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>
77 #include <asm/pgtable.h>
78 #include <asm/pgalloc.h>
79 #include <asm/uaccess.h>
80 #include <asm/mmu_context.h>
81 #include <asm/cacheflush.h>
82 #include <asm/tlbflush.h>
84 #include <trace/events/sched.h>
86 #define CREATE_TRACE_POINTS
87 #include <trace/events/task.h>
90 * Protected counters by write_lock_irq(&tasklist_lock)
92 unsigned long total_forks; /* Handle normal Linux uptimes. */
93 int nr_threads; /* The idle threads do not count.. */
95 int max_threads; /* tunable limit on nr_threads */
97 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
99 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
101 #ifdef CONFIG_PROVE_RCU
102 int lockdep_tasklist_lock_is_held(void)
104 return lockdep_is_held(&tasklist_lock);
106 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
107 #endif /* #ifdef CONFIG_PROVE_RCU */
109 int nr_processes(void)
114 for_each_possible_cpu(cpu)
115 total += per_cpu(process_counts, cpu);
120 void __weak arch_release_task_struct(struct task_struct *tsk)
124 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
125 static struct kmem_cache *task_struct_cachep;
127 static inline struct task_struct *alloc_task_struct_node(int node)
129 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
132 static inline void free_task_struct(struct task_struct *tsk)
134 kmem_cache_free(task_struct_cachep, tsk);
138 void __weak arch_release_thread_info(struct thread_info *ti)
142 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
145 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
146 * kmemcache based allocator.
148 # if THREAD_SIZE >= PAGE_SIZE
149 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
152 struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
155 return page ? page_address(page) : NULL;
158 static inline void free_thread_info(struct thread_info *ti)
160 free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
163 static struct kmem_cache *thread_info_cache;
165 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
168 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
171 static void free_thread_info(struct thread_info *ti)
173 kmem_cache_free(thread_info_cache, ti);
176 void thread_info_cache_init(void)
178 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
179 THREAD_SIZE, 0, NULL);
180 BUG_ON(thread_info_cache == NULL);
185 /* SLAB cache for signal_struct structures (tsk->signal) */
186 static struct kmem_cache *signal_cachep;
188 /* SLAB cache for sighand_struct structures (tsk->sighand) */
189 struct kmem_cache *sighand_cachep;
191 /* SLAB cache for files_struct structures (tsk->files) */
192 struct kmem_cache *files_cachep;
194 /* SLAB cache for fs_struct structures (tsk->fs) */
195 struct kmem_cache *fs_cachep;
197 /* SLAB cache for vm_area_struct structures */
198 struct kmem_cache *vm_area_cachep;
200 /* SLAB cache for mm_struct structures (tsk->mm) */
201 static struct kmem_cache *mm_cachep;
203 static void account_kernel_stack(struct thread_info *ti, int account)
205 struct zone *zone = page_zone(virt_to_page(ti));
207 mod_zone_page_state(zone, NR_KERNEL_STACK, account);
210 void free_task(struct task_struct *tsk)
212 account_kernel_stack(tsk->stack, -1);
213 arch_release_thread_info(tsk->stack);
214 free_thread_info(tsk->stack);
215 rt_mutex_debug_task_free(tsk);
216 ftrace_graph_exit_task(tsk);
217 put_seccomp_filter(tsk);
218 arch_release_task_struct(tsk);
219 free_task_struct(tsk);
221 EXPORT_SYMBOL(free_task);
223 static inline void free_signal_struct(struct signal_struct *sig)
225 taskstats_tgid_free(sig);
226 sched_autogroup_exit(sig);
227 kmem_cache_free(signal_cachep, sig);
230 static inline void put_signal_struct(struct signal_struct *sig)
232 if (atomic_dec_and_test(&sig->sigcnt))
233 free_signal_struct(sig);
236 void __put_task_struct(struct task_struct *tsk)
238 WARN_ON(!tsk->exit_state);
239 WARN_ON(atomic_read(&tsk->usage));
240 WARN_ON(tsk == current);
243 security_task_free(tsk);
245 delayacct_tsk_free(tsk);
246 put_signal_struct(tsk->signal);
248 if (!profile_handoff_task(tsk))
251 EXPORT_SYMBOL_GPL(__put_task_struct);
253 void __init __weak arch_task_cache_init(void) { }
255 void __init fork_init(unsigned long mempages)
257 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
258 #ifndef ARCH_MIN_TASKALIGN
259 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
261 /* create a slab on which task_structs can be allocated */
263 kmem_cache_create("task_struct", sizeof(struct task_struct),
264 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
267 /* do the arch specific task caches init */
268 arch_task_cache_init();
271 * The default maximum number of threads is set to a safe
272 * value: the thread structures can take up at most half
275 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
278 * we need to allow at least 20 threads to boot a system
280 if (max_threads < 20)
283 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
284 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
285 init_task.signal->rlim[RLIMIT_SIGPENDING] =
286 init_task.signal->rlim[RLIMIT_NPROC];
289 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
290 struct task_struct *src)
296 static struct task_struct *dup_task_struct(struct task_struct *orig)
298 struct task_struct *tsk;
299 struct thread_info *ti;
