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 <linux/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)
163 #ifdef CONFIG_VMAP_STACK
165 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
166 * flush. Try to minimize the number of calls by caching stacks.
168 #define NR_CACHED_STACKS 2
169 static DEFINE_PER_CPU(struct vm_struct
*, cached_stacks
[NR_CACHED_STACKS
]);
172 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
, int node
)
174 #ifdef CONFIG_VMAP_STACK
179 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
180 struct vm_struct
*s
= this_cpu_read(cached_stacks
[i
]);
184 this_cpu_write(cached_stacks
[i
], NULL
);
186 tsk
->stack_vm_area
= s
;
192 stack
= __vmalloc_node_range(THREAD_SIZE
, THREAD_SIZE
,
193 VMALLOC_START
, VMALLOC_END
,
194 THREADINFO_GFP
| __GFP_HIGHMEM
,
196 0, node
, __builtin_return_address(0));
199 * We can't call find_vm_area() in interrupt context, and
200 * free_thread_stack() can be called in interrupt context,
201 * so cache the vm_struct.
204 tsk
->stack_vm_area
= find_vm_area(stack
);
207 struct page
*page
= alloc_pages_node(node
, THREADINFO_GFP
,
210 return page
? page_address(page
) : NULL
;
214 static inline void free_thread_stack(struct task_struct
*tsk
)
216 #ifdef CONFIG_VMAP_STACK
217 if (task_stack_vm_area(tsk
)) {
221 local_irq_save(flags
);
222 for (i
= 0; i
< NR_CACHED_STACKS
; i
++) {
223 if (this_cpu_read(cached_stacks
[i
]))
226 this_cpu_write(cached_stacks
[i
], tsk
->stack_vm_area
);
227 local_irq_restore(flags
);
230 local_irq_restore(flags
);
232 vfree_atomic(tsk
->stack
);
237 __free_pages(virt_to_page(tsk
->stack
), THREAD_SIZE_ORDER
);
240 static struct kmem_cache
*thread_stack_cache
;
242 static unsigned long *alloc_thread_stack_node(struct task_struct
*tsk
,
245 return kmem_cache_alloc_node(thread_stack_cache
, THREADINFO_GFP
, node
);
248 static void free_thread_stack(struct task_struct
*tsk
)
250 kmem_cache_free(thread_stack_cache
, tsk
->stack
);
253 void thread_stack_cache_init(void)
255 thread_stack_cache
= kmem_cache_create("thread_stack", THREAD_SIZE
,
256 THREAD_SIZE
, 0, NULL
);
257 BUG_ON(thread_stack_cache
== NULL
);
262 /* SLAB cache for signal_struct structures (tsk->signal) */
263 static struct kmem_cache
*signal_cachep
;
265 /* SLAB cache for sighand_struct structures (tsk->sighand) */
266 struct kmem_cache
*sighand_cachep
;
268 /* SLAB cache for files_struct structures (tsk->files) */
269 struct kmem_cache
*files_cachep
;
271 /* SLAB cache for fs_struct structures (tsk->fs) */
272 struct kmem_cache
*fs_cachep
;
274 /* SLAB cache for vm_area_struct structures */
275 struct kmem_cache
*vm_area_cachep
;
277 /* SLAB cache for mm_struct structures (tsk->mm) */
278 static struct kmem_cache
*mm_cachep
;
280 static void account_kernel_stack(struct task_struct
*tsk
, int account
)
282 void *stack
= task_stack_page(tsk
);
283 struct vm_struct
*vm
= task_stack_vm_area(tsk
);
285 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK
) && PAGE_SIZE
% 1024 != 0);
290 BUG_ON(vm
->nr_pages
!= THREAD_SIZE
/ PAGE_SIZE
);
292 for (i
= 0; i
< THREAD_SIZE
/ PAGE_SIZE
; i
++) {
293 mod_zone_page_state(page_zone(vm
->pages
[i
]),
295 PAGE_SIZE
/ 1024 * account
);
298 /* All stack pages belong to the same memcg. */
299 memcg_kmem_update_page_stat(vm
->pages
[0], MEMCG_KERNEL_STACK_KB
,
300 account
* (THREAD_SIZE
/ 1024));
303 * All stack pages are in the same zone and belong to the
306 struct page
*first_page
= virt_to_page(stack
);
308 mod_zone_page_state(page_zone(first_page
), NR_KERNEL_STACK_KB
,
309 THREAD_SIZE
/ 1024 * account
);
311 memcg_kmem_update_page_stat(first_page
, MEMCG_KERNEL_STACK_KB
,
312 account
* (THREAD_SIZE
/ 1024));
316 static void release_task_stack(struct task_struct
*tsk
)
318 if (WARN_ON(tsk
->state
!= TASK_DEAD
))
319 return; /* Better to leak the stack than to free prematurely */
321 account_kernel_stack(tsk
, -1);
322 arch_release_thread_stack(tsk
->stack
);
323 free_thread_stack(tsk
);
325 #ifdef CONFIG_VMAP_STACK
326 tsk
->stack_vm_area
= NULL
;
330 #ifdef CONFIG_THREAD_INFO_IN_TASK
331 void put_task_stack(struct task_struct
*tsk
)
333 if (atomic_dec_and_test(&tsk
->stack_refcount
))
334 release_task_stack(tsk
);
338 void free_task(struct task_struct
*tsk
)
340 #ifndef CONFIG_THREAD_INFO_IN_TASK
342 * The task is finally done with both the stack and thread_info,
345 release_task_stack(tsk
);
348 * If the task had a separate stack allocation, it should be gone
351 WARN_ON_ONCE(atomic_read(&tsk
->stack_refcount
) != 0);
353 rt_mutex_debug_task_free(tsk
);
354 ftrace_graph_exit_task(tsk
);
355 put_seccomp_filter(tsk
);
356 arch_release_task_struct(tsk
);
357 if (tsk
->flags
& PF_KTHREAD
)
358 free_kthread_struct(tsk
);
359 free_task_struct(tsk
);
361 EXPORT_SYMBOL(free_task
);
363 static inline void free_signal_struct(struct signal_struct
*sig
)
365 taskstats_tgid_free(sig
);
366 sched_autogroup_exit(sig
);
368 * __mmdrop is not safe to call from softirq context on x86 due to
369 * pgd_dtor so postpone it to the async context
372 mmdrop_async(sig
->oom_mm
);
373 kmem_cache_free(signal_cachep
, sig
);
376 static inline void put_signal_struct(struct signal_struct
*sig
)
378 if (atomic_dec_and_test(&sig
->sigcnt
))
379 free_signal_struct(sig
);
382 void __put_task_struct(struct task_struct
*tsk
)
384 WARN_ON(!tsk
->exit_state
);
385 WARN_ON(atomic_read(&tsk
->usage
));
386 WARN_ON(tsk
== current
);
390 security_task_free(tsk
);
392 delayacct_tsk_free(tsk
);
393 put_signal_struct(tsk
->signal
);
395 if (!profile_handoff_task(tsk
))
398 EXPORT_SYMBOL_GPL(__put_task_struct
);
400 void __init __weak
arch_task_cache_init(void) { }
405 static void set_max_threads(unsigned int max_threads_suggested
)
410 * The number of threads shall be limited such that the thread
411 * structures may only consume a small part of the available memory.
