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/kmsan.h>
41 #include <linux/binfmts.h>
42 #include <linux/mman.h>
43 #include <linux/mmu_notifier.h>
46 #include <linux/mm_inline.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/syscall_user_dispatch.h>
57 #include <linux/jiffies.h>
58 #include <linux/futex.h>
59 #include <linux/compat.h>
60 #include <linux/kthread.h>
61 #include <linux/task_io_accounting_ops.h>
62 #include <linux/rcupdate.h>
63 #include <linux/ptrace.h>
64 #include <linux/mount.h>
65 #include <linux/audit.h>
66 #include <linux/memcontrol.h>
67 #include <linux/ftrace.h>
68 #include <linux/proc_fs.h>
69 #include <linux/profile.h>
70 #include <linux/rmap.h>
71 #include <linux/ksm.h>
72 #include <linux/acct.h>
73 #include <linux/userfaultfd_k.h>
74 #include <linux/tsacct_kern.h>
75 #include <linux/cn_proc.h>
76 #include <linux/freezer.h>
77 #include <linux/delayacct.h>
78 #include <linux/taskstats_kern.h>
79 #include <linux/tty.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.h>
100 #include <linux/stackprotector.h>
101 #include <linux/user_events.h>
102 #include <linux/iommu.h>
103 #include <linux/rseq.h>
104 #include <uapi/linux/pidfd.h>
106 #include <asm/pgalloc.h>
107 #include <linux/uaccess.h>
108 #include <asm/mmu_context.h>
109 #include <asm/cacheflush.h>
110 #include <asm/tlbflush.h>
112 #include <trace/events/sched.h>
114 #define CREATE_TRACE_POINTS
115 #include <trace/events/task.h>
118 * Minimum number of threads to boot the kernel
120 #define MIN_THREADS 20
123 * Maximum number of threads
125 #define MAX_THREADS FUTEX_TID_MASK
128 * Protected counters by write_lock_irq(&tasklist_lock)
130 unsigned long total_forks
; /* Handle normal Linux uptimes. */
131 int nr_threads
; /* The idle threads do not count.. */
133 static int max_threads
; /* tunable limit on nr_threads */
135 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
137 static const char * const resident_page_types
[] = {
138 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
139 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
140 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
141 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
144 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
146 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
148 #ifdef CONFIG_PROVE_RCU
149 int lockdep_tasklist_lock_is_held(void)
151 return lockdep_is_held(&tasklist_lock
);
153 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
154 #endif /* #ifdef CONFIG_PROVE_RCU */
156 int nr_processes(void)
161 for_each_possible_cpu(cpu
)
162 total
+= per_cpu(process_counts
, cpu
);
167 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
171 static struct kmem_cache
*task_struct_cachep
;
173 static inline struct task_struct
*alloc_task_struct_node(int node
)
175 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
178 static inline void free_task_struct(struct task_struct
*tsk
)
180 kmem_cache_free(task_struct_cachep
, tsk
);
184 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
185 * kmemcache based allocator.
187 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189 # ifdef CONFIG_VMAP_STACK
191 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
192 * flush. Try to minimize the number of calls by caching stacks.
194 #define NR_CACHED_STACKS 2
195 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
199 struct vm_struct
*stack_vm_area
;
202 static bool try_release_thread_stack_to_cache(struct vm_struct
*vm
)
206 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
207 if (this_cpu_cmpxchg(cached_stacks
[i
], NULL
, vm
) != NULL
)
214 static void thread_stack_free_rcu(struct rcu_head
*rh
)
216 struct vm_stack
*vm_stack
= container_of(rh
, struct vm_stack
, rcu
);
218 if (try_release_thread_stack_to_cache(vm_stack
->stack_vm_area
))
224 static void thread_stack_delayed_free(struct task_struct
*tsk
)
226 struct vm_stack
*vm_stack
= tsk
->stack
;
228 vm_stack
->stack_vm_area
= tsk
->stack_vm_area
;
229 call_rcu(&vm_stack
->rcu
, thread_stack_free_rcu
);
232 static int free_vm_stack_cache(unsigned int cpu
)
234 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
237 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
238 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
243 vfree(vm_stack
->addr
);
244 cached_vm_stacks
[i
] = NULL
;
250 static int memcg_charge_kernel_stack(struct vm_struct
*vm
)
256 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
258 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
259 ret
= memcg_kmem_charge_page(vm
->pages
[i
], GFP_KERNEL
, 0);
266 for (i
= 0; i
< nr_charged
; i
++)
267 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
271 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
273 struct vm_struct
*vm
;
277 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
280 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
285 /* Reset stack metadata. */
286 kasan_unpoison_range(s
->addr
, THREAD_SIZE
);
288 stack
= kasan_reset_tag(s
->addr
);
290 /* Clear stale pointers from reused stack. */
291 memset(stack
, 0, THREAD_SIZE
);
293 if (memcg_charge_kernel_stack(s
)) {
298 tsk
->stack_vm_area
= s
;
304 * Allocated stacks are cached and later reused by new threads,
305 * so memcg accounting is performed manually on assigning/releasing
306 * stacks to tasks. Drop __GFP_ACCOUNT.
308 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
309 VMALLOC_START
, VMALLOC_END
,
310 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
312 0, node
, __builtin_return_address(0));
316 vm
= find_vm_area(stack
);
317 if (memcg_charge_kernel_stack(vm
)) {
322 * We can't call find_vm_area() in interrupt context, and
323 * free_thread_stack() can be called in interrupt context,
324 * so cache the vm_struct.
326 tsk
->stack_vm_area
= vm
;
327 stack
= kasan_reset_tag(stack
);
332 static void free_thread_stack(struct task_struct
*tsk
)
334 if (!try_release_thread_stack_to_cache(tsk
->stack_vm_area
))
335 thread_stack_delayed_free(tsk
);
338 tsk
->stack_vm_area
= NULL
;
341 # else /* !CONFIG_VMAP_STACK */
343 static void thread_stack_free_rcu(struct rcu_head
*rh
)
345 __free_pages(virt_to_page(rh
), THREAD_SIZE_ORDER
);
348 static void thread_stack_delayed_free(struct task_struct
*tsk
)
350 struct rcu_head
*rh
= tsk
->stack
;
352 call_rcu(rh
, thread_stack_free_rcu
);
355 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
357 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
361 tsk
->stack
= kasan_reset_tag(page_address(page
));
367 static void free_thread_stack(struct task_struct
*tsk
)
369 thread_stack_delayed_free(tsk
);
373 # endif /* CONFIG_VMAP_STACK */
374 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
376 static struct kmem_cache
*thread_stack_cache
;
378 static void thread_stack_free_rcu(struct rcu_head
*rh
)
380 kmem_cache_free(thread_stack_cache
, rh
);
383 static void thread_stack_delayed_free(struct task_struct
*tsk
)
385 struct rcu_head
*rh
= tsk
->stack
;
387 call_rcu(rh
, thread_stack_free_rcu
);
390 static int alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
392 unsigned long *stack
;
393 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
394 stack
= kasan_reset_tag(stack
);
396 return stack
? 0 : -ENOMEM
;
399 static void free_thread_stack(struct task_struct
*tsk
)
401 thread_stack_delayed_free(tsk
);
405 void thread_stack_cache_init(void)
407 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
408 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
410 BUG_ON(thread_stack_cache
== NULL
);
413 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415 /* SLAB cache for signal_struct structures (tsk->signal) */
416 static struct kmem_cache
*signal_cachep
;
418 /* SLAB cache for sighand_struct structures (tsk->sighand) */
419 struct kmem_cache
*sighand_cachep
;
421 /* SLAB cache for files_struct structures (tsk->files) */
422 struct kmem_cache
*files_cachep
;
424 /* SLAB cache for fs_struct structures (tsk->fs) */
425 struct kmem_cache
*fs_cachep
;
427 /* SLAB cache for vm_area_struct structures */
428 static struct kmem_cache
*vm_area_cachep
;
430 /* SLAB cache for mm_struct structures (tsk->mm) */
431 static struct kmem_cache
*mm_cachep
;
433 #ifdef CONFIG_PER_VMA_LOCK
435 /* SLAB cache for vm_area_struct.lock */
436 static struct kmem_cache
*vma_lock_cachep
;
438 static bool vma_lock_alloc(struct vm_area_struct
*vma
)
440 vma
->vm_lock
= kmem_cache_alloc(vma_lock_cachep
, GFP_KERNEL
);
444 init_rwsem(&vma
->vm_lock
->lock
);
445 vma
->vm_lock_seq
= -1;
450 static inline void vma_lock_free(struct vm_area_struct
*vma
)
452 kmem_cache_free(vma_lock_cachep
, vma
->vm_lock
);
455 #else /* CONFIG_PER_VMA_LOCK */
457 static inline bool vma_lock_alloc(struct vm_area_struct
*vma
) { return true; }
458 static inline void vma_lock_free(struct vm_area_struct
*vma
) {}
460 #endif /* CONFIG_PER_VMA_LOCK */
462 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
464 struct vm_area_struct
*vma
;
466 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
471 if (!vma_lock_alloc(vma
)) {
472 kmem_cache_free(vm_area_cachep
, vma
);
479 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
481 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
486 ASSERT_EXCLUSIVE_WRITER(orig
->vm_flags
);
487 ASSERT_EXCLUSIVE_WRITER(orig
->vm_file
);
489 * orig->shared.rb may be modified concurrently, but the clone
490 * will be reinitialized.
