projects/linux·blob v6.12 · refs

3432 lines · 83.9 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/sched/ext.h>
27#include <linux/seq_file.h>
28#include <linux/rtmutex.h>
29#include <linux/init.h>
30#include <linux/unistd.h>
31#include <linux/module.h>
32#include <linux/vmalloc.h>
33#include <linux/completion.h>
34#include <linux/personality.h>
35#include <linux/mempolicy.h>
36#include <linux/sem.h>
37#include <linux/file.h>
38#include <linux/fdtable.h>
39#include <linux/iocontext.h>
40#include <linux/key.h>
41#include <linux/kmsan.h>
42#include <linux/binfmts.h>
43#include <linux/mman.h>
44#include <linux/mmu_notifier.h>
45#include <linux/fs.h>
46#include <linux/mm.h>
47#include <linux/mm_inline.h>
48#include <linux/memblock.h>
49#include <linux/nsproxy.h>
50#include <linux/capability.h>
51#include <linux/cpu.h>
52#include <linux/cgroup.h>
53#include <linux/security.h>
54#include <linux/hugetlb.h>
55#include <linux/seccomp.h>
56#include <linux/swap.h>
57#include <linux/syscalls.h>
58#include <linux/syscall_user_dispatch.h>
59#include <linux/jiffies.h>
60#include <linux/futex.h>
61#include <linux/compat.h>
62#include <linux/kthread.h>
63#include <linux/task_io_accounting_ops.h>
64#include <linux/rcupdate.h>
65#include <linux/ptrace.h>
66#include <linux/mount.h>
67#include <linux/audit.h>
68#include <linux/memcontrol.h>
69#include <linux/ftrace.h>
70#include <linux/proc_fs.h>
71#include <linux/profile.h>
72#include <linux/rmap.h>
73#include <linux/ksm.h>
74#include <linux/acct.h>
75#include <linux/userfaultfd_k.h>
76#include <linux/tsacct_kern.h>
77#include <linux/cn_proc.h>
78#include <linux/freezer.h>
79#include <linux/delayacct.h>
80#include <linux/taskstats_kern.h>
81#include <linux/tty.h>
82#include <linux/fs_struct.h>
83#include <linux/magic.h>
84#include <linux/perf_event.h>
85#include <linux/posix-timers.h>
86#include <linux/user-return-notifier.h>
87#include <linux/oom.h>
88#include <linux/khugepaged.h>
89#include <linux/signalfd.h>
90#include <linux/uprobes.h>
91#include <linux/aio.h>
92#include <linux/compiler.h>
93#include <linux/sysctl.h>
94#include <linux/kcov.h>
95#include <linux/livepatch.h>
96#include <linux/thread_info.h>
97#include <linux/stackleak.h>
98#include <linux/kasan.h>
99#include <linux/scs.h>
100#include <linux/io_uring.h>
101#include <linux/bpf.h>
102#include <linux/stackprotector.h>
103#include <linux/user_events.h>
104#include <linux/iommu.h>
105#include <linux/rseq.h>
106#include <uapi/linux/pidfd.h>
107#include <linux/pidfs.h>
108#include <linux/tick.h>
109
110#include <asm/pgalloc.h>
111#include <linux/uaccess.h>
112#include <asm/mmu_context.h>
113#include <asm/cacheflush.h>
114#include <asm/tlbflush.h>
115
116#include <trace/events/sched.h>
117
118#define CREATE_TRACE_POINTS
119#include <trace/events/task.h>
120
121#include <kunit/visibility.h>
122
123/*
124 * Minimum number of threads to boot the kernel
125 */
126#define MIN_THREADS 20
127
128/*
129 * Maximum number of threads
130 */
131#define MAX_THREADS FUTEX_TID_MASK
132
133/*
134 * Protected counters by write_lock_irq(&tasklist_lock)
135 */
136unsigned long total_forks; /* Handle normal Linux uptimes. */
137int nr_threads; /* The idle threads do not count.. */
138
139static int max_threads; /* tunable limit on nr_threads */
140
141#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
142
143static const char * const resident_page_types[] = {
144 NAMED_ARRAY_INDEX(MM_FILEPAGES),
145 NAMED_ARRAY_INDEX(MM_ANONPAGES),
146 NAMED_ARRAY_INDEX(MM_SWAPENTS),
147 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
148};
149
150DEFINE_PER_CPU(unsigned long, process_counts) = 0;
151
152__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
153
154#ifdef CONFIG_PROVE_RCU
155int lockdep_tasklist_lock_is_held(void)
156{
157 return lockdep_is_held(&tasklist_lock);
158}
159EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
160#endif /* #ifdef CONFIG_PROVE_RCU */
161
162int nr_processes(void)
163{
164 int cpu;
165 int total = 0;
166
167 for_each_possible_cpu(cpu)
168 total += per_cpu(process_counts, cpu);
169
170 return total;
171}
172
173void __weak arch_release_task_struct(struct task_struct *tsk)
174{
175}
176
177static struct kmem_cache *task_struct_cachep;
178
179static inline struct task_struct *alloc_task_struct_node(int node)
180{
181 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
182}
183
184static inline void free_task_struct(struct task_struct *tsk)
185{
186 kmem_cache_free(task_struct_cachep, tsk);
187}
188
189/*
190 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
191 * kmemcache based allocator.
192 */
193# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
194
195# ifdef CONFIG_VMAP_STACK
196/*
197 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
198 * flush. Try to minimize the number of calls by caching stacks.
199 */
200#define NR_CACHED_STACKS 2
201static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
202
203struct vm_stack {
204 struct rcu_head rcu;
205 struct vm_struct *stack_vm_area;
206};
207
208static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
209{
210 unsigned int i;
211
212 for (i = 0; i < NR_CACHED_STACKS; i++) {
213 struct vm_struct *tmp = NULL;
214
215 if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm))
216 return true;
217 }
218 return false;
219}
220
221static void thread_stack_free_rcu(struct rcu_head *rh)
222{
223 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
224
225 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
226 return;
227
228 vfree(vm_stack);
229}
230
231static void thread_stack_delayed_free(struct task_struct *tsk)
232{
233 struct vm_stack *vm_stack = tsk->stack;
234
235 vm_stack->stack_vm_area = tsk->stack_vm_area;
236 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
237}
238
239static int free_vm_stack_cache(unsigned int cpu)
240{
241 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
242 int i;
243
244 for (i = 0; i < NR_CACHED_STACKS; i++) {
245 struct vm_struct *vm_stack = cached_vm_stacks[i];
246
247 if (!vm_stack)
248 continue;
249
250 vfree(vm_stack->addr);
251 cached_vm_stacks[i] = NULL;
252 }
253
254 return 0;
255}
256
257static int memcg_charge_kernel_stack(struct vm_struct *vm)
258{
259 int i;
260 int ret;
261 int nr_charged = 0;
262
263 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
264
265 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
266 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
267 if (ret)
268 goto err;
269 nr_charged++;
270 }
271 return 0;
272err:
273 for (i = 0; i < nr_charged; i++)
274 memcg_kmem_uncharge_page(vm->pages[i], 0);
275 return ret;
276}
277
278static int alloc_thread_stack_node(struct task_struct *tsk, int node)
279{
280 struct vm_struct *vm;
281 void *stack;
282 int i;
283
284 for (i = 0; i < NR_CACHED_STACKS; i++) {
285 struct vm_struct *s;
286
287 s = this_cpu_xchg(cached_stacks[i], NULL);
288
289 if (!s)
290 continue;
291
292 /* Reset stack metadata. */
293 kasan_unpoison_range(s->addr, THREAD_SIZE);
294
295 stack = kasan_reset_tag(s->addr);
296
297 /* Clear stale pointers from reused stack. */
298 memset(stack, 0, THREAD_SIZE);
299
300 if (memcg_charge_kernel_stack(s)) {
301 vfree(s->addr);
302 return -ENOMEM;
303 }
304
305 tsk->stack_vm_area = s;
306 tsk->stack = stack;
307 return 0;
308 }
309
310 /*
311 * Allocated stacks are cached and later reused by new threads,
312 * so memcg accounting is performed manually on assigning/releasing
313 * stacks to tasks. Drop __GFP_ACCOUNT.
314 */
315 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
316 VMALLOC_START, VMALLOC_END,
317 THREADINFO_GFP & ~__GFP_ACCOUNT,
318 PAGE_KERNEL,
319 0, node, __builtin_return_address(0));
320 if (!stack)
321 return -ENOMEM;
322
323 vm = find_vm_area(stack);
324 if (memcg_charge_kernel_stack(vm)) {
325 vfree(stack);
326 return -ENOMEM;
327 }
328 /*
329 * We can't call find_vm_area() in interrupt context, and
330 * free_thread_stack() can be called in interrupt context,
331 * so cache the vm_struct.
332 */
333 tsk->stack_vm_area = vm;
334 stack = kasan_reset_tag(stack);
335 tsk->stack = stack;
336 return 0;
337}
338
339static void free_thread_stack(struct task_struct *tsk)
340{
341 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
342 thread_stack_delayed_free(tsk);
343
344 tsk->stack = NULL;
345 tsk->stack_vm_area = NULL;
346}
347
348# else /* !CONFIG_VMAP_STACK */
349
350static void thread_stack_free_rcu(struct rcu_head *rh)
351{
352 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
353}
354
355static void thread_stack_delayed_free(struct task_struct *tsk)
356{
357 struct rcu_head *rh = tsk->stack;
358
359 call_rcu(rh, thread_stack_free_rcu);
360}
361
362static int alloc_thread_stack_node(struct task_struct *tsk, int node)
363{
364 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
365 THREAD_SIZE_ORDER);
366
367 if (likely(page)) {
368 tsk->stack = kasan_reset_tag(page_address(page));
369 return 0;
370 }
371 return -ENOMEM;
372}
373
374static void free_thread_stack(struct task_struct *tsk)
375{
376 thread_stack_delayed_free(tsk);
377 tsk->stack = NULL;
378}
379
380# endif /* CONFIG_VMAP_STACK */
381# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
382
383static struct kmem_cache *thread_stack_cache;
384
385static void thread_stack_free_rcu(struct rcu_head *rh)
386{
387 kmem_cache_free(thread_stack_cache, rh);
388}
389
390static void thread_stack_delayed_free(struct task_struct *tsk)
391{
392 struct rcu_head *rh = tsk->stack;
393
394 call_rcu(rh, thread_stack_free_rcu);
395}
396
397static int alloc_thread_stack_node(struct task_struct *tsk, int node)
398{
399 unsigned long *stack;
400 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
401 stack = kasan_reset_tag(stack);
402 tsk->stack = stack;
403 return stack ? 0 : -ENOMEM;
404}
405
406static void free_thread_stack(struct task_struct *tsk)
407{
408 thread_stack_delayed_free(tsk);
409 tsk->stack = NULL;
410}
411
412void thread_stack_cache_init(void)
413{
414 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
415 THREAD_SIZE, THREAD_SIZE, 0, 0,
416 THREAD_SIZE, NULL);
417 BUG_ON(thread_stack_cache == NULL);
418}
419
420# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
421
422/* SLAB cache for signal_struct structures (tsk->signal) */
423static struct kmem_cache *signal_cachep;
424
425/* SLAB cache for sighand_struct structures (tsk->sighand) */
426struct kmem_cache *sighand_cachep;
427
428/* SLAB cache for files_struct structures (tsk->files) */
429struct kmem_cache *files_cachep;
430
431/* SLAB cache for fs_struct structures (tsk->fs) */
432struct kmem_cache *fs_cachep;
433
434/* SLAB cache for vm_area_struct structures */
435static struct kmem_cache *vm_area_cachep;
436
437/* SLAB cache for mm_struct structures (tsk->mm) */
438static struct kmem_cache *mm_cachep;
439
440#ifdef CONFIG_PER_VMA_LOCK
441
442/* SLAB cache for vm_area_struct.lock */
443static struct kmem_cache *vma_lock_cachep;
444
445static bool vma_lock_alloc(struct vm_area_struct *vma)
446{
447 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
448 if (!vma->vm_lock)
449 return false;
450
451 init_rwsem(&vma->vm_lock->lock);
452 vma->vm_lock_seq = -1;
453
454 return true;
455}
456
457static inline void vma_lock_free(struct vm_area_struct *vma)
458{
459 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
460}
461
462#else /* CONFIG_PER_VMA_LOCK */
463
464static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
465static inline void vma_lock_free(struct vm_area_struct *vma) {}
466
467#endif /* CONFIG_PER_VMA_LOCK */
468
469struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
470{
471 struct vm_area_struct *vma;
472
473 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
474 if (!vma)
475 return NULL;
476
477 vma_init(vma, mm);
478 if (!vma_lock_alloc(vma)) {
479 kmem_cache_free(vm_area_cachep, vma);
480 return NULL;
481 }
482
483 return vma;
484}
485
486struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
487{
488 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
489
490 if (!new)
491 return NULL;
492
493 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
494 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
495 /*
496 * orig->shared.rb may be modified concurrently, but the clone
497 * will be reinitialized.
