Loading...
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
128#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
129
130static const char * const resident_page_types[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
135};
136
137DEFINE_PER_CPU(unsigned long, process_counts) = 0;
138
139__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
140
141#ifdef CONFIG_PROVE_RCU
142int lockdep_tasklist_lock_is_held(void)
143{
144 return lockdep_is_held(&tasklist_lock);
145}
146EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
147#endif /* #ifdef CONFIG_PROVE_RCU */
148
149int nr_processes(void)
150{
151 int cpu;
152 int total = 0;
153
154 for_each_possible_cpu(cpu)
155 total += per_cpu(process_counts, cpu);
156
157 return total;
158}
159
160void __weak arch_release_task_struct(struct task_struct *tsk)
161{
162}
163
164#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165static struct kmem_cache *task_struct_cachep;
166
167static inline struct task_struct *alloc_task_struct_node(int node)
168{
169 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
170}
171
172static inline void free_task_struct(struct task_struct *tsk)
173{
174 kmem_cache_free(task_struct_cachep, tsk);
175}
176#endif
177
178#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
179
180/*
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
183 */
184# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
185
186#ifdef CONFIG_VMAP_STACK
187/*
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
190 */
191#define NR_CACHED_STACKS 2
192static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
193
194static int free_vm_stack_cache(unsigned int cpu)
195{
196 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
197 int i;
198
199 for (i = 0; i < NR_CACHED_STACKS; i++) {
200 struct vm_struct *vm_stack = cached_vm_stacks[i];
201
202 if (!vm_stack)
203 continue;
204
205 vfree(vm_stack->addr);
206 cached_vm_stacks[i] = NULL;
207 }
208
209 return 0;
210}
211#endif
212
213static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
214{
215#ifdef CONFIG_VMAP_STACK
216 void *stack;
217 int i;
218
219 for (i = 0; i < NR_CACHED_STACKS; i++) {
220 struct vm_struct *s;
221
222 s = this_cpu_xchg(cached_stacks[i], NULL);
223
224 if (!s)
225 continue;
226
227 /* Clear stale pointers from reused stack. */
228 memset(s->addr, 0, THREAD_SIZE);
229
230 tsk->stack_vm_area = s;
231 tsk->stack = s->addr;
232 return s->addr;
233 }
234
235 /*
236 * Allocated stacks are cached and later reused by new threads,
237 * so memcg accounting is performed manually on assigning/releasing
238 * stacks to tasks. Drop __GFP_ACCOUNT.
239 */
240 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
241 VMALLOC_START, VMALLOC_END,
242 THREADINFO_GFP & ~__GFP_ACCOUNT,
243 PAGE_KERNEL,
244 0, node, __builtin_return_address(0));
245
246 /*
247 * We can't call find_vm_area() in interrupt context, and
248 * free_thread_stack() can be called in interrupt context,
249 * so cache the vm_struct.
250 */
251 if (stack) {
252 tsk->stack_vm_area = find_vm_area(stack);
253 tsk->stack = stack;
254 }
255 return stack;
256#else
257 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
258 THREAD_SIZE_ORDER);
259
260 if (likely(page)) {
261 tsk->stack = page_address(page);
262 return tsk->stack;
263 }
264 return NULL;
265#endif
266}
267
268static inline void free_thread_stack(struct task_struct *tsk)
269{
270#ifdef CONFIG_VMAP_STACK
271 struct vm_struct *vm = task_stack_vm_area(tsk);
272
273 if (vm) {
274 int i;
275
276 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
277 mod_memcg_page_state(vm->pages[i],
278 MEMCG_KERNEL_STACK_KB,
279 -(int)(PAGE_SIZE / 1024));
280
281 memcg_kmem_uncharge(vm->pages[i], 0);
282 }
283
284 for (i = 0; i < NR_CACHED_STACKS; i++) {
285 if (this_cpu_cmpxchg(cached_stacks[i],
286 NULL, tsk->stack_vm_area) != NULL)
287 continue;
288
289 return;
290 }
291
292 vfree_atomic(tsk->stack);
293 return;
294 }
295#endif
296
297 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
298}
299# else
300static struct kmem_cache *thread_stack_cache;
301
302static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
303 int node)
304{
305 unsigned long *stack;
306 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
307 tsk->stack = stack;
308 return stack;
309}
310
311static void free_thread_stack(struct task_struct *tsk)
312{
313 kmem_cache_free(thread_stack_cache, tsk->stack);
314}
315
316void thread_stack_cache_init(void)
317{
318 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
319 THREAD_SIZE, THREAD_SIZE, 0, 0,
320 THREAD_SIZE, NULL);
321 BUG_ON(thread_stack_cache == NULL);
322}
323# endif
324#endif
325
326/* SLAB cache for signal_struct structures (tsk->signal) */
327static struct kmem_cache *signal_cachep;
328
329/* SLAB cache for sighand_struct structures (tsk->sighand) */
330struct kmem_cache *sighand_cachep;
331
332/* SLAB cache for files_struct structures (tsk->files) */
333struct kmem_cache *files_cachep;
334
335/* SLAB cache for fs_struct structures (tsk->fs) */
336struct kmem_cache *fs_cachep;
337
338/* SLAB cache for vm_area_struct structures */
339static struct kmem_cache *vm_area_cachep;
340
341/* SLAB cache for mm_struct structures (tsk->mm) */
342static struct kmem_cache *mm_cachep;
343
344struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
345{
346 struct vm_area_struct *vma;
347
348 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
349 if (vma)
350 vma_init(vma, mm);
351 return vma;
352}
353
354struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
355{
356 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357
358 if (new) {
359 *new = *orig;
360 INIT_LIST_HEAD(&new->anon_vma_chain);
361 }
362 return new;
363}
364
365void vm_area_free(struct vm_area_struct *vma)
366{
367 kmem_cache_free(vm_area_cachep, vma);
368}
369
370static void account_kernel_stack(struct task_struct *tsk, int account)
371{
372 void *stack = task_stack_page(tsk);
373 struct vm_struct *vm = task_stack_vm_area(tsk);
374
375 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
376
377 if (vm) {
378 int i;
379
380 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
381
382 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
383 mod_zone_page_state(page_zone(vm->pages[i]),
384 NR_KERNEL_STACK_KB,
385 PAGE_SIZE / 1024 * account);
386 }
387 } else {
388 /*
389 * All stack pages are in the same zone and belong to the
390 * same memcg.
391 */
392 struct page *first_page = virt_to_page(stack);
393
394 mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
395 THREAD_SIZE / 1024 * account);
396
397 mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
398 account * (THREAD_SIZE / 1024));
399 }
400}
401
402static int memcg_charge_kernel_stack(struct task_struct *tsk)
403{
404#ifdef CONFIG_VMAP_STACK
405 struct vm_struct *vm = task_stack_vm_area(tsk);
406 int ret;
407
408 if (vm) {
409 int i;
410
411 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
412 /*
413 * If memcg_kmem_charge() fails, page->mem_cgroup
414 * pointer is NULL, and both memcg_kmem_uncharge()
415 * and mod_memcg_page_state() in free_thread_stack()
416 * will ignore this page. So it's safe.
417 */
418 ret = memcg_kmem_charge(vm->pages[i], GFP_KERNEL, 0);
419 if (ret)
420 return ret;
421
422 mod_memcg_page_state(vm->pages[i],
423 MEMCG_KERNEL_STACK_KB,
424 PAGE_SIZE / 1024);
425 }
426 }
427#endif
428 return 0;
429}
430
431static void release_task_stack(struct task_struct *tsk)
432{
433 if (WARN_ON(tsk->state != TASK_DEAD))
434 return; /* Better to leak the stack than to free prematurely */
435
436 account_kernel_stack(tsk, -1);
437 free_thread_stack(tsk);
438 tsk->stack = NULL;
439#ifdef CONFIG_VMAP_STACK
440 tsk->stack_vm_area = NULL;
441#endif
442}
443
444#ifdef CONFIG_THREAD_INFO_IN_TASK
445void put_task_stack(struct task_struct *tsk)
446{
447 if (refcount_dec_and_test(&tsk->stack_refcount))
448 release_task_stack(tsk);
449}
450#endif
451
452void free_task(struct task_struct *tsk)
453{
454#ifndef CONFIG_THREAD_INFO_IN_TASK
455 /*
456 * The task is finally done with both the stack and thread_info,
457 * so free both.
458 */
459 release_task_stack(tsk);
460#else
461 /*
462 * If the task had a separate stack allocation, it should be gone
463 * by now.
464 */
465 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
466#endif
467 rt_mutex_debug_task_free(tsk);
468 ftrace_graph_exit_task(tsk);
469 put_seccomp_filter(tsk);
470 arch_release_task_struct(tsk);
471 if (tsk->flags & PF_KTHREAD)
472 free_kthread_struct(tsk);
473 free_task_struct(tsk);
474}
475EXPORT_SYMBOL(free_task);
476
477#ifdef CONFIG_MMU
478static __latent_entropy int dup_mmap(struct mm_struct *mm,
479 struct mm_struct *oldmm)
480{
481 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
482 struct rb_node **rb_link, *rb_parent;
483 int retval;
484 unsigned long charge;
485 LIST_HEAD(uf);
486
487 uprobe_start_dup_mmap();
488 if (down_write_killable(&oldmm->mmap_sem)) {
489 retval = -EINTR;
490 goto fail_uprobe_end;
491 }
492 flush_cache_dup_mm(oldmm);
493 uprobe_dup_mmap(oldmm, mm);
494 /*
495 * Not linked in yet - no deadlock potential:
496 */
497 down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
498
499 /* No ordering required: file already has been exposed. */
500 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
501
502 mm->total_vm = oldmm->total_vm;
503 mm->data_vm = oldmm->data_vm;
504 mm->exec_vm = oldmm->exec_vm;
505 mm->stack_vm = oldmm->stack_vm;
506
507 rb_link = &mm->mm_rb.rb_node;
508 rb_parent = NULL;
509 pprev = &mm->mmap;
510 retval = ksm_fork(mm, oldmm);
511 if (retval)
512 goto out;
513 retval = khugepaged_fork(mm, oldmm);
514 if (retval)
515 goto out;
516
517 prev = NULL;
518 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
519 struct file *file;
520
521 if (mpnt->vm_flags & VM_DONTCOPY) {
522 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
523 continue;
524 }
525 charge = 0;
526 /*
527 * Don't duplicate many vmas if we've been oom-killed (for
528 * example)
529 */
530 if (fatal_signal_pending(current)) {
531 retval = -EINTR;
532 goto out;
533 }
534 if (mpnt->vm_flags & VM_ACCOUNT) {
535 unsigned long len = vma_pages(mpnt);
536
537 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
538 goto fail_nomem;
539 charge = len;
540 }
541 tmp = vm_area_dup(mpnt);
542 if (!tmp)
543 goto fail_nomem;
544 retval = vma_dup_policy(mpnt, tmp);
545 if (retval)
546 goto fail_nomem_policy;
547 tmp->vm_mm = mm;
548 retval = dup_userfaultfd(tmp, &uf);
549 if (retval)
550 goto fail_nomem_anon_vma_fork;
551 if (tmp->vm_flags & VM_WIPEONFORK) {
552 /* VM_WIPEONFORK gets a clean slate in the child. */
553 tmp->anon_vma = NULL;
554 if (anon_vma_prepare(tmp))
555 goto fail_nomem_anon_vma_fork;
556 } else if (anon_vma_fork(tmp, mpnt))
557 goto fail_nomem_anon_vma_fork;
558 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
559 tmp->vm_next = tmp->vm_prev = NULL;
560 file = tmp->vm_file;
561 if (file) {
562 struct inode *inode = file_inode(file);
563 struct address_space *mapping = file->f_mapping;
564
565 get_file(file);
566 if (tmp->vm_flags & VM_DENYWRITE)
567 atomic_dec(&inode->i_writecount);
568 i_mmap_lock_write(mapping);
569 if (tmp->vm_flags & VM_SHARED)
570 atomic_inc(&mapping->i_mmap_writable);
571 flush_dcache_mmap_lock(mapping);
572 /* insert tmp into the share list, just after mpnt */
573 vma_interval_tree_insert_after(tmp, mpnt,
574 &mapping->i_mmap);
575 flush_dcache_mmap_unlock(mapping);
576 i_mmap_unlock_write(mapping);
577 }
578
579 /*
580 * Clear hugetlb-related page reserves for children. This only
581 * affects MAP_PRIVATE mappings. Faults generated by the child
582 * are not guaranteed to succeed, even if read-only
583 */
584 if (is_vm_hugetlb_page(tmp))
585 reset_vma_resv_huge_pages(tmp);
586
587 /*
588 * Link in the new vma and copy the page table entries.
589 */
590 *pprev = tmp;
591 pprev = &tmp->vm_next;
592 tmp->vm_prev = prev;
593 prev = tmp;
594
595 __vma_link_rb(mm, tmp, rb_link, rb_parent);
596 rb_link = &tmp->vm_rb.rb_right;
597 rb_parent = &tmp->vm_rb;
598
599 mm->map_count++;
600 if (!(tmp->vm_flags & VM_WIPEONFORK))
601 retval = copy_page_range(mm, oldmm, mpnt);
602
603 if (tmp->vm_ops && tmp->vm_ops->open)
604 tmp->vm_ops->open(tmp);
605
606 if (retval)
607 goto out;
608 }
609 /* a new mm has just been created */
610 retval = arch_dup_mmap(oldmm, mm);
611out:
612 up_write(&mm->mmap_sem);
613 flush_tlb_mm(oldmm);
614 up_write(&oldmm->mmap_sem);
615 dup_userfaultfd_complete(&uf);
616fail_uprobe_end:
617 uprobe_end_dup_mmap();
618 return retval;
619fail_nomem_anon_vma_fork:
620 mpol_put(vma_policy(tmp));
621fail_nomem_policy:
622 vm_area_free(tmp);
623fail_nomem:
624 retval = -ENOMEM;
625 vm_unacct_memory(charge);
626 goto out;
627}
628
629static inline int mm_alloc_pgd(struct mm_struct *mm)
630{
631 mm->pgd = pgd_alloc(mm);
632 if (unlikely(!mm->pgd))
633 return -ENOMEM;
634 return 0;
635}
636
637static inline void mm_free_pgd(struct mm_struct *mm)
638{
639 pgd_free(mm, mm->pgd);
640}
641#else
642static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
643{
644 down_write(&oldmm->mmap_sem);
645 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
646 up_write(&oldmm->mmap_sem);
647 return 0;
648}
649#define mm_alloc_pgd(mm) (0)
650#define mm_free_pgd(mm)
651#endif /* CONFIG_MMU */
652
653static void check_mm(struct mm_struct *mm)
654{
655 int i;
656
657 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
658 "Please make sure 'struct resident_page_types[]' is updated as well");
659
660 for (i = 0; i < NR_MM_COUNTERS; i++) {
661 long x = atomic_long_read(&mm->rss_stat.count[i]);
662
663 if (unlikely(x))
664 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
665 mm, resident_page_types[i], x);
666 }
667
668 if (mm_pgtables_bytes(mm))
669 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
670 mm_pgtables_bytes(mm));
671
672#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
673 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
674#endif
675}
676
677#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
678#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
679
680/*
681 * Called when the last reference to the mm
682 * is dropped: either by a lazy thread or by
683 * mmput. Free the page directory and the mm.
684 */
685void __mmdrop(struct mm_struct *mm)
686{
687 BUG_ON(mm == &init_mm);
688 WARN_ON_ONCE(mm == current->mm);
689 WARN_ON_ONCE(mm == current->active_mm);
690 mm_free_pgd(mm);
691 destroy_context(mm);
692 mmu_notifier_mm_destroy(mm);
693 check_mm(mm);
694 put_user_ns(mm->user_ns);
695 free_mm(mm);
696}
697EXPORT_SYMBOL_GPL(__mmdrop);
698
699static void mmdrop_async_fn(struct work_struct *work)
700{
701 struct mm_struct *mm;
702
703 mm = container_of(work, struct mm_struct, async_put_work);
704 __mmdrop(mm);
705}
706
707static void mmdrop_async(struct mm_struct *mm)
708{
709 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
710 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
711 schedule_work(&mm->async_put_work);
712 }
713}
714
715static inline void free_signal_struct(struct signal_struct *sig)
716{
717 taskstats_tgid_free(sig);
718 sched_autogroup_exit(sig);
719 /*
720 * __mmdrop is not safe to call from softirq context on x86 due to
721 * pgd_dtor so postpone it to the async context
722 */
723 if (sig->oom_mm)
724 mmdrop_async(sig->oom_mm);
725 kmem_cache_free(signal_cachep, sig);
726}
727
728static inline void put_signal_struct(struct signal_struct *sig)
729{
730 if (refcount_dec_and_test(&sig->sigcnt))
731 free_signal_struct(sig);
732}
733
734void __put_task_struct(struct task_struct *tsk)
735{
736 WARN_ON(!tsk->exit_state);
737 WARN_ON(refcount_read(&tsk->usage));
738 WARN_ON(tsk == current);
739
740 cgroup_free(tsk);
741 task_numa_free(tsk, true);
742 security_task_free(tsk);
743 exit_creds(tsk);
744 delayacct_tsk_free(tsk);
745 put_signal_struct(tsk->signal);
746
747 if (!profile_handoff_task(tsk))
748 free_task(tsk);
749}
750EXPORT_SYMBOL_GPL(__put_task_struct);
751
752void __init __weak arch_task_cache_init(void) { }
753
754/*
755 * set_max_threads
756 */
757static void set_max_threads(unsigned int max_threads_suggested)
758{
759 u64 threads;
760 unsigned long nr_pages = totalram_pages();
761
762 /*
763 * The number of threads shall be limited such that the thread
764 * structures may only consume a small part of the available memory.
765 */
766 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
767 threads = MAX_THREADS;
768 else
769 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
770 (u64) THREAD_SIZE * 8UL);
771
772 if (threads > max_threads_suggested)
773 threads = max_threads_suggested;
774
775 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
776}
777
778#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
779/* Initialized by the architecture: */
780int arch_task_struct_size __read_mostly;
781#endif
782
783#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
784static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
785{
786 /* Fetch thread_struct whitelist for the architecture. */
787 arch_thread_struct_whitelist(offset, size);
788
789 /*
790 * Handle zero-sized whitelist or empty thread_struct, otherwise
791 * adjust offset to position of thread_struct in task_struct.
792 */
793 if (unlikely(*size == 0))
794 *offset = 0;
795 else
796 *offset += offsetof(struct task_struct, thread);
797}
798#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
799
800void __init fork_init(void)
801{
802 int i;
803#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
804#ifndef ARCH_MIN_TASKALIGN
805#define ARCH_MIN_TASKALIGN 0
806#endif
807 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
808 unsigned long useroffset, usersize;
809
810 /* create a slab on which task_structs can be allocated */
811 task_struct_whitelist(&useroffset, &usersize);
812 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
813 arch_task_struct_size, align,
814 SLAB_PANIC|SLAB_ACCOUNT,
815 useroffset, usersize, NULL);
816#endif
817
818 /* do the arch specific task caches init */
819 arch_task_cache_init();
820
821 set_max_threads(MAX_THREADS);
822
823 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
824 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
825 init_task.signal->rlim[RLIMIT_SIGPENDING] =
826 init_task.signal->rlim[RLIMIT_NPROC];
827
828 for (i = 0; i < UCOUNT_COUNTS; i++) {
829 init_user_ns.ucount_max[i] = max_threads/2;
830 }
831
832#ifdef CONFIG_VMAP_STACK
833 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
834 NULL, free_vm_stack_cache);
835#endif
836
837 lockdep_init_task(&init_task);
838 uprobes_init();
839}
840
841int __weak arch_dup_task_struct(struct task_struct *dst,
842 struct task_struct *src)
843{
844 *dst = *src;
845 return 0;
846}
847
848void set_task_stack_end_magic(struct task_struct *tsk)
849{
850 unsigned long *stackend;
851
852 stackend = end_of_stack(tsk);
853 *stackend = STACK_END_MAGIC; /* for overflow detection */
854}
855
856static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
857{
858 struct task_struct *tsk;
859 unsigned long *stack;
860 struct vm_struct *stack_vm_area __maybe_unused;
861 int err;
862
863 if (node == NUMA_NO_NODE)
864 node = tsk_fork_get_node(orig);
865 tsk = alloc_task_struct_node(node);
866 if (!tsk)
867 return NULL;
868
869 stack = alloc_thread_stack_node(tsk, node);
870 if (!stack)
871 goto free_tsk;
872
873 if (memcg_charge_kernel_stack(tsk))
874 goto free_stack;
875
876 stack_vm_area = task_stack_vm_area(tsk);
877
878 err = arch_dup_task_struct(tsk, orig);
879
880 /*
881 * arch_dup_task_struct() clobbers the stack-related fields. Make
882 * sure they're properly initialized before using any stack-related
883 * functions again.
884 */
885 tsk->stack = stack;
886#ifdef CONFIG_VMAP_STACK
887 tsk->stack_vm_area = stack_vm_area;
888#endif
889#ifdef CONFIG_THREAD_INFO_IN_TASK
890 refcount_set(&tsk->stack_refcount, 1);
891#endif
892
893 if (err)
894 goto free_stack;
895
896#ifdef CONFIG_SECCOMP
897 /*
898 * We must handle setting up seccomp filters once we're under
899 * the sighand lock in case orig has changed between now and
900 * then. Until then, filter must be NULL to avoid messing up
901 * the usage counts on the error path calling free_task.
902 */
903 tsk->seccomp.filter = NULL;
904#endif
905
906 setup_thread_stack(tsk, orig);
907 clear_user_return_notifier(tsk);
908 clear_tsk_need_resched(tsk);
909 set_task_stack_end_magic(tsk);
910
911#ifdef CONFIG_STACKPROTECTOR
912 tsk->stack_canary = get_random_canary();
913#endif
914 if (orig->cpus_ptr == &orig->cpus_mask)
915 tsk->cpus_ptr = &tsk->cpus_mask;
916
917 /*
918 * One for the user space visible state that goes away when reaped.
919 * One for the scheduler.
