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
);
1038 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1039 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1041 mm
->flags
= default_dump_filter
;
1045 if (mm_alloc_pgd(mm
))
1048 if (init_new_context(p
, mm
))
1049 goto fail_nocontext
;
1051 mm
->user_ns
= get_user_ns(user_ns
);
1062 * Allocate and initialize an mm_struct.
1064 struct mm_struct
*mm_alloc(void)
1066 struct mm_struct
*mm
;
1072 memset(mm
, 0, sizeof(*mm
));
1073 return mm_init(mm
, current
, current_user_ns());
1076 static inline void __mmput(struct mm_struct
*mm
)
1078 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1080 uprobe_clear_state(mm
);
1083 khugepaged_exit(mm
); /* must run before exit_mmap */
1085 mm_put_huge_zero_page(mm
);
1086 set_mm_exe_file(mm
, NULL
);
1087 if (!list_empty(&mm
->mmlist
)) {
1088 spin_lock(&mmlist_lock
);
1089 list_del(&mm
->mmlist
);
1090 spin_unlock(&mmlist_lock
);
1093 module_put(mm
->binfmt
->module
);
1098 * Decrement the use count and release all resources for an mm.
1100 void mmput(struct mm_struct
*mm
)
1104 if (atomic_dec_and_test(&mm
->mm_users
))
1107 EXPORT_SYMBOL_GPL(mmput
);
1110 static void mmput_async_fn(struct work_struct
*work
)
1112 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1118 void mmput_async(struct mm_struct
*mm
)
1120 if (atomic_dec_and_test(&mm
->mm_users
)) {
1121 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1122 schedule_work(&mm
->async_put_work
);
1125 EXPORT_SYMBOL(mmput_async
);
1129 * set_mm_exe_file - change a reference to the mm's executable file
1131 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1133 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1134 * invocations: in mmput() nobody alive left, in execve task is single
1135 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1136 * mm->exe_file, but does so without using set_mm_exe_file() in order
1137 * to do avoid the need for any locks.
1139 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1141 struct file
*old_exe_file
;
1144 * It is safe to dereference the exe_file without RCU as
1145 * this function is only called if nobody else can access
1146 * this mm -- see comment above for justification.
1148 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1151 get_file(new_exe_file
);
1152 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1158 * get_mm_exe_file - acquire a reference to the mm's executable file
1160 * Returns %NULL if mm has no associated executable file.
1161 * User must release file via fput().
1163 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1165 struct file
*exe_file
;
1168 exe_file
= rcu_dereference(mm
->exe_file
);
1169 if (exe_file
&& !get_file_rcu(exe_file
))
1174 EXPORT_SYMBOL(get_mm_exe_file
);
1177 * get_task_exe_file - acquire a reference to the task's executable file
1179 * Returns %NULL if task's mm (if any) has no associated executable file or
1180 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1181 * User must release file via fput().
1183 struct file
*get_task_exe_file(struct task_struct
*task
)
1185 struct file
*exe_file
= NULL
;
1186 struct mm_struct
*mm
;
1191 if (!(task
->flags
& PF_KTHREAD
))
1192 exe_file
= get_mm_exe_file(mm
);
1197 EXPORT_SYMBOL(get_task_exe_file
);
1200 * get_task_mm - acquire a reference to the task's mm
1202 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1203 * this kernel workthread has transiently adopted a user mm with use_mm,
1204 * to do its AIO) is not set and if so returns a reference to it, after
1205 * bumping up the use count. User must release the mm via mmput()
1206 * after use. Typically used by /proc and ptrace.
1208 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1210 struct mm_struct
*mm
;
1215 if (task
->flags
& PF_KTHREAD
)
1223 EXPORT_SYMBOL_GPL(get_task_mm
);
1225 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1227 struct mm_struct
*mm
;
1230 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1232 return ERR_PTR(err
);
1234 mm
= get_task_mm(task
);
1235 if (mm
&& mm
!= current
->mm
&&
1236 !ptrace_may_access(task
, mode
)) {
1238 mm
= ERR_PTR(-EACCES
);
1240 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1245 static void complete_vfork_done(struct task_struct
*tsk
)
1247 struct completion
*vfork
;
1250 vfork
= tsk
->vfork_done
;
1251 if (likely(vfork
)) {
1252 tsk
->vfork_done
= NULL
;
1258 static int wait_for_vfork_done(struct task_struct
*child
,
1259 struct completion
*vfork
)
1263 freezer_do_not_count();
1264 cgroup_enter_frozen();
1265 killed
= wait_for_completion_killable(vfork
);
1266 cgroup_leave_frozen(false);
1271 child
->vfork_done
= NULL
;
1275 put_task_struct(child
);
1279 /* Please note the differences between mmput and mm_release.
1280 * mmput is called whenever we stop holding onto a mm_struct,
1281 * error success whatever.
1283 * mm_release is called after a mm_struct has been removed
1284 * from the current process.
1286 * This difference is important for error handling, when we
1287 * only half set up a mm_struct for a new process and need to restore
1288 * the old one. Because we mmput the new mm_struct before
1289 * restoring the old one. . .
1290 * Eric Biederman 10 January 1998
1292 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1294 uprobe_free_utask(tsk
);
1296 /* Get rid of any cached register state */
1297 deactivate_mm(tsk
, mm
);
1300 * Signal userspace if we're not exiting with a core dump
1301 * because we want to leave the value intact for debugging
1304 if (tsk
->clear_child_tid
) {
1305 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1306 atomic_read(&mm
->mm_users
) > 1) {
1308 * We don't check the error code - if userspace has
1309 * not set up a proper pointer then tough luck.
