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/sched/autogroup.h>
16#include <linux/sched/mm.h>
17#include <linux/sched/coredump.h>
18#include <linux/sched/user.h>
19#include <linux/sched/numa_balancing.h>
20#include <linux/sched/stat.h>
21#include <linux/sched/task.h>
22#include <linux/sched/task_stack.h>
23#include <linux/sched/cputime.h>
24#include <linux/rtmutex.h>
25#include <linux/init.h>
26#include <linux/unistd.h>
27#include <linux/module.h>
28#include <linux/vmalloc.h>
29#include <linux/completion.h>
30#include <linux/personality.h>
31#include <linux/mempolicy.h>
32#include <linux/sem.h>
33#include <linux/file.h>
34#include <linux/fdtable.h>
35#include <linux/iocontext.h>
36#include <linux/key.h>
37#include <linux/binfmts.h>
38#include <linux/mman.h>
39#include <linux/mmu_notifier.h>
40#include <linux/hmm.h>
41#include <linux/fs.h>
42#include <linux/mm.h>
43#include <linux/vmacache.h>
44#include <linux/nsproxy.h>
45#include <linux/capability.h>
46#include <linux/cpu.h>
47#include <linux/cgroup.h>
48#include <linux/security.h>
49#include <linux/hugetlb.h>
50#include <linux/seccomp.h>
51#include <linux/swap.h>
52#include <linux/syscalls.h>
53#include <linux/jiffies.h>
54#include <linux/futex.h>
55#include <linux/compat.h>
56#include <linux/kthread.h>
57#include <linux/task_io_accounting_ops.h>
58#include <linux/rcupdate.h>
59#include <linux/ptrace.h>
60#include <linux/mount.h>
61#include <linux/audit.h>
62#include <linux/memcontrol.h>
63#include <linux/ftrace.h>
64#include <linux/proc_fs.h>
65#include <linux/profile.h>
66#include <linux/rmap.h>
67#include <linux/ksm.h>
68#include <linux/acct.h>
69#include <linux/userfaultfd_k.h>
70#include <linux/tsacct_kern.h>
71#include <linux/cn_proc.h>
72#include <linux/freezer.h>
73#include <linux/delayacct.h>
74#include <linux/taskstats_kern.h>
75#include <linux/random.h>
76#include <linux/tty.h>
77#include <linux/blkdev.h>
78#include <linux/fs_struct.h>
79#include <linux/magic.h>
80#include <linux/perf_event.h>
81#include <linux/posix-timers.h>
82#include <linux/user-return-notifier.h>
83#include <linux/oom.h>
84#include <linux/khugepaged.h>
85#include <linux/signalfd.h>
86#include <linux/uprobes.h>
87#include <linux/aio.h>
88#include <linux/compiler.h>
89#include <linux/sysctl.h>
90#include <linux/kcov.h>
91#include <linux/livepatch.h>
92#include <linux/thread_info.h>
93#include <linux/stackleak.h>
94
95#include <asm/pgtable.h>
96#include <asm/pgalloc.h>
97#include <linux/uaccess.h>
98#include <asm/mmu_context.h>
99#include <asm/cacheflush.h>
100#include <asm/tlbflush.h>
101
102#include <trace/events/sched.h>
103
104#define CREATE_TRACE_POINTS
105#include <trace/events/task.h>
106
107/*
108 * Minimum number of threads to boot the kernel
109 */
110#define MIN_THREADS 20
111
112/*
113 * Maximum number of threads
114 */
115#define MAX_THREADS FUTEX_TID_MASK
116
117/*
118 * Protected counters by write_lock_irq(&tasklist_lock)
119 */
120unsigned long total_forks; /* Handle normal Linux uptimes. */
121int nr_threads; /* The idle threads do not count.. */
122
123int max_threads; /* tunable limit on nr_threads */
124
125DEFINE_PER_CPU(unsigned long, process_counts) = 0;
126
127__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
128
129#ifdef CONFIG_PROVE_RCU
130int lockdep_tasklist_lock_is_held(void)
131{
132 return lockdep_is_held(&tasklist_lock);
133}
134EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
135#endif /* #ifdef CONFIG_PROVE_RCU */
136
137int nr_processes(void)
138{
139 int cpu;
140 int total = 0;
141
142 for_each_possible_cpu(cpu)
143 total += per_cpu(process_counts, cpu);
144
145 return total;
146}
147
148void __weak arch_release_task_struct(struct task_struct *tsk)
149{
150}
151
152#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
153static struct kmem_cache *task_struct_cachep;
154
155static inline struct task_struct *alloc_task_struct_node(int node)
156{
157 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
158}
159
160static inline void free_task_struct(struct task_struct *tsk)
161{
162 kmem_cache_free(task_struct_cachep, tsk);
163}
164#endif
165
166#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
167
168/*
169 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
170 * kmemcache based allocator.
171 */
172# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
173
174#ifdef CONFIG_VMAP_STACK
175/*
176 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
177 * flush. Try to minimize the number of calls by caching stacks.
178 */
179#define NR_CACHED_STACKS 2
180static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
181
182static int free_vm_stack_cache(unsigned int cpu)
183{
184 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
185 int i;
186
187 for (i = 0; i < NR_CACHED_STACKS; i++) {
188 struct vm_struct *vm_stack = cached_vm_stacks[i];
189
190 if (!vm_stack)
191 continue;
192
193 vfree(vm_stack->addr);
194 cached_vm_stacks[i] = NULL;
195 }
196
197 return 0;
198}
199#endif
200
201static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
202{
203#ifdef CONFIG_VMAP_STACK
204 void *stack;
205 int i;
206
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 struct vm_struct *s;
209
210 s = this_cpu_xchg(cached_stacks[i], NULL);
211
212 if (!s)
213 continue;
214
215 /* Clear stale pointers from reused stack. */
216 memset(s->addr, 0, THREAD_SIZE);
217
218 tsk->stack_vm_area = s;
219 tsk->stack = s->addr;
220 return s->addr;
221 }
222
223 /*
224 * Allocated stacks are cached and later reused by new threads,
225 * so memcg accounting is performed manually on assigning/releasing
226 * stacks to tasks. Drop __GFP_ACCOUNT.
227 */
228 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
229 VMALLOC_START, VMALLOC_END,
230 THREADINFO_GFP & ~__GFP_ACCOUNT,
231 PAGE_KERNEL,
232 0, node, __builtin_return_address(0));
233
234 /*
235 * We can't call find_vm_area() in interrupt context, and
236 * free_thread_stack() can be called in interrupt context,
237 * so cache the vm_struct.
238 */
239 if (stack) {
240 tsk->stack_vm_area = find_vm_area(stack);
241 tsk->stack = stack;
242 }
243 return stack;
244#else
245 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
246 THREAD_SIZE_ORDER);
247
248 return page ? page_address(page) : NULL;
249#endif
250}
251
252static inline void free_thread_stack(struct task_struct *tsk)
253{
254#ifdef CONFIG_VMAP_STACK
255 struct vm_struct *vm = task_stack_vm_area(tsk);
256
257 if (vm) {
258 int i;
259
260 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
261 mod_memcg_page_state(vm->pages[i],
262 MEMCG_KERNEL_STACK_KB,
263 -(int)(PAGE_SIZE / 1024));
264
265 memcg_kmem_uncharge(vm->pages[i], 0);
266 }
267
268 for (i = 0; i < NR_CACHED_STACKS; i++) {
269 if (this_cpu_cmpxchg(cached_stacks[i],
270 NULL, tsk->stack_vm_area) != NULL)
271 continue;
272
273 return;
274 }
275
276 vfree_atomic(tsk->stack);
277 return;
278 }
279#endif
280
281 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
282}
283# else
284static struct kmem_cache *thread_stack_cache;
285
286static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
287 int node)
288{
289 unsigned long *stack;
290 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
291 tsk->stack = stack;
292 return stack;
293}
294
295static void free_thread_stack(struct task_struct *tsk)
296{
297 kmem_cache_free(thread_stack_cache, tsk->stack);
298}
299
300void thread_stack_cache_init(void)
301{
302 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
303 THREAD_SIZE, THREAD_SIZE, 0, 0,
304 THREAD_SIZE, NULL);
305 BUG_ON(thread_stack_cache == NULL);
306}
307# endif
308#endif
309
310/* SLAB cache for signal_struct structures (tsk->signal) */
311static struct kmem_cache *signal_cachep;
312
313/* SLAB cache for sighand_struct structures (tsk->sighand) */
314struct kmem_cache *sighand_cachep;
315
316/* SLAB cache for files_struct structures (tsk->files) */
317struct kmem_cache *files_cachep;
318
319/* SLAB cache for fs_struct structures (tsk->fs) */
320struct kmem_cache *fs_cachep;
321
322/* SLAB cache for vm_area_struct structures */
323static struct kmem_cache *vm_area_cachep;
324
325/* SLAB cache for mm_struct structures (tsk->mm) */
326static struct kmem_cache *mm_cachep;
327
328struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
329{
330 struct vm_area_struct *vma;
331
332 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
333 if (vma)
334 vma_init(vma, mm);
335 return vma;
336}
337
338struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
339{
340 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
341
342 if (new) {
343 *new = *orig;
344 INIT_LIST_HEAD(&new->anon_vma_chain);
345 }
346 return new;
347}
348
349void vm_area_free(struct vm_area_struct *vma)
350{
351 kmem_cache_free(vm_area_cachep, vma);
352}
353
354static void account_kernel_stack(struct task_struct *tsk, int account)
355{
356 void *stack = task_stack_page(tsk);
357 struct vm_struct *vm = task_stack_vm_area(tsk);
358
359 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
360
361 if (vm) {
362 int i;
363
364 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
365
366 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
367 mod_zone_page_state(page_zone(vm->pages[i]),
368 NR_KERNEL_STACK_KB,
369 PAGE_SIZE / 1024 * account);
370 }
371 } else {
372 /*
373 * All stack pages are in the same zone and belong to the
374 * same memcg.
