projects/linux·blob v6.3-rc3 · refs

3324 lines · 80.2 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 for_each_vma(old_vmi, mpnt) {
621 struct file *file;
622
623 if (mpnt->vm_flags & VM_DONTCOPY) {
624 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
625 continue;
626 }
627 charge = 0;
628 /*
629 * Don't duplicate many vmas if we've been oom-killed (for
630 * example)
631 */
632 if (fatal_signal_pending(current)) {
633 retval = -EINTR;
634 goto loop_out;
635 }
636 if (mpnt->vm_flags & VM_ACCOUNT) {
637 unsigned long len = vma_pages(mpnt);
638
639 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
640 goto fail_nomem;
641 charge = len;
642 }
643 tmp = vm_area_dup(mpnt);
644 if (!tmp)
645 goto fail_nomem;
646 retval = vma_dup_policy(mpnt, tmp);
647 if (retval)
648 goto fail_nomem_policy;
649 tmp->vm_mm = mm;
650 retval = dup_userfaultfd(tmp, &uf);
651 if (retval)
652 goto fail_nomem_anon_vma_fork;
653 if (tmp->vm_flags & VM_WIPEONFORK) {
654 /*
655 * VM_WIPEONFORK gets a clean slate in the child.
656 * Don't prepare anon_vma until fault since we don't
657 * copy page for current vma.
658 */
659 tmp->anon_vma = NULL;
660 } else if (anon_vma_fork(tmp, mpnt))
661 goto fail_nomem_anon_vma_fork;
662 vm_flags_clear(tmp, VM_LOCKED_MASK);
663 file = tmp->vm_file;
664 if (file) {
665 struct address_space *mapping = file->f_mapping;
666
667 get_file(file);
668 i_mmap_lock_write(mapping);
669 if (tmp->vm_flags & VM_SHARED)
670 mapping_allow_writable(mapping);
671 flush_dcache_mmap_lock(mapping);
672 /* insert tmp into the share list, just after mpnt */
673 vma_interval_tree_insert_after(tmp, mpnt,
674 &mapping->i_mmap);
675 flush_dcache_mmap_unlock(mapping);
676 i_mmap_unlock_write(mapping);
677 }
678
679 /*
680 * Copy/update hugetlb private vma information.
681 */
682 if (is_vm_hugetlb_page(tmp))
683 hugetlb_dup_vma_private(tmp);
684
685 /* Link the vma into the MT */
686 if (vma_iter_bulk_store(&vmi, tmp))
687 goto fail_nomem_vmi_store;
688
689 mm->map_count++;
690 if (!(tmp->vm_flags & VM_WIPEONFORK))
691 retval = copy_page_range(tmp, mpnt);
692
693 if (tmp->vm_ops && tmp->vm_ops->open)
694 tmp->vm_ops->open(tmp);
695
696 if (retval)
697 goto loop_out;
698 }
699 /* a new mm has just been created */
700 retval = arch_dup_mmap(oldmm, mm);
701loop_out:
702 vma_iter_free(&vmi);
703out:
704 mmap_write_unlock(mm);
705 flush_tlb_mm(oldmm);
706 mmap_write_unlock(oldmm);
707 dup_userfaultfd_complete(&uf);
708fail_uprobe_end:
709 uprobe_end_dup_mmap();
710 return retval;
711
712fail_nomem_vmi_store:
713 unlink_anon_vmas(tmp);
714fail_nomem_anon_vma_fork:
715 mpol_put(vma_policy(tmp));
716fail_nomem_policy:
717 vm_area_free(tmp);
718fail_nomem:
719 retval = -ENOMEM;
720 vm_unacct_memory(charge);
721 goto loop_out;
722}
723
724static inline int mm_alloc_pgd(struct mm_struct *mm)
725{
726 mm->pgd = pgd_alloc(mm);
727 if (unlikely(!mm->pgd))
728 return -ENOMEM;
729 return 0;
730}
731
732static inline void mm_free_pgd(struct mm_struct *mm)
733{
734 pgd_free(mm, mm->pgd);
735}
736#else
737static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
738{
739 mmap_write_lock(oldmm);
740 dup_mm_exe_file(mm, oldmm);
741 mmap_write_unlock(oldmm);
742 return 0;
743}
744#define mm_alloc_pgd(mm) (0)
745#define mm_free_pgd(mm)
746#endif /* CONFIG_MMU */
747
748static void check_mm(struct mm_struct *mm)
749{
750 int i;
751
752 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
753 "Please make sure 'struct resident_page_types[]' is updated as well");
754
755 for (i = 0; i < NR_MM_COUNTERS; i++) {
756 long x = percpu_counter_sum(&mm->rss_stat[i]);
757
758 if (likely(!x))
759 continue;
760
761 /* Making sure this is not due to race with CPU offlining. */
762 x = percpu_counter_sum_all(&mm->rss_stat[i]);
763 if (unlikely(x))
764 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
765 mm, resident_page_types[i], x);
766 }
767
768 if (mm_pgtables_bytes(mm))
769 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
770 mm_pgtables_bytes(mm));
771
772#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
773 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
774#endif
775}
776
777#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
778#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
779
780/*
781 * Called when the last reference to the mm
782 * is dropped: either by a lazy thread or by
783 * mmput. Free the page directory and the mm.
784 */
785void __mmdrop(struct mm_struct *mm)
786{
787 int i;
788
789 BUG_ON(mm == &init_mm);
790 WARN_ON_ONCE(mm == current->mm);
791 WARN_ON_ONCE(mm == current->active_mm);
792 mm_free_pgd(mm);
793 destroy_context(mm);
794 mmu_notifier_subscriptions_destroy(mm);
795 check_mm(mm);
796 put_user_ns(mm->user_ns);
797 mm_pasid_drop(mm);
798
799 for (i = 0; i < NR_MM_COUNTERS; i++)
800 percpu_counter_destroy(&mm->rss_stat[i]);
801 free_mm(mm);
802}
803EXPORT_SYMBOL_GPL(__mmdrop);
804
805static void mmdrop_async_fn(struct work_struct *work)
806{
807 struct mm_struct *mm;
808
809 mm = container_of(work, struct mm_struct, async_put_work);
810 __mmdrop(mm);
811}
812
813static void mmdrop_async(struct mm_struct *mm)
814{
815 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
816 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
817 schedule_work(&mm->async_put_work);
818 }
819}
820
821static inline void free_signal_struct(struct signal_struct *sig)
822{
823 taskstats_tgid_free(sig);
824 sched_autogroup_exit(sig);
825 /*
826 * __mmdrop is not safe to call from softirq context on x86 due to
827 * pgd_dtor so postpone it to the async context
828 */
829 if (sig->oom_mm)
830 mmdrop_async(sig->oom_mm);
831 kmem_cache_free(signal_cachep, sig);
832}
833
834static inline void put_signal_struct(struct signal_struct *sig)
835{
836 if (refcount_dec_and_test(&sig->sigcnt))
837 free_signal_struct(sig);
838}
839
840void __put_task_struct(struct task_struct *tsk)
841{
842 WARN_ON(!tsk->exit_state);
843 WARN_ON(refcount_read(&tsk->usage));
844 WARN_ON(tsk == current);
845
846 io_uring_free(tsk);
847 cgroup_free(tsk);
848 task_numa_free(tsk, true);
849 security_task_free(tsk);
850 bpf_task_storage_free(tsk);
851 exit_creds(tsk);
852 delayacct_tsk_free(tsk);
853 put_signal_struct(tsk->signal);
854 sched_core_free(tsk);
855 free_task(tsk);
856}
857EXPORT_SYMBOL_GPL(__put_task_struct);
858
859void __init __weak arch_task_cache_init(void) { }
860
861/*
862 * set_max_threads
863 */
864static void set_max_threads(unsigned int max_threads_suggested)
865{
866 u64 threads;
867 unsigned long nr_pages = totalram_pages();
868
869 /*
870 * The number of threads shall be limited such that the thread
871 * structures may only consume a small part of the available memory.
872 */
873 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
874 threads = MAX_THREADS;
875 else
876 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
877 (u64) THREAD_SIZE * 8UL);
878
879 if (threads > max_threads_suggested)
880 threads = max_threads_suggested;
881
882 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
883}
884
885#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
886/* Initialized by the architecture: */
887int arch_task_struct_size __read_mostly;
888#endif
889
890#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
891static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
892{
893 /* Fetch thread_struct whitelist for the architecture. */
894 arch_thread_struct_whitelist(offset, size);
895
896 /*
897 * Handle zero-sized whitelist or empty thread_struct, otherwise
898 * adjust offset to position of thread_struct in task_struct.
899 */
900 if (unlikely(*size == 0))
901 *offset = 0;
902 else
903 *offset += offsetof(struct task_struct, thread);
904}
905#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
906
907void __init fork_init(void)
908{
909 int i;
910#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
911#ifndef ARCH_MIN_TASKALIGN
912#define ARCH_MIN_TASKALIGN 0
913#endif
914 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
915 unsigned long useroffset, usersize;
916
917 /* create a slab on which task_structs can be allocated */
918 task_struct_whitelist(&useroffset, &usersize);
919 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
920 arch_task_struct_size, align,
921 SLAB_PANIC|SLAB_ACCOUNT,
922 useroffset, usersize, NULL);
923#endif
924
925 /* do the arch specific task caches init */
926 arch_task_cache_init();
927
928 set_max_threads(MAX_THREADS);
929
930 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
931 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
932 init_task.signal->rlim[RLIMIT_SIGPENDING] =
933 init_task.signal->rlim[RLIMIT_NPROC];
934
935 for (i = 0; i < UCOUNT_COUNTS; i++)
936 init_user_ns.ucount_max[i] = max_threads/2;
937
938 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
939 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
940 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
941 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
942
943#ifdef CONFIG_VMAP_STACK
944 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
945 NULL, free_vm_stack_cache);
946#endif
947
948 scs_init();
949
950 lockdep_init_task(&init_task);
951 uprobes_init();
952}
953
954int __weak arch_dup_task_struct(struct task_struct *dst,
955 struct task_struct *src)
956{
957 *dst = *src;
958 return 0;
959}
960
961void set_task_stack_end_magic(struct task_struct *tsk)
962{
963 unsigned long *stackend;
964
965 stackend = end_of_stack(tsk);
966 *stackend = STACK_END_MAGIC; /* for overflow detection */
967}
968
969static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
970{
971 struct task_struct *tsk;
972 int err;
973
974 if (node == NUMA_NO_NODE)
975 node = tsk_fork_get_node(orig);
976 tsk = alloc_task_struct_node(node);
977 if (!tsk)
978 return NULL;
979
980 err = arch_dup_task_struct(tsk, orig);
981 if (err)
982 goto free_tsk;
983
984 err = alloc_thread_stack_node(tsk, node);
985 if (err)
986 goto free_tsk;
987
988#ifdef CONFIG_THREAD_INFO_IN_TASK
989 refcount_set(&tsk->stack_refcount, 1);
990#endif
991 account_kernel_stack(tsk, 1);
992
993 err = scs_prepare(tsk, node);
994 if (err)
995 goto free_stack;
996
997#ifdef CONFIG_SECCOMP
998 /*
999 * We must handle setting up seccomp filters once we're under
1000 * the sighand lock in case orig has changed between now and
1001 * then. Until then, filter must be NULL to avoid messing up
1002 * the usage counts on the error path calling free_task.
