);
2002 p
->mempolicy
= NULL
;
2003 goto bad_fork_cleanup_threadgroup_lock
;
2006 #ifdef CONFIG_CPUSETS
2007 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2008 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2009 seqcount_init(&p
->mems_allowed_seq
);
2011 #ifdef CONFIG_TRACE_IRQFLAGS
2013 p
->hardirqs_enabled
= 0;
2014 p
->hardirq_enable_ip
= 0;
2015 p
->hardirq_enable_event
= 0;
2016 p
->hardirq_disable_ip
= _THIS_IP_
;
2017 p
->hardirq_disable_event
= 0;
2018 p
->softirqs_enabled
= 1;
2019 p
->softirq_enable_ip
= _THIS_IP_
;
2020 p
->softirq_enable_event
= 0;
2021 p
->softirq_disable_ip
= 0;
2022 p
->softirq_disable_event
= 0;
2023 p
->hardirq_context
= 0;
2024 p
->softirq_context
= 0;
2027 p
->pagefault_disabled
= 0;
2029 #ifdef CONFIG_LOCKDEP
2030 lockdep_init_task(p
);
2033 #ifdef CONFIG_DEBUG_MUTEXES
2034 p
->blocked_on
= NULL
; /* not blocked yet */
2036 #ifdef CONFIG_BCACHE
2037 p
->sequential_io
= 0;
2038 p
->sequential_io_avg
= 0;
2041 /* Perform scheduler related setup. Assign this task to a CPU. */
2042 retval
= sched_fork(clone_flags
, p
);
2044 goto bad_fork_cleanup_policy
;
2046 retval
= perf_event_init_task(p
);
2048 goto bad_fork_cleanup_policy
;
2049 retval
= audit_alloc(p
);
2051 goto bad_fork_cleanup_perf
;
2052 /* copy all the process information */
2054 retval
= security_task_alloc(p
, clone_flags
);
2056 goto bad_fork_cleanup_audit
;
2057 retval
= copy_semundo(clone_flags
, p
);
2059 goto bad_fork_cleanup_security
;
2060 retval
= copy_files(clone_flags
, p
);
2062 goto bad_fork_cleanup_semundo
;
2063 retval
= copy_fs(clone_flags
, p
);
2065 goto bad_fork_cleanup_files
;
2066 retval
= copy_sighand(clone_flags
, p
);
2068 goto bad_fork_cleanup_fs
;
2069 retval
= copy_signal(clone_flags
, p
);
2071 goto bad_fork_cleanup_sighand
;
2072 retval
= copy_mm(clone_flags
, p
);
2074 goto bad_fork_cleanup_signal
;
2075 retval
= copy_namespaces(clone_flags
, p
);
2077 goto bad_fork_cleanup_mm
;
2078 retval
= copy_io(clone_flags
, p
);
2080 goto bad_fork_cleanup_namespaces
;
2081 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2084 goto bad_fork_cleanup_io
;
2086 stackleak_task_init(p
);
2088 if (pid
!= &init_struct_pid
) {
2089 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2090 args
->set_tid_size
);
2092 retval
= PTR_ERR(pid
);
2093 goto bad_fork_cleanup_thread
;
2098 * This has to happen after we've potentially unshared the file
2099 * descriptor table (so that the pidfd doesn't leak into the child
2100 * if the fd table isn't shared).
2102 if (clone_flags
& CLONE_PIDFD
) {
2103 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2105 goto bad_fork_free_pid
;
2109 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2110 O_RDWR
| O_CLOEXEC
);
2111 if (IS_ERR(pidfile
)) {
2112 put_unused_fd(pidfd
);
2113 retval
= PTR_ERR(pidfile
);
2114 goto bad_fork_free_pid
;
2116 get_pid(pid
); /* held by pidfile now */
2118 retval
= put_user(pidfd
, args
->pidfd
);
2120 goto bad_fork_put_pidfd
;
2129 * sigaltstack should be cleared when sharing the same VM
2131 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2135 * Syscall tracing and stepping should be turned off in the
2136 * child regardless of CLONE_PTRACE.
