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/user.h>
20#include <linux/sched/numa_balancing.h>
21#include <linux/sched/stat.h>
22#include <linux/sched/task.h>
23#include <linux/sched/task_stack.h>
24#include <linux/sched/cputime.h>
25#include <linux/sched/ext.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/memblock.h>
48#include <linux/nsproxy.h>
49#include <linux/capability.h>
50#include <linux/cpu.h>
51#include <linux/cgroup.h>
52#include <linux/security.h>
53#include <linux/hugetlb.h>
54#include <linux/seccomp.h>
55#include <linux/swap.h>
56#include <linux/syscalls.h>
57#include <linux/syscall_user_dispatch.h>
58#include <linux/jiffies.h>
59#include <linux/futex.h>
60#include <linux/compat.h>
61#include <linux/kthread.h>
62#include <linux/task_io_accounting_ops.h>
63#include <linux/rcupdate.h>
64#include <linux/ptrace.h>
65#include <linux/mount.h>
66#include <linux/audit.h>
67#include <linux/memcontrol.h>
68#include <linux/ftrace.h>
69#include <linux/proc_fs.h>
70#include <linux/profile.h>
71#include <linux/rmap.h>
72#include <linux/ksm.h>
73#include <linux/acct.h>
74#include <linux/userfaultfd_k.h>
75#include <linux/tsacct_kern.h>
76#include <linux/cn_proc.h>
77#include <linux/freezer.h>
78#include <linux/delayacct.h>
79#include <linux/taskstats_kern.h>
80#include <linux/tty.h>
81#include <linux/fs_struct.h>
82#include <linux/magic.h>
83#include <linux/perf_event.h>
84#include <linux/posix-timers.h>
85#include <linux/user-return-notifier.h>
86#include <linux/oom.h>
87#include <linux/khugepaged.h>
88#include <linux/signalfd.h>
89#include <linux/uprobes.h>
90#include <linux/aio.h>
91#include <linux/compiler.h>
92#include <linux/sysctl.h>
93#include <linux/kcov.h>
94#include <linux/livepatch.h>
95#include <linux/thread_info.h>
96#include <linux/stackleak.h>
97#include <linux/kasan.h>
98#include <linux/scs.h>
99#include <linux/io_uring.h>
100#include <linux/bpf.h>
101#include <linux/stackprotector.h>
102#include <linux/user_events.h>
103#include <linux/iommu.h>
104#include <linux/rseq.h>
105#include <uapi/linux/pidfd.h>
106#include <linux/pidfs.h>
107#include <linux/tick.h>
108
109#include <asm/pgalloc.h>
110#include <linux/uaccess.h>
111#include <asm/mmu_context.h>
112#include <asm/cacheflush.h>
113#include <asm/tlbflush.h>
114
115#include <trace/events/sched.h>
116
117#define CREATE_TRACE_POINTS
118#include <trace/events/task.h>
119
120#include <kunit/visibility.h>
121
122/*
123 * Minimum number of threads to boot the kernel
124 */
125#define MIN_THREADS 20
126
127/*
128 * Maximum number of threads
129 */
130#define MAX_THREADS FUTEX_TID_MASK
131
132/*
133 * Protected counters by write_lock_irq(&tasklist_lock)
134 */
135unsigned long total_forks; /* Handle normal Linux uptimes. */
136int nr_threads; /* The idle threads do not count.. */
137
138static int max_threads; /* tunable limit on nr_threads */
139
140#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
141
142static const char * const resident_page_types[] = {
143 NAMED_ARRAY_INDEX(MM_FILEPAGES),
144 NAMED_ARRAY_INDEX(MM_ANONPAGES),
145 NAMED_ARRAY_INDEX(MM_SWAPENTS),
146 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
147};
148
149DEFINE_PER_CPU(unsigned long, process_counts) = 0;
150
151__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
152
153#ifdef CONFIG_PROVE_RCU
154int lockdep_tasklist_lock_is_held(void)
155{
156 return lockdep_is_held(&tasklist_lock);
157}
158EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
159#endif /* #ifdef CONFIG_PROVE_RCU */
160
161int nr_processes(void)
162{
163 int cpu;
164 int total = 0;
165
166 for_each_possible_cpu(cpu)
167 total += per_cpu(process_counts, cpu);
168
169 return total;
170}
171
172void __weak arch_release_task_struct(struct task_struct *tsk)
173{
174}
175
176static struct kmem_cache *task_struct_cachep;
177
178static inline struct task_struct *alloc_task_struct_node(int node)
179{
180 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
181}
182
183static inline void free_task_struct(struct task_struct *tsk)
184{
185 kmem_cache_free(task_struct_cachep, tsk);
186}
187
188/*
189 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
190 * kmemcache based allocator.
191 */
192# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
193
194# ifdef CONFIG_VMAP_STACK
195/*
196 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
197 * flush. Try to minimize the number of calls by caching stacks.
198 */
199#define NR_CACHED_STACKS 2
200static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
201
202struct vm_stack {
203 struct rcu_head rcu;
204 struct vm_struct *stack_vm_area;
205};
206
207static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
208{
209 unsigned int i;
210
211 for (i = 0; i < NR_CACHED_STACKS; i++) {
212 struct vm_struct *tmp = NULL;
213
214 if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm))
215 return true;
216 }
217 return false;
218}
219
220static void thread_stack_free_rcu(struct rcu_head *rh)
221{
222 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
223
224 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
225 return;
226
227 vfree(vm_stack);
228}
229
230static void thread_stack_delayed_free(struct task_struct *tsk)
231{
232 struct vm_stack *vm_stack = tsk->stack;
233
234 vm_stack->stack_vm_area = tsk->stack_vm_area;
235 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
236}
237
238static int free_vm_stack_cache(unsigned int cpu)
239{
240 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
241 int i;
242
243 for (i = 0; i < NR_CACHED_STACKS; i++) {
244 struct vm_struct *vm_stack = cached_vm_stacks[i];
245
246 if (!vm_stack)
247 continue;
248
249 vfree(vm_stack->addr);
250 cached_vm_stacks[i] = NULL;
251 }
252
253 return 0;
254}
255
256static int memcg_charge_kernel_stack(struct vm_struct *vm)
257{
258 int i;
259 int ret;
260 int nr_charged = 0;
261
262 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
263
264 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
265 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
266 if (ret)
267 goto err;
268 nr_charged++;
269 }
270 return 0;
271err:
272 for (i = 0; i < nr_charged; i++)
273 memcg_kmem_uncharge_page(vm->pages[i], 0);
274 return ret;
275}
276
277static int alloc_thread_stack_node(struct task_struct *tsk, int node)
278{
279 struct vm_struct *vm;
280 void *stack;
281 int i;
282
283 for (i = 0; i < NR_CACHED_STACKS; i++) {
284 struct vm_struct *s;
285
286 s = this_cpu_xchg(cached_stacks[i], NULL);
287
288 if (!s)
289 continue;
290
291 /* Reset stack metadata. */
292 kasan_unpoison_range(s->addr, THREAD_SIZE);
293
294 stack = kasan_reset_tag(s->addr);
295
296 /* Clear stale pointers from reused stack. */
297 memset(stack, 0, THREAD_SIZE);
298
299 if (memcg_charge_kernel_stack(s)) {
300 vfree(s->addr);
301 return -ENOMEM;
302 }
303
304 tsk->stack_vm_area = s;
305 tsk->stack = stack;
306 return 0;
307 }
308
309 /*
310 * Allocated stacks are cached and later reused by new threads,
311 * so memcg accounting is performed manually on assigning/releasing
312 * stacks to tasks. Drop __GFP_ACCOUNT.
313 */
314 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
315 VMALLOC_START, VMALLOC_END,
316 THREADINFO_GFP & ~__GFP_ACCOUNT,
317 PAGE_KERNEL,
318 0, node, __builtin_return_address(0));
319 if (!stack)
320 return -ENOMEM;
321
322 vm = find_vm_area(stack);
323 if (memcg_charge_kernel_stack(vm)) {
324 vfree(stack);
325 return -ENOMEM;
326 }
327 /*
328 * We can't call find_vm_area() in interrupt context, and
329 * free_thread_stack() can be called in interrupt context,
330 * so cache the vm_struct.
331 */
332 tsk->stack_vm_area = vm;
333 stack = kasan_reset_tag(stack);
334 tsk->stack = stack;
335 return 0;
336}
337
338static void free_thread_stack(struct task_struct *tsk)
339{
340 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
341 thread_stack_delayed_free(tsk);
342
343 tsk->stack = NULL;
344 tsk->stack_vm_area = NULL;
345}
346
347# else /* !CONFIG_VMAP_STACK */
348
349static void thread_stack_free_rcu(struct rcu_head *rh)
350{
351 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
352}
353
354static void thread_stack_delayed_free(struct task_struct *tsk)
355{
356 struct rcu_head *rh = tsk->stack;
357
358 call_rcu(rh, thread_stack_free_rcu);
359}
360
361static int alloc_thread_stack_node(struct task_struct *tsk, int node)
362{
363 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
364 THREAD_SIZE_ORDER);
365
366 if (likely(page)) {
367 tsk->stack = kasan_reset_tag(page_address(page));
368 return 0;
369 }
370 return -ENOMEM;
371}
372
373static void free_thread_stack(struct task_struct *tsk)
374{
375 thread_stack_delayed_free(tsk);
376 tsk->stack = NULL;
377}
378
379# endif /* CONFIG_VMAP_STACK */
380# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
381
382static struct kmem_cache *thread_stack_cache;
383
384static void thread_stack_free_rcu(struct rcu_head *rh)
385{
386 kmem_cache_free(thread_stack_cache, rh);
387}
388
389static void thread_stack_delayed_free(struct task_struct *tsk)
390{
391 struct rcu_head *rh = tsk->stack;
392
393 call_rcu(rh, thread_stack_free_rcu);
394}
395
396static int alloc_thread_stack_node(struct task_struct *tsk, int node)
397{
398 unsigned long *stack;
399 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
400 stack = kasan_reset_tag(stack);
401 tsk->stack = stack;
402 return stack ? 0 : -ENOMEM;
403}
404
405static void free_thread_stack(struct task_struct *tsk)
406{
407 thread_stack_delayed_free(tsk);
408 tsk->stack = NULL;
409}
410
411void thread_stack_cache_init(void)
412{
413 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
414 THREAD_SIZE, THREAD_SIZE, 0, 0,
415 THREAD_SIZE, NULL);
416 BUG_ON(thread_stack_cache == NULL);
417}
418
419# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
420
421/* SLAB cache for signal_struct structures (tsk->signal) */
422static struct kmem_cache *signal_cachep;
423
424/* SLAB cache for sighand_struct structures (tsk->sighand) */
425struct kmem_cache *sighand_cachep;
426
427/* SLAB cache for files_struct structures (tsk->files) */
428struct kmem_cache *files_cachep;
429
430/* SLAB cache for fs_struct structures (tsk->fs) */
431struct kmem_cache *fs_cachep;
432
433/* SLAB cache for vm_area_struct structures */
434static struct kmem_cache *vm_area_cachep;
435
436/* SLAB cache for mm_struct structures (tsk->mm) */
437static struct kmem_cache *mm_cachep;
438
439#ifdef CONFIG_PER_VMA_LOCK
440
441/* SLAB cache for vm_area_struct.lock */
442static struct kmem_cache *vma_lock_cachep;
443
444static bool vma_lock_alloc(struct vm_area_struct *vma)
445{
446 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
447 if (!vma->vm_lock)
448 return false;
449
450 init_rwsem(&vma->vm_lock->lock);
451 vma->vm_lock_seq = -1;
452
453 return true;
454}
455
456static inline void vma_lock_free(struct vm_area_struct *vma)
457{
458 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
459}
460
461#else /* CONFIG_PER_VMA_LOCK */
462
463static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
464static inline void vma_lock_free(struct vm_area_struct *vma) {}
465
466#endif /* CONFIG_PER_VMA_LOCK */
467
468struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
469{
470 struct vm_area_struct *vma;
471
472 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
473 if (!vma)
474 return NULL;
475
476 vma_init(vma, mm);
477 if (!vma_lock_alloc(vma)) {
478 kmem_cache_free(vm_area_cachep, vma);
479 return NULL;
480 }
481
482 return vma;
483}
484
485struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
486{
487 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
488
489 if (!new)
490 return NULL;
491
492 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
493 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
494 /*
495 * orig->shared.rb may be modified concurrently, but the clone
496 * will be reinitialized.