492 data_race(memcpy(new, orig
, sizeof(*new)));
493 if (!vma_lock_alloc(new)) {
494 kmem_cache_free(vm_area_cachep
, new);
497 INIT_LIST_HEAD(&new->anon_vma_chain
);
498 vma_numab_state_init(new);
499 dup_anon_vma_name(orig
, new);
504 void __vm_area_free(struct vm_area_struct
*vma
)
506 vma_numab_state_free(vma
);
507 free_anon_vma_name(vma
);
509 kmem_cache_free(vm_area_cachep
, vma
);
512 #ifdef CONFIG_PER_VMA_LOCK
513 static void vm_area_free_rcu_cb(struct rcu_head
*head
)
515 struct vm_area_struct
*vma
= container_of(head
, struct vm_area_struct
,
518 /* The vma should not be locked while being destroyed. */
519 VM_BUG_ON_VMA(rwsem_is_locked(&vma
->vm_lock
->lock
), vma
);
524 void vm_area_free(struct vm_area_struct
*vma
)
526 #ifdef CONFIG_PER_VMA_LOCK
527 call_rcu(&vma
->vm_rcu
, vm_area_free_rcu_cb
);
533 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
535 if (IS_ENABLED(CONFIG_VMAP_STACK
)) {
536 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
539 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
540 mod_lruvec_page_state(vm
->pages
[i
], NR_KERNEL_STACK_KB
,
541 account
* (PAGE_SIZE
/ 1024));
543 void *stack
= task_stack_page(tsk
);
545 /* All stack pages are in the same node. */
546 mod_lruvec_kmem_state(stack
, NR_KERNEL_STACK_KB
,
547 account
* (THREAD_SIZE
/ 1024));
551 void exit_task_stack_account(struct task_struct
*tsk
)
553 account_kernel_stack(tsk
, -1);
555 if (IS_ENABLED(CONFIG_VMAP_STACK
)) {
556 struct vm_struct
*vm
;
559 vm
= task_stack_vm_area(tsk
);
560 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++)
561 memcg_kmem_uncharge_page(vm
->pages
[i
], 0);
565 static void release_task_stack(struct task_struct
*tsk
)
567 if (WARN_ON(READ_ONCE(tsk
->__state
) != TASK_DEAD
))
568 return; /* Better to leak the stack than to free prematurely */
570 free_thread_stack(tsk
);
573 #ifdef CONFIG_THREAD_INFO_IN_TASK
574 void put_task_stack(struct task_struct
*tsk
)
576 if (refcount_dec_and_test(&tsk
->stack_refcount
))
577 release_task_stack(tsk
);
581 void free_task(struct task_struct
*tsk
)
583 #ifdef CONFIG_SECCOMP
584 WARN_ON_ONCE(tsk
->seccomp
.filter
);
586 release_user_cpus_ptr(tsk
);
589 #ifndef CONFIG_THREAD_INFO_IN_TASK
591 * The task is finally done with both the stack and thread_info,
594 release_task_stack(tsk
);
597 * If the task had a separate stack allocation, it should be gone
600 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
602 rt_mutex_debug_task_free(tsk
);
603 ftrace_graph_exit_task(tsk
);
604 arch_release_task_struct(tsk
);
605 if (tsk
->flags
& PF_KTHREAD
)
606 free_kthread_struct(tsk
);
607 bpf_task_storage_free(tsk
);
608 free_task_struct(tsk
);
610 EXPORT_SYMBOL(free_task
);
612 static void dup_mm_exe_file(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
614 struct file
*exe_file
;
616 exe_file
= get_mm_exe_file(oldmm
);
617 RCU_INIT_POINTER(mm
->exe_file
, exe_file
);
619 * We depend on the oldmm having properly denied write access to the
622 if (exe_file
&& deny_write_access(exe_file
))
623 pr_warn_once("deny_write_access() failed in %s\n", __func__
);
627 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
628 struct mm_struct
*oldmm
)
630 struct vm_area_struct
*mpnt
, *tmp
;
632 unsigned long charge
= 0;
634 VMA_ITERATOR(vmi
, mm
, 0);
636 uprobe_start_dup_mmap();
637 if (mmap_write_lock_killable(oldmm
)) {
639 goto fail_uprobe_end
;
641 flush_cache_dup_mm(oldmm
);
642 uprobe_dup_mmap(oldmm
, mm
);
644 * Not linked in yet - no deadlock potential:
646 mmap_write_lock_nested(mm
, SINGLE_DEPTH_NESTING
);
648 /* No ordering required: file already has been exposed. */
649 dup_mm_exe_file(mm
, oldmm
);
651 mm
->total_vm
= oldmm
->total_vm
;
652 mm
->data_vm
= oldmm
->data_vm
;
653 mm
->exec_vm
= oldmm
->exec_vm
;
654 mm
->stack_vm
= oldmm
->stack_vm
;
656 retval
= ksm_fork(mm
, oldmm
);
659 khugepaged_fork(mm
, oldmm
);
661 /* Use __mt_dup() to efficiently build an identical maple tree. */
662 retval
= __mt_dup(&oldmm
->mm_mt
, &mm
->mm_mt
, GFP_KERNEL
);
663 if (unlikely(retval
))
666 mt_clear_in_rcu(vmi
.mas
.tree
);
667 for_each_vma(vmi
, mpnt
) {
670 vma_start_write(mpnt
);
671 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
672 retval
= vma_iter_clear_gfp(&vmi
, mpnt
->vm_start
,
673 mpnt
->vm_end
, GFP_KERNEL
);
677 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
682 * Don't duplicate many vmas if we've been oom-killed (for
685 if (fatal_signal_pending(current
)) {
689 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
690 unsigned long len
= vma_pages(mpnt
);
692 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
696 tmp
= vm_area_dup(mpnt
);
699 retval
= vma_dup_policy(mpnt
, tmp
);
701 goto fail_nomem_policy
;
703 retval
= dup_userfaultfd(tmp
, &uf
);
705 goto fail_nomem_anon_vma_fork
;
706 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
708 * VM_WIPEONFORK gets a clean slate in the child.
709 * Don't prepare anon_vma until fault since we don't
710 * copy page for current vma.
712 tmp
->anon_vma
= NULL
;
713 } else if (anon_vma_fork(tmp
, mpnt
))
714 goto fail_nomem_anon_vma_fork
;
715 vm_flags_clear(tmp
, VM_LOCKED_MASK
);
718 struct address_space
*mapping
= file
->f_mapping
;
721 i_mmap_lock_write(mapping
);
722 if (vma_is_shared_maywrite(tmp
))
723 mapping_allow_writable(mapping
);
724 flush_dcache_mmap_lock(mapping
);
725 /* insert tmp into the share list, just after mpnt */
726 vma_interval_tree_insert_after(tmp
, mpnt
,
728 flush_dcache_mmap_unlock(mapping
);
729 i_mmap_unlock_write(mapping
);
733 * Copy/update hugetlb private vma information.
735 if (is_vm_hugetlb_page(tmp
))
736 hugetlb_dup_vma_private(tmp
);
739 * Link the vma into the MT. After using __mt_dup(), memory
740 * allocation is not necessary here, so it cannot fail.
742 vma_iter_bulk_store(&vmi
, tmp
);
745 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
746 retval
= copy_page_range(tmp
, mpnt
);
748 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
749 tmp
->vm_ops
->open(tmp
);
752 mpnt
= vma_next(&vmi
);
756 /* a new mm has just been created */
757 retval
= arch_dup_mmap(oldmm
, mm
);
761 mt_set_in_rcu(vmi
.mas
.tree
);
764 * The entire maple tree has already been duplicated. If the
765 * mmap duplication fails, mark the failure point with
766 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
767 * stop releasing VMAs that have not been duplicated after this
770 mas_set_range(&vmi
.mas
, mpnt
->vm_start
, mpnt
->vm_end
- 1);
771 mas_store(&vmi
.mas
, XA_ZERO_ENTRY
);
774 mmap_write_unlock(mm
);
776 mmap_write_unlock(oldmm
);
777 dup_userfaultfd_complete(&uf
);
779 uprobe_end_dup_mmap();
782 fail_nomem_anon_vma_fork
:
783 mpol_put(vma_policy(tmp
));
788 vm_unacct_memory(charge
);
792 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
794 mm
->pgd
= pgd_alloc(mm
);
795 if (unlikely(!mm
->pgd
))
800 static inline void mm_free_pgd(struct mm_struct
*mm
)
802 pgd_free(mm
, mm
->pgd
);
805 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
807 mmap_write_lock(oldmm
);
808 dup_mm_exe_file(mm
, oldmm
);
809 mmap_write_unlock(oldmm
);
812 #define mm_alloc_pgd(mm) (0)
813 #define mm_free_pgd(mm)
814 #endif /* CONFIG_MMU */
816 static void check_mm(struct mm_struct
*mm
)
820 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
821 "Please make sure 'struct resident_page_types[]' is updated as well");
823 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
824 long x
= percpu_counter_sum(&mm
->rss_stat
[i
]);
827 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
828 mm
, resident_page_types
[i
], x
);
831 if (mm_pgtables_bytes(mm
))
832 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
833 mm_pgtables_bytes(mm
));
835 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
836 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
840 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
841 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
843 static void do_check_lazy_tlb(void *arg
)
845 struct mm_struct
*mm
= arg
;
847 WARN_ON_ONCE(current
->active_mm
== mm
);
850 static void do_shoot_lazy_tlb(void *arg
)
852 struct mm_struct
*mm
= arg
;
854 if (current
->active_mm
== mm
) {
855 WARN_ON_ONCE(current
->mm
);
856 current
->active_mm
= &init_mm
;
857 switch_mm(mm
, &init_mm
, current
);
861 static void cleanup_lazy_tlbs(struct mm_struct
*mm
)
863 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN
)) {
865 * In this case, lazy tlb mms are refounted and would not reach
866 * __mmdrop until all CPUs have switched away and mmdrop()ed.
872 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
873 * requires lazy mm users to switch to another mm when the refcount
874 * drops to zero, before the mm is freed. This requires IPIs here to
875 * switch kernel threads to init_mm.
877 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
878 * switch with the final userspace teardown TLB flush which leaves the
879 * mm lazy on this CPU but no others, reducing the need for additional
880 * IPIs here. There are cases where a final IPI is still required here,
881 * such as the final mmdrop being performed on a different CPU than the
882 * one exiting, or kernel threads using the mm when userspace exits.
884 * IPI overheads have not found to be expensive, but they could be
885 * reduced in a number of possible ways, for example (roughly
886 * increasing order of complexity):
887 * - The last lazy reference created by exit_mm() could instead switch
888 * to init_mm, however it's probable this will run on the same CPU
889 * immediately afterwards, so this may not reduce IPIs much.
890 * - A batch of mms requiring IPIs could be gathered and freed at once.
891 * - CPUs store active_mm where it can be remotely checked without a
892 * lock, to filter out false-positives in the cpumask.
893 * - After mm_users or mm_count reaches zero, switching away from the
894 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
895 * with some batching or delaying of the final IPIs.
896 * - A delayed freeing and RCU-like quiescing sequence based on mm
897 * switching to avoid IPIs completely.
899 on_each_cpu_mask(mm_cpumask(mm
), do_shoot_lazy_tlb
, (void *)mm
, 1);
900 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES
))
901 on_each_cpu(do_check_lazy_tlb
, (void *)mm
, 1);
905 * Called when the last reference to the mm
906 * is dropped: either by a lazy thread or by
907 * mmput. Free the page directory and the mm.