920 */
921 refcount_set(&tsk->rcu_users, 2);
922 /* One for the rcu users */
923 refcount_set(&tsk->usage, 1);
924#ifdef CONFIG_BLK_DEV_IO_TRACE
925 tsk->btrace_seq = 0;
926#endif
927 tsk->splice_pipe = NULL;
928 tsk->task_frag.page = NULL;
929 tsk->wake_q.next = NULL;
930
931 account_kernel_stack(tsk, 1);
932
933 kcov_task_init(tsk);
934
935#ifdef CONFIG_FAULT_INJECTION
936 tsk->fail_nth = 0;
937#endif
938
939#ifdef CONFIG_BLK_CGROUP
940 tsk->throttle_queue = NULL;
941 tsk->use_memdelay = 0;
942#endif
943
944#ifdef CONFIG_MEMCG
945 tsk->active_memcg = NULL;
946#endif
947 return tsk;
948
949free_stack:
950 free_thread_stack(tsk);
951free_tsk:
952 free_task_struct(tsk);
953 return NULL;
954}
955
956__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
957
958static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
959
960static int __init coredump_filter_setup(char *s)
961{
962 default_dump_filter =
963 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
964 MMF_DUMP_FILTER_MASK;
965 return 1;
966}
967
968__setup("coredump_filter=", coredump_filter_setup);
969
970#include <linux/init_task.h>
971
972static void mm_init_aio(struct mm_struct *mm)
973{
974#ifdef CONFIG_AIO
975 spin_lock_init(&mm->ioctx_lock);
976 mm->ioctx_table = NULL;
977#endif
978}
979
980static __always_inline void mm_clear_owner(struct mm_struct *mm,
981 struct task_struct *p)
982{
983#ifdef CONFIG_MEMCG
984 if (mm->owner == p)
985 WRITE_ONCE(mm->owner, NULL);
986#endif
987}
988
989static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
990{
991#ifdef CONFIG_MEMCG
992 mm->owner = p;
993#endif
994}
995
996static void mm_init_uprobes_state(struct mm_struct *mm)
997{
998#ifdef CONFIG_UPROBES
999 mm->uprobes_state.xol_area = NULL;
1000#endif
1001}
1002
1003static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1004 struct user_namespace *user_ns)
1005{
1006 mm->mmap = NULL;
1007 mm->mm_rb = RB_ROOT;
1008 mm->vmacache_seqnum = 0;
1009 atomic_set(&mm->mm_users, 1);
1010 atomic_set(&mm->mm_count, 1);
1011 init_rwsem(&mm->mmap_sem);
1012 INIT_LIST_HEAD(&mm->mmlist);
1013 mm->core_state = NULL;
1014 mm_pgtables_bytes_init(mm);
1015 mm->map_count = 0;
1016 mm->locked_vm = 0;
1017 atomic64_set(&mm->pinned_vm, 0);
1018 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1019 spin_lock_init(&mm->page_table_lock);
1020 spin_lock_init(&mm->arg_lock);
1021 mm_init_cpumask(mm);
1022 mm_init_aio(mm);
1023 mm_init_owner(mm, p);
1024 RCU_INIT_POINTER(mm->exe_file, NULL);
1025 mmu_notifier_mm_init(mm);
1026 init_tlb_flush_pending(mm);
1027#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1028 mm->pmd_huge_pte = NULL;
1029#endif
1030 mm_init_uprobes_state(mm);
1031
1032 if (current->mm) {
1033 mm->flags = current->mm->flags & MMF_INIT_MASK;
1034 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1035 } else {
1036 mm->flags = default_dump_filter;
1037 mm->def_flags = 0;
1038 }
1039
1040 if (mm_alloc_pgd(mm))
1041 goto fail_nopgd;
1042
1043 if (init_new_context(p, mm))
1044 goto fail_nocontext;
1045
1046 mm->user_ns = get_user_ns(user_ns);
1047 return mm;
1048
1049fail_nocontext:
1050 mm_free_pgd(mm);
1051fail_nopgd:
1052 free_mm(mm);
1053 return NULL;
1054}
1055
1056/*
1057 * Allocate and initialize an mm_struct.
1058 */
1059struct mm_struct *mm_alloc(void)
1060{
1061 struct mm_struct *mm;
1062
1063 mm = allocate_mm();
1064 if (!mm)
1065 return NULL;
1066
1067 memset(mm, 0, sizeof(*mm));
1068 return mm_init(mm, current, current_user_ns());
1069}
1070
1071static inline void __mmput(struct mm_struct *mm)
1072{
1073 VM_BUG_ON(atomic_read(&mm->mm_users));
1074
1075 uprobe_clear_state(mm);
1076 exit_aio(mm);
1077 ksm_exit(mm);
1078 khugepaged_exit(mm); /* must run before exit_mmap */
1079 exit_mmap(mm);
1080 mm_put_huge_zero_page(mm);
1081 set_mm_exe_file(mm, NULL);
1082 if (!list_empty(&mm->mmlist)) {
1083 spin_lock(&mmlist_lock);
1084 list_del(&mm->mmlist);
1085 spin_unlock(&mmlist_lock);
1086 }
1087 if (mm->binfmt)
1088 module_put(mm->binfmt->module);
1089 mmdrop(mm);
1090}
1091
1092/*
1093 * Decrement the use count and release all resources for an mm.
1094 */
1095void mmput(struct mm_struct *mm)
1096{
1097 might_sleep();
1098
1099 if (atomic_dec_and_test(&mm->mm_users))
1100 __mmput(mm);
1101}
1102EXPORT_SYMBOL_GPL(mmput);
1103
1104#ifdef CONFIG_MMU
1105static void mmput_async_fn(struct work_struct *work)
1106{
1107 struct mm_struct *mm = container_of(work, struct mm_struct,
1108 async_put_work);
1109
1110 __mmput(mm);
1111}
1112
1113void mmput_async(struct mm_struct *mm)
1114{
1115 if (atomic_dec_and_test(&mm->mm_users)) {
1116 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1117 schedule_work(&mm->async_put_work);
1118 }
1119}
1120#endif
1121
1122/**
1123 * set_mm_exe_file - change a reference to the mm's executable file
1124 *
1125 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1126 *
1127 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1128 * invocations: in mmput() nobody alive left, in execve task is single
1129 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1130 * mm->exe_file, but does so without using set_mm_exe_file() in order
1131 * to do avoid the need for any locks.
1132 */
1133void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1134{
1135 struct file *old_exe_file;
1136
1137 /*
1138 * It is safe to dereference the exe_file without RCU as
1139 * this function is only called if nobody else can access
1140 * this mm -- see comment above for justification.
1141 */
1142 old_exe_file = rcu_dereference_raw(mm->exe_file);
1143
1144 if (new_exe_file)
1145 get_file(new_exe_file);
1146 rcu_assign_pointer(mm->exe_file, new_exe_file);
1147 if (old_exe_file)
1148 fput(old_exe_file);
1149}
1150
1151/**
1152 * get_mm_exe_file - acquire a reference to the mm's executable file
1153 *
1154 * Returns %NULL if mm has no associated executable file.
1155 * User must release file via fput().
1156 */
1157struct file *get_mm_exe_file(struct mm_struct *mm)
1158{
1159 struct file *exe_file;
1160
1161 rcu_read_lock();
1162 exe_file = rcu_dereference(mm->exe_file);
1163 if (exe_file && !get_file_rcu(exe_file))
1164 exe_file = NULL;
1165 rcu_read_unlock();
1166 return exe_file;
1167}
1168EXPORT_SYMBOL(get_mm_exe_file);
1169
1170/**
1171 * get_task_exe_file - acquire a reference to the task's executable file
1172 *
1173 * Returns %NULL if task's mm (if any) has no associated executable file or
1174 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1175 * User must release file via fput().
1176 */
1177struct file *get_task_exe_file(struct task_struct *task)
1178{
1179 struct file *exe_file = NULL;
1180 struct mm_struct *mm;
1181
1182 task_lock(task);
1183 mm = task->mm;
1184 if (mm) {
1185 if (!(task->flags & PF_KTHREAD))
1186 exe_file = get_mm_exe_file(mm);
1187 }
1188 task_unlock(task);
1189 return exe_file;
1190}
1191EXPORT_SYMBOL(get_task_exe_file);
1192
1193/**
1194 * get_task_mm - acquire a reference to the task's mm
1195 *
1196 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1197 * this kernel workthread has transiently adopted a user mm with use_mm,
1198 * to do its AIO) is not set and if so returns a reference to it, after
1199 * bumping up the use count. User must release the mm via mmput()
1200 * after use. Typically used by /proc and ptrace.
1201 */
1202struct mm_struct *get_task_mm(struct task_struct *task)
1203{
1204 struct mm_struct *mm;
1205
1206 task_lock(task);
1207 mm = task->mm;
1208 if (mm) {
1209 if (task->flags & PF_KTHREAD)
1210 mm = NULL;
1211 else
1212 mmget(mm);
1213 }
1214 task_unlock(task);
1215 return mm;
1216}
1217EXPORT_SYMBOL_GPL(get_task_mm);
1218
1219struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1220{
1221 struct mm_struct *mm;
1222 int err;
1223
1224 err = mutex_lock_killable(&task->signal->cred_guard_mutex);
1225 if (err)
1226 return ERR_PTR(err);
1227
1228 mm = get_task_mm(task);
1229 if (mm && mm != current->mm &&
1230 !ptrace_may_access(task, mode)) {
1231 mmput(mm);
1232 mm = ERR_PTR(-EACCES);
1233 }
1234 mutex_unlock(&task->signal->cred_guard_mutex);
1235
1236 return mm;
1237}
1238
1239static void complete_vfork_done(struct task_struct *tsk)
1240{
1241 struct completion *vfork;
1242
1243 task_lock(tsk);
1244 vfork = tsk->vfork_done;
1245 if (likely(vfork)) {
1246 tsk->vfork_done = NULL;
1247 complete(vfork);
1248 }
1249 task_unlock(tsk);
1250}
1251
1252static int wait_for_vfork_done(struct task_struct *child,
1253 struct completion *vfork)
1254{
1255 int killed;
1256
1257 freezer_do_not_count();
1258 cgroup_enter_frozen();
1259 killed = wait_for_completion_killable(vfork);
1260 cgroup_leave_frozen(false);
1261 freezer_count();
1262
1263 if (killed) {
1264 task_lock(child);
1265 child->vfork_done = NULL;
1266 task_unlock(child);
1267 }
1268
1269 put_task_struct(child);
1270 return killed;
1271}
1272
1273/* Please note the differences between mmput and mm_release.
1274 * mmput is called whenever we stop holding onto a mm_struct,
1275 * error success whatever.
1276 *
1277 * mm_release is called after a mm_struct has been removed
1278 * from the current process.
1279 *
1280 * This difference is important for error handling, when we
1281 * only half set up a mm_struct for a new process and need to restore
1282 * the old one. Because we mmput the new mm_struct before
1283 * restoring the old one. . .
1284 * Eric Biederman 10 January 1998
1285 */
1286void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1287{
1288 /* Get rid of any futexes when releasing the mm */
1289#ifdef CONFIG_FUTEX
1290 if (unlikely(tsk->robust_list)) {
1291 exit_robust_list(tsk);
1292 tsk->robust_list = NULL;
1293 }
1294#ifdef CONFIG_COMPAT
1295 if (unlikely(tsk->compat_robust_list)) {
1296 compat_exit_robust_list(tsk);
1297 tsk->compat_robust_list = NULL;
1298 }
1299#endif
1300 if (unlikely(!list_empty(&tsk->pi_state_list)))
1301 exit_pi_state_list(tsk);
1302#endif
1303
1304 uprobe_free_utask(tsk);
1305
1306 /* Get rid of any cached register state */
1307 deactivate_mm(tsk, mm);
1308
1309 /*
1310 * Signal userspace if we're not exiting with a core dump
1311 * because we want to leave the value intact for debugging
1312 * purposes.
1313 */
1314 if (tsk->clear_child_tid) {
1315 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1316 atomic_read(&mm->mm_users) > 1) {
1317 /*
1318 * We don't check the error code - if userspace has
1319 * not set up a proper pointer then tough luck.
1320 */
1321 put_user(0, tsk->clear_child_tid);
1322 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1323 1, NULL, NULL, 0, 0);
1324 }
1325 tsk->clear_child_tid = NULL;
1326 }
1327
1328 /*
1329 * All done, finally we can wake up parent and return this mm to him.
1330 * Also kthread_stop() uses this completion for synchronization.
1331 */
1332 if (tsk->vfork_done)
1333 complete_vfork_done(tsk);
1334}
1335
1336/**
1337 * dup_mm() - duplicates an existing mm structure
1338 * @tsk: the task_struct with which the new mm will be associated.
1339 * @oldmm: the mm to duplicate.
1340 *
1341 * Allocates a new mm structure and duplicates the provided @oldmm structure
1342 * content into it.
1343 *
1344 * Return: the duplicated mm or NULL on failure.
1345 */
1346static struct mm_struct *dup_mm(struct task_struct *tsk,
1347 struct mm_struct *oldmm)
1348{
1349 struct mm_struct *mm;
1350 int err;
1351
1352 mm = allocate_mm();
1353 if (!mm)
1354 goto fail_nomem;
1355
1356 memcpy(mm, oldmm, sizeof(*mm));
1357
1358 if (!mm_init(mm, tsk, mm->user_ns))
1359 goto fail_nomem;
1360
1361 err = dup_mmap(mm, oldmm);
1362 if (err)
1363 goto free_pt;
1364
1365 mm->hiwater_rss = get_mm_rss(mm);
1366 mm->hiwater_vm = mm->total_vm;
1367
1368 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1369 goto free_pt;
1370
1371 return mm;
1372
1373free_pt:
1374 /* don't put binfmt in mmput, we haven't got module yet */
1375 mm->binfmt = NULL;
1376 mm_init_owner(mm, NULL);
1377 mmput(mm);
1378
1379fail_nomem:
1380 return NULL;
1381}
1382
1383static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1384{
1385 struct mm_struct *mm, *oldmm;
1386 int retval;
1387
1388 tsk->min_flt = tsk->maj_flt = 0;
1389 tsk->nvcsw = tsk->nivcsw = 0;
1390#ifdef CONFIG_DETECT_HUNG_TASK
1391 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1392 tsk->last_switch_time = 0;
1393#endif
1394
1395 tsk->mm = NULL;
1396 tsk->active_mm = NULL;
1397
1398 /*
1399 * Are we cloning a kernel thread?
1400 *
1401 * We need to steal a active VM for that..
1402 */
1403 oldmm = current->mm;
1404 if (!oldmm)
1405 return 0;
1406
1407 /* initialize the new vmacache entries */
1408 vmacache_flush(tsk);
1409
1410 if (clone_flags & CLONE_VM) {
1411 mmget(oldmm);
1412 mm = oldmm;
1413 goto good_mm;
1414 }
1415
1416 retval = -ENOMEM;
1417 mm = dup_mm(tsk, current->mm);
1418 if (!mm)
1419 goto fail_nomem;
1420
1421good_mm:
1422 tsk->mm = mm;
1423 tsk->active_mm = mm;
1424 return 0;
1425
1426fail_nomem:
1427 return retval;
1428}
1429
1430static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1431{
1432 struct fs_struct *fs = current->fs;
1433 if (clone_flags & CLONE_FS) {
1434 /* tsk->fs is already what we want */
1435 spin_lock(&fs->lock);
1436 if (fs->in_exec) {
1437 spin_unlock(&fs->lock);
1438 return -EAGAIN;
1439 }
1440 fs->users++;
1441 spin_unlock(&fs->lock);
1442 return 0;
1443 }
1444 tsk->fs = copy_fs_struct(fs);
1445 if (!tsk->fs)
1446 return -ENOMEM;
1447 return 0;
1448}
1449
1450static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1451{
1452 struct files_struct *oldf, *newf;
1453 int error = 0;
1454
1455 /*
1456 * A background process may not have any files ...
1457 */
1458 oldf = current->files;
1459 if (!oldf)
1460 goto out;
1461
1462 if (clone_flags & CLONE_FILES) {
1463 atomic_inc(&oldf->count);
1464 goto out;
1465 }
1466
1467 newf = dup_fd(oldf, &error);
1468 if (!newf)
1469 goto out;
1470
1471 tsk->files = newf;
1472 error = 0;
1473out:
1474 return error;
1475}
1476
1477static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1478{
1479#ifdef CONFIG_BLOCK
1480 struct io_context *ioc = current->io_context;
1481 struct io_context *new_ioc;
1482
1483 if (!ioc)
1484 return 0;
1485 /*
1486 * Share io context with parent, if CLONE_IO is set
1487 */
1488 if (clone_flags & CLONE_IO) {
1489 ioc_task_link(ioc);
1490 tsk->io_context = ioc;
1491 } else if (ioprio_valid(ioc->ioprio)) {
1492 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1493 if (unlikely(!new_ioc))
1494 return -ENOMEM;
1495
1496 new_ioc->ioprio = ioc->ioprio;
1497 put_io_context(new_ioc);
1498 }
1499#endif
1500 return 0;
1501}
1502
1503static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1504{
1505 struct sighand_struct *sig;
1506
1507 if (clone_flags & CLONE_SIGHAND) {
1508 refcount_inc(¤t->sighand->count);
1509 return 0;
1510 }
1511 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1512 rcu_assign_pointer(tsk->sighand, sig);
1513 if (!sig)
1514 return -ENOMEM;
1515
1516 refcount_set(&sig->count, 1);
1517 spin_lock_irq(¤t->sighand->siglock);
1518 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1519 spin_unlock_irq(¤t->sighand->siglock);
1520 return 0;
1521}
1522
1523void __cleanup_sighand(struct sighand_struct *sighand)
1524{
1525 if (refcount_dec_and_test(&sighand->count)) {
1526 signalfd_cleanup(sighand);
1527 /*
1528 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1529 * without an RCU grace period, see __lock_task_sighand().
1530 */
1531 kmem_cache_free(sighand_cachep, sighand);
1532 }
1533}
1534
1535/*
1536 * Initialize POSIX timer handling for a thread group.
1537 */
1538static void posix_cpu_timers_init_group(struct signal_struct *sig)
1539{
1540 struct posix_cputimers *pct = &sig->posix_cputimers;
1541 unsigned long cpu_limit;
1542
1543 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1544 posix_cputimers_group_init(pct, cpu_limit);
1545}
1546
1547static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1548{
1549 struct signal_struct *sig;
1550
1551 if (clone_flags & CLONE_THREAD)
1552 return 0;
1553
1554 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1555 tsk->signal = sig;
1556 if (!sig)
1557 return -ENOMEM;
1558
1559 sig->nr_threads = 1;
1560 atomic_set(&sig->live, 1);
1561 refcount_set(&sig->sigcnt, 1);
1562
1563 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1564 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1565 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1566
1567 init_waitqueue_head(&sig->wait_chldexit);
1568 sig->curr_target = tsk;
1569 init_sigpending(&sig->shared_pending);
1570 INIT_HLIST_HEAD(&sig->multiprocess);
1571 seqlock_init(&sig->stats_lock);
1572 prev_cputime_init(&sig->prev_cputime);
1573
1574#ifdef CONFIG_POSIX_TIMERS
1575 INIT_LIST_HEAD(&sig->posix_timers);
1576 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1577 sig->real_timer.function = it_real_fn;
1578#endif
1579
1580 task_lock(current->group_leader);
1581 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1582 task_unlock(current->group_leader);
1583
1584 posix_cpu_timers_init_group(sig);
1585
1586 tty_audit_fork(sig);
1587 sched_autogroup_fork(sig);
1588
1589 sig->oom_score_adj = current->signal->oom_score_adj;
1590 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1591
1592 mutex_init(&sig->cred_guard_mutex);
1593
1594 return 0;
1595}
1596
1597static void copy_seccomp(struct task_struct *p)
1598{
1599#ifdef CONFIG_SECCOMP
1600 /*
1601 * Must be called with sighand->lock held, which is common to
1602 * all threads in the group. Holding cred_guard_mutex is not
1603 * needed because this new task is not yet running and cannot
1604 * be racing exec.
1605 */
1606 assert_spin_locked(¤t->sighand->siglock);
1607
1608 /* Ref-count the new filter user, and assign it. */
1609 get_seccomp_filter(current);
1610 p->seccomp = current->seccomp;
1611
1612 /*
1613 * Explicitly enable no_new_privs here in case it got set
1614 * between the task_struct being duplicated and holding the
1615 * sighand lock. The seccomp state and nnp must be in sync.
1616 */
1617 if (task_no_new_privs(current))
1618 task_set_no_new_privs(p);
1619
1620 /*
1621 * If the parent gained a seccomp mode after copying thread
1622 * flags and between before we held the sighand lock, we have
1623 * to manually enable the seccomp thread flag here.
1624 */
1625 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1626 set_tsk_thread_flag(p, TIF_SECCOMP);
1627#endif
1628}
1629
1630SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1631{
1632 current->clear_child_tid = tidptr;
1633
1634 return task_pid_vnr(current);
1635}
1636
1637static void rt_mutex_init_task(struct task_struct *p)
1638{
1639 raw_spin_lock_init(&p->pi_lock);
1640#ifdef CONFIG_RT_MUTEXES
1641 p->pi_waiters = RB_ROOT_CACHED;
1642 p->pi_top_task = NULL;
1643 p->pi_blocked_on = NULL;
1644#endif
1645}
1646
1647static inline void init_task_pid_links(struct task_struct *task)
1648{
1649 enum pid_type type;
1650
1651 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1652 INIT_HLIST_NODE(&task->pid_links[type]);
1653 }
1654}
1655
1656static inline void
1657init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1658{
1659 if (type == PIDTYPE_PID)
1660 task->thread_pid = pid;
1661 else
1662 task->signal->pids[type] = pid;
1663}
1664
1665static inline void rcu_copy_process(struct task_struct *p)
1666{
1667#ifdef CONFIG_PREEMPT_RCU
1668 p->rcu_read_lock_nesting = 0;
1669 p->rcu_read_unlock_special.s = 0;
1670 p->rcu_blocked_node = NULL;
1671 INIT_LIST_HEAD(&p->rcu_node_entry);
1672#endif /* #ifdef CONFIG_PREEMPT_RCU */
1673#ifdef CONFIG_TASKS_RCU
1674 p->rcu_tasks_holdout = false;
1675 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1676 p->rcu_tasks_idle_cpu = -1;
1677#endif /* #ifdef CONFIG_TASKS_RCU */
1678}
1679
1680struct pid *pidfd_pid(const struct file *file)
1681{
1682 if (file->f_op == &pidfd_fops)
1683 return file->private_data;
1684
1685 return ERR_PTR(-EBADF);
1686}
1687
1688static int pidfd_release(struct inode *inode, struct file *file)
1689{
1690 struct pid *pid = file->private_data;
1691
1692 file->private_data = NULL;
1693 put_pid(pid);
1694 return 0;
1695}
1696
1697#ifdef CONFIG_PROC_FS
1698static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1699{
1700 struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1701 struct pid *pid = f->private_data;
1702
1703 seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1704 seq_putc(m, '\n');
1705}
1706#endif
1707
1708/*
1709 * Poll support for process exit notification.
