* [jpoimboe:objtool-diff 2/2] kernel/fork.c: linux/livepatch.h is included more than once.
@ 2024-05-25 6:40 kernel test robot
2024-05-27 1:47 ` Liu, Yujie
0 siblings, 1 reply; 2+ messages in thread
From: kernel test robot @ 2024-05-25 6:40 UTC (permalink / raw)
To: oe-kbuild; +Cc: lkp
::::::
:::::: Manual check reason: "low confidence bisect report"
::::::
BCC: lkp@intel.com
CC: oe-kbuild-all@lists.linux.dev
TO: Josh Poimboeuf <jpoimboe@kernel.org>
tree: https://git.kernel.org/pub/scm/linux/kernel/git/jpoimboe/linux.git objtool-diff
head: 745009dc796e56fc87e911138d801679ecd3576e
commit: 745009dc796e56fc87e911138d801679ecd3576e [2/2] test
:::::: branch date: 6 hours ago
:::::: commit date: 6 hours ago
compiler: clang version 18.1.5 (https://github.com/llvm/llvm-project 617a15a9eac96088ae5e9134248d8236e34b91b1)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/r/202405251400.UdnwcgiL-lkp@intel.com/
includecheck warnings: (new ones prefixed by >>)
>> kernel/fork.c: linux/livepatch.h is included more than once.
--
>> drivers/input/joydev.c: linux/module.h is included more than once.
--
>> drivers/input/misc/pcspkr.c: linux/module.h is included more than once.
vim +93 kernel/fork.c
> 93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.h>
100 #include <linux/stackprotector.h>
101 #include <linux/user_events.h>
102 #include <linux/iommu.h>
103 #include <linux/rseq.h>
104 #include <uapi/linux/pidfd.h>
105 #include <linux/pidfs.h>
106
107 #include <asm/pgalloc.h>
108 #include <linux/uaccess.h>
109 #include <asm/mmu_context.h>
110 #include <asm/cacheflush.h>
111 #include <asm/tlbflush.h>
112
113 #include <trace/events/sched.h>
114
115 #define CREATE_TRACE_POINTS
116 #include <trace/events/task.h>
117
118 /*
119 * Minimum number of threads to boot the kernel
120 */
121 #define MIN_THREADS 20
122
123 /*
124 * Maximum number of threads
125 */
126 #define MAX_THREADS FUTEX_TID_MASK
127
128 /*
129 * Protected counters by write_lock_irq(&tasklist_lock)
130 */
131 unsigned long total_forks; /* Handle normal Linux uptimes. */
132 int nr_threads; /* The idle threads do not count.. */
133
134 static int max_threads; /* tunable limit on nr_threads */
135
136 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
137
138 static const char * const resident_page_types[] = {
139 NAMED_ARRAY_INDEX(MM_FILEPAGES),
140 NAMED_ARRAY_INDEX(MM_ANONPAGES),
141 NAMED_ARRAY_INDEX(MM_SWAPENTS),
142 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
143 };
144
145 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
146
147 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
148
149 #ifdef CONFIG_PROVE_RCU
150 int lockdep_tasklist_lock_is_held(void)
151 {
152 return lockdep_is_held(&tasklist_lock);
153 }
154 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
155 #endif /* #ifdef CONFIG_PROVE_RCU */
156
157 int nr_processes(void)
158 {
159 int cpu;
160 int total = 0;
161
162 for_each_possible_cpu(cpu)
163 total += per_cpu(process_counts, cpu);
164
165 return total;
166 }
167
168 void __weak arch_release_task_struct(struct task_struct *tsk)
169 {
170 }
171
172 static struct kmem_cache *task_struct_cachep;
173
174 static inline struct task_struct *alloc_task_struct_node(int node)
175 {
176 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
177 }
178
179 static inline void free_task_struct(struct task_struct *tsk)
180 {
181 kmem_cache_free(task_struct_cachep, tsk);
182 }
183
184 /*
185 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
186 * kmemcache based allocator.
187 */
188 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189
190 # ifdef CONFIG_VMAP_STACK
191 /*
192 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
193 * flush. Try to minimize the number of calls by caching stacks.
194 */
195 #define NR_CACHED_STACKS 2
196 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
197
198 struct vm_stack {
199 struct rcu_head rcu;
200 struct vm_struct *stack_vm_area;
201 };
202
203 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
204 {
205 unsigned int i;
206
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
209 continue;
210 return true;
211 }
212 return false;
213 }
214
215 static void thread_stack_free_rcu(struct rcu_head *rh)
216 {
217 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218
219 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
220 return;
221
222 vfree(vm_stack);
223 }
224
225 static void thread_stack_delayed_free(struct task_struct *tsk)
226 {
227 struct vm_stack *vm_stack = tsk->stack;
228
229 vm_stack->stack_vm_area = tsk->stack_vm_area;
230 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231 }
232
233 static int free_vm_stack_cache(unsigned int cpu)
234 {
235 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 int i;
237
238 for (i = 0; i < NR_CACHED_STACKS; i++) {
239 struct vm_struct *vm_stack = cached_vm_stacks[i];
240
241 if (!vm_stack)
242 continue;
243
244 vfree(vm_stack->addr);
245 cached_vm_stacks[i] = NULL;
246 }
247
248 return 0;
249 }
250
251 static int memcg_charge_kernel_stack(struct vm_struct *vm)
252 {
253 int i;
254 int ret;
255 int nr_charged = 0;
256
257 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258
259 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
260 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
261 if (ret)
262 goto err;
263 nr_charged++;
264 }
265 return 0;
266 err:
267 for (i = 0; i < nr_charged; i++)
268 memcg_kmem_uncharge_page(vm->pages[i], 0);
269 return ret;
270 }
271
272 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
273 {
274 struct vm_struct *vm;
275 void *stack;
276 int i;
277
278 for (i = 0; i < NR_CACHED_STACKS; i++) {
279 struct vm_struct *s;
280
281 s = this_cpu_xchg(cached_stacks[i], NULL);
282
283 if (!s)
284 continue;
285
286 /* Reset stack metadata. */
287 kasan_unpoison_range(s->addr, THREAD_SIZE);
288
289 stack = kasan_reset_tag(s->addr);
290
291 /* Clear stale pointers from reused stack. */
292 memset(stack, 0, THREAD_SIZE);
293
294 if (memcg_charge_kernel_stack(s)) {
295 vfree(s->addr);
296 return -ENOMEM;
297 }
298
299 tsk->stack_vm_area = s;
300 tsk->stack = stack;
301 return 0;
302 }
303
304 /*
305 * Allocated stacks are cached and later reused by new threads,
306 * so memcg accounting is performed manually on assigning/releasing
307 * stacks to tasks. Drop __GFP_ACCOUNT.
308 */
309 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
310 VMALLOC_START, VMALLOC_END,
311 THREADINFO_GFP & ~__GFP_ACCOUNT,
312 PAGE_KERNEL,
313 0, node, __builtin_return_address(0));
314 if (!stack)
315 return -ENOMEM;
316
317 vm = find_vm_area(stack);
318 if (memcg_charge_kernel_stack(vm)) {
319 vfree(stack);
320 return -ENOMEM;
321 }
322 /*
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
326 */
327 tsk->stack_vm_area = vm;
328 stack = kasan_reset_tag(stack);
329 tsk->stack = stack;
330 return 0;
331 }
332
333 static void free_thread_stack(struct task_struct *tsk)
334 {
335 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
337
338 tsk->stack = NULL;
339 tsk->stack_vm_area = NULL;
340 }
341
342 # else /* !CONFIG_VMAP_STACK */
343
344 static void thread_stack_free_rcu(struct rcu_head *rh)
345 {
346 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
347 }
348
349 static void thread_stack_delayed_free(struct task_struct *tsk)
350 {
351 struct rcu_head *rh = tsk->stack;
352
353 call_rcu(rh, thread_stack_free_rcu);
354 }
355
356 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
357 {
358 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
359 THREAD_SIZE_ORDER);
360
361 if (likely(page)) {
362 tsk->stack = kasan_reset_tag(page_address(page));
363 return 0;
364 }
365 return -ENOMEM;
366 }
367
368 static void free_thread_stack(struct task_struct *tsk)
369 {
370 thread_stack_delayed_free(tsk);
371 tsk->stack = NULL;
372 }
373
374 # endif /* CONFIG_VMAP_STACK */
375 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
376
377 static struct kmem_cache *thread_stack_cache;
378
379 static void thread_stack_free_rcu(struct rcu_head *rh)
380 {
381 kmem_cache_free(thread_stack_cache, rh);
382 }
383
384 static void thread_stack_delayed_free(struct task_struct *tsk)
385 {
386 struct rcu_head *rh = tsk->stack;
387
388 call_rcu(rh, thread_stack_free_rcu);
389 }
390
391 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
392 {
393 unsigned long *stack;
394 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
395 stack = kasan_reset_tag(stack);
396 tsk->stack = stack;
397 return stack ? 0 : -ENOMEM;
398 }
399
400 static void free_thread_stack(struct task_struct *tsk)
401 {
402 thread_stack_delayed_free(tsk);
403 tsk->stack = NULL;
404 }
405
406 void thread_stack_cache_init(void)
407 {
408 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
409 THREAD_SIZE, THREAD_SIZE, 0, 0,
410 THREAD_SIZE, NULL);
411 BUG_ON(thread_stack_cache == NULL);
412 }
413
414 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415
416 /* SLAB cache for signal_struct structures (tsk->signal) */
417 static struct kmem_cache *signal_cachep;
418
419 /* SLAB cache for sighand_struct structures (tsk->sighand) */
420 struct kmem_cache *sighand_cachep;
421
422 /* SLAB cache for files_struct structures (tsk->files) */
423 struct kmem_cache *files_cachep;
424
425 /* SLAB cache for fs_struct structures (tsk->fs) */
426 struct kmem_cache *fs_cachep;
427
428 /* SLAB cache for vm_area_struct structures */
429 static struct kmem_cache *vm_area_cachep;
430
431 /* SLAB cache for mm_struct structures (tsk->mm) */
432 static struct kmem_cache *mm_cachep;
433
434 #ifdef CONFIG_PER_VMA_LOCK
435
436 /* SLAB cache for vm_area_struct.lock */
437 static struct kmem_cache *vma_lock_cachep;
438
439 static bool vma_lock_alloc(struct vm_area_struct *vma)
440 {
441 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
442 if (!vma->vm_lock)
443 return false;
444
445 init_rwsem(&vma->vm_lock->lock);
446 vma->vm_lock_seq = -1;
447
448 return true;
449 }
450
451 static inline void vma_lock_free(struct vm_area_struct *vma)
452 {
453 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
454 }
455
456 #else /* CONFIG_PER_VMA_LOCK */
457
458 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
459 static inline void vma_lock_free(struct vm_area_struct *vma) {}
460
461 #endif /* CONFIG_PER_VMA_LOCK */
462
463 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
464 {
465 struct vm_area_struct *vma;
466
467 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
468 if (!vma)
469 return NULL;
470
471 vma_init(vma, mm);
472 if (!vma_lock_alloc(vma)) {
473 kmem_cache_free(vm_area_cachep, vma);
474 return NULL;
475 }
476
477 return vma;
478 }
479
480 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
481 {
482 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
483
484 if (!new)
485 return NULL;
486
487 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
488 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
489 /*
490 * orig->shared.rb may be modified concurrently, but the clone
491 * will be reinitialized.
492 */
493 data_race(memcpy(new, orig, sizeof(*new)));
494 if (!vma_lock_alloc(new)) {
495 kmem_cache_free(vm_area_cachep, new);
496 return NULL;
497 }
498 INIT_LIST_HEAD(&new->anon_vma_chain);
499 vma_numab_state_init(new);
500 dup_anon_vma_name(orig, new);
501
502 return new;
503 }
504
505 void __vm_area_free(struct vm_area_struct *vma)
506 {
507 vma_numab_state_free(vma);
508 free_anon_vma_name(vma);
509 vma_lock_free(vma);
510 kmem_cache_free(vm_area_cachep, vma);
511 }
512
513 #ifdef CONFIG_PER_VMA_LOCK
514 static void vm_area_free_rcu_cb(struct rcu_head *head)
515 {
516 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
517 vm_rcu);
518
519 /* The vma should not be locked while being destroyed. */
520 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
521 __vm_area_free(vma);
522 }
523 #endif
524
525 void vm_area_free(struct vm_area_struct *vma)
526 {
527 #ifdef CONFIG_PER_VMA_LOCK
528 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
529 #else
530 __vm_area_free(vma);
531 #endif
532 }
533
534 static void account_kernel_stack(struct task_struct *tsk, int account)
535 {
536 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
537 struct vm_struct *vm = task_stack_vm_area(tsk);
538 int i;
539
540 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
541 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
542 account * (PAGE_SIZE / 1024));
543 } else {
544 void *stack = task_stack_page(tsk);
545
546 /* All stack pages are in the same node. */
547 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
548 account * (THREAD_SIZE / 1024));
549 }
550 }
551
552 void exit_task_stack_account(struct task_struct *tsk)
553 {
554 account_kernel_stack(tsk, -1);
555
556 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
557 struct vm_struct *vm;
558 int i;
559
560 vm = task_stack_vm_area(tsk);
561 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
562 memcg_kmem_uncharge_page(vm->pages[i], 0);
563 }
564 }
565
566 static void release_task_stack(struct task_struct *tsk)
567 {
568 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
569 return; /* Better to leak the stack than to free prematurely */
570
571 free_thread_stack(tsk);
572 }
573
574 #ifdef CONFIG_THREAD_INFO_IN_TASK
575 void put_task_stack(struct task_struct *tsk)
576 {
577 if (refcount_dec_and_test(&tsk->stack_refcount))
578 release_task_stack(tsk);
579 }
580 #endif
581
582 void free_task(struct task_struct *tsk)
583 {
584 #ifdef CONFIG_SECCOMP
585 WARN_ON_ONCE(tsk->seccomp.filter);
586 #endif
587 release_user_cpus_ptr(tsk);
588 scs_release(tsk);
589
590 #ifndef CONFIG_THREAD_INFO_IN_TASK
591 /*
592 * The task is finally done with both the stack and thread_info,
593 * so free both.
594 */
595 release_task_stack(tsk);
596 #else
597 /*
598 * If the task had a separate stack allocation, it should be gone
599 * by now.
600 */
601 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
602 #endif
603 rt_mutex_debug_task_free(tsk);
604 ftrace_graph_exit_task(tsk);
605 arch_release_task_struct(tsk);
606 if (tsk->flags & PF_KTHREAD)
607 free_kthread_struct(tsk);
608 bpf_task_storage_free(tsk);
609 free_task_struct(tsk);
610 }
611 EXPORT_SYMBOL(free_task);
612
613 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
614 {
615 struct file *exe_file;
616
617 exe_file = get_mm_exe_file(oldmm);
618 RCU_INIT_POINTER(mm->exe_file, exe_file);
619 /*
620 * We depend on the oldmm having properly denied write access to the
621 * exe_file already.
622 */
623 if (exe_file && deny_write_access(exe_file))
624 pr_warn_once("deny_write_access() failed in %s\n", __func__);
625 }
626
627 #ifdef CONFIG_MMU
628 static __latent_entropy int dup_mmap(struct mm_struct *mm,
629 struct mm_struct *oldmm)
630 {
631 struct vm_area_struct *mpnt, *tmp;
632 int retval;
633 unsigned long charge = 0;
634 LIST_HEAD(uf);
635 VMA_ITERATOR(vmi, mm, 0);
636
637 uprobe_start_dup_mmap();
638 if (mmap_write_lock_killable(oldmm)) {
639 retval = -EINTR;
640 goto fail_uprobe_end;
641 }
642 flush_cache_dup_mm(oldmm);
643 uprobe_dup_mmap(oldmm, mm);
644 /*
645 * Not linked in yet - no deadlock potential:
646 */
647 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
648
649 /* No ordering required: file already has been exposed. */
650 dup_mm_exe_file(mm, oldmm);
651
652 mm->total_vm = oldmm->total_vm;
653 mm->data_vm = oldmm->data_vm;
654 mm->exec_vm = oldmm->exec_vm;
655 mm->stack_vm = oldmm->stack_vm;
656
657 retval = ksm_fork(mm, oldmm);
658 if (retval)
659 goto out;
660 khugepaged_fork(mm, oldmm);
661
662 /* Use __mt_dup() to efficiently build an identical maple tree. */
663 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
664 if (unlikely(retval))
665 goto out;
666
667 mt_clear_in_rcu(vmi.mas.tree);
668 for_each_vma(vmi, mpnt) {
669 struct file *file;
670
671 vma_start_write(mpnt);
672 if (mpnt->vm_flags & VM_DONTCOPY) {
673 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
674 mpnt->vm_end, GFP_KERNEL);
675 if (retval)
676 goto loop_out;
677
678 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
679 continue;
680 }
681 charge = 0;
682 /*
683 * Don't duplicate many vmas if we've been oom-killed (for
684 * example)
685 */
686 if (fatal_signal_pending(current)) {
687 retval = -EINTR;
688 goto loop_out;
689 }
690 if (mpnt->vm_flags & VM_ACCOUNT) {
691 unsigned long len = vma_pages(mpnt);
692
693 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
694 goto fail_nomem;
695 charge = len;
696 }
697 tmp = vm_area_dup(mpnt);
698 if (!tmp)
699 goto fail_nomem;
700 retval = vma_dup_policy(mpnt, tmp);
701 if (retval)
702 goto fail_nomem_policy;
703 tmp->vm_mm = mm;
704 retval = dup_userfaultfd(tmp, &uf);
705 if (retval)
706 goto fail_nomem_anon_vma_fork;
707 if (tmp->vm_flags & VM_WIPEONFORK) {
708 /*
709 * VM_WIPEONFORK gets a clean slate in the child.
710 * Don't prepare anon_vma until fault since we don't
711 * copy page for current vma.
712 */
713 tmp->anon_vma = NULL;
714 } else if (anon_vma_fork(tmp, mpnt))
715 goto fail_nomem_anon_vma_fork;
716 vm_flags_clear(tmp, VM_LOCKED_MASK);
717 /*
718 * Copy/update hugetlb private vma information.
719 */
720 if (is_vm_hugetlb_page(tmp))
721 hugetlb_dup_vma_private(tmp);
722
723 /*
724 * Link the vma into the MT. After using __mt_dup(), memory
725 * allocation is not necessary here, so it cannot fail.
