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/jiffies.h>
57#include <linux/futex.h>
58#include <linux/compat.h>
59#include <linux/kthread.h>
60#include <linux/task_io_accounting_ops.h>
61#include <linux/rcupdate.h>
62#include <linux/ptrace.h>
63#include <linux/mount.h>
64#include <linux/audit.h>
65#include <linux/memcontrol.h>
66#include <linux/ftrace.h>
67#include <linux/proc_fs.h>
68#include <linux/profile.h>
69#include <linux/rmap.h>
70#include <linux/ksm.h>
71#include <linux/acct.h>
72#include <linux/userfaultfd_k.h>
73#include <linux/tsacct_kern.h>
74#include <linux/cn_proc.h>
75#include <linux/freezer.h>
76#include <linux/delayacct.h>
77#include <linux/taskstats_kern.h>
78#include <linux/tty.h>
79#include <linux/fs_struct.h>
80#include <linux/magic.h>
81#include <linux/perf_event.h>
82#include <linux/posix-timers.h>
83#include <linux/user-return-notifier.h>
84#include <linux/oom.h>
85#include <linux/khugepaged.h>
86#include <linux/signalfd.h>
87#include <linux/uprobes.h>
88#include <linux/aio.h>
89#include <linux/compiler.h>
90#include <linux/sysctl.h>
91#include <linux/kcov.h>
92#include <linux/livepatch.h>
93#include <linux/thread_info.h>
94#include <linux/stackleak.h>
95#include <linux/kasan.h>
96#include <linux/scs.h>
97#include <linux/io_uring.h>
98#include <linux/bpf.h>
99#include <linux/stackprotector.h>
100
101#include <asm/pgalloc.h>
102#include <linux/uaccess.h>
103#include <asm/mmu_context.h>
104#include <asm/cacheflush.h>
105#include <asm/tlbflush.h>
106
107#include <trace/events/sched.h>
108
109#define CREATE_TRACE_POINTS
110#include <trace/events/task.h>
111
112/*
113 * Minimum number of threads to boot the kernel
114 */
115#define MIN_THREADS 20
116
117/*
118 * Maximum number of threads
119 */
120#define MAX_THREADS FUTEX_TID_MASK
121
122/*
123 * Protected counters by write_lock_irq(&tasklist_lock)
124 */
125unsigned long total_forks; /* Handle normal Linux uptimes. */
126int nr_threads; /* The idle threads do not count.. */
127
128static int max_threads; /* tunable limit on nr_threads */
129
130#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
131
132static const char * const resident_page_types[] = {
133 NAMED_ARRAY_INDEX(MM_FILEPAGES),
134 NAMED_ARRAY_INDEX(MM_ANONPAGES),
135 NAMED_ARRAY_INDEX(MM_SWAPENTS),
136 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
137};
138
139DEFINE_PER_CPU(unsigned long, process_counts) = 0;
140
141__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
142
143#ifdef CONFIG_PROVE_RCU
144int lockdep_tasklist_lock_is_held(void)
145{
146 return lockdep_is_held(&tasklist_lock);
147}
148EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
149#endif /* #ifdef CONFIG_PROVE_RCU */
150
151int nr_processes(void)
152{
153 int cpu;
154 int total = 0;
155
156 for_each_possible_cpu(cpu)
157 total += per_cpu(process_counts, cpu);
158
159 return total;
160}
161
162void __weak arch_release_task_struct(struct task_struct *tsk)
163{
164}
165
166#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
167static struct kmem_cache *task_struct_cachep;
168
169static inline struct task_struct *alloc_task_struct_node(int node)
170{
171 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
172}
173
174static inline void free_task_struct(struct task_struct *tsk)
175{
176 kmem_cache_free(task_struct_cachep, tsk);
177}
178#endif
179
180#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
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
254 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
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 }
262 return 0;
263err:
264 /*
265 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
266 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
267 * ignore this page.
268 */
269 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
270 memcg_kmem_uncharge_page(vm->pages[i], 0);
271 return ret;
272}
273
274static int alloc_thread_stack_node(struct task_struct *tsk, int node)
275{
276 struct vm_struct *vm;
277 void *stack;
278 int i;
279
280 for (i = 0; i < NR_CACHED_STACKS; i++) {
281 struct vm_struct *s;
282
283 s = this_cpu_xchg(cached_stacks[i], NULL);
284
285 if (!s)
286 continue;
287
288 /* Reset stack metadata. */
289 kasan_unpoison_range(s->addr, THREAD_SIZE);
290
291 stack = kasan_reset_tag(s->addr);
292
293 /* Clear stale pointers from reused stack. */
294 memset(stack, 0, THREAD_SIZE);
295
296 if (memcg_charge_kernel_stack(s)) {
297 vfree(s->addr);
298 return -ENOMEM;
299 }
300
301 tsk->stack_vm_area = s;
302 tsk->stack = stack;
303 return 0;
304 }
305
306 /*
307 * Allocated stacks are cached and later reused by new threads,
308 * so memcg accounting is performed manually on assigning/releasing
309 * stacks to tasks. Drop __GFP_ACCOUNT.
310 */
311 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
312 VMALLOC_START, VMALLOC_END,
313 THREADINFO_GFP & ~__GFP_ACCOUNT,
314 PAGE_KERNEL,
315 0, node, __builtin_return_address(0));
316 if (!stack)
317 return -ENOMEM;
318
319 vm = find_vm_area(stack);
320 if (memcg_charge_kernel_stack(vm)) {
321 vfree(stack);
322 return -ENOMEM;
323 }
324 /*
325 * We can't call find_vm_area() in interrupt context, and
326 * free_thread_stack() can be called in interrupt context,
327 * so cache the vm_struct.
328 */
329 tsk->stack_vm_area = vm;
330 stack = kasan_reset_tag(stack);
331 tsk->stack = stack;
332 return 0;
333}
334
335static void free_thread_stack(struct task_struct *tsk)
336{
337 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
338 thread_stack_delayed_free(tsk);
339
340 tsk->stack = NULL;
341 tsk->stack_vm_area = NULL;
342}
343
344# else /* !CONFIG_VMAP_STACK */
345
346static void thread_stack_free_rcu(struct rcu_head *rh)
347{
348 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
349}
350
351static void thread_stack_delayed_free(struct task_struct *tsk)
352{
353 struct rcu_head *rh = tsk->stack;
354
355 call_rcu(rh, thread_stack_free_rcu);
356}
357
358static int alloc_thread_stack_node(struct task_struct *tsk, int node)
359{
360 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
361 THREAD_SIZE_ORDER);
362
363 if (likely(page)) {
364 tsk->stack = kasan_reset_tag(page_address(page));
365 return 0;
366 }
367 return -ENOMEM;
368}
369
370static void free_thread_stack(struct task_struct *tsk)
371{
372 thread_stack_delayed_free(tsk);
373 tsk->stack = NULL;
374}
375
376# endif /* CONFIG_VMAP_STACK */
377# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
378
379static struct kmem_cache *thread_stack_cache;
380
381static void thread_stack_free_rcu(struct rcu_head *rh)
382{
383 kmem_cache_free(thread_stack_cache, rh);
384}
385
386static void thread_stack_delayed_free(struct task_struct *tsk)
387{
388 struct rcu_head *rh = tsk->stack;
389
390 call_rcu(rh, thread_stack_free_rcu);
391}
392
393static int alloc_thread_stack_node(struct task_struct *tsk, int node)
394{
395 unsigned long *stack;
396 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
397 stack = kasan_reset_tag(stack);
398 tsk->stack = stack;
399 return stack ? 0 : -ENOMEM;
400}
401
402static void free_thread_stack(struct task_struct *tsk)
403{
404 thread_stack_delayed_free(tsk);
405 tsk->stack = NULL;
406}
407
408void thread_stack_cache_init(void)
409{
410 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
411 THREAD_SIZE, THREAD_SIZE, 0, 0,
412 THREAD_SIZE, NULL);
413 BUG_ON(thread_stack_cache == NULL);
414}
415
416# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
417#else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
418
419static int alloc_thread_stack_node(struct task_struct *tsk, int node)
420{
421 unsigned long *stack;
422
423 stack = arch_alloc_thread_stack_node(tsk, node);
424 tsk->stack = stack;
425 return stack ? 0 : -ENOMEM;
426}
427
428static void free_thread_stack(struct task_struct *tsk)
429{
430 arch_free_thread_stack(tsk);
431 tsk->stack = NULL;
432}
433
434#endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
435
436/* SLAB cache for signal_struct structures (tsk->signal) */
437static struct kmem_cache *signal_cachep;
438
439/* SLAB cache for sighand_struct structures (tsk->sighand) */
440struct kmem_cache *sighand_cachep;
441
442/* SLAB cache for files_struct structures (tsk->files) */
443struct kmem_cache *files_cachep;
444
445/* SLAB cache for fs_struct structures (tsk->fs) */
446struct kmem_cache *fs_cachep;
447
448/* SLAB cache for vm_area_struct structures */
449static struct kmem_cache *vm_area_cachep;
450
451/* SLAB cache for mm_struct structures (tsk->mm) */
452static struct kmem_cache *mm_cachep;
453
454struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
455{
456 struct vm_area_struct *vma;
457
458 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
459 if (vma)
460 vma_init(vma, mm);
461 return vma;
462}
463
464struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
465{
466 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
467
468 if (new) {
469 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
470 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
471 /*
472 * orig->shared.rb may be modified concurrently, but the clone
473 * will be reinitialized.
474 */
475 *new = data_race(*orig);
476 INIT_LIST_HEAD(&new->anon_vma_chain);
477 dup_anon_vma_name(orig, new);
478 }
479 return new;
480}
481
482void vm_area_free(struct vm_area_struct *vma)
483{
484 free_anon_vma_name(vma);
485 kmem_cache_free(vm_area_cachep, vma);
486}
487
488static void account_kernel_stack(struct task_struct *tsk, int account)
489{
490 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
491 struct vm_struct *vm = task_stack_vm_area(tsk);
492 int i;
493
494 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
495 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
496 account * (PAGE_SIZE / 1024));
497 } else {
498 void *stack = task_stack_page(tsk);
499
500 /* All stack pages are in the same node. */
501 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
502 account * (THREAD_SIZE / 1024));
503 }
504}
505
506void exit_task_stack_account(struct task_struct *tsk)
507{
508 account_kernel_stack(tsk, -1);
509
510 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
511 struct vm_struct *vm;
512 int i;
513
514 vm = task_stack_vm_area(tsk);
515 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
516 memcg_kmem_uncharge_page(vm->pages[i], 0);
517 }
518}
519
520static void release_task_stack(struct task_struct *tsk)
521{
522 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
523 return; /* Better to leak the stack than to free prematurely */
524
525 free_thread_stack(tsk);
526}
527
528#ifdef CONFIG_THREAD_INFO_IN_TASK
529void put_task_stack(struct task_struct *tsk)
530{
531 if (refcount_dec_and_test(&tsk->stack_refcount))
532 release_task_stack(tsk);
533}
534#endif
535
536void free_task(struct task_struct *tsk)
537{
538#ifdef CONFIG_SECCOMP
539 WARN_ON_ONCE(tsk->seccomp.filter);
540#endif
541 release_user_cpus_ptr(tsk);
542 scs_release(tsk);
543
544#ifndef CONFIG_THREAD_INFO_IN_TASK
545 /*
546 * The task is finally done with both the stack and thread_info,
547 * so free both.
548 */
549 release_task_stack(tsk);
550#else
551 /*
552 * If the task had a separate stack allocation, it should be gone
553 * by now.
554 */
555 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
556#endif
557 rt_mutex_debug_task_free(tsk);
558 ftrace_graph_exit_task(tsk);
559 arch_release_task_struct(tsk);
560 if (tsk->flags & PF_KTHREAD)
561 free_kthread_struct(tsk);
562 free_task_struct(tsk);
563}
564EXPORT_SYMBOL(free_task);
565
566static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
567{
568 struct file *exe_file;
569
570 exe_file = get_mm_exe_file(oldmm);
571 RCU_INIT_POINTER(mm->exe_file, exe_file);
572 /*
573 * We depend on the oldmm having properly denied write access to the
574 * exe_file already.
575 */
576 if (exe_file && deny_write_access(exe_file))
577 pr_warn_once("deny_write_access() failed in %s\n", __func__);
578}
579
580#ifdef CONFIG_MMU
581static __latent_entropy int dup_mmap(struct mm_struct *mm,
582 struct mm_struct *oldmm)
583{
584 struct vm_area_struct *mpnt, *tmp;
585 int retval;
586 unsigned long charge = 0;
587 LIST_HEAD(uf);
588 MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
589 MA_STATE(mas, &mm->mm_mt, 0, 0);
590
591 uprobe_start_dup_mmap();
592 if (mmap_write_lock_killable(oldmm)) {
593 retval = -EINTR;
594 goto fail_uprobe_end;
595 }
596 flush_cache_dup_mm(oldmm);
597 uprobe_dup_mmap(oldmm, mm);
598 /*
599 * Not linked in yet - no deadlock potential:
600 */
601 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
602
603 /* No ordering required: file already has been exposed. */
604 dup_mm_exe_file(mm, oldmm);
605
606 mm->total_vm = oldmm->total_vm;
607 mm->data_vm = oldmm->data_vm;
608 mm->exec_vm = oldmm->exec_vm;
609 mm->stack_vm = oldmm->stack_vm;
610
611 retval = ksm_fork(mm, oldmm);
612 if (retval)
613 goto out;
614 khugepaged_fork(mm, oldmm);
615
616 retval = mas_expected_entries(&mas, oldmm->map_count);
617 if (retval)
618 goto out;
619
620 mas_for_each(&old_mas, mpnt, ULONG_MAX) {
621 struct file *file;
622
623 if (mpnt->vm_flags & VM_DONTCOPY) {
624 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
625 continue;
626 }
627 charge = 0;
628 /*
629 * Don't duplicate many vmas if we've been oom-killed (for
630 * example)
631 */
632 if (fatal_signal_pending(current)) {
633 retval = -EINTR;
634 goto loop_out;
635 }
636 if (mpnt->vm_flags & VM_ACCOUNT) {
637 unsigned long len = vma_pages(mpnt);
638
639 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
640 goto fail_nomem;
641 charge = len;
642 }
643 tmp = vm_area_dup(mpnt);
644 if (!tmp)
645 goto fail_nomem;
646 retval = vma_dup_policy(mpnt, tmp);
647 if (retval)
648 goto fail_nomem_policy;
649 tmp->vm_mm = mm;
650 retval = dup_userfaultfd(tmp, &uf);
651 if (retval)
652 goto fail_nomem_anon_vma_fork;
653 if (tmp->vm_flags & VM_WIPEONFORK) {
654 /*
655 * VM_WIPEONFORK gets a clean slate in the child.
656 * Don't prepare anon_vma until fault since we don't
657 * copy page for current vma.
658 */
659 tmp->anon_vma = NULL;
660 } else if (anon_vma_fork(tmp, mpnt))
661 goto fail_nomem_anon_vma_fork;
662 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
663 file = tmp->vm_file;
664 if (file) {
665 struct address_space *mapping = file->f_mapping;
666
667 get_file(file);
668 i_mmap_lock_write(mapping);
669 if (tmp->vm_flags & VM_SHARED)
670 mapping_allow_writable(mapping);
671 flush_dcache_mmap_lock(mapping);
672 /* insert tmp into the share list, just after mpnt */
673 vma_interval_tree_insert_after(tmp, mpnt,
674 &mapping->i_mmap);
675 flush_dcache_mmap_unlock(mapping);
676 i_mmap_unlock_write(mapping);
677 }
678
679 /*
680 * Copy/update hugetlb private vma information.
681 */
682 if (is_vm_hugetlb_page(tmp))
683 hugetlb_dup_vma_private(tmp);
684
685 /* Link the vma into the MT */
686 mas.index = tmp->vm_start;
687 mas.last = tmp->vm_end - 1;
688 mas_store(&mas, tmp);
689 if (mas_is_err(&mas))
690 goto fail_nomem_mas_store;
691
692 mm->map_count++;
693 if (!(tmp->vm_flags & VM_WIPEONFORK))
694 retval = copy_page_range(tmp, mpnt);
695
696 if (tmp->vm_ops && tmp->vm_ops->open)
697 tmp->vm_ops->open(tmp);
698
699 if (retval)
700 goto loop_out;
701 }
702 /* a new mm has just been created */
703 retval = arch_dup_mmap(oldmm, mm);
704loop_out:
705 mas_destroy(&mas);
706out:
707 mmap_write_unlock(mm);
708 flush_tlb_mm(oldmm);
709 mmap_write_unlock(oldmm);
710 dup_userfaultfd_complete(&uf);
711fail_uprobe_end:
712 uprobe_end_dup_mmap();
713 return retval;
714
715fail_nomem_mas_store:
716 unlink_anon_vmas(tmp);
717fail_nomem_anon_vma_fork:
718 mpol_put(vma_policy(tmp));
719fail_nomem_policy:
720 vm_area_free(tmp);
721fail_nomem:
722 retval = -ENOMEM;
723 vm_unacct_memory(charge);
724 goto loop_out;
725}
726
727static inline int mm_alloc_pgd(struct mm_struct *mm)
728{
729 mm->pgd = pgd_alloc(mm);
730 if (unlikely(!mm->pgd))
731 return -ENOMEM;
732 return 0;
733}
734
735static inline void mm_free_pgd(struct mm_struct *mm)
736{
737 pgd_free(mm, mm->pgd);
738}
739#else
740static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
741{
742 mmap_write_lock(oldmm);
743 dup_mm_exe_file(mm, oldmm);
744 mmap_write_unlock(oldmm);
745 return 0;
746}
747#define mm_alloc_pgd(mm) (0)
748#define mm_free_pgd(mm)
749#endif /* CONFIG_MMU */
750
751static void check_mm(struct mm_struct *mm)
752{
753 int i;
754
755 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
756 "Please make sure 'struct resident_page_types[]' is updated as well");
757
758 for (i = 0; i < NR_MM_COUNTERS; i++) {
759 long x = percpu_counter_sum(&mm->rss_stat[i]);
760
761 if (likely(!x))
762 continue;
763
764 /* Making sure this is not due to race with CPU offlining. */
765 x = percpu_counter_sum_all(&mm->rss_stat[i]);
766 if (unlikely(x))
767 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
768 mm, resident_page_types[i], x);
769 }
770
771 if (mm_pgtables_bytes(mm))
772 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
773 mm_pgtables_bytes(mm));
774
775#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
776 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
777#endif
778}
779
780#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
781#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
782
783/*
784 * Called when the last reference to the mm
785 * is dropped: either by a lazy thread or by
786 * mmput. Free the page directory and the mm.
787 */
788void __mmdrop(struct mm_struct *mm)
789{
790 int i;
791
792 BUG_ON(mm == &init_mm);
793 WARN_ON_ONCE(mm == current->mm);
794 WARN_ON_ONCE(mm == current->active_mm);
795 mm_free_pgd(mm);
796 destroy_context(mm);
797 mmu_notifier_subscriptions_destroy(mm);
798 check_mm(mm);
799 put_user_ns(mm->user_ns);
800 mm_pasid_drop(mm);
801
802 for (i = 0; i < NR_MM_COUNTERS; i++)
803 percpu_counter_destroy(&mm->rss_stat[i]);
804 free_mm(mm);
805}
806EXPORT_SYMBOL_GPL(__mmdrop);
807
808static void mmdrop_async_fn(struct work_struct *work)
809{
810 struct mm_struct *mm;
811
812 mm = container_of(work, struct mm_struct, async_put_work);
813 __mmdrop(mm);
814}
815
816static void mmdrop_async(struct mm_struct *mm)
817{
818 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
819 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
820 schedule_work(&mm->async_put_work);
821 }
822}
823
824static inline void free_signal_struct(struct signal_struct *sig)
825{
826 taskstats_tgid_free(sig);
827 sched_autogroup_exit(sig);
828 /*
829 * __mmdrop is not safe to call from softirq context on x86 due to
830 * pgd_dtor so postpone it to the async context
831 */
832 if (sig->oom_mm)
833 mmdrop_async(sig->oom_mm);
834 kmem_cache_free(signal_cachep, sig);
835}
836
837static inline void put_signal_struct(struct signal_struct *sig)
838{
839 if (refcount_dec_and_test(&sig->sigcnt))
840 free_signal_struct(sig);
841}
842
843void __put_task_struct(struct task_struct *tsk)
844{
845 WARN_ON(!tsk->exit_state);
846 WARN_ON(refcount_read(&tsk->usage));
847 WARN_ON(tsk == current);
848
849 io_uring_free(tsk);
850 cgroup_free(tsk);
851 task_numa_free(tsk, true);
852 security_task_free(tsk);
853 bpf_task_storage_free(tsk);
854 exit_creds(tsk);
855 delayacct_tsk_free(tsk);
856 put_signal_struct(tsk->signal);
857 sched_core_free(tsk);
858 free_task(tsk);
859}
860EXPORT_SYMBOL_GPL(__put_task_struct);
861
862void __init __weak arch_task_cache_init(void) { }
863
864/*
865 * set_max_threads
866 */
867static void set_max_threads(unsigned int max_threads_suggested)
868{
869 u64 threads;
870 unsigned long nr_pages = totalram_pages();
871
872 /*
873 * The number of threads shall be limited such that the thread
874 * structures may only consume a small part of the available memory.
875 */
876 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
877 threads = MAX_THREADS;
878 else
879 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
880 (u64) THREAD_SIZE * 8UL);
881
882 if (threads > max_threads_suggested)
883 threads = max_threads_suggested;
884
885 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
886}
887
888#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
889/* Initialized by the architecture: */
890int arch_task_struct_size __read_mostly;
891#endif
892
893#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
894static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
895{
896 /* Fetch thread_struct whitelist for the architecture. */
897 arch_thread_struct_whitelist(offset, size);
898
899 /*
900 * Handle zero-sized whitelist or empty thread_struct, otherwise
901 * adjust offset to position of thread_struct in task_struct.
