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>
111 * Minimum number of threads to boot the kernel
113 #define MIN_THREADS 20
116 * Maximum number of threads
118 #define MAX_THREADS FUTEX_TID_MASK
121 * Protected counters by write_lock_irq(&tasklist_lock)
123 unsigned long total_forks
; /* Handle normal Linux uptimes. */
124 int nr_threads
; /* The idle threads do not count.. */
126 static int max_threads
; /* tunable limit on nr_threads */
128 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
130 static const char * const resident_page_types
[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES
),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES
),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS
),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES
),
137 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
139 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
141 #ifdef CONFIG_PROVE_RCU
142 int lockdep_tasklist_lock_is_held(void)
144 return lockdep_is_held(&tasklist_lock
);
146 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
147 #endif /* #ifdef CONFIG_PROVE_RCU */
149 int nr_processes(void)
154 for_each_possible_cpu(cpu
)
155 total
+= per_cpu(process_counts
, cpu
);
160 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
164 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165 static struct kmem_cache
*task_struct_cachep
;
167 static inline struct task_struct
*alloc_task_struct_node(int node
)
169 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
172 static inline void free_task_struct(struct task_struct
*tsk
)
174 kmem_cache_free(task_struct_cachep
, tsk
);
178 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
184 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 #ifdef CONFIG_VMAP_STACK
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
191 #define NR_CACHED_STACKS 2
192 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
194 static int free_vm_stack_cache(unsigned int cpu
)
196 struct vm_struct
**cached_vm_stacks
= per_cpu_ptr(cached_stacks
, cpu
);
199 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
200 struct vm_struct
*vm_stack
= cached_vm_stacks
[i
];
205 vfree(vm_stack
->addr
);
206 cached_vm_stacks
[i
] = NULL
;
213 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
215 #ifdef CONFIG_VMAP_STACK
219 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
222 s
= this_cpu_xchg(cached_stacks
[i
], NULL
);
227 /* Clear stale pointers from reused stack. */
228 memset(s
->addr
, 0, THREAD_SIZE
);
230 tsk
->stack_vm_area
= s
;
231 tsk
->stack
= s
->addr
;
236 * Allocated stacks are cached and later reused by new threads,
237 * so memcg accounting is performed manually on assigning/releasing
238 * stacks to tasks. Drop __GFP_ACCOUNT.
240 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_ALIGN
,
241 VMALLOC_START
, VMALLOC_END
,
242 THREADINFO_GFP
& ~__GFP_ACCOUNT
,
244 0, node
, __builtin_return_address(0));
247 * We can't call find_vm_area() in interrupt context, and
248 * free_thread_stack() can be called in interrupt context,
249 * so cache the vm_struct.
252 tsk
->stack_vm_area
= find_vm_area(stack
);
257 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
261 tsk
->stack
= page_address(page
);
268 static inline void free_thread_stack(struct task_struct
*tsk
)
270 #ifdef CONFIG_VMAP_STACK
271 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
276 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
277 mod_memcg_page_state(vm
->pages
[i
],
278 MEMCG_KERNEL_STACK_KB
,
279 -(int)(PAGE_SIZE
/ 1024));
281 memcg_kmem_uncharge(vm
->pages
[i
], 0);
284 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
285 if (this_cpu_cmpxchg(cached_stacks
[i
],
286 NULL
, tsk
->stack_vm_area
) != NULL
)
292 vfree_atomic(tsk
->stack
);
297 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
300 static struct kmem_cache
*thread_stack_cache
;
302 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
305 unsigned long *stack
;
306 stack
= kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
311 static void free_thread_stack(struct task_struct
*tsk
)
313 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
316 void thread_stack_cache_init(void)
318 thread_stack_cache
= kmem_cache_create_usercopy("thread_stack",
319 THREAD_SIZE
, THREAD_SIZE
, 0, 0,
321 BUG_ON(thread_stack_cache
== NULL
);
326 /* SLAB cache for signal_struct structures (tsk->signal) */
327 static struct kmem_cache
*signal_cachep
;
329 /* SLAB cache for sighand_struct structures (tsk->sighand) */
330 struct kmem_cache
*sighand_cachep
;
332 /* SLAB cache for files_struct structures (tsk->files) */
333 struct kmem_cache
*files_cachep
;
335 /* SLAB cache for fs_struct structures (tsk->fs) */
336 struct kmem_cache
*fs_cachep
;
338 /* SLAB cache for vm_area_struct structures */
339 static struct kmem_cache
*vm_area_cachep
;
341 /* SLAB cache for mm_struct structures (tsk->mm) */
342 static struct kmem_cache
*mm_cachep
;
344 struct vm_area_struct
*vm_area_alloc(struct mm_struct
*mm
)
346 struct vm_area_struct
*vma
;
348 vma
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
354 struct vm_area_struct
*vm_area_dup(struct vm_area_struct
*orig
)
356 struct vm_area_struct
*new = kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
360 INIT_LIST_HEAD(&new->anon_vma_chain
);
365 void vm_area_free(struct vm_area_struct
*vma
)
367 kmem_cache_free(vm_area_cachep
, vma
);
370 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
372 void *stack
= task_stack_page(tsk
);
373 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
375 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
380 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
382 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
383 mod_zone_page_state(page_zone(vm
->pages
[i
]),
385 PAGE_SIZE
/ 1024 * account
);
389 * All stack pages are in the same zone and belong to the
392 struct page
*first_page
= virt_to_page(stack
);
394 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
395 THREAD_SIZE
/ 1024 * account
);
397 mod_memcg_page_state(first_page
, MEMCG_KERNEL_STACK_KB
,
398 account
* (THREAD_SIZE
/ 1024));
402 static int memcg_charge_kernel_stack(struct task_struct
*tsk
)
404 #ifdef CONFIG_VMAP_STACK
405 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
411 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
413 * If memcg_kmem_charge() fails, page->mem_cgroup
414 * pointer is NULL, and both memcg_kmem_uncharge()
415 * and mod_memcg_page_state() in free_thread_stack()
416 * will ignore this page. So it's safe.
418 ret
= memcg_kmem_charge(vm
->pages
[i
], GFP_KERNEL
, 0);
422 mod_memcg_page_state(vm
->pages
[i
],
423 MEMCG_KERNEL_STACK_KB
,
431 static void release_task_stack(struct task_struct
*tsk
)
433 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
434 return; /* Better to leak the stack than to free prematurely */
436 account_kernel_stack(tsk
, -1);
437 free_thread_stack(tsk
);
439 #ifdef CONFIG_VMAP_STACK
440 tsk
->stack_vm_area
= NULL
;
444 #ifdef CONFIG_THREAD_INFO_IN_TASK
445 void put_task_stack(struct task_struct
*tsk
)
447 if (refcount_dec_and_test(&tsk
->stack_refcount
))
448 release_task_stack(tsk
);
452 void free_task(struct task_struct
*tsk
)
454 #ifndef CONFIG_THREAD_INFO_IN_TASK
456 * The task is finally done with both the stack and thread_info,
459 release_task_stack(tsk
);
462 * If the task had a separate stack allocation, it should be gone
465 WARN_ON_ONCE(refcount_read(&tsk
->stack_refcount
) != 0);
467 rt_mutex_debug_task_free(tsk
);
468 ftrace_graph_exit_task(tsk
);
469 put_seccomp_filter(tsk
);
470 arch_release_task_struct(tsk
);
471 if (tsk
->flags
& PF_KTHREAD
)
472 free_kthread_struct(tsk
);
473 free_task_struct(tsk
);
475 EXPORT_SYMBOL(free_task
);
478 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
479 struct mm_struct
*oldmm
)
481 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
482 struct rb_node
**rb_link
, *rb_parent
;
484 unsigned long charge
;
487 uprobe_start_dup_mmap();
488 if (down_write_killable(&oldmm
->mmap_sem
)) {
490 goto fail_uprobe_end
;
492 flush_cache_dup_mm(oldmm
);
493 uprobe_dup_mmap(oldmm
, mm
);
495 * Not linked in yet - no deadlock potential:
497 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
499 /* No ordering required: file already has been exposed. */
500 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
502 mm
->total_vm
= oldmm
->total_vm
;
503 mm
->data_vm
= oldmm
->data_vm
;
504 mm
->exec_vm
= oldmm
->exec_vm
;
505 mm
->stack_vm
= oldmm
->stack_vm
;
507 rb_link
= &mm
->mm_rb
.rb_node
;
510 retval
= ksm_fork(mm
, oldmm
);
513 retval
= khugepaged_fork(mm
, oldmm
);
518 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
521 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
522 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
527 * Don't duplicate many vmas if we've been oom-killed (for
530 if (fatal_signal_pending(current
)) {
534 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
535 unsigned long len
= vma_pages(mpnt
);
537 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
541 tmp
= vm_area_dup(mpnt
);
544 retval
= vma_dup_policy(mpnt
, tmp
);
546 goto fail_nomem_policy
;
548 retval
= dup_userfaultfd(tmp
, &uf
);
550 goto fail_nomem_anon_vma_fork
;
551 if (tmp
->vm_flags
& VM_WIPEONFORK
) {
552 /* VM_WIPEONFORK gets a clean slate in the child. */
553 tmp
->anon_vma
= NULL
;
554 if (anon_vma_prepare(tmp
))
555 goto fail_nomem_anon_vma_fork
;
556 } else if (anon_vma_fork(tmp
, mpnt
))
557 goto fail_nomem_anon_vma_fork
;
558 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
559 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
562 struct inode
*inode
= file_inode(file
);
563 struct address_space
*mapping
= file
->f_mapping
;
566 if (tmp
->vm_flags
& VM_DENYWRITE
)
567 atomic_dec(&inode
->i_writecount
);
568 i_mmap_lock_write(mapping
);
569 if (tmp
->vm_flags
& VM_SHARED
)
570 atomic_inc(&mapping
->i_mmap_writable
);
571 flush_dcache_mmap_lock(mapping
);
572 /* insert tmp into the share list, just after mpnt */
573 vma_interval_tree_insert_after(tmp
, mpnt
,
575 flush_dcache_mmap_unlock(mapping
);
576 i_mmap_unlock_write(mapping
);
580 * Clear hugetlb-related page reserves for children. This only
581 * affects MAP_PRIVATE mappings. Faults generated by the child
582 * are not guaranteed to succeed, even if read-only
584 if (is_vm_hugetlb_page(tmp
))
585 reset_vma_resv_huge_pages(tmp
);
588 * Link in the new vma and copy the page table entries.
