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>
110 #ifdef CONFIG_USER_NS
111 extern int unprivileged_userns_clone
;
113 #define unprivileged_userns_clone 0
117 * Minimum number of threads to boot the kernel
119 #define MIN_THREADS 20
122 * Maximum number of threads
124 #define MAX_THREADS FUTEX_TID_MASK
127 * Protected counters by write_lock_irq(&tasklist_lock)
129 unsigned long total_forks
; /* Handle normal Linux uptimes. */
130 int nr_threads
; /* The idle threads do not count.. */
132 static int max_threads
; /* tunable limit on nr_threads */
134 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
136 static const char * const resident_page_types
[] = {
137 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
138 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
139 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
140 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
143 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
145 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
147 #ifdef CONFIG_PROVE_RCU
148 int lockdep_tasklist_lock_is_held(void)
150 return lockdep_is_held(&tasklist_lock
);
152 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
153 #endif /* #ifdef CONFIG_PROVE_RCU */
155 int nr_processes(void)
160 for_each_possible_cpu(cpu
)
161 total
+= per_cpu(process_counts
, cpu
);
166 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
170 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
171 static struct kmem_cache
*task_struct_cachep
;
173 static inline struct task_struct
*alloc_task_struct_node(int node
)
175 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
178 static inline void free_task_struct(struct task_struct
*tsk
)
180 kmem_cache_free(task_struct_cachep
, tsk
);
184 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
187 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
188 * kmemcache based allocator.
190 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
192 #ifdef CONFIG_VMAP_STACK
194 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
195 * flush. Try to minimize the number of calls by caching stacks.
197 #define NR_CACHED_STACKS 2
198 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
200 static int free_vm_stack_cache(unsigned int cpu
)
202 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
205 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
206 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
211 vfree(vm_stack
->addr
);
212 cached_vm_stacks
[i
] = NULL
;
219 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
221 #ifdef CONFIG_VMAP_STACK
225 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
228 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
233 /* Mark stack accessible for KASAN. */
234 kasan_unpoison_range(s
->addr
, THREAD_SIZE
);
236 /* Clear stale pointers from reused stack. */
237 memset(s
->addr
, 0, THREAD_SIZE
);
239 tsk
->stack_vm_area
= s
;
240 tsk
->stack
= s
->addr
;
245 * Allocated stacks are cached and later reused by new threads,
246 * so memcg accounting is performed manually on assigning/releasing
247 * stacks to tasks. Drop __GFP_ACCOUNT.
249 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
250 VMALLOC_START
, VMALLOC_END
,
251 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
253 0, node
, __builtin_return_address(0));
256 * We can't call find_vm_area() in interrupt context, and
257 * free_thread_stack() can be called in interrupt context,
258 * so cache the vm_struct.
261 tsk
->stack_vm_area
= find_vm_area(stack
);
266 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
270 tsk
->stack
= kasan_reset_tag(page_address(page
));
277 static inline void free_thread_stack(struct task_struct
*tsk
)
279 #ifdef CONFIG_VMAP_STACK
280 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
285 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
286 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
288 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
289 if (this_cpu_cmpxchg(cached_stacks
[i
],
290 NULL
, tsk
->stack_vm_area
) != NULL
)
296 vfree_atomic(tsk
->stack
);
301 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
304 static struct kmem_cache
*thread_stack_cache
;
306 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
309 unsigned long *stack
;
310 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
311 stack
= kasan_reset_tag(stack
);
316 static void free_thread_stack(struct task_struct
*tsk
)
318 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
321 void thread_stack_cache_init(void)
323 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
324 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
326 BUG_ON(thread_stack_cache
== NULL
);
331 /* SLAB cache for signal_struct structures (tsk->signal) */
332 static struct kmem_cache
*signal_cachep
;
334 /* SLAB cache for sighand_struct structures (tsk->sighand) */
335 struct kmem_cache
*sighand_cachep
;
337 /* SLAB cache for files_struct structures (tsk->files) */
338 struct kmem_cache
*files_cachep
;
340 /* SLAB cache for fs_struct structures (tsk->fs) */
341 struct kmem_cache
*fs_cachep
;
343 /* SLAB cache for vm_area_struct structures */
344 static struct kmem_cache
*vm_area_cachep
;
346 /* SLAB cache for mm_struct structures (tsk->mm) */
347 static struct kmem_cache
*mm_cachep
;
349 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
351 struct vm_area_struct
*vma
;
353 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
359 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
361 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
364 ASSERT_EXCLUSIVE_WRITER(orig
->vm_flags
);
365 ASSERT_EXCLUSIVE_WRITER(orig
->vm_file
);
367 * orig->shared.rb may be modified concurrently, but the clone
368 * will be reinitialized.
370 *new = data_race(*orig
);
371 INIT_LIST_HEAD(&new->anon_vma_chain
);
372 new->vm_next
= new->vm_prev
= NULL
;
377 void vm_area_free(struct vm_area_struct
*vma
)
379 kmem_cache_free(vm_area_cachep
, vma
);
382 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
384 void *stack
= task_stack_page(tsk
);
385 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
388 /* All stack pages are in the same node. */
390 mod_lruvec_page_state(vm
->pages
[0], NR_KERNEL_STACK_KB
,
391 account
* (THREAD_SIZE
/ 1024));
393 mod_lruvec_kmem_state(stack
, NR_KERNEL_STACK_KB
,
394 account
* (THREAD_SIZE
/ 1024));
397 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
399 #ifdef CONFIG_VMAP_STACK
400 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
403 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
408 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
410 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
412 * If memcg_kmem_charge_page() fails, page's
413 * memory cgroup pointer is NULL, and
414 * memcg_kmem_uncharge_page() in free_thread_stack()
415 * will ignore this page.
417 ret
= memcg_kmem_charge_page(vm
->pages
[i
], GFP_KERNEL
,
427 static void release_task_stack(struct task_struct
*tsk
)
429 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
430 return; /* Better to leak the stack than to free prematurely */
432 account_kernel_stack(tsk
, -1);
433 free_thread_stack(tsk
);
435 #ifdef CONFIG_VMAP_STACK
436 tsk
->stack_vm_area
= NULL
;
440 #ifdef CONFIG_THREAD_INFO_IN_TASK
441 void put_task_stack(struct task_struct
*tsk
)
443 if (refcount_dec_and_test(&tsk
->stack_refcount
))
444 release_task_stack(tsk
);
448 void free_task(struct task_struct
*tsk
)
452 #ifndef CONFIG_THREAD_INFO_IN_TASK
454 * The task is finally done with both the stack and thread_info,
457 release_task_stack(tsk
);
460 * If the task had a separate stack allocation, it should be gone
463 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
465 rt_mutex_debug_task_free(tsk
);
466 ftrace_graph_exit_task(tsk
);
467 arch_release_task_struct(tsk
);
468 if (tsk
->flags
& PF_KTHREAD
)
469 free_kthread_struct(tsk
);
470 free_task_struct(tsk
);
472 EXPORT_SYMBOL(free_task
);
475 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
476 struct mm_struct
*oldmm
)
478 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
479 struct rb_node
**rb_link
, *rb_parent
;
481 unsigned long charge
;
484 uprobe_start_dup_mmap();
485 if (mmap_write_lock_killable(oldmm
)) {
487 goto fail_uprobe_end
;
489 flush_cache_dup_mm(oldmm
);
490 uprobe_dup_mmap(oldmm
, mm
);
492 * Not linked in yet - no deadlock potential:
494 mmap_write_lock_nested(mm
, SINGLE_DEPTH_NESTING
);
496 /* No ordering required: file already has been exposed. */
497 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
499 mm
->total_vm
= oldmm
->total_vm
;
500 mm
->data_vm
= oldmm
->data_vm
;
501 mm
->exec_vm
= oldmm
->exec_vm
;
502 mm
->stack_vm
= oldmm
->stack_vm
;
504 rb_link
= &mm
->mm_rb
.rb_node
;
507 retval
= ksm_fork(mm
, oldmm
);
510 retval
= khugepaged_fork(mm
, oldmm
);
515 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
518 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
519 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
524 * Don't duplicate many vmas if we've been oom-killed (for
527 if (fatal_signal_pending(current
)) {
531 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
532 unsigned long len
= vma_pages(mpnt
);
534 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
538 tmp
= vm_area_dup(mpnt
);
541 retval
= vma_dup_policy(mpnt
, tmp
);
543 goto fail_nomem_policy
;
545 retval
= dup_userfaultfd(tmp
, &uf
);
547 goto fail_nomem_anon_vma_fork
;
548 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
550 * VM_WIPEONFORK gets a clean slate in the child.
551 * Don't prepare anon_vma until fault since we don't
552 * copy page for current vma.