909 void __mmdrop(struct mm_struct
*mm
)
911 BUG_ON(mm
== &init_mm
);
912 WARN_ON_ONCE(mm
== current
->mm
);
914 /* Ensure no CPUs are using this as their lazy tlb mm */
915 cleanup_lazy_tlbs(mm
);
917 WARN_ON_ONCE(mm
== current
->active_mm
);
920 mmu_notifier_subscriptions_destroy(mm
);
922 put_user_ns(mm
->user_ns
);
925 percpu_counter_destroy_many(mm
->rss_stat
, NR_MM_COUNTERS
);
929 EXPORT_SYMBOL_GPL(__mmdrop
);
931 static void mmdrop_async_fn(struct work_struct
*work
)
933 struct mm_struct
*mm
;
935 mm
= container_of(work
, struct mm_struct
, async_put_work
);
939 static void mmdrop_async(struct mm_struct
*mm
)
941 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
942 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
943 schedule_work(&mm
->async_put_work
);
947 static inline void free_signal_struct(struct signal_struct
*sig
)
949 taskstats_tgid_free(sig
);
950 sched_autogroup_exit(sig
);
952 * __mmdrop is not safe to call from softirq context on x86 due to
953 * pgd_dtor so postpone it to the async context
956 mmdrop_async(sig
->oom_mm
);
957 kmem_cache_free(signal_cachep
, sig
);
960 static inline void put_signal_struct(struct signal_struct
*sig
)
962 if (refcount_dec_and_test(&sig
->sigcnt
))
963 free_signal_struct(sig
);
966 void __put_task_struct(struct task_struct
*tsk
)
968 WARN_ON(!tsk
->exit_state
);
969 WARN_ON(refcount_read(&tsk
->usage
));
970 WARN_ON(tsk
== current
);
974 task_numa_free(tsk
, true);
975 security_task_free(tsk
);
977 delayacct_tsk_free(tsk
);
978 put_signal_struct(tsk
->signal
);
979 sched_core_free(tsk
);
982 EXPORT_SYMBOL_GPL(__put_task_struct
);
984 void __put_task_struct_rcu_cb(struct rcu_head
*rhp
)
986 struct task_struct
*task
= container_of(rhp
, struct task_struct
, rcu
);
988 __put_task_struct(task
);
990 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb
);
992 void __init __weak
arch_task_cache_init(void) { }
997 static void set_max_threads(unsigned int max_threads_suggested
)
1000 unsigned long nr_pages
= totalram_pages();
1003 * The number of threads shall be limited such that the thread
1004 * structures may only consume a small part of the available memory.
1006 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
1007 threads
= MAX_THREADS
;
1009 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
1010 (u64
) THREAD_SIZE
* 8UL);
1012 if (threads
> max_threads_suggested
)
1013 threads
= max_threads_suggested
;
1015 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
1018 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1019 /* Initialized by the architecture: */
1020 int arch_task_struct_size __read_mostly
;
1023 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
1025 /* Fetch thread_struct whitelist for the architecture. */
1026 arch_thread_struct_whitelist(offset
, size
);
1029 * Handle zero-sized whitelist or empty thread_struct, otherwise
1030 * adjust offset to position of thread_struct in task_struct.
1032 if (unlikely(*size
== 0))
1035 *offset
+= offsetof(struct task_struct
, thread
);
1038 void __init
fork_init(void)
1041 #ifndef ARCH_MIN_TASKALIGN
1042 #define ARCH_MIN_TASKALIGN 0
1044 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
1045 unsigned long useroffset
, usersize
;
1047 /* create a slab on which task_structs can be allocated */
1048 task_struct_whitelist(&useroffset
, &usersize
);
1049 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
1050 arch_task_struct_size
, align
,
1051 SLAB_PANIC
|SLAB_ACCOUNT
,
1052 useroffset
, usersize
, NULL
);
1054 /* do the arch specific task caches init */
1055 arch_task_cache_init();
1057 set_max_threads(MAX_THREADS
);
1059 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
1060 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
1061 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
1062 init_task
.signal
->rlim
[RLIMIT_NPROC
];
1064 for (i
= 0; i
< UCOUNT_COUNTS
; i
++)
1065 init_user_ns
.ucount_max
[i
] = max_threads
/2;
1067 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_NPROC
, RLIM_INFINITY
);
1068 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_MSGQUEUE
, RLIM_INFINITY
);
1069 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_SIGPENDING
, RLIM_INFINITY
);
1070 set_userns_rlimit_max(&init_user_ns
, UCOUNT_RLIMIT_MEMLOCK
, RLIM_INFINITY
);
1072 #ifdef CONFIG_VMAP_STACK
1073 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
1074 NULL
, free_vm_stack_cache
);
1079 lockdep_init_task(&init_task
);
1083 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
1084 struct task_struct
*src
)
1090 void set_task_stack_end_magic(struct task_struct
*tsk
)
1092 unsigned long *stackend
;
1094 stackend
= end_of_stack(tsk
);
1095 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
1098 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
1100 struct task_struct
*tsk
;
1103 if (node
== NUMA_NO_NODE
)
1104 node
= tsk_fork_get_node(orig
);
1105 tsk
= alloc_task_struct_node(node
);
1109 err
= arch_dup_task_struct(tsk
, orig
);
1113 err
= alloc_thread_stack_node(tsk
, node
);
1117 #ifdef CONFIG_THREAD_INFO_IN_TASK
1118 refcount_set(&tsk
->stack_refcount
, 1);
1120 account_kernel_stack(tsk
, 1);
1122 err
= scs_prepare(tsk
, node
);
1126 #ifdef CONFIG_SECCOMP
1128 * We must handle setting up seccomp filters once we're under
1129 * the sighand lock in case orig has changed between now and
1130 * then. Until then, filter must be NULL to avoid messing up
1131 * the usage counts on the error path calling free_task.
1133 tsk
->seccomp
.filter
= NULL
;
1136 setup_thread_stack(tsk
, orig
);
1137 clear_user_return_notifier(tsk
);
1138 clear_tsk_need_resched(tsk
);
1139 set_task_stack_end_magic(tsk
);
1140 clear_syscall_work_syscall_user_dispatch(tsk
);
1142 #ifdef CONFIG_STACKPROTECTOR
1143 tsk
->stack_canary
= get_random_canary();
1145 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
1146 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
1147 dup_user_cpus_ptr(tsk
, orig
, node
);
1150 * One for the user space visible state that goes away when reaped.
1151 * One for the scheduler.
1153 refcount_set(&tsk
->rcu_users
, 2);
1154 /* One for the rcu users */
1155 refcount_set(&tsk
->usage
, 1);
1156 #ifdef CONFIG_BLK_DEV_IO_TRACE
1157 tsk
->btrace_seq
= 0;
1159 tsk
->splice_pipe
= NULL
;
1160 tsk
->task_frag
.page
= NULL
;
1161 tsk
->wake_q
.next
= NULL
;
1162 tsk
->worker_private
= NULL
;
1164 kcov_task_init(tsk
);
1165 kmsan_task_create(tsk
);
1166 kmap_local_fork(tsk
);
1168 #ifdef CONFIG_FAULT_INJECTION
1172 #ifdef CONFIG_BLK_CGROUP
1173 tsk
->throttle_disk
= NULL
;
1174 tsk
->use_memdelay
= 0;
1177 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1178 tsk
->pasid_activated
= 0;
1182 tsk
->active_memcg
= NULL
;
1185 #ifdef CONFIG_CPU_SUP_INTEL
1186 tsk
->reported_split_lock
= 0;
1189 #ifdef CONFIG_SCHED_MM_CID
1191 tsk
->last_mm_cid
= -1;
1192 tsk
->mm_cid_active
= 0;
1193 tsk
->migrate_from_cpu
= -1;
1198 exit_task_stack_account(tsk
);
1199 free_thread_stack(tsk
);
1201 free_task_struct(tsk
);
1205 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
1207 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
1209 static int __init
coredump_filter_setup(char *s
)
1211 default_dump_filter
=
1212 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
1213 MMF_DUMP_FILTER_MASK
;
1217 __setup("coredump_filter=", coredump_filter_setup
);
1219 #include <linux/init_task.h>
1221 static void mm_init_aio(struct mm_struct
*mm
)
1224 spin_lock_init(&mm
->ioctx_lock
);
1225 mm
->ioctx_table
= NULL
;
1229 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
1230 struct task_struct
*p
)
1234 WRITE_ONCE(mm
->owner
, NULL
);
1238 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
1245 static void mm_init_uprobes_state(struct mm_struct
*mm
)
1247 #ifdef CONFIG_UPROBES
1248 mm
->uprobes_state
.xol_area
= NULL
;
1252 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1253 struct user_namespace
*user_ns
)
1255 mt_init_flags(&mm
->mm_mt
, MM_MT_FLAGS
);
1256 mt_set_external_lock(&mm
->mm_mt
, &mm
->mmap_lock
);
1257 atomic_set(&mm
->mm_users
, 1);
1258 atomic_set(&mm
->mm_count
, 1);
1259 seqcount_init(&mm
->write_protect_seq
);
1261 INIT_LIST_HEAD(&mm
->mmlist
);
1262 #ifdef CONFIG_PER_VMA_LOCK
1263 mm
->mm_lock_seq
= 0;
1265 mm_pgtables_bytes_init(mm
);
1268 atomic64_set(&mm
->pinned_vm
, 0);
1269 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1270 spin_lock_init(&mm
->page_table_lock
);
1271 spin_lock_init(&mm
->arg_lock
);
1272 mm_init_cpumask(mm
);
1274 mm_init_owner(mm
, p
);
1276 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1277 mmu_notifier_subscriptions_init(mm
);
1278 init_tlb_flush_pending(mm
);
1279 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1280 mm
->pmd_huge_pte
= NULL
;
1282 mm_init_uprobes_state(mm
);
1283 hugetlb_count_init(mm
);
1286 mm
->flags
= mmf_init_flags(current
->mm
->flags
);
1287 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1289 mm
->flags
= default_dump_filter
;
1293 if (mm_alloc_pgd(mm
))
1296 if (init_new_context(p
, mm
))
1297 goto fail_nocontext
;
1299 if (mm_alloc_cid(mm
))
1302 if (percpu_counter_init_many(mm
->rss_stat
, 0, GFP_KERNEL_ACCOUNT
,
1306 mm
->user_ns
= get_user_ns(user_ns
);
1307 lru_gen_init_mm(mm
);
1313 destroy_context(mm
);
1322 * Allocate and initialize an mm_struct.
