projects/linux·blob v5.2 · refs

2791 lines · 66.4 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/binfmts.h>
41#include <linux/mman.h>
42#include <linux/mmu_notifier.h>
43#include <linux/hmm.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/vmacache.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/jiffies.h>
57#include <linux/futex.h>
58#include <linux/compat.h>
59#include <linux/kthread.h>
60#include <linux/task_io_accounting_ops.h>
61#include <linux/rcupdate.h>
62#include <linux/ptrace.h>
63#include <linux/mount.h>
64#include <linux/audit.h>
65#include <linux/memcontrol.h>
66#include <linux/ftrace.h>
67#include <linux/proc_fs.h>
68#include <linux/profile.h>
69#include <linux/rmap.h>
70#include <linux/ksm.h>
71#include <linux/acct.h>
72#include <linux/userfaultfd_k.h>
73#include <linux/tsacct_kern.h>
74#include <linux/cn_proc.h>
75#include <linux/freezer.h>
76#include <linux/delayacct.h>
77#include <linux/taskstats_kern.h>
78#include <linux/random.h>
79#include <linux/tty.h>
80#include <linux/blkdev.h>
81#include <linux/fs_struct.h>
82#include <linux/magic.h>
83#include <linux/perf_event.h>
84#include <linux/posix-timers.h>
85#include <linux/user-return-notifier.h>
86#include <linux/oom.h>
87#include <linux/khugepaged.h>
88#include <linux/signalfd.h>
89#include <linux/uprobes.h>
90#include <linux/aio.h>
91#include <linux/compiler.h>
92#include <linux/sysctl.h>
93#include <linux/kcov.h>
94#include <linux/livepatch.h>
95#include <linux/thread_info.h>
96#include <linux/stackleak.h>
97
98#include <asm/pgtable.h>
99#include <asm/pgalloc.h>
100#include <linux/uaccess.h>
101#include <asm/mmu_context.h>
102#include <asm/cacheflush.h>
103#include <asm/tlbflush.h>
104
105#include <trace/events/sched.h>
106
107#define CREATE_TRACE_POINTS
108#include <trace/events/task.h>
109
110/*
111 * Minimum number of threads to boot the kernel
112 */
113#define MIN_THREADS 20
114
115/*
116 * Maximum number of threads
117 */
118#define MAX_THREADS FUTEX_TID_MASK
119
120/*
121 * Protected counters by write_lock_irq(&tasklist_lock)
122 */
123unsigned long total_forks; /* Handle normal Linux uptimes. */
124int nr_threads; /* The idle threads do not count.. */
125
126static int max_threads; /* tunable limit on nr_threads */
127
128DEFINE_PER_CPU(unsigned long, process_counts) = 0;
129
130__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
131
132#ifdef CONFIG_PROVE_RCU
133int lockdep_tasklist_lock_is_held(void)
134{
135 return lockdep_is_held(&tasklist_lock);
136}
137EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
138#endif /* #ifdef CONFIG_PROVE_RCU */
139
140int nr_processes(void)
141{
142 int cpu;
143 int total = 0;
144
145 for_each_possible_cpu(cpu)
146 total += per_cpu(process_counts, cpu);
147
148 return total;
149}
150
151void __weak arch_release_task_struct(struct task_struct *tsk)
152{
153}
154
155#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
156static struct kmem_cache *task_struct_cachep;
157
158static inline struct task_struct *alloc_task_struct_node(int node)
159{
160 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
161}
162
163static inline void free_task_struct(struct task_struct *tsk)
164{
165 kmem_cache_free(task_struct_cachep, tsk);
166}
167#endif
168
169#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
170
171/*
172 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
173 * kmemcache based allocator.
174 */
175# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
176
177#ifdef CONFIG_VMAP_STACK
178/*
179 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
180 * flush. Try to minimize the number of calls by caching stacks.
181 */
182#define NR_CACHED_STACKS 2
183static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
184
185static int free_vm_stack_cache(unsigned int cpu)
186{
187 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
188 int i;
189
190 for (i = 0; i < NR_CACHED_STACKS; i++) {
191 struct vm_struct *vm_stack = cached_vm_stacks[i];
192
193 if (!vm_stack)
194 continue;
195
196 vfree(vm_stack->addr);
197 cached_vm_stacks[i] = NULL;
198 }
199
200 return 0;
201}
202#endif
203
204static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
205{
206#ifdef CONFIG_VMAP_STACK
207 void *stack;
208 int i;
209
210 for (i = 0; i < NR_CACHED_STACKS; i++) {
211 struct vm_struct *s;
212
213 s = this_cpu_xchg(cached_stacks[i], NULL);
214
215 if (!s)
216 continue;
217
218 /* Clear stale pointers from reused stack. */
219 memset(s->addr, 0, THREAD_SIZE);
220
221 tsk->stack_vm_area = s;
222 tsk->stack = s->addr;
223 return s->addr;
224 }
225
226 /*
227 * Allocated stacks are cached and later reused by new threads,
228 * so memcg accounting is performed manually on assigning/releasing
229 * stacks to tasks. Drop __GFP_ACCOUNT.
230 */
231 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
232 VMALLOC_START, VMALLOC_END,
233 THREADINFO_GFP & ~__GFP_ACCOUNT,
234 PAGE_KERNEL,
235 0, node, __builtin_return_address(0));
236
237 /*
238 * We can't call find_vm_area() in interrupt context, and
239 * free_thread_stack() can be called in interrupt context,
240 * so cache the vm_struct.
241 */
242 if (stack) {
243 tsk->stack_vm_area = find_vm_area(stack);
244 tsk->stack = stack;
245 }
246 return stack;
247#else
248 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
249 THREAD_SIZE_ORDER);
250
251 if (likely(page)) {
252 tsk->stack = page_address(page);
253 return tsk->stack;
254 }
255 return NULL;
256#endif
257}
258
259static inline void free_thread_stack(struct task_struct *tsk)
260{
261#ifdef CONFIG_VMAP_STACK
262 struct vm_struct *vm = task_stack_vm_area(tsk);
263
264 if (vm) {
265 int i;
266
267 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
268 mod_memcg_page_state(vm->pages[i],
269 MEMCG_KERNEL_STACK_KB,
270 -(int)(PAGE_SIZE / 1024));
271
272 memcg_kmem_uncharge(vm->pages[i], 0);
273 }
274
275 for (i = 0; i < NR_CACHED_STACKS; i++) {
276 if (this_cpu_cmpxchg(cached_stacks[i],
277 NULL, tsk->stack_vm_area) != NULL)
278 continue;
279
280 return;
281 }
282
283 vfree_atomic(tsk->stack);
284 return;
285 }
286#endif
287
288 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
289}
290# else
291static struct kmem_cache *thread_stack_cache;
292
293static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
294 int node)
295{
296 unsigned long *stack;
297 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
298 tsk->stack = stack;
299 return stack;
300}
301
302static void free_thread_stack(struct task_struct *tsk)
303{
304 kmem_cache_free(thread_stack_cache, tsk->stack);
305}
306
307void thread_stack_cache_init(void)
308{
309 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
310 THREAD_SIZE, THREAD_SIZE, 0, 0,
311 THREAD_SIZE, NULL);
312 BUG_ON(thread_stack_cache == NULL);
313}
314# endif
315#endif
316
317/* SLAB cache for signal_struct structures (tsk->signal) */
318static struct kmem_cache *signal_cachep;
319
320/* SLAB cache for sighand_struct structures (tsk->sighand) */
321struct kmem_cache *sighand_cachep;
322
323/* SLAB cache for files_struct structures (tsk->files) */
324struct kmem_cache *files_cachep;
325
326/* SLAB cache for fs_struct structures (tsk->fs) */
327struct kmem_cache *fs_cachep;
328
329/* SLAB cache for vm_area_struct structures */
330static struct kmem_cache *vm_area_cachep;
331
332/* SLAB cache for mm_struct structures (tsk->mm) */
333static struct kmem_cache *mm_cachep;
334
335struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
336{
337 struct vm_area_struct *vma;
338
339 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
340 if (vma)
341 vma_init(vma, mm);
342 return vma;
343}
344
345struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
346{
347 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
348
349 if (new) {
350 *new = *orig;
351 INIT_LIST_HEAD(&new->anon_vma_chain);
352 }
353 return new;
354}
355
356void vm_area_free(struct vm_area_struct *vma)
357{
358 kmem_cache_free(vm_area_cachep, vma);
359}
360
361static void account_kernel_stack(struct task_struct *tsk, int account)
362{
363 void *stack = task_stack_page(tsk);
364 struct vm_struct *vm = task_stack_vm_area(tsk);
365
366 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
367
368 if (vm) {
369 int i;
370
371 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
372
373 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
374 mod_zone_page_state(page_zone(vm->pages[i]),
375 NR_KERNEL_STACK_KB,
376 PAGE_SIZE / 1024 * account);
377 }
378 } else {
379 /*
380 * All stack pages are in the same zone and belong to the
381 * same memcg.
382 */
383 struct page *first_page = virt_to_page(stack);
384
385 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
386 THREAD_SIZE / 1024 * account);
387
388 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
389 account * (THREAD_SIZE / 1024));
390 }
391}
392
393static int memcg_charge_kernel_stack(struct task_struct *tsk)
394{
395#ifdef CONFIG_VMAP_STACK
396 struct vm_struct *vm = task_stack_vm_area(tsk);
397 int ret;
398
399 if (vm) {
400 int i;
401
402 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
403 /*
404 * If memcg_kmem_charge() fails, page->mem_cgroup
405 * pointer is NULL, and both memcg_kmem_uncharge()
406 * and mod_memcg_page_state() in free_thread_stack()
407 * will ignore this page. So it's safe.
408 */
409 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
410 if (ret)
411 return ret;
412
413 mod_memcg_page_state(vm->pages[i],
414 MEMCG_KERNEL_STACK_KB,
415 PAGE_SIZE / 1024);
416 }
417 }
418#endif
419 return 0;
420}
421
422static void release_task_stack(struct task_struct *tsk)
423{
424 if (WARN_ON(tsk->state != TASK_DEAD))
425 return; /* Better to leak the stack than to free prematurely */
426
427 account_kernel_stack(tsk, -1);
428 free_thread_stack(tsk);
429 tsk->stack = NULL;
430#ifdef CONFIG_VMAP_STACK
431 tsk->stack_vm_area = NULL;
432#endif
433}
434
435#ifdef CONFIG_THREAD_INFO_IN_TASK
436void put_task_stack(struct task_struct *tsk)
437{
438 if (refcount_dec_and_test(&tsk->stack_refcount))
439 release_task_stack(tsk);
440}
441#endif
442
443void free_task(struct task_struct *tsk)
444{
445#ifndef CONFIG_THREAD_INFO_IN_TASK
446 /*
447 * The task is finally done with both the stack and thread_info,
448 * so free both.