1003 */
1004 tsk->seccomp.filter = NULL;
1005#endif
1006
1007 setup_thread_stack(tsk, orig);
1008 clear_user_return_notifier(tsk);
1009 clear_tsk_need_resched(tsk);
1010 set_task_stack_end_magic(tsk);
1011 clear_syscall_work_syscall_user_dispatch(tsk);
1012
1013#ifdef CONFIG_STACKPROTECTOR
1014 tsk->stack_canary = get_random_canary();
1015#endif
1016 if (orig->cpus_ptr == &orig->cpus_mask)
1017 tsk->cpus_ptr = &tsk->cpus_mask;
1018 dup_user_cpus_ptr(tsk, orig, node);
1019
1020 /*
1021 * One for the user space visible state that goes away when reaped.
1022 * One for the scheduler.
1023 */
1024 refcount_set(&tsk->rcu_users, 2);
1025 /* One for the rcu users */
1026 refcount_set(&tsk->usage, 1);
1027#ifdef CONFIG_BLK_DEV_IO_TRACE
1028 tsk->btrace_seq = 0;
1029#endif
1030 tsk->splice_pipe = NULL;
1031 tsk->task_frag.page = NULL;
1032 tsk->wake_q.next = NULL;
1033 tsk->worker_private = NULL;
1034
1035 kcov_task_init(tsk);
1036 kmsan_task_create(tsk);
1037 kmap_local_fork(tsk);
1038
1039#ifdef CONFIG_FAULT_INJECTION
1040 tsk->fail_nth = 0;
1041#endif
1042
1043#ifdef CONFIG_BLK_CGROUP
1044 tsk->throttle_disk = NULL;
1045 tsk->use_memdelay = 0;
1046#endif
1047
1048#ifdef CONFIG_IOMMU_SVA
1049 tsk->pasid_activated = 0;
1050#endif
1051
1052#ifdef CONFIG_MEMCG
1053 tsk->active_memcg = NULL;
1054#endif
1055
1056#ifdef CONFIG_CPU_SUP_INTEL
1057 tsk->reported_split_lock = 0;
1058#endif
1059
1060#ifdef CONFIG_SCHED_MM_CID
1061 tsk->mm_cid = -1;
1062 tsk->mm_cid_active = 0;
1063#endif
1064 return tsk;
1065
1066free_stack:
1067 exit_task_stack_account(tsk);
1068 free_thread_stack(tsk);
1069free_tsk:
1070 free_task_struct(tsk);
1071 return NULL;
1072}
1073
1074__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1075
1076static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1077
1078static int __init coredump_filter_setup(char *s)
1079{
1080 default_dump_filter =
1081 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1082 MMF_DUMP_FILTER_MASK;
1083 return 1;
1084}
1085
1086__setup("coredump_filter=", coredump_filter_setup);
1087
1088#include <linux/init_task.h>
1089
1090static void mm_init_aio(struct mm_struct *mm)
1091{
1092#ifdef CONFIG_AIO
1093 spin_lock_init(&mm->ioctx_lock);
1094 mm->ioctx_table = NULL;
1095#endif
1096}
1097
1098static __always_inline void mm_clear_owner(struct mm_struct *mm,
1099 struct task_struct *p)
1100{
1101#ifdef CONFIG_MEMCG
1102 if (mm->owner == p)
1103 WRITE_ONCE(mm->owner, NULL);
1104#endif
1105}
1106
1107static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1108{
1109#ifdef CONFIG_MEMCG
1110 mm->owner = p;
1111#endif
1112}
1113
1114static void mm_init_uprobes_state(struct mm_struct *mm)
1115{
1116#ifdef CONFIG_UPROBES
1117 mm->uprobes_state.xol_area = NULL;
1118#endif
1119}
1120
1121static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1122 struct user_namespace *user_ns)
1123{
1124 int i;
1125
1126 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1127 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1128 atomic_set(&mm->mm_users, 1);
1129 atomic_set(&mm->mm_count, 1);
1130 seqcount_init(&mm->write_protect_seq);
1131 mmap_init_lock(mm);
1132 INIT_LIST_HEAD(&mm->mmlist);
1133 mm_pgtables_bytes_init(mm);
1134 mm->map_count = 0;
1135 mm->locked_vm = 0;
1136 atomic64_set(&mm->pinned_vm, 0);
1137 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1138 spin_lock_init(&mm->page_table_lock);
1139 spin_lock_init(&mm->arg_lock);
1140 mm_init_cpumask(mm);
1141 mm_init_aio(mm);
1142 mm_init_owner(mm, p);
1143 mm_pasid_init(mm);
1144 RCU_INIT_POINTER(mm->exe_file, NULL);
1145 mmu_notifier_subscriptions_init(mm);
1146 init_tlb_flush_pending(mm);
1147#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1148 mm->pmd_huge_pte = NULL;
1149#endif
1150 mm_init_uprobes_state(mm);
1151 hugetlb_count_init(mm);
1152
1153 if (current->mm) {
1154 mm->flags = current->mm->flags & MMF_INIT_MASK;
1155 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1156 } else {
1157 mm->flags = default_dump_filter;
1158 mm->def_flags = 0;
1159 }
1160
1161 if (mm_alloc_pgd(mm))
1162 goto fail_nopgd;
1163
1164 if (init_new_context(p, mm))
1165 goto fail_nocontext;
1166
1167 for (i = 0; i < NR_MM_COUNTERS; i++)
1168 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1169 goto fail_pcpu;
1170
1171 mm->user_ns = get_user_ns(user_ns);
1172 lru_gen_init_mm(mm);
1173 mm_init_cid(mm);
1174 return mm;
1175
1176fail_pcpu:
1177 while (i > 0)
1178 percpu_counter_destroy(&mm->rss_stat[--i]);
1179fail_nocontext:
1180 mm_free_pgd(mm);
1181fail_nopgd:
1182 free_mm(mm);
1183 return NULL;
1184}
1185
1186/*
1187 * Allocate and initialize an mm_struct.
1188 */
1189struct mm_struct *mm_alloc(void)
1190{
1191 struct mm_struct *mm;
1192
1193 mm = allocate_mm();
1194 if (!mm)
1195 return NULL;
1196
1197 memset(mm, 0, sizeof(*mm));
1198 return mm_init(mm, current, current_user_ns());
1199}
1200
1201static inline void __mmput(struct mm_struct *mm)
1202{
1203 VM_BUG_ON(atomic_read(&mm->mm_users));
1204
1205 uprobe_clear_state(mm);
1206 exit_aio(mm);
1207 ksm_exit(mm);
1208 khugepaged_exit(mm); /* must run before exit_mmap */
1209 exit_mmap(mm);
1210 mm_put_huge_zero_page(mm);
1211 set_mm_exe_file(mm, NULL);
1212 if (!list_empty(&mm->mmlist)) {
1213 spin_lock(&mmlist_lock);
1214 list_del(&mm->mmlist);
1215 spin_unlock(&mmlist_lock);
1216 }
1217 if (mm->binfmt)
1218 module_put(mm->binfmt->module);
1219 lru_gen_del_mm(mm);
1220 mmdrop(mm);
1221}
1222
1223/*
1224 * Decrement the use count and release all resources for an mm.
1225 */
1226void mmput(struct mm_struct *mm)
1227{
1228 might_sleep();
1229
1230 if (atomic_dec_and_test(&mm->mm_users))
1231 __mmput(mm);
1232}
1233EXPORT_SYMBOL_GPL(mmput);
1234
1235#ifdef CONFIG_MMU
1236static void mmput_async_fn(struct work_struct *work)
1237{
1238 struct mm_struct *mm = container_of(work, struct mm_struct,
1239 async_put_work);
1240
1241 __mmput(mm);
1242}
1243
1244void mmput_async(struct mm_struct *mm)
1245{
1246 if (atomic_dec_and_test(&mm->mm_users)) {
1247 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1248 schedule_work(&mm->async_put_work);
1249 }
1250}
1251EXPORT_SYMBOL_GPL(mmput_async);
1252#endif
1253
1254/**
1255 * set_mm_exe_file - change a reference to the mm's executable file
1256 *
1257 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1258 *
1259 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1260 * invocations: in mmput() nobody alive left, in execve task is single
1261 * threaded.
1262 *
1263 * Can only fail if new_exe_file != NULL.
1264 */
1265int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1266{
1267 struct file *old_exe_file;
1268
1269 /*
1270 * It is safe to dereference the exe_file without RCU as
1271 * this function is only called if nobody else can access
1272 * this mm -- see comment above for justification.
1273 */
1274 old_exe_file = rcu_dereference_raw(mm->exe_file);
1275
1276 if (new_exe_file) {
1277 /*
1278 * We expect the caller (i.e., sys_execve) to already denied
1279 * write access, so this is unlikely to fail.
1280 */
1281 if (unlikely(deny_write_access(new_exe_file)))
1282 return -EACCES;
1283 get_file(new_exe_file);
1284 }
1285 rcu_assign_pointer(mm->exe_file, new_exe_file);
1286 if (old_exe_file) {
1287 allow_write_access(old_exe_file);
1288 fput(old_exe_file);
1289 }
1290 return 0;
1291}
1292
1293/**
1294 * replace_mm_exe_file - replace a reference to the mm's executable file
1295 *
1296 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1297 * dealing with concurrent invocation and without grabbing the mmap lock in
1298 * write mode.
