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_info(struct thread_info
*ti
)
155 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
158 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159 * kmemcache based allocator.
161 # if THREAD_SIZE >= PAGE_SIZE
162 static struct thread_info
*alloc_thread_info_node(struct task_struct
*tsk
,
165 struct page
*page
= alloc_kmem_pages_node(node
, THREADINFO_GFP
,
169 memcg_kmem_update_page_stat(page
, MEMCG_KERNEL_STACK
,
170 1 << THREAD_SIZE_ORDER
);
172 return page
? page_address(page
) : NULL
;
175 static inline void free_thread_info(struct thread_info
*ti
)
177 struct page
*page
= virt_to_page(ti
);
179 memcg_kmem_update_page_stat(page
, MEMCG_KERNEL_STACK
,
180 -(1 << THREAD_SIZE_ORDER
));
181 __free_kmem_pages(page
, THREAD_SIZE_ORDER
);
184 static struct kmem_cache
*thread_info_cache
;
186 static struct thread_info
*alloc_thread_info_node(struct task_struct
*tsk
,
189 return kmem_cache_alloc_node(thread_info_cache
, THREADINFO_GFP
, node
);
192 static void free_thread_info(struct thread_info
*ti
)
194 kmem_cache_free(thread_info_cache
, ti
);
197 void thread_info_cache_init(void)
199 thread_info_cache
= kmem_cache_create("thread_info", THREAD_SIZE
,
200 THREAD_SIZE
, 0, NULL
);
201 BUG_ON(thread_info_cache
== NULL
);
206 /* SLAB cache for signal_struct structures (tsk->signal) */
207 static struct kmem_cache
*signal_cachep
;
209 /* SLAB cache for sighand_struct structures (tsk->sighand) */
210 struct kmem_cache
*sighand_cachep
;
212 /* SLAB cache for files_struct structures (tsk->files) */
213 struct kmem_cache
*files_cachep
;
215 /* SLAB cache for fs_struct structures (tsk->fs) */
216 struct kmem_cache
*fs_cachep
;
218 /* SLAB cache for vm_area_struct structures */
219 struct kmem_cache
*vm_area_cachep
;
221 /* SLAB cache for mm_struct structures (tsk->mm) */
222 static struct kmem_cache
*mm_cachep
;
224 static void account_kernel_stack(struct thread_info
*ti
, int account
)
226 struct zone
*zone
= page_zone(virt_to_page(ti
));
228 mod_zone_page_state(zone
, NR_KERNEL_STACK
, account
);
231 void free_task(struct task_struct
*tsk
)
233 account_kernel_stack(tsk
->stack
, -1);
234 arch_release_thread_info(tsk
->stack
);
235 free_thread_info(tsk
->stack
);
236 rt_mutex_debug_task_free(tsk
);
237 ftrace_graph_exit_task(tsk
);
238 put_seccomp_filter(tsk
);
239 arch_release_task_struct(tsk
);
240 free_task_struct(tsk
);
242 EXPORT_SYMBOL(free_task
);
244 static inline void free_signal_struct(struct signal_struct
*sig
)
246 taskstats_tgid_free(sig
);
247 sched_autogroup_exit(sig
);
248 kmem_cache_free(signal_cachep
, sig
);
251 static inline void put_signal_struct(struct signal_struct
*sig
)
253 if (atomic_dec_and_test(&sig
->sigcnt
))
254 free_signal_struct(sig
);
257 void __put_task_struct(struct task_struct
*tsk
)
259 WARN_ON(!tsk
->exit_state
);
260 WARN_ON(atomic_read(&tsk
->usage
));
261 WARN_ON(tsk
== current
);
265 security_task_free(tsk
);
267 delayacct_tsk_free(tsk
);
268 put_signal_struct(tsk
->signal
);
270 if (!profile_handoff_task(tsk
))
273 EXPORT_SYMBOL_GPL(__put_task_struct
);
275 void __init __weak
arch_task_cache_init(void) { }
280 static void set_max_threads(unsigned int max_threads_suggested
)
285 * The number of threads shall be limited such that the thread
286 * structures may only consume a small part of the available memory.
288 if (fls64(totalram_pages
) + fls64(PAGE_SIZE
) > 64)
289 threads
= MAX_THREADS
;
291 threads
= div64_u64((u64
) totalram_pages
* (u64
) PAGE_SIZE
,
292 (u64
) THREAD_SIZE
* 8UL);
294 if (threads
> max_threads_suggested
)
295 threads
= max_threads_suggested
;
297 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
300 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
301 /* Initialized by the architecture: */
302 int arch_task_struct_size __read_mostly
;
305 void __init
fork_init(void)
307 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
308 #ifndef ARCH_MIN_TASKALIGN
309 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
311 /* create a slab on which task_structs can be allocated */
312 task_struct_cachep
= kmem_cache_create("task_struct",
313 arch_task_struct_size
, ARCH_MIN_TASKALIGN
,
314 SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
, NULL
);
317 /* do the arch specific task caches init */
318 arch_task_cache_init();
320 set_max_threads(MAX_THREADS
);
322 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
323 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
324 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
325 init_task
.signal
->rlim
[RLIMIT_NPROC
];
328 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
329 struct task_struct
*src
)
335 void set_task_stack_end_magic(struct task_struct
*tsk
)
337 unsigned long *stackend
;
339 stackend
= end_of_stack(tsk
);
340 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
343 static struct task_struct
*dup_task_struct(struct task_struct
*orig
)
345 struct task_struct
*tsk
;
346 struct thread_info
*ti
;
347 int node
= tsk_fork_get_node(orig
);
350 tsk
= alloc_task_struct_node(node
);
354 ti
= alloc_thread_info_node(tsk
, node
);
358 err
= arch_dup_task_struct(tsk
, orig
);
363 #ifdef CONFIG_SECCOMP
365 * We must handle setting up seccomp filters once we're under
366 * the sighand lock in case orig has changed between now and
367 * then. Until then, filter must be NULL to avoid messing up
368 * the usage counts on the error path calling free_task.
