4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * 'fork.c' contains the help-routines for the 'fork' system call
9 * (see also entry.S and others).
10 * Fork is rather simple, once you get the hang of it, but the memory
11 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
14 #include <linux/slab.h>
15 #include <linux/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/numa_balancing.h>
20 #include <linux/rtmutex.h>
21 #include <linux/init.h>
22 #include <linux/unistd.h>
23 #include <linux/module.h>
24 #include <linux/vmalloc.h>
25 #include <linux/completion.h>
26 #include <linux/personality.h>
27 #include <linux/mempolicy.h>
28 #include <linux/sem.h>
29 #include <linux/file.h>
30 #include <linux/fdtable.h>
31 #include <linux/iocontext.h>
32 #include <linux/key.h>
33 #include <linux/binfmts.h>
34 #include <linux/mman.h>
35 #include <linux/mmu_notifier.h>
38 #include <linux/vmacache.h>
39 #include <linux/nsproxy.h>
40 #include <linux/capability.h>
41 #include <linux/cpu.h>
42 #include <linux/cgroup.h>
43 #include <linux/security.h>
44 #include <linux/hugetlb.h>
45 #include <linux/seccomp.h>
46 #include <linux/swap.h>
47 #include <linux/syscalls.h>
48 #include <linux/jiffies.h>
49 #include <linux/futex.h>
50 #include <linux/compat.h>
51 #include <linux/kthread.h>
52 #include <linux/task_io_accounting_ops.h>
53 #include <linux/rcupdate.h>
54 #include <linux/ptrace.h>
55 #include <linux/mount.h>
56 #include <linux/audit.h>
57 #include <linux/memcontrol.h>
58 #include <linux/ftrace.h>
59 #include <linux/proc_fs.h>
60 #include <linux/profile.h>
61 #include <linux/rmap.h>
62 #include <linux/ksm.h>
63 #include <linux/acct.h>
64 #include <linux/userfaultfd_k.h>
65 #include <linux/tsacct_kern.h>
66 #include <linux/cn_proc.h>
67 #include <linux/freezer.h>
68 #include <linux/delayacct.h>
69 #include <linux/taskstats_kern.h>
70 #include <linux/random.h>
71 #include <linux/tty.h>
72 #include <linux/blkdev.h>
73 #include <linux/fs_struct.h>
74 #include <linux/magic.h>
75 #include <linux/perf_event.h>
76 #include <linux/posix-timers.h>
77 #include <linux/user-return-notifier.h>
78 #include <linux/oom.h>
79 #include <linux/khugepaged.h>
80 #include <linux/signalfd.h>
81 #include <linux/uprobes.h>
82 #include <linux/aio.h>
83 #include <linux/compiler.h>
84 #include <linux/sysctl.h>
85 #include <linux/kcov.h>
87 #include <asm/pgtable.h>
88 #include <asm/pgalloc.h>
89 #include <linux/uaccess.h>
90 #include <asm/mmu_context.h>
91 #include <asm/cacheflush.h>
92 #include <asm/tlbflush.h>
94 #include <trace/events/sched.h>
96 #define CREATE_TRACE_POINTS
97 #include <trace/events/task.h>
100 * Minimum number of threads to boot the kernel
102 #define MIN_THREADS 20
105 * Maximum number of threads
107 #define MAX_THREADS FUTEX_TID_MASK
110 * Protected counters by write_lock_irq(&tasklist_lock)
112 unsigned long total_forks
; /* Handle normal Linux uptimes. */
113 int nr_threads
; /* The idle threads do not count.. */
115 int max_threads
; /* tunable limit on nr_threads */
117 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
119 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
121 #ifdef CONFIG_PROVE_RCU
122 int lockdep_tasklist_lock_is_held(void)
124 return lockdep_is_held(&tasklist_lock
);
126 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
127 #endif /* #ifdef CONFIG_PROVE_RCU */
129 int nr_processes(void)
134 for_each_possible_cpu(cpu
)
135 total
+= per_cpu(process_counts
, cpu
);
140 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
144 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
145 static struct kmem_cache
*task_struct_cachep
;
147 static inline struct task_struct
*alloc_task_struct_node(int node
)
149 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
152 static inline void free_task_struct(struct task_struct
*tsk
)
154 kmem_cache_free(task_struct_cachep
, tsk
);
158 void __weak
arch_release_thread_stack(unsigned long *stack
)
162 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
165 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
166 * kmemcache based allocator.
168 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
170 #ifdef CONFIG_VMAP_STACK
172 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
173 * flush. Try to minimize the number of calls by caching stacks.
175 #define NR_CACHED_STACKS 2
176 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
179 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
181 #ifdef CONFIG_VMAP_STACK
186 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
187 struct vm_struct
*s
= this_cpu_read(cached_stacks
[i
]);
191 this_cpu_write(cached_stacks
[i
], NULL
);
193 tsk
->stack_vm_area
= s
;
199 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_SIZE
,
200 VMALLOC_START
, VMALLOC_END
,
201 THREADINFO_GFP
| __GFP_HIGHMEM
,
203 0, node
, __builtin_return_address(0));
206 * We can't call find_vm_area() in interrupt context, and
207 * free_thread_stack() can be called in interrupt context,
208 * so cache the vm_struct.
