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>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73 #include <linux/aio.h>
75 #include <asm/pgtable.h>
76 #include <asm/pgalloc.h>
77 #include <asm/uaccess.h>
78 #include <asm/mmu_context.h>
79 #include <asm/cacheflush.h>
80 #include <asm/tlbflush.h>
82 #include <trace/events/sched.h>
84 #define CREATE_TRACE_POINTS
85 #include <trace/events/task.h>
88 * Protected counters by write_lock_irq(&tasklist_lock)
90 unsigned long total_forks; /* Handle normal Linux uptimes. */
91 int nr_threads; /* The idle threads do not count.. */
93 int max_threads; /* tunable limit on nr_threads */
95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
99 #ifdef CONFIG_PROVE_RCU
100 int lockdep_tasklist_lock_is_held(void)
102 return lockdep_is_held(&tasklist_lock);
104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
105 #endif /* #ifdef CONFIG_PROVE_RCU */
107 int nr_processes(void)
112 for_each_possible_cpu(cpu)
113 total += per_cpu(process_counts, cpu);
118 void __weak arch_release_task_struct(struct task_struct *tsk)
122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
123 static struct kmem_cache *task_struct_cachep;
125 static inline struct task_struct *alloc_task_struct_node(int node)
127 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
130 static inline void free_task_struct(struct task_struct *tsk)
132 kmem_cache_free(task_struct_cachep, tsk);
136 void __weak arch_release_thread_info(struct thread_info *ti)
140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
143 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
144 * kmemcache based allocator.
146 # if THREAD_SIZE >= PAGE_SIZE
147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
150 struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
153 return page ? page_address(page) : NULL;
156 static inline void free_thread_info(struct thread_info *ti)
158 free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
161 static struct kmem_cache *thread_info_cache;
163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
166 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
169 static void free_thread_info(struct thread_info *ti)
171 kmem_cache_free(thread_info_cache, ti);
174 void thread_info_cache_init(void)
176 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
177 THREAD_SIZE, 0, NULL);
178 BUG_ON(thread_info_cache == NULL);
183 /* SLAB cache for signal_struct structures (tsk->signal) */
184 static struct kmem_cache *signal_cachep;
186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
187 struct kmem_cache *sighand_cachep;
189 /* SLAB cache for files_struct structures (tsk->files) */
190 struct kmem_cache *files_cachep;
192 /* SLAB cache for fs_struct structures (tsk->fs) */
193 struct kmem_cache *fs_cachep;
195 /* SLAB cache for vm_area_struct structures */
196 struct kmem_cache *vm_area_cachep;
198 /* SLAB cache for mm_struct structures (tsk->mm) */
199 static struct kmem_cache *mm_cachep;
201 /* Notifier list called when a task struct is freed */
202 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
204 static void account_kernel_stack(struct thread_info *ti, int account)
206 struct zone *zone = page_zone(virt_to_page(ti));
208 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
211 void free_task(struct task_struct *tsk)
213 account_kernel_stack(tsk->stack, -1);
214 arch_release_thread_info(tsk->stack);
215 free_thread_info(tsk->stack);
216 rt_mutex_debug_task_free(tsk);
217 ftrace_graph_exit_task(tsk);
218 put_seccomp_filter(tsk);
219 arch_release_task_struct(tsk);
220 free_task_struct(tsk);
222 EXPORT_SYMBOL(free_task);
224 static inline void free_signal_struct(struct signal_struct *sig)
226 taskstats_tgid_free(sig);
227 sched_autogroup_exit(sig);
228 kmem_cache_free(signal_cachep, sig);
231 static inline void put_signal_struct(struct signal_struct *sig)
233 if (atomic_dec_and_test(&sig->sigcnt))
234 free_signal_struct(sig);
237 int task_free_register(struct notifier_block *n)
239 return atomic_notifier_chain_register(&task_free_notifier, n);
241 EXPORT_SYMBOL(task_free_register);
243 int task_free_unregister(struct notifier_block *n)
245 return atomic_notifier_chain_unregister(&task_free_notifier, n);
247 EXPORT_SYMBOL(task_free_unregister);
249 void __put_task_struct(struct task_struct *tsk)
251 WARN_ON(!tsk->exit_state);
252 WARN_ON(atomic_read(&tsk->usage));
253 WARN_ON(tsk == current);
255 security_task_free(tsk);
257 delayacct_tsk_free(tsk);
258 put_signal_struct(tsk->signal);
260 atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
261 if (!profile_handoff_task(tsk))
264 EXPORT_SYMBOL_GPL(__put_task_struct);
266 void __init __weak arch_task_cache_init(void) { }
268 void __init fork_init(unsigned long mempages)
270 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
271 #ifndef ARCH_MIN_TASKALIGN
272 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
274 /* create a slab on which task_structs can be allocated */
276 kmem_cache_create("task_struct", sizeof(struct task_struct),
277 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
280 /* do the arch specific task caches init */
281 arch_task_cache_init();
284 * The default maximum number of threads is set to a safe
285 * value: the thread structures can take up at most half
288 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
291 * we need to allow at least 20 threads to boot a system
293 if (max_threads < 20)
296 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
297 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
298 init_task.signal->rlim[RLIMIT_SIGPENDING] =
299 init_task.signal->rlim[RLIMIT_NPROC];
302 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
303 struct task_struct *src)
309 static struct task_struct *dup_task_struct(struct task_struct *orig)
311 struct task_struct *tsk;
312 struct thread_info *ti;
313 unsigned long *stackend;
314 int node = tsk_fork_get_node(orig);
317 tsk = alloc_task_struct_node(node);
321 ti = alloc_thread_info_node(tsk, node);
325 err = arch_dup_task_struct(tsk, orig);
330 #ifdef CONFIG_SECCOMP
332 * We must handle setting up seccomp filters once we're under
333 * the sighand lock in case orig has changed between now and
334 * then. Until then, filter must be NULL to avoid messing up
335 * the usage counts on the error path calling free_task.