375 */
376 struct page *first_page = virt_to_page(stack);
377
378 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
379 THREAD_SIZE / 1024 * account);
380
381 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
382 account * (THREAD_SIZE / 1024));
383 }
384}
385
386static int memcg_charge_kernel_stack(struct task_struct *tsk)
387{
388#ifdef CONFIG_VMAP_STACK
389 struct vm_struct *vm = task_stack_vm_area(tsk);
390 int ret;
391
392 if (vm) {
393 int i;
394
395 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
396 /*
397 * If memcg_kmem_charge() fails, page->mem_cgroup
398 * pointer is NULL, and both memcg_kmem_uncharge()
399 * and mod_memcg_page_state() in free_thread_stack()
400 * will ignore this page. So it's safe.
401 */
402 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
403 if (ret)
404 return ret;
405
406 mod_memcg_page_state(vm->pages[i],
407 MEMCG_KERNEL_STACK_KB,
408 PAGE_SIZE / 1024);
409 }
410 }
411#endif
412 return 0;
413}
414
415static void release_task_stack(struct task_struct *tsk)
416{
417 if (WARN_ON(tsk->state != TASK_DEAD))
418 return; /* Better to leak the stack than to free prematurely */
419
420 account_kernel_stack(tsk, -1);
421 free_thread_stack(tsk);
422 tsk->stack = NULL;
423#ifdef CONFIG_VMAP_STACK
424 tsk->stack_vm_area = NULL;
425#endif
426}
427
428#ifdef CONFIG_THREAD_INFO_IN_TASK
429void put_task_stack(struct task_struct *tsk)
430{
431 if (refcount_dec_and_test(&tsk->stack_refcount))
432 release_task_stack(tsk);
433}
434#endif
435
436void free_task(struct task_struct *tsk)
437{
438#ifndef CONFIG_THREAD_INFO_IN_TASK
439 /*
440 * The task is finally done with both the stack and thread_info,
441 * so free both.
442 */
443 release_task_stack(tsk);
444#else
445 /*
446 * If the task had a separate stack allocation, it should be gone
447 * by now.
448 */
449 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
450#endif
451 rt_mutex_debug_task_free(tsk);
452 ftrace_graph_exit_task(tsk);
453 put_seccomp_filter(tsk);
454 arch_release_task_struct(tsk);
455 if (tsk->flags & PF_KTHREAD)
456 free_kthread_struct(tsk);
457 free_task_struct(tsk);
458}
459EXPORT_SYMBOL(free_task);
460
461#ifdef CONFIG_MMU
462static __latent_entropy int dup_mmap(struct mm_struct *mm,
463 struct mm_struct *oldmm)
464{
465 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
466 struct rb_node **rb_link, *rb_parent;
467 int retval;
468 unsigned long charge;
469 LIST_HEAD(uf);
470
471 uprobe_start_dup_mmap();
472 if (down_write_killable(&oldmm->mmap_sem)) {
473 retval = -EINTR;
474 goto fail_uprobe_end;
475 }
476 flush_cache_dup_mm(oldmm);
477 uprobe_dup_mmap(oldmm, mm);
478 /*
479 * Not linked in yet - no deadlock potential:
480 */
481 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
482
483 /* No ordering required: file already has been exposed. */
484 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
485
486 mm->total_vm = oldmm->total_vm;
487 mm->data_vm = oldmm->data_vm;
488 mm->exec_vm = oldmm->exec_vm;
489 mm->stack_vm = oldmm->stack_vm;
490
491 rb_link = &mm->mm_rb.rb_node;
492 rb_parent = NULL;
493 pprev = &mm->mmap;
494 retval = ksm_fork(mm, oldmm);
495 if (retval)
496 goto out;
497 retval = khugepaged_fork(mm, oldmm);
498 if (retval)
499 goto out;
500
501 prev = NULL;
502 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
503 struct file *file;
504
505 if (mpnt->vm_flags & VM_DONTCOPY) {
506 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
507 continue;
508 }
509 charge = 0;
510 /*
511 * Don't duplicate many vmas if we've been oom-killed (for
512 * example)
513 */
514 if (fatal_signal_pending(current)) {
515 retval = -EINTR;
516 goto out;
517 }
518 if (mpnt->vm_flags & VM_ACCOUNT) {
519 unsigned long len = vma_pages(mpnt);
520
521 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
522 goto fail_nomem;
523 charge = len;
524 }
525 tmp = vm_area_dup(mpnt);
526 if (!tmp)
527 goto fail_nomem;
528 retval = vma_dup_policy(mpnt, tmp);
529 if (retval)
530 goto fail_nomem_policy;
531 tmp->vm_mm = mm;
532 retval = dup_userfaultfd(tmp, &uf);
533 if (retval)
534 goto fail_nomem_anon_vma_fork;
535 if (tmp->vm_flags & VM_WIPEONFORK) {
536 /* VM_WIPEONFORK gets a clean slate in the child. */
537 tmp->anon_vma = NULL;
538 if (anon_vma_prepare(tmp))
539 goto fail_nomem_anon_vma_fork;
540 } else if (anon_vma_fork(tmp, mpnt))
541 goto fail_nomem_anon_vma_fork;
542 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
543 tmp->vm_next = tmp->vm_prev = NULL;
544 file = tmp->vm_file;
545 if (file) {
546 struct inode *inode = file_inode(file);
547 struct address_space *mapping = file->f_mapping;
548
549 get_file(file);
550 if (tmp->vm_flags & VM_DENYWRITE)
551 atomic_dec(&inode->i_writecount);
552 i_mmap_lock_write(mapping);
553 if (tmp->vm_flags & VM_SHARED)
554 atomic_inc(&mapping->i_mmap_writable);
555 flush_dcache_mmap_lock(mapping);
556 /* insert tmp into the share list, just after mpnt */
557 vma_interval_tree_insert_after(tmp, mpnt,
558 &mapping->i_mmap);
559 flush_dcache_mmap_unlock(mapping);
560 i_mmap_unlock_write(mapping);
561 }
562
563 /*
564 * Clear hugetlb-related page reserves for children. This only
565 * affects MAP_PRIVATE mappings. Faults generated by the child
566 * are not guaranteed to succeed, even if read-only
567 */
568 if (is_vm_hugetlb_page(tmp))
569 reset_vma_resv_huge_pages(tmp);
570
571 /*
572 * Link in the new vma and copy the page table entries.
573 */
574 *pprev = tmp;
575 pprev = &tmp->vm_next;
576 tmp->vm_prev = prev;
577 prev = tmp;
578
579 __vma_link_rb(mm, tmp, rb_link, rb_parent);
580 rb_link = &tmp->vm_rb.rb_right;
581 rb_parent = &tmp->vm_rb;
582
583 mm->map_count++;
584 if (!(tmp->vm_flags & VM_WIPEONFORK))
585 retval = copy_page_range(mm, oldmm, mpnt);
586
587 if (tmp->vm_ops && tmp->vm_ops->open)
588 tmp->vm_ops->open(tmp);
589
590 if (retval)
591 goto out;
592 }
593 /* a new mm has just been created */
594 retval = arch_dup_mmap(oldmm, mm);
595out:
596 up_write(&mm->mmap_sem);
597 flush_tlb_mm(oldmm);
598 up_write(&oldmm->mmap_sem);
599 dup_userfaultfd_complete(&uf);
600fail_uprobe_end:
601 uprobe_end_dup_mmap();
602 return retval;
603fail_nomem_anon_vma_fork:
604 mpol_put(vma_policy(tmp));
605fail_nomem_policy:
606 vm_area_free(tmp);
607fail_nomem:
608 retval = -ENOMEM;
609 vm_unacct_memory(charge);
610 goto out;
611}
612
613static inline int mm_alloc_pgd(struct mm_struct *mm)
614{
615 mm->pgd = pgd_alloc(mm);
616 if (unlikely(!mm->pgd))
617 return -ENOMEM;
618 return 0;
619}
620
621static inline void mm_free_pgd(struct mm_struct *mm)
622{
623 pgd_free(mm, mm->pgd);
624}
625#else
626static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
627{
628 down_write(&oldmm->mmap_sem);
629 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
630 up_write(&oldmm->mmap_sem);
631 return 0;
632}
633#define mm_alloc_pgd(mm) (0)
634#define mm_free_pgd(mm)
635#endif /* CONFIG_MMU */
636
637static void check_mm(struct mm_struct *mm)
638{
639 int i;
640
641 for (i = 0; i < NR_MM_COUNTERS; i++) {
642 long x = atomic_long_read(&mm->rss_stat.count[i]);
643
644 if (unlikely(x))
645 printk(KERN_ALERT "BUG: Bad rss-counter state "
646 "mm:%p idx:%d val:%ld\n", mm, i, x);
647 }
648
649 if (mm_pgtables_bytes(mm))
650 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
651 mm_pgtables_bytes(mm));
652
653#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
654 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
655#endif
656}
657
658#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
659#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
660
661/*
662 * Called when the last reference to the mm
663 * is dropped: either by a lazy thread or by
664 * mmput. Free the page directory and the mm.