370 tsk
->seccomp
.filter
= NULL
;
373 setup_thread_stack(tsk
, orig
);
374 clear_user_return_notifier(tsk
);
375 clear_tsk_need_resched(tsk
);
376 set_task_stack_end_magic(tsk
);
378 #ifdef CONFIG_CC_STACKPROTECTOR
379 tsk
->stack_canary
= get_random_int();
383 * One for us, one for whoever does the "release_task()" (usually
386 atomic_set(&tsk
->usage
, 2);
387 #ifdef CONFIG_BLK_DEV_IO_TRACE
390 tsk
->splice_pipe
= NULL
;
391 tsk
->task_frag
.page
= NULL
;
392 tsk
->wake_q
.next
= NULL
;
394 account_kernel_stack(ti
, 1);
401 free_thread_info(ti
);
403 free_task_struct(tsk
);
408 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
410 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
411 struct rb_node
**rb_link
, *rb_parent
;
413 unsigned long charge
;
415 uprobe_start_dup_mmap();
416 down_write(&oldmm
->mmap_sem
);
417 flush_cache_dup_mm(oldmm
);
418 uprobe_dup_mmap(oldmm
, mm
);
420 * Not linked in yet - no deadlock potential:
422 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
424 /* No ordering required: file already has been exposed. */
425 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
427 mm
->total_vm
= oldmm
->total_vm
;
428 mm
->data_vm
= oldmm
->data_vm
;
429 mm
->exec_vm
= oldmm
->exec_vm
;
430 mm
->stack_vm
= oldmm
->stack_vm
;
432 rb_link
= &mm
->mm_rb
.rb_node
;
435 retval
= ksm_fork(mm
, oldmm
);
438 retval
= khugepaged_fork(mm
, oldmm
);
443 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
446 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
447 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
451 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
452 unsigned long len
= vma_pages(mpnt
);
454 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
458 tmp
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
462 INIT_LIST_HEAD(&tmp
->anon_vma_chain
);
463 retval
= vma_dup_policy(mpnt
, tmp
);
465 goto fail_nomem_policy
;
467 if (anon_vma_fork(tmp
, mpnt
))
468 goto fail_nomem_anon_vma_fork
;
470 ~(VM_LOCKED
|VM_LOCKONFAULT
|VM_UFFD_MISSING
|VM_UFFD_WP
);
471 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
472 tmp
->vm_userfaultfd_ctx
= NULL_VM_UFFD_CTX
;
475 struct inode
*inode
= file_inode(file
);
476 struct address_space
*mapping
= file
->f_mapping
;
479 if (tmp
->vm_flags
& VM_DENYWRITE
)
480 atomic_dec(&inode
->i_writecount
);
481 i_mmap_lock_write(mapping
);
482 if (tmp
->vm_flags
& VM_SHARED
)
483 atomic_inc(&mapping
->i_mmap_writable
);
484 flush_dcache_mmap_lock(mapping
);
485 /* insert tmp into the share list, just after mpnt */
486 vma_interval_tree_insert_after(tmp
, mpnt
,
488 flush_dcache_mmap_unlock(mapping
);
489 i_mmap_unlock_write(mapping
);
493 * Clear hugetlb-related page reserves for children. This only
494 * affects MAP_PRIVATE mappings. Faults generated by the child
495 * are not guaranteed to succeed, even if read-only
497 if (is_vm_hugetlb_page(tmp
))
498 reset_vma_resv_huge_pages(tmp
);
501 * Link in the new vma and copy the page table entries.
504 pprev
= &tmp
->vm_next
;
508 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
509 rb_link
= &tmp
->vm_rb
.rb_right
;
510 rb_parent
= &tmp
->vm_rb
;
513 retval
= copy_page_range(mm
, oldmm
, mpnt
);
515 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
516 tmp
->vm_ops
->open(tmp
);
521 /* a new mm has just been created */
522 arch_dup_mmap(oldmm
, mm
);
525 up_write(&mm
->mmap_sem
);
527 up_write(&oldmm
->mmap_sem
);
528 uprobe_end_dup_mmap();
530 fail_nomem_anon_vma_fork
:
531 mpol_put(vma_policy(tmp
));
533 kmem_cache_free(vm_area_cachep
, tmp
);
536 vm_unacct_memory(charge
);
540 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
542 mm
->pgd
= pgd_alloc(mm
);
543 if (unlikely(!mm
->pgd
))
548 static inline void mm_free_pgd(struct mm_struct
*mm
)
550 pgd_free(mm
, mm
->pgd
);
553 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
555 down_write(&oldmm
->mmap_sem
);
556 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
557 up_write(&oldmm
->mmap_sem
);
560 #define mm_alloc_pgd(mm) (0)
561 #define mm_free_pgd(mm)
562 #endif /* CONFIG_MMU */
564 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
566 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
567 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
569 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
571 static int __init
coredump_filter_setup(char *s
)
573 default_dump_filter
=
574 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
575 MMF_DUMP_FILTER_MASK
;
579 __setup("coredump_filter=", coredump_filter_setup
);
581 #include <linux/init_task.h>
583 static void mm_init_aio(struct mm_struct
*mm
)
586 spin_lock_init(&mm
->ioctx_lock
);
587 mm
->ioctx_table
= NULL
;
591 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
598 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
)
602 mm
->vmacache_seqnum
= 0;
603 atomic_set(&mm
->mm_users
, 1);
604 atomic_set(&mm
->mm_count
, 1);
605 init_rwsem(&mm
->mmap_sem
);
606 INIT_LIST_HEAD(&mm
->mmlist
);
607 mm
->core_state
= NULL
;
608 atomic_long_set(&mm
->nr_ptes
, 0);
613 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
614 spin_lock_init(&mm
->page_table_lock
);
617 mm_init_owner(mm
, p
);
618 mmu_notifier_mm_init(mm
);
619 clear_tlb_flush_pending(mm
);
620 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
621 mm
->pmd_huge_pte
= NULL
;
625 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
626 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
628 mm
->flags
= default_dump_filter
;
632 if (mm_alloc_pgd(mm
))
635 if (init_new_context(p
, mm
))
647 static void check_mm(struct mm_struct
*mm
)
651 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
652 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
655 printk(KERN_ALERT
"BUG: Bad rss-counter state "
656 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
659 if (atomic_long_read(&mm
->nr_ptes
))
660 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
661 atomic_long_read(&mm
->nr_ptes
));
663 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
666 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
667 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
672 * Allocate and initialize an mm_struct.