300 unsigned long *stackend;
301 int node = tsk_fork_get_node(orig);
304 tsk = alloc_task_struct_node(node);
308 ti = alloc_thread_info_node(tsk, node);
312 err = arch_dup_task_struct(tsk, orig);
318 setup_thread_stack(tsk, orig);
319 clear_user_return_notifier(tsk);
320 clear_tsk_need_resched(tsk);
321 stackend = end_of_stack(tsk);
322 *stackend = STACK_END_MAGIC; /* for overflow detection */
324 #ifdef CONFIG_CC_STACKPROTECTOR
325 tsk->stack_canary = get_random_int();
329 * One for us, one for whoever does the "release_task()" (usually
332 atomic_set(&tsk->usage, 2);
333 #ifdef CONFIG_BLK_DEV_IO_TRACE
336 tsk->splice_pipe = NULL;
337 tsk->task_frag.page = NULL;
339 account_kernel_stack(ti, 1);
344 free_thread_info(ti);
346 free_task_struct(tsk);
351 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
353 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
354 struct rb_node **rb_link, *rb_parent;
356 unsigned long charge;
358 uprobe_start_dup_mmap();
359 down_write(&oldmm->mmap_sem);
360 flush_cache_dup_mm(oldmm);
361 uprobe_dup_mmap(oldmm, mm);
363 * Not linked in yet - no deadlock potential:
365 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
369 mm->vmacache_seqnum = 0;
371 cpumask_clear(mm_cpumask(mm));
373 rb_link = &mm->mm_rb.rb_node;
376 retval = ksm_fork(mm, oldmm);
379 retval = khugepaged_fork(mm, oldmm);
384 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
387 if (mpnt->vm_flags & VM_DONTCOPY) {
388 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
393 if (mpnt->vm_flags & VM_ACCOUNT) {
394 unsigned long len = vma_pages(mpnt);
396 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
400 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
404 INIT_LIST_HEAD(&tmp->anon_vma_chain);
405 retval = vma_dup_policy(mpnt, tmp);
407 goto fail_nomem_policy;
409 if (anon_vma_fork(tmp, mpnt))
410 goto fail_nomem_anon_vma_fork;
411 tmp->vm_flags &= ~VM_LOCKED;
412 tmp->vm_next = tmp->vm_prev = NULL;
415 struct inode *inode = file_inode(file);
416 struct address_space *mapping = file->f_mapping;
419 if (tmp->vm_flags & VM_DENYWRITE)
420 atomic_dec(&inode->i_writecount);
421 mutex_lock(&mapping->i_mmap_mutex);
422 if (tmp->vm_flags & VM_SHARED)
423 mapping->i_mmap_writable++;
424 flush_dcache_mmap_lock(mapping);
425 /* insert tmp into the share list, just after mpnt */
426 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
427 vma_nonlinear_insert(tmp,
428 &mapping->i_mmap_nonlinear);
430 vma_interval_tree_insert_after(tmp, mpnt,
432 flush_dcache_mmap_unlock(mapping);
433 mutex_unlock(&mapping->i_mmap_mutex);
437 * Clear hugetlb-related page reserves for children. This only
438 * affects MAP_PRIVATE mappings. Faults generated by the child
439 * are not guaranteed to succeed, even if read-only
441 if (is_vm_hugetlb_page(tmp))
442 reset_vma_resv_huge_pages(tmp);
445 * Link in the new vma and copy the page table entries.
448 pprev = &tmp->vm_next;
452 __vma_link_rb(mm, tmp, rb_link, rb_parent);
453 rb_link = &tmp->vm_rb.rb_right;
454 rb_parent = &tmp->vm_rb;
457 retval = copy_page_range(mm, oldmm, mpnt);
459 if (tmp->vm_ops && tmp->vm_ops->open)
460 tmp->vm_ops->open(tmp);
465 /* a new mm has just been created */
466 arch_dup_mmap(oldmm, mm);
469 up_write(&mm->mmap_sem);
471 up_write(&oldmm->mmap_sem);
472 uprobe_end_dup_mmap();
474 fail_nomem_anon_vma_fork:
475 mpol_put(vma_policy(tmp));
477 kmem_cache_free(vm_area_cachep, tmp);
480 vm_unacct_memory(charge);
484 static inline int mm_alloc_pgd(struct mm_struct *mm)
486 mm->pgd = pgd_alloc(mm);
487 if (unlikely(!mm->pgd))
492 static inline void mm_free_pgd(struct mm_struct *mm)
494 pgd_free(mm, mm->pgd);
497 #define dup_mmap(mm, oldmm) (0)
498 #define mm_alloc_pgd(mm) (0)
499 #define mm_free_pgd(mm)
500 #endif /* CONFIG_MMU */
502 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
504 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
505 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
507 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
509 static int __init coredump_filter_setup(char *s)
511 default_dump_filter =
512 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
513 MMF_DUMP_FILTER_MASK;
517 __setup("coredump_filter=", coredump_filter_setup);
519 #include <linux/init_task.h>
521 static void mm_init_aio(struct mm_struct *mm)
524 spin_lock_init(&mm->ioctx_lock);
525 mm->ioctx_table = NULL;
529 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
531 atomic_set(&mm->mm_users, 1);
532 atomic_set(&mm->mm_count, 1);
533 init_rwsem(&mm->mmap_sem);
534 INIT_LIST_HEAD(&mm->mmlist);
535 mm->core_state = NULL;
536 atomic_long_set(&mm->nr_ptes, 0);
537 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
538 spin_lock_init(&mm->page_table_lock);
540 mm_init_owner(mm, p);
541 clear_tlb_flush_pending(mm);
544 mm->flags = current->mm->flags & MMF_INIT_MASK;
545 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
547 mm->flags = default_dump_filter;
551 if (likely(!mm_alloc_pgd(mm))) {
552 mmu_notifier_mm_init(mm);
560 static void check_mm(struct mm_struct *mm)
564 for (i = 0; i < NR_MM_COUNTERS; i++) {
565 long x = atomic_long_read(&mm->rss_stat.count[i]);
568 printk(KERN_ALERT "BUG: Bad rss-counter state "
569 "mm:%p idx:%d val:%ld\n", mm, i, x);
572 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
573 VM_BUG_ON(mm->pmd_huge_pte);
578 * Allocate and initialize an mm_struct.