413 if (fls64(totalram_pages
) + fls64(PAGE_SIZE
) > 64)
414 threads
= MAX_THREADS
;
416 threads
= div64_u64((u64
) totalram_pages
* (u64
) PAGE_SIZE
,
417 (u64
) THREAD_SIZE
* 8UL);
419 if (threads
> max_threads_suggested
)
420 threads
= max_threads_suggested
;
422 max_threads
= clamp_t(u64
, threads
, MIN_THREADS
, MAX_THREADS
);
425 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
426 /* Initialized by the architecture: */
427 int arch_task_struct_size __read_mostly
;
430 void __init
fork_init(void)
433 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
434 #ifndef ARCH_MIN_TASKALIGN
435 #define ARCH_MIN_TASKALIGN 0
437 int align
= max_t(int, L1_CACHE_BYTES
, ARCH_MIN_TASKALIGN
);
439 /* create a slab on which task_structs can be allocated */
440 task_struct_cachep
= kmem_cache_create("task_struct",
441 arch_task_struct_size
, align
,
442 SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
, NULL
);
445 /* do the arch specific task caches init */
446 arch_task_cache_init();
448 set_max_threads(MAX_THREADS
);
450 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_cur
= max_threads
/2;
451 init_task
.signal
->rlim
[RLIMIT_NPROC
].rlim_max
= max_threads
/2;
452 init_task
.signal
->rlim
[RLIMIT_SIGPENDING
] =
453 init_task
.signal
->rlim
[RLIMIT_NPROC
];
455 for (i
= 0; i
< UCOUNT_COUNTS
; i
++) {
456 init_user_ns
.ucount_max
[i
] = max_threads
/2;
460 int __weak
arch_dup_task_struct(struct task_struct
*dst
,
461 struct task_struct
*src
)
467 void set_task_stack_end_magic(struct task_struct
*tsk
)
469 unsigned long *stackend
;
471 stackend
= end_of_stack(tsk
);
472 *stackend
= STACK_END_MAGIC
; /* for overflow detection */
475 static struct task_struct
*dup_task_struct(struct task_struct
*orig
, int node
)
477 struct task_struct
*tsk
;
478 unsigned long *stack
;
479 struct vm_struct
*stack_vm_area
;
482 if (node
== NUMA_NO_NODE
)
483 node
= tsk_fork_get_node(orig
);
484 tsk
= alloc_task_struct_node(node
);
488 stack
= alloc_thread_stack_node(tsk
, node
);
492 stack_vm_area
= task_stack_vm_area(tsk
);
494 err
= arch_dup_task_struct(tsk
, orig
);
497 * arch_dup_task_struct() clobbers the stack-related fields. Make
498 * sure they're properly initialized before using any stack-related
502 #ifdef CONFIG_VMAP_STACK
503 tsk
->stack_vm_area
= stack_vm_area
;
505 #ifdef CONFIG_THREAD_INFO_IN_TASK
506 atomic_set(&tsk
->stack_refcount
, 1);
512 #ifdef CONFIG_SECCOMP
514 * We must handle setting up seccomp filters once we're under
515 * the sighand lock in case orig has changed between now and
516 * then. Until then, filter must be NULL to avoid messing up
517 * the usage counts on the error path calling free_task.
519 tsk
->seccomp
.filter
= NULL
;
522 setup_thread_stack(tsk
, orig
);
523 clear_user_return_notifier(tsk
);
524 clear_tsk_need_resched(tsk
);
525 set_task_stack_end_magic(tsk
);
527 #ifdef CONFIG_CC_STACKPROTECTOR
528 tsk
->stack_canary
= get_random_int();
532 * One for us, one for whoever does the "release_task()" (usually
535 atomic_set(&tsk
->usage
, 2);
536 #ifdef CONFIG_BLK_DEV_IO_TRACE
539 tsk
->splice_pipe
= NULL
;
540 tsk
->task_frag
.page
= NULL
;
541 tsk
->wake_q
.next
= NULL
;
543 account_kernel_stack(tsk
, 1);
550 free_thread_stack(tsk
);
552 free_task_struct(tsk
);
557 static __latent_entropy
int dup_mmap(struct mm_struct
*mm
,
558 struct mm_struct
*oldmm
)
560 struct vm_area_struct
*mpnt
, *tmp
, *prev
, **pprev
;
561 struct rb_node
**rb_link
, *rb_parent
;
563 unsigned long charge
;
565 uprobe_start_dup_mmap();
566 if (down_write_killable(&oldmm
->mmap_sem
)) {
568 goto fail_uprobe_end
;
570 flush_cache_dup_mm(oldmm
);
571 uprobe_dup_mmap(oldmm
, mm
);
573 * Not linked in yet - no deadlock potential:
575 down_write_nested(&mm
->mmap_sem
, SINGLE_DEPTH_NESTING
);
577 /* No ordering required: file already has been exposed. */
578 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
580 mm
->total_vm
= oldmm
->total_vm
;
581 mm
->data_vm
= oldmm
->data_vm
;
582 mm
->exec_vm
= oldmm
->exec_vm
;
583 mm
->stack_vm
= oldmm
->stack_vm
;
585 rb_link
= &mm
->mm_rb
.rb_node
;
588 retval
= ksm_fork(mm
, oldmm
);
591 retval
= khugepaged_fork(mm
, oldmm
);
596 for (mpnt
= oldmm
->mmap
; mpnt
; mpnt
= mpnt
->vm_next
) {
599 if (mpnt
->vm_flags
& VM_DONTCOPY
) {
600 vm_stat_account(mm
, mpnt
->vm_flags
, -vma_pages(mpnt
));
604 if (mpnt
->vm_flags
& VM_ACCOUNT
) {
605 unsigned long len
= vma_pages(mpnt
);
607 if (security_vm_enough_memory_mm(oldmm
, len
)) /* sic */
611 tmp
= kmem_cache_alloc(vm_area_cachep
, GFP_KERNEL
);
615 INIT_LIST_HEAD(&tmp
->anon_vma_chain
);
616 retval
= vma_dup_policy(mpnt
, tmp
);
618 goto fail_nomem_policy
;
620 if (anon_vma_fork(tmp
, mpnt
))
621 goto fail_nomem_anon_vma_fork
;
623 ~(VM_LOCKED
|VM_LOCKONFAULT
|VM_UFFD_MISSING
|VM_UFFD_WP
);
624 tmp
->vm_next
= tmp
->vm_prev
= NULL
;
625 tmp
->vm_userfaultfd_ctx
= NULL_VM_UFFD_CTX
;
628 struct inode
*inode
= file_inode(file
);
629 struct address_space
*mapping
= file
->f_mapping
;
632 if (tmp
->vm_flags
& VM_DENYWRITE
)
633 atomic_dec(&inode
->i_writecount
);
634 i_mmap_lock_write(mapping
);
635 if (tmp
->vm_flags
& VM_SHARED
)
636 atomic_inc(&mapping
->i_mmap_writable
);
637 flush_dcache_mmap_lock(mapping
);
638 /* insert tmp into the share list, just after mpnt */
639 vma_interval_tree_insert_after(tmp
, mpnt
,
641 flush_dcache_mmap_unlock(mapping
);
642 i_mmap_unlock_write(mapping
);
646 * Clear hugetlb-related page reserves for children. This only
647 * affects MAP_PRIVATE mappings. Faults generated by the child
648 * are not guaranteed to succeed, even if read-only
650 if (is_vm_hugetlb_page(tmp
))
651 reset_vma_resv_huge_pages(tmp
);
654 * Link in the new vma and copy the page table entries.
