Root/kernel/fork.c

1/*
2 * linux/kernel/fork.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
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()'
12 */
13
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>
30#include <linux/fs.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>
74
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>
81
82#include <trace/events/sched.h>
83
84#define CREATE_TRACE_POINTS
85#include <trace/events/task.h>
86
87/*
88 * Protected counters by write_lock_irq(&tasklist_lock)
89 */
90unsigned long total_forks; /* Handle normal Linux uptimes. */
91int nr_threads; /* The idle threads do not count.. */
92
93int max_threads; /* tunable limit on nr_threads */
94
95DEFINE_PER_CPU(unsigned long, process_counts) = 0;
96
97__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
98
99#ifdef CONFIG_PROVE_RCU
100int lockdep_tasklist_lock_is_held(void)
101{
102    return lockdep_is_held(&tasklist_lock);
103}
104EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
105#endif /* #ifdef CONFIG_PROVE_RCU */
106
107int nr_processes(void)
108{
109    int cpu;
110    int total = 0;
111
112    for_each_possible_cpu(cpu)
113        total += per_cpu(process_counts, cpu);
114
115    return total;
116}
117
118void __weak arch_release_task_struct(struct task_struct *tsk)
119{
120}
121
122#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
123static struct kmem_cache *task_struct_cachep;
124
125static inline struct task_struct *alloc_task_struct_node(int node)
126{
127    return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
128}
129
130static inline void free_task_struct(struct task_struct *tsk)
131{
132    kmem_cache_free(task_struct_cachep, tsk);
133}
134#endif
135
136void __weak arch_release_thread_info(struct thread_info *ti)
137{
138}
139
140#ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
141
142/*
143 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
144 * kmemcache based allocator.
145 */
146# if THREAD_SIZE >= PAGE_SIZE
147static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
148                          int node)
149{
150    struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
151                         THREAD_SIZE_ORDER);
152
153    return page ? page_address(page) : NULL;
154}
155
156static inline void free_thread_info(struct thread_info *ti)
157{
158    free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
159}
160# else
161static struct kmem_cache *thread_info_cache;
162
163static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
164                          int node)
165{
166    return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
167}
168
169static void free_thread_info(struct thread_info *ti)
170{
171    kmem_cache_free(thread_info_cache, ti);
172}
173
174void thread_info_cache_init(void)
175{
176    thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
177                          THREAD_SIZE, 0, NULL);
178    BUG_ON(thread_info_cache == NULL);
179}
180# endif
181#endif
182
183/* SLAB cache for signal_struct structures (tsk->signal) */
184static struct kmem_cache *signal_cachep;
185
186/* SLAB cache for sighand_struct structures (tsk->sighand) */
187struct kmem_cache *sighand_cachep;
188
189/* SLAB cache for files_struct structures (tsk->files) */
190struct kmem_cache *files_cachep;
191
192/* SLAB cache for fs_struct structures (tsk->fs) */
193struct kmem_cache *fs_cachep;
194
195/* SLAB cache for vm_area_struct structures */
196struct kmem_cache *vm_area_cachep;
197
198/* SLAB cache for mm_struct structures (tsk->mm) */
199static struct kmem_cache *mm_cachep;
200
201static void account_kernel_stack(struct thread_info *ti, int account)
202{
203    struct zone *zone = page_zone(virt_to_page(ti));
204
205    mod_zone_page_state(zone, NR_KERNEL_STACK, account);
206}
207
208void free_task(struct task_struct *tsk)
209{
210    account_kernel_stack(tsk->stack, -1);
211    arch_release_thread_info(tsk->stack);
212    free_thread_info(tsk->stack);
213    rt_mutex_debug_task_free(tsk);
214    ftrace_graph_exit_task(tsk);
215    put_seccomp_filter(tsk);
216    arch_release_task_struct(tsk);
217    free_task_struct(tsk);
218}
219EXPORT_SYMBOL(free_task);
220
221static inline void free_signal_struct(struct signal_struct *sig)
222{
223    taskstats_tgid_free(sig);
224    sched_autogroup_exit(sig);
225    kmem_cache_free(signal_cachep, sig);
226}
227
228static inline void put_signal_struct(struct signal_struct *sig)
229{
230    if (atomic_dec_and_test(&sig->sigcnt))
231        free_signal_struct(sig);
232}
233
234void __put_task_struct(struct task_struct *tsk)
235{
236    WARN_ON(!tsk->exit_state);
237    WARN_ON(atomic_read(&tsk->usage));
238    WARN_ON(tsk == current);
239
240    security_task_free(tsk);
241    exit_creds(tsk);
242    delayacct_tsk_free(tsk);
243    put_signal_struct(tsk->signal);
244
245    if (!profile_handoff_task(tsk))
246        free_task(tsk);
247}
248EXPORT_SYMBOL_GPL(__put_task_struct);
249
250void __init __weak arch_task_cache_init(void) { }
251
252void __init fork_init(unsigned long mempages)
253{
254#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
255#ifndef ARCH_MIN_TASKALIGN
256#define ARCH_MIN_TASKALIGN L1_CACHE_BYTES
257#endif
258    /* create a slab on which task_structs can be allocated */
259    task_struct_cachep =
260        kmem_cache_create("task_struct", sizeof(struct task_struct),
261            ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
262#endif
263
264    /* do the arch specific task caches init */
265    arch_task_cache_init();
266
267    /*
268     * The default maximum number of threads is set to a safe
269     * value: the thread structures can take up at most half
270     * of memory.