726 */
727 vma_iter_bulk_store(&vmi, tmp);
728
729 mm->map_count++;
730
731 if (tmp->vm_ops && tmp->vm_ops->open)
732 tmp->vm_ops->open(tmp);
733
734 file = tmp->vm_file;
735 if (file) {
736 struct address_space *mapping = file->f_mapping;
737
738 get_file(file);
739 i_mmap_lock_write(mapping);
740 if (vma_is_shared_maywrite(tmp))
741 mapping_allow_writable(mapping);
742 flush_dcache_mmap_lock(mapping);
743 /* insert tmp into the share list, just after mpnt */
744 vma_interval_tree_insert_after(tmp, mpnt,
745 &mapping->i_mmap);
746 flush_dcache_mmap_unlock(mapping);
747 i_mmap_unlock_write(mapping);
748 }
749
750 if (!(tmp->vm_flags & VM_WIPEONFORK))
751 retval = copy_page_range(tmp, mpnt);
752
753 if (retval) {
754 mpnt = vma_next(&vmi);
755 goto loop_out;
756 }
757 }
758 /* a new mm has just been created */
759 retval = arch_dup_mmap(oldmm, mm);
760 loop_out:
761 vma_iter_free(&vmi);
762 if (!retval) {
763 mt_set_in_rcu(vmi.mas.tree);
764 } else if (mpnt) {
765 /*
766 * The entire maple tree has already been duplicated. If the
767 * mmap duplication fails, mark the failure point with
768 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
769 * stop releasing VMAs that have not been duplicated after this
770 * point.
771 */
772 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
773 mas_store(&vmi.mas, XA_ZERO_ENTRY);
774 }
775 out:
776 mmap_write_unlock(mm);
777 flush_tlb_mm(oldmm);
778 mmap_write_unlock(oldmm);
779 dup_userfaultfd_complete(&uf);
780 fail_uprobe_end:
781 uprobe_end_dup_mmap();
782 return retval;
783
784 fail_nomem_anon_vma_fork:
785 mpol_put(vma_policy(tmp));
786 fail_nomem_policy:
787 vm_area_free(tmp);
788 fail_nomem:
789 retval = -ENOMEM;
790 vm_unacct_memory(charge);
791 goto loop_out;
792 }
793
794 static inline int mm_alloc_pgd(struct mm_struct *mm)
795 {
796 mm->pgd = pgd_alloc(mm);
797 if (unlikely(!mm->pgd))
798 return -ENOMEM;
799 return 0;
800 }
801
802 static inline void mm_free_pgd(struct mm_struct *mm)
803 {
804 pgd_free(mm, mm->pgd);
805 }
806 #else
807 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
808 {
809 mmap_write_lock(oldmm);
810 dup_mm_exe_file(mm, oldmm);
811 mmap_write_unlock(oldmm);
812 return 0;
813 }
814 #define mm_alloc_pgd(mm) (0)
815 #define mm_free_pgd(mm)
816 #endif /* CONFIG_MMU */
817
818 static void check_mm(struct mm_struct *mm)
819 {
820 int i;
821
822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
823 "Please make sure 'struct resident_page_types[]' is updated as well");
824
825 for (i = 0; i < NR_MM_COUNTERS; i++) {
826 long x = percpu_counter_sum(&mm->rss_stat[i]);
827
828 if (unlikely(x))
829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
830 mm, resident_page_types[i], x);
831 }
832
833 if (mm_pgtables_bytes(mm))
834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
835 mm_pgtables_bytes(mm));
836
837 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
839 #endif
840 }
841
842 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
843 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
844
845 static void do_check_lazy_tlb(void *arg)
846 {
847 struct mm_struct *mm = arg;
848
849 WARN_ON_ONCE(current->active_mm == mm);
850 }
851
852 static void do_shoot_lazy_tlb(void *arg)
853 {
854 struct mm_struct *mm = arg;
855
856 if (current->active_mm == mm) {
857 WARN_ON_ONCE(current->mm);
858 current->active_mm = &init_mm;
859 switch_mm(mm, &init_mm, current);
860 }
861 }
862
863 static void cleanup_lazy_tlbs(struct mm_struct *mm)
864 {
865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
866 /*
867 * In this case, lazy tlb mms are refounted and would not reach
868 * __mmdrop until all CPUs have switched away and mmdrop()ed.
869 */
870 return;
871 }
872
873 /*
874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
875 * requires lazy mm users to switch to another mm when the refcount
876 * drops to zero, before the mm is freed. This requires IPIs here to
877 * switch kernel threads to init_mm.
878 *
879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
880 * switch with the final userspace teardown TLB flush which leaves the
881 * mm lazy on this CPU but no others, reducing the need for additional
882 * IPIs here. There are cases where a final IPI is still required here,
883 * such as the final mmdrop being performed on a different CPU than the
884 * one exiting, or kernel threads using the mm when userspace exits.
885 *
886 * IPI overheads have not found to be expensive, but they could be
887 * reduced in a number of possible ways, for example (roughly
888 * increasing order of complexity):
889 * - The last lazy reference created by exit_mm() could instead switch
890 * to init_mm, however it's probable this will run on the same CPU
891 * immediately afterwards, so this may not reduce IPIs much.
892 * - A batch of mms requiring IPIs could be gathered and freed at once.
893 * - CPUs store active_mm where it can be remotely checked without a
894 * lock, to filter out false-positives in the cpumask.
895 * - After mm_users or mm_count reaches zero, switching away from the
896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
897 * with some batching or delaying of the final IPIs.
898 * - A delayed freeing and RCU-like quiescing sequence based on mm
899 * switching to avoid IPIs completely.
900 */
901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
904 }
905
906 /*
907 * Called when the last reference to the mm
908 * is dropped: either by a lazy thread or by
909 * mmput. Free the page directory and the mm.
910 */
911 void __mmdrop(struct mm_struct *mm)
912 {
913 BUG_ON(mm == &init_mm);
914 WARN_ON_ONCE(mm == current->mm);
915
916 /* Ensure no CPUs are using this as their lazy tlb mm */
917 cleanup_lazy_tlbs(mm);
918
919 WARN_ON_ONCE(mm == current->active_mm);
920 mm_free_pgd(mm);
921 destroy_context(mm);
922 mmu_notifier_subscriptions_destroy(mm);
923 check_mm(mm);
924 put_user_ns(mm->user_ns);
925 mm_pasid_drop(mm);
926 mm_destroy_cid(mm);
927 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
928
929 free_mm(mm);
930 }
931 EXPORT_SYMBOL_GPL(__mmdrop);
932
933 static void mmdrop_async_fn(struct work_struct *work)
934 {
935 struct mm_struct *mm;
936
937 mm = container_of(work, struct mm_struct, async_put_work);
938 __mmdrop(mm);
939 }
940
941 static void mmdrop_async(struct mm_struct *mm)
942 {
943 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
944 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
945 schedule_work(&mm->async_put_work);
946 }
947 }
948
949 static inline void free_signal_struct(struct signal_struct *sig)
950 {
951 taskstats_tgid_free(sig);
952 sched_autogroup_exit(sig);
953 /*
954 * __mmdrop is not safe to call from softirq context on x86 due to
955 * pgd_dtor so postpone it to the async context
956 */
957 if (sig->oom_mm)
958 mmdrop_async(sig->oom_mm);
959 kmem_cache_free(signal_cachep, sig);
960 }
961
962 static inline void put_signal_struct(struct signal_struct *sig)
963 {
964 if (refcount_dec_and_test(&sig->sigcnt))
965 free_signal_struct(sig);
966 }
967
968 void __put_task_struct(struct task_struct *tsk)
969 {
970 WARN_ON(!tsk->exit_state);
971 WARN_ON(refcount_read(&tsk->usage));
972 WARN_ON(tsk == current);
973
974 io_uring_free(tsk);
975 cgroup_free(tsk);
976 task_numa_free(tsk, true);
977 security_task_free(tsk);
978 exit_creds(tsk);
979 delayacct_tsk_free(tsk);
980 put_signal_struct(tsk->signal);
981 sched_core_free(tsk);
982 free_task(tsk);
983 }
984 EXPORT_SYMBOL_GPL(__put_task_struct);
985
986 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
987 {
988 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
989
990 __put_task_struct(task);
991 }
992 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
993
994 void __init __weak arch_task_cache_init(void) { }
995
996 /*
997 * set_max_threads
998 */
999 static void set_max_threads(unsigned int max_threads_suggested)
1000 {
1001 u64 threads;
1002 unsigned long nr_pages = totalram_pages();
1003
1004 /*
1005 * The number of threads shall be limited such that the thread
1006 * structures may only consume a small part of the available memory.
1007 */
1008 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1009 threads = MAX_THREADS;
1010 else
1011 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1012 (u64) THREAD_SIZE * 8UL);
1013
1014 if (threads > max_threads_suggested)
1015 threads = max_threads_suggested;
1016
1017 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1018 }
1019
1020 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1021 /* Initialized by the architecture: */
1022 int arch_task_struct_size __read_mostly;
1023 #endif
1024
1025 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1026 {
1027 /* Fetch thread_struct whitelist for the architecture. */
1028 arch_thread_struct_whitelist(offset, size);
1029
1030 /*
1031 * Handle zero-sized whitelist or empty thread_struct, otherwise
1032 * adjust offset to position of thread_struct in task_struct.
1033 */
1034 if (unlikely(*size == 0))
1035 *offset = 0;
1036 else
1037 *offset += offsetof(struct task_struct, thread);
1038 }
1039
1040 void __init fork_init(void)
1041 {
1042 int i;
1043 #ifndef ARCH_MIN_TASKALIGN
1044 #define ARCH_MIN_TASKALIGN 0
1045 #endif
1046 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1047 unsigned long useroffset, usersize;
1048
1049 /* create a slab on which task_structs can be allocated */
1050 task_struct_whitelist(&useroffset, &usersize);
1051 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1052 arch_task_struct_size, align,
1053 SLAB_PANIC|SLAB_ACCOUNT,
1054 useroffset, usersize, NULL);
1055
1056 /* do the arch specific task caches init */
1057 arch_task_cache_init();
1058
1059 set_max_threads(MAX_THREADS);
1060
1061 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1064 init_task.signal->rlim[RLIMIT_NPROC];
1065
1066 for (i = 0; i < UCOUNT_COUNTS; i++)
1067 init_user_ns.ucount_max[i] = max_threads/2;
1068
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1073
1074 #ifdef CONFIG_VMAP_STACK
1075 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1076 NULL, free_vm_stack_cache);
1077 #endif
1078
1079 scs_init();
1080
1081 lockdep_init_task(&init_task);
1082 uprobes_init();
1083 }
1084
1085 int __weak arch_dup_task_struct(struct task_struct *dst,
1086 struct task_struct *src)
1087 {
1088 *dst = *src;
1089 return 0;
1090 }
1091
1092 void set_task_stack_end_magic(struct task_struct *tsk)
1093 {
1094 unsigned long *stackend;
1095
1096 stackend = end_of_stack(tsk);
1097 *stackend = STACK_END_MAGIC; /* for overflow detection */
1098 }
1099
1100 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1101 {
1102 struct task_struct *tsk;
1103 int err;
1104
1105 if (node == NUMA_NO_NODE)
1106 node = tsk_fork_get_node(orig);
1107 tsk = alloc_task_struct_node(node);
1108 if (!tsk)
1109 return NULL;
1110
1111 err = arch_dup_task_struct(tsk, orig);
1112 if (err)
1113 goto free_tsk;
1114
1115 err = alloc_thread_stack_node(tsk, node);
1116 if (err)
1117 goto free_tsk;
1118
1119 #ifdef CONFIG_THREAD_INFO_IN_TASK
1120 refcount_set(&tsk->stack_refcount, 1);
1121 #endif
1122 account_kernel_stack(tsk, 1);
1123
1124 err = scs_prepare(tsk, node);
1125 if (err)
1126 goto free_stack;
1127
1128 #ifdef CONFIG_SECCOMP
1129 /*
1130 * We must handle setting up seccomp filters once we're under
1131 * the sighand lock in case orig has changed between now and
1132 * then. Until then, filter must be NULL to avoid messing up
1133 * the usage counts on the error path calling free_task.
1134 */
1135 tsk->seccomp.filter = NULL;
1136 #endif
1137
1138 setup_thread_stack(tsk, orig);
1139 clear_user_return_notifier(tsk);
1140 clear_tsk_need_resched(tsk);
1141 set_task_stack_end_magic(tsk);
1142 clear_syscall_work_syscall_user_dispatch(tsk);
1143
1144 #ifdef CONFIG_STACKPROTECTOR
1145 tsk->stack_canary = get_random_canary();
1146 #endif
1147 if (orig->cpus_ptr == &orig->cpus_mask)
1148 tsk->cpus_ptr = &tsk->cpus_mask;
1149 dup_user_cpus_ptr(tsk, orig, node);
1150
1151 /*
1152 * One for the user space visible state that goes away when reaped.
1153 * One for the scheduler.
1154 */
1155 refcount_set(&tsk->rcu_users, 2);
1156 /* One for the rcu users */
1157 refcount_set(&tsk->usage, 1);
1158 #ifdef CONFIG_BLK_DEV_IO_TRACE
1159 tsk->btrace_seq = 0;
1160 #endif
1161 tsk->splice_pipe = NULL;
1162 tsk->task_frag.page = NULL;
1163 tsk->wake_q.next = NULL;
1164 tsk->worker_private = NULL;
1165
1166 kcov_task_init(tsk);
1167 kmsan_task_create(tsk);
1168 kmap_local_fork(tsk);
1169
1170 #ifdef CONFIG_FAULT_INJECTION
1171 tsk->fail_nth = 0;
1172 #endif
1173
1174 #ifdef CONFIG_BLK_CGROUP
1175 tsk->throttle_disk = NULL;
1176 tsk->use_memdelay = 0;
1177 #endif
1178
1179 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1180 tsk->pasid_activated = 0;
1181 #endif
1182
1183 #ifdef CONFIG_MEMCG
1184 tsk->active_memcg = NULL;
1185 #endif
1186
1187 #ifdef CONFIG_CPU_SUP_INTEL
1188 tsk->reported_split_lock = 0;
1189 #endif
1190
1191 #ifdef CONFIG_SCHED_MM_CID
1192 tsk->mm_cid = -1;
1193 tsk->last_mm_cid = -1;
1194 tsk->mm_cid_active = 0;
1195 tsk->migrate_from_cpu = -1;
1196 #endif
1197 return tsk;
1198
1199 free_stack:
1200 exit_task_stack_account(tsk);
1201 free_thread_stack(tsk);
1202 free_tsk:
1203 free_task_struct(tsk);
1204 return NULL;
1205 }
1206
1207 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1208
1209 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1210
1211 static int __init coredump_filter_setup(char *s)
1212 {
1213 default_dump_filter =
1214 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1215 MMF_DUMP_FILTER_MASK;
1216 return 1;
1217 }
1218
1219 __setup("coredump_filter=", coredump_filter_setup);
1220
1221 #include <linux/init_task.h>
1222
1223 static void mm_init_aio(struct mm_struct *mm)
1224 {
1225 #ifdef CONFIG_AIO
1226 spin_lock_init(&mm->ioctx_lock);
1227 mm->ioctx_table = NULL;
1228 #endif
1229 }
1230
1231 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1232 struct task_struct *p)
1233 {
1234 #ifdef CONFIG_MEMCG
1235 if (mm->owner == p)
1236 WRITE_ONCE(mm->owner, NULL);
1237 #endif
1238 }
1239
1240 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1241 {
1242 #ifdef CONFIG_MEMCG
1243 mm->owner = p;
1244 #endif
1245 }
1246
1247 static void mm_init_uprobes_state(struct mm_struct *mm)
1248 {
1249 #ifdef CONFIG_UPROBES
1250 mm->uprobes_state.xol_area = NULL;
1251 #endif
1252 }
1253
1254 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1255 struct user_namespace *user_ns)
1256 {
1257 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1258 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1259 atomic_set(&mm->mm_users, 1);
1260 atomic_set(&mm->mm_count, 1);
1261 seqcount_init(&mm->write_protect_seq);
1262 mmap_init_lock(mm);
1263 INIT_LIST_HEAD(&mm->mmlist);
1264 #ifdef CONFIG_PER_VMA_LOCK
1265 mm->mm_lock_seq = 0;
1266 #endif
1267 mm_pgtables_bytes_init(mm);
1268 mm->map_count = 0;
1269 mm->locked_vm = 0;
1270 atomic64_set(&mm->pinned_vm, 0);
1271 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1272 spin_lock_init(&mm->page_table_lock);
1273 spin_lock_init(&mm->arg_lock);
1274 mm_init_cpumask(mm);
1275 mm_init_aio(mm);
1276 mm_init_owner(mm, p);
1277 mm_pasid_init(mm);
1278 RCU_INIT_POINTER(mm->exe_file, NULL);
1279 mmu_notifier_subscriptions_init(mm);
1280 init_tlb_flush_pending(mm);
1281 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1282 mm->pmd_huge_pte = NULL;
1283 #endif
1284 mm_init_uprobes_state(mm);
1285 hugetlb_count_init(mm);
1286
1287 if (current->mm) {
1288 mm->flags = mmf_init_flags(current->mm->flags);
1289 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1290 } else {
1291 mm->flags = default_dump_filter;
1292 mm->def_flags = 0;
1293 }
1294
1295 if (mm_alloc_pgd(mm))
1296 goto fail_nopgd;
1297
1298 if (init_new_context(p, mm))
1299 goto fail_nocontext;
1300
1301 if (mm_alloc_cid(mm))
1302 goto fail_cid;
1303
1304 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1305 NR_MM_COUNTERS))
1306 goto fail_pcpu;
1307
1308 mm->user_ns = get_user_ns(user_ns);
1309 lru_gen_init_mm(mm);
1310 return mm;
1311
1312 fail_pcpu:
1313 mm_destroy_cid(mm);
1314 fail_cid:
1315 destroy_context(mm);
1316 fail_nocontext:
1317 mm_free_pgd(mm);
1318 fail_nopgd:
1319 free_mm(mm);
1320 return NULL;
1321 }
1322
1323 /*
1324 * Allocate and initialize an mm_struct.
