1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
43 #include <linux/hmm.h>
46 #include <linux/vmacache.h>
47 #include <linux/nsproxy.h>
48 #include <linux/capability.h>
49 #include <linux/cpu.h>
50 #include <linux/cgroup.h>
51 #include <linux/security.h>
52 #include <linux/hugetlb.h>
53 #include <linux/seccomp.h>
54 #include <linux/swap.h>
55 #include <linux/syscalls.h>
56 #include <linux/jiffies.h>
57 #include <linux/futex.h>
58 #include <linux/compat.h>
59 #include <linux/kthread.h>
60 #include <linux/task_io_accounting_ops.h>
61 #include <linux/rcupdate.h>
62 #include <linux/ptrace.h>
63 #include <linux/mount.h>
64 #include <linux/audit.h>
65 #include <linux/memcontrol.h>
66 #include <linux/ftrace.h>
67 #include <linux/proc_fs.h>
68 #include <linux/profile.h>
69 #include <linux/rmap.h>
70 #include <linux/ksm.h>
71 #include <linux/acct.h>
72 #include <linux/userfaultfd_k.h>
73 #include <linux/tsacct_kern.h>
74 #include <linux/cn_proc.h>
75 #include <linux/freezer.h>
76 #include <linux/delayacct.h>
77 #include <linux/taskstats_kern.h>
78 #include <linux/random.h>
79 #include <linux/tty.h>
80 #include <linux/blkdev.h>
81 #include <linux/fs_struct.h>
82 #include <linux/magic.h>
83 #include <linux/perf_event.h>
84 #include <linux/posix-timers.h>
85 #include <linux/user-return-notifier.h>
86 #include <linux/oom.h>
87 #include <linux/khugepaged.h>
88 #include <linux/signalfd.h>
89 #include <linux/uprobes.h>
90 #include <linux/aio.h>
91 #include <linux/compiler.h>
92 #include <linux/sysctl.h>
93 #include <linux/kcov.h>
94 #include <linux/livepatch.h>
95 #include <linux/thread_info.h>
96 #include <linux/stackleak.h>
98 #include <asm/pgtable.h>
99 #include <asm/pgalloc.h>
100 #include <linux/uaccess.h>
101 #include <asm/mmu_context.h>
102 #include <asm/cacheflush.h>
103 #include <asm/tlbflush.h>
105 #include <trace/events/sched.h>
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/task.h>
109 #ifdef CONFIG_USER_NS
110 extern int unprivileged_userns_clone
;
112 #define unprivileged_userns_clone 0
116 * Minimum number of threads to boot the kernel
118 #define MIN_THREADS 20
121 * Maximum number of threads
123 #define MAX_THREADS FUTEX_TID_MASK
126 * Protected counters by write_lock_irq(&tasklist_lock)
128 unsigned long total_forks
; /* Handle normal Linux uptimes. */
129 int nr_threads
; /* The idle threads do not count.. */
131 static int max_threads
; /* tunable limit on nr_threads */
133 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
135 static const char * const resident_page_types
[] = {
136 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
137 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
138 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
139 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
142 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
144 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
146 #ifdef CONFIG_PROVE_RCU
147 int lockdep_tasklist_lock_is_held(void)
149 return lockdep_is_held(&tasklist_lock
);
151 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
152 #endif /* #ifdef CONFIG_PROVE_RCU */
154 int nr_processes(void)
159 for_each_possible_cpu(cpu
)
160 total
+= per_cpu(process_counts
, cpu
);
165 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
169 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
170 static struct kmem_cache
*task_struct_cachep
;
172 static inline struct task_struct
*alloc_task_struct_node(int node
)
174 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
177 static inline void free_task_struct(struct task_struct
*tsk
)
179 kmem_cache_free(task_struct_cachep
, tsk
);
183 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
186 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
187 * kmemcache based allocator.
189 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
191 #ifdef CONFIG_VMAP_STACK
193 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
194 * flush. Try to minimize the number of calls by caching stacks.
196 #define NR_CACHED_STACKS 2
197 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
199 static int free_vm_stack_cache(unsigned int cpu
)
201 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
204 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
205 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
210 vfree(vm_stack
->addr
);
211 cached_vm_stacks
[i
] = NULL
;
218 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
220 #ifdef CONFIG_VMAP_STACK
224 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
227 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
232 /* Clear stale pointers from reused stack. */
233 memset(s
->addr
, 0, THREAD_SIZE
);
235 tsk
->stack_vm_area
= s
;
236 tsk
->stack
= s
->addr
;
241 * Allocated stacks are cached and later reused by new threads,
242 * so memcg accounting is performed manually on assigning/releasing
243 * stacks to tasks. Drop __GFP_ACCOUNT.
245 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
246 VMALLOC_START
, VMALLOC_END
,
247 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
249 0, node
, __builtin_return_address(0));
252 * We can't call find_vm_area() in interrupt context, and
253 * free_thread_stack() can be called in interrupt context,
254 * so cache the vm_struct.
257 tsk
->stack_vm_area
= find_vm_area(stack
);
262 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
266 tsk
->stack
= page_address(page
);
273 static inline void free_thread_stack(struct task_struct
*tsk
)
275 #ifdef CONFIG_VMAP_STACK
276 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
281 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
282 mod_memcg_page_state(vm
->pages
[i
],
283 MEMCG_KERNEL_STACK_KB
,
284 -(int)(PAGE_SIZE
/ 1024));
286 memcg_kmem_uncharge(vm
->pages
[i
], 0);
289 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
290 if (this_cpu_cmpxchg(cached_stacks
[i
],
291 NULL
, tsk
->stack_vm_area
) != NULL
)
297 vfree_atomic(tsk
->stack
);
302 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
305 static struct kmem_cache
*thread_stack_cache
;
307 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
310 unsigned long *stack
;
311 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
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
);
365 INIT_LIST_HEAD(&new->anon_vma_chain
);
370 void vm_area_free(struct vm_area_struct
*vma
)
372 kmem_cache_free(vm_area_cachep
, vma
);
375 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
377 void *stack
= task_stack_page(tsk
);
378 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
380 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
385 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
387 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
388 mod_zone_page_state(page_zone(vm
->pages
[i
]),
390 PAGE_SIZE
/ 1024 * account
);
394 * All stack pages are in the same zone and belong to the
397 struct page
*first_page
= virt_to_page(stack
);
399 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
400 THREAD_SIZE
/ 1024 * account
);
402 mod_memcg_obj_state(stack
, MEMCG_KERNEL_STACK_KB
,
403 account
* (THREAD_SIZE
/ 1024));
407 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
409 #ifdef CONFIG_VMAP_STACK
410 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
416 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
418 * If memcg_kmem_charge() fails, page->mem_cgroup
419 * pointer is NULL, and both memcg_kmem_uncharge()
420 * and mod_memcg_page_state() in free_thread_stack()
421 * will ignore this page. So it's safe.