1311 put_user(0, tsk
->clear_child_tid
);
1312 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1313 1, NULL
, NULL
, 0, 0);
1315 tsk
->clear_child_tid
= NULL
;
1319 * All done, finally we can wake up parent and return this mm to him.
1320 * Also kthread_stop() uses this completion for synchronization.
1322 if (tsk
->vfork_done
)
1323 complete_vfork_done(tsk
);
1326 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1328 futex_exit_release(tsk
);
1329 mm_release(tsk
, mm
);
1332 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1334 futex_exec_release(tsk
);
1335 mm_release(tsk
, mm
);
1339 * dup_mm() - duplicates an existing mm structure
1340 * @tsk: the task_struct with which the new mm will be associated.
1341 * @oldmm: the mm to duplicate.
1343 * Allocates a new mm structure and duplicates the provided @oldmm structure
1346 * Return: the duplicated mm or NULL on failure.
1348 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1349 struct mm_struct
*oldmm
)
1351 struct mm_struct
*mm
;
1358 memcpy(mm
, oldmm
, sizeof(*mm
));
1360 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1363 err
= dup_mmap(mm
, oldmm
);
1367 mm
->hiwater_rss
= get_mm_rss(mm
);
1368 mm
->hiwater_vm
= mm
->total_vm
;
1370 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1376 /* don't put binfmt in mmput, we haven't got module yet */
1378 mm_init_owner(mm
, NULL
);
1385 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1387 struct mm_struct
*mm
, *oldmm
;
1390 tsk
->min_flt
= tsk
->maj_flt
= 0;
1391 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1392 #ifdef CONFIG_DETECT_HUNG_TASK
1393 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1394 tsk
->last_switch_time
= 0;
1398 tsk
->active_mm
= NULL
;
1401 * Are we cloning a kernel thread?
1403 * We need to steal a active VM for that..
1405 oldmm
= current
->mm
;
1409 /* initialize the new vmacache entries */
1410 vmacache_flush(tsk
);
1412 if (clone_flags
& CLONE_VM
) {
1419 mm
= dup_mm(tsk
, current
->mm
);
1425 tsk
->active_mm
= mm
;
1432 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1434 struct fs_struct
*fs
= current
->fs
;
1435 if (clone_flags
& CLONE_FS
) {
1436 /* tsk->fs is already what we want */
1437 spin_lock(&fs
->lock
);
1439 spin_unlock(&fs
->lock
);
1443 spin_unlock(&fs
->lock
);
1446 tsk
->fs
= copy_fs_struct(fs
);
1452 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1454 struct files_struct
*oldf
, *newf
;
1458 * A background process may not have any files ...
1460 oldf
= current
->files
;
1464 if (clone_flags
& CLONE_FILES
) {
1465 atomic_inc(&oldf
->count
);
1469 newf
= dup_fd(oldf
, &error
);
1479 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1482 struct io_context
*ioc
= current
->io_context
;
1483 struct io_context
*new_ioc
;
1488 * Share io context with parent, if CLONE_IO is set
1490 if (clone_flags
& CLONE_IO
) {
1492 tsk
->io_context
= ioc
;
1493 } else if (ioprio_valid(ioc
->ioprio
)) {
1494 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1495 if (unlikely(!new_ioc
))
1498 new_ioc
->ioprio
= ioc
->ioprio
;
1499 put_io_context(new_ioc
);
1505 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1507 struct sighand_struct
*sig
;
1509 if (clone_flags
& CLONE_SIGHAND
) {
1510 refcount_inc(¤t
->sighand
->count
);
1513 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1514 rcu_assign_pointer(tsk
->sighand
, sig
);
1518 refcount_set(&sig
->count
, 1);
1519 spin_lock_irq(¤t
->sighand
->siglock
);
1520 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1521 spin_unlock_irq(¤t
->sighand
->siglock
);
1525 void __cleanup_sighand(struct sighand_struct
*sighand
)
1527 if (refcount_dec_and_test(&sighand
->count
)) {
1528 signalfd_cleanup(sighand
);
1530 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1531 * without an RCU grace period, see __lock_task_sighand().
1533 kmem_cache_free(sighand_cachep
, sighand
);
1538 * Initialize POSIX timer handling for a thread group.
1540 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1542 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1543 unsigned long cpu_limit
;
1545 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1546 posix_cputimers_group_init(pct
, cpu_limit
);
1549 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1551 struct signal_struct
*sig
;
1553 if (clone_flags
& CLONE_THREAD
)
1556 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1561 sig
->nr_threads
= 1;
1562 atomic_set(&sig
->live
, 1);
1563 refcount_set(&sig
->sigcnt
, 1);
1565 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1566 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1567 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1569 init_waitqueue_head(&sig
->wait_chldexit
);
1570 sig
->curr_target
= tsk
;
1571 init_sigpending(&sig
->shared_pending
);
1572 INIT_HLIST_HEAD(&sig
->multiprocess
);
1573 seqlock_init(&sig
->stats_lock
);
1574 prev_cputime_init(&sig
->prev_cputime
);
1576 #ifdef CONFIG_POSIX_TIMERS
1577 INIT_LIST_HEAD(&sig
->posix_timers
);
1578 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1579 sig
->real_timer
.function
= it_real_fn
;
1582 task_lock(current
->group_leader
);
1583 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1584 task_unlock(current
->group_leader
);
1586 posix_cpu_timers_init_group(sig
);
1588 tty_audit_fork(sig
);
1589 sched_autogroup_fork(sig
);
1591 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1592 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1594 mutex_init(&sig
->cred_guard_mutex
);
1599 static void copy_seccomp(struct task_struct
*p
)
1601 #ifdef CONFIG_SECCOMP
1603 * Must be called with sighand->lock held, which is common to
1604 * all threads in the group. Holding cred_guard_mutex is not
1605 * needed because this new task is not yet running and cannot
1608 assert_spin_locked(¤t
->sighand
->siglock
);
1610 /* Ref-count the new filter user, and assign it. */
1611 get_seccomp_filter(current
);
1612 p
->seccomp
= current
->seccomp
;
1615 * Explicitly enable no_new_privs here in case it got set
1616 * between the task_struct being duplicated and holding the
1617 * sighand lock. The seccomp state and nnp must be in sync.
