1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
100 #include <asm/pgalloc.h>
101 #include <linux/uaccess.h>
102 #include <asm/mmu_context.h>
103 #include <asm/cacheflush.h>
104 #include <asm/tlbflush.h>
106 #include <trace/events/sched.h>
108 #define CREATE_TRACE_POINTS
109 #include <trace/events/task.h>
112 * Minimum number of threads to boot the kernel
114 #define MIN_THREADS 20
117 * Maximum number of threads
119 #define MAX_THREADS FUTEX_TID_MASK
122 * Protected counters by write_lock_irq(&tasklist_lock)
124 unsigned long total_forks
; /* Handle normal Linux uptimes. */
125 int nr_threads
; /* The idle threads do not count.. */
127 static int max_threads
; /* tunable limit on nr_threads */
129 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
131 static const char * const resident_page_types
[] = {
132 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
133 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
134 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
135 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
138 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
140 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
142 #ifdef CONFIG_PROVE_RCU
143 int lockdep_tasklist_lock_is_held(void)
145 return lockdep_is_held(&tasklist_lock
);
147 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
148 #endif /* #ifdef CONFIG_PROVE_RCU */
150 int nr_processes(void)
155 for_each_possible_cpu(cpu
)
156 total
+= per_cpu(process_counts
, cpu
);
161 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
165 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
166 static struct kmem_cache
*task_struct_cachep
;
168 static inline struct task_struct
*alloc_task_struct_node(int node
)
170 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
173 static inline void free_task_struct(struct task_struct
*tsk
)
175 kmem_cache_free(task_struct_cachep
, tsk
);
179 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
182 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
183 * kmemcache based allocator.
185 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
187 #ifdef CONFIG_VMAP_STACK
189 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
190 * flush. Try to minimize the number of calls by caching stacks.
192 #define NR_CACHED_STACKS 2
193 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
195 static int free_vm_stack_cache(unsigned int cpu
)
197 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
200 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
201 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
206 vfree(vm_stack
->addr
);
207 cached_vm_stacks
[i
] = NULL
;
214 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
216 #ifdef CONFIG_VMAP_STACK
220 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
223 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
228 /* Clear the KASAN shadow of the stack. */
229 kasan_unpoison_shadow(s
->addr
, THREAD_SIZE
);
231 /* Clear stale pointers from reused stack. */
232 memset(s
->addr
, 0, THREAD_SIZE
);
234 tsk
->stack_vm_area
= s
;
235 tsk
->stack
= s
->addr
;
240 * Allocated stacks are cached and later reused by new threads,
241 * so memcg accounting is performed manually on assigning/releasing
242 * stacks to tasks. Drop __GFP_ACCOUNT.
244 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
245 VMALLOC_START
, VMALLOC_END
,
246 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
248 0, node
, __builtin_return_address(0));
251 * We can't call find_vm_area() in interrupt context, and
252 * free_thread_stack() can be called in interrupt context,
253 * so cache the vm_struct.
256 tsk
->stack_vm_area
= find_vm_area(stack
);
261 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
265 tsk
->stack
= kasan_reset_tag(page_address(page
));
272 static inline void free_thread_stack(struct task_struct
*tsk
)
274 #ifdef CONFIG_VMAP_STACK
275 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
280 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
281 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
283 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
284 if (this_cpu_cmpxchg(cached_stacks
[i
],
285 NULL
, tsk
->stack_vm_area
) != NULL
)
291 vfree_atomic(tsk
->stack
);
296 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
299 static struct kmem_cache
*thread_stack_cache
;
301 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
304 unsigned long *stack
;
305 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
306 stack
= kasan_reset_tag(stack
);
311 static void free_thread_stack(struct task_struct
*tsk
)
313 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
316 void thread_stack_cache_init(void)
318 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
319 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
321 BUG_ON(thread_stack_cache
== NULL
);
326 /* SLAB cache for signal_struct structures (tsk->signal) */
327 static struct kmem_cache
*signal_cachep
;
329 /* SLAB cache for sighand_struct structures (tsk->sighand) */
330 struct kmem_cache
*sighand_cachep
;
332 /* SLAB cache for files_struct structures (tsk->files) */
333 struct kmem_cache
*files_cachep
;
335 /* SLAB cache for fs_struct structures (tsk->fs) */
336 struct kmem_cache
*fs_cachep
;
338 /* SLAB cache for vm_area_struct structures */
339 static struct kmem_cache
*vm_area_cachep
;
341 /* SLAB cache for mm_struct structures (tsk->mm) */
342 static struct kmem_cache
*mm_cachep
;
344 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
346 struct vm_area_struct
*vma
;
348 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
354 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
356 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
359 ASSERT_EXCLUSIVE_WRITER(orig
->vm_flags
);
360 ASSERT_EXCLUSIVE_WRITER(orig
->vm_file
);
362 * orig->shared.rb may be modified concurrently, but the clone
363 * will be reinitialized.
365 *new = data_race(*orig
);
366 INIT_LIST_HEAD(&new->anon_vma_chain
);
367 new->vm_next
= new->vm_prev
= NULL
;
372 void vm_area_free(struct vm_area_struct
*vma
)
374 kmem_cache_free(vm_area_cachep
, vma
);
377 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
379 void *stack
= task_stack_page(tsk
);
380 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
383 /* All stack pages are in the same node. */
385 mod_lruvec_page_state(vm
->pages
[0], NR_KERNEL_STACK_KB
,
386 account
* (THREAD_SIZE
/ 1024));
388 mod_lruvec_slab_state(stack
, NR_KERNEL_STACK_KB
,
389 account
* (THREAD_SIZE
/ 1024));
392 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
394 #ifdef CONFIG_VMAP_STACK
395 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
398 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
403 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
405 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
407 * If memcg_kmem_charge_page() fails, page->mem_cgroup
408 * pointer is NULL, and memcg_kmem_uncharge_page() in
409 * free_thread_stack() will ignore this page.
411 ret
= memcg_kmem_charge_page(vm
->pages
[i
], GFP_KERNEL
,
421 static void release_task_stack(struct task_struct
*tsk
)
423 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
424 return; /* Better to leak the stack than to free prematurely */
426 account_kernel_stack(tsk
, -1);
427 free_thread_stack(tsk
);
429 #ifdef CONFIG_VMAP_STACK
430 tsk
->stack_vm_area
= NULL
;
434 #ifdef CONFIG_THREAD_INFO_IN_TASK
435 void put_task_stack(struct task_struct
*tsk
)
437 if (refcount_dec_and_test(&tsk
->stack_refcount
))
438 release_task_stack(tsk
);
442 void free_task(struct task_struct
*tsk
)
446 #ifndef CONFIG_THREAD_INFO_IN_TASK
448 * The task is finally done with both the stack and thread_info,
451 release_task_stack(tsk
);
454 * If the task had a separate stack allocation, it should be gone
457 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
459 rt_mutex_debug_task_free(tsk
);
460 ftrace_graph_exit_task(tsk
);
461 arch_release_task_struct(tsk
);
462 if (tsk
->flags
& PF_KTHREAD
)
463 free_kthread_struct(tsk
);
464 free_task_struct(tsk
);
466 EXPORT_SYMBOL(free_task
);
469 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
470 struct mm_struct
*oldmm
)
472 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
473 struct rb_node
**rb_link
, *rb_parent
;
475 unsigned long charge
;
478 uprobe_start_dup_mmap();
479 if (mmap_write_lock_killable(oldmm
)) {
481 goto fail_uprobe_end
;
483 flush_cache_dup_mm(oldmm
);
484 uprobe_dup_mmap(oldmm
, mm
);
486 * Not linked in yet - no deadlock potential:
488 mmap_write_lock_nested(mm
, SINGLE_DEPTH_NESTING
);
490 /* No ordering required: file already has been exposed. */
491 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
493 mm
->total_vm
= oldmm
->total_vm
;
494 mm
->data_vm
= oldmm
->data_vm
;
495 mm
->exec_vm
= oldmm
->exec_vm
;
496 mm
->stack_vm
= oldmm
->stack_vm
;
498 rb_link
= &mm
->mm_rb
.rb_node
;
501 retval
= ksm_fork(mm
, oldmm
);
504 retval
= khugepaged_fork(mm
, oldmm
);
509 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
512 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
513 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
518 * Don't duplicate many vmas if we've been oom-killed (for
521 if (fatal_signal_pending(current
)) {
525 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
526 unsigned long len
= vma_pages(mpnt
);
528 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
532 tmp
= vm_area_dup(mpnt
);
535 retval
= vma_dup_policy(mpnt
, tmp
);
537 goto fail_nomem_policy
;
539 retval
= dup_userfaultfd(tmp
, &uf
);
541 goto fail_nomem_anon_vma_fork
;
542 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
544 * VM_WIPEONFORK gets a clean slate in the child.