1710 */
1711static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1712{
1713 struct task_struct *task;
1714 struct pid *pid = file->private_data;
1715 __poll_t poll_flags = 0;
1716
1717 poll_wait(file, &pid->wait_pidfd, pts);
1718
1719 rcu_read_lock();
1720 task = pid_task(pid, PIDTYPE_PID);
1721 /*
1722 * Inform pollers only when the whole thread group exits.
1723 * If the thread group leader exits before all other threads in the
1724 * group, then poll(2) should block, similar to the wait(2) family.
1725 */
1726 if (!task || (task->exit_state && thread_group_empty(task)))
1727 poll_flags = EPOLLIN | EPOLLRDNORM;
1728 rcu_read_unlock();
1729
1730 return poll_flags;
1731}
1732
1733const struct file_operations pidfd_fops = {
1734 .release = pidfd_release,
1735 .poll = pidfd_poll,
1736#ifdef CONFIG_PROC_FS
1737 .show_fdinfo = pidfd_show_fdinfo,
1738#endif
1739};
1740
1741static void __delayed_free_task(struct rcu_head *rhp)
1742{
1743 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1744
1745 free_task(tsk);
1746}
1747
1748static __always_inline void delayed_free_task(struct task_struct *tsk)
1749{
1750 if (IS_ENABLED(CONFIG_MEMCG))
1751 call_rcu(&tsk->rcu, __delayed_free_task);
1752 else
1753 free_task(tsk);
1754}
1755
1756/*
1757 * This creates a new process as a copy of the old one,
1758 * but does not actually start it yet.
1759 *
1760 * It copies the registers, and all the appropriate
1761 * parts of the process environment (as per the clone
1762 * flags). The actual kick-off is left to the caller.
1763 */
1764static __latent_entropy struct task_struct *copy_process(
1765 struct pid *pid,
1766 int trace,
1767 int node,
1768 struct kernel_clone_args *args)
1769{
1770 int pidfd = -1, retval;
1771 struct task_struct *p;
1772 struct multiprocess_signals delayed;
1773 struct file *pidfile = NULL;
1774 u64 clone_flags = args->flags;
1775
1776 /*
1777 * Don't allow sharing the root directory with processes in a different
1778 * namespace
1779 */
1780 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1781 return ERR_PTR(-EINVAL);
1782
1783 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1784 return ERR_PTR(-EINVAL);
1785
1786 /*
1787 * Thread groups must share signals as well, and detached threads
1788 * can only be started up within the thread group.
1789 */
1790 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1791 return ERR_PTR(-EINVAL);
1792
1793 /*
1794 * Shared signal handlers imply shared VM. By way of the above,
1795 * thread groups also imply shared VM. Blocking this case allows
1796 * for various simplifications in other code.
1797 */
1798 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1799 return ERR_PTR(-EINVAL);
1800
1801 /*
1802 * Siblings of global init remain as zombies on exit since they are
1803 * not reaped by their parent (swapper). To solve this and to avoid
1804 * multi-rooted process trees, prevent global and container-inits
1805 * from creating siblings.
1806 */
1807 if ((clone_flags & CLONE_PARENT) &&
1808 current->signal->flags & SIGNAL_UNKILLABLE)
1809 return ERR_PTR(-EINVAL);
1810
1811 /*
1812 * If the new process will be in a different pid or user namespace
1813 * do not allow it to share a thread group with the forking task.
1814 */
1815 if (clone_flags & CLONE_THREAD) {
1816 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1817 (task_active_pid_ns(current) !=
1818 current->nsproxy->pid_ns_for_children))
1819 return ERR_PTR(-EINVAL);
1820 }
1821
1822 if (clone_flags & CLONE_PIDFD) {
1823 /*
1824 * - CLONE_DETACHED is blocked so that we can potentially
1825 * reuse it later for CLONE_PIDFD.
1826 * - CLONE_THREAD is blocked until someone really needs it.
1827 */
1828 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1829 return ERR_PTR(-EINVAL);
1830 }
1831
1832 /*
1833 * Force any signals received before this point to be delivered
1834 * before the fork happens. Collect up signals sent to multiple
1835 * processes that happen during the fork and delay them so that
1836 * they appear to happen after the fork.
1837 */
1838 sigemptyset(&delayed.signal);
1839 INIT_HLIST_NODE(&delayed.node);
1840
1841 spin_lock_irq(¤t->sighand->siglock);
1842 if (!(clone_flags & CLONE_THREAD))
1843 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1844 recalc_sigpending();
1845 spin_unlock_irq(¤t->sighand->siglock);
1846 retval = -ERESTARTNOINTR;
1847 if (signal_pending(current))
1848 goto fork_out;
1849
1850 retval = -ENOMEM;
1851 p = dup_task_struct(current, node);
1852 if (!p)
1853 goto fork_out;
1854
1855 /*
1856 * This _must_ happen before we call free_task(), i.e. before we jump
1857 * to any of the bad_fork_* labels. This is to avoid freeing
1858 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1859 * kernel threads (PF_KTHREAD).
1860 */
1861 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1862 /*
1863 * Clear TID on mm_release()?
1864 */
1865 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1866
1867 ftrace_graph_init_task(p);
1868
1869 rt_mutex_init_task(p);
1870
1871#ifdef CONFIG_PROVE_LOCKING
1872 DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1873 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1874#endif
1875 retval = -EAGAIN;
1876 if (atomic_read(&p->real_cred->user->processes) >=
1877 task_rlimit(p, RLIMIT_NPROC)) {
1878 if (p->real_cred->user != INIT_USER &&
1879 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1880 goto bad_fork_free;
1881 }
1882 current->flags &= ~PF_NPROC_EXCEEDED;
1883
1884 retval = copy_creds(p, clone_flags);
1885 if (retval < 0)
1886 goto bad_fork_free;
1887
1888 /*
1889 * If multiple threads are within copy_process(), then this check
1890 * triggers too late. This doesn't hurt, the check is only there
1891 * to stop root fork bombs.
1892 */
1893 retval = -EAGAIN;
1894 if (nr_threads >= max_threads)
1895 goto bad_fork_cleanup_count;
1896
1897 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1898 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1899 p->flags |= PF_FORKNOEXEC;
1900 INIT_LIST_HEAD(&p->children);
1901 INIT_LIST_HEAD(&p->sibling);
1902 rcu_copy_process(p);
1903 p->vfork_done = NULL;
1904 spin_lock_init(&p->alloc_lock);
1905
1906 init_sigpending(&p->pending);
1907
1908 p->utime = p->stime = p->gtime = 0;
1909#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1910 p->utimescaled = p->stimescaled = 0;
1911#endif
1912 prev_cputime_init(&p->prev_cputime);
1913
1914#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1915 seqcount_init(&p->vtime.seqcount);
1916 p->vtime.starttime = 0;
1917 p->vtime.state = VTIME_INACTIVE;
1918#endif
1919
1920#if defined(SPLIT_RSS_COUNTING)
1921 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1922#endif
1923
1924 p->default_timer_slack_ns = current->timer_slack_ns;
1925
1926#ifdef CONFIG_PSI
1927 p->psi_flags = 0;
1928#endif
1929
1930 task_io_accounting_init(&p->ioac);
1931 acct_clear_integrals(p);
1932
1933 posix_cputimers_init(&p->posix_cputimers);
1934
1935 p->io_context = NULL;
1936 audit_set_context(p, NULL);
1937 cgroup_fork(p);
1938#ifdef CONFIG_NUMA
1939 p->mempolicy = mpol_dup(p->mempolicy);
1940 if (IS_ERR(p->mempolicy)) {
1941 retval = PTR_ERR(p->mempolicy);
1942 p->mempolicy = NULL;
1943 goto bad_fork_cleanup_threadgroup_lock;
1944 }
1945#endif
1946#ifdef CONFIG_CPUSETS
1947 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1948 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1949 seqcount_init(&p->mems_allowed_seq);
1950#endif
1951#ifdef CONFIG_TRACE_IRQFLAGS
1952 p->irq_events = 0;
1953 p->hardirqs_enabled = 0;
1954 p->hardirq_enable_ip = 0;
1955 p->hardirq_enable_event = 0;
1956 p->hardirq_disable_ip = _THIS_IP_;
1957 p->hardirq_disable_event = 0;
1958 p->softirqs_enabled = 1;
1959 p->softirq_enable_ip = _THIS_IP_;
1960 p->softirq_enable_event = 0;
1961 p->softirq_disable_ip = 0;
1962 p->softirq_disable_event = 0;
1963 p->hardirq_context = 0;
1964 p->softirq_context = 0;
1965#endif
1966
1967 p->pagefault_disabled = 0;
1968
1969#ifdef CONFIG_LOCKDEP
1970 lockdep_init_task(p);
1971#endif
1972
1973#ifdef CONFIG_DEBUG_MUTEXES
1974 p->blocked_on = NULL; /* not blocked yet */
1975#endif
1976#ifdef CONFIG_BCACHE
1977 p->sequential_io = 0;
1978 p->sequential_io_avg = 0;
1979#endif
1980
1981 /* Perform scheduler related setup. Assign this task to a CPU. */
1982 retval = sched_fork(clone_flags, p);
1983 if (retval)
1984 goto bad_fork_cleanup_policy;
1985
1986 retval = perf_event_init_task(p);
1987 if (retval)
1988 goto bad_fork_cleanup_policy;
1989 retval = audit_alloc(p);
1990 if (retval)
1991 goto bad_fork_cleanup_perf;
1992 /* copy all the process information */
1993 shm_init_task(p);
1994 retval = security_task_alloc(p, clone_flags);
1995 if (retval)
1996 goto bad_fork_cleanup_audit;
1997 retval = copy_semundo(clone_flags, p);
1998 if (retval)
1999 goto bad_fork_cleanup_security;
2000 retval = copy_files(clone_flags, p);
2001 if (retval)
2002 goto bad_fork_cleanup_semundo;
2003 retval = copy_fs(clone_flags, p);
2004 if (retval)
2005 goto bad_fork_cleanup_files;
2006 retval = copy_sighand(clone_flags, p);
2007 if (retval)
2008 goto bad_fork_cleanup_fs;
2009 retval = copy_signal(clone_flags, p);
2010 if (retval)
2011 goto bad_fork_cleanup_sighand;
2012 retval = copy_mm(clone_flags, p);
2013 if (retval)
2014 goto bad_fork_cleanup_signal;
2015 retval = copy_namespaces(clone_flags, p);
2016 if (retval)
2017 goto bad_fork_cleanup_mm;
2018 retval = copy_io(clone_flags, p);
2019 if (retval)
2020 goto bad_fork_cleanup_namespaces;
2021 retval = copy_thread_tls(clone_flags, args->stack, args->stack_size, p,
2022 args->tls);
2023 if (retval)
2024 goto bad_fork_cleanup_io;
2025
2026 stackleak_task_init(p);
2027
2028 if (pid != &init_struct_pid) {
2029 pid = alloc_pid(p->nsproxy->pid_ns_for_children);
2030 if (IS_ERR(pid)) {
2031 retval = PTR_ERR(pid);
2032 goto bad_fork_cleanup_thread;
2033 }
2034 }
2035
2036 /*
2037 * This has to happen after we've potentially unshared the file
2038 * descriptor table (so that the pidfd doesn't leak into the child
2039 * if the fd table isn't shared).
2040 */
2041 if (clone_flags & CLONE_PIDFD) {
2042 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2043 if (retval < 0)
2044 goto bad_fork_free_pid;
2045
2046 pidfd = retval;
2047
2048 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2049 O_RDWR | O_CLOEXEC);
2050 if (IS_ERR(pidfile)) {
2051 put_unused_fd(pidfd);
2052 retval = PTR_ERR(pidfile);
2053 goto bad_fork_free_pid;
2054 }
2055 get_pid(pid); /* held by pidfile now */
2056
2057 retval = put_user(pidfd, args->pidfd);
2058 if (retval)
2059 goto bad_fork_put_pidfd;
2060 }
2061
2062#ifdef CONFIG_BLOCK
2063 p->plug = NULL;
2064#endif
2065#ifdef CONFIG_FUTEX
2066 p->robust_list = NULL;
2067#ifdef CONFIG_COMPAT
2068 p->compat_robust_list = NULL;
2069#endif
2070 INIT_LIST_HEAD(&p->pi_state_list);
2071 p->pi_state_cache = NULL;
2072#endif
2073 /*
2074 * sigaltstack should be cleared when sharing the same VM
2075 */
2076 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2077 sas_ss_reset(p);
2078
2079 /*
2080 * Syscall tracing and stepping should be turned off in the
2081 * child regardless of CLONE_PTRACE.
2082 */
2083 user_disable_single_step(p);
2084 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2085#ifdef TIF_SYSCALL_EMU
2086 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2087#endif
2088 clear_tsk_latency_tracing(p);
2089
2090 /* ok, now we should be set up.. */
2091 p->pid = pid_nr(pid);
2092 if (clone_flags & CLONE_THREAD) {
2093 p->exit_signal = -1;
2094 p->group_leader = current->group_leader;
2095 p->tgid = current->tgid;
2096 } else {
2097 if (clone_flags & CLONE_PARENT)
2098 p->exit_signal = current->group_leader->exit_signal;
2099 else
2100 p->exit_signal = args->exit_signal;
2101 p->group_leader = p;
2102 p->tgid = p->pid;
2103 }
2104
2105 p->nr_dirtied = 0;
2106 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2107 p->dirty_paused_when = 0;
2108
2109 p->pdeath_signal = 0;
2110 INIT_LIST_HEAD(&p->thread_group);
2111 p->task_works = NULL;
2112
2113 cgroup_threadgroup_change_begin(current);
2114 /*
2115 * Ensure that the cgroup subsystem policies allow the new process to be
2116 * forked. It should be noted the the new process's css_set can be changed
2117 * between here and cgroup_post_fork() if an organisation operation is in
2118 * progress.
2119 */
2120 retval = cgroup_can_fork(p);
2121 if (retval)
2122 goto bad_fork_cgroup_threadgroup_change_end;
2123
2124 /*
2125 * From this point on we must avoid any synchronous user-space
2126 * communication until we take the tasklist-lock. In particular, we do
2127 * not want user-space to be able to predict the process start-time by
2128 * stalling fork(2) after we recorded the start_time but before it is
2129 * visible to the system.
2130 */
2131
2132 p->start_time = ktime_get_ns();
2133 p->real_start_time = ktime_get_boottime_ns();
2134
2135 /*
2136 * Make it visible to the rest of the system, but dont wake it up yet.
2137 * Need tasklist lock for parent etc handling!
2138 */
2139 write_lock_irq(&tasklist_lock);
2140
2141 /* CLONE_PARENT re-uses the old parent */
2142 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2143 p->real_parent = current->real_parent;
2144 p->parent_exec_id = current->parent_exec_id;
2145 } else {
2146 p->real_parent = current;
2147 p->parent_exec_id = current->self_exec_id;
2148 }
2149
2150 klp_copy_process(p);
2151
2152 spin_lock(¤t->sighand->siglock);
2153
2154 /*
2155 * Copy seccomp details explicitly here, in case they were changed
2156 * before holding sighand lock.
2157 */
2158 copy_seccomp(p);
2159
2160 rseq_fork(p, clone_flags);
2161
2162 /* Don't start children in a dying pid namespace */
2163 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2164 retval = -ENOMEM;
2165 goto bad_fork_cancel_cgroup;
2166 }
2167
2168 /* Let kill terminate clone/fork in the middle */
2169 if (fatal_signal_pending(current)) {
2170 retval = -EINTR;
2171 goto bad_fork_cancel_cgroup;
2172 }
2173
2174 /* past the last point of failure */
2175 if (pidfile)
2176 fd_install(pidfd, pidfile);
2177
2178 init_task_pid_links(p);
2179 if (likely(p->pid)) {
2180 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2181
2182 init_task_pid(p, PIDTYPE_PID, pid);
2183 if (thread_group_leader(p)) {
2184 init_task_pid(p, PIDTYPE_TGID, pid);
2185 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2186 init_task_pid(p, PIDTYPE_SID, task_session(current));
2187
2188 if (is_child_reaper(pid)) {
2189 ns_of_pid(pid)->child_reaper = p;
2190 p->signal->flags |= SIGNAL_UNKILLABLE;
2191 }
2192 p->signal->shared_pending.signal = delayed.signal;
2193 p->signal->tty = tty_kref_get(current->signal->tty);
2194 /*
2195 * Inherit has_child_subreaper flag under the same
2196 * tasklist_lock with adding child to the process tree
2197 * for propagate_has_child_subreaper optimization.
2198 */
2199 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2200 p->real_parent->signal->is_child_subreaper;
2201 list_add_tail(&p->sibling, &p->real_parent->children);
2202 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2203 attach_pid(p, PIDTYPE_TGID);
2204 attach_pid(p, PIDTYPE_PGID);
2205 attach_pid(p, PIDTYPE_SID);
2206 __this_cpu_inc(process_counts);
2207 } else {
2208 current->signal->nr_threads++;
2209 atomic_inc(¤t->signal->live);
2210 refcount_inc(¤t->signal->sigcnt);
2211 task_join_group_stop(p);
2212 list_add_tail_rcu(&p->thread_group,
2213 &p->group_leader->thread_group);
2214 list_add_tail_rcu(&p->thread_node,
2215 &p->signal->thread_head);
2216 }
2217 attach_pid(p, PIDTYPE_PID);
2218 nr_threads++;
2219 }
2220 total_forks++;
2221 hlist_del_init(&delayed.node);
2222 spin_unlock(¤t->sighand->siglock);
2223 syscall_tracepoint_update(p);
2224 write_unlock_irq(&tasklist_lock);
2225
2226 proc_fork_connector(p);
2227 cgroup_post_fork(p);
2228 cgroup_threadgroup_change_end(current);
2229 perf_event_fork(p);
2230
2231 trace_task_newtask(p, clone_flags);
2232 uprobe_copy_process(p, clone_flags);
2233
2234 return p;
2235
2236bad_fork_cancel_cgroup:
2237 spin_unlock(¤t->sighand->siglock);
2238 write_unlock_irq(&tasklist_lock);
2239 cgroup_cancel_fork(p);
2240bad_fork_cgroup_threadgroup_change_end:
2241 cgroup_threadgroup_change_end(current);
2242bad_fork_put_pidfd:
2243 if (clone_flags & CLONE_PIDFD) {
2244 fput(pidfile);
2245 put_unused_fd(pidfd);
2246 }
2247bad_fork_free_pid:
2248 if (pid != &init_struct_pid)
2249 free_pid(pid);
2250bad_fork_cleanup_thread:
2251 exit_thread(p);
2252bad_fork_cleanup_io:
2253 if (p->io_context)
2254 exit_io_context(p);
2255bad_fork_cleanup_namespaces:
2256 exit_task_namespaces(p);
2257bad_fork_cleanup_mm:
2258 if (p->mm) {
2259 mm_clear_owner(p->mm, p);
2260 mmput(p->mm);
2261 }
2262bad_fork_cleanup_signal:
2263 if (!(clone_flags & CLONE_THREAD))
2264 free_signal_struct(p->signal);
2265bad_fork_cleanup_sighand:
2266 __cleanup_sighand(p->sighand);
2267bad_fork_cleanup_fs:
2268 exit_fs(p); /* blocking */
2269bad_fork_cleanup_files:
2270 exit_files(p); /* blocking */
2271bad_fork_cleanup_semundo:
2272 exit_sem(p);
2273bad_fork_cleanup_security:
2274 security_task_free(p);
2275bad_fork_cleanup_audit:
2276 audit_free(p);
2277bad_fork_cleanup_perf:
2278 perf_event_free_task(p);
2279bad_fork_cleanup_policy:
2280 lockdep_free_task(p);
2281#ifdef CONFIG_NUMA
2282 mpol_put(p->mempolicy);
2283bad_fork_cleanup_threadgroup_lock:
2284#endif
2285 delayacct_tsk_free(p);
2286bad_fork_cleanup_count:
2287 atomic_dec(&p->cred->user->processes);
2288 exit_creds(p);
2289bad_fork_free:
2290 p->state = TASK_DEAD;
2291 put_task_stack(p);
2292 delayed_free_task(p);
2293fork_out:
2294 spin_lock_irq(¤t->sighand->siglock);
2295 hlist_del_init(&delayed.node);
2296 spin_unlock_irq(¤t->sighand->siglock);
2297 return ERR_PTR(retval);
2298}
2299
2300static inline void init_idle_pids(struct task_struct *idle)
2301{
2302 enum pid_type type;
2303
2304 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2305 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2306 init_task_pid(idle, type, &init_struct_pid);
2307 }
2308}
2309
2310struct task_struct *fork_idle(int cpu)
2311{
2312 struct task_struct *task;
2313 struct kernel_clone_args args = {
2314 .flags = CLONE_VM,
2315 };
2316
2317 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2318 if (!IS_ERR(task)) {
2319 init_idle_pids(task);
2320 init_idle(task, cpu);
2321 }
2322
2323 return task;
2324}
2325
2326struct mm_struct *copy_init_mm(void)
2327{
2328 return dup_mm(NULL, &init_mm);
2329}
2330
2331/*
2332 * Ok, this is the main fork-routine.