423 ret
= memcg_kmem_charge(vm
->pages
[i
], GFP_KERNEL
, 0);
427 mod_memcg_page_state(vm
->pages
[i
],
428 MEMCG_KERNEL_STACK_KB
,
436 static void release_task_stack(struct task_struct
*tsk
)
438 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
439 return; /* Better to leak the stack than to free prematurely */
441 account_kernel_stack(tsk
, -1);
442 free_thread_stack(tsk
);
444 #ifdef CONFIG_VMAP_STACK
445 tsk
->stack_vm_area
= NULL
;
449 #ifdef CONFIG_THREAD_INFO_IN_TASK
450 void put_task_stack(struct task_struct
*tsk
)
452 if (refcount_dec_and_test(&tsk
->stack_refcount
))
453 release_task_stack(tsk
);
457 void free_task(struct task_struct
*tsk
)
459 #ifndef CONFIG_THREAD_INFO_IN_TASK
461 * The task is finally done with both the stack and thread_info,
464 release_task_stack(tsk
);
467 * If the task had a separate stack allocation, it should be gone
470 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
472 rt_mutex_debug_task_free(tsk
);
473 ftrace_graph_exit_task(tsk
);
474 put_seccomp_filter(tsk
);
475 arch_release_task_struct(tsk
);
476 if (tsk
->flags
& PF_KTHREAD
)
477 free_kthread_struct(tsk
);
478 free_task_struct(tsk
);
480 EXPORT_SYMBOL(free_task
);
483 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
484 struct mm_struct
*oldmm
)
486 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
487 struct rb_node
**rb_link
, *rb_parent
;
489 unsigned long charge
;
492 uprobe_start_dup_mmap();
493 if (down_write_killable(&oldmm
->mmap_sem
)) {
495 goto fail_uprobe_end
;
497 flush_cache_dup_mm(oldmm
);
498 uprobe_dup_mmap(oldmm
, mm
);
500 * Not linked in yet - no deadlock potential:
502 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
504 /* No ordering required: file already has been exposed. */
505 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
507 mm
->total_vm
= oldmm
->total_vm
;
508 mm
->data_vm
= oldmm
->data_vm
;
509 mm
->exec_vm
= oldmm
->exec_vm
;
510 mm
->stack_vm
= oldmm
->stack_vm
;
512 rb_link
= &mm
->mm_rb
.rb_node
;
515 retval
= ksm_fork(mm
, oldmm
);
518 retval
= khugepaged_fork(mm
, oldmm
);
523 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
526 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
527 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
532 * Don't duplicate many vmas if we've been oom-killed (for
535 if (fatal_signal_pending(current
)) {
539 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
540 unsigned long len
= vma_pages(mpnt
);
542 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
546 tmp
= vm_area_dup(mpnt
);
549 retval
= vma_dup_policy(mpnt
, tmp
);
551 goto fail_nomem_policy
;
553 retval
= dup_userfaultfd(tmp
, &uf
);
555 goto fail_nomem_anon_vma_fork
;
556 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
557 /* VM_WIPEONFORK gets a clean slate in the child. */
558 tmp
->anon_vma
= NULL
;
559 if (anon_vma_prepare(tmp
))
560 goto fail_nomem_anon_vma_fork
;
561 } else if (anon_vma_fork(tmp
, mpnt
))
562 goto fail_nomem_anon_vma_fork
;
563 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
564 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
567 struct inode
*inode
= file_inode(file
);
568 struct address_space
*mapping
= file
->f_mapping
;
571 if (tmp
->vm_flags
& VM_DENYWRITE
)
572 atomic_dec(&inode
->i_writecount
);
573 i_mmap_lock_write(mapping
);
574 if (tmp
->vm_flags
& VM_SHARED
)
575 atomic_inc(&mapping
->i_mmap_writable
);
576 flush_dcache_mmap_lock(mapping
);
577 /* insert tmp into the share list, just after mpnt */
578 vma_interval_tree_insert_after(tmp
, mpnt
,
580 flush_dcache_mmap_unlock(mapping
);
581 i_mmap_unlock_write(mapping
);
585 * Clear hugetlb-related page reserves for children. This only
586 * affects MAP_PRIVATE mappings. Faults generated by the child
587 * are not guaranteed to succeed, even if read-only
589 if (is_vm_hugetlb_page(tmp
))
590 reset_vma_resv_huge_pages(tmp
);
593 * Link in the new vma and copy the page table entries.
596 pprev
= &tmp
->vm_next
;
600 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
601 rb_link
= &tmp
->vm_rb
.rb_right
;
602 rb_parent
= &tmp
->vm_rb
;
605 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
606 retval
= copy_page_range(mm
, oldmm
, mpnt
);
608 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
609 tmp
->vm_ops
->open(tmp
);
614 /* a new mm has just been created */
615 retval
= arch_dup_mmap(oldmm
, mm
);
617 up_write(&mm
->mmap_sem
);
619 up_write(&oldmm
->mmap_sem
);
620 dup_userfaultfd_complete(&uf
);
622 uprobe_end_dup_mmap();
624 fail_nomem_anon_vma_fork
:
625 mpol_put(vma_policy(tmp
));
630 vm_unacct_memory(charge
);
634 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
636 mm
->pgd
= pgd_alloc(mm
);
637 if (unlikely(!mm
->pgd
))
642 static inline void mm_free_pgd(struct mm_struct
*mm
)
644 pgd_free(mm
, mm
->pgd
);
647 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
649 down_write(&oldmm
->mmap_sem
);
650 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
651 up_write(&oldmm
->mmap_sem
);
654 #define mm_alloc_pgd(mm) (0)
655 #define mm_free_pgd(mm)
656 #endif /* CONFIG_MMU */
658 static void check_mm(struct mm_struct
*mm
)
662 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
663 "Please make sure 'struct resident_page_types[]' is updated as well");
665 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
666 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
669 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
670 mm
, resident_page_types
[i
], x
);
673 if (mm_pgtables_bytes(mm
))
674 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
675 mm_pgtables_bytes(mm
));
677 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
678 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
682 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
683 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
686 * Called when the last reference to the mm
687 * is dropped: either by a lazy thread or by
688 * mmput. Free the page directory and the mm.
690 void __mmdrop(struct mm_struct
*mm
)
692 BUG_ON(mm
== &init_mm
);
693 WARN_ON_ONCE(mm
== current
->mm
);
694 WARN_ON_ONCE(mm
== current
->active_mm
);
697 mmu_notifier_mm_destroy(mm
);
699 put_user_ns(mm
->user_ns
);
702 EXPORT_SYMBOL_GPL(__mmdrop
);
704 static void mmdrop_async_fn(struct work_struct
*work
)
706 struct mm_struct
*mm
;
708 mm
= container_of(work
, struct mm_struct
, async_put_work
);
712 static void mmdrop_async(struct mm_struct
*mm
)
714 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
715 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
716 schedule_work(&mm
->async_put_work
);
720 static inline void free_signal_struct(struct signal_struct
*sig
)
722 taskstats_tgid_free(sig
);
723 sched_autogroup_exit(sig
);
725 * __mmdrop is not safe to call from softirq context on x86 due to
726 * pgd_dtor so postpone it to the async context
729 mmdrop_async(sig
->oom_mm
);
730 kmem_cache_free(signal_cachep
, sig
);
733 static inline void put_signal_struct(struct signal_struct
*sig
)
735 if (refcount_dec_and_test(&sig
->sigcnt
))
736 free_signal_struct(sig
);
739 void __put_task_struct(struct task_struct
*tsk
)
741 WARN_ON(!tsk
->exit_state
);
742 WARN_ON(refcount_read(&tsk
->usage
));
743 WARN_ON(tsk
== current
);
746 task_numa_free(tsk
, true);
747 security_task_free(tsk
);
749 delayacct_tsk_free(tsk
);
750 put_signal_struct(tsk
->signal
);
752 if (!profile_handoff_task(tsk
))
755 EXPORT_SYMBOL_GPL(__put_task_struct
);
757 void __init __weak
arch_task_cache_init(void) { }
762 static void set_max_threads(unsigned int max_threads_suggested
)
765 unsigned long nr_pages
= totalram_pages();
768 * The number of threads shall be limited such that the thread
769 * structures may only consume a small part of the available memory.
771 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
772 threads
= MAX_THREADS
;
774 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
775 (u64
) THREAD_SIZE
* 8UL);
777 if (threads
> max_threads_suggested
)
778 threads
= max_threads_suggested
;
780 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
783 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
784 /* Initialized by the architecture: */
785 int arch_task_struct_size __read_mostly
;
788 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
789 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
791 /* Fetch thread_struct whitelist for the architecture. */
792 arch_thread_struct_whitelist(offset
, size
);
795 * Handle zero-sized whitelist or empty thread_struct, otherwise
796 * adjust offset to position of thread_struct in task_struct.