449 */
450 release_task_stack(tsk);
451#else
452 /*
453 * If the task had a separate stack allocation, it should be gone
454 * by now.
455 */
456 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
457#endif
458 rt_mutex_debug_task_free(tsk);
459 ftrace_graph_exit_task(tsk);
460 put_seccomp_filter(tsk);
461 arch_release_task_struct(tsk);
462 if (tsk->flags & PF_KTHREAD)
463 free_kthread_struct(tsk);
464 free_task_struct(tsk);
465}
466EXPORT_SYMBOL(free_task);
467
468#ifdef CONFIG_MMU
469static __latent_entropy int dup_mmap(struct mm_struct *mm,
470 struct mm_struct *oldmm)
471{
472 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
473 struct rb_node **rb_link, *rb_parent;
474 int retval;
475 unsigned long charge;
476 LIST_HEAD(uf);
477
478 uprobe_start_dup_mmap();
479 if (down_write_killable(&oldmm->mmap_sem)) {
480 retval = -EINTR;
481 goto fail_uprobe_end;
482 }
483 flush_cache_dup_mm(oldmm);
484 uprobe_dup_mmap(oldmm, mm);
485 /*
486 * Not linked in yet - no deadlock potential:
487 */
488 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
489
490 /* No ordering required: file already has been exposed. */
491 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
492
493 mm->total_vm = oldmm->total_vm;
494 mm->data_vm = oldmm->data_vm;
495 mm->exec_vm = oldmm->exec_vm;
496 mm->stack_vm = oldmm->stack_vm;
497
498 rb_link = &mm->mm_rb.rb_node;
499 rb_parent = NULL;
500 pprev = &mm->mmap;
501 retval = ksm_fork(mm, oldmm);
502 if (retval)
503 goto out;
504 retval = khugepaged_fork(mm, oldmm);
505 if (retval)
506 goto out;
507
508 prev = NULL;
509 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
510 struct file *file;
511
512 if (mpnt->vm_flags & VM_DONTCOPY) {
513 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
514 continue;
515 }
516 charge = 0;
517 /*
518 * Don't duplicate many vmas if we've been oom-killed (for
519 * example)
520 */
521 if (fatal_signal_pending(current)) {
522 retval = -EINTR;
523 goto out;
524 }
525 if (mpnt->vm_flags & VM_ACCOUNT) {
526 unsigned long len = vma_pages(mpnt);
527
528 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
529 goto fail_nomem;
530 charge = len;
531 }
532 tmp = vm_area_dup(mpnt);
533 if (!tmp)
534 goto fail_nomem;
535 retval = vma_dup_policy(mpnt, tmp);
536 if (retval)
537 goto fail_nomem_policy;
538 tmp->vm_mm = mm;
539 retval = dup_userfaultfd(tmp, &uf);
540 if (retval)
541 goto fail_nomem_anon_vma_fork;
542 if (tmp->vm_flags & VM_WIPEONFORK) {
543 /* VM_WIPEONFORK gets a clean slate in the child. */
544 tmp->anon_vma = NULL;
545 if (anon_vma_prepare(tmp))
546 goto fail_nomem_anon_vma_fork;
547 } else if (anon_vma_fork(tmp, mpnt))
548 goto fail_nomem_anon_vma_fork;
549 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
550 tmp->vm_next = tmp->vm_prev = NULL;
551 file = tmp->vm_file;
552 if (file) {
553 struct inode *inode = file_inode(file);
554 struct address_space *mapping = file->f_mapping;
555
556 get_file(file);
557 if (tmp->vm_flags & VM_DENYWRITE)
558 atomic_dec(&inode->i_writecount);
559 i_mmap_lock_write(mapping);
560 if (tmp->vm_flags & VM_SHARED)
561 atomic_inc(&mapping->i_mmap_writable);
562 flush_dcache_mmap_lock(mapping);
563 /* insert tmp into the share list, just after mpnt */
564 vma_interval_tree_insert_after(tmp, mpnt,
565 &mapping->i_mmap);
566 flush_dcache_mmap_unlock(mapping);
567 i_mmap_unlock_write(mapping);
568 }
569
570 /*
571 * Clear hugetlb-related page reserves for children. This only
572 * affects MAP_PRIVATE mappings. Faults generated by the child
573 * are not guaranteed to succeed, even if read-only
574 */
575 if (is_vm_hugetlb_page(tmp))
576 reset_vma_resv_huge_pages(tmp);
577
578 /*
579 * Link in the new vma and copy the page table entries.
580 */
581 *pprev = tmp;
582 pprev = &tmp->vm_next;
583 tmp->vm_prev = prev;
584 prev = tmp;
585
586 __vma_link_rb(mm, tmp, rb_link, rb_parent);
587 rb_link = &tmp->vm_rb.rb_right;
588 rb_parent = &tmp->vm_rb;
589
590 mm->map_count++;
591 if (!(tmp->vm_flags & VM_WIPEONFORK))
592 retval = copy_page_range(mm, oldmm, mpnt);
593
594 if (tmp->vm_ops && tmp->vm_ops->open)
595 tmp->vm_ops->open(tmp);
596
597 if (retval)
598 goto out;
599 }
600 /* a new mm has just been created */
601 retval = arch_dup_mmap(oldmm, mm);
602out:
603 up_write(&mm->mmap_sem);
604 flush_tlb_mm(oldmm);
605 up_write(&oldmm->mmap_sem);
606 dup_userfaultfd_complete(&uf);
607fail_uprobe_end:
608 uprobe_end_dup_mmap();
609 return retval;
610fail_nomem_anon_vma_fork:
611 mpol_put(vma_policy(tmp));
612fail_nomem_policy:
613 vm_area_free(tmp);
614fail_nomem:
615 retval = -ENOMEM;
616 vm_unacct_memory(charge);
617 goto out;
618}
619
620static inline int mm_alloc_pgd(struct mm_struct *mm)
621{
622 mm->pgd = pgd_alloc(mm);
623 if (unlikely(!mm->pgd))
624 return -ENOMEM;
625 return 0;
626}
627
628static inline void mm_free_pgd(struct mm_struct *mm)
629{
630 pgd_free(mm, mm->pgd);
631}
632#else
633static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
634{
635 down_write(&oldmm->mmap_sem);
636 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
637 up_write(&oldmm->mmap_sem);
638 return 0;
639}
640#define mm_alloc_pgd(mm) (0)
641#define mm_free_pgd(mm)
642#endif /* CONFIG_MMU */
643
644static void check_mm(struct mm_struct *mm)
645{
646 int i;
647
648 for (i = 0; i < NR_MM_COUNTERS; i++) {
649 long x = atomic_long_read(&mm->rss_stat.count[i]);
650
651 if (unlikely(x))
652 printk(KERN_ALERT "BUG: Bad rss-counter state "
653 "mm:%p idx:%d val:%ld\n", mm, i, x);
654 }
655
656 if (mm_pgtables_bytes(mm))
657 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
658 mm_pgtables_bytes(mm));
659
660#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
661 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
662#endif
663}
664
665#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
666#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
667
668/*
669 * Called when the last reference to the mm
670 * is dropped: either by a lazy thread or by
671 * mmput. Free the page directory and the mm.
672 */
673void __mmdrop(struct mm_struct *mm)
674{
675 BUG_ON(mm == &init_mm);
676 WARN_ON_ONCE(mm == current->mm);
677 WARN_ON_ONCE(mm == current->active_mm);
678 mm_free_pgd(mm);
679 destroy_context(mm);
680 hmm_mm_destroy(mm);
681 mmu_notifier_mm_destroy(mm);
682 check_mm(mm);
683 put_user_ns(mm->user_ns);
684 free_mm(mm);
685}
686EXPORT_SYMBOL_GPL(__mmdrop);
687
688static void mmdrop_async_fn(struct work_struct *work)
689{
690 struct mm_struct *mm;
691
692 mm = container_of(work, struct mm_struct, async_put_work);
693 __mmdrop(mm);
694}
695
696static void mmdrop_async(struct mm_struct *mm)
697{
698 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
699 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
700 schedule_work(&mm->async_put_work);
701 }
702}
703
704static inline void free_signal_struct(struct signal_struct *sig)
705{
706 taskstats_tgid_free(sig);
707 sched_autogroup_exit(sig);
708 /*
709 * __mmdrop is not safe to call from softirq context on x86 due to
710 * pgd_dtor so postpone it to the async context
711 */
712 if (sig->oom_mm)
713 mmdrop_async(sig->oom_mm);
714 kmem_cache_free(signal_cachep, sig);
715}
716
717static inline void put_signal_struct(struct signal_struct *sig)
718{
719 if (refcount_dec_and_test(&sig->sigcnt))
720 free_signal_struct(sig);
721}
722
723void __put_task_struct(struct task_struct *tsk)
724{
725 WARN_ON(!tsk->exit_state);
726 WARN_ON(refcount_read(&tsk->usage));
727 WARN_ON(tsk == current);
728
729 cgroup_free(tsk);
730 task_numa_free(tsk);
731 security_task_free(tsk);
732 exit_creds(tsk);
733 delayacct_tsk_free(tsk);
734 put_signal_struct(tsk->signal);
735
736 if (!profile_handoff_task(tsk))
737 free_task(tsk);
738}
739EXPORT_SYMBOL_GPL(__put_task_struct);
740
741void __init __weak arch_task_cache_init(void) { }
742
743/*
744 * set_max_threads
745 */
746static void set_max_threads(unsigned int max_threads_suggested)
747{
748 u64 threads;
749 unsigned long nr_pages = totalram_pages();
750
751 /*
752 * The number of threads shall be limited such that the thread
753 * structures may only consume a small part of the available memory.
754 */
755 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
756 threads = MAX_THREADS;
757 else
758 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
759 (u64) THREAD_SIZE * 8UL);
760
761 if (threads > max_threads_suggested)
762 threads = max_threads_suggested;
763
764 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
765}
766
767#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
768/* Initialized by the architecture: */
769int arch_task_struct_size __read_mostly;
770#endif
771
772static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
773{
774 /* Fetch thread_struct whitelist for the architecture. */
775 arch_thread_struct_whitelist(offset, size);
776
777 /*
778 * Handle zero-sized whitelist or empty thread_struct, otherwise
779 * adjust offset to position of thread_struct in task_struct.
