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>
78 #include <asm/pgtable.h>
79 #include <asm/pgalloc.h>
80 #include <asm/uaccess.h>
81 #include <asm/mmu_context.h>
82 #include <asm/cacheflush.h>
83 #include <asm/tlbflush.h>
85 #include <trace/events/sched.h>
87 #define CREATE_TRACE_POINTS
88 #include <trace/events/task.h>
91 * Minimum number of threads to boot the kernel
93 #define MIN_THREADS 20
96 * Maximum number of threads
98 #define MAX_THREADS FUTEX_TID_MASK
101 * Protected counters by write_lock_irq(&tasklist_lock)
103 unsigned long total_forks; /* Handle normal Linux uptimes. */
104 int nr_threads; /* The idle threads do not count.. */
106 int max_threads; /* tunable limit on nr_threads */
108 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
110 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
112 #ifdef CONFIG_PROVE_RCU
113 int lockdep_tasklist_lock_is_held(void)
115 return lockdep_is_held(&tasklist_lock);
117 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
118 #endif /* #ifdef CONFIG_PROVE_RCU */
120 int nr_processes(void)
125 for_each_possible_cpu(cpu)
126 total += per_cpu(process_counts, cpu);
131 void __weak arch_release_task_struct(struct task_struct *tsk)
135 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
136 static struct kmem_cache *task_struct_cachep;
138 static inline struct task_struct *alloc_task_struct_node(int node)
140 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
143 static inline void free_task_struct(struct task_struct *tsk)
145 kmem_cache_free(task_struct_cachep, tsk);
149 void __weak arch_release_thread_info(struct thread_info *ti)
153 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
156 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
157 * kmemcache based allocator.
159 # if THREAD_SIZE >= PAGE_SIZE
160 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
163 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
166 return page ? page_address(page) : NULL;
169 static inline void free_thread_info(struct thread_info *ti)
171 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
174 static struct kmem_cache *thread_info_cache;
176 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
179 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
182 static void free_thread_info(struct thread_info *ti)
184 kmem_cache_free(thread_info_cache, ti);
187 void thread_info_cache_init(void)
189 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
190 THREAD_SIZE, 0, NULL);
191 BUG_ON(thread_info_cache == NULL);
196 /* SLAB cache for signal_struct structures (tsk->signal) */
197 static struct kmem_cache *signal_cachep;
199 /* SLAB cache for sighand_struct structures (tsk->sighand) */
200 struct kmem_cache *sighand_cachep;
202 /* SLAB cache for files_struct structures (tsk->files) */
203 struct kmem_cache *files_cachep;
205 /* SLAB cache for fs_struct structures (tsk->fs) */
206 struct kmem_cache *fs_cachep;
208 /* SLAB cache for vm_area_struct structures */
209 struct kmem_cache *vm_area_cachep;
211 /* SLAB cache for mm_struct structures (tsk->mm) */
212 static struct kmem_cache *mm_cachep;
214 static void account_kernel_stack(struct thread_info *ti, int account)
216 struct zone *zone = page_zone(virt_to_page(ti));
218 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
221 void free_task(struct task_struct *tsk)
223 account_kernel_stack(tsk->stack, -1);
224 arch_release_thread_info(tsk->stack);
225 free_thread_info(tsk->stack);
226 rt_mutex_debug_task_free(tsk);
227 ftrace_graph_exit_task(tsk);
228 put_seccomp_filter(tsk);
229 arch_release_task_struct(tsk);
230 free_task_struct(tsk);
232 EXPORT_SYMBOL(free_task);
234 static inline void free_signal_struct(struct signal_struct *sig)
236 taskstats_tgid_free(sig);
237 sched_autogroup_exit(sig);
238 kmem_cache_free(signal_cachep, sig);
241 static inline void put_signal_struct(struct signal_struct *sig)
243 if (atomic_dec_and_test(&sig->sigcnt))
244 free_signal_struct(sig);
247 void __put_task_struct(struct task_struct *tsk)
249 WARN_ON(!tsk->exit_state);
250 WARN_ON(atomic_read(&tsk->usage));
251 WARN_ON(tsk == current);
254 security_task_free(tsk);
256 delayacct_tsk_free(tsk);
257 put_signal_struct(tsk->signal);
259 if (!profile_handoff_task(tsk))
262 EXPORT_SYMBOL_GPL(__put_task_struct);
264 void __init __weak arch_task_cache_init(void) { }
269 static void set_max_threads(void)
274 * The number of threads shall be limited such that the thread
275 * structures may only consume a small part of the available memory.
277 if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
278 threads = MAX_THREADS;
280 threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
281 (u64) THREAD_SIZE * 8UL);
283 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
286 void __init fork_init(void)
288 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
289 #ifndef ARCH_MIN_TASKALIGN
290 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
292 /* create a slab on which task_structs can be allocated */
294 kmem_cache_create("task_struct", sizeof(struct task_struct),
295 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
298 /* do the arch specific task caches init */
299 arch_task_cache_init();
303 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
304 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
305 init_task.signal->rlim[RLIMIT_SIGPENDING] =
306 init_task.signal->rlim[RLIMIT_NPROC];
309 int __weak arch_dup_task_struct(struct task_struct *dst,
310 struct task_struct *src)
316 void set_task_stack_end_magic(struct task_struct *tsk)
318 unsigned long *stackend;
320 stackend = end_of_stack(tsk);
321 *stackend = STACK_END_MAGIC; /* for overflow detection */
324 static struct task_struct *dup_task_struct(struct task_struct *orig)
326 struct task_struct *tsk;
327 struct thread_info *ti;
328 int node = tsk_fork_get_node(orig);
331 tsk = alloc_task_struct_node(node);
335 ti = alloc_thread_info_node(tsk, node);
339 err = arch_dup_task_struct(tsk, orig);
344 #ifdef CONFIG_SECCOMP
346 * We must handle setting up seccomp filters once we're under
347 * the sighand lock in case orig has changed between now and
348 * then. Until then, filter must be NULL to avoid messing up
349 * the usage counts on the error path calling free_task.