798 if (unlikely(*size
== 0))
801 *offset
+= offsetof(struct task_struct
, thread
);
803 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
805 void __init
fork_init(void)
808 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
809 #ifndef ARCH_MIN_TASKALIGN
810 #define ARCH_MIN_TASKALIGN 0
812 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
813 unsigned long useroffset
, usersize
;
815 /* create a slab on which task_structs can be allocated */
816 task_struct_whitelist(&useroffset
, &usersize
);
817 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
818 arch_task_struct_size
, align
,
819 SLAB_PANIC
|SLAB_ACCOUNT
,
820 useroffset
, usersize
, NULL
);
823 /* do the arch specific task caches init */
824 arch_task_cache_init();
826 set_max_threads(MAX_THREADS
);
828 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
829 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
830 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
831 init_task
.signal
->rlim
[RLIMIT_NPROC
];
833 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
834 init_user_ns
.ucount_max
[i
] = max_threads
/2;
837 #ifdef CONFIG_VMAP_STACK
838 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
839 NULL
, free_vm_stack_cache
);
842 lockdep_init_task(&init_task
);
846 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
847 struct task_struct
*src
)
853 void set_task_stack_end_magic(struct task_struct
*tsk
)
855 unsigned long *stackend
;
857 stackend
= end_of_stack(tsk
);
858 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
861 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
863 struct task_struct
*tsk
;
864 unsigned long *stack
;
865 struct vm_struct
*stack_vm_area __maybe_unused
;
868 if (node
== NUMA_NO_NODE
)
869 node
= tsk_fork_get_node(orig
);
870 tsk
= alloc_task_struct_node(node
);
874 stack
= alloc_thread_stack_node(tsk
, node
);
878 if (memcg_charge_kernel_stack(tsk
))
881 stack_vm_area
= task_stack_vm_area(tsk
);
883 err
= arch_dup_task_struct(tsk
, orig
);
886 * arch_dup_task_struct() clobbers the stack-related fields. Make
887 * sure they're properly initialized before using any stack-related
891 #ifdef CONFIG_VMAP_STACK
892 tsk
->stack_vm_area
= stack_vm_area
;
894 #ifdef CONFIG_THREAD_INFO_IN_TASK
895 refcount_set(&tsk
->stack_refcount
, 1);
901 #ifdef CONFIG_SECCOMP
903 * We must handle setting up seccomp filters once we're under
904 * the sighand lock in case orig has changed between now and
905 * then. Until then, filter must be NULL to avoid messing up
906 * the usage counts on the error path calling free_task.
908 tsk
->seccomp
.filter
= NULL
;
911 setup_thread_stack(tsk
, orig
);
912 clear_user_return_notifier(tsk
);
913 clear_tsk_need_resched(tsk
);
914 set_task_stack_end_magic(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);
940 #ifdef CONFIG_FAULT_INJECTION
944 #ifdef CONFIG_BLK_CGROUP
945 tsk
->throttle_queue
= NULL
;
946 tsk
->use_memdelay
= 0;
950 tsk
->active_memcg
= NULL
;
955 free_thread_stack(tsk
);
957 free_task_struct(tsk
);
961 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
963 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
965 static int __init
coredump_filter_setup(char *s
)
967 default_dump_filter
=
968 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
969 MMF_DUMP_FILTER_MASK
;
973 __setup("coredump_filter=", coredump_filter_setup
);
975 #include <linux/init_task.h>
977 static void mm_init_aio(struct mm_struct
*mm
)
980 spin_lock_init(&mm
->ioctx_lock
);
981 mm
->ioctx_table
= NULL
;
985 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
986 struct task_struct
*p
)
990 WRITE_ONCE(mm
->owner
, NULL
);
994 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
1001 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1003 #ifdef CONFIG_UPROBES
1004 mm
->uprobes_state
.xol_area
= NULL
;
1008 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1009 struct user_namespace
*user_ns
)
1012 mm
->mm_rb
= RB_ROOT
;
1013 mm
->vmacache_seqnum
= 0;
1014 atomic_set(&mm
->mm_users
, 1);
1015 atomic_set(&mm
->mm_count
, 1);
1016 init_rwsem(&mm
->mmap_sem
);
1017 INIT_LIST_HEAD(&mm
->mmlist
);
1018 mm
->core_state
= NULL
;
1019 mm_pgtables_bytes_init(mm
);
1022 atomic64_set(&mm
->pinned_vm
, 0);
1023 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1024 spin_lock_init(&mm
->page_table_lock
);
1025 spin_lock_init(&mm
->arg_lock
);
1026 mm_init_cpumask(mm
);
1028 mm_init_owner(mm
, p
);
1029 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1030 mmu_notifier_mm_init(mm
);
1031 init_tlb_flush_pending(mm
);
1032 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1033 mm
->pmd_huge_pte
= NULL
;
1035 mm_init_uprobes_state(mm
);
1036 hugetlb_count_init(mm
);
1039 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1040 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1042 mm
->flags
= default_dump_filter
;
1046 if (mm_alloc_pgd(mm
))
1049 if (init_new_context(p
, mm
))
1050 goto fail_nocontext
;
1052 mm
->user_ns
= get_user_ns(user_ns
);
1063 * Allocate and initialize an mm_struct.
1065 struct mm_struct
*mm_alloc(void)
1067 struct mm_struct
*mm
;
1073 memset(mm
, 0, sizeof(*mm
));
1074 return mm_init(mm
, current
, current_user_ns());
1077 static inline void __mmput(struct mm_struct
*mm
)
1079 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1081 uprobe_clear_state(mm
);
1084 khugepaged_exit(mm
); /* must run before exit_mmap */
1086 mm_put_huge_zero_page(mm
);
1087 set_mm_exe_file(mm
, NULL
);
1088 if (!list_empty(&mm
->mmlist
)) {
1089 spin_lock(&mmlist_lock
);
1090 list_del(&mm
->mmlist
);
1091 spin_unlock(&mmlist_lock
);
1094 module_put(mm
->binfmt
->module
);
1099 * Decrement the use count and release all resources for an mm.
1101 void mmput(struct mm_struct
*mm
)
1105 if (atomic_dec_and_test(&mm
->mm_users
))
1108 EXPORT_SYMBOL_GPL(mmput
);
1111 static void mmput_async_fn(struct work_struct
*work
)
1113 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1119 void mmput_async(struct mm_struct
*mm
)
1121 if (atomic_dec_and_test(&mm
->mm_users
)) {
1122 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1123 schedule_work(&mm
->async_put_work
);
1126 EXPORT_SYMBOL(mmput_async
);
1130 * set_mm_exe_file - change a reference to the mm's executable file
1132 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1134 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1135 * invocations: in mmput() nobody alive left, in execve task is single
1136 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1137 * mm->exe_file, but does so without using set_mm_exe_file() in order
1138 * to do avoid the need for any locks.
1140 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1142 struct file
*old_exe_file
;
1145 * It is safe to dereference the exe_file without RCU as
1146 * this function is only called if nobody else can access
1147 * this mm -- see comment above for justification.
1149 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1152 get_file(new_exe_file
);
1153 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1159 * get_mm_exe_file - acquire a reference to the mm's executable file
1161 * Returns %NULL if mm has no associated executable file.
1162 * User must release file via fput().
1164 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1166 struct file
*exe_file
;
1169 exe_file
= rcu_dereference(mm
->exe_file
);
1170 if (exe_file
&& !get_file_rcu(exe_file
))
1175 EXPORT_SYMBOL(get_mm_exe_file
);
1178 * get_task_exe_file - acquire a reference to the task's executable file
1180 * Returns %NULL if task's mm (if any) has no associated executable file or
1181 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1182 * User must release file via fput().
1184 struct file
*get_task_exe_file(struct task_struct
*task
)
1186 struct file
*exe_file
= NULL
;
1187 struct mm_struct
*mm
;
1192 if (!(task
->flags
& PF_KTHREAD
))
1193 exe_file
= get_mm_exe_file(mm
);
1198 EXPORT_SYMBOL(get_task_exe_file
);
1201 * get_task_mm - acquire a reference to the task's mm
1203 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1204 * this kernel workthread has transiently adopted a user mm with use_mm,
1205 * to do its AIO) is not set and if so returns a reference to it, after
1206 * bumping up the use count. User must release the mm via mmput()
1207 * after use. Typically used by /proc and ptrace.
1209 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1211 struct mm_struct
*mm
;
1216 if (task
->flags
& PF_KTHREAD
)
1224 EXPORT_SYMBOL_GPL(get_task_mm
);
1226 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1228 struct mm_struct
*mm
;
1231 err
= down_read_killable(&task
->signal
->exec_update_lock
);
1233 return ERR_PTR(err
);
1235 mm
= get_task_mm(task
);
1236 if (mm
&& mm
!= current
->mm
&&
1237 !ptrace_may_access(task
, mode
)) {
1239 mm
= ERR_PTR(-EACCES
);
1241 up_read(&task
->signal
->exec_update_lock
);
1246 static void complete_vfork_done(struct task_struct
*tsk
)
1248 struct completion
*vfork
;
1251 vfork
= tsk
->vfork_done
;
1252 if (likely(vfork
)) {
1253 tsk
->vfork_done
= NULL
;
1259 static int wait_for_vfork_done(struct task_struct
*child
,
1260 struct completion
*vfork
)
1264 freezer_do_not_count();
1265 cgroup_enter_frozen();
1266 killed
= wait_for_completion_killable(vfork
);
1267 cgroup_leave_frozen(false);
1272 child
->vfork_done
= NULL
;
1276 put_task_struct(child
);
1280 /* Please note the differences between mmput and mm_release.
