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>
43 #include <linux/hmm.h>
46 #include <linux/vmacache.h>
47 #include <linux/nsproxy.h>
48 #include <linux/capability.h>
49 #include <linux/cpu.h>
50 #include <linux/cgroup.h>
51 #include <linux/security.h>
52 #include <linux/hugetlb.h>
53 #include <linux/seccomp.h>
54 #include <linux/swap.h>
55 #include <linux/syscalls.h>
56 #include <linux/jiffies.h>
57 #include <linux/futex.h>
58 #include <linux/compat.h>
59 #include <linux/kthread.h>
60 #include <linux/task_io_accounting_ops.h>
61 #include <linux/rcupdate.h>
62 #include <linux/ptrace.h>
63 #include <linux/mount.h>
64 #include <linux/audit.h>
65 #include <linux/memcontrol.h>
66 #include <linux/ftrace.h>
67 #include <linux/proc_fs.h>
68 #include <linux/profile.h>
69 #include <linux/rmap.h>
70 #include <linux/ksm.h>
71 #include <linux/acct.h>
72 #include <linux/userfaultfd_k.h>
73 #include <linux/tsacct_kern.h>
74 #include <linux/cn_proc.h>
75 #include <linux/freezer.h>
76 #include <linux/delayacct.h>
77 #include <linux/taskstats_kern.h>
78 #include <linux/random.h>
79 #include <linux/tty.h>
80 #include <linux/blkdev.h>
81 #include <linux/fs_struct.h>
82 #include <linux/magic.h>
83 #include <linux/perf_event.h>
84 #include <linux/posix-timers.h>
85 #include <linux/user-return-notifier.h>
86 #include <linux/oom.h>
87 #include <linux/khugepaged.h>
88 #include <linux/signalfd.h>
89 #include <linux/uprobes.h>
90 #include <linux/aio.h>
91 #include <linux/compiler.h>
92 #include <linux/sysctl.h>
93 #include <linux/kcov.h>
94 #include <linux/livepatch.h>
95 #include <linux/thread_info.h>
96 #include <linux/stackleak.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>
109 #ifdef CONFIG_USER_NS
110 extern int unprivileged_userns_clone
;
112 #define unprivileged_userns_clone 0
116 * Minimum number of threads to boot the kernel
118 #define MIN_THREADS 20
121 * Maximum number of threads
123 #define MAX_THREADS FUTEX_TID_MASK
126 * Protected counters by write_lock_irq(&tasklist_lock)
128 unsigned long total_forks
; /* Handle normal Linux uptimes. */
129 int nr_threads
; /* The idle threads do not count.. */
131 static int max_threads
; /* tunable limit on nr_threads */
133 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
135 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
137 #ifdef CONFIG_PROVE_RCU
138 int lockdep_tasklist_lock_is_held(void)
140 return lockdep_is_held(&tasklist_lock
);
142 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
143 #endif /* #ifdef CONFIG_PROVE_RCU */
145 int nr_processes(void)
150 for_each_possible_cpu(cpu
)
151 total
+= per_cpu(process_counts
, cpu
);
156 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
160 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
161 static struct kmem_cache
*task_struct_cachep
;
163 static inline struct task_struct
*alloc_task_struct_node(int node
)
165 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
168 static inline void free_task_struct(struct task_struct
*tsk
)
170 kmem_cache_free(task_struct_cachep
, tsk
);
174 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
177 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
178 * kmemcache based allocator.
180 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
182 #ifdef CONFIG_VMAP_STACK
184 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
185 * flush. Try to minimize the number of calls by caching stacks.
187 #define NR_CACHED_STACKS 2
188 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
190 static int free_vm_stack_cache(unsigned int cpu
)
192 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
195 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
196 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
201 vfree(vm_stack
->addr
);
202 cached_vm_stacks
[i
] = NULL
;
209 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
211 #ifdef CONFIG_VMAP_STACK
215 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
218 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
223 /* Clear stale pointers from reused stack. */
224 memset(s
->addr
, 0, THREAD_SIZE
);
226 tsk
->stack_vm_area
= s
;
227 tsk
->stack
= s
->addr
;
232 * Allocated stacks are cached and later reused by new threads,
233 * so memcg accounting is performed manually on assigning/releasing
234 * stacks to tasks. Drop __GFP_ACCOUNT.
236 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
237 VMALLOC_START
, VMALLOC_END
,
238 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
240 0, node
, __builtin_return_address(0));
243 * We can't call find_vm_area() in interrupt context, and
244 * free_thread_stack() can be called in interrupt context,
245 * so cache the vm_struct.
248 tsk
->stack_vm_area
= find_vm_area(stack
);
253 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
257 tsk
->stack
= page_address(page
);
264 static inline void free_thread_stack(struct task_struct
*tsk
)
266 #ifdef CONFIG_VMAP_STACK
267 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
272 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
273 mod_memcg_page_state(vm
->pages
[i
],
274 MEMCG_KERNEL_STACK_KB
,
275 -(int)(PAGE_SIZE
/ 1024));
277 memcg_kmem_uncharge(vm
->pages
[i
], 0);
280 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
281 if (this_cpu_cmpxchg(cached_stacks
[i
],
282 NULL
, tsk
->stack_vm_area
) != NULL
)
288 vfree_atomic(tsk
->stack
);
293 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
296 static struct kmem_cache
*thread_stack_cache
;
298 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
301 unsigned long *stack
;
302 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
307 static void free_thread_stack(struct task_struct
*tsk
)
309 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
312 void thread_stack_cache_init(void)
314 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
315 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
317 BUG_ON(thread_stack_cache
== NULL
);
322 /* SLAB cache for signal_struct structures (tsk->signal) */
323 static struct kmem_cache
*signal_cachep
;
325 /* SLAB cache for sighand_struct structures (tsk->sighand) */
326 struct kmem_cache
*sighand_cachep
;
328 /* SLAB cache for files_struct structures (tsk->files) */
329 struct kmem_cache
*files_cachep
;
331 /* SLAB cache for fs_struct structures (tsk->fs) */
332 struct kmem_cache
*fs_cachep
;
334 /* SLAB cache for vm_area_struct structures */
335 static struct kmem_cache
*vm_area_cachep
;
337 /* SLAB cache for mm_struct structures (tsk->mm) */
338 static struct kmem_cache
*mm_cachep
;
340 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
342 struct vm_area_struct
*vma
;
344 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
350 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
352 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
356 INIT_LIST_HEAD(&new->anon_vma_chain
);
361 void vm_area_free(struct vm_area_struct
*vma
)
363 kmem_cache_free(vm_area_cachep
, vma
);
366 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
368 void *stack
= task_stack_page(tsk
);
369 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
371 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
376 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
378 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
379 mod_zone_page_state(page_zone(vm
->pages
[i
]),
381 PAGE_SIZE
/ 1024 * account
);
385 * All stack pages are in the same zone and belong to the
388 struct page
*first_page
= virt_to_page(stack
);
390 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
391 THREAD_SIZE
/ 1024 * account
);
393 mod_memcg_page_state(first_page
, MEMCG_KERNEL_STACK_KB
,
394 account
* (THREAD_SIZE
/ 1024));
398 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
400 #ifdef CONFIG_VMAP_STACK
401 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
407 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
409 * If memcg_kmem_charge() fails, page->mem_cgroup
410 * pointer is NULL, and both memcg_kmem_uncharge()
411 * and mod_memcg_page_state() in free_thread_stack()
412 * will ignore this page. So it's safe.