1619 if (task_no_new_privs(current
))
1620 task_set_no_new_privs(p
);
1623 * If the parent gained a seccomp mode after copying thread
1624 * flags and between before we held the sighand lock, we have
1625 * to manually enable the seccomp thread flag here.
1627 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1628 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1632 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1634 current
->clear_child_tid
= tidptr
;
1636 return task_pid_vnr(current
);
1639 static void rt_mutex_init_task(struct task_struct
*p
)
1641 raw_spin_lock_init(&p
->pi_lock
);
1642 #ifdef CONFIG_RT_MUTEXES
1643 p
->pi_waiters
= RB_ROOT_CACHED
;
1644 p
->pi_top_task
= NULL
;
1645 p
->pi_blocked_on
= NULL
;
1649 static inline void init_task_pid_links(struct task_struct
*task
)
1653 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1654 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1659 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1661 if (type
== PIDTYPE_PID
)
1662 task
->thread_pid
= pid
;
1664 task
->signal
->pids
[type
] = pid
;
1667 static inline void rcu_copy_process(struct task_struct
*p
)
1669 #ifdef CONFIG_PREEMPT_RCU
1670 p
->rcu_read_lock_nesting
= 0;
1671 p
->rcu_read_unlock_special
.s
= 0;
1672 p
->rcu_blocked_node
= NULL
;
1673 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1674 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1675 #ifdef CONFIG_TASKS_RCU
1676 p
->rcu_tasks_holdout
= false;
1677 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1678 p
->rcu_tasks_idle_cpu
= -1;
1679 #endif /* #ifdef CONFIG_TASKS_RCU */
1682 struct pid
*pidfd_pid(const struct file
*file
)
1684 if (file
->f_op
== &pidfd_fops
)
1685 return file
->private_data
;
1687 return ERR_PTR(-EBADF
);
1690 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1692 struct pid
*pid
= file
->private_data
;
1694 file
->private_data
= NULL
;
1699 #ifdef CONFIG_PROC_FS
1700 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1702 struct pid_namespace
*ns
= proc_pid_ns(file_inode(m
->file
));
1703 struct pid
*pid
= f
->private_data
;
1705 seq_put_decimal_ull(m
, "Pid:\t", pid_nr_ns(pid
, ns
));
1711 * Poll support for process exit notification.
1713 static __poll_t
pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1715 struct task_struct
*task
;
1716 struct pid
*pid
= file
->private_data
;
1717 __poll_t poll_flags
= 0;
1719 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1722 task
= pid_task(pid
, PIDTYPE_PID
);
1724 * Inform pollers only when the whole thread group exits.
1725 * If the thread group leader exits before all other threads in the
1726 * group, then poll(2) should block, similar to the wait(2) family.
1728 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1729 poll_flags
= EPOLLIN
| EPOLLRDNORM
;
1735 const struct file_operations pidfd_fops
= {
1736 .release
= pidfd_release
,
1738 #ifdef CONFIG_PROC_FS
1739 .show_fdinfo
= pidfd_show_fdinfo
,
1743 static void __delayed_free_task(struct rcu_head
*rhp
)
1745 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1750 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1752 if (IS_ENABLED(CONFIG_MEMCG
))
1753 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1759 * This creates a new process as a copy of the old one,
1760 * but does not actually start it yet.
1762 * It copies the registers, and all the appropriate
1763 * parts of the process environment (as per the clone
1764 * flags). The actual kick-off is left to the caller.
1766 static __latent_entropy
struct task_struct
*copy_process(
1770 struct kernel_clone_args
*args
)
1772 int pidfd
= -1, retval
;
1773 struct task_struct
*p
;
1774 struct multiprocess_signals delayed
;
1775 struct file
*pidfile
= NULL
;
1776 u64 clone_flags
= args
->flags
;
1779 * Don't allow sharing the root directory with processes in a different
1782 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1783 return ERR_PTR(-EINVAL
);
1785 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1786 return ERR_PTR(-EINVAL
);
1788 if ((clone_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
)
1789 if (!capable(CAP_SYS_ADMIN
))
1790 return ERR_PTR(-EPERM
);
1793 * Thread groups must share signals as well, and detached threads
1794 * can only be started up within the thread group.
1796 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1797 return ERR_PTR(-EINVAL
);
1800 * Shared signal handlers imply shared VM. By way of the above,
1801 * thread groups also imply shared VM. Blocking this case allows
1802 * for various simplifications in other code.
1804 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1805 return ERR_PTR(-EINVAL
);
1808 * Siblings of global init remain as zombies on exit since they are
1809 * not reaped by their parent (swapper). To solve this and to avoid
1810 * multi-rooted process trees, prevent global and container-inits
1811 * from creating siblings.
1813 if ((clone_flags
& CLONE_PARENT
) &&
1814 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1815 return ERR_PTR(-EINVAL
);
1818 * If the new process will be in a different pid or user namespace
1819 * do not allow it to share a thread group with the forking task.
1821 if (clone_flags
& CLONE_THREAD
) {
1822 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1823 (task_active_pid_ns(current
) !=
1824 current
->nsproxy
->pid_ns_for_children
))
1825 return ERR_PTR(-EINVAL
);
1828 if (clone_flags
& CLONE_PIDFD
) {
1830 * - CLONE_DETACHED is blocked so that we can potentially
1831 * reuse it later for CLONE_PIDFD.