1281 * mmput is called whenever we stop holding onto a mm_struct,
1282 * error success whatever.
1284 * mm_release is called after a mm_struct has been removed
1285 * from the current process.
1287 * This difference is important for error handling, when we
1288 * only half set up a mm_struct for a new process and need to restore
1289 * the old one. Because we mmput the new mm_struct before
1290 * restoring the old one. . .
1291 * Eric Biederman 10 January 1998
1293 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1295 uprobe_free_utask(tsk
);
1297 /* Get rid of any cached register state */
1298 deactivate_mm(tsk
, mm
);
1301 * Signal userspace if we're not exiting with a core dump
1302 * because we want to leave the value intact for debugging
1305 if (tsk
->clear_child_tid
) {
1306 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1307 atomic_read(&mm
->mm_users
) > 1) {
1309 * We don't check the error code - if userspace has
1310 * not set up a proper pointer then tough luck.
1312 put_user(0, tsk
->clear_child_tid
);
1313 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1314 1, NULL
, NULL
, 0, 0);
1316 tsk
->clear_child_tid
= NULL
;
1320 * All done, finally we can wake up parent and return this mm to him.
1321 * Also kthread_stop() uses this completion for synchronization.
1323 if (tsk
->vfork_done
)
1324 complete_vfork_done(tsk
);
1327 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1329 futex_exit_release(tsk
);
1330 mm_release(tsk
, mm
);
1333 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1335 futex_exec_release(tsk
);
1336 mm_release(tsk
, mm
);
1340 * dup_mm() - duplicates an existing mm structure
1341 * @tsk: the task_struct with which the new mm will be associated.
1342 * @oldmm: the mm to duplicate.
1344 * Allocates a new mm structure and duplicates the provided @oldmm structure
1347 * Return: the duplicated mm or NULL on failure.
1349 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1350 struct mm_struct
*oldmm
)
1352 struct mm_struct
*mm
;
1359 memcpy(mm
, oldmm
, sizeof(*mm
));
1361 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1364 err
= dup_mmap(mm
, oldmm
);
1368 mm
->hiwater_rss
= get_mm_rss(mm
);
1369 mm
->hiwater_vm
= mm
->total_vm
;
1371 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1377 /* don't put binfmt in mmput, we haven't got module yet */
1379 mm_init_owner(mm
, NULL
);
1386 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1388 struct mm_struct
*mm
, *oldmm
;
1391 tsk
->min_flt
= tsk
->maj_flt
= 0;
1392 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1393 #ifdef CONFIG_DETECT_HUNG_TASK
1394 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1395 tsk
->last_switch_time
= 0;
1399 tsk
->active_mm
= NULL
;
1402 * Are we cloning a kernel thread?
1404 * We need to steal a active VM for that..
1406 oldmm
= current
->mm
;
1410 /* initialize the new vmacache entries */
1411 vmacache_flush(tsk
);
1413 if (clone_flags
& CLONE_VM
) {
1420 mm
= dup_mm(tsk
, current
->mm
);
1426 tsk
->active_mm
= mm
;
1433 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1435 struct fs_struct
*fs
= current
->fs
;
1436 if (clone_flags
& CLONE_FS
) {
1437 /* tsk->fs is already what we want */
1438 spin_lock(&fs
->lock
);
1440 spin_unlock(&fs
->lock
);
1444 spin_unlock(&fs
->lock
);
1447 tsk
->fs
= copy_fs_struct(fs
);
1453 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1455 struct files_struct
*oldf
, *newf
;
1459 * A background process may not have any files ...
1461 oldf
= current
->files
;
1465 if (clone_flags
& CLONE_FILES
) {
1466 atomic_inc(&oldf
->count
);
1470 newf
= dup_fd(oldf
, &error
);
1480 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1483 struct io_context
*ioc
= current
->io_context
;
1484 struct io_context
*new_ioc
;
1489 * Share io context with parent, if CLONE_IO is set
1491 if (clone_flags
& CLONE_IO
) {
1493 tsk
->io_context
= ioc
;
1494 } else if (ioprio_valid(ioc
->ioprio
)) {
1495 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1496 if (unlikely(!new_ioc
))
1499 new_ioc
->ioprio
= ioc
->ioprio
;
1500 put_io_context(new_ioc
);
1506 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1508 struct sighand_struct
*sig
;
1510 if (clone_flags
& CLONE_SIGHAND
) {
1511 refcount_inc(¤t
->sighand
->count
);
1514 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1515 rcu_assign_pointer(tsk
->sighand
, sig
);
1519 refcount_set(&sig
->count
, 1);
1520 spin_lock_irq(¤t
->sighand
->siglock
);
1521 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1522 spin_unlock_irq(¤t
->sighand
->siglock
);
1526 void __cleanup_sighand(struct sighand_struct
*sighand
)
1528 if (refcount_dec_and_test(&sighand
->count
)) {
1529 signalfd_cleanup(sighand
);
1531 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1532 * without an RCU grace period, see __lock_task_sighand().
1534 kmem_cache_free(sighand_cachep
, sighand
);
1539 * Initialize POSIX timer handling for a thread group.
1541 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1543 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1544 unsigned long cpu_limit
;
1546 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1547 posix_cputimers_group_init(pct
, cpu_limit
);
1550 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1552 struct signal_struct
*sig
;
1554 if (clone_flags
& CLONE_THREAD
)
1557 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1562 sig
->nr_threads
= 1;
1563 atomic_set(&sig
->live
, 1);
1564 refcount_set(&sig
->sigcnt
, 1);
1566 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1567 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1568 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1570 init_waitqueue_head(&sig
->wait_chldexit
);
1571 sig
->curr_target
= tsk
;
1572 init_sigpending(&sig
->shared_pending
);
1573 INIT_HLIST_HEAD(&sig
->multiprocess
);
1574 seqlock_init(&sig
->stats_lock
);
1575 prev_cputime_init(&sig
->prev_cputime
);
1577 #ifdef CONFIG_POSIX_TIMERS
1578 INIT_LIST_HEAD(&sig
->posix_timers
);
1579 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1580 sig
->real_timer
.function
= it_real_fn
;
1583 task_lock(current
->group_leader
);
1584 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1585 task_unlock(current
->group_leader
);
1587 posix_cpu_timers_init_group(sig
);
1589 tty_audit_fork(sig
);
1590 sched_autogroup_fork(sig
);
1592 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1593 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1595 mutex_init(&sig
->cred_guard_mutex
);
1596 init_rwsem(&sig
->exec_update_lock
);
1601 static void copy_seccomp(struct task_struct
*p
)
1603 #ifdef CONFIG_SECCOMP
1605 * Must be called with sighand->lock held, which is common to
1606 * all threads in the group. Holding cred_guard_mutex is not
1607 * needed because this new task is not yet running and cannot
1610 assert_spin_locked(¤t
->sighand
->siglock
);
1612 /* Ref-count the new filter user, and assign it. */
1613 get_seccomp_filter(current
);
1614 p
->seccomp
= current
->seccomp
;
1617 * Explicitly enable no_new_privs here in case it got set
1618 * between the task_struct being duplicated and holding the
1619 * sighand lock. The seccomp state and nnp must be in sync.
1621 if (task_no_new_privs(current
))
1622 task_set_no_new_privs(p
);
1625 * If the parent gained a seccomp mode after copying thread
1626 * flags and between before we held the sighand lock, we have
1627 * to manually enable the seccomp thread flag here.