497 */
498 data_race(memcpy(new, orig, sizeof(*new)));
499 if (!vma_lock_alloc(new)) {
500 kmem_cache_free(vm_area_cachep, new);
501 return NULL;
502 }
503 INIT_LIST_HEAD(&new->anon_vma_chain);
504 vma_numab_state_init(new);
505 dup_anon_vma_name(orig, new);
506
507 return new;
508}
509
510void __vm_area_free(struct vm_area_struct *vma)
511{
512 vma_numab_state_free(vma);
513 free_anon_vma_name(vma);
514 vma_lock_free(vma);
515 kmem_cache_free(vm_area_cachep, vma);
516}
517
518#ifdef CONFIG_PER_VMA_LOCK
519static void vm_area_free_rcu_cb(struct rcu_head *head)
520{
521 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
522 vm_rcu);
523
524 /* The vma should not be locked while being destroyed. */
525 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
526 __vm_area_free(vma);
527}
528#endif
529
530void vm_area_free(struct vm_area_struct *vma)
531{
532#ifdef CONFIG_PER_VMA_LOCK
533 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
534#else
535 __vm_area_free(vma);
536#endif
537}
538
539static void account_kernel_stack(struct task_struct *tsk, int account)
540{
541 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
542 struct vm_struct *vm = task_stack_vm_area(tsk);
543 int i;
544
545 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
546 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
547 account * (PAGE_SIZE / 1024));
548 } else {
549 void *stack = task_stack_page(tsk);
550
551 /* All stack pages are in the same node. */
552 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
553 account * (THREAD_SIZE / 1024));
554 }
555}
556
557void exit_task_stack_account(struct task_struct *tsk)
558{
559 account_kernel_stack(tsk, -1);
560
561 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
562 struct vm_struct *vm;
563 int i;
564
565 vm = task_stack_vm_area(tsk);
566 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
567 memcg_kmem_uncharge_page(vm->pages[i], 0);
568 }
569}
570
571static void release_task_stack(struct task_struct *tsk)
572{
573 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
574 return; /* Better to leak the stack than to free prematurely */
575
576 free_thread_stack(tsk);
577}
578
579#ifdef CONFIG_THREAD_INFO_IN_TASK
580void put_task_stack(struct task_struct *tsk)
581{
582 if (refcount_dec_and_test(&tsk->stack_refcount))
583 release_task_stack(tsk);
584}
585#endif
586
587void free_task(struct task_struct *tsk)
588{
589#ifdef CONFIG_SECCOMP
590 WARN_ON_ONCE(tsk->seccomp.filter);
591#endif
592 release_user_cpus_ptr(tsk);
593 scs_release(tsk);
594
595#ifndef CONFIG_THREAD_INFO_IN_TASK
596 /*
597 * The task is finally done with both the stack and thread_info,
598 * so free both.
599 */
600 release_task_stack(tsk);
601#else
602 /*
603 * If the task had a separate stack allocation, it should be gone
604 * by now.
605 */
606 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
607#endif
608 rt_mutex_debug_task_free(tsk);
609 ftrace_graph_exit_task(tsk);
610 arch_release_task_struct(tsk);
611 if (tsk->flags & PF_KTHREAD)
612 free_kthread_struct(tsk);
613 bpf_task_storage_free(tsk);
614 free_task_struct(tsk);
615}
616EXPORT_SYMBOL(free_task);
617
618static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
619{
620 struct file *exe_file;
621
622 exe_file = get_mm_exe_file(oldmm);
623 RCU_INIT_POINTER(mm->exe_file, exe_file);
624 /*
625 * We depend on the oldmm having properly denied write access to the
626 * exe_file already.
627 */
628 if (exe_file && deny_write_access(exe_file))
629 pr_warn_once("deny_write_access() failed in %s\n", __func__);
630}
631
632#ifdef CONFIG_MMU
633static __latent_entropy int dup_mmap(struct mm_struct *mm,
634 struct mm_struct *oldmm)
635{
636 struct vm_area_struct *mpnt, *tmp;
637 int retval;
638 unsigned long charge = 0;
639 LIST_HEAD(uf);
640 VMA_ITERATOR(vmi, mm, 0);
641
642 if (mmap_write_lock_killable(oldmm))
643 return -EINTR;
644 flush_cache_dup_mm(oldmm);
645 uprobe_dup_mmap(oldmm, mm);
646 /*
647 * Not linked in yet - no deadlock potential:
648 */
649 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
650
651 /* No ordering required: file already has been exposed. */
652 dup_mm_exe_file(mm, oldmm);
653
654 mm->total_vm = oldmm->total_vm;
655 mm->data_vm = oldmm->data_vm;
656 mm->exec_vm = oldmm->exec_vm;
657 mm->stack_vm = oldmm->stack_vm;
658
659 /* Use __mt_dup() to efficiently build an identical maple tree. */
660 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
661 if (unlikely(retval))
662 goto out;
663
664 mt_clear_in_rcu(vmi.mas.tree);
665 for_each_vma(vmi, mpnt) {
666 struct file *file;
667
668 vma_start_write(mpnt);
669 if (mpnt->vm_flags & VM_DONTCOPY) {
670 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
671 mpnt->vm_end, GFP_KERNEL);
672 if (retval)
673 goto loop_out;
674
675 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
676 continue;
677 }
678 charge = 0;
679 /*
680 * Don't duplicate many vmas if we've been oom-killed (for
681 * example)
682 */
683 if (fatal_signal_pending(current)) {
684 retval = -EINTR;
685 goto loop_out;
686 }
687 if (mpnt->vm_flags & VM_ACCOUNT) {
688 unsigned long len = vma_pages(mpnt);
689
690 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
691 goto fail_nomem;
692 charge = len;
693 }
694 tmp = vm_area_dup(mpnt);
695 if (!tmp)
696 goto fail_nomem;
697 retval = vma_dup_policy(mpnt, tmp);
698 if (retval)
699 goto fail_nomem_policy;
700 tmp->vm_mm = mm;
701 retval = dup_userfaultfd(tmp, &uf);
702 if (retval)
703 goto fail_nomem_anon_vma_fork;
704 if (tmp->vm_flags & VM_WIPEONFORK) {
705 /*
706 * VM_WIPEONFORK gets a clean slate in the child.
707 * Don't prepare anon_vma until fault since we don't
708 * copy page for current vma.
709 */
710 tmp->anon_vma = NULL;
711 } else if (anon_vma_fork(tmp, mpnt))
712 goto fail_nomem_anon_vma_fork;
713 vm_flags_clear(tmp, VM_LOCKED_MASK);
714 /*
715 * Copy/update hugetlb private vma information.
716 */
717 if (is_vm_hugetlb_page(tmp))
718 hugetlb_dup_vma_private(tmp);
719
720 /*
721 * Link the vma into the MT. After using __mt_dup(), memory
722 * allocation is not necessary here, so it cannot fail.
723 */
724 vma_iter_bulk_store(&vmi, tmp);
725
726 mm->map_count++;
727
728 if (tmp->vm_ops && tmp->vm_ops->open)
729 tmp->vm_ops->open(tmp);
730
731 file = tmp->vm_file;
732 if (file) {
733 struct address_space *mapping = file->f_mapping;
734
735 get_file(file);
736 i_mmap_lock_write(mapping);
737 if (vma_is_shared_maywrite(tmp))
738 mapping_allow_writable(mapping);
739 flush_dcache_mmap_lock(mapping);
740 /* insert tmp into the share list, just after mpnt */
741 vma_interval_tree_insert_after(tmp, mpnt,
742 &mapping->i_mmap);
743 flush_dcache_mmap_unlock(mapping);
744 i_mmap_unlock_write(mapping);
745 }
746
747 if (!(tmp->vm_flags & VM_WIPEONFORK))
748 retval = copy_page_range(tmp, mpnt);
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 ksm_fork(mm, oldmm);
762 khugepaged_fork(mm, oldmm);
763 } else {
764
765 /*
766 * The entire maple tree has already been duplicated. If the
767 * mmap duplication fails, mark the failure point with
768 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
769 * stop releasing VMAs that have not been duplicated after this
770 * point.
771 */
772 if (mpnt) {
773 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
774 mas_store(&vmi.mas, XA_ZERO_ENTRY);
775 /* Avoid OOM iterating a broken tree */
776 set_bit(MMF_OOM_SKIP, &mm->flags);
777 }
778 /*
779 * The mm_struct is going to exit, but the locks will be dropped
780 * first. Set the mm_struct as unstable is advisable as it is
781 * not fully initialised.
782 */
783 set_bit(MMF_UNSTABLE, &mm->flags);
784 }
785out:
786 mmap_write_unlock(mm);
787 flush_tlb_mm(oldmm);
788 mmap_write_unlock(oldmm);
789 if (!retval)
790 dup_userfaultfd_complete(&uf);
791 else
792 dup_userfaultfd_fail(&uf);
793 return retval;
794
795fail_nomem_anon_vma_fork:
796 mpol_put(vma_policy(tmp));
797fail_nomem_policy:
798 vm_area_free(tmp);
799fail_nomem:
800 retval = -ENOMEM;
801 vm_unacct_memory(charge);
802 goto loop_out;
803}
804
805static inline int mm_alloc_pgd(struct mm_struct *mm)
806{
807 mm->pgd = pgd_alloc(mm);
808 if (unlikely(!mm->pgd))
809 return -ENOMEM;
810 return 0;
811}
812
813static inline void mm_free_pgd(struct mm_struct *mm)
814{
815 pgd_free(mm, mm->pgd);
816}
817#else
818static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
819{
820 mmap_write_lock(oldmm);
821 dup_mm_exe_file(mm, oldmm);
822 mmap_write_unlock(oldmm);
823 return 0;
824}
825#define mm_alloc_pgd(mm) (0)
826#define mm_free_pgd(mm)
827#endif /* CONFIG_MMU */
828
829static void check_mm(struct mm_struct *mm)
830{
831 int i;
832
833 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
834 "Please make sure 'struct resident_page_types[]' is updated as well");
835
836 for (i = 0; i < NR_MM_COUNTERS; i++) {
837 long x = percpu_counter_sum(&mm->rss_stat[i]);
838
839 if (unlikely(x))
840 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
841 mm, resident_page_types[i], x);
842 }
843
844 if (mm_pgtables_bytes(mm))
845 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
846 mm_pgtables_bytes(mm));
847
848#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
849 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
850#endif
851}
852
853#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
854#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
855
856static void do_check_lazy_tlb(void *arg)
857{
858 struct mm_struct *mm = arg;
859
860 WARN_ON_ONCE(current->active_mm == mm);
861}
862
863static void do_shoot_lazy_tlb(void *arg)
864{
865 struct mm_struct *mm = arg;
866
867 if (current->active_mm == mm) {
868 WARN_ON_ONCE(current->mm);
869 current->active_mm = &init_mm;
870 switch_mm(mm, &init_mm, current);
871 }
872}
873
874static void cleanup_lazy_tlbs(struct mm_struct *mm)
875{
876 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
877 /*
878 * In this case, lazy tlb mms are refounted and would not reach
879 * __mmdrop until all CPUs have switched away and mmdrop()ed.
880 */
881 return;
882 }
883
884 /*
885 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
886 * requires lazy mm users to switch to another mm when the refcount
887 * drops to zero, before the mm is freed. This requires IPIs here to
888 * switch kernel threads to init_mm.
889 *
890 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
891 * switch with the final userspace teardown TLB flush which leaves the
892 * mm lazy on this CPU but no others, reducing the need for additional
893 * IPIs here. There are cases where a final IPI is still required here,
894 * such as the final mmdrop being performed on a different CPU than the
895 * one exiting, or kernel threads using the mm when userspace exits.
896 *
897 * IPI overheads have not found to be expensive, but they could be
898 * reduced in a number of possible ways, for example (roughly
899 * increasing order of complexity):
900 * - The last lazy reference created by exit_mm() could instead switch
901 * to init_mm, however it's probable this will run on the same CPU
902 * immediately afterwards, so this may not reduce IPIs much.
903 * - A batch of mms requiring IPIs could be gathered and freed at once.
904 * - CPUs store active_mm where it can be remotely checked without a
905 * lock, to filter out false-positives in the cpumask.
906 * - After mm_users or mm_count reaches zero, switching away from the
907 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
908 * with some batching or delaying of the final IPIs.
909 * - A delayed freeing and RCU-like quiescing sequence based on mm
910 * switching to avoid IPIs completely.
911 */
912 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
913 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
914 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
915}
916
917/*
918 * Called when the last reference to the mm
919 * is dropped: either by a lazy thread or by
920 * mmput. Free the page directory and the mm.