545 * Don't prepare anon_vma until fault since we don't
546 * copy page for current vma.
548 tmp
->anon_vma
= NULL
;
549 } else if (anon_vma_fork(tmp
, mpnt
))
550 goto fail_nomem_anon_vma_fork
;
551 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
554 struct inode
*inode
= file_inode(file
);
555 struct address_space
*mapping
= file
->f_mapping
;
558 if (tmp
->vm_flags
& VM_DENYWRITE
)
559 put_write_access(inode
);
560 i_mmap_lock_write(mapping
);
561 if (tmp
->vm_flags
& VM_SHARED
)
562 mapping_allow_writable(mapping
);
563 flush_dcache_mmap_lock(mapping
);
564 /* insert tmp into the share list, just after mpnt */
565 vma_interval_tree_insert_after(tmp
, mpnt
,
567 flush_dcache_mmap_unlock(mapping
);
568 i_mmap_unlock_write(mapping
);
572 * Clear hugetlb-related page reserves for children. This only
573 * affects MAP_PRIVATE mappings. Faults generated by the child
574 * are not guaranteed to succeed, even if read-only
576 if (is_vm_hugetlb_page(tmp
))
577 reset_vma_resv_huge_pages(tmp
);
580 * Link in the new vma and copy the page table entries.
583 pprev
= &tmp
->vm_next
;
587 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
588 rb_link
= &tmp
->vm_rb
.rb_right
;
589 rb_parent
= &tmp
->vm_rb
;
592 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
593 retval
= copy_page_range(tmp
, mpnt
);
595 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
596 tmp
->vm_ops
->open(tmp
);
601 /* a new mm has just been created */
602 retval
= arch_dup_mmap(oldmm
, mm
);
604 mmap_write_unlock(mm
);
606 mmap_write_unlock(oldmm
);
607 dup_userfaultfd_complete(&uf
);
609 uprobe_end_dup_mmap();
611 fail_nomem_anon_vma_fork
:
612 mpol_put(vma_policy(tmp
));
617 vm_unacct_memory(charge
);
621 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
623 mm
->pgd
= pgd_alloc(mm
);
624 if (unlikely(!mm
->pgd
))
629 static inline void mm_free_pgd(struct mm_struct
*mm
)
631 pgd_free(mm
, mm
->pgd
);
634 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
636 mmap_write_lock(oldmm
);
637 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
638 mmap_write_unlock(oldmm
);
641 #define mm_alloc_pgd(mm) (0)
642 #define mm_free_pgd(mm)
643 #endif /* CONFIG_MMU */
645 static void check_mm(struct mm_struct
*mm
)
649 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
650 "Please make sure 'struct resident_page_types[]' is updated as well");
652 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
653 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
656 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
657 mm
, resident_page_types
[i
], x
);
660 if (mm_pgtables_bytes(mm
))
661 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
662 mm_pgtables_bytes(mm
));
664 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
665 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
669 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
670 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
673 * Called when the last reference to the mm
674 * is dropped: either by a lazy thread or by
675 * mmput. Free the page directory and the mm.
677 void __mmdrop(struct mm_struct
*mm
)
679 BUG_ON(mm
== &init_mm
);
680 WARN_ON_ONCE(mm
== current
->mm
);
681 WARN_ON_ONCE(mm
== current
->active_mm
);
684 mmu_notifier_subscriptions_destroy(mm
);
686 put_user_ns(mm
->user_ns
);
689 EXPORT_SYMBOL_GPL(__mmdrop
);
691 static void mmdrop_async_fn(struct work_struct
*work
)
693 struct mm_struct
*mm
;
695 mm
= container_of(work
, struct mm_struct
, async_put_work
);
699 static void mmdrop_async(struct mm_struct
*mm
)
701 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
702 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
703 schedule_work(&mm
->async_put_work
);
707 static inline void free_signal_struct(struct signal_struct
*sig
)
709 taskstats_tgid_free(sig
);
710 sched_autogroup_exit(sig
);
712 * __mmdrop is not safe to call from softirq context on x86 due to
713 * pgd_dtor so postpone it to the async context
716 mmdrop_async(sig
->oom_mm
);
717 kmem_cache_free(signal_cachep
, sig
);
720 static inline void put_signal_struct(struct signal_struct
*sig
)
722 if (refcount_dec_and_test(&sig
->sigcnt
))
723 free_signal_struct(sig
);
726 void __put_task_struct(struct task_struct
*tsk
)
728 WARN_ON(!tsk
->exit_state
);
729 WARN_ON(refcount_read(&tsk
->usage
));
730 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 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
777 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
779 /* Fetch thread_struct whitelist for the architecture. */
780 arch_thread_struct_whitelist(offset
, size
);
783 * Handle zero-sized whitelist or empty thread_struct, otherwise
784 * adjust offset to position of thread_struct in task_struct.
786 if (unlikely(*size
== 0))
789 *offset
+= offsetof(struct task_struct
, thread
);
791 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
793 void __init
fork_init(void)
796 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
797 #ifndef ARCH_MIN_TASKALIGN
798 #define ARCH_MIN_TASKALIGN 0
800 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
801 unsigned long useroffset
, usersize
;
803 /* create a slab on which task_structs can be allocated */
804 task_struct_whitelist(&useroffset
, &usersize
);
805 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
806 arch_task_struct_size
, align
,
807 SLAB_PANIC
|SLAB_ACCOUNT
,
808 useroffset
, usersize
, NULL
);
811 /* do the arch specific task caches init */
812 arch_task_cache_init();
814 set_max_threads(MAX_THREADS
);
816 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
817 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
818 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
819 init_task
.signal
->rlim
[RLIMIT_NPROC
];
821 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
822 init_user_ns
.ucount_max
[i
] = max_threads
/2;
825 #ifdef CONFIG_VMAP_STACK
826 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
827 NULL
, free_vm_stack_cache
);
832 lockdep_init_task(&init_task
);
836 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
837 struct task_struct
*src
)
843 void set_task_stack_end_magic(struct task_struct
*tsk
)
845 unsigned long *stackend
;
847 stackend
= end_of_stack(tsk
);
848 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
851 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
853 struct task_struct
*tsk
;
854 unsigned long *stack
;
855 struct vm_struct
*stack_vm_area __maybe_unused
;
858 if (node
== NUMA_NO_NODE
)
859 node
= tsk_fork_get_node(orig
);
860 tsk
= alloc_task_struct_node(node
);
864 stack
= alloc_thread_stack_node(tsk
, node
);
868 if (memcg_charge_kernel_stack(tsk
))
871 stack_vm_area
= task_stack_vm_area(tsk
);
873 err
= arch_dup_task_struct(tsk
, orig
);
876 * arch_dup_task_struct() clobbers the stack-related fields. Make
877 * sure they're properly initialized before using any stack-related
881 #ifdef CONFIG_VMAP_STACK
882 tsk
->stack_vm_area
= stack_vm_area
;
884 #ifdef CONFIG_THREAD_INFO_IN_TASK
885 refcount_set(&tsk
->stack_refcount
, 1);
891 err
= scs_prepare(tsk
, node
);
895 #ifdef CONFIG_SECCOMP
897 * We must handle setting up seccomp filters once we're under
898 * the sighand lock in case orig has changed between now and
899 * then. Until then, filter must be NULL to avoid messing up
900 * the usage counts on the error path calling free_task.
902 tsk
->seccomp
.filter
= NULL
;
905 setup_thread_stack(tsk
, orig
);
906 clear_user_return_notifier(tsk
);
907 clear_tsk_need_resched(tsk
);
908 set_task_stack_end_magic(tsk
);
910 #ifdef CONFIG_STACKPROTECTOR
911 tsk
->stack_canary
= get_random_canary();
913 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
914 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
917 * One for the user space visible state that goes away when reaped.
918 * One for the scheduler.