498 */
499 data_race(memcpy(new, orig, sizeof(*new)));
500 if (!vma_lock_alloc(new)) {
501 kmem_cache_free(vm_area_cachep, new);
502 return NULL;
503 }
504 INIT_LIST_HEAD(&new->anon_vma_chain);
505 vma_numab_state_init(new);
506 dup_anon_vma_name(orig, new);
507
508 return new;
509}
510
511void __vm_area_free(struct vm_area_struct *vma)
512{
513 vma_numab_state_free(vma);
514 free_anon_vma_name(vma);
515 vma_lock_free(vma);
516 kmem_cache_free(vm_area_cachep, vma);
517}
518
519#ifdef CONFIG_PER_VMA_LOCK
520static void vm_area_free_rcu_cb(struct rcu_head *head)
521{
522 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
523 vm_rcu);
524
525 /* The vma should not be locked while being destroyed. */
526 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
527 __vm_area_free(vma);
528}
529#endif
530
531void vm_area_free(struct vm_area_struct *vma)
532{
533#ifdef CONFIG_PER_VMA_LOCK
534 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
535#else
536 __vm_area_free(vma);
537#endif
538}
539
540static void account_kernel_stack(struct task_struct *tsk, int account)
541{
542 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
543 struct vm_struct *vm = task_stack_vm_area(tsk);
544 int i;
545
546 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
547 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
548 account * (PAGE_SIZE / 1024));
549 } else {
550 void *stack = task_stack_page(tsk);
551
552 /* All stack pages are in the same node. */
553 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
554 account * (THREAD_SIZE / 1024));
555 }
556}
557
558void exit_task_stack_account(struct task_struct *tsk)
559{
560 account_kernel_stack(tsk, -1);
561
562 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
563 struct vm_struct *vm;
564 int i;
565
566 vm = task_stack_vm_area(tsk);
567 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
568 memcg_kmem_uncharge_page(vm->pages[i], 0);
569 }
570}
571
572static void release_task_stack(struct task_struct *tsk)
573{
574 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
575 return; /* Better to leak the stack than to free prematurely */
576
577 free_thread_stack(tsk);
578}
579
580#ifdef CONFIG_THREAD_INFO_IN_TASK
581void put_task_stack(struct task_struct *tsk)
582{
583 if (refcount_dec_and_test(&tsk->stack_refcount))
584 release_task_stack(tsk);
585}
586#endif
587
588void free_task(struct task_struct *tsk)
589{
590#ifdef CONFIG_SECCOMP
591 WARN_ON_ONCE(tsk->seccomp.filter);
592#endif
593 release_user_cpus_ptr(tsk);
594 scs_release(tsk);
595
596#ifndef CONFIG_THREAD_INFO_IN_TASK
597 /*
598 * The task is finally done with both the stack and thread_info,
599 * so free both.
600 */
601 release_task_stack(tsk);
602#else
603 /*
604 * If the task had a separate stack allocation, it should be gone
605 * by now.
606 */
607 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
608#endif
609 rt_mutex_debug_task_free(tsk);
610 ftrace_graph_exit_task(tsk);
611 arch_release_task_struct(tsk);
612 if (tsk->flags & PF_KTHREAD)
613 free_kthread_struct(tsk);
614 bpf_task_storage_free(tsk);
615 free_task_struct(tsk);
616}
617EXPORT_SYMBOL(free_task);
618
619static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
620{
621 struct file *exe_file;
622
623 exe_file = get_mm_exe_file(oldmm);
624 RCU_INIT_POINTER(mm->exe_file, exe_file);
625}
626
627#ifdef CONFIG_MMU
628static __latent_entropy int dup_mmap(struct mm_struct *mm,
629 struct mm_struct *oldmm)
630{
631 struct vm_area_struct *mpnt, *tmp;
632 int retval;
633 unsigned long charge = 0;
634 LIST_HEAD(uf);
635 VMA_ITERATOR(vmi, mm, 0);
636
637 uprobe_start_dup_mmap();
638 if (mmap_write_lock_killable(oldmm)) {
639 retval = -EINTR;
640 goto fail_uprobe_end;
641 }
642 flush_cache_dup_mm(oldmm);
643 uprobe_dup_mmap(oldmm, mm);
644 /*
645 * Not linked in yet - no deadlock potential:
646 */
647 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
648
649 /* No ordering required: file already has been exposed. */
650 dup_mm_exe_file(mm, oldmm);
651
652 mm->total_vm = oldmm->total_vm;
653 mm->data_vm = oldmm->data_vm;
654 mm->exec_vm = oldmm->exec_vm;
655 mm->stack_vm = oldmm->stack_vm;
656
657 /* Use __mt_dup() to efficiently build an identical maple tree. */
658 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
659 if (unlikely(retval))
660 goto out;
661
662 mt_clear_in_rcu(vmi.mas.tree);
663 for_each_vma(vmi, mpnt) {
664 struct file *file;
665
666 vma_start_write(mpnt);
667 if (mpnt->vm_flags & VM_DONTCOPY) {
668 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
669 mpnt->vm_end, GFP_KERNEL);
670 if (retval)
671 goto loop_out;
672
673 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
674 continue;
675 }
676 charge = 0;
677 /*
678 * Don't duplicate many vmas if we've been oom-killed (for
679 * example)
680 */
681 if (fatal_signal_pending(current)) {
682 retval = -EINTR;
683 goto loop_out;
684 }
685 if (mpnt->vm_flags & VM_ACCOUNT) {
686 unsigned long len = vma_pages(mpnt);
687
688 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
689 goto fail_nomem;
690 charge = len;
691 }
692 tmp = vm_area_dup(mpnt);
693 if (!tmp)
694 goto fail_nomem;
695 retval = vma_dup_policy(mpnt, tmp);
696 if (retval)
697 goto fail_nomem_policy;
698 tmp->vm_mm = mm;
699 retval = dup_userfaultfd(tmp, &uf);
700 if (retval)
701 goto fail_nomem_anon_vma_fork;
702 if (tmp->vm_flags & VM_WIPEONFORK) {
703 /*
704 * VM_WIPEONFORK gets a clean slate in the child.
705 * Don't prepare anon_vma until fault since we don't
706 * copy page for current vma.
707 */
708 tmp->anon_vma = NULL;
709 } else if (anon_vma_fork(tmp, mpnt))
710 goto fail_nomem_anon_vma_fork;
711 vm_flags_clear(tmp, VM_LOCKED_MASK);
712 /*
713 * Copy/update hugetlb private vma information.
714 */
715 if (is_vm_hugetlb_page(tmp))
716 hugetlb_dup_vma_private(tmp);
717
718 /*
719 * Link the vma into the MT. After using __mt_dup(), memory
720 * allocation is not necessary here, so it cannot fail.
721 */
722 vma_iter_bulk_store(&vmi, tmp);
723
724 mm->map_count++;
725
726 if (tmp->vm_ops && tmp->vm_ops->open)
727 tmp->vm_ops->open(tmp);
728
729 file = tmp->vm_file;
730 if (file) {
731 struct address_space *mapping = file->f_mapping;
732
733 get_file(file);
734 i_mmap_lock_write(mapping);
735 if (vma_is_shared_maywrite(tmp))
736 mapping_allow_writable(mapping);
737 flush_dcache_mmap_lock(mapping);
738 /* insert tmp into the share list, just after mpnt */
739 vma_interval_tree_insert_after(tmp, mpnt,
740 &mapping->i_mmap);
741 flush_dcache_mmap_unlock(mapping);
742 i_mmap_unlock_write(mapping);
743 }
744
745 if (!(tmp->vm_flags & VM_WIPEONFORK))
746 retval = copy_page_range(tmp, mpnt);
747
748 if (retval) {
749 mpnt = vma_next(&vmi);
750 goto loop_out;
751 }
752 }
753 /* a new mm has just been created */
754 retval = arch_dup_mmap(oldmm, mm);
755loop_out:
756 vma_iter_free(&vmi);
757 if (!retval) {
758 mt_set_in_rcu(vmi.mas.tree);
759 ksm_fork(mm, oldmm);
760 khugepaged_fork(mm, oldmm);
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 if (!retval)
777 dup_userfaultfd_complete(&uf);
778 else
779 dup_userfaultfd_fail(&uf);
780fail_uprobe_end:
781 uprobe_end_dup_mmap();
782 return retval;
783
784fail_nomem_anon_vma_fork:
785 mpol_put(vma_policy(tmp));
786fail_nomem_policy:
787 vm_area_free(tmp);
788fail_nomem:
789 retval = -ENOMEM;
790 vm_unacct_memory(charge);
791 goto loop_out;
792}
793
794static inline int mm_alloc_pgd(struct mm_struct *mm)
795{
796 mm->pgd = pgd_alloc(mm);
797 if (unlikely(!mm->pgd))
798 return -ENOMEM;
799 return 0;
800}
801
802static inline void mm_free_pgd(struct mm_struct *mm)
803{
804 pgd_free(mm, mm->pgd);
805}
806#else
807static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
808{
809 mmap_write_lock(oldmm);
810 dup_mm_exe_file(mm, oldmm);
811 mmap_write_unlock(oldmm);
812 return 0;
813}
814#define mm_alloc_pgd(mm) (0)
815#define mm_free_pgd(mm)
816#endif /* CONFIG_MMU */
817
818static void check_mm(struct mm_struct *mm)
819{
820 int i;
821
822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
823 "Please make sure 'struct resident_page_types[]' is updated as well");
824
825 for (i = 0; i < NR_MM_COUNTERS; i++) {
826 long x = percpu_counter_sum(&mm->rss_stat[i]);
827
828 if (unlikely(x))
829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
830 mm, resident_page_types[i], x);
831 }
832
833 if (mm_pgtables_bytes(mm))
834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
835 mm_pgtables_bytes(mm));
836
837#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
839#endif
840}
841
842#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
843#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
844
845static void do_check_lazy_tlb(void *arg)
846{
847 struct mm_struct *mm = arg;
848
849 WARN_ON_ONCE(current->active_mm == mm);
850}
851
852static void do_shoot_lazy_tlb(void *arg)
853{
854 struct mm_struct *mm = arg;
855
856 if (current->active_mm == mm) {
857 WARN_ON_ONCE(current->mm);
858 current->active_mm = &init_mm;
859 switch_mm(mm, &init_mm, current);
860 }
861}
862
863static void cleanup_lazy_tlbs(struct mm_struct *mm)
864{
865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
866 /*
867 * In this case, lazy tlb mms are refounted and would not reach
868 * __mmdrop until all CPUs have switched away and mmdrop()ed.
869 */
870 return;
871 }
872
873 /*
874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
875 * requires lazy mm users to switch to another mm when the refcount
876 * drops to zero, before the mm is freed. This requires IPIs here to
877 * switch kernel threads to init_mm.
878 *
879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
880 * switch with the final userspace teardown TLB flush which leaves the
881 * mm lazy on this CPU but no others, reducing the need for additional
882 * IPIs here. There are cases where a final IPI is still required here,
883 * such as the final mmdrop being performed on a different CPU than the
884 * one exiting, or kernel threads using the mm when userspace exits.
885 *
886 * IPI overheads have not found to be expensive, but they could be
887 * reduced in a number of possible ways, for example (roughly
888 * increasing order of complexity):
889 * - The last lazy reference created by exit_mm() could instead switch
890 * to init_mm, however it's probable this will run on the same CPU
891 * immediately afterwards, so this may not reduce IPIs much.
892 * - A batch of mms requiring IPIs could be gathered and freed at once.
893 * - CPUs store active_mm where it can be remotely checked without a
894 * lock, to filter out false-positives in the cpumask.
895 * - After mm_users or mm_count reaches zero, switching away from the
896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
897 * with some batching or delaying of the final IPIs.
898 * - A delayed freeing and RCU-like quiescing sequence based on mm
899 * switching to avoid IPIs completely.
900 */
901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
904}
905
906/*
907 * Called when the last reference to the mm
908 * is dropped: either by a lazy thread or by
909 * mmput. Free the page directory and the mm.