1324 struct mm_struct
*mm_alloc(void)
1326 struct mm_struct
*mm
;
1332 memset(mm
, 0, sizeof(*mm
));
1333 return mm_init(mm
, current
, current_user_ns());
1336 static inline void __mmput(struct mm_struct
*mm
)
1338 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1340 uprobe_clear_state(mm
);
1343 khugepaged_exit(mm
); /* must run before exit_mmap */
1345 mm_put_huge_zero_page(mm
);
1346 set_mm_exe_file(mm
, NULL
);
1347 if (!list_empty(&mm
->mmlist
)) {
1348 spin_lock(&mmlist_lock
);
1349 list_del(&mm
->mmlist
);
1350 spin_unlock(&mmlist_lock
);
1353 module_put(mm
->binfmt
->module
);
1359 * Decrement the use count and release all resources for an mm.
1361 void mmput(struct mm_struct
*mm
)
1365 if (atomic_dec_and_test(&mm
->mm_users
))
1368 EXPORT_SYMBOL_GPL(mmput
);
1371 static void mmput_async_fn(struct work_struct
*work
)
1373 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1379 void mmput_async(struct mm_struct
*mm
)
1381 if (atomic_dec_and_test(&mm
->mm_users
)) {
1382 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1383 schedule_work(&mm
->async_put_work
);
1386 EXPORT_SYMBOL_GPL(mmput_async
);
1390 * set_mm_exe_file - change a reference to the mm's executable file
1391 * @mm: The mm to change.
1392 * @new_exe_file: The new file to use.
1394 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1396 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1397 * invocations: in mmput() nobody alive left, in execve it happens before
1398 * the new mm is made visible to anyone.
1400 * Can only fail if new_exe_file != NULL.
1402 int set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1404 struct file
*old_exe_file
;
1407 * It is safe to dereference the exe_file without RCU as
1408 * this function is only called if nobody else can access
1409 * this mm -- see comment above for justification.
1411 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1415 * We expect the caller (i.e., sys_execve) to already denied
1416 * write access, so this is unlikely to fail.
1418 if (unlikely(deny_write_access(new_exe_file
)))
1420 get_file(new_exe_file
);
1422 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1424 allow_write_access(old_exe_file
);
1431 * replace_mm_exe_file - replace a reference to the mm's executable file
1432 * @mm: The mm to change.
1433 * @new_exe_file: The new file to use.
1435 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1437 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1439 int replace_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1441 struct vm_area_struct
*vma
;
1442 struct file
*old_exe_file
;
1445 /* Forbid mm->exe_file change if old file still mapped. */
1446 old_exe_file
= get_mm_exe_file(mm
);
1448 VMA_ITERATOR(vmi
, mm
, 0);
1450 for_each_vma(vmi
, vma
) {
1453 if (path_equal(&vma
->vm_file
->f_path
,
1454 &old_exe_file
->f_path
)) {
1459 mmap_read_unlock(mm
);
1465 ret
= deny_write_access(new_exe_file
);
1468 get_file(new_exe_file
);
1470 /* set the new file */
1471 mmap_write_lock(mm
);
1472 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1473 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1474 mmap_write_unlock(mm
);
1477 allow_write_access(old_exe_file
);
1484 * get_mm_exe_file - acquire a reference to the mm's executable file
1485 * @mm: The mm of interest.
1487 * Returns %NULL if mm has no associated executable file.
1488 * User must release file via fput().
1490 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1492 struct file
*exe_file
;
1495 exe_file
= get_file_rcu(&mm
->exe_file
);
1501 * get_task_exe_file - acquire a reference to the task's executable file
1504 * Returns %NULL if task's mm (if any) has no associated executable file or
1505 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1506 * User must release file via fput().
1508 struct file
*get_task_exe_file(struct task_struct
*task
)
1510 struct file
*exe_file
= NULL
;
1511 struct mm_struct
*mm
;
1516 if (!(task
->flags
& PF_KTHREAD
))
1517 exe_file
= get_mm_exe_file(mm
);
1524 * get_task_mm - acquire a reference to the task's mm
1527 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1528 * this kernel workthread has transiently adopted a user mm with use_mm,
1529 * to do its AIO) is not set and if so returns a reference to it, after
1530 * bumping up the use count. User must release the mm via mmput()
1531 * after use. Typically used by /proc and ptrace.
1533 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1535 struct mm_struct
*mm
;
1540 if (task
->flags
& PF_KTHREAD
)
1548 EXPORT_SYMBOL_GPL(get_task_mm
);
1550 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1552 struct mm_struct
*mm
;
1555 err
= down_read_killable(&task
->signal
->exec_update_lock
);
1557 return ERR_PTR(err
);
1559 mm
= get_task_mm(task
);
1560 if (mm
&& mm
!= current
->mm
&&
1561 !ptrace_may_access(task
, mode
)) {
1563 mm
= ERR_PTR(-EACCES
);
1565 up_read(&task
->signal
->exec_update_lock
);
1570 static void complete_vfork_done(struct task_struct
*tsk
)
1572 struct completion
*vfork
;
1575 vfork
= tsk
->vfork_done
;
1576 if (likely(vfork
)) {
1577 tsk
->vfork_done
= NULL
;
1583 static int wait_for_vfork_done(struct task_struct
*child
,
1584 struct completion
*vfork
)
1586 unsigned int state
= TASK_KILLABLE
|TASK_FREEZABLE
;
1589 cgroup_enter_frozen();
1590 killed
= wait_for_completion_state(vfork
, state
);
1591 cgroup_leave_frozen(false);
1595 child
->vfork_done
= NULL
;
1599 put_task_struct(child
);
1603 /* Please note the differences between mmput and mm_release.
1604 * mmput is called whenever we stop holding onto a mm_struct,
1605 * error success whatever.
1607 * mm_release is called after a mm_struct has been removed
1608 * from the current process.
1610 * This difference is important for error handling, when we
1611 * only half set up a mm_struct for a new process and need to restore
1612 * the old one. Because we mmput the new mm_struct before
1613 * restoring the old one. . .
1614 * Eric Biederman 10 January 1998
1616 static void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1618 uprobe_free_utask(tsk
);
1620 /* Get rid of any cached register state */
1621 deactivate_mm(tsk
, mm
);
1624 * Signal userspace if we're not exiting with a core dump
1625 * because we want to leave the value intact for debugging
1628 if (tsk
->clear_child_tid
) {
1629 if (atomic_read(&mm
->mm_users
) > 1) {
1631 * We don't check the error code - if userspace has
1632 * not set up a proper pointer then tough luck.
1634 put_user(0, tsk
->clear_child_tid
);
1635 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1636 1, NULL
, NULL
, 0, 0);
1638 tsk
->clear_child_tid
= NULL
;
1642 * All done, finally we can wake up parent and return this mm to him.
1643 * Also kthread_stop() uses this completion for synchronization.
1645 if (tsk
->vfork_done
)
1646 complete_vfork_done(tsk
);
1649 void exit_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1651 futex_exit_release(tsk
);
1652 mm_release(tsk
, mm
);
1655 void exec_mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1657 futex_exec_release(tsk
);
1658 mm_release(tsk
, mm
);
1662 * dup_mm() - duplicates an existing mm structure
1663 * @tsk: the task_struct with which the new mm will be associated.
1664 * @oldmm: the mm to duplicate.
1666 * Allocates a new mm structure and duplicates the provided @oldmm structure
1669 * Return: the duplicated mm or NULL on failure.
1671 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1672 struct mm_struct
*oldmm
)
1674 struct mm_struct
*mm
;
1681 memcpy(mm
, oldmm
, sizeof(*mm
));
1683 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1686 err
= dup_mmap(mm
, oldmm
);
1690 mm
->hiwater_rss
= get_mm_rss(mm
);
1691 mm
->hiwater_vm
= mm
->total_vm
;
1693 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1699 /* don't put binfmt in mmput, we haven't got module yet */
1701 mm_init_owner(mm
, NULL
);
1708 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1710 struct mm_struct
*mm
, *oldmm
;
1712 tsk
->min_flt
= tsk
->maj_flt
= 0;
1713 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1714 #ifdef CONFIG_DETECT_HUNG_TASK
1715 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1716 tsk
->last_switch_time
= 0;
1720 tsk
->active_mm
= NULL
;
1723 * Are we cloning a kernel thread?
1725 * We need to steal a active VM for that..
1727 oldmm
= current
->mm
;
1731 if (clone_flags
& CLONE_VM
) {
1735 mm
= dup_mm(tsk
, current
->mm
);
1741 tsk
->active_mm
= mm
;
1742 sched_mm_cid_fork(tsk
);
1746 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1748 struct fs_struct
*fs
= current
->fs
;
1749 if (clone_flags
& CLONE_FS
) {
1750 /* tsk->fs is already what we want */
1751 spin_lock(&fs
->lock
);
1753 spin_unlock(&fs
->lock
);
1757 spin_unlock(&fs
->lock
);
1760 tsk
->fs
= copy_fs_struct(fs
);
1766 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
,
1769 struct files_struct
*oldf
, *newf
;
1773 * A background process may not have any files ...
1775 oldf
= current
->files
;
1784 if (clone_flags
& CLONE_FILES
) {
1785 atomic_inc(&oldf
->count
);
1789 newf
= dup_fd(oldf
, NR_OPEN_MAX
, &error
);
1799 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1801 struct sighand_struct
*sig
;
1803 if (clone_flags
& CLONE_SIGHAND
) {
1804 refcount_inc(¤t
->sighand
->count
);
1807 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1808 RCU_INIT_POINTER(tsk
->sighand
, sig
);
1812 refcount_set(&sig
->count
, 1);
1813 spin_lock_irq(¤t
->sighand
->siglock
);
1814 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1815 spin_unlock_irq(¤t
->sighand
->siglock
);
1817 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1818 if (clone_flags
& CLONE_CLEAR_SIGHAND
)
1819 flush_signal_handlers(tsk
, 0);
1824 void __cleanup_sighand(struct sighand_struct
*sighand
)
1826 if (refcount_dec_and_test(&sighand
->count
)) {
1827 signalfd_cleanup(sighand
);
1829 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1830 * without an RCU grace period, see __lock_task_sighand().
1832 kmem_cache_free(sighand_cachep
, sighand
);
1837 * Initialize POSIX timer handling for a thread group.
