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/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.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/syscall_user_dispatch.h>
57#include <linux/jiffies.h>
58#include <linux/futex.h>
59#include <linux/compat.h>
60#include <linux/kthread.h>
61#include <linux/task_io_accounting_ops.h>
62#include <linux/rcupdate.h>
63#include <linux/ptrace.h>
64#include <linux/mount.h>
65#include <linux/audit.h>
66#include <linux/memcontrol.h>
67#include <linux/ftrace.h>
68#include <linux/proc_fs.h>
69#include <linux/profile.h>
70#include <linux/rmap.h>
71#include <linux/ksm.h>
72#include <linux/acct.h>
73#include <linux/userfaultfd_k.h>
74#include <linux/tsacct_kern.h>
75#include <linux/cn_proc.h>
76#include <linux/freezer.h>
77#include <linux/delayacct.h>
78#include <linux/taskstats_kern.h>
79#include <linux/tty.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#include <linux/stackprotector.h>
101#include <linux/user_events.h>
102#include <linux/iommu.h>
103#include <linux/rseq.h>
104
105#include <asm/pgalloc.h>
106#include <linux/uaccess.h>
107#include <asm/mmu_context.h>
108#include <asm/cacheflush.h>
109#include <asm/tlbflush.h>
110
111#include <trace/events/sched.h>
112
113#define CREATE_TRACE_POINTS
114#include <trace/events/task.h>
115
116/*
117 * Minimum number of threads to boot the kernel
118 */
119#define MIN_THREADS 20
120
121/*
122 * Maximum number of threads
123 */
124#define MAX_THREADS FUTEX_TID_MASK
125
126/*
127 * Protected counters by write_lock_irq(&tasklist_lock)
128 */
129unsigned long total_forks; /* Handle normal Linux uptimes. */
130int nr_threads; /* The idle threads do not count.. */
131
132static int max_threads; /* tunable limit on nr_threads */
133
134#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
135
136static const char * const resident_page_types[] = {
137 NAMED_ARRAY_INDEX(MM_FILEPAGES),
138 NAMED_ARRAY_INDEX(MM_ANONPAGES),
139 NAMED_ARRAY_INDEX(MM_SWAPENTS),
140 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
141};
142
143DEFINE_PER_CPU(unsigned long, process_counts) = 0;
144
145__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
146
147#ifdef CONFIG_PROVE_RCU
148int lockdep_tasklist_lock_is_held(void)
149{
150 return lockdep_is_held(&tasklist_lock);
151}
152EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
153#endif /* #ifdef CONFIG_PROVE_RCU */
154
155int nr_processes(void)
156{
157 int cpu;
158 int total = 0;
159
160 for_each_possible_cpu(cpu)
161 total += per_cpu(process_counts, cpu);
162
163 return total;
164}
165
166void __weak arch_release_task_struct(struct task_struct *tsk)
167{
168}
169
170static struct kmem_cache *task_struct_cachep;
171
172static inline struct task_struct *alloc_task_struct_node(int node)
173{
174 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
175}
176
177static inline void free_task_struct(struct task_struct *tsk)
178{
179 kmem_cache_free(task_struct_cachep, tsk);
180}
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
196struct vm_stack {
197 struct rcu_head rcu;
198 struct vm_struct *stack_vm_area;
199};
200
201static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
202{
203 unsigned int i;
204
205 for (i = 0; i < NR_CACHED_STACKS; i++) {
206 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
207 continue;
208 return true;
209 }
210 return false;
211}
212
213static void thread_stack_free_rcu(struct rcu_head *rh)
214{
215 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
216
217 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
218 return;
219
220 vfree(vm_stack);
221}
222
223static void thread_stack_delayed_free(struct task_struct *tsk)
224{
225 struct vm_stack *vm_stack = tsk->stack;
226
227 vm_stack->stack_vm_area = tsk->stack_vm_area;
228 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
229}
230
231static int free_vm_stack_cache(unsigned int cpu)
232{
233 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
234 int i;
235
236 for (i = 0; i < NR_CACHED_STACKS; i++) {
237 struct vm_struct *vm_stack = cached_vm_stacks[i];
238
239 if (!vm_stack)
240 continue;
241
242 vfree(vm_stack->addr);
243 cached_vm_stacks[i] = NULL;
244 }
245
246 return 0;
247}
248
249static int memcg_charge_kernel_stack(struct vm_struct *vm)
250{
251 int i;
252 int ret;
253 int nr_charged = 0;
254
255 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
256
257 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
258 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
259 if (ret)
260 goto err;
261 nr_charged++;
262 }
263 return 0;
264err:
265 for (i = 0; i < nr_charged; i++)
266 memcg_kmem_uncharge_page(vm->pages[i], 0);
267 return ret;
268}
269
270static int alloc_thread_stack_node(struct task_struct *tsk, int node)
271{
272 struct vm_struct *vm;
273 void *stack;
274 int i;
275
276 for (i = 0; i < NR_CACHED_STACKS; i++) {
277 struct vm_struct *s;
278
279 s = this_cpu_xchg(cached_stacks[i], NULL);
280
281 if (!s)
282 continue;
283
284 /* Reset stack metadata. */
285 kasan_unpoison_range(s->addr, THREAD_SIZE);
286
287 stack = kasan_reset_tag(s->addr);
288
289 /* Clear stale pointers from reused stack. */
290 memset(stack, 0, THREAD_SIZE);
291
292 if (memcg_charge_kernel_stack(s)) {
293 vfree(s->addr);
294 return -ENOMEM;
295 }
296
297 tsk->stack_vm_area = s;
298 tsk->stack = stack;
299 return 0;
300 }
301
302 /*
303 * Allocated stacks are cached and later reused by new threads,
304 * so memcg accounting is performed manually on assigning/releasing
305 * stacks to tasks. Drop __GFP_ACCOUNT.
306 */
307 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
308 VMALLOC_START, VMALLOC_END,
309 THREADINFO_GFP & ~__GFP_ACCOUNT,
310 PAGE_KERNEL,
311 0, node, __builtin_return_address(0));
312 if (!stack)
313 return -ENOMEM;
314
315 vm = find_vm_area(stack);
316 if (memcg_charge_kernel_stack(vm)) {
317 vfree(stack);
318 return -ENOMEM;
319 }
320 /*
321 * We can't call find_vm_area() in interrupt context, and
322 * free_thread_stack() can be called in interrupt context,
323 * so cache the vm_struct.
324 */
325 tsk->stack_vm_area = vm;
326 stack = kasan_reset_tag(stack);
327 tsk->stack = stack;
328 return 0;
329}
330
331static void free_thread_stack(struct task_struct *tsk)
332{
333 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
334 thread_stack_delayed_free(tsk);
335
336 tsk->stack = NULL;
337 tsk->stack_vm_area = NULL;
338}
339
340# else /* !CONFIG_VMAP_STACK */
341
342static void thread_stack_free_rcu(struct rcu_head *rh)
343{
344 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
345}
346
347static void thread_stack_delayed_free(struct task_struct *tsk)
348{
349 struct rcu_head *rh = tsk->stack;
350
351 call_rcu(rh, thread_stack_free_rcu);
352}
353
354static int alloc_thread_stack_node(struct task_struct *tsk, int node)
355{
356 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
357 THREAD_SIZE_ORDER);
358
359 if (likely(page)) {
360 tsk->stack = kasan_reset_tag(page_address(page));
361 return 0;
362 }
363 return -ENOMEM;
364}
365
366static void free_thread_stack(struct task_struct *tsk)
367{
368 thread_stack_delayed_free(tsk);
369 tsk->stack = NULL;
370}
371
372# endif /* CONFIG_VMAP_STACK */
373# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
374
375static struct kmem_cache *thread_stack_cache;
376
377static void thread_stack_free_rcu(struct rcu_head *rh)
378{
379 kmem_cache_free(thread_stack_cache, rh);
380}
381
382static void thread_stack_delayed_free(struct task_struct *tsk)
383{
384 struct rcu_head *rh = tsk->stack;
385
386 call_rcu(rh, thread_stack_free_rcu);
387}
388
389static int alloc_thread_stack_node(struct task_struct *tsk, int node)
390{
391 unsigned long *stack;
392 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
393 stack = kasan_reset_tag(stack);
394 tsk->stack = stack;
395 return stack ? 0 : -ENOMEM;
396}
397
398static void free_thread_stack(struct task_struct *tsk)
399{
400 thread_stack_delayed_free(tsk);
401 tsk->stack = NULL;
402}
403
404void thread_stack_cache_init(void)
405{
406 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
407 THREAD_SIZE, THREAD_SIZE, 0, 0,
408 THREAD_SIZE, NULL);
409 BUG_ON(thread_stack_cache == NULL);
410}
411
412# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
413
414/* SLAB cache for signal_struct structures (tsk->signal) */
415static struct kmem_cache *signal_cachep;
416
417/* SLAB cache for sighand_struct structures (tsk->sighand) */
418struct kmem_cache *sighand_cachep;
419
420/* SLAB cache for files_struct structures (tsk->files) */
421struct kmem_cache *files_cachep;
422
423/* SLAB cache for fs_struct structures (tsk->fs) */
424struct kmem_cache *fs_cachep;
425
426/* SLAB cache for vm_area_struct structures */
427static struct kmem_cache *vm_area_cachep;
428
429/* SLAB cache for mm_struct structures (tsk->mm) */
430static struct kmem_cache *mm_cachep;
431
432#ifdef CONFIG_PER_VMA_LOCK
433
434/* SLAB cache for vm_area_struct.lock */
435static struct kmem_cache *vma_lock_cachep;
436
437static bool vma_lock_alloc(struct vm_area_struct *vma)
438{
439 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
440 if (!vma->vm_lock)
441 return false;
442
443 init_rwsem(&vma->vm_lock->lock);
444 vma->vm_lock_seq = -1;
445
446 return true;
447}
448
449static inline void vma_lock_free(struct vm_area_struct *vma)
450{
451 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
452}
453
454#else /* CONFIG_PER_VMA_LOCK */
455
456static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
457static inline void vma_lock_free(struct vm_area_struct *vma) {}
458
459#endif /* CONFIG_PER_VMA_LOCK */
460
461struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
462{
463 struct vm_area_struct *vma;
464
465 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
466 if (!vma)
467 return NULL;
468
469 vma_init(vma, mm);
470 if (!vma_lock_alloc(vma)) {
471 kmem_cache_free(vm_area_cachep, vma);
472 return NULL;
473 }
474
475 return vma;
476}
477
478struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
479{
480 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
481
482 if (!new)
483 return NULL;
484
485 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
486 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
487 /*
488 * orig->shared.rb may be modified concurrently, but the clone
489 * will be reinitialized.
490 */
491 data_race(memcpy(new, orig, sizeof(*new)));
492 if (!vma_lock_alloc(new)) {
493 kmem_cache_free(vm_area_cachep, new);
494 return NULL;
495 }
496 INIT_LIST_HEAD(&new->anon_vma_chain);
497 vma_numab_state_init(new);
498 dup_anon_vma_name(orig, new);
499
500 return new;
501}
502
503void __vm_area_free(struct vm_area_struct *vma)
504{
505 vma_numab_state_free(vma);
506 free_anon_vma_name(vma);
507 vma_lock_free(vma);
508 kmem_cache_free(vm_area_cachep, vma);
509}
510
511#ifdef CONFIG_PER_VMA_LOCK
512static void vm_area_free_rcu_cb(struct rcu_head *head)
513{
514 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
515 vm_rcu);
516
517 /* The vma should not be locked while being destroyed. */
518 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
519 __vm_area_free(vma);
520}
521#endif
522
523void vm_area_free(struct vm_area_struct *vma)
524{
525#ifdef CONFIG_PER_VMA_LOCK
526 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
527#else
528 __vm_area_free(vma);
529#endif
530}
531
532static void account_kernel_stack(struct task_struct *tsk, int account)
533{
534 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
535 struct vm_struct *vm = task_stack_vm_area(tsk);
536 int i;
537
538 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
539 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
540 account * (PAGE_SIZE / 1024));
541 } else {
542 void *stack = task_stack_page(tsk);
543
544 /* All stack pages are in the same node. */
545 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
546 account * (THREAD_SIZE / 1024));
547 }
548}
549
550void exit_task_stack_account(struct task_struct *tsk)
551{
552 account_kernel_stack(tsk, -1);
553
554 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
555 struct vm_struct *vm;
556 int i;
557
558 vm = task_stack_vm_area(tsk);
559 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
560 memcg_kmem_uncharge_page(vm->pages[i], 0);
561 }
562}
563
564static void release_task_stack(struct task_struct *tsk)
565{
566 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
567 return; /* Better to leak the stack than to free prematurely */
568
569 free_thread_stack(tsk);
570}
571
572#ifdef CONFIG_THREAD_INFO_IN_TASK
573void put_task_stack(struct task_struct *tsk)
574{
575 if (refcount_dec_and_test(&tsk->stack_refcount))
576 release_task_stack(tsk);
577}
578#endif
579
580void free_task(struct task_struct *tsk)
581{
582#ifdef CONFIG_SECCOMP
583 WARN_ON_ONCE(tsk->seccomp.filter);
584#endif
585 release_user_cpus_ptr(tsk);
586 scs_release(tsk);
587
588#ifndef CONFIG_THREAD_INFO_IN_TASK
589 /*
590 * The task is finally done with both the stack and thread_info,
591 * so free both.