920 refcount_set(&tsk
->rcu_users
, 2);
921 /* One for the rcu users */
922 refcount_set(&tsk
->usage
, 1);
923 #ifdef CONFIG_BLK_DEV_IO_TRACE
926 tsk
->splice_pipe
= NULL
;
927 tsk
->task_frag
.page
= NULL
;
928 tsk
->wake_q
.next
= NULL
;
930 account_kernel_stack(tsk
, 1);
934 #ifdef CONFIG_FAULT_INJECTION
938 #ifdef CONFIG_BLK_CGROUP
939 tsk
->throttle_queue
= NULL
;
940 tsk
->use_memdelay
= 0;
944 tsk
->active_memcg
= NULL
;
949 free_thread_stack(tsk
);
951 free_task_struct(tsk
);
955 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
957 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
959 static int __init
coredump_filter_setup(char *s
)
961 default_dump_filter
=
962 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
963 MMF_DUMP_FILTER_MASK
;
967 __setup("coredump_filter=", coredump_filter_setup
);
969 #include <linux/init_task.h>
971 static void mm_init_aio(struct mm_struct
*mm
)
974 spin_lock_init(&mm
->ioctx_lock
);
975 mm
->ioctx_table
= NULL
;
979 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
980 struct task_struct
*p
)
984 WRITE_ONCE(mm
->owner
, NULL
);
988 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
995 static void mm_init_uprobes_state(struct mm_struct
*mm
)
997 #ifdef CONFIG_UPROBES
998 mm
->uprobes_state
.xol_area
= NULL
;
1002 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1003 struct user_namespace
*user_ns
)
1006 mm
->mm_rb
= RB_ROOT
;
1007 mm
->vmacache_seqnum
= 0;
1008 atomic_set(&mm
->mm_users
, 1);
1009 atomic_set(&mm
->mm_count
, 1);
1011 INIT_LIST_HEAD(&mm
->mmlist
);
1012 mm
->core_state
= NULL
;
1013 mm_pgtables_bytes_init(mm
);
1016 atomic_set(&mm
->has_pinned
, 0);
1017 atomic64_set(&mm
->pinned_vm
, 0);
1018 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1019 spin_lock_init(&mm
->page_table_lock
);
1020 spin_lock_init(&mm
->arg_lock
);
1021 mm_init_cpumask(mm
);
1023 mm_init_owner(mm
, p
);
1024 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1025 mmu_notifier_subscriptions_init(mm
);
1026 init_tlb_flush_pending(mm
);
1027 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1028 mm
->pmd_huge_pte
= NULL
;
1030 mm_init_uprobes_state(mm
);
1033 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1034 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1036 mm
->flags
= default_dump_filter
;
1040 if (mm_alloc_pgd(mm
))
1043 if (init_new_context(p
, mm
))
1044 goto fail_nocontext
;
1046 mm
->user_ns
= get_user_ns(user_ns
);
1057 * Allocate and initialize an mm_struct.
1059 struct mm_struct
*mm_alloc(void)
1061 struct mm_struct
*mm
;
1067 memset(mm
, 0, sizeof(*mm
));
1068 return mm_init(mm
, current
, current_user_ns());
1071 static inline void __mmput(struct mm_struct
*mm
)
1073 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1075 uprobe_clear_state(mm
);
1078 khugepaged_exit(mm
); /* must run before exit_mmap */
1080 mm_put_huge_zero_page(mm
);
1081 set_mm_exe_file(mm
, NULL
);
1082 if (!list_empty(&mm
->mmlist
)) {
1083 spin_lock(&mmlist_lock
);
1084 list_del(&mm
->mmlist
);
1085 spin_unlock(&mmlist_lock
);
1088 module_put(mm
->binfmt
->module
);
1093 * Decrement the use count and release all resources for an mm.
1095 void mmput(struct mm_struct
*mm
)
1099 if (atomic_dec_and_test(&mm
->mm_users
))
1102 EXPORT_SYMBOL_GPL(mmput
);
1105 static void mmput_async_fn(struct work_struct
*work
)
1107 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1113 void mmput_async(struct mm_struct
*mm
)
1115 if (atomic_dec_and_test(&mm
->mm_users
)) {
1116 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1117 schedule_work(&mm
->async_put_work
);
1123 * set_mm_exe_file - change a reference to the mm's executable file
1125 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1127 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1128 * invocations: in mmput() nobody alive left, in execve task is single
1129 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1130 * mm->exe_file, but does so without using set_mm_exe_file() in order
1131 * to do avoid the need for any locks.
1133 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1135 struct file
*old_exe_file
;
1138 * It is safe to dereference the exe_file without RCU as
1139 * this function is only called if nobody else can access
1140 * this mm -- see comment above for justification.
1142 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1145 get_file(new_exe_file
);
1146 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1152 * get_mm_exe_file - acquire a reference to the mm's executable file
1154 * Returns %NULL if mm has no associated executable file.
1155 * User must release file via fput().
1157 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1159 struct file
*exe_file
;
1162 exe_file
= rcu_dereference(mm
->exe_file
);
1163 if (exe_file
&& !get_file_rcu(exe_file
))
1168 EXPORT_SYMBOL(get_mm_exe_file
);
1171 * get_task_exe_file - acquire a reference to the task's executable file
1173 * Returns %NULL if task's mm (if any) has no associated executable file or
1174 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1175 * User must release file via fput().
1177 struct file
*get_task_exe_file(struct task_struct
*task
)
1179 struct file
*exe_file
= NULL
;
1180 struct mm_struct
*mm
;
1185 if (!(task
->flags
& PF_KTHREAD
))
1186 exe_file
= get_mm_exe_file(mm
);
1191 EXPORT_SYMBOL(get_task_exe_file
);
1194 * get_task_mm - acquire a reference to the task's mm
1196 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1197 * this kernel workthread has transiently adopted a user mm with use_mm,
1198 * to do its AIO) is not set and if so returns a reference to it, after
1199 * bumping up the use count. User must release the mm via mmput()
1200 * after use. Typically used by /proc and ptrace.
1202 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1204 struct mm_struct
*mm
;
1209 if (task
->flags
& PF_KTHREAD
)
1217 EXPORT_SYMBOL_GPL(get_task_mm
);
1219 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1221 struct mm_struct
*mm
;
1224 err
= mutex_lock_killable(&task
->signal
->exec_update_mutex
);
1226 return ERR_PTR(err
);
1228 mm
= get_task_mm(task
);
1229 if (mm
&& mm
!= current
->mm
&&
1230 !ptrace_may_access(task
, mode
)) {
1232 mm
= ERR_PTR(-EACCES
);
1234 mutex_unlock(&task
->signal
->exec_update_mutex
);
1239 static void complete_vfork_done(struct task_struct
*tsk
)
1241 struct completion
*vfork
;
1244 vfork
= tsk
->vfork_done
;
1245 if (likely(vfork
)) {
1246 tsk
->vfork_done
= NULL
;
1252 static int wait_for_vfork_done(struct task_struct
*child
,
1253 struct completion
*vfork
)
1257 freezer_do_not_count();
1258 cgroup_enter_frozen();
1259 killed
= wait_for_completion_killable(vfork
);
1260 cgroup_leave_frozen(false);
1265 child
->vfork_done
= NULL
;
1269 put_task_struct(child
);
1273 /* Please note the differences between mmput and mm_release.
1274 * mmput is called whenever we stop holding onto a mm_struct,
1275 * error success whatever.
1277 * mm_release is called after a mm_struct has been removed
1278 * from the current process.
1280 * This difference is important for error handling, when we
1281 * only half set up a mm_struct for a new process and need to restore
1282 * the old one. Because we mmput the new mm_struct before
1283 * restoring the old one. . .
1284 * Eric Biederman 10 January 1998
1286 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1288 uprobe_free_utask(tsk
);
1290 /* Get rid of any cached register state */
1291 deactivate_mm(tsk
, mm
);
1294 * Signal userspace if we're not exiting with a core dump
1295 * because we want to leave the value intact for debugging
1298 if (tsk
->clear_child_tid
) {
1299 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1300 atomic_read(&mm
->mm_users
) > 1) {
1302 * We don't check the error code - if userspace has
1303 * not set up a proper pointer then tough luck.
1305 put_user(0, tsk
->clear_child_tid
);
1306 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1307 1, NULL
, NULL
, 0, 0);
1309 tsk
->clear_child_tid
= NULL
;
1313 * All done, finally we can wake up parent and return this mm to him.
1314 * Also kthread_stop() uses this completion for synchronization.
