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/sched/stat.h>
21 #include <linux/rtmutex.h>
22 #include <linux/init.h>
23 #include <linux/unistd.h>
24 #include <linux/module.h>
25 #include <linux/vmalloc.h>
26 #include <linux/completion.h>
27 #include <linux/personality.h>
28 #include <linux/mempolicy.h>
29 #include <linux/sem.h>
30 #include <linux/file.h>
31 #include <linux/fdtable.h>
32 #include <linux/iocontext.h>
33 #include <linux/key.h>
34 #include <linux/binfmts.h>
35 #include <linux/mman.h>
36 #include <linux/mmu_notifier.h>
39 #include <linux/vmacache.h>
40 #include <linux/nsproxy.h>
41 #include <linux/capability.h>
42 #include <linux/cpu.h>
43 #include <linux/cgroup.h>
44 #include <linux/security.h>
45 #include <linux/hugetlb.h>
46 #include <linux/seccomp.h>
47 #include <linux/swap.h>
48 #include <linux/syscalls.h>
49 #include <linux/jiffies.h>
50 #include <linux/futex.h>
51 #include <linux/compat.h>
52 #include <linux/kthread.h>
53 #include <linux/task_io_accounting_ops.h>
54 #include <linux/rcupdate.h>
55 #include <linux/ptrace.h>
56 #include <linux/mount.h>
57 #include <linux/audit.h>
58 #include <linux/memcontrol.h>
59 #include <linux/ftrace.h>
60 #include <linux/proc_fs.h>
61 #include <linux/profile.h>
62 #include <linux/rmap.h>
63 #include <linux/ksm.h>
64 #include <linux/acct.h>
65 #include <linux/userfaultfd_k.h>
66 #include <linux/tsacct_kern.h>
67 #include <linux/cn_proc.h>
68 #include <linux/freezer.h>
69 #include <linux/delayacct.h>
70 #include <linux/taskstats_kern.h>
71 #include <linux/random.h>
72 #include <linux/tty.h>
73 #include <linux/blkdev.h>
74 #include <linux/fs_struct.h>
75 #include <linux/magic.h>
76 #include <linux/perf_event.h>
77 #include <linux/posix-timers.h>
78 #include <linux/user-return-notifier.h>
79 #include <linux/oom.h>
80 #include <linux/khugepaged.h>
81 #include <linux/signalfd.h>
82 #include <linux/uprobes.h>
83 #include <linux/aio.h>
84 #include <linux/compiler.h>
85 #include <linux/sysctl.h>
86 #include <linux/kcov.h>
88 #include <asm/pgtable.h>
89 #include <asm/pgalloc.h>
90 #include <linux/uaccess.h>
91 #include <asm/mmu_context.h>
92 #include <asm/cacheflush.h>
93 #include <asm/tlbflush.h>
95 #include <trace/events/sched.h>
97 #define CREATE_TRACE_POINTS
98 #include <trace/events/task.h>
101 * Minimum number of threads to boot the kernel
103 #define MIN_THREADS 20
106 * Maximum number of threads
108 #define MAX_THREADS FUTEX_TID_MASK
111 * Protected counters by write_lock_irq(&tasklist_lock)
113 unsigned long total_forks
; /* Handle normal Linux uptimes. */
114 int nr_threads
; /* The idle threads do not count.. */
116 int max_threads
; /* tunable limit on nr_threads */
118 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
120 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
122 #ifdef CONFIG_PROVE_RCU
123 int lockdep_tasklist_lock_is_held(void)
125 return lockdep_is_held(&tasklist_lock
);
127 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
128 #endif /* #ifdef CONFIG_PROVE_RCU */
130 int nr_processes(void)
135 for_each_possible_cpu(cpu
)
136 total
+= per_cpu(process_counts
, cpu
);
141 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
145 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
146 static struct kmem_cache
*task_struct_cachep
;
148 static inline struct task_struct
*alloc_task_struct_node(int node
)
150 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
153 static inline void free_task_struct(struct task_struct
*tsk
)
155 kmem_cache_free(task_struct_cachep
, tsk
);
159 void __weak
arch_release_thread_stack(unsigned long *stack
)
163 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
166 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
167 * kmemcache based allocator.
169 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
171 #ifdef CONFIG_VMAP_STACK
173 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
174 * flush. Try to minimize the number of calls by caching stacks.
176 #define NR_CACHED_STACKS 2
177 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
180 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
182 #ifdef CONFIG_VMAP_STACK
187 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
188 struct vm_struct
*s
= this_cpu_read(cached_stacks
[i
]);
192 this_cpu_write(cached_stacks
[i
], NULL
);
194 tsk
->stack_vm_area
= s
;
200 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_SIZE
,
201 VMALLOC_START
, VMALLOC_END
,
202 THREADINFO_GFP
| __GFP_HIGHMEM
,
204 0, node
, __builtin_return_address(0));
207 * We can't call find_vm_area() in interrupt context, and
208 * free_thread_stack() can be called in interrupt context,
209 * so cache the vm_struct.
212 tsk
->stack_vm_area
= find_vm_area(stack
);
215 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
218 return page
? page_address(page
) : NULL
;
222 static inline void free_thread_stack(struct task_struct
*tsk
)
224 #ifdef CONFIG_VMAP_STACK
225 if (task_stack_vm_area(tsk
)) {
229 local_irq_save(flags
);
230 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
231 if (this_cpu_read(cached_stacks
[i
]))
234 this_cpu_write(cached_stacks
[i
], tsk
->stack_vm_area
);
235 local_irq_restore(flags
);
238 local_irq_restore(flags
);
240 vfree_atomic(tsk
->stack
);
245 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
248 static struct kmem_cache
*thread_stack_cache
;
250 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
253 return kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
256 static void free_thread_stack(struct task_struct
*tsk
)
258 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
261 void thread_stack_cache_init(void)
263 thread_stack_cache
= kmem_cache_create("thread_stack", THREAD_SIZE
,
264 THREAD_SIZE
, 0, NULL
);
265 BUG_ON(thread_stack_cache
== NULL
);
270 /* SLAB cache for signal_struct structures (tsk->signal) */
271 static struct kmem_cache
*signal_cachep
;
273 /* SLAB cache for sighand_struct structures (tsk->sighand) */
274 struct kmem_cache
*sighand_cachep
;