780 */
781 if (unlikely(*size == 0))
782 *offset = 0;
783 else
784 *offset += offsetof(struct task_struct, thread);
785}
786
787void __init fork_init(void)
788{
789 int i;
790#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
791#ifndef ARCH_MIN_TASKALIGN
792#define ARCH_MIN_TASKALIGN 0
793#endif
794 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
795 unsigned long useroffset, usersize;
796
797 /* create a slab on which task_structs can be allocated */
798 task_struct_whitelist(&useroffset, &usersize);
799 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
800 arch_task_struct_size, align,
801 SLAB_PANIC|SLAB_ACCOUNT,
802 useroffset, usersize, NULL);
803#endif
804
805 /* do the arch specific task caches init */
806 arch_task_cache_init();
807
808 set_max_threads(MAX_THREADS);
809
810 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
811 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
812 init_task.signal->rlim[RLIMIT_SIGPENDING] =
813 init_task.signal->rlim[RLIMIT_NPROC];
814
815 for (i = 0; i < UCOUNT_COUNTS; i++) {
816 init_user_ns.ucount_max[i] = max_threads/2;
817 }
818
819#ifdef CONFIG_VMAP_STACK
820 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
821 NULL, free_vm_stack_cache);
822#endif
823
824 lockdep_init_task(&init_task);
825 uprobes_init();
826}
827
828int __weak arch_dup_task_struct(struct task_struct *dst,
829 struct task_struct *src)
830{
831 *dst = *src;
832 return 0;
833}
834
835void set_task_stack_end_magic(struct task_struct *tsk)
836{
837 unsigned long *stackend;
838
839 stackend = end_of_stack(tsk);
840 *stackend = STACK_END_MAGIC; /* for overflow detection */
841}
842
843static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
844{
845 struct task_struct *tsk;
846 unsigned long *stack;
847 struct vm_struct *stack_vm_area __maybe_unused;
848 int err;
849
850 if (node == NUMA_NO_NODE)
851 node = tsk_fork_get_node(orig);
852 tsk = alloc_task_struct_node(node);
853 if (!tsk)
854 return NULL;
855
856 stack = alloc_thread_stack_node(tsk, node);
857 if (!stack)
858 goto free_tsk;
859
860 if (memcg_charge_kernel_stack(tsk))
861 goto free_stack;
862
863 stack_vm_area = task_stack_vm_area(tsk);
864
865 err = arch_dup_task_struct(tsk, orig);
866
867 /*
868 * arch_dup_task_struct() clobbers the stack-related fields. Make
869 * sure they're properly initialized before using any stack-related
870 * functions again.
871 */
872 tsk->stack = stack;
873#ifdef CONFIG_VMAP_STACK
874 tsk->stack_vm_area = stack_vm_area;
875#endif
876#ifdef CONFIG_THREAD_INFO_IN_TASK
877 refcount_set(&tsk->stack_refcount, 1);
878#endif
879
880 if (err)
881 goto free_stack;
882
883#ifdef CONFIG_SECCOMP
884 /*
885 * We must handle setting up seccomp filters once we're under
886 * the sighand lock in case orig has changed between now and
887 * then. Until then, filter must be NULL to avoid messing up
888 * the usage counts on the error path calling free_task.
889 */
890 tsk->seccomp.filter = NULL;
891#endif
892
893 setup_thread_stack(tsk, orig);
894 clear_user_return_notifier(tsk);
895 clear_tsk_need_resched(tsk);
896 set_task_stack_end_magic(tsk);
897
898#ifdef CONFIG_STACKPROTECTOR
899 tsk->stack_canary = get_random_canary();
900#endif
901
902 /*
903 * One for us, one for whoever does the "release_task()" (usually
904 * parent)
905 */
906 refcount_set(&tsk->usage, 2);
907#ifdef CONFIG_BLK_DEV_IO_TRACE
908 tsk->btrace_seq = 0;
909#endif
910 tsk->splice_pipe = NULL;
911 tsk->task_frag.page = NULL;
912 tsk->wake_q.next = NULL;
913
914 account_kernel_stack(tsk, 1);
915
916 kcov_task_init(tsk);
917
918#ifdef CONFIG_FAULT_INJECTION
919 tsk->fail_nth = 0;
920#endif
921
922#ifdef CONFIG_BLK_CGROUP
923 tsk->throttle_queue = NULL;
924 tsk->use_memdelay = 0;
925#endif
926
927#ifdef CONFIG_MEMCG
928 tsk->active_memcg = NULL;
929#endif
930 return tsk;
931
932free_stack:
933 free_thread_stack(tsk);
934free_tsk:
935 free_task_struct(tsk);
936 return NULL;
937}
938
939__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
940
941static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
942
943static int __init coredump_filter_setup(char *s)
944{
945 default_dump_filter =
946 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
947 MMF_DUMP_FILTER_MASK;
948 return 1;
949}
950
951__setup("coredump_filter=", coredump_filter_setup);
952
953#include <linux/init_task.h>
954
955static void mm_init_aio(struct mm_struct *mm)
956{
957#ifdef CONFIG_AIO
958 spin_lock_init(&mm->ioctx_lock);
959 mm->ioctx_table = NULL;
960#endif
961}
962
963static __always_inline void mm_clear_owner(struct mm_struct *mm,
964 struct task_struct *p)
965{
966#ifdef CONFIG_MEMCG
967 if (mm->owner == p)
968 WRITE_ONCE(mm->owner, NULL);
969#endif
970}
971
972static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
973{
974#ifdef CONFIG_MEMCG
975 mm->owner = p;
976#endif
977}
978
979static void mm_init_uprobes_state(struct mm_struct *mm)
980{
981#ifdef CONFIG_UPROBES
982 mm->uprobes_state.xol_area = NULL;
983#endif
984}
985
986static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
987 struct user_namespace *user_ns)
988{
989 mm->mmap = NULL;
990 mm->mm_rb = RB_ROOT;
991 mm->vmacache_seqnum = 0;
992 atomic_set(&mm->mm_users, 1);
993 atomic_set(&mm->mm_count, 1);
994 init_rwsem(&mm->mmap_sem);
995 INIT_LIST_HEAD(&mm->mmlist);
996 mm->core_state = NULL;
997 mm_pgtables_bytes_init(mm);
998 mm->map_count = 0;
999 mm->locked_vm = 0;
1000 atomic64_set(&mm->pinned_vm, 0);
1001 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1002 spin_lock_init(&mm->page_table_lock);
1003 spin_lock_init(&mm->arg_lock);
1004 mm_init_cpumask(mm);
1005 mm_init_aio(mm);
1006 mm_init_owner(mm, p);
1007 RCU_INIT_POINTER(mm->exe_file, NULL);
1008 mmu_notifier_mm_init(mm);
1009 hmm_mm_init(mm);
1010 init_tlb_flush_pending(mm);
1011#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1012 mm->pmd_huge_pte = NULL;
1013#endif
1014 mm_init_uprobes_state(mm);
1015
1016 if (current->mm) {
1017 mm->flags = current->mm->flags & MMF_INIT_MASK;
1018 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1019 } else {
1020 mm->flags = default_dump_filter;
1021 mm->def_flags = 0;
1022 }
1023
1024 if (mm_alloc_pgd(mm))
1025 goto fail_nopgd;
1026
1027 if (init_new_context(p, mm))
1028 goto fail_nocontext;
1029
1030 mm->user_ns = get_user_ns(user_ns);
1031 return mm;
1032
1033fail_nocontext:
1034 mm_free_pgd(mm);
1035fail_nopgd:
1036 free_mm(mm);
1037 return NULL;
1038}
1039
1040/*
1041 * Allocate and initialize an mm_struct.
1042 */
1043struct mm_struct *mm_alloc(void)
1044{
1045 struct mm_struct *mm;
1046
1047 mm = allocate_mm();
1048 if (!mm)
1049 return NULL;
1050
1051 memset(mm, 0, sizeof(*mm));
1052 return mm_init(mm, current, current_user_ns());
1053}
1054
1055static inline void __mmput(struct mm_struct *mm)
1056{
1057 VM_BUG_ON(atomic_read(&mm->mm_users));
1058
1059 uprobe_clear_state(mm);
1060 exit_aio(mm);
1061 ksm_exit(mm);
1062 khugepaged_exit(mm); /* must run before exit_mmap */
1063 exit_mmap(mm);
1064 mm_put_huge_zero_page(mm);
1065 set_mm_exe_file(mm, NULL);
1066 if (!list_empty(&mm->mmlist)) {
1067 spin_lock(&mmlist_lock);
1068 list_del(&mm->mmlist);
1069 spin_unlock(&mmlist_lock);
1070 }
1071 if (mm->binfmt)
1072 module_put(mm->binfmt->module);
1073 mmdrop(mm);
1074}
1075
1076/*
1077 * Decrement the use count and release all resources for an mm.
1078 */
1079void mmput(struct mm_struct *mm)
1080{
1081 might_sleep();
1082
1083 if (atomic_dec_and_test(&mm->mm_users))
1084 __mmput(mm);
1085}
1086EXPORT_SYMBOL_GPL(mmput);
1087
1088#ifdef CONFIG_MMU
1089static void mmput_async_fn(struct work_struct *work)
1090{
1091 struct mm_struct *mm = container_of(work, struct mm_struct,
1092 async_put_work);
1093
1094 __mmput(mm);
1095}
1096
1097void mmput_async(struct mm_struct *mm)
1098{
1099 if (atomic_dec_and_test(&mm->mm_users)) {
1100 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1101 schedule_work(&mm->async_put_work);
1102 }
1103}
1104#endif
1105
1106/**
1107 * set_mm_exe_file - change a reference to the mm's executable file
1108 *
1109 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1110 *
1111 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1112 * invocations: in mmput() nobody alive left, in execve task is single
1113 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1114 * mm->exe_file, but does so without using set_mm_exe_file() in order
1115 * to do avoid the need for any locks.
1116 */
1117void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1118{
1119 struct file *old_exe_file;
1120
1121 /*
1122 * It is safe to dereference the exe_file without RCU as
1123 * this function is only called if nobody else can access
1124 * this mm -- see comment above for justification.