211 tsk
->stack_vm_area
= find_vm_area(stack
);
214 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
217 return page
? page_address(page
) : NULL
;
221 static inline void free_thread_stack(struct task_struct
*tsk
)
223 #ifdef CONFIG_VMAP_STACK
224 if (task_stack_vm_area(tsk
)) {
228 local_irq_save(flags
);
229 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
230 if (this_cpu_read(cached_stacks
[i
]))
233 this_cpu_write(cached_stacks
[i
], tsk
->stack_vm_area
);
234 local_irq_restore(flags
);
237 local_irq_restore(flags
);
239 vfree_atomic(tsk
->stack
);
244 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
247 static struct kmem_cache
*thread_stack_cache
;
249 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
252 return kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
255 static void free_thread_stack(struct task_struct
*tsk
)
257 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
260 void thread_stack_cache_init(void)
262 thread_stack_cache
= kmem_cache_create("thread_stack", THREAD_SIZE
,
263 THREAD_SIZE
, 0, NULL
);
264 BUG_ON(thread_stack_cache
== NULL
);
269 /* SLAB cache for signal_struct structures (tsk->signal) */
270 static struct kmem_cache
*signal_cachep
;
272 /* SLAB cache for sighand_struct structures (tsk->sighand) */
273 struct kmem_cache
*sighand_cachep
;
275 /* SLAB cache for files_struct structures (tsk->files) */
276 struct kmem_cache
*files_cachep
;
278 /* SLAB cache for fs_struct structures (tsk->fs) */
279 struct kmem_cache
*fs_cachep
;
281 /* SLAB cache for vm_area_struct structures */
282 struct kmem_cache
*vm_area_cachep
;
284 /* SLAB cache for mm_struct structures (tsk->mm) */
285 static struct kmem_cache
*mm_cachep
;
287 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
289 void *stack
= task_stack_page(tsk
);
290 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
292 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
297 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
299 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
300 mod_zone_page_state(page_zone(vm
->pages
[i
]),
302 PAGE_SIZE
/ 1024 * account
);
305 /* All stack pages belong to the same memcg. */
306 memcg_kmem_update_page_stat(vm
->pages
[0], MEMCG_KERNEL_STACK_KB
,
307 account
* (THREAD_SIZE
/ 1024));
310 * All stack pages are in the same zone and belong to the
313 struct page
*first_page
= virt_to_page(stack
);
315 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
316 THREAD_SIZE
/ 1024 * account
);
318 memcg_kmem_update_page_stat(first_page
, MEMCG_KERNEL_STACK_KB
,
319 account
* (THREAD_SIZE
/ 1024));
323 static void release_task_stack(struct task_struct
*tsk
)
325 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
326 return; /* Better to leak the stack than to free prematurely */
328 account_kernel_stack(tsk
, -1);
329 arch_release_thread_stack(tsk
->stack
);
330 free_thread_stack(tsk
);
332 #ifdef CONFIG_VMAP_STACK
333 tsk
->stack_vm_area
= NULL
;
337 #ifdef CONFIG_THREAD_INFO_IN_TASK
338 void put_task_stack(struct task_struct
*tsk
)
340 if (atomic_dec_and_test(&tsk
->stack_refcount
))
341 release_task_stack(tsk
);
345 void free_task(struct task_struct
*tsk
)
347 #ifndef CONFIG_THREAD_INFO_IN_TASK
349 * The task is finally done with both the stack and thread_info,
352 release_task_stack(tsk
);
355 * If the task had a separate stack allocation, it should be gone
358 WARN_ON_ONCE(atomic_read(&tsk
->stack_refcount
) != 0);
360 rt_mutex_debug_task_free(tsk
);
361 ftrace_graph_exit_task(tsk
);
362 put_seccomp_filter(tsk
);
363 arch_release_task_struct(tsk
);
364 if (tsk
->flags
& PF_KTHREAD
)
365 free_kthread_struct(tsk
);
366 free_task_struct(tsk
);
368 EXPORT_SYMBOL(free_task
);
370 static inline void free_signal_struct(struct signal_struct
*sig
)
372 taskstats_tgid_free(sig
);
373 sched_autogroup_exit(sig
);
375 * __mmdrop is not safe to call from softirq context on x86 due to
376 * pgd_dtor so postpone it to the async context
379 mmdrop_async(sig
->oom_mm
);
380 kmem_cache_free(signal_cachep
, sig
);
383 static inline void put_signal_struct(struct signal_struct
*sig
)
385 if (atomic_dec_and_test(&sig
->sigcnt
))
386 free_signal_struct(sig
);
389 void __put_task_struct(struct task_struct
*tsk
)
391 WARN_ON(!tsk
->exit_state
);
392 WARN_ON(atomic_read(&tsk
->usage
));
393 WARN_ON(tsk
== current
);
397 security_task_free(tsk
);
399 delayacct_tsk_free(tsk
);
400 put_signal_struct(tsk
->signal
);
402 if (!profile_handoff_task(tsk
))
405 EXPORT_SYMBOL_GPL(__put_task_struct
);
407 void __init __weak
arch_task_cache_init(void) { }
412 static void set_max_threads(unsigned int max_threads_suggested
)
417 * The number of threads shall be limited such that the thread
418 * structures may only consume a small part of the available memory.
420 if (fls64(totalram_pages
) + fls64(PAGE_SIZE
) > 64)
421 threads
= MAX_THREADS
;
423 threads
= div64_u64((u64
) totalram_pages
* (u64
) PAGE_SIZE
,
424 (u64
) THREAD_SIZE
* 8UL);
426 if (threads
> max_threads_suggested
)
427 threads
= max_threads_suggested
;
429 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
432 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
433 /* Initialized by the architecture: */
434 int arch_task_struct_size __read_mostly
;
437 void __init
fork_init(void)
440 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
441 #ifndef ARCH_MIN_TASKALIGN
442 #define ARCH_MIN_TASKALIGN 0
444 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
446 /* create a slab on which task_structs can be allocated */
447 task_struct_cachep
= kmem_cache_create("task_struct",
448 arch_task_struct_size
, align
,
449 SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
, NULL
);
452 /* do the arch specific task caches init */
453 arch_task_cache_init();
455 set_max_threads(MAX_THREADS
);
457 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
458 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
459 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
460 init_task
.signal
->rlim
[RLIMIT_NPROC
];
462 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
463 init_user_ns
.ucount_max
[i
] = max_threads
/2;
467 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
468 struct task_struct
*src
)
474 void set_task_stack_end_magic(struct task_struct
*tsk
)
476 unsigned long *stackend
;
478 stackend
= end_of_stack(tsk
);
479 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
482 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
484 struct task_struct
*tsk
;
485 unsigned long *stack
;
486 struct vm_struct
*stack_vm_area
;
489 if (node
== NUMA_NO_NODE
)
490 node
= tsk_fork_get_node(orig
);
491 tsk
= alloc_task_struct_node(node
);
495 stack
= alloc_thread_stack_node(tsk
, node
);
499 stack_vm_area
= task_stack_vm_area(tsk
);
501 err
= arch_dup_task_struct(tsk
, orig
);
504 * arch_dup_task_struct() clobbers the stack-related fields. Make
505 * sure they're properly initialized before using any stack-related
509 #ifdef CONFIG_VMAP_STACK
510 tsk
->stack_vm_area
= stack_vm_area
;
512 #ifdef CONFIG_THREAD_INFO_IN_TASK
513 atomic_set(&tsk
->stack_refcount
, 1);
519 #ifdef CONFIG_SECCOMP
521 * We must handle setting up seccomp filters once we're under
522 * the sighand lock in case orig has changed between now and
523 * then. Until then, filter must be NULL to avoid messing up
524 * the usage counts on the error path calling free_task.