591 pprev
= &tmp
->vm_next
;
595 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
596 rb_link
= &tmp
->vm_rb
.rb_right
;
597 rb_parent
= &tmp
->vm_rb
;
600 if (!(tmp
->vm_flags
& VM_WIPEONFORK
))
601 retval
= copy_page_range(mm
, oldmm
, mpnt
);
603 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
604 tmp
->vm_ops
->open(tmp
);
609 /* a new mm has just been created */
610 retval
= arch_dup_mmap(oldmm
, mm
);
612 up_write(&mm
->mmap_sem
);
614 up_write(&oldmm
->mmap_sem
);
615 dup_userfaultfd_complete(&uf
);
617 uprobe_end_dup_mmap();
619 fail_nomem_anon_vma_fork
:
620 mpol_put(vma_policy(tmp
));
625 vm_unacct_memory(charge
);
629 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
631 mm
->pgd
= pgd_alloc(mm
);
632 if (unlikely(!mm
->pgd
))
637 static inline void mm_free_pgd(struct mm_struct
*mm
)
639 pgd_free(mm
, mm
->pgd
);
642 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
644 down_write(&oldmm
->mmap_sem
);
645 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
646 up_write(&oldmm
->mmap_sem
);
649 #define mm_alloc_pgd(mm) (0)
650 #define mm_free_pgd(mm)
651 #endif /* CONFIG_MMU */
653 static void check_mm(struct mm_struct
*mm
)
657 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types
) != NR_MM_COUNTERS
,
658 "Please make sure 'struct resident_page_types[]' is updated as well");
660 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
661 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
664 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
665 mm
, resident_page_types
[i
], x
);
668 if (mm_pgtables_bytes(mm
))
669 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
670 mm_pgtables_bytes(mm
));
672 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
673 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
677 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
678 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
681 * Called when the last reference to the mm
682 * is dropped: either by a lazy thread or by
683 * mmput. Free the page directory and the mm.
685 void __mmdrop(struct mm_struct
*mm
)
687 BUG_ON(mm
== &init_mm
);
688 WARN_ON_ONCE(mm
== current
->mm
);
689 WARN_ON_ONCE(mm
== current
->active_mm
);
692 mmu_notifier_mm_destroy(mm
);
694 put_user_ns(mm
->user_ns
);
697 EXPORT_SYMBOL_GPL(__mmdrop
);
699 static void mmdrop_async_fn(struct work_struct
*work
)
701 struct mm_struct
*mm
;
703 mm
= container_of(work
, struct mm_struct
, async_put_work
);
707 static void mmdrop_async(struct mm_struct
*mm
)
709 if (unlikely(atomic_dec_and_test(&mm
->mm_count
))) {
710 INIT_WORK(&mm
->async_put_work
, mmdrop_async_fn
);
711 schedule_work(&mm
->async_put_work
);
715 static inline void free_signal_struct(struct signal_struct
*sig
)
717 taskstats_tgid_free(sig
);
718 sched_autogroup_exit(sig
);
720 * __mmdrop is not safe to call from softirq context on x86 due to
721 * pgd_dtor so postpone it to the async context
724 mmdrop_async(sig
->oom_mm
);
725 kmem_cache_free(signal_cachep
, sig
);
728 static inline void put_signal_struct(struct signal_struct
*sig
)
730 if (refcount_dec_and_test(&sig
->sigcnt
))
731 free_signal_struct(sig
);
734 void __put_task_struct(struct task_struct
*tsk
)
736 WARN_ON(!tsk
->exit_state
);
737 WARN_ON(refcount_read(&tsk
->usage
));
738 WARN_ON(tsk
== current
);
741 task_numa_free(tsk
, true);
742 security_task_free(tsk
);
744 delayacct_tsk_free(tsk
);
745 put_signal_struct(tsk
->signal
);
747 if (!profile_handoff_task(tsk
))
750 EXPORT_SYMBOL_GPL(__put_task_struct
);
752 void __init __weak
arch_task_cache_init(void) { }
757 static void set_max_threads(unsigned int max_threads_suggested
)
760 unsigned long nr_pages
= totalram_pages();
763 * The number of threads shall be limited such that the thread
764 * structures may only consume a small part of the available memory.
766 if (fls64(nr_pages
) + fls64(PAGE_SIZE
) > 64)
767 threads
= MAX_THREADS
;
769 threads
= div64_u64((u64
) nr_pages
* (u64
) PAGE_SIZE
,
770 (u64
) THREAD_SIZE
* 8UL);
772 if (threads
> max_threads_suggested
)
773 threads
= max_threads_suggested
;
775 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
778 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
779 /* Initialized by the architecture: */
780 int arch_task_struct_size __read_mostly
;
783 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
784 static void task_struct_whitelist(unsigned long *offset
, unsigned long *size
)
786 /* Fetch thread_struct whitelist for the architecture. */
787 arch_thread_struct_whitelist(offset
, size
);
790 * Handle zero-sized whitelist or empty thread_struct, otherwise
791 * adjust offset to position of thread_struct in task_struct.
793 if (unlikely(*size
== 0))
796 *offset
+= offsetof(struct task_struct
, thread
);
798 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
800 void __init
fork_init(void)
803 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
804 #ifndef ARCH_MIN_TASKALIGN
805 #define ARCH_MIN_TASKALIGN 0
807 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
808 unsigned long useroffset
, usersize
;
810 /* create a slab on which task_structs can be allocated */
811 task_struct_whitelist(&useroffset
, &usersize
);
812 task_struct_cachep
= kmem_cache_create_usercopy("task_struct",
813 arch_task_struct_size
, align
,
814 SLAB_PANIC
|SLAB_ACCOUNT
,
815 useroffset
, usersize
, NULL
);
818 /* do the arch specific task caches init */
819 arch_task_cache_init();
821 set_max_threads(MAX_THREADS
);
823 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
824 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
825 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
826 init_task
.signal
->rlim
[RLIMIT_NPROC
];
828 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
829 init_user_ns
.ucount_max
[i
] = max_threads
/2;
832 #ifdef CONFIG_VMAP_STACK
833 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN
, "fork:vm_stack_cache",
834 NULL
, free_vm_stack_cache
);
837 lockdep_init_task(&init_task
);
841 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
842 struct task_struct
*src
)
848 void set_task_stack_end_magic(struct task_struct
*tsk
)
850 unsigned long *stackend
;
852 stackend
= end_of_stack(tsk
);
853 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
856 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
858 struct task_struct
*tsk
;
859 unsigned long *stack
;
860 struct vm_struct
*stack_vm_area __maybe_unused
;
863 if (node
== NUMA_NO_NODE
)
864 node
= tsk_fork_get_node(orig
);
865 tsk
= alloc_task_struct_node(node
);
869 stack
= alloc_thread_stack_node(tsk
, node
);
873 if (memcg_charge_kernel_stack(tsk
))
876 stack_vm_area
= task_stack_vm_area(tsk
);
878 err
= arch_dup_task_struct(tsk
, orig
);
881 * arch_dup_task_struct() clobbers the stack-related fields. Make
882 * sure they're properly initialized before using any stack-related
886 #ifdef CONFIG_VMAP_STACK
887 tsk
->stack_vm_area
= stack_vm_area
;
889 #ifdef CONFIG_THREAD_INFO_IN_TASK
890 refcount_set(&tsk
->stack_refcount
, 1);
896 #ifdef CONFIG_SECCOMP
898 * We must handle setting up seccomp filters once we're under
899 * the sighand lock in case orig has changed between now and
900 * then. Until then, filter must be NULL to avoid messing up
901 * the usage counts on the error path calling free_task.