921 */
922void __mmdrop(struct mm_struct *mm)
923{
924 BUG_ON(mm == &init_mm);
925 WARN_ON_ONCE(mm == current->mm);
926
927 /* Ensure no CPUs are using this as their lazy tlb mm */
928 cleanup_lazy_tlbs(mm);
929
930 WARN_ON_ONCE(mm == current->active_mm);
931 mm_free_pgd(mm);
932 destroy_context(mm);
933 mmu_notifier_subscriptions_destroy(mm);
934 check_mm(mm);
935 put_user_ns(mm->user_ns);
936 mm_pasid_drop(mm);
937 mm_destroy_cid(mm);
938 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
939
940 free_mm(mm);
941}
942EXPORT_SYMBOL_GPL(__mmdrop);
943
944static void mmdrop_async_fn(struct work_struct *work)
945{
946 struct mm_struct *mm;
947
948 mm = container_of(work, struct mm_struct, async_put_work);
949 __mmdrop(mm);
950}
951
952static void mmdrop_async(struct mm_struct *mm)
953{
954 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
955 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
956 schedule_work(&mm->async_put_work);
957 }
958}
959
960static inline void free_signal_struct(struct signal_struct *sig)
961{
962 taskstats_tgid_free(sig);
963 sched_autogroup_exit(sig);
964 /*
965 * __mmdrop is not safe to call from softirq context on x86 due to
966 * pgd_dtor so postpone it to the async context
967 */
968 if (sig->oom_mm)
969 mmdrop_async(sig->oom_mm);
970 kmem_cache_free(signal_cachep, sig);
971}
972
973static inline void put_signal_struct(struct signal_struct *sig)
974{
975 if (refcount_dec_and_test(&sig->sigcnt))
976 free_signal_struct(sig);
977}
978
979void __put_task_struct(struct task_struct *tsk)
980{
981 WARN_ON(!tsk->exit_state);
982 WARN_ON(refcount_read(&tsk->usage));
983 WARN_ON(tsk == current);
984
985 sched_ext_free(tsk);
986 io_uring_free(tsk);
987 cgroup_free(tsk);
988 task_numa_free(tsk, true);
989 security_task_free(tsk);
990 exit_creds(tsk);
991 delayacct_tsk_free(tsk);
992 put_signal_struct(tsk->signal);
993 sched_core_free(tsk);
994 free_task(tsk);
995}
996EXPORT_SYMBOL_GPL(__put_task_struct);
997
998void __put_task_struct_rcu_cb(struct rcu_head *rhp)
999{
1000 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
1001
1002 __put_task_struct(task);
1003}
1004EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
1005
1006void __init __weak arch_task_cache_init(void) { }
1007
1008/*
1009 * set_max_threads
1010 */
1011static void __init set_max_threads(unsigned int max_threads_suggested)
1012{
1013 u64 threads;
1014 unsigned long nr_pages = memblock_estimated_nr_free_pages();
1015
1016 /*
1017 * The number of threads shall be limited such that the thread
1018 * structures may only consume a small part of the available memory.
1019 */
1020 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1021 threads = MAX_THREADS;
1022 else
1023 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1024 (u64) THREAD_SIZE * 8UL);
1025
1026 if (threads > max_threads_suggested)
1027 threads = max_threads_suggested;
1028
1029 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1030}
1031
1032#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1033/* Initialized by the architecture: */
1034int arch_task_struct_size __read_mostly;
1035#endif
1036
1037static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
1038{
1039 /* Fetch thread_struct whitelist for the architecture. */
1040 arch_thread_struct_whitelist(offset, size);
1041
1042 /*
1043 * Handle zero-sized whitelist or empty thread_struct, otherwise
1044 * adjust offset to position of thread_struct in task_struct.
1045 */
1046 if (unlikely(*size == 0))
1047 *offset = 0;
1048 else
1049 *offset += offsetof(struct task_struct, thread);
1050}
1051
1052void __init fork_init(void)
1053{
1054 int i;
1055#ifndef ARCH_MIN_TASKALIGN
1056#define ARCH_MIN_TASKALIGN 0
1057#endif
1058 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1059 unsigned long useroffset, usersize;
1060
1061 /* create a slab on which task_structs can be allocated */
1062 task_struct_whitelist(&useroffset, &usersize);
1063 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1064 arch_task_struct_size, align,
1065 SLAB_PANIC|SLAB_ACCOUNT,
1066 useroffset, usersize, NULL);
1067
1068 /* do the arch specific task caches init */
1069 arch_task_cache_init();
1070
1071 set_max_threads(MAX_THREADS);
1072
1073 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1074 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1075 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1076 init_task.signal->rlim[RLIMIT_NPROC];
1077
1078 for (i = 0; i < UCOUNT_COUNTS; i++)
1079 init_user_ns.ucount_max[i] = max_threads/2;
1080
1081 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1082 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1083 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1084 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1085
1086#ifdef CONFIG_VMAP_STACK
1087 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1088 NULL, free_vm_stack_cache);
1089#endif
1090
1091 scs_init();
1092
1093 lockdep_init_task(&init_task);
1094 uprobes_init();
1095}
1096
1097int __weak arch_dup_task_struct(struct task_struct *dst,
1098 struct task_struct *src)
1099{
1100 *dst = *src;
1101 return 0;
1102}
1103
1104void set_task_stack_end_magic(struct task_struct *tsk)
1105{
1106 unsigned long *stackend;
1107
1108 stackend = end_of_stack(tsk);
1109 *stackend = STACK_END_MAGIC; /* for overflow detection */
1110}
1111
1112static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1113{
1114 struct task_struct *tsk;
1115 int err;
1116
1117 if (node == NUMA_NO_NODE)
1118 node = tsk_fork_get_node(orig);
1119 tsk = alloc_task_struct_node(node);
1120 if (!tsk)
1121 return NULL;
1122
1123 err = arch_dup_task_struct(tsk, orig);
1124 if (err)
1125 goto free_tsk;
1126
1127 err = alloc_thread_stack_node(tsk, node);
1128 if (err)
1129 goto free_tsk;
1130
1131#ifdef CONFIG_THREAD_INFO_IN_TASK
1132 refcount_set(&tsk->stack_refcount, 1);
1133#endif
1134 account_kernel_stack(tsk, 1);
1135
1136 err = scs_prepare(tsk, node);
1137 if (err)
1138 goto free_stack;
1139
1140#ifdef CONFIG_SECCOMP
1141 /*
1142 * We must handle setting up seccomp filters once we're under
1143 * the sighand lock in case orig has changed between now and
1144 * then. Until then, filter must be NULL to avoid messing up
1145 * the usage counts on the error path calling free_task.
1146 */
1147 tsk->seccomp.filter = NULL;
1148#endif
1149
1150 setup_thread_stack(tsk, orig);
1151 clear_user_return_notifier(tsk);
1152 clear_tsk_need_resched(tsk);
1153 set_task_stack_end_magic(tsk);
1154 clear_syscall_work_syscall_user_dispatch(tsk);
1155
1156#ifdef CONFIG_STACKPROTECTOR
1157 tsk->stack_canary = get_random_canary();
1158#endif
1159 if (orig->cpus_ptr == &orig->cpus_mask)
1160 tsk->cpus_ptr = &tsk->cpus_mask;
1161 dup_user_cpus_ptr(tsk, orig, node);
1162
1163 /*
1164 * One for the user space visible state that goes away when reaped.
1165 * One for the scheduler.
1166 */
1167 refcount_set(&tsk->rcu_users, 2);
1168 /* One for the rcu users */
1169 refcount_set(&tsk->usage, 1);
1170#ifdef CONFIG_BLK_DEV_IO_TRACE
1171 tsk->btrace_seq = 0;
1172#endif
1173 tsk->splice_pipe = NULL;
1174 tsk->task_frag.page = NULL;
1175 tsk->wake_q.next = NULL;
1176 tsk->worker_private = NULL;
1177
1178 kcov_task_init(tsk);
1179 kmsan_task_create(tsk);
1180 kmap_local_fork(tsk);
1181
1182#ifdef CONFIG_FAULT_INJECTION
1183 tsk->fail_nth = 0;
1184#endif
1185
1186#ifdef CONFIG_BLK_CGROUP
1187 tsk->throttle_disk = NULL;
1188 tsk->use_memdelay = 0;
1189#endif
1190
1191#ifdef CONFIG_ARCH_HAS_CPU_PASID
1192 tsk->pasid_activated = 0;
1193#endif
1194
1195#ifdef CONFIG_MEMCG
1196 tsk->active_memcg = NULL;
1197#endif
1198
1199#ifdef CONFIG_X86_BUS_LOCK_DETECT
1200 tsk->reported_split_lock = 0;
1201#endif
1202
1203#ifdef CONFIG_SCHED_MM_CID
1204 tsk->mm_cid = -1;
1205 tsk->last_mm_cid = -1;
1206 tsk->mm_cid_active = 0;
1207 tsk->migrate_from_cpu = -1;
1208#endif
1209 return tsk;
1210
1211free_stack:
1212 exit_task_stack_account(tsk);
1213 free_thread_stack(tsk);
1214free_tsk:
1215 free_task_struct(tsk);
1216 return NULL;
1217}
1218
1219__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1220
1221static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1222
1223static int __init coredump_filter_setup(char *s)
1224{
1225 default_dump_filter =
1226 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1227 MMF_DUMP_FILTER_MASK;
1228 return 1;
1229}
1230
1231__setup("coredump_filter=", coredump_filter_setup);
1232
1233#include <linux/init_task.h>
1234
1235static void mm_init_aio(struct mm_struct *mm)
1236{
1237#ifdef CONFIG_AIO
1238 spin_lock_init(&mm->ioctx_lock);
1239 mm->ioctx_table = NULL;
1240#endif
1241}
1242
1243static __always_inline void mm_clear_owner(struct mm_struct *mm,
1244 struct task_struct *p)
1245{
1246#ifdef CONFIG_MEMCG
1247 if (mm->owner == p)
1248 WRITE_ONCE(mm->owner, NULL);
1249#endif
1250}
1251
1252static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1253{
1254#ifdef CONFIG_MEMCG
1255 mm->owner = p;
1256#endif
1257}
1258
1259static void mm_init_uprobes_state(struct mm_struct *mm)
1260{
1261#ifdef CONFIG_UPROBES
1262 mm->uprobes_state.xol_area = NULL;
1263#endif
1264}
1265
1266static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1267 struct user_namespace *user_ns)
1268{
1269 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1270 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1271 atomic_set(&mm->mm_users, 1);
1272 atomic_set(&mm->mm_count, 1);
1273 seqcount_init(&mm->write_protect_seq);
1274 mmap_init_lock(mm);
1275 INIT_LIST_HEAD(&mm->mmlist);
1276#ifdef CONFIG_PER_VMA_LOCK
1277 mm->mm_lock_seq = 0;
1278#endif
1279 mm_pgtables_bytes_init(mm);
1280 mm->map_count = 0;
1281 mm->locked_vm = 0;
1282 atomic64_set(&mm->pinned_vm, 0);
1283 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1284 spin_lock_init(&mm->page_table_lock);
1285 spin_lock_init(&mm->arg_lock);
1286 mm_init_cpumask(mm);
1287 mm_init_aio(mm);
1288 mm_init_owner(mm, p);
1289 mm_pasid_init(mm);
1290 RCU_INIT_POINTER(mm->exe_file, NULL);
1291 mmu_notifier_subscriptions_init(mm);
1292 init_tlb_flush_pending(mm);
1293#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1294 mm->pmd_huge_pte = NULL;
1295#endif
1296 mm_init_uprobes_state(mm);
1297 hugetlb_count_init(mm);
1298
1299 if (current->mm) {
1300 mm->flags = mmf_init_flags(current->mm->flags);
1301 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1302 } else {
1303 mm->flags = default_dump_filter;
1304 mm->def_flags = 0;
1305 }
1306
1307 if (mm_alloc_pgd(mm))
1308 goto fail_nopgd;
1309
1310 if (init_new_context(p, mm))
1311 goto fail_nocontext;
1312
1313 if (mm_alloc_cid(mm, p))
1314 goto fail_cid;
1315
1316 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1317 NR_MM_COUNTERS))
1318 goto fail_pcpu;
1319
1320 mm->user_ns = get_user_ns(user_ns);
1321 lru_gen_init_mm(mm);
1322 return mm;
1323
1324fail_pcpu:
1325 mm_destroy_cid(mm);
1326fail_cid:
1327 destroy_context(mm);
1328fail_nocontext:
1329 mm_free_pgd(mm);
1330fail_nopgd:
1331 free_mm(mm);
1332 return NULL;
1333}
1334
1335/*
1336 * Allocate and initialize an mm_struct.
1337 */
1338struct mm_struct *mm_alloc(void)
1339{
1340 struct mm_struct *mm;
1341
1342 mm = allocate_mm();
1343 if (!mm)
1344 return NULL;
1345
1346 memset(mm, 0, sizeof(*mm));
1347 return mm_init(mm, current, current_user_ns());
1348}
1349EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1350
1351static inline void __mmput(struct mm_struct *mm)
1352{
1353 VM_BUG_ON(atomic_read(&mm->mm_users));
1354
1355 uprobe_clear_state(mm);
1356 exit_aio(mm);
1357 ksm_exit(mm);
1358 khugepaged_exit(mm); /* must run before exit_mmap */
1359 exit_mmap(mm);
1360 mm_put_huge_zero_folio(mm);
1361 set_mm_exe_file(mm, NULL);
1362 if (!list_empty(&mm->mmlist)) {
1363 spin_lock(&mmlist_lock);
1364 list_del(&mm->mmlist);
1365 spin_unlock(&mmlist_lock);
1366 }
1367 if (mm->binfmt)
1368 module_put(mm->binfmt->module);
1369 lru_gen_del_mm(mm);
1370 mmdrop(mm);
1371}
1372
1373/*
1374 * Decrement the use count and release all resources for an mm.