665 */
666void __mmdrop(struct mm_struct *mm)
667{
668 BUG_ON(mm == &init_mm);
669 WARN_ON_ONCE(mm == current->mm);
670 WARN_ON_ONCE(mm == current->active_mm);
671 mm_free_pgd(mm);
672 destroy_context(mm);
673 hmm_mm_destroy(mm);
674 mmu_notifier_mm_destroy(mm);
675 check_mm(mm);
676 put_user_ns(mm->user_ns);
677 free_mm(mm);
678}
679EXPORT_SYMBOL_GPL(__mmdrop);
680
681static void mmdrop_async_fn(struct work_struct *work)
682{
683 struct mm_struct *mm;
684
685 mm = container_of(work, struct mm_struct, async_put_work);
686 __mmdrop(mm);
687}
688
689static void mmdrop_async(struct mm_struct *mm)
690{
691 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
692 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
693 schedule_work(&mm->async_put_work);
694 }
695}
696
697static inline void free_signal_struct(struct signal_struct *sig)
698{
699 taskstats_tgid_free(sig);
700 sched_autogroup_exit(sig);
701 /*
702 * __mmdrop is not safe to call from softirq context on x86 due to
703 * pgd_dtor so postpone it to the async context
704 */
705 if (sig->oom_mm)
706 mmdrop_async(sig->oom_mm);
707 kmem_cache_free(signal_cachep, sig);
708}
709
710static inline void put_signal_struct(struct signal_struct *sig)
711{
712 if (refcount_dec_and_test(&sig->sigcnt))
713 free_signal_struct(sig);
714}
715
716void __put_task_struct(struct task_struct *tsk)
717{
718 WARN_ON(!tsk->exit_state);
719 WARN_ON(refcount_read(&tsk->usage));
720 WARN_ON(tsk == current);
721
722 cgroup_free(tsk);
723 task_numa_free(tsk);
724 security_task_free(tsk);
725 exit_creds(tsk);
726 delayacct_tsk_free(tsk);
727 put_signal_struct(tsk->signal);
728
729 if (!profile_handoff_task(tsk))
730 free_task(tsk);
731}
732EXPORT_SYMBOL_GPL(__put_task_struct);
733
734void __init __weak arch_task_cache_init(void) { }
735
736/*
737 * set_max_threads
738 */
739static void set_max_threads(unsigned int max_threads_suggested)
740{
741 u64 threads;
742 unsigned long nr_pages = totalram_pages();
743
744 /*
745 * The number of threads shall be limited such that the thread
746 * structures may only consume a small part of the available memory.
747 */
748 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
749 threads = MAX_THREADS;
750 else
751 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
752 (u64) THREAD_SIZE * 8UL);
753
754 if (threads > max_threads_suggested)
755 threads = max_threads_suggested;
756
757 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
758}
759
760#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
761/* Initialized by the architecture: */
762int arch_task_struct_size __read_mostly;
763#endif
764
765static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
766{
767 /* Fetch thread_struct whitelist for the architecture. */
768 arch_thread_struct_whitelist(offset, size);
769
770 /*
771 * Handle zero-sized whitelist or empty thread_struct, otherwise
772 * adjust offset to position of thread_struct in task_struct.
773 */
774 if (unlikely(*size == 0))
775 *offset = 0;
776 else
777 *offset += offsetof(struct task_struct, thread);
778}
779
780void __init fork_init(void)
781{
782 int i;
783#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
784#ifndef ARCH_MIN_TASKALIGN
785#define ARCH_MIN_TASKALIGN 0
786#endif
787 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
788 unsigned long useroffset, usersize;
789
790 /* create a slab on which task_structs can be allocated */
791 task_struct_whitelist(&useroffset, &usersize);
792 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
793 arch_task_struct_size, align,
794 SLAB_PANIC|SLAB_ACCOUNT,
795 useroffset, usersize, NULL);
796#endif
797
798 /* do the arch specific task caches init */
799 arch_task_cache_init();
800
801 set_max_threads(MAX_THREADS);
802
803 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
804 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
805 init_task.signal->rlim[RLIMIT_SIGPENDING] =
806 init_task.signal->rlim[RLIMIT_NPROC];
807
808 for (i = 0; i < UCOUNT_COUNTS; i++) {
809 init_user_ns.ucount_max[i] = max_threads/2;
810 }
811
812#ifdef CONFIG_VMAP_STACK
813 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
814 NULL, free_vm_stack_cache);
815#endif
816
817 lockdep_init_task(&init_task);
818}
819
820int __weak arch_dup_task_struct(struct task_struct *dst,
821 struct task_struct *src)
822{
823 *dst = *src;
824 return 0;
825}
826
827void set_task_stack_end_magic(struct task_struct *tsk)
828{
829 unsigned long *stackend;
830
831 stackend = end_of_stack(tsk);
832 *stackend = STACK_END_MAGIC; /* for overflow detection */
833}
834
835static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
836{
837 struct task_struct *tsk;
838 unsigned long *stack;
839 struct vm_struct *stack_vm_area __maybe_unused;
840 int err;
841
842 if (node == NUMA_NO_NODE)
843 node = tsk_fork_get_node(orig);
844 tsk = alloc_task_struct_node(node);
845 if (!tsk)
846 return NULL;
847
848 stack = alloc_thread_stack_node(tsk, node);
849 if (!stack)
850 goto free_tsk;
851
852 if (memcg_charge_kernel_stack(tsk))
853 goto free_stack;
854
855 stack_vm_area = task_stack_vm_area(tsk);
856
857 err = arch_dup_task_struct(tsk, orig);
858
859 /*
860 * arch_dup_task_struct() clobbers the stack-related fields. Make
861 * sure they're properly initialized before using any stack-related
862 * functions again.
863 */
864 tsk->stack = stack;
865#ifdef CONFIG_VMAP_STACK
866 tsk->stack_vm_area = stack_vm_area;
867#endif
868#ifdef CONFIG_THREAD_INFO_IN_TASK
869 refcount_set(&tsk->stack_refcount, 1);
870#endif
871
872 if (err)
873 goto free_stack;
874
875#ifdef CONFIG_SECCOMP
876 /*
877 * We must handle setting up seccomp filters once we're under
878 * the sighand lock in case orig has changed between now and
879 * then. Until then, filter must be NULL to avoid messing up
880 * the usage counts on the error path calling free_task.
881 */
882 tsk->seccomp.filter = NULL;
883#endif
884
885 setup_thread_stack(tsk, orig);
886 clear_user_return_notifier(tsk);
887 clear_tsk_need_resched(tsk);
888 set_task_stack_end_magic(tsk);
889
890#ifdef CONFIG_STACKPROTECTOR
891 tsk->stack_canary = get_random_canary();
892#endif
893
894 /*
895 * One for us, one for whoever does the "release_task()" (usually
896 * parent)
897 */
898 refcount_set(&tsk->usage, 2);
899#ifdef CONFIG_BLK_DEV_IO_TRACE
900 tsk->btrace_seq = 0;
901#endif
902 tsk->splice_pipe = NULL;
903 tsk->task_frag.page = NULL;
904 tsk->wake_q.next = NULL;
905
906 account_kernel_stack(tsk, 1);
907
908 kcov_task_init(tsk);
909
910#ifdef CONFIG_FAULT_INJECTION
911 tsk->fail_nth = 0;
912#endif
913
914#ifdef CONFIG_BLK_CGROUP
915 tsk->throttle_queue = NULL;
916 tsk->use_memdelay = 0;
917#endif
918
919#ifdef CONFIG_MEMCG
920 tsk->active_memcg = NULL;
921#endif
922 return tsk;
923
924free_stack:
925 free_thread_stack(tsk);
926free_tsk:
927 free_task_struct(tsk);
928 return NULL;
929}
930
931__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
932
933static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
934
935static int __init coredump_filter_setup(char *s)
936{
937 default_dump_filter =
938 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
939 MMF_DUMP_FILTER_MASK;
940 return 1;
941}
942
943__setup("coredump_filter=", coredump_filter_setup);
944
945#include <linux/init_task.h>
946
947static void mm_init_aio(struct mm_struct *mm)
948{
949#ifdef CONFIG_AIO
950 spin_lock_init(&mm->ioctx_lock);
951 mm->ioctx_table = NULL;
952#endif
953}
954
955static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
956{
957#ifdef CONFIG_MEMCG
958 mm->owner = p;
959#endif
960}
961
962static void mm_init_uprobes_state(struct mm_struct *mm)
963{
964#ifdef CONFIG_UPROBES
965 mm->uprobes_state.xol_area = NULL;
966#endif
967}
968
969static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
970 struct user_namespace *user_ns)
971{
972 mm->mmap = NULL;
973 mm->mm_rb = RB_ROOT;
974 mm->vmacache_seqnum = 0;
975 atomic_set(&mm->mm_users, 1);
976 atomic_set(&mm->mm_count, 1);
977 init_rwsem(&mm->mmap_sem);
978 INIT_LIST_HEAD(&mm->mmlist);
979 mm->core_state = NULL;
980 mm_pgtables_bytes_init(mm);
981 mm->map_count = 0;
982 mm->locked_vm = 0;
983 atomic64_set(&mm->pinned_vm, 0);
984 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
985 spin_lock_init(&mm->page_table_lock);
986 spin_lock_init(&mm->arg_lock);
987 mm_init_cpumask(mm);
988 mm_init_aio(mm);
989 mm_init_owner(mm, p);
990 RCU_INIT_POINTER(mm->exe_file, NULL);
991 mmu_notifier_mm_init(mm);
992 hmm_mm_init(mm);
993 init_tlb_flush_pending(mm);
994#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
995 mm->pmd_huge_pte = NULL;
996#endif
997 mm_init_uprobes_state(mm);
998
999 if (current->mm) {
1000 mm->flags = current->mm->flags & MMF_INIT_MASK;
1001 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1002 } else {
1003 mm->flags = default_dump_filter;
1004 mm->def_flags = 0;
1005 }
1006
1007 if (mm_alloc_pgd(mm))
1008 goto fail_nopgd;
1009
1010 if (init_new_context(p, mm))
1011 goto fail_nocontext;
1012
1013 mm->user_ns = get_user_ns(user_ns);
1014 return mm;
1015
1016fail_nocontext:
1017 mm_free_pgd(mm);
1018fail_nopgd:
1019 free_mm(mm);
1020 return NULL;
1021}
1022
1023/*
1024 * Allocate and initialize an mm_struct.
1025 */
1026struct mm_struct *mm_alloc(void)
1027{
1028 struct mm_struct *mm;
1029
1030 mm = allocate_mm();
1031 if (!mm)
1032 return NULL;
1033
1034 memset(mm, 0, sizeof(*mm));
1035 return mm_init(mm, current, current_user_ns());
1036}
1037
1038static inline void __mmput(struct mm_struct *mm)
1039{
1040 VM_BUG_ON(atomic_read(&mm->mm_users));
1041
1042 uprobe_clear_state(mm);
1043 exit_aio(mm);
1044 ksm_exit(mm);
1045 khugepaged_exit(mm); /* must run before exit_mmap */
1046 exit_mmap(mm);
1047 mm_put_huge_zero_page(mm);
1048 set_mm_exe_file(mm, NULL);
1049 if (!list_empty(&mm->mmlist)) {
1050 spin_lock(&mmlist_lock);
1051 list_del(&mm->mmlist);
1052 spin_unlock(&mmlist_lock);
1053 }
1054 if (mm->binfmt)
1055 module_put(mm->binfmt->module);
1056 mmdrop(mm);
1057}
1058
1059/*
1060 * Decrement the use count and release all resources for an mm.