1629 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1630 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1634 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1636 current
->clear_child_tid
= tidptr
;
1638 return task_pid_vnr(current
);
1641 static void rt_mutex_init_task(struct task_struct
*p
)
1643 raw_spin_lock_init(&p
->pi_lock
);
1644 #ifdef CONFIG_RT_MUTEXES
1645 p
->pi_waiters
= RB_ROOT_CACHED
;
1646 p
->pi_top_task
= NULL
;
1647 p
->pi_blocked_on
= NULL
;
1651 static inline void init_task_pid_links(struct task_struct
*task
)
1655 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1656 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1661 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1663 if (type
== PIDTYPE_PID
)
1664 task
->thread_pid
= pid
;
1666 task
->signal
->pids
[type
] = pid
;
1669 static inline void rcu_copy_process(struct task_struct
*p
)
1671 #ifdef CONFIG_PREEMPT_RCU
1672 p
->rcu_read_lock_nesting
= 0;
1673 p
->rcu_read_unlock_special
.s
= 0;
1674 p
->rcu_blocked_node
= NULL
;
1675 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1676 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1677 #ifdef CONFIG_TASKS_RCU
1678 p
->rcu_tasks_holdout
= false;
1679 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1680 p
->rcu_tasks_idle_cpu
= -1;
1681 #endif /* #ifdef CONFIG_TASKS_RCU */
1684 struct pid
*pidfd_pid(const struct file
*file
)
1686 if (file
->f_op
== &pidfd_fops
)
1687 return file
->private_data
;
1689 return ERR_PTR(-EBADF
);
1692 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1694 struct pid
*pid
= file
->private_data
;
1696 file
->private_data
= NULL
;
1701 #ifdef CONFIG_PROC_FS
1702 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1704 struct pid_namespace
*ns
= proc_pid_ns(file_inode(m
->file
));
1705 struct pid
*pid
= f
->private_data
;
1707 seq_put_decimal_ull(m
, "Pid:\t", pid_nr_ns(pid
, ns
));
1713 * Poll support for process exit notification.
1715 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1717 struct task_struct
*task
;
1718 struct pid
*pid
= file
->private_data
;
1719 __poll_t poll_flags
= 0;
1721 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1724 task
= pid_task(pid
, PIDTYPE_PID
);
1726 * Inform pollers only when the whole thread group exits.
1727 * If the thread group leader exits before all other threads in the
1728 * group, then poll(2) should block, similar to the wait(2) family.
1730 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1731 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1737 const struct file_operations pidfd_fops
= {
1738 .release
= pidfd_release
,
1740 #ifdef CONFIG_PROC_FS
1741 .show_fdinfo
= pidfd_show_fdinfo
,
1745 static void __delayed_free_task(struct rcu_head
*rhp
)
1747 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1752 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1754 if (IS_ENABLED(CONFIG_MEMCG
))
1755 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1760 static void copy_oom_score_adj(u64 clone_flags
, struct task_struct
*tsk
)
1762 /* Skip if kernel thread */
1766 /* Skip if spawning a thread or using vfork */
1767 if ((clone_flags
& (CLONE_VM
| CLONE_THREAD
| CLONE_VFORK
)) != CLONE_VM
)
1770 /* We need to synchronize with __set_oom_adj */
1771 mutex_lock(&oom_adj_mutex
);
1772 set_bit(MMF_MULTIPROCESS
, &tsk
->mm
->flags
);
1773 /* Update the values in case they were changed after copy_signal */
1774 tsk
->signal
->oom_score_adj
= current
->signal
->oom_score_adj
;
1775 tsk
->signal
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1776 mutex_unlock(&oom_adj_mutex
);
1780 * This creates a new process as a copy of the old one,
1781 * but does not actually start it yet.
1783 * It copies the registers, and all the appropriate
1784 * parts of the process environment (as per the clone
1785 * flags). The actual kick-off is left to the caller.
1787 static __latent_entropy
struct task_struct
*copy_process(
1791 struct kernel_clone_args
*args
)
1793 int pidfd
= -1, retval
;
1794 struct task_struct
*p
;
1795 struct multiprocess_signals delayed
;
1796 struct file
*pidfile
= NULL
;
1797 u64 clone_flags
= args
->flags
;
1800 * Don't allow sharing the root directory with processes in a different
1803 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1804 return ERR_PTR(-EINVAL
);
1806 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1807 return ERR_PTR(-EINVAL
);
1809 if ((clone_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
)
1810 if (!capable(CAP_SYS_ADMIN
))
1811 return ERR_PTR(-EPERM
);
1814 * Thread groups must share signals as well, and detached threads
1815 * can only be started up within the thread group.
1817 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1818 return ERR_PTR(-EINVAL
);
1821 * Shared signal handlers imply shared VM. By way of the above,
1822 * thread groups also imply shared VM. Blocking this case allows
1823 * for various simplifications in other code.
1825 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1826 return ERR_PTR(-EINVAL
);
1829 * Siblings of global init remain as zombies on exit since they are
1830 * not reaped by their parent (swapper). To solve this and to avoid
1831 * multi-rooted process trees, prevent global and container-inits
1832 * from creating siblings.
1834 if ((clone_flags
& CLONE_PARENT
) &&
1835 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1836 return ERR_PTR(-EINVAL
);
1839 * If the new process will be in a different pid or user namespace
1840 * do not allow it to share a thread group with the forking task.
1842 if (clone_flags
& CLONE_THREAD
) {
1843 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1844 (task_active_pid_ns(current
) !=
1845 current
->nsproxy
->pid_ns_for_children
))
1846 return ERR_PTR(-EINVAL
);
1849 if (clone_flags
& CLONE_PIDFD
) {
1851 * - CLONE_DETACHED is blocked so that we can potentially
1852 * reuse it later for CLONE_PIDFD.
1853 * - CLONE_THREAD is blocked until someone really needs it.
1855 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1856 return ERR_PTR(-EINVAL
);
1860 * Force any signals received before this point to be delivered
1861 * before the fork happens. Collect up signals sent to multiple
1862 * processes that happen during the fork and delay them so that
1863 * they appear to happen after the fork.
1865 sigemptyset(&delayed
.signal
);
1866 INIT_HLIST_NODE(&delayed
.node
);
1868 spin_lock_irq(¤t
->sighand
->siglock
);
1869 if (!(clone_flags
& CLONE_THREAD
))
1870 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1871 recalc_sigpending();
1872 spin_unlock_irq(¤t
->sighand
->siglock
);
1873 retval
= -ERESTARTNOINTR
;
1874 if (signal_pending(current
))
1878 p
= dup_task_struct(current
, node
);
1883 * This _must_ happen before we call free_task(), i.e. before we jump
1884 * to any of the bad_fork_* labels. This is to avoid freeing
1885 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1886 * kernel threads (PF_KTHREAD).
1888 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1890 * Clear TID on mm_release()?
1892 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1894 ftrace_graph_init_task(p
);
1896 rt_mutex_init_task(p
);
1898 #ifdef CONFIG_PROVE_LOCKING
1899 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1900 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1903 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1904 task_rlimit(p
, RLIMIT_NPROC
)) {
1905 if (p
->real_cred
->user
!= INIT_USER
&&
1906 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1909 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1911 retval
= copy_creds(p
, clone_flags
);
1916 * If multiple threads are within copy_process(), then this check
1917 * triggers too late. This doesn't hurt, the check is only there
1918 * to stop root fork bombs.