1839 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1841 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1842 unsigned long cpu_limit
;
1844 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1845 posix_cputimers_group_init(pct
, cpu_limit
);
1848 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1850 struct signal_struct
*sig
;
1852 if (clone_flags
& CLONE_THREAD
)
1855 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1860 sig
->nr_threads
= 1;
1861 sig
->quick_threads
= 1;
1862 atomic_set(&sig
->live
, 1);
1863 refcount_set(&sig
->sigcnt
, 1);
1865 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1866 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1867 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1869 init_waitqueue_head(&sig
->wait_chldexit
);
1870 sig
->curr_target
= tsk
;
1871 init_sigpending(&sig
->shared_pending
);
1872 INIT_HLIST_HEAD(&sig
->multiprocess
);
1873 seqlock_init(&sig
->stats_lock
);
1874 prev_cputime_init(&sig
->prev_cputime
);
1876 #ifdef CONFIG_POSIX_TIMERS
1877 INIT_LIST_HEAD(&sig
->posix_timers
);
1878 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1879 sig
->real_timer
.function
= it_real_fn
;
1882 task_lock(current
->group_leader
);
1883 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1884 task_unlock(current
->group_leader
);
1886 posix_cpu_timers_init_group(sig
);
1888 tty_audit_fork(sig
);
1889 sched_autogroup_fork(sig
);
1891 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1892 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1894 mutex_init(&sig
->cred_guard_mutex
);
1895 init_rwsem(&sig
->exec_update_lock
);
1900 static void copy_seccomp(struct task_struct
*p
)
1902 #ifdef CONFIG_SECCOMP
1904 * Must be called with sighand->lock held, which is common to
1905 * all threads in the group. Holding cred_guard_mutex is not
1906 * needed because this new task is not yet running and cannot
1909 assert_spin_locked(¤t
->sighand
->siglock
);
1911 /* Ref-count the new filter user, and assign it. */
1912 get_seccomp_filter(current
);
1913 p
->seccomp
= current
->seccomp
;
1916 * Explicitly enable no_new_privs here in case it got set
1917 * between the task_struct being duplicated and holding the
1918 * sighand lock. The seccomp state and nnp must be in sync.
1920 if (task_no_new_privs(current
))
1921 task_set_no_new_privs(p
);
1924 * If the parent gained a seccomp mode after copying thread
1925 * flags and between before we held the sighand lock, we have
1926 * to manually enable the seccomp thread flag here.
1928 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1929 set_task_syscall_work(p
, SECCOMP
);
1933 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1935 current
->clear_child_tid
= tidptr
;
1937 return task_pid_vnr(current
);
1940 static void rt_mutex_init_task(struct task_struct
*p
)
1942 raw_spin_lock_init(&p
->pi_lock
);
1943 #ifdef CONFIG_RT_MUTEXES
1944 p
->pi_waiters
= RB_ROOT_CACHED
;
1945 p
->pi_top_task
= NULL
;
1946 p
->pi_blocked_on
= NULL
;
1950 static inline void init_task_pid_links(struct task_struct
*task
)
1954 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
)
1955 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1959 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1961 if (type
== PIDTYPE_PID
)
1962 task
->thread_pid
= pid
;
1964 task
->signal
->pids
[type
] = pid
;
1967 static inline void rcu_copy_process(struct task_struct
*p
)
1969 #ifdef CONFIG_PREEMPT_RCU
1970 p
->rcu_read_lock_nesting
= 0;
1971 p
->rcu_read_unlock_special
.s
= 0;
1972 p
->rcu_blocked_node
= NULL
;
1973 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1974 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1975 #ifdef CONFIG_TASKS_RCU
1976 p
->rcu_tasks_holdout
= false;
1977 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1978 p
->rcu_tasks_idle_cpu
= -1;
1979 #endif /* #ifdef CONFIG_TASKS_RCU */
1980 #ifdef CONFIG_TASKS_TRACE_RCU
1981 p
->trc_reader_nesting
= 0;
1982 p
->trc_reader_special
.s
= 0;
1983 INIT_LIST_HEAD(&p
->trc_holdout_list
);
1984 INIT_LIST_HEAD(&p
->trc_blkd_node
);
1985 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1988 struct pid
*pidfd_pid(const struct file
*file
)
1990 if (file
->f_op
== &pidfd_fops
)
1991 return file
->private_data
;
1993 return ERR_PTR(-EBADF
);
1997 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1998 * @pid: the struct pid for which to create a pidfd
1999 * @flags: flags of the new @pidfd
2000 * @ret: Where to return the file for the pidfd.
2002 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2003 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2005 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2006 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2007 * pidfd file are prepared.
2009 * If this function returns successfully the caller is responsible to either
2010 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2011 * order to install the pidfd into its file descriptor table or they must use
2012 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2015 * This function is useful when a pidfd must already be reserved but there
2016 * might still be points of failure afterwards and the caller wants to ensure
2017 * that no pidfd is leaked into its file descriptor table.
2019 * Return: On success, a reserved pidfd is returned from the function and a new
2020 * pidfd file is returned in the last argument to the function. On
2021 * error, a negative error code is returned from the function and the
2022 * last argument remains unchanged.
2024 static int __pidfd_prepare(struct pid
*pid
, unsigned int flags
, struct file
**ret
)
2027 struct file
*pidfd_file
;
2029 pidfd
= get_unused_fd_flags(O_CLOEXEC
);
2033 pidfd_file
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2035 if (IS_ERR(pidfd_file
)) {
2036 put_unused_fd(pidfd
);
2037 return PTR_ERR(pidfd_file
);
2039 get_pid(pid
); /* held by pidfd_file now */
2041 * anon_inode_getfile() ignores everything outside of the
2042 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2044 pidfd_file
->f_flags
|= (flags
& PIDFD_THREAD
);
2050 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2051 * @pid: the struct pid for which to create a pidfd
2052 * @flags: flags of the new @pidfd
2053 * @ret: Where to return the pidfd.
2055 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2056 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2058 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2059 * task identified by @pid must be a thread-group leader.
2061 * If this function returns successfully the caller is responsible to either
2062 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2063 * order to install the pidfd into its file descriptor table or they must use
2064 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2067 * This function is useful when a pidfd must already be reserved but there
2068 * might still be points of failure afterwards and the caller wants to ensure
2069 * that no pidfd is leaked into its file descriptor table.
2071 * Return: On success, a reserved pidfd is returned from the function and a new
2072 * pidfd file is returned in the last argument to the function. On
2073 * error, a negative error code is returned from the function and the
2074 * last argument remains unchanged.
2076 int pidfd_prepare(struct pid
*pid
, unsigned int flags
, struct file
**ret
)
2078 bool thread
= flags
& PIDFD_THREAD
;
2080 if (!pid
|| !pid_has_task(pid
, thread
? PIDTYPE_PID
: PIDTYPE_TGID
))
2083 return __pidfd_prepare(pid
, flags
, ret
);
2086 static void __delayed_free_task(struct rcu_head
*rhp
)
2088 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
2093 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
2095 if (IS_ENABLED(CONFIG_MEMCG
))
2096 call_rcu(&tsk
->rcu
, __delayed_free_task
);
2101 static void copy_oom_score_adj(u64 clone_flags
, struct task_struct
*tsk
)
2103 /* Skip if kernel thread */
2107 /* Skip if spawning a thread or using vfork */
2108 if ((clone_flags
& (CLONE_VM
| CLONE_THREAD
| CLONE_VFORK
)) != CLONE_VM
)
2111 /* We need to synchronize with __set_oom_adj */
2112 mutex_lock(&oom_adj_mutex
);
2113 set_bit(MMF_MULTIPROCESS
, &tsk
->mm
->flags
);
2114 /* Update the values in case they were changed after copy_signal */
2115 tsk
->signal
->oom_score_adj
= current
->signal
->oom_score_adj
;
2116 tsk
->signal
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
2117 mutex_unlock(&oom_adj_mutex
);
2121 static void rv_task_fork(struct task_struct
*p
)
2125 for (i
= 0; i
< RV_PER_TASK_MONITORS
; i
++)
2126 p
->rv
[i
].da_mon
.monitoring
= false;
2129 #define rv_task_fork(p) do {} while (0)
2133 * This creates a new process as a copy of the old one,
2134 * but does not actually start it yet.
2136 * It copies the registers, and all the appropriate
2137 * parts of the process environment (as per the clone
2138 * flags). The actual kick-off is left to the caller.
2140 __latent_entropy
struct task_struct
*copy_process(
2144 struct kernel_clone_args
*args
)
2146 int pidfd
= -1, retval
;
2147 struct task_struct
*p
;
2148 struct multiprocess_signals delayed
;
2149 struct file
*pidfile
= NULL
;
2150 const u64 clone_flags
= args
->flags
;
2151 struct nsproxy
*nsp
= current
->nsproxy
;
2154 * Don't allow sharing the root directory with processes in a different
2157 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
2158 return ERR_PTR(-EINVAL
);
2160 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
2161 return ERR_PTR(-EINVAL
);
2164 * Thread groups must share signals as well, and detached threads
2165 * can only be started up within the thread group.
2167 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
2168 return ERR_PTR(-EINVAL
);
2171 * Shared signal handlers imply shared VM. By way of the above,
2172 * thread groups also imply shared VM. Blocking this case allows
2173 * for various simplifications in other code.
2175 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
2176 return ERR_PTR(-EINVAL
);
2179 * Siblings of global init remain as zombies on exit since they are
2180 * not reaped by their parent (swapper). To solve this and to avoid
2181 * multi-rooted process trees, prevent global and container-inits
2182 * from creating siblings.
2184 if ((clone_flags
& CLONE_PARENT
) &&
2185 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
2186 return ERR_PTR(-EINVAL
);
2189 * If the new process will be in a different pid or user namespace
2190 * do not allow it to share a thread group with the forking task.
2192 if (clone_flags
& CLONE_THREAD
) {
2193 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
2194 (task_active_pid_ns(current
) != nsp
->pid_ns_for_children
))
2195 return ERR_PTR(-EINVAL
);
2198 if (clone_flags
& CLONE_PIDFD
) {
2200 * - CLONE_DETACHED is blocked so that we can potentially
2201 * reuse it later for CLONE_PIDFD.
2203 if (clone_flags
& CLONE_DETACHED
)
2204 return ERR_PTR(-EINVAL
);
2208 * Force any signals received before this point to be delivered
2209 * before the fork happens. Collect up signals sent to multiple
2210 * processes that happen during the fork and delay them so that
2211 * they appear to happen after the fork.
2213 sigemptyset(&delayed
.signal
);
2214 INIT_HLIST_NODE(&delayed
.node
);
2216 spin_lock_irq(¤t
->sighand
->siglock
);
2217 if (!(clone_flags
& CLONE_THREAD
))
2218 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
2219 recalc_sigpending();
2220 spin_unlock_irq(¤t
->sighand
->siglock
);
2221 retval
= -ERESTARTNOINTR
;
2222 if (task_sigpending(current
))
2226 p
= dup_task_struct(current
, node
);
2229 p
->flags
&= ~PF_KTHREAD
;
2231 p
->flags
|= PF_KTHREAD
;
2232 if (args
->user_worker
) {
2234 * Mark us a user worker, and block any signal that isn't
2237 p
->flags
|= PF_USER_WORKER
;
2238 siginitsetinv(&p
->blocked
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2240 if (args
->io_thread
)
2241 p
->flags
|= PF_IO_WORKER
;
2244 strscpy_pad(p
->comm
, args
->name
, sizeof(p
->comm
));
2246 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
2248 * Clear TID on mm_release()?