1061 */
1062void mmput(struct mm_struct *mm)
1063{
1064 might_sleep();
1065
1066 if (atomic_dec_and_test(&mm->mm_users))
1067 __mmput(mm);
1068}
1069EXPORT_SYMBOL_GPL(mmput);
1070
1071#ifdef CONFIG_MMU
1072static void mmput_async_fn(struct work_struct *work)
1073{
1074 struct mm_struct *mm = container_of(work, struct mm_struct,
1075 async_put_work);
1076
1077 __mmput(mm);
1078}
1079
1080void mmput_async(struct mm_struct *mm)
1081{
1082 if (atomic_dec_and_test(&mm->mm_users)) {
1083 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1084 schedule_work(&mm->async_put_work);
1085 }
1086}
1087#endif
1088
1089/**
1090 * set_mm_exe_file - change a reference to the mm's executable file
1091 *
1092 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1093 *
1094 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1095 * invocations: in mmput() nobody alive left, in execve task is single
1096 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1097 * mm->exe_file, but does so without using set_mm_exe_file() in order
1098 * to do avoid the need for any locks.
1099 */
1100void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1101{
1102 struct file *old_exe_file;
1103
1104 /*
1105 * It is safe to dereference the exe_file without RCU as
1106 * this function is only called if nobody else can access
1107 * this mm -- see comment above for justification.
1108 */
1109 old_exe_file = rcu_dereference_raw(mm->exe_file);
1110
1111 if (new_exe_file)
1112 get_file(new_exe_file);
1113 rcu_assign_pointer(mm->exe_file, new_exe_file);
1114 if (old_exe_file)
1115 fput(old_exe_file);
1116}
1117
1118/**
1119 * get_mm_exe_file - acquire a reference to the mm's executable file
1120 *
1121 * Returns %NULL if mm has no associated executable file.
1122 * User must release file via fput().
1123 */
1124struct file *get_mm_exe_file(struct mm_struct *mm)
1125{
1126 struct file *exe_file;
1127
1128 rcu_read_lock();
1129 exe_file = rcu_dereference(mm->exe_file);
1130 if (exe_file && !get_file_rcu(exe_file))
1131 exe_file = NULL;
1132 rcu_read_unlock();
1133 return exe_file;
1134}
1135EXPORT_SYMBOL(get_mm_exe_file);
1136
1137/**
1138 * get_task_exe_file - acquire a reference to the task's executable file
1139 *
1140 * Returns %NULL if task's mm (if any) has no associated executable file or
1141 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1142 * User must release file via fput().
1143 */
1144struct file *get_task_exe_file(struct task_struct *task)
1145{
1146 struct file *exe_file = NULL;
1147 struct mm_struct *mm;
1148
1149 task_lock(task);
1150 mm = task->mm;
1151 if (mm) {
1152 if (!(task->flags & PF_KTHREAD))
1153 exe_file = get_mm_exe_file(mm);
1154 }
1155 task_unlock(task);
1156 return exe_file;
1157}
1158EXPORT_SYMBOL(get_task_exe_file);
1159
1160/**
1161 * get_task_mm - acquire a reference to the task's mm
1162 *
1163 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1164 * this kernel workthread has transiently adopted a user mm with use_mm,
1165 * to do its AIO) is not set and if so returns a reference to it, after
1166 * bumping up the use count. User must release the mm via mmput()
1167 * after use. Typically used by /proc and ptrace.
1168 */
1169struct mm_struct *get_task_mm(struct task_struct *task)
1170{
1171 struct mm_struct *mm;
1172
1173 task_lock(task);
1174 mm = task->mm;
1175 if (mm) {
1176 if (task->flags & PF_KTHREAD)
1177 mm = NULL;
1178 else
1179 mmget(mm);
1180 }
1181 task_unlock(task);
1182 return mm;
1183}
1184EXPORT_SYMBOL_GPL(get_task_mm);
1185
1186struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1187{
1188 struct mm_struct *mm;
1189 int err;
1190
1191 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1192 if (err)
1193 return ERR_PTR(err);
1194
1195 mm = get_task_mm(task);
1196 if (mm && mm != current->mm &&
1197 !ptrace_may_access(task, mode)) {
1198 mmput(mm);
1199 mm = ERR_PTR(-EACCES);
1200 }
1201 mutex_unlock(&task->signal->cred_guard_mutex);
1202
1203 return mm;
1204}
1205
1206static void complete_vfork_done(struct task_struct *tsk)
1207{
1208 struct completion *vfork;
1209
1210 task_lock(tsk);
1211 vfork = tsk->vfork_done;
1212 if (likely(vfork)) {
1213 tsk->vfork_done = NULL;
1214 complete(vfork);
1215 }
1216 task_unlock(tsk);
1217}
1218
1219static int wait_for_vfork_done(struct task_struct *child,
1220 struct completion *vfork)
1221{
1222 int killed;
1223
1224 freezer_do_not_count();
1225 killed = wait_for_completion_killable(vfork);
1226 freezer_count();
1227
1228 if (killed) {
1229 task_lock(child);
1230 child->vfork_done = NULL;
1231 task_unlock(child);
1232 }
1233
1234 put_task_struct(child);
1235 return killed;
1236}
1237
1238/* Please note the differences between mmput and mm_release.
1239 * mmput is called whenever we stop holding onto a mm_struct,
1240 * error success whatever.
1241 *
1242 * mm_release is called after a mm_struct has been removed
1243 * from the current process.
1244 *
1245 * This difference is important for error handling, when we
1246 * only half set up a mm_struct for a new process and need to restore
1247 * the old one. Because we mmput the new mm_struct before
1248 * restoring the old one. . .
1249 * Eric Biederman 10 January 1998
1250 */
1251void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1252{
1253 /* Get rid of any futexes when releasing the mm */
1254#ifdef CONFIG_FUTEX
1255 if (unlikely(tsk->robust_list)) {
1256 exit_robust_list(tsk);
1257 tsk->robust_list = NULL;
1258 }
1259#ifdef CONFIG_COMPAT
1260 if (unlikely(tsk->compat_robust_list)) {
1261 compat_exit_robust_list(tsk);
1262 tsk->compat_robust_list = NULL;
1263 }
1264#endif
1265 if (unlikely(!list_empty(&tsk->pi_state_list)))
1266 exit_pi_state_list(tsk);
1267#endif
1268
1269 uprobe_free_utask(tsk);
1270
1271 /* Get rid of any cached register state */
1272 deactivate_mm(tsk, mm);
1273
1274 /*
1275 * Signal userspace if we're not exiting with a core dump
1276 * because we want to leave the value intact for debugging
1277 * purposes.
1278 */
1279 if (tsk->clear_child_tid) {
1280 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1281 atomic_read(&mm->mm_users) > 1) {
1282 /*
1283 * We don't check the error code - if userspace has
1284 * not set up a proper pointer then tough luck.
1285 */
1286 put_user(0, tsk->clear_child_tid);
1287 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1288 1, NULL, NULL, 0, 0);
1289 }
1290 tsk->clear_child_tid = NULL;
1291 }
1292
1293 /*
1294 * All done, finally we can wake up parent and return this mm to him.
1295 * Also kthread_stop() uses this completion for synchronization.
1296 */
1297 if (tsk->vfork_done)
1298 complete_vfork_done(tsk);
1299}
1300
1301/*
1302 * Allocate a new mm structure and copy contents from the
1303 * mm structure of the passed in task structure.
1304 */
1305static struct mm_struct *dup_mm(struct task_struct *tsk)
1306{
1307 struct mm_struct *mm, *oldmm = current->mm;
1308 int err;
1309
1310 mm = allocate_mm();
1311 if (!mm)
1312 goto fail_nomem;
1313
1314 memcpy(mm, oldmm, sizeof(*mm));
1315
1316 if (!mm_init(mm, tsk, mm->user_ns))
1317 goto fail_nomem;
1318
1319 err = dup_mmap(mm, oldmm);
1320 if (err)
1321 goto free_pt;
1322
1323 mm->hiwater_rss = get_mm_rss(mm);
1324 mm->hiwater_vm = mm->total_vm;
1325
1326 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1327 goto free_pt;
1328
1329 return mm;
1330
1331free_pt:
1332 /* don't put binfmt in mmput, we haven't got module yet */
1333 mm->binfmt = NULL;
1334 mmput(mm);
1335
1336fail_nomem:
1337 return NULL;
1338}
1339
1340static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1341{
1342 struct mm_struct *mm, *oldmm;
1343 int retval;
1344
1345 tsk->min_flt = tsk->maj_flt = 0;
1346 tsk->nvcsw = tsk->nivcsw = 0;
1347#ifdef CONFIG_DETECT_HUNG_TASK
1348 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1349 tsk->last_switch_time = 0;
1350#endif
1351
1352 tsk->mm = NULL;
1353 tsk->active_mm = NULL;
1354
1355 /*
1356 * Are we cloning a kernel thread?
1357 *
1358 * We need to steal a active VM for that..
