projects/linux·blob v6.4-rc2 · refs

3563 lines · 87.6 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
256 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
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 }
264 return 0;
265err:
266 /*
267 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
268 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
269 * ignore this page.
270 */
271 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
272 memcg_kmem_uncharge_page(vm->pages[i], 0);
273 return ret;
274}
275
276static int alloc_thread_stack_node(struct task_struct *tsk, int node)
277{
278 struct vm_struct *vm;
279 void *stack;
280 int i;
281
282 for (i = 0; i < NR_CACHED_STACKS; i++) {
283 struct vm_struct *s;
284
285 s = this_cpu_xchg(cached_stacks[i], NULL);
286
287 if (!s)
288 continue;
289
290 /* Reset stack metadata. */
291 kasan_unpoison_range(s->addr, THREAD_SIZE);
292
293 stack = kasan_reset_tag(s->addr);
294
295 /* Clear stale pointers from reused stack. */
296 memset(stack, 0, THREAD_SIZE);
297
298 if (memcg_charge_kernel_stack(s)) {
299 vfree(s->addr);
300 return -ENOMEM;
301 }
302
303 tsk->stack_vm_area = s;
304 tsk->stack = stack;
305 return 0;
306 }
307
308 /*
309 * Allocated stacks are cached and later reused by new threads,
310 * so memcg accounting is performed manually on assigning/releasing
311 * stacks to tasks. Drop __GFP_ACCOUNT.
312 */
313 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
314 VMALLOC_START, VMALLOC_END,
315 THREADINFO_GFP & ~__GFP_ACCOUNT,
316 PAGE_KERNEL,
317 0, node, __builtin_return_address(0));
318 if (!stack)
319 return -ENOMEM;
320
321 vm = find_vm_area(stack);
322 if (memcg_charge_kernel_stack(vm)) {
323 vfree(stack);
324 return -ENOMEM;
325 }
326 /*
327 * We can't call find_vm_area() in interrupt context, and
328 * free_thread_stack() can be called in interrupt context,
329 * so cache the vm_struct.
330 */
331 tsk->stack_vm_area = vm;
332 stack = kasan_reset_tag(stack);
333 tsk->stack = stack;
334 return 0;
335}
336
337static void free_thread_stack(struct task_struct *tsk)
338{
339 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
340 thread_stack_delayed_free(tsk);
341
342 tsk->stack = NULL;
343 tsk->stack_vm_area = NULL;
344}
345
346# else /* !CONFIG_VMAP_STACK */
347
348static void thread_stack_free_rcu(struct rcu_head *rh)
349{
350 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
351}
352
353static void thread_stack_delayed_free(struct task_struct *tsk)
354{
355 struct rcu_head *rh = tsk->stack;
356
357 call_rcu(rh, thread_stack_free_rcu);
358}
359
360static int alloc_thread_stack_node(struct task_struct *tsk, int node)
361{
362 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
363 THREAD_SIZE_ORDER);
364
365 if (likely(page)) {
366 tsk->stack = kasan_reset_tag(page_address(page));
367 return 0;
368 }
369 return -ENOMEM;
370}
371
372static void free_thread_stack(struct task_struct *tsk)
373{
374 thread_stack_delayed_free(tsk);
375 tsk->stack = NULL;
376}
377
378# endif /* CONFIG_VMAP_STACK */
379# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
380
381static struct kmem_cache *thread_stack_cache;
382
383static void thread_stack_free_rcu(struct rcu_head *rh)
384{
385 kmem_cache_free(thread_stack_cache, rh);
386}
387
388static void thread_stack_delayed_free(struct task_struct *tsk)
389{
390 struct rcu_head *rh = tsk->stack;
391
392 call_rcu(rh, thread_stack_free_rcu);
393}
394
395static int alloc_thread_stack_node(struct task_struct *tsk, int node)
396{
397 unsigned long *stack;
398 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
399 stack = kasan_reset_tag(stack);
400 tsk->stack = stack;
401 return stack ? 0 : -ENOMEM;
402}
403
404static void free_thread_stack(struct task_struct *tsk)
405{
406 thread_stack_delayed_free(tsk);
407 tsk->stack = NULL;
408}
409
410void thread_stack_cache_init(void)
411{
412 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
413 THREAD_SIZE, THREAD_SIZE, 0, 0,
414 THREAD_SIZE, NULL);
415 BUG_ON(thread_stack_cache == NULL);
416}
417
418# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
419#else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
420
421static int alloc_thread_stack_node(struct task_struct *tsk, int node)
422{
423 unsigned long *stack;
424
425 stack = arch_alloc_thread_stack_node(tsk, node);
426 tsk->stack = stack;
427 return stack ? 0 : -ENOMEM;
428}
429
430static void free_thread_stack(struct task_struct *tsk)
431{
432 arch_free_thread_stack(tsk);
433 tsk->stack = NULL;
434}
435
436#endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
437
438/* SLAB cache for signal_struct structures (tsk->signal) */
439static struct kmem_cache *signal_cachep;
440
441/* SLAB cache for sighand_struct structures (tsk->sighand) */
442struct kmem_cache *sighand_cachep;
443
444/* SLAB cache for files_struct structures (tsk->files) */
445struct kmem_cache *files_cachep;
446
447/* SLAB cache for fs_struct structures (tsk->fs) */
448struct kmem_cache *fs_cachep;
449
450/* SLAB cache for vm_area_struct structures */
451static struct kmem_cache *vm_area_cachep;
452
453/* SLAB cache for mm_struct structures (tsk->mm) */
454static struct kmem_cache *mm_cachep;
455
456#ifdef CONFIG_PER_VMA_LOCK
457
458/* SLAB cache for vm_area_struct.lock */
459static struct kmem_cache *vma_lock_cachep;
460
461static bool vma_lock_alloc(struct vm_area_struct *vma)
462{
463 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
464 if (!vma->vm_lock)
465 return false;
466
467 init_rwsem(&vma->vm_lock->lock);
468 vma->vm_lock_seq = -1;
469
470 return true;
471}
472
473static inline void vma_lock_free(struct vm_area_struct *vma)
474{
475 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
476}
477
478#else /* CONFIG_PER_VMA_LOCK */
479
480static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
481static inline void vma_lock_free(struct vm_area_struct *vma) {}
482
483#endif /* CONFIG_PER_VMA_LOCK */
484
485struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
486{
487 struct vm_area_struct *vma;
488
489 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
490 if (!vma)
491 return NULL;
492
493 vma_init(vma, mm);
494 if (!vma_lock_alloc(vma)) {
495 kmem_cache_free(vm_area_cachep, vma);
496 return NULL;
497 }
498
499 return vma;
500}
501
502struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
503{
504 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
505
506 if (!new)
507 return NULL;
508
509 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
510 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
511 /*
512 * orig->shared.rb may be modified concurrently, but the clone
513 * will be reinitialized.
514 */
515 data_race(memcpy(new, orig, sizeof(*new)));
516 if (!vma_lock_alloc(new)) {
517 kmem_cache_free(vm_area_cachep, new);
518 return NULL;
519 }
520 INIT_LIST_HEAD(&new->anon_vma_chain);
521 vma_numab_state_init(new);
522 dup_anon_vma_name(orig, new);
523
524 return new;
525}
526
527void __vm_area_free(struct vm_area_struct *vma)
528{
529 vma_numab_state_free(vma);
530 free_anon_vma_name(vma);
531 vma_lock_free(vma);
532 kmem_cache_free(vm_area_cachep, vma);
533}
534
535#ifdef CONFIG_PER_VMA_LOCK
536static void vm_area_free_rcu_cb(struct rcu_head *head)
537{
538 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
539 vm_rcu);
540
541 /* The vma should not be locked while being destroyed. */
542 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
543 __vm_area_free(vma);
544}
545#endif
546
547void vm_area_free(struct vm_area_struct *vma)
548{
549#ifdef CONFIG_PER_VMA_LOCK
550 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
551#else
552 __vm_area_free(vma);
553#endif
554}
555
556static void account_kernel_stack(struct task_struct *tsk, int account)
557{
558 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
559 struct vm_struct *vm = task_stack_vm_area(tsk);
560 int i;
561
562 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
563 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
564 account * (PAGE_SIZE / 1024));
565 } else {
566 void *stack = task_stack_page(tsk);
567
568 /* All stack pages are in the same node. */
569 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
570 account * (THREAD_SIZE / 1024));
571 }
572}
573
574void exit_task_stack_account(struct task_struct *tsk)
575{
576 account_kernel_stack(tsk, -1);
577
578 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
579 struct vm_struct *vm;
580 int i;
581
582 vm = task_stack_vm_area(tsk);
583 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
584 memcg_kmem_uncharge_page(vm->pages[i], 0);
585 }
586}
587
588static void release_task_stack(struct task_struct *tsk)
589{
590 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
591 return; /* Better to leak the stack than to free prematurely */
592
593 free_thread_stack(tsk);
594}
595
596#ifdef CONFIG_THREAD_INFO_IN_TASK
597void put_task_stack(struct task_struct *tsk)
598{
599 if (refcount_dec_and_test(&tsk->stack_refcount))
600 release_task_stack(tsk);
601}
602#endif
603
604void free_task(struct task_struct *tsk)
605{
606#ifdef CONFIG_SECCOMP
607 WARN_ON_ONCE(tsk->seccomp.filter);
608#endif
609 release_user_cpus_ptr(tsk);
610 scs_release(tsk);
611
612#ifndef CONFIG_THREAD_INFO_IN_TASK
613 /*
614 * The task is finally done with both the stack and thread_info,
615 * so free both.
616 */
617 release_task_stack(tsk);
618#else
619 /*
620 * If the task had a separate stack allocation, it should be gone
621 * by now.
622 */
623 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
624#endif
625 rt_mutex_debug_task_free(tsk);
626 ftrace_graph_exit_task(tsk);
627 arch_release_task_struct(tsk);
628 if (tsk->flags & PF_KTHREAD)
629 free_kthread_struct(tsk);
630 free_task_struct(tsk);
631}
632EXPORT_SYMBOL(free_task);
633
634static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
635{
636 struct file *exe_file;
637
638 exe_file = get_mm_exe_file(oldmm);
639 RCU_INIT_POINTER(mm->exe_file, exe_file);
640 /*
641 * We depend on the oldmm having properly denied write access to the
642 * exe_file already.
643 */
644 if (exe_file && deny_write_access(exe_file))
645 pr_warn_once("deny_write_access() failed in %s\n", __func__);
646}
647
648#ifdef CONFIG_MMU
649static __latent_entropy int dup_mmap(struct mm_struct *mm,
650 struct mm_struct *oldmm)
651{
652 struct vm_area_struct *mpnt, *tmp;
653 int retval;
654 unsigned long charge = 0;
655 LIST_HEAD(uf);
656 VMA_ITERATOR(old_vmi, oldmm, 0);
657 VMA_ITERATOR(vmi, mm, 0);
658
659 uprobe_start_dup_mmap();
660 if (mmap_write_lock_killable(oldmm)) {
661 retval = -EINTR;
662 goto fail_uprobe_end;
663 }
664 flush_cache_dup_mm(oldmm);
665 uprobe_dup_mmap(oldmm, mm);
666 /*
667 * Not linked in yet - no deadlock potential:
668 */
669 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
670
671 /* No ordering required: file already has been exposed. */
672 dup_mm_exe_file(mm, oldmm);
673
674 mm->total_vm = oldmm->total_vm;
675 mm->data_vm = oldmm->data_vm;
676 mm->exec_vm = oldmm->exec_vm;
677 mm->stack_vm = oldmm->stack_vm;
678
679 retval = ksm_fork(mm, oldmm);
680 if (retval)
681 goto out;
682 khugepaged_fork(mm, oldmm);
683
684 retval = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
685 if (retval)
686 goto out;
687
688 mt_clear_in_rcu(vmi.mas.tree);
689 for_each_vma(old_vmi, mpnt) {
690 struct file *file;
691
692 if (mpnt->vm_flags & VM_DONTCOPY) {
693 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
694 continue;
695 }
696 charge = 0;
697 /*
698 * Don't duplicate many vmas if we've been oom-killed (for
699 * example)
700 */
701 if (fatal_signal_pending(current)) {
702 retval = -EINTR;
703 goto loop_out;
704 }
705 if (mpnt->vm_flags & VM_ACCOUNT) {
706 unsigned long len = vma_pages(mpnt);
707
708 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
709 goto fail_nomem;
710 charge = len;
711 }
712 tmp = vm_area_dup(mpnt);
713 if (!tmp)
714 goto fail_nomem;
715 retval = vma_dup_policy(mpnt, tmp);
716 if (retval)
717 goto fail_nomem_policy;
718 tmp->vm_mm = mm;
719 retval = dup_userfaultfd(tmp, &uf);
720 if (retval)
721 goto fail_nomem_anon_vma_fork;
722 if (tmp->vm_flags & VM_WIPEONFORK) {
723 /*
724 * VM_WIPEONFORK gets a clean slate in the child.