1316 if (tsk
->vfork_done
)
1317 complete_vfork_done(tsk
);
1320 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1322 futex_exit_release(tsk
);
1323 mm_release(tsk
, mm
);
1326 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1328 futex_exec_release(tsk
);
1329 mm_release(tsk
, mm
);
1333 * dup_mm() - duplicates an existing mm structure
1334 * @tsk: the task_struct with which the new mm will be associated.
1335 * @oldmm: the mm to duplicate.
1337 * Allocates a new mm structure and duplicates the provided @oldmm structure
1340 * Return: the duplicated mm or NULL on failure.
1342 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1343 struct mm_struct
*oldmm
)
1345 struct mm_struct
*mm
;
1352 memcpy(mm
, oldmm
, sizeof(*mm
));
1354 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1357 err
= dup_mmap(mm
, oldmm
);
1361 mm
->hiwater_rss
= get_mm_rss(mm
);
1362 mm
->hiwater_vm
= mm
->total_vm
;
1364 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1370 /* don't put binfmt in mmput, we haven't got module yet */
1372 mm_init_owner(mm
, NULL
);
1379 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1381 struct mm_struct
*mm
, *oldmm
;
1384 tsk
->min_flt
= tsk
->maj_flt
= 0;
1385 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1386 #ifdef CONFIG_DETECT_HUNG_TASK
1387 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1388 tsk
->last_switch_time
= 0;
1392 tsk
->active_mm
= NULL
;
1395 * Are we cloning a kernel thread?
1397 * We need to steal a active VM for that..
1399 oldmm
= current
->mm
;
1403 /* initialize the new vmacache entries */
1404 vmacache_flush(tsk
);
1406 if (clone_flags
& CLONE_VM
) {
1413 mm
= dup_mm(tsk
, current
->mm
);
1419 tsk
->active_mm
= mm
;
1426 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1428 struct fs_struct
*fs
= current
->fs
;
1429 if (clone_flags
& CLONE_FS
) {
1430 /* tsk->fs is already what we want */
1431 spin_lock(&fs
->lock
);
1433 spin_unlock(&fs
->lock
);
1437 spin_unlock(&fs
->lock
);
1440 tsk
->fs
= copy_fs_struct(fs
);
1446 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1448 struct files_struct
*oldf
, *newf
;
1452 * A background process may not have any files ...
1454 oldf
= current
->files
;
1458 if (clone_flags
& CLONE_FILES
) {
1459 atomic_inc(&oldf
->count
);
1463 newf
= dup_fd(oldf
, NR_OPEN_MAX
, &error
);
1473 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1476 struct io_context
*ioc
= current
->io_context
;
1477 struct io_context
*new_ioc
;
1482 * Share io context with parent, if CLONE_IO is set
1484 if (clone_flags
& CLONE_IO
) {
1486 tsk
->io_context
= ioc
;
1487 } else if (ioprio_valid(ioc
->ioprio
)) {
1488 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1489 if (unlikely(!new_ioc
))
1492 new_ioc
->ioprio
= ioc
->ioprio
;
1493 put_io_context(new_ioc
);
1499 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1501 struct sighand_struct
*sig
;
1503 if (clone_flags
& CLONE_SIGHAND
) {
1504 refcount_inc(¤t
->sighand
->count
);
1507 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1508 RCU_INIT_POINTER(tsk
->sighand
, sig
);
1512 refcount_set(&sig
->count
, 1);
1513 spin_lock_irq(¤t
->sighand
->siglock
);
1514 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1515 spin_unlock_irq(¤t
->sighand
->siglock
);
1517 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1518 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1519 flush_signal_handlers(tsk
, 0);
1524 void __cleanup_sighand(struct sighand_struct
*sighand
)
1526 if (refcount_dec_and_test(&sighand
->count
)) {
1527 signalfd_cleanup(sighand
);
1529 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1530 * without an RCU grace period, see __lock_task_sighand().
1532 kmem_cache_free(sighand_cachep
, sighand
);
1537 * Initialize POSIX timer handling for a thread group.
1539 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1541 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1542 unsigned long cpu_limit
;
1544 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1545 posix_cputimers_group_init(pct
, cpu_limit
);
1548 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1550 struct signal_struct
*sig
;
1552 if (clone_flags
& CLONE_THREAD
)
1555 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1560 sig
->nr_threads
= 1;
1561 atomic_set(&sig
->live
, 1);
1562 refcount_set(&sig
->sigcnt
, 1);
1564 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1565 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1566 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1568 init_waitqueue_head(&sig
->wait_chldexit
);
1569 sig
->curr_target
= tsk
;
1570 init_sigpending(&sig
->shared_pending
);
1571 INIT_HLIST_HEAD(&sig
->multiprocess
);
1572 seqlock_init(&sig
->stats_lock
);
1573 prev_cputime_init(&sig
->prev_cputime
);
1575 #ifdef CONFIG_POSIX_TIMERS
1576 INIT_LIST_HEAD(&sig
->posix_timers
);
1577 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1578 sig
->real_timer
.function
= it_real_fn
;
1581 task_lock(current
->group_leader
);
1582 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1583 task_unlock(current
->group_leader
);
1585 posix_cpu_timers_init_group(sig
);
1587 tty_audit_fork(sig
);
1588 sched_autogroup_fork(sig
);
1590 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1591 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1593 mutex_init(&sig
->cred_guard_mutex
);
1594 mutex_init(&sig
->exec_update_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 static inline void init_task_pid_links(struct task_struct
*task
)
1653 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1654 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1659 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1661 if (type
== PIDTYPE_PID
)
1662 task
->thread_pid
= pid
;
1664 task
->signal
->pids
[type
] = pid
;
1667 static inline void rcu_copy_process(struct task_struct
*p
)
1669 #ifdef CONFIG_PREEMPT_RCU
1670 p
->rcu_read_lock_nesting
= 0;
1671 p
->rcu_read_unlock_special
.s
= 0;
1672 p
->rcu_blocked_node
= NULL
;
1673 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1674 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1675 #ifdef CONFIG_TASKS_RCU
1676 p
->rcu_tasks_holdout
= false;
1677 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1678 p
->rcu_tasks_idle_cpu
= -1;
1679 #endif /* #ifdef CONFIG_TASKS_RCU */
1680 #ifdef CONFIG_TASKS_TRACE_RCU
1681 p
->trc_reader_nesting
= 0;
1682 p
->trc_reader_special
.s
= 0;
1683 INIT_LIST_HEAD(&p
->trc_holdout_list
);
1684 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1687 struct pid
*pidfd_pid(const struct file
*file
)
1689 if (file
->f_op
== &pidfd_fops
)
1690 return file
->private_data
;
1692 return ERR_PTR(-EBADF
);
1695 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1697 struct pid
*pid
= file
->private_data
;
1699 file
->private_data
= NULL
;
1704 #ifdef CONFIG_PROC_FS
1706 * pidfd_show_fdinfo - print information about a pidfd
1707 * @m: proc fdinfo file
1708 * @f: file referencing a pidfd
1711 * This function will print the pid that a given pidfd refers to in the
1712 * pid namespace of the procfs instance.
1713 * If the pid namespace of the process is not a descendant of the pid
1714 * namespace of the procfs instance 0 will be shown as its pid. This is
1715 * similar to calling getppid() on a process whose parent is outside of
1716 * its pid namespace.
1719 * If pid namespaces are supported then this function will also print
1720 * the pid of a given pidfd refers to for all descendant pid namespaces
1721 * starting from the current pid namespace of the instance, i.e. the
1722 * Pid field and the first entry in the NSpid field will be identical.
1723 * If the pid namespace of the process is not a descendant of the pid
1724 * namespace of the procfs instance 0 will be shown as its first NSpid
1725 * entry and no others will be shown.