1125 */
1126 old_exe_file = rcu_dereference_raw(mm->exe_file);
1127
1128 if (new_exe_file)
1129 get_file(new_exe_file);
1130 rcu_assign_pointer(mm->exe_file, new_exe_file);
1131 if (old_exe_file)
1132 fput(old_exe_file);
1133}
1134
1135/**
1136 * get_mm_exe_file - acquire a reference to the mm's executable file
1137 *
1138 * Returns %NULL if mm has no associated executable file.
1139 * User must release file via fput().
1140 */
1141struct file *get_mm_exe_file(struct mm_struct *mm)
1142{
1143 struct file *exe_file;
1144
1145 rcu_read_lock();
1146 exe_file = rcu_dereference(mm->exe_file);
1147 if (exe_file && !get_file_rcu(exe_file))
1148 exe_file = NULL;
1149 rcu_read_unlock();
1150 return exe_file;
1151}
1152EXPORT_SYMBOL(get_mm_exe_file);
1153
1154/**
1155 * get_task_exe_file - acquire a reference to the task's executable file
1156 *
1157 * Returns %NULL if task's mm (if any) has no associated executable file or
1158 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1159 * User must release file via fput().
1160 */
1161struct file *get_task_exe_file(struct task_struct *task)
1162{
1163 struct file *exe_file = NULL;
1164 struct mm_struct *mm;
1165
1166 task_lock(task);
1167 mm = task->mm;
1168 if (mm) {
1169 if (!(task->flags & PF_KTHREAD))
1170 exe_file = get_mm_exe_file(mm);
1171 }
1172 task_unlock(task);
1173 return exe_file;
1174}
1175EXPORT_SYMBOL(get_task_exe_file);
1176
1177/**
1178 * get_task_mm - acquire a reference to the task's mm
1179 *
1180 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1181 * this kernel workthread has transiently adopted a user mm with use_mm,
1182 * to do its AIO) is not set and if so returns a reference to it, after
1183 * bumping up the use count. User must release the mm via mmput()
1184 * after use. Typically used by /proc and ptrace.
1185 */
1186struct mm_struct *get_task_mm(struct task_struct *task)
1187{
1188 struct mm_struct *mm;
1189
1190 task_lock(task);
1191 mm = task->mm;
1192 if (mm) {
1193 if (task->flags & PF_KTHREAD)
1194 mm = NULL;
1195 else
1196 mmget(mm);
1197 }
1198 task_unlock(task);
1199 return mm;
1200}
1201EXPORT_SYMBOL_GPL(get_task_mm);
1202
1203struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1204{
1205 struct mm_struct *mm;
1206 int err;
1207
1208 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1209 if (err)
1210 return ERR_PTR(err);
1211
1212 mm = get_task_mm(task);
1213 if (mm && mm != current->mm &&
1214 !ptrace_may_access(task, mode)) {
1215 mmput(mm);
1216 mm = ERR_PTR(-EACCES);
1217 }
1218 mutex_unlock(&task->signal->cred_guard_mutex);
1219
1220 return mm;
1221}
1222
1223static void complete_vfork_done(struct task_struct *tsk)
1224{
1225 struct completion *vfork;
1226
1227 task_lock(tsk);
1228 vfork = tsk->vfork_done;
1229 if (likely(vfork)) {
1230 tsk->vfork_done = NULL;
1231 complete(vfork);
1232 }
1233 task_unlock(tsk);
1234}
1235
1236static int wait_for_vfork_done(struct task_struct *child,
1237 struct completion *vfork)
1238{
1239 int killed;
1240
1241 freezer_do_not_count();
1242 cgroup_enter_frozen();
1243 killed = wait_for_completion_killable(vfork);
1244 cgroup_leave_frozen(false);
1245 freezer_count();
1246
1247 if (killed) {
1248 task_lock(child);
1249 child->vfork_done = NULL;
1250 task_unlock(child);
1251 }
1252
1253 put_task_struct(child);
1254 return killed;
1255}
1256
1257/* Please note the differences between mmput and mm_release.
1258 * mmput is called whenever we stop holding onto a mm_struct,
1259 * error success whatever.
1260 *
1261 * mm_release is called after a mm_struct has been removed
1262 * from the current process.
1263 *
1264 * This difference is important for error handling, when we
1265 * only half set up a mm_struct for a new process and need to restore
1266 * the old one. Because we mmput the new mm_struct before
1267 * restoring the old one. . .
1268 * Eric Biederman 10 January 1998
1269 */
1270void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1271{
1272 /* Get rid of any futexes when releasing the mm */
1273#ifdef CONFIG_FUTEX
1274 if (unlikely(tsk->robust_list)) {
1275 exit_robust_list(tsk);
1276 tsk->robust_list = NULL;
1277 }
1278#ifdef CONFIG_COMPAT
1279 if (unlikely(tsk->compat_robust_list)) {
1280 compat_exit_robust_list(tsk);
1281 tsk->compat_robust_list = NULL;
1282 }
1283#endif
1284 if (unlikely(!list_empty(&tsk->pi_state_list)))
1285 exit_pi_state_list(tsk);
1286#endif
1287
1288 uprobe_free_utask(tsk);
1289
1290 /* Get rid of any cached register state */
1291 deactivate_mm(tsk, mm);
1292
1293 /*
1294 * Signal userspace if we're not exiting with a core dump
1295 * because we want to leave the value intact for debugging
1296 * purposes.
1297 */
1298 if (tsk->clear_child_tid) {
1299 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1300 atomic_read(&mm->mm_users) > 1) {
1301 /*
1302 * We don't check the error code - if userspace has
1303 * not set up a proper pointer then tough luck.
1304 */
1305 put_user(0, tsk->clear_child_tid);
1306 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1307 1, NULL, NULL, 0, 0);
1308 }
1309 tsk->clear_child_tid = NULL;
1310 }
1311
1312 /*
1313 * All done, finally we can wake up parent and return this mm to him.
1314 * Also kthread_stop() uses this completion for synchronization.
1315 */
1316 if (tsk->vfork_done)
1317 complete_vfork_done(tsk);
1318}
1319
1320/**
1321 * dup_mm() - duplicates an existing mm structure
1322 * @tsk: the task_struct with which the new mm will be associated.
1323 * @oldmm: the mm to duplicate.
1324 *
1325 * Allocates a new mm structure and duplicates the provided @oldmm structure
1326 * content into it.
1327 *
1328 * Return: the duplicated mm or NULL on failure.
1329 */
1330static struct mm_struct *dup_mm(struct task_struct *tsk,
1331 struct mm_struct *oldmm)
1332{
1333 struct mm_struct *mm;
1334 int err;
1335
1336 mm = allocate_mm();
1337 if (!mm)
1338 goto fail_nomem;
1339
1340 memcpy(mm, oldmm, sizeof(*mm));
1341
1342 if (!mm_init(mm, tsk, mm->user_ns))
1343 goto fail_nomem;
1344
1345 err = dup_mmap(mm, oldmm);
1346 if (err)
1347 goto free_pt;
1348
1349 mm->hiwater_rss = get_mm_rss(mm);
1350 mm->hiwater_vm = mm->total_vm;
1351
1352 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1353 goto free_pt;
1354
1355 return mm;
1356
1357free_pt:
1358 /* don't put binfmt in mmput, we haven't got module yet */
1359 mm->binfmt = NULL;
1360 mm_init_owner(mm, NULL);
1361 mmput(mm);
1362
1363fail_nomem:
1364 return NULL;
1365}
1366
1367static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1368{
1369 struct mm_struct *mm, *oldmm;
1370 int retval;
1371
1372 tsk->min_flt = tsk->maj_flt = 0;
1373 tsk->nvcsw = tsk->nivcsw = 0;
1374#ifdef CONFIG_DETECT_HUNG_TASK
1375 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1376 tsk->last_switch_time = 0;
1377#endif
1378
1379 tsk->mm = NULL;
1380 tsk->active_mm = NULL;
1381
1382 /*
1383 * Are we cloning a kernel thread?
1384 *
1385 * We need to steal a active VM for that..
1386 */
1387 oldmm = current->mm;
1388 if (!oldmm)
1389 return 0;
1390
1391 /* initialize the new vmacache entries */
1392 vmacache_flush(tsk);
1393
1394 if (clone_flags & CLONE_VM) {
1395 mmget(oldmm);
1396 mm = oldmm;
1397 goto good_mm;
1398 }
1399
1400 retval = -ENOMEM;
1401 mm = dup_mm(tsk, current->mm);
1402 if (!mm)
1403 goto fail_nomem;
1404
1405good_mm:
1406 tsk->mm = mm;
1407 tsk->active_mm = mm;
1408 return 0;
1409
1410fail_nomem:
1411 return retval;
1412}
1413
1414static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1415{
1416 struct fs_struct *fs = current->fs;
1417 if (clone_flags & CLONE_FS) {
1418 /* tsk->fs is already what we want */
1419 spin_lock(&fs->lock);
1420 if (fs->in_exec) {
1421 spin_unlock(&fs->lock);
1422 return -EAGAIN;
1423 }
1424 fs->users++;
1425 spin_unlock(&fs->lock);
1426 return 0;
1427 }
1428 tsk->fs = copy_fs_struct(fs);
1429 if (!tsk->fs)
1430 return -ENOMEM;
1431 return 0;
1432}
1433
1434static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1435{
1436 struct files_struct *oldf, *newf;
1437 int error = 0;
1438
1439 /*
1440 * A background process may not have any files ...
1441 */
1442 oldf = current->files;
1443 if (!oldf)
1444 goto out;
1445
1446 if (clone_flags & CLONE_FILES) {
1447 atomic_inc(&oldf->count);
1448 goto out;
1449 }
1450
1451 newf = dup_fd(oldf, &error);
1452 if (!newf)
1453 goto out;
1454
1455 tsk->files = newf;
1456 error = 0;
1457out:
1458 return error;
1459}
1460
1461static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1462{
1463#ifdef CONFIG_BLOCK
1464 struct io_context *ioc = current->io_context;
1465 struct io_context *new_ioc;
1466
1467 if (!ioc)
1468 return 0;
1469 /*
1470 * Share io context with parent, if CLONE_IO is set
1471 */
1472 if (clone_flags & CLONE_IO) {
1473 ioc_task_link(ioc);
1474 tsk->io_context = ioc;
1475 } else if (ioprio_valid(ioc->ioprio)) {
1476 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1477 if (unlikely(!new_ioc))
1478 return -ENOMEM;
1479
1480 new_ioc->ioprio = ioc->ioprio;
1481 put_io_context(new_ioc);
1482 }
1483#endif
1484 return 0;
1485}
1486
1487static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1488{
1489 struct sighand_struct *sig;
1490
1491 if (clone_flags & CLONE_SIGHAND) {
1492 refcount_inc(&current->sighand->count);
1493 return 0;
1494 }
1495 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1496 rcu_assign_pointer(tsk->sighand, sig);
1497 if (!sig)
1498 return -ENOMEM;
1499
1500 refcount_set(&sig->count, 1);
1501 spin_lock_irq(&current->sighand->siglock);
1502 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1503 spin_unlock_irq(&current->sighand->siglock);
1504 return 0;
1505}
1506
1507void __cleanup_sighand(struct sighand_struct *sighand)
1508{
1509 if (refcount_dec_and_test(&sighand->count)) {
1510 signalfd_cleanup(sighand);
1511 /*
1512 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1513 * without an RCU grace period, see __lock_task_sighand().