1325 */
1326 struct mm_struct *mm_alloc(void)
1327 {
1328 struct mm_struct *mm;
1329
1330 mm = allocate_mm();
1331 if (!mm)
1332 return NULL;
1333
1334 memset(mm, 0, sizeof(*mm));
1335 return mm_init(mm, current, current_user_ns());
1336 }
1337
1338 static inline void __mmput(struct mm_struct *mm)
1339 {
1340 VM_BUG_ON(atomic_read(&mm->mm_users));
1341
1342 uprobe_clear_state(mm);
1343 exit_aio(mm);
1344 ksm_exit(mm);
1345 khugepaged_exit(mm); /* must run before exit_mmap */
1346 exit_mmap(mm);
1347 mm_put_huge_zero_page(mm);
1348 set_mm_exe_file(mm, NULL);
1349 if (!list_empty(&mm->mmlist)) {
1350 spin_lock(&mmlist_lock);
1351 list_del(&mm->mmlist);
1352 spin_unlock(&mmlist_lock);
1353 }
1354 if (mm->binfmt)
1355 module_put(mm->binfmt->module);
1356 lru_gen_del_mm(mm);
1357 mmdrop(mm);
1358 }
1359
1360 /*
1361 * Decrement the use count and release all resources for an mm.
1362 */
1363 void mmput(struct mm_struct *mm)
1364 {
1365 might_sleep();
1366
1367 if (atomic_dec_and_test(&mm->mm_users))
1368 __mmput(mm);
1369 }
1370 EXPORT_SYMBOL_GPL(mmput);
1371
1372 #ifdef CONFIG_MMU
1373 static void mmput_async_fn(struct work_struct *work)
1374 {
1375 struct mm_struct *mm = container_of(work, struct mm_struct,
1376 async_put_work);
1377
1378 __mmput(mm);
1379 }
1380
1381 void mmput_async(struct mm_struct *mm)
1382 {
1383 if (atomic_dec_and_test(&mm->mm_users)) {
1384 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1385 schedule_work(&mm->async_put_work);
1386 }
1387 }
1388 EXPORT_SYMBOL_GPL(mmput_async);
1389 #endif
1390
1391 /**
1392 * set_mm_exe_file - change a reference to the mm's executable file
1393 * @mm: The mm to change.
1394 * @new_exe_file: The new file to use.
1395 *
1396 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1397 *
1398 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1399 * invocations: in mmput() nobody alive left, in execve it happens before
1400 * the new mm is made visible to anyone.
1401 *
1402 * Can only fail if new_exe_file != NULL.
1403 */
1404 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1405 {
1406 struct file *old_exe_file;
1407
1408 /*
1409 * It is safe to dereference the exe_file without RCU as
1410 * this function is only called if nobody else can access
1411 * this mm -- see comment above for justification.
1412 */
1413 old_exe_file = rcu_dereference_raw(mm->exe_file);
1414
1415 if (new_exe_file) {
1416 /*
1417 * We expect the caller (i.e., sys_execve) to already denied
1418 * write access, so this is unlikely to fail.
1419 */
1420 if (unlikely(deny_write_access(new_exe_file)))
1421 return -EACCES;
1422 get_file(new_exe_file);
1423 }
1424 rcu_assign_pointer(mm->exe_file, new_exe_file);
1425 if (old_exe_file) {
1426 allow_write_access(old_exe_file);
1427 fput(old_exe_file);
1428 }
1429 return 0;
1430 }
1431
1432 /**
1433 * replace_mm_exe_file - replace a reference to the mm's executable file
1434 * @mm: The mm to change.
1435 * @new_exe_file: The new file to use.
1436 *
1437 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1438 *
1439 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1440 */
1441 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1442 {
1443 struct vm_area_struct *vma;
1444 struct file *old_exe_file;
1445 int ret = 0;
1446
1447 /* Forbid mm->exe_file change if old file still mapped. */
1448 old_exe_file = get_mm_exe_file(mm);
1449 if (old_exe_file) {
1450 VMA_ITERATOR(vmi, mm, 0);
1451 mmap_read_lock(mm);
1452 for_each_vma(vmi, vma) {
1453 if (!vma->vm_file)
1454 continue;
1455 if (path_equal(&vma->vm_file->f_path,
1456 &old_exe_file->f_path)) {
1457 ret = -EBUSY;
1458 break;
1459 }
1460 }
1461 mmap_read_unlock(mm);
1462 fput(old_exe_file);
1463 if (ret)
1464 return ret;
1465 }
1466
1467 ret = deny_write_access(new_exe_file);
1468 if (ret)
1469 return -EACCES;
1470 get_file(new_exe_file);
1471
1472 /* set the new file */
1473 mmap_write_lock(mm);
1474 old_exe_file = rcu_dereference_raw(mm->exe_file);
1475 rcu_assign_pointer(mm->exe_file, new_exe_file);
1476 mmap_write_unlock(mm);
1477
1478 if (old_exe_file) {
1479 allow_write_access(old_exe_file);
1480 fput(old_exe_file);
1481 }
1482 return 0;
1483 }
1484
1485 /**
1486 * get_mm_exe_file - acquire a reference to the mm's executable file
1487 * @mm: The mm of interest.
1488 *
1489 * Returns %NULL if mm has no associated executable file.
1490 * User must release file via fput().
1491 */
1492 struct file *get_mm_exe_file(struct mm_struct *mm)
1493 {
1494 struct file *exe_file;
1495
1496 rcu_read_lock();
1497 exe_file = get_file_rcu(&mm->exe_file);
1498 rcu_read_unlock();
1499 return exe_file;
1500 }
1501
1502 /**
1503 * get_task_exe_file - acquire a reference to the task's executable file
1504 * @task: The task.
1505 *
1506 * Returns %NULL if task's mm (if any) has no associated executable file or
1507 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1508 * User must release file via fput().
1509 */
1510 struct file *get_task_exe_file(struct task_struct *task)
1511 {
1512 struct file *exe_file = NULL;
1513 struct mm_struct *mm;
1514
1515 task_lock(task);
1516 mm = task->mm;
1517 if (mm) {
1518 if (!(task->flags & PF_KTHREAD))
1519 exe_file = get_mm_exe_file(mm);
1520 }
1521 task_unlock(task);
1522 return exe_file;
1523 }
1524
1525 /**
1526 * get_task_mm - acquire a reference to the task's mm
1527 * @task: The task.
1528 *
1529 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1530 * this kernel workthread has transiently adopted a user mm with use_mm,
1531 * to do its AIO) is not set and if so returns a reference to it, after
1532 * bumping up the use count. User must release the mm via mmput()
1533 * after use. Typically used by /proc and ptrace.
1534 */
1535 struct mm_struct *get_task_mm(struct task_struct *task)
1536 {
1537 struct mm_struct *mm;
1538
1539 task_lock(task);
1540 mm = task->mm;
1541 if (mm) {
1542 if (task->flags & PF_KTHREAD)
1543 mm = NULL;
1544 else
1545 mmget(mm);
1546 }
1547 task_unlock(task);
1548 return mm;
1549 }
1550 EXPORT_SYMBOL_GPL(get_task_mm);
1551
1552 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1553 {
1554 struct mm_struct *mm;
1555 int err;
1556
1557 err = down_read_killable(&task->signal->exec_update_lock);
1558 if (err)
1559 return ERR_PTR(err);
1560
1561 mm = get_task_mm(task);
1562 if (mm && mm != current->mm &&
1563 !ptrace_may_access(task, mode)) {
1564 mmput(mm);
1565 mm = ERR_PTR(-EACCES);
1566 }
1567 up_read(&task->signal->exec_update_lock);
1568
1569 return mm;
1570 }
1571
1572 static void complete_vfork_done(struct task_struct *tsk)
1573 {
1574 struct completion *vfork;
1575
1576 task_lock(tsk);
1577 vfork = tsk->vfork_done;
1578 if (likely(vfork)) {
1579 tsk->vfork_done = NULL;
1580 complete(vfork);
1581 }
1582 task_unlock(tsk);
1583 }
1584
1585 static int wait_for_vfork_done(struct task_struct *child,
1586 struct completion *vfork)
1587 {
1588 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1589 int killed;
1590
1591 cgroup_enter_frozen();
1592 killed = wait_for_completion_state(vfork, state);
1593 cgroup_leave_frozen(false);
1594
1595 if (killed) {
1596 task_lock(child);
1597 child->vfork_done = NULL;
1598 task_unlock(child);
1599 }
1600
1601 put_task_struct(child);
1602 return killed;
1603 }
1604
1605 /* Please note the differences between mmput and mm_release.
1606 * mmput is called whenever we stop holding onto a mm_struct,
1607 * error success whatever.
1608 *
1609 * mm_release is called after a mm_struct has been removed
1610 * from the current process.
1611 *
1612 * This difference is important for error handling, when we
1613 * only half set up a mm_struct for a new process and need to restore
1614 * the old one. Because we mmput the new mm_struct before
1615 * restoring the old one. . .
1616 * Eric Biederman 10 January 1998
1617 */
1618 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1619 {
1620 uprobe_free_utask(tsk);
1621
1622 /* Get rid of any cached register state */
1623 deactivate_mm(tsk, mm);
1624
1625 /*
1626 * Signal userspace if we're not exiting with a core dump
1627 * because we want to leave the value intact for debugging
1628 * purposes.
1629 */
1630 if (tsk->clear_child_tid) {
1631 if (atomic_read(&mm->mm_users) > 1) {
1632 /*
1633 * We don't check the error code - if userspace has
1634 * not set up a proper pointer then tough luck.
1635 */
1636 put_user(0, tsk->clear_child_tid);
1637 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1638 1, NULL, NULL, 0, 0);
1639 }
1640 tsk->clear_child_tid = NULL;
1641 }
1642
1643 /*
1644 * All done, finally we can wake up parent and return this mm to him.
1645 * Also kthread_stop() uses this completion for synchronization.
1646 */
1647 if (tsk->vfork_done)
1648 complete_vfork_done(tsk);
1649 }
1650
1651 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1652 {
1653 futex_exit_release(tsk);
1654 mm_release(tsk, mm);
1655 }
1656
1657 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1658 {
1659 futex_exec_release(tsk);
1660 mm_release(tsk, mm);
1661 }
1662
1663 /**
1664 * dup_mm() - duplicates an existing mm structure
1665 * @tsk: the task_struct with which the new mm will be associated.
1666 * @oldmm: the mm to duplicate.
1667 *
1668 * Allocates a new mm structure and duplicates the provided @oldmm structure
1669 * content into it.
1670 *
1671 * Return: the duplicated mm or NULL on failure.
1672 */
1673 static struct mm_struct *dup_mm(struct task_struct *tsk,
1674 struct mm_struct *oldmm)
1675 {
1676 struct mm_struct *mm;
1677 int err;
1678
1679 mm = allocate_mm();
1680 if (!mm)
1681 goto fail_nomem;
1682
1683 memcpy(mm, oldmm, sizeof(*mm));
1684
1685 if (!mm_init(mm, tsk, mm->user_ns))
1686 goto fail_nomem;
1687
1688 err = dup_mmap(mm, oldmm);
1689 if (err)
1690 goto free_pt;
1691
1692 mm->hiwater_rss = get_mm_rss(mm);
1693 mm->hiwater_vm = mm->total_vm;
1694
1695 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1696 goto free_pt;
1697
1698 return mm;
1699
1700 free_pt:
1701 /* don't put binfmt in mmput, we haven't got module yet */
1702 mm->binfmt = NULL;
1703 mm_init_owner(mm, NULL);
1704 mmput(mm);
1705
1706 fail_nomem:
1707 return NULL;
1708 }
1709
1710 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1711 {
1712 struct mm_struct *mm, *oldmm;
1713
1714 tsk->min_flt = tsk->maj_flt = 0;
1715 tsk->nvcsw = tsk->nivcsw = 0;
1716 #ifdef CONFIG_DETECT_HUNG_TASK
1717 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1718 tsk->last_switch_time = 0;
1719 #endif
1720
1721 tsk->mm = NULL;
1722 tsk->active_mm = NULL;
1723
1724 /*
1725 * Are we cloning a kernel thread?
1726 *
1727 * We need to steal a active VM for that..
1728 */
1729 oldmm = current->mm;
1730 if (!oldmm)
1731 return 0;
1732
1733 if (clone_flags & CLONE_VM) {
1734 mmget(oldmm);
1735 mm = oldmm;
1736 } else {
1737 mm = dup_mm(tsk, current->mm);
1738 if (!mm)
1739 return -ENOMEM;
1740 }
1741
1742 tsk->mm = mm;
1743 tsk->active_mm = mm;
1744 sched_mm_cid_fork(tsk);
1745 return 0;
1746 }
1747
1748 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1749 {
1750 struct fs_struct *fs = current->fs;
1751 if (clone_flags & CLONE_FS) {
1752 /* tsk->fs is already what we want */
1753 spin_lock(&fs->lock);
1754 /* "users" and "in_exec" locked for check_unsafe_exec() */
1755 if (fs->in_exec) {
1756 spin_unlock(&fs->lock);
1757 return -EAGAIN;
1758 }
1759 fs->users++;
1760 spin_unlock(&fs->lock);
1761 return 0;
1762 }
1763 tsk->fs = copy_fs_struct(fs);
1764 if (!tsk->fs)
1765 return -ENOMEM;
1766 return 0;
1767 }
1768
1769 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1770 int no_files)
1771 {
1772 struct files_struct *oldf, *newf;
1773 int error = 0;
1774
1775 /*
1776 * A background process may not have any files ...
1777 */
1778 oldf = current->files;
1779 if (!oldf)
1780 goto out;
1781
1782 if (no_files) {
1783 tsk->files = NULL;
1784 goto out;
1785 }
1786
1787 if (clone_flags & CLONE_FILES) {
1788 atomic_inc(&oldf->count);
1789 goto out;
1790 }
1791
1792 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1793 if (!newf)
1794 goto out;
1795
1796 tsk->files = newf;
1797 error = 0;
1798 out:
1799 return error;
1800 }
1801
1802 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1803 {
1804 struct sighand_struct *sig;
1805
1806 if (clone_flags & CLONE_SIGHAND) {
1807 refcount_inc(¤t->sighand->count);
1808 return 0;
1809 }
1810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1811 RCU_INIT_POINTER(tsk->sighand, sig);
1812 if (!sig)
1813 return -ENOMEM;
1814
1815 refcount_set(&sig->count, 1);
1816 spin_lock_irq(¤t->sighand->siglock);
1817 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1818 spin_unlock_irq(¤t->sighand->siglock);
1819
1820 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1821 if (clone_flags & CLONE_CLEAR_SIGHAND)
1822 flush_signal_handlers(tsk, 0);
1823
1824 return 0;
1825 }
1826
1827 void __cleanup_sighand(struct sighand_struct *sighand)
1828 {
1829 if (refcount_dec_and_test(&sighand->count)) {
1830 signalfd_cleanup(sighand);
1831 /*
1832 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1833 * without an RCU grace period, see __lock_task_sighand().
1834 */
1835 kmem_cache_free(sighand_cachep, sighand);
1836 }
1837 }
1838
1839 /*
1840 * Initialize POSIX timer handling for a thread group.
1841 */
1842 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1843 {
1844 struct posix_cputimers *pct = &sig->posix_cputimers;
1845 unsigned long cpu_limit;
1846
1847 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1848 posix_cputimers_group_init(pct, cpu_limit);
1849 }
1850
1851 void kpatch_foo(void);
1852 void kpatch_foo(void)
1853 {
1854 if (!jiffies)
1855 printk("kpatch copy signal\n");
1856 }
1857
1858 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1859 {
1860 struct signal_struct *sig;
1861
1862 kpatch_foo();
1863
1864 if (clone_flags & CLONE_THREAD)
1865 return 0;
1866
1867 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1868 tsk->signal = sig;
1869 if (!sig)
1870 return -ENOMEM;
1871
1872 sig->nr_threads = 1;
1873 sig->quick_threads = 1;
1874 atomic_set(&sig->live, 1);
1875 refcount_set(&sig->sigcnt, 1);
1876
1877 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1878 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1879 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1880
1881 init_waitqueue_head(&sig->wait_chldexit);
1882 sig->curr_target = tsk;
1883 init_sigpending(&sig->shared_pending);
1884 INIT_HLIST_HEAD(&sig->multiprocess);
1885 seqlock_init(&sig->stats_lock);
1886 prev_cputime_init(&sig->prev_cputime);
1887
1888 #ifdef CONFIG_POSIX_TIMERS
1889 INIT_LIST_HEAD(&sig->posix_timers);
1890 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1891 sig->real_timer.function = it_real_fn;
1892 #endif
1893
1894 task_lock(current->group_leader);
1895 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1896 task_unlock(current->group_leader);
1897
1898 posix_cpu_timers_init_group(sig);
1899
1900 tty_audit_fork(sig);
1901 sched_autogroup_fork(sig);
1902
1903 sig->oom_score_adj = current->signal->oom_score_adj;
1904 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1905
1906 mutex_init(&sig->cred_guard_mutex);
1907 init_rwsem(&sig->exec_update_lock);
1908
1909 return 0;
1910 }
1911
1912 static void copy_seccomp(struct task_struct *p)
1913 {
1914 #ifdef CONFIG_SECCOMP
1915 /*
1916 * Must be called with sighand->lock held, which is common to
1917 * all threads in the group. Holding cred_guard_mutex is not
1918 * needed because this new task is not yet running and cannot
1919 * be racing exec.
1920 */
1921 assert_spin_locked(¤t->sighand->siglock);
1922
1923 /* Ref-count the new filter user, and assign it. */
1924 get_seccomp_filter(current);
1925 p->seccomp = current->seccomp;
1926
1927 /*
1928 * Explicitly enable no_new_privs here in case it got set
1929 * between the task_struct being duplicated and holding the
1930 * sighand lock. The seccomp state and nnp must be in sync.
1931 */
1932 if (task_no_new_privs(current))
1933 task_set_no_new_privs(p);
1934
1935 /*
1936 * If the parent gained a seccomp mode after copying thread
1937 * flags and between before we held the sighand lock, we have
1938 * to manually enable the seccomp thread flag here.