674 struct mm_struct
*mm_alloc(void)
676 struct mm_struct
*mm
;
682 memset(mm
, 0, sizeof(*mm
));
683 return mm_init(mm
, current
);
687 * Called when the last reference to the mm
688 * is dropped: either by a lazy thread or by
689 * mmput. Free the page directory and the mm.
691 void __mmdrop(struct mm_struct
*mm
)
693 BUG_ON(mm
== &init_mm
);
696 mmu_notifier_mm_destroy(mm
);
700 EXPORT_SYMBOL_GPL(__mmdrop
);
702 static inline void __mmput(struct mm_struct
*mm
)
704 VM_BUG_ON(atomic_read(&mm
->mm_users
));
706 uprobe_clear_state(mm
);
709 khugepaged_exit(mm
); /* must run before exit_mmap */
711 set_mm_exe_file(mm
, NULL
);
712 if (!list_empty(&mm
->mmlist
)) {
713 spin_lock(&mmlist_lock
);
714 list_del(&mm
->mmlist
);
715 spin_unlock(&mmlist_lock
);
718 module_put(mm
->binfmt
->module
);
723 * Decrement the use count and release all resources for an mm.
725 void mmput(struct mm_struct
*mm
)
729 if (atomic_dec_and_test(&mm
->mm_users
))
732 EXPORT_SYMBOL_GPL(mmput
);
734 static void mmput_async_fn(struct work_struct
*work
)
736 struct mm_struct
*mm
= container_of(work
, struct mm_struct
, async_put_work
);
740 void mmput_async(struct mm_struct
*mm
)
742 if (atomic_dec_and_test(&mm
->mm_users
)) {
743 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
744 schedule_work(&mm
->async_put_work
);
749 * set_mm_exe_file - change a reference to the mm's executable file
751 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
753 * Main users are mmput() and sys_execve(). Callers prevent concurrent
754 * invocations: in mmput() nobody alive left, in execve task is single
755 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
756 * mm->exe_file, but does so without using set_mm_exe_file() in order
757 * to do avoid the need for any locks.
759 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
761 struct file
*old_exe_file
;
764 * It is safe to dereference the exe_file without RCU as
765 * this function is only called if nobody else can access
766 * this mm -- see comment above for justification.
768 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
771 get_file(new_exe_file
);
772 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
778 * get_mm_exe_file - acquire a reference to the mm's executable file
780 * Returns %NULL if mm has no associated executable file.
781 * User must release file via fput().
783 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
785 struct file
*exe_file
;
788 exe_file
= rcu_dereference(mm
->exe_file
);
789 if (exe_file
&& !get_file_rcu(exe_file
))
794 EXPORT_SYMBOL(get_mm_exe_file
);
797 * get_task_mm - acquire a reference to the task's mm
799 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
800 * this kernel workthread has transiently adopted a user mm with use_mm,
801 * to do its AIO) is not set and if so returns a reference to it, after
802 * bumping up the use count. User must release the mm via mmput()
803 * after use. Typically used by /proc and ptrace.
805 struct mm_struct
*get_task_mm(struct task_struct
*task
)
807 struct mm_struct
*mm
;
812 if (task
->flags
& PF_KTHREAD
)
815 atomic_inc(&mm
->mm_users
);
820 EXPORT_SYMBOL_GPL(get_task_mm
);
822 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
824 struct mm_struct
*mm
;
827 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
831 mm
= get_task_mm(task
);
832 if (mm
&& mm
!= current
->mm
&&
833 !ptrace_may_access(task
, mode
)) {
835 mm
= ERR_PTR(-EACCES
);
837 mutex_unlock(&task
->signal
->cred_guard_mutex
);
842 static void complete_vfork_done(struct task_struct
*tsk
)
844 struct completion
*vfork
;
847 vfork
= tsk
->vfork_done
;
849 tsk
->vfork_done
= NULL
;
855 static int wait_for_vfork_done(struct task_struct
*child
,
856 struct completion
*vfork
)
860 freezer_do_not_count();
861 killed
= wait_for_completion_killable(vfork
);
866 child
->vfork_done
= NULL
;
870 put_task_struct(child
);
874 /* Please note the differences between mmput and mm_release.
875 * mmput is called whenever we stop holding onto a mm_struct,
876 * error success whatever.
878 * mm_release is called after a mm_struct has been removed
879 * from the current process.
881 * This difference is important for error handling, when we
882 * only half set up a mm_struct for a new process and need to restore
883 * the old one. Because we mmput the new mm_struct before
884 * restoring the old one. . .
885 * Eric Biederman 10 January 1998
887 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
889 /* Get rid of any futexes when releasing the mm */
891 if (unlikely(tsk
->robust_list
)) {
892 exit_robust_list(tsk
);
893 tsk
->robust_list
= NULL
;
896 if (unlikely(tsk
->compat_robust_list
)) {
897 compat_exit_robust_list(tsk
);
898 tsk
->compat_robust_list
= NULL
;
901 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
902 exit_pi_state_list(tsk
);
905 uprobe_free_utask(tsk
);
907 /* Get rid of any cached register state */
908 deactivate_mm(tsk
, mm
);
911 * If we're exiting normally, clear a user-space tid field if
912 * requested. We leave this alone when dying by signal, to leave
913 * the value intact in a core dump, and to save the unnecessary
914 * trouble, say, a killed vfork parent shouldn't touch this mm.
915 * Userland only wants this done for a sys_exit.
917 if (tsk
->clear_child_tid
) {
918 if (!(tsk
->flags
& PF_SIGNALED
) &&
919 atomic_read(&mm
->mm_users
) > 1) {
921 * We don't check the error code - if userspace has
922 * not set up a proper pointer then tough luck.
924 put_user(0, tsk
->clear_child_tid
);
925 sys_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
928 tsk
->clear_child_tid
= NULL
;
932 * All done, finally we can wake up parent and return this mm to him.