414 ret
= memcg_kmem_charge(vm
->pages
[i
], GFP_KERNEL
, 0);
418 mod_memcg_page_state(vm
->pages
[i
],
419 MEMCG_KERNEL_STACK_KB
,
427 static void release_task_stack(struct task_struct
*tsk
)
429 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
430 return; /* Better to leak the stack than to free prematurely */
432 account_kernel_stack(tsk
, -1);
433 free_thread_stack(tsk
);
435 #ifdef CONFIG_VMAP_STACK
436 tsk
->stack_vm_area
= NULL
;
440 #ifdef CONFIG_THREAD_INFO_IN_TASK
441 void put_task_stack(struct task_struct
*tsk
)
443 if (refcount_dec_and_test(&tsk
->stack_refcount
))
444 release_task_stack(tsk
);
448 void free_task(struct task_struct
*tsk
)
450 #ifndef CONFIG_THREAD_INFO_IN_TASK
452 * The task is finally done with both the stack and thread_info,
455 release_task_stack(tsk
);
458 * If the task had a separate stack allocation, it should be gone
461 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
463 rt_mutex_debug_task_free(tsk
);
464 ftrace_graph_exit_task(tsk
);
465 put_seccomp_filter(tsk
);
466 arch_release_task_struct(tsk
);
467 if (tsk
->flags
& PF_KTHREAD
)
468 free_kthread_struct(tsk
);
469 free_task_struct(tsk
);
471 EXPORT_SYMBOL(free_task
);
474 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
475 struct mm_struct
*oldmm
)
477 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
478 struct rb_node
**rb_link
, *rb_parent
;
480 unsigned long charge
;
483 uprobe_start_dup_mmap();
484 if (down_write_killable(&oldmm
->mmap_sem
)) {
486 goto fail_uprobe_end
;
488 flush_cache_dup_mm(oldmm
);
489 uprobe_dup_mmap(oldmm
, mm
);
491 * Not linked in yet - no deadlock potential:
493 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
495 /* No ordering required: file already has been exposed. */
496 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
498 mm
->total_vm
= oldmm
->total_vm
;
499 mm
->data_vm
= oldmm
->data_vm
;
500 mm
->exec_vm
= oldmm
->exec_vm
;
501 mm
->stack_vm
= oldmm
->stack_vm
;
503 rb_link
= &mm
->mm_rb
.rb_node
;
506 retval
= ksm_fork(mm
, oldmm
);
509 retval
= khugepaged_fork(mm
, oldmm
);
514 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
517 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
518 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
523 * Don't duplicate many vmas if we've been oom-killed (for
526 if (fatal_signal_pending(current
)) {
530 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
531 unsigned long len
= vma_pages(mpnt
);
533 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
537 tmp
= vm_area_dup(mpnt
);
540 retval
= vma_dup_policy(mpnt
, tmp
);
542 goto fail_nomem_policy
;
544 retval
= dup_userfaultfd(tmp
, &uf
);
546 goto fail_nomem_anon_vma_fork
;
547 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
548 /* VM_WIPEONFORK gets a clean slate in the child. */
549 tmp
->anon_vma
= NULL
;
550 if (anon_vma_prepare(tmp
))
551 goto fail_nomem_anon_vma_fork
;
552 } else if (anon_vma_fork(tmp
, mpnt
))
553 goto fail_nomem_anon_vma_fork
;
554 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
555 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
558 struct inode
*inode
= file_inode(file
);
559 struct address_space
*mapping
= file
->f_mapping
;
562 if (tmp
->vm_flags
& VM_DENYWRITE
)
563 atomic_dec(&inode
->i_writecount
);
564 i_mmap_lock_write(mapping
);
565 if (tmp
->vm_flags
& VM_SHARED
)
566 atomic_inc(&mapping
->i_mmap_writable
);
567 flush_dcache_mmap_lock(mapping
);
568 /* insert tmp into the share list, just after mpnt */
569 vma_interval_tree_insert_after(tmp
, mpnt
,
571 flush_dcache_mmap_unlock(mapping
);
572 i_mmap_unlock_write(mapping
);
576 * Clear hugetlb-related page reserves for children. This only
577 * affects MAP_PRIVATE mappings. Faults generated by the child
578 * are not guaranteed to succeed, even if read-only
580 if (is_vm_hugetlb_page(tmp
))
581 reset_vma_resv_huge_pages(tmp
);
584 * Link in the new vma and copy the page table entries.
587 pprev
= &tmp
->vm_next
;
591 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
592 rb_link
= &tmp
->vm_rb
.rb_right
;
593 rb_parent
= &tmp
->vm_rb
;
596 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
597 retval
= copy_page_range(mm
, oldmm
, mpnt
);
599 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
600 tmp
->vm_ops
->open(tmp
);
605 /* a new mm has just been created */
606 retval
= arch_dup_mmap(oldmm
, mm
);
608 up_write(&mm
->mmap_sem
);
610 up_write(&oldmm
->mmap_sem
);
611 dup_userfaultfd_complete(&uf
);
613 uprobe_end_dup_mmap();
615 fail_nomem_anon_vma_fork
:
616 mpol_put(vma_policy(tmp
));
621 vm_unacct_memory(charge
);
625 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
627 mm
->pgd
= pgd_alloc(mm
);
628 if (unlikely(!mm
->pgd
))
633 static inline void mm_free_pgd(struct mm_struct
*mm
)
635 pgd_free(mm
, mm
->pgd
);
638 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
640 down_write(&oldmm
->mmap_sem
);
641 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
642 up_write(&oldmm
->mmap_sem
);
645 #define mm_alloc_pgd(mm) (0)
646 #define mm_free_pgd(mm)
647 #endif /* CONFIG_MMU */
649 static void check_mm(struct mm_struct
*mm
)
653 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
654 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
657 printk(KERN_ALERT
"BUG: Bad rss-counter state "
658 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
661 if (mm_pgtables_bytes(mm
))
662 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
663 mm_pgtables_bytes(mm
));
665 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
666 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
670 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
671 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
674 * Called when the last reference to the mm
675 * is dropped: either by a lazy thread or by
676 * mmput. Free the page directory and the mm.
678 void __mmdrop(struct mm_struct
*mm
)
680 BUG_ON(mm
== &init_mm
);
681 WARN_ON_ONCE(mm
== current
->mm
);
682 WARN_ON_ONCE(mm
== current
->active_mm
);
685 mmu_notifier_mm_destroy(mm
);
687 put_user_ns(mm
->user_ns
);
690 EXPORT_SYMBOL_GPL(__mmdrop
);
692 static void mmdrop_async_fn(struct work_struct
*work
)
694 struct mm_struct
*mm
;
696 mm
= container_of(work
, struct mm_struct
, async_put_work
);
700 static void mmdrop_async(struct mm_struct
*mm
)
702 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
703 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
704 schedule_work(&mm
->async_put_work
);
708 static inline void free_signal_struct(struct signal_struct
*sig
)
710 taskstats_tgid_free(sig
);
711 sched_autogroup_exit(sig
);
713 * __mmdrop is not safe to call from softirq context on x86 due to
714 * pgd_dtor so postpone it to the async context
717 mmdrop_async(sig
->oom_mm
);
718 kmem_cache_free(signal_cachep
, sig
);
721 static inline void put_signal_struct(struct signal_struct
*sig
)
723 if (refcount_dec_and_test(&sig
->sigcnt
))
724 free_signal_struct(sig
);
727 void __put_task_struct(struct task_struct
*tsk
)
729 WARN_ON(!tsk
->exit_state
);
730 WARN_ON(refcount_read(&tsk
->usage
));
731 WARN_ON(tsk
== current
);
734 task_numa_free(tsk
, true);
735 security_task_free(tsk
);
737 delayacct_tsk_free(tsk
);
738 put_signal_struct(tsk
->signal
);
740 if (!profile_handoff_task(tsk
))
743 EXPORT_SYMBOL_GPL(__put_task_struct
);
745 void __init __weak
arch_task_cache_init(void) { }
750 static void set_max_threads(unsigned int max_threads_suggested
)
753 unsigned long nr_pages
= totalram_pages();
756 * The number of threads shall be limited such that the thread
757 * structures may only consume a small part of the available memory.
759 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
760 threads
= MAX_THREADS
;
762 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
763 (u64
) THREAD_SIZE
* 8UL);
765 if (threads
> max_threads_suggested
)
766 threads
= max_threads_suggested
;
768 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
771 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
772 /* Initialized by the architecture: */
773 int arch_task_struct_size __read_mostly
;
776 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
778 /* Fetch thread_struct whitelist for the architecture. */
779 arch_thread_struct_whitelist(offset
, size
);
782 * Handle zero-sized whitelist or empty thread_struct, otherwise
783 * adjust offset to position of thread_struct in task_struct.