657 pprev
= &tmp
->vm_next
;
661 __vma_link_rb(mm
, tmp
, rb_link
, rb_parent
);
662 rb_link
= &tmp
->vm_rb
.rb_right
;
663 rb_parent
= &tmp
->vm_rb
;
666 retval
= copy_page_range(mm
, oldmm
, mpnt
);
668 if (tmp
->vm_ops
&& tmp
->vm_ops
->open
)
669 tmp
->vm_ops
->open(tmp
);
674 /* a new mm has just been created */
675 arch_dup_mmap(oldmm
, mm
);
678 up_write(&mm
->mmap_sem
);
680 up_write(&oldmm
->mmap_sem
);
682 uprobe_end_dup_mmap();
684 fail_nomem_anon_vma_fork
:
685 mpol_put(vma_policy(tmp
));
687 kmem_cache_free(vm_area_cachep
, tmp
);
690 vm_unacct_memory(charge
);
694 static inline int mm_alloc_pgd(struct mm_struct
*mm
)
696 mm
->pgd
= pgd_alloc(mm
);
697 if (unlikely(!mm
->pgd
))
702 static inline void mm_free_pgd(struct mm_struct
*mm
)
704 pgd_free(mm
, mm
->pgd
);
707 static int dup_mmap(struct mm_struct
*mm
, struct mm_struct
*oldmm
)
709 down_write(&oldmm
->mmap_sem
);
710 RCU_INIT_POINTER(mm
->exe_file
, get_mm_exe_file(oldmm
));
711 up_write(&oldmm
->mmap_sem
);
714 #define mm_alloc_pgd(mm) (0)
715 #define mm_free_pgd(mm)
716 #endif /* CONFIG_MMU */
718 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(mmlist_lock
);
720 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
721 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
723 static unsigned long default_dump_filter
= MMF_DUMP_FILTER_DEFAULT
;
725 static int __init
coredump_filter_setup(char *s
)
727 default_dump_filter
=
728 (simple_strtoul(s
, NULL
, 0) << MMF_DUMP_FILTER_SHIFT
) &
729 MMF_DUMP_FILTER_MASK
;
733 __setup("coredump_filter=", coredump_filter_setup
);
735 #include <linux/init_task.h>
737 static void mm_init_aio(struct mm_struct
*mm
)
740 spin_lock_init(&mm
->ioctx_lock
);
741 mm
->ioctx_table
= NULL
;
745 static void mm_init_owner(struct mm_struct
*mm
, struct task_struct
*p
)
752 static struct mm_struct
*mm_init(struct mm_struct
*mm
, struct task_struct
*p
,
753 struct user_namespace
*user_ns
)
757 mm
->vmacache_seqnum
= 0;
758 atomic_set(&mm
->mm_users
, 1);
759 atomic_set(&mm
->mm_count
, 1);
760 init_rwsem(&mm
->mmap_sem
);
761 INIT_LIST_HEAD(&mm
->mmlist
);
762 mm
->core_state
= NULL
;
763 atomic_long_set(&mm
->nr_ptes
, 0);
768 memset(&mm
->rss_stat
, 0, sizeof(mm
->rss_stat
));
769 spin_lock_init(&mm
->page_table_lock
);
772 mm_init_owner(mm
, p
);
773 mmu_notifier_mm_init(mm
);
774 clear_tlb_flush_pending(mm
);
775 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
776 mm
->pmd_huge_pte
= NULL
;
780 mm
->flags
= current
->mm
->flags
& MMF_INIT_MASK
;
781 mm
->def_flags
= current
->mm
->def_flags
& VM_INIT_DEF_MASK
;
783 mm
->flags
= default_dump_filter
;
787 if (mm_alloc_pgd(mm
))
790 if (init_new_context(p
, mm
))
793 mm
->user_ns
= get_user_ns(user_ns
);
803 static void check_mm(struct mm_struct
*mm
)
807 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
808 long x
= atomic_long_read(&mm
->rss_stat
.count
[i
]);
811 printk(KERN_ALERT
"BUG: Bad rss-counter state "
812 "mm:%p idx:%d val:%ld\n", mm
, i
, x
);
815 if (atomic_long_read(&mm
->nr_ptes
))
816 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
817 atomic_long_read(&mm
->nr_ptes
));
819 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
822 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
823 VM_BUG_ON_MM(mm
->pmd_huge_pte
, mm
);
828 * Allocate and initialize an mm_struct.
830 struct mm_struct
*mm_alloc(void)
832 struct mm_struct
*mm
;
838 memset(mm
, 0, sizeof(*mm
));
839 return mm_init(mm
, current
, current_user_ns());
843 * Called when the last reference to the mm
844 * is dropped: either by a lazy thread or by
845 * mmput. Free the page directory and the mm.
847 void __mmdrop(struct mm_struct
*mm
)
849 BUG_ON(mm
== &init_mm
);
852 mmu_notifier_mm_destroy(mm
);
854 put_user_ns(mm
->user_ns
);
857 EXPORT_SYMBOL_GPL(__mmdrop
);
859 static inline void __mmput(struct mm_struct
*mm
)
861 VM_BUG_ON(atomic_read(&mm
->mm_users
));
863 uprobe_clear_state(mm
);
866 khugepaged_exit(mm
); /* must run before exit_mmap */
868 mm_put_huge_zero_page(mm
);
869 set_mm_exe_file(mm
, NULL
);
870 if (!list_empty(&mm
->mmlist
)) {
871 spin_lock(&mmlist_lock
);
872 list_del(&mm
->mmlist
);
873 spin_unlock(&mmlist_lock
);
876 module_put(mm
->binfmt
->module
);
877 set_bit(MMF_OOM_SKIP
, &mm
->flags
);
882 * Decrement the use count and release all resources for an mm.
884 void mmput(struct mm_struct
*mm
)
888 if (atomic_dec_and_test(&mm
->mm_users
))
891 EXPORT_SYMBOL_GPL(mmput
);
894 static void mmput_async_fn(struct work_struct
*work
)
896 struct mm_struct
*mm
= container_of(work
, struct mm_struct
, async_put_work
);
900 void mmput_async(struct mm_struct
*mm
)
902 if (atomic_dec_and_test(&mm
->mm_users
)) {
903 INIT_WORK(&mm
->async_put_work
, mmput_async_fn
);
904 schedule_work(&mm
->async_put_work
);
910 * set_mm_exe_file - change a reference to the mm's executable file
912 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
914 * Main users are mmput() and sys_execve(). Callers prevent concurrent
915 * invocations: in mmput() nobody alive left, in execve task is single
916 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
917 * mm->exe_file, but does so without using set_mm_exe_file() in order
918 * to do avoid the need for any locks.
920 void set_mm_exe_file(struct mm_struct
*mm
, struct file
*new_exe_file
)
922 struct file
*old_exe_file
;
925 * It is safe to dereference the exe_file without RCU as
926 * this function is only called if nobody else can access
927 * this mm -- see comment above for justification.
929 old_exe_file
= rcu_dereference_raw(mm
->exe_file
);
932 get_file(new_exe_file
);
933 rcu_assign_pointer(mm
->exe_file
, new_exe_file
);
939 * get_mm_exe_file - acquire a reference to the mm's executable file
941 * Returns %NULL if mm has no associated executable file.
942 * User must release file via fput().
944 struct file
*get_mm_exe_file(struct mm_struct
*mm
)
946 struct file
*exe_file
;
949 exe_file
= rcu_dereference(mm
->exe_file
);
950 if (exe_file
&& !get_file_rcu(exe_file
))
955 EXPORT_SYMBOL(get_mm_exe_file
);
958 * get_task_exe_file - acquire a reference to the task's executable file
960 * Returns %NULL if task's mm (if any) has no associated executable file or
961 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
962 * User must release file via fput().
964 struct file
*get_task_exe_file(struct task_struct
*task
)
966 struct file
*exe_file
= NULL
;
967 struct mm_struct
*mm
;
972 if (!(task
->flags
& PF_KTHREAD
))
973 exe_file
= get_mm_exe_file(mm
);
978 EXPORT_SYMBOL(get_task_exe_file
);
981 * get_task_mm - acquire a reference to the task's mm
983 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
984 * this kernel workthread has transiently adopted a user mm with use_mm,
985 * to do its AIO) is not set and if so returns a reference to it, after
986 * bumping up the use count. User must release the mm via mmput()
987 * after use. Typically used by /proc and ptrace.
989 struct mm_struct
*get_task_mm(struct task_struct
*task
)
991 struct mm_struct
*mm
;
996 if (task
->flags
& PF_KTHREAD
)
999 atomic_inc(&mm
->mm_users
);
1004 EXPORT_SYMBOL_GPL(get_task_mm
);
1006 struct mm_struct
*mm_access(struct task_struct
*task
, unsigned int mode
)
1008 struct mm_struct
*mm
;
1011 err
= mutex_lock_killable(&task
->signal
->cred_guard_mutex
);
1013 return ERR_PTR(err
);
1015 mm
= get_task_mm(task
);
1016 if (mm
&& mm
!= current
->mm
&&
1017 !ptrace_may_access(task
, mode
)) {
1019 mm
= ERR_PTR(-EACCES
);
1021 mutex_unlock(&task
->signal
->cred_guard_mutex
);
1026 static void complete_vfork_done(struct task_struct
*tsk
)
1028 struct completion
*vfork
;
1031 vfork
= tsk
->vfork_done
;
1032 if (likely(vfork
)) {
1033 tsk
->vfork_done
= NULL
;
1039 static int wait_for_vfork_done(struct task_struct
*child
,
1040 struct completion
*vfork
)
1044 freezer_do_not_count();
1045 killed
= wait_for_completion_killable(vfork
);
1050 child
->vfork_done
= NULL
;
1054 put_task_struct(child
);
1058 /* Please note the differences between mmput and mm_release.