910 */
911void __mmdrop(struct mm_struct *mm)
912{
913 BUG_ON(mm == &init_mm);
914 WARN_ON_ONCE(mm == current->mm);
915
916 /* Ensure no CPUs are using this as their lazy tlb mm */
917 cleanup_lazy_tlbs(mm);
918
919 WARN_ON_ONCE(mm == current->active_mm);
920 mm_free_pgd(mm);
921 destroy_context(mm);
922 mmu_notifier_subscriptions_destroy(mm);
923 check_mm(mm);
924 put_user_ns(mm->user_ns);
925 mm_pasid_drop(mm);
926 mm_destroy_cid(mm);
927 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
928
929 free_mm(mm);
930}
931EXPORT_SYMBOL_GPL(__mmdrop);
932
933static void mmdrop_async_fn(struct work_struct *work)
934{
935 struct mm_struct *mm;
936
937 mm = container_of(work, struct mm_struct, async_put_work);
938 __mmdrop(mm);
939}
940
941static void mmdrop_async(struct mm_struct *mm)
942{
943 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
944 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
945 schedule_work(&mm->async_put_work);
946 }
947}
948
949static inline void free_signal_struct(struct signal_struct *sig)
950{
951 taskstats_tgid_free(sig);
952 sched_autogroup_exit(sig);
953 /*
954 * __mmdrop is not safe to call from softirq context on x86 due to
955 * pgd_dtor so postpone it to the async context
956 */
957 if (sig->oom_mm)
958 mmdrop_async(sig->oom_mm);
959 kmem_cache_free(signal_cachep, sig);
960}
961
962static inline void put_signal_struct(struct signal_struct *sig)
963{
964 if (refcount_dec_and_test(&sig->sigcnt))
965 free_signal_struct(sig);
966}
967
968void __put_task_struct(struct task_struct *tsk)
969{
970 WARN_ON(!tsk->exit_state);
971 WARN_ON(refcount_read(&tsk->usage));
972 WARN_ON(tsk == current);
973
974 sched_ext_free(tsk);
975 io_uring_free(tsk);
976 cgroup_free(tsk);
977 task_numa_free(tsk, true);
978 security_task_free(tsk);
979 exit_creds(tsk);
980 delayacct_tsk_free(tsk);
981 put_signal_struct(tsk->signal);
982 sched_core_free(tsk);
983 free_task(tsk);
984}
985EXPORT_SYMBOL_GPL(__put_task_struct);
986
987void __put_task_struct_rcu_cb(struct rcu_head *rhp)
988{
989 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
990
991 __put_task_struct(task);
992}
993EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
994
995void __init __weak arch_task_cache_init(void) { }
996
997/*
998 * set_max_threads
999 */
1000static void __init set_max_threads(unsigned int max_threads_suggested)
1001{
1002 u64 threads;
1003 unsigned long nr_pages = memblock_estimated_nr_free_pages();
1004
1005 /*
1006 * The number of threads shall be limited such that the thread
1007 * structures may only consume a small part of the available memory.
1008 */
1009 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1010 threads = MAX_THREADS;
1011 else
1012 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1013 (u64) THREAD_SIZE * 8UL);
1014
1015 if (threads > max_threads_suggested)
1016 threads = max_threads_suggested;
1017
1018 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1019}
1020
1021#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1022/* Initialized by the architecture: */
1023int arch_task_struct_size __read_mostly;
1024#endif
1025
1026static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
1027{
1028 /* Fetch thread_struct whitelist for the architecture. */
1029 arch_thread_struct_whitelist(offset, size);
1030
1031 /*
1032 * Handle zero-sized whitelist or empty thread_struct, otherwise
1033 * adjust offset to position of thread_struct in task_struct.
1034 */
1035 if (unlikely(*size == 0))
1036 *offset = 0;
1037 else
1038 *offset += offsetof(struct task_struct, thread);
1039}
1040
1041void __init fork_init(void)
1042{
1043 int i;
1044#ifndef ARCH_MIN_TASKALIGN
1045#define ARCH_MIN_TASKALIGN 0
1046#endif
1047 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1048 unsigned long useroffset, usersize;
1049
1050 /* create a slab on which task_structs can be allocated */
1051 task_struct_whitelist(&useroffset, &usersize);
1052 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1053 arch_task_struct_size, align,
1054 SLAB_PANIC|SLAB_ACCOUNT,
1055 useroffset, usersize, NULL);
1056
1057 /* do the arch specific task caches init */
1058 arch_task_cache_init();
1059
1060 set_max_threads(MAX_THREADS);
1061
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1064 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1065 init_task.signal->rlim[RLIMIT_NPROC];
1066
1067 for (i = 0; i < UCOUNT_COUNTS; i++)
1068 init_user_ns.ucount_max[i] = max_threads/2;
1069
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1073 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1074
1075#ifdef CONFIG_VMAP_STACK
1076 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1077 NULL, free_vm_stack_cache);
1078#endif
1079
1080 scs_init();
1081
1082 lockdep_init_task(&init_task);
1083 uprobes_init();
1084}
1085
1086int __weak arch_dup_task_struct(struct task_struct *dst,
1087 struct task_struct *src)
1088{
1089 *dst = *src;
1090 return 0;
1091}
1092
1093void set_task_stack_end_magic(struct task_struct *tsk)
1094{
1095 unsigned long *stackend;
1096
1097 stackend = end_of_stack(tsk);
1098 *stackend = STACK_END_MAGIC; /* for overflow detection */
1099}
1100
1101static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1102{
1103 struct task_struct *tsk;
1104 int err;
1105
1106 if (node == NUMA_NO_NODE)
1107 node = tsk_fork_get_node(orig);
1108 tsk = alloc_task_struct_node(node);
1109 if (!tsk)
1110 return NULL;
1111
1112 err = arch_dup_task_struct(tsk, orig);
1113 if (err)
1114 goto free_tsk;
1115
1116 err = alloc_thread_stack_node(tsk, node);
1117 if (err)
1118 goto free_tsk;
1119
1120#ifdef CONFIG_THREAD_INFO_IN_TASK
1121 refcount_set(&tsk->stack_refcount, 1);
1122#endif
1123 account_kernel_stack(tsk, 1);
1124
1125 err = scs_prepare(tsk, node);
1126 if (err)
1127 goto free_stack;
1128
1129#ifdef CONFIG_SECCOMP
1130 /*
1131 * We must handle setting up seccomp filters once we're under
1132 * the sighand lock in case orig has changed between now and
1133 * then. Until then, filter must be NULL to avoid messing up
1134 * the usage counts on the error path calling free_task.
1135 */
1136 tsk->seccomp.filter = NULL;
1137#endif
1138
1139 setup_thread_stack(tsk, orig);
1140 clear_user_return_notifier(tsk);
1141 clear_tsk_need_resched(tsk);
1142 set_task_stack_end_magic(tsk);
1143 clear_syscall_work_syscall_user_dispatch(tsk);
1144
1145#ifdef CONFIG_STACKPROTECTOR
1146 tsk->stack_canary = get_random_canary();
1147#endif
1148 if (orig->cpus_ptr == &orig->cpus_mask)
1149 tsk->cpus_ptr = &tsk->cpus_mask;
1150 dup_user_cpus_ptr(tsk, orig, node);
1151
1152 /*
1153 * One for the user space visible state that goes away when reaped.
1154 * One for the scheduler.
1155 */
1156 refcount_set(&tsk->rcu_users, 2);
1157 /* One for the rcu users */
1158 refcount_set(&tsk->usage, 1);
1159#ifdef CONFIG_BLK_DEV_IO_TRACE
1160 tsk->btrace_seq = 0;
1161#endif
1162 tsk->splice_pipe = NULL;
1163 tsk->task_frag.page = NULL;
1164 tsk->wake_q.next = NULL;
1165 tsk->worker_private = NULL;
1166
1167 kcov_task_init(tsk);
1168 kmsan_task_create(tsk);
1169 kmap_local_fork(tsk);
1170
1171#ifdef CONFIG_FAULT_INJECTION
1172 tsk->fail_nth = 0;
1173#endif
1174
1175#ifdef CONFIG_BLK_CGROUP
1176 tsk->throttle_disk = NULL;
1177 tsk->use_memdelay = 0;
1178#endif
1179
1180#ifdef CONFIG_ARCH_HAS_CPU_PASID
1181 tsk->pasid_activated = 0;
1182#endif
1183
1184#ifdef CONFIG_MEMCG
1185 tsk->active_memcg = NULL;
1186#endif
1187
1188#ifdef CONFIG_CPU_SUP_INTEL
1189 tsk->reported_split_lock = 0;
1190#endif
1191
1192#ifdef CONFIG_SCHED_MM_CID
1193 tsk->mm_cid = -1;
1194 tsk->last_mm_cid = -1;
1195 tsk->mm_cid_active = 0;
1196 tsk->migrate_from_cpu = -1;
1197#endif
1198 return tsk;
1199
1200free_stack:
1201 exit_task_stack_account(tsk);
1202 free_thread_stack(tsk);
1203free_tsk:
1204 free_task_struct(tsk);
1205 return NULL;
1206}
1207
1208__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1209
1210static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1211
1212static int __init coredump_filter_setup(char *s)
1213{
1214 default_dump_filter =
1215 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1216 MMF_DUMP_FILTER_MASK;
1217 return 1;
1218}
1219
1220__setup("coredump_filter=", coredump_filter_setup);
1221
1222#include <linux/init_task.h>
1223
1224static void mm_init_aio(struct mm_struct *mm)
1225{
1226#ifdef CONFIG_AIO
1227 spin_lock_init(&mm->ioctx_lock);
1228 mm->ioctx_table = NULL;
1229#endif
1230}
1231
1232static __always_inline void mm_clear_owner(struct mm_struct *mm,
1233 struct task_struct *p)
1234{
1235#ifdef CONFIG_MEMCG
1236 if (mm->owner == p)
1237 WRITE_ONCE(mm->owner, NULL);
1238#endif
1239}
1240
1241static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1242{
1243#ifdef CONFIG_MEMCG
1244 mm->owner = p;
1245#endif
1246}
1247
1248static void mm_init_uprobes_state(struct mm_struct *mm)
1249{
1250#ifdef CONFIG_UPROBES
1251 mm->uprobes_state.xol_area = NULL;
1252#endif
1253}
1254
1255static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1256 struct user_namespace *user_ns)
1257{
1258 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1259 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1260 atomic_set(&mm->mm_users, 1);
1261 atomic_set(&mm->mm_count, 1);
1262 seqcount_init(&mm->write_protect_seq);
1263 mmap_init_lock(mm);
1264 INIT_LIST_HEAD(&mm->mmlist);
1265#ifdef CONFIG_PER_VMA_LOCK
1266 mm->mm_lock_seq = 0;
1267#endif
1268 mm_pgtables_bytes_init(mm);
1269 mm->map_count = 0;
1270 mm->locked_vm = 0;
1271 atomic64_set(&mm->pinned_vm, 0);
1272 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1273 spin_lock_init(&mm->page_table_lock);
1274 spin_lock_init(&mm->arg_lock);
1275 mm_init_cpumask(mm);
1276 mm_init_aio(mm);
1277 mm_init_owner(mm, p);
1278 mm_pasid_init(mm);
1279 RCU_INIT_POINTER(mm->exe_file, NULL);
1280 mmu_notifier_subscriptions_init(mm);
1281 init_tlb_flush_pending(mm);
1282#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1283 mm->pmd_huge_pte = NULL;
1284#endif
1285 mm_init_uprobes_state(mm);
1286 hugetlb_count_init(mm);
1287
1288 if (current->mm) {
1289 mm->flags = mmf_init_flags(current->mm->flags);
1290 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1291 } else {
1292 mm->flags = default_dump_filter;
1293 mm->def_flags = 0;
1294 }
1295
1296 if (mm_alloc_pgd(mm))
1297 goto fail_nopgd;
1298
1299 if (init_new_context(p, mm))
1300 goto fail_nocontext;
1301
1302 if (mm_alloc_cid(mm))
1303 goto fail_cid;
1304
1305 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1306 NR_MM_COUNTERS))
1307 goto fail_pcpu;
1308
1309 mm->user_ns = get_user_ns(user_ns);
1310 lru_gen_init_mm(mm);
1311 return mm;
1312
1313fail_pcpu:
1314 mm_destroy_cid(mm);
1315fail_cid:
1316 destroy_context(mm);
1317fail_nocontext:
1318 mm_free_pgd(mm);
1319fail_nopgd:
1320 free_mm(mm);
1321 return NULL;
1322}
1323
1324/*
1325 * Allocate and initialize an mm_struct.
