projects/linux·blob v6.5 · refs

3557 lines · 87.4 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/jiffies.h>
57#include <linux/futex.h>
58#include <linux/compat.h>
59#include <linux/kthread.h>
60#include <linux/task_io_accounting_ops.h>
61#include <linux/rcupdate.h>
62#include <linux/ptrace.h>
63#include <linux/mount.h>
64#include <linux/audit.h>
65#include <linux/memcontrol.h>
66#include <linux/ftrace.h>
67#include <linux/proc_fs.h>
68#include <linux/profile.h>
69#include <linux/rmap.h>
70#include <linux/ksm.h>
71#include <linux/acct.h>
72#include <linux/userfaultfd_k.h>
73#include <linux/tsacct_kern.h>
74#include <linux/cn_proc.h>
75#include <linux/freezer.h>
76#include <linux/delayacct.h>
77#include <linux/taskstats_kern.h>
78#include <linux/tty.h>
79#include <linux/fs_struct.h>
80#include <linux/magic.h>
81#include <linux/perf_event.h>
82#include <linux/posix-timers.h>
83#include <linux/user-return-notifier.h>
84#include <linux/oom.h>
85#include <linux/khugepaged.h>
86#include <linux/signalfd.h>
87#include <linux/uprobes.h>
88#include <linux/aio.h>
89#include <linux/compiler.h>
90#include <linux/sysctl.h>
91#include <linux/kcov.h>
92#include <linux/livepatch.h>
93#include <linux/thread_info.h>
94#include <linux/stackleak.h>
95#include <linux/kasan.h>
96#include <linux/scs.h>
97#include <linux/io_uring.h>
98#include <linux/bpf.h>
99#include <linux/stackprotector.h>
100#include <linux/user_events.h>
101#include <linux/iommu.h>
102
103#include <asm/pgalloc.h>
104#include <linux/uaccess.h>
105#include <asm/mmu_context.h>
106#include <asm/cacheflush.h>
107#include <asm/tlbflush.h>
108
109#include <trace/events/sched.h>
110
111#define CREATE_TRACE_POINTS
112#include <trace/events/task.h>
113
114/*
115 * Minimum number of threads to boot the kernel
116 */
117#define MIN_THREADS 20
118
119/*
120 * Maximum number of threads
121 */
122#define MAX_THREADS FUTEX_TID_MASK
123
124/*
125 * Protected counters by write_lock_irq(&tasklist_lock)
126 */
127unsigned long total_forks; /* Handle normal Linux uptimes. */
128int nr_threads; /* The idle threads do not count.. */
129
130static int max_threads; /* tunable limit on nr_threads */
131
132#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
133
134static const char * const resident_page_types[] = {
135 NAMED_ARRAY_INDEX(MM_FILEPAGES),
136 NAMED_ARRAY_INDEX(MM_ANONPAGES),
137 NAMED_ARRAY_INDEX(MM_SWAPENTS),
138 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
139};
140
141DEFINE_PER_CPU(unsigned long, process_counts) = 0;
142
143__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
144
145#ifdef CONFIG_PROVE_RCU
146int lockdep_tasklist_lock_is_held(void)
147{
148 return lockdep_is_held(&tasklist_lock);
149}
150EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
151#endif /* #ifdef CONFIG_PROVE_RCU */
152
153int nr_processes(void)
154{
155 int cpu;
156 int total = 0;
157
158 for_each_possible_cpu(cpu)
159 total += per_cpu(process_counts, cpu);
160
161 return total;
162}
163
164void __weak arch_release_task_struct(struct task_struct *tsk)
165{
166}
167
168#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
169static struct kmem_cache *task_struct_cachep;
170
171static inline struct task_struct *alloc_task_struct_node(int node)
172{
173 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
174}
175
176static inline void free_task_struct(struct task_struct *tsk)
177{
178 kmem_cache_free(task_struct_cachep, tsk);
179}
180#endif
181
182#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
183
184/*
185 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
186 * kmemcache based allocator.
187 */
188# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189
190# ifdef CONFIG_VMAP_STACK
191/*
192 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
193 * flush. Try to minimize the number of calls by caching stacks.
194 */
195#define NR_CACHED_STACKS 2
196static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
197
198struct vm_stack {
199 struct rcu_head rcu;
200 struct vm_struct *stack_vm_area;
201};
202
203static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
204{
205 unsigned int i;
206
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
209 continue;
210 return true;
211 }
212 return false;
213}
214
215static void thread_stack_free_rcu(struct rcu_head *rh)
216{
217 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218
219 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
220 return;
221
222 vfree(vm_stack);
223}
224
225static void thread_stack_delayed_free(struct task_struct *tsk)
226{
227 struct vm_stack *vm_stack = tsk->stack;
228
229 vm_stack->stack_vm_area = tsk->stack_vm_area;
230 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231}
232
233static int free_vm_stack_cache(unsigned int cpu)
234{
235 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 int i;
237
238 for (i = 0; i < NR_CACHED_STACKS; i++) {
239 struct vm_struct *vm_stack = cached_vm_stacks[i];
240
241 if (!vm_stack)
242 continue;
243
244 vfree(vm_stack->addr);
245 cached_vm_stacks[i] = NULL;
246 }
247
248 return 0;
249}
250
251static int memcg_charge_kernel_stack(struct vm_struct *vm)
252{
253 int i;
254 int ret;
255 int nr_charged = 0;
256
257 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258
259 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
260 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
261 if (ret)
262 goto err;
263 nr_charged++;
264 }
265 return 0;
266err:
267 for (i = 0; i < nr_charged; i++)
268 memcg_kmem_uncharge_page(vm->pages[i], 0);
269 return ret;
270}
271
272static int alloc_thread_stack_node(struct task_struct *tsk, int node)
273{
274 struct vm_struct *vm;
275 void *stack;
276 int i;
277
278 for (i = 0; i < NR_CACHED_STACKS; i++) {
279 struct vm_struct *s;
280
281 s = this_cpu_xchg(cached_stacks[i], NULL);
282
283 if (!s)
284 continue;
285
286 /* Reset stack metadata. */
287 kasan_unpoison_range(s->addr, THREAD_SIZE);
288
289 stack = kasan_reset_tag(s->addr);
290
291 /* Clear stale pointers from reused stack. */
292 memset(stack, 0, THREAD_SIZE);
293
294 if (memcg_charge_kernel_stack(s)) {
295 vfree(s->addr);
296 return -ENOMEM;
297 }
298
299 tsk->stack_vm_area = s;
300 tsk->stack = stack;
301 return 0;
302 }
303
304 /*
305 * Allocated stacks are cached and later reused by new threads,
306 * so memcg accounting is performed manually on assigning/releasing
307 * stacks to tasks. Drop __GFP_ACCOUNT.
308 */
309 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
310 VMALLOC_START, VMALLOC_END,
311 THREADINFO_GFP & ~__GFP_ACCOUNT,
312 PAGE_KERNEL,
313 0, node, __builtin_return_address(0));
314 if (!stack)
315 return -ENOMEM;
316
317 vm = find_vm_area(stack);
318 if (memcg_charge_kernel_stack(vm)) {
319 vfree(stack);
320 return -ENOMEM;
321 }
322 /*
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
326 */
327 tsk->stack_vm_area = vm;
328 stack = kasan_reset_tag(stack);
329 tsk->stack = stack;
330 return 0;
331}
332
333static void free_thread_stack(struct task_struct *tsk)
334{
335 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
337
338 tsk->stack = NULL;
339 tsk->stack_vm_area = NULL;
340}
341
342# else /* !CONFIG_VMAP_STACK */
343
344static void thread_stack_free_rcu(struct rcu_head *rh)
345{
346 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
347}
348
349static void thread_stack_delayed_free(struct task_struct *tsk)
350{
351 struct rcu_head *rh = tsk->stack;
352
353 call_rcu(rh, thread_stack_free_rcu);
354}
355
356static int alloc_thread_stack_node(struct task_struct *tsk, int node)
357{
358 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
359 THREAD_SIZE_ORDER);
360
361 if (likely(page)) {
362 tsk->stack = kasan_reset_tag(page_address(page));
363 return 0;
364 }
365 return -ENOMEM;
366}
367
368static void free_thread_stack(struct task_struct *tsk)
369{
370 thread_stack_delayed_free(tsk);
371 tsk->stack = NULL;
372}
373
374# endif /* CONFIG_VMAP_STACK */
375# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
376
377static struct kmem_cache *thread_stack_cache;
378
379static void thread_stack_free_rcu(struct rcu_head *rh)
380{
381 kmem_cache_free(thread_stack_cache, rh);
382}
383
384static void thread_stack_delayed_free(struct task_struct *tsk)
385{
386 struct rcu_head *rh = tsk->stack;
387
388 call_rcu(rh, thread_stack_free_rcu);
389}
390
391static int alloc_thread_stack_node(struct task_struct *tsk, int node)
392{
393 unsigned long *stack;
394 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
395 stack = kasan_reset_tag(stack);
396 tsk->stack = stack;
397 return stack ? 0 : -ENOMEM;
398}
399
400static void free_thread_stack(struct task_struct *tsk)
401{
402 thread_stack_delayed_free(tsk);
403 tsk->stack = NULL;
404}
405
406void thread_stack_cache_init(void)
407{
408 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
409 THREAD_SIZE, THREAD_SIZE, 0, 0,
410 THREAD_SIZE, NULL);
411 BUG_ON(thread_stack_cache == NULL);
412}
413
414# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415#else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
416
417static int alloc_thread_stack_node(struct task_struct *tsk, int node)
418{
419 unsigned long *stack;
420
421 stack = arch_alloc_thread_stack_node(tsk, node);
422 tsk->stack = stack;
423 return stack ? 0 : -ENOMEM;
424}
425
426static void free_thread_stack(struct task_struct *tsk)
427{
428 arch_free_thread_stack(tsk);
429 tsk->stack = NULL;
430}
431
432#endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
433
434/* SLAB cache for signal_struct structures (tsk->signal) */
435static struct kmem_cache *signal_cachep;
436
437/* SLAB cache for sighand_struct structures (tsk->sighand) */
438struct kmem_cache *sighand_cachep;
439
440/* SLAB cache for files_struct structures (tsk->files) */
441struct kmem_cache *files_cachep;
442
443/* SLAB cache for fs_struct structures (tsk->fs) */
444struct kmem_cache *fs_cachep;
445
446/* SLAB cache for vm_area_struct structures */
447static struct kmem_cache *vm_area_cachep;
448
449/* SLAB cache for mm_struct structures (tsk->mm) */
450static struct kmem_cache *mm_cachep;
451
452#ifdef CONFIG_PER_VMA_LOCK
453
454/* SLAB cache for vm_area_struct.lock */
455static struct kmem_cache *vma_lock_cachep;
456
457static bool vma_lock_alloc(struct vm_area_struct *vma)
458{
459 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
460 if (!vma->vm_lock)
461 return false;
462
463 init_rwsem(&vma->vm_lock->lock);
464 vma->vm_lock_seq = -1;
465
466 return true;
467}
468
469static inline void vma_lock_free(struct vm_area_struct *vma)
470{
471 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
472}
473
474#else /* CONFIG_PER_VMA_LOCK */
475
476static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
477static inline void vma_lock_free(struct vm_area_struct *vma) {}
478
479#endif /* CONFIG_PER_VMA_LOCK */
480
481struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
482{
483 struct vm_area_struct *vma;
484
485 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
486 if (!vma)
487 return NULL;
488
489 vma_init(vma, mm);
490 if (!vma_lock_alloc(vma)) {
491 kmem_cache_free(vm_area_cachep, vma);
492 return NULL;
493 }
494
495 return vma;
496}
497
498struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
499{
500 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
501
502 if (!new)
503 return NULL;
504
505 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
506 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
507 /*
508 * orig->shared.rb may be modified concurrently, but the clone
509 * will be reinitialized.