337 tsk->seccomp.filter = NULL;
340 setup_thread_stack(tsk, orig);
341 clear_user_return_notifier(tsk);
342 clear_tsk_need_resched(tsk);
343 stackend = end_of_stack(tsk);
344 *stackend = STACK_END_MAGIC; /* for overflow detection */
346 #ifdef CONFIG_CC_STACKPROTECTOR
347 tsk->stack_canary = get_random_int();
351 * One for us, one for whoever does the "release_task()" (usually
354 atomic_set(&tsk->usage, 2);
355 #ifdef CONFIG_BLK_DEV_IO_TRACE
358 tsk->splice_pipe = NULL;
359 tsk->task_frag.page = NULL;
361 account_kernel_stack(ti, 1);
366 free_thread_info(ti);
368 free_task_struct(tsk);
373 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
375 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
376 struct rb_node **rb_link, *rb_parent;
378 unsigned long charge;
379 struct mempolicy *pol;
381 uprobe_start_dup_mmap();
382 down_write(&oldmm->mmap_sem);
383 flush_cache_dup_mm(oldmm);
384 uprobe_dup_mmap(oldmm, mm);
386 * Not linked in yet - no deadlock potential:
388 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
392 mm->mmap_cache = NULL;
393 mm->free_area_cache = oldmm->mmap_base;
394 mm->cached_hole_size = ~0UL;
396 cpumask_clear(mm_cpumask(mm));
398 rb_link = &mm->mm_rb.rb_node;
401 retval = ksm_fork(mm, oldmm);
404 retval = khugepaged_fork(mm, oldmm);
409 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
412 if (mpnt->vm_flags & VM_DONTCOPY) {
413 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
418 if (mpnt->vm_flags & VM_ACCOUNT) {
419 unsigned long len = vma_pages(mpnt);
421 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
425 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
429 INIT_LIST_HEAD(&tmp->anon_vma_chain);
430 pol = mpol_dup(vma_policy(mpnt));
431 retval = PTR_ERR(pol);
433 goto fail_nomem_policy;
434 vma_set_policy(tmp, pol);
436 if (anon_vma_fork(tmp, mpnt))
437 goto fail_nomem_anon_vma_fork;
438 tmp->vm_flags &= ~VM_LOCKED;
439 tmp->vm_next = tmp->vm_prev = NULL;
442 struct inode *inode = file_inode(file);
443 struct address_space *mapping = file->f_mapping;
446 if (tmp->vm_flags & VM_DENYWRITE)
447 atomic_dec(&inode->i_writecount);
448 mutex_lock(&mapping->i_mmap_mutex);
449 if (tmp->vm_flags & VM_SHARED)
450 mapping->i_mmap_writable++;
451 flush_dcache_mmap_lock(mapping);
452 /* insert tmp into the share list, just after mpnt */
453 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
454 vma_nonlinear_insert(tmp,
455 &mapping->i_mmap_nonlinear);
457 vma_interval_tree_insert_after(tmp, mpnt,
459 flush_dcache_mmap_unlock(mapping);
460 mutex_unlock(&mapping->i_mmap_mutex);
464 * Clear hugetlb-related page reserves for children. This only
465 * affects MAP_PRIVATE mappings. Faults generated by the child
466 * are not guaranteed to succeed, even if read-only
468 if (is_vm_hugetlb_page(tmp))
469 reset_vma_resv_huge_pages(tmp);
472 * Link in the new vma and copy the page table entries.
475 pprev = &tmp->vm_next;
479 __vma_link_rb(mm, tmp, rb_link, rb_parent);
480 rb_link = &tmp->vm_rb.rb_right;
481 rb_parent = &tmp->vm_rb;
484 retval = copy_page_range(mm, oldmm, mpnt);
486 if (tmp->vm_ops && tmp->vm_ops->open)
487 tmp->vm_ops->open(tmp);
492 /* a new mm has just been created */
493 arch_dup_mmap(oldmm, mm);
496 up_write(&mm->mmap_sem);
498 up_write(&oldmm->mmap_sem);
499 uprobe_end_dup_mmap();
501 fail_nomem_anon_vma_fork:
504 kmem_cache_free(vm_area_cachep, tmp);
507 vm_unacct_memory(charge);
511 static inline int mm_alloc_pgd(struct mm_struct *mm)
513 mm->pgd = pgd_alloc(mm);
514 if (unlikely(!mm->pgd))
519 static inline void mm_free_pgd(struct mm_struct *mm)
521 pgd_free(mm, mm->pgd);
524 #define dup_mmap(mm, oldmm) (0)
525 #define mm_alloc_pgd(mm) (0)
526 #define mm_free_pgd(mm)
527 #endif /* CONFIG_MMU */
529 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
531 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
532 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
534 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
536 static int __init coredump_filter_setup(char *s)
538 default_dump_filter =
539 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
540 MMF_DUMP_FILTER_MASK;
544 __setup("coredump_filter=", coredump_filter_setup);
546 #include <linux/init_task.h>
548 static void mm_init_aio(struct mm_struct *mm)
551 spin_lock_init(&mm->ioctx_lock);
552 INIT_HLIST_HEAD(&mm->ioctx_list);
556 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
558 atomic_set(&mm->mm_users, 1);
559 atomic_set(&mm->mm_count, 1);
560 init_rwsem(&mm->mmap_sem);
561 INIT_LIST_HEAD(&mm->mmlist);
562 mm->flags = (current->mm) ?
