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