276 /* SLAB cache for files_struct structures (tsk->files) */
277 struct kmem_cache
*files_cachep
;
279 /* SLAB cache for fs_struct structures (tsk->fs) */
280 struct kmem_cache
*fs_cachep
;
282 /* SLAB cache for vm_area_struct structures */
283 struct kmem_cache
*vm_area_cachep
;
285 /* SLAB cache for mm_struct structures (tsk->mm) */
286 static struct kmem_cache
*mm_cachep
;
288 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
290 void *stack
= task_stack_page(tsk
);
291 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
293 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
298 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
300 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
301 mod_zone_page_state(page_zone(vm
->pages
[i
]),
303 PAGE_SIZE
/ 1024 * account
);
306 /* All stack pages belong to the same memcg. */
307 memcg_kmem_update_page_stat(vm
->pages
[0], MEMCG_KERNEL_STACK_KB
,
308 account
* (THREAD_SIZE
/ 1024));
311 * All stack pages are in the same zone and belong to the
314 struct page
*first_page
= virt_to_page(stack
);
316 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
317 THREAD_SIZE
/ 1024 * account
);
319 memcg_kmem_update_page_stat(first_page
, MEMCG_KERNEL_STACK_KB
,
320 account
* (THREAD_SIZE
/ 1024));
324 static void release_task_stack(struct task_struct
*tsk
)
326 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
327 return; /* Better to leak the stack than to free prematurely */
329 account_kernel_stack(tsk
, -1);
330 arch_release_thread_stack(tsk
->stack
);
331 free_thread_stack(tsk
);
333 #ifdef CONFIG_VMAP_STACK
334 tsk
->stack_vm_area
= NULL
;
338 #ifdef CONFIG_THREAD_INFO_IN_TASK
339 void put_task_stack(struct task_struct
*tsk
)
341 if (atomic_dec_and_test(&tsk
->stack_refcount
))
342 release_task_stack(tsk
);
346 void free_task(struct task_struct
*tsk
)
348 #ifndef CONFIG_THREAD_INFO_IN_TASK
350 * The task is finally done with both the stack and thread_info,
353 release_task_stack(tsk
);
356 * If the task had a separate stack allocation, it should be gone
359 WARN_ON_ONCE(atomic_read(&tsk
->stack_refcount
) != 0);
361 rt_mutex_debug_task_free(tsk
);
362 ftrace_graph_exit_task(tsk
);
363 put_seccomp_filter(tsk
);
364 arch_release_task_struct(tsk
);
365 if (tsk
->flags
& PF_KTHREAD
)
366 free_kthread_struct(tsk
);
367 free_task_struct(tsk
);
369 EXPORT_SYMBOL(free_task
);
371 static inline void free_signal_struct(struct signal_struct
*sig
)
373 taskstats_tgid_free(sig
);
374 sched_autogroup_exit(sig
);
376 * __mmdrop is not safe to call from softirq context on x86 due to
377 * pgd_dtor so postpone it to the async context
380 mmdrop_async(sig
->oom_mm
);
381 kmem_cache_free(signal_cachep
, sig
);
384 static inline void put_signal_struct(struct signal_struct
*sig
)
386 if (atomic_dec_and_test(&sig
->sigcnt
))
387 free_signal_struct(sig
);
390 void __put_task_struct(struct task_struct
*tsk
)
392 WARN_ON(!tsk
->exit_state
);
393 WARN_ON(atomic_read(&tsk
->usage
));
394 WARN_ON(tsk
== current
);
398 security_task_free(tsk
);
400 delayacct_tsk_free(tsk
);
401 put_signal_struct(tsk
->signal
);
403 if (!profile_handoff_task(tsk
))
406 EXPORT_SYMBOL_GPL(__put_task_struct
);
408 void __init __weak
arch_task_cache_init(void) { }
413 static void set_max_threads(unsigned int max_threads_suggested
)
418 * The number of threads shall be limited such that the thread
419 * structures may only consume a small part of the available memory.
421 if (fls64(totalram_pages
) + fls64(PAGE_SIZE
) > 64)
422 threads
= MAX_THREADS
;
424 threads
= div64_u64((u64
) totalram_pages
* (u64
) PAGE_SIZE
,
425 (u64
) THREAD_SIZE
* 8UL);
427 if (threads
> max_threads_suggested
)
428 threads
= max_threads_suggested
;
430 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
433 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
434 /* Initialized by the architecture: */
435 int arch_task_struct_size __read_mostly
;
438 void __init
fork_init(void)
441 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
442 #ifndef ARCH_MIN_TASKALIGN
443 #define ARCH_MIN_TASKALIGN 0
445 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
447 /* create a slab on which task_structs can be allocated */
448 task_struct_cachep
= kmem_cache_create("task_struct",
449 arch_task_struct_size
, align
,
450 SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
, NULL
);
453 /* do the arch specific task caches init */
454 arch_task_cache_init();
456 set_max_threads(MAX_THREADS
);
458 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
459 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
460 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
461 init_task
.signal
->rlim
[RLIMIT_NPROC
];
463 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
464 init_user_ns
.ucount_max
[i
] = max_threads
/2;
468 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
469 struct task_struct
*src
)
475 void set_task_stack_end_magic(struct task_struct
*tsk
)
477 unsigned long *stackend
;
479 stackend
= end_of_stack(tsk
);
480 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
483 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
485 struct task_struct
*tsk
;
486 unsigned long *stack
;
487 struct vm_struct
*stack_vm_area
;
490 if (node
== NUMA_NO_NODE
)
491 node
= tsk_fork_get_node(orig
);
492 tsk
= alloc_task_struct_node(node
);
496 stack
= alloc_thread_stack_node(tsk
, node
);
500 stack_vm_area
= task_stack_vm_area(tsk
);
502 err
= arch_dup_task_struct(tsk
, orig
);
505 * arch_dup_task_struct() clobbers the stack-related fields. Make
506 * sure they're properly initialized before using any stack-related
510 #ifdef CONFIG_VMAP_STACK
511 tsk
->stack_vm_area
= stack_vm_area
;
513 #ifdef CONFIG_THREAD_INFO_IN_TASK
514 atomic_set(&tsk
->stack_refcount
, 1);
520 #ifdef CONFIG_SECCOMP
522 * We must handle setting up seccomp filters once we're under
523 * the sighand lock in case orig has changed between now and
524 * then. Until then, filter must be NULL to avoid messing up
525 * the usage counts on the error path calling free_task.