902 */
903 if (unlikely(*size == 0))
904 *offset = 0;
905 else
906 *offset += offsetof(struct task_struct, thread);
907}
908#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
909
910void __init fork_init(void)
911{
912 int i;
913#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
914#ifndef ARCH_MIN_TASKALIGN
915#define ARCH_MIN_TASKALIGN 0
916#endif
917 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
918 unsigned long useroffset, usersize;
919
920 /* create a slab on which task_structs can be allocated */
921 task_struct_whitelist(&useroffset, &usersize);
922 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
923 arch_task_struct_size, align,
924 SLAB_PANIC|SLAB_ACCOUNT,
925 useroffset, usersize, NULL);
926#endif
927
928 /* do the arch specific task caches init */
929 arch_task_cache_init();
930
931 set_max_threads(MAX_THREADS);
932
933 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
934 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
935 init_task.signal->rlim[RLIMIT_SIGPENDING] =
936 init_task.signal->rlim[RLIMIT_NPROC];
937
938 for (i = 0; i < UCOUNT_COUNTS; i++)
939 init_user_ns.ucount_max[i] = max_threads/2;
940
941 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
942 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
943 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
944 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
945
946#ifdef CONFIG_VMAP_STACK
947 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
948 NULL, free_vm_stack_cache);
949#endif
950
951 scs_init();
952
953 lockdep_init_task(&init_task);
954 uprobes_init();
955}
956
957int __weak arch_dup_task_struct(struct task_struct *dst,
958 struct task_struct *src)
959{
960 *dst = *src;
961 return 0;
962}
963
964void set_task_stack_end_magic(struct task_struct *tsk)
965{
966 unsigned long *stackend;
967
968 stackend = end_of_stack(tsk);
969 *stackend = STACK_END_MAGIC; /* for overflow detection */
970}
971
972static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
973{
974 struct task_struct *tsk;
975 int err;
976
977 if (node == NUMA_NO_NODE)
978 node = tsk_fork_get_node(orig);
979 tsk = alloc_task_struct_node(node);
980 if (!tsk)
981 return NULL;
982
983 err = arch_dup_task_struct(tsk, orig);
984 if (err)
985 goto free_tsk;
986
987 err = alloc_thread_stack_node(tsk, node);
988 if (err)
989 goto free_tsk;
990
991#ifdef CONFIG_THREAD_INFO_IN_TASK
992 refcount_set(&tsk->stack_refcount, 1);
993#endif
994 account_kernel_stack(tsk, 1);
995
996 err = scs_prepare(tsk, node);
997 if (err)
998 goto free_stack;
999
1000#ifdef CONFIG_SECCOMP
1001 /*
1002 * We must handle setting up seccomp filters once we're under
1003 * the sighand lock in case orig has changed between now and
1004 * then. Until then, filter must be NULL to avoid messing up
1005 * the usage counts on the error path calling free_task.
1006 */
1007 tsk->seccomp.filter = NULL;
1008#endif
1009
1010 setup_thread_stack(tsk, orig);
1011 clear_user_return_notifier(tsk);
1012 clear_tsk_need_resched(tsk);
1013 set_task_stack_end_magic(tsk);
1014 clear_syscall_work_syscall_user_dispatch(tsk);
1015
1016#ifdef CONFIG_STACKPROTECTOR
1017 tsk->stack_canary = get_random_canary();
1018#endif
1019 if (orig->cpus_ptr == &orig->cpus_mask)
1020 tsk->cpus_ptr = &tsk->cpus_mask;
1021 dup_user_cpus_ptr(tsk, orig, node);
1022
1023 /*
1024 * One for the user space visible state that goes away when reaped.
1025 * One for the scheduler.
1026 */
1027 refcount_set(&tsk->rcu_users, 2);
1028 /* One for the rcu users */
1029 refcount_set(&tsk->usage, 1);
1030#ifdef CONFIG_BLK_DEV_IO_TRACE
1031 tsk->btrace_seq = 0;
1032#endif
1033 tsk->splice_pipe = NULL;
1034 tsk->task_frag.page = NULL;
1035 tsk->wake_q.next = NULL;
1036 tsk->worker_private = NULL;
1037
1038 kcov_task_init(tsk);
1039 kmsan_task_create(tsk);
1040 kmap_local_fork(tsk);
1041
1042#ifdef CONFIG_FAULT_INJECTION
1043 tsk->fail_nth = 0;
1044#endif
1045
1046#ifdef CONFIG_BLK_CGROUP
1047 tsk->throttle_queue = NULL;
1048 tsk->use_memdelay = 0;
1049#endif
1050
1051#ifdef CONFIG_IOMMU_SVA
1052 tsk->pasid_activated = 0;
1053#endif
1054
1055#ifdef CONFIG_MEMCG
1056 tsk->active_memcg = NULL;
1057#endif
1058
1059#ifdef CONFIG_CPU_SUP_INTEL
1060 tsk->reported_split_lock = 0;
1061#endif
1062
1063 return tsk;
1064
1065free_stack:
1066 exit_task_stack_account(tsk);
1067 free_thread_stack(tsk);
1068free_tsk:
1069 free_task_struct(tsk);
1070 return NULL;
1071}
1072
1073__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1074
1075static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1076
1077static int __init coredump_filter_setup(char *s)
1078{
1079 default_dump_filter =
1080 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1081 MMF_DUMP_FILTER_MASK;
1082 return 1;
1083}
1084
1085__setup("coredump_filter=", coredump_filter_setup);
1086
1087#include <linux/init_task.h>
1088
1089static void mm_init_aio(struct mm_struct *mm)
1090{
1091#ifdef CONFIG_AIO
1092 spin_lock_init(&mm->ioctx_lock);
1093 mm->ioctx_table = NULL;
1094#endif
1095}
1096
1097static __always_inline void mm_clear_owner(struct mm_struct *mm,
1098 struct task_struct *p)
1099{
1100#ifdef CONFIG_MEMCG
1101 if (mm->owner == p)
1102 WRITE_ONCE(mm->owner, NULL);
1103#endif
1104}
1105
1106static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1107{
1108#ifdef CONFIG_MEMCG
1109 mm->owner = p;
1110#endif
1111}
1112
1113static void mm_init_uprobes_state(struct mm_struct *mm)
1114{
1115#ifdef CONFIG_UPROBES
1116 mm->uprobes_state.xol_area = NULL;
1117#endif
1118}
1119
1120static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1121 struct user_namespace *user_ns)
1122{
1123 int i;
1124
1125 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1126 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1127 atomic_set(&mm->mm_users, 1);
1128 atomic_set(&mm->mm_count, 1);
1129 seqcount_init(&mm->write_protect_seq);
1130 mmap_init_lock(mm);
1131 INIT_LIST_HEAD(&mm->mmlist);
1132 mm_pgtables_bytes_init(mm);
1133 mm->map_count = 0;
1134 mm->locked_vm = 0;
1135 atomic64_set(&mm->pinned_vm, 0);
1136 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1137 spin_lock_init(&mm->page_table_lock);
1138 spin_lock_init(&mm->arg_lock);
1139 mm_init_cpumask(mm);
1140 mm_init_aio(mm);
1141 mm_init_owner(mm, p);
1142 mm_pasid_init(mm);
1143 RCU_INIT_POINTER(mm->exe_file, NULL);
1144 mmu_notifier_subscriptions_init(mm);
1145 init_tlb_flush_pending(mm);
1146#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1147 mm->pmd_huge_pte = NULL;
1148#endif
1149 mm_init_uprobes_state(mm);
1150 hugetlb_count_init(mm);
1151
1152 if (current->mm) {
1153 mm->flags = current->mm->flags & MMF_INIT_MASK;
1154 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1155 } else {
1156 mm->flags = default_dump_filter;
1157 mm->def_flags = 0;
1158 }
1159
1160 if (mm_alloc_pgd(mm))
1161 goto fail_nopgd;
1162
1163 if (init_new_context(p, mm))
1164 goto fail_nocontext;
1165
1166 for (i = 0; i < NR_MM_COUNTERS; i++)
1167 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1168 goto fail_pcpu;
1169
1170 mm->user_ns = get_user_ns(user_ns);
1171 lru_gen_init_mm(mm);
1172 return mm;
1173
1174fail_pcpu:
1175 while (i > 0)
1176 percpu_counter_destroy(&mm->rss_stat[--i]);
1177fail_nocontext:
1178 mm_free_pgd(mm);
1179fail_nopgd:
1180 free_mm(mm);
1181 return NULL;
1182}
1183
1184/*
1185 * Allocate and initialize an mm_struct.
1186 */
1187struct mm_struct *mm_alloc(void)
1188{
1189 struct mm_struct *mm;
1190
1191 mm = allocate_mm();
1192 if (!mm)
1193 return NULL;
1194
1195 memset(mm, 0, sizeof(*mm));
1196 return mm_init(mm, current, current_user_ns());
1197}
1198
1199static inline void __mmput(struct mm_struct *mm)
1200{
1201 VM_BUG_ON(atomic_read(&mm->mm_users));
1202
1203 uprobe_clear_state(mm);
1204 exit_aio(mm);
1205 ksm_exit(mm);
1206 khugepaged_exit(mm); /* must run before exit_mmap */
1207 exit_mmap(mm);
1208 mm_put_huge_zero_page(mm);
1209 set_mm_exe_file(mm, NULL);
1210 if (!list_empty(&mm->mmlist)) {
1211 spin_lock(&mmlist_lock);
1212 list_del(&mm->mmlist);
1213 spin_unlock(&mmlist_lock);
1214 }
1215 if (mm->binfmt)
1216 module_put(mm->binfmt->module);
1217 lru_gen_del_mm(mm);
1218 mmdrop(mm);
1219}
1220
1221/*
1222 * Decrement the use count and release all resources for an mm.
1223 */
1224void mmput(struct mm_struct *mm)
1225{
1226 might_sleep();
1227
1228 if (atomic_dec_and_test(&mm->mm_users))
1229 __mmput(mm);
1230}
1231EXPORT_SYMBOL_GPL(mmput);
1232
1233#ifdef CONFIG_MMU
1234static void mmput_async_fn(struct work_struct *work)
1235{
1236 struct mm_struct *mm = container_of(work, struct mm_struct,
1237 async_put_work);
1238
1239 __mmput(mm);
1240}
1241
1242void mmput_async(struct mm_struct *mm)
1243{
1244 if (atomic_dec_and_test(&mm->mm_users)) {
1245 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1246 schedule_work(&mm->async_put_work);
1247 }
1248}
1249EXPORT_SYMBOL_GPL(mmput_async);
1250#endif
1251
1252/**
1253 * set_mm_exe_file - change a reference to the mm's executable file
1254 *
1255 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1256 *
1257 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1258 * invocations: in mmput() nobody alive left, in execve task is single
1259 * threaded.
1260 *
1261 * Can only fail if new_exe_file != NULL.
1262 */
1263int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1264{
1265 struct file *old_exe_file;
1266
1267 /*
1268 * It is safe to dereference the exe_file without RCU as
1269 * this function is only called if nobody else can access
1270 * this mm -- see comment above for justification.
1271 */
1272 old_exe_file = rcu_dereference_raw(mm->exe_file);
1273
1274 if (new_exe_file) {
1275 /*
1276 * We expect the caller (i.e., sys_execve) to already denied
1277 * write access, so this is unlikely to fail.
1278 */
1279 if (unlikely(deny_write_access(new_exe_file)))
1280 return -EACCES;
1281 get_file(new_exe_file);
1282 }
1283 rcu_assign_pointer(mm->exe_file, new_exe_file);
1284 if (old_exe_file) {
1285 allow_write_access(old_exe_file);
1286 fput(old_exe_file);
1287 }
1288 return 0;
1289}
1290
1291/**
1292 * replace_mm_exe_file - replace a reference to the mm's executable file
1293 *
1294 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1295 * dealing with concurrent invocation and without grabbing the mmap lock in
1296 * write mode.
1297 *
1298 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1299 */
1300int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1301{
1302 struct vm_area_struct *vma;
1303 struct file *old_exe_file;
1304 int ret = 0;
1305
1306 /* Forbid mm->exe_file change if old file still mapped. */
1307 old_exe_file = get_mm_exe_file(mm);
1308 if (old_exe_file) {
1309 VMA_ITERATOR(vmi, mm, 0);
1310 mmap_read_lock(mm);
1311 for_each_vma(vmi, vma) {
1312 if (!vma->vm_file)
1313 continue;
1314 if (path_equal(&vma->vm_file->f_path,
1315 &old_exe_file->f_path)) {
1316 ret = -EBUSY;
1317 break;
1318 }
1319 }
1320 mmap_read_unlock(mm);
1321 fput(old_exe_file);
1322 if (ret)
1323 return ret;
1324 }
1325
1326 /* set the new file, lockless */
1327 ret = deny_write_access(new_exe_file);
1328 if (ret)
1329 return -EACCES;
1330 get_file(new_exe_file);
1331
1332 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1333 if (old_exe_file) {
1334 /*
1335 * Don't race with dup_mmap() getting the file and disallowing
1336 * write access while someone might open the file writable.
1337 */
1338 mmap_read_lock(mm);
1339 allow_write_access(old_exe_file);
1340 fput(old_exe_file);
1341 mmap_read_unlock(mm);
1342 }
1343 return 0;
1344}
1345
1346/**
1347 * get_mm_exe_file - acquire a reference to the mm's executable file
1348 *
1349 * Returns %NULL if mm has no associated executable file.
1350 * User must release file via fput().
1351 */
1352struct file *get_mm_exe_file(struct mm_struct *mm)
1353{
1354 struct file *exe_file;
1355
1356 rcu_read_lock();
1357 exe_file = rcu_dereference(mm->exe_file);
1358 if (exe_file && !get_file_rcu(exe_file))
1359 exe_file = NULL;
1360 rcu_read_unlock();
1361 return exe_file;
1362}
1363
1364/**
1365 * get_task_exe_file - acquire a reference to the task's executable file
1366 *
1367 * Returns %NULL if task's mm (if any) has no associated executable file or
1368 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1369 * User must release file via fput().
1370 */
1371struct file *get_task_exe_file(struct task_struct *task)
1372{
1373 struct file *exe_file = NULL;
1374 struct mm_struct *mm;
1375
1376 task_lock(task);
1377 mm = task->mm;
1378 if (mm) {
1379 if (!(task->flags & PF_KTHREAD))
1380 exe_file = get_mm_exe_file(mm);
1381 }
1382 task_unlock(task);
1383 return exe_file;
1384}
1385
1386/**
1387 * get_task_mm - acquire a reference to the task's mm
1388 *
1389 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1390 * this kernel workthread has transiently adopted a user mm with use_mm,
1391 * to do its AIO) is not set and if so returns a reference to it, after
1392 * bumping up the use count. User must release the mm via mmput()
1393 * after use. Typically used by /proc and ptrace.
1394 */
1395struct mm_struct *get_task_mm(struct task_struct *task)
1396{
1397 struct mm_struct *mm;
1398
1399 task_lock(task);
1400 mm = task->mm;
1401 if (mm) {
1402 if (task->flags & PF_KTHREAD)
1403 mm = NULL;
1404 else
1405 mmget(mm);
1406 }
1407 task_unlock(task);
1408 return mm;
1409}
1410EXPORT_SYMBOL_GPL(get_task_mm);
1411
1412struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1413{
1414 struct mm_struct *mm;
1415 int err;
1416
1417 err = down_read_killable(&task->signal->exec_update_lock);
1418 if (err)
1419 return ERR_PTR(err);
1420
1421 mm = get_task_mm(task);
1422 if (mm && mm != current->mm &&
1423 !ptrace_may_access(task, mode)) {
1424 mmput(mm);
1425 mm = ERR_PTR(-EACCES);
1426 }
1427 up_read(&task->signal->exec_update_lock);
1428
1429 return mm;
1430}
1431
1432static void complete_vfork_done(struct task_struct *tsk)
1433{
1434 struct completion *vfork;
1435
1436 task_lock(tsk);
1437 vfork = tsk->vfork_done;
1438 if (likely(vfork)) {
1439 tsk->vfork_done = NULL;
1440 complete(vfork);
1441 }
1442 task_unlock(tsk);
1443}
1444
1445static int wait_for_vfork_done(struct task_struct *child,
1446 struct completion *vfork)
1447{
1448 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1449 int killed;
1450
1451 cgroup_enter_frozen();
1452 killed = wait_for_completion_state(vfork, state);
1453 cgroup_leave_frozen(false);
1454
1455 if (killed) {
1456 task_lock(child);
1457 child->vfork_done = NULL;
1458 task_unlock(child);
1459 }
1460
1461 put_task_struct(child);
1462 return killed;
1463}
1464
1465/* Please note the differences between mmput and mm_release.
1466 * mmput is called whenever we stop holding onto a mm_struct,
1467 * error success whatever.
1468 *
1469 * mm_release is called after a mm_struct has been removed
1470 * from the current process.
1471 *
1472 * This difference is important for error handling, when we
1473 * only half set up a mm_struct for a new process and need to restore
1474 * the old one. Because we mmput the new mm_struct before
1475 * restoring the old one. . .
1476 * Eric Biederman 10 January 1998
1477 */
1478static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1479{
1480 uprobe_free_utask(tsk);
1481
1482 /* Get rid of any cached register state */
1483 deactivate_mm(tsk, mm);
1484
1485 /*
1486 * Signal userspace if we're not exiting with a core dump
1487 * because we want to leave the value intact for debugging
1488 * purposes.
1489 */
1490 if (tsk->clear_child_tid) {
1491 if (atomic_read(&mm->mm_users) > 1) {
1492 /*
1493 * We don't check the error code - if userspace has
1494 * not set up a proper pointer then tough luck.
1495 */
1496 put_user(0, tsk->clear_child_tid);
1497 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1498 1, NULL, NULL, 0, 0);
1499 }
1500 tsk->clear_child_tid = NULL;
1501 }
1502
1503 /*
1504 * All done, finally we can wake up parent and return this mm to him.
1505 * Also kthread_stop() uses this completion for synchronization.
1506 */
1507 if (tsk->vfork_done)
1508 complete_vfork_done(tsk);
1509}
1510
1511void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1512{
1513 futex_exit_release(tsk);
1514 mm_release(tsk, mm);
1515}
1516
1517void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1518{
1519 futex_exec_release(tsk);
1520 mm_release(tsk, mm);
1521}
1522
1523/**
1524 * dup_mm() - duplicates an existing mm structure
1525 * @tsk: the task_struct with which the new mm will be associated.
1526 * @oldmm: the mm to duplicate.
1527 *
1528 * Allocates a new mm structure and duplicates the provided @oldmm structure
1529 * content into it.
1530 *
1531 * Return: the duplicated mm or NULL on failure.
1532 */
1533static struct mm_struct *dup_mm(struct task_struct *tsk,
1534 struct mm_struct *oldmm)
1535{
1536 struct mm_struct *mm;
1537 int err;
1538
1539 mm = allocate_mm();
1540 if (!mm)
1541 goto fail_nomem;
1542
1543 memcpy(mm, oldmm, sizeof(*mm));
1544
1545 if (!mm_init(mm, tsk, mm->user_ns))
1546 goto fail_nomem;
1547
1548 err = dup_mmap(mm, oldmm);
1549 if (err)
1550 goto free_pt;
1551
1552 mm->hiwater_rss = get_mm_rss(mm);
1553 mm->hiwater_vm = mm->total_vm;
1554
1555 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1556 goto free_pt;
1557
1558 return mm;
1559
1560free_pt:
1561 /* don't put binfmt in mmput, we haven't got module yet */
1562 mm->binfmt = NULL;
1563 mm_init_owner(mm, NULL);
1564 mmput(mm);
1565
1566fail_nomem:
1567 return NULL;
1568}
1569
1570static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1571{
1572 struct mm_struct *mm, *oldmm;
1573
1574 tsk->min_flt = tsk->maj_flt = 0;
1575 tsk->nvcsw = tsk->nivcsw = 0;
1576#ifdef CONFIG_DETECT_HUNG_TASK
1577 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1578 tsk->last_switch_time = 0;
1579#endif
1580
1581 tsk->mm = NULL;
1582 tsk->active_mm = NULL;
1583
1584 /*
1585 * Are we cloning a kernel thread?
1586 *
1587 * We need to steal a active VM for that..
1588 */
1589 oldmm = current->mm;
1590 if (!oldmm)
1591 return 0;
1592
1593 if (clone_flags & CLONE_VM) {
1594 mmget(oldmm);
1595 mm = oldmm;
1596 } else {
1597 mm = dup_mm(tsk, current->mm);
1598 if (!mm)
1599 return -ENOMEM;
1600 }
1601
1602 tsk->mm = mm;
1603 tsk->active_mm = mm;
1604 return 0;
1605}
1606
1607static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1608{
1609 struct fs_struct *fs = current->fs;
1610 if (clone_flags & CLONE_FS) {
1611 /* tsk->fs is already what we want */
1612 spin_lock(&fs->lock);
1613 if (fs->in_exec) {
1614 spin_unlock(&fs->lock);
1615 return -EAGAIN;
1616 }
1617 fs->users++;
1618 spin_unlock(&fs->lock);
1619 return 0;
1620 }
1621 tsk->fs = copy_fs_struct(fs);
1622 if (!tsk->fs)
1623 return -ENOMEM;
1624 return 0;
1625}
1626
1627static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1628{
1629 struct files_struct *oldf, *newf;
1630 int error = 0;
1631
1632 /*
1633 * A background process may not have any files ...
1634 */
1635 oldf = current->files;
1636 if (!oldf)
1637 goto out;
1638
1639 if (clone_flags & CLONE_FILES) {
1640 atomic_inc(&oldf->count);
1641 goto out;
1642 }
1643
1644 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1645 if (!newf)
1646 goto out;
1647
1648 tsk->files = newf;
1649 error = 0;
1650out:
1651 return error;
1652}
1653
1654static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1655{
1656 struct sighand_struct *sig;
1657
1658 if (clone_flags & CLONE_SIGHAND) {
1659 refcount_inc(¤t->sighand->count);
1660 return 0;
1661 }
1662 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1663 RCU_INIT_POINTER(tsk->sighand, sig);
1664 if (!sig)
1665 return -ENOMEM;
1666
1667 refcount_set(&sig->count, 1);
1668 spin_lock_irq(¤t->sighand->siglock);
1669 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1670 spin_unlock_irq(¤t->sighand->siglock);
1671
1672 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1673 if (clone_flags & CLONE_CLEAR_SIGHAND)
1674 flush_signal_handlers(tsk, 0);
1675
1676 return 0;
1677}
1678
1679void __cleanup_sighand(struct sighand_struct *sighand)
1680{
1681 if (refcount_dec_and_test(&sighand->count)) {
1682 signalfd_cleanup(sighand);
1683 /*
1684 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1685 * without an RCU grace period, see __lock_task_sighand().
1686 */
1687 kmem_cache_free(sighand_cachep, sighand);
1688 }
1689}
1690
1691/*
1692 * Initialize POSIX timer handling for a thread group.
