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 * Protected counters by write_lock_irq(&tasklist_lock)
93 unsigned long total_forks; /* Handle normal Linux uptimes. */
94 int nr_threads; /* The idle threads do not count.. */
96 int max_threads; /* tunable limit on nr_threads */
98 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
100 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
102 #ifdef CONFIG_PROVE_RCU
103 int lockdep_tasklist_lock_is_held(void)
105 return lockdep_is_held(&tasklist_lock);
107 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
108 #endif /* #ifdef CONFIG_PROVE_RCU */
110 int nr_processes(void)
115 for_each_possible_cpu(cpu)
116 total += per_cpu(process_counts, cpu);
121 void __weak arch_release_task_struct(struct task_struct *tsk)
125 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
126 static struct kmem_cache *task_struct_cachep;
128 static inline struct task_struct *alloc_task_struct_node(int node)
130 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
133 static inline void free_task_struct(struct task_struct *tsk)
135 kmem_cache_free(task_struct_cachep, tsk);
139 void __weak arch_release_thread_info(struct thread_info *ti)
143 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
146 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
147 * kmemcache based allocator.
149 # if THREAD_SIZE >= PAGE_SIZE
150 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
153 struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
156 return page ? page_address(page) : NULL;
159 static inline void free_thread_info(struct thread_info *ti)
161 free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
164 static struct kmem_cache *thread_info_cache;
166 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
169 return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
172 static void free_thread_info(struct thread_info *ti)
174 kmem_cache_free(thread_info_cache, ti);
177 void thread_info_cache_init(void)
179 thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
180 THREAD_SIZE, 0, NULL);
181 BUG_ON(thread_info_cache == NULL);
186 /* SLAB cache for signal_struct structures (tsk->signal) */
187 static struct kmem_cache *signal_cachep;
189 /* SLAB cache for sighand_struct structures (tsk->sighand) */
190 struct kmem_cache *sighand_cachep;
192 /* SLAB cache for files_struct structures (tsk->files) */
193 struct kmem_cache *files_cachep;
195 /* SLAB cache for fs_struct structures (tsk->fs) */
196 struct kmem_cache *fs_cachep;
198 /* SLAB cache for vm_area_struct structures */
199 struct kmem_cache *vm_area_cachep;
201 /* SLAB cache for mm_struct structures (tsk->mm) */
202 static struct kmem_cache *mm_cachep;
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 void __put_task_struct(struct task_struct *tsk)
239 WARN_ON(!tsk->exit_state);
240 WARN_ON(atomic_read(&tsk->usage));
241 WARN_ON(tsk == current);
244 security_task_free(tsk);
246 delayacct_tsk_free(tsk);
247 put_signal_struct(tsk->signal);
249 if (!profile_handoff_task(tsk))
252 EXPORT_SYMBOL_GPL(__put_task_struct);
254 void __init __weak arch_task_cache_init(void) { }
256 void __init fork_init(unsigned long mempages)
258 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
259 #ifndef ARCH_MIN_TASKALIGN
260 #define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
262 /* create a slab on which task_structs can be allocated */
264 kmem_cache_create("task_struct", sizeof(struct task_struct),
265 ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
268 /* do the arch specific task caches init */
269 arch_task_cache_init();
272 * The default maximum number of threads is set to a safe
273 * value: the thread structures can take up at most half
276 max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
279 * we need to allow at least 20 threads to boot a system
281 if (max_threads < 20)
284 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
285 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
286 init_task.signal->rlim[RLIMIT_SIGPENDING] =
287 init_task.signal->rlim[RLIMIT_NPROC];
290 int __weak arch_dup_task_struct(struct task_struct *dst,
291 struct task_struct *src)
297 static struct task_struct *dup_task_struct(struct task_struct *orig)
299 struct task_struct *tsk;
300 struct thread_info *ti;
301 unsigned long *stackend;
302 int node = tsk_fork_get_node(orig);
305 tsk = alloc_task_struct_node(node);
309 ti = alloc_thread_info_node(tsk, node);
313 err = arch_dup_task_struct(tsk, orig);
318 #ifdef CONFIG_SECCOMP
320 * We must handle setting up seccomp filters once we're under
321 * the sighand lock in case orig has changed between now and
322 * then. Until then, filter must be NULL to avoid messing up
323 * the usage counts on the error path calling free_task.