1326 */
1327struct mm_struct *mm_alloc(void)
1328{
1329 struct mm_struct *mm;
1330
1331 mm = allocate_mm();
1332 if (!mm)
1333 return NULL;
1334
1335 memset(mm, 0, sizeof(*mm));
1336 return mm_init(mm, current, current_user_ns());
1337}
1338EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1339
1340static inline void __mmput(struct mm_struct *mm)
1341{
1342 VM_BUG_ON(atomic_read(&mm->mm_users));
1343
1344 uprobe_clear_state(mm);
1345 exit_aio(mm);
1346 ksm_exit(mm);
1347 khugepaged_exit(mm); /* must run before exit_mmap */
1348 exit_mmap(mm);
1349 mm_put_huge_zero_folio(mm);
1350 set_mm_exe_file(mm, NULL);
1351 if (!list_empty(&mm->mmlist)) {
1352 spin_lock(&mmlist_lock);
1353 list_del(&mm->mmlist);
1354 spin_unlock(&mmlist_lock);
1355 }
1356 if (mm->binfmt)
1357 module_put(mm->binfmt->module);
1358 lru_gen_del_mm(mm);
1359 mmdrop(mm);
1360}
1361
1362/*
1363 * Decrement the use count and release all resources for an mm.
1364 */
1365void mmput(struct mm_struct *mm)
1366{
1367 might_sleep();
1368
1369 if (atomic_dec_and_test(&mm->mm_users))
1370 __mmput(mm);
1371}
1372EXPORT_SYMBOL_GPL(mmput);
1373
1374#ifdef CONFIG_MMU
1375static void mmput_async_fn(struct work_struct *work)
1376{
1377 struct mm_struct *mm = container_of(work, struct mm_struct,
1378 async_put_work);
1379
1380 __mmput(mm);
1381}
1382
1383void mmput_async(struct mm_struct *mm)
1384{
1385 if (atomic_dec_and_test(&mm->mm_users)) {
1386 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1387 schedule_work(&mm->async_put_work);
1388 }
1389}
1390EXPORT_SYMBOL_GPL(mmput_async);
1391#endif
1392
1393/**
1394 * set_mm_exe_file - change a reference to the mm's executable file
1395 * @mm: The mm to change.
1396 * @new_exe_file: The new file to use.
1397 *
1398 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1399 *
1400 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1401 * invocations: in mmput() nobody alive left, in execve it happens before
1402 * the new mm is made visible to anyone.
1403 *
1404 * Can only fail if new_exe_file != NULL.
1405 */
1406int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1407{
1408 struct file *old_exe_file;
1409
1410 /*
1411 * It is safe to dereference the exe_file without RCU as
1412 * this function is only called if nobody else can access
1413 * this mm -- see comment above for justification.
1414 */
1415 old_exe_file = rcu_dereference_raw(mm->exe_file);
1416
1417 if (new_exe_file)
1418 get_file(new_exe_file);
1419 rcu_assign_pointer(mm->exe_file, new_exe_file);
1420 if (old_exe_file)
1421 fput(old_exe_file);
1422 return 0;
1423}
1424
1425/**
1426 * replace_mm_exe_file - replace a reference to the mm's executable file
1427 * @mm: The mm to change.
1428 * @new_exe_file: The new file to use.
1429 *
1430 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1431 *
1432 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1433 */
1434int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1435{
1436 struct vm_area_struct *vma;
1437 struct file *old_exe_file;
1438 int ret = 0;
1439
1440 /* Forbid mm->exe_file change if old file still mapped. */
1441 old_exe_file = get_mm_exe_file(mm);
1442 if (old_exe_file) {
1443 VMA_ITERATOR(vmi, mm, 0);
1444 mmap_read_lock(mm);
1445 for_each_vma(vmi, vma) {
1446 if (!vma->vm_file)
1447 continue;
1448 if (path_equal(&vma->vm_file->f_path,
1449 &old_exe_file->f_path)) {
1450 ret = -EBUSY;
1451 break;
1452 }
1453 }
1454 mmap_read_unlock(mm);
1455 fput(old_exe_file);
1456 if (ret)
1457 return ret;
1458 }
1459
1460 get_file(new_exe_file);
1461
1462 /* set the new file */
1463 mmap_write_lock(mm);
1464 old_exe_file = rcu_dereference_raw(mm->exe_file);
1465 rcu_assign_pointer(mm->exe_file, new_exe_file);
1466 mmap_write_unlock(mm);
1467
1468 if (old_exe_file)
1469 fput(old_exe_file);
1470 return 0;
1471}
1472
1473/**
1474 * get_mm_exe_file - acquire a reference to the mm's executable file
1475 * @mm: The mm of interest.
1476 *
1477 * Returns %NULL if mm has no associated executable file.
1478 * User must release file via fput().
1479 */
1480struct file *get_mm_exe_file(struct mm_struct *mm)
1481{
1482 struct file *exe_file;
1483
1484 rcu_read_lock();
1485 exe_file = get_file_rcu(&mm->exe_file);
1486 rcu_read_unlock();
1487 return exe_file;
1488}
1489
1490/**
1491 * get_task_exe_file - acquire a reference to the task's executable file
1492 * @task: The task.
1493 *
1494 * Returns %NULL if task's mm (if any) has no associated executable file or
1495 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1496 * User must release file via fput().
1497 */
1498struct file *get_task_exe_file(struct task_struct *task)
1499{
1500 struct file *exe_file = NULL;
1501 struct mm_struct *mm;
1502
1503 task_lock(task);
1504 mm = task->mm;
1505 if (mm) {
1506 if (!(task->flags & PF_KTHREAD))
1507 exe_file = get_mm_exe_file(mm);
1508 }
1509 task_unlock(task);
1510 return exe_file;
1511}
1512
1513/**
1514 * get_task_mm - acquire a reference to the task's mm
1515 * @task: The task.
1516 *
1517 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1518 * this kernel workthread has transiently adopted a user mm with use_mm,
1519 * to do its AIO) is not set and if so returns a reference to it, after
1520 * bumping up the use count. User must release the mm via mmput()
1521 * after use. Typically used by /proc and ptrace.
1522 */
1523struct mm_struct *get_task_mm(struct task_struct *task)
1524{
1525 struct mm_struct *mm;
1526
1527 if (task->flags & PF_KTHREAD)
1528 return NULL;
1529
1530 task_lock(task);
1531 mm = task->mm;
1532 if (mm)
1533 mmget(mm);
1534 task_unlock(task);
1535 return mm;
1536}
1537EXPORT_SYMBOL_GPL(get_task_mm);
1538
1539struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1540{
1541 struct mm_struct *mm;
1542 int err;
1543
1544 err = down_read_killable(&task->signal->exec_update_lock);
1545 if (err)
1546 return ERR_PTR(err);
1547
1548 mm = get_task_mm(task);
1549 if (mm && mm != current->mm &&
1550 !ptrace_may_access(task, mode)) {
1551 mmput(mm);
1552 mm = ERR_PTR(-EACCES);
1553 }
1554 up_read(&task->signal->exec_update_lock);
1555
1556 return mm;
1557}
1558
1559static void complete_vfork_done(struct task_struct *tsk)
1560{
1561 struct completion *vfork;
1562
1563 task_lock(tsk);
1564 vfork = tsk->vfork_done;
1565 if (likely(vfork)) {
1566 tsk->vfork_done = NULL;
1567 complete(vfork);
1568 }
1569 task_unlock(tsk);
1570}
1571
1572static int wait_for_vfork_done(struct task_struct *child,
1573 struct completion *vfork)
1574{
1575 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1576 int killed;
1577
1578 cgroup_enter_frozen();
1579 killed = wait_for_completion_state(vfork, state);
1580 cgroup_leave_frozen(false);
1581
1582 if (killed) {
1583 task_lock(child);
1584 child->vfork_done = NULL;
1585 task_unlock(child);
1586 }
1587
1588 put_task_struct(child);
1589 return killed;
1590}
1591
1592/* Please note the differences between mmput and mm_release.
1593 * mmput is called whenever we stop holding onto a mm_struct,
1594 * error success whatever.
1595 *
1596 * mm_release is called after a mm_struct has been removed
1597 * from the current process.
1598 *
1599 * This difference is important for error handling, when we
1600 * only half set up a mm_struct for a new process and need to restore
1601 * the old one. Because we mmput the new mm_struct before
1602 * restoring the old one. . .
1603 * Eric Biederman 10 January 1998
1604 */
1605static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1606{
1607 uprobe_free_utask(tsk);
1608
1609 /* Get rid of any cached register state */
1610 deactivate_mm(tsk, mm);
1611
1612 /*
1613 * Signal userspace if we're not exiting with a core dump
1614 * because we want to leave the value intact for debugging
1615 * purposes.
1616 */
1617 if (tsk->clear_child_tid) {
1618 if (atomic_read(&mm->mm_users) > 1) {
1619 /*
1620 * We don't check the error code - if userspace has
1621 * not set up a proper pointer then tough luck.
1622 */
1623 put_user(0, tsk->clear_child_tid);
1624 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1625 1, NULL, NULL, 0, 0);
1626 }
1627 tsk->clear_child_tid = NULL;
1628 }
1629
1630 /*
1631 * All done, finally we can wake up parent and return this mm to him.
1632 * Also kthread_stop() uses this completion for synchronization.
1633 */
1634 if (tsk->vfork_done)
1635 complete_vfork_done(tsk);
1636}
1637
1638void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1639{
1640 futex_exit_release(tsk);
1641 mm_release(tsk, mm);
1642}
1643
1644void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1645{
1646 futex_exec_release(tsk);
1647 mm_release(tsk, mm);
1648}
1649
1650/**
1651 * dup_mm() - duplicates an existing mm structure
1652 * @tsk: the task_struct with which the new mm will be associated.
1653 * @oldmm: the mm to duplicate.
1654 *
1655 * Allocates a new mm structure and duplicates the provided @oldmm structure
1656 * content into it.
1657 *
1658 * Return: the duplicated mm or NULL on failure.
1659 */
1660static struct mm_struct *dup_mm(struct task_struct *tsk,
1661 struct mm_struct *oldmm)
1662{
1663 struct mm_struct *mm;
1664 int err;
1665
1666 mm = allocate_mm();
1667 if (!mm)
1668 goto fail_nomem;
1669
1670 memcpy(mm, oldmm, sizeof(*mm));
1671
1672 if (!mm_init(mm, tsk, mm->user_ns))
1673 goto fail_nomem;
1674
1675 err = dup_mmap(mm, oldmm);
1676 if (err)
1677 goto free_pt;
1678
1679 mm->hiwater_rss = get_mm_rss(mm);
1680 mm->hiwater_vm = mm->total_vm;
1681
1682 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1683 goto free_pt;
1684
1685 return mm;
1686
1687free_pt:
1688 /* don't put binfmt in mmput, we haven't got module yet */
1689 mm->binfmt = NULL;
1690 mm_init_owner(mm, NULL);
1691 mmput(mm);
1692
1693fail_nomem:
1694 return NULL;
1695}
1696
1697static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1698{
1699 struct mm_struct *mm, *oldmm;
1700
1701 tsk->min_flt = tsk->maj_flt = 0;
1702 tsk->nvcsw = tsk->nivcsw = 0;
1703#ifdef CONFIG_DETECT_HUNG_TASK
1704 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1705 tsk->last_switch_time = 0;
1706#endif
1707
1708 tsk->mm = NULL;
1709 tsk->active_mm = NULL;
1710
1711 /*
1712 * Are we cloning a kernel thread?
1713 *
1714 * We need to steal a active VM for that..
1715 */
1716 oldmm = current->mm;
1717 if (!oldmm)
1718 return 0;
1719
1720 if (clone_flags & CLONE_VM) {
1721 mmget(oldmm);
1722 mm = oldmm;
1723 } else {
1724 mm = dup_mm(tsk, current->mm);
1725 if (!mm)
1726 return -ENOMEM;
1727 }
1728
1729 tsk->mm = mm;
1730 tsk->active_mm = mm;
1731 sched_mm_cid_fork(tsk);
1732 return 0;
1733}
1734
1735static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1736{
1737 struct fs_struct *fs = current->fs;
1738 if (clone_flags & CLONE_FS) {
1739 /* tsk->fs is already what we want */
1740 spin_lock(&fs->lock);
1741 /* "users" and "in_exec" locked for check_unsafe_exec() */
1742 if (fs->in_exec) {
1743 spin_unlock(&fs->lock);
1744 return -EAGAIN;
1745 }
1746 fs->users++;
1747 spin_unlock(&fs->lock);
1748 return 0;
1749 }
1750 tsk->fs = copy_fs_struct(fs);
1751 if (!tsk->fs)
1752 return -ENOMEM;
1753 return 0;
1754}
1755
1756static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1757 int no_files)
1758{
1759 struct files_struct *oldf, *newf;
1760
1761 /*
1762 * A background process may not have any files ...
