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/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/delayacct.h>
62 #include <linux/taskstats_kern.h>
63 #include <linux/random.h>
64 #include <linux/tty.h>
65 #include <linux/blkdev.h>
66 #include <linux/fs_struct.h>
67 #include <linux/magic.h>
68 #include <linux/perf_event.h>
69 #include <linux/posix-timers.h>
70 #include <linux/user-return-notifier.h>
71 #include <linux/oom.h>
72 #include <linux/khugepaged.h>
73 #include <linux/signalfd.h>
74 #include <linux/uprobes.h>
75 #include <linux/aio.h>
76 #include <linux/compiler.h>
77 #include <linux/sysctl.h>
78 #include <linux/kcov.h>
80 #include <asm/pgtable.h>
81 #include <asm/pgalloc.h>
82 #include <asm/uaccess.h>
83 #include <asm/mmu_context.h>
84 #include <asm/cacheflush.h>
85 #include <asm/tlbflush.h>
87 #include <trace/events/sched.h>
89 #define CREATE_TRACE_POINTS
90 #include <trace/events/task.h>
93 * Minimum number of threads to boot the kernel
95 #define MIN_THREADS 20
98 * Maximum number of threads
100 #define MAX_THREADS FUTEX_TID_MASK
103 * Protected counters by write_lock_irq(&tasklist_lock)
105 unsigned long total_forks
; /* Handle normal Linux uptimes. */
106 int nr_threads
; /* The idle threads do not count.. */
108 int max_threads
; /* tunable limit on nr_threads */
110 DEFINE_PER_CPU(unsigned long, process_counts
) = 0;
112 __cacheline_aligned
DEFINE_RWLOCK(tasklist_lock
); /* outer */
114 #ifdef CONFIG_PROVE_RCU
115 int lockdep_tasklist_lock_is_held(void)
117 return lockdep_is_held(&tasklist_lock
);
119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held
);
120 #endif /* #ifdef CONFIG_PROVE_RCU */
122 int nr_processes(void)
127 for_each_possible_cpu(cpu
)
128 total
+= per_cpu(process_counts
, cpu
);
133 void __weak
arch_release_task_struct(struct task_struct
*tsk
)
137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
138 static struct kmem_cache
*task_struct_cachep
;
140 static inline struct task_struct
*alloc_task_struct_node(int node
)
142 return kmem_cache_alloc_node(task_struct_cachep
, GFP_KERNEL
, node
);
145 static inline void free_task_struct(struct task_struct
*tsk
)
147 kmem_cache_free(task_struct_cachep
, tsk
);
151 void __weak
arch_release_thread_stack(unsigned long *stack
)
155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
158 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159 * kmemcache based allocator.
161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
162 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
164 #ifdef CONFIG_VMAP_STACK
165 void *stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_SIZE
,
166 VMALLOC_START
, VMALLOC_END
,
167 THREADINFO_GFP
| __GFP_HIGHMEM
,
170 __builtin_return_address(0));
173 * We can't call find_vm_area() in interrupt context, and
174 * free_thread_stack() can be called in interrupt context,
175 * so cache the vm_struct.
178 tsk
->stack_vm_area
= find_vm_area(stack
);
181 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
184 return page
? page_address(page
) : NULL
;
188 static inline void free_thread_stack(struct task_struct
*tsk
)
190 if (task_stack_vm_area(tsk
))
193 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
196 static struct kmem_cache
*thread_stack_cache
;
198 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
201 return kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
204 static void free_thread_stack(struct task_struct
*tsk
)
206 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
209 void thread_stack_cache_init(void)
211 thread_stack_cache
= kmem_cache_create("thread_stack", THREAD_SIZE
,
212 THREAD_SIZE
, 0, NULL
);
213 BUG_ON(thread_stack_cache
== NULL
);
218 /* SLAB cache for signal_struct structures (tsk->signal) */
219 static struct kmem_cache
*signal_cachep
;
221 /* SLAB cache for sighand_struct structures (tsk->sighand) */
222 struct kmem_cache
*sighand_cachep
;
224 /* SLAB cache for files_struct structures (tsk->files) */
225 struct kmem_cache
*files_cachep
;
227 /* SLAB cache for fs_struct structures (tsk->fs) */
228 struct kmem_cache
*fs_cachep
;
230 /* SLAB cache for vm_area_struct structures */
231 struct kmem_cache
*vm_area_cachep
;
233 /* SLAB cache for mm_struct structures (tsk->mm) */
234 static struct kmem_cache
*mm_cachep
;
236 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
238 void *stack
= task_stack_page(tsk
);
239 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
241 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
246 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
248 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
249 mod_zone_page_state(page_zone(vm
->pages
[i
]),
251 PAGE_SIZE
/ 1024 * account
);
254 /* All stack pages belong to the same memcg. */
255 memcg_kmem_update_page_stat(vm
->pages
[0], MEMCG_KERNEL_STACK_KB
,
256 account
* (THREAD_SIZE
/ 1024));
259 * All stack pages are in the same zone and belong to the
262 struct page
*first_page
= virt_to_page(stack
);
264 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
265 THREAD_SIZE
/ 1024 * account
);
267 memcg_kmem_update_page_stat(first_page
, MEMCG_KERNEL_STACK_KB
,
268 account
* (THREAD_SIZE
/ 1024));
272 void free_task(struct task_struct
*tsk
)
274 account_kernel_stack(tsk
, -1);
275 arch_release_thread_stack(tsk
->stack
);
276 free_thread_stack(tsk
);
277 rt_mutex_debug_task_free(tsk
);
278 ftrace_graph_exit_task(tsk
);
279 put_seccomp_filter(tsk
);
280 arch_release_task_struct(tsk
);
281 free_task_struct(tsk
);
283 EXPORT_SYMBOL(free_task
);
285 static inline void free_signal_struct(struct signal_struct
*sig
)
287 taskstats_tgid_free(sig
);
288 sched_autogroup_exit(sig
);
290 * __mmdrop is not safe to call from softirq context on x86 due to
291 * pgd_dtor so postpone it to the async context
294 mmdrop_async(sig
->oom_mm
);
295 kmem_cache_free(signal_cachep
, sig
);
298 static inline void put_signal_struct(struct signal_struct
*sig
)
300 if (atomic_dec_and_test(&sig
->sigcnt
))
301 free_signal_struct(sig
);
304 void __put_task_struct(struct task_struct
*tsk
)
306 WARN_ON(!tsk
->exit_state
);
307 WARN_ON(atomic_read(&tsk
->usage
));
308 WARN_ON(tsk
== current
);
312 security_task_free(tsk
);
314 delayacct_tsk_free(tsk
);
315 put_signal_struct(tsk
->signal
);
317 if (!profile_handoff_task(tsk
))
320 EXPORT_SYMBOL_GPL(__put_task_struct
);
322 void __init __weak
arch_task_cache_init(void) { }
327 static void set_max_threads(unsigned int max_threads_suggested
)
332 * The number of threads shall be limited such that the thread
333 * structures may only consume a small part of the available memory.