1375 */
1376void mmput(struct mm_struct *mm)
1377{
1378 might_sleep();
1379
1380 if (atomic_dec_and_test(&mm->mm_users))
1381 __mmput(mm);
1382}
1383EXPORT_SYMBOL_GPL(mmput);
1384
1385#ifdef CONFIG_MMU
1386static void mmput_async_fn(struct work_struct *work)
1387{
1388 struct mm_struct *mm = container_of(work, struct mm_struct,
1389 async_put_work);
1390
1391 __mmput(mm);
1392}
1393
1394void mmput_async(struct mm_struct *mm)
1395{
1396 if (atomic_dec_and_test(&mm->mm_users)) {
1397 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1398 schedule_work(&mm->async_put_work);
1399 }
1400}
1401EXPORT_SYMBOL_GPL(mmput_async);
1402#endif
1403
1404/**
1405 * set_mm_exe_file - change a reference to the mm's executable file
1406 * @mm: The mm to change.
1407 * @new_exe_file: The new file to use.
1408 *
1409 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1410 *
1411 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1412 * invocations: in mmput() nobody alive left, in execve it happens before
1413 * the new mm is made visible to anyone.
1414 *
1415 * Can only fail if new_exe_file != NULL.
1416 */
1417int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1418{
1419 struct file *old_exe_file;
1420
1421 /*
1422 * It is safe to dereference the exe_file without RCU as
1423 * this function is only called if nobody else can access
1424 * this mm -- see comment above for justification.
1425 */
1426 old_exe_file = rcu_dereference_raw(mm->exe_file);
1427
1428 if (new_exe_file) {
1429 /*
1430 * We expect the caller (i.e., sys_execve) to already denied
1431 * write access, so this is unlikely to fail.
1432 */
1433 if (unlikely(deny_write_access(new_exe_file)))
1434 return -EACCES;
1435 get_file(new_exe_file);
1436 }
1437 rcu_assign_pointer(mm->exe_file, new_exe_file);
1438 if (old_exe_file) {
1439 allow_write_access(old_exe_file);
1440 fput(old_exe_file);
1441 }
1442 return 0;
1443}
1444
1445/**
1446 * replace_mm_exe_file - replace a reference to the mm's executable file
1447 * @mm: The mm to change.
1448 * @new_exe_file: The new file to use.
1449 *
1450 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1451 *
1452 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1453 */
1454int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1455{
1456 struct vm_area_struct *vma;
1457 struct file *old_exe_file;
1458 int ret = 0;
1459
1460 /* Forbid mm->exe_file change if old file still mapped. */
1461 old_exe_file = get_mm_exe_file(mm);
1462 if (old_exe_file) {
1463 VMA_ITERATOR(vmi, mm, 0);
1464 mmap_read_lock(mm);
1465 for_each_vma(vmi, vma) {
1466 if (!vma->vm_file)
1467 continue;
1468 if (path_equal(&vma->vm_file->f_path,
1469 &old_exe_file->f_path)) {
1470 ret = -EBUSY;
1471 break;
1472 }
1473 }
1474 mmap_read_unlock(mm);
1475 fput(old_exe_file);
1476 if (ret)
1477 return ret;
1478 }
1479
1480 ret = deny_write_access(new_exe_file);
1481 if (ret)
1482 return -EACCES;
1483 get_file(new_exe_file);
1484
1485 /* set the new file */
1486 mmap_write_lock(mm);
1487 old_exe_file = rcu_dereference_raw(mm->exe_file);
1488 rcu_assign_pointer(mm->exe_file, new_exe_file);
1489 mmap_write_unlock(mm);
1490
1491 if (old_exe_file) {
1492 allow_write_access(old_exe_file);
1493 fput(old_exe_file);
1494 }
1495 return 0;
1496}
1497
1498/**
1499 * get_mm_exe_file - acquire a reference to the mm's executable file
1500 * @mm: The mm of interest.
1501 *
1502 * Returns %NULL if mm has no associated executable file.
1503 * User must release file via fput().
1504 */
1505struct file *get_mm_exe_file(struct mm_struct *mm)
1506{
1507 struct file *exe_file;
1508
1509 rcu_read_lock();
1510 exe_file = get_file_rcu(&mm->exe_file);
1511 rcu_read_unlock();
1512 return exe_file;
1513}
1514
1515/**
1516 * get_task_exe_file - acquire a reference to the task's executable file
1517 * @task: The task.
1518 *
1519 * Returns %NULL if task's mm (if any) has no associated executable file or
1520 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1521 * User must release file via fput().
1522 */
1523struct file *get_task_exe_file(struct task_struct *task)
1524{
1525 struct file *exe_file = NULL;
1526 struct mm_struct *mm;
1527
1528 task_lock(task);
1529 mm = task->mm;
1530 if (mm) {
1531 if (!(task->flags & PF_KTHREAD))
1532 exe_file = get_mm_exe_file(mm);
1533 }
1534 task_unlock(task);
1535 return exe_file;
1536}
1537
1538/**
1539 * get_task_mm - acquire a reference to the task's mm
1540 * @task: The task.
1541 *
1542 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1543 * this kernel workthread has transiently adopted a user mm with use_mm,
1544 * to do its AIO) is not set and if so returns a reference to it, after
1545 * bumping up the use count. User must release the mm via mmput()
1546 * after use. Typically used by /proc and ptrace.
1547 */
1548struct mm_struct *get_task_mm(struct task_struct *task)
1549{
1550 struct mm_struct *mm;
1551
1552 if (task->flags & PF_KTHREAD)
1553 return NULL;
1554
1555 task_lock(task);
1556 mm = task->mm;
1557 if (mm)
1558 mmget(mm);
1559 task_unlock(task);
1560 return mm;
1561}
1562EXPORT_SYMBOL_GPL(get_task_mm);
1563
1564struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1565{
1566 struct mm_struct *mm;
1567 int err;
1568
1569 err = down_read_killable(&task->signal->exec_update_lock);
1570 if (err)
1571 return ERR_PTR(err);
1572
1573 mm = get_task_mm(task);
1574 if (!mm) {
1575 mm = ERR_PTR(-ESRCH);
1576 } else if (mm != current->mm && !ptrace_may_access(task, mode)) {
1577 mmput(mm);
1578 mm = ERR_PTR(-EACCES);
1579 }
1580 up_read(&task->signal->exec_update_lock);
1581
1582 return mm;
1583}
1584
1585static void complete_vfork_done(struct task_struct *tsk)
1586{
1587 struct completion *vfork;
1588
1589 task_lock(tsk);
1590 vfork = tsk->vfork_done;
1591 if (likely(vfork)) {
1592 tsk->vfork_done = NULL;
1593 complete(vfork);
1594 }
1595 task_unlock(tsk);
1596}
1597
1598static int wait_for_vfork_done(struct task_struct *child,
1599 struct completion *vfork)
1600{
1601 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1602 int killed;
1603
1604 cgroup_enter_frozen();
1605 killed = wait_for_completion_state(vfork, state);
1606 cgroup_leave_frozen(false);
1607
1608 if (killed) {
1609 task_lock(child);
1610 child->vfork_done = NULL;
1611 task_unlock(child);
1612 }
1613
1614 put_task_struct(child);
1615 return killed;
1616}
1617
1618/* Please note the differences between mmput and mm_release.
1619 * mmput is called whenever we stop holding onto a mm_struct,
1620 * error success whatever.
1621 *
1622 * mm_release is called after a mm_struct has been removed
1623 * from the current process.
1624 *
1625 * This difference is important for error handling, when we
1626 * only half set up a mm_struct for a new process and need to restore
1627 * the old one. Because we mmput the new mm_struct before
1628 * restoring the old one. . .
1629 * Eric Biederman 10 January 1998
1630 */
1631static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1632{
1633 uprobe_free_utask(tsk);
1634
1635 /* Get rid of any cached register state */
1636 deactivate_mm(tsk, mm);
1637
1638 /*
1639 * Signal userspace if we're not exiting with a core dump
1640 * because we want to leave the value intact for debugging
1641 * purposes.
1642 */
1643 if (tsk->clear_child_tid) {
1644 if (atomic_read(&mm->mm_users) > 1) {
1645 /*
1646 * We don't check the error code - if userspace has
1647 * not set up a proper pointer then tough luck.
1648 */
1649 put_user(0, tsk->clear_child_tid);
1650 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1651 1, NULL, NULL, 0, 0);
1652 }
1653 tsk->clear_child_tid = NULL;
1654 }
1655
1656 /*
1657 * All done, finally we can wake up parent and return this mm to him.
1658 * Also kthread_stop() uses this completion for synchronization.
1659 */
1660 if (tsk->vfork_done)
1661 complete_vfork_done(tsk);
1662}
1663
1664void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1665{
1666 futex_exit_release(tsk);
1667 mm_release(tsk, mm);
1668}
1669
1670void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1671{
1672 futex_exec_release(tsk);
1673 mm_release(tsk, mm);
1674}
1675
1676/**
1677 * dup_mm() - duplicates an existing mm structure
1678 * @tsk: the task_struct with which the new mm will be associated.
1679 * @oldmm: the mm to duplicate.
1680 *
1681 * Allocates a new mm structure and duplicates the provided @oldmm structure
1682 * content into it.
1683 *
1684 * Return: the duplicated mm or NULL on failure.
1685 */
1686static struct mm_struct *dup_mm(struct task_struct *tsk,
1687 struct mm_struct *oldmm)
1688{
1689 struct mm_struct *mm;
1690 int err;
1691
1692 mm = allocate_mm();
1693 if (!mm)
1694 goto fail_nomem;
1695
1696 memcpy(mm, oldmm, sizeof(*mm));
1697
1698 if (!mm_init(mm, tsk, mm->user_ns))
1699 goto fail_nomem;
1700
1701 uprobe_start_dup_mmap();
1702 err = dup_mmap(mm, oldmm);
1703 if (err)
1704 goto free_pt;
1705 uprobe_end_dup_mmap();
1706
1707 mm->hiwater_rss = get_mm_rss(mm);
1708 mm->hiwater_vm = mm->total_vm;
1709
1710 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1711 goto free_pt;
1712
1713 return mm;
1714
1715free_pt:
1716 /* don't put binfmt in mmput, we haven't got module yet */
1717 mm->binfmt = NULL;
1718 mm_init_owner(mm, NULL);
1719 mmput(mm);
1720 if (err)
1721 uprobe_end_dup_mmap();
1722
1723fail_nomem:
1724 return NULL;
1725}
1726
1727static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1728{
1729 struct mm_struct *mm, *oldmm;
1730
1731 tsk->min_flt = tsk->maj_flt = 0;
1732 tsk->nvcsw = tsk->nivcsw = 0;
1733#ifdef CONFIG_DETECT_HUNG_TASK
1734 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1735 tsk->last_switch_time = 0;
1736#endif
1737
1738 tsk->mm = NULL;
1739 tsk->active_mm = NULL;
1740
1741 /*
1742 * Are we cloning a kernel thread?
1743 *
1744 * We need to steal a active VM for that..
1745 */
1746 oldmm = current->mm;
1747 if (!oldmm)
1748 return 0;
1749
1750 if (clone_flags & CLONE_VM) {
1751 mmget(oldmm);
1752 mm = oldmm;
1753 } else {
1754 mm = dup_mm(tsk, current->mm);
1755 if (!mm)
1756 return -ENOMEM;
1757 }
1758
1759 tsk->mm = mm;
1760 tsk->active_mm = mm;
1761 sched_mm_cid_fork(tsk);
1762 return 0;
1763}
1764
1765static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1766{
1767 struct fs_struct *fs = current->fs;
1768 if (clone_flags & CLONE_FS) {
1769 /* tsk->fs is already what we want */
1770 spin_lock(&fs->lock);
1771 /* "users" and "in_exec" locked for check_unsafe_exec() */
1772 if (fs->in_exec) {
1773 spin_unlock(&fs->lock);
1774 return -EAGAIN;
1775 }
1776 fs->users++;
1777 spin_unlock(&fs->lock);
1778 return 0;
1779 }
1780 tsk->fs = copy_fs_struct(fs);
1781 if (!tsk->fs)
1782 return -ENOMEM;
1783 return 0;
1784}
1785
1786static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1787 int no_files)
1788{
1789 struct files_struct *oldf, *newf;
1790
1791 /*
1792 * A background process may not have any files ...