903 tsk
->seccomp
.filter
= NULL
;
906 setup_thread_stack(tsk
, orig
);
907 clear_user_return_notifier(tsk
);
908 clear_tsk_need_resched(tsk
);
909 set_task_stack_end_magic(tsk
);
911 #ifdef CONFIG_STACKPROTECTOR
912 tsk
->stack_canary
= get_random_canary();
914 if (orig
->cpus_ptr
== &orig
->cpus_mask
)
915 tsk
->cpus_ptr
= &tsk
->cpus_mask
;
918 * One for us, one for whoever does the "release_task()" (usually
921 refcount_set(&tsk
->usage
, 2);
922 #ifdef CONFIG_BLK_DEV_IO_TRACE
925 tsk
->splice_pipe
= NULL
;
926 tsk
->task_frag
.page
= NULL
;
927 tsk
->wake_q
.next
= NULL
;
929 account_kernel_stack(tsk
, 1);
933 #ifdef CONFIG_FAULT_INJECTION
937 #ifdef CONFIG_BLK_CGROUP
938 tsk
->throttle_queue
= NULL
;
939 tsk
->use_memdelay
= 0;
943 tsk
->active_memcg
= NULL
;
948 free_thread_stack(tsk
);
950 free_task_struct(tsk
);
954 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
956 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
958 static int __init
coredump_filter_setup(char *s
)
960 default_dump_filter
=
961 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
962 MMF_DUMP_FILTER_MASK
;
966 __setup("coredump_filter=", coredump_filter_setup
);
968 #include <linux/init_task.h>
970 static void mm_init_aio(struct mm_struct
*mm
)
973 spin_lock_init(&mm
->ioctx_lock
);
974 mm
->ioctx_table
= NULL
;
978 static __always_inline
void mm_clear_owner(struct mm_struct
*mm
,
979 struct task_struct
*p
)
983 WRITE_ONCE(mm
->owner
, NULL
);
987 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
994 static void mm_init_uprobes_state(struct mm_struct
*mm
)
996 #ifdef CONFIG_UPROBES
997 mm
->uprobes_state
.xol_area
= NULL
;
1001 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
1002 struct user_namespace
*user_ns
)
1005 mm
->mm_rb
= RB_ROOT
;
1006 mm
->vmacache_seqnum
= 0;
1007 atomic_set(&mm
->mm_users
, 1);
1008 atomic_set(&mm
->mm_count
, 1);
1009 init_rwsem(&mm
->mmap_sem
);
1010 INIT_LIST_HEAD(&mm
->mmlist
);
1011 mm
->core_state
= NULL
;
1012 mm_pgtables_bytes_init(mm
);
1015 atomic64_set(&mm
->pinned_vm
, 0);
1016 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
1017 spin_lock_init(&mm
->page_table_lock
);
1018 spin_lock_init(&mm
->arg_lock
);
1019 mm_init_cpumask(mm
);
1021 mm_init_owner(mm
, p
);
1022 RCU_INIT_POINTER(mm
->exe_file
, NULL
);
1023 mmu_notifier_mm_init(mm
);
1024 init_tlb_flush_pending(mm
);
1025 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1026 mm
->pmd_huge_pte
= NULL
;
1028 mm_init_uprobes_state(mm
);
1031 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
1032 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
1034 mm
->flags
= default_dump_filter
;
1038 if (mm_alloc_pgd(mm
))
1041 if (init_new_context(p
, mm
))
1042 goto fail_nocontext
;
1044 mm
->user_ns
= get_user_ns(user_ns
);
1055 * Allocate and initialize an mm_struct.
1057 struct mm_struct
*mm_alloc(void)
1059 struct mm_struct
*mm
;
1065 memset(mm
, 0, sizeof(*mm
));
1066 return mm_init(mm
, current
, current_user_ns());
1069 static inline void __mmput(struct mm_struct
*mm
)
1071 VM_BUG_ON(atomic_read(&mm
->mm_users
));
1073 uprobe_clear_state(mm
);
1076 khugepaged_exit(mm
); /* must run before exit_mmap */
1078 mm_put_huge_zero_page(mm
);
1079 set_mm_exe_file(mm
, NULL
);
1080 if (!list_empty(&mm
->mmlist
)) {
1081 spin_lock(&mmlist_lock
);
1082 list_del(&mm
->mmlist
);
1083 spin_unlock(&mmlist_lock
);
1086 module_put(mm
->binfmt
->module
);
1091 * Decrement the use count and release all resources for an mm.
1093 void mmput(struct mm_struct
*mm
)
1097 if (atomic_dec_and_test(&mm
->mm_users
))
1100 EXPORT_SYMBOL_GPL(mmput
);
1103 static void mmput_async_fn(struct work_struct
*work
)
1105 struct mm_struct
*mm
= container_of(work
, struct mm_struct
,
1111 void mmput_async(struct mm_struct
*mm
)
1113 if (atomic_dec_and_test(&mm
->mm_users
)) {
1114 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
1115 schedule_work(&mm
->async_put_work
);
1121 * set_mm_exe_file - change a reference to the mm's executable file
1123 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1125 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1126 * invocations: in mmput() nobody alive left, in execve task is single
1127 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1128 * mm->exe_file, but does so without using set_mm_exe_file() in order
1129 * to do avoid the need for any locks.
1131 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
1133 struct file
*old_exe_file
;
1136 * It is safe to dereference the exe_file without RCU as
1137 * this function is only called if nobody else can access
1138 * this mm -- see comment above for justification.
1140 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
1143 get_file(new_exe_file
);
1144 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
1150 * get_mm_exe_file - acquire a reference to the mm's executable file
1152 * Returns %NULL if mm has no associated executable file.
1153 * User must release file via fput().
1155 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
1157 struct file
*exe_file
;
1160 exe_file
= rcu_dereference(mm
->exe_file
);
1161 if (exe_file
&& !get_file_rcu(exe_file
))
1166 EXPORT_SYMBOL(get_mm_exe_file
);
1169 * get_task_exe_file - acquire a reference to the task's executable file
1171 * Returns %NULL if task's mm (if any) has no associated executable file or
1172 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1173 * User must release file via fput().
1175 struct file
*get_task_exe_file(struct task_struct
*task
)
1177 struct file
*exe_file
= NULL
;
1178 struct mm_struct
*mm
;
1183 if (!(task
->flags
& PF_KTHREAD
))
1184 exe_file
= get_mm_exe_file(mm
);
1189 EXPORT_SYMBOL(get_task_exe_file
);
1192 * get_task_mm - acquire a reference to the task's mm
1194 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1195 * this kernel workthread has transiently adopted a user mm with use_mm,
1196 * to do its AIO) is not set and if so returns a reference to it, after
1197 * bumping up the use count. User must release the mm via mmput()
1198 * after use. Typically used by /proc and ptrace.
1200 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1202 struct mm_struct
*mm
;
1207 if (task
->flags
& PF_KTHREAD
)
1215 EXPORT_SYMBOL_GPL(get_task_mm
);
1217 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1219 struct mm_struct
*mm
;
1222 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1224 return ERR_PTR(err
);
1226 mm
= get_task_mm(task
);
1227 if (mm
&& mm
!= current
->mm
&&
1228 !ptrace_may_access(task
, mode
)) {
1230 mm
= ERR_PTR(-EACCES
);
1232 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1237 static void complete_vfork_done(struct task_struct
*tsk
)
1239 struct completion
*vfork
;
1242 vfork
= tsk
->vfork_done
;
1243 if (likely(vfork
)) {
1244 tsk
->vfork_done
= NULL
;
1250 static int wait_for_vfork_done(struct task_struct
*child
,
1251 struct completion
*vfork
)
1255 freezer_do_not_count();
1256 cgroup_enter_frozen();
1257 killed
= wait_for_completion_killable(vfork
);
1258 cgroup_leave_frozen(false);
1263 child
->vfork_done
= NULL
;
1267 put_task_struct(child
);
1271 /* Please note the differences between mmput and mm_release.
1272 * mmput is called whenever we stop holding onto a mm_struct,
1273 * error success whatever.
