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/mm_inline.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/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>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.h>
100 #include <linux/sched/mm.h>
102 #include <asm/pgalloc.h>
103 #include <linux/uaccess.h>
104 #include <asm/mmu_context.h>
105 #include <asm/cacheflush.h>
106 #include <asm/tlbflush.h>
108 #include <trace/events/sched.h>
110 #define CREATE_TRACE_POINTS
111 #include <trace/events/task.h>
114 * Minimum number of threads to boot the kernel
116 #define MIN_THREADS 20
119 * Maximum number of threads
121 #define MAX_THREADS FUTEX_TID_MASK
124 * Protected counters by write_lock_irq(&tasklist_lock)
126 unsigned long total_forks
; /* Handle normal Linux uptimes. */
127 int nr_threads
; /* The idle threads do not count.. */
129 static int max_threads
; /* tunable limit on nr_threads */
131 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
133 static const char * const resident_page_types
[] = {
134 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
135 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
136 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
137 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
140 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
142 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
144 #ifdef CONFIG_PROVE_RCU
145 int lockdep_tasklist_lock_is_held(void)
147 return lockdep_is_held(&tasklist_lock
);
149 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
150 #endif /* #ifdef CONFIG_PROVE_RCU */
152 int nr_processes(void)
157 for_each_possible_cpu(cpu
)
158 total
+= per_cpu(process_counts
, cpu
);
163 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
167 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
168 static struct kmem_cache
*task_struct_cachep
;
170 static inline struct task_struct
*alloc_task_struct_node(int node
)
172 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
175 static inline void free_task_struct(struct task_struct
*tsk
)
177 kmem_cache_free(task_struct_cachep
, tsk
);
181 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
184 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
185 * kmemcache based allocator.
187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189 # ifdef CONFIG_VMAP_STACK
191 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
192 * flush. Try to minimize the number of calls by caching stacks.
194 #define NR_CACHED_STACKS 2
195 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
199 struct vm_struct
*stack_vm_area
;
202 static bool try_release_thread_stack_to_cache(struct vm_struct
*vm
)
206 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
207 if (this_cpu_cmpxchg(cached_stacks
[i
], NULL
, vm
) != NULL
)
214 static void thread_stack_free_rcu(struct rcu_head
*rh
)
216 struct vm_stack
*vm_stack
= container_of(rh
, struct vm_stack
, rcu
);
218 if (try_release_thread_stack_to_cache(vm_stack
->stack_vm_area
))
224 static void thread_stack_delayed_free(struct task_struct
*tsk
)
226 struct vm_stack
*vm_stack
= tsk
->stack
;
228 vm_stack
->stack_vm_area
= tsk
->stack_vm_area
;
229 call_rcu(&vm_stack
->rcu
, thread_stack_free_rcu
);
232 static int free_vm_stack_cache(unsigned int cpu
)
234 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
237 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
238 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
243 vfree(vm_stack
->addr
);
244 cached_vm_stacks
[i
] = NULL
;
250 static int memcg_charge_kernel_stack(struct vm_struct
*vm
)
255 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
256 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
258 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
259 ret
= memcg_kmem_charge_page(vm
->pages
[i
], GFP_KERNEL
, 0);
266 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
267 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
270 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
271 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
275 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
277 struct vm_struct
*vm
;
281 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
284 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
289 /* Reset stack metadata. */
290 kasan_unpoison_range(s
->addr
, THREAD_SIZE
);
292 stack
= kasan_reset_tag(s
->addr
);
294 /* Clear stale pointers from reused stack. */
295 memset(stack
, 0, THREAD_SIZE
);
297 if (memcg_charge_kernel_stack(s
)) {
302 tsk
->stack_vm_area
= s
;
308 * Allocated stacks are cached and later reused by new threads,
309 * so memcg accounting is performed manually on assigning/releasing
310 * stacks to tasks. Drop __GFP_ACCOUNT.
312 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
313 VMALLOC_START
, VMALLOC_END
,
314 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
316 0, node
, __builtin_return_address(0));
320 vm
= find_vm_area(stack
);
321 if (memcg_charge_kernel_stack(vm
)) {
326 * We can't call find_vm_area() in interrupt context, and
327 * free_thread_stack() can be called in interrupt context,
328 * so cache the vm_struct.
330 tsk
->stack_vm_area
= vm
;
331 stack
= kasan_reset_tag(stack
);
336 static void free_thread_stack(struct task_struct
*tsk
)
338 if (!try_release_thread_stack_to_cache(tsk
->stack_vm_area
))
339 thread_stack_delayed_free(tsk
);
342 tsk
->stack_vm_area
= NULL
;
345 # else /* !CONFIG_VMAP_STACK */
347 static void thread_stack_free_rcu(struct rcu_head
*rh
)
349 __free_pages(virt_to_page(rh
), THREAD_SIZE_ORDER
);
352 static void thread_stack_delayed_free(struct task_struct
*tsk
)
354 struct rcu_head
*rh
= tsk
->stack
;
356 call_rcu(rh
, thread_stack_free_rcu
);
359 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
361 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
365 tsk
->stack
= kasan_reset_tag(page_address(page
));
371 static void free_thread_stack(struct task_struct
*tsk
)
373 thread_stack_delayed_free(tsk
);
377 # endif /* CONFIG_VMAP_STACK */
378 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
380 static struct kmem_cache
*thread_stack_cache
;
382 static void thread_stack_free_rcu(struct rcu_head
*rh
)
384 kmem_cache_free(thread_stack_cache
, rh
);
387 static void thread_stack_delayed_free(struct task_struct
*tsk
)
389 struct rcu_head
*rh
= tsk
->stack
;
391 call_rcu(rh
, thread_stack_free_rcu
);
394 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
396 unsigned long *stack
;
397 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
398 stack
= kasan_reset_tag(stack
);
400 return stack
? 0 : -ENOMEM
;
403 static void free_thread_stack(struct task_struct
*tsk
)
405 thread_stack_delayed_free(tsk
);
409 void thread_stack_cache_init(void)
411 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
412 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
414 BUG_ON(thread_stack_cache
== NULL
);
417 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
418 #else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
420 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
422 unsigned long *stack
;
424 stack
= arch_alloc_thread_stack_node(tsk
, node
);
426 return stack
? 0 : -ENOMEM
;
429 static void free_thread_stack(struct task_struct
*tsk
)
431 arch_free_thread_stack(tsk
);
435 #endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
437 /* SLAB cache for signal_struct structures (tsk->signal) */
438 static struct kmem_cache
*signal_cachep
;
440 /* SLAB cache for sighand_struct structures (tsk->sighand) */
441 struct kmem_cache
*sighand_cachep
;
443 /* SLAB cache for files_struct structures (tsk->files) */
444 struct kmem_cache
*files_cachep
;
446 /* SLAB cache for fs_struct structures (tsk->fs) */
447 struct kmem_cache
*fs_cachep
;
449 /* SLAB cache for vm_area_struct structures */
450 static struct kmem_cache
*vm_area_cachep
;
452 /* SLAB cache for mm_struct structures (tsk->mm) */
453 static struct kmem_cache
*mm_cachep
;
455 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
457 struct vm_area_struct
*vma
;
459 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
465 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
467 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
470 ASSERT_EXCLUSIVE_WRITER(orig
->vm_flags
);
471 ASSERT_EXCLUSIVE_WRITER(orig
->vm_file
);
473 * orig->shared.rb may be modified concurrently, but the clone
474 * will be reinitialized.
