Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/syscall_user_dispatch.h>
57#include <linux/jiffies.h>
58#include <linux/futex.h>
59#include <linux/compat.h>
60#include <linux/kthread.h>
61#include <linux/task_io_accounting_ops.h>
62#include <linux/rcupdate.h>
63#include <linux/ptrace.h>
64#include <linux/mount.h>
65#include <linux/audit.h>
66#include <linux/memcontrol.h>
67#include <linux/ftrace.h>
68#include <linux/proc_fs.h>
69#include <linux/profile.h>
70#include <linux/rmap.h>
71#include <linux/ksm.h>
72#include <linux/acct.h>
73#include <linux/userfaultfd_k.h>
74#include <linux/tsacct_kern.h>
75#include <linux/cn_proc.h>
76#include <linux/freezer.h>
77#include <linux/delayacct.h>
78#include <linux/taskstats_kern.h>
79#include <linux/tty.h>
80#include <linux/fs_struct.h>
81#include <linux/magic.h>
82#include <linux/perf_event.h>
83#include <linux/posix-timers.h>
84#include <linux/user-return-notifier.h>
85#include <linux/oom.h>
86#include <linux/khugepaged.h>
87#include <linux/signalfd.h>
88#include <linux/uprobes.h>
89#include <linux/aio.h>
90#include <linux/compiler.h>
91#include <linux/sysctl.h>
92#include <linux/kcov.h>
93#include <linux/livepatch.h>
94#include <linux/thread_info.h>
95#include <linux/stackleak.h>
96#include <linux/kasan.h>
97#include <linux/scs.h>
98#include <linux/io_uring.h>
99#include <linux/bpf.h>
100#include <linux/stackprotector.h>
101#include <linux/user_events.h>
102#include <linux/iommu.h>
103#include <linux/rseq.h>
104
105#include <asm/pgalloc.h>
106#include <linux/uaccess.h>
107#include <asm/mmu_context.h>
108#include <asm/cacheflush.h>
109#include <asm/tlbflush.h>
110
111#include <trace/events/sched.h>
112
113#define CREATE_TRACE_POINTS
114#include <trace/events/task.h>
115
116/*
117 * Minimum number of threads to boot the kernel
118 */
119#define MIN_THREADS 20
120
121/*
122 * Maximum number of threads
123 */
124#define MAX_THREADS FUTEX_TID_MASK
125
126/*
127 * Protected counters by write_lock_irq(&tasklist_lock)
128 */
129unsigned long total_forks; /* Handle normal Linux uptimes. */
130int nr_threads; /* The idle threads do not count.. */
131
132static int max_threads; /* tunable limit on nr_threads */
133
134#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
135
136static const char * const resident_page_types[] = {
137 NAMED_ARRAY_INDEX(MM_FILEPAGES),
138 NAMED_ARRAY_INDEX(MM_ANONPAGES),
139 NAMED_ARRAY_INDEX(MM_SWAPENTS),
140 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
141};
142
143DEFINE_PER_CPU(unsigned long, process_counts) = 0;
144
145__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
146
147#ifdef CONFIG_PROVE_RCU
148int lockdep_tasklist_lock_is_held(void)
149{
150 return lockdep_is_held(&tasklist_lock);
151}
152EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
153#endif /* #ifdef CONFIG_PROVE_RCU */
154
155int nr_processes(void)
156{
157 int cpu;
158 int total = 0;
159
160 for_each_possible_cpu(cpu)
161 total += per_cpu(process_counts, cpu);
162
163 return total;
164}
165
166void __weak arch_release_task_struct(struct task_struct *tsk)
167{
168}
169
170static struct kmem_cache *task_struct_cachep;
171
172static inline struct task_struct *alloc_task_struct_node(int node)
173{
174 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
175}
176
177static inline void free_task_struct(struct task_struct *tsk)
178{
179 kmem_cache_free(task_struct_cachep, tsk);
180}
181
182/*
183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
184 * kmemcache based allocator.
185 */
186# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
187
188# ifdef CONFIG_VMAP_STACK
189/*
190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
191 * flush. Try to minimize the number of calls by caching stacks.
192 */
193#define NR_CACHED_STACKS 2
194static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
195
196struct vm_stack {
197 struct rcu_head rcu;
198 struct vm_struct *stack_vm_area;
199};
200
201static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
202{
203 unsigned int i;
204
205 for (i = 0; i < NR_CACHED_STACKS; i++) {
206 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
207 continue;
208 return true;
209 }
210 return false;
211}
212
213static void thread_stack_free_rcu(struct rcu_head *rh)
214{
215 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
216
217 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
218 return;
219
220 vfree(vm_stack);
221}
222
223static void thread_stack_delayed_free(struct task_struct *tsk)
224{
225 struct vm_stack *vm_stack = tsk->stack;
226
227 vm_stack->stack_vm_area = tsk->stack_vm_area;
228 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
229}
230
231static int free_vm_stack_cache(unsigned int cpu)
232{
233 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
234 int i;
235
236 for (i = 0; i < NR_CACHED_STACKS; i++) {
237 struct vm_struct *vm_stack = cached_vm_stacks[i];
238
239 if (!vm_stack)
240 continue;
241
242 vfree(vm_stack->addr);
243 cached_vm_stacks[i] = NULL;
244 }
245
246 return 0;
247}
248
249static int memcg_charge_kernel_stack(struct vm_struct *vm)
250{
251 int i;
252 int ret;
253 int nr_charged = 0;
254
255 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
256
257 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
258 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
259 if (ret)
260 goto err;
261 nr_charged++;
262 }
263 return 0;
264err:
265 for (i = 0; i < nr_charged; i++)
266 memcg_kmem_uncharge_page(vm->pages[i], 0);
267 return ret;
268}
269
270static int alloc_thread_stack_node(struct task_struct *tsk, int node)
271{
272 struct vm_struct *vm;
273 void *stack;
274 int i;
275
276 for (i = 0; i < NR_CACHED_STACKS; i++) {
277 struct vm_struct *s;
278
279 s = this_cpu_xchg(cached_stacks[i], NULL);
280
281 if (!s)
282 continue;
283
284 /* Reset stack metadata. */
285 kasan_unpoison_range(s->addr, THREAD_SIZE);
286
287 stack = kasan_reset_tag(s->addr);
288
289 /* Clear stale pointers from reused stack. */
290 memset(stack, 0, THREAD_SIZE);
291
292 if (memcg_charge_kernel_stack(s)) {
293 vfree(s->addr);
294 return -ENOMEM;
295 }
296
297 tsk->stack_vm_area = s;
298 tsk->stack = stack;
299 return 0;
300 }
301
302 /*
303 * Allocated stacks are cached and later reused by new threads,
304 * so memcg accounting is performed manually on assigning/releasing
305 * stacks to tasks. Drop __GFP_ACCOUNT.
306 */
307 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
308 VMALLOC_START, VMALLOC_END,
309 THREADINFO_GFP & ~__GFP_ACCOUNT,
310 PAGE_KERNEL,
311 0, node, __builtin_return_address(0));
312 if (!stack)
313 return -ENOMEM;
314
315 vm = find_vm_area(stack);
316 if (memcg_charge_kernel_stack(vm)) {
317 vfree(stack);
318 return -ENOMEM;
319 }
320 /*
321 * We can't call find_vm_area() in interrupt context, and
322 * free_thread_stack() can be called in interrupt context,
323 * so cache the vm_struct.
324 */
325 tsk->stack_vm_area = vm;
326 stack = kasan_reset_tag(stack);
327 tsk->stack = stack;
328 return 0;
329}
330
331static void free_thread_stack(struct task_struct *tsk)
332{
333 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
334 thread_stack_delayed_free(tsk);
335
336 tsk->stack = NULL;
337 tsk->stack_vm_area = NULL;
338}
339
340# else /* !CONFIG_VMAP_STACK */
341
342static void thread_stack_free_rcu(struct rcu_head *rh)
343{
344 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
345}
346
347static void thread_stack_delayed_free(struct task_struct *tsk)
348{
349 struct rcu_head *rh = tsk->stack;
350
351 call_rcu(rh, thread_stack_free_rcu);
352}
353
354static int alloc_thread_stack_node(struct task_struct *tsk, int node)
355{
356 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
357 THREAD_SIZE_ORDER);
358
359 if (likely(page)) {
360 tsk->stack = kasan_reset_tag(page_address(page));
361 return 0;
362 }
363 return -ENOMEM;
364}
365
366static void free_thread_stack(struct task_struct *tsk)
367{
368 thread_stack_delayed_free(tsk);
369 tsk->stack = NULL;
370}
371
372# endif /* CONFIG_VMAP_STACK */
373# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
374
375static struct kmem_cache *thread_stack_cache;
376
377static void thread_stack_free_rcu(struct rcu_head *rh)
378{
379 kmem_cache_free(thread_stack_cache, rh);
380}
381
382static void thread_stack_delayed_free(struct task_struct *tsk)
383{
384 struct rcu_head *rh = tsk->stack;
385
386 call_rcu(rh, thread_stack_free_rcu);
387}
388
389static int alloc_thread_stack_node(struct task_struct *tsk, int node)
390{
391 unsigned long *stack;
392 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
393 stack = kasan_reset_tag(stack);
394 tsk->stack = stack;
395 return stack ? 0 : -ENOMEM;
396}
397
398static void free_thread_stack(struct task_struct *tsk)
399{
400 thread_stack_delayed_free(tsk);
401 tsk->stack = NULL;
402}
403
404void thread_stack_cache_init(void)
405{
406 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
407 THREAD_SIZE, THREAD_SIZE, 0, 0,
408 THREAD_SIZE, NULL);
409 BUG_ON(thread_stack_cache == NULL);
410}
411
412# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
413
414/* SLAB cache for signal_struct structures (tsk->signal) */
415static struct kmem_cache *signal_cachep;
416
417/* SLAB cache for sighand_struct structures (tsk->sighand) */
418struct kmem_cache *sighand_cachep;
419
420/* SLAB cache for files_struct structures (tsk->files) */
421struct kmem_cache *files_cachep;
422
423/* SLAB cache for fs_struct structures (tsk->fs) */
424struct kmem_cache *fs_cachep;
425
426/* SLAB cache for vm_area_struct structures */
427static struct kmem_cache *vm_area_cachep;
428
429/* SLAB cache for mm_struct structures (tsk->mm) */
430static struct kmem_cache *mm_cachep;
431
432#ifdef CONFIG_PER_VMA_LOCK
433
434/* SLAB cache for vm_area_struct.lock */
435static struct kmem_cache *vma_lock_cachep;
436
437static bool vma_lock_alloc(struct vm_area_struct *vma)
438{
439 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
440 if (!vma->vm_lock)
441 return false;
442
443 init_rwsem(&vma->vm_lock->lock);
444 vma->vm_lock_seq = -1;
445
446 return true;
447}
448
449static inline void vma_lock_free(struct vm_area_struct *vma)
450{
451 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
452}
453
454#else /* CONFIG_PER_VMA_LOCK */
455
456static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
457static inline void vma_lock_free(struct vm_area_struct *vma) {}
458
459#endif /* CONFIG_PER_VMA_LOCK */
460
461struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
462{
463 struct vm_area_struct *vma;
464
465 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
466 if (!vma)
467 return NULL;
468
469 vma_init(vma, mm);
470 if (!vma_lock_alloc(vma)) {
471 kmem_cache_free(vm_area_cachep, vma);
472 return NULL;
473 }
474
475 return vma;
476}
477
478struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
479{
480 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
481
482 if (!new)
483 return NULL;
484
485 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
486 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
487 /*
488 * orig->shared.rb may be modified concurrently, but the clone
489 * will be reinitialized.
490 */
491 data_race(memcpy(new, orig, sizeof(*new)));
492 if (!vma_lock_alloc(new)) {
493 kmem_cache_free(vm_area_cachep, new);
494 return NULL;
495 }
496 INIT_LIST_HEAD(&new->anon_vma_chain);
497 vma_numab_state_init(new);
498 dup_anon_vma_name(orig, new);
499
500 return new;
501}
502
503void __vm_area_free(struct vm_area_struct *vma)
504{
505 vma_numab_state_free(vma);
506 free_anon_vma_name(vma);
507 vma_lock_free(vma);
508 kmem_cache_free(vm_area_cachep, vma);
509}
510
511#ifdef CONFIG_PER_VMA_LOCK
512static void vm_area_free_rcu_cb(struct rcu_head *head)
513{
514 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
515 vm_rcu);
516
517 /* The vma should not be locked while being destroyed. */
518 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
519 __vm_area_free(vma);
520}
521#endif
522
523void vm_area_free(struct vm_area_struct *vma)
524{
525#ifdef CONFIG_PER_VMA_LOCK
526 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
527#else
528 __vm_area_free(vma);
529#endif
530}
531
532static void account_kernel_stack(struct task_struct *tsk, int account)
533{
534 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
535 struct vm_struct *vm = task_stack_vm_area(tsk);
536 int i;
537
538 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
539 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
540 account * (PAGE_SIZE / 1024));
541 } else {
542 void *stack = task_stack_page(tsk);
543
544 /* All stack pages are in the same node. */
545 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
546 account * (THREAD_SIZE / 1024));
547 }
548}
549
550void exit_task_stack_account(struct task_struct *tsk)
551{
552 account_kernel_stack(tsk, -1);
553
554 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
555 struct vm_struct *vm;
556 int i;
557
558 vm = task_stack_vm_area(tsk);
559 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
560 memcg_kmem_uncharge_page(vm->pages[i], 0);
561 }
562}
563
564static void release_task_stack(struct task_struct *tsk)
565{
566 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
567 return; /* Better to leak the stack than to free prematurely */
568
569 free_thread_stack(tsk);
570}
571
572#ifdef CONFIG_THREAD_INFO_IN_TASK
573void put_task_stack(struct task_struct *tsk)
574{
575 if (refcount_dec_and_test(&tsk->stack_refcount))
576 release_task_stack(tsk);
577}
578#endif
579
580void free_task(struct task_struct *tsk)
581{
582#ifdef CONFIG_SECCOMP
583 WARN_ON_ONCE(tsk->seccomp.filter);
584#endif
585 release_user_cpus_ptr(tsk);
586 scs_release(tsk);
587
588#ifndef CONFIG_THREAD_INFO_IN_TASK
589 /*
590 * The task is finally done with both the stack and thread_info,
591 * so free both.
592 */
593 release_task_stack(tsk);
594#else
595 /*
596 * If the task had a separate stack allocation, it should be gone
597 * by now.
598 */
599 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
600#endif
601 rt_mutex_debug_task_free(tsk);
602 ftrace_graph_exit_task(tsk);
603 arch_release_task_struct(tsk);
604 if (tsk->flags & PF_KTHREAD)
605 free_kthread_struct(tsk);
606 bpf_task_storage_free(tsk);
607 free_task_struct(tsk);
608}
609EXPORT_SYMBOL(free_task);
610
611static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
612{
613 struct file *exe_file;
614
615 exe_file = get_mm_exe_file(oldmm);
616 RCU_INIT_POINTER(mm->exe_file, exe_file);
617 /*
618 * We depend on the oldmm having properly denied write access to the
619 * exe_file already.
620 */
621 if (exe_file && deny_write_access(exe_file))
622 pr_warn_once("deny_write_access() failed in %s\n", __func__);
623}
624
625#ifdef CONFIG_MMU
626static __latent_entropy int dup_mmap(struct mm_struct *mm,
627 struct mm_struct *oldmm)
628{
629 struct vm_area_struct *mpnt, *tmp;
630 int retval;
631 unsigned long charge = 0;
632 LIST_HEAD(uf);
633 VMA_ITERATOR(vmi, mm, 0);
634
635 uprobe_start_dup_mmap();
636 if (mmap_write_lock_killable(oldmm)) {
637 retval = -EINTR;
638 goto fail_uprobe_end;
639 }
640 flush_cache_dup_mm(oldmm);
641 uprobe_dup_mmap(oldmm, mm);
642 /*
643 * Not linked in yet - no deadlock potential:
644 */
645 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
646
647 /* No ordering required: file already has been exposed. */
648 dup_mm_exe_file(mm, oldmm);
649
650 mm->total_vm = oldmm->total_vm;
651 mm->data_vm = oldmm->data_vm;
652 mm->exec_vm = oldmm->exec_vm;
653 mm->stack_vm = oldmm->stack_vm;
654
655 retval = ksm_fork(mm, oldmm);
656 if (retval)
657 goto out;
658 khugepaged_fork(mm, oldmm);
659
660 /* Use __mt_dup() to efficiently build an identical maple tree. */
661 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
662 if (unlikely(retval))
663 goto out;
664
665 mt_clear_in_rcu(vmi.mas.tree);
666 for_each_vma(vmi, mpnt) {
667 struct file *file;
668
669 vma_start_write(mpnt);
670 if (mpnt->vm_flags & VM_DONTCOPY) {
671 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
672 mpnt->vm_end, GFP_KERNEL);
673 if (retval)
674 goto loop_out;
675
676 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
677 continue;
678 }
679 charge = 0;
680 /*
681 * Don't duplicate many vmas if we've been oom-killed (for
682 * example)
683 */
684 if (fatal_signal_pending(current)) {
685 retval = -EINTR;
686 goto loop_out;
687 }
688 if (mpnt->vm_flags & VM_ACCOUNT) {
689 unsigned long len = vma_pages(mpnt);
690
691 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
692 goto fail_nomem;
693 charge = len;
694 }
695 tmp = vm_area_dup(mpnt);
696 if (!tmp)
697 goto fail_nomem;
698 retval = vma_dup_policy(mpnt, tmp);
699 if (retval)
700 goto fail_nomem_policy;
701 tmp->vm_mm = mm;
702 retval = dup_userfaultfd(tmp, &uf);
703 if (retval)
704 goto fail_nomem_anon_vma_fork;
705 if (tmp->vm_flags & VM_WIPEONFORK) {
706 /*
707 * VM_WIPEONFORK gets a clean slate in the child.
708 * Don't prepare anon_vma until fault since we don't
709 * copy page for current vma.
710 */
711 tmp->anon_vma = NULL;
712 } else if (anon_vma_fork(tmp, mpnt))
713 goto fail_nomem_anon_vma_fork;
714 vm_flags_clear(tmp, VM_LOCKED_MASK);
715 file = tmp->vm_file;
716 if (file) {
717 struct address_space *mapping = file->f_mapping;
718
719 get_file(file);
720 i_mmap_lock_write(mapping);
721 if (vma_is_shared_maywrite(tmp))
722 mapping_allow_writable(mapping);
723 flush_dcache_mmap_lock(mapping);
724 /* insert tmp into the share list, just after mpnt */
725 vma_interval_tree_insert_after(tmp, mpnt,
726 &mapping->i_mmap);
727 flush_dcache_mmap_unlock(mapping);
728 i_mmap_unlock_write(mapping);
729 }
730
731 /*
732 * Copy/update hugetlb private vma information.
733 */
734 if (is_vm_hugetlb_page(tmp))
735 hugetlb_dup_vma_private(tmp);
736
737 /*
738 * Link the vma into the MT. After using __mt_dup(), memory
739 * allocation is not necessary here, so it cannot fail.
740 */
741 vma_iter_bulk_store(&vmi, tmp);
742
743 mm->map_count++;
744 if (!(tmp->vm_flags & VM_WIPEONFORK))
745 retval = copy_page_range(tmp, mpnt);
746
747 if (tmp->vm_ops && tmp->vm_ops->open)
748 tmp->vm_ops->open(tmp);
749
750 if (retval) {
751 mpnt = vma_next(&vmi);
752 goto loop_out;
753 }
754 }
755 /* a new mm has just been created */
756 retval = arch_dup_mmap(oldmm, mm);
757loop_out:
758 vma_iter_free(&vmi);
759 if (!retval) {
760 mt_set_in_rcu(vmi.mas.tree);
761 } else if (mpnt) {
762 /*
763 * The entire maple tree has already been duplicated. If the
764 * mmap duplication fails, mark the failure point with
765 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
766 * stop releasing VMAs that have not been duplicated after this
767 * point.
768 */
769 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
770 mas_store(&vmi.mas, XA_ZERO_ENTRY);
771 }
772out:
773 mmap_write_unlock(mm);
774 flush_tlb_mm(oldmm);
775 mmap_write_unlock(oldmm);
776 dup_userfaultfd_complete(&uf);
777fail_uprobe_end:
778 uprobe_end_dup_mmap();
779 return retval;
780
781fail_nomem_anon_vma_fork:
782 mpol_put(vma_policy(tmp));
783fail_nomem_policy:
784 vm_area_free(tmp);
785fail_nomem:
786 retval = -ENOMEM;
787 vm_unacct_memory(charge);
788 goto loop_out;
789}
790
791static inline int mm_alloc_pgd(struct mm_struct *mm)
792{
793 mm->pgd = pgd_alloc(mm);
794 if (unlikely(!mm->pgd))
795 return -ENOMEM;
796 return 0;
797}
798
799static inline void mm_free_pgd(struct mm_struct *mm)
800{
801 pgd_free(mm, mm->pgd);
802}
803#else
804static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
805{
806 mmap_write_lock(oldmm);
807 dup_mm_exe_file(mm, oldmm);
808 mmap_write_unlock(oldmm);
809 return 0;
810}
811#define mm_alloc_pgd(mm) (0)
812#define mm_free_pgd(mm)
813#endif /* CONFIG_MMU */
814
815static void check_mm(struct mm_struct *mm)
816{
817 int i;
818
819 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
820 "Please make sure 'struct resident_page_types[]' is updated as well");
821
822 for (i = 0; i < NR_MM_COUNTERS; i++) {
823 long x = percpu_counter_sum(&mm->rss_stat[i]);
824
825 if (unlikely(x))
826 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
827 mm, resident_page_types[i], x);
828 }
829
830 if (mm_pgtables_bytes(mm))
831 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
832 mm_pgtables_bytes(mm));
833
834#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
835 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
836#endif
837}
838
839#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
840#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
841
842static void do_check_lazy_tlb(void *arg)
843{
844 struct mm_struct *mm = arg;
845
846 WARN_ON_ONCE(current->active_mm == mm);
847}
848
849static void do_shoot_lazy_tlb(void *arg)
850{
851 struct mm_struct *mm = arg;
852
853 if (current->active_mm == mm) {
854 WARN_ON_ONCE(current->mm);
855 current->active_mm = &init_mm;
856 switch_mm(mm, &init_mm, current);
857 }
858}
859
860static void cleanup_lazy_tlbs(struct mm_struct *mm)
861{
862 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
863 /*
864 * In this case, lazy tlb mms are refounted and would not reach
865 * __mmdrop until all CPUs have switched away and mmdrop()ed.
866 */
867 return;
868 }
869
870 /*
871 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
872 * requires lazy mm users to switch to another mm when the refcount
873 * drops to zero, before the mm is freed. This requires IPIs here to
874 * switch kernel threads to init_mm.
875 *
876 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
877 * switch with the final userspace teardown TLB flush which leaves the
878 * mm lazy on this CPU but no others, reducing the need for additional
879 * IPIs here. There are cases where a final IPI is still required here,
880 * such as the final mmdrop being performed on a different CPU than the
881 * one exiting, or kernel threads using the mm when userspace exits.
882 *
883 * IPI overheads have not found to be expensive, but they could be
884 * reduced in a number of possible ways, for example (roughly
885 * increasing order of complexity):
886 * - The last lazy reference created by exit_mm() could instead switch
887 * to init_mm, however it's probable this will run on the same CPU
888 * immediately afterwards, so this may not reduce IPIs much.
889 * - A batch of mms requiring IPIs could be gathered and freed at once.
890 * - CPUs store active_mm where it can be remotely checked without a
891 * lock, to filter out false-positives in the cpumask.
892 * - After mm_users or mm_count reaches zero, switching away from the
893 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
894 * with some batching or delaying of the final IPIs.
895 * - A delayed freeing and RCU-like quiescing sequence based on mm
896 * switching to avoid IPIs completely.
897 */
898 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
899 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
900 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
901}
902
903/*
904 * Called when the last reference to the mm
905 * is dropped: either by a lazy thread or by
906 * mmput. Free the page directory and the mm.
