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
);
289 kmem_cache_free(signal_cachep
, sig
);
292 static inline void put_signal_struct(struct signal_struct
*sig
)
294 if (atomic_dec_and_test(&sig
->sigcnt
))
295 free_signal_struct(sig
);
298 void __put_task_struct(struct task_struct
*tsk
)
300 WARN_ON(!tsk
->exit_state
);
301 WARN_ON(atomic_read(&tsk
->usage
));
302 WARN_ON(tsk
== current
);
306 security_task_free(tsk
);
308 delayacct_tsk_free(tsk
);
309 put_signal_struct(tsk
->signal
);
311 if (!profile_handoff_task(tsk
))
314 EXPORT_SYMBOL_GPL(__put_task_struct
);
316 void __init __weak
arch_task_cache_init(void) { }
321 static void set_max_threads(unsigned int max_threads_suggested
)
326 * The number of threads shall be limited such that the thread
327 * structures may only consume a small part of the available memory.
329 if (fls64(totalram_pages
) + fls64(PAGE_SIZE
) > 64)
330 threads
= MAX_THREADS
;
332 threads
= div64_u64((u64
) totalram_pages
* (u64
) PAGE_SIZE
,
333 (u64
) THREAD_SIZE
* 8UL);
335 if (threads
> max_threads_suggested
)
336 threads
= max_threads_suggested
;
338 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
341 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
342 /* Initialized by the architecture: */
343 int arch_task_struct_size __read_mostly
;
346 void __init
fork_init(void)
348 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
349 #ifndef ARCH_MIN_TASKALIGN
350 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
352 /* create a slab on which task_structs can be allocated */
353 task_struct_cachep
= kmem_cache_create("task_struct",
354 arch_task_struct_size
, ARCH_MIN_TASKALIGN
,
355 SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
, NULL
);
358 /* do the arch specific task caches init */
359 arch_task_cache_init();
361 set_max_threads(MAX_THREADS
);
363 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
364 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
365 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
366 init_task
.signal
->rlim
[RLIMIT_NPROC
];
369 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
370 struct task_struct
*src
)
376 void set_task_stack_end_magic(struct task_struct
*tsk
)
378 unsigned long *stackend
;
380 stackend
= end_of_stack(tsk
);
381 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
384 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
386 struct task_struct
*tsk
;
387 unsigned long *stack
;
388 struct vm_struct
*stack_vm_area
;
391 if (node
== NUMA_NO_NODE
)
392 node
= tsk_fork_get_node(orig
);
393 tsk
= alloc_task_struct_node(node
);
397 stack
= alloc_thread_stack_node(tsk
, node
);
401 stack_vm_area
= task_stack_vm_area(tsk
);
403 err
= arch_dup_task_struct(tsk
, orig
);
406 * arch_dup_task_struct() clobbers the stack-related fields. Make
407 * sure they're properly initialized before using any stack-related
411 #ifdef CONFIG_VMAP_STACK
412 tsk
->stack_vm_area
= stack_vm_area
;
418 #ifdef CONFIG_SECCOMP
420 * We must handle setting up seccomp filters once we're under
421 * the sighand lock in case orig has changed between now and
422 * then. Until then, filter must be NULL to avoid messing up
423 * the usage counts on the error path calling free_task.
425 tsk
->seccomp
.filter
= NULL
;
428 setup_thread_stack(tsk
, orig
);
429 clear_user_return_notifier(tsk
);
430 clear_tsk_need_resched(tsk
);
431 set_task_stack_end_magic(tsk
);
433 #ifdef CONFIG_CC_STACKPROTECTOR
434 tsk
->stack_canary
= get_random_int();
438 * One for us, one for whoever does the "release_task()" (usually
441 atomic_set(&tsk
->usage
, 2);
442 #ifdef CONFIG_BLK_DEV_IO_TRACE
445 tsk
->splice_pipe
= NULL
;
446 tsk
->task_frag
.page
= NULL
;
447 tsk
->wake_q
.next
= NULL
;
449 account_kernel_stack(tsk
, 1);
456 free_thread_stack(tsk
);
458 free_task_struct(tsk
);
463 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
465 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
466 struct rb_node
**rb_link
, *rb_parent
;
468 unsigned long charge
;
470 uprobe_start_dup_mmap();
471 if (down_write_killable(&oldmm
->mmap_sem
)) {
473 goto fail_uprobe_end
;
475 flush_cache_dup_mm(oldmm
);
476 uprobe_dup_mmap(oldmm
, mm
);
478 * Not linked in yet - no deadlock potential:
480 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
482 /* No ordering required: file already has been exposed. */
483 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
485 mm
->total_vm
= oldmm
->total_vm
;
486 mm
->data_vm
= oldmm
->data_vm
;
487 mm
->exec_vm
= oldmm
->exec_vm
;
488 mm
->stack_vm
= oldmm
->stack_vm
;
490 rb_link
= &mm
->mm_rb
.rb_node
;
493 retval
= ksm_fork(mm
, oldmm
);
496 retval
= khugepaged_fork(mm
, oldmm
);
501 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
504 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
505 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
509 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
510 unsigned long len
= vma_pages(mpnt
);
512 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
516 tmp
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
520 INIT_LIST_HEAD(&tmp
->anon_vma_chain
);
521 retval
= vma_dup_policy(mpnt
, tmp
);
523 goto fail_nomem_policy
;
525 if (anon_vma_fork(tmp
, mpnt
))
526 goto fail_nomem_anon_vma_fork
;
528 ~(VM_LOCKED
|VM_LOCKONFAULT
|VM_UFFD_MISSING
|VM_UFFD_WP
);
529 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
530 tmp
->vm_userfaultfd_ctx
= NULL_VM_UFFD_CTX
;
533 struct inode
*inode
= file_inode(file
);
534 struct address_space
*mapping
= file
->f_mapping
;
537 if (tmp
->vm_flags
& VM_DENYWRITE
)
538 atomic_dec(&inode
->i_writecount
);
539 i_mmap_lock_write(mapping
);
540 if (tmp
->vm_flags
& VM_SHARED
)
541 atomic_inc(&mapping
->i_mmap_writable
);
542 flush_dcache_mmap_lock(mapping
);
543 /* insert tmp into the share list, just after mpnt */
544 vma_interval_tree_insert_after(tmp
, mpnt
,
546 flush_dcache_mmap_unlock(mapping
);
547 i_mmap_unlock_write(mapping
);
551 * Clear hugetlb-related page reserves for children. This only
552 * affects MAP_PRIVATE mappings. Faults generated by the child
553 * are not guaranteed to succeed, even if read-only
555 if (is_vm_hugetlb_page(tmp
))
556 reset_vma_resv_huge_pages(tmp
);
559 * Link in the new vma and copy the page table entries.