1726 * Note that this differs from the Pid and NSpid fields in
1727 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1728 * the pid namespace of the procfs instance. The difference becomes
1729 * obvious when sending around a pidfd between pid namespaces from a
1730 * different branch of the tree, i.e. where no ancestoral relation is
1731 * present between the pid namespaces:
1732 * - create two new pid namespaces ns1 and ns2 in the initial pid
1733 * namespace (also take care to create new mount namespaces in the
1734 * new pid namespace and mount procfs)
1735 * - create a process with a pidfd in ns1
1736 * - send pidfd from ns1 to ns2
1737 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1738 * have exactly one entry, which is 0
1740 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1742 struct pid
*pid
= f
->private_data
;
1743 struct pid_namespace
*ns
;
1746 if (likely(pid_has_task(pid
, PIDTYPE_PID
))) {
1747 ns
= proc_pid_ns(file_inode(m
->file
)->i_sb
);
1748 nr
= pid_nr_ns(pid
, ns
);
1751 seq_put_decimal_ll(m
, "Pid:\t", nr
);
1753 #ifdef CONFIG_PID_NS
1754 seq_put_decimal_ll(m
, "\nNSpid:\t", nr
);
1758 /* If nr is non-zero it means that 'pid' is valid and that
1759 * ns, i.e. the pid namespace associated with the procfs
1760 * instance, is in the pid namespace hierarchy of pid.
1761 * Start at one below the already printed level.
1763 for (i
= ns
->level
+ 1; i
<= pid
->level
; i
++)
1764 seq_put_decimal_ll(m
, "\t", pid
->numbers
[i
].nr
);
1772 * Poll support for process exit notification.
1774 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1776 struct pid
*pid
= file
->private_data
;
1777 __poll_t poll_flags
= 0;
1779 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1782 * Inform pollers only when the whole thread group exits.
1783 * If the thread group leader exits before all other threads in the
1784 * group, then poll(2) should block, similar to the wait(2) family.
1786 if (thread_group_exited(pid
))
1787 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1792 const struct file_operations pidfd_fops
= {
1793 .release
= pidfd_release
,
1795 #ifdef CONFIG_PROC_FS
1796 .show_fdinfo
= pidfd_show_fdinfo
,
1800 static void __delayed_free_task(struct rcu_head
*rhp
)
1802 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1807 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1809 if (IS_ENABLED(CONFIG_MEMCG
))
1810 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1815 static void copy_oom_score_adj(u64 clone_flags
, struct task_struct
*tsk
)
1817 /* Skip if kernel thread */
1821 /* Skip if spawning a thread or using vfork */
1822 if ((clone_flags
& (CLONE_VM
| CLONE_THREAD
| CLONE_VFORK
)) != CLONE_VM
)
1825 /* We need to synchronize with __set_oom_adj */
1826 mutex_lock(&oom_adj_mutex
);
1827 set_bit(MMF_MULTIPROCESS
, &tsk
->mm
->flags
);
1828 /* Update the values in case they were changed after copy_signal */
1829 tsk
->signal
->oom_score_adj
= current
->signal
->oom_score_adj
;
1830 tsk
->signal
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1831 mutex_unlock(&oom_adj_mutex
);
1835 * This creates a new process as a copy of the old one,
1836 * but does not actually start it yet.
1838 * It copies the registers, and all the appropriate
1839 * parts of the process environment (as per the clone
1840 * flags). The actual kick-off is left to the caller.
1842 static __latent_entropy
struct task_struct
*copy_process(
1846 struct kernel_clone_args
*args
)
1848 int pidfd
= -1, retval
;
1849 struct task_struct
*p
;
1850 struct multiprocess_signals delayed
;
1851 struct file
*pidfile
= NULL
;
1852 u64 clone_flags
= args
->flags
;
1853 struct nsproxy
*nsp
= current
->nsproxy
;
1856 * Don't allow sharing the root directory with processes in a different
1859 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1860 return ERR_PTR(-EINVAL
);
1862 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1863 return ERR_PTR(-EINVAL
);
1866 * Thread groups must share signals as well, and detached threads
1867 * can only be started up within the thread group.
1869 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1870 return ERR_PTR(-EINVAL
);
1873 * Shared signal handlers imply shared VM. By way of the above,
1874 * thread groups also imply shared VM. Blocking this case allows
1875 * for various simplifications in other code.
1877 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1878 return ERR_PTR(-EINVAL
);
1881 * Siblings of global init remain as zombies on exit since they are
1882 * not reaped by their parent (swapper). To solve this and to avoid
1883 * multi-rooted process trees, prevent global and container-inits
1884 * from creating siblings.
1886 if ((clone_flags
& CLONE_PARENT
) &&
1887 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1888 return ERR_PTR(-EINVAL
);
1891 * If the new process will be in a different pid or user namespace
1892 * do not allow it to share a thread group with the forking task.
1894 if (clone_flags
& CLONE_THREAD
) {
1895 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1896 (task_active_pid_ns(current
) != nsp
->pid_ns_for_children
))
1897 return ERR_PTR(-EINVAL
);
1901 * If the new process will be in a different time namespace
1902 * do not allow it to share VM or a thread group with the forking task.
1904 if (clone_flags
& (CLONE_THREAD
| CLONE_VM
)) {
1905 if (nsp
->time_ns
!= nsp
->time_ns_for_children
)
1906 return ERR_PTR(-EINVAL
);
1909 if (clone_flags
& CLONE_PIDFD
) {
1911 * - CLONE_DETACHED is blocked so that we can potentially
1912 * reuse it later for CLONE_PIDFD.
1913 * - CLONE_THREAD is blocked until someone really needs it.
1915 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1916 return ERR_PTR(-EINVAL
);
1920 * Force any signals received before this point to be delivered
1921 * before the fork happens. Collect up signals sent to multiple
1922 * processes that happen during the fork and delay them so that
1923 * they appear to happen after the fork.
1925 sigemptyset(&delayed
.signal
);
1926 INIT_HLIST_NODE(&delayed
.node
);
1928 spin_lock_irq(¤t
->sighand
->siglock
);
1929 if (!(clone_flags
& CLONE_THREAD
))
1930 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1931 recalc_sigpending();
1932 spin_unlock_irq(¤t
->sighand
->siglock
);
1933 retval
= -ERESTARTNOINTR
;
1934 if (signal_pending(current
))
1938 p
= dup_task_struct(current
, node
);
1943 * This _must_ happen before we call free_task(), i.e. before we jump
1944 * to any of the bad_fork_* labels. This is to avoid freeing
1945 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1946 * kernel threads (PF_KTHREAD).
1948 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1950 * Clear TID on mm_release()?
1952 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1954 ftrace_graph_init_task(p
);
1956 rt_mutex_init_task(p
);
1958 lockdep_assert_irqs_enabled();
1959 #ifdef CONFIG_PROVE_LOCKING
1960 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1963 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1964 task_rlimit(p
, RLIMIT_NPROC
)) {
1965 if (p
->real_cred
->user
!= INIT_USER
&&
1966 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1969 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1971 retval
= copy_creds(p
, clone_flags
);
1976 * If multiple threads are within copy_process(), then this check
1977 * triggers too late. This doesn't hurt, the check is only there
1978 * to stop root fork bombs.