907 */
908void __mmdrop(struct mm_struct *mm)
909{
910 BUG_ON(mm == &init_mm);
911 WARN_ON_ONCE(mm == current->mm);
912
913 /* Ensure no CPUs are using this as their lazy tlb mm */
914 cleanup_lazy_tlbs(mm);
915
916 WARN_ON_ONCE(mm == current->active_mm);
917 mm_free_pgd(mm);
918 destroy_context(mm);
919 mmu_notifier_subscriptions_destroy(mm);
920 check_mm(mm);
921 put_user_ns(mm->user_ns);
922 mm_pasid_drop(mm);
923 mm_destroy_cid(mm);
924 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
925
926 free_mm(mm);
927}
928EXPORT_SYMBOL_GPL(__mmdrop);
929
930static void mmdrop_async_fn(struct work_struct *work)
931{
932 struct mm_struct *mm;
933
934 mm = container_of(work, struct mm_struct, async_put_work);
935 __mmdrop(mm);
936}
937
938static void mmdrop_async(struct mm_struct *mm)
939{
940 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
941 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
942 schedule_work(&mm->async_put_work);
943 }
944}
945
946static inline void free_signal_struct(struct signal_struct *sig)
947{
948 taskstats_tgid_free(sig);
949 sched_autogroup_exit(sig);
950 /*
951 * __mmdrop is not safe to call from softirq context on x86 due to
952 * pgd_dtor so postpone it to the async context
953 */
954 if (sig->oom_mm)
955 mmdrop_async(sig->oom_mm);
956 kmem_cache_free(signal_cachep, sig);
957}
958
959static inline void put_signal_struct(struct signal_struct *sig)
960{
961 if (refcount_dec_and_test(&sig->sigcnt))
962 free_signal_struct(sig);
963}
964
965void __put_task_struct(struct task_struct *tsk)
966{
967 WARN_ON(!tsk->exit_state);
968 WARN_ON(refcount_read(&tsk->usage));
969 WARN_ON(tsk == current);
970
971 io_uring_free(tsk);
972 cgroup_free(tsk);
973 task_numa_free(tsk, true);
974 security_task_free(tsk);
975 exit_creds(tsk);
976 delayacct_tsk_free(tsk);
977 put_signal_struct(tsk->signal);
978 sched_core_free(tsk);
979 free_task(tsk);
980}
981EXPORT_SYMBOL_GPL(__put_task_struct);
982
983void __put_task_struct_rcu_cb(struct rcu_head *rhp)
984{
985 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
986
987 __put_task_struct(task);
988}
989EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
990
991void __init __weak arch_task_cache_init(void) { }
992
993/*
994 * set_max_threads
995 */
996static void set_max_threads(unsigned int max_threads_suggested)
997{
998 u64 threads;
999 unsigned long nr_pages = totalram_pages();
1000
1001 /*
1002 * The number of threads shall be limited such that the thread
1003 * structures may only consume a small part of the available memory.
1004 */
1005 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1006 threads = MAX_THREADS;
1007 else
1008 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1009 (u64) THREAD_SIZE * 8UL);
1010
1011 if (threads > max_threads_suggested)
1012 threads = max_threads_suggested;
1013
1014 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1015}
1016
1017#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1018/* Initialized by the architecture: */
1019int arch_task_struct_size __read_mostly;
1020#endif
1021
1022static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1023{
1024 /* Fetch thread_struct whitelist for the architecture. */
1025 arch_thread_struct_whitelist(offset, size);
1026
1027 /*
1028 * Handle zero-sized whitelist or empty thread_struct, otherwise
1029 * adjust offset to position of thread_struct in task_struct.
1030 */
1031 if (unlikely(*size == 0))
1032 *offset = 0;
1033 else
1034 *offset += offsetof(struct task_struct, thread);
1035}
1036
1037void __init fork_init(void)
1038{
1039 int i;
1040#ifndef ARCH_MIN_TASKALIGN
1041#define ARCH_MIN_TASKALIGN 0
1042#endif
1043 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1044 unsigned long useroffset, usersize;
1045
1046 /* create a slab on which task_structs can be allocated */
1047 task_struct_whitelist(&useroffset, &usersize);
1048 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1049 arch_task_struct_size, align,
1050 SLAB_PANIC|SLAB_ACCOUNT,
1051 useroffset, usersize, NULL);
1052
1053 /* do the arch specific task caches init */
1054 arch_task_cache_init();
1055
1056 set_max_threads(MAX_THREADS);
1057
1058 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1059 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1060 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1061 init_task.signal->rlim[RLIMIT_NPROC];
1062
1063 for (i = 0; i < UCOUNT_COUNTS; i++)
1064 init_user_ns.ucount_max[i] = max_threads/2;
1065
1066 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1067 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1068 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1070
1071#ifdef CONFIG_VMAP_STACK
1072 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1073 NULL, free_vm_stack_cache);
1074#endif
1075
1076 scs_init();
1077
1078 lockdep_init_task(&init_task);
1079 uprobes_init();
1080}
1081
1082int __weak arch_dup_task_struct(struct task_struct *dst,
1083 struct task_struct *src)
1084{
1085 *dst = *src;
1086 return 0;
1087}
1088
1089void set_task_stack_end_magic(struct task_struct *tsk)
1090{
1091 unsigned long *stackend;
1092
1093 stackend = end_of_stack(tsk);
1094 *stackend = STACK_END_MAGIC; /* for overflow detection */
1095}
1096
1097static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1098{
1099 struct task_struct *tsk;
1100 int err;
1101
1102 if (node == NUMA_NO_NODE)
1103 node = tsk_fork_get_node(orig);
1104 tsk = alloc_task_struct_node(node);
1105 if (!tsk)
1106 return NULL;
1107
1108 err = arch_dup_task_struct(tsk, orig);
1109 if (err)
1110 goto free_tsk;
1111
1112 err = alloc_thread_stack_node(tsk, node);
1113 if (err)
1114 goto free_tsk;
1115
1116#ifdef CONFIG_THREAD_INFO_IN_TASK
1117 refcount_set(&tsk->stack_refcount, 1);
1118#endif
1119 account_kernel_stack(tsk, 1);
1120
1121 err = scs_prepare(tsk, node);
1122 if (err)
1123 goto free_stack;
1124
1125#ifdef CONFIG_SECCOMP
1126 /*
1127 * We must handle setting up seccomp filters once we're under
1128 * the sighand lock in case orig has changed between now and
1129 * then. Until then, filter must be NULL to avoid messing up
1130 * the usage counts on the error path calling free_task.
1131 */
1132 tsk->seccomp.filter = NULL;
1133#endif
1134
1135 setup_thread_stack(tsk, orig);
1136 clear_user_return_notifier(tsk);
1137 clear_tsk_need_resched(tsk);
1138 set_task_stack_end_magic(tsk);
1139 clear_syscall_work_syscall_user_dispatch(tsk);
1140
1141#ifdef CONFIG_STACKPROTECTOR
1142 tsk->stack_canary = get_random_canary();
1143#endif
1144 if (orig->cpus_ptr == &orig->cpus_mask)
1145 tsk->cpus_ptr = &tsk->cpus_mask;
1146 dup_user_cpus_ptr(tsk, orig, node);
1147
1148 /*
1149 * One for the user space visible state that goes away when reaped.
1150 * One for the scheduler.
1151 */
1152 refcount_set(&tsk->rcu_users, 2);
1153 /* One for the rcu users */
1154 refcount_set(&tsk->usage, 1);
1155#ifdef CONFIG_BLK_DEV_IO_TRACE
1156 tsk->btrace_seq = 0;
1157#endif
1158 tsk->splice_pipe = NULL;
1159 tsk->task_frag.page = NULL;
1160 tsk->wake_q.next = NULL;
1161 tsk->worker_private = NULL;
1162
1163 kcov_task_init(tsk);
1164 kmsan_task_create(tsk);
1165 kmap_local_fork(tsk);
1166
1167#ifdef CONFIG_FAULT_INJECTION
1168 tsk->fail_nth = 0;
1169#endif
1170
1171#ifdef CONFIG_BLK_CGROUP
1172 tsk->throttle_disk = NULL;
1173 tsk->use_memdelay = 0;
1174#endif
1175
1176#ifdef CONFIG_ARCH_HAS_CPU_PASID
1177 tsk->pasid_activated = 0;
1178#endif
1179
1180#ifdef CONFIG_MEMCG
1181 tsk->active_memcg = NULL;
1182#endif
1183
1184#ifdef CONFIG_CPU_SUP_INTEL
1185 tsk->reported_split_lock = 0;
1186#endif
1187
1188#ifdef CONFIG_SCHED_MM_CID
1189 tsk->mm_cid = -1;
1190 tsk->last_mm_cid = -1;
1191 tsk->mm_cid_active = 0;
1192 tsk->migrate_from_cpu = -1;
1193#endif
1194 return tsk;
1195
1196free_stack:
1197 exit_task_stack_account(tsk);
1198 free_thread_stack(tsk);
1199free_tsk:
1200 free_task_struct(tsk);
1201 return NULL;
1202}
1203
1204__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1205
1206static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1207
1208static int __init coredump_filter_setup(char *s)
1209{
1210 default_dump_filter =
1211 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1212 MMF_DUMP_FILTER_MASK;
1213 return 1;
1214}
1215
1216__setup("coredump_filter=", coredump_filter_setup);
1217
1218#include <linux/init_task.h>
1219
1220static void mm_init_aio(struct mm_struct *mm)
1221{
1222#ifdef CONFIG_AIO
1223 spin_lock_init(&mm->ioctx_lock);
1224 mm->ioctx_table = NULL;
1225#endif
1226}
1227
1228static __always_inline void mm_clear_owner(struct mm_struct *mm,
1229 struct task_struct *p)
1230{
1231#ifdef CONFIG_MEMCG
1232 if (mm->owner == p)
1233 WRITE_ONCE(mm->owner, NULL);
1234#endif
1235}
1236
1237static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1238{
1239#ifdef CONFIG_MEMCG
1240 mm->owner = p;
1241#endif
1242}
1243
1244static void mm_init_uprobes_state(struct mm_struct *mm)
1245{
1246#ifdef CONFIG_UPROBES
1247 mm->uprobes_state.xol_area = NULL;
1248#endif
1249}
1250
1251static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1252 struct user_namespace *user_ns)
1253{
1254 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1255 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1256 atomic_set(&mm->mm_users, 1);
1257 atomic_set(&mm->mm_count, 1);
1258 seqcount_init(&mm->write_protect_seq);
1259 mmap_init_lock(mm);
1260 INIT_LIST_HEAD(&mm->mmlist);
1261#ifdef CONFIG_PER_VMA_LOCK
1262 mm->mm_lock_seq = 0;
1263#endif
1264 mm_pgtables_bytes_init(mm);
1265 mm->map_count = 0;
1266 mm->locked_vm = 0;
1267 atomic64_set(&mm->pinned_vm, 0);
1268 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1269 spin_lock_init(&mm->page_table_lock);
1270 spin_lock_init(&mm->arg_lock);
1271 mm_init_cpumask(mm);
1272 mm_init_aio(mm);
1273 mm_init_owner(mm, p);
1274 mm_pasid_init(mm);
1275 RCU_INIT_POINTER(mm->exe_file, NULL);
1276 mmu_notifier_subscriptions_init(mm);
1277 init_tlb_flush_pending(mm);
1278#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1279 mm->pmd_huge_pte = NULL;
1280#endif
1281 mm_init_uprobes_state(mm);
1282 hugetlb_count_init(mm);
1283
1284 if (current->mm) {
1285 mm->flags = mmf_init_flags(current->mm->flags);
1286 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1287 } else {
1288 mm->flags = default_dump_filter;
1289 mm->def_flags = 0;
1290 }
1291
1292 if (mm_alloc_pgd(mm))
1293 goto fail_nopgd;
1294
1295 if (init_new_context(p, mm))
1296 goto fail_nocontext;
1297
1298 if (mm_alloc_cid(mm))
1299 goto fail_cid;
1300
1301 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1302 NR_MM_COUNTERS))
1303 goto fail_pcpu;
1304
1305 mm->user_ns = get_user_ns(user_ns);
1306 lru_gen_init_mm(mm);
1307 return mm;
1308
1309fail_pcpu:
1310 mm_destroy_cid(mm);
1311fail_cid:
1312 destroy_context(mm);
1313fail_nocontext:
1314 mm_free_pgd(mm);
1315fail_nopgd:
1316 free_mm(mm);
1317 return NULL;
1318}
1319
1320/*
1321 * Allocate and initialize an mm_struct.
1322 */
1323struct mm_struct *mm_alloc(void)
1324{
1325 struct mm_struct *mm;
1326
1327 mm = allocate_mm();
1328 if (!mm)
1329 return NULL;
1330
1331 memset(mm, 0, sizeof(*mm));
1332 return mm_init(mm, current, current_user_ns());
1333}
1334
1335static inline void __mmput(struct mm_struct *mm)
1336{
1337 VM_BUG_ON(atomic_read(&mm->mm_users));
1338
1339 uprobe_clear_state(mm);
1340 exit_aio(mm);
1341 ksm_exit(mm);
1342 khugepaged_exit(mm); /* must run before exit_mmap */
1343 exit_mmap(mm);
1344 mm_put_huge_zero_page(mm);
1345 set_mm_exe_file(mm, NULL);
1346 if (!list_empty(&mm->mmlist)) {
1347 spin_lock(&mmlist_lock);
1348 list_del(&mm->mmlist);
1349 spin_unlock(&mmlist_lock);
1350 }
1351 if (mm->binfmt)
1352 module_put(mm->binfmt->module);
1353 lru_gen_del_mm(mm);
1354 mmdrop(mm);
1355}
1356
1357/*
1358 * Decrement the use count and release all resources for an mm.
1359 */
1360void mmput(struct mm_struct *mm)
1361{
1362 might_sleep();
1363
1364 if (atomic_dec_and_test(&mm->mm_users))
1365 __mmput(mm);
1366}
1367EXPORT_SYMBOL_GPL(mmput);
1368
1369#ifdef CONFIG_MMU
1370static void mmput_async_fn(struct work_struct *work)
1371{
1372 struct mm_struct *mm = container_of(work, struct mm_struct,
1373 async_put_work);
1374
1375 __mmput(mm);
1376}
1377
1378void mmput_async(struct mm_struct *mm)
1379{
1380 if (atomic_dec_and_test(&mm->mm_users)) {
1381 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1382 schedule_work(&mm->async_put_work);
1383 }
1384}
1385EXPORT_SYMBOL_GPL(mmput_async);
1386#endif
1387
1388/**
1389 * set_mm_exe_file - change a reference to the mm's executable file
1390 * @mm: The mm to change.
1391 * @new_exe_file: The new file to use.
1392 *
1393 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1394 *
1395 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1396 * invocations: in mmput() nobody alive left, in execve it happens before
1397 * the new mm is made visible to anyone.
1398 *
1399 * Can only fail if new_exe_file != NULL.
1400 */
1401int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1402{
1403 struct file *old_exe_file;
1404
1405 /*
1406 * It is safe to dereference the exe_file without RCU as
1407 * this function is only called if nobody else can access
1408 * this mm -- see comment above for justification.
1409 */
1410 old_exe_file = rcu_dereference_raw(mm->exe_file);
1411
1412 if (new_exe_file) {
1413 /*
1414 * We expect the caller (i.e., sys_execve) to already denied
1415 * write access, so this is unlikely to fail.
1416 */
1417 if (unlikely(deny_write_access(new_exe_file)))
1418 return -EACCES;
1419 get_file(new_exe_file);
1420 }
1421 rcu_assign_pointer(mm->exe_file, new_exe_file);
1422 if (old_exe_file) {
1423 allow_write_access(old_exe_file);
1424 fput(old_exe_file);
1425 }
1426 return 0;
1427}
1428
1429/**
1430 * replace_mm_exe_file - replace a reference to the mm's executable file
1431 * @mm: The mm to change.
1432 * @new_exe_file: The new file to use.
1433 *
1434 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1435 *
1436 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1437 */
1438int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1439{
1440 struct vm_area_struct *vma;
1441 struct file *old_exe_file;
1442 int ret = 0;
1443
1444 /* Forbid mm->exe_file change if old file still mapped. */
1445 old_exe_file = get_mm_exe_file(mm);
1446 if (old_exe_file) {
1447 VMA_ITERATOR(vmi, mm, 0);
1448 mmap_read_lock(mm);
1449 for_each_vma(vmi, vma) {
1450 if (!vma->vm_file)
1451 continue;
1452 if (path_equal(&vma->vm_file->f_path,
1453 &old_exe_file->f_path)) {
1454 ret = -EBUSY;
1455 break;
1456 }
1457 }
1458 mmap_read_unlock(mm);
1459 fput(old_exe_file);
1460 if (ret)
1461 return ret;
1462 }
1463
1464 ret = deny_write_access(new_exe_file);
1465 if (ret)
1466 return -EACCES;
1467 get_file(new_exe_file);
1468
1469 /* set the new file */
1470 mmap_write_lock(mm);
1471 old_exe_file = rcu_dereference_raw(mm->exe_file);
1472 rcu_assign_pointer(mm->exe_file, new_exe_file);
1473 mmap_write_unlock(mm);
1474
1475 if (old_exe_file) {
1476 allow_write_access(old_exe_file);
1477 fput(old_exe_file);
1478 }
1479 return 0;
1480}
1481
1482/**
1483 * get_mm_exe_file - acquire a reference to the mm's executable file
1484 * @mm: The mm of interest.
1485 *
1486 * Returns %NULL if mm has no associated executable file.
1487 * User must release file via fput().
1488 */
1489struct file *get_mm_exe_file(struct mm_struct *mm)
1490{
1491 struct file *exe_file;
1492
1493 rcu_read_lock();
1494 exe_file = get_file_rcu(&mm->exe_file);
1495 rcu_read_unlock();
1496 return exe_file;
1497}
1498
1499/**
1500 * get_task_exe_file - acquire a reference to the task's executable file
1501 * @task: The task.
1502 *
1503 * Returns %NULL if task's mm (if any) has no associated executable file or
1504 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1505 * User must release file via fput().
1506 */
1507struct file *get_task_exe_file(struct task_struct *task)
1508{
1509 struct file *exe_file = NULL;
1510 struct mm_struct *mm;
1511
1512 task_lock(task);
1513 mm = task->mm;
1514 if (mm) {
1515 if (!(task->flags & PF_KTHREAD))
1516 exe_file = get_mm_exe_file(mm);
1517 }
1518 task_unlock(task);
1519 return exe_file;
1520}
1521
1522/**
1523 * get_task_mm - acquire a reference to the task's mm
1524 * @task: The task.
1525 *
1526 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1527 * this kernel workthread has transiently adopted a user mm with use_mm,
1528 * to do its AIO) is not set and if so returns a reference to it, after
1529 * bumping up the use count. User must release the mm via mmput()
1530 * after use. Typically used by /proc and ptrace.
1531 */
1532struct mm_struct *get_task_mm(struct task_struct *task)
1533{
1534 struct mm_struct *mm;
1535
1536 task_lock(task);
1537 mm = task->mm;
1538 if (mm) {
1539 if (task->flags & PF_KTHREAD)
1540 mm = NULL;
1541 else
1542 mmget(mm);
1543 }
1544 task_unlock(task);
1545 return mm;
1546}
1547EXPORT_SYMBOL_GPL(get_task_mm);
1548
1549struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1550{
1551 struct mm_struct *mm;
1552 int err;
1553
1554 err = down_read_killable(&task->signal->exec_update_lock);
1555 if (err)
1556 return ERR_PTR(err);
1557
1558 mm = get_task_mm(task);
1559 if (mm && mm != current->mm &&
1560 !ptrace_may_access(task, mode)) {
1561 mmput(mm);
1562 mm = ERR_PTR(-EACCES);
1563 }
1564 up_read(&task->signal->exec_update_lock);
1565
1566 return mm;
1567}
1568
1569static void complete_vfork_done(struct task_struct *tsk)
1570{
1571 struct completion *vfork;
1572
1573 task_lock(tsk);
1574 vfork = tsk->vfork_done;
1575 if (likely(vfork)) {
1576 tsk->vfork_done = NULL;
1577 complete(vfork);
1578 }
1579 task_unlock(tsk);
1580}
1581
1582static int wait_for_vfork_done(struct task_struct *child,
1583 struct completion *vfork)
1584{
1585 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1586 int killed;
1587
1588 cgroup_enter_frozen();
1589 killed = wait_for_completion_state(vfork, state);
1590 cgroup_leave_frozen(false);
1591
1592 if (killed) {
1593 task_lock(child);
1594 child->vfork_done = NULL;
1595 task_unlock(child);
1596 }
1597
1598 put_task_struct(child);
1599 return killed;
1600}
1601
1602/* Please note the differences between mmput and mm_release.
1603 * mmput is called whenever we stop holding onto a mm_struct,
1604 * error success whatever.
1605 *
1606 * mm_release is called after a mm_struct has been removed
1607 * from the current process.
1608 *
1609 * This difference is important for error handling, when we
1610 * only half set up a mm_struct for a new process and need to restore
1611 * the old one. Because we mmput the new mm_struct before
1612 * restoring the old one. . .
1613 * Eric Biederman 10 January 1998
1614 */
1615static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1616{
1617 uprobe_free_utask(tsk);
1618
1619 /* Get rid of any cached register state */
1620 deactivate_mm(tsk, mm);
1621
1622 /*
1623 * Signal userspace if we're not exiting with a core dump
1624 * because we want to leave the value intact for debugging
1625 * purposes.
1626 */
1627 if (tsk->clear_child_tid) {
1628 if (atomic_read(&mm->mm_users) > 1) {
1629 /*
1630 * We don't check the error code - if userspace has
1631 * not set up a proper pointer then tough luck.
1632 */
1633 put_user(0, tsk->clear_child_tid);
1634 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1635 1, NULL, NULL, 0, 0);
1636 }
1637 tsk->clear_child_tid = NULL;
1638 }
1639
1640 /*
1641 * All done, finally we can wake up parent and return this mm to him.
1642 * Also kthread_stop() uses this completion for synchronization.
1643 */
1644 if (tsk->vfork_done)
1645 complete_vfork_done(tsk);
1646}
1647
1648void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1649{
1650 futex_exit_release(tsk);
1651 mm_release(tsk, mm);
1652}
1653
1654void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1655{
1656 futex_exec_release(tsk);
1657 mm_release(tsk, mm);
1658}
1659
1660/**
1661 * dup_mm() - duplicates an existing mm structure
1662 * @tsk: the task_struct with which the new mm will be associated.
1663 * @oldmm: the mm to duplicate.
1664 *
1665 * Allocates a new mm structure and duplicates the provided @oldmm structure
1666 * content into it.
1667 *
1668 * Return: the duplicated mm or NULL on failure.
1669 */
1670static struct mm_struct *dup_mm(struct task_struct *tsk,
1671 struct mm_struct *oldmm)
1672{
1673 struct mm_struct *mm;
1674 int err;
1675
1676 mm = allocate_mm();
1677 if (!mm)
1678 goto fail_nomem;
1679
1680 memcpy(mm, oldmm, sizeof(*mm));
1681
1682 if (!mm_init(mm, tsk, mm->user_ns))
1683 goto fail_nomem;
1684
1685 err = dup_mmap(mm, oldmm);
1686 if (err)
1687 goto free_pt;
1688
1689 mm->hiwater_rss = get_mm_rss(mm);
1690 mm->hiwater_vm = mm->total_vm;
1691
1692 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1693 goto free_pt;
1694
1695 return mm;
1696
1697free_pt:
1698 /* don't put binfmt in mmput, we haven't got module yet */
1699 mm->binfmt = NULL;
1700 mm_init_owner(mm, NULL);
1701 mmput(mm);
1702
1703fail_nomem:
1704 return NULL;
1705}
1706
1707static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1708{
1709 struct mm_struct *mm, *oldmm;
1710
1711 tsk->min_flt = tsk->maj_flt = 0;
1712 tsk->nvcsw = tsk->nivcsw = 0;
1713#ifdef CONFIG_DETECT_HUNG_TASK
1714 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1715 tsk->last_switch_time = 0;
1716#endif
1717
1718 tsk->mm = NULL;
1719 tsk->active_mm = NULL;
1720
1721 /*
1722 * Are we cloning a kernel thread?
1723 *
1724 * We need to steal a active VM for that..
1725 */
1726 oldmm = current->mm;
1727 if (!oldmm)
1728 return 0;
1729
1730 if (clone_flags & CLONE_VM) {
1731 mmget(oldmm);
1732 mm = oldmm;
1733 } else {
1734 mm = dup_mm(tsk, current->mm);
1735 if (!mm)
1736 return -ENOMEM;
1737 }
1738
1739 tsk->mm = mm;
1740 tsk->active_mm = mm;
1741 sched_mm_cid_fork(tsk);
1742 return 0;
1743}
1744
1745static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1746{
1747 struct fs_struct *fs = current->fs;
1748 if (clone_flags & CLONE_FS) {
1749 /* tsk->fs is already what we want */
1750 spin_lock(&fs->lock);
1751 /* "users" and "in_exec" locked for check_unsafe_exec() */
1752 if (fs->in_exec) {
1753 spin_unlock(&fs->lock);
1754 return -EAGAIN;
1755 }
1756 fs->users++;
1757 spin_unlock(&fs->lock);
1758 return 0;
1759 }
1760 tsk->fs = copy_fs_struct(fs);
1761 if (!tsk->fs)
1762 return -ENOMEM;
1763 return 0;
1764}
1765
1766static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1767 int no_files)
1768{
1769 struct files_struct *oldf, *newf;
1770 int error = 0;
1771
1772 /*
1773 * A background process may not have any files ...
1774 */
1775 oldf = current->files;
1776 if (!oldf)
1777 goto out;
1778
1779 if (no_files) {
1780 tsk->files = NULL;
1781 goto out;
1782 }
1783
1784 if (clone_flags & CLONE_FILES) {
1785 atomic_inc(&oldf->count);
1786 goto out;
1787 }
1788
1789 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1790 if (!newf)
1791 goto out;
1792
1793 tsk->files = newf;
1794 error = 0;
1795out:
1796 return error;
1797}
1798
1799static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1800{
1801 struct sighand_struct *sig;
1802
1803 if (clone_flags & CLONE_SIGHAND) {
1804 refcount_inc(¤t->sighand->count);
1805 return 0;
1806 }
1807 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1808 RCU_INIT_POINTER(tsk->sighand, sig);
1809 if (!sig)
1810 return -ENOMEM;
1811
1812 refcount_set(&sig->count, 1);
1813 spin_lock_irq(¤t->sighand->siglock);
1814 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1815 spin_unlock_irq(¤t->sighand->siglock);
1816
1817 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1818 if (clone_flags & CLONE_CLEAR_SIGHAND)
1819 flush_signal_handlers(tsk, 0);
1820
1821 return 0;
1822}
1823
1824void __cleanup_sighand(struct sighand_struct *sighand)
1825{
1826 if (refcount_dec_and_test(&sighand->count)) {
1827 signalfd_cleanup(sighand);
1828 /*
1829 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1830 * without an RCU grace period, see __lock_task_sighand().