2333 *
2334 * It copies the process, and if successful kick-starts
2335 * it and waits for it to finish using the VM if required.
2336 *
2337 * args->exit_signal is expected to be checked for sanity by the caller.
2338 */
2339long _do_fork(struct kernel_clone_args *args)
2340{
2341 u64 clone_flags = args->flags;
2342 struct completion vfork;
2343 struct pid *pid;
2344 struct task_struct *p;
2345 int trace = 0;
2346 long nr;
2347
2348 /*
2349 * Determine whether and which event to report to ptracer. When
2350 * called from kernel_thread or CLONE_UNTRACED is explicitly
2351 * requested, no event is reported; otherwise, report if the event
2352 * for the type of forking is enabled.
2353 */
2354 if (!(clone_flags & CLONE_UNTRACED)) {
2355 if (clone_flags & CLONE_VFORK)
2356 trace = PTRACE_EVENT_VFORK;
2357 else if (args->exit_signal != SIGCHLD)
2358 trace = PTRACE_EVENT_CLONE;
2359 else
2360 trace = PTRACE_EVENT_FORK;
2361
2362 if (likely(!ptrace_event_enabled(current, trace)))
2363 trace = 0;
2364 }
2365
2366 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2367 add_latent_entropy();
2368
2369 if (IS_ERR(p))
2370 return PTR_ERR(p);
2371
2372 /*
2373 * Do this prior waking up the new thread - the thread pointer
2374 * might get invalid after that point, if the thread exits quickly.
2375 */
2376 trace_sched_process_fork(current, p);
2377
2378 pid = get_task_pid(p, PIDTYPE_PID);
2379 nr = pid_vnr(pid);
2380
2381 if (clone_flags & CLONE_PARENT_SETTID)
2382 put_user(nr, args->parent_tid);
2383
2384 if (clone_flags & CLONE_VFORK) {
2385 p->vfork_done = &vfork;
2386 init_completion(&vfork);
2387 get_task_struct(p);
2388 }
2389
2390 wake_up_new_task(p);
2391
2392 /* forking complete and child started to run, tell ptracer */
2393 if (unlikely(trace))
2394 ptrace_event_pid(trace, pid);
2395
2396 if (clone_flags & CLONE_VFORK) {
2397 if (!wait_for_vfork_done(p, &vfork))
2398 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2399 }
2400
2401 put_pid(pid);
2402 return nr;
2403}
2404
2405bool legacy_clone_args_valid(const struct kernel_clone_args *kargs)
2406{
2407 /* clone(CLONE_PIDFD) uses parent_tidptr to return a pidfd */
2408 if ((kargs->flags & CLONE_PIDFD) &&
2409 (kargs->flags & CLONE_PARENT_SETTID))
2410 return false;
2411
2412 return true;
2413}
2414
2415#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2416/* For compatibility with architectures that call do_fork directly rather than
2417 * using the syscall entry points below. */
2418long do_fork(unsigned long clone_flags,
2419 unsigned long stack_start,
2420 unsigned long stack_size,
2421 int __user *parent_tidptr,
2422 int __user *child_tidptr)
2423{
2424 struct kernel_clone_args args = {
2425 .flags = (clone_flags & ~CSIGNAL),
2426 .pidfd = parent_tidptr,
2427 .child_tid = child_tidptr,
2428 .parent_tid = parent_tidptr,
2429 .exit_signal = (clone_flags & CSIGNAL),
2430 .stack = stack_start,
2431 .stack_size = stack_size,
2432 };
2433
2434 if (!legacy_clone_args_valid(&args))
2435 return -EINVAL;
2436
2437 return _do_fork(&args);
2438}
2439#endif
2440
2441/*
2442 * Create a kernel thread.
2443 */
2444pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2445{
2446 struct kernel_clone_args args = {
2447 .flags = ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2448 .exit_signal = (flags & CSIGNAL),
2449 .stack = (unsigned long)fn,
2450 .stack_size = (unsigned long)arg,
2451 };
2452
2453 return _do_fork(&args);
2454}
2455
2456#ifdef __ARCH_WANT_SYS_FORK
2457SYSCALL_DEFINE0(fork)
2458{
2459#ifdef CONFIG_MMU
2460 struct kernel_clone_args args = {
2461 .exit_signal = SIGCHLD,
2462 };
2463
2464 return _do_fork(&args);
2465#else
2466 /* can not support in nommu mode */
2467 return -EINVAL;
2468#endif
2469}
2470#endif
2471
2472#ifdef __ARCH_WANT_SYS_VFORK
2473SYSCALL_DEFINE0(vfork)
2474{
2475 struct kernel_clone_args args = {
2476 .flags = CLONE_VFORK | CLONE_VM,
2477 .exit_signal = SIGCHLD,
2478 };
2479
2480 return _do_fork(&args);
2481}
2482#endif
2483
2484#ifdef __ARCH_WANT_SYS_CLONE
2485#ifdef CONFIG_CLONE_BACKWARDS
2486SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2487 int __user *, parent_tidptr,
2488 unsigned long, tls,
2489 int __user *, child_tidptr)
2490#elif defined(CONFIG_CLONE_BACKWARDS2)
2491SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2492 int __user *, parent_tidptr,
2493 int __user *, child_tidptr,
2494 unsigned long, tls)
2495#elif defined(CONFIG_CLONE_BACKWARDS3)
2496SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2497 int, stack_size,
2498 int __user *, parent_tidptr,
2499 int __user *, child_tidptr,
2500 unsigned long, tls)
2501#else
2502SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2503 int __user *, parent_tidptr,
2504 int __user *, child_tidptr,
2505 unsigned long, tls)
2506#endif
2507{
2508 struct kernel_clone_args args = {
2509 .flags = (clone_flags & ~CSIGNAL),
2510 .pidfd = parent_tidptr,
2511 .child_tid = child_tidptr,
2512 .parent_tid = parent_tidptr,
2513 .exit_signal = (clone_flags & CSIGNAL),
2514 .stack = newsp,
2515 .tls = tls,
2516 };
2517
2518 if (!legacy_clone_args_valid(&args))
2519 return -EINVAL;
2520
2521 return _do_fork(&args);
2522}
2523#endif
2524
2525#ifdef __ARCH_WANT_SYS_CLONE3
2526noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2527 struct clone_args __user *uargs,
2528 size_t usize)
2529{
2530 int err;
2531 struct clone_args args;
2532
2533 if (unlikely(usize > PAGE_SIZE))
2534 return -E2BIG;
2535 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2536 return -EINVAL;
2537
2538 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2539 if (err)
2540 return err;
2541
2542 /*
2543 * Verify that higher 32bits of exit_signal are unset and that
2544 * it is a valid signal
2545 */
2546 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2547 !valid_signal(args.exit_signal)))
2548 return -EINVAL;
2549
2550 *kargs = (struct kernel_clone_args){
2551 .flags = args.flags,
2552 .pidfd = u64_to_user_ptr(args.pidfd),
2553 .child_tid = u64_to_user_ptr(args.child_tid),
2554 .parent_tid = u64_to_user_ptr(args.parent_tid),
2555 .exit_signal = args.exit_signal,
2556 .stack = args.stack,
2557 .stack_size = args.stack_size,
2558 .tls = args.tls,
2559 };
2560
2561 return 0;
2562}
2563
2564/**
2565 * clone3_stack_valid - check and prepare stack
2566 * @kargs: kernel clone args
2567 *
2568 * Verify that the stack arguments userspace gave us are sane.
2569 * In addition, set the stack direction for userspace since it's easy for us to
2570 * determine.
2571 */
2572static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2573{
2574 if (kargs->stack == 0) {
2575 if (kargs->stack_size > 0)
2576 return false;
2577 } else {
2578 if (kargs->stack_size == 0)
2579 return false;
2580
2581 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2582 return false;
2583
2584#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2585 kargs->stack += kargs->stack_size;
2586#endif
2587 }
2588
2589 return true;
2590}
2591
2592static bool clone3_args_valid(struct kernel_clone_args *kargs)
2593{
2594 /*
2595 * All lower bits of the flag word are taken.
2596 * Verify that no other unknown flags are passed along.
2597 */
2598 if (kargs->flags & ~CLONE_LEGACY_FLAGS)
2599 return false;
2600
2601 /*
2602 * - make the CLONE_DETACHED bit reuseable for clone3
2603 * - make the CSIGNAL bits reuseable for clone3
2604 */
2605 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2606 return false;
2607
2608 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2609 kargs->exit_signal)
2610 return false;
2611
2612 if (!clone3_stack_valid(kargs))
2613 return false;
2614
2615 return true;
2616}
2617
2618/**
2619 * clone3 - create a new process with specific properties
2620 * @uargs: argument structure
2621 * @size: size of @uargs
2622 *
2623 * clone3() is the extensible successor to clone()/clone2().
2624 * It takes a struct as argument that is versioned by its size.
2625 *
2626 * Return: On success, a positive PID for the child process.
2627 * On error, a negative errno number.
2628 */
2629SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2630{
2631 int err;
2632
2633 struct kernel_clone_args kargs;
2634
2635 err = copy_clone_args_from_user(&kargs, uargs, size);
2636 if (err)
2637 return err;
2638
2639 if (!clone3_args_valid(&kargs))
2640 return -EINVAL;
2641
2642 return _do_fork(&kargs);
2643}
2644#endif
2645
2646void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2647{
2648 struct task_struct *leader, *parent, *child;
2649 int res;
2650
2651 read_lock(&tasklist_lock);
2652 leader = top = top->group_leader;
2653down:
2654 for_each_thread(leader, parent) {
2655 list_for_each_entry(child, &parent->children, sibling) {
2656 res = visitor(child, data);
2657 if (res) {
2658 if (res < 0)
2659 goto out;
2660 leader = child;
2661 goto down;
2662 }
2663up:
2664 ;
2665 }
2666 }
2667
2668 if (leader != top) {
2669 child = leader;
2670 parent = child->real_parent;
2671 leader = parent->group_leader;
2672 goto up;
2673 }
2674out:
2675 read_unlock(&tasklist_lock);
2676}
2677
2678#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2679#define ARCH_MIN_MMSTRUCT_ALIGN 0
2680#endif
2681
2682static void sighand_ctor(void *data)
2683{
2684 struct sighand_struct *sighand = data;
2685
2686 spin_lock_init(&sighand->siglock);
2687 init_waitqueue_head(&sighand->signalfd_wqh);
2688}
2689
2690void __init proc_caches_init(void)
2691{
2692 unsigned int mm_size;
2693
2694 sighand_cachep = kmem_cache_create("sighand_cache",
2695 sizeof(struct sighand_struct), 0,
2696 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2697 SLAB_ACCOUNT, sighand_ctor);
2698 signal_cachep = kmem_cache_create("signal_cache",
2699 sizeof(struct signal_struct), 0,
2700 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2701 NULL);
2702 files_cachep = kmem_cache_create("files_cache",
2703 sizeof(struct files_struct), 0,
2704 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2705 NULL);
2706 fs_cachep = kmem_cache_create("fs_cache",
2707 sizeof(struct fs_struct), 0,
2708 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2709 NULL);
2710
2711 /*
2712 * The mm_cpumask is located at the end of mm_struct, and is
2713 * dynamically sized based on the maximum CPU number this system
2714 * can have, taking hotplug into account (nr_cpu_ids).
2715 */
2716 mm_size = sizeof(struct mm_struct) + cpumask_size();
2717
2718 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2719 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2720 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2721 offsetof(struct mm_struct, saved_auxv),
2722 sizeof_field(struct mm_struct, saved_auxv),
2723 NULL);
2724 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2725 mmap_init();
2726 nsproxy_cache_init();
2727}
2728
2729/*
2730 * Check constraints on flags passed to the unshare system call.
2731 */
2732static int check_unshare_flags(unsigned long unshare_flags)
2733{
2734 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2735 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2736 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2737 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2738 return -EINVAL;
2739 /*
2740 * Not implemented, but pretend it works if there is nothing
2741 * to unshare. Note that unsharing the address space or the
2742 * signal handlers also need to unshare the signal queues (aka
2743 * CLONE_THREAD).
2744 */
2745 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2746 if (!thread_group_empty(current))
2747 return -EINVAL;
2748 }
2749 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2750 if (refcount_read(¤t->sighand->count) > 1)
2751 return -EINVAL;
2752 }
2753 if (unshare_flags & CLONE_VM) {
2754 if (!current_is_single_threaded())
2755 return -EINVAL;
2756 }
2757
2758 return 0;
2759}
2760
2761/*
2762 * Unshare the filesystem structure if it is being shared
2763 */
2764static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2765{
2766 struct fs_struct *fs = current->fs;
2767
2768 if (!(unshare_flags & CLONE_FS) || !fs)
2769 return 0;
2770
2771 /* don't need lock here; in the worst case we'll do useless copy */
2772 if (fs->users == 1)
2773 return 0;
2774
2775 *new_fsp = copy_fs_struct(fs);
2776 if (!*new_fsp)
2777 return -ENOMEM;
2778
2779 return 0;
2780}
2781
2782/*
2783 * Unshare file descriptor table if it is being shared
2784 */
2785static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2786{
2787 struct files_struct *fd = current->files;
2788 int error = 0;
2789
2790 if ((unshare_flags & CLONE_FILES) &&
2791 (fd && atomic_read(&fd->count) > 1)) {
2792 *new_fdp = dup_fd(fd, &error);
2793 if (!*new_fdp)
2794 return error;
2795 }
2796
2797 return 0;
2798}
2799
2800/*
2801 * unshare allows a process to 'unshare' part of the process
2802 * context which was originally shared using clone. copy_*
2803 * functions used by do_fork() cannot be used here directly
2804 * because they modify an inactive task_struct that is being
2805 * constructed. Here we are modifying the current, active,
2806 * task_struct.
2807 */
2808int ksys_unshare(unsigned long unshare_flags)
2809{
2810 struct fs_struct *fs, *new_fs = NULL;
2811 struct files_struct *fd, *new_fd = NULL;
2812 struct cred *new_cred = NULL;
2813 struct nsproxy *new_nsproxy = NULL;
2814 int do_sysvsem = 0;
2815 int err;
2816
2817 /*
2818 * If unsharing a user namespace must also unshare the thread group
2819 * and unshare the filesystem root and working directories.
2820 */
2821 if (unshare_flags & CLONE_NEWUSER)
2822 unshare_flags |= CLONE_THREAD | CLONE_FS;
2823 /*
2824 * If unsharing vm, must also unshare signal handlers.
2825 */
2826 if (unshare_flags & CLONE_VM)
2827 unshare_flags |= CLONE_SIGHAND;
2828 /*
2829 * If unsharing a signal handlers, must also unshare the signal queues.
2830 */
2831 if (unshare_flags & CLONE_SIGHAND)
2832 unshare_flags |= CLONE_THREAD;
2833 /*
2834 * If unsharing namespace, must also unshare filesystem information.
2835 */
2836 if (unshare_flags & CLONE_NEWNS)
2837 unshare_flags |= CLONE_FS;
2838
2839 err = check_unshare_flags(unshare_flags);
2840 if (err)
2841 goto bad_unshare_out;
2842 /*
2843 * CLONE_NEWIPC must also detach from the undolist: after switching
2844 * to a new ipc namespace, the semaphore arrays from the old
2845 * namespace are unreachable.
2846 */
2847 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2848 do_sysvsem = 1;
2849 err = unshare_fs(unshare_flags, &new_fs);
2850 if (err)
2851 goto bad_unshare_out;
2852 err = unshare_fd(unshare_flags, &new_fd);
2853 if (err)
2854 goto bad_unshare_cleanup_fs;
2855 err = unshare_userns(unshare_flags, &new_cred);
2856 if (err)
2857 goto bad_unshare_cleanup_fd;
2858 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2859 new_cred, new_fs);
2860 if (err)
2861 goto bad_unshare_cleanup_cred;
2862
2863 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2864 if (do_sysvsem) {
2865 /*
2866 * CLONE_SYSVSEM is equivalent to sys_exit().
2867 */
2868 exit_sem(current);
2869 }
2870 if (unshare_flags & CLONE_NEWIPC) {
2871 /* Orphan segments in old ns (see sem above). */
2872 exit_shm(current);
2873 shm_init_task(current);
2874 }
2875
2876 if (new_nsproxy)
2877 switch_task_namespaces(current, new_nsproxy);
2878
2879 task_lock(current);
2880
2881 if (new_fs) {
2882 fs = current->fs;
2883 spin_lock(&fs->lock);
2884 current->fs = new_fs;
2885 if (--fs->users)
2886 new_fs = NULL;
2887 else
2888 new_fs = fs;
2889 spin_unlock(&fs->lock);
2890 }
2891
2892 if (new_fd) {
2893 fd = current->files;
2894 current->files = new_fd;
2895 new_fd = fd;
2896 }
2897
2898 task_unlock(current);
2899
2900 if (new_cred) {
2901 /* Install the new user namespace */
2902 commit_creds(new_cred);
2903 new_cred = NULL;
2904 }
2905 }
2906
2907 perf_event_namespaces(current);
2908
2909bad_unshare_cleanup_cred:
2910 if (new_cred)
2911 put_cred(new_cred);
2912bad_unshare_cleanup_fd:
2913 if (new_fd)
2914 put_files_struct(new_fd);
2915
2916bad_unshare_cleanup_fs:
2917 if (new_fs)
2918 free_fs_struct(new_fs);
2919
2920bad_unshare_out:
2921 return err;
2922}
2923
2924SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2925{
2926 return ksys_unshare(unshare_flags);
2927}
2928
2929/*
2930 * Helper to unshare the files of the current task.
2931 * We don't want to expose copy_files internals to
2932 * the exec layer of the kernel.
2933 */
2934
2935int unshare_files(struct files_struct **displaced)
2936{
2937 struct task_struct *task = current;
2938 struct files_struct *copy = NULL;
2939 int error;
2940
2941 error = unshare_fd(CLONE_FILES, ©);
2942 if (error || !copy) {
2943 *displaced = NULL;
2944 return error;
2945 }
2946 *displaced = task->files;
2947 task_lock(task);
2948 task->files = copy;
2949 task_unlock(task);
2950 return 0;
2951}
2952
2953int sysctl_max_threads(struct ctl_table *table, int write,
2954 void __user *buffer, size_t *lenp, loff_t *ppos)
2955{
2956 struct ctl_table t;
2957 int ret;
2958 int threads = max_threads;
2959 int min = 1;
2960 int max = MAX_THREADS;
2961
2962 t = *table;
2963 t.data = &threads;
2964 t.extra1 = &min;
2965 t.extra2 = &max;
2966
2967 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2968 if (ret || !write)
2969 return ret;
2970
2971 max_threads = threads;
2972
2973 return 0;
2974}
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/fs.h>
44#include <linux/mm.h>
45#include <linux/vmacache.h>
46#include <linux/nsproxy.h>
47#include <linux/capability.h>
48#include <linux/cpu.h>
49#include <linux/cgroup.h>
50#include <linux/security.h>
51#include <linux/hugetlb.h>
52#include <linux/seccomp.h>
53#include <linux/swap.h>
54#include <linux/syscalls.h>
55#include <linux/jiffies.h>
56#include <linux/futex.h>
57#include <linux/compat.h>
58#include <linux/kthread.h>
59#include <linux/task_io_accounting_ops.h>
60#include <linux/rcupdate.h>
61#include <linux/ptrace.h>
62#include <linux/mount.h>
63#include <linux/audit.h>
64#include <linux/memcontrol.h>
65#include <linux/ftrace.h>
66#include <linux/proc_fs.h>
67#include <linux/profile.h>
68#include <linux/rmap.h>
69#include <linux/ksm.h>
70#include <linux/acct.h>
71#include <linux/userfaultfd_k.h>
72#include <linux/tsacct_kern.h>
73#include <linux/cn_proc.h>
74#include <linux/freezer.h>
75#include <linux/delayacct.h>
76#include <linux/taskstats_kern.h>
77#include <linux/random.h>
78#include <linux/tty.h>
79#include <linux/blkdev.h>
80#include <linux/fs_struct.h>
81#include <linux/magic.h>
82#include <linux/perf_event.h>
83#include <linux/posix-timers.h>
84#include <linux/user-return-notifier.h>
85#include <linux/oom.h>
86#include <linux/khugepaged.h>
87#include <linux/signalfd.h>
88#include <linux/uprobes.h>
89#include <linux/aio.h>
90#include <linux/compiler.h>
91#include <linux/sysctl.h>
92#include <linux/kcov.h>
93#include <linux/livepatch.h>
94#include <linux/thread_info.h>
95#include <linux/stackleak.h>
96#include <linux/kasan.h>
97#include <linux/scs.h>
98#include <linux/io_uring.h>
99#include <linux/bpf.h>
100
101#include <asm/pgalloc.h>
102#include <linux/uaccess.h>
103#include <asm/mmu_context.h>
104#include <asm/cacheflush.h>
105#include <asm/tlbflush.h>
106
107#include <trace/events/sched.h>
108
109#define CREATE_TRACE_POINTS
110#include <trace/events/task.h>
111
112/*
113 * Minimum number of threads to boot the kernel
114 */
115#define MIN_THREADS 20
116
117/*
118 * Maximum number of threads
119 */
120#define MAX_THREADS FUTEX_TID_MASK
121
122/*
123 * Protected counters by write_lock_irq(&tasklist_lock)
124 */
125unsigned long total_forks; /* Handle normal Linux uptimes. */
126int nr_threads; /* The idle threads do not count.. */
127
128static int max_threads; /* tunable limit on nr_threads */
129
130#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
131
132static const char * const resident_page_types[] = {
133 NAMED_ARRAY_INDEX(MM_FILEPAGES),
134 NAMED_ARRAY_INDEX(MM_ANONPAGES),
135 NAMED_ARRAY_INDEX(MM_SWAPENTS),
136 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
137};
138
139DEFINE_PER_CPU(unsigned long, process_counts) = 0;
140
141__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
142
143#ifdef CONFIG_PROVE_RCU
144int lockdep_tasklist_lock_is_held(void)
145{
146 return lockdep_is_held(&tasklist_lock);
147}
148EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
149#endif /* #ifdef CONFIG_PROVE_RCU */
150
151int nr_processes(void)
152{
153 int cpu;
154 int total = 0;
155
156 for_each_possible_cpu(cpu)
157 total += per_cpu(process_counts, cpu);
158
159 return total;
160}
161
162void __weak arch_release_task_struct(struct task_struct *tsk)
163{
164}
165
166#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
167static struct kmem_cache *task_struct_cachep;
168
169static inline struct task_struct *alloc_task_struct_node(int node)
170{
171 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
172}
173
174static inline void free_task_struct(struct task_struct *tsk)
175{
176 kmem_cache_free(task_struct_cachep, tsk);
177}
178#endif
179
180#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181
182/*
183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
184 * kmemcache based allocator.