325 tsk->seccomp.filter = NULL;
328 setup_thread_stack(tsk, orig);
329 clear_user_return_notifier(tsk);
330 clear_tsk_need_resched(tsk);
331 stackend = end_of_stack(tsk);
332 *stackend = STACK_END_MAGIC; /* for overflow detection */
334 #ifdef CONFIG_CC_STACKPROTECTOR
335 tsk->stack_canary = get_random_int();
339 * One for us, one for whoever does the "release_task()" (usually
342 atomic_set(&tsk->usage, 2);
343 #ifdef CONFIG_BLK_DEV_IO_TRACE
346 tsk->splice_pipe = NULL;
347 tsk->task_frag.page = NULL;
349 account_kernel_stack(ti, 1);
354 free_thread_info(ti);
356 free_task_struct(tsk);
361 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
363 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
364 struct rb_node **rb_link, *rb_parent;
366 unsigned long charge;
368 uprobe_start_dup_mmap();
369 down_write(&oldmm->mmap_sem);
370 flush_cache_dup_mm(oldmm);
371 uprobe_dup_mmap(oldmm, mm);
373 * Not linked in yet - no deadlock potential:
375 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
379 mm->vmacache_seqnum = 0;
381 cpumask_clear(mm_cpumask(mm));
383 rb_link = &mm->mm_rb.rb_node;
386 retval = ksm_fork(mm, oldmm);
389 retval = khugepaged_fork(mm, oldmm);
394 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
397 if (mpnt->vm_flags & VM_DONTCOPY) {
398 vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
403 if (mpnt->vm_flags & VM_ACCOUNT) {
404 unsigned long len = vma_pages(mpnt);
406 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
410 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
414 INIT_LIST_HEAD(&tmp->anon_vma_chain);
415 retval = vma_dup_policy(mpnt, tmp);
417 goto fail_nomem_policy;
419 if (anon_vma_fork(tmp, mpnt))
420 goto fail_nomem_anon_vma_fork;
421 tmp->vm_flags &= ~VM_LOCKED;
422 tmp->vm_next = tmp->vm_prev = NULL;
425 struct inode *inode = file_inode(file);
426 struct address_space *mapping = file->f_mapping;
429 if (tmp->vm_flags & VM_DENYWRITE)
430 atomic_dec(&inode->i_writecount);
431 mutex_lock(&mapping->i_mmap_mutex);
432 if (tmp->vm_flags & VM_SHARED)
433 mapping->i_mmap_writable++;
434 flush_dcache_mmap_lock(mapping);
435 /* insert tmp into the share list, just after mpnt */
436 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
437 vma_nonlinear_insert(tmp,
438 &mapping->i_mmap_nonlinear);
440 vma_interval_tree_insert_after(tmp, mpnt,
442 flush_dcache_mmap_unlock(mapping);
443 mutex_unlock(&mapping->i_mmap_mutex);
447 * Clear hugetlb-related page reserves for children. This only
448 * affects MAP_PRIVATE mappings. Faults generated by the child
449 * are not guaranteed to succeed, even if read-only
451 if (is_vm_hugetlb_page(tmp))
452 reset_vma_resv_huge_pages(tmp);
455 * Link in the new vma and copy the page table entries.
458 pprev = &tmp->vm_next;
462 __vma_link_rb(mm, tmp, rb_link, rb_parent);
463 rb_link = &tmp->vm_rb.rb_right;
464 rb_parent = &tmp->vm_rb;
467 retval = copy_page_range(mm, oldmm, mpnt);
469 if (tmp->vm_ops && tmp->vm_ops->open)
470 tmp->vm_ops->open(tmp);
475 /* a new mm has just been created */
476 arch_dup_mmap(oldmm, mm);
479 up_write(&mm->mmap_sem);
481 up_write(&oldmm->mmap_sem);
482 uprobe_end_dup_mmap();
484 fail_nomem_anon_vma_fork:
485 mpol_put(vma_policy(tmp));
487 kmem_cache_free(vm_area_cachep, tmp);
490 vm_unacct_memory(charge);
494 static inline int mm_alloc_pgd(struct mm_struct *mm)
496 mm->pgd = pgd_alloc(mm);
497 if (unlikely(!mm->pgd))
502 static inline void mm_free_pgd(struct mm_struct *mm)
504 pgd_free(mm, mm->pgd);
507 #define dup_mmap(mm, oldmm) (0)
508 #define mm_alloc_pgd(mm) (0)
509 #define mm_free_pgd(mm)
510 #endif /* CONFIG_MMU */
512 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
514 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
515 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
517 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
519 static int __init coredump_filter_setup(char *s)
521 default_dump_filter =
522 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
523 MMF_DUMP_FILTER_MASK;
527 __setup("coredump_filter=", coredump_filter_setup);
529 #include <linux/init_task.h>
531 static void mm_init_aio(struct mm_struct *mm)
534 spin_lock_init(&mm->ioctx_lock);
535 mm->ioctx_table = NULL;
539 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
541 atomic_set(&mm->mm_users, 1);
542 atomic_set(&mm->mm_count, 1);
543 init_rwsem(&mm->mmap_sem);
544 INIT_LIST_HEAD(&mm->mmlist);
545 mm->core_state = NULL;
546 atomic_long_set(&mm->nr_ptes, 0);
547 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
548 spin_lock_init(&mm->page_table_lock);
550 mm_init_owner(mm, p);
551 clear_tlb_flush_pending(mm);
554 mm->flags = current->mm->flags & MMF_INIT_MASK;
555 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
557 mm->flags = default_dump_filter;
561 if (likely(!mm_alloc_pgd(mm))) {
562 mmu_notifier_mm_init(mm);
570 static void check_mm(struct mm_struct *mm)
574 for (i = 0; i < NR_MM_COUNTERS; i++) {
575 long x = atomic_long_read(&mm->rss_stat.count[i]);
578 printk(KERN_ALERT "BUG: Bad rss-counter state "
579 "mm:%p idx:%d val:%ld\n", mm, i, x);
582 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
583 VM_BUG_ON(mm->pmd_huge_pte);
588 * Allocate and initialize an mm_struct.
590 struct mm_struct *mm_alloc(void)
592 struct mm_struct *mm;
598 memset(mm, 0, sizeof(*mm));
600 return mm_init(mm, current);
604 * Called when the last reference to the mm
605 * is dropped: either by a lazy thread or by
606 * mmput. Free the page directory and the mm.