1831 */
1832 kmem_cache_free(sighand_cachep, sighand);
1833 }
1834}
1835
1836/*
1837 * Initialize POSIX timer handling for a thread group.
1838 */
1839static void posix_cpu_timers_init_group(struct signal_struct *sig)
1840{
1841 struct posix_cputimers *pct = &sig->posix_cputimers;
1842 unsigned long cpu_limit;
1843
1844 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1845 posix_cputimers_group_init(pct, cpu_limit);
1846}
1847
1848static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1849{
1850 struct signal_struct *sig;
1851
1852 if (clone_flags & CLONE_THREAD)
1853 return 0;
1854
1855 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1856 tsk->signal = sig;
1857 if (!sig)
1858 return -ENOMEM;
1859
1860 sig->nr_threads = 1;
1861 sig->quick_threads = 1;
1862 atomic_set(&sig->live, 1);
1863 refcount_set(&sig->sigcnt, 1);
1864
1865 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1866 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1867 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1868
1869 init_waitqueue_head(&sig->wait_chldexit);
1870 sig->curr_target = tsk;
1871 init_sigpending(&sig->shared_pending);
1872 INIT_HLIST_HEAD(&sig->multiprocess);
1873 seqlock_init(&sig->stats_lock);
1874 prev_cputime_init(&sig->prev_cputime);
1875
1876#ifdef CONFIG_POSIX_TIMERS
1877 INIT_LIST_HEAD(&sig->posix_timers);
1878 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1879 sig->real_timer.function = it_real_fn;
1880#endif
1881
1882 task_lock(current->group_leader);
1883 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1884 task_unlock(current->group_leader);
1885
1886 posix_cpu_timers_init_group(sig);
1887
1888 tty_audit_fork(sig);
1889 sched_autogroup_fork(sig);
1890
1891 sig->oom_score_adj = current->signal->oom_score_adj;
1892 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1893
1894 mutex_init(&sig->cred_guard_mutex);
1895 init_rwsem(&sig->exec_update_lock);
1896
1897 return 0;
1898}
1899
1900static void copy_seccomp(struct task_struct *p)
1901{
1902#ifdef CONFIG_SECCOMP
1903 /*
1904 * Must be called with sighand->lock held, which is common to
1905 * all threads in the group. Holding cred_guard_mutex is not
1906 * needed because this new task is not yet running and cannot
1907 * be racing exec.
1908 */
1909 assert_spin_locked(¤t->sighand->siglock);
1910
1911 /* Ref-count the new filter user, and assign it. */
1912 get_seccomp_filter(current);
1913 p->seccomp = current->seccomp;
1914
1915 /*
1916 * Explicitly enable no_new_privs here in case it got set
1917 * between the task_struct being duplicated and holding the
1918 * sighand lock. The seccomp state and nnp must be in sync.
1919 */
1920 if (task_no_new_privs(current))
1921 task_set_no_new_privs(p);
1922
1923 /*
1924 * If the parent gained a seccomp mode after copying thread
1925 * flags and between before we held the sighand lock, we have
1926 * to manually enable the seccomp thread flag here.
1927 */
1928 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1929 set_task_syscall_work(p, SECCOMP);
1930#endif
1931}
1932
1933SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1934{
1935 current->clear_child_tid = tidptr;
1936
1937 return task_pid_vnr(current);
1938}
1939
1940static void rt_mutex_init_task(struct task_struct *p)
1941{
1942 raw_spin_lock_init(&p->pi_lock);
1943#ifdef CONFIG_RT_MUTEXES
1944 p->pi_waiters = RB_ROOT_CACHED;
1945 p->pi_top_task = NULL;
1946 p->pi_blocked_on = NULL;
1947#endif
1948}
1949
1950static inline void init_task_pid_links(struct task_struct *task)
1951{
1952 enum pid_type type;
1953
1954 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1955 INIT_HLIST_NODE(&task->pid_links[type]);
1956}
1957
1958static inline void
1959init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1960{
1961 if (type == PIDTYPE_PID)
1962 task->thread_pid = pid;
1963 else
1964 task->signal->pids[type] = pid;
1965}
1966
1967static inline void rcu_copy_process(struct task_struct *p)
1968{
1969#ifdef CONFIG_PREEMPT_RCU
1970 p->rcu_read_lock_nesting = 0;
1971 p->rcu_read_unlock_special.s = 0;
1972 p->rcu_blocked_node = NULL;
1973 INIT_LIST_HEAD(&p->rcu_node_entry);
1974#endif /* #ifdef CONFIG_PREEMPT_RCU */
1975#ifdef CONFIG_TASKS_RCU
1976 p->rcu_tasks_holdout = false;
1977 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1978 p->rcu_tasks_idle_cpu = -1;
1979#endif /* #ifdef CONFIG_TASKS_RCU */
1980#ifdef CONFIG_TASKS_TRACE_RCU
1981 p->trc_reader_nesting = 0;
1982 p->trc_reader_special.s = 0;
1983 INIT_LIST_HEAD(&p->trc_holdout_list);
1984 INIT_LIST_HEAD(&p->trc_blkd_node);
1985#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1986}
1987
1988struct pid *pidfd_pid(const struct file *file)
1989{
1990 if (file->f_op == &pidfd_fops)
1991 return file->private_data;
1992
1993 return ERR_PTR(-EBADF);
1994}
1995
1996static int pidfd_release(struct inode *inode, struct file *file)
1997{
1998 struct pid *pid = file->private_data;
1999
2000 file->private_data = NULL;
2001 put_pid(pid);
2002 return 0;
2003}
2004
2005#ifdef CONFIG_PROC_FS
2006/**
2007 * pidfd_show_fdinfo - print information about a pidfd
2008 * @m: proc fdinfo file
2009 * @f: file referencing a pidfd
2010 *
2011 * Pid:
2012 * This function will print the pid that a given pidfd refers to in the
2013 * pid namespace of the procfs instance.
2014 * If the pid namespace of the process is not a descendant of the pid
2015 * namespace of the procfs instance 0 will be shown as its pid. This is
2016 * similar to calling getppid() on a process whose parent is outside of
2017 * its pid namespace.
2018 *
2019 * NSpid:
2020 * If pid namespaces are supported then this function will also print
2021 * the pid of a given pidfd refers to for all descendant pid namespaces
2022 * starting from the current pid namespace of the instance, i.e. the
2023 * Pid field and the first entry in the NSpid field will be identical.
2024 * If the pid namespace of the process is not a descendant of the pid
2025 * namespace of the procfs instance 0 will be shown as its first NSpid
2026 * entry and no others will be shown.
2027 * Note that this differs from the Pid and NSpid fields in
2028 * /proc/<pid>/status where Pid and NSpid are always shown relative to
2029 * the pid namespace of the procfs instance. The difference becomes
2030 * obvious when sending around a pidfd between pid namespaces from a
2031 * different branch of the tree, i.e. where no ancestral relation is
2032 * present between the pid namespaces:
2033 * - create two new pid namespaces ns1 and ns2 in the initial pid
2034 * namespace (also take care to create new mount namespaces in the
2035 * new pid namespace and mount procfs)
2036 * - create a process with a pidfd in ns1
2037 * - send pidfd from ns1 to ns2
2038 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
2039 * have exactly one entry, which is 0
2040 */
2041static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
2042{
2043 struct pid *pid = f->private_data;
2044 struct pid_namespace *ns;
2045 pid_t nr = -1;
2046
2047 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
2048 ns = proc_pid_ns(file_inode(m->file)->i_sb);
2049 nr = pid_nr_ns(pid, ns);
2050 }
2051
2052 seq_put_decimal_ll(m, "Pid:\t", nr);
2053
2054#ifdef CONFIG_PID_NS
2055 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
2056 if (nr > 0) {
2057 int i;
2058
2059 /* If nr is non-zero it means that 'pid' is valid and that
2060 * ns, i.e. the pid namespace associated with the procfs
2061 * instance, is in the pid namespace hierarchy of pid.
2062 * Start at one below the already printed level.
2063 */
2064 for (i = ns->level + 1; i <= pid->level; i++)
2065 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
2066 }
2067#endif
2068 seq_putc(m, '\n');
2069}
2070#endif
2071
2072/*
2073 * Poll support for process exit notification.
2074 */
2075static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
2076{
2077 struct pid *pid = file->private_data;
2078 __poll_t poll_flags = 0;
2079
2080 poll_wait(file, &pid->wait_pidfd, pts);
2081
2082 /*
2083 * Inform pollers only when the whole thread group exits.
2084 * If the thread group leader exits before all other threads in the
2085 * group, then poll(2) should block, similar to the wait(2) family.
2086 */
2087 if (thread_group_exited(pid))
2088 poll_flags = EPOLLIN | EPOLLRDNORM;
2089
2090 return poll_flags;
2091}
2092
2093const struct file_operations pidfd_fops = {
2094 .release = pidfd_release,
2095 .poll = pidfd_poll,
2096#ifdef CONFIG_PROC_FS
2097 .show_fdinfo = pidfd_show_fdinfo,
2098#endif
2099};
2100
2101/**
2102 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2103 * @pid: the struct pid for which to create a pidfd
2104 * @flags: flags of the new @pidfd
2105 * @ret: Where to return the file for the pidfd.
2106 *
2107 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2108 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2109 *
2110 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2111 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2112 * pidfd file are prepared.
2113 *
2114 * If this function returns successfully the caller is responsible to either
2115 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2116 * order to install the pidfd into its file descriptor table or they must use
2117 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2118 * respectively.
2119 *
2120 * This function is useful when a pidfd must already be reserved but there
2121 * might still be points of failure afterwards and the caller wants to ensure
2122 * that no pidfd is leaked into its file descriptor table.
2123 *
2124 * Return: On success, a reserved pidfd is returned from the function and a new
2125 * pidfd file is returned in the last argument to the function. On
2126 * error, a negative error code is returned from the function and the
2127 * last argument remains unchanged.
2128 */
2129static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2130{
2131 int pidfd;
2132 struct file *pidfd_file;
2133
2134 if (flags & ~(O_NONBLOCK | O_RDWR | O_CLOEXEC))
2135 return -EINVAL;
2136
2137 pidfd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2138 if (pidfd < 0)
2139 return pidfd;
2140
2141 pidfd_file = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2142 flags | O_RDWR | O_CLOEXEC);
2143 if (IS_ERR(pidfd_file)) {
2144 put_unused_fd(pidfd);
2145 return PTR_ERR(pidfd_file);
2146 }
2147 get_pid(pid); /* held by pidfd_file now */
2148 *ret = pidfd_file;
2149 return pidfd;
2150}
2151
2152/**
2153 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2154 * @pid: the struct pid for which to create a pidfd
2155 * @flags: flags of the new @pidfd
2156 * @ret: Where to return the pidfd.
2157 *
2158 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2159 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2160 *
2161 * The helper verifies that @pid is used as a thread group leader.
2162 *
2163 * If this function returns successfully the caller is responsible to either
2164 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2165 * order to install the pidfd into its file descriptor table or they must use
2166 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2167 * respectively.
2168 *
2169 * This function is useful when a pidfd must already be reserved but there
2170 * might still be points of failure afterwards and the caller wants to ensure
2171 * that no pidfd is leaked into its file descriptor table.
2172 *
2173 * Return: On success, a reserved pidfd is returned from the function and a new
2174 * pidfd file is returned in the last argument to the function. On
2175 * error, a negative error code is returned from the function and the
2176 * last argument remains unchanged.
2177 */
2178int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2179{
2180 if (!pid || !pid_has_task(pid, PIDTYPE_TGID))
2181 return -EINVAL;
2182
2183 return __pidfd_prepare(pid, flags, ret);
2184}
2185
2186static void __delayed_free_task(struct rcu_head *rhp)
2187{
2188 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2189
2190 free_task(tsk);
2191}
2192
2193static __always_inline void delayed_free_task(struct task_struct *tsk)
2194{
2195 if (IS_ENABLED(CONFIG_MEMCG))
2196 call_rcu(&tsk->rcu, __delayed_free_task);
2197 else
2198 free_task(tsk);
2199}
2200
2201static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2202{
2203 /* Skip if kernel thread */
2204 if (!tsk->mm)
2205 return;
2206
2207 /* Skip if spawning a thread or using vfork */
2208 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2209 return;
2210
2211 /* We need to synchronize with __set_oom_adj */
2212 mutex_lock(&oom_adj_mutex);
2213 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2214 /* Update the values in case they were changed after copy_signal */
2215 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2216 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2217 mutex_unlock(&oom_adj_mutex);
2218}
2219
2220#ifdef CONFIG_RV
2221static void rv_task_fork(struct task_struct *p)
2222{
2223 int i;
2224
2225 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2226 p->rv[i].da_mon.monitoring = false;
2227}
2228#else
2229#define rv_task_fork(p) do {} while (0)
2230#endif
2231
2232/*
2233 * This creates a new process as a copy of the old one,
2234 * but does not actually start it yet.
2235 *
2236 * It copies the registers, and all the appropriate
2237 * parts of the process environment (as per the clone
2238 * flags). The actual kick-off is left to the caller.
2239 */
2240__latent_entropy struct task_struct *copy_process(
2241 struct pid *pid,
2242 int trace,
2243 int node,
2244 struct kernel_clone_args *args)
2245{
2246 int pidfd = -1, retval;
2247 struct task_struct *p;
2248 struct multiprocess_signals delayed;
2249 struct file *pidfile = NULL;
2250 const u64 clone_flags = args->flags;
2251 struct nsproxy *nsp = current->nsproxy;
2252
2253 /*
2254 * Don't allow sharing the root directory with processes in a different
2255 * namespace
2256 */
2257 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2258 return ERR_PTR(-EINVAL);
2259
2260 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2261 return ERR_PTR(-EINVAL);
2262
2263 /*
2264 * Thread groups must share signals as well, and detached threads
2265 * can only be started up within the thread group.
2266 */
2267 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2268 return ERR_PTR(-EINVAL);
2269
2270 /*
2271 * Shared signal handlers imply shared VM. By way of the above,
2272 * thread groups also imply shared VM. Blocking this case allows
2273 * for various simplifications in other code.
2274 */
2275 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2276 return ERR_PTR(-EINVAL);
2277
2278 /*
2279 * Siblings of global init remain as zombies on exit since they are
2280 * not reaped by their parent (swapper). To solve this and to avoid
2281 * multi-rooted process trees, prevent global and container-inits
2282 * from creating siblings.
2283 */
2284 if ((clone_flags & CLONE_PARENT) &&
2285 current->signal->flags & SIGNAL_UNKILLABLE)
2286 return ERR_PTR(-EINVAL);
2287
2288 /*
2289 * If the new process will be in a different pid or user namespace
2290 * do not allow it to share a thread group with the forking task.
2291 */
2292 if (clone_flags & CLONE_THREAD) {
2293 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2294 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2295 return ERR_PTR(-EINVAL);
2296 }
2297
2298 if (clone_flags & CLONE_PIDFD) {
2299 /*
2300 * - CLONE_DETACHED is blocked so that we can potentially
2301 * reuse it later for CLONE_PIDFD.
2302 * - CLONE_THREAD is blocked until someone really needs it.
2303 */
2304 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2305 return ERR_PTR(-EINVAL);
2306 }
2307
2308 /*
2309 * Force any signals received before this point to be delivered
2310 * before the fork happens. Collect up signals sent to multiple
2311 * processes that happen during the fork and delay them so that
2312 * they appear to happen after the fork.
2313 */
2314 sigemptyset(&delayed.signal);
2315 INIT_HLIST_NODE(&delayed.node);
2316
2317 spin_lock_irq(¤t->sighand->siglock);
2318 if (!(clone_flags & CLONE_THREAD))
2319 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2320 recalc_sigpending();
2321 spin_unlock_irq(¤t->sighand->siglock);
2322 retval = -ERESTARTNOINTR;
2323 if (task_sigpending(current))
2324 goto fork_out;
2325
2326 retval = -ENOMEM;
2327 p = dup_task_struct(current, node);
2328 if (!p)
2329 goto fork_out;
2330 p->flags &= ~PF_KTHREAD;
2331 if (args->kthread)
2332 p->flags |= PF_KTHREAD;
2333 if (args->user_worker) {
2334 /*
2335 * Mark us a user worker, and block any signal that isn't
2336 * fatal or STOP
2337 */
2338 p->flags |= PF_USER_WORKER;
2339 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2340 }
2341 if (args->io_thread)
2342 p->flags |= PF_IO_WORKER;
2343
2344 if (args->name)
2345 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2346
2347 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2348 /*
2349 * Clear TID on mm_release()?
2350 */
2351 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2352
2353 ftrace_graph_init_task(p);
2354
2355 rt_mutex_init_task(p);
2356
2357 lockdep_assert_irqs_enabled();
2358#ifdef CONFIG_PROVE_LOCKING
2359 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2360#endif
2361 retval = copy_creds(p, clone_flags);
2362 if (retval < 0)
2363 goto bad_fork_free;
2364
2365 retval = -EAGAIN;
2366 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2367 if (p->real_cred->user != INIT_USER &&
2368 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2369 goto bad_fork_cleanup_count;
2370 }
2371 current->flags &= ~PF_NPROC_EXCEEDED;
2372
2373 /*
2374 * If multiple threads are within copy_process(), then this check
2375 * triggers too late. This doesn't hurt, the check is only there
2376 * to stop root fork bombs.
2377 */
2378 retval = -EAGAIN;
2379 if (data_race(nr_threads >= max_threads))
2380 goto bad_fork_cleanup_count;
2381
2382 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2383 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2384 p->flags |= PF_FORKNOEXEC;
2385 INIT_LIST_HEAD(&p->children);
2386 INIT_LIST_HEAD(&p->sibling);
2387 rcu_copy_process(p);
2388 p->vfork_done = NULL;
2389 spin_lock_init(&p->alloc_lock);
2390
2391 init_sigpending(&p->pending);
2392
2393 p->utime = p->stime = p->gtime = 0;
2394#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2395 p->utimescaled = p->stimescaled = 0;
2396#endif
2397 prev_cputime_init(&p->prev_cputime);
2398
2399#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2400 seqcount_init(&p->vtime.seqcount);
2401 p->vtime.starttime = 0;
2402 p->vtime.state = VTIME_INACTIVE;
2403#endif
2404
2405#ifdef CONFIG_IO_URING
2406 p->io_uring = NULL;
2407#endif
2408
2409 p->default_timer_slack_ns = current->timer_slack_ns;
2410
2411#ifdef CONFIG_PSI
2412 p->psi_flags = 0;
2413#endif
2414
2415 task_io_accounting_init(&p->ioac);
2416 acct_clear_integrals(p);
2417
2418 posix_cputimers_init(&p->posix_cputimers);
2419
2420 p->io_context = NULL;
2421 audit_set_context(p, NULL);
2422 cgroup_fork(p);
2423 if (args->kthread) {
2424 if (!set_kthread_struct(p))
2425 goto bad_fork_cleanup_delayacct;
2426 }
2427#ifdef CONFIG_NUMA
2428 p->mempolicy = mpol_dup(p->mempolicy);
2429 if (IS_ERR(p->mempolicy)) {
2430 retval = PTR_ERR(p->mempolicy);
2431 p->mempolicy = NULL;
2432 goto bad_fork_cleanup_delayacct;
2433 }
2434#endif
2435#ifdef CONFIG_CPUSETS
2436 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2437 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2438 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2439#endif
2440#ifdef CONFIG_TRACE_IRQFLAGS
2441 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2442 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2443 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2444 p->softirqs_enabled = 1;
2445 p->softirq_context = 0;
2446#endif
2447
2448 p->pagefault_disabled = 0;
2449
2450#ifdef CONFIG_LOCKDEP
2451 lockdep_init_task(p);
2452#endif
2453
2454#ifdef CONFIG_DEBUG_MUTEXES
2455 p->blocked_on = NULL; /* not blocked yet */
2456#endif
2457#ifdef CONFIG_BCACHE
2458 p->sequential_io = 0;
2459 p->sequential_io_avg = 0;
2460#endif
2461#ifdef CONFIG_BPF_SYSCALL
2462 RCU_INIT_POINTER(p->bpf_storage, NULL);
2463 p->bpf_ctx = NULL;
2464#endif
2465
2466 /* Perform scheduler related setup. Assign this task to a CPU. */
2467 retval = sched_fork(clone_flags, p);
2468 if (retval)
2469 goto bad_fork_cleanup_policy;
2470
2471 retval = perf_event_init_task(p, clone_flags);
2472 if (retval)
2473 goto bad_fork_cleanup_policy;
2474 retval = audit_alloc(p);
2475 if (retval)
2476 goto bad_fork_cleanup_perf;
2477 /* copy all the process information */
2478 shm_init_task(p);
2479 retval = security_task_alloc(p, clone_flags);
2480 if (retval)
2481 goto bad_fork_cleanup_audit;
2482 retval = copy_semundo(clone_flags, p);
2483 if (retval)
2484 goto bad_fork_cleanup_security;
2485 retval = copy_files(clone_flags, p, args->no_files);
2486 if (retval)
2487 goto bad_fork_cleanup_semundo;
2488 retval = copy_fs(clone_flags, p);
2489 if (retval)
2490 goto bad_fork_cleanup_files;
2491 retval = copy_sighand(clone_flags, p);
2492 if (retval)
2493 goto bad_fork_cleanup_fs;
2494 retval = copy_signal(clone_flags, p);
2495 if (retval)
2496 goto bad_fork_cleanup_sighand;
2497 retval = copy_mm(clone_flags, p);
2498 if (retval)
2499 goto bad_fork_cleanup_signal;
2500 retval = copy_namespaces(clone_flags, p);
2501 if (retval)
2502 goto bad_fork_cleanup_mm;
2503 retval = copy_io(clone_flags, p);
2504 if (retval)
2505 goto bad_fork_cleanup_namespaces;
2506 retval = copy_thread(p, args);
2507 if (retval)
2508 goto bad_fork_cleanup_io;
2509
2510 stackleak_task_init(p);
2511
2512 if (pid != &init_struct_pid) {
2513 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2514 args->set_tid_size);
2515 if (IS_ERR(pid)) {
2516 retval = PTR_ERR(pid);
2517 goto bad_fork_cleanup_thread;
2518 }
2519 }
2520
2521 /*
2522 * This has to happen after we've potentially unshared the file
2523 * descriptor table (so that the pidfd doesn't leak into the child
2524 * if the fd table isn't shared).
2525 */
2526 if (clone_flags & CLONE_PIDFD) {
2527 /* Note that no task has been attached to @pid yet. */
2528 retval = __pidfd_prepare(pid, O_RDWR | O_CLOEXEC, &pidfile);
2529 if (retval < 0)
2530 goto bad_fork_free_pid;
2531 pidfd = retval;
2532
2533 retval = put_user(pidfd, args->pidfd);
2534 if (retval)
2535 goto bad_fork_put_pidfd;
2536 }
2537
2538#ifdef CONFIG_BLOCK
2539 p->plug = NULL;
2540#endif
2541 futex_init_task(p);
2542
2543 /*
2544 * sigaltstack should be cleared when sharing the same VM
2545 */
2546 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2547 sas_ss_reset(p);
2548
2549 /*
2550 * Syscall tracing and stepping should be turned off in the
2551 * child regardless of CLONE_PTRACE.
2552 */
2553 user_disable_single_step(p);
2554 clear_task_syscall_work(p, SYSCALL_TRACE);
2555#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2556 clear_task_syscall_work(p, SYSCALL_EMU);
2557#endif
2558 clear_tsk_latency_tracing(p);
2559
2560 /* ok, now we should be set up.. */
2561 p->pid = pid_nr(pid);
2562 if (clone_flags & CLONE_THREAD) {
2563 p->group_leader = current->group_leader;
2564 p->tgid = current->tgid;
2565 } else {
2566 p->group_leader = p;
2567 p->tgid = p->pid;
2568 }
2569
2570 p->nr_dirtied = 0;
2571 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2572 p->dirty_paused_when = 0;
2573
2574 p->pdeath_signal = 0;
2575 p->task_works = NULL;
2576 clear_posix_cputimers_work(p);
2577
2578#ifdef CONFIG_KRETPROBES
2579 p->kretprobe_instances.first = NULL;
2580#endif
2581#ifdef CONFIG_RETHOOK
2582 p->rethooks.first = NULL;
2583#endif
2584
2585 /*
2586 * Ensure that the cgroup subsystem policies allow the new process to be
2587 * forked. It should be noted that the new process's css_set can be changed
2588 * between here and cgroup_post_fork() if an organisation operation is in
2589 * progress.
2590 */
2591 retval = cgroup_can_fork(p, args);
2592 if (retval)
2593 goto bad_fork_put_pidfd;
2594
2595 /*
2596 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2597 * the new task on the correct runqueue. All this *before* the task
2598 * becomes visible.
2599 *
2600 * This isn't part of ->can_fork() because while the re-cloning is
2601 * cgroup specific, it unconditionally needs to place the task on a
2602 * runqueue.
2603 */
2604 sched_cgroup_fork(p, args);
2605
2606 /*
2607 * From this point on we must avoid any synchronous user-space
2608 * communication until we take the tasklist-lock. In particular, we do
2609 * not want user-space to be able to predict the process start-time by
2610 * stalling fork(2) after we recorded the start_time but before it is
2611 * visible to the system.