510 */
511 data_race(memcpy(new, orig, sizeof(*new)));
512 if (!vma_lock_alloc(new)) {
513 kmem_cache_free(vm_area_cachep, new);
514 return NULL;
515 }
516 INIT_LIST_HEAD(&new->anon_vma_chain);
517 vma_numab_state_init(new);
518 dup_anon_vma_name(orig, new);
519
520 return new;
521}
522
523void __vm_area_free(struct vm_area_struct *vma)
524{
525 vma_numab_state_free(vma);
526 free_anon_vma_name(vma);
527 vma_lock_free(vma);
528 kmem_cache_free(vm_area_cachep, vma);
529}
530
531#ifdef CONFIG_PER_VMA_LOCK
532static void vm_area_free_rcu_cb(struct rcu_head *head)
533{
534 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
535 vm_rcu);
536
537 /* The vma should not be locked while being destroyed. */
538 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
539 __vm_area_free(vma);
540}
541#endif
542
543void vm_area_free(struct vm_area_struct *vma)
544{
545#ifdef CONFIG_PER_VMA_LOCK
546 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
547#else
548 __vm_area_free(vma);
549#endif
550}
551
552static void account_kernel_stack(struct task_struct *tsk, int account)
553{
554 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
555 struct vm_struct *vm = task_stack_vm_area(tsk);
556 int i;
557
558 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
559 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
560 account * (PAGE_SIZE / 1024));
561 } else {
562 void *stack = task_stack_page(tsk);
563
564 /* All stack pages are in the same node. */
565 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
566 account * (THREAD_SIZE / 1024));
567 }
568}
569
570void exit_task_stack_account(struct task_struct *tsk)
571{
572 account_kernel_stack(tsk, -1);
573
574 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
575 struct vm_struct *vm;
576 int i;
577
578 vm = task_stack_vm_area(tsk);
579 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
580 memcg_kmem_uncharge_page(vm->pages[i], 0);
581 }
582}
583
584static void release_task_stack(struct task_struct *tsk)
585{
586 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
587 return; /* Better to leak the stack than to free prematurely */
588
589 free_thread_stack(tsk);
590}
591
592#ifdef CONFIG_THREAD_INFO_IN_TASK
593void put_task_stack(struct task_struct *tsk)
594{
595 if (refcount_dec_and_test(&tsk->stack_refcount))
596 release_task_stack(tsk);
597}
598#endif
599
600void free_task(struct task_struct *tsk)
601{
602#ifdef CONFIG_SECCOMP
603 WARN_ON_ONCE(tsk->seccomp.filter);
604#endif
605 release_user_cpus_ptr(tsk);
606 scs_release(tsk);
607
608#ifndef CONFIG_THREAD_INFO_IN_TASK
609 /*
610 * The task is finally done with both the stack and thread_info,
611 * so free both.
612 */
613 release_task_stack(tsk);
614#else
615 /*
616 * If the task had a separate stack allocation, it should be gone
617 * by now.
618 */
619 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
620#endif
621 rt_mutex_debug_task_free(tsk);
622 ftrace_graph_exit_task(tsk);
623 arch_release_task_struct(tsk);
624 if (tsk->flags & PF_KTHREAD)
625 free_kthread_struct(tsk);
626 bpf_task_storage_free(tsk);
627 free_task_struct(tsk);
628}
629EXPORT_SYMBOL(free_task);
630
631static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
632{
633 struct file *exe_file;
634
635 exe_file = get_mm_exe_file(oldmm);
636 RCU_INIT_POINTER(mm->exe_file, exe_file);
637 /*
638 * We depend on the oldmm having properly denied write access to the
639 * exe_file already.
640 */
641 if (exe_file && deny_write_access(exe_file))
642 pr_warn_once("deny_write_access() failed in %s\n", __func__);
643}
644
645#ifdef CONFIG_MMU
646static __latent_entropy int dup_mmap(struct mm_struct *mm,
647 struct mm_struct *oldmm)
648{
649 struct vm_area_struct *mpnt, *tmp;
650 int retval;
651 unsigned long charge = 0;
652 LIST_HEAD(uf);
653 VMA_ITERATOR(old_vmi, oldmm, 0);
654 VMA_ITERATOR(vmi, mm, 0);
655
656 uprobe_start_dup_mmap();
657 if (mmap_write_lock_killable(oldmm)) {
658 retval = -EINTR;
659 goto fail_uprobe_end;
660 }
661 flush_cache_dup_mm(oldmm);
662 uprobe_dup_mmap(oldmm, mm);
663 /*
664 * Not linked in yet - no deadlock potential:
665 */
666 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
667
668 /* No ordering required: file already has been exposed. */
669 dup_mm_exe_file(mm, oldmm);
670
671 mm->total_vm = oldmm->total_vm;
672 mm->data_vm = oldmm->data_vm;
673 mm->exec_vm = oldmm->exec_vm;
674 mm->stack_vm = oldmm->stack_vm;
675
676 retval = ksm_fork(mm, oldmm);
677 if (retval)
678 goto out;
679 khugepaged_fork(mm, oldmm);
680
681 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
682 if (retval)
683 goto out;
684
685 mt_clear_in_rcu(vmi.mas.tree);
686 for_each_vma(old_vmi, mpnt) {
687 struct file *file;
688
689 vma_start_write(mpnt);
690 if (mpnt->vm_flags & VM_DONTCOPY) {
691 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
692 continue;
693 }
694 charge = 0;
695 /*
696 * Don't duplicate many vmas if we've been oom-killed (for
697 * example)
698 */
699 if (fatal_signal_pending(current)) {
700 retval = -EINTR;
701 goto loop_out;
702 }
703 if (mpnt->vm_flags & VM_ACCOUNT) {
704 unsigned long len = vma_pages(mpnt);
705
706 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
707 goto fail_nomem;
708 charge = len;
709 }
710 tmp = vm_area_dup(mpnt);
711 if (!tmp)
712 goto fail_nomem;
713 retval = vma_dup_policy(mpnt, tmp);
714 if (retval)
715 goto fail_nomem_policy;
716 tmp->vm_mm = mm;
717 retval = dup_userfaultfd(tmp, &uf);
718 if (retval)
719 goto fail_nomem_anon_vma_fork;
720 if (tmp->vm_flags & VM_WIPEONFORK) {
721 /*
722 * VM_WIPEONFORK gets a clean slate in the child.
723 * Don't prepare anon_vma until fault since we don't
724 * copy page for current vma.
725 */
726 tmp->anon_vma = NULL;
727 } else if (anon_vma_fork(tmp, mpnt))
728 goto fail_nomem_anon_vma_fork;
729 vm_flags_clear(tmp, VM_LOCKED_MASK);
730 file = tmp->vm_file;
731 if (file) {
732 struct address_space *mapping = file->f_mapping;
733
734 get_file(file);
735 i_mmap_lock_write(mapping);
736 if (tmp->vm_flags & VM_SHARED)
737 mapping_allow_writable(mapping);
738 flush_dcache_mmap_lock(mapping);
739 /* insert tmp into the share list, just after mpnt */
740 vma_interval_tree_insert_after(tmp, mpnt,
741 &mapping->i_mmap);
742 flush_dcache_mmap_unlock(mapping);
743 i_mmap_unlock_write(mapping);
744 }
745
746 /*
747 * Copy/update hugetlb private vma information.
748 */
749 if (is_vm_hugetlb_page(tmp))
750 hugetlb_dup_vma_private(tmp);
751
752 /* Link the vma into the MT */
753 if (vma_iter_bulk_store(&vmi, tmp))
754 goto fail_nomem_vmi_store;
755
756 mm->map_count++;
757 if (!(tmp->vm_flags & VM_WIPEONFORK))
758 retval = copy_page_range(tmp, mpnt);
759
760 if (tmp->vm_ops && tmp->vm_ops->open)
761 tmp->vm_ops->open(tmp);
762
763 if (retval)
764 goto loop_out;
765 }
766 /* a new mm has just been created */
767 retval = arch_dup_mmap(oldmm, mm);
768loop_out:
769 vma_iter_free(&vmi);
770 if (!retval)
771 mt_set_in_rcu(vmi.mas.tree);
772out:
773 mmap_write_unlock(mm);
774 flush_tlb_mm(oldmm);
775 mmap_write_unlock(oldmm);
776 dup_userfaultfd_complete(&uf);
777fail_uprobe_end:
778 uprobe_end_dup_mmap();
779 return retval;
780
781fail_nomem_vmi_store:
782 unlink_anon_vmas(tmp);
783fail_nomem_anon_vma_fork:
784 mpol_put(vma_policy(tmp));
785fail_nomem_policy:
786 vm_area_free(tmp);
787fail_nomem:
788 retval = -ENOMEM;
789 vm_unacct_memory(charge);
790 goto loop_out;
791}
792
793static inline int mm_alloc_pgd(struct mm_struct *mm)
794{
795 mm->pgd = pgd_alloc(mm);
796 if (unlikely(!mm->pgd))
797 return -ENOMEM;
798 return 0;
799}
800
801static inline void mm_free_pgd(struct mm_struct *mm)
802{
803 pgd_free(mm, mm->pgd);
804}
805#else
806static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
807{
808 mmap_write_lock(oldmm);
809 dup_mm_exe_file(mm, oldmm);
810 mmap_write_unlock(oldmm);
811 return 0;
812}
813#define mm_alloc_pgd(mm) (0)
814#define mm_free_pgd(mm)
815#endif /* CONFIG_MMU */
816
817static void check_mm(struct mm_struct *mm)
818{
819 int i;
820
821 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
822 "Please make sure 'struct resident_page_types[]' is updated as well");
823
824 for (i = 0; i < NR_MM_COUNTERS; i++) {
825 long x = percpu_counter_sum(&mm->rss_stat[i]);
826
827 if (unlikely(x))
828 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
829 mm, resident_page_types[i], x);
830 }
831
832 if (mm_pgtables_bytes(mm))
833 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
834 mm_pgtables_bytes(mm));
835
836#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
837 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
838#endif
839}
840
841#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
842#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
843
844static void do_check_lazy_tlb(void *arg)
845{
846 struct mm_struct *mm = arg;
847
848 WARN_ON_ONCE(current->active_mm == mm);
849}
850
851static void do_shoot_lazy_tlb(void *arg)
852{
853 struct mm_struct *mm = arg;
854
855 if (current->active_mm == mm) {
856 WARN_ON_ONCE(current->mm);
857 current->active_mm = &init_mm;
858 switch_mm(mm, &init_mm, current);
859 }
860}
861
862static void cleanup_lazy_tlbs(struct mm_struct *mm)
863{
864 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
865 /*
866 * In this case, lazy tlb mms are refounted and would not reach
867 * __mmdrop until all CPUs have switched away and mmdrop()ed.
868 */
869 return;
870 }
871
872 /*
873 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
874 * requires lazy mm users to switch to another mm when the refcount
875 * drops to zero, before the mm is freed. This requires IPIs here to
876 * switch kernel threads to init_mm.
877 *
878 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
879 * switch with the final userspace teardown TLB flush which leaves the
880 * mm lazy on this CPU but no others, reducing the need for additional
881 * IPIs here. There are cases where a final IPI is still required here,
882 * such as the final mmdrop being performed on a different CPU than the
883 * one exiting, or kernel threads using the mm when userspace exits.
884 *
885 * IPI overheads have not found to be expensive, but they could be
886 * reduced in a number of possible ways, for example (roughly
887 * increasing order of complexity):
888 * - The last lazy reference created by exit_mm() could instead switch
889 * to init_mm, however it's probable this will run on the same CPU
890 * immediately afterwards, so this may not reduce IPIs much.
891 * - A batch of mms requiring IPIs could be gathered and freed at once.
892 * - CPUs store active_mm where it can be remotely checked without a
893 * lock, to filter out false-positives in the cpumask.
894 * - After mm_users or mm_count reaches zero, switching away from the
895 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
896 * with some batching or delaying of the final IPIs.
897 * - A delayed freeing and RCU-like quiescing sequence based on mm
898 * switching to avoid IPIs completely.
899 */
900 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
901 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
902 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
903}
904
905/*
906 * Called when the last reference to the mm
907 * is dropped: either by a lazy thread or by
908 * mmput. Free the page directory and the mm.
909 */
910void __mmdrop(struct mm_struct *mm)
911{
912 int i;
913
914 BUG_ON(mm == &init_mm);
915 WARN_ON_ONCE(mm == current->mm);
916
917 /* Ensure no CPUs are using this as their lazy tlb mm */
918 cleanup_lazy_tlbs(mm);
919
920 WARN_ON_ONCE(mm == current->active_mm);
921 mm_free_pgd(mm);
922 destroy_context(mm);
923 mmu_notifier_subscriptions_destroy(mm);
924 check_mm(mm);
925 put_user_ns(mm->user_ns);
926 mm_pasid_drop(mm);
927 mm_destroy_cid(mm);
928
929 for (i = 0; i < NR_MM_COUNTERS; i++)
930 percpu_counter_destroy(&mm->rss_stat[i]);
931 free_mm(mm);
932}
933EXPORT_SYMBOL_GPL(__mmdrop);
934
935static void mmdrop_async_fn(struct work_struct *work)
936{
937 struct mm_struct *mm;
938
939 mm = container_of(work, struct mm_struct, async_put_work);
940 __mmdrop(mm);
941}
942
943static void mmdrop_async(struct mm_struct *mm)
944{
945 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
946 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
947 schedule_work(&mm->async_put_work);
948 }
949}
950
951static inline void free_signal_struct(struct signal_struct *sig)
952{
953 taskstats_tgid_free(sig);
954 sched_autogroup_exit(sig);
955 /*
956 * __mmdrop is not safe to call from softirq context on x86 due to
957 * pgd_dtor so postpone it to the async context
958 */
959 if (sig->oom_mm)
960 mmdrop_async(sig->oom_mm);
961 kmem_cache_free(signal_cachep, sig);
962}
963
964static inline void put_signal_struct(struct signal_struct *sig)
965{
966 if (refcount_dec_and_test(&sig->sigcnt))
967 free_signal_struct(sig);
968}
969
970void __put_task_struct(struct task_struct *tsk)
971{
972 WARN_ON(!tsk->exit_state);
973 WARN_ON(refcount_read(&tsk->usage));
974 WARN_ON(tsk == current);
975
976 io_uring_free(tsk);
977 cgroup_free(tsk);
978 task_numa_free(tsk, true);
979 security_task_free(tsk);
980 exit_creds(tsk);
981 delayacct_tsk_free(tsk);
982 put_signal_struct(tsk->signal);
983 sched_core_free(tsk);
984 free_task(tsk);
985}
986EXPORT_SYMBOL_GPL(__put_task_struct);
987
988void __init __weak arch_task_cache_init(void) { }
989
990/*
991 * set_max_threads
992 */
993static void set_max_threads(unsigned int max_threads_suggested)
994{
995 u64 threads;
996 unsigned long nr_pages = totalram_pages();
997
998 /*
999 * The number of threads shall be limited such that the thread
1000 * structures may only consume a small part of the available memory.