271     */
272    max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
273
274    /*
275     * we need to allow at least 20 threads to boot a system
276     */
277    if (max_threads < 20)
278        max_threads = 20;
279
280    init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
281    init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
282    init_task.signal->rlim[RLIMIT_SIGPENDING] =
283        init_task.signal->rlim[RLIMIT_NPROC];
284}
285
286int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
287                           struct task_struct *src)
288{
289    *dst = *src;
290    return 0;
291}
292
293static struct task_struct *dup_task_struct(struct task_struct *orig)
294{
295    struct task_struct *tsk;
296    struct thread_info *ti;
297    unsigned long *stackend;
298    int node = tsk_fork_get_node(orig);
299    int err;
300
301    tsk = alloc_task_struct_node(node);
302    if (!tsk)
303        return NULL;
304
305    ti = alloc_thread_info_node(tsk, node);
306    if (!ti)
307        goto free_tsk;
308
309    err = arch_dup_task_struct(tsk, orig);
310    if (err)
311        goto free_ti;
312
313    tsk->stack = ti;
314
315    setup_thread_stack(tsk, orig);
316    clear_user_return_notifier(tsk);
317    clear_tsk_need_resched(tsk);
318    stackend = end_of_stack(tsk);
319    *stackend = STACK_END_MAGIC; /* for overflow detection */
320
321#ifdef CONFIG_CC_STACKPROTECTOR
322    tsk->stack_canary = get_random_int();
323#endif
324
325    /*
326     * One for us, one for whoever does the "release_task()" (usually
327     * parent)
328     */
329    atomic_set(&tsk->usage, 2);
330#ifdef CONFIG_BLK_DEV_IO_TRACE
331    tsk->btrace_seq = 0;
332#endif
333    tsk->splice_pipe = NULL;
334    tsk->task_frag.page = NULL;
335
336    account_kernel_stack(ti, 1);
337
338    return tsk;
339
340free_ti:
341    free_thread_info(ti);
342free_tsk:
343    free_task_struct(tsk);
344    return NULL;
345}
346
347#ifdef CONFIG_MMU
348static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
349{
350    struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
351    struct rb_node **rb_link, *rb_parent;
352    int retval;
353    unsigned long charge;
354
355    uprobe_start_dup_mmap();
356    down_write(&oldmm->mmap_sem);
357    flush_cache_dup_mm(oldmm);
358    uprobe_dup_mmap(oldmm, mm);
359    /*
360     * Not linked in yet - no deadlock potential:
361     */
362    down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
363
364    mm->locked_vm = 0;
365    mm->mmap = NULL;
366    mm->mmap_cache = NULL;
367    mm->map_count = 0;
368    cpumask_clear(mm_cpumask(mm));
369    mm->mm_rb = RB_ROOT;
370    rb_link = &mm->mm_rb.rb_node;
371    rb_parent = NULL;
372    pprev = &mm->mmap;
373    retval = ksm_fork(mm, oldmm);
374    if (retval)
375        goto out;
376    retval = khugepaged_fork(mm, oldmm);
377    if (retval)
378        goto out;
379
380    prev = NULL;
381    for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
382        struct file *file;
383
384        if (mpnt->vm_flags & VM_DONTCOPY) {
385            vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
386                            -vma_pages(mpnt));
387            continue;
388        }
389        charge = 0;
390        if (mpnt->vm_flags & VM_ACCOUNT) {
391            unsigned long len = vma_pages(mpnt);
392
393            if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
394                goto fail_nomem;
395            charge = len;
396        }
397        tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
398        if (!tmp)
399            goto fail_nomem;
400        *tmp = *mpnt;
401        INIT_LIST_HEAD(&tmp->anon_vma_chain);
402        retval = vma_dup_policy(mpnt, tmp);
403        if (retval)
404            goto fail_nomem_policy;
405        tmp->vm_mm = mm;
406        if (anon_vma_fork(tmp, mpnt))
407            goto fail_nomem_anon_vma_fork;
408        tmp->vm_flags &= ~VM_LOCKED;
409        tmp->vm_next = tmp->vm_prev = NULL;
410        file = tmp->vm_file;
411        if (file) {
412            struct inode *inode = file_inode(file);
413            struct address_space *mapping = file->f_mapping;
414
415            get_file(file);
416            if (tmp->vm_flags & VM_DENYWRITE)
417                atomic_dec(&inode->i_writecount);
418            mutex_lock(&mapping->i_mmap_mutex);
419            if (tmp->vm_flags & VM_SHARED)
420                mapping->i_mmap_writable++;
421            flush_dcache_mmap_lock(mapping);
422            /* insert tmp into the share list, just after mpnt */
423            if (unlikely(tmp->vm_flags & VM_NONLINEAR))
424                vma_nonlinear_insert(tmp,
425                        &mapping->i_mmap_nonlinear);
426            else
427                vma_interval_tree_insert_after(tmp, mpnt,
428                            &mapping->i_mmap);
429            flush_dcache_mmap_unlock(mapping);
430            mutex_unlock(&mapping->i_mmap_mutex);
431        }
432
433        /*
434         * Clear hugetlb-related page reserves for children. This only
435         * affects MAP_PRIVATE mappings. Faults generated by the child
436         * are not guaranteed to succeed, even if read-only
437         */
438        if (is_vm_hugetlb_page(tmp))
439            reset_vma_resv_huge_pages(tmp);
440
441        /*
442         * Link in the new vma and copy the page table entries.
443         */
444        *pprev = tmp;
445        pprev = &tmp->vm_next;
446        tmp->vm_prev = prev;
447        prev = tmp;
448
449        __vma_link_rb(mm, tmp, rb_link, rb_parent);
450        rb_link = &tmp->vm_rb.rb_right;
451        rb_parent = &tmp->vm_rb;
452
453        mm->map_count++;
454        retval = copy_page_range(mm, oldmm, mpnt);
455
456        if (tmp->vm_ops && tmp->vm_ops->open)
457            tmp->vm_ops->open(tmp);
458
459        if (retval)
460            goto out;
461    }
462    /* a new mm has just been created */
463    arch_dup_mmap(oldmm, mm);
464    retval = 0;
465out:
466    up_write(&mm->mmap_sem);
467    flush_tlb_mm(oldmm);
468    up_write(&oldmm->mmap_sem);
469    uprobe_end_dup_mmap();
470    return retval;
471fail_nomem_anon_vma_fork:
472    mpol_put(vma_policy(tmp));
473fail_nomem_policy:
474    kmem_cache_free(vm_area_cachep, tmp);
475fail_nomem:
476    retval = -ENOMEM;
477    vm_unacct_memory(charge);
478    goto out;
479}
480
481static inline int mm_alloc_pgd(struct mm_struct *mm)
482{
483    mm->pgd = pgd_alloc(mm);
484    if (unlikely(!mm->pgd))
485        return -ENOMEM;
486    return 0;
487}
488
489static inline void mm_free_pgd(struct mm_struct *mm)
490{
491    pgd_free(mm, mm->pgd);
492}
493#else
494#define dup_mmap(mm, oldmm) (0)
495#define mm_alloc_pgd(mm) (0)
496#define mm_free_pgd(mm)
497#endif /* CONFIG_MMU */
498
499__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
500
501#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
502#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
503
504static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
505
506static int __init coredump_filter_setup(char *s)
507{
508    default_dump_filter =
509        (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
510        MMF_DUMP_FILTER_MASK;
511    return 1;
512}
513
514__setup("coredump_filter=", coredump_filter_setup);
515
516#include <linux/init_task.h>
517
518static void mm_init_aio(struct mm_struct *mm)
519{
520#ifdef CONFIG_AIO
521    spin_lock_init(&mm->ioctx_lock);
522    mm->ioctx_table = NULL;
523#endif
524}
525
526static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
527{
528    atomic_set(&mm->mm_users, 1);
529    atomic_set(&mm->mm_count, 1);
530    init_rwsem(&mm->mmap_sem);
531    INIT_LIST_HEAD(&mm->mmlist);
532    mm->flags = (current->mm) ?
533        (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
534    mm->core_state = NULL;
535    atomic_long_set(&mm->nr_ptes, 0);
536    memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
537    spin_lock_init(&mm->page_table_lock);
538    mm_init_aio(mm);
539    mm_init_owner(mm, p);
540    clear_tlb_flush_pending(mm);
541
542    if (likely(!mm_alloc_pgd(mm))) {
543        mm->def_flags = 0;
544        mmu_notifier_mm_init(mm);
545        return mm;
546    }
547
548    free_mm(mm);
549    return NULL;
550}
551
552static void check_mm(struct mm_struct *mm)
553{
554    int i;
555
556    for (i = 0; i < NR_MM_COUNTERS; i++) {
557        long x = atomic_long_read(&mm->rss_stat.count[i]);
558
559        if (unlikely(x))
560            printk(KERN_ALERT "BUG: Bad rss-counter state "
561                      "mm:%p idx:%d val:%ld\n", mm, i, x);
562    }
563
564#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
565    VM_BUG_ON(mm->pmd_huge_pte);
566#endif
567}
568
569/*
570 * Allocate and initialize an mm_struct.