1693 */
1694static void posix_cpu_timers_init_group(struct signal_struct *sig)
1695{
1696 struct posix_cputimers *pct = &sig->posix_cputimers;
1697 unsigned long cpu_limit;
1698
1699 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1700 posix_cputimers_group_init(pct, cpu_limit);
1701}
1702
1703static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1704{
1705 struct signal_struct *sig;
1706
1707 if (clone_flags & CLONE_THREAD)
1708 return 0;
1709
1710 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1711 tsk->signal = sig;
1712 if (!sig)
1713 return -ENOMEM;
1714
1715 sig->nr_threads = 1;
1716 sig->quick_threads = 1;
1717 atomic_set(&sig->live, 1);
1718 refcount_set(&sig->sigcnt, 1);
1719
1720 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1721 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1722 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1723
1724 init_waitqueue_head(&sig->wait_chldexit);
1725 sig->curr_target = tsk;
1726 init_sigpending(&sig->shared_pending);
1727 INIT_HLIST_HEAD(&sig->multiprocess);
1728 seqlock_init(&sig->stats_lock);
1729 prev_cputime_init(&sig->prev_cputime);
1730
1731#ifdef CONFIG_POSIX_TIMERS
1732 INIT_LIST_HEAD(&sig->posix_timers);
1733 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1734 sig->real_timer.function = it_real_fn;
1735#endif
1736
1737 task_lock(current->group_leader);
1738 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1739 task_unlock(current->group_leader);
1740
1741 posix_cpu_timers_init_group(sig);
1742
1743 tty_audit_fork(sig);
1744 sched_autogroup_fork(sig);
1745
1746 sig->oom_score_adj = current->signal->oom_score_adj;
1747 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1748
1749 mutex_init(&sig->cred_guard_mutex);
1750 init_rwsem(&sig->exec_update_lock);
1751
1752 return 0;
1753}
1754
1755static void copy_seccomp(struct task_struct *p)
1756{
1757#ifdef CONFIG_SECCOMP
1758 /*
1759 * Must be called with sighand->lock held, which is common to
1760 * all threads in the group. Holding cred_guard_mutex is not
1761 * needed because this new task is not yet running and cannot
1762 * be racing exec.
1763 */
1764 assert_spin_locked(¤t->sighand->siglock);
1765
1766 /* Ref-count the new filter user, and assign it. */
1767 get_seccomp_filter(current);
1768 p->seccomp = current->seccomp;
1769
1770 /*
1771 * Explicitly enable no_new_privs here in case it got set
1772 * between the task_struct being duplicated and holding the
1773 * sighand lock. The seccomp state and nnp must be in sync.
1774 */
1775 if (task_no_new_privs(current))
1776 task_set_no_new_privs(p);
1777
1778 /*
1779 * If the parent gained a seccomp mode after copying thread
1780 * flags and between before we held the sighand lock, we have
1781 * to manually enable the seccomp thread flag here.
1782 */
1783 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1784 set_task_syscall_work(p, SECCOMP);
1785#endif
1786}
1787
1788SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1789{
1790 current->clear_child_tid = tidptr;
1791
1792 return task_pid_vnr(current);
1793}
1794
1795static void rt_mutex_init_task(struct task_struct *p)
1796{
1797 raw_spin_lock_init(&p->pi_lock);
1798#ifdef CONFIG_RT_MUTEXES
1799 p->pi_waiters = RB_ROOT_CACHED;
1800 p->pi_top_task = NULL;
1801 p->pi_blocked_on = NULL;
1802#endif
1803}
1804
1805static inline void init_task_pid_links(struct task_struct *task)
1806{
1807 enum pid_type type;
1808
1809 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1810 INIT_HLIST_NODE(&task->pid_links[type]);
1811}
1812
1813static inline void
1814init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1815{
1816 if (type == PIDTYPE_PID)
1817 task->thread_pid = pid;
1818 else
1819 task->signal->pids[type] = pid;
1820}
1821
1822static inline void rcu_copy_process(struct task_struct *p)
1823{
1824#ifdef CONFIG_PREEMPT_RCU
1825 p->rcu_read_lock_nesting = 0;
1826 p->rcu_read_unlock_special.s = 0;
1827 p->rcu_blocked_node = NULL;
1828 INIT_LIST_HEAD(&p->rcu_node_entry);
1829#endif /* #ifdef CONFIG_PREEMPT_RCU */
1830#ifdef CONFIG_TASKS_RCU
1831 p->rcu_tasks_holdout = false;
1832 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1833 p->rcu_tasks_idle_cpu = -1;
1834#endif /* #ifdef CONFIG_TASKS_RCU */
1835#ifdef CONFIG_TASKS_TRACE_RCU
1836 p->trc_reader_nesting = 0;
1837 p->trc_reader_special.s = 0;
1838 INIT_LIST_HEAD(&p->trc_holdout_list);
1839 INIT_LIST_HEAD(&p->trc_blkd_node);
1840#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1841}
1842
1843struct pid *pidfd_pid(const struct file *file)
1844{
1845 if (file->f_op == &pidfd_fops)
1846 return file->private_data;
1847
1848 return ERR_PTR(-EBADF);
1849}
1850
1851static int pidfd_release(struct inode *inode, struct file *file)
1852{
1853 struct pid *pid = file->private_data;
1854
1855 file->private_data = NULL;
1856 put_pid(pid);
1857 return 0;
1858}
1859
1860#ifdef CONFIG_PROC_FS
1861/**
1862 * pidfd_show_fdinfo - print information about a pidfd
1863 * @m: proc fdinfo file
1864 * @f: file referencing a pidfd
1865 *
1866 * Pid:
1867 * This function will print the pid that a given pidfd refers to in the
1868 * pid namespace of the procfs instance.
1869 * If the pid namespace of the process is not a descendant of the pid
1870 * namespace of the procfs instance 0 will be shown as its pid. This is
1871 * similar to calling getppid() on a process whose parent is outside of
1872 * its pid namespace.
1873 *
1874 * NSpid:
1875 * If pid namespaces are supported then this function will also print
1876 * the pid of a given pidfd refers to for all descendant pid namespaces
1877 * starting from the current pid namespace of the instance, i.e. the
1878 * Pid field and the first entry in the NSpid field will be identical.
1879 * If the pid namespace of the process is not a descendant of the pid
1880 * namespace of the procfs instance 0 will be shown as its first NSpid
1881 * entry and no others will be shown.
1882 * Note that this differs from the Pid and NSpid fields in
1883 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1884 * the pid namespace of the procfs instance. The difference becomes
1885 * obvious when sending around a pidfd between pid namespaces from a
1886 * different branch of the tree, i.e. where no ancestral relation is
1887 * present between the pid namespaces:
1888 * - create two new pid namespaces ns1 and ns2 in the initial pid
1889 * namespace (also take care to create new mount namespaces in the
1890 * new pid namespace and mount procfs)
1891 * - create a process with a pidfd in ns1
1892 * - send pidfd from ns1 to ns2
1893 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1894 * have exactly one entry, which is 0
1895 */
1896static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1897{
1898 struct pid *pid = f->private_data;
1899 struct pid_namespace *ns;
1900 pid_t nr = -1;
1901
1902 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1903 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1904 nr = pid_nr_ns(pid, ns);
1905 }
1906
1907 seq_put_decimal_ll(m, "Pid:\t", nr);
1908
1909#ifdef CONFIG_PID_NS
1910 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1911 if (nr > 0) {
1912 int i;
1913
1914 /* If nr is non-zero it means that 'pid' is valid and that
1915 * ns, i.e. the pid namespace associated with the procfs
1916 * instance, is in the pid namespace hierarchy of pid.
1917 * Start at one below the already printed level.
1918 */
1919 for (i = ns->level + 1; i <= pid->level; i++)
1920 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1921 }
1922#endif
1923 seq_putc(m, '\n');
1924}
1925#endif
1926
1927/*
1928 * Poll support for process exit notification.
1929 */
1930static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1931{
1932 struct pid *pid = file->private_data;
1933 __poll_t poll_flags = 0;
1934
1935 poll_wait(file, &pid->wait_pidfd, pts);
1936
1937 /*
1938 * Inform pollers only when the whole thread group exits.
1939 * If the thread group leader exits before all other threads in the
1940 * group, then poll(2) should block, similar to the wait(2) family.
1941 */
1942 if (thread_group_exited(pid))
1943 poll_flags = EPOLLIN | EPOLLRDNORM;
1944
1945 return poll_flags;
1946}
1947
1948const struct file_operations pidfd_fops = {
1949 .release = pidfd_release,
1950 .poll = pidfd_poll,
1951#ifdef CONFIG_PROC_FS
1952 .show_fdinfo = pidfd_show_fdinfo,
1953#endif
1954};
1955
1956static void __delayed_free_task(struct rcu_head *rhp)
1957{
1958 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1959
1960 free_task(tsk);
1961}
1962
1963static __always_inline void delayed_free_task(struct task_struct *tsk)
1964{
1965 if (IS_ENABLED(CONFIG_MEMCG))
1966 call_rcu(&tsk->rcu, __delayed_free_task);
1967 else
1968 free_task(tsk);
1969}
1970
1971static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1972{
1973 /* Skip if kernel thread */
1974 if (!tsk->mm)
1975 return;
1976
1977 /* Skip if spawning a thread or using vfork */
1978 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1979 return;
1980
1981 /* We need to synchronize with __set_oom_adj */
1982 mutex_lock(&oom_adj_mutex);
1983 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1984 /* Update the values in case they were changed after copy_signal */
1985 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1986 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1987 mutex_unlock(&oom_adj_mutex);
1988}
1989
1990#ifdef CONFIG_RV
1991static void rv_task_fork(struct task_struct *p)
1992{
1993 int i;
1994
1995 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1996 p->rv[i].da_mon.monitoring = false;
1997}
1998#else
1999#define rv_task_fork(p) do {} while (0)
2000#endif
2001
2002/*
2003 * This creates a new process as a copy of the old one,
2004 * but does not actually start it yet.
2005 *
2006 * It copies the registers, and all the appropriate
2007 * parts of the process environment (as per the clone
2008 * flags). The actual kick-off is left to the caller.
2009 */
2010static __latent_entropy struct task_struct *copy_process(
2011 struct pid *pid,
2012 int trace,
2013 int node,
2014 struct kernel_clone_args *args)
2015{
2016 int pidfd = -1, retval;
2017 struct task_struct *p;
2018 struct multiprocess_signals delayed;
2019 struct file *pidfile = NULL;
2020 const u64 clone_flags = args->flags;
2021 struct nsproxy *nsp = current->nsproxy;
2022
2023 /*
2024 * Don't allow sharing the root directory with processes in a different
2025 * namespace
2026 */
2027 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2028 return ERR_PTR(-EINVAL);
2029
2030 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2031 return ERR_PTR(-EINVAL);
2032
2033 /*
2034 * Thread groups must share signals as well, and detached threads
2035 * can only be started up within the thread group.
2036 */
2037 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2038 return ERR_PTR(-EINVAL);
2039
2040 /*
2041 * Shared signal handlers imply shared VM. By way of the above,
2042 * thread groups also imply shared VM. Blocking this case allows
2043 * for various simplifications in other code.
2044 */
2045 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2046 return ERR_PTR(-EINVAL);
2047
2048 /*
2049 * Siblings of global init remain as zombies on exit since they are
2050 * not reaped by their parent (swapper). To solve this and to avoid
2051 * multi-rooted process trees, prevent global and container-inits
2052 * from creating siblings.
2053 */
2054 if ((clone_flags & CLONE_PARENT) &&
2055 current->signal->flags & SIGNAL_UNKILLABLE)
2056 return ERR_PTR(-EINVAL);
2057
2058 /*
2059 * If the new process will be in a different pid or user namespace
2060 * do not allow it to share a thread group with the forking task.
2061 */
2062 if (clone_flags & CLONE_THREAD) {
2063 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2064 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2065 return ERR_PTR(-EINVAL);
2066 }
2067
2068 if (clone_flags & CLONE_PIDFD) {
2069 /*
2070 * - CLONE_DETACHED is blocked so that we can potentially
2071 * reuse it later for CLONE_PIDFD.
2072 * - CLONE_THREAD is blocked until someone really needs it.
2073 */
2074 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2075 return ERR_PTR(-EINVAL);
2076 }
2077
2078 /*
2079 * Force any signals received before this point to be delivered
2080 * before the fork happens. Collect up signals sent to multiple
2081 * processes that happen during the fork and delay them so that
2082 * they appear to happen after the fork.
2083 */
2084 sigemptyset(&delayed.signal);
2085 INIT_HLIST_NODE(&delayed.node);
2086
2087 spin_lock_irq(¤t->sighand->siglock);
2088 if (!(clone_flags & CLONE_THREAD))
2089 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2090 recalc_sigpending();
2091 spin_unlock_irq(¤t->sighand->siglock);
2092 retval = -ERESTARTNOINTR;
2093 if (task_sigpending(current))
2094 goto fork_out;
2095
2096 retval = -ENOMEM;
2097 p = dup_task_struct(current, node);
2098 if (!p)
2099 goto fork_out;
2100 p->flags &= ~PF_KTHREAD;
2101 if (args->kthread)
2102 p->flags |= PF_KTHREAD;
2103 if (args->io_thread) {
2104 /*
2105 * Mark us an IO worker, and block any signal that isn't
2106 * fatal or STOP
2107 */
2108 p->flags |= PF_IO_WORKER;
2109 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2110 }
2111
2112 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2113 /*
2114 * Clear TID on mm_release()?
2115 */
2116 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2117
2118 ftrace_graph_init_task(p);
2119
2120 rt_mutex_init_task(p);
2121
2122 lockdep_assert_irqs_enabled();
2123#ifdef CONFIG_PROVE_LOCKING
2124 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2125#endif
2126 retval = copy_creds(p, clone_flags);
2127 if (retval < 0)
2128 goto bad_fork_free;
2129
2130 retval = -EAGAIN;
2131 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2132 if (p->real_cred->user != INIT_USER &&
2133 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2134 goto bad_fork_cleanup_count;
2135 }
2136 current->flags &= ~PF_NPROC_EXCEEDED;
2137
2138 /*
2139 * If multiple threads are within copy_process(), then this check
2140 * triggers too late. This doesn't hurt, the check is only there
2141 * to stop root fork bombs.
2142 */
2143 retval = -EAGAIN;
2144 if (data_race(nr_threads >= max_threads))
2145 goto bad_fork_cleanup_count;
2146
2147 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2148 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2149 p->flags |= PF_FORKNOEXEC;
2150 INIT_LIST_HEAD(&p->children);
2151 INIT_LIST_HEAD(&p->sibling);
2152 rcu_copy_process(p);
2153 p->vfork_done = NULL;
2154 spin_lock_init(&p->alloc_lock);
2155
2156 init_sigpending(&p->pending);
2157
2158 p->utime = p->stime = p->gtime = 0;
2159#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2160 p->utimescaled = p->stimescaled = 0;
2161#endif
2162 prev_cputime_init(&p->prev_cputime);
2163
2164#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2165 seqcount_init(&p->vtime.seqcount);
2166 p->vtime.starttime = 0;
2167 p->vtime.state = VTIME_INACTIVE;
2168#endif
2169
2170#ifdef CONFIG_IO_URING
2171 p->io_uring = NULL;
2172#endif
2173
2174#if defined(SPLIT_RSS_COUNTING)
2175 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2176#endif
2177
2178 p->default_timer_slack_ns = current->timer_slack_ns;
2179
2180#ifdef CONFIG_PSI
2181 p->psi_flags = 0;
2182#endif
2183
2184 task_io_accounting_init(&p->ioac);
2185 acct_clear_integrals(p);
2186
2187 posix_cputimers_init(&p->posix_cputimers);
2188
2189 p->io_context = NULL;
2190 audit_set_context(p, NULL);
2191 cgroup_fork(p);
2192 if (args->kthread) {
2193 if (!set_kthread_struct(p))
2194 goto bad_fork_cleanup_delayacct;
2195 }
2196#ifdef CONFIG_NUMA
2197 p->mempolicy = mpol_dup(p->mempolicy);
2198 if (IS_ERR(p->mempolicy)) {
2199 retval = PTR_ERR(p->mempolicy);
2200 p->mempolicy = NULL;
2201 goto bad_fork_cleanup_delayacct;
2202 }
2203#endif
2204#ifdef CONFIG_CPUSETS
2205 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2206 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2207 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2208#endif
2209#ifdef CONFIG_TRACE_IRQFLAGS
2210 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2211 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2212 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2213 p->softirqs_enabled = 1;
2214 p->softirq_context = 0;
2215#endif
2216
2217 p->pagefault_disabled = 0;
2218
2219#ifdef CONFIG_LOCKDEP
2220 lockdep_init_task(p);
2221#endif
2222
2223#ifdef CONFIG_DEBUG_MUTEXES
2224 p->blocked_on = NULL; /* not blocked yet */
2225#endif
2226#ifdef CONFIG_BCACHE
2227 p->sequential_io = 0;
2228 p->sequential_io_avg = 0;
2229#endif
2230#ifdef CONFIG_BPF_SYSCALL
2231 RCU_INIT_POINTER(p->bpf_storage, NULL);
2232 p->bpf_ctx = NULL;
2233#endif
2234
2235 /* Perform scheduler related setup. Assign this task to a CPU. */
2236 retval = sched_fork(clone_flags, p);
2237 if (retval)
2238 goto bad_fork_cleanup_policy;
2239
2240 retval = perf_event_init_task(p, clone_flags);
2241 if (retval)
2242 goto bad_fork_cleanup_policy;
2243 retval = audit_alloc(p);
2244 if (retval)
2245 goto bad_fork_cleanup_perf;
2246 /* copy all the process information */
2247 shm_init_task(p);
2248 retval = security_task_alloc(p, clone_flags);
2249 if (retval)
2250 goto bad_fork_cleanup_audit;
2251 retval = copy_semundo(clone_flags, p);
2252 if (retval)
2253 goto bad_fork_cleanup_security;
2254 retval = copy_files(clone_flags, p);
2255 if (retval)
2256 goto bad_fork_cleanup_semundo;
2257 retval = copy_fs(clone_flags, p);
2258 if (retval)
2259 goto bad_fork_cleanup_files;
2260 retval = copy_sighand(clone_flags, p);
2261 if (retval)
2262 goto bad_fork_cleanup_fs;
2263 retval = copy_signal(clone_flags, p);
2264 if (retval)
2265 goto bad_fork_cleanup_sighand;
2266 retval = copy_mm(clone_flags, p);
2267 if (retval)
2268 goto bad_fork_cleanup_signal;
2269 retval = copy_namespaces(clone_flags, p);
2270 if (retval)
2271 goto bad_fork_cleanup_mm;
2272 retval = copy_io(clone_flags, p);
2273 if (retval)
2274 goto bad_fork_cleanup_namespaces;
2275 retval = copy_thread(p, args);
2276 if (retval)
2277 goto bad_fork_cleanup_io;
2278
2279 stackleak_task_init(p);
2280
2281 if (pid != &init_struct_pid) {
2282 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2283 args->set_tid_size);
2284 if (IS_ERR(pid)) {
2285 retval = PTR_ERR(pid);
2286 goto bad_fork_cleanup_thread;
2287 }
2288 }
2289
2290 /*
2291 * This has to happen after we've potentially unshared the file
2292 * descriptor table (so that the pidfd doesn't leak into the child
2293 * if the fd table isn't shared).
2294 */
2295 if (clone_flags & CLONE_PIDFD) {
2296 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2297 if (retval < 0)
2298 goto bad_fork_free_pid;
2299
2300 pidfd = retval;
2301
2302 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2303 O_RDWR | O_CLOEXEC);
2304 if (IS_ERR(pidfile)) {
2305 put_unused_fd(pidfd);
2306 retval = PTR_ERR(pidfile);
2307 goto bad_fork_free_pid;
2308 }
2309 get_pid(pid); /* held by pidfile now */
2310
2311 retval = put_user(pidfd, args->pidfd);
2312 if (retval)
2313 goto bad_fork_put_pidfd;
2314 }
2315
2316#ifdef CONFIG_BLOCK
2317 p->plug = NULL;
2318#endif
2319 futex_init_task(p);
2320
2321 /*
2322 * sigaltstack should be cleared when sharing the same VM
2323 */
2324 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2325 sas_ss_reset(p);
2326
2327 /*
2328 * Syscall tracing and stepping should be turned off in the
2329 * child regardless of CLONE_PTRACE.
2330 */
2331 user_disable_single_step(p);
2332 clear_task_syscall_work(p, SYSCALL_TRACE);
2333#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2334 clear_task_syscall_work(p, SYSCALL_EMU);
2335#endif
2336 clear_tsk_latency_tracing(p);
2337
2338 /* ok, now we should be set up.. */
2339 p->pid = pid_nr(pid);
2340 if (clone_flags & CLONE_THREAD) {
2341 p->group_leader = current->group_leader;
2342 p->tgid = current->tgid;
2343 } else {
2344 p->group_leader = p;
2345 p->tgid = p->pid;
2346 }
2347
2348 p->nr_dirtied = 0;
2349 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2350 p->dirty_paused_when = 0;
2351
2352 p->pdeath_signal = 0;
2353 INIT_LIST_HEAD(&p->thread_group);
2354 p->task_works = NULL;
2355 clear_posix_cputimers_work(p);
2356
2357#ifdef CONFIG_KRETPROBES
2358 p->kretprobe_instances.first = NULL;
2359#endif
2360#ifdef CONFIG_RETHOOK
2361 p->rethooks.first = NULL;
2362#endif
2363
2364 /*
2365 * Ensure that the cgroup subsystem policies allow the new process to be
2366 * forked. It should be noted that the new process's css_set can be changed
2367 * between here and cgroup_post_fork() if an organisation operation is in
2368 * progress.
2369 */
2370 retval = cgroup_can_fork(p, args);
2371 if (retval)
2372 goto bad_fork_put_pidfd;
2373
2374 /*
2375 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2376 * the new task on the correct runqueue. All this *before* the task
2377 * becomes visible.
2378 *
2379 * This isn't part of ->can_fork() because while the re-cloning is
2380 * cgroup specific, it unconditionally needs to place the task on a
2381 * runqueue.
2382 */
2383 sched_cgroup_fork(p, args);
2384
2385 /*
2386 * From this point on we must avoid any synchronous user-space
2387 * communication until we take the tasklist-lock. In particular, we do
2388 * not want user-space to be able to predict the process start-time by
2389 * stalling fork(2) after we recorded the start_time but before it is
2390 * visible to the system.
2391 */
2392
2393 p->start_time = ktime_get_ns();
2394 p->start_boottime = ktime_get_boottime_ns();
2395
2396 /*
2397 * Make it visible to the rest of the system, but dont wake it up yet.
2398 * Need tasklist lock for parent etc handling!
2399 */
2400 write_lock_irq(&tasklist_lock);
2401
2402 /* CLONE_PARENT re-uses the old parent */
2403 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2404 p->real_parent = current->real_parent;
2405 p->parent_exec_id = current->parent_exec_id;
2406 if (clone_flags & CLONE_THREAD)
2407 p->exit_signal = -1;
2408 else
2409 p->exit_signal = current->group_leader->exit_signal;
2410 } else {
2411 p->real_parent = current;
2412 p->parent_exec_id = current->self_exec_id;
2413 p->exit_signal = args->exit_signal;
2414 }
2415
2416 klp_copy_process(p);
2417
2418 sched_core_fork(p);
2419
2420 spin_lock(¤t->sighand->siglock);
2421
2422 rv_task_fork(p);
2423
2424 rseq_fork(p, clone_flags);
2425
2426 /* Don't start children in a dying pid namespace */
2427 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2428 retval = -ENOMEM;
2429 goto bad_fork_cancel_cgroup;
2430 }
2431
2432 /* Let kill terminate clone/fork in the middle */
2433 if (fatal_signal_pending(current)) {
2434 retval = -EINTR;
2435 goto bad_fork_cancel_cgroup;
2436 }
2437
2438 /* No more failure paths after this point. */
2439
2440 /*
2441 * Copy seccomp details explicitly here, in case they were changed
2442 * before holding sighand lock.