185 */
186# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
187
188#ifdef CONFIG_VMAP_STACK
189/*
190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
191 * flush. Try to minimize the number of calls by caching stacks.
192 */
193#define NR_CACHED_STACKS 2
194static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
195
196static int free_vm_stack_cache(unsigned int cpu)
197{
198 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
199 int i;
200
201 for (i = 0; i < NR_CACHED_STACKS; i++) {
202 struct vm_struct *vm_stack = cached_vm_stacks[i];
203
204 if (!vm_stack)
205 continue;
206
207 vfree(vm_stack->addr);
208 cached_vm_stacks[i] = NULL;
209 }
210
211 return 0;
212}
213#endif
214
215static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
216{
217#ifdef CONFIG_VMAP_STACK
218 void *stack;
219 int i;
220
221 for (i = 0; i < NR_CACHED_STACKS; i++) {
222 struct vm_struct *s;
223
224 s = this_cpu_xchg(cached_stacks[i], NULL);
225
226 if (!s)
227 continue;
228
229 /* Mark stack accessible for KASAN. */
230 kasan_unpoison_range(s->addr, THREAD_SIZE);
231
232 /* Clear stale pointers from reused stack. */
233 memset(s->addr, 0, THREAD_SIZE);
234
235 tsk->stack_vm_area = s;
236 tsk->stack = s->addr;
237 return s->addr;
238 }
239
240 /*
241 * Allocated stacks are cached and later reused by new threads,
242 * so memcg accounting is performed manually on assigning/releasing
243 * stacks to tasks. Drop __GFP_ACCOUNT.
244 */
245 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
246 VMALLOC_START, VMALLOC_END,
247 THREADINFO_GFP & ~__GFP_ACCOUNT,
248 PAGE_KERNEL,
249 0, node, __builtin_return_address(0));
250
251 /*
252 * We can't call find_vm_area() in interrupt context, and
253 * free_thread_stack() can be called in interrupt context,
254 * so cache the vm_struct.
255 */
256 if (stack) {
257 tsk->stack_vm_area = find_vm_area(stack);
258 tsk->stack = stack;
259 }
260 return stack;
261#else
262 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
263 THREAD_SIZE_ORDER);
264
265 if (likely(page)) {
266 tsk->stack = kasan_reset_tag(page_address(page));
267 return tsk->stack;
268 }
269 return NULL;
270#endif
271}
272
273static inline void free_thread_stack(struct task_struct *tsk)
274{
275#ifdef CONFIG_VMAP_STACK
276 struct vm_struct *vm = task_stack_vm_area(tsk);
277
278 if (vm) {
279 int i;
280
281 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
282 memcg_kmem_uncharge_page(vm->pages[i], 0);
283
284 for (i = 0; i < NR_CACHED_STACKS; i++) {
285 if (this_cpu_cmpxchg(cached_stacks[i],
286 NULL, tsk->stack_vm_area) != NULL)
287 continue;
288
289 return;
290 }
291
292 vfree_atomic(tsk->stack);
293 return;
294 }
295#endif
296
297 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
298}
299# else
300static struct kmem_cache *thread_stack_cache;
301
302static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
303 int node)
304{
305 unsigned long *stack;
306 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
307 stack = kasan_reset_tag(stack);
308 tsk->stack = stack;
309 return stack;
310}
311
312static void free_thread_stack(struct task_struct *tsk)
313{
314 kmem_cache_free(thread_stack_cache, tsk->stack);
315}
316
317void thread_stack_cache_init(void)
318{
319 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
320 THREAD_SIZE, THREAD_SIZE, 0, 0,
321 THREAD_SIZE, NULL);
322 BUG_ON(thread_stack_cache == NULL);
323}
324# endif
325#endif
326
327/* SLAB cache for signal_struct structures (tsk->signal) */
328static struct kmem_cache *signal_cachep;
329
330/* SLAB cache for sighand_struct structures (tsk->sighand) */
331struct kmem_cache *sighand_cachep;
332
333/* SLAB cache for files_struct structures (tsk->files) */
334struct kmem_cache *files_cachep;
335
336/* SLAB cache for fs_struct structures (tsk->fs) */
337struct kmem_cache *fs_cachep;
338
339/* SLAB cache for vm_area_struct structures */
340static struct kmem_cache *vm_area_cachep;
341
342/* SLAB cache for mm_struct structures (tsk->mm) */
343static struct kmem_cache *mm_cachep;
344
345struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
346{
347 struct vm_area_struct *vma;
348
349 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
350 if (vma)
351 vma_init(vma, mm);
352 return vma;
353}
354
355struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
356{
357 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
358
359 if (new) {
360 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
361 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
362 /*
363 * orig->shared.rb may be modified concurrently, but the clone
364 * will be reinitialized.
365 */
366 *new = data_race(*orig);
367 INIT_LIST_HEAD(&new->anon_vma_chain);
368 new->vm_next = new->vm_prev = NULL;
369 }
370 return new;
371}
372
373void vm_area_free(struct vm_area_struct *vma)
374{
375 kmem_cache_free(vm_area_cachep, vma);
376}
377
378static void account_kernel_stack(struct task_struct *tsk, int account)
379{
380 void *stack = task_stack_page(tsk);
381 struct vm_struct *vm = task_stack_vm_area(tsk);
382
383 if (vm) {
384 int i;
385
386 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
387 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
388 account * (PAGE_SIZE / 1024));
389 } else {
390 /* All stack pages are in the same node. */
391 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
392 account * (THREAD_SIZE / 1024));
393 }
394}
395
396static int memcg_charge_kernel_stack(struct task_struct *tsk)
397{
398#ifdef CONFIG_VMAP_STACK
399 struct vm_struct *vm = task_stack_vm_area(tsk);
400 int ret;
401
402 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
403
404 if (vm) {
405 int i;
406
407 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
408
409 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
410 /*
411 * If memcg_kmem_charge_page() fails, page's
412 * memory cgroup pointer is NULL, and
413 * memcg_kmem_uncharge_page() in free_thread_stack()
414 * will ignore this page.
415 */
416 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
417 0);
418 if (ret)
419 return ret;
420 }
421 }
422#endif
423 return 0;
424}
425
426static void release_task_stack(struct task_struct *tsk)
427{
428 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
429 return; /* Better to leak the stack than to free prematurely */
430
431 account_kernel_stack(tsk, -1);
432 free_thread_stack(tsk);
433 tsk->stack = NULL;
434#ifdef CONFIG_VMAP_STACK
435 tsk->stack_vm_area = NULL;
436#endif
437}
438
439#ifdef CONFIG_THREAD_INFO_IN_TASK
440void put_task_stack(struct task_struct *tsk)
441{
442 if (refcount_dec_and_test(&tsk->stack_refcount))
443 release_task_stack(tsk);
444}
445#endif
446
447void free_task(struct task_struct *tsk)
448{
449 scs_release(tsk);
450
451#ifndef CONFIG_THREAD_INFO_IN_TASK
452 /*
453 * The task is finally done with both the stack and thread_info,
454 * so free both.
455 */
456 release_task_stack(tsk);
457#else
458 /*
459 * If the task had a separate stack allocation, it should be gone
460 * by now.
461 */
462 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
463#endif
464 rt_mutex_debug_task_free(tsk);
465 ftrace_graph_exit_task(tsk);
466 arch_release_task_struct(tsk);
467 if (tsk->flags & PF_KTHREAD)
468 free_kthread_struct(tsk);
469 free_task_struct(tsk);
470}
471EXPORT_SYMBOL(free_task);
472
473#ifdef CONFIG_MMU
474static __latent_entropy int dup_mmap(struct mm_struct *mm,
475 struct mm_struct *oldmm)
476{
477 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
478 struct rb_node **rb_link, *rb_parent;
479 int retval;
480 unsigned long charge;
481 LIST_HEAD(uf);
482
483 uprobe_start_dup_mmap();
484 if (mmap_write_lock_killable(oldmm)) {
485 retval = -EINTR;
486 goto fail_uprobe_end;
487 }
488 flush_cache_dup_mm(oldmm);
489 uprobe_dup_mmap(oldmm, mm);
490 /*
491 * Not linked in yet - no deadlock potential:
492 */
493 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
494
495 /* No ordering required: file already has been exposed. */
496 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
497
498 mm->total_vm = oldmm->total_vm;
499 mm->data_vm = oldmm->data_vm;
500 mm->exec_vm = oldmm->exec_vm;
501 mm->stack_vm = oldmm->stack_vm;
502
503 rb_link = &mm->mm_rb.rb_node;
504 rb_parent = NULL;
505 pprev = &mm->mmap;
506 retval = ksm_fork(mm, oldmm);
507 if (retval)
508 goto out;
509 retval = khugepaged_fork(mm, oldmm);
510 if (retval)
511 goto out;
512
513 prev = NULL;
514 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
515 struct file *file;
516
517 if (mpnt->vm_flags & VM_DONTCOPY) {
518 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
519 continue;
520 }
521 charge = 0;
522 /*
523 * Don't duplicate many vmas if we've been oom-killed (for
524 * example)
525 */
526 if (fatal_signal_pending(current)) {
527 retval = -EINTR;
528 goto out;
529 }
530 if (mpnt->vm_flags & VM_ACCOUNT) {
531 unsigned long len = vma_pages(mpnt);
532
533 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
534 goto fail_nomem;
535 charge = len;
536 }
537 tmp = vm_area_dup(mpnt);
538 if (!tmp)
539 goto fail_nomem;
540 retval = vma_dup_policy(mpnt, tmp);
541 if (retval)
542 goto fail_nomem_policy;
543 tmp->vm_mm = mm;
544 retval = dup_userfaultfd(tmp, &uf);
545 if (retval)
546 goto fail_nomem_anon_vma_fork;
547 if (tmp->vm_flags & VM_WIPEONFORK) {
548 /*
549 * VM_WIPEONFORK gets a clean slate in the child.
550 * Don't prepare anon_vma until fault since we don't
551 * copy page for current vma.
552 */
553 tmp->anon_vma = NULL;
554 } else if (anon_vma_fork(tmp, mpnt))
555 goto fail_nomem_anon_vma_fork;
556 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
557 file = tmp->vm_file;
558 if (file) {
559 struct inode *inode = file_inode(file);
560 struct address_space *mapping = file->f_mapping;
561
562 get_file(file);
563 if (tmp->vm_flags & VM_DENYWRITE)
564 put_write_access(inode);
565 i_mmap_lock_write(mapping);
566 if (tmp->vm_flags & VM_SHARED)
567 mapping_allow_writable(mapping);
568 flush_dcache_mmap_lock(mapping);
569 /* insert tmp into the share list, just after mpnt */
570 vma_interval_tree_insert_after(tmp, mpnt,
571 &mapping->i_mmap);
572 flush_dcache_mmap_unlock(mapping);
573 i_mmap_unlock_write(mapping);
574 }
575
576 /*
577 * Clear hugetlb-related page reserves for children. This only
578 * affects MAP_PRIVATE mappings. Faults generated by the child
579 * are not guaranteed to succeed, even if read-only
580 */
581 if (is_vm_hugetlb_page(tmp))
582 reset_vma_resv_huge_pages(tmp);
583
584 /*
585 * Link in the new vma and copy the page table entries.
586 */
587 *pprev = tmp;
588 pprev = &tmp->vm_next;
589 tmp->vm_prev = prev;
590 prev = tmp;
591
592 __vma_link_rb(mm, tmp, rb_link, rb_parent);
593 rb_link = &tmp->vm_rb.rb_right;
594 rb_parent = &tmp->vm_rb;
595
596 mm->map_count++;
597 if (!(tmp->vm_flags & VM_WIPEONFORK))
598 retval = copy_page_range(tmp, mpnt);
599
600 if (tmp->vm_ops && tmp->vm_ops->open)
601 tmp->vm_ops->open(tmp);
602
603 if (retval)
604 goto out;
605 }
606 /* a new mm has just been created */
607 retval = arch_dup_mmap(oldmm, mm);
608out:
609 mmap_write_unlock(mm);
610 flush_tlb_mm(oldmm);
611 mmap_write_unlock(oldmm);
612 dup_userfaultfd_complete(&uf);
613fail_uprobe_end:
614 uprobe_end_dup_mmap();
615 return retval;
616fail_nomem_anon_vma_fork:
617 mpol_put(vma_policy(tmp));
618fail_nomem_policy:
619 vm_area_free(tmp);
620fail_nomem:
621 retval = -ENOMEM;
622 vm_unacct_memory(charge);
623 goto out;
624}
625
626static inline int mm_alloc_pgd(struct mm_struct *mm)
627{
628 mm->pgd = pgd_alloc(mm);
629 if (unlikely(!mm->pgd))
630 return -ENOMEM;
631 return 0;
632}
633
634static inline void mm_free_pgd(struct mm_struct *mm)
635{
636 pgd_free(mm, mm->pgd);
637}
638#else
639static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
640{
641 mmap_write_lock(oldmm);
642 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
643 mmap_write_unlock(oldmm);
644 return 0;
645}
646#define mm_alloc_pgd(mm) (0)
647#define mm_free_pgd(mm)
648#endif /* CONFIG_MMU */
649
650static void check_mm(struct mm_struct *mm)
651{
652 int i;
653
654 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
655 "Please make sure 'struct resident_page_types[]' is updated as well");
656
657 for (i = 0; i < NR_MM_COUNTERS; i++) {
658 long x = atomic_long_read(&mm->rss_stat.count[i]);
659
660 if (unlikely(x))
661 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
662 mm, resident_page_types[i], x);
663 }
664
665 if (mm_pgtables_bytes(mm))
666 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
667 mm_pgtables_bytes(mm));
668
669#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
670 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
671#endif
672}
673
674#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
675#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
676
677/*
678 * Called when the last reference to the mm
679 * is dropped: either by a lazy thread or by
680 * mmput. Free the page directory and the mm.
681 */
682void __mmdrop(struct mm_struct *mm)
683{
684 BUG_ON(mm == &init_mm);
685 WARN_ON_ONCE(mm == current->mm);
686 WARN_ON_ONCE(mm == current->active_mm);
687 mm_free_pgd(mm);
688 destroy_context(mm);
689 mmu_notifier_subscriptions_destroy(mm);
690 check_mm(mm);
691 put_user_ns(mm->user_ns);
692 free_mm(mm);
693}
694EXPORT_SYMBOL_GPL(__mmdrop);
695
696static void mmdrop_async_fn(struct work_struct *work)
697{
698 struct mm_struct *mm;
699
700 mm = container_of(work, struct mm_struct, async_put_work);
701 __mmdrop(mm);
702}
703
704static void mmdrop_async(struct mm_struct *mm)
705{
706 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
707 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
708 schedule_work(&mm->async_put_work);
709 }
710}
711
712static inline void free_signal_struct(struct signal_struct *sig)
713{
714 taskstats_tgid_free(sig);
715 sched_autogroup_exit(sig);
716 /*
717 * __mmdrop is not safe to call from softirq context on x86 due to
718 * pgd_dtor so postpone it to the async context
719 */
720 if (sig->oom_mm)
721 mmdrop_async(sig->oom_mm);
722 kmem_cache_free(signal_cachep, sig);
723}
724
725static inline void put_signal_struct(struct signal_struct *sig)
726{
727 if (refcount_dec_and_test(&sig->sigcnt))
728 free_signal_struct(sig);
729}
730
731void __put_task_struct(struct task_struct *tsk)
732{
733 WARN_ON(!tsk->exit_state);
734 WARN_ON(refcount_read(&tsk->usage));
735 WARN_ON(tsk == current);
736
737 io_uring_free(tsk);
738 cgroup_free(tsk);
739 task_numa_free(tsk, true);
740 security_task_free(tsk);
741 bpf_task_storage_free(tsk);
742 exit_creds(tsk);
743 delayacct_tsk_free(tsk);
744 put_signal_struct(tsk->signal);
745 sched_core_free(tsk);
746
747 if (!profile_handoff_task(tsk))
748 free_task(tsk);
749}
750EXPORT_SYMBOL_GPL(__put_task_struct);
751
752void __init __weak arch_task_cache_init(void) { }
753
754/*
755 * set_max_threads
756 */
757static void set_max_threads(unsigned int max_threads_suggested)
758{
759 u64 threads;
760 unsigned long nr_pages = totalram_pages();
761
762 /*
763 * The number of threads shall be limited such that the thread
764 * structures may only consume a small part of the available memory.
765 */
766 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
767 threads = MAX_THREADS;
768 else
769 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
770 (u64) THREAD_SIZE * 8UL);
771
772 if (threads > max_threads_suggested)
773 threads = max_threads_suggested;
774
775 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
776}
777
778#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
779/* Initialized by the architecture: */
780int arch_task_struct_size __read_mostly;
781#endif
782
783#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
784static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
785{
786 /* Fetch thread_struct whitelist for the architecture. */
787 arch_thread_struct_whitelist(offset, size);
788
789 /*
790 * Handle zero-sized whitelist or empty thread_struct, otherwise
791 * adjust offset to position of thread_struct in task_struct.
792 */
793 if (unlikely(*size == 0))
794 *offset = 0;
795 else
796 *offset += offsetof(struct task_struct, thread);
797}
798#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
799
800void __init fork_init(void)
801{
802 int i;
803#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
804#ifndef ARCH_MIN_TASKALIGN
805#define ARCH_MIN_TASKALIGN 0
806#endif
807 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
808 unsigned long useroffset, usersize;
809
810 /* create a slab on which task_structs can be allocated */
811 task_struct_whitelist(&useroffset, &usersize);
812 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
813 arch_task_struct_size, align,
814 SLAB_PANIC|SLAB_ACCOUNT,
815 useroffset, usersize, NULL);
816#endif
817
818 /* do the arch specific task caches init */
819 arch_task_cache_init();
820
821 set_max_threads(MAX_THREADS);
822
823 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
824 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
825 init_task.signal->rlim[RLIMIT_SIGPENDING] =
826 init_task.signal->rlim[RLIMIT_NPROC];
827
828 for (i = 0; i < MAX_PER_NAMESPACE_UCOUNTS; i++)
829 init_user_ns.ucount_max[i] = max_threads/2;
830
831 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
832 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
833 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
834 set_rlimit_ucount_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
835
836#ifdef CONFIG_VMAP_STACK
837 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
838 NULL, free_vm_stack_cache);
839#endif
840
841 scs_init();
842
843 lockdep_init_task(&init_task);
844 uprobes_init();
845}
846
847int __weak arch_dup_task_struct(struct task_struct *dst,
848 struct task_struct *src)
849{
850 *dst = *src;
851 return 0;
852}
853
854void set_task_stack_end_magic(struct task_struct *tsk)
855{
856 unsigned long *stackend;
857
858 stackend = end_of_stack(tsk);
859 *stackend = STACK_END_MAGIC; /* for overflow detection */
860}
861
862static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
863{
864 struct task_struct *tsk;
865 unsigned long *stack;
866 struct vm_struct *stack_vm_area __maybe_unused;
867 int err;
868
869 if (node == NUMA_NO_NODE)
870 node = tsk_fork_get_node(orig);
871 tsk = alloc_task_struct_node(node);
872 if (!tsk)
873 return NULL;
874
875 stack = alloc_thread_stack_node(tsk, node);
876 if (!stack)
877 goto free_tsk;
878
879 if (memcg_charge_kernel_stack(tsk))
880 goto free_stack;
881
882 stack_vm_area = task_stack_vm_area(tsk);
883
884 err = arch_dup_task_struct(tsk, orig);
885
886 /*
887 * arch_dup_task_struct() clobbers the stack-related fields. Make
888 * sure they're properly initialized before using any stack-related
889 * functions again.
890 */
891 tsk->stack = stack;
892#ifdef CONFIG_VMAP_STACK
893 tsk->stack_vm_area = stack_vm_area;
894#endif
895#ifdef CONFIG_THREAD_INFO_IN_TASK
896 refcount_set(&tsk->stack_refcount, 1);
897#endif
898
899 if (err)
900 goto free_stack;
901
902 err = scs_prepare(tsk, node);
903 if (err)
904 goto free_stack;
905
906#ifdef CONFIG_SECCOMP
907 /*
908 * We must handle setting up seccomp filters once we're under
909 * the sighand lock in case orig has changed between now and
910 * then. Until then, filter must be NULL to avoid messing up
911 * the usage counts on the error path calling free_task.
912 */
913 tsk->seccomp.filter = NULL;
914#endif
915
916 setup_thread_stack(tsk, orig);
917 clear_user_return_notifier(tsk);
918 clear_tsk_need_resched(tsk);
919 set_task_stack_end_magic(tsk);
920 clear_syscall_work_syscall_user_dispatch(tsk);
921
922#ifdef CONFIG_STACKPROTECTOR
923 tsk->stack_canary = get_random_canary();
924#endif
925 if (orig->cpus_ptr == &orig->cpus_mask)
926 tsk->cpus_ptr = &tsk->cpus_mask;
927
928 /*
929 * One for the user space visible state that goes away when reaped.
930 * One for the scheduler.