562 pprev
= &tmp
->vm_next
;
566 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
567 rb_link
= &tmp
->vm_rb
.rb_right
;
568 rb_parent
= &tmp
->vm_rb
;
571 retval
= copy_page_range(mm
, oldmm
, mpnt
);
573 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
574 tmp
->vm_ops
->open(tmp
);
579 /* a new mm has just been created */
580 arch_dup_mmap(oldmm
, mm
);
583 up_write(&mm
->mmap_sem
);
585 up_write(&oldmm
->mmap_sem
);
587 uprobe_end_dup_mmap();
589 fail_nomem_anon_vma_fork
:
590 mpol_put(vma_policy(tmp
));
592 kmem_cache_free(vm_area_cachep
, tmp
);
595 vm_unacct_memory(charge
);
599 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
601 mm
->pgd
= pgd_alloc(mm
);
602 if (unlikely(!mm
->pgd
))
607 static inline void mm_free_pgd(struct mm_struct
*mm
)
609 pgd_free(mm
, mm
->pgd
);
612 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
614 down_write(&oldmm
->mmap_sem
);
615 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
616 up_write(&oldmm
->mmap_sem
);
619 #define mm_alloc_pgd(mm) (0)
620 #define mm_free_pgd(mm)
621 #endif /* CONFIG_MMU */
623 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
625 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
626 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
628 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
630 static int __init
coredump_filter_setup(char *s
)
632 default_dump_filter
=
633 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
634 MMF_DUMP_FILTER_MASK
;
638 __setup("coredump_filter=", coredump_filter_setup
);
640 #include <linux/init_task.h>
642 static void mm_init_aio(struct mm_struct
*mm
)
645 spin_lock_init(&mm
->ioctx_lock
);
646 mm
->ioctx_table
= NULL
;
650 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
657 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
)
661 mm
->vmacache_seqnum
= 0;
662 atomic_set(&mm
->mm_users
, 1);
663 atomic_set(&mm
->mm_count
, 1);
664 init_rwsem(&mm
->mmap_sem
);
665 INIT_LIST_HEAD(&mm
->mmlist
);
666 mm
->core_state
= NULL
;
667 atomic_long_set(&mm
->nr_ptes
, 0);
672 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
673 spin_lock_init(&mm
->page_table_lock
);
676 mm_init_owner(mm
, p
);
677 mmu_notifier_mm_init(mm
);
678 clear_tlb_flush_pending(mm
);
679 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
680 mm
->pmd_huge_pte
= NULL
;
684 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
685 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
687 mm
->flags
= default_dump_filter
;
691 if (mm_alloc_pgd(mm
))
694 if (init_new_context(p
, mm
))
706 static void check_mm(struct mm_struct
*mm
)
710 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
711 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
714 printk(KERN_ALERT
"BUG: Bad rss-counter state "
715 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
718 if (atomic_long_read(&mm
->nr_ptes
))
719 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
720 atomic_long_read(&mm
->nr_ptes
));
722 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
725 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
726 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
731 * Allocate and initialize an mm_struct.
733 struct mm_struct
*mm_alloc(void)
735 struct mm_struct
*mm
;
741 memset(mm
, 0, sizeof(*mm
));
742 return mm_init(mm
, current
);
746 * Called when the last reference to the mm
747 * is dropped: either by a lazy thread or by
748 * mmput. Free the page directory and the mm.
750 void __mmdrop(struct mm_struct
*mm
)
752 BUG_ON(mm
== &init_mm
);
755 mmu_notifier_mm_destroy(mm
);
759 EXPORT_SYMBOL_GPL(__mmdrop
);
761 static inline void __mmput(struct mm_struct
*mm
)
763 VM_BUG_ON(atomic_read(&mm
->mm_users
));
765 uprobe_clear_state(mm
);
768 khugepaged_exit(mm
); /* must run before exit_mmap */
770 set_mm_exe_file(mm
, NULL
);
771 if (!list_empty(&mm
->mmlist
)) {
772 spin_lock(&mmlist_lock
);
773 list_del(&mm
->mmlist
);
774 spin_unlock(&mmlist_lock
);
777 module_put(mm
->binfmt
->module
);
782 * Decrement the use count and release all resources for an mm.
784 void mmput(struct mm_struct
*mm
)
788 if (atomic_dec_and_test(&mm
->mm_users
))
791 EXPORT_SYMBOL_GPL(mmput
);
794 static void mmput_async_fn(struct work_struct
*work
)
796 struct mm_struct
*mm
= container_of(work
, struct mm_struct
, async_put_work
);
800 void mmput_async(struct mm_struct
*mm
)
802 if (atomic_dec_and_test(&mm
->mm_users
)) {
803 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
804 schedule_work(&mm
->async_put_work
);
810 * set_mm_exe_file - change a reference to the mm's executable file
812 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
814 * Main users are mmput() and sys_execve(). Callers prevent concurrent
815 * invocations: in mmput() nobody alive left, in execve task is single
816 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
817 * mm->exe_file, but does so without using set_mm_exe_file() in order
818 * to do avoid the need for any locks.