1921 if (nr_threads
>= max_threads
)
1922 goto bad_fork_cleanup_count
;
1924 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1925 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1926 p
->flags
|= PF_FORKNOEXEC
;
1927 INIT_LIST_HEAD(&p
->children
);
1928 INIT_LIST_HEAD(&p
->sibling
);
1929 rcu_copy_process(p
);
1930 p
->vfork_done
= NULL
;
1931 spin_lock_init(&p
->alloc_lock
);
1933 init_sigpending(&p
->pending
);
1935 p
->utime
= p
->stime
= p
->gtime
= 0;
1936 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1937 p
->utimescaled
= p
->stimescaled
= 0;
1939 prev_cputime_init(&p
->prev_cputime
);
1941 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1942 seqcount_init(&p
->vtime
.seqcount
);
1943 p
->vtime
.starttime
= 0;
1944 p
->vtime
.state
= VTIME_INACTIVE
;
1947 #if defined(SPLIT_RSS_COUNTING)
1948 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1951 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1957 task_io_accounting_init(&p
->ioac
);
1958 acct_clear_integrals(p
);
1960 posix_cputimers_init(&p
->posix_cputimers
);
1962 p
->io_context
= NULL
;
1963 audit_set_context(p
, NULL
);
1966 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1967 if (IS_ERR(p
->mempolicy
)) {
1968 retval
= PTR_ERR(p
->mempolicy
);
1969 p
->mempolicy
= NULL
;
1970 goto bad_fork_cleanup_threadgroup_lock
;
1973 #ifdef CONFIG_CPUSETS
1974 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1975 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1976 seqcount_init(&p
->mems_allowed_seq
);
1978 #ifdef CONFIG_TRACE_IRQFLAGS
1980 p
->hardirqs_enabled
= 0;
1981 p
->hardirq_enable_ip
= 0;
1982 p
->hardirq_enable_event
= 0;
1983 p
->hardirq_disable_ip
= _THIS_IP_
;
1984 p
->hardirq_disable_event
= 0;
1985 p
->softirqs_enabled
= 1;
1986 p
->softirq_enable_ip
= _THIS_IP_
;
1987 p
->softirq_enable_event
= 0;
1988 p
->softirq_disable_ip
= 0;
1989 p
->softirq_disable_event
= 0;
1990 p
->hardirq_context
= 0;
1991 p
->softirq_context
= 0;
1994 p
->pagefault_disabled
= 0;
1996 #ifdef CONFIG_LOCKDEP
1997 lockdep_init_task(p
);
2000 #ifdef CONFIG_DEBUG_MUTEXES
2001 p
->blocked_on
= NULL
; /* not blocked yet */
2003 #ifdef CONFIG_BCACHE
2004 p
->sequential_io
= 0;
2005 p
->sequential_io_avg
= 0;
2008 /* Perform scheduler related setup. Assign this task to a CPU. */
2009 retval
= sched_fork(clone_flags
, p
);
2011 goto bad_fork_cleanup_policy
;
2013 retval
= perf_event_init_task(p
);
2015 goto bad_fork_cleanup_policy
;
2016 retval
= audit_alloc(p
);
2018 goto bad_fork_cleanup_perf
;
2019 /* copy all the process information */
2021 retval
= security_task_alloc(p
, clone_flags
);
2023 goto bad_fork_cleanup_audit
;
2024 retval
= copy_semundo(clone_flags
, p
);
2026 goto bad_fork_cleanup_security
;
2027 retval
= copy_files(clone_flags
, p
);
2029 goto bad_fork_cleanup_semundo
;
2030 retval
= copy_fs(clone_flags
, p
);
2032 goto bad_fork_cleanup_files
;
2033 retval
= copy_sighand(clone_flags
, p
);
2035 goto bad_fork_cleanup_fs
;
2036 retval
= copy_signal(clone_flags
, p
);
2038 goto bad_fork_cleanup_sighand
;
2039 retval
= copy_mm(clone_flags
, p
);
2041 goto bad_fork_cleanup_signal
;
2042 retval
= copy_namespaces(clone_flags
, p
);
2044 goto bad_fork_cleanup_mm
;
2045 retval
= copy_io(clone_flags
, p
);
2047 goto bad_fork_cleanup_namespaces
;
2048 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2051 goto bad_fork_cleanup_io
;
2053 stackleak_task_init(p
);
2055 if (pid
!= &init_struct_pid
) {
2056 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
2058 retval
= PTR_ERR(pid
);
2059 goto bad_fork_cleanup_thread
;
2064 * This has to happen after we've potentially unshared the file
2065 * descriptor table (so that the pidfd doesn't leak into the child
2066 * if the fd table isn't shared).
2068 if (clone_flags
& CLONE_PIDFD
) {
2069 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2071 goto bad_fork_free_pid
;
2075 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2076 O_RDWR
| O_CLOEXEC
);
2077 if (IS_ERR(pidfile
)) {
2078 put_unused_fd(pidfd
);
2079 retval
= PTR_ERR(pidfile
);
2080 goto bad_fork_free_pid
;
2082 get_pid(pid
); /* held by pidfile now */
2084 retval
= put_user(pidfd
, args
->pidfd
);
2086 goto bad_fork_put_pidfd
;
2095 * sigaltstack should be cleared when sharing the same VM
2097 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2101 * Syscall tracing and stepping should be turned off in the
2102 * child regardless of CLONE_PTRACE.
2104 user_disable_single_step(p
);
2105 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2106 #ifdef TIF_SYSCALL_EMU
2107 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2109 clear_tsk_latency_tracing(p
);
2111 /* ok, now we should be set up.. */
2112 p
->pid
= pid_nr(pid
);
2113 if (clone_flags
& CLONE_THREAD
) {
2114 p
->group_leader
= current
->group_leader
;
2115 p
->tgid
= current
->tgid
;
2117 p
->group_leader
= p
;
2122 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2123 p
->dirty_paused_when
= 0;
2125 p
->pdeath_signal
= 0;
2126 INIT_LIST_HEAD(&p
->thread_group
);
2127 p
->task_works
= NULL
;
2129 cgroup_threadgroup_change_begin(current
);
2131 * Ensure that the cgroup subsystem policies allow the new process to be
2132 * forked. It should be noted the the new process's css_set can be changed
2133 * between here and cgroup_post_fork() if an organisation operation is in
2136 retval
= cgroup_can_fork(p
);
2138 goto bad_fork_cgroup_threadgroup_change_end
;
2141 * From this point on we must avoid any synchronous user-space
2142 * communication until we take the tasklist-lock. In particular, we do
2143 * not want user-space to be able to predict the process start-time by
2144 * stalling fork(2) after we recorded the start_time but before it is
2145 * visible to the system.
2148 p
->start_time
= ktime_get_ns();
2149 p
->real_start_time
= ktime_get_boottime_ns();
2152 * Make it visible to the rest of the system, but dont wake it up yet.
2153 * Need tasklist lock for parent etc handling!
2155 write_lock_irq(&tasklist_lock
);
2157 /* CLONE_PARENT re-uses the old parent */
2158 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2159 p
->real_parent
= current
->real_parent
;
2160 p
->parent_exec_id
= current
->parent_exec_id
;
2161 if (clone_flags
& CLONE_THREAD
)
2162 p
->exit_signal
= -1;
2164 p
->exit_signal
= current
->group_leader
->exit_signal
;
2166 p
->real_parent
= current
;
2167 p
->parent_exec_id
= current
->self_exec_id
;
2168 p
->exit_signal
= args
->exit_signal
;
2171 klp_copy_process(p
);
2173 spin_lock(¤t
->sighand
->siglock
);
2176 * Copy seccomp details explicitly here, in case they were changed
2177 * before holding sighand lock.
2181 rseq_fork(p
, clone_flags
);
2183 /* Don't start children in a dying pid namespace */
2184 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2186 goto bad_fork_cancel_cgroup
;
2189 /* Let kill terminate clone/fork in the middle */
2190 if (fatal_signal_pending(current
)) {
2192 goto bad_fork_cancel_cgroup
;
2195 init_task_pid_links(p
);
2196 if (likely(p
->pid
)) {
2197 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2199 init_task_pid(p
, PIDTYPE_PID
, pid
);
2200 if (thread_group_leader(p
)) {
2201 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2202 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2203 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2205 if (is_child_reaper(pid
)) {
2206 ns_of_pid(pid
)->child_reaper
= p
;
2207 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2209 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2210 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2212 * Inherit has_child_subreaper flag under the same
2213 * tasklist_lock with adding child to the process tree
2214 * for propagate_has_child_subreaper optimization.