1832 * - CLONE_THREAD is blocked until someone really needs it.
1834 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1835 return ERR_PTR(-EINVAL
);
1839 * Force any signals received before this point to be delivered
1840 * before the fork happens. Collect up signals sent to multiple
1841 * processes that happen during the fork and delay them so that
1842 * they appear to happen after the fork.
1844 sigemptyset(&delayed
.signal
);
1845 INIT_HLIST_NODE(&delayed
.node
);
1847 spin_lock_irq(¤t
->sighand
->siglock
);
1848 if (!(clone_flags
& CLONE_THREAD
))
1849 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1850 recalc_sigpending();
1851 spin_unlock_irq(¤t
->sighand
->siglock
);
1852 retval
= -ERESTARTNOINTR
;
1853 if (signal_pending(current
))
1857 p
= dup_task_struct(current
, node
);
1862 * This _must_ happen before we call free_task(), i.e. before we jump
1863 * to any of the bad_fork_* labels. This is to avoid freeing
1864 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1865 * kernel threads (PF_KTHREAD).
1867 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1869 * Clear TID on mm_release()?
1871 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1873 ftrace_graph_init_task(p
);
1875 rt_mutex_init_task(p
);
1877 #ifdef CONFIG_PROVE_LOCKING
1878 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1879 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1882 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1883 task_rlimit(p
, RLIMIT_NPROC
)) {
1884 if (p
->real_cred
->user
!= INIT_USER
&&
1885 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1888 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1890 retval
= copy_creds(p
, clone_flags
);
1895 * If multiple threads are within copy_process(), then this check
1896 * triggers too late. This doesn't hurt, the check is only there
1897 * to stop root fork bombs.
1900 if (nr_threads
>= max_threads
)
1901 goto bad_fork_cleanup_count
;
1903 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1904 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1905 p
->flags
|= PF_FORKNOEXEC
;
1906 INIT_LIST_HEAD(&p
->children
);
1907 INIT_LIST_HEAD(&p
->sibling
);
1908 rcu_copy_process(p
);
1909 p
->vfork_done
= NULL
;
1910 spin_lock_init(&p
->alloc_lock
);
1912 init_sigpending(&p
->pending
);
1914 p
->utime
= p
->stime
= p
->gtime
= 0;
1915 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1916 p
->utimescaled
= p
->stimescaled
= 0;
1918 prev_cputime_init(&p
->prev_cputime
);
1920 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1921 seqcount_init(&p
->vtime
.seqcount
);
1922 p
->vtime
.starttime
= 0;
1923 p
->vtime
.state
= VTIME_INACTIVE
;
1926 #if defined(SPLIT_RSS_COUNTING)
1927 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1930 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1936 task_io_accounting_init(&p
->ioac
);
1937 acct_clear_integrals(p
);
1939 posix_cputimers_init(&p
->posix_cputimers
);
1941 p
->io_context
= NULL
;
1942 audit_set_context(p
, NULL
);
1945 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1946 if (IS_ERR(p
->mempolicy
)) {
1947 retval
= PTR_ERR(p
->mempolicy
);
1948 p
->mempolicy
= NULL
;
1949 goto bad_fork_cleanup_threadgroup_lock
;
1952 #ifdef CONFIG_CPUSETS
1953 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1954 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1955 seqcount_init(&p
->mems_allowed_seq
);
1957 #ifdef CONFIG_TRACE_IRQFLAGS
1959 p
->hardirqs_enabled
= 0;
1960 p
->hardirq_enable_ip
= 0;
1961 p
->hardirq_enable_event
= 0;
1962 p
->hardirq_disable_ip
= _THIS_IP_
;
1963 p
->hardirq_disable_event
= 0;
1964 p
->softirqs_enabled
= 1;
1965 p
->softirq_enable_ip
= _THIS_IP_
;
1966 p
->softirq_enable_event
= 0;
1967 p
->softirq_disable_ip
= 0;
1968 p
->softirq_disable_event
= 0;
1969 p
->hardirq_context
= 0;
1970 p
->softirq_context
= 0;
1973 p
->pagefault_disabled
= 0;
1975 #ifdef CONFIG_LOCKDEP
1976 lockdep_init_task(p
);
1979 #ifdef CONFIG_DEBUG_MUTEXES
1980 p
->blocked_on
= NULL
; /* not blocked yet */
1982 #ifdef CONFIG_BCACHE
1983 p
->sequential_io
= 0;
1984 p
->sequential_io_avg
= 0;
1987 /* Perform scheduler related setup. Assign this task to a CPU. */
1988 retval
= sched_fork(clone_flags
, p
);
1990 goto bad_fork_cleanup_policy
;
1992 retval
= perf_event_init_task(p
);
1994 goto bad_fork_cleanup_policy
;
1995 retval
= audit_alloc(p
);
1997 goto bad_fork_cleanup_perf
;
1998 /* copy all the process information */
2000 retval
= security_task_alloc(p
, clone_flags
);
2002 goto bad_fork_cleanup_audit
;
2003 retval
= copy_semundo(clone_flags
, p
);
2005 goto bad_fork_cleanup_security
;
2006 retval
= copy_files(clone_flags
, p
);
2008 goto bad_fork_cleanup_semundo
;
2009 retval
= copy_fs(clone_flags
, p
);
2011 goto bad_fork_cleanup_files
;
2012 retval
= copy_sighand(clone_flags
, p
);
2014 goto bad_fork_cleanup_fs
;
2015 retval
= copy_signal(clone_flags
, p
);
2017 goto bad_fork_cleanup_sighand
;
2018 retval
= copy_mm(clone_flags
, p
);
2020 goto bad_fork_cleanup_signal
;
2021 retval
= copy_namespaces(clone_flags
, p
);
2023 goto bad_fork_cleanup_mm
;
2024 retval
= copy_io(clone_flags
, p
);
2026 goto bad_fork_cleanup_namespaces
;
2027 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2030 goto bad_fork_cleanup_io
;
2032 stackleak_task_init(p
);
2034 if (pid
!= &init_struct_pid
) {
2035 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
2037 retval
= PTR_ERR(pid
);
2038 goto bad_fork_cleanup_thread
;
2043 * This has to happen after we've potentially unshared the file
2044 * descriptor table (so that the pidfd doesn't leak into the child
2045 * if the fd table isn't shared).
