Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/jiffies.h>
57#include <linux/futex.h>
58#include <linux/compat.h>
59#include <linux/kthread.h>
60#include <linux/task_io_accounting_ops.h>
61#include <linux/rcupdate.h>
62#include <linux/ptrace.h>
63#include <linux/mount.h>
64#include <linux/audit.h>
65#include <linux/memcontrol.h>
66#include <linux/ftrace.h>
67#include <linux/proc_fs.h>
68#include <linux/profile.h>
69#include <linux/rmap.h>
70#include <linux/ksm.h>
71#include <linux/acct.h>
72#include <linux/userfaultfd_k.h>
73#include <linux/tsacct_kern.h>
74#include <linux/cn_proc.h>
75#include <linux/freezer.h>
76#include <linux/delayacct.h>
77#include <linux/taskstats_kern.h>
78#include <linux/tty.h>
79#include <linux/fs_struct.h>
80#include <linux/magic.h>
81#include <linux/perf_event.h>
82#include <linux/posix-timers.h>
83#include <linux/user-return-notifier.h>
84#include <linux/oom.h>
85#include <linux/khugepaged.h>
86#include <linux/signalfd.h>
87#include <linux/uprobes.h>
88#include <linux/aio.h>
89#include <linux/compiler.h>
90#include <linux/sysctl.h>
91#include <linux/kcov.h>
92#include <linux/livepatch.h>
93#include <linux/thread_info.h>
94#include <linux/stackleak.h>
95#include <linux/kasan.h>
96#include <linux/scs.h>
97#include <linux/io_uring.h>
98#include <linux/bpf.h>
99#include <linux/stackprotector.h>
100
101#include <asm/pgalloc.h>
102#include <linux/uaccess.h>
103#include <asm/mmu_context.h>
104#include <asm/cacheflush.h>
105#include <asm/tlbflush.h>
106
107#include <trace/events/sched.h>
108
109#define CREATE_TRACE_POINTS
110#include <trace/events/task.h>
111
112/*
113 * Minimum number of threads to boot the kernel
114 */
115#define MIN_THREADS 20
116
117/*
118 * Maximum number of threads
119 */
120#define MAX_THREADS FUTEX_TID_MASK
121
122/*
123 * Protected counters by write_lock_irq(&tasklist_lock)
124 */
125unsigned long total_forks; /* Handle normal Linux uptimes. */
126int nr_threads; /* The idle threads do not count.. */
127
128static int max_threads; /* tunable limit on nr_threads */
129
130#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
131
132static const char * const resident_page_types[] = {
133 NAMED_ARRAY_INDEX(MM_FILEPAGES),
134 NAMED_ARRAY_INDEX(MM_ANONPAGES),
135 NAMED_ARRAY_INDEX(MM_SWAPENTS),
136 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
137};
138
139DEFINE_PER_CPU(unsigned long, process_counts) = 0;
140
141__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
142
143#ifdef CONFIG_PROVE_RCU
144int lockdep_tasklist_lock_is_held(void)
145{
146 return lockdep_is_held(&tasklist_lock);
147}
148EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
149#endif /* #ifdef CONFIG_PROVE_RCU */
150
151int nr_processes(void)
152{
153 int cpu;
154 int total = 0;
155
156 for_each_possible_cpu(cpu)
157 total += per_cpu(process_counts, cpu);
158
159 return total;
160}
161
162void __weak arch_release_task_struct(struct task_struct *tsk)
163{
164}
165
166#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
167static struct kmem_cache *task_struct_cachep;
168
169static inline struct task_struct *alloc_task_struct_node(int node)
170{
171 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
172}
173
174static inline void free_task_struct(struct task_struct *tsk)
175{
176 kmem_cache_free(task_struct_cachep, tsk);
177}
178#endif
179
180#ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
181
182/*
183 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
184 * kmemcache based allocator.
185 */
186# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
187
188# ifdef CONFIG_VMAP_STACK
189/*
190 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
191 * flush. Try to minimize the number of calls by caching stacks.
192 */
193#define NR_CACHED_STACKS 2
194static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
195
196struct vm_stack {
197 struct rcu_head rcu;
198 struct vm_struct *stack_vm_area;
199};
200
201static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
202{
203 unsigned int i;
204
205 for (i = 0; i < NR_CACHED_STACKS; i++) {
206 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
207 continue;
208 return true;
209 }
210 return false;
211}
212
213static void thread_stack_free_rcu(struct rcu_head *rh)
214{
215 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
216
217 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
218 return;
219
220 vfree(vm_stack);
221}
222
223static void thread_stack_delayed_free(struct task_struct *tsk)
224{
225 struct vm_stack *vm_stack = tsk->stack;
226
227 vm_stack->stack_vm_area = tsk->stack_vm_area;
228 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
229}
230
231static int free_vm_stack_cache(unsigned int cpu)
232{
233 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
234 int i;
235
236 for (i = 0; i < NR_CACHED_STACKS; i++) {
237 struct vm_struct *vm_stack = cached_vm_stacks[i];
238
239 if (!vm_stack)
240 continue;
241
242 vfree(vm_stack->addr);
243 cached_vm_stacks[i] = NULL;
244 }
245
246 return 0;
247}
248
249static int memcg_charge_kernel_stack(struct vm_struct *vm)
250{
251 int i;
252 int ret;
253
254 BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
255 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
256
257 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
258 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
259 if (ret)
260 goto err;
261 }
262 return 0;
263err:
264 /*
265 * If memcg_kmem_charge_page() fails, page's memory cgroup pointer is
266 * NULL, and memcg_kmem_uncharge_page() in free_thread_stack() will
267 * ignore this page.
268 */
269 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
270 memcg_kmem_uncharge_page(vm->pages[i], 0);
271 return ret;
272}
273
274static int alloc_thread_stack_node(struct task_struct *tsk, int node)
275{
276 struct vm_struct *vm;
277 void *stack;
278 int i;
279
280 for (i = 0; i < NR_CACHED_STACKS; i++) {
281 struct vm_struct *s;
282
283 s = this_cpu_xchg(cached_stacks[i], NULL);
284
285 if (!s)
286 continue;
287
288 /* Reset stack metadata. */
289 kasan_unpoison_range(s->addr, THREAD_SIZE);
290
291 stack = kasan_reset_tag(s->addr);
292
293 /* Clear stale pointers from reused stack. */
294 memset(stack, 0, THREAD_SIZE);
295
296 if (memcg_charge_kernel_stack(s)) {
297 vfree(s->addr);
298 return -ENOMEM;
299 }
300
301 tsk->stack_vm_area = s;
302 tsk->stack = stack;
303 return 0;
304 }
305
306 /*
307 * Allocated stacks are cached and later reused by new threads,
308 * so memcg accounting is performed manually on assigning/releasing
309 * stacks to tasks. Drop __GFP_ACCOUNT.
310 */
311 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
312 VMALLOC_START, VMALLOC_END,
313 THREADINFO_GFP & ~__GFP_ACCOUNT,
314 PAGE_KERNEL,
315 0, node, __builtin_return_address(0));
316 if (!stack)
317 return -ENOMEM;
318
319 vm = find_vm_area(stack);
320 if (memcg_charge_kernel_stack(vm)) {
321 vfree(stack);
322 return -ENOMEM;
323 }
324 /*
325 * We can't call find_vm_area() in interrupt context, and
326 * free_thread_stack() can be called in interrupt context,
327 * so cache the vm_struct.
328 */
329 tsk->stack_vm_area = vm;
330 stack = kasan_reset_tag(stack);
331 tsk->stack = stack;
332 return 0;
333}
334
335static void free_thread_stack(struct task_struct *tsk)
336{
337 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
338 thread_stack_delayed_free(tsk);
339
340 tsk->stack = NULL;
341 tsk->stack_vm_area = NULL;
342}
343
344# else /* !CONFIG_VMAP_STACK */
345
346static void thread_stack_free_rcu(struct rcu_head *rh)
347{
348 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
349}
350
351static void thread_stack_delayed_free(struct task_struct *tsk)
352{
353 struct rcu_head *rh = tsk->stack;
354
355 call_rcu(rh, thread_stack_free_rcu);
356}
357
358static int alloc_thread_stack_node(struct task_struct *tsk, int node)
359{
360 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
361 THREAD_SIZE_ORDER);
362
363 if (likely(page)) {
364 tsk->stack = kasan_reset_tag(page_address(page));
365 return 0;
366 }
367 return -ENOMEM;
368}
369
370static void free_thread_stack(struct task_struct *tsk)
371{
372 thread_stack_delayed_free(tsk);
373 tsk->stack = NULL;
374}
375
376# endif /* CONFIG_VMAP_STACK */
377# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
378
379static struct kmem_cache *thread_stack_cache;
380
381static void thread_stack_free_rcu(struct rcu_head *rh)
382{
383 kmem_cache_free(thread_stack_cache, rh);
384}
385
386static void thread_stack_delayed_free(struct task_struct *tsk)
387{
388 struct rcu_head *rh = tsk->stack;
389
390 call_rcu(rh, thread_stack_free_rcu);
391}
392
393static int alloc_thread_stack_node(struct task_struct *tsk, int node)
394{
395 unsigned long *stack;
396 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
397 stack = kasan_reset_tag(stack);
398 tsk->stack = stack;
399 return stack ? 0 : -ENOMEM;
400}
401
402static void free_thread_stack(struct task_struct *tsk)
403{
404 thread_stack_delayed_free(tsk);
405 tsk->stack = NULL;
406}
407
408void thread_stack_cache_init(void)
409{
410 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
411 THREAD_SIZE, THREAD_SIZE, 0, 0,
412 THREAD_SIZE, NULL);
413 BUG_ON(thread_stack_cache == NULL);
414}
415
416# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
417#else /* CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
418
419static int alloc_thread_stack_node(struct task_struct *tsk, int node)
420{
421 unsigned long *stack;
422
423 stack = arch_alloc_thread_stack_node(tsk, node);
424 tsk->stack = stack;
425 return stack ? 0 : -ENOMEM;
426}
427
428static void free_thread_stack(struct task_struct *tsk)
429{
430 arch_free_thread_stack(tsk);
431 tsk->stack = NULL;
432}
433
434#endif /* !CONFIG_ARCH_THREAD_STACK_ALLOCATOR */
435
436/* SLAB cache for signal_struct structures (tsk->signal) */
437static struct kmem_cache *signal_cachep;
438
439/* SLAB cache for sighand_struct structures (tsk->sighand) */
440struct kmem_cache *sighand_cachep;
441
442/* SLAB cache for files_struct structures (tsk->files) */
443struct kmem_cache *files_cachep;
444
445/* SLAB cache for fs_struct structures (tsk->fs) */
446struct kmem_cache *fs_cachep;
447
448/* SLAB cache for vm_area_struct structures */
449static struct kmem_cache *vm_area_cachep;
450
451/* SLAB cache for mm_struct structures (tsk->mm) */
452static struct kmem_cache *mm_cachep;
453
454struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
455{
456 struct vm_area_struct *vma;
457
458 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
459 if (vma)
460 vma_init(vma, mm);
461 return vma;
462}
463
464struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
465{
466 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
467
468 if (new) {
469 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
470 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
471 /*
472 * orig->shared.rb may be modified concurrently, but the clone
473 * will be reinitialized.
474 */
475 *new = data_race(*orig);
476 INIT_LIST_HEAD(&new->anon_vma_chain);
477 dup_anon_vma_name(orig, new);
478 }
479 return new;
480}
481
482void vm_area_free(struct vm_area_struct *vma)
483{
484 free_anon_vma_name(vma);
485 kmem_cache_free(vm_area_cachep, vma);
486}
487
488static void account_kernel_stack(struct task_struct *tsk, int account)
489{
490 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
491 struct vm_struct *vm = task_stack_vm_area(tsk);
492 int i;
493
494 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
495 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
496 account * (PAGE_SIZE / 1024));
497 } else {
498 void *stack = task_stack_page(tsk);
499
500 /* All stack pages are in the same node. */
501 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
502 account * (THREAD_SIZE / 1024));
503 }
504}
505
506void exit_task_stack_account(struct task_struct *tsk)
507{
508 account_kernel_stack(tsk, -1);
509
510 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
511 struct vm_struct *vm;
512 int i;
513
514 vm = task_stack_vm_area(tsk);
515 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
516 memcg_kmem_uncharge_page(vm->pages[i], 0);
517 }
518}
519
520static void release_task_stack(struct task_struct *tsk)
521{
522 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
523 return; /* Better to leak the stack than to free prematurely */
524
525 free_thread_stack(tsk);
526}
527
528#ifdef CONFIG_THREAD_INFO_IN_TASK
529void put_task_stack(struct task_struct *tsk)
530{
531 if (refcount_dec_and_test(&tsk->stack_refcount))
532 release_task_stack(tsk);
533}
534#endif
535
536void free_task(struct task_struct *tsk)
537{
538#ifdef CONFIG_SECCOMP
539 WARN_ON_ONCE(tsk->seccomp.filter);
540#endif
541 release_user_cpus_ptr(tsk);
542 scs_release(tsk);
543
544#ifndef CONFIG_THREAD_INFO_IN_TASK
545 /*
546 * The task is finally done with both the stack and thread_info,
547 * so free both.
548 */
549 release_task_stack(tsk);
550#else
551 /*
552 * If the task had a separate stack allocation, it should be gone
553 * by now.
554 */
555 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
556#endif
557 rt_mutex_debug_task_free(tsk);
558 ftrace_graph_exit_task(tsk);
559 arch_release_task_struct(tsk);
560 if (tsk->flags & PF_KTHREAD)
561 free_kthread_struct(tsk);
562 free_task_struct(tsk);
563}
564EXPORT_SYMBOL(free_task);
565
566static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
567{
568 struct file *exe_file;
569
570 exe_file = get_mm_exe_file(oldmm);
571 RCU_INIT_POINTER(mm->exe_file, exe_file);
572 /*
573 * We depend on the oldmm having properly denied write access to the
574 * exe_file already.
575 */
576 if (exe_file && deny_write_access(exe_file))
577 pr_warn_once("deny_write_access() failed in %s\n", __func__);
578}
579
580#ifdef CONFIG_MMU
581static __latent_entropy int dup_mmap(struct mm_struct *mm,
582 struct mm_struct *oldmm)
583{
584 struct vm_area_struct *mpnt, *tmp;
585 int retval;
586 unsigned long charge = 0;
587 LIST_HEAD(uf);
588 MA_STATE(old_mas, &oldmm->mm_mt, 0, 0);
589 MA_STATE(mas, &mm->mm_mt, 0, 0);
590
591 uprobe_start_dup_mmap();
592 if (mmap_write_lock_killable(oldmm)) {
593 retval = -EINTR;
594 goto fail_uprobe_end;
595 }
596 flush_cache_dup_mm(oldmm);
597 uprobe_dup_mmap(oldmm, mm);
598 /*
599 * Not linked in yet - no deadlock potential:
600 */
601 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
602
603 /* No ordering required: file already has been exposed. */
604 dup_mm_exe_file(mm, oldmm);
605
606 mm->total_vm = oldmm->total_vm;
607 mm->data_vm = oldmm->data_vm;
608 mm->exec_vm = oldmm->exec_vm;
609 mm->stack_vm = oldmm->stack_vm;
610
611 retval = ksm_fork(mm, oldmm);
612 if (retval)
613 goto out;
614 khugepaged_fork(mm, oldmm);
615
616 retval = mas_expected_entries(&mas, oldmm->map_count);
617 if (retval)
618 goto out;
619
620 mas_for_each(&old_mas, mpnt, ULONG_MAX) {
621 struct file *file;
622
623 if (mpnt->vm_flags & VM_DONTCOPY) {
624 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
625 continue;
626 }
627 charge = 0;
628 /*
629 * Don't duplicate many vmas if we've been oom-killed (for
630 * example)
631 */
632 if (fatal_signal_pending(current)) {
633 retval = -EINTR;
634 goto loop_out;
635 }
636 if (mpnt->vm_flags & VM_ACCOUNT) {
637 unsigned long len = vma_pages(mpnt);
638
639 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
640 goto fail_nomem;
641 charge = len;
642 }
643 tmp = vm_area_dup(mpnt);
644 if (!tmp)
645 goto fail_nomem;
646 retval = vma_dup_policy(mpnt, tmp);
647 if (retval)
648 goto fail_nomem_policy;
649 tmp->vm_mm = mm;
650 retval = dup_userfaultfd(tmp, &uf);
651 if (retval)
652 goto fail_nomem_anon_vma_fork;
653 if (tmp->vm_flags & VM_WIPEONFORK) {
654 /*
655 * VM_WIPEONFORK gets a clean slate in the child.
656 * Don't prepare anon_vma until fault since we don't
657 * copy page for current vma.
658 */
659 tmp->anon_vma = NULL;
660 } else if (anon_vma_fork(tmp, mpnt))
661 goto fail_nomem_anon_vma_fork;
662 tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
663 file = tmp->vm_file;
664 if (file) {
665 struct address_space *mapping = file->f_mapping;
666
667 get_file(file);
668 i_mmap_lock_write(mapping);
669 if (tmp->vm_flags & VM_SHARED)
670 mapping_allow_writable(mapping);
671 flush_dcache_mmap_lock(mapping);
672 /* insert tmp into the share list, just after mpnt */
673 vma_interval_tree_insert_after(tmp, mpnt,
674 &mapping->i_mmap);
675 flush_dcache_mmap_unlock(mapping);
676 i_mmap_unlock_write(mapping);
677 }
678
679 /*
680 * Copy/update hugetlb private vma information.
681 */
682 if (is_vm_hugetlb_page(tmp))
683 hugetlb_dup_vma_private(tmp);
684
685 /* Link the vma into the MT */
686 mas.index = tmp->vm_start;
687 mas.last = tmp->vm_end - 1;
688 mas_store(&mas, tmp);
689 if (mas_is_err(&mas))
690 goto fail_nomem_mas_store;
691
692 mm->map_count++;
693 if (!(tmp->vm_flags & VM_WIPEONFORK))
694 retval = copy_page_range(tmp, mpnt);
695
696 if (tmp->vm_ops && tmp->vm_ops->open)
697 tmp->vm_ops->open(tmp);
698
699 if (retval)
700 goto loop_out;
701 }
702 /* a new mm has just been created */
703 retval = arch_dup_mmap(oldmm, mm);
704loop_out:
705 mas_destroy(&mas);
706out:
707 mmap_write_unlock(mm);
708 flush_tlb_mm(oldmm);
709 mmap_write_unlock(oldmm);
710 dup_userfaultfd_complete(&uf);
711fail_uprobe_end:
712 uprobe_end_dup_mmap();
713 return retval;
714
715fail_nomem_mas_store:
716 unlink_anon_vmas(tmp);
717fail_nomem_anon_vma_fork:
718 mpol_put(vma_policy(tmp));
719fail_nomem_policy:
720 vm_area_free(tmp);
721fail_nomem:
722 retval = -ENOMEM;
723 vm_unacct_memory(charge);
724 goto loop_out;
725}
726
727static inline int mm_alloc_pgd(struct mm_struct *mm)
728{
729 mm->pgd = pgd_alloc(mm);
730 if (unlikely(!mm->pgd))
731 return -ENOMEM;
732 return 0;
733}
734
735static inline void mm_free_pgd(struct mm_struct *mm)
736{
737 pgd_free(mm, mm->pgd);
738}
739#else
740static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
741{
742 mmap_write_lock(oldmm);
743 dup_mm_exe_file(mm, oldmm);
744 mmap_write_unlock(oldmm);
745 return 0;
746}
747#define mm_alloc_pgd(mm) (0)
748#define mm_free_pgd(mm)
749#endif /* CONFIG_MMU */
750
751static void check_mm(struct mm_struct *mm)
752{
753 int i;
754
755 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
756 "Please make sure 'struct resident_page_types[]' is updated as well");
757
758 for (i = 0; i < NR_MM_COUNTERS; i++) {
759 long x = percpu_counter_sum(&mm->rss_stat[i]);
760
761 if (likely(!x))
762 continue;
763
764 /* Making sure this is not due to race with CPU offlining. */
765 x = percpu_counter_sum_all(&mm->rss_stat[i]);
766 if (unlikely(x))
767 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
768 mm, resident_page_types[i], x);
769 }
770
771 if (mm_pgtables_bytes(mm))
772 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
773 mm_pgtables_bytes(mm));
774
775#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
776 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
777#endif
778}
779
780#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
781#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
782
783/*
784 * Called when the last reference to the mm
785 * is dropped: either by a lazy thread or by
786 * mmput. Free the page directory and the mm.
787 */
788void __mmdrop(struct mm_struct *mm)
789{
790 int i;
791
792 BUG_ON(mm == &init_mm);
793 WARN_ON_ONCE(mm == current->mm);
794 WARN_ON_ONCE(mm == current->active_mm);
795 mm_free_pgd(mm);
796 destroy_context(mm);
797 mmu_notifier_subscriptions_destroy(mm);
798 check_mm(mm);
799 put_user_ns(mm->user_ns);
800 mm_pasid_drop(mm);
801
802 for (i = 0; i < NR_MM_COUNTERS; i++)
803 percpu_counter_destroy(&mm->rss_stat[i]);
804 free_mm(mm);
805}
806EXPORT_SYMBOL_GPL(__mmdrop);
807
808static void mmdrop_async_fn(struct work_struct *work)
809{
810 struct mm_struct *mm;
811
812 mm = container_of(work, struct mm_struct, async_put_work);
813 __mmdrop(mm);
814}
815
816static void mmdrop_async(struct mm_struct *mm)
817{
818 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
819 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
820 schedule_work(&mm->async_put_work);
821 }
822}
823
824static inline void free_signal_struct(struct signal_struct *sig)
825{
826 taskstats_tgid_free(sig);
827 sched_autogroup_exit(sig);
828 /*
829 * __mmdrop is not safe to call from softirq context on x86 due to
830 * pgd_dtor so postpone it to the async context
831 */
832 if (sig->oom_mm)
833 mmdrop_async(sig->oom_mm);
834 kmem_cache_free(signal_cachep, sig);
835}
836
837static inline void put_signal_struct(struct signal_struct *sig)
838{
839 if (refcount_dec_and_test(&sig->sigcnt))
840 free_signal_struct(sig);
841}
842
843void __put_task_struct(struct task_struct *tsk)
844{
845 WARN_ON(!tsk->exit_state);
846 WARN_ON(refcount_read(&tsk->usage));
847 WARN_ON(tsk == current);
848
849 io_uring_free(tsk);
850 cgroup_free(tsk);
851 task_numa_free(tsk, true);
852 security_task_free(tsk);
853 bpf_task_storage_free(tsk);
854 exit_creds(tsk);
855 delayacct_tsk_free(tsk);
856 put_signal_struct(tsk->signal);
857 sched_core_free(tsk);
858 free_task(tsk);
859}
860EXPORT_SYMBOL_GPL(__put_task_struct);
861
862void __init __weak arch_task_cache_init(void) { }
863
864/*
865 * set_max_threads
866 */
867static void set_max_threads(unsigned int max_threads_suggested)
868{
869 u64 threads;
870 unsigned long nr_pages = totalram_pages();
871
872 /*
873 * The number of threads shall be limited such that the thread
874 * structures may only consume a small part of the available memory.
875 */
876 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
877 threads = MAX_THREADS;
878 else
879 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
880 (u64) THREAD_SIZE * 8UL);
881
882 if (threads > max_threads_suggested)
883 threads = max_threads_suggested;
884
885 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
886}
887
888#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
889/* Initialized by the architecture: */
890int arch_task_struct_size __read_mostly;
891#endif
892
893#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
894static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
895{
896 /* Fetch thread_struct whitelist for the architecture. */
897 arch_thread_struct_whitelist(offset, size);
898
899 /*
900 * Handle zero-sized whitelist or empty thread_struct, otherwise
901 * adjust offset to position of thread_struct in task_struct.