1514 */
1515 kmem_cache_free(sighand_cachep, sighand);
1516 }
1517}
1518
1519#ifdef CONFIG_POSIX_TIMERS
1520/*
1521 * Initialize POSIX timer handling for a thread group.
1522 */
1523static void posix_cpu_timers_init_group(struct signal_struct *sig)
1524{
1525 unsigned long cpu_limit;
1526
1527 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1528 if (cpu_limit != RLIM_INFINITY) {
1529 sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1530 sig->cputimer.running = true;
1531 }
1532
1533 /* The timer lists. */
1534 INIT_LIST_HEAD(&sig->cpu_timers[0]);
1535 INIT_LIST_HEAD(&sig->cpu_timers[1]);
1536 INIT_LIST_HEAD(&sig->cpu_timers[2]);
1537}
1538#else
1539static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1540#endif
1541
1542static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1543{
1544 struct signal_struct *sig;
1545
1546 if (clone_flags & CLONE_THREAD)
1547 return 0;
1548
1549 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1550 tsk->signal = sig;
1551 if (!sig)
1552 return -ENOMEM;
1553
1554 sig->nr_threads = 1;
1555 atomic_set(&sig->live, 1);
1556 refcount_set(&sig->sigcnt, 1);
1557
1558 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1559 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1560 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1561
1562 init_waitqueue_head(&sig->wait_chldexit);
1563 sig->curr_target = tsk;
1564 init_sigpending(&sig->shared_pending);
1565 INIT_HLIST_HEAD(&sig->multiprocess);
1566 seqlock_init(&sig->stats_lock);
1567 prev_cputime_init(&sig->prev_cputime);
1568
1569#ifdef CONFIG_POSIX_TIMERS
1570 INIT_LIST_HEAD(&sig->posix_timers);
1571 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1572 sig->real_timer.function = it_real_fn;
1573#endif
1574
1575 task_lock(current->group_leader);
1576 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1577 task_unlock(current->group_leader);
1578
1579 posix_cpu_timers_init_group(sig);
1580
1581 tty_audit_fork(sig);
1582 sched_autogroup_fork(sig);
1583
1584 sig->oom_score_adj = current->signal->oom_score_adj;
1585 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1586
1587 mutex_init(&sig->cred_guard_mutex);
1588
1589 return 0;
1590}
1591
1592static void copy_seccomp(struct task_struct *p)
1593{
1594#ifdef CONFIG_SECCOMP
1595 /*
1596 * Must be called with sighand->lock held, which is common to
1597 * all threads in the group. Holding cred_guard_mutex is not
1598 * needed because this new task is not yet running and cannot
1599 * be racing exec.
1600 */
1601 assert_spin_locked(&current->sighand->siglock);
1602
1603 /* Ref-count the new filter user, and assign it. */
1604 get_seccomp_filter(current);
1605 p->seccomp = current->seccomp;
1606
1607 /*
1608 * Explicitly enable no_new_privs here in case it got set
1609 * between the task_struct being duplicated and holding the
1610 * sighand lock. The seccomp state and nnp must be in sync.
1611 */
1612 if (task_no_new_privs(current))
1613 task_set_no_new_privs(p);
1614
1615 /*
1616 * If the parent gained a seccomp mode after copying thread
1617 * flags and between before we held the sighand lock, we have
1618 * to manually enable the seccomp thread flag here.
1619 */
1620 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1621 set_tsk_thread_flag(p, TIF_SECCOMP);
1622#endif
1623}
1624
1625SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1626{
1627 current->clear_child_tid = tidptr;
1628
1629 return task_pid_vnr(current);
1630}
1631
1632static void rt_mutex_init_task(struct task_struct *p)
1633{
1634 raw_spin_lock_init(&p->pi_lock);
1635#ifdef CONFIG_RT_MUTEXES
1636 p->pi_waiters = RB_ROOT_CACHED;
1637 p->pi_top_task = NULL;
1638 p->pi_blocked_on = NULL;
1639#endif
1640}
1641
1642#ifdef CONFIG_POSIX_TIMERS
1643/*
1644 * Initialize POSIX timer handling for a single task.
1645 */
1646static void posix_cpu_timers_init(struct task_struct *tsk)
1647{
1648 tsk->cputime_expires.prof_exp = 0;
1649 tsk->cputime_expires.virt_exp = 0;
1650 tsk->cputime_expires.sched_exp = 0;
1651 INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1652 INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1653 INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1654}
1655#else
1656static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1657#endif
1658
1659static inline void init_task_pid_links(struct task_struct *task)
1660{
1661 enum pid_type type;
1662
1663 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1664 INIT_HLIST_NODE(&task->pid_links[type]);
1665 }
1666}
1667
1668static inline void
1669init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1670{
1671 if (type == PIDTYPE_PID)
1672 task->thread_pid = pid;
1673 else
1674 task->signal->pids[type] = pid;
1675}
1676
1677static inline void rcu_copy_process(struct task_struct *p)
1678{
1679#ifdef CONFIG_PREEMPT_RCU
1680 p->rcu_read_lock_nesting = 0;
1681 p->rcu_read_unlock_special.s = 0;
1682 p->rcu_blocked_node = NULL;
1683 INIT_LIST_HEAD(&p->rcu_node_entry);
1684#endif /* #ifdef CONFIG_PREEMPT_RCU */
1685#ifdef CONFIG_TASKS_RCU
1686 p->rcu_tasks_holdout = false;
1687 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1688 p->rcu_tasks_idle_cpu = -1;
1689#endif /* #ifdef CONFIG_TASKS_RCU */
1690}
1691
1692static int pidfd_release(struct inode *inode, struct file *file)
1693{
1694 struct pid *pid = file->private_data;
1695
1696 file->private_data = NULL;
1697 put_pid(pid);
1698 return 0;
1699}
1700
1701#ifdef CONFIG_PROC_FS
1702static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1703{
1704 struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1705 struct pid *pid = f->private_data;
1706
1707 seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1708 seq_putc(m, '\n');
1709}
1710#endif
1711
1712const struct file_operations pidfd_fops = {
1713 .release = pidfd_release,
1714#ifdef CONFIG_PROC_FS
1715 .show_fdinfo = pidfd_show_fdinfo,
1716#endif
1717};
1718
1719static void __delayed_free_task(struct rcu_head *rhp)
1720{
1721 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1722
1723 free_task(tsk);
1724}
1725
1726static __always_inline void delayed_free_task(struct task_struct *tsk)
1727{
1728 if (IS_ENABLED(CONFIG_MEMCG))
1729 call_rcu(&tsk->rcu, __delayed_free_task);
1730 else
1731 free_task(tsk);
1732}
1733
1734/*
1735 * This creates a new process as a copy of the old one,
1736 * but does not actually start it yet.
1737 *
1738 * It copies the registers, and all the appropriate
1739 * parts of the process environment (as per the clone
1740 * flags). The actual kick-off is left to the caller.
1741 */
1742static __latent_entropy struct task_struct *copy_process(
1743 unsigned long clone_flags,
1744 unsigned long stack_start,
1745 unsigned long stack_size,
1746 int __user *parent_tidptr,
1747 int __user *child_tidptr,
1748 struct pid *pid,
1749 int trace,
1750 unsigned long tls,
1751 int node)
1752{
1753 int pidfd = -1, retval;
1754 struct task_struct *p;
1755 struct multiprocess_signals delayed;
1756 struct file *pidfile = NULL;
1757
1758 /*
1759 * Don't allow sharing the root directory with processes in a different
1760 * namespace
1761 */
1762 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1763 return ERR_PTR(-EINVAL);
1764
1765 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1766 return ERR_PTR(-EINVAL);
1767
1768 /*
1769 * Thread groups must share signals as well, and detached threads
1770 * can only be started up within the thread group.
1771 */
1772 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1773 return ERR_PTR(-EINVAL);
1774
1775 /*
1776 * Shared signal handlers imply shared VM. By way of the above,
1777 * thread groups also imply shared VM. Blocking this case allows
1778 * for various simplifications in other code.
1779 */
1780 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1781 return ERR_PTR(-EINVAL);
1782
1783 /*
1784 * Siblings of global init remain as zombies on exit since they are
1785 * not reaped by their parent (swapper). To solve this and to avoid
1786 * multi-rooted process trees, prevent global and container-inits
1787 * from creating siblings.
1788 */
1789 if ((clone_flags & CLONE_PARENT) &&
1790 current->signal->flags & SIGNAL_UNKILLABLE)
1791 return ERR_PTR(-EINVAL);
1792
1793 /*
1794 * If the new process will be in a different pid or user namespace
1795 * do not allow it to share a thread group with the forking task.
1796 */
1797 if (clone_flags & CLONE_THREAD) {
1798 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1799 (task_active_pid_ns(current) !=
1800 current->nsproxy->pid_ns_for_children))
1801 return ERR_PTR(-EINVAL);
1802 }
1803
1804 if (clone_flags & CLONE_PIDFD) {
1805 /*
1806 * - CLONE_PARENT_SETTID is useless for pidfds and also
1807 * parent_tidptr is used to return pidfds.
1808 * - CLONE_DETACHED is blocked so that we can potentially
1809 * reuse it later for CLONE_PIDFD.
1810 * - CLONE_THREAD is blocked until someone really needs it.
1811 */
1812 if (clone_flags &
1813 (CLONE_DETACHED | CLONE_PARENT_SETTID | CLONE_THREAD))
1814 return ERR_PTR(-EINVAL);
1815 }
1816
1817 /*
1818 * Force any signals received before this point to be delivered
1819 * before the fork happens. Collect up signals sent to multiple
1820 * processes that happen during the fork and delay them so that
1821 * they appear to happen after the fork.