2047 if (clone_flags
& CLONE_PIDFD
) {
2048 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2050 goto bad_fork_free_pid
;
2054 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2055 O_RDWR
| O_CLOEXEC
);
2056 if (IS_ERR(pidfile
)) {
2057 put_unused_fd(pidfd
);
2058 retval
= PTR_ERR(pidfile
);
2059 goto bad_fork_free_pid
;
2061 get_pid(pid
); /* held by pidfile now */
2063 retval
= put_user(pidfd
, args
->pidfd
);
2065 goto bad_fork_put_pidfd
;
2074 * sigaltstack should be cleared when sharing the same VM
2076 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2080 * Syscall tracing and stepping should be turned off in the
2081 * child regardless of CLONE_PTRACE.
2083 user_disable_single_step(p
);
2084 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2085 #ifdef TIF_SYSCALL_EMU
2086 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2088 clear_tsk_latency_tracing(p
);
2090 /* ok, now we should be set up.. */
2091 p
->pid
= pid_nr(pid
);
2092 if (clone_flags
& CLONE_THREAD
) {
2093 p
->exit_signal
= -1;
2094 p
->group_leader
= current
->group_leader
;
2095 p
->tgid
= current
->tgid
;
2097 if (clone_flags
& CLONE_PARENT
)
2098 p
->exit_signal
= current
->group_leader
->exit_signal
;
2100 p
->exit_signal
= args
->exit_signal
;
2101 p
->group_leader
= p
;
2106 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2107 p
->dirty_paused_when
= 0;
2109 p
->pdeath_signal
= 0;
2110 INIT_LIST_HEAD(&p
->thread_group
);
2111 p
->task_works
= NULL
;
2113 cgroup_threadgroup_change_begin(current
);
2115 * Ensure that the cgroup subsystem policies allow the new process to be
2116 * forked. It should be noted the the new process's css_set can be changed
2117 * between here and cgroup_post_fork() if an organisation operation is in
2120 retval
= cgroup_can_fork(p
);
2122 goto bad_fork_cgroup_threadgroup_change_end
;
2125 * From this point on we must avoid any synchronous user-space
2126 * communication until we take the tasklist-lock. In particular, we do
2127 * not want user-space to be able to predict the process start-time by
2128 * stalling fork(2) after we recorded the start_time but before it is
2129 * visible to the system.
2132 p
->start_time
= ktime_get_ns();
2133 p
->real_start_time
= ktime_get_boottime_ns();
2136 * Make it visible to the rest of the system, but dont wake it up yet.
2137 * Need tasklist lock for parent etc handling!
2139 write_lock_irq(&tasklist_lock
);
2141 /* CLONE_PARENT re-uses the old parent */
2142 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2143 p
->real_parent
= current
->real_parent
;
2144 p
->parent_exec_id
= current
->parent_exec_id
;
2146 p
->real_parent
= current
;
2147 p
->parent_exec_id
= current
->self_exec_id
;
2150 klp_copy_process(p
);
2152 spin_lock(¤t
->sighand
->siglock
);
2155 * Copy seccomp details explicitly here, in case they were changed
2156 * before holding sighand lock.
2160 rseq_fork(p
, clone_flags
);
2162 /* Don't start children in a dying pid namespace */
2163 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2165 goto bad_fork_cancel_cgroup
;
2168 /* Let kill terminate clone/fork in the middle */
2169 if (fatal_signal_pending(current
)) {
2171 goto bad_fork_cancel_cgroup
;
2174 /* past the last point of failure */
2176 fd_install(pidfd
, pidfile
);
2178 init_task_pid_links(p
);
2179 if (likely(p
->pid
)) {
2180 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2182 init_task_pid(p
, PIDTYPE_PID
, pid
);
2183 if (thread_group_leader(p
)) {
2184 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2185 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2186 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2188 if (is_child_reaper(pid
)) {
2189 ns_of_pid(pid
)->child_reaper
= p
;
2190 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2192 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2193 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2195 * Inherit has_child_subreaper flag under the same
2196 * tasklist_lock with adding child to the process tree
2197 * for propagate_has_child_subreaper optimization.