563 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
564 mm->core_state = NULL;
566 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
567 spin_lock_init(&mm->page_table_lock);
568 mm->free_area_cache = TASK_UNMAPPED_BASE;
569 mm->cached_hole_size = ~0UL;
571 mm_init_owner(mm, p);
573 if (likely(!mm_alloc_pgd(mm))) {
575 mmu_notifier_mm_init(mm);
583 static void check_mm(struct mm_struct *mm)
587 for (i = 0; i < NR_MM_COUNTERS; i++) {
588 long x = atomic_long_read(&mm->rss_stat.count[i]);
591 printk(KERN_ALERT "BUG: Bad rss-counter state "
592 "mm:%p idx:%d val:%ld\n", mm, i, x);
595 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
596 VM_BUG_ON(mm->pmd_huge_pte);
601 * Allocate and initialize an mm_struct.
603 struct mm_struct *mm_alloc(void)
605 struct mm_struct *mm;
611 memset(mm, 0, sizeof(*mm));
613 return mm_init(mm, current);
617 * Called when the last reference to the mm
618 * is dropped: either by a lazy thread or by
619 * mmput. Free the page directory and the mm.
621 void __mmdrop(struct mm_struct *mm)
623 BUG_ON(mm == &init_mm);
626 mmu_notifier_mm_destroy(mm);
630 EXPORT_SYMBOL_GPL(__mmdrop);
633 * Decrement the use count and release all resources for an mm.
635 void mmput(struct mm_struct *mm)
639 if (atomic_dec_and_test(&mm->mm_users)) {
640 uprobe_clear_state(mm);
643 khugepaged_exit(mm); /* must run before exit_mmap */
645 set_mm_exe_file(mm, NULL);
646 if (!list_empty(&mm->mmlist)) {
647 spin_lock(&mmlist_lock);
648 list_del(&mm->mmlist);
649 spin_unlock(&mmlist_lock);
652 module_put(mm->binfmt->module);
656 EXPORT_SYMBOL_GPL(mmput);
658 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
661 get_file(new_exe_file);
664 mm->exe_file = new_exe_file;
667 struct file *get_mm_exe_file(struct mm_struct *mm)
669 struct file *exe_file;
671 /* We need mmap_sem to protect against races with removal of exe_file */
672 down_read(&mm->mmap_sem);
673 exe_file = mm->exe_file;
676 up_read(&mm->mmap_sem);
680 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
682 /* It's safe to write the exe_file pointer without exe_file_lock because
683 * this is called during fork when the task is not yet in /proc */
684 newmm->exe_file = get_mm_exe_file(oldmm);
688 * get_task_mm - acquire a reference to the task's mm
690 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
691 * this kernel workthread has transiently adopted a user mm with use_mm,
692 * to do its AIO) is not set and if so returns a reference to it, after
693 * bumping up the use count. User must release the mm via mmput()
694 * after use. Typically used by /proc and ptrace.
696 struct mm_struct *get_task_mm(struct task_struct *task)
698 struct mm_struct *mm;
703 if (task->flags & PF_KTHREAD)
706 atomic_inc(&mm->mm_users);
711 EXPORT_SYMBOL_GPL(get_task_mm);
713 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
715 struct mm_struct *mm;
718 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
722 mm = get_task_mm(task);
723 if (mm && mm != current->mm &&
724 !ptrace_may_access(task, mode) &&
725 !capable(CAP_SYS_RESOURCE)) {
727 mm = ERR_PTR(-EACCES);
729 mutex_unlock(&task->signal->cred_guard_mutex);
734 static void complete_vfork_done(struct task_struct *tsk)
736 struct completion *vfork;
739 vfork = tsk->vfork_done;
741 tsk->vfork_done = NULL;
747 static int wait_for_vfork_done(struct task_struct *child,
748 struct completion *vfork)
752 freezer_do_not_count();
753 killed = wait_for_completion_killable(vfork);
758 child->vfork_done = NULL;
762 put_task_struct(child);
766 /* Please note the differences between mmput and mm_release.
767 * mmput is called whenever we stop holding onto a mm_struct,
768 * error success whatever.
770 * mm_release is called after a mm_struct has been removed
771 * from the current process.
773 * This difference is important for error handling, when we
774 * only half set up a mm_struct for a new process and need to restore
775 * the old one. Because we mmput the new mm_struct before
776 * restoring the old one. . .
777 * Eric Biederman 10 January 1998
779 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
781 /* Get rid of any futexes when releasing the mm */
783 if (unlikely(tsk->robust_list)) {
784 exit_robust_list(tsk);
785 tsk->robust_list = NULL;
788 if (unlikely(tsk->compat_robust_list)) {
789 compat_exit_robust_list(tsk);
790 tsk->compat_robust_list = NULL;
793 if (unlikely(!list_empty(&tsk->pi_state_list)))
794 exit_pi_state_list(tsk);
797 uprobe_free_utask(tsk);
799 /* Get rid of any cached register state */
800 deactivate_mm(tsk, mm);
803 * If we're exiting normally, clear a user-space tid field if
804 * requested. We leave this alone when dying by signal, to leave
805 * the value intact in a core dump, and to save the unnecessary
806 * trouble, say, a killed vfork parent shouldn't touch this mm.