580 struct mm_struct *mm_alloc(void)
582 struct mm_struct *mm;
588 memset(mm, 0, sizeof(*mm));
590 return mm_init(mm, current);
594 * Called when the last reference to the mm
595 * is dropped: either by a lazy thread or by
596 * mmput. Free the page directory and the mm.
598 void __mmdrop(struct mm_struct *mm)
600 BUG_ON(mm == &init_mm);
603 mmu_notifier_mm_destroy(mm);
607 EXPORT_SYMBOL_GPL(__mmdrop);
610 * Decrement the use count and release all resources for an mm.
612 void mmput(struct mm_struct *mm)
616 if (atomic_dec_and_test(&mm->mm_users)) {
617 uprobe_clear_state(mm);
620 khugepaged_exit(mm); /* must run before exit_mmap */
622 set_mm_exe_file(mm, NULL);
623 if (!list_empty(&mm->mmlist)) {
624 spin_lock(&mmlist_lock);
625 list_del(&mm->mmlist);
626 spin_unlock(&mmlist_lock);
629 module_put(mm->binfmt->module);
633 EXPORT_SYMBOL_GPL(mmput);
635 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
638 get_file(new_exe_file);
641 mm->exe_file = new_exe_file;
644 struct file *get_mm_exe_file(struct mm_struct *mm)
646 struct file *exe_file;
648 /* We need mmap_sem to protect against races with removal of exe_file */
649 down_read(&mm->mmap_sem);
650 exe_file = mm->exe_file;
653 up_read(&mm->mmap_sem);
657 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
659 /* It's safe to write the exe_file pointer without exe_file_lock because
660 * this is called during fork when the task is not yet in /proc */
661 newmm->exe_file = get_mm_exe_file(oldmm);
665 * get_task_mm - acquire a reference to the task's mm
667 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
668 * this kernel workthread has transiently adopted a user mm with use_mm,
669 * to do its AIO) is not set and if so returns a reference to it, after
670 * bumping up the use count. User must release the mm via mmput()
671 * after use. Typically used by /proc and ptrace.
673 struct mm_struct *get_task_mm(struct task_struct *task)
675 struct mm_struct *mm;
680 if (task->flags & PF_KTHREAD)
683 atomic_inc(&mm->mm_users);
688 EXPORT_SYMBOL_GPL(get_task_mm);
690 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
692 struct mm_struct *mm;
695 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
699 mm = get_task_mm(task);
700 if (mm && mm != current->mm &&
701 !ptrace_may_access(task, mode)) {
703 mm = ERR_PTR(-EACCES);
705 mutex_unlock(&task->signal->cred_guard_mutex);
710 static void complete_vfork_done(struct task_struct *tsk)
712 struct completion *vfork;
715 vfork = tsk->vfork_done;
717 tsk->vfork_done = NULL;
723 static int wait_for_vfork_done(struct task_struct *child,
724 struct completion *vfork)
728 freezer_do_not_count();
729 killed = wait_for_completion_killable(vfork);
734 child->vfork_done = NULL;
738 put_task_struct(child);
742 /* Please note the differences between mmput and mm_release.
743 * mmput is called whenever we stop holding onto a mm_struct,
744 * error success whatever.
746 * mm_release is called after a mm_struct has been removed
747 * from the current process.
749 * This difference is important for error handling, when we
750 * only half set up a mm_struct for a new process and need to restore
751 * the old one. Because we mmput the new mm_struct before
752 * restoring the old one. . .
753 * Eric Biederman 10 January 1998
755 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
757 /* Get rid of any futexes when releasing the mm */
759 if (unlikely(tsk->robust_list)) {
760 exit_robust_list(tsk);
761 tsk->robust_list = NULL;
764 if (unlikely(tsk->compat_robust_list)) {
765 compat_exit_robust_list(tsk);
766 tsk->compat_robust_list = NULL;
769 if (unlikely(!list_empty(&tsk->pi_state_list)))
770 exit_pi_state_list(tsk);
773 uprobe_free_utask(tsk);
775 /* Get rid of any cached register state */
776 deactivate_mm(tsk, mm);
779 * If we're exiting normally, clear a user-space tid field if
780 * requested. We leave this alone when dying by signal, to leave
781 * the value intact in a core dump, and to save the unnecessary
782 * trouble, say, a killed vfork parent shouldn't touch this mm.
783 * Userland only wants this done for a sys_exit.
785 if (tsk->clear_child_tid) {
786 if (!(tsk->flags & PF_SIGNALED) &&
787 atomic_read(&mm->mm_users) > 1) {
789 * We don't check the error code - if userspace has
790 * not set up a proper pointer then tough luck.