1763 */
1764 oldf = current->files;
1765 if (!oldf)
1766 return 0;
1767
1768 if (no_files) {
1769 tsk->files = NULL;
1770 return 0;
1771 }
1772
1773 if (clone_flags & CLONE_FILES) {
1774 atomic_inc(&oldf->count);
1775 return 0;
1776 }
1777
1778 newf = dup_fd(oldf, NULL);
1779 if (IS_ERR(newf))
1780 return PTR_ERR(newf);
1781
1782 tsk->files = newf;
1783 return 0;
1784}
1785
1786static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1787{
1788 struct sighand_struct *sig;
1789
1790 if (clone_flags & CLONE_SIGHAND) {
1791 refcount_inc(&current->sighand->count);
1792 return 0;
1793 }
1794 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1795 RCU_INIT_POINTER(tsk->sighand, sig);
1796 if (!sig)
1797 return -ENOMEM;
1798
1799 refcount_set(&sig->count, 1);
1800 spin_lock_irq(&current->sighand->siglock);
1801 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1802 spin_unlock_irq(&current->sighand->siglock);
1803
1804 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1805 if (clone_flags & CLONE_CLEAR_SIGHAND)
1806 flush_signal_handlers(tsk, 0);
1807
1808 return 0;
1809}
1810
1811void __cleanup_sighand(struct sighand_struct *sighand)
1812{
1813 if (refcount_dec_and_test(&sighand->count)) {
1814 signalfd_cleanup(sighand);
1815 /*
1816 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1817 * without an RCU grace period, see __lock_task_sighand().
1818 */
1819 kmem_cache_free(sighand_cachep, sighand);
1820 }
1821}
1822
1823/*
1824 * Initialize POSIX timer handling for a thread group.
1825 */
1826static void posix_cpu_timers_init_group(struct signal_struct *sig)
1827{
1828 struct posix_cputimers *pct = &sig->posix_cputimers;
1829 unsigned long cpu_limit;
1830
1831 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1832 posix_cputimers_group_init(pct, cpu_limit);
1833}
1834
1835static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1836{
1837 struct signal_struct *sig;
1838
1839 if (clone_flags & CLONE_THREAD)
1840 return 0;
1841
1842 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1843 tsk->signal = sig;
1844 if (!sig)
1845 return -ENOMEM;
1846
1847 sig->nr_threads = 1;
1848 sig->quick_threads = 1;
1849 atomic_set(&sig->live, 1);
1850 refcount_set(&sig->sigcnt, 1);
1851
1852 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1853 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1854 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1855
1856 init_waitqueue_head(&sig->wait_chldexit);
1857 sig->curr_target = tsk;
1858 init_sigpending(&sig->shared_pending);
1859 INIT_HLIST_HEAD(&sig->multiprocess);
1860 seqlock_init(&sig->stats_lock);
1861 prev_cputime_init(&sig->prev_cputime);
1862
1863#ifdef CONFIG_POSIX_TIMERS
1864 INIT_HLIST_HEAD(&sig->posix_timers);
1865 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1866 sig->real_timer.function = it_real_fn;
1867#endif
1868
1869 task_lock(current->group_leader);
1870 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1871 task_unlock(current->group_leader);
1872
1873 posix_cpu_timers_init_group(sig);
1874
1875 tty_audit_fork(sig);
1876 sched_autogroup_fork(sig);
1877
1878 sig->oom_score_adj = current->signal->oom_score_adj;
1879 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1880
1881 mutex_init(&sig->cred_guard_mutex);
1882 init_rwsem(&sig->exec_update_lock);
1883
1884 return 0;
1885}
1886
1887static void copy_seccomp(struct task_struct *p)
1888{
1889#ifdef CONFIG_SECCOMP
1890 /*
1891 * Must be called with sighand->lock held, which is common to
1892 * all threads in the group. Holding cred_guard_mutex is not
1893 * needed because this new task is not yet running and cannot
1894 * be racing exec.
1895 */
1896 assert_spin_locked(&current->sighand->siglock);
1897
1898 /* Ref-count the new filter user, and assign it. */
1899 get_seccomp_filter(current);
1900 p->seccomp = current->seccomp;
1901
1902 /*
1903 * Explicitly enable no_new_privs here in case it got set
1904 * between the task_struct being duplicated and holding the
1905 * sighand lock. The seccomp state and nnp must be in sync.
1906 */
1907 if (task_no_new_privs(current))
1908 task_set_no_new_privs(p);
1909
1910 /*
1911 * If the parent gained a seccomp mode after copying thread
1912 * flags and between before we held the sighand lock, we have
1913 * to manually enable the seccomp thread flag here.
1914 */
1915 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1916 set_task_syscall_work(p, SECCOMP);
1917#endif
1918}
1919
1920SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1921{
1922 current->clear_child_tid = tidptr;
1923
1924 return task_pid_vnr(current);
1925}
1926
1927static void rt_mutex_init_task(struct task_struct *p)
1928{
1929 raw_spin_lock_init(&p->pi_lock);
1930#ifdef CONFIG_RT_MUTEXES
1931 p->pi_waiters = RB_ROOT_CACHED;
1932 p->pi_top_task = NULL;
1933 p->pi_blocked_on = NULL;
1934#endif
1935}
1936
1937static inline void init_task_pid_links(struct task_struct *task)
1938{
1939 enum pid_type type;
1940
1941 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1942 INIT_HLIST_NODE(&task->pid_links[type]);
1943}
1944
1945static inline void
1946init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1947{
1948 if (type == PIDTYPE_PID)
1949 task->thread_pid = pid;
1950 else
1951 task->signal->pids[type] = pid;
1952}
1953
1954static inline void rcu_copy_process(struct task_struct *p)
1955{
1956#ifdef CONFIG_PREEMPT_RCU
1957 p->rcu_read_lock_nesting = 0;
1958 p->rcu_read_unlock_special.s = 0;
1959 p->rcu_blocked_node = NULL;
1960 INIT_LIST_HEAD(&p->rcu_node_entry);
1961#endif /* #ifdef CONFIG_PREEMPT_RCU */
1962#ifdef CONFIG_TASKS_RCU
1963 p->rcu_tasks_holdout = false;
1964 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1965 p->rcu_tasks_idle_cpu = -1;
1966 INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1967#endif /* #ifdef CONFIG_TASKS_RCU */
1968#ifdef CONFIG_TASKS_TRACE_RCU
1969 p->trc_reader_nesting = 0;
1970 p->trc_reader_special.s = 0;
1971 INIT_LIST_HEAD(&p->trc_holdout_list);
1972 INIT_LIST_HEAD(&p->trc_blkd_node);
1973#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1974}
1975
1976/**
1977 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1978 * @pid: the struct pid for which to create a pidfd
1979 * @flags: flags of the new @pidfd
1980 * @ret: Where to return the file for the pidfd.
1981 *
1982 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1983 * caller's file descriptor table. The pidfd is reserved but not installed yet.
1984 *
1985 * The helper doesn't perform checks on @pid which makes it useful for pidfds
1986 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
1987 * pidfd file are prepared.
1988 *
1989 * If this function returns successfully the caller is responsible to either
1990 * call fd_install() passing the returned pidfd and pidfd file as arguments in
1991 * order to install the pidfd into its file descriptor table or they must use
1992 * put_unused_fd() and fput() on the returned pidfd and pidfd file
1993 * respectively.
1994 *
1995 * This function is useful when a pidfd must already be reserved but there
1996 * might still be points of failure afterwards and the caller wants to ensure
1997 * that no pidfd is leaked into its file descriptor table.
1998 *
1999 * Return: On success, a reserved pidfd is returned from the function and a new
2000 * pidfd file is returned in the last argument to the function. On
2001 * error, a negative error code is returned from the function and the
2002 * last argument remains unchanged.
2003 */
2004static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2005{
2006 int pidfd;
2007 struct file *pidfd_file;
2008
2009 pidfd = get_unused_fd_flags(O_CLOEXEC);
2010 if (pidfd < 0)
2011 return pidfd;
2012
2013 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2014 if (IS_ERR(pidfd_file)) {
2015 put_unused_fd(pidfd);
2016 return PTR_ERR(pidfd_file);
2017 }
2018 /*
2019 * anon_inode_getfile() ignores everything outside of the
2020 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2021 */
2022 pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2023 *ret = pidfd_file;
2024 return pidfd;
2025}
2026
2027/**
2028 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2029 * @pid: the struct pid for which to create a pidfd
2030 * @flags: flags of the new @pidfd
2031 * @ret: Where to return the pidfd.
2032 *
2033 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2034 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2035 *
2036 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2037 * task identified by @pid must be a thread-group leader.
2038 *
2039 * If this function returns successfully the caller is responsible to either
2040 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2041 * order to install the pidfd into its file descriptor table or they must use
2042 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2043 * respectively.
2044 *
2045 * This function is useful when a pidfd must already be reserved but there
2046 * might still be points of failure afterwards and the caller wants to ensure
2047 * that no pidfd is leaked into its file descriptor table.
2048 *
2049 * Return: On success, a reserved pidfd is returned from the function and a new
2050 * pidfd file is returned in the last argument to the function. On
2051 * error, a negative error code is returned from the function and the
2052 * last argument remains unchanged.
2053 */
2054int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2055{
2056 bool thread = flags & PIDFD_THREAD;
2057
2058 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2059 return -EINVAL;
2060
2061 return __pidfd_prepare(pid, flags, ret);
2062}
2063
2064static void __delayed_free_task(struct rcu_head *rhp)
2065{
2066 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2067
2068 free_task(tsk);
2069}
2070
2071static __always_inline void delayed_free_task(struct task_struct *tsk)
2072{
2073 if (IS_ENABLED(CONFIG_MEMCG))
2074 call_rcu(&tsk->rcu, __delayed_free_task);
2075 else
2076 free_task(tsk);
2077}
2078
2079static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2080{
2081 /* Skip if kernel thread */
2082 if (!tsk->mm)
2083 return;
2084
2085 /* Skip if spawning a thread or using vfork */
2086 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2087 return;
2088
2089 /* We need to synchronize with __set_oom_adj */
2090 mutex_lock(&oom_adj_mutex);
2091 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2092 /* Update the values in case they were changed after copy_signal */
2093 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2094 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2095 mutex_unlock(&oom_adj_mutex);
2096}
2097
2098#ifdef CONFIG_RV
2099static void rv_task_fork(struct task_struct *p)
2100{
2101 int i;
2102
2103 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2104 p->rv[i].da_mon.monitoring = false;
2105}
2106#else
2107#define rv_task_fork(p) do {} while (0)
2108#endif
2109
2110/*
2111 * This creates a new process as a copy of the old one,
2112 * but does not actually start it yet.
2113 *
2114 * It copies the registers, and all the appropriate
2115 * parts of the process environment (as per the clone
2116 * flags). The actual kick-off is left to the caller.
2117 */
2118__latent_entropy struct task_struct *copy_process(
2119 struct pid *pid,
2120 int trace,
2121 int node,
2122 struct kernel_clone_args *args)
2123{
2124 int pidfd = -1, retval;
2125 struct task_struct *p;
2126 struct multiprocess_signals delayed;
2127 struct file *pidfile = NULL;
2128 const u64 clone_flags = args->flags;
2129 struct nsproxy *nsp = current->nsproxy;
2130
2131 /*
2132 * Don't allow sharing the root directory with processes in a different
2133 * namespace
2134 */
2135 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2136 return ERR_PTR(-EINVAL);
2137
2138 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2139 return ERR_PTR(-EINVAL);
2140
2141 /*
2142 * Thread groups must share signals as well, and detached threads
2143 * can only be started up within the thread group.
2144 */
2145 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2146 return ERR_PTR(-EINVAL);
2147
2148 /*
2149 * Shared signal handlers imply shared VM. By way of the above,
2150 * thread groups also imply shared VM. Blocking this case allows
2151 * for various simplifications in other code.
2152 */
2153 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2154 return ERR_PTR(-EINVAL);
2155
2156 /*
2157 * Siblings of global init remain as zombies on exit since they are
2158 * not reaped by their parent (swapper). To solve this and to avoid
2159 * multi-rooted process trees, prevent global and container-inits
2160 * from creating siblings.
2161 */
2162 if ((clone_flags & CLONE_PARENT) &&
2163 current->signal->flags & SIGNAL_UNKILLABLE)
2164 return ERR_PTR(-EINVAL);
2165
2166 /*
2167 * If the new process will be in a different pid or user namespace
2168 * do not allow it to share a thread group with the forking task.
2169 */
2170 if (clone_flags & CLONE_THREAD) {
2171 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2172 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2173 return ERR_PTR(-EINVAL);
2174 }
2175
2176 if (clone_flags & CLONE_PIDFD) {
2177 /*
2178 * - CLONE_DETACHED is blocked so that we can potentially
2179 * reuse it later for CLONE_PIDFD.
2180 */
2181 if (clone_flags & CLONE_DETACHED)
2182 return ERR_PTR(-EINVAL);
2183 }
2184
2185 /*
2186 * Force any signals received before this point to be delivered
2187 * before the fork happens. Collect up signals sent to multiple
2188 * processes that happen during the fork and delay them so that
2189 * they appear to happen after the fork.