571 */
572struct mm_struct *mm_alloc(void)
573{
574    struct mm_struct *mm;
575
576    mm = allocate_mm();
577    if (!mm)
578        return NULL;
579
580    memset(mm, 0, sizeof(*mm));
581    mm_init_cpumask(mm);
582    return mm_init(mm, current);
583}
584
585/*
586 * Called when the last reference to the mm
587 * is dropped: either by a lazy thread or by
588 * mmput. Free the page directory and the mm.
589 */
590void __mmdrop(struct mm_struct *mm)
591{
592    BUG_ON(mm == &init_mm);
593    mm_free_pgd(mm);
594    destroy_context(mm);
595    mmu_notifier_mm_destroy(mm);
596    check_mm(mm);
597    free_mm(mm);
598}
599EXPORT_SYMBOL_GPL(__mmdrop);
600
601/*
602 * Decrement the use count and release all resources for an mm.
603 */
604void mmput(struct mm_struct *mm)
605{
606    might_sleep();
607
608    if (atomic_dec_and_test(&mm->mm_users)) {
609        uprobe_clear_state(mm);
610        exit_aio(mm);
611        ksm_exit(mm);
612        khugepaged_exit(mm); /* must run before exit_mmap */
613        exit_mmap(mm);
614        set_mm_exe_file(mm, NULL);
615        if (!list_empty(&mm->mmlist)) {
616            spin_lock(&mmlist_lock);
617            list_del(&mm->mmlist);
618            spin_unlock(&mmlist_lock);
619        }
620        if (mm->binfmt)
621            module_put(mm->binfmt->module);
622        mmdrop(mm);
623    }
624}
625EXPORT_SYMBOL_GPL(mmput);
626
627void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
628{
629    if (new_exe_file)
630        get_file(new_exe_file);
631    if (mm->exe_file)
632        fput(mm->exe_file);
633    mm->exe_file = new_exe_file;
634}
635
636struct file *get_mm_exe_file(struct mm_struct *mm)
637{
638    struct file *exe_file;
639
640    /* We need mmap_sem to protect against races with removal of exe_file */
641    down_read(&mm->mmap_sem);
642    exe_file = mm->exe_file;
643    if (exe_file)
644        get_file(exe_file);
645    up_read(&mm->mmap_sem);
646    return exe_file;
647}
648
649static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
650{
651    /* It's safe to write the exe_file pointer without exe_file_lock because
652     * this is called during fork when the task is not yet in /proc */
653    newmm->exe_file = get_mm_exe_file(oldmm);
654}
655
656/**
657 * get_task_mm - acquire a reference to the task's mm
658 *
659 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
660 * this kernel workthread has transiently adopted a user mm with use_mm,
661 * to do its AIO) is not set and if so returns a reference to it, after
662 * bumping up the use count. User must release the mm via mmput()
663 * after use. Typically used by /proc and ptrace.
664 */
665struct mm_struct *get_task_mm(struct task_struct *task)
666{
667    struct mm_struct *mm;
668
669    task_lock(task);
670    mm = task->mm;
671    if (mm) {
672        if (task->flags & PF_KTHREAD)
673            mm = NULL;
674        else
675            atomic_inc(&mm->mm_users);
676    }
677    task_unlock(task);
678    return mm;
679}
680EXPORT_SYMBOL_GPL(get_task_mm);
681
682struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
683{
684    struct mm_struct *mm;
685    int err;
686
687    err = mutex_lock_killable(&task->signal->cred_guard_mutex);
688    if (err)
689        return ERR_PTR(err);
690
691    mm = get_task_mm(task);
692    if (mm && mm != current->mm &&
693            !ptrace_may_access(task, mode)) {
694        mmput(mm);
695        mm = ERR_PTR(-EACCES);
696    }
697    mutex_unlock(&task->signal->cred_guard_mutex);
698
699    return mm;
700}
701
702static void complete_vfork_done(struct task_struct *tsk)
703{
704    struct completion *vfork;
705
706    task_lock(tsk);
707    vfork = tsk->vfork_done;
708    if (likely(vfork)) {
709        tsk->vfork_done = NULL;
710        complete(vfork);
711    }
712    task_unlock(tsk);
713}
714
715static int wait_for_vfork_done(struct task_struct *child,
716                struct completion *vfork)
717{
718    int killed;
719
720    freezer_do_not_count();
721    killed = wait_for_completion_killable(vfork);
722    freezer_count();
723
724    if (killed) {
725        task_lock(child);
726        child->vfork_done = NULL;
727        task_unlock(child);
728    }
729
730    put_task_struct(child);
731    return killed;
732}
733
734/* Please note the differences between mmput and mm_release.
735 * mmput is called whenever we stop holding onto a mm_struct,
736 * error success whatever.
737 *
738 * mm_release is called after a mm_struct has been removed
739 * from the current process.
740 *
741 * This difference is important for error handling, when we
742 * only half set up a mm_struct for a new process and need to restore
743 * the old one. Because we mmput the new mm_struct before
744 * restoring the old one. . .
745 * Eric Biederman 10 January 1998
746 */
747void mm_release(struct task_struct *tsk, struct mm_struct *mm)
748{
749    /* Get rid of any futexes when releasing the mm */
750#ifdef CONFIG_FUTEX
751    if (unlikely(tsk->robust_list)) {
752        exit_robust_list(tsk);
753        tsk->robust_list = NULL;
754    }
755#ifdef CONFIG_COMPAT
756    if (unlikely(tsk->compat_robust_list)) {
757        compat_exit_robust_list(tsk);
758        tsk->compat_robust_list = NULL;
759    }
760#endif
761    if (unlikely(!list_empty(&tsk->pi_state_list)))
762        exit_pi_state_list(tsk);
763#endif
764
765    uprobe_free_utask(tsk);
766
767    /* Get rid of any cached register state */
768    deactivate_mm(tsk, mm);
769
770    /*
771     * If we're exiting normally, clear a user-space tid field if
772     * requested. We leave this alone when dying by signal, to leave
773     * the value intact in a core dump, and to save the unnecessary
774     * trouble, say, a killed vfork parent shouldn't touch this mm.
775     * Userland only wants this done for a sys_exit.
776     */
777    if (tsk->clear_child_tid) {
778        if (!(tsk->flags & PF_SIGNALED) &&
779            atomic_read(&mm->mm_users) > 1) {
780            /*
781             * We don't check the error code - if userspace has
782             * not set up a proper pointer then tough luck.
783             */
784            put_user(0, tsk->clear_child_tid);
785            sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
786                    1, NULL, NULL, 0);
787        }
788        tsk->clear_child_tid = NULL;
789    }
790
791    /*
792     * All done, finally we can wake up parent and return this mm to him.
793     * Also kthread_stop() uses this completion for synchronization.
794     */
795    if (tsk->vfork_done)
796        complete_vfork_done(tsk);
797}
798
799/*
800 * Allocate a new mm structure and copy contents from the
801 * mm structure of the passed in task structure.