792 put_user(0, tsk->clear_child_tid);
793 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
796 tsk->clear_child_tid = NULL;
800 * All done, finally we can wake up parent and return this mm to him.
801 * Also kthread_stop() uses this completion for synchronization.
804 complete_vfork_done(tsk);
808 * Allocate a new mm structure and copy contents from the
809 * mm structure of the passed in task structure.
811 static struct mm_struct *dup_mm(struct task_struct *tsk)
813 struct mm_struct *mm, *oldmm = current->mm;
820 memcpy(mm, oldmm, sizeof(*mm));
823 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
824 mm->pmd_huge_pte = NULL;
826 if (!mm_init(mm, tsk))
829 if (init_new_context(tsk, mm))
832 dup_mm_exe_file(oldmm, mm);
834 err = dup_mmap(mm, oldmm);
838 mm->hiwater_rss = get_mm_rss(mm);
839 mm->hiwater_vm = mm->total_vm;
841 if (mm->binfmt && !try_module_get(mm->binfmt->module))
847 /* don't put binfmt in mmput, we haven't got module yet */
856 * If init_new_context() failed, we cannot use mmput() to free the mm
857 * because it calls destroy_context()
864 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
866 struct mm_struct *mm, *oldmm;
869 tsk->min_flt = tsk->maj_flt = 0;
870 tsk->nvcsw = tsk->nivcsw = 0;
871 #ifdef CONFIG_DETECT_HUNG_TASK
872 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
876 tsk->active_mm = NULL;
879 * Are we cloning a kernel thread?
881 * We need to steal a active VM for that..
887 /* initialize the new vmacache entries */
890 if (clone_flags & CLONE_VM) {
891 atomic_inc(&oldmm->mm_users);
910 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
912 struct fs_struct *fs = current->fs;
913 if (clone_flags & CLONE_FS) {
914 /* tsk->fs is already what we want */
915 spin_lock(&fs->lock);
917 spin_unlock(&fs->lock);
921 spin_unlock(&fs->lock);
924 tsk->fs = copy_fs_struct(fs);
930 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
932 struct files_struct *oldf, *newf;
936 * A background process may not have any files ...
938 oldf = current->files;
942 if (clone_flags & CLONE_FILES) {
943 atomic_inc(&oldf->count);
947 newf = dup_fd(oldf, &error);
957 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
960 struct io_context *ioc = current->io_context;
961 struct io_context *new_ioc;
966 * Share io context with parent, if CLONE_IO is set
968 if (clone_flags & CLONE_IO) {
970 tsk->io_context = ioc;
971 } else if (ioprio_valid(ioc->ioprio)) {
972 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
973 if (unlikely(!new_ioc))
976 new_ioc->ioprio = ioc->ioprio;
977 put_io_context(new_ioc);
983 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
985 struct sighand_struct *sig;
987 if (clone_flags & CLONE_SIGHAND) {
988 atomic_inc(¤t->sighand->count);
991 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
992 rcu_assign_pointer(tsk->sighand, sig);
995 atomic_set(&sig->count, 1);
996 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1000 void __cleanup_sighand(struct sighand_struct *sighand)
1002 if (atomic_dec_and_test(&sighand->count)) {
1003 signalfd_cleanup(sighand);
1004 kmem_cache_free(sighand_cachep, sighand);
1010 * Initialize POSIX timer handling for a thread group.
1012 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1014 unsigned long cpu_limit;
1016 /* Thread group counters. */
1017 thread_group_cputime_init(sig);
1019 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1020 if (cpu_limit != RLIM_INFINITY) {
1021 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1022 sig->cputimer.running = 1;
1025 /* The timer lists. */
1026 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1027 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1028 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1031 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1033 struct signal_struct *sig;
1035 if (clone_flags & CLONE_THREAD)
1038 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1043 sig->nr_threads = 1;
1044 atomic_set(&sig->live, 1);
1045 atomic_set(&sig->sigcnt, 1);
1047 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1048 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1049 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1051 init_waitqueue_head(&sig->wait_chldexit);
1052 sig->curr_target = tsk;
1053 init_sigpending(&sig->shared_pending);
1054 INIT_LIST_HEAD(&sig->posix_timers);
1056 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1057 sig->real_timer.function = it_real_fn;
1059 task_lock(current->group_leader);
1060 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1061 task_unlock(current->group_leader);
1063 posix_cpu_timers_init_group(sig);
1065 tty_audit_fork(sig);
1066 sched_autogroup_fork(sig);
1068 #ifdef CONFIG_CGROUPS
1069 init_rwsem(&sig->group_rwsem);
1072 sig->oom_score_adj = current->signal->oom_score_adj;
1073 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1075 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1076 current->signal->is_child_subreaper;
1078 mutex_init(&sig->cred_guard_mutex);
1083 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1085 unsigned long new_flags = p->flags;
1087 new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1088 new_flags |= PF_FORKNOEXEC;
1089 p->flags = new_flags;
1092 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1094 current->clear_child_tid = tidptr;
1096 return task_pid_vnr(current);
1099 static void rt_mutex_init_task(struct task_struct *p)
1101 raw_spin_lock_init(&p->pi_lock);
1102 #ifdef CONFIG_RT_MUTEXES
1103 p->pi_waiters = RB_ROOT;
1104 p->pi_waiters_leftmost = NULL;
1105 p->pi_blocked_on = NULL;
1106 p->pi_top_task = NULL;
1110 #ifdef CONFIG_MM_OWNER
1111 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1115 #endif /* CONFIG_MM_OWNER */
1118 * Initialize POSIX timer handling for a single task.