592 */
593 release_task_stack(tsk);
594#else
595 /*
596 * If the task had a separate stack allocation, it should be gone
597 * by now.
598 */
599 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
600#endif
601 rt_mutex_debug_task_free(tsk);
602 ftrace_graph_exit_task(tsk);
603 arch_release_task_struct(tsk);
604 if (tsk->flags & PF_KTHREAD)
605 free_kthread_struct(tsk);
606 bpf_task_storage_free(tsk);
607 free_task_struct(tsk);
608}
609EXPORT_SYMBOL(free_task);
610
611static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
612{
613 struct file *exe_file;
614
615 exe_file = get_mm_exe_file(oldmm);
616 RCU_INIT_POINTER(mm->exe_file, exe_file);
617 /*
618 * We depend on the oldmm having properly denied write access to the
619 * exe_file already.
620 */
621 if (exe_file && deny_write_access(exe_file))
622 pr_warn_once("deny_write_access() failed in %s\n", __func__);
623}
624
625#ifdef CONFIG_MMU
626static __latent_entropy int dup_mmap(struct mm_struct *mm,
627 struct mm_struct *oldmm)
628{
629 struct vm_area_struct *mpnt, *tmp;
630 int retval;
631 unsigned long charge = 0;
632 LIST_HEAD(uf);
633 VMA_ITERATOR(vmi, mm, 0);
634
635 uprobe_start_dup_mmap();
636 if (mmap_write_lock_killable(oldmm)) {
637 retval = -EINTR;
638 goto fail_uprobe_end;
639 }
640 flush_cache_dup_mm(oldmm);
641 uprobe_dup_mmap(oldmm, mm);
642 /*
643 * Not linked in yet - no deadlock potential:
644 */
645 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
646
647 /* No ordering required: file already has been exposed. */
648 dup_mm_exe_file(mm, oldmm);
649
650 mm->total_vm = oldmm->total_vm;
651 mm->data_vm = oldmm->data_vm;
652 mm->exec_vm = oldmm->exec_vm;
653 mm->stack_vm = oldmm->stack_vm;
654
655 retval = ksm_fork(mm, oldmm);
656 if (retval)
657 goto out;
658 khugepaged_fork(mm, oldmm);
659
660 /* Use __mt_dup() to efficiently build an identical maple tree. */
661 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
662 if (unlikely(retval))
663 goto out;
664
665 mt_clear_in_rcu(vmi.mas.tree);
666 for_each_vma(vmi, mpnt) {
667 struct file *file;
668
669 vma_start_write(mpnt);
670 if (mpnt->vm_flags & VM_DONTCOPY) {
671 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
672 mpnt->vm_end, GFP_KERNEL);
673 if (retval)
674 goto loop_out;
675
676 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
677 continue;
678 }
679 charge = 0;
680 /*
681 * Don't duplicate many vmas if we've been oom-killed (for
682 * example)
683 */
684 if (fatal_signal_pending(current)) {
685 retval = -EINTR;
686 goto loop_out;
687 }
688 if (mpnt->vm_flags & VM_ACCOUNT) {
689 unsigned long len = vma_pages(mpnt);
690
691 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
692 goto fail_nomem;
693 charge = len;
694 }
695 tmp = vm_area_dup(mpnt);
696 if (!tmp)
697 goto fail_nomem;
698 retval = vma_dup_policy(mpnt, tmp);
699 if (retval)
700 goto fail_nomem_policy;
701 tmp->vm_mm = mm;
702 retval = dup_userfaultfd(tmp, &uf);
703 if (retval)
704 goto fail_nomem_anon_vma_fork;
705 if (tmp->vm_flags & VM_WIPEONFORK) {
706 /*
707 * VM_WIPEONFORK gets a clean slate in the child.
708 * Don't prepare anon_vma until fault since we don't
709 * copy page for current vma.
710 */
711 tmp->anon_vma = NULL;
712 } else if (anon_vma_fork(tmp, mpnt))
713 goto fail_nomem_anon_vma_fork;
714 vm_flags_clear(tmp, VM_LOCKED_MASK);
715 file = tmp->vm_file;
716 if (file) {
717 struct address_space *mapping = file->f_mapping;
718
719 get_file(file);
720 i_mmap_lock_write(mapping);
721 if (vma_is_shared_maywrite(tmp))
722 mapping_allow_writable(mapping);
723 flush_dcache_mmap_lock(mapping);
724 /* insert tmp into the share list, just after mpnt */
725 vma_interval_tree_insert_after(tmp, mpnt,
726 &mapping->i_mmap);
727 flush_dcache_mmap_unlock(mapping);
728 i_mmap_unlock_write(mapping);
729 }
730
731 /*
732 * Copy/update hugetlb private vma information.
733 */
734 if (is_vm_hugetlb_page(tmp))
735 hugetlb_dup_vma_private(tmp);
736
737 /*
738 * Link the vma into the MT. After using __mt_dup(), memory
739 * allocation is not necessary here, so it cannot fail.
740 */
741 vma_iter_bulk_store(&vmi, tmp);
742
743 mm->map_count++;
744 if (!(tmp->vm_flags & VM_WIPEONFORK))
745 retval = copy_page_range(tmp, mpnt);
746
747 if (tmp->vm_ops && tmp->vm_ops->open)
748 tmp->vm_ops->open(tmp);
749
750 if (retval) {
751 mpnt = vma_next(&vmi);
752 goto loop_out;
753 }
754 }
755 /* a new mm has just been created */
756 retval = arch_dup_mmap(oldmm, mm);
757loop_out:
758 vma_iter_free(&vmi);
759 if (!retval) {
760 mt_set_in_rcu(vmi.mas.tree);
761 } else if (mpnt) {
762 /*
763 * The entire maple tree has already been duplicated. If the
764 * mmap duplication fails, mark the failure point with
765 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
766 * stop releasing VMAs that have not been duplicated after this
767 * point.
768 */
769 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
770 mas_store(&vmi.mas, XA_ZERO_ENTRY);
771 }
772out:
773 mmap_write_unlock(mm);
774 flush_tlb_mm(oldmm);
775 mmap_write_unlock(oldmm);
776 dup_userfaultfd_complete(&uf);
777fail_uprobe_end:
778 uprobe_end_dup_mmap();
779 return retval;
780
781fail_nomem_anon_vma_fork:
782 mpol_put(vma_policy(tmp));
783fail_nomem_policy:
784 vm_area_free(tmp);
785fail_nomem:
786 retval = -ENOMEM;
787 vm_unacct_memory(charge);
788 goto loop_out;
789}
790
791static inline int mm_alloc_pgd(struct mm_struct *mm)
792{
793 mm->pgd = pgd_alloc(mm);
794 if (unlikely(!mm->pgd))
795 return -ENOMEM;
796 return 0;
797}
798
799static inline void mm_free_pgd(struct mm_struct *mm)
800{
801 pgd_free(mm, mm->pgd);
802}
803#else
804static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
805{
806 mmap_write_lock(oldmm);
807 dup_mm_exe_file(mm, oldmm);
808 mmap_write_unlock(oldmm);
809 return 0;
810}
811#define mm_alloc_pgd(mm) (0)
812#define mm_free_pgd(mm)
813#endif /* CONFIG_MMU */
814
815static void check_mm(struct mm_struct *mm)
816{
817 int i;
818
819 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
820 "Please make sure 'struct resident_page_types[]' is updated as well");
821
822 for (i = 0; i < NR_MM_COUNTERS; i++) {
823 long x = percpu_counter_sum(&mm->rss_stat[i]);
824
825 if (unlikely(x))
826 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
827 mm, resident_page_types[i], x);
828 }
829
830 if (mm_pgtables_bytes(mm))
831 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
832 mm_pgtables_bytes(mm));
833
834#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
835 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
836#endif
837}
838
839#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
840#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
841
842static void do_check_lazy_tlb(void *arg)
843{
844 struct mm_struct *mm = arg;
845
846 WARN_ON_ONCE(current->active_mm == mm);
847}
848
849static void do_shoot_lazy_tlb(void *arg)
850{
851 struct mm_struct *mm = arg;
852
853 if (current->active_mm == mm) {
854 WARN_ON_ONCE(current->mm);
855 current->active_mm = &init_mm;
856 switch_mm(mm, &init_mm, current);
857 }
858}
859
860static void cleanup_lazy_tlbs(struct mm_struct *mm)
861{
862 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
863 /*
864 * In this case, lazy tlb mms are refounted and would not reach
865 * __mmdrop until all CPUs have switched away and mmdrop()ed.
866 */
867 return;
868 }
869
870 /*
871 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
872 * requires lazy mm users to switch to another mm when the refcount
873 * drops to zero, before the mm is freed. This requires IPIs here to
874 * switch kernel threads to init_mm.
875 *
876 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
877 * switch with the final userspace teardown TLB flush which leaves the
878 * mm lazy on this CPU but no others, reducing the need for additional
879 * IPIs here. There are cases where a final IPI is still required here,
880 * such as the final mmdrop being performed on a different CPU than the
881 * one exiting, or kernel threads using the mm when userspace exits.
882 *
883 * IPI overheads have not found to be expensive, but they could be
884 * reduced in a number of possible ways, for example (roughly
885 * increasing order of complexity):
886 * - The last lazy reference created by exit_mm() could instead switch
887 * to init_mm, however it's probable this will run on the same CPU
888 * immediately afterwards, so this may not reduce IPIs much.
889 * - A batch of mms requiring IPIs could be gathered and freed at once.
890 * - CPUs store active_mm where it can be remotely checked without a
891 * lock, to filter out false-positives in the cpumask.
892 * - After mm_users or mm_count reaches zero, switching away from the
893 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
894 * with some batching or delaying of the final IPIs.
895 * - A delayed freeing and RCU-like quiescing sequence based on mm
896 * switching to avoid IPIs completely.
897 */
898 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
899 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
900 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
901}
902
903/*
904 * Called when the last reference to the mm
905 * is dropped: either by a lazy thread or by
906 * mmput. Free the page directory and the mm.
907 */
908void __mmdrop(struct mm_struct *mm)
909{
910 BUG_ON(mm == &init_mm);
911 WARN_ON_ONCE(mm == current->mm);
912
913 /* Ensure no CPUs are using this as their lazy tlb mm */
914 cleanup_lazy_tlbs(mm);
915
916 WARN_ON_ONCE(mm == current->active_mm);
917 mm_free_pgd(mm);
918 destroy_context(mm);
919 mmu_notifier_subscriptions_destroy(mm);
920 check_mm(mm);
921 put_user_ns(mm->user_ns);
922 mm_pasid_drop(mm);
923 mm_destroy_cid(mm);
924 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
925
926 free_mm(mm);
927}
928EXPORT_SYMBOL_GPL(__mmdrop);
929
930static void mmdrop_async_fn(struct work_struct *work)
931{
932 struct mm_struct *mm;
933
934 mm = container_of(work, struct mm_struct, async_put_work);
935 __mmdrop(mm);
936}
937
938static void mmdrop_async(struct mm_struct *mm)
939{
940 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
941 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
942 schedule_work(&mm->async_put_work);
943 }
944}
945
946static inline void free_signal_struct(struct signal_struct *sig)
947{
948 taskstats_tgid_free(sig);
949 sched_autogroup_exit(sig);
950 /*
951 * __mmdrop is not safe to call from softirq context on x86 due to
952 * pgd_dtor so postpone it to the async context
953 */
954 if (sig->oom_mm)
955 mmdrop_async(sig->oom_mm);
956 kmem_cache_free(signal_cachep, sig);
957}
958
959static inline void put_signal_struct(struct signal_struct *sig)
960{
961 if (refcount_dec_and_test(&sig->sigcnt))
962 free_signal_struct(sig);
963}
964
965void __put_task_struct(struct task_struct *tsk)
966{
967 WARN_ON(!tsk->exit_state);
968 WARN_ON(refcount_read(&tsk->usage));
969 WARN_ON(tsk == current);
970
971 io_uring_free(tsk);
972 cgroup_free(tsk);
973 task_numa_free(tsk, true);
974 security_task_free(tsk);
975 exit_creds(tsk);
976 delayacct_tsk_free(tsk);
977 put_signal_struct(tsk->signal);
978 sched_core_free(tsk);
979 free_task(tsk);
980}
981EXPORT_SYMBOL_GPL(__put_task_struct);
982
983void __put_task_struct_rcu_cb(struct rcu_head *rhp)
984{
985 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
986
987 __put_task_struct(task);
988}
989EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
990
991void __init __weak arch_task_cache_init(void) { }
992
993/*
994 * set_max_threads
995 */
996static void set_max_threads(unsigned int max_threads_suggested)
997{
998 u64 threads;
999 unsigned long nr_pages = totalram_pages();
1000
1001 /*
1002 * The number of threads shall be limited such that the thread
1003 * structures may only consume a small part of the available memory.