1299 *
1300 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1301 */
1302int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1303{
1304 struct vm_area_struct *vma;
1305 struct file *old_exe_file;
1306 int ret = 0;
1307
1308 /* Forbid mm->exe_file change if old file still mapped. */
1309 old_exe_file = get_mm_exe_file(mm);
1310 if (old_exe_file) {
1311 VMA_ITERATOR(vmi, mm, 0);
1312 mmap_read_lock(mm);
1313 for_each_vma(vmi, vma) {
1314 if (!vma->vm_file)
1315 continue;
1316 if (path_equal(&vma->vm_file->f_path,
1317 &old_exe_file->f_path)) {
1318 ret = -EBUSY;
1319 break;
1320 }
1321 }
1322 mmap_read_unlock(mm);
1323 fput(old_exe_file);
1324 if (ret)
1325 return ret;
1326 }
1327
1328 /* set the new file, lockless */
1329 ret = deny_write_access(new_exe_file);
1330 if (ret)
1331 return -EACCES;
1332 get_file(new_exe_file);
1333
1334 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1335 if (old_exe_file) {
1336 /*
1337 * Don't race with dup_mmap() getting the file and disallowing
1338 * write access while someone might open the file writable.
1339 */
1340 mmap_read_lock(mm);
1341 allow_write_access(old_exe_file);
1342 fput(old_exe_file);
1343 mmap_read_unlock(mm);
1344 }
1345 return 0;
1346}
1347
1348/**
1349 * get_mm_exe_file - acquire a reference to the mm's executable file
1350 *
1351 * Returns %NULL if mm has no associated executable file.
1352 * User must release file via fput().
1353 */
1354struct file *get_mm_exe_file(struct mm_struct *mm)
1355{
1356 struct file *exe_file;
1357
1358 rcu_read_lock();
1359 exe_file = rcu_dereference(mm->exe_file);
1360 if (exe_file && !get_file_rcu(exe_file))
1361 exe_file = NULL;
1362 rcu_read_unlock();
1363 return exe_file;
1364}
1365
1366/**
1367 * get_task_exe_file - acquire a reference to the task's executable file
1368 *
1369 * Returns %NULL if task's mm (if any) has no associated executable file or
1370 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1371 * User must release file via fput().
1372 */
1373struct file *get_task_exe_file(struct task_struct *task)
1374{
1375 struct file *exe_file = NULL;
1376 struct mm_struct *mm;
1377
1378 task_lock(task);
1379 mm = task->mm;
1380 if (mm) {
1381 if (!(task->flags & PF_KTHREAD))
1382 exe_file = get_mm_exe_file(mm);
1383 }
1384 task_unlock(task);
1385 return exe_file;
1386}
1387
1388/**
1389 * get_task_mm - acquire a reference to the task's mm
1390 *
1391 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1392 * this kernel workthread has transiently adopted a user mm with use_mm,
1393 * to do its AIO) is not set and if so returns a reference to it, after
1394 * bumping up the use count. User must release the mm via mmput()
1395 * after use. Typically used by /proc and ptrace.
1396 */
1397struct mm_struct *get_task_mm(struct task_struct *task)
1398{
1399 struct mm_struct *mm;
1400
1401 task_lock(task);
1402 mm = task->mm;
1403 if (mm) {
1404 if (task->flags & PF_KTHREAD)
1405 mm = NULL;
1406 else
1407 mmget(mm);
1408 }
1409 task_unlock(task);
1410 return mm;
1411}
1412EXPORT_SYMBOL_GPL(get_task_mm);
1413
1414struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1415{
1416 struct mm_struct *mm;
1417 int err;
1418
1419 err = down_read_killable(&task->signal->exec_update_lock);
1420 if (err)
1421 return ERR_PTR(err);
1422
1423 mm = get_task_mm(task);
1424 if (mm && mm != current->mm &&
1425 !ptrace_may_access(task, mode)) {
1426 mmput(mm);
1427 mm = ERR_PTR(-EACCES);
1428 }
1429 up_read(&task->signal->exec_update_lock);
1430
1431 return mm;
1432}
1433
1434static void complete_vfork_done(struct task_struct *tsk)
1435{
1436 struct completion *vfork;
1437
1438 task_lock(tsk);
1439 vfork = tsk->vfork_done;
1440 if (likely(vfork)) {
1441 tsk->vfork_done = NULL;
1442 complete(vfork);
1443 }
1444 task_unlock(tsk);
1445}
1446
1447static int wait_for_vfork_done(struct task_struct *child,
1448 struct completion *vfork)
1449{
1450 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1451 int killed;
1452
1453 cgroup_enter_frozen();
1454 killed = wait_for_completion_state(vfork, state);
1455 cgroup_leave_frozen(false);
1456
1457 if (killed) {
1458 task_lock(child);
1459 child->vfork_done = NULL;
1460 task_unlock(child);
1461 }
1462
1463 put_task_struct(child);
1464 return killed;
1465}
1466
1467/* Please note the differences between mmput and mm_release.
1468 * mmput is called whenever we stop holding onto a mm_struct,
1469 * error success whatever.
1470 *
1471 * mm_release is called after a mm_struct has been removed
1472 * from the current process.
1473 *
1474 * This difference is important for error handling, when we
1475 * only half set up a mm_struct for a new process and need to restore
1476 * the old one. Because we mmput the new mm_struct before
1477 * restoring the old one. . .
1478 * Eric Biederman 10 January 1998
1479 */
1480static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1481{
1482 uprobe_free_utask(tsk);
1483
1484 /* Get rid of any cached register state */
1485 deactivate_mm(tsk, mm);
1486
1487 /*
1488 * Signal userspace if we're not exiting with a core dump
1489 * because we want to leave the value intact for debugging
1490 * purposes.
1491 */
1492 if (tsk->clear_child_tid) {
1493 if (atomic_read(&mm->mm_users) > 1) {
1494 /*
1495 * We don't check the error code - if userspace has
1496 * not set up a proper pointer then tough luck.
1497 */
1498 put_user(0, tsk->clear_child_tid);
1499 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1500 1, NULL, NULL, 0, 0);
1501 }
1502 tsk->clear_child_tid = NULL;
1503 }
1504
1505 /*
1506 * All done, finally we can wake up parent and return this mm to him.
1507 * Also kthread_stop() uses this completion for synchronization.
1508 */
1509 if (tsk->vfork_done)
1510 complete_vfork_done(tsk);
1511}
1512
1513void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1514{
1515 futex_exit_release(tsk);
1516 mm_release(tsk, mm);
1517}
1518
1519void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1520{
1521 futex_exec_release(tsk);
1522 mm_release(tsk, mm);
1523}
1524
1525/**
1526 * dup_mm() - duplicates an existing mm structure
1527 * @tsk: the task_struct with which the new mm will be associated.
1528 * @oldmm: the mm to duplicate.
1529 *
1530 * Allocates a new mm structure and duplicates the provided @oldmm structure
1531 * content into it.
1532 *
1533 * Return: the duplicated mm or NULL on failure.
1534 */
1535static struct mm_struct *dup_mm(struct task_struct *tsk,
1536 struct mm_struct *oldmm)
1537{
1538 struct mm_struct *mm;
1539 int err;
1540
1541 mm = allocate_mm();
1542 if (!mm)
1543 goto fail_nomem;
1544
1545 memcpy(mm, oldmm, sizeof(*mm));
1546
1547 if (!mm_init(mm, tsk, mm->user_ns))
1548 goto fail_nomem;
1549
1550 err = dup_mmap(mm, oldmm);
1551 if (err)
1552 goto free_pt;
1553
1554 mm->hiwater_rss = get_mm_rss(mm);
1555 mm->hiwater_vm = mm->total_vm;
1556
1557 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1558 goto free_pt;
1559
1560 return mm;
1561
1562free_pt:
1563 /* don't put binfmt in mmput, we haven't got module yet */
1564 mm->binfmt = NULL;
1565 mm_init_owner(mm, NULL);
1566 mmput(mm);
1567
1568fail_nomem:
1569 return NULL;
1570}
1571
1572static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1573{
1574 struct mm_struct *mm, *oldmm;
1575
1576 tsk->min_flt = tsk->maj_flt = 0;
1577 tsk->nvcsw = tsk->nivcsw = 0;
1578#ifdef CONFIG_DETECT_HUNG_TASK
1579 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1580 tsk->last_switch_time = 0;
1581#endif
1582
1583 tsk->mm = NULL;
1584 tsk->active_mm = NULL;
1585
1586 /*
1587 * Are we cloning a kernel thread?
1588 *
1589 * We need to steal a active VM for that..
1590 */
1591 oldmm = current->mm;
1592 if (!oldmm)
1593 return 0;
1594
1595 if (clone_flags & CLONE_VM) {
1596 mmget(oldmm);
1597 mm = oldmm;
1598 } else {
1599 mm = dup_mm(tsk, current->mm);
1600 if (!mm)
1601 return -ENOMEM;
1602 }
1603
1604 tsk->mm = mm;
1605 tsk->active_mm = mm;
1606 sched_mm_cid_fork(tsk);
1607 return 0;
1608}
1609
1610static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1611{
1612 struct fs_struct *fs = current->fs;
1613 if (clone_flags & CLONE_FS) {
1614 /* tsk->fs is already what we want */
1615 spin_lock(&fs->lock);
1616 if (fs->in_exec) {
1617 spin_unlock(&fs->lock);
1618 return -EAGAIN;
1619 }
1620 fs->users++;
1621 spin_unlock(&fs->lock);
1622 return 0;
1623 }
1624 tsk->fs = copy_fs_struct(fs);
1625 if (!tsk->fs)
1626 return -ENOMEM;
1627 return 0;
1628}
1629
1630static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1631{
1632 struct files_struct *oldf, *newf;
1633 int error = 0;
1634
1635 /*
1636 * A background process may not have any files ...