931 */
932 refcount_set(&tsk->rcu_users, 2);
933 /* One for the rcu users */
934 refcount_set(&tsk->usage, 1);
935#ifdef CONFIG_BLK_DEV_IO_TRACE
936 tsk->btrace_seq = 0;
937#endif
938 tsk->splice_pipe = NULL;
939 tsk->task_frag.page = NULL;
940 tsk->wake_q.next = NULL;
941 tsk->pf_io_worker = NULL;
942
943 account_kernel_stack(tsk, 1);
944
945 kcov_task_init(tsk);
946 kmap_local_fork(tsk);
947
948#ifdef CONFIG_FAULT_INJECTION
949 tsk->fail_nth = 0;
950#endif
951
952#ifdef CONFIG_BLK_CGROUP
953 tsk->throttle_queue = NULL;
954 tsk->use_memdelay = 0;
955#endif
956
957#ifdef CONFIG_MEMCG
958 tsk->active_memcg = NULL;
959#endif
960 return tsk;
961
962free_stack:
963 free_thread_stack(tsk);
964free_tsk:
965 free_task_struct(tsk);
966 return NULL;
967}
968
969__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
970
971static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
972
973static int __init coredump_filter_setup(char *s)
974{
975 default_dump_filter =
976 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
977 MMF_DUMP_FILTER_MASK;
978 return 1;
979}
980
981__setup("coredump_filter=", coredump_filter_setup);
982
983#include <linux/init_task.h>
984
985static void mm_init_aio(struct mm_struct *mm)
986{
987#ifdef CONFIG_AIO
988 spin_lock_init(&mm->ioctx_lock);
989 mm->ioctx_table = NULL;
990#endif
991}
992
993static __always_inline void mm_clear_owner(struct mm_struct *mm,
994 struct task_struct *p)
995{
996#ifdef CONFIG_MEMCG
997 if (mm->owner == p)
998 WRITE_ONCE(mm->owner, NULL);
999#endif
1000}
1001
1002static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1003{
1004#ifdef CONFIG_MEMCG
1005 mm->owner = p;
1006#endif
1007}
1008
1009static void mm_init_pasid(struct mm_struct *mm)
1010{
1011#ifdef CONFIG_IOMMU_SUPPORT
1012 mm->pasid = INIT_PASID;
1013#endif
1014}
1015
1016static void mm_init_uprobes_state(struct mm_struct *mm)
1017{
1018#ifdef CONFIG_UPROBES
1019 mm->uprobes_state.xol_area = NULL;
1020#endif
1021}
1022
1023static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1024 struct user_namespace *user_ns)
1025{
1026 mm->mmap = NULL;
1027 mm->mm_rb = RB_ROOT;
1028 mm->vmacache_seqnum = 0;
1029 atomic_set(&mm->mm_users, 1);
1030 atomic_set(&mm->mm_count, 1);
1031 seqcount_init(&mm->write_protect_seq);
1032 mmap_init_lock(mm);
1033 INIT_LIST_HEAD(&mm->mmlist);
1034 mm->core_state = NULL;
1035 mm_pgtables_bytes_init(mm);
1036 mm->map_count = 0;
1037 mm->locked_vm = 0;
1038 atomic64_set(&mm->pinned_vm, 0);
1039 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1040 spin_lock_init(&mm->page_table_lock);
1041 spin_lock_init(&mm->arg_lock);
1042 mm_init_cpumask(mm);
1043 mm_init_aio(mm);
1044 mm_init_owner(mm, p);
1045 mm_init_pasid(mm);
1046 RCU_INIT_POINTER(mm->exe_file, NULL);
1047 mmu_notifier_subscriptions_init(mm);
1048 init_tlb_flush_pending(mm);
1049#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1050 mm->pmd_huge_pte = NULL;
1051#endif
1052 mm_init_uprobes_state(mm);
1053 hugetlb_count_init(mm);
1054
1055 if (current->mm) {
1056 mm->flags = current->mm->flags & MMF_INIT_MASK;
1057 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1058 } else {
1059 mm->flags = default_dump_filter;
1060 mm->def_flags = 0;
1061 }
1062
1063 if (mm_alloc_pgd(mm))
1064 goto fail_nopgd;
1065
1066 if (init_new_context(p, mm))
1067 goto fail_nocontext;
1068
1069 mm->user_ns = get_user_ns(user_ns);
1070 return mm;
1071
1072fail_nocontext:
1073 mm_free_pgd(mm);
1074fail_nopgd:
1075 free_mm(mm);
1076 return NULL;
1077}
1078
1079/*
1080 * Allocate and initialize an mm_struct.
1081 */
1082struct mm_struct *mm_alloc(void)
1083{
1084 struct mm_struct *mm;
1085
1086 mm = allocate_mm();
1087 if (!mm)
1088 return NULL;
1089
1090 memset(mm, 0, sizeof(*mm));
1091 return mm_init(mm, current, current_user_ns());
1092}
1093
1094static inline void __mmput(struct mm_struct *mm)
1095{
1096 VM_BUG_ON(atomic_read(&mm->mm_users));
1097
1098 uprobe_clear_state(mm);
1099 exit_aio(mm);
1100 ksm_exit(mm);
1101 khugepaged_exit(mm); /* must run before exit_mmap */
1102 exit_mmap(mm);
1103 mm_put_huge_zero_page(mm);
1104 set_mm_exe_file(mm, NULL);
1105 if (!list_empty(&mm->mmlist)) {
1106 spin_lock(&mmlist_lock);
1107 list_del(&mm->mmlist);
1108 spin_unlock(&mmlist_lock);
1109 }
1110 if (mm->binfmt)
1111 module_put(mm->binfmt->module);
1112 mmdrop(mm);
1113}
1114
1115/*
1116 * Decrement the use count and release all resources for an mm.
1117 */
1118void mmput(struct mm_struct *mm)
1119{
1120 might_sleep();
1121
1122 if (atomic_dec_and_test(&mm->mm_users))
1123 __mmput(mm);
1124}
1125EXPORT_SYMBOL_GPL(mmput);
1126
1127#ifdef CONFIG_MMU
1128static void mmput_async_fn(struct work_struct *work)
1129{
1130 struct mm_struct *mm = container_of(work, struct mm_struct,
1131 async_put_work);
1132
1133 __mmput(mm);
1134}
1135
1136void mmput_async(struct mm_struct *mm)
1137{
1138 if (atomic_dec_and_test(&mm->mm_users)) {
1139 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1140 schedule_work(&mm->async_put_work);
1141 }
1142}
1143#endif
1144
1145/**
1146 * set_mm_exe_file - change a reference to the mm's executable file
1147 *
1148 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1149 *
1150 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1151 * invocations: in mmput() nobody alive left, in execve task is single
1152 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1153 * mm->exe_file, but does so without using set_mm_exe_file() in order
1154 * to avoid the need for any locks.
1155 */
1156void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1157{
1158 struct file *old_exe_file;
1159
1160 /*
1161 * It is safe to dereference the exe_file without RCU as
1162 * this function is only called if nobody else can access
1163 * this mm -- see comment above for justification.
1164 */
1165 old_exe_file = rcu_dereference_raw(mm->exe_file);
1166
1167 if (new_exe_file)
1168 get_file(new_exe_file);
1169 rcu_assign_pointer(mm->exe_file, new_exe_file);
1170 if (old_exe_file)
1171 fput(old_exe_file);
1172}
1173
1174/**
1175 * get_mm_exe_file - acquire a reference to the mm's executable file
1176 *
1177 * Returns %NULL if mm has no associated executable file.
1178 * User must release file via fput().
1179 */
1180struct file *get_mm_exe_file(struct mm_struct *mm)
1181{
1182 struct file *exe_file;
1183
1184 rcu_read_lock();
1185 exe_file = rcu_dereference(mm->exe_file);
1186 if (exe_file && !get_file_rcu(exe_file))
1187 exe_file = NULL;
1188 rcu_read_unlock();
1189 return exe_file;
1190}
1191EXPORT_SYMBOL(get_mm_exe_file);
1192
1193/**
1194 * get_task_exe_file - acquire a reference to the task's executable file
1195 *
1196 * Returns %NULL if task's mm (if any) has no associated executable file or
1197 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1198 * User must release file via fput().
1199 */
1200struct file *get_task_exe_file(struct task_struct *task)
1201{
1202 struct file *exe_file = NULL;
1203 struct mm_struct *mm;
1204
1205 task_lock(task);
1206 mm = task->mm;
1207 if (mm) {
1208 if (!(task->flags & PF_KTHREAD))
1209 exe_file = get_mm_exe_file(mm);
1210 }
1211 task_unlock(task);
1212 return exe_file;
1213}
1214EXPORT_SYMBOL(get_task_exe_file);
1215
1216/**
1217 * get_task_mm - acquire a reference to the task's mm
1218 *
1219 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1220 * this kernel workthread has transiently adopted a user mm with use_mm,
1221 * to do its AIO) is not set and if so returns a reference to it, after
1222 * bumping up the use count. User must release the mm via mmput()
1223 * after use. Typically used by /proc and ptrace.
1224 */
1225struct mm_struct *get_task_mm(struct task_struct *task)
1226{
1227 struct mm_struct *mm;
1228
1229 task_lock(task);
1230 mm = task->mm;
1231 if (mm) {
1232 if (task->flags & PF_KTHREAD)
1233 mm = NULL;
1234 else
1235 mmget(mm);
1236 }
1237 task_unlock(task);
1238 return mm;
1239}
1240EXPORT_SYMBOL_GPL(get_task_mm);
1241
1242struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1243{
1244 struct mm_struct *mm;
1245 int err;
1246
1247 err = down_read_killable(&task->signal->exec_update_lock);
1248 if (err)
1249 return ERR_PTR(err);
1250
1251 mm = get_task_mm(task);
1252 if (mm && mm != current->mm &&
1253 !ptrace_may_access(task, mode)) {
1254 mmput(mm);
1255 mm = ERR_PTR(-EACCES);
1256 }
1257 up_read(&task->signal->exec_update_lock);
1258
1259 return mm;
1260}
1261
1262static void complete_vfork_done(struct task_struct *tsk)
1263{
1264 struct completion *vfork;
1265
1266 task_lock(tsk);
1267 vfork = tsk->vfork_done;
1268 if (likely(vfork)) {
1269 tsk->vfork_done = NULL;
1270 complete(vfork);
1271 }
1272 task_unlock(tsk);
1273}
1274
1275static int wait_for_vfork_done(struct task_struct *child,
1276 struct completion *vfork)
1277{
1278 int killed;
1279
1280 freezer_do_not_count();
1281 cgroup_enter_frozen();
1282 killed = wait_for_completion_killable(vfork);
1283 cgroup_leave_frozen(false);
1284 freezer_count();
1285
1286 if (killed) {
1287 task_lock(child);
1288 child->vfork_done = NULL;
1289 task_unlock(child);
1290 }
1291
1292 put_task_struct(child);
1293 return killed;
1294}
1295
1296/* Please note the differences between mmput and mm_release.
1297 * mmput is called whenever we stop holding onto a mm_struct,
1298 * error success whatever.
1299 *
1300 * mm_release is called after a mm_struct has been removed
1301 * from the current process.
1302 *
1303 * This difference is important for error handling, when we
1304 * only half set up a mm_struct for a new process and need to restore
1305 * the old one. Because we mmput the new mm_struct before
1306 * restoring the old one. . .
1307 * Eric Biederman 10 January 1998
1308 */
1309static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1310{
1311 uprobe_free_utask(tsk);
1312
1313 /* Get rid of any cached register state */
1314 deactivate_mm(tsk, mm);
1315
1316 /*
1317 * Signal userspace if we're not exiting with a core dump
1318 * because we want to leave the value intact for debugging
1319 * purposes.
1320 */
1321 if (tsk->clear_child_tid) {
1322 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1323 atomic_read(&mm->mm_users) > 1) {
1324 /*
1325 * We don't check the error code - if userspace has
1326 * not set up a proper pointer then tough luck.
1327 */
1328 put_user(0, tsk->clear_child_tid);
1329 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1330 1, NULL, NULL, 0, 0);
1331 }
1332 tsk->clear_child_tid = NULL;
1333 }
1334
1335 /*
1336 * All done, finally we can wake up parent and return this mm to him.
1337 * Also kthread_stop() uses this completion for synchronization.
1338 */
1339 if (tsk->vfork_done)
1340 complete_vfork_done(tsk);
1341}
1342
1343void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1344{
1345 futex_exit_release(tsk);
1346 mm_release(tsk, mm);
1347}
1348
1349void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1350{
1351 futex_exec_release(tsk);
1352 mm_release(tsk, mm);
1353}
1354
1355/**
1356 * dup_mm() - duplicates an existing mm structure
1357 * @tsk: the task_struct with which the new mm will be associated.
1358 * @oldmm: the mm to duplicate.
1359 *
1360 * Allocates a new mm structure and duplicates the provided @oldmm structure
1361 * content into it.
1362 *
1363 * Return: the duplicated mm or NULL on failure.
1364 */
1365static struct mm_struct *dup_mm(struct task_struct *tsk,
1366 struct mm_struct *oldmm)
1367{
1368 struct mm_struct *mm;
1369 int err;
1370
1371 mm = allocate_mm();
1372 if (!mm)
1373 goto fail_nomem;
1374
1375 memcpy(mm, oldmm, sizeof(*mm));
1376
1377 if (!mm_init(mm, tsk, mm->user_ns))
1378 goto fail_nomem;
1379
1380 err = dup_mmap(mm, oldmm);
1381 if (err)
1382 goto free_pt;
1383
1384 mm->hiwater_rss = get_mm_rss(mm);
1385 mm->hiwater_vm = mm->total_vm;
1386
1387 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1388 goto free_pt;
1389
1390 return mm;
1391
1392free_pt:
1393 /* don't put binfmt in mmput, we haven't got module yet */
1394 mm->binfmt = NULL;
1395 mm_init_owner(mm, NULL);
1396 mmput(mm);
1397
1398fail_nomem:
1399 return NULL;
1400}
1401
1402static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1403{
1404 struct mm_struct *mm, *oldmm;
1405
1406 tsk->min_flt = tsk->maj_flt = 0;
1407 tsk->nvcsw = tsk->nivcsw = 0;
1408#ifdef CONFIG_DETECT_HUNG_TASK
1409 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1410 tsk->last_switch_time = 0;
1411#endif
1412
1413 tsk->mm = NULL;
1414 tsk->active_mm = NULL;
1415
1416 /*
1417 * Are we cloning a kernel thread?
1418 *
1419 * We need to steal a active VM for that..
1420 */
1421 oldmm = current->mm;
1422 if (!oldmm)
1423 return 0;
1424
1425 /* initialize the new vmacache entries */
1426 vmacache_flush(tsk);
1427
1428 if (clone_flags & CLONE_VM) {
1429 mmget(oldmm);
1430 mm = oldmm;
1431 } else {
1432 mm = dup_mm(tsk, current->mm);
1433 if (!mm)
1434 return -ENOMEM;
1435 }
1436
1437 tsk->mm = mm;
1438 tsk->active_mm = mm;
1439 return 0;
1440}
1441
1442static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1443{
1444 struct fs_struct *fs = current->fs;
1445 if (clone_flags & CLONE_FS) {
1446 /* tsk->fs is already what we want */
1447 spin_lock(&fs->lock);
1448 if (fs->in_exec) {
1449 spin_unlock(&fs->lock);
1450 return -EAGAIN;
1451 }
1452 fs->users++;
1453 spin_unlock(&fs->lock);
1454 return 0;
1455 }
1456 tsk->fs = copy_fs_struct(fs);
1457 if (!tsk->fs)
1458 return -ENOMEM;
1459 return 0;
1460}
1461
1462static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1463{
1464 struct files_struct *oldf, *newf;
1465 int error = 0;
1466
1467 /*
1468 * A background process may not have any files ...
1469 */
1470 oldf = current->files;
1471 if (!oldf)
1472 goto out;
1473
1474 if (clone_flags & CLONE_FILES) {
1475 atomic_inc(&oldf->count);
1476 goto out;
1477 }
1478
1479 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1480 if (!newf)
1481 goto out;
1482
1483 tsk->files = newf;
1484 error = 0;
1485out:
1486 return error;
1487}
1488
1489static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1490{
1491#ifdef CONFIG_BLOCK
1492 struct io_context *ioc = current->io_context;
1493 struct io_context *new_ioc;
1494
1495 if (!ioc)
1496 return 0;
1497 /*
1498 * Share io context with parent, if CLONE_IO is set
1499 */
1500 if (clone_flags & CLONE_IO) {
1501 ioc_task_link(ioc);
1502 tsk->io_context = ioc;
1503 } else if (ioprio_valid(ioc->ioprio)) {
1504 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1505 if (unlikely(!new_ioc))
1506 return -ENOMEM;
1507
1508 new_ioc->ioprio = ioc->ioprio;
1509 put_io_context(new_ioc);
1510 }
1511#endif
1512 return 0;
1513}
1514
1515static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1516{
1517 struct sighand_struct *sig;
1518
1519 if (clone_flags & CLONE_SIGHAND) {
1520 refcount_inc(¤t->sighand->count);
1521 return 0;
1522 }
1523 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1524 RCU_INIT_POINTER(tsk->sighand, sig);
1525 if (!sig)
1526 return -ENOMEM;
1527
1528 refcount_set(&sig->count, 1);
1529 spin_lock_irq(¤t->sighand->siglock);
1530 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1531 spin_unlock_irq(¤t->sighand->siglock);
1532
1533 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1534 if (clone_flags & CLONE_CLEAR_SIGHAND)
1535 flush_signal_handlers(tsk, 0);
1536
1537 return 0;
1538}
1539
1540void __cleanup_sighand(struct sighand_struct *sighand)
1541{
1542 if (refcount_dec_and_test(&sighand->count)) {
1543 signalfd_cleanup(sighand);
1544 /*
1545 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1546 * without an RCU grace period, see __lock_task_sighand().
1547 */
1548 kmem_cache_free(sighand_cachep, sighand);
1549 }
1550}
1551
1552/*
1553 * Initialize POSIX timer handling for a thread group.
1554 */
1555static void posix_cpu_timers_init_group(struct signal_struct *sig)
1556{
1557 struct posix_cputimers *pct = &sig->posix_cputimers;
1558 unsigned long cpu_limit;
1559
1560 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1561 posix_cputimers_group_init(pct, cpu_limit);
1562}
1563
1564static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1565{
1566 struct signal_struct *sig;
1567
1568 if (clone_flags & CLONE_THREAD)
1569 return 0;
1570
1571 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1572 tsk->signal = sig;
1573 if (!sig)
1574 return -ENOMEM;
1575
1576 sig->nr_threads = 1;
1577 atomic_set(&sig->live, 1);
1578 refcount_set(&sig->sigcnt, 1);
1579
1580 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1581 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1582 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1583
1584 init_waitqueue_head(&sig->wait_chldexit);
1585 sig->curr_target = tsk;
1586 init_sigpending(&sig->shared_pending);
1587 INIT_HLIST_HEAD(&sig->multiprocess);
1588 seqlock_init(&sig->stats_lock);
1589 prev_cputime_init(&sig->prev_cputime);
1590
1591#ifdef CONFIG_POSIX_TIMERS
1592 INIT_LIST_HEAD(&sig->posix_timers);
1593 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1594 sig->real_timer.function = it_real_fn;
1595#endif
1596
1597 task_lock(current->group_leader);
1598 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1599 task_unlock(current->group_leader);
1600
1601 posix_cpu_timers_init_group(sig);
1602
1603 tty_audit_fork(sig);
1604 sched_autogroup_fork(sig);
1605
1606 sig->oom_score_adj = current->signal->oom_score_adj;
1607 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1608
1609 mutex_init(&sig->cred_guard_mutex);
1610 init_rwsem(&sig->exec_update_lock);
1611
1612 return 0;
1613}
1614
1615static void copy_seccomp(struct task_struct *p)
1616{
1617#ifdef CONFIG_SECCOMP
1618 /*
1619 * Must be called with sighand->lock held, which is common to
1620 * all threads in the group. Holding cred_guard_mutex is not
1621 * needed because this new task is not yet running and cannot
1622 * be racing exec.
1623 */
1624 assert_spin_locked(¤t->sighand->siglock);
1625
1626 /* Ref-count the new filter user, and assign it. */
1627 get_seccomp_filter(current);
1628 p->seccomp = current->seccomp;
1629
1630 /*
1631 * Explicitly enable no_new_privs here in case it got set
1632 * between the task_struct being duplicated and holding the
1633 * sighand lock. The seccomp state and nnp must be in sync.
1634 */
1635 if (task_no_new_privs(current))
1636 task_set_no_new_privs(p);
1637
1638 /*
1639 * If the parent gained a seccomp mode after copying thread
1640 * flags and between before we held the sighand lock, we have
1641 * to manually enable the seccomp thread flag here.
1642 */
1643 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1644 set_task_syscall_work(p, SECCOMP);
1645#endif
1646}
1647
1648SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1649{
1650 current->clear_child_tid = tidptr;
1651
1652 return task_pid_vnr(current);
1653}
1654
1655static void rt_mutex_init_task(struct task_struct *p)
1656{
1657 raw_spin_lock_init(&p->pi_lock);
1658#ifdef CONFIG_RT_MUTEXES
1659 p->pi_waiters = RB_ROOT_CACHED;
1660 p->pi_top_task = NULL;
1661 p->pi_blocked_on = NULL;
1662#endif
1663}
1664
1665static inline void init_task_pid_links(struct task_struct *task)
1666{
1667 enum pid_type type;
1668
1669 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1670 INIT_HLIST_NODE(&task->pid_links[type]);
1671}
1672
1673static inline void
1674init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1675{
1676 if (type == PIDTYPE_PID)
1677 task->thread_pid = pid;
1678 else
1679 task->signal->pids[type] = pid;
1680}
1681
1682static inline void rcu_copy_process(struct task_struct *p)
1683{
1684#ifdef CONFIG_PREEMPT_RCU
1685 p->rcu_read_lock_nesting = 0;
1686 p->rcu_read_unlock_special.s = 0;
1687 p->rcu_blocked_node = NULL;
1688 INIT_LIST_HEAD(&p->rcu_node_entry);
1689#endif /* #ifdef CONFIG_PREEMPT_RCU */
1690#ifdef CONFIG_TASKS_RCU
1691 p->rcu_tasks_holdout = false;
1692 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1693 p->rcu_tasks_idle_cpu = -1;
1694#endif /* #ifdef CONFIG_TASKS_RCU */
1695#ifdef CONFIG_TASKS_TRACE_RCU
1696 p->trc_reader_nesting = 0;
1697 p->trc_reader_special.s = 0;
1698 INIT_LIST_HEAD(&p->trc_holdout_list);
1699#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1700}
1701
1702struct pid *pidfd_pid(const struct file *file)
1703{
1704 if (file->f_op == &pidfd_fops)
1705 return file->private_data;
1706
1707 return ERR_PTR(-EBADF);
1708}
1709
1710static int pidfd_release(struct inode *inode, struct file *file)
1711{
1712 struct pid *pid = file->private_data;
1713
1714 file->private_data = NULL;
1715 put_pid(pid);
1716 return 0;
1717}
1718
1719#ifdef CONFIG_PROC_FS
1720/**
1721 * pidfd_show_fdinfo - print information about a pidfd
1722 * @m: proc fdinfo file
1723 * @f: file referencing a pidfd
1724 *
1725 * Pid:
1726 * This function will print the pid that a given pidfd refers to in the
1727 * pid namespace of the procfs instance.