526 tsk
->seccomp
.filter
= NULL
;
529 setup_thread_stack(tsk
, orig
);
530 clear_user_return_notifier(tsk
);
531 clear_tsk_need_resched(tsk
);
532 set_task_stack_end_magic(tsk
);
534 #ifdef CONFIG_CC_STACKPROTECTOR
535 tsk
->stack_canary
= get_random_int();
539 * One for us, one for whoever does the "release_task()" (usually
542 atomic_set(&tsk
->usage
, 2);
543 #ifdef CONFIG_BLK_DEV_IO_TRACE
546 tsk
->splice_pipe
= NULL
;
547 tsk
->task_frag
.page
= NULL
;
548 tsk
->wake_q
.next
= NULL
;
550 account_kernel_stack(tsk
, 1);
557 free_thread_stack(tsk
);
559 free_task_struct(tsk
);
564 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
565 struct mm_struct
*oldmm
)
567 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
568 struct rb_node
**rb_link
, *rb_parent
;
570 unsigned long charge
;
573 uprobe_start_dup_mmap();
574 if (down_write_killable(&oldmm
->mmap_sem
)) {
576 goto fail_uprobe_end
;
578 flush_cache_dup_mm(oldmm
);
579 uprobe_dup_mmap(oldmm
, mm
);
581 * Not linked in yet - no deadlock potential:
583 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
585 /* No ordering required: file already has been exposed. */
586 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
588 mm
->total_vm
= oldmm
->total_vm
;
589 mm
->data_vm
= oldmm
->data_vm
;
590 mm
->exec_vm
= oldmm
->exec_vm
;
591 mm
->stack_vm
= oldmm
->stack_vm
;
593 rb_link
= &mm
->mm_rb
.rb_node
;
596 retval
= ksm_fork(mm
, oldmm
);
599 retval
= khugepaged_fork(mm
, oldmm
);
604 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
607 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
608 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
612 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
613 unsigned long len
= vma_pages(mpnt
);
615 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
619 tmp
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
623 INIT_LIST_HEAD(&tmp
->anon_vma_chain
);
624 retval
= vma_dup_policy(mpnt
, tmp
);
626 goto fail_nomem_policy
;
628 retval
= dup_userfaultfd(tmp
, &uf
);
630 goto fail_nomem_anon_vma_fork
;
631 if (anon_vma_fork(tmp
, mpnt
))
632 goto fail_nomem_anon_vma_fork
;
633 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
634 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
637 struct inode
*inode
= file_inode(file
);
638 struct address_space
*mapping
= file
->f_mapping
;
641 if (tmp
->vm_flags
& VM_DENYWRITE
)
642 atomic_dec(&inode
->i_writecount
);
643 i_mmap_lock_write(mapping
);
644 if (tmp
->vm_flags
& VM_SHARED
)
645 atomic_inc(&mapping
->i_mmap_writable
);
646 flush_dcache_mmap_lock(mapping
);
647 /* insert tmp into the share list, just after mpnt */
648 vma_interval_tree_insert_after(tmp
, mpnt
,
650 flush_dcache_mmap_unlock(mapping
);
651 i_mmap_unlock_write(mapping
);
655 * Clear hugetlb-related page reserves for children. This only
656 * affects MAP_PRIVATE mappings. Faults generated by the child
657 * are not guaranteed to succeed, even if read-only
659 if (is_vm_hugetlb_page(tmp
))
660 reset_vma_resv_huge_pages(tmp
);
663 * Link in the new vma and copy the page table entries.
666 pprev
= &tmp
->vm_next
;
670 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
671 rb_link
= &tmp
->vm_rb
.rb_right
;
672 rb_parent
= &tmp
->vm_rb
;
675 retval
= copy_page_range(mm
, oldmm
, mpnt
);
677 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
678 tmp
->vm_ops
->open(tmp
);
683 /* a new mm has just been created */
684 arch_dup_mmap(oldmm
, mm
);
687 up_write(&mm
->mmap_sem
);
689 up_write(&oldmm
->mmap_sem
);
690 dup_userfaultfd_complete(&uf
);
692 uprobe_end_dup_mmap();
694 fail_nomem_anon_vma_fork
:
695 mpol_put(vma_policy(tmp
));
697 kmem_cache_free(vm_area_cachep
, tmp
);
700 vm_unacct_memory(charge
);
704 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
706 mm
->pgd
= pgd_alloc(mm
);
707 if (unlikely(!mm
->pgd
))
712 static inline void mm_free_pgd(struct mm_struct
*mm
)
714 pgd_free(mm
, mm
->pgd
);
717 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
719 down_write(&oldmm
->mmap_sem
);
720 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
721 up_write(&oldmm
->mmap_sem
);
724 #define mm_alloc_pgd(mm) (0)
725 #define mm_free_pgd(mm)
726 #endif /* CONFIG_MMU */
728 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
730 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
731 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
733 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
735 static int __init
coredump_filter_setup(char *s
)
737 default_dump_filter
=
738 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
739 MMF_DUMP_FILTER_MASK
;
743 __setup("coredump_filter=", coredump_filter_setup
);
745 #include <linux/init_task.h>
747 static void mm_init_aio(struct mm_struct
*mm
)
750 spin_lock_init(&mm
->ioctx_lock
);
751 mm
->ioctx_table
= NULL
;
755 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
762 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
763 struct user_namespace
*user_ns
)
767 mm
->vmacache_seqnum
= 0;
768 atomic_set(&mm
->mm_users
, 1);
769 atomic_set(&mm
->mm_count
, 1);
770 init_rwsem(&mm
->mmap_sem
);
771 INIT_LIST_HEAD(&mm
->mmlist
);
772 mm
->core_state
= NULL
;
773 atomic_long_set(&mm
->nr_ptes
, 0);
778 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
779 spin_lock_init(&mm
->page_table_lock
);
782 mm_init_owner(mm
, p
);
783 mmu_notifier_mm_init(mm
);
784 clear_tlb_flush_pending(mm
);
785 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
786 mm
->pmd_huge_pte
= NULL
;
790 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
791 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
793 mm
->flags
= default_dump_filter
;
797 if (mm_alloc_pgd(mm
))
800 if (init_new_context(p
, mm
))
803 mm
->user_ns
= get_user_ns(user_ns
);
813 static void check_mm(struct mm_struct
*mm
)
817 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
818 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
821 printk(KERN_ALERT
"BUG: Bad rss-counter state "
822 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
825 if (atomic_long_read(&mm
->nr_ptes
))
826 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
827 atomic_long_read(&mm
->nr_ptes
));
829 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
832 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
833 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
838 * Allocate and initialize an mm_struct.
840 struct mm_struct
*mm_alloc(void)
842 struct mm_struct
*mm
;
848 memset(mm
, 0, sizeof(*mm
));
849 return mm_init(mm
, current
, current_user_ns());
853 * Called when the last reference to the mm
854 * is dropped: either by a lazy thread or by
855 * mmput. Free the page directory and the mm.
857 void __mmdrop(struct mm_struct
*mm
)
859 BUG_ON(mm
== &init_mm
);
862 mmu_notifier_mm_destroy(mm
);
864 put_user_ns(mm
->user_ns
);
867 EXPORT_SYMBOL_GPL(__mmdrop
);
869 static inline void __mmput(struct mm_struct
*mm
)
871 VM_BUG_ON(atomic_read(&mm
->mm_users
));
873 uprobe_clear_state(mm
);
876 khugepaged_exit(mm
); /* must run before exit_mmap */
878 mm_put_huge_zero_page(mm
);
879 set_mm_exe_file(mm
, NULL
);
880 if (!list_empty(&mm
->mmlist
)) {
881 spin_lock(&mmlist_lock
);
882 list_del(&mm
->mmlist
);
883 spin_unlock(&mmlist_lock
);
886 module_put(mm
->binfmt
->module
);
887 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
892 * Decrement the use count and release all resources for an mm.
894 void mmput(struct mm_struct
*mm
)
898 if (atomic_dec_and_test(&mm
->mm_users
))
901 EXPORT_SYMBOL_GPL(mmput
);
904 static void mmput_async_fn(struct work_struct
*work
)
906 struct mm_struct
*mm
= container_of(work
, struct mm_struct
, async_put_work
);
910 void mmput_async(struct mm_struct
*mm
)
912 if (atomic_dec_and_test(&mm
->mm_users
)) {
913 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
914 schedule_work(&mm
->async_put_work
);
920 * set_mm_exe_file - change a reference to the mm's executable file
922 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
924 * Main users are mmput() and sys_execve(). Callers prevent concurrent
925 * invocations: in mmput() nobody alive left, in execve task is single
926 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
927 * mm->exe_file, but does so without using set_mm_exe_file() in order
928 * to do avoid the need for any locks.