820 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
822 struct file
*old_exe_file
;
825 * It is safe to dereference the exe_file without RCU as
826 * this function is only called if nobody else can access
827 * this mm -- see comment above for justification.
829 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
832 get_file(new_exe_file
);
833 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
839 * get_mm_exe_file - acquire a reference to the mm's executable file
841 * Returns %NULL if mm has no associated executable file.
842 * User must release file via fput().
844 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
846 struct file
*exe_file
;
849 exe_file
= rcu_dereference(mm
->exe_file
);
850 if (exe_file
&& !get_file_rcu(exe_file
))
855 EXPORT_SYMBOL(get_mm_exe_file
);
858 * get_task_exe_file - acquire a reference to the task's executable file
860 * Returns %NULL if task's mm (if any) has no associated executable file or
861 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
862 * User must release file via fput().
864 struct file
*get_task_exe_file(struct task_struct
*task
)
866 struct file
*exe_file
= NULL
;
867 struct mm_struct
*mm
;
872 if (!(task
->flags
& PF_KTHREAD
))
873 exe_file
= get_mm_exe_file(mm
);
878 EXPORT_SYMBOL(get_task_exe_file
);
881 * get_task_mm - acquire a reference to the task's mm
883 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
884 * this kernel workthread has transiently adopted a user mm with use_mm,
885 * to do its AIO) is not set and if so returns a reference to it, after
886 * bumping up the use count. User must release the mm via mmput()
887 * after use. Typically used by /proc and ptrace.
889 struct mm_struct
*get_task_mm(struct task_struct
*task
)
891 struct mm_struct
*mm
;
896 if (task
->flags
& PF_KTHREAD
)
899 atomic_inc(&mm
->mm_users
);
904 EXPORT_SYMBOL_GPL(get_task_mm
);
906 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
908 struct mm_struct
*mm
;
911 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
915 mm
= get_task_mm(task
);
916 if (mm
&& mm
!= current
->mm
&&
917 !ptrace_may_access(task
, mode
)) {
919 mm
= ERR_PTR(-EACCES
);
921 mutex_unlock(&task
->signal
->cred_guard_mutex
);
926 static void complete_vfork_done(struct task_struct
*tsk
)
928 struct completion
*vfork
;
931 vfork
= tsk
->vfork_done
;
933 tsk
->vfork_done
= NULL
;
939 static int wait_for_vfork_done(struct task_struct
*child
,
940 struct completion
*vfork
)
944 freezer_do_not_count();
945 killed
= wait_for_completion_killable(vfork
);
950 child
->vfork_done
= NULL
;
954 put_task_struct(child
);
958 /* Please note the differences between mmput and mm_release.
959 * mmput is called whenever we stop holding onto a mm_struct,
960 * error success whatever.
962 * mm_release is called after a mm_struct has been removed
963 * from the current process.
965 * This difference is important for error handling, when we
966 * only half set up a mm_struct for a new process and need to restore
967 * the old one. Because we mmput the new mm_struct before
968 * restoring the old one. . .
969 * Eric Biederman 10 January 1998
971 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
973 /* Get rid of any futexes when releasing the mm */
975 if (unlikely(tsk
->robust_list
)) {
976 exit_robust_list(tsk
);
977 tsk
->robust_list
= NULL
;
980 if (unlikely(tsk
->compat_robust_list
)) {
981 compat_exit_robust_list(tsk
);
982 tsk
->compat_robust_list
= NULL
;
985 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
986 exit_pi_state_list(tsk
);
989 uprobe_free_utask(tsk
);
991 /* Get rid of any cached register state */
992 deactivate_mm(tsk
, mm
);
995 * Signal userspace if we're not exiting with a core dump
996 * because we want to leave the value intact for debugging
999 if (tsk
->clear_child_tid
) {
1000 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1001 atomic_read(&mm
->mm_users
) > 1) {
1003 * We don't check the error code - if userspace has
1004 * not set up a proper pointer then tough luck.
1006 put_user(0, tsk
->clear_child_tid
);
1007 sys_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1010 tsk
->clear_child_tid
= NULL
;
1014 * All done, finally we can wake up parent and return this mm to him.
1015 * Also kthread_stop() uses this completion for synchronization.
1017 if (tsk
->vfork_done
)
1018 complete_vfork_done(tsk
);
1022 * Allocate a new mm structure and copy contents from the
1023 * mm structure of the passed in task structure.
1025 static struct mm_struct
*dup_mm(struct task_struct
*tsk
)
1027 struct mm_struct
*mm
, *oldmm
= current
->mm
;
1034 memcpy(mm
, oldmm
, sizeof(*mm
));
1036 if (!mm_init(mm
, tsk
))
1039 err
= dup_mmap(mm
, oldmm
);
1043 mm
->hiwater_rss
= get_mm_rss(mm
);
1044 mm
->hiwater_vm
= mm
->total_vm
;
1046 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1052 /* don't put binfmt in mmput, we haven't got module yet */
1060 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1062 struct mm_struct
*mm
, *oldmm
;
1065 tsk
->min_flt
= tsk
->maj_flt
= 0;
1066 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1067 #ifdef CONFIG_DETECT_HUNG_TASK
1068 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1072 tsk
->active_mm
= NULL
;
1075 * Are we cloning a kernel thread?
1077 * We need to steal a active VM for that..