1637 */
1638 oldf = current->files;
1639 if (!oldf)
1640 goto out;
1641
1642 if (clone_flags & CLONE_FILES) {
1643 atomic_inc(&oldf->count);
1644 goto out;
1645 }
1646
1647 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1648 if (!newf)
1649 goto out;
1650
1651 tsk->files = newf;
1652 error = 0;
1653out:
1654 return error;
1655}
1656
1657static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1658{
1659 struct sighand_struct *sig;
1660
1661 if (clone_flags & CLONE_SIGHAND) {
1662 refcount_inc(&current->sighand->count);
1663 return 0;
1664 }
1665 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1666 RCU_INIT_POINTER(tsk->sighand, sig);
1667 if (!sig)
1668 return -ENOMEM;
1669
1670 refcount_set(&sig->count, 1);
1671 spin_lock_irq(&current->sighand->siglock);
1672 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1673 spin_unlock_irq(&current->sighand->siglock);
1674
1675 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1676 if (clone_flags & CLONE_CLEAR_SIGHAND)
1677 flush_signal_handlers(tsk, 0);
1678
1679 return 0;
1680}
1681
1682void __cleanup_sighand(struct sighand_struct *sighand)
1683{
1684 if (refcount_dec_and_test(&sighand->count)) {
1685 signalfd_cleanup(sighand);
1686 /*
1687 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1688 * without an RCU grace period, see __lock_task_sighand().
1689 */
1690 kmem_cache_free(sighand_cachep, sighand);
1691 }
1692}
1693
1694/*
1695 * Initialize POSIX timer handling for a thread group.
1696 */
1697static void posix_cpu_timers_init_group(struct signal_struct *sig)
1698{
1699 struct posix_cputimers *pct = &sig->posix_cputimers;
1700 unsigned long cpu_limit;
1701
1702 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1703 posix_cputimers_group_init(pct, cpu_limit);
1704}
1705
1706static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1707{
1708 struct signal_struct *sig;
1709
1710 if (clone_flags & CLONE_THREAD)
1711 return 0;
1712
1713 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1714 tsk->signal = sig;
1715 if (!sig)
1716 return -ENOMEM;
1717
1718 sig->nr_threads = 1;
1719 sig->quick_threads = 1;
1720 atomic_set(&sig->live, 1);
1721 refcount_set(&sig->sigcnt, 1);
1722
1723 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1724 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1725 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1726
1727 init_waitqueue_head(&sig->wait_chldexit);
1728 sig->curr_target = tsk;
1729 init_sigpending(&sig->shared_pending);
1730 INIT_HLIST_HEAD(&sig->multiprocess);
1731 seqlock_init(&sig->stats_lock);
1732 prev_cputime_init(&sig->prev_cputime);
1733
1734#ifdef CONFIG_POSIX_TIMERS
1735 INIT_LIST_HEAD(&sig->posix_timers);
1736 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1737 sig->real_timer.function = it_real_fn;
1738#endif
1739
1740 task_lock(current->group_leader);
1741 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1742 task_unlock(current->group_leader);
1743
1744 posix_cpu_timers_init_group(sig);
1745
1746 tty_audit_fork(sig);
1747 sched_autogroup_fork(sig);
1748
1749 sig->oom_score_adj = current->signal->oom_score_adj;
1750 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1751
1752 mutex_init(&sig->cred_guard_mutex);
1753 init_rwsem(&sig->exec_update_lock);
1754
1755 return 0;
1756}
1757
1758static void copy_seccomp(struct task_struct *p)
1759{
1760#ifdef CONFIG_SECCOMP
1761 /*
1762 * Must be called with sighand->lock held, which is common to
1763 * all threads in the group. Holding cred_guard_mutex is not
1764 * needed because this new task is not yet running and cannot
1765 * be racing exec.
1766 */
1767 assert_spin_locked(&current->sighand->siglock);
1768
1769 /* Ref-count the new filter user, and assign it. */
1770 get_seccomp_filter(current);
1771 p->seccomp = current->seccomp;
1772
1773 /*
1774 * Explicitly enable no_new_privs here in case it got set
1775 * between the task_struct being duplicated and holding the
1776 * sighand lock. The seccomp state and nnp must be in sync.
1777 */
1778 if (task_no_new_privs(current))
1779 task_set_no_new_privs(p);
1780
1781 /*
1782 * If the parent gained a seccomp mode after copying thread
1783 * flags and between before we held the sighand lock, we have
1784 * to manually enable the seccomp thread flag here.
1785 */
1786 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1787 set_task_syscall_work(p, SECCOMP);
1788#endif
1789}
1790
1791SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1792{
1793 current->clear_child_tid = tidptr;
1794
1795 return task_pid_vnr(current);
1796}
1797
1798static void rt_mutex_init_task(struct task_struct *p)
1799{
1800 raw_spin_lock_init(&p->pi_lock);
1801#ifdef CONFIG_RT_MUTEXES
1802 p->pi_waiters = RB_ROOT_CACHED;
1803 p->pi_top_task = NULL;
1804 p->pi_blocked_on = NULL;
1805#endif
1806}
1807
1808static inline void init_task_pid_links(struct task_struct *task)
1809{
1810 enum pid_type type;
1811
1812 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1813 INIT_HLIST_NODE(&task->pid_links[type]);
1814}
1815
1816static inline void
1817init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1818{
1819 if (type == PIDTYPE_PID)
1820 task->thread_pid = pid;
1821 else
1822 task->signal->pids[type] = pid;
1823}
1824
1825static inline void rcu_copy_process(struct task_struct *p)
1826{
1827#ifdef CONFIG_PREEMPT_RCU
1828 p->rcu_read_lock_nesting = 0;
1829 p->rcu_read_unlock_special.s = 0;
1830 p->rcu_blocked_node = NULL;
1831 INIT_LIST_HEAD(&p->rcu_node_entry);
1832#endif /* #ifdef CONFIG_PREEMPT_RCU */
1833#ifdef CONFIG_TASKS_RCU
1834 p->rcu_tasks_holdout = false;
1835 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1836 p->rcu_tasks_idle_cpu = -1;
1837#endif /* #ifdef CONFIG_TASKS_RCU */
1838#ifdef CONFIG_TASKS_TRACE_RCU
1839 p->trc_reader_nesting = 0;
1840 p->trc_reader_special.s = 0;
1841 INIT_LIST_HEAD(&p->trc_holdout_list);
1842 INIT_LIST_HEAD(&p->trc_blkd_node);
1843#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1844}
1845
1846struct pid *pidfd_pid(const struct file *file)
1847{
1848 if (file->f_op == &pidfd_fops)
1849 return file->private_data;
1850
1851 return ERR_PTR(-EBADF);
1852}
1853
1854static int pidfd_release(struct inode *inode, struct file *file)
1855{
1856 struct pid *pid = file->private_data;
1857
1858 file->private_data = NULL;
1859 put_pid(pid);
1860 return 0;
1861}
1862
1863#ifdef CONFIG_PROC_FS
1864/**
1865 * pidfd_show_fdinfo - print information about a pidfd
1866 * @m: proc fdinfo file
1867 * @f: file referencing a pidfd
1868 *
1869 * Pid:
1870 * This function will print the pid that a given pidfd refers to in the
1871 * pid namespace of the procfs instance.
1872 * If the pid namespace of the process is not a descendant of the pid
1873 * namespace of the procfs instance 0 will be shown as its pid. This is
1874 * similar to calling getppid() on a process whose parent is outside of
1875 * its pid namespace.
1876 *
1877 * NSpid:
1878 * If pid namespaces are supported then this function will also print
1879 * the pid of a given pidfd refers to for all descendant pid namespaces
1880 * starting from the current pid namespace of the instance, i.e. the
1881 * Pid field and the first entry in the NSpid field will be identical.
1882 * If the pid namespace of the process is not a descendant of the pid
1883 * namespace of the procfs instance 0 will be shown as its first NSpid
1884 * entry and no others will be shown.
1885 * Note that this differs from the Pid and NSpid fields in
1886 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1887 * the pid namespace of the procfs instance. The difference becomes
1888 * obvious when sending around a pidfd between pid namespaces from a
1889 * different branch of the tree, i.e. where no ancestral relation is
1890 * present between the pid namespaces:
1891 * - create two new pid namespaces ns1 and ns2 in the initial pid
1892 * namespace (also take care to create new mount namespaces in the
1893 * new pid namespace and mount procfs)
1894 * - create a process with a pidfd in ns1
1895 * - send pidfd from ns1 to ns2
1896 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1897 * have exactly one entry, which is 0
1898 */
1899static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1900{
1901 struct pid *pid = f->private_data;
1902 struct pid_namespace *ns;
1903 pid_t nr = -1;
1904
1905 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1906 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1907 nr = pid_nr_ns(pid, ns);
1908 }
1909
1910 seq_put_decimal_ll(m, "Pid:\t", nr);
1911
1912#ifdef CONFIG_PID_NS
1913 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1914 if (nr > 0) {
1915 int i;
1916
1917 /* If nr is non-zero it means that 'pid' is valid and that
1918 * ns, i.e. the pid namespace associated with the procfs
1919 * instance, is in the pid namespace hierarchy of pid.
1920 * Start at one below the already printed level.
1921 */
1922 for (i = ns->level + 1; i <= pid->level; i++)
1923 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1924 }
1925#endif
1926 seq_putc(m, '\n');
1927}
1928#endif
1929
1930/*
1931 * Poll support for process exit notification.
1932 */
1933static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1934{
1935 struct pid *pid = file->private_data;
1936 __poll_t poll_flags = 0;
1937
1938 poll_wait(file, &pid->wait_pidfd, pts);
1939
1940 /*
1941 * Inform pollers only when the whole thread group exits.
1942 * If the thread group leader exits before all other threads in the
1943 * group, then poll(2) should block, similar to the wait(2) family.
1944 */
1945 if (thread_group_exited(pid))
1946 poll_flags = EPOLLIN | EPOLLRDNORM;
1947
1948 return poll_flags;
1949}
1950
1951const struct file_operations pidfd_fops = {
1952 .release = pidfd_release,
1953 .poll = pidfd_poll,
1954#ifdef CONFIG_PROC_FS
1955 .show_fdinfo = pidfd_show_fdinfo,
1956#endif
1957};
1958
1959static void __delayed_free_task(struct rcu_head *rhp)
1960{
1961 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1962
1963 free_task(tsk);
1964}
1965
1966static __always_inline void delayed_free_task(struct task_struct *tsk)
1967{
1968 if (IS_ENABLED(CONFIG_MEMCG))
1969 call_rcu(&tsk->rcu, __delayed_free_task);
1970 else
1971 free_task(tsk);
1972}
1973
1974static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1975{
1976 /* Skip if kernel thread */
1977 if (!tsk->mm)
1978 return;
1979
1980 /* Skip if spawning a thread or using vfork */
1981 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1982 return;
1983
1984 /* We need to synchronize with __set_oom_adj */
1985 mutex_lock(&oom_adj_mutex);
1986 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1987 /* Update the values in case they were changed after copy_signal */
1988 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1989 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1990 mutex_unlock(&oom_adj_mutex);
1991}
1992
1993#ifdef CONFIG_RV
1994static void rv_task_fork(struct task_struct *p)
1995{
1996 int i;
1997
1998 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1999 p->rv[i].da_mon.monitoring = false;
2000}
2001#else
2002#define rv_task_fork(p) do {} while (0)
2003#endif
2004
2005/*
2006 * This creates a new process as a copy of the old one,
2007 * but does not actually start it yet.