476 *new = data_race(*orig
);
477 INIT_LIST_HEAD(&new->anon_vma_chain
);
478 new->vm_next
= new->vm_prev
= NULL
;
479 dup_anon_vma_name(orig
, new);
484 void vm_area_free(struct vm_area_struct
*vma
)
486 free_anon_vma_name(vma
);
487 kmem_cache_free(vm_area_cachep
, vma
);
490 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
492 if (IS_ENABLED(CONFIG_VMAP_STACK
)) {
493 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
496 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
497 mod_lruvec_page_state(vm
->pages
[i
], NR_KERNEL_STACK_KB
,
498 account
* (PAGE_SIZE
/ 1024));
500 void *stack
= task_stack_page(tsk
);
502 /* All stack pages are in the same node. */
503 mod_lruvec_kmem_state(stack
, NR_KERNEL_STACK_KB
,
504 account
* (THREAD_SIZE
/ 1024));
508 void exit_task_stack_account(struct task_struct
*tsk
)
510 account_kernel_stack(tsk
, -1);
512 if (IS_ENABLED(CONFIG_VMAP_STACK
)) {
513 struct vm_struct
*vm
;
516 vm
= task_stack_vm_area(tsk
);
517 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
518 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
522 static void release_task_stack(struct task_struct
*tsk
)
524 if (WARN_ON(READ_ONCE(tsk
->__state
) != TASK_DEAD
))
525 return; /* Better to leak the stack than to free prematurely */
527 free_thread_stack(tsk
);
530 #ifdef CONFIG_THREAD_INFO_IN_TASK
531 void put_task_stack(struct task_struct
*tsk
)
533 if (refcount_dec_and_test(&tsk
->stack_refcount
))
534 release_task_stack(tsk
);
538 void free_task(struct task_struct
*tsk
)
540 release_user_cpus_ptr(tsk
);
543 #ifndef CONFIG_THREAD_INFO_IN_TASK
545 * The task is finally done with both the stack and thread_info,
548 release_task_stack(tsk
);
551 * If the task had a separate stack allocation, it should be gone
554 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
556 rt_mutex_debug_task_free(tsk
);
557 ftrace_graph_exit_task(tsk
);
558 arch_release_task_struct(tsk
);
559 if (tsk
->flags
& PF_KTHREAD
)
560 free_kthread_struct(tsk
);
561 free_task_struct(tsk
);
563 EXPORT_SYMBOL(free_task
);
565 static void dup_mm_exe_file(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
567 struct file
*exe_file
;
569 exe_file
= get_mm_exe_file(oldmm
);
570 RCU_INIT_POINTER(mm
->exe_file
, exe_file
);
572 * We depend on the oldmm having properly denied write access to the
575 if (exe_file
&& deny_write_access(exe_file
))
576 pr_warn_once("deny_write_access() failed in %s\n", __func__
);
580 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
581 struct mm_struct
*oldmm
)
583 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
584 struct rb_node
**rb_link
, *rb_parent
;
586 unsigned long charge
;
589 uprobe_start_dup_mmap();
590 if (mmap_write_lock_killable(oldmm
)) {
592 goto fail_uprobe_end
;
594 flush_cache_dup_mm(oldmm
);
595 uprobe_dup_mmap(oldmm
, mm
);
597 * Not linked in yet - no deadlock potential:
599 mmap_write_lock_nested(mm
, SINGLE_DEPTH_NESTING
);
601 /* No ordering required: file already has been exposed. */
602 dup_mm_exe_file(mm
, oldmm
);
604 mm
->total_vm
= oldmm
->total_vm
;
605 mm
->data_vm
= oldmm
->data_vm
;
606 mm
->exec_vm
= oldmm
->exec_vm
;
607 mm
->stack_vm
= oldmm
->stack_vm
;
609 rb_link
= &mm
->mm_rb
.rb_node
;
612 retval
= ksm_fork(mm
, oldmm
);
615 khugepaged_fork(mm
, oldmm
);
618 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
621 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
622 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
627 * Don't duplicate many vmas if we've been oom-killed (for
630 if (fatal_signal_pending(current
)) {
634 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
635 unsigned long len
= vma_pages(mpnt
);
637 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
641 tmp
= vm_area_dup(mpnt
);
644 retval
= vma_dup_policy(mpnt
, tmp
);
646 goto fail_nomem_policy
;
648 retval
= dup_userfaultfd(tmp
, &uf
);
650 goto fail_nomem_anon_vma_fork
;
651 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
653 * VM_WIPEONFORK gets a clean slate in the child.
654 * Don't prepare anon_vma until fault since we don't
655 * copy page for current vma.
657 tmp
->anon_vma
= NULL
;
658 } else if (anon_vma_fork(tmp
, mpnt
))
659 goto fail_nomem_anon_vma_fork
;
660 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
663 struct address_space
*mapping
= file
->f_mapping
;
666 i_mmap_lock_write(mapping
);
667 if (tmp
->vm_flags
& VM_SHARED
)
668 mapping_allow_writable(mapping
);
669 flush_dcache_mmap_lock(mapping
);
670 /* insert tmp into the share list, just after mpnt */
671 vma_interval_tree_insert_after(tmp
, mpnt
,
673 flush_dcache_mmap_unlock(mapping
);
674 i_mmap_unlock_write(mapping
);
678 * Clear hugetlb-related page reserves for children. This only
679 * affects MAP_PRIVATE mappings. Faults generated by the child
680 * are not guaranteed to succeed, even if read-only
682 if (is_vm_hugetlb_page(tmp
))
683 reset_vma_resv_huge_pages(tmp
);
686 * Link in the new vma and copy the page table entries.
689 pprev
= &tmp
->vm_next
;
693 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
694 rb_link
= &tmp
->vm_rb
.rb_right
;
695 rb_parent
= &tmp
->vm_rb
;
698 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
699 retval
= copy_page_range(tmp
, mpnt
);
701 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
702 tmp
->vm_ops
->open(tmp
);
707 /* a new mm has just been created */
708 retval
= arch_dup_mmap(oldmm
, mm
);
710 mmap_write_unlock(mm
);
712 mmap_write_unlock(oldmm
);
713 dup_userfaultfd_complete(&uf
);
715 uprobe_end_dup_mmap();
717 fail_nomem_anon_vma_fork
:
718 mpol_put(vma_policy(tmp
));
723 vm_unacct_memory(charge
);
727 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
729 mm
->pgd
= pgd_alloc(mm
);
730 if (unlikely(!mm
->pgd
))
735 static inline void mm_free_pgd(struct mm_struct
*mm
)
737 pgd_free(mm
, mm
->pgd
);
740 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
742 mmap_write_lock(oldmm
);
743 dup_mm_exe_file(mm
, oldmm
);
744 mmap_write_unlock(oldmm
);
747 #define mm_alloc_pgd(mm) (0)
748 #define mm_free_pgd(mm)
749 #endif /* CONFIG_MMU */
751 static void check_mm(struct mm_struct
*mm
)
755 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
756 "Please make sure 'struct resident_page_types[]' is updated as well");
758 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
759 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
762 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
763 mm
, resident_page_types
[i
], x
);
766 if (mm_pgtables_bytes(mm
))
767 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
768 mm_pgtables_bytes(mm
));
770 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
771 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
775 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
776 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
779 * Called when the last reference to the mm
780 * is dropped: either by a lazy thread or by
781 * mmput. Free the page directory and the mm.
783 void __mmdrop(struct mm_struct
*mm
)
785 BUG_ON(mm
== &init_mm
);
786 WARN_ON_ONCE(mm
== current
->mm
);
787 WARN_ON_ONCE(mm
== current
->active_mm
);
790 mmu_notifier_subscriptions_destroy(mm
);
792 put_user_ns(mm
->user_ns
);
796 EXPORT_SYMBOL_GPL(__mmdrop
);
798 static void mmdrop_async_fn(struct work_struct
*work
)
800 struct mm_struct
*mm
;
802 mm
= container_of(work
, struct mm_struct
, async_put_work
);
806 static void mmdrop_async(struct mm_struct
*mm
)
808 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
809 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
810 schedule_work(&mm
->async_put_work
);
814 static inline void free_signal_struct(struct signal_struct
*sig
)
816 taskstats_tgid_free(sig
);
817 sched_autogroup_exit(sig
);
819 * __mmdrop is not safe to call from softirq context on x86 due to
820 * pgd_dtor so postpone it to the async context
823 mmdrop_async(sig
->oom_mm
);
824 kmem_cache_free(signal_cachep
, sig
);
827 static inline void put_signal_struct(struct signal_struct
*sig
)
829 if (refcount_dec_and_test(&sig
->sigcnt
))
830 free_signal_struct(sig
);
833 void __put_task_struct(struct task_struct
*tsk
)
835 WARN_ON(!tsk
->exit_state
);
836 WARN_ON(refcount_read(&tsk
->usage
));
837 WARN_ON(tsk
== current
);
841 task_numa_free(tsk
, true);
842 security_task_free(tsk
);
843 bpf_task_storage_free(tsk
);
845 delayacct_tsk_free(tsk
);
846 put_signal_struct(tsk
->signal
);
847 sched_core_free(tsk
);
850 EXPORT_SYMBOL_GPL(__put_task_struct
);
852 void __init __weak
arch_task_cache_init(void) { }
857 static void set_max_threads(unsigned int max_threads_suggested
)
860 unsigned long nr_pages
= totalram_pages();
863 * The number of threads shall be limited such that the thread
864 * structures may only consume a small part of the available memory.
866 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
867 threads
= MAX_THREADS
;
869 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
870 (u64
) THREAD_SIZE
* 8UL);
872 if (threads
> max_threads_suggested
)
873 threads
= max_threads_suggested
;
875 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
878 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
879 /* Initialized by the architecture: */
880 int arch_task_struct_size __read_mostly
;
883 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
884 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
886 /* Fetch thread_struct whitelist for the architecture. */
887 arch_thread_struct_whitelist(offset
, size
);
890 * Handle zero-sized whitelist or empty thread_struct, otherwise
891 * adjust offset to position of thread_struct in task_struct.
893 if (unlikely(*size
== 0))
896 *offset
+= offsetof(struct task_struct
, thread
);
898 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
900 void __init
fork_init(void)
903 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
904 #ifndef ARCH_MIN_TASKALIGN
905 #define ARCH_MIN_TASKALIGN 0
907 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
908 unsigned long useroffset
, usersize
;
910 /* create a slab on which task_structs can be allocated */
911 task_struct_whitelist(&useroffset
, &usersize
);
912 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
913 arch_task_struct_size
, align
,
914 SLAB_PANIC
|SLAB_ACCOUNT
,
915 useroffset
, usersize
, NULL
);
918 /* do the arch specific task caches init */
919 arch_task_cache_init();
921 set_max_threads(MAX_THREADS
);
923 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
924 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
925 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
926 init_task
.signal
->rlim
[RLIMIT_NPROC
];
928 for (i
= 0; i
< UCOUNT_COUNTS
; i
++)
929 init_user_ns
.ucount_max
[i
] = max_threads
/2;
931 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_NPROC
, RLIM_INFINITY
);
932 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_MSGQUEUE
, RLIM_INFINITY
);
933 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_SIGPENDING
, RLIM_INFINITY
);
934 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_MEMLOCK
, RLIM_INFINITY
);
936 #ifdef CONFIG_VMAP_STACK
937 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
938 NULL
, free_vm_stack_cache
);
943 lockdep_init_task(&init_task
);
947 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
948 struct task_struct
*src
)
954 void set_task_stack_end_magic(struct task_struct
*tsk
)
956 unsigned long *stackend
;
958 stackend
= end_of_stack(tsk
);
959 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
962 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
964 struct task_struct
*tsk
;
967 if (node
== NUMA_NO_NODE
)
968 node
= tsk_fork_get_node(orig
);
969 tsk
= alloc_task_struct_node(node
);
973 err
= arch_dup_task_struct(tsk
, orig
);
977 err
= alloc_thread_stack_node(tsk
, node
);
981 #ifdef CONFIG_THREAD_INFO_IN_TASK
982 refcount_set(&tsk
->stack_refcount
, 1);
984 account_kernel_stack(tsk
, 1);
986 err
= scs_prepare(tsk
, node
);
990 #ifdef CONFIG_SECCOMP
992 * We must handle setting up seccomp filters once we're under
993 * the sighand lock in case orig has changed between now and
994 * then. Until then, filter must be NULL to avoid messing up
995 * the usage counts on the error path calling free_task.