933 * Also kthread_stop() uses this completion for synchronization.
936 complete_vfork_done(tsk
);
940 * Allocate a new mm structure and copy contents from the
941 * mm structure of the passed in task structure.
943 static struct mm_struct
*dup_mm(struct task_struct
*tsk
)
945 struct mm_struct
*mm
, *oldmm
= current
->mm
;
952 memcpy(mm
, oldmm
, sizeof(*mm
));
954 if (!mm_init(mm
, tsk
))
957 err
= dup_mmap(mm
, oldmm
);
961 mm
->hiwater_rss
= get_mm_rss(mm
);
962 mm
->hiwater_vm
= mm
->total_vm
;
964 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
970 /* don't put binfmt in mmput, we haven't got module yet */
978 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
980 struct mm_struct
*mm
, *oldmm
;
983 tsk
->min_flt
= tsk
->maj_flt
= 0;
984 tsk
->nvcsw
= tsk
->nivcsw
= 0;
985 #ifdef CONFIG_DETECT_HUNG_TASK
986 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
990 tsk
->active_mm
= NULL
;
993 * Are we cloning a kernel thread?
995 * We need to steal a active VM for that..
1001 /* initialize the new vmacache entries */
1002 vmacache_flush(tsk
);
1004 if (clone_flags
& CLONE_VM
) {
1005 atomic_inc(&oldmm
->mm_users
);
1017 tsk
->active_mm
= mm
;
1024 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1026 struct fs_struct
*fs
= current
->fs
;
1027 if (clone_flags
& CLONE_FS
) {
1028 /* tsk->fs is already what we want */
1029 spin_lock(&fs
->lock
);
1031 spin_unlock(&fs
->lock
);
1035 spin_unlock(&fs
->lock
);
1038 tsk
->fs
= copy_fs_struct(fs
);
1044 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1046 struct files_struct
*oldf
, *newf
;
1050 * A background process may not have any files ...
1052 oldf
= current
->files
;
1056 if (clone_flags
& CLONE_FILES
) {
1057 atomic_inc(&oldf
->count
);
1061 newf
= dup_fd(oldf
, &error
);
1071 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1074 struct io_context
*ioc
= current
->io_context
;
1075 struct io_context
*new_ioc
;
1080 * Share io context with parent, if CLONE_IO is set
1082 if (clone_flags
& CLONE_IO
) {
1084 tsk
->io_context
= ioc
;
1085 } else if (ioprio_valid(ioc
->ioprio
)) {
1086 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1087 if (unlikely(!new_ioc
))
1090 new_ioc
->ioprio
= ioc
->ioprio
;
1091 put_io_context(new_ioc
);
1097 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1099 struct sighand_struct
*sig
;
1101 if (clone_flags
& CLONE_SIGHAND
) {
1102 atomic_inc(¤t
->sighand
->count
);
1105 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1106 rcu_assign_pointer(tsk
->sighand
, sig
);
1110 atomic_set(&sig
->count
, 1);
1111 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1115 void __cleanup_sighand(struct sighand_struct
*sighand
)
1117 if (atomic_dec_and_test(&sighand
->count
)) {
1118 signalfd_cleanup(sighand
);
1120 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1121 * without an RCU grace period, see __lock_task_sighand().
1123 kmem_cache_free(sighand_cachep
, sighand
);
1128 * Initialize POSIX timer handling for a thread group.
1130 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1132 unsigned long cpu_limit
;
1134 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1135 if (cpu_limit
!= RLIM_INFINITY
) {
1136 sig
->cputime_expires
.prof_exp
= secs_to_cputime(cpu_limit
);
1137 sig
->cputimer
.running
= true;
1140 /* The timer lists. */
1141 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1142 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1143 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1146 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1148 struct signal_struct
*sig
;
1150 if (clone_flags
& CLONE_THREAD
)
1153 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1158 sig
->nr_threads
= 1;
1159 atomic_set(&sig
->live
, 1);
1160 atomic_set(&sig
->sigcnt
, 1);
1162 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1163 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1164 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1166 init_waitqueue_head(&sig
->wait_chldexit
);
1167 sig
->curr_target
= tsk
;
1168 init_sigpending(&sig
->shared_pending
);
1169 INIT_LIST_HEAD(&sig
->posix_timers
);
1170 seqlock_init(&sig
->stats_lock
);
1171 prev_cputime_init(&sig
->prev_cputime
);
1173 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1174 sig
->real_timer
.function
= it_real_fn
;
1176 task_lock(current
->group_leader
);
1177 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1178 task_unlock(current
->group_leader
);
1180 posix_cpu_timers_init_group(sig
);
1182 tty_audit_fork(sig
);
1183 sched_autogroup_fork(sig
);
1185 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1186 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1188 sig
->has_child_subreaper
= current
->signal
->has_child_subreaper
||
1189 current
->signal
->is_child_subreaper
;
1191 mutex_init(&sig
->cred_guard_mutex
);
1196 static void copy_seccomp(struct task_struct
*p
)
1198 #ifdef CONFIG_SECCOMP
1200 * Must be called with sighand->lock held, which is common to
1201 * all threads in the group. Holding cred_guard_mutex is not
1202 * needed because this new task is not yet running and cannot
1205 assert_spin_locked(¤t
->sighand
->siglock
);
1207 /* Ref-count the new filter user, and assign it. */
1208 get_seccomp_filter(current
);
1209 p
->seccomp
= current
->seccomp
;
1212 * Explicitly enable no_new_privs here in case it got set
1213 * between the task_struct being duplicated and holding the
1214 * sighand lock. The seccomp state and nnp must be in sync.
1216 if (task_no_new_privs(current
))
1217 task_set_no_new_privs(p
);
1220 * If the parent gained a seccomp mode after copying thread
1221 * flags and between before we held the sighand lock, we have
1222 * to manually enable the seccomp thread flag here.