1001 */
1002 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1003 threads = MAX_THREADS;
1004 else
1005 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1006 (u64) THREAD_SIZE * 8UL);
1007
1008 if (threads > max_threads_suggested)
1009 threads = max_threads_suggested;
1010
1011 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1012}
1013
1014#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1015/* Initialized by the architecture: */
1016int arch_task_struct_size __read_mostly;
1017#endif
1018
1019#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1020static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1021{
1022 /* Fetch thread_struct whitelist for the architecture. */
1023 arch_thread_struct_whitelist(offset, size);
1024
1025 /*
1026 * Handle zero-sized whitelist or empty thread_struct, otherwise
1027 * adjust offset to position of thread_struct in task_struct.
1028 */
1029 if (unlikely(*size == 0))
1030 *offset = 0;
1031 else
1032 *offset += offsetof(struct task_struct, thread);
1033}
1034#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
1035
1036void __init fork_init(void)
1037{
1038 int i;
1039#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1040#ifndef ARCH_MIN_TASKALIGN
1041#define ARCH_MIN_TASKALIGN 0
1042#endif
1043 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1044 unsigned long useroffset, usersize;
1045
1046 /* create a slab on which task_structs can be allocated */
1047 task_struct_whitelist(&useroffset, &usersize);
1048 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1049 arch_task_struct_size, align,
1050 SLAB_PANIC|SLAB_ACCOUNT,
1051 useroffset, usersize, NULL);
1052#endif
1053
1054 /* do the arch specific task caches init */
1055 arch_task_cache_init();
1056
1057 set_max_threads(MAX_THREADS);
1058
1059 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1060 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1061 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1062 init_task.signal->rlim[RLIMIT_NPROC];
1063
1064 for (i = 0; i < UCOUNT_COUNTS; i++)
1065 init_user_ns.ucount_max[i] = max_threads/2;
1066
1067 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1068 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1071
1072#ifdef CONFIG_VMAP_STACK
1073 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1074 NULL, free_vm_stack_cache);
1075#endif
1076
1077 scs_init();
1078
1079 lockdep_init_task(&init_task);
1080 uprobes_init();
1081}
1082
1083int __weak arch_dup_task_struct(struct task_struct *dst,
1084 struct task_struct *src)
1085{
1086 *dst = *src;
1087 return 0;
1088}
1089
1090void set_task_stack_end_magic(struct task_struct *tsk)
1091{
1092 unsigned long *stackend;
1093
1094 stackend = end_of_stack(tsk);
1095 *stackend = STACK_END_MAGIC; /* for overflow detection */
1096}
1097
1098static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1099{
1100 struct task_struct *tsk;
1101 int err;
1102
1103 if (node == NUMA_NO_NODE)
1104 node = tsk_fork_get_node(orig);
1105 tsk = alloc_task_struct_node(node);
1106 if (!tsk)
1107 return NULL;
1108
1109 err = arch_dup_task_struct(tsk, orig);
1110 if (err)
1111 goto free_tsk;
1112
1113 err = alloc_thread_stack_node(tsk, node);
1114 if (err)
1115 goto free_tsk;
1116
1117#ifdef CONFIG_THREAD_INFO_IN_TASK
1118 refcount_set(&tsk->stack_refcount, 1);
1119#endif
1120 account_kernel_stack(tsk, 1);
1121
1122 err = scs_prepare(tsk, node);
1123 if (err)
1124 goto free_stack;
1125
1126#ifdef CONFIG_SECCOMP
1127 /*
1128 * We must handle setting up seccomp filters once we're under
1129 * the sighand lock in case orig has changed between now and
1130 * then. Until then, filter must be NULL to avoid messing up
1131 * the usage counts on the error path calling free_task.
1132 */
1133 tsk->seccomp.filter = NULL;
1134#endif
1135
1136 setup_thread_stack(tsk, orig);
1137 clear_user_return_notifier(tsk);
1138 clear_tsk_need_resched(tsk);
1139 set_task_stack_end_magic(tsk);
1140 clear_syscall_work_syscall_user_dispatch(tsk);
1141
1142#ifdef CONFIG_STACKPROTECTOR
1143 tsk->stack_canary = get_random_canary();
1144#endif
1145 if (orig->cpus_ptr == &orig->cpus_mask)
1146 tsk->cpus_ptr = &tsk->cpus_mask;
1147 dup_user_cpus_ptr(tsk, orig, node);
1148
1149 /*
1150 * One for the user space visible state that goes away when reaped.
1151 * One for the scheduler.
1152 */
1153 refcount_set(&tsk->rcu_users, 2);
1154 /* One for the rcu users */
1155 refcount_set(&tsk->usage, 1);
1156#ifdef CONFIG_BLK_DEV_IO_TRACE
1157 tsk->btrace_seq = 0;
1158#endif
1159 tsk->splice_pipe = NULL;
1160 tsk->task_frag.page = NULL;
1161 tsk->wake_q.next = NULL;
1162 tsk->worker_private = NULL;
1163
1164 kcov_task_init(tsk);
1165 kmsan_task_create(tsk);
1166 kmap_local_fork(tsk);
1167
1168#ifdef CONFIG_FAULT_INJECTION
1169 tsk->fail_nth = 0;
1170#endif
1171
1172#ifdef CONFIG_BLK_CGROUP
1173 tsk->throttle_disk = NULL;
1174 tsk->use_memdelay = 0;
1175#endif
1176
1177#ifdef CONFIG_IOMMU_SVA
1178 tsk->pasid_activated = 0;
1179#endif
1180
1181#ifdef CONFIG_MEMCG
1182 tsk->active_memcg = NULL;
1183#endif
1184
1185#ifdef CONFIG_CPU_SUP_INTEL
1186 tsk->reported_split_lock = 0;
1187#endif
1188
1189#ifdef CONFIG_SCHED_MM_CID
1190 tsk->mm_cid = -1;
1191 tsk->last_mm_cid = -1;
1192 tsk->mm_cid_active = 0;
1193 tsk->migrate_from_cpu = -1;
1194#endif
1195 return tsk;
1196
1197free_stack:
1198 exit_task_stack_account(tsk);
1199 free_thread_stack(tsk);
1200free_tsk:
1201 free_task_struct(tsk);
1202 return NULL;
1203}
1204
1205__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1206
1207static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1208
1209static int __init coredump_filter_setup(char *s)
1210{
1211 default_dump_filter =
1212 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1213 MMF_DUMP_FILTER_MASK;
1214 return 1;
1215}
1216
1217__setup("coredump_filter=", coredump_filter_setup);
1218
1219#include <linux/init_task.h>
1220
1221static void mm_init_aio(struct mm_struct *mm)
1222{
1223#ifdef CONFIG_AIO
1224 spin_lock_init(&mm->ioctx_lock);
1225 mm->ioctx_table = NULL;
1226#endif
1227}
1228
1229static __always_inline void mm_clear_owner(struct mm_struct *mm,
1230 struct task_struct *p)
1231{
1232#ifdef CONFIG_MEMCG
1233 if (mm->owner == p)
1234 WRITE_ONCE(mm->owner, NULL);
1235#endif
1236}
1237
1238static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1239{
1240#ifdef CONFIG_MEMCG
1241 mm->owner = p;
1242#endif
1243}
1244
1245static void mm_init_uprobes_state(struct mm_struct *mm)
1246{
1247#ifdef CONFIG_UPROBES
1248 mm->uprobes_state.xol_area = NULL;
1249#endif
1250}
1251
1252static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1253 struct user_namespace *user_ns)
1254{
1255 int i;
1256
1257 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1258 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1259 atomic_set(&mm->mm_users, 1);
1260 atomic_set(&mm->mm_count, 1);
1261 seqcount_init(&mm->write_protect_seq);
1262 mmap_init_lock(mm);
1263 INIT_LIST_HEAD(&mm->mmlist);
1264#ifdef CONFIG_PER_VMA_LOCK
1265 mm->mm_lock_seq = 0;
1266#endif
1267 mm_pgtables_bytes_init(mm);
1268 mm->map_count = 0;
1269 mm->locked_vm = 0;
1270 atomic64_set(&mm->pinned_vm, 0);
1271 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1272 spin_lock_init(&mm->page_table_lock);
1273 spin_lock_init(&mm->arg_lock);
1274 mm_init_cpumask(mm);
1275 mm_init_aio(mm);
1276 mm_init_owner(mm, p);
1277 mm_pasid_init(mm);
1278 RCU_INIT_POINTER(mm->exe_file, NULL);
1279 mmu_notifier_subscriptions_init(mm);
1280 init_tlb_flush_pending(mm);
1281#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1282 mm->pmd_huge_pte = NULL;
1283#endif
1284 mm_init_uprobes_state(mm);
1285 hugetlb_count_init(mm);
1286
1287 if (current->mm) {
1288 mm->flags = current->mm->flags & MMF_INIT_MASK;
1289 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1290 } else {
1291 mm->flags = default_dump_filter;
1292 mm->def_flags = 0;
1293 }
1294
1295 if (mm_alloc_pgd(mm))
1296 goto fail_nopgd;
1297
1298 if (init_new_context(p, mm))
1299 goto fail_nocontext;
1300
1301 if (mm_alloc_cid(mm))
1302 goto fail_cid;
1303
1304 for (i = 0; i < NR_MM_COUNTERS; i++)
1305 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1306 goto fail_pcpu;
1307
1308 mm->user_ns = get_user_ns(user_ns);
1309 lru_gen_init_mm(mm);
1310 return mm;
1311
1312fail_pcpu:
1313 while (i > 0)
1314 percpu_counter_destroy(&mm->rss_stat[--i]);
1315 mm_destroy_cid(mm);
1316fail_cid:
1317 destroy_context(mm);
1318fail_nocontext:
1319 mm_free_pgd(mm);
1320fail_nopgd:
1321 free_mm(mm);
1322 return NULL;
1323}
1324
1325/*
1326 * Allocate and initialize an mm_struct.
1327 */
1328struct mm_struct *mm_alloc(void)
1329{
1330 struct mm_struct *mm;
1331
1332 mm = allocate_mm();
1333 if (!mm)
1334 return NULL;
1335
1336 memset(mm, 0, sizeof(*mm));
1337 return mm_init(mm, current, current_user_ns());
1338}
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_page(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 *
1396 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1397 *
1398 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1399 * invocations: in mmput() nobody alive left, in execve task is single
1400 * threaded.
1401 *
1402 * Can only fail if new_exe_file != NULL.
1403 */
1404int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1405{
1406 struct file *old_exe_file;
1407
1408 /*
1409 * It is safe to dereference the exe_file without RCU as
1410 * this function is only called if nobody else can access
1411 * this mm -- see comment above for justification.
1412 */
1413 old_exe_file = rcu_dereference_raw(mm->exe_file);
1414
1415 if (new_exe_file) {
1416 /*
1417 * We expect the caller (i.e., sys_execve) to already denied
1418 * write access, so this is unlikely to fail.
1419 */
1420 if (unlikely(deny_write_access(new_exe_file)))
1421 return -EACCES;
1422 get_file(new_exe_file);
1423 }
1424 rcu_assign_pointer(mm->exe_file, new_exe_file);
1425 if (old_exe_file) {
1426 allow_write_access(old_exe_file);
1427 fput(old_exe_file);
1428 }
1429 return 0;
1430}
1431
1432/**
1433 * replace_mm_exe_file - replace a reference to the mm's executable file
1434 *
1435 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1436 * dealing with concurrent invocation and without grabbing the mmap lock in
1437 * write mode.