608 void __mmdrop(struct mm_struct *mm)
610 BUG_ON(mm == &init_mm);
613 mmu_notifier_mm_destroy(mm);
617 EXPORT_SYMBOL_GPL(__mmdrop);
620 * Decrement the use count and release all resources for an mm.
622 void mmput(struct mm_struct *mm)
626 if (atomic_dec_and_test(&mm->mm_users)) {
627 uprobe_clear_state(mm);
630 khugepaged_exit(mm); /* must run before exit_mmap */
632 set_mm_exe_file(mm, NULL);
633 if (!list_empty(&mm->mmlist)) {
634 spin_lock(&mmlist_lock);
635 list_del(&mm->mmlist);
636 spin_unlock(&mmlist_lock);
639 module_put(mm->binfmt->module);
643 EXPORT_SYMBOL_GPL(mmput);
645 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
648 get_file(new_exe_file);
651 mm->exe_file = new_exe_file;
654 struct file *get_mm_exe_file(struct mm_struct *mm)
656 struct file *exe_file;
658 /* We need mmap_sem to protect against races with removal of exe_file */
659 down_read(&mm->mmap_sem);
660 exe_file = mm->exe_file;
663 up_read(&mm->mmap_sem);
667 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
669 /* It's safe to write the exe_file pointer without exe_file_lock because
670 * this is called during fork when the task is not yet in /proc */
671 newmm->exe_file = get_mm_exe_file(oldmm);
675 * get_task_mm - acquire a reference to the task's mm
677 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
678 * this kernel workthread has transiently adopted a user mm with use_mm,
679 * to do its AIO) is not set and if so returns a reference to it, after
680 * bumping up the use count. User must release the mm via mmput()
681 * after use. Typically used by /proc and ptrace.
683 struct mm_struct *get_task_mm(struct task_struct *task)
685 struct mm_struct *mm;
690 if (task->flags & PF_KTHREAD)
693 atomic_inc(&mm->mm_users);
698 EXPORT_SYMBOL_GPL(get_task_mm);
700 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
702 struct mm_struct *mm;
705 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
709 mm = get_task_mm(task);
710 if (mm && mm != current->mm &&
711 !ptrace_may_access(task, mode)) {
713 mm = ERR_PTR(-EACCES);
715 mutex_unlock(&task->signal->cred_guard_mutex);
720 static void complete_vfork_done(struct task_struct *tsk)
722 struct completion *vfork;
725 vfork = tsk->vfork_done;
727 tsk->vfork_done = NULL;
733 static int wait_for_vfork_done(struct task_struct *child,
734 struct completion *vfork)
738 freezer_do_not_count();
739 killed = wait_for_completion_killable(vfork);
744 child->vfork_done = NULL;
748 put_task_struct(child);
752 /* Please note the differences between mmput and mm_release.
753 * mmput is called whenever we stop holding onto a mm_struct,
754 * error success whatever.
756 * mm_release is called after a mm_struct has been removed
757 * from the current process.
759 * This difference is important for error handling, when we
760 * only half set up a mm_struct for a new process and need to restore
761 * the old one. Because we mmput the new mm_struct before
762 * restoring the old one. . .
763 * Eric Biederman 10 January 1998
765 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
767 /* Get rid of any futexes when releasing the mm */
769 if (unlikely(tsk->robust_list)) {
770 exit_robust_list(tsk);
771 tsk->robust_list = NULL;
774 if (unlikely(tsk->compat_robust_list)) {
775 compat_exit_robust_list(tsk);
776 tsk->compat_robust_list = NULL;
779 if (unlikely(!list_empty(&tsk->pi_state_list)))
780 exit_pi_state_list(tsk);
783 uprobe_free_utask(tsk);
785 /* Get rid of any cached register state */
786 deactivate_mm(tsk, mm);
789 * If we're exiting normally, clear a user-space tid field if
790 * requested. We leave this alone when dying by signal, to leave
791 * the value intact in a core dump, and to save the unnecessary
792 * trouble, say, a killed vfork parent shouldn't touch this mm.
793 * Userland only wants this done for a sys_exit.
795 if (tsk->clear_child_tid) {
796 if (!(tsk->flags & PF_SIGNALED) &&
797 atomic_read(&mm->mm_users) > 1) {
799 * We don't check the error code - if userspace has
800 * not set up a proper pointer then tough luck.
802 put_user(0, tsk->clear_child_tid);
803 sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
806 tsk->clear_child_tid = NULL;
810 * All done, finally we can wake up parent and return this mm to him.
811 * Also kthread_stop() uses this completion for synchronization.
814 complete_vfork_done(tsk);
818 * Allocate a new mm structure and copy contents from the
819 * mm structure of the passed in task structure.
821 static struct mm_struct *dup_mm(struct task_struct *tsk)
823 struct mm_struct *mm, *oldmm = current->mm;
830 memcpy(mm, oldmm, sizeof(*mm));
833 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
834 mm->pmd_huge_pte = NULL;
836 if (!mm_init(mm, tsk))
839 if (init_new_context(tsk, mm))
842 dup_mm_exe_file(oldmm, mm);
844 err = dup_mmap(mm, oldmm);
848 mm->hiwater_rss = get_mm_rss(mm);
849 mm->hiwater_vm = mm->total_vm;
851 if (mm->binfmt && !try_module_get(mm->binfmt->module))
857 /* don't put binfmt in mmput, we haven't got module yet */
866 * If init_new_context() failed, we cannot use mmput() to free the mm
867 * because it calls destroy_context()
874 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
876 struct mm_struct *mm, *oldmm;
879 tsk->min_flt = tsk->maj_flt = 0;
880 tsk->nvcsw = tsk->nivcsw = 0;
881 #ifdef CONFIG_DETECT_HUNG_TASK
882 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
886 tsk->active_mm = NULL;
889 * Are we cloning a kernel thread?