902 */
903 if (unlikely(*size == 0))
904 *offset = 0;
905 else
906 *offset += offsetof(struct task_struct, thread);
907}
908#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
909
910void __init fork_init(void)
911{
912 int i;
913#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
914#ifndef ARCH_MIN_TASKALIGN
915#define ARCH_MIN_TASKALIGN 0
916#endif
917 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
918 unsigned long useroffset, usersize;
919
920 /* create a slab on which task_structs can be allocated */
921 task_struct_whitelist(&useroffset, &usersize);
922 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
923 arch_task_struct_size, align,
924 SLAB_PANIC|SLAB_ACCOUNT,
925 useroffset, usersize, NULL);
926#endif
927
928 /* do the arch specific task caches init */
929 arch_task_cache_init();
930
931 set_max_threads(MAX_THREADS);
932
933 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
934 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
935 init_task.signal->rlim[RLIMIT_SIGPENDING] =
936 init_task.signal->rlim[RLIMIT_NPROC];
937
938 for (i = 0; i < UCOUNT_COUNTS; i++)
939 init_user_ns.ucount_max[i] = max_threads/2;
940
941 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
942 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
943 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
944 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
945
946#ifdef CONFIG_VMAP_STACK
947 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
948 NULL, free_vm_stack_cache);
949#endif
950
951 scs_init();
952
953 lockdep_init_task(&init_task);
954 uprobes_init();
955}
956
957int __weak arch_dup_task_struct(struct task_struct *dst,
958 struct task_struct *src)
959{
960 *dst = *src;
961 return 0;
962}
963
964void set_task_stack_end_magic(struct task_struct *tsk)
965{
966 unsigned long *stackend;
967
968 stackend = end_of_stack(tsk);
969 *stackend = STACK_END_MAGIC; /* for overflow detection */
970}
971
972static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
973{
974 struct task_struct *tsk;
975 int err;
976
977 if (node == NUMA_NO_NODE)
978 node = tsk_fork_get_node(orig);
979 tsk = alloc_task_struct_node(node);
980 if (!tsk)
981 return NULL;
982
983 err = arch_dup_task_struct(tsk, orig);
984 if (err)
985 goto free_tsk;
986
987 err = alloc_thread_stack_node(tsk, node);
988 if (err)
989 goto free_tsk;
990
991#ifdef CONFIG_THREAD_INFO_IN_TASK
992 refcount_set(&tsk->stack_refcount, 1);
993#endif
994 account_kernel_stack(tsk, 1);
995
996 err = scs_prepare(tsk, node);
997 if (err)
998 goto free_stack;
999
1000#ifdef CONFIG_SECCOMP
1001 /*
1002 * We must handle setting up seccomp filters once we're under
1003 * the sighand lock in case orig has changed between now and
1004 * then. Until then, filter must be NULL to avoid messing up
1005 * the usage counts on the error path calling free_task.
1006 */
1007 tsk->seccomp.filter = NULL;
1008#endif
1009
1010 setup_thread_stack(tsk, orig);
1011 clear_user_return_notifier(tsk);
1012 clear_tsk_need_resched(tsk);
1013 set_task_stack_end_magic(tsk);
1014 clear_syscall_work_syscall_user_dispatch(tsk);
1015
1016#ifdef CONFIG_STACKPROTECTOR
1017 tsk->stack_canary = get_random_canary();
1018#endif
1019 if (orig->cpus_ptr == &orig->cpus_mask)
1020 tsk->cpus_ptr = &tsk->cpus_mask;
1021 dup_user_cpus_ptr(tsk, orig, node);
1022
1023 /*
1024 * One for the user space visible state that goes away when reaped.
1025 * One for the scheduler.
1026 */
1027 refcount_set(&tsk->rcu_users, 2);
1028 /* One for the rcu users */
1029 refcount_set(&tsk->usage, 1);
1030#ifdef CONFIG_BLK_DEV_IO_TRACE
1031 tsk->btrace_seq = 0;
1032#endif
1033 tsk->splice_pipe = NULL;
1034 tsk->task_frag.page = NULL;
1035 tsk->wake_q.next = NULL;
1036 tsk->worker_private = NULL;
1037
1038 kcov_task_init(tsk);
1039 kmsan_task_create(tsk);
1040 kmap_local_fork(tsk);
1041
1042#ifdef CONFIG_FAULT_INJECTION
1043 tsk->fail_nth = 0;
1044#endif
1045
1046#ifdef CONFIG_BLK_CGROUP
1047 tsk->throttle_queue = NULL;
1048 tsk->use_memdelay = 0;
1049#endif
1050
1051#ifdef CONFIG_IOMMU_SVA
1052 tsk->pasid_activated = 0;
1053#endif
1054
1055#ifdef CONFIG_MEMCG
1056 tsk->active_memcg = NULL;
1057#endif
1058
1059#ifdef CONFIG_CPU_SUP_INTEL
1060 tsk->reported_split_lock = 0;
1061#endif
1062
1063 return tsk;
1064
1065free_stack:
1066 exit_task_stack_account(tsk);
1067 free_thread_stack(tsk);
1068free_tsk:
1069 free_task_struct(tsk);
1070 return NULL;
1071}
1072
1073__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1074
1075static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1076
1077static int __init coredump_filter_setup(char *s)
1078{
1079 default_dump_filter =
1080 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1081 MMF_DUMP_FILTER_MASK;
1082 return 1;
1083}
1084
1085__setup("coredump_filter=", coredump_filter_setup);
1086
1087#include <linux/init_task.h>
1088
1089static void mm_init_aio(struct mm_struct *mm)
1090{
1091#ifdef CONFIG_AIO
1092 spin_lock_init(&mm->ioctx_lock);
1093 mm->ioctx_table = NULL;
1094#endif
1095}
1096
1097static __always_inline void mm_clear_owner(struct mm_struct *mm,
1098 struct task_struct *p)
1099{
1100#ifdef CONFIG_MEMCG
1101 if (mm->owner == p)
1102 WRITE_ONCE(mm->owner, NULL);
1103#endif
1104}
1105
1106static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1107{
1108#ifdef CONFIG_MEMCG
1109 mm->owner = p;
1110#endif
1111}
1112
1113static void mm_init_uprobes_state(struct mm_struct *mm)
1114{
1115#ifdef CONFIG_UPROBES
1116 mm->uprobes_state.xol_area = NULL;
1117#endif
1118}
1119
1120static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1121 struct user_namespace *user_ns)
1122{
1123 int i;
1124
1125 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1126 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1127 atomic_set(&mm->mm_users, 1);
1128 atomic_set(&mm->mm_count, 1);
1129 seqcount_init(&mm->write_protect_seq);
1130 mmap_init_lock(mm);
1131 INIT_LIST_HEAD(&mm->mmlist);
1132 mm_pgtables_bytes_init(mm);
1133 mm->map_count = 0;
1134 mm->locked_vm = 0;
1135 atomic64_set(&mm->pinned_vm, 0);
1136 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1137 spin_lock_init(&mm->page_table_lock);
1138 spin_lock_init(&mm->arg_lock);
1139 mm_init_cpumask(mm);
1140 mm_init_aio(mm);
1141 mm_init_owner(mm, p);
1142 mm_pasid_init(mm);
1143 RCU_INIT_POINTER(mm->exe_file, NULL);
1144 mmu_notifier_subscriptions_init(mm);
1145 init_tlb_flush_pending(mm);
1146#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1147 mm->pmd_huge_pte = NULL;
1148#endif
1149 mm_init_uprobes_state(mm);
1150 hugetlb_count_init(mm);
1151
1152 if (current->mm) {
1153 mm->flags = current->mm->flags & MMF_INIT_MASK;
1154 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1155 } else {
1156 mm->flags = default_dump_filter;
1157 mm->def_flags = 0;
1158 }
1159
1160 if (mm_alloc_pgd(mm))
1161 goto fail_nopgd;
1162
1163 if (init_new_context(p, mm))
1164 goto fail_nocontext;
1165
1166 for (i = 0; i < NR_MM_COUNTERS; i++)
1167 if (percpu_counter_init(&mm->rss_stat[i], 0, GFP_KERNEL_ACCOUNT))
1168 goto fail_pcpu;
1169
1170 mm->user_ns = get_user_ns(user_ns);
1171 lru_gen_init_mm(mm);
1172 return mm;
1173
1174fail_pcpu:
1175 while (i > 0)
1176 percpu_counter_destroy(&mm->rss_stat[--i]);
1177fail_nocontext:
1178 mm_free_pgd(mm);
1179fail_nopgd:
1180 free_mm(mm);
1181 return NULL;
1182}
1183
1184/*
1185 * Allocate and initialize an mm_struct.
1186 */
1187struct mm_struct *mm_alloc(void)
1188{
1189 struct mm_struct *mm;
1190
1191 mm = allocate_mm();
1192 if (!mm)
1193 return NULL;
1194
1195 memset(mm, 0, sizeof(*mm));
1196 return mm_init(mm, current, current_user_ns());
1197}
1198
1199static inline void __mmput(struct mm_struct *mm)
1200{
1201 VM_BUG_ON(atomic_read(&mm->mm_users));
1202
1203 uprobe_clear_state(mm);
1204 exit_aio(mm);
1205 ksm_exit(mm);
1206 khugepaged_exit(mm); /* must run before exit_mmap */
1207 exit_mmap(mm);
1208 mm_put_huge_zero_page(mm);
1209 set_mm_exe_file(mm, NULL);
1210 if (!list_empty(&mm->mmlist)) {
1211 spin_lock(&mmlist_lock);
1212 list_del(&mm->mmlist);
1213 spin_unlock(&mmlist_lock);
1214 }
1215 if (mm->binfmt)
1216 module_put(mm->binfmt->module);
1217 lru_gen_del_mm(mm);
1218 mmdrop(mm);
1219}
1220
1221/*
1222 * Decrement the use count and release all resources for an mm.
1223 */
1224void mmput(struct mm_struct *mm)
1225{
1226 might_sleep();
1227
1228 if (atomic_dec_and_test(&mm->mm_users))
1229 __mmput(mm);
1230}
1231EXPORT_SYMBOL_GPL(mmput);
1232
1233#ifdef CONFIG_MMU
1234static void mmput_async_fn(struct work_struct *work)
1235{
1236 struct mm_struct *mm = container_of(work, struct mm_struct,
1237 async_put_work);
1238
1239 __mmput(mm);
1240}
1241
1242void mmput_async(struct mm_struct *mm)
1243{
1244 if (atomic_dec_and_test(&mm->mm_users)) {
1245 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1246 schedule_work(&mm->async_put_work);
1247 }
1248}
1249EXPORT_SYMBOL_GPL(mmput_async);
1250#endif
1251
1252/**
1253 * set_mm_exe_file - change a reference to the mm's executable file
1254 *
1255 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1256 *
1257 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1258 * invocations: in mmput() nobody alive left, in execve task is single
1259 * threaded.
1260 *
1261 * Can only fail if new_exe_file != NULL.
1262 */
1263int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1264{
1265 struct file *old_exe_file;
1266
1267 /*
1268 * It is safe to dereference the exe_file without RCU as
1269 * this function is only called if nobody else can access
1270 * this mm -- see comment above for justification.
1271 */
1272 old_exe_file = rcu_dereference_raw(mm->exe_file);
1273
1274 if (new_exe_file) {
1275 /*
1276 * We expect the caller (i.e., sys_execve) to already denied
1277 * write access, so this is unlikely to fail.
1278 */
1279 if (unlikely(deny_write_access(new_exe_file)))
1280 return -EACCES;
1281 get_file(new_exe_file);
1282 }
1283 rcu_assign_pointer(mm->exe_file, new_exe_file);
1284 if (old_exe_file) {
1285 allow_write_access(old_exe_file);
1286 fput(old_exe_file);
1287 }
1288 return 0;
1289}
1290
1291/**
1292 * replace_mm_exe_file - replace a reference to the mm's executable file
1293 *
1294 * This changes mm's executable file (shown as symlink /proc/[pid]/exe),
1295 * dealing with concurrent invocation and without grabbing the mmap lock in
1296 * write mode.
1297 *
1298 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1299 */
1300int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1301{
1302 struct vm_area_struct *vma;
1303 struct file *old_exe_file;
1304 int ret = 0;
1305
1306 /* Forbid mm->exe_file change if old file still mapped. */
1307 old_exe_file = get_mm_exe_file(mm);
1308 if (old_exe_file) {
1309 VMA_ITERATOR(vmi, mm, 0);
1310 mmap_read_lock(mm);
1311 for_each_vma(vmi, vma) {
1312 if (!vma->vm_file)
1313 continue;
1314 if (path_equal(&vma->vm_file->f_path,
1315 &old_exe_file->f_path)) {
1316 ret = -EBUSY;
1317 break;
1318 }
1319 }
1320 mmap_read_unlock(mm);
1321 fput(old_exe_file);
1322 if (ret)
1323 return ret;
1324 }
1325
1326 /* set the new file, lockless */
1327 ret = deny_write_access(new_exe_file);
1328 if (ret)
1329 return -EACCES;
1330 get_file(new_exe_file);
1331
1332 old_exe_file = xchg(&mm->exe_file, new_exe_file);
1333 if (old_exe_file) {
1334 /*
1335 * Don't race with dup_mmap() getting the file and disallowing
1336 * write access while someone might open the file writable.
1337 */
1338 mmap_read_lock(mm);
1339 allow_write_access(old_exe_file);
1340 fput(old_exe_file);
1341 mmap_read_unlock(mm);
1342 }
1343 return 0;
1344}
1345
1346/**
1347 * get_mm_exe_file - acquire a reference to the mm's executable file
1348 *
1349 * Returns %NULL if mm has no associated executable file.
1350 * User must release file via fput().
1351 */
1352struct file *get_mm_exe_file(struct mm_struct *mm)
1353{
1354 struct file *exe_file;
1355
1356 rcu_read_lock();
1357 exe_file = rcu_dereference(mm->exe_file);
1358 if (exe_file && !get_file_rcu(exe_file))
1359 exe_file = NULL;
1360 rcu_read_unlock();
1361 return exe_file;
1362}
1363
1364/**
1365 * get_task_exe_file - acquire a reference to the task's executable file
1366 *
1367 * Returns %NULL if task's mm (if any) has no associated executable file or
1368 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1369 * User must release file via fput().
1370 */
1371struct file *get_task_exe_file(struct task_struct *task)
1372{
1373 struct file *exe_file = NULL;
1374 struct mm_struct *mm;
1375
1376 task_lock(task);
1377 mm = task->mm;
1378 if (mm) {
1379 if (!(task->flags & PF_KTHREAD))
1380 exe_file = get_mm_exe_file(mm);
1381 }
1382 task_unlock(task);
1383 return exe_file;
1384}
1385
1386/**
1387 * get_task_mm - acquire a reference to the task's mm
1388 *
1389 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1390 * this kernel workthread has transiently adopted a user mm with use_mm,
1391 * to do its AIO) is not set and if so returns a reference to it, after
1392 * bumping up the use count. User must release the mm via mmput()
1393 * after use. Typically used by /proc and ptrace.
1394 */
1395struct mm_struct *get_task_mm(struct task_struct *task)
1396{
1397 struct mm_struct *mm;
1398
1399 task_lock(task);
1400 mm = task->mm;
1401 if (mm) {
1402 if (task->flags & PF_KTHREAD)
1403 mm = NULL;
1404 else
1405 mmget(mm);
1406 }
1407 task_unlock(task);
1408 return mm;
1409}
1410EXPORT_SYMBOL_GPL(get_task_mm);
1411
1412struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1413{
1414 struct mm_struct *mm;
1415 int err;
1416
1417 err = down_read_killable(&task->signal->exec_update_lock);
1418 if (err)
1419 return ERR_PTR(err);
1420
1421 mm = get_task_mm(task);
1422 if (mm && mm != current->mm &&
1423 !ptrace_may_access(task, mode)) {
1424 mmput(mm);
1425 mm = ERR_PTR(-EACCES);
1426 }
1427 up_read(&task->signal->exec_update_lock);
1428
1429 return mm;
1430}
1431
1432static void complete_vfork_done(struct task_struct *tsk)
1433{
1434 struct completion *vfork;
1435
1436 task_lock(tsk);
1437 vfork = tsk->vfork_done;
1438 if (likely(vfork)) {
1439 tsk->vfork_done = NULL;
1440 complete(vfork);
1441 }
1442 task_unlock(tsk);
1443}
1444
1445static int wait_for_vfork_done(struct task_struct *child,
1446 struct completion *vfork)
1447{
1448 unsigned int state = TASK_UNINTERRUPTIBLE|TASK_KILLABLE|TASK_FREEZABLE;
1449 int killed;
1450
1451 cgroup_enter_frozen();
1452 killed = wait_for_completion_state(vfork, state);
1453 cgroup_leave_frozen(false);
1454
1455 if (killed) {
1456 task_lock(child);
1457 child->vfork_done = NULL;
1458 task_unlock(child);
1459 }
1460
1461 put_task_struct(child);
1462 return killed;
1463}
1464
1465/* Please note the differences between mmput and mm_release.
1466 * mmput is called whenever we stop holding onto a mm_struct,
1467 * error success whatever.
1468 *
1469 * mm_release is called after a mm_struct has been removed
1470 * from the current process.
1471 *
1472 * This difference is important for error handling, when we
1473 * only half set up a mm_struct for a new process and need to restore
1474 * the old one. Because we mmput the new mm_struct before
1475 * restoring the old one. . .
1476 * Eric Biederman 10 January 1998
1477 */
1478static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1479{
1480 uprobe_free_utask(tsk);
1481
1482 /* Get rid of any cached register state */
1483 deactivate_mm(tsk, mm);
1484
1485 /*
1486 * Signal userspace if we're not exiting with a core dump
1487 * because we want to leave the value intact for debugging
1488 * purposes.
1489 */
1490 if (tsk->clear_child_tid) {
1491 if (atomic_read(&mm->mm_users) > 1) {
1492 /*
1493 * We don't check the error code - if userspace has
1494 * not set up a proper pointer then tough luck.
1495 */
1496 put_user(0, tsk->clear_child_tid);
1497 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1498 1, NULL, NULL, 0, 0);
1499 }
1500 tsk->clear_child_tid = NULL;
1501 }
1502
1503 /*
1504 * All done, finally we can wake up parent and return this mm to him.
1505 * Also kthread_stop() uses this completion for synchronization.
1506 */
1507 if (tsk->vfork_done)
1508 complete_vfork_done(tsk);
1509}
1510
1511void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1512{
1513 futex_exit_release(tsk);
1514 mm_release(tsk, mm);
1515}
1516
1517void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1518{
1519 futex_exec_release(tsk);
1520 mm_release(tsk, mm);
1521}
1522
1523/**
1524 * dup_mm() - duplicates an existing mm structure
1525 * @tsk: the task_struct with which the new mm will be associated.
1526 * @oldmm: the mm to duplicate.
1527 *
1528 * Allocates a new mm structure and duplicates the provided @oldmm structure
1529 * content into it.
1530 *
1531 * Return: the duplicated mm or NULL on failure.
1532 */
1533static struct mm_struct *dup_mm(struct task_struct *tsk,
1534 struct mm_struct *oldmm)
1535{
1536 struct mm_struct *mm;
1537 int err;
1538
1539 mm = allocate_mm();
1540 if (!mm)
1541 goto fail_nomem;
1542
1543 memcpy(mm, oldmm, sizeof(*mm));
1544
1545 if (!mm_init(mm, tsk, mm->user_ns))
1546 goto fail_nomem;
1547
1548 err = dup_mmap(mm, oldmm);
1549 if (err)
1550 goto free_pt;
1551
1552 mm->hiwater_rss = get_mm_rss(mm);
1553 mm->hiwater_vm = mm->total_vm;
1554
1555 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1556 goto free_pt;
1557
1558 return mm;
1559
1560free_pt:
1561 /* don't put binfmt in mmput, we haven't got module yet */
1562 mm->binfmt = NULL;
1563 mm_init_owner(mm, NULL);
1564 mmput(mm);
1565
1566fail_nomem:
1567 return NULL;
1568}
1569
1570static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1571{
1572 struct mm_struct *mm, *oldmm;
1573
1574 tsk->min_flt = tsk->maj_flt = 0;
1575 tsk->nvcsw = tsk->nivcsw = 0;
1576#ifdef CONFIG_DETECT_HUNG_TASK
1577 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1578 tsk->last_switch_time = 0;
1579#endif
1580
1581 tsk->mm = NULL;
1582 tsk->active_mm = NULL;
1583
1584 /*
1585 * Are we cloning a kernel thread?
1586 *
1587 * We need to steal a active VM for that..
1588 */
1589 oldmm = current->mm;
1590 if (!oldmm)
1591 return 0;
1592
1593 if (clone_flags & CLONE_VM) {
1594 mmget(oldmm);
1595 mm = oldmm;
1596 } else {
1597 mm = dup_mm(tsk, current->mm);
1598 if (!mm)
1599 return -ENOMEM;
1600 }
1601
1602 tsk->mm = mm;
1603 tsk->active_mm = mm;
1604 return 0;
1605}
1606
1607static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1608{
1609 struct fs_struct *fs = current->fs;
1610 if (clone_flags & CLONE_FS) {
1611 /* tsk->fs is already what we want */
1612 spin_lock(&fs->lock);
1613 if (fs->in_exec) {
1614 spin_unlock(&fs->lock);
1615 return -EAGAIN;
1616 }
1617 fs->users++;
1618 spin_unlock(&fs->lock);
1619 return 0;
1620 }
1621 tsk->fs = copy_fs_struct(fs);
1622 if (!tsk->fs)
1623 return -ENOMEM;
1624 return 0;
1625}
1626
1627static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1628{
1629 struct files_struct *oldf, *newf;
1630 int error = 0;
1631
1632 /*
1633 * A background process may not have any files ...
1634 */
1635 oldf = current->files;
1636 if (!oldf)
1637 goto out;
1638
1639 if (clone_flags & CLONE_FILES) {
1640 atomic_inc(&oldf->count);
1641 goto out;
1642 }
1643
1644 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1645 if (!newf)
1646 goto out;
1647
1648 tsk->files = newf;
1649 error = 0;
1650out:
1651 return error;
1652}
1653
1654static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1655{
1656 struct sighand_struct *sig;
1657
1658 if (clone_flags & CLONE_SIGHAND) {
1659 refcount_inc(¤t->sighand->count);
1660 return 0;
1661 }
1662 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1663 RCU_INIT_POINTER(tsk->sighand, sig);
1664 if (!sig)
1665 return -ENOMEM;
1666
1667 refcount_set(&sig->count, 1);
1668 spin_lock_irq(¤t->sighand->siglock);
1669 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1670 spin_unlock_irq(¤t->sighand->siglock);
1671
1672 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1673 if (clone_flags & CLONE_CLEAR_SIGHAND)
1674 flush_signal_handlers(tsk, 0);
1675
1676 return 0;
1677}
1678
1679void __cleanup_sighand(struct sighand_struct *sighand)
1680{
1681 if (refcount_dec_and_test(&sighand->count)) {
1682 signalfd_cleanup(sighand);
1683 /*
1684 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1685 * without an RCU grace period, see __lock_task_sighand().
1686 */
1687 kmem_cache_free(sighand_cachep, sighand);
1688 }
1689}
1690
1691/*
1692 * Initialize POSIX timer handling for a thread group.
1693 */
1694static void posix_cpu_timers_init_group(struct signal_struct *sig)
1695{
1696 struct posix_cputimers *pct = &sig->posix_cputimers;
1697 unsigned long cpu_limit;
1698
1699 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1700 posix_cputimers_group_init(pct, cpu_limit);
1701}
1702
1703static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1704{
1705 struct signal_struct *sig;
1706
1707 if (clone_flags & CLONE_THREAD)
1708 return 0;
1709
1710 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1711 tsk->signal = sig;
1712 if (!sig)
1713 return -ENOMEM;
1714
1715 sig->nr_threads = 1;
1716 sig->quick_threads = 1;
1717 atomic_set(&sig->live, 1);
1718 refcount_set(&sig->sigcnt, 1);
1719
1720 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1721 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1722 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1723
1724 init_waitqueue_head(&sig->wait_chldexit);
1725 sig->curr_target = tsk;
1726 init_sigpending(&sig->shared_pending);
1727 INIT_HLIST_HEAD(&sig->multiprocess);
1728 seqlock_init(&sig->stats_lock);
1729 prev_cputime_init(&sig->prev_cputime);
1730
1731#ifdef CONFIG_POSIX_TIMERS
1732 INIT_LIST_HEAD(&sig->posix_timers);
1733 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1734 sig->real_timer.function = it_real_fn;
1735#endif
1736
1737 task_lock(current->group_leader);
1738 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1739 task_unlock(current->group_leader);
1740
1741 posix_cpu_timers_init_group(sig);
1742
1743 tty_audit_fork(sig);
1744 sched_autogroup_fork(sig);
1745
1746 sig->oom_score_adj = current->signal->oom_score_adj;
1747 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1748
1749 mutex_init(&sig->cred_guard_mutex);
1750 init_rwsem(&sig->exec_update_lock);
1751
1752 return 0;
1753}
1754
1755static void copy_seccomp(struct task_struct *p)
1756{
1757#ifdef CONFIG_SECCOMP
1758 /*
1759 * Must be called with sighand->lock held, which is common to
1760 * all threads in the group. Holding cred_guard_mutex is not
1761 * needed because this new task is not yet running and cannot
1762 * be racing exec.
1763 */
1764 assert_spin_locked(¤t->sighand->siglock);
1765
1766 /* Ref-count the new filter user, and assign it. */
1767 get_seccomp_filter(current);
1768 p->seccomp = current->seccomp;
1769
1770 /*
1771 * Explicitly enable no_new_privs here in case it got set
1772 * between the task_struct being duplicated and holding the
1773 * sighand lock. The seccomp state and nnp must be in sync.