1224 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1225 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1229 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1231 current
->clear_child_tid
= tidptr
;
1233 return task_pid_vnr(current
);
1236 static void rt_mutex_init_task(struct task_struct
*p
)
1238 raw_spin_lock_init(&p
->pi_lock
);
1239 #ifdef CONFIG_RT_MUTEXES
1240 p
->pi_waiters
= RB_ROOT
;
1241 p
->pi_waiters_leftmost
= NULL
;
1242 p
->pi_blocked_on
= NULL
;
1247 * Initialize POSIX timer handling for a single task.
1249 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1251 tsk
->cputime_expires
.prof_exp
= 0;
1252 tsk
->cputime_expires
.virt_exp
= 0;
1253 tsk
->cputime_expires
.sched_exp
= 0;
1254 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1255 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1256 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1260 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1262 task
->pids
[type
].pid
= pid
;
1266 * This creates a new process as a copy of the old one,
1267 * but does not actually start it yet.
1269 * It copies the registers, and all the appropriate
1270 * parts of the process environment (as per the clone
1271 * flags). The actual kick-off is left to the caller.
1273 static struct task_struct
*copy_process(unsigned long clone_flags
,
1274 unsigned long stack_start
,
1275 unsigned long stack_size
,
1276 int __user
*child_tidptr
,
1282 struct task_struct
*p
;
1284 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1285 return ERR_PTR(-EINVAL
);
1287 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1288 return ERR_PTR(-EINVAL
);
1291 * Thread groups must share signals as well, and detached threads
1292 * can only be started up within the thread group.
1294 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1295 return ERR_PTR(-EINVAL
);
1298 * Shared signal handlers imply shared VM. By way of the above,
1299 * thread groups also imply shared VM. Blocking this case allows
1300 * for various simplifications in other code.
1302 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1303 return ERR_PTR(-EINVAL
);
1306 * Siblings of global init remain as zombies on exit since they are
1307 * not reaped by their parent (swapper). To solve this and to avoid
1308 * multi-rooted process trees, prevent global and container-inits
1309 * from creating siblings.
1311 if ((clone_flags
& CLONE_PARENT
) &&
1312 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1313 return ERR_PTR(-EINVAL
);
1316 * If the new process will be in a different pid or user namespace
1317 * do not allow it to share a thread group with the forking task.
1319 if (clone_flags
& CLONE_THREAD
) {
1320 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1321 (task_active_pid_ns(current
) !=
1322 current
->nsproxy
->pid_ns_for_children
))
1323 return ERR_PTR(-EINVAL
);
1326 retval
= security_task_create(clone_flags
);
1331 p
= dup_task_struct(current
);
1335 ftrace_graph_init_task(p
);
1337 rt_mutex_init_task(p
);
1339 #ifdef CONFIG_PROVE_LOCKING
1340 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1341 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1344 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1345 task_rlimit(p
, RLIMIT_NPROC
)) {
1346 if (p
->real_cred
->user
!= INIT_USER
&&
1347 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1350 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1352 retval
= copy_creds(p
, clone_flags
);
1357 * If multiple threads are within copy_process(), then this check
1358 * triggers too late. This doesn't hurt, the check is only there
1359 * to stop root fork bombs.
1362 if (nr_threads
>= max_threads
)
1363 goto bad_fork_cleanup_count
;
1365 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1366 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
);
1367 p
->flags
|= PF_FORKNOEXEC
;
1368 INIT_LIST_HEAD(&p
->children
);
1369 INIT_LIST_HEAD(&p
->sibling
);
1370 rcu_copy_process(p
);
1371 p
->vfork_done
= NULL
;
1372 spin_lock_init(&p
->alloc_lock
);
1374 init_sigpending(&p
->pending
);
1376 p
->utime
= p
->stime
= p
->gtime
= 0;
1377 p
->utimescaled
= p
->stimescaled
= 0;
1378 prev_cputime_init(&p
->prev_cputime
);
1380 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1381 seqcount_init(&p
->vtime_seqcount
);
1383 p
->vtime_snap_whence
= VTIME_INACTIVE
;
1386 #if defined(SPLIT_RSS_COUNTING)
1387 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1390 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1392 task_io_accounting_init(&p
->ioac
);
1393 acct_clear_integrals(p
);
1395 posix_cpu_timers_init(p
);
1397 p
->start_time
= ktime_get_ns();
1398 p
->real_start_time
= ktime_get_boot_ns();
1399 p
->io_context
= NULL
;
1400 p
->audit_context
= NULL
;
1401 threadgroup_change_begin(current
);
1404 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1405 if (IS_ERR(p
->mempolicy
)) {
1406 retval
= PTR_ERR(p
->mempolicy
);
1407 p
->mempolicy
= NULL
;
1408 goto bad_fork_cleanup_threadgroup_lock
;
1411 #ifdef CONFIG_CPUSETS
1412 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1413 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1414 seqcount_init(&p
->mems_allowed_seq
);
1416 #ifdef CONFIG_TRACE_IRQFLAGS
1418 p
->hardirqs_enabled
= 0;
1419 p
->hardirq_enable_ip
= 0;
1420 p
->hardirq_enable_event
= 0;
1421 p
->hardirq_disable_ip
= _THIS_IP_
;
1422 p
->hardirq_disable_event
= 0;
1423 p
->softirqs_enabled
= 1;
1424 p
->softirq_enable_ip
= _THIS_IP_
;
1425 p
->softirq_enable_event
= 0;
1426 p
->softirq_disable_ip
= 0;
1427 p
->softirq_disable_event
= 0;
1428 p
->hardirq_context
= 0;
1429 p
->softirq_context
= 0;
1432 p
->pagefault_disabled
= 0;
1434 #ifdef CONFIG_LOCKDEP
1435 p
->lockdep_depth
= 0; /* no locks held yet */
1436 p
->curr_chain_key
= 0;
1437 p
->lockdep_recursion
= 0;
1440 #ifdef CONFIG_DEBUG_MUTEXES
1441 p
->blocked_on
= NULL
; /* not blocked yet */
1443 #ifdef CONFIG_BCACHE
1444 p
->sequential_io
= 0;
1445 p
->sequential_io_avg
= 0;
1448 /* Perform scheduler related setup. Assign this task to a CPU. */
1449 retval
= sched_fork(clone_flags
, p
);
1451 goto bad_fork_cleanup_policy
;
1453 retval
= perf_event_init_task(p
);
1455 goto bad_fork_cleanup_policy
;
1456 retval
= audit_alloc(p
);
1458 goto bad_fork_cleanup_perf
;
1459 /* copy all the process information */
1461 retval
= copy_semundo(clone_flags
, p
);
1463 goto bad_fork_cleanup_audit
;
1464 retval
= copy_files(clone_flags
, p
);
1466 goto bad_fork_cleanup_semundo
;
1467 retval
= copy_fs(clone_flags
, p
);
1469 goto bad_fork_cleanup_files
;
1470 retval
= copy_sighand(clone_flags
, p
);
1472 goto bad_fork_cleanup_fs
;
1473 retval
= copy_signal(clone_flags
, p
);
1475 goto bad_fork_cleanup_sighand
;
1476 retval
= copy_mm(clone_flags
, p
);
1478 goto bad_fork_cleanup_signal
;
1479 retval
= copy_namespaces(clone_flags
, p
);
1481 goto bad_fork_cleanup_mm
;
1482 retval
= copy_io(clone_flags
, p
);
1484 goto bad_fork_cleanup_namespaces
;
1485 retval
= copy_thread_tls(clone_flags
, stack_start
, stack_size
, p
, tls
);
1487 goto bad_fork_cleanup_io
;
1489 if (pid
!= &init_struct_pid
) {
1490 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
1492 retval
= PTR_ERR(pid
);
1493 goto bad_fork_cleanup_thread
;
1497 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
1499 * Clear TID on mm_release()?