891 * We need to steal a active VM for that..
897 /* initialize the new vmacache entries */
900 if (clone_flags & CLONE_VM) {
901 atomic_inc(&oldmm->mm_users);
920 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
922 struct fs_struct *fs = current->fs;
923 if (clone_flags & CLONE_FS) {
924 /* tsk->fs is already what we want */
925 spin_lock(&fs->lock);
927 spin_unlock(&fs->lock);
931 spin_unlock(&fs->lock);
934 tsk->fs = copy_fs_struct(fs);
940 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
942 struct files_struct *oldf, *newf;
946 * A background process may not have any files ...
948 oldf = current->files;
952 if (clone_flags & CLONE_FILES) {
953 atomic_inc(&oldf->count);
957 newf = dup_fd(oldf, &error);
967 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
970 struct io_context *ioc = current->io_context;
971 struct io_context *new_ioc;
976 * Share io context with parent, if CLONE_IO is set
978 if (clone_flags & CLONE_IO) {
980 tsk->io_context = ioc;
981 } else if (ioprio_valid(ioc->ioprio)) {
982 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
983 if (unlikely(!new_ioc))
986 new_ioc->ioprio = ioc->ioprio;
987 put_io_context(new_ioc);
993 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
995 struct sighand_struct *sig;
997 if (clone_flags & CLONE_SIGHAND) {
998 atomic_inc(¤t->sighand->count);
1001 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1002 rcu_assign_pointer(tsk->sighand, sig);
1005 atomic_set(&sig->count, 1);
1006 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1010 void __cleanup_sighand(struct sighand_struct *sighand)
1012 if (atomic_dec_and_test(&sighand->count)) {
1013 signalfd_cleanup(sighand);
1014 kmem_cache_free(sighand_cachep, sighand);
1020 * Initialize POSIX timer handling for a thread group.
1022 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1024 unsigned long cpu_limit;
1026 /* Thread group counters. */
1027 thread_group_cputime_init(sig);
1029 cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1030 if (cpu_limit != RLIM_INFINITY) {
1031 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1032 sig->cputimer.running = 1;
1035 /* The timer lists. */
1036 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1037 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1038 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1041 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1043 struct signal_struct *sig;
1045 if (clone_flags & CLONE_THREAD)
1048 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1053 sig->nr_threads = 1;
1054 atomic_set(&sig->live, 1);
1055 atomic_set(&sig->sigcnt, 1);
1057 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1058 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1059 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1061 init_waitqueue_head(&sig->wait_chldexit);
1062 sig->curr_target = tsk;
1063 init_sigpending(&sig->shared_pending);
1064 INIT_LIST_HEAD(&sig->posix_timers);
1066 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1067 sig->real_timer.function = it_real_fn;
1069 task_lock(current->group_leader);
1070 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1071 task_unlock(current->group_leader);
1073 posix_cpu_timers_init_group(sig);
1075 tty_audit_fork(sig);
1076 sched_autogroup_fork(sig);
1078 #ifdef CONFIG_CGROUPS
1079 init_rwsem(&sig->group_rwsem);
1082 sig->oom_score_adj = current->signal->oom_score_adj;
1083 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1085 sig->has_child_subreaper = current->signal->has_child_subreaper ||
1086 current->signal->is_child_subreaper;
1088 mutex_init(&sig->cred_guard_mutex);
1093 static void copy_seccomp(struct task_struct *p)
1095 #ifdef CONFIG_SECCOMP
1097 * Must be called with sighand->lock held, which is common to
1098 * all threads in the group. Holding cred_guard_mutex is not
1099 * needed because this new task is not yet running and cannot
1102 BUG_ON(!spin_is_locked(¤t->sighand->siglock));
1104 /* Ref-count the new filter user, and assign it. */
1105 get_seccomp_filter(current);
1106 p->seccomp = current->seccomp;
1109 * Explicitly enable no_new_privs here in case it got set
1110 * between the task_struct being duplicated and holding the
1111 * sighand lock. The seccomp state and nnp must be in sync.
1113 if (task_no_new_privs(current))
1114 task_set_no_new_privs(p);
1117 * If the parent gained a seccomp mode after copying thread
1118 * flags and between before we held the sighand lock, we have
1119 * to manually enable the seccomp thread flag here.
1121 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1122 set_tsk_thread_flag(p, TIF_SECCOMP);
1126 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1128 current->clear_child_tid = tidptr;
1130 return task_pid_vnr(current);
1133 static void rt_mutex_init_task(struct task_struct *p)
1135 raw_spin_lock_init(&p->pi_lock);
1136 #ifdef CONFIG_RT_MUTEXES
1137 p->pi_waiters = RB_ROOT;
1138 p->pi_waiters_leftmost = NULL;
1139 p->pi_blocked_on = NULL;
1144 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1148 #endif /* CONFIG_MEMCG */
1151 * Initialize POSIX timer handling for a single task.