807 * Userland only wants this done for a sys_exit.
809 if (tsk->clear_child_tid) {
810 if (!(tsk->flags & PF_SIGNALED) &&
811 atomic_read(&mm->mm_users) > 1) {
813 * We don't check the error code - if userspace has
814 * not set up a proper pointer then tough luck.
816 put_user(0, tsk->clear_child_tid);
817 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
820 tsk->clear_child_tid = NULL;
824 * All done, finally we can wake up parent and return this mm to him.
825 * Also kthread_stop() uses this completion for synchronization.
828 complete_vfork_done(tsk);
832 * Allocate a new mm structure and copy contents from the
833 * mm structure of the passed in task structure.
835 struct mm_struct *dup_mm(struct task_struct *tsk)
837 struct mm_struct *mm, *oldmm = current->mm;
847 memcpy(mm, oldmm, sizeof(*mm));
850 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
851 mm->pmd_huge_pte = NULL;
853 #ifdef CONFIG_NUMA_BALANCING
854 mm->first_nid = NUMA_PTE_SCAN_INIT;
856 if (!mm_init(mm, tsk))
859 if (init_new_context(tsk, mm))
862 dup_mm_exe_file(oldmm, mm);
864 err = dup_mmap(mm, oldmm);
868 mm->hiwater_rss = get_mm_rss(mm);
869 mm->hiwater_vm = mm->total_vm;
871 if (mm->binfmt && !try_module_get(mm->binfmt->module))
877 /* don't put binfmt in mmput, we haven't got module yet */
886 * If init_new_context() failed, we cannot use mmput() to free the mm
887 * because it calls destroy_context()
894 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
896 struct mm_struct *mm, *oldmm;
899 tsk->min_flt = tsk->maj_flt = 0;
900 tsk->nvcsw = tsk->nivcsw = 0;
901 #ifdef CONFIG_DETECT_HUNG_TASK
902 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
906 tsk->active_mm = NULL;
909 * Are we cloning a kernel thread?
911 * We need to steal a active VM for that..
917 if (clone_flags & CLONE_VM) {
918 atomic_inc(&oldmm->mm_users);
937 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
939 struct fs_struct *fs = current->fs;
940 if (clone_flags & CLONE_FS) {
941 /* tsk->fs is already what we want */
942 spin_lock(&fs->lock);
944 spin_unlock(&fs->lock);
948 spin_unlock(&fs->lock);
951 tsk->fs = copy_fs_struct(fs);
957 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
959 struct files_struct *oldf, *newf;
963 * A background process may not have any files ...
965 oldf = current->files;
969 if (clone_flags & CLONE_FILES) {
970 atomic_inc(&oldf->count);
974 newf = dup_fd(oldf, &error);
984 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
987 struct io_context *ioc = current->io_context;
988 struct io_context *new_ioc;
993 * Share io context with parent, if CLONE_IO is set
995 if (clone_flags & CLONE_IO) {
997 tsk->io_context = ioc;
998 } else if (ioprio_valid(ioc->ioprio)) {
999 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1000 if (unlikely(!new_ioc))
1003 new_ioc->ioprio = ioc->ioprio;
1004 put_io_context(new_ioc);
1010 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1012 struct sighand_struct *sig;
1014 if (clone_flags & CLONE_SIGHAND) {
1015 atomic_inc(¤t->sighand->count);
1018 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1019 rcu_assign_pointer(tsk->sighand, sig);
1022 atomic_set(&sig->count, 1);
1023 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1027 void __cleanup_sighand(struct sighand_struct *sighand)
1029 if (atomic_dec_and_test(&sighand->count)) {
1030 signalfd_cleanup(sighand);
1031 kmem_cache_free(sighand_cachep, sighand);
1037 * Initialize POSIX timer handling for a thread group.
1039 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1041 unsigned long cpu_limit;
1043 /* Thread group counters. */
1044 thread_group_cputime_init(sig);
1046 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1047 if (cpu_limit != RLIM_INFINITY) {
1048 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1049 sig->cputimer.running = 1;
1052 /* The timer lists. */
1053 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1054 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1055 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1058 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1060 struct signal_struct *sig;
1062 if (clone_flags & CLONE_THREAD)
1065 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1070 sig->nr_threads = 1;
1071 atomic_set(&sig->live, 1);
1072 atomic_set(&sig->sigcnt, 1);
1073 init_waitqueue_head(&sig->wait_chldexit);
1074 sig->curr_target = tsk;
1075 init_sigpending(&sig->shared_pending);
1076 INIT_LIST_HEAD(&sig->posix_timers);
1078 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1079 sig->real_timer.function = it_real_fn;
1081 task_lock(current->group_leader);
1082 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1083 task_unlock(current->group_leader);
1085 posix_cpu_timers_init_group(sig);
1087 tty_audit_fork(sig);
1088 sched_autogroup_fork(sig);
1090 #ifdef CONFIG_CGROUPS
1091 init_rwsem(&sig->group_rwsem);
1094 sig->oom_score_adj = current->signal->oom_score_adj;
1095 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1097 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1098 current->signal->is_child_subreaper;
1100 mutex_init(&sig->cred_guard_mutex);
1105 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1107 unsigned long new_flags = p->flags;
1109 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1110 new_flags |= PF_FORKNOEXEC;
1111 p->flags = new_flags;
1114 static void copy_seccomp(struct task_struct *p)
1116 #ifdef CONFIG_SECCOMP
1118 * Must be called with sighand->lock held, which is common to
1119 * all threads in the group. Holding cred_guard_mutex is not
1120 * needed because this new task is not yet running and cannot
1123 BUG_ON(!spin_is_locked(¤t->sighand->siglock));
1125 /* Ref-count the new filter user, and assign it. */
1126 get_seccomp_filter(current);
1127 p->seccomp = current->seccomp;
1130 * Explicitly enable no_new_privs here in case it got set
1131 * between the task_struct being duplicated and holding the
1132 * sighand lock. The seccomp state and nnp must be in sync.