725 * Don't prepare anon_vma until fault since we don't
726 * copy page for current vma.
727 */
728 tmp->anon_vma = NULL;
729 } else if (anon_vma_fork(tmp, mpnt))
730 goto fail_nomem_anon_vma_fork;
731 vm_flags_clear(tmp, VM_LOCKED_MASK);
732 file = tmp->vm_file;
733 if (file) {
734 struct address_space *mapping = file->f_mapping;
735
736 get_file(file);
737 i_mmap_lock_write(mapping);
738 if (tmp->vm_flags & VM_SHARED)
739 mapping_allow_writable(mapping);
740 flush_dcache_mmap_lock(mapping);
741 /* insert tmp into the share list, just after mpnt */
742 vma_interval_tree_insert_after(tmp, mpnt,
743 &mapping->i_mmap);
744 flush_dcache_mmap_unlock(mapping);
745 i_mmap_unlock_write(mapping);
746 }
747
748 /*
749 * Copy/update hugetlb private vma information.
750 */
751 if (is_vm_hugetlb_page(tmp))
752 hugetlb_dup_vma_private(tmp);
753
754 /* Link the vma into the MT */
755 if (vma_iter_bulk_store(&vmi, tmp))
756 goto fail_nomem_vmi_store;
757
758 mm->map_count++;
759 if (!(tmp->vm_flags & VM_WIPEONFORK))
760 retval = copy_page_range(tmp, mpnt);
761
762 if (tmp->vm_ops && tmp->vm_ops->open)
763 tmp->vm_ops->open(tmp);
764
765 if (retval)
766 goto loop_out;
767 }
768 /* a new mm has just been created */
769 retval = arch_dup_mmap(oldmm, mm);
770loop_out:
771 vma_iter_free(&vmi);
772 if (!retval)
773 mt_set_in_rcu(vmi.mas.tree);
774out:
775 mmap_write_unlock(mm);
776 flush_tlb_mm(oldmm);
777 mmap_write_unlock(oldmm);
778 dup_userfaultfd_complete(&uf);
779fail_uprobe_end:
780 uprobe_end_dup_mmap();
781 return retval;
782
783fail_nomem_vmi_store:
784 unlink_anon_vmas(tmp);
785fail_nomem_anon_vma_fork:
786 mpol_put(vma_policy(tmp));
787fail_nomem_policy:
788 vm_area_free(tmp);
789fail_nomem:
790 retval = -ENOMEM;
791 vm_unacct_memory(charge);
792 goto loop_out;
793}
794
795static inline int mm_alloc_pgd(struct mm_struct *mm)
796{
797 mm->pgd = pgd_alloc(mm);
798 if (unlikely(!mm->pgd))
799 return -ENOMEM;
800 return 0;
801}
802
803static inline void mm_free_pgd(struct mm_struct *mm)
804{
805 pgd_free(mm, mm->pgd);
806}
807#else
808static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
809{
810 mmap_write_lock(oldmm);
811 dup_mm_exe_file(mm, oldmm);
812 mmap_write_unlock(oldmm);
813 return 0;
814}
815#define mm_alloc_pgd(mm) (0)
816#define mm_free_pgd(mm)
817#endif /* CONFIG_MMU */
818
819static void check_mm(struct mm_struct *mm)
820{
821 int i;
822
823 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
824 "Please make sure 'struct resident_page_types[]' is updated as well");
825
826 for (i = 0; i < NR_MM_COUNTERS; i++) {
827 long x = percpu_counter_sum(&mm->rss_stat[i]);
828
829 if (unlikely(x))
830 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
831 mm, resident_page_types[i], x);
832 }
833
834 if (mm_pgtables_bytes(mm))
835 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
836 mm_pgtables_bytes(mm));
837
838#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
839 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
840#endif
841}
842
843#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
844#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
845
846static void do_check_lazy_tlb(void *arg)
847{
848 struct mm_struct *mm = arg;
849
850 WARN_ON_ONCE(current->active_mm == mm);
851}
852
853static void do_shoot_lazy_tlb(void *arg)
854{
855 struct mm_struct *mm = arg;
856
857 if (current->active_mm == mm) {
858 WARN_ON_ONCE(current->mm);
859 current->active_mm = &init_mm;
860 switch_mm(mm, &init_mm, current);
861 }
862}
863
864static void cleanup_lazy_tlbs(struct mm_struct *mm)
865{
866 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
867 /*
868 * In this case, lazy tlb mms are refounted and would not reach
869 * __mmdrop until all CPUs have switched away and mmdrop()ed.
870 */
871 return;
872 }
873
874 /*
875 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
876 * requires lazy mm users to switch to another mm when the refcount
877 * drops to zero, before the mm is freed. This requires IPIs here to
878 * switch kernel threads to init_mm.
879 *
880 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
881 * switch with the final userspace teardown TLB flush which leaves the
882 * mm lazy on this CPU but no others, reducing the need for additional
883 * IPIs here. There are cases where a final IPI is still required here,
884 * such as the final mmdrop being performed on a different CPU than the
885 * one exiting, or kernel threads using the mm when userspace exits.
886 *
887 * IPI overheads have not found to be expensive, but they could be
888 * reduced in a number of possible ways, for example (roughly
889 * increasing order of complexity):
890 * - The last lazy reference created by exit_mm() could instead switch
891 * to init_mm, however it's probable this will run on the same CPU
892 * immediately afterwards, so this may not reduce IPIs much.
893 * - A batch of mms requiring IPIs could be gathered and freed at once.
894 * - CPUs store active_mm where it can be remotely checked without a
895 * lock, to filter out false-positives in the cpumask.
896 * - After mm_users or mm_count reaches zero, switching away from the
897 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
898 * with some batching or delaying of the final IPIs.
899 * - A delayed freeing and RCU-like quiescing sequence based on mm
900 * switching to avoid IPIs completely.
901 */
902 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
903 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
904 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
905}
906
907/*
908 * Called when the last reference to the mm
909 * is dropped: either by a lazy thread or by
910 * mmput. Free the page directory and the mm.
911 */
912void __mmdrop(struct mm_struct *mm)
913{
914 int i;
915
916 BUG_ON(mm == &init_mm);
917 WARN_ON_ONCE(mm == current->mm);
918
919 /* Ensure no CPUs are using this as their lazy tlb mm */
920 cleanup_lazy_tlbs(mm);
921
922 WARN_ON_ONCE(mm == current->active_mm);
923 mm_free_pgd(mm);
924 destroy_context(mm);
925 mmu_notifier_subscriptions_destroy(mm);
926 check_mm(mm);
927 put_user_ns(mm->user_ns);
928 mm_pasid_drop(mm);
929 mm_destroy_cid(mm);
930
931 for (i = 0; i < NR_MM_COUNTERS; i++)
932 percpu_counter_destroy(&mm->rss_stat[i]);
933 free_mm(mm);
934}
935EXPORT_SYMBOL_GPL(__mmdrop);
936
937static void mmdrop_async_fn(struct work_struct *work)
938{
939 struct mm_struct *mm;
940
941 mm = container_of(work, struct mm_struct, async_put_work);
942 __mmdrop(mm);
943}
944
945static void mmdrop_async(struct mm_struct *mm)
946{
947 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
948 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
949 schedule_work(&mm->async_put_work);
950 }
951}
952
953static inline void free_signal_struct(struct signal_struct *sig)
954{
955 taskstats_tgid_free(sig);
956 sched_autogroup_exit(sig);
957 /*
958 * __mmdrop is not safe to call from softirq context on x86 due to
959 * pgd_dtor so postpone it to the async context
960 */
961 if (sig->oom_mm)
962 mmdrop_async(sig->oom_mm);
963 kmem_cache_free(signal_cachep, sig);
964}
965
966static inline void put_signal_struct(struct signal_struct *sig)
967{
968 if (refcount_dec_and_test(&sig->sigcnt))
969 free_signal_struct(sig);
970}
971
972void __put_task_struct(struct task_struct *tsk)
973{
974 WARN_ON(!tsk->exit_state);
975 WARN_ON(refcount_read(&tsk->usage));
976 WARN_ON(tsk == current);
977
978 io_uring_free(tsk);
979 cgroup_free(tsk);
980 task_numa_free(tsk, true);
981 security_task_free(tsk);
982 bpf_task_storage_free(tsk);
983 exit_creds(tsk);
984 delayacct_tsk_free(tsk);
985 put_signal_struct(tsk->signal);
986 sched_core_free(tsk);
987 free_task(tsk);
988}
989EXPORT_SYMBOL_GPL(__put_task_struct);
990
991void __init __weak arch_task_cache_init(void) { }
992
993/*
994 * set_max_threads
995 */
996static void set_max_threads(unsigned int max_threads_suggested)
997{
998 u64 threads;
999 unsigned long nr_pages = totalram_pages();
1000
1001 /*
1002 * The number of threads shall be limited such that the thread
1003 * structures may only consume a small part of the available memory.
1004 */
1005 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1006 threads = MAX_THREADS;
1007 else
1008 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1009 (u64) THREAD_SIZE * 8UL);
1010
1011 if (threads > max_threads_suggested)
1012 threads = max_threads_suggested;
1013
1014 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1015}
1016
1017#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1018/* Initialized by the architecture: */
1019int arch_task_struct_size __read_mostly;
1020#endif
1021
1022#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1023static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1024{
1025 /* Fetch thread_struct whitelist for the architecture. */
1026 arch_thread_struct_whitelist(offset, size);
1027
1028 /*
1029 * Handle zero-sized whitelist or empty thread_struct, otherwise
1030 * adjust offset to position of thread_struct in task_struct.
1031 */
1032 if (unlikely(*size == 0))
1033 *offset = 0;
1034 else
1035 *offset += offsetof(struct task_struct, thread);
1036}
1037#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
1038
1039void __init fork_init(void)
1040{
1041 int i;
1042#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
1043#ifndef ARCH_MIN_TASKALIGN
1044#define ARCH_MIN_TASKALIGN 0
1045#endif
1046 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1047 unsigned long useroffset, usersize;
1048
1049 /* create a slab on which task_structs can be allocated */
1050 task_struct_whitelist(&useroffset, &usersize);
1051 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1052 arch_task_struct_size, align,
1053 SLAB_PANIC|SLAB_ACCOUNT,
1054 useroffset, usersize, NULL);
1055#endif
1056
1057 /* do the arch specific task caches init */
1058 arch_task_cache_init();
1059
1060 set_max_threads(MAX_THREADS);
1061
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1064 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1065 init_task.signal->rlim[RLIMIT_NPROC];
1066
1067 for (i = 0; i < UCOUNT_COUNTS; i++)
1068 init_user_ns.ucount_max[i] = max_threads/2;
1069
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1073 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1074
1075#ifdef CONFIG_VMAP_STACK
1076 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1077 NULL, free_vm_stack_cache);
1078#endif
1079
1080 scs_init();
1081
1082 lockdep_init_task(&init_task);
1083 uprobes_init();
1084}
1085
1086int __weak arch_dup_task_struct(struct task_struct *dst,
1087 struct task_struct *src)
1088{
1089 *dst = *src;
1090 return 0;
1091}
1092
1093void set_task_stack_end_magic(struct task_struct *tsk)
1094{
1095 unsigned long *stackend;
1096
1097 stackend = end_of_stack(tsk);
1098 *stackend = STACK_END_MAGIC; /* for overflow detection */
1099}
1100
1101static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1102{
1103 struct task_struct *tsk;
1104 int err;
1105
1106 if (node == NUMA_NO_NODE)
1107 node = tsk_fork_get_node(orig);
1108 tsk = alloc_task_struct_node(node);
1109 if (!tsk)
1110 return NULL;
1111
1112 err = arch_dup_task_struct(tsk, orig);
1113 if (err)
1114 goto free_tsk;
1115
1116 err = alloc_thread_stack_node(tsk, node);
1117 if (err)
1118 goto free_tsk;
1119
1120#ifdef CONFIG_THREAD_INFO_IN_TASK
1121 refcount_set(&tsk->stack_refcount, 1);
1122#endif
1123 account_kernel_stack(tsk, 1);
1124
1125 err = scs_prepare(tsk, node);
1126 if (err)
1127 goto free_stack;
1128
1129#ifdef CONFIG_SECCOMP
1130 /*
1131 * We must handle setting up seccomp filters once we're under
1132 * the sighand lock in case orig has changed between now and
1133 * then. Until then, filter must be NULL to avoid messing up
1134 * the usage counts on the error path calling free_task.