2008 *
2009 * It copies the registers, and all the appropriate
2010 * parts of the process environment (as per the clone
2011 * flags). The actual kick-off is left to the caller.
2012 */
2013static __latent_entropy struct task_struct *copy_process(
2014 struct pid *pid,
2015 int trace,
2016 int node,
2017 struct kernel_clone_args *args)
2018{
2019 int pidfd = -1, retval;
2020 struct task_struct *p;
2021 struct multiprocess_signals delayed;
2022 struct file *pidfile = NULL;
2023 const u64 clone_flags = args->flags;
2024 struct nsproxy *nsp = current->nsproxy;
2025
2026 /*
2027 * Don't allow sharing the root directory with processes in a different
2028 * namespace
2029 */
2030 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2031 return ERR_PTR(-EINVAL);
2032
2033 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2034 return ERR_PTR(-EINVAL);
2035
2036 /*
2037 * Thread groups must share signals as well, and detached threads
2038 * can only be started up within the thread group.
2039 */
2040 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2041 return ERR_PTR(-EINVAL);
2042
2043 /*
2044 * Shared signal handlers imply shared VM. By way of the above,
2045 * thread groups also imply shared VM. Blocking this case allows
2046 * for various simplifications in other code.
2047 */
2048 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2049 return ERR_PTR(-EINVAL);
2050
2051 /*
2052 * Siblings of global init remain as zombies on exit since they are
2053 * not reaped by their parent (swapper). To solve this and to avoid
2054 * multi-rooted process trees, prevent global and container-inits
2055 * from creating siblings.
2056 */
2057 if ((clone_flags & CLONE_PARENT) &&
2058 current->signal->flags & SIGNAL_UNKILLABLE)
2059 return ERR_PTR(-EINVAL);
2060
2061 /*
2062 * If the new process will be in a different pid or user namespace
2063 * do not allow it to share a thread group with the forking task.
2064 */
2065 if (clone_flags & CLONE_THREAD) {
2066 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2067 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2068 return ERR_PTR(-EINVAL);
2069 }
2070
2071 if (clone_flags & CLONE_PIDFD) {
2072 /*
2073 * - CLONE_DETACHED is blocked so that we can potentially
2074 * reuse it later for CLONE_PIDFD.
2075 * - CLONE_THREAD is blocked until someone really needs it.
2076 */
2077 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2078 return ERR_PTR(-EINVAL);
2079 }
2080
2081 /*
2082 * Force any signals received before this point to be delivered
2083 * before the fork happens. Collect up signals sent to multiple
2084 * processes that happen during the fork and delay them so that
2085 * they appear to happen after the fork.
2086 */
2087 sigemptyset(&delayed.signal);
2088 INIT_HLIST_NODE(&delayed.node);
2089
2090 spin_lock_irq(&current->sighand->siglock);
2091 if (!(clone_flags & CLONE_THREAD))
2092 hlist_add_head(&delayed.node, &current->signal->multiprocess);
2093 recalc_sigpending();
2094 spin_unlock_irq(&current->sighand->siglock);
2095 retval = -ERESTARTNOINTR;
2096 if (task_sigpending(current))
2097 goto fork_out;
2098
2099 retval = -ENOMEM;
2100 p = dup_task_struct(current, node);
2101 if (!p)
2102 goto fork_out;
2103 p->flags &= ~PF_KTHREAD;
2104 if (args->kthread)
2105 p->flags |= PF_KTHREAD;
2106 if (args->io_thread) {
2107 /*
2108 * Mark us an IO worker, and block any signal that isn't
2109 * fatal or STOP
2110 */
2111 p->flags |= PF_IO_WORKER;
2112 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2113 }
2114
2115 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2116 /*
2117 * Clear TID on mm_release()?
2118 */
2119 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2120
2121 ftrace_graph_init_task(p);
2122
2123 rt_mutex_init_task(p);
2124
2125 lockdep_assert_irqs_enabled();
2126#ifdef CONFIG_PROVE_LOCKING
2127 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2128#endif
2129 retval = copy_creds(p, clone_flags);
2130 if (retval < 0)
2131 goto bad_fork_free;
2132
2133 retval = -EAGAIN;
2134 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2135 if (p->real_cred->user != INIT_USER &&
2136 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2137 goto bad_fork_cleanup_count;
2138 }
2139 current->flags &= ~PF_NPROC_EXCEEDED;
2140
2141 /*
2142 * If multiple threads are within copy_process(), then this check
2143 * triggers too late. This doesn't hurt, the check is only there
2144 * to stop root fork bombs.
2145 */
2146 retval = -EAGAIN;
2147 if (data_race(nr_threads >= max_threads))
2148 goto bad_fork_cleanup_count;
2149
2150 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2151 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2152 p->flags |= PF_FORKNOEXEC;
2153 INIT_LIST_HEAD(&p->children);
2154 INIT_LIST_HEAD(&p->sibling);
2155 rcu_copy_process(p);
2156 p->vfork_done = NULL;
2157 spin_lock_init(&p->alloc_lock);
2158
2159 init_sigpending(&p->pending);
2160
2161 p->utime = p->stime = p->gtime = 0;
2162#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2163 p->utimescaled = p->stimescaled = 0;
2164#endif
2165 prev_cputime_init(&p->prev_cputime);
2166
2167#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2168 seqcount_init(&p->vtime.seqcount);
2169 p->vtime.starttime = 0;
2170 p->vtime.state = VTIME_INACTIVE;
2171#endif
2172
2173#ifdef CONFIG_IO_URING
2174 p->io_uring = NULL;
2175#endif
2176
2177#if defined(SPLIT_RSS_COUNTING)
2178 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2179#endif
2180
2181 p->default_timer_slack_ns = current->timer_slack_ns;
2182
2183#ifdef CONFIG_PSI
2184 p->psi_flags = 0;
2185#endif
2186
2187 task_io_accounting_init(&p->ioac);
2188 acct_clear_integrals(p);
2189
2190 posix_cputimers_init(&p->posix_cputimers);
2191
2192 p->io_context = NULL;
2193 audit_set_context(p, NULL);
2194 cgroup_fork(p);
2195 if (args->kthread) {
2196 if (!set_kthread_struct(p))
2197 goto bad_fork_cleanup_delayacct;
2198 }
2199#ifdef CONFIG_NUMA
2200 p->mempolicy = mpol_dup(p->mempolicy);
2201 if (IS_ERR(p->mempolicy)) {
2202 retval = PTR_ERR(p->mempolicy);
2203 p->mempolicy = NULL;
2204 goto bad_fork_cleanup_delayacct;
2205 }
2206#endif
2207#ifdef CONFIG_CPUSETS
2208 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2209 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2210 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2211#endif
2212#ifdef CONFIG_TRACE_IRQFLAGS
2213 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2214 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2215 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2216 p->softirqs_enabled = 1;
2217 p->softirq_context = 0;
2218#endif
2219
2220 p->pagefault_disabled = 0;
2221
2222#ifdef CONFIG_LOCKDEP
2223 lockdep_init_task(p);
2224#endif
2225
2226#ifdef CONFIG_DEBUG_MUTEXES
2227 p->blocked_on = NULL; /* not blocked yet */
2228#endif
2229#ifdef CONFIG_BCACHE
2230 p->sequential_io = 0;
2231 p->sequential_io_avg = 0;
2232#endif
2233#ifdef CONFIG_BPF_SYSCALL
2234 RCU_INIT_POINTER(p->bpf_storage, NULL);
2235 p->bpf_ctx = NULL;
2236#endif
2237
2238 /* Perform scheduler related setup. Assign this task to a CPU. */
2239 retval = sched_fork(clone_flags, p);
2240 if (retval)
2241 goto bad_fork_cleanup_policy;
2242
2243 retval = perf_event_init_task(p, clone_flags);
2244 if (retval)
2245 goto bad_fork_cleanup_policy;
2246 retval = audit_alloc(p);
2247 if (retval)
2248 goto bad_fork_cleanup_perf;
2249 /* copy all the process information */
2250 shm_init_task(p);
2251 retval = security_task_alloc(p, clone_flags);
2252 if (retval)
2253 goto bad_fork_cleanup_audit;
2254 retval = copy_semundo(clone_flags, p);
2255 if (retval)
2256 goto bad_fork_cleanup_security;
2257 retval = copy_files(clone_flags, p);
2258 if (retval)
2259 goto bad_fork_cleanup_semundo;
2260 retval = copy_fs(clone_flags, p);
2261 if (retval)
2262 goto bad_fork_cleanup_files;
2263 retval = copy_sighand(clone_flags, p);
2264 if (retval)
2265 goto bad_fork_cleanup_fs;
2266 retval = copy_signal(clone_flags, p);
2267 if (retval)
2268 goto bad_fork_cleanup_sighand;
2269 retval = copy_mm(clone_flags, p);
2270 if (retval)
2271 goto bad_fork_cleanup_signal;
2272 retval = copy_namespaces(clone_flags, p);
2273 if (retval)
2274 goto bad_fork_cleanup_mm;
2275 retval = copy_io(clone_flags, p);
2276 if (retval)
2277 goto bad_fork_cleanup_namespaces;
2278 retval = copy_thread(p, args);
2279 if (retval)
2280 goto bad_fork_cleanup_io;
2281
2282 stackleak_task_init(p);
2283
2284 if (pid != &init_struct_pid) {
2285 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2286 args->set_tid_size);
2287 if (IS_ERR(pid)) {
2288 retval = PTR_ERR(pid);
2289 goto bad_fork_cleanup_thread;
2290 }
2291 }
2292
2293 /*
2294 * This has to happen after we've potentially unshared the file
2295 * descriptor table (so that the pidfd doesn't leak into the child
2296 * if the fd table isn't shared).