1120 static void posix_cpu_timers_init(struct task_struct *tsk)
1122 tsk->cputime_expires.prof_exp = 0;
1123 tsk->cputime_expires.virt_exp = 0;
1124 tsk->cputime_expires.sched_exp = 0;
1125 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1126 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1127 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1131 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1133 task->pids[type].pid = pid;
1137 * This creates a new process as a copy of the old one,
1138 * but does not actually start it yet.
1140 * It copies the registers, and all the appropriate
1141 * parts of the process environment (as per the clone
1142 * flags). The actual kick-off is left to the caller.
1144 static struct task_struct *copy_process(unsigned long clone_flags,
1145 unsigned long stack_start,
1146 unsigned long stack_size,
1147 int __user *child_tidptr,
1152 struct task_struct *p;
1154 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1155 return ERR_PTR(-EINVAL);
1157 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1158 return ERR_PTR(-EINVAL);
1161 * Thread groups must share signals as well, and detached threads
1162 * can only be started up within the thread group.
1164 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1165 return ERR_PTR(-EINVAL);
1168 * Shared signal handlers imply shared VM. By way of the above,
1169 * thread groups also imply shared VM. Blocking this case allows
1170 * for various simplifications in other code.
1172 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1173 return ERR_PTR(-EINVAL);
1176 * Siblings of global init remain as zombies on exit since they are
1177 * not reaped by their parent (swapper). To solve this and to avoid
1178 * multi-rooted process trees, prevent global and container-inits
1179 * from creating siblings.
1181 if ((clone_flags & CLONE_PARENT) &&
1182 current->signal->flags & SIGNAL_UNKILLABLE)
1183 return ERR_PTR(-EINVAL);
1186 * If the new process will be in a different pid or user namespace
1187 * do not allow it to share a thread group or signal handlers or
1188 * parent with the forking task.
1190 if (clone_flags & CLONE_SIGHAND) {
1191 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1192 (task_active_pid_ns(current) !=
1193 current->nsproxy->pid_ns_for_children))
1194 return ERR_PTR(-EINVAL);
1197 retval = security_task_create(clone_flags);
1202 p = dup_task_struct(current);
1206 ftrace_graph_init_task(p);
1207 get_seccomp_filter(p);
1209 rt_mutex_init_task(p);
1211 #ifdef CONFIG_PROVE_LOCKING
1212 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1213 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1216 if (atomic_read(&p->real_cred->user->processes) >=
1217 task_rlimit(p, RLIMIT_NPROC)) {
1218 if (p->real_cred->user != INIT_USER &&
1219 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1222 current->flags &= ~PF_NPROC_EXCEEDED;
1224 retval = copy_creds(p, clone_flags);
1229 * If multiple threads are within copy_process(), then this check
1230 * triggers too late. This doesn't hurt, the check is only there
1231 * to stop root fork bombs.
1234 if (nr_threads >= max_threads)
1235 goto bad_fork_cleanup_count;
1237 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1238 goto bad_fork_cleanup_count;
1240 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1241 copy_flags(clone_flags, p);
1242 INIT_LIST_HEAD(&p->children);
1243 INIT_LIST_HEAD(&p->sibling);
1244 rcu_copy_process(p);
1245 p->vfork_done = NULL;
1246 spin_lock_init(&p->alloc_lock);
1248 init_sigpending(&p->pending);
1250 p->utime = p->stime = p->gtime = 0;
1251 p->utimescaled = p->stimescaled = 0;
1252 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1253 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1255 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1256 seqlock_init(&p->vtime_seqlock);
1258 p->vtime_snap_whence = VTIME_SLEEPING;
1261 #if defined(SPLIT_RSS_COUNTING)
1262 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1265 p->default_timer_slack_ns = current->timer_slack_ns;
1267 task_io_accounting_init(&p->ioac);
1268 acct_clear_integrals(p);
1270 posix_cpu_timers_init(p);
1272 do_posix_clock_monotonic_gettime(&p->start_time);
1273 p->real_start_time = p->start_time;
1274 monotonic_to_bootbased(&p->real_start_time);
1275 p->io_context = NULL;
1276 p->audit_context = NULL;
1277 if (clone_flags & CLONE_THREAD)
1278 threadgroup_change_begin(current);
1281 p->mempolicy = mpol_dup(p->mempolicy);
1282 if (IS_ERR(p->mempolicy)) {