351 tsk->seccomp.filter = NULL;
354 setup_thread_stack(tsk, orig);
355 clear_user_return_notifier(tsk);
356 clear_tsk_need_resched(tsk);
357 set_task_stack_end_magic(tsk);
359 #ifdef CONFIG_CC_STACKPROTECTOR
360 tsk->stack_canary = get_random_int();
364 * One for us, one for whoever does the "release_task()" (usually
367 atomic_set(&tsk->usage, 2);
368 #ifdef CONFIG_BLK_DEV_IO_TRACE
371 tsk->splice_pipe = NULL;
372 tsk->task_frag.page = NULL;
374 account_kernel_stack(ti, 1);
379 free_thread_info(ti);
381 free_task_struct(tsk);
386 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
388 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
389 struct rb_node **rb_link, *rb_parent;
391 unsigned long charge;
393 uprobe_start_dup_mmap();
394 down_write(&oldmm->mmap_sem);
395 flush_cache_dup_mm(oldmm);
396 uprobe_dup_mmap(oldmm, mm);
398 * Not linked in yet - no deadlock potential:
400 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
402 mm->total_vm = oldmm->total_vm;
403 mm->shared_vm = oldmm->shared_vm;
404 mm->exec_vm = oldmm->exec_vm;
405 mm->stack_vm = oldmm->stack_vm;
407 rb_link = &mm->mm_rb.rb_node;
410 retval = ksm_fork(mm, oldmm);
413 retval = khugepaged_fork(mm, oldmm);
418 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
421 if (mpnt->vm_flags & VM_DONTCOPY) {
422 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
427 if (mpnt->vm_flags & VM_ACCOUNT) {
428 unsigned long len = vma_pages(mpnt);
430 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
434 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
438 INIT_LIST_HEAD(&tmp->anon_vma_chain);
439 retval = vma_dup_policy(mpnt, tmp);
441 goto fail_nomem_policy;
443 if (anon_vma_fork(tmp, mpnt))
444 goto fail_nomem_anon_vma_fork;
445 tmp->vm_flags &= ~VM_LOCKED;
446 tmp->vm_next = tmp->vm_prev = NULL;
449 struct inode *inode = file_inode(file);
450 struct address_space *mapping = file->f_mapping;
453 if (tmp->vm_flags & VM_DENYWRITE)
454 atomic_dec(&inode->i_writecount);
455 i_mmap_lock_write(mapping);
456 if (tmp->vm_flags & VM_SHARED)
457 atomic_inc(&mapping->i_mmap_writable);
458 flush_dcache_mmap_lock(mapping);
459 /* insert tmp into the share list, just after mpnt */
460 vma_interval_tree_insert_after(tmp, mpnt,
462 flush_dcache_mmap_unlock(mapping);
463 i_mmap_unlock_write(mapping);
467 * Clear hugetlb-related page reserves for children. This only
468 * affects MAP_PRIVATE mappings. Faults generated by the child
469 * are not guaranteed to succeed, even if read-only
471 if (is_vm_hugetlb_page(tmp))
472 reset_vma_resv_huge_pages(tmp);
475 * Link in the new vma and copy the page table entries.
478 pprev = &tmp->vm_next;
482 __vma_link_rb(mm, tmp, rb_link, rb_parent);
483 rb_link = &tmp->vm_rb.rb_right;
484 rb_parent = &tmp->vm_rb;
487 retval = copy_page_range(mm, oldmm, mpnt);
489 if (tmp->vm_ops && tmp->vm_ops->open)
490 tmp->vm_ops->open(tmp);
495 /* a new mm has just been created */
496 arch_dup_mmap(oldmm, mm);
499 up_write(&mm->mmap_sem);
501 up_write(&oldmm->mmap_sem);
502 uprobe_end_dup_mmap();
504 fail_nomem_anon_vma_fork:
505 mpol_put(vma_policy(tmp));
507 kmem_cache_free(vm_area_cachep, tmp);
510 vm_unacct_memory(charge);
514 static inline int mm_alloc_pgd(struct mm_struct *mm)
516 mm->pgd = pgd_alloc(mm);
517 if (unlikely(!mm->pgd))
522 static inline void mm_free_pgd(struct mm_struct *mm)
524 pgd_free(mm, mm->pgd);
527 #define dup_mmap(mm, oldmm) (0)
528 #define mm_alloc_pgd(mm) (0)
529 #define mm_free_pgd(mm)
530 #endif /* CONFIG_MMU */
532 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
534 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
535 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
537 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
539 static int __init coredump_filter_setup(char *s)
541 default_dump_filter =
542 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
543 MMF_DUMP_FILTER_MASK;
547 __setup("coredump_filter=", coredump_filter_setup);
549 #include <linux/init_task.h>
551 static void mm_init_aio(struct mm_struct *mm)
554 spin_lock_init(&mm->ioctx_lock);
555 mm->ioctx_table = NULL;
559 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
566 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
570 mm->vmacache_seqnum = 0;
571 atomic_set(&mm->mm_users, 1);
572 atomic_set(&mm->mm_count, 1);
573 init_rwsem(&mm->mmap_sem);
574 INIT_LIST_HEAD(&mm->mmlist);
575 mm->core_state = NULL;
576 atomic_long_set(&mm->nr_ptes, 0);
581 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
582 spin_lock_init(&mm->page_table_lock);
585 mm_init_owner(mm, p);
586 mmu_notifier_mm_init(mm);
587 clear_tlb_flush_pending(mm);
588 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
589 mm->pmd_huge_pte = NULL;
593 mm->flags = current->mm->flags & MMF_INIT_MASK;
594 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
596 mm->flags = default_dump_filter;
600 if (mm_alloc_pgd(mm))
603 if (init_new_context(p, mm))
615 static void check_mm(struct mm_struct *mm)
619 for (i = 0; i < NR_MM_COUNTERS; i++) {
620 long x = atomic_long_read(&mm->rss_stat.count[i]);
623 printk(KERN_ALERT "BUG: Bad rss-counter state "
624 "mm:%p idx:%d val:%ld\n", mm, i, x);
627 if (atomic_long_read(&mm->nr_ptes))
628 pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
629 atomic_long_read(&mm->nr_ptes));
631 pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
634 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
635 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
640 * Allocate and initialize an mm_struct.