335 if (fls64(totalram_pages
) + fls64(PAGE_SIZE
) > 64)
336 threads
= MAX_THREADS
;
338 threads
= div64_u64((u64
) totalram_pages
* (u64
) PAGE_SIZE
,
339 (u64
) THREAD_SIZE
* 8UL);
341 if (threads
> max_threads_suggested
)
342 threads
= max_threads_suggested
;
344 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
347 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
348 /* Initialized by the architecture: */
349 int arch_task_struct_size __read_mostly
;
352 void __init
fork_init(void)
354 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
355 #ifndef ARCH_MIN_TASKALIGN
356 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
358 /* create a slab on which task_structs can be allocated */
359 task_struct_cachep
= kmem_cache_create("task_struct",
360 arch_task_struct_size
, ARCH_MIN_TASKALIGN
,
361 SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
, NULL
);
364 /* do the arch specific task caches init */
365 arch_task_cache_init();
367 set_max_threads(MAX_THREADS
);
369 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
370 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
371 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
372 init_task
.signal
->rlim
[RLIMIT_NPROC
];
375 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
376 struct task_struct
*src
)
382 void set_task_stack_end_magic(struct task_struct
*tsk
)
384 unsigned long *stackend
;
386 stackend
= end_of_stack(tsk
);
387 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
390 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
392 struct task_struct
*tsk
;
393 unsigned long *stack
;
394 struct vm_struct
*stack_vm_area
;
397 if (node
== NUMA_NO_NODE
)
398 node
= tsk_fork_get_node(orig
);
399 tsk
= alloc_task_struct_node(node
);
403 stack
= alloc_thread_stack_node(tsk
, node
);
407 stack_vm_area
= task_stack_vm_area(tsk
);
409 err
= arch_dup_task_struct(tsk
, orig
);
412 * arch_dup_task_struct() clobbers the stack-related fields. Make
413 * sure they're properly initialized before using any stack-related
417 #ifdef CONFIG_VMAP_STACK
418 tsk
->stack_vm_area
= stack_vm_area
;
424 #ifdef CONFIG_SECCOMP
426 * We must handle setting up seccomp filters once we're under
427 * the sighand lock in case orig has changed between now and
428 * then. Until then, filter must be NULL to avoid messing up
429 * the usage counts on the error path calling free_task.
431 tsk
->seccomp
.filter
= NULL
;
434 setup_thread_stack(tsk
, orig
);
435 clear_user_return_notifier(tsk
);
436 clear_tsk_need_resched(tsk
);
437 set_task_stack_end_magic(tsk
);
439 #ifdef CONFIG_CC_STACKPROTECTOR
440 tsk
->stack_canary
= get_random_int();
444 * One for us, one for whoever does the "release_task()" (usually
447 atomic_set(&tsk
->usage
, 2);
448 #ifdef CONFIG_BLK_DEV_IO_TRACE
451 tsk
->splice_pipe
= NULL
;
452 tsk
->task_frag
.page
= NULL
;
453 tsk
->wake_q
.next
= NULL
;
455 account_kernel_stack(tsk
, 1);
462 free_thread_stack(tsk
);
464 free_task_struct(tsk
);
469 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
470 struct mm_struct
*oldmm
)
472 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
473 struct rb_node
**rb_link
, *rb_parent
;
475 unsigned long charge
;
477 uprobe_start_dup_mmap();
478 if (down_write_killable(&oldmm
->mmap_sem
)) {
480 goto fail_uprobe_end
;
482 flush_cache_dup_mm(oldmm
);
483 uprobe_dup_mmap(oldmm
, mm
);
485 * Not linked in yet - no deadlock potential:
487 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
489 /* No ordering required: file already has been exposed. */
490 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
492 mm
->total_vm
= oldmm
->total_vm
;
493 mm
->data_vm
= oldmm
->data_vm
;
494 mm
->exec_vm
= oldmm
->exec_vm
;
495 mm
->stack_vm
= oldmm
->stack_vm
;
497 rb_link
= &mm
->mm_rb
.rb_node
;
500 retval
= ksm_fork(mm
, oldmm
);
503 retval
= khugepaged_fork(mm
, oldmm
);
508 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
511 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
512 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
516 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
517 unsigned long len
= vma_pages(mpnt
);
519 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
523 tmp
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
527 INIT_LIST_HEAD(&tmp
->anon_vma_chain
);
528 retval
= vma_dup_policy(mpnt
, tmp
);
530 goto fail_nomem_policy
;
532 if (anon_vma_fork(tmp
, mpnt
))
533 goto fail_nomem_anon_vma_fork
;
535 ~(VM_LOCKED
|VM_LOCKONFAULT
|VM_UFFD_MISSING
|VM_UFFD_WP
);
536 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
537 tmp
->vm_userfaultfd_ctx
= NULL_VM_UFFD_CTX
;
540 struct inode
*inode
= file_inode(file
);
541 struct address_space
*mapping
= file
->f_mapping
;
544 if (tmp
->vm_flags
& VM_DENYWRITE
)
545 atomic_dec(&inode
->i_writecount
);
546 i_mmap_lock_write(mapping
);
547 if (tmp
->vm_flags
& VM_SHARED
)
548 atomic_inc(&mapping
->i_mmap_writable
);
549 flush_dcache_mmap_lock(mapping
);
550 /* insert tmp into the share list, just after mpnt */
551 vma_interval_tree_insert_after(tmp
, mpnt
,
553 flush_dcache_mmap_unlock(mapping
);
554 i_mmap_unlock_write(mapping
);
558 * Clear hugetlb-related page reserves for children. This only
559 * affects MAP_PRIVATE mappings. Faults generated by the child
560 * are not guaranteed to succeed, even if read-only
562 if (is_vm_hugetlb_page(tmp
))
563 reset_vma_resv_huge_pages(tmp
);
566 * Link in the new vma and copy the page table entries.