1501 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
1506 p
->robust_list
= NULL
;
1507 #ifdef CONFIG_COMPAT
1508 p
->compat_robust_list
= NULL
;
1510 INIT_LIST_HEAD(&p
->pi_state_list
);
1511 p
->pi_state_cache
= NULL
;
1514 * sigaltstack should be cleared when sharing the same VM
1516 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
1520 * Syscall tracing and stepping should be turned off in the
1521 * child regardless of CLONE_PTRACE.
1523 user_disable_single_step(p
);
1524 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
1525 #ifdef TIF_SYSCALL_EMU
1526 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1528 clear_all_latency_tracing(p
);
1530 /* ok, now we should be set up.. */
1531 p
->pid
= pid_nr(pid
);
1532 if (clone_flags
& CLONE_THREAD
) {
1533 p
->exit_signal
= -1;
1534 p
->group_leader
= current
->group_leader
;
1535 p
->tgid
= current
->tgid
;
1537 if (clone_flags
& CLONE_PARENT
)
1538 p
->exit_signal
= current
->group_leader
->exit_signal
;
1540 p
->exit_signal
= (clone_flags
& CSIGNAL
);
1541 p
->group_leader
= p
;
1546 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
1547 p
->dirty_paused_when
= 0;
1549 p
->pdeath_signal
= 0;
1550 INIT_LIST_HEAD(&p
->thread_group
);
1551 p
->task_works
= NULL
;
1554 * Ensure that the cgroup subsystem policies allow the new process to be
1555 * forked. It should be noted the the new process's css_set can be changed
1556 * between here and cgroup_post_fork() if an organisation operation is in
1559 retval
= cgroup_can_fork(p
);
1561 goto bad_fork_free_pid
;
1564 * Make it visible to the rest of the system, but dont wake it up yet.
1565 * Need tasklist lock for parent etc handling!
1567 write_lock_irq(&tasklist_lock
);
1569 /* CLONE_PARENT re-uses the old parent */
1570 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
1571 p
->real_parent
= current
->real_parent
;
1572 p
->parent_exec_id
= current
->parent_exec_id
;
1574 p
->real_parent
= current
;
1575 p
->parent_exec_id
= current
->self_exec_id
;
1578 spin_lock(¤t
->sighand
->siglock
);
1581 * Copy seccomp details explicitly here, in case they were changed
1582 * before holding sighand lock.
1587 * Process group and session signals need to be delivered to just the
1588 * parent before the fork or both the parent and the child after the
1589 * fork. Restart if a signal comes in before we add the new process to
1590 * it's process group.
1591 * A fatal signal pending means that current will exit, so the new
1592 * thread can't slip out of an OOM kill (or normal SIGKILL).
1594 recalc_sigpending();
1595 if (signal_pending(current
)) {
1596 spin_unlock(¤t
->sighand
->siglock
);
1597 write_unlock_irq(&tasklist_lock
);
1598 retval
= -ERESTARTNOINTR
;
1599 goto bad_fork_cancel_cgroup
;
1602 if (likely(p
->pid
)) {
1603 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
1605 init_task_pid(p
, PIDTYPE_PID
, pid
);
1606 if (thread_group_leader(p
)) {
1607 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
1608 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
1610 if (is_child_reaper(pid
)) {
1611 ns_of_pid(pid
)->child_reaper
= p
;
1612 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
1615 p
->signal
->leader_pid
= pid
;
1616 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
1617 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
1618 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
1619 attach_pid(p
, PIDTYPE_PGID
);
1620 attach_pid(p
, PIDTYPE_SID
);
1621 __this_cpu_inc(process_counts
);
1623 current
->signal
->nr_threads
++;
1624 atomic_inc(¤t
->signal
->live
);
1625 atomic_inc(¤t
->signal
->sigcnt
);
1626 list_add_tail_rcu(&p
->thread_group
,
1627 &p
->group_leader
->thread_group
);
1628 list_add_tail_rcu(&p
->thread_node
,
1629 &p
->signal
->thread_head
);
1631 attach_pid(p
, PIDTYPE_PID
);
1636 spin_unlock(¤t
->sighand
->siglock
);
1637 syscall_tracepoint_update(p
);
1638 write_unlock_irq(&tasklist_lock
);
1640 proc_fork_connector(p
);
1641 cgroup_post_fork(p
);
1642 threadgroup_change_end(current
);
1645 trace_task_newtask(p
, clone_flags
);
1646 uprobe_copy_process(p
, clone_flags
);
1650 bad_fork_cancel_cgroup
:
1651 cgroup_cancel_fork(p
);
1653 if (pid
!= &init_struct_pid
)
1655 bad_fork_cleanup_thread
:
1657 bad_fork_cleanup_io
:
1660 bad_fork_cleanup_namespaces
:
1661 exit_task_namespaces(p
);
1662 bad_fork_cleanup_mm
:
1665 bad_fork_cleanup_signal
:
1666 if (!(clone_flags
& CLONE_THREAD
))
1667 free_signal_struct(p
->signal
);
1668 bad_fork_cleanup_sighand
:
1669 __cleanup_sighand(p
->sighand
);
1670 bad_fork_cleanup_fs
:
1671 exit_fs(p
); /* blocking */
1672 bad_fork_cleanup_files
:
1673 exit_files(p
); /* blocking */
1674 bad_fork_cleanup_semundo
:
1676 bad_fork_cleanup_audit
:
1678 bad_fork_cleanup_perf
:
1679 perf_event_free_task(p
);
1680 bad_fork_cleanup_policy
:
1682 mpol_put(p
->mempolicy
);
1683 bad_fork_cleanup_threadgroup_lock
:
1685 threadgroup_change_end(current
);
1686 delayacct_tsk_free(p
);
1687 bad_fork_cleanup_count
:
1688 atomic_dec(&p
->cred
->user
->processes
);
1693 return ERR_PTR(retval
);
1696 static inline void init_idle_pids(struct pid_link
*links
)
1700 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1701 INIT_HLIST_NODE(&links
