projects/linux·blob v6.3 · refs

3323 lines · 80.1 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/jiffies.h>
57#include <linux/futex.h>
58#include <linux/compat.h>
59#include <linux/kthread.h>
60#include <linux/task_io_accounting_ops.h>
61#include <linux/rcupdate.h>
62#include <linux/ptrace.h>
63#include <linux/mount.h>
64#include <linux/audit.h>
65#include <linux/memcontrol.h>
66#include <linux/ftrace.h>
67#include <linux/proc_fs.h>
68#include <linux/profile.h>
69#include <linux/rmap.h>
70#include <linux/ksm.h>
71#include <linux/acct.h>
72#include <linux/userfaultfd_k.h>
73#include <linux/tsacct_kern.h>
74#include <linux/cn_proc.h>
75#include <linux/freezer.h>
76#include <linux/delayacct.h>
77#include <linux/taskstats_kern.h>
78#include <linux/tty.h>
79#include <linux/fs_struct.h>
80#include <linux/magic.h>
81#include <linux/perf_event.h>
82#include <linux/posix-timers.h>
83#include <linux/user-return-notifier.h>
84#include <linux/oom.h>
85#include <linux/khugepaged.h>
86#include <linux/signalfd.h>
87#include <linux/uprobes.h>
88#include <linux/aio.h>
89#include <linux/compiler.h>
90#include <linux/sysctl.h>
91#include <linux/kcov.h>
92#include <linux/livepatch.h>
93#include <linux/thread_info.h>
94#include <linux/stackleak.h>
95#include <linux/kasan.h>
96#include <linux/scs.h>
97#include <linux/io_uring.h>
98#include <linux/bpf.h>
99#include <linux/stackprotector.h>
100
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 data_race(memcpy(new, orig, sizeof(*new)));
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 VMA_ITERATOR(old_vmi, oldmm, 0);
589 VMA_ITERATOR(vmi, mm, 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 = vma_iter_bulk_alloc(&vmi, oldmm->map_count);
617 if (retval)
618 goto out;
619
620 mt_clear_in_rcu(vmi.mas.tree);
621 for_each_vma(old_vmi, mpnt) {
622 struct file *file;
623
624 if (mpnt->vm_flags & VM_DONTCOPY) {
625 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
626 continue;
627 }
628 charge = 0;
629 /*
630 * Don't duplicate many vmas if we've been oom-killed (for
631 * example)
632 */
633 if (fatal_signal_pending(current)) {
634 retval = -EINTR;
635 goto loop_out;
636 }
637 if (mpnt->vm_flags & VM_ACCOUNT) {
638 unsigned long len = vma_pages(mpnt);
639
640 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
641 goto fail_nomem;
642 charge = len;
643 }
644 tmp = vm_area_dup(mpnt);
645 if (!tmp)
646 goto fail_nomem;
647 retval = vma_dup_policy(mpnt, tmp);
648 if (retval)
649 goto fail_nomem_policy;
650 tmp->vm_mm = mm;
651 retval = dup_userfaultfd(tmp, &uf);
652 if (retval)
653 goto fail_nomem_anon_vma_fork;
654 if (tmp->vm_flags & VM_WIPEONFORK) {
655 /*
656 * VM_WIPEONFORK gets a clean slate in the child.
657 * Don't prepare anon_vma until fault since we don't
658 * copy page for current vma.
659 */
660 tmp->anon_vma = NULL;
661 } else if (anon_vma_fork(tmp, mpnt))
662 goto fail_nomem_anon_vma_fork;
663 vm_flags_clear(tmp, VM_LOCKED_MASK);
664 file = tmp->vm_file;
665 if (file) {
666 struct address_space *mapping = file->f_mapping;
667
668 get_file(file);
669 i_mmap_lock_write(mapping);
670 if (tmp->vm_flags & VM_SHARED)
671 mapping_allow_writable(mapping);
672 flush_dcache_mmap_lock(mapping);
673 /* insert tmp into the share list, just after mpnt */
674 vma_interval_tree_insert_after(tmp, mpnt,
675 &mapping->i_mmap);
676 flush_dcache_mmap_unlock(mapping);
677 i_mmap_unlock_write(mapping);
678 }
679
680 /*
681 * Copy/update hugetlb private vma information.
682 */
683 if (is_vm_hugetlb_page(tmp))
684 hugetlb_dup_vma_private(tmp);
685
686 /* Link the vma into the MT */
687 if (vma_iter_bulk_store(&vmi, tmp))
688 goto fail_nomem_vmi_store;
689
690 mm->map_count++;
691 if (!(tmp->vm_flags & VM_WIPEONFORK))
692 retval = copy_page_range(tmp, mpnt);
693
694 if (tmp->vm_ops && tmp->vm_ops->open)
695 tmp->vm_ops->open(tmp);
696
697 if (retval)
698 goto loop_out;
699 }
700 /* a new mm has just been created */
701 retval = arch_dup_mmap(oldmm, mm);
702loop_out:
703 vma_iter_free(&vmi);
704 if (!retval)
705 mt_set_in_rcu(vmi.mas.tree);
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_vmi_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 (unlikely(x))
762 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
763 mm, resident_page_types[i], x);
764 }
765
766 if (mm_pgtables_bytes(mm))
767 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
768 mm_pgtables_bytes(mm));
769
770#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
771 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
772#endif
773}
774
775#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
776#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
777
778/*
779 * Called when the last reference to the mm
780 * is dropped: either by a lazy thread or by
781 * mmput. Free the page directory and the mm.
782 */
783void __mmdrop(struct mm_struct *mm)
784{
785 int i;
786
787 BUG_ON(mm == &init_mm);
788 WARN_ON_ONCE(mm == current->mm);
789 WARN_ON_ONCE(mm == current->active_mm);
790 mm_free_pgd(mm);
791 destroy_context(mm);
792 mmu_notifier_subscriptions_destroy(mm);
793 check_mm(mm);
794 put_user_ns(mm->user_ns);
795 mm_pasid_drop(mm);
796
797 for (i = 0; i < NR_MM_COUNTERS; i++)
798 percpu_counter_destroy(&mm->rss_stat[i]);
799 free_mm(mm);
800}
801EXPORT_SYMBOL_GPL(__mmdrop);
802
803static void mmdrop_async_fn(struct work_struct *work)
804{
805 struct mm_struct *mm;
806
807 mm = container_of(work, struct mm_struct, async_put_work);
808 __mmdrop(mm);
809}
810
811static void mmdrop_async(struct mm_struct *mm)
812{
813 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
814 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
815 schedule_work(&mm->async_put_work);
816 }
817}
818
819static inline void free_signal_struct(struct signal_struct *sig)
820{
821 taskstats_tgid_free(sig);
822 sched_autogroup_exit(sig);
823 /*
824 * __mmdrop is not safe to call from softirq context on x86 due to
825 * pgd_dtor so postpone it to the async context
826 */
827 if (sig->oom_mm)
828 mmdrop_async(sig->oom_mm);
829 kmem_cache_free(signal_cachep, sig);
830}
831
832static inline void put_signal_struct(struct signal_struct *sig)
833{
834 if (refcount_dec_and_test(&sig->sigcnt))
835 free_signal_struct(sig);
836}
837
838void __put_task_struct(struct task_struct *tsk)
839{
840 WARN_ON(!tsk->exit_state);
841 WARN_ON(refcount_read(&tsk->usage));
842 WARN_ON(tsk == current);
843
844 io_uring_free(tsk);
845 cgroup_free(tsk);
846 task_numa_free(tsk, true);
847 security_task_free(tsk);
848 bpf_task_storage_free(tsk);
849 exit_creds(tsk);
850 delayacct_tsk_free(tsk);
851 put_signal_struct(tsk->signal);
852 sched_core_free(tsk);
853 free_task(tsk);
854}
855EXPORT_SYMBOL_GPL(__put_task_struct);
856
857void __init __weak arch_task_cache_init(void) { }
858
859/*
860 * set_max_threads
861 */
862static void set_max_threads(unsigned int max_threads_suggested)
863{
864 u64 threads;
865 unsigned long nr_pages = totalram_pages();
866
867 /*
868 * The number of threads shall be limited such that the thread
869 * structures may only consume a small part of the available memory.
870 */
871 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
872 threads = MAX_THREADS;
873 else
874 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
875 (u64) THREAD_SIZE * 8UL);
876
877 if (threads > max_threads_suggested)
878 threads = max_threads_suggested;
879
880 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
881}
882
883#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
884/* Initialized by the architecture: */
885int arch_task_struct_size __read_mostly;
886#endif
887
888#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
889static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
890{
891 /* Fetch thread_struct whitelist for the architecture. */
892 arch_thread_struct_whitelist(offset, size);
893
894 /*
895 * Handle zero-sized whitelist or empty thread_struct, otherwise
896 * adjust offset to position of thread_struct in task_struct.
897 */
898 if (unlikely(*size == 0))
899 *offset = 0;
900 else
901 *offset += offsetof(struct task_struct, thread);
902}
903#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
904
905void __init fork_init(void)
906{
907 int i;
908#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
909#ifndef ARCH_MIN_TASKALIGN
910#define ARCH_MIN_TASKALIGN 0
911#endif
912 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
913 unsigned long useroffset, usersize;
914
915 /* create a slab on which task_structs can be allocated */
916 task_struct_whitelist(&useroffset, &usersize);
917 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
918 arch_task_struct_size, align,
919 SLAB_PANIC|SLAB_ACCOUNT,
920 useroffset, usersize, NULL);
921#endif
922
923 /* do the arch specific task caches init */
924 arch_task_cache_init();
925
926 set_max_threads(MAX_THREADS);
927
928 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
929 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
930 init_task.signal->rlim[RLIMIT_SIGPENDING] =
931 init_task.signal->rlim[RLIMIT_NPROC];
932
933 for (i = 0; i < UCOUNT_COUNTS; i++)
934 init_user_ns.ucount_max[i] = max_threads/2;
935
936 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
937 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
938 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
939 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
940
941#ifdef CONFIG_VMAP_STACK
942 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
943 NULL, free_vm_stack_cache);
944#endif
945
946 scs_init();
947
948 lockdep_init_task(&init_task);
949 uprobes_init();
950}
951
952int __weak arch_dup_task_struct(struct task_struct *dst,
953 struct task_struct *src)
954{
955 *dst = *src;
956 return 0;
957}
958
959void set_task_stack_end_magic(struct task_struct *tsk)
960{
961 unsigned long *stackend;
962
963 stackend = end_of_stack(tsk);
964 *stackend = STACK_END_MAGIC; /* for overflow detection */
965}
966
967static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
968{
969 struct task_struct *tsk;
970 int err;
971
972 if (node == NUMA_NO_NODE)
973 node = tsk_fork_get_node(orig);
974 tsk = alloc_task_struct_node(node);
975 if (!tsk)
976 return NULL;
977
978 err = arch_dup_task_struct(tsk, orig);
979 if (err)
980 goto free_tsk;
981
982 err = alloc_thread_stack_node(tsk, node);
983 if (err)
984 goto free_tsk;
985
986#ifdef CONFIG_THREAD_INFO_IN_TASK
987 refcount_set(&tsk->stack_refcount, 1);
988#endif
989 account_kernel_stack(tsk, 1);
990
991 err = scs_prepare(tsk, node);
992 if (err)
993 goto free_stack;
994
995#ifdef CONFIG_SECCOMP
996 /*
997 * We must handle setting up seccomp filters once we're under
998 * the sighand lock in case orig has changed between now and
999 * then. Until then, filter must be NULL to avoid messing up
1000 * the usage counts on the error path calling free_task.