1079 oldmm
= current
->mm
;
1083 /* initialize the new vmacache entries */
1084 vmacache_flush(tsk
);
1086 if (clone_flags
& CLONE_VM
) {
1087 atomic_inc(&oldmm
->mm_users
);
1099 tsk
->active_mm
= mm
;
1106 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1108 struct fs_struct
*fs
= current
->fs
;
1109 if (clone_flags
& CLONE_FS
) {
1110 /* tsk->fs is already what we want */
1111 spin_lock(&fs
->lock
);
1113 spin_unlock(&fs
->lock
);
1117 spin_unlock(&fs
->lock
);
1120 tsk
->fs
= copy_fs_struct(fs
);
1126 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1128 struct files_struct
*oldf
, *newf
;
1132 * A background process may not have any files ...
1134 oldf
= current
->files
;
1138 if (clone_flags
& CLONE_FILES
) {
1139 atomic_inc(&oldf
->count
);
1143 newf
= dup_fd(oldf
, &error
);
1153 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1156 struct io_context
*ioc
= current
->io_context
;
1157 struct io_context
*new_ioc
;
1162 * Share io context with parent, if CLONE_IO is set
1164 if (clone_flags
& CLONE_IO
) {
1166 tsk
->io_context
= ioc
;
1167 } else if (ioprio_valid(ioc
->ioprio
)) {
1168 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1169 if (unlikely(!new_ioc
))
1172 new_ioc
->ioprio
= ioc
->ioprio
;
1173 put_io_context(new_ioc
);
1179 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1181 struct sighand_struct
*sig
;
1183 if (clone_flags
& CLONE_SIGHAND
) {
1184 atomic_inc(¤t
->sighand
->count
);
1187 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1188 rcu_assign_pointer(tsk
->sighand
, sig
);
1192 atomic_set(&sig
->count
, 1);
1193 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1197 void __cleanup_sighand(struct sighand_struct
*sighand
)
1199 if (atomic_dec_and_test(&sighand
->count
)) {
1200 signalfd_cleanup(sighand
);
1202 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1203 * without an RCU grace period, see __lock_task_sighand().
1205 kmem_cache_free(sighand_cachep
, sighand
);
1210 * Initialize POSIX timer handling for a thread group.
1212 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1214 unsigned long cpu_limit
;
1216 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1217 if (cpu_limit
!= RLIM_INFINITY
) {
1218 sig
->cputime_expires
.prof_exp
= secs_to_cputime(cpu_limit
);
1219 sig
->cputimer
.running
= true;
1222 /* The timer lists. */
1223 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1224 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1225 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1228 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1230 struct signal_struct
*sig
;
1232 if (clone_flags
& CLONE_THREAD
)
1235 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1240 sig
->nr_threads
= 1;
1241 atomic_set(&sig
->live
, 1);
1242 atomic_set(&sig
->sigcnt
, 1);
1244 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1245 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1246 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1248 init_waitqueue_head(&sig
->wait_chldexit
);
1249 sig
->curr_target
= tsk
;
1250 init_sigpending(&sig
->shared_pending
);
1251 INIT_LIST_HEAD(&sig
->posix_timers
);
1252 seqlock_init(&sig
->stats_lock
);
1253 prev_cputime_init(&sig
->prev_cputime
);
1255 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1256 sig
->real_timer
.function
= it_real_fn
;
1258 task_lock(current
->group_leader
);
1259 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1260 task_unlock(current
->group_leader
);
1262 posix_cpu_timers_init_group(sig
);
1264 tty_audit_fork(sig
);
1265 sched_autogroup_fork(sig
);
1267 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1268 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1270 sig
->has_child_subreaper
= current
->signal
->has_child_subreaper
||
1271 current
->signal
->is_child_subreaper
;
1273 mutex_init(&sig
->cred_guard_mutex
);
1278 static void copy_seccomp(struct task_struct
*p
)
1280 #ifdef CONFIG_SECCOMP
1282 * Must be called with sighand->lock held, which is common to
1283 * all threads in the group. Holding cred_guard_mutex is not
1284 * needed because this new task is not yet running and cannot
1287 assert_spin_locked(¤t
->sighand
->siglock
);
1289 /* Ref-count the new filter user, and assign it. */
1290 get_seccomp_filter(current
);
1291 p
->seccomp
= current
->seccomp
;
1294 * Explicitly enable no_new_privs here in case it got set
1295 * between the task_struct being duplicated and holding the
1296 * sighand lock. The seccomp state and nnp must be in sync.
1298 if (task_no_new_privs(current
))
1299 task_set_no_new_privs(p
);
1302 * If the parent gained a seccomp mode after copying thread
1303 * flags and between before we held the sighand lock, we have
1304 * to manually enable the seccomp thread flag here.
1306 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1307 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1311 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1313 current
->clear_child_tid
= tidptr
;
1315 return task_pid_vnr(current
);
1318 static void rt_mutex_init_task(struct task_struct
*p
)
1320 raw_spin_lock_init(&p
->pi_lock
);
1321 #ifdef CONFIG_RT_MUTEXES
1322 p
->pi_waiters
= RB_ROOT
;
1323 p
->pi_waiters_leftmost
= NULL
;
1324 p
->pi_blocked_on
= NULL
;
1329 * Initialize POSIX timer handling for a single task.
1331 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1333 tsk
->cputime_expires
.prof_exp
= 0;
1334 tsk
->cputime_expires
.virt_exp
= 0;
1335 tsk
->cputime_expires
.sched_exp
= 0;
1336 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1337 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1338 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1342 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1344 task
->pids
[type
].pid
= pid
;
1348 * This creates a new process as a copy of the old one,
1349 * but does not actually start it yet.
1351 * It copies the registers, and all the appropriate
1352 * parts of the process environment (as per the clone
1353 * flags). The actual kick-off is left to the caller.
1355 static struct task_struct
*copy_process(unsigned long clone_flags
,
1356 unsigned long stack_start
,
1357 unsigned long stack_size
,
1358 int __user
*child_tidptr
,
1365 struct task_struct
*p
;
1367 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1368 return ERR_PTR(-EINVAL
);
1370 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1371 return ERR_PTR(-EINVAL
);
1374 * Thread groups must share signals as well, and detached threads
1375 * can only be started up within the thread group.