1438 *
1439 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1440 */
1441int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1442{
1443 struct vm_area_struct *vma;
1444 struct file *old_exe_file;
1445 int ret = 0;
1446
1447 /* Forbid mm->exe_file change if old file still mapped. */
1448 old_exe_file = get_mm_exe_file(mm);
1449 if (old_exe_file) {
1450 VMA_ITERATOR(vmi, mm, 0);
1451 mmap_read_lock(mm);
1452 for_each_vma(vmi, vma) {
1453 if (!vma->vm_file)
1454 continue;
1455 if (path_equal(&vma->vm_file->f_path,
1456 &old_exe_file->f_path)) {
1457 ret = -EBUSY;
1458 break;
1459 }
1460 }
1461 mmap_read_unlock(mm);
1462 fput(old_exe_file);
1463 if (ret)
1464 return ret;
1465 }
1466
1467 /* set the new file, lockless */
1468 ret = deny_write_access(new_exe_file);
1469 if (ret)
1470 return -EACCES;
1471 get_file(new_exe_file);
1472
1473 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1474 if (old_exe_file) {
1475 /*
1476 * Don't race with dup_mmap() getting the file and disallowing
1477 * write access while someone might open the file writable.
1478 */
1479 mmap_read_lock(mm);
1480 allow_write_access(old_exe_file);
1481 fput(old_exe_file);
1482 mmap_read_unlock(mm);
1483 }
1484 return 0;
1485}
1486
1487/**
1488 * get_mm_exe_file - acquire a reference to the mm's executable file
1489 *
1490 * Returns %NULL if mm has no associated executable file.
1491 * User must release file via fput().
1492 */
1493struct file *get_mm_exe_file(struct mm_struct *mm)
1494{
1495 struct file *exe_file;
1496
1497 rcu_read_lock();
1498 exe_file = rcu_dereference(mm->exe_file);
1499 if (exe_file && !get_file_rcu(exe_file))
1500 exe_file = NULL;
1501 rcu_read_unlock();
1502 return exe_file;
1503}
1504
1505/**
1506 * get_task_exe_file - acquire a reference to the task's executable file
1507 *
1508 * Returns %NULL if task's mm (if any) has no associated executable file or
1509 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1510 * User must release file via fput().
1511 */
1512struct file *get_task_exe_file(struct task_struct *task)
1513{
1514 struct file *exe_file = NULL;
1515 struct mm_struct *mm;
1516
1517 task_lock(task);
1518 mm = task->mm;
1519 if (mm) {
1520 if (!(task->flags & PF_KTHREAD))
1521 exe_file = get_mm_exe_file(mm);
1522 }
1523 task_unlock(task);
1524 return exe_file;
1525}
1526
1527/**
1528 * get_task_mm - acquire a reference to the task's mm
1529 *
1530 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1531 * this kernel workthread has transiently adopted a user mm with use_mm,
1532 * to do its AIO) is not set and if so returns a reference to it, after
1533 * bumping up the use count. User must release the mm via mmput()
1534 * after use. Typically used by /proc and ptrace.
1535 */
1536struct mm_struct *get_task_mm(struct task_struct *task)
1537{
1538 struct mm_struct *mm;
1539
1540 task_lock(task);
1541 mm = task->mm;
1542 if (mm) {
1543 if (task->flags & PF_KTHREAD)
1544 mm = NULL;
1545 else
1546 mmget(mm);
1547 }
1548 task_unlock(task);
1549 return mm;
1550}
1551EXPORT_SYMBOL_GPL(get_task_mm);
1552
1553struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1554{
1555 struct mm_struct *mm;
1556 int err;
1557
1558 err = down_read_killable(&task->signal->exec_update_lock);
1559 if (err)
1560 return ERR_PTR(err);
1561
1562 mm = get_task_mm(task);
1563 if (mm && mm != current->mm &&
1564 !ptrace_may_access(task, mode)) {
1565 mmput(mm);
1566 mm = ERR_PTR(-EACCES);
1567 }
1568 up_read(&task->signal->exec_update_lock);
1569
1570 return mm;
1571}
1572
1573static void complete_vfork_done(struct task_struct *tsk)
1574{
1575 struct completion *vfork;
1576
1577 task_lock(tsk);
1578 vfork = tsk->vfork_done;
1579 if (likely(vfork)) {
1580 tsk->vfork_done = NULL;
1581 complete(vfork);
1582 }
1583 task_unlock(tsk);
1584}
1585
1586static int wait_for_vfork_done(struct task_struct *child,
1587 struct completion *vfork)
1588{
1589 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1590 int killed;
1591
1592 cgroup_enter_frozen();
1593 killed = wait_for_completion_state(vfork, state);
1594 cgroup_leave_frozen(false);
1595
1596 if (killed) {
1597 task_lock(child);
1598 child->vfork_done = NULL;
1599 task_unlock(child);
1600 }
1601
1602 put_task_struct(child);
1603 return killed;
1604}
1605
1606/* Please note the differences between mmput and mm_release.
1607 * mmput is called whenever we stop holding onto a mm_struct,
1608 * error success whatever.
1609 *
1610 * mm_release is called after a mm_struct has been removed
1611 * from the current process.
1612 *
1613 * This difference is important for error handling, when we
1614 * only half set up a mm_struct for a new process and need to restore
1615 * the old one. Because we mmput the new mm_struct before
1616 * restoring the old one. . .
1617 * Eric Biederman 10 January 1998
1618 */
1619static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1620{
1621 uprobe_free_utask(tsk);
1622
1623 /* Get rid of any cached register state */
1624 deactivate_mm(tsk, mm);
1625
1626 /*
1627 * Signal userspace if we're not exiting with a core dump
1628 * because we want to leave the value intact for debugging
1629 * purposes.
1630 */
1631 if (tsk->clear_child_tid) {
1632 if (atomic_read(&mm->mm_users) > 1) {
1633 /*
1634 * We don't check the error code - if userspace has
1635 * not set up a proper pointer then tough luck.
1636 */
1637 put_user(0, tsk->clear_child_tid);
1638 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1639 1, NULL, NULL, 0, 0);
1640 }
1641 tsk->clear_child_tid = NULL;
1642 }
1643
1644 /*
1645 * All done, finally we can wake up parent and return this mm to him.
1646 * Also kthread_stop() uses this completion for synchronization.
1647 */
1648 if (tsk->vfork_done)
1649 complete_vfork_done(tsk);
1650}
1651
1652void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1653{
1654 futex_exit_release(tsk);
1655 mm_release(tsk, mm);
1656}
1657
1658void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1659{
1660 futex_exec_release(tsk);
1661 mm_release(tsk, mm);
1662}
1663
1664/**
1665 * dup_mm() - duplicates an existing mm structure
1666 * @tsk: the task_struct with which the new mm will be associated.
1667 * @oldmm: the mm to duplicate.
1668 *
1669 * Allocates a new mm structure and duplicates the provided @oldmm structure
1670 * content into it.
1671 *
1672 * Return: the duplicated mm or NULL on failure.
1673 */
1674static struct mm_struct *dup_mm(struct task_struct *tsk,
1675 struct mm_struct *oldmm)
1676{
1677 struct mm_struct *mm;
1678 int err;
1679
1680 mm = allocate_mm();
1681 if (!mm)
1682 goto fail_nomem;
1683
1684 memcpy(mm, oldmm, sizeof(*mm));
1685
1686 if (!mm_init(mm, tsk, mm->user_ns))
1687 goto fail_nomem;
1688
1689 err = dup_mmap(mm, oldmm);
1690 if (err)
1691 goto free_pt;
1692
1693 mm->hiwater_rss = get_mm_rss(mm);
1694 mm->hiwater_vm = mm->total_vm;
1695
1696 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1697 goto free_pt;
1698
1699 return mm;
1700
1701free_pt:
1702 /* don't put binfmt in mmput, we haven't got module yet */
1703 mm->binfmt = NULL;
1704 mm_init_owner(mm, NULL);
1705 mmput(mm);
1706
1707fail_nomem:
1708 return NULL;
1709}
1710
1711static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1712{
1713 struct mm_struct *mm, *oldmm;
1714
1715 tsk->min_flt = tsk->maj_flt = 0;
1716 tsk->nvcsw = tsk->nivcsw = 0;
1717#ifdef CONFIG_DETECT_HUNG_TASK
1718 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1719 tsk->last_switch_time = 0;
1720#endif
1721
1722 tsk->mm = NULL;
1723 tsk->active_mm = NULL;
1724
1725 /*
1726 * Are we cloning a kernel thread?
1727 *
1728 * We need to steal a active VM for that..
1729 */
1730 oldmm = current->mm;
1731 if (!oldmm)
1732 return 0;
1733
1734 if (clone_flags & CLONE_VM) {
1735 mmget(oldmm);
1736 mm = oldmm;
1737 } else {
1738 mm = dup_mm(tsk, current->mm);
1739 if (!mm)
1740 return -ENOMEM;
1741 }
1742
1743 tsk->mm = mm;
1744 tsk->active_mm = mm;
1745 sched_mm_cid_fork(tsk);
1746 return 0;
1747}
1748
1749static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1750{
1751 struct fs_struct *fs = current->fs;
1752 if (clone_flags & CLONE_FS) {
1753 /* tsk->fs is already what we want */
1754 spin_lock(&fs->lock);
1755 if (fs->in_exec) {
1756 spin_unlock(&fs->lock);
1757 return -EAGAIN;
1758 }
1759 fs->users++;
1760 spin_unlock(&fs->lock);
1761 return 0;
1762 }
1763 tsk->fs = copy_fs_struct(fs);
1764 if (!tsk->fs)
1765 return -ENOMEM;
1766 return 0;
1767}
1768
1769static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1770 int no_files)
1771{
1772 struct files_struct *oldf, *newf;
1773 int error = 0;
1774
1775 /*
1776 * A background process may not have any files ...
1777 */
1778 oldf = current->files;
1779 if (!oldf)
1780 goto out;
1781
1782 if (no_files) {
1783 tsk->files = NULL;
1784 goto out;
1785 }
1786
1787 if (clone_flags & CLONE_FILES) {
1788 atomic_inc(&oldf->count);
1789 goto out;
1790 }
1791
1792 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1793 if (!newf)
1794 goto out;
1795
1796 tsk->files = newf;
1797 error = 0;
1798out:
1799 return error;
1800}
1801
1802static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1803{
1804 struct sighand_struct *sig;
1805
1806 if (clone_flags & CLONE_SIGHAND) {
1807 refcount_inc(&current->sighand->count);
1808 return 0;
1809 }
1810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1811 RCU_INIT_POINTER(tsk->sighand, sig);
1812 if (!sig)
1813 return -ENOMEM;
1814
1815 refcount_set(&sig->count, 1);
1816 spin_lock_irq(&current->sighand->siglock);
1817 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1818 spin_unlock_irq(&current->sighand->siglock);
1819
1820 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1821 if (clone_flags & CLONE_CLEAR_SIGHAND)
1822 flush_signal_handlers(tsk, 0);
1823
1824 return 0;
1825}
1826
1827void __cleanup_sighand(struct sighand_struct *sighand)
1828{
1829 if (refcount_dec_and_test(&sighand->count)) {
1830 signalfd_cleanup(sighand);
1831 /*
1832 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1833 * without an RCU grace period, see __lock_task_sighand().
1834 */
1835 kmem_cache_free(sighand_cachep, sighand);
1836 }
1837}
1838
1839/*
1840 * Initialize POSIX timer handling for a thread group.
1841 */
1842static void posix_cpu_timers_init_group(struct signal_struct *sig)
1843{
1844 struct posix_cputimers *pct = &sig->posix_cputimers;
1845 unsigned long cpu_limit;
1846
1847 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1848 posix_cputimers_group_init(pct, cpu_limit);
1849}
1850
1851static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1852{
1853 struct signal_struct *sig;
1854
1855 if (clone_flags & CLONE_THREAD)
1856 return 0;
1857
1858 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1859 tsk->signal = sig;
1860 if (!sig)
1861 return -ENOMEM;
1862
1863 sig->nr_threads = 1;
1864 sig->quick_threads = 1;
1865 atomic_set(&sig->live, 1);
1866 refcount_set(&sig->sigcnt, 1);
1867
1868 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1869 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1870 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1871
1872 init_waitqueue_head(&sig->wait_chldexit);
1873 sig->curr_target = tsk;
1874 init_sigpending(&sig->shared_pending);
1875 INIT_HLIST_HEAD(&sig->multiprocess);
1876 seqlock_init(&sig->stats_lock);
1877 prev_cputime_init(&sig->prev_cputime);
1878
1879#ifdef CONFIG_POSIX_TIMERS
1880 INIT_LIST_HEAD(&sig->posix_timers);
1881 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1882 sig->real_timer.function = it_real_fn;
1883#endif
1884
1885 task_lock(current->group_leader);
1886 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1887 task_unlock(current->group_leader);
1888
1889 posix_cpu_timers_init_group(sig);
1890
1891 tty_audit_fork(sig);
1892 sched_autogroup_fork(sig);
1893
1894 sig->oom_score_adj = current->signal->oom_score_adj;
1895 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1896
1897 mutex_init(&sig->cred_guard_mutex);
1898 init_rwsem(&sig->exec_update_lock);
1899
1900 return 0;
1901}
1902
1903static void copy_seccomp(struct task_struct *p)
1904{
1905#ifdef CONFIG_SECCOMP
1906 /*
1907 * Must be called with sighand->lock held, which is common to
1908 * all threads in the group. Holding cred_guard_mutex is not
1909 * needed because this new task is not yet running and cannot
1910 * be racing exec.