2138 user_disable_single_step(p
);
2139 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2140 #ifdef TIF_SYSCALL_EMU
2141 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2143 clear_tsk_latency_tracing(p
);
2145 /* ok, now we should be set up.. */
2146 p
->pid
= pid_nr(pid
);
2147 if (clone_flags
& CLONE_THREAD
) {
2148 p
->exit_signal
= -1;
2149 p
->group_leader
= current
->group_leader
;
2150 p
->tgid
= current
->tgid
;
2152 if (clone_flags
& CLONE_PARENT
)
2153 p
->exit_signal
= current
->group_leader
->exit_signal
;
2155 p
->exit_signal
= args
->exit_signal
;
2156 p
->group_leader
= p
;
2161 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2162 p
->dirty_paused_when
= 0;
2164 p
->pdeath_signal
= 0;
2165 INIT_LIST_HEAD(&p
->thread_group
);
2166 p
->task_works
= NULL
;
2168 cgroup_threadgroup_change_begin(current
);
2170 * Ensure that the cgroup subsystem policies allow the new process to be
2171 * forked. It should be noted the the new process's css_set can be changed
2172 * between here and cgroup_post_fork() if an organisation operation is in
2175 retval
= cgroup_can_fork(p
);
2177 goto bad_fork_cgroup_threadgroup_change_end
;
2180 * From this point on we must avoid any synchronous user-space
2181 * communication until we take the tasklist-lock. In particular, we do
2182 * not want user-space to be able to predict the process start-time by
2183 * stalling fork(2) after we recorded the start_time but before it is
2184 * visible to the system.
2187 p
->start_time
= ktime_get_ns();
2188 p
->start_boottime
= ktime_get_boottime_ns();
2191 * Make it visible to the rest of the system, but dont wake it up yet.
2192 * Need tasklist lock for parent etc handling!
2194 write_lock_irq(&tasklist_lock
);
2196 /* CLONE_PARENT re-uses the old parent */
2197 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2198 p
->real_parent
= current
->real_parent
;
2199 p
->parent_exec_id
= current
->parent_exec_id
;
2201 p
->real_parent
= current
;
2202 p
->parent_exec_id
= current
->self_exec_id
;
2205 klp_copy_process(p
);
2207 spin_lock(¤t
->sighand
->siglock
);
2210 * Copy seccomp details explicitly here, in case they were changed
2211 * before holding sighand lock.
2215 rseq_fork(p
, clone_flags
);
2217 /* Don't start children in a dying pid namespace */
2218 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2220 goto bad_fork_cancel_cgroup
;
2223 /* Let kill terminate clone/fork in the middle */
2224 if (fatal_signal_pending(current
)) {
2226 goto bad_fork_cancel_cgroup
;
2229 /* past the last point of failure */
2231 fd_install(pidfd
, pidfile
);
2233 init_task_pid_links(p
);
2234 if (likely(p
->pid
)) {
2235 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2237 init_task_pid(p
, PIDTYPE_PID
, pid
);
2238 if (thread_group_leader(p
)) {
2239 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2240 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2241 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2243 if (is_child_reaper(pid
)) {
2244 ns_of_pid(pid
)->child_reaper
= p
;
2245 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2247 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2248 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2250 * Inherit has_child_subreaper flag under the same
2251 * tasklist_lock with adding child to the process tree
2252 * for propagate_has_child_subreaper optimization.