1134 if (task_no_new_privs(current))
1135 task_set_no_new_privs(p);
1138 * If the parent gained a seccomp mode after copying thread
1139 * flags and between before we held the sighand lock, we have
1140 * to manually enable the seccomp thread flag here.
1142 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1143 set_tsk_thread_flag(p, TIF_SECCOMP);
1147 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1149 current->clear_child_tid = tidptr;
1151 return task_pid_vnr(current);
1154 static void rt_mutex_init_task(struct task_struct *p)
1156 raw_spin_lock_init(&p->pi_lock);
1157 #ifdef CONFIG_RT_MUTEXES
1158 plist_head_init(&p->pi_waiters);
1159 p->pi_blocked_on = NULL;
1163 #ifdef CONFIG_MM_OWNER
1164 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1168 #endif /* CONFIG_MM_OWNER */
1171 * Initialize POSIX timer handling for a single task.
1173 static void posix_cpu_timers_init(struct task_struct *tsk)
1175 tsk->cputime_expires.prof_exp = 0;
1176 tsk->cputime_expires.virt_exp = 0;
1177 tsk->cputime_expires.sched_exp = 0;
1178 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1179 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1180 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1184 * This creates a new process as a copy of the old one,
1185 * but does not actually start it yet.
1187 * It copies the registers, and all the appropriate
1188 * parts of the process environment (as per the clone
1189 * flags). The actual kick-off is left to the caller.
1191 static struct task_struct *copy_process(unsigned long clone_flags,
1192 unsigned long stack_start,
1193 unsigned long stack_size,
1194 int __user *child_tidptr,
1199 struct task_struct *p;
1201 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1202 return ERR_PTR(-EINVAL);
1204 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1205 return ERR_PTR(-EINVAL);
1208 * Thread groups must share signals as well, and detached threads
1209 * can only be started up within the thread group.
1211 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1212 return ERR_PTR(-EINVAL);
1215 * Shared signal handlers imply shared VM. By way of the above,
1216 * thread groups also imply shared VM. Blocking this case allows
1217 * for various simplifications in other code.
1219 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1220 return ERR_PTR(-EINVAL);
1223 * Siblings of global init remain as zombies on exit since they are
1224 * not reaped by their parent (swapper). To solve this and to avoid
1225 * multi-rooted process trees, prevent global and container-inits
1226 * from creating siblings.
1228 if ((clone_flags & CLONE_PARENT) &&
1229 current->signal->flags & SIGNAL_UNKILLABLE)
1230 return ERR_PTR(-EINVAL);
1233 * If the new process will be in a different pid namespace
1234 * don't allow the creation of threads.
1236 if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
1237 (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1238 return ERR_PTR(-EINVAL);
1240 retval = security_task_create(clone_flags);
1245 p = dup_task_struct(current);
1249 ftrace_graph_init_task(p);
1251 rt_mutex_init_task(p);
1253 #ifdef CONFIG_PROVE_LOCKING
1254 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1255 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1258 if (atomic_read(&p->real_cred->user->processes) >=
1259 task_rlimit(p, RLIMIT_NPROC)) {
1260 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1261 p->real_cred->user != INIT_USER)
1264 current->flags &= ~PF_NPROC_EXCEEDED;
1266 retval = copy_creds(p, clone_flags);
1271 * If multiple threads are within copy_process(), then this check
1272 * triggers too late. This doesn't hurt, the check is only there
1273 * to stop root fork bombs.
1276 if (nr_threads >= max_threads)
1277 goto bad_fork_cleanup_count;
1279 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1280 goto bad_fork_cleanup_count;
1283 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1284 copy_flags(clone_flags, p);
1285 INIT_LIST_HEAD(&p->children);
1286 INIT_LIST_HEAD(&p->sibling);
1287 rcu_copy_process(p);
1288 p->vfork_done = NULL;
1289 spin_lock_init(&p->alloc_lock);
1291 init_sigpending(&p->pending);
1293 p->utime = p->stime = p->gtime = 0;
1294 p->utimescaled = p->stimescaled = 0;
1295 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1296 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1298 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1299 seqlock_init(&p->vtime_seqlock);
1301 p->vtime_snap_whence = VTIME_SLEEPING;
1304 #if defined(SPLIT_RSS_COUNTING)
1305 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1308 p->default_timer_slack_ns = current->timer_slack_ns;
1310 task_io_accounting_init(&p->ioac);
1311 acct_clear_integrals(p);
1313 posix_cpu_timers_init(p);
1315 do_posix_clock_monotonic_gettime(&p->start_time);