1911 */
1912 assert_spin_locked(&current->sighand->siglock);
1913
1914 /* Ref-count the new filter user, and assign it. */
1915 get_seccomp_filter(current);
1916 p->seccomp = current->seccomp;
1917
1918 /*
1919 * Explicitly enable no_new_privs here in case it got set
1920 * between the task_struct being duplicated and holding the
1921 * sighand lock. The seccomp state and nnp must be in sync.
1922 */
1923 if (task_no_new_privs(current))
1924 task_set_no_new_privs(p);
1925
1926 /*
1927 * If the parent gained a seccomp mode after copying thread
1928 * flags and between before we held the sighand lock, we have
1929 * to manually enable the seccomp thread flag here.
1930 */
1931 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1932 set_task_syscall_work(p, SECCOMP);
1933#endif
1934}
1935
1936SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1937{
1938 current->clear_child_tid = tidptr;
1939
1940 return task_pid_vnr(current);
1941}
1942
1943static void rt_mutex_init_task(struct task_struct *p)
1944{
1945 raw_spin_lock_init(&p->pi_lock);
1946#ifdef CONFIG_RT_MUTEXES
1947 p->pi_waiters = RB_ROOT_CACHED;
1948 p->pi_top_task = NULL;
1949 p->pi_blocked_on = NULL;
1950#endif
1951}
1952
1953static inline void init_task_pid_links(struct task_struct *task)
1954{
1955 enum pid_type type;
1956
1957 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1958 INIT_HLIST_NODE(&task->pid_links[type]);
1959}
1960
1961static inline void
1962init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1963{
1964 if (type == PIDTYPE_PID)
1965 task->thread_pid = pid;
1966 else
1967 task->signal->pids[type] = pid;
1968}
1969
1970static inline void rcu_copy_process(struct task_struct *p)
1971{
1972#ifdef CONFIG_PREEMPT_RCU
1973 p->rcu_read_lock_nesting = 0;
1974 p->rcu_read_unlock_special.s = 0;
1975 p->rcu_blocked_node = NULL;
1976 INIT_LIST_HEAD(&p->rcu_node_entry);
1977#endif /* #ifdef CONFIG_PREEMPT_RCU */
1978#ifdef CONFIG_TASKS_RCU
1979 p->rcu_tasks_holdout = false;
1980 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1981 p->rcu_tasks_idle_cpu = -1;
1982#endif /* #ifdef CONFIG_TASKS_RCU */
1983#ifdef CONFIG_TASKS_TRACE_RCU
1984 p->trc_reader_nesting = 0;
1985 p->trc_reader_special.s = 0;
1986 INIT_LIST_HEAD(&p->trc_holdout_list);
1987 INIT_LIST_HEAD(&p->trc_blkd_node);
1988#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1989}
1990
1991struct pid *pidfd_pid(const struct file *file)
1992{
1993 if (file->f_op == &pidfd_fops)
1994 return file->private_data;
1995
1996 return ERR_PTR(-EBADF);
1997}
1998
1999static int pidfd_release(struct inode *inode, struct file *file)
2000{
2001 struct pid *pid = file->private_data;
2002
2003 file->private_data = NULL;
2004 put_pid(pid);
2005 return 0;
2006}
2007
2008#ifdef CONFIG_PROC_FS
2009/**
2010 * pidfd_show_fdinfo - print information about a pidfd
2011 * @m: proc fdinfo file
2012 * @f: file referencing a pidfd
2013 *
2014 * Pid:
2015 * This function will print the pid that a given pidfd refers to in the
2016 * pid namespace of the procfs instance.
2017 * If the pid namespace of the process is not a descendant of the pid
2018 * namespace of the procfs instance 0 will be shown as its pid. This is
2019 * similar to calling getppid() on a process whose parent is outside of
2020 * its pid namespace.
2021 *
2022 * NSpid:
2023 * If pid namespaces are supported then this function will also print
2024 * the pid of a given pidfd refers to for all descendant pid namespaces
2025 * starting from the current pid namespace of the instance, i.e. the
2026 * Pid field and the first entry in the NSpid field will be identical.
2027 * If the pid namespace of the process is not a descendant of the pid
2028 * namespace of the procfs instance 0 will be shown as its first NSpid
2029 * entry and no others will be shown.
2030 * Note that this differs from the Pid and NSpid fields in
2031 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2032 * the pid namespace of the procfs instance. The difference becomes
2033 * obvious when sending around a pidfd between pid namespaces from a
2034 * different branch of the tree, i.e. where no ancestral relation is
2035 * present between the pid namespaces:
2036 * - create two new pid namespaces ns1 and ns2 in the initial pid
2037 * namespace (also take care to create new mount namespaces in the
2038 * new pid namespace and mount procfs)
2039 * - create a process with a pidfd in ns1
2040 * - send pidfd from ns1 to ns2
2041 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2042 * have exactly one entry, which is 0
2043 */
2044static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2045{
2046 struct pid *pid = f->private_data;
2047 struct pid_namespace *ns;
2048 pid_t nr = -1;
2049
2050 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2051 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2052 nr = pid_nr_ns(pid, ns);
2053 }
2054
2055 seq_put_decimal_ll(m, "Pid:\t", nr);
2056
2057#ifdef CONFIG_PID_NS
2058 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2059 if (nr > 0) {
2060 int i;
2061
2062 /* If nr is non-zero it means that 'pid' is valid and that
2063 * ns, i.e. the pid namespace associated with the procfs
2064 * instance, is in the pid namespace hierarchy of pid.
2065 * Start at one below the already printed level.
2066 */
2067 for (i = ns->level + 1; i <= pid->level; i++)
2068 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2069 }
2070#endif
2071 seq_putc(m, '\n');
2072}
2073#endif
2074
2075/*
2076 * Poll support for process exit notification.
2077 */
2078static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2079{
2080 struct pid *pid = file->private_data;
2081 __poll_t poll_flags = 0;
2082
2083 poll_wait(file, &pid->wait_pidfd, pts);
2084
2085 /*
2086 * Inform pollers only when the whole thread group exits.
2087 * If the thread group leader exits before all other threads in the
2088 * group, then poll(2) should block, similar to the wait(2) family.
2089 */
2090 if (thread_group_exited(pid))
2091 poll_flags = EPOLLIN | EPOLLRDNORM;
2092
2093 return poll_flags;
2094}
2095
2096const struct file_operations pidfd_fops = {
2097 .release = pidfd_release,
2098 .poll = pidfd_poll,
2099#ifdef CONFIG_PROC_FS
2100 .show_fdinfo = pidfd_show_fdinfo,
2101#endif
2102};
2103
2104/**
2105 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2106 * @pid: the struct pid for which to create a pidfd
2107 * @flags: flags of the new @pidfd
2108 * @pidfd: the pidfd to return
2109 *
2110 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2111 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2112
2113 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2114 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2115 * pidfd file are prepared.
2116 *
2117 * If this function returns successfully the caller is responsible to either
2118 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2119 * order to install the pidfd into its file descriptor table or they must use
2120 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2121 * respectively.
2122 *
2123 * This function is useful when a pidfd must already be reserved but there
2124 * might still be points of failure afterwards and the caller wants to ensure
2125 * that no pidfd is leaked into its file descriptor table.
2126 *
2127 * Return: On success, a reserved pidfd is returned from the function and a new
2128 * pidfd file is returned in the last argument to the function. On
2129 * error, a negative error code is returned from the function and the
2130 * last argument remains unchanged.
2131 */
2132static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2133{
2134 int pidfd;
2135 struct file *pidfd_file;
2136
2137 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2138 return -EINVAL;
2139
2140 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2141 if (pidfd < 0)
2142 return pidfd;
2143
2144 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2145 flags | O_RDWR | O_CLOEXEC);
2146 if (IS_ERR(pidfd_file)) {
2147 put_unused_fd(pidfd);
2148 return PTR_ERR(pidfd_file);
2149 }
2150 get_pid(pid); /* held by pidfd_file now */
2151 *ret = pidfd_file;
2152 return pidfd;
2153}
2154
2155/**
2156 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2157 * @pid: the struct pid for which to create a pidfd
2158 * @flags: flags of the new @pidfd
2159 * @pidfd: the pidfd to return
2160 *
2161 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2162 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2163 *
2164 * The helper verifies that @pid is used as a thread group leader.
2165 *
2166 * If this function returns successfully the caller is responsible to either
2167 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2168 * order to install the pidfd into its file descriptor table or they must use
2169 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2170 * respectively.
2171 *
2172 * This function is useful when a pidfd must already be reserved but there
2173 * might still be points of failure afterwards and the caller wants to ensure
2174 * that no pidfd is leaked into its file descriptor table.
2175 *
2176 * Return: On success, a reserved pidfd is returned from the function and a new
2177 * pidfd file is returned in the last argument to the function. On
2178 * error, a negative error code is returned from the function and the
2179 * last argument remains unchanged.
2180 */
2181int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2182{
2183 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2184 return -EINVAL;
2185
2186 return __pidfd_prepare(pid, flags, ret);
2187}
2188
2189static void __delayed_free_task(struct rcu_head *rhp)
2190{
2191 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2192
2193 free_task(tsk);
2194}
2195
2196static __always_inline void delayed_free_task(struct task_struct *tsk)
2197{
2198 if (IS_ENABLED(CONFIG_MEMCG))
2199 call_rcu(&tsk->rcu, __delayed_free_task);
2200 else
2201 free_task(tsk);
2202}
2203
2204static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2205{
2206 /* Skip if kernel thread */
2207 if (!tsk->mm)
2208 return;
2209
2210 /* Skip if spawning a thread or using vfork */
2211 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2212 return;
2213
2214 /* We need to synchronize with __set_oom_adj */
2215 mutex_lock(&oom_adj_mutex);
2216 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2217 /* Update the values in case they were changed after copy_signal */
2218 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2219 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2220 mutex_unlock(&oom_adj_mutex);
2221}
2222
2223#ifdef CONFIG_RV
2224static void rv_task_fork(struct task_struct *p)
2225{
2226 int i;
2227
2228 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2229 p->rv[i].da_mon.monitoring = false;
2230}
2231#else
2232#define rv_task_fork(p) do {} while (0)
2233#endif
2234
2235/*
2236 * This creates a new process as a copy of the old one,
2237 * but does not actually start it yet.
2238 *
2239 * It copies the registers, and all the appropriate
2240 * parts of the process environment (as per the clone
2241 * flags). The actual kick-off is left to the caller.
2242 */
2243__latent_entropy struct task_struct *copy_process(
2244 struct pid *pid,
2245 int trace,
2246 int node,
2247 struct kernel_clone_args *args)
2248{
2249 int pidfd = -1, retval;
2250 struct task_struct *p;
2251 struct multiprocess_signals delayed;
2252 struct file *pidfile = NULL;
2253 const u64 clone_flags = args->flags;
2254 struct nsproxy *nsp = current->nsproxy;
2255
2256 /*
2257 * Don't allow sharing the root directory with processes in a different
2258 * namespace
2259 */
2260 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2261 return ERR_PTR(-EINVAL);
2262
2263 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2264 return ERR_PTR(-EINVAL);
2265
2266 /*
2267 * Thread groups must share signals as well, and detached threads
2268 * can only be started up within the thread group.
2269 */
2270 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2271 return ERR_PTR(-EINVAL);
2272
2273 /*
2274 * Shared signal handlers imply shared VM. By way of the above,
2275 * thread groups also imply shared VM. Blocking this case allows
2276 * for various simplifications in other code.
2277 */
2278 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2279 return ERR_PTR(-EINVAL);
2280
2281 /*
2282 * Siblings of global init remain as zombies on exit since they are
2283 * not reaped by their parent (swapper). To solve this and to avoid
2284 * multi-rooted process trees, prevent global and container-inits
2285 * from creating siblings.
2286 */
2287 if ((clone_flags & CLONE_PARENT) &&
2288 current->signal->flags & SIGNAL_UNKILLABLE)
2289 return ERR_PTR(-EINVAL);
2290
2291 /*
2292 * If the new process will be in a different pid or user namespace
2293 * do not allow it to share a thread group with the forking task.
2294 */
2295 if (clone_flags & CLONE_THREAD) {
2296 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2297 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2298 return ERR_PTR(-EINVAL);
2299 }
2300
2301 if (clone_flags & CLONE_PIDFD) {
2302 /*
2303 * - CLONE_DETACHED is blocked so that we can potentially
2304 * reuse it later for CLONE_PIDFD.