1377 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1378 return ERR_PTR(-EINVAL
);
1381 * Shared signal handlers imply shared VM. By way of the above,
1382 * thread groups also imply shared VM. Blocking this case allows
1383 * for various simplifications in other code.
1385 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1386 return ERR_PTR(-EINVAL
);
1389 * Siblings of global init remain as zombies on exit since they are
1390 * not reaped by their parent (swapper). To solve this and to avoid
1391 * multi-rooted process trees, prevent global and container-inits
1392 * from creating siblings.
1394 if ((clone_flags
& CLONE_PARENT
) &&
1395 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1396 return ERR_PTR(-EINVAL
);
1399 * If the new process will be in a different pid or user namespace
1400 * do not allow it to share a thread group with the forking task.
1402 if (clone_flags
& CLONE_THREAD
) {
1403 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1404 (task_active_pid_ns(current
) !=
1405 current
->nsproxy
->pid_ns_for_children
))
1406 return ERR_PTR(-EINVAL
);
1409 retval
= security_task_create(clone_flags
);
1414 p
= dup_task_struct(current
, node
);
1418 ftrace_graph_init_task(p
);
1420 rt_mutex_init_task(p
);
1422 #ifdef CONFIG_PROVE_LOCKING
1423 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1424 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1427 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1428 task_rlimit(p
, RLIMIT_NPROC
)) {
1429 if (p
->real_cred
->user
!= INIT_USER
&&
1430 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1433 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1435 retval
= copy_creds(p
, clone_flags
);
1440 * If multiple threads are within copy_process(), then this check
1441 * triggers too late. This doesn't hurt, the check is only there
1442 * to stop root fork bombs.
1445 if (nr_threads
>= max_threads
)
1446 goto bad_fork_cleanup_count
;
1448 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1449 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
);
1450 p
->flags
|= PF_FORKNOEXEC
;
1451 INIT_LIST_HEAD(&p
->children
);
1452 INIT_LIST_HEAD(&p
->sibling
);
1453 rcu_copy_process(p
);
1454 p
->vfork_done
= NULL
;
1455 spin_lock_init(&p
->alloc_lock
);
1457 init_sigpending(&p
->pending
);
1459 p
->utime
= p
->stime
= p
->gtime
= 0;
1460 p
->utimescaled
= p
->stimescaled
= 0;
1461 prev_cputime_init(&p
->prev_cputime
);
1463 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1464 seqcount_init(&p
->vtime_seqcount
);
1466 p
->vtime_snap_whence
= VTIME_INACTIVE
;
1469 #if defined(SPLIT_RSS_COUNTING)
1470 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1473 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1475 task_io_accounting_init(&p
->ioac
);
1476 acct_clear_integrals(p
);
1478 posix_cpu_timers_init(p
);
1480 p
->start_time
= ktime_get_ns();
1481 p
->real_start_time
= ktime_get_boot_ns();
1482 p
->io_context
= NULL
;
1483 p
->audit_context
= NULL
;
1486 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1487 if (IS_ERR(p
->mempolicy
)) {
1488 retval
= PTR_ERR(p
->mempolicy
);
1489 p
->mempolicy
= NULL
;
1490 goto bad_fork_cleanup_threadgroup_lock
;
1493 #ifdef CONFIG_CPUSETS
1494 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1495 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1496 seqcount_init(&p
->mems_allowed_seq
);
1498 #ifdef CONFIG_TRACE_IRQFLAGS
1500 p
->hardirqs_enabled
= 0;
1501 p
->hardirq_enable_ip
= 0;
1502 p
->hardirq_enable_event
= 0;
1503 p
->hardirq_disable_ip
= _THIS_IP_
;
1504 p
->hardirq_disable_event
= 0;
1505 p
->softirqs_enabled
= 1;
1506 p
->softirq_enable_ip
= _THIS_IP_
;
1507 p
->softirq_enable_event
= 0;
1508 p
->softirq_disable_ip
= 0;
1509 p
->softirq_disable_event
= 0;
1510 p
->hardirq_context
= 0;
1511 p
->softirq_context
= 0;
1514 p
->pagefault_disabled
= 0;
1516 #ifdef CONFIG_LOCKDEP
1517 p
->lockdep_depth
= 0; /* no locks held yet */
1518 p
->curr_chain_key
= 0;
1519 p
->lockdep_recursion
= 0;
1522 #ifdef CONFIG_DEBUG_MUTEXES
1523 p
->blocked_on
= NULL
; /* not blocked yet */
1525 #ifdef CONFIG_BCACHE
1526 p
->sequential_io
= 0;
1527 p
->sequential_io_avg
= 0;
1530 /* Perform scheduler related setup. Assign this task to a CPU. */
1531 retval
= sched_fork(clone_flags
, p
);
1533 goto bad_fork_cleanup_policy
;
1535 retval
= perf_event_init_task(p
);
1537 goto bad_fork_cleanup_policy
;
1538 retval
= audit_alloc(p
);
1540 goto bad_fork_cleanup_perf
;
1541 /* copy all the process information */
1543 retval
= copy_semundo(clone_flags
, p
);
1545 goto bad_fork_cleanup_audit
;
1546 retval
= copy_files(clone_flags
, p
);
1548 goto bad_fork_cleanup_semundo
;
1549 retval
= copy_fs(clone_flags
, p
);
1551 goto bad_fork_cleanup_files
;
1552 retval
= copy_sighand(clone_flags
, p
);
1554 goto bad_fork_cleanup_fs
;
1555 retval
= copy_signal(clone_flags
, p
);
1557 goto bad_fork_cleanup_sighand
;
1558 retval
= copy_mm(clone_flags
, p
);
1560 goto bad_fork_cleanup_signal
;
1561 retval
= copy_namespaces(clone_flags
, p
);
1563 goto bad_fork_cleanup_mm
;
1564 retval
= copy_io(clone_flags
, p
);
1566 goto bad_fork_cleanup_namespaces
;
1567 retval
= copy_thread_tls(clone_flags
, stack_start
, stack_size
, p
, tls
);
1569 goto bad_fork_cleanup_io
;
1571 if (pid
!= &init_struct_pid
) {
1572 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
1574 retval
= PTR_ERR(pid
);
1575 goto bad_fork_cleanup_thread
;
1579 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
1581 * Clear TID on mm_release()?