2190 */
2191 sigemptyset(&delayed.signal);
2192 INIT_HLIST_NODE(&delayed.node);
2193
2194 spin_lock_irq(&current->sighand->siglock);
2195 if (!(clone_flags & CLONE_THREAD))
2196 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2197 recalc_sigpending();
2198 spin_unlock_irq(&current->sighand->siglock);
2199 retval = -ERESTARTNOINTR;
2200 if (task_sigpending(current))
2201 goto fork_out;
2202
2203 retval = -ENOMEM;
2204 p = dup_task_struct(current, node);
2205 if (!p)
2206 goto fork_out;
2207 p->flags &= ~PF_KTHREAD;
2208 if (args->kthread)
2209 p->flags |= PF_KTHREAD;
2210 if (args->user_worker) {
2211 /*
2212 * Mark us a user worker, and block any signal that isn't
2213 * fatal or STOP
2214 */
2215 p->flags |= PF_USER_WORKER;
2216 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2217 }
2218 if (args->io_thread)
2219 p->flags |= PF_IO_WORKER;
2220
2221 if (args->name)
2222 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2223
2224 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2225 /*
2226 * Clear TID on mm_release()?
2227 */
2228 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2229
2230 ftrace_graph_init_task(p);
2231
2232 rt_mutex_init_task(p);
2233
2234 lockdep_assert_irqs_enabled();
2235#ifdef CONFIG_PROVE_LOCKING
2236 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2237#endif
2238 retval = copy_creds(p, clone_flags);
2239 if (retval < 0)
2240 goto bad_fork_free;
2241
2242 retval = -EAGAIN;
2243 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2244 if (p->real_cred->user != INIT_USER &&
2245 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2246 goto bad_fork_cleanup_count;
2247 }
2248 current->flags &= ~PF_NPROC_EXCEEDED;
2249
2250 /*
2251 * If multiple threads are within copy_process(), then this check
2252 * triggers too late. This doesn't hurt, the check is only there
2253 * to stop root fork bombs.
2254 */
2255 retval = -EAGAIN;
2256 if (data_race(nr_threads >= max_threads))
2257 goto bad_fork_cleanup_count;
2258
2259 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2260 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2261 p->flags |= PF_FORKNOEXEC;
2262 INIT_LIST_HEAD(&p->children);
2263 INIT_LIST_HEAD(&p->sibling);
2264 rcu_copy_process(p);
2265 p->vfork_done = NULL;
2266 spin_lock_init(&p->alloc_lock);
2267
2268 init_sigpending(&p->pending);
2269
2270 p->utime = p->stime = p->gtime = 0;
2271#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2272 p->utimescaled = p->stimescaled = 0;
2273#endif
2274 prev_cputime_init(&p->prev_cputime);
2275
2276#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2277 seqcount_init(&p->vtime.seqcount);
2278 p->vtime.starttime = 0;
2279 p->vtime.state = VTIME_INACTIVE;
2280#endif
2281
2282#ifdef CONFIG_IO_URING
2283 p->io_uring = NULL;
2284#endif
2285
2286 p->default_timer_slack_ns = current->timer_slack_ns;
2287
2288#ifdef CONFIG_PSI
2289 p->psi_flags = 0;
2290#endif
2291
2292 task_io_accounting_init(&p->ioac);
2293 acct_clear_integrals(p);
2294
2295 posix_cputimers_init(&p->posix_cputimers);
2296 tick_dep_init_task(p);
2297
2298 p->io_context = NULL;
2299 audit_set_context(p, NULL);
2300 cgroup_fork(p);
2301 if (args->kthread) {
2302 if (!set_kthread_struct(p))
2303 goto bad_fork_cleanup_delayacct;
2304 }
2305#ifdef CONFIG_NUMA
2306 p->mempolicy = mpol_dup(p->mempolicy);
2307 if (IS_ERR(p->mempolicy)) {
2308 retval = PTR_ERR(p->mempolicy);
2309 p->mempolicy = NULL;
2310 goto bad_fork_cleanup_delayacct;
2311 }
2312#endif
2313#ifdef CONFIG_CPUSETS
2314 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2315 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2316#endif
2317#ifdef CONFIG_TRACE_IRQFLAGS
2318 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2319 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2320 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2321 p->softirqs_enabled = 1;
2322 p->softirq_context = 0;
2323#endif
2324
2325 p->pagefault_disabled = 0;
2326
2327#ifdef CONFIG_LOCKDEP
2328 lockdep_init_task(p);
2329#endif
2330
2331#ifdef CONFIG_DEBUG_MUTEXES
2332 p->blocked_on = NULL; /* not blocked yet */
2333#endif
2334#ifdef CONFIG_BCACHE
2335 p->sequential_io = 0;
2336 p->sequential_io_avg = 0;
2337#endif
2338#ifdef CONFIG_BPF_SYSCALL
2339 RCU_INIT_POINTER(p->bpf_storage, NULL);
2340 p->bpf_ctx = NULL;
2341#endif
2342
2343 /* Perform scheduler related setup. Assign this task to a CPU. */
2344 retval = sched_fork(clone_flags, p);
2345 if (retval)
2346 goto bad_fork_cleanup_policy;
2347
2348 retval = perf_event_init_task(p, clone_flags);
2349 if (retval)
2350 goto bad_fork_sched_cancel_fork;
2351 retval = audit_alloc(p);
2352 if (retval)
2353 goto bad_fork_cleanup_perf;
2354 /* copy all the process information */
2355 shm_init_task(p);
2356 retval = security_task_alloc(p, clone_flags);
2357 if (retval)
2358 goto bad_fork_cleanup_audit;
2359 retval = copy_semundo(clone_flags, p);
2360 if (retval)
2361 goto bad_fork_cleanup_security;
2362 retval = copy_files(clone_flags, p, args->no_files);
2363 if (retval)
2364 goto bad_fork_cleanup_semundo;
2365 retval = copy_fs(clone_flags, p);
2366 if (retval)
2367 goto bad_fork_cleanup_files;
2368 retval = copy_sighand(clone_flags, p);
2369 if (retval)
2370 goto bad_fork_cleanup_fs;
2371 retval = copy_signal(clone_flags, p);
2372 if (retval)
2373 goto bad_fork_cleanup_sighand;
2374 retval = copy_mm(clone_flags, p);
2375 if (retval)
2376 goto bad_fork_cleanup_signal;
2377 retval = copy_namespaces(clone_flags, p);
2378 if (retval)
2379 goto bad_fork_cleanup_mm;
2380 retval = copy_io(clone_flags, p);
2381 if (retval)
2382 goto bad_fork_cleanup_namespaces;
2383 retval = copy_thread(p, args);
2384 if (retval)
2385 goto bad_fork_cleanup_io;
2386
2387 stackleak_task_init(p);
2388
2389 if (pid != &init_struct_pid) {
2390 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2391 args->set_tid_size);
2392 if (IS_ERR(pid)) {
2393 retval = PTR_ERR(pid);
2394 goto bad_fork_cleanup_thread;
2395 }
2396 }
2397
2398 /*
2399 * This has to happen after we've potentially unshared the file
2400 * descriptor table (so that the pidfd doesn't leak into the child
2401 * if the fd table isn't shared).
2402 */
2403 if (clone_flags & CLONE_PIDFD) {
2404 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2405
2406 /* Note that no task has been attached to @pid yet. */
2407 retval = __pidfd_prepare(pid, flags, &pidfile);
2408 if (retval < 0)
2409 goto bad_fork_free_pid;
2410 pidfd = retval;
2411
2412 retval = put_user(pidfd, args->pidfd);
2413 if (retval)
2414 goto bad_fork_put_pidfd;
2415 }
2416
2417#ifdef CONFIG_BLOCK
2418 p->plug = NULL;
2419#endif
2420 futex_init_task(p);
2421
2422 /*
2423 * sigaltstack should be cleared when sharing the same VM
2424 */
2425 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2426 sas_ss_reset(p);
2427
2428 /*
2429 * Syscall tracing and stepping should be turned off in the
2430 * child regardless of CLONE_PTRACE.
2431 */
2432 user_disable_single_step(p);
2433 clear_task_syscall_work(p, SYSCALL_TRACE);
2434#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2435 clear_task_syscall_work(p, SYSCALL_EMU);
2436#endif
2437 clear_tsk_latency_tracing(p);
2438
2439 /* ok, now we should be set up.. */
2440 p->pid = pid_nr(pid);
2441 if (clone_flags & CLONE_THREAD) {
2442 p->group_leader = current->group_leader;
2443 p->tgid = current->tgid;
2444 } else {
2445 p->group_leader = p;
2446 p->tgid = p->pid;
2447 }
2448
2449 p->nr_dirtied = 0;
2450 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2451 p->dirty_paused_when = 0;
2452
2453 p->pdeath_signal = 0;
2454 p->task_works = NULL;
2455 clear_posix_cputimers_work(p);
2456
2457#ifdef CONFIG_KRETPROBES
2458 p->kretprobe_instances.first = NULL;
2459#endif
2460#ifdef CONFIG_RETHOOK
2461 p->rethooks.first = NULL;
2462#endif
2463
2464 /*
2465 * Ensure that the cgroup subsystem policies allow the new process to be
2466 * forked. It should be noted that the new process's css_set can be changed
2467 * between here and cgroup_post_fork() if an organisation operation is in
2468 * progress.
2469 */
2470 retval = cgroup_can_fork(p, args);
2471 if (retval)
2472 goto bad_fork_put_pidfd;
2473
2474 /*
2475 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2476 * the new task on the correct runqueue. All this *before* the task
2477 * becomes visible.
2478 *
2479 * This isn't part of ->can_fork() because while the re-cloning is
2480 * cgroup specific, it unconditionally needs to place the task on a
2481 * runqueue.
2482 */
2483 retval = sched_cgroup_fork(p, args);
2484 if (retval)
2485 goto bad_fork_cancel_cgroup;
2486
2487 /*
2488 * From this point on we must avoid any synchronous user-space
2489 * communication until we take the tasklist-lock. In particular, we do
2490 * not want user-space to be able to predict the process start-time by
2491 * stalling fork(2) after we recorded the start_time but before it is
2492 * visible to the system.
2493 */
2494
2495 p->start_time = ktime_get_ns();
2496 p->start_boottime = ktime_get_boottime_ns();
2497
2498 /*
2499 * Make it visible to the rest of the system, but dont wake it up yet.
2500 * Need tasklist lock for parent etc handling!
2501 */
2502 write_lock_irq(&tasklist_lock);
2503
2504 /* CLONE_PARENT re-uses the old parent */
2505 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2506 p->real_parent = current->real_parent;
2507 p->parent_exec_id = current->parent_exec_id;
2508 if (clone_flags & CLONE_THREAD)
2509 p->exit_signal = -1;
2510 else
2511 p->exit_signal = current->group_leader->exit_signal;
2512 } else {
2513 p->real_parent = current;
2514 p->parent_exec_id = current->self_exec_id;
2515 p->exit_signal = args->exit_signal;
2516 }
2517
2518 klp_copy_process(p);
2519
2520 sched_core_fork(p);
2521
2522 spin_lock(&current->sighand->siglock);
2523
2524 rv_task_fork(p);
2525
2526 rseq_fork(p, clone_flags);
2527
2528 /* Don't start children in a dying pid namespace */
2529 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2530 retval = -ENOMEM;
2531 goto bad_fork_core_free;
2532 }
2533
2534 /* Let kill terminate clone/fork in the middle */
2535 if (fatal_signal_pending(current)) {
2536 retval = -EINTR;
2537 goto bad_fork_core_free;
2538 }
2539
2540 /* No more failure paths after this point. */
2541
2542 /*
2543 * Copy seccomp details explicitly here, in case they were changed
2544 * before holding sighand lock.
2545 */
2546 copy_seccomp(p);
2547
2548 init_task_pid_links(p);
2549 if (likely(p->pid)) {
2550 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2551
2552 init_task_pid(p, PIDTYPE_PID, pid);
2553 if (thread_group_leader(p)) {
2554 init_task_pid(p, PIDTYPE_TGID, pid);
2555 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2556 init_task_pid(p, PIDTYPE_SID, task_session(current));
2557
2558 if (is_child_reaper(pid)) {
2559 ns_of_pid(pid)->child_reaper = p;
2560 p->signal->flags |= SIGNAL_UNKILLABLE;
2561 }
2562 p->signal->shared_pending.signal = delayed.signal;
2563 p->signal->tty = tty_kref_get(current->signal->tty);
2564 /*
2565 * Inherit has_child_subreaper flag under the same
2566 * tasklist_lock with adding child to the process tree
2567 * for propagate_has_child_subreaper optimization.