1359 */
1360 oldmm = current->mm;
1361 if (!oldmm)
1362 return 0;
1363
1364 /* initialize the new vmacache entries */
1365 vmacache_flush(tsk);
1366
1367 if (clone_flags & CLONE_VM) {
1368 mmget(oldmm);
1369 mm = oldmm;
1370 goto good_mm;
1371 }
1372
1373 retval = -ENOMEM;
1374 mm = dup_mm(tsk);
1375 if (!mm)
1376 goto fail_nomem;
1377
1378good_mm:
1379 tsk->mm = mm;
1380 tsk->active_mm = mm;
1381 return 0;
1382
1383fail_nomem:
1384 return retval;
1385}
1386
1387static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1388{
1389 struct fs_struct *fs = current->fs;
1390 if (clone_flags & CLONE_FS) {
1391 /* tsk->fs is already what we want */
1392 spin_lock(&fs->lock);
1393 if (fs->in_exec) {
1394 spin_unlock(&fs->lock);
1395 return -EAGAIN;
1396 }
1397 fs->users++;
1398 spin_unlock(&fs->lock);
1399 return 0;
1400 }
1401 tsk->fs = copy_fs_struct(fs);
1402 if (!tsk->fs)
1403 return -ENOMEM;
1404 return 0;
1405}
1406
1407static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1408{
1409 struct files_struct *oldf, *newf;
1410 int error = 0;
1411
1412 /*
1413 * A background process may not have any files ...
1414 */
1415 oldf = current->files;
1416 if (!oldf)
1417 goto out;
1418
1419 if (clone_flags & CLONE_FILES) {
1420 atomic_inc(&oldf->count);
1421 goto out;
1422 }
1423
1424 newf = dup_fd(oldf, &error);
1425 if (!newf)
1426 goto out;
1427
1428 tsk->files = newf;
1429 error = 0;
1430out:
1431 return error;
1432}
1433
1434static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1435{
1436#ifdef CONFIG_BLOCK
1437 struct io_context *ioc = current->io_context;
1438 struct io_context *new_ioc;
1439
1440 if (!ioc)
1441 return 0;
1442 /*
1443 * Share io context with parent, if CLONE_IO is set
1444 */
1445 if (clone_flags & CLONE_IO) {
1446 ioc_task_link(ioc);
1447 tsk->io_context = ioc;
1448 } else if (ioprio_valid(ioc->ioprio)) {
1449 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1450 if (unlikely(!new_ioc))
1451 return -ENOMEM;
1452
1453 new_ioc->ioprio = ioc->ioprio;
1454 put_io_context(new_ioc);
1455 }
1456#endif
1457 return 0;
1458}
1459
1460static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1461{
1462 struct sighand_struct *sig;
1463
1464 if (clone_flags & CLONE_SIGHAND) {
1465 refcount_inc(&current->sighand->count);
1466 return 0;
1467 }
1468 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1469 rcu_assign_pointer(tsk->sighand, sig);
1470 if (!sig)
1471 return -ENOMEM;
1472
1473 refcount_set(&sig->count, 1);
1474 spin_lock_irq(&current->sighand->siglock);
1475 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1476 spin_unlock_irq(&current->sighand->siglock);
1477 return 0;
1478}
1479
1480void __cleanup_sighand(struct sighand_struct *sighand)
1481{
1482 if (refcount_dec_and_test(&sighand->count)) {
1483 signalfd_cleanup(sighand);
1484 /*
1485 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1486 * without an RCU grace period, see __lock_task_sighand().
1487 */
1488 kmem_cache_free(sighand_cachep, sighand);
1489 }
1490}
1491
1492#ifdef CONFIG_POSIX_TIMERS
1493/*
1494 * Initialize POSIX timer handling for a thread group.
1495 */
1496static void posix_cpu_timers_init_group(struct signal_struct *sig)
1497{
1498 unsigned long cpu_limit;
1499
1500 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1501 if (cpu_limit != RLIM_INFINITY) {
1502 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1503 sig->cputimer.running = true;
1504 }
1505
1506 /* The timer lists. */
1507 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1508 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1509 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1510}
1511#else
1512static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1513#endif
1514
1515static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1516{
1517 struct signal_struct *sig;
1518
1519 if (clone_flags & CLONE_THREAD)
1520 return 0;
1521
1522 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1523 tsk->signal = sig;
1524 if (!sig)
1525 return -ENOMEM;
1526
1527 sig->nr_threads = 1;
1528 atomic_set(&sig->live, 1);
1529 refcount_set(&sig->sigcnt, 1);
1530
1531 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1532 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1533 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1534
1535 init_waitqueue_head(&sig->wait_chldexit);
1536 sig->curr_target = tsk;
1537 init_sigpending(&sig->shared_pending);
1538 INIT_HLIST_HEAD(&sig->multiprocess);
1539 seqlock_init(&sig->stats_lock);
1540 prev_cputime_init(&sig->prev_cputime);
1541
1542#ifdef CONFIG_POSIX_TIMERS
1543 INIT_LIST_HEAD(&sig->posix_timers);
1544 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1545 sig->real_timer.function = it_real_fn;
1546#endif
1547
1548 task_lock(current->group_leader);
1549 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1550 task_unlock(current->group_leader);
1551
1552 posix_cpu_timers_init_group(sig);
1553
1554 tty_audit_fork(sig);
1555 sched_autogroup_fork(sig);
1556
1557 sig->oom_score_adj = current->signal->oom_score_adj;
1558 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1559
1560 mutex_init(&sig->cred_guard_mutex);
1561
1562 return 0;
1563}
1564
1565static void copy_seccomp(struct task_struct *p)
1566{
1567#ifdef CONFIG_SECCOMP
1568 /*
1569 * Must be called with sighand->lock held, which is common to
1570 * all threads in the group. Holding cred_guard_mutex is not
1571 * needed because this new task is not yet running and cannot
1572 * be racing exec.
1573 */
1574 assert_spin_locked(&current->sighand->siglock);
1575
1576 /* Ref-count the new filter user, and assign it. */
1577 get_seccomp_filter(current);
1578 p->seccomp = current->seccomp;
1579
1580 /*
1581 * Explicitly enable no_new_privs here in case it got set
1582 * between the task_struct being duplicated and holding the
1583 * sighand lock. The seccomp state and nnp must be in sync.
1584 */
1585 if (task_no_new_privs(current))
1586 task_set_no_new_privs(p);
1587
1588 /*
1589 * If the parent gained a seccomp mode after copying thread
1590 * flags and between before we held the sighand lock, we have
1591 * to manually enable the seccomp thread flag here.
1592 */
1593 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1594 set_tsk_thread_flag(p, TIF_SECCOMP);
1595#endif
1596}
1597
1598SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1599{
1600 current->clear_child_tid = tidptr;
1601
1602 return task_pid_vnr(current);
1603}
1604
1605static void rt_mutex_init_task(struct task_struct *p)
1606{
1607 raw_spin_lock_init(&p->pi_lock);
1608#ifdef CONFIG_RT_MUTEXES
1609 p->pi_waiters = RB_ROOT_CACHED;
1610 p->pi_top_task = NULL;
1611 p->pi_blocked_on = NULL;
1612#endif
1613}
1614
1615#ifdef CONFIG_POSIX_TIMERS
1616/*
1617 * Initialize POSIX timer handling for a single task.
1618 */
1619static void posix_cpu_timers_init(struct task_struct *tsk)
1620{
1621 tsk->cputime_expires.prof_exp = 0;
1622 tsk->cputime_expires.virt_exp = 0;
1623 tsk->cputime_expires.sched_exp = 0;
1624 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1625 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1626 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1627}
1628#else
1629static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1630#endif
1631
1632static inline void init_task_pid_links(struct task_struct *task)
1633{
1634 enum pid_type type;
1635
1636 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1637 INIT_HLIST_NODE(&task->pid_links[type]);
1638 }
1639}
1640
1641static inline void
1642init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1643{
1644 if (type == PIDTYPE_PID)
1645 task->thread_pid = pid;
1646 else
1647 task->signal->pids[type] = pid;
1648}
1649
1650static inline void rcu_copy_process(struct task_struct *p)
1651{
1652#ifdef CONFIG_PREEMPT_RCU
1653 p->rcu_read_lock_nesting = 0;
1654 p->rcu_read_unlock_special.s = 0;
1655 p->rcu_blocked_node = NULL;
1656 INIT_LIST_HEAD(&p->rcu_node_entry);
1657#endif /* #ifdef CONFIG_PREEMPT_RCU */
1658#ifdef CONFIG_TASKS_RCU
1659 p->rcu_tasks_holdout = false;
1660 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1661 p->rcu_tasks_idle_cpu = -1;
1662#endif /* #ifdef CONFIG_TASKS_RCU */
1663}
1664
1665/*
1666 * This creates a new process as a copy of the old one,
1667 * but does not actually start it yet.
1668 *
1669 * It copies the registers, and all the appropriate
1670 * parts of the process environment (as per the clone
1671 * flags). The actual kick-off is left to the caller.
1672 */
1673static __latent_entropy struct task_struct *copy_process(
1674 unsigned long clone_flags,
1675 unsigned long stack_start,
1676 unsigned long stack_size,
1677 int __user *child_tidptr,
1678 struct pid *pid,
1679 int trace,
1680 unsigned long tls,
1681 int node)
1682{
1683 int retval;
1684 struct task_struct *p;
1685 struct multiprocess_signals delayed;
1686
1687 /*
1688 * Don't allow sharing the root directory with processes in a different
1689 * namespace
1690 */
1691 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1692 return ERR_PTR(-EINVAL);
1693
1694 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1695 return ERR_PTR(-EINVAL);
1696
1697 /*
1698 * Thread groups must share signals as well, and detached threads
1699 * can only be started up within the thread group.
1700 */
1701 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1702 return ERR_PTR(-EINVAL);
1703
1704 /*
1705 * Shared signal handlers imply shared VM. By way of the above,
1706 * thread groups also imply shared VM. Blocking this case allows
1707 * for various simplifications in other code.
1708 */
1709 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1710 return ERR_PTR(-EINVAL);
1711
1712 /*
1713 * Siblings of global init remain as zombies on exit since they are
1714 * not reaped by their parent (swapper). To solve this and to avoid
1715 * multi-rooted process trees, prevent global and container-inits
1716 * from creating siblings.
1717 */
1718 if ((clone_flags & CLONE_PARENT) &&
1719 current->signal->flags & SIGNAL_UNKILLABLE)
1720 return ERR_PTR(-EINVAL);
1721
1722 /*
1723 * If the new process will be in a different pid or user namespace
1724 * do not allow it to share a thread group with the forking task.