569 pprev
= &tmp
->vm_next
;
573 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
574 rb_link
= &tmp
->vm_rb
.rb_right
;
575 rb_parent
= &tmp
->vm_rb
;
578 retval
= copy_page_range(mm
, oldmm
, mpnt
);
580 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
581 tmp
->vm_ops
->open(tmp
);
586 /* a new mm has just been created */
587 arch_dup_mmap(oldmm
, mm
);
590 up_write(&mm
->mmap_sem
);
592 up_write(&oldmm
->mmap_sem
);
594 uprobe_end_dup_mmap();
596 fail_nomem_anon_vma_fork
:
597 mpol_put(vma_policy(tmp
));
599 kmem_cache_free(vm_area_cachep
, tmp
);
602 vm_unacct_memory(charge
);
606 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
608 mm
->pgd
= pgd_alloc(mm
);
609 if (unlikely(!mm
->pgd
))
614 static inline void mm_free_pgd(struct mm_struct
*mm
)
616 pgd_free(mm
, mm
->pgd
);
619 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
621 down_write(&oldmm
->mmap_sem
);
622 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
623 up_write(&oldmm
->mmap_sem
);
626 #define mm_alloc_pgd(mm) (0)
627 #define mm_free_pgd(mm)
628 #endif /* CONFIG_MMU */
630 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
632 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
633 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
635 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
637 static int __init
coredump_filter_setup(char *s
)
639 default_dump_filter
=
640 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
641 MMF_DUMP_FILTER_MASK
;
645 __setup("coredump_filter=", coredump_filter_setup
);
647 #include <linux/init_task.h>
649 static void mm_init_aio(struct mm_struct
*mm
)
652 spin_lock_init(&mm
->ioctx_lock
);
653 mm
->ioctx_table
= NULL
;
657 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
664 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
)
668 mm
->vmacache_seqnum
= 0;
669 atomic_set(&mm
->mm_users
, 1);
670 atomic_set(&mm
->mm_count
, 1);
671 init_rwsem(&mm
->mmap_sem
);
672 INIT_LIST_HEAD(&mm
->mmlist
);
673 mm
->core_state
= NULL
;
674 atomic_long_set(&mm
->nr_ptes
, 0);
679 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
680 spin_lock_init(&mm
->page_table_lock
);
683 mm_init_owner(mm
, p
);
684 mmu_notifier_mm_init(mm
);
685 clear_tlb_flush_pending(mm
);
686 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
687 mm
->pmd_huge_pte
= NULL
;
691 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
692 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
694 mm
->flags
= default_dump_filter
;
698 if (mm_alloc_pgd(mm
))
701 if (init_new_context(p
, mm
))
713 static void check_mm(struct mm_struct
*mm
)
717 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
718 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
721 printk(KERN_ALERT
"BUG: Bad rss-counter state "
722 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
725 if (atomic_long_read(&mm
->nr_ptes
))
726 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
727 atomic_long_read(&mm
->nr_ptes
));
729 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
732 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
733 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
738 * Allocate and initialize an mm_struct.
740 struct mm_struct
*mm_alloc(void)
742 struct mm_struct
*mm
;
748 memset(mm
, 0, sizeof(*mm
));
749 return mm_init(mm
, current
);
753 * Called when the last reference to the mm
754 * is dropped: either by a lazy thread or by
755 * mmput. Free the page directory and the mm.
757 void __mmdrop(struct mm_struct
*mm
)
759 BUG_ON(mm
== &init_mm
);
762 mmu_notifier_mm_destroy(mm
);
766 EXPORT_SYMBOL_GPL(__mmdrop
);
768 static inline void __mmput(struct mm_struct
*mm
)
770 VM_BUG_ON(atomic_read(&mm
->mm_users
));
772 uprobe_clear_state(mm
);
775 khugepaged_exit(mm
); /* must run before exit_mmap */
777 mm_put_huge_zero_page(mm
);
778 set_mm_exe_file(mm
, NULL
);
779 if (!list_empty(&mm
->mmlist
)) {
780 spin_lock(&mmlist_lock
);
781 list_del(&mm
->mmlist
);
782 spin_unlock(&mmlist_lock
);
785 module_put(mm
->binfmt
->module
);
786 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
791 * Decrement the use count and release all resources for an mm.
793 void mmput(struct mm_struct
*mm
)
797 if (atomic_dec_and_test(&mm
->mm_users
))
800 EXPORT_SYMBOL_GPL(mmput
);
803 static void mmput_async_fn(struct work_struct
*work
)
805 struct mm_struct
*mm
= container_of(work
, struct mm_struct
, async_put_work
);
809 void mmput_async(struct mm_struct
*mm
)
811 if (atomic_dec_and_test(&mm
->mm_users
)) {
812 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
813 schedule_work(&mm
->async_put_work
);
819 * set_mm_exe_file - change a reference to the mm's executable file
821 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
823 * Main users are mmput() and sys_execve(). Callers prevent concurrent
824 * invocations: in mmput() nobody alive left, in execve task is single
825 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
826 * mm->exe_file, but does so without using set_mm_exe_file() in order
827 * to do avoid the need for any locks.
829 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
831 struct file
*old_exe_file
;
834 * It is safe to dereference the exe_file without RCU as
835 * this function is only called if nobody else can access
836 * this mm -- see comment above for justification.
838 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
841 get_file(new_exe_file
);
842 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
848 * get_mm_exe_file - acquire a reference to the mm's executable file
850 * Returns %NULL if mm has no associated executable file.
851 * User must release file via fput().
853 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
855 struct file
*exe_file
;
858 exe_file
= rcu_dereference(mm
->exe_file
);
859 if (exe_file
&& !get_file_rcu(exe_file
))
864 EXPORT_SYMBOL(get_mm_exe_file
);
867 * get_task_exe_file - acquire a reference to the task's executable file
869 * Returns %NULL if task's mm (if any) has no associated executable file or
870 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
871 * User must release file via fput().
873 struct file
*get_task_exe_file(struct task_struct
*task
)
875 struct file
*exe_file
= NULL
;
876 struct mm_struct
*mm
;
881 if (!(task
->flags
& PF_KTHREAD
))
882 exe_file
= get_mm_exe_file(mm
);
887 EXPORT_SYMBOL(get_task_exe_file
);
890 * get_task_mm - acquire a reference to the task's mm
892 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
893 * this kernel workthread has transiently adopted a user mm with use_mm,
894 * to do its AIO) is not set and if so returns a reference to it, after
895 * bumping up the use count. User must release the mm via mmput()
896 * after use. Typically used by /proc and ptrace.