785 if (unlikely(*size
== 0))
788 *offset
+= offsetof(struct task_struct
, thread
);
791 void __init
fork_init(void)
794 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
795 #ifndef ARCH_MIN_TASKALIGN
796 #define ARCH_MIN_TASKALIGN 0
798 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
799 unsigned long useroffset
, usersize
;
801 /* create a slab on which task_structs can be allocated */
802 task_struct_whitelist(&useroffset
, &usersize
);
803 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
804 arch_task_struct_size
, align
,
805 SLAB_PANIC
|SLAB_ACCOUNT
,
806 useroffset
, usersize
, NULL
);
809 /* do the arch specific task caches init */
810 arch_task_cache_init();
812 set_max_threads(MAX_THREADS
);
814 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
815 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
816 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
817 init_task
.signal
->rlim
[RLIMIT_NPROC
];
819 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
820 init_user_ns
.ucount_max
[i
] = max_threads
/2;
823 #ifdef CONFIG_VMAP_STACK
824 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
825 NULL
, free_vm_stack_cache
);
828 lockdep_init_task(&init_task
);
832 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
833 struct task_struct
*src
)
839 void set_task_stack_end_magic(struct task_struct
*tsk
)
841 unsigned long *stackend
;
843 stackend
= end_of_stack(tsk
);
844 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
847 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
849 struct task_struct
*tsk
;
850 unsigned long *stack
;
851 struct vm_struct
*stack_vm_area __maybe_unused
;
854 if (node
== NUMA_NO_NODE
)
855 node
= tsk_fork_get_node(orig
);
856 tsk
= alloc_task_struct_node(node
);
860 stack
= alloc_thread_stack_node(tsk
, node
);
864 if (memcg_charge_kernel_stack(tsk
))
867 stack_vm_area
= task_stack_vm_area(tsk
);
869 err
= arch_dup_task_struct(tsk
, orig
);
872 * arch_dup_task_struct() clobbers the stack-related fields. Make
873 * sure they're properly initialized before using any stack-related
877 #ifdef CONFIG_VMAP_STACK
878 tsk
->stack_vm_area
= stack_vm_area
;
880 #ifdef CONFIG_THREAD_INFO_IN_TASK
881 refcount_set(&tsk
->stack_refcount
, 1);
887 #ifdef CONFIG_SECCOMP
889 * We must handle setting up seccomp filters once we're under
890 * the sighand lock in case orig has changed between now and
891 * then. Until then, filter must be NULL to avoid messing up
892 * the usage counts on the error path calling free_task.
894 tsk
->seccomp
.filter
= NULL
;
897 setup_thread_stack(tsk
, orig
);
898 clear_user_return_notifier(tsk
);
899 clear_tsk_need_resched(tsk
);
900 set_task_stack_end_magic(tsk
);
902 #ifdef CONFIG_STACKPROTECTOR
903 tsk
->stack_canary
= get_random_canary();
905 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
906 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
909 * One for us, one for whoever does the "release_task()" (usually
912 refcount_set(&tsk
->usage
, 2);
913 #ifdef CONFIG_BLK_DEV_IO_TRACE
916 tsk
->splice_pipe
= NULL
;
917 tsk
->task_frag
.page
= NULL
;
918 tsk
->wake_q
.next
= NULL
;
920 account_kernel_stack(tsk
, 1);
924 #ifdef CONFIG_FAULT_INJECTION
928 #ifdef CONFIG_BLK_CGROUP
929 tsk
->throttle_queue
= NULL
;
930 tsk
->use_memdelay
= 0;
934 tsk
->active_memcg
= NULL
;
939 free_thread_stack(tsk
);
941 free_task_struct(tsk
);
945 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
947 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
949 static int __init
coredump_filter_setup(char *s
)
951 default_dump_filter
=
952 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
953 MMF_DUMP_FILTER_MASK
;
957 __setup("coredump_filter=", coredump_filter_setup
);
959 #include <linux/init_task.h>
961 static void mm_init_aio(struct mm_struct
*mm
)
964 spin_lock_init(&mm
->ioctx_lock
);
965 mm
->ioctx_table
= NULL
;
969 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
970 struct task_struct
*p
)
974 WRITE_ONCE(mm
->owner
, NULL
);
978 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
985 static void mm_init_uprobes_state(struct mm_struct
*mm
)
987 #ifdef CONFIG_UPROBES
988 mm
->uprobes_state
.xol_area
= NULL
;
992 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
993 struct user_namespace
*user_ns
)
997 mm
->vmacache_seqnum
= 0;
998 atomic_set(&mm
->mm_users
, 1);
999 atomic_set(&mm
->mm_count
, 1);
1000 init_rwsem(&mm
->mmap_sem
);
1001 INIT_LIST_HEAD(&mm
->mmlist
);
1002 mm
->core_state
= NULL
;
1003 mm_pgtables_bytes_init(mm
);
1006 atomic64_set(&mm
->pinned_vm
, 0);
1007 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1008 spin_lock_init(&mm
->page_table_lock
);
1009 spin_lock_init(&mm
->arg_lock
);
1010 mm_init_cpumask(mm
);
1012 mm_init_owner(mm
, p
);
1013 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1014 mmu_notifier_mm_init(mm
);
1016 init_tlb_flush_pending(mm
);
1017 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1018 mm
->pmd_huge_pte
= NULL
;
1020 mm_init_uprobes_state(mm
);
1023 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1024 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1026 mm
->flags
= default_dump_filter
;
1030 if (mm_alloc_pgd(mm
))
1033 if (init_new_context(p
, mm
))
1034 goto fail_nocontext
;
1036 mm
->user_ns
= get_user_ns(user_ns
);
1047 * Allocate and initialize an mm_struct.
1049 struct mm_struct
*mm_alloc(void)
1051 struct mm_struct
*mm
;
1057 memset(mm
, 0, sizeof(*mm
));
1058 return mm_init(mm
, current
, current_user_ns());
1061 static inline void __mmput(struct mm_struct
*mm
)
1063 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1065 uprobe_clear_state(mm
);
1068 khugepaged_exit(mm
); /* must run before exit_mmap */
1070 mm_put_huge_zero_page(mm
);
1071 set_mm_exe_file(mm
, NULL
);
1072 if (!list_empty(&mm
->mmlist
)) {
1073 spin_lock(&mmlist_lock
);
1074 list_del(&mm
->mmlist
);
1075 spin_unlock(&mmlist_lock
);
1078 module_put(mm
->binfmt
->module
);
1083 * Decrement the use count and release all resources for an mm.
1085 void mmput(struct mm_struct
*mm
)
1089 if (atomic_dec_and_test(&mm
->mm_users
))
1092 EXPORT_SYMBOL_GPL(mmput
);
1095 static void mmput_async_fn(struct work_struct
*work
)
1097 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1103 void mmput_async(struct mm_struct
*mm
)
1105 if (atomic_dec_and_test(&mm
->mm_users
)) {
1106 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1107 schedule_work(&mm
->async_put_work
);
1110 EXPORT_SYMBOL(mmput_async
);
1114 * set_mm_exe_file - change a reference to the mm's executable file
1116 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1118 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1119 * invocations: in mmput() nobody alive left, in execve task is single
1120 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1121 * mm->exe_file, but does so without using set_mm_exe_file() in order
1122 * to do avoid the need for any locks.
1124 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1126 struct file
*old_exe_file
;
1129 * It is safe to dereference the exe_file without RCU as
1130 * this function is only called if nobody else can access
1131 * this mm -- see comment above for justification.
1133 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1136 get_file(new_exe_file
);
1137 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1143 * get_mm_exe_file - acquire a reference to the mm's executable file
1145 * Returns %NULL if mm has no associated executable file.
1146 * User must release file via fput().
1148 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1150 struct file
*exe_file
;
1153 exe_file
= rcu_dereference(mm
->exe_file
);
1154 if (exe_file
&& !get_file_rcu(exe_file
))
1159 EXPORT_SYMBOL(get_mm_exe_file
);
1162 * get_task_exe_file - acquire a reference to the task's executable file
1164 * Returns %NULL if task's mm (if any) has no associated executable file or
1165 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1166 * User must release file via fput().
1168 struct file
*get_task_exe_file(struct task_struct
*task
)
1170 struct file
*exe_file
= NULL
;
1171 struct mm_struct
*mm
;
1176 if (!(task
->flags
& PF_KTHREAD
))
1177 exe_file
= get_mm_exe_file(mm
);
1182 EXPORT_SYMBOL(get_task_exe_file
);
1185 * get_task_mm - acquire a reference to the task's mm
1187 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1188 * this kernel workthread has transiently adopted a user mm with use_mm,
1189 * to do its AIO) is not set and if so returns a reference to it, after
1190 * bumping up the use count. User must release the mm via mmput()
1191 * after use. Typically used by /proc and ptrace.