802 */
803struct mm_struct *dup_mm(struct task_struct *tsk)
804{
805    struct mm_struct *mm, *oldmm = current->mm;
806    int err;
807
808    if (!oldmm)
809        return NULL;
810
811    mm = allocate_mm();
812    if (!mm)
813        goto fail_nomem;
814
815    memcpy(mm, oldmm, sizeof(*mm));
816    mm_init_cpumask(mm);
817
818#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
819    mm->pmd_huge_pte = NULL;
820#endif
821    if (!mm_init(mm, tsk))
822        goto fail_nomem;
823
824    if (init_new_context(tsk, mm))
825        goto fail_nocontext;
826
827    dup_mm_exe_file(oldmm, mm);
828
829    err = dup_mmap(mm, oldmm);
830    if (err)
831        goto free_pt;
832
833    mm->hiwater_rss = get_mm_rss(mm);
834    mm->hiwater_vm = mm->total_vm;
835
836    if (mm->binfmt && !try_module_get(mm->binfmt->module))
837        goto free_pt;
838
839    return mm;
840
841free_pt:
842    /* don't put binfmt in mmput, we haven't got module yet */
843    mm->binfmt = NULL;
844    mmput(mm);
845
846fail_nomem:
847    return NULL;
848
849fail_nocontext:
850    /*
851     * If init_new_context() failed, we cannot use mmput() to free the mm
852     * because it calls destroy_context()
853     */
854    mm_free_pgd(mm);
855    free_mm(mm);
856    return NULL;
857}
858
859static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
860{
861    struct mm_struct *mm, *oldmm;
862    int retval;
863
864    tsk->min_flt = tsk->maj_flt = 0;
865    tsk->nvcsw = tsk->nivcsw = 0;
866#ifdef CONFIG_DETECT_HUNG_TASK
867    tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
868#endif
869
870    tsk->mm = NULL;
871    tsk->active_mm = NULL;
872
873    /*
874     * Are we cloning a kernel thread?
875     *
876     * We need to steal a active VM for that..
877     */
878    oldmm = current->mm;
879    if (!oldmm)
880        return 0;
881
882    if (clone_flags & CLONE_VM) {
883        atomic_inc(&oldmm->mm_users);
884        mm = oldmm;
885        goto good_mm;
886    }
887
888    retval = -ENOMEM;
889    mm = dup_mm(tsk);
890    if (!mm)
891        goto fail_nomem;
892
893good_mm:
894    tsk->mm = mm;
895    tsk->active_mm = mm;
896    return 0;
897
898fail_nomem:
899    return retval;
900}
901
902static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
903{
904    struct fs_struct *fs = current->fs;
905    if (clone_flags & CLONE_FS) {
906        /* tsk->fs is already what we want */
907        spin_lock(&fs->lock);
908        if (fs->in_exec) {
909            spin_unlock(&fs->lock);
910            return -EAGAIN;
911        }
912        fs->users++;
913        spin_unlock(&fs->lock);
914        return 0;
915    }
916    tsk->fs = copy_fs_struct(fs);
917    if (!tsk->fs)
918        return -ENOMEM;
919    return 0;
920}
921
922static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
923{
924    struct files_struct *oldf, *newf;
925    int error = 0;
926
927    /*
928     * A background process may not have any files ...
929     */
930    oldf = current->files;
931    if (!oldf)
932        goto out;
933
934    if (clone_flags & CLONE_FILES) {
935        atomic_inc(&oldf->count);
936        goto out;
937    }
938
939    newf = dup_fd(oldf, &error);
940    if (!newf)
941        goto out;
942
943    tsk->files = newf;
944    error = 0;
945out:
946    return error;
947}
948
949static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
950{
951#ifdef CONFIG_BLOCK
952    struct io_context *ioc = current->io_context;
953    struct io_context *new_ioc;
954
955    if (!ioc)
956        return 0;
957    /*
958     * Share io context with parent, if CLONE_IO is set
959     */
960    if (clone_flags & CLONE_IO) {
961        ioc_task_link(ioc);
962        tsk->io_context = ioc;
963    } else if (ioprio_valid(ioc->ioprio)) {
964        new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
965        if (unlikely(!new_ioc))
966            return -ENOMEM;
967
968        new_ioc->ioprio = ioc->ioprio;
969        put_io_context(new_ioc);
970    }
971#endif
972    return 0;
973}
974
975static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
976{
977    struct sighand_struct *sig;
978
979    if (clone_flags & CLONE_SIGHAND) {
980        atomic_inc(&current->sighand->count);
981        return 0;
982    }
983    sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
984    rcu_assign_pointer(tsk->sighand, sig);
985    if (!sig)
986        return -ENOMEM;
987    atomic_set(&sig->count, 1);
988    memcpy(sig->action, current->sighand->action, sizeof(sig->action));
989    return 0;
990}
991
992void __cleanup_sighand(struct sighand_struct *sighand)
993{
994    if (atomic_dec_and_test(&sighand->count)) {
995        signalfd_cleanup(sighand);
996        kmem_cache_free(sighand_cachep, sighand);
997    }
998}
999
1000
1001/*
1002 * Initialize POSIX timer handling for a thread group.
1003 */
1004static void posix_cpu_timers_init_group(struct signal_struct *sig)
1005{
1006    unsigned long cpu_limit;
1007
1008    /* Thread group counters. */
1009    thread_group_cputime_init(sig);
1010
1011    cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1012    if (cpu_limit != RLIM_INFINITY) {
1013        sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1014        sig->cputimer.running = 1;
1015    }
1016
1017    /* The timer lists. */
1018    INIT_LIST_HEAD(&sig->cpu_timers[0]);
1019    INIT_LIST_HEAD(&sig->cpu_timers[1]);
1020    INIT_LIST_HEAD(&sig->cpu_timers[2]);
1021}
1022
1023static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1024{
1025    struct signal_struct *sig;
1026
1027    if (clone_flags & CLONE_THREAD)
1028        return 0;
1029
1030    sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1031    tsk->signal = sig;
1032    if (!sig)
1033        return -ENOMEM;
1034
1035    sig->nr_threads = 1;
1036    atomic_set(&sig->live, 1);
1037    atomic_set(&sig->sigcnt, 1);
1038    init_waitqueue_head(&sig->wait_chldexit);
1039    sig->curr_target = tsk;
1040    init_sigpending(&sig->shared_pending);
1041    INIT_LIST_HEAD(&sig->posix_timers);
1042
1043    hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1044    sig->real_timer.function = it_real_fn;
1045
1046    task_lock(current->group_leader);
1047    memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1048    task_unlock(current->group_leader);
1049
1050    posix_cpu_timers_init_group(sig);
1051
1052    tty_audit_fork(sig);
1053    sched_autogroup_fork(sig);
1054
1055#ifdef CONFIG_CGROUPS
1056    init_rwsem(&sig->group_rwsem);
1057#endif
1058
1059    sig->oom_score_adj = current->signal->oom_score_adj;
1060    sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1061
1062    sig->has_child_subreaper = current->signal->has_child_subreaper ||
1063                   current->signal->is_child_subreaper;
1064
1065    mutex_init(&sig->cred_guard_mutex);
1066
1067    return 0;
1068}
1069
1070static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1071{
1072    unsigned long new_flags = p->flags;
1073
1074    new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1075    new_flags |= PF_FORKNOEXEC;
1076    p->flags = new_flags;
1077}
1078
1079SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1080{
1081    current->clear_child_tid = tidptr;
1082
1083    return task_pid_vnr(current);
1084}
1085
1086static void rt_mutex_init_task(struct task_struct *p)
1087{
1088    raw_spin_lock_init(&p->pi_lock);
1089#ifdef CONFIG_RT_MUTEXES
1090    plist_head_init(&p->pi_waiters);
1091    p->pi_blocked_on = NULL;
1092#endif
1093}
1094
1095#ifdef CONFIG_MM_OWNER
1096void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1097{
1098    mm->owner = p;
1099}
1100#endif /* CONFIG_MM_OWNER */
1101
1102/*
1103 * Initialize POSIX timer handling for a single task.