2254 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2255 p
->real_parent
->signal
->is_child_subreaper
;
2256 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2257 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2258 attach_pid(p
, PIDTYPE_TGID
);
2259 attach_pid(p
, PIDTYPE_PGID
);
2260 attach_pid(p
, PIDTYPE_SID
);
2261 __this_cpu_inc(process_counts
);
2263 current
->signal
->nr_threads
++;
2264 atomic_inc(¤t
->signal
->live
);
2265 refcount_inc(¤t
->signal
->sigcnt
);
2266 task_join_group_stop(p
);
2267 list_add_tail_rcu(&p
->thread_group
,
2268 &p
->group_leader
->thread_group
);
2269 list_add_tail_rcu(&p
->thread_node
,
2270 &p
->signal
->thread_head
);
2272 attach_pid(p
, PIDTYPE_PID
);
2276 hlist_del_init(&delayed
.node
);
2277 spin_unlock(¤t
->sighand
->siglock
);
2278 syscall_tracepoint_update(p
);
2279 write_unlock_irq(&tasklist_lock
);
2281 proc_fork_connector(p
);
2282 cgroup_post_fork(p
);
2283 cgroup_threadgroup_change_end(current
);
2286 trace_task_newtask(p
, clone_flags
);
2287 uprobe_copy_process(p
, clone_flags
);
2291 bad_fork_cancel_cgroup
:
2292 spin_unlock(¤t
->sighand
->siglock
);
2293 write_unlock_irq(&tasklist_lock
);
2294 cgroup_cancel_fork(p
);
2295 bad_fork_cgroup_threadgroup_change_end
:
2296 cgroup_threadgroup_change_end(current
);
2298 if (clone_flags
& CLONE_PIDFD
) {
2300 put_unused_fd(pidfd
);
2303 if (pid
!= &init_struct_pid
)
2305 bad_fork_cleanup_thread
:
2307 bad_fork_cleanup_io
:
2310 bad_fork_cleanup_namespaces
:
2311 exit_task_namespaces(p
);
2312 bad_fork_cleanup_mm
:
2314 mm_clear_owner(p
->mm
, p
);
2317 bad_fork_cleanup_signal
:
2318 if (!(clone_flags
& CLONE_THREAD
))
2319 free_signal_struct(p
->signal
);
2320 bad_fork_cleanup_sighand
:
2321 __cleanup_sighand(p
->sighand
);
2322 bad_fork_cleanup_fs
:
2323 exit_fs(p
); /* blocking */
2324 bad_fork_cleanup_files
:
2325 exit_files(p
); /* blocking */
2326 bad_fork_cleanup_semundo
:
2328 bad_fork_cleanup_security
:
2329 security_task_free(p
);
2330 bad_fork_cleanup_audit
:
2332 bad_fork_cleanup_perf
:
2333 perf_event_free_task(p
);
2334 bad_fork_cleanup_policy
:
2335 lockdep_free_task(p
);
2337 mpol_put(p
->mempolicy
);
2338 bad_fork_cleanup_threadgroup_lock
:
2340 delayacct_tsk_free(p
);
2341 bad_fork_cleanup_count
:
2342 atomic_dec(&p
->cred
->user
->processes
);
2345 p
->state
= TASK_DEAD
;
2347 delayed_free_task(p
);
2349 spin_lock_irq(¤t
->sighand
->siglock
);
2350 hlist_del_init(&delayed
.node
);
2351 spin_unlock_irq(¤t
->sighand
->siglock
);
2352 return ERR_PTR(retval
);
2355 static inline void init_idle_pids(struct task_struct
*idle
)
2359 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2360 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2361 init_task_pid(idle
, type
, &init_struct_pid
);
2365 struct task_struct
*fork_idle(int cpu
)
2367 struct task_struct
*task
;
2368 struct kernel_clone_args args
= {
2372 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2373 if (!IS_ERR(task
)) {
2374 init_idle_pids(task
);
2375 init_idle(task
, cpu
);
2381 struct mm_struct
*copy_init_mm(void)
2383 return dup_mm(NULL
, &init_mm
);
2387 * Ok, this is the main fork-routine.
2389 * It copies the process, and if successful kick-starts
2390 * it and waits for it to finish using the VM if required.
2392 * args->exit_signal is expected to be checked for sanity by the caller.
2394 long _do_fork(struct kernel_clone_args
*args
)
2396 u64 clone_flags
= args
->flags
;
2397 struct completion vfork
;
2399 struct task_struct
*p
;
2404 * Determine whether and which event to report to ptracer. When
2405 * called from kernel_thread or CLONE_UNTRACED is explicitly
2406 * requested, no event is reported; otherwise, report if the event
2407 * for the type of forking is enabled.
2409 if (!(clone_flags
& CLONE_UNTRACED
)) {
2410 if (clone_flags
& CLONE_VFORK
)
2411 trace
= PTRACE_EVENT_VFORK
;
2412 else if (args
->exit_signal
!= SIGCHLD
)
2413 trace
= PTRACE_EVENT_CLONE
;
2415 trace
= PTRACE_EVENT_FORK
;
2417 if (likely(!ptrace_event_enabled(current
, trace
)))
2421 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2422 add_latent_entropy();
2428 * Do this prior waking up the new thread - the thread pointer
2429 * might get invalid after that point, if the thread exits quickly.