2612 */
2613
2614 p->start_time = ktime_get_ns();
2615 p->start_boottime = ktime_get_boottime_ns();
2616
2617 /*
2618 * Make it visible to the rest of the system, but dont wake it up yet.
2619 * Need tasklist lock for parent etc handling!
2620 */
2621 write_lock_irq(&tasklist_lock);
2622
2623 /* CLONE_PARENT re-uses the old parent */
2624 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2625 p->real_parent = current->real_parent;
2626 p->parent_exec_id = current->parent_exec_id;
2627 if (clone_flags & CLONE_THREAD)
2628 p->exit_signal = -1;
2629 else
2630 p->exit_signal = current->group_leader->exit_signal;
2631 } else {
2632 p->real_parent = current;
2633 p->parent_exec_id = current->self_exec_id;
2634 p->exit_signal = args->exit_signal;
2635 }
2636
2637 klp_copy_process(p);
2638
2639 sched_core_fork(p);
2640
2641 spin_lock(¤t->sighand->siglock);
2642
2643 rv_task_fork(p);
2644
2645 rseq_fork(p, clone_flags);
2646
2647 /* Don't start children in a dying pid namespace */
2648 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2649 retval = -ENOMEM;
2650 goto bad_fork_cancel_cgroup;
2651 }
2652
2653 /* Let kill terminate clone/fork in the middle */
2654 if (fatal_signal_pending(current)) {
2655 retval = -EINTR;
2656 goto bad_fork_cancel_cgroup;
2657 }
2658
2659 /* No more failure paths after this point. */
2660
2661 /*
2662 * Copy seccomp details explicitly here, in case they were changed
2663 * before holding sighand lock.
2664 */
2665 copy_seccomp(p);
2666
2667 init_task_pid_links(p);
2668 if (likely(p->pid)) {
2669 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2670
2671 init_task_pid(p, PIDTYPE_PID, pid);
2672 if (thread_group_leader(p)) {
2673 init_task_pid(p, PIDTYPE_TGID, pid);
2674 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2675 init_task_pid(p, PIDTYPE_SID, task_session(current));
2676
2677 if (is_child_reaper(pid)) {
2678 ns_of_pid(pid)->child_reaper = p;
2679 p->signal->flags |= SIGNAL_UNKILLABLE;
2680 }
2681 p->signal->shared_pending.signal = delayed.signal;
2682 p->signal->tty = tty_kref_get(current->signal->tty);
2683 /*
2684 * Inherit has_child_subreaper flag under the same
2685 * tasklist_lock with adding child to the process tree
2686 * for propagate_has_child_subreaper optimization.
2687 */
2688 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2689 p->real_parent->signal->is_child_subreaper;
2690 list_add_tail(&p->sibling, &p->real_parent->children);
2691 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2692 attach_pid(p, PIDTYPE_TGID);
2693 attach_pid(p, PIDTYPE_PGID);
2694 attach_pid(p, PIDTYPE_SID);
2695 __this_cpu_inc(process_counts);
2696 } else {
2697 current->signal->nr_threads++;
2698 current->signal->quick_threads++;
2699 atomic_inc(¤t->signal->live);
2700 refcount_inc(¤t->signal->sigcnt);
2701 task_join_group_stop(p);
2702 list_add_tail_rcu(&p->thread_node,
2703 &p->signal->thread_head);
2704 }
2705 attach_pid(p, PIDTYPE_PID);
2706 nr_threads++;
2707 }
2708 total_forks++;
2709 hlist_del_init(&delayed.node);
2710 spin_unlock(¤t->sighand->siglock);
2711 syscall_tracepoint_update(p);
2712 write_unlock_irq(&tasklist_lock);
2713
2714 if (pidfile)
2715 fd_install(pidfd, pidfile);
2716
2717 proc_fork_connector(p);
2718 sched_post_fork(p);
2719 cgroup_post_fork(p, args);
2720 perf_event_fork(p);
2721
2722 trace_task_newtask(p, clone_flags);
2723 uprobe_copy_process(p, clone_flags);
2724 user_events_fork(p, clone_flags);
2725
2726 copy_oom_score_adj(clone_flags, p);
2727
2728 return p;
2729
2730bad_fork_cancel_cgroup:
2731 sched_core_free(p);
2732 spin_unlock(¤t->sighand->siglock);
2733 write_unlock_irq(&tasklist_lock);
2734 cgroup_cancel_fork(p, args);
2735bad_fork_put_pidfd:
2736 if (clone_flags & CLONE_PIDFD) {
2737 fput(pidfile);
2738 put_unused_fd(pidfd);
2739 }
2740bad_fork_free_pid:
2741 if (pid != &init_struct_pid)
2742 free_pid(pid);
2743bad_fork_cleanup_thread:
2744 exit_thread(p);
2745bad_fork_cleanup_io:
2746 if (p->io_context)
2747 exit_io_context(p);
2748bad_fork_cleanup_namespaces:
2749 exit_task_namespaces(p);
2750bad_fork_cleanup_mm:
2751 if (p->mm) {
2752 mm_clear_owner(p->mm, p);
2753 mmput(p->mm);
2754 }
2755bad_fork_cleanup_signal:
2756 if (!(clone_flags & CLONE_THREAD))
2757 free_signal_struct(p->signal);
2758bad_fork_cleanup_sighand:
2759 __cleanup_sighand(p->sighand);
2760bad_fork_cleanup_fs:
2761 exit_fs(p); /* blocking */
2762bad_fork_cleanup_files:
2763 exit_files(p); /* blocking */
2764bad_fork_cleanup_semundo:
2765 exit_sem(p);
2766bad_fork_cleanup_security:
2767 security_task_free(p);
2768bad_fork_cleanup_audit:
2769 audit_free(p);
2770bad_fork_cleanup_perf:
2771 perf_event_free_task(p);
2772bad_fork_cleanup_policy:
2773 lockdep_free_task(p);
2774#ifdef CONFIG_NUMA
2775 mpol_put(p->mempolicy);
2776#endif
2777bad_fork_cleanup_delayacct:
2778 delayacct_tsk_free(p);
2779bad_fork_cleanup_count:
2780 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2781 exit_creds(p);
2782bad_fork_free:
2783 WRITE_ONCE(p->__state, TASK_DEAD);
2784 exit_task_stack_account(p);
2785 put_task_stack(p);
2786 delayed_free_task(p);
2787fork_out:
2788 spin_lock_irq(¤t->sighand->siglock);
2789 hlist_del_init(&delayed.node);
2790 spin_unlock_irq(¤t->sighand->siglock);
2791 return ERR_PTR(retval);
2792}
2793
2794static inline void init_idle_pids(struct task_struct *idle)
2795{
2796 enum pid_type type;
2797
2798 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2799 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2800 init_task_pid(idle, type, &init_struct_pid);
2801 }
2802}
2803
2804static int idle_dummy(void *dummy)
2805{
2806 /* This function is never called */
2807 return 0;
2808}
2809
2810struct task_struct * __init fork_idle(int cpu)
2811{
2812 struct task_struct *task;
2813 struct kernel_clone_args args = {
2814 .flags = CLONE_VM,
2815 .fn = &idle_dummy,
2816 .fn_arg = NULL,
2817 .kthread = 1,
2818 .idle = 1,
2819 };
2820
2821 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2822 if (!IS_ERR(task)) {
2823 init_idle_pids(task);
2824 init_idle(task, cpu);
2825 }
2826
2827 return task;
2828}
2829
2830/*
2831 * This is like kernel_clone(), but shaved down and tailored to just
2832 * creating io_uring workers. It returns a created task, or an error pointer.
2833 * The returned task is inactive, and the caller must fire it up through
2834 * wake_up_new_task(p). All signals are blocked in the created task.
2835 */
2836struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2837{
2838 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2839 CLONE_IO;
2840 struct kernel_clone_args args = {
2841 .flags = ((lower_32_bits(flags) | CLONE_VM |
2842 CLONE_UNTRACED) & ~CSIGNAL),
2843 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2844 .fn = fn,
2845 .fn_arg = arg,
2846 .io_thread = 1,
2847 .user_worker = 1,
2848 };
2849
2850 return copy_process(NULL, 0, node, &args);
2851}
2852
2853/*
2854 * Ok, this is the main fork-routine.
2855 *
2856 * It copies the process, and if successful kick-starts
2857 * it and waits for it to finish using the VM if required.
2858 *
2859 * args->exit_signal is expected to be checked for sanity by the caller.
2860 */
2861pid_t kernel_clone(struct kernel_clone_args *args)
2862{
2863 u64 clone_flags = args->flags;
2864 struct completion vfork;
2865 struct pid *pid;
2866 struct task_struct *p;
2867 int trace = 0;
2868 pid_t nr;
2869
2870 /*
2871 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2872 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2873 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2874 * field in struct clone_args and it still doesn't make sense to have
2875 * them both point at the same memory location. Performing this check
2876 * here has the advantage that we don't need to have a separate helper
2877 * to check for legacy clone().
2878 */
2879 if ((args->flags & CLONE_PIDFD) &&
2880 (args->flags & CLONE_PARENT_SETTID) &&
2881 (args->pidfd == args->parent_tid))
2882 return -EINVAL;
2883
2884 /*
2885 * Determine whether and which event to report to ptracer. When
2886 * called from kernel_thread or CLONE_UNTRACED is explicitly
2887 * requested, no event is reported; otherwise, report if the event
2888 * for the type of forking is enabled.
2889 */
2890 if (!(clone_flags & CLONE_UNTRACED)) {
2891 if (clone_flags & CLONE_VFORK)
2892 trace = PTRACE_EVENT_VFORK;
2893 else if (args->exit_signal != SIGCHLD)
2894 trace = PTRACE_EVENT_CLONE;
2895 else
2896 trace = PTRACE_EVENT_FORK;
2897
2898 if (likely(!ptrace_event_enabled(current, trace)))
2899 trace = 0;
2900 }
2901
2902 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2903 add_latent_entropy();
2904
2905 if (IS_ERR(p))
2906 return PTR_ERR(p);
2907
2908 /*
2909 * Do this prior waking up the new thread - the thread pointer
2910 * might get invalid after that point, if the thread exits quickly.
2911 */
2912 trace_sched_process_fork(current, p);
2913
2914 pid = get_task_pid(p, PIDTYPE_PID);
2915 nr = pid_vnr(pid);
2916
2917 if (clone_flags & CLONE_PARENT_SETTID)
2918 put_user(nr, args->parent_tid);
2919
2920 if (clone_flags & CLONE_VFORK) {
2921 p->vfork_done = &vfork;
2922 init_completion(&vfork);
2923 get_task_struct(p);
2924 }
2925
2926 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2927 /* lock the task to synchronize with memcg migration */
2928 task_lock(p);
2929 lru_gen_add_mm(p->mm);
2930 task_unlock(p);
2931 }
2932
2933 wake_up_new_task(p);
2934
2935 /* forking complete and child started to run, tell ptracer */
2936 if (unlikely(trace))
2937 ptrace_event_pid(trace, pid);
2938
2939 if (clone_flags & CLONE_VFORK) {
2940 if (!wait_for_vfork_done(p, &vfork))
2941 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2942 }
2943
2944 put_pid(pid);
2945 return nr;
2946}
2947
2948/*
2949 * Create a kernel thread.
2950 */
2951pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2952 unsigned long flags)
2953{
2954 struct kernel_clone_args args = {
2955 .flags = ((lower_32_bits(flags) | CLONE_VM |
2956 CLONE_UNTRACED) & ~CSIGNAL),
2957 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2958 .fn = fn,
2959 .fn_arg = arg,
2960 .name = name,
2961 .kthread = 1,
2962 };
2963
2964 return kernel_clone(&args);
2965}
2966
2967/*
2968 * Create a user mode thread.
2969 */
2970pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2971{
2972 struct kernel_clone_args args = {
2973 .flags = ((lower_32_bits(flags) | CLONE_VM |
2974 CLONE_UNTRACED) & ~CSIGNAL),
2975 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2976 .fn = fn,
2977 .fn_arg = arg,
2978 };
2979
2980 return kernel_clone(&args);
2981}
2982
2983#ifdef __ARCH_WANT_SYS_FORK
2984SYSCALL_DEFINE0(fork)
2985{
2986#ifdef CONFIG_MMU
2987 struct kernel_clone_args args = {
2988 .exit_signal = SIGCHLD,
2989 };
2990
2991 return kernel_clone(&args);
2992#else
2993 /* can not support in nommu mode */
2994 return -EINVAL;
2995#endif
2996}
2997#endif
2998
2999#ifdef __ARCH_WANT_SYS_VFORK
3000SYSCALL_DEFINE0(vfork)
3001{
3002 struct kernel_clone_args args = {
3003 .flags = CLONE_VFORK | CLONE_VM,
3004 .exit_signal = SIGCHLD,
3005 };
3006
3007 return kernel_clone(&args);
3008}
3009#endif
3010
3011#ifdef __ARCH_WANT_SYS_CLONE
3012#ifdef CONFIG_CLONE_BACKWARDS
3013SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3014 int __user *, parent_tidptr,
3015 unsigned long, tls,
3016 int __user *, child_tidptr)
3017#elif defined(CONFIG_CLONE_BACKWARDS2)
3018SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
3019 int __user *, parent_tidptr,
3020 int __user *, child_tidptr,
3021 unsigned long, tls)
3022#elif defined(CONFIG_CLONE_BACKWARDS3)
3023SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
3024 int, stack_size,
3025 int __user *, parent_tidptr,
3026 int __user *, child_tidptr,
3027 unsigned long, tls)
3028#else
3029SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
3030 int __user *, parent_tidptr,
3031 int __user *, child_tidptr,
3032 unsigned long, tls)
3033#endif
3034{
3035 struct kernel_clone_args args = {
3036 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
3037 .pidfd = parent_tidptr,
3038 .child_tid = child_tidptr,
3039 .parent_tid = parent_tidptr,
3040 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
3041 .stack = newsp,
3042 .tls = tls,
3043 };
3044
3045 return kernel_clone(&args);
3046}
3047#endif
3048
3049#ifdef __ARCH_WANT_SYS_CLONE3
3050
3051noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
3052 struct clone_args __user *uargs,
3053 size_t usize)
3054{
3055 int err;
3056 struct clone_args args;
3057 pid_t *kset_tid = kargs->set_tid;
3058
3059 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
3060 CLONE_ARGS_SIZE_VER0);
3061 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
3062 CLONE_ARGS_SIZE_VER1);
3063 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
3064 CLONE_ARGS_SIZE_VER2);
3065 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
3066
3067 if (unlikely(usize > PAGE_SIZE))
3068 return -E2BIG;
3069 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
3070 return -EINVAL;
3071
3072 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
3073 if (err)
3074 return err;
3075
3076 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
3077 return -EINVAL;
3078
3079 if (unlikely(!args.set_tid && args.set_tid_size > 0))
3080 return -EINVAL;
3081
3082 if (unlikely(args.set_tid && args.set_tid_size == 0))
3083 return -EINVAL;
3084
3085 /*
3086 * Verify that higher 32bits of exit_signal are unset and that
3087 * it is a valid signal
3088 */
3089 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3090 !valid_signal(args.exit_signal)))
3091 return -EINVAL;
3092
3093 if ((args.flags & CLONE_INTO_CGROUP) &&
3094 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3095 return -EINVAL;
3096
3097 *kargs = (struct kernel_clone_args){
3098 .flags = args.flags,
3099 .pidfd = u64_to_user_ptr(args.pidfd),
3100 .child_tid = u64_to_user_ptr(args.child_tid),
3101 .parent_tid = u64_to_user_ptr(args.parent_tid),
3102 .exit_signal = args.exit_signal,
3103 .stack = args.stack,
3104 .stack_size = args.stack_size,
3105 .tls = args.tls,
3106 .set_tid_size = args.set_tid_size,
3107 .cgroup = args.cgroup,
3108 };
3109
3110 if (args.set_tid &&
3111 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3112 (kargs->set_tid_size * sizeof(pid_t))))
3113 return -EFAULT;
3114
3115 kargs->set_tid = kset_tid;
3116
3117 return 0;
3118}
3119
3120/**
3121 * clone3_stack_valid - check and prepare stack
3122 * @kargs: kernel clone args
3123 *
3124 * Verify that the stack arguments userspace gave us are sane.
3125 * In addition, set the stack direction for userspace since it's easy for us to
3126 * determine.
3127 */
3128static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3129{
3130 if (kargs->stack == 0) {
3131 if (kargs->stack_size > 0)
3132 return false;
3133 } else {
3134 if (kargs->stack_size == 0)
3135 return false;
3136
3137 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3138 return false;
3139
3140#if !defined(CONFIG_STACK_GROWSUP)
3141 kargs->stack += kargs->stack_size;
3142#endif
3143 }
3144
3145 return true;
3146}
3147
3148static bool clone3_args_valid(struct kernel_clone_args *kargs)
3149{
3150 /* Verify that no unknown flags are passed along. */
3151 if (kargs->flags &
3152 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3153 return false;
3154
3155 /*
3156 * - make the CLONE_DETACHED bit reusable for clone3
3157 * - make the CSIGNAL bits reusable for clone3
3158 */
3159 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3160 return false;
3161
3162 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3163 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3164 return false;
3165
3166 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3167 kargs->exit_signal)
3168 return false;
3169
3170 if (!clone3_stack_valid(kargs))
3171 return false;
3172
3173 return true;
3174}
3175
3176/**
3177 * sys_clone3 - create a new process with specific properties
3178 * @uargs: argument structure
3179 * @size: size of @uargs
3180 *
3181 * clone3() is the extensible successor to clone()/clone2().
3182 * It takes a struct as argument that is versioned by its size.
3183 *
3184 * Return: On success, a positive PID for the child process.
3185 * On error, a negative errno number.
3186 */
3187SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3188{
3189 int err;
3190
3191 struct kernel_clone_args kargs;
3192 pid_t set_tid[MAX_PID_NS_LEVEL];
3193
3194 kargs.set_tid = set_tid;
3195
3196 err = copy_clone_args_from_user(&kargs, uargs, size);
3197 if (err)
3198 return err;
3199
3200 if (!clone3_args_valid(&kargs))
3201 return -EINVAL;
3202
3203 return kernel_clone(&kargs);
3204}
3205#endif
3206
3207void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3208{
3209 struct task_struct *leader, *parent, *child;
3210 int res;
3211
3212 read_lock(&tasklist_lock);
3213 leader = top = top->group_leader;
3214down:
3215 for_each_thread(leader, parent) {
3216 list_for_each_entry(child, &parent->children, sibling) {
3217 res = visitor(child, data);
3218 if (res) {
3219 if (res < 0)
3220 goto out;
3221 leader = child;
3222 goto down;
3223 }
3224up:
3225 ;
3226 }
3227 }
3228
3229 if (leader != top) {
3230 child = leader;
3231 parent = child->real_parent;
3232 leader = parent->group_leader;
3233 goto up;
3234 }
3235out:
3236 read_unlock(&tasklist_lock);
3237}
3238
3239#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3240#define ARCH_MIN_MMSTRUCT_ALIGN 0
3241#endif
3242
3243static void sighand_ctor(void *data)
3244{
3245 struct sighand_struct *sighand = data;
3246
3247 spin_lock_init(&sighand->siglock);
3248 init_waitqueue_head(&sighand->signalfd_wqh);
3249}
3250
3251void __init mm_cache_init(void)
3252{
3253 unsigned int mm_size;
3254
3255 /*
3256 * The mm_cpumask is located at the end of mm_struct, and is
3257 * dynamically sized based on the maximum CPU number this system
3258 * can have, taking hotplug into account (nr_cpu_ids).
3259 */
3260 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3261
3262 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3263 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3264 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3265 offsetof(struct mm_struct, saved_auxv),
3266 sizeof_field(struct mm_struct, saved_auxv),
3267 NULL);
3268}
3269
3270void __init proc_caches_init(void)
3271{
3272 sighand_cachep = kmem_cache_create("sighand_cache",
3273 sizeof(struct sighand_struct), 0,
3274 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3275 SLAB_ACCOUNT, sighand_ctor);
3276 signal_cachep = kmem_cache_create("signal_cache",
3277 sizeof(struct signal_struct), 0,
3278 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3279 NULL);
3280 files_cachep = kmem_cache_create("files_cache",
3281 sizeof(struct files_struct), 0,
3282 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3283 NULL);
3284 fs_cachep = kmem_cache_create("fs_cache",
3285 sizeof(struct fs_struct), 0,
3286 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3287 NULL);
3288
3289 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3290#ifdef CONFIG_PER_VMA_LOCK
3291 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3292#endif
3293 mmap_init();
3294 nsproxy_cache_init();
3295}
3296
3297/*
3298 * Check constraints on flags passed to the unshare system call.
3299 */
3300static int check_unshare_flags(unsigned long unshare_flags)
3301{
3302 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3303 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3304 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3305 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3306 CLONE_NEWTIME))
3307 return -EINVAL;
3308 /*
3309 * Not implemented, but pretend it works if there is nothing
3310 * to unshare. Note that unsharing the address space or the
3311 * signal handlers also need to unshare the signal queues (aka
3312 * CLONE_THREAD).
3313 */
3314 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3315 if (!thread_group_empty(current))
3316 return -EINVAL;
3317 }
3318 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3319 if (refcount_read(¤t->sighand->count) > 1)
3320 return -EINVAL;
3321 }
3322 if (unshare_flags & CLONE_VM) {
3323 if (!current_is_single_threaded())
3324 return -EINVAL;
3325 }
3326
3327 return 0;
3328}
3329
3330/*
3331 * Unshare the filesystem structure if it is being shared
3332 */
3333static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3334{
3335 struct fs_struct *fs = current->fs;
3336
3337 if (!(unshare_flags & CLONE_FS) || !fs)
3338 return 0;
3339
3340 /* don't need lock here; in the worst case we'll do useless copy */
3341 if (fs->users == 1)
3342 return 0;
3343
3344 *new_fsp = copy_fs_struct(fs);
3345 if (!*new_fsp)
3346 return -ENOMEM;
3347
3348 return 0;
3349}
3350
3351/*
3352 * Unshare file descriptor table if it is being shared
3353 */
3354int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3355 struct files_struct **new_fdp)
3356{
3357 struct files_struct *fd = current->files;
3358 int error = 0;
3359
3360 if ((unshare_flags & CLONE_FILES) &&
3361 (fd && atomic_read(&fd->count) > 1)) {
3362 *new_fdp = dup_fd(fd, max_fds, &error);
3363 if (!*new_fdp)
3364 return error;
3365 }
3366
3367 return 0;
3368}
3369
3370/*
3371 * unshare allows a process to 'unshare' part of the process
3372 * context which was originally shared using clone. copy_*
3373 * functions used by kernel_clone() cannot be used here directly
3374 * because they modify an inactive task_struct that is being
3375 * constructed. Here we are modifying the current, active,
3376 * task_struct.
3377 */
3378int ksys_unshare(unsigned long unshare_flags)
3379{
3380 struct fs_struct *fs, *new_fs = NULL;
3381 struct files_struct *new_fd = NULL;
3382 struct cred *new_cred = NULL;
3383 struct nsproxy *new_nsproxy = NULL;
3384 int do_sysvsem = 0;
3385 int err;
3386
3387 /*
3388 * If unsharing a user namespace must also unshare the thread group
3389 * and unshare the filesystem root and working directories.
3390 */
3391 if (unshare_flags & CLONE_NEWUSER)
3392 unshare_flags |= CLONE_THREAD | CLONE_FS;
3393 /*
3394 * If unsharing vm, must also unshare signal handlers.
3395 */
3396 if (unshare_flags & CLONE_VM)
3397 unshare_flags |= CLONE_SIGHAND;
3398 /*
3399 * If unsharing a signal handlers, must also unshare the signal queues.
3400 */
3401 if (unshare_flags & CLONE_SIGHAND)
3402 unshare_flags |= CLONE_THREAD;
3403 /*
3404 * If unsharing namespace, must also unshare filesystem information.
3405 */
3406 if (unshare_flags & CLONE_NEWNS)
3407 unshare_flags |= CLONE_FS;
3408
3409 err = check_unshare_flags(unshare_flags);
3410 if (err)
3411 goto bad_unshare_out;
3412 /*
3413 * CLONE_NEWIPC must also detach from the undolist: after switching
3414 * to a new ipc namespace, the semaphore arrays from the old
3415 * namespace are unreachable.
3416 */
3417 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3418 do_sysvsem = 1;
3419 err = unshare_fs(unshare_flags, &new_fs);
3420 if (err)
3421 goto bad_unshare_out;
3422 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3423 if (err)
3424 goto bad_unshare_cleanup_fs;
3425 err = unshare_userns(unshare_flags, &new_cred);
3426 if (err)
3427 goto bad_unshare_cleanup_fd;
3428 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3429 new_cred, new_fs);
3430 if (err)
3431 goto bad_unshare_cleanup_cred;
3432
3433 if (new_cred) {
3434 err = set_cred_ucounts(new_cred);
3435 if (err)
3436 goto bad_unshare_cleanup_cred;
3437 }
3438
3439 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3440 if (do_sysvsem) {
3441 /*
3442 * CLONE_SYSVSEM is equivalent to sys_exit().