1774 */
1775 if (task_no_new_privs(current))
1776 task_set_no_new_privs(p);
1777
1778 /*
1779 * If the parent gained a seccomp mode after copying thread
1780 * flags and between before we held the sighand lock, we have
1781 * to manually enable the seccomp thread flag here.
1782 */
1783 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1784 set_task_syscall_work(p, SECCOMP);
1785#endif
1786}
1787
1788SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1789{
1790 current->clear_child_tid = tidptr;
1791
1792 return task_pid_vnr(current);
1793}
1794
1795static void rt_mutex_init_task(struct task_struct *p)
1796{
1797 raw_spin_lock_init(&p->pi_lock);
1798#ifdef CONFIG_RT_MUTEXES
1799 p->pi_waiters = RB_ROOT_CACHED;
1800 p->pi_top_task = NULL;
1801 p->pi_blocked_on = NULL;
1802#endif
1803}
1804
1805static inline void init_task_pid_links(struct task_struct *task)
1806{
1807 enum pid_type type;
1808
1809 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1810 INIT_HLIST_NODE(&task->pid_links[type]);
1811}
1812
1813static inline void
1814init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1815{
1816 if (type == PIDTYPE_PID)
1817 task->thread_pid = pid;
1818 else
1819 task->signal->pids[type] = pid;
1820}
1821
1822static inline void rcu_copy_process(struct task_struct *p)
1823{
1824#ifdef CONFIG_PREEMPT_RCU
1825 p->rcu_read_lock_nesting = 0;
1826 p->rcu_read_unlock_special.s = 0;
1827 p->rcu_blocked_node = NULL;
1828 INIT_LIST_HEAD(&p->rcu_node_entry);
1829#endif /* #ifdef CONFIG_PREEMPT_RCU */
1830#ifdef CONFIG_TASKS_RCU
1831 p->rcu_tasks_holdout = false;
1832 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1833 p->rcu_tasks_idle_cpu = -1;
1834#endif /* #ifdef CONFIG_TASKS_RCU */
1835#ifdef CONFIG_TASKS_TRACE_RCU
1836 p->trc_reader_nesting = 0;
1837 p->trc_reader_special.s = 0;
1838 INIT_LIST_HEAD(&p->trc_holdout_list);
1839 INIT_LIST_HEAD(&p->trc_blkd_node);
1840#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1841}
1842
1843struct pid *pidfd_pid(const struct file *file)
1844{
1845 if (file->f_op == &pidfd_fops)
1846 return file->private_data;
1847
1848 return ERR_PTR(-EBADF);
1849}
1850
1851static int pidfd_release(struct inode *inode, struct file *file)
1852{
1853 struct pid *pid = file->private_data;
1854
1855 file->private_data = NULL;
1856 put_pid(pid);
1857 return 0;
1858}
1859
1860#ifdef CONFIG_PROC_FS
1861/**
1862 * pidfd_show_fdinfo - print information about a pidfd
1863 * @m: proc fdinfo file
1864 * @f: file referencing a pidfd
1865 *
1866 * Pid:
1867 * This function will print the pid that a given pidfd refers to in the
1868 * pid namespace of the procfs instance.
1869 * If the pid namespace of the process is not a descendant of the pid
1870 * namespace of the procfs instance 0 will be shown as its pid. This is
1871 * similar to calling getppid() on a process whose parent is outside of
1872 * its pid namespace.
1873 *
1874 * NSpid:
1875 * If pid namespaces are supported then this function will also print
1876 * the pid of a given pidfd refers to for all descendant pid namespaces
1877 * starting from the current pid namespace of the instance, i.e. the
1878 * Pid field and the first entry in the NSpid field will be identical.
1879 * If the pid namespace of the process is not a descendant of the pid
1880 * namespace of the procfs instance 0 will be shown as its first NSpid
1881 * entry and no others will be shown.
1882 * Note that this differs from the Pid and NSpid fields in
1883 * /proc/<pid>/status where Pid and NSpid are always shown relative to
1884 * the pid namespace of the procfs instance. The difference becomes
1885 * obvious when sending around a pidfd between pid namespaces from a
1886 * different branch of the tree, i.e. where no ancestral relation is
1887 * present between the pid namespaces:
1888 * - create two new pid namespaces ns1 and ns2 in the initial pid
1889 * namespace (also take care to create new mount namespaces in the
1890 * new pid namespace and mount procfs)
1891 * - create a process with a pidfd in ns1
1892 * - send pidfd from ns1 to ns2
1893 * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1894 * have exactly one entry, which is 0
1895 */
1896static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1897{
1898 struct pid *pid = f->private_data;
1899 struct pid_namespace *ns;
1900 pid_t nr = -1;
1901
1902 if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1903 ns = proc_pid_ns(file_inode(m->file)->i_sb);
1904 nr = pid_nr_ns(pid, ns);
1905 }
1906
1907 seq_put_decimal_ll(m, "Pid:\t", nr);
1908
1909#ifdef CONFIG_PID_NS
1910 seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1911 if (nr > 0) {
1912 int i;
1913
1914 /* If nr is non-zero it means that 'pid' is valid and that
1915 * ns, i.e. the pid namespace associated with the procfs
1916 * instance, is in the pid namespace hierarchy of pid.
1917 * Start at one below the already printed level.
1918 */
1919 for (i = ns->level + 1; i <= pid->level; i++)
1920 seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1921 }
1922#endif
1923 seq_putc(m, '\n');
1924}
1925#endif
1926
1927/*
1928 * Poll support for process exit notification.
1929 */
1930static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1931{
1932 struct pid *pid = file->private_data;
1933 __poll_t poll_flags = 0;
1934
1935 poll_wait(file, &pid->wait_pidfd, pts);
1936
1937 /*
1938 * Inform pollers only when the whole thread group exits.
1939 * If the thread group leader exits before all other threads in the
1940 * group, then poll(2) should block, similar to the wait(2) family.
1941 */
1942 if (thread_group_exited(pid))
1943 poll_flags = EPOLLIN | EPOLLRDNORM;
1944
1945 return poll_flags;
1946}
1947
1948const struct file_operations pidfd_fops = {
1949 .release = pidfd_release,
1950 .poll = pidfd_poll,
1951#ifdef CONFIG_PROC_FS
1952 .show_fdinfo = pidfd_show_fdinfo,
1953#endif
1954};
1955
1956static void __delayed_free_task(struct rcu_head *rhp)
1957{
1958 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1959
1960 free_task(tsk);
1961}
1962
1963static __always_inline void delayed_free_task(struct task_struct *tsk)
1964{
1965 if (IS_ENABLED(CONFIG_MEMCG))
1966 call_rcu(&tsk->rcu, __delayed_free_task);
1967 else
1968 free_task(tsk);
1969}
1970
1971static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1972{
1973 /* Skip if kernel thread */
1974 if (!tsk->mm)
1975 return;
1976
1977 /* Skip if spawning a thread or using vfork */
1978 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1979 return;
1980
1981 /* We need to synchronize with __set_oom_adj */
1982 mutex_lock(&oom_adj_mutex);
1983 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1984 /* Update the values in case they were changed after copy_signal */
1985 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1986 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1987 mutex_unlock(&oom_adj_mutex);
1988}
1989
1990#ifdef CONFIG_RV
1991static void rv_task_fork(struct task_struct *p)
1992{
1993 int i;
1994
1995 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
1996 p->rv[i].da_mon.monitoring = false;
1997}
1998#else
1999#define rv_task_fork(p) do {} while (0)
2000#endif
2001
2002/*
2003 * This creates a new process as a copy of the old one,
2004 * but does not actually start it yet.
2005 *
2006 * It copies the registers, and all the appropriate
2007 * parts of the process environment (as per the clone
2008 * flags). The actual kick-off is left to the caller.
2009 */
2010static __latent_entropy struct task_struct *copy_process(
2011 struct pid *pid,
2012 int trace,
2013 int node,
2014 struct kernel_clone_args *args)
2015{
2016 int pidfd = -1, retval;
2017 struct task_struct *p;
2018 struct multiprocess_signals delayed;
2019 struct file *pidfile = NULL;
2020 const u64 clone_flags = args->flags;
2021 struct nsproxy *nsp = current->nsproxy;
2022
2023 /*
2024 * Don't allow sharing the root directory with processes in a different
2025 * namespace
2026 */
2027 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2028 return ERR_PTR(-EINVAL);
2029
2030 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2031 return ERR_PTR(-EINVAL);
2032
2033 /*
2034 * Thread groups must share signals as well, and detached threads
2035 * can only be started up within the thread group.
2036 */
2037 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2038 return ERR_PTR(-EINVAL);
2039
2040 /*
2041 * Shared signal handlers imply shared VM. By way of the above,
2042 * thread groups also imply shared VM. Blocking this case allows
2043 * for various simplifications in other code.
2044 */
2045 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2046 return ERR_PTR(-EINVAL);
2047
2048 /*
2049 * Siblings of global init remain as zombies on exit since they are
2050 * not reaped by their parent (swapper). To solve this and to avoid
2051 * multi-rooted process trees, prevent global and container-inits
2052 * from creating siblings.
2053 */
2054 if ((clone_flags & CLONE_PARENT) &&
2055 current->signal->flags & SIGNAL_UNKILLABLE)
2056 return ERR_PTR(-EINVAL);
2057
2058 /*
2059 * If the new process will be in a different pid or user namespace
2060 * do not allow it to share a thread group with the forking task.
2061 */
2062 if (clone_flags & CLONE_THREAD) {
2063 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2064 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2065 return ERR_PTR(-EINVAL);
2066 }
2067
2068 if (clone_flags & CLONE_PIDFD) {
2069 /*
2070 * - CLONE_DETACHED is blocked so that we can potentially
2071 * reuse it later for CLONE_PIDFD.
2072 * - CLONE_THREAD is blocked until someone really needs it.
2073 */
2074 if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
2075 return ERR_PTR(-EINVAL);
2076 }
2077
2078 /*
2079 * Force any signals received before this point to be delivered
2080 * before the fork happens. Collect up signals sent to multiple
2081 * processes that happen during the fork and delay them so that
2082 * they appear to happen after the fork.
2083 */
2084 sigemptyset(&delayed.signal);
2085 INIT_HLIST_NODE(&delayed.node);
2086
2087 spin_lock_irq(¤t->sighand->siglock);
2088 if (!(clone_flags & CLONE_THREAD))
2089 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2090 recalc_sigpending();
2091 spin_unlock_irq(¤t->sighand->siglock);
2092 retval = -ERESTARTNOINTR;
2093 if (task_sigpending(current))
2094 goto fork_out;
2095
2096 retval = -ENOMEM;
2097 p = dup_task_struct(current, node);
2098 if (!p)
2099 goto fork_out;
2100 p->flags &= ~PF_KTHREAD;
2101 if (args->kthread)
2102 p->flags |= PF_KTHREAD;
2103 if (args->io_thread) {
2104 /*
2105 * Mark us an IO worker, and block any signal that isn't
2106 * fatal or STOP
2107 */
2108 p->flags |= PF_IO_WORKER;
2109 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2110 }
2111
2112 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2113 /*
2114 * Clear TID on mm_release()?
2115 */
2116 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2117
2118 ftrace_graph_init_task(p);
2119
2120 rt_mutex_init_task(p);
2121
2122 lockdep_assert_irqs_enabled();
2123#ifdef CONFIG_PROVE_LOCKING
2124 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2125#endif
2126 retval = copy_creds(p, clone_flags);
2127 if (retval < 0)
2128 goto bad_fork_free;
2129
2130 retval = -EAGAIN;
2131 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2132 if (p->real_cred->user != INIT_USER &&
2133 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2134 goto bad_fork_cleanup_count;
2135 }
2136 current->flags &= ~PF_NPROC_EXCEEDED;
2137
2138 /*
2139 * If multiple threads are within copy_process(), then this check
2140 * triggers too late. This doesn't hurt, the check is only there
2141 * to stop root fork bombs.
2142 */
2143 retval = -EAGAIN;
2144 if (data_race(nr_threads >= max_threads))
2145 goto bad_fork_cleanup_count;
2146
2147 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2148 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2149 p->flags |= PF_FORKNOEXEC;
2150 INIT_LIST_HEAD(&p->children);
2151 INIT_LIST_HEAD(&p->sibling);
2152 rcu_copy_process(p);
2153 p->vfork_done = NULL;
2154 spin_lock_init(&p->alloc_lock);
2155
2156 init_sigpending(&p->pending);
2157
2158 p->utime = p->stime = p->gtime = 0;
2159#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2160 p->utimescaled = p->stimescaled = 0;
2161#endif
2162 prev_cputime_init(&p->prev_cputime);
2163
2164#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2165 seqcount_init(&p->vtime.seqcount);
2166 p->vtime.starttime = 0;
2167 p->vtime.state = VTIME_INACTIVE;
2168#endif
2169
2170#ifdef CONFIG_IO_URING
2171 p->io_uring = NULL;
2172#endif
2173
2174#if defined(SPLIT_RSS_COUNTING)
2175 memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2176#endif
2177
2178 p->default_timer_slack_ns = current->timer_slack_ns;
2179
2180#ifdef CONFIG_PSI
2181 p->psi_flags = 0;
2182#endif
2183
2184 task_io_accounting_init(&p->ioac);
2185 acct_clear_integrals(p);
2186
2187 posix_cputimers_init(&p->posix_cputimers);
2188
2189 p->io_context = NULL;
2190 audit_set_context(p, NULL);
2191 cgroup_fork(p);
2192 if (args->kthread) {
2193 if (!set_kthread_struct(p))
2194 goto bad_fork_cleanup_delayacct;
2195 }
2196#ifdef CONFIG_NUMA
2197 p->mempolicy = mpol_dup(p->mempolicy);
2198 if (IS_ERR(p->mempolicy)) {
2199 retval = PTR_ERR(p->mempolicy);
2200 p->mempolicy = NULL;
2201 goto bad_fork_cleanup_delayacct;
2202 }
2203#endif
2204#ifdef CONFIG_CPUSETS
2205 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2206 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2207 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2208#endif
2209#ifdef CONFIG_TRACE_IRQFLAGS
2210 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2211 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2212 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2213 p->softirqs_enabled = 1;
2214 p->softirq_context = 0;
2215#endif
2216
2217 p->pagefault_disabled = 0;
2218
2219#ifdef CONFIG_LOCKDEP
2220 lockdep_init_task(p);
2221#endif
2222
2223#ifdef CONFIG_DEBUG_MUTEXES
2224 p->blocked_on = NULL; /* not blocked yet */
2225#endif
2226#ifdef CONFIG_BCACHE
2227 p->sequential_io = 0;
2228 p->sequential_io_avg = 0;
2229#endif
2230#ifdef CONFIG_BPF_SYSCALL
2231 RCU_INIT_POINTER(p->bpf_storage, NULL);
2232 p->bpf_ctx = NULL;
2233#endif
2234
2235 /* Perform scheduler related setup. Assign this task to a CPU. */
2236 retval = sched_fork(clone_flags, p);
2237 if (retval)
2238 goto bad_fork_cleanup_policy;
2239
2240 retval = perf_event_init_task(p, clone_flags);
2241 if (retval)
2242 goto bad_fork_cleanup_policy;
2243 retval = audit_alloc(p);
2244 if (retval)
2245 goto bad_fork_cleanup_perf;
2246 /* copy all the process information */
2247 shm_init_task(p);
2248 retval = security_task_alloc(p, clone_flags);
2249 if (retval)
2250 goto bad_fork_cleanup_audit;
2251 retval = copy_semundo(clone_flags, p);
2252 if (retval)
2253 goto bad_fork_cleanup_security;
2254 retval = copy_files(clone_flags, p);
2255 if (retval)
2256 goto bad_fork_cleanup_semundo;
2257 retval = copy_fs(clone_flags, p);
2258 if (retval)
2259 goto bad_fork_cleanup_files;
2260 retval = copy_sighand(clone_flags, p);
2261 if (retval)
2262 goto bad_fork_cleanup_fs;
2263 retval = copy_signal(clone_flags, p);
2264 if (retval)
2265 goto bad_fork_cleanup_sighand;
2266 retval = copy_mm(clone_flags, p);
2267 if (retval)
2268 goto bad_fork_cleanup_signal;
2269 retval = copy_namespaces(clone_flags, p);
2270 if (retval)
2271 goto bad_fork_cleanup_mm;
2272 retval = copy_io(clone_flags, p);
2273 if (retval)
2274 goto bad_fork_cleanup_namespaces;
2275 retval = copy_thread(p, args);
2276 if (retval)
2277 goto bad_fork_cleanup_io;
2278
2279 stackleak_task_init(p);
2280
2281 if (pid != &init_struct_pid) {
2282 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2283 args->set_tid_size);
2284 if (IS_ERR(pid)) {
2285 retval = PTR_ERR(pid);
2286 goto bad_fork_cleanup_thread;
2287 }
2288 }
2289
2290 /*
2291 * This has to happen after we've potentially unshared the file
2292 * descriptor table (so that the pidfd doesn't leak into the child
2293 * if the fd table isn't shared).
2294 */
2295 if (clone_flags & CLONE_PIDFD) {
2296 retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2297 if (retval < 0)
2298 goto bad_fork_free_pid;
2299
2300 pidfd = retval;
2301
2302 pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2303 O_RDWR | O_CLOEXEC);
2304 if (IS_ERR(pidfile)) {
2305 put_unused_fd(pidfd);
2306 retval = PTR_ERR(pidfile);
2307 goto bad_fork_free_pid;
2308 }
2309 get_pid(pid); /* held by pidfile now */
2310
2311 retval = put_user(pidfd, args->pidfd);
2312 if (retval)
2313 goto bad_fork_put_pidfd;
2314 }
2315
2316#ifdef CONFIG_BLOCK
2317 p->plug = NULL;
2318#endif
2319 futex_init_task(p);
2320
2321 /*
2322 * sigaltstack should be cleared when sharing the same VM
2323 */
2324 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2325 sas_ss_reset(p);
2326
2327 /*
2328 * Syscall tracing and stepping should be turned off in the
2329 * child regardless of CLONE_PTRACE.
2330 */
2331 user_disable_single_step(p);
2332 clear_task_syscall_work(p, SYSCALL_TRACE);
2333#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2334 clear_task_syscall_work(p, SYSCALL_EMU);
2335#endif
2336 clear_tsk_latency_tracing(p);
2337
2338 /* ok, now we should be set up.. */
2339 p->pid = pid_nr(pid);
2340 if (clone_flags & CLONE_THREAD) {
2341 p->group_leader = current->group_leader;
2342 p->tgid = current->tgid;
2343 } else {
2344 p->group_leader = p;
2345 p->tgid = p->pid;
2346 }
2347
2348 p->nr_dirtied = 0;
2349 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2350 p->dirty_paused_when = 0;
2351
2352 p->pdeath_signal = 0;
2353 INIT_LIST_HEAD(&p->thread_group);
2354 p->task_works = NULL;
2355 clear_posix_cputimers_work(p);
2356
2357#ifdef CONFIG_KRETPROBES
2358 p->kretprobe_instances.first = NULL;
2359#endif
2360#ifdef CONFIG_RETHOOK
2361 p->rethooks.first = NULL;
2362#endif
2363
2364 /*
2365 * Ensure that the cgroup subsystem policies allow the new process to be
2366 * forked. It should be noted that the new process's css_set can be changed
2367 * between here and cgroup_post_fork() if an organisation operation is in
2368 * progress.
2369 */
2370 retval = cgroup_can_fork(p, args);
2371 if (retval)
2372 goto bad_fork_put_pidfd;
2373
2374 /*
2375 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2376 * the new task on the correct runqueue. All this *before* the task
2377 * becomes visible.
2378 *
2379 * This isn't part of ->can_fork() because while the re-cloning is
2380 * cgroup specific, it unconditionally needs to place the task on a
2381 * runqueue.
2382 */
2383 sched_cgroup_fork(p, args);
2384
2385 /*
2386 * From this point on we must avoid any synchronous user-space
2387 * communication until we take the tasklist-lock. In particular, we do
2388 * not want user-space to be able to predict the process start-time by
2389 * stalling fork(2) after we recorded the start_time but before it is
2390 * visible to the system.
2391 */
2392
2393 p->start_time = ktime_get_ns();
2394 p->start_boottime = ktime_get_boottime_ns();
2395
2396 /*
2397 * Make it visible to the rest of the system, but dont wake it up yet.
2398 * Need tasklist lock for parent etc handling!
2399 */
2400 write_lock_irq(&tasklist_lock);
2401
2402 /* CLONE_PARENT re-uses the old parent */
2403 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2404 p->real_parent = current->real_parent;
2405 p->parent_exec_id = current->parent_exec_id;
2406 if (clone_flags & CLONE_THREAD)
2407 p->exit_signal = -1;
2408 else
2409 p->exit_signal = current->group_leader->exit_signal;
2410 } else {
2411 p->real_parent = current;
2412 p->parent_exec_id = current->self_exec_id;
2413 p->exit_signal = args->exit_signal;
2414 }
2415
2416 klp_copy_process(p);
2417
2418 sched_core_fork(p);
2419
2420 spin_lock(¤t->sighand->siglock);
2421
2422 rv_task_fork(p);
2423
2424 rseq_fork(p, clone_flags);
2425
2426 /* Don't start children in a dying pid namespace */
2427 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2428 retval = -ENOMEM;
2429 goto bad_fork_cancel_cgroup;
2430 }
2431
2432 /* Let kill terminate clone/fork in the middle */
2433 if (fatal_signal_pending(current)) {
2434 retval = -EINTR;
2435 goto bad_fork_cancel_cgroup;
2436 }
2437
2438 /* No more failure paths after this point. */
2439
2440 /*
2441 * Copy seccomp details explicitly here, in case they were changed
2442 * before holding sighand lock.
2443 */
2444 copy_seccomp(p);
2445
2446 init_task_pid_links(p);
2447 if (likely(p->pid)) {
2448 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2449
2450 init_task_pid(p, PIDTYPE_PID, pid);
2451 if (thread_group_leader(p)) {
2452 init_task_pid(p, PIDTYPE_TGID, pid);
2453 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2454 init_task_pid(p, PIDTYPE_SID, task_session(current));
2455
2456 if (is_child_reaper(pid)) {
2457 ns_of_pid(pid)->child_reaper = p;
2458 p->signal->flags |= SIGNAL_UNKILLABLE;
2459 }
2460 p->signal->shared_pending.signal = delayed.signal;
2461 p->signal->tty = tty_kref_get(current->signal->tty);
2462 /*
2463 * Inherit has_child_subreaper flag under the same
2464 * tasklist_lock with adding child to the process tree
2465 * for propagate_has_child_subreaper optimization.