1793 */
1794 oldf = current->files;
1795 if (!oldf)
1796 return 0;
1797
1798 if (no_files) {
1799 tsk->files = NULL;
1800 return 0;
1801 }
1802
1803 if (clone_flags & CLONE_FILES) {
1804 atomic_inc(&oldf->count);
1805 return 0;
1806 }
1807
1808 newf = dup_fd(oldf, NULL);
1809 if (IS_ERR(newf))
1810 return PTR_ERR(newf);
1811
1812 tsk->files = newf;
1813 return 0;
1814}
1815
1816static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1817{
1818 struct sighand_struct *sig;
1819
1820 if (clone_flags & CLONE_SIGHAND) {
1821 refcount_inc(¤t->sighand->count);
1822 return 0;
1823 }
1824 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1825 RCU_INIT_POINTER(tsk->sighand, sig);
1826 if (!sig)
1827 return -ENOMEM;
1828
1829 refcount_set(&sig->count, 1);
1830 spin_lock_irq(¤t->sighand->siglock);
1831 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1832 spin_unlock_irq(¤t->sighand->siglock);
1833
1834 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1835 if (clone_flags & CLONE_CLEAR_SIGHAND)
1836 flush_signal_handlers(tsk, 0);
1837
1838 return 0;
1839}
1840
1841void __cleanup_sighand(struct sighand_struct *sighand)
1842{
1843 if (refcount_dec_and_test(&sighand->count)) {
1844 signalfd_cleanup(sighand);
1845 /*
1846 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1847 * without an RCU grace period, see __lock_task_sighand().
1848 */
1849 kmem_cache_free(sighand_cachep, sighand);
1850 }
1851}
1852
1853/*
1854 * Initialize POSIX timer handling for a thread group.
1855 */
1856static void posix_cpu_timers_init_group(struct signal_struct *sig)
1857{
1858 struct posix_cputimers *pct = &sig->posix_cputimers;
1859 unsigned long cpu_limit;
1860
1861 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1862 posix_cputimers_group_init(pct, cpu_limit);
1863}
1864
1865static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1866{
1867 struct signal_struct *sig;
1868
1869 if (clone_flags & CLONE_THREAD)
1870 return 0;
1871
1872 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1873 tsk->signal = sig;
1874 if (!sig)
1875 return -ENOMEM;
1876
1877 sig->nr_threads = 1;
1878 sig->quick_threads = 1;
1879 atomic_set(&sig->live, 1);
1880 refcount_set(&sig->sigcnt, 1);
1881
1882 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1883 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1884 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1885
1886 init_waitqueue_head(&sig->wait_chldexit);
1887 sig->curr_target = tsk;
1888 init_sigpending(&sig->shared_pending);
1889 INIT_HLIST_HEAD(&sig->multiprocess);
1890 seqlock_init(&sig->stats_lock);
1891 prev_cputime_init(&sig->prev_cputime);
1892
1893#ifdef CONFIG_POSIX_TIMERS
1894 INIT_HLIST_HEAD(&sig->posix_timers);
1895 INIT_HLIST_HEAD(&sig->ignored_posix_timers);
1896 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1897 sig->real_timer.function = it_real_fn;
1898#endif
1899
1900 task_lock(current->group_leader);
1901 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1902 task_unlock(current->group_leader);
1903
1904 posix_cpu_timers_init_group(sig);
1905
1906 tty_audit_fork(sig);
1907 sched_autogroup_fork(sig);
1908
1909 sig->oom_score_adj = current->signal->oom_score_adj;
1910 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1911
1912 mutex_init(&sig->cred_guard_mutex);
1913 init_rwsem(&sig->exec_update_lock);
1914
1915 return 0;
1916}
1917
1918static void copy_seccomp(struct task_struct *p)
1919{
1920#ifdef CONFIG_SECCOMP
1921 /*
1922 * Must be called with sighand->lock held, which is common to
1923 * all threads in the group. Holding cred_guard_mutex is not
1924 * needed because this new task is not yet running and cannot
1925 * be racing exec.
1926 */
1927 assert_spin_locked(¤t->sighand->siglock);
1928
1929 /* Ref-count the new filter user, and assign it. */
1930 get_seccomp_filter(current);
1931 p->seccomp = current->seccomp;
1932
1933 /*
1934 * Explicitly enable no_new_privs here in case it got set
1935 * between the task_struct being duplicated and holding the
1936 * sighand lock. The seccomp state and nnp must be in sync.
1937 */
1938 if (task_no_new_privs(current))
1939 task_set_no_new_privs(p);
1940
1941 /*
1942 * If the parent gained a seccomp mode after copying thread
1943 * flags and between before we held the sighand lock, we have
1944 * to manually enable the seccomp thread flag here.
1945 */
1946 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1947 set_task_syscall_work(p, SECCOMP);
1948#endif
1949}
1950
1951SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1952{
1953 current->clear_child_tid = tidptr;
1954
1955 return task_pid_vnr(current);
1956}
1957
1958static void rt_mutex_init_task(struct task_struct *p)
1959{
1960 raw_spin_lock_init(&p->pi_lock);
1961#ifdef CONFIG_RT_MUTEXES
1962 p->pi_waiters = RB_ROOT_CACHED;
1963 p->pi_top_task = NULL;
1964 p->pi_blocked_on = NULL;
1965#endif
1966}
1967
1968static inline void init_task_pid_links(struct task_struct *task)
1969{
1970 enum pid_type type;
1971
1972 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1973 INIT_HLIST_NODE(&task->pid_links[type]);
1974}
1975
1976static inline void
1977init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1978{
1979 if (type == PIDTYPE_PID)
1980 task->thread_pid = pid;
1981 else
1982 task->signal->pids[type] = pid;
1983}
1984
1985static inline void rcu_copy_process(struct task_struct *p)
1986{
1987#ifdef CONFIG_PREEMPT_RCU
1988 p->rcu_read_lock_nesting = 0;
1989 p->rcu_read_unlock_special.s = 0;
1990 p->rcu_blocked_node = NULL;
1991 INIT_LIST_HEAD(&p->rcu_node_entry);
1992#endif /* #ifdef CONFIG_PREEMPT_RCU */
1993#ifdef CONFIG_TASKS_RCU
1994 p->rcu_tasks_holdout = false;
1995 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1996 p->rcu_tasks_idle_cpu = -1;
1997 INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1998#endif /* #ifdef CONFIG_TASKS_RCU */
1999#ifdef CONFIG_TASKS_TRACE_RCU
2000 p->trc_reader_nesting = 0;
2001 p->trc_reader_special.s = 0;
2002 INIT_LIST_HEAD(&p->trc_holdout_list);
2003 INIT_LIST_HEAD(&p->trc_blkd_node);
2004#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
2005}
2006
2007/**
2008 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2009 * @pid: the struct pid for which to create a pidfd
2010 * @flags: flags of the new @pidfd
2011 * @ret: Where to return the file for the pidfd.
2012 *
2013 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2014 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2015 *
2016 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2017 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2018 * pidfd file are prepared.
2019 *
2020 * If this function returns successfully the caller is responsible to either
2021 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2022 * order to install the pidfd into its file descriptor table or they must use
2023 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2024 * respectively.
2025 *
2026 * This function is useful when a pidfd must already be reserved but there
2027 * might still be points of failure afterwards and the caller wants to ensure
2028 * that no pidfd is leaked into its file descriptor table.
2029 *
2030 * Return: On success, a reserved pidfd is returned from the function and a new
2031 * pidfd file is returned in the last argument to the function. On
2032 * error, a negative error code is returned from the function and the
2033 * last argument remains unchanged.
2034 */
2035static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2036{
2037 int pidfd;
2038 struct file *pidfd_file;
2039
2040 pidfd = get_unused_fd_flags(O_CLOEXEC);
2041 if (pidfd < 0)
2042 return pidfd;
2043
2044 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2045 if (IS_ERR(pidfd_file)) {
2046 put_unused_fd(pidfd);
2047 return PTR_ERR(pidfd_file);
2048 }
2049 /*
2050 * anon_inode_getfile() ignores everything outside of the
2051 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2052 */
2053 pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2054 *ret = pidfd_file;
2055 return pidfd;
2056}
2057
2058/**
2059 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2060 * @pid: the struct pid for which to create a pidfd
2061 * @flags: flags of the new @pidfd
2062 * @ret: Where to return the pidfd.
2063 *
2064 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2065 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2066 *
2067 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2068 * task identified by @pid must be a thread-group leader.
2069 *
2070 * If this function returns successfully the caller is responsible to either
2071 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2072 * order to install the pidfd into its file descriptor table or they must use
2073 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2074 * respectively.
2075 *
2076 * This function is useful when a pidfd must already be reserved but there
2077 * might still be points of failure afterwards and the caller wants to ensure
2078 * that no pidfd is leaked into its file descriptor table.
2079 *
2080 * Return: On success, a reserved pidfd is returned from the function and a new
2081 * pidfd file is returned in the last argument to the function. On
2082 * error, a negative error code is returned from the function and the
2083 * last argument remains unchanged.
2084 */
2085int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2086{
2087 bool thread = flags & PIDFD_THREAD;
2088
2089 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2090 return -EINVAL;
2091
2092 return __pidfd_prepare(pid, flags, ret);
2093}
2094
2095static void __delayed_free_task(struct rcu_head *rhp)
2096{
2097 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2098
2099 free_task(tsk);
2100}
2101
2102static __always_inline void delayed_free_task(struct task_struct *tsk)
2103{
2104 if (IS_ENABLED(CONFIG_MEMCG))
2105 call_rcu(&tsk->rcu, __delayed_free_task);
2106 else
2107 free_task(tsk);
2108}
2109
2110static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2111{
2112 /* Skip if kernel thread */
2113 if (!tsk->mm)
2114 return;
2115
2116 /* Skip if spawning a thread or using vfork */
2117 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2118 return;
2119
2120 /* We need to synchronize with __set_oom_adj */
2121 mutex_lock(&oom_adj_mutex);
2122 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2123 /* Update the values in case they were changed after copy_signal */
2124 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2125 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2126 mutex_unlock(&oom_adj_mutex);
2127}
2128
2129#ifdef CONFIG_RV
2130static void rv_task_fork(struct task_struct *p)
2131{
2132 int i;
2133
2134 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2135 p->rv[i].da_mon.monitoring = false;
2136}
2137#else
2138#define rv_task_fork(p) do {} while (0)
2139#endif
2140
2141/*
2142 * This creates a new process as a copy of the old one,
2143 * but does not actually start it yet.
2144 *
2145 * It copies the registers, and all the appropriate
2146 * parts of the process environment (as per the clone
2147 * flags). The actual kick-off is left to the caller.
2148 */
2149__latent_entropy struct task_struct *copy_process(
2150 struct pid *pid,
2151 int trace,
2152 int node,
2153 struct kernel_clone_args *args)
2154{
2155 int pidfd = -1, retval;
2156 struct task_struct *p;
2157 struct multiprocess_signals delayed;
2158 struct file *pidfile = NULL;
2159 const u64 clone_flags = args->flags;
2160 struct nsproxy *nsp = current->nsproxy;
2161
2162 /*
2163 * Don't allow sharing the root directory with processes in a different
2164 * namespace
2165 */
2166 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2167 return ERR_PTR(-EINVAL);
2168
2169 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2170 return ERR_PTR(-EINVAL);
2171
2172 /*
2173 * Thread groups must share signals as well, and detached threads
2174 * can only be started up within the thread group.
2175 */
2176 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2177 return ERR_PTR(-EINVAL);
2178
2179 /*
2180 * Shared signal handlers imply shared VM. By way of the above,
2181 * thread groups also imply shared VM. Blocking this case allows
2182 * for various simplifications in other code.
2183 */
2184 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2185 return ERR_PTR(-EINVAL);
2186
2187 /*
2188 * Siblings of global init remain as zombies on exit since they are
2189 * not reaped by their parent (swapper). To solve this and to avoid
2190 * multi-rooted process trees, prevent global and container-inits
2191 * from creating siblings.
2192 */
2193 if ((clone_flags & CLONE_PARENT) &&
2194 current->signal->flags & SIGNAL_UNKILLABLE)
2195 return ERR_PTR(-EINVAL);
2196
2197 /*
2198 * If the new process will be in a different pid or user namespace
2199 * do not allow it to share a thread group with the forking task.
2200 */
2201 if (clone_flags & CLONE_THREAD) {
2202 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2203 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2204 return ERR_PTR(-EINVAL);
2205 }
2206
2207 if (clone_flags & CLONE_PIDFD) {
2208 /*
2209 * - CLONE_DETACHED is blocked so that we can potentially
2210 * reuse it later for CLONE_PIDFD.
2211 */
2212 if (clone_flags & CLONE_DETACHED)
2213 return ERR_PTR(-EINVAL);
2214 }
2215
2216 /*
2217 * Force any signals received before this point to be delivered
2218 * before the fork happens. Collect up signals sent to multiple
2219 * processes that happen during the fork and delay them so that
2220 * they appear to happen after the fork.