2443 */
2444 copy_seccomp(p);
2445
2446 init_task_pid_links(p);
2447 if (likely(p->pid)) {
2448 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2449
2450 init_task_pid(p, PIDTYPE_PID, pid);
2451 if (thread_group_leader(p)) {
2452 init_task_pid(p, PIDTYPE_TGID, pid);
2453 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2454 init_task_pid(p, PIDTYPE_SID, task_session(current));
2455
2456 if (is_child_reaper(pid)) {
2457 ns_of_pid(pid)->child_reaper = p;
2458 p->signal->flags |= SIGNAL_UNKILLABLE;
2459 }
2460 p->signal->shared_pending.signal = delayed.signal;
2461 p->signal->tty = tty_kref_get(current->signal->tty);
2462 /*
2463 * Inherit has_child_subreaper flag under the same
2464 * tasklist_lock with adding child to the process tree
2465 * for propagate_has_child_subreaper optimization.
2466 */
2467 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2468 p->real_parent->signal->is_child_subreaper;
2469 list_add_tail(&p->sibling, &p->real_parent->children);
2470 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2471 attach_pid(p, PIDTYPE_TGID);
2472 attach_pid(p, PIDTYPE_PGID);
2473 attach_pid(p, PIDTYPE_SID);
2474 __this_cpu_inc(process_counts);
2475 } else {
2476 current->signal->nr_threads++;
2477 current->signal->quick_threads++;
2478 atomic_inc(¤t->signal->live);
2479 refcount_inc(¤t->signal->sigcnt);
2480 task_join_group_stop(p);
2481 list_add_tail_rcu(&p->thread_group,
2482 &p->group_leader->thread_group);
2483 list_add_tail_rcu(&p->thread_node,
2484 &p->signal->thread_head);
2485 }
2486 attach_pid(p, PIDTYPE_PID);
2487 nr_threads++;
2488 }
2489 total_forks++;
2490 hlist_del_init(&delayed.node);
2491 spin_unlock(¤t->sighand->siglock);
2492 syscall_tracepoint_update(p);
2493 write_unlock_irq(&tasklist_lock);
2494
2495 if (pidfile)
2496 fd_install(pidfd, pidfile);
2497
2498 proc_fork_connector(p);
2499 sched_post_fork(p);
2500 cgroup_post_fork(p, args);
2501 perf_event_fork(p);
2502
2503 trace_task_newtask(p, clone_flags);
2504 uprobe_copy_process(p, clone_flags);
2505
2506 copy_oom_score_adj(clone_flags, p);
2507
2508 return p;
2509
2510bad_fork_cancel_cgroup:
2511 sched_core_free(p);
2512 spin_unlock(¤t->sighand->siglock);
2513 write_unlock_irq(&tasklist_lock);
2514 cgroup_cancel_fork(p, args);
2515bad_fork_put_pidfd:
2516 if (clone_flags & CLONE_PIDFD) {
2517 fput(pidfile);
2518 put_unused_fd(pidfd);
2519 }
2520bad_fork_free_pid:
2521 if (pid != &init_struct_pid)
2522 free_pid(pid);
2523bad_fork_cleanup_thread:
2524 exit_thread(p);
2525bad_fork_cleanup_io:
2526 if (p->io_context)
2527 exit_io_context(p);
2528bad_fork_cleanup_namespaces:
2529 exit_task_namespaces(p);
2530bad_fork_cleanup_mm:
2531 if (p->mm) {
2532 mm_clear_owner(p->mm, p);
2533 mmput(p->mm);
2534 }
2535bad_fork_cleanup_signal:
2536 if (!(clone_flags & CLONE_THREAD))
2537 free_signal_struct(p->signal);
2538bad_fork_cleanup_sighand:
2539 __cleanup_sighand(p->sighand);
2540bad_fork_cleanup_fs:
2541 exit_fs(p); /* blocking */
2542bad_fork_cleanup_files:
2543 exit_files(p); /* blocking */
2544bad_fork_cleanup_semundo:
2545 exit_sem(p);
2546bad_fork_cleanup_security:
2547 security_task_free(p);
2548bad_fork_cleanup_audit:
2549 audit_free(p);
2550bad_fork_cleanup_perf:
2551 perf_event_free_task(p);
2552bad_fork_cleanup_policy:
2553 lockdep_free_task(p);
2554#ifdef CONFIG_NUMA
2555 mpol_put(p->mempolicy);
2556#endif
2557bad_fork_cleanup_delayacct:
2558 delayacct_tsk_free(p);
2559bad_fork_cleanup_count:
2560 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2561 exit_creds(p);
2562bad_fork_free:
2563 WRITE_ONCE(p->__state, TASK_DEAD);
2564 exit_task_stack_account(p);
2565 put_task_stack(p);
2566 delayed_free_task(p);
2567fork_out:
2568 spin_lock_irq(¤t->sighand->siglock);
2569 hlist_del_init(&delayed.node);
2570 spin_unlock_irq(¤t->sighand->siglock);
2571 return ERR_PTR(retval);
2572}
2573
2574static inline void init_idle_pids(struct task_struct *idle)
2575{
2576 enum pid_type type;
2577
2578 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2579 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2580 init_task_pid(idle, type, &init_struct_pid);
2581 }
2582}
2583
2584static int idle_dummy(void *dummy)
2585{
2586 /* This function is never called */
2587 return 0;
2588}
2589
2590struct task_struct * __init fork_idle(int cpu)
2591{
2592 struct task_struct *task;
2593 struct kernel_clone_args args = {
2594 .flags = CLONE_VM,
2595 .fn = &idle_dummy,
2596 .fn_arg = NULL,
2597 .kthread = 1,
2598 .idle = 1,
2599 };
2600
2601 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2602 if (!IS_ERR(task)) {
2603 init_idle_pids(task);
2604 init_idle(task, cpu);
2605 }
2606
2607 return task;
2608}
2609
2610/*
2611 * This is like kernel_clone(), but shaved down and tailored to just
2612 * creating io_uring workers. It returns a created task, or an error pointer.
2613 * The returned task is inactive, and the caller must fire it up through
2614 * wake_up_new_task(p). All signals are blocked in the created task.
2615 */
2616struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2617{
2618 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2619 CLONE_IO;
2620 struct kernel_clone_args args = {
2621 .flags = ((lower_32_bits(flags) | CLONE_VM |
2622 CLONE_UNTRACED) & ~CSIGNAL),
2623 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2624 .fn = fn,
2625 .fn_arg = arg,
2626 .io_thread = 1,
2627 };
2628
2629 return copy_process(NULL, 0, node, &args);
2630}
2631
2632/*
2633 * Ok, this is the main fork-routine.
2634 *
2635 * It copies the process, and if successful kick-starts
2636 * it and waits for it to finish using the VM if required.
2637 *
2638 * args->exit_signal is expected to be checked for sanity by the caller.
2639 */
2640pid_t kernel_clone(struct kernel_clone_args *args)
2641{
2642 u64 clone_flags = args->flags;
2643 struct completion vfork;
2644 struct pid *pid;
2645 struct task_struct *p;
2646 int trace = 0;
2647 pid_t nr;
2648
2649 /*
2650 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2651 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2652 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2653 * field in struct clone_args and it still doesn't make sense to have
2654 * them both point at the same memory location. Performing this check
2655 * here has the advantage that we don't need to have a separate helper
2656 * to check for legacy clone().
2657 */
2658 if ((args->flags & CLONE_PIDFD) &&
2659 (args->flags & CLONE_PARENT_SETTID) &&
2660 (args->pidfd == args->parent_tid))
2661 return -EINVAL;
2662
2663 /*
2664 * Determine whether and which event to report to ptracer. When
2665 * called from kernel_thread or CLONE_UNTRACED is explicitly
2666 * requested, no event is reported; otherwise, report if the event
2667 * for the type of forking is enabled.
2668 */
2669 if (!(clone_flags & CLONE_UNTRACED)) {
2670 if (clone_flags & CLONE_VFORK)
2671 trace = PTRACE_EVENT_VFORK;
2672 else if (args->exit_signal != SIGCHLD)
2673 trace = PTRACE_EVENT_CLONE;
2674 else
2675 trace = PTRACE_EVENT_FORK;
2676
2677 if (likely(!ptrace_event_enabled(current, trace)))
2678 trace = 0;
2679 }
2680
2681 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2682 add_latent_entropy();
2683
2684 if (IS_ERR(p))
2685 return PTR_ERR(p);
2686
2687 /*
2688 * Do this prior waking up the new thread - the thread pointer
2689 * might get invalid after that point, if the thread exits quickly.
2690 */
2691 trace_sched_process_fork(current, p);
2692
2693 pid = get_task_pid(p, PIDTYPE_PID);
2694 nr = pid_vnr(pid);
2695
2696 if (clone_flags & CLONE_PARENT_SETTID)
2697 put_user(nr, args->parent_tid);
2698
2699 if (clone_flags & CLONE_VFORK) {
2700 p->vfork_done = &vfork;
2701 init_completion(&vfork);
2702 get_task_struct(p);
2703 }
2704
2705 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2706 /* lock the task to synchronize with memcg migration */
2707 task_lock(p);
2708 lru_gen_add_mm(p->mm);
2709 task_unlock(p);
2710 }
2711
2712 wake_up_new_task(p);
2713
2714 /* forking complete and child started to run, tell ptracer */
2715 if (unlikely(trace))
2716 ptrace_event_pid(trace, pid);
2717
2718 if (clone_flags & CLONE_VFORK) {
2719 if (!wait_for_vfork_done(p, &vfork))
2720 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2721 }
2722
2723 put_pid(pid);
2724 return nr;
2725}
2726
2727/*
2728 * Create a kernel thread.
2729 */
2730pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2731{
2732 struct kernel_clone_args args = {
2733 .flags = ((lower_32_bits(flags) | CLONE_VM |
2734 CLONE_UNTRACED) & ~CSIGNAL),
2735 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2736 .fn = fn,
2737 .fn_arg = arg,
2738 .kthread = 1,
2739 };
2740
2741 return kernel_clone(&args);
2742}
2743
2744/*
2745 * Create a user mode thread.
2746 */
2747pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2748{
2749 struct kernel_clone_args args = {
2750 .flags = ((lower_32_bits(flags) | CLONE_VM |
2751 CLONE_UNTRACED) & ~CSIGNAL),
2752 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2753 .fn = fn,
2754 .fn_arg = arg,
2755 };
2756
2757 return kernel_clone(&args);
2758}
2759
2760#ifdef __ARCH_WANT_SYS_FORK
2761SYSCALL_DEFINE0(fork)
2762{
2763#ifdef CONFIG_MMU
2764 struct kernel_clone_args args = {
2765 .exit_signal = SIGCHLD,
2766 };
2767
2768 return kernel_clone(&args);
2769#else
2770 /* can not support in nommu mode */
2771 return -EINVAL;
2772#endif
2773}
2774#endif
2775
2776#ifdef __ARCH_WANT_SYS_VFORK
2777SYSCALL_DEFINE0(vfork)
2778{
2779 struct kernel_clone_args args = {
2780 .flags = CLONE_VFORK | CLONE_VM,
2781 .exit_signal = SIGCHLD,
2782 };
2783
2784 return kernel_clone(&args);
2785}
2786#endif
2787
2788#ifdef __ARCH_WANT_SYS_CLONE
2789#ifdef CONFIG_CLONE_BACKWARDS
2790SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2791 int __user *, parent_tidptr,
2792 unsigned long, tls,
2793 int __user *, child_tidptr)
2794#elif defined(CONFIG_CLONE_BACKWARDS2)
2795SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2796 int __user *, parent_tidptr,
2797 int __user *, child_tidptr,
2798 unsigned long, tls)
2799#elif defined(CONFIG_CLONE_BACKWARDS3)
2800SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2801 int, stack_size,
2802 int __user *, parent_tidptr,
2803 int __user *, child_tidptr,
2804 unsigned long, tls)
2805#else
2806SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2807 int __user *, parent_tidptr,
2808 int __user *, child_tidptr,
2809 unsigned long, tls)
2810#endif
2811{
2812 struct kernel_clone_args args = {
2813 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2814 .pidfd = parent_tidptr,
2815 .child_tid = child_tidptr,
2816 .parent_tid = parent_tidptr,
2817 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2818 .stack = newsp,
2819 .tls = tls,
2820 };
2821
2822 return kernel_clone(&args);
2823}
2824#endif
2825
2826#ifdef __ARCH_WANT_SYS_CLONE3
2827
2828noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2829 struct clone_args __user *uargs,
2830 size_t usize)
2831{
2832 int err;
2833 struct clone_args args;
2834 pid_t *kset_tid = kargs->set_tid;
2835
2836 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2837 CLONE_ARGS_SIZE_VER0);
2838 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2839 CLONE_ARGS_SIZE_VER1);
2840 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2841 CLONE_ARGS_SIZE_VER2);
2842 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2843
2844 if (unlikely(usize > PAGE_SIZE))
2845 return -E2BIG;
2846 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2847 return -EINVAL;
2848
2849 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2850 if (err)
2851 return err;
2852
2853 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2854 return -EINVAL;
2855
2856 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2857 return -EINVAL;
2858
2859 if (unlikely(args.set_tid && args.set_tid_size == 0))
2860 return -EINVAL;
2861
2862 /*
2863 * Verify that higher 32bits of exit_signal are unset and that
2864 * it is a valid signal
2865 */
2866 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2867 !valid_signal(args.exit_signal)))
2868 return -EINVAL;
2869
2870 if ((args.flags & CLONE_INTO_CGROUP) &&
2871 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2872 return -EINVAL;
2873
2874 *kargs = (struct kernel_clone_args){
2875 .flags = args.flags,
2876 .pidfd = u64_to_user_ptr(args.pidfd),
2877 .child_tid = u64_to_user_ptr(args.child_tid),
2878 .parent_tid = u64_to_user_ptr(args.parent_tid),
2879 .exit_signal = args.exit_signal,
2880 .stack = args.stack,
2881 .stack_size = args.stack_size,
2882 .tls = args.tls,
2883 .set_tid_size = args.set_tid_size,
2884 .cgroup = args.cgroup,
2885 };
2886
2887 if (args.set_tid &&
2888 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2889 (kargs->set_tid_size * sizeof(pid_t))))
2890 return -EFAULT;
2891
2892 kargs->set_tid = kset_tid;
2893
2894 return 0;
2895}
2896
2897/**
2898 * clone3_stack_valid - check and prepare stack
2899 * @kargs: kernel clone args
2900 *
2901 * Verify that the stack arguments userspace gave us are sane.
2902 * In addition, set the stack direction for userspace since it's easy for us to
2903 * determine.
2904 */
2905static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2906{
2907 if (kargs->stack == 0) {
2908 if (kargs->stack_size > 0)
2909 return false;
2910 } else {
2911 if (kargs->stack_size == 0)
2912 return false;
2913
2914 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2915 return false;
2916
2917#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2918 kargs->stack += kargs->stack_size;
2919#endif
2920 }
2921
2922 return true;
2923}
2924
2925static bool clone3_args_valid(struct kernel_clone_args *kargs)
2926{
2927 /* Verify that no unknown flags are passed along. */
2928 if (kargs->flags &
2929 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2930 return false;
2931
2932 /*
2933 * - make the CLONE_DETACHED bit reusable for clone3
2934 * - make the CSIGNAL bits reusable for clone3
2935 */
2936 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2937 return false;
2938
2939 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2940 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2941 return false;
2942
2943 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2944 kargs->exit_signal)
2945 return false;
2946
2947 if (!clone3_stack_valid(kargs))
2948 return false;
2949
2950 return true;
2951}
2952
2953/**
2954 * clone3 - create a new process with specific properties
2955 * @uargs: argument structure
2956 * @size: size of @uargs
2957 *
2958 * clone3() is the extensible successor to clone()/clone2().
2959 * It takes a struct as argument that is versioned by its size.
2960 *
2961 * Return: On success, a positive PID for the child process.
2962 * On error, a negative errno number.
2963 */
2964SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2965{
2966 int err;
2967
2968 struct kernel_clone_args kargs;
2969 pid_t set_tid[MAX_PID_NS_LEVEL];
2970
2971 kargs.set_tid = set_tid;
2972
2973 err = copy_clone_args_from_user(&kargs, uargs, size);
2974 if (err)
2975 return err;
2976
2977 if (!clone3_args_valid(&kargs))
2978 return -EINVAL;
2979
2980 return kernel_clone(&kargs);
2981}
2982#endif
2983
2984void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2985{
2986 struct task_struct *leader, *parent, *child;
2987 int res;
2988
2989 read_lock(&tasklist_lock);
2990 leader = top = top->group_leader;
2991down:
2992 for_each_thread(leader, parent) {
2993 list_for_each_entry(child, &parent->children, sibling) {
2994 res = visitor(child, data);
2995 if (res) {
2996 if (res < 0)
2997 goto out;
2998 leader = child;
2999 goto down;
3000 }
3001up:
3002 ;
3003 }
3004 }
3005
3006 if (leader != top) {
3007 child = leader;
3008 parent = child->real_parent;
3009 leader = parent->group_leader;
3010 goto up;
3011 }
3012out:
3013 read_unlock(&tasklist_lock);
3014}
3015
3016#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3017#define ARCH_MIN_MMSTRUCT_ALIGN 0
3018#endif
3019
3020static void sighand_ctor(void *data)
3021{
3022 struct sighand_struct *sighand = data;
3023
3024 spin_lock_init(&sighand->siglock);
3025 init_waitqueue_head(&sighand->signalfd_wqh);
3026}
3027
3028void __init mm_cache_init(void)
3029{
3030 unsigned int mm_size;
3031
3032 /*
3033 * The mm_cpumask is located at the end of mm_struct, and is
3034 * dynamically sized based on the maximum CPU number this system
3035 * can have, taking hotplug into account (nr_cpu_ids).
3036 */
3037 mm_size = sizeof(struct mm_struct) + cpumask_size();
3038
3039 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3040 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3041 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3042 offsetof(struct mm_struct, saved_auxv),
3043 sizeof_field(struct mm_struct, saved_auxv),
3044 NULL);
3045}
3046
3047void __init proc_caches_init(void)
3048{
3049 sighand_cachep = kmem_cache_create("sighand_cache",
3050 sizeof(struct sighand_struct), 0,
3051 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3052 SLAB_ACCOUNT, sighand_ctor);
3053 signal_cachep = kmem_cache_create("signal_cache",
3054 sizeof(struct signal_struct), 0,
3055 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3056 NULL);
3057 files_cachep = kmem_cache_create("files_cache",
3058 sizeof(struct files_struct), 0,
3059 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3060 NULL);
3061 fs_cachep = kmem_cache_create("fs_cache",
3062 sizeof(struct fs_struct), 0,
3063 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3064 NULL);
3065
3066 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3067 mmap_init();
3068 nsproxy_cache_init();
3069}
3070
3071/*
3072 * Check constraints on flags passed to the unshare system call.
3073 */
3074static int check_unshare_flags(unsigned long unshare_flags)
3075{
3076 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3077 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3078 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3079 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3080 CLONE_NEWTIME))
3081 return -EINVAL;
3082 /*
3083 * Not implemented, but pretend it works if there is nothing
3084 * to unshare. Note that unsharing the address space or the
3085 * signal handlers also need to unshare the signal queues (aka
3086 * CLONE_THREAD).
3087 */
3088 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3089 if (!thread_group_empty(current))
3090 return -EINVAL;
3091 }
3092 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3093 if (refcount_read(¤t->sighand->count) > 1)
3094 return -EINVAL;
3095 }
3096 if (unshare_flags & CLONE_VM) {
3097 if (!current_is_single_threaded())
3098 return -EINVAL;
3099 }
3100
3101 return 0;
3102}
3103
3104/*
3105 * Unshare the filesystem structure if it is being shared
3106 */
3107static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3108{
3109 struct fs_struct *fs = current->fs;
3110
3111 if (!(unshare_flags & CLONE_FS) || !fs)
3112 return 0;
3113
3114 /* don't need lock here; in the worst case we'll do useless copy */
3115 if (fs->users == 1)
3116 return 0;
3117
3118 *new_fsp = copy_fs_struct(fs);
3119 if (!*new_fsp)
3120 return -ENOMEM;
3121
3122 return 0;
3123}
3124
3125/*
3126 * Unshare file descriptor table if it is being shared
3127 */
3128int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3129 struct files_struct **new_fdp)
3130{
3131 struct files_struct *fd = current->files;
3132 int error = 0;
3133
3134 if ((unshare_flags & CLONE_FILES) &&
3135 (fd && atomic_read(&fd->count) > 1)) {
3136 *new_fdp = dup_fd(fd, max_fds, &error);
3137 if (!*new_fdp)
3138 return error;
3139 }
3140
3141 return 0;
3142}
3143
3144/*
3145 * unshare allows a process to 'unshare' part of the process
3146 * context which was originally shared using clone. copy_*
3147 * functions used by kernel_clone() cannot be used here directly
3148 * because they modify an inactive task_struct that is being
3149 * constructed. Here we are modifying the current, active,
3150 * task_struct.
3151 */
3152int ksys_unshare(unsigned long unshare_flags)
3153{
3154 struct fs_struct *fs, *new_fs = NULL;
3155 struct files_struct *new_fd = NULL;
3156 struct cred *new_cred = NULL;
3157 struct nsproxy *new_nsproxy = NULL;
3158 int do_sysvsem = 0;
3159 int err;
3160
3161 /*
3162 * If unsharing a user namespace must also unshare the thread group
3163 * and unshare the filesystem root and working directories.
3164 */
3165 if (unshare_flags & CLONE_NEWUSER)
3166 unshare_flags |= CLONE_THREAD | CLONE_FS;
3167 /*
3168 * If unsharing vm, must also unshare signal handlers.
3169 */
3170 if (unshare_flags & CLONE_VM)
3171 unshare_flags |= CLONE_SIGHAND;
3172 /*
3173 * If unsharing a signal handlers, must also unshare the signal queues.
3174 */
3175 if (unshare_flags & CLONE_SIGHAND)
3176 unshare_flags |= CLONE_THREAD;
3177 /*
3178 * If unsharing namespace, must also unshare filesystem information.
3179 */
3180 if (unshare_flags & CLONE_NEWNS)
3181 unshare_flags |= CLONE_FS;
3182
3183 err = check_unshare_flags(unshare_flags);
3184 if (err)
3185 goto bad_unshare_out;
3186 /*
3187 * CLONE_NEWIPC must also detach from the undolist: after switching
3188 * to a new ipc namespace, the semaphore arrays from the old
3189 * namespace are unreachable.