2305 * - CLONE_THREAD is blocked until someone really needs it.
2306 */
2307 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2308 return ERR_PTR(-EINVAL);
2309 }
2310
2311 /*
2312 * Force any signals received before this point to be delivered
2313 * before the fork happens. Collect up signals sent to multiple
2314 * processes that happen during the fork and delay them so that
2315 * they appear to happen after the fork.
2316 */
2317 sigemptyset(&delayed.signal);
2318 INIT_HLIST_NODE(&delayed.node);
2319
2320 spin_lock_irq(&current->sighand->siglock);
2321 if (!(clone_flags & CLONE_THREAD))
2322 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2323 recalc_sigpending();
2324 spin_unlock_irq(&current->sighand->siglock);
2325 retval = -ERESTARTNOINTR;
2326 if (task_sigpending(current))
2327 goto fork_out;
2328
2329 retval = -ENOMEM;
2330 p = dup_task_struct(current, node);
2331 if (!p)
2332 goto fork_out;
2333 p->flags &= ~PF_KTHREAD;
2334 if (args->kthread)
2335 p->flags |= PF_KTHREAD;
2336 if (args->user_worker) {
2337 /*
2338 * Mark us a user worker, and block any signal that isn't
2339 * fatal or STOP
2340 */
2341 p->flags |= PF_USER_WORKER;
2342 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2343 }
2344 if (args->io_thread)
2345 p->flags |= PF_IO_WORKER;
2346
2347 if (args->name)
2348 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2349
2350 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2351 /*
2352 * Clear TID on mm_release()?
2353 */
2354 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2355
2356 ftrace_graph_init_task(p);
2357
2358 rt_mutex_init_task(p);
2359
2360 lockdep_assert_irqs_enabled();
2361#ifdef CONFIG_PROVE_LOCKING
2362 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2363#endif
2364 retval = copy_creds(p, clone_flags);
2365 if (retval < 0)
2366 goto bad_fork_free;
2367
2368 retval = -EAGAIN;
2369 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2370 if (p->real_cred->user != INIT_USER &&
2371 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2372 goto bad_fork_cleanup_count;
2373 }
2374 current->flags &= ~PF_NPROC_EXCEEDED;
2375
2376 /*
2377 * If multiple threads are within copy_process(), then this check
2378 * triggers too late. This doesn't hurt, the check is only there
2379 * to stop root fork bombs.
2380 */
2381 retval = -EAGAIN;
2382 if (data_race(nr_threads >= max_threads))
2383 goto bad_fork_cleanup_count;
2384
2385 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2386 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2387 p->flags |= PF_FORKNOEXEC;
2388 INIT_LIST_HEAD(&p->children);
2389 INIT_LIST_HEAD(&p->sibling);
2390 rcu_copy_process(p);
2391 p->vfork_done = NULL;
2392 spin_lock_init(&p->alloc_lock);
2393
2394 init_sigpending(&p->pending);
2395
2396 p->utime = p->stime = p->gtime = 0;
2397#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2398 p->utimescaled = p->stimescaled = 0;
2399#endif
2400 prev_cputime_init(&p->prev_cputime);
2401
2402#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2403 seqcount_init(&p->vtime.seqcount);
2404 p->vtime.starttime = 0;
2405 p->vtime.state = VTIME_INACTIVE;
2406#endif
2407
2408#ifdef CONFIG_IO_URING
2409 p->io_uring = NULL;
2410#endif
2411
2412#if defined(SPLIT_RSS_COUNTING)
2413 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2414#endif
2415
2416 p->default_timer_slack_ns = current->timer_slack_ns;
2417
2418#ifdef CONFIG_PSI
2419 p->psi_flags = 0;
2420#endif
2421
2422 task_io_accounting_init(&p->ioac);
2423 acct_clear_integrals(p);
2424
2425 posix_cputimers_init(&p->posix_cputimers);
2426
2427 p->io_context = NULL;
2428 audit_set_context(p, NULL);
2429 cgroup_fork(p);
2430 if (args->kthread) {
2431 if (!set_kthread_struct(p))
2432 goto bad_fork_cleanup_delayacct;
2433 }
2434#ifdef CONFIG_NUMA
2435 p->mempolicy = mpol_dup(p->mempolicy);
2436 if (IS_ERR(p->mempolicy)) {
2437 retval = PTR_ERR(p->mempolicy);
2438 p->mempolicy = NULL;
2439 goto bad_fork_cleanup_delayacct;
2440 }
2441#endif
2442#ifdef CONFIG_CPUSETS
2443 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2444 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2445 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2446#endif
2447#ifdef CONFIG_TRACE_IRQFLAGS
2448 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2449 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2450 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2451 p->softirqs_enabled = 1;
2452 p->softirq_context = 0;
2453#endif
2454
2455 p->pagefault_disabled = 0;
2456
2457#ifdef CONFIG_LOCKDEP
2458 lockdep_init_task(p);
2459#endif
2460
2461#ifdef CONFIG_DEBUG_MUTEXES
2462 p->blocked_on = NULL; /* not blocked yet */
2463#endif
2464#ifdef CONFIG_BCACHE
2465 p->sequential_io = 0;
2466 p->sequential_io_avg = 0;
2467#endif
2468#ifdef CONFIG_BPF_SYSCALL
2469 RCU_INIT_POINTER(p->bpf_storage, NULL);
2470 p->bpf_ctx = NULL;
2471#endif
2472
2473 /* Perform scheduler related setup. Assign this task to a CPU. */
2474 retval = sched_fork(clone_flags, p);
2475 if (retval)
2476 goto bad_fork_cleanup_policy;
2477
2478 retval = perf_event_init_task(p, clone_flags);
2479 if (retval)
2480 goto bad_fork_cleanup_policy;
2481 retval = audit_alloc(p);
2482 if (retval)
2483 goto bad_fork_cleanup_perf;
2484 /* copy all the process information */
2485 shm_init_task(p);
2486 retval = security_task_alloc(p, clone_flags);
2487 if (retval)
2488 goto bad_fork_cleanup_audit;
2489 retval = copy_semundo(clone_flags, p);
2490 if (retval)
2491 goto bad_fork_cleanup_security;
2492 retval = copy_files(clone_flags, p, args->no_files);
2493 if (retval)
2494 goto bad_fork_cleanup_semundo;
2495 retval = copy_fs(clone_flags, p);
2496 if (retval)
2497 goto bad_fork_cleanup_files;
2498 retval = copy_sighand(clone_flags, p);
2499 if (retval)
2500 goto bad_fork_cleanup_fs;
2501 retval = copy_signal(clone_flags, p);
2502 if (retval)
2503 goto bad_fork_cleanup_sighand;
2504 retval = copy_mm(clone_flags, p);
2505 if (retval)
2506 goto bad_fork_cleanup_signal;
2507 retval = copy_namespaces(clone_flags, p);
2508 if (retval)
2509 goto bad_fork_cleanup_mm;
2510 retval = copy_io(clone_flags, p);
2511 if (retval)
2512 goto bad_fork_cleanup_namespaces;
2513 retval = copy_thread(p, args);
2514 if (retval)
2515 goto bad_fork_cleanup_io;
2516
2517 stackleak_task_init(p);
2518
2519 if (pid != &init_struct_pid) {
2520 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2521 args->set_tid_size);
2522 if (IS_ERR(pid)) {
2523 retval = PTR_ERR(pid);
2524 goto bad_fork_cleanup_thread;
2525 }
2526 }
2527
2528 /*
2529 * This has to happen after we've potentially unshared the file
2530 * descriptor table (so that the pidfd doesn't leak into the child
2531 * if the fd table isn't shared).
2532 */
2533 if (clone_flags & CLONE_PIDFD) {
2534 /* Note that no task has been attached to @pid yet. */
2535 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2536 if (retval < 0)
2537 goto bad_fork_free_pid;
2538 pidfd = retval;
2539
2540 retval = put_user(pidfd, args->pidfd);
2541 if (retval)
2542 goto bad_fork_put_pidfd;
2543 }
2544
2545#ifdef CONFIG_BLOCK
2546 p->plug = NULL;
2547#endif
2548 futex_init_task(p);
2549
2550 /*
2551 * sigaltstack should be cleared when sharing the same VM
2552 */
2553 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2554 sas_ss_reset(p);
2555
2556 /*
2557 * Syscall tracing and stepping should be turned off in the
2558 * child regardless of CLONE_PTRACE.
2559 */
2560 user_disable_single_step(p);
2561 clear_task_syscall_work(p, SYSCALL_TRACE);
2562#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2563 clear_task_syscall_work(p, SYSCALL_EMU);
2564#endif
2565 clear_tsk_latency_tracing(p);
2566
2567 /* ok, now we should be set up.. */
2568 p->pid = pid_nr(pid);
2569 if (clone_flags & CLONE_THREAD) {
2570 p->group_leader = current->group_leader;
2571 p->tgid = current->tgid;
2572 } else {
2573 p->group_leader = p;
2574 p->tgid = p->pid;
2575 }
2576
2577 p->nr_dirtied = 0;
2578 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2579 p->dirty_paused_when = 0;
2580
2581 p->pdeath_signal = 0;
2582 INIT_LIST_HEAD(&p->thread_group);
2583 p->task_works = NULL;
2584 clear_posix_cputimers_work(p);
2585
2586#ifdef CONFIG_KRETPROBES
2587 p->kretprobe_instances.first = NULL;
2588#endif
2589#ifdef CONFIG_RETHOOK
2590 p->rethooks.first = NULL;
2591#endif
2592
2593 /*
2594 * Ensure that the cgroup subsystem policies allow the new process to be
2595 * forked. It should be noted that the new process's css_set can be changed
2596 * between here and cgroup_post_fork() if an organisation operation is in
2597 * progress.
2598 */
2599 retval = cgroup_can_fork(p, args);
2600 if (retval)
2601 goto bad_fork_put_pidfd;
2602
2603 /*
2604 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2605 * the new task on the correct runqueue. All this *before* the task
2606 * becomes visible.
2607 *
2608 * This isn't part of ->can_fork() because while the re-cloning is
2609 * cgroup specific, it unconditionally needs to place the task on a
2610 * runqueue.
2611 */
2612 sched_cgroup_fork(p, args);
2613
2614 /*
2615 * From this point on we must avoid any synchronous user-space
2616 * communication until we take the tasklist-lock. In particular, we do
2617 * not want user-space to be able to predict the process start-time by
2618 * stalling fork(2) after we recorded the start_time but before it is
2619 * visible to the system.
2620 */
2621
2622 p->start_time = ktime_get_ns();
2623 p->start_boottime = ktime_get_boottime_ns();
2624
2625 /*
2626 * Make it visible to the rest of the system, but dont wake it up yet.
2627 * Need tasklist lock for parent etc handling!
2628 */
2629 write_lock_irq(&tasklist_lock);
2630
2631 /* CLONE_PARENT re-uses the old parent */
2632 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2633 p->real_parent = current->real_parent;
2634 p->parent_exec_id = current->parent_exec_id;
2635 if (clone_flags & CLONE_THREAD)
2636 p->exit_signal = -1;
2637 else
2638 p->exit_signal = current->group_leader->exit_signal;
2639 } else {
2640 p->real_parent = current;
2641 p->parent_exec_id = current->self_exec_id;
2642 p->exit_signal = args->exit_signal;
2643 }
2644
2645 klp_copy_process(p);
2646
2647 sched_core_fork(p);
2648
2649 spin_lock(&current->sighand->siglock);
2650
2651 rv_task_fork(p);
2652
2653 rseq_fork(p, clone_flags);
2654
2655 /* Don't start children in a dying pid namespace */
2656 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2657 retval = -ENOMEM;
2658 goto bad_fork_cancel_cgroup;
2659 }
2660
2661 /* Let kill terminate clone/fork in the middle */
2662 if (fatal_signal_pending(current)) {
2663 retval = -EINTR;
2664 goto bad_fork_cancel_cgroup;
2665 }
2666
2667 /* No more failure paths after this point. */
2668
2669 /*
2670 * Copy seccomp details explicitly here, in case they were changed
2671 * before holding sighand lock.
2672 */
2673 copy_seccomp(p);
2674
2675 init_task_pid_links(p);
2676 if (likely(p->pid)) {
2677 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2678
2679 init_task_pid(p, PIDTYPE_PID, pid);
2680 if (thread_group_leader(p)) {
2681 init_task_pid(p, PIDTYPE_TGID, pid);
2682 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2683 init_task_pid(p, PIDTYPE_SID, task_session(current));
2684
2685 if (is_child_reaper(pid)) {
2686 ns_of_pid(pid)->child_reaper = p;
2687 p->signal->flags |= SIGNAL_UNKILLABLE;
2688 }
2689 p->signal->shared_pending.signal = delayed.signal;
2690 p->signal->tty = tty_kref_get(current->signal->tty);
2691 /*
2692 * Inherit has_child_subreaper flag under the same
2693 * tasklist_lock with adding child to the process tree
2694 * for propagate_has_child_subreaper optimization.