997 tsk
->seccomp
.filter
= NULL
;
1000 setup_thread_stack(tsk
, orig
);
1001 clear_user_return_notifier(tsk
);
1002 clear_tsk_need_resched(tsk
);
1003 set_task_stack_end_magic(tsk
);
1004 clear_syscall_work_syscall_user_dispatch(tsk
);
1006 #ifdef CONFIG_STACKPROTECTOR
1007 tsk
->stack_canary
= get_random_canary();
1009 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
1010 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
1011 dup_user_cpus_ptr(tsk
, orig
, node
);
1014 * One for the user space visible state that goes away when reaped.
1015 * One for the scheduler.
1017 refcount_set(&tsk
->rcu_users
, 2);
1018 /* One for the rcu users */
1019 refcount_set(&tsk
->usage
, 1);
1020 #ifdef CONFIG_BLK_DEV_IO_TRACE
1021 tsk
->btrace_seq
= 0;
1023 tsk
->splice_pipe
= NULL
;
1024 tsk
->task_frag
.page
= NULL
;
1025 tsk
->wake_q
.next
= NULL
;
1026 tsk
->worker_private
= NULL
;
1028 kcov_task_init(tsk
);
1029 kmap_local_fork(tsk
);
1031 #ifdef CONFIG_FAULT_INJECTION
1035 #ifdef CONFIG_BLK_CGROUP
1036 tsk
->throttle_queue
= NULL
;
1037 tsk
->use_memdelay
= 0;
1040 #ifdef CONFIG_IOMMU_SVA
1041 tsk
->pasid_activated
= 0;
1045 tsk
->active_memcg
= NULL
;
1048 #ifdef CONFIG_CPU_SUP_INTEL
1049 tsk
->reported_split_lock
= 0;
1055 exit_task_stack_account(tsk
);
1056 free_thread_stack(tsk
);
1058 free_task_struct(tsk
);
1062 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
1064 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
1066 static int __init
coredump_filter_setup(char *s
)
1068 default_dump_filter
=
1069 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
1070 MMF_DUMP_FILTER_MASK
;
1074 __setup("coredump_filter=", coredump_filter_setup
);
1076 #include <linux/init_task.h>
1078 static void mm_init_aio(struct mm_struct
*mm
)
1081 spin_lock_init(&mm
->ioctx_lock
);
1082 mm
->ioctx_table
= NULL
;
1086 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
1087 struct task_struct
*p
)
1091 WRITE_ONCE(mm
->owner
, NULL
);
1095 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
1102 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1104 #ifdef CONFIG_UPROBES
1105 mm
->uprobes_state
.xol_area
= NULL
;
1109 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1110 struct user_namespace
*user_ns
)
1113 mm
->mm_rb
= RB_ROOT
;
1114 mm
->vmacache_seqnum
= 0;
1115 atomic_set(&mm
->mm_users
, 1);
1116 atomic_set(&mm
->mm_count
, 1);
1117 seqcount_init(&mm
->write_protect_seq
);
1119 INIT_LIST_HEAD(&mm
->mmlist
);
1120 mm_pgtables_bytes_init(mm
);
1123 atomic64_set(&mm
->pinned_vm
, 0);
1124 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1125 spin_lock_init(&mm
->page_table_lock
);
1126 spin_lock_init(&mm
->arg_lock
);
1127 mm_init_cpumask(mm
);
1129 mm_init_owner(mm
, p
);
1131 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1132 mmu_notifier_subscriptions_init(mm
);
1133 init_tlb_flush_pending(mm
);
1134 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1135 mm
->pmd_huge_pte
= NULL
;
1137 mm_init_uprobes_state(mm
);
1138 hugetlb_count_init(mm
);
1141 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1142 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1144 mm
->flags
= default_dump_filter
;
1148 if (mm_alloc_pgd(mm
))
1151 if (init_new_context(p
, mm
))
1152 goto fail_nocontext
;
1154 mm
->user_ns
= get_user_ns(user_ns
);
1165 * Allocate and initialize an mm_struct.
1167 struct mm_struct
*mm_alloc(void)
1169 struct mm_struct
*mm
;
1175 memset(mm
, 0, sizeof(*mm
));
1176 return mm_init(mm
, current
, current_user_ns());
1179 static inline void __mmput(struct mm_struct
*mm
)
1181 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1183 uprobe_clear_state(mm
);
1186 khugepaged_exit(mm
); /* must run before exit_mmap */
1188 mm_put_huge_zero_page(mm
);
1189 set_mm_exe_file(mm
, NULL
);
1190 if (!list_empty(&mm
->mmlist
)) {
1191 spin_lock(&mmlist_lock
);
1192 list_del(&mm
->mmlist
);
1193 spin_unlock(&mmlist_lock
);
1196 module_put(mm
->binfmt
->module
);
1201 * Decrement the use count and release all resources for an mm.
1203 void mmput(struct mm_struct
*mm
)
1207 if (atomic_dec_and_test(&mm
->mm_users
))
1210 EXPORT_SYMBOL_GPL(mmput
);
1213 static void mmput_async_fn(struct work_struct
*work
)
1215 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1221 void mmput_async(struct mm_struct
*mm
)
1223 if (atomic_dec_and_test(&mm
->mm_users
)) {
1224 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1225 schedule_work(&mm
->async_put_work
);
1228 EXPORT_SYMBOL_GPL(mmput_async
);
1232 * set_mm_exe_file - change a reference to the mm's executable file
1234 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1236 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1237 * invocations: in mmput() nobody alive left, in execve task is single
1240 * Can only fail if new_exe_file != NULL.
1242 int set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1244 struct file
*old_exe_file
;
1247 * It is safe to dereference the exe_file without RCU as
1248 * this function is only called if nobody else can access
1249 * this mm -- see comment above for justification.
1251 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1255 * We expect the caller (i.e., sys_execve) to already denied
1256 * write access, so this is unlikely to fail.
1258 if (unlikely(deny_write_access(new_exe_file
)))
1260 get_file(new_exe_file
);
1262 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1264 allow_write_access(old_exe_file
);
1271 * replace_mm_exe_file - replace a reference to the mm's executable file
1273 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1274 * dealing with concurrent invocation and without grabbing the mmap lock in
1277 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1279 int replace_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1281 struct vm_area_struct
*vma
;
1282 struct file
*old_exe_file
;
1285 /* Forbid mm->exe_file change if old file still mapped. */
1286 old_exe_file
= get_mm_exe_file(mm
);
1289 for (vma
= mm
->mmap
; vma
&& !ret
; vma
= vma
->vm_next
) {
1292 if (path_equal(&vma
->vm_file
->f_path
,
1293 &old_exe_file
->f_path
))
1296 mmap_read_unlock(mm
);
1302 /* set the new file, lockless */
1303 ret
= deny_write_access(new_exe_file
);
1306 get_file(new_exe_file
);
1308 old_exe_file
= xchg(&mm
->exe_file
, new_exe_file
);
1311 * Don't race with dup_mmap() getting the file and disallowing
1312 * write access while someone might open the file writable.
1315 allow_write_access(old_exe_file
);
1317 mmap_read_unlock(mm
);
1323 * get_mm_exe_file - acquire a reference to the mm's executable file
1325 * Returns %NULL if mm has no associated executable file.
1326 * User must release file via fput().
1328 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1330 struct file
*exe_file
;
1333 exe_file
= rcu_dereference(mm
->exe_file
);
1334 if (exe_file
&& !get_file_rcu(exe_file
))
1341 * get_task_exe_file - acquire a reference to the task's executable file
1343 * Returns %NULL if task's mm (if any) has no associated executable file or
1344 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1345 * User must release file via fput().
1347 struct file
*get_task_exe_file(struct task_struct
*task
)
1349 struct file
*exe_file
= NULL
;
1350 struct mm_struct
*mm
;
1355 if (!(task
->flags
& PF_KTHREAD
))
1356 exe_file
= get_mm_exe_file(mm
);
1363 * get_task_mm - acquire a reference to the task's mm
1365 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1366 * this kernel workthread has transiently adopted a user mm with use_mm,
1367 * to do its AIO) is not set and if so returns a reference to it, after
1368 * bumping up the use count. User must release the mm via mmput()
1369 * after use. Typically used by /proc and ptrace.