2199 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2200 p
->real_parent
->signal
->is_child_subreaper
;
2201 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2202 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2203 attach_pid(p
, PIDTYPE_TGID
);
2204 attach_pid(p
, PIDTYPE_PGID
);
2205 attach_pid(p
, PIDTYPE_SID
);
2206 __this_cpu_inc(process_counts
);
2208 current
->signal
->nr_threads
++;
2209 atomic_inc(¤t
->signal
->live
);
2210 refcount_inc(¤t
->signal
->sigcnt
);
2211 task_join_group_stop(p
);
2212 list_add_tail_rcu(&p
->thread_group
,
2213 &p
->group_leader
->thread_group
);
2214 list_add_tail_rcu(&p
->thread_node
,
2215 &p
->signal
->thread_head
);
2217 attach_pid(p
, PIDTYPE_PID
);
2221 hlist_del_init(&delayed
.node
);
2222 spin_unlock(¤t
->sighand
->siglock
);
2223 syscall_tracepoint_update(p
);
2224 write_unlock_irq(&tasklist_lock
);
2226 proc_fork_connector(p
);
2227 cgroup_post_fork(p
);
2228 cgroup_threadgroup_change_end(current
);
2231 trace_task_newtask(p
, clone_flags
);
2232 uprobe_copy_process(p
, clone_flags
);
2236 bad_fork_cancel_cgroup
:
2237 spin_unlock(¤t
->sighand
->siglock
);
2238 write_unlock_irq(&tasklist_lock
);
2239 cgroup_cancel_fork(p
);
2240 bad_fork_cgroup_threadgroup_change_end
:
2241 cgroup_threadgroup_change_end(current
);
2243 if (clone_flags
& CLONE_PIDFD
) {
2245 put_unused_fd(pidfd
);
2248 if (pid
!= &init_struct_pid
)
2250 bad_fork_cleanup_thread
:
2252 bad_fork_cleanup_io
:
2255 bad_fork_cleanup_namespaces
:
2256 exit_task_namespaces(p
);
2257 bad_fork_cleanup_mm
:
2259 mm_clear_owner(p
->mm
, p
);
2262 bad_fork_cleanup_signal
:
2263 if (!(clone_flags
& CLONE_THREAD
))
2264 free_signal_struct(p
->signal
);
2265 bad_fork_cleanup_sighand
:
2266 __cleanup_sighand(p
->sighand
);
2267 bad_fork_cleanup_fs
:
2268 exit_fs(p
); /* blocking */
2269 bad_fork_cleanup_files
:
2270 exit_files(p
); /* blocking */
2271 bad_fork_cleanup_semundo
:
2273 bad_fork_cleanup_security
:
2274 security_task_free(p
);
2275 bad_fork_cleanup_audit
:
2277 bad_fork_cleanup_perf
:
2278 perf_event_free_task(p
);
2279 bad_fork_cleanup_policy
:
2280 lockdep_free_task(p
);
2282 mpol_put(p
->mempolicy
);
2283 bad_fork_cleanup_threadgroup_lock
:
2285 delayacct_tsk_free(p
);
2286 bad_fork_cleanup_count
:
2287 atomic_dec(&p
->cred
->user
->processes
);
2290 p
->state
= TASK_DEAD
;
2292 delayed_free_task(p
);
2294 spin_lock_irq(¤t
->sighand
->siglock
);
2295 hlist_del_init(&delayed
.node
);
2296 spin_unlock_irq(¤t
->sighand
->siglock
);
2297 return ERR_PTR(retval
);
2300 static inline void init_idle_pids(struct task_struct
*idle
)
2304 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2305 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2306 init_task_pid(idle
, type
, &init_struct_pid
);
2310 struct task_struct
*fork_idle(int cpu
)
2312 struct task_struct
*task
;
2313 struct kernel_clone_args args
= {
2317 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2318 if (!IS_ERR(task
)) {
2319 init_idle_pids(task
);
2320 init_idle(task
, cpu
);
2326 struct mm_struct
*copy_init_mm(void)
2328 return dup_mm(NULL
, &init_mm
);
2332 * Ok, this is the main fork-routine.
2334 * It copies the process, and if successful kick-starts
2335 * it and waits for it to finish using the VM if required.
2337 * args->exit_signal is expected to be checked for sanity by the caller.
2339 long _do_fork(struct kernel_clone_args
*args
)
2341 u64 clone_flags
= args
->flags
;
2342 struct completion vfork
;
2344 struct task_struct
*p
;
2349 * Determine whether and which event to report to ptracer. When
2350 * called from kernel_thread or CLONE_UNTRACED is explicitly
2351 * requested, no event is reported; otherwise, report if the event
2352 * for the type of forking is enabled.
2354 if (!(clone_flags
& CLONE_UNTRACED
)) {
2355 if (clone_flags
& CLONE_VFORK
)
2356 trace
= PTRACE_EVENT_VFORK
;
2357 else if (args
->exit_signal
!= SIGCHLD
)
2358 trace
= PTRACE_EVENT_CLONE
;
2360 trace
= PTRACE_EVENT_FORK
;
2362 if (likely(!ptrace_event_enabled(current
, trace
)))
2366 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2367 add_latent_entropy();
2373 * Do this prior waking up the new thread - the thread pointer
2374 * might get invalid after that point, if the thread exits quickly.
2376 trace_sched_process_fork(current
, p
);
2378 pid
= get_task_pid(p
, PIDTYPE_PID
);
2381 if (clone_flags
& CLONE_PARENT_SETTID
)
2382 put_user(nr
, args
->parent_tid
);
2384 if (clone_flags
& CLONE_VFORK
) {
2385 p
->vfork_done
= &vfork
;
2386 init_completion(&vfork
);
2390 wake_up_new_task(p
);
2392 /* forking complete and child started to run, tell ptracer */
2393 if (unlikely(trace
))
2394 ptrace_event_pid(trace
, pid
);
2396 if (clone_flags
& CLONE_VFORK
) {
2397 if (!wait_for_vfork_done(p
, &vfork
))
2398 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2405 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2407 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2408 if ((kargs
->flags
& CLONE_PIDFD
) &&
2409 (kargs
->flags
& CLONE_PARENT_SETTID
))
2415 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2416 /* For compatibility with architectures that call do_fork directly rather than
2417 * using the syscall entry points below. */
2418 long do_fork(unsigned long clone_flags
,
2419 unsigned long stack_start
,
2420 unsigned long stack_size
,
2421 int __user
*parent_tidptr
,
2422 int __user
*child_tidptr
)
2424 struct kernel_clone_args args
= {
2425 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2426 .pidfd
= parent_tidptr
,
2427 .child_tid
= child_tidptr
,
2428 .parent_tid
= parent_tidptr
,
2429 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2430 .stack
= stack_start
,
2431 .stack_size
= stack_size
,
2434 if (!legacy_clone_args_valid(&args
))
2437 return _do_fork(&args
);
2442 * Create a kernel thread.