1316 p->real_start_time = p->start_time;
1317 monotonic_to_bootbased(&p->real_start_time);
1318 p->io_context = NULL;
1319 p->audit_context = NULL;
1320 if (clone_flags & CLONE_THREAD)
1321 threadgroup_change_begin(current);
1324 p->mempolicy = mpol_dup(p->mempolicy);
1325 if (IS_ERR(p->mempolicy)) {
1326 retval = PTR_ERR(p->mempolicy);
1327 p->mempolicy = NULL;
1328 goto bad_fork_cleanup_cgroup;
1330 mpol_fix_fork_child_flag(p);
1332 #ifdef CONFIG_CPUSETS
1333 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1334 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1335 seqcount_init(&p->mems_allowed_seq);
1337 #ifdef CONFIG_TRACE_IRQFLAGS
1339 p->hardirqs_enabled = 0;
1340 p->hardirq_enable_ip = 0;
1341 p->hardirq_enable_event = 0;
1342 p->hardirq_disable_ip = _THIS_IP_;
1343 p->hardirq_disable_event = 0;
1344 p->softirqs_enabled = 1;
1345 p->softirq_enable_ip = _THIS_IP_;
1346 p->softirq_enable_event = 0;
1347 p->softirq_disable_ip = 0;
1348 p->softirq_disable_event = 0;
1349 p->hardirq_context = 0;
1350 p->softirq_context = 0;
1352 #ifdef CONFIG_LOCKDEP
1353 p->lockdep_depth = 0; /* no locks held yet */
1354 p->curr_chain_key = 0;
1355 p->lockdep_recursion = 0;
1358 #ifdef CONFIG_DEBUG_MUTEXES
1359 p->blocked_on = NULL; /* not blocked yet */
1362 p->memcg_batch.do_batch = 0;
1363 p->memcg_batch.memcg = NULL;
1365 #ifdef CONFIG_BCACHE
1366 p->sequential_io = 0;
1367 p->sequential_io_avg = 0;
1370 /* Perform scheduler related setup. Assign this task to a CPU. */
1373 retval = perf_event_init_task(p);
1375 goto bad_fork_cleanup_policy;
1376 retval = audit_alloc(p);
1378 goto bad_fork_cleanup_policy;
1379 /* copy all the process information */
1380 retval = copy_semundo(clone_flags, p);
1382 goto bad_fork_cleanup_audit;
1383 retval = copy_files(clone_flags, p);
1385 goto bad_fork_cleanup_semundo;
1386 retval = copy_fs(clone_flags, p);
1388 goto bad_fork_cleanup_files;
1389 retval = copy_sighand(clone_flags, p);
1391 goto bad_fork_cleanup_fs;
1392 retval = copy_signal(clone_flags, p);
1394 goto bad_fork_cleanup_sighand;
1395 retval = copy_mm(clone_flags, p);
1397 goto bad_fork_cleanup_signal;
1398 retval = copy_namespaces(clone_flags, p);
1400 goto bad_fork_cleanup_mm;
1401 retval = copy_io(clone_flags, p);
1403 goto bad_fork_cleanup_namespaces;
1404 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1406 goto bad_fork_cleanup_io;
1408 if (pid != &init_struct_pid) {
1410 pid = alloc_pid(p->nsproxy->pid_ns);
1412 goto bad_fork_cleanup_io;
1415 p->pid = pid_nr(pid);
1417 if (clone_flags & CLONE_THREAD)
1418 p->tgid = current->tgid;
1420 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1422 * Clear TID on mm_release()?
1424 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1429 p->robust_list = NULL;
1430 #ifdef CONFIG_COMPAT
1431 p->compat_robust_list = NULL;
1433 INIT_LIST_HEAD(&p->pi_state_list);
1434 p->pi_state_cache = NULL;
1436 uprobe_copy_process(p);
1438 * sigaltstack should be cleared when sharing the same VM
1440 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1441 p->sas_ss_sp = p->sas_ss_size = 0;
1444 * Syscall tracing and stepping should be turned off in the
1445 * child regardless of CLONE_PTRACE.
1447 user_disable_single_step(p);
1448 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1449 #ifdef TIF_SYSCALL_EMU
1450 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1452 clear_all_latency_tracing(p);
1454 /* ok, now we should be set up.. */
1455 if (clone_flags & CLONE_THREAD)
1456 p->exit_signal = -1;
1457 else if (clone_flags & CLONE_PARENT)
1458 p->exit_signal = current->group_leader->exit_signal;
1460 p->exit_signal = (clone_flags & CSIGNAL);
1462 p->pdeath_signal = 0;
1466 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1467 p->dirty_paused_when = 0;
1470 * Ok, make it visible to the rest of the system.
1471 * We dont wake it up yet.
1473 p->group_leader = p;
1474 INIT_LIST_HEAD(&p->thread_group);
1475 p->task_works = NULL;
1477 /* Need tasklist lock for parent etc handling! */
1478 write_lock_irq(&tasklist_lock);
1480 /* CLONE_PARENT re-uses the old parent */
1481 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1482 p->real_parent = current->real_parent;
1483 p->parent_exec_id = current->parent_exec_id;
1485 p->real_parent = current;
1486 p->parent_exec_id = current->self_exec_id;
1489 spin_lock(¤t->sighand->siglock);
1492 * Copy seccomp details explicitly here, in case they were changed
1493 * before holding sighand lock.
1498 * Process group and session signals need to be delivered to just the
1499 * parent before the fork or both the parent and the child after the
1500 * fork. Restart if a signal comes in before we add the new process to
1501 * it's process group.
1502 * A fatal signal pending means that current will exit, so the new
1503 * thread can't slip out of an OOM kill (or normal SIGKILL).