1583 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
1588 p
->robust_list
= NULL
;
1589 #ifdef CONFIG_COMPAT
1590 p
->compat_robust_list
= NULL
;
1592 INIT_LIST_HEAD(&p
->pi_state_list
);
1593 p
->pi_state_cache
= NULL
;
1596 * sigaltstack should be cleared when sharing the same VM
1598 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
1602 * Syscall tracing and stepping should be turned off in the
1603 * child regardless of CLONE_PTRACE.
1605 user_disable_single_step(p
);
1606 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
1607 #ifdef TIF_SYSCALL_EMU
1608 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1610 clear_all_latency_tracing(p
);
1612 /* ok, now we should be set up.. */
1613 p
->pid
= pid_nr(pid
);
1614 if (clone_flags
& CLONE_THREAD
) {
1615 p
->exit_signal
= -1;
1616 p
->group_leader
= current
->group_leader
;
1617 p
->tgid
= current
->tgid
;
1619 if (clone_flags
& CLONE_PARENT
)
1620 p
->exit_signal
= current
->group_leader
->exit_signal
;
1622 p
->exit_signal
= (clone_flags
& CSIGNAL
);
1623 p
->group_leader
= p
;
1628 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
1629 p
->dirty_paused_when
= 0;
1631 p
->pdeath_signal
= 0;
1632 INIT_LIST_HEAD(&p
->thread_group
);
1633 p
->task_works
= NULL
;
1635 threadgroup_change_begin(current
);
1637 * Ensure that the cgroup subsystem policies allow the new process to be
1638 * forked. It should be noted the the new process's css_set can be changed
1639 * between here and cgroup_post_fork() if an organisation operation is in
1642 retval
= cgroup_can_fork(p
);
1644 goto bad_fork_free_pid
;
1647 * Make it visible to the rest of the system, but dont wake it up yet.
1648 * Need tasklist lock for parent etc handling!
1650 write_lock_irq(&tasklist_lock
);
1652 /* CLONE_PARENT re-uses the old parent */
1653 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
1654 p
->real_parent
= current
->real_parent
;
1655 p
->parent_exec_id
= current
->parent_exec_id
;
1657 p
->real_parent
= current
;
1658 p
->parent_exec_id
= current
->self_exec_id
;
1661 spin_lock(¤t
->sighand
->siglock
);
1664 * Copy seccomp details explicitly here, in case they were changed
1665 * before holding sighand lock.
1670 * Process group and session signals need to be delivered to just the
1671 * parent before the fork or both the parent and the child after the
1672 * fork. Restart if a signal comes in before we add the new process to
1673 * it's process group.
1674 * A fatal signal pending means that current will exit, so the new
1675 * thread can't slip out of an OOM kill (or normal SIGKILL).
1677 recalc_sigpending();
1678 if (signal_pending(current
)) {
1679 spin_unlock(¤t
->sighand
->siglock
);
1680 write_unlock_irq(&tasklist_lock
);
1681 retval
= -ERESTARTNOINTR
;
1682 goto bad_fork_cancel_cgroup
;
1685 if (likely(p
->pid
)) {
1686 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
1688 init_task_pid(p
, PIDTYPE_PID
, pid
);
1689 if (thread_group_leader(p
)) {
1690 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
1691 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
1693 if (is_child_reaper(pid
)) {
1694 ns_of_pid(pid
)->child_reaper
= p
;
1695 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
1698 p
->signal
->leader_pid
= pid
;
1699 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
1700 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
1701 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
1702 attach_pid(p
, PIDTYPE_PGID
);
1703 attach_pid(p
, PIDTYPE_SID
);
1704 __this_cpu_inc(process_counts
);
1706 current
->signal
->nr_threads
++;
1707 atomic_inc(¤t
->signal
->live
);
1708 atomic_inc(¤t
->signal
->sigcnt
);
1709 list_add_tail_rcu(&p
->thread_group
,
1710 &p
->group_leader
->thread_group
);
1711 list_add_tail_rcu(&p
->thread_node
,
1712 &p
->signal
->thread_head
);
1714 attach_pid(p
, PIDTYPE_PID
);
1719 spin_unlock(¤t
->sighand
->siglock
);
1720 syscall_tracepoint_update(p
);
1721 write_unlock_irq(&tasklist_lock
);
1723 proc_fork_connector(p
);
1724 cgroup_post_fork(p
);
1725 threadgroup_change_end(current
);
1728 trace_task_newtask(p
, clone_flags
);
1729 uprobe_copy_process(p
, clone_flags
);
1733 bad_fork_cancel_cgroup
:
1734 cgroup_cancel_fork(p
);
1736 threadgroup_change_end(current
);
1737 if (pid
!= &init_struct_pid
)
1739 bad_fork_cleanup_thread
:
1741 bad_fork_cleanup_io
:
1744 bad_fork_cleanup_namespaces
:
1745 exit_task_namespaces(p
);
1746 bad_fork_cleanup_mm
:
1749 bad_fork_cleanup_signal
:
1750 if (!(clone_flags
& CLONE_THREAD
))
1751 free_signal_struct(p
->signal
);
1752 bad_fork_cleanup_sighand
:
1753 __cleanup_sighand(p
->sighand
);
1754 bad_fork_cleanup_fs
:
1755 exit_fs(p
); /* blocking */
1756 bad_fork_cleanup_files
:
1757 exit_files(p
); /* blocking */
1758 bad_fork_cleanup_semundo
:
1760 bad_fork_cleanup_audit
:
1762 bad_fork_cleanup_perf
:
1763 perf_event_free_task(p
);
1764 bad_fork_cleanup_policy
:
1766 mpol_put(p
->mempolicy
);
1767 bad_fork_cleanup_threadgroup_lock
:
1769 delayacct_tsk_free(p
);
1770 bad_fork_cleanup_count
:
1771 atomic_dec(&p
->cred
->user
->processes
);
1776 return ERR_PTR(retval
);
1779 static inline void init_idle_pids(struct pid_link
*links
)
1783 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1784 INIT_HLIST_NODE(&links
[type
].node
); /* not really needed */
1785 links
[type
].pid
= &init_struct_pid
;
1789 struct task_struct
*fork_idle(int cpu
)
1791 struct task_struct
*task
;
1792 task
= copy_process(CLONE_VM
, 0, 0, NULL
, &init_struct_pid
, 0, 0,
1794 if (!IS_ERR(task
)) {
1795 init_idle_pids(task
->pids
);
1796 init_idle(task
, cpu
);
1803 * Ok, this is the main fork-routine.