554 tmp
->anon_vma
= NULL
;
555 } else if (anon_vma_fork(tmp
, mpnt
))
556 goto fail_nomem_anon_vma_fork
;
557 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
560 struct inode
*inode
= file_inode(file
);
561 struct address_space
*mapping
= file
->f_mapping
;
564 if (tmp
->vm_flags
& VM_DENYWRITE
)
565 put_write_access(inode
);
566 i_mmap_lock_write(mapping
);
567 if (tmp
->vm_flags
& VM_SHARED
)
568 mapping_allow_writable(mapping
);
569 flush_dcache_mmap_lock(mapping
);
570 /* insert tmp into the share list, just after mpnt */
571 vma_interval_tree_insert_after(tmp
, mpnt
,
573 flush_dcache_mmap_unlock(mapping
);
574 i_mmap_unlock_write(mapping
);
578 * Clear hugetlb-related page reserves for children. This only
579 * affects MAP_PRIVATE mappings. Faults generated by the child
580 * are not guaranteed to succeed, even if read-only
582 if (is_vm_hugetlb_page(tmp
))
583 reset_vma_resv_huge_pages(tmp
);
586 * Link in the new vma and copy the page table entries.
589 pprev
= &tmp
->vm_next
;
593 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
594 rb_link
= &tmp
->vm_rb
.rb_right
;
595 rb_parent
= &tmp
->vm_rb
;
598 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
599 retval
= copy_page_range(tmp
, mpnt
);
601 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
602 tmp
->vm_ops
->open(tmp
);
607 /* a new mm has just been created */
608 retval
= arch_dup_mmap(oldmm
, mm
);
610 mmap_write_unlock(mm
);
612 mmap_write_unlock(oldmm
);
613 dup_userfaultfd_complete(&uf
);
615 uprobe_end_dup_mmap();
617 fail_nomem_anon_vma_fork
:
618 mpol_put(vma_policy(tmp
));
623 vm_unacct_memory(charge
);
627 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
629 mm
->pgd
= pgd_alloc(mm
);
630 if (unlikely(!mm
->pgd
))
635 static inline void mm_free_pgd(struct mm_struct
*mm
)
637 pgd_free(mm
, mm
->pgd
);
640 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
642 mmap_write_lock(oldmm
);
643 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
644 mmap_write_unlock(oldmm
);
647 #define mm_alloc_pgd(mm) (0)
648 #define mm_free_pgd(mm)
649 #endif /* CONFIG_MMU */
651 static void check_mm(struct mm_struct
*mm
)
655 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
656 "Please make sure 'struct resident_page_types[]' is updated as well");
658 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
659 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
662 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
663 mm
, resident_page_types
[i
], x
);
666 if (mm_pgtables_bytes(mm
))
667 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
668 mm_pgtables_bytes(mm
));
670 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
671 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
675 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
676 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
679 * Called when the last reference to the mm
680 * is dropped: either by a lazy thread or by
681 * mmput. Free the page directory and the mm.
683 void __mmdrop(struct mm_struct
*mm
)
685 BUG_ON(mm
== &init_mm
);
686 WARN_ON_ONCE(mm
== current
->mm
);
687 WARN_ON_ONCE(mm
== current
->active_mm
);
690 mmu_notifier_subscriptions_destroy(mm
);
692 put_user_ns(mm
->user_ns
);
695 EXPORT_SYMBOL_GPL(__mmdrop
);
697 static void mmdrop_async_fn(struct work_struct
*work
)
699 struct mm_struct
*mm
;
701 mm
= container_of(work
, struct mm_struct
, async_put_work
);
705 static void mmdrop_async(struct mm_struct
*mm
)
707 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
708 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
709 schedule_work(&mm
->async_put_work
);
713 static inline void free_signal_struct(struct signal_struct
*sig
)
715 taskstats_tgid_free(sig
);
716 sched_autogroup_exit(sig
);
718 * __mmdrop is not safe to call from softirq context on x86 due to
719 * pgd_dtor so postpone it to the async context
722 mmdrop_async(sig
->oom_mm
);
723 kmem_cache_free(signal_cachep
, sig
);
726 static inline void put_signal_struct(struct signal_struct
*sig
)
728 if (refcount_dec_and_test(&sig
->sigcnt
))
729 free_signal_struct(sig
);
732 void __put_task_struct(struct task_struct
*tsk
)
734 WARN_ON(!tsk
->exit_state
);
735 WARN_ON(refcount_read(&tsk
->usage
));
736 WARN_ON(tsk
== current
);
740 task_numa_free(tsk
, true);
741 security_task_free(tsk
);
743 delayacct_tsk_free(tsk
);
744 put_signal_struct(tsk
->signal
);
746 if (!profile_handoff_task(tsk
))
749 EXPORT_SYMBOL_GPL(__put_task_struct
);
751 void __init __weak
arch_task_cache_init(void) { }
756 static void set_max_threads(unsigned int max_threads_suggested
)
759 unsigned long nr_pages
= totalram_pages();
762 * The number of threads shall be limited such that the thread
763 * structures may only consume a small part of the available memory.
765 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
766 threads
= MAX_THREADS
;
768 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
769 (u64
) THREAD_SIZE
* 8UL);
771 if (threads
> max_threads_suggested
)
772 threads
= max_threads_suggested
;
774 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
777 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
778 /* Initialized by the architecture: */
779 int arch_task_struct_size __read_mostly
;
782 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
783 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
785 /* Fetch thread_struct whitelist for the architecture. */
786 arch_thread_struct_whitelist(offset
, size
);
789 * Handle zero-sized whitelist or empty thread_struct, otherwise
790 * adjust offset to position of thread_struct in task_struct.
792 if (unlikely(*size
== 0))
795 *offset
+= offsetof(struct task_struct
, thread
);
797 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
799 void __init
fork_init(void)
802 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
803 #ifndef ARCH_MIN_TASKALIGN
804 #define ARCH_MIN_TASKALIGN 0
806 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
807 unsigned long useroffset
, usersize
;
809 /* create a slab on which task_structs can be allocated */
810 task_struct_whitelist(&useroffset
, &usersize
);
811 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
812 arch_task_struct_size
, align
,
813 SLAB_PANIC
|SLAB_ACCOUNT
,
814 useroffset
, usersize
, NULL
);
817 /* do the arch specific task caches init */
818 arch_task_cache_init();
820 set_max_threads(MAX_THREADS
);
822 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
823 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
824 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
825 init_task
.signal
->rlim
[RLIMIT_NPROC
];
827 for (i
= 0; i
< UCOUNT_COUNTS
; i
++)
828 init_user_ns
.ucount_max
[i
] = max_threads
/2;
830 #ifdef CONFIG_VMAP_STACK
831 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
832 NULL
, free_vm_stack_cache
);
837 lockdep_init_task(&init_task
);
841 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
842 struct task_struct
*src
)
848 void set_task_stack_end_magic(struct task_struct
*tsk
)
850 unsigned long *stackend
;
852 stackend
= end_of_stack(tsk
);
853 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
856 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
858 struct task_struct
*tsk
;
859 unsigned long *stack
;
860 struct vm_struct
*stack_vm_area __maybe_unused
;
863 if (node
== NUMA_NO_NODE
)
864 node
= tsk_fork_get_node(orig
);
865 tsk
= alloc_task_struct_node(node
);
869 stack
= alloc_thread_stack_node(tsk
, node
);
873 if (memcg_charge_kernel_stack(tsk
))
876 stack_vm_area
= task_stack_vm_area(tsk
);
878 err
= arch_dup_task_struct(tsk
, orig
);
881 * arch_dup_task_struct() clobbers the stack-related fields. Make
882 * sure they're properly initialized before using any stack-related
886 #ifdef CONFIG_VMAP_STACK
887 tsk
->stack_vm_area
= stack_vm_area
;
889 #ifdef CONFIG_THREAD_INFO_IN_TASK
890 refcount_set(&tsk
->stack_refcount
, 1);
896 err
= scs_prepare(tsk
, node
);
900 #ifdef CONFIG_SECCOMP
902 * We must handle setting up seccomp filters once we're under
903 * the sighand lock in case orig has changed between now and
904 * then. Until then, filter must be NULL to avoid messing up
905 * the usage counts on the error path calling free_task.
907 tsk
->seccomp
.filter
= NULL
;
910 setup_thread_stack(tsk
, orig
);
911 clear_user_return_notifier(tsk
);
912 clear_tsk_need_resched(tsk
);
913 set_task_stack_end_magic(tsk
);
914 clear_syscall_work_syscall_user_dispatch(tsk
);
916 #ifdef CONFIG_STACKPROTECTOR
917 tsk
->stack_canary
= get_random_canary();
919 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
920 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
923 * One for the user space visible state that goes away when reaped.
924 * One for the scheduler.