1371 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1373 struct mm_struct
*mm
;
1378 if (task
->flags
& PF_KTHREAD
)
1386 EXPORT_SYMBOL_GPL(get_task_mm
);
1388 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1390 struct mm_struct
*mm
;
1393 err
= down_read_killable(&task
->signal
->exec_update_lock
);
1395 return ERR_PTR(err
);
1397 mm
= get_task_mm(task
);
1398 if (mm
&& mm
!= current
->mm
&&
1399 !ptrace_may_access(task
, mode
)) {
1401 mm
= ERR_PTR(-EACCES
);
1403 up_read(&task
->signal
->exec_update_lock
);
1408 static void complete_vfork_done(struct task_struct
*tsk
)
1410 struct completion
*vfork
;
1413 vfork
= tsk
->vfork_done
;
1414 if (likely(vfork
)) {
1415 tsk
->vfork_done
= NULL
;
1421 static int wait_for_vfork_done(struct task_struct
*child
,
1422 struct completion
*vfork
)
1424 unsigned int state
= TASK_UNINTERRUPTIBLE
|TASK_KILLABLE
|TASK_FREEZABLE
;
1427 cgroup_enter_frozen();
1428 killed
= wait_for_completion_state(vfork
, state
);
1429 cgroup_leave_frozen(false);
1433 child
->vfork_done
= NULL
;
1437 put_task_struct(child
);
1441 /* Please note the differences between mmput and mm_release.
1442 * mmput is called whenever we stop holding onto a mm_struct,
1443 * error success whatever.
1445 * mm_release is called after a mm_struct has been removed
1446 * from the current process.
1448 * This difference is important for error handling, when we
1449 * only half set up a mm_struct for a new process and need to restore
1450 * the old one. Because we mmput the new mm_struct before
1451 * restoring the old one. . .
1452 * Eric Biederman 10 January 1998
1454 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1456 uprobe_free_utask(tsk
);
1458 /* Get rid of any cached register state */
1459 deactivate_mm(tsk
, mm
);
1462 * Signal userspace if we're not exiting with a core dump
1463 * because we want to leave the value intact for debugging
1466 if (tsk
->clear_child_tid
) {
1467 if (atomic_read(&mm
->mm_users
) > 1) {
1469 * We don't check the error code - if userspace has
1470 * not set up a proper pointer then tough luck.
1472 put_user(0, tsk
->clear_child_tid
);
1473 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1474 1, NULL
, NULL
, 0, 0);
1476 tsk
->clear_child_tid
= NULL
;
1480 * All done, finally we can wake up parent and return this mm to him.
1481 * Also kthread_stop() uses this completion for synchronization.
1483 if (tsk
->vfork_done
)
1484 complete_vfork_done(tsk
);
1487 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1489 futex_exit_release(tsk
);
1490 mm_release(tsk
, mm
);
1493 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1495 futex_exec_release(tsk
);
1496 mm_release(tsk
, mm
);
1500 * dup_mm() - duplicates an existing mm structure
1501 * @tsk: the task_struct with which the new mm will be associated.
1502 * @oldmm: the mm to duplicate.
1504 * Allocates a new mm structure and duplicates the provided @oldmm structure
1507 * Return: the duplicated mm or NULL on failure.
1509 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1510 struct mm_struct
*oldmm
)
1512 struct mm_struct
*mm
;
1519 memcpy(mm
, oldmm
, sizeof(*mm
));
1521 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1524 err
= dup_mmap(mm
, oldmm
);
1528 mm
->hiwater_rss
= get_mm_rss(mm
);
1529 mm
->hiwater_vm
= mm
->total_vm
;
1531 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1537 /* don't put binfmt in mmput, we haven't got module yet */
1539 mm_init_owner(mm
, NULL
);
1546 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1548 struct mm_struct
*mm
, *oldmm
;
1550 tsk
->min_flt
= tsk
->maj_flt
= 0;
1551 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1552 #ifdef CONFIG_DETECT_HUNG_TASK
1553 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1554 tsk
->last_switch_time
= 0;
1558 tsk
->active_mm
= NULL
;
1561 * Are we cloning a kernel thread?
1563 * We need to steal a active VM for that..
1565 oldmm
= current
->mm
;
1569 /* initialize the new vmacache entries */
1570 vmacache_flush(tsk
);
1572 if (clone_flags
& CLONE_VM
) {
1576 mm
= dup_mm(tsk
, current
->mm
);
1582 tsk
->active_mm
= mm
;
1586 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1588 struct fs_struct
*fs
= current
->fs
;
1589 if (clone_flags
& CLONE_FS
) {
1590 /* tsk->fs is already what we want */
1591 spin_lock(&fs
->lock
);
1593 spin_unlock(&fs
->lock
);
1597 spin_unlock(&fs
->lock
);
1600 tsk
->fs
= copy_fs_struct(fs
);
1606 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1608 struct files_struct
*oldf
, *newf
;
1612 * A background process may not have any files ...
1614 oldf
= current
->files
;
1618 if (clone_flags
& CLONE_FILES
) {
1619 atomic_inc(&oldf
->count
);
1623 newf
= dup_fd(oldf
, NR_OPEN_MAX
, &error
);
1633 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1635 struct sighand_struct
*sig
;
1637 if (clone_flags
& CLONE_SIGHAND
) {
1638 refcount_inc(¤t
->sighand
->count
);
1641 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1642 RCU_INIT_POINTER(tsk
->sighand
, sig
);
1646 refcount_set(&sig
->count
, 1);
1647 spin_lock_irq(¤t
->sighand
->siglock
);
1648 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1649 spin_unlock_irq(¤t
->sighand
->siglock
);
1651 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1652 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1653 flush_signal_handlers(tsk
, 0);
1658 void __cleanup_sighand(struct sighand_struct
*sighand
)
1660 if (refcount_dec_and_test(&sighand
->count
)) {
1661 signalfd_cleanup(sighand
);
1663 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1664 * without an RCU grace period, see __lock_task_sighand().
1666 kmem_cache_free(sighand_cachep
, sighand
);
1671 * Initialize POSIX timer handling for a thread group.
1673 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1675 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1676 unsigned long cpu_limit
;
1678 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1679 posix_cputimers_group_init(pct
, cpu_limit
);
1682 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1684 struct signal_struct
*sig
;
1686 if (clone_flags
& CLONE_THREAD
)
1689 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1694 sig
->nr_threads
= 1;
1695 sig
->quick_threads
= 1;
1696 atomic_set(&sig
->live
, 1);
1697 refcount_set(&sig
->sigcnt
, 1);
1699 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1700 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1701 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1703 init_waitqueue_head(&sig
->wait_chldexit
);
1704 sig
->curr_target
= tsk
;
1705 init_sigpending(&sig
->shared_pending
);
1706 INIT_HLIST_HEAD(&sig
->multiprocess
);
1707 seqlock_init(&sig
->stats_lock
);
1708 prev_cputime_init(&sig
->prev_cputime
);
1710 #ifdef CONFIG_POSIX_TIMERS
1711 INIT_LIST_HEAD(&sig
->posix_timers
);
1712 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1713 sig
->real_timer
.function
= it_real_fn
;
1716 task_lock(current
->group_leader
);
1717 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1718 task_unlock(current
->group_leader
);
1720 posix_cpu_timers_init_group(sig
);
1722 tty_audit_fork(sig
);
1723 sched_autogroup_fork(sig
);
1725 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1726 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1728 mutex_init(&sig
->cred_guard_mutex
);
1729 init_rwsem(&sig
->exec_update_lock
);
1734 static void copy_seccomp(struct task_struct
*p
)
1736 #ifdef CONFIG_SECCOMP
1738 * Must be called with sighand->lock held, which is common to
1739 * all threads in the group. Holding cred_guard_mutex is not
1740 * needed because this new task is not yet running and cannot
1743 assert_spin_locked(¤t
->sighand
->siglock
);
1745 /* Ref-count the new filter user, and assign it. */
1746 get_seccomp_filter(current
);
1747 p
->seccomp
= current
->seccomp
;
1750 * Explicitly enable no_new_privs here in case it got set
1751 * between the task_struct being duplicated and holding the
1752 * sighand lock. The seccomp state and nnp must be in sync.
1754 if (task_no_new_privs(current
))
1755 task_set_no_new_privs(p
);
1758 * If the parent gained a seccomp mode after copying thread
1759 * flags and between before we held the sighand lock, we have
1760 * to manually enable the seccomp thread flag here.
1762 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1763 set_task_syscall_work(p
, SECCOMP
);
1767 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1769 current
->clear_child_tid
= tidptr
;
1771 return task_pid_vnr(current
);
1774 static void rt_mutex_init_task(struct task_struct
*p
)
1776 raw_spin_lock_init(&p
->pi_lock
);
1777 #ifdef CONFIG_RT_MUTEXES
1778 p
->pi_waiters
= RB_ROOT_CACHED
;
1779 p
->pi_top_task
= NULL
;
1780 p
->pi_blocked_on
= NULL
;
1784 static inline void init_task_pid_links(struct task_struct
*task
)
1788 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
)
1789 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1793 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1795 if (type
== PIDTYPE_PID
)
1796 task
->thread_pid
= pid
;
1798 task
->signal
->pids
[type
] = pid
;
1801 static inline void rcu_copy_process(struct task_struct
*p
)
1803 #ifdef CONFIG_PREEMPT_RCU
1804 p
->rcu_read_lock_nesting
= 0;
1805 p
->rcu_read_unlock_special
.s
= 0;
1806 p
->rcu_blocked_node
= NULL
;
1807 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1808 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1809 #ifdef CONFIG_TASKS_RCU
1810 p
->rcu_tasks_holdout
= false;
1811 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1812 p
->rcu_tasks_idle_cpu
= -1;
1813 #endif /* #ifdef CONFIG_TASKS_RCU */
1814 #ifdef CONFIG_TASKS_TRACE_RCU
1815 p
->trc_reader_nesting
= 0;
1816 p
->trc_reader_special
.s
= 0;
1817 INIT_LIST_HEAD(&p
->trc_holdout_list
);
1818 INIT_LIST_HEAD(&p
->trc_blkd_node
);
1819 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1822 struct pid
*pidfd_pid(const struct file
*file
)
1824 if (file
->f_op
== &pidfd_fops
)
1825 return file
->private_data
;
1827 return ERR_PTR(-EBADF
);
1830 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1832 struct pid
*pid
= file
->private_data
;
1834 file
->private_data
= NULL
;
1839 #ifdef CONFIG_PROC_FS
1841 * pidfd_show_fdinfo - print information about a pidfd
1842 * @m: proc fdinfo file
1843 * @f: file referencing a pidfd
1846 * This function will print the pid that a given pidfd refers to in the
1847 * pid namespace of the procfs instance.