2695 */
2696 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2697 p->real_parent->signal->is_child_subreaper;
2698 list_add_tail(&p->sibling, &p->real_parent->children);
2699 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2700 attach_pid(p, PIDTYPE_TGID);
2701 attach_pid(p, PIDTYPE_PGID);
2702 attach_pid(p, PIDTYPE_SID);
2703 __this_cpu_inc(process_counts);
2704 } else {
2705 current->signal->nr_threads++;
2706 current->signal->quick_threads++;
2707 atomic_inc(&current->signal->live);
2708 refcount_inc(&current->signal->sigcnt);
2709 task_join_group_stop(p);
2710 list_add_tail_rcu(&p->thread_group,
2711 &p->group_leader->thread_group);
2712 list_add_tail_rcu(&p->thread_node,
2713 &p->signal->thread_head);
2714 }
2715 attach_pid(p, PIDTYPE_PID);
2716 nr_threads++;
2717 }
2718 total_forks++;
2719 hlist_del_init(&delayed.node);
2720 spin_unlock(&current->sighand->siglock);
2721 syscall_tracepoint_update(p);
2722 write_unlock_irq(&tasklist_lock);
2723
2724 if (pidfile)
2725 fd_install(pidfd, pidfile);
2726
2727 proc_fork_connector(p);
2728 sched_post_fork(p);
2729 cgroup_post_fork(p, args);
2730 perf_event_fork(p);
2731
2732 trace_task_newtask(p, clone_flags);
2733 uprobe_copy_process(p, clone_flags);
2734 user_events_fork(p, clone_flags);
2735
2736 copy_oom_score_adj(clone_flags, p);
2737
2738 return p;
2739
2740bad_fork_cancel_cgroup:
2741 sched_core_free(p);
2742 spin_unlock(&current->sighand->siglock);
2743 write_unlock_irq(&tasklist_lock);
2744 cgroup_cancel_fork(p, args);
2745bad_fork_put_pidfd:
2746 if (clone_flags & CLONE_PIDFD) {
2747 fput(pidfile);
2748 put_unused_fd(pidfd);
2749 }
2750bad_fork_free_pid:
2751 if (pid != &init_struct_pid)
2752 free_pid(pid);
2753bad_fork_cleanup_thread:
2754 exit_thread(p);
2755bad_fork_cleanup_io:
2756 if (p->io_context)
2757 exit_io_context(p);
2758bad_fork_cleanup_namespaces:
2759 exit_task_namespaces(p);
2760bad_fork_cleanup_mm:
2761 if (p->mm) {
2762 mm_clear_owner(p->mm, p);
2763 mmput(p->mm);
2764 }
2765bad_fork_cleanup_signal:
2766 if (!(clone_flags & CLONE_THREAD))
2767 free_signal_struct(p->signal);
2768bad_fork_cleanup_sighand:
2769 __cleanup_sighand(p->sighand);
2770bad_fork_cleanup_fs:
2771 exit_fs(p); /* blocking */
2772bad_fork_cleanup_files:
2773 exit_files(p); /* blocking */
2774bad_fork_cleanup_semundo:
2775 exit_sem(p);
2776bad_fork_cleanup_security:
2777 security_task_free(p);
2778bad_fork_cleanup_audit:
2779 audit_free(p);
2780bad_fork_cleanup_perf:
2781 perf_event_free_task(p);
2782bad_fork_cleanup_policy:
2783 lockdep_free_task(p);
2784#ifdef CONFIG_NUMA
2785 mpol_put(p->mempolicy);
2786#endif
2787bad_fork_cleanup_delayacct:
2788 delayacct_tsk_free(p);
2789bad_fork_cleanup_count:
2790 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2791 exit_creds(p);
2792bad_fork_free:
2793 WRITE_ONCE(p->__state, TASK_DEAD);
2794 exit_task_stack_account(p);
2795 put_task_stack(p);
2796 delayed_free_task(p);
2797fork_out:
2798 spin_lock_irq(&current->sighand->siglock);
2799 hlist_del_init(&delayed.node);
2800 spin_unlock_irq(&current->sighand->siglock);
2801 return ERR_PTR(retval);
2802}
2803
2804static inline void init_idle_pids(struct task_struct *idle)
2805{
2806 enum pid_type type;
2807
2808 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2809 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2810 init_task_pid(idle, type, &init_struct_pid);
2811 }
2812}
2813
2814static int idle_dummy(void *dummy)
2815{
2816 /* This function is never called */
2817 return 0;
2818}
2819
2820struct task_struct * __init fork_idle(int cpu)
2821{
2822 struct task_struct *task;
2823 struct kernel_clone_args args = {
2824 .flags = CLONE_VM,
2825 .fn = &idle_dummy,
2826 .fn_arg = NULL,
2827 .kthread = 1,
2828 .idle = 1,
2829 };
2830
2831 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2832 if (!IS_ERR(task)) {
2833 init_idle_pids(task);
2834 init_idle(task, cpu);
2835 }
2836
2837 return task;
2838}
2839
2840/*
2841 * This is like kernel_clone(), but shaved down and tailored to just
2842 * creating io_uring workers. It returns a created task, or an error pointer.
2843 * The returned task is inactive, and the caller must fire it up through
2844 * wake_up_new_task(p). All signals are blocked in the created task.
2845 */
2846struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2847{
2848 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2849 CLONE_IO;
2850 struct kernel_clone_args args = {
2851 .flags = ((lower_32_bits(flags) | CLONE_VM |
2852 CLONE_UNTRACED) & ~CSIGNAL),
2853 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2854 .fn = fn,
2855 .fn_arg = arg,
2856 .io_thread = 1,
2857 .user_worker = 1,
2858 };
2859
2860 return copy_process(NULL, 0, node, &args);
2861}
2862
2863/*
2864 * Ok, this is the main fork-routine.
2865 *
2866 * It copies the process, and if successful kick-starts
2867 * it and waits for it to finish using the VM if required.
2868 *
2869 * args->exit_signal is expected to be checked for sanity by the caller.
2870 */
2871pid_t kernel_clone(struct kernel_clone_args *args)
2872{
2873 u64 clone_flags = args->flags;
2874 struct completion vfork;
2875 struct pid *pid;
2876 struct task_struct *p;
2877 int trace = 0;
2878 pid_t nr;
2879
2880 /*
2881 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2882 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2883 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2884 * field in struct clone_args and it still doesn't make sense to have
2885 * them both point at the same memory location. Performing this check
2886 * here has the advantage that we don't need to have a separate helper
2887 * to check for legacy clone().
2888 */
2889 if ((args->flags & CLONE_PIDFD) &&
2890 (args->flags & CLONE_PARENT_SETTID) &&
2891 (args->pidfd == args->parent_tid))
2892 return -EINVAL;
2893
2894 /*
2895 * Determine whether and which event to report to ptracer. When
2896 * called from kernel_thread or CLONE_UNTRACED is explicitly
2897 * requested, no event is reported; otherwise, report if the event
2898 * for the type of forking is enabled.
2899 */
2900 if (!(clone_flags & CLONE_UNTRACED)) {
2901 if (clone_flags & CLONE_VFORK)
2902 trace = PTRACE_EVENT_VFORK;
2903 else if (args->exit_signal != SIGCHLD)
2904 trace = PTRACE_EVENT_CLONE;
2905 else
2906 trace = PTRACE_EVENT_FORK;
2907
2908 if (likely(!ptrace_event_enabled(current, trace)))
2909 trace = 0;
2910 }
2911
2912 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2913 add_latent_entropy();
2914
2915 if (IS_ERR(p))
2916 return PTR_ERR(p);
2917
2918 /*
2919 * Do this prior waking up the new thread - the thread pointer
2920 * might get invalid after that point, if the thread exits quickly.
2921 */
2922 trace_sched_process_fork(current, p);
2923
2924 pid = get_task_pid(p, PIDTYPE_PID);
2925 nr = pid_vnr(pid);
2926
2927 if (clone_flags & CLONE_PARENT_SETTID)
2928 put_user(nr, args->parent_tid);
2929
2930 if (clone_flags & CLONE_VFORK) {
2931 p->vfork_done = &vfork;
2932 init_completion(&vfork);
2933 get_task_struct(p);
2934 }
2935
2936 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2937 /* lock the task to synchronize with memcg migration */
2938 task_lock(p);
2939 lru_gen_add_mm(p->mm);
2940 task_unlock(p);
2941 }
2942
2943 wake_up_new_task(p);
2944
2945 /* forking complete and child started to run, tell ptracer */
2946 if (unlikely(trace))
2947 ptrace_event_pid(trace, pid);
2948
2949 if (clone_flags & CLONE_VFORK) {
2950 if (!wait_for_vfork_done(p, &vfork))
2951 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2952 }
2953
2954 put_pid(pid);
2955 return nr;
2956}
2957
2958/*
2959 * Create a kernel thread.
2960 */
2961pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2962 unsigned long flags)
2963{
2964 struct kernel_clone_args args = {
2965 .flags = ((lower_32_bits(flags) | CLONE_VM |
2966 CLONE_UNTRACED) & ~CSIGNAL),
2967 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2968 .fn = fn,
2969 .fn_arg = arg,
2970 .name = name,
2971 .kthread = 1,
2972 };
2973
2974 return kernel_clone(&args);
2975}
2976
2977/*
2978 * Create a user mode thread.
2979 */
2980pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2981{
2982 struct kernel_clone_args args = {
2983 .flags = ((lower_32_bits(flags) | CLONE_VM |
2984 CLONE_UNTRACED) & ~CSIGNAL),
2985 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2986 .fn = fn,
2987 .fn_arg = arg,
2988 };
2989
2990 return kernel_clone(&args);
2991}
2992
2993#ifdef __ARCH_WANT_SYS_FORK
2994SYSCALL_DEFINE0(fork)
2995{
2996#ifdef CONFIG_MMU
2997 struct kernel_clone_args args = {
2998 .exit_signal = SIGCHLD,
2999 };
3000
3001 return kernel_clone(&args);
3002#else
3003 /* can not support in nommu mode */
3004 return -EINVAL;
3005#endif
3006}
3007#endif
3008
3009#ifdef __ARCH_WANT_SYS_VFORK
3010SYSCALL_DEFINE0(vfork)
3011{
3012 struct kernel_clone_args args = {
3013 .flags = CLONE_VFORK | CLONE_VM,
3014 .exit_signal = SIGCHLD,
3015 };
3016
3017 return kernel_clone(&args);
3018}
3019#endif
3020
3021#ifdef __ARCH_WANT_SYS_CLONE
3022#ifdef CONFIG_CLONE_BACKWARDS
3023SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3024 int __user *, parent_tidptr,
3025 unsigned long, tls,
3026 int __user *, child_tidptr)
3027#elif defined(CONFIG_CLONE_BACKWARDS2)
3028SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3029 int __user *, parent_tidptr,
3030 int __user *, child_tidptr,
3031 unsigned long, tls)
3032#elif defined(CONFIG_CLONE_BACKWARDS3)
3033SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3034 int, stack_size,
3035 int __user *, parent_tidptr,
3036 int __user *, child_tidptr,
3037 unsigned long, tls)
3038#else
3039SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3040 int __user *, parent_tidptr,
3041 int __user *, child_tidptr,
3042 unsigned long, tls)
3043#endif
3044{
3045 struct kernel_clone_args args = {
3046 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3047 .pidfd = parent_tidptr,
3048 .child_tid = child_tidptr,
3049 .parent_tid = parent_tidptr,
3050 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3051 .stack = newsp,
3052 .tls = tls,
3053 };
3054
3055 return kernel_clone(&args);
3056}
3057#endif
3058
3059#ifdef __ARCH_WANT_SYS_CLONE3
3060
3061noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3062 struct clone_args __user *uargs,
3063 size_t usize)
3064{
3065 int err;
3066 struct clone_args args;
3067 pid_t *kset_tid = kargs->set_tid;
3068
3069 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3070 CLONE_ARGS_SIZE_VER0);
3071 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3072 CLONE_ARGS_SIZE_VER1);
3073 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3074 CLONE_ARGS_SIZE_VER2);
3075 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3076
3077 if (unlikely(usize > PAGE_SIZE))
3078 return -E2BIG;
3079 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3080 return -EINVAL;
3081
3082 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3083 if (err)
3084 return err;
3085
3086 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3087 return -EINVAL;
3088
3089 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3090 return -EINVAL;
3091
3092 if (unlikely(args.set_tid && args.set_tid_size == 0))
3093 return -EINVAL;
3094
3095 /*
3096 * Verify that higher 32bits of exit_signal are unset and that
3097 * it is a valid signal
3098 */
3099 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3100 !valid_signal(args.exit_signal)))
3101 return -EINVAL;
3102
3103 if ((args.flags & CLONE_INTO_CGROUP) &&
3104 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3105 return -EINVAL;
3106
3107 *kargs = (struct kernel_clone_args){
3108 .flags = args.flags,
3109 .pidfd = u64_to_user_ptr(args.pidfd),
3110 .child_tid = u64_to_user_ptr(args.child_tid),
3111 .parent_tid = u64_to_user_ptr(args.parent_tid),
3112 .exit_signal = args.exit_signal,
3113 .stack = args.stack,
3114 .stack_size = args.stack_size,
3115 .tls = args.tls,
3116 .set_tid_size = args.set_tid_size,
3117 .cgroup = args.cgroup,
3118 };
3119
3120 if (args.set_tid &&
3121 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3122 (kargs->set_tid_size * sizeof(pid_t))))
3123 return -EFAULT;
3124
3125 kargs->set_tid = kset_tid;
3126
3127 return 0;
3128}
3129
3130/**
3131 * clone3_stack_valid - check and prepare stack
3132 * @kargs: kernel clone args
3133 *
3134 * Verify that the stack arguments userspace gave us are sane.