1104 */
1105static void posix_cpu_timers_init(struct task_struct *tsk)
1106{
1107    tsk->cputime_expires.prof_exp = 0;
1108    tsk->cputime_expires.virt_exp = 0;
1109    tsk->cputime_expires.sched_exp = 0;
1110    INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1111    INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1112    INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1113}
1114
1115static inline void
1116init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1117{
1118     task->pids[type].pid = pid;
1119}
1120
1121/*
1122 * This creates a new process as a copy of the old one,
1123 * but does not actually start it yet.
1124 *
1125 * It copies the registers, and all the appropriate
1126 * parts of the process environment (as per the clone
1127 * flags). The actual kick-off is left to the caller.
1128 */
1129static struct task_struct *copy_process(unsigned long clone_flags,
1130                    unsigned long stack_start,
1131                    unsigned long stack_size,
1132                    int __user *child_tidptr,
1133                    struct pid *pid,
1134                    int trace)
1135{
1136    int retval;
1137    struct task_struct *p;
1138
1139    if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1140        return ERR_PTR(-EINVAL);
1141
1142    if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1143        return ERR_PTR(-EINVAL);
1144
1145    /*
1146     * Thread groups must share signals as well, and detached threads
1147     * can only be started up within the thread group.
1148     */
1149    if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1150        return ERR_PTR(-EINVAL);
1151
1152    /*
1153     * Shared signal handlers imply shared VM. By way of the above,
1154     * thread groups also imply shared VM. Blocking this case allows
1155     * for various simplifications in other code.
1156     */
1157    if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1158        return ERR_PTR(-EINVAL);
1159
1160    /*
1161     * Siblings of global init remain as zombies on exit since they are
1162     * not reaped by their parent (swapper). To solve this and to avoid
1163     * multi-rooted process trees, prevent global and container-inits
1164     * from creating siblings.
1165     */
1166    if ((clone_flags & CLONE_PARENT) &&
1167                current->signal->flags & SIGNAL_UNKILLABLE)
1168        return ERR_PTR(-EINVAL);
1169
1170    /*
1171     * If the new process will be in a different pid or user namespace
1172     * do not allow it to share a thread group or signal handlers or
1173     * parent with the forking task.
1174     */
1175    if (clone_flags & CLONE_SIGHAND) {
1176        if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1177            (task_active_pid_ns(current) !=
1178                current->nsproxy->pid_ns_for_children))
1179            return ERR_PTR(-EINVAL);
1180    }
1181
1182    retval = security_task_create(clone_flags);
1183    if (retval)
1184        goto fork_out;
1185
1186    retval = -ENOMEM;
1187    p = dup_task_struct(current);
1188    if (!p)
1189        goto fork_out;
1190
1191    ftrace_graph_init_task(p);
1192    get_seccomp_filter(p);
1193
1194    rt_mutex_init_task(p);
1195
1196#ifdef CONFIG_PROVE_LOCKING
1197    DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1198    DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1199#endif
1200    retval = -EAGAIN;
1201    if (atomic_read(&p->real_cred->user->processes) >=
1202            task_rlimit(p, RLIMIT_NPROC)) {
1203        if (p->real_cred->user != INIT_USER &&
1204            !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1205            goto bad_fork_free;
1206    }
1207    current->flags &= ~PF_NPROC_EXCEEDED;
1208
1209    retval = copy_creds(p, clone_flags);
1210    if (retval < 0)
1211        goto bad_fork_free;
1212
1213    /*
1214     * If multiple threads are within copy_process(), then this check
1215     * triggers too late. This doesn't hurt, the check is only there
1216     * to stop root fork bombs.
1217     */
1218    retval = -EAGAIN;
1219    if (nr_threads >= max_threads)
1220        goto bad_fork_cleanup_count;
1221
1222    if (!try_module_get(task_thread_info(p)->exec_domain->module))
1223        goto bad_fork_cleanup_count;
1224
1225    p->did_exec = 0;
1226    delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1227    copy_flags(clone_flags, p);
1228    INIT_LIST_HEAD(&p->children);
1229    INIT_LIST_HEAD(&p->sibling);
1230    rcu_copy_process(p);
1231    p->vfork_done = NULL;
1232    spin_lock_init(&p->alloc_lock);
1233
1234    init_sigpending(&p->pending);
1235
1236    p->utime = p->stime = p->gtime = 0;
1237    p->utimescaled = p->stimescaled = 0;
1238#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1239    p->prev_cputime.utime = p->prev_cputime.stime = 0;
1240#endif
1241#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1242    seqlock_init(&p->vtime_seqlock);
1243    p->vtime_snap = 0;
1244    p->vtime_snap_whence = VTIME_SLEEPING;
1245#endif
1246
1247#if defined(SPLIT_RSS_COUNTING)
1248    memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1249#endif
1250
1251    p->default_timer_slack_ns = current->timer_slack_ns;
1252
1253    task_io_accounting_init(&p->ioac);
1254    acct_clear_integrals(p);
1255
1256    posix_cpu_timers_init(p);
1257
1258    do_posix_clock_monotonic_gettime(&p->start_time);
1259    p->real_start_time = p->start_time;
1260    monotonic_to_bootbased(&p->real_start_time);
1261    p->io_context = NULL;
1262    p->audit_context = NULL;
1263    if (clone_flags & CLONE_THREAD)
1264        threadgroup_change_begin(current);
1265    cgroup_fork(p);
1266#ifdef CONFIG_NUMA
1267    p->mempolicy = mpol_dup(p->mempolicy);
1268    if (IS_ERR(p->mempolicy)) {