1848 * If the pid namespace of the process is not a descendant of the pid
1849 * namespace of the procfs instance 0 will be shown as its pid. This is
1850 * similar to calling getppid() on a process whose parent is outside of
1851 * its pid namespace.
1854 * If pid namespaces are supported then this function will also print
1855 * the pid of a given pidfd refers to for all descendant pid namespaces
1856 * starting from the current pid namespace of the instance, i.e. the
1857 * Pid field and the first entry in the NSpid field will be identical.
1858 * If the pid namespace of the process is not a descendant of the pid
1859 * namespace of the procfs instance 0 will be shown as its first NSpid
1860 * entry and no others will be shown.
1861 * Note that this differs from the Pid and NSpid fields in
1862 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1863 * the pid namespace of the procfs instance. The difference becomes
1864 * obvious when sending around a pidfd between pid namespaces from a
1865 * different branch of the tree, i.e. where no ancestral relation is
1866 * present between the pid namespaces:
1867 * - create two new pid namespaces ns1 and ns2 in the initial pid
1868 * namespace (also take care to create new mount namespaces in the
1869 * new pid namespace and mount procfs)
1870 * - create a process with a pidfd in ns1
1871 * - send pidfd from ns1 to ns2
1872 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1873 * have exactly one entry, which is 0
1875 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1877 struct pid
*pid
= f
->private_data
;
1878 struct pid_namespace
*ns
;
1881 if (likely(pid_has_task(pid
, PIDTYPE_PID
))) {
1882 ns
= proc_pid_ns(file_inode(m
->file
)->i_sb
);
1883 nr
= pid_nr_ns(pid
, ns
);
1886 seq_put_decimal_ll(m
, "Pid:\t", nr
);
1888 #ifdef CONFIG_PID_NS
1889 seq_put_decimal_ll(m
, "\nNSpid:\t", nr
);
1893 /* If nr is non-zero it means that 'pid' is valid and that
1894 * ns, i.e. the pid namespace associated with the procfs
1895 * instance, is in the pid namespace hierarchy of pid.
1896 * Start at one below the already printed level.
1898 for (i
= ns
->level
+ 1; i
<= pid
->level
; i
++)
1899 seq_put_decimal_ll(m
, "\t", pid
->numbers
[i
].nr
);
1907 * Poll support for process exit notification.
1909 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1911 struct pid
*pid
= file
->private_data
;
1912 __poll_t poll_flags
= 0;
1914 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1917 * Inform pollers only when the whole thread group exits.
1918 * If the thread group leader exits before all other threads in the
1919 * group, then poll(2) should block, similar to the wait(2) family.
1921 if (thread_group_exited(pid
))
1922 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1927 const struct file_operations pidfd_fops
= {
1928 .release
= pidfd_release
,
1930 #ifdef CONFIG_PROC_FS
1931 .show_fdinfo
= pidfd_show_fdinfo
,
1935 static void __delayed_free_task(struct rcu_head
*rhp
)
1937 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1942 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1944 if (IS_ENABLED(CONFIG_MEMCG
))
1945 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1950 static void copy_oom_score_adj(u64 clone_flags
, struct task_struct
*tsk
)
1952 /* Skip if kernel thread */
1956 /* Skip if spawning a thread or using vfork */
1957 if ((clone_flags
& (CLONE_VM
| CLONE_THREAD
| CLONE_VFORK
)) != CLONE_VM
)
1960 /* We need to synchronize with __set_oom_adj */
1961 mutex_lock(&oom_adj_mutex
);
1962 set_bit(MMF_MULTIPROCESS
, &tsk
->mm
->flags
);
1963 /* Update the values in case they were changed after copy_signal */
1964 tsk
->signal
->oom_score_adj
= current
->signal
->oom_score_adj
;
1965 tsk
->signal
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1966 mutex_unlock(&oom_adj_mutex
);
1970 static void rv_task_fork(struct task_struct
*p
)
1974 for (i
= 0; i
< RV_PER_TASK_MONITORS
; i
++)
1975 p
->rv
[i
].da_mon
.monitoring
= false;
1978 #define rv_task_fork(p) do {} while (0)
1982 * This creates a new process as a copy of the old one,
1983 * but does not actually start it yet.
1985 * It copies the registers, and all the appropriate
1986 * parts of the process environment (as per the clone
1987 * flags). The actual kick-off is left to the caller.
1989 static __latent_entropy
struct task_struct
*copy_process(
1993 struct kernel_clone_args
*args
)
1995 int pidfd
= -1, retval
;
1996 struct task_struct
*p
;
1997 struct multiprocess_signals delayed
;
1998 struct file
*pidfile
= NULL
;
1999 const u64 clone_flags
= args
->flags
;
2000 struct nsproxy
*nsp
= current
->nsproxy
;
2003 * Don't allow sharing the root directory with processes in a different
2006 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
2007 return ERR_PTR(-EINVAL
);
2009 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
2010 return ERR_PTR(-EINVAL
);
2013 * Thread groups must share signals as well, and detached threads
2014 * can only be started up within the thread group.
2016 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
2017 return ERR_PTR(-EINVAL
);
2020 * Shared signal handlers imply shared VM. By way of the above,
2021 * thread groups also imply shared VM. Blocking this case allows
2022 * for various simplifications in other code.
2024 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
2025 return ERR_PTR(-EINVAL
);
2028 * Siblings of global init remain as zombies on exit since they are
2029 * not reaped by their parent (swapper). To solve this and to avoid
2030 * multi-rooted process trees, prevent global and container-inits
2031 * from creating siblings.
2033 if ((clone_flags
& CLONE_PARENT
) &&
2034 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
2035 return ERR_PTR(-EINVAL
);
2038 * If the new process will be in a different pid or user namespace
2039 * do not allow it to share a thread group with the forking task.
2041 if (clone_flags
& CLONE_THREAD
) {
2042 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
2043 (task_active_pid_ns(current
) != nsp
->pid_ns_for_children
))
2044 return ERR_PTR(-EINVAL
);
2048 * If the new process will be in a different time namespace
2049 * do not allow it to share VM or a thread group with the forking task.
2051 if (clone_flags
& (CLONE_THREAD
| CLONE_VM
)) {
2052 if (nsp
->time_ns
!= nsp
->time_ns_for_children
)
2053 return ERR_PTR(-EINVAL
);
2056 if (clone_flags
& CLONE_PIDFD
) {
2058 * - CLONE_DETACHED is blocked so that we can potentially
2059 * reuse it later for CLONE_PIDFD.
2060 * - CLONE_THREAD is blocked until someone really needs it.
2062 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
2063 return ERR_PTR(-EINVAL
);
2067 * Force any signals received before this point to be delivered
2068 * before the fork happens. Collect up signals sent to multiple
2069 * processes that happen during the fork and delay them so that
2070 * they appear to happen after the fork.
2072 sigemptyset(&delayed
.signal
);
2073 INIT_HLIST_NODE(&delayed
.node
);
2075 spin_lock_irq(¤t
->sighand
->siglock
);
2076 if (!(clone_flags
& CLONE_THREAD
))
2077 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
2078 recalc_sigpending();
2079 spin_unlock_irq(¤t
->sighand
->siglock
);
2080 retval
= -ERESTARTNOINTR
;
2081 if (task_sigpending(current
))
2085 p
= dup_task_struct(current
, node
);
2088 p
->flags
&= ~PF_KTHREAD
;
2090 p
->flags
|= PF_KTHREAD
;
2091 if (args
->io_thread
) {
2093 * Mark us an IO worker, and block any signal that isn't
2096 p
->flags
|= PF_IO_WORKER
;
2097 siginitsetinv(&p
->blocked
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2100 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
2102 * Clear TID on mm_release()?
2104 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
2106 ftrace_graph_init_task(p
);
2108 rt_mutex_init_task(p
);
2110 lockdep_assert_irqs_enabled();
2111 #ifdef CONFIG_PROVE_LOCKING
2112 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
2114 retval
= copy_creds(p
, clone_flags
);
2119 if (is_rlimit_overlimit(task_ucounts(p
), UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
))) {
2120 if (p
->real_cred
->user
!= INIT_USER
&&
2121 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
2122 goto bad_fork_cleanup_count
;
2124 current
->flags
&= ~PF_NPROC_EXCEEDED
;
2127 * If multiple threads are within copy_process(), then this check
2128 * triggers too late. This doesn't hurt, the check is only there
2129 * to stop root fork bombs.