527 tsk
->seccomp
.filter
= NULL
;
530 setup_thread_stack(tsk
, orig
);
531 clear_user_return_notifier(tsk
);
532 clear_tsk_need_resched(tsk
);
533 set_task_stack_end_magic(tsk
);
535 #ifdef CONFIG_CC_STACKPROTECTOR
536 tsk
->stack_canary
= get_random_int();
540 * One for us, one for whoever does the "release_task()" (usually
543 atomic_set(&tsk
->usage
, 2);
544 #ifdef CONFIG_BLK_DEV_IO_TRACE
547 tsk
->splice_pipe
= NULL
;
548 tsk
->task_frag
.page
= NULL
;
549 tsk
->wake_q
.next
= NULL
;
551 account_kernel_stack(tsk
, 1);
558 free_thread_stack(tsk
);
560 free_task_struct(tsk
);
565 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
566 struct mm_struct
*oldmm
)
568 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
569 struct rb_node
**rb_link
, *rb_parent
;
571 unsigned long charge
;
574 uprobe_start_dup_mmap();
575 if (down_write_killable(&oldmm
->mmap_sem
)) {
577 goto fail_uprobe_end
;
579 flush_cache_dup_mm(oldmm
);
580 uprobe_dup_mmap(oldmm
, mm
);
582 * Not linked in yet - no deadlock potential:
584 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
586 /* No ordering required: file already has been exposed. */
587 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
589 mm
->total_vm
= oldmm
->total_vm
;
590 mm
->data_vm
= oldmm
->data_vm
;
591 mm
->exec_vm
= oldmm
->exec_vm
;
592 mm
->stack_vm
= oldmm
->stack_vm
;
594 rb_link
= &mm
->mm_rb
.rb_node
;
597 retval
= ksm_fork(mm
, oldmm
);
600 retval
= khugepaged_fork(mm
, oldmm
);
605 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
608 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
609 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
613 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
614 unsigned long len
= vma_pages(mpnt
);
616 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
620 tmp
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
624 INIT_LIST_HEAD(&tmp
->anon_vma_chain
);
625 retval
= vma_dup_policy(mpnt
, tmp
);
627 goto fail_nomem_policy
;
629 retval
= dup_userfaultfd(tmp
, &uf
);
631 goto fail_nomem_anon_vma_fork
;
632 if (anon_vma_fork(tmp
, mpnt
))
633 goto fail_nomem_anon_vma_fork
;
634 tmp
->vm_flags
&= ~(VM_LOCKED
| VM_LOCKONFAULT
);
635 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
638 struct inode
*inode
= file_inode(file
);
639 struct address_space
*mapping
= file
->f_mapping
;
642 if (tmp
->vm_flags
& VM_DENYWRITE
)
643 atomic_dec(&inode
->i_writecount
);
644 i_mmap_lock_write(mapping
);
645 if (tmp
->vm_flags
& VM_SHARED
)
646 atomic_inc(&mapping
->i_mmap_writable
);
647 flush_dcache_mmap_lock(mapping
);
648 /* insert tmp into the share list, just after mpnt */
649 vma_interval_tree_insert_after(tmp
, mpnt
,
651 flush_dcache_mmap_unlock(mapping
);
652 i_mmap_unlock_write(mapping
);
656 * Clear hugetlb-related page reserves for children. This only
657 * affects MAP_PRIVATE mappings. Faults generated by the child
658 * are not guaranteed to succeed, even if read-only
660 if (is_vm_hugetlb_page(tmp
))
661 reset_vma_resv_huge_pages(tmp
);
664 * Link in the new vma and copy the page table entries.
667 pprev
= &tmp
->vm_next
;
671 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
672 rb_link
= &tmp
->vm_rb
.rb_right
;
673 rb_parent
= &tmp
->vm_rb
;
676 retval
= copy_page_range(mm
, oldmm
, mpnt
);
678 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
679 tmp
->vm_ops
->open(tmp
);
684 /* a new mm has just been created */
685 arch_dup_mmap(oldmm
, mm
);
688 up_write(&mm
->mmap_sem
);
690 up_write(&oldmm
->mmap_sem
);
691 dup_userfaultfd_complete(&uf
);
693 uprobe_end_dup_mmap();
695 fail_nomem_anon_vma_fork
:
696 mpol_put(vma_policy(tmp
));
698 kmem_cache_free(vm_area_cachep
, tmp
);
701 vm_unacct_memory(charge
);
705 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
707 mm
->pgd
= pgd_alloc(mm
);
708 if (unlikely(!mm
->pgd
))
713 static inline void mm_free_pgd(struct mm_struct
*mm
)
715 pgd_free(mm
, mm
->pgd
);
718 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
720 down_write(&oldmm
->mmap_sem
);
721 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
722 up_write(&oldmm
->mmap_sem
);
725 #define mm_alloc_pgd(mm) (0)
726 #define mm_free_pgd(mm)
727 #endif /* CONFIG_MMU */
729 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
731 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
732 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
734 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
736 static int __init
coredump_filter_setup(char *s
)
738 default_dump_filter
=
739 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
740 MMF_DUMP_FILTER_MASK
;
744 __setup("coredump_filter=", coredump_filter_setup
);
746 #include <linux/init_task.h>
748 static void mm_init_aio(struct mm_struct
*mm
)
751 spin_lock_init(&mm
->ioctx_lock
);
752 mm
->ioctx_table
= NULL
;
756 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
763 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
764 struct user_namespace
*user_ns
)
768 mm
->vmacache_seqnum
= 0;
769 atomic_set(&mm
->mm_users
, 1);
770 atomic_set(&mm
->mm_count
, 1);
771 init_rwsem(&mm
->mmap_sem
);
772 INIT_LIST_HEAD(&mm
->mmlist
);
773 mm
->core_state
= NULL
;
774 atomic_long_set(&mm
->nr_ptes
, 0);
779 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
780 spin_lock_init(&mm
->page_table_lock
);
783 mm_init_owner(mm
, p
);
784 mmu_notifier_mm_init(mm
);
785 clear_tlb_flush_pending(mm
);
786 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
787 mm
->pmd_huge_pte
= NULL
;
791 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
792 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
794 mm
->flags
= default_dump_filter
;
798 if (mm_alloc_pgd(mm
))
801 if (init_new_context(p
, mm
))
804 mm
->user_ns
= get_user_ns(user_ns
);
814 static void check_mm(struct mm_struct
*mm
)
818 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
819 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
822 printk(KERN_ALERT
"BUG: Bad rss-counter state "
823 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
826 if (atomic_long_read(&mm
->nr_ptes
))
827 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
828 atomic_long_read(&mm
->nr_ptes
));
830 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
833 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
834 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
839 * Allocate and initialize an mm_struct.
841 struct mm_struct
*mm_alloc(void)
843 struct mm_struct
*mm
;
849 memset(mm
, 0, sizeof(*mm
));
850 return mm_init(mm
, current
, current_user_ns());
854 * Called when the last reference to the mm
855 * is dropped: either by a lazy thread or by
856 * mmput. Free the page directory and the mm.
858 void __mmdrop(struct mm_struct
*mm
)
860 BUG_ON(mm
== &init_mm
);
863 mmu_notifier_mm_destroy(mm
);
865 put_user_ns(mm
->user_ns
);
868 EXPORT_SYMBOL_GPL(__mmdrop
);
870 static inline void __mmput(struct mm_struct
*mm
)
872 VM_BUG_ON(atomic_read(&mm
->mm_users
));
874 uprobe_clear_state(mm
);
877 khugepaged_exit(mm
); /* must run before exit_mmap */
879 mm_put_huge_zero_page(mm
);
880 set_mm_exe_file(mm
, NULL
);
881 if (!list_empty(&mm
->mmlist
)) {
882 spin_lock(&mmlist_lock
);
883 list_del(&mm
->mmlist
);
884 spin_unlock(&mmlist_lock
);
887 module_put(mm
->binfmt
->module
);
888 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
893 * Decrement the use count and release all resources for an mm.
895 void mmput(struct mm_struct
*mm
)
899 if (atomic_dec_and_test(&mm
->mm_users
))
902 EXPORT_SYMBOL_GPL(mmput
);
905 static void mmput_async_fn(struct work_struct
*work
)
907 struct mm_struct
*mm
= container_of(work
, struct mm_struct
, async_put_work
);
911 void mmput_async(struct mm_struct
*mm
)
913 if (atomic_dec_and_test(&mm
->mm_users
)) {
914 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
915 schedule_work(&mm
->async_put_work
);
921 * set_mm_exe_file - change a reference to the mm's executable file
923 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
925 * Main users are mmput() and sys_execve(). Callers prevent concurrent
926 * invocations: in mmput() nobody alive left, in execve task is single
927 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
928 * mm->exe_file, but does so without using set_mm_exe_file() in order
929 * to do avoid the need for any locks.