642 struct mm_struct *mm_alloc(void)
644 struct mm_struct *mm;
650 memset(mm, 0, sizeof(*mm));
651 return mm_init(mm, current);
655 * Called when the last reference to the mm
656 * is dropped: either by a lazy thread or by
657 * mmput. Free the page directory and the mm.
659 void __mmdrop(struct mm_struct *mm)
661 BUG_ON(mm == &init_mm);
664 mmu_notifier_mm_destroy(mm);
668 EXPORT_SYMBOL_GPL(__mmdrop);
671 * Decrement the use count and release all resources for an mm.
673 void mmput(struct mm_struct *mm)
677 if (atomic_dec_and_test(&mm->mm_users)) {
678 uprobe_clear_state(mm);
681 khugepaged_exit(mm); /* must run before exit_mmap */
683 set_mm_exe_file(mm, NULL);
684 if (!list_empty(&mm->mmlist)) {
685 spin_lock(&mmlist_lock);
686 list_del(&mm->mmlist);
687 spin_unlock(&mmlist_lock);
690 module_put(mm->binfmt->module);
694 EXPORT_SYMBOL_GPL(mmput);
696 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
699 get_file(new_exe_file);
702 mm->exe_file = new_exe_file;
705 struct file *get_mm_exe_file(struct mm_struct *mm)
707 struct file *exe_file;
709 /* We need mmap_sem to protect against races with removal of exe_file */
710 down_read(&mm->mmap_sem);
711 exe_file = mm->exe_file;
714 up_read(&mm->mmap_sem);
718 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
720 /* It's safe to write the exe_file pointer without exe_file_lock because
721 * this is called during fork when the task is not yet in /proc */
722 newmm->exe_file = get_mm_exe_file(oldmm);
726 * get_task_mm - acquire a reference to the task's mm
728 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
729 * this kernel workthread has transiently adopted a user mm with use_mm,
730 * to do its AIO) is not set and if so returns a reference to it, after
731 * bumping up the use count. User must release the mm via mmput()
732 * after use. Typically used by /proc and ptrace.
734 struct mm_struct *get_task_mm(struct task_struct *task)
736 struct mm_struct *mm;
741 if (task->flags & PF_KTHREAD)
744 atomic_inc(&mm->mm_users);
749 EXPORT_SYMBOL_GPL(get_task_mm);
751 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
753 struct mm_struct *mm;
756 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
760 mm = get_task_mm(task);
761 if (mm && mm != current->mm &&
762 !ptrace_may_access(task, mode)) {
764 mm = ERR_PTR(-EACCES);
766 mutex_unlock(&task->signal->cred_guard_mutex);
771 static void complete_vfork_done(struct task_struct *tsk)
773 struct completion *vfork;
776 vfork = tsk->vfork_done;
778 tsk->vfork_done = NULL;
784 static int wait_for_vfork_done(struct task_struct *child,
785 struct completion *vfork)
789 freezer_do_not_count();
790 killed = wait_for_completion_killable(vfork);
795 child->vfork_done = NULL;
799 put_task_struct(child);
803 /* Please note the differences between mmput and mm_release.
804 * mmput is called whenever we stop holding onto a mm_struct,
805 * error success whatever.
807 * mm_release is called after a mm_struct has been removed
808 * from the current process.
810 * This difference is important for error handling, when we
811 * only half set up a mm_struct for a new process and need to restore
812 * the old one. Because we mmput the new mm_struct before
813 * restoring the old one. . .
814 * Eric Biederman 10 January 1998
816 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
818 /* Get rid of any futexes when releasing the mm */
820 if (unlikely(tsk->robust_list)) {
821 exit_robust_list(tsk);
822 tsk->robust_list = NULL;
825 if (unlikely(tsk->compat_robust_list)) {
826 compat_exit_robust_list(tsk);
827 tsk->compat_robust_list = NULL;
830 if (unlikely(!list_empty(&tsk->pi_state_list)))
831 exit_pi_state_list(tsk);
834 uprobe_free_utask(tsk);
836 /* Get rid of any cached register state */
837 deactivate_mm(tsk, mm);
840 * If we're exiting normally, clear a user-space tid field if
841 * requested. We leave this alone when dying by signal, to leave
842 * the value intact in a core dump, and to save the unnecessary
843 * trouble, say, a killed vfork parent shouldn't touch this mm.