1004 */
1005 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1006 threads = MAX_THREADS;
1007 else
1008 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1009 (u64) THREAD_SIZE * 8UL);
1010
1011 if (threads > max_threads_suggested)
1012 threads = max_threads_suggested;
1013
1014 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1015}
1016
1017#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1018/* Initialized by the architecture: */
1019int arch_task_struct_size __read_mostly;
1020#endif
1021
1022static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1023{
1024 /* Fetch thread_struct whitelist for the architecture. */
1025 arch_thread_struct_whitelist(offset, size);
1026
1027 /*
1028 * Handle zero-sized whitelist or empty thread_struct, otherwise
1029 * adjust offset to position of thread_struct in task_struct.
1030 */
1031 if (unlikely(*size == 0))
1032 *offset = 0;
1033 else
1034 *offset += offsetof(struct task_struct, thread);
1035}
1036
1037void __init fork_init(void)
1038{
1039 int i;
1040#ifndef ARCH_MIN_TASKALIGN
1041#define ARCH_MIN_TASKALIGN 0
1042#endif
1043 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1044 unsigned long useroffset, usersize;
1045
1046 /* create a slab on which task_structs can be allocated */
1047 task_struct_whitelist(&useroffset, &usersize);
1048 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1049 arch_task_struct_size, align,
1050 SLAB_PANIC|SLAB_ACCOUNT,
1051 useroffset, usersize, NULL);
1052
1053 /* do the arch specific task caches init */
1054 arch_task_cache_init();
1055
1056 set_max_threads(MAX_THREADS);
1057
1058 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1059 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1060 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1061 init_task.signal->rlim[RLIMIT_NPROC];
1062
1063 for (i = 0; i < UCOUNT_COUNTS; i++)
1064 init_user_ns.ucount_max[i] = max_threads/2;
1065
1066 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1067 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1068 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1070
1071#ifdef CONFIG_VMAP_STACK
1072 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1073 NULL, free_vm_stack_cache);
1074#endif
1075
1076 scs_init();
1077
1078 lockdep_init_task(&init_task);
1079 uprobes_init();
1080}
1081
1082int __weak arch_dup_task_struct(struct task_struct *dst,
1083 struct task_struct *src)
1084{
1085 *dst = *src;
1086 return 0;
1087}
1088
1089void set_task_stack_end_magic(struct task_struct *tsk)
1090{
1091 unsigned long *stackend;
1092
1093 stackend = end_of_stack(tsk);
1094 *stackend = STACK_END_MAGIC; /* for overflow detection */
1095}
1096
1097static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1098{
1099 struct task_struct *tsk;
1100 int err;
1101
1102 if (node == NUMA_NO_NODE)
1103 node = tsk_fork_get_node(orig);
1104 tsk = alloc_task_struct_node(node);
1105 if (!tsk)
1106 return NULL;
1107
1108 err = arch_dup_task_struct(tsk, orig);
1109 if (err)
1110 goto free_tsk;
1111
1112 err = alloc_thread_stack_node(tsk, node);
1113 if (err)
1114 goto free_tsk;
1115
1116#ifdef CONFIG_THREAD_INFO_IN_TASK
1117 refcount_set(&tsk->stack_refcount, 1);
1118#endif
1119 account_kernel_stack(tsk, 1);
1120
1121 err = scs_prepare(tsk, node);
1122 if (err)
1123 goto free_stack;
1124
1125#ifdef CONFIG_SECCOMP
1126 /*
1127 * We must handle setting up seccomp filters once we're under
1128 * the sighand lock in case orig has changed between now and
1129 * then. Until then, filter must be NULL to avoid messing up
1130 * the usage counts on the error path calling free_task.
1131 */
1132 tsk->seccomp.filter = NULL;
1133#endif
1134
1135 setup_thread_stack(tsk, orig);
1136 clear_user_return_notifier(tsk);
1137 clear_tsk_need_resched(tsk);
1138 set_task_stack_end_magic(tsk);
1139 clear_syscall_work_syscall_user_dispatch(tsk);
1140
1141#ifdef CONFIG_STACKPROTECTOR
1142 tsk->stack_canary = get_random_canary();
1143#endif
1144 if (orig->cpus_ptr == &orig->cpus_mask)
1145 tsk->cpus_ptr = &tsk->cpus_mask;
1146 dup_user_cpus_ptr(tsk, orig, node);
1147
1148 /*
1149 * One for the user space visible state that goes away when reaped.
1150 * One for the scheduler.
1151 */
1152 refcount_set(&tsk->rcu_users, 2);
1153 /* One for the rcu users */
1154 refcount_set(&tsk->usage, 1);
1155#ifdef CONFIG_BLK_DEV_IO_TRACE
1156 tsk->btrace_seq = 0;
1157#endif
1158 tsk->splice_pipe = NULL;
1159 tsk->task_frag.page = NULL;
1160 tsk->wake_q.next = NULL;
1161 tsk->worker_private = NULL;
1162
1163 kcov_task_init(tsk);
1164 kmsan_task_create(tsk);
1165 kmap_local_fork(tsk);
1166
1167#ifdef CONFIG_FAULT_INJECTION
1168 tsk->fail_nth = 0;
1169#endif
1170
1171#ifdef CONFIG_BLK_CGROUP
1172 tsk->throttle_disk = NULL;
1173 tsk->use_memdelay = 0;
1174#endif
1175
1176#ifdef CONFIG_ARCH_HAS_CPU_PASID
1177 tsk->pasid_activated = 0;
1178#endif
1179
1180#ifdef CONFIG_MEMCG
1181 tsk->active_memcg = NULL;
1182#endif
1183
1184#ifdef CONFIG_CPU_SUP_INTEL
1185 tsk->reported_split_lock = 0;
1186#endif
1187
1188#ifdef CONFIG_SCHED_MM_CID
1189 tsk->mm_cid = -1;
1190 tsk->last_mm_cid = -1;
1191 tsk->mm_cid_active = 0;
1192 tsk->migrate_from_cpu = -1;
1193#endif
1194 return tsk;
1195
1196free_stack:
1197 exit_task_stack_account(tsk);
1198 free_thread_stack(tsk);
1199free_tsk:
1200 free_task_struct(tsk);
1201 return NULL;
1202}
1203
1204__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1205
1206static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1207
1208static int __init coredump_filter_setup(char *s)
1209{
1210 default_dump_filter =
1211 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1212 MMF_DUMP_FILTER_MASK;
1213 return 1;
1214}
1215
1216__setup("coredump_filter=", coredump_filter_setup);
1217
1218#include <linux/init_task.h>
1219
1220static void mm_init_aio(struct mm_struct *mm)
1221{
1222#ifdef CONFIG_AIO
1223 spin_lock_init(&mm->ioctx_lock);
1224 mm->ioctx_table = NULL;
1225#endif
1226}
1227
1228static __always_inline void mm_clear_owner(struct mm_struct *mm,
1229 struct task_struct *p)
1230{
1231#ifdef CONFIG_MEMCG
1232 if (mm->owner == p)
1233 WRITE_ONCE(mm->owner, NULL);
1234#endif
1235}
1236
1237static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1238{
1239#ifdef CONFIG_MEMCG
1240 mm->owner = p;
1241#endif
1242}
1243
1244static void mm_init_uprobes_state(struct mm_struct *mm)
1245{
1246#ifdef CONFIG_UPROBES
1247 mm->uprobes_state.xol_area = NULL;
1248#endif
1249}
1250
1251static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1252 struct user_namespace *user_ns)
1253{
1254 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1255 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1256 atomic_set(&mm->mm_users, 1);
1257 atomic_set(&mm->mm_count, 1);
1258 seqcount_init(&mm->write_protect_seq);
1259 mmap_init_lock(mm);
1260 INIT_LIST_HEAD(&mm->mmlist);
1261#ifdef CONFIG_PER_VMA_LOCK
1262 mm->mm_lock_seq = 0;
1263#endif
1264 mm_pgtables_bytes_init(mm);
1265 mm->map_count = 0;
1266 mm->locked_vm = 0;
1267 atomic64_set(&mm->pinned_vm, 0);
1268 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1269 spin_lock_init(&mm->page_table_lock);
1270 spin_lock_init(&mm->arg_lock);
1271 mm_init_cpumask(mm);
1272 mm_init_aio(mm);
1273 mm_init_owner(mm, p);
1274 mm_pasid_init(mm);
1275 RCU_INIT_POINTER(mm->exe_file, NULL);
1276 mmu_notifier_subscriptions_init(mm);
1277 init_tlb_flush_pending(mm);
1278#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1279 mm->pmd_huge_pte = NULL;
1280#endif
1281 mm_init_uprobes_state(mm);
1282 hugetlb_count_init(mm);
1283
1284 if (current->mm) {
1285 mm->flags = mmf_init_flags(current->mm->flags);
1286 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1287 } else {
1288 mm->flags = default_dump_filter;
1289 mm->def_flags = 0;
1290 }
1291
1292 if (mm_alloc_pgd(mm))
1293 goto fail_nopgd;
1294
1295 if (init_new_context(p, mm))
1296 goto fail_nocontext;
1297
1298 if (mm_alloc_cid(mm))
1299 goto fail_cid;
1300
1301 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1302 NR_MM_COUNTERS))
1303 goto fail_pcpu;
1304
1305 mm->user_ns = get_user_ns(user_ns);
1306 lru_gen_init_mm(mm);
1307 return mm;
1308
1309fail_pcpu:
1310 mm_destroy_cid(mm);
1311fail_cid:
1312 destroy_context(mm);
1313fail_nocontext:
1314 mm_free_pgd(mm);
1315fail_nopgd:
1316 free_mm(mm);
1317 return NULL;
1318}
1319
1320/*
1321 * Allocate and initialize an mm_struct.
1322 */
1323struct mm_struct *mm_alloc(void)
1324{
1325 struct mm_struct *mm;
1326
1327 mm = allocate_mm();
1328 if (!mm)
1329 return NULL;
1330
1331 memset(mm, 0, sizeof(*mm));
1332 return mm_init(mm, current, current_user_ns());
1333}
1334
1335static inline void __mmput(struct mm_struct *mm)
1336{
1337 VM_BUG_ON(atomic_read(&mm->mm_users));
1338
1339 uprobe_clear_state(mm);
1340 exit_aio(mm);
1341 ksm_exit(mm);
1342 khugepaged_exit(mm); /* must run before exit_mmap */
1343 exit_mmap(mm);
1344 mm_put_huge_zero_page(mm);
1345 set_mm_exe_file(mm, NULL);
1346 if (!list_empty(&mm->mmlist)) {
1347 spin_lock(&mmlist_lock);
1348 list_del(&mm->mmlist);
1349 spin_unlock(&mmlist_lock);
1350 }
1351 if (mm->binfmt)
1352 module_put(mm->binfmt->module);
1353 lru_gen_del_mm(mm);
1354 mmdrop(mm);
1355}
1356
1357/*
1358 * Decrement the use count and release all resources for an mm.
1359 */
1360void mmput(struct mm_struct *mm)
1361{
1362 might_sleep();
1363
1364 if (atomic_dec_and_test(&mm->mm_users))
1365 __mmput(mm);
1366}
1367EXPORT_SYMBOL_GPL(mmput);
1368
1369#ifdef CONFIG_MMU
1370static void mmput_async_fn(struct work_struct *work)
1371{
1372 struct mm_struct *mm = container_of(work, struct mm_struct,
1373 async_put_work);
1374
1375 __mmput(mm);
1376}
1377
1378void mmput_async(struct mm_struct *mm)
1379{
1380 if (atomic_dec_and_test(&mm->mm_users)) {
1381 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1382 schedule_work(&mm->async_put_work);
1383 }
1384}
1385EXPORT_SYMBOL_GPL(mmput_async);
1386#endif
1387
1388/**
1389 * set_mm_exe_file - change a reference to the mm's executable file
1390 * @mm: The mm to change.
1391 * @new_exe_file: The new file to use.
1392 *
1393 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1394 *
1395 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1396 * invocations: in mmput() nobody alive left, in execve it happens before
1397 * the new mm is made visible to anyone.
1398 *
1399 * Can only fail if new_exe_file != NULL.
1400 */
1401int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1402{
1403 struct file *old_exe_file;
1404
1405 /*
1406 * It is safe to dereference the exe_file without RCU as
1407 * this function is only called if nobody else can access
1408 * this mm -- see comment above for justification.
1409 */
1410 old_exe_file = rcu_dereference_raw(mm->exe_file);
1411
1412 if (new_exe_file) {
1413 /*
1414 * We expect the caller (i.e., sys_execve) to already denied
1415 * write access, so this is unlikely to fail.