931 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
933 struct file
*old_exe_file
;
936 * It is safe to dereference the exe_file without RCU as
937 * this function is only called if nobody else can access
938 * this mm -- see comment above for justification.
940 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
943 get_file(new_exe_file
);
944 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
950 * get_mm_exe_file - acquire a reference to the mm's executable file
952 * Returns %NULL if mm has no associated executable file.
953 * User must release file via fput().
955 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
957 struct file
*exe_file
;
960 exe_file
= rcu_dereference(mm
->exe_file
);
961 if (exe_file
&& !get_file_rcu(exe_file
))
966 EXPORT_SYMBOL(get_mm_exe_file
);
969 * get_task_exe_file - acquire a reference to the task's executable file
971 * Returns %NULL if task's mm (if any) has no associated executable file or
972 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
973 * User must release file via fput().
975 struct file
*get_task_exe_file(struct task_struct
*task
)
977 struct file
*exe_file
= NULL
;
978 struct mm_struct
*mm
;
983 if (!(task
->flags
& PF_KTHREAD
))
984 exe_file
= get_mm_exe_file(mm
);
989 EXPORT_SYMBOL(get_task_exe_file
);
992 * get_task_mm - acquire a reference to the task's mm
994 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
995 * this kernel workthread has transiently adopted a user mm with use_mm,
996 * to do its AIO) is not set and if so returns a reference to it, after
997 * bumping up the use count. User must release the mm via mmput()
998 * after use. Typically used by /proc and ptrace.
1000 struct mm_struct
*get_task_mm(struct task_struct
*task
)
1002 struct mm_struct
*mm
;
1007 if (task
->flags
& PF_KTHREAD
)
1015 EXPORT_SYMBOL_GPL(get_task_mm
);
1017 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1019 struct mm_struct
*mm
;
1022 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1024 return ERR_PTR(err
);
1026 mm
= get_task_mm(task
);
1027 if (mm
&& mm
!= current
->mm
&&
1028 !ptrace_may_access(task
, mode
)) {
1030 mm
= ERR_PTR(-EACCES
);
1032 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1037 static void complete_vfork_done(struct task_struct
*tsk
)
1039 struct completion
*vfork
;
1042 vfork
= tsk
->vfork_done
;
1043 if (likely(vfork
)) {
1044 tsk
->vfork_done
= NULL
;
1050 static int wait_for_vfork_done(struct task_struct
*child
,
1051 struct completion
*vfork
)
1055 freezer_do_not_count();
1056 killed
= wait_for_completion_killable(vfork
);
1061 child
->vfork_done
= NULL
;
1065 put_task_struct(child
);
1069 /* Please note the differences between mmput and mm_release.
1070 * mmput is called whenever we stop holding onto a mm_struct,
1071 * error success whatever.
1073 * mm_release is called after a mm_struct has been removed
1074 * from the current process.
1076 * This difference is important for error handling, when we
1077 * only half set up a mm_struct for a new process and need to restore
1078 * the old one. Because we mmput the new mm_struct before
1079 * restoring the old one. . .
1080 * Eric Biederman 10 January 1998
1082 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1084 /* Get rid of any futexes when releasing the mm */
1086 if (unlikely(tsk
->robust_list
)) {
1087 exit_robust_list(tsk
);
1088 tsk
->robust_list
= NULL
;
1090 #ifdef CONFIG_COMPAT
1091 if (unlikely(tsk
->compat_robust_list
)) {
1092 compat_exit_robust_list(tsk
);
1093 tsk
->compat_robust_list
= NULL
;
1096 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
1097 exit_pi_state_list(tsk
);
1100 uprobe_free_utask(tsk
);
1102 /* Get rid of any cached register state */
1103 deactivate_mm(tsk
, mm
);
1106 * Signal userspace if we're not exiting with a core dump
1107 * because we want to leave the value intact for debugging
1110 if (tsk
->clear_child_tid
) {
1111 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1112 atomic_read(&mm
->mm_users
) > 1) {
1114 * We don't check the error code - if userspace has
1115 * not set up a proper pointer then tough luck.
1117 put_user(0, tsk
->clear_child_tid
);
1118 sys_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1121 tsk
->clear_child_tid
= NULL
;
1125 * All done, finally we can wake up parent and return this mm to him.
1126 * Also kthread_stop() uses this completion for synchronization.
1128 if (tsk
->vfork_done
)
1129 complete_vfork_done(tsk
);
1133 * Allocate a new mm structure and copy contents from the
1134 * mm structure of the passed in task structure.
1136 static struct mm_struct
*dup_mm(struct task_struct
*tsk
)
1138 struct mm_struct
*mm
, *oldmm
= current
->mm
;
1145 memcpy(mm
, oldmm
, sizeof(*mm
));
1147 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1150 err
= dup_mmap(mm
, oldmm
);
1154 mm
->hiwater_rss
= get_mm_rss(mm
);
1155 mm
->hiwater_vm
= mm
->total_vm
;
1157 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1163 /* don't put binfmt in mmput, we haven't got module yet */
1171 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1173 struct mm_struct
*mm
, *oldmm
;
1176 tsk
->min_flt
= tsk
->maj_flt
= 0;
1177 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1178 #ifdef CONFIG_DETECT_HUNG_TASK
1179 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1183 tsk
->active_mm
= NULL
;
1186 * Are we cloning a kernel thread?
1188 * We need to steal a active VM for that..
1190 oldmm
= current
->mm
;
1194 /* initialize the new vmacache entries */
1195 vmacache_flush(tsk
);
1197 if (clone_flags
& CLONE_VM
) {
1210 tsk
->active_mm
= mm
;
1217 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1219 struct fs_struct
*fs
= current
->fs
;
1220 if (clone_flags
& CLONE_FS
) {
1221 /* tsk->fs is already what we want */
1222 spin_lock(&fs
->lock
);
1224 spin_unlock(&fs
->lock
);
1228 spin_unlock(&fs
->lock
);
1231 tsk
->fs
= copy_fs_struct(fs
);
1237 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1239 struct files_struct
*oldf
, *newf
;
1243 * A background process may not have any files ...
1245 oldf
= current
->files
;
1249 if (clone_flags
& CLONE_FILES
) {
1250 atomic_inc(&oldf
->count
);
1254 newf
= dup_fd(oldf
, &error
);
1264 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1267 struct io_context
*ioc
= current
->io_context
;
1268 struct io_context
*new_ioc
;
1273 * Share io context with parent, if CLONE_IO is set
1275 if (clone_flags
& CLONE_IO
) {
1277 tsk
->io_context
= ioc
;
1278 } else if (ioprio_valid(ioc
->ioprio
)) {
1279 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1280 if (unlikely(!new_ioc
))
1283 new_ioc
->ioprio
= ioc
->ioprio
;
1284 put_io_context(new_ioc
);
1290 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1292 struct sighand_struct
*sig
;
1294 if (clone_flags
& CLONE_SIGHAND
) {
1295 atomic_inc(¤t
->sighand
->count
);
1298 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1299 rcu_assign_pointer(tsk
->sighand
, sig
);
1303 atomic_set(&sig
->count
, 1);
1304 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1308 void __cleanup_sighand(struct sighand_struct
*sighand
)
1310 if (atomic_dec_and_test(&sighand
->count
)) {
1311 signalfd_cleanup(sighand
);
1313 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1314 * without an RCU grace period, see __lock_task_sighand().