1939 */
1940 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1941 set_task_syscall_work(p, SECCOMP);
1942 #endif
1943 }
1944
1945 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1946 {
1947 current->clear_child_tid = tidptr;
1948
1949 return task_pid_vnr(current);
1950 }
1951
1952 static void rt_mutex_init_task(struct task_struct *p)
1953 {
1954 raw_spin_lock_init(&p->pi_lock);
1955 #ifdef CONFIG_RT_MUTEXES
1956 p->pi_waiters = RB_ROOT_CACHED;
1957 p->pi_top_task = NULL;
1958 p->pi_blocked_on = NULL;
1959 #endif
1960 }
1961
1962 static inline void init_task_pid_links(struct task_struct *task)
1963 {
1964 enum pid_type type;
1965
1966 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1967 INIT_HLIST_NODE(&task->pid_links[type]);
1968 }
1969
1970 static inline void
1971 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1972 {
1973 if (type == PIDTYPE_PID)
1974 task->thread_pid = pid;
1975 else
1976 task->signal->pids[type] = pid;
1977 }
1978
1979 static inline void rcu_copy_process(struct task_struct *p)
1980 {
1981 #ifdef CONFIG_PREEMPT_RCU
1982 p->rcu_read_lock_nesting = 0;
1983 p->rcu_read_unlock_special.s = 0;
1984 p->rcu_blocked_node = NULL;
1985 INIT_LIST_HEAD(&p->rcu_node_entry);
1986 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1987 #ifdef CONFIG_TASKS_RCU
1988 p->rcu_tasks_holdout = false;
1989 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1990 p->rcu_tasks_idle_cpu = -1;
1991 INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1992 #endif /* #ifdef CONFIG_TASKS_RCU */
1993 #ifdef CONFIG_TASKS_TRACE_RCU
1994 p->trc_reader_nesting = 0;
1995 p->trc_reader_special.s = 0;
1996 INIT_LIST_HEAD(&p->trc_holdout_list);
1997 INIT_LIST_HEAD(&p->trc_blkd_node);
1998 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1999 }
2000
2001 /**
2002 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2003 * @pid: the struct pid for which to create a pidfd
2004 * @flags: flags of the new @pidfd
2005 * @ret: Where to return the file for the pidfd.
2006 *
2007 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2008 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2009 *
2010 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2011 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2012 * pidfd file are prepared.
2013 *
2014 * If this function returns successfully the caller is responsible to either
2015 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2016 * order to install the pidfd into its file descriptor table or they must use
2017 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2018 * respectively.
2019 *
2020 * This function is useful when a pidfd must already be reserved but there
2021 * might still be points of failure afterwards and the caller wants to ensure
2022 * that no pidfd is leaked into its file descriptor table.
2023 *
2024 * Return: On success, a reserved pidfd is returned from the function and a new
2025 * pidfd file is returned in the last argument to the function. On
2026 * error, a negative error code is returned from the function and the
2027 * last argument remains unchanged.
2028 */
2029 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2030 {
2031 int pidfd;
2032 struct file *pidfd_file;
2033
2034 pidfd = get_unused_fd_flags(O_CLOEXEC);
2035 if (pidfd < 0)
2036 return pidfd;
2037
2038 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2039 if (IS_ERR(pidfd_file)) {
2040 put_unused_fd(pidfd);
2041 return PTR_ERR(pidfd_file);
2042 }
2043 /*
2044 * anon_inode_getfile() ignores everything outside of the
2045 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2046 */
2047 pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2048 *ret = pidfd_file;
2049 return pidfd;
2050 }
2051
2052 /**
2053 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2054 * @pid: the struct pid for which to create a pidfd
2055 * @flags: flags of the new @pidfd
2056 * @ret: Where to return the pidfd.
2057 *
2058 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2059 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2060 *
2061 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2062 * task identified by @pid must be a thread-group leader.
2063 *
2064 * If this function returns successfully the caller is responsible to either
2065 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2066 * order to install the pidfd into its file descriptor table or they must use
2067 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2068 * respectively.
2069 *
2070 * This function is useful when a pidfd must already be reserved but there
2071 * might still be points of failure afterwards and the caller wants to ensure
2072 * that no pidfd is leaked into its file descriptor table.
2073 *
2074 * Return: On success, a reserved pidfd is returned from the function and a new
2075 * pidfd file is returned in the last argument to the function. On
2076 * error, a negative error code is returned from the function and the
2077 * last argument remains unchanged.
2078 */
2079 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2080 {
2081 bool thread = flags & PIDFD_THREAD;
2082
2083 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2084 return -EINVAL;
2085
2086 return __pidfd_prepare(pid, flags, ret);
2087 }
2088
2089 static void __delayed_free_task(struct rcu_head *rhp)
2090 {
2091 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2092
2093 free_task(tsk);
2094 }
2095
2096 static __always_inline void delayed_free_task(struct task_struct *tsk)
2097 {
2098 if (IS_ENABLED(CONFIG_MEMCG))
2099 call_rcu(&tsk->rcu, __delayed_free_task);
2100 else
2101 free_task(tsk);
2102 }
2103
2104 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2105 {
2106 /* Skip if kernel thread */
2107 if (!tsk->mm)
2108 return;
2109
2110 /* Skip if spawning a thread or using vfork */
2111 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2112 return;
2113
2114 /* We need to synchronize with __set_oom_adj */
2115 mutex_lock(&oom_adj_mutex);
2116 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2117 /* Update the values in case they were changed after copy_signal */
2118 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2119 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2120 mutex_unlock(&oom_adj_mutex);
2121 }
2122
2123 #ifdef CONFIG_RV
2124 static void rv_task_fork(struct task_struct *p)
2125 {
2126 int i;
2127
2128 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2129 p->rv[i].da_mon.monitoring = false;
2130 }
2131 #else
2132 #define rv_task_fork(p) do {} while (0)
2133 #endif
2134
2135 /*
2136 * This creates a new process as a copy of the old one,
2137 * but does not actually start it yet.
2138 *
2139 * It copies the registers, and all the appropriate
2140 * parts of the process environment (as per the clone
2141 * flags). The actual kick-off is left to the caller.
2142 */
2143 __latent_entropy struct task_struct *copy_process(
2144 struct pid *pid,
2145 int trace,
2146 int node,
2147 struct kernel_clone_args *args)
2148 {
2149 int pidfd = -1, retval;
2150 struct task_struct *p;
2151 struct multiprocess_signals delayed;
2152 struct file *pidfile = NULL;
2153 const u64 clone_flags = args->flags;
2154 struct nsproxy *nsp = current->nsproxy;
2155
2156 /*
2157 * Don't allow sharing the root directory with processes in a different
2158 * namespace
2159 */
2160 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2161 return ERR_PTR(-EINVAL);
2162
2163 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2164 return ERR_PTR(-EINVAL);
2165
2166 /*
2167 * Thread groups must share signals as well, and detached threads
2168 * can only be started up within the thread group.
2169 */
2170 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2171 return ERR_PTR(-EINVAL);
2172
2173 /*
2174 * Shared signal handlers imply shared VM. By way of the above,
2175 * thread groups also imply shared VM. Blocking this case allows
2176 * for various simplifications in other code.
2177 */
2178 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2179 return ERR_PTR(-EINVAL);
2180
2181 /*
2182 * Siblings of global init remain as zombies on exit since they are
2183 * not reaped by their parent (swapper). To solve this and to avoid
2184 * multi-rooted process trees, prevent global and container-inits
2185 * from creating siblings.
2186 */
2187 if ((clone_flags & CLONE_PARENT) &&
2188 current->signal->flags & SIGNAL_UNKILLABLE)
2189 return ERR_PTR(-EINVAL);
2190
2191 /*
2192 * If the new process will be in a different pid or user namespace
2193 * do not allow it to share a thread group with the forking task.
2194 */
2195 if (clone_flags & CLONE_THREAD) {
2196 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2197 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2198 return ERR_PTR(-EINVAL);
2199 }
2200
2201 if (clone_flags & CLONE_PIDFD) {
2202 /*
2203 * - CLONE_DETACHED is blocked so that we can potentially
2204 * reuse it later for CLONE_PIDFD.
2205 */
2206 if (clone_flags & CLONE_DETACHED)
2207 return ERR_PTR(-EINVAL);
2208 }
2209
2210 /*
2211 * Force any signals received before this point to be delivered
2212 * before the fork happens. Collect up signals sent to multiple
2213 * processes that happen during the fork and delay them so that
2214 * they appear to happen after the fork.
2215 */
2216 sigemptyset(&delayed.signal);
2217 INIT_HLIST_NODE(&delayed.node);
2218
2219 spin_lock_irq(¤t->sighand->siglock);
2220 if (!(clone_flags & CLONE_THREAD))
2221 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2222 recalc_sigpending();
2223 spin_unlock_irq(¤t->sighand->siglock);
2224 retval = -ERESTARTNOINTR;
2225 if (task_sigpending(current))
2226 goto fork_out;
2227
2228 retval = -ENOMEM;
2229 p = dup_task_struct(current, node);
2230 if (!p)
2231 goto fork_out;
2232 p->flags &= ~PF_KTHREAD;
2233 if (args->kthread)
2234 p->flags |= PF_KTHREAD;
2235 if (args->user_worker) {
2236 /*
2237 * Mark us a user worker, and block any signal that isn't
2238 * fatal or STOP
2239 */
2240 p->flags |= PF_USER_WORKER;
2241 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2242 }
2243 if (args->io_thread)
2244 p->flags |= PF_IO_WORKER;
2245
2246 if (args->name)
2247 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2248
2249 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2250 /*
2251 * Clear TID on mm_release()?
2252 */
2253 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2254
2255 ftrace_graph_init_task(p);
2256
2257 rt_mutex_init_task(p);
2258
2259 lockdep_assert_irqs_enabled();
2260 #ifdef CONFIG_PROVE_LOCKING
2261 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2262 #endif
2263 retval = copy_creds(p, clone_flags);
2264 if (retval < 0)
2265 goto bad_fork_free;
2266
2267 retval = -EAGAIN;
2268 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2269 if (p->real_cred->user != INIT_USER &&
2270 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2271 goto bad_fork_cleanup_count;
2272 }
2273 current->flags &= ~PF_NPROC_EXCEEDED;
2274
2275 /*
2276 * If multiple threads are within copy_process(), then this check
2277 * triggers too late. This doesn't hurt, the check is only there
2278 * to stop root fork bombs.
2279 */
2280 retval = -EAGAIN;
2281 if (data_race(nr_threads >= max_threads))
2282 goto bad_fork_cleanup_count;
2283
2284 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2285 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2286 p->flags |= PF_FORKNOEXEC;
2287 INIT_LIST_HEAD(&p->children);
2288 INIT_LIST_HEAD(&p->sibling);
2289 rcu_copy_process(p);
2290 p->vfork_done = NULL;
2291 spin_lock_init(&p->alloc_lock);
2292
2293 init_sigpending(&p->pending);
2294
2295 p->utime = p->stime = p->gtime = 0;
2296 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2297 p->utimescaled = p->stimescaled = 0;
2298 #endif
2299 prev_cputime_init(&p->prev_cputime);
2300
2301 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2302 seqcount_init(&p->vtime.seqcount);
2303 p->vtime.starttime = 0;
2304 p->vtime.state = VTIME_INACTIVE;
2305 #endif
2306
2307 #ifdef CONFIG_IO_URING
2308 p->io_uring = NULL;
2309 #endif
2310
2311 p->default_timer_slack_ns = current->timer_slack_ns;
2312
2313 #ifdef CONFIG_PSI
2314 p->psi_flags = 0;
2315 #endif
2316
2317 task_io_accounting_init(&p->ioac);
2318 acct_clear_integrals(p);
2319
2320 posix_cputimers_init(&p->posix_cputimers);
2321
2322 p->io_context = NULL;
2323 audit_set_context(p, NULL);
2324 cgroup_fork(p);
2325 if (args->kthread) {
2326 if (!set_kthread_struct(p))
2327 goto bad_fork_cleanup_delayacct;
2328 }
2329 #ifdef CONFIG_NUMA
2330 p->mempolicy = mpol_dup(p->mempolicy);
2331 if (IS_ERR(p->mempolicy)) {
2332 retval = PTR_ERR(p->mempolicy);
2333 p->mempolicy = NULL;
2334 goto bad_fork_cleanup_delayacct;
2335 }
2336 #endif
2337 #ifdef CONFIG_CPUSETS
2338 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2339 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2340 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2341 #endif
2342 #ifdef CONFIG_TRACE_IRQFLAGS
2343 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2344 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2345 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2346 p->softirqs_enabled = 1;
2347 p->softirq_context = 0;
2348 #endif
2349
2350 p->pagefault_disabled = 0;
2351
2352 #ifdef CONFIG_LOCKDEP
2353 lockdep_init_task(p);
2354 #endif
2355
2356 #ifdef CONFIG_DEBUG_MUTEXES
2357 p->blocked_on = NULL; /* not blocked yet */
2358 #endif
2359 #ifdef CONFIG_BCACHE
2360 p->sequential_io = 0;
2361 p->sequential_io_avg = 0;
2362 #endif
2363 #ifdef CONFIG_BPF_SYSCALL
2364 RCU_INIT_POINTER(p->bpf_storage, NULL);
2365 p->bpf_ctx = NULL;
2366 #endif
2367
2368 /* Perform scheduler related setup. Assign this task to a CPU. */
2369 retval = sched_fork(clone_flags, p);
2370 if (retval)
2371 goto bad_fork_cleanup_policy;
2372
2373 retval = perf_event_init_task(p, clone_flags);
2374 if (retval)
2375 goto bad_fork_cleanup_policy;
2376 retval = audit_alloc(p);
2377 if (retval)
2378 goto bad_fork_cleanup_perf;
2379 /* copy all the process information */
2380 shm_init_task(p);
2381 retval = security_task_alloc(p, clone_flags);
2382 if (retval)
2383 goto bad_fork_cleanup_audit;
2384 retval = copy_semundo(clone_flags, p);
2385 if (retval)
2386 goto bad_fork_cleanup_security;
2387 retval = copy_files(clone_flags, p, args->no_files);
2388 if (retval)
2389 goto bad_fork_cleanup_semundo;
2390 retval = copy_fs(clone_flags, p);
2391 if (retval)
2392 goto bad_fork_cleanup_files;
2393 retval = copy_sighand(clone_flags, p);
2394 if (retval)
2395 goto bad_fork_cleanup_fs;
2396 retval = copy_signal(clone_flags, p);
2397 if (retval)
2398 goto bad_fork_cleanup_sighand;
2399 retval = copy_mm(clone_flags, p);
2400 if (retval)
2401 goto bad_fork_cleanup_signal;
2402 retval = copy_namespaces(clone_flags, p);
2403 if (retval)
2404 goto bad_fork_cleanup_mm;
2405 retval = copy_io(clone_flags, p);
2406 if (retval)
2407 goto bad_fork_cleanup_namespaces;
2408 retval = copy_thread(p, args);
2409 if (retval)
2410 goto bad_fork_cleanup_io;
2411
2412 stackleak_task_init(p);
2413
2414 if (pid != &init_struct_pid) {
2415 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2416 args->set_tid_size);
2417 if (IS_ERR(pid)) {
2418 retval = PTR_ERR(pid);
2419 goto bad_fork_cleanup_thread;
2420 }
2421 }
2422
2423 /*
2424 * This has to happen after we've potentially unshared the file
2425 * descriptor table (so that the pidfd doesn't leak into the child
2426 * if the fd table isn't shared).
2427 */
2428 if (clone_flags & CLONE_PIDFD) {
2429 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2430
2431 /* Note that no task has been attached to @pid yet. */
2432 retval = __pidfd_prepare(pid, flags, &pidfile);
2433 if (retval < 0)
2434 goto bad_fork_free_pid;
2435 pidfd = retval;
2436
2437 retval = put_user(pidfd, args->pidfd);
2438 if (retval)
2439 goto bad_fork_put_pidfd;
2440 }
2441
2442 #ifdef CONFIG_BLOCK
2443 p->plug = NULL;
2444 #endif
2445 futex_init_task(p);
2446
2447 /*
2448 * sigaltstack should be cleared when sharing the same VM
2449 */
2450 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2451 sas_ss_reset(p);
2452
2453 /*
2454 * Syscall tracing and stepping should be turned off in the
2455 * child regardless of CLONE_PTRACE.
2456 */
2457 user_disable_single_step(p);
2458 clear_task_syscall_work(p, SYSCALL_TRACE);
2459 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2460 clear_task_syscall_work(p, SYSCALL_EMU);
2461 #endif
2462 clear_tsk_latency_tracing(p);
2463
2464 /* ok, now we should be set up.. */
2465 p->pid = pid_nr(pid);
2466 if (clone_flags & CLONE_THREAD) {
2467 p->group_leader = current->group_leader;
2468 p->tgid = current->tgid;
2469 } else {
2470 p->group_leader = p;
2471 p->tgid = p->pid;
2472 }
2473
2474 p->nr_dirtied = 0;
2475 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2476 p->dirty_paused_when = 0;
2477
2478 p->pdeath_signal = 0;
2479 p->task_works = NULL;
2480 clear_posix_cputimers_work(p);
2481
2482 #ifdef CONFIG_KRETPROBES
2483 p->kretprobe_instances.first = NULL;
2484 #endif
2485 #ifdef CONFIG_RETHOOK
2486 p->rethooks.first = NULL;
2487 #endif
2488
2489 /*
2490 * Ensure that the cgroup subsystem policies allow the new process to be
2491 * forked. It should be noted that the new process's css_set can be changed
2492 * between here and cgroup_post_fork() if an organisation operation is in
2493 * progress.
2494 */
2495 retval = cgroup_can_fork(p, args);
2496 if (retval)
2497 goto bad_fork_put_pidfd;
2498
2499 /*
2500 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2501 * the new task on the correct runqueue. All this *before* the task
2502 * becomes visible.
2503 *
2504 * This isn't part of ->can_fork() because while the re-cloning is
2505 * cgroup specific, it unconditionally needs to place the task on a
2506 * runqueue.
2507 */
2508 sched_cgroup_fork(p, args);
2509
2510 /*
2511 * From this point on we must avoid any synchronous user-space
2512 * communication until we take the tasklist-lock. In particular, we do
2513 * not want user-space to be able to predict the process start-time by
2514 * stalling fork(2) after we recorded the start_time but before it is
2515 * visible to the system.
2516 */
2517
2518 p->start_time = ktime_get_ns();
2519 p->start_boottime = ktime_get_boottime_ns();
2520
2521 /*
2522 * Make it visible to the rest of the system, but dont wake it up yet.
2523 * Need tasklist lock for parent etc handling!