2297 */
2298 if (clone_flags & CLONE_PIDFD) {
2299 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2300 if (retval < 0)
2301 goto bad_fork_free_pid;
2302
2303 pidfd = retval;
2304
2305 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2306 O_RDWR | O_CLOEXEC);
2307 if (IS_ERR(pidfile)) {
2308 put_unused_fd(pidfd);
2309 retval = PTR_ERR(pidfile);
2310 goto bad_fork_free_pid;
2311 }
2312 get_pid(pid); /* held by pidfile now */
2313
2314 retval = put_user(pidfd, args->pidfd);
2315 if (retval)
2316 goto bad_fork_put_pidfd;
2317 }
2318
2319#ifdef CONFIG_BLOCK
2320 p->plug = NULL;
2321#endif
2322 futex_init_task(p);
2323
2324 /*
2325 * sigaltstack should be cleared when sharing the same VM
2326 */
2327 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2328 sas_ss_reset(p);
2329
2330 /*
2331 * Syscall tracing and stepping should be turned off in the
2332 * child regardless of CLONE_PTRACE.
2333 */
2334 user_disable_single_step(p);
2335 clear_task_syscall_work(p, SYSCALL_TRACE);
2336#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2337 clear_task_syscall_work(p, SYSCALL_EMU);
2338#endif
2339 clear_tsk_latency_tracing(p);
2340
2341 /* ok, now we should be set up.. */
2342 p->pid = pid_nr(pid);
2343 if (clone_flags & CLONE_THREAD) {
2344 p->group_leader = current->group_leader;
2345 p->tgid = current->tgid;
2346 } else {
2347 p->group_leader = p;
2348 p->tgid = p->pid;
2349 }
2350
2351 p->nr_dirtied = 0;
2352 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2353 p->dirty_paused_when = 0;
2354
2355 p->pdeath_signal = 0;
2356 INIT_LIST_HEAD(&p->thread_group);
2357 p->task_works = NULL;
2358 clear_posix_cputimers_work(p);
2359
2360#ifdef CONFIG_KRETPROBES
2361 p->kretprobe_instances.first = NULL;
2362#endif
2363#ifdef CONFIG_RETHOOK
2364 p->rethooks.first = NULL;
2365#endif
2366
2367 /*
2368 * Ensure that the cgroup subsystem policies allow the new process to be
2369 * forked. It should be noted that the new process's css_set can be changed
2370 * between here and cgroup_post_fork() if an organisation operation is in
2371 * progress.
2372 */
2373 retval = cgroup_can_fork(p, args);
2374 if (retval)
2375 goto bad_fork_put_pidfd;
2376
2377 /*
2378 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2379 * the new task on the correct runqueue. All this *before* the task
2380 * becomes visible.
2381 *
2382 * This isn't part of ->can_fork() because while the re-cloning is
2383 * cgroup specific, it unconditionally needs to place the task on a
2384 * runqueue.
2385 */
2386 sched_cgroup_fork(p, args);
2387
2388 /*
2389 * From this point on we must avoid any synchronous user-space
2390 * communication until we take the tasklist-lock. In particular, we do
2391 * not want user-space to be able to predict the process start-time by
2392 * stalling fork(2) after we recorded the start_time but before it is
2393 * visible to the system.
2394 */
2395
2396 p->start_time = ktime_get_ns();
2397 p->start_boottime = ktime_get_boottime_ns();
2398
2399 /*
2400 * Make it visible to the rest of the system, but dont wake it up yet.
2401 * Need tasklist lock for parent etc handling!
2402 */
2403 write_lock_irq(&tasklist_lock);
2404
2405 /* CLONE_PARENT re-uses the old parent */
2406 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2407 p->real_parent = current->real_parent;
2408 p->parent_exec_id = current->parent_exec_id;
2409 if (clone_flags & CLONE_THREAD)
2410 p->exit_signal = -1;
2411 else
2412 p->exit_signal = current->group_leader->exit_signal;
2413 } else {
2414 p->real_parent = current;
2415 p->parent_exec_id = current->self_exec_id;
2416 p->exit_signal = args->exit_signal;
2417 }
2418
2419 klp_copy_process(p);
2420
2421 sched_core_fork(p);
2422
2423 spin_lock(&current->sighand->siglock);
2424
2425 rv_task_fork(p);
2426
2427 rseq_fork(p, clone_flags);
2428
2429 /* Don't start children in a dying pid namespace */
2430 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2431 retval = -ENOMEM;
2432 goto bad_fork_cancel_cgroup;
2433 }
2434
2435 /* Let kill terminate clone/fork in the middle */
2436 if (fatal_signal_pending(current)) {
2437 retval = -EINTR;
2438 goto bad_fork_cancel_cgroup;
2439 }
2440
2441 /* No more failure paths after this point. */
2442
2443 /*
2444 * Copy seccomp details explicitly here, in case they were changed
2445 * before holding sighand lock.
2446 */
2447 copy_seccomp(p);
2448
2449 init_task_pid_links(p);
2450 if (likely(p->pid)) {
2451 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2452
2453 init_task_pid(p, PIDTYPE_PID, pid);
2454 if (thread_group_leader(p)) {
2455 init_task_pid(p, PIDTYPE_TGID, pid);
2456 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2457 init_task_pid(p, PIDTYPE_SID, task_session(current));
2458
2459 if (is_child_reaper(pid)) {
2460 ns_of_pid(pid)->child_reaper = p;
2461 p->signal->flags |= SIGNAL_UNKILLABLE;
2462 }
2463 p->signal->shared_pending.signal = delayed.signal;
2464 p->signal->tty = tty_kref_get(current->signal->tty);
2465 /*
2466 * Inherit has_child_subreaper flag under the same
2467 * tasklist_lock with adding child to the process tree
2468 * for propagate_has_child_subreaper optimization.
2469 */
2470 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2471 p->real_parent->signal->is_child_subreaper;
2472 list_add_tail(&p->sibling, &p->real_parent->children);
2473 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2474 attach_pid(p, PIDTYPE_TGID);
2475 attach_pid(p, PIDTYPE_PGID);
2476 attach_pid(p, PIDTYPE_SID);
2477 __this_cpu_inc(process_counts);
2478 } else {
2479 current->signal->nr_threads++;
2480 current->signal->quick_threads++;
2481 atomic_inc(&current->signal->live);
2482 refcount_inc(&current->signal->sigcnt);
2483 task_join_group_stop(p);
2484 list_add_tail_rcu(&p->thread_group,
2485 &p->group_leader->thread_group);
2486 list_add_tail_rcu(&p->thread_node,
2487 &p->signal->thread_head);
2488 }
2489 attach_pid(p, PIDTYPE_PID);
2490 nr_threads++;
2491 }
2492 total_forks++;
2493 hlist_del_init(&delayed.node);
2494 spin_unlock(&current->sighand->siglock);
2495 syscall_tracepoint_update(p);
2496 write_unlock_irq(&tasklist_lock);
2497
2498 if (pidfile)
2499 fd_install(pidfd, pidfile);
2500
2501 proc_fork_connector(p);
2502 sched_post_fork(p);
2503 cgroup_post_fork(p, args);
2504 perf_event_fork(p);
2505
2506 trace_task_newtask(p, clone_flags);
2507 uprobe_copy_process(p, clone_flags);
2508
2509 copy_oom_score_adj(clone_flags, p);
2510
2511 return p;
2512
2513bad_fork_cancel_cgroup:
2514 sched_core_free(p);
2515 spin_unlock(&current->sighand->siglock);
2516 write_unlock_irq(&tasklist_lock);
2517 cgroup_cancel_fork(p, args);
2518bad_fork_put_pidfd:
2519 if (clone_flags & CLONE_PIDFD) {
2520 fput(pidfile);
2521 put_unused_fd(pidfd);
2522 }
2523bad_fork_free_pid:
2524 if (pid != &init_struct_pid)
2525 free_pid(pid);
2526bad_fork_cleanup_thread:
2527 exit_thread(p);
2528bad_fork_cleanup_io:
2529 if (p->io_context)
2530 exit_io_context(p);
2531bad_fork_cleanup_namespaces:
2532 exit_task_namespaces(p);
2533bad_fork_cleanup_mm:
2534 if (p->mm) {
2535 mm_clear_owner(p->mm, p);
2536 mmput(p->mm);
2537 }
2538bad_fork_cleanup_signal:
2539 if (!(clone_flags & CLONE_THREAD))
2540 free_signal_struct(p->signal);
2541bad_fork_cleanup_sighand:
2542 __cleanup_sighand(p->sighand);
2543bad_fork_cleanup_fs:
2544 exit_fs(p); /* blocking */
2545bad_fork_cleanup_files:
2546 exit_files(p); /* blocking */
2547bad_fork_cleanup_semundo:
2548 exit_sem(p);
2549bad_fork_cleanup_security:
2550 security_task_free(p);
2551bad_fork_cleanup_audit:
2552 audit_free(p);
2553bad_fork_cleanup_perf:
2554 perf_event_free_task(p);
2555bad_fork_cleanup_policy:
2556 lockdep_free_task(p);
2557#ifdef CONFIG_NUMA
2558 mpol_put(p->mempolicy);
2559#endif
2560bad_fork_cleanup_delayacct:
2561 delayacct_tsk_free(p);
2562bad_fork_cleanup_count:
2563 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2564 exit_creds(p);
2565bad_fork_free:
2566 WRITE_ONCE(p->__state, TASK_DEAD);
2567 exit_task_stack_account(p);
2568 put_task_stack(p);
2569 delayed_free_task(p);
2570fork_out:
2571 spin_lock_irq(&current->sighand->siglock);
2572 hlist_del_init(&delayed.node);
2573 spin_unlock_irq(&current->sighand->siglock);
2574 return ERR_PTR(retval);
2575}
2576
2577static inline void init_idle_pids(struct task_struct *idle)
2578{
2579 enum pid_type type;
2580
2581 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2582 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2583 init_task_pid(idle, type, &init_struct_pid);
2584 }
2585}
2586
2587static int idle_dummy(void *dummy)
2588{
2589 /* This function is never called */
2590 return 0;
2591}
2592
2593struct task_struct * __init fork_idle(int cpu)
2594{
2595 struct task_struct *task;
2596 struct kernel_clone_args args = {
2597 .flags = CLONE_VM,
2598 .fn = &idle_dummy,
2599 .fn_arg = NULL,
2600 .kthread = 1,
2601 .idle = 1,
2602 };
2603
2604 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2605 if (!IS_ERR(task)) {
2606 init_idle_pids(task);
2607 init_idle(task, cpu);
2608 }
2609
2610 return task;
2611}
2612
2613/*
2614 * This is like kernel_clone(), but shaved down and tailored to just
2615 * creating io_uring workers. It returns a created task, or an error pointer.