898 struct mm_struct
*get_task_mm(struct task_struct
*task
)
900 struct mm_struct
*mm
;
905 if (task
->flags
& PF_KTHREAD
)
908 atomic_inc(&mm
->mm_users
);
913 EXPORT_SYMBOL_GPL(get_task_mm
);
915 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
917 struct mm_struct
*mm
;
920 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
924 mm
= get_task_mm(task
);
925 if (mm
&& mm
!= current
->mm
&&
926 !ptrace_may_access(task
, mode
)) {
928 mm
= ERR_PTR(-EACCES
);
930 mutex_unlock(&task
->signal
->cred_guard_mutex
);
935 static void complete_vfork_done(struct task_struct
*tsk
)
937 struct completion
*vfork
;
940 vfork
= tsk
->vfork_done
;
942 tsk
->vfork_done
= NULL
;
948 static int wait_for_vfork_done(struct task_struct
*child
,
949 struct completion
*vfork
)
953 freezer_do_not_count();
954 killed
= wait_for_completion_killable(vfork
);
959 child
->vfork_done
= NULL
;
963 put_task_struct(child
);
967 /* Please note the differences between mmput and mm_release.
968 * mmput is called whenever we stop holding onto a mm_struct,
969 * error success whatever.
971 * mm_release is called after a mm_struct has been removed
972 * from the current process.
974 * This difference is important for error handling, when we
975 * only half set up a mm_struct for a new process and need to restore
976 * the old one. Because we mmput the new mm_struct before
977 * restoring the old one. . .
978 * Eric Biederman 10 January 1998
980 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
982 /* Get rid of any futexes when releasing the mm */
984 if (unlikely(tsk
->robust_list
)) {
985 exit_robust_list(tsk
);
986 tsk
->robust_list
= NULL
;
989 if (unlikely(tsk
->compat_robust_list
)) {
990 compat_exit_robust_list(tsk
);
991 tsk
->compat_robust_list
= NULL
;
994 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
995 exit_pi_state_list(tsk
);
998 uprobe_free_utask(tsk
);
1000 /* Get rid of any cached register state */
1001 deactivate_mm(tsk
, mm
);
1004 * Signal userspace if we're not exiting with a core dump
1005 * because we want to leave the value intact for debugging
1008 if (tsk
->clear_child_tid
) {
1009 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1010 atomic_read(&mm
->mm_users
) > 1) {
1012 * We don't check the error code - if userspace has
1013 * not set up a proper pointer then tough luck.
1015 put_user(0, tsk
->clear_child_tid
);
1016 sys_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1019 tsk
->clear_child_tid
= NULL
;
1023 * All done, finally we can wake up parent and return this mm to him.
1024 * Also kthread_stop() uses this completion for synchronization.
1026 if (tsk
->vfork_done
)
1027 complete_vfork_done(tsk
);
1031 * Allocate a new mm structure and copy contents from the
1032 * mm structure of the passed in task structure.
1034 static struct mm_struct
*dup_mm(struct task_struct
*tsk
)
1036 struct mm_struct
*mm
, *oldmm
= current
->mm
;
1043 memcpy(mm
, oldmm
, sizeof(*mm
));
1045 if (!mm_init(mm
, tsk
))
1048 err
= dup_mmap(mm
, oldmm
);
1052 mm
->hiwater_rss
= get_mm_rss(mm
);
1053 mm
->hiwater_vm
= mm
->total_vm
;
1055 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1061 /* don't put binfmt in mmput, we haven't got module yet */
1069 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1071 struct mm_struct
*mm
, *oldmm
;
1074 tsk
->min_flt
= tsk
->maj_flt
= 0;
1075 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1076 #ifdef CONFIG_DETECT_HUNG_TASK
1077 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1081 tsk
->active_mm
= NULL
;
1084 * Are we cloning a kernel thread?
1086 * We need to steal a active VM for that..
1088 oldmm
= current
->mm
;
1092 /* initialize the new vmacache entries */
1093 vmacache_flush(tsk
);
1095 if (clone_flags
& CLONE_VM
) {
1096 atomic_inc(&oldmm
->mm_users
);
1108 tsk
->active_mm
= mm
;
1115 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1117 struct fs_struct
*fs
= current
->fs
;
1118 if (clone_flags
& CLONE_FS
) {
1119 /* tsk->fs is already what we want */
1120 spin_lock(&fs
->lock
);
1122 spin_unlock(&fs
->lock
);
1126 spin_unlock(&fs
->lock
);
1129 tsk
->fs
= copy_fs_struct(fs
);
1135 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1137 struct files_struct
*oldf
, *newf
;
1141 * A background process may not have any files ...
1143 oldf
= current
->files
;
1147 if (clone_flags
& CLONE_FILES
) {
1148 atomic_inc(&oldf
->count
);
1152 newf
= dup_fd(oldf
, &error
);
1162 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1165 struct io_context
*ioc
= current
->io_context
;
1166 struct io_context
*new_ioc
;
1171 * Share io context with parent, if CLONE_IO is set
1173 if (clone_flags
& CLONE_IO
) {
1175 tsk
->io_context
= ioc
;
1176 } else if (ioprio_valid(ioc
->ioprio
)) {
1177 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1178 if (unlikely(!new_ioc
))
1181 new_ioc
->ioprio
= ioc
->ioprio
;
1182 put_io_context(new_ioc
);
1188 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1190 struct sighand_struct
*sig
;
1192 if (clone_flags
& CLONE_SIGHAND
) {
1193 atomic_inc(¤t
->sighand
->count
);
1196 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1197 rcu_assign_pointer(tsk
->sighand
, sig
);
1201 atomic_set(&sig
->count
, 1);
1202 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1206 void __cleanup_sighand(struct sighand_struct
*sighand
)
1208 if (atomic_dec_and_test(&sighand
->count
)) {
1209 signalfd_cleanup(sighand
);
1211 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1212 * without an RCU grace period, see __lock_task_sighand().
1214 kmem_cache_free(sighand_cachep
, sighand
);
1219 * Initialize POSIX timer handling for a thread group.