2568 */
2569 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2570 p->real_parent->signal->is_child_subreaper;
2571 list_add_tail(&p->sibling, &p->real_parent->children);
2572 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2573 attach_pid(p, PIDTYPE_TGID);
2574 attach_pid(p, PIDTYPE_PGID);
2575 attach_pid(p, PIDTYPE_SID);
2576 __this_cpu_inc(process_counts);
2577 } else {
2578 current->signal->nr_threads++;
2579 current->signal->quick_threads++;
2580 atomic_inc(&current->signal->live);
2581 refcount_inc(&current->signal->sigcnt);
2582 task_join_group_stop(p);
2583 list_add_tail_rcu(&p->thread_node,
2584 &p->signal->thread_head);
2585 }
2586 attach_pid(p, PIDTYPE_PID);
2587 nr_threads++;
2588 }
2589 total_forks++;
2590 hlist_del_init(&delayed.node);
2591 spin_unlock(&current->sighand->siglock);
2592 syscall_tracepoint_update(p);
2593 write_unlock_irq(&tasklist_lock);
2594
2595 if (pidfile)
2596 fd_install(pidfd, pidfile);
2597
2598 proc_fork_connector(p);
2599 sched_post_fork(p);
2600 cgroup_post_fork(p, args);
2601 perf_event_fork(p);
2602
2603 trace_task_newtask(p, clone_flags);
2604 uprobe_copy_process(p, clone_flags);
2605 user_events_fork(p, clone_flags);
2606
2607 copy_oom_score_adj(clone_flags, p);
2608
2609 return p;
2610
2611bad_fork_core_free:
2612 sched_core_free(p);
2613 spin_unlock(&current->sighand->siglock);
2614 write_unlock_irq(&tasklist_lock);
2615bad_fork_cancel_cgroup:
2616 cgroup_cancel_fork(p, args);
2617bad_fork_put_pidfd:
2618 if (clone_flags & CLONE_PIDFD) {
2619 fput(pidfile);
2620 put_unused_fd(pidfd);
2621 }
2622bad_fork_free_pid:
2623 if (pid != &init_struct_pid)
2624 free_pid(pid);
2625bad_fork_cleanup_thread:
2626 exit_thread(p);
2627bad_fork_cleanup_io:
2628 if (p->io_context)
2629 exit_io_context(p);
2630bad_fork_cleanup_namespaces:
2631 exit_task_namespaces(p);
2632bad_fork_cleanup_mm:
2633 if (p->mm) {
2634 mm_clear_owner(p->mm, p);
2635 mmput(p->mm);
2636 }
2637bad_fork_cleanup_signal:
2638 if (!(clone_flags & CLONE_THREAD))
2639 free_signal_struct(p->signal);
2640bad_fork_cleanup_sighand:
2641 __cleanup_sighand(p->sighand);
2642bad_fork_cleanup_fs:
2643 exit_fs(p); /* blocking */
2644bad_fork_cleanup_files:
2645 exit_files(p); /* blocking */
2646bad_fork_cleanup_semundo:
2647 exit_sem(p);
2648bad_fork_cleanup_security:
2649 security_task_free(p);
2650bad_fork_cleanup_audit:
2651 audit_free(p);
2652bad_fork_cleanup_perf:
2653 perf_event_free_task(p);
2654bad_fork_sched_cancel_fork:
2655 sched_cancel_fork(p);
2656bad_fork_cleanup_policy:
2657 lockdep_free_task(p);
2658#ifdef CONFIG_NUMA
2659 mpol_put(p->mempolicy);
2660#endif
2661bad_fork_cleanup_delayacct:
2662 delayacct_tsk_free(p);
2663bad_fork_cleanup_count:
2664 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2665 exit_creds(p);
2666bad_fork_free:
2667 WRITE_ONCE(p->__state, TASK_DEAD);
2668 exit_task_stack_account(p);
2669 put_task_stack(p);
2670 delayed_free_task(p);
2671fork_out:
2672 spin_lock_irq(&current->sighand->siglock);
2673 hlist_del_init(&delayed.node);
2674 spin_unlock_irq(&current->sighand->siglock);
2675 return ERR_PTR(retval);
2676}
2677
2678static inline void init_idle_pids(struct task_struct *idle)
2679{
2680 enum pid_type type;
2681
2682 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2683 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2684 init_task_pid(idle, type, &init_struct_pid);
2685 }
2686}
2687
2688static int idle_dummy(void *dummy)
2689{
2690 /* This function is never called */
2691 return 0;
2692}
2693
2694struct task_struct * __init fork_idle(int cpu)
2695{
2696 struct task_struct *task;
2697 struct kernel_clone_args args = {
2698 .flags = CLONE_VM,
2699 .fn = &idle_dummy,
2700 .fn_arg = NULL,
2701 .kthread = 1,
2702 .idle = 1,
2703 };
2704
2705 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2706 if (!IS_ERR(task)) {
2707 init_idle_pids(task);
2708 init_idle(task, cpu);
2709 }
2710
2711 return task;
2712}
2713
2714/*
2715 * This is like kernel_clone(), but shaved down and tailored to just
2716 * creating io_uring workers. It returns a created task, or an error pointer.
2717 * The returned task is inactive, and the caller must fire it up through
2718 * wake_up_new_task(p). All signals are blocked in the created task.
2719 */
2720struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2721{
2722 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2723 CLONE_IO;
2724 struct kernel_clone_args args = {
2725 .flags = ((lower_32_bits(flags) | CLONE_VM |
2726 CLONE_UNTRACED) & ~CSIGNAL),
2727 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2728 .fn = fn,
2729 .fn_arg = arg,
2730 .io_thread = 1,
2731 .user_worker = 1,
2732 };
2733
2734 return copy_process(NULL, 0, node, &args);
2735}
2736
2737/*
2738 * Ok, this is the main fork-routine.
2739 *
2740 * It copies the process, and if successful kick-starts
2741 * it and waits for it to finish using the VM if required.
2742 *
2743 * args->exit_signal is expected to be checked for sanity by the caller.
2744 */
2745pid_t kernel_clone(struct kernel_clone_args *args)
2746{
2747 u64 clone_flags = args->flags;
2748 struct completion vfork;
2749 struct pid *pid;
2750 struct task_struct *p;
2751 int trace = 0;
2752 pid_t nr;
2753
2754 /*
2755 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2756 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2757 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2758 * field in struct clone_args and it still doesn't make sense to have
2759 * them both point at the same memory location. Performing this check
2760 * here has the advantage that we don't need to have a separate helper
2761 * to check for legacy clone().
2762 */
2763 if ((clone_flags & CLONE_PIDFD) &&
2764 (clone_flags & CLONE_PARENT_SETTID) &&
2765 (args->pidfd == args->parent_tid))
2766 return -EINVAL;
2767
2768 /*
2769 * Determine whether and which event to report to ptracer. When
2770 * called from kernel_thread or CLONE_UNTRACED is explicitly
2771 * requested, no event is reported; otherwise, report if the event
2772 * for the type of forking is enabled.
2773 */
2774 if (!(clone_flags & CLONE_UNTRACED)) {
2775 if (clone_flags & CLONE_VFORK)
2776 trace = PTRACE_EVENT_VFORK;
2777 else if (args->exit_signal != SIGCHLD)
2778 trace = PTRACE_EVENT_CLONE;
2779 else
2780 trace = PTRACE_EVENT_FORK;
2781
2782 if (likely(!ptrace_event_enabled(current, trace)))
2783 trace = 0;
2784 }
2785
2786 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2787 add_latent_entropy();
2788
2789 if (IS_ERR(p))
2790 return PTR_ERR(p);
2791
2792 /*
2793 * Do this prior waking up the new thread - the thread pointer
2794 * might get invalid after that point, if the thread exits quickly.
2795 */
2796 trace_sched_process_fork(current, p);
2797
2798 pid = get_task_pid(p, PIDTYPE_PID);
2799 nr = pid_vnr(pid);
2800
2801 if (clone_flags & CLONE_PARENT_SETTID)
2802 put_user(nr, args->parent_tid);
2803
2804 if (clone_flags & CLONE_VFORK) {
2805 p->vfork_done = &vfork;
2806 init_completion(&vfork);
2807 get_task_struct(p);
2808 }
2809
2810 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2811 /* lock the task to synchronize with memcg migration */
2812 task_lock(p);
2813 lru_gen_add_mm(p->mm);
2814 task_unlock(p);
2815 }
2816
2817 wake_up_new_task(p);
2818
2819 /* forking complete and child started to run, tell ptracer */
2820 if (unlikely(trace))
2821 ptrace_event_pid(trace, pid);
2822
2823 if (clone_flags & CLONE_VFORK) {
2824 if (!wait_for_vfork_done(p, &vfork))
2825 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2826 }
2827
2828 put_pid(pid);
2829 return nr;
2830}
2831
2832/*
2833 * Create a kernel thread.
2834 */
2835pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2836 unsigned long flags)
2837{
2838 struct kernel_clone_args args = {
2839 .flags = ((lower_32_bits(flags) | CLONE_VM |
2840 CLONE_UNTRACED) & ~CSIGNAL),
2841 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2842 .fn = fn,
2843 .fn_arg = arg,
2844 .name = name,
2845 .kthread = 1,
2846 };
2847
2848 return kernel_clone(&args);
2849}
2850
2851/*
2852 * Create a user mode thread.
2853 */
2854pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2855{
2856 struct kernel_clone_args args = {
2857 .flags = ((lower_32_bits(flags) | CLONE_VM |
2858 CLONE_UNTRACED) & ~CSIGNAL),
2859 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2860 .fn = fn,
2861 .fn_arg = arg,
2862 };
2863
2864 return kernel_clone(&args);
2865}
2866
2867#ifdef __ARCH_WANT_SYS_FORK
2868SYSCALL_DEFINE0(fork)
2869{
2870#ifdef CONFIG_MMU
2871 struct kernel_clone_args args = {
2872 .exit_signal = SIGCHLD,
2873 };
2874
2875 return kernel_clone(&args);
2876#else
2877 /* can not support in nommu mode */
2878 return -EINVAL;
2879#endif
2880}
2881#endif
2882
2883#ifdef __ARCH_WANT_SYS_VFORK
2884SYSCALL_DEFINE0(vfork)
2885{
2886 struct kernel_clone_args args = {
2887 .flags = CLONE_VFORK | CLONE_VM,
2888 .exit_signal = SIGCHLD,
2889 };
2890
2891 return kernel_clone(&args);
2892}
2893#endif
2894
2895#ifdef __ARCH_WANT_SYS_CLONE
2896#ifdef CONFIG_CLONE_BACKWARDS
2897SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2898 int __user *, parent_tidptr,
2899 unsigned long, tls,
2900 int __user *, child_tidptr)
2901#elif defined(CONFIG_CLONE_BACKWARDS2)
2902SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2903 int __user *, parent_tidptr,
2904 int __user *, child_tidptr,
2905 unsigned long, tls)
2906#elif defined(CONFIG_CLONE_BACKWARDS3)
2907SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2908 int, stack_size,
2909 int __user *, parent_tidptr,
2910 int __user *, child_tidptr,
2911 unsigned long, tls)
2912#else
2913SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2914 int __user *, parent_tidptr,
2915 int __user *, child_tidptr,
2916 unsigned long, tls)
2917#endif
2918{
2919 struct kernel_clone_args args = {
2920 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2921 .pidfd = parent_tidptr,
2922 .child_tid = child_tidptr,
2923 .parent_tid = parent_tidptr,
2924 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2925 .stack = newsp,
2926 .tls = tls,
2927 };
2928
2929 return kernel_clone(&args);
2930}
2931#endif
2932
2933noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2934 struct clone_args __user *uargs,
2935 size_t usize)
2936{
2937 int err;
2938 struct clone_args args;
2939 pid_t *kset_tid = kargs->set_tid;
2940
2941 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2942 CLONE_ARGS_SIZE_VER0);
2943 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2944 CLONE_ARGS_SIZE_VER1);
2945 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2946 CLONE_ARGS_SIZE_VER2);
2947 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2948
2949 if (unlikely(usize > PAGE_SIZE))
2950 return -E2BIG;
2951 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2952 return -EINVAL;
2953
2954 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2955 if (err)
2956 return err;
2957
2958 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2959 return -EINVAL;
2960
2961 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2962 return -EINVAL;
2963
2964 if (unlikely(args.set_tid && args.set_tid_size == 0))
2965 return -EINVAL;
2966
2967 /*
2968 * Verify that higher 32bits of exit_signal are unset and that
2969 * it is a valid signal
2970 */
2971 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2972 !valid_signal(args.exit_signal)))
2973 return -EINVAL;
2974
2975 if ((args.flags & CLONE_INTO_CGROUP) &&
2976 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2977 return -EINVAL;
2978
2979 *kargs = (struct kernel_clone_args){
2980 .flags = args.flags,
2981 .pidfd = u64_to_user_ptr(args.pidfd),
2982 .child_tid = u64_to_user_ptr(args.child_tid),
2983 .parent_tid = u64_to_user_ptr(args.parent_tid),
2984 .exit_signal = args.exit_signal,
2985 .stack = args.stack,
2986 .stack_size = args.stack_size,
2987 .tls = args.tls,
2988 .set_tid_size = args.set_tid_size,
2989 .cgroup = args.cgroup,
2990 };
2991
2992 if (args.set_tid &&
2993 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2994 (kargs->set_tid_size * sizeof(pid_t))))
2995 return -EFAULT;
2996
2997 kargs->set_tid = kset_tid;
2998
2999 return 0;
3000}
3001
3002/**
3003 * clone3_stack_valid - check and prepare stack
3004 * @kargs: kernel clone args
3005 *
3006 * Verify that the stack arguments userspace gave us are sane.