1981 if (data_race(nr_threads
>= max_threads
))
1982 goto bad_fork_cleanup_count
;
1984 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1985 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1986 p
->flags
|= PF_FORKNOEXEC
;
1987 INIT_LIST_HEAD(&p
->children
);
1988 INIT_LIST_HEAD(&p
->sibling
);
1989 rcu_copy_process(p
);
1990 p
->vfork_done
= NULL
;
1991 spin_lock_init(&p
->alloc_lock
);
1993 init_sigpending(&p
->pending
);
1995 p
->utime
= p
->stime
= p
->gtime
= 0;
1996 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1997 p
->utimescaled
= p
->stimescaled
= 0;
1999 prev_cputime_init(&p
->prev_cputime
);
2001 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2002 seqcount_init(&p
->vtime
.seqcount
);
2003 p
->vtime
.starttime
= 0;
2004 p
->vtime
.state
= VTIME_INACTIVE
;
2007 #ifdef CONFIG_IO_URING
2011 #if defined(SPLIT_RSS_COUNTING)
2012 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
2015 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
2021 task_io_accounting_init(&p
->ioac
);
2022 acct_clear_integrals(p
);
2024 posix_cputimers_init(&p
->posix_cputimers
);
2026 p
->io_context
= NULL
;
2027 audit_set_context(p
, NULL
);
2030 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2031 if (IS_ERR(p
->mempolicy
)) {
2032 retval
= PTR_ERR(p
->mempolicy
);
2033 p
->mempolicy
= NULL
;
2034 goto bad_fork_cleanup_threadgroup_lock
;
2037 #ifdef CONFIG_CPUSETS
2038 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2039 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2040 seqcount_spinlock_init(&p
->mems_allowed_seq
, &p
->alloc_lock
);
2042 #ifdef CONFIG_TRACE_IRQFLAGS
2043 memset(&p
->irqtrace
, 0, sizeof(p
->irqtrace
));
2044 p
->irqtrace
.hardirq_disable_ip
= _THIS_IP_
;
2045 p
->irqtrace
.softirq_enable_ip
= _THIS_IP_
;
2046 p
->softirqs_enabled
= 1;
2047 p
->softirq_context
= 0;
2050 p
->pagefault_disabled
= 0;
2052 #ifdef CONFIG_LOCKDEP
2053 lockdep_init_task(p
);
2056 #ifdef CONFIG_DEBUG_MUTEXES
2057 p
->blocked_on
= NULL
; /* not blocked yet */
2059 #ifdef CONFIG_BCACHE
2060 p
->sequential_io
= 0;
2061 p
->sequential_io_avg
= 0;
2064 /* Perform scheduler related setup. Assign this task to a CPU. */
2065 retval
= sched_fork(clone_flags
, p
);
2067 goto bad_fork_cleanup_policy
;
2069 retval
= perf_event_init_task(p
);
2071 goto bad_fork_cleanup_policy
;
2072 retval
= audit_alloc(p
);
2074 goto bad_fork_cleanup_perf
;
2075 /* copy all the process information */
2077 retval
= security_task_alloc(p
, clone_flags
);
2079 goto bad_fork_cleanup_audit
;
2080 retval
= copy_semundo(clone_flags
, p
);
2082 goto bad_fork_cleanup_security
;
2083 retval
= copy_files(clone_flags
, p
);
2085 goto bad_fork_cleanup_semundo
;
2086 retval
= copy_fs(clone_flags
, p
);
2088 goto bad_fork_cleanup_files
;
2089 retval
= copy_sighand(clone_flags
, p
);
2091 goto bad_fork_cleanup_fs
;
2092 retval
= copy_signal(clone_flags
, p
);
2094 goto bad_fork_cleanup_sighand
;
2095 retval
= copy_mm(clone_flags
, p
);
2097 goto bad_fork_cleanup_signal
;
2098 retval
= copy_namespaces(clone_flags
, p
);
2100 goto bad_fork_cleanup_mm
;
2101 retval
= copy_io(clone_flags
, p
);
2103 goto bad_fork_cleanup_namespaces
;
2104 retval
= copy_thread(clone_flags
, args
->stack
, args
->stack_size
, p
, args
->tls
);
2106 goto bad_fork_cleanup_io
;
2108 stackleak_task_init(p
);
2110 if (pid
!= &init_struct_pid
) {
2111 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2112 args
->set_tid_size
);
2114 retval
= PTR_ERR(pid
);
2115 goto bad_fork_cleanup_thread
;
2120 * This has to happen after we've potentially unshared the file
2121 * descriptor table (so that the pidfd doesn't leak into the child
2122 * if the fd table isn't shared).
2124 if (clone_flags
& CLONE_PIDFD
) {
2125 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2127 goto bad_fork_free_pid
;
2131 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2132 O_RDWR
| O_CLOEXEC
);
2133 if (IS_ERR(pidfile
)) {
2134 put_unused_fd(pidfd
);
2135 retval
= PTR_ERR(pidfile
);
2136 goto bad_fork_free_pid
;
2138 get_pid(pid
); /* held by pidfile now */
2140 retval
= put_user(pidfd
, args
->pidfd
);
2142 goto bad_fork_put_pidfd
;
2151 * sigaltstack should be cleared when sharing the same VM
2153 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2157 * Syscall tracing and stepping should be turned off in the
2158 * child regardless of CLONE_PTRACE.
2160 user_disable_single_step(p
);
2161 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2162 #ifdef TIF_SYSCALL_EMU
2163 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2165 clear_tsk_latency_tracing(p
);
2167 /* ok, now we should be set up.. */
2168 p
->pid
= pid_nr(pid
);
2169 if (clone_flags
& CLONE_THREAD
) {
2170 p
->group_leader
= current
->group_leader
;
2171 p
->tgid
= current
->tgid
;
2173 p
->group_leader
= p
;
2178 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2179 p
->dirty_paused_when
= 0;
2181 p
->pdeath_signal
= 0;
2182 INIT_LIST_HEAD(&p
->thread_group
);
2183 p
->task_works
= NULL
;
2186 * Ensure that the cgroup subsystem policies allow the new process to be
2187 * forked. It should be noted that the new process's css_set can be changed
2188 * between here and cgroup_post_fork() if an organisation operation is in
2191 retval
= cgroup_can_fork(p
, args
);
2193 goto bad_fork_put_pidfd
;
2196 * From this point on we must avoid any synchronous user-space
2197 * communication until we take the tasklist-lock. In particular, we do
2198 * not want user-space to be able to predict the process start-time by
2199 * stalling fork(2) after we recorded the start_time but before it is
2200 * visible to the system.
2203 p
->start_time
= ktime_get_ns();
2204 p
->start_boottime
= ktime_get_boottime_ns();
2207 * Make it visible to the rest of the system, but dont wake it up yet.
2208 * Need tasklist lock for parent etc handling!
2210 write_lock_irq(&tasklist_lock
);
2212 /* CLONE_PARENT re-uses the old parent */
2213 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2214 p
->real_parent
= current
->real_parent
;
2215 p
->parent_exec_id
= current
->parent_exec_id
;
2216 if (clone_flags
& CLONE_THREAD
)
2217 p
->exit_signal
= -1;
2219 p
->exit_signal
= current
->group_leader
->exit_signal
;
2221 p
->real_parent
= current
;
2222 p
->parent_exec_id
= current
->self_exec_id
;
2223 p
->exit_signal
= args
->exit_signal
;
2226 klp_copy_process(p
);
2228 spin_lock(¤t
->sighand
->siglock
);
2231 * Copy seccomp details explicitly here, in case they were changed
2232 * before holding sighand lock.
2236 rseq_fork(p
, clone_flags
);
2238 /* Don't start children in a dying pid namespace */
2239 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2241 goto bad_fork_cancel_cgroup
;
2244 /* Let kill terminate clone/fork in the middle */
2245 if (fatal_signal_pending(current
)) {
2247 goto bad_fork_cancel_cgroup
;
2250 /* past the last point of failure */
2252 fd_install(pidfd
, pidfile
);
2254 init_task_pid_links(p
);
2255 if (likely(p
->pid
)) {
2256 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2258 init_task_pid(p
, PIDTYPE_PID
, pid
);
2259 if (thread_group_leader(p
)) {
2260 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2261 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2262 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2264 if (is_child_reaper(pid
)) {
2265 ns_of_pid(pid
)->child_reaper
= p
;
2266 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2268 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2269 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2271 * Inherit has_child_subreaper flag under the same
2272 * tasklist_lock with adding child to the process tree
2273 * for propagate_has_child_subreaper optimization.