1416 */
1417 if (unlikely(deny_write_access(new_exe_file)))
1418 return -EACCES;
1419 get_file(new_exe_file);
1420 }
1421 rcu_assign_pointer(mm->exe_file, new_exe_file);
1422 if (old_exe_file) {
1423 allow_write_access(old_exe_file);
1424 fput(old_exe_file);
1425 }
1426 return 0;
1427}
1428
1429/**
1430 * replace_mm_exe_file - replace a reference to the mm's executable file
1431 * @mm: The mm to change.
1432 * @new_exe_file: The new file to use.
1433 *
1434 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1435 *
1436 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1437 */
1438int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1439{
1440 struct vm_area_struct *vma;
1441 struct file *old_exe_file;
1442 int ret = 0;
1443
1444 /* Forbid mm->exe_file change if old file still mapped. */
1445 old_exe_file = get_mm_exe_file(mm);
1446 if (old_exe_file) {
1447 VMA_ITERATOR(vmi, mm, 0);
1448 mmap_read_lock(mm);
1449 for_each_vma(vmi, vma) {
1450 if (!vma->vm_file)
1451 continue;
1452 if (path_equal(&vma->vm_file->f_path,
1453 &old_exe_file->f_path)) {
1454 ret = -EBUSY;
1455 break;
1456 }
1457 }
1458 mmap_read_unlock(mm);
1459 fput(old_exe_file);
1460 if (ret)
1461 return ret;
1462 }
1463
1464 ret = deny_write_access(new_exe_file);
1465 if (ret)
1466 return -EACCES;
1467 get_file(new_exe_file);
1468
1469 /* set the new file */
1470 mmap_write_lock(mm);
1471 old_exe_file = rcu_dereference_raw(mm->exe_file);
1472 rcu_assign_pointer(mm->exe_file, new_exe_file);
1473 mmap_write_unlock(mm);
1474
1475 if (old_exe_file) {
1476 allow_write_access(old_exe_file);
1477 fput(old_exe_file);
1478 }
1479 return 0;
1480}
1481
1482/**
1483 * get_mm_exe_file - acquire a reference to the mm's executable file
1484 * @mm: The mm of interest.
1485 *
1486 * Returns %NULL if mm has no associated executable file.
1487 * User must release file via fput().
1488 */
1489struct file *get_mm_exe_file(struct mm_struct *mm)
1490{
1491 struct file *exe_file;
1492
1493 rcu_read_lock();
1494 exe_file = get_file_rcu(&mm->exe_file);
1495 rcu_read_unlock();
1496 return exe_file;
1497}
1498
1499/**
1500 * get_task_exe_file - acquire a reference to the task's executable file
1501 * @task: The task.
1502 *
1503 * Returns %NULL if task's mm (if any) has no associated executable file or
1504 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1505 * User must release file via fput().
1506 */
1507struct file *get_task_exe_file(struct task_struct *task)
1508{
1509 struct file *exe_file = NULL;
1510 struct mm_struct *mm;
1511
1512 task_lock(task);
1513 mm = task->mm;
1514 if (mm) {
1515 if (!(task->flags & PF_KTHREAD))
1516 exe_file = get_mm_exe_file(mm);
1517 }
1518 task_unlock(task);
1519 return exe_file;
1520}
1521
1522/**
1523 * get_task_mm - acquire a reference to the task's mm
1524 * @task: The task.
1525 *
1526 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1527 * this kernel workthread has transiently adopted a user mm with use_mm,
1528 * to do its AIO) is not set and if so returns a reference to it, after
1529 * bumping up the use count. User must release the mm via mmput()
1530 * after use. Typically used by /proc and ptrace.
1531 */
1532struct mm_struct *get_task_mm(struct task_struct *task)
1533{
1534 struct mm_struct *mm;
1535
1536 task_lock(task);
1537 mm = task->mm;
1538 if (mm) {
1539 if (task->flags & PF_KTHREAD)
1540 mm = NULL;
1541 else
1542 mmget(mm);
1543 }
1544 task_unlock(task);
1545 return mm;
1546}
1547EXPORT_SYMBOL_GPL(get_task_mm);
1548
1549struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1550{
1551 struct mm_struct *mm;
1552 int err;
1553
1554 err = down_read_killable(&task->signal->exec_update_lock);
1555 if (err)
1556 return ERR_PTR(err);
1557
1558 mm = get_task_mm(task);
1559 if (mm && mm != current->mm &&
1560 !ptrace_may_access(task, mode)) {
1561 mmput(mm);
1562 mm = ERR_PTR(-EACCES);
1563 }
1564 up_read(&task->signal->exec_update_lock);
1565
1566 return mm;
1567}
1568
1569static void complete_vfork_done(struct task_struct *tsk)
1570{
1571 struct completion *vfork;
1572
1573 task_lock(tsk);
1574 vfork = tsk->vfork_done;
1575 if (likely(vfork)) {
1576 tsk->vfork_done = NULL;
1577 complete(vfork);
1578 }
1579 task_unlock(tsk);
1580}
1581
1582static int wait_for_vfork_done(struct task_struct *child,
1583 struct completion *vfork)
1584{
1585 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1586 int killed;
1587
1588 cgroup_enter_frozen();
1589 killed = wait_for_completion_state(vfork, state);
1590 cgroup_leave_frozen(false);
1591
1592 if (killed) {
1593 task_lock(child);
1594 child->vfork_done = NULL;
1595 task_unlock(child);
1596 }
1597
1598 put_task_struct(child);
1599 return killed;
1600}
1601
1602/* Please note the differences between mmput and mm_release.
1603 * mmput is called whenever we stop holding onto a mm_struct,
1604 * error success whatever.
1605 *
1606 * mm_release is called after a mm_struct has been removed
1607 * from the current process.
1608 *
1609 * This difference is important for error handling, when we
1610 * only half set up a mm_struct for a new process and need to restore
1611 * the old one. Because we mmput the new mm_struct before
1612 * restoring the old one. . .
1613 * Eric Biederman 10 January 1998
1614 */
1615static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1616{
1617 uprobe_free_utask(tsk);
1618
1619 /* Get rid of any cached register state */
1620 deactivate_mm(tsk, mm);
1621
1622 /*
1623 * Signal userspace if we're not exiting with a core dump
1624 * because we want to leave the value intact for debugging
1625 * purposes.
1626 */
1627 if (tsk->clear_child_tid) {
1628 if (atomic_read(&mm->mm_users) > 1) {
1629 /*
1630 * We don't check the error code - if userspace has
1631 * not set up a proper pointer then tough luck.
1632 */
1633 put_user(0, tsk->clear_child_tid);
1634 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1635 1, NULL, NULL, 0, 0);
1636 }
1637 tsk->clear_child_tid = NULL;
1638 }
1639
1640 /*
1641 * All done, finally we can wake up parent and return this mm to him.
1642 * Also kthread_stop() uses this completion for synchronization.
1643 */
1644 if (tsk->vfork_done)
1645 complete_vfork_done(tsk);
1646}
1647
1648void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1649{
1650 futex_exit_release(tsk);
1651 mm_release(tsk, mm);
1652}
1653
1654void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1655{
1656 futex_exec_release(tsk);
1657 mm_release(tsk, mm);
1658}
1659
1660/**
1661 * dup_mm() - duplicates an existing mm structure
1662 * @tsk: the task_struct with which the new mm will be associated.
1663 * @oldmm: the mm to duplicate.
1664 *
1665 * Allocates a new mm structure and duplicates the provided @oldmm structure
1666 * content into it.
1667 *
1668 * Return: the duplicated mm or NULL on failure.
1669 */
1670static struct mm_struct *dup_mm(struct task_struct *tsk,
1671 struct mm_struct *oldmm)
1672{
1673 struct mm_struct *mm;
1674 int err;
1675
1676 mm = allocate_mm();
1677 if (!mm)
1678 goto fail_nomem;
1679
1680 memcpy(mm, oldmm, sizeof(*mm));
1681
1682 if (!mm_init(mm, tsk, mm->user_ns))
1683 goto fail_nomem;
1684
1685 err = dup_mmap(mm, oldmm);
1686 if (err)
1687 goto free_pt;
1688
1689 mm->hiwater_rss = get_mm_rss(mm);
1690 mm->hiwater_vm = mm->total_vm;
1691
1692 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1693 goto free_pt;
1694
1695 return mm;
1696
1697free_pt:
1698 /* don't put binfmt in mmput, we haven't got module yet */
1699 mm->binfmt = NULL;
1700 mm_init_owner(mm, NULL);
1701 mmput(mm);
1702
1703fail_nomem:
1704 return NULL;
1705}
1706
1707static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1708{
1709 struct mm_struct *mm, *oldmm;
1710
1711 tsk->min_flt = tsk->maj_flt = 0;
1712 tsk->nvcsw = tsk->nivcsw = 0;
1713#ifdef CONFIG_DETECT_HUNG_TASK
1714 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1715 tsk->last_switch_time = 0;
1716#endif
1717
1718 tsk->mm = NULL;
1719 tsk->active_mm = NULL;
1720
1721 /*
1722 * Are we cloning a kernel thread?
1723 *
1724 * We need to steal a active VM for that..
1725 */
1726 oldmm = current->mm;
1727 if (!oldmm)
1728 return 0;
1729
1730 if (clone_flags & CLONE_VM) {
1731 mmget(oldmm);
1732 mm = oldmm;
1733 } else {
1734 mm = dup_mm(tsk, current->mm);
1735 if (!mm)
1736 return -ENOMEM;
1737 }
1738
1739 tsk->mm = mm;
1740 tsk->active_mm = mm;
1741 sched_mm_cid_fork(tsk);
1742 return 0;
1743}
1744
1745static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1746{
1747 struct fs_struct *fs = current->fs;
1748 if (clone_flags & CLONE_FS) {
1749 /* tsk->fs is already what we want */
1750 spin_lock(&fs->lock);
1751 /* "users" and "in_exec" locked for check_unsafe_exec() */
1752 if (fs->in_exec) {
1753 spin_unlock(&fs->lock);
1754 return -EAGAIN;
1755 }
1756 fs->users++;
1757 spin_unlock(&fs->lock);
1758 return 0;
1759 }
1760 tsk->fs = copy_fs_struct(fs);
1761 if (!tsk->fs)
1762 return -ENOMEM;
1763 return 0;
1764}
1765
1766static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1767 int no_files)
1768{
1769 struct files_struct *oldf, *newf;
1770 int error = 0;
1771
1772 /*
1773 * A background process may not have any files ...
1774 */
1775 oldf = current->files;
1776 if (!oldf)
1777 goto out;
1778
1779 if (no_files) {
1780 tsk->files = NULL;
1781 goto out;
1782 }
1783
1784 if (clone_flags & CLONE_FILES) {
1785 atomic_inc(&oldf->count);
1786 goto out;
1787 }
1788
1789 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1790 if (!newf)
1791 goto out;
1792
1793 tsk->files = newf;
1794 error = 0;
1795out:
1796 return error;
1797}
1798
1799static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1800{
1801 struct sighand_struct *sig;
1802
1803 if (clone_flags & CLONE_SIGHAND) {
1804 refcount_inc(¤t->sighand->count);
1805 return 0;
1806 }
1807 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1808 RCU_INIT_POINTER(tsk->sighand, sig);
1809 if (!sig)
1810 return -ENOMEM;
1811
1812 refcount_set(&sig->count, 1);
1813 spin_lock_irq(¤t->sighand->siglock);
1814 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1815 spin_unlock_irq(¤t->sighand->siglock);
1816
1817 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1818 if (clone_flags & CLONE_CLEAR_SIGHAND)
1819 flush_signal_handlers(tsk, 0);
1820
1821 return 0;
1822}
1823
1824void __cleanup_sighand(struct sighand_struct *sighand)
1825{
1826 if (refcount_dec_and_test(&sighand->count)) {
1827 signalfd_cleanup(sighand);
1828 /*
1829 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1830 * without an RCU grace period, see __lock_task_sighand().
1831 */
1832 kmem_cache_free(sighand_cachep, sighand);
1833 }
1834}
1835
1836/*
1837 * Initialize POSIX timer handling for a thread group.