2132 if (data_race(nr_threads
>= max_threads
))
2133 goto bad_fork_cleanup_count
;
2135 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
2136 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
| PF_NO_SETAFFINITY
);
2137 p
->flags
|= PF_FORKNOEXEC
;
2138 INIT_LIST_HEAD(&p
->children
);
2139 INIT_LIST_HEAD(&p
->sibling
);
2140 rcu_copy_process(p
);
2141 p
->vfork_done
= NULL
;
2142 spin_lock_init(&p
->alloc_lock
);
2144 init_sigpending(&p
->pending
);
2146 p
->utime
= p
->stime
= p
->gtime
= 0;
2147 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2148 p
->utimescaled
= p
->stimescaled
= 0;
2150 prev_cputime_init(&p
->prev_cputime
);
2152 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2153 seqcount_init(&p
->vtime
.seqcount
);
2154 p
->vtime
.starttime
= 0;
2155 p
->vtime
.state
= VTIME_INACTIVE
;
2158 #ifdef CONFIG_IO_URING
2162 #if defined(SPLIT_RSS_COUNTING)
2163 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
2166 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
2172 task_io_accounting_init(&p
->ioac
);
2173 acct_clear_integrals(p
);
2175 posix_cputimers_init(&p
->posix_cputimers
);
2177 p
->io_context
= NULL
;
2178 audit_set_context(p
, NULL
);
2180 if (args
->kthread
) {
2181 if (!set_kthread_struct(p
))
2182 goto bad_fork_cleanup_delayacct
;
2185 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2186 if (IS_ERR(p
->mempolicy
)) {
2187 retval
= PTR_ERR(p
->mempolicy
);
2188 p
->mempolicy
= NULL
;
2189 goto bad_fork_cleanup_delayacct
;
2192 #ifdef CONFIG_CPUSETS
2193 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2194 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2195 seqcount_spinlock_init(&p
->mems_allowed_seq
, &p
->alloc_lock
);
2197 #ifdef CONFIG_TRACE_IRQFLAGS
2198 memset(&p
->irqtrace
, 0, sizeof(p
->irqtrace
));
2199 p
->irqtrace
.hardirq_disable_ip
= _THIS_IP_
;
2200 p
->irqtrace
.softirq_enable_ip
= _THIS_IP_
;
2201 p
->softirqs_enabled
= 1;
2202 p
->softirq_context
= 0;
2205 p
->pagefault_disabled
= 0;
2207 #ifdef CONFIG_LOCKDEP
2208 lockdep_init_task(p
);
2211 #ifdef CONFIG_DEBUG_MUTEXES
2212 p
->blocked_on
= NULL
; /* not blocked yet */
2214 #ifdef CONFIG_BCACHE
2215 p
->sequential_io
= 0;
2216 p
->sequential_io_avg
= 0;
2218 #ifdef CONFIG_BPF_SYSCALL
2219 RCU_INIT_POINTER(p
->bpf_storage
, NULL
);
2223 /* Perform scheduler related setup. Assign this task to a CPU. */
2224 retval
= sched_fork(clone_flags
, p
);
2226 goto bad_fork_cleanup_policy
;
2228 retval
= perf_event_init_task(p
, clone_flags
);
2230 goto bad_fork_cleanup_policy
;
2231 retval
= audit_alloc(p
);
2233 goto bad_fork_cleanup_perf
;
2234 /* copy all the process information */
2236 retval
= security_task_alloc(p
, clone_flags
);
2238 goto bad_fork_cleanup_audit
;
2239 retval
= copy_semundo(clone_flags
, p
);
2241 goto bad_fork_cleanup_security
;
2242 retval
= copy_files(clone_flags
, p
);
2244 goto bad_fork_cleanup_semundo
;
2245 retval
= copy_fs(clone_flags
, p
);
2247 goto bad_fork_cleanup_files
;
2248 retval
= copy_sighand(clone_flags
, p
);
2250 goto bad_fork_cleanup_fs
;
2251 retval
= copy_signal(clone_flags
, p
);
2253 goto bad_fork_cleanup_sighand
;
2254 retval
= copy_mm(clone_flags
, p
);
2256 goto bad_fork_cleanup_signal
;
2257 retval
= copy_namespaces(clone_flags
, p
);
2259 goto bad_fork_cleanup_mm
;
2260 retval
= copy_io(clone_flags
, p
);
2262 goto bad_fork_cleanup_namespaces
;
2263 retval
= copy_thread(p
, args
);
2265 goto bad_fork_cleanup_io
;
2267 stackleak_task_init(p
);
2269 if (pid
!= &init_struct_pid
) {
2270 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2271 args
->set_tid_size
);
2273 retval
= PTR_ERR(pid
);
2274 goto bad_fork_cleanup_thread
;
2279 * This has to happen after we've potentially unshared the file
2280 * descriptor table (so that the pidfd doesn't leak into the child
2281 * if the fd table isn't shared).
2283 if (clone_flags
& CLONE_PIDFD
) {
2284 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2286 goto bad_fork_free_pid
;
2290 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2291 O_RDWR
| O_CLOEXEC
);
2292 if (IS_ERR(pidfile
)) {
2293 put_unused_fd(pidfd
);
2294 retval
= PTR_ERR(pidfile
);
2295 goto bad_fork_free_pid
;
2297 get_pid(pid
); /* held by pidfile now */
2299 retval
= put_user(pidfd
, args
->pidfd
);
2301 goto bad_fork_put_pidfd
;
2310 * sigaltstack should be cleared when sharing the same VM
2312 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2316 * Syscall tracing and stepping should be turned off in the
2317 * child regardless of CLONE_PTRACE.
2319 user_disable_single_step(p
);
2320 clear_task_syscall_work(p
, SYSCALL_TRACE
);
2321 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2322 clear_task_syscall_work(p
, SYSCALL_EMU
);
2324 clear_tsk_latency_tracing(p
);
2326 /* ok, now we should be set up.. */
2327 p
->pid
= pid_nr(pid
);
2328 if (clone_flags
& CLONE_THREAD
) {
2329 p
->group_leader
= current
->group_leader
;
2330 p
->tgid
= current
->tgid
;
2332 p
->group_leader
= p
;
2337 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2338 p
->dirty_paused_when
= 0;
2340 p
->pdeath_signal
= 0;
2341 INIT_LIST_HEAD(&p
->thread_group
);
2342 p
->task_works
= NULL
;
2343 clear_posix_cputimers_work(p
);
2345 #ifdef CONFIG_KRETPROBES
2346 p
->kretprobe_instances
.first
= NULL
;
2348 #ifdef CONFIG_RETHOOK
2349 p
->rethooks
.first
= NULL
;
2353 * Ensure that the cgroup subsystem policies allow the new process to be
2354 * forked. It should be noted that the new process's css_set can be changed
2355 * between here and cgroup_post_fork() if an organisation operation is in
2358 retval
= cgroup_can_fork(p
, args
);
2360 goto bad_fork_put_pidfd
;
2363 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2364 * the new task on the correct runqueue. All this *before* the task
2367 * This isn't part of ->can_fork() because while the re-cloning is
2368 * cgroup specific, it unconditionally needs to place the task on a
2371 sched_cgroup_fork(p
, args
);
2374 * From this point on we must avoid any synchronous user-space
2375 * communication until we take the tasklist-lock. In particular, we do
2376 * not want user-space to be able to predict the process start-time by
2377 * stalling fork(2) after we recorded the start_time but before it is
2378 * visible to the system.
2381 p
->start_time
= ktime_get_ns();
2382 p
->start_boottime
= ktime_get_boottime_ns();
2385 * Make it visible to the rest of the system, but dont wake it up yet.
2386 * Need tasklist lock for parent etc handling!
2388 write_lock_irq(&tasklist_lock
);
2390 /* CLONE_PARENT re-uses the old parent */
2391 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2392 p
->real_parent
= current
->real_parent
;
2393 p
->parent_exec_id
= current
->parent_exec_id
;
2394 if (clone_flags
& CLONE_THREAD
)
2395 p
->exit_signal
= -1;
2397 p
->exit_signal
= current
->group_leader
->exit_signal
;
2399 p
->real_parent
= current
;
2400 p
->parent_exec_id
= current
->self_exec_id
;
2401 p
->exit_signal
= args
->exit_signal
;
2404 klp_copy_process(p
);
2408 spin_lock(¤t
->sighand
->siglock
);
2411 * Copy seccomp details explicitly here, in case they were changed
2412 * before holding sighand lock.
2418 rseq_fork(p
, clone_flags
);
2420 /* Don't start children in a dying pid namespace */
2421 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2423 goto bad_fork_cancel_cgroup
;
2426 /* Let kill terminate clone/fork in the middle */
2427 if (fatal_signal_pending(current
)) {
2429 goto bad_fork_cancel_cgroup
;
2432 init_task_pid_links(p
);
2433 if (likely(p
->pid
)) {
2434 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2436 init_task_pid(p
, PIDTYPE_PID
, pid
);
2437 if (thread_group_leader(p
)) {
2438 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2439 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2440 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2442 if (is_child_reaper(pid
)) {
2443 ns_of_pid(pid
)->child_reaper
= p
;
2444 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2446 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2447 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2449 * Inherit has_child_subreaper flag under the same
2450 * tasklist_lock with adding child to the process tree
2451 * for propagate_has_child_subreaper optimization.