1275 * mm_release is called after a mm_struct has been removed
1276 * from the current process.
1278 * This difference is important for error handling, when we
1279 * only half set up a mm_struct for a new process and need to restore
1280 * the old one. Because we mmput the new mm_struct before
1281 * restoring the old one. . .
1282 * Eric Biederman 10 January 1998
1284 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1286 /* Get rid of any futexes when releasing the mm */
1288 if (unlikely(tsk
->robust_list
)) {
1289 exit_robust_list(tsk
);
1290 tsk
->robust_list
= NULL
;
1292 #ifdef CONFIG_COMPAT
1293 if (unlikely(tsk
->compat_robust_list
)) {
1294 compat_exit_robust_list(tsk
);
1295 tsk
->compat_robust_list
= NULL
;
1298 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
1299 exit_pi_state_list(tsk
);
1302 uprobe_free_utask(tsk
);
1304 /* Get rid of any cached register state */
1305 deactivate_mm(tsk
, mm
);
1308 * Signal userspace if we're not exiting with a core dump
1309 * because we want to leave the value intact for debugging
1312 if (tsk
->clear_child_tid
) {
1313 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1314 atomic_read(&mm
->mm_users
) > 1) {
1316 * We don't check the error code - if userspace has
1317 * not set up a proper pointer then tough luck.
1319 put_user(0, tsk
->clear_child_tid
);
1320 do_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1321 1, NULL
, NULL
, 0, 0);
1323 tsk
->clear_child_tid
= NULL
;
1327 * All done, finally we can wake up parent and return this mm to him.
1328 * Also kthread_stop() uses this completion for synchronization.
1330 if (tsk
->vfork_done
)
1331 complete_vfork_done(tsk
);
1335 * dup_mm() - duplicates an existing mm structure
1336 * @tsk: the task_struct with which the new mm will be associated.
1337 * @oldmm: the mm to duplicate.
1339 * Allocates a new mm structure and duplicates the provided @oldmm structure
1342 * Return: the duplicated mm or NULL on failure.
1344 static struct mm_struct
*dup_mm(struct task_struct
*tsk
,
1345 struct mm_struct
*oldmm
)
1347 struct mm_struct
*mm
;
1354 memcpy(mm
, oldmm
, sizeof(*mm
));
1356 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1359 err
= dup_mmap(mm
, oldmm
);
1363 mm
->hiwater_rss
= get_mm_rss(mm
);
1364 mm
->hiwater_vm
= mm
->total_vm
;
1366 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1372 /* don't put binfmt in mmput, we haven't got module yet */
1374 mm_init_owner(mm
, NULL
);
1381 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1383 struct mm_struct
*mm
, *oldmm
;
1386 tsk
->min_flt
= tsk
->maj_flt
= 0;
1387 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1388 #ifdef CONFIG_DETECT_HUNG_TASK
1389 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1390 tsk
->last_switch_time
= 0;
1394 tsk
->active_mm
= NULL
;
1397 * Are we cloning a kernel thread?
1399 * We need to steal a active VM for that..
1401 oldmm
= current
->mm
;
1405 /* initialize the new vmacache entries */
1406 vmacache_flush(tsk
);
1408 if (clone_flags
& CLONE_VM
) {
1415 mm
= dup_mm(tsk
, current
->mm
);
1421 tsk
->active_mm
= mm
;
1428 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1430 struct fs_struct
*fs
= current
->fs
;
1431 if (clone_flags
& CLONE_FS
) {
1432 /* tsk->fs is already what we want */
1433 spin_lock(&fs
->lock
);
1435 spin_unlock(&fs
->lock
);
1439 spin_unlock(&fs
->lock
);
1442 tsk
->fs
= copy_fs_struct(fs
);
1448 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1450 struct files_struct
*oldf
, *newf
;
1454 * A background process may not have any files ...
1456 oldf
= current
->files
;
1460 if (clone_flags
& CLONE_FILES
) {
1461 atomic_inc(&oldf
->count
);
1465 newf
= dup_fd(oldf
, &error
);
1475 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1478 struct io_context
*ioc
= current
->io_context
;
1479 struct io_context
*new_ioc
;
1484 * Share io context with parent, if CLONE_IO is set
1486 if (clone_flags
& CLONE_IO
) {
1488 tsk
->io_context
= ioc
;
1489 } else if (ioprio_valid(ioc
->ioprio
)) {
1490 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1491 if (unlikely(!new_ioc
))
1494 new_ioc
->ioprio
= ioc
->ioprio
;
1495 put_io_context(new_ioc
);
1501 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1503 struct sighand_struct
*sig
;
1505 if (clone_flags
& CLONE_SIGHAND
) {
1506 refcount_inc(¤t
->sighand
->count
);
1509 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1510 rcu_assign_pointer(tsk
->sighand
, sig
);
1514 refcount_set(&sig
->count
, 1);
1515 spin_lock_irq(¤t
->sighand
->siglock
);
1516 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1517 spin_unlock_irq(¤t
->sighand
->siglock
);
1521 void __cleanup_sighand(struct sighand_struct
*sighand
)
1523 if (refcount_dec_and_test(&sighand
->count
)) {
1524 signalfd_cleanup(sighand
);
1526 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1527 * without an RCU grace period, see __lock_task_sighand().
1529 kmem_cache_free(sighand_cachep
, sighand
);
1534 * Initialize POSIX timer handling for a thread group.
1536 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1538 struct posix_cputimers
*pct
= &sig
->posix_cputimers
;
1539 unsigned long cpu_limit
;
1541 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1542 posix_cputimers_group_init(pct
, cpu_limit
);
1545 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1547 struct signal_struct
*sig
;
1549 if (clone_flags
& CLONE_THREAD
)
1552 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1557 sig
->nr_threads
= 1;
1558 atomic_set(&sig
->live
, 1);
1559 refcount_set(&sig
->sigcnt
, 1);
1561 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1562 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1563 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1565 init_waitqueue_head(&sig
->wait_chldexit
);
1566 sig
->curr_target
= tsk
;
1567 init_sigpending(&sig
->shared_pending
);
1568 INIT_HLIST_HEAD(&sig
->multiprocess
);
1569 seqlock_init(&sig
->stats_lock
);
1570 prev_cputime_init(&sig
->prev_cputime
);
1572 #ifdef CONFIG_POSIX_TIMERS
1573 INIT_LIST_HEAD(&sig
->posix_timers
);
1574 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1575 sig
->real_timer
.function
= it_real_fn
;
1578 task_lock(current
->group_leader
);
1579 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1580 task_unlock(current
->group_leader
);
1582 posix_cpu_timers_init_group(sig
);
1584 tty_audit_fork(sig
);
1585 sched_autogroup_fork(sig
);
1587 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1588 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1590 mutex_init(&sig
->cred_guard_mutex
);
1595 static void copy_seccomp(struct task_struct
*p
)
1597 #ifdef CONFIG_SECCOMP
1599 * Must be called with sighand->lock held, which is common to
1600 * all threads in the group. Holding cred_guard_mutex is not
1601 * needed because this new task is not yet running and cannot
1604 assert_spin_locked(¤t
->sighand
->siglock
);
1606 /* Ref-count the new filter user, and assign it. */
1607 get_seccomp_filter(current
);
1608 p
->seccomp
= current
->seccomp
;
1611 * Explicitly enable no_new_privs here in case it got set
1612 * between the task_struct being duplicated and holding the
1613 * sighand lock. The seccomp state and nnp must be in sync.
1615 if (task_no_new_privs(current
))
1616 task_set_no_new_privs(p
);
1619 * If the parent gained a seccomp mode after copying thread
1620 * flags and between before we held the sighand lock, we have
1621 * to manually enable the seccomp thread flag here.