2466 */
2467 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2468 p->real_parent->signal->is_child_subreaper;
2469 list_add_tail(&p->sibling, &p->real_parent->children);
2470 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2471 attach_pid(p, PIDTYPE_TGID);
2472 attach_pid(p, PIDTYPE_PGID);
2473 attach_pid(p, PIDTYPE_SID);
2474 __this_cpu_inc(process_counts);
2475 } else {
2476 current->signal->nr_threads++;
2477 current->signal->quick_threads++;
2478 atomic_inc(¤t->signal->live);
2479 refcount_inc(¤t->signal->sigcnt);
2480 task_join_group_stop(p);
2481 list_add_tail_rcu(&p->thread_group,
2482 &p->group_leader->thread_group);
2483 list_add_tail_rcu(&p->thread_node,
2484 &p->signal->thread_head);
2485 }
2486 attach_pid(p, PIDTYPE_PID);
2487 nr_threads++;
2488 }
2489 total_forks++;
2490 hlist_del_init(&delayed.node);
2491 spin_unlock(¤t->sighand->siglock);
2492 syscall_tracepoint_update(p);
2493 write_unlock_irq(&tasklist_lock);
2494
2495 if (pidfile)
2496 fd_install(pidfd, pidfile);
2497
2498 proc_fork_connector(p);
2499 sched_post_fork(p);
2500 cgroup_post_fork(p, args);
2501 perf_event_fork(p);
2502
2503 trace_task_newtask(p, clone_flags);
2504 uprobe_copy_process(p, clone_flags);
2505
2506 copy_oom_score_adj(clone_flags, p);
2507
2508 return p;
2509
2510bad_fork_cancel_cgroup:
2511 sched_core_free(p);
2512 spin_unlock(¤t->sighand->siglock);
2513 write_unlock_irq(&tasklist_lock);
2514 cgroup_cancel_fork(p, args);
2515bad_fork_put_pidfd:
2516 if (clone_flags & CLONE_PIDFD) {
2517 fput(pidfile);
2518 put_unused_fd(pidfd);
2519 }
2520bad_fork_free_pid:
2521 if (pid != &init_struct_pid)
2522 free_pid(pid);
2523bad_fork_cleanup_thread:
2524 exit_thread(p);
2525bad_fork_cleanup_io:
2526 if (p->io_context)
2527 exit_io_context(p);
2528bad_fork_cleanup_namespaces:
2529 exit_task_namespaces(p);
2530bad_fork_cleanup_mm:
2531 if (p->mm) {
2532 mm_clear_owner(p->mm, p);
2533 mmput(p->mm);
2534 }
2535bad_fork_cleanup_signal:
2536 if (!(clone_flags & CLONE_THREAD))
2537 free_signal_struct(p->signal);
2538bad_fork_cleanup_sighand:
2539 __cleanup_sighand(p->sighand);
2540bad_fork_cleanup_fs:
2541 exit_fs(p); /* blocking */
2542bad_fork_cleanup_files:
2543 exit_files(p); /* blocking */
2544bad_fork_cleanup_semundo:
2545 exit_sem(p);
2546bad_fork_cleanup_security:
2547 security_task_free(p);
2548bad_fork_cleanup_audit:
2549 audit_free(p);
2550bad_fork_cleanup_perf:
2551 perf_event_free_task(p);
2552bad_fork_cleanup_policy:
2553 lockdep_free_task(p);
2554#ifdef CONFIG_NUMA
2555 mpol_put(p->mempolicy);
2556#endif
2557bad_fork_cleanup_delayacct:
2558 delayacct_tsk_free(p);
2559bad_fork_cleanup_count:
2560 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2561 exit_creds(p);
2562bad_fork_free:
2563 WRITE_ONCE(p->__state, TASK_DEAD);
2564 exit_task_stack_account(p);
2565 put_task_stack(p);
2566 delayed_free_task(p);
2567fork_out:
2568 spin_lock_irq(¤t->sighand->siglock);
2569 hlist_del_init(&delayed.node);
2570 spin_unlock_irq(¤t->sighand->siglock);
2571 return ERR_PTR(retval);
2572}
2573
2574static inline void init_idle_pids(struct task_struct *idle)
2575{
2576 enum pid_type type;
2577
2578 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2579 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2580 init_task_pid(idle, type, &init_struct_pid);
2581 }
2582}
2583
2584static int idle_dummy(void *dummy)
2585{
2586 /* This function is never called */
2587 return 0;
2588}
2589
2590struct task_struct * __init fork_idle(int cpu)
2591{
2592 struct task_struct *task;
2593 struct kernel_clone_args args = {
2594 .flags = CLONE_VM,
2595 .fn = &idle_dummy,
2596 .fn_arg = NULL,
2597 .kthread = 1,
2598 .idle = 1,
2599 };
2600
2601 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2602 if (!IS_ERR(task)) {
2603 init_idle_pids(task);
2604 init_idle(task, cpu);
2605 }
2606
2607 return task;
2608}
2609
2610/*
2611 * This is like kernel_clone(), but shaved down and tailored to just
2612 * creating io_uring workers. It returns a created task, or an error pointer.
2613 * The returned task is inactive, and the caller must fire it up through
2614 * wake_up_new_task(p). All signals are blocked in the created task.
2615 */
2616struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2617{
2618 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2619 CLONE_IO;
2620 struct kernel_clone_args args = {
2621 .flags = ((lower_32_bits(flags) | CLONE_VM |
2622 CLONE_UNTRACED) & ~CSIGNAL),
2623 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2624 .fn = fn,
2625 .fn_arg = arg,
2626 .io_thread = 1,
2627 };
2628
2629 return copy_process(NULL, 0, node, &args);
2630}
2631
2632/*
2633 * Ok, this is the main fork-routine.
2634 *
2635 * It copies the process, and if successful kick-starts
2636 * it and waits for it to finish using the VM if required.
2637 *
2638 * args->exit_signal is expected to be checked for sanity by the caller.
2639 */
2640pid_t kernel_clone(struct kernel_clone_args *args)
2641{
2642 u64 clone_flags = args->flags;
2643 struct completion vfork;
2644 struct pid *pid;
2645 struct task_struct *p;
2646 int trace = 0;
2647 pid_t nr;
2648
2649 /*
2650 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2651 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2652 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2653 * field in struct clone_args and it still doesn't make sense to have
2654 * them both point at the same memory location. Performing this check
2655 * here has the advantage that we don't need to have a separate helper
2656 * to check for legacy clone().
2657 */
2658 if ((args->flags & CLONE_PIDFD) &&
2659 (args->flags & CLONE_PARENT_SETTID) &&
2660 (args->pidfd == args->parent_tid))
2661 return -EINVAL;
2662
2663 /*
2664 * Determine whether and which event to report to ptracer. When
2665 * called from kernel_thread or CLONE_UNTRACED is explicitly
2666 * requested, no event is reported; otherwise, report if the event
2667 * for the type of forking is enabled.
2668 */
2669 if (!(clone_flags & CLONE_UNTRACED)) {
2670 if (clone_flags & CLONE_VFORK)
2671 trace = PTRACE_EVENT_VFORK;
2672 else if (args->exit_signal != SIGCHLD)
2673 trace = PTRACE_EVENT_CLONE;
2674 else
2675 trace = PTRACE_EVENT_FORK;
2676
2677 if (likely(!ptrace_event_enabled(current, trace)))
2678 trace = 0;
2679 }
2680
2681 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2682 add_latent_entropy();
2683
2684 if (IS_ERR(p))
2685 return PTR_ERR(p);
2686
2687 /*
2688 * Do this prior waking up the new thread - the thread pointer
2689 * might get invalid after that point, if the thread exits quickly.
2690 */
2691 trace_sched_process_fork(current, p);
2692
2693 pid = get_task_pid(p, PIDTYPE_PID);
2694 nr = pid_vnr(pid);
2695
2696 if (clone_flags & CLONE_PARENT_SETTID)
2697 put_user(nr, args->parent_tid);
2698
2699 if (clone_flags & CLONE_VFORK) {
2700 p->vfork_done = &vfork;
2701 init_completion(&vfork);
2702 get_task_struct(p);
2703 }
2704
2705 if (IS_ENABLED(CONFIG_LRU_GEN) && !(clone_flags & CLONE_VM)) {
2706 /* lock the task to synchronize with memcg migration */
2707 task_lock(p);
2708 lru_gen_add_mm(p->mm);
2709 task_unlock(p);
2710 }
2711
2712 wake_up_new_task(p);
2713
2714 /* forking complete and child started to run, tell ptracer */
2715 if (unlikely(trace))
2716 ptrace_event_pid(trace, pid);
2717
2718 if (clone_flags & CLONE_VFORK) {
2719 if (!wait_for_vfork_done(p, &vfork))
2720 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2721 }
2722
2723 put_pid(pid);
2724 return nr;
2725}
2726
2727/*
2728 * Create a kernel thread.
2729 */
2730pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2731{
2732 struct kernel_clone_args args = {
2733 .flags = ((lower_32_bits(flags) | CLONE_VM |
2734 CLONE_UNTRACED) & ~CSIGNAL),
2735 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2736 .fn = fn,
2737 .fn_arg = arg,
2738 .kthread = 1,
2739 };
2740
2741 return kernel_clone(&args);
2742}
2743
2744/*
2745 * Create a user mode thread.
2746 */
2747pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2748{
2749 struct kernel_clone_args args = {
2750 .flags = ((lower_32_bits(flags) | CLONE_VM |
2751 CLONE_UNTRACED) & ~CSIGNAL),
2752 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2753 .fn = fn,
2754 .fn_arg = arg,
2755 };
2756
2757 return kernel_clone(&args);
2758}
2759
2760#ifdef __ARCH_WANT_SYS_FORK
2761SYSCALL_DEFINE0(fork)
2762{
2763#ifdef CONFIG_MMU
2764 struct kernel_clone_args args = {
2765 .exit_signal = SIGCHLD,
2766 };
2767
2768 return kernel_clone(&args);
2769#else
2770 /* can not support in nommu mode */
2771 return -EINVAL;
2772#endif
2773}
2774#endif
2775
2776#ifdef __ARCH_WANT_SYS_VFORK
2777SYSCALL_DEFINE0(vfork)
2778{
2779 struct kernel_clone_args args = {
2780 .flags = CLONE_VFORK | CLONE_VM,
2781 .exit_signal = SIGCHLD,
2782 };
2783
2784 return kernel_clone(&args);
2785}
2786#endif
2787
2788#ifdef __ARCH_WANT_SYS_CLONE
2789#ifdef CONFIG_CLONE_BACKWARDS
2790SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2791 int __user *, parent_tidptr,
2792 unsigned long, tls,
2793 int __user *, child_tidptr)
2794#elif defined(CONFIG_CLONE_BACKWARDS2)
2795SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2796 int __user *, parent_tidptr,
2797 int __user *, child_tidptr,
2798 unsigned long, tls)
2799#elif defined(CONFIG_CLONE_BACKWARDS3)
2800SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2801 int, stack_size,
2802 int __user *, parent_tidptr,
2803 int __user *, child_tidptr,
2804 unsigned long, tls)
2805#else
2806SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2807 int __user *, parent_tidptr,
2808 int __user *, child_tidptr,
2809 unsigned long, tls)
2810#endif
2811{
2812 struct kernel_clone_args args = {
2813 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2814 .pidfd = parent_tidptr,
2815 .child_tid = child_tidptr,
2816 .parent_tid = parent_tidptr,
2817 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2818 .stack = newsp,
2819 .tls = tls,
2820 };
2821
2822 return kernel_clone(&args);
2823}
2824#endif
2825
2826#ifdef __ARCH_WANT_SYS_CLONE3
2827
2828noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2829 struct clone_args __user *uargs,
2830 size_t usize)
2831{
2832 int err;
2833 struct clone_args args;
2834 pid_t *kset_tid = kargs->set_tid;
2835
2836 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2837 CLONE_ARGS_SIZE_VER0);
2838 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2839 CLONE_ARGS_SIZE_VER1);
2840 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2841 CLONE_ARGS_SIZE_VER2);
2842 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2843
2844 if (unlikely(usize > PAGE_SIZE))
2845 return -E2BIG;
2846 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2847 return -EINVAL;
2848
2849 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2850 if (err)
2851 return err;
2852
2853 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2854 return -EINVAL;
2855
2856 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2857 return -EINVAL;
2858
2859 if (unlikely(args.set_tid && args.set_tid_size == 0))
2860 return -EINVAL;
2861
2862 /*
2863 * Verify that higher 32bits of exit_signal are unset and that
2864 * it is a valid signal
2865 */
2866 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2867 !valid_signal(args.exit_signal)))
2868 return -EINVAL;
2869
2870 if ((args.flags & CLONE_INTO_CGROUP) &&
2871 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2872 return -EINVAL;
2873
2874 *kargs = (struct kernel_clone_args){
2875 .flags = args.flags,
2876 .pidfd = u64_to_user_ptr(args.pidfd),
2877 .child_tid = u64_to_user_ptr(args.child_tid),
2878 .parent_tid = u64_to_user_ptr(args.parent_tid),
2879 .exit_signal = args.exit_signal,
2880 .stack = args.stack,
2881 .stack_size = args.stack_size,
2882 .tls = args.tls,
2883 .set_tid_size = args.set_tid_size,
2884 .cgroup = args.cgroup,
2885 };
2886
2887 if (args.set_tid &&
2888 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2889 (kargs->set_tid_size * sizeof(pid_t))))
2890 return -EFAULT;
2891
2892 kargs->set_tid = kset_tid;
2893
2894 return 0;
2895}
2896
2897/**
2898 * clone3_stack_valid - check and prepare stack
2899 * @kargs: kernel clone args
2900 *
2901 * Verify that the stack arguments userspace gave us are sane.
2902 * In addition, set the stack direction for userspace since it's easy for us to
2903 * determine.
2904 */
2905static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2906{
2907 if (kargs->stack == 0) {
2908 if (kargs->stack_size > 0)
2909 return false;
2910 } else {
2911 if (kargs->stack_size == 0)
2912 return false;
2913
2914 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2915 return false;
2916
2917#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2918 kargs->stack += kargs->stack_size;
2919#endif
2920 }
2921
2922 return true;
2923}
2924
2925static bool clone3_args_valid(struct kernel_clone_args *kargs)
2926{
2927 /* Verify that no unknown flags are passed along. */
2928 if (kargs->flags &
2929 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2930 return false;
2931
2932 /*
2933 * - make the CLONE_DETACHED bit reusable for clone3
2934 * - make the CSIGNAL bits reusable for clone3
2935 */
2936 if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2937 return false;
2938
2939 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2940 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2941 return false;
2942
2943 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2944 kargs->exit_signal)
2945 return false;
2946
2947 if (!clone3_stack_valid(kargs))
2948 return false;
2949
2950 return true;
2951}
2952
2953/**
2954 * clone3 - create a new process with specific properties
2955 * @uargs: argument structure
2956 * @size: size of @uargs
2957 *
2958 * clone3() is the extensible successor to clone()/clone2().
2959 * It takes a struct as argument that is versioned by its size.
2960 *
2961 * Return: On success, a positive PID for the child process.
2962 * On error, a negative errno number.
2963 */
2964SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2965{
2966 int err;
2967
2968 struct kernel_clone_args kargs;
2969 pid_t set_tid[MAX_PID_NS_LEVEL];
2970
2971 kargs.set_tid = set_tid;
2972
2973 err = copy_clone_args_from_user(&kargs, uargs, size);
2974 if (err)
2975 return err;
2976
2977 if (!clone3_args_valid(&kargs))
2978 return -EINVAL;
2979
2980 return kernel_clone(&kargs);
2981}
2982#endif
2983
2984void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2985{
2986 struct task_struct *leader, *parent, *child;
2987 int res;
2988
2989 read_lock(&tasklist_lock);
2990 leader = top = top->group_leader;
2991down:
2992 for_each_thread(leader, parent) {
2993 list_for_each_entry(child, &parent->children, sibling) {
2994 res = visitor(child, data);
2995 if (res) {
2996 if (res < 0)
2997 goto out;
2998 leader = child;
2999 goto down;
3000 }
3001up:
3002 ;
3003 }
3004 }
3005
3006 if (leader != top) {
3007 child = leader;
3008 parent = child->real_parent;
3009 leader = parent->group_leader;
3010 goto up;
3011 }
3012out:
3013 read_unlock(&tasklist_lock);
3014}
3015
3016#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3017#define ARCH_MIN_MMSTRUCT_ALIGN 0
3018#endif
3019
3020static void sighand_ctor(void *data)
3021{
3022 struct sighand_struct *sighand = data;
3023
3024 spin_lock_init(&sighand->siglock);
3025 init_waitqueue_head(&sighand->signalfd_wqh);
3026}
3027
3028void __init mm_cache_init(void)
3029{
3030 unsigned int mm_size;
3031
3032 /*
3033 * The mm_cpumask is located at the end of mm_struct, and is
3034 * dynamically sized based on the maximum CPU number this system
3035 * can have, taking hotplug into account (nr_cpu_ids).
3036 */
3037 mm_size = sizeof(struct mm_struct) + cpumask_size();
3038
3039 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3040 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3041 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3042 offsetof(struct mm_struct, saved_auxv),
3043 sizeof_field(struct mm_struct, saved_auxv),
3044 NULL);
3045}
3046
3047void __init proc_caches_init(void)
3048{
3049 sighand_cachep = kmem_cache_create("sighand_cache",
3050 sizeof(struct sighand_struct), 0,
3051 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3052 SLAB_ACCOUNT, sighand_ctor);
3053 signal_cachep = kmem_cache_create("signal_cache",
3054 sizeof(struct signal_struct), 0,
3055 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3056 NULL);
3057 files_cachep = kmem_cache_create("files_cache",
3058 sizeof(struct files_struct), 0,
3059 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3060 NULL);
3061 fs_cachep = kmem_cache_create("fs_cache",
3062 sizeof(struct fs_struct), 0,
3063 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3064 NULL);
3065
3066 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3067 mmap_init();
3068 nsproxy_cache_init();
3069}
3070
3071/*
3072 * Check constraints on flags passed to the unshare system call.
3073 */
3074static int check_unshare_flags(unsigned long unshare_flags)
3075{
3076 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3077 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3078 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3079 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3080 CLONE_NEWTIME))
3081 return -EINVAL;
3082 /*
3083 * Not implemented, but pretend it works if there is nothing
3084 * to unshare. Note that unsharing the address space or the
3085 * signal handlers also need to unshare the signal queues (aka
3086 * CLONE_THREAD).
3087 */
3088 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3089 if (!thread_group_empty(current))
3090 return -EINVAL;
3091 }
3092 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3093 if (refcount_read(¤t->sighand->count) > 1)
3094 return -EINVAL;
3095 }
3096 if (unshare_flags & CLONE_VM) {
3097 if (!current_is_single_threaded())
3098 return -EINVAL;
3099 }
3100
3101 return 0;
3102}
3103
3104/*
3105 * Unshare the filesystem structure if it is being shared
3106 */
3107static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3108{
3109 struct fs_struct *fs = current->fs;
3110
3111 if (!(unshare_flags & CLONE_FS) || !fs)
3112 return 0;
3113
3114 /* don't need lock here; in the worst case we'll do useless copy */
3115 if (fs->users == 1)
3116 return 0;
3117
3118 *new_fsp = copy_fs_struct(fs);
3119 if (!*new_fsp)
3120 return -ENOMEM;
3121
3122 return 0;
3123}
3124
3125/*
3126 * Unshare file descriptor table if it is being shared
3127 */
3128int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3129 struct files_struct **new_fdp)
3130{
3131 struct files_struct *fd = current->files;
3132 int error = 0;
3133
3134 if ((unshare_flags & CLONE_FILES) &&
3135 (fd && atomic_read(&fd->count) > 1)) {
3136 *new_fdp = dup_fd(fd, max_fds, &error);
3137 if (!*new_fdp)
3138 return error;
3139 }
3140
3141 return 0;
3142}
3143
3144/*
3145 * unshare allows a process to 'unshare' part of the process
3146 * context which was originally shared using clone. copy_*
3147 * functions used by kernel_clone() cannot be used here directly
3148 * because they modify an inactive task_struct that is being
3149 * constructed. Here we are modifying the current, active,
3150 * task_struct.
3151 */
3152int ksys_unshare(unsigned long unshare_flags)
3153{
3154 struct fs_struct *fs, *new_fs = NULL;
3155 struct files_struct *new_fd = NULL;
3156 struct cred *new_cred = NULL;
3157 struct nsproxy *new_nsproxy = NULL;
3158 int do_sysvsem = 0;
3159 int err;
3160
3161 /*
3162 * If unsharing a user namespace must also unshare the thread group
3163 * and unshare the filesystem root and working directories.
3164 */
3165 if (unshare_flags & CLONE_NEWUSER)
3166 unshare_flags |= CLONE_THREAD | CLONE_FS;
3167 /*
3168 * If unsharing vm, must also unshare signal handlers.
3169 */
3170 if (unshare_flags & CLONE_VM)
3171 unshare_flags |= CLONE_SIGHAND;
3172 /*
3173 * If unsharing a signal handlers, must also unshare the signal queues.
3174 */
3175 if (unshare_flags & CLONE_SIGHAND)
3176 unshare_flags |= CLONE_THREAD;
3177 /*
3178 * If unsharing namespace, must also unshare filesystem information.
3179 */
3180 if (unshare_flags & CLONE_NEWNS)
3181 unshare_flags |= CLONE_FS;
3182
3183 err = check_unshare_flags(unshare_flags);
3184 if (err)
3185 goto bad_unshare_out;
3186 /*
3187 * CLONE_NEWIPC must also detach from the undolist: after switching
3188 * to a new ipc namespace, the semaphore arrays from the old
3189 * namespace are unreachable.
3190 */
3191 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3192 do_sysvsem = 1;
3193 err = unshare_fs(unshare_flags, &new_fs);
3194 if (err)
3195 goto bad_unshare_out;
3196 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3197 if (err)
3198 goto bad_unshare_cleanup_fs;
3199 err = unshare_userns(unshare_flags, &new_cred);
3200 if (err)
3201 goto bad_unshare_cleanup_fd;
3202 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3203 new_cred, new_fs);
3204 if (err)
3205 goto bad_unshare_cleanup_cred;
3206
3207 if (new_cred) {
3208 err = set_cred_ucounts(new_cred);
3209 if (err)
3210 goto bad_unshare_cleanup_cred;
3211 }
3212
3213 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3214 if (do_sysvsem) {
3215 /*
3216 * CLONE_SYSVSEM is equivalent to sys_exit().
3217 */
3218 exit_sem(current);
3219 }
3220 if (unshare_flags & CLONE_NEWIPC) {
3221 /* Orphan segments in old ns (see sem above). */
3222 exit_shm(current);
3223 shm_init_task(current);
3224 }
3225
3226 if (new_nsproxy)
3227 switch_task_namespaces(current, new_nsproxy);
3228
3229 task_lock(current);
3230
3231 if (new_fs) {
3232 fs = current->fs;
3233 spin_lock(&fs->lock);
3234 current->fs = new_fs;
3235 if (--fs->users)
3236 new_fs = NULL;
3237 else
3238 new_fs = fs;
3239 spin_unlock(&fs->lock);
3240 }
3241
3242 if (new_fd)
3243 swap(current->files, new_fd);
3244
3245 task_unlock(current);
3246
3247 if (new_cred) {
3248 /* Install the new user namespace */
3249 commit_creds(new_cred);
3250 new_cred = NULL;
3251 }
3252 }
3253
3254 perf_event_namespaces(current);
3255
3256bad_unshare_cleanup_cred:
3257 if (new_cred)
3258 put_cred(new_cred);
3259bad_unshare_cleanup_fd:
3260 if (new_fd)
3261 put_files_struct(new_fd);
3262
3263bad_unshare_cleanup_fs:
3264 if (new_fs)
3265 free_fs_struct(new_fs);
3266
3267bad_unshare_out:
3268 return err;
3269}
3270
3271SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3272{
3273 return ksys_unshare(unshare_flags);
3274}
3275
3276/*
3277 * Helper to unshare the files of the current task.
3278 * We don't want to expose copy_files internals to
3279 * the exec layer of the kernel.
3280 */
3281
3282int unshare_files(void)
3283{
3284 struct task_struct *task = current;
3285 struct files_struct *old, *copy = NULL;
3286 int error;
3287
3288 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3289 if (error || !copy)
3290 return error;
3291
3292 old = task->files;
3293 task_lock(task);
3294 task->files = copy;
3295 task_unlock(task);
3296 put_files_struct(old);
3297 return 0;
3298}
3299
3300int sysctl_max_threads(struct ctl_table *table, int write,
3301 void *buffer, size_t *lenp, loff_t *ppos)
3302{
3303 struct ctl_table t;
3304 int ret;
3305 int threads = max_threads;
3306 int min = 1;
3307 int max = MAX_THREADS;
3308
3309 t = *table;
3310 t.data = &threads;
3311 t.extra1 = &min;
3312 t.extra2 = &max;
3313
3314 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3315 if (ret || !write)
3316 return ret;
3317
3318 max_threads = threads;
3319
3320 return 0;
3321}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/fork.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * 'fork.c' contains the help-routines for the 'fork' system call
10 * (see also entry.S and others).