3443 */
3444 exit_sem(current);
3445 }
3446 if (unshare_flags & CLONE_NEWIPC) {
3447 /* Orphan segments in old ns (see sem above). */
3448 exit_shm(current);
3449 shm_init_task(current);
3450 }
3451
3452 if (new_nsproxy)
3453 switch_task_namespaces(current, new_nsproxy);
3454
3455 task_lock(current);
3456
3457 if (new_fs) {
3458 fs = current->fs;
3459 spin_lock(&fs->lock);
3460 current->fs = new_fs;
3461 if (--fs->users)
3462 new_fs = NULL;
3463 else
3464 new_fs = fs;
3465 spin_unlock(&fs->lock);
3466 }
3467
3468 if (new_fd)
3469 swap(current->files, new_fd);
3470
3471 task_unlock(current);
3472
3473 if (new_cred) {
3474 /* Install the new user namespace */
3475 commit_creds(new_cred);
3476 new_cred = NULL;
3477 }
3478 }
3479
3480 perf_event_namespaces(current);
3481
3482bad_unshare_cleanup_cred:
3483 if (new_cred)
3484 put_cred(new_cred);
3485bad_unshare_cleanup_fd:
3486 if (new_fd)
3487 put_files_struct(new_fd);
3488
3489bad_unshare_cleanup_fs:
3490 if (new_fs)
3491 free_fs_struct(new_fs);
3492
3493bad_unshare_out:
3494 return err;
3495}
3496
3497SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3498{
3499 return ksys_unshare(unshare_flags);
3500}
3501
3502/*
3503 * Helper to unshare the files of the current task.
3504 * We don't want to expose copy_files internals to
3505 * the exec layer of the kernel.
3506 */
3507
3508int unshare_files(void)
3509{
3510 struct task_struct *task = current;
3511 struct files_struct *old, *copy = NULL;
3512 int error;
3513
3514 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3515 if (error || !copy)
3516 return error;
3517
3518 old = task->files;
3519 task_lock(task);
3520 task->files = copy;
3521 task_unlock(task);
3522 put_files_struct(old);
3523 return 0;
3524}
3525
3526int sysctl_max_threads(struct ctl_table *table, int write,
3527 void *buffer, size_t *lenp, loff_t *ppos)
3528{
3529 struct ctl_table t;
3530 int ret;
3531 int threads = max_threads;
3532 int min = 1;
3533 int max = MAX_THREADS;
3534
3535 t = *table;
3536 t.data = &threads;
3537 t.extra1 = &min;
3538 t.extra2 = &max;
3539
3540 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3541 if (ret || !write)
3542 return ret;
3543
3544 max_threads = threads;
3545
3546 return 0;
3547}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/syscall_user_dispatch.h>
57#include <linux/jiffies.h>
58#include <linux/futex.h>
59#include <linux/compat.h>
60#include <linux/kthread.h>
61#include <linux/task_io_accounting_ops.h>
62#include <linux/rcupdate.h>
63#include <linux/ptrace.h>
64#include <linux/mount.h>
65#include <linux/audit.h>
66#include <linux/memcontrol.h>
67#include <linux/ftrace.h>
68#include <linux/proc_fs.h>
69#include <linux/profile.h>
70#include <linux/rmap.h>
71#include <linux/ksm.h>
72#include <linux/acct.h>
73#include <linux/userfaultfd_k.h>
74#include <linux/tsacct_kern.h>
75#include <linux/cn_proc.h>
76#include <linux/freezer.h>
77#include <linux/delayacct.h>
78#include <linux/taskstats_kern.h>
79#include <linux/tty.h>
80#include <linux/fs_struct.h>
81#include <linux/magic.h>
82#include <linux/perf_event.h>
83#include <linux/posix-timers.h>
84#include <linux/user-return-notifier.h>
85#include <linux/oom.h>
86#include <linux/khugepaged.h>
87#include <linux/signalfd.h>
88#include <linux/uprobes.h>
89#include <linux/aio.h>
90#include <linux/compiler.h>
91#include <linux/sysctl.h>
92#include <linux/kcov.h>
93#include <linux/livepatch.h>
94#include <linux/thread_info.h>
95#include <linux/stackleak.h>
96#include <linux/kasan.h>
97#include <linux/scs.h>
98#include <linux/io_uring.h>
99#include <linux/bpf.h>
100#include <linux/stackprotector.h>
101#include <linux/user_events.h>
102#include <linux/iommu.h>
103#include <linux/rseq.h>
104#include <uapi/linux/pidfd.h>
105#include <linux/pidfs.h>
106
107#include <asm/pgalloc.h>
108#include <linux/uaccess.h>
109#include <asm/mmu_context.h>
110#include <asm/cacheflush.h>
111#include <asm/tlbflush.h>
112
113#include <trace/events/sched.h>
114
115#define CREATE_TRACE_POINTS
116#include <trace/events/task.h>
117
118/*
119 * Minimum number of threads to boot the kernel
120 */
121#define MIN_THREADS 20
122
123/*
124 * Maximum number of threads
125 */
126#define MAX_THREADS FUTEX_TID_MASK
127
128/*
129 * Protected counters by write_lock_irq(&tasklist_lock)
130 */
131unsigned long total_forks; /* Handle normal Linux uptimes. */
132int nr_threads; /* The idle threads do not count.. */
133
134static int max_threads; /* tunable limit on nr_threads */
135
136#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
137
138static const char * const resident_page_types[] = {
139 NAMED_ARRAY_INDEX(MM_FILEPAGES),
140 NAMED_ARRAY_INDEX(MM_ANONPAGES),
141 NAMED_ARRAY_INDEX(MM_SWAPENTS),
142 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
143};
144
145DEFINE_PER_CPU(unsigned long, process_counts) = 0;
146
147__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
148
149#ifdef CONFIG_PROVE_RCU
150int lockdep_tasklist_lock_is_held(void)
151{
152 return lockdep_is_held(&tasklist_lock);
153}
154EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
155#endif /* #ifdef CONFIG_PROVE_RCU */
156
157int nr_processes(void)
158{
159 int cpu;
160 int total = 0;
161
162 for_each_possible_cpu(cpu)
163 total += per_cpu(process_counts, cpu);
164
165 return total;
166}
167
168void __weak arch_release_task_struct(struct task_struct *tsk)
169{
170}
171
172static struct kmem_cache *task_struct_cachep;
173
174static inline struct task_struct *alloc_task_struct_node(int node)
175{
176 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
177}
178
179static inline void free_task_struct(struct task_struct *tsk)
180{
181 kmem_cache_free(task_struct_cachep, tsk);
182}
183
184/*
185 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
186 * kmemcache based allocator.
187 */
188# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189
190# ifdef CONFIG_VMAP_STACK
191/*
192 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
193 * flush. Try to minimize the number of calls by caching stacks.
194 */
195#define NR_CACHED_STACKS 2
196static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
197
198struct vm_stack {
199 struct rcu_head rcu;
200 struct vm_struct *stack_vm_area;
201};
202
203static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
204{
205 unsigned int i;
206
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
209 continue;
210 return true;
211 }
212 return false;
213}
214
215static void thread_stack_free_rcu(struct rcu_head *rh)
216{
217 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218
219 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
220 return;
221
222 vfree(vm_stack);
223}
224
225static void thread_stack_delayed_free(struct task_struct *tsk)
226{
227 struct vm_stack *vm_stack = tsk->stack;
228
229 vm_stack->stack_vm_area = tsk->stack_vm_area;
230 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231}
232
233static int free_vm_stack_cache(unsigned int cpu)
234{
235 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 int i;
237
238 for (i = 0; i < NR_CACHED_STACKS; i++) {
239 struct vm_struct *vm_stack = cached_vm_stacks[i];
240
241 if (!vm_stack)
242 continue;
243
244 vfree(vm_stack->addr);
245 cached_vm_stacks[i] = NULL;
246 }
247
248 return 0;
249}
250
251static int memcg_charge_kernel_stack(struct vm_struct *vm)
252{
253 int i;
254 int ret;
255 int nr_charged = 0;
256
257 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258
259 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
260 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
261 if (ret)
262 goto err;
263 nr_charged++;
264 }
265 return 0;
266err:
267 for (i = 0; i < nr_charged; i++)
268 memcg_kmem_uncharge_page(vm->pages[i], 0);
269 return ret;
270}
271
272static int alloc_thread_stack_node(struct task_struct *tsk, int node)
273{
274 struct vm_struct *vm;
275 void *stack;
276 int i;
277
278 for (i = 0; i < NR_CACHED_STACKS; i++) {
279 struct vm_struct *s;
280
281 s = this_cpu_xchg(cached_stacks[i], NULL);
282
283 if (!s)
284 continue;
285
286 /* Reset stack metadata. */
287 kasan_unpoison_range(s->addr, THREAD_SIZE);
288
289 stack = kasan_reset_tag(s->addr);
290
291 /* Clear stale pointers from reused stack. */
292 memset(stack, 0, THREAD_SIZE);
293
294 if (memcg_charge_kernel_stack(s)) {
295 vfree(s->addr);
296 return -ENOMEM;
297 }
298
299 tsk->stack_vm_area = s;
300 tsk->stack = stack;
301 return 0;
302 }
303
304 /*
305 * Allocated stacks are cached and later reused by new threads,
306 * so memcg accounting is performed manually on assigning/releasing
307 * stacks to tasks. Drop __GFP_ACCOUNT.
308 */
309 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
310 VMALLOC_START, VMALLOC_END,
311 THREADINFO_GFP & ~__GFP_ACCOUNT,
312 PAGE_KERNEL,
313 0, node, __builtin_return_address(0));
314 if (!stack)
315 return -ENOMEM;
316
317 vm = find_vm_area(stack);
318 if (memcg_charge_kernel_stack(vm)) {
319 vfree(stack);
320 return -ENOMEM;
321 }
322 /*
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
326 */
327 tsk->stack_vm_area = vm;
328 stack = kasan_reset_tag(stack);
329 tsk->stack = stack;
330 return 0;
331}
332
333static void free_thread_stack(struct task_struct *tsk)
334{
335 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
337
338 tsk->stack = NULL;
339 tsk->stack_vm_area = NULL;
340}
341
342# else /* !CONFIG_VMAP_STACK */
343
344static void thread_stack_free_rcu(struct rcu_head *rh)
345{
346 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
347}
348
349static void thread_stack_delayed_free(struct task_struct *tsk)
350{
351 struct rcu_head *rh = tsk->stack;
352
353 call_rcu(rh, thread_stack_free_rcu);
354}
355
356static int alloc_thread_stack_node(struct task_struct *tsk, int node)
357{
358 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
359 THREAD_SIZE_ORDER);
360
361 if (likely(page)) {
362 tsk->stack = kasan_reset_tag(page_address(page));
363 return 0;
364 }
365 return -ENOMEM;
366}
367
368static void free_thread_stack(struct task_struct *tsk)
369{
370 thread_stack_delayed_free(tsk);
371 tsk->stack = NULL;
372}
373
374# endif /* CONFIG_VMAP_STACK */
375# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
376
377static struct kmem_cache *thread_stack_cache;
378
379static void thread_stack_free_rcu(struct rcu_head *rh)
380{
381 kmem_cache_free(thread_stack_cache, rh);
382}
383
384static void thread_stack_delayed_free(struct task_struct *tsk)
385{
386 struct rcu_head *rh = tsk->stack;
387
388 call_rcu(rh, thread_stack_free_rcu);
389}
390
391static int alloc_thread_stack_node(struct task_struct *tsk, int node)
392{
393 unsigned long *stack;
394 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
395 stack = kasan_reset_tag(stack);
396 tsk->stack = stack;
397 return stack ? 0 : -ENOMEM;
398}
399
400static void free_thread_stack(struct task_struct *tsk)
401{
402 thread_stack_delayed_free(tsk);
403 tsk->stack = NULL;
404}
405
406void thread_stack_cache_init(void)
407{
408 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
409 THREAD_SIZE, THREAD_SIZE, 0, 0,
410 THREAD_SIZE, NULL);
411 BUG_ON(thread_stack_cache == NULL);
412}
413
414# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415
416/* SLAB cache for signal_struct structures (tsk->signal) */
417static struct kmem_cache *signal_cachep;
418
419/* SLAB cache for sighand_struct structures (tsk->sighand) */
420struct kmem_cache *sighand_cachep;
421
422/* SLAB cache for files_struct structures (tsk->files) */
423struct kmem_cache *files_cachep;
424
425/* SLAB cache for fs_struct structures (tsk->fs) */
426struct kmem_cache *fs_cachep;
427
428/* SLAB cache for vm_area_struct structures */
429static struct kmem_cache *vm_area_cachep;
430
431/* SLAB cache for mm_struct structures (tsk->mm) */
432static struct kmem_cache *mm_cachep;
433
434#ifdef CONFIG_PER_VMA_LOCK
435
436/* SLAB cache for vm_area_struct.lock */
437static struct kmem_cache *vma_lock_cachep;
438
439static bool vma_lock_alloc(struct vm_area_struct *vma)
440{
441 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
442 if (!vma->vm_lock)
443 return false;
444
445 init_rwsem(&vma->vm_lock->lock);
446 vma->vm_lock_seq = -1;
447
448 return true;
449}
450
451static inline void vma_lock_free(struct vm_area_struct *vma)
452{
453 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
454}
455
456#else /* CONFIG_PER_VMA_LOCK */
457
458static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
459static inline void vma_lock_free(struct vm_area_struct *vma) {}
460
461#endif /* CONFIG_PER_VMA_LOCK */
462
463struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
464{
465 struct vm_area_struct *vma;
466
467 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
468 if (!vma)
469 return NULL;
470
471 vma_init(vma, mm);
472 if (!vma_lock_alloc(vma)) {
473 kmem_cache_free(vm_area_cachep, vma);
474 return NULL;
475 }
476
477 return vma;
478}
479
480struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
481{
482 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
483
484 if (!new)
485 return NULL;
486
487 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
488 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
489 /*
490 * orig->shared.rb may be modified concurrently, but the clone
491 * will be reinitialized.
492 */
493 data_race(memcpy(new, orig, sizeof(*new)));
494 if (!vma_lock_alloc(new)) {
495 kmem_cache_free(vm_area_cachep, new);
496 return NULL;
497 }
498 INIT_LIST_HEAD(&new->anon_vma_chain);
499 vma_numab_state_init(new);
500 dup_anon_vma_name(orig, new);
501
502 return new;
503}
504
505void __vm_area_free(struct vm_area_struct *vma)
506{
507 vma_numab_state_free(vma);
508 free_anon_vma_name(vma);
509 vma_lock_free(vma);
510 kmem_cache_free(vm_area_cachep, vma);
511}
512
513#ifdef CONFIG_PER_VMA_LOCK
514static void vm_area_free_rcu_cb(struct rcu_head *head)
515{
516 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
517 vm_rcu);
518
519 /* The vma should not be locked while being destroyed. */
520 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
521 __vm_area_free(vma);
522}
523#endif
524
525void vm_area_free(struct vm_area_struct *vma)
526{
527#ifdef CONFIG_PER_VMA_LOCK
528 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
529#else
530 __vm_area_free(vma);
531#endif
532}
533
534static void account_kernel_stack(struct task_struct *tsk, int account)
535{
536 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
537 struct vm_struct *vm = task_stack_vm_area(tsk);
538 int i;
539
540 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
541 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
542 account * (PAGE_SIZE / 1024));
543 } else {
544 void *stack = task_stack_page(tsk);
545
546 /* All stack pages are in the same node. */
547 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
548 account * (THREAD_SIZE / 1024));
549 }
550}
551
552void exit_task_stack_account(struct task_struct *tsk)
553{
554 account_kernel_stack(tsk, -1);
555
556 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
557 struct vm_struct *vm;
558 int i;
559
560 vm = task_stack_vm_area(tsk);
561 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
562 memcg_kmem_uncharge_page(vm->pages[i], 0);
563 }
564}
565
566static void release_task_stack(struct task_struct *tsk)
567{
568 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
569 return; /* Better to leak the stack than to free prematurely */
570
571 free_thread_stack(tsk);
572}
573
574#ifdef CONFIG_THREAD_INFO_IN_TASK
575void put_task_stack(struct task_struct *tsk)
576{
577 if (refcount_dec_and_test(&tsk->stack_refcount))
578 release_task_stack(tsk);
579}
580#endif
581
582void free_task(struct task_struct *tsk)
583{
584#ifdef CONFIG_SECCOMP
585 WARN_ON_ONCE(tsk->seccomp.filter);
586#endif
587 release_user_cpus_ptr(tsk);
588 scs_release(tsk);
589
590#ifndef CONFIG_THREAD_INFO_IN_TASK
591 /*
592 * The task is finally done with both the stack and thread_info,
593 * so free both.
594 */
595 release_task_stack(tsk);
596#else
597 /*
598 * If the task had a separate stack allocation, it should be gone
599 * by now.
600 */
601 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
602#endif
603 rt_mutex_debug_task_free(tsk);
604 ftrace_graph_exit_task(tsk);
605 arch_release_task_struct(tsk);
606 if (tsk->flags & PF_KTHREAD)
607 free_kthread_struct(tsk);
608 bpf_task_storage_free(tsk);
609 free_task_struct(tsk);
610}
611EXPORT_SYMBOL(free_task);
612
613static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
614{
615 struct file *exe_file;
616
617 exe_file = get_mm_exe_file(oldmm);
618 RCU_INIT_POINTER(mm->exe_file, exe_file);
619 /*
620 * We depend on the oldmm having properly denied write access to the
621 * exe_file already.
622 */
623 if (exe_file && deny_write_access(exe_file))
624 pr_warn_once("deny_write_access() failed in %s\n", __func__);
625}
626
627#ifdef CONFIG_MMU
628static __latent_entropy int dup_mmap(struct mm_struct *mm,
629 struct mm_struct *oldmm)
630{
631 struct vm_area_struct *mpnt, *tmp;
632 int retval;
633 unsigned long charge = 0;
634 LIST_HEAD(uf);
635 VMA_ITERATOR(vmi, mm, 0);
636
637 uprobe_start_dup_mmap();
638 if (mmap_write_lock_killable(oldmm)) {
639 retval = -EINTR;
640 goto fail_uprobe_end;
641 }
642 flush_cache_dup_mm(oldmm);
643 uprobe_dup_mmap(oldmm, mm);
644 /*
645 * Not linked in yet - no deadlock potential:
646 */
647 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
648
649 /* No ordering required: file already has been exposed. */
650 dup_mm_exe_file(mm, oldmm);
651
652 mm->total_vm = oldmm->total_vm;
653 mm->data_vm = oldmm->data_vm;
654 mm->exec_vm = oldmm->exec_vm;
655 mm->stack_vm = oldmm->stack_vm;
656
657 retval = ksm_fork(mm, oldmm);
658 if (retval)
659 goto out;
660 khugepaged_fork(mm, oldmm);
661
662 /* Use __mt_dup() to efficiently build an identical maple tree. */
663 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
664 if (unlikely(retval))
665 goto out;
666
667 mt_clear_in_rcu(vmi.mas.tree);
668 for_each_vma(vmi, mpnt) {
669 struct file *file;
670
671 vma_start_write(mpnt);
672 if (mpnt->vm_flags & VM_DONTCOPY) {
673 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
674 mpnt->vm_end, GFP_KERNEL);
675 if (retval)
676 goto loop_out;
677
678 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
679 continue;
680 }
681 charge = 0;
682 /*
683 * Don't duplicate many vmas if we've been oom-killed (for
684 * example)
685 */
686 if (fatal_signal_pending(current)) {
687 retval = -EINTR;
688 goto loop_out;
689 }
690 if (mpnt->vm_flags & VM_ACCOUNT) {
691 unsigned long len = vma_pages(mpnt);
692
693 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
694 goto fail_nomem;
695 charge = len;
696 }
697 tmp = vm_area_dup(mpnt);
698 if (!tmp)
699 goto fail_nomem;
700 retval = vma_dup_policy(mpnt, tmp);
701 if (retval)
702 goto fail_nomem_policy;
703 tmp->vm_mm = mm;
704 retval = dup_userfaultfd(tmp, &uf);
705 if (retval)
706 goto fail_nomem_anon_vma_fork;
707 if (tmp->vm_flags & VM_WIPEONFORK) {
708 /*
709 * VM_WIPEONFORK gets a clean slate in the child.
710 * Don't prepare anon_vma until fault since we don't
711 * copy page for current vma.
712 */
713 tmp->anon_vma = NULL;
714 } else if (anon_vma_fork(tmp, mpnt))
715 goto fail_nomem_anon_vma_fork;
716 vm_flags_clear(tmp, VM_LOCKED_MASK);
717 /*
718 * Copy/update hugetlb private vma information.
719 */
720 if (is_vm_hugetlb_page(tmp))
721 hugetlb_dup_vma_private(tmp);
722
723 /*
724 * Link the vma into the MT. After using __mt_dup(), memory
725 * allocation is not necessary here, so it cannot fail.
726 */
727 vma_iter_bulk_store(&vmi, tmp);
728
729 mm->map_count++;
730
731 if (tmp->vm_ops && tmp->vm_ops->open)
732 tmp->vm_ops->open(tmp);
733
734 file = tmp->vm_file;
735 if (file) {
736 struct address_space *mapping = file->f_mapping;
737
738 get_file(file);
739 i_mmap_lock_write(mapping);
740 if (vma_is_shared_maywrite(tmp))
741 mapping_allow_writable(mapping);
742 flush_dcache_mmap_lock(mapping);
743 /* insert tmp into the share list, just after mpnt */
744 vma_interval_tree_insert_after(tmp, mpnt,
745 &mapping->i_mmap);
746 flush_dcache_mmap_unlock(mapping);
747 i_mmap_unlock_write(mapping);
748 }
749
750 if (!(tmp->vm_flags & VM_WIPEONFORK))
751 retval = copy_page_range(tmp, mpnt);
752
753 if (retval) {
754 mpnt = vma_next(&vmi);
755 goto loop_out;
756 }
757 }
758 /* a new mm has just been created */
759 retval = arch_dup_mmap(oldmm, mm);
760loop_out:
761 vma_iter_free(&vmi);
762 if (!retval) {
763 mt_set_in_rcu(vmi.mas.tree);
764 } else if (mpnt) {
765 /*
766 * The entire maple tree has already been duplicated. If the
767 * mmap duplication fails, mark the failure point with
768 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
769 * stop releasing VMAs that have not been duplicated after this
770 * point.
771 */
772 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
773 mas_store(&vmi.mas, XA_ZERO_ENTRY);
774 }
775out:
776 mmap_write_unlock(mm);
777 flush_tlb_mm(oldmm);
778 mmap_write_unlock(oldmm);
779 dup_userfaultfd_complete(&uf);
780fail_uprobe_end:
781 uprobe_end_dup_mmap();
782 return retval;
783
784fail_nomem_anon_vma_fork:
785 mpol_put(vma_policy(tmp));
786fail_nomem_policy:
787 vm_area_free(tmp);
788fail_nomem:
789 retval = -ENOMEM;
790 vm_unacct_memory(charge);
791 goto loop_out;
792}
793
794static inline int mm_alloc_pgd(struct mm_struct *mm)
795{
796 mm->pgd = pgd_alloc(mm);
797 if (unlikely(!mm->pgd))
798 return -ENOMEM;
799 return 0;
800}
801
802static inline void mm_free_pgd(struct mm_struct *mm)
803{
804 pgd_free(mm, mm->pgd);
805}
806#else
807static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
808{
809 mmap_write_lock(oldmm);
810 dup_mm_exe_file(mm, oldmm);
811 mmap_write_unlock(oldmm);
812 return 0;
813}
814#define mm_alloc_pgd(mm) (0)
815#define mm_free_pgd(mm)
816#endif /* CONFIG_MMU */
817
818static void check_mm(struct mm_struct *mm)
819{
820 int i;
821
822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
823 "Please make sure 'struct resident_page_types[]' is updated as well");
824
825 for (i = 0; i < NR_MM_COUNTERS; i++) {
826 long x = percpu_counter_sum(&mm->rss_stat[i]);
827
828 if (unlikely(x))
829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
830 mm, resident_page_types[i], x);
831 }
832
833 if (mm_pgtables_bytes(mm))
834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
835 mm_pgtables_bytes(mm));
836
837#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
839#endif
840}
841
842#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
843#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
844
845static void do_check_lazy_tlb(void *arg)
846{
847 struct mm_struct *mm = arg;
848
849 WARN_ON_ONCE(current->active_mm == mm);
850}
851
852static void do_shoot_lazy_tlb(void *arg)
853{
854 struct mm_struct *mm = arg;
855
856 if (current->active_mm == mm) {
857 WARN_ON_ONCE(current->mm);
858 current->active_mm = &init_mm;
859 switch_mm(mm, &init_mm, current);
860 }
861}
862
863static void cleanup_lazy_tlbs(struct mm_struct *mm)
864{
865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
866 /*
867 * In this case, lazy tlb mms are refounted and would not reach
868 * __mmdrop until all CPUs have switched away and mmdrop()ed.
869 */
870 return;
871 }
872
873 /*
874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
875 * requires lazy mm users to switch to another mm when the refcount
876 * drops to zero, before the mm is freed. This requires IPIs here to
877 * switch kernel threads to init_mm.
878 *
879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
880 * switch with the final userspace teardown TLB flush which leaves the
881 * mm lazy on this CPU but no others, reducing the need for additional
882 * IPIs here. There are cases where a final IPI is still required here,
883 * such as the final mmdrop being performed on a different CPU than the
884 * one exiting, or kernel threads using the mm when userspace exits.