1623 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1624 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1628 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1630 current
->clear_child_tid
= tidptr
;
1632 return task_pid_vnr(current
);
1635 static void rt_mutex_init_task(struct task_struct
*p
)
1637 raw_spin_lock_init(&p
->pi_lock
);
1638 #ifdef CONFIG_RT_MUTEXES
1639 p
->pi_waiters
= RB_ROOT_CACHED
;
1640 p
->pi_top_task
= NULL
;
1641 p
->pi_blocked_on
= NULL
;
1645 static inline void init_task_pid_links(struct task_struct
*task
)
1649 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1650 INIT_HLIST_NODE(&task
->pid_links
[type
]);
1655 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1657 if (type
== PIDTYPE_PID
)
1658 task
->thread_pid
= pid
;
1660 task
->signal
->pids
[type
] = pid
;
1663 static inline void rcu_copy_process(struct task_struct
*p
)
1665 #ifdef CONFIG_PREEMPT_RCU
1666 p
->rcu_read_lock_nesting
= 0;
1667 p
->rcu_read_unlock_special
.s
= 0;
1668 p
->rcu_blocked_node
= NULL
;
1669 INIT_LIST_HEAD(&p
->rcu_node_entry
);
1670 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1671 #ifdef CONFIG_TASKS_RCU
1672 p
->rcu_tasks_holdout
= false;
1673 INIT_LIST_HEAD(&p
->rcu_tasks_holdout_list
);
1674 p
->rcu_tasks_idle_cpu
= -1;
1675 #endif /* #ifdef CONFIG_TASKS_RCU */
1678 struct pid
*pidfd_pid(const struct file
*file
)
1680 if (file
->f_op
== &pidfd_fops
)
1681 return file
->private_data
;
1683 return ERR_PTR(-EBADF
);
1686 static int pidfd_release(struct inode
*inode
, struct file
*file
)
1688 struct pid
*pid
= file
->private_data
;
1690 file
->private_data
= NULL
;
1695 #ifdef CONFIG_PROC_FS
1696 static void pidfd_show_fdinfo(struct seq_file
*m
, struct file
*f
)
1698 struct pid_namespace
*ns
= proc_pid_ns(file_inode(m
->file
));
1699 struct pid
*pid
= f
->private_data
;
1701 seq_put_decimal_ull(m
, "Pid:\t", pid_nr_ns(pid
, ns
));
1707 * Poll support for process exit notification.
1709 static unsigned int pidfd_poll(struct file
*file
, struct poll_table_struct
*pts
)
1711 struct task_struct
*task
;
1712 struct pid
*pid
= file
->private_data
;
1715 poll_wait(file
, &pid
->wait_pidfd
, pts
);
1718 task
= pid_task(pid
, PIDTYPE_PID
);
1720 * Inform pollers only when the whole thread group exits.
1721 * If the thread group leader exits before all other threads in the
1722 * group, then poll(2) should block, similar to the wait(2) family.
1724 if (!task
|| (task
->exit_state
&& thread_group_empty(task
)))
1725 poll_flags
= POLLIN
| POLLRDNORM
;
1731 const struct file_operations pidfd_fops
= {
1732 .release
= pidfd_release
,
1734 #ifdef CONFIG_PROC_FS
1735 .show_fdinfo
= pidfd_show_fdinfo
,
1739 static void __delayed_free_task(struct rcu_head
*rhp
)
1741 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
1746 static __always_inline
void delayed_free_task(struct task_struct
*tsk
)
1748 if (IS_ENABLED(CONFIG_MEMCG
))
1749 call_rcu(&tsk
->rcu
, __delayed_free_task
);
1755 * This creates a new process as a copy of the old one,
1756 * but does not actually start it yet.
1758 * It copies the registers, and all the appropriate
1759 * parts of the process environment (as per the clone
1760 * flags). The actual kick-off is left to the caller.
1762 static __latent_entropy
struct task_struct
*copy_process(
1766 struct kernel_clone_args
*args
)
1768 int pidfd
= -1, retval
;
1769 struct task_struct
*p
;
1770 struct multiprocess_signals delayed
;
1771 struct file
*pidfile
= NULL
;
1772 u64 clone_flags
= args
->flags
;
1775 * Don't allow sharing the root directory with processes in a different
1778 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1779 return ERR_PTR(-EINVAL
);
1781 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1782 return ERR_PTR(-EINVAL
);
1785 * Thread groups must share signals as well, and detached threads
1786 * can only be started up within the thread group.
1788 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1789 return ERR_PTR(-EINVAL
);
1792 * Shared signal handlers imply shared VM. By way of the above,
1793 * thread groups also imply shared VM. Blocking this case allows
1794 * for various simplifications in other code.
1796 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1797 return ERR_PTR(-EINVAL
);
1800 * Siblings of global init remain as zombies on exit since they are
1801 * not reaped by their parent (swapper). To solve this and to avoid
1802 * multi-rooted process trees, prevent global and container-inits
1803 * from creating siblings.
1805 if ((clone_flags
& CLONE_PARENT
) &&
1806 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1807 return ERR_PTR(-EINVAL
);
1810 * If the new process will be in a different pid or user namespace
1811 * do not allow it to share a thread group with the forking task.
1813 if (clone_flags
& CLONE_THREAD
) {
1814 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1815 (task_active_pid_ns(current
) !=
1816 current
->nsproxy
->pid_ns_for_children
))
1817 return ERR_PTR(-EINVAL
);
1820 if (clone_flags
& CLONE_PIDFD
) {
1822 * - CLONE_DETACHED is blocked so that we can potentially
1823 * reuse it later for CLONE_PIDFD.
1824 * - CLONE_THREAD is blocked until someone really needs it.
1826 if (clone_flags
& (CLONE_DETACHED
| CLONE_THREAD
))
1827 return ERR_PTR(-EINVAL
);
1831 * Force any signals received before this point to be delivered
1832 * before the fork happens. Collect up signals sent to multiple
1833 * processes that happen during the fork and delay them so that
1834 * they appear to happen after the fork.
1836 sigemptyset(&delayed
.signal
);
1837 INIT_HLIST_NODE(&delayed
.node
);
1839 spin_lock_irq(¤t
->sighand
->siglock
);
1840 if (!(clone_flags
& CLONE_THREAD
))
1841 hlist_add_head(&delayed
.node
, ¤t
->signal
->multiprocess
);
1842 recalc_sigpending();
1843 spin_unlock_irq(¤t
->sighand
->siglock
);
1844 retval
= -ERESTARTNOINTR
;
1845 if (signal_pending(current
))
1849 p
= dup_task_struct(current
, node
);
1854 * This _must_ happen before we call free_task(), i.e. before we jump
1855 * to any of the bad_fork_* labels. This is to avoid freeing
1856 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1857 * kernel threads (PF_KTHREAD).
1859 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? args
->child_tid
: NULL
;
1861 * Clear TID on mm_release()?
1863 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? args
->child_tid
: NULL
;
1865 ftrace_graph_init_task(p
);
1867 rt_mutex_init_task(p
);
1869 #ifdef CONFIG_PROVE_LOCKING
1870 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1871 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1874 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1875 task_rlimit(p
, RLIMIT_NPROC
)) {
1876 if (p
->real_cred
->user
!= INIT_USER
&&
1877 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1880 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1882 retval
= copy_creds(p
, clone_flags
);
1887 * If multiple threads are within copy_process(), then this check
1888 * triggers too late. This doesn't hurt, the check is only there
1889 * to stop root fork bombs.