11 * Fork is rather simple, once you get the hang of it, but the memory
12 * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13 */
14
15#include <linux/anon_inodes.h>
16#include <linux/slab.h>
17#include <linux/sched/autogroup.h>
18#include <linux/sched/mm.h>
19#include <linux/sched/coredump.h>
20#include <linux/sched/user.h>
21#include <linux/sched/numa_balancing.h>
22#include <linux/sched/stat.h>
23#include <linux/sched/task.h>
24#include <linux/sched/task_stack.h>
25#include <linux/sched/cputime.h>
26#include <linux/seq_file.h>
27#include <linux/rtmutex.h>
28#include <linux/init.h>
29#include <linux/unistd.h>
30#include <linux/module.h>
31#include <linux/vmalloc.h>
32#include <linux/completion.h>
33#include <linux/personality.h>
34#include <linux/mempolicy.h>
35#include <linux/sem.h>
36#include <linux/file.h>
37#include <linux/fdtable.h>
38#include <linux/iocontext.h>
39#include <linux/key.h>
40#include <linux/kmsan.h>
41#include <linux/binfmts.h>
42#include <linux/mman.h>
43#include <linux/mmu_notifier.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/mm_inline.h>
47#include <linux/nsproxy.h>
48#include <linux/capability.h>
49#include <linux/cpu.h>
50#include <linux/cgroup.h>
51#include <linux/security.h>
52#include <linux/hugetlb.h>
53#include <linux/seccomp.h>
54#include <linux/swap.h>
55#include <linux/syscalls.h>
56#include <linux/syscall_user_dispatch.h>
57#include <linux/jiffies.h>
58#include <linux/futex.h>
59#include <linux/compat.h>
60#include <linux/kthread.h>
61#include <linux/task_io_accounting_ops.h>
62#include <linux/rcupdate.h>
63#include <linux/ptrace.h>
64#include <linux/mount.h>
65#include <linux/audit.h>
66#include <linux/memcontrol.h>
67#include <linux/ftrace.h>
68#include <linux/proc_fs.h>
69#include <linux/profile.h>
70#include <linux/rmap.h>
71#include <linux/ksm.h>
72#include <linux/acct.h>
73#include <linux/userfaultfd_k.h>
74#include <linux/tsacct_kern.h>
75#include <linux/cn_proc.h>
76#include <linux/freezer.h>
77#include <linux/delayacct.h>
78#include <linux/taskstats_kern.h>
79#include <linux/tty.h>
80#include <linux/fs_struct.h>
81#include <linux/magic.h>
82#include <linux/perf_event.h>
83#include <linux/posix-timers.h>
84#include <linux/user-return-notifier.h>
85#include <linux/oom.h>
86#include <linux/khugepaged.h>
87#include <linux/signalfd.h>
88#include <linux/uprobes.h>
89#include <linux/aio.h>
90#include <linux/compiler.h>
91#include <linux/sysctl.h>
92#include <linux/kcov.h>
93#include <linux/livepatch.h>
94#include <linux/thread_info.h>
95#include <linux/stackleak.h>
96#include <linux/kasan.h>
97#include <linux/scs.h>
98#include <linux/io_uring.h>
99#include <linux/bpf.h>
100#include <linux/stackprotector.h>
101#include <linux/user_events.h>
102#include <linux/iommu.h>
103#include <linux/rseq.h>
104#include <uapi/linux/pidfd.h>
105#include <linux/pidfs.h>
106
107#include <asm/pgalloc.h>
108#include <linux/uaccess.h>
109#include <asm/mmu_context.h>
110#include <asm/cacheflush.h>
111#include <asm/tlbflush.h>
112
113#include <trace/events/sched.h>
114
115#define CREATE_TRACE_POINTS
116#include <trace/events/task.h>
117
118/*
119 * Minimum number of threads to boot the kernel
120 */
121#define MIN_THREADS 20
122
123/*
124 * Maximum number of threads
125 */
126#define MAX_THREADS FUTEX_TID_MASK
127
128/*
129 * Protected counters by write_lock_irq(&tasklist_lock)
130 */
131unsigned long total_forks; /* Handle normal Linux uptimes. */
132int nr_threads; /* The idle threads do not count.. */
133
134static int max_threads; /* tunable limit on nr_threads */
135
136#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
137
138static const char * const resident_page_types[] = {
139 NAMED_ARRAY_INDEX(MM_FILEPAGES),
140 NAMED_ARRAY_INDEX(MM_ANONPAGES),
141 NAMED_ARRAY_INDEX(MM_SWAPENTS),
142 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
143};
144
145DEFINE_PER_CPU(unsigned long, process_counts) = 0;
146
147__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
148
149#ifdef CONFIG_PROVE_RCU
150int lockdep_tasklist_lock_is_held(void)
151{
152 return lockdep_is_held(&tasklist_lock);
153}
154EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
155#endif /* #ifdef CONFIG_PROVE_RCU */
156
157int nr_processes(void)
158{
159 int cpu;
160 int total = 0;
161
162 for_each_possible_cpu(cpu)
163 total += per_cpu(process_counts, cpu);
164
165 return total;
166}
167
168void __weak arch_release_task_struct(struct task_struct *tsk)
169{
170}
171
172static struct kmem_cache *task_struct_cachep;
173
174static inline struct task_struct *alloc_task_struct_node(int node)
175{
176 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
177}
178
179static inline void free_task_struct(struct task_struct *tsk)
180{
181 kmem_cache_free(task_struct_cachep, tsk);
182}
183
184/*
185 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
186 * kmemcache based allocator.
187 */
188# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189
190# ifdef CONFIG_VMAP_STACK
191/*
192 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
193 * flush. Try to minimize the number of calls by caching stacks.
194 */
195#define NR_CACHED_STACKS 2
196static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
197
198struct vm_stack {
199 struct rcu_head rcu;
200 struct vm_struct *stack_vm_area;
201};
202
203static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
204{
205 unsigned int i;
206
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
209 continue;
210 return true;
211 }
212 return false;
213}
214
215static void thread_stack_free_rcu(struct rcu_head *rh)
216{
217 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218
219 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
220 return;
221
222 vfree(vm_stack);
223}
224
225static void thread_stack_delayed_free(struct task_struct *tsk)
226{
227 struct vm_stack *vm_stack = tsk->stack;
228
229 vm_stack->stack_vm_area = tsk->stack_vm_area;
230 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231}
232
233static int free_vm_stack_cache(unsigned int cpu)
234{
235 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 int i;
237
238 for (i = 0; i < NR_CACHED_STACKS; i++) {
239 struct vm_struct *vm_stack = cached_vm_stacks[i];
240
241 if (!vm_stack)
242 continue;
243
244 vfree(vm_stack->addr);
245 cached_vm_stacks[i] = NULL;
246 }
247
248 return 0;
249}
250
251static int memcg_charge_kernel_stack(struct vm_struct *vm)
252{
253 int i;
254 int ret;
255 int nr_charged = 0;
256
257 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258
259 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
260 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
261 if (ret)
262 goto err;
263 nr_charged++;
264 }
265 return 0;
266err:
267 for (i = 0; i < nr_charged; i++)
268 memcg_kmem_uncharge_page(vm->pages[i], 0);
269 return ret;
270}
271
272static int alloc_thread_stack_node(struct task_struct *tsk, int node)
273{
274 struct vm_struct *vm;
275 void *stack;
276 int i;
277
278 for (i = 0; i < NR_CACHED_STACKS; i++) {
279 struct vm_struct *s;
280
281 s = this_cpu_xchg(cached_stacks[i], NULL);
282
283 if (!s)
284 continue;
285
286 /* Reset stack metadata. */
287 kasan_unpoison_range(s->addr, THREAD_SIZE);
288
289 stack = kasan_reset_tag(s->addr);
290
291 /* Clear stale pointers from reused stack. */
292 memset(stack, 0, THREAD_SIZE);
293
294 if (memcg_charge_kernel_stack(s)) {
295 vfree(s->addr);
296 return -ENOMEM;
297 }
298
299 tsk->stack_vm_area = s;
300 tsk->stack = stack;
301 return 0;
302 }
303
304 /*
305 * Allocated stacks are cached and later reused by new threads,
306 * so memcg accounting is performed manually on assigning/releasing
307 * stacks to tasks. Drop __GFP_ACCOUNT.
308 */
309 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
310 VMALLOC_START, VMALLOC_END,
311 THREADINFO_GFP & ~__GFP_ACCOUNT,
312 PAGE_KERNEL,
313 0, node, __builtin_return_address(0));
314 if (!stack)
315 return -ENOMEM;
316
317 vm = find_vm_area(stack);
318 if (memcg_charge_kernel_stack(vm)) {
319 vfree(stack);
320 return -ENOMEM;
321 }
322 /*
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
326 */
327 tsk->stack_vm_area = vm;
328 stack = kasan_reset_tag(stack);
329 tsk->stack = stack;
330 return 0;
331}
332
333static void free_thread_stack(struct task_struct *tsk)
334{
335 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
337
338 tsk->stack = NULL;
339 tsk->stack_vm_area = NULL;
340}
341
342# else /* !CONFIG_VMAP_STACK */
343
344static void thread_stack_free_rcu(struct rcu_head *rh)
345{
346 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
347}
348
349static void thread_stack_delayed_free(struct task_struct *tsk)
350{
351 struct rcu_head *rh = tsk->stack;
352
353 call_rcu(rh, thread_stack_free_rcu);
354}
355
356static int alloc_thread_stack_node(struct task_struct *tsk, int node)
357{
358 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
359 THREAD_SIZE_ORDER);
360
361 if (likely(page)) {
362 tsk->stack = kasan_reset_tag(page_address(page));
363 return 0;
364 }
365 return -ENOMEM;
366}
367
368static void free_thread_stack(struct task_struct *tsk)
369{
370 thread_stack_delayed_free(tsk);
371 tsk->stack = NULL;
372}
373
374# endif /* CONFIG_VMAP_STACK */
375# else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
376
377static struct kmem_cache *thread_stack_cache;
378
379static void thread_stack_free_rcu(struct rcu_head *rh)
380{
381 kmem_cache_free(thread_stack_cache, rh);
382}
383
384static void thread_stack_delayed_free(struct task_struct *tsk)
385{
386 struct rcu_head *rh = tsk->stack;
387
388 call_rcu(rh, thread_stack_free_rcu);
389}
390
391static int alloc_thread_stack_node(struct task_struct *tsk, int node)
392{
393 unsigned long *stack;
394 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
395 stack = kasan_reset_tag(stack);
396 tsk->stack = stack;
397 return stack ? 0 : -ENOMEM;
398}
399
400static void free_thread_stack(struct task_struct *tsk)
401{
402 thread_stack_delayed_free(tsk);
403 tsk->stack = NULL;
404}
405
406void thread_stack_cache_init(void)
407{
408 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
409 THREAD_SIZE, THREAD_SIZE, 0, 0,
410 THREAD_SIZE, NULL);
411 BUG_ON(thread_stack_cache == NULL);
412}
413
414# endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415
416/* SLAB cache for signal_struct structures (tsk->signal) */
417static struct kmem_cache *signal_cachep;
418
419/* SLAB cache for sighand_struct structures (tsk->sighand) */
420struct kmem_cache *sighand_cachep;
421
422/* SLAB cache for files_struct structures (tsk->files) */
423struct kmem_cache *files_cachep;
424
425/* SLAB cache for fs_struct structures (tsk->fs) */
426struct kmem_cache *fs_cachep;
427
428/* SLAB cache for vm_area_struct structures */
429static struct kmem_cache *vm_area_cachep;
430
431/* SLAB cache for mm_struct structures (tsk->mm) */
432static struct kmem_cache *mm_cachep;
433
434#ifdef CONFIG_PER_VMA_LOCK
435
436/* SLAB cache for vm_area_struct.lock */
437static struct kmem_cache *vma_lock_cachep;
438
439static bool vma_lock_alloc(struct vm_area_struct *vma)
440{
441 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
442 if (!vma->vm_lock)
443 return false;
444
445 init_rwsem(&vma->vm_lock->lock);
446 vma->vm_lock_seq = -1;
447
448 return true;
449}
450
451static inline void vma_lock_free(struct vm_area_struct *vma)
452{
453 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
454}
455
456#else /* CONFIG_PER_VMA_LOCK */
457
458static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
459static inline void vma_lock_free(struct vm_area_struct *vma) {}
460
461#endif /* CONFIG_PER_VMA_LOCK */
462
463struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
464{
465 struct vm_area_struct *vma;
466
467 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
468 if (!vma)
469 return NULL;
470
471 vma_init(vma, mm);
472 if (!vma_lock_alloc(vma)) {
473 kmem_cache_free(vm_area_cachep, vma);
474 return NULL;
475 }
476
477 return vma;
478}
479
480struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
481{
482 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
483
484 if (!new)
485 return NULL;
486
487 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
488 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
489 /*
490 * orig->shared.rb may be modified concurrently, but the clone
491 * will be reinitialized.
492 */
493 data_race(memcpy(new, orig, sizeof(*new)));
494 if (!vma_lock_alloc(new)) {
495 kmem_cache_free(vm_area_cachep, new);
496 return NULL;
497 }
498 INIT_LIST_HEAD(&new->anon_vma_chain);
499 vma_numab_state_init(new);
500 dup_anon_vma_name(orig, new);
501
502 return new;
503}
504
505void __vm_area_free(struct vm_area_struct *vma)
506{
507 vma_numab_state_free(vma);
508 free_anon_vma_name(vma);
509 vma_lock_free(vma);
510 kmem_cache_free(vm_area_cachep, vma);
511}
512
513#ifdef CONFIG_PER_VMA_LOCK
514static void vm_area_free_rcu_cb(struct rcu_head *head)
515{
516 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
517 vm_rcu);
518
519 /* The vma should not be locked while being destroyed. */
520 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
521 __vm_area_free(vma);
522}
523#endif
524
525void vm_area_free(struct vm_area_struct *vma)
526{
527#ifdef CONFIG_PER_VMA_LOCK
528 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
529#else
530 __vm_area_free(vma);
531#endif
532}
533
534static void account_kernel_stack(struct task_struct *tsk, int account)
535{
536 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
537 struct vm_struct *vm = task_stack_vm_area(tsk);
538 int i;
539
540 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
541 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
542 account * (PAGE_SIZE / 1024));
543 } else {
544 void *stack = task_stack_page(tsk);
545
546 /* All stack pages are in the same node. */
547 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
548 account * (THREAD_SIZE / 1024));
549 }
550}
551
552void exit_task_stack_account(struct task_struct *tsk)
553{
554 account_kernel_stack(tsk, -1);
555
556 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
557 struct vm_struct *vm;
558 int i;
559
560 vm = task_stack_vm_area(tsk);
561 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
562 memcg_kmem_uncharge_page(vm->pages[i], 0);
563 }
564}
565
566static void release_task_stack(struct task_struct *tsk)
567{
568 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
569 return; /* Better to leak the stack than to free prematurely */
570
571 free_thread_stack(tsk);
572}
573
574#ifdef CONFIG_THREAD_INFO_IN_TASK
575void put_task_stack(struct task_struct *tsk)
576{
577 if (refcount_dec_and_test(&tsk->stack_refcount))
578 release_task_stack(tsk);
579}
580#endif
581
582void free_task(struct task_struct *tsk)
583{
584#ifdef CONFIG_SECCOMP
585 WARN_ON_ONCE(tsk->seccomp.filter);
586#endif
587 release_user_cpus_ptr(tsk);
588 scs_release(tsk);
589
590#ifndef CONFIG_THREAD_INFO_IN_TASK
591 /*
592 * The task is finally done with both the stack and thread_info,
593 * so free both.
594 */
595 release_task_stack(tsk);
596#else
597 /*
598 * If the task had a separate stack allocation, it should be gone
599 * by now.
600 */
601 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
602#endif
603 rt_mutex_debug_task_free(tsk);
604 ftrace_graph_exit_task(tsk);
605 arch_release_task_struct(tsk);
606 if (tsk->flags & PF_KTHREAD)
607 free_kthread_struct(tsk);
608 bpf_task_storage_free(tsk);
609 free_task_struct(tsk);
610}
611EXPORT_SYMBOL(free_task);
612
613static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
614{
615 struct file *exe_file;
616
617 exe_file = get_mm_exe_file(oldmm);
618 RCU_INIT_POINTER(mm->exe_file, exe_file);
619 /*
620 * We depend on the oldmm having properly denied write access to the
621 * exe_file already.
622 */
623 if (exe_file && deny_write_access(exe_file))
624 pr_warn_once("deny_write_access() failed in %s\n", __func__);
625}
626
627#ifdef CONFIG_MMU
628static __latent_entropy int dup_mmap(struct mm_struct *mm,
629 struct mm_struct *oldmm)
630{
631 struct vm_area_struct *mpnt, *tmp;
632 int retval;
633 unsigned long charge = 0;
634 LIST_HEAD(uf);
635 VMA_ITERATOR(vmi, mm, 0);
636
637 uprobe_start_dup_mmap();
638 if (mmap_write_lock_killable(oldmm)) {
639 retval = -EINTR;
640 goto fail_uprobe_end;
641 }
642 flush_cache_dup_mm(oldmm);
643 uprobe_dup_mmap(oldmm, mm);
644 /*
645 * Not linked in yet - no deadlock potential:
646 */
647 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
648
649 /* No ordering required: file already has been exposed. */
650 dup_mm_exe_file(mm, oldmm);
651
652 mm->total_vm = oldmm->total_vm;
653 mm->data_vm = oldmm->data_vm;
654 mm->exec_vm = oldmm->exec_vm;
655 mm->stack_vm = oldmm->stack_vm;
656
657 retval = ksm_fork(mm, oldmm);
658 if (retval)
659 goto out;
660 khugepaged_fork(mm, oldmm);
661
662 /* Use __mt_dup() to efficiently build an identical maple tree. */
663 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
664 if (unlikely(retval))
665 goto out;
666
667 mt_clear_in_rcu(vmi.mas.tree);
668 for_each_vma(vmi, mpnt) {
669 struct file *file;
670
671 vma_start_write(mpnt);
672 if (mpnt->vm_flags & VM_DONTCOPY) {
673 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
674 mpnt->vm_end, GFP_KERNEL);
675 if (retval)
676 goto loop_out;
677
678 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
679 continue;
680 }
681 charge = 0;
682 /*
683 * Don't duplicate many vmas if we've been oom-killed (for
684 * example)
685 */
686 if (fatal_signal_pending(current)) {
687 retval = -EINTR;
688 goto loop_out;
689 }
690 if (mpnt->vm_flags & VM_ACCOUNT) {
691 unsigned long len = vma_pages(mpnt);
692
693 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
694 goto fail_nomem;
695 charge = len;
696 }
697 tmp = vm_area_dup(mpnt);
698 if (!tmp)
699 goto fail_nomem;
700 retval = vma_dup_policy(mpnt, tmp);
701 if (retval)
702 goto fail_nomem_policy;
703 tmp->vm_mm = mm;
704 retval = dup_userfaultfd(tmp, &uf);
705 if (retval)
706 goto fail_nomem_anon_vma_fork;
707 if (tmp->vm_flags & VM_WIPEONFORK) {
708 /*
709 * VM_WIPEONFORK gets a clean slate in the child.
710 * Don't prepare anon_vma until fault since we don't
711 * copy page for current vma.
712 */
713 tmp->anon_vma = NULL;
714 } else if (anon_vma_fork(tmp, mpnt))
715 goto fail_nomem_anon_vma_fork;
716 vm_flags_clear(tmp, VM_LOCKED_MASK);
717 /*
718 * Copy/update hugetlb private vma information.
719 */
720 if (is_vm_hugetlb_page(tmp))
721 hugetlb_dup_vma_private(tmp);
722
723 /*
724 * Link the vma into the MT. After using __mt_dup(), memory
725 * allocation is not necessary here, so it cannot fail.
726 */
727 vma_iter_bulk_store(&vmi, tmp);
728
729 mm->map_count++;
730
731 if (tmp->vm_ops && tmp->vm_ops->open)
732 tmp->vm_ops->open(tmp);
733
734 file = tmp->vm_file;
735 if (file) {
736 struct address_space *mapping = file->f_mapping;
737
738 get_file(file);
739 i_mmap_lock_write(mapping);
740 if (vma_is_shared_maywrite(tmp))
741 mapping_allow_writable(mapping);
742 flush_dcache_mmap_lock(mapping);
743 /* insert tmp into the share list, just after mpnt */
744 vma_interval_tree_insert_after(tmp, mpnt,
745 &mapping->i_mmap);
746 flush_dcache_mmap_unlock(mapping);
747 i_mmap_unlock_write(mapping);
748 }
749
750 if (!(tmp->vm_flags & VM_WIPEONFORK))
751 retval = copy_page_range(tmp, mpnt);
752
753 if (retval) {
754 mpnt = vma_next(&vmi);
755 goto loop_out;
756 }
757 }
758 /* a new mm has just been created */
759 retval = arch_dup_mmap(oldmm, mm);
760loop_out:
761 vma_iter_free(&vmi);
762 if (!retval) {
763 mt_set_in_rcu(vmi.mas.tree);
764 } else if (mpnt) {
765 /*
766 * The entire maple tree has already been duplicated. If the
767 * mmap duplication fails, mark the failure point with
768 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
769 * stop releasing VMAs that have not been duplicated after this
770 * point.
771 */
772 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
773 mas_store(&vmi.mas, XA_ZERO_ENTRY);
774 }
775out:
776 mmap_write_unlock(mm);
777 flush_tlb_mm(oldmm);
778 mmap_write_unlock(oldmm);
779 dup_userfaultfd_complete(&uf);
780fail_uprobe_end:
781 uprobe_end_dup_mmap();
782 return retval;
783
784fail_nomem_anon_vma_fork:
785 mpol_put(vma_policy(tmp));
786fail_nomem_policy:
787 vm_area_free(tmp);
788fail_nomem:
789 retval = -ENOMEM;
790 vm_unacct_memory(charge);
791 goto loop_out;
792}
793
794static inline int mm_alloc_pgd(struct mm_struct *mm)
795{
796 mm->pgd = pgd_alloc(mm);
797 if (unlikely(!mm->pgd))
798 return -ENOMEM;
799 return 0;
800}
801
802static inline void mm_free_pgd(struct mm_struct *mm)
803{
804 pgd_free(mm, mm->pgd);
805}
806#else
807static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
808{
809 mmap_write_lock(oldmm);
810 dup_mm_exe_file(mm, oldmm);
811 mmap_write_unlock(oldmm);
812 return 0;
813}
814#define mm_alloc_pgd(mm) (0)
815#define mm_free_pgd(mm)
816#endif /* CONFIG_MMU */
817
818static void check_mm(struct mm_struct *mm)
819{
820 int i;
821
822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
823 "Please make sure 'struct resident_page_types[]' is updated as well");
824
825 for (i = 0; i < NR_MM_COUNTERS; i++) {
826 long x = percpu_counter_sum(&mm->rss_stat[i]);
827
828 if (unlikely(x))
829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
830 mm, resident_page_types[i], x);
831 }
832
833 if (mm_pgtables_bytes(mm))
834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
835 mm_pgtables_bytes(mm));
836
837#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
839#endif
840}
841
842#define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
843#define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
844
845static void do_check_lazy_tlb(void *arg)
846{
847 struct mm_struct *mm = arg;
848
849 WARN_ON_ONCE(current->active_mm == mm);
850}
851
852static void do_shoot_lazy_tlb(void *arg)
853{
854 struct mm_struct *mm = arg;
855
856 if (current->active_mm == mm) {
857 WARN_ON_ONCE(current->mm);
858 current->active_mm = &init_mm;
859 switch_mm(mm, &init_mm, current);
860 }
861}
862
863static void cleanup_lazy_tlbs(struct mm_struct *mm)
864{
865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
866 /*
867 * In this case, lazy tlb mms are refounted and would not reach
868 * __mmdrop until all CPUs have switched away and mmdrop()ed.
869 */
870 return;
871 }
872
873 /*
874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
875 * requires lazy mm users to switch to another mm when the refcount
876 * drops to zero, before the mm is freed. This requires IPIs here to
877 * switch kernel threads to init_mm.
878 *
879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
880 * switch with the final userspace teardown TLB flush which leaves the
881 * mm lazy on this CPU but no others, reducing the need for additional
882 * IPIs here. There are cases where a final IPI is still required here,
883 * such as the final mmdrop being performed on a different CPU than the
884 * one exiting, or kernel threads using the mm when userspace exits.
885 *
886 * IPI overheads have not found to be expensive, but they could be
887 * reduced in a number of possible ways, for example (roughly
888 * increasing order of complexity):
889 * - The last lazy reference created by exit_mm() could instead switch
890 * to init_mm, however it's probable this will run on the same CPU
891 * immediately afterwards, so this may not reduce IPIs much.
892 * - A batch of mms requiring IPIs could be gathered and freed at once.
893 * - CPUs store active_mm where it can be remotely checked without a
894 * lock, to filter out false-positives in the cpumask.
895 * - After mm_users or mm_count reaches zero, switching away from the
896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
897 * with some batching or delaying of the final IPIs.
898 * - A delayed freeing and RCU-like quiescing sequence based on mm
899 * switching to avoid IPIs completely.
900 */
901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
904}
905
906/*
907 * Called when the last reference to the mm
908 * is dropped: either by a lazy thread or by
909 * mmput. Free the page directory and the mm.
910 */
911void __mmdrop(struct mm_struct *mm)
912{
913 BUG_ON(mm == &init_mm);
914 WARN_ON_ONCE(mm == current->mm);
915
916 /* Ensure no CPUs are using this as their lazy tlb mm */
917 cleanup_lazy_tlbs(mm);
918
919 WARN_ON_ONCE(mm == current->active_mm);
920 mm_free_pgd(mm);
921 destroy_context(mm);
922 mmu_notifier_subscriptions_destroy(mm);
923 check_mm(mm);
924 put_user_ns(mm->user_ns);
925 mm_pasid_drop(mm);
926 mm_destroy_cid(mm);
927 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
928
929 free_mm(mm);
930}
931EXPORT_SYMBOL_GPL(__mmdrop);
932
933static void mmdrop_async_fn(struct work_struct *work)
934{
935 struct mm_struct *mm;
936
937 mm = container_of(work, struct mm_struct, async_put_work);
938 __mmdrop(mm);
939}
940
941static void mmdrop_async(struct mm_struct *mm)
942{
943 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
944 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
945 schedule_work(&mm->async_put_work);
946 }
947}
948
949static inline void free_signal_struct(struct signal_struct *sig)
950{
951 taskstats_tgid_free(sig);
952 sched_autogroup_exit(sig);
953 /*
954 * __mmdrop is not safe to call from softirq context on x86 due to
955 * pgd_dtor so postpone it to the async context
956 */
957 if (sig->oom_mm)
958 mmdrop_async(sig->oom_mm);
959 kmem_cache_free(signal_cachep, sig);
960}
961
962static inline void put_signal_struct(struct signal_struct *sig)
963{
964 if (refcount_dec_and_test(&sig->sigcnt))
965 free_signal_struct(sig);
966}
967
968void __put_task_struct(struct task_struct *tsk)
969{
970 WARN_ON(!tsk->exit_state);
971 WARN_ON(refcount_read(&tsk->usage));
972 WARN_ON(tsk == current);
973
974 io_uring_free(tsk);
975 cgroup_free(tsk);
976 task_numa_free(tsk, true);
977 security_task_free(tsk);
978 exit_creds(tsk);
979 delayacct_tsk_free(tsk);
980 put_signal_struct(tsk->signal);
981 sched_core_free(tsk);
982 free_task(tsk);
983}
984EXPORT_SYMBOL_GPL(__put_task_struct);
985
986void __put_task_struct_rcu_cb(struct rcu_head *rhp)
987{
988 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
989
990 __put_task_struct(task);
991}
992EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
993
994void __init __weak arch_task_cache_init(void) { }
995
996/*
997 * set_max_threads
998 */
999static void set_max_threads(unsigned int max_threads_suggested)
1000{
1001 u64 threads;
1002 unsigned long nr_pages = totalram_pages();
1003
1004 /*
1005 * The number of threads shall be limited such that the thread
1006 * structures may only consume a small part of the available memory.