2221 */
2222 sigemptyset(&delayed.signal);
2223 INIT_HLIST_NODE(&delayed.node);
2224
2225 spin_lock_irq(¤t->sighand->siglock);
2226 if (!(clone_flags & CLONE_THREAD))
2227 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2228 recalc_sigpending();
2229 spin_unlock_irq(¤t->sighand->siglock);
2230 retval = -ERESTARTNOINTR;
2231 if (task_sigpending(current))
2232 goto fork_out;
2233
2234 retval = -ENOMEM;
2235 p = dup_task_struct(current, node);
2236 if (!p)
2237 goto fork_out;
2238 p->flags &= ~PF_KTHREAD;
2239 if (args->kthread)
2240 p->flags |= PF_KTHREAD;
2241 if (args->user_worker) {
2242 /*
2243 * Mark us a user worker, and block any signal that isn't
2244 * fatal or STOP
2245 */
2246 p->flags |= PF_USER_WORKER;
2247 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2248 }
2249 if (args->io_thread)
2250 p->flags |= PF_IO_WORKER;
2251
2252 if (args->name)
2253 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2254
2255 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2256 /*
2257 * Clear TID on mm_release()?
2258 */
2259 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2260
2261 ftrace_graph_init_task(p);
2262
2263 rt_mutex_init_task(p);
2264
2265 lockdep_assert_irqs_enabled();
2266#ifdef CONFIG_PROVE_LOCKING
2267 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2268#endif
2269 retval = copy_creds(p, clone_flags);
2270 if (retval < 0)
2271 goto bad_fork_free;
2272
2273 retval = -EAGAIN;
2274 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2275 if (p->real_cred->user != INIT_USER &&
2276 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2277 goto bad_fork_cleanup_count;
2278 }
2279 current->flags &= ~PF_NPROC_EXCEEDED;
2280
2281 /*
2282 * If multiple threads are within copy_process(), then this check
2283 * triggers too late. This doesn't hurt, the check is only there
2284 * to stop root fork bombs.
2285 */
2286 retval = -EAGAIN;
2287 if (data_race(nr_threads >= max_threads))
2288 goto bad_fork_cleanup_count;
2289
2290 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2291 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2292 p->flags |= PF_FORKNOEXEC;
2293 INIT_LIST_HEAD(&p->children);
2294 INIT_LIST_HEAD(&p->sibling);
2295 rcu_copy_process(p);
2296 p->vfork_done = NULL;
2297 spin_lock_init(&p->alloc_lock);
2298
2299 init_sigpending(&p->pending);
2300
2301 p->utime = p->stime = p->gtime = 0;
2302#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2303 p->utimescaled = p->stimescaled = 0;
2304#endif
2305 prev_cputime_init(&p->prev_cputime);
2306
2307#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2308 seqcount_init(&p->vtime.seqcount);
2309 p->vtime.starttime = 0;
2310 p->vtime.state = VTIME_INACTIVE;
2311#endif
2312
2313#ifdef CONFIG_IO_URING
2314 p->io_uring = NULL;
2315#endif
2316
2317 p->default_timer_slack_ns = current->timer_slack_ns;
2318
2319#ifdef CONFIG_PSI
2320 p->psi_flags = 0;
2321#endif
2322
2323 task_io_accounting_init(&p->ioac);
2324 acct_clear_integrals(p);
2325
2326 posix_cputimers_init(&p->posix_cputimers);
2327 tick_dep_init_task(p);
2328
2329 p->io_context = NULL;
2330 audit_set_context(p, NULL);
2331 cgroup_fork(p);
2332 if (args->kthread) {
2333 if (!set_kthread_struct(p))
2334 goto bad_fork_cleanup_delayacct;
2335 }
2336#ifdef CONFIG_NUMA
2337 p->mempolicy = mpol_dup(p->mempolicy);
2338 if (IS_ERR(p->mempolicy)) {
2339 retval = PTR_ERR(p->mempolicy);
2340 p->mempolicy = NULL;
2341 goto bad_fork_cleanup_delayacct;
2342 }
2343#endif
2344#ifdef CONFIG_CPUSETS
2345 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2346 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2347#endif
2348#ifdef CONFIG_TRACE_IRQFLAGS
2349 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2350 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2351 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2352 p->softirqs_enabled = 1;
2353 p->softirq_context = 0;
2354#endif
2355
2356 p->pagefault_disabled = 0;
2357
2358#ifdef CONFIG_LOCKDEP
2359 lockdep_init_task(p);
2360#endif
2361
2362#ifdef CONFIG_DEBUG_MUTEXES
2363 p->blocked_on = NULL; /* not blocked yet */
2364#endif
2365#ifdef CONFIG_BCACHE
2366 p->sequential_io = 0;
2367 p->sequential_io_avg = 0;
2368#endif
2369#ifdef CONFIG_BPF_SYSCALL
2370 RCU_INIT_POINTER(p->bpf_storage, NULL);
2371 p->bpf_ctx = NULL;
2372#endif
2373
2374 /* Perform scheduler related setup. Assign this task to a CPU. */
2375 retval = sched_fork(clone_flags, p);
2376 if (retval)
2377 goto bad_fork_cleanup_policy;
2378
2379 retval = perf_event_init_task(p, clone_flags);
2380 if (retval)
2381 goto bad_fork_sched_cancel_fork;
2382 retval = audit_alloc(p);
2383 if (retval)
2384 goto bad_fork_cleanup_perf;
2385 /* copy all the process information */
2386 shm_init_task(p);
2387 retval = security_task_alloc(p, clone_flags);
2388 if (retval)
2389 goto bad_fork_cleanup_audit;
2390 retval = copy_semundo(clone_flags, p);
2391 if (retval)
2392 goto bad_fork_cleanup_security;
2393 retval = copy_files(clone_flags, p, args->no_files);
2394 if (retval)
2395 goto bad_fork_cleanup_semundo;
2396 retval = copy_fs(clone_flags, p);
2397 if (retval)
2398 goto bad_fork_cleanup_files;
2399 retval = copy_sighand(clone_flags, p);
2400 if (retval)
2401 goto bad_fork_cleanup_fs;
2402 retval = copy_signal(clone_flags, p);
2403 if (retval)
2404 goto bad_fork_cleanup_sighand;
2405 retval = copy_mm(clone_flags, p);
2406 if (retval)
2407 goto bad_fork_cleanup_signal;
2408 retval = copy_namespaces(clone_flags, p);
2409 if (retval)
2410 goto bad_fork_cleanup_mm;
2411 retval = copy_io(clone_flags, p);
2412 if (retval)
2413 goto bad_fork_cleanup_namespaces;
2414 retval = copy_thread(p, args);
2415 if (retval)
2416 goto bad_fork_cleanup_io;
2417
2418 stackleak_task_init(p);
2419
2420 if (pid != &init_struct_pid) {
2421 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2422 args->set_tid_size);
2423 if (IS_ERR(pid)) {
2424 retval = PTR_ERR(pid);
2425 goto bad_fork_cleanup_thread;
2426 }
2427 }
2428
2429 /*
2430 * This has to happen after we've potentially unshared the file
2431 * descriptor table (so that the pidfd doesn't leak into the child
2432 * if the fd table isn't shared).
2433 */
2434 if (clone_flags & CLONE_PIDFD) {
2435 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2436
2437 /* Note that no task has been attached to @pid yet. */
2438 retval = __pidfd_prepare(pid, flags, &pidfile);
2439 if (retval < 0)
2440 goto bad_fork_free_pid;
2441 pidfd = retval;
2442
2443 retval = put_user(pidfd, args->pidfd);
2444 if (retval)
2445 goto bad_fork_put_pidfd;
2446 }
2447
2448#ifdef CONFIG_BLOCK
2449 p->plug = NULL;
2450#endif
2451 futex_init_task(p);
2452
2453 /*
2454 * sigaltstack should be cleared when sharing the same VM
2455 */
2456 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2457 sas_ss_reset(p);
2458
2459 /*
2460 * Syscall tracing and stepping should be turned off in the
2461 * child regardless of CLONE_PTRACE.
2462 */
2463 user_disable_single_step(p);
2464 clear_task_syscall_work(p, SYSCALL_TRACE);
2465#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2466 clear_task_syscall_work(p, SYSCALL_EMU);
2467#endif
2468 clear_tsk_latency_tracing(p);
2469
2470 /* ok, now we should be set up.. */
2471 p->pid = pid_nr(pid);
2472 if (clone_flags & CLONE_THREAD) {
2473 p->group_leader = current->group_leader;
2474 p->tgid = current->tgid;
2475 } else {
2476 p->group_leader = p;
2477 p->tgid = p->pid;
2478 }
2479
2480 p->nr_dirtied = 0;
2481 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2482 p->dirty_paused_when = 0;
2483
2484 p->pdeath_signal = 0;
2485 p->task_works = NULL;
2486 clear_posix_cputimers_work(p);
2487
2488#ifdef CONFIG_KRETPROBES
2489 p->kretprobe_instances.first = NULL;
2490#endif
2491#ifdef CONFIG_RETHOOK
2492 p->rethooks.first = NULL;
2493#endif
2494
2495 /*
2496 * Ensure that the cgroup subsystem policies allow the new process to be
2497 * forked. It should be noted that the new process's css_set can be changed
2498 * between here and cgroup_post_fork() if an organisation operation is in
2499 * progress.
2500 */
2501 retval = cgroup_can_fork(p, args);
2502 if (retval)
2503 goto bad_fork_put_pidfd;
2504
2505 /*
2506 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2507 * the new task on the correct runqueue. All this *before* the task
2508 * becomes visible.
2509 *
2510 * This isn't part of ->can_fork() because while the re-cloning is
2511 * cgroup specific, it unconditionally needs to place the task on a
2512 * runqueue.
2513 */
2514 retval = sched_cgroup_fork(p, args);
2515 if (retval)
2516 goto bad_fork_cancel_cgroup;
2517
2518 /*
2519 * From this point on we must avoid any synchronous user-space
2520 * communication until we take the tasklist-lock. In particular, we do
2521 * not want user-space to be able to predict the process start-time by
2522 * stalling fork(2) after we recorded the start_time but before it is
2523 * visible to the system.
2524 */
2525
2526 p->start_time = ktime_get_ns();
2527 p->start_boottime = ktime_get_boottime_ns();
2528
2529 /*
2530 * Make it visible to the rest of the system, but dont wake it up yet.
2531 * Need tasklist lock for parent etc handling!
2532 */
2533 write_lock_irq(&tasklist_lock);
2534
2535 /* CLONE_PARENT re-uses the old parent */
2536 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2537 p->real_parent = current->real_parent;
2538 p->parent_exec_id = current->parent_exec_id;
2539 if (clone_flags & CLONE_THREAD)
2540 p->exit_signal = -1;
2541 else
2542 p->exit_signal = current->group_leader->exit_signal;
2543 } else {
2544 p->real_parent = current;
2545 p->parent_exec_id = current->self_exec_id;
2546 p->exit_signal = args->exit_signal;
2547 }
2548
2549 klp_copy_process(p);
2550
2551 sched_core_fork(p);
2552
2553 spin_lock(¤t->sighand->siglock);
2554
2555 rv_task_fork(p);
2556
2557 rseq_fork(p, clone_flags);
2558
2559 /* Don't start children in a dying pid namespace */
2560 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2561 retval = -ENOMEM;
2562 goto bad_fork_core_free;
2563 }
2564
2565 /* Let kill terminate clone/fork in the middle */
2566 if (fatal_signal_pending(current)) {
2567 retval = -EINTR;
2568 goto bad_fork_core_free;
2569 }
2570
2571 /* No more failure paths after this point. */
2572
2573 /*
2574 * Copy seccomp details explicitly here, in case they were changed
2575 * before holding sighand lock.
2576 */
2577 copy_seccomp(p);
2578
2579 init_task_pid_links(p);
2580 if (likely(p->pid)) {
2581 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2582
2583 init_task_pid(p, PIDTYPE_PID, pid);
2584 if (thread_group_leader(p)) {
2585 init_task_pid(p, PIDTYPE_TGID, pid);
2586 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2587 init_task_pid(p, PIDTYPE_SID, task_session(current));
2588
2589 if (is_child_reaper(pid)) {
2590 ns_of_pid(pid)->child_reaper = p;
2591 p->signal->flags |= SIGNAL_UNKILLABLE;
2592 }
2593 p->signal->shared_pending.signal = delayed.signal;
2594 p->signal->tty = tty_kref_get(current->signal->tty);
2595 /*
2596 * Inherit has_child_subreaper flag under the same
2597 * tasklist_lock with adding child to the process tree
2598 * for propagate_has_child_subreaper optimization.