2250 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
2252 ftrace_graph_init_task(p
);
2254 rt_mutex_init_task(p
);
2256 lockdep_assert_irqs_enabled();
2257 #ifdef CONFIG_PROVE_LOCKING
2258 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
2260 retval
= copy_creds(p
, clone_flags
);
2265 if (is_rlimit_overlimit(task_ucounts(p
), UCOUNT_RLIMIT_NPROC
, rlimit(RLIMIT_NPROC
))) {
2266 if (p
->real_cred
->user
!= INIT_USER
&&
2267 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
2268 goto bad_fork_cleanup_count
;
2270 current
->flags
&= ~PF_NPROC_EXCEEDED
;
2273 * If multiple threads are within copy_process(), then this check
2274 * triggers too late. This doesn't hurt, the check is only there
2275 * to stop root fork bombs.
2278 if (data_race(nr_threads
>= max_threads
))
2279 goto bad_fork_cleanup_count
;
2281 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
2282 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
| PF_NO_SETAFFINITY
);
2283 p
->flags
|= PF_FORKNOEXEC
;
2284 INIT_LIST_HEAD(&p
->children
);
2285 INIT_LIST_HEAD(&p
->sibling
);
2286 rcu_copy_process(p
);
2287 p
->vfork_done
= NULL
;
2288 spin_lock_init(&p
->alloc_lock
);
2290 init_sigpending(&p
->pending
);
2292 p
->utime
= p
->stime
= p
->gtime
= 0;
2293 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2294 p
->utimescaled
= p
->stimescaled
= 0;
2296 prev_cputime_init(&p
->prev_cputime
);
2298 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2299 seqcount_init(&p
->vtime
.seqcount
);
2300 p
->vtime
.starttime
= 0;
2301 p
->vtime
.state
= VTIME_INACTIVE
;
2304 #ifdef CONFIG_IO_URING
2308 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
2314 task_io_accounting_init(&p
->ioac
);
2315 acct_clear_integrals(p
);
2317 posix_cputimers_init(&p
->posix_cputimers
);
2319 p
->io_context
= NULL
;
2320 audit_set_context(p
, NULL
);
2322 if (args
->kthread
) {
2323 if (!set_kthread_struct(p
))
2324 goto bad_fork_cleanup_delayacct
;
2327 p
->mempolicy
= mpol_dup(p
->mempolicy
);
2328 if (IS_ERR(p
->mempolicy
)) {
2329 retval
= PTR_ERR(p
->mempolicy
);
2330 p
->mempolicy
= NULL
;
2331 goto bad_fork_cleanup_delayacct
;
2334 #ifdef CONFIG_CPUSETS
2335 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
2336 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
2337 seqcount_spinlock_init(&p
->mems_allowed_seq
, &p
->alloc_lock
);
2339 #ifdef CONFIG_TRACE_IRQFLAGS
2340 memset(&p
->irqtrace
, 0, sizeof(p
->irqtrace
));
2341 p
->irqtrace
.hardirq_disable_ip
= _THIS_IP_
;
2342 p
->irqtrace
.softirq_enable_ip
= _THIS_IP_
;
2343 p
->softirqs_enabled
= 1;
2344 p
->softirq_context
= 0;
2347 p
->pagefault_disabled
= 0;
2349 #ifdef CONFIG_LOCKDEP
2350 lockdep_init_task(p
);
2353 #ifdef CONFIG_DEBUG_MUTEXES
2354 p
->blocked_on
= NULL
; /* not blocked yet */
2356 #ifdef CONFIG_BCACHE
2357 p
->sequential_io
= 0;
2358 p
->sequential_io_avg
= 0;
2360 #ifdef CONFIG_BPF_SYSCALL
2361 RCU_INIT_POINTER(p
->bpf_storage
, NULL
);
2365 /* Perform scheduler related setup. Assign this task to a CPU. */
2366 retval
= sched_fork(clone_flags
, p
);
2368 goto bad_fork_cleanup_policy
;
2370 retval
= perf_event_init_task(p
, clone_flags
);
2372 goto bad_fork_cleanup_policy
;
2373 retval
= audit_alloc(p
);
2375 goto bad_fork_cleanup_perf
;
2376 /* copy all the process information */
2378 retval
= security_task_alloc(p
, clone_flags
);
2380 goto bad_fork_cleanup_audit
;
2381 retval
= copy_semundo(clone_flags
, p
);
2383 goto bad_fork_cleanup_security
;
2384 retval
= copy_files(clone_flags
, p
, args
->no_files
);
2386 goto bad_fork_cleanup_semundo
;
2387 retval
= copy_fs(clone_flags
, p
);
2389 goto bad_fork_cleanup_files
;
2390 retval
= copy_sighand(clone_flags
, p
);
2392 goto bad_fork_cleanup_fs
;
2393 retval
= copy_signal(clone_flags
, p
);
2395 goto bad_fork_cleanup_sighand
;
2396 retval
= copy_mm(clone_flags
, p
);
2398 goto bad_fork_cleanup_signal
;
2399 retval
= copy_namespaces(clone_flags
, p
);
2401 goto bad_fork_cleanup_mm
;
2402 retval
= copy_io(clone_flags
, p
);
2404 goto bad_fork_cleanup_namespaces
;
2405 retval
= copy_thread(p
, args
);
2407 goto bad_fork_cleanup_io
;
2409 stackleak_task_init(p
);
2411 if (pid
!= &init_struct_pid
) {
2412 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
, args
->set_tid
,
2413 args
->set_tid_size
);
2415 retval
= PTR_ERR(pid
);
2416 goto bad_fork_cleanup_thread
;
2421 * This has to happen after we've potentially unshared the file
2422 * descriptor table (so that the pidfd doesn't leak into the child
2423 * if the fd table isn't shared).
2425 if (clone_flags
& CLONE_PIDFD
) {
2426 int flags
= (clone_flags
& CLONE_THREAD
) ? PIDFD_THREAD
: 0;
2428 /* Note that no task has been attached to @pid yet. */
2429 retval
= __pidfd_prepare(pid
, flags
, &pidfile
);
2431 goto bad_fork_free_pid
;
2434 retval
= put_user(pidfd
, args
->pidfd
);
2436 goto bad_fork_put_pidfd
;
2445 * sigaltstack should be cleared when sharing the same VM
2447 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2451 * Syscall tracing and stepping should be turned off in the
2452 * child regardless of CLONE_PTRACE.
2454 user_disable_single_step(p
);
2455 clear_task_syscall_work(p
, SYSCALL_TRACE
);
2456 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2457 clear_task_syscall_work(p
, SYSCALL_EMU
);
2459 clear_tsk_latency_tracing(p
);
2461 /* ok, now we should be set up.. */
2462 p
->pid
= pid_nr(pid
);
2463 if (clone_flags
& CLONE_THREAD
) {
2464 p
->group_leader
= current
->group_leader
;
2465 p
->tgid
= current
->tgid
;
2467 p
->group_leader
= p
;
2472 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2473 p
->dirty_paused_when
= 0;
2475 p
->pdeath_signal
= 0;
2476 p
->task_works
= NULL
;
2477 clear_posix_cputimers_work(p
);
2479 #ifdef CONFIG_KRETPROBES
2480 p
->kretprobe_instances
.first
= NULL
;
2482 #ifdef CONFIG_RETHOOK
2483 p
->rethooks
.first
= NULL
;
2487 * Ensure that the cgroup subsystem policies allow the new process to be
2488 * forked. It should be noted that the new process's css_set can be changed
2489 * between here and cgroup_post_fork() if an organisation operation is in
2492 retval
= cgroup_can_fork(p
, args
);
2494 goto bad_fork_put_pidfd
;
2497 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2498 * the new task on the correct runqueue. All this *before* the task
2501 * This isn't part of ->can_fork() because while the re-cloning is
2502 * cgroup specific, it unconditionally needs to place the task on a
2505 sched_cgroup_fork(p
, args
);
2508 * From this point on we must avoid any synchronous user-space
2509 * communication until we take the tasklist-lock. In particular, we do
2510 * not want user-space to be able to predict the process start-time by
2511 * stalling fork(2) after we recorded the start_time but before it is
2512 * visible to the system.
2515 p
->start_time
= ktime_get_ns();
2516 p
->start_boottime
= ktime_get_boottime_ns();
2519 * Make it visible to the rest of the system, but dont wake it up yet.
2520 * Need tasklist lock for parent etc handling!
2522 write_lock_irq(&tasklist_lock
);
2524 /* CLONE_PARENT re-uses the old parent */
2525 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2526 p
->real_parent
= current
->real_parent
;
2527 p
->parent_exec_id
= current
->parent_exec_id
;
2528 if (clone_flags
& CLONE_THREAD
)
2529 p
->exit_signal
= -1;
2531 p
->exit_signal
= current
->group_leader
->exit_signal
;
2533 p
->real_parent
= current
;
2534 p
->parent_exec_id
= current
->self_exec_id
;
2535 p
->exit_signal
= args
->exit_signal
;
2538 klp_copy_process(p
);
2542 spin_lock(¤t
->sighand
->siglock
);
2546 rseq_fork(p
, clone_flags
);
2548 /* Don't start children in a dying pid namespace */
2549 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2551 goto bad_fork_cancel_cgroup
;
2554 /* Let kill terminate clone/fork in the middle */
2555 if (fatal_signal_pending(current
)) {
2557 goto bad_fork_cancel_cgroup
;
2560 /* No more failure paths after this point. */
2563 * Copy seccomp details explicitly here, in case they were changed
2564 * before holding sighand lock.
2568 init_task_pid_links(p
);
2569 if (likely(p
->pid
)) {
2570 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2572 init_task_pid(p
, PIDTYPE_PID
, pid
);
2573 if (thread_group_leader(p
)) {
2574 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2575 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2576 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2578 if (is_child_reaper(pid
)) {
2579 ns_of_pid(pid
)->child_reaper
= p
;
2580 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2582 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2583 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2585 * Inherit has_child_subreaper flag under the same
2586 * tasklist_lock with adding child to the process tree
2587 * for propagate_has_child_subreaper optimization.