2453 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2454 p
->real_parent
->signal
->is_child_subreaper
;
2455 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2456 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2457 attach_pid(p
, PIDTYPE_TGID
);
2458 attach_pid(p
, PIDTYPE_PGID
);
2459 attach_pid(p
, PIDTYPE_SID
);
2460 __this_cpu_inc(process_counts
);
2462 current
->signal
->nr_threads
++;
2463 current
->signal
->quick_threads
++;
2464 atomic_inc(¤t
->signal
->live
);
2465 refcount_inc(¤t
->signal
->sigcnt
);
2466 task_join_group_stop(p
);
2467 list_add_tail_rcu(&p
->thread_group
,
2468 &p
->group_leader
->thread_group
);
2469 list_add_tail_rcu(&p
->thread_node
,
2470 &p
->signal
->thread_head
);
2472 attach_pid(p
, PIDTYPE_PID
);
2476 hlist_del_init(&delayed
.node
);
2477 spin_unlock(¤t
->sighand
->siglock
);
2478 syscall_tracepoint_update(p
);
2479 write_unlock_irq(&tasklist_lock
);
2482 fd_install(pidfd
, pidfile
);
2484 proc_fork_connector(p
);
2486 cgroup_post_fork(p
, args
);
2489 trace_task_newtask(p
, clone_flags
);
2490 uprobe_copy_process(p
, clone_flags
);
2492 copy_oom_score_adj(clone_flags
, p
);
2496 bad_fork_cancel_cgroup
:
2498 spin_unlock(¤t
->sighand
->siglock
);
2499 write_unlock_irq(&tasklist_lock
);
2500 cgroup_cancel_fork(p
, args
);
2502 if (clone_flags
& CLONE_PIDFD
) {
2504 put_unused_fd(pidfd
);
2507 if (pid
!= &init_struct_pid
)
2509 bad_fork_cleanup_thread
:
2511 bad_fork_cleanup_io
:
2514 bad_fork_cleanup_namespaces
:
2515 exit_task_namespaces(p
);
2516 bad_fork_cleanup_mm
:
2518 mm_clear_owner(p
->mm
, p
);
2521 bad_fork_cleanup_signal
:
2522 if (!(clone_flags
& CLONE_THREAD
))
2523 free_signal_struct(p
->signal
);
2524 bad_fork_cleanup_sighand
:
2525 __cleanup_sighand(p
->sighand
);
2526 bad_fork_cleanup_fs
:
2527 exit_fs(p
); /* blocking */
2528 bad_fork_cleanup_files
:
2529 exit_files(p
); /* blocking */
2530 bad_fork_cleanup_semundo
:
2532 bad_fork_cleanup_security
:
2533 security_task_free(p
);
2534 bad_fork_cleanup_audit
:
2536 bad_fork_cleanup_perf
:
2537 perf_event_free_task(p
);
2538 bad_fork_cleanup_policy
:
2539 lockdep_free_task(p
);
2541 mpol_put(p
->mempolicy
);
2543 bad_fork_cleanup_delayacct
:
2544 delayacct_tsk_free(p
);
2545 bad_fork_cleanup_count
:
2546 dec_rlimit_ucounts(task_ucounts(p
), UCOUNT_RLIMIT_NPROC
, 1);
2549 WRITE_ONCE(p
->__state
, TASK_DEAD
);
2550 exit_task_stack_account(p
);
2552 delayed_free_task(p
);
2554 spin_lock_irq(¤t
->sighand
->siglock
);
2555 hlist_del_init(&delayed
.node
);
2556 spin_unlock_irq(¤t
->sighand
->siglock
);
2557 return ERR_PTR(retval
);
2560 static inline void init_idle_pids(struct task_struct
*idle
)
2564 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2565 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2566 init_task_pid(idle
, type
, &init_struct_pid
);
2570 static int idle_dummy(void *dummy
)
2572 /* This function is never called */
2576 struct task_struct
* __init
fork_idle(int cpu
)
2578 struct task_struct
*task
;
2579 struct kernel_clone_args args
= {
2587 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2588 if (!IS_ERR(task
)) {
2589 init_idle_pids(task
);
2590 init_idle(task
, cpu
);
2596 struct mm_struct
*copy_init_mm(void)
2598 return dup_mm(NULL
, &init_mm
);
2602 * This is like kernel_clone(), but shaved down and tailored to just
2603 * creating io_uring workers. It returns a created task, or an error pointer.
2604 * The returned task is inactive, and the caller must fire it up through
2605 * wake_up_new_task(p). All signals are blocked in the created task.
2607 struct task_struct
*create_io_thread(int (*fn
)(void *), void *arg
, int node
)
2609 unsigned long flags
= CLONE_FS
|CLONE_FILES
|CLONE_SIGHAND
|CLONE_THREAD
|
2611 struct kernel_clone_args args
= {
2612 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2613 CLONE_UNTRACED
) & ~CSIGNAL
),
2614 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2620 return copy_process(NULL
, 0, node
, &args
);
2624 * Ok, this is the main fork-routine.
2626 * It copies the process, and if successful kick-starts
2627 * it and waits for it to finish using the VM if required.
2629 * args->exit_signal is expected to be checked for sanity by the caller.
2631 pid_t
kernel_clone(struct kernel_clone_args
*args
)
2633 u64 clone_flags
= args
->flags
;
2634 struct completion vfork
;
2636 struct task_struct
*p
;
2641 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2642 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2643 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2644 * field in struct clone_args and it still doesn't make sense to have
2645 * them both point at the same memory location. Performing this check
2646 * here has the advantage that we don't need to have a separate helper
2647 * to check for legacy clone().
2649 if ((args
->flags
& CLONE_PIDFD
) &&
2650 (args
->flags
& CLONE_PARENT_SETTID
) &&
2651 (args
->pidfd
== args
->parent_tid
))
2655 * Determine whether and which event to report to ptracer. When
2656 * called from kernel_thread or CLONE_UNTRACED is explicitly
2657 * requested, no event is reported; otherwise, report if the event
2658 * for the type of forking is enabled.
2660 if (!(clone_flags
& CLONE_UNTRACED
)) {
2661 if (clone_flags
& CLONE_VFORK
)
2662 trace
= PTRACE_EVENT_VFORK
;
2663 else if (args
->exit_signal
!= SIGCHLD
)
2664 trace
= PTRACE_EVENT_CLONE
;
2666 trace
= PTRACE_EVENT_FORK
;
2668 if (likely(!ptrace_event_enabled(current
, trace
)))
2672 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2673 add_latent_entropy();
2679 * Do this prior waking up the new thread - the thread pointer
2680 * might get invalid after that point, if the thread exits quickly.
2682 trace_sched_process_fork(current
, p
);
2684 pid
= get_task_pid(p
, PIDTYPE_PID
);
2687 if (clone_flags
& CLONE_PARENT_SETTID
)
2688 put_user(nr
, args
->parent_tid
);
2690 if (clone_flags
& CLONE_VFORK
) {
2691 p
->vfork_done
= &vfork
;
2692 init_completion(&vfork
);
2696 wake_up_new_task(p
);
2698 /* forking complete and child started to run, tell ptracer */
2699 if (unlikely(trace
))
2700 ptrace_event_pid(trace
, pid
);
2702 if (clone_flags
& CLONE_VFORK
) {
2703 if (!wait_for_vfork_done(p
, &vfork
))
2704 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2712 * Create a kernel thread.
2714 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2716 struct kernel_clone_args args
= {
2717 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2718 CLONE_UNTRACED
) & ~CSIGNAL
),
2719 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2725 return kernel_clone(&args
);
2729 * Create a user mode thread.
2731 pid_t
user_mode_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2733 struct kernel_clone_args args
= {
2734 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2735 CLONE_UNTRACED
) & ~CSIGNAL
),
2736 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2741 return kernel_clone(&args
);
2744 #ifdef __ARCH_WANT_SYS_FORK
2745 SYSCALL_DEFINE0(fork
)
2748 struct kernel_clone_args args
= {
2749 .exit_signal
= SIGCHLD
,
2752 return kernel_clone(&args
);
2754 /* can not support in nommu mode */
2760 #ifdef __ARCH_WANT_SYS_VFORK
2761 SYSCALL_DEFINE0(vfork
)
2763 struct kernel_clone_args args
= {
2764 .flags
= CLONE_VFORK
| CLONE_VM
,
2765 .exit_signal
= SIGCHLD
,
2768 return kernel_clone(&args
);
2772 #ifdef __ARCH_WANT_SYS_CLONE
2773 #ifdef CONFIG_CLONE_BACKWARDS
2774 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2775 int __user
*, parent_tidptr
,
2777 int __user
*, child_tidptr
)
2778 #elif defined(CONFIG_CLONE_BACKWARDS2)
2779 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2780 int __user
*, parent_tidptr
,
2781 int __user
*, child_tidptr
,
2783 #elif defined(CONFIG_CLONE_BACKWARDS3)
2784 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2786 int __user
*, parent_tidptr
,
2787 int __user
*, child_tidptr
,
2790 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2791 int __user
*, parent_tidptr
,
2792 int __user
*, child_tidptr
,
2796 struct kernel_clone_args args
= {
2797 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2798 .pidfd
= parent_tidptr
,
2799 .child_tid
= child_tidptr
,
2800 .parent_tid
= parent_tidptr
,
2801 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2806 return kernel_clone(&args
);
2810 #ifdef __ARCH_WANT_SYS_CLONE3
2812 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2813 struct clone_args __user
*uargs
,
2817 struct clone_args args
;
2818 pid_t
*kset_tid
= kargs
->set_tid
;
2820 BUILD_BUG_ON(offsetofend(struct clone_args
, tls
) !=
2821 CLONE_ARGS_SIZE_VER0
);
2822 BUILD_BUG_ON(offsetofend(struct clone_args
, set_tid_size
) !=
2823 CLONE_ARGS_SIZE_VER1
);
2824 BUILD_BUG_ON(offsetofend(struct clone_args
, cgroup
) !=
2825 CLONE_ARGS_SIZE_VER2
);
2826 BUILD_BUG_ON(sizeof(struct clone_args
) != CLONE_ARGS_SIZE_VER2
);
2828 if (unlikely(usize
> PAGE_SIZE
))
2830 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2833 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2837 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2840 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2843 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2847 * Verify that higher 32bits of exit_signal are unset and that
2848 * it is a valid signal
2850 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2851 !valid_signal(args
.exit_signal
)))
2854 if ((args
.flags
& CLONE_INTO_CGROUP
) &&
2855 (args
.cgroup
> INT_MAX
|| usize
< CLONE_ARGS_SIZE_VER2
))
2858 *kargs
= (struct kernel_clone_args
){
2859 .flags
= args
.flags
,
2860 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2861 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2862 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2863 .exit_signal
= args
.exit_signal
,
2864 .stack
= args
.stack
,
2865 .stack_size
= args
.stack_size
,
2867 .set_tid_size
= args
.set_tid_size
,
2868 .cgroup
= args
.cgroup
,
2872 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
2873 (kargs
->set_tid_size
* sizeof(pid_t
))))
2876 kargs
->set_tid
= kset_tid
;
2882 * clone3_stack_valid - check and prepare stack
2883 * @kargs: kernel clone args
2885 * Verify that the stack arguments userspace gave us are sane.