1838 */
1839static void posix_cpu_timers_init_group(struct signal_struct *sig)
1840{
1841 struct posix_cputimers *pct = &sig->posix_cputimers;
1842 unsigned long cpu_limit;
1843
1844 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1845 posix_cputimers_group_init(pct, cpu_limit);
1846}
1847
1848static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1849{
1850 struct signal_struct *sig;
1851
1852 if (clone_flags & CLONE_THREAD)
1853 return 0;
1854
1855 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1856 tsk->signal = sig;
1857 if (!sig)
1858 return -ENOMEM;
1859
1860 sig->nr_threads = 1;
1861 sig->quick_threads = 1;
1862 atomic_set(&sig->live, 1);
1863 refcount_set(&sig->sigcnt, 1);
1864
1865 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1866 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1867 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1868
1869 init_waitqueue_head(&sig->wait_chldexit);
1870 sig->curr_target = tsk;
1871 init_sigpending(&sig->shared_pending);
1872 INIT_HLIST_HEAD(&sig->multiprocess);
1873 seqlock_init(&sig->stats_lock);
1874 prev_cputime_init(&sig->prev_cputime);
1875
1876#ifdef CONFIG_POSIX_TIMERS
1877 INIT_LIST_HEAD(&sig->posix_timers);
1878 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1879 sig->real_timer.function = it_real_fn;
1880#endif
1881
1882 task_lock(current->group_leader);
1883 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1884 task_unlock(current->group_leader);
1885
1886 posix_cpu_timers_init_group(sig);
1887
1888 tty_audit_fork(sig);
1889 sched_autogroup_fork(sig);
1890
1891 sig->oom_score_adj = current->signal->oom_score_adj;
1892 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1893
1894 mutex_init(&sig->cred_guard_mutex);
1895 init_rwsem(&sig->exec_update_lock);
1896
1897 return 0;
1898}
1899
1900static void copy_seccomp(struct task_struct *p)
1901{
1902#ifdef CONFIG_SECCOMP
1903 /*
1904 * Must be called with sighand->lock held, which is common to
1905 * all threads in the group. Holding cred_guard_mutex is not
1906 * needed because this new task is not yet running and cannot
1907 * be racing exec.
1908 */
1909 assert_spin_locked(¤t->sighand->siglock);
1910
1911 /* Ref-count the new filter user, and assign it. */
1912 get_seccomp_filter(current);
1913 p->seccomp = current->seccomp;
1914
1915 /*
1916 * Explicitly enable no_new_privs here in case it got set
1917 * between the task_struct being duplicated and holding the
1918 * sighand lock. The seccomp state and nnp must be in sync.
1919 */
1920 if (task_no_new_privs(current))
1921 task_set_no_new_privs(p);
1922
1923 /*
1924 * If the parent gained a seccomp mode after copying thread
1925 * flags and between before we held the sighand lock, we have
1926 * to manually enable the seccomp thread flag here.
1927 */
1928 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1929 set_task_syscall_work(p, SECCOMP);
1930#endif
1931}
1932
1933SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1934{
1935 current->clear_child_tid = tidptr;
1936
1937 return task_pid_vnr(current);
1938}
1939
1940static void rt_mutex_init_task(struct task_struct *p)
1941{
1942 raw_spin_lock_init(&p->pi_lock);
1943#ifdef CONFIG_RT_MUTEXES
1944 p->pi_waiters = RB_ROOT_CACHED;
1945 p->pi_top_task = NULL;
1946 p->pi_blocked_on = NULL;
1947#endif
1948}
1949
1950static inline void init_task_pid_links(struct task_struct *task)
1951{
1952 enum pid_type type;
1953
1954 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1955 INIT_HLIST_NODE(&task->pid_links[type]);
1956}
1957
1958static inline void
1959init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1960{
1961 if (type == PIDTYPE_PID)
1962 task->thread_pid = pid;
1963 else
1964 task->signal->pids[type] = pid;
1965}
1966
1967static inline void rcu_copy_process(struct task_struct *p)
1968{
1969#ifdef CONFIG_PREEMPT_RCU
1970 p->rcu_read_lock_nesting = 0;
1971 p->rcu_read_unlock_special.s = 0;
1972 p->rcu_blocked_node = NULL;
1973 INIT_LIST_HEAD(&p->rcu_node_entry);
1974#endif /* #ifdef CONFIG_PREEMPT_RCU */
1975#ifdef CONFIG_TASKS_RCU
1976 p->rcu_tasks_holdout = false;
1977 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1978 p->rcu_tasks_idle_cpu = -1;
1979#endif /* #ifdef CONFIG_TASKS_RCU */
1980#ifdef CONFIG_TASKS_TRACE_RCU
1981 p->trc_reader_nesting = 0;
1982 p->trc_reader_special.s = 0;
1983 INIT_LIST_HEAD(&p->trc_holdout_list);
1984 INIT_LIST_HEAD(&p->trc_blkd_node);
1985#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1986}
1987
1988struct pid *pidfd_pid(const struct file *file)
1989{
1990 if (file->f_op == &pidfd_fops)
1991 return file->private_data;
1992
1993 return ERR_PTR(-EBADF);
1994}
1995
1996static int pidfd_release(struct inode *inode, struct file *file)
1997{
1998 struct pid *pid = file->private_data;
1999
2000 file->private_data = NULL;
2001 put_pid(pid);
2002 return 0;
2003}
2004
2005#ifdef CONFIG_PROC_FS
2006/**
2007 * pidfd_show_fdinfo - print information about a pidfd
2008 * @m: proc fdinfo file
2009 * @f: file referencing a pidfd
2010 *
2011 * Pid:
2012 * This function will print the pid that a given pidfd refers to in the
2013 * pid namespace of the procfs instance.
2014 * If the pid namespace of the process is not a descendant of the pid
2015 * namespace of the procfs instance 0 will be shown as its pid. This is
2016 * similar to calling getppid() on a process whose parent is outside of
2017 * its pid namespace.
2018 *
2019 * NSpid:
2020 * If pid namespaces are supported then this function will also print
2021 * the pid of a given pidfd refers to for all descendant pid namespaces
2022 * starting from the current pid namespace of the instance, i.e. the
2023 * Pid field and the first entry in the NSpid field will be identical.
2024 * If the pid namespace of the process is not a descendant of the pid
2025 * namespace of the procfs instance 0 will be shown as its first NSpid
2026 * entry and no others will be shown.
2027 * Note that this differs from the Pid and NSpid fields in
2028 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2029 * the pid namespace of the procfs instance. The difference becomes
2030 * obvious when sending around a pidfd between pid namespaces from a
2031 * different branch of the tree, i.e. where no ancestral relation is
2032 * present between the pid namespaces:
2033 * - create two new pid namespaces ns1 and ns2 in the initial pid
2034 * namespace (also take care to create new mount namespaces in the
2035 * new pid namespace and mount procfs)
2036 * - create a process with a pidfd in ns1
2037 * - send pidfd from ns1 to ns2
2038 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2039 * have exactly one entry, which is 0
2040 */
2041static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2042{
2043 struct pid *pid = f->private_data;
2044 struct pid_namespace *ns;
2045 pid_t nr = -1;
2046
2047 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2048 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2049 nr = pid_nr_ns(pid, ns);
2050 }
2051
2052 seq_put_decimal_ll(m, "Pid:\t", nr);
2053
2054#ifdef CONFIG_PID_NS
2055 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2056 if (nr > 0) {
2057 int i;
2058
2059 /* If nr is non-zero it means that 'pid' is valid and that
2060 * ns, i.e. the pid namespace associated with the procfs
2061 * instance, is in the pid namespace hierarchy of pid.
2062 * Start at one below the already printed level.
2063 */
2064 for (i = ns->level + 1; i <= pid->level; i++)
2065 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2066 }
2067#endif
2068 seq_putc(m, '\n');
2069}
2070#endif
2071
2072/*
2073 * Poll support for process exit notification.
2074 */
2075static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2076{
2077 struct pid *pid = file->private_data;
2078 __poll_t poll_flags = 0;
2079
2080 poll_wait(file, &pid->wait_pidfd, pts);
2081
2082 /*
2083 * Inform pollers only when the whole thread group exits.
2084 * If the thread group leader exits before all other threads in the
2085 * group, then poll(2) should block, similar to the wait(2) family.
2086 */
2087 if (thread_group_exited(pid))
2088 poll_flags = EPOLLIN | EPOLLRDNORM;
2089
2090 return poll_flags;
2091}
2092
2093const struct file_operations pidfd_fops = {
2094 .release = pidfd_release,
2095 .poll = pidfd_poll,
2096#ifdef CONFIG_PROC_FS
2097 .show_fdinfo = pidfd_show_fdinfo,
2098#endif
2099};
2100
2101/**
2102 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2103 * @pid: the struct pid for which to create a pidfd
2104 * @flags: flags of the new @pidfd
2105 * @ret: Where to return the file for the pidfd.
2106 *
2107 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2108 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2109 *
2110 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2111 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2112 * pidfd file are prepared.
2113 *
2114 * If this function returns successfully the caller is responsible to either
2115 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2116 * order to install the pidfd into its file descriptor table or they must use
2117 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2118 * respectively.
2119 *
2120 * This function is useful when a pidfd must already be reserved but there
2121 * might still be points of failure afterwards and the caller wants to ensure
2122 * that no pidfd is leaked into its file descriptor table.
2123 *
2124 * Return: On success, a reserved pidfd is returned from the function and a new
2125 * pidfd file is returned in the last argument to the function. On
2126 * error, a negative error code is returned from the function and the
2127 * last argument remains unchanged.
2128 */
2129static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2130{
2131 int pidfd;
2132 struct file *pidfd_file;
2133
2134 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2135 return -EINVAL;
2136
2137 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2138 if (pidfd < 0)
2139 return pidfd;
2140
2141 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2142 flags | O_RDWR | O_CLOEXEC);
2143 if (IS_ERR(pidfd_file)) {
2144 put_unused_fd(pidfd);
2145 return PTR_ERR(pidfd_file);
2146 }
2147 get_pid(pid); /* held by pidfd_file now */
2148 *ret = pidfd_file;
2149 return pidfd;
2150}
2151
2152/**
2153 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2154 * @pid: the struct pid for which to create a pidfd
2155 * @flags: flags of the new @pidfd
2156 * @ret: Where to return the pidfd.
2157 *
2158 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2159 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2160 *
2161 * The helper verifies that @pid is used as a thread group leader.
2162 *
2163 * If this function returns successfully the caller is responsible to either
2164 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2165 * order to install the pidfd into its file descriptor table or they must use
2166 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2167 * respectively.
2168 *
2169 * This function is useful when a pidfd must already be reserved but there
2170 * might still be points of failure afterwards and the caller wants to ensure
2171 * that no pidfd is leaked into its file descriptor table.
2172 *
2173 * Return: On success, a reserved pidfd is returned from the function and a new
2174 * pidfd file is returned in the last argument to the function. On
2175 * error, a negative error code is returned from the function and the
2176 * last argument remains unchanged.
2177 */
2178int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2179{
2180 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2181 return -EINVAL;
2182
2183 return __pidfd_prepare(pid, flags, ret);
2184}
2185
2186static void __delayed_free_task(struct rcu_head *rhp)
2187{
2188 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2189
2190 free_task(tsk);
2191}
2192
2193static __always_inline void delayed_free_task(struct task_struct *tsk)
2194{
2195 if (IS_ENABLED(CONFIG_MEMCG))
2196 call_rcu(&tsk->rcu, __delayed_free_task);
2197 else
2198 free_task(tsk);
2199}
2200
2201static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2202{
2203 /* Skip if kernel thread */
2204 if (!tsk->mm)
2205 return;
2206
2207 /* Skip if spawning a thread or using vfork */
2208 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2209 return;
2210
2211 /* We need to synchronize with __set_oom_adj */
2212 mutex_lock(&oom_adj_mutex);
2213 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2214 /* Update the values in case they were changed after copy_signal */
2215 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2216 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2217 mutex_unlock(&oom_adj_mutex);
2218}
2219
2220#ifdef CONFIG_RV
2221static void rv_task_fork(struct task_struct *p)
2222{
2223 int i;
2224
2225 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2226 p->rv[i].da_mon.monitoring = false;
2227}
2228#else
2229#define rv_task_fork(p) do {} while (0)
2230#endif
2231
2232/*
2233 * This creates a new process as a copy of the old one,
2234 * but does not actually start it yet.
2235 *
2236 * It copies the registers, and all the appropriate
2237 * parts of the process environment (as per the clone
2238 * flags). The actual kick-off is left to the caller.
2239 */
2240__latent_entropy struct task_struct *copy_process(
2241 struct pid *pid,
2242 int trace,
2243 int node,
2244 struct kernel_clone_args *args)
2245{
2246 int pidfd = -1, retval;
2247 struct task_struct *p;
2248 struct multiprocess_signals delayed;
2249 struct file *pidfile = NULL;
2250 const u64 clone_flags = args->flags;
2251 struct nsproxy *nsp = current->nsproxy;
2252
2253 /*
2254 * Don't allow sharing the root directory with processes in a different
2255 * namespace
2256 */
2257 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2258 return ERR_PTR(-EINVAL);
2259
2260 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2261 return ERR_PTR(-EINVAL);
2262
2263 /*
2264 * Thread groups must share signals as well, and detached threads
2265 * can only be started up within the thread group.
2266 */
2267 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2268 return ERR_PTR(-EINVAL);
2269
2270 /*
2271 * Shared signal handlers imply shared VM. By way of the above,
2272 * thread groups also imply shared VM. Blocking this case allows
2273 * for various simplifications in other code.
2274 */
2275 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2276 return ERR_PTR(-EINVAL);
2277
2278 /*
2279 * Siblings of global init remain as zombies on exit since they are
2280 * not reaped by their parent (swapper). To solve this and to avoid
2281 * multi-rooted process trees, prevent global and container-inits
2282 * from creating siblings.