885 *
886 * IPI overheads have not found to be expensive, but they could be
887 * reduced in a number of possible ways, for example (roughly
888 * increasing order of complexity):
889 * - The last lazy reference created by exit_mm() could instead switch
890 * to init_mm, however it's probable this will run on the same CPU
891 * immediately afterwards, so this may not reduce IPIs much.
892 * - A batch of mms requiring IPIs could be gathered and freed at once.
893 * - CPUs store active_mm where it can be remotely checked without a
894 * lock, to filter out false-positives in the cpumask.
895 * - After mm_users or mm_count reaches zero, switching away from the
896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
897 * with some batching or delaying of the final IPIs.
898 * - A delayed freeing and RCU-like quiescing sequence based on mm
899 * switching to avoid IPIs completely.
900 */
901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
904}
905
906/*
907 * Called when the last reference to the mm
908 * is dropped: either by a lazy thread or by
909 * mmput. Free the page directory and the mm.
910 */
911void __mmdrop(struct mm_struct *mm)
912{
913 BUG_ON(mm == &init_mm);
914 WARN_ON_ONCE(mm == current->mm);
915
916 /* Ensure no CPUs are using this as their lazy tlb mm */
917 cleanup_lazy_tlbs(mm);
918
919 WARN_ON_ONCE(mm == current->active_mm);
920 mm_free_pgd(mm);
921 destroy_context(mm);
922 mmu_notifier_subscriptions_destroy(mm);
923 check_mm(mm);
924 put_user_ns(mm->user_ns);
925 mm_pasid_drop(mm);
926 mm_destroy_cid(mm);
927 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
928
929 free_mm(mm);
930}
931EXPORT_SYMBOL_GPL(__mmdrop);
932
933static void mmdrop_async_fn(struct work_struct *work)
934{
935 struct mm_struct *mm;
936
937 mm = container_of(work, struct mm_struct, async_put_work);
938 __mmdrop(mm);
939}
940
941static void mmdrop_async(struct mm_struct *mm)
942{
943 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
944 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
945 schedule_work(&mm->async_put_work);
946 }
947}
948
949static inline void free_signal_struct(struct signal_struct *sig)
950{
951 taskstats_tgid_free(sig);
952 sched_autogroup_exit(sig);
953 /*
954 * __mmdrop is not safe to call from softirq context on x86 due to
955 * pgd_dtor so postpone it to the async context
956 */
957 if (sig->oom_mm)
958 mmdrop_async(sig->oom_mm);
959 kmem_cache_free(signal_cachep, sig);
960}
961
962static inline void put_signal_struct(struct signal_struct *sig)
963{
964 if (refcount_dec_and_test(&sig->sigcnt))
965 free_signal_struct(sig);
966}
967
968void __put_task_struct(struct task_struct *tsk)
969{
970 WARN_ON(!tsk->exit_state);
971 WARN_ON(refcount_read(&tsk->usage));
972 WARN_ON(tsk == current);
973
974 io_uring_free(tsk);
975 cgroup_free(tsk);
976 task_numa_free(tsk, true);
977 security_task_free(tsk);
978 exit_creds(tsk);
979 delayacct_tsk_free(tsk);
980 put_signal_struct(tsk->signal);
981 sched_core_free(tsk);
982 free_task(tsk);
983}
984EXPORT_SYMBOL_GPL(__put_task_struct);
985
986void __put_task_struct_rcu_cb(struct rcu_head *rhp)
987{
988 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
989
990 __put_task_struct(task);
991}
992EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
993
994void __init __weak arch_task_cache_init(void) { }
995
996/*
997 * set_max_threads
998 */
999static void set_max_threads(unsigned int max_threads_suggested)
1000{
1001 u64 threads;
1002 unsigned long nr_pages = totalram_pages();
1003
1004 /*
1005 * The number of threads shall be limited such that the thread
1006 * structures may only consume a small part of the available memory.
1007 */
1008 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1009 threads = MAX_THREADS;
1010 else
1011 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1012 (u64) THREAD_SIZE * 8UL);
1013
1014 if (threads > max_threads_suggested)
1015 threads = max_threads_suggested;
1016
1017 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1018}
1019
1020#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1021/* Initialized by the architecture: */
1022int arch_task_struct_size __read_mostly;
1023#endif
1024
1025static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1026{
1027 /* Fetch thread_struct whitelist for the architecture. */
1028 arch_thread_struct_whitelist(offset, size);
1029
1030 /*
1031 * Handle zero-sized whitelist or empty thread_struct, otherwise
1032 * adjust offset to position of thread_struct in task_struct.
1033 */
1034 if (unlikely(*size == 0))
1035 *offset = 0;
1036 else
1037 *offset += offsetof(struct task_struct, thread);
1038}
1039
1040void __init fork_init(void)
1041{
1042 int i;
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
1056 /* do the arch specific task caches init */
1057 arch_task_cache_init();
1058
1059 set_max_threads(MAX_THREADS);
1060
1061 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1064 init_task.signal->rlim[RLIMIT_NPROC];
1065
1066 for (i = 0; i < UCOUNT_COUNTS; i++)
1067 init_user_ns.ucount_max[i] = max_threads/2;
1068
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1073
1074#ifdef CONFIG_VMAP_STACK
1075 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1076 NULL, free_vm_stack_cache);
1077#endif
1078
1079 scs_init();
1080
1081 lockdep_init_task(&init_task);
1082 uprobes_init();
1083}
1084
1085int __weak arch_dup_task_struct(struct task_struct *dst,
1086 struct task_struct *src)
1087{
1088 *dst = *src;
1089 return 0;
1090}
1091
1092void set_task_stack_end_magic(struct task_struct *tsk)
1093{
1094 unsigned long *stackend;
1095
1096 stackend = end_of_stack(tsk);
1097 *stackend = STACK_END_MAGIC; /* for overflow detection */
1098}
1099
1100static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1101{
1102 struct task_struct *tsk;
1103 int err;
1104
1105 if (node == NUMA_NO_NODE)
1106 node = tsk_fork_get_node(orig);
1107 tsk = alloc_task_struct_node(node);
1108 if (!tsk)
1109 return NULL;
1110
1111 err = arch_dup_task_struct(tsk, orig);
1112 if (err)
1113 goto free_tsk;
1114
1115 err = alloc_thread_stack_node(tsk, node);
1116 if (err)
1117 goto free_tsk;
1118
1119#ifdef CONFIG_THREAD_INFO_IN_TASK
1120 refcount_set(&tsk->stack_refcount, 1);
1121#endif
1122 account_kernel_stack(tsk, 1);
1123
1124 err = scs_prepare(tsk, node);
1125 if (err)
1126 goto free_stack;
1127
1128#ifdef CONFIG_SECCOMP
1129 /*
1130 * We must handle setting up seccomp filters once we're under
1131 * the sighand lock in case orig has changed between now and
1132 * then. Until then, filter must be NULL to avoid messing up
1133 * the usage counts on the error path calling free_task.
1134 */
1135 tsk->seccomp.filter = NULL;
1136#endif
1137
1138 setup_thread_stack(tsk, orig);
1139 clear_user_return_notifier(tsk);
1140 clear_tsk_need_resched(tsk);
1141 set_task_stack_end_magic(tsk);
1142 clear_syscall_work_syscall_user_dispatch(tsk);
1143
1144#ifdef CONFIG_STACKPROTECTOR
1145 tsk->stack_canary = get_random_canary();
1146#endif
1147 if (orig->cpus_ptr == &orig->cpus_mask)
1148 tsk->cpus_ptr = &tsk->cpus_mask;
1149 dup_user_cpus_ptr(tsk, orig, node);
1150
1151 /*
1152 * One for the user space visible state that goes away when reaped.
1153 * One for the scheduler.
1154 */
1155 refcount_set(&tsk->rcu_users, 2);
1156 /* One for the rcu users */
1157 refcount_set(&tsk->usage, 1);
1158#ifdef CONFIG_BLK_DEV_IO_TRACE
1159 tsk->btrace_seq = 0;
1160#endif
1161 tsk->splice_pipe = NULL;
1162 tsk->task_frag.page = NULL;
1163 tsk->wake_q.next = NULL;
1164 tsk->worker_private = NULL;
1165
1166 kcov_task_init(tsk);
1167 kmsan_task_create(tsk);
1168 kmap_local_fork(tsk);
1169
1170#ifdef CONFIG_FAULT_INJECTION
1171 tsk->fail_nth = 0;
1172#endif
1173
1174#ifdef CONFIG_BLK_CGROUP
1175 tsk->throttle_disk = NULL;
1176 tsk->use_memdelay = 0;
1177#endif
1178
1179#ifdef CONFIG_ARCH_HAS_CPU_PASID
1180 tsk->pasid_activated = 0;
1181#endif
1182
1183#ifdef CONFIG_MEMCG
1184 tsk->active_memcg = NULL;
1185#endif
1186
1187#ifdef CONFIG_CPU_SUP_INTEL
1188 tsk->reported_split_lock = 0;
1189#endif
1190
1191#ifdef CONFIG_SCHED_MM_CID
1192 tsk->mm_cid = -1;
1193 tsk->last_mm_cid = -1;
1194 tsk->mm_cid_active = 0;
1195 tsk->migrate_from_cpu = -1;
1196#endif
1197 return tsk;
1198
1199free_stack:
1200 exit_task_stack_account(tsk);
1201 free_thread_stack(tsk);
1202free_tsk:
1203 free_task_struct(tsk);
1204 return NULL;
1205}
1206
1207__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1208
1209static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1210
1211static int __init coredump_filter_setup(char *s)
1212{
1213 default_dump_filter =
1214 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1215 MMF_DUMP_FILTER_MASK;
1216 return 1;
1217}
1218
1219__setup("coredump_filter=", coredump_filter_setup);
1220
1221#include <linux/init_task.h>
1222
1223static void mm_init_aio(struct mm_struct *mm)
1224{
1225#ifdef CONFIG_AIO
1226 spin_lock_init(&mm->ioctx_lock);
1227 mm->ioctx_table = NULL;
1228#endif
1229}
1230
1231static __always_inline void mm_clear_owner(struct mm_struct *mm,
1232 struct task_struct *p)
1233{
1234#ifdef CONFIG_MEMCG
1235 if (mm->owner == p)
1236 WRITE_ONCE(mm->owner, NULL);
1237#endif
1238}
1239
1240static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1241{
1242#ifdef CONFIG_MEMCG
1243 mm->owner = p;
1244#endif
1245}
1246
1247static void mm_init_uprobes_state(struct mm_struct *mm)
1248{
1249#ifdef CONFIG_UPROBES
1250 mm->uprobes_state.xol_area = NULL;
1251#endif
1252}
1253
1254static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1255 struct user_namespace *user_ns)
1256{
1257 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1258 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1259 atomic_set(&mm->mm_users, 1);
1260 atomic_set(&mm->mm_count, 1);
1261 seqcount_init(&mm->write_protect_seq);
1262 mmap_init_lock(mm);
1263 INIT_LIST_HEAD(&mm->mmlist);
1264#ifdef CONFIG_PER_VMA_LOCK
1265 mm->mm_lock_seq = 0;
1266#endif
1267 mm_pgtables_bytes_init(mm);
1268 mm->map_count = 0;
1269 mm->locked_vm = 0;
1270 atomic64_set(&mm->pinned_vm, 0);
1271 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1272 spin_lock_init(&mm->page_table_lock);
1273 spin_lock_init(&mm->arg_lock);
1274 mm_init_cpumask(mm);
1275 mm_init_aio(mm);
1276 mm_init_owner(mm, p);
1277 mm_pasid_init(mm);
1278 RCU_INIT_POINTER(mm->exe_file, NULL);
1279 mmu_notifier_subscriptions_init(mm);
1280 init_tlb_flush_pending(mm);
1281#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1282 mm->pmd_huge_pte = NULL;
1283#endif
1284 mm_init_uprobes_state(mm);
1285 hugetlb_count_init(mm);
1286
1287 if (current->mm) {
1288 mm->flags = mmf_init_flags(current->mm->flags);
1289 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1290 } else {
1291 mm->flags = default_dump_filter;
1292 mm->def_flags = 0;
1293 }
1294
1295 if (mm_alloc_pgd(mm))
1296 goto fail_nopgd;
1297
1298 if (init_new_context(p, mm))
1299 goto fail_nocontext;
1300
1301 if (mm_alloc_cid(mm))
1302 goto fail_cid;
1303
1304 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1305 NR_MM_COUNTERS))
1306 goto fail_pcpu;
1307
1308 mm->user_ns = get_user_ns(user_ns);
1309 lru_gen_init_mm(mm);
1310 return mm;
1311
1312fail_pcpu:
1313 mm_destroy_cid(mm);
1314fail_cid:
1315 destroy_context(mm);
1316fail_nocontext:
1317 mm_free_pgd(mm);
1318fail_nopgd:
1319 free_mm(mm);
1320 return NULL;
1321}
1322
1323/*
1324 * Allocate and initialize an mm_struct.
1325 */
1326struct mm_struct *mm_alloc(void)
1327{
1328 struct mm_struct *mm;
1329
1330 mm = allocate_mm();
1331 if (!mm)
1332 return NULL;
1333
1334 memset(mm, 0, sizeof(*mm));
1335 return mm_init(mm, current, current_user_ns());
1336}
1337
1338static inline void __mmput(struct mm_struct *mm)
1339{
1340 VM_BUG_ON(atomic_read(&mm->mm_users));
1341
1342 uprobe_clear_state(mm);
1343 exit_aio(mm);
1344 ksm_exit(mm);
1345 khugepaged_exit(mm); /* must run before exit_mmap */
1346 exit_mmap(mm);
1347 mm_put_huge_zero_page(mm);
1348 set_mm_exe_file(mm, NULL);
1349 if (!list_empty(&mm->mmlist)) {
1350 spin_lock(&mmlist_lock);
1351 list_del(&mm->mmlist);
1352 spin_unlock(&mmlist_lock);
1353 }
1354 if (mm->binfmt)
1355 module_put(mm->binfmt->module);
1356 lru_gen_del_mm(mm);
1357 mmdrop(mm);
1358}
1359
1360/*
1361 * Decrement the use count and release all resources for an mm.
1362 */
1363void mmput(struct mm_struct *mm)
1364{
1365 might_sleep();
1366
1367 if (atomic_dec_and_test(&mm->mm_users))
1368 __mmput(mm);
1369}
1370EXPORT_SYMBOL_GPL(mmput);
1371
1372#ifdef CONFIG_MMU
1373static void mmput_async_fn(struct work_struct *work)
1374{
1375 struct mm_struct *mm = container_of(work, struct mm_struct,
1376 async_put_work);
1377
1378 __mmput(mm);
1379}
1380
1381void mmput_async(struct mm_struct *mm)
1382{
1383 if (atomic_dec_and_test(&mm->mm_users)) {
1384 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1385 schedule_work(&mm->async_put_work);
1386 }
1387}
1388EXPORT_SYMBOL_GPL(mmput_async);
1389#endif
1390
1391/**
1392 * set_mm_exe_file - change a reference to the mm's executable file
1393 * @mm: The mm to change.
1394 * @new_exe_file: The new file to use.
1395 *
1396 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1397 *
1398 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1399 * invocations: in mmput() nobody alive left, in execve it happens before
1400 * the new mm is made visible to anyone.
1401 *
1402 * Can only fail if new_exe_file != NULL.
1403 */
1404int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1405{
1406 struct file *old_exe_file;
1407
1408 /*
1409 * It is safe to dereference the exe_file without RCU as
1410 * this function is only called if nobody else can access
1411 * this mm -- see comment above for justification.
1412 */
1413 old_exe_file = rcu_dereference_raw(mm->exe_file);
1414
1415 if (new_exe_file) {
1416 /*
1417 * We expect the caller (i.e., sys_execve) to already denied
1418 * write access, so this is unlikely to fail.
1419 */
1420 if (unlikely(deny_write_access(new_exe_file)))
1421 return -EACCES;
1422 get_file(new_exe_file);
1423 }
1424 rcu_assign_pointer(mm->exe_file, new_exe_file);
1425 if (old_exe_file) {
1426 allow_write_access(old_exe_file);
1427 fput(old_exe_file);
1428 }
1429 return 0;
1430}
1431
1432/**
1433 * replace_mm_exe_file - replace a reference to the mm's executable file
1434 * @mm: The mm to change.
1435 * @new_exe_file: The new file to use.
1436 *
1437 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1438 *
1439 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1440 */
1441int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1442{
1443 struct vm_area_struct *vma;
1444 struct file *old_exe_file;
1445 int ret = 0;
1446
1447 /* Forbid mm->exe_file change if old file still mapped. */
1448 old_exe_file = get_mm_exe_file(mm);
1449 if (old_exe_file) {
1450 VMA_ITERATOR(vmi, mm, 0);
1451 mmap_read_lock(mm);
1452 for_each_vma(vmi, vma) {
1453 if (!vma->vm_file)
1454 continue;
1455 if (path_equal(&vma->vm_file->f_path,
1456 &old_exe_file->f_path)) {
1457 ret = -EBUSY;
1458 break;
1459 }
1460 }
1461 mmap_read_unlock(mm);
1462 fput(old_exe_file);
1463 if (ret)
1464 return ret;
1465 }
1466
1467 ret = deny_write_access(new_exe_file);
1468 if (ret)
1469 return -EACCES;
1470 get_file(new_exe_file);
1471
1472 /* set the new file */
1473 mmap_write_lock(mm);
1474 old_exe_file = rcu_dereference_raw(mm->exe_file);
1475 rcu_assign_pointer(mm->exe_file, new_exe_file);
1476 mmap_write_unlock(mm);
1477
1478 if (old_exe_file) {
1479 allow_write_access(old_exe_file);
1480 fput(old_exe_file);
1481 }
1482 return 0;
1483}
1484
1485/**
1486 * get_mm_exe_file - acquire a reference to the mm's executable file
1487 * @mm: The mm of interest.
1488 *
1489 * Returns %NULL if mm has no associated executable file.
1490 * User must release file via fput().
1491 */
1492struct file *get_mm_exe_file(struct mm_struct *mm)
1493{
1494 struct file *exe_file;
1495
1496 rcu_read_lock();
1497 exe_file = get_file_rcu(&mm->exe_file);
1498 rcu_read_unlock();
1499 return exe_file;
1500}
1501
1502/**
1503 * get_task_exe_file - acquire a reference to the task's executable file
1504 * @task: The task.
1505 *
1506 * Returns %NULL if task's mm (if any) has no associated executable file or
1507 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1508 * User must release file via fput().
1509 */
1510struct file *get_task_exe_file(struct task_struct *task)
1511{
1512 struct file *exe_file = NULL;
1513 struct mm_struct *mm;
1514
1515 task_lock(task);
1516 mm = task->mm;
1517 if (mm) {
1518 if (!(task->flags & PF_KTHREAD))
1519 exe_file = get_mm_exe_file(mm);
1520 }
1521 task_unlock(task);
1522 return exe_file;
1523}
1524
1525/**
1526 * get_task_mm - acquire a reference to the task's mm
1527 * @task: The task.
1528 *
1529 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1530 * this kernel workthread has transiently adopted a user mm with use_mm,
1531 * to do its AIO) is not set and if so returns a reference to it, after
1532 * bumping up the use count. User must release the mm via mmput()
1533 * after use. Typically used by /proc and ptrace.
1534 */
1535struct mm_struct *get_task_mm(struct task_struct *task)
1536{
1537 struct mm_struct *mm;
1538
1539 task_lock(task);
1540 mm = task->mm;
1541 if (mm) {
1542 if (task->flags & PF_KTHREAD)
1543 mm = NULL;
1544 else
1545 mmget(mm);
1546 }
1547 task_unlock(task);
1548 return mm;
1549}
1550EXPORT_SYMBOL_GPL(get_task_mm);
1551
1552struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1553{
1554 struct mm_struct *mm;
1555 int err;
1556
1557 err = down_read_killable(&task->signal->exec_update_lock);
1558 if (err)
1559 return ERR_PTR(err);
1560
1561 mm = get_task_mm(task);
1562 if (mm && mm != current->mm &&
1563 !ptrace_may_access(task, mode)) {
1564 mmput(mm);
1565 mm = ERR_PTR(-EACCES);
1566 }
1567 up_read(&task->signal->exec_update_lock);
1568
1569 return mm;
1570}
1571
1572static void complete_vfork_done(struct task_struct *tsk)
1573{
1574 struct completion *vfork;
1575
1576 task_lock(tsk);
1577 vfork = tsk->vfork_done;
1578 if (likely(vfork)) {
1579 tsk->vfork_done = NULL;
1580 complete(vfork);
1581 }
1582 task_unlock(tsk);
1583}
1584
1585static int wait_for_vfork_done(struct task_struct *child,
1586 struct completion *vfork)
1587{
1588 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1589 int killed;
1590
1591 cgroup_enter_frozen();
1592 killed = wait_for_completion_state(vfork, state);
1593 cgroup_leave_frozen(false);
1594
1595 if (killed) {
1596 task_lock(child);
1597 child->vfork_done = NULL;
1598 task_unlock(child);
1599 }
1600
1601 put_task_struct(child);
1602 return killed;
1603}
1604
1605/* Please note the differences between mmput and mm_release.
1606 * mmput is called whenever we stop holding onto a mm_struct,
1607 * error success whatever.
1608 *
1609 * mm_release is called after a mm_struct has been removed
1610 * from the current process.
1611 *
1612 * This difference is important for error handling, when we
1613 * only half set up a mm_struct for a new process and need to restore
1614 * the old one. Because we mmput the new mm_struct before
1615 * restoring the old one. . .
1616 * Eric Biederman 10 January 1998
1617 */
1618static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1619{
1620 uprobe_free_utask(tsk);
1621
1622 /* Get rid of any cached register state */
1623 deactivate_mm(tsk, mm);
1624
1625 /*
1626 * Signal userspace if we're not exiting with a core dump
1627 * because we want to leave the value intact for debugging
1628 * purposes.
1629 */
1630 if (tsk->clear_child_tid) {
1631 if (atomic_read(&mm->mm_users) > 1) {
1632 /*
1633 * We don't check the error code - if userspace has
1634 * not set up a proper pointer then tough luck.
1635 */
1636 put_user(0, tsk->clear_child_tid);
1637 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1638 1, NULL, NULL, 0, 0);
1639 }
1640 tsk->clear_child_tid = NULL;
1641 }
1642
1643 /*
1644 * All done, finally we can wake up parent and return this mm to him.
1645 * Also kthread_stop() uses this completion for synchronization.
1646 */
1647 if (tsk->vfork_done)
1648 complete_vfork_done(tsk);
1649}
1650
1651void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1652{
1653 futex_exit_release(tsk);
1654 mm_release(tsk, mm);
1655}
1656
1657void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1658{
1659 futex_exec_release(tsk);
1660 mm_release(tsk, mm);
1661}
1662
1663/**
1664 * dup_mm() - duplicates an existing mm structure
1665 * @tsk: the task_struct with which the new mm will be associated.
1666 * @oldmm: the mm to duplicate.
1667 *
1668 * Allocates a new mm structure and duplicates the provided @oldmm structure
1669 * content into it.
1670 *
1671 * Return: the duplicated mm or NULL on failure.
1672 */
1673static struct mm_struct *dup_mm(struct task_struct *tsk,
1674 struct mm_struct *oldmm)
1675{
1676 struct mm_struct *mm;
1677 int err;
1678
1679 mm = allocate_mm();
1680 if (!mm)
1681 goto fail_nomem;
1682
1683 memcpy(mm, oldmm, sizeof(*mm));
1684
1685 if (!mm_init(mm, tsk, mm->user_ns))
1686 goto fail_nomem;
1687
1688 err = dup_mmap(mm, oldmm);
1689 if (err)
1690 goto free_pt;
1691
1692 mm->hiwater_rss = get_mm_rss(mm);
1693 mm->hiwater_vm = mm->total_vm;
1694
1695 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1696 goto free_pt;
1697
1698 return mm;
1699
1700free_pt:
1701 /* don't put binfmt in mmput, we haven't got module yet */
1702 mm->binfmt = NULL;
1703 mm_init_owner(mm, NULL);
1704 mmput(mm);
1705
1706fail_nomem:
1707 return NULL;
1708}
1709
1710static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1711{
1712 struct mm_struct *mm, *oldmm;
1713
1714 tsk->min_flt = tsk->maj_flt = 0;
1715 tsk->nvcsw = tsk->nivcsw = 0;
1716#ifdef CONFIG_DETECT_HUNG_TASK
1717 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1718 tsk->last_switch_time = 0;
1719#endif
1720
1721 tsk->mm = NULL;
1722 tsk->active_mm = NULL;
1723
1724 /*
1725 * Are we cloning a kernel thread?
1726 *
1727 * We need to steal a active VM for that..
1728 */
1729 oldmm = current->mm;
1730 if (!oldmm)
1731 return 0;
1732
1733 if (clone_flags & CLONE_VM) {
1734 mmget(oldmm);
1735 mm = oldmm;
1736 } else {
1737 mm = dup_mm(tsk, current->mm);
1738 if (!mm)
1739 return -ENOMEM;
1740 }
1741
1742 tsk->mm = mm;
1743 tsk->active_mm = mm;
1744 sched_mm_cid_fork(tsk);
1745 return 0;
1746}
1747
1748static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1749{
1750 struct fs_struct *fs = current->fs;
1751 if (clone_flags & CLONE_FS) {
1752 /* tsk->fs is already what we want */
1753 spin_lock(&fs->lock);
1754 /* "users" and "in_exec" locked for check_unsafe_exec() */
1755 if (fs->in_exec) {
1756 spin_unlock(&fs->lock);
1757 return -EAGAIN;
1758 }
1759 fs->users++;
1760 spin_unlock(&fs->lock);
1761 return 0;
1762 }
1763 tsk->fs = copy_fs_struct(fs);
1764 if (!tsk->fs)
1765 return -ENOMEM;
1766 return 0;
1767}
1768
1769static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1770 int no_files)
1771{
1772 struct files_struct *oldf, *newf;
1773 int error = 0;
1774
1775 /*
1776 * A background process may not have any files ...