1316 kmem_cache_free(sighand_cachep
, sighand
);
1320 #ifdef CONFIG_POSIX_TIMERS
1322 * Initialize POSIX timer handling for a thread group.
1324 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1326 unsigned long cpu_limit
;
1328 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1329 if (cpu_limit
!= RLIM_INFINITY
) {
1330 sig
->cputime_expires
.prof_exp
= cpu_limit
* NSEC_PER_SEC
;
1331 sig
->cputimer
.running
= true;
1334 /* The timer lists. */
1335 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1336 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1337 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1340 static inline void posix_cpu_timers_init_group(struct signal_struct
*sig
) { }
1343 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1345 struct signal_struct
*sig
;
1347 if (clone_flags
& CLONE_THREAD
)
1350 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1355 sig
->nr_threads
= 1;
1356 atomic_set(&sig
->live
, 1);
1357 atomic_set(&sig
->sigcnt
, 1);
1359 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1360 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1361 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1363 init_waitqueue_head(&sig
->wait_chldexit
);
1364 sig
->curr_target
= tsk
;
1365 init_sigpending(&sig
->shared_pending
);
1366 seqlock_init(&sig
->stats_lock
);
1367 prev_cputime_init(&sig
->prev_cputime
);
1369 #ifdef CONFIG_POSIX_TIMERS
1370 INIT_LIST_HEAD(&sig
->posix_timers
);
1371 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1372 sig
->real_timer
.function
= it_real_fn
;
1375 task_lock(current
->group_leader
);
1376 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1377 task_unlock(current
->group_leader
);
1379 posix_cpu_timers_init_group(sig
);
1381 tty_audit_fork(sig
);
1382 sched_autogroup_fork(sig
);
1384 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1385 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1387 mutex_init(&sig
->cred_guard_mutex
);
1392 static void copy_seccomp(struct task_struct
*p
)
1394 #ifdef CONFIG_SECCOMP
1396 * Must be called with sighand->lock held, which is common to
1397 * all threads in the group. Holding cred_guard_mutex is not
1398 * needed because this new task is not yet running and cannot
1401 assert_spin_locked(¤t
->sighand
->siglock
);
1403 /* Ref-count the new filter user, and assign it. */
1404 get_seccomp_filter(current
);
1405 p
->seccomp
= current
->seccomp
;
1408 * Explicitly enable no_new_privs here in case it got set
1409 * between the task_struct being duplicated and holding the
1410 * sighand lock. The seccomp state and nnp must be in sync.
1412 if (task_no_new_privs(current
))
1413 task_set_no_new_privs(p
);
1416 * If the parent gained a seccomp mode after copying thread
1417 * flags and between before we held the sighand lock, we have
1418 * to manually enable the seccomp thread flag here.
1420 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1421 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1425 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1427 current
->clear_child_tid
= tidptr
;
1429 return task_pid_vnr(current
);
1432 static void rt_mutex_init_task(struct task_struct
*p
)
1434 raw_spin_lock_init(&p
->pi_lock
);
1435 #ifdef CONFIG_RT_MUTEXES
1436 p
->pi_waiters
= RB_ROOT
;
1437 p
->pi_waiters_leftmost
= NULL
;
1438 p
->pi_blocked_on
= NULL
;
1442 #ifdef CONFIG_POSIX_TIMERS
1444 * Initialize POSIX timer handling for a single task.
1446 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1448 tsk
->cputime_expires
.prof_exp
= 0;
1449 tsk
->cputime_expires
.virt_exp
= 0;
1450 tsk
->cputime_expires
.sched_exp
= 0;
1451 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1452 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1453 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1456 static inline void posix_cpu_timers_init(struct task_struct
*tsk
) { }
1460 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1462 task
->pids
[type
].pid
= pid
;
1466 * This creates a new process as a copy of the old one,
1467 * but does not actually start it yet.
1469 * It copies the registers, and all the appropriate
1470 * parts of the process environment (as per the clone
1471 * flags). The actual kick-off is left to the caller.
1473 static __latent_entropy
struct task_struct
*copy_process(
1474 unsigned long clone_flags
,
1475 unsigned long stack_start
,
1476 unsigned long stack_size
,
1477 int __user
*child_tidptr
,
1484 struct task_struct
*p
;
1486 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1487 return ERR_PTR(-EINVAL
);
1489 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1490 return ERR_PTR(-EINVAL
);
1493 * Thread groups must share signals as well, and detached threads
1494 * can only be started up within the thread group.
1496 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1497 return ERR_PTR(-EINVAL
);
1500 * Shared signal handlers imply shared VM. By way of the above,
1501 * thread groups also imply shared VM. Blocking this case allows
1502 * for various simplifications in other code.
1504 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1505 return ERR_PTR(-EINVAL
);
1508 * Siblings of global init remain as zombies on exit since they are
1509 * not reaped by their parent (swapper). To solve this and to avoid
1510 * multi-rooted process trees, prevent global and container-inits
1511 * from creating siblings.
1513 if ((clone_flags
& CLONE_PARENT
) &&
1514 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1515 return ERR_PTR(-EINVAL
);
1518 * If the new process will be in a different pid or user namespace
1519 * do not allow it to share a thread group with the forking task.
1521 if (clone_flags
& CLONE_THREAD
) {
1522 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1523 (task_active_pid_ns(current
) !=
1524 current
->nsproxy
->pid_ns_for_children
))
1525 return ERR_PTR(-EINVAL
);
1528 retval
= security_task_create(clone_flags
);
1533 p
= dup_task_struct(current
, node
);
1537 ftrace_graph_init_task(p
);
1539 rt_mutex_init_task(p
);
1541 #ifdef CONFIG_PROVE_LOCKING
1542 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1543 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1546 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1547 task_rlimit(p
, RLIMIT_NPROC
)) {
1548 if (p
->real_cred
->user
!= INIT_USER
&&
1549 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1552 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1554 retval
= copy_creds(p
, clone_flags
);
1559 * If multiple threads are within copy_process(), then this check
1560 * triggers too late. This doesn't hurt, the check is only there
1561 * to stop root fork bombs.