2589 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2590 p
->real_parent
->signal
->is_child_subreaper
;
2591 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2592 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2593 attach_pid(p
, PIDTYPE_TGID
);
2594 attach_pid(p
, PIDTYPE_PGID
);
2595 attach_pid(p
, PIDTYPE_SID
);
2596 __this_cpu_inc(process_counts
);
2598 current
->signal
->nr_threads
++;
2599 current
->signal
->quick_threads
++;
2600 atomic_inc(¤t
->signal
->live
);
2601 refcount_inc(¤t
->signal
->sigcnt
);
2602 task_join_group_stop(p
);
2603 list_add_tail_rcu(&p
->thread_node
,
2604 &p
->signal
->thread_head
);
2606 attach_pid(p
, PIDTYPE_PID
);
2610 hlist_del_init(&delayed
.node
);
2611 spin_unlock(¤t
->sighand
->siglock
);
2612 syscall_tracepoint_update(p
);
2613 write_unlock_irq(&tasklist_lock
);
2616 fd_install(pidfd
, pidfile
);
2618 proc_fork_connector(p
);
2620 cgroup_post_fork(p
, args
);
2623 trace_task_newtask(p
, clone_flags
);
2624 uprobe_copy_process(p
, clone_flags
);
2625 user_events_fork(p
, clone_flags
);
2627 copy_oom_score_adj(clone_flags
, p
);
2631 bad_fork_cancel_cgroup
:
2633 spin_unlock(¤t
->sighand
->siglock
);
2634 write_unlock_irq(&tasklist_lock
);
2635 cgroup_cancel_fork(p
, args
);
2637 if (clone_flags
& CLONE_PIDFD
) {
2639 put_unused_fd(pidfd
);
2642 if (pid
!= &init_struct_pid
)
2644 bad_fork_cleanup_thread
:
2646 bad_fork_cleanup_io
:
2649 bad_fork_cleanup_namespaces
:
2650 exit_task_namespaces(p
);
2651 bad_fork_cleanup_mm
:
2653 mm_clear_owner(p
->mm
, p
);
2656 bad_fork_cleanup_signal
:
2657 if (!(clone_flags
& CLONE_THREAD
))
2658 free_signal_struct(p
->signal
);
2659 bad_fork_cleanup_sighand
:
2660 __cleanup_sighand(p
->sighand
);
2661 bad_fork_cleanup_fs
:
2662 exit_fs(p
); /* blocking */
2663 bad_fork_cleanup_files
:
2664 exit_files(p
); /* blocking */
2665 bad_fork_cleanup_semundo
:
2667 bad_fork_cleanup_security
:
2668 security_task_free(p
);
2669 bad_fork_cleanup_audit
:
2671 bad_fork_cleanup_perf
:
2672 perf_event_free_task(p
);
2673 bad_fork_cleanup_policy
:
2674 lockdep_free_task(p
);
2676 mpol_put(p
->mempolicy
);
2678 bad_fork_cleanup_delayacct
:
2679 delayacct_tsk_free(p
);
2680 bad_fork_cleanup_count
:
2681 dec_rlimit_ucounts(task_ucounts(p
), UCOUNT_RLIMIT_NPROC
, 1);
2684 WRITE_ONCE(p
->__state
, TASK_DEAD
);
2685 exit_task_stack_account(p
);
2687 delayed_free_task(p
);
2689 spin_lock_irq(¤t
->sighand
->siglock
);
2690 hlist_del_init(&delayed
.node
);
2691 spin_unlock_irq(¤t
->sighand
->siglock
);
2692 return ERR_PTR(retval
);
2695 static inline void init_idle_pids(struct task_struct
*idle
)
2699 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2700 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2701 init_task_pid(idle
, type
, &init_struct_pid
);
2705 static int idle_dummy(void *dummy
)
2707 /* This function is never called */
2711 struct task_struct
* __init
fork_idle(int cpu
)
2713 struct task_struct
*task
;
2714 struct kernel_clone_args args
= {
2722 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2723 if (!IS_ERR(task
)) {
2724 init_idle_pids(task
);
2725 init_idle(task
, cpu
);
2732 * This is like kernel_clone(), but shaved down and tailored to just
2733 * creating io_uring workers. It returns a created task, or an error pointer.
2734 * The returned task is inactive, and the caller must fire it up through
2735 * wake_up_new_task(p). All signals are blocked in the created task.
2737 struct task_struct
*create_io_thread(int (*fn
)(void *), void *arg
, int node
)
2739 unsigned long flags
= CLONE_FS
|CLONE_FILES
|CLONE_SIGHAND
|CLONE_THREAD
|
2741 struct kernel_clone_args args
= {
2742 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2743 CLONE_UNTRACED
) & ~CSIGNAL
),
2744 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2751 return copy_process(NULL
, 0, node
, &args
);
2755 * Ok, this is the main fork-routine.
2757 * It copies the process, and if successful kick-starts
2758 * it and waits for it to finish using the VM if required.
2760 * args->exit_signal is expected to be checked for sanity by the caller.
2762 pid_t
kernel_clone(struct kernel_clone_args
*args
)
2764 u64 clone_flags
= args
->flags
;
2765 struct completion vfork
;
2767 struct task_struct
*p
;
2772 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2773 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2774 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2775 * field in struct clone_args and it still doesn't make sense to have
2776 * them both point at the same memory location. Performing this check
2777 * here has the advantage that we don't need to have a separate helper
2778 * to check for legacy clone().
2780 if ((clone_flags
& CLONE_PIDFD
) &&
2781 (clone_flags
& CLONE_PARENT_SETTID
) &&
2782 (args
->pidfd
== args
->parent_tid
))
2786 * Determine whether and which event to report to ptracer. When
2787 * called from kernel_thread or CLONE_UNTRACED is explicitly
2788 * requested, no event is reported; otherwise, report if the event
2789 * for the type of forking is enabled.
2791 if (!(clone_flags
& CLONE_UNTRACED
)) {
2792 if (clone_flags
& CLONE_VFORK
)
2793 trace
= PTRACE_EVENT_VFORK
;
2794 else if (args
->exit_signal
!= SIGCHLD
)
2795 trace
= PTRACE_EVENT_CLONE
;
2797 trace
= PTRACE_EVENT_FORK
;
2799 if (likely(!ptrace_event_enabled(current
, trace
)))
2803 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2804 add_latent_entropy();
2810 * Do this prior waking up the new thread - the thread pointer
2811 * might get invalid after that point, if the thread exits quickly.
2813 trace_sched_process_fork(current
, p
);
2815 pid
= get_task_pid(p
, PIDTYPE_PID
);
2818 if (clone_flags
& CLONE_PARENT_SETTID
)
2819 put_user(nr
, args
->parent_tid
);
2821 if (clone_flags
& CLONE_VFORK
) {
2822 p
->vfork_done
= &vfork
;
2823 init_completion(&vfork
);
2827 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU
) && !(clone_flags
& CLONE_VM
)) {
2828 /* lock the task to synchronize with memcg migration */
2830 lru_gen_add_mm(p
->mm
);
2834 wake_up_new_task(p
);
2836 /* forking complete and child started to run, tell ptracer */
2837 if (unlikely(trace
))
2838 ptrace_event_pid(trace
, pid
);
2840 if (clone_flags
& CLONE_VFORK
) {
2841 if (!wait_for_vfork_done(p
, &vfork
))
2842 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2850 * Create a kernel thread.
2852 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, const char *name
,
2853 unsigned long flags
)
2855 struct kernel_clone_args args
= {
2856 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2857 CLONE_UNTRACED
) & ~CSIGNAL
),
2858 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2865 return kernel_clone(&args
);
2869 * Create a user mode thread.
2871 pid_t
user_mode_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2873 struct kernel_clone_args args
= {
2874 .flags
= ((lower_32_bits(flags
) | CLONE_VM
|
2875 CLONE_UNTRACED
) & ~CSIGNAL
),
2876 .exit_signal
= (lower_32_bits(flags
) & CSIGNAL
),
2881 return kernel_clone(&args
);
2884 #ifdef __ARCH_WANT_SYS_FORK
2885 SYSCALL_DEFINE0(fork
)
2888 struct kernel_clone_args args
= {
2889 .exit_signal
= SIGCHLD
,
2892 return kernel_clone(&args
);
2894 /* can not support in nommu mode */
2900 #ifdef __ARCH_WANT_SYS_VFORK
2901 SYSCALL_DEFINE0(vfork
)
2903 struct kernel_clone_args args
= {
2904 .flags
= CLONE_VFORK
| CLONE_VM
,
2905 .exit_signal
= SIGCHLD
,
2908 return kernel_clone(&args
);
2912 #ifdef __ARCH_WANT_SYS_CLONE
2913 #ifdef CONFIG_CLONE_BACKWARDS
2914 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2915 int __user
*, parent_tidptr
,
2917 int __user
*, child_tidptr
)
2918 #elif defined(CONFIG_CLONE_BACKWARDS2)
2919 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2920 int __user
*, parent_tidptr
,
2921 int __user
*, child_tidptr
,
2923 #elif defined(CONFIG_CLONE_BACKWARDS3)
2924 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2926 int __user
*, parent_tidptr
,
2927 int __user
*, child_tidptr
,
2930 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2931 int __user
*, parent_tidptr
,
2932 int __user
*, child_tidptr
,
2936 struct kernel_clone_args args
= {
2937 .flags
= (lower_32_bits(clone_flags
) & ~CSIGNAL
),
2938 .pidfd
= parent_tidptr
,
2939 .child_tid
= child_tidptr
,
2940 .parent_tid
= parent_tidptr
,
2941 .exit_signal
= (lower_32_bits(clone_flags
) & CSIGNAL
),
2946 return kernel_clone(&args
);
2950 #ifdef __ARCH_WANT_SYS_CLONE3
2952 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2953 struct clone_args __user
*uargs
,
2957 struct clone_args args
;
2958 pid_t
*kset_tid
= kargs
->set_tid
;
2960 BUILD_BUG_ON(offsetofend(struct clone_args
, tls
) !=
2961 CLONE_ARGS_SIZE_VER0
);
2962 BUILD_BUG_ON(offsetofend(struct clone_args
, set_tid_size
) !=
2963 CLONE_ARGS_SIZE_VER1
);
2964 BUILD_BUG_ON(offsetofend(struct clone_args
, cgroup
) !=
2965 CLONE_ARGS_SIZE_VER2
);
2966 BUILD_BUG_ON(sizeof(struct clone_args
) != CLONE_ARGS_SIZE_VER2
);
2968 if (unlikely(usize
> PAGE_SIZE
))
2970 if (unlikely(usize
< CLONE_ARGS_SIZE_VER0
))
2973 err
= copy_struct_from_user(&args
, sizeof(args
), uargs
, usize
);
2977 if (unlikely(args
.set_tid_size
> MAX_PID_NS_LEVEL
))
2980 if (unlikely(!args
.set_tid
&& args
.set_tid_size
> 0))
2983 if (unlikely(args
.set_tid
&& args
.set_tid_size
== 0))
2987 * Verify that higher 32bits of exit_signal are unset and that
2988 * it is a valid signal
2990 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2991 !valid_signal(args
.exit_signal
)))
2994 if ((args
.flags
& CLONE_INTO_CGROUP
) &&
2995 (args
.cgroup
> INT_MAX
|| usize
< CLONE_ARGS_SIZE_VER2
))
2998 *kargs
= (struct kernel_clone_args
){
2999 .flags
= args
.flags
,
3000 .pidfd
= u64_to_user_ptr(args
.pidfd
),
3001 .child_tid
= u64_to_user_ptr(args
.child_tid
),
3002 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
3003 .exit_signal
= args
.exit_signal
,
3004 .stack
= args
.stack
,
3005 .stack_size
= args
.stack_size
,
3007 .set_tid_size
= args
.set_tid_size
,
3008 .cgroup
= args
.cgroup
,
3012 copy_from_user(kset_tid
, u64_to_user_ptr(args
.set_tid
),
3013 (kargs
->set_tid_size
* sizeof(pid_t
))))
3016 kargs
->set_tid
= kset_tid
;
3022 * clone3_stack_valid - check and prepare stack
3023 * @kargs: kernel clone args
3025 * Verify that the stack arguments userspace gave us are sane.