1892 if (nr_threads
>= max_threads
)
1893 goto bad_fork_cleanup_count
;
1895 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1896 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1897 p
->flags
|= PF_FORKNOEXEC
;
1898 INIT_LIST_HEAD(&p
->children
);
1899 INIT_LIST_HEAD(&p
->sibling
);
1900 rcu_copy_process(p
);
1901 p
->vfork_done
= NULL
;
1902 spin_lock_init(&p
->alloc_lock
);
1904 init_sigpending(&p
->pending
);
1906 p
->utime
= p
->stime
= p
->gtime
= 0;
1907 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1908 p
->utimescaled
= p
->stimescaled
= 0;
1910 prev_cputime_init(&p
->prev_cputime
);
1912 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1913 seqcount_init(&p
->vtime
.seqcount
);
1914 p
->vtime
.starttime
= 0;
1915 p
->vtime
.state
= VTIME_INACTIVE
;
1918 #if defined(SPLIT_RSS_COUNTING)
1919 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1922 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1928 task_io_accounting_init(&p
->ioac
);
1929 acct_clear_integrals(p
);
1931 posix_cputimers_init(&p
->posix_cputimers
);
1933 p
->io_context
= NULL
;
1934 audit_set_context(p
, NULL
);
1937 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1938 if (IS_ERR(p
->mempolicy
)) {
1939 retval
= PTR_ERR(p
->mempolicy
);
1940 p
->mempolicy
= NULL
;
1941 goto bad_fork_cleanup_threadgroup_lock
;
1944 #ifdef CONFIG_CPUSETS
1945 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1946 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1947 seqcount_init(&p
->mems_allowed_seq
);
1949 #ifdef CONFIG_TRACE_IRQFLAGS
1951 p
->hardirqs_enabled
= 0;
1952 p
->hardirq_enable_ip
= 0;
1953 p
->hardirq_enable_event
= 0;
1954 p
->hardirq_disable_ip
= _THIS_IP_
;
1955 p
->hardirq_disable_event
= 0;
1956 p
->softirqs_enabled
= 1;
1957 p
->softirq_enable_ip
= _THIS_IP_
;
1958 p
->softirq_enable_event
= 0;
1959 p
->softirq_disable_ip
= 0;
1960 p
->softirq_disable_event
= 0;
1961 p
->hardirq_context
= 0;
1962 p
->softirq_context
= 0;
1965 p
->pagefault_disabled
= 0;
1967 #ifdef CONFIG_LOCKDEP
1968 lockdep_init_task(p
);
1971 #ifdef CONFIG_DEBUG_MUTEXES
1972 p
->blocked_on
= NULL
; /* not blocked yet */
1974 #ifdef CONFIG_BCACHE
1975 p
->sequential_io
= 0;
1976 p
->sequential_io_avg
= 0;
1979 /* Perform scheduler related setup. Assign this task to a CPU. */
1980 retval
= sched_fork(clone_flags
, p
);
1982 goto bad_fork_cleanup_policy
;
1984 retval
= perf_event_init_task(p
);
1986 goto bad_fork_cleanup_policy
;
1987 retval
= audit_alloc(p
);
1989 goto bad_fork_cleanup_perf
;
1990 /* copy all the process information */
1992 retval
= security_task_alloc(p
, clone_flags
);
1994 goto bad_fork_cleanup_audit
;
1995 retval
= copy_semundo(clone_flags
, p
);
1997 goto bad_fork_cleanup_security
;
1998 retval
= copy_files(clone_flags
, p
);
2000 goto bad_fork_cleanup_semundo
;
2001 retval
= copy_fs(clone_flags
, p
);
2003 goto bad_fork_cleanup_files
;
2004 retval
= copy_sighand(clone_flags
, p
);
2006 goto bad_fork_cleanup_fs
;
2007 retval
= copy_signal(clone_flags
, p
);
2009 goto bad_fork_cleanup_sighand
;
2010 retval
= copy_mm(clone_flags
, p
);
2012 goto bad_fork_cleanup_signal
;
2013 retval
= copy_namespaces(clone_flags
, p
);
2015 goto bad_fork_cleanup_mm
;
2016 retval
= copy_io(clone_flags
, p
);
2018 goto bad_fork_cleanup_namespaces
;
2019 retval
= copy_thread_tls(clone_flags
, args
->stack
, args
->stack_size
, p
,
2022 goto bad_fork_cleanup_io
;
2024 stackleak_task_init(p
);
2026 if (pid
!= &init_struct_pid
) {
2027 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
2029 retval
= PTR_ERR(pid
);
2030 goto bad_fork_cleanup_thread
;
2035 * This has to happen after we've potentially unshared the file
2036 * descriptor table (so that the pidfd doesn't leak into the child
2037 * if the fd table isn't shared).
2039 if (clone_flags
& CLONE_PIDFD
) {
2040 retval
= get_unused_fd_flags(O_RDWR
| O_CLOEXEC
);
2042 goto bad_fork_free_pid
;
2046 pidfile
= anon_inode_getfile("[pidfd]", &pidfd_fops
, pid
,
2047 O_RDWR
| O_CLOEXEC
);
2048 if (IS_ERR(pidfile
)) {
2049 put_unused_fd(pidfd
);
2050 retval
= PTR_ERR(pidfile
);
2051 goto bad_fork_free_pid
;
2053 get_pid(pid
); /* held by pidfile now */
2055 retval
= put_user(pidfd
, args
->pidfd
);
2057 goto bad_fork_put_pidfd
;
2064 p
->robust_list
= NULL
;
2065 #ifdef CONFIG_COMPAT
2066 p
->compat_robust_list
= NULL
;
2068 INIT_LIST_HEAD(&p
->pi_state_list
);
2069 p
->pi_state_cache
= NULL
;
2072 * sigaltstack should be cleared when sharing the same VM
2074 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
2078 * Syscall tracing and stepping should be turned off in the
2079 * child regardless of CLONE_PTRACE.
2081 user_disable_single_step(p
);
2082 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
2083 #ifdef TIF_SYSCALL_EMU
2084 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
2086 clear_tsk_latency_tracing(p
);
2088 /* ok, now we should be set up.. */
2089 p
->pid
= pid_nr(pid
);
2090 if (clone_flags
& CLONE_THREAD
) {
2091 p
->exit_signal
= -1;
2092 p
->group_leader
= current
->group_leader
;
2093 p
->tgid
= current
->tgid
;
2095 if (clone_flags
& CLONE_PARENT
)
2096 p
->exit_signal
= current
->group_leader
->exit_signal
;
2098 p
->exit_signal
= args
->exit_signal
;
2099 p
->group_leader
= p
;
2104 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
2105 p
->dirty_paused_when
= 0;
2107 p
->pdeath_signal
= 0;
2108 INIT_LIST_HEAD(&p
->thread_group
);
2109 p
->task_works
= NULL
;
2111 cgroup_threadgroup_change_begin(current
);
2113 * Ensure that the cgroup subsystem policies allow the new process to be
2114 * forked. It should be noted the the new process's css_set can be changed
2115 * between here and cgroup_post_fork() if an organisation operation is in
2118 retval
= cgroup_can_fork(p
);
2120 goto bad_fork_cgroup_threadgroup_change_end
;
2123 * From this point on we must avoid any synchronous user-space
2124 * communication until we take the tasklist-lock. In particular, we do
2125 * not want user-space to be able to predict the process start-time by
2126 * stalling fork(2) after we recorded the start_time but before it is
2127 * visible to the system.
2130 p
->start_time
= ktime_get_ns();
2131 p
->real_start_time
= ktime_get_boottime_ns();
2134 * Make it visible to the rest of the system, but dont wake it up yet.
2135 * Need tasklist lock for parent etc handling!
2137 write_lock_irq(&tasklist_lock
);
2139 /* CLONE_PARENT re-uses the old parent */
2140 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
2141 p
->real_parent
= current
->real_parent
;
2142 p
->parent_exec_id
= current
->parent_exec_id
;
2144 p
->real_parent
= current
;
2145 p
->parent_exec_id
= current
->self_exec_id
;
2148 klp_copy_process(p
);
2150 spin_lock(¤t
->sighand
->siglock
);
2153 * Copy seccomp details explicitly here, in case they were changed
2154 * before holding sighand lock.
2158 rseq_fork(p
, clone_flags
);
2160 /* Don't start children in a dying pid namespace */
2161 if (unlikely(!(ns_of_pid(pid
)->pid_allocated
& PIDNS_ADDING
))) {
2163 goto bad_fork_cancel_cgroup
;
2166 /* Let kill terminate clone/fork in the middle */
2167 if (fatal_signal_pending(current
)) {
2169 goto bad_fork_cancel_cgroup
;
2172 /* past the last point of failure */
2174 fd_install(pidfd
, pidfile
);
2176 init_task_pid_links(p
);
2177 if (likely(p
->pid
)) {
2178 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
2180 init_task_pid(p
, PIDTYPE_PID
, pid
);
2181 if (thread_group_leader(p
)) {
2182 init_task_pid(p
, PIDTYPE_TGID
, pid
);
2183 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
2184 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
2186 if (is_child_reaper(pid
)) {
2187 ns_of_pid(pid
)->child_reaper
= p
;
2188 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
2190 p
->signal
->shared_pending
.signal
= delayed
.signal
;
2191 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
2193 * Inherit has_child_subreaper flag under the same
2194 * tasklist_lock with adding child to the process tree
2195 * for propagate_has_child_subreaper optimization.