1007 */
1008 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1009 threads = MAX_THREADS;
1010 else
1011 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1012 (u64) THREAD_SIZE * 8UL);
1013
1014 if (threads > max_threads_suggested)
1015 threads = max_threads_suggested;
1016
1017 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1018}
1019
1020#ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1021/* Initialized by the architecture: */
1022int arch_task_struct_size __read_mostly;
1023#endif
1024
1025static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1026{
1027 /* Fetch thread_struct whitelist for the architecture. */
1028 arch_thread_struct_whitelist(offset, size);
1029
1030 /*
1031 * Handle zero-sized whitelist or empty thread_struct, otherwise
1032 * adjust offset to position of thread_struct in task_struct.
1033 */
1034 if (unlikely(*size == 0))
1035 *offset = 0;
1036 else
1037 *offset += offsetof(struct task_struct, thread);
1038}
1039
1040void __init fork_init(void)
1041{
1042 int i;
1043#ifndef ARCH_MIN_TASKALIGN
1044#define ARCH_MIN_TASKALIGN 0
1045#endif
1046 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1047 unsigned long useroffset, usersize;
1048
1049 /* create a slab on which task_structs can be allocated */
1050 task_struct_whitelist(&useroffset, &usersize);
1051 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1052 arch_task_struct_size, align,
1053 SLAB_PANIC|SLAB_ACCOUNT,
1054 useroffset, usersize, NULL);
1055
1056 /* do the arch specific task caches init */
1057 arch_task_cache_init();
1058
1059 set_max_threads(MAX_THREADS);
1060
1061 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1064 init_task.signal->rlim[RLIMIT_NPROC];
1065
1066 for (i = 0; i < UCOUNT_COUNTS; i++)
1067 init_user_ns.ucount_max[i] = max_threads/2;
1068
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1073
1074#ifdef CONFIG_VMAP_STACK
1075 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1076 NULL, free_vm_stack_cache);
1077#endif
1078
1079 scs_init();
1080
1081 lockdep_init_task(&init_task);
1082 uprobes_init();
1083}
1084
1085int __weak arch_dup_task_struct(struct task_struct *dst,
1086 struct task_struct *src)
1087{
1088 *dst = *src;
1089 return 0;
1090}
1091
1092void set_task_stack_end_magic(struct task_struct *tsk)
1093{
1094 unsigned long *stackend;
1095
1096 stackend = end_of_stack(tsk);
1097 *stackend = STACK_END_MAGIC; /* for overflow detection */
1098}
1099
1100static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1101{
1102 struct task_struct *tsk;
1103 int err;
1104
1105 if (node == NUMA_NO_NODE)
1106 node = tsk_fork_get_node(orig);
1107 tsk = alloc_task_struct_node(node);
1108 if (!tsk)
1109 return NULL;
1110
1111 err = arch_dup_task_struct(tsk, orig);
1112 if (err)
1113 goto free_tsk;
1114
1115 err = alloc_thread_stack_node(tsk, node);
1116 if (err)
1117 goto free_tsk;
1118
1119#ifdef CONFIG_THREAD_INFO_IN_TASK
1120 refcount_set(&tsk->stack_refcount, 1);
1121#endif
1122 account_kernel_stack(tsk, 1);
1123
1124 err = scs_prepare(tsk, node);
1125 if (err)
1126 goto free_stack;
1127
1128#ifdef CONFIG_SECCOMP
1129 /*
1130 * We must handle setting up seccomp filters once we're under
1131 * the sighand lock in case orig has changed between now and
1132 * then. Until then, filter must be NULL to avoid messing up
1133 * the usage counts on the error path calling free_task.
1134 */
1135 tsk->seccomp.filter = NULL;
1136#endif
1137
1138 setup_thread_stack(tsk, orig);
1139 clear_user_return_notifier(tsk);
1140 clear_tsk_need_resched(tsk);
1141 set_task_stack_end_magic(tsk);
1142 clear_syscall_work_syscall_user_dispatch(tsk);
1143
1144#ifdef CONFIG_STACKPROTECTOR
1145 tsk->stack_canary = get_random_canary();
1146#endif
1147 if (orig->cpus_ptr == &orig->cpus_mask)
1148 tsk->cpus_ptr = &tsk->cpus_mask;
1149 dup_user_cpus_ptr(tsk, orig, node);
1150
1151 /*
1152 * One for the user space visible state that goes away when reaped.
1153 * One for the scheduler.
1154 */
1155 refcount_set(&tsk->rcu_users, 2);
1156 /* One for the rcu users */
1157 refcount_set(&tsk->usage, 1);
1158#ifdef CONFIG_BLK_DEV_IO_TRACE
1159 tsk->btrace_seq = 0;
1160#endif
1161 tsk->splice_pipe = NULL;
1162 tsk->task_frag.page = NULL;
1163 tsk->wake_q.next = NULL;
1164 tsk->worker_private = NULL;
1165
1166 kcov_task_init(tsk);
1167 kmsan_task_create(tsk);
1168 kmap_local_fork(tsk);
1169
1170#ifdef CONFIG_FAULT_INJECTION
1171 tsk->fail_nth = 0;
1172#endif
1173
1174#ifdef CONFIG_BLK_CGROUP
1175 tsk->throttle_disk = NULL;
1176 tsk->use_memdelay = 0;
1177#endif
1178
1179#ifdef CONFIG_ARCH_HAS_CPU_PASID
1180 tsk->pasid_activated = 0;
1181#endif
1182
1183#ifdef CONFIG_MEMCG
1184 tsk->active_memcg = NULL;
1185#endif
1186
1187#ifdef CONFIG_CPU_SUP_INTEL
1188 tsk->reported_split_lock = 0;
1189#endif
1190
1191#ifdef CONFIG_SCHED_MM_CID
1192 tsk->mm_cid = -1;
1193 tsk->last_mm_cid = -1;
1194 tsk->mm_cid_active = 0;
1195 tsk->migrate_from_cpu = -1;
1196#endif
1197 return tsk;
1198
1199free_stack:
1200 exit_task_stack_account(tsk);
1201 free_thread_stack(tsk);
1202free_tsk:
1203 free_task_struct(tsk);
1204 return NULL;
1205}
1206
1207__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1208
1209static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1210
1211static int __init coredump_filter_setup(char *s)
1212{
1213 default_dump_filter =
1214 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1215 MMF_DUMP_FILTER_MASK;
1216 return 1;
1217}
1218
1219__setup("coredump_filter=", coredump_filter_setup);
1220
1221#include <linux/init_task.h>
1222
1223static void mm_init_aio(struct mm_struct *mm)
1224{
1225#ifdef CONFIG_AIO
1226 spin_lock_init(&mm->ioctx_lock);
1227 mm->ioctx_table = NULL;
1228#endif
1229}
1230
1231static __always_inline void mm_clear_owner(struct mm_struct *mm,
1232 struct task_struct *p)
1233{
1234#ifdef CONFIG_MEMCG
1235 if (mm->owner == p)
1236 WRITE_ONCE(mm->owner, NULL);
1237#endif
1238}
1239
1240static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1241{
1242#ifdef CONFIG_MEMCG
1243 mm->owner = p;
1244#endif
1245}
1246
1247static void mm_init_uprobes_state(struct mm_struct *mm)
1248{
1249#ifdef CONFIG_UPROBES
1250 mm->uprobes_state.xol_area = NULL;
1251#endif
1252}
1253
1254static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1255 struct user_namespace *user_ns)
1256{
1257 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1258 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1259 atomic_set(&mm->mm_users, 1);
1260 atomic_set(&mm->mm_count, 1);
1261 seqcount_init(&mm->write_protect_seq);
1262 mmap_init_lock(mm);
1263 INIT_LIST_HEAD(&mm->mmlist);
1264#ifdef CONFIG_PER_VMA_LOCK
1265 mm->mm_lock_seq = 0;
1266#endif
1267 mm_pgtables_bytes_init(mm);
1268 mm->map_count = 0;
1269 mm->locked_vm = 0;
1270 atomic64_set(&mm->pinned_vm, 0);
1271 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1272 spin_lock_init(&mm->page_table_lock);
1273 spin_lock_init(&mm->arg_lock);
1274 mm_init_cpumask(mm);
1275 mm_init_aio(mm);
1276 mm_init_owner(mm, p);
1277 mm_pasid_init(mm);
1278 RCU_INIT_POINTER(mm->exe_file, NULL);
1279 mmu_notifier_subscriptions_init(mm);
1280 init_tlb_flush_pending(mm);
1281#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1282 mm->pmd_huge_pte = NULL;
1283#endif
1284 mm_init_uprobes_state(mm);
1285 hugetlb_count_init(mm);
1286
1287 if (current->mm) {
1288 mm->flags = mmf_init_flags(current->mm->flags);
1289 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1290 } else {
1291 mm->flags = default_dump_filter;
1292 mm->def_flags = 0;
1293 }
1294
1295 if (mm_alloc_pgd(mm))
1296 goto fail_nopgd;
1297
1298 if (init_new_context(p, mm))
1299 goto fail_nocontext;
1300
1301 if (mm_alloc_cid(mm))
1302 goto fail_cid;
1303
1304 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1305 NR_MM_COUNTERS))
1306 goto fail_pcpu;
1307
1308 mm->user_ns = get_user_ns(user_ns);
1309 lru_gen_init_mm(mm);
1310 return mm;
1311
1312fail_pcpu:
1313 mm_destroy_cid(mm);
1314fail_cid:
1315 destroy_context(mm);
1316fail_nocontext:
1317 mm_free_pgd(mm);
1318fail_nopgd:
1319 free_mm(mm);
1320 return NULL;
1321}
1322
1323/*
1324 * Allocate and initialize an mm_struct.
1325 */
1326struct mm_struct *mm_alloc(void)
1327{
1328 struct mm_struct *mm;
1329
1330 mm = allocate_mm();
1331 if (!mm)
1332 return NULL;
1333
1334 memset(mm, 0, sizeof(*mm));
1335 return mm_init(mm, current, current_user_ns());
1336}
1337
1338static inline void __mmput(struct mm_struct *mm)
1339{
1340 VM_BUG_ON(atomic_read(&mm->mm_users));
1341
1342 uprobe_clear_state(mm);
1343 exit_aio(mm);
1344 ksm_exit(mm);
1345 khugepaged_exit(mm); /* must run before exit_mmap */
1346 exit_mmap(mm);
1347 mm_put_huge_zero_page(mm);
1348 set_mm_exe_file(mm, NULL);
1349 if (!list_empty(&mm->mmlist)) {
1350 spin_lock(&mmlist_lock);
1351 list_del(&mm->mmlist);
1352 spin_unlock(&mmlist_lock);
1353 }
1354 if (mm->binfmt)
1355 module_put(mm->binfmt->module);
1356 lru_gen_del_mm(mm);
1357 mmdrop(mm);
1358}
1359
1360/*
1361 * Decrement the use count and release all resources for an mm.
1362 */
1363void mmput(struct mm_struct *mm)
1364{
1365 might_sleep();
1366
1367 if (atomic_dec_and_test(&mm->mm_users))
1368 __mmput(mm);
1369}
1370EXPORT_SYMBOL_GPL(mmput);
1371
1372#ifdef CONFIG_MMU
1373static void mmput_async_fn(struct work_struct *work)
1374{
1375 struct mm_struct *mm = container_of(work, struct mm_struct,
1376 async_put_work);
1377
1378 __mmput(mm);
1379}
1380
1381void mmput_async(struct mm_struct *mm)
1382{
1383 if (atomic_dec_and_test(&mm->mm_users)) {
1384 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1385 schedule_work(&mm->async_put_work);
1386 }
1387}
1388EXPORT_SYMBOL_GPL(mmput_async);
1389#endif
1390
1391/**
1392 * set_mm_exe_file - change a reference to the mm's executable file
1393 * @mm: The mm to change.
1394 * @new_exe_file: The new file to use.
1395 *
1396 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1397 *
1398 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1399 * invocations: in mmput() nobody alive left, in execve it happens before
1400 * the new mm is made visible to anyone.
1401 *
1402 * Can only fail if new_exe_file != NULL.
1403 */
1404int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1405{
1406 struct file *old_exe_file;
1407
1408 /*
1409 * It is safe to dereference the exe_file without RCU as
1410 * this function is only called if nobody else can access
1411 * this mm -- see comment above for justification.
1412 */
1413 old_exe_file = rcu_dereference_raw(mm->exe_file);
1414
1415 if (new_exe_file) {
1416 /*
1417 * We expect the caller (i.e., sys_execve) to already denied
1418 * write access, so this is unlikely to fail.
1419 */
1420 if (unlikely(deny_write_access(new_exe_file)))
1421 return -EACCES;
1422 get_file(new_exe_file);
1423 }
1424 rcu_assign_pointer(mm->exe_file, new_exe_file);
1425 if (old_exe_file) {
1426 allow_write_access(old_exe_file);
1427 fput(old_exe_file);
1428 }
1429 return 0;
1430}
1431
1432/**
1433 * replace_mm_exe_file - replace a reference to the mm's executable file
1434 * @mm: The mm to change.
1435 * @new_exe_file: The new file to use.
1436 *
1437 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1438 *
1439 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1440 */
1441int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1442{
1443 struct vm_area_struct *vma;
1444 struct file *old_exe_file;
1445 int ret = 0;
1446
1447 /* Forbid mm->exe_file change if old file still mapped. */
1448 old_exe_file = get_mm_exe_file(mm);
1449 if (old_exe_file) {
1450 VMA_ITERATOR(vmi, mm, 0);
1451 mmap_read_lock(mm);
1452 for_each_vma(vmi, vma) {
1453 if (!vma->vm_file)
1454 continue;
1455 if (path_equal(&vma->vm_file->f_path,
1456 &old_exe_file->f_path)) {
1457 ret = -EBUSY;
1458 break;
1459 }
1460 }
1461 mmap_read_unlock(mm);
1462 fput(old_exe_file);
1463 if (ret)
1464 return ret;
1465 }
1466
1467 ret = deny_write_access(new_exe_file);
1468 if (ret)
1469 return -EACCES;
1470 get_file(new_exe_file);
1471
1472 /* set the new file */
1473 mmap_write_lock(mm);
1474 old_exe_file = rcu_dereference_raw(mm->exe_file);
1475 rcu_assign_pointer(mm->exe_file, new_exe_file);
1476 mmap_write_unlock(mm);
1477
1478 if (old_exe_file) {
1479 allow_write_access(old_exe_file);
1480 fput(old_exe_file);
1481 }
1482 return 0;
1483}
1484
1485/**
1486 * get_mm_exe_file - acquire a reference to the mm's executable file
1487 * @mm: The mm of interest.
1488 *
1489 * Returns %NULL if mm has no associated executable file.
1490 * User must release file via fput().
1491 */
1492struct file *get_mm_exe_file(struct mm_struct *mm)
1493{
1494 struct file *exe_file;
1495
1496 rcu_read_lock();
1497 exe_file = get_file_rcu(&mm->exe_file);
1498 rcu_read_unlock();
1499 return exe_file;
1500}
1501
1502/**
1503 * get_task_exe_file - acquire a reference to the task's executable file
1504 * @task: The task.
1505 *
1506 * Returns %NULL if task's mm (if any) has no associated executable file or
1507 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1508 * User must release file via fput().
1509 */
1510struct file *get_task_exe_file(struct task_struct *task)
1511{
1512 struct file *exe_file = NULL;
1513 struct mm_struct *mm;
1514
1515 task_lock(task);
1516 mm = task->mm;
1517 if (mm) {
1518 if (!(task->flags & PF_KTHREAD))
1519 exe_file = get_mm_exe_file(mm);
1520 }
1521 task_unlock(task);
1522 return exe_file;
1523}
1524
1525/**
1526 * get_task_mm - acquire a reference to the task's mm
1527 * @task: The task.
1528 *
1529 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1530 * this kernel workthread has transiently adopted a user mm with use_mm,
1531 * to do its AIO) is not set and if so returns a reference to it, after
1532 * bumping up the use count. User must release the mm via mmput()
1533 * after use. Typically used by /proc and ptrace.
1534 */
1535struct mm_struct *get_task_mm(struct task_struct *task)
1536{
1537 struct mm_struct *mm;
1538
1539 task_lock(task);
1540 mm = task->mm;
1541 if (mm) {
1542 if (task->flags & PF_KTHREAD)
1543 mm = NULL;
1544 else
1545 mmget(mm);
1546 }
1547 task_unlock(task);
1548 return mm;
1549}
1550EXPORT_SYMBOL_GPL(get_task_mm);
1551
1552struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1553{
1554 struct mm_struct *mm;
1555 int err;
1556
1557 err = down_read_killable(&task->signal->exec_update_lock);
1558 if (err)
1559 return ERR_PTR(err);
1560
1561 mm = get_task_mm(task);
1562 if (mm && mm != current->mm &&
1563 !ptrace_may_access(task, mode)) {
1564 mmput(mm);
1565 mm = ERR_PTR(-EACCES);
1566 }
1567 up_read(&task->signal->exec_update_lock);
1568
1569 return mm;
1570}
1571
1572static void complete_vfork_done(struct task_struct *tsk)
1573{
1574 struct completion *vfork;
1575
1576 task_lock(tsk);
1577 vfork = tsk->vfork_done;
1578 if (likely(vfork)) {
1579 tsk->vfork_done = NULL;
1580 complete(vfork);
1581 }
1582 task_unlock(tsk);
1583}
1584
1585static int wait_for_vfork_done(struct task_struct *child,
1586 struct completion *vfork)
1587{
1588 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1589 int killed;
1590
1591 cgroup_enter_frozen();
1592 killed = wait_for_completion_state(vfork, state);
1593 cgroup_leave_frozen(false);
1594
1595 if (killed) {
1596 task_lock(child);
1597 child->vfork_done = NULL;
1598 task_unlock(child);
1599 }
1600
1601 put_task_struct(child);
1602 return killed;
1603}
1604
1605/* Please note the differences between mmput and mm_release.
1606 * mmput is called whenever we stop holding onto a mm_struct,
1607 * error success whatever.
1608 *
1609 * mm_release is called after a mm_struct has been removed
1610 * from the current process.
1611 *
1612 * This difference is important for error handling, when we
1613 * only half set up a mm_struct for a new process and need to restore
1614 * the old one. Because we mmput the new mm_struct before
1615 * restoring the old one. . .
1616 * Eric Biederman 10 January 1998
1617 */
1618static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1619{
1620 uprobe_free_utask(tsk);
1621
1622 /* Get rid of any cached register state */
1623 deactivate_mm(tsk, mm);
1624
1625 /*
1626 * Signal userspace if we're not exiting with a core dump
1627 * because we want to leave the value intact for debugging
1628 * purposes.
1629 */
1630 if (tsk->clear_child_tid) {
1631 if (atomic_read(&mm->mm_users) > 1) {
1632 /*
1633 * We don't check the error code - if userspace has
1634 * not set up a proper pointer then tough luck.
1635 */
1636 put_user(0, tsk->clear_child_tid);
1637 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1638 1, NULL, NULL, 0, 0);
1639 }
1640 tsk->clear_child_tid = NULL;
1641 }
1642
1643 /*
1644 * All done, finally we can wake up parent and return this mm to him.
1645 * Also kthread_stop() uses this completion for synchronization.
1646 */
1647 if (tsk->vfork_done)
1648 complete_vfork_done(tsk);
1649}
1650
1651void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1652{
1653 futex_exit_release(tsk);
1654 mm_release(tsk, mm);
1655}
1656
1657void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1658{
1659 futex_exec_release(tsk);
1660 mm_release(tsk, mm);
1661}
1662
1663/**
1664 * dup_mm() - duplicates an existing mm structure
1665 * @tsk: the task_struct with which the new mm will be associated.
1666 * @oldmm: the mm to duplicate.
1667 *
1668 * Allocates a new mm structure and duplicates the provided @oldmm structure
1669 * content into it.
1670 *
1671 * Return: the duplicated mm or NULL on failure.
1672 */
1673static struct mm_struct *dup_mm(struct task_struct *tsk,
1674 struct mm_struct *oldmm)
1675{
1676 struct mm_struct *mm;
1677 int err;
1678
1679 mm = allocate_mm();
1680 if (!mm)
1681 goto fail_nomem;
1682
1683 memcpy(mm, oldmm, sizeof(*mm));
1684
1685 if (!mm_init(mm, tsk, mm->user_ns))
1686 goto fail_nomem;
1687
1688 err = dup_mmap(mm, oldmm);
1689 if (err)
1690 goto free_pt;
1691
1692 mm->hiwater_rss = get_mm_rss(mm);
1693 mm->hiwater_vm = mm->total_vm;
1694
1695 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1696 goto free_pt;
1697
1698 return mm;
1699
1700free_pt:
1701 /* don't put binfmt in mmput, we haven't got module yet */
1702 mm->binfmt = NULL;
1703 mm_init_owner(mm, NULL);
1704 mmput(mm);
1705
1706fail_nomem:
1707 return NULL;
1708}
1709
1710static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1711{
1712 struct mm_struct *mm, *oldmm;
1713
1714 tsk->min_flt = tsk->maj_flt = 0;
1715 tsk->nvcsw = tsk->nivcsw = 0;
1716#ifdef CONFIG_DETECT_HUNG_TASK
1717 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1718 tsk->last_switch_time = 0;
1719#endif
1720
1721 tsk->mm = NULL;
1722 tsk->active_mm = NULL;
1723
1724 /*
1725 * Are we cloning a kernel thread?
1726 *
1727 * We need to steal a active VM for that..
1728 */
1729 oldmm = current->mm;
1730 if (!oldmm)
1731 return 0;
1732
1733 if (clone_flags & CLONE_VM) {
1734 mmget(oldmm);
1735 mm = oldmm;
1736 } else {
1737 mm = dup_mm(tsk, current->mm);
1738 if (!mm)
1739 return -ENOMEM;
1740 }
1741
1742 tsk->mm = mm;
1743 tsk->active_mm = mm;
1744 sched_mm_cid_fork(tsk);
1745 return 0;
1746}
1747
1748static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1749{
1750 struct fs_struct *fs = current->fs;
1751 if (clone_flags & CLONE_FS) {
1752 /* tsk->fs is already what we want */
1753 spin_lock(&fs->lock);
1754 /* "users" and "in_exec" locked for check_unsafe_exec() */
1755 if (fs->in_exec) {
1756 spin_unlock(&fs->lock);
1757 return -EAGAIN;
1758 }
1759 fs->users++;
1760 spin_unlock(&fs->lock);
1761 return 0;
1762 }
1763 tsk->fs = copy_fs_struct(fs);
1764 if (!tsk->fs)
1765 return -ENOMEM;
1766 return 0;
1767}
1768
1769static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1770 int no_files)
1771{
1772 struct files_struct *oldf, *newf;
1773 int error = 0;
1774
1775 /*
1776 * A background process may not have any files ...
1777 */
1778 oldf = current->files;
1779 if (!oldf)
1780 goto out;
1781
1782 if (no_files) {
1783 tsk->files = NULL;
1784 goto out;
1785 }
1786
1787 if (clone_flags & CLONE_FILES) {
1788 atomic_inc(&oldf->count);
1789 goto out;
1790 }
1791
1792 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1793 if (!newf)
1794 goto out;
1795
1796 tsk->files = newf;
1797 error = 0;
1798out:
1799 return error;
1800}
1801
1802static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1803{
1804 struct sighand_struct *sig;
1805
1806 if (clone_flags & CLONE_SIGHAND) {
1807 refcount_inc(¤t->sighand->count);
1808 return 0;
1809 }
1810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1811 RCU_INIT_POINTER(tsk->sighand, sig);
1812 if (!sig)
1813 return -ENOMEM;
1814
1815 refcount_set(&sig->count, 1);
1816 spin_lock_irq(¤t->sighand->siglock);
1817 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1818 spin_unlock_irq(¤t->sighand->siglock);
1819
1820 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1821 if (clone_flags & CLONE_CLEAR_SIGHAND)
1822 flush_signal_handlers(tsk, 0);
1823
1824 return 0;
1825}
1826
1827void __cleanup_sighand(struct sighand_struct *sighand)
1828{
1829 if (refcount_dec_and_test(&sighand->count)) {
1830 signalfd_cleanup(sighand);
1831 /*
1832 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1833 * without an RCU grace period, see __lock_task_sighand().