1001 */
1002 tsk->seccomp.filter = NULL;
1003#endif
1004
1005 setup_thread_stack(tsk, orig);
1006 clear_user_return_notifier(tsk);
1007 clear_tsk_need_resched(tsk);
1008 set_task_stack_end_magic(tsk);
1009 clear_syscall_work_syscall_user_dispatch(tsk);
1010
1011#ifdef CONFIG_STACKPROTECTOR
1012 tsk->stack_canary = get_random_canary();
1013#endif
1014 if (orig->cpus_ptr == &orig->cpus_mask)
1015 tsk->cpus_ptr = &tsk->cpus_mask;
1016 dup_user_cpus_ptr(tsk, orig, node);
1017
1018 /*
1019 * One for the user space visible state that goes away when reaped.
1020 * One for the scheduler.
1021 */
1022 refcount_set(&tsk->rcu_users, 2);
1023 /* One for the rcu users */
1024 refcount_set(&tsk->usage, 1);
1025#ifdef CONFIG_BLK_DEV_IO_TRACE
1026 tsk->btrace_seq = 0;
1027#endif
1028 tsk->splice_pipe = NULL;
1029 tsk->task_frag.page = NULL;
1030 tsk->wake_q.next = NULL;
1031 tsk->worker_private = NULL;
1032
1033 kcov_task_init(tsk);
1034 kmsan_task_create(tsk);
1035 kmap_local_fork(tsk);
1036
1037#ifdef CONFIG_FAULT_INJECTION
1038 tsk->fail_nth = 0;
1039#endif
1040
1041#ifdef CONFIG_BLK_CGROUP
1042 tsk->throttle_disk = NULL;
1043 tsk->use_memdelay = 0;
1044#endif
1045
1046#ifdef CONFIG_IOMMU_SVA
1047 tsk->pasid_activated = 0;
1048#endif
1049
1050#ifdef CONFIG_MEMCG
1051 tsk->active_memcg = NULL;
1052#endif
1053
1054#ifdef CONFIG_CPU_SUP_INTEL
1055 tsk->reported_split_lock = 0;
1056#endif
1057
1058#ifdef CONFIG_SCHED_MM_CID
1059 tsk->mm_cid = -1;
1060 tsk->mm_cid_active = 0;
1061#endif
1062 return tsk;
1063
1064free_stack:
1065 exit_task_stack_account(tsk);
1066 free_thread_stack(tsk);
1067free_tsk:
1068 free_task_struct(tsk);
1069 return NULL;
1070}
1071
1072__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1073
1074static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1075
1076static int __init coredump_filter_setup(char *s)
1077{
1078 default_dump_filter =
1079 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1080 MMF_DUMP_FILTER_MASK;
1081 return 1;
1082}
1083
1084__setup("coredump_filter=", coredump_filter_setup);
1085
1086#include <linux/init_task.h>
1087
1088static void mm_init_aio(struct mm_struct *mm)
1089{
1090#ifdef CONFIG_AIO
1091 spin_lock_init(&mm->ioctx_lock);
1092 mm->ioctx_table = NULL;
1093#endif
1094}
1095
1096static __always_inline void mm_clear_owner(struct mm_struct *mm,
1097 struct task_struct *p)
1098{
1099#ifdef CONFIG_MEMCG
1100 if (mm->owner == p)
1101 WRITE_ONCE(mm->owner, NULL);
1102#endif
1103}
1104
1105static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1106{
1107#ifdef CONFIG_MEMCG
1108 mm->owner = p;
1109#endif
1110}
1111
1112static void mm_init_uprobes_state(struct mm_struct *mm)
1113{
1114#ifdef CONFIG_UPROBES
1115 mm->uprobes_state.xol_area = NULL;
1116#endif
1117}
1118
1119static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1120 struct user_namespace *user_ns)
1121{
1122 int i;
1123
1124 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1125 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1126 atomic_set(&mm->mm_users, 1);
1127 atomic_set(&mm->mm_count, 1);
1128 seqcount_init(&mm->write_protect_seq);
1129 mmap_init_lock(mm);
1130 INIT_LIST_HEAD(&mm->mmlist);
1131 mm_pgtables_bytes_init(mm);
1132 mm->map_count = 0;
1133 mm->locked_vm = 0;
1134 atomic64_set(&mm->pinned_vm, 0);
1135 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1136 spin_lock_init(&mm->page_table_lock);
1137 spin_lock_init(&mm->arg_lock);
1138 mm_init_cpumask(mm);
1139 mm_init_aio(mm);
1140 mm_init_owner(mm, p);
1141 mm_pasid_init(mm);
1142 RCU_INIT_POINTER(mm->exe_file, NULL);
1143 mmu_notifier_subscriptions_init(mm);
1144 init_tlb_flush_pending(mm);
1145#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1146 mm->pmd_huge_pte = NULL;
1147#endif
1148 mm_init_uprobes_state(mm);
1149 hugetlb_count_init(mm);
1150
1151 if (current->mm) {
1152 mm->flags = current->mm->flags & MMF_INIT_MASK;
1153 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1154 } else {
1155 mm->flags = default_dump_filter;
1156 mm->def_flags = 0;
1157 }
1158
1159 if (mm_alloc_pgd(mm))
1160 goto fail_nopgd;
1161
1162 if (init_new_context(p, mm))
1163 goto fail_nocontext;
1164
1165 for (i = 0; i < NR_MM_COUNTERS; i++)
1166 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1167 goto fail_pcpu;
1168
1169 mm->user_ns = get_user_ns(user_ns);
1170 lru_gen_init_mm(mm);
1171 mm_init_cid(mm);
1172 return mm;
1173
1174fail_pcpu:
1175 while (i > 0)
1176 percpu_counter_destroy(&mm->rss_stat[--i]);
1177 destroy_context(mm);
1178fail_nocontext:
1179 mm_free_pgd(mm);
1180fail_nopgd:
1181 free_mm(mm);
1182 return NULL;
1183}
1184
1185/*
1186 * Allocate and initialize an mm_struct.
1187 */
1188struct mm_struct *mm_alloc(void)
1189{
1190 struct mm_struct *mm;
1191
1192 mm = allocate_mm();
1193 if (!mm)
1194 return NULL;
1195
1196 memset(mm, 0, sizeof(*mm));
1197 return mm_init(mm, current, current_user_ns());
1198}
1199
1200static inline void __mmput(struct mm_struct *mm)
1201{
1202 VM_BUG_ON(atomic_read(&mm->mm_users));
1203
1204 uprobe_clear_state(mm);
1205 exit_aio(mm);
1206 ksm_exit(mm);
1207 khugepaged_exit(mm); /* must run before exit_mmap */
1208 exit_mmap(mm);
1209 mm_put_huge_zero_page(mm);
1210 set_mm_exe_file(mm, NULL);
1211 if (!list_empty(&mm->mmlist)) {
1212 spin_lock(&mmlist_lock);
1213 list_del(&mm->mmlist);
1214 spin_unlock(&mmlist_lock);
1215 }
1216 if (mm->binfmt)
1217 module_put(mm->binfmt->module);
1218 lru_gen_del_mm(mm);
1219 mmdrop(mm);
1220}
1221
1222/*
1223 * Decrement the use count and release all resources for an mm.
1224 */
1225void mmput(struct mm_struct *mm)
1226{
1227 might_sleep();
1228
1229 if (atomic_dec_and_test(&mm->mm_users))
1230 __mmput(mm);
1231}
1232EXPORT_SYMBOL_GPL(mmput);
1233
1234#ifdef CONFIG_MMU
1235static void mmput_async_fn(struct work_struct *work)
1236{
1237 struct mm_struct *mm = container_of(work, struct mm_struct,
1238 async_put_work);
1239
1240 __mmput(mm);
1241}
1242
1243void mmput_async(struct mm_struct *mm)
1244{
1245 if (atomic_dec_and_test(&mm->mm_users)) {
1246 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1247 schedule_work(&mm->async_put_work);
1248 }
1249}
1250EXPORT_SYMBOL_GPL(mmput_async);
1251#endif
1252
1253/**
1254 * set_mm_exe_file - change a reference to the mm's executable file
1255 *
1256 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1257 *
1258 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1259 * invocations: in mmput() nobody alive left, in execve task is single
1260 * threaded.
1261 *
1262 * Can only fail if new_exe_file != NULL.
1263 */
1264int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1265{
1266 struct file *old_exe_file;
1267
1268 /*
1269 * It is safe to dereference the exe_file without RCU as
1270 * this function is only called if nobody else can access
1271 * this mm -- see comment above for justification.
1272 */
1273 old_exe_file = rcu_dereference_raw(mm->exe_file);
1274
1275 if (new_exe_file) {
1276 /*
1277 * We expect the caller (i.e., sys_execve) to already denied
1278 * write access, so this is unlikely to fail.
1279 */
1280 if (unlikely(deny_write_access(new_exe_file)))
1281 return -EACCES;
1282 get_file(new_exe_file);
1283 }
1284 rcu_assign_pointer(mm->exe_file, new_exe_file);
1285 if (old_exe_file) {
1286 allow_write_access(old_exe_file);
1287 fput(old_exe_file);
1288 }
1289 return 0;
1290}
1291
1292/**
1293 * replace_mm_exe_file - replace a reference to the mm's executable file
1294 *
1295 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1296 * dealing with concurrent invocation and without grabbing the mmap lock in
1297 * write mode.