1193 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1195 struct mm_struct
*mm
;
1200 if (task
->flags
& PF_KTHREAD
)
1208 EXPORT_SYMBOL_GPL(get_task_mm
);
1210 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1212 struct mm_struct
*mm
;
1215 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1217 return ERR_PTR(err
);
1219 mm
= get_task_mm(task
);
1220 if (mm
&& mm
!= current
->mm
&&
1221 !ptrace_may_access(task
, mode
)) {
1223 mm
= ERR_PTR(-EACCES
);
1225 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1230 static void complete_vfork_done(struct task_struct
*tsk
)
1232 struct completion
*vfork
;
1235 vfork
= tsk
->vfork_done
;
1236 if (likely(vfork
)) {
1237 tsk
->vfork_done
= NULL
;
1243 static int wait_for_vfork_done(struct task_struct
*child
,
1244 struct completion
*vfork
)
1248 freezer_do_not_count();
1249 cgroup_enter_frozen();
1250 killed
= wait_for_completion_killable(vfork
);
1251 cgroup_leave_frozen(false);
1256 child
->vfork_done
= NULL
;
1260 put_task_struct(child
);
1264 /* Please note the differences between mmput and mm_release.
1265 * mmput is called whenever we stop holding onto a mm_struct,
1266 * error success whatever.
1268 * mm_release is called after a mm_struct has been removed
1269 * from the current process.
1271 * This difference is important for error handling, when we
1272 * only half set up a mm_struct for a new process and need to restore
1273 * the old one. Because we mmput the new mm_struct before
1274 * restoring the old one. . .
1275 * Eric Biederman 10 January 1998
1277 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1279 /* Get rid of any futexes when releasing the mm */
1281 if (unlikely(tsk
->robust_list
)) {
1282 exit_robust_list(tsk
);
1283 tsk
->robust_list
= NULL
;
1285 #ifdef CONFIG_COMPAT
1286 if (unlikely(tsk
->compat_robust_list
)) {
1287 compat_exit_robust_list(tsk
);
1288 tsk
->compat_robust_list
= NULL
;
1291 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
1292 exit_pi_state_list(tsk
);
1295 uprobe_free_utask(tsk
);
1297 /* Get rid of any cached register state */
1298 deactivate_mm(tsk
, mm
);
1301 * Signal userspace if we're not exiting with a core dump
1302 * because we want to leave the value intact for debugging
1305 if (tsk
->clear_child_tid
) {
1306 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1307 atomic_read(&mm
->mm_users
) > 1) {
1309 * We don't check the error code - if userspace has
1310 * not set up a proper pointer then tough luck.
1312 put_user(0, tsk
->clear_child_tid
);
1313 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1314 1, NULL
, NULL
, 0, 0);
1316 tsk
->clear_child_tid
= NULL
;
1320 * All done, finally we can wake up parent and return this mm to him.
1321 * Also kthread_stop() uses this completion for synchronization.
1323 if (tsk
->vfork_done
)
1324 complete_vfork_done(tsk
);
1328 * dup_mm() - duplicates an existing mm structure
1329 * @tsk: the task_struct with which the new mm will be associated.
1330 * @oldmm: the mm to duplicate.
1332 * Allocates a new mm structure and duplicates the provided @oldmm structure
1335 * Return: the duplicated mm or NULL on failure.
1337 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1338 struct mm_struct
*oldmm
)
1340 struct mm_struct
*mm
;
1347 memcpy(mm
, oldmm
, sizeof(*mm
));
1349 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1352 err
= dup_mmap(mm
, oldmm
);
1356 mm
->hiwater_rss
= get_mm_rss(mm
);
1357 mm
->hiwater_vm
= mm
->total_vm
;
1359 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1365 /* don't put binfmt in mmput, we haven't got module yet */
1367 mm_init_owner(mm
, NULL
);
1374 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1376 struct mm_struct
*mm
, *oldmm
;
1379 tsk
->min_flt
= tsk
->maj_flt
= 0;
1380 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1381 #ifdef CONFIG_DETECT_HUNG_TASK
1382 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1383 tsk
->last_switch_time
= 0;
1387 tsk
->active_mm
= NULL
;
1390 * Are we cloning a kernel thread?
1392 * We need to steal a active VM for that..
1394 oldmm
= current
->mm
;
1398 /* initialize the new vmacache entries */
1399 vmacache_flush(tsk
);
1401 if (clone_flags
& CLONE_VM
) {
1408 mm
= dup_mm(tsk
, current
->mm
);
1414 tsk
->active_mm
= mm
;
1421 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1423 struct fs_struct
*fs
= current
->fs
;
1424 if (clone_flags
& CLONE_FS
) {
1425 /* tsk->fs is already what we want */
1426 spin_lock(&fs
->lock
);
1428 spin_unlock(&fs
->lock
);
1432 spin_unlock(&fs
->lock
);
1435 tsk
->fs
= copy_fs_struct(fs
);
1441 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1443 struct files_struct
*oldf
, *newf
;
1447 * A background process may not have any files ...
1449 oldf
= current
->files
;
1453 if (clone_flags
& CLONE_FILES
) {
1454 atomic_inc(&oldf
->count
);
1458 newf
= dup_fd(oldf
, &error
);
1468 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1471 struct io_context
*ioc
= current
->io_context
;
1472 struct io_context
*new_ioc
;
1477 * Share io context with parent, if CLONE_IO is set
1479 if (clone_flags
& CLONE_IO
) {
1481 tsk
->io_context
= ioc
;
1482 } else if (ioprio_valid(ioc
->ioprio
)) {
1483 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1484 if (unlikely(!new_ioc
))
1487 new_ioc
->ioprio
= ioc
->ioprio
;
1488 put_io_context(new_ioc
);
1494 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1496 struct sighand_struct
*sig
;
1498 if (clone_flags
& CLONE_SIGHAND
) {
1499 refcount_inc(¤t
->sighand
->count
);
1502 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1503 rcu_assign_pointer(tsk
->sighand
, sig
);
1507 refcount_set(&sig
->count
, 1);
1508 spin_lock_irq(¤t
->sighand
->siglock
);
1509 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1510 spin_unlock_irq(¤t
->sighand
->siglock
);
1514 void __cleanup_sighand(struct sighand_struct
*sighand
)
1516 if (refcount_dec_and_test(&sighand
->count
)) {
1517 signalfd_cleanup(sighand
);
1519 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1520 * without an RCU grace period, see __lock_task_sighand().
1522 kmem_cache_free(sighand_cachep
, sighand
);
1526 #ifdef CONFIG_POSIX_TIMERS
1528 * Initialize POSIX timer handling for a thread group.
1530 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1532 unsigned long cpu_limit
;
1534 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1535 if (cpu_limit
!= RLIM_INFINITY
) {
1536 sig
->cputime_expires
.prof_exp
= cpu_limit
* NSEC_PER_SEC
;
1537 sig
->cputimer
.running
= true;
1540 /* The timer lists. */
1541 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1542 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1543 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1546 static inline void posix_cpu_timers_init_group(struct signal_struct
*sig
) { }
1549 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1551 struct signal_struct
*sig
;
1553 if (clone_flags
& CLONE_THREAD
)
1556 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1561 sig
->nr_threads
= 1;
1562 atomic_set(&sig
->live
, 1);
1563 refcount_set(&sig
->sigcnt
, 1);
1565 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1566 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1567 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1569 init_waitqueue_head(&sig
->wait_chldexit
);
1570 sig
->curr_target
= tsk
;
1571 init_sigpending(&sig
->shared_pending
);
1572 INIT_HLIST_HEAD(&sig
->multiprocess
);
1573 seqlock_init(&sig
->stats_lock
);
1574 prev_cputime_init(&sig
->prev_cputime
);
1576 #ifdef CONFIG_POSIX_TIMERS
1577 INIT_LIST_HEAD(&sig
->posix_timers
);
1578 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1579 sig
->real_timer
.function
= it_real_fn
;
1582 task_lock(current
->group_leader
);
1583 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1584 task_unlock(current
->group_leader
);
1586 posix_cpu_timers_init_group(sig
);
1588 tty_audit_fork(sig
);
1589 sched_autogroup_fork(sig
);
1591 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1592 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1594 mutex_init(&sig
->cred_guard_mutex
);
1599 static void copy_seccomp(struct task_struct
*p
)
1601 #ifdef CONFIG_SECCOMP
1603 * Must be called with sighand->lock held, which is common to
1604 * all threads in the group. Holding cred_guard_mutex is not
1605 * needed because this new task is not yet running and cannot
1608 assert_spin_locked(¤t
->sighand
->siglock
);
1610 /* Ref-count the new filter user, and assign it. */
1611 get_seccomp_filter(current
);
1612 p
->seccomp
= current
->seccomp
;
1615 * Explicitly enable no_new_privs here in case it got set
1616 * between the task_struct being duplicated and holding the
1617 * sighand lock. The seccomp state and nnp must be in sync.