2599 */
2600 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2601 p->real_parent->signal->is_child_subreaper;
2602 list_add_tail(&p->sibling, &p->real_parent->children);
2603 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2604 attach_pid(p, PIDTYPE_TGID);
2605 attach_pid(p, PIDTYPE_PGID);
2606 attach_pid(p, PIDTYPE_SID);
2607 __this_cpu_inc(process_counts);
2608 } else {
2609 current->signal->nr_threads++;
2610 current->signal->quick_threads++;
2611 atomic_inc(¤t->signal->live);
2612 refcount_inc(¤t->signal->sigcnt);
2613 task_join_group_stop(p);
2614 list_add_tail_rcu(&p->thread_node,
2615 &p->signal->thread_head);
2616 }
2617 attach_pid(p, PIDTYPE_PID);
2618 nr_threads++;
2619 }
2620 total_forks++;
2621 hlist_del_init(&delayed.node);
2622 spin_unlock(¤t->sighand->siglock);
2623 syscall_tracepoint_update(p);
2624 write_unlock_irq(&tasklist_lock);
2625
2626 if (pidfile)
2627 fd_install(pidfd, pidfile);
2628
2629 proc_fork_connector(p);
2630 sched_post_fork(p);
2631 cgroup_post_fork(p, args);
2632 perf_event_fork(p);
2633
2634 trace_task_newtask(p, clone_flags);
2635 uprobe_copy_process(p, clone_flags);
2636 user_events_fork(p, clone_flags);
2637
2638 copy_oom_score_adj(clone_flags, p);
2639
2640 return p;
2641
2642bad_fork_core_free:
2643 sched_core_free(p);
2644 spin_unlock(¤t->sighand->siglock);
2645 write_unlock_irq(&tasklist_lock);
2646bad_fork_cancel_cgroup:
2647 cgroup_cancel_fork(p, args);
2648bad_fork_put_pidfd:
2649 if (clone_flags & CLONE_PIDFD) {
2650 fput(pidfile);
2651 put_unused_fd(pidfd);
2652 }
2653bad_fork_free_pid:
2654 if (pid != &init_struct_pid)
2655 free_pid(pid);
2656bad_fork_cleanup_thread:
2657 exit_thread(p);
2658bad_fork_cleanup_io:
2659 if (p->io_context)
2660 exit_io_context(p);
2661bad_fork_cleanup_namespaces:
2662 exit_task_namespaces(p);
2663bad_fork_cleanup_mm:
2664 if (p->mm) {
2665 mm_clear_owner(p->mm, p);
2666 mmput(p->mm);
2667 }
2668bad_fork_cleanup_signal:
2669 if (!(clone_flags & CLONE_THREAD))
2670 free_signal_struct(p->signal);
2671bad_fork_cleanup_sighand:
2672 __cleanup_sighand(p->sighand);
2673bad_fork_cleanup_fs:
2674 exit_fs(p); /* blocking */
2675bad_fork_cleanup_files:
2676 exit_files(p); /* blocking */
2677bad_fork_cleanup_semundo:
2678 exit_sem(p);
2679bad_fork_cleanup_security:
2680 security_task_free(p);
2681bad_fork_cleanup_audit:
2682 audit_free(p);
2683bad_fork_cleanup_perf:
2684 perf_event_free_task(p);
2685bad_fork_sched_cancel_fork:
2686 sched_cancel_fork(p);
2687bad_fork_cleanup_policy:
2688 lockdep_free_task(p);
2689#ifdef CONFIG_NUMA
2690 mpol_put(p->mempolicy);
2691#endif
2692bad_fork_cleanup_delayacct:
2693 delayacct_tsk_free(p);
2694bad_fork_cleanup_count:
2695 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2696 exit_creds(p);
2697bad_fork_free:
2698 WRITE_ONCE(p->__state, TASK_DEAD);
2699 exit_task_stack_account(p);
2700 put_task_stack(p);
2701 delayed_free_task(p);
2702fork_out:
2703 spin_lock_irq(¤t->sighand->siglock);
2704 hlist_del_init(&delayed.node);
2705 spin_unlock_irq(¤t->sighand->siglock);
2706 return ERR_PTR(retval);
2707}
2708
2709static inline void init_idle_pids(struct task_struct *idle)
2710{
2711 enum pid_type type;
2712
2713 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2714 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2715 init_task_pid(idle, type, &init_struct_pid);
2716 }
2717}
2718
2719static int idle_dummy(void *dummy)
2720{
2721 /* This function is never called */
2722 return 0;
2723}
2724
2725struct task_struct * __init fork_idle(int cpu)
2726{
2727 struct task_struct *task;
2728 struct kernel_clone_args args = {
2729 .flags = CLONE_VM,
2730 .fn = &idle_dummy,
2731 .fn_arg = NULL,
2732 .kthread = 1,
2733 .idle = 1,
2734 };
2735
2736 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2737 if (!IS_ERR(task)) {
2738 init_idle_pids(task);
2739 init_idle(task, cpu);
2740 }
2741
2742 return task;
2743}
2744
2745/*
2746 * This is like kernel_clone(), but shaved down and tailored to just
2747 * creating io_uring workers. It returns a created task, or an error pointer.
2748 * The returned task is inactive, and the caller must fire it up through
2749 * wake_up_new_task(p). All signals are blocked in the created task.
2750 */
2751struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2752{
2753 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2754 CLONE_IO;
2755 struct kernel_clone_args args = {
2756 .flags = ((lower_32_bits(flags) | CLONE_VM |
2757 CLONE_UNTRACED) & ~CSIGNAL),
2758 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2759 .fn = fn,
2760 .fn_arg = arg,
2761 .io_thread = 1,
2762 .user_worker = 1,
2763 };
2764
2765 return copy_process(NULL, 0, node, &args);
2766}
2767
2768/*
2769 * Ok, this is the main fork-routine.
2770 *
2771 * It copies the process, and if successful kick-starts
2772 * it and waits for it to finish using the VM if required.
2773 *
2774 * args->exit_signal is expected to be checked for sanity by the caller.
2775 */
2776pid_t kernel_clone(struct kernel_clone_args *args)
2777{
2778 u64 clone_flags = args->flags;
2779 struct completion vfork;
2780 struct pid *pid;
2781 struct task_struct *p;
2782 int trace = 0;
2783 pid_t nr;
2784
2785 /*
2786 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2787 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2788 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2789 * field in struct clone_args and it still doesn't make sense to have
2790 * them both point at the same memory location. Performing this check
2791 * here has the advantage that we don't need to have a separate helper
2792 * to check for legacy clone().
2793 */
2794 if ((clone_flags & CLONE_PIDFD) &&
2795 (clone_flags & CLONE_PARENT_SETTID) &&
2796 (args->pidfd == args->parent_tid))
2797 return -EINVAL;
2798
2799 /*
2800 * Determine whether and which event to report to ptracer. When
2801 * called from kernel_thread or CLONE_UNTRACED is explicitly
2802 * requested, no event is reported; otherwise, report if the event
2803 * for the type of forking is enabled.
2804 */
2805 if (!(clone_flags & CLONE_UNTRACED)) {
2806 if (clone_flags & CLONE_VFORK)
2807 trace = PTRACE_EVENT_VFORK;
2808 else if (args->exit_signal != SIGCHLD)
2809 trace = PTRACE_EVENT_CLONE;
2810 else
2811 trace = PTRACE_EVENT_FORK;
2812
2813 if (likely(!ptrace_event_enabled(current, trace)))
2814 trace = 0;
2815 }
2816
2817 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2818 add_latent_entropy();
2819
2820 if (IS_ERR(p))
2821 return PTR_ERR(p);
2822
2823 /*
2824 * Do this prior waking up the new thread - the thread pointer
2825 * might get invalid after that point, if the thread exits quickly.
2826 */
2827 trace_sched_process_fork(current, p);
2828
2829 pid = get_task_pid(p, PIDTYPE_PID);
2830 nr = pid_vnr(pid);
2831
2832 if (clone_flags & CLONE_PARENT_SETTID)
2833 put_user(nr, args->parent_tid);
2834
2835 if (clone_flags & CLONE_VFORK) {
2836 p->vfork_done = &vfork;
2837 init_completion(&vfork);
2838 get_task_struct(p);
2839 }
2840
2841 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2842 /* lock the task to synchronize with memcg migration */
2843 task_lock(p);
2844 lru_gen_add_mm(p->mm);
2845 task_unlock(p);
2846 }
2847
2848 wake_up_new_task(p);
2849
2850 /* forking complete and child started to run, tell ptracer */
2851 if (unlikely(trace))
2852 ptrace_event_pid(trace, pid);
2853
2854 if (clone_flags & CLONE_VFORK) {
2855 if (!wait_for_vfork_done(p, &vfork))
2856 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2857 }
2858
2859 put_pid(pid);
2860 return nr;
2861}
2862
2863/*
2864 * Create a kernel thread.
2865 */
2866pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2867 unsigned long flags)
2868{
2869 struct kernel_clone_args args = {
2870 .flags = ((lower_32_bits(flags) | CLONE_VM |
2871 CLONE_UNTRACED) & ~CSIGNAL),
2872 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2873 .fn = fn,
2874 .fn_arg = arg,
2875 .name = name,
2876 .kthread = 1,
2877 };
2878
2879 return kernel_clone(&args);
2880}
2881
2882/*
2883 * Create a user mode thread.
2884 */
2885pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2886{
2887 struct kernel_clone_args args = {
2888 .flags = ((lower_32_bits(flags) | CLONE_VM |
2889 CLONE_UNTRACED) & ~CSIGNAL),
2890 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2891 .fn = fn,
2892 .fn_arg = arg,
2893 };
2894
2895 return kernel_clone(&args);
2896}
2897
2898#ifdef __ARCH_WANT_SYS_FORK
2899SYSCALL_DEFINE0(fork)
2900{
2901#ifdef CONFIG_MMU
2902 struct kernel_clone_args args = {
2903 .exit_signal = SIGCHLD,
2904 };
2905
2906 return kernel_clone(&args);
2907#else
2908 /* can not support in nommu mode */
2909 return -EINVAL;
2910#endif
2911}
2912#endif
2913
2914#ifdef __ARCH_WANT_SYS_VFORK
2915SYSCALL_DEFINE0(vfork)
2916{
2917 struct kernel_clone_args args = {
2918 .flags = CLONE_VFORK | CLONE_VM,
2919 .exit_signal = SIGCHLD,
2920 };
2921
2922 return kernel_clone(&args);
2923}
2924#endif
2925
2926#ifdef __ARCH_WANT_SYS_CLONE
2927#ifdef CONFIG_CLONE_BACKWARDS
2928SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2929 int __user *, parent_tidptr,
2930 unsigned long, tls,
2931 int __user *, child_tidptr)
2932#elif defined(CONFIG_CLONE_BACKWARDS2)
2933SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2934 int __user *, parent_tidptr,
2935 int __user *, child_tidptr,
2936 unsigned long, tls)
2937#elif defined(CONFIG_CLONE_BACKWARDS3)
2938SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2939 int, stack_size,
2940 int __user *, parent_tidptr,
2941 int __user *, child_tidptr,
2942 unsigned long, tls)
2943#else
2944SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2945 int __user *, parent_tidptr,
2946 int __user *, child_tidptr,
2947 unsigned long, tls)
2948#endif
2949{
2950 struct kernel_clone_args args = {
2951 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2952 .pidfd = parent_tidptr,
2953 .child_tid = child_tidptr,
2954 .parent_tid = parent_tidptr,
2955 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2956 .stack = newsp,
2957 .tls = tls,
2958 };
2959
2960 return kernel_clone(&args);
2961}
2962#endif
2963
2964noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2965 struct clone_args __user *uargs,
2966 size_t usize)
2967{
2968 int err;
2969 struct clone_args args;
2970 pid_t *kset_tid = kargs->set_tid;
2971
2972 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2973 CLONE_ARGS_SIZE_VER0);
2974 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2975 CLONE_ARGS_SIZE_VER1);
2976 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2977 CLONE_ARGS_SIZE_VER2);
2978 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2979
2980 if (unlikely(usize > PAGE_SIZE))
2981 return -E2BIG;
2982 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2983 return -EINVAL;
2984
2985 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2986 if (err)
2987 return err;
2988
2989 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2990 return -EINVAL;
2991
2992 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2993 return -EINVAL;
2994
2995 if (unlikely(args.set_tid && args.set_tid_size == 0))
2996 return -EINVAL;
2997
2998 /*
2999 * Verify that higher 32bits of exit_signal are unset and that
3000 * it is a valid signal
3001 */
3002 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3003 !valid_signal(args.exit_signal)))
3004 return -EINVAL;
3005
3006 if ((args.flags & CLONE_INTO_CGROUP) &&
3007 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3008 return -EINVAL;
3009
3010 *kargs = (struct kernel_clone_args){
3011 .flags = args.flags,
3012 .pidfd = u64_to_user_ptr(args.pidfd),
3013 .child_tid = u64_to_user_ptr(args.child_tid),
3014 .parent_tid = u64_to_user_ptr(args.parent_tid),
3015 .exit_signal = args.exit_signal,
3016 .stack = args.stack,
3017 .stack_size = args.stack_size,
3018 .tls = args.tls,
3019 .set_tid_size = args.set_tid_size,
3020 .cgroup = args.cgroup,
3021 };
3022
3023 if (args.set_tid &&
3024 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3025 (kargs->set_tid_size * sizeof(pid_t))))
3026 return -EFAULT;
3027
3028 kargs->set_tid = kset_tid;
3029
3030 return 0;
3031}
3032
3033/**
3034 * clone3_stack_valid - check and prepare stack
3035 * @kargs: kernel clone args
3036 *
3037 * Verify that the stack arguments userspace gave us are sane.