2444 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2446 struct kernel_clone_args args
= {
2447 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2448 CLONE_UNTRACED
) & ~CSIGNAL
),
2449 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2450 .stack
= (unsigned long)fn
,
2451 .stack_size
= (unsigned long)arg
,
2454 return _do_fork(&args
);
2457 #ifdef __ARCH_WANT_SYS_FORK
2458 SYSCALL_DEFINE0(fork
)
2461 struct kernel_clone_args args
= {
2462 .exit_signal
= SIGCHLD
,
2465 return _do_fork(&args
);
2467 /* can not support in nommu mode */
2473 #ifdef __ARCH_WANT_SYS_VFORK
2474 SYSCALL_DEFINE0(vfork
)
2476 struct kernel_clone_args args
= {
2477 .flags
= CLONE_VFORK
| CLONE_VM
,
2478 .exit_signal
= SIGCHLD
,
2481 return _do_fork(&args
);
2485 #ifdef __ARCH_WANT_SYS_CLONE
2486 #ifdef CONFIG_CLONE_BACKWARDS
2487 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2488 int __user
*, parent_tidptr
,
2490 int __user
*, child_tidptr
)
2491 #elif defined(CONFIG_CLONE_BACKWARDS2)
2492 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2493 int __user
*, parent_tidptr
,
2494 int __user
*, child_tidptr
,
2496 #elif defined(CONFIG_CLONE_BACKWARDS3)
2497 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2499 int __user
*, parent_tidptr
,
2500 int __user
*, child_tidptr
,
2503 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2504 int __user
*, parent_tidptr
,
2505 int __user
*, child_tidptr
,
2509 struct kernel_clone_args args
= {
2510 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2511 .pidfd
= parent_tidptr
,
2512 .child_tid
= child_tidptr
,
2513 .parent_tid
= parent_tidptr
,
2514 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2519 if (!legacy_clone_args_valid(&args
))
2522 return _do_fork(&args
);
2526 #ifdef __ARCH_WANT_SYS_CLONE3
2529 * copy_thread implementations handle CLONE_SETTLS by reading the TLS value from
2530 * the registers containing the syscall arguments for clone. This doesn't work
2531 * with clone3 since the TLS value is passed in clone_args instead.
2533 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2534 #error clone3 requires copy_thread_tls support in arch
2537 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2538 struct clone_args __user
*uargs
,
2542 struct clone_args args
;
2544 if (unlikely(usize
> PAGE_SIZE
))
2546 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2549 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2554 * Verify that higher 32bits of exit_signal are unset and that
2555 * it is a valid signal
2557 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2558 !valid_signal(args
.exit_signal
)))
2561 *kargs
= (struct kernel_clone_args
){
2562 .flags
= args
.flags
,
2563 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2564 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2565 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2566 .exit_signal
= args
.exit_signal
,
2567 .stack
= args
.stack
,
2568 .stack_size
= args
.stack_size
,
2576 * clone3_stack_valid - check and prepare stack
2577 * @kargs: kernel clone args
2579 * Verify that the stack arguments userspace gave us are sane.
2580 * In addition, set the stack direction for userspace since it's easy for us to
2583 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
2585 if (kargs
->stack
== 0) {
2586 if (kargs
->stack_size
> 0)
2589 if (kargs
->stack_size
== 0)
2592 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
2595 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2596 kargs
->stack
+= kargs
->stack_size
;
2603 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
2606 * All lower bits of the flag word are taken.
2607 * Verify that no other unknown flags are passed along.
2609 if (kargs
->flags
& ~CLONE_LEGACY_FLAGS
)
2613 * - make the CLONE_DETACHED bit reuseable for clone3
2614 * - make the CSIGNAL bits reuseable for clone3
2616 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2619 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2623 if (!clone3_stack_valid(kargs
))
2630 * clone3 - create a new process with specific properties
2631 * @uargs: argument structure
2632 * @size: size of @uargs
2634 * clone3() is the extensible successor to clone()/clone2().
2635 * It takes a struct as argument that is versioned by its size.
2637 * Return: On success, a positive PID for the child process.
2638 * On error, a negative errno number.
2640 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2644 struct kernel_clone_args kargs
;
2646 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2650 if (!clone3_args_valid(&kargs
))
2653 return _do_fork(&kargs
);
2657 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2659 struct task_struct
*leader
, *parent
, *child
;
2662 read_lock(&tasklist_lock
);
2663 leader
= top
= top
->group_leader
;
2665 for_each_thread(leader
, parent
) {
2666 list_for_each_entry(child
, &parent
->children
, sibling
) {
2667 res
= visitor(child
, data
);
2679 if (leader
!= top
) {
2681 parent
= child
->real_parent
;
2682 leader
= parent
->group_leader
;
2686 read_unlock(&tasklist_lock
);
2689 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2690 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2693 static void sighand_ctor(void *data
)
2695 struct sighand_struct
*sighand
= data
;
2697 spin_lock_init(&sighand
->siglock
);
2698 init_waitqueue_head(&sighand
->signalfd_wqh
);
2701 void __init
proc_caches_init(void)
2703 unsigned int mm_size
;
2705 sighand_cachep
= kmem_cache_create("sighand_cache",
2706 sizeof(struct sighand_struct
), 0,
2707 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2708 SLAB_ACCOUNT
, sighand_ctor
);
2709 signal_cachep
= kmem_cache_create("signal_cache",
2710 sizeof(struct signal_struct
), 0,
2711 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2713 files_cachep
= kmem_cache_create("files_cache",
2714 sizeof(struct files_struct
), 0,
2715 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2717 fs_cachep
= kmem_cache_create("fs_cache",
2718 sizeof(struct fs_struct
), 0,
2719 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2723 * The mm_cpumask is located at the end of mm_struct, and is
2724 * dynamically sized based on the maximum CPU number this system
2725 * can have, taking hotplug into account (nr_cpu_ids).