1135 */
1136 tsk->seccomp.filter = NULL;
1137#endif
1138
1139 setup_thread_stack(tsk, orig);
1140 clear_user_return_notifier(tsk);
1141 clear_tsk_need_resched(tsk);
1142 set_task_stack_end_magic(tsk);
1143 clear_syscall_work_syscall_user_dispatch(tsk);
1144
1145#ifdef CONFIG_STACKPROTECTOR
1146 tsk->stack_canary = get_random_canary();
1147#endif
1148 if (orig->cpus_ptr == &orig->cpus_mask)
1149 tsk->cpus_ptr = &tsk->cpus_mask;
1150 dup_user_cpus_ptr(tsk, orig, node);
1151
1152 /*
1153 * One for the user space visible state that goes away when reaped.
1154 * One for the scheduler.
1155 */
1156 refcount_set(&tsk->rcu_users, 2);
1157 /* One for the rcu users */
1158 refcount_set(&tsk->usage, 1);
1159#ifdef CONFIG_BLK_DEV_IO_TRACE
1160 tsk->btrace_seq = 0;
1161#endif
1162 tsk->splice_pipe = NULL;
1163 tsk->task_frag.page = NULL;
1164 tsk->wake_q.next = NULL;
1165 tsk->worker_private = NULL;
1166
1167 kcov_task_init(tsk);
1168 kmsan_task_create(tsk);
1169 kmap_local_fork(tsk);
1170
1171#ifdef CONFIG_FAULT_INJECTION
1172 tsk->fail_nth = 0;
1173#endif
1174
1175#ifdef CONFIG_BLK_CGROUP
1176 tsk->throttle_disk = NULL;
1177 tsk->use_memdelay = 0;
1178#endif
1179
1180#ifdef CONFIG_IOMMU_SVA
1181 tsk->pasid_activated = 0;
1182#endif
1183
1184#ifdef CONFIG_MEMCG
1185 tsk->active_memcg = NULL;
1186#endif
1187
1188#ifdef CONFIG_CPU_SUP_INTEL
1189 tsk->reported_split_lock = 0;
1190#endif
1191
1192#ifdef CONFIG_SCHED_MM_CID
1193 tsk->mm_cid = -1;
1194 tsk->last_mm_cid = -1;
1195 tsk->mm_cid_active = 0;
1196 tsk->migrate_from_cpu = -1;
1197#endif
1198 return tsk;
1199
1200free_stack:
1201 exit_task_stack_account(tsk);
1202 free_thread_stack(tsk);
1203free_tsk:
1204 free_task_struct(tsk);
1205 return NULL;
1206}
1207
1208__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1209
1210static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1211
1212static int __init coredump_filter_setup(char *s)
1213{
1214 default_dump_filter =
1215 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1216 MMF_DUMP_FILTER_MASK;
1217 return 1;
1218}
1219
1220__setup("coredump_filter=", coredump_filter_setup);
1221
1222#include <linux/init_task.h>
1223
1224static void mm_init_aio(struct mm_struct *mm)
1225{
1226#ifdef CONFIG_AIO
1227 spin_lock_init(&mm->ioctx_lock);
1228 mm->ioctx_table = NULL;
1229#endif
1230}
1231
1232static __always_inline void mm_clear_owner(struct mm_struct *mm,
1233 struct task_struct *p)
1234{
1235#ifdef CONFIG_MEMCG
1236 if (mm->owner == p)
1237 WRITE_ONCE(mm->owner, NULL);
1238#endif
1239}
1240
1241static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1242{
1243#ifdef CONFIG_MEMCG
1244 mm->owner = p;
1245#endif
1246}
1247
1248static void mm_init_uprobes_state(struct mm_struct *mm)
1249{
1250#ifdef CONFIG_UPROBES
1251 mm->uprobes_state.xol_area = NULL;
1252#endif
1253}
1254
1255static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1256 struct user_namespace *user_ns)
1257{
1258 int i;
1259
1260 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1261 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1262 atomic_set(&mm->mm_users, 1);
1263 atomic_set(&mm->mm_count, 1);
1264 seqcount_init(&mm->write_protect_seq);
1265 mmap_init_lock(mm);
1266 INIT_LIST_HEAD(&mm->mmlist);
1267#ifdef CONFIG_PER_VMA_LOCK
1268 mm->mm_lock_seq = 0;
1269#endif
1270 mm_pgtables_bytes_init(mm);
1271 mm->map_count = 0;
1272 mm->locked_vm = 0;
1273 atomic64_set(&mm->pinned_vm, 0);
1274 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1275 spin_lock_init(&mm->page_table_lock);
1276 spin_lock_init(&mm->arg_lock);
1277 mm_init_cpumask(mm);
1278 mm_init_aio(mm);
1279 mm_init_owner(mm, p);
1280 mm_pasid_init(mm);
1281 RCU_INIT_POINTER(mm->exe_file, NULL);
1282 mmu_notifier_subscriptions_init(mm);
1283 init_tlb_flush_pending(mm);
1284#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1285 mm->pmd_huge_pte = NULL;
1286#endif
1287 mm_init_uprobes_state(mm);
1288 hugetlb_count_init(mm);
1289
1290 if (current->mm) {
1291 mm->flags = current->mm->flags & MMF_INIT_MASK;
1292 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1293 } else {
1294 mm->flags = default_dump_filter;
1295 mm->def_flags = 0;
1296 }
1297
1298 if (mm_alloc_pgd(mm))
1299 goto fail_nopgd;
1300
1301 if (init_new_context(p, mm))
1302 goto fail_nocontext;
1303
1304 if (mm_alloc_cid(mm))
1305 goto fail_cid;
1306
1307 for (i = 0; i < NR_MM_COUNTERS; i++)
1308 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1309 goto fail_pcpu;
1310
1311 mm->user_ns = get_user_ns(user_ns);
1312 lru_gen_init_mm(mm);
1313 return mm;
1314
1315fail_pcpu:
1316 while (i > 0)
1317 percpu_counter_destroy(&mm->rss_stat[--i]);
1318 mm_destroy_cid(mm);
1319fail_cid:
1320 destroy_context(mm);
1321fail_nocontext:
1322 mm_free_pgd(mm);
1323fail_nopgd:
1324 free_mm(mm);
1325 return NULL;
1326}
1327
1328/*
1329 * Allocate and initialize an mm_struct.
1330 */
1331struct mm_struct *mm_alloc(void)
1332{
1333 struct mm_struct *mm;
1334
1335 mm = allocate_mm();
1336 if (!mm)
1337 return NULL;
1338
1339 memset(mm, 0, sizeof(*mm));
1340 return mm_init(mm, current, current_user_ns());
1341}
1342
1343static inline void __mmput(struct mm_struct *mm)
1344{
1345 VM_BUG_ON(atomic_read(&mm->mm_users));
1346
1347 uprobe_clear_state(mm);
1348 exit_aio(mm);
1349 ksm_exit(mm);
1350 khugepaged_exit(mm); /* must run before exit_mmap */
1351 exit_mmap(mm);
1352 mm_put_huge_zero_page(mm);
1353 set_mm_exe_file(mm, NULL);
1354 if (!list_empty(&mm->mmlist)) {
1355 spin_lock(&mmlist_lock);
1356 list_del(&mm->mmlist);
1357 spin_unlock(&mmlist_lock);
1358 }
1359 if (mm->binfmt)
1360 module_put(mm->binfmt->module);
1361 lru_gen_del_mm(mm);
1362 mmdrop(mm);
1363}
1364
1365/*
1366 * Decrement the use count and release all resources for an mm.
1367 */
1368void mmput(struct mm_struct *mm)
1369{
1370 might_sleep();
1371
1372 if (atomic_dec_and_test(&mm->mm_users))
1373 __mmput(mm);
1374}
1375EXPORT_SYMBOL_GPL(mmput);
1376
1377#ifdef CONFIG_MMU
1378static void mmput_async_fn(struct work_struct *work)
1379{
1380 struct mm_struct *mm = container_of(work, struct mm_struct,
1381 async_put_work);
1382
1383 __mmput(mm);
1384}
1385
1386void mmput_async(struct mm_struct *mm)
1387{
1388 if (atomic_dec_and_test(&mm->mm_users)) {
1389 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1390 schedule_work(&mm->async_put_work);
1391 }
1392}
1393EXPORT_SYMBOL_GPL(mmput_async);
1394#endif
1395
1396/**
1397 * set_mm_exe_file - change a reference to the mm's executable file
1398 *
1399 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1400 *
1401 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1402 * invocations: in mmput() nobody alive left, in execve task is single
1403 * threaded.
1404 *
1405 * Can only fail if new_exe_file != NULL.
1406 */
1407int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1408{
1409 struct file *old_exe_file;
1410
1411 /*
1412 * It is safe to dereference the exe_file without RCU as
1413 * this function is only called if nobody else can access
1414 * this mm -- see comment above for justification.
1415 */
1416 old_exe_file = rcu_dereference_raw(mm->exe_file);
1417
1418 if (new_exe_file) {
1419 /*
1420 * We expect the caller (i.e., sys_execve) to already denied
1421 * write access, so this is unlikely to fail.
1422 */
1423 if (unlikely(deny_write_access(new_exe_file)))
1424 return -EACCES;
1425 get_file(new_exe_file);
1426 }
1427 rcu_assign_pointer(mm->exe_file, new_exe_file);
1428 if (old_exe_file) {
1429 allow_write_access(old_exe_file);
1430 fput(old_exe_file);
1431 }
1432 return 0;
1433}
1434
1435/**
1436 * replace_mm_exe_file - replace a reference to the mm's executable file
1437 *
1438 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1439 * dealing with concurrent invocation and without grabbing the mmap lock in
1440 * write mode.
1441 *
1442 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1443 */
1444int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1445{
1446 struct vm_area_struct *vma;
1447 struct file *old_exe_file;
1448 int ret = 0;
1449
1450 /* Forbid mm->exe_file change if old file still mapped. */
1451 old_exe_file = get_mm_exe_file(mm);
1452 if (old_exe_file) {
1453 VMA_ITERATOR(vmi, mm, 0);
1454 mmap_read_lock(mm);
1455 for_each_vma(vmi, vma) {
1456 if (!vma->vm_file)
1457 continue;
1458 if (path_equal(&vma->vm_file->f_path,
1459 &old_exe_file->f_path)) {
1460 ret = -EBUSY;
1461 break;
1462 }
1463 }
1464 mmap_read_unlock(mm);
1465 fput(old_exe_file);
1466 if (ret)
1467 return ret;
1468 }
1469
1470 /* set the new file, lockless */
1471 ret = deny_write_access(new_exe_file);
1472 if (ret)
1473 return -EACCES;
1474 get_file(new_exe_file);
1475
1476 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1477 if (old_exe_file) {
1478 /*
1479 * Don't race with dup_mmap() getting the file and disallowing
1480 * write access while someone might open the file writable.
1481 */
1482 mmap_read_lock(mm);
1483 allow_write_access(old_exe_file);
1484 fput(old_exe_file);
1485 mmap_read_unlock(mm);
1486 }
1487 return 0;
1488}
1489
1490/**
1491 * get_mm_exe_file - acquire a reference to the mm's executable file
1492 *
1493 * Returns %NULL if mm has no associated executable file.
1494 * User must release file via fput().
1495 */
1496struct file *get_mm_exe_file(struct mm_struct *mm)
1497{
1498 struct file *exe_file;
1499
1500 rcu_read_lock();
1501 exe_file = rcu_dereference(mm->exe_file);
1502 if (exe_file && !get_file_rcu(exe_file))
1503 exe_file = NULL;
1504 rcu_read_unlock();
1505 return exe_file;
1506}
1507
1508/**
1509 * get_task_exe_file - acquire a reference to the task's executable file
1510 *
1511 * Returns %NULL if task's mm (if any) has no associated executable file or
1512 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1513 * User must release file via fput().