930 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
932 struct file
*old_exe_file
;
935 * It is safe to dereference the exe_file without RCU as
936 * this function is only called if nobody else can access
937 * this mm -- see comment above for justification.
939 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
942 get_file(new_exe_file
);
943 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
949 * get_mm_exe_file - acquire a reference to the mm's executable file
951 * Returns %NULL if mm has no associated executable file.
952 * User must release file via fput().
954 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
956 struct file
*exe_file
;
959 exe_file
= rcu_dereference(mm
->exe_file
);
960 if (exe_file
&& !get_file_rcu(exe_file
))
965 EXPORT_SYMBOL(get_mm_exe_file
);
968 * get_task_exe_file - acquire a reference to the task's executable file
970 * Returns %NULL if task's mm (if any) has no associated executable file or
971 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
972 * User must release file via fput().
974 struct file
*get_task_exe_file(struct task_struct
*task
)
976 struct file
*exe_file
= NULL
;
977 struct mm_struct
*mm
;
982 if (!(task
->flags
& PF_KTHREAD
))
983 exe_file
= get_mm_exe_file(mm
);
988 EXPORT_SYMBOL(get_task_exe_file
);
991 * get_task_mm - acquire a reference to the task's mm
993 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
994 * this kernel workthread has transiently adopted a user mm with use_mm,
995 * to do its AIO) is not set and if so returns a reference to it, after
996 * bumping up the use count. User must release the mm via mmput()
997 * after use. Typically used by /proc and ptrace.
999 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1001 struct mm_struct
*mm
;
1006 if (task
->flags
& PF_KTHREAD
)
1014 EXPORT_SYMBOL_GPL(get_task_mm
);
1016 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1018 struct mm_struct
*mm
;
1021 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1023 return ERR_PTR(err
);
1025 mm
= get_task_mm(task
);
1026 if (mm
&& mm
!= current
->mm
&&
1027 !ptrace_may_access(task
, mode
)) {
1029 mm
= ERR_PTR(-EACCES
);
1031 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1036 static void complete_vfork_done(struct task_struct
*tsk
)
1038 struct completion
*vfork
;
1041 vfork
= tsk
->vfork_done
;
1042 if (likely(vfork
)) {
1043 tsk
->vfork_done
= NULL
;
1049 static int wait_for_vfork_done(struct task_struct
*child
,
1050 struct completion
*vfork
)
1054 freezer_do_not_count();
1055 killed
= wait_for_completion_killable(vfork
);
1060 child
->vfork_done
= NULL
;
1064 put_task_struct(child
);
1068 /* Please note the differences between mmput and mm_release.
1069 * mmput is called whenever we stop holding onto a mm_struct,
1070 * error success whatever.
1072 * mm_release is called after a mm_struct has been removed
1073 * from the current process.
1075 * This difference is important for error handling, when we
1076 * only half set up a mm_struct for a new process and need to restore
1077 * the old one. Because we mmput the new mm_struct before
1078 * restoring the old one. . .
1079 * Eric Biederman 10 January 1998
1081 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1083 /* Get rid of any futexes when releasing the mm */
1085 if (unlikely(tsk
->robust_list
)) {
1086 exit_robust_list(tsk
);
1087 tsk
->robust_list
= NULL
;
1089 #ifdef CONFIG_COMPAT
1090 if (unlikely(tsk
->compat_robust_list
)) {
1091 compat_exit_robust_list(tsk
);
1092 tsk
->compat_robust_list
= NULL
;
1095 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
1096 exit_pi_state_list(tsk
);
1099 uprobe_free_utask(tsk
);
1101 /* Get rid of any cached register state */
1102 deactivate_mm(tsk
, mm
);
1105 * Signal userspace if we're not exiting with a core dump
1106 * because we want to leave the value intact for debugging
1109 if (tsk
->clear_child_tid
) {
1110 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1111 atomic_read(&mm
->mm_users
) > 1) {
1113 * We don't check the error code - if userspace has
1114 * not set up a proper pointer then tough luck.
1116 put_user(0, tsk
->clear_child_tid
);
1117 sys_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1120 tsk
->clear_child_tid
= NULL
;
1124 * All done, finally we can wake up parent and return this mm to him.
1125 * Also kthread_stop() uses this completion for synchronization.
1127 if (tsk
->vfork_done
)
1128 complete_vfork_done(tsk
);
1132 * Allocate a new mm structure and copy contents from the
1133 * mm structure of the passed in task structure.
1135 static struct mm_struct
*dup_mm(struct task_struct
*tsk
)
1137 struct mm_struct
*mm
, *oldmm
= current
->mm
;
1144 memcpy(mm
, oldmm
, sizeof(*mm
));
1146 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1149 err
= dup_mmap(mm
, oldmm
);
1153 mm
->hiwater_rss
= get_mm_rss(mm
);
1154 mm
->hiwater_vm
= mm
->total_vm
;
1156 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1162 /* don't put binfmt in mmput, we haven't got module yet */
1170 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1172 struct mm_struct
*mm
, *oldmm
;
1175 tsk
->min_flt
= tsk
->maj_flt
= 0;
1176 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1177 #ifdef CONFIG_DETECT_HUNG_TASK
1178 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1182 tsk
->active_mm
= NULL
;
1185 * Are we cloning a kernel thread?
1187 * We need to steal a active VM for that..
1189 oldmm
= current
->mm
;
1193 /* initialize the new vmacache entries */
1194 vmacache_flush(tsk
);
1196 if (clone_flags
& CLONE_VM
) {
1209 tsk
->active_mm
= mm
;
1216 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1218 struct fs_struct
*fs
= current
->fs
;
1219 if (clone_flags
& CLONE_FS
) {
1220 /* tsk->fs is already what we want */
1221 spin_lock(&fs
->lock
);
1223 spin_unlock(&fs
->lock
);
1227 spin_unlock(&fs
->lock
);
1230 tsk
->fs
= copy_fs_struct(fs
);
1236 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1238 struct files_struct
*oldf
, *newf
;
1242 * A background process may not have any files ...
1244 oldf
= current
->files
;
1248 if (clone_flags
& CLONE_FILES
) {
1249 atomic_inc(&oldf
->count
);
1253 newf
= dup_fd(oldf
, &error
);
1263 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1266 struct io_context
*ioc
= current
->io_context
;
1267 struct io_context
*new_ioc
;
1272 * Share io context with parent, if CLONE_IO is set
1274 if (clone_flags
& CLONE_IO
) {
1276 tsk
->io_context
= ioc
;
1277 } else if (ioprio_valid(ioc
->ioprio
)) {
1278 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1279 if (unlikely(!new_ioc
))
1282 new_ioc
->ioprio
= ioc
->ioprio
;
1283 put_io_context(new_ioc
);
1289 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1291 struct sighand_struct
*sig
;
1293 if (clone_flags
& CLONE_SIGHAND
) {
1294 atomic_inc(¤t
->sighand
->count
);
1297 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1298 rcu_assign_pointer(tsk
->sighand
, sig
);
1302 atomic_set(&sig
->count
, 1);
1303 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1307 void __cleanup_sighand(struct sighand_struct
*sighand
)
1309 if (atomic_dec_and_test(&sighand
->count
)) {
1310 signalfd_cleanup(sighand
);
1312 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1313 * without an RCU grace period, see __lock_task_sighand().