844 * Userland only wants this done for a sys_exit.
846 if (tsk->clear_child_tid) {
847 if (!(tsk->flags & PF_SIGNALED) &&
848 atomic_read(&mm->mm_users) > 1) {
850 * We don't check the error code - if userspace has
851 * not set up a proper pointer then tough luck.
853 put_user(0, tsk->clear_child_tid);
854 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
857 tsk->clear_child_tid = NULL;
861 * All done, finally we can wake up parent and return this mm to him.
862 * Also kthread_stop() uses this completion for synchronization.
865 complete_vfork_done(tsk);
869 * Allocate a new mm structure and copy contents from the
870 * mm structure of the passed in task structure.
872 static struct mm_struct *dup_mm(struct task_struct *tsk)
874 struct mm_struct *mm, *oldmm = current->mm;
881 memcpy(mm, oldmm, sizeof(*mm));
883 if (!mm_init(mm, tsk))
886 dup_mm_exe_file(oldmm, mm);
888 err = dup_mmap(mm, oldmm);
892 mm->hiwater_rss = get_mm_rss(mm);
893 mm->hiwater_vm = mm->total_vm;
895 if (mm->binfmt && !try_module_get(mm->binfmt->module))
901 /* don't put binfmt in mmput, we haven't got module yet */
909 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
911 struct mm_struct *mm, *oldmm;
914 tsk->min_flt = tsk->maj_flt = 0;
915 tsk->nvcsw = tsk->nivcsw = 0;
916 #ifdef CONFIG_DETECT_HUNG_TASK
917 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
921 tsk->active_mm = NULL;
924 * Are we cloning a kernel thread?
926 * We need to steal a active VM for that..
932 /* initialize the new vmacache entries */
935 if (clone_flags & CLONE_VM) {
936 atomic_inc(&oldmm->mm_users);
955 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
957 struct fs_struct *fs = current->fs;
958 if (clone_flags & CLONE_FS) {
959 /* tsk->fs is already what we want */
960 spin_lock(&fs->lock);
962 spin_unlock(&fs->lock);
966 spin_unlock(&fs->lock);
969 tsk->fs = copy_fs_struct(fs);
975 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
977 struct files_struct *oldf, *newf;
981 * A background process may not have any files ...
983 oldf = current->files;
987 if (clone_flags & CLONE_FILES) {
988 atomic_inc(&oldf->count);
992 newf = dup_fd(oldf, &error);
1002 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1005 struct io_context *ioc = current->io_context;
1006 struct io_context *new_ioc;
1011 * Share io context with parent, if CLONE_IO is set
1013 if (clone_flags & CLONE_IO) {
1015 tsk->io_context = ioc;
1016 } else if (ioprio_valid(ioc->ioprio)) {
1017 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1018 if (unlikely(!new_ioc))
1021 new_ioc->ioprio = ioc->ioprio;
1022 put_io_context(new_ioc);
1028 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1030 struct sighand_struct *sig;
1032 if (clone_flags & CLONE_SIGHAND) {
1033 atomic_inc(¤t->sighand->count);
1036 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1037 rcu_assign_pointer(tsk->sighand, sig);
1040 atomic_set(&sig->count, 1);
1041 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1045 void __cleanup_sighand(struct sighand_struct *sighand)
1047 if (atomic_dec_and_test(&sighand->count)) {
1048 signalfd_cleanup(sighand);
1050 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1051 * without an RCU grace period, see __lock_task_sighand().
1053 kmem_cache_free(sighand_cachep, sighand);
1058 * Initialize POSIX timer handling for a thread group.
1060 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1062 unsigned long cpu_limit;
1064 /* Thread group counters. */
1065 thread_group_cputime_init(sig);
1067 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1068 if (cpu_limit != RLIM_INFINITY) {
1069 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1070 sig->cputimer.running = 1;
1073 /* The timer lists. */
1074 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1075 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1076 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1079 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1081 struct signal_struct *sig;
1083 if (clone_flags & CLONE_THREAD)
1086 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1091 sig->nr_threads = 1;
1092 atomic_set(&sig->live, 1);
1093 atomic_set(&sig->sigcnt, 1);
1095 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1096 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1097 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1099 init_waitqueue_head(&sig->wait_chldexit);
1100 sig->curr_target = tsk;
1101 init_sigpending(&sig->shared_pending);
1102 INIT_LIST_HEAD(&sig->posix_timers);
1103 seqlock_init(&sig->stats_lock);
1105 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1106 sig->real_timer.function = it_real_fn;
1108 task_lock(current->group_leader);
1109 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1110 task_unlock(current->group_leader);
1112 posix_cpu_timers_init_group(sig);
1114 tty_audit_fork(sig);
1115 sched_autogroup_fork(sig);
1117 #ifdef CONFIG_CGROUPS
1118 init_rwsem(&sig->group_rwsem);
1121 sig->oom_score_adj = current->signal->oom_score_adj;
1122 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1124 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1125 current->signal->is_child_subreaper;
1127 mutex_init(&sig->cred_guard_mutex);
1132 static void copy_seccomp(struct task_struct *p)
1134 #ifdef CONFIG_SECCOMP
1136 * Must be called with sighand->lock held, which is common to
1137 * all threads in the group. Holding cred_guard_mutex is not
1138 * needed because this new task is not yet running and cannot
1141 assert_spin_locked(¤t->sighand->siglock);
1143 /* Ref-count the new filter user, and assign it. */
1144 get_seccomp_filter(current);
1145 p->seccomp = current->seccomp;
1148 * Explicitly enable no_new_privs here in case it got set
1149 * between the task_struct being duplicated and holding the
1150 * sighand lock. The seccomp state and nnp must be in sync.