1269        retval = PTR_ERR(p->mempolicy);
1270        p->mempolicy = NULL;
1271        goto bad_fork_cleanup_cgroup;
1272    }
1273    mpol_fix_fork_child_flag(p);
1274#endif
1275#ifdef CONFIG_CPUSETS
1276    p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1277    p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1278    seqcount_init(&p->mems_allowed_seq);
1279#endif
1280#ifdef CONFIG_TRACE_IRQFLAGS
1281    p->irq_events = 0;
1282    p->hardirqs_enabled = 0;
1283    p->hardirq_enable_ip = 0;
1284    p->hardirq_enable_event = 0;
1285    p->hardirq_disable_ip = _THIS_IP_;
1286    p->hardirq_disable_event = 0;
1287    p->softirqs_enabled = 1;
1288    p->softirq_enable_ip = _THIS_IP_;
1289    p->softirq_enable_event = 0;
1290    p->softirq_disable_ip = 0;
1291    p->softirq_disable_event = 0;
1292    p->hardirq_context = 0;
1293    p->softirq_context = 0;
1294#endif
1295#ifdef CONFIG_LOCKDEP
1296    p->lockdep_depth = 0; /* no locks held yet */
1297    p->curr_chain_key = 0;
1298    p->lockdep_recursion = 0;
1299#endif
1300
1301#ifdef CONFIG_DEBUG_MUTEXES
1302    p->blocked_on = NULL; /* not blocked yet */
1303#endif
1304#ifdef CONFIG_MEMCG
1305    p->memcg_batch.do_batch = 0;
1306    p->memcg_batch.memcg = NULL;
1307#endif
1308#ifdef CONFIG_BCACHE
1309    p->sequential_io = 0;
1310    p->sequential_io_avg = 0;
1311#endif
1312
1313    /* Perform scheduler related setup. Assign this task to a CPU. */
1314    sched_fork(clone_flags, p);
1315
1316    retval = perf_event_init_task(p);
1317    if (retval)
1318        goto bad_fork_cleanup_policy;
1319    retval = audit_alloc(p);
1320    if (retval)
1321        goto bad_fork_cleanup_policy;
1322    /* copy all the process information */
1323    retval = copy_semundo(clone_flags, p);
1324    if (retval)
1325        goto bad_fork_cleanup_audit;
1326    retval = copy_files(clone_flags, p);
1327    if (retval)
1328        goto bad_fork_cleanup_semundo;
1329    retval = copy_fs(clone_flags, p);
1330    if (retval)
1331        goto bad_fork_cleanup_files;
1332    retval = copy_sighand(clone_flags, p);
1333    if (retval)
1334        goto bad_fork_cleanup_fs;
1335    retval = copy_signal(clone_flags, p);
1336    if (retval)
1337        goto bad_fork_cleanup_sighand;
1338    retval = copy_mm(clone_flags, p);
1339    if (retval)
1340        goto bad_fork_cleanup_signal;
1341    retval = copy_namespaces(clone_flags, p);
1342    if (retval)
1343        goto bad_fork_cleanup_mm;
1344    retval = copy_io(clone_flags, p);
1345    if (retval)
1346        goto bad_fork_cleanup_namespaces;
1347    retval = copy_thread(clone_flags, stack_start, stack_size, p);
1348    if (retval)
1349        goto bad_fork_cleanup_io;
1350
1351    if (pid != &init_struct_pid) {
1352        retval = -ENOMEM;
1353        pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1354        if (!pid)
1355            goto bad_fork_cleanup_io;
1356    }
1357
1358    p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1359    /*
1360     * Clear TID on mm_release()?
1361     */
1362    p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1363#ifdef CONFIG_BLOCK
1364    p->plug = NULL;
1365#endif
1366#ifdef CONFIG_FUTEX
1367    p->robust_list = NULL;
1368#ifdef CONFIG_COMPAT
1369    p->compat_robust_list = NULL;
1370#endif
1371    INIT_LIST_HEAD(&p->pi_state_list);
1372    p->pi_state_cache = NULL;
1373#endif
1374    /*
1375     * sigaltstack should be cleared when sharing the same VM
1376     */
1377    if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1378        p->sas_ss_sp = p->sas_ss_size = 0;
1379
1380    /*
1381     * Syscall tracing and stepping should be turned off in the
1382     * child regardless of CLONE_PTRACE.
1383     */
1384    user_disable_single_step(p);
1385    clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1386#ifdef TIF_SYSCALL_EMU
1387    clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1388#endif
1389    clear_all_latency_tracing(p);
1390
1391    /* ok, now we should be set up.. */
1392    p->pid = pid_nr(pid);
1393    if (clone_flags & CLONE_THREAD) {
1394        p->exit_signal = -1;
1395        p->group_leader = current->group_leader;
1396        p->tgid = current->tgid;
1397    } else {
1398        if (clone_flags & CLONE_PARENT)
1399            p->exit_signal = current->group_leader->exit_signal;
1400        else
1401            p->exit_signal = (clone_flags & CSIGNAL);
1402        p->group_leader = p;
1403        p->tgid = p->pid;
1404    }
1405
1406    p->pdeath_signal = 0;
1407    p->exit_state = 0;
1408
1409    p->nr_dirtied = 0;
1410    p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1411    p->dirty_paused_when = 0;
1412
1413    INIT_LIST_HEAD(&p->thread_group);
1414    p->task_works = NULL;
1415
1416    /*
1417     * Make it visible to the rest of the system, but dont wake it up yet.
1418     * Need tasklist lock for parent etc handling!
1419     */
1420    write_lock_irq(&tasklist_lock);
1421
1422    /* CLONE_PARENT re-uses the old parent */
1423    if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1424        p->real_parent = current->real_parent;
1425        p->parent_exec_id = current->parent_exec_id;
1426    } else {
1427        p->real_parent = current;
1428        p->parent_exec_id = current->self_exec_id;
1429    }
1430
1431    spin_lock(&current->sighand->siglock);
1432
1433    /*
1434     * Process group and session signals need to be delivered to just the
1435     * parent before the fork or both the parent and the child after the
1436     * fork. Restart if a signal comes in before we add the new process to
1437     * it's process group.
1438     * A fatal signal pending means that current will exit, so the new
1439     * thread can't slip out of an OOM kill (or normal SIGKILL).