1315 kmem_cache_free(sighand_cachep
, sighand
);
1319 #ifdef CONFIG_POSIX_TIMERS
1321 * Initialize POSIX timer handling for a thread group.
1323 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1325 unsigned long cpu_limit
;
1327 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1328 if (cpu_limit
!= RLIM_INFINITY
) {
1329 sig
->cputime_expires
.prof_exp
= cpu_limit
* NSEC_PER_SEC
;
1330 sig
->cputimer
.running
= true;
1333 /* The timer lists. */
1334 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1335 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1336 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1339 static inline void posix_cpu_timers_init_group(struct signal_struct
*sig
) { }
1342 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1344 struct signal_struct
*sig
;
1346 if (clone_flags
& CLONE_THREAD
)
1349 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1354 sig
->nr_threads
= 1;
1355 atomic_set(&sig
->live
, 1);
1356 atomic_set(&sig
->sigcnt
, 1);
1358 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1359 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1360 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1362 init_waitqueue_head(&sig
->wait_chldexit
);
1363 sig
->curr_target
= tsk
;
1364 init_sigpending(&sig
->shared_pending
);
1365 seqlock_init(&sig
->stats_lock
);
1366 prev_cputime_init(&sig
->prev_cputime
);
1368 #ifdef CONFIG_POSIX_TIMERS
1369 INIT_LIST_HEAD(&sig
->posix_timers
);
1370 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1371 sig
->real_timer
.function
= it_real_fn
;
1374 task_lock(current
->group_leader
);
1375 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1376 task_unlock(current
->group_leader
);
1378 posix_cpu_timers_init_group(sig
);
1380 tty_audit_fork(sig
);
1381 sched_autogroup_fork(sig
);
1383 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1384 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1386 mutex_init(&sig
->cred_guard_mutex
);
1391 static void copy_seccomp(struct task_struct
*p
)
1393 #ifdef CONFIG_SECCOMP
1395 * Must be called with sighand->lock held, which is common to
1396 * all threads in the group. Holding cred_guard_mutex is not
1397 * needed because this new task is not yet running and cannot
1400 assert_spin_locked(¤t
->sighand
->siglock
);
1402 /* Ref-count the new filter user, and assign it. */
1403 get_seccomp_filter(current
);
1404 p
->seccomp
= current
->seccomp
;
1407 * Explicitly enable no_new_privs here in case it got set
1408 * between the task_struct being duplicated and holding the
1409 * sighand lock. The seccomp state and nnp must be in sync.
1411 if (task_no_new_privs(current
))
1412 task_set_no_new_privs(p
);
1415 * If the parent gained a seccomp mode after copying thread
1416 * flags and between before we held the sighand lock, we have
1417 * to manually enable the seccomp thread flag here.
1419 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1420 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1424 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1426 current
->clear_child_tid
= tidptr
;
1428 return task_pid_vnr(current
);
1431 static void rt_mutex_init_task(struct task_struct
*p
)
1433 raw_spin_lock_init(&p
->pi_lock
);
1434 #ifdef CONFIG_RT_MUTEXES
1435 p
->pi_waiters
= RB_ROOT
;
1436 p
->pi_waiters_leftmost
= NULL
;
1437 p
->pi_blocked_on
= NULL
;
1441 #ifdef CONFIG_POSIX_TIMERS
1443 * Initialize POSIX timer handling for a single task.
1445 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1447 tsk
->cputime_expires
.prof_exp
= 0;
1448 tsk
->cputime_expires
.virt_exp
= 0;
1449 tsk
->cputime_expires
.sched_exp
= 0;
1450 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1451 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1452 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1455 static inline void posix_cpu_timers_init(struct task_struct
*tsk
) { }
1459 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1461 task
->pids
[type
].pid
= pid
;
1465 * This creates a new process as a copy of the old one,
1466 * but does not actually start it yet.
1468 * It copies the registers, and all the appropriate
1469 * parts of the process environment (as per the clone
1470 * flags). The actual kick-off is left to the caller.
1472 static __latent_entropy
struct task_struct
*copy_process(
1473 unsigned long clone_flags
,
1474 unsigned long stack_start
,
1475 unsigned long stack_size
,
1476 int __user
*child_tidptr
,
1483 struct task_struct
*p
;
1485 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1486 return ERR_PTR(-EINVAL
);
1488 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1489 return ERR_PTR(-EINVAL
);
1492 * Thread groups must share signals as well, and detached threads
1493 * can only be started up within the thread group.
1495 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1496 return ERR_PTR(-EINVAL
);
1499 * Shared signal handlers imply shared VM. By way of the above,
1500 * thread groups also imply shared VM. Blocking this case allows
1501 * for various simplifications in other code.
1503 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1504 return ERR_PTR(-EINVAL
);
1507 * Siblings of global init remain as zombies on exit since they are
1508 * not reaped by their parent (swapper). To solve this and to avoid
1509 * multi-rooted process trees, prevent global and container-inits
1510 * from creating siblings.
1512 if ((clone_flags
& CLONE_PARENT
) &&
1513 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1514 return ERR_PTR(-EINVAL
);
1517 * If the new process will be in a different pid or user namespace
1518 * do not allow it to share a thread group with the forking task.
1520 if (clone_flags
& CLONE_THREAD
) {
1521 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1522 (task_active_pid_ns(current
) !=
1523 current
->nsproxy
->pid_ns_for_children
))
1524 return ERR_PTR(-EINVAL
);
1527 retval
= security_task_create(clone_flags
);
1532 p
= dup_task_struct(current
, node
);
1536 ftrace_graph_init_task(p
);
1538 rt_mutex_init_task(p
);
1540 #ifdef CONFIG_PROVE_LOCKING
1541 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1542 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1545 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1546 task_rlimit(p
, RLIMIT_NPROC
)) {
1547 if (p
->real_cred
->user
!= INIT_USER
&&
1548 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1551 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1553 retval
= copy_creds(p
, clone_flags
);
1558 * If multiple threads are within copy_process(), then this check
1559 * triggers too late. This doesn't hurt, the check is only there
1560 * to stop root fork bombs.