1059 * mmput is called whenever we stop holding onto a mm_struct,
1060 * error success whatever.
1062 * mm_release is called after a mm_struct has been removed
1063 * from the current process.
1065 * This difference is important for error handling, when we
1066 * only half set up a mm_struct for a new process and need to restore
1067 * the old one. Because we mmput the new mm_struct before
1068 * restoring the old one. . .
1069 * Eric Biederman 10 January 1998
1071 void mm_release(struct task_struct
*tsk
, struct mm_struct
*mm
)
1073 /* Get rid of any futexes when releasing the mm */
1075 if (unlikely(tsk
->robust_list
)) {
1076 exit_robust_list(tsk
);
1077 tsk
->robust_list
= NULL
;
1079 #ifdef CONFIG_COMPAT
1080 if (unlikely(tsk
->compat_robust_list
)) {
1081 compat_exit_robust_list(tsk
);
1082 tsk
->compat_robust_list
= NULL
;
1085 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
1086 exit_pi_state_list(tsk
);
1089 uprobe_free_utask(tsk
);
1091 /* Get rid of any cached register state */
1092 deactivate_mm(tsk
, mm
);
1095 * Signal userspace if we're not exiting with a core dump
1096 * because we want to leave the value intact for debugging
1099 if (tsk
->clear_child_tid
) {
1100 if (!(tsk
->signal
->flags
& SIGNAL_GROUP_COREDUMP
) &&
1101 atomic_read(&mm
->mm_users
) > 1) {
1103 * We don't check the error code - if userspace has
1104 * not set up a proper pointer then tough luck.
1106 put_user(0, tsk
->clear_child_tid
);
1107 sys_futex(tsk
->clear_child_tid
, FUTEX_WAKE
,
1110 tsk
->clear_child_tid
= NULL
;
1114 * All done, finally we can wake up parent and return this mm to him.
1115 * Also kthread_stop() uses this completion for synchronization.
1117 if (tsk
->vfork_done
)
1118 complete_vfork_done(tsk
);
1122 * Allocate a new mm structure and copy contents from the
1123 * mm structure of the passed in task structure.
1125 static struct mm_struct
*dup_mm(struct task_struct
*tsk
)
1127 struct mm_struct
*mm
, *oldmm
= current
->mm
;
1134 memcpy(mm
, oldmm
, sizeof(*mm
));
1136 if (!mm_init(mm
, tsk
, mm
->user_ns
))
1139 err
= dup_mmap(mm
, oldmm
);
1143 mm
->hiwater_rss
= get_mm_rss(mm
);
1144 mm
->hiwater_vm
= mm
->total_vm
;
1146 if (mm
->binfmt
&& !try_module_get(mm
->binfmt
->module
))
1152 /* don't put binfmt in mmput, we haven't got module yet */
1160 static int copy_mm(unsigned long clone_flags
, struct task_struct
*tsk
)
1162 struct mm_struct
*mm
, *oldmm
;
1165 tsk
->min_flt
= tsk
->maj_flt
= 0;
1166 tsk
->nvcsw
= tsk
->nivcsw
= 0;
1167 #ifdef CONFIG_DETECT_HUNG_TASK
1168 tsk
->last_switch_count
= tsk
->nvcsw
+ tsk
->nivcsw
;
1172 tsk
->active_mm
= NULL
;
1175 * Are we cloning a kernel thread?
1177 * We need to steal a active VM for that..
1179 oldmm
= current
->mm
;
1183 /* initialize the new vmacache entries */
1184 vmacache_flush(tsk
);
1186 if (clone_flags
& CLONE_VM
) {
1187 atomic_inc(&oldmm
->mm_users
);
1199 tsk
->active_mm
= mm
;
1206 static int copy_fs(unsigned long clone_flags
, struct task_struct
*tsk
)
1208 struct fs_struct
*fs
= current
->fs
;
1209 if (clone_flags
& CLONE_FS
) {
1210 /* tsk->fs is already what we want */
1211 spin_lock(&fs
->lock
);
1213 spin_unlock(&fs
->lock
);
1217 spin_unlock(&fs
->lock
);
1220 tsk
->fs
= copy_fs_struct(fs
);
1226 static int copy_files(unsigned long clone_flags
, struct task_struct
*tsk
)
1228 struct files_struct
*oldf
, *newf
;
1232 * A background process may not have any files ...
1234 oldf
= current
->files
;
1238 if (clone_flags
& CLONE_FILES
) {
1239 atomic_inc(&oldf
->count
);
1243 newf
= dup_fd(oldf
, &error
);
1253 static int copy_io(unsigned long clone_flags
, struct task_struct
*tsk
)
1256 struct io_context
*ioc
= current
->io_context
;
1257 struct io_context
*new_ioc
;
1262 * Share io context with parent, if CLONE_IO is set
1264 if (clone_flags
& CLONE_IO
) {
1266 tsk
->io_context
= ioc
;
1267 } else if (ioprio_valid(ioc
->ioprio
)) {
1268 new_ioc
= get_task_io_context(tsk
, GFP_KERNEL
, NUMA_NO_NODE
);
1269 if (unlikely(!new_ioc
))
1272 new_ioc
->ioprio
= ioc
->ioprio
;
1273 put_io_context(new_ioc
);
1279 static int copy_sighand(unsigned long clone_flags
, struct task_struct
*tsk
)
1281 struct sighand_struct
*sig
;
1283 if (clone_flags
& CLONE_SIGHAND
) {
1284 atomic_inc(¤t
->sighand
->count
);
1287 sig
= kmem_cache_alloc(sighand_cachep
, GFP_KERNEL
);
1288 rcu_assign_pointer(tsk
->sighand
, sig
);
1292 atomic_set(&sig
->count
, 1);
1293 memcpy(sig
->action
, current
->sighand
->action
, sizeof(sig
->action
));
1297 void __cleanup_sighand(struct sighand_struct
*sighand
)
1299 if (atomic_dec_and_test(&sighand
->count
)) {
1300 signalfd_cleanup(sighand
);
1302 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1303 * without an RCU grace period, see __lock_task_sighand().
1305 kmem_cache_free(sighand_cachep
, sighand
);
1309 #ifdef CONFIG_POSIX_TIMERS
1311 * Initialize POSIX timer handling for a thread group.