926 refcount_set(&tsk
->rcu_users
, 2);
927 /* One for the rcu users */
928 refcount_set(&tsk
->usage
, 1);
929 #ifdef CONFIG_BLK_DEV_IO_TRACE
932 tsk
->splice_pipe
= NULL
;
933 tsk
->task_frag
.page
= NULL
;
934 tsk
->wake_q
.next
= NULL
;
936 account_kernel_stack(tsk
, 1);
939 kmap_local_fork(tsk
);
941 #ifdef CONFIG_FAULT_INJECTION
945 #ifdef CONFIG_BLK_CGROUP
946 tsk
->throttle_queue
= NULL
;
947 tsk
->use_memdelay
= 0;
951 tsk
->active_memcg
= NULL
;
956 free_thread_stack(tsk
);
958 free_task_struct(tsk
);
962 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
964 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
966 static int __init
coredump_filter_setup(char *s
)
968 default_dump_filter
=
969 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
970 MMF_DUMP_FILTER_MASK
;
974 __setup("coredump_filter=", coredump_filter_setup
);
976 #include <linux/init_task.h>
978 static void mm_init_aio(struct mm_struct
*mm
)
981 spin_lock_init(&mm
->ioctx_lock
);
982 mm
->ioctx_table
= NULL
;
986 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
987 struct task_struct
*p
)
991 WRITE_ONCE(mm
->owner
, NULL
);
995 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
1002 static void mm_init_pasid(struct mm_struct
*mm
)
1004 #ifdef CONFIG_IOMMU_SUPPORT
1005 mm
->pasid
= INIT_PASID
;
1009 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1011 #ifdef CONFIG_UPROBES
1012 mm
->uprobes_state
.xol_area
= NULL
;
1016 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1017 struct user_namespace
*user_ns
)
1020 mm
->mm_rb
= RB_ROOT
;
1021 mm
->vmacache_seqnum
= 0;
1022 atomic_set(&mm
->mm_users
, 1);
1023 atomic_set(&mm
->mm_count
, 1);
1024 seqcount_init(&mm
->write_protect_seq
);
1026 INIT_LIST_HEAD(&mm
->mmlist
);
1027 mm
->core_state
= NULL
;
1028 mm_pgtables_bytes_init(mm
);
1031 atomic_set(&mm
->has_pinned
, 0);
1032 atomic64_set(&mm
->pinned_vm
, 0);
1033 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1034 spin_lock_init(&mm
->page_table_lock
);
1035 spin_lock_init(&mm
->arg_lock
);
1036 mm_init_cpumask(mm
);
1038 mm_init_owner(mm
, p
);
1040 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1041 mmu_notifier_subscriptions_init(mm
);
1042 init_tlb_flush_pending(mm
);
1043 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1044 mm
->pmd_huge_pte
= NULL
;
1046 mm_init_uprobes_state(mm
);
1049 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1050 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1052 mm
->flags
= default_dump_filter
;
1056 if (mm_alloc_pgd(mm
))
1059 if (init_new_context(p
, mm
))
1060 goto fail_nocontext
;
1062 mm
->user_ns
= get_user_ns(user_ns
);
1073 * Allocate and initialize an mm_struct.
1075 struct mm_struct
*mm_alloc(void)
1077 struct mm_struct
*mm
;
1083 memset(mm
, 0, sizeof(*mm
));
1084 return mm_init(mm
, current
, current_user_ns());
1087 static inline void __mmput(struct mm_struct
*mm
)
1089 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1091 uprobe_clear_state(mm
);
1094 khugepaged_exit(mm
); /* must run before exit_mmap */
1096 mm_put_huge_zero_page(mm
);
1097 set_mm_exe_file(mm
, NULL
);
1098 if (!list_empty(&mm
->mmlist
)) {
1099 spin_lock(&mmlist_lock
);
1100 list_del(&mm
->mmlist
);
1101 spin_unlock(&mmlist_lock
);
1104 module_put(mm
->binfmt
->module
);
1109 * Decrement the use count and release all resources for an mm.
1111 void mmput(struct mm_struct
*mm
)
1115 if (atomic_dec_and_test(&mm
->mm_users
))
1118 EXPORT_SYMBOL_GPL(mmput
);
1121 static void mmput_async_fn(struct work_struct
*work
)
1123 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1129 void mmput_async(struct mm_struct
*mm
)
1131 if (atomic_dec_and_test(&mm
->mm_users
)) {
1132 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1133 schedule_work(&mm
->async_put_work
);
1136 EXPORT_SYMBOL(mmput_async
);
1140 * set_mm_exe_file - change a reference to the mm's executable file
1142 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1144 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1145 * invocations: in mmput() nobody alive left, in execve task is single
1146 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1147 * mm->exe_file, but does so without using set_mm_exe_file() in order
1148 * to do avoid the need for any locks.
1150 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1152 struct file
*old_exe_file
;
1155 * It is safe to dereference the exe_file without RCU as
1156 * this function is only called if nobody else can access
1157 * this mm -- see comment above for justification.
1159 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1162 get_file(new_exe_file
);
1163 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1169 * get_mm_exe_file - acquire a reference to the mm's executable file
1171 * Returns %NULL if mm has no associated executable file.
1172 * User must release file via fput().
1174 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1176 struct file
*exe_file
;
1179 exe_file
= rcu_dereference(mm
->exe_file
);
1180 if (exe_file
&& !get_file_rcu(exe_file
))
1185 EXPORT_SYMBOL(get_mm_exe_file
);
1188 * get_task_exe_file - acquire a reference to the task's executable file
1190 * Returns %NULL if task's mm (if any) has no associated executable file or
1191 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1192 * User must release file via fput().
1194 struct file
*get_task_exe_file(struct task_struct
*task
)
1196 struct file
*exe_file
= NULL
;
1197 struct mm_struct
*mm
;
1202 if (!(task
->flags
& PF_KTHREAD
))
1203 exe_file
= get_mm_exe_file(mm
);
1208 EXPORT_SYMBOL(get_task_exe_file
);
1211 * get_task_mm - acquire a reference to the task's mm
1213 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1214 * this kernel workthread has transiently adopted a user mm with use_mm,
1215 * to do its AIO) is not set and if so returns a reference to it, after
1216 * bumping up the use count. User must release the mm via mmput()
1217 * after use. Typically used by /proc and ptrace.
1219 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1221 struct mm_struct
*mm
;
1226 if (task
->flags
& PF_KTHREAD
)
1234 EXPORT_SYMBOL_GPL(get_task_mm
);
1236 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1238 struct mm_struct
*mm
;
1241 err
= down_read_killable(&task
->signal
->exec_update_lock
);
1243 return ERR_PTR(err
);
1245 mm
= get_task_mm(task
);
1246 if (mm
&& mm
!= current
->mm
&&
1247 !ptrace_may_access(task
, mode
)) {
1249 mm
= ERR_PTR(-EACCES
);
1251 up_read(&task
->signal
->exec_update_lock
);
1256 static void complete_vfork_done(struct task_struct
*tsk
)
1258 struct completion
*vfork
;
1261 vfork
= tsk
->vfork_done
;
1262 if (likely(vfork
)) {
1263 tsk
->vfork_done
= NULL
;
1269 static int wait_for_vfork_done(struct task_struct
*child
,
1270 struct completion
*vfork
)
1274 freezer_do_not_count();
1275 cgroup_enter_frozen();
1276 killed
= wait_for_completion_killable(vfork
);
1277 cgroup_leave_frozen(false);
1282 child
->vfork_done
= NULL
;
1286 put_task_struct(child
);
1290 /* Please note the differences between mmput and mm_release.
1291 * mmput is called whenever we stop holding onto a mm_struct,
1292 * error success whatever.
1294 * mm_release is called after a mm_struct has been removed
1295 * from the current process.
1297 * This difference is important for error handling, when we
1298 * only half set up a mm_struct for a new process and need to restore
1299 * the old one. Because we mmput the new mm_struct before
1300 * restoring the old one. . .
1301 * Eric Biederman 10 January 1998
1303 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1305 uprobe_free_utask(tsk
);
1307 /* Get rid of any cached register state */
1308 deactivate_mm(tsk
, mm
);
1311 * Signal userspace if we're not exiting with a core dump
1312 * because we want to leave the value intact for debugging
1315 if (tsk
->clear_child_tid
) {
1316 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1317 atomic_read(&mm
->mm_users
) > 1) {
1319 * We don't check the error code - if userspace has
1320 * not set up a proper pointer then tough luck.
1322 put_user(0, tsk
->clear_child_tid
);
1323 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1324 1, NULL
, NULL
, 0, 0);
1326 tsk
->clear_child_tid
= NULL
;
1330 * All done, finally we can wake up parent and return this mm to him.
1331 * Also kthread_stop() uses this completion for synchronization.