2216 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2217 p
->real_parent
->signal
->is_child_subreaper
;
2218 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2219 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2220 attach_pid(p
, PIDTYPE_TGID
);
2221 attach_pid(p
, PIDTYPE_PGID
);
2222 attach_pid(p
, PIDTYPE_SID
);
2223 __this_cpu_inc(process_counts
);
2225 current
->signal
->nr_threads
++;
2226 atomic_inc(¤t
->signal
->live
);
2227 refcount_inc(¤t
->signal
->sigcnt
);
2228 task_join_group_stop(p
);
2229 list_add_tail_rcu(&p
->thread_group
,
2230 &p
->group_leader
->thread_group
);
2231 list_add_tail_rcu(&p
->thread_node
,
2232 &p
->signal
->thread_head
);
2234 attach_pid(p
, PIDTYPE_PID
);
2238 hlist_del_init(&delayed
.node
);
2239 spin_unlock(¤t
->sighand
->siglock
);
2240 syscall_tracepoint_update(p
);
2241 write_unlock_irq(&tasklist_lock
);
2244 fd_install(pidfd
, pidfile
);
2246 proc_fork_connector(p
);
2247 cgroup_post_fork(p
);
2248 cgroup_threadgroup_change_end(current
);
2251 trace_task_newtask(p
, clone_flags
);
2252 uprobe_copy_process(p
, clone_flags
);
2254 copy_oom_score_adj(clone_flags
, p
);
2258 bad_fork_cancel_cgroup
:
2259 spin_unlock(¤t
->sighand
->siglock
);
2260 write_unlock_irq(&tasklist_lock
);
2261 cgroup_cancel_fork(p
);
2262 bad_fork_cgroup_threadgroup_change_end
:
2263 cgroup_threadgroup_change_end(current
);
2265 if (clone_flags
& CLONE_PIDFD
) {
2267 put_unused_fd(pidfd
);
2270 if (pid
!= &init_struct_pid
)
2272 bad_fork_cleanup_thread
:
2274 bad_fork_cleanup_io
:
2277 bad_fork_cleanup_namespaces
:
2278 exit_task_namespaces(p
);
2279 bad_fork_cleanup_mm
:
2281 mm_clear_owner(p
->mm
, p
);
2284 bad_fork_cleanup_signal
:
2285 if (!(clone_flags
& CLONE_THREAD
))
2286 free_signal_struct(p
->signal
);
2287 bad_fork_cleanup_sighand
:
2288 __cleanup_sighand(p
->sighand
);
2289 bad_fork_cleanup_fs
:
2290 exit_fs(p
); /* blocking */
2291 bad_fork_cleanup_files
:
2292 exit_files(p
); /* blocking */
2293 bad_fork_cleanup_semundo
:
2295 bad_fork_cleanup_security
:
2296 security_task_free(p
);
2297 bad_fork_cleanup_audit
:
2299 bad_fork_cleanup_perf
:
2300 perf_event_free_task(p
);
2301 bad_fork_cleanup_policy
:
2302 lockdep_free_task(p
);
2304 mpol_put(p
->mempolicy
);
2305 bad_fork_cleanup_threadgroup_lock
:
2307 delayacct_tsk_free(p
);
2308 bad_fork_cleanup_count
:
2309 atomic_dec(&p
->cred
->user
->processes
);
2312 p
->state
= TASK_DEAD
;
2314 delayed_free_task(p
);
2316 spin_lock_irq(¤t
->sighand
->siglock
);
2317 hlist_del_init(&delayed
.node
);
2318 spin_unlock_irq(¤t
->sighand
->siglock
);
2319 return ERR_PTR(retval
);
2322 static inline void init_idle_pids(struct task_struct
*idle
)
2326 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2327 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2328 init_task_pid(idle
, type
, &init_struct_pid
);
2332 struct task_struct
*fork_idle(int cpu
)
2334 struct task_struct
*task
;
2335 struct kernel_clone_args args
= {
2339 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2340 if (!IS_ERR(task
)) {
2341 init_idle_pids(task
);
2342 init_idle(task
, cpu
);
2348 struct mm_struct
*copy_init_mm(void)
2350 return dup_mm(NULL
, &init_mm
);
2354 * Ok, this is the main fork-routine.
2356 * It copies the process, and if successful kick-starts
2357 * it and waits for it to finish using the VM if required.
2359 * args->exit_signal is expected to be checked for sanity by the caller.
2361 long _do_fork(struct kernel_clone_args
*args
)
2363 u64 clone_flags
= args
->flags
;
2364 struct completion vfork
;
2366 struct task_struct
*p
;
2371 * Determine whether and which event to report to ptracer. When
2372 * called from kernel_thread or CLONE_UNTRACED is explicitly
2373 * requested, no event is reported; otherwise, report if the event
2374 * for the type of forking is enabled.
2376 if (!(clone_flags
& CLONE_UNTRACED
)) {
2377 if (clone_flags
& CLONE_VFORK
)
2378 trace
= PTRACE_EVENT_VFORK
;
2379 else if (args
->exit_signal
!= SIGCHLD
)
2380 trace
= PTRACE_EVENT_CLONE
;
2382 trace
= PTRACE_EVENT_FORK
;
2384 if (likely(!ptrace_event_enabled(current
, trace
)))
2388 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2389 add_latent_entropy();
2395 * Do this prior waking up the new thread - the thread pointer
2396 * might get invalid after that point, if the thread exits quickly.
2398 trace_sched_process_fork(current
, p
);
2400 pid
= get_task_pid(p
, PIDTYPE_PID
);
2403 if (clone_flags
& CLONE_PARENT_SETTID
)
2404 put_user(nr
, args
->parent_tid
);
2406 if (clone_flags
& CLONE_VFORK
) {
2407 p
->vfork_done
= &vfork
;
2408 init_completion(&vfork
);
2412 wake_up_new_task(p
);
2414 /* forking complete and child started to run, tell ptracer */
2415 if (unlikely(trace
))
2416 ptrace_event_pid(trace
, pid
);
2418 if (clone_flags
& CLONE_VFORK
) {
2419 if (!wait_for_vfork_done(p
, &vfork
))
2420 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2427 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2429 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2430 if ((kargs
->flags
& CLONE_PIDFD
) &&
2431 (kargs
->flags
& CLONE_PARENT_SETTID
))
2437 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2438 /* For compatibility with architectures that call do_fork directly rather than
2439 * using the syscall entry points below. */
2440 long do_fork(unsigned long clone_flags
,
2441 unsigned long stack_start
,
2442 unsigned long stack_size
,
2443 int __user
*parent_tidptr
,
2444 int __user
*child_tidptr
)
2446 struct kernel_clone_args args
= {
2447 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2448 .pidfd
= parent_tidptr
,
2449 .child_tid
= child_tidptr
,
2450 .parent_tid
= parent_tidptr
,
2451 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2452 .stack
= stack_start
,
2453 .stack_size
= stack_size
,
2456 if (!legacy_clone_args_valid(&args
))
2459 return _do_fork(&args
);
2464 * Create a kernel thread.
2466 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2468 struct kernel_clone_args args
= {
2469 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2470 CLONE_UNTRACED
) & ~CSIGNAL
),
2471 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2472 .stack
= (unsigned long)fn
,
2473 .stack_size
= (unsigned long)arg
,
2476 return _do_fork(&args
);
2479 #ifdef __ARCH_WANT_SYS_FORK
2480 SYSCALL_DEFINE0(fork
)
2483 struct kernel_clone_args args
= {
2484 .exit_signal
= SIGCHLD
,
2487 return _do_fork(&args
);
2489 /* can not support in nommu mode */
2495 #ifdef __ARCH_WANT_SYS_VFORK
2496 SYSCALL_DEFINE0(vfork
)
2498 struct kernel_clone_args args
= {
2499 .flags
= CLONE_VFORK
| CLONE_VM
,
2500 .exit_signal
= SIGCHLD
,
2503 return _do_fork(&args
);
2507 #ifdef __ARCH_WANT_SYS_CLONE
2508 #ifdef CONFIG_CLONE_BACKWARDS
2509 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2510 int __user
*, parent_tidptr
,
2512 int __user
*, child_tidptr
)
2513 #elif defined(CONFIG_CLONE_BACKWARDS2)
2514 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2515 int __user
*, parent_tidptr
,
2516 int __user
*, child_tidptr
,
2518 #elif defined(CONFIG_CLONE_BACKWARDS3)
2519 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2521 int __user
*, parent_tidptr
,
2522 int __user
*, child_tidptr
,
2525 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2526 int __user
*, parent_tidptr
,
2527 int __user
*, child_tidptr
,
2531 struct kernel_clone_args args
= {
2532 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2533 .pidfd
= parent_tidptr
,
2534 .child_tid
= child_tidptr
,
2535 .parent_tid
= parent_tidptr
,
2536 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2541 if (!legacy_clone_args_valid(&args
))
2544 return _do_fork(&args
);
2548 #ifdef __ARCH_WANT_SYS_CLONE3
2551 * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2552 * the registers containing the syscall arguments for clone. This doesn't work
2553 * with clone3 since the TLS value is passed in clone_args instead.
2555 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2556 #error clone3 requires copy_thread_tls support in arch
2559 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2560 struct clone_args __user
*uargs
,
2564 struct clone_args args
;
2566 if (unlikely(usize
> PAGE_SIZE
))
2568 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2571 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2576 * Verify that higher 32bits of exit_signal are unset and that
2577 * it is a valid signal
2579 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2580 !valid_signal(args
.exit_signal
)))
2583 *kargs
= (struct kernel_clone_args
){
2584 .flags
= args
.flags
,
2585 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2586 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2587 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2588 .exit_signal
= args
.exit_signal
,
2589 .stack
= args
.stack
,
2590 .stack_size
= args
.stack_size
,
2598 * clone3_stack_valid - check and prepare stack
2599 * @kargs: kernel clone args
2601 * Verify that the stack arguments userspace gave us are sane.