1298 *
1299 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1300 */
1301int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1302{
1303 struct vm_area_struct *vma;
1304 struct file *old_exe_file;
1305 int ret = 0;
1306
1307 /* Forbid mm->exe_file change if old file still mapped. */
1308 old_exe_file = get_mm_exe_file(mm);
1309 if (old_exe_file) {
1310 VMA_ITERATOR(vmi, mm, 0);
1311 mmap_read_lock(mm);
1312 for_each_vma(vmi, vma) {
1313 if (!vma->vm_file)
1314 continue;
1315 if (path_equal(&vma->vm_file->f_path,
1316 &old_exe_file->f_path)) {
1317 ret = -EBUSY;
1318 break;
1319 }
1320 }
1321 mmap_read_unlock(mm);
1322 fput(old_exe_file);
1323 if (ret)
1324 return ret;
1325 }
1326
1327 /* set the new file, lockless */
1328 ret = deny_write_access(new_exe_file);
1329 if (ret)
1330 return -EACCES;
1331 get_file(new_exe_file);
1332
1333 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1334 if (old_exe_file) {
1335 /*
1336 * Don't race with dup_mmap() getting the file and disallowing
1337 * write access while someone might open the file writable.
1338 */
1339 mmap_read_lock(mm);
1340 allow_write_access(old_exe_file);
1341 fput(old_exe_file);
1342 mmap_read_unlock(mm);
1343 }
1344 return 0;
1345}
1346
1347/**
1348 * get_mm_exe_file - acquire a reference to the mm's executable file
1349 *
1350 * Returns %NULL if mm has no associated executable file.
1351 * User must release file via fput().
1352 */
1353struct file *get_mm_exe_file(struct mm_struct *mm)
1354{
1355 struct file *exe_file;
1356
1357 rcu_read_lock();
1358 exe_file = rcu_dereference(mm->exe_file);
1359 if (exe_file && !get_file_rcu(exe_file))
1360 exe_file = NULL;
1361 rcu_read_unlock();
1362 return exe_file;
1363}
1364
1365/**
1366 * get_task_exe_file - acquire a reference to the task's executable file
1367 *
1368 * Returns %NULL if task's mm (if any) has no associated executable file or
1369 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1370 * User must release file via fput().
1371 */
1372struct file *get_task_exe_file(struct task_struct *task)
1373{
1374 struct file *exe_file = NULL;
1375 struct mm_struct *mm;
1376
1377 task_lock(task);
1378 mm = task->mm;
1379 if (mm) {
1380 if (!(task->flags & PF_KTHREAD))
1381 exe_file = get_mm_exe_file(mm);
1382 }
1383 task_unlock(task);
1384 return exe_file;
1385}
1386
1387/**
1388 * get_task_mm - acquire a reference to the task's mm
1389 *
1390 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1391 * this kernel workthread has transiently adopted a user mm with use_mm,
1392 * to do its AIO) is not set and if so returns a reference to it, after
1393 * bumping up the use count. User must release the mm via mmput()
1394 * after use. Typically used by /proc and ptrace.
1395 */
1396struct mm_struct *get_task_mm(struct task_struct *task)
1397{
1398 struct mm_struct *mm;
1399
1400 task_lock(task);
1401 mm = task->mm;
1402 if (mm) {
1403 if (task->flags & PF_KTHREAD)
1404 mm = NULL;
1405 else
1406 mmget(mm);
1407 }
1408 task_unlock(task);
1409 return mm;
1410}
1411EXPORT_SYMBOL_GPL(get_task_mm);
1412
1413struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1414{
1415 struct mm_struct *mm;
1416 int err;
1417
1418 err = down_read_killable(&task->signal->exec_update_lock);
1419 if (err)
1420 return ERR_PTR(err);
1421
1422 mm = get_task_mm(task);
1423 if (mm && mm != current->mm &&
1424 !ptrace_may_access(task, mode)) {
1425 mmput(mm);
1426 mm = ERR_PTR(-EACCES);
1427 }
1428 up_read(&task->signal->exec_update_lock);
1429
1430 return mm;
1431}
1432
1433static void complete_vfork_done(struct task_struct *tsk)
1434{
1435 struct completion *vfork;
1436
1437 task_lock(tsk);
1438 vfork = tsk->vfork_done;
1439 if (likely(vfork)) {
1440 tsk->vfork_done = NULL;
1441 complete(vfork);
1442 }
1443 task_unlock(tsk);
1444}
1445
1446static int wait_for_vfork_done(struct task_struct *child,
1447 struct completion *vfork)
1448{
1449 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1450 int killed;
1451
1452 cgroup_enter_frozen();
1453 killed = wait_for_completion_state(vfork, state);
1454 cgroup_leave_frozen(false);
1455
1456 if (killed) {
1457 task_lock(child);
1458 child->vfork_done = NULL;
1459 task_unlock(child);
1460 }
1461
1462 put_task_struct(child);
1463 return killed;
1464}
1465
1466/* Please note the differences between mmput and mm_release.
1467 * mmput is called whenever we stop holding onto a mm_struct,
1468 * error success whatever.
1469 *
1470 * mm_release is called after a mm_struct has been removed
1471 * from the current process.
1472 *
1473 * This difference is important for error handling, when we
1474 * only half set up a mm_struct for a new process and need to restore
1475 * the old one. Because we mmput the new mm_struct before
1476 * restoring the old one. . .
1477 * Eric Biederman 10 January 1998
1478 */
1479static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1480{
1481 uprobe_free_utask(tsk);
1482
1483 /* Get rid of any cached register state */
1484 deactivate_mm(tsk, mm);
1485
1486 /*
1487 * Signal userspace if we're not exiting with a core dump
1488 * because we want to leave the value intact for debugging
1489 * purposes.
1490 */
1491 if (tsk->clear_child_tid) {
1492 if (atomic_read(&mm->mm_users) > 1) {
1493 /*
1494 * We don't check the error code - if userspace has
1495 * not set up a proper pointer then tough luck.
1496 */
1497 put_user(0, tsk->clear_child_tid);
1498 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1499 1, NULL, NULL, 0, 0);
1500 }
1501 tsk->clear_child_tid = NULL;
1502 }
1503
1504 /*
1505 * All done, finally we can wake up parent and return this mm to him.
1506 * Also kthread_stop() uses this completion for synchronization.
1507 */
1508 if (tsk->vfork_done)
1509 complete_vfork_done(tsk);
1510}
1511
1512void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1513{
1514 futex_exit_release(tsk);
1515 mm_release(tsk, mm);
1516}
1517
1518void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1519{
1520 futex_exec_release(tsk);
1521 mm_release(tsk, mm);
1522}
1523
1524/**
1525 * dup_mm() - duplicates an existing mm structure
1526 * @tsk: the task_struct with which the new mm will be associated.
1527 * @oldmm: the mm to duplicate.
1528 *
1529 * Allocates a new mm structure and duplicates the provided @oldmm structure
1530 * content into it.
1531 *
1532 * Return: the duplicated mm or NULL on failure.
1533 */
1534static struct mm_struct *dup_mm(struct task_struct *tsk,
1535 struct mm_struct *oldmm)
1536{
1537 struct mm_struct *mm;
1538 int err;
1539
1540 mm = allocate_mm();
1541 if (!mm)
1542 goto fail_nomem;
1543
1544 memcpy(mm, oldmm, sizeof(*mm));
1545
1546 if (!mm_init(mm, tsk, mm->user_ns))
1547 goto fail_nomem;
1548
1549 err = dup_mmap(mm, oldmm);
1550 if (err)
1551 goto free_pt;
1552
1553 mm->hiwater_rss = get_mm_rss(mm);
1554 mm->hiwater_vm = mm->total_vm;
1555
1556 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1557 goto free_pt;
1558
1559 return mm;
1560
1561free_pt:
1562 /* don't put binfmt in mmput, we haven't got module yet */
1563 mm->binfmt = NULL;
1564 mm_init_owner(mm, NULL);
1565 mmput(mm);
1566
1567fail_nomem:
1568 return NULL;
1569}
1570
1571static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1572{
1573 struct mm_struct *mm, *oldmm;
1574
1575 tsk->min_flt = tsk->maj_flt = 0;
1576 tsk->nvcsw = tsk->nivcsw = 0;
1577#ifdef CONFIG_DETECT_HUNG_TASK
1578 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1579 tsk->last_switch_time = 0;
1580#endif
1581
1582 tsk->mm = NULL;
1583 tsk->active_mm = NULL;
1584
1585 /*
1586 * Are we cloning a kernel thread?
1587 *
1588 * We need to steal a active VM for that..
1589 */
1590 oldmm = current->mm;
1591 if (!oldmm)
1592 return 0;
1593
1594 if (clone_flags & CLONE_VM) {
1595 mmget(oldmm);
1596 mm = oldmm;
1597 } else {
1598 mm = dup_mm(tsk, current->mm);
1599 if (!mm)
1600 return -ENOMEM;
1601 }
1602
1603 tsk->mm = mm;
1604 tsk->active_mm = mm;
1605 sched_mm_cid_fork(tsk);
1606 return 0;
1607}
1608
1609static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1610{
1611 struct fs_struct *fs = current->fs;
1612 if (clone_flags & CLONE_FS) {
1613 /* tsk->fs is already what we want */
1614 spin_lock(&fs->lock);
1615 if (fs->in_exec) {
1616 spin_unlock(&fs->lock);
1617 return -EAGAIN;
1618 }
1619 fs->users++;
1620 spin_unlock(&fs->lock);
1621 return 0;
1622 }
1623 tsk->fs = copy_fs_struct(fs);
1624 if (!tsk->fs)
1625 return -ENOMEM;
1626 return 0;
1627}
1628
1629static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1630{
1631 struct files_struct *oldf, *newf;
1632 int error = 0;
1633
1634 /*
1635 * A background process may not have any files ...