2524 */
2525 write_lock_irq(&tasklist_lock);
2526
2527 /* CLONE_PARENT re-uses the old parent */
2528 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2529 p->real_parent = current->real_parent;
2530 p->parent_exec_id = current->parent_exec_id;
2531 if (clone_flags & CLONE_THREAD)
2532 p->exit_signal = -1;
2533 else
2534 p->exit_signal = current->group_leader->exit_signal;
2535 } else {
2536 p->real_parent = current;
2537 p->parent_exec_id = current->self_exec_id;
2538 p->exit_signal = args->exit_signal;
2539 }
2540
2541 klp_copy_process(p);
2542
2543 sched_core_fork(p);
2544
2545 spin_lock(¤t->sighand->siglock);
2546
2547 rv_task_fork(p);
2548
2549 rseq_fork(p, clone_flags);
2550
2551 /* Don't start children in a dying pid namespace */
2552 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2553 retval = -ENOMEM;
2554 goto bad_fork_cancel_cgroup;
2555 }
2556
2557 /* Let kill terminate clone/fork in the middle */
2558 if (fatal_signal_pending(current)) {
2559 retval = -EINTR;
2560 goto bad_fork_cancel_cgroup;
2561 }
2562
2563 /* No more failure paths after this point. */
2564
2565 /*
2566 * Copy seccomp details explicitly here, in case they were changed
2567 * before holding sighand lock.
2568 */
2569 copy_seccomp(p);
2570
2571 init_task_pid_links(p);
2572 if (likely(p->pid)) {
2573 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2574
2575 init_task_pid(p, PIDTYPE_PID, pid);
2576 if (thread_group_leader(p)) {
2577 init_task_pid(p, PIDTYPE_TGID, pid);
2578 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2579 init_task_pid(p, PIDTYPE_SID, task_session(current));
2580
2581 if (is_child_reaper(pid)) {
2582 ns_of_pid(pid)->child_reaper = p;
2583 p->signal->flags |= SIGNAL_UNKILLABLE;
2584 }
2585 p->signal->shared_pending.signal = delayed.signal;
2586 p->signal->tty = tty_kref_get(current->signal->tty);
2587 /*
2588 * Inherit has_child_subreaper flag under the same
2589 * tasklist_lock with adding child to the process tree
2590 * for propagate_has_child_subreaper optimization.
2591 */
2592 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2593 p->real_parent->signal->is_child_subreaper;
2594 list_add_tail(&p->sibling, &p->real_parent->children);
2595 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2596 attach_pid(p, PIDTYPE_TGID);
2597 attach_pid(p, PIDTYPE_PGID);
2598 attach_pid(p, PIDTYPE_SID);
2599 __this_cpu_inc(process_counts);
2600 } else {
2601 current->signal->nr_threads++;
2602 current->signal->quick_threads++;
2603 atomic_inc(¤t->signal->live);
2604 refcount_inc(¤t->signal->sigcnt);
2605 task_join_group_stop(p);
2606 list_add_tail_rcu(&p->thread_node,
2607 &p->signal->thread_head);
2608 }
2609 attach_pid(p, PIDTYPE_PID);
2610 nr_threads++;
2611 }
2612 total_forks++;
2613 hlist_del_init(&delayed.node);
2614 spin_unlock(¤t->sighand->siglock);
2615 syscall_tracepoint_update(p);
2616 write_unlock_irq(&tasklist_lock);
2617
2618 if (pidfile)
2619 fd_install(pidfd, pidfile);
2620
2621 proc_fork_connector(p);
2622 sched_post_fork(p);
2623 cgroup_post_fork(p, args);
2624 perf_event_fork(p);
2625
2626 trace_task_newtask(p, clone_flags);
2627 uprobe_copy_process(p, clone_flags);
2628 user_events_fork(p, clone_flags);
2629
2630 copy_oom_score_adj(clone_flags, p);
2631
2632 return p;
2633
2634 bad_fork_cancel_cgroup:
2635 sched_core_free(p);
2636 spin_unlock(¤t->sighand->siglock);
2637 write_unlock_irq(&tasklist_lock);
2638 cgroup_cancel_fork(p, args);
2639 bad_fork_put_pidfd:
2640 if (clone_flags & CLONE_PIDFD) {
2641 fput(pidfile);
2642 put_unused_fd(pidfd);
2643 }
2644 bad_fork_free_pid:
2645 if (pid != &init_struct_pid)
2646 free_pid(pid);
2647 bad_fork_cleanup_thread:
2648 exit_thread(p);
2649 bad_fork_cleanup_io:
2650 if (p->io_context)
2651 exit_io_context(p);
2652 bad_fork_cleanup_namespaces:
2653 exit_task_namespaces(p);
2654 bad_fork_cleanup_mm:
2655 if (p->mm) {
2656 mm_clear_owner(p->mm, p);
2657 mmput(p->mm);
2658 }
2659 bad_fork_cleanup_signal:
2660 if (!(clone_flags & CLONE_THREAD))
2661 free_signal_struct(p->signal);
2662 bad_fork_cleanup_sighand:
2663 __cleanup_sighand(p->sighand);
2664 bad_fork_cleanup_fs:
2665 exit_fs(p); /* blocking */
2666 bad_fork_cleanup_files:
2667 exit_files(p); /* blocking */
2668 bad_fork_cleanup_semundo:
2669 exit_sem(p);
2670 bad_fork_cleanup_security:
2671 security_task_free(p);
2672 bad_fork_cleanup_audit:
2673 audit_free(p);
2674 bad_fork_cleanup_perf:
2675 perf_event_free_task(p);
2676 bad_fork_cleanup_policy:
2677 lockdep_free_task(p);
2678 #ifdef CONFIG_NUMA
2679 mpol_put(p->mempolicy);
2680 #endif
2681 bad_fork_cleanup_delayacct:
2682 delayacct_tsk_free(p);
2683 bad_fork_cleanup_count:
2684 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2685 exit_creds(p);
2686 bad_fork_free:
2687 WRITE_ONCE(p->__state, TASK_DEAD);
2688 exit_task_stack_account(p);
2689 put_task_stack(p);
2690 delayed_free_task(p);
2691 fork_out:
2692 spin_lock_irq(¤t->sighand->siglock);
2693 hlist_del_init(&delayed.node);
2694 spin_unlock_irq(¤t->sighand->siglock);
2695 return ERR_PTR(retval);
2696 }
2697
2698 static inline void init_idle_pids(struct task_struct *idle)
2699 {
2700 enum pid_type type;
2701
2702 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2703 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2704 init_task_pid(idle, type, &init_struct_pid);
2705 }
2706 }
2707
2708 static int idle_dummy(void *dummy)
2709 {
2710 /* This function is never called */
2711 return 0;
2712 }
2713
2714 struct task_struct * __init fork_idle(int cpu)
2715 {
2716 struct task_struct *task;
2717 struct kernel_clone_args args = {
2718 .flags = CLONE_VM,
2719 .fn = &idle_dummy,
2720 .fn_arg = NULL,
2721 .kthread = 1,
2722 .idle = 1,
2723 };
2724
2725 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2726 if (!IS_ERR(task)) {
2727 init_idle_pids(task);
2728 init_idle(task, cpu);
2729 }
2730
2731 return task;
2732 }
2733
2734 /*
2735 * This is like kernel_clone(), but shaved down and tailored to just
2736 * creating io_uring workers. It returns a created task, or an error pointer.
2737 * The returned task is inactive, and the caller must fire it up through
2738 * wake_up_new_task(p). All signals are blocked in the created task.
2739 */
2740 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2741 {
2742 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2743 CLONE_IO;
2744 struct kernel_clone_args args = {
2745 .flags = ((lower_32_bits(flags) | CLONE_VM |
2746 CLONE_UNTRACED) & ~CSIGNAL),
2747 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2748 .fn = fn,
2749 .fn_arg = arg,
2750 .io_thread = 1,
2751 .user_worker = 1,
2752 };
2753
2754 return copy_process(NULL, 0, node, &args);
2755 }
2756
2757 /*
2758 * Ok, this is the main fork-routine.
2759 *
2760 * It copies the process, and if successful kick-starts
2761 * it and waits for it to finish using the VM if required.
2762 *
2763 * args->exit_signal is expected to be checked for sanity by the caller.
2764 */
> 2765 #include <linux/livepatch.h>
2766 pid_t kernel_clone(struct kernel_clone_args *args)
2767 {
2768 u64 clone_flags = args->flags;
2769 struct completion vfork;
2770 struct pid *pid;
2771 struct task_struct *p;
2772 int trace = 0;
2773 pid_t nr;
2774 int *newpid;
2775 static int ctr = 0;
2776
2777 /*
2778 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2779 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2780 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2781 * field in struct clone_args and it still doesn't make sense to have
2782 * them both point at the same memory location. Performing this check
2783 * here has the advantage that we don't need to have a separate helper
2784 * to check for legacy clone().
2785 */
2786 if ((clone_flags & CLONE_PIDFD) &&
2787 (clone_flags & CLONE_PARENT_SETTID) &&
2788 (args->pidfd == args->parent_tid))
2789 return -EINVAL;
2790
2791 /*
2792 * Determine whether and which event to report to ptracer. When
2793 * called from kernel_thread or CLONE_UNTRACED is explicitly
2794 * requested, no event is reported; otherwise, report if the event
2795 * for the type of forking is enabled.
2796 */
2797 if (!(clone_flags & CLONE_UNTRACED)) {
2798 if (clone_flags & CLONE_VFORK)
2799 trace = PTRACE_EVENT_VFORK;
2800 else if (args->exit_signal != SIGCHLD)
2801 trace = PTRACE_EVENT_CLONE;
2802 else
2803 trace = PTRACE_EVENT_FORK;
2804
2805 if (likely(!ptrace_event_enabled(current, trace)))
2806 trace = 0;
2807 }
2808
2809 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2810 add_latent_entropy();
2811
2812 if (IS_ERR(p))
2813 return PTR_ERR(p);
2814
2815 newpid = klp_shadow_get_or_alloc(p, 0, sizeof(*newpid), GFP_KERNEL,
2816 NULL, NULL);
2817 if (newpid)
2818 *newpid = ctr++;
2819
2820 /*
2821 * Do this prior waking up the new thread - the thread pointer
2822 * might get invalid after that point, if the thread exits quickly.
2823 */
2824 trace_sched_process_fork(current, p);
2825
2826 pid = get_task_pid(p, PIDTYPE_PID);
2827 nr = pid_vnr(pid);
2828
2829 if (clone_flags & CLONE_PARENT_SETTID)
2830 put_user(nr, args->parent_tid);
2831
2832 if (clone_flags & CLONE_VFORK) {
2833 p->vfork_done = &vfork;
2834 init_completion(&vfork);
2835 get_task_struct(p);
2836 }
2837
2838 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2839 /* lock the task to synchronize with memcg migration */
2840 task_lock(p);
2841 lru_gen_add_mm(p->mm);
2842 task_unlock(p);
2843 }
2844
2845 wake_up_new_task(p);
2846
2847 /* forking complete and child started to run, tell ptracer */
2848 if (unlikely(trace))
2849 ptrace_event_pid(trace, pid);
2850
2851 if (clone_flags & CLONE_VFORK) {
2852 if (!wait_for_vfork_done(p, &vfork))
2853 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2854 }
2855
2856 put_pid(pid);
2857 return nr;
2858 }
2859
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
^ permalink raw reply [flat|nested] 2+ messages in thread* RE: [jpoimboe:objtool-diff 2/2] kernel/fork.c: linux/livepatch.h is included more than once.
2024-05-25 6:40 [jpoimboe:objtool-diff 2/2] kernel/fork.c: linux/livepatch.h is included more than once kernel test robot
@ 2024-05-27 1:47 ` Liu, Yujie
0 siblings, 0 replies; 2+ messages in thread
From: Liu, Yujie @ 2024-05-27 1:47 UTC (permalink / raw)
To: lkp, oe-kbuild@lists.linux.dev
not_report
informal patch.
commit 745009dc796e56fc87e911138d801679ecd3576e (HEAD, jpoimboe/objtool-diff)
Author: Josh Poimboeuf <jpoimboe@kernel.org>
Date: Fri May 24 17:21:01 2024 -0700
test
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
-----Original Message-----
From: lkp <lkp@intel.com>
Sent: Saturday, May 25, 2024 14:40
To: oe-kbuild@lists.linux.dev
Cc: lkp <lkp@intel.com>
Subject: [jpoimboe:objtool-diff 2/2] kernel/fork.c: linux/livepatch.h is included more than once.
::::::
:::::: Manual check reason: "low confidence bisect report"
::::::
BCC: lkp@intel.com
CC: oe-kbuild-all@lists.linux.dev
TO: Josh Poimboeuf <jpoimboe@kernel.org>
tree: https://git.kernel.org/pub/scm/linux/kernel/git/jpoimboe/linux.git objtool-diff
head: 745009dc796e56fc87e911138d801679ecd3576e
commit: 745009dc796e56fc87e911138d801679ecd3576e [2/2] test
:::::: branch date: 6 hours ago
:::::: commit date: 6 hours ago
compiler: clang version 18.1.5 (https://github.com/llvm/llvm-project 617a15a9eac96088ae5e9134248d8236e34b91b1)
If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/r/202405251400.UdnwcgiL-lkp@intel.com/
includecheck warnings: (new ones prefixed by >>)
>> kernel/fork.c: linux/livepatch.h is included more than once.
--
>> drivers/input/joydev.c: linux/module.h is included more than once.
--
>> drivers/input/misc/pcspkr.c: linux/module.h is included more than once.
vim +93 kernel/fork.c
> 93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 #include <linux/bpf.h>
100 #include <linux/stackprotector.h>
101 #include <linux/user_events.h>
102 #include <linux/iommu.h>
103 #include <linux/rseq.h>
104 #include <uapi/linux/pidfd.h>
105 #include <linux/pidfs.h>
106
107 #include <asm/pgalloc.h>
108 #include <linux/uaccess.h>
109 #include <asm/mmu_context.h>
110 #include <asm/cacheflush.h>
111 #include <asm/tlbflush.h>
112
113 #include <trace/events/sched.h>
114
115 #define CREATE_TRACE_POINTS
116 #include <trace/events/task.h>
117
118 /*
119 * Minimum number of threads to boot the kernel
120 */
121 #define MIN_THREADS 20
122
123 /*
124 * Maximum number of threads
125 */
126 #define MAX_THREADS FUTEX_TID_MASK
127
128 /*
129 * Protected counters by write_lock_irq(&tasklist_lock)
130 */
131 unsigned long total_forks; /* Handle normal Linux uptimes. */
132 int nr_threads; /* The idle threads do not count.. */
133
134 static int max_threads; /* tunable limit on nr_threads */
135
136 #define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
137
138 static const char * const resident_page_types[] = {
139 NAMED_ARRAY_INDEX(MM_FILEPAGES),
140 NAMED_ARRAY_INDEX(MM_ANONPAGES),
141 NAMED_ARRAY_INDEX(MM_SWAPENTS),
142 NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
143 };
144
145 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
146
147 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
148
149 #ifdef CONFIG_PROVE_RCU
150 int lockdep_tasklist_lock_is_held(void)
151 {
152 return lockdep_is_held(&tasklist_lock);
153 }
154 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
155 #endif /* #ifdef CONFIG_PROVE_RCU */
156
157 int nr_processes(void)
158 {
159 int cpu;
160 int total = 0;
161
162 for_each_possible_cpu(cpu)
163 total += per_cpu(process_counts, cpu);
164
165 return total;
166 }
167
168 void __weak arch_release_task_struct(struct task_struct *tsk)
169 {
170 }
171
172 static struct kmem_cache *task_struct_cachep;
173
174 static inline struct task_struct *alloc_task_struct_node(int node)
175 {
176 return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
177 }
178
179 static inline void free_task_struct(struct task_struct *tsk)
180 {
181 kmem_cache_free(task_struct_cachep, tsk);
182 }
183
184 /*
185 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
186 * kmemcache based allocator.
187 */
188 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
189
190 # ifdef CONFIG_VMAP_STACK
191 /*
192 * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
193 * flush. Try to minimize the number of calls by caching stacks.
194 */
195 #define NR_CACHED_STACKS 2
196 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
197
198 struct vm_stack {
199 struct rcu_head rcu;
200 struct vm_struct *stack_vm_area;
201 };
202
203 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
204 {
205 unsigned int i;
206
207 for (i = 0; i < NR_CACHED_STACKS; i++) {
208 if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
209 continue;
210 return true;
211 }
212 return false;
213 }
214
215 static void thread_stack_free_rcu(struct rcu_head *rh)
216 {
217 struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
218
219 if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
220 return;
221
222 vfree(vm_stack);
223 }
224
225 static void thread_stack_delayed_free(struct task_struct *tsk)
226 {
227 struct vm_stack *vm_stack = tsk->stack;
228
229 vm_stack->stack_vm_area = tsk->stack_vm_area;
230 call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
231 }
232
233 static int free_vm_stack_cache(unsigned int cpu)
234 {
235 struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
236 int i;
237
238 for (i = 0; i < NR_CACHED_STACKS; i++) {
239 struct vm_struct *vm_stack = cached_vm_stacks[i];
240
241 if (!vm_stack)
242 continue;
243
244 vfree(vm_stack->addr);
245 cached_vm_stacks[i] = NULL;
246 }
247
248 return 0;
249 }
250
251 static int memcg_charge_kernel_stack(struct vm_struct *vm)
252 {
253 int i;
254 int ret;
255 int nr_charged = 0;
256
257 BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
258
259 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
260 ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
261 if (ret)
262 goto err;
263 nr_charged++;
264 }
265 return 0;
266 err:
267 for (i = 0; i < nr_charged; i++)
268 memcg_kmem_uncharge_page(vm->pages[i], 0);
269 return ret;
270 }
271
272 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
273 {
274 struct vm_struct *vm;
275 void *stack;
276 int i;
277
278 for (i = 0; i < NR_CACHED_STACKS; i++) {
279 struct vm_struct *s;
280
281 s = this_cpu_xchg(cached_stacks[i], NULL);
282
283 if (!s)
284 continue;
285
286 /* Reset stack metadata. */
287 kasan_unpoison_range(s->addr, THREAD_SIZE);
288
289 stack = kasan_reset_tag(s->addr);
290
291 /* Clear stale pointers from reused stack. */
292 memset(stack, 0, THREAD_SIZE);
293
294 if (memcg_charge_kernel_stack(s)) {
295 vfree(s->addr);
296 return -ENOMEM;
297 }
298
299 tsk->stack_vm_area = s;
300 tsk->stack = stack;
301 return 0;
302 }
303
304 /*
305 * Allocated stacks are cached and later reused by new threads,
306 * so memcg accounting is performed manually on assigning/releasing
307 * stacks to tasks. Drop __GFP_ACCOUNT.