3026 * In addition, set the stack direction for userspace since it's easy for us to
3029 static inline bool clone3_stack_valid(struct kernel_clone_args
*kargs
)
3031 if (kargs
->stack
== 0) {
3032 if (kargs
->stack_size
> 0)
3035 if (kargs
->stack_size
== 0)
3038 if (!access_ok((void __user
*)kargs
->stack
, kargs
->stack_size
))
3041 #if !defined(CONFIG_STACK_GROWSUP)
3042 kargs
->stack
+= kargs
->stack_size
;
3049 static bool clone3_args_valid(struct kernel_clone_args
*kargs
)
3051 /* Verify that no unknown flags are passed along. */
3053 ~(CLONE_LEGACY_FLAGS
| CLONE_CLEAR_SIGHAND
| CLONE_INTO_CGROUP
))
3057 * - make the CLONE_DETACHED bit reusable for clone3
3058 * - make the CSIGNAL bits reusable for clone3
3060 if (kargs
->flags
& (CLONE_DETACHED
| (CSIGNAL
& (~CLONE_NEWTIME
))))
3063 if ((kargs
->flags
& (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
)) ==
3064 (CLONE_SIGHAND
| CLONE_CLEAR_SIGHAND
))
3067 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
3071 if (!clone3_stack_valid(kargs
))
3078 * sys_clone3 - create a new process with specific properties
3079 * @uargs: argument structure
3080 * @size: size of @uargs
3082 * clone3() is the extensible successor to clone()/clone2().
3083 * It takes a struct as argument that is versioned by its size.
3085 * Return: On success, a positive PID for the child process.
3086 * On error, a negative errno number.
3088 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
3092 struct kernel_clone_args kargs
;
3093 pid_t set_tid
[MAX_PID_NS_LEVEL
];
3095 kargs
.set_tid
= set_tid
;
3097 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
3101 if (!clone3_args_valid(&kargs
))
3104 return kernel_clone(&kargs
);
3108 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
3110 struct task_struct
*leader
, *parent
, *child
;
3113 read_lock(&tasklist_lock
);
3114 leader
= top
= top
->group_leader
;
3116 for_each_thread(leader
, parent
) {
3117 list_for_each_entry(child
, &parent
->children
, sibling
) {
3118 res
= visitor(child
, data
);
3130 if (leader
!= top
) {
3132 parent
= child
->real_parent
;
3133 leader
= parent
->group_leader
;
3137 read_unlock(&tasklist_lock
);
3140 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3141 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3144 static void sighand_ctor(void *data
)
3146 struct sighand_struct
*sighand
= data
;
3148 spin_lock_init(&sighand
->siglock
);
3149 init_waitqueue_head(&sighand
->signalfd_wqh
);
3152 void __init
mm_cache_init(void)
3154 unsigned int mm_size
;
3157 * The mm_cpumask is located at the end of mm_struct, and is
3158 * dynamically sized based on the maximum CPU number this system
3159 * can have, taking hotplug into account (nr_cpu_ids).
3161 mm_size
= sizeof(struct mm_struct
) + cpumask_size() + mm_cid_size();
3163 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
3164 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
3165 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3166 offsetof(struct mm_struct
, saved_auxv
),
3167 sizeof_field(struct mm_struct
, saved_auxv
),
3171 void __init
proc_caches_init(void)
3173 sighand_cachep
= kmem_cache_create("sighand_cache",
3174 sizeof(struct sighand_struct
), 0,
3175 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
3176 SLAB_ACCOUNT
, sighand_ctor
);
3177 signal_cachep
= kmem_cache_create("signal_cache",
3178 sizeof(struct signal_struct
), 0,
3179 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3181 files_cachep
= kmem_cache_create("files_cache",
3182 sizeof(struct files_struct
), 0,
3183 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3185 fs_cachep
= kmem_cache_create("fs_cache",
3186 sizeof(struct fs_struct
), 0,
3187 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
3190 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
3191 #ifdef CONFIG_PER_VMA_LOCK
3192 vma_lock_cachep
= KMEM_CACHE(vma_lock
, SLAB_PANIC
|SLAB_ACCOUNT
);
3195 nsproxy_cache_init();
3199 * Check constraints on flags passed to the unshare system call.
3201 static int check_unshare_flags(unsigned long unshare_flags
)
3203 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
3204 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
3205 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
3206 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
|
3210 * Not implemented, but pretend it works if there is nothing
3211 * to unshare. Note that unsharing the address space or the
3212 * signal handlers also need to unshare the signal queues (aka
3215 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
3216 if (!thread_group_empty(current
))
3219 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
3220 if (refcount_read(¤t
->sighand
->count
) > 1)
3223 if (unshare_flags
& CLONE_VM
) {
3224 if (!current_is_single_threaded())
3232 * Unshare the filesystem structure if it is being shared
3234 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
3236 struct fs_struct
*fs
= current
->fs
;
3238 if (!(unshare_flags
& CLONE_FS
) || !fs
)
3241 /* don't need lock here; in the worst case we'll do useless copy */
3245 *new_fsp
= copy_fs_struct(fs
);
3253 * Unshare file descriptor table if it is being shared
3255 int unshare_fd(unsigned long unshare_flags
, unsigned int max_fds
,
3256 struct files_struct
**new_fdp
)
3258 struct files_struct
*fd
= current
->files
;
3261 if ((unshare_flags
& CLONE_FILES
) &&
3262 (fd
&& atomic_read(&fd
->count
) > 1)) {
3263 *new_fdp
= dup_fd(fd
, max_fds
, &error
);
3272 * unshare allows a process to 'unshare' part of the process
3273 * context which was originally shared using clone. copy_*
3274 * functions used by kernel_clone() cannot be used here directly
3275 * because they modify an inactive task_struct that is being
3276 * constructed. Here we are modifying the current, active,
3279 int ksys_unshare(unsigned long unshare_flags
)
3281 struct fs_struct
*fs
, *new_fs
= NULL
;
3282 struct files_struct
*new_fd
= NULL
;
3283 struct cred
*new_cred
= NULL
;
3284 struct nsproxy
*new_nsproxy
= NULL
;
3289 * If unsharing a user namespace must also unshare the thread group
3290 * and unshare the filesystem root and working directories.
3292 if (unshare_flags
& CLONE_NEWUSER
)
3293 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
3295 * If unsharing vm, must also unshare signal handlers.
3297 if (unshare_flags
& CLONE_VM
)
3298 unshare_flags
|= CLONE_SIGHAND
;
3300 * If unsharing a signal handlers, must also unshare the signal queues.
3302 if (unshare_flags
& CLONE_SIGHAND
)
3303 unshare_flags
|= CLONE_THREAD
;
3305 * If unsharing namespace, must also unshare filesystem information.
3307 if (unshare_flags
& CLONE_NEWNS
)
3308 unshare_flags
|= CLONE_FS
;
3310 err
= check_unshare_flags(unshare_flags
);
3312 goto bad_unshare_out
;
3314 * CLONE_NEWIPC must also detach from the undolist: after switching
3315 * to a new ipc namespace, the semaphore arrays from the old
3316 * namespace are unreachable.
3318 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
3320 err
= unshare_fs(unshare_flags
, &new_fs
);
3322 goto bad_unshare_out
;
3323 err
= unshare_fd(unshare_flags
, NR_OPEN_MAX
, &new_fd
);
3325 goto bad_unshare_cleanup_fs
;
3326 err
= unshare_userns(unshare_flags
, &new_cred
);
3328 goto bad_unshare_cleanup_fd
;
3329 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
3332 goto bad_unshare_cleanup_cred
;
3335 err
= set_cred_ucounts(new_cred
);
3337 goto bad_unshare_cleanup_cred
;
3340 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
3343 * CLONE_SYSVSEM is equivalent to sys_exit().
3347 if (unshare_flags
& CLONE_NEWIPC
) {
3348 /* Orphan segments in old ns (see sem above). */
3350 shm_init_task(current
);
3354 switch_task_namespaces(current
, new_nsproxy
);
3360 spin_lock(&fs
->lock
);
3361 current
->fs
= new_fs
;
3366 spin_unlock(&fs
->lock
);
3370 swap(current
->files
, new_fd
);
3372 task_unlock(current
);
3375 /* Install the new user namespace */
3376 commit_creds(new_cred
);
3381 perf_event_namespaces(current
);
3383 bad_unshare_cleanup_cred
:
3386 bad_unshare_cleanup_fd
:
3388 put_files_struct(new_fd
);
3390 bad_unshare_cleanup_fs
:
3392 free_fs_struct(new_fs
);
3398 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
3400 return ksys_unshare(unshare_flags
);
3404 * Helper to unshare the files of the current task.
3405 * We don't want to expose copy_files internals to
3406 * the exec layer of the kernel.
3409 int unshare_files(void)
3411 struct task_struct
*task
= current
;
3412 struct files_struct
*old
, *copy
= NULL
;
3415 error
= unshare_fd(CLONE_FILES
, NR_OPEN_MAX
, ©
);
3423 put_files_struct(old
);
3427 int sysctl_max_threads(struct ctl_table
*table
, int write
,
3428 void *buffer
, size_t *lenp
, loff_t
*ppos
)
3432 int threads
= max_threads
;
3434 int max
= MAX_THREADS
;
3441 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
3445 max_threads
= threads
;