1805 * It copies the process, and if successful kick-starts
1806 * it and waits for it to finish using the VM if required.
1808 long _do_fork(unsigned long clone_flags
,
1809 unsigned long stack_start
,
1810 unsigned long stack_size
,
1811 int __user
*parent_tidptr
,
1812 int __user
*child_tidptr
,
1815 struct task_struct
*p
;
1820 * Determine whether and which event to report to ptracer. When
1821 * called from kernel_thread or CLONE_UNTRACED is explicitly
1822 * requested, no event is reported; otherwise, report if the event
1823 * for the type of forking is enabled.
1825 if (!(clone_flags
& CLONE_UNTRACED
)) {
1826 if (clone_flags
& CLONE_VFORK
)
1827 trace
= PTRACE_EVENT_VFORK
;
1828 else if ((clone_flags
& CSIGNAL
) != SIGCHLD
)
1829 trace
= PTRACE_EVENT_CLONE
;
1831 trace
= PTRACE_EVENT_FORK
;
1833 if (likely(!ptrace_event_enabled(current
, trace
)))
1837 p
= copy_process(clone_flags
, stack_start
, stack_size
,
1838 child_tidptr
, NULL
, trace
, tls
, NUMA_NO_NODE
);
1840 * Do this prior waking up the new thread - the thread pointer
1841 * might get invalid after that point, if the thread exits quickly.
1844 struct completion vfork
;
1847 trace_sched_process_fork(current
, p
);
1849 pid
= get_task_pid(p
, PIDTYPE_PID
);
1852 if (clone_flags
& CLONE_PARENT_SETTID
)
1853 put_user(nr
, parent_tidptr
);
1855 if (clone_flags
& CLONE_VFORK
) {
1856 p
->vfork_done
= &vfork
;
1857 init_completion(&vfork
);
1861 wake_up_new_task(p
);
1863 /* forking complete and child started to run, tell ptracer */
1864 if (unlikely(trace
))
1865 ptrace_event_pid(trace
, pid
);
1867 if (clone_flags
& CLONE_VFORK
) {
1868 if (!wait_for_vfork_done(p
, &vfork
))
1869 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
1879 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1880 /* For compatibility with architectures that call do_fork directly rather than
1881 * using the syscall entry points below. */
1882 long do_fork(unsigned long clone_flags
,
1883 unsigned long stack_start
,
1884 unsigned long stack_size
,
1885 int __user
*parent_tidptr
,
1886 int __user
*child_tidptr
)
1888 return _do_fork(clone_flags
, stack_start
, stack_size
,
1889 parent_tidptr
, child_tidptr
, 0);
1894 * Create a kernel thread.
1896 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
1898 return _do_fork(flags
|CLONE_VM
|CLONE_UNTRACED
, (unsigned long)fn
,
1899 (unsigned long)arg
, NULL
, NULL
, 0);
1902 #ifdef __ARCH_WANT_SYS_FORK
1903 SYSCALL_DEFINE0(fork
)
1906 return _do_fork(SIGCHLD
, 0, 0, NULL
, NULL
, 0);
1908 /* can not support in nommu mode */
1914 #ifdef __ARCH_WANT_SYS_VFORK
1915 SYSCALL_DEFINE0(vfork
)
1917 return _do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, 0,
1922 #ifdef __ARCH_WANT_SYS_CLONE
1923 #ifdef CONFIG_CLONE_BACKWARDS
1924 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1925 int __user
*, parent_tidptr
,
1927 int __user
*, child_tidptr
)
1928 #elif defined(CONFIG_CLONE_BACKWARDS2)
1929 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
1930 int __user
*, parent_tidptr
,
1931 int __user
*, child_tidptr
,
1933 #elif defined(CONFIG_CLONE_BACKWARDS3)
1934 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1936 int __user
*, parent_tidptr
,
1937 int __user
*, child_tidptr
,
1940 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
1941 int __user
*, parent_tidptr
,
1942 int __user
*, child_tidptr
,
1946 return _do_fork(clone_flags
, newsp
, 0, parent_tidptr
, child_tidptr
, tls
);
1950 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1951 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1954 static void sighand_ctor(void *data
)
1956 struct sighand_struct
*sighand
= data
;
1958 spin_lock_init(&sighand
->siglock
);
1959 init_waitqueue_head(&sighand
->signalfd_wqh
);
1962 void __init
proc_caches_init(void)
1964 sighand_cachep
= kmem_cache_create("sighand_cache",
1965 sizeof(struct sighand_struct
), 0,
1966 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_DESTROY_BY_RCU
|
1967 SLAB_NOTRACK
|SLAB_ACCOUNT
, sighand_ctor
);
1968 signal_cachep
= kmem_cache_create("signal_cache",
1969 sizeof(struct signal_struct
), 0,
1970 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1972 files_cachep
= kmem_cache_create("files_cache",
1973 sizeof(struct files_struct
), 0,
1974 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1976 fs_cachep
= kmem_cache_create("fs_cache",
1977 sizeof(struct fs_struct
), 0,
1978 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1981 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1982 * whole struct cpumask for the OFFSTACK case. We could change
1983 * this to *only* allocate as much of it as required by the
1984 * maximum number of CPU's we can ever have. The cpumask_allocation
1985 * is at the end of the structure, exactly for that reason.