1725 */
1726 if (clone_flags & CLONE_THREAD) {
1727 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1728 (task_active_pid_ns(current) !=
1729 current->nsproxy->pid_ns_for_children))
1730 return ERR_PTR(-EINVAL);
1731 }
1732
1733 /*
1734 * Force any signals received before this point to be delivered
1735 * before the fork happens. Collect up signals sent to multiple
1736 * processes that happen during the fork and delay them so that
1737 * they appear to happen after the fork.
1738 */
1739 sigemptyset(&delayed.signal);
1740 INIT_HLIST_NODE(&delayed.node);
1741
1742 spin_lock_irq(&current->sighand->siglock);
1743 if (!(clone_flags & CLONE_THREAD))
1744 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1745 recalc_sigpending();
1746 spin_unlock_irq(&current->sighand->siglock);
1747 retval = -ERESTARTNOINTR;
1748 if (signal_pending(current))
1749 goto fork_out;
1750
1751 retval = -ENOMEM;
1752 p = dup_task_struct(current, node);
1753 if (!p)
1754 goto fork_out;
1755
1756 /*
1757 * This _must_ happen before we call free_task(), i.e. before we jump
1758 * to any of the bad_fork_* labels. This is to avoid freeing
1759 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1760 * kernel threads (PF_KTHREAD).
1761 */
1762 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1763 /*
1764 * Clear TID on mm_release()?
1765 */
1766 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1767
1768 ftrace_graph_init_task(p);
1769
1770 rt_mutex_init_task(p);
1771
1772#ifdef CONFIG_PROVE_LOCKING
1773 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1774 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1775#endif
1776 retval = -EAGAIN;
1777 if (atomic_read(&p->real_cred->user->processes) >=
1778 task_rlimit(p, RLIMIT_NPROC)) {
1779 if (p->real_cred->user != INIT_USER &&
1780 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1781 goto bad_fork_free;
1782 }
1783 current->flags &= ~PF_NPROC_EXCEEDED;
1784
1785 retval = copy_creds(p, clone_flags);
1786 if (retval < 0)
1787 goto bad_fork_free;
1788
1789 /*
1790 * If multiple threads are within copy_process(), then this check
1791 * triggers too late. This doesn't hurt, the check is only there
1792 * to stop root fork bombs.
1793 */
1794 retval = -EAGAIN;
1795 if (nr_threads >= max_threads)
1796 goto bad_fork_cleanup_count;
1797
1798 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1799 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1800 p->flags |= PF_FORKNOEXEC;
1801 INIT_LIST_HEAD(&p->children);
1802 INIT_LIST_HEAD(&p->sibling);
1803 rcu_copy_process(p);
1804 p->vfork_done = NULL;
1805 spin_lock_init(&p->alloc_lock);
1806
1807 init_sigpending(&p->pending);
1808
1809 p->utime = p->stime = p->gtime = 0;
1810#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1811 p->utimescaled = p->stimescaled = 0;
1812#endif
1813 prev_cputime_init(&p->prev_cputime);
1814
1815#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1816 seqcount_init(&p->vtime.seqcount);
1817 p->vtime.starttime = 0;
1818 p->vtime.state = VTIME_INACTIVE;
1819#endif
1820
1821#if defined(SPLIT_RSS_COUNTING)
1822 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1823#endif
1824
1825 p->default_timer_slack_ns = current->timer_slack_ns;
1826
1827#ifdef CONFIG_PSI
1828 p->psi_flags = 0;
1829#endif
1830
1831 task_io_accounting_init(&p->ioac);
1832 acct_clear_integrals(p);
1833
1834 posix_cpu_timers_init(p);
1835
1836 p->io_context = NULL;
1837 audit_set_context(p, NULL);
1838 cgroup_fork(p);
1839#ifdef CONFIG_NUMA
1840 p->mempolicy = mpol_dup(p->mempolicy);
1841 if (IS_ERR(p->mempolicy)) {
1842 retval = PTR_ERR(p->mempolicy);
1843 p->mempolicy = NULL;
1844 goto bad_fork_cleanup_threadgroup_lock;
1845 }
1846#endif
1847#ifdef CONFIG_CPUSETS
1848 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1849 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1850 seqcount_init(&p->mems_allowed_seq);
1851#endif
1852#ifdef CONFIG_TRACE_IRQFLAGS
1853 p->irq_events = 0;
1854 p->hardirqs_enabled = 0;
1855 p->hardirq_enable_ip = 0;
1856 p->hardirq_enable_event = 0;
1857 p->hardirq_disable_ip = _THIS_IP_;
1858 p->hardirq_disable_event = 0;
1859 p->softirqs_enabled = 1;
1860 p->softirq_enable_ip = _THIS_IP_;
1861 p->softirq_enable_event = 0;
1862 p->softirq_disable_ip = 0;
1863 p->softirq_disable_event = 0;
1864 p->hardirq_context = 0;
1865 p->softirq_context = 0;
1866#endif
1867
1868 p->pagefault_disabled = 0;
1869
1870#ifdef CONFIG_LOCKDEP
1871 p->lockdep_depth = 0; /* no locks held yet */
1872 p->curr_chain_key = 0;
1873 p->lockdep_recursion = 0;
1874 lockdep_init_task(p);
1875#endif
1876
1877#ifdef CONFIG_DEBUG_MUTEXES
1878 p->blocked_on = NULL; /* not blocked yet */
1879#endif
1880#ifdef CONFIG_BCACHE
1881 p->sequential_io = 0;
1882 p->sequential_io_avg = 0;
1883#endif
1884
1885 /* Perform scheduler related setup. Assign this task to a CPU. */
1886 retval = sched_fork(clone_flags, p);
1887 if (retval)
1888 goto bad_fork_cleanup_policy;
1889
1890 retval = perf_event_init_task(p);
1891 if (retval)
1892 goto bad_fork_cleanup_policy;
1893 retval = audit_alloc(p);
1894 if (retval)
1895 goto bad_fork_cleanup_perf;
1896 /* copy all the process information */
1897 shm_init_task(p);
1898 retval = security_task_alloc(p, clone_flags);
1899 if (retval)
1900 goto bad_fork_cleanup_audit;
1901 retval = copy_semundo(clone_flags, p);
1902 if (retval)
1903 goto bad_fork_cleanup_security;
1904 retval = copy_files(clone_flags, p);
1905 if (retval)
1906 goto bad_fork_cleanup_semundo;
1907 retval = copy_fs(clone_flags, p);
1908 if (retval)
1909 goto bad_fork_cleanup_files;
1910 retval = copy_sighand(clone_flags, p);
1911 if (retval)
1912 goto bad_fork_cleanup_fs;
1913 retval = copy_signal(clone_flags, p);
1914 if (retval)
1915 goto bad_fork_cleanup_sighand;
1916 retval = copy_mm(clone_flags, p);
1917 if (retval)
1918 goto bad_fork_cleanup_signal;
1919 retval = copy_namespaces(clone_flags, p);
1920 if (retval)
1921 goto bad_fork_cleanup_mm;
1922 retval = copy_io(clone_flags, p);
1923 if (retval)
1924 goto bad_fork_cleanup_namespaces;
1925 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1926 if (retval)
1927 goto bad_fork_cleanup_io;
1928
1929 stackleak_task_init(p);
1930
1931 if (pid != &init_struct_pid) {
1932 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1933 if (IS_ERR(pid)) {
1934 retval = PTR_ERR(pid);
1935 goto bad_fork_cleanup_thread;
1936 }
1937 }
1938
1939#ifdef CONFIG_BLOCK
1940 p->plug = NULL;
1941#endif
1942#ifdef CONFIG_FUTEX
1943 p->robust_list = NULL;
1944#ifdef CONFIG_COMPAT
1945 p->compat_robust_list = NULL;
1946#endif
1947 INIT_LIST_HEAD(&p->pi_state_list);
1948 p->pi_state_cache = NULL;
1949#endif
1950 /*
1951 * sigaltstack should be cleared when sharing the same VM
1952 */
1953 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1954 sas_ss_reset(p);
1955
1956 /*
1957 * Syscall tracing and stepping should be turned off in the
1958 * child regardless of CLONE_PTRACE.
1959 */
1960 user_disable_single_step(p);
1961 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1962#ifdef TIF_SYSCALL_EMU
1963 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1964#endif
1965 clear_all_latency_tracing(p);
1966
1967 /* ok, now we should be set up.. */
1968 p->pid = pid_nr(pid);
1969 if (clone_flags & CLONE_THREAD) {
1970 p->exit_signal = -1;
1971 p->group_leader = current->group_leader;
1972 p->tgid = current->tgid;
1973 } else {
1974 if (clone_flags & CLONE_PARENT)
1975 p->exit_signal = current->group_leader->exit_signal;
1976 else
1977 p->exit_signal = (clone_flags & CSIGNAL);
1978 p->group_leader = p;
1979 p->tgid = p->pid;
1980 }
1981
1982 p->nr_dirtied = 0;
1983 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1984 p->dirty_paused_when = 0;
1985
1986 p->pdeath_signal = 0;
1987 INIT_LIST_HEAD(&p->thread_group);
1988 p->task_works = NULL;
1989
1990 cgroup_threadgroup_change_begin(current);
1991 /*
1992 * Ensure that the cgroup subsystem policies allow the new process to be
1993 * forked. It should be noted the the new process's css_set can be changed
1994 * between here and cgroup_post_fork() if an organisation operation is in
1995 * progress.
1996 */
1997 retval = cgroup_can_fork(p);
1998 if (retval)
1999 goto bad_fork_free_pid;
2000
2001 /*
2002 * From this point on we must avoid any synchronous user-space
2003 * communication until we take the tasklist-lock. In particular, we do
2004 * not want user-space to be able to predict the process start-time by
2005 * stalling fork(2) after we recorded the start_time but before it is
2006 * visible to the system.
2007 */
2008
2009 p->start_time = ktime_get_ns();
2010 p->real_start_time = ktime_get_boot_ns();
2011
2012 /*
2013 * Make it visible to the rest of the system, but dont wake it up yet.
2014 * Need tasklist lock for parent etc handling!