2886 * In addition, set the stack direction for userspace since it's easy for us to
2889 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2891 if (kargs
->stack
== 0) {
2892 if (kargs
->stack_size
> 0)
2895 if (kargs
->stack_size
== 0)
2898 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2901 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2902 kargs
->stack
+= kargs
->stack_size
;
2909 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2911 /* Verify that no unknown flags are passed along. */
2913 ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
| CLONE_INTO_CGROUP
))
2917 * - make the CLONE_DETACHED bit reusable for clone3
2918 * - make the CSIGNAL bits reusable for clone3
2920 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2923 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
2924 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
2927 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2931 if (!clone3_stack_valid(kargs
))
2938 * clone3 - create a new process with specific properties
2939 * @uargs: argument structure
2940 * @size: size of @uargs
2942 * clone3() is the extensible successor to clone()/clone2().
2943 * It takes a struct as argument that is versioned by its size.
2945 * Return: On success, a positive PID for the child process.
2946 * On error, a negative errno number.
2948 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2952 struct kernel_clone_args kargs
;
2953 pid_t set_tid
[MAX_PID_NS_LEVEL
];
2955 kargs
.set_tid
= set_tid
;
2957 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2961 if (!clone3_args_valid(&kargs
))
2964 return kernel_clone(&kargs
);
2968 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2970 struct task_struct
*leader
, *parent
, *child
;
2973 read_lock(&tasklist_lock
);
2974 leader
= top
= top
->group_leader
;
2976 for_each_thread(leader
, parent
) {
2977 list_for_each_entry(child
, &parent
->children
, sibling
) {
2978 res
= visitor(child
, data
);
2990 if (leader
!= top
) {
2992 parent
= child
->real_parent
;
2993 leader
= parent
->group_leader
;
2997 read_unlock(&tasklist_lock
);
3000 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3001 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3004 static void sighand_ctor(void *data
)
3006 struct sighand_struct
*sighand
= data
;
3008 spin_lock_init(&sighand
->siglock
);
3009 init_waitqueue_head(&sighand
->signalfd_wqh
);
3012 void __init
proc_caches_init(void)
3014 unsigned int mm_size
;
3016 sighand_cachep
= kmem_cache_create("sighand_cache",
3017 sizeof(struct sighand_struct
), 0,
3018 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
3019 SLAB_ACCOUNT
, sighand_ctor
);
3020 signal_cachep
= kmem_cache_create("signal_cache",
3021 sizeof(struct signal_struct
), 0,
3022 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3024 files_cachep
= kmem_cache_create("files_cache",
3025 sizeof(struct files_struct
), 0,
3026 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3028 fs_cachep
= kmem_cache_create("fs_cache",
3029 sizeof(struct fs_struct
), 0,
3030 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3034 * The mm_cpumask is located at the end of mm_struct, and is
3035 * dynamically sized based on the maximum CPU number this system
3036 * can have, taking hotplug into account (nr_cpu_ids).
3038 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
3040 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
3041 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
3042 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3043 offsetof(struct mm_struct
, saved_auxv
),
3044 sizeof_field(struct mm_struct
, saved_auxv
),
3046 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
3048 nsproxy_cache_init();
3052 * Check constraints on flags passed to the unshare system call.
3054 static int check_unshare_flags(unsigned long unshare_flags
)
3056 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
3057 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
3058 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
3059 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
|
3063 * Not implemented, but pretend it works if there is nothing
3064 * to unshare. Note that unsharing the address space or the
3065 * signal handlers also need to unshare the signal queues (aka
3068 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
3069 if (!thread_group_empty(current
))
3072 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
3073 if (refcount_read(¤t
->sighand
->count
) > 1)
3076 if (unshare_flags
& CLONE_VM
) {
3077 if (!current_is_single_threaded())
3085 * Unshare the filesystem structure if it is being shared
3087 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
3089 struct fs_struct
*fs
= current
->fs
;
3091 if (!(unshare_flags
& CLONE_FS
) || !fs
)
3094 /* don't need lock here; in the worst case we'll do useless copy */
3098 *new_fsp
= copy_fs_struct(fs
);
3106 * Unshare file descriptor table if it is being shared
3108 int unshare_fd(unsigned long unshare_flags
, unsigned int max_fds
,
3109 struct files_struct
**new_fdp
)
3111 struct files_struct
*fd
= current
->files
;
3114 if ((unshare_flags
& CLONE_FILES
) &&
3115 (fd
&& atomic_read(&fd
->count
) > 1)) {
3116 *new_fdp
= dup_fd(fd
, max_fds
, &error
);
3125 * unshare allows a process to 'unshare' part of the process
3126 * context which was originally shared using clone. copy_*
3127 * functions used by kernel_clone() cannot be used here directly
3128 * because they modify an inactive task_struct that is being
3129 * constructed. Here we are modifying the current, active,
3132 int ksys_unshare(unsigned long unshare_flags
)
3134 struct fs_struct
*fs
, *new_fs
= NULL
;
3135 struct files_struct
*new_fd
= NULL
;
3136 struct cred
*new_cred
= NULL
;
3137 struct nsproxy
*new_nsproxy
= NULL
;
3142 * If unsharing a user namespace must also unshare the thread group
3143 * and unshare the filesystem root and working directories.
3145 if (unshare_flags
& CLONE_NEWUSER
)
3146 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
3148 * If unsharing vm, must also unshare signal handlers.
3150 if (unshare_flags
& CLONE_VM
)
3151 unshare_flags
|= CLONE_SIGHAND
;
3153 * If unsharing a signal handlers, must also unshare the signal queues.
3155 if (unshare_flags
& CLONE_SIGHAND
)
3156 unshare_flags
|= CLONE_THREAD
;
3158 * If unsharing namespace, must also unshare filesystem information.
3160 if (unshare_flags
& CLONE_NEWNS
)
3161 unshare_flags
|= CLONE_FS
;
3163 err
= check_unshare_flags(unshare_flags
);
3165 goto bad_unshare_out
;
3167 * CLONE_NEWIPC must also detach from the undolist: after switching
3168 * to a new ipc namespace, the semaphore arrays from the old
3169 * namespace are unreachable.
3171 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
3173 err
= unshare_fs(unshare_flags
, &new_fs
);
3175 goto bad_unshare_out
;
3176 err
= unshare_fd(unshare_flags
, NR_OPEN_MAX
, &new_fd
);
3178 goto bad_unshare_cleanup_fs
;
3179 err
= unshare_userns(unshare_flags
, &new_cred
);
3181 goto bad_unshare_cleanup_fd
;
3182 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
3185 goto bad_unshare_cleanup_cred
;
3188 err
= set_cred_ucounts(new_cred
);
3190 goto bad_unshare_cleanup_cred
;
3193 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
3196 * CLONE_SYSVSEM is equivalent to sys_exit().
3200 if (unshare_flags
& CLONE_NEWIPC
) {
3201 /* Orphan segments in old ns (see sem above). */
3203 shm_init_task(current
);
3207 switch_task_namespaces(current
, new_nsproxy
);
3213 spin_lock(&fs
->lock
);
3214 current
->fs
= new_fs
;
3219 spin_unlock(&fs
->lock
);
3223 swap(current
->files
, new_fd
);
3225 task_unlock(current
);
3228 /* Install the new user namespace */
3229 commit_creds(new_cred
);
3234 perf_event_namespaces(current
);
3236 bad_unshare_cleanup_cred
:
3239 bad_unshare_cleanup_fd
:
3241 put_files_struct(new_fd
);
3243 bad_unshare_cleanup_fs
:
3245 free_fs_struct(new_fs
);
3251 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3253 return ksys_unshare(unshare_flags
);
3257 * Helper to unshare the files of the current task.
3258 * We don't want to expose copy_files internals to
3259 * the exec layer of the kernel.
3262 int unshare_files(void)
3264 struct task_struct
*task
= current
;
3265 struct files_struct
*old
, *copy
= NULL
;
3268 error
= unshare_fd(CLONE_FILES
, NR_OPEN_MAX
, ©
);
3276 put_files_struct(old
);
3280 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3281 void *buffer
, size_t *lenp
, loff_t
*ppos
)
3285 int threads
= max_threads
;
3287 int max
= MAX_THREADS
;
3294 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3298 max_threads
= threads
;