1777 */
1778 oldf = current->files;
1779 if (!oldf)
1780 goto out;
1781
1782 if (no_files) {
1783 tsk->files = NULL;
1784 goto out;
1785 }
1786
1787 if (clone_flags & CLONE_FILES) {
1788 atomic_inc(&oldf->count);
1789 goto out;
1790 }
1791
1792 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1793 if (!newf)
1794 goto out;
1795
1796 tsk->files = newf;
1797 error = 0;
1798out:
1799 return error;
1800}
1801
1802static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1803{
1804 struct sighand_struct *sig;
1805
1806 if (clone_flags & CLONE_SIGHAND) {
1807 refcount_inc(¤t->sighand->count);
1808 return 0;
1809 }
1810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1811 RCU_INIT_POINTER(tsk->sighand, sig);
1812 if (!sig)
1813 return -ENOMEM;
1814
1815 refcount_set(&sig->count, 1);
1816 spin_lock_irq(¤t->sighand->siglock);
1817 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1818 spin_unlock_irq(¤t->sighand->siglock);
1819
1820 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1821 if (clone_flags & CLONE_CLEAR_SIGHAND)
1822 flush_signal_handlers(tsk, 0);
1823
1824 return 0;
1825}
1826
1827void __cleanup_sighand(struct sighand_struct *sighand)
1828{
1829 if (refcount_dec_and_test(&sighand->count)) {
1830 signalfd_cleanup(sighand);
1831 /*
1832 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1833 * without an RCU grace period, see __lock_task_sighand().
1834 */
1835 kmem_cache_free(sighand_cachep, sighand);
1836 }
1837}
1838
1839/*
1840 * Initialize POSIX timer handling for a thread group.
1841 */
1842static void posix_cpu_timers_init_group(struct signal_struct *sig)
1843{
1844 struct posix_cputimers *pct = &sig->posix_cputimers;
1845 unsigned long cpu_limit;
1846
1847 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1848 posix_cputimers_group_init(pct, cpu_limit);
1849}
1850
1851static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1852{
1853 struct signal_struct *sig;
1854
1855 if (clone_flags & CLONE_THREAD)
1856 return 0;
1857
1858 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1859 tsk->signal = sig;
1860 if (!sig)
1861 return -ENOMEM;
1862
1863 sig->nr_threads = 1;
1864 sig->quick_threads = 1;
1865 atomic_set(&sig->live, 1);
1866 refcount_set(&sig->sigcnt, 1);
1867
1868 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1869 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1870 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1871
1872 init_waitqueue_head(&sig->wait_chldexit);
1873 sig->curr_target = tsk;
1874 init_sigpending(&sig->shared_pending);
1875 INIT_HLIST_HEAD(&sig->multiprocess);
1876 seqlock_init(&sig->stats_lock);
1877 prev_cputime_init(&sig->prev_cputime);
1878
1879#ifdef CONFIG_POSIX_TIMERS
1880 INIT_LIST_HEAD(&sig->posix_timers);
1881 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1882 sig->real_timer.function = it_real_fn;
1883#endif
1884
1885 task_lock(current->group_leader);
1886 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1887 task_unlock(current->group_leader);
1888
1889 posix_cpu_timers_init_group(sig);
1890
1891 tty_audit_fork(sig);
1892 sched_autogroup_fork(sig);
1893
1894 sig->oom_score_adj = current->signal->oom_score_adj;
1895 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1896
1897 mutex_init(&sig->cred_guard_mutex);
1898 init_rwsem(&sig->exec_update_lock);
1899
1900 return 0;
1901}
1902
1903static void copy_seccomp(struct task_struct *p)
1904{
1905#ifdef CONFIG_SECCOMP
1906 /*
1907 * Must be called with sighand->lock held, which is common to
1908 * all threads in the group. Holding cred_guard_mutex is not
1909 * needed because this new task is not yet running and cannot
1910 * be racing exec.
1911 */
1912 assert_spin_locked(¤t->sighand->siglock);
1913
1914 /* Ref-count the new filter user, and assign it. */
1915 get_seccomp_filter(current);
1916 p->seccomp = current->seccomp;
1917
1918 /*
1919 * Explicitly enable no_new_privs here in case it got set
1920 * between the task_struct being duplicated and holding the
1921 * sighand lock. The seccomp state and nnp must be in sync.
1922 */
1923 if (task_no_new_privs(current))
1924 task_set_no_new_privs(p);
1925
1926 /*
1927 * If the parent gained a seccomp mode after copying thread
1928 * flags and between before we held the sighand lock, we have
1929 * to manually enable the seccomp thread flag here.
1930 */
1931 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1932 set_task_syscall_work(p, SECCOMP);
1933#endif
1934}
1935
1936SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1937{
1938 current->clear_child_tid = tidptr;
1939
1940 return task_pid_vnr(current);
1941}
1942
1943static void rt_mutex_init_task(struct task_struct *p)
1944{
1945 raw_spin_lock_init(&p->pi_lock);
1946#ifdef CONFIG_RT_MUTEXES
1947 p->pi_waiters = RB_ROOT_CACHED;
1948 p->pi_top_task = NULL;
1949 p->pi_blocked_on = NULL;
1950#endif
1951}
1952
1953static inline void init_task_pid_links(struct task_struct *task)
1954{
1955 enum pid_type type;
1956
1957 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1958 INIT_HLIST_NODE(&task->pid_links[type]);
1959}
1960
1961static inline void
1962init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1963{
1964 if (type == PIDTYPE_PID)
1965 task->thread_pid = pid;
1966 else
1967 task->signal->pids[type] = pid;
1968}
1969
1970static inline void rcu_copy_process(struct task_struct *p)
1971{
1972#ifdef CONFIG_PREEMPT_RCU
1973 p->rcu_read_lock_nesting = 0;
1974 p->rcu_read_unlock_special.s = 0;
1975 p->rcu_blocked_node = NULL;
1976 INIT_LIST_HEAD(&p->rcu_node_entry);
1977#endif /* #ifdef CONFIG_PREEMPT_RCU */
1978#ifdef CONFIG_TASKS_RCU
1979 p->rcu_tasks_holdout = false;
1980 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1981 p->rcu_tasks_idle_cpu = -1;
1982 INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1983#endif /* #ifdef CONFIG_TASKS_RCU */
1984#ifdef CONFIG_TASKS_TRACE_RCU
1985 p->trc_reader_nesting = 0;
1986 p->trc_reader_special.s = 0;
1987 INIT_LIST_HEAD(&p->trc_holdout_list);
1988 INIT_LIST_HEAD(&p->trc_blkd_node);
1989#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1990}
1991
1992/**
1993 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1994 * @pid: the struct pid for which to create a pidfd
1995 * @flags: flags of the new @pidfd
1996 * @ret: Where to return the file for the pidfd.
1997 *
1998 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1999 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2000 *
2001 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2002 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2003 * pidfd file are prepared.
2004 *
2005 * If this function returns successfully the caller is responsible to either
2006 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2007 * order to install the pidfd into its file descriptor table or they must use
2008 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2009 * respectively.
2010 *
2011 * This function is useful when a pidfd must already be reserved but there
2012 * might still be points of failure afterwards and the caller wants to ensure
2013 * that no pidfd is leaked into its file descriptor table.
2014 *
2015 * Return: On success, a reserved pidfd is returned from the function and a new
2016 * pidfd file is returned in the last argument to the function. On
2017 * error, a negative error code is returned from the function and the
2018 * last argument remains unchanged.
2019 */
2020static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2021{
2022 int pidfd;
2023 struct file *pidfd_file;
2024
2025 pidfd = get_unused_fd_flags(O_CLOEXEC);
2026 if (pidfd < 0)
2027 return pidfd;
2028
2029 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2030 if (IS_ERR(pidfd_file)) {
2031 put_unused_fd(pidfd);
2032 return PTR_ERR(pidfd_file);
2033 }
2034 /*
2035 * anon_inode_getfile() ignores everything outside of the
2036 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2037 */
2038 pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2039 *ret = pidfd_file;
2040 return pidfd;
2041}
2042
2043/**
2044 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2045 * @pid: the struct pid for which to create a pidfd
2046 * @flags: flags of the new @pidfd
2047 * @ret: Where to return the pidfd.
2048 *
2049 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2050 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2051 *
2052 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2053 * task identified by @pid must be a thread-group leader.
2054 *
2055 * If this function returns successfully the caller is responsible to either
2056 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2057 * order to install the pidfd into its file descriptor table or they must use
2058 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2059 * respectively.
2060 *
2061 * This function is useful when a pidfd must already be reserved but there
2062 * might still be points of failure afterwards and the caller wants to ensure
2063 * that no pidfd is leaked into its file descriptor table.
2064 *
2065 * Return: On success, a reserved pidfd is returned from the function and a new
2066 * pidfd file is returned in the last argument to the function. On
2067 * error, a negative error code is returned from the function and the
2068 * last argument remains unchanged.
2069 */
2070int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2071{
2072 bool thread = flags & PIDFD_THREAD;
2073
2074 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2075 return -EINVAL;
2076
2077 return __pidfd_prepare(pid, flags, ret);
2078}
2079
2080static void __delayed_free_task(struct rcu_head *rhp)
2081{
2082 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2083
2084 free_task(tsk);
2085}
2086
2087static __always_inline void delayed_free_task(struct task_struct *tsk)
2088{
2089 if (IS_ENABLED(CONFIG_MEMCG))
2090 call_rcu(&tsk->rcu, __delayed_free_task);
2091 else
2092 free_task(tsk);
2093}
2094
2095static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2096{
2097 /* Skip if kernel thread */
2098 if (!tsk->mm)
2099 return;
2100
2101 /* Skip if spawning a thread or using vfork */
2102 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2103 return;
2104
2105 /* We need to synchronize with __set_oom_adj */
2106 mutex_lock(&oom_adj_mutex);
2107 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2108 /* Update the values in case they were changed after copy_signal */
2109 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2110 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2111 mutex_unlock(&oom_adj_mutex);
2112}
2113
2114#ifdef CONFIG_RV
2115static void rv_task_fork(struct task_struct *p)
2116{
2117 int i;
2118
2119 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2120 p->rv[i].da_mon.monitoring = false;
2121}
2122#else
2123#define rv_task_fork(p) do {} while (0)
2124#endif
2125
2126/*
2127 * This creates a new process as a copy of the old one,
2128 * but does not actually start it yet.
2129 *
2130 * It copies the registers, and all the appropriate
2131 * parts of the process environment (as per the clone
2132 * flags). The actual kick-off is left to the caller.
2133 */
2134__latent_entropy struct task_struct *copy_process(
2135 struct pid *pid,
2136 int trace,
2137 int node,
2138 struct kernel_clone_args *args)
2139{
2140 int pidfd = -1, retval;
2141 struct task_struct *p;
2142 struct multiprocess_signals delayed;
2143 struct file *pidfile = NULL;
2144 const u64 clone_flags = args->flags;
2145 struct nsproxy *nsp = current->nsproxy;
2146
2147 /*
2148 * Don't allow sharing the root directory with processes in a different
2149 * namespace
2150 */
2151 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2152 return ERR_PTR(-EINVAL);
2153
2154 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2155 return ERR_PTR(-EINVAL);
2156
2157 /*
2158 * Thread groups must share signals as well, and detached threads
2159 * can only be started up within the thread group.
2160 */
2161 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2162 return ERR_PTR(-EINVAL);
2163
2164 /*
2165 * Shared signal handlers imply shared VM. By way of the above,
2166 * thread groups also imply shared VM. Blocking this case allows
2167 * for various simplifications in other code.
2168 */
2169 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2170 return ERR_PTR(-EINVAL);
2171
2172 /*
2173 * Siblings of global init remain as zombies on exit since they are
2174 * not reaped by their parent (swapper). To solve this and to avoid
2175 * multi-rooted process trees, prevent global and container-inits
2176 * from creating siblings.
2177 */
2178 if ((clone_flags & CLONE_PARENT) &&
2179 current->signal->flags & SIGNAL_UNKILLABLE)
2180 return ERR_PTR(-EINVAL);
2181
2182 /*
2183 * If the new process will be in a different pid or user namespace
2184 * do not allow it to share a thread group with the forking task.
2185 */
2186 if (clone_flags & CLONE_THREAD) {
2187 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2188 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2189 return ERR_PTR(-EINVAL);
2190 }
2191
2192 if (clone_flags & CLONE_PIDFD) {
2193 /*
2194 * - CLONE_DETACHED is blocked so that we can potentially
2195 * reuse it later for CLONE_PIDFD.
2196 */
2197 if (clone_flags & CLONE_DETACHED)
2198 return ERR_PTR(-EINVAL);
2199 }
2200
2201 /*
2202 * Force any signals received before this point to be delivered
2203 * before the fork happens. Collect up signals sent to multiple
2204 * processes that happen during the fork and delay them so that
2205 * they appear to happen after the fork.
2206 */
2207 sigemptyset(&delayed.signal);
2208 INIT_HLIST_NODE(&delayed.node);
2209
2210 spin_lock_irq(¤t->sighand->siglock);
2211 if (!(clone_flags & CLONE_THREAD))
2212 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2213 recalc_sigpending();
2214 spin_unlock_irq(¤t->sighand->siglock);
2215 retval = -ERESTARTNOINTR;
2216 if (task_sigpending(current))
2217 goto fork_out;
2218
2219 retval = -ENOMEM;
2220 p = dup_task_struct(current, node);
2221 if (!p)
2222 goto fork_out;
2223 p->flags &= ~PF_KTHREAD;
2224 if (args->kthread)
2225 p->flags |= PF_KTHREAD;
2226 if (args->user_worker) {
2227 /*
2228 * Mark us a user worker, and block any signal that isn't
2229 * fatal or STOP
2230 */
2231 p->flags |= PF_USER_WORKER;
2232 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2233 }
2234 if (args->io_thread)
2235 p->flags |= PF_IO_WORKER;
2236
2237 if (args->name)
2238 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2239
2240 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2241 /*
2242 * Clear TID on mm_release()?
2243 */
2244 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2245
2246 ftrace_graph_init_task(p);
2247
2248 rt_mutex_init_task(p);
2249
2250 lockdep_assert_irqs_enabled();
2251#ifdef CONFIG_PROVE_LOCKING
2252 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2253#endif
2254 retval = copy_creds(p, clone_flags);
2255 if (retval < 0)
2256 goto bad_fork_free;
2257
2258 retval = -EAGAIN;
2259 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2260 if (p->real_cred->user != INIT_USER &&
2261 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2262 goto bad_fork_cleanup_count;
2263 }
2264 current->flags &= ~PF_NPROC_EXCEEDED;
2265
2266 /*
2267 * If multiple threads are within copy_process(), then this check
2268 * triggers too late. This doesn't hurt, the check is only there
2269 * to stop root fork bombs.
2270 */
2271 retval = -EAGAIN;
2272 if (data_race(nr_threads >= max_threads))
2273 goto bad_fork_cleanup_count;
2274
2275 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2276 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2277 p->flags |= PF_FORKNOEXEC;
2278 INIT_LIST_HEAD(&p->children);
2279 INIT_LIST_HEAD(&p->sibling);
2280 rcu_copy_process(p);
2281 p->vfork_done = NULL;
2282 spin_lock_init(&p->alloc_lock);
2283
2284 init_sigpending(&p->pending);
2285
2286 p->utime = p->stime = p->gtime = 0;
2287#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2288 p->utimescaled = p->stimescaled = 0;
2289#endif
2290 prev_cputime_init(&p->prev_cputime);
2291
2292#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2293 seqcount_init(&p->vtime.seqcount);
2294 p->vtime.starttime = 0;
2295 p->vtime.state = VTIME_INACTIVE;
2296#endif
2297
2298#ifdef CONFIG_IO_URING
2299 p->io_uring = NULL;
2300#endif
2301
2302 p->default_timer_slack_ns = current->timer_slack_ns;
2303
2304#ifdef CONFIG_PSI
2305 p->psi_flags = 0;
2306#endif
2307
2308 task_io_accounting_init(&p->ioac);
2309 acct_clear_integrals(p);
2310
2311 posix_cputimers_init(&p->posix_cputimers);
2312
2313 p->io_context = NULL;
2314 audit_set_context(p, NULL);
2315 cgroup_fork(p);
2316 if (args->kthread) {
2317 if (!set_kthread_struct(p))
2318 goto bad_fork_cleanup_delayacct;
2319 }
2320#ifdef CONFIG_NUMA
2321 p->mempolicy = mpol_dup(p->mempolicy);
2322 if (IS_ERR(p->mempolicy)) {
2323 retval = PTR_ERR(p->mempolicy);
2324 p->mempolicy = NULL;
2325 goto bad_fork_cleanup_delayacct;
2326 }
2327#endif
2328#ifdef CONFIG_CPUSETS
2329 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2330 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2331 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2332#endif
2333#ifdef CONFIG_TRACE_IRQFLAGS
2334 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2335 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2336 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2337 p->softirqs_enabled = 1;
2338 p->softirq_context = 0;
2339#endif
2340
2341 p->pagefault_disabled = 0;
2342
2343#ifdef CONFIG_LOCKDEP
2344 lockdep_init_task(p);
2345#endif
2346
2347#ifdef CONFIG_DEBUG_MUTEXES
2348 p->blocked_on = NULL; /* not blocked yet */
2349#endif
2350#ifdef CONFIG_BCACHE
2351 p->sequential_io = 0;
2352 p->sequential_io_avg = 0;
2353#endif
2354#ifdef CONFIG_BPF_SYSCALL
2355 RCU_INIT_POINTER(p->bpf_storage, NULL);
2356 p->bpf_ctx = NULL;
2357#endif
2358
2359 /* Perform scheduler related setup. Assign this task to a CPU. */
2360 retval = sched_fork(clone_flags, p);
2361 if (retval)
2362 goto bad_fork_cleanup_policy;
2363
2364 retval = perf_event_init_task(p, clone_flags);
2365 if (retval)
2366 goto bad_fork_cleanup_policy;
2367 retval = audit_alloc(p);
2368 if (retval)
2369 goto bad_fork_cleanup_perf;
2370 /* copy all the process information */
2371 shm_init_task(p);
2372 retval = security_task_alloc(p, clone_flags);
2373 if (retval)
2374 goto bad_fork_cleanup_audit;
2375 retval = copy_semundo(clone_flags, p);
2376 if (retval)
2377 goto bad_fork_cleanup_security;
2378 retval = copy_files(clone_flags, p, args->no_files);
2379 if (retval)
2380 goto bad_fork_cleanup_semundo;
2381 retval = copy_fs(clone_flags, p);
2382 if (retval)
2383 goto bad_fork_cleanup_files;
2384 retval = copy_sighand(clone_flags, p);
2385 if (retval)
2386 goto bad_fork_cleanup_fs;
2387 retval = copy_signal(clone_flags, p);
2388 if (retval)
2389 goto bad_fork_cleanup_sighand;
2390 retval = copy_mm(clone_flags, p);
2391 if (retval)
2392 goto bad_fork_cleanup_signal;
2393 retval = copy_namespaces(clone_flags, p);
2394 if (retval)
2395 goto bad_fork_cleanup_mm;
2396 retval = copy_io(clone_flags, p);
2397 if (retval)
2398 goto bad_fork_cleanup_namespaces;
2399 retval = copy_thread(p, args);
2400 if (retval)
2401 goto bad_fork_cleanup_io;
2402
2403 stackleak_task_init(p);
2404
2405 if (pid != &init_struct_pid) {
2406 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2407 args->set_tid_size);
2408 if (IS_ERR(pid)) {
2409 retval = PTR_ERR(pid);
2410 goto bad_fork_cleanup_thread;
2411 }
2412 }
2413
2414 /*
2415 * This has to happen after we've potentially unshared the file
2416 * descriptor table (so that the pidfd doesn't leak into the child
2417 * if the fd table isn't shared).
2418 */
2419 if (clone_flags & CLONE_PIDFD) {
2420 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2421
2422 /* Note that no task has been attached to @pid yet. */
2423 retval = __pidfd_prepare(pid, flags, &pidfile);
2424 if (retval < 0)
2425 goto bad_fork_free_pid;
2426 pidfd = retval;
2427
2428 retval = put_user(pidfd, args->pidfd);
2429 if (retval)
2430 goto bad_fork_put_pidfd;
2431 }
2432
2433#ifdef CONFIG_BLOCK
2434 p->plug = NULL;
2435#endif
2436 futex_init_task(p);
2437
2438 /*
2439 * sigaltstack should be cleared when sharing the same VM
2440 */
2441 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2442 sas_ss_reset(p);
2443
2444 /*
2445 * Syscall tracing and stepping should be turned off in the
2446 * child regardless of CLONE_PTRACE.
2447 */
2448 user_disable_single_step(p);
2449 clear_task_syscall_work(p, SYSCALL_TRACE);
2450#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2451 clear_task_syscall_work(p, SYSCALL_EMU);
2452#endif
2453 clear_tsk_latency_tracing(p);
2454
2455 /* ok, now we should be set up.. */
2456 p->pid = pid_nr(pid);
2457 if (clone_flags & CLONE_THREAD) {
2458 p->group_leader = current->group_leader;
2459 p->tgid = current->tgid;
2460 } else {
2461 p->group_leader = p;
2462 p->tgid = p->pid;
2463 }
2464
2465 p->nr_dirtied = 0;
2466 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2467 p->dirty_paused_when = 0;
2468
2469 p->pdeath_signal = 0;
2470 p->task_works = NULL;
2471 clear_posix_cputimers_work(p);
2472
2473#ifdef CONFIG_KRETPROBES
2474 p->kretprobe_instances.first = NULL;
2475#endif
2476#ifdef CONFIG_RETHOOK
2477 p->rethooks.first = NULL;
2478#endif
2479
2480 /*
2481 * Ensure that the cgroup subsystem policies allow the new process to be
2482 * forked. It should be noted that the new process's css_set can be changed
2483 * between here and cgroup_post_fork() if an organisation operation is in
2484 * progress.
2485 */
2486 retval = cgroup_can_fork(p, args);
2487 if (retval)
2488 goto bad_fork_put_pidfd;
2489
2490 /*
2491 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2492 * the new task on the correct runqueue. All this *before* the task
2493 * becomes visible.
2494 *
2495 * This isn't part of ->can_fork() because while the re-cloning is
2496 * cgroup specific, it unconditionally needs to place the task on a
2497 * runqueue.
2498 */
2499 sched_cgroup_fork(p, args);
2500
2501 /*
2502 * From this point on we must avoid any synchronous user-space
2503 * communication until we take the tasklist-lock. In particular, we do
2504 * not want user-space to be able to predict the process start-time by
2505 * stalling fork(2) after we recorded the start_time but before it is
2506 * visible to the system.
2507 */
2508
2509 p->start_time = ktime_get_ns();
2510 p->start_boottime = ktime_get_boottime_ns();
2511
2512 /*
2513 * Make it visible to the rest of the system, but dont wake it up yet.
2514 * Need tasklist lock for parent etc handling!
2515 */
2516 write_lock_irq(&tasklist_lock);
2517
2518 /* CLONE_PARENT re-uses the old parent */
2519 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2520 p->real_parent = current->real_parent;
2521 p->parent_exec_id = current->parent_exec_id;
2522 if (clone_flags & CLONE_THREAD)
2523 p->exit_signal = -1;
2524 else
2525 p->exit_signal = current->group_leader->exit_signal;
2526 } else {
2527 p->real_parent = current;
2528 p->parent_exec_id = current->self_exec_id;
2529 p->exit_signal = args->exit_signal;
2530 }
2531
2532 klp_copy_process(p);
2533
2534 sched_core_fork(p);
2535
2536 spin_lock(¤t->sighand->siglock);
2537
2538 rv_task_fork(p);
2539
2540 rseq_fork(p, clone_flags);
2541
2542 /* Don't start children in a dying pid namespace */
2543 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2544 retval = -ENOMEM;
2545 goto bad_fork_cancel_cgroup;
2546 }
2547
2548 /* Let kill terminate clone/fork in the middle */
2549 if (fatal_signal_pending(current)) {
2550 retval = -EINTR;
2551 goto bad_fork_cancel_cgroup;
2552 }
2553
2554 /* No more failure paths after this point. */
2555
2556 /*
2557 * Copy seccomp details explicitly here, in case they were changed
2558 * before holding sighand lock.
2559 */
2560 copy_seccomp(p);
2561
2562 init_task_pid_links(p);
2563 if (likely(p->pid)) {
2564 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2565
2566 init_task_pid(p, PIDTYPE_PID, pid);
2567 if (thread_group_leader(p)) {
2568 init_task_pid(p, PIDTYPE_TGID, pid);
2569 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2570 init_task_pid(p, PIDTYPE_SID, task_session(current));
2571
2572 if (is_child_reaper(pid)) {
2573 ns_of_pid(pid)->child_reaper = p;
2574 p->signal->flags |= SIGNAL_UNKILLABLE;
2575 }
2576 p->signal->shared_pending.signal = delayed.signal;
2577 p->signal->tty = tty_kref_get(current->signal->tty);
2578 /*
2579 * Inherit has_child_subreaper flag under the same
2580 * tasklist_lock with adding child to the process tree
2581 * for propagate_has_child_subreaper optimization.