1153 static void posix_cpu_timers_init(struct task_struct *tsk)
1155 tsk->cputime_expires.prof_exp = 0;
1156 tsk->cputime_expires.virt_exp = 0;
1157 tsk->cputime_expires.sched_exp = 0;
1158 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1159 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1160 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1164 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1166 task->pids[type].pid = pid;
1170 * This creates a new process as a copy of the old one,
1171 * but does not actually start it yet.
1173 * It copies the registers, and all the appropriate
1174 * parts of the process environment (as per the clone
1175 * flags). The actual kick-off is left to the caller.
1177 static struct task_struct *copy_process(unsigned long clone_flags,
1178 unsigned long stack_start,
1179 unsigned long stack_size,
1180 int __user *child_tidptr,
1185 struct task_struct *p;
1187 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1188 return ERR_PTR(-EINVAL);
1190 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1191 return ERR_PTR(-EINVAL);
1194 * Thread groups must share signals as well, and detached threads
1195 * can only be started up within the thread group.
1197 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1198 return ERR_PTR(-EINVAL);
1201 * Shared signal handlers imply shared VM. By way of the above,
1202 * thread groups also imply shared VM. Blocking this case allows
1203 * for various simplifications in other code.
1205 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1206 return ERR_PTR(-EINVAL);
1209 * Siblings of global init remain as zombies on exit since they are
1210 * not reaped by their parent (swapper). To solve this and to avoid
1211 * multi-rooted process trees, prevent global and container-inits
1212 * from creating siblings.
1214 if ((clone_flags & CLONE_PARENT) &&
1215 current->signal->flags & SIGNAL_UNKILLABLE)
1216 return ERR_PTR(-EINVAL);
1219 * If the new process will be in a different pid or user namespace
1220 * do not allow it to share a thread group or signal handlers or
1221 * parent with the forking task.
1223 if (clone_flags & CLONE_SIGHAND) {
1224 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1225 (task_active_pid_ns(current) !=
1226 current->nsproxy->pid_ns_for_children))
1227 return ERR_PTR(-EINVAL);
1230 retval = security_task_create(clone_flags);
1235 p = dup_task_struct(current);
1239 ftrace_graph_init_task(p);
1241 rt_mutex_init_task(p);
1243 #ifdef CONFIG_PROVE_LOCKING
1244 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1245 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1248 if (atomic_read(&p->real_cred->user->processes) >=
1249 task_rlimit(p, RLIMIT_NPROC)) {
1250 if (p->real_cred->user != INIT_USER &&
1251 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1254 current->flags &= ~PF_NPROC_EXCEEDED;
1256 retval = copy_creds(p, clone_flags);
1261 * If multiple threads are within copy_process(), then this check
1262 * triggers too late. This doesn't hurt, the check is only there
1263 * to stop root fork bombs.
1266 if (nr_threads >= max_threads)
1267 goto bad_fork_cleanup_count;
1269 if (!try_module_get(task_thread_info(p)->exec_domain->module))
1270 goto bad_fork_cleanup_count;
1272 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1273 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1274 p->flags |= PF_FORKNOEXEC;
1275 INIT_LIST_HEAD(&p->children);
1276 INIT_LIST_HEAD(&p->sibling);
1277 rcu_copy_process(p);
1278 p->vfork_done = NULL;
1279 spin_lock_init(&p->alloc_lock);
1281 init_sigpending(&p->pending);
1283 p->utime = p->stime = p->gtime = 0;
1284 p->utimescaled = p->stimescaled = 0;
1285 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1286 p->prev_cputime.utime = p->prev_cputime.stime = 0;
1288 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1289 seqlock_init(&p->vtime_seqlock);
1291 p->vtime_snap_whence = VTIME_SLEEPING;
1294 #if defined(SPLIT_RSS_COUNTING)
1295 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1298 p->default_timer_slack_ns = current->timer_slack_ns;
1300 task_io_accounting_init(&p->ioac);
1301 acct_clear_integrals(p);
1303 posix_cpu_timers_init(p);
1305 p->start_time = ktime_get_ns();
1306 p->real_start_time = ktime_get_boot_ns();
1307 p->io_context = NULL;
1308 p->audit_context = NULL;
1309 if (clone_flags & CLONE_THREAD)
1310 threadgroup_change_begin(current);
1313 p->mempolicy = mpol_dup(p->mempolicy);
1314 if (IS_ERR(p->mempolicy)) {
1315 retval = PTR_ERR(p->mempolicy);
1316 p->mempolicy = NULL;