2197 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
2198 p
->real_parent
->signal
->is_child_subreaper
;
2199 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
2200 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
2201 attach_pid(p
, PIDTYPE_TGID
);
2202 attach_pid(p
, PIDTYPE_PGID
);
2203 attach_pid(p
, PIDTYPE_SID
);
2204 __this_cpu_inc(process_counts
);
2206 current
->signal
->nr_threads
++;
2207 atomic_inc(¤t
->signal
->live
);
2208 refcount_inc(¤t
->signal
->sigcnt
);
2209 task_join_group_stop(p
);
2210 list_add_tail_rcu(&p
->thread_group
,
2211 &p
->group_leader
->thread_group
);
2212 list_add_tail_rcu(&p
->thread_node
,
2213 &p
->signal
->thread_head
);
2215 attach_pid(p
, PIDTYPE_PID
);
2219 hlist_del_init(&delayed
.node
);
2220 spin_unlock(¤t
->sighand
->siglock
);
2221 syscall_tracepoint_update(p
);
2222 write_unlock_irq(&tasklist_lock
);
2224 proc_fork_connector(p
);
2225 cgroup_post_fork(p
);
2226 cgroup_threadgroup_change_end(current
);
2229 trace_task_newtask(p
, clone_flags
);
2230 uprobe_copy_process(p
, clone_flags
);
2234 bad_fork_cancel_cgroup
:
2235 spin_unlock(¤t
->sighand
->siglock
);
2236 write_unlock_irq(&tasklist_lock
);
2237 cgroup_cancel_fork(p
);
2238 bad_fork_cgroup_threadgroup_change_end
:
2239 cgroup_threadgroup_change_end(current
);
2241 if (clone_flags
& CLONE_PIDFD
) {
2243 put_unused_fd(pidfd
);
2246 if (pid
!= &init_struct_pid
)
2248 bad_fork_cleanup_thread
:
2250 bad_fork_cleanup_io
:
2253 bad_fork_cleanup_namespaces
:
2254 exit_task_namespaces(p
);
2255 bad_fork_cleanup_mm
:
2257 mm_clear_owner(p
->mm
, p
);
2260 bad_fork_cleanup_signal
:
2261 if (!(clone_flags
& CLONE_THREAD
))
2262 free_signal_struct(p
->signal
);
2263 bad_fork_cleanup_sighand
:
2264 __cleanup_sighand(p
->sighand
);
2265 bad_fork_cleanup_fs
:
2266 exit_fs(p
); /* blocking */
2267 bad_fork_cleanup_files
:
2268 exit_files(p
); /* blocking */
2269 bad_fork_cleanup_semundo
:
2271 bad_fork_cleanup_security
:
2272 security_task_free(p
);
2273 bad_fork_cleanup_audit
:
2275 bad_fork_cleanup_perf
:
2276 perf_event_free_task(p
);
2277 bad_fork_cleanup_policy
:
2278 lockdep_free_task(p
);
2280 mpol_put(p
->mempolicy
);
2281 bad_fork_cleanup_threadgroup_lock
:
2283 delayacct_tsk_free(p
);
2284 bad_fork_cleanup_count
:
2285 atomic_dec(&p
->cred
->user
->processes
);
2288 p
->state
= TASK_DEAD
;
2290 delayed_free_task(p
);
2292 spin_lock_irq(¤t
->sighand
->siglock
);
2293 hlist_del_init(&delayed
.node
);
2294 spin_unlock_irq(¤t
->sighand
->siglock
);
2295 return ERR_PTR(retval
);
2298 static inline void init_idle_pids(struct task_struct
*idle
)
2302 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
2303 INIT_HLIST_NODE(&idle
->pid_links
[type
]); /* not really needed */
2304 init_task_pid(idle
, type
, &init_struct_pid
);
2308 struct task_struct
*fork_idle(int cpu
)
2310 struct task_struct
*task
;
2311 struct kernel_clone_args args
= {
2315 task
= copy_process(&init_struct_pid
, 0, cpu_to_node(cpu
), &args
);
2316 if (!IS_ERR(task
)) {
2317 init_idle_pids(task
);
2318 init_idle(task
, cpu
);
2324 struct mm_struct
*copy_init_mm(void)
2326 return dup_mm(NULL
, &init_mm
);
2330 * Ok, this is the main fork-routine.
2332 * It copies the process, and if successful kick-starts
2333 * it and waits for it to finish using the VM if required.
2335 * args->exit_signal is expected to be checked for sanity by the caller.
2337 long _do_fork(struct kernel_clone_args
*args
)
2339 u64 clone_flags
= args
->flags
;
2340 struct completion vfork
;
2342 struct task_struct
*p
;
2347 * Determine whether and which event to report to ptracer. When
2348 * called from kernel_thread or CLONE_UNTRACED is explicitly
2349 * requested, no event is reported; otherwise, report if the event
2350 * for the type of forking is enabled.
2352 if (!(clone_flags
& CLONE_UNTRACED
)) {
2353 if (clone_flags
& CLONE_VFORK
)
2354 trace
= PTRACE_EVENT_VFORK
;
2355 else if (args
->exit_signal
!= SIGCHLD
)
2356 trace
= PTRACE_EVENT_CLONE
;
2358 trace
= PTRACE_EVENT_FORK
;
2360 if (likely(!ptrace_event_enabled(current
, trace
)))
2364 p
= copy_process(NULL
, trace
, NUMA_NO_NODE
, args
);
2365 add_latent_entropy();
2371 * Do this prior waking up the new thread - the thread pointer
2372 * might get invalid after that point, if the thread exits quickly.
2374 trace_sched_process_fork(current
, p
);
2376 pid
= get_task_pid(p
, PIDTYPE_PID
);
2379 if (clone_flags
& CLONE_PARENT_SETTID
)
2380 put_user(nr
, args
->parent_tid
);
2382 if (clone_flags
& CLONE_VFORK
) {
2383 p
->vfork_done
= &vfork
;
2384 init_completion(&vfork
);
2388 wake_up_new_task(p
);
2390 /* forking complete and child started to run, tell ptracer */
2391 if (unlikely(trace
))
2392 ptrace_event_pid(trace
, pid
);
2394 if (clone_flags
& CLONE_VFORK
) {
2395 if (!wait_for_vfork_done(p
, &vfork
))
2396 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2403 bool legacy_clone_args_valid(const struct kernel_clone_args
*kargs
)
2405 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2406 if ((kargs
->flags
& CLONE_PIDFD
) &&
2407 (kargs
->flags
& CLONE_PARENT_SETTID
))
2413 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2414 /* For compatibility with architectures that call do_fork directly rather than
2415 * using the syscall entry points below. */
2416 long do_fork(unsigned long clone_flags
,
2417 unsigned long stack_start
,
2418 unsigned long stack_size
,
2419 int __user
*parent_tidptr
,
2420 int __user
*child_tidptr
)
2422 struct kernel_clone_args args
= {
2423 .flags
= (clone_flags
& ~CSIGNAL
),
2424 .pidfd
= parent_tidptr
,
2425 .child_tid
= child_tidptr
,
2426 .parent_tid
= parent_tidptr
,
2427 .exit_signal
= (clone_flags
& CSIGNAL
),
2428 .stack
= stack_start
,
2429 .stack_size
= stack_size
,
2432 if (!legacy_clone_args_valid(&args
))
2435 return _do_fork(&args
);
2440 * Create a kernel thread.
2442 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2444 struct kernel_clone_args args
= {
2445 .flags
= ((flags
| CLONE_VM
| CLONE_UNTRACED
) & ~CSIGNAL
),
2446 .exit_signal
= (flags
& CSIGNAL
),
2447 .stack
= (unsigned long)fn
,
2448 .stack_size
= (unsigned long)arg
,
2451 return _do_fork(&args
);
2454 #ifdef __ARCH_WANT_SYS_FORK
2455 SYSCALL_DEFINE0(fork
)
2458 struct kernel_clone_args args
= {
2459 .exit_signal
= SIGCHLD
,
2462 return _do_fork(&args
);
2464 /* can not support in nommu mode */
2470 #ifdef __ARCH_WANT_SYS_VFORK
2471 SYSCALL_DEFINE0(vfork
)
2473 struct kernel_clone_args args
= {
2474 .flags
= CLONE_VFORK
| CLONE_VM
,
2475 .exit_signal
= SIGCHLD
,
2478 return _do_fork(&args
);
2482 #ifdef __ARCH_WANT_SYS_CLONE
2483 #ifdef CONFIG_CLONE_BACKWARDS
2484 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2485 int __user
*, parent_tidptr
,
2487 int __user
*, child_tidptr
)
2488 #elif defined(CONFIG_CLONE_BACKWARDS2)
2489 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2490 int __user
*, parent_tidptr
,
2491 int __user
*, child_tidptr
,
2493 #elif defined(CONFIG_CLONE_BACKWARDS3)
2494 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2496 int __user
*, parent_tidptr
,
2497 int __user
*, child_tidptr
,
2500 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2501 int __user
*, parent_tidptr
,
2502 int __user
*, child_tidptr
,
2506 struct kernel_clone_args args
= {
2507 .flags
= (clone_flags
& ~CSIGNAL
),
2508 .pidfd
= parent_tidptr
,
2509 .child_tid
= child_tidptr
,
2510 .parent_tid
= parent_tidptr
,
2511 .exit_signal
= (clone_flags
& CSIGNAL
),
2516 if (!legacy_clone_args_valid(&args
))
2519 return _do_fork(&args
);
2523 #ifdef __ARCH_WANT_SYS_CLONE3
2524 noinline
static int copy_clone_args_from_user(struct kernel_clone_args
*kargs
,
2525 struct clone_args __user
*uargs
,
2528 struct clone_args args
;
2530 if (unlikely(size
> PAGE_SIZE
))
2533 if (unlikely(size
< sizeof(struct clone_args
)))
2536 if (unlikely(!access_ok(uargs
, size
)))
2539 if (size
> sizeof(struct clone_args
)) {
2540 unsigned char __user
*addr
;
2541 unsigned char __user
*end
;
2544 addr
= (void __user
*)uargs
+ sizeof(struct clone_args
);
2545 end
= (void __user
*)uargs
+ size
;
2547 for (; addr
< end
; addr
++) {
2548 if (get_user(val
, addr
))
2554 size
= sizeof(struct clone_args
);
2557 if (copy_from_user(&args
, uargs
, size
))
2561 * Verify that higher 32bits of exit_signal are unset and that
2562 * it is a valid signal
2564 if (unlikely((args
.exit_signal
& ~((u64
)CSIGNAL
)) ||
2565 !valid_signal(args
.exit_signal
)))
2568 *kargs
= (struct kernel_clone_args
){
2569 .flags
= args
.flags
,
2570 .pidfd
= u64_to_user_ptr(args
.pidfd
),
2571 .child_tid
= u64_to_user_ptr(args
.child_tid
),
2572 .parent_tid
= u64_to_user_ptr(args
.parent_tid
),
2573 .exit_signal
= args
.exit_signal
,
2574 .stack
= args
.stack
,
2575 .stack_size
= args
.stack_size
,
2582 static bool clone3_args_valid(const struct kernel_clone_args
*kargs
)
2585 * All lower bits of the flag word are taken.