1221 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1223 unsigned long cpu_limit
;
1225 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1226 if (cpu_limit
!= RLIM_INFINITY
) {
1227 sig
->cputime_expires
.prof_exp
= secs_to_cputime(cpu_limit
);
1228 sig
->cputimer
.running
= true;
1231 /* The timer lists. */
1232 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1233 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1234 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1237 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1239 struct signal_struct
*sig
;
1241 if (clone_flags
& CLONE_THREAD
)
1244 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1249 sig
->nr_threads
= 1;
1250 atomic_set(&sig
->live
, 1);
1251 atomic_set(&sig
->sigcnt
, 1);
1253 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1254 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1255 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1257 init_waitqueue_head(&sig
->wait_chldexit
);
1258 sig
->curr_target
= tsk
;
1259 init_sigpending(&sig
->shared_pending
);
1260 INIT_LIST_HEAD(&sig
->posix_timers
);
1261 seqlock_init(&sig
->stats_lock
);
1262 prev_cputime_init(&sig
->prev_cputime
);
1264 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1265 sig
->real_timer
.function
= it_real_fn
;
1267 task_lock(current
->group_leader
);
1268 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1269 task_unlock(current
->group_leader
);
1271 posix_cpu_timers_init_group(sig
);
1273 tty_audit_fork(sig
);
1274 sched_autogroup_fork(sig
);
1276 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1277 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1279 sig
->has_child_subreaper
= current
->signal
->has_child_subreaper
||
1280 current
->signal
->is_child_subreaper
;
1282 mutex_init(&sig
->cred_guard_mutex
);
1287 static void copy_seccomp(struct task_struct
*p
)
1289 #ifdef CONFIG_SECCOMP
1291 * Must be called with sighand->lock held, which is common to
1292 * all threads in the group. Holding cred_guard_mutex is not
1293 * needed because this new task is not yet running and cannot
1296 assert_spin_locked(¤t
->sighand
->siglock
);
1298 /* Ref-count the new filter user, and assign it. */
1299 get_seccomp_filter(current
);
1300 p
->seccomp
= current
->seccomp
;
1303 * Explicitly enable no_new_privs here in case it got set
1304 * between the task_struct being duplicated and holding the
1305 * sighand lock. The seccomp state and nnp must be in sync.
1307 if (task_no_new_privs(current
))
1308 task_set_no_new_privs(p
);
1311 * If the parent gained a seccomp mode after copying thread
1312 * flags and between before we held the sighand lock, we have
1313 * to manually enable the seccomp thread flag here.
1315 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1316 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1320 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1322 current
->clear_child_tid
= tidptr
;
1324 return task_pid_vnr(current
);
1327 static void rt_mutex_init_task(struct task_struct
*p
)
1329 raw_spin_lock_init(&p
->pi_lock
);
1330 #ifdef CONFIG_RT_MUTEXES
1331 p
->pi_waiters
= RB_ROOT
;
1332 p
->pi_waiters_leftmost
= NULL
;
1333 p
->pi_blocked_on
= NULL
;
1338 * Initialize POSIX timer handling for a single task.
1340 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1342 tsk
->cputime_expires
.prof_exp
= 0;
1343 tsk
->cputime_expires
.virt_exp
= 0;
1344 tsk
->cputime_expires
.sched_exp
= 0;
1345 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1346 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1347 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1351 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1353 task
->pids
[type
].pid
= pid
;
1357 * This creates a new process as a copy of the old one,
1358 * but does not actually start it yet.
1360 * It copies the registers, and all the appropriate
1361 * parts of the process environment (as per the clone
1362 * flags). The actual kick-off is left to the caller.
1364 static __latent_entropy
struct task_struct
*copy_process(
1365 unsigned long clone_flags
,
1366 unsigned long stack_start
,
1367 unsigned long stack_size
,
1368 int __user
*child_tidptr
,
1375 struct task_struct
*p
;
1377 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1378 return ERR_PTR(-EINVAL
);
1380 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1381 return ERR_PTR(-EINVAL
);
1384 * Thread groups must share signals as well, and detached threads
1385 * can only be started up within the thread group.
1387 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1388 return ERR_PTR(-EINVAL
);
1391 * Shared signal handlers imply shared VM. By way of the above,
1392 * thread groups also imply shared VM. Blocking this case allows
1393 * for various simplifications in other code.
1395 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1396 return ERR_PTR(-EINVAL
);
1399 * Siblings of global init remain as zombies on exit since they are
1400 * not reaped by their parent (swapper). To solve this and to avoid
1401 * multi-rooted process trees, prevent global and container-inits
1402 * from creating siblings.
1404 if ((clone_flags
& CLONE_PARENT
) &&
1405 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1406 return ERR_PTR(-EINVAL
);
1409 * If the new process will be in a different pid or user namespace
1410 * do not allow it to share a thread group with the forking task.
1412 if (clone_flags
& CLONE_THREAD
) {
1413 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1414 (task_active_pid_ns(current
) !=
1415 current
->nsproxy
->pid_ns_for_children
))
1416 return ERR_PTR(-EINVAL
);
1419 retval
= security_task_create(clone_flags
);
1424 p
= dup_task_struct(current
, node
);
1428 ftrace_graph_init_task(p
);
1430 rt_mutex_init_task(p
);
1432 #ifdef CONFIG_PROVE_LOCKING
1433 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1434 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1437 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1438 task_rlimit(p
, RLIMIT_NPROC
)) {
1439 if (p
->real_cred
->user
!= INIT_USER
&&
1440 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1443 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1445 retval
= copy_creds(p
, clone_flags
);
1450 * If multiple threads are within copy_process(), then this check
1451 * triggers too late. This doesn't hurt, the check is only there
1452 * to stop root fork bombs.