2015 */
2016 write_lock_irq(&tasklist_lock);
2017
2018 /* CLONE_PARENT re-uses the old parent */
2019 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2020 p->real_parent = current->real_parent;
2021 p->parent_exec_id = current->parent_exec_id;
2022 } else {
2023 p->real_parent = current;
2024 p->parent_exec_id = current->self_exec_id;
2025 }
2026
2027 klp_copy_process(p);
2028
2029 spin_lock(&current->sighand->siglock);
2030
2031 /*
2032 * Copy seccomp details explicitly here, in case they were changed
2033 * before holding sighand lock.
2034 */
2035 copy_seccomp(p);
2036
2037 rseq_fork(p, clone_flags);
2038
2039 /* Don't start children in a dying pid namespace */
2040 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2041 retval = -ENOMEM;
2042 goto bad_fork_cancel_cgroup;
2043 }
2044
2045 /* Let kill terminate clone/fork in the middle */
2046 if (fatal_signal_pending(current)) {
2047 retval = -EINTR;
2048 goto bad_fork_cancel_cgroup;
2049 }
2050
2051
2052 init_task_pid_links(p);
2053 if (likely(p->pid)) {
2054 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2055
2056 init_task_pid(p, PIDTYPE_PID, pid);
2057 if (thread_group_leader(p)) {
2058 init_task_pid(p, PIDTYPE_TGID, pid);
2059 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2060 init_task_pid(p, PIDTYPE_SID, task_session(current));
2061
2062 if (is_child_reaper(pid)) {
2063 ns_of_pid(pid)->child_reaper = p;
2064 p->signal->flags |= SIGNAL_UNKILLABLE;
2065 }
2066 p->signal->shared_pending.signal = delayed.signal;
2067 p->signal->tty = tty_kref_get(current->signal->tty);
2068 /*
2069 * Inherit has_child_subreaper flag under the same
2070 * tasklist_lock with adding child to the process tree
2071 * for propagate_has_child_subreaper optimization.
2072 */
2073 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2074 p->real_parent->signal->is_child_subreaper;
2075 list_add_tail(&p->sibling, &p->real_parent->children);
2076 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2077 attach_pid(p, PIDTYPE_TGID);
2078 attach_pid(p, PIDTYPE_PGID);
2079 attach_pid(p, PIDTYPE_SID);
2080 __this_cpu_inc(process_counts);
2081 } else {
2082 current->signal->nr_threads++;
2083 atomic_inc(&current->signal->live);
2084 refcount_inc(&current->signal->sigcnt);
2085 task_join_group_stop(p);
2086 list_add_tail_rcu(&p->thread_group,
2087 &p->group_leader->thread_group);
2088 list_add_tail_rcu(&p->thread_node,
2089 &p->signal->thread_head);
2090 }
2091 attach_pid(p, PIDTYPE_PID);
2092 nr_threads++;
2093 }
2094 total_forks++;
2095 hlist_del_init(&delayed.node);
2096 spin_unlock(&current->sighand->siglock);
2097 syscall_tracepoint_update(p);
2098 write_unlock_irq(&tasklist_lock);
2099
2100 proc_fork_connector(p);
2101 cgroup_post_fork(p);
2102 cgroup_threadgroup_change_end(current);
2103 perf_event_fork(p);
2104
2105 trace_task_newtask(p, clone_flags);
2106 uprobe_copy_process(p, clone_flags);
2107
2108 return p;
2109
2110bad_fork_cancel_cgroup:
2111 spin_unlock(&current->sighand->siglock);
2112 write_unlock_irq(&tasklist_lock);
2113 cgroup_cancel_fork(p);
2114bad_fork_free_pid:
2115 cgroup_threadgroup_change_end(current);
2116 if (pid != &init_struct_pid)
2117 free_pid(pid);
2118bad_fork_cleanup_thread:
2119 exit_thread(p);
2120bad_fork_cleanup_io:
2121 if (p->io_context)
2122 exit_io_context(p);
2123bad_fork_cleanup_namespaces:
2124 exit_task_namespaces(p);
2125bad_fork_cleanup_mm:
2126 if (p->mm)
2127 mmput(p->mm);
2128bad_fork_cleanup_signal:
2129 if (!(clone_flags & CLONE_THREAD))
2130 free_signal_struct(p->signal);
2131bad_fork_cleanup_sighand:
2132 __cleanup_sighand(p->sighand);
2133bad_fork_cleanup_fs:
2134 exit_fs(p); /* blocking */
2135bad_fork_cleanup_files:
2136 exit_files(p); /* blocking */
2137bad_fork_cleanup_semundo:
2138 exit_sem(p);
2139bad_fork_cleanup_security:
2140 security_task_free(p);
2141bad_fork_cleanup_audit:
2142 audit_free(p);
2143bad_fork_cleanup_perf:
2144 perf_event_free_task(p);
2145bad_fork_cleanup_policy:
2146 lockdep_free_task(p);
2147#ifdef CONFIG_NUMA
2148 mpol_put(p->mempolicy);
2149bad_fork_cleanup_threadgroup_lock:
2150#endif
2151 delayacct_tsk_free(p);
2152bad_fork_cleanup_count:
2153 atomic_dec(&p->cred->user->processes);
2154 exit_creds(p);
2155bad_fork_free:
2156 p->state = TASK_DEAD;
2157 put_task_stack(p);
2158 free_task(p);
2159fork_out:
2160 spin_lock_irq(&current->sighand->siglock);
2161 hlist_del_init(&delayed.node);
2162 spin_unlock_irq(&current->sighand->siglock);
2163 return ERR_PTR(retval);
2164}
2165
2166static inline void init_idle_pids(struct task_struct *idle)
2167{
2168 enum pid_type type;
2169
2170 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2171 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2172 init_task_pid(idle, type, &init_struct_pid);
2173 }
2174}
2175
2176struct task_struct *fork_idle(int cpu)
2177{
2178 struct task_struct *task;
2179 task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2180 cpu_to_node(cpu));
2181 if (!IS_ERR(task)) {
2182 init_idle_pids(task);
2183 init_idle(task, cpu);
2184 }
2185
2186 return task;
2187}
2188
2189/*
2190 * Ok, this is the main fork-routine.
2191 *
2192 * It copies the process, and if successful kick-starts
2193 * it and waits for it to finish using the VM if required.
2194 */
2195long _do_fork(unsigned long clone_flags,
2196 unsigned long stack_start,
2197 unsigned long stack_size,
2198 int __user *parent_tidptr,
2199 int __user *child_tidptr,
2200 unsigned long tls)
2201{
2202 struct completion vfork;
2203 struct pid *pid;
2204 struct task_struct *p;
2205 int trace = 0;
2206 long nr;
2207
2208 /*
2209 * Determine whether and which event to report to ptracer. When
2210 * called from kernel_thread or CLONE_UNTRACED is explicitly
2211 * requested, no event is reported; otherwise, report if the event
2212 * for the type of forking is enabled.
2213 */
2214 if (!(clone_flags & CLONE_UNTRACED)) {
2215 if (clone_flags & CLONE_VFORK)
2216 trace = PTRACE_EVENT_VFORK;
2217 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2218 trace = PTRACE_EVENT_CLONE;
2219 else
2220 trace = PTRACE_EVENT_FORK;
2221
2222 if (likely(!ptrace_event_enabled(current, trace)))
2223 trace = 0;
2224 }
2225
2226 p = copy_process(clone_flags, stack_start, stack_size,
2227 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2228 add_latent_entropy();
2229
2230 if (IS_ERR(p))
2231 return PTR_ERR(p);
2232
2233 /*
2234 * Do this prior waking up the new thread - the thread pointer
2235 * might get invalid after that point, if the thread exits quickly.
2236 */
2237 trace_sched_process_fork(current, p);
2238
2239 pid = get_task_pid(p, PIDTYPE_PID);
2240 nr = pid_vnr(pid);
2241
2242 if (clone_flags & CLONE_PARENT_SETTID)
2243 put_user(nr, parent_tidptr);
2244
2245 if (clone_flags & CLONE_VFORK) {
2246 p->vfork_done = &vfork;
2247 init_completion(&vfork);
2248 get_task_struct(p);
2249 }
2250
2251 wake_up_new_task(p);
2252
2253 /* forking complete and child started to run, tell ptracer */
2254 if (unlikely(trace))
2255 ptrace_event_pid(trace, pid);
2256
2257 if (clone_flags & CLONE_VFORK) {
2258 if (!wait_for_vfork_done(p, &vfork))
2259 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2260 }
2261
2262 put_pid(pid);
2263 return nr;
2264}
2265
2266#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2267/* For compatibility with architectures that call do_fork directly rather than
2268 * using the syscall entry points below. */
2269long do_fork(unsigned long clone_flags,
2270 unsigned long stack_start,
2271 unsigned long stack_size,
2272 int __user *parent_tidptr,
2273 int __user *child_tidptr)
2274{
2275 return _do_fork(clone_flags, stack_start, stack_size,
2276 parent_tidptr, child_tidptr, 0);
2277}
2278#endif
2279
2280/*
2281 * Create a kernel thread.