1564 if (nr_threads
>= max_threads
)
1565 goto bad_fork_cleanup_count
;
1567 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1568 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1569 p
->flags
|= PF_FORKNOEXEC
;
1570 INIT_LIST_HEAD(&p
->children
);
1571 INIT_LIST_HEAD(&p
->sibling
);
1572 rcu_copy_process(p
);
1573 p
->vfork_done
= NULL
;
1574 spin_lock_init(&p
->alloc_lock
);
1576 init_sigpending(&p
->pending
);
1578 p
->utime
= p
->stime
= p
->gtime
= 0;
1579 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1580 p
->utimescaled
= p
->stimescaled
= 0;
1582 prev_cputime_init(&p
->prev_cputime
);
1584 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1585 seqcount_init(&p
->vtime_seqcount
);
1587 p
->vtime_snap_whence
= VTIME_INACTIVE
;
1590 #if defined(SPLIT_RSS_COUNTING)
1591 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1594 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1596 task_io_accounting_init(&p
->ioac
);
1597 acct_clear_integrals(p
);
1599 posix_cpu_timers_init(p
);
1601 p
->start_time
= ktime_get_ns();
1602 p
->real_start_time
= ktime_get_boot_ns();
1603 p
->io_context
= NULL
;
1604 p
->audit_context
= NULL
;
1607 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1608 if (IS_ERR(p
->mempolicy
)) {
1609 retval
= PTR_ERR(p
->mempolicy
);
1610 p
->mempolicy
= NULL
;
1611 goto bad_fork_cleanup_threadgroup_lock
;
1614 #ifdef CONFIG_CPUSETS
1615 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1616 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1617 seqcount_init(&p
->mems_allowed_seq
);
1619 #ifdef CONFIG_TRACE_IRQFLAGS
1621 p
->hardirqs_enabled
= 0;
1622 p
->hardirq_enable_ip
= 0;
1623 p
->hardirq_enable_event
= 0;
1624 p
->hardirq_disable_ip
= _THIS_IP_
;
1625 p
->hardirq_disable_event
= 0;
1626 p
->softirqs_enabled
= 1;
1627 p
->softirq_enable_ip
= _THIS_IP_
;
1628 p
->softirq_enable_event
= 0;
1629 p
->softirq_disable_ip
= 0;
1630 p
->softirq_disable_event
= 0;
1631 p
->hardirq_context
= 0;
1632 p
->softirq_context
= 0;
1635 p
->pagefault_disabled
= 0;
1637 #ifdef CONFIG_LOCKDEP
1638 p
->lockdep_depth
= 0; /* no locks held yet */
1639 p
->curr_chain_key
= 0;
1640 p
->lockdep_recursion
= 0;
1643 #ifdef CONFIG_DEBUG_MUTEXES
1644 p
->blocked_on
= NULL
; /* not blocked yet */
1646 #ifdef CONFIG_BCACHE
1647 p
->sequential_io
= 0;
1648 p
->sequential_io_avg
= 0;
1651 /* Perform scheduler related setup. Assign this task to a CPU. */
1652 retval
= sched_fork(clone_flags
, p
);
1654 goto bad_fork_cleanup_policy
;
1656 retval
= perf_event_init_task(p
);
1658 goto bad_fork_cleanup_policy
;
1659 retval
= audit_alloc(p
);
1661 goto bad_fork_cleanup_perf
;
1662 /* copy all the process information */
1664 retval
= copy_semundo(clone_flags
, p
);
1666 goto bad_fork_cleanup_audit
;
1667 retval
= copy_files(clone_flags
, p
);
1669 goto bad_fork_cleanup_semundo
;
1670 retval
= copy_fs(clone_flags
, p
);
1672 goto bad_fork_cleanup_files
;
1673 retval
= copy_sighand(clone_flags
, p
);
1675 goto bad_fork_cleanup_fs
;
1676 retval
= copy_signal(clone_flags
, p
);
1678 goto bad_fork_cleanup_sighand
;
1679 retval
= copy_mm(clone_flags
, p
);
1681 goto bad_fork_cleanup_signal
;
1682 retval
= copy_namespaces(clone_flags
, p
);
1684 goto bad_fork_cleanup_mm
;
1685 retval
= copy_io(clone_flags
, p
);
1687 goto bad_fork_cleanup_namespaces
;
1688 retval
= copy_thread_tls(clone_flags
, stack_start
, stack_size
, p
, tls
);
1690 goto bad_fork_cleanup_io
;
1692 if (pid
!= &init_struct_pid
) {
1693 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
1695 retval
= PTR_ERR(pid
);
1696 goto bad_fork_cleanup_thread
;
1700 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
1702 * Clear TID on mm_release()?
1704 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
1709 p
->robust_list
= NULL
;
1710 #ifdef CONFIG_COMPAT
1711 p
->compat_robust_list
= NULL
;
1713 INIT_LIST_HEAD(&p
->pi_state_list
);
1714 p
->pi_state_cache
= NULL
;
1717 * sigaltstack should be cleared when sharing the same VM
1719 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
1723 * Syscall tracing and stepping should be turned off in the
1724 * child regardless of CLONE_PTRACE.
1726 user_disable_single_step(p
);
1727 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
1728 #ifdef TIF_SYSCALL_EMU
1729 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1731 clear_all_latency_tracing(p
);
1733 /* ok, now we should be set up.. */
1734 p
->pid
= pid_nr(pid
);
1735 if (clone_flags
& CLONE_THREAD
) {
1736 p
->exit_signal
= -1;
1737 p
->group_leader
= current
->group_leader
;
1738 p
->tgid
= current
->tgid
;
1740 if (clone_flags
& CLONE_PARENT
)
1741 p
->exit_signal
= current
->group_leader
->exit_signal
;
1743 p
->exit_signal
= (clone_flags
& CSIGNAL
);
1744 p
->group_leader
= p
;
1749 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
1750 p
->dirty_paused_when
= 0;
1752 p
->pdeath_signal
= 0;
1753 INIT_LIST_HEAD(&p
->thread_group
);
1754 p
->task_works
= NULL
;
1756 cgroup_threadgroup_change_begin(current
);
1758 * Ensure that the cgroup subsystem policies allow the new process to be
1759 * forked. It should be noted the the new process's css_set can be changed
1760 * between here and cgroup_post_fork() if an organisation operation is in
1763 retval
= cgroup_can_fork(p
);
1765 goto bad_fork_free_pid
;
1768 * Make it visible to the rest of the system, but dont wake it up yet.
1769 * Need tasklist lock for parent etc handling!
1771 write_lock_irq(&tasklist_lock
);
1773 /* CLONE_PARENT re-uses the old parent */
1774 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
1775 p
->real_parent
= current
->real_parent
;
1776 p
->parent_exec_id
= current
->parent_exec_id
;
1778 p
->real_parent
= current
;
1779 p
->parent_exec_id
= current
->self_exec_id
;
1782 spin_lock(¤t
->sighand
->siglock
);
1785 * Copy seccomp details explicitly here, in case they were changed
1786 * before holding sighand lock.
1791 * Process group and session signals need to be delivered to just the
1792 * parent before the fork or both the parent and the child after the
1793 * fork. Restart if a signal comes in before we add the new process to
1794 * it's process group.
1795 * A fatal signal pending means that current will exit, so the new
1796 * thread can't slip out of an OOM kill (or normal SIGKILL).
1798 recalc_sigpending();
1799 if (signal_pending(current
)) {
1800 spin_unlock(¤t
->sighand
->siglock
);
1801 write_unlock_irq(&tasklist_lock
);
1802 retval
= -ERESTARTNOINTR
;
1803 goto bad_fork_cancel_cgroup
;
1806 if (likely(p
->pid
)) {
1807 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
1809 init_task_pid(p
, PIDTYPE_PID
, pid
);
1810 if (thread_group_leader(p
)) {
1811 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
1812 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
1814 if (is_child_reaper(pid
)) {
1815 ns_of_pid(pid
)->child_reaper
= p
;
1816 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
1819 p
->signal
->leader_pid
= pid
;
1820 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
1822 * Inherit has_child_subreaper flag under the same
1823 * tasklist_lock with adding child to the process tree
1824 * for propagate_has_child_subreaper optimization.