1834 */
1835 kmem_cache_free(sighand_cachep, sighand);
1836 }
1837}
1838
1839/*
1840 * Initialize POSIX timer handling for a thread group.
1841 */
1842static void posix_cpu_timers_init_group(struct signal_struct *sig)
1843{
1844 struct posix_cputimers *pct = &sig->posix_cputimers;
1845 unsigned long cpu_limit;
1846
1847 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1848 posix_cputimers_group_init(pct, cpu_limit);
1849}
1850
1851static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1852{
1853 struct signal_struct *sig;
1854
1855 if (clone_flags & CLONE_THREAD)
1856 return 0;
1857
1858 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1859 tsk->signal = sig;
1860 if (!sig)
1861 return -ENOMEM;
1862
1863 sig->nr_threads = 1;
1864 sig->quick_threads = 1;
1865 atomic_set(&sig->live, 1);
1866 refcount_set(&sig->sigcnt, 1);
1867
1868 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1869 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1870 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1871
1872 init_waitqueue_head(&sig->wait_chldexit);
1873 sig->curr_target = tsk;
1874 init_sigpending(&sig->shared_pending);
1875 INIT_HLIST_HEAD(&sig->multiprocess);
1876 seqlock_init(&sig->stats_lock);
1877 prev_cputime_init(&sig->prev_cputime);
1878
1879#ifdef CONFIG_POSIX_TIMERS
1880 INIT_LIST_HEAD(&sig->posix_timers);
1881 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1882 sig->real_timer.function = it_real_fn;
1883#endif
1884
1885 task_lock(current->group_leader);
1886 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1887 task_unlock(current->group_leader);
1888
1889 posix_cpu_timers_init_group(sig);
1890
1891 tty_audit_fork(sig);
1892 sched_autogroup_fork(sig);
1893
1894 sig->oom_score_adj = current->signal->oom_score_adj;
1895 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1896
1897 mutex_init(&sig->cred_guard_mutex);
1898 init_rwsem(&sig->exec_update_lock);
1899
1900 return 0;
1901}
1902
1903static void copy_seccomp(struct task_struct *p)
1904{
1905#ifdef CONFIG_SECCOMP
1906 /*
1907 * Must be called with sighand->lock held, which is common to
1908 * all threads in the group. Holding cred_guard_mutex is not
1909 * needed because this new task is not yet running and cannot
1910 * be racing exec.
1911 */
1912 assert_spin_locked(¤t->sighand->siglock);
1913
1914 /* Ref-count the new filter user, and assign it. */
1915 get_seccomp_filter(current);
1916 p->seccomp = current->seccomp;
1917
1918 /*
1919 * Explicitly enable no_new_privs here in case it got set
1920 * between the task_struct being duplicated and holding the
1921 * sighand lock. The seccomp state and nnp must be in sync.
1922 */
1923 if (task_no_new_privs(current))
1924 task_set_no_new_privs(p);
1925
1926 /*
1927 * If the parent gained a seccomp mode after copying thread
1928 * flags and between before we held the sighand lock, we have
1929 * to manually enable the seccomp thread flag here.
1930 */
1931 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1932 set_task_syscall_work(p, SECCOMP);
1933#endif
1934}
1935
1936SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1937{
1938 current->clear_child_tid = tidptr;
1939
1940 return task_pid_vnr(current);
1941}
1942
1943static void rt_mutex_init_task(struct task_struct *p)
1944{
1945 raw_spin_lock_init(&p->pi_lock);
1946#ifdef CONFIG_RT_MUTEXES
1947 p->pi_waiters = RB_ROOT_CACHED;
1948 p->pi_top_task = NULL;
1949 p->pi_blocked_on = NULL;
1950#endif
1951}
1952
1953static inline void init_task_pid_links(struct task_struct *task)
1954{
1955 enum pid_type type;
1956
1957 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1958 INIT_HLIST_NODE(&task->pid_links[type]);
1959}
1960
1961static inline void
1962init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1963{
1964 if (type == PIDTYPE_PID)
1965 task->thread_pid = pid;
1966 else
1967 task->signal->pids[type] = pid;
1968}
1969
1970static inline void rcu_copy_process(struct task_struct *p)
1971{
1972#ifdef CONFIG_PREEMPT_RCU
1973 p->rcu_read_lock_nesting = 0;
1974 p->rcu_read_unlock_special.s = 0;
1975 p->rcu_blocked_node = NULL;
1976 INIT_LIST_HEAD(&p->rcu_node_entry);
1977#endif /* #ifdef CONFIG_PREEMPT_RCU */
1978#ifdef CONFIG_TASKS_RCU
1979 p->rcu_tasks_holdout = false;
1980 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1981 p->rcu_tasks_idle_cpu = -1;
1982 INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1983#endif /* #ifdef CONFIG_TASKS_RCU */
1984#ifdef CONFIG_TASKS_TRACE_RCU
1985 p->trc_reader_nesting = 0;
1986 p->trc_reader_special.s = 0;
1987 INIT_LIST_HEAD(&p->trc_holdout_list);
1988 INIT_LIST_HEAD(&p->trc_blkd_node);
1989#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1990}
1991
1992/**
1993 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1994 * @pid: the struct pid for which to create a pidfd
1995 * @flags: flags of the new @pidfd
1996 * @ret: Where to return the file for the pidfd.
1997 *
1998 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1999 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2000 *
2001 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2002 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2003 * pidfd file are prepared.
2004 *
2005 * If this function returns successfully the caller is responsible to either
2006 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2007 * order to install the pidfd into its file descriptor table or they must use
2008 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2009 * respectively.
2010 *
2011 * This function is useful when a pidfd must already be reserved but there
2012 * might still be points of failure afterwards and the caller wants to ensure
2013 * that no pidfd is leaked into its file descriptor table.
2014 *
2015 * Return: On success, a reserved pidfd is returned from the function and a new
2016 * pidfd file is returned in the last argument to the function. On
2017 * error, a negative error code is returned from the function and the
2018 * last argument remains unchanged.
2019 */
2020static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2021{
2022 int pidfd;
2023 struct file *pidfd_file;
2024
2025 pidfd = get_unused_fd_flags(O_CLOEXEC);
2026 if (pidfd < 0)
2027 return pidfd;
2028
2029 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2030 if (IS_ERR(pidfd_file)) {
2031 put_unused_fd(pidfd);
2032 return PTR_ERR(pidfd_file);
2033 }
2034 /*
2035 * anon_inode_getfile() ignores everything outside of the
2036 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2037 */
2038 pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2039 *ret = pidfd_file;
2040 return pidfd;
2041}
2042
2043/**
2044 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2045 * @pid: the struct pid for which to create a pidfd
2046 * @flags: flags of the new @pidfd
2047 * @ret: Where to return the pidfd.
2048 *
2049 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2050 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2051 *
2052 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2053 * task identified by @pid must be a thread-group leader.
2054 *
2055 * If this function returns successfully the caller is responsible to either
2056 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2057 * order to install the pidfd into its file descriptor table or they must use
2058 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2059 * respectively.
2060 *
2061 * This function is useful when a pidfd must already be reserved but there
2062 * might still be points of failure afterwards and the caller wants to ensure
2063 * that no pidfd is leaked into its file descriptor table.
2064 *
2065 * Return: On success, a reserved pidfd is returned from the function and a new
2066 * pidfd file is returned in the last argument to the function. On
2067 * error, a negative error code is returned from the function and the
2068 * last argument remains unchanged.
2069 */
2070int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2071{
2072 bool thread = flags & PIDFD_THREAD;
2073
2074 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2075 return -EINVAL;
2076
2077 return __pidfd_prepare(pid, flags, ret);
2078}
2079
2080static void __delayed_free_task(struct rcu_head *rhp)
2081{
2082 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2083
2084 free_task(tsk);
2085}
2086
2087static __always_inline void delayed_free_task(struct task_struct *tsk)
2088{
2089 if (IS_ENABLED(CONFIG_MEMCG))
2090 call_rcu(&tsk->rcu, __delayed_free_task);
2091 else
2092 free_task(tsk);
2093}
2094
2095static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2096{
2097 /* Skip if kernel thread */
2098 if (!tsk->mm)
2099 return;
2100
2101 /* Skip if spawning a thread or using vfork */
2102 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2103 return;
2104
2105 /* We need to synchronize with __set_oom_adj */
2106 mutex_lock(&oom_adj_mutex);
2107 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2108 /* Update the values in case they were changed after copy_signal */
2109 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2110 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2111 mutex_unlock(&oom_adj_mutex);
2112}
2113
2114#ifdef CONFIG_RV
2115static void rv_task_fork(struct task_struct *p)
2116{
2117 int i;
2118
2119 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2120 p->rv[i].da_mon.monitoring = false;
2121}
2122#else
2123#define rv_task_fork(p) do {} while (0)
2124#endif
2125
2126/*
2127 * This creates a new process as a copy of the old one,
2128 * but does not actually start it yet.
2129 *
2130 * It copies the registers, and all the appropriate
2131 * parts of the process environment (as per the clone
2132 * flags). The actual kick-off is left to the caller.
2133 */
2134__latent_entropy struct task_struct *copy_process(
2135 struct pid *pid,
2136 int trace,
2137 int node,
2138 struct kernel_clone_args *args)
2139{
2140 int pidfd = -1, retval;
2141 struct task_struct *p;
2142 struct multiprocess_signals delayed;
2143 struct file *pidfile = NULL;
2144 const u64 clone_flags = args->flags;
2145 struct nsproxy *nsp = current->nsproxy;
2146
2147 /*
2148 * Don't allow sharing the root directory with processes in a different
2149 * namespace
2150 */
2151 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2152 return ERR_PTR(-EINVAL);
2153
2154 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2155 return ERR_PTR(-EINVAL);
2156
2157 /*
2158 * Thread groups must share signals as well, and detached threads
2159 * can only be started up within the thread group.
2160 */
2161 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2162 return ERR_PTR(-EINVAL);
2163
2164 /*
2165 * Shared signal handlers imply shared VM. By way of the above,
2166 * thread groups also imply shared VM. Blocking this case allows
2167 * for various simplifications in other code.
2168 */
2169 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2170 return ERR_PTR(-EINVAL);
2171
2172 /*
2173 * Siblings of global init remain as zombies on exit since they are
2174 * not reaped by their parent (swapper). To solve this and to avoid
2175 * multi-rooted process trees, prevent global and container-inits
2176 * from creating siblings.
2177 */
2178 if ((clone_flags & CLONE_PARENT) &&
2179 current->signal->flags & SIGNAL_UNKILLABLE)
2180 return ERR_PTR(-EINVAL);
2181
2182 /*
2183 * If the new process will be in a different pid or user namespace
2184 * do not allow it to share a thread group with the forking task.
2185 */
2186 if (clone_flags & CLONE_THREAD) {
2187 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2188 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2189 return ERR_PTR(-EINVAL);
2190 }
2191
2192 if (clone_flags & CLONE_PIDFD) {
2193 /*
2194 * - CLONE_DETACHED is blocked so that we can potentially
2195 * reuse it later for CLONE_PIDFD.
2196 */
2197 if (clone_flags & CLONE_DETACHED)
2198 return ERR_PTR(-EINVAL);
2199 }
2200
2201 /*
2202 * Force any signals received before this point to be delivered
2203 * before the fork happens. Collect up signals sent to multiple
2204 * processes that happen during the fork and delay them so that
2205 * they appear to happen after the fork.
2206 */
2207 sigemptyset(&delayed.signal);
2208 INIT_HLIST_NODE(&delayed.node);
2209
2210 spin_lock_irq(¤t->sighand->siglock);
2211 if (!(clone_flags & CLONE_THREAD))
2212 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2213 recalc_sigpending();
2214 spin_unlock_irq(¤t->sighand->siglock);
2215 retval = -ERESTARTNOINTR;
2216 if (task_sigpending(current))
2217 goto fork_out;
2218
2219 retval = -ENOMEM;
2220 p = dup_task_struct(current, node);
2221 if (!p)
2222 goto fork_out;
2223 p->flags &= ~PF_KTHREAD;
2224 if (args->kthread)
2225 p->flags |= PF_KTHREAD;
2226 if (args->user_worker) {
2227 /*
2228 * Mark us a user worker, and block any signal that isn't
2229 * fatal or STOP
2230 */
2231 p->flags |= PF_USER_WORKER;
2232 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2233 }
2234 if (args->io_thread)
2235 p->flags |= PF_IO_WORKER;
2236
2237 if (args->name)
2238 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2239
2240 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2241 /*
2242 * Clear TID on mm_release()?
2243 */
2244 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2245
2246 ftrace_graph_init_task(p);
2247
2248 rt_mutex_init_task(p);
2249
2250 lockdep_assert_irqs_enabled();
2251#ifdef CONFIG_PROVE_LOCKING
2252 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2253#endif
2254 retval = copy_creds(p, clone_flags);
2255 if (retval < 0)
2256 goto bad_fork_free;
2257
2258 retval = -EAGAIN;
2259 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2260 if (p->real_cred->user != INIT_USER &&
2261 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2262 goto bad_fork_cleanup_count;
2263 }
2264 current->flags &= ~PF_NPROC_EXCEEDED;
2265
2266 /*
2267 * If multiple threads are within copy_process(), then this check
2268 * triggers too late. This doesn't hurt, the check is only there
2269 * to stop root fork bombs.
2270 */
2271 retval = -EAGAIN;
2272 if (data_race(nr_threads >= max_threads))
2273 goto bad_fork_cleanup_count;
2274
2275 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2276 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2277 p->flags |= PF_FORKNOEXEC;
2278 INIT_LIST_HEAD(&p->children);
2279 INIT_LIST_HEAD(&p->sibling);
2280 rcu_copy_process(p);
2281 p->vfork_done = NULL;
2282 spin_lock_init(&p->alloc_lock);
2283
2284 init_sigpending(&p->pending);
2285
2286 p->utime = p->stime = p->gtime = 0;
2287#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2288 p->utimescaled = p->stimescaled = 0;
2289#endif
2290 prev_cputime_init(&p->prev_cputime);
2291
2292#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2293 seqcount_init(&p->vtime.seqcount);
2294 p->vtime.starttime = 0;
2295 p->vtime.state = VTIME_INACTIVE;
2296#endif
2297
2298#ifdef CONFIG_IO_URING
2299 p->io_uring = NULL;
2300#endif
2301
2302 p->default_timer_slack_ns = current->timer_slack_ns;
2303
2304#ifdef CONFIG_PSI
2305 p->psi_flags = 0;
2306#endif
2307
2308 task_io_accounting_init(&p->ioac);
2309 acct_clear_integrals(p);
2310
2311 posix_cputimers_init(&p->posix_cputimers);
2312
2313 p->io_context = NULL;
2314 audit_set_context(p, NULL);
2315 cgroup_fork(p);
2316 if (args->kthread) {
2317 if (!set_kthread_struct(p))
2318 goto bad_fork_cleanup_delayacct;
2319 }
2320#ifdef CONFIG_NUMA
2321 p->mempolicy = mpol_dup(p->mempolicy);
2322 if (IS_ERR(p->mempolicy)) {
2323 retval = PTR_ERR(p->mempolicy);
2324 p->mempolicy = NULL;
2325 goto bad_fork_cleanup_delayacct;
2326 }
2327#endif
2328#ifdef CONFIG_CPUSETS
2329 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2330 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2331 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2332#endif
2333#ifdef CONFIG_TRACE_IRQFLAGS
2334 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2335 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2336 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2337 p->softirqs_enabled = 1;
2338 p->softirq_context = 0;
2339#endif
2340
2341 p->pagefault_disabled = 0;
2342
2343#ifdef CONFIG_LOCKDEP
2344 lockdep_init_task(p);
2345#endif
2346
2347#ifdef CONFIG_DEBUG_MUTEXES
2348 p->blocked_on = NULL; /* not blocked yet */
2349#endif
2350#ifdef CONFIG_BCACHE
2351 p->sequential_io = 0;
2352 p->sequential_io_avg = 0;
2353#endif
2354#ifdef CONFIG_BPF_SYSCALL
2355 RCU_INIT_POINTER(p->bpf_storage, NULL);
2356 p->bpf_ctx = NULL;
2357#endif
2358
2359 /* Perform scheduler related setup. Assign this task to a CPU. */
2360 retval = sched_fork(clone_flags, p);
2361 if (retval)
2362 goto bad_fork_cleanup_policy;
2363
2364 retval = perf_event_init_task(p, clone_flags);
2365 if (retval)
2366 goto bad_fork_cleanup_policy;
2367 retval = audit_alloc(p);
2368 if (retval)
2369 goto bad_fork_cleanup_perf;
2370 /* copy all the process information */
2371 shm_init_task(p);
2372 retval = security_task_alloc(p, clone_flags);
2373 if (retval)
2374 goto bad_fork_cleanup_audit;
2375 retval = copy_semundo(clone_flags, p);
2376 if (retval)
2377 goto bad_fork_cleanup_security;
2378 retval = copy_files(clone_flags, p, args->no_files);
2379 if (retval)
2380 goto bad_fork_cleanup_semundo;
2381 retval = copy_fs(clone_flags, p);
2382 if (retval)
2383 goto bad_fork_cleanup_files;
2384 retval = copy_sighand(clone_flags, p);
2385 if (retval)
2386 goto bad_fork_cleanup_fs;
2387 retval = copy_signal(clone_flags, p);
2388 if (retval)
2389 goto bad_fork_cleanup_sighand;
2390 retval = copy_mm(clone_flags, p);
2391 if (retval)
2392 goto bad_fork_cleanup_signal;
2393 retval = copy_namespaces(clone_flags, p);
2394 if (retval)
2395 goto bad_fork_cleanup_mm;
2396 retval = copy_io(clone_flags, p);
2397 if (retval)
2398 goto bad_fork_cleanup_namespaces;
2399 retval = copy_thread(p, args);
2400 if (retval)
2401 goto bad_fork_cleanup_io;
2402
2403 stackleak_task_init(p);
2404
2405 if (pid != &init_struct_pid) {
2406 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2407 args->set_tid_size);
2408 if (IS_ERR(pid)) {
2409 retval = PTR_ERR(pid);
2410 goto bad_fork_cleanup_thread;
2411 }
2412 }
2413
2414 /*
2415 * This has to happen after we've potentially unshared the file
2416 * descriptor table (so that the pidfd doesn't leak into the child
2417 * if the fd table isn't shared).
2418 */
2419 if (clone_flags & CLONE_PIDFD) {
2420 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2421
2422 /* Note that no task has been attached to @pid yet. */
2423 retval = __pidfd_prepare(pid, flags, &pidfile);
2424 if (retval < 0)
2425 goto bad_fork_free_pid;
2426 pidfd = retval;
2427
2428 retval = put_user(pidfd, args->pidfd);
2429 if (retval)
2430 goto bad_fork_put_pidfd;
2431 }
2432
2433#ifdef CONFIG_BLOCK
2434 p->plug = NULL;
2435#endif
2436 futex_init_task(p);
2437
2438 /*
2439 * sigaltstack should be cleared when sharing the same VM
2440 */
2441 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2442 sas_ss_reset(p);
2443
2444 /*
2445 * Syscall tracing and stepping should be turned off in the
2446 * child regardless of CLONE_PTRACE.
2447 */
2448 user_disable_single_step(p);
2449 clear_task_syscall_work(p, SYSCALL_TRACE);
2450#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2451 clear_task_syscall_work(p, SYSCALL_EMU);
2452#endif
2453 clear_tsk_latency_tracing(p);
2454
2455 /* ok, now we should be set up.. */
2456 p->pid = pid_nr(pid);
2457 if (clone_flags & CLONE_THREAD) {
2458 p->group_leader = current->group_leader;
2459 p->tgid = current->tgid;
2460 } else {
2461 p->group_leader = p;
2462 p->tgid = p->pid;
2463 }
2464
2465 p->nr_dirtied = 0;
2466 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2467 p->dirty_paused_when = 0;
2468
2469 p->pdeath_signal = 0;
2470 p->task_works = NULL;
2471 clear_posix_cputimers_work(p);
2472
2473#ifdef CONFIG_KRETPROBES
2474 p->kretprobe_instances.first = NULL;
2475#endif
2476#ifdef CONFIG_RETHOOK
2477 p->rethooks.first = NULL;
2478#endif
2479
2480 /*
2481 * Ensure that the cgroup subsystem policies allow the new process to be
2482 * forked. It should be noted that the new process's css_set can be changed
2483 * between here and cgroup_post_fork() if an organisation operation is in
2484 * progress.
2485 */
2486 retval = cgroup_can_fork(p, args);
2487 if (retval)
2488 goto bad_fork_put_pidfd;
2489
2490 /*
2491 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2492 * the new task on the correct runqueue. All this *before* the task
2493 * becomes visible.
2494 *
2495 * This isn't part of ->can_fork() because while the re-cloning is
2496 * cgroup specific, it unconditionally needs to place the task on a
2497 * runqueue.
2498 */
2499 sched_cgroup_fork(p, args);
2500
2501 /*
2502 * From this point on we must avoid any synchronous user-space
2503 * communication until we take the tasklist-lock. In particular, we do
2504 * not want user-space to be able to predict the process start-time by
2505 * stalling fork(2) after we recorded the start_time but before it is
2506 * visible to the system.
2507 */
2508
2509 p->start_time = ktime_get_ns();
2510 p->start_boottime = ktime_get_boottime_ns();
2511
2512 /*
2513 * Make it visible to the rest of the system, but dont wake it up yet.
2514 * Need tasklist lock for parent etc handling!
2515 */
2516 write_lock_irq(&tasklist_lock);
2517
2518 /* CLONE_PARENT re-uses the old parent */
2519 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2520 p->real_parent = current->real_parent;
2521 p->parent_exec_id = current->parent_exec_id;
2522 if (clone_flags & CLONE_THREAD)
2523 p->exit_signal = -1;
2524 else
2525 p->exit_signal = current->group_leader->exit_signal;
2526 } else {
2527 p->real_parent = current;
2528 p->parent_exec_id = current->self_exec_id;
2529 p->exit_signal = args->exit_signal;
2530 }
2531
2532 klp_copy_process(p);
2533
2534 sched_core_fork(p);
2535
2536 spin_lock(¤t->sighand->siglock);
2537
2538 rv_task_fork(p);
2539
2540 rseq_fork(p, clone_flags);
2541
2542 /* Don't start children in a dying pid namespace */
2543 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2544 retval = -ENOMEM;
2545 goto bad_fork_cancel_cgroup;
2546 }
2547
2548 /* Let kill terminate clone/fork in the middle */
2549 if (fatal_signal_pending(current)) {
2550 retval = -EINTR;
2551 goto bad_fork_cancel_cgroup;
2552 }
2553
2554 /* No more failure paths after this point. */
2555
2556 /*
2557 * Copy seccomp details explicitly here, in case they were changed
2558 * before holding sighand lock.
2559 */
2560 copy_seccomp(p);
2561
2562 init_task_pid_links(p);
2563 if (likely(p->pid)) {
2564 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2565
2566 init_task_pid(p, PIDTYPE_PID, pid);
2567 if (thread_group_leader(p)) {
2568 init_task_pid(p, PIDTYPE_TGID, pid);
2569 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2570 init_task_pid(p, PIDTYPE_SID, task_session(current));
2571
2572 if (is_child_reaper(pid)) {
2573 ns_of_pid(pid)->child_reaper = p;
2574 p->signal->flags |= SIGNAL_UNKILLABLE;
2575 }
2576 p->signal->shared_pending.signal = delayed.signal;
2577 p->signal->tty = tty_kref_get(current->signal->tty);
2578 /*
2579 * Inherit has_child_subreaper flag under the same
2580 * tasklist_lock with adding child to the process tree
2581 * for propagate_has_child_subreaper optimization.