1333 if (tsk
->vfork_done
)
1334 complete_vfork_done(tsk
);
1337 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1339 futex_exit_release(tsk
);
1340 mm_release(tsk
, mm
);
1343 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1345 futex_exec_release(tsk
);
1346 mm_release(tsk
, mm
);
1350 * dup_mm() - duplicates an existing mm structure
1351 * @tsk: the task_struct with which the new mm will be associated.
1352 * @oldmm: the mm to duplicate.
1354 * Allocates a new mm structure and duplicates the provided @oldmm structure
1357 * Return: the duplicated mm or NULL on failure.
1359 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1360 struct mm_struct
*oldmm
)
1362 struct mm_struct
*mm
;
1369 memcpy(mm
, oldmm
, sizeof(*mm
));
1371 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1374 err
= dup_mmap(mm
, oldmm
);
1378 mm
->hiwater_rss
= get_mm_rss(mm
);
1379 mm
->hiwater_vm
= mm
->total_vm
;
1381 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1387 /* don't put binfmt in mmput, we haven't got module yet */
1389 mm_init_owner(mm
, NULL
);
1396 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1398 struct mm_struct
*mm
, *oldmm
;
1401 tsk
->min_flt
= tsk
->maj_flt
= 0;
1402 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1403 #ifdef CONFIG_DETECT_HUNG_TASK
1404 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1405 tsk
->last_switch_time
= 0;
1409 tsk
->active_mm
= NULL
;
1412 * Are we cloning a kernel thread?
1414 * We need to steal a active VM for that..
1416 oldmm
= current
->mm
;
1420 /* initialize the new vmacache entries */
1421 vmacache_flush(tsk
);
1423 if (clone_flags
& CLONE_VM
) {
1430 mm
= dup_mm(tsk
, current
->mm
);
1436 tsk
->active_mm
= mm
;
1443 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1445 struct fs_struct
*fs
= current
->fs
;
1446 if (clone_flags
& CLONE_FS
) {
1447 /* tsk->fs is already what we want */
1448 spin_lock(&fs
->lock
);
1450 spin_unlock(&fs
->lock
);
1454 spin_unlock(&fs
->lock
);
1457 tsk
->fs
= copy_fs_struct(fs
);
1463 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1465 struct files_struct
*oldf
, *newf
;
1469 * A background process may not have any files ...
1471 oldf
= current
->files
;
1475 if (clone_flags
& CLONE_FILES
) {
1476 atomic_inc(&oldf
->count
);
1480 newf
= dup_fd(oldf
, NR_OPEN_MAX
, &error
);
1490 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1493 struct io_context
*ioc
= current
->io_context
;
1494 struct io_context
*new_ioc
;
1499 * Share io context with parent, if CLONE_IO is set
1501 if (clone_flags
& CLONE_IO
) {
1503 tsk
->io_context
= ioc
;
1504 } else if (ioprio_valid(ioc
->ioprio
)) {
1505 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1506 if (unlikely(!new_ioc
))
1509 new_ioc
->ioprio
= ioc
->ioprio
;
1510 put_io_context(new_ioc
);
1516 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1518 struct sighand_struct
*sig
;
1520 if (clone_flags
& CLONE_SIGHAND
) {
1521 refcount_inc(¤t
->sighand
->count
);
1524 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1525 RCU_INIT_POINTER(tsk
->sighand
, sig
);
1529 refcount_set(&sig
->count
, 1);
1530 spin_lock_irq(¤t
->sighand
->siglock
);
1531 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1532 spin_unlock_irq(¤t
->sighand
->siglock
);
1534 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1535 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1536 flush_signal_handlers(tsk
, 0);
1541 void __cleanup_sighand(struct sighand_struct
*sighand
)
1543 if (refcount_dec_and_test(&sighand
->count
)) {
1544 signalfd_cleanup(sighand
);
1546 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1547 * without an RCU grace period, see __lock_task_sighand().
1549 kmem_cache_free(sighand_cachep
, sighand
);
1554 * Initialize POSIX timer handling for a thread group.
1556 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1558 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1559 unsigned long cpu_limit
;
1561 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1562 posix_cputimers_group_init(pct
, cpu_limit
);
1565 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1567 struct signal_struct
*sig
;
1569 if (clone_flags
& CLONE_THREAD
)
1572 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1577 sig
->nr_threads
= 1;
1578 atomic_set(&sig
->live
, 1);
1579 refcount_set(&sig
->sigcnt
, 1);
1581 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1582 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1583 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1585 init_waitqueue_head(&sig
->wait_chldexit
);
1586 sig
->curr_target
= tsk
;
1587 init_sigpending(&sig
->shared_pending
);
1588 INIT_HLIST_HEAD(&sig
->multiprocess
);
1589 seqlock_init(&sig
->stats_lock
);
1590 prev_cputime_init(&sig
->prev_cputime
);
1592 #ifdef CONFIG_POSIX_TIMERS
1593 INIT_LIST_HEAD(&sig
->posix_timers
);
1594 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1595 sig
->real_timer
.function
= it_real_fn
;
1598 task_lock(current
->group_leader
);
1599 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1600 task_unlock(current
->group_leader
);
1602 posix_cpu_timers_init_group(sig
);
1604 tty_audit_fork(sig
);
1605 sched_autogroup_fork(sig
);
1607 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1608 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1610 mutex_init(&sig
->cred_guard_mutex
);
1611 init_rwsem(&sig
->exec_update_lock
);
1616 static void copy_seccomp(struct task_struct
*p
)
1618 #ifdef CONFIG_SECCOMP
1620 * Must be called with sighand->lock held, which is common to
1621 * all threads in the group. Holding cred_guard_mutex is not
1622 * needed because this new task is not yet running and cannot
1625 assert_spin_locked(¤t
->sighand
->siglock
);
1627 /* Ref-count the new filter user, and assign it. */
1628 get_seccomp_filter(current
);
1629 p
->seccomp
= current
->seccomp
;
1632 * Explicitly enable no_new_privs here in case it got set
1633 * between the task_struct being duplicated and holding the
1634 * sighand lock. The seccomp state and nnp must be in sync.
1636 if (task_no_new_privs(current
))
1637 task_set_no_new_privs(p
);
1640 * If the parent gained a seccomp mode after copying thread
1641 * flags and between before we held the sighand lock, we have
1642 * to manually enable the seccomp thread flag here.
1644 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1645 set_task_syscall_work(p
, SECCOMP
);
1649 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1651 current
->clear_child_tid
= tidptr
;
1653 return task_pid_vnr(current
);
1656 static void rt_mutex_init_task(struct task_struct
*p
)
1658 raw_spin_lock_init(&p
->pi_lock
);
1659 #ifdef CONFIG_RT_MUTEXES
1660 p
->pi_waiters
= RB_ROOT_CACHED
;
1661 p
->pi_top_task
= NULL
;
1662 p
->pi_blocked_on
= NULL
;
1666 static inline void init_task_pid_links(struct task_struct
*task
)
1670 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
)
1671 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1675 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1677 if (type
== PIDTYPE_PID
)
1678 task
->thread_pid
= pid
;
1680 task
->signal
->pids
[type
] = pid
;
1683 static inline void rcu_copy_process(struct task_struct
*p
)
1685 #ifdef CONFIG_PREEMPT_RCU
1686 p
->rcu_read_lock_nesting
= 0;
1687 p
->rcu_read_unlock_special
.s
= 0;
1688 p
->rcu_blocked_node
= NULL
;
1689 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1690 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1691 #ifdef CONFIG_TASKS_RCU
1692 p
->rcu_tasks_holdout
= false;
1693 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1694 p
->rcu_tasks_idle_cpu
= -1;
1695 #endif /* #ifdef CONFIG_TASKS_RCU */
1696 #ifdef CONFIG_TASKS_TRACE_RCU
1697 p
->trc_reader_nesting
= 0;
1698 p
->trc_reader_special
.s
= 0;
1699 INIT_LIST_HEAD(&p
->trc_holdout_list
);
1700 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1703 struct pid
*pidfd_pid(const struct file
*file
)
1705 if (file
->f_op
== &pidfd_fops
)
1706 return file
->private_data
;
1708 return ERR_PTR(-EBADF
);
1711 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1713 struct pid
*pid
= file
->private_data
;
1715 file
->private_data
= NULL
;
1720 #ifdef CONFIG_PROC_FS
1722 * pidfd_show_fdinfo - print information about a pidfd
1723 * @m: proc fdinfo file
1724 * @f: file referencing a pidfd
1727 * This function will print the pid that a given pidfd refers to in the
1728 * pid namespace of the procfs instance.
1729 * If the pid namespace of the process is not a descendant of the pid
1730 * namespace of the procfs instance 0 will be shown as its pid. This is
1731 * similar to calling getppid() on a process whose parent is outside of
1732 * its pid namespace.