1313 static void posix_cpu_timers_init_group(struct signal_struct
*sig
)
1315 unsigned long cpu_limit
;
1317 cpu_limit
= READ_ONCE(sig
->rlim
[RLIMIT_CPU
].rlim_cur
);
1318 if (cpu_limit
!= RLIM_INFINITY
) {
1319 sig
->cputime_expires
.prof_exp
= cpu_limit
* NSEC_PER_SEC
;
1320 sig
->cputimer
.running
= true;
1323 /* The timer lists. */
1324 INIT_LIST_HEAD(&sig
->cpu_timers
[0]);
1325 INIT_LIST_HEAD(&sig
->cpu_timers
[1]);
1326 INIT_LIST_HEAD(&sig
->cpu_timers
[2]);
1329 static inline void posix_cpu_timers_init_group(struct signal_struct
*sig
) { }
1332 static int copy_signal(unsigned long clone_flags
, struct task_struct
*tsk
)
1334 struct signal_struct
*sig
;
1336 if (clone_flags
& CLONE_THREAD
)
1339 sig
= kmem_cache_zalloc(signal_cachep
, GFP_KERNEL
);
1344 sig
->nr_threads
= 1;
1345 atomic_set(&sig
->live
, 1);
1346 atomic_set(&sig
->sigcnt
, 1);
1348 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1349 sig
->thread_head
= (struct list_head
)LIST_HEAD_INIT(tsk
->thread_node
);
1350 tsk
->thread_node
= (struct list_head
)LIST_HEAD_INIT(sig
->thread_head
);
1352 init_waitqueue_head(&sig
->wait_chldexit
);
1353 sig
->curr_target
= tsk
;
1354 init_sigpending(&sig
->shared_pending
);
1355 seqlock_init(&sig
->stats_lock
);
1356 prev_cputime_init(&sig
->prev_cputime
);
1358 #ifdef CONFIG_POSIX_TIMERS
1359 INIT_LIST_HEAD(&sig
->posix_timers
);
1360 hrtimer_init(&sig
->real_timer
, CLOCK_MONOTONIC
, HRTIMER_MODE_REL
);
1361 sig
->real_timer
.function
= it_real_fn
;
1364 task_lock(current
->group_leader
);
1365 memcpy(sig
->rlim
, current
->signal
->rlim
, sizeof sig
->rlim
);
1366 task_unlock(current
->group_leader
);
1368 posix_cpu_timers_init_group(sig
);
1370 tty_audit_fork(sig
);
1371 sched_autogroup_fork(sig
);
1373 sig
->oom_score_adj
= current
->signal
->oom_score_adj
;
1374 sig
->oom_score_adj_min
= current
->signal
->oom_score_adj_min
;
1376 sig
->has_child_subreaper
= current
->signal
->has_child_subreaper
||
1377 current
->signal
->is_child_subreaper
;
1379 mutex_init(&sig
->cred_guard_mutex
);
1384 static void copy_seccomp(struct task_struct
*p
)
1386 #ifdef CONFIG_SECCOMP
1388 * Must be called with sighand->lock held, which is common to
1389 * all threads in the group. Holding cred_guard_mutex is not
1390 * needed because this new task is not yet running and cannot
1393 assert_spin_locked(¤t
->sighand
->siglock
);
1395 /* Ref-count the new filter user, and assign it. */
1396 get_seccomp_filter(current
);
1397 p
->seccomp
= current
->seccomp
;
1400 * Explicitly enable no_new_privs here in case it got set
1401 * between the task_struct being duplicated and holding the
1402 * sighand lock. The seccomp state and nnp must be in sync.
1404 if (task_no_new_privs(current
))
1405 task_set_no_new_privs(p
);
1408 * If the parent gained a seccomp mode after copying thread
1409 * flags and between before we held the sighand lock, we have
1410 * to manually enable the seccomp thread flag here.
1412 if (p
->seccomp
.mode
!= SECCOMP_MODE_DISABLED
)
1413 set_tsk_thread_flag(p
, TIF_SECCOMP
);
1417 SYSCALL_DEFINE1(set_tid_address
, int __user
*, tidptr
)
1419 current
->clear_child_tid
= tidptr
;
1421 return task_pid_vnr(current
);
1424 static void rt_mutex_init_task(struct task_struct
*p
)
1426 raw_spin_lock_init(&p
->pi_lock
);
1427 #ifdef CONFIG_RT_MUTEXES
1428 p
->pi_waiters
= RB_ROOT
;
1429 p
->pi_waiters_leftmost
= NULL
;
1430 p
->pi_blocked_on
= NULL
;
1434 #ifdef CONFIG_POSIX_TIMERS
1436 * Initialize POSIX timer handling for a single task.
1438 static void posix_cpu_timers_init(struct task_struct
*tsk
)
1440 tsk
->cputime_expires
.prof_exp
= 0;
1441 tsk
->cputime_expires
.virt_exp
= 0;
1442 tsk
->cputime_expires
.sched_exp
= 0;
1443 INIT_LIST_HEAD(&tsk
->cpu_timers
[0]);
1444 INIT_LIST_HEAD(&tsk
->cpu_timers
[1]);
1445 INIT_LIST_HEAD(&tsk
->cpu_timers
[2]);
1448 static inline void posix_cpu_timers_init(struct task_struct
*tsk
) { }
1452 init_task_pid(struct task_struct
*task
, enum pid_type type
, struct pid
*pid
)
1454 task
->pids
[type
].pid
= pid
;
1458 * This creates a new process as a copy of the old one,
1459 * but does not actually start it yet.
1461 * It copies the registers, and all the appropriate
1462 * parts of the process environment (as per the clone
1463 * flags). The actual kick-off is left to the caller.
1465 static __latent_entropy
struct task_struct
*copy_process(
1466 unsigned long clone_flags
,
1467 unsigned long stack_start
,
1468 unsigned long stack_size
,
1469 int __user
*child_tidptr
,
1476 struct task_struct
*p
;
1478 if ((clone_flags
& (CLONE_NEWNS
|CLONE_FS
)) == (CLONE_NEWNS
|CLONE_FS
))
1479 return ERR_PTR(-EINVAL
);
1481 if ((clone_flags
& (CLONE_NEWUSER
|CLONE_FS
)) == (CLONE_NEWUSER
|CLONE_FS
))
1482 return ERR_PTR(-EINVAL
);
1485 * Thread groups must share signals as well, and detached threads
1486 * can only be started up within the thread group.
1488 if ((clone_flags
& CLONE_THREAD
) && !(clone_flags
& CLONE_SIGHAND
))
1489 return ERR_PTR(-EINVAL
);
1492 * Shared signal handlers imply shared VM. By way of the above,
1493 * thread groups also imply shared VM. Blocking this case allows
1494 * for various simplifications in other code.
1496 if ((clone_flags
& CLONE_SIGHAND
) && !(clone_flags
& CLONE_VM
))
1497 return ERR_PTR(-EINVAL
);
1500 * Siblings of global init remain as zombies on exit since they are
1501 * not reaped by their parent (swapper). To solve this and to avoid
1502 * multi-rooted process trees, prevent global and container-inits
1503 * from creating siblings.
1505 if ((clone_flags
& CLONE_PARENT
) &&
1506 current
->signal
->flags
& SIGNAL_UNKILLABLE
)
1507 return ERR_PTR(-EINVAL
);
1510 * If the new process will be in a different pid or user namespace
1511 * do not allow it to share a thread group with the forking task.
1513 if (clone_flags
& CLONE_THREAD
) {
1514 if ((clone_flags
& (CLONE_NEWUSER
| CLONE_NEWPID
)) ||
1515 (task_active_pid_ns(current
) !=
1516 current
->nsproxy
->pid_ns_for_children
))
1517 return ERR_PTR(-EINVAL
);
1520 retval
= security_task_create(clone_flags
);
1525 p
= dup_task_struct(current
, node
);
1529 ftrace_graph_init_task(p
);
1531 rt_mutex_init_task(p
);
1533 #ifdef CONFIG_PROVE_LOCKING
1534 DEBUG_LOCKS_WARN_ON(!p
->hardirqs_enabled
);
1535 DEBUG_LOCKS_WARN_ON(!p
->softirqs_enabled
);
1538 if (atomic_read(&p
->real_cred
->user
->processes
) >=
1539 task_rlimit(p
, RLIMIT_NPROC
)) {
1540 if (p
->real_cred
->user
!= INIT_USER
&&
1541 !capable(CAP_SYS_RESOURCE
) && !capable(CAP_SYS_ADMIN
))
1544 current
->flags
&= ~PF_NPROC_EXCEEDED
;
1546 retval
= copy_creds(p
, clone_flags
);
1551 * If multiple threads are within copy_process(), then this check
1552 * triggers too late. This doesn't hurt, the check is only there
1553 * to stop root fork bombs.