3190 */
3191 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3192 do_sysvsem = 1;
3193 err = unshare_fs(unshare_flags, &new_fs);
3194 if (err)
3195 goto bad_unshare_out;
3196 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3197 if (err)
3198 goto bad_unshare_cleanup_fs;
3199 err = unshare_userns(unshare_flags, &new_cred);
3200 if (err)
3201 goto bad_unshare_cleanup_fd;
3202 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3203 new_cred, new_fs);
3204 if (err)
3205 goto bad_unshare_cleanup_cred;
3206
3207 if (new_cred) {
3208 err = set_cred_ucounts(new_cred);
3209 if (err)
3210 goto bad_unshare_cleanup_cred;
3211 }
3212
3213 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3214 if (do_sysvsem) {
3215 /*
3216 * CLONE_SYSVSEM is equivalent to sys_exit().
3217 */
3218 exit_sem(current);
3219 }
3220 if (unshare_flags & CLONE_NEWIPC) {
3221 /* Orphan segments in old ns (see sem above). */
3222 exit_shm(current);
3223 shm_init_task(current);
3224 }
3225
3226 if (new_nsproxy)
3227 switch_task_namespaces(current, new_nsproxy);
3228
3229 task_lock(current);
3230
3231 if (new_fs) {
3232 fs = current->fs;
3233 spin_lock(&fs->lock);
3234 current->fs = new_fs;
3235 if (--fs->users)
3236 new_fs = NULL;
3237 else
3238 new_fs = fs;
3239 spin_unlock(&fs->lock);
3240 }
3241
3242 if (new_fd)
3243 swap(current->files, new_fd);
3244
3245 task_unlock(current);
3246
3247 if (new_cred) {
3248 /* Install the new user namespace */
3249 commit_creds(new_cred);
3250 new_cred = NULL;
3251 }
3252 }
3253
3254 perf_event_namespaces(current);
3255
3256bad_unshare_cleanup_cred:
3257 if (new_cred)
3258 put_cred(new_cred);
3259bad_unshare_cleanup_fd:
3260 if (new_fd)
3261 put_files_struct(new_fd);
3262
3263bad_unshare_cleanup_fs:
3264 if (new_fs)
3265 free_fs_struct(new_fs);
3266
3267bad_unshare_out:
3268 return err;
3269}
3270
3271SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3272{
3273 return ksys_unshare(unshare_flags);
3274}
3275
3276/*
3277 * Helper to unshare the files of the current task.
3278 * We don't want to expose copy_files internals to
3279 * the exec layer of the kernel.
3280 */
3281
3282int unshare_files(void)
3283{
3284 struct task_struct *task = current;
3285 struct files_struct *old, *copy = NULL;
3286 int error;
3287
3288 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3289 if (error || !copy)
3290 return error;
3291
3292 old = task->files;
3293 task_lock(task);
3294 task->files = copy;
3295 task_unlock(task);
3296 put_files_struct(old);
3297 return 0;
3298}
3299
3300int sysctl_max_threads(struct ctl_table *table, int write,
3301 void *buffer, size_t *lenp, loff_t *ppos)
3302{
3303 struct ctl_table t;
3304 int ret;
3305 int threads = max_threads;
3306 int min = 1;
3307 int max = MAX_THREADS;
3308
3309 t = *table;
3310 t.data = &threads;
3311 t.extra1 = &min;
3312 t.extra2 = &max;
3313
3314 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3315 if (ret || !write)
3316 return ret;
3317
3318 max_threads = threads;
3319
3320 return 0;
3321}
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/binfmts.h>
41#include <linux/mman.h>
42#include <linux/mmu_notifier.h>
43#include <linux/fs.h>
44#include <linux/mm.h>
45#include <linux/vmacache.h>
46#include <linux/nsproxy.h>
47#include <linux/capability.h>
48#include <linux/cpu.h>
49#include <linux/cgroup.h>
50#include <linux/security.h>
51#include <linux/hugetlb.h>
52#include <linux/seccomp.h>
53#include <linux/swap.h>
54#include <linux/syscalls.h>
55#include <linux/jiffies.h>
56#include <linux/futex.h>
57#include <linux/compat.h>
58#include <linux/kthread.h>
59#include <linux/task_io_accounting_ops.h>
60#include <linux/rcupdate.h>
61#include <linux/ptrace.h>
62#include <linux/mount.h>
63#include <linux/audit.h>
64#include <linux/memcontrol.h>
65#include <linux/ftrace.h>
66#include <linux/proc_fs.h>
67#include <linux/profile.h>
68#include <linux/rmap.h>
69#include <linux/ksm.h>
70#include <linux/acct.h>
71#include <linux/userfaultfd_k.h>
72#include <linux/tsacct_kern.h>
73#include <linux/cn_proc.h>
74#include <linux/freezer.h>
75#include <linux/delayacct.h>
76#include <linux/taskstats_kern.h>
77#include <linux/random.h>
78#include <linux/tty.h>
79#include <linux/blkdev.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
99#include <asm/pgalloc.h>
100#include <linux/uaccess.h>
101#include <asm/mmu_context.h>
102#include <asm/cacheflush.h>
103#include <asm/tlbflush.h>
104
105#include <trace/events/sched.h>
106
107#define CREATE_TRACE_POINTS
108#include <trace/events/task.h>
109
110/*
111 * Minimum number of threads to boot the kernel
112 */
113#define MIN_THREADS 20
114
115/*
116 * Maximum number of threads
117 */
118#define MAX_THREADS FUTEX_TID_MASK
119
120/*
121 * Protected counters by write_lock_irq(&tasklist_lock)
122 */
123unsigned long total_forks; /* Handle normal Linux uptimes. */
124int nr_threads; /* The idle threads do not count.. */
125
126static int max_threads; /* tunable limit on nr_threads */
127
128#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
129
130static const char * const resident_page_types[] = {
131 NAMED_ARRAY_INDEX(MM_FILEPAGES),
132 NAMED_ARRAY_INDEX(MM_ANONPAGES),
133 NAMED_ARRAY_INDEX(MM_SWAPENTS),
134 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
135};
136
137DEFINE_PER_CPU(unsigned long, process_counts) = 0;
138
139__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
140
141#ifdef CONFIG_PROVE_RCU
142int lockdep_tasklist_lock_is_held(void)
143{
144 return lockdep_is_held(&tasklist_lock);
145}
146EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
147#endif /* #ifdef CONFIG_PROVE_RCU */
148
149int nr_processes(void)
150{
151 int cpu;
152 int total = 0;
153
154 for_each_possible_cpu(cpu)
155 total += per_cpu(process_counts, cpu);
156
157 return total;
158}
159
160void __weak arch_release_task_struct(struct task_struct *tsk)
161{
162}
163
164#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
165static struct kmem_cache *task_struct_cachep;
166
167static inline struct task_struct *alloc_task_struct_node(int node)
168{
169 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
170}
171
172static inline void free_task_struct(struct task_struct *tsk)
173{
174 kmem_cache_free(task_struct_cachep, tsk);
175}
176#endif
177
178#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
179
180/*
181 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
182 * kmemcache based allocator.
183 */
184# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
185
186#ifdef CONFIG_VMAP_STACK
187/*
188 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
189 * flush. Try to minimize the number of calls by caching stacks.
190 */
191#define NR_CACHED_STACKS 2
192static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
193
194static int free_vm_stack_cache(unsigned int cpu)
195{
196 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
197 int i;
198
199 for (i = 0; i < NR_CACHED_STACKS; i++) {
200 struct vm_struct *vm_stack = cached_vm_stacks[i];
201
202 if (!vm_stack)
203 continue;
204
205 vfree(vm_stack->addr);
206 cached_vm_stacks[i] = NULL;
207 }
208
209 return 0;
210}
211#endif
212
213static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
214{
215#ifdef CONFIG_VMAP_STACK
216 void *stack;
217 int i;
218
219 for (i = 0; i < NR_CACHED_STACKS; i++) {
220 struct vm_struct *s;
221
222 s = this_cpu_xchg(cached_stacks[i], NULL);
223
224 if (!s)
225 continue;
226
227 /* Clear the KASAN shadow of the stack. */
228 kasan_unpoison_shadow(s->addr, THREAD_SIZE);
229
230 /* Clear stale pointers from reused stack. */
231 memset(s->addr, 0, THREAD_SIZE);
232
233 tsk->stack_vm_area = s;
234 tsk->stack = s->addr;
235 return s->addr;
236 }
237
238 /*
239 * Allocated stacks are cached and later reused by new threads,
240 * so memcg accounting is performed manually on assigning/releasing
241 * stacks to tasks. Drop __GFP_ACCOUNT.
242 */
243 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
244 VMALLOC_START, VMALLOC_END,
245 THREADINFO_GFP & ~__GFP_ACCOUNT,
246 PAGE_KERNEL,
247 0, node, __builtin_return_address(0));
248
249 /*
250 * We can't call find_vm_area() in interrupt context, and
251 * free_thread_stack() can be called in interrupt context,
252 * so cache the vm_struct.
253 */
254 if (stack) {
255 tsk->stack_vm_area = find_vm_area(stack);
256 tsk->stack = stack;
257 }
258 return stack;
259#else
260 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
261 THREAD_SIZE_ORDER);
262
263 if (likely(page)) {
264 tsk->stack = kasan_reset_tag(page_address(page));
265 return tsk->stack;
266 }
267 return NULL;
268#endif
269}
270
271static inline void free_thread_stack(struct task_struct *tsk)
272{
273#ifdef CONFIG_VMAP_STACK
274 struct vm_struct *vm = task_stack_vm_area(tsk);
275
276 if (vm) {
277 int i;
278
279 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
280 memcg_kmem_uncharge_page(vm->pages[i], 0);
281
282 for (i = 0; i < NR_CACHED_STACKS; i++) {
283 if (this_cpu_cmpxchg(cached_stacks[i],
284 NULL, tsk->stack_vm_area) != NULL)
285 continue;
286
287 return;
288 }
289
290 vfree_atomic(tsk->stack);
291 return;
292 }
293#endif
294
295 __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
296}
297# else
298static struct kmem_cache *thread_stack_cache;
299
300static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
301 int node)
302{
303 unsigned long *stack;
304 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
305 stack = kasan_reset_tag(stack);
306 tsk->stack = stack;
307 return stack;
308}
309
310static void free_thread_stack(struct task_struct *tsk)
311{
312 kmem_cache_free(thread_stack_cache, tsk->stack);
313}
314
315void thread_stack_cache_init(void)
316{
317 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
318 THREAD_SIZE, THREAD_SIZE, 0, 0,
319 THREAD_SIZE, NULL);
320 BUG_ON(thread_stack_cache == NULL);
321}
322# endif
323#endif
324
325/* SLAB cache for signal_struct structures (tsk->signal) */
326static struct kmem_cache *signal_cachep;
327
328/* SLAB cache for sighand_struct structures (tsk->sighand) */
329struct kmem_cache *sighand_cachep;
330
331/* SLAB cache for files_struct structures (tsk->files) */
332struct kmem_cache *files_cachep;
333
334/* SLAB cache for fs_struct structures (tsk->fs) */
335struct kmem_cache *fs_cachep;
336
337/* SLAB cache for vm_area_struct structures */
338static struct kmem_cache *vm_area_cachep;
339
340/* SLAB cache for mm_struct structures (tsk->mm) */
341static struct kmem_cache *mm_cachep;
342
343struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
344{
345 struct vm_area_struct *vma;
346
347 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
348 if (vma)
349 vma_init(vma, mm);
350 return vma;
351}
352
353struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
354{
355 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
356
357 if (new) {
358 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
359 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
360 /*
361 * orig->shared.rb may be modified concurrently, but the clone
362 * will be reinitialized.
363 */
364 *new = data_race(*orig);
365 INIT_LIST_HEAD(&new->anon_vma_chain);
366 new->vm_next = new->vm_prev = NULL;
367 }
368 return new;
369}
370
371void vm_area_free(struct vm_area_struct *vma)
372{
373 kmem_cache_free(vm_area_cachep, vma);
374}
375
376static void account_kernel_stack(struct task_struct *tsk, int account)
377{
378 void *stack = task_stack_page(tsk);
379 struct vm_struct *vm = task_stack_vm_area(tsk);
380
381
382 /* All stack pages are in the same node. */
383 if (vm)
384 mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
385 account * (THREAD_SIZE / 1024));
386 else
387 mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
388 account * (THREAD_SIZE / 1024));
389}
390
391static int memcg_charge_kernel_stack(struct task_struct *tsk)
392{
393#ifdef CONFIG_VMAP_STACK
394 struct vm_struct *vm = task_stack_vm_area(tsk);
395 int ret;
396
397 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
398
399 if (vm) {
400 int i;
401
402 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
403
404 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
405 /*
406 * If memcg_kmem_charge_page() fails, page->mem_cgroup
407 * pointer is NULL, and memcg_kmem_uncharge_page() in
408 * free_thread_stack() will ignore this page.
409 */
410 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
411 0);
412 if (ret)
413 return ret;
414 }
415 }
416#endif
417 return 0;
418}
419
420static void release_task_stack(struct task_struct *tsk)
421{
422 if (WARN_ON(tsk->state != TASK_DEAD))
423 return; /* Better to leak the stack than to free prematurely */
424
425 account_kernel_stack(tsk, -1);
426 free_thread_stack(tsk);
427 tsk->stack = NULL;
428#ifdef CONFIG_VMAP_STACK
429 tsk->stack_vm_area = NULL;
430#endif
431}
432
433#ifdef CONFIG_THREAD_INFO_IN_TASK
434void put_task_stack(struct task_struct *tsk)
435{
436 if (refcount_dec_and_test(&tsk->stack_refcount))
437 release_task_stack(tsk);
438}
439#endif
440
441void free_task(struct task_struct *tsk)
442{
443 scs_release(tsk);
444
445#ifndef CONFIG_THREAD_INFO_IN_TASK
446 /*
447 * The task is finally done with both the stack and thread_info,
448 * so free both.
449 */
450 release_task_stack(tsk);
451#else
452 /*
453 * If the task had a separate stack allocation, it should be gone
454 * by now.
455 */
456 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
457#endif
458 rt_mutex_debug_task_free(tsk);
459 ftrace_graph_exit_task(tsk);
460 arch_release_task_struct(tsk);
461 if (tsk->flags & PF_KTHREAD)
462 free_kthread_struct(tsk);
463 free_task_struct(tsk);
464}
465EXPORT_SYMBOL(free_task);
466
467#ifdef CONFIG_MMU
468static __latent_entropy int dup_mmap(struct mm_struct *mm,
469 struct mm_struct *oldmm)
470{
471 struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
472 struct rb_node **rb_link, *rb_parent;
473 int retval;
474 unsigned long charge;
475 LIST_HEAD(uf);
476
477 uprobe_start_dup_mmap();
478 if (mmap_write_lock_killable(oldmm)) {
479 retval = -EINTR;
480 goto fail_uprobe_end;
481 }
482 flush_cache_dup_mm(oldmm);
483 uprobe_dup_mmap(oldmm, mm);
484 /*
485 * Not linked in yet - no deadlock potential:
486 */
487 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
488
489 /* No ordering required: file already has been exposed. */
490 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
491
492 mm->total_vm = oldmm->total_vm;
493 mm->data_vm = oldmm->data_vm;
494 mm->exec_vm = oldmm->exec_vm;
495 mm->stack_vm = oldmm->stack_vm;
496
497 rb_link = &mm->mm_rb.rb_node;
498 rb_parent = NULL;
499 pprev = &mm->mmap;
500 retval = ksm_fork(mm, oldmm);
501 if (retval)
502 goto out;
503 retval = khugepaged_fork(mm, oldmm);
504 if (retval)
505 goto out;
506
507 prev = NULL;
508 for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
509 struct file *file;
510
511 if (mpnt->vm_flags & VM_DONTCOPY) {
512 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
513 continue;
514 }
515 charge = 0;
516 /*
517 * Don't duplicate many vmas if we've been oom-killed (for
518 * example)
519 */
520 if (fatal_signal_pending(current)) {
521 retval = -EINTR;
522 goto out;
523 }
524 if (mpnt->vm_flags & VM_ACCOUNT) {
525 unsigned long len = vma_pages(mpnt);
526
527 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
528 goto fail_nomem;
529 charge = len;
530 }
531 tmp = vm_area_dup(mpnt);
532 if (!tmp)
533 goto fail_nomem;
534 retval = vma_dup_policy(mpnt, tmp);
535 if (retval)
536 goto fail_nomem_policy;
537 tmp->vm_mm = mm;
538 retval = dup_userfaultfd(tmp, &uf);
539 if (retval)
540 goto fail_nomem_anon_vma_fork;
541 if (tmp->vm_flags & VM_WIPEONFORK) {
542 /*
543 * VM_WIPEONFORK gets a clean slate in the child.
544 * Don't prepare anon_vma until fault since we don't
545 * copy page for current vma.
546 */
547 tmp->anon_vma = NULL;
548 } else if (anon_vma_fork(tmp, mpnt))
549 goto fail_nomem_anon_vma_fork;
550 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
551 file = tmp->vm_file;
552 if (file) {
553 struct inode *inode = file_inode(file);
554 struct address_space *mapping = file->f_mapping;
555
556 get_file(file);
557 if (tmp->vm_flags & VM_DENYWRITE)
558 atomic_dec(&inode->i_writecount);
559 i_mmap_lock_write(mapping);
560 if (tmp->vm_flags & VM_SHARED)
561 atomic_inc(&mapping->i_mmap_writable);
562 flush_dcache_mmap_lock(mapping);
563 /* insert tmp into the share list, just after mpnt */
564 vma_interval_tree_insert_after(tmp, mpnt,
565 &mapping->i_mmap);
566 flush_dcache_mmap_unlock(mapping);
567 i_mmap_unlock_write(mapping);
568 }
569
570 /*
571 * Clear hugetlb-related page reserves for children. This only
572 * affects MAP_PRIVATE mappings. Faults generated by the child
573 * are not guaranteed to succeed, even if read-only
574 */
575 if (is_vm_hugetlb_page(tmp))
576 reset_vma_resv_huge_pages(tmp);
577
578 /*
579 * Link in the new vma and copy the page table entries.
580 */
581 *pprev = tmp;
582 pprev = &tmp->vm_next;
583 tmp->vm_prev = prev;
584 prev = tmp;
585
586 __vma_link_rb(mm, tmp, rb_link, rb_parent);
587 rb_link = &tmp->vm_rb.rb_right;
588 rb_parent = &tmp->vm_rb;
589
590 mm->map_count++;
591 if (!(tmp->vm_flags & VM_WIPEONFORK))
592 retval = copy_page_range(mm, oldmm, mpnt, tmp);
593
594 if (tmp->vm_ops && tmp->vm_ops->open)
595 tmp->vm_ops->open(tmp);
596
597 if (retval)
598 goto out;
599 }
600 /* a new mm has just been created */
601 retval = arch_dup_mmap(oldmm, mm);
602out:
603 mmap_write_unlock(mm);
604 flush_tlb_mm(oldmm);
605 mmap_write_unlock(oldmm);
606 dup_userfaultfd_complete(&uf);
607fail_uprobe_end:
608 uprobe_end_dup_mmap();
609 return retval;
610fail_nomem_anon_vma_fork:
611 mpol_put(vma_policy(tmp));
612fail_nomem_policy:
613 vm_area_free(tmp);
614fail_nomem:
615 retval = -ENOMEM;
616 vm_unacct_memory(charge);
617 goto out;
618}
619
620static inline int mm_alloc_pgd(struct mm_struct *mm)
621{
622 mm->pgd = pgd_alloc(mm);
623 if (unlikely(!mm->pgd))
624 return -ENOMEM;
625 return 0;
626}
627
628static inline void mm_free_pgd(struct mm_struct *mm)
629{
630 pgd_free(mm, mm->pgd);
631}
632#else
633static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
634{
635 mmap_write_lock(oldmm);
636 RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
637 mmap_write_unlock(oldmm);
638 return 0;
639}
640#define mm_alloc_pgd(mm) (0)
641#define mm_free_pgd(mm)
642#endif /* CONFIG_MMU */
643
644static void check_mm(struct mm_struct *mm)
645{
646 int i;
647
648 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
649 "Please make sure 'struct resident_page_types[]' is updated as well");
650
651 for (i = 0; i < NR_MM_COUNTERS; i++) {
652 long x = atomic_long_read(&mm->rss_stat.count[i]);
653
654 if (unlikely(x))
655 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
656 mm, resident_page_types[i], x);
657 }
658
659 if (mm_pgtables_bytes(mm))
660 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
661 mm_pgtables_bytes(mm));
662
663#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
664 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
665#endif
666}
667
668#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
669#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
670
671/*
672 * Called when the last reference to the mm
673 * is dropped: either by a lazy thread or by
674 * mmput. Free the page directory and the mm.
675 */
676void __mmdrop(struct mm_struct *mm)
677{
678 BUG_ON(mm == &init_mm);
679 WARN_ON_ONCE(mm == current->mm);
680 WARN_ON_ONCE(mm == current->active_mm);
681 mm_free_pgd(mm);
682 destroy_context(mm);
683 mmu_notifier_subscriptions_destroy(mm);
684 check_mm(mm);
685 put_user_ns(mm->user_ns);
686 free_mm(mm);
687}
688EXPORT_SYMBOL_GPL(__mmdrop);
689
690static void mmdrop_async_fn(struct work_struct *work)
691{
692 struct mm_struct *mm;
693
694 mm = container_of(work, struct mm_struct, async_put_work);
695 __mmdrop(mm);
696}
697
698static void mmdrop_async(struct mm_struct *mm)
699{
700 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
701 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
702 schedule_work(&mm->async_put_work);
703 }
704}
705
706static inline void free_signal_struct(struct signal_struct *sig)
707{
708 taskstats_tgid_free(sig);
709 sched_autogroup_exit(sig);
710 /*
711 * __mmdrop is not safe to call from softirq context on x86 due to
712 * pgd_dtor so postpone it to the async context
713 */
714 if (sig->oom_mm)
715 mmdrop_async(sig->oom_mm);
716 kmem_cache_free(signal_cachep, sig);
717}
718
719static inline void put_signal_struct(struct signal_struct *sig)
720{
721 if (refcount_dec_and_test(&sig->sigcnt))
722 free_signal_struct(sig);
723}
724
725void __put_task_struct(struct task_struct *tsk)
726{
727 WARN_ON(!tsk->exit_state);
728 WARN_ON(refcount_read(&tsk->usage));
729 WARN_ON(tsk == current);
730
731 cgroup_free(tsk);
732 task_numa_free(tsk, true);
733 security_task_free(tsk);
734 exit_creds(tsk);
735 delayacct_tsk_free(tsk);
736 put_signal_struct(tsk->signal);
737
738 if (!profile_handoff_task(tsk))
739 free_task(tsk);
740}
741EXPORT_SYMBOL_GPL(__put_task_struct);
742
743void __init __weak arch_task_cache_init(void) { }
744
745/*
746 * set_max_threads
747 */
748static void set_max_threads(unsigned int max_threads_suggested)
749{
750 u64 threads;
751 unsigned long nr_pages = totalram_pages();
752
753 /*
754 * The number of threads shall be limited such that the thread
755 * structures may only consume a small part of the available memory.