1636 */
1637 oldf = current->files;
1638 if (!oldf)
1639 goto out;
1640
1641 if (clone_flags & CLONE_FILES) {
1642 atomic_inc(&oldf->count);
1643 goto out;
1644 }
1645
1646 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1647 if (!newf)
1648 goto out;
1649
1650 tsk->files = newf;
1651 error = 0;
1652out:
1653 return error;
1654}
1655
1656static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1657{
1658 struct sighand_struct *sig;
1659
1660 if (clone_flags & CLONE_SIGHAND) {
1661 refcount_inc(&current->sighand->count);
1662 return 0;
1663 }
1664 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1665 RCU_INIT_POINTER(tsk->sighand, sig);
1666 if (!sig)
1667 return -ENOMEM;
1668
1669 refcount_set(&sig->count, 1);
1670 spin_lock_irq(&current->sighand->siglock);
1671 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1672 spin_unlock_irq(&current->sighand->siglock);
1673
1674 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1675 if (clone_flags & CLONE_CLEAR_SIGHAND)
1676 flush_signal_handlers(tsk, 0);
1677
1678 return 0;
1679}
1680
1681void __cleanup_sighand(struct sighand_struct *sighand)
1682{
1683 if (refcount_dec_and_test(&sighand->count)) {
1684 signalfd_cleanup(sighand);
1685 /*
1686 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1687 * without an RCU grace period, see __lock_task_sighand().
1688 */
1689 kmem_cache_free(sighand_cachep, sighand);
1690 }
1691}
1692
1693/*
1694 * Initialize POSIX timer handling for a thread group.
1695 */
1696static void posix_cpu_timers_init_group(struct signal_struct *sig)
1697{
1698 struct posix_cputimers *pct = &sig->posix_cputimers;
1699 unsigned long cpu_limit;
1700
1701 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1702 posix_cputimers_group_init(pct, cpu_limit);
1703}
1704
1705static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1706{
1707 struct signal_struct *sig;
1708
1709 if (clone_flags & CLONE_THREAD)
1710 return 0;
1711
1712 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1713 tsk->signal = sig;
1714 if (!sig)
1715 return -ENOMEM;
1716
1717 sig->nr_threads = 1;
1718 sig->quick_threads = 1;
1719 atomic_set(&sig->live, 1);
1720 refcount_set(&sig->sigcnt, 1);
1721
1722 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1723 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1724 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1725
1726 init_waitqueue_head(&sig->wait_chldexit);
1727 sig->curr_target = tsk;
1728 init_sigpending(&sig->shared_pending);
1729 INIT_HLIST_HEAD(&sig->multiprocess);
1730 seqlock_init(&sig->stats_lock);
1731 prev_cputime_init(&sig->prev_cputime);
1732
1733#ifdef CONFIG_POSIX_TIMERS
1734 INIT_LIST_HEAD(&sig->posix_timers);
1735 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1736 sig->real_timer.function = it_real_fn;
1737#endif
1738
1739 task_lock(current->group_leader);
1740 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1741 task_unlock(current->group_leader);
1742
1743 posix_cpu_timers_init_group(sig);
1744
1745 tty_audit_fork(sig);
1746 sched_autogroup_fork(sig);
1747
1748 sig->oom_score_adj = current->signal->oom_score_adj;
1749 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1750
1751 mutex_init(&sig->cred_guard_mutex);
1752 init_rwsem(&sig->exec_update_lock);
1753
1754 return 0;
1755}
1756
1757static void copy_seccomp(struct task_struct *p)
1758{
1759#ifdef CONFIG_SECCOMP
1760 /*
1761 * Must be called with sighand->lock held, which is common to
1762 * all threads in the group. Holding cred_guard_mutex is not
1763 * needed because this new task is not yet running and cannot
1764 * be racing exec.
1765 */
1766 assert_spin_locked(&current->sighand->siglock);
1767
1768 /* Ref-count the new filter user, and assign it. */
1769 get_seccomp_filter(current);
1770 p->seccomp = current->seccomp;
1771
1772 /*
1773 * Explicitly enable no_new_privs here in case it got set
1774 * between the task_struct being duplicated and holding the
1775 * sighand lock. The seccomp state and nnp must be in sync.
1776 */
1777 if (task_no_new_privs(current))
1778 task_set_no_new_privs(p);
1779
1780 /*
1781 * If the parent gained a seccomp mode after copying thread
1782 * flags and between before we held the sighand lock, we have
1783 * to manually enable the seccomp thread flag here.
1784 */
1785 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1786 set_task_syscall_work(p, SECCOMP);
1787#endif
1788}
1789
1790SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1791{
1792 current->clear_child_tid = tidptr;
1793
1794 return task_pid_vnr(current);
1795}
1796
1797static void rt_mutex_init_task(struct task_struct *p)
1798{
1799 raw_spin_lock_init(&p->pi_lock);
1800#ifdef CONFIG_RT_MUTEXES
1801 p->pi_waiters = RB_ROOT_CACHED;
1802 p->pi_top_task = NULL;
1803 p->pi_blocked_on = NULL;
1804#endif
1805}
1806
1807static inline void init_task_pid_links(struct task_struct *task)
1808{
1809 enum pid_type type;
1810
1811 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1812 INIT_HLIST_NODE(&task->pid_links[type]);
1813}
1814
1815static inline void
1816init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1817{
1818 if (type == PIDTYPE_PID)
1819 task->thread_pid = pid;
1820 else
1821 task->signal->pids[type] = pid;
1822}
1823
1824static inline void rcu_copy_process(struct task_struct *p)
1825{
1826#ifdef CONFIG_PREEMPT_RCU
1827 p->rcu_read_lock_nesting = 0;
1828 p->rcu_read_unlock_special.s = 0;
1829 p->rcu_blocked_node = NULL;
1830 INIT_LIST_HEAD(&p->rcu_node_entry);
1831#endif /* #ifdef CONFIG_PREEMPT_RCU */
1832#ifdef CONFIG_TASKS_RCU
1833 p->rcu_tasks_holdout = false;
1834 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1835 p->rcu_tasks_idle_cpu = -1;
1836#endif /* #ifdef CONFIG_TASKS_RCU */
1837#ifdef CONFIG_TASKS_TRACE_RCU
1838 p->trc_reader_nesting = 0;
1839 p->trc_reader_special.s = 0;
1840 INIT_LIST_HEAD(&p->trc_holdout_list);
1841 INIT_LIST_HEAD(&p->trc_blkd_node);
1842#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1843}
1844
1845struct pid *pidfd_pid(const struct file *file)
1846{
1847 if (file->f_op == &pidfd_fops)
1848 return file->private_data;
1849
1850 return ERR_PTR(-EBADF);
1851}
1852
1853static int pidfd_release(struct inode *inode, struct file *file)
1854{
1855 struct pid *pid = file->private_data;
1856
1857 file->private_data = NULL;
1858 put_pid(pid);
1859 return 0;
1860}
1861
1862#ifdef CONFIG_PROC_FS
1863/**
1864 * pidfd_show_fdinfo - print information about a pidfd
1865 * @m: proc fdinfo file
1866 * @f: file referencing a pidfd
1867 *
1868 * Pid:
1869 * This function will print the pid that a given pidfd refers to in the
1870 * pid namespace of the procfs instance.
1871 * If the pid namespace of the process is not a descendant of the pid
1872 * namespace of the procfs instance 0 will be shown as its pid. This is
1873 * similar to calling getppid() on a process whose parent is outside of
1874 * its pid namespace.
1875 *
1876 * NSpid:
1877 * If pid namespaces are supported then this function will also print
1878 * the pid of a given pidfd refers to for all descendant pid namespaces
1879 * starting from the current pid namespace of the instance, i.e. the
1880 * Pid field and the first entry in the NSpid field will be identical.
1881 * If the pid namespace of the process is not a descendant of the pid
1882 * namespace of the procfs instance 0 will be shown as its first NSpid
1883 * entry and no others will be shown.
1884 * Note that this differs from the Pid and NSpid fields in
1885 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1886 * the pid namespace of the procfs instance. The difference becomes
1887 * obvious when sending around a pidfd between pid namespaces from a
1888 * different branch of the tree, i.e. where no ancestral relation is
1889 * present between the pid namespaces:
1890 * - create two new pid namespaces ns1 and ns2 in the initial pid
1891 * namespace (also take care to create new mount namespaces in the
1892 * new pid namespace and mount procfs)
1893 * - create a process with a pidfd in ns1
1894 * - send pidfd from ns1 to ns2
1895 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1896 * have exactly one entry, which is 0
1897 */
1898static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1899{
1900 struct pid *pid = f->private_data;
1901 struct pid_namespace *ns;
1902 pid_t nr = -1;
1903
1904 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1905 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1906 nr = pid_nr_ns(pid, ns);
1907 }
1908
1909 seq_put_decimal_ll(m, "Pid:\t", nr);
1910
1911#ifdef CONFIG_PID_NS
1912 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1913 if (nr > 0) {
1914 int i;
1915
1916 /* If nr is non-zero it means that 'pid' is valid and that
1917 * ns, i.e. the pid namespace associated with the procfs
1918 * instance, is in the pid namespace hierarchy of pid.
1919 * Start at one below the already printed level.
1920 */
1921 for (i = ns->level + 1; i <= pid->level; i++)
1922 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1923 }
1924#endif
1925 seq_putc(m, '\n');
1926}
1927#endif
1928
1929/*
1930 * Poll support for process exit notification.
1931 */
1932static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1933{
1934 struct pid *pid = file->private_data;
1935 __poll_t poll_flags = 0;
1936
1937 poll_wait(file, &pid->wait_pidfd, pts);
1938
1939 /*
1940 * Inform pollers only when the whole thread group exits.
1941 * If the thread group leader exits before all other threads in the
1942 * group, then poll(2) should block, similar to the wait(2) family.
1943 */
1944 if (thread_group_exited(pid))
1945 poll_flags = EPOLLIN | EPOLLRDNORM;
1946
1947 return poll_flags;
1948}
1949
1950const struct file_operations pidfd_fops = {
1951 .release = pidfd_release,
1952 .poll = pidfd_poll,
1953#ifdef CONFIG_PROC_FS
1954 .show_fdinfo = pidfd_show_fdinfo,
1955#endif
1956};
1957
1958static void __delayed_free_task(struct rcu_head *rhp)
1959{
1960 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1961
1962 free_task(tsk);
1963}
1964
1965static __always_inline void delayed_free_task(struct task_struct *tsk)
1966{
1967 if (IS_ENABLED(CONFIG_MEMCG))
1968 call_rcu(&tsk->rcu, __delayed_free_task);
1969 else
1970 free_task(tsk);
1971}
1972
1973static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1974{
1975 /* Skip if kernel thread */
1976 if (!tsk->mm)
1977 return;
1978
1979 /* Skip if spawning a thread or using vfork */
1980 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1981 return;
1982
1983 /* We need to synchronize with __set_oom_adj */
1984 mutex_lock(&oom_adj_mutex);
1985 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1986 /* Update the values in case they were changed after copy_signal */
1987 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1988 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1989 mutex_unlock(&oom_adj_mutex);
1990}
1991
1992#ifdef CONFIG_RV
1993static void rv_task_fork(struct task_struct *p)
1994{
1995 int i;
1996
1997 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1998 p->rv[i].da_mon.monitoring = false;
1999}
2000#else
2001#define rv_task_fork(p) do {} while (0)
2002#endif
2003
2004/*
2005 * This creates a new process as a copy of the old one,
2006 * but does not actually start it yet.