1317 goto bad_fork_cleanup_threadgroup_lock;
1320 #ifdef CONFIG_CPUSETS
1321 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1322 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1323 seqcount_init(&p->mems_allowed_seq);
1325 #ifdef CONFIG_TRACE_IRQFLAGS
1327 p->hardirqs_enabled = 0;
1328 p->hardirq_enable_ip = 0;
1329 p->hardirq_enable_event = 0;
1330 p->hardirq_disable_ip = _THIS_IP_;
1331 p->hardirq_disable_event = 0;
1332 p->softirqs_enabled = 1;
1333 p->softirq_enable_ip = _THIS_IP_;
1334 p->softirq_enable_event = 0;
1335 p->softirq_disable_ip = 0;
1336 p->softirq_disable_event = 0;
1337 p->hardirq_context = 0;
1338 p->softirq_context = 0;
1340 #ifdef CONFIG_LOCKDEP
1341 p->lockdep_depth = 0; /* no locks held yet */
1342 p->curr_chain_key = 0;
1343 p->lockdep_recursion = 0;
1346 #ifdef CONFIG_DEBUG_MUTEXES
1347 p->blocked_on = NULL; /* not blocked yet */
1350 p->memcg_batch.do_batch = 0;
1351 p->memcg_batch.memcg = NULL;
1353 #ifdef CONFIG_BCACHE
1354 p->sequential_io = 0;
1355 p->sequential_io_avg = 0;
1358 /* Perform scheduler related setup. Assign this task to a CPU. */
1359 retval = sched_fork(clone_flags, p);
1361 goto bad_fork_cleanup_policy;
1363 retval = perf_event_init_task(p);
1365 goto bad_fork_cleanup_policy;
1366 retval = audit_alloc(p);
1368 goto bad_fork_cleanup_policy;
1369 /* copy all the process information */
1370 retval = copy_semundo(clone_flags, p);
1372 goto bad_fork_cleanup_audit;
1373 retval = copy_files(clone_flags, p);
1375 goto bad_fork_cleanup_semundo;
1376 retval = copy_fs(clone_flags, p);
1378 goto bad_fork_cleanup_files;
1379 retval = copy_sighand(clone_flags, p);
1381 goto bad_fork_cleanup_fs;
1382 retval = copy_signal(clone_flags, p);
1384 goto bad_fork_cleanup_sighand;
1385 retval = copy_mm(clone_flags, p);
1387 goto bad_fork_cleanup_signal;
1388 retval = copy_namespaces(clone_flags, p);
1390 goto bad_fork_cleanup_mm;
1391 retval = copy_io(clone_flags, p);
1393 goto bad_fork_cleanup_namespaces;
1394 retval = copy_thread(clone_flags, stack_start, stack_size, p);
1396 goto bad_fork_cleanup_io;
1398 if (pid != &init_struct_pid) {
1400 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1402 goto bad_fork_cleanup_io;
1405 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1407 * Clear TID on mm_release()?
1409 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1414 p->robust_list = NULL;
1415 #ifdef CONFIG_COMPAT
1416 p->compat_robust_list = NULL;
1418 INIT_LIST_HEAD(&p->pi_state_list);
1419 p->pi_state_cache = NULL;
1422 * sigaltstack should be cleared when sharing the same VM
1424 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1425 p->sas_ss_sp = p->sas_ss_size = 0;
1428 * Syscall tracing and stepping should be turned off in the
1429 * child regardless of CLONE_PTRACE.
1431 user_disable_single_step(p);
1432 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1433 #ifdef TIF_SYSCALL_EMU
1434 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1436 clear_all_latency_tracing(p);
1438 /* ok, now we should be set up.. */
1439 p->pid = pid_nr(pid);
1440 if (clone_flags & CLONE_THREAD) {
1441 p->exit_signal = -1;
1442 p->group_leader = current->group_leader;
1443 p->tgid = current->tgid;
1445 if (clone_flags & CLONE_PARENT)
1446 p->exit_signal = current->group_leader->exit_signal;
1448 p->exit_signal = (clone_flags & CSIGNAL);
1449 p->group_leader = p;
1454 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1455 p->dirty_paused_when = 0;
1457 p->pdeath_signal = 0;
1458 INIT_LIST_HEAD(&p->thread_group);
1459 p->task_works = NULL;
1462 * Make it visible to the rest of the system, but dont wake it up yet.
1463 * Need tasklist lock for parent etc handling!
1465 write_lock_irq(&tasklist_lock);
1467 /* CLONE_PARENT re-uses the old parent */
1468 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1469 p->real_parent = current->real_parent;
1470 p->parent_exec_id = current->parent_exec_id;
1472 p->real_parent = current;
1473 p->parent_exec_id = current->self_exec_id;
1476 spin_lock(¤t->sighand->siglock);
1479 * Copy seccomp details explicitly here, in case they were changed
1480 * before holding sighand lock.
1485 * Process group and session signals need to be delivered to just the
1486 * parent before the fork or both the parent and the child after the
1487 * fork. Restart if a signal comes in before we add the new process to
1488 * it's process group.
1489 * A fatal signal pending means that current will exit, so the new
1490 * thread can't slip out of an OOM kill (or normal SIGKILL).