[type
].node
); /* not really needed */
1702 links
[type
].pid
= &init_struct_pid
;
1706 struct task_struct
*fork_idle(int cpu
)
1708 struct task_struct
*task
;
1709 task
= copy_process(CLONE_VM
, 0, 0, NULL
, &init_struct_pid
, 0, 0);
1710 if (!IS_ERR(task
)) {
1711 init_idle_pids(task
->pids
);
1712 init_idle(task
, cpu
);
1719 * Ok, this is the main fork-routine.
1721 * It copies the process, and if successful kick-starts
1722 * it and waits for it to finish using the VM if required.
1724 long _do_fork(unsigned long clone_flags
,
1725 unsigned long stack_start
,
1726 unsigned long stack_size
,
1727 int __user
*parent_tidptr
,
1728 int __user
*child_tidptr
,
1731 struct task_struct
*p
;
1736 * Determine whether and which event to report to ptracer. When
1737 * called from kernel_thread or CLONE_UNTRACED is explicitly
1738 * requested, no event is reported; otherwise, report if the event
1739 * for the type of forking is enabled.
1741 if (!(clone_flags
& CLONE_UNTRACED
)) {
1742 if (clone_flags
& CLONE_VFORK
)
1743 trace
= PTRACE_EVENT_VFORK
;
1744 else if ((clone_flags
& CSIGNAL
) != SIGCHLD
)
1745 trace
= PTRACE_EVENT_CLONE
;
1747 trace
= PTRACE_EVENT_FORK
;
1749 if (likely(!ptrace_event_enabled(current
, trace
)))
1753 p
= copy_process(clone_flags
, stack_start
, stack_size
,
1754 child_tidptr
, NULL
, trace
, tls
);
1756 * Do this prior waking up the new thread - the thread pointer
1757 * might get invalid after that point, if the thread exits quickly.
1760 struct completion vfork
;
1763 trace_sched_process_fork(current
, p
);
1765 pid
= get_task_pid(p
, PIDTYPE_PID
);
1768 if (clone_flags
& CLONE_PARENT_SETTID
)
1769 put_user(nr
, parent_tidptr
);
1771 if (clone_flags
& CLONE_VFORK
) {
1772 p
->vfork_done
= &vfork
;
1773 init_completion(&vfork
);
1777 wake_up_new_task(p
);
1779 /* forking complete and child started to run, tell ptracer */
1780 if (unlikely(trace
))
1781 ptrace_event_pid(trace
, pid
);
1783 if (clone_flags
& CLONE_VFORK
) {
1784 if (!wait_for_vfork_done(p
, &vfork
))
1785 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
1795 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1796 /* For compatibility with architectures that call do_fork directly rather than
1797 * using the syscall entry points below. */
1798 long do_fork(unsigned long clone_flags
,
1799 unsigned long stack_start
,
1800 unsigned long stack_size
,
1801 int __user
*parent_tidptr
,
1802 int __user
*child_tidptr
)
1804 return _do_fork(clone_flags
, stack_start
, stack_size
,
1805 parent_tidptr
, child_tidptr
, 0);
1810 * Create a kernel thread.
1812 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
1814 return _do_fork(flags
|CLONE_VM
|CLONE_UNTRACED
, (unsigned long)fn
,
1815 (unsigned long)arg
, NULL
, NULL
, 0);
1818 #ifdef __ARCH_WANT_SYS_FORK
1819 SYSCALL_DEFINE0(fork
)
1822 return _do_fork(SIGCHLD
, 0, 0, NULL
, NULL
, 0);
1824 /* can not support in nommu mode */
1830 #ifdef __ARCH_WANT_SYS_VFORK
1831 SYSCALL_DEFINE0(vfork
)
1833 return _do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, 0,
1838 #ifdef __ARCH_WANT_SYS_CLONE
1839 #ifdef CONFIG_CLONE_BACKWARDS
1840 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1841 int __user
*, parent_tidptr
,
1843 int __user
*, child_tidptr
)
1844 #elif defined(CONFIG_CLONE_BACKWARDS2)
1845 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
1846 int __user
*, parent_tidptr
,
1847 int __user
*, child_tidptr
,
1849 #elif defined(CONFIG_CLONE_BACKWARDS3)
1850 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1852 int __user
*, parent_tidptr
,
1853 int __user
*, child_tidptr
,
1856 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1857 int __user
*, parent_tidptr
,
1858 int __user
*, child_tidptr
,
1862 return _do_fork(clone_flags
, newsp
, 0, parent_tidptr
, child_tidptr
, tls
);
1866 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1867 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1870 static void sighand_ctor(void *data
)
1872 struct sighand_struct
*sighand
= data
;
1874 spin_lock_init(&sighand
->siglock
);
1875 init_waitqueue_head(&sighand
->signalfd_wqh
);
1878 void __init
proc_caches_init(void)
1880 sighand_cachep
= kmem_cache_create("sighand_cache",
1881 sizeof(struct sighand_struct
), 0,
1882 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_DESTROY_BY_RCU
|
1883 SLAB_NOTRACK
|SLAB_ACCOUNT
, sighand_ctor
);
1884 signal_cachep
= kmem_cache_create("signal_cache",
1885 sizeof(struct signal_struct
), 0,
1886 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1888 files_cachep
= kmem_cache_create("files_cache",
1889 sizeof(struct files_struct
), 0,
1890 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1892 fs_cachep
= kmem_cache_create("fs_cache",
1893 sizeof(struct fs_struct
), 0,
1894 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1897 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1898 * whole struct cpumask for the OFFSTACK case. We could change
1899 * this to *only* allocate as much of it as required by the
1900 * maximum number of CPU's we can ever have. The cpumask_allocation
1901 * is at the end of the structure, exactly for that reason.