2007 *
2008 * It copies the registers, and all the appropriate
2009 * parts of the process environment (as per the clone
2010 * flags). The actual kick-off is left to the caller.
2011 */
2012static __latent_entropy struct task_struct *copy_process(
2013 struct pid *pid,
2014 int trace,
2015 int node,
2016 struct kernel_clone_args *args)
2017{
2018 int pidfd = -1, retval;
2019 struct task_struct *p;
2020 struct multiprocess_signals delayed;
2021 struct file *pidfile = NULL;
2022 const u64 clone_flags = args->flags;
2023 struct nsproxy *nsp = current->nsproxy;
2024
2025 /*
2026 * Don't allow sharing the root directory with processes in a different
2027 * namespace
2028 */
2029 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2030 return ERR_PTR(-EINVAL);
2031
2032 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2033 return ERR_PTR(-EINVAL);
2034
2035 /*
2036 * Thread groups must share signals as well, and detached threads
2037 * can only be started up within the thread group.
2038 */
2039 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2040 return ERR_PTR(-EINVAL);
2041
2042 /*
2043 * Shared signal handlers imply shared VM. By way of the above,
2044 * thread groups also imply shared VM. Blocking this case allows
2045 * for various simplifications in other code.
2046 */
2047 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2048 return ERR_PTR(-EINVAL);
2049
2050 /*
2051 * Siblings of global init remain as zombies on exit since they are
2052 * not reaped by their parent (swapper). To solve this and to avoid
2053 * multi-rooted process trees, prevent global and container-inits
2054 * from creating siblings.
2055 */
2056 if ((clone_flags & CLONE_PARENT) &&
2057 current->signal->flags & SIGNAL_UNKILLABLE)
2058 return ERR_PTR(-EINVAL);
2059
2060 /*
2061 * If the new process will be in a different pid or user namespace
2062 * do not allow it to share a thread group with the forking task.
2063 */
2064 if (clone_flags & CLONE_THREAD) {
2065 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2066 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2067 return ERR_PTR(-EINVAL);
2068 }
2069
2070 if (clone_flags & CLONE_PIDFD) {
2071 /*
2072 * - CLONE_DETACHED is blocked so that we can potentially
2073 * reuse it later for CLONE_PIDFD.
2074 * - CLONE_THREAD is blocked until someone really needs it.
2075 */
2076 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2077 return ERR_PTR(-EINVAL);
2078 }
2079
2080 /*
2081 * Force any signals received before this point to be delivered
2082 * before the fork happens. Collect up signals sent to multiple
2083 * processes that happen during the fork and delay them so that
2084 * they appear to happen after the fork.
2085 */
2086 sigemptyset(&delayed.signal);
2087 INIT_HLIST_NODE(&delayed.node);
2088
2089 spin_lock_irq(&current->sighand->siglock);
2090 if (!(clone_flags & CLONE_THREAD))
2091 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2092 recalc_sigpending();
2093 spin_unlock_irq(&current->sighand->siglock);
2094 retval = -ERESTARTNOINTR;
2095 if (task_sigpending(current))
2096 goto fork_out;
2097
2098 retval = -ENOMEM;
2099 p = dup_task_struct(current, node);
2100 if (!p)
2101 goto fork_out;
2102 p->flags &= ~PF_KTHREAD;
2103 if (args->kthread)
2104 p->flags |= PF_KTHREAD;
2105 if (args->io_thread) {
2106 /*
2107 * Mark us an IO worker, and block any signal that isn't
2108 * fatal or STOP
2109 */
2110 p->flags |= PF_IO_WORKER;
2111 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2112 }
2113
2114 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2115 /*
2116 * Clear TID on mm_release()?
2117 */
2118 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2119
2120 ftrace_graph_init_task(p);
2121
2122 rt_mutex_init_task(p);
2123
2124 lockdep_assert_irqs_enabled();
2125#ifdef CONFIG_PROVE_LOCKING
2126 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2127#endif
2128 retval = copy_creds(p, clone_flags);
2129 if (retval < 0)
2130 goto bad_fork_free;
2131
2132 retval = -EAGAIN;
2133 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2134 if (p->real_cred->user != INIT_USER &&
2135 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2136 goto bad_fork_cleanup_count;
2137 }
2138 current->flags &= ~PF_NPROC_EXCEEDED;
2139
2140 /*
2141 * If multiple threads are within copy_process(), then this check
2142 * triggers too late. This doesn't hurt, the check is only there
2143 * to stop root fork bombs.
2144 */
2145 retval = -EAGAIN;
2146 if (data_race(nr_threads >= max_threads))
2147 goto bad_fork_cleanup_count;
2148
2149 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2150 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2151 p->flags |= PF_FORKNOEXEC;
2152 INIT_LIST_HEAD(&p->children);
2153 INIT_LIST_HEAD(&p->sibling);
2154 rcu_copy_process(p);
2155 p->vfork_done = NULL;
2156 spin_lock_init(&p->alloc_lock);
2157
2158 init_sigpending(&p->pending);
2159
2160 p->utime = p->stime = p->gtime = 0;
2161#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2162 p->utimescaled = p->stimescaled = 0;
2163#endif
2164 prev_cputime_init(&p->prev_cputime);
2165
2166#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2167 seqcount_init(&p->vtime.seqcount);
2168 p->vtime.starttime = 0;
2169 p->vtime.state = VTIME_INACTIVE;
2170#endif
2171
2172#ifdef CONFIG_IO_URING
2173 p->io_uring = NULL;
2174#endif
2175
2176#if defined(SPLIT_RSS_COUNTING)
2177 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2178#endif
2179
2180 p->default_timer_slack_ns = current->timer_slack_ns;
2181
2182#ifdef CONFIG_PSI
2183 p->psi_flags = 0;
2184#endif
2185
2186 task_io_accounting_init(&p->ioac);
2187 acct_clear_integrals(p);
2188
2189 posix_cputimers_init(&p->posix_cputimers);
2190
2191 p->io_context = NULL;
2192 audit_set_context(p, NULL);
2193 cgroup_fork(p);
2194 if (args->kthread) {
2195 if (!set_kthread_struct(p))
2196 goto bad_fork_cleanup_delayacct;
2197 }
2198#ifdef CONFIG_NUMA
2199 p->mempolicy = mpol_dup(p->mempolicy);
2200 if (IS_ERR(p->mempolicy)) {
2201 retval = PTR_ERR(p->mempolicy);
2202 p->mempolicy = NULL;
2203 goto bad_fork_cleanup_delayacct;
2204 }
2205#endif
2206#ifdef CONFIG_CPUSETS
2207 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2208 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2209 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2210#endif
2211#ifdef CONFIG_TRACE_IRQFLAGS
2212 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2213 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2214 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2215 p->softirqs_enabled = 1;
2216 p->softirq_context = 0;
2217#endif
2218
2219 p->pagefault_disabled = 0;
2220
2221#ifdef CONFIG_LOCKDEP
2222 lockdep_init_task(p);
2223#endif
2224
2225#ifdef CONFIG_DEBUG_MUTEXES
2226 p->blocked_on = NULL; /* not blocked yet */
2227#endif
2228#ifdef CONFIG_BCACHE
2229 p->sequential_io = 0;
2230 p->sequential_io_avg = 0;
2231#endif
2232#ifdef CONFIG_BPF_SYSCALL
2233 RCU_INIT_POINTER(p->bpf_storage, NULL);
2234 p->bpf_ctx = NULL;
2235#endif
2236
2237 /* Perform scheduler related setup. Assign this task to a CPU. */
2238 retval = sched_fork(clone_flags, p);
2239 if (retval)
2240 goto bad_fork_cleanup_policy;
2241
2242 retval = perf_event_init_task(p, clone_flags);
2243 if (retval)
2244 goto bad_fork_cleanup_policy;
2245 retval = audit_alloc(p);
2246 if (retval)
2247 goto bad_fork_cleanup_perf;
2248 /* copy all the process information */
2249 shm_init_task(p);
2250 retval = security_task_alloc(p, clone_flags);
2251 if (retval)
2252 goto bad_fork_cleanup_audit;
2253 retval = copy_semundo(clone_flags, p);
2254 if (retval)
2255 goto bad_fork_cleanup_security;
2256 retval = copy_files(clone_flags, p);
2257 if (retval)
2258 goto bad_fork_cleanup_semundo;
2259 retval = copy_fs(clone_flags, p);
2260 if (retval)
2261 goto bad_fork_cleanup_files;
2262 retval = copy_sighand(clone_flags, p);
2263 if (retval)
2264 goto bad_fork_cleanup_fs;
2265 retval = copy_signal(clone_flags, p);
2266 if (retval)
2267 goto bad_fork_cleanup_sighand;
2268 retval = copy_mm(clone_flags, p);
2269 if (retval)
2270 goto bad_fork_cleanup_signal;
2271 retval = copy_namespaces(clone_flags, p);
2272 if (retval)
2273 goto bad_fork_cleanup_mm;
2274 retval = copy_io(clone_flags, p);
2275 if (retval)
2276 goto bad_fork_cleanup_namespaces;
2277 retval = copy_thread(p, args);
2278 if (retval)
2279 goto bad_fork_cleanup_io;
2280
2281 stackleak_task_init(p);
2282
2283 if (pid != &init_struct_pid) {
2284 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2285 args->set_tid_size);
2286 if (IS_ERR(pid)) {
2287 retval = PTR_ERR(pid);
2288 goto bad_fork_cleanup_thread;
2289 }
2290 }
2291
2292 /*
2293 * This has to happen after we've potentially unshared the file
2294 * descriptor table (so that the pidfd doesn't leak into the child
2295 * if the fd table isn't shared).