2602 * In addition, set the stack direction for userspace since it's easy for us to
2605 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2607 if (kargs
->stack
== 0) {
2608 if (kargs
->stack_size
> 0)
2611 if (kargs
->stack_size
== 0)
2614 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2617 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2618 kargs
->stack
+= kargs
->stack_size
;
2625 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2628 * All lower bits of the flag word are taken.
2629 * Verify that no other unknown flags are passed along.
2631 if (kargs
->flags
& ~CLONE_LEGACY_FLAGS
)
2635 * - make the CLONE_DETACHED bit reuseable for clone3
2636 * - make the CSIGNAL bits reuseable for clone3
2638 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2641 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2645 if (!clone3_stack_valid(kargs
))
2652 * clone3 - create a new process with specific properties
2653 * @uargs: argument structure
2654 * @size: size of @uargs
2656 * clone3() is the extensible successor to clone()/clone2().
2657 * It takes a struct as argument that is versioned by its size.
2659 * Return: On success, a positive PID for the child process.
2660 * On error, a negative errno number.
2662 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2666 struct kernel_clone_args kargs
;
2668 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2672 if (!clone3_args_valid(&kargs
))
2675 return _do_fork(&kargs
);
2679 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2681 struct task_struct
*leader
, *parent
, *child
;
2684 read_lock(&tasklist_lock
);
2685 leader
= top
= top
->group_leader
;
2687 for_each_thread(leader
, parent
) {
2688 list_for_each_entry(child
, &parent
->children
, sibling
) {
2689 res
= visitor(child
, data
);
2701 if (leader
!= top
) {
2703 parent
= child
->real_parent
;
2704 leader
= parent
->group_leader
;
2708 read_unlock(&tasklist_lock
);
2711 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2712 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2715 static void sighand_ctor(void *data
)
2717 struct sighand_struct
*sighand
= data
;
2719 spin_lock_init(&sighand
->siglock
);
2720 init_waitqueue_head(&sighand
->signalfd_wqh
);
2723 void __init
proc_caches_init(void)
2725 unsigned int mm_size
;
2727 sighand_cachep
= kmem_cache_create("sighand_cache",
2728 sizeof(struct sighand_struct
), 0,
2729 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2730 SLAB_ACCOUNT
, sighand_ctor
);
2731 signal_cachep
= kmem_cache_create("signal_cache",
2732 sizeof(struct signal_struct
), 0,
2733 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2735 files_cachep
= kmem_cache_create("files_cache",
2736 sizeof(struct files_struct
), 0,
2737 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2739 fs_cachep
= kmem_cache_create("fs_cache",
2740 sizeof(struct fs_struct
), 0,
2741 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2745 * The mm_cpumask is located at the end of mm_struct, and is
2746 * dynamically sized based on the maximum CPU number this system
2747 * can have, taking hotplug into account (nr_cpu_ids).
2749 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2751 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2752 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2753 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2754 offsetof(struct mm_struct
, saved_auxv
),
2755 sizeof_field(struct mm_struct
, saved_auxv
),
2757 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2759 nsproxy_cache_init();
2763 * Check constraints on flags passed to the unshare system call.
2765 static int check_unshare_flags(unsigned long unshare_flags
)
2767 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2768 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2769 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2770 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2773 * Not implemented, but pretend it works if there is nothing
2774 * to unshare. Note that unsharing the address space or the
2775 * signal handlers also need to unshare the signal queues (aka
2778 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2779 if (!thread_group_empty(current
))
2782 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2783 if (refcount_read(¤t
->sighand
->count
) > 1)
2786 if (unshare_flags
& CLONE_VM
) {
2787 if (!current_is_single_threaded())
2795 * Unshare the filesystem structure if it is being shared
2797 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2799 struct fs_struct
*fs
= current
->fs
;
2801 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2804 /* don't need lock here; in the worst case we'll do useless copy */
2808 *new_fsp
= copy_fs_struct(fs
);
2816 * Unshare file descriptor table if it is being shared
2818 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2820 struct files_struct
*fd
= current
->files
;
2823 if ((unshare_flags
& CLONE_FILES
) &&
2824 (fd
&& atomic_read(&fd
->count
) > 1)) {
2825 *new_fdp
= dup_fd(fd
, &error
);
2834 * unshare allows a process to 'unshare' part of the process
2835 * context which was originally shared using clone. copy_*
2836 * functions used by do_fork() cannot be used here directly
2837 * because they modify an inactive task_struct that is being
2838 * constructed. Here we are modifying the current, active,
2841 int ksys_unshare(unsigned long unshare_flags
)
2843 struct fs_struct
*fs
, *new_fs
= NULL
;
2844 struct files_struct
*fd
, *new_fd
= NULL
;
2845 struct cred
*new_cred
= NULL
;
2846 struct nsproxy
*new_nsproxy
= NULL
;
2851 * If unsharing a user namespace must also unshare the thread group
2852 * and unshare the filesystem root and working directories.
2854 if (unshare_flags
& CLONE_NEWUSER
)
2855 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2857 * If unsharing vm, must also unshare signal handlers.
2859 if (unshare_flags
& CLONE_VM
)
2860 unshare_flags
|= CLONE_SIGHAND
;
2862 * If unsharing a signal handlers, must also unshare the signal queues.
2864 if (unshare_flags
& CLONE_SIGHAND
)
2865 unshare_flags
|= CLONE_THREAD
;
2867 * If unsharing namespace, must also unshare filesystem information.
2869 if (unshare_flags
& CLONE_NEWNS
)
2870 unshare_flags
|= CLONE_FS
;
2872 if ((unshare_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
) {
2874 if (!capable(CAP_SYS_ADMIN
))
2875 goto bad_unshare_out
;
2878 err
= check_unshare_flags(unshare_flags
);
2880 goto bad_unshare_out
;
2882 * CLONE_NEWIPC must also detach from the undolist: after switching
2883 * to a new ipc namespace, the semaphore arrays from the old
2884 * namespace are unreachable.
2886 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2888 err
= unshare_fs(unshare_flags
, &new_fs
);
2890 goto bad_unshare_out
;
2891 err
= unshare_fd(unshare_flags
, &new_fd
);
2893 goto bad_unshare_cleanup_fs
;
2894 err
= unshare_userns(unshare_flags
, &new_cred
);
2896 goto bad_unshare_cleanup_fd
;
2897 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2900 goto bad_unshare_cleanup_cred
;
2902 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2905 * CLONE_SYSVSEM is equivalent to sys_exit().
2909 if (unshare_flags
& CLONE_NEWIPC
) {
2910 /* Orphan segments in old ns (see sem above). */
2912 shm_init_task(current
);
2916 switch_task_namespaces(current
, new_nsproxy
);
2922 spin_lock(&fs
->lock
);
2923 current
->fs
= new_fs
;
2928 spin_unlock(&fs
->lock
);
2932 fd
= current
->files
;
2933 current
->files
= new_fd
;
2937 task_unlock(current
);
2940 /* Install the new user namespace */
2941 commit_creds(new_cred
);
2946 perf_event_namespaces(current
);
2948 bad_unshare_cleanup_cred
:
2951 bad_unshare_cleanup_fd
:
2953 put_files_struct(new_fd
);
2955 bad_unshare_cleanup_fs
:
2957 free_fs_struct(new_fs
);
2963 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2965 return ksys_unshare(unshare_flags
);
2969 * Helper to unshare the files of the current task.
2970 * We don't want to expose copy_files internals to
2971 * the exec layer of the kernel.
2974 int unshare_files(struct files_struct
**displaced
)
2976 struct task_struct
*task
= current
;
2977 struct files_struct
*copy
= NULL
;
2980 error
= unshare_fd(CLONE_FILES
, ©
);
2981 if (error
|| !copy
) {
2985 *displaced
= task
->files
;
2992 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2993 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2997 int threads
= max_threads
;
2999 int max
= MAX_THREADS
;
3006 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3010 max_threads
= threads
;