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