2582 */
2583 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2584 p->real_parent->signal->is_child_subreaper;
2585 list_add_tail(&p->sibling, &p->real_parent->children);
2586 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2587 attach_pid(p, PIDTYPE_TGID);
2588 attach_pid(p, PIDTYPE_PGID);
2589 attach_pid(p, PIDTYPE_SID);
2590 __this_cpu_inc(process_counts);
2591 } else {
2592 current->signal->nr_threads++;
2593 current->signal->quick_threads++;
2594 atomic_inc(¤t->signal->live);
2595 refcount_inc(¤t->signal->sigcnt);
2596 task_join_group_stop(p);
2597 list_add_tail_rcu(&p->thread_node,
2598 &p->signal->thread_head);
2599 }
2600 attach_pid(p, PIDTYPE_PID);
2601 nr_threads++;
2602 }
2603 total_forks++;
2604 hlist_del_init(&delayed.node);
2605 spin_unlock(¤t->sighand->siglock);
2606 syscall_tracepoint_update(p);
2607 write_unlock_irq(&tasklist_lock);
2608
2609 if (pidfile)
2610 fd_install(pidfd, pidfile);
2611
2612 proc_fork_connector(p);
2613 sched_post_fork(p);
2614 cgroup_post_fork(p, args);
2615 perf_event_fork(p);
2616
2617 trace_task_newtask(p, clone_flags);
2618 uprobe_copy_process(p, clone_flags);
2619 user_events_fork(p, clone_flags);
2620
2621 copy_oom_score_adj(clone_flags, p);
2622
2623 return p;
2624
2625bad_fork_cancel_cgroup:
2626 sched_core_free(p);
2627 spin_unlock(¤t->sighand->siglock);
2628 write_unlock_irq(&tasklist_lock);
2629 cgroup_cancel_fork(p, args);
2630bad_fork_put_pidfd:
2631 if (clone_flags & CLONE_PIDFD) {
2632 fput(pidfile);
2633 put_unused_fd(pidfd);
2634 }
2635bad_fork_free_pid:
2636 if (pid != &init_struct_pid)
2637 free_pid(pid);
2638bad_fork_cleanup_thread:
2639 exit_thread(p);
2640bad_fork_cleanup_io:
2641 if (p->io_context)
2642 exit_io_context(p);
2643bad_fork_cleanup_namespaces:
2644 exit_task_namespaces(p);
2645bad_fork_cleanup_mm:
2646 if (p->mm) {
2647 mm_clear_owner(p->mm, p);
2648 mmput(p->mm);
2649 }
2650bad_fork_cleanup_signal:
2651 if (!(clone_flags & CLONE_THREAD))
2652 free_signal_struct(p->signal);
2653bad_fork_cleanup_sighand:
2654 __cleanup_sighand(p->sighand);
2655bad_fork_cleanup_fs:
2656 exit_fs(p); /* blocking */
2657bad_fork_cleanup_files:
2658 exit_files(p); /* blocking */
2659bad_fork_cleanup_semundo:
2660 exit_sem(p);
2661bad_fork_cleanup_security:
2662 security_task_free(p);
2663bad_fork_cleanup_audit:
2664 audit_free(p);
2665bad_fork_cleanup_perf:
2666 perf_event_free_task(p);
2667bad_fork_cleanup_policy:
2668 lockdep_free_task(p);
2669#ifdef CONFIG_NUMA
2670 mpol_put(p->mempolicy);
2671#endif
2672bad_fork_cleanup_delayacct:
2673 delayacct_tsk_free(p);
2674bad_fork_cleanup_count:
2675 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2676 exit_creds(p);
2677bad_fork_free:
2678 WRITE_ONCE(p->__state, TASK_DEAD);
2679 exit_task_stack_account(p);
2680 put_task_stack(p);
2681 delayed_free_task(p);
2682fork_out:
2683 spin_lock_irq(¤t->sighand->siglock);
2684 hlist_del_init(&delayed.node);
2685 spin_unlock_irq(¤t->sighand->siglock);
2686 return ERR_PTR(retval);
2687}
2688
2689static inline void init_idle_pids(struct task_struct *idle)
2690{
2691 enum pid_type type;
2692
2693 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2694 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2695 init_task_pid(idle, type, &init_struct_pid);
2696 }
2697}
2698
2699static int idle_dummy(void *dummy)
2700{
2701 /* This function is never called */
2702 return 0;
2703}
2704
2705struct task_struct * __init fork_idle(int cpu)
2706{
2707 struct task_struct *task;
2708 struct kernel_clone_args args = {
2709 .flags = CLONE_VM,
2710 .fn = &idle_dummy,
2711 .fn_arg = NULL,
2712 .kthread = 1,
2713 .idle = 1,
2714 };
2715
2716 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2717 if (!IS_ERR(task)) {
2718 init_idle_pids(task);
2719 init_idle(task, cpu);
2720 }
2721
2722 return task;
2723}
2724
2725/*
2726 * This is like kernel_clone(), but shaved down and tailored to just
2727 * creating io_uring workers. It returns a created task, or an error pointer.
2728 * The returned task is inactive, and the caller must fire it up through
2729 * wake_up_new_task(p). All signals are blocked in the created task.
2730 */
2731struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2732{
2733 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2734 CLONE_IO;
2735 struct kernel_clone_args args = {
2736 .flags = ((lower_32_bits(flags) | CLONE_VM |
2737 CLONE_UNTRACED) & ~CSIGNAL),
2738 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2739 .fn = fn,
2740 .fn_arg = arg,
2741 .io_thread = 1,
2742 .user_worker = 1,
2743 };
2744
2745 return copy_process(NULL, 0, node, &args);
2746}
2747
2748/*
2749 * Ok, this is the main fork-routine.
2750 *
2751 * It copies the process, and if successful kick-starts
2752 * it and waits for it to finish using the VM if required.
2753 *
2754 * args->exit_signal is expected to be checked for sanity by the caller.
2755 */
2756pid_t kernel_clone(struct kernel_clone_args *args)
2757{
2758 u64 clone_flags = args->flags;
2759 struct completion vfork;
2760 struct pid *pid;
2761 struct task_struct *p;
2762 int trace = 0;
2763 pid_t nr;
2764
2765 /*
2766 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2767 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2768 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2769 * field in struct clone_args and it still doesn't make sense to have
2770 * them both point at the same memory location. Performing this check
2771 * here has the advantage that we don't need to have a separate helper
2772 * to check for legacy clone().
2773 */
2774 if ((clone_flags & CLONE_PIDFD) &&
2775 (clone_flags & CLONE_PARENT_SETTID) &&
2776 (args->pidfd == args->parent_tid))
2777 return -EINVAL;
2778
2779 /*
2780 * Determine whether and which event to report to ptracer. When
2781 * called from kernel_thread or CLONE_UNTRACED is explicitly
2782 * requested, no event is reported; otherwise, report if the event
2783 * for the type of forking is enabled.
2784 */
2785 if (!(clone_flags & CLONE_UNTRACED)) {
2786 if (clone_flags & CLONE_VFORK)
2787 trace = PTRACE_EVENT_VFORK;
2788 else if (args->exit_signal != SIGCHLD)
2789 trace = PTRACE_EVENT_CLONE;
2790 else
2791 trace = PTRACE_EVENT_FORK;
2792
2793 if (likely(!ptrace_event_enabled(current, trace)))
2794 trace = 0;
2795 }
2796
2797 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2798 add_latent_entropy();
2799
2800 if (IS_ERR(p))
2801 return PTR_ERR(p);
2802
2803 /*
2804 * Do this prior waking up the new thread - the thread pointer
2805 * might get invalid after that point, if the thread exits quickly.
2806 */
2807 trace_sched_process_fork(current, p);
2808
2809 pid = get_task_pid(p, PIDTYPE_PID);
2810 nr = pid_vnr(pid);
2811
2812 if (clone_flags & CLONE_PARENT_SETTID)
2813 put_user(nr, args->parent_tid);
2814
2815 if (clone_flags & CLONE_VFORK) {
2816 p->vfork_done = &vfork;
2817 init_completion(&vfork);
2818 get_task_struct(p);
2819 }
2820
2821 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2822 /* lock the task to synchronize with memcg migration */
2823 task_lock(p);
2824 lru_gen_add_mm(p->mm);
2825 task_unlock(p);
2826 }
2827
2828 wake_up_new_task(p);
2829
2830 /* forking complete and child started to run, tell ptracer */
2831 if (unlikely(trace))
2832 ptrace_event_pid(trace, pid);
2833
2834 if (clone_flags & CLONE_VFORK) {
2835 if (!wait_for_vfork_done(p, &vfork))
2836 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2837 }
2838
2839 put_pid(pid);
2840 return nr;
2841}
2842
2843/*
2844 * Create a kernel thread.
2845 */
2846pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2847 unsigned long flags)
2848{
2849 struct kernel_clone_args args = {
2850 .flags = ((lower_32_bits(flags) | CLONE_VM |
2851 CLONE_UNTRACED) & ~CSIGNAL),
2852 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2853 .fn = fn,
2854 .fn_arg = arg,
2855 .name = name,
2856 .kthread = 1,
2857 };
2858
2859 return kernel_clone(&args);
2860}
2861
2862/*
2863 * Create a user mode thread.
2864 */
2865pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2866{
2867 struct kernel_clone_args args = {
2868 .flags = ((lower_32_bits(flags) | CLONE_VM |
2869 CLONE_UNTRACED) & ~CSIGNAL),
2870 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2871 .fn = fn,
2872 .fn_arg = arg,
2873 };
2874
2875 return kernel_clone(&args);
2876}
2877
2878#ifdef __ARCH_WANT_SYS_FORK
2879SYSCALL_DEFINE0(fork)
2880{
2881#ifdef CONFIG_MMU
2882 struct kernel_clone_args args = {
2883 .exit_signal = SIGCHLD,
2884 };
2885
2886 return kernel_clone(&args);
2887#else
2888 /* can not support in nommu mode */
2889 return -EINVAL;
2890#endif
2891}
2892#endif
2893
2894#ifdef __ARCH_WANT_SYS_VFORK
2895SYSCALL_DEFINE0(vfork)
2896{
2897 struct kernel_clone_args args = {
2898 .flags = CLONE_VFORK | CLONE_VM,
2899 .exit_signal = SIGCHLD,
2900 };
2901
2902 return kernel_clone(&args);
2903}
2904#endif
2905
2906#ifdef __ARCH_WANT_SYS_CLONE
2907#ifdef CONFIG_CLONE_BACKWARDS
2908SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2909 int __user *, parent_tidptr,
2910 unsigned long, tls,
2911 int __user *, child_tidptr)
2912#elif defined(CONFIG_CLONE_BACKWARDS2)
2913SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2914 int __user *, parent_tidptr,
2915 int __user *, child_tidptr,
2916 unsigned long, tls)
2917#elif defined(CONFIG_CLONE_BACKWARDS3)
2918SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2919 int, stack_size,
2920 int __user *, parent_tidptr,
2921 int __user *, child_tidptr,
2922 unsigned long, tls)
2923#else
2924SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2925 int __user *, parent_tidptr,
2926 int __user *, child_tidptr,
2927 unsigned long, tls)
2928#endif
2929{
2930 struct kernel_clone_args args = {
2931 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2932 .pidfd = parent_tidptr,
2933 .child_tid = child_tidptr,
2934 .parent_tid = parent_tidptr,
2935 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2936 .stack = newsp,
2937 .tls = tls,
2938 };
2939
2940 return kernel_clone(&args);
2941}
2942#endif
2943
2944#ifdef __ARCH_WANT_SYS_CLONE3
2945
2946noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2947 struct clone_args __user *uargs,
2948 size_t usize)
2949{
2950 int err;
2951 struct clone_args args;
2952 pid_t *kset_tid = kargs->set_tid;
2953
2954 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2955 CLONE_ARGS_SIZE_VER0);
2956 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2957 CLONE_ARGS_SIZE_VER1);
2958 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2959 CLONE_ARGS_SIZE_VER2);
2960 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2961
2962 if (unlikely(usize > PAGE_SIZE))
2963 return -E2BIG;
2964 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2965 return -EINVAL;
2966
2967 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2968 if (err)
2969 return err;
2970
2971 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2972 return -EINVAL;
2973
2974 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2975 return -EINVAL;
2976
2977 if (unlikely(args.set_tid && args.set_tid_size == 0))
2978 return -EINVAL;
2979
2980 /*
2981 * Verify that higher 32bits of exit_signal are unset and that
2982 * it is a valid signal
2983 */
2984 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2985 !valid_signal(args.exit_signal)))
2986 return -EINVAL;
2987
2988 if ((args.flags & CLONE_INTO_CGROUP) &&
2989 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2990 return -EINVAL;
2991
2992 *kargs = (struct kernel_clone_args){
2993 .flags = args.flags,
2994 .pidfd = u64_to_user_ptr(args.pidfd),
2995 .child_tid = u64_to_user_ptr(args.child_tid),
2996 .parent_tid = u64_to_user_ptr(args.parent_tid),
2997 .exit_signal = args.exit_signal,
2998 .stack = args.stack,
2999 .stack_size = args.stack_size,
3000 .tls = args.tls,
3001 .set_tid_size = args.set_tid_size,
3002 .cgroup = args.cgroup,
3003 };
3004
3005 if (args.set_tid &&
3006 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3007 (kargs->set_tid_size * sizeof(pid_t))))
3008 return -EFAULT;
3009
3010 kargs->set_tid = kset_tid;
3011
3012 return 0;
3013}
3014
3015/**
3016 * clone3_stack_valid - check and prepare stack
3017 * @kargs: kernel clone args
3018 *
3019 * Verify that the stack arguments userspace gave us are sane.
3020 * In addition, set the stack direction for userspace since it's easy for us to
3021 * determine.
3022 */
3023static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3024{
3025 if (kargs->stack == 0) {
3026 if (kargs->stack_size > 0)
3027 return false;
3028 } else {
3029 if (kargs->stack_size == 0)
3030 return false;
3031
3032 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3033 return false;
3034
3035#if !defined(CONFIG_STACK_GROWSUP)
3036 kargs->stack += kargs->stack_size;
3037#endif
3038 }
3039
3040 return true;
3041}
3042
3043static bool clone3_args_valid(struct kernel_clone_args *kargs)
3044{
3045 /* Verify that no unknown flags are passed along. */
3046 if (kargs->flags &
3047 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3048 return false;
3049
3050 /*
3051 * - make the CLONE_DETACHED bit reusable for clone3
3052 * - make the CSIGNAL bits reusable for clone3
3053 */
3054 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3055 return false;
3056
3057 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3058 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3059 return false;
3060
3061 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3062 kargs->exit_signal)
3063 return false;
3064
3065 if (!clone3_stack_valid(kargs))
3066 return false;
3067
3068 return true;
3069}
3070
3071/**
3072 * sys_clone3 - create a new process with specific properties
3073 * @uargs: argument structure
3074 * @size: size of @uargs
3075 *
3076 * clone3() is the extensible successor to clone()/clone2().
3077 * It takes a struct as argument that is versioned by its size.
3078 *
3079 * Return: On success, a positive PID for the child process.
3080 * On error, a negative errno number.
3081 */
3082SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3083{
3084 int err;
3085
3086 struct kernel_clone_args kargs;
3087 pid_t set_tid[MAX_PID_NS_LEVEL];
3088
3089 kargs.set_tid = set_tid;
3090
3091 err = copy_clone_args_from_user(&kargs, uargs, size);
3092 if (err)
3093 return err;
3094
3095 if (!clone3_args_valid(&kargs))
3096 return -EINVAL;
3097
3098 return kernel_clone(&kargs);
3099}
3100#endif
3101
3102void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3103{
3104 struct task_struct *leader, *parent, *child;
3105 int res;
3106
3107 read_lock(&tasklist_lock);
3108 leader = top = top->group_leader;
3109down:
3110 for_each_thread(leader, parent) {
3111 list_for_each_entry(child, &parent->children, sibling) {
3112 res = visitor(child, data);
3113 if (res) {
3114 if (res < 0)
3115 goto out;
3116 leader = child;
3117 goto down;
3118 }
3119up:
3120 ;
3121 }
3122 }
3123
3124 if (leader != top) {
3125 child = leader;
3126 parent = child->real_parent;
3127 leader = parent->group_leader;
3128 goto up;
3129 }
3130out:
3131 read_unlock(&tasklist_lock);
3132}
3133
3134#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3135#define ARCH_MIN_MMSTRUCT_ALIGN 0
3136#endif
3137
3138static void sighand_ctor(void *data)
3139{
3140 struct sighand_struct *sighand = data;
3141
3142 spin_lock_init(&sighand->siglock);
3143 init_waitqueue_head(&sighand->signalfd_wqh);
3144}
3145
3146void __init mm_cache_init(void)
3147{
3148 unsigned int mm_size;
3149
3150 /*
3151 * The mm_cpumask is located at the end of mm_struct, and is
3152 * dynamically sized based on the maximum CPU number this system
3153 * can have, taking hotplug into account (nr_cpu_ids).
3154 */
3155 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3156
3157 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3158 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3159 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3160 offsetof(struct mm_struct, saved_auxv),
3161 sizeof_field(struct mm_struct, saved_auxv),
3162 NULL);
3163}
3164
3165void __init proc_caches_init(void)
3166{
3167 sighand_cachep = kmem_cache_create("sighand_cache",
3168 sizeof(struct sighand_struct), 0,
3169 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3170 SLAB_ACCOUNT, sighand_ctor);
3171 signal_cachep = kmem_cache_create("signal_cache",
3172 sizeof(struct signal_struct), 0,
3173 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3174 NULL);
3175 files_cachep = kmem_cache_create("files_cache",
3176 sizeof(struct files_struct), 0,
3177 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3178 NULL);
3179 fs_cachep = kmem_cache_create("fs_cache",
3180 sizeof(struct fs_struct), 0,
3181 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3182 NULL);
3183
3184 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3185#ifdef CONFIG_PER_VMA_LOCK
3186 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3187#endif
3188 mmap_init();
3189 nsproxy_cache_init();
3190}
3191
3192/*
3193 * Check constraints on flags passed to the unshare system call.
3194 */
3195static int check_unshare_flags(unsigned long unshare_flags)
3196{
3197 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3198 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3199 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3200 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3201 CLONE_NEWTIME))
3202 return -EINVAL;
3203 /*
3204 * Not implemented, but pretend it works if there is nothing
3205 * to unshare. Note that unsharing the address space or the
3206 * signal handlers also need to unshare the signal queues (aka
3207 * CLONE_THREAD).
3208 */
3209 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3210 if (!thread_group_empty(current))
3211 return -EINVAL;
3212 }
3213 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3214 if (refcount_read(¤t->sighand->count) > 1)
3215 return -EINVAL;
3216 }
3217 if (unshare_flags & CLONE_VM) {
3218 if (!current_is_single_threaded())
3219 return -EINVAL;
3220 }
3221
3222 return 0;
3223}
3224
3225/*
3226 * Unshare the filesystem structure if it is being shared
3227 */
3228static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3229{
3230 struct fs_struct *fs = current->fs;
3231
3232 if (!(unshare_flags & CLONE_FS) || !fs)
3233 return 0;
3234
3235 /* don't need lock here; in the worst case we'll do useless copy */
3236 if (fs->users == 1)
3237 return 0;
3238
3239 *new_fsp = copy_fs_struct(fs);
3240 if (!*new_fsp)
3241 return -ENOMEM;
3242
3243 return 0;
3244}
3245
3246/*
3247 * Unshare file descriptor table if it is being shared
3248 */
3249int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3250 struct files_struct **new_fdp)
3251{
3252 struct files_struct *fd = current->files;
3253 int error = 0;
3254
3255 if ((unshare_flags & CLONE_FILES) &&
3256 (fd && atomic_read(&fd->count) > 1)) {
3257 *new_fdp = dup_fd(fd, max_fds, &error);
3258 if (!*new_fdp)
3259 return error;
3260 }
3261
3262 return 0;
3263}
3264
3265/*
3266 * unshare allows a process to 'unshare' part of the process
3267 * context which was originally shared using clone. copy_*
3268 * functions used by kernel_clone() cannot be used here directly
3269 * because they modify an inactive task_struct that is being
3270 * constructed. Here we are modifying the current, active,
3271 * task_struct.
3272 */
3273int ksys_unshare(unsigned long unshare_flags)
3274{
3275 struct fs_struct *fs, *new_fs = NULL;
3276 struct files_struct *new_fd = NULL;
3277 struct cred *new_cred = NULL;
3278 struct nsproxy *new_nsproxy = NULL;
3279 int do_sysvsem = 0;
3280 int err;
3281
3282 /*
3283 * If unsharing a user namespace must also unshare the thread group
3284 * and unshare the filesystem root and working directories.
3285 */
3286 if (unshare_flags & CLONE_NEWUSER)
3287 unshare_flags |= CLONE_THREAD | CLONE_FS;
3288 /*
3289 * If unsharing vm, must also unshare signal handlers.
3290 */
3291 if (unshare_flags & CLONE_VM)
3292 unshare_flags |= CLONE_SIGHAND;
3293 /*
3294 * If unsharing a signal handlers, must also unshare the signal queues.
3295 */
3296 if (unshare_flags & CLONE_SIGHAND)
3297 unshare_flags |= CLONE_THREAD;
3298 /*
3299 * If unsharing namespace, must also unshare filesystem information.
3300 */
3301 if (unshare_flags & CLONE_NEWNS)
3302 unshare_flags |= CLONE_FS;
3303
3304 err = check_unshare_flags(unshare_flags);
3305 if (err)
3306 goto bad_unshare_out;
3307 /*
3308 * CLONE_NEWIPC must also detach from the undolist: after switching
3309 * to a new ipc namespace, the semaphore arrays from the old
3310 * namespace are unreachable.
3311 */
3312 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3313 do_sysvsem = 1;
3314 err = unshare_fs(unshare_flags, &new_fs);
3315 if (err)
3316 goto bad_unshare_out;
3317 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3318 if (err)
3319 goto bad_unshare_cleanup_fs;
3320 err = unshare_userns(unshare_flags, &new_cred);
3321 if (err)
3322 goto bad_unshare_cleanup_fd;
3323 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3324 new_cred, new_fs);
3325 if (err)
3326 goto bad_unshare_cleanup_cred;
3327
3328 if (new_cred) {
3329 err = set_cred_ucounts(new_cred);
3330 if (err)
3331 goto bad_unshare_cleanup_cred;
3332 }
3333
3334 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3335 if (do_sysvsem) {
3336 /*
3337 * CLONE_SYSVSEM is equivalent to sys_exit().
3338 */
3339 exit_sem(current);
3340 }
3341 if (unshare_flags & CLONE_NEWIPC) {
3342 /* Orphan segments in old ns (see sem above). */
3343 exit_shm(current);
3344 shm_init_task(current);
3345 }
3346
3347 if (new_nsproxy)
3348 switch_task_namespaces(current, new_nsproxy);
3349
3350 task_lock(current);
3351
3352 if (new_fs) {
3353 fs = current->fs;
3354 spin_lock(&fs->lock);
3355 current->fs = new_fs;
3356 if (--fs->users)
3357 new_fs = NULL;
3358 else
3359 new_fs = fs;
3360 spin_unlock(&fs->lock);
3361 }
3362
3363 if (new_fd)
3364 swap(current->files, new_fd);
3365
3366 task_unlock(current);
3367
3368 if (new_cred) {
3369 /* Install the new user namespace */
3370 commit_creds(new_cred);
3371 new_cred = NULL;
3372 }
3373 }
3374
3375 perf_event_namespaces(current);
3376
3377bad_unshare_cleanup_cred:
3378 if (new_cred)
3379 put_cred(new_cred);
3380bad_unshare_cleanup_fd:
3381 if (new_fd)
3382 put_files_struct(new_fd);
3383
3384bad_unshare_cleanup_fs:
3385 if (new_fs)
3386 free_fs_struct(new_fs);
3387
3388bad_unshare_out:
3389 return err;
3390}
3391
3392SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3393{
3394 return ksys_unshare(unshare_flags);
3395}
3396
3397/*
3398 * Helper to unshare the files of the current task.
3399 * We don't want to expose copy_files internals to
3400 * the exec layer of the kernel.
3401 */
3402
3403int unshare_files(void)
3404{
3405 struct task_struct *task = current;
3406 struct files_struct *old, *copy = NULL;
3407 int error;
3408
3409 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3410 if (error || !copy)
3411 return error;
3412
3413 old = task->files;
3414 task_lock(task);
3415 task->files = copy;
3416 task_unlock(task);
3417 put_files_struct(old);
3418 return 0;
3419}
3420
3421int sysctl_max_threads(struct ctl_table *table, int write,
3422 void *buffer, size_t *lenp, loff_t *ppos)
3423{
3424 struct ctl_table t;
3425 int ret;
3426 int threads = max_threads;
3427 int min = 1;
3428 int max = MAX_THREADS;
3429
3430 t = *table;
3431 t.data = &threads;
3432 t.extra1 = &min;
3433 t.extra2 = &max;
3434
3435 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3436 if (ret || !write)
3437 return ret;
3438
3439 max_threads = threads;
3440
3441 return 0;
3442}