2283 */
2284 if ((clone_flags & CLONE_PARENT) &&
2285 current->signal->flags & SIGNAL_UNKILLABLE)
2286 return ERR_PTR(-EINVAL);
2287
2288 /*
2289 * If the new process will be in a different pid or user namespace
2290 * do not allow it to share a thread group with the forking task.
2291 */
2292 if (clone_flags & CLONE_THREAD) {
2293 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2294 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2295 return ERR_PTR(-EINVAL);
2296 }
2297
2298 if (clone_flags & CLONE_PIDFD) {
2299 /*
2300 * - CLONE_DETACHED is blocked so that we can potentially
2301 * reuse it later for CLONE_PIDFD.
2302 * - CLONE_THREAD is blocked until someone really needs it.
2303 */
2304 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2305 return ERR_PTR(-EINVAL);
2306 }
2307
2308 /*
2309 * Force any signals received before this point to be delivered
2310 * before the fork happens. Collect up signals sent to multiple
2311 * processes that happen during the fork and delay them so that
2312 * they appear to happen after the fork.
2313 */
2314 sigemptyset(&delayed.signal);
2315 INIT_HLIST_NODE(&delayed.node);
2316
2317 spin_lock_irq(¤t->sighand->siglock);
2318 if (!(clone_flags & CLONE_THREAD))
2319 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2320 recalc_sigpending();
2321 spin_unlock_irq(¤t->sighand->siglock);
2322 retval = -ERESTARTNOINTR;
2323 if (task_sigpending(current))
2324 goto fork_out;
2325
2326 retval = -ENOMEM;
2327 p = dup_task_struct(current, node);
2328 if (!p)
2329 goto fork_out;
2330 p->flags &= ~PF_KTHREAD;
2331 if (args->kthread)
2332 p->flags |= PF_KTHREAD;
2333 if (args->user_worker) {
2334 /*
2335 * Mark us a user worker, and block any signal that isn't
2336 * fatal or STOP
2337 */
2338 p->flags |= PF_USER_WORKER;
2339 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2340 }
2341 if (args->io_thread)
2342 p->flags |= PF_IO_WORKER;
2343
2344 if (args->name)
2345 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2346
2347 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2348 /*
2349 * Clear TID on mm_release()?
2350 */
2351 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2352
2353 ftrace_graph_init_task(p);
2354
2355 rt_mutex_init_task(p);
2356
2357 lockdep_assert_irqs_enabled();
2358#ifdef CONFIG_PROVE_LOCKING
2359 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2360#endif
2361 retval = copy_creds(p, clone_flags);
2362 if (retval < 0)
2363 goto bad_fork_free;
2364
2365 retval = -EAGAIN;
2366 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2367 if (p->real_cred->user != INIT_USER &&
2368 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2369 goto bad_fork_cleanup_count;
2370 }
2371 current->flags &= ~PF_NPROC_EXCEEDED;
2372
2373 /*
2374 * If multiple threads are within copy_process(), then this check
2375 * triggers too late. This doesn't hurt, the check is only there
2376 * to stop root fork bombs.
2377 */
2378 retval = -EAGAIN;
2379 if (data_race(nr_threads >= max_threads))
2380 goto bad_fork_cleanup_count;
2381
2382 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2383 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2384 p->flags |= PF_FORKNOEXEC;
2385 INIT_LIST_HEAD(&p->children);
2386 INIT_LIST_HEAD(&p->sibling);
2387 rcu_copy_process(p);
2388 p->vfork_done = NULL;
2389 spin_lock_init(&p->alloc_lock);
2390
2391 init_sigpending(&p->pending);
2392
2393 p->utime = p->stime = p->gtime = 0;
2394#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2395 p->utimescaled = p->stimescaled = 0;
2396#endif
2397 prev_cputime_init(&p->prev_cputime);
2398
2399#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2400 seqcount_init(&p->vtime.seqcount);
2401 p->vtime.starttime = 0;
2402 p->vtime.state = VTIME_INACTIVE;
2403#endif
2404
2405#ifdef CONFIG_IO_URING
2406 p->io_uring = NULL;
2407#endif
2408
2409 p->default_timer_slack_ns = current->timer_slack_ns;
2410
2411#ifdef CONFIG_PSI
2412 p->psi_flags = 0;
2413#endif
2414
2415 task_io_accounting_init(&p->ioac);
2416 acct_clear_integrals(p);
2417
2418 posix_cputimers_init(&p->posix_cputimers);
2419
2420 p->io_context = NULL;
2421 audit_set_context(p, NULL);
2422 cgroup_fork(p);
2423 if (args->kthread) {
2424 if (!set_kthread_struct(p))
2425 goto bad_fork_cleanup_delayacct;
2426 }
2427#ifdef CONFIG_NUMA
2428 p->mempolicy = mpol_dup(p->mempolicy);
2429 if (IS_ERR(p->mempolicy)) {
2430 retval = PTR_ERR(p->mempolicy);
2431 p->mempolicy = NULL;
2432 goto bad_fork_cleanup_delayacct;
2433 }
2434#endif
2435#ifdef CONFIG_CPUSETS
2436 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2437 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2438 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2439#endif
2440#ifdef CONFIG_TRACE_IRQFLAGS
2441 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2442 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2443 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2444 p->softirqs_enabled = 1;
2445 p->softirq_context = 0;
2446#endif
2447
2448 p->pagefault_disabled = 0;
2449
2450#ifdef CONFIG_LOCKDEP
2451 lockdep_init_task(p);
2452#endif
2453
2454#ifdef CONFIG_DEBUG_MUTEXES
2455 p->blocked_on = NULL; /* not blocked yet */
2456#endif
2457#ifdef CONFIG_BCACHE
2458 p->sequential_io = 0;
2459 p->sequential_io_avg = 0;
2460#endif
2461#ifdef CONFIG_BPF_SYSCALL
2462 RCU_INIT_POINTER(p->bpf_storage, NULL);
2463 p->bpf_ctx = NULL;
2464#endif
2465
2466 /* Perform scheduler related setup. Assign this task to a CPU. */
2467 retval = sched_fork(clone_flags, p);
2468 if (retval)
2469 goto bad_fork_cleanup_policy;
2470
2471 retval = perf_event_init_task(p, clone_flags);
2472 if (retval)
2473 goto bad_fork_cleanup_policy;
2474 retval = audit_alloc(p);
2475 if (retval)
2476 goto bad_fork_cleanup_perf;
2477 /* copy all the process information */
2478 shm_init_task(p);
2479 retval = security_task_alloc(p, clone_flags);
2480 if (retval)
2481 goto bad_fork_cleanup_audit;
2482 retval = copy_semundo(clone_flags, p);
2483 if (retval)
2484 goto bad_fork_cleanup_security;
2485 retval = copy_files(clone_flags, p, args->no_files);
2486 if (retval)
2487 goto bad_fork_cleanup_semundo;
2488 retval = copy_fs(clone_flags, p);
2489 if (retval)
2490 goto bad_fork_cleanup_files;
2491 retval = copy_sighand(clone_flags, p);
2492 if (retval)
2493 goto bad_fork_cleanup_fs;
2494 retval = copy_signal(clone_flags, p);
2495 if (retval)
2496 goto bad_fork_cleanup_sighand;
2497 retval = copy_mm(clone_flags, p);
2498 if (retval)
2499 goto bad_fork_cleanup_signal;
2500 retval = copy_namespaces(clone_flags, p);
2501 if (retval)
2502 goto bad_fork_cleanup_mm;
2503 retval = copy_io(clone_flags, p);
2504 if (retval)
2505 goto bad_fork_cleanup_namespaces;
2506 retval = copy_thread(p, args);
2507 if (retval)
2508 goto bad_fork_cleanup_io;
2509
2510 stackleak_task_init(p);
2511
2512 if (pid != &init_struct_pid) {
2513 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2514 args->set_tid_size);
2515 if (IS_ERR(pid)) {
2516 retval = PTR_ERR(pid);
2517 goto bad_fork_cleanup_thread;
2518 }
2519 }
2520
2521 /*
2522 * This has to happen after we've potentially unshared the file
2523 * descriptor table (so that the pidfd doesn't leak into the child
2524 * if the fd table isn't shared).
2525 */
2526 if (clone_flags & CLONE_PIDFD) {
2527 /* Note that no task has been attached to @pid yet. */
2528 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2529 if (retval < 0)
2530 goto bad_fork_free_pid;
2531 pidfd = retval;
2532
2533 retval = put_user(pidfd, args->pidfd);
2534 if (retval)
2535 goto bad_fork_put_pidfd;
2536 }
2537
2538#ifdef CONFIG_BLOCK
2539 p->plug = NULL;
2540#endif
2541 futex_init_task(p);
2542
2543 /*
2544 * sigaltstack should be cleared when sharing the same VM
2545 */
2546 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2547 sas_ss_reset(p);
2548
2549 /*
2550 * Syscall tracing and stepping should be turned off in the
2551 * child regardless of CLONE_PTRACE.
2552 */
2553 user_disable_single_step(p);
2554 clear_task_syscall_work(p, SYSCALL_TRACE);
2555#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2556 clear_task_syscall_work(p, SYSCALL_EMU);
2557#endif
2558 clear_tsk_latency_tracing(p);
2559
2560 /* ok, now we should be set up.. */
2561 p->pid = pid_nr(pid);
2562 if (clone_flags & CLONE_THREAD) {
2563 p->group_leader = current->group_leader;
2564 p->tgid = current->tgid;
2565 } else {
2566 p->group_leader = p;
2567 p->tgid = p->pid;
2568 }
2569
2570 p->nr_dirtied = 0;
2571 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2572 p->dirty_paused_when = 0;
2573
2574 p->pdeath_signal = 0;
2575 p->task_works = NULL;
2576 clear_posix_cputimers_work(p);
2577
2578#ifdef CONFIG_KRETPROBES
2579 p->kretprobe_instances.first = NULL;
2580#endif
2581#ifdef CONFIG_RETHOOK
2582 p->rethooks.first = NULL;
2583#endif
2584
2585 /*
2586 * Ensure that the cgroup subsystem policies allow the new process to be
2587 * forked. It should be noted that the new process's css_set can be changed
2588 * between here and cgroup_post_fork() if an organisation operation is in
2589 * progress.
2590 */
2591 retval = cgroup_can_fork(p, args);
2592 if (retval)
2593 goto bad_fork_put_pidfd;
2594
2595 /*
2596 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2597 * the new task on the correct runqueue. All this *before* the task
2598 * becomes visible.
2599 *
2600 * This isn't part of ->can_fork() because while the re-cloning is
2601 * cgroup specific, it unconditionally needs to place the task on a
2602 * runqueue.
2603 */
2604 sched_cgroup_fork(p, args);
2605
2606 /*
2607 * From this point on we must avoid any synchronous user-space
2608 * communication until we take the tasklist-lock. In particular, we do
2609 * not want user-space to be able to predict the process start-time by
2610 * stalling fork(2) after we recorded the start_time but before it is
2611 * visible to the system.
2612 */
2613
2614 p->start_time = ktime_get_ns();
2615 p->start_boottime = ktime_get_boottime_ns();
2616
2617 /*
2618 * Make it visible to the rest of the system, but dont wake it up yet.
2619 * Need tasklist lock for parent etc handling!
2620 */
2621 write_lock_irq(&tasklist_lock);
2622
2623 /* CLONE_PARENT re-uses the old parent */
2624 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2625 p->real_parent = current->real_parent;
2626 p->parent_exec_id = current->parent_exec_id;
2627 if (clone_flags & CLONE_THREAD)
2628 p->exit_signal = -1;
2629 else
2630 p->exit_signal = current->group_leader->exit_signal;
2631 } else {
2632 p->real_parent = current;
2633 p->parent_exec_id = current->self_exec_id;
2634 p->exit_signal = args->exit_signal;
2635 }
2636
2637 klp_copy_process(p);
2638
2639 sched_core_fork(p);
2640
2641 spin_lock(¤t->sighand->siglock);
2642
2643 rv_task_fork(p);
2644
2645 rseq_fork(p, clone_flags);
2646
2647 /* Don't start children in a dying pid namespace */
2648 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2649 retval = -ENOMEM;
2650 goto bad_fork_cancel_cgroup;
2651 }
2652
2653 /* Let kill terminate clone/fork in the middle */
2654 if (fatal_signal_pending(current)) {
2655 retval = -EINTR;
2656 goto bad_fork_cancel_cgroup;
2657 }
2658
2659 /* No more failure paths after this point. */
2660
2661 /*
2662 * Copy seccomp details explicitly here, in case they were changed
2663 * before holding sighand lock.