1728 * If the pid namespace of the process is not a descendant of the pid
1729 * namespace of the procfs instance 0 will be shown as its pid. This is
1730 * similar to calling getppid() on a process whose parent is outside of
1731 * its pid namespace.
1732 *
1733 * NSpid:
1734 * If pid namespaces are supported then this function will also print
1735 * the pid of a given pidfd refers to for all descendant pid namespaces
1736 * starting from the current pid namespace of the instance, i.e. the
1737 * Pid field and the first entry in the NSpid field will be identical.
1738 * If the pid namespace of the process is not a descendant of the pid
1739 * namespace of the procfs instance 0 will be shown as its first NSpid
1740 * entry and no others will be shown.
1741 * Note that this differs from the Pid and NSpid fields in
1742 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1743 * the pid namespace of the procfs instance. The difference becomes
1744 * obvious when sending around a pidfd between pid namespaces from a
1745 * different branch of the tree, i.e. where no ancestral relation is
1746 * present between the pid namespaces:
1747 * - create two new pid namespaces ns1 and ns2 in the initial pid
1748 * namespace (also take care to create new mount namespaces in the
1749 * new pid namespace and mount procfs)
1750 * - create a process with a pidfd in ns1
1751 * - send pidfd from ns1 to ns2
1752 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1753 * have exactly one entry, which is 0
1754 */
1755static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1756{
1757 struct pid *pid = f->private_data;
1758 struct pid_namespace *ns;
1759 pid_t nr = -1;
1760
1761 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1762 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1763 nr = pid_nr_ns(pid, ns);
1764 }
1765
1766 seq_put_decimal_ll(m, "Pid:\t", nr);
1767
1768#ifdef CONFIG_PID_NS
1769 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1770 if (nr > 0) {
1771 int i;
1772
1773 /* If nr is non-zero it means that 'pid' is valid and that
1774 * ns, i.e. the pid namespace associated with the procfs
1775 * instance, is in the pid namespace hierarchy of pid.
1776 * Start at one below the already printed level.
1777 */
1778 for (i = ns->level + 1; i <= pid->level; i++)
1779 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1780 }
1781#endif
1782 seq_putc(m, '\n');
1783}
1784#endif
1785
1786/*
1787 * Poll support for process exit notification.
1788 */
1789static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1790{
1791 struct pid *pid = file->private_data;
1792 __poll_t poll_flags = 0;
1793
1794 poll_wait(file, &pid->wait_pidfd, pts);
1795
1796 /*
1797 * Inform pollers only when the whole thread group exits.
1798 * If the thread group leader exits before all other threads in the
1799 * group, then poll(2) should block, similar to the wait(2) family.
1800 */
1801 if (thread_group_exited(pid))
1802 poll_flags = EPOLLIN | EPOLLRDNORM;
1803
1804 return poll_flags;
1805}
1806
1807const struct file_operations pidfd_fops = {
1808 .release = pidfd_release,
1809 .poll = pidfd_poll,
1810#ifdef CONFIG_PROC_FS
1811 .show_fdinfo = pidfd_show_fdinfo,
1812#endif
1813};
1814
1815static void __delayed_free_task(struct rcu_head *rhp)
1816{
1817 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1818
1819 free_task(tsk);
1820}
1821
1822static __always_inline void delayed_free_task(struct task_struct *tsk)
1823{
1824 if (IS_ENABLED(CONFIG_MEMCG))
1825 call_rcu(&tsk->rcu, __delayed_free_task);
1826 else
1827 free_task(tsk);
1828}
1829
1830static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1831{
1832 /* Skip if kernel thread */
1833 if (!tsk->mm)
1834 return;
1835
1836 /* Skip if spawning a thread or using vfork */
1837 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1838 return;
1839
1840 /* We need to synchronize with __set_oom_adj */
1841 mutex_lock(&oom_adj_mutex);
1842 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1843 /* Update the values in case they were changed after copy_signal */
1844 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1845 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1846 mutex_unlock(&oom_adj_mutex);
1847}
1848
1849/*
1850 * This creates a new process as a copy of the old one,
1851 * but does not actually start it yet.
1852 *
1853 * It copies the registers, and all the appropriate
1854 * parts of the process environment (as per the clone
1855 * flags). The actual kick-off is left to the caller.
1856 */
1857static __latent_entropy struct task_struct *copy_process(
1858 struct pid *pid,
1859 int trace,
1860 int node,
1861 struct kernel_clone_args *args)
1862{
1863 int pidfd = -1, retval;
1864 struct task_struct *p;
1865 struct multiprocess_signals delayed;
1866 struct file *pidfile = NULL;
1867 u64 clone_flags = args->flags;
1868 struct nsproxy *nsp = current->nsproxy;
1869
1870 /*
1871 * Don't allow sharing the root directory with processes in a different
1872 * namespace
1873 */
1874 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1875 return ERR_PTR(-EINVAL);
1876
1877 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1878 return ERR_PTR(-EINVAL);
1879
1880 /*
1881 * Thread groups must share signals as well, and detached threads
1882 * can only be started up within the thread group.
1883 */
1884 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1885 return ERR_PTR(-EINVAL);
1886
1887 /*
1888 * Shared signal handlers imply shared VM. By way of the above,
1889 * thread groups also imply shared VM. Blocking this case allows
1890 * for various simplifications in other code.
1891 */
1892 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1893 return ERR_PTR(-EINVAL);
1894
1895 /*
1896 * Siblings of global init remain as zombies on exit since they are
1897 * not reaped by their parent (swapper). To solve this and to avoid
1898 * multi-rooted process trees, prevent global and container-inits
1899 * from creating siblings.
1900 */
1901 if ((clone_flags & CLONE_PARENT) &&
1902 current->signal->flags & SIGNAL_UNKILLABLE)
1903 return ERR_PTR(-EINVAL);
1904
1905 /*
1906 * If the new process will be in a different pid or user namespace
1907 * do not allow it to share a thread group with the forking task.
1908 */
1909 if (clone_flags & CLONE_THREAD) {
1910 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1911 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1912 return ERR_PTR(-EINVAL);
1913 }
1914
1915 /*
1916 * If the new process will be in a different time namespace
1917 * do not allow it to share VM or a thread group with the forking task.
1918 */
1919 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1920 if (nsp->time_ns != nsp->time_ns_for_children)
1921 return ERR_PTR(-EINVAL);
1922 }
1923
1924 if (clone_flags & CLONE_PIDFD) {
1925 /*
1926 * - CLONE_DETACHED is blocked so that we can potentially
1927 * reuse it later for CLONE_PIDFD.
1928 * - CLONE_THREAD is blocked until someone really needs it.
1929 */
1930 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1931 return ERR_PTR(-EINVAL);
1932 }
1933
1934 /*
1935 * Force any signals received before this point to be delivered
1936 * before the fork happens. Collect up signals sent to multiple
1937 * processes that happen during the fork and delay them so that
1938 * they appear to happen after the fork.
1939 */
1940 sigemptyset(&delayed.signal);
1941 INIT_HLIST_NODE(&delayed.node);
1942
1943 spin_lock_irq(¤t->sighand->siglock);
1944 if (!(clone_flags & CLONE_THREAD))
1945 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1946 recalc_sigpending();
1947 spin_unlock_irq(¤t->sighand->siglock);
1948 retval = -ERESTARTNOINTR;
1949 if (task_sigpending(current))
1950 goto fork_out;
1951
1952 retval = -ENOMEM;
1953 p = dup_task_struct(current, node);
1954 if (!p)
1955 goto fork_out;
1956 if (args->io_thread) {
1957 /*
1958 * Mark us an IO worker, and block any signal that isn't
1959 * fatal or STOP
1960 */
1961 p->flags |= PF_IO_WORKER;
1962 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1963 }
1964
1965 /*
1966 * This _must_ happen before we call free_task(), i.e. before we jump
1967 * to any of the bad_fork_* labels. This is to avoid freeing
1968 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1969 * kernel threads (PF_KTHREAD).
1970 */
1971 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1972 /*
1973 * Clear TID on mm_release()?
1974 */
1975 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1976
1977 ftrace_graph_init_task(p);
1978
1979 rt_mutex_init_task(p);
1980
1981 lockdep_assert_irqs_enabled();
1982#ifdef CONFIG_PROVE_LOCKING
1983 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1984#endif
1985 retval = -EAGAIN;
1986 if (is_ucounts_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1987 if (p->real_cred->user != INIT_USER &&
1988 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1989 goto bad_fork_free;
1990 }
1991 current->flags &= ~PF_NPROC_EXCEEDED;
1992
1993 retval = copy_creds(p, clone_flags);
1994 if (retval < 0)
1995 goto bad_fork_free;
1996
1997 /*
1998 * If multiple threads are within copy_process(), then this check
1999 * triggers too late. This doesn't hurt, the check is only there
2000 * to stop root fork bombs.
2001 */
2002 retval = -EAGAIN;
2003 if (data_race(nr_threads >= max_threads))
2004 goto bad_fork_cleanup_count;
2005
2006 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2007 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2008 p->flags |= PF_FORKNOEXEC;
2009 INIT_LIST_HEAD(&p->children);
2010 INIT_LIST_HEAD(&p->sibling);
2011 rcu_copy_process(p);
2012 p->vfork_done = NULL;
2013 spin_lock_init(&p->alloc_lock);
2014
2015 init_sigpending(&p->pending);
2016
2017 p->utime = p->stime = p->gtime = 0;
2018#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2019 p->utimescaled = p->stimescaled = 0;
2020#endif
2021 prev_cputime_init(&p->prev_cputime);
2022
2023#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2024 seqcount_init(&p->vtime.seqcount);
2025 p->vtime.starttime = 0;
2026 p->vtime.state = VTIME_INACTIVE;
2027#endif
2028
2029#ifdef CONFIG_IO_URING
2030 p->io_uring = NULL;
2031#endif
2032
2033#if defined(SPLIT_RSS_COUNTING)
2034 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2035#endif
2036
2037 p->default_timer_slack_ns = current->timer_slack_ns;
2038
2039#ifdef CONFIG_PSI
2040 p->psi_flags = 0;
2041#endif
2042
2043 task_io_accounting_init(&p->ioac);
2044 acct_clear_integrals(p);
2045
2046 posix_cputimers_init(&p->posix_cputimers);
2047
2048 p->io_context = NULL;
2049 audit_set_context(p, NULL);
2050 cgroup_fork(p);
2051#ifdef CONFIG_NUMA
2052 p->mempolicy = mpol_dup(p->mempolicy);
2053 if (IS_ERR(p->mempolicy)) {
2054 retval = PTR_ERR(p->mempolicy);
2055 p->mempolicy = NULL;
2056 goto bad_fork_cleanup_threadgroup_lock;
2057 }
2058#endif
2059#ifdef CONFIG_CPUSETS
2060 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2061 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2062 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2063#endif
2064#ifdef CONFIG_TRACE_IRQFLAGS
2065 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2066 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2067 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2068 p->softirqs_enabled = 1;
2069 p->softirq_context = 0;
2070#endif
2071
2072 p->pagefault_disabled = 0;
2073
2074#ifdef CONFIG_LOCKDEP
2075 lockdep_init_task(p);
2076#endif
2077
2078#ifdef CONFIG_DEBUG_MUTEXES
2079 p->blocked_on = NULL; /* not blocked yet */
2080#endif
2081#ifdef CONFIG_BCACHE
2082 p->sequential_io = 0;
2083 p->sequential_io_avg = 0;
2084#endif
2085#ifdef CONFIG_BPF_SYSCALL
2086 RCU_INIT_POINTER(p->bpf_storage, NULL);
2087#endif
2088
2089 /* Perform scheduler related setup. Assign this task to a CPU. */
2090 retval = sched_fork(clone_flags, p);
2091 if (retval)
2092 goto bad_fork_cleanup_policy;
2093
2094 retval = perf_event_init_task(p, clone_flags);
2095 if (retval)
2096 goto bad_fork_cleanup_policy;
2097 retval = audit_alloc(p);
2098 if (retval)
2099 goto bad_fork_cleanup_perf;
2100 /* copy all the process information */
2101 shm_init_task(p);
2102 retval = security_task_alloc(p, clone_flags);
2103 if (retval)
2104 goto bad_fork_cleanup_audit;
2105 retval = copy_semundo(clone_flags, p);
2106 if (retval)
2107 goto bad_fork_cleanup_security;
2108 retval = copy_files(clone_flags, p);
2109 if (retval)
2110 goto bad_fork_cleanup_semundo;
2111 retval = copy_fs(clone_flags, p);
2112 if (retval)
2113 goto bad_fork_cleanup_files;
2114 retval = copy_sighand(clone_flags, p);
2115 if (retval)
2116 goto bad_fork_cleanup_fs;
2117 retval = copy_signal(clone_flags, p);
2118 if (retval)
2119 goto bad_fork_cleanup_sighand;
2120 retval = copy_mm(clone_flags, p);
2121 if (retval)
2122 goto bad_fork_cleanup_signal;
2123 retval = copy_namespaces(clone_flags, p);
2124 if (retval)
2125 goto bad_fork_cleanup_mm;
2126 retval = copy_io(clone_flags, p);
2127 if (retval)
2128 goto bad_fork_cleanup_namespaces;
2129 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2130 if (retval)
2131 goto bad_fork_cleanup_io;
2132
2133 stackleak_task_init(p);
2134
2135 if (pid != &init_struct_pid) {
2136 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2137 args->set_tid_size);
2138 if (IS_ERR(pid)) {
2139 retval = PTR_ERR(pid);
2140 goto bad_fork_cleanup_thread;
2141 }
2142 }
2143
2144 /*
2145 * This has to happen after we've potentially unshared the file
2146 * descriptor table (so that the pidfd doesn't leak into the child
2147 * if the fd table isn't shared).
2148 */
2149 if (clone_flags & CLONE_PIDFD) {
2150 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2151 if (retval < 0)
2152 goto bad_fork_free_pid;
2153
2154 pidfd = retval;
2155
2156 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2157 O_RDWR | O_CLOEXEC);
2158 if (IS_ERR(pidfile)) {
2159 put_unused_fd(pidfd);
2160 retval = PTR_ERR(pidfile);
2161 goto bad_fork_free_pid;
2162 }
2163 get_pid(pid); /* held by pidfile now */
2164
2165 retval = put_user(pidfd, args->pidfd);
2166 if (retval)
2167 goto bad_fork_put_pidfd;
2168 }
2169
2170#ifdef CONFIG_BLOCK
2171 p->plug = NULL;
2172#endif
2173 futex_init_task(p);
2174
2175 /*
2176 * sigaltstack should be cleared when sharing the same VM
2177 */
2178 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2179 sas_ss_reset(p);
2180
2181 /*
2182 * Syscall tracing and stepping should be turned off in the
2183 * child regardless of CLONE_PTRACE.
2184 */
2185 user_disable_single_step(p);
2186 clear_task_syscall_work(p, SYSCALL_TRACE);
2187#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2188 clear_task_syscall_work(p, SYSCALL_EMU);
2189#endif
2190 clear_tsk_latency_tracing(p);
2191
2192 /* ok, now we should be set up.. */
2193 p->pid = pid_nr(pid);
2194 if (clone_flags & CLONE_THREAD) {
2195 p->group_leader = current->group_leader;
2196 p->tgid = current->tgid;
2197 } else {
2198 p->group_leader = p;
2199 p->tgid = p->pid;
2200 }
2201
2202 p->nr_dirtied = 0;
2203 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2204 p->dirty_paused_when = 0;
2205
2206 p->pdeath_signal = 0;
2207 INIT_LIST_HEAD(&p->thread_group);
2208 p->task_works = NULL;
2209
2210#ifdef CONFIG_KRETPROBES
2211 p->kretprobe_instances.first = NULL;
2212#endif
2213
2214 /*
2215 * Ensure that the cgroup subsystem policies allow the new process to be
2216 * forked. It should be noted that the new process's css_set can be changed
2217 * between here and cgroup_post_fork() if an organisation operation is in
2218 * progress.
2219 */
2220 retval = cgroup_can_fork(p, args);
2221 if (retval)
2222 goto bad_fork_put_pidfd;
2223
2224 /*
2225 * From this point on we must avoid any synchronous user-space
2226 * communication until we take the tasklist-lock. In particular, we do
2227 * not want user-space to be able to predict the process start-time by
2228 * stalling fork(2) after we recorded the start_time but before it is
2229 * visible to the system.
2230 */
2231
2232 p->start_time = ktime_get_ns();
2233 p->start_boottime = ktime_get_boottime_ns();
2234
2235 /*
2236 * Make it visible to the rest of the system, but dont wake it up yet.
2237 * Need tasklist lock for parent etc handling!
2238 */
2239 write_lock_irq(&tasklist_lock);
2240
2241 /* CLONE_PARENT re-uses the old parent */
2242 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2243 p->real_parent = current->real_parent;
2244 p->parent_exec_id = current->parent_exec_id;
2245 if (clone_flags & CLONE_THREAD)
2246 p->exit_signal = -1;
2247 else
2248 p->exit_signal = current->group_leader->exit_signal;
2249 } else {
2250 p->real_parent = current;
2251 p->parent_exec_id = current->self_exec_id;
2252 p->exit_signal = args->exit_signal;
2253 }
2254
2255 klp_copy_process(p);
2256
2257 sched_core_fork(p);
2258
2259 spin_lock(¤t->sighand->siglock);
2260
2261 /*
2262 * Copy seccomp details explicitly here, in case they were changed
2263 * before holding sighand lock.
2264 */
2265 copy_seccomp(p);
2266
2267 rseq_fork(p, clone_flags);
2268
2269 /* Don't start children in a dying pid namespace */
2270 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2271 retval = -ENOMEM;
2272 goto bad_fork_cancel_cgroup;
2273 }
2274
2275 /* Let kill terminate clone/fork in the middle */
2276 if (fatal_signal_pending(current)) {
2277 retval = -EINTR;
2278 goto bad_fork_cancel_cgroup;
2279 }
2280
2281 /* past the last point of failure */
2282 if (pidfile)
2283 fd_install(pidfd, pidfile);
2284
2285 init_task_pid_links(p);
2286 if (likely(p->pid)) {
2287 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2288
2289 init_task_pid(p, PIDTYPE_PID, pid);
2290 if (thread_group_leader(p)) {
2291 init_task_pid(p, PIDTYPE_TGID, pid);
2292 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2293 init_task_pid(p, PIDTYPE_SID, task_session(current));
2294
2295 if (is_child_reaper(pid)) {
2296 ns_of_pid(pid)->child_reaper = p;
2297 p->signal->flags |= SIGNAL_UNKILLABLE;
2298 }
2299 p->signal->shared_pending.signal = delayed.signal;
2300 p->signal->tty = tty_kref_get(current->signal->tty);
2301 /*
2302 * Inherit has_child_subreaper flag under the same
2303 * tasklist_lock with adding child to the process tree
2304 * for propagate_has_child_subreaper optimization.