2586 * Verify that no other unknown flags are passed along.
2588 if (kargs
->flags
& ~CLONE_LEGACY_FLAGS
)
2592 * - make the CLONE_DETACHED bit reuseable for clone3
2593 * - make the CSIGNAL bits reuseable for clone3
2595 if (kargs
->flags
& (CLONE_DETACHED
| CSIGNAL
))
2598 if ((kargs
->flags
& (CLONE_THREAD
| CLONE_PARENT
)) &&
2605 SYSCALL_DEFINE2(clone3
, struct clone_args __user
*, uargs
, size_t, size
)
2609 struct kernel_clone_args kargs
;
2611 err
= copy_clone_args_from_user(&kargs
, uargs
, size
);
2615 if (!clone3_args_valid(&kargs
))
2618 return _do_fork(&kargs
);
2622 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2624 struct task_struct
*leader
, *parent
, *child
;
2627 read_lock(&tasklist_lock
);
2628 leader
= top
= top
->group_leader
;
2630 for_each_thread(leader
, parent
) {
2631 list_for_each_entry(child
, &parent
->children
, sibling
) {
2632 res
= visitor(child
, data
);
2644 if (leader
!= top
) {
2646 parent
= child
->real_parent
;
2647 leader
= parent
->group_leader
;
2651 read_unlock(&tasklist_lock
);
2654 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2655 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2658 static void sighand_ctor(void *data
)
2660 struct sighand_struct
*sighand
= data
;
2662 spin_lock_init(&sighand
->siglock
);
2663 init_waitqueue_head(&sighand
->signalfd_wqh
);
2666 void __init
proc_caches_init(void)
2668 unsigned int mm_size
;
2670 sighand_cachep
= kmem_cache_create("sighand_cache",
2671 sizeof(struct sighand_struct
), 0,
2672 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_TYPESAFE_BY_RCU
|
2673 SLAB_ACCOUNT
, sighand_ctor
);
2674 signal_cachep
= kmem_cache_create("signal_cache",
2675 sizeof(struct signal_struct
), 0,
2676 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2678 files_cachep
= kmem_cache_create("files_cache",
2679 sizeof(struct files_struct
), 0,
2680 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2682 fs_cachep
= kmem_cache_create("fs_cache",
2683 sizeof(struct fs_struct
), 0,
2684 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2688 * The mm_cpumask is located at the end of mm_struct, and is
2689 * dynamically sized based on the maximum CPU number this system
2690 * can have, taking hotplug into account (nr_cpu_ids).
2692 mm_size
= sizeof(struct mm_struct
) + cpumask_size();
2694 mm_cachep
= kmem_cache_create_usercopy("mm_struct",
2695 mm_size
, ARCH_MIN_MMSTRUCT_ALIGN
,
2696 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
,
2697 offsetof(struct mm_struct
, saved_auxv
),
2698 sizeof_field(struct mm_struct
, saved_auxv
),
2700 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2702 nsproxy_cache_init();
2706 * Check constraints on flags passed to the unshare system call.
2708 static int check_unshare_flags(unsigned long unshare_flags
)
2710 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2711 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2712 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2713 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2716 * Not implemented, but pretend it works if there is nothing
2717 * to unshare. Note that unsharing the address space or the
2718 * signal handlers also need to unshare the signal queues (aka
2721 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2722 if (!thread_group_empty(current
))
2725 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2726 if (refcount_read(¤t
->sighand
->count
) > 1)
2729 if (unshare_flags
& CLONE_VM
) {
2730 if (!current_is_single_threaded())
2738 * Unshare the filesystem structure if it is being shared
2740 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2742 struct fs_struct
*fs
= current
->fs
;
2744 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2747 /* don't need lock here; in the worst case we'll do useless copy */
2751 *new_fsp
= copy_fs_struct(fs
);
2759 * Unshare file descriptor table if it is being shared
2761 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2763 struct files_struct
*fd
= current
->files
;
2766 if ((unshare_flags
& CLONE_FILES
) &&
2767 (fd
&& atomic_read(&fd
->count
) > 1)) {
2768 *new_fdp
= dup_fd(fd
, &error
);
2777 * unshare allows a process to 'unshare' part of the process
2778 * context which was originally shared using clone. copy_*
2779 * functions used by do_fork() cannot be used here directly
2780 * because they modify an inactive task_struct that is being
2781 * constructed. Here we are modifying the current, active,
2784 int ksys_unshare(unsigned long unshare_flags
)
2786 struct fs_struct
*fs
, *new_fs
= NULL
;
2787 struct files_struct
*fd
, *new_fd
= NULL
;
2788 struct cred
*new_cred
= NULL
;
2789 struct nsproxy
*new_nsproxy
= NULL
;
2794 * If unsharing a user namespace must also unshare the thread group
2795 * and unshare the filesystem root and working directories.
2797 if (unshare_flags
& CLONE_NEWUSER
)
2798 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2800 * If unsharing vm, must also unshare signal handlers.
2802 if (unshare_flags
& CLONE_VM
)
2803 unshare_flags
|= CLONE_SIGHAND
;
2805 * If unsharing a signal handlers, must also unshare the signal queues.
2807 if (unshare_flags
& CLONE_SIGHAND
)
2808 unshare_flags
|= CLONE_THREAD
;
2810 * If unsharing namespace, must also unshare filesystem information.
2812 if (unshare_flags
& CLONE_NEWNS
)
2813 unshare_flags
|= CLONE_FS
;
2815 err
= check_unshare_flags(unshare_flags
);
2817 goto bad_unshare_out
;
2819 * CLONE_NEWIPC must also detach from the undolist: after switching
2820 * to a new ipc namespace, the semaphore arrays from the old
2821 * namespace are unreachable.
2823 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2825 err
= unshare_fs(unshare_flags
, &new_fs
);
2827 goto bad_unshare_out
;
2828 err
= unshare_fd(unshare_flags
, &new_fd
);
2830 goto bad_unshare_cleanup_fs
;
2831 err
= unshare_userns(unshare_flags
, &new_cred
);
2833 goto bad_unshare_cleanup_fd
;
2834 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2837 goto bad_unshare_cleanup_cred
;
2839 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2842 * CLONE_SYSVSEM is equivalent to sys_exit().
2846 if (unshare_flags
& CLONE_NEWIPC
) {
2847 /* Orphan segments in old ns (see sem above). */
2849 shm_init_task(current
);
2853 switch_task_namespaces(current
, new_nsproxy
);
2859 spin_lock(&fs
->lock
);
2860 current
->fs
= new_fs
;
2865 spin_unlock(&fs
->lock
);
2869 fd
= current
->files
;
2870 current
->files
= new_fd
;
2874 task_unlock(current
);
2877 /* Install the new user namespace */
2878 commit_creds(new_cred
);
2883 perf_event_namespaces(current
);
2885 bad_unshare_cleanup_cred
:
2888 bad_unshare_cleanup_fd
:
2890 put_files_struct(new_fd
);
2892 bad_unshare_cleanup_fs
:
2894 free_fs_struct(new_fs
);
2900 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2902 return ksys_unshare(unshare_flags
);
2906 * Helper to unshare the files of the current task.
2907 * We don't want to expose copy_files internals to
2908 * the exec layer of the kernel.
2911 int unshare_files(struct files_struct
**displaced
)
2913 struct task_struct
*task
= current
;
2914 struct files_struct
*copy
= NULL
;
2917 error
= unshare_fd(CLONE_FILES
, ©
);
2918 if (error
|| !copy
) {
2922 *displaced
= task
->files
;
2929 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2930 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2934 int threads
= max_threads
;
2935 int min
= MIN_THREADS
;
2936 int max
= MAX_THREADS
;
2943 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2947 set_max_threads(threads
);