2582 */
2583 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2584 p->real_parent->signal->is_child_subreaper;
2585 list_add_tail(&p->sibling, &p->real_parent->children);
2586 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2587 attach_pid(p, PIDTYPE_TGID);
2588 attach_pid(p, PIDTYPE_PGID);
2589 attach_pid(p, PIDTYPE_SID);
2590 __this_cpu_inc(process_counts);
2591 } else {
2592 current->signal->nr_threads++;
2593 current->signal->quick_threads++;
2594 atomic_inc(¤t->signal->live);
2595 refcount_inc(¤t->signal->sigcnt);
2596 task_join_group_stop(p);
2597 list_add_tail_rcu(&p->thread_node,
2598 &p->signal->thread_head);
2599 }
2600 attach_pid(p, PIDTYPE_PID);
2601 nr_threads++;
2602 }
2603 total_forks++;
2604 hlist_del_init(&delayed.node);
2605 spin_unlock(¤t->sighand->siglock);
2606 syscall_tracepoint_update(p);
2607 write_unlock_irq(&tasklist_lock);
2608
2609 if (pidfile)
2610 fd_install(pidfd, pidfile);
2611
2612 proc_fork_connector(p);
2613 sched_post_fork(p);
2614 cgroup_post_fork(p, args);
2615 perf_event_fork(p);
2616
2617 trace_task_newtask(p, clone_flags);
2618 uprobe_copy_process(p, clone_flags);
2619 user_events_fork(p, clone_flags);
2620
2621 copy_oom_score_adj(clone_flags, p);
2622
2623 return p;
2624
2625bad_fork_cancel_cgroup:
2626 sched_core_free(p);
2627 spin_unlock(¤t->sighand->siglock);
2628 write_unlock_irq(&tasklist_lock);
2629 cgroup_cancel_fork(p, args);
2630bad_fork_put_pidfd:
2631 if (clone_flags & CLONE_PIDFD) {
2632 fput(pidfile);
2633 put_unused_fd(pidfd);
2634 }
2635bad_fork_free_pid:
2636 if (pid != &init_struct_pid)
2637 free_pid(pid);
2638bad_fork_cleanup_thread:
2639 exit_thread(p);
2640bad_fork_cleanup_io:
2641 if (p->io_context)
2642 exit_io_context(p);
2643bad_fork_cleanup_namespaces:
2644 exit_task_namespaces(p);
2645bad_fork_cleanup_mm:
2646 if (p->mm) {
2647 mm_clear_owner(p->mm, p);
2648 mmput(p->mm);
2649 }
2650bad_fork_cleanup_signal:
2651 if (!(clone_flags & CLONE_THREAD))
2652 free_signal_struct(p->signal);
2653bad_fork_cleanup_sighand:
2654 __cleanup_sighand(p->sighand);
2655bad_fork_cleanup_fs:
2656 exit_fs(p); /* blocking */
2657bad_fork_cleanup_files:
2658 exit_files(p); /* blocking */
2659bad_fork_cleanup_semundo:
2660 exit_sem(p);
2661bad_fork_cleanup_security:
2662 security_task_free(p);
2663bad_fork_cleanup_audit:
2664 audit_free(p);
2665bad_fork_cleanup_perf:
2666 perf_event_free_task(p);
2667bad_fork_cleanup_policy:
2668 lockdep_free_task(p);
2669#ifdef CONFIG_NUMA
2670 mpol_put(p->mempolicy);
2671#endif
2672bad_fork_cleanup_delayacct:
2673 delayacct_tsk_free(p);
2674bad_fork_cleanup_count:
2675 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2676 exit_creds(p);
2677bad_fork_free:
2678 WRITE_ONCE(p->__state, TASK_DEAD);
2679 exit_task_stack_account(p);
2680 put_task_stack(p);
2681 delayed_free_task(p);
2682fork_out:
2683 spin_lock_irq(¤t->sighand->siglock);
2684 hlist_del_init(&delayed.node);
2685 spin_unlock_irq(¤t->sighand->siglock);
2686 return ERR_PTR(retval);
2687}
2688
2689static inline void init_idle_pids(struct task_struct *idle)
2690{
2691 enum pid_type type;
2692
2693 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2694 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2695 init_task_pid(idle, type, &init_struct_pid);
2696 }
2697}
2698
2699static int idle_dummy(void *dummy)
2700{
2701 /* This function is never called */
2702 return 0;
2703}
2704
2705struct task_struct * __init fork_idle(int cpu)
2706{
2707 struct task_struct *task;
2708 struct kernel_clone_args args = {
2709 .flags = CLONE_VM,
2710 .fn = &idle_dummy,
2711 .fn_arg = NULL,
2712 .kthread = 1,
2713 .idle = 1,
2714 };
2715
2716 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2717 if (!IS_ERR(task)) {
2718 init_idle_pids(task);
2719 init_idle(task, cpu);
2720 }
2721
2722 return task;
2723}
2724
2725/*
2726 * This is like kernel_clone(), but shaved down and tailored to just
2727 * creating io_uring workers. It returns a created task, or an error pointer.
2728 * The returned task is inactive, and the caller must fire it up through
2729 * wake_up_new_task(p). All signals are blocked in the created task.
2730 */
2731struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2732{
2733 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2734 CLONE_IO;
2735 struct kernel_clone_args args = {
2736 .flags = ((lower_32_bits(flags) | CLONE_VM |
2737 CLONE_UNTRACED) & ~CSIGNAL),
2738 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2739 .fn = fn,
2740 .fn_arg = arg,
2741 .io_thread = 1,
2742 .user_worker = 1,
2743 };
2744
2745 return copy_process(NULL, 0, node, &args);
2746}
2747
2748/*
2749 * Ok, this is the main fork-routine.
2750 *
2751 * It copies the process, and if successful kick-starts
2752 * it and waits for it to finish using the VM if required.
2753 *
2754 * args->exit_signal is expected to be checked for sanity by the caller.
2755 */
2756pid_t kernel_clone(struct kernel_clone_args *args)
2757{
2758 u64 clone_flags = args->flags;
2759 struct completion vfork;
2760 struct pid *pid;
2761 struct task_struct *p;
2762 int trace = 0;
2763 pid_t nr;
2764
2765 /*
2766 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2767 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2768 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2769 * field in struct clone_args and it still doesn't make sense to have
2770 * them both point at the same memory location. Performing this check
2771 * here has the advantage that we don't need to have a separate helper
2772 * to check for legacy clone().
2773 */
2774 if ((clone_flags & CLONE_PIDFD) &&
2775 (clone_flags & CLONE_PARENT_SETTID) &&
2776 (args->pidfd == args->parent_tid))
2777 return -EINVAL;
2778
2779 /*
2780 * Determine whether and which event to report to ptracer. When
2781 * called from kernel_thread or CLONE_UNTRACED is explicitly
2782 * requested, no event is reported; otherwise, report if the event
2783 * for the type of forking is enabled.
2784 */
2785 if (!(clone_flags & CLONE_UNTRACED)) {
2786 if (clone_flags & CLONE_VFORK)
2787 trace = PTRACE_EVENT_VFORK;
2788 else if (args->exit_signal != SIGCHLD)
2789 trace = PTRACE_EVENT_CLONE;
2790 else
2791 trace = PTRACE_EVENT_FORK;
2792
2793 if (likely(!ptrace_event_enabled(current, trace)))
2794 trace = 0;
2795 }
2796
2797 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2798 add_latent_entropy();
2799
2800 if (IS_ERR(p))
2801 return PTR_ERR(p);
2802
2803 /*
2804 * Do this prior waking up the new thread - the thread pointer
2805 * might get invalid after that point, if the thread exits quickly.
2806 */
2807 trace_sched_process_fork(current, p);
2808
2809 pid = get_task_pid(p, PIDTYPE_PID);
2810 nr = pid_vnr(pid);
2811
2812 if (clone_flags & CLONE_PARENT_SETTID)
2813 put_user(nr, args->parent_tid);
2814
2815 if (clone_flags & CLONE_VFORK) {
2816 p->vfork_done = &vfork;
2817 init_completion(&vfork);
2818 get_task_struct(p);
2819 }
2820
2821 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2822 /* lock the task to synchronize with memcg migration */
2823 task_lock(p);
2824 lru_gen_add_mm(p->mm);
2825 task_unlock(p);
2826 }
2827
2828 wake_up_new_task(p);
2829
2830 /* forking complete and child started to run, tell ptracer */
2831 if (unlikely(trace))
2832 ptrace_event_pid(trace, pid);
2833
2834 if (clone_flags & CLONE_VFORK) {
2835 if (!wait_for_vfork_done(p, &vfork))
2836 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2837 }
2838
2839 put_pid(pid);
2840 return nr;
2841}
2842
2843/*
2844 * Create a kernel thread.
2845 */
2846pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2847 unsigned long flags)
2848{
2849 struct kernel_clone_args args = {
2850 .flags = ((lower_32_bits(flags) | CLONE_VM |
2851 CLONE_UNTRACED) & ~CSIGNAL),
2852 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2853 .fn = fn,
2854 .fn_arg = arg,
2855 .name = name,
2856 .kthread = 1,
2857 };
2858
2859 return kernel_clone(&args);
2860}
2861
2862/*
2863 * Create a user mode thread.
2864 */
2865pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2866{
2867 struct kernel_clone_args args = {
2868 .flags = ((lower_32_bits(flags) | CLONE_VM |
2869 CLONE_UNTRACED) & ~CSIGNAL),
2870 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2871 .fn = fn,
2872 .fn_arg = arg,
2873 };
2874
2875 return kernel_clone(&args);
2876}
2877
2878#ifdef __ARCH_WANT_SYS_FORK
2879SYSCALL_DEFINE0(fork)
2880{
2881#ifdef CONFIG_MMU
2882 struct kernel_clone_args args = {
2883 .exit_signal = SIGCHLD,
2884 };
2885
2886 return kernel_clone(&args);
2887#else
2888 /* can not support in nommu mode */
2889 return -EINVAL;
2890#endif
2891}
2892#endif
2893
2894#ifdef __ARCH_WANT_SYS_VFORK
2895SYSCALL_DEFINE0(vfork)
2896{
2897 struct kernel_clone_args args = {
2898 .flags = CLONE_VFORK | CLONE_VM,
2899 .exit_signal = SIGCHLD,
2900 };
2901
2902 return kernel_clone(&args);
2903}
2904#endif
2905
2906#ifdef __ARCH_WANT_SYS_CLONE
2907#ifdef CONFIG_CLONE_BACKWARDS
2908SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2909 int __user *, parent_tidptr,
2910 unsigned long, tls,
2911 int __user *, child_tidptr)
2912#elif defined(CONFIG_CLONE_BACKWARDS2)
2913SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2914 int __user *, parent_tidptr,
2915 int __user *, child_tidptr,
2916 unsigned long, tls)
2917#elif defined(CONFIG_CLONE_BACKWARDS3)
2918SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2919 int, stack_size,
2920 int __user *, parent_tidptr,
2921 int __user *, child_tidptr,
2922 unsigned long, tls)
2923#else
2924SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2925 int __user *, parent_tidptr,
2926 int __user *, child_tidptr,
2927 unsigned long, tls)
2928#endif
2929{
2930 struct kernel_clone_args args = {
2931 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
2932 .pidfd = parent_tidptr,
2933 .child_tid = child_tidptr,
2934 .parent_tid = parent_tidptr,
2935 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
2936 .stack = newsp,
2937 .tls = tls,
2938 };
2939
2940 return kernel_clone(&args);
2941}
2942#endif
2943
2944#ifdef __ARCH_WANT_SYS_CLONE3
2945
2946noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2947 struct clone_args __user *uargs,
2948 size_t usize)
2949{
2950 int err;
2951 struct clone_args args;
2952 pid_t *kset_tid = kargs->set_tid;
2953
2954 BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2955 CLONE_ARGS_SIZE_VER0);
2956 BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2957 CLONE_ARGS_SIZE_VER1);
2958 BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2959 CLONE_ARGS_SIZE_VER2);
2960 BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2961
2962 if (unlikely(usize > PAGE_SIZE))
2963 return -E2BIG;
2964 if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2965 return -EINVAL;
2966
2967 err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2968 if (err)
2969 return err;
2970
2971 if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2972 return -EINVAL;
2973
2974 if (unlikely(!args.set_tid && args.set_tid_size > 0))
2975 return -EINVAL;
2976
2977 if (unlikely(args.set_tid && args.set_tid_size == 0))
2978 return -EINVAL;
2979
2980 /*
2981 * Verify that higher 32bits of exit_signal are unset and that
2982 * it is a valid signal
2983 */
2984 if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2985 !valid_signal(args.exit_signal)))
2986 return -EINVAL;
2987
2988 if ((args.flags & CLONE_INTO_CGROUP) &&
2989 (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2990 return -EINVAL;
2991
2992 *kargs = (struct kernel_clone_args){
2993 .flags = args.flags,
2994 .pidfd = u64_to_user_ptr(args.pidfd),
2995 .child_tid = u64_to_user_ptr(args.child_tid),
2996 .parent_tid = u64_to_user_ptr(args.parent_tid),
2997 .exit_signal = args.exit_signal,
2998 .stack = args.stack,
2999 .stack_size = args.stack_size,
3000 .tls = args.tls,
3001 .set_tid_size = args.set_tid_size,
3002 .cgroup = args.cgroup,
3003 };
3004
3005 if (args.set_tid &&
3006 copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3007 (kargs->set_tid_size * sizeof(pid_t))))
3008 return -EFAULT;
3009
3010 kargs->set_tid = kset_tid;
3011
3012 return 0;
3013}
3014
3015/**
3016 * clone3_stack_valid - check and prepare stack
3017 * @kargs: kernel clone args
3018 *
3019 * Verify that the stack arguments userspace gave us are sane.
3020 * In addition, set the stack direction for userspace since it's easy for us to
3021 * determine.
3022 */
3023static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3024{
3025 if (kargs->stack == 0) {
3026 if (kargs->stack_size > 0)
3027 return false;
3028 } else {
3029 if (kargs->stack_size == 0)
3030 return false;
3031
3032 if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3033 return false;
3034
3035#if !defined(CONFIG_STACK_GROWSUP)
3036 kargs->stack += kargs->stack_size;
3037#endif
3038 }
3039
3040 return true;
3041}
3042
3043static bool clone3_args_valid(struct kernel_clone_args *kargs)
3044{
3045 /* Verify that no unknown flags are passed along. */
3046 if (kargs->flags &
3047 ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3048 return false;
3049
3050 /*
3051 * - make the CLONE_DETACHED bit reusable for clone3
3052 * - make the CSIGNAL bits reusable for clone3
3053 */
3054 if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3055 return false;
3056
3057 if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3058 (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3059 return false;
3060
3061 if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3062 kargs->exit_signal)
3063 return false;
3064
3065 if (!clone3_stack_valid(kargs))
3066 return false;
3067
3068 return true;
3069}
3070
3071/**
3072 * sys_clone3 - create a new process with specific properties
3073 * @uargs: argument structure
3074 * @size: size of @uargs
3075 *
3076 * clone3() is the extensible successor to clone()/clone2().
3077 * It takes a struct as argument that is versioned by its size.
3078 *
3079 * Return: On success, a positive PID for the child process.
3080 * On error, a negative errno number.
3081 */
3082SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3083{
3084 int err;
3085
3086 struct kernel_clone_args kargs;
3087 pid_t set_tid[MAX_PID_NS_LEVEL];
3088
3089 kargs.set_tid = set_tid;
3090
3091 err = copy_clone_args_from_user(&kargs, uargs, size);
3092 if (err)
3093 return err;
3094
3095 if (!clone3_args_valid(&kargs))
3096 return -EINVAL;
3097
3098 return kernel_clone(&kargs);
3099}
3100#endif
3101
3102void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3103{
3104 struct task_struct *leader, *parent, *child;
3105 int res;
3106
3107 read_lock(&tasklist_lock);
3108 leader = top = top->group_leader;
3109down:
3110 for_each_thread(leader, parent) {
3111 list_for_each_entry(child, &parent->children, sibling) {
3112 res = visitor(child, data);
3113 if (res) {
3114 if (res < 0)
3115 goto out;
3116 leader = child;
3117 goto down;
3118 }
3119up:
3120 ;
3121 }
3122 }
3123
3124 if (leader != top) {
3125 child = leader;
3126 parent = child->real_parent;
3127 leader = parent->group_leader;
3128 goto up;
3129 }
3130out:
3131 read_unlock(&tasklist_lock);
3132}
3133
3134#ifndef ARCH_MIN_MMSTRUCT_ALIGN
3135#define ARCH_MIN_MMSTRUCT_ALIGN 0
3136#endif
3137
3138static void sighand_ctor(void *data)
3139{
3140 struct sighand_struct *sighand = data;
3141
3142 spin_lock_init(&sighand->siglock);
3143 init_waitqueue_head(&sighand->signalfd_wqh);
3144}
3145
3146void __init mm_cache_init(void)
3147{
3148 unsigned int mm_size;
3149
3150 /*
3151 * The mm_cpumask is located at the end of mm_struct, and is
3152 * dynamically sized based on the maximum CPU number this system
3153 * can have, taking hotplug into account (nr_cpu_ids).
3154 */
3155 mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3156
3157 mm_cachep = kmem_cache_create_usercopy("mm_struct",
3158 mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3159 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3160 offsetof(struct mm_struct, saved_auxv),
3161 sizeof_field(struct mm_struct, saved_auxv),
3162 NULL);
3163}
3164
3165void __init proc_caches_init(void)
3166{
3167 sighand_cachep = kmem_cache_create("sighand_cache",
3168 sizeof(struct sighand_struct), 0,
3169 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3170 SLAB_ACCOUNT, sighand_ctor);
3171 signal_cachep = kmem_cache_create("signal_cache",
3172 sizeof(struct signal_struct), 0,
3173 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3174 NULL);
3175 files_cachep = kmem_cache_create("files_cache",
3176 sizeof(struct files_struct), 0,
3177 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3178 NULL);
3179 fs_cachep = kmem_cache_create("fs_cache",
3180 sizeof(struct fs_struct), 0,
3181 SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3182 NULL);
3183
3184 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3185#ifdef CONFIG_PER_VMA_LOCK
3186 vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3187#endif
3188 mmap_init();
3189 nsproxy_cache_init();
3190}
3191
3192/*
3193 * Check constraints on flags passed to the unshare system call.
3194 */
3195static int check_unshare_flags(unsigned long unshare_flags)
3196{
3197 if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3198 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3199 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3200 CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3201 CLONE_NEWTIME))
3202 return -EINVAL;
3203 /*
3204 * Not implemented, but pretend it works if there is nothing
3205 * to unshare. Note that unsharing the address space or the
3206 * signal handlers also need to unshare the signal queues (aka
3207 * CLONE_THREAD).
3208 */
3209 if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3210 if (!thread_group_empty(current))
3211 return -EINVAL;
3212 }
3213 if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3214 if (refcount_read(¤t->sighand->count) > 1)
3215 return -EINVAL;
3216 }
3217 if (unshare_flags & CLONE_VM) {
3218 if (!current_is_single_threaded())
3219 return -EINVAL;
3220 }
3221
3222 return 0;
3223}
3224
3225/*
3226 * Unshare the filesystem structure if it is being shared
3227 */
3228static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3229{
3230 struct fs_struct *fs = current->fs;
3231
3232 if (!(unshare_flags & CLONE_FS) || !fs)
3233 return 0;
3234
3235 /* don't need lock here; in the worst case we'll do useless copy */
3236 if (fs->users == 1)
3237 return 0;
3238
3239 *new_fsp = copy_fs_struct(fs);
3240 if (!*new_fsp)
3241 return -ENOMEM;
3242
3243 return 0;
3244}
3245
3246/*
3247 * Unshare file descriptor table if it is being shared
3248 */
3249int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
3250 struct files_struct **new_fdp)
3251{
3252 struct files_struct *fd = current->files;
3253 int error = 0;
3254
3255 if ((unshare_flags & CLONE_FILES) &&
3256 (fd && atomic_read(&fd->count) > 1)) {
3257 *new_fdp = dup_fd(fd, max_fds, &error);
3258 if (!*new_fdp)
3259 return error;
3260 }
3261
3262 return 0;
3263}
3264
3265/*
3266 * unshare allows a process to 'unshare' part of the process
3267 * context which was originally shared using clone. copy_*
3268 * functions used by kernel_clone() cannot be used here directly
3269 * because they modify an inactive task_struct that is being
3270 * constructed. Here we are modifying the current, active,
3271 * task_struct.
3272 */
3273int ksys_unshare(unsigned long unshare_flags)
3274{
3275 struct fs_struct *fs, *new_fs = NULL;
3276 struct files_struct *new_fd = NULL;
3277 struct cred *new_cred = NULL;
3278 struct nsproxy *new_nsproxy = NULL;
3279 int do_sysvsem = 0;
3280 int err;
3281
3282 /*
3283 * If unsharing a user namespace must also unshare the thread group
3284 * and unshare the filesystem root and working directories.
3285 */
3286 if (unshare_flags & CLONE_NEWUSER)
3287 unshare_flags |= CLONE_THREAD | CLONE_FS;
3288 /*
3289 * If unsharing vm, must also unshare signal handlers.
3290 */
3291 if (unshare_flags & CLONE_VM)
3292 unshare_flags |= CLONE_SIGHAND;
3293 /*
3294 * If unsharing a signal handlers, must also unshare the signal queues.
3295 */
3296 if (unshare_flags & CLONE_SIGHAND)
3297 unshare_flags |= CLONE_THREAD;
3298 /*
3299 * If unsharing namespace, must also unshare filesystem information.
3300 */
3301 if (unshare_flags & CLONE_NEWNS)
3302 unshare_flags |= CLONE_FS;
3303
3304 err = check_unshare_flags(unshare_flags);
3305 if (err)
3306 goto bad_unshare_out;
3307 /*
3308 * CLONE_NEWIPC must also detach from the undolist: after switching
3309 * to a new ipc namespace, the semaphore arrays from the old
3310 * namespace are unreachable.
3311 */
3312 if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3313 do_sysvsem = 1;
3314 err = unshare_fs(unshare_flags, &new_fs);
3315 if (err)
3316 goto bad_unshare_out;
3317 err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
3318 if (err)
3319 goto bad_unshare_cleanup_fs;
3320 err = unshare_userns(unshare_flags, &new_cred);
3321 if (err)
3322 goto bad_unshare_cleanup_fd;
3323 err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3324 new_cred, new_fs);
3325 if (err)
3326 goto bad_unshare_cleanup_cred;
3327
3328 if (new_cred) {
3329 err = set_cred_ucounts(new_cred);
3330 if (err)
3331 goto bad_unshare_cleanup_cred;
3332 }
3333
3334 if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3335 if (do_sysvsem) {
3336 /*
3337 * CLONE_SYSVSEM is equivalent to sys_exit().
3338 */
3339 exit_sem(current);
3340 }
3341 if (unshare_flags & CLONE_NEWIPC) {
3342 /* Orphan segments in old ns (see sem above). */
3343 exit_shm(current);
3344 shm_init_task(current);
3345 }
3346
3347 if (new_nsproxy)
3348 switch_task_namespaces(current, new_nsproxy);
3349
3350 task_lock(current);
3351
3352 if (new_fs) {
3353 fs = current->fs;
3354 spin_lock(&fs->lock);
3355 current->fs = new_fs;
3356 if (--fs->users)
3357 new_fs = NULL;
3358 else
3359 new_fs = fs;
3360 spin_unlock(&fs->lock);
3361 }
3362
3363 if (new_fd)
3364 swap(current->files, new_fd);
3365
3366 task_unlock(current);
3367
3368 if (new_cred) {
3369 /* Install the new user namespace */
3370 commit_creds(new_cred);
3371 new_cred = NULL;
3372 }
3373 }
3374
3375 perf_event_namespaces(current);
3376
3377bad_unshare_cleanup_cred:
3378 if (new_cred)
3379 put_cred(new_cred);
3380bad_unshare_cleanup_fd:
3381 if (new_fd)
3382 put_files_struct(new_fd);
3383
3384bad_unshare_cleanup_fs:
3385 if (new_fs)
3386 free_fs_struct(new_fs);
3387
3388bad_unshare_out:
3389 return err;
3390}
3391
3392SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3393{
3394 return ksys_unshare(unshare_flags);
3395}
3396
3397/*
3398 * Helper to unshare the files of the current task.
3399 * We don't want to expose copy_files internals to
3400 * the exec layer of the kernel.
3401 */
3402
3403int unshare_files(void)
3404{
3405 struct task_struct *task = current;
3406 struct files_struct *old, *copy = NULL;
3407 int error;
3408
3409 error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, ©);
3410 if (error || !copy)
3411 return error;
3412
3413 old = task->files;
3414 task_lock(task);
3415 task->files = copy;
3416 task_unlock(task);
3417 put_files_struct(old);
3418 return 0;
3419}
3420
3421int sysctl_max_threads(struct ctl_table *table, int write,
3422 void *buffer, size_t *lenp, loff_t *ppos)
3423{
3424 struct ctl_table t;
3425 int ret;
3426 int threads = max_threads;
3427 int min = 1;
3428 int max = MAX_THREADS;
3429
3430 t = *table;
3431 t.data = &threads;
3432 t.extra1 = &min;
3433 t.extra2 = &max;
3434
3435 ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3436 if (ret || !write)
3437 return ret;
3438
3439 max_threads = threads;
3440
3441 return 0;
3442}