2305 */
2306 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2307 p->real_parent->signal->is_child_subreaper;
2308 list_add_tail(&p->sibling, &p->real_parent->children);
2309 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2310 attach_pid(p, PIDTYPE_TGID);
2311 attach_pid(p, PIDTYPE_PGID);
2312 attach_pid(p, PIDTYPE_SID);
2313 __this_cpu_inc(process_counts);
2314 } else {
2315 current->signal->nr_threads++;
2316 atomic_inc(¤t->signal->live);
2317 refcount_inc(¤t->signal->sigcnt);
2318 task_join_group_stop(p);
2319 list_add_tail_rcu(&p->thread_group,
2320 &p->group_leader->thread_group);
2321 list_add_tail_rcu(&p->thread_node,
2322 &p->signal->thread_head);
2323 }
2324 attach_pid(p, PIDTYPE_PID);
2325 nr_threads++;
2326 }
2327 total_forks++;
2328 hlist_del_init(&delayed.node);
2329 spin_unlock(¤t->sighand->siglock);
2330 syscall_tracepoint_update(p);
2331 write_unlock_irq(&tasklist_lock);
2332
2333 proc_fork_connector(p);
2334 sched_post_fork(p);
2335 cgroup_post_fork(p, args);
2336 perf_event_fork(p);
2337
2338 trace_task_newtask(p, clone_flags);
2339 uprobe_copy_process(p, clone_flags);
2340
2341 copy_oom_score_adj(clone_flags, p);
2342
2343 return p;
2344
2345bad_fork_cancel_cgroup:
2346 sched_core_free(p);
2347 spin_unlock(¤t->sighand->siglock);
2348 write_unlock_irq(&tasklist_lock);
2349 cgroup_cancel_fork(p, args);
2350bad_fork_put_pidfd:
2351 if (clone_flags & CLONE_PIDFD) {
2352 fput(pidfile);
2353 put_unused_fd(pidfd);
2354 }
2355bad_fork_free_pid:
2356 if (pid != &init_struct_pid)
2357 free_pid(pid);
2358bad_fork_cleanup_thread:
2359 exit_thread(p);
2360bad_fork_cleanup_io:
2361 if (p->io_context)
2362 exit_io_context(p);
2363bad_fork_cleanup_namespaces:
2364 exit_task_namespaces(p);
2365bad_fork_cleanup_mm:
2366 if (p->mm) {
2367 mm_clear_owner(p->mm, p);
2368 mmput(p->mm);
2369 }
2370bad_fork_cleanup_signal:
2371 if (!(clone_flags & CLONE_THREAD))
2372 free_signal_struct(p->signal);
2373bad_fork_cleanup_sighand:
2374 __cleanup_sighand(p->sighand);
2375bad_fork_cleanup_fs:
2376 exit_fs(p); /* blocking */
2377bad_fork_cleanup_files:
2378 exit_files(p); /* blocking */
2379bad_fork_cleanup_semundo:
2380 exit_sem(p);
2381bad_fork_cleanup_security:
2382 security_task_free(p);
2383bad_fork_cleanup_audit:
2384 audit_free(p);
2385bad_fork_cleanup_perf:
2386 perf_event_free_task(p);
2387bad_fork_cleanup_policy:
2388 lockdep_free_task(p);
2389#ifdef CONFIG_NUMA
2390 mpol_put(p->mempolicy);
2391bad_fork_cleanup_threadgroup_lock:
2392#endif
2393 delayacct_tsk_free(p);
2394bad_fork_cleanup_count:
2395 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2396 exit_creds(p);
2397bad_fork_free:
2398 WRITE_ONCE(p->__state, TASK_DEAD);
2399 put_task_stack(p);
2400 delayed_free_task(p);
2401fork_out:
2402 spin_lock_irq(¤t->sighand->siglock);
2403 hlist_del_init(&delayed.node);
2404 spin_unlock_irq(¤t->sighand->siglock);
2405 return ERR_PTR(retval);
2406}
2407
2408static inline void init_idle_pids(struct task_struct *idle)
2409{
2410 enum pid_type type;
2411
2412 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2413 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2414 init_task_pid(idle, type, &init_struct_pid);
2415 }
2416}
2417
2418struct task_struct * __init fork_idle(int cpu)
2419{
2420 struct task_struct *task;
2421 struct kernel_clone_args args = {
2422 .flags = CLONE_VM,
2423 };
2424
2425 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2426 if (!IS_ERR(task)) {
2427 init_idle_pids(task);
2428 init_idle(task, cpu);
2429 }
2430
2431 return task;
2432}
2433
2434struct mm_struct *copy_init_mm(void)
2435{
2436 return dup_mm(NULL, &init_mm);
2437}
2438
2439/*
2440 * This is like kernel_clone(), but shaved down and tailored to just
2441 * creating io_uring workers. It returns a created task, or an error pointer.
2442 * The returned task is inactive, and the caller must fire it up through
2443 * wake_up_new_task(p). All signals are blocked in the created task.
2444 */
2445struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2446{
2447 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2448 CLONE_IO;
2449 struct kernel_clone_args args = {
2450 .flags = ((lower_32_bits(flags) | CLONE_VM |
2451 CLONE_UNTRACED) & ~CSIGNAL),
2452 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2453 .stack = (unsigned long)fn,
2454 .stack_size = (unsigned long)arg,
2455 .io_thread = 1,
2456 };
2457
2458 return copy_process(NULL, 0, node, &args);
2459}
2460
2461/*
2462 * Ok, this is the main fork-routine.
2463 *
2464 * It copies the process, and if successful kick-starts
2465 * it and waits for it to finish using the VM if required.
2466 *
2467 * args->exit_signal is expected to be checked for sanity by the caller.
2468 */
2469pid_t kernel_clone(struct kernel_clone_args *args)
2470{
2471 u64 clone_flags = args->flags;
2472 struct completion vfork;
2473 struct pid *pid;
2474 struct task_struct *p;
2475 int trace = 0;
2476 pid_t nr;
2477
2478 /*
2479 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2480 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2481 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2482 * field in struct clone_args and it still doesn't make sense to have
2483 * them both point at the same memory location. Performing this check
2484 * here has the advantage that we don't need to have a separate helper
2485 * to check for legacy clone().
2486 */
2487 if ((args->flags & CLONE_PIDFD) &&
2488 (args->flags & CLONE_PARENT_SETTID) &&
2489 (args->pidfd == args->parent_tid))
2490 return -EINVAL;
2491
2492 /*
2493 * Determine whether and which event to report to ptracer. When
2494 * called from kernel_thread or CLONE_UNTRACED is explicitly
2495 * requested, no event is reported; otherwise, report if the event
2496 * for the type of forking is enabled.
2497 */
2498 if (!(clone_flags & CLONE_UNTRACED)) {
2499 if (clone_flags & CLONE_VFORK)
2500 trace = PTRACE_EVENT_VFORK;
2501 else if (args->exit_signal != SIGCHLD)
2502 trace = PTRACE_EVENT_CLONE;
2503 else
2504 trace = PTRACE_EVENT_FORK;
2505
2506 if (likely(!ptrace_event_enabled(current, trace)))
2507 trace = 0;
2508 }
2509
2510 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2511 add_latent_entropy();
2512
2513 if (IS_ERR(p))
2514 return PTR_ERR(p);
2515
2516 /*
2517 * Do this prior waking up the new thread - the thread pointer
2518 * might get invalid after that point, if the thread exits quickly.
2519 */
2520 trace_sched_process_fork(current, p);
2521
2522 pid = get_task_pid(p, PIDTYPE_PID);
2523 nr = pid_vnr(pid);
2524
2525 if (clone_flags & CLONE_PARENT_SETTID)
2526 put_user(nr, args->parent_tid);
2527
2528 if (clone_flags & CLONE_VFORK) {
2529 p->vfork_done = &vfork;
2530 init_completion(&vfork);
2531 get_task_struct(p);
2532 }
2533
2534 wake_up_new_task(p);
2535
2536 /* forking complete and child started to run, tell ptracer */
2537 if (unlikely(trace))
2538 ptrace_event_pid(trace, pid);
2539
2540 if (clone_flags & CLONE_VFORK) {
2541 if (!wait_for_vfork_done(p, &vfork))
2542 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2543 }
2544
2545 put_pid(pid);
2546 return nr;
2547}
2548
2549/*
2550 * Create a kernel thread.
2551 */
2552pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2553{
2554 struct kernel_clone_args args = {
2555 .flags = ((lower_32_bits(flags) | CLONE_VM |
2556 CLONE_UNTRACED) & ~CSIGNAL),
2557 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2558 .stack = (unsigned long)fn,
2559 .stack_size = (unsigned long)arg,
2560 };
2561
2562 return kernel_clone(&args);
2563}
2564
2565#ifdef __ARCH_WANT_SYS_FORK
2566SYSCALL_DEFINE0(fork)
2567{
2568#ifdef CONFIG_MMU
2569 struct kernel_clone_args args = {
2570 .exit_signal = SIGCHLD,
2571 };
2572
2573 return kernel_clone(&args);
2574#else
2575 /* can not support in nommu mode */
2576 return -EINVAL;
2577#endif
2578}
2579#endif
2580
2581#ifdef __ARCH_WANT_SYS_VFORK
2582SYSCALL_DEFINE0(vfork)
2583{
2584 struct kernel_clone_args args = {
2585 .flags = CLONE_VFORK | CLONE_VM,
2586 .exit_signal = SIGCHLD,
2587 };
2588
2589 return kernel_clone(&args);
2590}
2591#endif
2592
2593#ifdef __ARCH_WANT_SYS_CLONE
2594#ifdef CONFIG_CLONE_BACKWARDS
2595SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2596 int __user *, parent_tidptr,
2597 unsigned long, tls,
2598 int __user *, child_tidptr)
2599#elif defined(CONFIG_CLONE_BACKWARDS2)
2600SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2601 int __user *, parent_tidptr,
2602 int __user *, child_tidptr,
2603 unsigned long, tls)
2604#elif defined(CONFIG_CLONE_BACKWARDS3)
2605SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2606 int, stack_size,
2607 int __user *, parent_tidptr,
2608 int __user *, child_tidptr,
2609 unsigned long, tls)
2610#else
2611SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2612 int __user *, parent_tidptr,
2613 int __user *, child_tidptr,
2614 unsigned long, tls)
2615#endif
2616{
2617 struct kernel_clone_args args = {
2618 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2619 .pidfd = parent_tidptr,
2620 .child_tid = child_tidptr,
2621 .parent_tid = parent_tidptr,
2622 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2623 .stack = newsp,
2624 .tls = tls,
2625 };
2626
2627 return kernel_clone(&args);
2628}
2629#endif
2630
2631#ifdef __ARCH_WANT_SYS_CLONE3
2632
2633noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2634 struct clone_args __user *uargs,
2635 size_t usize)
2636{
2637 int err;
2638 struct clone_args args;
2639 pid_t *kset_tid = kargs->set_tid;
2640
2641 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2642 CLONE_ARGS_SIZE_VER0);
2643 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2644 CLONE_ARGS_SIZE_VER1);
2645 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2646 CLONE_ARGS_SIZE_VER2);
2647 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2648
2649 if (unlikely(usize > PAGE_SIZE))
2650 return -E2BIG;
2651 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2652 return -EINVAL;
2653
2654 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2655 if (err)
2656 return err;
2657
2658 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2659 return -EINVAL;
2660
2661 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2662 return -EINVAL;
2663
2664 if (unlikely(args.set_tid && args.set_tid_size == 0))
2665 return -EINVAL;
2666
2667 /*
2668 * Verify that higher 32bits of exit_signal are unset and that
2669 * it is a valid signal
2670 */
2671 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2672 !valid_signal(args.exit_signal)))
2673 return -EINVAL;
2674
2675 if ((args.flags & CLONE_INTO_CGROUP) &&
2676 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2677 return -EINVAL;
2678
2679 *kargs = (struct kernel_clone_args){
2680 .flags = args.flags,
2681 .pidfd = u64_to_user_ptr(args.pidfd),
2682 .child_tid = u64_to_user_ptr(args.child_tid),
2683 .parent_tid = u64_to_user_ptr(args.parent_tid),
2684 .exit_signal = args.exit_signal,
2685 .stack = args.stack,
2686 .stack_size = args.stack_size,
2687 .tls = args.tls,
2688 .set_tid_size = args.set_tid_size,
2689 .cgroup = args.cgroup,
2690 };
2691
2692 if (args.set_tid &&
2693 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2694 (kargs->set_tid_size * sizeof(pid_t))))
2695 return -EFAULT;
2696
2697 kargs->set_tid = kset_tid;
2698
2699 return 0;
2700}
2701
2702/**
2703 * clone3_stack_valid - check and prepare stack
2704 * @kargs: kernel clone args
2705 *
2706 * Verify that the stack arguments userspace gave us are sane.
2707 * In addition, set the stack direction for userspace since it's easy for us to
2708 * determine.
2709 */
2710static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2711{
2712 if (kargs->stack == 0) {
2713 if (kargs->stack_size > 0)
2714 return false;
2715 } else {
2716 if (kargs->stack_size == 0)
2717 return false;
2718
2719 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2720 return false;
2721
2722#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2723 kargs->stack += kargs->stack_size;
2724#endif
2725 }
2726
2727 return true;
2728}
2729
2730static bool clone3_args_valid(struct kernel_clone_args *kargs)
2731{
2732 /* Verify that no unknown flags are passed along. */
2733 if (kargs->flags &
2734 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2735 return false;
2736
2737 /*
2738 * - make the CLONE_DETACHED bit reusable for clone3
2739 * - make the CSIGNAL bits reusable for clone3
2740 */
2741 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2742 return false;
2743
2744 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2745 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2746 return false;
2747
2748 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2749 kargs->exit_signal)
2750 return false;
2751
2752 if (!clone3_stack_valid(kargs))
2753 return false;
2754
2755 return true;
2756}
2757
2758/**
2759 * clone3 - create a new process with specific properties
2760 * @uargs: argument structure
2761 * @size: size of @uargs
2762 *
2763 * clone3() is the extensible successor to clone()/clone2().
2764 * It takes a struct as argument that is versioned by its size.
2765 *
2766 * Return: On success, a positive PID for the child process.
2767 * On error, a negative errno number.
2768 */
2769SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2770{
2771 int err;
2772
2773 struct kernel_clone_args kargs;
2774 pid_t set_tid[MAX_PID_NS_LEVEL];
2775
2776 kargs.set_tid = set_tid;
2777
2778 err = copy_clone_args_from_user(&kargs, uargs, size);
2779 if (err)
2780 return err;
2781
2782 if (!clone3_args_valid(&kargs))
2783 return -EINVAL;
2784
2785 return kernel_clone(&kargs);
2786}
2787#endif
2788
2789void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2790{
2791 struct task_struct *leader, *parent, *child;
2792 int res;
2793
2794 read_lock(&tasklist_lock);
2795 leader = top = top->group_leader;
2796down:
2797 for_each_thread(leader, parent) {
2798 list_for_each_entry(child, &parent->children, sibling) {
2799 res = visitor(child, data);
2800 if (res) {
2801 if (res < 0)
2802 goto out;
2803 leader = child;
2804 goto down;
2805 }
2806up:
2807 ;
2808 }
2809 }
2810
2811 if (leader != top) {
2812 child = leader;
2813 parent = child->real_parent;
2814 leader = parent->group_leader;
2815 goto up;
2816 }
2817out:
2818 read_unlock(&tasklist_lock);
2819}
2820
2821#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2822#define ARCH_MIN_MMSTRUCT_ALIGN 0
2823#endif
2824
2825static void sighand_ctor(void *data)
2826{
2827 struct sighand_struct *sighand = data;
2828
2829 spin_lock_init(&sighand->siglock);
2830 init_waitqueue_head(&sighand->signalfd_wqh);
2831}
2832
2833void __init proc_caches_init(void)
2834{
2835 unsigned int mm_size;
2836
2837 sighand_cachep = kmem_cache_create("sighand_cache",
2838 sizeof(struct sighand_struct), 0,
2839 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2840 SLAB_ACCOUNT, sighand_ctor);
2841 signal_cachep = kmem_cache_create("signal_cache",
2842 sizeof(struct signal_struct), 0,
2843 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2844 NULL);
2845 files_cachep = kmem_cache_create("files_cache",
2846 sizeof(struct files_struct), 0,
2847 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2848 NULL);
2849 fs_cachep = kmem_cache_create("fs_cache",
2850 sizeof(struct fs_struct), 0,
2851 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2852 NULL);
2853
2854 /*
2855 * The mm_cpumask is located at the end of mm_struct, and is
2856 * dynamically sized based on the maximum CPU number this system
2857 * can have, taking hotplug into account (nr_cpu_ids).
2858 */
2859 mm_size = sizeof(struct mm_struct) + cpumask_size();
2860
2861 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2862 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2863 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2864 offsetof(struct mm_struct, saved_auxv),
2865 sizeof_field(struct mm_struct, saved_auxv),
2866 NULL);
2867 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2868 mmap_init();
2869 nsproxy_cache_init();
2870}
2871
2872/*
2873 * Check constraints on flags passed to the unshare system call.
2874 */
2875static int check_unshare_flags(unsigned long unshare_flags)
2876{
2877 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2878 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2879 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2880 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2881 CLONE_NEWTIME))
2882 return -EINVAL;
2883 /*
2884 * Not implemented, but pretend it works if there is nothing
2885 * to unshare. Note that unsharing the address space or the
2886 * signal handlers also need to unshare the signal queues (aka
2887 * CLONE_THREAD).
2888 */
2889 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2890 if (!thread_group_empty(current))
2891 return -EINVAL;
2892 }
2893 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2894 if (refcount_read(¤t->sighand->count) > 1)
2895 return -EINVAL;
2896 }
2897 if (unshare_flags & CLONE_VM) {
2898 if (!current_is_single_threaded())
2899 return -EINVAL;
2900 }
2901
2902 return 0;
2903}
2904
2905/*
2906 * Unshare the filesystem structure if it is being shared
2907 */
2908static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2909{
2910 struct fs_struct *fs = current->fs;
2911
2912 if (!(unshare_flags & CLONE_FS) || !fs)
2913 return 0;
2914
2915 /* don't need lock here; in the worst case we'll do useless copy */
2916 if (fs->users == 1)
2917 return 0;
2918
2919 *new_fsp = copy_fs_struct(fs);
2920 if (!*new_fsp)
2921 return -ENOMEM;
2922
2923 return 0;
2924}
2925
2926/*
2927 * Unshare file descriptor table if it is being shared
2928 */
2929int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2930 struct files_struct **new_fdp)
2931{
2932 struct files_struct *fd = current->files;
2933 int error = 0;
2934
2935 if ((unshare_flags & CLONE_FILES) &&
2936 (fd && atomic_read(&fd->count) > 1)) {
2937 *new_fdp = dup_fd(fd, max_fds, &error);
2938 if (!*new_fdp)
2939 return error;
2940 }
2941
2942 return 0;
2943}
2944
2945/*
2946 * unshare allows a process to 'unshare' part of the process
2947 * context which was originally shared using clone. copy_*
2948 * functions used by kernel_clone() cannot be used here directly
2949 * because they modify an inactive task_struct that is being
2950 * constructed. Here we are modifying the current, active,
2951 * task_struct.
2952 */
2953int ksys_unshare(unsigned long unshare_flags)
2954{
2955 struct fs_struct *fs, *new_fs = NULL;
2956 struct files_struct *fd, *new_fd = NULL;
2957 struct cred *new_cred = NULL;
2958 struct nsproxy *new_nsproxy = NULL;
2959 int do_sysvsem = 0;
2960 int err;
2961
2962 /*
2963 * If unsharing a user namespace must also unshare the thread group
2964 * and unshare the filesystem root and working directories.
2965 */
2966 if (unshare_flags & CLONE_NEWUSER)
2967 unshare_flags |= CLONE_THREAD | CLONE_FS;
2968 /*
2969 * If unsharing vm, must also unshare signal handlers.
2970 */
2971 if (unshare_flags & CLONE_VM)
2972 unshare_flags |= CLONE_SIGHAND;
2973 /*
2974 * If unsharing a signal handlers, must also unshare the signal queues.
2975 */
2976 if (unshare_flags & CLONE_SIGHAND)
2977 unshare_flags |= CLONE_THREAD;
2978 /*
2979 * If unsharing namespace, must also unshare filesystem information.
2980 */
2981 if (unshare_flags & CLONE_NEWNS)
2982 unshare_flags |= CLONE_FS;
2983
2984 err = check_unshare_flags(unshare_flags);
2985 if (err)
2986 goto bad_unshare_out;
2987 /*
2988 * CLONE_NEWIPC must also detach from the undolist: after switching
2989 * to a new ipc namespace, the semaphore arrays from the old
2990 * namespace are unreachable.
2991 */
2992 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2993 do_sysvsem = 1;
2994 err = unshare_fs(unshare_flags, &new_fs);
2995 if (err)
2996 goto bad_unshare_out;
2997 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2998 if (err)
2999 goto bad_unshare_cleanup_fs;
3000 err = unshare_userns(unshare_flags, &new_cred);
3001 if (err)
3002 goto bad_unshare_cleanup_fd;
3003 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3004 new_cred, new_fs);
3005 if (err)
3006 goto bad_unshare_cleanup_cred;
3007
3008 if (new_cred) {
3009 err = set_cred_ucounts(new_cred);
3010 if (err)
3011 goto bad_unshare_cleanup_cred;
3012 }
3013
3014 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3015 if (do_sysvsem) {
3016 /*
3017 * CLONE_SYSVSEM is equivalent to sys_exit().
3018 */
3019 exit_sem(current);
3020 }
3021 if (unshare_flags & CLONE_NEWIPC) {
3022 /* Orphan segments in old ns (see sem above). */
3023 exit_shm(current);
3024 shm_init_task(current);
3025 }
3026
3027 if (new_nsproxy)
3028 switch_task_namespaces(current, new_nsproxy);
3029
3030 task_lock(current);
3031
3032 if (new_fs) {
3033 fs = current->fs;
3034 spin_lock(&fs->lock);
3035 current->fs = new_fs;
3036 if (--fs->users)
3037 new_fs = NULL;
3038 else
3039 new_fs = fs;
3040 spin_unlock(&fs->lock);
3041 }
3042
3043 if (new_fd) {
3044 fd = current->files;
3045 current->files = new_fd;
3046 new_fd = fd;
3047 }
3048
3049 task_unlock(current);
3050
3051 if (new_cred) {
3052 /* Install the new user namespace */
3053 commit_creds(new_cred);
3054 new_cred = NULL;
3055 }
3056 }
3057
3058 perf_event_namespaces(current);
3059
3060bad_unshare_cleanup_cred:
3061 if (new_cred)
3062 put_cred(new_cred);
3063bad_unshare_cleanup_fd:
3064 if (new_fd)
3065 put_files_struct(new_fd);
3066
3067bad_unshare_cleanup_fs:
3068 if (new_fs)
3069 free_fs_struct(new_fs);
3070
3071bad_unshare_out:
3072 return err;
3073}
3074
3075SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3076{
3077 return ksys_unshare(unshare_flags);
3078}
3079
3080/*
3081 * Helper to unshare the files of the current task.
3082 * We don't want to expose copy_files internals to
3083 * the exec layer of the kernel.
3084 */
3085
3086int unshare_files(void)
3087{
3088 struct task_struct *task = current;
3089 struct files_struct *old, *copy = NULL;
3090 int error;
3091
3092 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3093 if (error || !copy)
3094 return error;
3095
3096 old = task->files;
3097 task_lock(task);
3098 task->files = copy;
3099 task_unlock(task);
3100 put_files_struct(old);
3101 return 0;
3102}
3103
3104int sysctl_max_threads(struct ctl_table *table, int write,
3105 void *buffer, size_t *lenp, loff_t *ppos)
3106{
3107 struct ctl_table t;
3108 int ret;
3109 int threads = max_threads;
3110 int min = 1;
3111 int max = MAX_THREADS;
3112
3113 t = *table;
3114 t.data = &threads;
3115 t.extra1 = &min;
3116 t.extra2 = &max;
3117
3118 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3119 if (ret || !write)
3120 return ret;
3121
3122 max_threads = threads;
3123
3124 return 0;
3125}