2431 trace_sched_process_fork(current
, p
);
2433 pid
= get_task_pid(p
, PIDTYPE_PID
);
2436 if (clone_flags
& CLONE_PARENT_SETTID
)
2437 put_user(nr
, args
->parent_tid
);
2439 if (clone_flags
& CLONE_VFORK
) {
2440 p
->vfork_done
= &vfork
;
2441 init_completion(&vfork
);
2445 wake_up_new_task(p
);
2447 /* forking complete and child started to run, tell ptracer */
2448 if (unlikely(trace
))
2449 ptrace_event_pid(trace
, pid
);
2451 if (clone_flags
& CLONE_VFORK
) {
2452 if (!wait_for_vfork_done(p
, &vfork
))
2453 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2460 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2462 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2463 if ((kargs
->flags
& CLONE_PIDFD
) &&
2464 (kargs
->flags
& CLONE_PARENT_SETTID
))
2470 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2471 /* For compatibility with architectures that call do_fork directly rather than
2472 * using the syscall entry points below. */
2473 long do_fork(unsigned long clone_flags
,
2474 unsigned long stack_start
,
2475 unsigned long stack_size
,
2476 int __user
*parent_tidptr
,
2477 int __user
*child_tidptr
)
2479 struct kernel_clone_args args
= {
2480 .flags
= (clone_flags
& ~CSIGNAL
),
2481 .pidfd
= parent_tidptr
,
2482 .child_tid
= child_tidptr
,
2483 .parent_tid
= parent_tidptr
,
2484 .exit_signal
= (clone_flags
& CSIGNAL
),
2485 .stack
= stack_start
,
2486 .stack_size
= stack_size
,
2489 if (!legacy_clone_args_valid(&args
))
2492 return _do_fork(&args
);
2497 * Create a kernel thread.
2499 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2501 struct kernel_clone_args args
= {
2502 .flags
= ((flags
| CLONE_VM
| CLONE_UNTRACED
) & ~CSIGNAL
),
2503 .exit_signal
= (flags
& CSIGNAL
),
2504 .stack
= (unsigned long)fn
,
2505 .stack_size
= (unsigned long)arg
,
2508 return _do_fork(&args
);
2511 #ifdef __ARCH_WANT_SYS_FORK
2512 SYSCALL_DEFINE0(fork
)
2515 struct kernel_clone_args args
= {
2516 .exit_signal
= SIGCHLD
,
2519 return _do_fork(&args
);
2521 /* can not support in nommu mode */
2527 #ifdef __ARCH_WANT_SYS_VFORK
2528 SYSCALL_DEFINE0(vfork
)
2530 struct kernel_clone_args args
= {
2531 .flags
= CLONE_VFORK
| CLONE_VM
,
2532 .exit_signal
= SIGCHLD
,
2535 return _do_fork(&args
);
2539 #ifdef __ARCH_WANT_SYS_CLONE
2540 #ifdef CONFIG_CLONE_BACKWARDS
2541 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2542 int __user
*, parent_tidptr
,
2544 int __user
*, child_tidptr
)
2545 #elif defined(CONFIG_CLONE_BACKWARDS2)
2546 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2547 int __user
*, parent_tidptr
,
2548 int __user
*, child_tidptr
,
2550 #elif defined(CONFIG_CLONE_BACKWARDS3)
2551 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2553 int __user
*, parent_tidptr
,
2554 int __user
*, child_tidptr
,
2557 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2558 int __user
*, parent_tidptr
,
2559 int __user
*, child_tidptr
,
2563 struct kernel_clone_args args
= {
2564 .flags
= (clone_flags
& ~CSIGNAL
),
2565 .pidfd
= parent_tidptr
,
2566 .child_tid
= child_tidptr
,
2567 .parent_tid
= parent_tidptr
,
2568 .exit_signal
= (clone_flags
& CSIGNAL
),
2573 if (!legacy_clone_args_valid(&args
))
2576 return _do_fork(&args
);
2580 #ifdef __ARCH_WANT_SYS_CLONE3
2581 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2582 struct clone_args __user
*uargs
,
2586 struct clone_args args
;
2587 pid_t
*kset_tid
= kargs
->set_tid
;
2589 if (unlikely(usize
> PAGE_SIZE
))
2591 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2594 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2598 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2601 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2604 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2608 * Verify that higher 32bits of exit_signal are unset and that
2609 * it is a valid signal
2611 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2612 !valid_signal(args
.exit_signal
)))
2615 *kargs
= (struct kernel_clone_args
){
2616 .flags
= args
.flags
,
2617 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2618 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2619 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2620 .exit_signal
= args
.exit_signal
,
2621 .stack
= args
.stack
,
2622 .stack_size
= args
.stack_size
,
2624 .set_tid_size
= args
.set_tid_size
,
2628 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
2629 (kargs
->set_tid_size
* sizeof(pid_t
))))
2632 kargs
->set_tid
= kset_tid
;
2638 * clone3_stack_valid - check and prepare stack
2639 * @kargs: kernel clone args
2641 * Verify that the stack arguments userspace gave us are sane.