3038 * In addition, set the stack direction for userspace since it's easy for us to
3039 * determine.
3040 */
3041static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3042{
3043 if (kargs->stack == 0) {
3044 if (kargs->stack_size > 0)
3045 return false;
3046 } else {
3047 if (kargs->stack_size == 0)
3048 return false;
3049
3050 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3051 return false;
3052
3053#if !defined(CONFIG_STACK_GROWSUP)
3054 kargs->stack += kargs->stack_size;
3055#endif
3056 }
3057
3058 return true;
3059}
3060
3061static bool clone3_args_valid(struct kernel_clone_args *kargs)
3062{
3063 /* Verify that no unknown flags are passed along. */
3064 if (kargs->flags &
3065 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3066 return false;
3067
3068 /*
3069 * - make the CLONE_DETACHED bit reusable for clone3
3070 * - make the CSIGNAL bits reusable for clone3
3071 */
3072 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3073 return false;
3074
3075 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3076 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3077 return false;
3078
3079 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3080 kargs->exit_signal)
3081 return false;
3082
3083 if (!clone3_stack_valid(kargs))
3084 return false;
3085
3086 return true;
3087}
3088
3089/**
3090 * sys_clone3 - create a new process with specific properties
3091 * @uargs: argument structure
3092 * @size: size of @uargs
3093 *
3094 * clone3() is the extensible successor to clone()/clone2().
3095 * It takes a struct as argument that is versioned by its size.
3096 *
3097 * Return: On success, a positive PID for the child process.
3098 * On error, a negative errno number.
3099 */
3100SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3101{
3102 int err;
3103
3104 struct kernel_clone_args kargs;
3105 pid_t set_tid[MAX_PID_NS_LEVEL];
3106
3107#ifdef __ARCH_BROKEN_SYS_CLONE3
3108#warning clone3() entry point is missing, please fix
3109 return -ENOSYS;
3110#endif
3111
3112 kargs.set_tid = set_tid;
3113
3114 err = copy_clone_args_from_user(&kargs, uargs, size);
3115 if (err)
3116 return err;
3117
3118 if (!clone3_args_valid(&kargs))
3119 return -EINVAL;
3120
3121 return kernel_clone(&kargs);
3122}
3123
3124void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3125{
3126 struct task_struct *leader, *parent, *child;
3127 int res;
3128
3129 read_lock(&tasklist_lock);
3130 leader = top = top->group_leader;
3131down:
3132 for_each_thread(leader, parent) {
3133 list_for_each_entry(child, &parent->children, sibling) {
3134 res = visitor(child, data);
3135 if (res) {
3136 if (res < 0)
3137 goto out;
3138 leader = child;
3139 goto down;
3140 }
3141up:
3142 ;
3143 }
3144 }
3145
3146 if (leader != top) {
3147 child = leader;
3148 parent = child->real_parent;
3149 leader = parent->group_leader;
3150 goto up;
3151 }
3152out:
3153 read_unlock(&tasklist_lock);
3154}
3155
3156#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3157#define ARCH_MIN_MMSTRUCT_ALIGN 0
3158#endif
3159
3160static void sighand_ctor(void *data)
3161{
3162 struct sighand_struct *sighand = data;
3163
3164 spin_lock_init(&sighand->siglock);
3165 init_waitqueue_head(&sighand->signalfd_wqh);
3166}
3167
3168void __init mm_cache_init(void)
3169{
3170 unsigned int mm_size;
3171
3172 /*
3173 * The mm_cpumask is located at the end of mm_struct, and is
3174 * dynamically sized based on the maximum CPU number this system
3175 * can have, taking hotplug into account (nr_cpu_ids).
3176 */
3177 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3178
3179 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3180 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3181 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3182 offsetof(struct mm_struct, saved_auxv),
3183 sizeof_field(struct mm_struct, saved_auxv),
3184 NULL);
3185}
3186
3187void __init proc_caches_init(void)
3188{
3189 sighand_cachep = kmem_cache_create("sighand_cache",
3190 sizeof(struct sighand_struct), 0,
3191 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3192 SLAB_ACCOUNT, sighand_ctor);
3193 signal_cachep = kmem_cache_create("signal_cache",
3194 sizeof(struct signal_struct), 0,
3195 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3196 NULL);
3197 files_cachep = kmem_cache_create("files_cache",
3198 sizeof(struct files_struct), 0,
3199 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3200 NULL);
3201 fs_cachep = kmem_cache_create("fs_cache",
3202 sizeof(struct fs_struct), 0,
3203 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3204 NULL);
3205
3206 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3207#ifdef CONFIG_PER_VMA_LOCK
3208 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3209#endif
3210 mmap_init();
3211 nsproxy_cache_init();
3212}
3213
3214/*
3215 * Check constraints on flags passed to the unshare system call.
3216 */
3217static int check_unshare_flags(unsigned long unshare_flags)
3218{
3219 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3220 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3221 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3222 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3223 CLONE_NEWTIME))
3224 return -EINVAL;
3225 /*
3226 * Not implemented, but pretend it works if there is nothing
3227 * to unshare. Note that unsharing the address space or the
3228 * signal handlers also need to unshare the signal queues (aka
3229 * CLONE_THREAD).
3230 */
3231 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3232 if (!thread_group_empty(current))
3233 return -EINVAL;
3234 }
3235 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3236 if (refcount_read(¤t->sighand->count) > 1)
3237 return -EINVAL;
3238 }
3239 if (unshare_flags & CLONE_VM) {
3240 if (!current_is_single_threaded())
3241 return -EINVAL;
3242 }
3243
3244 return 0;
3245}
3246
3247/*
3248 * Unshare the filesystem structure if it is being shared
3249 */
3250static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3251{
3252 struct fs_struct *fs = current->fs;
3253
3254 if (!(unshare_flags & CLONE_FS) || !fs)
3255 return 0;
3256
3257 /* don't need lock here; in the worst case we'll do useless copy */
3258 if (fs->users == 1)
3259 return 0;
3260
3261 *new_fsp = copy_fs_struct(fs);
3262 if (!*new_fsp)
3263 return -ENOMEM;
3264
3265 return 0;
3266}
3267
3268/*
3269 * Unshare file descriptor table if it is being shared
3270 */
3271static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
3272{
3273 struct files_struct *fd = current->files;
3274
3275 if ((unshare_flags & CLONE_FILES) &&
3276 (fd && atomic_read(&fd->count) > 1)) {
3277 fd = dup_fd(fd, NULL);
3278 if (IS_ERR(fd))
3279 return PTR_ERR(fd);
3280 *new_fdp = fd;
3281 }
3282
3283 return 0;
3284}
3285
3286/*
3287 * unshare allows a process to 'unshare' part of the process
3288 * context which was originally shared using clone. copy_*
3289 * functions used by kernel_clone() cannot be used here directly
3290 * because they modify an inactive task_struct that is being
3291 * constructed. Here we are modifying the current, active,
3292 * task_struct.
3293 */
3294int ksys_unshare(unsigned long unshare_flags)
3295{
3296 struct fs_struct *fs, *new_fs = NULL;
3297 struct files_struct *new_fd = NULL;
3298 struct cred *new_cred = NULL;
3299 struct nsproxy *new_nsproxy = NULL;
3300 int do_sysvsem = 0;
3301 int err;
3302
3303 /*
3304 * If unsharing a user namespace must also unshare the thread group
3305 * and unshare the filesystem root and working directories.
3306 */
3307 if (unshare_flags & CLONE_NEWUSER)
3308 unshare_flags |= CLONE_THREAD | CLONE_FS;
3309 /*
3310 * If unsharing vm, must also unshare signal handlers.
3311 */
3312 if (unshare_flags & CLONE_VM)
3313 unshare_flags |= CLONE_SIGHAND;
3314 /*
3315 * If unsharing a signal handlers, must also unshare the signal queues.
3316 */
3317 if (unshare_flags & CLONE_SIGHAND)
3318 unshare_flags |= CLONE_THREAD;
3319 /*
3320 * If unsharing namespace, must also unshare filesystem information.
3321 */
3322 if (unshare_flags & CLONE_NEWNS)
3323 unshare_flags |= CLONE_FS;
3324
3325 err = check_unshare_flags(unshare_flags);
3326 if (err)
3327 goto bad_unshare_out;
3328 /*
3329 * CLONE_NEWIPC must also detach from the undolist: after switching
3330 * to a new ipc namespace, the semaphore arrays from the old
3331 * namespace are unreachable.
3332 */
3333 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3334 do_sysvsem = 1;
3335 err = unshare_fs(unshare_flags, &new_fs);
3336 if (err)
3337 goto bad_unshare_out;
3338 err = unshare_fd(unshare_flags, &new_fd);
3339 if (err)
3340 goto bad_unshare_cleanup_fs;
3341 err = unshare_userns(unshare_flags, &new_cred);
3342 if (err)
3343 goto bad_unshare_cleanup_fd;
3344 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3345 new_cred, new_fs);
3346 if (err)
3347 goto bad_unshare_cleanup_cred;
3348
3349 if (new_cred) {
3350 err = set_cred_ucounts(new_cred);
3351 if (err)
3352 goto bad_unshare_cleanup_cred;
3353 }
3354
3355 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3356 if (do_sysvsem) {
3357 /*
3358 * CLONE_SYSVSEM is equivalent to sys_exit().
3359 */
3360 exit_sem(current);
3361 }
3362 if (unshare_flags & CLONE_NEWIPC) {
3363 /* Orphan segments in old ns (see sem above). */
3364 exit_shm(current);
3365 shm_init_task(current);
3366 }
3367
3368 if (new_nsproxy)
3369 switch_task_namespaces(current, new_nsproxy);
3370
3371 task_lock(current);
3372
3373 if (new_fs) {
3374 fs = current->fs;
3375 spin_lock(&fs->lock);
3376 current->fs = new_fs;
3377 if (--fs->users)
3378 new_fs = NULL;
3379 else
3380 new_fs = fs;
3381 spin_unlock(&fs->lock);
3382 }
3383
3384 if (new_fd)
3385 swap(current->files, new_fd);
3386
3387 task_unlock(current);
3388
3389 if (new_cred) {
3390 /* Install the new user namespace */
3391 commit_creds(new_cred);
3392 new_cred = NULL;
3393 }
3394 }
3395
3396 perf_event_namespaces(current);
3397
3398bad_unshare_cleanup_cred:
3399 if (new_cred)
3400 put_cred(new_cred);
3401bad_unshare_cleanup_fd:
3402 if (new_fd)
3403 put_files_struct(new_fd);
3404
3405bad_unshare_cleanup_fs:
3406 if (new_fs)
3407 free_fs_struct(new_fs);
3408
3409bad_unshare_out:
3410 return err;
3411}
3412
3413SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3414{
3415 return ksys_unshare(unshare_flags);
3416}
3417
3418/*
3419 * Helper to unshare the files of the current task.
3420 * We don't want to expose copy_files internals to
3421 * the exec layer of the kernel.
3422 */
3423
3424int unshare_files(void)
3425{
3426 struct task_struct *task = current;
3427 struct files_struct *old, *copy = NULL;
3428 int error;
3429
3430 error = unshare_fd(CLONE_FILES, ©);
3431 if (error || !copy)
3432 return error;
3433
3434 old = task->files;
3435 task_lock(task);
3436 task->files = copy;
3437 task_unlock(task);
3438 put_files_struct(old);
3439 return 0;
3440}
3441
3442int sysctl_max_threads(const struct ctl_table *table, int write,
3443 void *buffer, size_t *lenp, loff_t *ppos)
3444{
3445 struct ctl_table t;
3446 int ret;
3447 int threads = max_threads;
3448 int min = 1;
3449 int max = MAX_THREADS;
3450
3451 t = *table;
3452 t.data = &threads;
3453 t.extra1 = &min;
3454 t.extra2 = &max;
3455
3456 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3457 if (ret || !write)
3458 return ret;
3459
3460 max_threads = threads;
3461
3462 return 0;
3463}
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}