2296 */
2297 if (clone_flags & CLONE_PIDFD) {
2298 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2299 if (retval < 0)
2300 goto bad_fork_free_pid;
2301
2302 pidfd = retval;
2303
2304 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2305 O_RDWR | O_CLOEXEC);
2306 if (IS_ERR(pidfile)) {
2307 put_unused_fd(pidfd);
2308 retval = PTR_ERR(pidfile);
2309 goto bad_fork_free_pid;
2310 }
2311 get_pid(pid); /* held by pidfile now */
2312
2313 retval = put_user(pidfd, args->pidfd);
2314 if (retval)
2315 goto bad_fork_put_pidfd;
2316 }
2317
2318#ifdef CONFIG_BLOCK
2319 p->plug = NULL;
2320#endif
2321 futex_init_task(p);
2322
2323 /*
2324 * sigaltstack should be cleared when sharing the same VM
2325 */
2326 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2327 sas_ss_reset(p);
2328
2329 /*
2330 * Syscall tracing and stepping should be turned off in the
2331 * child regardless of CLONE_PTRACE.
2332 */
2333 user_disable_single_step(p);
2334 clear_task_syscall_work(p, SYSCALL_TRACE);
2335#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2336 clear_task_syscall_work(p, SYSCALL_EMU);
2337#endif
2338 clear_tsk_latency_tracing(p);
2339
2340 /* ok, now we should be set up.. */
2341 p->pid = pid_nr(pid);
2342 if (clone_flags & CLONE_THREAD) {
2343 p->group_leader = current->group_leader;
2344 p->tgid = current->tgid;
2345 } else {
2346 p->group_leader = p;
2347 p->tgid = p->pid;
2348 }
2349
2350 p->nr_dirtied = 0;
2351 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2352 p->dirty_paused_when = 0;
2353
2354 p->pdeath_signal = 0;
2355 INIT_LIST_HEAD(&p->thread_group);
2356 p->task_works = NULL;
2357 clear_posix_cputimers_work(p);
2358
2359#ifdef CONFIG_KRETPROBES
2360 p->kretprobe_instances.first = NULL;
2361#endif
2362#ifdef CONFIG_RETHOOK
2363 p->rethooks.first = NULL;
2364#endif
2365
2366 /*
2367 * Ensure that the cgroup subsystem policies allow the new process to be
2368 * forked. It should be noted that the new process's css_set can be changed
2369 * between here and cgroup_post_fork() if an organisation operation is in
2370 * progress.
2371 */
2372 retval = cgroup_can_fork(p, args);
2373 if (retval)
2374 goto bad_fork_put_pidfd;
2375
2376 /*
2377 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2378 * the new task on the correct runqueue. All this *before* the task
2379 * becomes visible.
2380 *
2381 * This isn't part of ->can_fork() because while the re-cloning is
2382 * cgroup specific, it unconditionally needs to place the task on a
2383 * runqueue.
2384 */
2385 sched_cgroup_fork(p, args);
2386
2387 /*
2388 * From this point on we must avoid any synchronous user-space
2389 * communication until we take the tasklist-lock. In particular, we do
2390 * not want user-space to be able to predict the process start-time by
2391 * stalling fork(2) after we recorded the start_time but before it is
2392 * visible to the system.
2393 */
2394
2395 p->start_time = ktime_get_ns();
2396 p->start_boottime = ktime_get_boottime_ns();
2397
2398 /*
2399 * Make it visible to the rest of the system, but dont wake it up yet.
2400 * Need tasklist lock for parent etc handling!
2401 */
2402 write_lock_irq(&tasklist_lock);
2403
2404 /* CLONE_PARENT re-uses the old parent */
2405 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2406 p->real_parent = current->real_parent;
2407 p->parent_exec_id = current->parent_exec_id;
2408 if (clone_flags & CLONE_THREAD)
2409 p->exit_signal = -1;
2410 else
2411 p->exit_signal = current->group_leader->exit_signal;
2412 } else {
2413 p->real_parent = current;
2414 p->parent_exec_id = current->self_exec_id;
2415 p->exit_signal = args->exit_signal;
2416 }
2417
2418 klp_copy_process(p);
2419
2420 sched_core_fork(p);
2421
2422 spin_lock(&current->sighand->siglock);
2423
2424 rv_task_fork(p);
2425
2426 rseq_fork(p, clone_flags);
2427
2428 /* Don't start children in a dying pid namespace */
2429 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2430 retval = -ENOMEM;
2431 goto bad_fork_cancel_cgroup;
2432 }
2433
2434 /* Let kill terminate clone/fork in the middle */
2435 if (fatal_signal_pending(current)) {
2436 retval = -EINTR;
2437 goto bad_fork_cancel_cgroup;
2438 }
2439
2440 /* No more failure paths after this point. */
2441
2442 /*
2443 * Copy seccomp details explicitly here, in case they were changed
2444 * before holding sighand lock.
2445 */
2446 copy_seccomp(p);
2447
2448 init_task_pid_links(p);
2449 if (likely(p->pid)) {
2450 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2451
2452 init_task_pid(p, PIDTYPE_PID, pid);
2453 if (thread_group_leader(p)) {
2454 init_task_pid(p, PIDTYPE_TGID, pid);
2455 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2456 init_task_pid(p, PIDTYPE_SID, task_session(current));
2457
2458 if (is_child_reaper(pid)) {
2459 ns_of_pid(pid)->child_reaper = p;
2460 p->signal->flags |= SIGNAL_UNKILLABLE;
2461 }
2462 p->signal->shared_pending.signal = delayed.signal;
2463 p->signal->tty = tty_kref_get(current->signal->tty);
2464 /*
2465 * Inherit has_child_subreaper flag under the same
2466 * tasklist_lock with adding child to the process tree
2467 * for propagate_has_child_subreaper optimization.
2468 */
2469 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2470 p->real_parent->signal->is_child_subreaper;
2471 list_add_tail(&p->sibling, &p->real_parent->children);
2472 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2473 attach_pid(p, PIDTYPE_TGID);
2474 attach_pid(p, PIDTYPE_PGID);
2475 attach_pid(p, PIDTYPE_SID);
2476 __this_cpu_inc(process_counts);
2477 } else {
2478 current->signal->nr_threads++;
2479 current->signal->quick_threads++;
2480 atomic_inc(&current->signal->live);
2481 refcount_inc(&current->signal->sigcnt);
2482 task_join_group_stop(p);
2483 list_add_tail_rcu(&p->thread_group,
2484 &p->group_leader->thread_group);
2485 list_add_tail_rcu(&p->thread_node,
2486 &p->signal->thread_head);
2487 }
2488 attach_pid(p, PIDTYPE_PID);
2489 nr_threads++;
2490 }
2491 total_forks++;
2492 hlist_del_init(&delayed.node);
2493 spin_unlock(&current->sighand->siglock);
2494 syscall_tracepoint_update(p);
2495 write_unlock_irq(&tasklist_lock);
2496
2497 if (pidfile)
2498 fd_install(pidfd, pidfile);
2499
2500 proc_fork_connector(p);
2501 sched_post_fork(p);
2502 cgroup_post_fork(p, args);
2503 perf_event_fork(p);
2504
2505 trace_task_newtask(p, clone_flags);
2506 uprobe_copy_process(p, clone_flags);
2507
2508 copy_oom_score_adj(clone_flags, p);
2509
2510 return p;
2511
2512bad_fork_cancel_cgroup:
2513 sched_core_free(p);
2514 spin_unlock(&current->sighand->siglock);
2515 write_unlock_irq(&tasklist_lock);
2516 cgroup_cancel_fork(p, args);
2517bad_fork_put_pidfd:
2518 if (clone_flags & CLONE_PIDFD) {
2519 fput(pidfile);
2520 put_unused_fd(pidfd);
2521 }
2522bad_fork_free_pid:
2523 if (pid != &init_struct_pid)
2524 free_pid(pid);
2525bad_fork_cleanup_thread:
2526 exit_thread(p);
2527bad_fork_cleanup_io:
2528 if (p->io_context)
2529 exit_io_context(p);
2530bad_fork_cleanup_namespaces:
2531 exit_task_namespaces(p);
2532bad_fork_cleanup_mm:
2533 if (p->mm) {
2534 mm_clear_owner(p->mm, p);
2535 mmput(p->mm);
2536 }
2537bad_fork_cleanup_signal:
2538 if (!(clone_flags & CLONE_THREAD))
2539 free_signal_struct(p->signal);
2540bad_fork_cleanup_sighand:
2541 __cleanup_sighand(p->sighand);
2542bad_fork_cleanup_fs:
2543 exit_fs(p); /* blocking */
2544bad_fork_cleanup_files:
2545 exit_files(p); /* blocking */
2546bad_fork_cleanup_semundo:
2547 exit_sem(p);
2548bad_fork_cleanup_security:
2549 security_task_free(p);
2550bad_fork_cleanup_audit:
2551 audit_free(p);
2552bad_fork_cleanup_perf:
2553 perf_event_free_task(p);
2554bad_fork_cleanup_policy:
2555 lockdep_free_task(p);
2556#ifdef CONFIG_NUMA
2557 mpol_put(p->mempolicy);
2558#endif
2559bad_fork_cleanup_delayacct:
2560 delayacct_tsk_free(p);
2561bad_fork_cleanup_count:
2562 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2563 exit_creds(p);
2564bad_fork_free:
2565 WRITE_ONCE(p->__state, TASK_DEAD);
2566 exit_task_stack_account(p);
2567 put_task_stack(p);
2568 delayed_free_task(p);
2569fork_out:
2570 spin_lock_irq(&current->sighand->siglock);
2571 hlist_del_init(&delayed.node);
2572 spin_unlock_irq(&current->sighand->siglock);
2573 return ERR_PTR(retval);
2574}
2575
2576static inline void init_idle_pids(struct task_struct *idle)
2577{
2578 enum pid_type type;
2579
2580 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2581 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2582 init_task_pid(idle, type, &init_struct_pid);
2583 }
2584}
2585
2586static int idle_dummy(void *dummy)
2587{
2588 /* This function is never called */
2589 return 0;
2590}
2591
2592struct task_struct * __init fork_idle(int cpu)
2593{
2594 struct task_struct *task;
2595 struct kernel_clone_args args = {
2596 .flags = CLONE_VM,
2597 .fn = &idle_dummy,
2598 .fn_arg = NULL,
2599 .kthread = 1,
2600 .idle = 1,
2601 };
2602
2603 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2604 if (!IS_ERR(task)) {
2605 init_idle_pids(task);
2606 init_idle(task, cpu);
2607 }
2608
2609 return task;
2610}
2611
2612/*
2613 * This is like kernel_clone(), but shaved down and tailored to just
2614 * creating io_uring workers. It returns a created task, or an error pointer.