1619 if (task_no_new_privs(current
))
1620 task_set_no_new_privs(p
);
1623 * If the parent gained a seccomp mode after copying thread
1624 * flags and between before we held the sighand lock, we have
1625 * to manually enable the seccomp thread flag here.
1627 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1628 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1632 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1634 current
->clear_child_tid
= tidptr
;
1636 return task_pid_vnr(current
);
1639 static void rt_mutex_init_task(struct task_struct
*p
)
1641 raw_spin_lock_init(&p
->pi_lock
);
1642 #ifdef CONFIG_RT_MUTEXES
1643 p
->pi_waiters
= RB_ROOT_CACHED
;
1644 p
->pi_top_task
= NULL
;
1645 p
->pi_blocked_on
= NULL
;
1649 #ifdef CONFIG_POSIX_TIMERS
1651 * Initialize POSIX timer handling for a single task.
1653 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1655 tsk
->cputime_expires
.prof_exp
= 0;
1656 tsk
->cputime_expires
.virt_exp
= 0;
1657 tsk
->cputime_expires
.sched_exp
= 0;
1658 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1659 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1660 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1663 static inline void posix_cpu_timers_init(struct task_struct
*tsk
) { }
1666 static inline void init_task_pid_links(struct task_struct
*task
)
1670 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1671 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1676 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1678 if (type
== PIDTYPE_PID
)
1679 task
->thread_pid
= pid
;
1681 task
->signal
->pids
[type
] = pid
;
1684 static inline void rcu_copy_process(struct task_struct
*p
)
1686 #ifdef CONFIG_PREEMPT_RCU
1687 p
->rcu_read_lock_nesting
= 0;
1688 p
->rcu_read_unlock_special
.s
= 0;
1689 p
->rcu_blocked_node
= NULL
;
1690 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1691 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1692 #ifdef CONFIG_TASKS_RCU
1693 p
->rcu_tasks_holdout
= false;
1694 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1695 p
->rcu_tasks_idle_cpu
= -1;
1696 #endif /* #ifdef CONFIG_TASKS_RCU */
1699 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1701 struct pid
*pid
= file
->private_data
;
1703 file
->private_data
= NULL
;
1708 #ifdef CONFIG_PROC_FS
1709 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1711 struct pid_namespace
*ns
= proc_pid_ns(file_inode(m
->file
));
1712 struct pid
*pid
= f
->private_data
;
1714 seq_put_decimal_ull(m
, "Pid:\t", pid_nr_ns(pid
, ns
));
1720 * Poll support for process exit notification.
1722 static unsigned int pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1724 struct task_struct
*task
;
1725 struct pid
*pid
= file
->private_data
;
1728 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1731 task
= pid_task(pid
, PIDTYPE_PID
);
1733 * Inform pollers only when the whole thread group exits.
1734 * If the thread group leader exits before all other threads in the
1735 * group, then poll(2) should block, similar to the wait(2) family.
1737 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1738 poll_flags
= POLLIN
| POLLRDNORM
;
1744 const struct file_operations pidfd_fops
= {
1745 .release
= pidfd_release
,
1747 #ifdef CONFIG_PROC_FS
1748 .show_fdinfo
= pidfd_show_fdinfo
,
1752 static void __delayed_free_task(struct rcu_head
*rhp
)
1754 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1759 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1761 if (IS_ENABLED(CONFIG_MEMCG
))
1762 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1768 * This creates a new process as a copy of the old one,
1769 * but does not actually start it yet.
1771 * It copies the registers, and all the appropriate
1772 * parts of the process environment (as per the clone
1773 * flags). The actual kick-off is left to the caller.
1775 static __latent_entropy
struct task_struct
*copy_process(
1779 struct kernel_clone_args
*args
)
1781 int pidfd
= -1, retval
;
1782 struct task_struct
*p
;
1783 struct multiprocess_signals delayed
;
1784 struct file
*pidfile
= NULL
;
1785 u64 clone_flags
= args
->flags
;
1788 * Don't allow sharing the root directory with processes in a different
1791 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1792 return ERR_PTR(-EINVAL
);
1794 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1795 return ERR_PTR(-EINVAL
);
1797 if ((clone_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
)
1798 if (!capable(CAP_SYS_ADMIN
))
1799 return ERR_PTR(-EPERM
);
1802 * Thread groups must share signals as well, and detached threads
1803 * can only be started up within the thread group.
1805 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1806 return ERR_PTR(-EINVAL
);
1809 * Shared signal handlers imply shared VM. By way of the above,
1810 * thread groups also imply shared VM. Blocking this case allows
1811 * for various simplifications in other code.
1813 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1814 return ERR_PTR(-EINVAL
);
1817 * Siblings of global init remain as zombies on exit since they are
1818 * not reaped by their parent (swapper). To solve this and to avoid
1819 * multi-rooted process trees, prevent global and container-inits
1820 * from creating siblings.
1822 if ((clone_flags
& CLONE_PARENT
) &&
1823 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1824 return ERR_PTR(-EINVAL
);
1827 * If the new process will be in a different pid or user namespace
1828 * do not allow it to share a thread group with the forking task.
1830 if (clone_flags
& CLONE_THREAD
) {
1831 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1832 (task_active_pid_ns(current
) !=
1833 current
->nsproxy
->pid_ns_for_children
))
1834 return ERR_PTR(-EINVAL
);
1837 if (clone_flags
& CLONE_PIDFD
) {
1839 * - CLONE_DETACHED is blocked so that we can potentially
1840 * reuse it later for CLONE_PIDFD.
1841 * - CLONE_THREAD is blocked until someone really needs it.
1843 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1844 return ERR_PTR(-EINVAL
);
1848 * Force any signals received before this point to be delivered
1849 * before the fork happens. Collect up signals sent to multiple
1850 * processes that happen during the fork and delay them so that
1851 * they appear to happen after the fork.
1853 sigemptyset(&delayed
.signal
);
1854 INIT_HLIST_NODE(&delayed
.node
);
1856 spin_lock_irq(¤t
->sighand
->siglock
);
1857 if (!(clone_flags
& CLONE_THREAD
))
1858 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1859 recalc_sigpending();
1860 spin_unlock_irq(¤t
->sighand
->siglock
);
1861 retval
= -ERESTARTNOINTR
;
1862 if (signal_pending(current
))
1866 p
= dup_task_struct(current
, node
);
1871 * This _must_ happen before we call free_task(), i.e. before we jump
1872 * to any of the bad_fork_* labels. This is to avoid freeing
1873 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1874 * kernel threads (PF_KTHREAD).
1876 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1878 * Clear TID on mm_release()?
1880 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1882 ftrace_graph_init_task(p
);
1884 rt_mutex_init_task(p
);
1886 #ifdef CONFIG_PROVE_LOCKING
1887 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1888 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1891 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1892 task_rlimit(p
, RLIMIT_NPROC
)) {
1893 if (p
->real_cred
->user
!= INIT_USER
&&
1894 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1897 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1899 retval
= copy_creds(p
, clone_flags
);
1904 * If multiple threads are within copy_process(), then this check
1905 * triggers too late. This doesn't hurt, the check is only there
1906 * to stop root fork bombs.