1152 if (task_no_new_privs(current))
1153 task_set_no_new_privs(p);
1156 * If the parent gained a seccomp mode after copying thread
1157 * flags and between before we held the sighand lock, we have
1158 * to manually enable the seccomp thread flag here.
1160 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1161 set_tsk_thread_flag(p, TIF_SECCOMP);
1165 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1167 current->clear_child_tid = tidptr;
1169 return task_pid_vnr(current);
1172 static void rt_mutex_init_task(struct task_struct *p)
1174 raw_spin_lock_init(&p->pi_lock);
1175 #ifdef CONFIG_RT_MUTEXES
1176 p->pi_waiters = RB_ROOT;
1177 p->pi_waiters_leftmost = NULL;
1178 p->pi_blocked_on = NULL;
1183 * Initialize POSIX timer handling for a single task.
1185 static void posix_cpu_timers_init(struct task_struct *tsk)
1187 tsk->cputime_expires.prof_exp = 0;
1188 tsk->cputime_expires.virt_exp = 0;
1189 tsk->cputime_expires.sched_exp = 0;
1190 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1191 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1192 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1196 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1198 task->pids[type].pid = pid;
1202 * This creates a new process as a copy of the old one,
1203 * but does not actually start it yet.
1205 * It copies the registers, and all the appropriate
1206 * parts of the process environment (as per the clone
1207 * flags). The actual kick-off is left to the caller.
1209 static struct task_struct *copy_process(unsigned long clone_flags,
1210 unsigned long stack_start,
1211 unsigned long stack_size,
1212 int __user *child_tidptr,
1217 struct task_struct *p;
1219 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1220 return ERR_PTR(-EINVAL);
1222 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1223 return ERR_PTR(-EINVAL);
1226 * Thread groups must share signals as well, and detached threads
1227 * can only be started up within the thread group.
1229 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1230 return ERR_PTR(-EINVAL);
1233 * Shared signal handlers imply shared VM. By way of the above,
1234 * thread groups also imply shared VM. Blocking this case allows
1235 * for various simplifications in other code.
1237 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1238 return ERR_PTR(-EINVAL);
1241 * Siblings of global init remain as zombies on exit since they are
1242 * not reaped by their parent (swapper). To solve this and to avoid
1243 * multi-rooted process trees, prevent global and container-inits
1244 * from creating siblings.
1246 if ((clone_flags & CLONE_PARENT) &&
1247 current->signal->flags & SIGNAL_UNKILLABLE)
1248 return ERR_PTR(-EINVAL);
1251 * If the new process will be in a different pid or user namespace
1252 * do not allow it to share a thread group or signal handlers or
1253 * parent with the forking task.
1255 if (clone_flags & CLONE_SIGHAND) {
1256 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1257 (task_active_pid_ns(current) !=
1258 current->nsproxy->pid_ns_for_children))
1259 return ERR_PTR(-EINVAL);
1262 retval = security_task_create(clone_flags);
1267 p = dup_task_struct(current);
1271 ftrace_graph_init_task(p);
1273 rt_mutex_init_task(p);
1275 #ifdef CONFIG_PROVE_LOCKING
1276 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1277 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1280 if (atomic_read(&p->real_cred->user->processes) >=
1281 task_rlimit(p, RLIMIT_NPROC)) {
1282 if (p->real_cred->user != INIT_USER &&
1283 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1286 current->flags &= ~PF_NPROC_EXCEEDED;
1288 retval = copy_creds(p, clone_flags);
1293 * If multiple threads are within copy_process(), then this check
1294 * triggers too late. This doesn't hurt, the check is only there
1295 * to stop root fork bombs.
1298 if (nr_threads >= max_threads)
1299 goto bad_fork_cleanup_count;
1301 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1302 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1303 p->flags |= PF_FORKNOEXEC;
1304 INIT_LIST_HEAD(&p->children);
1305 INIT_LIST_HEAD(&p->sibling);
1306 rcu_copy_process(p);
1307 p->vfork_done = NULL;
1308 spin_lock_init(&p->alloc_lock);
1310 init_sigpending(&p->pending);
1312 p->utime = p->stime = p->gtime = 0;
1313 p->utimescaled = p->stimescaled = 0;
1314 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1315 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1317 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1318 seqlock_init(&p->vtime_seqlock);
1320 p->vtime_snap_whence = VTIME_SLEEPING;
1323 #if defined(SPLIT_RSS_COUNTING)
1324 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1327 p->default_timer_slack_ns = current->timer_slack_ns;
1329 task_io_accounting_init(&p->ioac);
1330 acct_clear_integrals(p);
1332 posix_cpu_timers_init(p);