1492 recalc_sigpending();
1493 if (signal_pending(current)) {
1494 spin_unlock(¤t->sighand->siglock);
1495 write_unlock_irq(&tasklist_lock);
1496 retval = -ERESTARTNOINTR;
1497 goto bad_fork_free_pid;
1500 if (likely(p->pid)) {
1501 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1503 init_task_pid(p, PIDTYPE_PID, pid);
1504 if (thread_group_leader(p)) {
1505 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1506 init_task_pid(p, PIDTYPE_SID, task_session(current));
1508 if (is_child_reaper(pid)) {
1509 ns_of_pid(pid)->child_reaper = p;
1510 p->signal->flags |= SIGNAL_UNKILLABLE;
1513 p->signal->leader_pid = pid;
1514 p->signal->tty = tty_kref_get(current->signal->tty);
1515 list_add_tail(&p->sibling, &p->real_parent->children);
1516 list_add_tail_rcu(&p->tasks, &init_task.tasks);
1517 attach_pid(p, PIDTYPE_PGID);
1518 attach_pid(p, PIDTYPE_SID);
1519 __this_cpu_inc(process_counts);
1521 current->signal->nr_threads++;
1522 atomic_inc(¤t->signal->live);
1523 atomic_inc(¤t->signal->sigcnt);
1524 list_add_tail_rcu(&p->thread_group,
1525 &p->group_leader->thread_group);
1526 list_add_tail_rcu(&p->thread_node,
1527 &p->signal->thread_head);
1529 attach_pid(p, PIDTYPE_PID);
1534 spin_unlock(¤t->sighand->siglock);
1535 syscall_tracepoint_update(p);
1536 write_unlock_irq(&tasklist_lock);
1538 proc_fork_connector(p);
1539 cgroup_post_fork(p);
1540 if (clone_flags & CLONE_THREAD)
1541 threadgroup_change_end(current);
1544 trace_task_newtask(p, clone_flags);
1545 uprobe_copy_process(p, clone_flags);
1550 if (pid != &init_struct_pid)
1552 bad_fork_cleanup_io:
1555 bad_fork_cleanup_namespaces:
1556 exit_task_namespaces(p);
1557 bad_fork_cleanup_mm:
1560 bad_fork_cleanup_signal:
1561 if (!(clone_flags & CLONE_THREAD))
1562 free_signal_struct(p->signal);
1563 bad_fork_cleanup_sighand:
1564 __cleanup_sighand(p->sighand);
1565 bad_fork_cleanup_fs:
1566 exit_fs(p); /* blocking */
1567 bad_fork_cleanup_files:
1568 exit_files(p); /* blocking */
1569 bad_fork_cleanup_semundo:
1571 bad_fork_cleanup_audit:
1573 bad_fork_cleanup_policy:
1574 perf_event_free_task(p);
1576 mpol_put(p->mempolicy);
1577 bad_fork_cleanup_threadgroup_lock:
1579 if (clone_flags & CLONE_THREAD)
1580 threadgroup_change_end(current);
1581 delayacct_tsk_free(p);
1582 module_put(task_thread_info(p)->exec_domain->module);
1583 bad_fork_cleanup_count:
1584 atomic_dec(&p->cred->user->processes);
1589 return ERR_PTR(retval);
1592 static inline void init_idle_pids(struct pid_link *links)
1596 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1597 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1598 links[type].pid = &init_struct_pid;
1602 struct task_struct *fork_idle(int cpu)
1604 struct task_struct *task;
1605 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1606 if (!IS_ERR(task)) {
1607 init_idle_pids(task->pids);
1608 init_idle(task, cpu);
1615 * Ok, this is the main fork-routine.
1617 * It copies the process, and if successful kick-starts
1618 * it and waits for it to finish using the VM if required.
1620 long do_fork(unsigned long clone_flags,
1621 unsigned long stack_start,
1622 unsigned long stack_size,
1623 int __user *parent_tidptr,
1624 int __user *child_tidptr)
1626 struct task_struct *p;
1631 * Determine whether and which event to report to ptracer. When
1632 * called from kernel_thread or CLONE_UNTRACED is explicitly
1633 * requested, no event is reported; otherwise, report if the event
1634 * for the type of forking is enabled.
1636 if (!(clone_flags & CLONE_UNTRACED)) {
1637 if (clone_flags & CLONE_VFORK)
1638 trace = PTRACE_EVENT_VFORK;
1639 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1640 trace = PTRACE_EVENT_CLONE;
1642 trace = PTRACE_EVENT_FORK;
1644 if (likely(!ptrace_event_enabled(current, trace)))
1648 p = copy_process(clone_flags, stack_start, stack_size,
1649 child_tidptr, NULL, trace);
1651 * Do this prior waking up the new thread - the thread pointer
1652 * might get invalid after that point, if the thread exits quickly.
1655 struct completion vfork;
1658 trace_sched_process_fork(current, p);
1660 pid = get_task_pid(p, PIDTYPE_PID);
1663 if (clone_flags & CLONE_PARENT_SETTID)
1664 put_user(nr, parent_tidptr);
1666 if (clone_flags & CLONE_VFORK) {
1667 p->vfork_done = &vfork;
1668 init_completion(&vfork);
1672 wake_up_new_task(p);
1674 /* forking complete and child started to run, tell ptracer */
1675 if (unlikely(trace))
1676 ptrace_event_pid(trace, pid);
1678 if (clone_flags & CLONE_VFORK) {
1679 if (!wait_for_vfork_done(p, &vfork))
1680 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1691 * Create a kernel thread.