1514 */
1515struct file *get_task_exe_file(struct task_struct *task)
1516{
1517 struct file *exe_file = NULL;
1518 struct mm_struct *mm;
1519
1520 task_lock(task);
1521 mm = task->mm;
1522 if (mm) {
1523 if (!(task->flags & PF_KTHREAD))
1524 exe_file = get_mm_exe_file(mm);
1525 }
1526 task_unlock(task);
1527 return exe_file;
1528}
1529
1530/**
1531 * get_task_mm - acquire a reference to the task's mm
1532 *
1533 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1534 * this kernel workthread has transiently adopted a user mm with use_mm,
1535 * to do its AIO) is not set and if so returns a reference to it, after
1536 * bumping up the use count. User must release the mm via mmput()
1537 * after use. Typically used by /proc and ptrace.
1538 */
1539struct mm_struct *get_task_mm(struct task_struct *task)
1540{
1541 struct mm_struct *mm;
1542
1543 task_lock(task);
1544 mm = task->mm;
1545 if (mm) {
1546 if (task->flags & PF_KTHREAD)
1547 mm = NULL;
1548 else
1549 mmget(mm);
1550 }
1551 task_unlock(task);
1552 return mm;
1553}
1554EXPORT_SYMBOL_GPL(get_task_mm);
1555
1556struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1557{
1558 struct mm_struct *mm;
1559 int err;
1560
1561 err = down_read_killable(&task->signal->exec_update_lock);
1562 if (err)
1563 return ERR_PTR(err);
1564
1565 mm = get_task_mm(task);
1566 if (mm && mm != current->mm &&
1567 !ptrace_may_access(task, mode)) {
1568 mmput(mm);
1569 mm = ERR_PTR(-EACCES);
1570 }
1571 up_read(&task->signal->exec_update_lock);
1572
1573 return mm;
1574}
1575
1576static void complete_vfork_done(struct task_struct *tsk)
1577{
1578 struct completion *vfork;
1579
1580 task_lock(tsk);
1581 vfork = tsk->vfork_done;
1582 if (likely(vfork)) {
1583 tsk->vfork_done = NULL;
1584 complete(vfork);
1585 }
1586 task_unlock(tsk);
1587}
1588
1589static int wait_for_vfork_done(struct task_struct *child,
1590 struct completion *vfork)
1591{
1592 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1593 int killed;
1594
1595 cgroup_enter_frozen();
1596 killed = wait_for_completion_state(vfork, state);
1597 cgroup_leave_frozen(false);
1598
1599 if (killed) {
1600 task_lock(child);
1601 child->vfork_done = NULL;
1602 task_unlock(child);
1603 }
1604
1605 put_task_struct(child);
1606 return killed;
1607}
1608
1609/* Please note the differences between mmput and mm_release.
1610 * mmput is called whenever we stop holding onto a mm_struct,
1611 * error success whatever.
1612 *
1613 * mm_release is called after a mm_struct has been removed
1614 * from the current process.
1615 *
1616 * This difference is important for error handling, when we
1617 * only half set up a mm_struct for a new process and need to restore
1618 * the old one. Because we mmput the new mm_struct before
1619 * restoring the old one. . .
1620 * Eric Biederman 10 January 1998
1621 */
1622static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1623{
1624 uprobe_free_utask(tsk);
1625
1626 /* Get rid of any cached register state */
1627 deactivate_mm(tsk, mm);
1628
1629 /*
1630 * Signal userspace if we're not exiting with a core dump
1631 * because we want to leave the value intact for debugging
1632 * purposes.
1633 */
1634 if (tsk->clear_child_tid) {
1635 if (atomic_read(&mm->mm_users) > 1) {
1636 /*
1637 * We don't check the error code - if userspace has
1638 * not set up a proper pointer then tough luck.
1639 */
1640 put_user(0, tsk->clear_child_tid);
1641 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1642 1, NULL, NULL, 0, 0);
1643 }
1644 tsk->clear_child_tid = NULL;
1645 }
1646
1647 /*
1648 * All done, finally we can wake up parent and return this mm to him.
1649 * Also kthread_stop() uses this completion for synchronization.
1650 */
1651 if (tsk->vfork_done)
1652 complete_vfork_done(tsk);
1653}
1654
1655void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1656{
1657 futex_exit_release(tsk);
1658 mm_release(tsk, mm);
1659}
1660
1661void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1662{
1663 futex_exec_release(tsk);
1664 mm_release(tsk, mm);
1665}
1666
1667/**
1668 * dup_mm() - duplicates an existing mm structure
1669 * @tsk: the task_struct with which the new mm will be associated.
1670 * @oldmm: the mm to duplicate.
1671 *
1672 * Allocates a new mm structure and duplicates the provided @oldmm structure
1673 * content into it.
1674 *
1675 * Return: the duplicated mm or NULL on failure.
1676 */
1677static struct mm_struct *dup_mm(struct task_struct *tsk,
1678 struct mm_struct *oldmm)
1679{
1680 struct mm_struct *mm;
1681 int err;
1682
1683 mm = allocate_mm();
1684 if (!mm)
1685 goto fail_nomem;
1686
1687 memcpy(mm, oldmm, sizeof(*mm));
1688
1689 if (!mm_init(mm, tsk, mm->user_ns))
1690 goto fail_nomem;
1691
1692 err = dup_mmap(mm, oldmm);
1693 if (err)
1694 goto free_pt;
1695
1696 mm->hiwater_rss = get_mm_rss(mm);
1697 mm->hiwater_vm = mm->total_vm;
1698
1699 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1700 goto free_pt;
1701
1702 return mm;
1703
1704free_pt:
1705 /* don't put binfmt in mmput, we haven't got module yet */
1706 mm->binfmt = NULL;
1707 mm_init_owner(mm, NULL);
1708 mmput(mm);
1709
1710fail_nomem:
1711 return NULL;
1712}
1713
1714static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1715{
1716 struct mm_struct *mm, *oldmm;
1717
1718 tsk->min_flt = tsk->maj_flt = 0;
1719 tsk->nvcsw = tsk->nivcsw = 0;
1720#ifdef CONFIG_DETECT_HUNG_TASK
1721 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1722 tsk->last_switch_time = 0;
1723#endif
1724
1725 tsk->mm = NULL;
1726 tsk->active_mm = NULL;
1727
1728 /*
1729 * Are we cloning a kernel thread?
1730 *
1731 * We need to steal a active VM for that..
1732 */
1733 oldmm = current->mm;
1734 if (!oldmm)
1735 return 0;
1736
1737 if (clone_flags & CLONE_VM) {
1738 mmget(oldmm);
1739 mm = oldmm;
1740 } else {
1741 mm = dup_mm(tsk, current->mm);
1742 if (!mm)
1743 return -ENOMEM;
1744 }
1745
1746 tsk->mm = mm;
1747 tsk->active_mm = mm;
1748 sched_mm_cid_fork(tsk);
1749 return 0;
1750}
1751
1752static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1753{
1754 struct fs_struct *fs = current->fs;
1755 if (clone_flags & CLONE_FS) {
1756 /* tsk->fs is already what we want */
1757 spin_lock(&fs->lock);
1758 if (fs->in_exec) {
1759 spin_unlock(&fs->lock);
1760 return -EAGAIN;
1761 }
1762 fs->users++;
1763 spin_unlock(&fs->lock);
1764 return 0;
1765 }
1766 tsk->fs = copy_fs_struct(fs);
1767 if (!tsk->fs)
1768 return -ENOMEM;
1769 return 0;
1770}
1771
1772static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1773 int no_files)
1774{
1775 struct files_struct *oldf, *newf;
1776 int error = 0;
1777
1778 /*
1779 * A background process may not have any files ...
1780 */
1781 oldf = current->files;
1782 if (!oldf)
1783 goto out;
1784
1785 if (no_files) {
1786 tsk->files = NULL;
1787 goto out;
1788 }
1789
1790 if (clone_flags & CLONE_FILES) {
1791 atomic_inc(&oldf->count);
1792 goto out;
1793 }
1794
1795 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1796 if (!newf)
1797 goto out;
1798
1799 tsk->files = newf;
1800 error = 0;
1801out:
1802 return error;
1803}
1804
1805static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1806{
1807 struct sighand_struct *sig;
1808
1809 if (clone_flags & CLONE_SIGHAND) {
1810 refcount_inc(&current->sighand->count);
1811 return 0;
1812 }
1813 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1814 RCU_INIT_POINTER(tsk->sighand, sig);
1815 if (!sig)
1816 return -ENOMEM;
1817
1818 refcount_set(&sig->count, 1);
1819 spin_lock_irq(&current->sighand->siglock);
1820 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1821 spin_unlock_irq(&current->sighand->siglock);
1822
1823 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1824 if (clone_flags & CLONE_CLEAR_SIGHAND)
1825 flush_signal_handlers(tsk, 0);
1826
1827 return 0;
1828}
1829
1830void __cleanup_sighand(struct sighand_struct *sighand)
1831{
1832 if (refcount_dec_and_test(&sighand->count)) {
1833 signalfd_cleanup(sighand);
1834 /*
1835 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1836 * without an RCU grace period, see __lock_task_sighand().
1837 */
1838 kmem_cache_free(sighand_cachep, sighand);
1839 }
1840}
1841
1842/*
1843 * Initialize POSIX timer handling for a thread group.
1844 */
1845static void posix_cpu_timers_init_group(struct signal_struct *sig)
1846{
1847 struct posix_cputimers *pct = &sig->posix_cputimers;
1848 unsigned long cpu_limit;
1849
1850 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1851 posix_cputimers_group_init(pct, cpu_limit);
1852}
1853
1854static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1855{
1856 struct signal_struct *sig;
1857
1858 if (clone_flags & CLONE_THREAD)
1859 return 0;
1860
1861 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1862 tsk->signal = sig;
1863 if (!sig)
1864 return -ENOMEM;
1865
1866 sig->nr_threads = 1;
1867 sig->quick_threads = 1;
1868 atomic_set(&sig->live, 1);
1869 refcount_set(&sig->sigcnt, 1);
1870
1871 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1872 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1873 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1874
1875 init_waitqueue_head(&sig->wait_chldexit);
1876 sig->curr_target = tsk;
1877 init_sigpending(&sig->shared_pending);
1878 INIT_HLIST_HEAD(&sig->multiprocess);
1879 seqlock_init(&sig->stats_lock);
1880 prev_cputime_init(&sig->prev_cputime);
1881
1882#ifdef CONFIG_POSIX_TIMERS
1883 INIT_LIST_HEAD(&sig->posix_timers);
1884 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1885 sig->real_timer.function = it_real_fn;
1886#endif
1887
1888 task_lock(current->group_leader);
1889 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1890 task_unlock(current->group_leader);
1891
1892 posix_cpu_timers_init_group(sig);
1893
1894 tty_audit_fork(sig);
1895 sched_autogroup_fork(sig);
1896
1897 sig->oom_score_adj = current->signal->oom_score_adj;
1898 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1899
1900 mutex_init(&sig->cred_guard_mutex);
1901 init_rwsem(&sig->exec_update_lock);
1902
1903 return 0;
1904}
1905
1906static void copy_seccomp(struct task_struct *p)
1907{
1908#ifdef CONFIG_SECCOMP
1909 /*
1910 * Must be called with sighand->lock held, which is common to
1911 * all threads in the group. Holding cred_guard_mutex is not
1912 * needed because this new task is not yet running and cannot
1913 * be racing exec.
1914 */
1915 assert_spin_locked(&current->sighand->siglock);
1916
1917 /* Ref-count the new filter user, and assign it. */
1918 get_seccomp_filter(current);
1919 p->seccomp = current->seccomp;
1920
1921 /*
1922 * Explicitly enable no_new_privs here in case it got set
1923 * between the task_struct being duplicated and holding the
1924 * sighand lock. The seccomp state and nnp must be in sync.
1925 */
1926 if (task_no_new_privs(current))
1927 task_set_no_new_privs(p);
1928
1929 /*
1930 * If the parent gained a seccomp mode after copying thread
1931 * flags and between before we held the sighand lock, we have
1932 * to manually enable the seccomp thread flag here.