1283 retval = PTR_ERR(p->mempolicy);
1284 p->mempolicy = NULL;
1285 goto bad_fork_cleanup_threadgroup_lock;
1287 mpol_fix_fork_child_flag(p);
1289 #ifdef CONFIG_CPUSETS
1290 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1291 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1292 seqcount_init(&p->mems_allowed_seq);
1294 #ifdef CONFIG_TRACE_IRQFLAGS
1296 p->hardirqs_enabled = 0;
1297 p->hardirq_enable_ip = 0;
1298 p->hardirq_enable_event = 0;
1299 p->hardirq_disable_ip = _THIS_IP_;
1300 p->hardirq_disable_event = 0;
1301 p->softirqs_enabled = 1;
1302 p->softirq_enable_ip = _THIS_IP_;
1303 p->softirq_enable_event = 0;
1304 p->softirq_disable_ip = 0;
1305 p->softirq_disable_event = 0;
1306 p->hardirq_context = 0;
1307 p->softirq_context = 0;
1309 #ifdef CONFIG_LOCKDEP
1310 p->lockdep_depth = 0; /* no locks held yet */
1311 p->curr_chain_key = 0;
1312 p->lockdep_recursion = 0;
1315 #ifdef CONFIG_DEBUG_MUTEXES
1316 p->blocked_on = NULL; /* not blocked yet */
1319 p->memcg_batch.do_batch = 0;
1320 p->memcg_batch.memcg = NULL;
1322 #ifdef CONFIG_BCACHE
1323 p->sequential_io = 0;
1324 p->sequential_io_avg = 0;
1327 /* Perform scheduler related setup. Assign this task to a CPU. */
1328 retval = sched_fork(clone_flags, p);
1330 goto bad_fork_cleanup_policy;
1332 retval = perf_event_init_task(p);
1334 goto bad_fork_cleanup_policy;
1335 retval = audit_alloc(p);
1337 goto bad_fork_cleanup_policy;
1338 /* copy all the process information */
1339 retval = copy_semundo(clone_flags, p);
1341 goto bad_fork_cleanup_audit;
1342 retval = copy_files(clone_flags, p);
1344 goto bad_fork_cleanup_semundo;
1345 retval = copy_fs(clone_flags, p);
1347 goto bad_fork_cleanup_files;
1348 retval = copy_sighand(clone_flags, p);
1350 goto bad_fork_cleanup_fs;
1351 retval = copy_signal(clone_flags, p);
1353 goto bad_fork_cleanup_sighand;
1354 retval = copy_mm(clone_flags, p);
1356 goto bad_fork_cleanup_signal;
1357 retval = copy_namespaces(clone_flags, p);
1359 goto bad_fork_cleanup_mm;
1360 retval = copy_io(clone_flags, p);
1362 goto bad_fork_cleanup_namespaces;
1363 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1365 goto bad_fork_cleanup_io;
1367 if (pid != &init_struct_pid) {
1369 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1371 goto bad_fork_cleanup_io;
1374 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1376 * Clear TID on mm_release()?
1378 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1383 p->robust_list = NULL;
1384 #ifdef CONFIG_COMPAT
1385 p->compat_robust_list = NULL;
1387 INIT_LIST_HEAD(&p->pi_state_list);
1388 p->pi_state_cache = NULL;
1391 * sigaltstack should be cleared when sharing the same VM
1393 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1394 p->sas_ss_sp = p->sas_ss_size = 0;
1397 * Syscall tracing and stepping should be turned off in the
1398 * child regardless of CLONE_PTRACE.
1400 user_disable_single_step(p);
1401 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1402 #ifdef TIF_SYSCALL_EMU
1403 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1405 clear_all_latency_tracing(p);
1407 /* ok, now we should be set up.. */
1408 p->pid = pid_nr(pid);
1409 if (clone_flags & CLONE_THREAD) {
1410 p->exit_signal = -1;
1411 p->group_leader = current->group_leader;
1412 p->tgid = current->tgid;
1414 if (clone_flags & CLONE_PARENT)
1415 p->exit_signal = current->group_leader->exit_signal;
1417 p->exit_signal = (clone_flags & CSIGNAL);
1418 p->group_leader = p;
1423 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1424 p->dirty_paused_when = 0;
1426 p->pdeath_signal = 0;
1427 INIT_LIST_HEAD(&p->thread_group);
1428 p->task_works = NULL;
1431 * Make it visible to the rest of the system, but dont wake it up yet.
1432 * Need tasklist lock for parent etc handling!
1434 write_lock_irq(&tasklist_lock);
1436 /* CLONE_PARENT re-uses the old parent */
1437 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1438 p->real_parent = current->real_parent;
1439 p->parent_exec_id = current->parent_exec_id;
1441 p->real_parent = current;
1442 p->parent_exec_id = current->self_exec_id;
1445 spin_lock(¤t->sighand->siglock);
1448 * Process group and session signals need to be delivered to just the
1449 * parent before the fork or both the parent and the child after the
1450 * fork. Restart if a signal comes in before we add the new process to
1451 * it's process group.
1452 * A fatal signal pending means that current will exit, so the new
1453 * thread can't slip out of an OOM kill (or normal SIGKILL).