1693 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1695 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1696 (unsigned long)arg, NULL, NULL);
1699 #ifdef __ARCH_WANT_SYS_FORK
1700 SYSCALL_DEFINE0(fork)
1703 return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1705 /* can not support in nommu mode */
1711 #ifdef __ARCH_WANT_SYS_VFORK
1712 SYSCALL_DEFINE0(vfork)
1714 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1719 #ifdef __ARCH_WANT_SYS_CLONE
1720 #ifdef CONFIG_CLONE_BACKWARDS
1721 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1722 int __user *, parent_tidptr,
1724 int __user *, child_tidptr)
1725 #elif defined(CONFIG_CLONE_BACKWARDS2)
1726 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1727 int __user *, parent_tidptr,
1728 int __user *, child_tidptr,
1730 #elif defined(CONFIG_CLONE_BACKWARDS3)
1731 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1733 int __user *, parent_tidptr,
1734 int __user *, child_tidptr,
1737 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1738 int __user *, parent_tidptr,
1739 int __user *, child_tidptr,
1743 return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1747 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1748 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1751 static void sighand_ctor(void *data)
1753 struct sighand_struct *sighand = data;
1755 spin_lock_init(&sighand->siglock);
1756 init_waitqueue_head(&sighand->signalfd_wqh);
1759 void __init proc_caches_init(void)
1761 sighand_cachep = kmem_cache_create("sighand_cache",
1762 sizeof(struct sighand_struct), 0,
1763 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1764 SLAB_NOTRACK, sighand_ctor);
1765 signal_cachep = kmem_cache_create("signal_cache",
1766 sizeof(struct signal_struct), 0,
1767 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1768 files_cachep = kmem_cache_create("files_cache",
1769 sizeof(struct files_struct), 0,
1770 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1771 fs_cachep = kmem_cache_create("fs_cache",
1772 sizeof(struct fs_struct), 0,
1773 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1775 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1776 * whole struct cpumask for the OFFSTACK case. We could change
1777 * this to *only* allocate as much of it as required by the
1778 * maximum number of CPU's we can ever have. The cpumask_allocation
1779 * is at the end of the structure, exactly for that reason.
1781 mm_cachep = kmem_cache_create("mm_struct",
1782 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1783 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1784 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1786 nsproxy_cache_init();
1790 * Check constraints on flags passed to the unshare system call.
1792 static int check_unshare_flags(unsigned long unshare_flags)
1794 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1795 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1796 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1797 CLONE_NEWUSER|CLONE_NEWPID))
1800 * Not implemented, but pretend it works if there is nothing to
1801 * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1802 * needs to unshare vm.
1804 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1805 /* FIXME: get_task_mm() increments ->mm_users */
1806 if (atomic_read(¤t->mm->mm_users) > 1)
1814 * Unshare the filesystem structure if it is being shared
1816 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1818 struct fs_struct *fs = current->fs;
1820 if (!(unshare_flags & CLONE_FS) || !fs)
1823 /* don't need lock here; in the worst case we'll do useless copy */
1827 *new_fsp = copy_fs_struct(fs);
1835 * Unshare file descriptor table if it is being shared
1837 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1839 struct files_struct *fd = current->files;
1842 if ((unshare_flags & CLONE_FILES) &&
1843 (fd && atomic_read(&fd->count) > 1)) {
1844 *new_fdp = dup_fd(fd, &error);
1853 * unshare allows a process to 'unshare' part of the process
1854 * context which was originally shared using clone. copy_*
1855 * functions used by do_fork() cannot be used here directly
1856 * because they modify an inactive task_struct that is being
1857 * constructed. Here we are modifying the current, active,
1860 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1862 struct fs_struct *fs, *new_fs = NULL;
1863 struct files_struct *fd, *new_fd = NULL;
1864 struct cred *new_cred = NULL;
1865 struct nsproxy *new_nsproxy = NULL;
1870 * If unsharing a user namespace must also unshare the thread.
1872 if (unshare_flags & CLONE_NEWUSER)
1873 unshare_flags |= CLONE_THREAD | CLONE_FS;
1875 * If unsharing a thread from a thread group, must also unshare vm.
1877 if (unshare_flags & CLONE_THREAD)
1878 unshare_flags |= CLONE_VM;
1880 * If unsharing vm, must also unshare signal handlers.
1882 if (unshare_flags & CLONE_VM)
1883 unshare_flags |= CLONE_SIGHAND;
1885 * If unsharing namespace, must also unshare filesystem information.
1887 if (unshare_flags & CLONE_NEWNS)
1888 unshare_flags |= CLONE_FS;
1890 err = check_unshare_flags(unshare_flags);
1892 goto bad_unshare_out;
1894 * CLONE_NEWIPC must also detach from the undolist: after switching
1895 * to a new ipc namespace, the semaphore arrays from the old
1896 * namespace are unreachable.
1898 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1900 err = unshare_fs(unshare_flags, &new_fs);
1902 goto bad_unshare_out;
1903 err = unshare_fd(unshare_flags, &new_fd);
1905 goto bad_unshare_cleanup_fs;
1906 err = unshare_userns(unshare_flags, &new_cred);
1908 goto bad_unshare_cleanup_fd;
1909 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1912 goto bad_unshare_cleanup_cred;
1914 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1917 * CLONE_SYSVSEM is equivalent to sys_exit().
1923 switch_task_namespaces(current, new_nsproxy);
1929 spin_lock(&fs->lock);
1930 current->fs = new_fs;
1935 spin_unlock(&fs->lock);
1939 fd = current->files;
1940 current->files = new_fd;
1944 task_unlock(current);
1947 /* Install the new user namespace */
1948 commit_creds(new_cred);
1953 bad_unshare_cleanup_cred:
1956 bad_unshare_cleanup_fd:
1958 put_files_struct(new_fd);
1960 bad_unshare_cleanup_fs:
1962 free_fs_struct(new_fs);
1969 * Helper to unshare the files of the current task.
1970 * We don't want to expose copy_files internals to
1971 * the exec layer of the kernel.
1974 int unshare_files(struct files_struct **displaced)
1976 struct task_struct *task = current;
1977 struct files_struct *copy = NULL;
1980 error = unshare_fd(CLONE_FILES, ©);
1981 if (error || !copy) {
1985 *displaced = task->files;