3135 * In addition, set the stack direction for userspace since it's easy for us to
3136 * determine.
3137 */
3138static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3139{
3140 if (kargs->stack == 0) {
3141 if (kargs->stack_size > 0)
3142 return false;
3143 } else {
3144 if (kargs->stack_size == 0)
3145 return false;
3146
3147 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3148 return false;
3149
3150#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3151 kargs->stack += kargs->stack_size;
3152#endif
3153 }
3154
3155 return true;
3156}
3157
3158static bool clone3_args_valid(struct kernel_clone_args *kargs)
3159{
3160 /* Verify that no unknown flags are passed along. */
3161 if (kargs->flags &
3162 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3163 return false;
3164
3165 /*
3166 * - make the CLONE_DETACHED bit reusable for clone3
3167 * - make the CSIGNAL bits reusable for clone3
3168 */
3169 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3170 return false;
3171
3172 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3173 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3174 return false;
3175
3176 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3177 kargs->exit_signal)
3178 return false;
3179
3180 if (!clone3_stack_valid(kargs))
3181 return false;
3182
3183 return true;
3184}
3185
3186/**
3187 * clone3 - create a new process with specific properties
3188 * @uargs: argument structure
3189 * @size: size of @uargs
3190 *
3191 * clone3() is the extensible successor to clone()/clone2().
3192 * It takes a struct as argument that is versioned by its size.
3193 *
3194 * Return: On success, a positive PID for the child process.
3195 * On error, a negative errno number.
3196 */
3197SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3198{
3199 int err;
3200
3201 struct kernel_clone_args kargs;
3202 pid_t set_tid[MAX_PID_NS_LEVEL];
3203
3204 kargs.set_tid = set_tid;
3205
3206 err = copy_clone_args_from_user(&kargs, uargs, size);
3207 if (err)
3208 return err;
3209
3210 if (!clone3_args_valid(&kargs))
3211 return -EINVAL;
3212
3213 return kernel_clone(&kargs);
3214}
3215#endif
3216
3217void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3218{
3219 struct task_struct *leader, *parent, *child;
3220 int res;
3221
3222 read_lock(&tasklist_lock);
3223 leader = top = top->group_leader;
3224down:
3225 for_each_thread(leader, parent) {
3226 list_for_each_entry(child, &parent->children, sibling) {
3227 res = visitor(child, data);
3228 if (res) {
3229 if (res < 0)
3230 goto out;
3231 leader = child;
3232 goto down;
3233 }
3234up:
3235 ;
3236 }
3237 }
3238
3239 if (leader != top) {
3240 child = leader;
3241 parent = child->real_parent;
3242 leader = parent->group_leader;
3243 goto up;
3244 }
3245out:
3246 read_unlock(&tasklist_lock);
3247}
3248
3249#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3250#define ARCH_MIN_MMSTRUCT_ALIGN 0
3251#endif
3252
3253static void sighand_ctor(void *data)
3254{
3255 struct sighand_struct *sighand = data;
3256
3257 spin_lock_init(&sighand->siglock);
3258 init_waitqueue_head(&sighand->signalfd_wqh);
3259}
3260
3261void __init mm_cache_init(void)
3262{
3263 unsigned int mm_size;
3264
3265 /*
3266 * The mm_cpumask is located at the end of mm_struct, and is
3267 * dynamically sized based on the maximum CPU number this system
3268 * can have, taking hotplug into account (nr_cpu_ids).
3269 */
3270 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3271
3272 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3273 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3274 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3275 offsetof(struct mm_struct, saved_auxv),
3276 sizeof_field(struct mm_struct, saved_auxv),
3277 NULL);
3278}
3279
3280void __init proc_caches_init(void)
3281{
3282 sighand_cachep = kmem_cache_create("sighand_cache",
3283 sizeof(struct sighand_struct), 0,
3284 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3285 SLAB_ACCOUNT, sighand_ctor);
3286 signal_cachep = kmem_cache_create("signal_cache",
3287 sizeof(struct signal_struct), 0,
3288 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3289 NULL);
3290 files_cachep = kmem_cache_create("files_cache",
3291 sizeof(struct files_struct), 0,
3292 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3293 NULL);
3294 fs_cachep = kmem_cache_create("fs_cache",
3295 sizeof(struct fs_struct), 0,
3296 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3297 NULL);
3298
3299 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3300#ifdef CONFIG_PER_VMA_LOCK
3301 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3302#endif
3303 mmap_init();
3304 nsproxy_cache_init();
3305}
3306
3307/*
3308 * Check constraints on flags passed to the unshare system call.
3309 */
3310static int check_unshare_flags(unsigned long unshare_flags)
3311{
3312 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3313 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3314 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3315 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3316 CLONE_NEWTIME))
3317 return -EINVAL;
3318 /*
3319 * Not implemented, but pretend it works if there is nothing
3320 * to unshare. Note that unsharing the address space or the
3321 * signal handlers also need to unshare the signal queues (aka
3322 * CLONE_THREAD).
3323 */
3324 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3325 if (!thread_group_empty(current))
3326 return -EINVAL;
3327 }
3328 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3329 if (refcount_read(&current->sighand->count) > 1)
3330 return -EINVAL;
3331 }
3332 if (unshare_flags & CLONE_VM) {
3333 if (!current_is_single_threaded())
3334 return -EINVAL;
3335 }
3336
3337 return 0;
3338}
3339
3340/*
3341 * Unshare the filesystem structure if it is being shared
3342 */
3343static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3344{
3345 struct fs_struct *fs = current->fs;
3346
3347 if (!(unshare_flags & CLONE_FS) || !fs)
3348 return 0;
3349
3350 /* don't need lock here; in the worst case we'll do useless copy */
3351 if (fs->users == 1)
3352 return 0;
3353
3354 *new_fsp = copy_fs_struct(fs);
3355 if (!*new_fsp)
3356 return -ENOMEM;
3357
3358 return 0;
3359}
3360
3361/*
3362 * Unshare file descriptor table if it is being shared
3363 */
3364int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3365 struct files_struct **new_fdp)
3366{
3367 struct files_struct *fd = current->files;
3368 int error = 0;
3369
3370 if ((unshare_flags & CLONE_FILES) &&
3371 (fd && atomic_read(&fd->count) > 1)) {
3372 *new_fdp = dup_fd(fd, max_fds, &error);
3373 if (!*new_fdp)
3374 return error;
3375 }
3376
3377 return 0;
3378}
3379
3380/*
3381 * unshare allows a process to 'unshare' part of the process
3382 * context which was originally shared using clone. copy_*
3383 * functions used by kernel_clone() cannot be used here directly
3384 * because they modify an inactive task_struct that is being
3385 * constructed. Here we are modifying the current, active,
3386 * task_struct.
3387 */
3388int ksys_unshare(unsigned long unshare_flags)
3389{
3390 struct fs_struct *fs, *new_fs = NULL;
3391 struct files_struct *new_fd = NULL;
3392 struct cred *new_cred = NULL;
3393 struct nsproxy *new_nsproxy = NULL;
3394 int do_sysvsem = 0;
3395 int err;
3396
3397 /*
3398 * If unsharing a user namespace must also unshare the thread group
3399 * and unshare the filesystem root and working directories.
3400 */
3401 if (unshare_flags & CLONE_NEWUSER)
3402 unshare_flags |= CLONE_THREAD | CLONE_FS;
3403 /*
3404 * If unsharing vm, must also unshare signal handlers.
3405 */
3406 if (unshare_flags & CLONE_VM)
3407 unshare_flags |= CLONE_SIGHAND;
3408 /*
3409 * If unsharing a signal handlers, must also unshare the signal queues.
3410 */
3411 if (unshare_flags & CLONE_SIGHAND)
3412 unshare_flags |= CLONE_THREAD;
3413 /*
3414 * If unsharing namespace, must also unshare filesystem information.
3415 */
3416 if (unshare_flags & CLONE_NEWNS)
3417 unshare_flags |= CLONE_FS;
3418
3419 err = check_unshare_flags(unshare_flags);
3420 if (err)
3421 goto bad_unshare_out;
3422 /*
3423 * CLONE_NEWIPC must also detach from the undolist: after switching
3424 * to a new ipc namespace, the semaphore arrays from the old
3425 * namespace are unreachable.
3426 */
3427 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3428 do_sysvsem = 1;
3429 err = unshare_fs(unshare_flags, &new_fs);
3430 if (err)
3431 goto bad_unshare_out;
3432 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3433 if (err)
3434 goto bad_unshare_cleanup_fs;
3435 err = unshare_userns(unshare_flags, &new_cred);
3436 if (err)
3437 goto bad_unshare_cleanup_fd;
3438 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3439 new_cred, new_fs);
3440 if (err)
3441 goto bad_unshare_cleanup_cred;
3442
3443 if (new_cred) {
3444 err = set_cred_ucounts(new_cred);
3445 if (err)
3446 goto bad_unshare_cleanup_cred;
3447 }
3448
3449 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3450 if (do_sysvsem) {
3451 /*
3452 * CLONE_SYSVSEM is equivalent to sys_exit().
3453 */
3454 exit_sem(current);
3455 }
3456 if (unshare_flags & CLONE_NEWIPC) {
3457 /* Orphan segments in old ns (see sem above). */
3458 exit_shm(current);
3459 shm_init_task(current);
3460 }
3461
3462 if (new_nsproxy)
3463 switch_task_namespaces(current, new_nsproxy);
3464
3465 task_lock(current);
3466
3467 if (new_fs) {
3468 fs = current->fs;
3469 spin_lock(&fs->lock);
3470 current->fs = new_fs;
3471 if (--fs->users)
3472 new_fs = NULL;
3473 else
3474 new_fs = fs;
3475 spin_unlock(&fs->lock);
3476 }
3477
3478 if (new_fd)
3479 swap(current->files, new_fd);
3480
3481 task_unlock(current);
3482
3483 if (new_cred) {
3484 /* Install the new user namespace */
3485 commit_creds(new_cred);
3486 new_cred = NULL;
3487 }
3488 }
3489
3490 perf_event_namespaces(current);
3491
3492bad_unshare_cleanup_cred:
3493 if (new_cred)
3494 put_cred(new_cred);
3495bad_unshare_cleanup_fd:
3496 if (new_fd)
3497 put_files_struct(new_fd);
3498
3499bad_unshare_cleanup_fs:
3500 if (new_fs)
3501 free_fs_struct(new_fs);
3502
3503bad_unshare_out:
3504 return err;
3505}
3506
3507SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3508{
3509 return ksys_unshare(unshare_flags);
3510}
3511
3512/*
3513 * Helper to unshare the files of the current task.
3514 * We don't want to expose copy_files internals to
3515 * the exec layer of the kernel.
3516 */
3517
3518int unshare_files(void)
3519{
3520 struct task_struct *task = current;
3521 struct files_struct *old, *copy = NULL;
3522 int error;
3523
3524 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3525 if (error || !copy)
3526 return error;
3527
3528 old = task->files;
3529 task_lock(task);
3530 task->files = copy;
3531 task_unlock(task);
3532 put_files_struct(old);
3533 return 0;
3534}
3535
3536int sysctl_max_threads(struct ctl_table *table, int write,
3537 void *buffer, size_t *lenp, loff_t *ppos)
3538{
3539 struct ctl_table t;
3540 int ret;
3541 int threads = max_threads;
3542 int min = 1;
3543 int max = MAX_THREADS;
3544
3545 t = *table;
3546 t.data = &threads;
3547 t.extra1 = &min;
3548 t.extra2 = &max;
3549
3550 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3551 if (ret || !write)
3552 return ret;
3553
3554 max_threads = threads;
3555
3556 return 0;
3557}