1735 * If pid namespaces are supported then this function will also print
1736 * the pid of a given pidfd refers to for all descendant pid namespaces
1737 * starting from the current pid namespace of the instance, i.e. the
1738 * Pid field and the first entry in the NSpid field will be identical.
1739 * If the pid namespace of the process is not a descendant of the pid
1740 * namespace of the procfs instance 0 will be shown as its first NSpid
1741 * entry and no others will be shown.
1742 * Note that this differs from the Pid and NSpid fields in
1743 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1744 * the pid namespace of the procfs instance. The difference becomes
1745 * obvious when sending around a pidfd between pid namespaces from a
1746 * different branch of the tree, i.e. where no ancestoral relation is
1747 * present between the pid namespaces:
1748 * - create two new pid namespaces ns1 and ns2 in the initial pid
1749 * namespace (also take care to create new mount namespaces in the
1750 * new pid namespace and mount procfs)
1751 * - create a process with a pidfd in ns1
1752 * - send pidfd from ns1 to ns2
1753 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1754 * have exactly one entry, which is 0
1756 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1758 struct pid
*pid
= f
->private_data
;
1759 struct pid_namespace
*ns
;
1762 if (likely(pid_has_task(pid
, PIDTYPE_PID
))) {
1763 ns
= proc_pid_ns(file_inode(m
->file
)->i_sb
);
1764 nr
= pid_nr_ns(pid
, ns
);
1767 seq_put_decimal_ll(m
, "Pid:\t", nr
);
1769 #ifdef CONFIG_PID_NS
1770 seq_put_decimal_ll(m
, "\nNSpid:\t", nr
);
1774 /* If nr is non-zero it means that 'pid' is valid and that
1775 * ns, i.e. the pid namespace associated with the procfs
1776 * instance, is in the pid namespace hierarchy of pid.
1777 * Start at one below the already printed level.
1779 for (i
= ns
->level
+ 1; i
<= pid
->level
; i
++)
1780 seq_put_decimal_ll(m
, "\t", pid
->numbers
[i
].nr
);
1788 * Poll support for process exit notification.
1790 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1792 struct pid
*pid
= file
->private_data
;
1793 __poll_t poll_flags
= 0;
1795 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1798 * Inform pollers only when the whole thread group exits.
1799 * If the thread group leader exits before all other threads in the
1800 * group, then poll(2) should block, similar to the wait(2) family.
1802 if (thread_group_exited(pid
))
1803 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1808 const struct file_operations pidfd_fops
= {
1809 .release
= pidfd_release
,
1811 #ifdef CONFIG_PROC_FS
1812 .show_fdinfo
= pidfd_show_fdinfo
,
1816 static void __delayed_free_task(struct rcu_head
*rhp
)
1818 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1823 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1825 if (IS_ENABLED(CONFIG_MEMCG
))
1826 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1831 static void copy_oom_score_adj(u64 clone_flags
, struct task_struct
*tsk
)
1833 /* Skip if kernel thread */
1837 /* Skip if spawning a thread or using vfork */
1838 if ((clone_flags
& (CLONE_VM
| CLONE_THREAD
| CLONE_VFORK
)) != CLONE_VM
)
1841 /* We need to synchronize with __set_oom_adj */
1842 mutex_lock(&oom_adj_mutex
);
1843 set_bit(MMF_MULTIPROCESS
, &tsk
->mm
->flags
);
1844 /* Update the values in case they were changed after copy_signal */
1845 tsk
->signal
->oom_score_adj
= current
->signal
->oom_score_adj
;
1846 tsk
->signal
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1847 mutex_unlock(&oom_adj_mutex
);
1851 * This creates a new process as a copy of the old one,
1852 * but does not actually start it yet.
1854 * It copies the registers, and all the appropriate
1855 * parts of the process environment (as per the clone
1856 * flags). The actual kick-off is left to the caller.
1858 static __latent_entropy
struct task_struct
*copy_process(
1862 struct kernel_clone_args
*args
)
1864 int pidfd
= -1, retval
;
1865 struct task_struct
*p
;
1866 struct multiprocess_signals delayed
;
1867 struct file
*pidfile
= NULL
;
1868 u64 clone_flags
= args
->flags
;
1869 struct nsproxy
*nsp
= current
->nsproxy
;
1872 * Don't allow sharing the root directory with processes in a different
1875 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1876 return ERR_PTR(-EINVAL
);
1878 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1879 return ERR_PTR(-EINVAL
);
1881 if ((clone_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
)
1882 if (!capable(CAP_SYS_ADMIN
))
1883 return ERR_PTR(-EPERM
);
1886 * Thread groups must share signals as well, and detached threads
1887 * can only be started up within the thread group.
1889 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1890 return ERR_PTR(-EINVAL
);
1893 * Shared signal handlers imply shared VM. By way of the above,
1894 * thread groups also imply shared VM. Blocking this case allows
1895 * for various simplifications in other code.
1897 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1898 return ERR_PTR(-EINVAL
);
1901 * Siblings of global init remain as zombies on exit since they are
1902 * not reaped by their parent (swapper). To solve this and to avoid
1903 * multi-rooted process trees, prevent global and container-inits
1904 * from creating siblings.
1906 if ((clone_flags
& CLONE_PARENT
) &&
1907 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1908 return ERR_PTR(-EINVAL
);
1911 * If the new process will be in a different pid or user namespace
1912 * do not allow it to share a thread group with the forking task.
1914 if (clone_flags
& CLONE_THREAD
) {
1915 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1916 (task_active_pid_ns(current
) != nsp
->pid_ns_for_children
))
1917 return ERR_PTR(-EINVAL
);
1921 * If the new process will be in a different time namespace
1922 * do not allow it to share VM or a thread group with the forking task.
1924 if (clone_flags
& (CLONE_THREAD
| CLONE_VM
)) {
1925 if (nsp
->time_ns
!= nsp
->time_ns_for_children
)
1926 return ERR_PTR(-EINVAL
);
1929 if (clone_flags
& CLONE_PIDFD
) {
1931 * - CLONE_DETACHED is blocked so that we can potentially
1932 * reuse it later for CLONE_PIDFD.
1933 * - CLONE_THREAD is blocked until someone really needs it.
1935 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1936 return ERR_PTR(-EINVAL
);
1940 * Force any signals received before this point to be delivered
1941 * before the fork happens. Collect up signals sent to multiple
1942 * processes that happen during the fork and delay them so that
1943 * they appear to happen after the fork.
1945 sigemptyset(&delayed
.signal
);
1946 INIT_HLIST_NODE(&delayed
.node
);
1948 spin_lock_irq(¤t
->sighand
->siglock
);
1949 if (!(clone_flags
& CLONE_THREAD
))
1950 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1951 recalc_sigpending();
1952 spin_unlock_irq(¤t
->sighand
->siglock
);
1953 retval
= -ERESTARTNOINTR
;
1954 if (signal_pending(current
))
1958 p
= dup_task_struct(current
, node
);
1963 * This _must_ happen before we call free_task(), i.e. before we jump
1964 * to any of the bad_fork_* labels. This is to avoid freeing
1965 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1966 * kernel threads (PF_KTHREAD).
1968 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1970 * Clear TID on mm_release()?
1972 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1974 ftrace_graph_init_task(p
);
1976 rt_mutex_init_task(p
);
1978 lockdep_assert_irqs_enabled();
1979 #ifdef CONFIG_PROVE_LOCKING
1980 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1983 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1984 task_rlimit(p
, RLIMIT_NPROC
)) {
1985 if (p
->real_cred
->user
!= INIT_USER
&&
1986 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1989 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1991 retval
= copy_creds(p
, clone_flags
);
1996 * If multiple threads are within copy_process(), then this check
1997 * triggers too late. This doesn't hurt, the check is only there
1998 * to stop root fork bombs.