1822 */
1823 sigemptyset(&delayed.signal);
1824 INIT_HLIST_NODE(&delayed.node);
1825
1826 spin_lock_irq(&current->sighand->siglock);
1827 if (!(clone_flags & CLONE_THREAD))
1828 hlist_add_head(&delayed.node, &current->signal->multiprocess);
1829 recalc_sigpending();
1830 spin_unlock_irq(&current->sighand->siglock);
1831 retval = -ERESTARTNOINTR;
1832 if (signal_pending(current))
1833 goto fork_out;
1834
1835 retval = -ENOMEM;
1836 p = dup_task_struct(current, node);
1837 if (!p)
1838 goto fork_out;
1839
1840 /*
1841 * This _must_ happen before we call free_task(), i.e. before we jump
1842 * to any of the bad_fork_* labels. This is to avoid freeing
1843 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1844 * kernel threads (PF_KTHREAD).
1845 */
1846 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1847 /*
1848 * Clear TID on mm_release()?
1849 */
1850 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1851
1852 ftrace_graph_init_task(p);
1853
1854 rt_mutex_init_task(p);
1855
1856#ifdef CONFIG_PROVE_LOCKING
1857 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1858 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1859#endif
1860 retval = -EAGAIN;
1861 if (atomic_read(&p->real_cred->user->processes) >=
1862 task_rlimit(p, RLIMIT_NPROC)) {
1863 if (p->real_cred->user != INIT_USER &&
1864 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1865 goto bad_fork_free;
1866 }
1867 current->flags &= ~PF_NPROC_EXCEEDED;
1868
1869 retval = copy_creds(p, clone_flags);
1870 if (retval < 0)
1871 goto bad_fork_free;
1872
1873 /*
1874 * If multiple threads are within copy_process(), then this check
1875 * triggers too late. This doesn't hurt, the check is only there
1876 * to stop root fork bombs.
1877 */
1878 retval = -EAGAIN;
1879 if (nr_threads >= max_threads)
1880 goto bad_fork_cleanup_count;
1881
1882 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1883 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1884 p->flags |= PF_FORKNOEXEC;
1885 INIT_LIST_HEAD(&p->children);
1886 INIT_LIST_HEAD(&p->sibling);
1887 rcu_copy_process(p);
1888 p->vfork_done = NULL;
1889 spin_lock_init(&p->alloc_lock);
1890
1891 init_sigpending(&p->pending);
1892
1893 p->utime = p->stime = p->gtime = 0;
1894#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1895 p->utimescaled = p->stimescaled = 0;
1896#endif
1897 prev_cputime_init(&p->prev_cputime);
1898
1899#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1900 seqcount_init(&p->vtime.seqcount);
1901 p->vtime.starttime = 0;
1902 p->vtime.state = VTIME_INACTIVE;
1903#endif
1904
1905#if defined(SPLIT_RSS_COUNTING)
1906 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1907#endif
1908
1909 p->default_timer_slack_ns = current->timer_slack_ns;
1910
1911#ifdef CONFIG_PSI
1912 p->psi_flags = 0;
1913#endif
1914
1915 task_io_accounting_init(&p->ioac);
1916 acct_clear_integrals(p);
1917
1918 posix_cpu_timers_init(p);
1919
1920 p->io_context = NULL;
1921 audit_set_context(p, NULL);
1922 cgroup_fork(p);
1923#ifdef CONFIG_NUMA
1924 p->mempolicy = mpol_dup(p->mempolicy);
1925 if (IS_ERR(p->mempolicy)) {
1926 retval = PTR_ERR(p->mempolicy);
1927 p->mempolicy = NULL;
1928 goto bad_fork_cleanup_threadgroup_lock;
1929 }
1930#endif
1931#ifdef CONFIG_CPUSETS
1932 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1933 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1934 seqcount_init(&p->mems_allowed_seq);
1935#endif
1936#ifdef CONFIG_TRACE_IRQFLAGS
1937 p->irq_events = 0;
1938 p->hardirqs_enabled = 0;
1939 p->hardirq_enable_ip = 0;
1940 p->hardirq_enable_event = 0;
1941 p->hardirq_disable_ip = _THIS_IP_;
1942 p->hardirq_disable_event = 0;
1943 p->softirqs_enabled = 1;
1944 p->softirq_enable_ip = _THIS_IP_;
1945 p->softirq_enable_event = 0;
1946 p->softirq_disable_ip = 0;
1947 p->softirq_disable_event = 0;
1948 p->hardirq_context = 0;
1949 p->softirq_context = 0;
1950#endif
1951
1952 p->pagefault_disabled = 0;
1953
1954#ifdef CONFIG_LOCKDEP
1955 p->lockdep_depth = 0; /* no locks held yet */
1956 p->curr_chain_key = 0;
1957 p->lockdep_recursion = 0;
1958 lockdep_init_task(p);
1959#endif
1960
1961#ifdef CONFIG_DEBUG_MUTEXES
1962 p->blocked_on = NULL; /* not blocked yet */
1963#endif
1964#ifdef CONFIG_BCACHE
1965 p->sequential_io = 0;
1966 p->sequential_io_avg = 0;
1967#endif
1968
1969 /* Perform scheduler related setup. Assign this task to a CPU. */
1970 retval = sched_fork(clone_flags, p);
1971 if (retval)
1972 goto bad_fork_cleanup_policy;
1973
1974 retval = perf_event_init_task(p);
1975 if (retval)
1976 goto bad_fork_cleanup_policy;
1977 retval = audit_alloc(p);
1978 if (retval)
1979 goto bad_fork_cleanup_perf;
1980 /* copy all the process information */
1981 shm_init_task(p);
1982 retval = security_task_alloc(p, clone_flags);
1983 if (retval)
1984 goto bad_fork_cleanup_audit;
1985 retval = copy_semundo(clone_flags, p);
1986 if (retval)
1987 goto bad_fork_cleanup_security;
1988 retval = copy_files(clone_flags, p);
1989 if (retval)
1990 goto bad_fork_cleanup_semundo;
1991 retval = copy_fs(clone_flags, p);
1992 if (retval)
1993 goto bad_fork_cleanup_files;
1994 retval = copy_sighand(clone_flags, p);
1995 if (retval)
1996 goto bad_fork_cleanup_fs;
1997 retval = copy_signal(clone_flags, p);
1998 if (retval)
1999 goto bad_fork_cleanup_sighand;
2000 retval = copy_mm(clone_flags, p);
2001 if (retval)
2002 goto bad_fork_cleanup_signal;
2003 retval = copy_namespaces(clone_flags, p);
2004 if (retval)
2005 goto bad_fork_cleanup_mm;
2006 retval = copy_io(clone_flags, p);
2007 if (retval)
2008 goto bad_fork_cleanup_namespaces;
2009 retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
2010 if (retval)
2011 goto bad_fork_cleanup_io;
2012
2013 stackleak_task_init(p);
2014
2015 if (pid != &init_struct_pid) {
2016 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2017 if (IS_ERR(pid)) {
2018 retval = PTR_ERR(pid);
2019 goto bad_fork_cleanup_thread;
2020 }
2021 }
2022
2023 /*
2024 * This has to happen after we've potentially unshared the file
2025 * descriptor table (so that the pidfd doesn't leak into the child
2026 * if the fd table isn't shared).
2027 */
2028 if (clone_flags & CLONE_PIDFD) {
2029 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2030 if (retval < 0)
2031 goto bad_fork_free_pid;
2032
2033 pidfd = retval;
2034
2035 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2036 O_RDWR | O_CLOEXEC);
2037 if (IS_ERR(pidfile)) {
2038 put_unused_fd(pidfd);
2039 retval = PTR_ERR(pidfile);
2040 goto bad_fork_free_pid;
2041 }
2042 get_pid(pid); /* held by pidfile now */
2043
2044 retval = put_user(pidfd, parent_tidptr);
2045 if (retval)
2046 goto bad_fork_put_pidfd;
2047 }
2048
2049#ifdef CONFIG_BLOCK
2050 p->plug = NULL;
2051#endif
2052#ifdef CONFIG_FUTEX
2053 p->robust_list = NULL;
2054#ifdef CONFIG_COMPAT
2055 p->compat_robust_list = NULL;
2056#endif
2057 INIT_LIST_HEAD(&p->pi_state_list);
2058 p->pi_state_cache = NULL;
2059#endif
2060 /*
2061 * sigaltstack should be cleared when sharing the same VM
2062 */
2063 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2064 sas_ss_reset(p);
2065
2066 /*
2067 * Syscall tracing and stepping should be turned off in the
2068 * child regardless of CLONE_PTRACE.
2069 */
2070 user_disable_single_step(p);
2071 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2072#ifdef TIF_SYSCALL_EMU
2073 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2074#endif
2075 clear_tsk_latency_tracing(p);
2076
2077 /* ok, now we should be set up.. */
2078 p->pid = pid_nr(pid);
2079 if (clone_flags & CLONE_THREAD) {
2080 p->exit_signal = -1;
2081 p->group_leader = current->group_leader;
2082 p->tgid = current->tgid;
2083 } else {
2084 if (clone_flags & CLONE_PARENT)
2085 p->exit_signal = current->group_leader->exit_signal;
2086 else
2087 p->exit_signal = (clone_flags & CSIGNAL);
2088 p->group_leader = p;
2089 p->tgid = p->pid;
2090 }
2091
2092 p->nr_dirtied = 0;
2093 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2094 p->dirty_paused_when = 0;
2095
2096 p->pdeath_signal = 0;
2097 INIT_LIST_HEAD(&p->thread_group);
2098 p->task_works = NULL;
2099
2100 cgroup_threadgroup_change_begin(current);
2101 /*
2102 * Ensure that the cgroup subsystem policies allow the new process to be
2103 * forked. It should be noted the the new process's css_set can be changed
2104 * between here and cgroup_post_fork() if an organisation operation is in
2105 * progress.
2106 */
2107 retval = cgroup_can_fork(p);
2108 if (retval)
2109 goto bad_fork_cgroup_threadgroup_change_end;
2110
2111 /*
2112 * From this point on we must avoid any synchronous user-space
2113 * communication until we take the tasklist-lock. In particular, we do
2114 * not want user-space to be able to predict the process start-time by
2115 * stalling fork(2) after we recorded the start_time but before it is
2116 * visible to the system.