2615 * The returned task is inactive, and the caller must fire it up through
2616 * wake_up_new_task(p). All signals are blocked in the created task.
2617 */
2618struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2619{
2620 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2621 CLONE_IO;
2622 struct kernel_clone_args args = {
2623 .flags = ((lower_32_bits(flags) | CLONE_VM |
2624 CLONE_UNTRACED) & ~CSIGNAL),
2625 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2626 .fn = fn,
2627 .fn_arg = arg,
2628 .io_thread = 1,
2629 };
2630
2631 return copy_process(NULL, 0, node, &args);
2632}
2633
2634/*
2635 * Ok, this is the main fork-routine.
2636 *
2637 * It copies the process, and if successful kick-starts
2638 * it and waits for it to finish using the VM if required.
2639 *
2640 * args->exit_signal is expected to be checked for sanity by the caller.
2641 */
2642pid_t kernel_clone(struct kernel_clone_args *args)
2643{
2644 u64 clone_flags = args->flags;
2645 struct completion vfork;
2646 struct pid *pid;
2647 struct task_struct *p;
2648 int trace = 0;
2649 pid_t nr;
2650
2651 /*
2652 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2653 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2654 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2655 * field in struct clone_args and it still doesn't make sense to have
2656 * them both point at the same memory location. Performing this check
2657 * here has the advantage that we don't need to have a separate helper
2658 * to check for legacy clone().
2659 */
2660 if ((args->flags & CLONE_PIDFD) &&
2661 (args->flags & CLONE_PARENT_SETTID) &&
2662 (args->pidfd == args->parent_tid))
2663 return -EINVAL;
2664
2665 /*
2666 * Determine whether and which event to report to ptracer. When
2667 * called from kernel_thread or CLONE_UNTRACED is explicitly
2668 * requested, no event is reported; otherwise, report if the event
2669 * for the type of forking is enabled.
2670 */
2671 if (!(clone_flags & CLONE_UNTRACED)) {
2672 if (clone_flags & CLONE_VFORK)
2673 trace = PTRACE_EVENT_VFORK;
2674 else if (args->exit_signal != SIGCHLD)
2675 trace = PTRACE_EVENT_CLONE;
2676 else
2677 trace = PTRACE_EVENT_FORK;
2678
2679 if (likely(!ptrace_event_enabled(current, trace)))
2680 trace = 0;
2681 }
2682
2683 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2684 add_latent_entropy();
2685
2686 if (IS_ERR(p))
2687 return PTR_ERR(p);
2688
2689 /*
2690 * Do this prior waking up the new thread - the thread pointer
2691 * might get invalid after that point, if the thread exits quickly.
2692 */
2693 trace_sched_process_fork(current, p);
2694
2695 pid = get_task_pid(p, PIDTYPE_PID);
2696 nr = pid_vnr(pid);
2697
2698 if (clone_flags & CLONE_PARENT_SETTID)
2699 put_user(nr, args->parent_tid);
2700
2701 if (clone_flags & CLONE_VFORK) {
2702 p->vfork_done = &vfork;
2703 init_completion(&vfork);
2704 get_task_struct(p);
2705 }
2706
2707 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2708 /* lock the task to synchronize with memcg migration */
2709 task_lock(p);
2710 lru_gen_add_mm(p->mm);
2711 task_unlock(p);
2712 }
2713
2714 wake_up_new_task(p);
2715
2716 /* forking complete and child started to run, tell ptracer */
2717 if (unlikely(trace))
2718 ptrace_event_pid(trace, pid);
2719
2720 if (clone_flags & CLONE_VFORK) {
2721 if (!wait_for_vfork_done(p, &vfork))
2722 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2723 }
2724
2725 put_pid(pid);
2726 return nr;
2727}
2728
2729/*
2730 * Create a kernel thread.
2731 */
2732pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2733{
2734 struct kernel_clone_args args = {
2735 .flags = ((lower_32_bits(flags) | CLONE_VM |
2736 CLONE_UNTRACED) & ~CSIGNAL),
2737 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2738 .fn = fn,
2739 .fn_arg = arg,
2740 .kthread = 1,
2741 };
2742
2743 return kernel_clone(&args);
2744}
2745
2746/*
2747 * Create a user mode thread.
2748 */
2749pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2750{
2751 struct kernel_clone_args args = {
2752 .flags = ((lower_32_bits(flags) | CLONE_VM |
2753 CLONE_UNTRACED) & ~CSIGNAL),
2754 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2755 .fn = fn,
2756 .fn_arg = arg,
2757 };
2758
2759 return kernel_clone(&args);
2760}
2761
2762#ifdef __ARCH_WANT_SYS_FORK
2763SYSCALL_DEFINE0(fork)
2764{
2765#ifdef CONFIG_MMU
2766 struct kernel_clone_args args = {
2767 .exit_signal = SIGCHLD,
2768 };
2769
2770 return kernel_clone(&args);
2771#else
2772 /* can not support in nommu mode */
2773 return -EINVAL;
2774#endif
2775}
2776#endif
2777
2778#ifdef __ARCH_WANT_SYS_VFORK
2779SYSCALL_DEFINE0(vfork)
2780{
2781 struct kernel_clone_args args = {
2782 .flags = CLONE_VFORK | CLONE_VM,
2783 .exit_signal = SIGCHLD,
2784 };
2785
2786 return kernel_clone(&args);
2787}
2788#endif
2789
2790#ifdef __ARCH_WANT_SYS_CLONE
2791#ifdef CONFIG_CLONE_BACKWARDS
2792SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2793 int __user *, parent_tidptr,
2794 unsigned long, tls,
2795 int __user *, child_tidptr)
2796#elif defined(CONFIG_CLONE_BACKWARDS2)
2797SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2798 int __user *, parent_tidptr,
2799 int __user *, child_tidptr,
2800 unsigned long, tls)
2801#elif defined(CONFIG_CLONE_BACKWARDS3)
2802SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2803 int, stack_size,
2804 int __user *, parent_tidptr,
2805 int __user *, child_tidptr,
2806 unsigned long, tls)
2807#else
2808SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2809 int __user *, parent_tidptr,
2810 int __user *, child_tidptr,
2811 unsigned long, tls)
2812#endif
2813{
2814 struct kernel_clone_args args = {
2815 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2816 .pidfd = parent_tidptr,
2817 .child_tid = child_tidptr,
2818 .parent_tid = parent_tidptr,
2819 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2820 .stack = newsp,
2821 .tls = tls,
2822 };
2823
2824 return kernel_clone(&args);
2825}
2826#endif
2827
2828#ifdef __ARCH_WANT_SYS_CLONE3
2829
2830noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2831 struct clone_args __user *uargs,
2832 size_t usize)
2833{
2834 int err;
2835 struct clone_args args;
2836 pid_t *kset_tid = kargs->set_tid;
2837
2838 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2839 CLONE_ARGS_SIZE_VER0);
2840 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2841 CLONE_ARGS_SIZE_VER1);
2842 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2843 CLONE_ARGS_SIZE_VER2);
2844 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2845
2846 if (unlikely(usize > PAGE_SIZE))
2847 return -E2BIG;
2848 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2849 return -EINVAL;
2850
2851 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2852 if (err)
2853 return err;
2854
2855 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2856 return -EINVAL;
2857
2858 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2859 return -EINVAL;
2860
2861 if (unlikely(args.set_tid && args.set_tid_size == 0))
2862 return -EINVAL;
2863
2864 /*
2865 * Verify that higher 32bits of exit_signal are unset and that
2866 * it is a valid signal
2867 */
2868 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2869 !valid_signal(args.exit_signal)))
2870 return -EINVAL;
2871
2872 if ((args.flags & CLONE_INTO_CGROUP) &&
2873 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2874 return -EINVAL;
2875
2876 *kargs = (struct kernel_clone_args){
2877 .flags = args.flags,
2878 .pidfd = u64_to_user_ptr(args.pidfd),
2879 .child_tid = u64_to_user_ptr(args.child_tid),
2880 .parent_tid = u64_to_user_ptr(args.parent_tid),
2881 .exit_signal = args.exit_signal,
2882 .stack = args.stack,
2883 .stack_size = args.stack_size,
2884 .tls = args.tls,
2885 .set_tid_size = args.set_tid_size,
2886 .cgroup = args.cgroup,
2887 };
2888
2889 if (args.set_tid &&
2890 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2891 (kargs->set_tid_size * sizeof(pid_t))))
2892 return -EFAULT;
2893
2894 kargs->set_tid = kset_tid;
2895
2896 return 0;
2897}
2898
2899/**
2900 * clone3_stack_valid - check and prepare stack
2901 * @kargs: kernel clone args
2902 *
2903 * Verify that the stack arguments userspace gave us are sane.
2904 * In addition, set the stack direction for userspace since it's easy for us to
2905 * determine.
2906 */
2907static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2908{
2909 if (kargs->stack == 0) {
2910 if (kargs->stack_size > 0)
2911 return false;
2912 } else {
2913 if (kargs->stack_size == 0)
2914 return false;
2915
2916 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2917 return false;
2918
2919#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2920 kargs->stack += kargs->stack_size;
2921#endif
2922 }
2923
2924 return true;
2925}
2926
2927static bool clone3_args_valid(struct kernel_clone_args *kargs)
2928{
2929 /* Verify that no unknown flags are passed along. */
2930 if (kargs->flags &
2931 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2932 return false;
2933
2934 /*
2935 * - make the CLONE_DETACHED bit reusable for clone3
2936 * - make the CSIGNAL bits reusable for clone3
2937 */
2938 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2939 return false;
2940
2941 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2942 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2943 return false;
2944
2945 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2946 kargs->exit_signal)
2947 return false;
2948
2949 if (!clone3_stack_valid(kargs))
2950 return false;
2951
2952 return true;
2953}
2954
2955/**
2956 * clone3 - create a new process with specific properties
2957 * @uargs: argument structure
2958 * @size: size of @uargs
2959 *
2960 * clone3() is the extensible successor to clone()/clone2().