1556 if (nr_threads
>= max_threads
)
1557 goto bad_fork_cleanup_count
;
1559 delayacct_tsk_init(p
); /* Must remain after dup_task_struct() */
1560 p
->flags
&= ~(PF_SUPERPRIV
| PF_WQ_WORKER
| PF_IDLE
);
1561 p
->flags
|= PF_FORKNOEXEC
;
1562 INIT_LIST_HEAD(&p
->children
);
1563 INIT_LIST_HEAD(&p
->sibling
);
1564 rcu_copy_process(p
);
1565 p
->vfork_done
= NULL
;
1566 spin_lock_init(&p
->alloc_lock
);
1568 init_sigpending(&p
->pending
);
1570 p
->utime
= p
->stime
= p
->gtime
= 0;
1571 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1572 p
->utimescaled
= p
->stimescaled
= 0;
1574 prev_cputime_init(&p
->prev_cputime
);
1576 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1577 seqcount_init(&p
->vtime_seqcount
);
1579 p
->vtime_snap_whence
= VTIME_INACTIVE
;
1582 #if defined(SPLIT_RSS_COUNTING)
1583 memset(&p
->rss_stat
, 0, sizeof(p
->rss_stat
));
1586 p
->default_timer_slack_ns
= current
->timer_slack_ns
;
1588 task_io_accounting_init(&p
->ioac
);
1589 acct_clear_integrals(p
);
1591 posix_cpu_timers_init(p
);
1593 p
->start_time
= ktime_get_ns();
1594 p
->real_start_time
= ktime_get_boot_ns();
1595 p
->io_context
= NULL
;
1596 p
->audit_context
= NULL
;
1599 p
->mempolicy
= mpol_dup(p
->mempolicy
);
1600 if (IS_ERR(p
->mempolicy
)) {
1601 retval
= PTR_ERR(p
->mempolicy
);
1602 p
->mempolicy
= NULL
;
1603 goto bad_fork_cleanup_threadgroup_lock
;
1606 #ifdef CONFIG_CPUSETS
1607 p
->cpuset_mem_spread_rotor
= NUMA_NO_NODE
;
1608 p
->cpuset_slab_spread_rotor
= NUMA_NO_NODE
;
1609 seqcount_init(&p
->mems_allowed_seq
);
1611 #ifdef CONFIG_TRACE_IRQFLAGS
1613 p
->hardirqs_enabled
= 0;
1614 p
->hardirq_enable_ip
= 0;
1615 p
->hardirq_enable_event
= 0;
1616 p
->hardirq_disable_ip
= _THIS_IP_
;
1617 p
->hardirq_disable_event
= 0;
1618 p
->softirqs_enabled
= 1;
1619 p
->softirq_enable_ip
= _THIS_IP_
;
1620 p
->softirq_enable_event
= 0;
1621 p
->softirq_disable_ip
= 0;
1622 p
->softirq_disable_event
= 0;
1623 p
->hardirq_context
= 0;
1624 p
->softirq_context
= 0;
1627 p
->pagefault_disabled
= 0;
1629 #ifdef CONFIG_LOCKDEP
1630 p
->lockdep_depth
= 0; /* no locks held yet */
1631 p
->curr_chain_key
= 0;
1632 p
->lockdep_recursion
= 0;
1635 #ifdef CONFIG_DEBUG_MUTEXES
1636 p
->blocked_on
= NULL
; /* not blocked yet */
1638 #ifdef CONFIG_BCACHE
1639 p
->sequential_io
= 0;
1640 p
->sequential_io_avg
= 0;
1643 /* Perform scheduler related setup. Assign this task to a CPU. */
1644 retval
= sched_fork(clone_flags
, p
);
1646 goto bad_fork_cleanup_policy
;
1648 retval
= perf_event_init_task(p
);
1650 goto bad_fork_cleanup_policy
;
1651 retval
= audit_alloc(p
);
1653 goto bad_fork_cleanup_perf
;
1654 /* copy all the process information */
1656 retval
= copy_semundo(clone_flags
, p
);
1658 goto bad_fork_cleanup_audit
;
1659 retval
= copy_files(clone_flags
, p
);
1661 goto bad_fork_cleanup_semundo
;
1662 retval
= copy_fs(clone_flags
, p
);
1664 goto bad_fork_cleanup_files
;
1665 retval
= copy_sighand(clone_flags
, p
);
1667 goto bad_fork_cleanup_fs
;
1668 retval
= copy_signal(clone_flags
, p
);
1670 goto bad_fork_cleanup_sighand
;
1671 retval
= copy_mm(clone_flags
, p
);
1673 goto bad_fork_cleanup_signal
;
1674 retval
= copy_namespaces(clone_flags
, p
);
1676 goto bad_fork_cleanup_mm
;
1677 retval
= copy_io(clone_flags
, p
);
1679 goto bad_fork_cleanup_namespaces
;
1680 retval
= copy_thread_tls(clone_flags
, stack_start
, stack_size
, p
, tls
);
1682 goto bad_fork_cleanup_io
;
1684 if (pid
!= &init_struct_pid
) {
1685 pid
= alloc_pid(p
->nsproxy
->pid_ns_for_children
);
1687 retval
= PTR_ERR(pid
);
1688 goto bad_fork_cleanup_thread
;
1692 p
->set_child_tid
= (clone_flags
& CLONE_CHILD_SETTID
) ? child_tidptr
: NULL
;
1694 * Clear TID on mm_release()?
1696 p
->clear_child_tid
= (clone_flags
& CLONE_CHILD_CLEARTID
) ? child_tidptr
: NULL
;
1701 p
->robust_list
= NULL
;
1702 #ifdef CONFIG_COMPAT
1703 p
->compat_robust_list
= NULL
;
1705 INIT_LIST_HEAD(&p
->pi_state_list
);
1706 p
->pi_state_cache
= NULL
;
1709 * sigaltstack should be cleared when sharing the same VM
1711 if ((clone_flags
& (CLONE_VM
|CLONE_VFORK
)) == CLONE_VM
)
1715 * Syscall tracing and stepping should be turned off in the
1716 * child regardless of CLONE_PTRACE.
1718 user_disable_single_step(p
);
1719 clear_tsk_thread_flag(p
, TIF_SYSCALL_TRACE
);
1720 #ifdef TIF_SYSCALL_EMU
1721 clear_tsk_thread_flag(p
, TIF_SYSCALL_EMU
);
1723 clear_all_latency_tracing(p
);
1725 /* ok, now we should be set up.. */
1726 p
->pid
= pid_nr(pid
);
1727 if (clone_flags
& CLONE_THREAD
) {
1728 p
->exit_signal
= -1;
1729 p
->group_leader
= current
->group_leader
;
1730 p
->tgid
= current
->tgid
;
1732 if (clone_flags
& CLONE_PARENT
)
1733 p
->exit_signal
= current
->group_leader
->exit_signal
;
1735 p
->exit_signal
= (clone_flags
& CSIGNAL
);
1736 p
->group_leader
= p
;
1741 p
->nr_dirtied_pause
= 128 >> (PAGE_SHIFT
- 10);
1742 p
->dirty_paused_when
= 0;
1744 p
->pdeath_signal
= 0;
1745 INIT_LIST_HEAD(&p
->thread_group
);
1746 p
->task_works
= NULL
;
1748 threadgroup_change_begin(current
);
1750 * Ensure that the cgroup subsystem policies allow the new process to be
1751 * forked. It should be noted the the new process's css_set can be changed
1752 * between here and cgroup_post_fork() if an organisation operation is in
1755 retval
= cgroup_can_fork(p
);
1757 goto bad_fork_free_pid
;
1760 * Make it visible to the rest of the system, but dont wake it up yet.
1761 * Need tasklist lock for parent etc handling!
1763 write_lock_irq(&tasklist_lock
);
1765 /* CLONE_PARENT re-uses the old parent */
1766 if (clone_flags
& (CLONE_PARENT
|CLONE_THREAD
)) {
1767 p
->real_parent
= current
->real_parent
;
1768 p
->parent_exec_id
= current
->parent_exec_id
;
1770 p
->real_parent
= current
;
1771 p
->parent_exec_id
= current
->self_exec_id
;
1774 spin_lock(¤t
->sighand
->siglock
);
1777 * Copy seccomp details explicitly here, in case they were changed
1778 * before holding sighand lock.
1783 * Process group and session signals need to be delivered to just the
1784 * parent before the fork or both the parent and the child after the
1785 * fork. Restart if a signal comes in before we add the new process to
1786 * it's process group.
1787 * A fatal signal pending means that current will exit, so the new
1788 * thread can't slip out of an OOM kill (or normal SIGKILL).