2616 * The returned task is inactive, and the caller must fire it up through
2617 * wake_up_new_task(p). All signals are blocked in the created task.
2618 */
2619struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2620{
2621 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2622 CLONE_IO;
2623 struct kernel_clone_args args = {
2624 .flags = ((lower_32_bits(flags) | CLONE_VM |
2625 CLONE_UNTRACED) & ~CSIGNAL),
2626 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2627 .fn = fn,
2628 .fn_arg = arg,
2629 .io_thread = 1,
2630 };
2631
2632 return copy_process(NULL, 0, node, &args);
2633}
2634
2635/*
2636 * Ok, this is the main fork-routine.
2637 *
2638 * It copies the process, and if successful kick-starts
2639 * it and waits for it to finish using the VM if required.
2640 *
2641 * args->exit_signal is expected to be checked for sanity by the caller.
2642 */
2643pid_t kernel_clone(struct kernel_clone_args *args)
2644{
2645 u64 clone_flags = args->flags;
2646 struct completion vfork;
2647 struct pid *pid;
2648 struct task_struct *p;
2649 int trace = 0;
2650 pid_t nr;
2651
2652 /*
2653 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2654 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2655 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2656 * field in struct clone_args and it still doesn't make sense to have
2657 * them both point at the same memory location. Performing this check
2658 * here has the advantage that we don't need to have a separate helper
2659 * to check for legacy clone().
2660 */
2661 if ((args->flags & CLONE_PIDFD) &&
2662 (args->flags & CLONE_PARENT_SETTID) &&
2663 (args->pidfd == args->parent_tid))
2664 return -EINVAL;
2665
2666 /*
2667 * Determine whether and which event to report to ptracer. When
2668 * called from kernel_thread or CLONE_UNTRACED is explicitly
2669 * requested, no event is reported; otherwise, report if the event
2670 * for the type of forking is enabled.
2671 */
2672 if (!(clone_flags & CLONE_UNTRACED)) {
2673 if (clone_flags & CLONE_VFORK)
2674 trace = PTRACE_EVENT_VFORK;
2675 else if (args->exit_signal != SIGCHLD)
2676 trace = PTRACE_EVENT_CLONE;
2677 else
2678 trace = PTRACE_EVENT_FORK;
2679
2680 if (likely(!ptrace_event_enabled(current, trace)))
2681 trace = 0;
2682 }
2683
2684 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2685 add_latent_entropy();
2686
2687 if (IS_ERR(p))
2688 return PTR_ERR(p);
2689
2690 /*
2691 * Do this prior waking up the new thread - the thread pointer
2692 * might get invalid after that point, if the thread exits quickly.
2693 */
2694 trace_sched_process_fork(current, p);
2695
2696 pid = get_task_pid(p, PIDTYPE_PID);
2697 nr = pid_vnr(pid);
2698
2699 if (clone_flags & CLONE_PARENT_SETTID)
2700 put_user(nr, args->parent_tid);
2701
2702 if (clone_flags & CLONE_VFORK) {
2703 p->vfork_done = &vfork;
2704 init_completion(&vfork);
2705 get_task_struct(p);
2706 }
2707
2708 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2709 /* lock the task to synchronize with memcg migration */
2710 task_lock(p);
2711 lru_gen_add_mm(p->mm);
2712 task_unlock(p);
2713 }
2714
2715 wake_up_new_task(p);
2716
2717 /* forking complete and child started to run, tell ptracer */
2718 if (unlikely(trace))
2719 ptrace_event_pid(trace, pid);
2720
2721 if (clone_flags & CLONE_VFORK) {
2722 if (!wait_for_vfork_done(p, &vfork))
2723 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2724 }
2725
2726 put_pid(pid);
2727 return nr;
2728}
2729
2730/*
2731 * Create a kernel thread.
2732 */
2733pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2734{
2735 struct kernel_clone_args args = {
2736 .flags = ((lower_32_bits(flags) | CLONE_VM |
2737 CLONE_UNTRACED) & ~CSIGNAL),
2738 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2739 .fn = fn,
2740 .fn_arg = arg,
2741 .kthread = 1,
2742 };
2743
2744 return kernel_clone(&args);
2745}
2746
2747/*
2748 * Create a user mode thread.
2749 */
2750pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2751{
2752 struct kernel_clone_args args = {
2753 .flags = ((lower_32_bits(flags) | CLONE_VM |
2754 CLONE_UNTRACED) & ~CSIGNAL),
2755 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2756 .fn = fn,
2757 .fn_arg = arg,
2758 };
2759
2760 return kernel_clone(&args);
2761}
2762
2763#ifdef __ARCH_WANT_SYS_FORK
2764SYSCALL_DEFINE0(fork)
2765{
2766#ifdef CONFIG_MMU
2767 struct kernel_clone_args args = {
2768 .exit_signal = SIGCHLD,
2769 };
2770
2771 return kernel_clone(&args);
2772#else
2773 /* can not support in nommu mode */
2774 return -EINVAL;
2775#endif
2776}
2777#endif
2778
2779#ifdef __ARCH_WANT_SYS_VFORK
2780SYSCALL_DEFINE0(vfork)
2781{
2782 struct kernel_clone_args args = {
2783 .flags = CLONE_VFORK | CLONE_VM,
2784 .exit_signal = SIGCHLD,
2785 };
2786
2787 return kernel_clone(&args);
2788}
2789#endif
2790
2791#ifdef __ARCH_WANT_SYS_CLONE
2792#ifdef CONFIG_CLONE_BACKWARDS
2793SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2794 int __user *, parent_tidptr,
2795 unsigned long, tls,
2796 int __user *, child_tidptr)
2797#elif defined(CONFIG_CLONE_BACKWARDS2)
2798SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2799 int __user *, parent_tidptr,
2800 int __user *, child_tidptr,
2801 unsigned long, tls)
2802#elif defined(CONFIG_CLONE_BACKWARDS3)
2803SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2804 int, stack_size,
2805 int __user *, parent_tidptr,
2806 int __user *, child_tidptr,
2807 unsigned long, tls)
2808#else
2809SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2810 int __user *, parent_tidptr,
2811 int __user *, child_tidptr,
2812 unsigned long, tls)
2813#endif
2814{
2815 struct kernel_clone_args args = {
2816 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2817 .pidfd = parent_tidptr,
2818 .child_tid = child_tidptr,
2819 .parent_tid = parent_tidptr,
2820 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2821 .stack = newsp,
2822 .tls = tls,
2823 };
2824
2825 return kernel_clone(&args);
2826}
2827#endif
2828
2829#ifdef __ARCH_WANT_SYS_CLONE3
2830
2831noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2832 struct clone_args __user *uargs,
2833 size_t usize)
2834{
2835 int err;
2836 struct clone_args args;
2837 pid_t *kset_tid = kargs->set_tid;
2838
2839 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2840 CLONE_ARGS_SIZE_VER0);
2841 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2842 CLONE_ARGS_SIZE_VER1);
2843 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2844 CLONE_ARGS_SIZE_VER2);
2845 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2846
2847 if (unlikely(usize > PAGE_SIZE))
2848 return -E2BIG;
2849 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2850 return -EINVAL;
2851
2852 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2853 if (err)
2854 return err;
2855
2856 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2857 return -EINVAL;
2858
2859 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2860 return -EINVAL;
2861
2862 if (unlikely(args.set_tid && args.set_tid_size == 0))
2863 return -EINVAL;
2864
2865 /*
2866 * Verify that higher 32bits of exit_signal are unset and that
2867 * it is a valid signal
2868 */
2869 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2870 !valid_signal(args.exit_signal)))
2871 return -EINVAL;
2872
2873 if ((args.flags & CLONE_INTO_CGROUP) &&
2874 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2875 return -EINVAL;
2876
2877 *kargs = (struct kernel_clone_args){
2878 .flags = args.flags,
2879 .pidfd = u64_to_user_ptr(args.pidfd),
2880 .child_tid = u64_to_user_ptr(args.child_tid),
2881 .parent_tid = u64_to_user_ptr(args.parent_tid),
2882 .exit_signal = args.exit_signal,
2883 .stack = args.stack,
2884 .stack_size = args.stack_size,
2885 .tls = args.tls,
2886 .set_tid_size = args.set_tid_size,
2887 .cgroup = args.cgroup,
2888 };
2889
2890 if (args.set_tid &&
2891 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2892 (kargs->set_tid_size * sizeof(pid_t))))
2893 return -EFAULT;
2894
2895 kargs->set_tid = kset_tid;
2896
2897 return 0;
2898}
2899
2900/**
2901 * clone3_stack_valid - check and prepare stack
2902 * @kargs: kernel clone args
2903 *
2904 * Verify that the stack arguments userspace gave us are sane.
2905 * In addition, set the stack direction for userspace since it's easy for us to
2906 * determine.
2907 */
2908static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2909{
2910 if (kargs->stack == 0) {
2911 if (kargs->stack_size > 0)
2912 return false;
2913 } else {
2914 if (kargs->stack_size == 0)
2915 return false;
2916
2917 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2918 return false;
2919
2920#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2921 kargs->stack += kargs->stack_size;
2922#endif
2923 }
2924
2925 return true;
2926}
2927
2928static bool clone3_args_valid(struct kernel_clone_args *kargs)
2929{
2930 /* Verify that no unknown flags are passed along. */
2931 if (kargs->flags &
2932 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2933 return false;
2934
2935 /*
2936 * - make the CLONE_DETACHED bit reusable for clone3
2937 * - make the CSIGNAL bits reusable for clone3
2938 */
2939 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
2940 return false;
2941
2942 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2943 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2944 return false;
2945
2946 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2947 kargs->exit_signal)
2948 return false;
2949
2950 if (!clone3_stack_valid(kargs))
2951 return false;
2952
2953 return true;
2954}
2955
2956/**
2957 * clone3 - create a new process with specific properties
2958 * @uargs: argument structure
2959 * @size: size of @uargs
2960 *
2961 * clone3() is the extensible successor to clone()/clone2().