308 */
309 stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
310 VMALLOC_START, VMALLOC_END,
311 THREADINFO_GFP & ~__GFP_ACCOUNT,
312 PAGE_KERNEL,
313 0, node, __builtin_return_address(0));
314 if (!stack)
315 return -ENOMEM;
316
317 vm = find_vm_area(stack);
318 if (memcg_charge_kernel_stack(vm)) {
319 vfree(stack);
320 return -ENOMEM;
321 }
322 /*
323 * We can't call find_vm_area() in interrupt context, and
324 * free_thread_stack() can be called in interrupt context,
325 * so cache the vm_struct.
326 */
327 tsk->stack_vm_area = vm;
328 stack = kasan_reset_tag(stack);
329 tsk->stack = stack;
330 return 0;
331 }
332
333 static void free_thread_stack(struct task_struct *tsk)
334 {
335 if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
336 thread_stack_delayed_free(tsk);
337
338 tsk->stack = NULL;
339 tsk->stack_vm_area = NULL;
340 }
341
342 # else /* !CONFIG_VMAP_STACK */
343
344 static void thread_stack_free_rcu(struct rcu_head *rh)
345 {
346 __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
347 }
348
349 static void thread_stack_delayed_free(struct task_struct *tsk)
350 {
351 struct rcu_head *rh = tsk->stack;
352
353 call_rcu(rh, thread_stack_free_rcu);
354 }
355
356 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
357 {
358 struct page *page = alloc_pages_node(node, THREADINFO_GFP,
359 THREAD_SIZE_ORDER);
360
361 if (likely(page)) {
362 tsk->stack = kasan_reset_tag(page_address(page));
363 return 0;
364 }
365 return -ENOMEM;
366 }
367
368 static void free_thread_stack(struct task_struct *tsk)
369 {
370 thread_stack_delayed_free(tsk);
371 tsk->stack = NULL;
372 }
373
374 # endif /* CONFIG_VMAP_STACK */
375 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
376
377 static struct kmem_cache *thread_stack_cache;
378
379 static void thread_stack_free_rcu(struct rcu_head *rh)
380 {
381 kmem_cache_free(thread_stack_cache, rh);
382 }
383
384 static void thread_stack_delayed_free(struct task_struct *tsk)
385 {
386 struct rcu_head *rh = tsk->stack;
387
388 call_rcu(rh, thread_stack_free_rcu);
389 }
390
391 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
392 {
393 unsigned long *stack;
394 stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
395 stack = kasan_reset_tag(stack);
396 tsk->stack = stack;
397 return stack ? 0 : -ENOMEM;
398 }
399
400 static void free_thread_stack(struct task_struct *tsk)
401 {
402 thread_stack_delayed_free(tsk);
403 tsk->stack = NULL;
404 }
405
406 void thread_stack_cache_init(void)
407 {
408 thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
409 THREAD_SIZE, THREAD_SIZE, 0, 0,
410 THREAD_SIZE, NULL);
411 BUG_ON(thread_stack_cache == NULL);
412 }
413
414 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
415
416 /* SLAB cache for signal_struct structures (tsk->signal) */
417 static struct kmem_cache *signal_cachep;
418
419 /* SLAB cache for sighand_struct structures (tsk->sighand) */
420 struct kmem_cache *sighand_cachep;
421
422 /* SLAB cache for files_struct structures (tsk->files) */
423 struct kmem_cache *files_cachep;
424
425 /* SLAB cache for fs_struct structures (tsk->fs) */
426 struct kmem_cache *fs_cachep;
427
428 /* SLAB cache for vm_area_struct structures */
429 static struct kmem_cache *vm_area_cachep;
430
431 /* SLAB cache for mm_struct structures (tsk->mm) */
432 static struct kmem_cache *mm_cachep;
433
434 #ifdef CONFIG_PER_VMA_LOCK
435
436 /* SLAB cache for vm_area_struct.lock */
437 static struct kmem_cache *vma_lock_cachep;
438
439 static bool vma_lock_alloc(struct vm_area_struct *vma)
440 {
441 vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
442 if (!vma->vm_lock)
443 return false;
444
445 init_rwsem(&vma->vm_lock->lock);
446 vma->vm_lock_seq = -1;
447
448 return true;
449 }
450
451 static inline void vma_lock_free(struct vm_area_struct *vma)
452 {
453 kmem_cache_free(vma_lock_cachep, vma->vm_lock);
454 }
455
456 #else /* CONFIG_PER_VMA_LOCK */
457
458 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
459 static inline void vma_lock_free(struct vm_area_struct *vma) {}
460
461 #endif /* CONFIG_PER_VMA_LOCK */
462
463 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
464 {
465 struct vm_area_struct *vma;
466
467 vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
468 if (!vma)
469 return NULL;
470
471 vma_init(vma, mm);
472 if (!vma_lock_alloc(vma)) {
473 kmem_cache_free(vm_area_cachep, vma);
474 return NULL;
475 }
476
477 return vma;
478 }
479
480 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
481 {
482 struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
483
484 if (!new)
485 return NULL;
486
487 ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
488 ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
489 /*
490 * orig->shared.rb may be modified concurrently, but the clone
491 * will be reinitialized.
492 */
493 data_race(memcpy(new, orig, sizeof(*new)));
494 if (!vma_lock_alloc(new)) {
495 kmem_cache_free(vm_area_cachep, new);
496 return NULL;
497 }
498 INIT_LIST_HEAD(&new->anon_vma_chain);
499 vma_numab_state_init(new);
500 dup_anon_vma_name(orig, new);
501
502 return new;
503 }
504
505 void __vm_area_free(struct vm_area_struct *vma)
506 {
507 vma_numab_state_free(vma);
508 free_anon_vma_name(vma);
509 vma_lock_free(vma);
510 kmem_cache_free(vm_area_cachep, vma);
511 }
512
513 #ifdef CONFIG_PER_VMA_LOCK
514 static void vm_area_free_rcu_cb(struct rcu_head *head)
515 {
516 struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
517 vm_rcu);
518
519 /* The vma should not be locked while being destroyed. */
520 VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
521 __vm_area_free(vma);
522 }
523 #endif
524
525 void vm_area_free(struct vm_area_struct *vma)
526 {
527 #ifdef CONFIG_PER_VMA_LOCK
528 call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
529 #else
530 __vm_area_free(vma);
531 #endif
532 }
533
534 static void account_kernel_stack(struct task_struct *tsk, int account)
535 {
536 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
537 struct vm_struct *vm = task_stack_vm_area(tsk);
538 int i;
539
540 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
541 mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
542 account * (PAGE_SIZE / 1024));
543 } else {
544 void *stack = task_stack_page(tsk);
545
546 /* All stack pages are in the same node. */
547 mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
548 account * (THREAD_SIZE / 1024));
549 }
550 }
551
552 void exit_task_stack_account(struct task_struct *tsk)
553 {
554 account_kernel_stack(tsk, -1);
555
556 if (IS_ENABLED(CONFIG_VMAP_STACK)) {
557 struct vm_struct *vm;
558 int i;
559
560 vm = task_stack_vm_area(tsk);
561 for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
562 memcg_kmem_uncharge_page(vm->pages[i], 0);
563 }
564 }
565
566 static void release_task_stack(struct task_struct *tsk)
567 {
568 if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
569 return; /* Better to leak the stack than to free prematurely */
570
571 free_thread_stack(tsk);
572 }
573
574 #ifdef CONFIG_THREAD_INFO_IN_TASK
575 void put_task_stack(struct task_struct *tsk)
576 {
577 if (refcount_dec_and_test(&tsk->stack_refcount))
578 release_task_stack(tsk);
579 }
580 #endif
581
582 void free_task(struct task_struct *tsk)
583 {
584 #ifdef CONFIG_SECCOMP
585 WARN_ON_ONCE(tsk->seccomp.filter);
586 #endif
587 release_user_cpus_ptr(tsk);
588 scs_release(tsk);
589
590 #ifndef CONFIG_THREAD_INFO_IN_TASK
591 /*
592 * The task is finally done with both the stack and thread_info,
593 * so free both.
594 */
595 release_task_stack(tsk);
596 #else
597 /*
598 * If the task had a separate stack allocation, it should be gone
599 * by now.
600 */
601 WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
602 #endif
603 rt_mutex_debug_task_free(tsk);
604 ftrace_graph_exit_task(tsk);
605 arch_release_task_struct(tsk);
606 if (tsk->flags & PF_KTHREAD)
607 free_kthread_struct(tsk);
608 bpf_task_storage_free(tsk);
609 free_task_struct(tsk);
610 }
611 EXPORT_SYMBOL(free_task);
612
613 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
614 {
615 struct file *exe_file;
616
617 exe_file = get_mm_exe_file(oldmm);
618 RCU_INIT_POINTER(mm->exe_file, exe_file);
619 /*
620 * We depend on the oldmm having properly denied write access to the
621 * exe_file already.
622 */
623 if (exe_file && deny_write_access(exe_file))
624 pr_warn_once("deny_write_access() failed in %s\n", __func__);
625 }
626
627 #ifdef CONFIG_MMU
628 static __latent_entropy int dup_mmap(struct mm_struct *mm,
629 struct mm_struct *oldmm)
630 {
631 struct vm_area_struct *mpnt, *tmp;
632 int retval;
633 unsigned long charge = 0;
634 LIST_HEAD(uf);
635 VMA_ITERATOR(vmi, mm, 0);
636
637 uprobe_start_dup_mmap();
638 if (mmap_write_lock_killable(oldmm)) {
639 retval = -EINTR;
640 goto fail_uprobe_end;
641 }
642 flush_cache_dup_mm(oldmm);
643 uprobe_dup_mmap(oldmm, mm);
644 /*
645 * Not linked in yet - no deadlock potential:
646 */
647 mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
648
649 /* No ordering required: file already has been exposed. */
650 dup_mm_exe_file(mm, oldmm);
651
652 mm->total_vm = oldmm->total_vm;
653 mm->data_vm = oldmm->data_vm;
654 mm->exec_vm = oldmm->exec_vm;
655 mm->stack_vm = oldmm->stack_vm;
656
657 retval = ksm_fork(mm, oldmm);
658 if (retval)
659 goto out;
660 khugepaged_fork(mm, oldmm);
661
662 /* Use __mt_dup() to efficiently build an identical maple tree. */
663 retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
664 if (unlikely(retval))
665 goto out;
666
667 mt_clear_in_rcu(vmi.mas.tree);
668 for_each_vma(vmi, mpnt) {
669 struct file *file;
670
671 vma_start_write(mpnt);
672 if (mpnt->vm_flags & VM_DONTCOPY) {
673 retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
674 mpnt->vm_end, GFP_KERNEL);
675 if (retval)
676 goto loop_out;
677
678 vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
679 continue;
680 }
681 charge = 0;
682 /*
683 * Don't duplicate many vmas if we've been oom-killed (for
684 * example)
685 */
686 if (fatal_signal_pending(current)) {
687 retval = -EINTR;
688 goto loop_out;
689 }
690 if (mpnt->vm_flags & VM_ACCOUNT) {
691 unsigned long len = vma_pages(mpnt);
692
693 if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
694 goto fail_nomem;
695 charge = len;
696 }
697 tmp = vm_area_dup(mpnt);
698 if (!tmp)
699 goto fail_nomem;
700 retval = vma_dup_policy(mpnt, tmp);
701 if (retval)
702 goto fail_nomem_policy;
703 tmp->vm_mm = mm;
704 retval = dup_userfaultfd(tmp, &uf);
705 if (retval)
706 goto fail_nomem_anon_vma_fork;
707 if (tmp->vm_flags & VM_WIPEONFORK) {
708 /*
709 * VM_WIPEONFORK gets a clean slate in the child.
710 * Don't prepare anon_vma until fault since we don't
711 * copy page for current vma.
712 */
713 tmp->anon_vma = NULL;
714 } else if (anon_vma_fork(tmp, mpnt))
715 goto fail_nomem_anon_vma_fork;
716 vm_flags_clear(tmp, VM_LOCKED_MASK);
717 /*
718 * Copy/update hugetlb private vma information.
719 */
720 if (is_vm_hugetlb_page(tmp))
721 hugetlb_dup_vma_private(tmp);
722
723 /*
724 * Link the vma into the MT. After using __mt_dup(), memory
725 * allocation is not necessary here, so it cannot fail.
726 */
727 vma_iter_bulk_store(&vmi, tmp);
728
729 mm->map_count++;
730
731 if (tmp->vm_ops && tmp->vm_ops->open)
732 tmp->vm_ops->open(tmp);
733
734 file = tmp->vm_file;
735 if (file) {
736 struct address_space *mapping = file->f_mapping;
737
738 get_file(file);
739 i_mmap_lock_write(mapping);
740 if (vma_is_shared_maywrite(tmp))
741 mapping_allow_writable(mapping);
742 flush_dcache_mmap_lock(mapping);
743 /* insert tmp into the share list, just after mpnt */
744 vma_interval_tree_insert_after(tmp, mpnt,
745 &mapping->i_mmap);
746 flush_dcache_mmap_unlock(mapping);
747 i_mmap_unlock_write(mapping);
748 }
749
750 if (!(tmp->vm_flags & VM_WIPEONFORK))
751 retval = copy_page_range(tmp, mpnt);
752
753 if (retval) {
754 mpnt = vma_next(&vmi);
755 goto loop_out;
756 }
757 }
758 /* a new mm has just been created */
759 retval = arch_dup_mmap(oldmm, mm);
760 loop_out:
761 vma_iter_free(&vmi);
762 if (!retval) {
763 mt_set_in_rcu(vmi.mas.tree);
764 } else if (mpnt) {
765 /*
766 * The entire maple tree has already been duplicated. If the
767 * mmap duplication fails, mark the failure point with
768 * XA_ZERO_ENTRY. In exit_mmap(), if this marker is encountered,
769 * stop releasing VMAs that have not been duplicated after this
770 * point.
771 */
772 mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
773 mas_store(&vmi.mas, XA_ZERO_ENTRY);
774 }
775 out:
776 mmap_write_unlock(mm);
777 flush_tlb_mm(oldmm);
778 mmap_write_unlock(oldmm);
779 dup_userfaultfd_complete(&uf);
780 fail_uprobe_end:
781 uprobe_end_dup_mmap();
782 return retval;
783
784 fail_nomem_anon_vma_fork:
785 mpol_put(vma_policy(tmp));
786 fail_nomem_policy:
787 vm_area_free(tmp);
788 fail_nomem:
789 retval = -ENOMEM;
790 vm_unacct_memory(charge);
791 goto loop_out;
792 }
793
794 static inline int mm_alloc_pgd(struct mm_struct *mm)
795 {
796 mm->pgd = pgd_alloc(mm);
797 if (unlikely(!mm->pgd))
798 return -ENOMEM;
799 return 0;
800 }
801
802 static inline void mm_free_pgd(struct mm_struct *mm)
803 {
804 pgd_free(mm, mm->pgd);
805 }
806 #else
807 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
808 {
809 mmap_write_lock(oldmm);
810 dup_mm_exe_file(mm, oldmm);
811 mmap_write_unlock(oldmm);
812 return 0;
813 }
814 #define mm_alloc_pgd(mm) (0)
815 #define mm_free_pgd(mm)
816 #endif /* CONFIG_MMU */
817
818 static void check_mm(struct mm_struct *mm)
819 {
820 int i;
821
822 BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
823 "Please make sure 'struct resident_page_types[]' is updated as well");
824
825 for (i = 0; i < NR_MM_COUNTERS; i++) {
826 long x = percpu_counter_sum(&mm->rss_stat[i]);
827
828 if (unlikely(x))
829 pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
830 mm, resident_page_types[i], x);
831 }
832
833 if (mm_pgtables_bytes(mm))
834 pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
835 mm_pgtables_bytes(mm));
836
837 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
838 VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
839 #endif
840 }
841
842 #define allocate_mm() (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
843 #define free_mm(mm) (kmem_cache_free(mm_cachep, (mm)))
844
845 static void do_check_lazy_tlb(void *arg)
846 {
847 struct mm_struct *mm = arg;
848
849 WARN_ON_ONCE(current->active_mm == mm);
850 }
851
852 static void do_shoot_lazy_tlb(void *arg)
853 {
854 struct mm_struct *mm = arg;
855
856 if (current->active_mm == mm) {
857 WARN_ON_ONCE(current->mm);
858 current->active_mm = &init_mm;
859 switch_mm(mm, &init_mm, current);
860 }
861 }
862
863 static void cleanup_lazy_tlbs(struct mm_struct *mm)
864 {
865 if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
866 /*
867 * In this case, lazy tlb mms are refounted and would not reach
868 * __mmdrop until all CPUs have switched away and mmdrop()ed.
869 */
870 return;
871 }
872
873 /*
874 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
875 * requires lazy mm users to switch to another mm when the refcount
876 * drops to zero, before the mm is freed. This requires IPIs here to
877 * switch kernel threads to init_mm.
878 *
879 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
880 * switch with the final userspace teardown TLB flush which leaves the
881 * mm lazy on this CPU but no others, reducing the need for additional
882 * IPIs here. There are cases where a final IPI is still required here,
883 * such as the final mmdrop being performed on a different CPU than the
884 * one exiting, or kernel threads using the mm when userspace exits.
885 *
886 * IPI overheads have not found to be expensive, but they could be
887 * reduced in a number of possible ways, for example (roughly
888 * increasing order of complexity):
889 * - The last lazy reference created by exit_mm() could instead switch
890 * to init_mm, however it's probable this will run on the same CPU
891 * immediately afterwards, so this may not reduce IPIs much.
892 * - A batch of mms requiring IPIs could be gathered and freed at once.
893 * - CPUs store active_mm where it can be remotely checked without a
894 * lock, to filter out false-positives in the cpumask.
895 * - After mm_users or mm_count reaches zero, switching away from the
896 * mm could clear mm_cpumask to reduce some IPIs, perhaps together
897 * with some batching or delaying of the final IPIs.
898 * - A delayed freeing and RCU-like quiescing sequence based on mm
899 * switching to avoid IPIs completely.