1933 */
1934 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1935 set_task_syscall_work(p, SECCOMP);
1936#endif
1937}
1938
1939SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1940{
1941 current->clear_child_tid = tidptr;
1942
1943 return task_pid_vnr(current);
1944}
1945
1946static void rt_mutex_init_task(struct task_struct *p)
1947{
1948 raw_spin_lock_init(&p->pi_lock);
1949#ifdef CONFIG_RT_MUTEXES
1950 p->pi_waiters = RB_ROOT_CACHED;
1951 p->pi_top_task = NULL;
1952 p->pi_blocked_on = NULL;
1953#endif
1954}
1955
1956static inline void init_task_pid_links(struct task_struct *task)
1957{
1958 enum pid_type type;
1959
1960 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1961 INIT_HLIST_NODE(&task->pid_links[type]);
1962}
1963
1964static inline void
1965init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1966{
1967 if (type == PIDTYPE_PID)
1968 task->thread_pid = pid;
1969 else
1970 task->signal->pids[type] = pid;
1971}
1972
1973static inline void rcu_copy_process(struct task_struct *p)
1974{
1975#ifdef CONFIG_PREEMPT_RCU
1976 p->rcu_read_lock_nesting = 0;
1977 p->rcu_read_unlock_special.s = 0;
1978 p->rcu_blocked_node = NULL;
1979 INIT_LIST_HEAD(&p->rcu_node_entry);
1980#endif /* #ifdef CONFIG_PREEMPT_RCU */
1981#ifdef CONFIG_TASKS_RCU
1982 p->rcu_tasks_holdout = false;
1983 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1984 p->rcu_tasks_idle_cpu = -1;
1985#endif /* #ifdef CONFIG_TASKS_RCU */
1986#ifdef CONFIG_TASKS_TRACE_RCU
1987 p->trc_reader_nesting = 0;
1988 p->trc_reader_special.s = 0;
1989 INIT_LIST_HEAD(&p->trc_holdout_list);
1990 INIT_LIST_HEAD(&p->trc_blkd_node);
1991#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1992}
1993
1994struct pid *pidfd_pid(const struct file *file)
1995{
1996 if (file->f_op == &pidfd_fops)
1997 return file->private_data;
1998
1999 return ERR_PTR(-EBADF);
2000}
2001
2002static int pidfd_release(struct inode *inode, struct file *file)
2003{
2004 struct pid *pid = file->private_data;
2005
2006 file->private_data = NULL;
2007 put_pid(pid);
2008 return 0;
2009}
2010
2011#ifdef CONFIG_PROC_FS
2012/**
2013 * pidfd_show_fdinfo - print information about a pidfd
2014 * @m: proc fdinfo file
2015 * @f: file referencing a pidfd
2016 *
2017 * Pid:
2018 * This function will print the pid that a given pidfd refers to in the
2019 * pid namespace of the procfs instance.
2020 * If the pid namespace of the process is not a descendant of the pid
2021 * namespace of the procfs instance 0 will be shown as its pid. This is
2022 * similar to calling getppid() on a process whose parent is outside of
2023 * its pid namespace.
2024 *
2025 * NSpid:
2026 * If pid namespaces are supported then this function will also print
2027 * the pid of a given pidfd refers to for all descendant pid namespaces
2028 * starting from the current pid namespace of the instance, i.e. the
2029 * Pid field and the first entry in the NSpid field will be identical.
2030 * If the pid namespace of the process is not a descendant of the pid
2031 * namespace of the procfs instance 0 will be shown as its first NSpid
2032 * entry and no others will be shown.
2033 * Note that this differs from the Pid and NSpid fields in
2034 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2035 * the pid namespace of the procfs instance. The difference becomes
2036 * obvious when sending around a pidfd between pid namespaces from a
2037 * different branch of the tree, i.e. where no ancestral relation is
2038 * present between the pid namespaces:
2039 * - create two new pid namespaces ns1 and ns2 in the initial pid
2040 * namespace (also take care to create new mount namespaces in the
2041 * new pid namespace and mount procfs)
2042 * - create a process with a pidfd in ns1
2043 * - send pidfd from ns1 to ns2
2044 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2045 * have exactly one entry, which is 0
2046 */
2047static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2048{
2049 struct pid *pid = f->private_data;
2050 struct pid_namespace *ns;
2051 pid_t nr = -1;
2052
2053 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2054 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2055 nr = pid_nr_ns(pid, ns);
2056 }
2057
2058 seq_put_decimal_ll(m, "Pid:\t", nr);
2059
2060#ifdef CONFIG_PID_NS
2061 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2062 if (nr > 0) {
2063 int i;
2064
2065 /* If nr is non-zero it means that 'pid' is valid and that
2066 * ns, i.e. the pid namespace associated with the procfs
2067 * instance, is in the pid namespace hierarchy of pid.
2068 * Start at one below the already printed level.
2069 */
2070 for (i = ns->level + 1; i <= pid->level; i++)
2071 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2072 }
2073#endif
2074 seq_putc(m, '\n');
2075}
2076#endif
2077
2078/*
2079 * Poll support for process exit notification.
2080 */
2081static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2082{
2083 struct pid *pid = file->private_data;
2084 __poll_t poll_flags = 0;
2085
2086 poll_wait(file, &pid->wait_pidfd, pts);
2087
2088 /*
2089 * Inform pollers only when the whole thread group exits.
2090 * If the thread group leader exits before all other threads in the
2091 * group, then poll(2) should block, similar to the wait(2) family.
2092 */
2093 if (thread_group_exited(pid))
2094 poll_flags = EPOLLIN | EPOLLRDNORM;
2095
2096 return poll_flags;
2097}
2098
2099const struct file_operations pidfd_fops = {
2100 .release = pidfd_release,
2101 .poll = pidfd_poll,
2102#ifdef CONFIG_PROC_FS
2103 .show_fdinfo = pidfd_show_fdinfo,
2104#endif
2105};
2106
2107/**
2108 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2109 * @pid: the struct pid for which to create a pidfd
2110 * @flags: flags of the new @pidfd
2111 * @pidfd: the pidfd to return
2112 *
2113 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2114 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2115
2116 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2117 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2118 * pidfd file are prepared.
2119 *
2120 * If this function returns successfully the caller is responsible to either
2121 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2122 * order to install the pidfd into its file descriptor table or they must use
2123 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2124 * respectively.
2125 *
2126 * This function is useful when a pidfd must already be reserved but there
2127 * might still be points of failure afterwards and the caller wants to ensure
2128 * that no pidfd is leaked into its file descriptor table.
2129 *
2130 * Return: On success, a reserved pidfd is returned from the function and a new
2131 * pidfd file is returned in the last argument to the function. On
2132 * error, a negative error code is returned from the function and the
2133 * last argument remains unchanged.
2134 */
2135static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2136{
2137 int pidfd;
2138 struct file *pidfd_file;
2139
2140 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2141 return -EINVAL;
2142
2143 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2144 if (pidfd < 0)
2145 return pidfd;
2146
2147 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2148 flags | O_RDWR | O_CLOEXEC);
2149 if (IS_ERR(pidfd_file)) {
2150 put_unused_fd(pidfd);
2151 return PTR_ERR(pidfd_file);
2152 }
2153 get_pid(pid); /* held by pidfd_file now */
2154 *ret = pidfd_file;
2155 return pidfd;
2156}
2157
2158/**
2159 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2160 * @pid: the struct pid for which to create a pidfd
2161 * @flags: flags of the new @pidfd
2162 * @pidfd: the pidfd to return
2163 *
2164 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2165 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2166 *
2167 * The helper verifies that @pid is used as a thread group leader.
2168 *
2169 * If this function returns successfully the caller is responsible to either
2170 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2171 * order to install the pidfd into its file descriptor table or they must use
2172 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2173 * respectively.
2174 *
2175 * This function is useful when a pidfd must already be reserved but there
2176 * might still be points of failure afterwards and the caller wants to ensure
2177 * that no pidfd is leaked into its file descriptor table.
2178 *
2179 * Return: On success, a reserved pidfd is returned from the function and a new
2180 * pidfd file is returned in the last argument to the function. On
2181 * error, a negative error code is returned from the function and the
2182 * last argument remains unchanged.
2183 */
2184int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2185{
2186 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2187 return -EINVAL;
2188
2189 return __pidfd_prepare(pid, flags, ret);
2190}
2191
2192static void __delayed_free_task(struct rcu_head *rhp)
2193{
2194 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2195
2196 free_task(tsk);
2197}
2198
2199static __always_inline void delayed_free_task(struct task_struct *tsk)
2200{
2201 if (IS_ENABLED(CONFIG_MEMCG))
2202 call_rcu(&tsk->rcu, __delayed_free_task);
2203 else
2204 free_task(tsk);
2205}
2206
2207static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2208{
2209 /* Skip if kernel thread */
2210 if (!tsk->mm)
2211 return;
2212
2213 /* Skip if spawning a thread or using vfork */
2214 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2215 return;
2216
2217 /* We need to synchronize with __set_oom_adj */
2218 mutex_lock(&oom_adj_mutex);
2219 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2220 /* Update the values in case they were changed after copy_signal */
2221 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2222 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2223 mutex_unlock(&oom_adj_mutex);
2224}
2225
2226#ifdef CONFIG_RV
2227static void rv_task_fork(struct task_struct *p)
2228{
2229 int i;
2230
2231 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2232 p->rv[i].da_mon.monitoring = false;
2233}
2234#else
2235#define rv_task_fork(p) do {} while (0)
2236#endif
2237
2238/*
2239 * This creates a new process as a copy of the old one,
2240 * but does not actually start it yet.
2241 *
2242 * It copies the registers, and all the appropriate
2243 * parts of the process environment (as per the clone
2244 * flags). The actual kick-off is left to the caller.
2245 */
2246__latent_entropy struct task_struct *copy_process(
2247 struct pid *pid,
2248 int trace,
2249 int node,
2250 struct kernel_clone_args *args)
2251{
2252 int pidfd = -1, retval;
2253 struct task_struct *p;
2254 struct multiprocess_signals delayed;
2255 struct file *pidfile = NULL;
2256 const u64 clone_flags = args->flags;
2257 struct nsproxy *nsp = current->nsproxy;
2258
2259 /*
2260 * Don't allow sharing the root directory with processes in a different
2261 * namespace
2262 */
2263 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2264 return ERR_PTR(-EINVAL);
2265
2266 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2267 return ERR_PTR(-EINVAL);
2268
2269 /*
2270 * Thread groups must share signals as well, and detached threads
2271 * can only be started up within the thread group.
2272 */
2273 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2274 return ERR_PTR(-EINVAL);
2275
2276 /*
2277 * Shared signal handlers imply shared VM. By way of the above,
2278 * thread groups also imply shared VM. Blocking this case allows
2279 * for various simplifications in other code.
2280 */
2281 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2282 return ERR_PTR(-EINVAL);
2283
2284 /*
2285 * Siblings of global init remain as zombies on exit since they are
2286 * not reaped by their parent (swapper). To solve this and to avoid
2287 * multi-rooted process trees, prevent global and container-inits
2288 * from creating siblings.
2289 */
2290 if ((clone_flags & CLONE_PARENT) &&
2291 current->signal->flags & SIGNAL_UNKILLABLE)
2292 return ERR_PTR(-EINVAL);
2293
2294 /*
2295 * If the new process will be in a different pid or user namespace
2296 * do not allow it to share a thread group with the forking task.
2297 */
2298 if (clone_flags & CLONE_THREAD) {
2299 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2300 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2301 return ERR_PTR(-EINVAL);
2302 }
2303
2304 if (clone_flags & CLONE_PIDFD) {
2305 /*
2306 * - CLONE_DETACHED is blocked so that we can potentially
2307 * reuse it later for CLONE_PIDFD.
2308 * - CLONE_THREAD is blocked until someone really needs it.
2309 */
2310 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2311 return ERR_PTR(-EINVAL);
2312 }
2313
2314 /*
2315 * Force any signals received before this point to be delivered
2316 * before the fork happens. Collect up signals sent to multiple
2317 * processes that happen during the fork and delay them so that
2318 * they appear to happen after the fork.
2319 */
2320 sigemptyset(&delayed.signal);
2321 INIT_HLIST_NODE(&delayed.node);
2322
2323 spin_lock_irq(&current->sighand->siglock);
2324 if (!(clone_flags & CLONE_THREAD))
2325 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2326 recalc_sigpending();
2327 spin_unlock_irq(&current->sighand->siglock);
2328 retval = -ERESTARTNOINTR;
2329 if (task_sigpending(current))
2330 goto fork_out;
2331
2332 retval = -ENOMEM;
2333 p = dup_task_struct(current, node);
2334 if (!p)
2335 goto fork_out;
2336 p->flags &= ~PF_KTHREAD;
2337 if (args->kthread)
2338 p->flags |= PF_KTHREAD;
2339 if (args->user_worker)
2340 p->flags |= PF_USER_WORKER;
2341 if (args->io_thread) {
2342 /*
2343 * Mark us an IO worker, and block any signal that isn't
2344 * fatal or STOP
2345 */
2346 p->flags |= PF_IO_WORKER;
2347 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2348 }
2349
2350 if (args->name)
2351 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2352
2353 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2354 /*
2355 * Clear TID on mm_release()?