1826 p
->signal
->has_child_subreaper
= p
->real_parent
->signal
->has_child_subreaper
||
1827 p
->real_parent
->signal
->is_child_subreaper
;
1828 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
1829 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
1830 attach_pid(p
, PIDTYPE_PGID
);
1831 attach_pid(p
, PIDTYPE_SID
);
1832 __this_cpu_inc(process_counts
);
1834 current
->signal
->nr_threads
++;
1835 atomic_inc(¤t
->signal
->live
);
1836 atomic_inc(¤t
->signal
->sigcnt
);
1837 list_add_tail_rcu(&p
->thread_group
,
1838 &p
->group_leader
->thread_group
);
1839 list_add_tail_rcu(&p
->thread_node
,
1840 &p
->signal
->thread_head
);
1842 attach_pid(p
, PIDTYPE_PID
);
1847 spin_unlock(¤t
->sighand
->siglock
);
1848 syscall_tracepoint_update(p
);
1849 write_unlock_irq(&tasklist_lock
);
1851 proc_fork_connector(p
);
1852 cgroup_post_fork(p
);
1853 cgroup_threadgroup_change_end(current
);
1856 trace_task_newtask(p
, clone_flags
);
1857 uprobe_copy_process(p
, clone_flags
);
1861 bad_fork_cancel_cgroup
:
1862 cgroup_cancel_fork(p
);
1864 cgroup_threadgroup_change_end(current
);
1865 if (pid
!= &init_struct_pid
)
1867 bad_fork_cleanup_thread
:
1869 bad_fork_cleanup_io
:
1872 bad_fork_cleanup_namespaces
:
1873 exit_task_namespaces(p
);
1874 bad_fork_cleanup_mm
:
1877 bad_fork_cleanup_signal
:
1878 if (!(clone_flags
& CLONE_THREAD
))
1879 free_signal_struct(p
->signal
);
1880 bad_fork_cleanup_sighand
:
1881 __cleanup_sighand(p
->sighand
);
1882 bad_fork_cleanup_fs
:
1883 exit_fs(p
); /* blocking */
1884 bad_fork_cleanup_files
:
1885 exit_files(p
); /* blocking */
1886 bad_fork_cleanup_semundo
:
1888 bad_fork_cleanup_audit
:
1890 bad_fork_cleanup_perf
:
1891 perf_event_free_task(p
);
1892 bad_fork_cleanup_policy
:
1894 mpol_put(p
->mempolicy
);
1895 bad_fork_cleanup_threadgroup_lock
:
1897 delayacct_tsk_free(p
);
1898 bad_fork_cleanup_count
:
1899 atomic_dec(&p
->cred
->user
->processes
);
1902 p
->state
= TASK_DEAD
;
1906 return ERR_PTR(retval
);
1909 static inline void init_idle_pids(struct pid_link
*links
)
1913 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1914 INIT_HLIST_NODE(&links
[type
].node
); /* not really needed */
1915 links
[type
].pid
= &init_struct_pid
;
1919 struct task_struct
*fork_idle(int cpu
)
1921 struct task_struct
*task
;
1922 task
= copy_process(CLONE_VM
, 0, 0, NULL
, &init_struct_pid
, 0, 0,
1924 if (!IS_ERR(task
)) {
1925 init_idle_pids(task
->pids
);
1926 init_idle(task
, cpu
);
1933 * Ok, this is the main fork-routine.
1935 * It copies the process, and if successful kick-starts
1936 * it and waits for it to finish using the VM if required.
1938 long _do_fork(unsigned long clone_flags
,
1939 unsigned long stack_start
,
1940 unsigned long stack_size
,
1941 int __user
*parent_tidptr
,
1942 int __user
*child_tidptr
,
1945 struct task_struct
*p
;
1950 * Determine whether and which event to report to ptracer. When
1951 * called from kernel_thread or CLONE_UNTRACED is explicitly
1952 * requested, no event is reported; otherwise, report if the event
1953 * for the type of forking is enabled.
1955 if (!(clone_flags
& CLONE_UNTRACED
)) {
1956 if (clone_flags
& CLONE_VFORK
)
1957 trace
= PTRACE_EVENT_VFORK
;
1958 else if ((clone_flags
& CSIGNAL
) != SIGCHLD
)
1959 trace
= PTRACE_EVENT_CLONE
;
1961 trace
= PTRACE_EVENT_FORK
;
1963 if (likely(!ptrace_event_enabled(current
, trace
)))
1967 p
= copy_process(clone_flags
, stack_start
, stack_size
,
1968 child_tidptr
, NULL
, trace
, tls
, NUMA_NO_NODE
);
1969 add_latent_entropy();
1971 * Do this prior waking up the new thread - the thread pointer
1972 * might get invalid after that point, if the thread exits quickly.
1975 struct completion vfork
;
1978 trace_sched_process_fork(current
, p
);
1980 pid
= get_task_pid(p
, PIDTYPE_PID
);
1983 if (clone_flags
& CLONE_PARENT_SETTID
)
1984 put_user(nr
, parent_tidptr
);
1986 if (clone_flags
& CLONE_VFORK
) {
1987 p
->vfork_done
= &vfork
;
1988 init_completion(&vfork
);
1992 wake_up_new_task(p
);
1994 /* forking complete and child started to run, tell ptracer */
1995 if (unlikely(trace
))
1996 ptrace_event_pid(trace
, pid
);
1998 if (clone_flags
& CLONE_VFORK
) {
1999 if (!wait_for_vfork_done(p
, &vfork
))
2000 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
2010 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2011 /* For compatibility with architectures that call do_fork directly rather than
2012 * using the syscall entry points below. */
2013 long do_fork(unsigned long clone_flags
,
2014 unsigned long stack_start
,
2015 unsigned long stack_size
,
2016 int __user
*parent_tidptr
,
2017 int __user
*child_tidptr
)
2019 return _do_fork(clone_flags
, stack_start
, stack_size
,
2020 parent_tidptr
, child_tidptr
, 0);
2025 * Create a kernel thread.