2642 * In addition, set the stack direction for userspace since it's easy for us to
2645 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2647 if (kargs
->stack
== 0) {
2648 if (kargs
->stack_size
> 0)
2651 if (kargs
->stack_size
== 0)
2654 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2657 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2658 kargs
->stack
+= kargs
->stack_size
;
2665 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2667 /* Verify that no unknown flags are passed along. */
2668 if (kargs
->flags
& ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
))
2672 * - make the CLONE_DETACHED bit reuseable for clone3
2673 * - make the CSIGNAL bits reuseable for clone3
2675 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2678 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
2679 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
2682 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2686 if (!clone3_stack_valid(kargs
))
2693 * clone3 - create a new process with specific properties
2694 * @uargs: argument structure
2695 * @size: size of @uargs
2697 * clone3() is the extensible successor to clone()/clone2().
2698 * It takes a struct as argument that is versioned by its size.
2700 * Return: On success, a positive PID for the child process.
2701 * On error, a negative errno number.
2703 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2707 struct kernel_clone_args kargs
;
2708 pid_t set_tid
[MAX_PID_NS_LEVEL
];
2710 kargs
.set_tid
= set_tid
;
2712 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2716 if (!clone3_args_valid(&kargs
))
2719 return _do_fork(&kargs
);
2723 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2725 struct task_struct
*leader
, *parent
, *child
;
2728 read_lock(&tasklist_lock
);
2729 leader
= top
= top
->group_leader
;
2731 for_each_thread(leader
, parent
) {
2732 list_for_each_entry(child
, &parent
->children
, sibling
) {
2733 res
= visitor(child
, data
);
2745 if (leader
!= top
) {
2747 parent
= child
->real_parent
;
2748 leader
= parent
->group_leader
;
2752 read_unlock(&tasklist_lock
);
2755 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2756 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2759 static void sighand_ctor(void *data
)
2761 struct sighand_struct
*sighand
= data
;
2763 spin_lock_init(&sighand
->siglock
);
2764 init_waitqueue_head(&sighand
->signalfd_wqh
);
2767 void __init
proc_caches_init(void)
2769 unsigned int mm_size
;
2771 sighand_cachep
= kmem_cache_create("sighand_cache",
2772 sizeof(struct sighand_struct
), 0,
2773 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2774 SLAB_ACCOUNT
, sighand_ctor
);
2775 signal_cachep
= kmem_cache_create("signal_cache",
2776 sizeof(struct signal_struct
), 0,
2777 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2779 files_cachep
= kmem_cache_create("files_cache",
2780 sizeof(struct files_struct
), 0,
2781 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2783 fs_cachep
= kmem_cache_create("fs_cache",
2784 sizeof(struct fs_struct
), 0,
2785 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2789 * The mm_cpumask is located at the end of mm_struct, and is
2790 * dynamically sized based on the maximum CPU number this system
2791 * can have, taking hotplug into account (nr_cpu_ids).
2793 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2795 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2796 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2797 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2798 offsetof(struct mm_struct
, saved_auxv
),
2799 sizeof_field(struct mm_struct
, saved_auxv
),
2801 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2803 nsproxy_cache_init();
2807 * Check constraints on flags passed to the unshare system call.