1455 if (nr_threads
>= max_threads
)
1456 goto bad_fork_cleanup_count
;
1458 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1459 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
);
1460 p
->flags
|= PF_FORKNOEXEC
;
1461 INIT_LIST_HEAD(&p
->children
);
1462 INIT_LIST_HEAD(&p
->sibling
);
1463 rcu_copy_process(p
);
1464 p
->vfork_done
= NULL
;
1465 spin_lock_init(&p
->alloc_lock
);
1467 init_sigpending(&p
->pending
);
1469 p
->utime
= p
->stime
= p
->gtime
= 0;
1470 p
->utimescaled
= p
->stimescaled
= 0;
1471 prev_cputime_init(&p
->prev_cputime
);
1473 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1474 seqcount_init(&p
->vtime_seqcount
);
1476 p
->vtime_snap_whence
= VTIME_INACTIVE
;
1479 #if defined(SPLIT_RSS_COUNTING)
1480 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1483 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1485 task_io_accounting_init(&p
->ioac
);
1486 acct_clear_integrals(p
);
1488 posix_cpu_timers_init(p
);
1490 p
->start_time
= ktime_get_ns();
1491 p
->real_start_time
= ktime_get_boot_ns();
1492 p
->io_context
= NULL
;
1493 p
->audit_context
= NULL
;
1496 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1497 if (IS_ERR(p
->mempolicy
)) {
1498 retval
= PTR_ERR(p
->mempolicy
);
1499 p
->mempolicy
= NULL
;
1500 goto bad_fork_cleanup_threadgroup_lock
;
1503 #ifdef CONFIG_CPUSETS
1504 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1505 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1506 seqcount_init(&p
->mems_allowed_seq
);
1508 #ifdef CONFIG_TRACE_IRQFLAGS
1510 p
->hardirqs_enabled
= 0;
1511 p
->hardirq_enable_ip
= 0;
1512 p
->hardirq_enable_event
= 0;
1513 p
->hardirq_disable_ip
= _THIS_IP_
;
1514 p
->hardirq_disable_event
= 0;
1515 p
->softirqs_enabled
= 1;
1516 p
->softirq_enable_ip
= _THIS_IP_
;
1517 p
->softirq_enable_event
= 0;
1518 p
->softirq_disable_ip
= 0;
1519 p
->softirq_disable_event
= 0;
1520 p
->hardirq_context
= 0;
1521 p
->softirq_context
= 0;
1524 p
->pagefault_disabled
= 0;
1526 #ifdef CONFIG_LOCKDEP
1527 p
->lockdep_depth
= 0; /* no locks held yet */
1528 p
->curr_chain_key
= 0;
1529 p
->lockdep_recursion
= 0;
1532 #ifdef CONFIG_DEBUG_MUTEXES
1533 p
->blocked_on
= NULL
; /* not blocked yet */
1535 #ifdef CONFIG_BCACHE
1536 p
->sequential_io
= 0;
1537 p
->sequential_io_avg
= 0;
1540 /* Perform scheduler related setup. Assign this task to a CPU. */
1541 retval
= sched_fork(clone_flags
, p
);
1543 goto bad_fork_cleanup_policy
;
1545 retval
= perf_event_init_task(p
);
1547 goto bad_fork_cleanup_policy
;
1548 retval
= audit_alloc(p
);
1550 goto bad_fork_cleanup_perf
;
1551 /* copy all the process information */
1553 retval
= copy_semundo(clone_flags
, p
);
1555 goto bad_fork_cleanup_audit
;
1556 retval
= copy_files(clone_flags
, p
);
1558 goto bad_fork_cleanup_semundo
;
1559 retval
= copy_fs(clone_flags
, p
);
1561 goto bad_fork_cleanup_files
;
1562 retval
= copy_sighand(clone_flags
, p
);
1564 goto bad_fork_cleanup_fs
;
1565 retval
= copy_signal(clone_flags
, p
);
1567 goto bad_fork_cleanup_sighand
;
1568 retval
= copy_mm(clone_flags
, p
);
1570 goto bad_fork_cleanup_signal
;
1571 retval
= copy_namespaces(clone_flags
, p
);
1573 goto bad_fork_cleanup_mm
;
1574 retval
= copy_io(clone_flags
, p
);
1576 goto bad_fork_cleanup_namespaces
;
1577 retval
= copy_thread_tls(clone_flags
, stack_start
, stack_size
, p
, tls
);
1579 goto bad_fork_cleanup_io
;
1581 if (pid
!= &init_struct_pid
) {
1582 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
1584 retval
= PTR_ERR(pid
);
1585 goto bad_fork_cleanup_thread
;
1589 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
1591 * Clear TID on mm_release()?
1593 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
1598 p
->robust_list
= NULL
;
1599 #ifdef CONFIG_COMPAT
1600 p
->compat_robust_list
= NULL
;
1602 INIT_LIST_HEAD(&p
->pi_state_list
);
1603 p
->pi_state_cache
= NULL
;
1606 * sigaltstack should be cleared when sharing the same VM
1608 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
1612 * Syscall tracing and stepping should be turned off in the
1613 * child regardless of CLONE_PTRACE.
1615 user_disable_single_step(p
);
1616 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
1617 #ifdef TIF_SYSCALL_EMU
1618 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1620 clear_all_latency_tracing(p
);
1622 /* ok, now we should be set up.. */
1623 p
->pid
= pid_nr(pid
);
1624 if (clone_flags
& CLONE_THREAD
) {
1625 p
->exit_signal
= -1;
1626 p
->group_leader
= current
->group_leader
;
1627 p
->tgid
= current
->tgid
;
1629 if (clone_flags
& CLONE_PARENT
)
1630 p
->exit_signal
= current
->group_leader
->exit_signal
;
1632 p
->exit_signal
= (clone_flags
& CSIGNAL
);
1633 p
->group_leader
= p
;
1638 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
1639 p
->dirty_paused_when
= 0;
1641 p
->pdeath_signal
= 0;
1642 INIT_LIST_HEAD(&p
->thread_group
);
1643 p
->task_works
= NULL
;
1645 threadgroup_change_begin(current
);
1647 * Ensure that the cgroup subsystem policies allow the new process to be
1648 * forked. It should be noted the the new process's css_set can be changed
1649 * between here and cgroup_post_fork() if an organisation operation is in
1652 retval
= cgroup_can_fork(p
);
1654 goto bad_fork_free_pid
;
1657 * Make it visible to the rest of the system, but dont wake it up yet.
1658 * Need tasklist lock for parent etc handling!
1660 write_lock_irq(&tasklist_lock
);
1662 /* CLONE_PARENT re-uses the old parent */
1663 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
1664 p
->real_parent
= current
->real_parent
;
1665 p
->parent_exec_id
= current
->parent_exec_id
;
1667 p
->real_parent
= current
;
1668 p
->parent_exec_id
= current
->self_exec_id
;
1671 spin_lock(¤t
->sighand
->siglock
);
1674 * Copy seccomp details explicitly here, in case they were changed
1675 * before holding sighand lock.
1680 * Process group and session signals need to be delivered to just the
1681 * parent before the fork or both the parent and the child after the
1682 * fork. Restart if a signal comes in before we add the new process to
1683 * it's process group.
1684 * A fatal signal pending means that current will exit, so the new
1685 * thread can't slip out of an OOM kill (or normal SIGKILL).