1563 if (nr_threads
>= max_threads
)
1564 goto bad_fork_cleanup_count
;
1566 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1567 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1568 p
->flags
|= PF_FORKNOEXEC
;
1569 INIT_LIST_HEAD(&p
->children
);
1570 INIT_LIST_HEAD(&p
->sibling
);
1571 rcu_copy_process(p
);
1572 p
->vfork_done
= NULL
;
1573 spin_lock_init(&p
->alloc_lock
);
1575 init_sigpending(&p
->pending
);
1577 p
->utime
= p
->stime
= p
->gtime
= 0;
1578 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1579 p
->utimescaled
= p
->stimescaled
= 0;
1581 prev_cputime_init(&p
->prev_cputime
);
1583 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1584 seqcount_init(&p
->vtime_seqcount
);
1586 p
->vtime_snap_whence
= VTIME_INACTIVE
;
1589 #if defined(SPLIT_RSS_COUNTING)
1590 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1593 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1595 task_io_accounting_init(&p
->ioac
);
1596 acct_clear_integrals(p
);
1598 posix_cpu_timers_init(p
);
1600 p
->start_time
= ktime_get_ns();
1601 p
->real_start_time
= ktime_get_boot_ns();
1602 p
->io_context
= NULL
;
1603 p
->audit_context
= NULL
;
1606 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1607 if (IS_ERR(p
->mempolicy
)) {
1608 retval
= PTR_ERR(p
->mempolicy
);
1609 p
->mempolicy
= NULL
;
1610 goto bad_fork_cleanup_threadgroup_lock
;
1613 #ifdef CONFIG_CPUSETS
1614 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1615 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1616 seqcount_init(&p
->mems_allowed_seq
);
1618 #ifdef CONFIG_TRACE_IRQFLAGS
1620 p
->hardirqs_enabled
= 0;
1621 p
->hardirq_enable_ip
= 0;
1622 p
->hardirq_enable_event
= 0;
1623 p
->hardirq_disable_ip
= _THIS_IP_
;
1624 p
->hardirq_disable_event
= 0;
1625 p
->softirqs_enabled
= 1;
1626 p
->softirq_enable_ip
= _THIS_IP_
;
1627 p
->softirq_enable_event
= 0;
1628 p
->softirq_disable_ip
= 0;
1629 p
->softirq_disable_event
= 0;
1630 p
->hardirq_context
= 0;
1631 p
->softirq_context
= 0;
1634 p
->pagefault_disabled
= 0;
1636 #ifdef CONFIG_LOCKDEP
1637 p
->lockdep_depth
= 0; /* no locks held yet */
1638 p
->curr_chain_key
= 0;
1639 p
->lockdep_recursion
= 0;
1642 #ifdef CONFIG_DEBUG_MUTEXES
1643 p
->blocked_on
= NULL
; /* not blocked yet */
1645 #ifdef CONFIG_BCACHE
1646 p
->sequential_io
= 0;
1647 p
->sequential_io_avg
= 0;
1650 /* Perform scheduler related setup. Assign this task to a CPU. */
1651 retval
= sched_fork(clone_flags
, p
);
1653 goto bad_fork_cleanup_policy
;
1655 retval
= perf_event_init_task(p
);
1657 goto bad_fork_cleanup_policy
;
1658 retval
= audit_alloc(p
);
1660 goto bad_fork_cleanup_perf
;
1661 /* copy all the process information */
1663 retval
= copy_semundo(clone_flags
, p
);
1665 goto bad_fork_cleanup_audit
;
1666 retval
= copy_files(clone_flags
, p
);
1668 goto bad_fork_cleanup_semundo
;
1669 retval
= copy_fs(clone_flags
, p
);
1671 goto bad_fork_cleanup_files
;
1672 retval
= copy_sighand(clone_flags
, p
);
1674 goto bad_fork_cleanup_fs
;
1675 retval
= copy_signal(clone_flags
, p
);
1677 goto bad_fork_cleanup_sighand
;
1678 retval
= copy_mm(clone_flags
, p
);
1680 goto bad_fork_cleanup_signal
;
1681 retval
= copy_namespaces(clone_flags
, p
);
1683 goto bad_fork_cleanup_mm
;
1684 retval
= copy_io(clone_flags
, p
);
1686 goto bad_fork_cleanup_namespaces
;
1687 retval
= copy_thread_tls(clone_flags
, stack_start
, stack_size
, p
, tls
);
1689 goto bad_fork_cleanup_io
;
1691 if (pid
!= &init_struct_pid
) {
1692 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
1694 retval
= PTR_ERR(pid
);
1695 goto bad_fork_cleanup_thread
;
1699 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
1701 * Clear TID on mm_release()?
1703 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
1708 p
->robust_list
= NULL
;
1709 #ifdef CONFIG_COMPAT
1710 p
->compat_robust_list
= NULL
;
1712 INIT_LIST_HEAD(&p
->pi_state_list
);
1713 p
->pi_state_cache
= NULL
;
1716 * sigaltstack should be cleared when sharing the same VM
1718 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
1722 * Syscall tracing and stepping should be turned off in the
1723 * child regardless of CLONE_PTRACE.
1725 user_disable_single_step(p
);
1726 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
1727 #ifdef TIF_SYSCALL_EMU
1728 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1730 clear_all_latency_tracing(p
);
1732 /* ok, now we should be set up.. */
1733 p
->pid
= pid_nr(pid
);
1734 if (clone_flags
& CLONE_THREAD
) {
1735 p
->exit_signal
= -1;
1736 p
->group_leader
= current
->group_leader
;
1737 p
->tgid
= current
->tgid
;
1739 if (clone_flags
& CLONE_PARENT
)
1740 p
->exit_signal
= current
->group_leader
->exit_signal
;
1742 p
->exit_signal
= (clone_flags
& CSIGNAL
);
1743 p
->group_leader
= p
;
1748 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
1749 p
->dirty_paused_when
= 0;
1751 p
->pdeath_signal
= 0;
1752 INIT_LIST_HEAD(&p
->thread_group
);
1753 p
->task_works
= NULL
;
1755 cgroup_threadgroup_change_begin(current
);
1757 * Ensure that the cgroup subsystem policies allow the new process to be
1758 * forked. It should be noted the the new process's css_set can be changed
1759 * between here and cgroup_post_fork() if an organisation operation is in
1762 retval
= cgroup_can_fork(p
);
1764 goto bad_fork_free_pid
;
1767 * Make it visible to the rest of the system, but dont wake it up yet.
1768 * Need tasklist lock for parent etc handling!
1770 write_lock_irq(&tasklist_lock
);
1772 /* CLONE_PARENT re-uses the old parent */
1773 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
1774 p
->real_parent
= current
->real_parent
;
1775 p
->parent_exec_id
= current
->parent_exec_id
;
1777 p
->real_parent
= current
;
1778 p
->parent_exec_id
= current
->self_exec_id
;
1781 spin_lock(¤t
->sighand
->siglock
);
1784 * Copy seccomp details explicitly here, in case they were changed
1785 * before holding sighand lock.
1790 * Process group and session signals need to be delivered to just the
1791 * parent before the fork or both the parent and the child after the
1792 * fork. Restart if a signal comes in before we add the new process to
1793 * it's process group.
1794 * A fatal signal pending means that current will exit, so the new
1795 * thread can't slip out of an OOM kill (or normal SIGKILL).
1797 recalc_sigpending();
1798 if (signal_pending(current
)) {
1799 spin_unlock(¤t
->sighand
->siglock
);
1800 write_unlock_irq(&tasklist_lock
);
1801 retval
= -ERESTARTNOINTR
;
1802 goto bad_fork_cancel_cgroup
;
1805 if (likely(p
->pid
)) {
1806 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
1808 init_task_pid(p
, PIDTYPE_PID
, pid
);
1809 if (thread_group_leader(p
)) {
1810 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
1811 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
1813 if (is_child_reaper(pid
)) {
1814 ns_of_pid(pid
)->child_reaper
= p
;
1815 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
1818 p
->signal
->leader_pid
= pid
;
1819 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
1821 * Inherit has_child_subreaper flag under the same
1822 * tasklist_lock with adding child to the process tree
1823 * for propagate_has_child_subreaper optimization.