1909 if (nr_threads
>= max_threads
)
1910 goto bad_fork_cleanup_count
;
1912 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1913 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1914 p
->flags
|= PF_FORKNOEXEC
;
1915 INIT_LIST_HEAD(&p
->children
);
1916 INIT_LIST_HEAD(&p
->sibling
);
1917 rcu_copy_process(p
);
1918 p
->vfork_done
= NULL
;
1919 spin_lock_init(&p
->alloc_lock
);
1921 init_sigpending(&p
->pending
);
1923 p
->utime
= p
->stime
= p
->gtime
= 0;
1924 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1925 p
->utimescaled
= p
->stimescaled
= 0;
1927 prev_cputime_init(&p
->prev_cputime
);
1929 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1930 seqcount_init(&p
->vtime
.seqcount
);
1931 p
->vtime
.starttime
= 0;
1932 p
->vtime
.state
= VTIME_INACTIVE
;
1935 #if defined(SPLIT_RSS_COUNTING)
1936 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1939 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1945 task_io_accounting_init(&p
->ioac
);
1946 acct_clear_integrals(p
);
1948 posix_cpu_timers_init(p
);
1950 p
->io_context
= NULL
;
1951 audit_set_context(p
, NULL
);
1954 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1955 if (IS_ERR(p
->mempolicy
)) {
1956 retval
= PTR_ERR(p
->mempolicy
);
1957 p
->mempolicy
= NULL
;
1958 goto bad_fork_cleanup_threadgroup_lock
;
1961 #ifdef CONFIG_CPUSETS
1962 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1963 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1964 seqcount_init(&p
->mems_allowed_seq
);
1966 #ifdef CONFIG_TRACE_IRQFLAGS
1968 p
->hardirqs_enabled
= 0;
1969 p
->hardirq_enable_ip
= 0;
1970 p
->hardirq_enable_event
= 0;
1971 p
->hardirq_disable_ip
= _THIS_IP_
;
1972 p
->hardirq_disable_event
= 0;
1973 p
->softirqs_enabled
= 1;
1974 p
->softirq_enable_ip
= _THIS_IP_
;
1975 p
->softirq_enable_event
= 0;
1976 p
->softirq_disable_ip
= 0;
1977 p
->softirq_disable_event
= 0;
1978 p
->hardirq_context
= 0;
1979 p
->softirq_context
= 0;
1982 p
->pagefault_disabled
= 0;
1984 #ifdef CONFIG_LOCKDEP
1985 lockdep_init_task(p
);
1988 #ifdef CONFIG_DEBUG_MUTEXES
1989 p
->blocked_on
= NULL
; /* not blocked yet */
1991 #ifdef CONFIG_BCACHE
1992 p
->sequential_io
= 0;
1993 p
->sequential_io_avg
= 0;
1996 /* Perform scheduler related setup. Assign this task to a CPU. */
1997 retval
= sched_fork(clone_flags
, p
);
1999 goto bad_fork_cleanup_policy
;
2001 retval
= perf_event_init_task(p
);
2003 goto bad_fork_cleanup_policy
;
2004 retval
= audit_alloc(p
);
2006 goto bad_fork_cleanup_perf
;
2007 /* copy all the process information */
2009 retval
= security_task_alloc(p
, clone_flags
);
2011 goto bad_fork_cleanup_audit
;
2012 retval
= copy_semundo(clone_flags
, p
);
2014 goto bad_fork_cleanup_security
;
2015 retval
= copy_files(clone_flags
, p
);
2017 goto bad_fork_cleanup_semundo
;
2018 retval
= copy_fs(clone_flags
, p
);
2020 goto bad_fork_cleanup_files
;
2021 retval
= copy_sighand(clone_flags
, p
);
2023 goto bad_fork_cleanup_fs
;
2024 retval
= copy_signal(clone_flags
, p
);
2026 goto bad_fork_cleanup_sighand
;
2027 retval
= copy_mm(clone_flags
, p
);
2029 goto bad_fork_cleanup_signal
;
2030 retval
= copy_namespaces(clone_flags
, p
);
2032 goto bad_fork_cleanup_mm
;
2033 retval
= copy_io(clone_flags
, p
);
2035 goto bad_fork_cleanup_namespaces
;
2036 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2039 goto bad_fork_cleanup_io
;
2041 stackleak_task_init(p
);
2043 if (pid
!= &init_struct_pid
) {
2044 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
2046 retval
= PTR_ERR(pid
);
2047 goto bad_fork_cleanup_thread
;
2052 * This has to happen after we've potentially unshared the file
2053 * descriptor table (so that the pidfd doesn't leak into the child
2054 * if the fd table isn't shared).
2056 if (clone_flags
& CLONE_PIDFD
) {
2057 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2059 goto bad_fork_free_pid
;
2063 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2064 O_RDWR
| O_CLOEXEC
);
2065 if (IS_ERR(pidfile
)) {
2066 put_unused_fd(pidfd
);
2067 retval
= PTR_ERR(pidfile
);
2068 goto bad_fork_free_pid
;
2070 get_pid(pid
); /* held by pidfile now */
2072 retval
= put_user(pidfd
, args
->pidfd
);
2074 goto bad_fork_put_pidfd
;
2081 p
->robust_list
= NULL
;
2082 #ifdef CONFIG_COMPAT
2083 p
->compat_robust_list
= NULL
;
2085 INIT_LIST_HEAD(&p
->pi_state_list
);
2086 p
->pi_state_cache
= NULL
;
2089 * sigaltstack should be cleared when sharing the same VM
2091 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2095 * Syscall tracing and stepping should be turned off in the
2096 * child regardless of CLONE_PTRACE.
2098 user_disable_single_step(p
);
2099 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2100 #ifdef TIF_SYSCALL_EMU
2101 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2103 clear_tsk_latency_tracing(p
);
2105 /* ok, now we should be set up.. */
2106 p
->pid
= pid_nr(pid
);
2107 if (clone_flags
& CLONE_THREAD
) {
2108 p
->exit_signal
= -1;
2109 p
->group_leader
= current
->group_leader
;
2110 p
->tgid
= current
->tgid
;
2112 if (clone_flags
& CLONE_PARENT
)
2113 p
->exit_signal
= current
->group_leader
->exit_signal
;
2115 p
->exit_signal
= args
->exit_signal
;
2116 p
->group_leader
= p
;
2121 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2122 p
->dirty_paused_when
= 0;
2124 p
->pdeath_signal
= 0;
2125 INIT_LIST_HEAD(&p
->thread_group
);
2126 p
->task_works
= NULL
;
2128 cgroup_threadgroup_change_begin(current
);
2130 * Ensure that the cgroup subsystem policies allow the new process to be
2131 * forked. It should be noted the the new process's css_set can be changed
2132 * between here and cgroup_post_fork() if an organisation operation is in
2135 retval
= cgroup_can_fork(p
);
2137 goto bad_fork_cgroup_threadgroup_change_end
;
2140 * From this point on we must avoid any synchronous user-space
2141 * communication until we take the tasklist-lock. In particular, we do
2142 * not want user-space to be able to predict the process start-time by
2143 * stalling fork(2) after we recorded the start_time but before it is
2144 * visible to the system.
2147 p
->start_time
= ktime_get_ns();
2148 p
->real_start_time
= ktime_get_boottime_ns();
2151 * Make it visible to the rest of the system, but dont wake it up yet.
2152 * Need tasklist lock for parent etc handling!
2154 write_lock_irq(&tasklist_lock
);
2156 /* CLONE_PARENT re-uses the old parent */
2157 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2158 p
->real_parent
= current
->real_parent
;
2159 p
->parent_exec_id
= current
->parent_exec_id
;
2161 p
->real_parent
= current
;
2162 p
->parent_exec_id
= current
->self_exec_id
;
2165 klp_copy_process(p
);
2167 spin_lock(¤t
->sighand
->siglock
);
2170 * Copy seccomp details explicitly here, in case they were changed
2171 * before holding sighand lock.
2175 rseq_fork(p
, clone_flags
);
2177 /* Don't start children in a dying pid namespace */
2178 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2180 goto bad_fork_cancel_cgroup
;
2183 /* Let kill terminate clone/fork in the middle */
2184 if (fatal_signal_pending(current
)) {
2186 goto bad_fork_cancel_cgroup
;
2189 /* past the last point of failure */
2191 fd_install(pidfd
, pidfile
);
2193 init_task_pid_links(p
);
2194 if (likely(p
->pid
)) {
2195 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2197 init_task_pid(p
, PIDTYPE_PID
, pid
);
2198 if (thread_group_leader(p
)) {
2199 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2200 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2201 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2203 if (is_child_reaper(pid
)) {
2204 ns_of_pid(pid
)->child_reaper
= p
;
2205 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2207 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2208 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2210 * Inherit has_child_subreaper flag under the same
2211 * tasklist_lock with adding child to the process tree
2212 * for propagate_has_child_subreaper optimization.