2001 if (data_race(nr_threads
>= max_threads
))
2002 goto bad_fork_cleanup_count
;
2004 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
2005 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
2006 p
->flags
|= PF_FORKNOEXEC
;
2007 INIT_LIST_HEAD(&p
->children
);
2008 INIT_LIST_HEAD(&p
->sibling
);
2009 rcu_copy_process(p
);
2010 p
->vfork_done
= NULL
;
2011 spin_lock_init(&p
->alloc_lock
);
2013 init_sigpending(&p
->pending
);
2015 p
->utime
= p
->stime
= p
->gtime
= 0;
2016 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2017 p
->utimescaled
= p
->stimescaled
= 0;
2019 prev_cputime_init(&p
->prev_cputime
);
2021 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2022 seqcount_init(&p
->vtime
.seqcount
);
2023 p
->vtime
.starttime
= 0;
2024 p
->vtime
.state
= VTIME_INACTIVE
;
2027 #ifdef CONFIG_IO_URING
2031 #if defined(SPLIT_RSS_COUNTING)
2032 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
2035 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
2041 task_io_accounting_init(&p
->ioac
);
2042 acct_clear_integrals(p
);
2044 posix_cputimers_init(&p
->posix_cputimers
);
2046 p
->io_context
= NULL
;
2047 audit_set_context(p
, NULL
);
2050 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2051 if (IS_ERR(p
->mempolicy
)) {
2052 retval
= PTR_ERR(p
->mempolicy
);
2053 p
->mempolicy
= NULL
;
2054 goto bad_fork_cleanup_threadgroup_lock
;
2057 #ifdef CONFIG_CPUSETS
2058 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2059 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2060 seqcount_spinlock_init(&p
->mems_allowed_seq
, &p
->alloc_lock
);
2062 #ifdef CONFIG_TRACE_IRQFLAGS
2063 memset(&p
->irqtrace
, 0, sizeof(p
->irqtrace
));
2064 p
->irqtrace
.hardirq_disable_ip
= _THIS_IP_
;
2065 p
->irqtrace
.softirq_enable_ip
= _THIS_IP_
;
2066 p
->softirqs_enabled
= 1;
2067 p
->softirq_context
= 0;
2070 p
->pagefault_disabled
= 0;
2072 #ifdef CONFIG_LOCKDEP
2073 lockdep_init_task(p
);
2076 #ifdef CONFIG_DEBUG_MUTEXES
2077 p
->blocked_on
= NULL
; /* not blocked yet */
2079 #ifdef CONFIG_BCACHE
2080 p
->sequential_io
= 0;
2081 p
->sequential_io_avg
= 0;
2084 /* Perform scheduler related setup. Assign this task to a CPU. */
2085 retval
= sched_fork(clone_flags
, p
);
2087 goto bad_fork_cleanup_policy
;
2089 retval
= perf_event_init_task(p
);
2091 goto bad_fork_cleanup_policy
;
2092 retval
= audit_alloc(p
);
2094 goto bad_fork_cleanup_perf
;
2095 /* copy all the process information */
2097 retval
= security_task_alloc(p
, clone_flags
);
2099 goto bad_fork_cleanup_audit
;
2100 retval
= copy_semundo(clone_flags
, p
);
2102 goto bad_fork_cleanup_security
;
2103 retval
= copy_files(clone_flags
, p
);
2105 goto bad_fork_cleanup_semundo
;
2106 retval
= copy_fs(clone_flags
, p
);
2108 goto bad_fork_cleanup_files
;
2109 retval
= copy_sighand(clone_flags
, p
);
2111 goto bad_fork_cleanup_fs
;
2112 retval
= copy_signal(clone_flags
, p
);
2114 goto bad_fork_cleanup_sighand
;
2115 retval
= copy_mm(clone_flags
, p
);
2117 goto bad_fork_cleanup_signal
;
2118 retval
= copy_namespaces(clone_flags
, p
);
2120 goto bad_fork_cleanup_mm
;
2121 retval
= copy_io(clone_flags
, p
);
2123 goto bad_fork_cleanup_namespaces
;
2124 retval
= copy_thread(clone_flags
, args
->stack
, args
->stack_size
, p
, args
->tls
);
2126 goto bad_fork_cleanup_io
;
2128 stackleak_task_init(p
);
2130 if (pid
!= &init_struct_pid
) {
2131 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2132 args
->set_tid_size
);
2134 retval
= PTR_ERR(pid
);
2135 goto bad_fork_cleanup_thread
;
2140 * This has to happen after we've potentially unshared the file
2141 * descriptor table (so that the pidfd doesn't leak into the child
2142 * if the fd table isn't shared).
2144 if (clone_flags
& CLONE_PIDFD
) {
2145 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2147 goto bad_fork_free_pid
;
2151 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2152 O_RDWR
| O_CLOEXEC
);
2153 if (IS_ERR(pidfile
)) {
2154 put_unused_fd(pidfd
);
2155 retval
= PTR_ERR(pidfile
);
2156 goto bad_fork_free_pid
;
2158 get_pid(pid
); /* held by pidfile now */
2160 retval
= put_user(pidfd
, args
->pidfd
);
2162 goto bad_fork_put_pidfd
;
2171 * sigaltstack should be cleared when sharing the same VM
2173 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2177 * Syscall tracing and stepping should be turned off in the
2178 * child regardless of CLONE_PTRACE.
2180 user_disable_single_step(p
);
2181 clear_task_syscall_work(p
, SYSCALL_TRACE
);
2182 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2183 clear_task_syscall_work(p
, SYSCALL_EMU
);
2185 clear_tsk_latency_tracing(p
);
2187 /* ok, now we should be set up.. */
2188 p
->pid
= pid_nr(pid
);
2189 if (clone_flags
& CLONE_THREAD
) {
2190 p
->group_leader
= current
->group_leader
;
2191 p
->tgid
= current
->tgid
;
2193 p
->group_leader
= p
;
2198 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2199 p
->dirty_paused_when
= 0;
2201 p
->pdeath_signal
= 0;
2202 INIT_LIST_HEAD(&p
->thread_group
);
2203 p
->task_works
= NULL
;
2205 #ifdef CONFIG_KRETPROBES
2206 p
->kretprobe_instances
.first
= NULL
;
2210 * Ensure that the cgroup subsystem policies allow the new process to be
2211 * forked. It should be noted that the new process's css_set can be changed
2212 * between here and cgroup_post_fork() if an organisation operation is in
2215 retval
= cgroup_can_fork(p
, args
);
2217 goto bad_fork_put_pidfd
;
2220 * From this point on we must avoid any synchronous user-space
2221 * communication until we take the tasklist-lock. In particular, we do
2222 * not want user-space to be able to predict the process start-time by
2223 * stalling fork(2) after we recorded the start_time but before it is
2224 * visible to the system.
2227 p
->start_time
= ktime_get_ns();
2228 p
->start_boottime
= ktime_get_boottime_ns();
2231 * Make it visible to the rest of the system, but dont wake it up yet.
2232 * Need tasklist lock for parent etc handling!
2234 write_lock_irq(&tasklist_lock
);
2236 /* CLONE_PARENT re-uses the old parent */
2237 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2238 p
->real_parent
= current
->real_parent
;
2239 p
->parent_exec_id
= current
->parent_exec_id
;
2240 if (clone_flags
& CLONE_THREAD
)
2241 p
->exit_signal
= -1;
2243 p
->exit_signal
= current
->group_leader
->exit_signal
;
2245 p
->real_parent
= current
;
2246 p
->parent_exec_id
= current
->self_exec_id
;
2247 p
->exit_signal
= args
->exit_signal
;
2250 klp_copy_process(p
);
2252 spin_lock(¤t
->sighand
->siglock
);
2255 * Copy seccomp details explicitly here, in case they were changed
2256 * before holding sighand lock.
2260 rseq_fork(p
, clone_flags
);
2262 /* Don't start children in a dying pid namespace */
2263 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2265 goto bad_fork_cancel_cgroup
;
2268 /* Let kill terminate clone/fork in the middle */
2269 if (fatal_signal_pending(current
)) {
2271 goto bad_fork_cancel_cgroup
;
2274 /* past the last point of failure */
2276 fd_install(pidfd
, pidfile
);
2278 init_task_pid_links(p
);
2279 if (likely(p
->pid
)) {
2280 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2282 init_task_pid(p
, PIDTYPE_PID
, pid
);
2283 if (thread_group_leader(p
)) {
2284 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2285 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2286 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2288 if (is_child_reaper(pid
)) {
2289 ns_of_pid(pid
)->child_reaper
= p
;
2290 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2292 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2293 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2295 * Inherit has_child_subreaper flag under the same
2296 * tasklist_lock with adding child to the process tree
2297 * for propagate_has_child_subreaper optimization.