2275 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2276 p
->real_parent
->signal
->is_child_subreaper
;
2277 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2278 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2279 attach_pid(p
, PIDTYPE_TGID
);
2280 attach_pid(p
, PIDTYPE_PGID
);
2281 attach_pid(p
, PIDTYPE_SID
);
2282 __this_cpu_inc(process_counts
);
2284 current
->signal
->nr_threads
++;
2285 atomic_inc(¤t
->signal
->live
);
2286 refcount_inc(¤t
->signal
->sigcnt
);
2287 task_join_group_stop(p
);
2288 list_add_tail_rcu(&p
->thread_group
,
2289 &p
->group_leader
->thread_group
);
2290 list_add_tail_rcu(&p
->thread_node
,
2291 &p
->signal
->thread_head
);
2293 attach_pid(p
, PIDTYPE_PID
);
2297 hlist_del_init(&delayed
.node
);
2298 spin_unlock(¤t
->sighand
->siglock
);
2299 syscall_tracepoint_update(p
);
2300 write_unlock_irq(&tasklist_lock
);
2302 proc_fork_connector(p
);
2304 cgroup_post_fork(p
, args
);
2307 trace_task_newtask(p
, clone_flags
);
2308 uprobe_copy_process(p
, clone_flags
);
2310 copy_oom_score_adj(clone_flags
, p
);
2314 bad_fork_cancel_cgroup
:
2315 spin_unlock(¤t
->sighand
->siglock
);
2316 write_unlock_irq(&tasklist_lock
);
2317 cgroup_cancel_fork(p
, args
);
2319 if (clone_flags
& CLONE_PIDFD
) {
2321 put_unused_fd(pidfd
);
2324 if (pid
!= &init_struct_pid
)
2326 bad_fork_cleanup_thread
:
2328 bad_fork_cleanup_io
:
2331 bad_fork_cleanup_namespaces
:
2332 exit_task_namespaces(p
);
2333 bad_fork_cleanup_mm
:
2335 mm_clear_owner(p
->mm
, p
);
2338 bad_fork_cleanup_signal
:
2339 if (!(clone_flags
& CLONE_THREAD
))
2340 free_signal_struct(p
->signal
);
2341 bad_fork_cleanup_sighand
:
2342 __cleanup_sighand(p
->sighand
);
2343 bad_fork_cleanup_fs
:
2344 exit_fs(p
); /* blocking */
2345 bad_fork_cleanup_files
:
2346 exit_files(p
); /* blocking */
2347 bad_fork_cleanup_semundo
:
2349 bad_fork_cleanup_security
:
2350 security_task_free(p
);
2351 bad_fork_cleanup_audit
:
2353 bad_fork_cleanup_perf
:
2354 perf_event_free_task(p
);
2355 bad_fork_cleanup_policy
:
2356 lockdep_free_task(p
);
2358 mpol_put(p
->mempolicy
);
2359 bad_fork_cleanup_threadgroup_lock
:
2361 delayacct_tsk_free(p
);
2362 bad_fork_cleanup_count
:
2363 atomic_dec(&p
->cred
->user
->processes
);
2366 p
->state
= TASK_DEAD
;
2368 delayed_free_task(p
);
2370 spin_lock_irq(¤t
->sighand
->siglock
);
2371 hlist_del_init(&delayed
.node
);
2372 spin_unlock_irq(¤t
->sighand
->siglock
);
2373 return ERR_PTR(retval
);
2376 static inline void init_idle_pids(struct task_struct
*idle
)
2380 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2381 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2382 init_task_pid(idle
, type
, &init_struct_pid
);
2386 struct task_struct
*fork_idle(int cpu
)
2388 struct task_struct
*task
;
2389 struct kernel_clone_args args
= {
2393 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2394 if (!IS_ERR(task
)) {
2395 init_idle_pids(task
);
2396 init_idle(task
, cpu
);
2402 struct mm_struct
*copy_init_mm(void)
2404 return dup_mm(NULL
, &init_mm
);
2408 * Ok, this is the main fork-routine.
2410 * It copies the process, and if successful kick-starts
2411 * it and waits for it to finish using the VM if required.
2413 * args->exit_signal is expected to be checked for sanity by the caller.
2415 pid_t
kernel_clone(struct kernel_clone_args
*args
)
2417 u64 clone_flags
= args
->flags
;
2418 struct completion vfork
;
2420 struct task_struct
*p
;
2425 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2426 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2427 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2428 * field in struct clone_args and it still doesn't make sense to have
2429 * them both point at the same memory location. Performing this check
2430 * here has the advantage that we don't need to have a separate helper
2431 * to check for legacy clone().
2433 if ((args
->flags
& CLONE_PIDFD
) &&
2434 (args
->flags
& CLONE_PARENT_SETTID
) &&
2435 (args
->pidfd
== args
->parent_tid
))
2439 * Determine whether and which event to report to ptracer. When
2440 * called from kernel_thread or CLONE_UNTRACED is explicitly
2441 * requested, no event is reported; otherwise, report if the event
2442 * for the type of forking is enabled.
2444 if (!(clone_flags
& CLONE_UNTRACED
)) {
2445 if (clone_flags
& CLONE_VFORK
)
2446 trace
= PTRACE_EVENT_VFORK
;
2447 else if (args
->exit_signal
!= SIGCHLD
)
2448 trace
= PTRACE_EVENT_CLONE
;
2450 trace
= PTRACE_EVENT_FORK
;
2452 if (likely(!ptrace_event_enabled(current
, trace
)))
2456 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2457 add_latent_entropy();
2463 * Do this prior waking up the new thread - the thread pointer
2464 * might get invalid after that point, if the thread exits quickly.
2466 trace_sched_process_fork(current
, p
);
2468 pid
= get_task_pid(p
, PIDTYPE_PID
);
2471 if (clone_flags
& CLONE_PARENT_SETTID
)
2472 put_user(nr
, args
->parent_tid
);
2474 if (clone_flags
& CLONE_VFORK
) {
2475 p
->vfork_done
= &vfork
;
2476 init_completion(&vfork
);
2480 wake_up_new_task(p
);
2482 /* forking complete and child started to run, tell ptracer */
2483 if (unlikely(trace
))
2484 ptrace_event_pid(trace
, pid
);
2486 if (clone_flags
& CLONE_VFORK
) {
2487 if (!wait_for_vfork_done(p
, &vfork
))
2488 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2496 * Create a kernel thread.
2498 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2500 struct kernel_clone_args args
= {
2501 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2502 CLONE_UNTRACED
) & ~CSIGNAL
),
2503 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2504 .stack
= (unsigned long)fn
,
2505 .stack_size
= (unsigned long)arg
,
2508 return kernel_clone(&args
);
2511 #ifdef __ARCH_WANT_SYS_FORK
2512 SYSCALL_DEFINE0(fork
)
2515 struct kernel_clone_args args
= {
2516 .exit_signal
= SIGCHLD
,
2519 return kernel_clone(&args
);
2521 /* can not support in nommu mode */
2527 #ifdef __ARCH_WANT_SYS_VFORK
2528 SYSCALL_DEFINE0(vfork
)
2530 struct kernel_clone_args args
= {
2531 .flags
= CLONE_VFORK
| CLONE_VM
,
2532 .exit_signal
= SIGCHLD
,
2535 return kernel_clone(&args
);
2539 #ifdef __ARCH_WANT_SYS_CLONE
2540 #ifdef CONFIG_CLONE_BACKWARDS
2541 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2542 int __user
*, parent_tidptr
,
2544 int __user
*, child_tidptr
)
2545 #elif defined(CONFIG_CLONE_BACKWARDS2)
2546 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2547 int __user
*, parent_tidptr
,
2548 int __user
*, child_tidptr
,
2550 #elif defined(CONFIG_CLONE_BACKWARDS3)
2551 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2553 int __user
*, parent_tidptr
,
2554 int __user
*, child_tidptr
,
2557 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2558 int __user
*, parent_tidptr
,
2559 int __user
*, child_tidptr
,
2563 struct kernel_clone_args args
= {
2564 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2565 .pidfd
= parent_tidptr
,
2566 .child_tid
= child_tidptr
,
2567 .parent_tid
= parent_tidptr
,
2568 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2573 return kernel_clone(&args
);
2577 #ifdef __ARCH_WANT_SYS_CLONE3
2579 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2580 struct clone_args __user
*uargs
,
2584 struct clone_args args
;
2585 pid_t
*kset_tid
= kargs
->set_tid
;
2587 BUILD_BUG_ON(offsetofend(struct clone_args
, tls
) !=
2588 CLONE_ARGS_SIZE_VER0
);
2589 BUILD_BUG_ON(offsetofend(struct clone_args
, set_tid_size
) !=
2590 CLONE_ARGS_SIZE_VER1
);
2591 BUILD_BUG_ON(offsetofend(struct clone_args
, cgroup
) !=
2592 CLONE_ARGS_SIZE_VER2
);
2593 BUILD_BUG_ON(sizeof(struct clone_args
) != CLONE_ARGS_SIZE_VER2
);
2595 if (unlikely(usize
> PAGE_SIZE
))
2597 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2600 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2604 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2607 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2610 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2614 * Verify that higher 32bits of exit_signal are unset and that
2615 * it is a valid signal
2617 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2618 !valid_signal(args
.exit_signal
)))
2621 if ((args
.flags
& CLONE_INTO_CGROUP
) &&
2622 (args
.cgroup
> INT_MAX
|| usize
< CLONE_ARGS_SIZE_VER2
))
2625 *kargs
= (struct kernel_clone_args
){
2626 .flags
= args
.flags
,
2627 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2628 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2629 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2630 .exit_signal
= args
.exit_signal
,
2631 .stack
= args
.stack
,
2632 .stack_size
= args
.stack_size
,
2634 .set_tid_size
= args
.set_tid_size
,
2635 .cgroup
= args
.cgroup
,
2639 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
2640 (kargs
->set_tid_size
* sizeof(pid_t
))))
2643 kargs
->set_tid
= kset_tid
;
2649 * clone3_stack_valid - check and prepare stack
2650 * @kargs: kernel clone args
2652 * Verify that the stack arguments userspace gave us are sane.