756 */
757 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
758 threads = MAX_THREADS;
759 else
760 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
761 (u64) THREAD_SIZE * 8UL);
762
763 if (threads > max_threads_suggested)
764 threads = max_threads_suggested;
765
766 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
767}
768
769#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
770/* Initialized by the architecture: */
771int arch_task_struct_size __read_mostly;
772#endif
773
774#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
775static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
776{
777 /* Fetch thread_struct whitelist for the architecture. */
778 arch_thread_struct_whitelist(offset, size);
779
780 /*
781 * Handle zero-sized whitelist or empty thread_struct, otherwise
782 * adjust offset to position of thread_struct in task_struct.
783 */
784 if (unlikely(*size == 0))
785 *offset = 0;
786 else
787 *offset += offsetof(struct task_struct, thread);
788}
789#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
790
791void __init fork_init(void)
792{
793 int i;
794#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
795#ifndef ARCH_MIN_TASKALIGN
796#define ARCH_MIN_TASKALIGN 0
797#endif
798 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
799 unsigned long useroffset, usersize;
800
801 /* create a slab on which task_structs can be allocated */
802 task_struct_whitelist(&useroffset, &usersize);
803 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
804 arch_task_struct_size, align,
805 SLAB_PANIC|SLAB_ACCOUNT,
806 useroffset, usersize, NULL);
807#endif
808
809 /* do the arch specific task caches init */
810 arch_task_cache_init();
811
812 set_max_threads(MAX_THREADS);
813
814 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
815 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
816 init_task.signal->rlim[RLIMIT_SIGPENDING] =
817 init_task.signal->rlim[RLIMIT_NPROC];
818
819 for (i = 0; i < UCOUNT_COUNTS; i++) {
820 init_user_ns.ucount_max[i] = max_threads/2;
821 }
822
823#ifdef CONFIG_VMAP_STACK
824 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
825 NULL, free_vm_stack_cache);
826#endif
827
828 scs_init();
829
830 lockdep_init_task(&init_task);
831 uprobes_init();
832}
833
834int __weak arch_dup_task_struct(struct task_struct *dst,
835 struct task_struct *src)
836{
837 *dst = *src;
838 return 0;
839}
840
841void set_task_stack_end_magic(struct task_struct *tsk)
842{
843 unsigned long *stackend;
844
845 stackend = end_of_stack(tsk);
846 *stackend = STACK_END_MAGIC; /* for overflow detection */
847}
848
849static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
850{
851 struct task_struct *tsk;
852 unsigned long *stack;
853 struct vm_struct *stack_vm_area __maybe_unused;
854 int err;
855
856 if (node == NUMA_NO_NODE)
857 node = tsk_fork_get_node(orig);
858 tsk = alloc_task_struct_node(node);
859 if (!tsk)
860 return NULL;
861
862 stack = alloc_thread_stack_node(tsk, node);
863 if (!stack)
864 goto free_tsk;
865
866 if (memcg_charge_kernel_stack(tsk))
867 goto free_stack;
868
869 stack_vm_area = task_stack_vm_area(tsk);
870
871 err = arch_dup_task_struct(tsk, orig);
872
873 /*
874 * arch_dup_task_struct() clobbers the stack-related fields. Make
875 * sure they're properly initialized before using any stack-related
876 * functions again.
877 */
878 tsk->stack = stack;
879#ifdef CONFIG_VMAP_STACK
880 tsk->stack_vm_area = stack_vm_area;
881#endif
882#ifdef CONFIG_THREAD_INFO_IN_TASK
883 refcount_set(&tsk->stack_refcount, 1);
884#endif
885
886 if (err)
887 goto free_stack;
888
889 err = scs_prepare(tsk, node);
890 if (err)
891 goto free_stack;
892
893#ifdef CONFIG_SECCOMP
894 /*
895 * We must handle setting up seccomp filters once we're under
896 * the sighand lock in case orig has changed between now and
897 * then. Until then, filter must be NULL to avoid messing up
898 * the usage counts on the error path calling free_task.
899 */
900 tsk->seccomp.filter = NULL;
901#endif
902
903 setup_thread_stack(tsk, orig);
904 clear_user_return_notifier(tsk);
905 clear_tsk_need_resched(tsk);
906 set_task_stack_end_magic(tsk);
907
908#ifdef CONFIG_STACKPROTECTOR
909 tsk->stack_canary = get_random_canary();
910#endif
911 if (orig->cpus_ptr == &orig->cpus_mask)
912 tsk->cpus_ptr = &tsk->cpus_mask;
913
914 /*
915 * One for the user space visible state that goes away when reaped.
916 * One for the scheduler.
917 */
918 refcount_set(&tsk->rcu_users, 2);
919 /* One for the rcu users */
920 refcount_set(&tsk->usage, 1);
921#ifdef CONFIG_BLK_DEV_IO_TRACE
922 tsk->btrace_seq = 0;
923#endif
924 tsk->splice_pipe = NULL;
925 tsk->task_frag.page = NULL;
926 tsk->wake_q.next = NULL;
927
928 account_kernel_stack(tsk, 1);
929
930 kcov_task_init(tsk);
931
932#ifdef CONFIG_FAULT_INJECTION
933 tsk->fail_nth = 0;
934#endif
935
936#ifdef CONFIG_BLK_CGROUP
937 tsk->throttle_queue = NULL;
938 tsk->use_memdelay = 0;
939#endif
940
941#ifdef CONFIG_MEMCG
942 tsk->active_memcg = NULL;
943#endif
944 return tsk;
945
946free_stack:
947 free_thread_stack(tsk);
948free_tsk:
949 free_task_struct(tsk);
950 return NULL;
951}
952
953__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
954
955static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
956
957static int __init coredump_filter_setup(char *s)
958{
959 default_dump_filter =
960 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
961 MMF_DUMP_FILTER_MASK;
962 return 1;
963}
964
965__setup("coredump_filter=", coredump_filter_setup);
966
967#include <linux/init_task.h>
968
969static void mm_init_aio(struct mm_struct *mm)
970{
971#ifdef CONFIG_AIO
972 spin_lock_init(&mm->ioctx_lock);
973 mm->ioctx_table = NULL;
974#endif
975}
976
977static __always_inline void mm_clear_owner(struct mm_struct *mm,
978 struct task_struct *p)
979{
980#ifdef CONFIG_MEMCG
981 if (mm->owner == p)
982 WRITE_ONCE(mm->owner, NULL);
983#endif
984}
985
986static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
987{
988#ifdef CONFIG_MEMCG
989 mm->owner = p;
990#endif
991}
992
993static void mm_init_uprobes_state(struct mm_struct *mm)
994{
995#ifdef CONFIG_UPROBES
996 mm->uprobes_state.xol_area = NULL;
997#endif
998}
999
1000static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1001 struct user_namespace *user_ns)
1002{
1003 mm->mmap = NULL;
1004 mm->mm_rb = RB_ROOT;
1005 mm->vmacache_seqnum = 0;
1006 atomic_set(&mm->mm_users, 1);
1007 atomic_set(&mm->mm_count, 1);
1008 mmap_init_lock(mm);
1009 INIT_LIST_HEAD(&mm->mmlist);
1010 mm->core_state = NULL;
1011 mm_pgtables_bytes_init(mm);
1012 mm->map_count = 0;
1013 mm->locked_vm = 0;
1014 atomic_set(&mm->has_pinned, 0);
1015 atomic64_set(&mm->pinned_vm, 0);
1016 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1017 spin_lock_init(&mm->page_table_lock);
1018 spin_lock_init(&mm->arg_lock);
1019 mm_init_cpumask(mm);
1020 mm_init_aio(mm);
1021 mm_init_owner(mm, p);
1022 RCU_INIT_POINTER(mm->exe_file, NULL);
1023 mmu_notifier_subscriptions_init(mm);
1024 init_tlb_flush_pending(mm);
1025#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1026 mm->pmd_huge_pte = NULL;
1027#endif
1028 mm_init_uprobes_state(mm);
1029
1030 if (current->mm) {
1031 mm->flags = current->mm->flags & MMF_INIT_MASK;
1032 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1033 } else {
1034 mm->flags = default_dump_filter;
1035 mm->def_flags = 0;
1036 }
1037
1038 if (mm_alloc_pgd(mm))
1039 goto fail_nopgd;
1040
1041 if (init_new_context(p, mm))
1042 goto fail_nocontext;
1043
1044 mm->user_ns = get_user_ns(user_ns);
1045 return mm;
1046
1047fail_nocontext:
1048 mm_free_pgd(mm);
1049fail_nopgd:
1050 free_mm(mm);
1051 return NULL;
1052}
1053
1054/*
1055 * Allocate and initialize an mm_struct.
1056 */
1057struct mm_struct *mm_alloc(void)
1058{
1059 struct mm_struct *mm;
1060
1061 mm = allocate_mm();
1062 if (!mm)
1063 return NULL;
1064
1065 memset(mm, 0, sizeof(*mm));
1066 return mm_init(mm, current, current_user_ns());
1067}
1068
1069static inline void __mmput(struct mm_struct *mm)
1070{
1071 VM_BUG_ON(atomic_read(&mm->mm_users));
1072
1073 uprobe_clear_state(mm);
1074 exit_aio(mm);
1075 ksm_exit(mm);
1076 khugepaged_exit(mm); /* must run before exit_mmap */
1077 exit_mmap(mm);
1078 mm_put_huge_zero_page(mm);
1079 set_mm_exe_file(mm, NULL);
1080 if (!list_empty(&mm->mmlist)) {
1081 spin_lock(&mmlist_lock);
1082 list_del(&mm->mmlist);
1083 spin_unlock(&mmlist_lock);
1084 }
1085 if (mm->binfmt)
1086 module_put(mm->binfmt->module);
1087 mmdrop(mm);
1088}
1089
1090/*
1091 * Decrement the use count and release all resources for an mm.
1092 */
1093void mmput(struct mm_struct *mm)
1094{
1095 might_sleep();
1096
1097 if (atomic_dec_and_test(&mm->mm_users))
1098 __mmput(mm);
1099}
1100EXPORT_SYMBOL_GPL(mmput);
1101
1102#ifdef CONFIG_MMU
1103static void mmput_async_fn(struct work_struct *work)
1104{
1105 struct mm_struct *mm = container_of(work, struct mm_struct,
1106 async_put_work);
1107
1108 __mmput(mm);
1109}
1110
1111void mmput_async(struct mm_struct *mm)
1112{
1113 if (atomic_dec_and_test(&mm->mm_users)) {
1114 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1115 schedule_work(&mm->async_put_work);
1116 }
1117}
1118#endif
1119
1120/**
1121 * set_mm_exe_file - change a reference to the mm's executable file
1122 *
1123 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1124 *
1125 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1126 * invocations: in mmput() nobody alive left, in execve task is single
1127 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1128 * mm->exe_file, but does so without using set_mm_exe_file() in order
1129 * to do avoid the need for any locks.
1130 */
1131void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1132{
1133 struct file *old_exe_file;
1134
1135 /*
1136 * It is safe to dereference the exe_file without RCU as
1137 * this function is only called if nobody else can access
1138 * this mm -- see comment above for justification.
1139 */
1140 old_exe_file = rcu_dereference_raw(mm->exe_file);
1141
1142 if (new_exe_file)
1143 get_file(new_exe_file);
1144 rcu_assign_pointer(mm->exe_file, new_exe_file);
1145 if (old_exe_file)
1146 fput(old_exe_file);
1147}
1148
1149/**
1150 * get_mm_exe_file - acquire a reference to the mm's executable file
1151 *
1152 * Returns %NULL if mm has no associated executable file.
1153 * User must release file via fput().
1154 */
1155struct file *get_mm_exe_file(struct mm_struct *mm)
1156{
1157 struct file *exe_file;
1158
1159 rcu_read_lock();
1160 exe_file = rcu_dereference(mm->exe_file);
1161 if (exe_file && !get_file_rcu(exe_file))
1162 exe_file = NULL;
1163 rcu_read_unlock();
1164 return exe_file;
1165}
1166EXPORT_SYMBOL(get_mm_exe_file);
1167
1168/**
1169 * get_task_exe_file - acquire a reference to the task's executable file
1170 *
1171 * Returns %NULL if task's mm (if any) has no associated executable file or
1172 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1173 * User must release file via fput().
1174 */
1175struct file *get_task_exe_file(struct task_struct *task)
1176{
1177 struct file *exe_file = NULL;
1178 struct mm_struct *mm;
1179
1180 task_lock(task);
1181 mm = task->mm;
1182 if (mm) {
1183 if (!(task->flags & PF_KTHREAD))
1184 exe_file = get_mm_exe_file(mm);
1185 }
1186 task_unlock(task);
1187 return exe_file;
1188}
1189EXPORT_SYMBOL(get_task_exe_file);
1190
1191/**
1192 * get_task_mm - acquire a reference to the task's mm
1193 *
1194 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1195 * this kernel workthread has transiently adopted a user mm with use_mm,
1196 * to do its AIO) is not set and if so returns a reference to it, after
1197 * bumping up the use count. User must release the mm via mmput()
1198 * after use. Typically used by /proc and ptrace.
1199 */
1200struct mm_struct *get_task_mm(struct task_struct *task)
1201{
1202 struct mm_struct *mm;
1203
1204 task_lock(task);
1205 mm = task->mm;
1206 if (mm) {
1207 if (task->flags & PF_KTHREAD)
1208 mm = NULL;
1209 else
1210 mmget(mm);
1211 }
1212 task_unlock(task);
1213 return mm;
1214}
1215EXPORT_SYMBOL_GPL(get_task_mm);
1216
1217struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1218{
1219 struct mm_struct *mm;
1220 int err;
1221
1222 err = mutex_lock_killable(&task->signal->exec_update_mutex);
1223 if (err)
1224 return ERR_PTR(err);
1225
1226 mm = get_task_mm(task);
1227 if (mm && mm != current->mm &&
1228 !ptrace_may_access(task, mode)) {
1229 mmput(mm);
1230 mm = ERR_PTR(-EACCES);
1231 }
1232 mutex_unlock(&task->signal->exec_update_mutex);
1233
1234 return mm;
1235}
1236
1237static void complete_vfork_done(struct task_struct *tsk)
1238{
1239 struct completion *vfork;
1240
1241 task_lock(tsk);
1242 vfork = tsk->vfork_done;
1243 if (likely(vfork)) {
1244 tsk->vfork_done = NULL;
1245 complete(vfork);
1246 }
1247 task_unlock(tsk);
1248}
1249
1250static int wait_for_vfork_done(struct task_struct *child,
1251 struct completion *vfork)
1252{
1253 int killed;
1254
1255 freezer_do_not_count();
1256 cgroup_enter_frozen();
1257 killed = wait_for_completion_killable(vfork);
1258 cgroup_leave_frozen(false);
1259 freezer_count();
1260
1261 if (killed) {
1262 task_lock(child);
1263 child->vfork_done = NULL;
1264 task_unlock(child);
1265 }
1266
1267 put_task_struct(child);
1268 return killed;
1269}
1270
1271/* Please note the differences between mmput and mm_release.
1272 * mmput is called whenever we stop holding onto a mm_struct,
1273 * error success whatever.
1274 *
1275 * mm_release is called after a mm_struct has been removed
1276 * from the current process.
1277 *
1278 * This difference is important for error handling, when we
1279 * only half set up a mm_struct for a new process and need to restore
1280 * the old one. Because we mmput the new mm_struct before
1281 * restoring the old one. . .
1282 * Eric Biederman 10 January 1998
1283 */
1284static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1285{
1286 uprobe_free_utask(tsk);
1287
1288 /* Get rid of any cached register state */
1289 deactivate_mm(tsk, mm);
1290
1291 /*
1292 * Signal userspace if we're not exiting with a core dump
1293 * because we want to leave the value intact for debugging
1294 * purposes.
1295 */
1296 if (tsk->clear_child_tid) {
1297 if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1298 atomic_read(&mm->mm_users) > 1) {
1299 /*
1300 * We don't check the error code - if userspace has
1301 * not set up a proper pointer then tough luck.
1302 */
1303 put_user(0, tsk->clear_child_tid);
1304 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1305 1, NULL, NULL, 0, 0);
1306 }
1307 tsk->clear_child_tid = NULL;
1308 }
1309
1310 /*
1311 * All done, finally we can wake up parent and return this mm to him.
1312 * Also kthread_stop() uses this completion for synchronization.
1313 */
1314 if (tsk->vfork_done)
1315 complete_vfork_done(tsk);
1316}
1317
1318void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1319{
1320 futex_exit_release(tsk);
1321 mm_release(tsk, mm);
1322}
1323
1324void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1325{
1326 futex_exec_release(tsk);
1327 mm_release(tsk, mm);
1328}
1329
1330/**
1331 * dup_mm() - duplicates an existing mm structure
1332 * @tsk: the task_struct with which the new mm will be associated.
1333 * @oldmm: the mm to duplicate.
1334 *
1335 * Allocates a new mm structure and duplicates the provided @oldmm structure
1336 * content into it.
1337 *
1338 * Return: the duplicated mm or NULL on failure.
1339 */
1340static struct mm_struct *dup_mm(struct task_struct *tsk,
1341 struct mm_struct *oldmm)
1342{
1343 struct mm_struct *mm;
1344 int err;
1345
1346 mm = allocate_mm();
1347 if (!mm)
1348 goto fail_nomem;
1349
1350 memcpy(mm, oldmm, sizeof(*mm));
1351
1352 if (!mm_init(mm, tsk, mm->user_ns))
1353 goto fail_nomem;
1354
1355 err = dup_mmap(mm, oldmm);
1356 if (err)
1357 goto free_pt;
1358
1359 mm->hiwater_rss = get_mm_rss(mm);
1360 mm->hiwater_vm = mm->total_vm;
1361
1362 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1363 goto free_pt;
1364
1365 return mm;
1366
1367free_pt:
1368 /* don't put binfmt in mmput, we haven't got module yet */
1369 mm->binfmt = NULL;
1370 mm_init_owner(mm, NULL);
1371 mmput(mm);
1372
1373fail_nomem:
1374 return NULL;
1375}
1376
1377static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1378{
1379 struct mm_struct *mm, *oldmm;
1380 int retval;
1381
1382 tsk->min_flt = tsk->maj_flt = 0;
1383 tsk->nvcsw = tsk->nivcsw = 0;
1384#ifdef CONFIG_DETECT_HUNG_TASK
1385 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1386 tsk->last_switch_time = 0;
1387#endif
1388
1389 tsk->mm = NULL;
1390 tsk->active_mm = NULL;
1391
1392 /*
1393 * Are we cloning a kernel thread?
1394 *
1395 * We need to steal a active VM for that..
1396 */
1397 oldmm = current->mm;
1398 if (!oldmm)
1399 return 0;
1400
1401 /* initialize the new vmacache entries */
1402 vmacache_flush(tsk);
1403
1404 if (clone_flags & CLONE_VM) {
1405 mmget(oldmm);
1406 mm = oldmm;
1407 goto good_mm;
1408 }
1409
1410 retval = -ENOMEM;
1411 mm = dup_mm(tsk, current->mm);
1412 if (!mm)
1413 goto fail_nomem;
1414
1415good_mm:
1416 tsk->mm = mm;
1417 tsk->active_mm = mm;
1418 return 0;
1419
1420fail_nomem:
1421 return retval;
1422}
1423
1424static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1425{
1426 struct fs_struct *fs = current->fs;
1427 if (clone_flags & CLONE_FS) {
1428 /* tsk->fs is already what we want */
1429 spin_lock(&fs->lock);
1430 if (fs->in_exec) {
1431 spin_unlock(&fs->lock);
1432 return -EAGAIN;
1433 }
1434 fs->users++;
1435 spin_unlock(&fs->lock);
1436 return 0;
1437 }
1438 tsk->fs = copy_fs_struct(fs);
1439 if (!tsk->fs)
1440 return -ENOMEM;
1441 return 0;
1442}
1443
1444static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1445{
1446 struct files_struct *oldf, *newf;
1447 int error = 0;
1448
1449 /*
1450 * A background process may not have any files ...
1451 */
1452 oldf = current->files;
1453 if (!oldf)
1454 goto out;
1455
1456 if (clone_flags & CLONE_FILES) {
1457 atomic_inc(&oldf->count);
1458 goto out;
1459 }
1460
1461 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1462 if (!newf)
1463 goto out;
1464
1465 tsk->files = newf;
1466 error = 0;
1467out:
1468 return error;
1469}
1470
1471static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1472{
1473#ifdef CONFIG_BLOCK
1474 struct io_context *ioc = current->io_context;
1475 struct io_context *new_ioc;
1476
1477 if (!ioc)
1478 return 0;
1479 /*
1480 * Share io context with parent, if CLONE_IO is set
1481 */
1482 if (clone_flags & CLONE_IO) {
1483 ioc_task_link(ioc);
1484 tsk->io_context = ioc;
1485 } else if (ioprio_valid(ioc->ioprio)) {
1486 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1487 if (unlikely(!new_ioc))
1488 return -ENOMEM;
1489
1490 new_ioc->ioprio = ioc->ioprio;
1491 put_io_context(new_ioc);
1492 }
1493#endif
1494 return 0;
1495}
1496
1497static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1498{
1499 struct sighand_struct *sig;
1500
1501 if (clone_flags & CLONE_SIGHAND) {
1502 refcount_inc(¤t->sighand->count);
1503 return 0;
1504 }
1505 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1506 RCU_INIT_POINTER(tsk->sighand, sig);
1507 if (!sig)
1508 return -ENOMEM;
1509
1510 refcount_set(&sig->count, 1);
1511 spin_lock_irq(¤t->sighand->siglock);
1512 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1513 spin_unlock_irq(¤t->sighand->siglock);
1514
1515 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1516 if (clone_flags & CLONE_CLEAR_SIGHAND)
1517 flush_signal_handlers(tsk, 0);
1518
1519 return 0;
1520}
1521
1522void __cleanup_sighand(struct sighand_struct *sighand)
1523{
1524 if (refcount_dec_and_test(&sighand->count)) {
1525 signalfd_cleanup(sighand);
1526 /*
1527 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1528 * without an RCU grace period, see __lock_task_sighand().
1529 */
1530 kmem_cache_free(sighand_cachep, sighand);
1531 }
1532}
1533
1534/*
1535 * Initialize POSIX timer handling for a thread group.