1687 recalc_sigpending();
1688 if (signal_pending(current
)) {
1689 spin_unlock(¤t
->sighand
->siglock
);
1690 write_unlock_irq(&tasklist_lock
);
1691 retval
= -ERESTARTNOINTR
;
1692 goto bad_fork_cancel_cgroup
;
1695 if (likely(p
->pid
)) {
1696 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
1698 init_task_pid(p
, PIDTYPE_PID
, pid
);
1699 if (thread_group_leader(p
)) {
1700 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
1701 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
1703 if (is_child_reaper(pid
)) {
1704 ns_of_pid(pid
)->child_reaper
= p
;
1705 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
1708 p
->signal
->leader_pid
= pid
;
1709 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
1710 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
1711 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
1712 attach_pid(p
, PIDTYPE_PGID
);
1713 attach_pid(p
, PIDTYPE_SID
);
1714 __this_cpu_inc(process_counts
);
1716 current
->signal
->nr_threads
++;
1717 atomic_inc(¤t
->signal
->live
);
1718 atomic_inc(¤t
->signal
->sigcnt
);
1719 list_add_tail_rcu(&p
->thread_group
,
1720 &p
->group_leader
->thread_group
);
1721 list_add_tail_rcu(&p
->thread_node
,
1722 &p
->signal
->thread_head
);
1724 attach_pid(p
, PIDTYPE_PID
);
1729 spin_unlock(¤t
->sighand
->siglock
);
1730 syscall_tracepoint_update(p
);
1731 write_unlock_irq(&tasklist_lock
);
1733 proc_fork_connector(p
);
1734 cgroup_post_fork(p
);
1735 threadgroup_change_end(current
);
1738 trace_task_newtask(p
, clone_flags
);
1739 uprobe_copy_process(p
, clone_flags
);
1743 bad_fork_cancel_cgroup
:
1744 cgroup_cancel_fork(p
);
1746 threadgroup_change_end(current
);
1747 if (pid
!= &init_struct_pid
)
1749 bad_fork_cleanup_thread
:
1751 bad_fork_cleanup_io
:
1754 bad_fork_cleanup_namespaces
:
1755 exit_task_namespaces(p
);
1756 bad_fork_cleanup_mm
:
1759 bad_fork_cleanup_signal
:
1760 if (!(clone_flags
& CLONE_THREAD
))
1761 free_signal_struct(p
->signal
);
1762 bad_fork_cleanup_sighand
:
1763 __cleanup_sighand(p
->sighand
);
1764 bad_fork_cleanup_fs
:
1765 exit_fs(p
); /* blocking */
1766 bad_fork_cleanup_files
:
1767 exit_files(p
); /* blocking */
1768 bad_fork_cleanup_semundo
:
1770 bad_fork_cleanup_audit
:
1772 bad_fork_cleanup_perf
:
1773 perf_event_free_task(p
);
1774 bad_fork_cleanup_policy
:
1776 mpol_put(p
->mempolicy
);
1777 bad_fork_cleanup_threadgroup_lock
:
1779 delayacct_tsk_free(p
);
1780 bad_fork_cleanup_count
:
1781 atomic_dec(&p
->cred
->user
->processes
);
1786 return ERR_PTR(retval
);
1789 static inline void init_idle_pids(struct pid_link
*links
)
1793 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1794 INIT_HLIST_NODE(&links
[type
].node
); /* not really needed */
1795 links
[type
].pid
= &init_struct_pid
;
1799 struct task_struct
*fork_idle(int cpu
)
1801 struct task_struct
*task
;
1802 task
= copy_process(CLONE_VM
, 0, 0, NULL
, &init_struct_pid
, 0, 0,
1804 if (!IS_ERR(task
)) {
1805 init_idle_pids(task
->pids
);
1806 init_idle(task
, cpu
);
1813 * Ok, this is the main fork-routine.
1815 * It copies the process, and if successful kick-starts
1816 * it and waits for it to finish using the VM if required.
1818 long _do_fork(unsigned long clone_flags
,
1819 unsigned long stack_start
,
1820 unsigned long stack_size
,
1821 int __user
*parent_tidptr
,
1822 int __user
*child_tidptr
,
1825 struct task_struct
*p
;
1830 * Determine whether and which event to report to ptracer. When
1831 * called from kernel_thread or CLONE_UNTRACED is explicitly
1832 * requested, no event is reported; otherwise, report if the event
1833 * for the type of forking is enabled.
1835 if (!(clone_flags
& CLONE_UNTRACED
)) {
1836 if (clone_flags
& CLONE_VFORK
)
1837 trace
= PTRACE_EVENT_VFORK
;
1838 else if ((clone_flags
& CSIGNAL
) != SIGCHLD
)
1839 trace
= PTRACE_EVENT_CLONE
;
1841 trace
= PTRACE_EVENT_FORK
;
1843 if (likely(!ptrace_event_enabled(current
, trace
)))
1847 p
= copy_process(clone_flags
, stack_start
, stack_size
,
1848 child_tidptr
, NULL
, trace
, tls
, NUMA_NO_NODE
);
1849 add_latent_entropy();
1851 * Do this prior waking up the new thread - the thread pointer
1852 * might get invalid after that point, if the thread exits quickly.
1855 struct completion vfork
;
1858 trace_sched_process_fork(current
, p
);
1860 pid
= get_task_pid(p
, PIDTYPE_PID
);
1863 if (clone_flags
& CLONE_PARENT_SETTID
)
1864 put_user(nr
, parent_tidptr
);
1866 if (clone_flags
& CLONE_VFORK
) {
1867 p
->vfork_done
= &vfork
;
1868 init_completion(&vfork
);
1872 wake_up_new_task(p
);
1874 /* forking complete and child started to run, tell ptracer */
1875 if (unlikely(trace
))
1876 ptrace_event_pid(trace
, pid
);
1878 if (clone_flags
& CLONE_VFORK
) {
1879 if (!wait_for_vfork_done(p
, &vfork
))
1880 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
1890 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1891 /* For compatibility with architectures that call do_fork directly rather than
1892 * using the syscall entry points below. */
1893 long do_fork(unsigned long clone_flags
,
1894 unsigned long stack_start
,
1895 unsigned long stack_size
,
1896 int __user
*parent_tidptr
,
1897 int __user
*child_tidptr
)
1899 return _do_fork(clone_flags
, stack_start
, stack_size
,
1900 parent_tidptr
, child_tidptr
, 0);
1905 * Create a kernel thread.