1334 p->start_time = ktime_get_ns();
1335 p->real_start_time = ktime_get_boot_ns();
1336 p->io_context = NULL;
1337 p->audit_context = NULL;
1338 if (clone_flags & CLONE_THREAD)
1339 threadgroup_change_begin(current);
1342 p->mempolicy = mpol_dup(p->mempolicy);
1343 if (IS_ERR(p->mempolicy)) {
1344 retval = PTR_ERR(p->mempolicy);
1345 p->mempolicy = NULL;
1346 goto bad_fork_cleanup_threadgroup_lock;
1349 #ifdef CONFIG_CPUSETS
1350 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1351 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1352 seqcount_init(&p->mems_allowed_seq);
1354 #ifdef CONFIG_TRACE_IRQFLAGS
1356 p->hardirqs_enabled = 0;
1357 p->hardirq_enable_ip = 0;
1358 p->hardirq_enable_event = 0;
1359 p->hardirq_disable_ip = _THIS_IP_;
1360 p->hardirq_disable_event = 0;
1361 p->softirqs_enabled = 1;
1362 p->softirq_enable_ip = _THIS_IP_;
1363 p->softirq_enable_event = 0;
1364 p->softirq_disable_ip = 0;
1365 p->softirq_disable_event = 0;
1366 p->hardirq_context = 0;
1367 p->softirq_context = 0;
1369 #ifdef CONFIG_LOCKDEP
1370 p->lockdep_depth = 0; /* no locks held yet */
1371 p->curr_chain_key = 0;
1372 p->lockdep_recursion = 0;
1375 #ifdef CONFIG_DEBUG_MUTEXES
1376 p->blocked_on = NULL; /* not blocked yet */
1378 #ifdef CONFIG_BCACHE
1379 p->sequential_io = 0;
1380 p->sequential_io_avg = 0;
1383 /* Perform scheduler related setup. Assign this task to a CPU. */
1384 retval = sched_fork(clone_flags, p);
1386 goto bad_fork_cleanup_policy;
1388 retval = perf_event_init_task(p);
1390 goto bad_fork_cleanup_policy;
1391 retval = audit_alloc(p);
1393 goto bad_fork_cleanup_perf;
1394 /* copy all the process information */
1396 retval = copy_semundo(clone_flags, p);
1398 goto bad_fork_cleanup_audit;
1399 retval = copy_files(clone_flags, p);
1401 goto bad_fork_cleanup_semundo;
1402 retval = copy_fs(clone_flags, p);
1404 goto bad_fork_cleanup_files;
1405 retval = copy_sighand(clone_flags, p);
1407 goto bad_fork_cleanup_fs;
1408 retval = copy_signal(clone_flags, p);
1410 goto bad_fork_cleanup_sighand;
1411 retval = copy_mm(clone_flags, p);
1413 goto bad_fork_cleanup_signal;
1414 retval = copy_namespaces(clone_flags, p);
1416 goto bad_fork_cleanup_mm;
1417 retval = copy_io(clone_flags, p);
1419 goto bad_fork_cleanup_namespaces;
1420 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1422 goto bad_fork_cleanup_io;
1424 if (pid != &init_struct_pid) {
1425 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1427 retval = PTR_ERR(pid);
1428 goto bad_fork_cleanup_io;
1432 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1434 * Clear TID on mm_release()?
1436 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1441 p->robust_list = NULL;
1442 #ifdef CONFIG_COMPAT
1443 p->compat_robust_list = NULL;
1445 INIT_LIST_HEAD(&p->pi_state_list);
1446 p->pi_state_cache = NULL;
1449 * sigaltstack should be cleared when sharing the same VM
1451 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1452 p->sas_ss_sp = p->sas_ss_size = 0;
1455 * Syscall tracing and stepping should be turned off in the
1456 * child regardless of CLONE_PTRACE.
1458 user_disable_single_step(p);
1459 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1460 #ifdef TIF_SYSCALL_EMU
1461 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1463 clear_all_latency_tracing(p);
1465 /* ok, now we should be set up.. */
1466 p->pid = pid_nr(pid);
1467 if (clone_flags & CLONE_THREAD) {
1468 p->exit_signal = -1;
1469 p->group_leader = current->group_leader;
1470 p->tgid = current->tgid;
1472 if (clone_flags & CLONE_PARENT)
1473 p->exit_signal = current->group_leader->exit_signal;
1475 p->exit_signal = (clone_flags & CSIGNAL);
1476 p->group_leader = p;
1481 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1482 p->dirty_paused_when = 0;
1484 p->pdeath_signal = 0;
1485 INIT_LIST_HEAD(&p->thread_group);
1486 p->task_works = NULL;
1489 * Make it visible to the rest of the system, but dont wake it up yet.
1490 * Need tasklist lock for parent etc handling!
1492 write_lock_irq(&tasklist_lock);
1494 /* CLONE_PARENT re-uses the old parent */
1495 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1496 p->real_parent = current->real_parent;
1497 p->parent_exec_id = current->parent_exec_id;
1499 p->real_parent = current;
1500 p->parent_exec_id = current->self_exec_id;
1503 spin_lock(¤t->sighand->siglock);
1506 * Copy seccomp details explicitly here, in case they were changed
1507 * before holding sighand lock.
1512 * Process group and session signals need to be delivered to just the
1513 * parent before the fork or both the parent and the child after the
1514 * fork. Restart if a signal comes in before we add the new process to
1515 * it's process group.
1516 * A fatal signal pending means that current will exit, so the new
1517 * thread can't slip out of an OOM kill (or normal SIGKILL).