2214 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2215 p
->real_parent
->signal
->is_child_subreaper
;
2216 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2217 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2218 attach_pid(p
, PIDTYPE_TGID
);
2219 attach_pid(p
, PIDTYPE_PGID
);
2220 attach_pid(p
, PIDTYPE_SID
);
2221 __this_cpu_inc(process_counts
);
2223 current
->signal
->nr_threads
++;
2224 atomic_inc(¤t
->signal
->live
);
2225 refcount_inc(¤t
->signal
->sigcnt
);
2226 task_join_group_stop(p
);
2227 list_add_tail_rcu(&p
->thread_group
,
2228 &p
->group_leader
->thread_group
);
2229 list_add_tail_rcu(&p
->thread_node
,
2230 &p
->signal
->thread_head
);
2232 attach_pid(p
, PIDTYPE_PID
);
2236 hlist_del_init(&delayed
.node
);
2237 spin_unlock(¤t
->sighand
->siglock
);
2238 syscall_tracepoint_update(p
);
2239 write_unlock_irq(&tasklist_lock
);
2241 proc_fork_connector(p
);
2242 cgroup_post_fork(p
);
2243 cgroup_threadgroup_change_end(current
);
2246 trace_task_newtask(p
, clone_flags
);
2247 uprobe_copy_process(p
, clone_flags
);
2251 bad_fork_cancel_cgroup
:
2252 spin_unlock(¤t
->sighand
->siglock
);
2253 write_unlock_irq(&tasklist_lock
);
2254 cgroup_cancel_fork(p
);
2255 bad_fork_cgroup_threadgroup_change_end
:
2256 cgroup_threadgroup_change_end(current
);
2258 if (clone_flags
& CLONE_PIDFD
) {
2260 put_unused_fd(pidfd
);
2263 if (pid
!= &init_struct_pid
)
2265 bad_fork_cleanup_thread
:
2267 bad_fork_cleanup_io
:
2270 bad_fork_cleanup_namespaces
:
2271 exit_task_namespaces(p
);
2272 bad_fork_cleanup_mm
:
2274 mm_clear_owner(p
->mm
, p
);
2277 bad_fork_cleanup_signal
:
2278 if (!(clone_flags
& CLONE_THREAD
))
2279 free_signal_struct(p
->signal
);
2280 bad_fork_cleanup_sighand
:
2281 __cleanup_sighand(p
->sighand
);
2282 bad_fork_cleanup_fs
:
2283 exit_fs(p
); /* blocking */
2284 bad_fork_cleanup_files
:
2285 exit_files(p
); /* blocking */
2286 bad_fork_cleanup_semundo
:
2288 bad_fork_cleanup_security
:
2289 security_task_free(p
);
2290 bad_fork_cleanup_audit
:
2292 bad_fork_cleanup_perf
:
2293 perf_event_free_task(p
);
2294 bad_fork_cleanup_policy
:
2295 lockdep_free_task(p
);
2297 mpol_put(p
->mempolicy
);
2298 bad_fork_cleanup_threadgroup_lock
:
2300 delayacct_tsk_free(p
);
2301 bad_fork_cleanup_count
:
2302 atomic_dec(&p
->cred
->user
->processes
);
2305 p
->state
= TASK_DEAD
;
2307 delayed_free_task(p
);
2309 spin_lock_irq(¤t
->sighand
->siglock
);
2310 hlist_del_init(&delayed
.node
);
2311 spin_unlock_irq(¤t
->sighand
->siglock
);
2312 return ERR_PTR(retval
);
2315 static inline void init_idle_pids(struct task_struct
*idle
)
2319 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2320 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2321 init_task_pid(idle
, type
, &init_struct_pid
);
2325 struct task_struct
*fork_idle(int cpu
)
2327 struct task_struct
*task
;
2328 struct kernel_clone_args args
= {
2332 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2333 if (!IS_ERR(task
)) {
2334 init_idle_pids(task
);
2335 init_idle(task
, cpu
);
2341 struct mm_struct
*copy_init_mm(void)
2343 return dup_mm(NULL
, &init_mm
);
2347 * Ok, this is the main fork-routine.
2349 * It copies the process, and if successful kick-starts
2350 * it and waits for it to finish using the VM if required.
2352 * args->exit_signal is expected to be checked for sanity by the caller.
2354 long _do_fork(struct kernel_clone_args
*args
)
2356 u64 clone_flags
= args
->flags
;
2357 struct completion vfork
;
2359 struct task_struct
*p
;
2364 * Determine whether and which event to report to ptracer. When
2365 * called from kernel_thread or CLONE_UNTRACED is explicitly
2366 * requested, no event is reported; otherwise, report if the event
2367 * for the type of forking is enabled.
2369 if (!(clone_flags
& CLONE_UNTRACED
)) {
2370 if (clone_flags
& CLONE_VFORK
)
2371 trace
= PTRACE_EVENT_VFORK
;
2372 else if (args
->exit_signal
!= SIGCHLD
)
2373 trace
= PTRACE_EVENT_CLONE
;
2375 trace
= PTRACE_EVENT_FORK
;
2377 if (likely(!ptrace_event_enabled(current
, trace
)))
2381 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2382 add_latent_entropy();
2388 * Do this prior waking up the new thread - the thread pointer
2389 * might get invalid after that point, if the thread exits quickly.
2391 trace_sched_process_fork(current
, p
);
2393 pid
= get_task_pid(p
, PIDTYPE_PID
);
2396 if (clone_flags
& CLONE_PARENT_SETTID
)
2397 put_user(nr
, args
->parent_tid
);
2399 if (clone_flags
& CLONE_VFORK
) {
2400 p
->vfork_done
= &vfork
;
2401 init_completion(&vfork
);
2405 wake_up_new_task(p
);
2407 /* forking complete and child started to run, tell ptracer */
2408 if (unlikely(trace
))
2409 ptrace_event_pid(trace
, pid
);
2411 if (clone_flags
& CLONE_VFORK
) {
2412 if (!wait_for_vfork_done(p
, &vfork
))
2413 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2420 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2422 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2423 if ((kargs
->flags
& CLONE_PIDFD
) &&
2424 (kargs
->flags
& CLONE_PARENT_SETTID
))
2430 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2431 /* For compatibility with architectures that call do_fork directly rather than
2432 * using the syscall entry points below. */
2433 long do_fork(unsigned long clone_flags
,
2434 unsigned long stack_start
,
2435 unsigned long stack_size
,
2436 int __user
*parent_tidptr
,
2437 int __user
*child_tidptr
)
2439 struct kernel_clone_args args
= {
2440 .flags
= (clone_flags
& ~CSIGNAL
),
2441 .pidfd
= parent_tidptr
,
2442 .child_tid
= child_tidptr
,
2443 .parent_tid
= parent_tidptr
,
2444 .exit_signal
= (clone_flags
& CSIGNAL
),
2445 .stack
= stack_start
,
2446 .stack_size
= stack_size
,
2449 if (!legacy_clone_args_valid(&args
))
2452 return _do_fork(&args
);
2457 * Create a kernel thread.
2459 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2461 struct kernel_clone_args args
= {
2462 .flags
= ((flags
| CLONE_VM
| CLONE_UNTRACED
) & ~CSIGNAL
),
2463 .exit_signal
= (flags
& CSIGNAL
),
2464 .stack
= (unsigned long)fn
,
2465 .stack_size
= (unsigned long)arg
,
2468 return _do_fork(&args
);
2471 #ifdef __ARCH_WANT_SYS_FORK
2472 SYSCALL_DEFINE0(fork
)
2475 struct kernel_clone_args args
= {
2476 .exit_signal
= SIGCHLD
,
2479 return _do_fork(&args
);
2481 /* can not support in nommu mode */
2487 #ifdef __ARCH_WANT_SYS_VFORK
2488 SYSCALL_DEFINE0(vfork
)
2490 struct kernel_clone_args args
= {
2491 .flags
= CLONE_VFORK
| CLONE_VM
,
2492 .exit_signal
= SIGCHLD
,
2495 return _do_fork(&args
);
2499 #ifdef __ARCH_WANT_SYS_CLONE
2500 #ifdef CONFIG_CLONE_BACKWARDS
2501 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2502 int __user
*, parent_tidptr
,
2504 int __user
*, child_tidptr
)
2505 #elif defined(CONFIG_CLONE_BACKWARDS2)
2506 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2507 int __user
*, parent_tidptr
,
2508 int __user
*, child_tidptr
,
2510 #elif defined(CONFIG_CLONE_BACKWARDS3)
2511 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2513 int __user
*, parent_tidptr
,
2514 int __user
*, child_tidptr
,
2517 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2518 int __user
*, parent_tidptr
,
2519 int __user
*, child_tidptr
,
2523 struct kernel_clone_args args
= {
2524 .flags
= (clone_flags
& ~CSIGNAL
),
2525 .pidfd
= parent_tidptr
,
2526 .child_tid
= child_tidptr
,
2527 .parent_tid
= parent_tidptr
,
2528 .exit_signal
= (clone_flags
& CSIGNAL
),
2533 if (!legacy_clone_args_valid(&args
))
2536 return _do_fork(&args
);
2540 #ifdef __ARCH_WANT_SYS_CLONE3
2541 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2542 struct clone_args __user
*uargs
,
2545 struct clone_args args
;
2547 if (unlikely(size
> PAGE_SIZE
))
2550 if (unlikely(size
< sizeof(struct clone_args
)))
2553 if (unlikely(!access_ok(uargs
, size
)))
2556 if (size
> sizeof(struct clone_args
)) {
2557 unsigned char __user
*addr
;
2558 unsigned char __user
*end
;
2561 addr
= (void __user
*)uargs
+ sizeof(struct clone_args
);
2562 end
= (void __user
*)uargs
+ size
;
2564 for (; addr
< end
; addr
++) {
2565 if (get_user(val
, addr
))
2571 size
= sizeof(struct clone_args
);
2574 if (copy_from_user(&args
, uargs
, size
))
2578 * Verify that higher 32bits of exit_signal are unset and that
2579 * it is a valid signal
2581 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2582 !valid_signal(args
.exit_signal
)))
2585 *kargs
= (struct kernel_clone_args
){
2586 .flags
= args
.flags
,
2587 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2588 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2589 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2590 .exit_signal
= args
.exit_signal
,
2591 .stack
= args
.stack
,
2592 .stack_size
= args
.stack_size
,
2600 * clone3_stack_valid - check and prepare stack
2601 * @kargs: kernel clone args
2603 * Verify that the stack arguments userspace gave us are sane.