2117 */
2118
2119 p->start_time = ktime_get_ns();
2120 p->real_start_time = ktime_get_boot_ns();
2121
2122 /*
2123 * Make it visible to the rest of the system, but dont wake it up yet.
2124 * Need tasklist lock for parent etc handling!
2125 */
2126 write_lock_irq(&tasklist_lock);
2127
2128 /* CLONE_PARENT re-uses the old parent */
2129 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2130 p->real_parent = current->real_parent;
2131 p->parent_exec_id = current->parent_exec_id;
2132 } else {
2133 p->real_parent = current;
2134 p->parent_exec_id = current->self_exec_id;
2135 }
2136
2137 klp_copy_process(p);
2138
2139 spin_lock(&current->sighand->siglock);
2140
2141 /*
2142 * Copy seccomp details explicitly here, in case they were changed
2143 * before holding sighand lock.
2144 */
2145 copy_seccomp(p);
2146
2147 rseq_fork(p, clone_flags);
2148
2149 /* Don't start children in a dying pid namespace */
2150 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2151 retval = -ENOMEM;
2152 goto bad_fork_cancel_cgroup;
2153 }
2154
2155 /* Let kill terminate clone/fork in the middle */
2156 if (fatal_signal_pending(current)) {
2157 retval = -EINTR;
2158 goto bad_fork_cancel_cgroup;
2159 }
2160
2161 /* past the last point of failure */
2162 if (pidfile)
2163 fd_install(pidfd, pidfile);
2164
2165 init_task_pid_links(p);
2166 if (likely(p->pid)) {
2167 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2168
2169 init_task_pid(p, PIDTYPE_PID, pid);
2170 if (thread_group_leader(p)) {
2171 init_task_pid(p, PIDTYPE_TGID, pid);
2172 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2173 init_task_pid(p, PIDTYPE_SID, task_session(current));
2174
2175 if (is_child_reaper(pid)) {
2176 ns_of_pid(pid)->child_reaper = p;
2177 p->signal->flags |= SIGNAL_UNKILLABLE;
2178 }
2179 p->signal->shared_pending.signal = delayed.signal;
2180 p->signal->tty = tty_kref_get(current->signal->tty);
2181 /*
2182 * Inherit has_child_subreaper flag under the same
2183 * tasklist_lock with adding child to the process tree
2184 * for propagate_has_child_subreaper optimization.
2185 */
2186 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2187 p->real_parent->signal->is_child_subreaper;
2188 list_add_tail(&p->sibling, &p->real_parent->children);
2189 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2190 attach_pid(p, PIDTYPE_TGID);
2191 attach_pid(p, PIDTYPE_PGID);
2192 attach_pid(p, PIDTYPE_SID);
2193 __this_cpu_inc(process_counts);
2194 } else {
2195 current->signal->nr_threads++;
2196 atomic_inc(&current->signal->live);
2197 refcount_inc(&current->signal->sigcnt);
2198 task_join_group_stop(p);
2199 list_add_tail_rcu(&p->thread_group,
2200 &p->group_leader->thread_group);
2201 list_add_tail_rcu(&p->thread_node,
2202 &p->signal->thread_head);
2203 }
2204 attach_pid(p, PIDTYPE_PID);
2205 nr_threads++;
2206 }
2207 total_forks++;
2208 hlist_del_init(&delayed.node);
2209 spin_unlock(&current->sighand->siglock);
2210 syscall_tracepoint_update(p);
2211 write_unlock_irq(&tasklist_lock);
2212
2213 proc_fork_connector(p);
2214 cgroup_post_fork(p);
2215 cgroup_threadgroup_change_end(current);
2216 perf_event_fork(p);
2217
2218 trace_task_newtask(p, clone_flags);
2219 uprobe_copy_process(p, clone_flags);
2220
2221 return p;
2222
2223bad_fork_cancel_cgroup:
2224 spin_unlock(&current->sighand->siglock);
2225 write_unlock_irq(&tasklist_lock);
2226 cgroup_cancel_fork(p);
2227bad_fork_cgroup_threadgroup_change_end:
2228 cgroup_threadgroup_change_end(current);
2229bad_fork_put_pidfd:
2230 if (clone_flags & CLONE_PIDFD) {
2231 fput(pidfile);
2232 put_unused_fd(pidfd);
2233 }
2234bad_fork_free_pid:
2235 if (pid != &init_struct_pid)
2236 free_pid(pid);
2237bad_fork_cleanup_thread:
2238 exit_thread(p);
2239bad_fork_cleanup_io:
2240 if (p->io_context)
2241 exit_io_context(p);
2242bad_fork_cleanup_namespaces:
2243 exit_task_namespaces(p);
2244bad_fork_cleanup_mm:
2245 if (p->mm) {
2246 mm_clear_owner(p->mm, p);
2247 mmput(p->mm);
2248 }
2249bad_fork_cleanup_signal:
2250 if (!(clone_flags & CLONE_THREAD))
2251 free_signal_struct(p->signal);
2252bad_fork_cleanup_sighand:
2253 __cleanup_sighand(p->sighand);
2254bad_fork_cleanup_fs:
2255 exit_fs(p); /* blocking */
2256bad_fork_cleanup_files:
2257 exit_files(p); /* blocking */
2258bad_fork_cleanup_semundo:
2259 exit_sem(p);
2260bad_fork_cleanup_security:
2261 security_task_free(p);
2262bad_fork_cleanup_audit:
2263 audit_free(p);
2264bad_fork_cleanup_perf:
2265 perf_event_free_task(p);
2266bad_fork_cleanup_policy:
2267 lockdep_free_task(p);
2268#ifdef CONFIG_NUMA
2269 mpol_put(p->mempolicy);
2270bad_fork_cleanup_threadgroup_lock:
2271#endif
2272 delayacct_tsk_free(p);
2273bad_fork_cleanup_count:
2274 atomic_dec(&p->cred->user->processes);
2275 exit_creds(p);
2276bad_fork_free:
2277 p->state = TASK_DEAD;
2278 put_task_stack(p);
2279 delayed_free_task(p);
2280fork_out:
2281 spin_lock_irq(&current->sighand->siglock);
2282 hlist_del_init(&delayed.node);
2283 spin_unlock_irq(&current->sighand->siglock);
2284 return ERR_PTR(retval);
2285}
2286
2287static inline void init_idle_pids(struct task_struct *idle)
2288{
2289 enum pid_type type;
2290
2291 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2292 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2293 init_task_pid(idle, type, &init_struct_pid);
2294 }
2295}
2296
2297struct task_struct *fork_idle(int cpu)
2298{
2299 struct task_struct *task;
2300 task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2301 cpu_to_node(cpu));
2302 if (!IS_ERR(task)) {
2303 init_idle_pids(task);
2304 init_idle(task, cpu);
2305 }
2306
2307 return task;
2308}
2309
2310struct mm_struct *copy_init_mm(void)
2311{
2312 return dup_mm(NULL, &init_mm);
2313}
2314
2315/*
2316 * Ok, this is the main fork-routine.
2317 *
2318 * It copies the process, and if successful kick-starts
2319 * it and waits for it to finish using the VM if required.
2320 */
2321long _do_fork(unsigned long clone_flags,
2322 unsigned long stack_start,
2323 unsigned long stack_size,
2324 int __user *parent_tidptr,
2325 int __user *child_tidptr,
2326 unsigned long tls)
2327{
2328 struct completion vfork;
2329 struct pid *pid;
2330 struct task_struct *p;
2331 int trace = 0;
2332 long nr;
2333
2334 /*
2335 * Determine whether and which event to report to ptracer. When
2336 * called from kernel_thread or CLONE_UNTRACED is explicitly
2337 * requested, no event is reported; otherwise, report if the event
2338 * for the type of forking is enabled.
2339 */
2340 if (!(clone_flags & CLONE_UNTRACED)) {
2341 if (clone_flags & CLONE_VFORK)
2342 trace = PTRACE_EVENT_VFORK;
2343 else if ((clone_flags & CSIGNAL) != SIGCHLD)
2344 trace = PTRACE_EVENT_CLONE;
2345 else
2346 trace = PTRACE_EVENT_FORK;
2347
2348 if (likely(!ptrace_event_enabled(current, trace)))
2349 trace = 0;
2350 }
2351
2352 p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2353 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2354 add_latent_entropy();
2355
2356 if (IS_ERR(p))
2357 return PTR_ERR(p);
2358
2359 /*
2360 * Do this prior waking up the new thread - the thread pointer
2361 * might get invalid after that point, if the thread exits quickly.
2362 */
2363 trace_sched_process_fork(current, p);
2364
2365 pid = get_task_pid(p, PIDTYPE_PID);
2366 nr = pid_vnr(pid);
2367
2368 if (clone_flags & CLONE_PARENT_SETTID)
2369 put_user(nr, parent_tidptr);
2370
2371 if (clone_flags & CLONE_VFORK) {
2372 p->vfork_done = &vfork;
2373 init_completion(&vfork);
2374 get_task_struct(p);
2375 }
2376
2377 wake_up_new_task(p);
2378
2379 /* forking complete and child started to run, tell ptracer */
2380 if (unlikely(trace))
2381 ptrace_event_pid(trace, pid);
2382
2383 if (clone_flags & CLONE_VFORK) {
2384 if (!wait_for_vfork_done(p, &vfork))
2385 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2386 }
2387
2388 put_pid(pid);
2389 return nr;
2390}
2391
2392#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2393/* For compatibility with architectures that call do_fork directly rather than
2394 * using the syscall entry points below. */
2395long do_fork(unsigned long clone_flags,
2396 unsigned long stack_start,
2397 unsigned long stack_size,
2398 int __user *parent_tidptr,
2399 int __user *child_tidptr)
2400{
2401 return _do_fork(clone_flags, stack_start, stack_size,
2402 parent_tidptr, child_tidptr, 0);
2403}
2404#endif
2405
2406/*
2407 * Create a kernel thread.