900 */
901 on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
902 if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
903 on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
904 }
905
906 /*
907 * Called when the last reference to the mm
908 * is dropped: either by a lazy thread or by
909 * mmput. Free the page directory and the mm.
910 */
911 void __mmdrop(struct mm_struct *mm)
912 {
913 BUG_ON(mm == &init_mm);
914 WARN_ON_ONCE(mm == current->mm);
915
916 /* Ensure no CPUs are using this as their lazy tlb mm */
917 cleanup_lazy_tlbs(mm);
918
919 WARN_ON_ONCE(mm == current->active_mm);
920 mm_free_pgd(mm);
921 destroy_context(mm);
922 mmu_notifier_subscriptions_destroy(mm);
923 check_mm(mm);
924 put_user_ns(mm->user_ns);
925 mm_pasid_drop(mm);
926 mm_destroy_cid(mm);
927 percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
928
929 free_mm(mm);
930 }
931 EXPORT_SYMBOL_GPL(__mmdrop);
932
933 static void mmdrop_async_fn(struct work_struct *work)
934 {
935 struct mm_struct *mm;
936
937 mm = container_of(work, struct mm_struct, async_put_work);
938 __mmdrop(mm);
939 }
940
941 static void mmdrop_async(struct mm_struct *mm)
942 {
943 if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
944 INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
945 schedule_work(&mm->async_put_work);
946 }
947 }
948
949 static inline void free_signal_struct(struct signal_struct *sig)
950 {
951 taskstats_tgid_free(sig);
952 sched_autogroup_exit(sig);
953 /*
954 * __mmdrop is not safe to call from softirq context on x86 due to
955 * pgd_dtor so postpone it to the async context
956 */
957 if (sig->oom_mm)
958 mmdrop_async(sig->oom_mm);
959 kmem_cache_free(signal_cachep, sig);
960 }
961
962 static inline void put_signal_struct(struct signal_struct *sig)
963 {
964 if (refcount_dec_and_test(&sig->sigcnt))
965 free_signal_struct(sig);
966 }
967
968 void __put_task_struct(struct task_struct *tsk)
969 {
970 WARN_ON(!tsk->exit_state);
971 WARN_ON(refcount_read(&tsk->usage));
972 WARN_ON(tsk == current);
973
974 io_uring_free(tsk);
975 cgroup_free(tsk);
976 task_numa_free(tsk, true);
977 security_task_free(tsk);
978 exit_creds(tsk);
979 delayacct_tsk_free(tsk);
980 put_signal_struct(tsk->signal);
981 sched_core_free(tsk);
982 free_task(tsk);
983 }
984 EXPORT_SYMBOL_GPL(__put_task_struct);
985
986 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
987 {
988 struct task_struct *task = container_of(rhp, struct task_struct, rcu);
989
990 __put_task_struct(task);
991 }
992 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
993
994 void __init __weak arch_task_cache_init(void) { }
995
996 /*
997 * set_max_threads
998 */
999 static void set_max_threads(unsigned int max_threads_suggested)
1000 {
1001 u64 threads;
1002 unsigned long nr_pages = totalram_pages();
1003
1004 /*
1005 * The number of threads shall be limited such that the thread
1006 * structures may only consume a small part of the available memory.
1007 */
1008 if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1009 threads = MAX_THREADS;
1010 else
1011 threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1012 (u64) THREAD_SIZE * 8UL);
1013
1014 if (threads > max_threads_suggested)
1015 threads = max_threads_suggested;
1016
1017 max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1018 }
1019
1020 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1021 /* Initialized by the architecture: */
1022 int arch_task_struct_size __read_mostly;
1023 #endif
1024
1025 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
1026 {
1027 /* Fetch thread_struct whitelist for the architecture. */
1028 arch_thread_struct_whitelist(offset, size);
1029
1030 /*
1031 * Handle zero-sized whitelist or empty thread_struct, otherwise
1032 * adjust offset to position of thread_struct in task_struct.
1033 */
1034 if (unlikely(*size == 0))
1035 *offset = 0;
1036 else
1037 *offset += offsetof(struct task_struct, thread);
1038 }
1039
1040 void __init fork_init(void)
1041 {
1042 int i;
1043 #ifndef ARCH_MIN_TASKALIGN
1044 #define ARCH_MIN_TASKALIGN 0
1045 #endif
1046 int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1047 unsigned long useroffset, usersize;
1048
1049 /* create a slab on which task_structs can be allocated */
1050 task_struct_whitelist(&useroffset, &usersize);
1051 task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1052 arch_task_struct_size, align,
1053 SLAB_PANIC|SLAB_ACCOUNT,
1054 useroffset, usersize, NULL);
1055
1056 /* do the arch specific task caches init */
1057 arch_task_cache_init();
1058
1059 set_max_threads(MAX_THREADS);
1060
1061 init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1062 init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1063 init_task.signal->rlim[RLIMIT_SIGPENDING] =
1064 init_task.signal->rlim[RLIMIT_NPROC];
1065
1066 for (i = 0; i < UCOUNT_COUNTS; i++)
1067 init_user_ns.ucount_max[i] = max_threads/2;
1068
1069 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC, RLIM_INFINITY);
1070 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE, RLIM_INFINITY);
1071 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1072 set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK, RLIM_INFINITY);
1073
1074 #ifdef CONFIG_VMAP_STACK
1075 cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1076 NULL, free_vm_stack_cache);
1077 #endif
1078
1079 scs_init();
1080
1081 lockdep_init_task(&init_task);
1082 uprobes_init();
1083 }
1084
1085 int __weak arch_dup_task_struct(struct task_struct *dst,
1086 struct task_struct *src)
1087 {
1088 *dst = *src;
1089 return 0;
1090 }
1091
1092 void set_task_stack_end_magic(struct task_struct *tsk)
1093 {
1094 unsigned long *stackend;
1095
1096 stackend = end_of_stack(tsk);
1097 *stackend = STACK_END_MAGIC; /* for overflow detection */
1098 }
1099
1100 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1101 {
1102 struct task_struct *tsk;
1103 int err;
1104
1105 if (node == NUMA_NO_NODE)
1106 node = tsk_fork_get_node(orig);
1107 tsk = alloc_task_struct_node(node);
1108 if (!tsk)
1109 return NULL;
1110
1111 err = arch_dup_task_struct(tsk, orig);
1112 if (err)
1113 goto free_tsk;
1114
1115 err = alloc_thread_stack_node(tsk, node);
1116 if (err)
1117 goto free_tsk;
1118
1119 #ifdef CONFIG_THREAD_INFO_IN_TASK
1120 refcount_set(&tsk->stack_refcount, 1);
1121 #endif
1122 account_kernel_stack(tsk, 1);
1123
1124 err = scs_prepare(tsk, node);
1125 if (err)
1126 goto free_stack;
1127
1128 #ifdef CONFIG_SECCOMP
1129 /*
1130 * We must handle setting up seccomp filters once we're under
1131 * the sighand lock in case orig has changed between now and
1132 * then. Until then, filter must be NULL to avoid messing up
1133 * the usage counts on the error path calling free_task.
1134 */
1135 tsk->seccomp.filter = NULL;
1136 #endif
1137
1138 setup_thread_stack(tsk, orig);
1139 clear_user_return_notifier(tsk);
1140 clear_tsk_need_resched(tsk);
1141 set_task_stack_end_magic(tsk);
1142 clear_syscall_work_syscall_user_dispatch(tsk);
1143
1144 #ifdef CONFIG_STACKPROTECTOR
1145 tsk->stack_canary = get_random_canary();
1146 #endif
1147 if (orig->cpus_ptr == &orig->cpus_mask)
1148 tsk->cpus_ptr = &tsk->cpus_mask;
1149 dup_user_cpus_ptr(tsk, orig, node);
1150
1151 /*
1152 * One for the user space visible state that goes away when reaped.
1153 * One for the scheduler.
1154 */
1155 refcount_set(&tsk->rcu_users, 2);
1156 /* One for the rcu users */
1157 refcount_set(&tsk->usage, 1);
1158 #ifdef CONFIG_BLK_DEV_IO_TRACE
1159 tsk->btrace_seq = 0;
1160 #endif
1161 tsk->splice_pipe = NULL;
1162 tsk->task_frag.page = NULL;
1163 tsk->wake_q.next = NULL;
1164 tsk->worker_private = NULL;
1165
1166 kcov_task_init(tsk);
1167 kmsan_task_create(tsk);
1168 kmap_local_fork(tsk);
1169
1170 #ifdef CONFIG_FAULT_INJECTION
1171 tsk->fail_nth = 0;
1172 #endif
1173
1174 #ifdef CONFIG_BLK_CGROUP
1175 tsk->throttle_disk = NULL;
1176 tsk->use_memdelay = 0;
1177 #endif
1178
1179 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1180 tsk->pasid_activated = 0;
1181 #endif
1182
1183 #ifdef CONFIG_MEMCG
1184 tsk->active_memcg = NULL;
1185 #endif
1186
1187 #ifdef CONFIG_CPU_SUP_INTEL
1188 tsk->reported_split_lock = 0;
1189 #endif
1190
1191 #ifdef CONFIG_SCHED_MM_CID
1192 tsk->mm_cid = -1;
1193 tsk->last_mm_cid = -1;
1194 tsk->mm_cid_active = 0;
1195 tsk->migrate_from_cpu = -1;
1196 #endif
1197 return tsk;
1198
1199 free_stack:
1200 exit_task_stack_account(tsk);
1201 free_thread_stack(tsk);
1202 free_tsk:
1203 free_task_struct(tsk);
1204 return NULL;
1205 }
1206
1207 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1208
1209 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1210
1211 static int __init coredump_filter_setup(char *s)
1212 {
1213 default_dump_filter =
1214 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1215 MMF_DUMP_FILTER_MASK;
1216 return 1;
1217 }
1218
1219 __setup("coredump_filter=", coredump_filter_setup);
1220
1221 #include <linux/init_task.h>
1222
1223 static void mm_init_aio(struct mm_struct *mm)
1224 {
1225 #ifdef CONFIG_AIO
1226 spin_lock_init(&mm->ioctx_lock);
1227 mm->ioctx_table = NULL;
1228 #endif
1229 }
1230
1231 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1232 struct task_struct *p)
1233 {
1234 #ifdef CONFIG_MEMCG
1235 if (mm->owner == p)
1236 WRITE_ONCE(mm->owner, NULL);
1237 #endif
1238 }
1239
1240 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1241 {
1242 #ifdef CONFIG_MEMCG
1243 mm->owner = p;
1244 #endif
1245 }
1246
1247 static void mm_init_uprobes_state(struct mm_struct *mm)
1248 {
1249 #ifdef CONFIG_UPROBES
1250 mm->uprobes_state.xol_area = NULL;
1251 #endif
1252 }
1253
1254 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1255 struct user_namespace *user_ns)
1256 {
1257 mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1258 mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1259 atomic_set(&mm->mm_users, 1);
1260 atomic_set(&mm->mm_count, 1);
1261 seqcount_init(&mm->write_protect_seq);
1262 mmap_init_lock(mm);
1263 INIT_LIST_HEAD(&mm->mmlist);
1264 #ifdef CONFIG_PER_VMA_LOCK
1265 mm->mm_lock_seq = 0;
1266 #endif
1267 mm_pgtables_bytes_init(mm);
1268 mm->map_count = 0;
1269 mm->locked_vm = 0;
1270 atomic64_set(&mm->pinned_vm, 0);
1271 memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1272 spin_lock_init(&mm->page_table_lock);
1273 spin_lock_init(&mm->arg_lock);
1274 mm_init_cpumask(mm);
1275 mm_init_aio(mm);
1276 mm_init_owner(mm, p);
1277 mm_pasid_init(mm);
1278 RCU_INIT_POINTER(mm->exe_file, NULL);
1279 mmu_notifier_subscriptions_init(mm);
1280 init_tlb_flush_pending(mm);
1281 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1282 mm->pmd_huge_pte = NULL;
1283 #endif
1284 mm_init_uprobes_state(mm);
1285 hugetlb_count_init(mm);
1286
1287 if (current->mm) {
1288 mm->flags = mmf_init_flags(current->mm->flags);
1289 mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1290 } else {
1291 mm->flags = default_dump_filter;
1292 mm->def_flags = 0;
1293 }
1294
1295 if (mm_alloc_pgd(mm))
1296 goto fail_nopgd;
1297
1298 if (init_new_context(p, mm))
1299 goto fail_nocontext;
1300
1301 if (mm_alloc_cid(mm))
1302 goto fail_cid;
1303
1304 if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1305 NR_MM_COUNTERS))
1306 goto fail_pcpu;
1307
1308 mm->user_ns = get_user_ns(user_ns);
1309 lru_gen_init_mm(mm);
1310 return mm;
1311
1312 fail_pcpu:
1313 mm_destroy_cid(mm);
1314 fail_cid:
1315 destroy_context(mm);
1316 fail_nocontext:
1317 mm_free_pgd(mm);
1318 fail_nopgd:
1319 free_mm(mm);
1320 return NULL;
1321 }
1322
1323 /*
1324 * Allocate and initialize an mm_struct.
1325 */
1326 struct mm_struct *mm_alloc(void)
1327 {
1328 struct mm_struct *mm;
1329
1330 mm = allocate_mm();
1331 if (!mm)
1332 return NULL;
1333
1334 memset(mm, 0, sizeof(*mm));
1335 return mm_init(mm, current, current_user_ns());
1336 }
1337
1338 static inline void __mmput(struct mm_struct *mm)
1339 {
1340 VM_BUG_ON(atomic_read(&mm->mm_users));
1341
1342 uprobe_clear_state(mm);
1343 exit_aio(mm);
1344 ksm_exit(mm);
1345 khugepaged_exit(mm); /* must run before exit_mmap */
1346 exit_mmap(mm);
1347 mm_put_huge_zero_page(mm);
1348 set_mm_exe_file(mm, NULL);
1349 if (!list_empty(&mm->mmlist)) {
1350 spin_lock(&mmlist_lock);
1351 list_del(&mm->mmlist);
1352 spin_unlock(&mmlist_lock);
1353 }
1354 if (mm->binfmt)
1355 module_put(mm->binfmt->module);
1356 lru_gen_del_mm(mm);
1357 mmdrop(mm);
1358 }
1359
1360 /*
1361 * Decrement the use count and release all resources for an mm.
1362 */
1363 void mmput(struct mm_struct *mm)
1364 {
1365 might_sleep();
1366
1367 if (atomic_dec_and_test(&mm->mm_users))
1368 __mmput(mm);
1369 }
1370 EXPORT_SYMBOL_GPL(mmput);
1371
1372 #ifdef CONFIG_MMU
1373 static void mmput_async_fn(struct work_struct *work)
1374 {
1375 struct mm_struct *mm = container_of(work, struct mm_struct,
1376 async_put_work);
1377
1378 __mmput(mm);
1379 }
1380
1381 void mmput_async(struct mm_struct *mm)
1382 {
1383 if (atomic_dec_and_test(&mm->mm_users)) {
1384 INIT_WORK(&mm->async_put_work, mmput_async_fn);
1385 schedule_work(&mm->async_put_work);
1386 }
1387 }
1388 EXPORT_SYMBOL_GPL(mmput_async);
1389 #endif
1390
1391 /**
1392 * set_mm_exe_file - change a reference to the mm's executable file
1393 * @mm: The mm to change.
1394 * @new_exe_file: The new file to use.
1395 *
1396 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1397 *
1398 * Main users are mmput() and sys_execve(). Callers prevent concurrent
1399 * invocations: in mmput() nobody alive left, in execve it happens before
1400 * the new mm is made visible to anyone.
1401 *
1402 * Can only fail if new_exe_file != NULL.
1403 */
1404 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1405 {
1406 struct file *old_exe_file;
1407
1408 /*
1409 * It is safe to dereference the exe_file without RCU as
1410 * this function is only called if nobody else can access
1411 * this mm -- see comment above for justification.
1412 */
1413 old_exe_file = rcu_dereference_raw(mm->exe_file);
1414
1415 if (new_exe_file) {
1416 /*
1417 * We expect the caller (i.e., sys_execve) to already denied
1418 * write access, so this is unlikely to fail.
1419 */
1420 if (unlikely(deny_write_access(new_exe_file)))
1421 return -EACCES;
1422 get_file(new_exe_file);
1423 }
1424 rcu_assign_pointer(mm->exe_file, new_exe_file);
1425 if (old_exe_file) {
1426 allow_write_access(old_exe_file);
1427 fput(old_exe_file);
1428 }
1429 return 0;
1430 }
1431
1432 /**
1433 * replace_mm_exe_file - replace a reference to the mm's executable file
1434 * @mm: The mm to change.
1435 * @new_exe_file: The new file to use.
1436 *
1437 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1438 *
1439 * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1440 */
1441 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1442 {
1443 struct vm_area_struct *vma;
1444 struct file *old_exe_file;
1445 int ret = 0;
1446
1447 /* Forbid mm->exe_file change if old file still mapped. */
1448 old_exe_file = get_mm_exe_file(mm);
1449 if (old_exe_file) {
1450 VMA_ITERATOR(vmi, mm, 0);
1451 mmap_read_lock(mm);
1452 for_each_vma(vmi, vma) {
1453 if (!vma->vm_file)
1454 continue;
1455 if (path_equal(&vma->vm_file->f_path,
1456 &old_exe_file->f_path)) {
1457 ret = -EBUSY;
1458 break;
1459 }
1460 }
1461 mmap_read_unlock(mm);
1462 fput(old_exe_file);
1463 if (ret)
1464 return ret;
1465 }
1466
1467 ret = deny_write_access(new_exe_file);
1468 if (ret)
1469 return -EACCES;
1470 get_file(new_exe_file);
1471
1472 /* set the new file */
1473 mmap_write_lock(mm);
1474 old_exe_file = rcu_dereference_raw(mm->exe_file);
1475 rcu_assign_pointer(mm->exe_file, new_exe_file);
1476 mmap_write_unlock(mm);
1477
1478 if (old_exe_file) {
1479 allow_write_access(old_exe_file);
1480 fput(old_exe_file);
1481 }
1482 return 0;
1483 }
1484
1485 /**
1486 * get_mm_exe_file - acquire a reference to the mm's executable file
1487 * @mm: The mm of interest.
1488 *
1489 * Returns %NULL if mm has no associated executable file.
1490 * User must release file via fput().