2604 * In addition, set the stack direction for userspace since it's easy for us to
2607 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2609 if (kargs
->stack
== 0) {
2610 if (kargs
->stack_size
> 0)
2613 if (kargs
->stack_size
== 0)
2616 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2619 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2620 kargs
->stack
+= kargs
->stack_size
;
2627 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2630 * All lower bits of the flag word are taken.
2631 * Verify that no other unknown flags are passed along.
2633 if (kargs
->flags
& ~CLONE_LEGACY_FLAGS
)
2637 * - make the CLONE_DETACHED bit reuseable for clone3
2638 * - make the CSIGNAL bits reuseable for clone3
2640 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2643 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2647 if (!clone3_stack_valid(kargs
))
2653 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2657 struct kernel_clone_args kargs
;
2659 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2663 if (!clone3_args_valid(&kargs
))
2666 return _do_fork(&kargs
);
2670 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2672 struct task_struct
*leader
, *parent
, *child
;
2675 read_lock(&tasklist_lock
);
2676 leader
= top
= top
->group_leader
;
2678 for_each_thread(leader
, parent
) {
2679 list_for_each_entry(child
, &parent
->children
, sibling
) {
2680 res
= visitor(child
, data
);
2692 if (leader
!= top
) {
2694 parent
= child
->real_parent
;
2695 leader
= parent
->group_leader
;
2699 read_unlock(&tasklist_lock
);
2702 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2703 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2706 static void sighand_ctor(void *data
)
2708 struct sighand_struct
*sighand
= data
;
2710 spin_lock_init(&sighand
->siglock
);
2711 init_waitqueue_head(&sighand
->signalfd_wqh
);
2714 void __init
proc_caches_init(void)
2716 unsigned int mm_size
;
2718 sighand_cachep
= kmem_cache_create("sighand_cache",
2719 sizeof(struct sighand_struct
), 0,
2720 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2721 SLAB_ACCOUNT
, sighand_ctor
);
2722 signal_cachep
= kmem_cache_create("signal_cache",
2723 sizeof(struct signal_struct
), 0,
2724 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2726 files_cachep
= kmem_cache_create("files_cache",
2727 sizeof(struct files_struct
), 0,
2728 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2730 fs_cachep
= kmem_cache_create("fs_cache",
2731 sizeof(struct fs_struct
), 0,
2732 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2736 * The mm_cpumask is located at the end of mm_struct, and is
2737 * dynamically sized based on the maximum CPU number this system
2738 * can have, taking hotplug into account (nr_cpu_ids).
2740 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2742 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2743 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2744 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2745 offsetof(struct mm_struct
, saved_auxv
),
2746 sizeof_field(struct mm_struct
, saved_auxv
),
2748 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2750 nsproxy_cache_init();
2754 * Check constraints on flags passed to the unshare system call.
2756 static int check_unshare_flags(unsigned long unshare_flags
)
2758 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2759 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2760 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2761 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2764 * Not implemented, but pretend it works if there is nothing
2765 * to unshare. Note that unsharing the address space or the
2766 * signal handlers also need to unshare the signal queues (aka
2769 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2770 if (!thread_group_empty(current
))
2773 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2774 if (refcount_read(¤t
->sighand
->count
) > 1)
2777 if (unshare_flags
& CLONE_VM
) {
2778 if (!current_is_single_threaded())
2786 * Unshare the filesystem structure if it is being shared
2788 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2790 struct fs_struct
*fs
= current
->fs
;
2792 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2795 /* don't need lock here; in the worst case we'll do useless copy */
2799 *new_fsp
= copy_fs_struct(fs
);
2807 * Unshare file descriptor table if it is being shared
2809 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2811 struct files_struct
*fd
= current
->files
;
2814 if ((unshare_flags
& CLONE_FILES
) &&
2815 (fd
&& atomic_read(&fd
->count
) > 1)) {
2816 *new_fdp
= dup_fd(fd
, &error
);
2825 * unshare allows a process to 'unshare' part of the process
2826 * context which was originally shared using clone. copy_*
2827 * functions used by do_fork() cannot be used here directly
2828 * because they modify an inactive task_struct that is being
2829 * constructed. Here we are modifying the current, active,
2832 int ksys_unshare(unsigned long unshare_flags
)
2834 struct fs_struct
*fs
, *new_fs
= NULL
;
2835 struct files_struct
*fd
, *new_fd
= NULL
;
2836 struct cred
*new_cred
= NULL
;
2837 struct nsproxy
*new_nsproxy
= NULL
;
2842 * If unsharing a user namespace must also unshare the thread group
2843 * and unshare the filesystem root and working directories.
2845 if (unshare_flags
& CLONE_NEWUSER
)
2846 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2848 * If unsharing vm, must also unshare signal handlers.
2850 if (unshare_flags
& CLONE_VM
)
2851 unshare_flags
|= CLONE_SIGHAND
;
2853 * If unsharing a signal handlers, must also unshare the signal queues.
2855 if (unshare_flags
& CLONE_SIGHAND
)
2856 unshare_flags
|= CLONE_THREAD
;
2858 * If unsharing namespace, must also unshare filesystem information.
2860 if (unshare_flags
& CLONE_NEWNS
)
2861 unshare_flags
|= CLONE_FS
;
2863 if ((unshare_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
) {
2865 if (!capable(CAP_SYS_ADMIN
))
2866 goto bad_unshare_out
;
2869 err
= check_unshare_flags(unshare_flags
);
2871 goto bad_unshare_out
;
2873 * CLONE_NEWIPC must also detach from the undolist: after switching
2874 * to a new ipc namespace, the semaphore arrays from the old
2875 * namespace are unreachable.
2877 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2879 err
= unshare_fs(unshare_flags
, &new_fs
);
2881 goto bad_unshare_out
;
2882 err
= unshare_fd(unshare_flags
, &new_fd
);
2884 goto bad_unshare_cleanup_fs
;
2885 err
= unshare_userns(unshare_flags
, &new_cred
);
2887 goto bad_unshare_cleanup_fd
;
2888 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2891 goto bad_unshare_cleanup_cred
;
2893 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2896 * CLONE_SYSVSEM is equivalent to sys_exit().
2900 if (unshare_flags
& CLONE_NEWIPC
) {
2901 /* Orphan segments in old ns (see sem above). */
2903 shm_init_task(current
);
2907 switch_task_namespaces(current
, new_nsproxy
);
2913 spin_lock(&fs
->lock
);
2914 current
->fs
= new_fs
;
2919 spin_unlock(&fs
->lock
);
2923 fd
= current
->files
;
2924 current
->files
= new_fd
;
2928 task_unlock(current
);
2931 /* Install the new user namespace */
2932 commit_creds(new_cred
);
2937 perf_event_namespaces(current
);
2939 bad_unshare_cleanup_cred
:
2942 bad_unshare_cleanup_fd
:
2944 put_files_struct(new_fd
);
2946 bad_unshare_cleanup_fs
:
2948 free_fs_struct(new_fs
);
2954 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2956 return ksys_unshare(unshare_flags
);
2960 * Helper to unshare the files of the current task.
2961 * We don't want to expose copy_files internals to
2962 * the exec layer of the kernel.
2965 int unshare_files(struct files_struct
**displaced
)
2967 struct task_struct
*task
= current
;
2968 struct files_struct
*copy
= NULL
;
2971 error
= unshare_fd(CLONE_FILES
, ©
);
2972 if (error
|| !copy
) {
2976 *displaced
= task
->files
;
2983 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2984 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2988 int threads
= max_threads
;
2990 int max
= MAX_THREADS
;
2997 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3001 max_threads
= threads
;