2356 */
2357 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2358
2359 ftrace_graph_init_task(p);
2360
2361 rt_mutex_init_task(p);
2362
2363 lockdep_assert_irqs_enabled();
2364#ifdef CONFIG_PROVE_LOCKING
2365 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2366#endif
2367 retval = copy_creds(p, clone_flags);
2368 if (retval < 0)
2369 goto bad_fork_free;
2370
2371 retval = -EAGAIN;
2372 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2373 if (p->real_cred->user != INIT_USER &&
2374 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2375 goto bad_fork_cleanup_count;
2376 }
2377 current->flags &= ~PF_NPROC_EXCEEDED;
2378
2379 /*
2380 * If multiple threads are within copy_process(), then this check
2381 * triggers too late. This doesn't hurt, the check is only there
2382 * to stop root fork bombs.
2383 */
2384 retval = -EAGAIN;
2385 if (data_race(nr_threads >= max_threads))
2386 goto bad_fork_cleanup_count;
2387
2388 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2389 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2390 p->flags |= PF_FORKNOEXEC;
2391 INIT_LIST_HEAD(&p->children);
2392 INIT_LIST_HEAD(&p->sibling);
2393 rcu_copy_process(p);
2394 p->vfork_done = NULL;
2395 spin_lock_init(&p->alloc_lock);
2396
2397 init_sigpending(&p->pending);
2398
2399 p->utime = p->stime = p->gtime = 0;
2400#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2401 p->utimescaled = p->stimescaled = 0;
2402#endif
2403 prev_cputime_init(&p->prev_cputime);
2404
2405#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2406 seqcount_init(&p->vtime.seqcount);
2407 p->vtime.starttime = 0;
2408 p->vtime.state = VTIME_INACTIVE;
2409#endif
2410
2411#ifdef CONFIG_IO_URING
2412 p->io_uring = NULL;
2413#endif
2414
2415#if defined(SPLIT_RSS_COUNTING)
2416 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2417#endif
2418
2419 p->default_timer_slack_ns = current->timer_slack_ns;
2420
2421#ifdef CONFIG_PSI
2422 p->psi_flags = 0;
2423#endif
2424
2425 task_io_accounting_init(&p->ioac);
2426 acct_clear_integrals(p);
2427
2428 posix_cputimers_init(&p->posix_cputimers);
2429
2430 p->io_context = NULL;
2431 audit_set_context(p, NULL);
2432 cgroup_fork(p);
2433 if (args->kthread) {
2434 if (!set_kthread_struct(p))
2435 goto bad_fork_cleanup_delayacct;
2436 }
2437#ifdef CONFIG_NUMA
2438 p->mempolicy = mpol_dup(p->mempolicy);
2439 if (IS_ERR(p->mempolicy)) {
2440 retval = PTR_ERR(p->mempolicy);
2441 p->mempolicy = NULL;
2442 goto bad_fork_cleanup_delayacct;
2443 }
2444#endif
2445#ifdef CONFIG_CPUSETS
2446 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2447 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2448 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2449#endif
2450#ifdef CONFIG_TRACE_IRQFLAGS
2451 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2452 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2453 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2454 p->softirqs_enabled = 1;
2455 p->softirq_context = 0;
2456#endif
2457
2458 p->pagefault_disabled = 0;
2459
2460#ifdef CONFIG_LOCKDEP
2461 lockdep_init_task(p);
2462#endif
2463
2464#ifdef CONFIG_DEBUG_MUTEXES
2465 p->blocked_on = NULL; /* not blocked yet */
2466#endif
2467#ifdef CONFIG_BCACHE
2468 p->sequential_io = 0;
2469 p->sequential_io_avg = 0;
2470#endif
2471#ifdef CONFIG_BPF_SYSCALL
2472 RCU_INIT_POINTER(p->bpf_storage, NULL);
2473 p->bpf_ctx = NULL;
2474#endif
2475
2476 /* Perform scheduler related setup. Assign this task to a CPU. */
2477 retval = sched_fork(clone_flags, p);
2478 if (retval)
2479 goto bad_fork_cleanup_policy;
2480
2481 retval = perf_event_init_task(p, clone_flags);
2482 if (retval)
2483 goto bad_fork_cleanup_policy;
2484 retval = audit_alloc(p);
2485 if (retval)
2486 goto bad_fork_cleanup_perf;
2487 /* copy all the process information */
2488 shm_init_task(p);
2489 retval = security_task_alloc(p, clone_flags);
2490 if (retval)
2491 goto bad_fork_cleanup_audit;
2492 retval = copy_semundo(clone_flags, p);
2493 if (retval)
2494 goto bad_fork_cleanup_security;
2495 retval = copy_files(clone_flags, p, args->no_files);
2496 if (retval)
2497 goto bad_fork_cleanup_semundo;
2498 retval = copy_fs(clone_flags, p);
2499 if (retval)
2500 goto bad_fork_cleanup_files;
2501 retval = copy_sighand(clone_flags, p);
2502 if (retval)
2503 goto bad_fork_cleanup_fs;
2504 retval = copy_signal(clone_flags, p);
2505 if (retval)
2506 goto bad_fork_cleanup_sighand;
2507 retval = copy_mm(clone_flags, p);
2508 if (retval)
2509 goto bad_fork_cleanup_signal;
2510 retval = copy_namespaces(clone_flags, p);
2511 if (retval)
2512 goto bad_fork_cleanup_mm;
2513 retval = copy_io(clone_flags, p);
2514 if (retval)
2515 goto bad_fork_cleanup_namespaces;
2516 retval = copy_thread(p, args);
2517 if (retval)
2518 goto bad_fork_cleanup_io;
2519
2520 if (args->ignore_signals)
2521 ignore_signals(p);
2522
2523 stackleak_task_init(p);
2524
2525 if (pid != &init_struct_pid) {
2526 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2527 args->set_tid_size);
2528 if (IS_ERR(pid)) {
2529 retval = PTR_ERR(pid);
2530 goto bad_fork_cleanup_thread;
2531 }
2532 }
2533
2534 /*
2535 * This has to happen after we've potentially unshared the file
2536 * descriptor table (so that the pidfd doesn't leak into the child
2537 * if the fd table isn't shared).
2538 */
2539 if (clone_flags & CLONE_PIDFD) {
2540 /* Note that no task has been attached to @pid yet. */
2541 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2542 if (retval < 0)
2543 goto bad_fork_free_pid;
2544 pidfd = retval;
2545
2546 retval = put_user(pidfd, args->pidfd);
2547 if (retval)
2548 goto bad_fork_put_pidfd;
2549 }
2550
2551#ifdef CONFIG_BLOCK
2552 p->plug = NULL;
2553#endif
2554 futex_init_task(p);
2555
2556 /*
2557 * sigaltstack should be cleared when sharing the same VM
2558 */
2559 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2560 sas_ss_reset(p);
2561
2562 /*
2563 * Syscall tracing and stepping should be turned off in the
2564 * child regardless of CLONE_PTRACE.
2565 */
2566 user_disable_single_step(p);
2567 clear_task_syscall_work(p, SYSCALL_TRACE);
2568#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2569 clear_task_syscall_work(p, SYSCALL_EMU);
2570#endif
2571 clear_tsk_latency_tracing(p);
2572
2573 /* ok, now we should be set up.. */
2574 p->pid = pid_nr(pid);
2575 if (clone_flags & CLONE_THREAD) {
2576 p->group_leader = current->group_leader;
2577 p->tgid = current->tgid;
2578 } else {
2579 p->group_leader = p;
2580 p->tgid = p->pid;
2581 }
2582
2583 p->nr_dirtied = 0;
2584 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2585 p->dirty_paused_when = 0;
2586
2587 p->pdeath_signal = 0;
2588 INIT_LIST_HEAD(&p->thread_group);
2589 p->task_works = NULL;
2590 clear_posix_cputimers_work(p);
2591
2592#ifdef CONFIG_KRETPROBES
2593 p->kretprobe_instances.first = NULL;
2594#endif
2595#ifdef CONFIG_RETHOOK
2596 p->rethooks.first = NULL;
2597#endif
2598
2599 /*
2600 * Ensure that the cgroup subsystem policies allow the new process to be
2601 * forked. It should be noted that the new process's css_set can be changed
2602 * between here and cgroup_post_fork() if an organisation operation is in
2603 * progress.
2604 */
2605 retval = cgroup_can_fork(p, args);
2606 if (retval)
2607 goto bad_fork_put_pidfd;
2608
2609 /*
2610 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2611 * the new task on the correct runqueue. All this *before* the task
2612 * becomes visible.
2613 *
2614 * This isn't part of ->can_fork() because while the re-cloning is
2615 * cgroup specific, it unconditionally needs to place the task on a
2616 * runqueue.
2617 */
2618 sched_cgroup_fork(p, args);
2619
2620 /*
2621 * From this point on we must avoid any synchronous user-space
2622 * communication until we take the tasklist-lock. In particular, we do
2623 * not want user-space to be able to predict the process start-time by
2624 * stalling fork(2) after we recorded the start_time but before it is
2625 * visible to the system.
2626 */
2627
2628 p->start_time = ktime_get_ns();
2629 p->start_boottime = ktime_get_boottime_ns();
2630
2631 /*
2632 * Make it visible to the rest of the system, but dont wake it up yet.
2633 * Need tasklist lock for parent etc handling!
2634 */
2635 write_lock_irq(&tasklist_lock);
2636
2637 /* CLONE_PARENT re-uses the old parent */
2638 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2639 p->real_parent = current->real_parent;
2640 p->parent_exec_id = current->parent_exec_id;
2641 if (clone_flags & CLONE_THREAD)
2642 p->exit_signal = -1;
2643 else
2644 p->exit_signal = current->group_leader->exit_signal;
2645 } else {
2646 p->real_parent = current;
2647 p->parent_exec_id = current->self_exec_id;
2648 p->exit_signal = args->exit_signal;
2649 }
2650
2651 klp_copy_process(p);
2652
2653 sched_core_fork(p);
2654
2655 spin_lock(&current->sighand->siglock);
2656
2657 rv_task_fork(p);
2658
2659 rseq_fork(p, clone_flags);
2660
2661 /* Don't start children in a dying pid namespace */
2662 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2663 retval = -ENOMEM;
2664 goto bad_fork_cancel_cgroup;
2665 }
2666
2667 /* Let kill terminate clone/fork in the middle */
2668 if (fatal_signal_pending(current)) {
2669 retval = -EINTR;
2670 goto bad_fork_cancel_cgroup;
2671 }
2672
2673 /* No more failure paths after this point. */
2674
2675 /*
2676 * Copy seccomp details explicitly here, in case they were changed
2677 * before holding sighand lock.
2678 */
2679 copy_seccomp(p);
2680
2681 init_task_pid_links(p);
2682 if (likely(p->pid)) {
2683 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2684
2685 init_task_pid(p, PIDTYPE_PID, pid);
2686 if (thread_group_leader(p)) {
2687 init_task_pid(p, PIDTYPE_TGID, pid);
2688 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2689 init_task_pid(p, PIDTYPE_SID, task_session(current));
2690
2691 if (is_child_reaper(pid)) {
2692 ns_of_pid(pid)->child_reaper = p;
2693 p->signal->flags |= SIGNAL_UNKILLABLE;
2694 }
2695 p->signal->shared_pending.signal = delayed.signal;
2696 p->signal->tty = tty_kref_get(current->signal->tty);
2697 /*
2698 * Inherit has_child_subreaper flag under the same
2699 * tasklist_lock with adding child to the process tree
2700 * for propagate_has_child_subreaper optimization.