1825 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
1826 p
->real_parent
->signal
->is_child_subreaper
;
1827 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
1828 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
1829 attach_pid(p
, PIDTYPE_PGID
);
1830 attach_pid(p
, PIDTYPE_SID
);
1831 __this_cpu_inc(process_counts
);
1833 current
->signal
->nr_threads
++;
1834 atomic_inc(¤t
->signal
->live
);
1835 atomic_inc(¤t
->signal
->sigcnt
);
1836 list_add_tail_rcu(&p
->thread_group
,
1837 &p
->group_leader
->thread_group
);
1838 list_add_tail_rcu(&p
->thread_node
,
1839 &p
->signal
->thread_head
);
1841 attach_pid(p
, PIDTYPE_PID
);
1846 spin_unlock(¤t
->sighand
->siglock
);
1847 syscall_tracepoint_update(p
);
1848 write_unlock_irq(&tasklist_lock
);
1850 proc_fork_connector(p
);
1851 cgroup_post_fork(p
);
1852 cgroup_threadgroup_change_end(current
);
1855 trace_task_newtask(p
, clone_flags
);
1856 uprobe_copy_process(p
, clone_flags
);
1860 bad_fork_cancel_cgroup
:
1861 cgroup_cancel_fork(p
);
1863 cgroup_threadgroup_change_end(current
);
1864 if (pid
!= &init_struct_pid
)
1866 bad_fork_cleanup_thread
:
1868 bad_fork_cleanup_io
:
1871 bad_fork_cleanup_namespaces
:
1872 exit_task_namespaces(p
);
1873 bad_fork_cleanup_mm
:
1876 bad_fork_cleanup_signal
:
1877 if (!(clone_flags
& CLONE_THREAD
))
1878 free_signal_struct(p
->signal
);
1879 bad_fork_cleanup_sighand
:
1880 __cleanup_sighand(p
->sighand
);
1881 bad_fork_cleanup_fs
:
1882 exit_fs(p
); /* blocking */
1883 bad_fork_cleanup_files
:
1884 exit_files(p
); /* blocking */
1885 bad_fork_cleanup_semundo
:
1887 bad_fork_cleanup_audit
:
1889 bad_fork_cleanup_perf
:
1890 perf_event_free_task(p
);
1891 bad_fork_cleanup_policy
:
1893 mpol_put(p
->mempolicy
);
1894 bad_fork_cleanup_threadgroup_lock
:
1896 delayacct_tsk_free(p
);
1897 bad_fork_cleanup_count
:
1898 atomic_dec(&p
->cred
->user
->processes
);
1901 p
->state
= TASK_DEAD
;
1905 return ERR_PTR(retval
);
1908 static inline void init_idle_pids(struct pid_link
*links
)
1912 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1913 INIT_HLIST_NODE(&links
[type
].node
); /* not really needed */
1914 links
[type
].pid
= &init_struct_pid
;
1918 struct task_struct
*fork_idle(int cpu
)
1920 struct task_struct
*task
;
1921 task
= copy_process(CLONE_VM
, 0, 0, NULL
, &init_struct_pid
, 0, 0,
1923 if (!IS_ERR(task
)) {
1924 init_idle_pids(task
->pids
);
1925 init_idle(task
, cpu
);
1932 * Ok, this is the main fork-routine.
1934 * It copies the process, and if successful kick-starts
1935 * it and waits for it to finish using the VM if required.
1937 long _do_fork(unsigned long clone_flags
,
1938 unsigned long stack_start
,
1939 unsigned long stack_size
,
1940 int __user
*parent_tidptr
,
1941 int __user
*child_tidptr
,
1944 struct task_struct
*p
;
1949 * Determine whether and which event to report to ptracer. When
1950 * called from kernel_thread or CLONE_UNTRACED is explicitly
1951 * requested, no event is reported; otherwise, report if the event
1952 * for the type of forking is enabled.
1954 if (!(clone_flags
& CLONE_UNTRACED
)) {
1955 if (clone_flags
& CLONE_VFORK
)
1956 trace
= PTRACE_EVENT_VFORK
;
1957 else if ((clone_flags
& CSIGNAL
) != SIGCHLD
)
1958 trace
= PTRACE_EVENT_CLONE
;
1960 trace
= PTRACE_EVENT_FORK
;
1962 if (likely(!ptrace_event_enabled(current
, trace
)))
1966 p
= copy_process(clone_flags
, stack_start
, stack_size
,
1967 child_tidptr
, NULL
, trace
, tls
, NUMA_NO_NODE
);
1968 add_latent_entropy();
1970 * Do this prior waking up the new thread - the thread pointer
1971 * might get invalid after that point, if the thread exits quickly.
1974 struct completion vfork
;
1977 trace_sched_process_fork(current
, p
);
1979 pid
= get_task_pid(p
, PIDTYPE_PID
);
1982 if (clone_flags
& CLONE_PARENT_SETTID
)
1983 put_user(nr
, parent_tidptr
);
1985 if (clone_flags
& CLONE_VFORK
) {
1986 p
->vfork_done
= &vfork
;
1987 init_completion(&vfork
);
1991 wake_up_new_task(p
);
1993 /* forking complete and child started to run, tell ptracer */
1994 if (unlikely(trace
))
1995 ptrace_event_pid(trace
, pid
);
1997 if (clone_flags
& CLONE_VFORK
) {
1998 if (!wait_for_vfork_done(p
, &vfork
))
1999 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2009 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2010 /* For compatibility with architectures that call do_fork directly rather than
2011 * using the syscall entry points below. */
2012 long do_fork(unsigned long clone_flags
,
2013 unsigned long stack_start
,
2014 unsigned long stack_size
,
2015 int __user
*parent_tidptr
,
2016 int __user
*child_tidptr
)
2018 return _do_fork(clone_flags
, stack_start
, stack_size
,
2019 parent_tidptr
, child_tidptr
, 0);
2024 * Create a kernel thread.