1440    */
1441    recalc_sigpending();
1442    if (signal_pending(current)) {
1443        spin_unlock(&current->sighand->siglock);
1444        write_unlock_irq(&tasklist_lock);
1445        retval = -ERESTARTNOINTR;
1446        goto bad_fork_free_pid;
1447    }
1448
1449    if (likely(p->pid)) {
1450        ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1451
1452        init_task_pid(p, PIDTYPE_PID, pid);
1453        if (thread_group_leader(p)) {
1454            init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1455            init_task_pid(p, PIDTYPE_SID, task_session(current));
1456
1457            if (is_child_reaper(pid)) {
1458                ns_of_pid(pid)->child_reaper = p;
1459                p->signal->flags |= SIGNAL_UNKILLABLE;
1460            }
1461
1462            p->signal->leader_pid = pid;
1463            p->signal->tty = tty_kref_get(current->signal->tty);
1464            list_add_tail(&p->sibling, &p->real_parent->children);
1465            list_add_tail_rcu(&p->tasks, &init_task.tasks);
1466            attach_pid(p, PIDTYPE_PGID);
1467            attach_pid(p, PIDTYPE_SID);
1468            __this_cpu_inc(process_counts);
1469        } else {
1470            current->signal->nr_threads++;
1471            atomic_inc(&current->signal->live);
1472            atomic_inc(&current->signal->sigcnt);
1473            list_add_tail_rcu(&p->thread_group,
1474                      &p->group_leader->thread_group);
1475        }
1476        attach_pid(p, PIDTYPE_PID);
1477        nr_threads++;
1478    }
1479
1480    total_forks++;
1481    spin_unlock(&current->sighand->siglock);
1482    write_unlock_irq(&tasklist_lock);
1483    proc_fork_connector(p);
1484    cgroup_post_fork(p);
1485    if (clone_flags & CLONE_THREAD)
1486        threadgroup_change_end(current);
1487    perf_event_fork(p);
1488
1489    trace_task_newtask(p, clone_flags);
1490    uprobe_copy_process(p, clone_flags);
1491
1492    return p;
1493
1494bad_fork_free_pid:
1495    if (pid != &init_struct_pid)
1496        free_pid(pid);
1497bad_fork_cleanup_io:
1498    if (p->io_context)
1499        exit_io_context(p);
1500bad_fork_cleanup_namespaces:
1501    exit_task_namespaces(p);
1502bad_fork_cleanup_mm:
1503    if (p->mm)
1504        mmput(p->mm);
1505bad_fork_cleanup_signal:
1506    if (!(clone_flags & CLONE_THREAD))
1507        free_signal_struct(p->signal);
1508bad_fork_cleanup_sighand:
1509    __cleanup_sighand(p->sighand);
1510bad_fork_cleanup_fs:
1511    exit_fs(p); /* blocking */
1512bad_fork_cleanup_files:
1513    exit_files(p); /* blocking */
1514bad_fork_cleanup_semundo:
1515    exit_sem(p);
1516bad_fork_cleanup_audit:
1517    audit_free(p);
1518bad_fork_cleanup_policy:
1519    perf_event_free_task(p);
1520#ifdef CONFIG_NUMA
1521    mpol_put(p->mempolicy);
1522bad_fork_cleanup_cgroup:
1523#endif
1524    if (clone_flags & CLONE_THREAD)
1525        threadgroup_change_end(current);
1526    cgroup_exit(p, 0);
1527    delayacct_tsk_free(p);
1528    module_put(task_thread_info(p)->exec_domain->module);
1529bad_fork_cleanup_count:
1530    atomic_dec(&p->cred->user->processes);
1531    exit_creds(p);
1532bad_fork_free:
1533    free_task(p);
1534fork_out:
1535    return ERR_PTR(retval);
1536}
1537
1538static inline void init_idle_pids(struct pid_link *links)
1539{
1540    enum pid_type type;
1541
1542    for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1543        INIT_HLIST_NODE(&links[type].node); /* not really needed */
1544        links[type].pid = &init_struct_pid;
1545    }
1546}
1547
1548struct task_struct *fork_idle(int cpu)
1549{
1550    struct task_struct *task;
1551    task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1552    if (!IS_ERR(task)) {
1553        init_idle_pids(task->pids);
1554        init_idle(task, cpu);
1555    }
1556
1557    return task;
1558}
1559
1560/*
1561 * Ok, this is the main fork-routine.
1562 *
1563 * It copies the process, and if successful kick-starts
1564 * it and waits for it to finish using the VM if required.
1565 */
1566long do_fork(unsigned long clone_flags,
1567          unsigned long stack_start,
1568          unsigned long stack_size,
1569          int __user *parent_tidptr,
1570          int __user *child_tidptr)
1571{
1572    struct task_struct *p;
1573    int trace = 0;
1574    long nr;
1575
1576    /*
1577     * Determine whether and which event to report to ptracer. When
1578     * called from kernel_thread or CLONE_UNTRACED is explicitly
1579     * requested, no event is reported; otherwise, report if the event
1580     * for the type of forking is enabled.
1581     */
1582    if (!(clone_flags & CLONE_UNTRACED)) {
1583        if (clone_flags & CLONE_VFORK)
1584            trace = PTRACE_EVENT_VFORK;
1585        else if ((clone_flags & CSIGNAL) != SIGCHLD)
1586            trace = PTRACE_EVENT_CLONE;
1587        else
1588            trace = PTRACE_EVENT_FORK;
1589
1590        if (likely(!ptrace_event_enabled(current, trace)))
1591            trace = 0;
1592    }
1593
1594    p = copy_process(clone_flags, stack_start, stack_size,
1595             child_tidptr, NULL, trace);
1596    /*
1597     * Do this prior waking up the new thread - the thread pointer
1598     * might get invalid after that point, if the thread exits quickly.
1599     */
1600    if (!IS_ERR(p)) {
1601        struct completion vfork;
1602
1603        trace_sched_process_fork(current, p);
1604
1605        nr = task_pid_vnr(p);
1606
1607        if (clone_flags & CLONE_PARENT_SETTID)
1608            put_user(nr, parent_tidptr);
1609
1610        if (clone_flags & CLONE_VFORK) {
1611            p->vfork_done = &vfork;
1612            init_completion(&vfork);
1613            get_task_struct(p);
1614        }
1615
1616        wake_up_new_task(p);
1617
1618        /* forking complete and child started to run, tell ptracer */
1619        if (unlikely(trace))
1620            ptrace_event(trace, nr);
1621
1622        if (clone_flags & CLONE_VFORK) {
1623            if (!wait_for_vfork_done(p, &vfork))
1624                ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1625        }
1626    } else {
1627        nr = PTR_ERR(p);
1628    }
1629    return nr;
1630}
1631
1632/*
1633 * Create a kernel thread.
1634 */
1635pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1636{
1637    return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1638        (unsigned long)arg, NULL, NULL);
1639}
1640
1641#ifdef __ARCH_WANT_SYS_FORK
1642SYSCALL_DEFINE0(fork)
1643{
1644#ifdef CONFIG_MMU
1645    return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1646#else
1647    /* can not support in nommu mode */
1648    return(-EINVAL);
1649#endif
1650}
1651#endif
1652
1653#ifdef __ARCH_WANT_SYS_VFORK
1654SYSCALL_DEFINE0(vfork)
1655{
1656    return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1657            0, NULL, NULL);
1658}
1659#endif
1660
1661#ifdef __ARCH_WANT_SYS_CLONE
1662#ifdef CONFIG_CLONE_BACKWARDS
1663SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1664         int __user *, parent_tidptr,
1665         int, tls_val,
1666         int __user *, child_tidptr)
1667#elif defined(CONFIG_CLONE_BACKWARDS2)
1668SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1669         int __user *, parent_tidptr,
1670         int __user *, child_tidptr,
1671         int, tls_val)
1672#elif defined(CONFIG_CLONE_BACKWARDS3)
1673SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1674        int, stack_size,
1675        int __user *, parent_tidptr,
1676        int __user *, child_tidptr,
1677        int, tls_val)
1678#else
1679SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1680         int __user *, parent_tidptr,
1681         int __user *, child_tidptr,
1682         int, tls_val)
1683#endif
1684{
1685    return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1686}
1687#endif
1688
1689#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1690#define ARCH_MIN_MMSTRUCT_ALIGN 0
1691#endif
1692
1693static void sighand_ctor(void *data)
1694{
1695    struct sighand_struct *sighand = data;
1696
1697    spin_lock_init(&sighand->siglock);
1698    init_waitqueue_head(&sighand->signalfd_wqh);
1699}
1700
1701void __init proc_caches_init(void)
1702{
1703    sighand_cachep = kmem_cache_create("sighand_cache",
1704            sizeof(struct sighand_struct), 0,
1705            SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1706            SLAB_NOTRACK, sighand_ctor);
1707    signal_cachep = kmem_cache_create("signal_cache",
1708            sizeof(struct signal_struct), 0,
1709            SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1710    files_cachep = kmem_cache_create("files_cache",
1711            sizeof(struct files_struct), 0,
1712            SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1713    fs_cachep = kmem_cache_create("fs_cache",
1714            sizeof(struct fs_struct), 0,
1715            SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1716    /*
1717     * FIXME! The "sizeof(struct mm_struct)" currently includes the
1718     * whole struct cpumask for the OFFSTACK case. We could change
1719     * this to *only* allocate as much of it as required by the
1720     * maximum number of CPU's we can ever have. The cpumask_allocation
1721     * is at the end of the structure, exactly for that reason.