2701 */
2702 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2703 p->real_parent->signal->is_child_subreaper;
2704 list_add_tail(&p->sibling, &p->real_parent->children);
2705 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2706 attach_pid(p, PIDTYPE_TGID);
2707 attach_pid(p, PIDTYPE_PGID);
2708 attach_pid(p, PIDTYPE_SID);
2709 __this_cpu_inc(process_counts);
2710 } else {
2711 current->signal->nr_threads++;
2712 current->signal->quick_threads++;
2713 atomic_inc(&current->signal->live);
2714 refcount_inc(&current->signal->sigcnt);
2715 task_join_group_stop(p);
2716 list_add_tail_rcu(&p->thread_group,
2717 &p->group_leader->thread_group);
2718 list_add_tail_rcu(&p->thread_node,
2719 &p->signal->thread_head);
2720 }
2721 attach_pid(p, PIDTYPE_PID);
2722 nr_threads++;
2723 }
2724 total_forks++;
2725 hlist_del_init(&delayed.node);
2726 spin_unlock(&current->sighand->siglock);
2727 syscall_tracepoint_update(p);
2728 write_unlock_irq(&tasklist_lock);
2729
2730 if (pidfile)
2731 fd_install(pidfd, pidfile);
2732
2733 proc_fork_connector(p);
2734 sched_post_fork(p);
2735 cgroup_post_fork(p, args);
2736 perf_event_fork(p);
2737
2738 trace_task_newtask(p, clone_flags);
2739 uprobe_copy_process(p, clone_flags);
2740 user_events_fork(p, clone_flags);
2741
2742 copy_oom_score_adj(clone_flags, p);
2743
2744 return p;
2745
2746bad_fork_cancel_cgroup:
2747 sched_core_free(p);
2748 spin_unlock(&current->sighand->siglock);
2749 write_unlock_irq(&tasklist_lock);
2750 cgroup_cancel_fork(p, args);
2751bad_fork_put_pidfd:
2752 if (clone_flags & CLONE_PIDFD) {
2753 fput(pidfile);
2754 put_unused_fd(pidfd);
2755 }
2756bad_fork_free_pid:
2757 if (pid != &init_struct_pid)
2758 free_pid(pid);
2759bad_fork_cleanup_thread:
2760 exit_thread(p);
2761bad_fork_cleanup_io:
2762 if (p->io_context)
2763 exit_io_context(p);
2764bad_fork_cleanup_namespaces:
2765 exit_task_namespaces(p);
2766bad_fork_cleanup_mm:
2767 if (p->mm) {
2768 mm_clear_owner(p->mm, p);
2769 mmput(p->mm);
2770 }
2771bad_fork_cleanup_signal:
2772 if (!(clone_flags & CLONE_THREAD))
2773 free_signal_struct(p->signal);
2774bad_fork_cleanup_sighand:
2775 __cleanup_sighand(p->sighand);
2776bad_fork_cleanup_fs:
2777 exit_fs(p); /* blocking */
2778bad_fork_cleanup_files:
2779 exit_files(p); /* blocking */
2780bad_fork_cleanup_semundo:
2781 exit_sem(p);
2782bad_fork_cleanup_security:
2783 security_task_free(p);
2784bad_fork_cleanup_audit:
2785 audit_free(p);
2786bad_fork_cleanup_perf:
2787 perf_event_free_task(p);
2788bad_fork_cleanup_policy:
2789 lockdep_free_task(p);
2790#ifdef CONFIG_NUMA
2791 mpol_put(p->mempolicy);
2792#endif
2793bad_fork_cleanup_delayacct:
2794 delayacct_tsk_free(p);
2795bad_fork_cleanup_count:
2796 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2797 exit_creds(p);
2798bad_fork_free:
2799 WRITE_ONCE(p->__state, TASK_DEAD);
2800 exit_task_stack_account(p);
2801 put_task_stack(p);
2802 delayed_free_task(p);
2803fork_out:
2804 spin_lock_irq(&current->sighand->siglock);
2805 hlist_del_init(&delayed.node);
2806 spin_unlock_irq(&current->sighand->siglock);
2807 return ERR_PTR(retval);
2808}
2809
2810static inline void init_idle_pids(struct task_struct *idle)
2811{
2812 enum pid_type type;
2813
2814 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2815 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2816 init_task_pid(idle, type, &init_struct_pid);
2817 }
2818}
2819
2820static int idle_dummy(void *dummy)
2821{
2822 /* This function is never called */
2823 return 0;
2824}
2825
2826struct task_struct * __init fork_idle(int cpu)
2827{
2828 struct task_struct *task;
2829 struct kernel_clone_args args = {
2830 .flags = CLONE_VM,
2831 .fn = &idle_dummy,
2832 .fn_arg = NULL,
2833 .kthread = 1,
2834 .idle = 1,
2835 };
2836
2837 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2838 if (!IS_ERR(task)) {
2839 init_idle_pids(task);
2840 init_idle(task, cpu);
2841 }
2842
2843 return task;
2844}
2845
2846/*
2847 * This is like kernel_clone(), but shaved down and tailored to just
2848 * creating io_uring workers. It returns a created task, or an error pointer.
2849 * The returned task is inactive, and the caller must fire it up through
2850 * wake_up_new_task(p). All signals are blocked in the created task.
2851 */
2852struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2853{
2854 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2855 CLONE_IO;
2856 struct kernel_clone_args args = {
2857 .flags = ((lower_32_bits(flags) | CLONE_VM |
2858 CLONE_UNTRACED) & ~CSIGNAL),
2859 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2860 .fn = fn,
2861 .fn_arg = arg,
2862 .io_thread = 1,
2863 .user_worker = 1,
2864 };
2865
2866 return copy_process(NULL, 0, node, &args);
2867}
2868
2869/*
2870 * Ok, this is the main fork-routine.
2871 *
2872 * It copies the process, and if successful kick-starts
2873 * it and waits for it to finish using the VM if required.
2874 *
2875 * args->exit_signal is expected to be checked for sanity by the caller.
2876 */
2877pid_t kernel_clone(struct kernel_clone_args *args)
2878{
2879 u64 clone_flags = args->flags;
2880 struct completion vfork;
2881 struct pid *pid;
2882 struct task_struct *p;
2883 int trace = 0;
2884 pid_t nr;
2885
2886 /*
2887 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2888 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2889 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2890 * field in struct clone_args and it still doesn't make sense to have
2891 * them both point at the same memory location. Performing this check
2892 * here has the advantage that we don't need to have a separate helper
2893 * to check for legacy clone().
2894 */
2895 if ((args->flags & CLONE_PIDFD) &&
2896 (args->flags & CLONE_PARENT_SETTID) &&
2897 (args->pidfd == args->parent_tid))
2898 return -EINVAL;
2899
2900 /*
2901 * Determine whether and which event to report to ptracer. When
2902 * called from kernel_thread or CLONE_UNTRACED is explicitly
2903 * requested, no event is reported; otherwise, report if the event
2904 * for the type of forking is enabled.
2905 */
2906 if (!(clone_flags & CLONE_UNTRACED)) {
2907 if (clone_flags & CLONE_VFORK)
2908 trace = PTRACE_EVENT_VFORK;
2909 else if (args->exit_signal != SIGCHLD)
2910 trace = PTRACE_EVENT_CLONE;
2911 else
2912 trace = PTRACE_EVENT_FORK;
2913
2914 if (likely(!ptrace_event_enabled(current, trace)))
2915 trace = 0;
2916 }
2917
2918 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2919 add_latent_entropy();
2920
2921 if (IS_ERR(p))
2922 return PTR_ERR(p);
2923
2924 /*
2925 * Do this prior waking up the new thread - the thread pointer
2926 * might get invalid after that point, if the thread exits quickly.
2927 */
2928 trace_sched_process_fork(current, p);
2929
2930 pid = get_task_pid(p, PIDTYPE_PID);
2931 nr = pid_vnr(pid);
2932
2933 if (clone_flags & CLONE_PARENT_SETTID)
2934 put_user(nr, args->parent_tid);
2935
2936 if (clone_flags & CLONE_VFORK) {
2937 p->vfork_done = &vfork;
2938 init_completion(&vfork);
2939 get_task_struct(p);
2940 }
2941
2942 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2943 /* lock the task to synchronize with memcg migration */
2944 task_lock(p);
2945 lru_gen_add_mm(p->mm);
2946 task_unlock(p);
2947 }
2948
2949 wake_up_new_task(p);
2950
2951 /* forking complete and child started to run, tell ptracer */
2952 if (unlikely(trace))
2953 ptrace_event_pid(trace, pid);
2954
2955 if (clone_flags & CLONE_VFORK) {
2956 if (!wait_for_vfork_done(p, &vfork))
2957 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2958 }
2959
2960 put_pid(pid);
2961 return nr;
2962}
2963
2964/*
2965 * Create a kernel thread.
2966 */
2967pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2968 unsigned long flags)
2969{
2970 struct kernel_clone_args args = {
2971 .flags = ((lower_32_bits(flags) | CLONE_VM |
2972 CLONE_UNTRACED) & ~CSIGNAL),
2973 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2974 .fn = fn,
2975 .fn_arg = arg,
2976 .name = name,
2977 .kthread = 1,
2978 };
2979
2980 return kernel_clone(&args);
2981}
2982
2983/*
2984 * Create a user mode thread.
2985 */
2986pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2987{
2988 struct kernel_clone_args args = {
2989 .flags = ((lower_32_bits(flags) | CLONE_VM |
2990 CLONE_UNTRACED) & ~CSIGNAL),
2991 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2992 .fn = fn,
2993 .fn_arg = arg,
2994 };
2995
2996 return kernel_clone(&args);
2997}
2998
2999#ifdef __ARCH_WANT_SYS_FORK
3000SYSCALL_DEFINE0(fork)
3001{
3002#ifdef CONFIG_MMU
3003 struct kernel_clone_args args = {
3004 .exit_signal = SIGCHLD,
3005 };
3006
3007 return kernel_clone(&args);
3008#else
3009 /* can not support in nommu mode */
3010 return -EINVAL;
3011#endif
3012}
3013#endif
3014
3015#ifdef __ARCH_WANT_SYS_VFORK
3016SYSCALL_DEFINE0(vfork)
3017{
3018 struct kernel_clone_args args = {
3019 .flags = CLONE_VFORK | CLONE_VM,
3020 .exit_signal = SIGCHLD,
3021 };
3022
3023 return kernel_clone(&args);
3024}
3025#endif
3026
3027#ifdef __ARCH_WANT_SYS_CLONE
3028#ifdef CONFIG_CLONE_BACKWARDS
3029SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3030 int __user *, parent_tidptr,
3031 unsigned long, tls,
3032 int __user *, child_tidptr)
3033#elif defined(CONFIG_CLONE_BACKWARDS2)
3034SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3035 int __user *, parent_tidptr,
3036 int __user *, child_tidptr,
3037 unsigned long, tls)
3038#elif defined(CONFIG_CLONE_BACKWARDS3)
3039SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3040 int, stack_size,
3041 int __user *, parent_tidptr,
3042 int __user *, child_tidptr,
3043 unsigned long, tls)
3044#else
3045SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3046 int __user *, parent_tidptr,
3047 int __user *, child_tidptr,
3048 unsigned long, tls)
3049#endif
3050{
3051 struct kernel_clone_args args = {
3052 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3053 .pidfd = parent_tidptr,
3054 .child_tid = child_tidptr,
3055 .parent_tid = parent_tidptr,
3056 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3057 .stack = newsp,
3058 .tls = tls,
3059 };
3060
3061 return kernel_clone(&args);
3062}
3063#endif
3064
3065#ifdef __ARCH_WANT_SYS_CLONE3
3066
3067noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3068 struct clone_args __user *uargs,
3069 size_t usize)
3070{
3071 int err;
3072 struct clone_args args;
3073 pid_t *kset_tid = kargs->set_tid;
3074
3075 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3076 CLONE_ARGS_SIZE_VER0);
3077 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3078 CLONE_ARGS_SIZE_VER1);
3079 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3080 CLONE_ARGS_SIZE_VER2);
3081 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3082
3083 if (unlikely(usize > PAGE_SIZE))
3084 return -E2BIG;
3085 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3086 return -EINVAL;
3087
3088 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3089 if (err)
3090 return err;
3091
3092 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3093 return -EINVAL;
3094
3095 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3096 return -EINVAL;
3097
3098 if (unlikely(args.set_tid && args.set_tid_size == 0))
3099 return -EINVAL;
3100
3101 /*
3102 * Verify that higher 32bits of exit_signal are unset and that
3103 * it is a valid signal
3104 */
3105 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3106 !valid_signal(args.exit_signal)))
3107 return -EINVAL;
3108
3109 if ((args.flags & CLONE_INTO_CGROUP) &&
3110 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3111 return -EINVAL;
3112
3113 *kargs = (struct kernel_clone_args){
3114 .flags = args.flags,
3115 .pidfd = u64_to_user_ptr(args.pidfd),
3116 .child_tid = u64_to_user_ptr(args.child_tid),
3117 .parent_tid = u64_to_user_ptr(args.parent_tid),
3118 .exit_signal = args.exit_signal,
3119 .stack = args.stack,
3120 .stack_size = args.stack_size,
3121 .tls = args.tls,
3122 .set_tid_size = args.set_tid_size,
3123 .cgroup = args.cgroup,
3124 };
3125
3126 if (args.set_tid &&
3127 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3128 (kargs->set_tid_size * sizeof(pid_t))))
3129 return -EFAULT;
3130
3131 kargs->set_tid = kset_tid;
3132
3133 return 0;
3134}
3135
3136/**
3137 * clone3_stack_valid - check and prepare stack
3138 * @kargs: kernel clone args
3139 *
3140 * Verify that the stack arguments userspace gave us are sane.