1519 recalc_sigpending();
1520 if (signal_pending(current)) {
1521 spin_unlock(¤t->sighand->siglock);
1522 write_unlock_irq(&tasklist_lock);
1523 retval = -ERESTARTNOINTR;
1524 goto bad_fork_free_pid;
1527 if (likely(p->pid)) {
1528 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1530 init_task_pid(p, PIDTYPE_PID, pid);
1531 if (thread_group_leader(p)) {
1532 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1533 init_task_pid(p, PIDTYPE_SID, task_session(current));
1535 if (is_child_reaper(pid)) {
1536 ns_of_pid(pid)->child_reaper = p;
1537 p->signal->flags |= SIGNAL_UNKILLABLE;
1540 p->signal->leader_pid = pid;
1541 p->signal->tty = tty_kref_get(current->signal->tty);
1542 list_add_tail(&p->sibling, &p->real_parent->children);
1543 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1544 attach_pid(p, PIDTYPE_PGID);
1545 attach_pid(p, PIDTYPE_SID);
1546 __this_cpu_inc(process_counts);
1548 current->signal->nr_threads++;
1549 atomic_inc(¤t->signal->live);
1550 atomic_inc(¤t->signal->sigcnt);
1551 list_add_tail_rcu(&p->thread_group,
1552 &p->group_leader->thread_group);
1553 list_add_tail_rcu(&p->thread_node,
1554 &p->signal->thread_head);
1556 attach_pid(p, PIDTYPE_PID);
1561 spin_unlock(¤t->sighand->siglock);
1562 syscall_tracepoint_update(p);
1563 write_unlock_irq(&tasklist_lock);
1565 proc_fork_connector(p);
1566 cgroup_post_fork(p);
1567 if (clone_flags & CLONE_THREAD)
1568 threadgroup_change_end(current);
1571 trace_task_newtask(p, clone_flags);
1572 uprobe_copy_process(p, clone_flags);
1577 if (pid != &init_struct_pid)
1579 bad_fork_cleanup_io:
1582 bad_fork_cleanup_namespaces:
1583 exit_task_namespaces(p);
1584 bad_fork_cleanup_mm:
1587 bad_fork_cleanup_signal:
1588 if (!(clone_flags & CLONE_THREAD))
1589 free_signal_struct(p->signal);
1590 bad_fork_cleanup_sighand:
1591 __cleanup_sighand(p->sighand);
1592 bad_fork_cleanup_fs:
1593 exit_fs(p); /* blocking */
1594 bad_fork_cleanup_files:
1595 exit_files(p); /* blocking */
1596 bad_fork_cleanup_semundo:
1598 bad_fork_cleanup_audit:
1600 bad_fork_cleanup_perf:
1601 perf_event_free_task(p);
1602 bad_fork_cleanup_policy:
1604 mpol_put(p->mempolicy);
1605 bad_fork_cleanup_threadgroup_lock:
1607 if (clone_flags & CLONE_THREAD)
1608 threadgroup_change_end(current);
1609 delayacct_tsk_free(p);
1610 bad_fork_cleanup_count:
1611 atomic_dec(&p->cred->user->processes);
1616 return ERR_PTR(retval);
1619 static inline void init_idle_pids(struct pid_link *links)
1623 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1624 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1625 links[type].pid = &init_struct_pid;
1629 struct task_struct *fork_idle(int cpu)
1631 struct task_struct *task;
1632 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1633 if (!IS_ERR(task)) {
1634 init_idle_pids(task->pids);
1635 init_idle(task, cpu);
1642 * Ok, this is the main fork-routine.
1644 * It copies the process, and if successful kick-starts
1645 * it and waits for it to finish using the VM if required.
1647 long do_fork(unsigned long clone_flags,
1648 unsigned long stack_start,
1649 unsigned long stack_size,
1650 int __user *parent_tidptr,
1651 int __user *child_tidptr)
1653 struct task_struct *p;
1658 * Determine whether and which event to report to ptracer. When
1659 * called from kernel_thread or CLONE_UNTRACED is explicitly
1660 * requested, no event is reported; otherwise, report if the event
1661 * for the type of forking is enabled.
1663 if (!(clone_flags & CLONE_UNTRACED)) {
1664 if (clone_flags & CLONE_VFORK)
1665 trace = PTRACE_EVENT_VFORK;
1666 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1667 trace = PTRACE_EVENT_CLONE;
1669 trace = PTRACE_EVENT_FORK;
1671 if (likely(!ptrace_event_enabled(current, trace)))
1675 p = copy_process(clone_flags, stack_start, stack_size,
1676 child_tidptr, NULL, trace);
1678 * Do this prior waking up the new thread - the thread pointer
1679 * might get invalid after that point, if the thread exits quickly.
1682 struct completion vfork;
1685 trace_sched_process_fork(current, p);
1687 pid = get_task_pid(p, PIDTYPE_PID);
1690 if (clone_flags & CLONE_PARENT_SETTID)
1691 put_user(nr, parent_tidptr);
1693 if (clone_flags & CLONE_VFORK) {
1694 p->vfork_done = &vfork;
1695 init_completion(&vfork);
1699 wake_up_new_task(p);
1701 /* forking complete and child started to run, tell ptracer */
1702 if (unlikely(trace))
1703 ptrace_event_pid(trace, pid);
1705 if (clone_flags & CLONE_VFORK) {
1706 if (!wait_for_vfork_done(p, &vfork))
1707 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1718 * Create a kernel thread.