1907 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
1909 return _do_fork(flags
|CLONE_VM
|CLONE_UNTRACED
, (unsigned long)fn
,
1910 (unsigned long)arg
, NULL
, NULL
, 0);
1913 #ifdef __ARCH_WANT_SYS_FORK
1914 SYSCALL_DEFINE0(fork
)
1917 return _do_fork(SIGCHLD
, 0, 0, NULL
, NULL
, 0);
1919 /* can not support in nommu mode */
1925 #ifdef __ARCH_WANT_SYS_VFORK
1926 SYSCALL_DEFINE0(vfork
)
1928 return _do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, 0,
1933 #ifdef __ARCH_WANT_SYS_CLONE
1934 #ifdef CONFIG_CLONE_BACKWARDS
1935 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1936 int __user
*, parent_tidptr
,
1938 int __user
*, child_tidptr
)
1939 #elif defined(CONFIG_CLONE_BACKWARDS2)
1940 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
1941 int __user
*, parent_tidptr
,
1942 int __user
*, child_tidptr
,
1944 #elif defined(CONFIG_CLONE_BACKWARDS3)
1945 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1947 int __user
*, parent_tidptr
,
1948 int __user
*, child_tidptr
,
1951 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1952 int __user
*, parent_tidptr
,
1953 int __user
*, child_tidptr
,
1957 return _do_fork(clone_flags
, newsp
, 0, parent_tidptr
, child_tidptr
, tls
);
1961 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1962 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1965 static void sighand_ctor(void *data
)
1967 struct sighand_struct
*sighand
= data
;
1969 spin_lock_init(&sighand
->siglock
);
1970 init_waitqueue_head(&sighand
->signalfd_wqh
);
1973 void __init
proc_caches_init(void)
1975 sighand_cachep
= kmem_cache_create("sighand_cache",
1976 sizeof(struct sighand_struct
), 0,
1977 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_DESTROY_BY_RCU
|
1978 SLAB_NOTRACK
|SLAB_ACCOUNT
, sighand_ctor
);
1979 signal_cachep
= kmem_cache_create("signal_cache",
1980 sizeof(struct signal_struct
), 0,
1981 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1983 files_cachep
= kmem_cache_create("files_cache",
1984 sizeof(struct files_struct
), 0,
1985 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1987 fs_cachep
= kmem_cache_create("fs_cache",
1988 sizeof(struct fs_struct
), 0,
1989 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1992 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1993 * whole struct cpumask for the OFFSTACK case. We could change
1994 * this to *only* allocate as much of it as required by the
1995 * maximum number of CPU's we can ever have. The cpumask_allocation
1996 * is at the end of the structure, exactly for that reason.
1998 mm_cachep
= kmem_cache_create("mm_struct",
1999 sizeof(struct mm_struct
), ARCH_MIN_MMSTRUCT_ALIGN
,
2000 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2002 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2004 nsproxy_cache_init();
2008 * Check constraints on flags passed to the unshare system call.
2010 static int check_unshare_flags(unsigned long unshare_flags
)
2012 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2013 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2014 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2015 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2018 * Not implemented, but pretend it works if there is nothing
2019 * to unshare. Note that unsharing the address space or the
2020 * signal handlers also need to unshare the signal queues (aka
2023 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2024 if (!thread_group_empty(current
))
2027 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2028 if (atomic_read(¤t
->sighand
->count
) > 1)
2031 if (unshare_flags
& CLONE_VM
) {
2032 if (!current_is_single_threaded())
2040 * Unshare the filesystem structure if it is being shared
2042 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2044 struct fs_struct
*fs
= current
->fs
;
2046 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2049 /* don't need lock here; in the worst case we'll do useless copy */
2053 *new_fsp
= copy_fs_struct(fs
);
2061 * Unshare file descriptor table if it is being shared
2063 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2065 struct files_struct
*fd
= current
->files
;
2068 if ((unshare_flags
& CLONE_FILES
) &&
2069 (fd
&& atomic_read(&fd
->count
) > 1)) {
2070 *new_fdp
= dup_fd(fd
, &error
);
2079 * unshare allows a process to 'unshare' part of the process
2080 * context which was originally shared using clone. copy_*
2081 * functions used by do_fork() cannot be used here directly
2082 * because they modify an inactive task_struct that is being
2083 * constructed. Here we are modifying the current, active,
2086 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2088 struct fs_struct
*fs
, *new_fs
= NULL
;
2089 struct files_struct
*fd
, *new_fd
= NULL
;
2090 struct cred
*new_cred
= NULL
;
2091 struct nsproxy
*new_nsproxy
= NULL
;
2096 * If unsharing a user namespace must also unshare the thread group
2097 * and unshare the filesystem root and working directories.
2099 if (unshare_flags
& CLONE_NEWUSER
)
2100 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2102 * If unsharing vm, must also unshare signal handlers.
2104 if (unshare_flags
& CLONE_VM
)
2105 unshare_flags
|= CLONE_SIGHAND
;
2107 * If unsharing a signal handlers, must also unshare the signal queues.
2109 if (unshare_flags
& CLONE_SIGHAND
)
2110 unshare_flags
|= CLONE_THREAD
;
2112 * If unsharing namespace, must also unshare filesystem information.
2114 if (unshare_flags
& CLONE_NEWNS
)
2115 unshare_flags
|= CLONE_FS
;
2117 err
= check_unshare_flags(unshare_flags
);
2119 goto bad_unshare_out
;
2121 * CLONE_NEWIPC must also detach from the undolist: after switching
2122 * to a new ipc namespace, the semaphore arrays from the old
2123 * namespace are unreachable.
2125 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2127 err
= unshare_fs(unshare_flags
, &new_fs
);
2129 goto bad_unshare_out
;
2130 err
= unshare_fd(unshare_flags
, &new_fd
);
2132 goto bad_unshare_cleanup_fs
;
2133 err
= unshare_userns(unshare_flags
, &new_cred
);
2135 goto bad_unshare_cleanup_fd
;
2136 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2139 goto bad_unshare_cleanup_cred
;
2141 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2144 * CLONE_SYSVSEM is equivalent to sys_exit().
2148 if (unshare_flags
& CLONE_NEWIPC
) {
2149 /* Orphan segments in old ns (see sem above). */
2151 shm_init_task(current
);
2155 switch_task_namespaces(current
, new_nsproxy
);
2161 spin_lock(&fs
->lock
);
2162 current
->fs
= new_fs
;
2167 spin_unlock(&fs
->lock
);
2171 fd
= current
->files
;
2172 current
->files
= new_fd
;
2176 task_unlock(current
);
2179 /* Install the new user namespace */
2180 commit_creds(new_cred
);
2185 bad_unshare_cleanup_cred
:
2188 bad_unshare_cleanup_fd
:
2190 put_files_struct(new_fd
);
2192 bad_unshare_cleanup_fs
:
2194 free_fs_struct(new_fs
);
2201 * Helper to unshare the files of the current task.
2202 * We don't want to expose copy_files internals to
2203 * the exec layer of the kernel.
2206 int unshare_files(struct files_struct
**displaced
)
2208 struct task_struct
*task
= current
;
2209 struct files_struct
*copy
= NULL
;
2212 error
= unshare_fd(CLONE_FILES
, ©
);
2213 if (error
|| !copy
) {
2217 *displaced
= task
->files
;
2224 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2225 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2229 int threads
= max_threads
;
2230 int min
= MIN_THREADS
;
2231 int max
= MAX_THREADS
;
2238 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2242 set_max_threads(threads
);