3141 * In addition, set the stack direction for userspace since it's easy for us to
3142 * determine.
3143 */
3144static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3145{
3146 if (kargs->stack == 0) {
3147 if (kargs->stack_size > 0)
3148 return false;
3149 } else {
3150 if (kargs->stack_size == 0)
3151 return false;
3152
3153 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3154 return false;
3155
3156#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
3157 kargs->stack += kargs->stack_size;
3158#endif
3159 }
3160
3161 return true;
3162}
3163
3164static bool clone3_args_valid(struct kernel_clone_args *kargs)
3165{
3166 /* Verify that no unknown flags are passed along. */
3167 if (kargs->flags &
3168 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3169 return false;
3170
3171 /*
3172 * - make the CLONE_DETACHED bit reusable for clone3
3173 * - make the CSIGNAL bits reusable for clone3
3174 */
3175 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3176 return false;
3177
3178 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3179 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3180 return false;
3181
3182 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3183 kargs->exit_signal)
3184 return false;
3185
3186 if (!clone3_stack_valid(kargs))
3187 return false;
3188
3189 return true;
3190}
3191
3192/**
3193 * clone3 - create a new process with specific properties
3194 * @uargs: argument structure
3195 * @size: size of @uargs
3196 *
3197 * clone3() is the extensible successor to clone()/clone2().
3198 * It takes a struct as argument that is versioned by its size.
3199 *
3200 * Return: On success, a positive PID for the child process.
3201 * On error, a negative errno number.
3202 */
3203SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3204{
3205 int err;
3206
3207 struct kernel_clone_args kargs;
3208 pid_t set_tid[MAX_PID_NS_LEVEL];
3209
3210 kargs.set_tid = set_tid;
3211
3212 err = copy_clone_args_from_user(&kargs, uargs, size);
3213 if (err)
3214 return err;
3215
3216 if (!clone3_args_valid(&kargs))
3217 return -EINVAL;
3218
3219 return kernel_clone(&kargs);
3220}
3221#endif
3222
3223void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3224{
3225 struct task_struct *leader, *parent, *child;
3226 int res;
3227
3228 read_lock(&tasklist_lock);
3229 leader = top = top->group_leader;
3230down:
3231 for_each_thread(leader, parent) {
3232 list_for_each_entry(child, &parent->children, sibling) {
3233 res = visitor(child, data);
3234 if (res) {
3235 if (res < 0)
3236 goto out;
3237 leader = child;
3238 goto down;
3239 }
3240up:
3241 ;
3242 }
3243 }
3244
3245 if (leader != top) {
3246 child = leader;
3247 parent = child->real_parent;
3248 leader = parent->group_leader;
3249 goto up;
3250 }
3251out:
3252 read_unlock(&tasklist_lock);
3253}
3254
3255#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3256#define ARCH_MIN_MMSTRUCT_ALIGN 0
3257#endif
3258
3259static void sighand_ctor(void *data)
3260{
3261 struct sighand_struct *sighand = data;
3262
3263 spin_lock_init(&sighand->siglock);
3264 init_waitqueue_head(&sighand->signalfd_wqh);
3265}
3266
3267void __init mm_cache_init(void)
3268{
3269 unsigned int mm_size;
3270
3271 /*
3272 * The mm_cpumask is located at the end of mm_struct, and is
3273 * dynamically sized based on the maximum CPU number this system
3274 * can have, taking hotplug into account (nr_cpu_ids).
3275 */
3276 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3277
3278 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3279 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3280 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3281 offsetof(struct mm_struct, saved_auxv),
3282 sizeof_field(struct mm_struct, saved_auxv),
3283 NULL);
3284}
3285
3286void __init proc_caches_init(void)
3287{
3288 sighand_cachep = kmem_cache_create("sighand_cache",
3289 sizeof(struct sighand_struct), 0,
3290 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3291 SLAB_ACCOUNT, sighand_ctor);
3292 signal_cachep = kmem_cache_create("signal_cache",
3293 sizeof(struct signal_struct), 0,
3294 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3295 NULL);
3296 files_cachep = kmem_cache_create("files_cache",
3297 sizeof(struct files_struct), 0,
3298 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3299 NULL);
3300 fs_cachep = kmem_cache_create("fs_cache",
3301 sizeof(struct fs_struct), 0,
3302 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3303 NULL);
3304
3305 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3306#ifdef CONFIG_PER_VMA_LOCK
3307 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3308#endif
3309 mmap_init();
3310 nsproxy_cache_init();
3311}
3312
3313/*
3314 * Check constraints on flags passed to the unshare system call.
3315 */
3316static int check_unshare_flags(unsigned long unshare_flags)
3317{
3318 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3319 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3320 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3321 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3322 CLONE_NEWTIME))
3323 return -EINVAL;
3324 /*
3325 * Not implemented, but pretend it works if there is nothing
3326 * to unshare. Note that unsharing the address space or the
3327 * signal handlers also need to unshare the signal queues (aka
3328 * CLONE_THREAD).
3329 */
3330 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3331 if (!thread_group_empty(current))
3332 return -EINVAL;
3333 }
3334 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3335 if (refcount_read(&current->sighand->count) > 1)
3336 return -EINVAL;
3337 }
3338 if (unshare_flags & CLONE_VM) {
3339 if (!current_is_single_threaded())
3340 return -EINVAL;
3341 }
3342
3343 return 0;
3344}
3345
3346/*
3347 * Unshare the filesystem structure if it is being shared
3348 */
3349static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3350{
3351 struct fs_struct *fs = current->fs;
3352
3353 if (!(unshare_flags & CLONE_FS) || !fs)
3354 return 0;
3355
3356 /* don't need lock here; in the worst case we'll do useless copy */
3357 if (fs->users == 1)
3358 return 0;
3359
3360 *new_fsp = copy_fs_struct(fs);
3361 if (!*new_fsp)
3362 return -ENOMEM;
3363
3364 return 0;
3365}
3366
3367/*
3368 * Unshare file descriptor table if it is being shared
3369 */
3370int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3371 struct files_struct **new_fdp)
3372{
3373 struct files_struct *fd = current->files;
3374 int error = 0;
3375
3376 if ((unshare_flags & CLONE_FILES) &&
3377 (fd && atomic_read(&fd->count) > 1)) {
3378 *new_fdp = dup_fd(fd, max_fds, &error);
3379 if (!*new_fdp)
3380 return error;
3381 }
3382
3383 return 0;
3384}
3385
3386/*
3387 * unshare allows a process to 'unshare' part of the process
3388 * context which was originally shared using clone. copy_*
3389 * functions used by kernel_clone() cannot be used here directly
3390 * because they modify an inactive task_struct that is being
3391 * constructed. Here we are modifying the current, active,
3392 * task_struct.
3393 */
3394int ksys_unshare(unsigned long unshare_flags)
3395{
3396 struct fs_struct *fs, *new_fs = NULL;
3397 struct files_struct *new_fd = NULL;
3398 struct cred *new_cred = NULL;
3399 struct nsproxy *new_nsproxy = NULL;
3400 int do_sysvsem = 0;
3401 int err;
3402
3403 /*
3404 * If unsharing a user namespace must also unshare the thread group
3405 * and unshare the filesystem root and working directories.
3406 */
3407 if (unshare_flags & CLONE_NEWUSER)
3408 unshare_flags |= CLONE_THREAD | CLONE_FS;
3409 /*
3410 * If unsharing vm, must also unshare signal handlers.
3411 */
3412 if (unshare_flags & CLONE_VM)
3413 unshare_flags |= CLONE_SIGHAND;
3414 /*
3415 * If unsharing a signal handlers, must also unshare the signal queues.
3416 */
3417 if (unshare_flags & CLONE_SIGHAND)
3418 unshare_flags |= CLONE_THREAD;
3419 /*
3420 * If unsharing namespace, must also unshare filesystem information.
3421 */
3422 if (unshare_flags & CLONE_NEWNS)
3423 unshare_flags |= CLONE_FS;
3424
3425 err = check_unshare_flags(unshare_flags);
3426 if (err)
3427 goto bad_unshare_out;
3428 /*
3429 * CLONE_NEWIPC must also detach from the undolist: after switching
3430 * to a new ipc namespace, the semaphore arrays from the old
3431 * namespace are unreachable.
3432 */
3433 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3434 do_sysvsem = 1;
3435 err = unshare_fs(unshare_flags, &new_fs);
3436 if (err)
3437 goto bad_unshare_out;
3438 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3439 if (err)
3440 goto bad_unshare_cleanup_fs;
3441 err = unshare_userns(unshare_flags, &new_cred);
3442 if (err)
3443 goto bad_unshare_cleanup_fd;
3444 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3445 new_cred, new_fs);
3446 if (err)
3447 goto bad_unshare_cleanup_cred;
3448
3449 if (new_cred) {
3450 err = set_cred_ucounts(new_cred);
3451 if (err)
3452 goto bad_unshare_cleanup_cred;
3453 }
3454
3455 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3456 if (do_sysvsem) {
3457 /*
3458 * CLONE_SYSVSEM is equivalent to sys_exit().
3459 */
3460 exit_sem(current);
3461 }
3462 if (unshare_flags & CLONE_NEWIPC) {
3463 /* Orphan segments in old ns (see sem above). */
3464 exit_shm(current);
3465 shm_init_task(current);
3466 }
3467
3468 if (new_nsproxy)
3469 switch_task_namespaces(current, new_nsproxy);
3470
3471 task_lock(current);
3472
3473 if (new_fs) {
3474 fs = current->fs;
3475 spin_lock(&fs->lock);
3476 current->fs = new_fs;
3477 if (--fs->users)
3478 new_fs = NULL;
3479 else
3480 new_fs = fs;
3481 spin_unlock(&fs->lock);
3482 }
3483
3484 if (new_fd)
3485 swap(current->files, new_fd);
3486
3487 task_unlock(current);
3488
3489 if (new_cred) {
3490 /* Install the new user namespace */
3491 commit_creds(new_cred);
3492 new_cred = NULL;
3493 }
3494 }
3495
3496 perf_event_namespaces(current);
3497
3498bad_unshare_cleanup_cred:
3499 if (new_cred)
3500 put_cred(new_cred);
3501bad_unshare_cleanup_fd:
3502 if (new_fd)
3503 put_files_struct(new_fd);
3504
3505bad_unshare_cleanup_fs:
3506 if (new_fs)
3507 free_fs_struct(new_fs);
3508
3509bad_unshare_out:
3510 return err;
3511}
3512
3513SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3514{
3515 return ksys_unshare(unshare_flags);
3516}
3517
3518/*
3519 * Helper to unshare the files of the current task.
3520 * We don't want to expose copy_files internals to
3521 * the exec layer of the kernel.
3522 */
3523
3524int unshare_files(void)
3525{
3526 struct task_struct *task = current;
3527 struct files_struct *old, *copy = NULL;
3528 int error;
3529
3530 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3531 if (error || !copy)
3532 return error;
3533
3534 old = task->files;
3535 task_lock(task);
3536 task->files = copy;
3537 task_unlock(task);
3538 put_files_struct(old);
3539 return 0;
3540}
3541
3542int sysctl_max_threads(struct ctl_table *table, int write,
3543 void *buffer, size_t *lenp, loff_t *ppos)
3544{
3545 struct ctl_table t;
3546 int ret;
3547 int threads = max_threads;
3548 int min = 1;
3549 int max = MAX_THREADS;
3550
3551 t = *table;
3552 t.data = &threads;
3553 t.extra1 = &min;
3554 t.extra2 = &max;
3555
3556 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3557 if (ret || !write)
3558 return ret;
3559
3560 max_threads = threads;
3561
3562 return 0;
3563}