1790 recalc_sigpending();
1791 if (signal_pending(current
)) {
1792 spin_unlock(¤t
->sighand
->siglock
);
1793 write_unlock_irq(&tasklist_lock
);
1794 retval
= -ERESTARTNOINTR
;
1795 goto bad_fork_cancel_cgroup
;
1798 if (likely(p
->pid
)) {
1799 ptrace_init_task(p
, (clone_flags
& CLONE_PTRACE
) || trace
);
1801 init_task_pid(p
, PIDTYPE_PID
, pid
);
1802 if (thread_group_leader(p
)) {
1803 init_task_pid(p
, PIDTYPE_PGID
, task_pgrp(current
));
1804 init_task_pid(p
, PIDTYPE_SID
, task_session(current
));
1806 if (is_child_reaper(pid
)) {
1807 ns_of_pid(pid
)->child_reaper
= p
;
1808 p
->signal
->flags
|= SIGNAL_UNKILLABLE
;
1811 p
->signal
->leader_pid
= pid
;
1812 p
->signal
->tty
= tty_kref_get(current
->signal
->tty
);
1813 list_add_tail(&p
->sibling
, &p
->real_parent
->children
);
1814 list_add_tail_rcu(&p
->tasks
, &init_task
.tasks
);
1815 attach_pid(p
, PIDTYPE_PGID
);
1816 attach_pid(p
, PIDTYPE_SID
);
1817 __this_cpu_inc(process_counts
);
1819 current
->signal
->nr_threads
++;
1820 atomic_inc(¤t
->signal
->live
);
1821 atomic_inc(¤t
->signal
->sigcnt
);
1822 list_add_tail_rcu(&p
->thread_group
,
1823 &p
->group_leader
->thread_group
);
1824 list_add_tail_rcu(&p
->thread_node
,
1825 &p
->signal
->thread_head
);
1827 attach_pid(p
, PIDTYPE_PID
);
1832 spin_unlock(¤t
->sighand
->siglock
);
1833 syscall_tracepoint_update(p
);
1834 write_unlock_irq(&tasklist_lock
);
1836 proc_fork_connector(p
);
1837 cgroup_post_fork(p
);
1838 threadgroup_change_end(current
);
1841 trace_task_newtask(p
, clone_flags
);
1842 uprobe_copy_process(p
, clone_flags
);
1846 bad_fork_cancel_cgroup
:
1847 cgroup_cancel_fork(p
);
1849 threadgroup_change_end(current
);
1850 if (pid
!= &init_struct_pid
)
1852 bad_fork_cleanup_thread
:
1854 bad_fork_cleanup_io
:
1857 bad_fork_cleanup_namespaces
:
1858 exit_task_namespaces(p
);
1859 bad_fork_cleanup_mm
:
1862 bad_fork_cleanup_signal
:
1863 if (!(clone_flags
& CLONE_THREAD
))
1864 free_signal_struct(p
->signal
);
1865 bad_fork_cleanup_sighand
:
1866 __cleanup_sighand(p
->sighand
);
1867 bad_fork_cleanup_fs
:
1868 exit_fs(p
); /* blocking */
1869 bad_fork_cleanup_files
:
1870 exit_files(p
); /* blocking */
1871 bad_fork_cleanup_semundo
:
1873 bad_fork_cleanup_audit
:
1875 bad_fork_cleanup_perf
:
1876 perf_event_free_task(p
);
1877 bad_fork_cleanup_policy
:
1879 mpol_put(p
->mempolicy
);
1880 bad_fork_cleanup_threadgroup_lock
:
1882 delayacct_tsk_free(p
);
1883 bad_fork_cleanup_count
:
1884 atomic_dec(&p
->cred
->user
->processes
);
1887 p
->state
= TASK_DEAD
;
1891 return ERR_PTR(retval
);
1894 static inline void init_idle_pids(struct pid_link
*links
)
1898 for (type
= PIDTYPE_PID
; type
< PIDTYPE_MAX
; ++type
) {
1899 INIT_HLIST_NODE(&links
[type
].node
); /* not really needed */
1900 links
[type
].pid
= &init_struct_pid
;
1904 struct task_struct
*fork_idle(int cpu
)
1906 struct task_struct
*task
;
1907 task
= copy_process(CLONE_VM
, 0, 0, NULL
, &init_struct_pid
, 0, 0,
1909 if (!IS_ERR(task
)) {
1910 init_idle_pids(task
->pids
);
1911 init_idle(task
, cpu
);
1918 * Ok, this is the main fork-routine.
1920 * It copies the process, and if successful kick-starts
1921 * it and waits for it to finish using the VM if required.
1923 long _do_fork(unsigned long clone_flags
,
1924 unsigned long stack_start
,
1925 unsigned long stack_size
,
1926 int __user
*parent_tidptr
,
1927 int __user
*child_tidptr
,
1930 struct task_struct
*p
;
1935 * Determine whether and which event to report to ptracer. When
1936 * called from kernel_thread or CLONE_UNTRACED is explicitly
1937 * requested, no event is reported; otherwise, report if the event
1938 * for the type of forking is enabled.
1940 if (!(clone_flags
& CLONE_UNTRACED
)) {
1941 if (clone_flags
& CLONE_VFORK
)
1942 trace
= PTRACE_EVENT_VFORK
;
1943 else if ((clone_flags
& CSIGNAL
) != SIGCHLD
)
1944 trace
= PTRACE_EVENT_CLONE
;
1946 trace
= PTRACE_EVENT_FORK
;
1948 if (likely(!ptrace_event_enabled(current
, trace
)))
1952 p
= copy_process(clone_flags
, stack_start
, stack_size
,
1953 child_tidptr
, NULL
, trace
, tls
, NUMA_NO_NODE
);
1954 add_latent_entropy();
1956 * Do this prior waking up the new thread - the thread pointer
1957 * might get invalid after that point, if the thread exits quickly.
1960 struct completion vfork
;
1963 trace_sched_process_fork(current
, p
);
1965 pid
= get_task_pid(p
, PIDTYPE_PID
);
1968 if (clone_flags
& CLONE_PARENT_SETTID
)
1969 put_user(nr
, parent_tidptr
);
1971 if (clone_flags
& CLONE_VFORK
) {
1972 p
->vfork_done
= &vfork
;
1973 init_completion(&vfork
);
1977 wake_up_new_task(p
);
1979 /* forking complete and child started to run, tell ptracer */
1980 if (unlikely(trace
))
1981 ptrace_event_pid(trace
, pid
);
1983 if (clone_flags
& CLONE_VFORK
) {
1984 if (!wait_for_vfork_done(p
, &vfork
))
1985 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE
, pid
);
1995 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1996 /* For compatibility with architectures that call do_fork directly rather than
1997 * using the syscall entry points below. */
1998 long do_fork(unsigned long clone_flags
,
1999 unsigned long stack_start
,
2000 unsigned long stack_size
,
2001 int __user
*parent_tidptr
,
2002 int __user
*child_tidptr
)
2004 return _do_fork(clone_flags
, stack_start
, stack_size
,
2005 parent_tidptr
, child_tidptr
, 0);
2010 * Create a kernel thread.