2282 */
2283pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2284{
2285 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2286 (unsigned long)arg, NULL, NULL, 0);
2287}
2288
2289#ifdef __ARCH_WANT_SYS_FORK
2290SYSCALL_DEFINE0(fork)
2291{
2292#ifdef CONFIG_MMU
2293 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2294#else
2295 /* can not support in nommu mode */
2296 return -EINVAL;
2297#endif
2298}
2299#endif
2300
2301#ifdef __ARCH_WANT_SYS_VFORK
2302SYSCALL_DEFINE0(vfork)
2303{
2304 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2305 0, NULL, NULL, 0);
2306}
2307#endif
2308
2309#ifdef __ARCH_WANT_SYS_CLONE
2310#ifdef CONFIG_CLONE_BACKWARDS
2311SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2312 int __user *, parent_tidptr,
2313 unsigned long, tls,
2314 int __user *, child_tidptr)
2315#elif defined(CONFIG_CLONE_BACKWARDS2)
2316SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2317 int __user *, parent_tidptr,
2318 int __user *, child_tidptr,
2319 unsigned long, tls)
2320#elif defined(CONFIG_CLONE_BACKWARDS3)
2321SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2322 int, stack_size,
2323 int __user *, parent_tidptr,
2324 int __user *, child_tidptr,
2325 unsigned long, tls)
2326#else
2327SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2328 int __user *, parent_tidptr,
2329 int __user *, child_tidptr,
2330 unsigned long, tls)
2331#endif
2332{
2333 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2334}
2335#endif
2336
2337void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2338{
2339 struct task_struct *leader, *parent, *child;
2340 int res;
2341
2342 read_lock(&tasklist_lock);
2343 leader = top = top->group_leader;
2344down:
2345 for_each_thread(leader, parent) {
2346 list_for_each_entry(child, &parent->children, sibling) {
2347 res = visitor(child, data);
2348 if (res) {
2349 if (res < 0)
2350 goto out;
2351 leader = child;
2352 goto down;
2353 }
2354up:
2355 ;
2356 }
2357 }
2358
2359 if (leader != top) {
2360 child = leader;
2361 parent = child->real_parent;
2362 leader = parent->group_leader;
2363 goto up;
2364 }
2365out:
2366 read_unlock(&tasklist_lock);
2367}
2368
2369#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2370#define ARCH_MIN_MMSTRUCT_ALIGN 0
2371#endif
2372
2373static void sighand_ctor(void *data)
2374{
2375 struct sighand_struct *sighand = data;
2376
2377 spin_lock_init(&sighand->siglock);
2378 init_waitqueue_head(&sighand->signalfd_wqh);
2379}
2380
2381void __init proc_caches_init(void)
2382{
2383 unsigned int mm_size;
2384
2385 sighand_cachep = kmem_cache_create("sighand_cache",
2386 sizeof(struct sighand_struct), 0,
2387 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2388 SLAB_ACCOUNT, sighand_ctor);
2389 signal_cachep = kmem_cache_create("signal_cache",
2390 sizeof(struct signal_struct), 0,
2391 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2392 NULL);
2393 files_cachep = kmem_cache_create("files_cache",
2394 sizeof(struct files_struct), 0,
2395 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2396 NULL);
2397 fs_cachep = kmem_cache_create("fs_cache",
2398 sizeof(struct fs_struct), 0,
2399 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2400 NULL);
2401
2402 /*
2403 * The mm_cpumask is located at the end of mm_struct, and is
2404 * dynamically sized based on the maximum CPU number this system
2405 * can have, taking hotplug into account (nr_cpu_ids).
2406 */
2407 mm_size = sizeof(struct mm_struct) + cpumask_size();
2408
2409 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2410 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2411 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2412 offsetof(struct mm_struct, saved_auxv),
2413 sizeof_field(struct mm_struct, saved_auxv),
2414 NULL);
2415 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2416 mmap_init();
2417 nsproxy_cache_init();
2418}
2419
2420/*
2421 * Check constraints on flags passed to the unshare system call.
2422 */
2423static int check_unshare_flags(unsigned long unshare_flags)
2424{
2425 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2426 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2427 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2428 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2429 return -EINVAL;
2430 /*
2431 * Not implemented, but pretend it works if there is nothing
2432 * to unshare. Note that unsharing the address space or the
2433 * signal handlers also need to unshare the signal queues (aka
2434 * CLONE_THREAD).
2435 */
2436 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2437 if (!thread_group_empty(current))
2438 return -EINVAL;
2439 }
2440 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2441 if (refcount_read(&current->sighand->count) > 1)
2442 return -EINVAL;
2443 }
2444 if (unshare_flags & CLONE_VM) {
2445 if (!current_is_single_threaded())
2446 return -EINVAL;
2447 }
2448
2449 return 0;
2450}
2451
2452/*
2453 * Unshare the filesystem structure if it is being shared
2454 */
2455static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2456{
2457 struct fs_struct *fs = current->fs;
2458
2459 if (!(unshare_flags & CLONE_FS) || !fs)
2460 return 0;
2461
2462 /* don't need lock here; in the worst case we'll do useless copy */
2463 if (fs->users == 1)
2464 return 0;
2465
2466 *new_fsp = copy_fs_struct(fs);
2467 if (!*new_fsp)
2468 return -ENOMEM;
2469
2470 return 0;
2471}
2472
2473/*
2474 * Unshare file descriptor table if it is being shared
2475 */
2476static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2477{
2478 struct files_struct *fd = current->files;
2479 int error = 0;
2480
2481 if ((unshare_flags & CLONE_FILES) &&
2482 (fd && atomic_read(&fd->count) > 1)) {
2483 *new_fdp = dup_fd(fd, &error);
2484 if (!*new_fdp)
2485 return error;
2486 }
2487
2488 return 0;
2489}
2490
2491/*
2492 * unshare allows a process to 'unshare' part of the process
2493 * context which was originally shared using clone. copy_*
2494 * functions used by do_fork() cannot be used here directly
2495 * because they modify an inactive task_struct that is being
2496 * constructed. Here we are modifying the current, active,
2497 * task_struct.
2498 */
2499int ksys_unshare(unsigned long unshare_flags)
2500{
2501 struct fs_struct *fs, *new_fs = NULL;
2502 struct files_struct *fd, *new_fd = NULL;
2503 struct cred *new_cred = NULL;
2504 struct nsproxy *new_nsproxy = NULL;
2505 int do_sysvsem = 0;
2506 int err;
2507
2508 /*
2509 * If unsharing a user namespace must also unshare the thread group
2510 * and unshare the filesystem root and working directories.
2511 */
2512 if (unshare_flags & CLONE_NEWUSER)
2513 unshare_flags |= CLONE_THREAD | CLONE_FS;
2514 /*
2515 * If unsharing vm, must also unshare signal handlers.
2516 */
2517 if (unshare_flags & CLONE_VM)
2518 unshare_flags |= CLONE_SIGHAND;
2519 /*
2520 * If unsharing a signal handlers, must also unshare the signal queues.
2521 */
2522 if (unshare_flags & CLONE_SIGHAND)
2523 unshare_flags |= CLONE_THREAD;
2524 /*
2525 * If unsharing namespace, must also unshare filesystem information.
2526 */
2527 if (unshare_flags & CLONE_NEWNS)
2528 unshare_flags |= CLONE_FS;
2529
2530 err = check_unshare_flags(unshare_flags);
2531 if (err)
2532 goto bad_unshare_out;
2533 /*
2534 * CLONE_NEWIPC must also detach from the undolist: after switching
2535 * to a new ipc namespace, the semaphore arrays from the old
2536 * namespace are unreachable.
2537 */
2538 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2539 do_sysvsem = 1;
2540 err = unshare_fs(unshare_flags, &new_fs);
2541 if (err)
2542 goto bad_unshare_out;
2543 err = unshare_fd(unshare_flags, &new_fd);
2544 if (err)
2545 goto bad_unshare_cleanup_fs;
2546 err = unshare_userns(unshare_flags, &new_cred);
2547 if (err)
2548 goto bad_unshare_cleanup_fd;
2549 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2550 new_cred, new_fs);
2551 if (err)
2552 goto bad_unshare_cleanup_cred;
2553
2554 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2555 if (do_sysvsem) {
2556 /*
2557 * CLONE_SYSVSEM is equivalent to sys_exit().
2558 */
2559 exit_sem(current);
2560 }
2561 if (unshare_flags & CLONE_NEWIPC) {
2562 /* Orphan segments in old ns (see sem above). */
2563 exit_shm(current);
2564 shm_init_task(current);
2565 }
2566
2567 if (new_nsproxy)
2568 switch_task_namespaces(current, new_nsproxy);
2569
2570 task_lock(current);
2571
2572 if (new_fs) {
2573 fs = current->fs;
2574 spin_lock(&fs->lock);
2575 current->fs = new_fs;
2576 if (--fs->users)
2577 new_fs = NULL;
2578 else
2579 new_fs = fs;
2580 spin_unlock(&fs->lock);
2581 }
2582
2583 if (new_fd) {
2584 fd = current->files;
2585 current->files = new_fd;
2586 new_fd = fd;
2587 }
2588
2589 task_unlock(current);
2590
2591 if (new_cred) {
2592 /* Install the new user namespace */
2593 commit_creds(new_cred);
2594 new_cred = NULL;
2595 }
2596 }
2597
2598 perf_event_namespaces(current);
2599
2600bad_unshare_cleanup_cred:
2601 if (new_cred)
2602 put_cred(new_cred);
2603bad_unshare_cleanup_fd:
2604 if (new_fd)
2605 put_files_struct(new_fd);
2606
2607bad_unshare_cleanup_fs:
2608 if (new_fs)
2609 free_fs_struct(new_fs);
2610
2611bad_unshare_out:
2612 return err;
2613}
2614
2615SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2616{
2617 return ksys_unshare(unshare_flags);
2618}
2619
2620/*
2621 * Helper to unshare the files of the current task.
2622 * We don't want to expose copy_files internals to
2623 * the exec layer of the kernel.
2624 */
2625
2626int unshare_files(struct files_struct **displaced)
2627{
2628 struct task_struct *task = current;
2629 struct files_struct *copy = NULL;
2630 int error;
2631
2632 error = unshare_fd(CLONE_FILES, &copy);
2633 if (error || !copy) {
2634 *displaced = NULL;
2635 return error;
2636 }
2637 *displaced = task->files;
2638 task_lock(task);
2639 task->files = copy;
2640 task_unlock(task);
2641 return 0;
2642}
2643
2644int sysctl_max_threads(struct ctl_table *table, int write,
2645 void __user *buffer, size_t *lenp, loff_t *ppos)
2646{
2647 struct ctl_table t;
2648 int ret;
2649 int threads = max_threads;
2650 int min = MIN_THREADS;
2651 int max = MAX_THREADS;
2652
2653 t = *table;
2654 t.data = &threads;
2655 t.extra1 = &min;
2656 t.extra2 = &max;
2657
2658 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2659 if (ret || !write)
2660 return ret;
2661
2662 set_max_threads(threads);
2663
2664 return 0;
2665}
2666