3007 * In addition, set the stack direction for userspace since it's easy for us to
3008 * determine.
3009 */
3010static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3011{
3012 if (kargs->stack == 0) {
3013 if (kargs->stack_size > 0)
3014 return false;
3015 } else {
3016 if (kargs->stack_size == 0)
3017 return false;
3018
3019 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3020 return false;
3021
3022#if !defined(CONFIG_STACK_GROWSUP)
3023 kargs->stack += kargs->stack_size;
3024#endif
3025 }
3026
3027 return true;
3028}
3029
3030static bool clone3_args_valid(struct kernel_clone_args *kargs)
3031{
3032 /* Verify that no unknown flags are passed along. */
3033 if (kargs->flags &
3034 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3035 return false;
3036
3037 /*
3038 * - make the CLONE_DETACHED bit reusable for clone3
3039 * - make the CSIGNAL bits reusable for clone3
3040 */
3041 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3042 return false;
3043
3044 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3045 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3046 return false;
3047
3048 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3049 kargs->exit_signal)
3050 return false;
3051
3052 if (!clone3_stack_valid(kargs))
3053 return false;
3054
3055 return true;
3056}
3057
3058/**
3059 * sys_clone3 - create a new process with specific properties
3060 * @uargs: argument structure
3061 * @size: size of @uargs
3062 *
3063 * clone3() is the extensible successor to clone()/clone2().
3064 * It takes a struct as argument that is versioned by its size.
3065 *
3066 * Return: On success, a positive PID for the child process.
3067 * On error, a negative errno number.
3068 */
3069SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3070{
3071 int err;
3072
3073 struct kernel_clone_args kargs;
3074 pid_t set_tid[MAX_PID_NS_LEVEL];
3075
3076#ifdef __ARCH_BROKEN_SYS_CLONE3
3077#warning clone3() entry point is missing, please fix
3078 return -ENOSYS;
3079#endif
3080
3081 kargs.set_tid = set_tid;
3082
3083 err = copy_clone_args_from_user(&kargs, uargs, size);
3084 if (err)
3085 return err;
3086
3087 if (!clone3_args_valid(&kargs))
3088 return -EINVAL;
3089
3090 return kernel_clone(&kargs);
3091}
3092
3093void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3094{
3095 struct task_struct *leader, *parent, *child;
3096 int res;
3097
3098 read_lock(&tasklist_lock);
3099 leader = top = top->group_leader;
3100down:
3101 for_each_thread(leader, parent) {
3102 list_for_each_entry(child, &parent->children, sibling) {
3103 res = visitor(child, data);
3104 if (res) {
3105 if (res < 0)
3106 goto out;
3107 leader = child;
3108 goto down;
3109 }
3110up:
3111 ;
3112 }
3113 }
3114
3115 if (leader != top) {
3116 child = leader;
3117 parent = child->real_parent;
3118 leader = parent->group_leader;
3119 goto up;
3120 }
3121out:
3122 read_unlock(&tasklist_lock);
3123}
3124
3125#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3126#define ARCH_MIN_MMSTRUCT_ALIGN 0
3127#endif
3128
3129static void sighand_ctor(void *data)
3130{
3131 struct sighand_struct *sighand = data;
3132
3133 spin_lock_init(&sighand->siglock);
3134 init_waitqueue_head(&sighand->signalfd_wqh);
3135}
3136
3137void __init mm_cache_init(void)
3138{
3139 unsigned int mm_size;
3140
3141 /*
3142 * The mm_cpumask is located at the end of mm_struct, and is
3143 * dynamically sized based on the maximum CPU number this system
3144 * can have, taking hotplug into account (nr_cpu_ids).
3145 */
3146 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3147
3148 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3149 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3150 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3151 offsetof(struct mm_struct, saved_auxv),
3152 sizeof_field(struct mm_struct, saved_auxv),
3153 NULL);
3154}
3155
3156void __init proc_caches_init(void)
3157{
3158 sighand_cachep = kmem_cache_create("sighand_cache",
3159 sizeof(struct sighand_struct), 0,
3160 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3161 SLAB_ACCOUNT, sighand_ctor);
3162 signal_cachep = kmem_cache_create("signal_cache",
3163 sizeof(struct signal_struct), 0,
3164 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3165 NULL);
3166 files_cachep = kmem_cache_create("files_cache",
3167 sizeof(struct files_struct), 0,
3168 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3169 NULL);
3170 fs_cachep = kmem_cache_create("fs_cache",
3171 sizeof(struct fs_struct), 0,
3172 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3173 NULL);
3174
3175 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3176#ifdef CONFIG_PER_VMA_LOCK
3177 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3178#endif
3179 mmap_init();
3180 nsproxy_cache_init();
3181}
3182
3183/*
3184 * Check constraints on flags passed to the unshare system call.
3185 */
3186static int check_unshare_flags(unsigned long unshare_flags)
3187{
3188 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3189 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3190 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3191 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3192 CLONE_NEWTIME))
3193 return -EINVAL;
3194 /*
3195 * Not implemented, but pretend it works if there is nothing
3196 * to unshare. Note that unsharing the address space or the
3197 * signal handlers also need to unshare the signal queues (aka
3198 * CLONE_THREAD).
3199 */
3200 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3201 if (!thread_group_empty(current))
3202 return -EINVAL;
3203 }
3204 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3205 if (refcount_read(&current->sighand->count) > 1)
3206 return -EINVAL;
3207 }
3208 if (unshare_flags & CLONE_VM) {
3209 if (!current_is_single_threaded())
3210 return -EINVAL;
3211 }
3212
3213 return 0;
3214}
3215
3216/*
3217 * Unshare the filesystem structure if it is being shared
3218 */
3219static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3220{
3221 struct fs_struct *fs = current->fs;
3222
3223 if (!(unshare_flags & CLONE_FS) || !fs)
3224 return 0;
3225
3226 /* don't need lock here; in the worst case we'll do useless copy */
3227 if (fs->users == 1)
3228 return 0;
3229
3230 *new_fsp = copy_fs_struct(fs);
3231 if (!*new_fsp)
3232 return -ENOMEM;
3233
3234 return 0;
3235}
3236
3237/*
3238 * Unshare file descriptor table if it is being shared
3239 */
3240static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
3241{
3242 struct files_struct *fd = current->files;
3243
3244 if ((unshare_flags & CLONE_FILES) &&
3245 (fd && atomic_read(&fd->count) > 1)) {
3246 fd = dup_fd(fd, NULL);
3247 if (IS_ERR(fd))
3248 return PTR_ERR(fd);
3249 *new_fdp = fd;
3250 }
3251
3252 return 0;
3253}
3254
3255/*
3256 * unshare allows a process to 'unshare' part of the process
3257 * context which was originally shared using clone. copy_*
3258 * functions used by kernel_clone() cannot be used here directly
3259 * because they modify an inactive task_struct that is being
3260 * constructed. Here we are modifying the current, active,
3261 * task_struct.
3262 */
3263int ksys_unshare(unsigned long unshare_flags)
3264{
3265 struct fs_struct *fs, *new_fs = NULL;
3266 struct files_struct *new_fd = NULL;
3267 struct cred *new_cred = NULL;
3268 struct nsproxy *new_nsproxy = NULL;
3269 int do_sysvsem = 0;
3270 int err;
3271
3272 /*
3273 * If unsharing a user namespace must also unshare the thread group
3274 * and unshare the filesystem root and working directories.
3275 */
3276 if (unshare_flags & CLONE_NEWUSER)
3277 unshare_flags |= CLONE_THREAD | CLONE_FS;
3278 /*
3279 * If unsharing vm, must also unshare signal handlers.
3280 */
3281 if (unshare_flags & CLONE_VM)
3282 unshare_flags |= CLONE_SIGHAND;
3283 /*
3284 * If unsharing a signal handlers, must also unshare the signal queues.
3285 */
3286 if (unshare_flags & CLONE_SIGHAND)
3287 unshare_flags |= CLONE_THREAD;
3288 /*
3289 * If unsharing namespace, must also unshare filesystem information.
3290 */
3291 if (unshare_flags & CLONE_NEWNS)
3292 unshare_flags |= CLONE_FS;
3293
3294 err = check_unshare_flags(unshare_flags);
3295 if (err)
3296 goto bad_unshare_out;
3297 /*
3298 * CLONE_NEWIPC must also detach from the undolist: after switching
3299 * to a new ipc namespace, the semaphore arrays from the old
3300 * namespace are unreachable.
3301 */
3302 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3303 do_sysvsem = 1;
3304 err = unshare_fs(unshare_flags, &new_fs);
3305 if (err)
3306 goto bad_unshare_out;
3307 err = unshare_fd(unshare_flags, &new_fd);
3308 if (err)
3309 goto bad_unshare_cleanup_fs;
3310 err = unshare_userns(unshare_flags, &new_cred);
3311 if (err)
3312 goto bad_unshare_cleanup_fd;
3313 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3314 new_cred, new_fs);
3315 if (err)
3316 goto bad_unshare_cleanup_cred;
3317
3318 if (new_cred) {
3319 err = set_cred_ucounts(new_cred);
3320 if (err)
3321 goto bad_unshare_cleanup_cred;
3322 }
3323
3324 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3325 if (do_sysvsem) {
3326 /*
3327 * CLONE_SYSVSEM is equivalent to sys_exit().
3328 */
3329 exit_sem(current);
3330 }
3331 if (unshare_flags & CLONE_NEWIPC) {
3332 /* Orphan segments in old ns (see sem above). */
3333 exit_shm(current);
3334 shm_init_task(current);
3335 }
3336
3337 if (new_nsproxy)
3338 switch_task_namespaces(current, new_nsproxy);
3339
3340 task_lock(current);
3341
3342 if (new_fs) {
3343 fs = current->fs;
3344 spin_lock(&fs->lock);
3345 current->fs = new_fs;
3346 if (--fs->users)
3347 new_fs = NULL;
3348 else
3349 new_fs = fs;
3350 spin_unlock(&fs->lock);
3351 }
3352
3353 if (new_fd)
3354 swap(current->files, new_fd);
3355
3356 task_unlock(current);
3357
3358 if (new_cred) {
3359 /* Install the new user namespace */
3360 commit_creds(new_cred);
3361 new_cred = NULL;
3362 }
3363 }
3364
3365 perf_event_namespaces(current);
3366
3367bad_unshare_cleanup_cred:
3368 if (new_cred)
3369 put_cred(new_cred);
3370bad_unshare_cleanup_fd:
3371 if (new_fd)
3372 put_files_struct(new_fd);
3373
3374bad_unshare_cleanup_fs:
3375 if (new_fs)
3376 free_fs_struct(new_fs);
3377
3378bad_unshare_out:
3379 return err;
3380}
3381
3382SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3383{
3384 return ksys_unshare(unshare_flags);
3385}
3386
3387/*
3388 * Helper to unshare the files of the current task.
3389 * We don't want to expose copy_files internals to
3390 * the exec layer of the kernel.
3391 */
3392
3393int unshare_files(void)
3394{
3395 struct task_struct *task = current;
3396 struct files_struct *old, *copy = NULL;
3397 int error;
3398
3399 error = unshare_fd(CLONE_FILES, &copy);
3400 if (error || !copy)
3401 return error;
3402
3403 old = task->files;
3404 task_lock(task);
3405 task->files = copy;
3406 task_unlock(task);
3407 put_files_struct(old);
3408 return 0;
3409}
3410
3411int sysctl_max_threads(const struct ctl_table *table, int write,
3412 void *buffer, size_t *lenp, loff_t *ppos)
3413{
3414 struct ctl_table t;
3415 int ret;
3416 int threads = max_threads;
3417 int min = 1;
3418 int max = MAX_THREADS;
3419
3420 t = *table;
3421 t.data = &threads;
3422 t.extra1 = &min;
3423 t.extra2 = &max;
3424
3425 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3426 if (ret || !write)
3427 return ret;
3428
3429 max_threads = threads;
3430
3431 return 0;
3432}