2408 */
2409pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2410{
2411 return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2412 (unsigned long)arg, NULL, NULL, 0);
2413}
2414
2415#ifdef __ARCH_WANT_SYS_FORK
2416SYSCALL_DEFINE0(fork)
2417{
2418#ifdef CONFIG_MMU
2419 return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2420#else
2421 /* can not support in nommu mode */
2422 return -EINVAL;
2423#endif
2424}
2425#endif
2426
2427#ifdef __ARCH_WANT_SYS_VFORK
2428SYSCALL_DEFINE0(vfork)
2429{
2430 return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2431 0, NULL, NULL, 0);
2432}
2433#endif
2434
2435#ifdef __ARCH_WANT_SYS_CLONE
2436#ifdef CONFIG_CLONE_BACKWARDS
2437SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2438 int __user *, parent_tidptr,
2439 unsigned long, tls,
2440 int __user *, child_tidptr)
2441#elif defined(CONFIG_CLONE_BACKWARDS2)
2442SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2443 int __user *, parent_tidptr,
2444 int __user *, child_tidptr,
2445 unsigned long, tls)
2446#elif defined(CONFIG_CLONE_BACKWARDS3)
2447SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2448 int, stack_size,
2449 int __user *, parent_tidptr,
2450 int __user *, child_tidptr,
2451 unsigned long, tls)
2452#else
2453SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2454 int __user *, parent_tidptr,
2455 int __user *, child_tidptr,
2456 unsigned long, tls)
2457#endif
2458{
2459 return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2460}
2461#endif
2462
2463void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2464{
2465 struct task_struct *leader, *parent, *child;
2466 int res;
2467
2468 read_lock(&tasklist_lock);
2469 leader = top = top->group_leader;
2470down:
2471 for_each_thread(leader, parent) {
2472 list_for_each_entry(child, &parent->children, sibling) {
2473 res = visitor(child, data);
2474 if (res) {
2475 if (res < 0)
2476 goto out;
2477 leader = child;
2478 goto down;
2479 }
2480up:
2481 ;
2482 }
2483 }
2484
2485 if (leader != top) {
2486 child = leader;
2487 parent = child->real_parent;
2488 leader = parent->group_leader;
2489 goto up;
2490 }
2491out:
2492 read_unlock(&tasklist_lock);
2493}
2494
2495#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2496#define ARCH_MIN_MMSTRUCT_ALIGN 0
2497#endif
2498
2499static void sighand_ctor(void *data)
2500{
2501 struct sighand_struct *sighand = data;
2502
2503 spin_lock_init(&sighand->siglock);
2504 init_waitqueue_head(&sighand->signalfd_wqh);
2505}
2506
2507void __init proc_caches_init(void)
2508{
2509 unsigned int mm_size;
2510
2511 sighand_cachep = kmem_cache_create("sighand_cache",
2512 sizeof(struct sighand_struct), 0,
2513 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2514 SLAB_ACCOUNT, sighand_ctor);
2515 signal_cachep = kmem_cache_create("signal_cache",
2516 sizeof(struct signal_struct), 0,
2517 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2518 NULL);
2519 files_cachep = kmem_cache_create("files_cache",
2520 sizeof(struct files_struct), 0,
2521 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2522 NULL);
2523 fs_cachep = kmem_cache_create("fs_cache",
2524 sizeof(struct fs_struct), 0,
2525 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2526 NULL);
2527
2528 /*
2529 * The mm_cpumask is located at the end of mm_struct, and is
2530 * dynamically sized based on the maximum CPU number this system
2531 * can have, taking hotplug into account (nr_cpu_ids).
2532 */
2533 mm_size = sizeof(struct mm_struct) + cpumask_size();
2534
2535 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2536 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2537 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2538 offsetof(struct mm_struct, saved_auxv),
2539 sizeof_field(struct mm_struct, saved_auxv),
2540 NULL);
2541 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2542 mmap_init();
2543 nsproxy_cache_init();
2544}
2545
2546/*
2547 * Check constraints on flags passed to the unshare system call.
2548 */
2549static int check_unshare_flags(unsigned long unshare_flags)
2550{
2551 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2552 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2553 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2554 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2555 return -EINVAL;
2556 /*
2557 * Not implemented, but pretend it works if there is nothing
2558 * to unshare. Note that unsharing the address space or the
2559 * signal handlers also need to unshare the signal queues (aka
2560 * CLONE_THREAD).
2561 */
2562 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2563 if (!thread_group_empty(current))
2564 return -EINVAL;
2565 }
2566 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2567 if (refcount_read(&current->sighand->count) > 1)
2568 return -EINVAL;
2569 }
2570 if (unshare_flags & CLONE_VM) {
2571 if (!current_is_single_threaded())
2572 return -EINVAL;
2573 }
2574
2575 return 0;
2576}
2577
2578/*
2579 * Unshare the filesystem structure if it is being shared
2580 */
2581static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2582{
2583 struct fs_struct *fs = current->fs;
2584
2585 if (!(unshare_flags & CLONE_FS) || !fs)
2586 return 0;
2587
2588 /* don't need lock here; in the worst case we'll do useless copy */
2589 if (fs->users == 1)
2590 return 0;
2591
2592 *new_fsp = copy_fs_struct(fs);
2593 if (!*new_fsp)
2594 return -ENOMEM;
2595
2596 return 0;
2597}
2598
2599/*
2600 * Unshare file descriptor table if it is being shared
2601 */
2602static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2603{
2604 struct files_struct *fd = current->files;
2605 int error = 0;
2606
2607 if ((unshare_flags & CLONE_FILES) &&
2608 (fd && atomic_read(&fd->count) > 1)) {
2609 *new_fdp = dup_fd(fd, &error);
2610 if (!*new_fdp)
2611 return error;
2612 }
2613
2614 return 0;
2615}
2616
2617/*
2618 * unshare allows a process to 'unshare' part of the process
2619 * context which was originally shared using clone. copy_*
2620 * functions used by do_fork() cannot be used here directly
2621 * because they modify an inactive task_struct that is being
2622 * constructed. Here we are modifying the current, active,
2623 * task_struct.
2624 */
2625int ksys_unshare(unsigned long unshare_flags)
2626{
2627 struct fs_struct *fs, *new_fs = NULL;
2628 struct files_struct *fd, *new_fd = NULL;
2629 struct cred *new_cred = NULL;
2630 struct nsproxy *new_nsproxy = NULL;
2631 int do_sysvsem = 0;
2632 int err;
2633
2634 /*
2635 * If unsharing a user namespace must also unshare the thread group
2636 * and unshare the filesystem root and working directories.
2637 */
2638 if (unshare_flags & CLONE_NEWUSER)
2639 unshare_flags |= CLONE_THREAD | CLONE_FS;
2640 /*
2641 * If unsharing vm, must also unshare signal handlers.
2642 */
2643 if (unshare_flags & CLONE_VM)
2644 unshare_flags |= CLONE_SIGHAND;
2645 /*
2646 * If unsharing a signal handlers, must also unshare the signal queues.
2647 */
2648 if (unshare_flags & CLONE_SIGHAND)
2649 unshare_flags |= CLONE_THREAD;
2650 /*
2651 * If unsharing namespace, must also unshare filesystem information.
2652 */
2653 if (unshare_flags & CLONE_NEWNS)
2654 unshare_flags |= CLONE_FS;
2655
2656 err = check_unshare_flags(unshare_flags);
2657 if (err)
2658 goto bad_unshare_out;
2659 /*
2660 * CLONE_NEWIPC must also detach from the undolist: after switching
2661 * to a new ipc namespace, the semaphore arrays from the old
2662 * namespace are unreachable.
2663 */
2664 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2665 do_sysvsem = 1;
2666 err = unshare_fs(unshare_flags, &new_fs);
2667 if (err)
2668 goto bad_unshare_out;
2669 err = unshare_fd(unshare_flags, &new_fd);
2670 if (err)
2671 goto bad_unshare_cleanup_fs;
2672 err = unshare_userns(unshare_flags, &new_cred);
2673 if (err)
2674 goto bad_unshare_cleanup_fd;
2675 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2676 new_cred, new_fs);
2677 if (err)
2678 goto bad_unshare_cleanup_cred;
2679
2680 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2681 if (do_sysvsem) {
2682 /*
2683 * CLONE_SYSVSEM is equivalent to sys_exit().
2684 */
2685 exit_sem(current);
2686 }
2687 if (unshare_flags & CLONE_NEWIPC) {
2688 /* Orphan segments in old ns (see sem above). */
2689 exit_shm(current);
2690 shm_init_task(current);
2691 }
2692
2693 if (new_nsproxy)
2694 switch_task_namespaces(current, new_nsproxy);
2695
2696 task_lock(current);
2697
2698 if (new_fs) {
2699 fs = current->fs;
2700 spin_lock(&fs->lock);
2701 current->fs = new_fs;
2702 if (--fs->users)
2703 new_fs = NULL;
2704 else
2705 new_fs = fs;
2706 spin_unlock(&fs->lock);
2707 }
2708
2709 if (new_fd) {
2710 fd = current->files;
2711 current->files = new_fd;
2712 new_fd = fd;
2713 }
2714
2715 task_unlock(current);
2716
2717 if (new_cred) {
2718 /* Install the new user namespace */
2719 commit_creds(new_cred);
2720 new_cred = NULL;
2721 }
2722 }
2723
2724 perf_event_namespaces(current);
2725
2726bad_unshare_cleanup_cred:
2727 if (new_cred)
2728 put_cred(new_cred);
2729bad_unshare_cleanup_fd:
2730 if (new_fd)
2731 put_files_struct(new_fd);
2732
2733bad_unshare_cleanup_fs:
2734 if (new_fs)
2735 free_fs_struct(new_fs);
2736
2737bad_unshare_out:
2738 return err;
2739}
2740
2741SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2742{
2743 return ksys_unshare(unshare_flags);
2744}
2745
2746/*
2747 * Helper to unshare the files of the current task.
2748 * We don't want to expose copy_files internals to
2749 * the exec layer of the kernel.
2750 */
2751
2752int unshare_files(struct files_struct **displaced)
2753{
2754 struct task_struct *task = current;
2755 struct files_struct *copy = NULL;
2756 int error;
2757
2758 error = unshare_fd(CLONE_FILES, &copy);
2759 if (error || !copy) {
2760 *displaced = NULL;
2761 return error;
2762 }
2763 *displaced = task->files;
2764 task_lock(task);
2765 task->files = copy;
2766 task_unlock(task);
2767 return 0;
2768}
2769
2770int sysctl_max_threads(struct ctl_table *table, int write,
2771 void __user *buffer, size_t *lenp, loff_t *ppos)
2772{
2773 struct ctl_table t;
2774 int ret;
2775 int threads = max_threads;
2776 int min = MIN_THREADS;
2777 int max = MAX_THREADS;
2778
2779 t = *table;
2780 t.data = &threads;
2781 t.extra1 = &min;
2782 t.extra2 = &max;
2783
2784 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2785 if (ret || !write)
2786 return ret;
2787
2788 set_max_threads(threads);
2789
2790 return 0;
2791}