1455 recalc_sigpending();
1456 if (signal_pending(current)) {
1457 spin_unlock(¤t->sighand->siglock);
1458 write_unlock_irq(&tasklist_lock);
1459 retval = -ERESTARTNOINTR;
1460 goto bad_fork_free_pid;
1463 if (likely(p->pid)) {
1464 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1466 init_task_pid(p, PIDTYPE_PID, pid);
1467 if (thread_group_leader(p)) {
1468 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1469 init_task_pid(p, PIDTYPE_SID, task_session(current));
1471 if (is_child_reaper(pid)) {
1472 ns_of_pid(pid)->child_reaper = p;
1473 p->signal->flags |= SIGNAL_UNKILLABLE;
1476 p->signal->leader_pid = pid;
1477 p->signal->tty = tty_kref_get(current->signal->tty);
1478 list_add_tail(&p->sibling, &p->real_parent->children);
1479 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1480 attach_pid(p, PIDTYPE_PGID);
1481 attach_pid(p, PIDTYPE_SID);
1482 __this_cpu_inc(process_counts);
1484 current->signal->nr_threads++;
1485 atomic_inc(¤t->signal->live);
1486 atomic_inc(¤t->signal->sigcnt);
1487 list_add_tail_rcu(&p->thread_group,
1488 &p->group_leader->thread_group);
1489 list_add_tail_rcu(&p->thread_node,
1490 &p->signal->thread_head);
1492 attach_pid(p, PIDTYPE_PID);
1497 spin_unlock(¤t->sighand->siglock);
1498 write_unlock_irq(&tasklist_lock);
1499 proc_fork_connector(p);
1500 cgroup_post_fork(p);
1501 if (clone_flags & CLONE_THREAD)
1502 threadgroup_change_end(current);
1505 trace_task_newtask(p, clone_flags);
1506 uprobe_copy_process(p, clone_flags);
1511 if (pid != &init_struct_pid)
1513 bad_fork_cleanup_io:
1516 bad_fork_cleanup_namespaces:
1517 exit_task_namespaces(p);
1518 bad_fork_cleanup_mm:
1521 bad_fork_cleanup_signal:
1522 if (!(clone_flags & CLONE_THREAD))
1523 free_signal_struct(p->signal);
1524 bad_fork_cleanup_sighand:
1525 __cleanup_sighand(p->sighand);
1526 bad_fork_cleanup_fs:
1527 exit_fs(p); /* blocking */
1528 bad_fork_cleanup_files:
1529 exit_files(p); /* blocking */
1530 bad_fork_cleanup_semundo:
1532 bad_fork_cleanup_audit:
1534 bad_fork_cleanup_policy:
1535 perf_event_free_task(p);
1537 mpol_put(p->mempolicy);
1538 bad_fork_cleanup_threadgroup_lock:
1540 if (clone_flags & CLONE_THREAD)
1541 threadgroup_change_end(current);
1542 delayacct_tsk_free(p);
1543 module_put(task_thread_info(p)->exec_domain->module);
1544 bad_fork_cleanup_count:
1545 atomic_dec(&p->cred->user->processes);
1550 return ERR_PTR(retval);
1553 static inline void init_idle_pids(struct pid_link *links)
1557 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1558 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1559 links[type].pid = &init_struct_pid;
1563 struct task_struct *fork_idle(int cpu)
1565 struct task_struct *task;
1566 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1567 if (!IS_ERR(task)) {
1568 init_idle_pids(task->pids);
1569 init_idle(task, cpu);
1576 * Ok, this is the main fork-routine.
1578 * It copies the process, and if successful kick-starts
1579 * it and waits for it to finish using the VM if required.
1581 long do_fork(unsigned long clone_flags,
1582 unsigned long stack_start,
1583 unsigned long stack_size,
1584 int __user *parent_tidptr,
1585 int __user *child_tidptr)
1587 struct task_struct *p;
1592 * Determine whether and which event to report to ptracer. When
1593 * called from kernel_thread or CLONE_UNTRACED is explicitly
1594 * requested, no event is reported; otherwise, report if the event
1595 * for the type of forking is enabled.
1597 if (!(clone_flags & CLONE_UNTRACED)) {
1598 if (clone_flags & CLONE_VFORK)
1599 trace = PTRACE_EVENT_VFORK;
1600 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1601 trace = PTRACE_EVENT_CLONE;
1603 trace = PTRACE_EVENT_FORK;
1605 if (likely(!ptrace_event_enabled(current, trace)))
1609 p = copy_process(clone_flags, stack_start, stack_size,
1610 child_tidptr, NULL, trace);
1612 * Do this prior waking up the new thread - the thread pointer
1613 * might get invalid after that point, if the thread exits quickly.
1616 struct completion vfork;
1618 trace_sched_process_fork(current, p);
1620 nr = task_pid_vnr(p);
1622 if (clone_flags & CLONE_PARENT_SETTID)
1623 put_user(nr, parent_tidptr);
1625 if (clone_flags & CLONE_VFORK) {
1626 p->vfork_done = &vfork;
1627 init_completion(&vfork);
1631 wake_up_new_task(p);
1633 /* forking complete and child started to run, tell ptracer */
1634 if (unlikely(trace))
1635 ptrace_event(trace, nr);
1637 if (clone_flags & CLONE_VFORK) {
1638 if (!wait_for_vfork_done(p, &vfork))
1639 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1648 * Create a kernel thread.