1536 */
1537static void posix_cpu_timers_init_group(struct signal_struct *sig)
1538{
1539 struct posix_cputimers *pct = &sig->posix_cputimers;
1540 unsigned long cpu_limit;
1541
1542 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1543 posix_cputimers_group_init(pct, cpu_limit);
1544}
1545
1546static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1547{
1548 struct signal_struct *sig;
1549
1550 if (clone_flags & CLONE_THREAD)
1551 return 0;
1552
1553 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1554 tsk->signal = sig;
1555 if (!sig)
1556 return -ENOMEM;
1557
1558 sig->nr_threads = 1;
1559 atomic_set(&sig->live, 1);
1560 refcount_set(&sig->sigcnt, 1);
1561
1562 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1563 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1564 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1565
1566 init_waitqueue_head(&sig->wait_chldexit);
1567 sig->curr_target = tsk;
1568 init_sigpending(&sig->shared_pending);
1569 INIT_HLIST_HEAD(&sig->multiprocess);
1570 seqlock_init(&sig->stats_lock);
1571 prev_cputime_init(&sig->prev_cputime);
1572
1573#ifdef CONFIG_POSIX_TIMERS
1574 INIT_LIST_HEAD(&sig->posix_timers);
1575 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1576 sig->real_timer.function = it_real_fn;
1577#endif
1578
1579 task_lock(current->group_leader);
1580 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1581 task_unlock(current->group_leader);
1582
1583 posix_cpu_timers_init_group(sig);
1584
1585 tty_audit_fork(sig);
1586 sched_autogroup_fork(sig);
1587
1588 sig->oom_score_adj = current->signal->oom_score_adj;
1589 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1590
1591 mutex_init(&sig->cred_guard_mutex);
1592 mutex_init(&sig->exec_update_mutex);
1593
1594 return 0;
1595}
1596
1597static void copy_seccomp(struct task_struct *p)
1598{
1599#ifdef CONFIG_SECCOMP
1600 /*
1601 * Must be called with sighand->lock held, which is common to
1602 * all threads in the group. Holding cred_guard_mutex is not
1603 * needed because this new task is not yet running and cannot
1604 * be racing exec.
1605 */
1606 assert_spin_locked(¤t->sighand->siglock);
1607
1608 /* Ref-count the new filter user, and assign it. */
1609 get_seccomp_filter(current);
1610 p->seccomp = current->seccomp;
1611
1612 /*
1613 * Explicitly enable no_new_privs here in case it got set
1614 * between the task_struct being duplicated and holding the
1615 * sighand lock. The seccomp state and nnp must be in sync.
1616 */
1617 if (task_no_new_privs(current))
1618 task_set_no_new_privs(p);
1619
1620 /*
1621 * If the parent gained a seccomp mode after copying thread
1622 * flags and between before we held the sighand lock, we have
1623 * to manually enable the seccomp thread flag here.
1624 */
1625 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1626 set_tsk_thread_flag(p, TIF_SECCOMP);
1627#endif
1628}
1629
1630SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1631{
1632 current->clear_child_tid = tidptr;
1633
1634 return task_pid_vnr(current);
1635}
1636
1637static void rt_mutex_init_task(struct task_struct *p)
1638{
1639 raw_spin_lock_init(&p->pi_lock);
1640#ifdef CONFIG_RT_MUTEXES
1641 p->pi_waiters = RB_ROOT_CACHED;
1642 p->pi_top_task = NULL;
1643 p->pi_blocked_on = NULL;
1644#endif
1645}
1646
1647static inline void init_task_pid_links(struct task_struct *task)
1648{
1649 enum pid_type type;
1650
1651 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1652 INIT_HLIST_NODE(&task->pid_links[type]);
1653 }
1654}
1655
1656static inline void
1657init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1658{
1659 if (type == PIDTYPE_PID)
1660 task->thread_pid = pid;
1661 else
1662 task->signal->pids[type] = pid;
1663}
1664
1665static inline void rcu_copy_process(struct task_struct *p)
1666{
1667#ifdef CONFIG_PREEMPT_RCU
1668 p->rcu_read_lock_nesting = 0;
1669 p->rcu_read_unlock_special.s = 0;
1670 p->rcu_blocked_node = NULL;
1671 INIT_LIST_HEAD(&p->rcu_node_entry);
1672#endif /* #ifdef CONFIG_PREEMPT_RCU */
1673#ifdef CONFIG_TASKS_RCU
1674 p->rcu_tasks_holdout = false;
1675 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1676 p->rcu_tasks_idle_cpu = -1;
1677#endif /* #ifdef CONFIG_TASKS_RCU */
1678#ifdef CONFIG_TASKS_TRACE_RCU
1679 p->trc_reader_nesting = 0;
1680 p->trc_reader_special.s = 0;
1681 INIT_LIST_HEAD(&p->trc_holdout_list);
1682#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1683}
1684
1685struct pid *pidfd_pid(const struct file *file)
1686{
1687 if (file->f_op == &pidfd_fops)
1688 return file->private_data;
1689
1690 return ERR_PTR(-EBADF);
1691}
1692
1693static int pidfd_release(struct inode *inode, struct file *file)
1694{
1695 struct pid *pid = file->private_data;
1696
1697 file->private_data = NULL;
1698 put_pid(pid);
1699 return 0;
1700}
1701
1702#ifdef CONFIG_PROC_FS
1703/**
1704 * pidfd_show_fdinfo - print information about a pidfd
1705 * @m: proc fdinfo file
1706 * @f: file referencing a pidfd
1707 *
1708 * Pid:
1709 * This function will print the pid that a given pidfd refers to in the
1710 * pid namespace of the procfs instance.
1711 * If the pid namespace of the process is not a descendant of the pid
1712 * namespace of the procfs instance 0 will be shown as its pid. This is
1713 * similar to calling getppid() on a process whose parent is outside of
1714 * its pid namespace.
1715 *
1716 * NSpid:
1717 * If pid namespaces are supported then this function will also print
1718 * the pid of a given pidfd refers to for all descendant pid namespaces
1719 * starting from the current pid namespace of the instance, i.e. the
1720 * Pid field and the first entry in the NSpid field will be identical.
1721 * If the pid namespace of the process is not a descendant of the pid
1722 * namespace of the procfs instance 0 will be shown as its first NSpid
1723 * entry and no others will be shown.
1724 * Note that this differs from the Pid and NSpid fields in
1725 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1726 * the pid namespace of the procfs instance. The difference becomes
1727 * obvious when sending around a pidfd between pid namespaces from a
1728 * different branch of the tree, i.e. where no ancestoral relation is
1729 * present between the pid namespaces:
1730 * - create two new pid namespaces ns1 and ns2 in the initial pid
1731 * namespace (also take care to create new mount namespaces in the
1732 * new pid namespace and mount procfs)
1733 * - create a process with a pidfd in ns1
1734 * - send pidfd from ns1 to ns2
1735 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1736 * have exactly one entry, which is 0
1737 */
1738static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1739{
1740 struct pid *pid = f->private_data;
1741 struct pid_namespace *ns;
1742 pid_t nr = -1;
1743
1744 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1745 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1746 nr = pid_nr_ns(pid, ns);
1747 }
1748
1749 seq_put_decimal_ll(m, "Pid:\t", nr);
1750
1751#ifdef CONFIG_PID_NS
1752 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1753 if (nr > 0) {
1754 int i;
1755
1756 /* If nr is non-zero it means that 'pid' is valid and that
1757 * ns, i.e. the pid namespace associated with the procfs
1758 * instance, is in the pid namespace hierarchy of pid.
1759 * Start at one below the already printed level.
1760 */
1761 for (i = ns->level + 1; i <= pid->level; i++)
1762 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1763 }
1764#endif
1765 seq_putc(m, '\n');
1766}
1767#endif
1768
1769/*
1770 * Poll support for process exit notification.
1771 */
1772static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1773{
1774 struct pid *pid = file->private_data;
1775 __poll_t poll_flags = 0;
1776
1777 poll_wait(file, &pid->wait_pidfd, pts);
1778
1779 /*
1780 * Inform pollers only when the whole thread group exits.
1781 * If the thread group leader exits before all other threads in the
1782 * group, then poll(2) should block, similar to the wait(2) family.
1783 */
1784 if (thread_group_exited(pid))
1785 poll_flags = EPOLLIN | EPOLLRDNORM;
1786
1787 return poll_flags;
1788}
1789
1790const struct file_operations pidfd_fops = {
1791 .release = pidfd_release,
1792 .poll = pidfd_poll,
1793#ifdef CONFIG_PROC_FS
1794 .show_fdinfo = pidfd_show_fdinfo,
1795#endif
1796};
1797
1798static void __delayed_free_task(struct rcu_head *rhp)
1799{
1800 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1801
1802 free_task(tsk);
1803}
1804
1805static __always_inline void delayed_free_task(struct task_struct *tsk)
1806{
1807 if (IS_ENABLED(CONFIG_MEMCG))
1808 call_rcu(&tsk->rcu, __delayed_free_task);
1809 else
1810 free_task(tsk);
1811}
1812
1813/*
1814 * This creates a new process as a copy of the old one,
1815 * but does not actually start it yet.
1816 *
1817 * It copies the registers, and all the appropriate
1818 * parts of the process environment (as per the clone
1819 * flags). The actual kick-off is left to the caller.
1820 */
1821static __latent_entropy struct task_struct *copy_process(
1822 struct pid *pid,
1823 int trace,
1824 int node,
1825 struct kernel_clone_args *args)
1826{
1827 int pidfd = -1, retval;
1828 struct task_struct *p;
1829 struct multiprocess_signals delayed;
1830 struct file *pidfile = NULL;
1831 u64 clone_flags = args->flags;
1832 struct nsproxy *nsp = current->nsproxy;
1833
1834 /*
1835 * Don't allow sharing the root directory with processes in a different
1836 * namespace
1837 */
1838 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1839 return ERR_PTR(-EINVAL);
1840
1841 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1842 return ERR_PTR(-EINVAL);
1843
1844 /*
1845 * Thread groups must share signals as well, and detached threads
1846 * can only be started up within the thread group.
1847 */
1848 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1849 return ERR_PTR(-EINVAL);
1850
1851 /*
1852 * Shared signal handlers imply shared VM. By way of the above,
1853 * thread groups also imply shared VM. Blocking this case allows
1854 * for various simplifications in other code.
1855 */
1856 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1857 return ERR_PTR(-EINVAL);
1858
1859 /*
1860 * Siblings of global init remain as zombies on exit since they are
1861 * not reaped by their parent (swapper). To solve this and to avoid
1862 * multi-rooted process trees, prevent global and container-inits
1863 * from creating siblings.
1864 */
1865 if ((clone_flags & CLONE_PARENT) &&
1866 current->signal->flags & SIGNAL_UNKILLABLE)
1867 return ERR_PTR(-EINVAL);
1868
1869 /*
1870 * If the new process will be in a different pid or user namespace
1871 * do not allow it to share a thread group with the forking task.
1872 */
1873 if (clone_flags & CLONE_THREAD) {
1874 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1875 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1876 return ERR_PTR(-EINVAL);
1877 }
1878
1879 /*
1880 * If the new process will be in a different time namespace
1881 * do not allow it to share VM or a thread group with the forking task.
1882 */
1883 if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1884 if (nsp->time_ns != nsp->time_ns_for_children)
1885 return ERR_PTR(-EINVAL);
1886 }
1887
1888 if (clone_flags & CLONE_PIDFD) {
1889 /*
1890 * - CLONE_DETACHED is blocked so that we can potentially
1891 * reuse it later for CLONE_PIDFD.
1892 * - CLONE_THREAD is blocked until someone really needs it.
1893 */
1894 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1895 return ERR_PTR(-EINVAL);
1896 }
1897
1898 /*
1899 * Force any signals received before this point to be delivered
1900 * before the fork happens. Collect up signals sent to multiple
1901 * processes that happen during the fork and delay them so that
1902 * they appear to happen after the fork.
1903 */
1904 sigemptyset(&delayed.signal);
1905 INIT_HLIST_NODE(&delayed.node);
1906
1907 spin_lock_irq(¤t->sighand->siglock);
1908 if (!(clone_flags & CLONE_THREAD))
1909 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
1910 recalc_sigpending();
1911 spin_unlock_irq(¤t->sighand->siglock);
1912 retval = -ERESTARTNOINTR;
1913 if (signal_pending(current))
1914 goto fork_out;
1915
1916 retval = -ENOMEM;
1917 p = dup_task_struct(current, node);
1918 if (!p)
1919 goto fork_out;
1920
1921 /*
1922 * This _must_ happen before we call free_task(), i.e. before we jump
1923 * to any of the bad_fork_* labels. This is to avoid freeing
1924 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1925 * kernel threads (PF_KTHREAD).
1926 */
1927 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1928 /*
1929 * Clear TID on mm_release()?
1930 */
1931 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1932
1933 ftrace_graph_init_task(p);
1934
1935 rt_mutex_init_task(p);
1936
1937 lockdep_assert_irqs_enabled();
1938#ifdef CONFIG_PROVE_LOCKING
1939 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1940#endif
1941 retval = -EAGAIN;
1942 if (atomic_read(&p->real_cred->user->processes) >=
1943 task_rlimit(p, RLIMIT_NPROC)) {
1944 if (p->real_cred->user != INIT_USER &&
1945 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1946 goto bad_fork_free;
1947 }
1948 current->flags &= ~PF_NPROC_EXCEEDED;
1949
1950 retval = copy_creds(p, clone_flags);
1951 if (retval < 0)
1952 goto bad_fork_free;
1953
1954 /*
1955 * If multiple threads are within copy_process(), then this check
1956 * triggers too late. This doesn't hurt, the check is only there
1957 * to stop root fork bombs.
1958 */
1959 retval = -EAGAIN;
1960 if (data_race(nr_threads >= max_threads))
1961 goto bad_fork_cleanup_count;
1962
1963 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
1964 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1965 p->flags |= PF_FORKNOEXEC;
1966 INIT_LIST_HEAD(&p->children);
1967 INIT_LIST_HEAD(&p->sibling);
1968 rcu_copy_process(p);
1969 p->vfork_done = NULL;
1970 spin_lock_init(&p->alloc_lock);
1971
1972 init_sigpending(&p->pending);
1973
1974 p->utime = p->stime = p->gtime = 0;
1975#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1976 p->utimescaled = p->stimescaled = 0;
1977#endif
1978 prev_cputime_init(&p->prev_cputime);
1979
1980#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1981 seqcount_init(&p->vtime.seqcount);
1982 p->vtime.starttime = 0;
1983 p->vtime.state = VTIME_INACTIVE;
1984#endif
1985
1986#if defined(SPLIT_RSS_COUNTING)
1987 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1988#endif
1989
1990 p->default_timer_slack_ns = current->timer_slack_ns;
1991
1992#ifdef CONFIG_PSI
1993 p->psi_flags = 0;
1994#endif
1995
1996 task_io_accounting_init(&p->ioac);
1997 acct_clear_integrals(p);
1998
1999 posix_cputimers_init(&p->posix_cputimers);
2000
2001 p->io_context = NULL;
2002 audit_set_context(p, NULL);
2003 cgroup_fork(p);
2004#ifdef CONFIG_NUMA
2005 p->mempolicy = mpol_dup(p->mempolicy);
2006 if (IS_ERR(p->mempolicy)) {
2007 retval = PTR_ERR(p->mempolicy);
2008 p->mempolicy = NULL;
2009 goto bad_fork_cleanup_threadgroup_lock;
2010 }
2011#endif
2012#ifdef CONFIG_CPUSETS
2013 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2014 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2015 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2016#endif
2017#ifdef CONFIG_TRACE_IRQFLAGS
2018 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2019 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2020 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2021 p->softirqs_enabled = 1;
2022 p->softirq_context = 0;
2023#endif
2024
2025 p->pagefault_disabled = 0;
2026
2027#ifdef CONFIG_LOCKDEP
2028 lockdep_init_task(p);
2029#endif
2030
2031#ifdef CONFIG_DEBUG_MUTEXES
2032 p->blocked_on = NULL; /* not blocked yet */
2033#endif
2034#ifdef CONFIG_BCACHE
2035 p->sequential_io = 0;
2036 p->sequential_io_avg = 0;
2037#endif
2038
2039 /* Perform scheduler related setup. Assign this task to a CPU. */
2040 retval = sched_fork(clone_flags, p);
2041 if (retval)
2042 goto bad_fork_cleanup_policy;
2043
2044 retval = perf_event_init_task(p);
2045 if (retval)
2046 goto bad_fork_cleanup_policy;
2047 retval = audit_alloc(p);
2048 if (retval)
2049 goto bad_fork_cleanup_perf;
2050 /* copy all the process information */
2051 shm_init_task(p);
2052 retval = security_task_alloc(p, clone_flags);
2053 if (retval)
2054 goto bad_fork_cleanup_audit;
2055 retval = copy_semundo(clone_flags, p);
2056 if (retval)
2057 goto bad_fork_cleanup_security;
2058 retval = copy_files(clone_flags, p);
2059 if (retval)
2060 goto bad_fork_cleanup_semundo;
2061 retval = copy_fs(clone_flags, p);
2062 if (retval)
2063 goto bad_fork_cleanup_files;
2064 retval = copy_sighand(clone_flags, p);
2065 if (retval)
2066 goto bad_fork_cleanup_fs;
2067 retval = copy_signal(clone_flags, p);
2068 if (retval)
2069 goto bad_fork_cleanup_sighand;
2070 retval = copy_mm(clone_flags, p);
2071 if (retval)
2072 goto bad_fork_cleanup_signal;
2073 retval = copy_namespaces(clone_flags, p);
2074 if (retval)
2075 goto bad_fork_cleanup_mm;
2076 retval = copy_io(clone_flags, p);
2077 if (retval)
2078 goto bad_fork_cleanup_namespaces;
2079 retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2080 if (retval)
2081 goto bad_fork_cleanup_io;
2082
2083 stackleak_task_init(p);
2084
2085 if (pid != &init_struct_pid) {
2086 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2087 args->set_tid_size);
2088 if (IS_ERR(pid)) {
2089 retval = PTR_ERR(pid);
2090 goto bad_fork_cleanup_thread;
2091 }
2092 }
2093
2094 /*
2095 * This has to happen after we've potentially unshared the file
2096 * descriptor table (so that the pidfd doesn't leak into the child
2097 * if the fd table isn't shared).
2098 */
2099 if (clone_flags & CLONE_PIDFD) {
2100 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2101 if (retval < 0)
2102 goto bad_fork_free_pid;
2103
2104 pidfd = retval;
2105
2106 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2107 O_RDWR | O_CLOEXEC);
2108 if (IS_ERR(pidfile)) {
2109 put_unused_fd(pidfd);
2110 retval = PTR_ERR(pidfile);
2111 goto bad_fork_free_pid;
2112 }
2113 get_pid(pid); /* held by pidfile now */
2114
2115 retval = put_user(pidfd, args->pidfd);
2116 if (retval)
2117 goto bad_fork_put_pidfd;
2118 }
2119
2120#ifdef CONFIG_BLOCK
2121 p->plug = NULL;
2122#endif
2123 futex_init_task(p);
2124
2125 /*
2126 * sigaltstack should be cleared when sharing the same VM
2127 */
2128 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2129 sas_ss_reset(p);
2130
2131 /*
2132 * Syscall tracing and stepping should be turned off in the
2133 * child regardless of CLONE_PTRACE.
2134 */
2135 user_disable_single_step(p);
2136 clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
2137#ifdef TIF_SYSCALL_EMU
2138 clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2139#endif
2140 clear_tsk_latency_tracing(p);
2141
2142 /* ok, now we should be set up.. */
2143 p->pid = pid_nr(pid);
2144 if (clone_flags & CLONE_THREAD) {
2145 p->exit_signal = -1;
2146 p->group_leader = current->group_leader;
2147 p->tgid = current->tgid;
2148 } else {
2149 if (clone_flags & CLONE_PARENT)
2150 p->exit_signal = current->group_leader->exit_signal;
2151 else
2152 p->exit_signal = args->exit_signal;
2153 p->group_leader = p;
2154 p->tgid = p->pid;
2155 }
2156
2157 p->nr_dirtied = 0;
2158 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2159 p->dirty_paused_when = 0;
2160
2161 p->pdeath_signal = 0;
2162 INIT_LIST_HEAD(&p->thread_group);
2163 p->task_works = NULL;
2164
2165 /*
2166 * Ensure that the cgroup subsystem policies allow the new process to be
2167 * forked. It should be noted the the new process's css_set can be changed
2168 * between here and cgroup_post_fork() if an organisation operation is in
2169 * progress.
2170 */
2171 retval = cgroup_can_fork(p, args);
2172 if (retval)
2173 goto bad_fork_put_pidfd;
2174
2175 /*
2176 * From this point on we must avoid any synchronous user-space
2177 * communication until we take the tasklist-lock. In particular, we do
2178 * not want user-space to be able to predict the process start-time by
2179 * stalling fork(2) after we recorded the start_time but before it is
2180 * visible to the system.
2181 */
2182
2183 p->start_time = ktime_get_ns();
2184 p->start_boottime = ktime_get_boottime_ns();
2185
2186 /*
2187 * Make it visible to the rest of the system, but dont wake it up yet.
2188 * Need tasklist lock for parent etc handling!
2189 */
2190 write_lock_irq(&tasklist_lock);
2191
2192 /* CLONE_PARENT re-uses the old parent */
2193 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2194 p->real_parent = current->real_parent;
2195 p->parent_exec_id = current->parent_exec_id;
2196 } else {
2197 p->real_parent = current;
2198 p->parent_exec_id = current->self_exec_id;
2199 }
2200
2201 klp_copy_process(p);
2202
2203 spin_lock(¤t->sighand->siglock);
2204
2205 /*
2206 * Copy seccomp details explicitly here, in case they were changed
2207 * before holding sighand lock.
2208 */
2209 copy_seccomp(p);
2210
2211 rseq_fork(p, clone_flags);
2212
2213 /* Don't start children in a dying pid namespace */
2214 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2215 retval = -ENOMEM;
2216 goto bad_fork_cancel_cgroup;
2217 }
2218
2219 /* Let kill terminate clone/fork in the middle */
2220 if (fatal_signal_pending(current)) {
2221 retval = -EINTR;
2222 goto bad_fork_cancel_cgroup;
2223 }
2224
2225 /* past the last point of failure */
2226 if (pidfile)
2227 fd_install(pidfd, pidfile);
2228
2229 init_task_pid_links(p);
2230 if (likely(p->pid)) {
2231 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2232
2233 init_task_pid(p, PIDTYPE_PID, pid);
2234 if (thread_group_leader(p)) {
2235 init_task_pid(p, PIDTYPE_TGID, pid);
2236 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2237 init_task_pid(p, PIDTYPE_SID, task_session(current));
2238
2239 if (is_child_reaper(pid)) {
2240 ns_of_pid(pid)->child_reaper = p;
2241 p->signal->flags |= SIGNAL_UNKILLABLE;
2242 }
2243 p->signal->shared_pending.signal = delayed.signal;
2244 p->signal->tty = tty_kref_get(current->signal->tty);
2245 /*
2246 * Inherit has_child_subreaper flag under the same
2247 * tasklist_lock with adding child to the process tree
2248 * for propagate_has_child_subreaper optimization.