1987 mm_cachep
= kmem_cache_create("mm_struct",
1988 sizeof(struct mm_struct
), ARCH_MIN_MMSTRUCT_ALIGN
,
1989 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
1991 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
1993 nsproxy_cache_init();
1997 * Check constraints on flags passed to the unshare system call.
1999 static int check_unshare_flags(unsigned long unshare_flags
)
2001 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2002 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2003 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2004 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2007 * Not implemented, but pretend it works if there is nothing
2008 * to unshare. Note that unsharing the address space or the
2009 * signal handlers also need to unshare the signal queues (aka
2012 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2013 if (!thread_group_empty(current
))
2016 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2017 if (atomic_read(¤t
->sighand
->count
) > 1)
2020 if (unshare_flags
& CLONE_VM
) {
2021 if (!current_is_single_threaded())
2029 * Unshare the filesystem structure if it is being shared
2031 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2033 struct fs_struct
*fs
= current
->fs
;
2035 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2038 /* don't need lock here; in the worst case we'll do useless copy */
2042 *new_fsp
= copy_fs_struct(fs
);
2050 * Unshare file descriptor table if it is being shared
2052 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2054 struct files_struct
*fd
= current
->files
;
2057 if ((unshare_flags
& CLONE_FILES
) &&
2058 (fd
&& atomic_read(&fd
->count
) > 1)) {
2059 *new_fdp
= dup_fd(fd
, &error
);
2068 * unshare allows a process to 'unshare' part of the process
2069 * context which was originally shared using clone. copy_*
2070 * functions used by do_fork() cannot be used here directly
2071 * because they modify an inactive task_struct that is being
2072 * constructed. Here we are modifying the current, active,
2075 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2077 struct fs_struct
*fs
, *new_fs
= NULL
;
2078 struct files_struct
*fd
, *new_fd
= NULL
;
2079 struct cred
*new_cred
= NULL
;
2080 struct nsproxy
*new_nsproxy
= NULL
;
2085 * If unsharing a user namespace must also unshare the thread group
2086 * and unshare the filesystem root and working directories.
2088 if (unshare_flags
& CLONE_NEWUSER
)
2089 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2091 * If unsharing vm, must also unshare signal handlers.
2093 if (unshare_flags
& CLONE_VM
)
2094 unshare_flags
|= CLONE_SIGHAND
;
2096 * If unsharing a signal handlers, must also unshare the signal queues.
2098 if (unshare_flags
& CLONE_SIGHAND
)
2099 unshare_flags
|= CLONE_THREAD
;
2101 * If unsharing namespace, must also unshare filesystem information.
2103 if (unshare_flags
& CLONE_NEWNS
)
2104 unshare_flags
|= CLONE_FS
;
2106 err
= check_unshare_flags(unshare_flags
);
2108 goto bad_unshare_out
;
2110 * CLONE_NEWIPC must also detach from the undolist: after switching
2111 * to a new ipc namespace, the semaphore arrays from the old
2112 * namespace are unreachable.
2114 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2116 err
= unshare_fs(unshare_flags
, &new_fs
);
2118 goto bad_unshare_out
;
2119 err
= unshare_fd(unshare_flags
, &new_fd
);
2121 goto bad_unshare_cleanup_fs
;
2122 err
= unshare_userns(unshare_flags
, &new_cred
);
2124 goto bad_unshare_cleanup_fd
;
2125 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2128 goto bad_unshare_cleanup_cred
;
2130 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2133 * CLONE_SYSVSEM is equivalent to sys_exit().
2137 if (unshare_flags
& CLONE_NEWIPC
) {
2138 /* Orphan segments in old ns (see sem above). */
2140 shm_init_task(current
);
2144 switch_task_namespaces(current
, new_nsproxy
);
2150 spin_lock(&fs
->lock
);
2151 current
->fs
= new_fs
;
2156 spin_unlock(&fs
->lock
);
2160 fd
= current
->files
;
2161 current
->files
= new_fd
;
2165 task_unlock(current
);
2168 /* Install the new user namespace */
2169 commit_creds(new_cred
);
2174 bad_unshare_cleanup_cred
:
2177 bad_unshare_cleanup_fd
:
2179 put_files_struct(new_fd
);
2181 bad_unshare_cleanup_fs
:
2183 free_fs_struct(new_fs
);
2190 * Helper to unshare the files of the current task.
2191 * We don't want to expose copy_files internals to
2192 * the exec layer of the kernel.
2195 int unshare_files(struct files_struct
**displaced
)
2197 struct task_struct
*task
= current
;
2198 struct files_struct
*copy
= NULL
;
2201 error
= unshare_fd(CLONE_FILES
, ©
);
2202 if (error
|| !copy
) {
2206 *displaced
= task
->files
;
2213 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2214 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2218 int threads
= max_threads
;
2219 int min
= MIN_THREADS
;
2220 int max
= MAX_THREADS
;
2227 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2231 set_max_threads(threads
);