1650 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1652 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1653 (unsigned long)arg, NULL, NULL);
1656 #ifdef __ARCH_WANT_SYS_FORK
1657 SYSCALL_DEFINE0(fork)
1660 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1662 /* can not support in nommu mode */
1668 #ifdef __ARCH_WANT_SYS_VFORK
1669 SYSCALL_DEFINE0(vfork)
1671 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1676 #ifdef __ARCH_WANT_SYS_CLONE
1677 #ifdef CONFIG_CLONE_BACKWARDS
1678 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1679 int __user *, parent_tidptr,
1681 int __user *, child_tidptr)
1682 #elif defined(CONFIG_CLONE_BACKWARDS2)
1683 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1684 int __user *, parent_tidptr,
1685 int __user *, child_tidptr,
1687 #elif defined(CONFIG_CLONE_BACKWARDS3)
1688 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1690 int __user *, parent_tidptr,
1691 int __user *, child_tidptr,
1694 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1695 int __user *, parent_tidptr,
1696 int __user *, child_tidptr,
1700 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1704 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1705 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1708 static void sighand_ctor(void *data)
1710 struct sighand_struct *sighand = data;
1712 spin_lock_init(&sighand->siglock);
1713 init_waitqueue_head(&sighand->signalfd_wqh);
1716 void __init proc_caches_init(void)
1718 sighand_cachep = kmem_cache_create("sighand_cache",
1719 sizeof(struct sighand_struct), 0,
1720 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1721 SLAB_NOTRACK, sighand_ctor);
1722 signal_cachep = kmem_cache_create("signal_cache",
1723 sizeof(struct signal_struct), 0,
1724 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1725 files_cachep = kmem_cache_create("files_cache",
1726 sizeof(struct files_struct), 0,
1727 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1728 fs_cachep = kmem_cache_create("fs_cache",
1729 sizeof(struct fs_struct), 0,
1730 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1732 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1733 * whole struct cpumask for the OFFSTACK case. We could change
1734 * this to *only* allocate as much of it as required by the
1735 * maximum number of CPU's we can ever have. The cpumask_allocation
1736 * is at the end of the structure, exactly for that reason.
1738 mm_cachep = kmem_cache_create("mm_struct",
1739 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1740 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1741 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1743 nsproxy_cache_init();
1747 * Check constraints on flags passed to the unshare system call.
1749 static int check_unshare_flags(unsigned long unshare_flags)
1751 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1752 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1753 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1754 CLONE_NEWUSER|CLONE_NEWPID))
1757 * Not implemented, but pretend it works if there is nothing to
1758 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1759 * needs to unshare vm.
1761 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1762 /* FIXME: get_task_mm() increments ->mm_users */
1763 if (atomic_read(¤t->mm->mm_users) > 1)
1771 * Unshare the filesystem structure if it is being shared
1773 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1775 struct fs_struct *fs = current->fs;
1777 if (!(unshare_flags & CLONE_FS) || !fs)
1780 /* don't need lock here; in the worst case we'll do useless copy */
1784 *new_fsp = copy_fs_struct(fs);
1792 * Unshare file descriptor table if it is being shared
1794 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1796 struct files_struct *fd = current->files;
1799 if ((unshare_flags & CLONE_FILES) &&
1800 (fd && atomic_read(&fd->count) > 1)) {
1801 *new_fdp = dup_fd(fd, &error);
1810 * unshare allows a process to 'unshare' part of the process
1811 * context which was originally shared using clone. copy_*
1812 * functions used by do_fork() cannot be used here directly
1813 * because they modify an inactive task_struct that is being
1814 * constructed. Here we are modifying the current, active,
1817 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1819 struct fs_struct *fs, *new_fs = NULL;
1820 struct files_struct *fd, *new_fd = NULL;
1821 struct cred *new_cred = NULL;
1822 struct nsproxy *new_nsproxy = NULL;
1827 * If unsharing a user namespace must also unshare the thread.
1829 if (unshare_flags & CLONE_NEWUSER)
1830 unshare_flags |= CLONE_THREAD | CLONE_FS;
1832 * If unsharing a thread from a thread group, must also unshare vm.
1834 if (unshare_flags & CLONE_THREAD)
1835 unshare_flags |= CLONE_VM;
1837 * If unsharing vm, must also unshare signal handlers.
1839 if (unshare_flags & CLONE_VM)
1840 unshare_flags |= CLONE_SIGHAND;
1842 * If unsharing namespace, must also unshare filesystem information.
1844 if (unshare_flags & CLONE_NEWNS)
1845 unshare_flags |= CLONE_FS;
1847 err = check_unshare_flags(unshare_flags);
1849 goto bad_unshare_out;
1851 * CLONE_NEWIPC must also detach from the undolist: after switching
1852 * to a new ipc namespace, the semaphore arrays from the old
1853 * namespace are unreachable.
1855 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1857 err = unshare_fs(unshare_flags, &new_fs);
1859 goto bad_unshare_out;
1860 err = unshare_fd(unshare_flags, &new_fd);
1862 goto bad_unshare_cleanup_fs;
1863 err = unshare_userns(unshare_flags, &new_cred);
1865 goto bad_unshare_cleanup_fd;
1866 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1869 goto bad_unshare_cleanup_cred;
1871 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1874 * CLONE_SYSVSEM is equivalent to sys_exit().
1880 switch_task_namespaces(current, new_nsproxy);
1886 spin_lock(&fs->lock);
1887 current->fs = new_fs;
1892 spin_unlock(&fs->lock);
1896 fd = current->files;
1897 current->files = new_fd;
1901 task_unlock(current);
1904 /* Install the new user namespace */
1905 commit_creds(new_cred);
1910 bad_unshare_cleanup_cred:
1913 bad_unshare_cleanup_fd:
1915 put_files_struct(new_fd);
1917 bad_unshare_cleanup_fs:
1919 free_fs_struct(new_fs);
1926 * Helper to unshare the files of the current task.
1927 * We don't want to expose copy_files internals to
1928 * the exec layer of the kernel.
1931 int unshare_files(struct files_struct **displaced)
1933 struct task_struct *task = current;
1934 struct files_struct *copy = NULL;
1937 error = unshare_fd(CLONE_FILES, ©);
1938 if (error || !copy) {
1942 *displaced = task->files;