2961 * It takes a struct as argument that is versioned by its size.
2962 *
2963 * Return: On success, a positive PID for the child process.
2964 * On error, a negative errno number.
2965 */
2966SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2967{
2968 int err;
2969
2970 struct kernel_clone_args kargs;
2971 pid_t set_tid[MAX_PID_NS_LEVEL];
2972
2973 kargs.set_tid = set_tid;
2974
2975 err = copy_clone_args_from_user(&kargs, uargs, size);
2976 if (err)
2977 return err;
2978
2979 if (!clone3_args_valid(&kargs))
2980 return -EINVAL;
2981
2982 return kernel_clone(&kargs);
2983}
2984#endif
2985
2986void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2987{
2988 struct task_struct *leader, *parent, *child;
2989 int res;
2990
2991 read_lock(&tasklist_lock);
2992 leader = top = top->group_leader;
2993down:
2994 for_each_thread(leader, parent) {
2995 list_for_each_entry(child, &parent->children, sibling) {
2996 res = visitor(child, data);
2997 if (res) {
2998 if (res < 0)
2999 goto out;
3000 leader = child;
3001 goto down;
3002 }
3003up:
3004 ;
3005 }
3006 }
3007
3008 if (leader != top) {
3009 child = leader;
3010 parent = child->real_parent;
3011 leader = parent->group_leader;
3012 goto up;
3013 }
3014out:
3015 read_unlock(&tasklist_lock);
3016}
3017
3018#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3019#define ARCH_MIN_MMSTRUCT_ALIGN 0
3020#endif
3021
3022static void sighand_ctor(void *data)
3023{
3024 struct sighand_struct *sighand = data;
3025
3026 spin_lock_init(&sighand->siglock);
3027 init_waitqueue_head(&sighand->signalfd_wqh);
3028}
3029
3030void __init mm_cache_init(void)
3031{
3032 unsigned int mm_size;
3033
3034 /*
3035 * The mm_cpumask is located at the end of mm_struct, and is
3036 * dynamically sized based on the maximum CPU number this system
3037 * can have, taking hotplug into account (nr_cpu_ids).
3038 */
3039 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3040
3041 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3042 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3043 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3044 offsetof(struct mm_struct, saved_auxv),
3045 sizeof_field(struct mm_struct, saved_auxv),
3046 NULL);
3047}
3048
3049void __init proc_caches_init(void)
3050{
3051 sighand_cachep = kmem_cache_create("sighand_cache",
3052 sizeof(struct sighand_struct), 0,
3053 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3054 SLAB_ACCOUNT, sighand_ctor);
3055 signal_cachep = kmem_cache_create("signal_cache",
3056 sizeof(struct signal_struct), 0,
3057 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3058 NULL);
3059 files_cachep = kmem_cache_create("files_cache",
3060 sizeof(struct files_struct), 0,
3061 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3062 NULL);
3063 fs_cachep = kmem_cache_create("fs_cache",
3064 sizeof(struct fs_struct), 0,
3065 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3066 NULL);
3067
3068 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3069 mmap_init();
3070 nsproxy_cache_init();
3071}
3072
3073/*
3074 * Check constraints on flags passed to the unshare system call.
3075 */
3076static int check_unshare_flags(unsigned long unshare_flags)
3077{
3078 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3079 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3080 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3081 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3082 CLONE_NEWTIME))
3083 return -EINVAL;
3084 /*
3085 * Not implemented, but pretend it works if there is nothing
3086 * to unshare. Note that unsharing the address space or the
3087 * signal handlers also need to unshare the signal queues (aka
3088 * CLONE_THREAD).
3089 */
3090 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3091 if (!thread_group_empty(current))
3092 return -EINVAL;
3093 }
3094 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3095 if (refcount_read(&current->sighand->count) > 1)
3096 return -EINVAL;
3097 }
3098 if (unshare_flags & CLONE_VM) {
3099 if (!current_is_single_threaded())
3100 return -EINVAL;
3101 }
3102
3103 return 0;
3104}
3105
3106/*
3107 * Unshare the filesystem structure if it is being shared
3108 */
3109static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3110{
3111 struct fs_struct *fs = current->fs;
3112
3113 if (!(unshare_flags & CLONE_FS) || !fs)
3114 return 0;
3115
3116 /* don't need lock here; in the worst case we'll do useless copy */
3117 if (fs->users == 1)
3118 return 0;
3119
3120 *new_fsp = copy_fs_struct(fs);
3121 if (!*new_fsp)
3122 return -ENOMEM;
3123
3124 return 0;
3125}
3126
3127/*
3128 * Unshare file descriptor table if it is being shared
3129 */
3130int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3131 struct files_struct **new_fdp)
3132{
3133 struct files_struct *fd = current->files;
3134 int error = 0;
3135
3136 if ((unshare_flags & CLONE_FILES) &&
3137 (fd && atomic_read(&fd->count) > 1)) {
3138 *new_fdp = dup_fd(fd, max_fds, &error);
3139 if (!*new_fdp)
3140 return error;
3141 }
3142
3143 return 0;
3144}
3145
3146/*
3147 * unshare allows a process to 'unshare' part of the process
3148 * context which was originally shared using clone. copy_*
3149 * functions used by kernel_clone() cannot be used here directly
3150 * because they modify an inactive task_struct that is being
3151 * constructed. Here we are modifying the current, active,
3152 * task_struct.
3153 */
3154int ksys_unshare(unsigned long unshare_flags)
3155{
3156 struct fs_struct *fs, *new_fs = NULL;
3157 struct files_struct *new_fd = NULL;
3158 struct cred *new_cred = NULL;
3159 struct nsproxy *new_nsproxy = NULL;
3160 int do_sysvsem = 0;
3161 int err;
3162
3163 /*
3164 * If unsharing a user namespace must also unshare the thread group
3165 * and unshare the filesystem root and working directories.
3166 */
3167 if (unshare_flags & CLONE_NEWUSER)
3168 unshare_flags |= CLONE_THREAD | CLONE_FS;
3169 /*
3170 * If unsharing vm, must also unshare signal handlers.
3171 */
3172 if (unshare_flags & CLONE_VM)
3173 unshare_flags |= CLONE_SIGHAND;
3174 /*
3175 * If unsharing a signal handlers, must also unshare the signal queues.
3176 */
3177 if (unshare_flags & CLONE_SIGHAND)
3178 unshare_flags |= CLONE_THREAD;
3179 /*
3180 * If unsharing namespace, must also unshare filesystem information.
3181 */
3182 if (unshare_flags & CLONE_NEWNS)
3183 unshare_flags |= CLONE_FS;
3184
3185 err = check_unshare_flags(unshare_flags);
3186 if (err)
3187 goto bad_unshare_out;
3188 /*
3189 * CLONE_NEWIPC must also detach from the undolist: after switching
3190 * to a new ipc namespace, the semaphore arrays from the old
3191 * namespace are unreachable.
3192 */
3193 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3194 do_sysvsem = 1;
3195 err = unshare_fs(unshare_flags, &new_fs);
3196 if (err)
3197 goto bad_unshare_out;
3198 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3199 if (err)
3200 goto bad_unshare_cleanup_fs;
3201 err = unshare_userns(unshare_flags, &new_cred);
3202 if (err)
3203 goto bad_unshare_cleanup_fd;
3204 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3205 new_cred, new_fs);
3206 if (err)
3207 goto bad_unshare_cleanup_cred;
3208
3209 if (new_cred) {
3210 err = set_cred_ucounts(new_cred);
3211 if (err)
3212 goto bad_unshare_cleanup_cred;
3213 }
3214
3215 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3216 if (do_sysvsem) {
3217 /*
3218 * CLONE_SYSVSEM is equivalent to sys_exit().
3219 */
3220 exit_sem(current);
3221 }
3222 if (unshare_flags & CLONE_NEWIPC) {
3223 /* Orphan segments in old ns (see sem above). */
3224 exit_shm(current);
3225 shm_init_task(current);
3226 }
3227
3228 if (new_nsproxy)
3229 switch_task_namespaces(current, new_nsproxy);
3230
3231 task_lock(current);
3232
3233 if (new_fs) {
3234 fs = current->fs;
3235 spin_lock(&fs->lock);
3236 current->fs = new_fs;
3237 if (--fs->users)
3238 new_fs = NULL;
3239 else
3240 new_fs = fs;
3241 spin_unlock(&fs->lock);
3242 }
3243
3244 if (new_fd)
3245 swap(current->files, new_fd);
3246
3247 task_unlock(current);
3248
3249 if (new_cred) {
3250 /* Install the new user namespace */
3251 commit_creds(new_cred);
3252 new_cred = NULL;
3253 }
3254 }
3255
3256 perf_event_namespaces(current);
3257
3258bad_unshare_cleanup_cred:
3259 if (new_cred)
3260 put_cred(new_cred);
3261bad_unshare_cleanup_fd:
3262 if (new_fd)
3263 put_files_struct(new_fd);
3264
3265bad_unshare_cleanup_fs:
3266 if (new_fs)
3267 free_fs_struct(new_fs);
3268
3269bad_unshare_out:
3270 return err;
3271}
3272
3273SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3274{
3275 return ksys_unshare(unshare_flags);
3276}
3277
3278/*
3279 * Helper to unshare the files of the current task.
3280 * We don't want to expose copy_files internals to
3281 * the exec layer of the kernel.
3282 */
3283
3284int unshare_files(void)
3285{
3286 struct task_struct *task = current;
3287 struct files_struct *old, *copy = NULL;
3288 int error;
3289
3290 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3291 if (error || !copy)
3292 return error;
3293
3294 old = task->files;
3295 task_lock(task);
3296 task->files = copy;
3297 task_unlock(task);
3298 put_files_struct(old);
3299 return 0;
3300}
3301
3302int sysctl_max_threads(struct ctl_table *table, int write,
3303 void *buffer, size_t *lenp, loff_t *ppos)
3304{
3305 struct ctl_table t;
3306 int ret;
3307 int threads = max_threads;
3308 int min = 1;
3309 int max = MAX_THREADS;
3310
3311 t = *table;
3312 t.data = &threads;
3313 t.extra1 = &min;
3314 t.extra2 = &max;
3315
3316 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3317 if (ret || !write)
3318 return ret;
3319
3320 max_threads = threads;
3321
3322 return 0;
3323}