1505 recalc_sigpending();
1506 if (signal_pending(current)) {
1507 spin_unlock(¤t->sighand->siglock);
1508 write_unlock_irq(&tasklist_lock);
1509 retval = -ERESTARTNOINTR;
1510 goto bad_fork_free_pid;
1513 if (clone_flags & CLONE_THREAD) {
1514 current->signal->nr_threads++;
1515 atomic_inc(¤t->signal->live);
1516 atomic_inc(¤t->signal->sigcnt);
1517 p->group_leader = current->group_leader;
1518 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1521 if (likely(p->pid)) {
1522 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1524 if (thread_group_leader(p)) {
1525 if (is_child_reaper(pid)) {
1526 ns_of_pid(pid)->child_reaper = p;
1527 p->signal->flags |= SIGNAL_UNKILLABLE;
1530 p->signal->leader_pid = pid;
1531 p->signal->tty = tty_kref_get(current->signal->tty);
1532 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1533 attach_pid(p, PIDTYPE_SID, task_session(current));
1534 list_add_tail(&p->sibling, &p->real_parent->children);
1535 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1536 __this_cpu_inc(process_counts);
1538 attach_pid(p, PIDTYPE_PID, pid);
1543 spin_unlock(¤t->sighand->siglock);
1544 write_unlock_irq(&tasklist_lock);
1545 proc_fork_connector(p);
1546 cgroup_post_fork(p);
1547 if (clone_flags & CLONE_THREAD)
1548 threadgroup_change_end(current);
1551 trace_task_newtask(p, clone_flags);
1556 if (pid != &init_struct_pid)
1558 bad_fork_cleanup_io:
1561 bad_fork_cleanup_namespaces:
1562 exit_task_namespaces(p);
1563 bad_fork_cleanup_mm:
1566 bad_fork_cleanup_signal:
1567 if (!(clone_flags & CLONE_THREAD))
1568 free_signal_struct(p->signal);
1569 bad_fork_cleanup_sighand:
1570 __cleanup_sighand(p->sighand);
1571 bad_fork_cleanup_fs:
1572 exit_fs(p); /* blocking */
1573 bad_fork_cleanup_files:
1574 exit_files(p); /* blocking */
1575 bad_fork_cleanup_semundo:
1577 bad_fork_cleanup_audit:
1579 bad_fork_cleanup_policy:
1580 perf_event_free_task(p);
1582 mpol_put(p->mempolicy);
1583 bad_fork_cleanup_cgroup:
1585 if (clone_flags & CLONE_THREAD)
1586 threadgroup_change_end(current);
1588 delayacct_tsk_free(p);
1589 module_put(task_thread_info(p)->exec_domain->module);
1590 bad_fork_cleanup_count:
1591 atomic_dec(&p->cred->user->processes);
1596 return ERR_PTR(retval);
1599 static inline void init_idle_pids(struct pid_link *links)
1603 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1604 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1605 links[type].pid = &init_struct_pid;
1609 struct task_struct * __cpuinit fork_idle(int cpu)
1611 struct task_struct *task;
1612 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1613 if (!IS_ERR(task)) {
1614 init_idle_pids(task->pids);
1615 init_idle(task, cpu);
1622 * Ok, this is the main fork-routine.
1624 * It copies the process, and if successful kick-starts
1625 * it and waits for it to finish using the VM if required.
1627 long do_fork(unsigned long clone_flags,
1628 unsigned long stack_start,
1629 unsigned long stack_size,
1630 int __user *parent_tidptr,
1631 int __user *child_tidptr)
1633 struct task_struct *p;
1638 * Do some preliminary argument and permissions checking before we
1639 * actually start allocating stuff
1641 if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1642 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1647 * Determine whether and which event to report to ptracer. When
1648 * called from kernel_thread or CLONE_UNTRACED is explicitly
1649 * requested, no event is reported; otherwise, report if the event
1650 * for the type of forking is enabled.
1652 if (!(clone_flags & CLONE_UNTRACED)) {
1653 if (clone_flags & CLONE_VFORK)
1654 trace = PTRACE_EVENT_VFORK;
1655 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1656 trace = PTRACE_EVENT_CLONE;
1658 trace = PTRACE_EVENT_FORK;
1660 if (likely(!ptrace_event_enabled(current, trace)))
1664 p = copy_process(clone_flags, stack_start, stack_size,
1665 child_tidptr, NULL, trace);
1667 * Do this prior waking up the new thread - the thread pointer
1668 * might get invalid after that point, if the thread exits quickly.
1671 struct completion vfork;
1673 trace_sched_process_fork(current, p);
1675 nr = task_pid_vnr(p);
1677 if (clone_flags & CLONE_PARENT_SETTID)
1678 put_user(nr, parent_tidptr);
1680 if (clone_flags & CLONE_VFORK) {
1681 p->vfork_done = &vfork;
1682 init_completion(&vfork);
1686 wake_up_new_task(p);
1688 /* forking complete and child started to run, tell ptracer */
1689 if (unlikely(trace))
1690 ptrace_event(trace, nr);
1692 if (clone_flags & CLONE_VFORK) {
1693 if (!wait_for_vfork_done(p, &vfork))
1694 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1703 * Create a kernel thread.