1903 mm_cachep
= kmem_cache_create("mm_struct",
1904 sizeof(struct mm_struct
), ARCH_MIN_MMSTRUCT_ALIGN
,
1905 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1907 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
1909 nsproxy_cache_init();
1913 * Check constraints on flags passed to the unshare system call.
1915 static int check_unshare_flags(unsigned long unshare_flags
)
1917 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
1918 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
1919 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
1920 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
1923 * Not implemented, but pretend it works if there is nothing
1924 * to unshare. Note that unsharing the address space or the
1925 * signal handlers also need to unshare the signal queues (aka
1928 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
1929 if (!thread_group_empty(current
))
1932 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
1933 if (atomic_read(¤t
->sighand
->count
) > 1)
1936 if (unshare_flags
& CLONE_VM
) {
1937 if (!current_is_single_threaded())
1945 * Unshare the filesystem structure if it is being shared
1947 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
1949 struct fs_struct
*fs
= current
->fs
;
1951 if (!(unshare_flags
& CLONE_FS
) || !fs
)
1954 /* don't need lock here; in the worst case we'll do useless copy */
1958 *new_fsp
= copy_fs_struct(fs
);
1966 * Unshare file descriptor table if it is being shared
1968 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
1970 struct files_struct
*fd
= current
->files
;
1973 if ((unshare_flags
& CLONE_FILES
) &&
1974 (fd
&& atomic_read(&fd
->count
) > 1)) {
1975 *new_fdp
= dup_fd(fd
, &error
);
1984 * unshare allows a process to 'unshare' part of the process
1985 * context which was originally shared using clone. copy_*
1986 * functions used by do_fork() cannot be used here directly
1987 * because they modify an inactive task_struct that is being
1988 * constructed. Here we are modifying the current, active,
1991 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
1993 struct fs_struct
*fs
, *new_fs
= NULL
;
1994 struct files_struct
*fd
, *new_fd
= NULL
;
1995 struct cred
*new_cred
= NULL
;
1996 struct nsproxy
*new_nsproxy
= NULL
;
2001 * If unsharing a user namespace must also unshare the thread group
2002 * and unshare the filesystem root and working directories.
2004 if (unshare_flags
& CLONE_NEWUSER
)
2005 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2007 * If unsharing vm, must also unshare signal handlers.
2009 if (unshare_flags
& CLONE_VM
)
2010 unshare_flags
|= CLONE_SIGHAND
;
2012 * If unsharing a signal handlers, must also unshare the signal queues.
2014 if (unshare_flags
& CLONE_SIGHAND
)
2015 unshare_flags
|= CLONE_THREAD
;
2017 * If unsharing namespace, must also unshare filesystem information.
2019 if (unshare_flags
& CLONE_NEWNS
)
2020 unshare_flags
|= CLONE_FS
;
2022 err
= check_unshare_flags(unshare_flags
);
2024 goto bad_unshare_out
;
2026 * CLONE_NEWIPC must also detach from the undolist: after switching
2027 * to a new ipc namespace, the semaphore arrays from the old
2028 * namespace are unreachable.
2030 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2032 err
= unshare_fs(unshare_flags
, &new_fs
);
2034 goto bad_unshare_out
;
2035 err
= unshare_fd(unshare_flags
, &new_fd
);
2037 goto bad_unshare_cleanup_fs
;
2038 err
= unshare_userns(unshare_flags
, &new_cred
);
2040 goto bad_unshare_cleanup_fd
;
2041 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2044 goto bad_unshare_cleanup_cred
;
2046 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2049 * CLONE_SYSVSEM is equivalent to sys_exit().
2053 if (unshare_flags
& CLONE_NEWIPC
) {
2054 /* Orphan segments in old ns (see sem above). */
2056 shm_init_task(current
);
2060 switch_task_namespaces(current
, new_nsproxy
);
2066 spin_lock(&fs
->lock
);
2067 current
->fs
= new_fs
;
2072 spin_unlock(&fs
->lock
);
2076 fd
= current
->files
;
2077 current
->files
= new_fd
;
2081 task_unlock(current
);
2084 /* Install the new user namespace */
2085 commit_creds(new_cred
);
2090 bad_unshare_cleanup_cred
:
2093 bad_unshare_cleanup_fd
:
2095 put_files_struct(new_fd
);
2097 bad_unshare_cleanup_fs
:
2099 free_fs_struct(new_fs
);
2106 * Helper to unshare the files of the current task.
2107 * We don't want to expose copy_files internals to
2108 * the exec layer of the kernel.
2111 int unshare_files(struct files_struct
**displaced
)
2113 struct task_struct
*task
= current
;
2114 struct files_struct
*copy
= NULL
;
2117 error
= unshare_fd(CLONE_FILES
, ©
);
2118 if (error
|| !copy
) {
2122 *displaced
= task
->files
;
2129 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2130 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2134 int threads
= max_threads
;
2135 int min
= MIN_THREADS
;
2136 int max
= MAX_THREADS
;
2143 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2147 set_max_threads(threads
);