2012 pid_t
kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
2014 return _do_fork(flags
|CLONE_VM
|CLONE_UNTRACED
, (unsigned long)fn
,
2015 (unsigned long)arg
, NULL
, NULL
, 0);
2018 #ifdef __ARCH_WANT_SYS_FORK
2019 SYSCALL_DEFINE0(fork
)
2022 return _do_fork(SIGCHLD
, 0, 0, NULL
, NULL
, 0);
2024 /* can not support in nommu mode */
2030 #ifdef __ARCH_WANT_SYS_VFORK
2031 SYSCALL_DEFINE0(vfork
)
2033 return _do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, 0,
2038 #ifdef __ARCH_WANT_SYS_CLONE
2039 #ifdef CONFIG_CLONE_BACKWARDS
2040 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2041 int __user
*, parent_tidptr
,
2043 int __user
*, child_tidptr
)
2044 #elif defined(CONFIG_CLONE_BACKWARDS2)
2045 SYSCALL_DEFINE5(clone
, unsigned long, newsp
, unsigned long, clone_flags
,
2046 int __user
*, parent_tidptr
,
2047 int __user
*, child_tidptr
,
2049 #elif defined(CONFIG_CLONE_BACKWARDS3)
2050 SYSCALL_DEFINE6(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2052 int __user
*, parent_tidptr
,
2053 int __user
*, child_tidptr
,
2056 SYSCALL_DEFINE5(clone
, unsigned long, clone_flags
, unsigned long, newsp
,
2057 int __user
*, parent_tidptr
,
2058 int __user
*, child_tidptr
,
2062 return _do_fork(clone_flags
, newsp
, 0, parent_tidptr
, child_tidptr
, tls
);
2066 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2067 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2070 static void sighand_ctor(void *data
)
2072 struct sighand_struct
*sighand
= data
;
2074 spin_lock_init(&sighand
->siglock
);
2075 init_waitqueue_head(&sighand
->signalfd_wqh
);
2078 void __init
proc_caches_init(void)
2080 sighand_cachep
= kmem_cache_create("sighand_cache",
2081 sizeof(struct sighand_struct
), 0,
2082 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_DESTROY_BY_RCU
|
2083 SLAB_NOTRACK
|SLAB_ACCOUNT
, sighand_ctor
);
2084 signal_cachep
= kmem_cache_create("signal_cache",
2085 sizeof(struct signal_struct
), 0,
2086 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2088 files_cachep
= kmem_cache_create("files_cache",
2089 sizeof(struct files_struct
), 0,
2090 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2092 fs_cachep
= kmem_cache_create("fs_cache",
2093 sizeof(struct fs_struct
), 0,
2094 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2097 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2098 * whole struct cpumask for the OFFSTACK case. We could change
2099 * this to *only* allocate as much of it as required by the
2100 * maximum number of CPU's we can ever have. The cpumask_allocation
2101 * is at the end of the structure, exactly for that reason.
2103 mm_cachep
= kmem_cache_create("mm_struct",
2104 sizeof(struct mm_struct
), ARCH_MIN_MMSTRUCT_ALIGN
,
2105 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_NOTRACK
|SLAB_ACCOUNT
,
2107 vm_area_cachep
= KMEM_CACHE(vm_area_struct
, SLAB_PANIC
|SLAB_ACCOUNT
);
2109 nsproxy_cache_init();
2113 * Check constraints on flags passed to the unshare system call.
2115 static int check_unshare_flags(unsigned long unshare_flags
)
2117 if (unshare_flags
& ~(CLONE_THREAD
|CLONE_FS
|CLONE_NEWNS
|CLONE_SIGHAND
|
2118 CLONE_VM
|CLONE_FILES
|CLONE_SYSVSEM
|
2119 CLONE_NEWUTS
|CLONE_NEWIPC
|CLONE_NEWNET
|
2120 CLONE_NEWUSER
|CLONE_NEWPID
|CLONE_NEWCGROUP
))
2123 * Not implemented, but pretend it works if there is nothing
2124 * to unshare. Note that unsharing the address space or the
2125 * signal handlers also need to unshare the signal queues (aka
2128 if (unshare_flags
& (CLONE_THREAD
| CLONE_SIGHAND
| CLONE_VM
)) {
2129 if (!thread_group_empty(current
))
2132 if (unshare_flags
& (CLONE_SIGHAND
| CLONE_VM
)) {
2133 if (atomic_read(¤t
->sighand
->count
) > 1)
2136 if (unshare_flags
& CLONE_VM
) {
2137 if (!current_is_single_threaded())
2145 * Unshare the filesystem structure if it is being shared
2147 static int unshare_fs(unsigned long unshare_flags
, struct fs_struct
**new_fsp
)
2149 struct fs_struct
*fs
= current
->fs
;
2151 if (!(unshare_flags
& CLONE_FS
) || !fs
)
2154 /* don't need lock here; in the worst case we'll do useless copy */
2158 *new_fsp
= copy_fs_struct(fs
);
2166 * Unshare file descriptor table if it is being shared
2168 static int unshare_fd(unsigned long unshare_flags
, struct files_struct
**new_fdp
)
2170 struct files_struct
*fd
= current
->files
;
2173 if ((unshare_flags
& CLONE_FILES
) &&
2174 (fd
&& atomic_read(&fd
->count
) > 1)) {
2175 *new_fdp
= dup_fd(fd
, &error
);
2184 * unshare allows a process to 'unshare' part of the process
2185 * context which was originally shared using clone. copy_*
2186 * functions used by do_fork() cannot be used here directly
2187 * because they modify an inactive task_struct that is being
2188 * constructed. Here we are modifying the current, active,
2191 SYSCALL_DEFINE1(unshare
, unsigned long, unshare_flags
)
2193 struct fs_struct
*fs
, *new_fs
= NULL
;
2194 struct files_struct
*fd
, *new_fd
= NULL
;
2195 struct cred
*new_cred
= NULL
;
2196 struct nsproxy
*new_nsproxy
= NULL
;
2201 * If unsharing a user namespace must also unshare the thread group
2202 * and unshare the filesystem root and working directories.
2204 if (unshare_flags
& CLONE_NEWUSER
)
2205 unshare_flags
|= CLONE_THREAD
| CLONE_FS
;
2207 * If unsharing vm, must also unshare signal handlers.
2209 if (unshare_flags
& CLONE_VM
)
2210 unshare_flags
|= CLONE_SIGHAND
;
2212 * If unsharing a signal handlers, must also unshare the signal queues.
2214 if (unshare_flags
& CLONE_SIGHAND
)
2215 unshare_flags
|= CLONE_THREAD
;
2217 * If unsharing namespace, must also unshare filesystem information.
2219 if (unshare_flags
& CLONE_NEWNS
)
2220 unshare_flags
|= CLONE_FS
;
2222 err
= check_unshare_flags(unshare_flags
);
2224 goto bad_unshare_out
;
2226 * CLONE_NEWIPC must also detach from the undolist: after switching
2227 * to a new ipc namespace, the semaphore arrays from the old
2228 * namespace are unreachable.
2230 if (unshare_flags
& (CLONE_NEWIPC
|CLONE_SYSVSEM
))
2232 err
= unshare_fs(unshare_flags
, &new_fs
);
2234 goto bad_unshare_out
;
2235 err
= unshare_fd(unshare_flags
, &new_fd
);
2237 goto bad_unshare_cleanup_fs
;
2238 err
= unshare_userns(unshare_flags
, &new_cred
);
2240 goto bad_unshare_cleanup_fd
;
2241 err
= unshare_nsproxy_namespaces(unshare_flags
, &new_nsproxy
,
2244 goto bad_unshare_cleanup_cred
;
2246 if (new_fs
|| new_fd
|| do_sysvsem
|| new_cred
|| new_nsproxy
) {
2249 * CLONE_SYSVSEM is equivalent to sys_exit().
2253 if (unshare_flags
& CLONE_NEWIPC
) {
2254 /* Orphan segments in old ns (see sem above). */
2256 shm_init_task(current
);
2260 switch_task_namespaces(current
, new_nsproxy
);
2266 spin_lock(&fs
->lock
);
2267 current
->fs
= new_fs
;
2272 spin_unlock(&fs
->lock
);
2276 fd
= current
->files
;
2277 current
->files
= new_fd
;
2281 task_unlock(current
);
2284 /* Install the new user namespace */
2285 commit_creds(new_cred
);
2290 bad_unshare_cleanup_cred
:
2293 bad_unshare_cleanup_fd
:
2295 put_files_struct(new_fd
);
2297 bad_unshare_cleanup_fs
:
2299 free_fs_struct(new_fs
);
2306 * Helper to unshare the files of the current task.
2307 * We don't want to expose copy_files internals to
2308 * the exec layer of the kernel.
2311 int unshare_files(struct files_struct
**displaced
)
2313 struct task_struct
*task
= current
;
2314 struct files_struct
*copy
= NULL
;
2317 error
= unshare_fd(CLONE_FILES
, ©
);
2318 if (error
|| !copy
) {
2322 *displaced
= task
->files
;
2329 int sysctl_max_threads(struct ctl_table
*table
, int write
,
2330 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
2334 int threads
= max_threads
;
2335 int min
= MIN_THREADS
;
2336 int max
= MAX_THREADS
;
2343 ret
= proc_dointvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
2347 set_max_threads(threads
);