2664 */
2665 copy_seccomp(p);
2666
2667 init_task_pid_links(p);
2668 if (likely(p->pid)) {
2669 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2670
2671 init_task_pid(p, PIDTYPE_PID, pid);
2672 if (thread_group_leader(p)) {
2673 init_task_pid(p, PIDTYPE_TGID, pid);
2674 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2675 init_task_pid(p, PIDTYPE_SID, task_session(current));
2676
2677 if (is_child_reaper(pid)) {
2678 ns_of_pid(pid)->child_reaper = p;
2679 p->signal->flags |= SIGNAL_UNKILLABLE;
2680 }
2681 p->signal->shared_pending.signal = delayed.signal;
2682 p->signal->tty = tty_kref_get(current->signal->tty);
2683 /*
2684 * Inherit has_child_subreaper flag under the same
2685 * tasklist_lock with adding child to the process tree
2686 * for propagate_has_child_subreaper optimization.
2687 */
2688 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2689 p->real_parent->signal->is_child_subreaper;
2690 list_add_tail(&p->sibling, &p->real_parent->children);
2691 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2692 attach_pid(p, PIDTYPE_TGID);
2693 attach_pid(p, PIDTYPE_PGID);
2694 attach_pid(p, PIDTYPE_SID);
2695 __this_cpu_inc(process_counts);
2696 } else {
2697 current->signal->nr_threads++;
2698 current->signal->quick_threads++;
2699 atomic_inc(¤t->signal->live);
2700 refcount_inc(¤t->signal->sigcnt);
2701 task_join_group_stop(p);
2702 list_add_tail_rcu(&p->thread_node,
2703 &p->signal->thread_head);
2704 }
2705 attach_pid(p, PIDTYPE_PID);
2706 nr_threads++;
2707 }
2708 total_forks++;
2709 hlist_del_init(&delayed.node);
2710 spin_unlock(¤t->sighand->siglock);
2711 syscall_tracepoint_update(p);
2712 write_unlock_irq(&tasklist_lock);
2713
2714 if (pidfile)
2715 fd_install(pidfd, pidfile);
2716
2717 proc_fork_connector(p);
2718 sched_post_fork(p);
2719 cgroup_post_fork(p, args);
2720 perf_event_fork(p);
2721
2722 trace_task_newtask(p, clone_flags);
2723 uprobe_copy_process(p, clone_flags);
2724 user_events_fork(p, clone_flags);
2725
2726 copy_oom_score_adj(clone_flags, p);
2727
2728 return p;
2729
2730bad_fork_cancel_cgroup:
2731 sched_core_free(p);
2732 spin_unlock(¤t->sighand->siglock);
2733 write_unlock_irq(&tasklist_lock);
2734 cgroup_cancel_fork(p, args);
2735bad_fork_put_pidfd:
2736 if (clone_flags & CLONE_PIDFD) {
2737 fput(pidfile);
2738 put_unused_fd(pidfd);
2739 }
2740bad_fork_free_pid:
2741 if (pid != &init_struct_pid)
2742 free_pid(pid);
2743bad_fork_cleanup_thread:
2744 exit_thread(p);
2745bad_fork_cleanup_io:
2746 if (p->io_context)
2747 exit_io_context(p);
2748bad_fork_cleanup_namespaces:
2749 exit_task_namespaces(p);
2750bad_fork_cleanup_mm:
2751 if (p->mm) {
2752 mm_clear_owner(p->mm, p);
2753 mmput(p->mm);
2754 }
2755bad_fork_cleanup_signal:
2756 if (!(clone_flags & CLONE_THREAD))
2757 free_signal_struct(p->signal);
2758bad_fork_cleanup_sighand:
2759 __cleanup_sighand(p->sighand);
2760bad_fork_cleanup_fs:
2761 exit_fs(p); /* blocking */
2762bad_fork_cleanup_files:
2763 exit_files(p); /* blocking */
2764bad_fork_cleanup_semundo:
2765 exit_sem(p);
2766bad_fork_cleanup_security:
2767 security_task_free(p);
2768bad_fork_cleanup_audit:
2769 audit_free(p);
2770bad_fork_cleanup_perf:
2771 perf_event_free_task(p);
2772bad_fork_cleanup_policy:
2773 lockdep_free_task(p);
2774#ifdef CONFIG_NUMA
2775 mpol_put(p->mempolicy);
2776#endif
2777bad_fork_cleanup_delayacct:
2778 delayacct_tsk_free(p);
2779bad_fork_cleanup_count:
2780 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2781 exit_creds(p);
2782bad_fork_free:
2783 WRITE_ONCE(p->__state, TASK_DEAD);
2784 exit_task_stack_account(p);
2785 put_task_stack(p);
2786 delayed_free_task(p);
2787fork_out:
2788 spin_lock_irq(¤t->sighand->siglock);
2789 hlist_del_init(&delayed.node);
2790 spin_unlock_irq(¤t->sighand->siglock);
2791 return ERR_PTR(retval);
2792}
2793
2794static inline void init_idle_pids(struct task_struct *idle)
2795{
2796 enum pid_type type;
2797
2798 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2799 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2800 init_task_pid(idle, type, &init_struct_pid);
2801 }
2802}
2803
2804static int idle_dummy(void *dummy)
2805{
2806 /* This function is never called */
2807 return 0;
2808}
2809
2810struct task_struct * __init fork_idle(int cpu)
2811{
2812 struct task_struct *task;
2813 struct kernel_clone_args args = {
2814 .flags = CLONE_VM,
2815 .fn = &idle_dummy,
2816 .fn_arg = NULL,
2817 .kthread = 1,
2818 .idle = 1,
2819 };
2820
2821 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2822 if (!IS_ERR(task)) {
2823 init_idle_pids(task);
2824 init_idle(task, cpu);
2825 }
2826
2827 return task;
2828}
2829
2830/*
2831 * This is like kernel_clone(), but shaved down and tailored to just
2832 * creating io_uring workers. It returns a created task, or an error pointer.
2833 * The returned task is inactive, and the caller must fire it up through
2834 * wake_up_new_task(p). All signals are blocked in the created task.
2835 */
2836struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2837{
2838 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2839 CLONE_IO;
2840 struct kernel_clone_args args = {
2841 .flags = ((lower_32_bits(flags) | CLONE_VM |
2842 CLONE_UNTRACED) & ~CSIGNAL),
2843 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2844 .fn = fn,
2845 .fn_arg = arg,
2846 .io_thread = 1,
2847 .user_worker = 1,
2848 };
2849
2850 return copy_process(NULL, 0, node, &args);
2851}
2852
2853/*
2854 * Ok, this is the main fork-routine.
2855 *
2856 * It copies the process, and if successful kick-starts
2857 * it and waits for it to finish using the VM if required.
2858 *
2859 * args->exit_signal is expected to be checked for sanity by the caller.
2860 */
2861pid_t kernel_clone(struct kernel_clone_args *args)
2862{
2863 u64 clone_flags = args->flags;
2864 struct completion vfork;
2865 struct pid *pid;
2866 struct task_struct *p;
2867 int trace = 0;
2868 pid_t nr;
2869
2870 /*
2871 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2872 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2873 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2874 * field in struct clone_args and it still doesn't make sense to have
2875 * them both point at the same memory location. Performing this check
2876 * here has the advantage that we don't need to have a separate helper
2877 * to check for legacy clone().
2878 */
2879 if ((args->flags & CLONE_PIDFD) &&
2880 (args->flags & CLONE_PARENT_SETTID) &&
2881 (args->pidfd == args->parent_tid))
2882 return -EINVAL;
2883
2884 /*
2885 * Determine whether and which event to report to ptracer. When
2886 * called from kernel_thread or CLONE_UNTRACED is explicitly
2887 * requested, no event is reported; otherwise, report if the event
2888 * for the type of forking is enabled.
2889 */
2890 if (!(clone_flags & CLONE_UNTRACED)) {
2891 if (clone_flags & CLONE_VFORK)
2892 trace = PTRACE_EVENT_VFORK;
2893 else if (args->exit_signal != SIGCHLD)
2894 trace = PTRACE_EVENT_CLONE;
2895 else
2896 trace = PTRACE_EVENT_FORK;
2897
2898 if (likely(!ptrace_event_enabled(current, trace)))
2899 trace = 0;
2900 }
2901
2902 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2903 add_latent_entropy();
2904
2905 if (IS_ERR(p))
2906 return PTR_ERR(p);
2907
2908 /*
2909 * Do this prior waking up the new thread - the thread pointer
2910 * might get invalid after that point, if the thread exits quickly.
2911 */
2912 trace_sched_process_fork(current, p);
2913
2914 pid = get_task_pid(p, PIDTYPE_PID);
2915 nr = pid_vnr(pid);
2916
2917 if (clone_flags & CLONE_PARENT_SETTID)
2918 put_user(nr, args->parent_tid);
2919
2920 if (clone_flags & CLONE_VFORK) {
2921 p->vfork_done = &vfork;
2922 init_completion(&vfork);
2923 get_task_struct(p);
2924 }
2925
2926 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2927 /* lock the task to synchronize with memcg migration */
2928 task_lock(p);
2929 lru_gen_add_mm(p->mm);
2930 task_unlock(p);
2931 }
2932
2933 wake_up_new_task(p);
2934
2935 /* forking complete and child started to run, tell ptracer */
2936 if (unlikely(trace))
2937 ptrace_event_pid(trace, pid);
2938
2939 if (clone_flags & CLONE_VFORK) {
2940 if (!wait_for_vfork_done(p, &vfork))
2941 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2942 }
2943
2944 put_pid(pid);
2945 return nr;
2946}
2947
2948/*
2949 * Create a kernel thread.
2950 */
2951pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2952 unsigned long flags)
2953{
2954 struct kernel_clone_args args = {
2955 .flags = ((lower_32_bits(flags) | CLONE_VM |
2956 CLONE_UNTRACED) & ~CSIGNAL),
2957 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2958 .fn = fn,
2959 .fn_arg = arg,
2960 .name = name,
2961 .kthread = 1,
2962 };
2963
2964 return kernel_clone(&args);
2965}
2966
2967/*
2968 * Create a user mode thread.
2969 */
2970pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2971{
2972 struct kernel_clone_args args = {
2973 .flags = ((lower_32_bits(flags) | CLONE_VM |
2974 CLONE_UNTRACED) & ~CSIGNAL),
2975 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2976 .fn = fn,
2977 .fn_arg = arg,
2978 };
2979
2980 return kernel_clone(&args);
2981}
2982
2983#ifdef __ARCH_WANT_SYS_FORK
2984SYSCALL_DEFINE0(fork)
2985{
2986#ifdef CONFIG_MMU
2987 struct kernel_clone_args args = {
2988 .exit_signal = SIGCHLD,
2989 };
2990
2991 return kernel_clone(&args);
2992#else
2993 /* can not support in nommu mode */
2994 return -EINVAL;
2995#endif
2996}
2997#endif
2998
2999#ifdef __ARCH_WANT_SYS_VFORK
3000SYSCALL_DEFINE0(vfork)
3001{
3002 struct kernel_clone_args args = {
3003 .flags = CLONE_VFORK | CLONE_VM,
3004 .exit_signal = SIGCHLD,
3005 };
3006
3007 return kernel_clone(&args);
3008}
3009#endif
3010
3011#ifdef __ARCH_WANT_SYS_CLONE
3012#ifdef CONFIG_CLONE_BACKWARDS
3013SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3014 int __user *, parent_tidptr,
3015 unsigned long, tls,
3016 int __user *, child_tidptr)
3017#elif defined(CONFIG_CLONE_BACKWARDS2)
3018SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3019 int __user *, parent_tidptr,
3020 int __user *, child_tidptr,
3021 unsigned long, tls)
3022#elif defined(CONFIG_CLONE_BACKWARDS3)
3023SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3024 int, stack_size,
3025 int __user *, parent_tidptr,
3026 int __user *, child_tidptr,
3027 unsigned long, tls)
3028#else
3029SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3030 int __user *, parent_tidptr,
3031 int __user *, child_tidptr,
3032 unsigned long, tls)
3033#endif
3034{
3035 struct kernel_clone_args args = {
3036 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3037 .pidfd = parent_tidptr,
3038 .child_tid = child_tidptr,
3039 .parent_tid = parent_tidptr,
3040 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3041 .stack = newsp,
3042 .tls = tls,
3043 };
3044
3045 return kernel_clone(&args);
3046}
3047#endif
3048
3049#ifdef __ARCH_WANT_SYS_CLONE3
3050
3051noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3052 struct clone_args __user *uargs,
3053 size_t usize)
3054{
3055 int err;
3056 struct clone_args args;
3057 pid_t *kset_tid = kargs->set_tid;
3058
3059 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3060 CLONE_ARGS_SIZE_VER0);
3061 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3062 CLONE_ARGS_SIZE_VER1);
3063 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3064 CLONE_ARGS_SIZE_VER2);
3065 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3066
3067 if (unlikely(usize > PAGE_SIZE))
3068 return -E2BIG;
3069 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3070 return -EINVAL;
3071
3072 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3073 if (err)
3074 return err;
3075
3076 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3077 return -EINVAL;
3078
3079 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3080 return -EINVAL;
3081
3082 if (unlikely(args.set_tid && args.set_tid_size == 0))
3083 return -EINVAL;
3084
3085 /*
3086 * Verify that higher 32bits of exit_signal are unset and that
3087 * it is a valid signal
3088 */
3089 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3090 !valid_signal(args.exit_signal)))
3091 return -EINVAL;
3092
3093 if ((args.flags & CLONE_INTO_CGROUP) &&
3094 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3095 return -EINVAL;
3096
3097 *kargs = (struct kernel_clone_args){
3098 .flags = args.flags,
3099 .pidfd = u64_to_user_ptr(args.pidfd),
3100 .child_tid = u64_to_user_ptr(args.child_tid),
3101 .parent_tid = u64_to_user_ptr(args.parent_tid),
3102 .exit_signal = args.exit_signal,
3103 .stack = args.stack,
3104 .stack_size = args.stack_size,
3105 .tls = args.tls,
3106 .set_tid_size = args.set_tid_size,
3107 .cgroup = args.cgroup,
3108 };
3109
3110 if (args.set_tid &&
3111 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3112 (kargs->set_tid_size * sizeof(pid_t))))
3113 return -EFAULT;
3114
3115 kargs->set_tid = kset_tid;
3116
3117 return 0;
3118}
3119
3120/**
3121 * clone3_stack_valid - check and prepare stack
3122 * @kargs: kernel clone args
3123 *
3124 * Verify that the stack arguments userspace gave us are sane.