2249 */
2250 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2251 p->real_parent->signal->is_child_subreaper;
2252 list_add_tail(&p->sibling, &p->real_parent->children);
2253 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2254 attach_pid(p, PIDTYPE_TGID);
2255 attach_pid(p, PIDTYPE_PGID);
2256 attach_pid(p, PIDTYPE_SID);
2257 __this_cpu_inc(process_counts);
2258 } else {
2259 current->signal->nr_threads++;
2260 atomic_inc(¤t->signal->live);
2261 refcount_inc(¤t->signal->sigcnt);
2262 task_join_group_stop(p);
2263 list_add_tail_rcu(&p->thread_group,
2264 &p->group_leader->thread_group);
2265 list_add_tail_rcu(&p->thread_node,
2266 &p->signal->thread_head);
2267 }
2268 attach_pid(p, PIDTYPE_PID);
2269 nr_threads++;
2270 }
2271 total_forks++;
2272 hlist_del_init(&delayed.node);
2273 spin_unlock(¤t->sighand->siglock);
2274 syscall_tracepoint_update(p);
2275 write_unlock_irq(&tasklist_lock);
2276
2277 proc_fork_connector(p);
2278 sched_post_fork(p);
2279 cgroup_post_fork(p, args);
2280 perf_event_fork(p);
2281
2282 trace_task_newtask(p, clone_flags);
2283 uprobe_copy_process(p, clone_flags);
2284
2285 return p;
2286
2287bad_fork_cancel_cgroup:
2288 spin_unlock(¤t->sighand->siglock);
2289 write_unlock_irq(&tasklist_lock);
2290 cgroup_cancel_fork(p, args);
2291bad_fork_put_pidfd:
2292 if (clone_flags & CLONE_PIDFD) {
2293 fput(pidfile);
2294 put_unused_fd(pidfd);
2295 }
2296bad_fork_free_pid:
2297 if (pid != &init_struct_pid)
2298 free_pid(pid);
2299bad_fork_cleanup_thread:
2300 exit_thread(p);
2301bad_fork_cleanup_io:
2302 if (p->io_context)
2303 exit_io_context(p);
2304bad_fork_cleanup_namespaces:
2305 exit_task_namespaces(p);
2306bad_fork_cleanup_mm:
2307 if (p->mm) {
2308 mm_clear_owner(p->mm, p);
2309 mmput(p->mm);
2310 }
2311bad_fork_cleanup_signal:
2312 if (!(clone_flags & CLONE_THREAD))
2313 free_signal_struct(p->signal);
2314bad_fork_cleanup_sighand:
2315 __cleanup_sighand(p->sighand);
2316bad_fork_cleanup_fs:
2317 exit_fs(p); /* blocking */
2318bad_fork_cleanup_files:
2319 exit_files(p); /* blocking */
2320bad_fork_cleanup_semundo:
2321 exit_sem(p);
2322bad_fork_cleanup_security:
2323 security_task_free(p);
2324bad_fork_cleanup_audit:
2325 audit_free(p);
2326bad_fork_cleanup_perf:
2327 perf_event_free_task(p);
2328bad_fork_cleanup_policy:
2329 lockdep_free_task(p);
2330#ifdef CONFIG_NUMA
2331 mpol_put(p->mempolicy);
2332bad_fork_cleanup_threadgroup_lock:
2333#endif
2334 delayacct_tsk_free(p);
2335bad_fork_cleanup_count:
2336 atomic_dec(&p->cred->user->processes);
2337 exit_creds(p);
2338bad_fork_free:
2339 p->state = TASK_DEAD;
2340 put_task_stack(p);
2341 delayed_free_task(p);
2342fork_out:
2343 spin_lock_irq(¤t->sighand->siglock);
2344 hlist_del_init(&delayed.node);
2345 spin_unlock_irq(¤t->sighand->siglock);
2346 return ERR_PTR(retval);
2347}
2348
2349static inline void init_idle_pids(struct task_struct *idle)
2350{
2351 enum pid_type type;
2352
2353 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2354 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2355 init_task_pid(idle, type, &init_struct_pid);
2356 }
2357}
2358
2359struct task_struct *fork_idle(int cpu)
2360{
2361 struct task_struct *task;
2362 struct kernel_clone_args args = {
2363 .flags = CLONE_VM,
2364 };
2365
2366 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2367 if (!IS_ERR(task)) {
2368 init_idle_pids(task);
2369 init_idle(task, cpu);
2370 }
2371
2372 return task;
2373}
2374
2375struct mm_struct *copy_init_mm(void)
2376{
2377 return dup_mm(NULL, &init_mm);
2378}
2379
2380/*
2381 * Ok, this is the main fork-routine.
2382 *
2383 * It copies the process, and if successful kick-starts
2384 * it and waits for it to finish using the VM if required.
2385 *
2386 * args->exit_signal is expected to be checked for sanity by the caller.
2387 */
2388long _do_fork(struct kernel_clone_args *args)
2389{
2390 u64 clone_flags = args->flags;
2391 struct completion vfork;
2392 struct pid *pid;
2393 struct task_struct *p;
2394 int trace = 0;
2395 long nr;
2396
2397 /*
2398 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2399 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2400 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2401 * field in struct clone_args and it still doesn't make sense to have
2402 * them both point at the same memory location. Performing this check
2403 * here has the advantage that we don't need to have a separate helper
2404 * to check for legacy clone().
2405 */
2406 if ((args->flags & CLONE_PIDFD) &&
2407 (args->flags & CLONE_PARENT_SETTID) &&
2408 (args->pidfd == args->parent_tid))
2409 return -EINVAL;
2410
2411 /*
2412 * Determine whether and which event to report to ptracer. When
2413 * called from kernel_thread or CLONE_UNTRACED is explicitly
2414 * requested, no event is reported; otherwise, report if the event
2415 * for the type of forking is enabled.
2416 */
2417 if (!(clone_flags & CLONE_UNTRACED)) {
2418 if (clone_flags & CLONE_VFORK)
2419 trace = PTRACE_EVENT_VFORK;
2420 else if (args->exit_signal != SIGCHLD)
2421 trace = PTRACE_EVENT_CLONE;
2422 else
2423 trace = PTRACE_EVENT_FORK;
2424
2425 if (likely(!ptrace_event_enabled(current, trace)))
2426 trace = 0;
2427 }
2428
2429 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2430 add_latent_entropy();
2431
2432 if (IS_ERR(p))
2433 return PTR_ERR(p);
2434
2435 /*
2436 * Do this prior waking up the new thread - the thread pointer
2437 * might get invalid after that point, if the thread exits quickly.
2438 */
2439 trace_sched_process_fork(current, p);
2440
2441 pid = get_task_pid(p, PIDTYPE_PID);
2442 nr = pid_vnr(pid);
2443
2444 if (clone_flags & CLONE_PARENT_SETTID)
2445 put_user(nr, args->parent_tid);
2446
2447 if (clone_flags & CLONE_VFORK) {
2448 p->vfork_done = &vfork;
2449 init_completion(&vfork);
2450 get_task_struct(p);
2451 }
2452
2453 wake_up_new_task(p);
2454
2455 /* forking complete and child started to run, tell ptracer */
2456 if (unlikely(trace))
2457 ptrace_event_pid(trace, pid);
2458
2459 if (clone_flags & CLONE_VFORK) {
2460 if (!wait_for_vfork_done(p, &vfork))
2461 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2462 }
2463
2464 put_pid(pid);
2465 return nr;
2466}
2467
2468/*
2469 * Create a kernel thread.
2470 */
2471pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2472{
2473 struct kernel_clone_args args = {
2474 .flags = ((lower_32_bits(flags) | CLONE_VM |
2475 CLONE_UNTRACED) & ~CSIGNAL),
2476 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2477 .stack = (unsigned long)fn,
2478 .stack_size = (unsigned long)arg,
2479 };
2480
2481 return _do_fork(&args);
2482}
2483
2484#ifdef __ARCH_WANT_SYS_FORK
2485SYSCALL_DEFINE0(fork)
2486{
2487#ifdef CONFIG_MMU
2488 struct kernel_clone_args args = {
2489 .exit_signal = SIGCHLD,
2490 };
2491
2492 return _do_fork(&args);
2493#else
2494 /* can not support in nommu mode */
2495 return -EINVAL;
2496#endif
2497}
2498#endif
2499
2500#ifdef __ARCH_WANT_SYS_VFORK
2501SYSCALL_DEFINE0(vfork)
2502{
2503 struct kernel_clone_args args = {
2504 .flags = CLONE_VFORK | CLONE_VM,
2505 .exit_signal = SIGCHLD,
2506 };
2507
2508 return _do_fork(&args);
2509}
2510#endif
2511
2512#ifdef __ARCH_WANT_SYS_CLONE
2513#ifdef CONFIG_CLONE_BACKWARDS
2514SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2515 int __user *, parent_tidptr,
2516 unsigned long, tls,
2517 int __user *, child_tidptr)
2518#elif defined(CONFIG_CLONE_BACKWARDS2)
2519SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2520 int __user *, parent_tidptr,
2521 int __user *, child_tidptr,
2522 unsigned long, tls)
2523#elif defined(CONFIG_CLONE_BACKWARDS3)
2524SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2525 int, stack_size,
2526 int __user *, parent_tidptr,
2527 int __user *, child_tidptr,
2528 unsigned long, tls)
2529#else
2530SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2531 int __user *, parent_tidptr,
2532 int __user *, child_tidptr,
2533 unsigned long, tls)
2534#endif
2535{
2536 struct kernel_clone_args args = {
2537 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2538 .pidfd = parent_tidptr,
2539 .child_tid = child_tidptr,
2540 .parent_tid = parent_tidptr,
2541 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2542 .stack = newsp,
2543 .tls = tls,
2544 };
2545
2546 return _do_fork(&args);
2547}
2548#endif
2549
2550#ifdef __ARCH_WANT_SYS_CLONE3
2551
2552noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2553 struct clone_args __user *uargs,
2554 size_t usize)
2555{
2556 int err;
2557 struct clone_args args;
2558 pid_t *kset_tid = kargs->set_tid;
2559
2560 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2561 CLONE_ARGS_SIZE_VER0);
2562 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2563 CLONE_ARGS_SIZE_VER1);
2564 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2565 CLONE_ARGS_SIZE_VER2);
2566 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2567
2568 if (unlikely(usize > PAGE_SIZE))
2569 return -E2BIG;
2570 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2571 return -EINVAL;
2572
2573 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2574 if (err)
2575 return err;
2576
2577 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2578 return -EINVAL;
2579
2580 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2581 return -EINVAL;
2582
2583 if (unlikely(args.set_tid && args.set_tid_size == 0))
2584 return -EINVAL;
2585
2586 /*
2587 * Verify that higher 32bits of exit_signal are unset and that
2588 * it is a valid signal
2589 */
2590 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2591 !valid_signal(args.exit_signal)))
2592 return -EINVAL;
2593
2594 if ((args.flags & CLONE_INTO_CGROUP) &&
2595 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2596 return -EINVAL;
2597
2598 *kargs = (struct kernel_clone_args){
2599 .flags = args.flags,
2600 .pidfd = u64_to_user_ptr(args.pidfd),
2601 .child_tid = u64_to_user_ptr(args.child_tid),
2602 .parent_tid = u64_to_user_ptr(args.parent_tid),
2603 .exit_signal = args.exit_signal,
2604 .stack = args.stack,
2605 .stack_size = args.stack_size,
2606 .tls = args.tls,
2607 .set_tid_size = args.set_tid_size,
2608 .cgroup = args.cgroup,
2609 };
2610
2611 if (args.set_tid &&
2612 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2613 (kargs->set_tid_size * sizeof(pid_t))))
2614 return -EFAULT;
2615
2616 kargs->set_tid = kset_tid;
2617
2618 return 0;
2619}
2620
2621/**
2622 * clone3_stack_valid - check and prepare stack
2623 * @kargs: kernel clone args
2624 *
2625 * Verify that the stack arguments userspace gave us are sane.
2626 * In addition, set the stack direction for userspace since it's easy for us to
2627 * determine.
2628 */
2629static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2630{
2631 if (kargs->stack == 0) {
2632 if (kargs->stack_size > 0)
2633 return false;
2634 } else {
2635 if (kargs->stack_size == 0)
2636 return false;
2637
2638 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2639 return false;
2640
2641#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2642 kargs->stack += kargs->stack_size;
2643#endif
2644 }
2645
2646 return true;
2647}
2648
2649static bool clone3_args_valid(struct kernel_clone_args *kargs)
2650{
2651 /* Verify that no unknown flags are passed along. */
2652 if (kargs->flags &
2653 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2654 return false;
2655
2656 /*
2657 * - make the CLONE_DETACHED bit reuseable for clone3
2658 * - make the CSIGNAL bits reuseable for clone3
2659 */
2660 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2661 return false;
2662
2663 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2664 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2665 return false;
2666
2667 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2668 kargs->exit_signal)
2669 return false;
2670
2671 if (!clone3_stack_valid(kargs))
2672 return false;
2673
2674 return true;
2675}
2676
2677/**
2678 * clone3 - create a new process with specific properties
2679 * @uargs: argument structure
2680 * @size: size of @uargs
2681 *
2682 * clone3() is the extensible successor to clone()/clone2().
2683 * It takes a struct as argument that is versioned by its size.
2684 *
2685 * Return: On success, a positive PID for the child process.
2686 * On error, a negative errno number.
2687 */
2688SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2689{
2690 int err;
2691
2692 struct kernel_clone_args kargs;
2693 pid_t set_tid[MAX_PID_NS_LEVEL];
2694
2695 kargs.set_tid = set_tid;
2696
2697 err = copy_clone_args_from_user(&kargs, uargs, size);
2698 if (err)
2699 return err;
2700
2701 if (!clone3_args_valid(&kargs))
2702 return -EINVAL;
2703
2704 return _do_fork(&kargs);
2705}
2706#endif
2707
2708void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2709{
2710 struct task_struct *leader, *parent, *child;
2711 int res;
2712
2713 read_lock(&tasklist_lock);
2714 leader = top = top->group_leader;
2715down:
2716 for_each_thread(leader, parent) {
2717 list_for_each_entry(child, &parent->children, sibling) {
2718 res = visitor(child, data);
2719 if (res) {
2720 if (res < 0)
2721 goto out;
2722 leader = child;
2723 goto down;
2724 }
2725up:
2726 ;
2727 }
2728 }
2729
2730 if (leader != top) {
2731 child = leader;
2732 parent = child->real_parent;
2733 leader = parent->group_leader;
2734 goto up;
2735 }
2736out:
2737 read_unlock(&tasklist_lock);
2738}
2739
2740#ifndef ARCH_MIN_MMSTRUCT_ALIGN
2741#define ARCH_MIN_MMSTRUCT_ALIGN 0
2742#endif
2743
2744static void sighand_ctor(void *data)
2745{
2746 struct sighand_struct *sighand = data;
2747
2748 spin_lock_init(&sighand->siglock);
2749 init_waitqueue_head(&sighand->signalfd_wqh);
2750}
2751
2752void __init proc_caches_init(void)
2753{
2754 unsigned int mm_size;
2755
2756 sighand_cachep = kmem_cache_create("sighand_cache",
2757 sizeof(struct sighand_struct), 0,
2758 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2759 SLAB_ACCOUNT, sighand_ctor);
2760 signal_cachep = kmem_cache_create("signal_cache",
2761 sizeof(struct signal_struct), 0,
2762 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2763 NULL);
2764 files_cachep = kmem_cache_create("files_cache",
2765 sizeof(struct files_struct), 0,
2766 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2767 NULL);
2768 fs_cachep = kmem_cache_create("fs_cache",
2769 sizeof(struct fs_struct), 0,
2770 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2771 NULL);
2772
2773 /*
2774 * The mm_cpumask is located at the end of mm_struct, and is
2775 * dynamically sized based on the maximum CPU number this system
2776 * can have, taking hotplug into account (nr_cpu_ids).
2777 */
2778 mm_size = sizeof(struct mm_struct) + cpumask_size();
2779
2780 mm_cachep = kmem_cache_create_usercopy("mm_struct",
2781 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2782 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2783 offsetof(struct mm_struct, saved_auxv),
2784 sizeof_field(struct mm_struct, saved_auxv),
2785 NULL);
2786 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2787 mmap_init();
2788 nsproxy_cache_init();
2789}
2790
2791/*
2792 * Check constraints on flags passed to the unshare system call.
2793 */
2794static int check_unshare_flags(unsigned long unshare_flags)
2795{
2796 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2797 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2798 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2799 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2800 CLONE_NEWTIME))
2801 return -EINVAL;
2802 /*
2803 * Not implemented, but pretend it works if there is nothing
2804 * to unshare. Note that unsharing the address space or the
2805 * signal handlers also need to unshare the signal queues (aka
2806 * CLONE_THREAD).
2807 */
2808 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2809 if (!thread_group_empty(current))
2810 return -EINVAL;
2811 }
2812 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2813 if (refcount_read(¤t->sighand->count) > 1)
2814 return -EINVAL;
2815 }
2816 if (unshare_flags & CLONE_VM) {
2817 if (!current_is_single_threaded())
2818 return -EINVAL;
2819 }
2820
2821 return 0;
2822}
2823
2824/*
2825 * Unshare the filesystem structure if it is being shared
2826 */
2827static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2828{
2829 struct fs_struct *fs = current->fs;
2830
2831 if (!(unshare_flags & CLONE_FS) || !fs)
2832 return 0;
2833
2834 /* don't need lock here; in the worst case we'll do useless copy */
2835 if (fs->users == 1)
2836 return 0;
2837
2838 *new_fsp = copy_fs_struct(fs);
2839 if (!*new_fsp)
2840 return -ENOMEM;
2841
2842 return 0;
2843}
2844
2845/*
2846 * Unshare file descriptor table if it is being shared
2847 */
2848int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2849 struct files_struct **new_fdp)
2850{
2851 struct files_struct *fd = current->files;
2852 int error = 0;
2853
2854 if ((unshare_flags & CLONE_FILES) &&
2855 (fd && atomic_read(&fd->count) > 1)) {
2856 *new_fdp = dup_fd(fd, max_fds, &error);
2857 if (!*new_fdp)
2858 return error;
2859 }
2860
2861 return 0;
2862}
2863
2864/*
2865 * unshare allows a process to 'unshare' part of the process
2866 * context which was originally shared using clone. copy_*
2867 * functions used by _do_fork() cannot be used here directly
2868 * because they modify an inactive task_struct that is being
2869 * constructed. Here we are modifying the current, active,
2870 * task_struct.
2871 */
2872int ksys_unshare(unsigned long unshare_flags)
2873{
2874 struct fs_struct *fs, *new_fs = NULL;
2875 struct files_struct *fd, *new_fd = NULL;
2876 struct cred *new_cred = NULL;
2877 struct nsproxy *new_nsproxy = NULL;
2878 int do_sysvsem = 0;
2879 int err;
2880
2881 /*
2882 * If unsharing a user namespace must also unshare the thread group
2883 * and unshare the filesystem root and working directories.
2884 */
2885 if (unshare_flags & CLONE_NEWUSER)
2886 unshare_flags |= CLONE_THREAD | CLONE_FS;
2887 /*
2888 * If unsharing vm, must also unshare signal handlers.
2889 */
2890 if (unshare_flags & CLONE_VM)
2891 unshare_flags |= CLONE_SIGHAND;
2892 /*
2893 * If unsharing a signal handlers, must also unshare the signal queues.
2894 */
2895 if (unshare_flags & CLONE_SIGHAND)
2896 unshare_flags |= CLONE_THREAD;
2897 /*
2898 * If unsharing namespace, must also unshare filesystem information.
2899 */
2900 if (unshare_flags & CLONE_NEWNS)
2901 unshare_flags |= CLONE_FS;
2902
2903 err = check_unshare_flags(unshare_flags);
2904 if (err)
2905 goto bad_unshare_out;
2906 /*
2907 * CLONE_NEWIPC must also detach from the undolist: after switching
2908 * to a new ipc namespace, the semaphore arrays from the old
2909 * namespace are unreachable.
2910 */
2911 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2912 do_sysvsem = 1;
2913 err = unshare_fs(unshare_flags, &new_fs);
2914 if (err)
2915 goto bad_unshare_out;
2916 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2917 if (err)
2918 goto bad_unshare_cleanup_fs;
2919 err = unshare_userns(unshare_flags, &new_cred);
2920 if (err)
2921 goto bad_unshare_cleanup_fd;
2922 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2923 new_cred, new_fs);
2924 if (err)
2925 goto bad_unshare_cleanup_cred;
2926
2927 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2928 if (do_sysvsem) {
2929 /*
2930 * CLONE_SYSVSEM is equivalent to sys_exit().
2931 */
2932 exit_sem(current);
2933 }
2934 if (unshare_flags & CLONE_NEWIPC) {
2935 /* Orphan segments in old ns (see sem above). */
2936 exit_shm(current);
2937 shm_init_task(current);
2938 }
2939
2940 if (new_nsproxy)
2941 switch_task_namespaces(current, new_nsproxy);
2942
2943 task_lock(current);
2944
2945 if (new_fs) {
2946 fs = current->fs;
2947 spin_lock(&fs->lock);
2948 current->fs = new_fs;
2949 if (--fs->users)
2950 new_fs = NULL;
2951 else
2952 new_fs = fs;
2953 spin_unlock(&fs->lock);
2954 }
2955
2956 if (new_fd) {
2957 fd = current->files;
2958 current->files = new_fd;
2959 new_fd = fd;
2960 }
2961
2962 task_unlock(current);
2963
2964 if (new_cred) {
2965 /* Install the new user namespace */
2966 commit_creds(new_cred);
2967 new_cred = NULL;
2968 }
2969 }
2970
2971 perf_event_namespaces(current);
2972
2973bad_unshare_cleanup_cred:
2974 if (new_cred)
2975 put_cred(new_cred);
2976bad_unshare_cleanup_fd:
2977 if (new_fd)
2978 put_files_struct(new_fd);
2979
2980bad_unshare_cleanup_fs:
2981 if (new_fs)
2982 free_fs_struct(new_fs);
2983
2984bad_unshare_out:
2985 return err;
2986}
2987
2988SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2989{
2990 return ksys_unshare(unshare_flags);
2991}
2992
2993/*
2994 * Helper to unshare the files of the current task.
2995 * We don't want to expose copy_files internals to
2996 * the exec layer of the kernel.
2997 */
2998
2999int unshare_files(struct files_struct **displaced)
3000{
3001 struct task_struct *task = current;
3002 struct files_struct *copy = NULL;
3003 int error;
3004
3005 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3006 if (error || !copy) {
3007 *displaced = NULL;
3008 return error;
3009 }
3010 *displaced = task->files;
3011 task_lock(task);
3012 task->files = copy;
3013 task_unlock(task);
3014 return 0;
3015}
3016
3017int sysctl_max_threads(struct ctl_table *table, int write,
3018 void *buffer, size_t *lenp, loff_t *ppos)
3019{
3020 struct ctl_table t;
3021 int ret;
3022 int threads = max_threads;
3023 int min = 1;
3024 int max = MAX_THREADS;
3025
3026 t = *table;
3027 t.data = &threads;
3028 t.extra1 = &min;
3029 t.extra2 = &max;
3030
3031 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3032 if (ret || !write)
3033 return ret;
3034
3035 max_threads = threads;
3036
3037 return 0;
3038}