2727 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2729 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2730 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2731 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2732 offsetof(struct mm_struct
, saved_auxv
),
2733 sizeof_field(struct mm_struct
, saved_auxv
),
2735 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2737 nsproxy_cache_init();
2741 * Check constraints on flags passed to the unshare system call.
2743 static int check_unshare_flags(unsigned long unshare_flags
)
2745 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2746 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2747 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2748 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2751 * Not implemented, but pretend it works if there is nothing
2752 * to unshare. Note that unsharing the address space or the
2753 * signal handlers also need to unshare the signal queues (aka
2756 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2757 if (!thread_group_empty(current
))
2760 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2761 if (refcount_read(¤t
->sighand
->count
) > 1)
2764 if (unshare_flags
& CLONE_VM
) {
2765 if (!current_is_single_threaded())
2773 * Unshare the filesystem structure if it is being shared
2775 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2777 struct fs_struct
*fs
= current
->fs
;
2779 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2782 /* don't need lock here; in the worst case we'll do useless copy */
2786 *new_fsp
= copy_fs_struct(fs
);
2794 * Unshare file descriptor table if it is being shared
2796 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2798 struct files_struct
*fd
= current
->files
;
2801 if ((unshare_flags
& CLONE_FILES
) &&
2802 (fd
&& atomic_read(&fd
->count
) > 1)) {
2803 *new_fdp
= dup_fd(fd
, &error
);
2812 * unshare allows a process to 'unshare' part of the process
2813 * context which was originally shared using clone. copy_*
2814 * functions used by do_fork() cannot be used here directly
2815 * because they modify an inactive task_struct that is being
2816 * constructed. Here we are modifying the current, active,
2819 int ksys_unshare(unsigned long unshare_flags
)
2821 struct fs_struct
*fs
, *new_fs
= NULL
;
2822 struct files_struct
*fd
, *new_fd
= NULL
;
2823 struct cred
*new_cred
= NULL
;
2824 struct nsproxy
*new_nsproxy
= NULL
;
2829 * If unsharing a user namespace must also unshare the thread group
2830 * and unshare the filesystem root and working directories.
2832 if (unshare_flags
& CLONE_NEWUSER
)
2833 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2835 * If unsharing vm, must also unshare signal handlers.
2837 if (unshare_flags
& CLONE_VM
)
2838 unshare_flags
|= CLONE_SIGHAND
;
2840 * If unsharing a signal handlers, must also unshare the signal queues.
2842 if (unshare_flags
& CLONE_SIGHAND
)
2843 unshare_flags
|= CLONE_THREAD
;
2845 * If unsharing namespace, must also unshare filesystem information.
2847 if (unshare_flags
& CLONE_NEWNS
)
2848 unshare_flags
|= CLONE_FS
;
2850 if ((unshare_flags
& CLONE_NEWUSER
) && !unprivileged_userns_clone
) {
2852 if (!capable(CAP_SYS_ADMIN
))
2853 goto bad_unshare_out
;
2856 err
= check_unshare_flags(unshare_flags
);
2858 goto bad_unshare_out
;
2860 * CLONE_NEWIPC must also detach from the undolist: after switching
2861 * to a new ipc namespace, the semaphore arrays from the old
2862 * namespace are unreachable.
2864 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2866 err
= unshare_fs(unshare_flags
, &new_fs
);
2868 goto bad_unshare_out
;
2869 err
= unshare_fd(unshare_flags
, &new_fd
);
2871 goto bad_unshare_cleanup_fs
;
2872 err
= unshare_userns(unshare_flags
, &new_cred
);
2874 goto bad_unshare_cleanup_fd
;
2875 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2878 goto bad_unshare_cleanup_cred
;
2880 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2883 * CLONE_SYSVSEM is equivalent to sys_exit().
2887 if (unshare_flags
& CLONE_NEWIPC
) {
2888 /* Orphan segments in old ns (see sem above). */
2890 shm_init_task(current
);
2894 switch_task_namespaces(current
, new_nsproxy
);
2900 spin_lock(&fs
->lock
);
2901 current
->fs
= new_fs
;
2906 spin_unlock(&fs
->lock
);
2910 fd
= current
->files
;
2911 current
->files
= new_fd
;
2915 task_unlock(current
);
2918 /* Install the new user namespace */
2919 commit_creds(new_cred
);
2924 perf_event_namespaces(current
);
2926 bad_unshare_cleanup_cred
:
2929 bad_unshare_cleanup_fd
:
2931 put_files_struct(new_fd
);
2933 bad_unshare_cleanup_fs
:
2935 free_fs_struct(new_fs
);
2941 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2943 return ksys_unshare(unshare_flags
);
2947 * Helper to unshare the files of the current task.
2948 * We don't want to expose copy_files internals to
2949 * the exec layer of the kernel.
2952 int unshare_files(struct files_struct
**displaced
)
2954 struct task_struct
*task
= current
;
2955 struct files_struct
*copy
= NULL
;
2958 error
= unshare_fd(CLONE_FILES
, ©
);
2959 if (error
|| !copy
) {
2963 *displaced
= task
->files
;
2970 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2971 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2975 int threads
= max_threads
;
2977 int max
= MAX_THREADS
;
2984 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2988 max_threads
= threads
;