2027 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2029 return _do_fork(flags
|CLONE_VM
|CLONE_UNTRACED
, (unsigned long)fn
,
2030 (unsigned long)arg
, NULL
, NULL
, 0);
2033 #ifdef __ARCH_WANT_SYS_FORK
2034 SYSCALL_DEFINE0(fork
)
2037 return _do_fork(SIGCHLD
, 0, 0, NULL
, NULL
, 0);
2039 /* can not support in nommu mode */
2045 #ifdef __ARCH_WANT_SYS_VFORK
2046 SYSCALL_DEFINE0(vfork
)
2048 return _do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, 0,
2053 #ifdef __ARCH_WANT_SYS_CLONE
2054 #ifdef CONFIG_CLONE_BACKWARDS
2055 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2056 int __user
*, parent_tidptr
,
2058 int __user
*, child_tidptr
)
2059 #elif defined(CONFIG_CLONE_BACKWARDS2)
2060 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2061 int __user
*, parent_tidptr
,
2062 int __user
*, child_tidptr
,
2064 #elif defined(CONFIG_CLONE_BACKWARDS3)
2065 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2067 int __user
*, parent_tidptr
,
2068 int __user
*, child_tidptr
,
2071 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2072 int __user
*, parent_tidptr
,
2073 int __user
*, child_tidptr
,
2077 return _do_fork(clone_flags
, newsp
, 0, parent_tidptr
, child_tidptr
, tls
);
2081 void walk_process_tree(struct task_struct
*top
, proc_visitor visitor
, void *data
)
2083 struct task_struct
*leader
, *parent
, *child
;
2086 read_lock(&tasklist_lock
);
2087 leader
= top
= top
->group_leader
;
2089 for_each_thread(leader
, parent
) {
2090 list_for_each_entry(child
, &parent
->children
, sibling
) {
2091 res
= visitor(child
, data
);
2103 if (leader
!= top
) {
2105 parent
= child
->real_parent
;
2106 leader
= parent
->group_leader
;
2110 read_unlock(&tasklist_lock
);
2113 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2114 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2117 static void sighand_ctor(void *data
)
2119 struct sighand_struct
*sighand
= data
;
2121 spin_lock_init(&sighand
->siglock
);
2122 init_waitqueue_head(&sighand
->signalfd_wqh
);
2125 void __init
proc_caches_init(void)
2127 sighand_cachep
= kmem_cache_create("sighand_cache",
2128 sizeof(struct sighand_struct
), 0,
2129 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_DESTROY_BY_RCU
|
2130 SLAB_NOTRACK
|SLAB_ACCOUNT
, sighand_ctor
);
2131 signal_cachep
= kmem_cache_create("signal_cache",
2132 sizeof(struct signal_struct
), 0,
2133 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2135 files_cachep
= kmem_cache_create("files_cache",
2136 sizeof(struct files_struct
), 0,
2137 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2139 fs_cachep
= kmem_cache_create("fs_cache",
2140 sizeof(struct fs_struct
), 0,
2141 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2144 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2145 * whole struct cpumask for the OFFSTACK case. We could change
2146 * this to *only* allocate as much of it as required by the
2147 * maximum number of CPU's we can ever have. The cpumask_allocation
2148 * is at the end of the structure, exactly for that reason.
2150 mm_cachep
= kmem_cache_create("mm_struct",
2151 sizeof(struct mm_struct
), ARCH_MIN_MMSTRUCT_ALIGN
,
2152 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2154 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2156 nsproxy_cache_init();
2160 * Check constraints on flags passed to the unshare system call.
2162 static int check_unshare_flags(unsigned long unshare_flags
)
2164 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2165 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2166 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2167 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2170 * Not implemented, but pretend it works if there is nothing
2171 * to unshare. Note that unsharing the address space or the
2172 * signal handlers also need to unshare the signal queues (aka
2175 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2176 if (!thread_group_empty(current
))
2179 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2180 if (atomic_read(¤t
->sighand
->count
) > 1)
2183 if (unshare_flags
& CLONE_VM
) {
2184 if (!current_is_single_threaded())
2192 * Unshare the filesystem structure if it is being shared
2194 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2196 struct fs_struct
*fs
= current
->fs
;
2198 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2201 /* don't need lock here; in the worst case we'll do useless copy */
2205 *new_fsp
= copy_fs_struct(fs
);
2213 * Unshare file descriptor table if it is being shared
2215 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2217 struct files_struct
*fd
= current
->files
;
2220 if ((unshare_flags
& CLONE_FILES
) &&
2221 (fd
&& atomic_read(&fd
->count
) > 1)) {
2222 *new_fdp
= dup_fd(fd
, &error
);
2231 * unshare allows a process to 'unshare' part of the process
2232 * context which was originally shared using clone. copy_*
2233 * functions used by do_fork() cannot be used here directly
2234 * because they modify an inactive task_struct that is being
2235 * constructed. Here we are modifying the current, active,
2238 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2240 struct fs_struct
*fs
, *new_fs
= NULL
;
2241 struct files_struct
*fd
, *new_fd
= NULL
;
2242 struct cred
*new_cred
= NULL
;
2243 struct nsproxy
*new_nsproxy
= NULL
;
2248 * If unsharing a user namespace must also unshare the thread group
2249 * and unshare the filesystem root and working directories.
2251 if (unshare_flags
& CLONE_NEWUSER
)
2252 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2254 * If unsharing vm, must also unshare signal handlers.
2256 if (unshare_flags
& CLONE_VM
)
2257 unshare_flags
|= CLONE_SIGHAND
;
2259 * If unsharing a signal handlers, must also unshare the signal queues.
2261 if (unshare_flags
& CLONE_SIGHAND
)
2262 unshare_flags
|= CLONE_THREAD
;
2264 * If unsharing namespace, must also unshare filesystem information.
2266 if (unshare_flags
& CLONE_NEWNS
)
2267 unshare_flags
|= CLONE_FS
;
2269 err
= check_unshare_flags(unshare_flags
);
2271 goto bad_unshare_out
;
2273 * CLONE_NEWIPC must also detach from the undolist: after switching
2274 * to a new ipc namespace, the semaphore arrays from the old
2275 * namespace are unreachable.
2277 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2279 err
= unshare_fs(unshare_flags
, &new_fs
);
2281 goto bad_unshare_out
;
2282 err
= unshare_fd(unshare_flags
, &new_fd
);
2284 goto bad_unshare_cleanup_fs
;
2285 err
= unshare_userns(unshare_flags
, &new_cred
);
2287 goto bad_unshare_cleanup_fd
;
2288 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2291 goto bad_unshare_cleanup_cred
;
2293 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2296 * CLONE_SYSVSEM is equivalent to sys_exit().
2300 if (unshare_flags
& CLONE_NEWIPC
) {
2301 /* Orphan segments in old ns (see sem above). */
2303 shm_init_task(current
);
2307 switch_task_namespaces(current
, new_nsproxy
);
2313 spin_lock(&fs
->lock
);
2314 current
->fs
= new_fs
;
2319 spin_unlock(&fs
->lock
);
2323 fd
= current
->files
;
2324 current
->files
= new_fd
;
2328 task_unlock(current
);
2331 /* Install the new user namespace */
2332 commit_creds(new_cred
);
2337 bad_unshare_cleanup_cred
:
2340 bad_unshare_cleanup_fd
:
2342 put_files_struct(new_fd
);
2344 bad_unshare_cleanup_fs
:
2346 free_fs_struct(new_fs
);
2353 * Helper to unshare the files of the current task.
2354 * We don't want to expose copy_files internals to
2355 * the exec layer of the kernel.
2358 int unshare_files(struct files_struct
**displaced
)
2360 struct task_struct
*task
= current
;
2361 struct files_struct
*copy
= NULL
;
2364 error
= unshare_fd(CLONE_FILES
, ©
);
2365 if (error
|| !copy
) {
2369 *displaced
= task
->files
;
2376 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2377 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2381 int threads
= max_threads
;
2382 int min
= MIN_THREADS
;
2383 int max
= MAX_THREADS
;
2390 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2394 set_max_threads(threads
);