1720 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1722 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1723 (unsigned long)arg, NULL, NULL);
1726 #ifdef __ARCH_WANT_SYS_FORK
1727 SYSCALL_DEFINE0(fork)
1730 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1732 /* can not support in nommu mode */
1738 #ifdef __ARCH_WANT_SYS_VFORK
1739 SYSCALL_DEFINE0(vfork)
1741 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1746 #ifdef __ARCH_WANT_SYS_CLONE
1747 #ifdef CONFIG_CLONE_BACKWARDS
1748 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1749 int __user *, parent_tidptr,
1751 int __user *, child_tidptr)
1752 #elif defined(CONFIG_CLONE_BACKWARDS2)
1753 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1754 int __user *, parent_tidptr,
1755 int __user *, child_tidptr,
1757 #elif defined(CONFIG_CLONE_BACKWARDS3)
1758 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1760 int __user *, parent_tidptr,
1761 int __user *, child_tidptr,
1764 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1765 int __user *, parent_tidptr,
1766 int __user *, child_tidptr,
1770 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1774 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1775 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1778 static void sighand_ctor(void *data)
1780 struct sighand_struct *sighand = data;
1782 spin_lock_init(&sighand->siglock);
1783 init_waitqueue_head(&sighand->signalfd_wqh);
1786 void __init proc_caches_init(void)
1788 sighand_cachep = kmem_cache_create("sighand_cache",
1789 sizeof(struct sighand_struct), 0,
1790 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1791 SLAB_NOTRACK, sighand_ctor);
1792 signal_cachep = kmem_cache_create("signal_cache",
1793 sizeof(struct signal_struct), 0,
1794 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1795 files_cachep = kmem_cache_create("files_cache",
1796 sizeof(struct files_struct), 0,
1797 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1798 fs_cachep = kmem_cache_create("fs_cache",
1799 sizeof(struct fs_struct), 0,
1800 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1802 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1803 * whole struct cpumask for the OFFSTACK case. We could change
1804 * this to *only* allocate as much of it as required by the
1805 * maximum number of CPU's we can ever have. The cpumask_allocation
1806 * is at the end of the structure, exactly for that reason.
1808 mm_cachep = kmem_cache_create("mm_struct",
1809 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1810 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1811 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1813 nsproxy_cache_init();
1817 * Check constraints on flags passed to the unshare system call.
1819 static int check_unshare_flags(unsigned long unshare_flags)
1821 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1822 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1823 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1824 CLONE_NEWUSER|CLONE_NEWPID))
1827 * Not implemented, but pretend it works if there is nothing to
1828 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1829 * needs to unshare vm.
1831 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1832 /* FIXME: get_task_mm() increments ->mm_users */
1833 if (atomic_read(¤t->mm->mm_users) > 1)
1841 * Unshare the filesystem structure if it is being shared
1843 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1845 struct fs_struct *fs = current->fs;
1847 if (!(unshare_flags & CLONE_FS) || !fs)
1850 /* don't need lock here; in the worst case we'll do useless copy */
1854 *new_fsp = copy_fs_struct(fs);
1862 * Unshare file descriptor table if it is being shared
1864 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1866 struct files_struct *fd = current->files;
1869 if ((unshare_flags & CLONE_FILES) &&
1870 (fd && atomic_read(&fd->count) > 1)) {
1871 *new_fdp = dup_fd(fd, &error);
1880 * unshare allows a process to 'unshare' part of the process
1881 * context which was originally shared using clone. copy_*
1882 * functions used by do_fork() cannot be used here directly
1883 * because they modify an inactive task_struct that is being
1884 * constructed. Here we are modifying the current, active,
1887 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1889 struct fs_struct *fs, *new_fs = NULL;
1890 struct files_struct *fd, *new_fd = NULL;
1891 struct cred *new_cred = NULL;
1892 struct nsproxy *new_nsproxy = NULL;
1897 * If unsharing a user namespace must also unshare the thread.
1899 if (unshare_flags & CLONE_NEWUSER)
1900 unshare_flags |= CLONE_THREAD | CLONE_FS;
1902 * If unsharing a thread from a thread group, must also unshare vm.
1904 if (unshare_flags & CLONE_THREAD)
1905 unshare_flags |= CLONE_VM;
1907 * If unsharing vm, must also unshare signal handlers.
1909 if (unshare_flags & CLONE_VM)
1910 unshare_flags |= CLONE_SIGHAND;
1912 * If unsharing namespace, must also unshare filesystem information.
1914 if (unshare_flags & CLONE_NEWNS)
1915 unshare_flags |= CLONE_FS;
1917 err = check_unshare_flags(unshare_flags);
1919 goto bad_unshare_out;
1921 * CLONE_NEWIPC must also detach from the undolist: after switching
1922 * to a new ipc namespace, the semaphore arrays from the old
1923 * namespace are unreachable.
1925 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1927 err = unshare_fs(unshare_flags, &new_fs);
1929 goto bad_unshare_out;
1930 err = unshare_fd(unshare_flags, &new_fd);
1932 goto bad_unshare_cleanup_fs;
1933 err = unshare_userns(unshare_flags, &new_cred);
1935 goto bad_unshare_cleanup_fd;
1936 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1939 goto bad_unshare_cleanup_cred;
1941 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1944 * CLONE_SYSVSEM is equivalent to sys_exit().
1948 if (unshare_flags & CLONE_NEWIPC) {
1949 /* Orphan segments in old ns (see sem above). */
1951 shm_init_task(current);
1955 switch_task_namespaces(current, new_nsproxy);
1961 spin_lock(&fs->lock);
1962 current->fs = new_fs;
1967 spin_unlock(&fs->lock);
1971 fd = current->files;
1972 current->files = new_fd;
1976 task_unlock(current);
1979 /* Install the new user namespace */
1980 commit_creds(new_cred);
1985 bad_unshare_cleanup_cred:
1988 bad_unshare_cleanup_fd:
1990 put_files_struct(new_fd);
1992 bad_unshare_cleanup_fs:
1994 free_fs_struct(new_fs);
2001 * Helper to unshare the files of the current task.
2002 * We don't want to expose copy_files internals to
2003 * the exec layer of the kernel.
2006 int unshare_files(struct files_struct **displaced)
2008 struct task_struct *task = current;
2009 struct files_struct *copy = NULL;
2012 error = unshare_fd(CLONE_FILES, ©);
2013 if (error || !copy) {
2017 *displaced = task->files;