1705 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1707 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1708 (unsigned long)arg, NULL, NULL);
1711 #ifdef __ARCH_WANT_SYS_FORK
1712 SYSCALL_DEFINE0(fork)
1715 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1717 /* can not support in nommu mode */
1723 #ifdef __ARCH_WANT_SYS_VFORK
1724 SYSCALL_DEFINE0(vfork)
1726 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1731 #ifdef __ARCH_WANT_SYS_CLONE
1732 #ifdef CONFIG_CLONE_BACKWARDS
1733 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1734 int __user *, parent_tidptr,
1736 int __user *, child_tidptr)
1737 #elif defined(CONFIG_CLONE_BACKWARDS2)
1738 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1739 int __user *, parent_tidptr,
1740 int __user *, child_tidptr,
1743 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1744 int __user *, parent_tidptr,
1745 int __user *, child_tidptr,
1749 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1753 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1754 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1757 static void sighand_ctor(void *data)
1759 struct sighand_struct *sighand = data;
1761 spin_lock_init(&sighand->siglock);
1762 init_waitqueue_head(&sighand->signalfd_wqh);
1765 void __init proc_caches_init(void)
1767 sighand_cachep = kmem_cache_create("sighand_cache",
1768 sizeof(struct sighand_struct), 0,
1769 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1770 SLAB_NOTRACK, sighand_ctor);
1771 signal_cachep = kmem_cache_create("signal_cache",
1772 sizeof(struct signal_struct), 0,
1773 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1774 files_cachep = kmem_cache_create("files_cache",
1775 sizeof(struct files_struct), 0,
1776 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1777 fs_cachep = kmem_cache_create("fs_cache",
1778 sizeof(struct fs_struct), 0,
1779 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1781 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1782 * whole struct cpumask for the OFFSTACK case. We could change
1783 * this to *only* allocate as much of it as required by the
1784 * maximum number of CPU's we can ever have. The cpumask_allocation
1785 * is at the end of the structure, exactly for that reason.
1787 mm_cachep = kmem_cache_create("mm_struct",
1788 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1789 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1790 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1792 nsproxy_cache_init();
1796 * Check constraints on flags passed to the unshare system call.
1798 static int check_unshare_flags(unsigned long unshare_flags)
1800 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1801 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1802 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1803 CLONE_NEWUSER|CLONE_NEWPID))
1806 * Not implemented, but pretend it works if there is nothing to
1807 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1808 * needs to unshare vm.
1810 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1811 /* FIXME: get_task_mm() increments ->mm_users */
1812 if (atomic_read(¤t->mm->mm_users) > 1)
1820 * Unshare the filesystem structure if it is being shared
1822 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1824 struct fs_struct *fs = current->fs;
1826 if (!(unshare_flags & CLONE_FS) || !fs)
1829 /* don't need lock here; in the worst case we'll do useless copy */
1833 *new_fsp = copy_fs_struct(fs);
1841 * Unshare file descriptor table if it is being shared
1843 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1845 struct files_struct *fd = current->files;
1848 if ((unshare_flags & CLONE_FILES) &&
1849 (fd && atomic_read(&fd->count) > 1)) {
1850 *new_fdp = dup_fd(fd, &error);
1859 * unshare allows a process to 'unshare' part of the process
1860 * context which was originally shared using clone. copy_*
1861 * functions used by do_fork() cannot be used here directly
1862 * because they modify an inactive task_struct that is being
1863 * constructed. Here we are modifying the current, active,
1866 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1868 struct fs_struct *fs, *new_fs = NULL;
1869 struct files_struct *fd, *new_fd = NULL;
1870 struct cred *new_cred = NULL;
1871 struct nsproxy *new_nsproxy = NULL;
1876 * If unsharing a user namespace must also unshare the thread.
1878 if (unshare_flags & CLONE_NEWUSER)
1879 unshare_flags |= CLONE_THREAD | CLONE_FS;
1881 * If unsharing a pid namespace must also unshare the thread.
1883 if (unshare_flags & CLONE_NEWPID)
1884 unshare_flags |= CLONE_THREAD;
1886 * If unsharing a thread from a thread group, must also unshare vm.
1888 if (unshare_flags & CLONE_THREAD)
1889 unshare_flags |= CLONE_VM;
1891 * If unsharing vm, must also unshare signal handlers.
1893 if (unshare_flags & CLONE_VM)
1894 unshare_flags |= CLONE_SIGHAND;
1896 * If unsharing namespace, must also unshare filesystem information.
1898 if (unshare_flags & CLONE_NEWNS)
1899 unshare_flags |= CLONE_FS;
1901 err = check_unshare_flags(unshare_flags);
1903 goto bad_unshare_out;
1905 * CLONE_NEWIPC must also detach from the undolist: after switching
1906 * to a new ipc namespace, the semaphore arrays from the old
1907 * namespace are unreachable.
1909 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1911 err = unshare_fs(unshare_flags, &new_fs);
1913 goto bad_unshare_out;
1914 err = unshare_fd(unshare_flags, &new_fd);
1916 goto bad_unshare_cleanup_fs;
1917 err = unshare_userns(unshare_flags, &new_cred);
1919 goto bad_unshare_cleanup_fd;
1920 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1923 goto bad_unshare_cleanup_cred;
1925 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1928 * CLONE_SYSVSEM is equivalent to sys_exit().
1934 switch_task_namespaces(current, new_nsproxy);
1940 spin_lock(&fs->lock);
1941 current->fs = new_fs;
1946 spin_unlock(&fs->lock);
1950 fd = current->files;
1951 current->files = new_fd;
1955 task_unlock(current);
1958 /* Install the new user namespace */
1959 commit_creds(new_cred);
1964 bad_unshare_cleanup_cred:
1967 bad_unshare_cleanup_fd:
1969 put_files_struct(new_fd);
1971 bad_unshare_cleanup_fs:
1973 free_fs_struct(new_fs);
1980 * Helper to unshare the files of the current task.
1981 * We don't want to expose copy_files internals to
1982 * the exec layer of the kernel.
1985 int unshare_files(struct files_struct **displaced)
1987 struct task_struct *task = current;
1988 struct files_struct *copy = NULL;
1991 error = unshare_fd(CLONE_FILES, ©);
1992 if (error || !copy) {
1996 *displaced = task->files;