2026 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2028 return _do_fork(flags
|CLONE_VM
|CLONE_UNTRACED
, (unsigned long)fn
,
2029 (unsigned long)arg
, NULL
, NULL
, 0);
2032 #ifdef __ARCH_WANT_SYS_FORK
2033 SYSCALL_DEFINE0(fork
)
2036 return _do_fork(SIGCHLD
, 0, 0, NULL
, NULL
, 0);
2038 /* can not support in nommu mode */
2044 #ifdef __ARCH_WANT_SYS_VFORK
2045 SYSCALL_DEFINE0(vfork
)
2047 return _do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, 0,
2052 #ifdef __ARCH_WANT_SYS_CLONE
2053 #ifdef CONFIG_CLONE_BACKWARDS
2054 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2055 int __user
*, parent_tidptr
,
2057 int __user
*, child_tidptr
)
2058 #elif defined(CONFIG_CLONE_BACKWARDS2)
2059 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2060 int __user
*, parent_tidptr
,
2061 int __user
*, child_tidptr
,
2063 #elif defined(CONFIG_CLONE_BACKWARDS3)
2064 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2066 int __user
*, parent_tidptr
,
2067 int __user
*, child_tidptr
,
2070 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2071 int __user
*, parent_tidptr
,
2072 int __user
*, child_tidptr
,
2076 return _do_fork(clone_flags
, newsp
, 0, parent_tidptr
, child_tidptr
, tls
);
2080 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2082 struct task_struct
*leader
, *parent
, *child
;
2085 read_lock(&tasklist_lock
);
2086 leader
= top
= top
->group_leader
;
2088 for_each_thread(leader
, parent
) {
2089 list_for_each_entry(child
, &parent
->children
, sibling
) {
2090 res
= visitor(child
, data
);
2102 if (leader
!= top
) {
2104 parent
= child
->real_parent
;
2105 leader
= parent
->group_leader
;
2109 read_unlock(&tasklist_lock
);
2112 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2113 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2116 static void sighand_ctor(void *data
)
2118 struct sighand_struct
*sighand
= data
;
2120 spin_lock_init(&sighand
->siglock
);
2121 init_waitqueue_head(&sighand
->signalfd_wqh
);
2124 void __init
proc_caches_init(void)
2126 sighand_cachep
= kmem_cache_create("sighand_cache",
2127 sizeof(struct sighand_struct
), 0,
2128 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_DESTROY_BY_RCU
|
2129 SLAB_NOTRACK
|SLAB_ACCOUNT
, sighand_ctor
);
2130 signal_cachep
= kmem_cache_create("signal_cache",
2131 sizeof(struct signal_struct
), 0,
2132 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2134 files_cachep
= kmem_cache_create("files_cache",
2135 sizeof(struct files_struct
), 0,
2136 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2138 fs_cachep
= kmem_cache_create("fs_cache",
2139 sizeof(struct fs_struct
), 0,
2140 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2143 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2144 * whole struct cpumask for the OFFSTACK case. We could change
2145 * this to *only* allocate as much of it as required by the
2146 * maximum number of CPU's we can ever have. The cpumask_allocation
2147 * is at the end of the structure, exactly for that reason.
2149 mm_cachep
= kmem_cache_create("mm_struct",
2150 sizeof(struct mm_struct
), ARCH_MIN_MMSTRUCT_ALIGN
,
2151 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2153 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2155 nsproxy_cache_init();
2159 * Check constraints on flags passed to the unshare system call.
2161 static int check_unshare_flags(unsigned long unshare_flags
)
2163 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2164 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2165 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2166 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2169 * Not implemented, but pretend it works if there is nothing
2170 * to unshare. Note that unsharing the address space or the
2171 * signal handlers also need to unshare the signal queues (aka
2174 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2175 if (!thread_group_empty(current
))
2178 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2179 if (atomic_read(¤t
->sighand
->count
) > 1)
2182 if (unshare_flags
& CLONE_VM
) {
2183 if (!current_is_single_threaded())
2191 * Unshare the filesystem structure if it is being shared
2193 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2195 struct fs_struct
*fs
= current
->fs
;
2197 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2200 /* don't need lock here; in the worst case we'll do useless copy */
2204 *new_fsp
= copy_fs_struct(fs
);
2212 * Unshare file descriptor table if it is being shared
2214 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2216 struct files_struct
*fd
= current
->files
;
2219 if ((unshare_flags
& CLONE_FILES
) &&
2220 (fd
&& atomic_read(&fd
->count
) > 1)) {
2221 *new_fdp
= dup_fd(fd
, &error
);
2230 * unshare allows a process to 'unshare' part of the process
2231 * context which was originally shared using clone. copy_*
2232 * functions used by do_fork() cannot be used here directly
2233 * because they modify an inactive task_struct that is being
2234 * constructed. Here we are modifying the current, active,
2237 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2239 struct fs_struct
*fs
, *new_fs
= NULL
;
2240 struct files_struct
*fd
, *new_fd
= NULL
;
2241 struct cred
*new_cred
= NULL
;
2242 struct nsproxy
*new_nsproxy
= NULL
;
2247 * If unsharing a user namespace must also unshare the thread group
2248 * and unshare the filesystem root and working directories.
2250 if (unshare_flags
& CLONE_NEWUSER
)
2251 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2253 * If unsharing vm, must also unshare signal handlers.
2255 if (unshare_flags
& CLONE_VM
)
2256 unshare_flags
|= CLONE_SIGHAND
;
2258 * If unsharing a signal handlers, must also unshare the signal queues.
2260 if (unshare_flags
& CLONE_SIGHAND
)
2261 unshare_flags
|= CLONE_THREAD
;
2263 * If unsharing namespace, must also unshare filesystem information.
2265 if (unshare_flags
& CLONE_NEWNS
)
2266 unshare_flags
|= CLONE_FS
;
2268 err
= check_unshare_flags(unshare_flags
);
2270 goto bad_unshare_out
;
2272 * CLONE_NEWIPC must also detach from the undolist: after switching
2273 * to a new ipc namespace, the semaphore arrays from the old
2274 * namespace are unreachable.
2276 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2278 err
= unshare_fs(unshare_flags
, &new_fs
);
2280 goto bad_unshare_out
;
2281 err
= unshare_fd(unshare_flags
, &new_fd
);
2283 goto bad_unshare_cleanup_fs
;
2284 err
= unshare_userns(unshare_flags
, &new_cred
);
2286 goto bad_unshare_cleanup_fd
;
2287 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2290 goto bad_unshare_cleanup_cred
;
2292 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2295 * CLONE_SYSVSEM is equivalent to sys_exit().
2299 if (unshare_flags
& CLONE_NEWIPC
) {
2300 /* Orphan segments in old ns (see sem above). */
2302 shm_init_task(current
);
2306 switch_task_namespaces(current
, new_nsproxy
);
2312 spin_lock(&fs
->lock
);
2313 current
->fs
= new_fs
;
2318 spin_unlock(&fs
->lock
);
2322 fd
= current
->files
;
2323 current
->files
= new_fd
;
2327 task_unlock(current
);
2330 /* Install the new user namespace */
2331 commit_creds(new_cred
);
2336 bad_unshare_cleanup_cred
:
2339 bad_unshare_cleanup_fd
:
2341 put_files_struct(new_fd
);
2343 bad_unshare_cleanup_fs
:
2345 free_fs_struct(new_fs
);
2352 * Helper to unshare the files of the current task.
2353 * We don't want to expose copy_files internals to
2354 * the exec layer of the kernel.
2357 int unshare_files(struct files_struct
**displaced
)
2359 struct task_struct
*task
= current
;
2360 struct files_struct
*copy
= NULL
;
2363 error
= unshare_fd(CLONE_FILES
, ©
);
2364 if (error
|| !copy
) {
2368 *displaced
= task
->files
;
2375 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2376 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2380 int threads
= max_threads
;
2381 int min
= MIN_THREADS
;
2382 int max
= MAX_THREADS
;
2389 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2393 set_max_threads(threads
);