1722     */
1723    mm_cachep = kmem_cache_create("mm_struct",
1724            sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1725            SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1726    vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1727    mmap_init();
1728    nsproxy_cache_init();
1729}
1730
1731/*
1732 * Check constraints on flags passed to the unshare system call.
1733 */
1734static int check_unshare_flags(unsigned long unshare_flags)
1735{
1736    if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1737                CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1738                CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1739                CLONE_NEWUSER|CLONE_NEWPID))
1740        return -EINVAL;
1741    /*
1742     * Not implemented, but pretend it works if there is nothing to
1743     * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1744     * needs to unshare vm.
1745     */
1746    if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1747        /* FIXME: get_task_mm() increments ->mm_users */
1748        if (atomic_read(&current->mm->mm_users) > 1)
1749            return -EINVAL;
1750    }
1751
1752    return 0;
1753}
1754
1755/*
1756 * Unshare the filesystem structure if it is being shared
1757 */
1758static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1759{
1760    struct fs_struct *fs = current->fs;
1761
1762    if (!(unshare_flags & CLONE_FS) || !fs)
1763        return 0;
1764
1765    /* don't need lock here; in the worst case we'll do useless copy */
1766    if (fs->users == 1)
1767        return 0;
1768
1769    *new_fsp = copy_fs_struct(fs);
1770    if (!*new_fsp)
1771        return -ENOMEM;
1772
1773    return 0;
1774}
1775
1776/*
1777 * Unshare file descriptor table if it is being shared
1778 */
1779static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1780{
1781    struct files_struct *fd = current->files;
1782    int error = 0;
1783
1784    if ((unshare_flags & CLONE_FILES) &&
1785        (fd && atomic_read(&fd->count) > 1)) {
1786        *new_fdp = dup_fd(fd, &error);
1787        if (!*new_fdp)
1788            return error;
1789    }
1790
1791    return 0;
1792}
1793
1794/*
1795 * unshare allows a process to 'unshare' part of the process
1796 * context which was originally shared using clone. copy_*
1797 * functions used by do_fork() cannot be used here directly
1798 * because they modify an inactive task_struct that is being
1799 * constructed. Here we are modifying the current, active,
1800 * task_struct.
1801 */
1802SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1803{
1804    struct fs_struct *fs, *new_fs = NULL;
1805    struct files_struct *fd, *new_fd = NULL;
1806    struct cred *new_cred = NULL;
1807    struct nsproxy *new_nsproxy = NULL;
1808    int do_sysvsem = 0;
1809    int err;
1810
1811    /*
1812     * If unsharing a user namespace must also unshare the thread.
1813     */
1814    if (unshare_flags & CLONE_NEWUSER)
1815        unshare_flags |= CLONE_THREAD | CLONE_FS;
1816    /*
1817     * If unsharing a thread from a thread group, must also unshare vm.
1818     */
1819    if (unshare_flags & CLONE_THREAD)
1820        unshare_flags |= CLONE_VM;
1821    /*
1822     * If unsharing vm, must also unshare signal handlers.
1823     */
1824    if (unshare_flags & CLONE_VM)
1825        unshare_flags |= CLONE_SIGHAND;
1826    /*
1827     * If unsharing namespace, must also unshare filesystem information.
1828     */
1829    if (unshare_flags & CLONE_NEWNS)
1830        unshare_flags |= CLONE_FS;
1831
1832    err = check_unshare_flags(unshare_flags);
1833    if (err)
1834        goto bad_unshare_out;
1835    /*
1836     * CLONE_NEWIPC must also detach from the undolist: after switching
1837     * to a new ipc namespace, the semaphore arrays from the old
1838     * namespace are unreachable.
1839     */
1840    if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1841        do_sysvsem = 1;
1842    err = unshare_fs(unshare_flags, &new_fs);
1843    if (err)
1844        goto bad_unshare_out;
1845    err = unshare_fd(unshare_flags, &new_fd);
1846    if (err)
1847        goto bad_unshare_cleanup_fs;
1848    err = unshare_userns(unshare_flags, &new_cred);
1849    if (err)
1850        goto bad_unshare_cleanup_fd;
1851    err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1852                     new_cred, new_fs);
1853    if (err)
1854        goto bad_unshare_cleanup_cred;
1855
1856    if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1857        if (do_sysvsem) {
1858            /*
1859             * CLONE_SYSVSEM is equivalent to sys_exit().
1860             */
1861            exit_sem(current);
1862        }
1863
1864        if (new_nsproxy)
1865            switch_task_namespaces(current, new_nsproxy);
1866
1867        task_lock(current);
1868
1869        if (new_fs) {
1870            fs = current->fs;
1871            spin_lock(&fs->lock);
1872            current->fs = new_fs;
1873            if (--fs->users)
1874                new_fs = NULL;
1875            else
1876                new_fs = fs;
1877            spin_unlock(&fs->lock);
1878        }
1879
1880        if (new_fd) {
1881            fd = current->files;
1882            current->files = new_fd;
1883            new_fd = fd;
1884        }
1885
1886        task_unlock(current);
1887
1888        if (new_cred) {
1889            /* Install the new user namespace */
1890            commit_creds(new_cred);
1891            new_cred = NULL;
1892        }
1893    }
1894
1895bad_unshare_cleanup_cred:
1896    if (new_cred)
1897        put_cred(new_cred);
1898bad_unshare_cleanup_fd:
1899    if (new_fd)
1900        put_files_struct(new_fd);
1901
1902bad_unshare_cleanup_fs:
1903    if (new_fs)
1904        free_fs_struct(new_fs);
1905
1906bad_unshare_out:
1907    return err;
1908}
1909
1910/*
1911 * Helper to unshare the files of the current task.
1912 * We don't want to expose copy_files internals to
1913 * the exec layer of the kernel.
1914 */
1915
1916int unshare_files(struct files_struct **displaced)
1917{
1918    struct task_struct *task = current;
1919    struct files_struct *copy = NULL;
1920    int error;
1921
1922    error = unshare_fd(CLONE_FILES, &copy);
1923    if (error || !copy) {
1924        *displaced = NULL;
1925        return error;
1926    }
1927    *displaced = task->files;
1928    task_lock(task);
1929    task->files = copy;
1930    task_unlock(task);
1931    return 0;
1932}
1933

Archive Download this file



interactive