3125 * In addition, set the stack direction for userspace since it's easy for us to
3126 * determine.
3127 */
3128static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3129{
3130 if (kargs->stack == 0) {
3131 if (kargs->stack_size > 0)
3132 return false;
3133 } else {
3134 if (kargs->stack_size == 0)
3135 return false;
3136
3137 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3138 return false;
3139
3140#if !defined(CONFIG_STACK_GROWSUP)
3141 kargs->stack += kargs->stack_size;
3142#endif
3143 }
3144
3145 return true;
3146}
3147
3148static bool clone3_args_valid(struct kernel_clone_args *kargs)
3149{
3150 /* Verify that no unknown flags are passed along. */
3151 if (kargs->flags &
3152 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3153 return false;
3154
3155 /*
3156 * - make the CLONE_DETACHED bit reusable for clone3
3157 * - make the CSIGNAL bits reusable for clone3
3158 */
3159 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3160 return false;
3161
3162 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3163 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3164 return false;
3165
3166 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3167 kargs->exit_signal)
3168 return false;
3169
3170 if (!clone3_stack_valid(kargs))
3171 return false;
3172
3173 return true;
3174}
3175
3176/**
3177 * sys_clone3 - create a new process with specific properties
3178 * @uargs: argument structure
3179 * @size: size of @uargs
3180 *
3181 * clone3() is the extensible successor to clone()/clone2().
3182 * It takes a struct as argument that is versioned by its size.
3183 *
3184 * Return: On success, a positive PID for the child process.
3185 * On error, a negative errno number.
3186 */
3187SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3188{
3189 int err;
3190
3191 struct kernel_clone_args kargs;
3192 pid_t set_tid[MAX_PID_NS_LEVEL];
3193
3194 kargs.set_tid = set_tid;
3195
3196 err = copy_clone_args_from_user(&kargs, uargs, size);
3197 if (err)
3198 return err;
3199
3200 if (!clone3_args_valid(&kargs))
3201 return -EINVAL;
3202
3203 return kernel_clone(&kargs);
3204}
3205#endif
3206
3207void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3208{
3209 struct task_struct *leader, *parent, *child;
3210 int res;
3211
3212 read_lock(&tasklist_lock);
3213 leader = top = top->group_leader;
3214down:
3215 for_each_thread(leader, parent) {
3216 list_for_each_entry(child, &parent->children, sibling) {
3217 res = visitor(child, data);
3218 if (res) {
3219 if (res < 0)
3220 goto out;
3221 leader = child;
3222 goto down;
3223 }
3224up:
3225 ;
3226 }
3227 }
3228
3229 if (leader != top) {
3230 child = leader;
3231 parent = child->real_parent;
3232 leader = parent->group_leader;
3233 goto up;
3234 }
3235out:
3236 read_unlock(&tasklist_lock);
3237}
3238
3239#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3240#define ARCH_MIN_MMSTRUCT_ALIGN 0
3241#endif
3242
3243static void sighand_ctor(void *data)
3244{
3245 struct sighand_struct *sighand = data;
3246
3247 spin_lock_init(&sighand->siglock);
3248 init_waitqueue_head(&sighand->signalfd_wqh);
3249}
3250
3251void __init mm_cache_init(void)
3252{
3253 unsigned int mm_size;
3254
3255 /*
3256 * The mm_cpumask is located at the end of mm_struct, and is
3257 * dynamically sized based on the maximum CPU number this system
3258 * can have, taking hotplug into account (nr_cpu_ids).
3259 */
3260 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3261
3262 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3263 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3264 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3265 offsetof(struct mm_struct, saved_auxv),
3266 sizeof_field(struct mm_struct, saved_auxv),
3267 NULL);
3268}
3269
3270void __init proc_caches_init(void)
3271{
3272 sighand_cachep = kmem_cache_create("sighand_cache",
3273 sizeof(struct sighand_struct), 0,
3274 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3275 SLAB_ACCOUNT, sighand_ctor);
3276 signal_cachep = kmem_cache_create("signal_cache",
3277 sizeof(struct signal_struct), 0,
3278 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3279 NULL);
3280 files_cachep = kmem_cache_create("files_cache",
3281 sizeof(struct files_struct), 0,
3282 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3283 NULL);
3284 fs_cachep = kmem_cache_create("fs_cache",
3285 sizeof(struct fs_struct), 0,
3286 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3287 NULL);
3288
3289 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3290#ifdef CONFIG_PER_VMA_LOCK
3291 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3292#endif
3293 mmap_init();
3294 nsproxy_cache_init();
3295}
3296
3297/*
3298 * Check constraints on flags passed to the unshare system call.
3299 */
3300static int check_unshare_flags(unsigned long unshare_flags)
3301{
3302 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3303 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3304 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3305 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3306 CLONE_NEWTIME))
3307 return -EINVAL;
3308 /*
3309 * Not implemented, but pretend it works if there is nothing
3310 * to unshare. Note that unsharing the address space or the
3311 * signal handlers also need to unshare the signal queues (aka
3312 * CLONE_THREAD).
3313 */
3314 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3315 if (!thread_group_empty(current))
3316 return -EINVAL;
3317 }
3318 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3319 if (refcount_read(¤t->sighand->count) > 1)
3320 return -EINVAL;
3321 }
3322 if (unshare_flags & CLONE_VM) {
3323 if (!current_is_single_threaded())
3324 return -EINVAL;
3325 }
3326
3327 return 0;
3328}
3329
3330/*
3331 * Unshare the filesystem structure if it is being shared
3332 */
3333static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3334{
3335 struct fs_struct *fs = current->fs;
3336
3337 if (!(unshare_flags & CLONE_FS) || !fs)
3338 return 0;
3339
3340 /* don't need lock here; in the worst case we'll do useless copy */
3341 if (fs->users == 1)
3342 return 0;
3343
3344 *new_fsp = copy_fs_struct(fs);
3345 if (!*new_fsp)
3346 return -ENOMEM;
3347
3348 return 0;
3349}
3350
3351/*
3352 * Unshare file descriptor table if it is being shared
3353 */
3354int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3355 struct files_struct **new_fdp)
3356{
3357 struct files_struct *fd = current->files;
3358 int error = 0;
3359
3360 if ((unshare_flags & CLONE_FILES) &&
3361 (fd && atomic_read(&fd->count) > 1)) {
3362 *new_fdp = dup_fd(fd, max_fds, &error);
3363 if (!*new_fdp)
3364 return error;
3365 }
3366
3367 return 0;
3368}
3369
3370/*
3371 * unshare allows a process to 'unshare' part of the process
3372 * context which was originally shared using clone. copy_*
3373 * functions used by kernel_clone() cannot be used here directly
3374 * because they modify an inactive task_struct that is being
3375 * constructed. Here we are modifying the current, active,
3376 * task_struct.
3377 */
3378int ksys_unshare(unsigned long unshare_flags)
3379{
3380 struct fs_struct *fs, *new_fs = NULL;
3381 struct files_struct *new_fd = NULL;
3382 struct cred *new_cred = NULL;
3383 struct nsproxy *new_nsproxy = NULL;
3384 int do_sysvsem = 0;
3385 int err;
3386
3387 /*
3388 * If unsharing a user namespace must also unshare the thread group
3389 * and unshare the filesystem root and working directories.
3390 */
3391 if (unshare_flags & CLONE_NEWUSER)
3392 unshare_flags |= CLONE_THREAD | CLONE_FS;
3393 /*
3394 * If unsharing vm, must also unshare signal handlers.
3395 */
3396 if (unshare_flags & CLONE_VM)
3397 unshare_flags |= CLONE_SIGHAND;
3398 /*
3399 * If unsharing a signal handlers, must also unshare the signal queues.
3400 */
3401 if (unshare_flags & CLONE_SIGHAND)
3402 unshare_flags |= CLONE_THREAD;
3403 /*
3404 * If unsharing namespace, must also unshare filesystem information.
3405 */
3406 if (unshare_flags & CLONE_NEWNS)
3407 unshare_flags |= CLONE_FS;
3408
3409 err = check_unshare_flags(unshare_flags);
3410 if (err)
3411 goto bad_unshare_out;
3412 /*
3413 * CLONE_NEWIPC must also detach from the undolist: after switching
3414 * to a new ipc namespace, the semaphore arrays from the old
3415 * namespace are unreachable.
3416 */
3417 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3418 do_sysvsem = 1;
3419 err = unshare_fs(unshare_flags, &new_fs);
3420 if (err)
3421 goto bad_unshare_out;
3422 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3423 if (err)
3424 goto bad_unshare_cleanup_fs;
3425 err = unshare_userns(unshare_flags, &new_cred);
3426 if (err)
3427 goto bad_unshare_cleanup_fd;
3428 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3429 new_cred, new_fs);
3430 if (err)
3431 goto bad_unshare_cleanup_cred;
3432
3433 if (new_cred) {
3434 err = set_cred_ucounts(new_cred);
3435 if (err)
3436 goto bad_unshare_cleanup_cred;
3437 }
3438
3439 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3440 if (do_sysvsem) {
3441 /*
3442 * CLONE_SYSVSEM is equivalent to sys_exit().
3443 */
3444 exit_sem(current);
3445 }
3446 if (unshare_flags & CLONE_NEWIPC) {
3447 /* Orphan segments in old ns (see sem above). */
3448 exit_shm(current);
3449 shm_init_task(current);
3450 }
3451
3452 if (new_nsproxy)
3453 switch_task_namespaces(current, new_nsproxy);
3454
3455 task_lock(current);
3456
3457 if (new_fs) {
3458 fs = current->fs;
3459 spin_lock(&fs->lock);
3460 current->fs = new_fs;
3461 if (--fs->users)
3462 new_fs = NULL;
3463 else
3464 new_fs = fs;
3465 spin_unlock(&fs->lock);
3466 }
3467
3468 if (new_fd)
3469 swap(current->files, new_fd);
3470
3471 task_unlock(current);
3472
3473 if (new_cred) {
3474 /* Install the new user namespace */
3475 commit_creds(new_cred);
3476 new_cred = NULL;
3477 }
3478 }
3479
3480 perf_event_namespaces(current);
3481
3482bad_unshare_cleanup_cred:
3483 if (new_cred)
3484 put_cred(new_cred);
3485bad_unshare_cleanup_fd:
3486 if (new_fd)
3487 put_files_struct(new_fd);
3488
3489bad_unshare_cleanup_fs:
3490 if (new_fs)
3491 free_fs_struct(new_fs);
3492
3493bad_unshare_out:
3494 return err;
3495}
3496
3497SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3498{
3499 return ksys_unshare(unshare_flags);
3500}
3501
3502/*
3503 * Helper to unshare the files of the current task.
3504 * We don't want to expose copy_files internals to
3505 * the exec layer of the kernel.
3506 */
3507
3508int unshare_files(void)
3509{
3510 struct task_struct *task = current;
3511 struct files_struct *old, *copy = NULL;
3512 int error;
3513
3514 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3515 if (error || !copy)
3516 return error;
3517
3518 old = task->files;
3519 task_lock(task);
3520 task->files = copy;
3521 task_unlock(task);
3522 put_files_struct(old);
3523 return 0;
3524}
3525
3526int sysctl_max_threads(struct ctl_table *table, int write,
3527 void *buffer, size_t *lenp, loff_t *ppos)
3528{
3529 struct ctl_table t;
3530 int ret;
3531 int threads = max_threads;
3532 int min = 1;
3533 int max = MAX_THREADS;
3534
3535 t = *table;
3536 t.data = &threads;
3537 t.extra1 = &min;
3538 t.extra2 = &max;
3539
3540 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3541 if (ret || !write)
3542 return ret;
3543
3544 max_threads = threads;
3545
3546 return 0;
3547}
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}