2809 static int check_unshare_flags(unsigned long unshare_flags
)
2811 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2812 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2813 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2814 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2817 * Not implemented, but pretend it works if there is nothing
2818 * to unshare. Note that unsharing the address space or the
2819 * signal handlers also need to unshare the signal queues (aka
2822 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2823 if (!thread_group_empty(current
))
2826 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2827 if (refcount_read(¤t
->sighand
->count
) > 1)
2830 if (unshare_flags
& CLONE_VM
) {
2831 if (!current_is_single_threaded())
2839 * Unshare the filesystem structure if it is being shared
2841 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2843 struct fs_struct
*fs
= current
->fs
;
2845 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2848 /* don't need lock here; in the worst case we'll do useless copy */
2852 *new_fsp
= copy_fs_struct(fs
);
2860 * Unshare file descriptor table if it is being shared
2862 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2864 struct files_struct
*fd
= current
->files
;
2867 if ((unshare_flags
& CLONE_FILES
) &&
2868 (fd
&& atomic_read(&fd
->count
) > 1)) {
2869 *new_fdp
= dup_fd(fd
, &error
);
2878 * unshare allows a process to 'unshare' part of the process
2879 * context which was originally shared using clone. copy_*
2880 * functions used by do_fork() cannot be used here directly
2881 * because they modify an inactive task_struct that is being
2882 * constructed. Here we are modifying the current, active,
2885 int ksys_unshare(unsigned long unshare_flags
)
2887 struct fs_struct
*fs
, *new_fs
= NULL
;
2888 struct files_struct
*fd
, *new_fd
= NULL
;
2889 struct cred
*new_cred
= NULL
;
2890 struct nsproxy
*new_nsproxy
= NULL
;
2895 * If unsharing a user namespace must also unshare the thread group
2896 * and unshare the filesystem root and working directories.
2898 if (unshare_flags
& CLONE_NEWUSER
)
2899 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2901 * If unsharing vm, must also unshare signal handlers.
2903 if (unshare_flags
& CLONE_VM
)
2904 unshare_flags
|= CLONE_SIGHAND
;
2906 * If unsharing a signal handlers, must also unshare the signal queues.
2908 if (unshare_flags
& CLONE_SIGHAND
)
2909 unshare_flags
|= CLONE_THREAD
;
2911 * If unsharing namespace, must also unshare filesystem information.
2913 if (unshare_flags
& CLONE_NEWNS
)
2914 unshare_flags
|= CLONE_FS
;
2916 err
= check_unshare_flags(unshare_flags
);
2918 goto bad_unshare_out
;
2920 * CLONE_NEWIPC must also detach from the undolist: after switching
2921 * to a new ipc namespace, the semaphore arrays from the old
2922 * namespace are unreachable.
2924 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2926 err
= unshare_fs(unshare_flags
, &new_fs
);
2928 goto bad_unshare_out
;
2929 err
= unshare_fd(unshare_flags
, &new_fd
);
2931 goto bad_unshare_cleanup_fs
;
2932 err
= unshare_userns(unshare_flags
, &new_cred
);
2934 goto bad_unshare_cleanup_fd
;
2935 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2938 goto bad_unshare_cleanup_cred
;
2940 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2943 * CLONE_SYSVSEM is equivalent to sys_exit().
2947 if (unshare_flags
& CLONE_NEWIPC
) {
2948 /* Orphan segments in old ns (see sem above). */
2950 shm_init_task(current
);
2954 switch_task_namespaces(current
, new_nsproxy
);
2960 spin_lock(&fs
->lock
);
2961 current
->fs
= new_fs
;
2966 spin_unlock(&fs
->lock
);
2970 fd
= current
->files
;
2971 current
->files
= new_fd
;
2975 task_unlock(current
);
2978 /* Install the new user namespace */
2979 commit_creds(new_cred
);
2984 perf_event_namespaces(current
);
2986 bad_unshare_cleanup_cred
:
2989 bad_unshare_cleanup_fd
:
2991 put_files_struct(new_fd
);
2993 bad_unshare_cleanup_fs
:
2995 free_fs_struct(new_fs
);
3001 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3003 return ksys_unshare(unshare_flags
);
3007 * Helper to unshare the files of the current task.
3008 * We don't want to expose copy_files internals to
3009 * the exec layer of the kernel.
3012 int unshare_files(struct files_struct
**displaced
)
3014 struct task_struct
*task
= current
;
3015 struct files_struct
*copy
= NULL
;
3018 error
= unshare_fd(CLONE_FILES
, ©
);
3019 if (error
|| !copy
) {
3023 *displaced
= task
->files
;
3030 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3031 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
3035 int threads
= max_threads
;
3037 int max
= MAX_THREADS
;
3044 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3048 max_threads
= threads
;