2653 * In addition, set the stack direction for userspace since it's easy for us to
2656 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2658 if (kargs
->stack
== 0) {
2659 if (kargs
->stack_size
> 0)
2662 if (kargs
->stack_size
== 0)
2665 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2668 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2669 kargs
->stack
+= kargs
->stack_size
;
2676 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2678 /* Verify that no unknown flags are passed along. */
2680 ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
| CLONE_INTO_CGROUP
))
2684 * - make the CLONE_DETACHED bit reuseable for clone3
2685 * - make the CSIGNAL bits reuseable for clone3
2687 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2690 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
2691 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
2694 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2698 if (!clone3_stack_valid(kargs
))
2705 * clone3 - create a new process with specific properties
2706 * @uargs: argument structure
2707 * @size: size of @uargs
2709 * clone3() is the extensible successor to clone()/clone2().
2710 * It takes a struct as argument that is versioned by its size.
2712 * Return: On success, a positive PID for the child process.
2713 * On error, a negative errno number.
2715 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2719 struct kernel_clone_args kargs
;
2720 pid_t set_tid
[MAX_PID_NS_LEVEL
];
2722 kargs
.set_tid
= set_tid
;
2724 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2728 if (!clone3_args_valid(&kargs
))
2731 return kernel_clone(&kargs
);
2735 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2737 struct task_struct
*leader
, *parent
, *child
;
2740 read_lock(&tasklist_lock
);
2741 leader
= top
= top
->group_leader
;
2743 for_each_thread(leader
, parent
) {
2744 list_for_each_entry(child
, &parent
->children
, sibling
) {
2745 res
= visitor(child
, data
);
2757 if (leader
!= top
) {
2759 parent
= child
->real_parent
;
2760 leader
= parent
->group_leader
;
2764 read_unlock(&tasklist_lock
);
2767 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2768 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2771 static void sighand_ctor(void *data
)
2773 struct sighand_struct
*sighand
= data
;
2775 spin_lock_init(&sighand
->siglock
);
2776 init_waitqueue_head(&sighand
->signalfd_wqh
);
2779 void __init
proc_caches_init(void)
2781 unsigned int mm_size
;
2783 sighand_cachep
= kmem_cache_create("sighand_cache",
2784 sizeof(struct sighand_struct
), 0,
2785 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2786 SLAB_ACCOUNT
, sighand_ctor
);
2787 signal_cachep
= kmem_cache_create("signal_cache",
2788 sizeof(struct signal_struct
), 0,
2789 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2791 files_cachep
= kmem_cache_create("files_cache",
2792 sizeof(struct files_struct
), 0,
2793 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2795 fs_cachep
= kmem_cache_create("fs_cache",
2796 sizeof(struct fs_struct
), 0,
2797 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2801 * The mm_cpumask is located at the end of mm_struct, and is
2802 * dynamically sized based on the maximum CPU number this system
2803 * can have, taking hotplug into account (nr_cpu_ids).
2805 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2807 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2808 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2809 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2810 offsetof(struct mm_struct
, saved_auxv
),
2811 sizeof_field(struct mm_struct
, saved_auxv
),
2813 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2815 nsproxy_cache_init();
2819 * Check constraints on flags passed to the unshare system call.
2821 static int check_unshare_flags(unsigned long unshare_flags
)
2823 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2824 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2825 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2826 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
|
2830 * Not implemented, but pretend it works if there is nothing
2831 * to unshare. Note that unsharing the address space or the
2832 * signal handlers also need to unshare the signal queues (aka
2835 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2836 if (!thread_group_empty(current
))
2839 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2840 if (refcount_read(¤t
->sighand
->count
) > 1)
2843 if (unshare_flags
& CLONE_VM
) {
2844 if (!current_is_single_threaded())
2852 * Unshare the filesystem structure if it is being shared
2854 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2856 struct fs_struct
*fs
= current
->fs
;
2858 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2861 /* don't need lock here; in the worst case we'll do useless copy */
2865 *new_fsp
= copy_fs_struct(fs
);
2873 * Unshare file descriptor table if it is being shared
2875 int unshare_fd(unsigned long unshare_flags
, unsigned int max_fds
,
2876 struct files_struct
**new_fdp
)
2878 struct files_struct
*fd
= current
->files
;
2881 if ((unshare_flags
& CLONE_FILES
) &&
2882 (fd
&& atomic_read(&fd
->count
) > 1)) {
2883 *new_fdp
= dup_fd(fd
, max_fds
, &error
);
2892 * unshare allows a process to 'unshare' part of the process
2893 * context which was originally shared using clone. copy_*
2894 * functions used by kernel_clone() cannot be used here directly
2895 * because they modify an inactive task_struct that is being
2896 * constructed. Here we are modifying the current, active,
2899 int ksys_unshare(unsigned long unshare_flags
)
2901 struct fs_struct
*fs
, *new_fs
= NULL
;
2902 struct files_struct
*fd
, *new_fd
= NULL
;
2903 struct cred
*new_cred
= NULL
;
2904 struct nsproxy
*new_nsproxy
= NULL
;
2909 * If unsharing a user namespace must also unshare the thread group
2910 * and unshare the filesystem root and working directories.
2912 if (unshare_flags
& CLONE_NEWUSER
)
2913 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2915 * If unsharing vm, must also unshare signal handlers.
2917 if (unshare_flags
& CLONE_VM
)
2918 unshare_flags
|= CLONE_SIGHAND
;
2920 * If unsharing a signal handlers, must also unshare the signal queues.
2922 if (unshare_flags
& CLONE_SIGHAND
)
2923 unshare_flags
|= CLONE_THREAD
;
2925 * If unsharing namespace, must also unshare filesystem information.
2927 if (unshare_flags
& CLONE_NEWNS
)
2928 unshare_flags
|= CLONE_FS
;
2930 err
= check_unshare_flags(unshare_flags
);
2932 goto bad_unshare_out
;
2934 * CLONE_NEWIPC must also detach from the undolist: after switching
2935 * to a new ipc namespace, the semaphore arrays from the old
2936 * namespace are unreachable.
2938 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2940 err
= unshare_fs(unshare_flags
, &new_fs
);
2942 goto bad_unshare_out
;
2943 err
= unshare_fd(unshare_flags
, NR_OPEN_MAX
, &new_fd
);
2945 goto bad_unshare_cleanup_fs
;
2946 err
= unshare_userns(unshare_flags
, &new_cred
);
2948 goto bad_unshare_cleanup_fd
;
2949 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2952 goto bad_unshare_cleanup_cred
;
2954 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2957 * CLONE_SYSVSEM is equivalent to sys_exit().
2961 if (unshare_flags
& CLONE_NEWIPC
) {
2962 /* Orphan segments in old ns (see sem above). */
2964 shm_init_task(current
);
2968 switch_task_namespaces(current
, new_nsproxy
);
2974 spin_lock(&fs
->lock
);
2975 current
->fs
= new_fs
;
2980 spin_unlock(&fs
->lock
);
2984 fd
= current
->files
;
2985 current
->files
= new_fd
;
2989 task_unlock(current
);
2992 /* Install the new user namespace */
2993 commit_creds(new_cred
);
2998 perf_event_namespaces(current
);
3000 bad_unshare_cleanup_cred
:
3003 bad_unshare_cleanup_fd
:
3005 put_files_struct(new_fd
);
3007 bad_unshare_cleanup_fs
:
3009 free_fs_struct(new_fs
);
3015 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3017 return ksys_unshare(unshare_flags
);
3021 * Helper to unshare the files of the current task.
3022 * We don't want to expose copy_files internals to
3023 * the exec layer of the kernel.
3026 int unshare_files(struct files_struct
**displaced
)
3028 struct task_struct
*task
= current
;
3029 struct files_struct
*copy
= NULL
;
3032 error
= unshare_fd(CLONE_FILES
, NR_OPEN_MAX
, ©
);
3033 if (error
|| !copy
) {
3037 *displaced
= task
->files
;
3044 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3045 void *buffer
, size_t *lenp
, loff_t
*ppos
)
3049 int threads
= max_threads
;
3051 int max
= MAX_THREADS
;
3058 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3062 max_threads
= threads
;