2962 * It takes a struct as argument that is versioned by its size.
2963 *
2964 * Return: On success, a positive PID for the child process.
2965 * On error, a negative errno number.
2966 */
2967SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2968{
2969 int err;
2970
2971 struct kernel_clone_args kargs;
2972 pid_t set_tid[MAX_PID_NS_LEVEL];
2973
2974 kargs.set_tid = set_tid;
2975
2976 err = copy_clone_args_from_user(&kargs, uargs, size);
2977 if (err)
2978 return err;
2979
2980 if (!clone3_args_valid(&kargs))
2981 return -EINVAL;
2982
2983 return kernel_clone(&kargs);
2984}
2985#endif
2986
2987void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2988{
2989 struct task_struct *leader, *parent, *child;
2990 int res;
2991
2992 read_lock(&tasklist_lock);
2993 leader = top = top->group_leader;
2994down:
2995 for_each_thread(leader, parent) {
2996 list_for_each_entry(child, &parent->children, sibling) {
2997 res = visitor(child, data);
2998 if (res) {
2999 if (res < 0)
3000 goto out;
3001 leader = child;
3002 goto down;
3003 }
3004up:
3005 ;
3006 }
3007 }
3008
3009 if (leader != top) {
3010 child = leader;
3011 parent = child->real_parent;
3012 leader = parent->group_leader;
3013 goto up;
3014 }
3015out:
3016 read_unlock(&tasklist_lock);
3017}
3018
3019#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3020#define ARCH_MIN_MMSTRUCT_ALIGN 0
3021#endif
3022
3023static void sighand_ctor(void *data)
3024{
3025 struct sighand_struct *sighand = data;
3026
3027 spin_lock_init(&sighand->siglock);
3028 init_waitqueue_head(&sighand->signalfd_wqh);
3029}
3030
3031void __init mm_cache_init(void)
3032{
3033 unsigned int mm_size;
3034
3035 /*
3036 * The mm_cpumask is located at the end of mm_struct, and is
3037 * dynamically sized based on the maximum CPU number this system
3038 * can have, taking hotplug into account (nr_cpu_ids).
3039 */
3040 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3041
3042 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3043 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3044 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3045 offsetof(struct mm_struct, saved_auxv),
3046 sizeof_field(struct mm_struct, saved_auxv),
3047 NULL);
3048}
3049
3050void __init proc_caches_init(void)
3051{
3052 sighand_cachep = kmem_cache_create("sighand_cache",
3053 sizeof(struct sighand_struct), 0,
3054 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3055 SLAB_ACCOUNT, sighand_ctor);
3056 signal_cachep = kmem_cache_create("signal_cache",
3057 sizeof(struct signal_struct), 0,
3058 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3059 NULL);
3060 files_cachep = kmem_cache_create("files_cache",
3061 sizeof(struct files_struct), 0,
3062 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3063 NULL);
3064 fs_cachep = kmem_cache_create("fs_cache",
3065 sizeof(struct fs_struct), 0,
3066 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3067 NULL);
3068
3069 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3070 mmap_init();
3071 nsproxy_cache_init();
3072}
3073
3074/*
3075 * Check constraints on flags passed to the unshare system call.
3076 */
3077static int check_unshare_flags(unsigned long unshare_flags)
3078{
3079 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3080 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3081 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3082 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3083 CLONE_NEWTIME))
3084 return -EINVAL;
3085 /*
3086 * Not implemented, but pretend it works if there is nothing
3087 * to unshare. Note that unsharing the address space or the
3088 * signal handlers also need to unshare the signal queues (aka
3089 * CLONE_THREAD).
3090 */
3091 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3092 if (!thread_group_empty(current))
3093 return -EINVAL;
3094 }
3095 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3096 if (refcount_read(&current->sighand->count) > 1)
3097 return -EINVAL;
3098 }
3099 if (unshare_flags & CLONE_VM) {
3100 if (!current_is_single_threaded())
3101 return -EINVAL;
3102 }
3103
3104 return 0;
3105}
3106
3107/*
3108 * Unshare the filesystem structure if it is being shared
3109 */
3110static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3111{
3112 struct fs_struct *fs = current->fs;
3113
3114 if (!(unshare_flags & CLONE_FS) || !fs)
3115 return 0;
3116
3117 /* don't need lock here; in the worst case we'll do useless copy */
3118 if (fs->users == 1)
3119 return 0;
3120
3121 *new_fsp = copy_fs_struct(fs);
3122 if (!*new_fsp)
3123 return -ENOMEM;
3124
3125 return 0;
3126}
3127
3128/*
3129 * Unshare file descriptor table if it is being shared
3130 */
3131int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3132 struct files_struct **new_fdp)
3133{
3134 struct files_struct *fd = current->files;
3135 int error = 0;
3136
3137 if ((unshare_flags & CLONE_FILES) &&
3138 (fd && atomic_read(&fd->count) > 1)) {
3139 *new_fdp = dup_fd(fd, max_fds, &error);
3140 if (!*new_fdp)
3141 return error;
3142 }
3143
3144 return 0;
3145}
3146
3147/*
3148 * unshare allows a process to 'unshare' part of the process
3149 * context which was originally shared using clone. copy_*
3150 * functions used by kernel_clone() cannot be used here directly
3151 * because they modify an inactive task_struct that is being
3152 * constructed. Here we are modifying the current, active,
3153 * task_struct.
3154 */
3155int ksys_unshare(unsigned long unshare_flags)
3156{
3157 struct fs_struct *fs, *new_fs = NULL;
3158 struct files_struct *new_fd = NULL;
3159 struct cred *new_cred = NULL;
3160 struct nsproxy *new_nsproxy = NULL;
3161 int do_sysvsem = 0;
3162 int err;
3163
3164 /*
3165 * If unsharing a user namespace must also unshare the thread group
3166 * and unshare the filesystem root and working directories.
3167 */
3168 if (unshare_flags & CLONE_NEWUSER)
3169 unshare_flags |= CLONE_THREAD | CLONE_FS;
3170 /*
3171 * If unsharing vm, must also unshare signal handlers.
3172 */
3173 if (unshare_flags & CLONE_VM)
3174 unshare_flags |= CLONE_SIGHAND;
3175 /*
3176 * If unsharing a signal handlers, must also unshare the signal queues.
3177 */
3178 if (unshare_flags & CLONE_SIGHAND)
3179 unshare_flags |= CLONE_THREAD;
3180 /*
3181 * If unsharing namespace, must also unshare filesystem information.
3182 */
3183 if (unshare_flags & CLONE_NEWNS)
3184 unshare_flags |= CLONE_FS;
3185
3186 err = check_unshare_flags(unshare_flags);
3187 if (err)
3188 goto bad_unshare_out;
3189 /*
3190 * CLONE_NEWIPC must also detach from the undolist: after switching
3191 * to a new ipc namespace, the semaphore arrays from the old
3192 * namespace are unreachable.
3193 */
3194 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3195 do_sysvsem = 1;
3196 err = unshare_fs(unshare_flags, &new_fs);
3197 if (err)
3198 goto bad_unshare_out;
3199 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3200 if (err)
3201 goto bad_unshare_cleanup_fs;
3202 err = unshare_userns(unshare_flags, &new_cred);
3203 if (err)
3204 goto bad_unshare_cleanup_fd;
3205 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3206 new_cred, new_fs);
3207 if (err)
3208 goto bad_unshare_cleanup_cred;
3209
3210 if (new_cred) {
3211 err = set_cred_ucounts(new_cred);
3212 if (err)
3213 goto bad_unshare_cleanup_cred;
3214 }
3215
3216 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3217 if (do_sysvsem) {
3218 /*
3219 * CLONE_SYSVSEM is equivalent to sys_exit().
3220 */
3221 exit_sem(current);
3222 }
3223 if (unshare_flags & CLONE_NEWIPC) {
3224 /* Orphan segments in old ns (see sem above). */
3225 exit_shm(current);
3226 shm_init_task(current);
3227 }
3228
3229 if (new_nsproxy)
3230 switch_task_namespaces(current, new_nsproxy);
3231
3232 task_lock(current);
3233
3234 if (new_fs) {
3235 fs = current->fs;
3236 spin_lock(&fs->lock);
3237 current->fs = new_fs;
3238 if (--fs->users)
3239 new_fs = NULL;
3240 else
3241 new_fs = fs;
3242 spin_unlock(&fs->lock);
3243 }
3244
3245 if (new_fd)
3246 swap(current->files, new_fd);
3247
3248 task_unlock(current);
3249
3250 if (new_cred) {
3251 /* Install the new user namespace */
3252 commit_creds(new_cred);
3253 new_cred = NULL;
3254 }
3255 }
3256
3257 perf_event_namespaces(current);
3258
3259bad_unshare_cleanup_cred:
3260 if (new_cred)
3261 put_cred(new_cred);
3262bad_unshare_cleanup_fd:
3263 if (new_fd)
3264 put_files_struct(new_fd);
3265
3266bad_unshare_cleanup_fs:
3267 if (new_fs)
3268 free_fs_struct(new_fs);
3269
3270bad_unshare_out:
3271 return err;
3272}
3273
3274SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3275{
3276 return ksys_unshare(unshare_flags);
3277}
3278
3279/*
3280 * Helper to unshare the files of the current task.
3281 * We don't want to expose copy_files internals to
3282 * the exec layer of the kernel.
3283 */
3284
3285int unshare_files(void)
3286{
3287 struct task_struct *task = current;
3288 struct files_struct *old, *copy = NULL;
3289 int error;
3290
3291 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3292 if (error || !copy)
3293 return error;
3294
3295 old = task->files;
3296 task_lock(task);
3297 task->files = copy;
3298 task_unlock(task);
3299 put_files_struct(old);
3300 return 0;
3301}
3302
3303int sysctl_max_threads(struct ctl_table *table, int write,
3304 void *buffer, size_t *lenp, loff_t *ppos)
3305{
3306 struct ctl_table t;
3307 int ret;
3308 int threads = max_threads;
3309 int min = 1;
3310 int max = MAX_THREADS;
3311
3312 t = *table;
3313 t.data = &threads;
3314 t.extra1 = &min;
3315 t.extra2 = &max;
3316
3317 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3318 if (ret || !write)
3319 return ret;
3320
3321 max_threads = threads;
3322
3323 return 0;
3324}