2299 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2300 p
->real_parent
->signal
->is_child_subreaper
;
2301 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2302 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2303 attach_pid(p
, PIDTYPE_TGID
);
2304 attach_pid(p
, PIDTYPE_PGID
);
2305 attach_pid(p
, PIDTYPE_SID
);
2306 __this_cpu_inc(process_counts
);
2308 current
->signal
->nr_threads
++;
2309 atomic_inc(¤t
->signal
->live
);
2310 refcount_inc(¤t
->signal
->sigcnt
);
2311 task_join_group_stop(p
);
2312 list_add_tail_rcu(&p
->thread_group
,
2313 &p
->group_leader
->thread_group
);
2314 list_add_tail_rcu(&p
->thread_node
,
2315 &p
->signal
->thread_head
);
2317 attach_pid(p
, PIDTYPE_PID
);
2321 hlist_del_init(&delayed
.node
);
2322 spin_unlock(¤t
->sighand
->siglock
);
2323 syscall_tracepoint_update(p
);
2324 write_unlock_irq(&tasklist_lock
);
2326 proc_fork_connector(p
);
2328 cgroup_post_fork(p
, args
);
2331 trace_task_newtask(p
, clone_flags
);
2332 uprobe_copy_process(p
, clone_flags
);
2334 copy_oom_score_adj(clone_flags
, p
);
2338 bad_fork_cancel_cgroup
:
2339 spin_unlock(¤t
->sighand
->siglock
);
2340 write_unlock_irq(&tasklist_lock
);
2341 cgroup_cancel_fork(p
, args
);
2343 if (clone_flags
& CLONE_PIDFD
) {
2345 put_unused_fd(pidfd
);
2348 if (pid
!= &init_struct_pid
)
2350 bad_fork_cleanup_thread
:
2352 bad_fork_cleanup_io
:
2355 bad_fork_cleanup_namespaces
:
2356 exit_task_namespaces(p
);
2357 bad_fork_cleanup_mm
:
2359 mm_clear_owner(p
->mm
, p
);
2362 bad_fork_cleanup_signal
:
2363 if (!(clone_flags
& CLONE_THREAD
))
2364 free_signal_struct(p
->signal
);
2365 bad_fork_cleanup_sighand
:
2366 __cleanup_sighand(p
->sighand
);
2367 bad_fork_cleanup_fs
:
2368 exit_fs(p
); /* blocking */
2369 bad_fork_cleanup_files
:
2370 exit_files(p
); /* blocking */
2371 bad_fork_cleanup_semundo
:
2373 bad_fork_cleanup_security
:
2374 security_task_free(p
);
2375 bad_fork_cleanup_audit
:
2377 bad_fork_cleanup_perf
:
2378 perf_event_free_task(p
);
2379 bad_fork_cleanup_policy
:
2380 lockdep_free_task(p
);
2382 mpol_put(p
->mempolicy
);
2383 bad_fork_cleanup_threadgroup_lock
:
2385 delayacct_tsk_free(p
);
2386 bad_fork_cleanup_count
:
2387 atomic_dec(&p
->cred
->user
->processes
);
2390 p
->state
= TASK_DEAD
;
2392 delayed_free_task(p
);
2394 spin_lock_irq(¤t
->sighand
->siglock
);
2395 hlist_del_init(&delayed
.node
);
2396 spin_unlock_irq(¤t
->sighand
->siglock
);
2397 return ERR_PTR(retval
);
2400 static inline void init_idle_pids(struct task_struct
*idle
)
2404 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2405 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2406 init_task_pid(idle
, type
, &init_struct_pid
);
2410 struct task_struct
*fork_idle(int cpu
)
2412 struct task_struct
*task
;
2413 struct kernel_clone_args args
= {
2417 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2418 if (!IS_ERR(task
)) {
2419 init_idle_pids(task
);
2420 init_idle(task
, cpu
);
2426 struct mm_struct
*copy_init_mm(void)
2428 return dup_mm(NULL
, &init_mm
);
2432 * Ok, this is the main fork-routine.
2434 * It copies the process, and if successful kick-starts
2435 * it and waits for it to finish using the VM if required.
2437 * args->exit_signal is expected to be checked for sanity by the caller.
2439 pid_t
kernel_clone(struct kernel_clone_args
*args
)
2441 u64 clone_flags
= args
->flags
;
2442 struct completion vfork
;
2444 struct task_struct
*p
;
2449 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2450 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2451 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2452 * field in struct clone_args and it still doesn't make sense to have
2453 * them both point at the same memory location. Performing this check
2454 * here has the advantage that we don't need to have a separate helper
2455 * to check for legacy clone().
2457 if ((args
->flags
& CLONE_PIDFD
) &&
2458 (args
->flags
& CLONE_PARENT_SETTID
) &&
2459 (args
->pidfd
== args
->parent_tid
))
2463 * Determine whether and which event to report to ptracer. When
2464 * called from kernel_thread or CLONE_UNTRACED is explicitly
2465 * requested, no event is reported; otherwise, report if the event
2466 * for the type of forking is enabled.
2468 if (!(clone_flags
& CLONE_UNTRACED
)) {
2469 if (clone_flags
& CLONE_VFORK
)
2470 trace
= PTRACE_EVENT_VFORK
;
2471 else if (args
->exit_signal
!= SIGCHLD
)
2472 trace
= PTRACE_EVENT_CLONE
;
2474 trace
= PTRACE_EVENT_FORK
;
2476 if (likely(!ptrace_event_enabled(current
, trace
)))
2480 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2481 add_latent_entropy();
2487 * Do this prior waking up the new thread - the thread pointer
2488 * might get invalid after that point, if the thread exits quickly.
2490 trace_sched_process_fork(current
, p
);
2492 pid
= get_task_pid(p
, PIDTYPE_PID
);
2495 if (clone_flags
& CLONE_PARENT_SETTID
)
2496 put_user(nr
, args
->parent_tid
);
2498 if (clone_flags
& CLONE_VFORK
) {
2499 p
->vfork_done
= &vfork
;
2500 init_completion(&vfork
);
2504 wake_up_new_task(p
);
2506 /* forking complete and child started to run, tell ptracer */
2507 if (unlikely(trace
))
2508 ptrace_event_pid(trace
, pid
);
2510 if (clone_flags
& CLONE_VFORK
) {
2511 if (!wait_for_vfork_done(p
, &vfork
))
2512 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2520 * Create a kernel thread.
2522 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2524 struct kernel_clone_args args
= {
2525 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2526 CLONE_UNTRACED
) & ~CSIGNAL
),
2527 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2528 .stack
= (unsigned long)fn
,
2529 .stack_size
= (unsigned long)arg
,
2532 return kernel_clone(&args
);
2535 #ifdef __ARCH_WANT_SYS_FORK
2536 SYSCALL_DEFINE0(fork
)
2539 struct kernel_clone_args args
= {
2540 .exit_signal
= SIGCHLD
,
2543 return kernel_clone(&args
);
2545 /* can not support in nommu mode */
2551 #ifdef __ARCH_WANT_SYS_VFORK
2552 SYSCALL_DEFINE0(vfork
)
2554 struct kernel_clone_args args
= {
2555 .flags
= CLONE_VFORK
| CLONE_VM
,
2556 .exit_signal
= SIGCHLD
,
2559 return kernel_clone(&args
);
2563 #ifdef __ARCH_WANT_SYS_CLONE
2564 #ifdef CONFIG_CLONE_BACKWARDS
2565 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2566 int __user
*, parent_tidptr
,
2568 int __user
*, child_tidptr
)
2569 #elif defined(CONFIG_CLONE_BACKWARDS2)
2570 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2571 int __user
*, parent_tidptr
,
2572 int __user
*, child_tidptr
,
2574 #elif defined(CONFIG_CLONE_BACKWARDS3)
2575 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2577 int __user
*, parent_tidptr
,
2578 int __user
*, child_tidptr
,
2581 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2582 int __user
*, parent_tidptr
,
2583 int __user
*, child_tidptr
,
2587 struct kernel_clone_args args
= {
2588 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2589 .pidfd
= parent_tidptr
,
2590 .child_tid
= child_tidptr
,
2591 .parent_tid
= parent_tidptr
,
2592 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2597 return kernel_clone(&args
);
2601 #ifdef __ARCH_WANT_SYS_CLONE3
2603 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2604 struct clone_args __user
*uargs
,
2608 struct clone_args args
;
2609 pid_t
*kset_tid
= kargs
->set_tid
;
2611 BUILD_BUG_ON(offsetofend(struct clone_args
, tls
) !=
2612 CLONE_ARGS_SIZE_VER0
);
2613 BUILD_BUG_ON(offsetofend(struct clone_args
, set_tid_size
) !=
2614 CLONE_ARGS_SIZE_VER1
);
2615 BUILD_BUG_ON(offsetofend(struct clone_args
, cgroup
) !=
2616 CLONE_ARGS_SIZE_VER2
);
2617 BUILD_BUG_ON(sizeof(struct clone_args
) != CLONE_ARGS_SIZE_VER2
);
2619 if (unlikely(usize
> PAGE_SIZE
))
2621 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2624 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2628 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2631 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2634 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2638 * Verify that higher 32bits of exit_signal are unset and that
2639 * it is a valid signal
2641 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2642 !valid_signal(args
.exit_signal
)))
2645 if ((args
.flags
& CLONE_INTO_CGROUP
) &&
2646 (args
.cgroup
> INT_MAX
|| usize
< CLONE_ARGS_SIZE_VER2
))
2649 *kargs
= (struct kernel_clone_args
){
2650 .flags
= args
.flags
,
2651 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2652 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2653 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2654 .exit_signal
= args
.exit_signal
,
2655 .stack
= args
.stack
,
2656 .stack_size
= args
.stack_size
,
2658 .set_tid_size
= args
.set_tid_size
,
2659 .cgroup
= args
.cgroup
,
2663 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
2664 (kargs
->set_tid_size
* sizeof(pid_t
))))
2667 kargs
->set_tid
= kset_tid
;
2673 * clone3_stack_valid - check and prepare stack
2674 * @kargs: kernel clone args
2676 * Verify that the stack arguments userspace gave us are sane.
2677 * In addition, set the stack direction for userspace since it's easy for us to
2680 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2682 if (kargs
->stack
== 0) {
2683 if (kargs
->stack_size
> 0)
2686 if (kargs
->stack_size
== 0)
2689 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2692 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2693 kargs
->stack
+= kargs
->stack_size
;
2700 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2702 /* Verify that no unknown flags are passed along. */
2704 ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
| CLONE_INTO_CGROUP
))
2708 * - make the CLONE_DETACHED bit reuseable for clone3
2709 * - make the CSIGNAL bits reuseable for clone3
2711 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2714 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
2715 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
2718 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2722 if (!clone3_stack_valid(kargs
))
2729 * clone3 - create a new process with specific properties
2730 * @uargs: argument structure
2731 * @size: size of @uargs
2733 * clone3() is the extensible successor to clone()/clone2().
2734 * It takes a struct as argument that is versioned by its size.
2736 * Return: On success, a positive PID for the child process.
2737 * On error, a negative errno number.
2739 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2743 struct kernel_clone_args kargs
;
2744 pid_t set_tid
[MAX_PID_NS_LEVEL
];
2746 kargs
.set_tid
= set_tid
;
2748 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2752 if (!clone3_args_valid(&kargs
))
2755 return kernel_clone(&kargs
);
2759 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2761 struct task_struct
*leader
, *parent
, *child
;
2764 read_lock(&tasklist_lock
);
2765 leader
= top
= top
->group_leader
;
2767 for_each_thread(leader
, parent
) {
2768 list_for_each_entry(child
, &parent
->children
, sibling
) {
2769 res
= visitor(child
, data
);
2781 if (leader
!= top
) {
2783 parent
= child
->real_parent
;
2784 leader
= parent
->group_leader
;
2788 read_unlock(&tasklist_lock
);
2791 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2792 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2795 static void sighand_ctor(void *data
)
2797 struct sighand_struct
*sighand
= data
;
2799 spin_lock_init(&sighand
->siglock
);
2800 init_waitqueue_head(&sighand
->signalfd_wqh
);
2803 void __init
proc_caches_init(void)
2805 unsigned int mm_size
;
2807 sighand_cachep
= kmem_cache_create("sighand_cache",
2808 sizeof(struct sighand_struct
), 0,
2809 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2810 SLAB_ACCOUNT
, sighand_ctor
);
2811 signal_cachep
= kmem_cache_create("signal_cache",
2812 sizeof(struct signal_struct
), 0,
2813 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2815 files_cachep
= kmem_cache_create("files_cache",
2816 sizeof(struct files_struct
), 0,
2817 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2819 fs_cachep
= kmem_cache_create("fs_cache",
2820 sizeof(struct fs_struct
), 0,
2821 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2825 * The mm_cpumask is located at the end of mm_struct, and is
2826 * dynamically sized based on the maximum CPU number this system
2827 * can have, taking hotplug into account (nr_cpu_ids).
2829 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2831 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2832 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2833 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2834 offsetof(struct mm_struct
, saved_auxv
),
2835 sizeof_field(struct mm_struct
, saved_auxv
),
2837 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2839 nsproxy_cache_init();
2843 * Check constraints on flags passed to the unshare system call.
2845 static int check_unshare_flags(unsigned long unshare_flags
)
2847 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2848 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2849 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2850 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
|
2854 * Not implemented, but pretend it works if there is nothing
2855 * to unshare. Note that unsharing the address space or the
2856 * signal handlers also need to unshare the signal queues (aka
2859 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2860 if (!thread_group_empty(current
))
2863 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2864 if (refcount_read(¤t
->sighand
->count
) > 1)
2867 if (unshare_flags
& CLONE_VM
) {
2868 if (!current_is_single_threaded())
2876 * Unshare the filesystem structure if it is being shared
2878 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2880 struct fs_struct
*fs
= current
->fs
;
2882 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2885 /* don't need lock here; in the worst case we'll do useless copy */
2889 *new_fsp
= copy_fs_struct(fs
);
2897 * Unshare file descriptor table if it is being shared
2899 int unshare_fd(unsigned long unshare_flags
, unsigned int max_fds
,
2900 struct files_struct
**new_fdp
)
2902 struct files_struct
*fd
= current
->files
;
2905 if ((unshare_flags
& CLONE_FILES
) &&
2906 (fd
&& atomic_read(&fd
->count
) > 1)) {
2907 *new_fdp
= dup_fd(fd
, max_fds
, &error
);
2916 * unshare allows a process to 'unshare' part of the process
2917 * context which was originally shared using clone. copy_*
2918 * functions used by kernel_clone() cannot be used here directly
2919 * because they modify an inactive task_struct that is being
2920 * constructed. Here we are modifying the current, active,
2923 int ksys_unshare(unsigned long unshare_flags
)
2925 struct fs_struct
*fs
, *new_fs
= NULL
;
2926 struct files_struct
*fd
, *new_fd
= NULL
;
2927 struct cred
*new_cred
= NULL
;
2928 struct nsproxy
*new_nsproxy
= NULL
;
2933 * If unsharing a user namespace must also unshare the thread group
2934 * and unshare the filesystem root and working directories.
2936 if (unshare_flags
& CLONE_NEWUSER
)
2937 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2939 * If unsharing vm, must also unshare signal handlers.
2941 if (unshare_flags
& CLONE_VM
)
2942 unshare_flags
|= CLONE_SIGHAND
;
2944 * If unsharing a signal handlers, must also unshare the signal queues.
2946 if (unshare_flags
& CLONE_SIGHAND
)
2947 unshare_flags
|= CLONE_THREAD
;
2949 * If unsharing namespace, must also unshare filesystem information.
2951 if (unshare_flags
& CLONE_NEWNS
)
2952 unshare_flags
|= CLONE_FS
;
2954 if ((unshare_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
) {
2956 if (!capable(CAP_SYS_ADMIN
))
2957 goto bad_unshare_out
;
2960 err
= check_unshare_flags(unshare_flags
);
2962 goto bad_unshare_out
;
2964 * CLONE_NEWIPC must also detach from the undolist: after switching
2965 * to a new ipc namespace, the semaphore arrays from the old
2966 * namespace are unreachable.
2968 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2970 err
= unshare_fs(unshare_flags
, &new_fs
);
2972 goto bad_unshare_out
;
2973 err
= unshare_fd(unshare_flags
, NR_OPEN_MAX
, &new_fd
);
2975 goto bad_unshare_cleanup_fs
;
2976 err
= unshare_userns(unshare_flags
, &new_cred
);
2978 goto bad_unshare_cleanup_fd
;
2979 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2982 goto bad_unshare_cleanup_cred
;
2984 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2987 * CLONE_SYSVSEM is equivalent to sys_exit().
2991 if (unshare_flags
& CLONE_NEWIPC
) {
2992 /* Orphan segments in old ns (see sem above). */
2994 shm_init_task(current
);
2998 switch_task_namespaces(current
, new_nsproxy
);
3004 spin_lock(&fs
->lock
);
3005 current
->fs
= new_fs
;
3010 spin_unlock(&fs
->lock
);
3014 fd
= current
->files
;
3015 current
->files
= new_fd
;
3019 task_unlock(current
);
3022 /* Install the new user namespace */
3023 commit_creds(new_cred
);
3028 perf_event_namespaces(current
);
3030 bad_unshare_cleanup_cred
:
3033 bad_unshare_cleanup_fd
:
3035 put_files_struct(new_fd
);
3037 bad_unshare_cleanup_fs
:
3039 free_fs_struct(new_fs
);
3045 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3047 return ksys_unshare(unshare_flags
);
3051 * Helper to unshare the files of the current task.
3052 * We don't want to expose copy_files internals to
3053 * the exec layer of the kernel.
3056 int unshare_files(void)
3058 struct task_struct
*task
= current
;
3059 struct files_struct
*old
, *copy
= NULL
;
3062 error
= unshare_fd(CLONE_FILES
, NR_OPEN_MAX
, ©
);
3070 put_files_struct(old
);
3074 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3075 void *buffer
, size_t *lenp
, loff_t
*ppos
)
3079 int threads
= max_threads
;
3081 int max
= MAX_THREADS
;
3088 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3092 max_threads
= threads
;