1491 */
1492 struct file *get_mm_exe_file(struct mm_struct *mm)
1493 {
1494 struct file *exe_file;
1495
1496 rcu_read_lock();
1497 exe_file = get_file_rcu(&mm->exe_file);
1498 rcu_read_unlock();
1499 return exe_file;
1500 }
1501
1502 /**
1503 * get_task_exe_file - acquire a reference to the task's executable file
1504 * @task: The task.
1505 *
1506 * Returns %NULL if task's mm (if any) has no associated executable file or
1507 * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1508 * User must release file via fput().
1509 */
1510 struct file *get_task_exe_file(struct task_struct *task)
1511 {
1512 struct file *exe_file = NULL;
1513 struct mm_struct *mm;
1514
1515 task_lock(task);
1516 mm = task->mm;
1517 if (mm) {
1518 if (!(task->flags & PF_KTHREAD))
1519 exe_file = get_mm_exe_file(mm);
1520 }
1521 task_unlock(task);
1522 return exe_file;
1523 }
1524
1525 /**
1526 * get_task_mm - acquire a reference to the task's mm
1527 * @task: The task.
1528 *
1529 * Returns %NULL if the task has no mm. Checks PF_KTHREAD (meaning
1530 * this kernel workthread has transiently adopted a user mm with use_mm,
1531 * to do its AIO) is not set and if so returns a reference to it, after
1532 * bumping up the use count. User must release the mm via mmput()
1533 * after use. Typically used by /proc and ptrace.
1534 */
1535 struct mm_struct *get_task_mm(struct task_struct *task)
1536 {
1537 struct mm_struct *mm;
1538
1539 task_lock(task);
1540 mm = task->mm;
1541 if (mm) {
1542 if (task->flags & PF_KTHREAD)
1543 mm = NULL;
1544 else
1545 mmget(mm);
1546 }
1547 task_unlock(task);
1548 return mm;
1549 }
1550 EXPORT_SYMBOL_GPL(get_task_mm);
1551
1552 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1553 {
1554 struct mm_struct *mm;
1555 int err;
1556
1557 err = down_read_killable(&task->signal->exec_update_lock);
1558 if (err)
1559 return ERR_PTR(err);
1560
1561 mm = get_task_mm(task);
1562 if (mm && mm != current->mm &&
1563 !ptrace_may_access(task, mode)) {
1564 mmput(mm);
1565 mm = ERR_PTR(-EACCES);
1566 }
1567 up_read(&task->signal->exec_update_lock);
1568
1569 return mm;
1570 }
1571
1572 static void complete_vfork_done(struct task_struct *tsk)
1573 {
1574 struct completion *vfork;
1575
1576 task_lock(tsk);
1577 vfork = tsk->vfork_done;
1578 if (likely(vfork)) {
1579 tsk->vfork_done = NULL;
1580 complete(vfork);
1581 }
1582 task_unlock(tsk);
1583 }
1584
1585 static int wait_for_vfork_done(struct task_struct *child,
1586 struct completion *vfork)
1587 {
1588 unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1589 int killed;
1590
1591 cgroup_enter_frozen();
1592 killed = wait_for_completion_state(vfork, state);
1593 cgroup_leave_frozen(false);
1594
1595 if (killed) {
1596 task_lock(child);
1597 child->vfork_done = NULL;
1598 task_unlock(child);
1599 }
1600
1601 put_task_struct(child);
1602 return killed;
1603 }
1604
1605 /* Please note the differences between mmput and mm_release.
1606 * mmput is called whenever we stop holding onto a mm_struct,
1607 * error success whatever.
1608 *
1609 * mm_release is called after a mm_struct has been removed
1610 * from the current process.
1611 *
1612 * This difference is important for error handling, when we
1613 * only half set up a mm_struct for a new process and need to restore
1614 * the old one. Because we mmput the new mm_struct before
1615 * restoring the old one. . .
1616 * Eric Biederman 10 January 1998
1617 */
1618 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1619 {
1620 uprobe_free_utask(tsk);
1621
1622 /* Get rid of any cached register state */
1623 deactivate_mm(tsk, mm);
1624
1625 /*
1626 * Signal userspace if we're not exiting with a core dump
1627 * because we want to leave the value intact for debugging
1628 * purposes.
1629 */
1630 if (tsk->clear_child_tid) {
1631 if (atomic_read(&mm->mm_users) > 1) {
1632 /*
1633 * We don't check the error code - if userspace has
1634 * not set up a proper pointer then tough luck.
1635 */
1636 put_user(0, tsk->clear_child_tid);
1637 do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1638 1, NULL, NULL, 0, 0);
1639 }
1640 tsk->clear_child_tid = NULL;
1641 }
1642
1643 /*
1644 * All done, finally we can wake up parent and return this mm to him.
1645 * Also kthread_stop() uses this completion for synchronization.
1646 */
1647 if (tsk->vfork_done)
1648 complete_vfork_done(tsk);
1649 }
1650
1651 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1652 {
1653 futex_exit_release(tsk);
1654 mm_release(tsk, mm);
1655 }
1656
1657 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1658 {
1659 futex_exec_release(tsk);
1660 mm_release(tsk, mm);
1661 }
1662
1663 /**
1664 * dup_mm() - duplicates an existing mm structure
1665 * @tsk: the task_struct with which the new mm will be associated.
1666 * @oldmm: the mm to duplicate.
1667 *
1668 * Allocates a new mm structure and duplicates the provided @oldmm structure
1669 * content into it.
1670 *
1671 * Return: the duplicated mm or NULL on failure.
1672 */
1673 static struct mm_struct *dup_mm(struct task_struct *tsk,
1674 struct mm_struct *oldmm)
1675 {
1676 struct mm_struct *mm;
1677 int err;
1678
1679 mm = allocate_mm();
1680 if (!mm)
1681 goto fail_nomem;
1682
1683 memcpy(mm, oldmm, sizeof(*mm));
1684
1685 if (!mm_init(mm, tsk, mm->user_ns))
1686 goto fail_nomem;
1687
1688 err = dup_mmap(mm, oldmm);
1689 if (err)
1690 goto free_pt;
1691
1692 mm->hiwater_rss = get_mm_rss(mm);
1693 mm->hiwater_vm = mm->total_vm;
1694
1695 if (mm->binfmt && !try_module_get(mm->binfmt->module))
1696 goto free_pt;
1697
1698 return mm;
1699
1700 free_pt:
1701 /* don't put binfmt in mmput, we haven't got module yet */
1702 mm->binfmt = NULL;
1703 mm_init_owner(mm, NULL);
1704 mmput(mm);
1705
1706 fail_nomem:
1707 return NULL;
1708 }
1709
1710 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1711 {
1712 struct mm_struct *mm, *oldmm;
1713
1714 tsk->min_flt = tsk->maj_flt = 0;
1715 tsk->nvcsw = tsk->nivcsw = 0;
1716 #ifdef CONFIG_DETECT_HUNG_TASK
1717 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1718 tsk->last_switch_time = 0;
1719 #endif
1720
1721 tsk->mm = NULL;
1722 tsk->active_mm = NULL;
1723
1724 /*
1725 * Are we cloning a kernel thread?
1726 *
1727 * We need to steal a active VM for that..
1728 */
1729 oldmm = current->mm;
1730 if (!oldmm)
1731 return 0;
1732
1733 if (clone_flags & CLONE_VM) {
1734 mmget(oldmm);
1735 mm = oldmm;
1736 } else {
1737 mm = dup_mm(tsk, current->mm);
1738 if (!mm)
1739 return -ENOMEM;
1740 }
1741
1742 tsk->mm = mm;
1743 tsk->active_mm = mm;
1744 sched_mm_cid_fork(tsk);
1745 return 0;
1746 }
1747
1748 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1749 {
1750 struct fs_struct *fs = current->fs;
1751 if (clone_flags & CLONE_FS) {
1752 /* tsk->fs is already what we want */
1753 spin_lock(&fs->lock);
1754 /* "users" and "in_exec" locked for check_unsafe_exec() */
1755 if (fs->in_exec) {
1756 spin_unlock(&fs->lock);
1757 return -EAGAIN;
1758 }
1759 fs->users++;
1760 spin_unlock(&fs->lock);
1761 return 0;
1762 }
1763 tsk->fs = copy_fs_struct(fs);
1764 if (!tsk->fs)
1765 return -ENOMEM;
1766 return 0;
1767 }
1768
1769 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1770 int no_files)
1771 {
1772 struct files_struct *oldf, *newf;
1773 int error = 0;
1774
1775 /*
1776 * A background process may not have any files ...
1777 */
1778 oldf = current->files;
1779 if (!oldf)
1780 goto out;
1781
1782 if (no_files) {
1783 tsk->files = NULL;
1784 goto out;
1785 }
1786
1787 if (clone_flags & CLONE_FILES) {
1788 atomic_inc(&oldf->count);
1789 goto out;
1790 }
1791
1792 newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1793 if (!newf)
1794 goto out;
1795
1796 tsk->files = newf;
1797 error = 0;
1798 out:
1799 return error;
1800 }
1801
1802 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1803 {
1804 struct sighand_struct *sig;
1805
1806 if (clone_flags & CLONE_SIGHAND) {
1807 refcount_inc(¤t->sighand->count);
1808 return 0;
1809 }
1810 sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1811 RCU_INIT_POINTER(tsk->sighand, sig);
1812 if (!sig)
1813 return -ENOMEM;
1814
1815 refcount_set(&sig->count, 1);
1816 spin_lock_irq(¤t->sighand->siglock);
1817 memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1818 spin_unlock_irq(¤t->sighand->siglock);
1819
1820 /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1821 if (clone_flags & CLONE_CLEAR_SIGHAND)
1822 flush_signal_handlers(tsk, 0);
1823
1824 return 0;
1825 }
1826
1827 void __cleanup_sighand(struct sighand_struct *sighand)
1828 {
1829 if (refcount_dec_and_test(&sighand->count)) {
1830 signalfd_cleanup(sighand);
1831 /*
1832 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1833 * without an RCU grace period, see __lock_task_sighand().
1834 */
1835 kmem_cache_free(sighand_cachep, sighand);
1836 }
1837 }
1838
1839 /*
1840 * Initialize POSIX timer handling for a thread group.
1841 */
1842 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1843 {
1844 struct posix_cputimers *pct = &sig->posix_cputimers;
1845 unsigned long cpu_limit;
1846
1847 cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1848 posix_cputimers_group_init(pct, cpu_limit);
1849 }
1850
1851 void kpatch_foo(void);
1852 void kpatch_foo(void)
1853 {
1854 if (!jiffies)
1855 printk("kpatch copy signal\n");
1856 }
1857
1858 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1859 {
1860 struct signal_struct *sig;
1861
1862 kpatch_foo();
1863
1864 if (clone_flags & CLONE_THREAD)
1865 return 0;
1866
1867 sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1868 tsk->signal = sig;
1869 if (!sig)
1870 return -ENOMEM;
1871
1872 sig->nr_threads = 1;
1873 sig->quick_threads = 1;
1874 atomic_set(&sig->live, 1);
1875 refcount_set(&sig->sigcnt, 1);
1876
1877 /* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1878 sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1879 tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1880
1881 init_waitqueue_head(&sig->wait_chldexit);
1882 sig->curr_target = tsk;
1883 init_sigpending(&sig->shared_pending);
1884 INIT_HLIST_HEAD(&sig->multiprocess);
1885 seqlock_init(&sig->stats_lock);
1886 prev_cputime_init(&sig->prev_cputime);
1887
1888 #ifdef CONFIG_POSIX_TIMERS
1889 INIT_LIST_HEAD(&sig->posix_timers);
1890 hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1891 sig->real_timer.function = it_real_fn;
1892 #endif
1893
1894 task_lock(current->group_leader);
1895 memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1896 task_unlock(current->group_leader);
1897
1898 posix_cpu_timers_init_group(sig);
1899
1900 tty_audit_fork(sig);
1901 sched_autogroup_fork(sig);
1902
1903 sig->oom_score_adj = current->signal->oom_score_adj;
1904 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1905
1906 mutex_init(&sig->cred_guard_mutex);
1907 init_rwsem(&sig->exec_update_lock);
1908
1909 return 0;
1910 }
1911
1912 static void copy_seccomp(struct task_struct *p)
1913 {
1914 #ifdef CONFIG_SECCOMP
1915 /*
1916 * Must be called with sighand->lock held, which is common to
1917 * all threads in the group. Holding cred_guard_mutex is not
1918 * needed because this new task is not yet running and cannot
1919 * be racing exec.
1920 */
1921 assert_spin_locked(¤t->sighand->siglock);
1922
1923 /* Ref-count the new filter user, and assign it. */
1924 get_seccomp_filter(current);
1925 p->seccomp = current->seccomp;
1926
1927 /*
1928 * Explicitly enable no_new_privs here in case it got set
1929 * between the task_struct being duplicated and holding the
1930 * sighand lock. The seccomp state and nnp must be in sync.
1931 */
1932 if (task_no_new_privs(current))
1933 task_set_no_new_privs(p);
1934
1935 /*
1936 * If the parent gained a seccomp mode after copying thread
1937 * flags and between before we held the sighand lock, we have
1938 * to manually enable the seccomp thread flag here.
1939 */
1940 if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1941 set_task_syscall_work(p, SECCOMP);
1942 #endif
1943 }
1944
1945 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1946 {
1947 current->clear_child_tid = tidptr;
1948
1949 return task_pid_vnr(current);
1950 }
1951
1952 static void rt_mutex_init_task(struct task_struct *p)
1953 {
1954 raw_spin_lock_init(&p->pi_lock);
1955 #ifdef CONFIG_RT_MUTEXES
1956 p->pi_waiters = RB_ROOT_CACHED;
1957 p->pi_top_task = NULL;
1958 p->pi_blocked_on = NULL;
1959 #endif
1960 }
1961
1962 static inline void init_task_pid_links(struct task_struct *task)
1963 {
1964 enum pid_type type;
1965
1966 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1967 INIT_HLIST_NODE(&task->pid_links[type]);
1968 }
1969
1970 static inline void
1971 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1972 {
1973 if (type == PIDTYPE_PID)
1974 task->thread_pid = pid;
1975 else
1976 task->signal->pids[type] = pid;
1977 }
1978
1979 static inline void rcu_copy_process(struct task_struct *p)
1980 {
1981 #ifdef CONFIG_PREEMPT_RCU
1982 p->rcu_read_lock_nesting = 0;
1983 p->rcu_read_unlock_special.s = 0;
1984 p->rcu_blocked_node = NULL;
1985 INIT_LIST_HEAD(&p->rcu_node_entry);
1986 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1987 #ifdef CONFIG_TASKS_RCU
1988 p->rcu_tasks_holdout = false;
1989 INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1990 p->rcu_tasks_idle_cpu = -1;
1991 INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1992 #endif /* #ifdef CONFIG_TASKS_RCU */
1993 #ifdef CONFIG_TASKS_TRACE_RCU
1994 p->trc_reader_nesting = 0;
1995 p->trc_reader_special.s = 0;
1996 INIT_LIST_HEAD(&p->trc_holdout_list);
1997 INIT_LIST_HEAD(&p->trc_blkd_node);
1998 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1999 }
2000
2001 /**
2002 * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2003 * @pid: the struct pid for which to create a pidfd
2004 * @flags: flags of the new @pidfd
2005 * @ret: Where to return the file for the pidfd.
2006 *
2007 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2008 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2009 *
2010 * The helper doesn't perform checks on @pid which makes it useful for pidfds
2011 * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2012 * pidfd file are prepared.
2013 *
2014 * If this function returns successfully the caller is responsible to either
2015 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2016 * order to install the pidfd into its file descriptor table or they must use
2017 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2018 * respectively.
2019 *
2020 * This function is useful when a pidfd must already be reserved but there
2021 * might still be points of failure afterwards and the caller wants to ensure
2022 * that no pidfd is leaked into its file descriptor table.
2023 *
2024 * Return: On success, a reserved pidfd is returned from the function and a new
2025 * pidfd file is returned in the last argument to the function. On
2026 * error, a negative error code is returned from the function and the
2027 * last argument remains unchanged.
2028 */
2029 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2030 {
2031 int pidfd;
2032 struct file *pidfd_file;
2033
2034 pidfd = get_unused_fd_flags(O_CLOEXEC);
2035 if (pidfd < 0)
2036 return pidfd;
2037
2038 pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2039 if (IS_ERR(pidfd_file)) {
2040 put_unused_fd(pidfd);
2041 return PTR_ERR(pidfd_file);
2042 }
2043 /*
2044 * anon_inode_getfile() ignores everything outside of the
2045 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2046 */
2047 pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2048 *ret = pidfd_file;
2049 return pidfd;
2050 }
2051
2052 /**
2053 * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2054 * @pid: the struct pid for which to create a pidfd
2055 * @flags: flags of the new @pidfd
2056 * @ret: Where to return the pidfd.
2057 *
2058 * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2059 * caller's file descriptor table. The pidfd is reserved but not installed yet.
2060 *
2061 * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2062 * task identified by @pid must be a thread-group leader.
2063 *
2064 * If this function returns successfully the caller is responsible to either
2065 * call fd_install() passing the returned pidfd and pidfd file as arguments in
2066 * order to install the pidfd into its file descriptor table or they must use
2067 * put_unused_fd() and fput() on the returned pidfd and pidfd file
2068 * respectively.
2069 *
2070 * This function is useful when a pidfd must already be reserved but there
2071 * might still be points of failure afterwards and the caller wants to ensure
2072 * that no pidfd is leaked into its file descriptor table.
2073 *
2074 * Return: On success, a reserved pidfd is returned from the function and a new
2075 * pidfd file is returned in the last argument to the function. On
2076 * error, a negative error code is returned from the function and the
2077 * last argument remains unchanged.
2078 */
2079 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2080 {
2081 bool thread = flags & PIDFD_THREAD;
2082
2083 if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2084 return -EINVAL;
2085
2086 return __pidfd_prepare(pid, flags, ret);
2087 }
2088
2089 static void __delayed_free_task(struct rcu_head *rhp)
2090 {
2091 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2092
2093 free_task(tsk);
2094 }
2095
2096 static __always_inline void delayed_free_task(struct task_struct *tsk)
2097 {
2098 if (IS_ENABLED(CONFIG_MEMCG))
2099 call_rcu(&tsk->rcu, __delayed_free_task);
2100 else
2101 free_task(tsk);
2102 }
2103
2104 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2105 {
2106 /* Skip if kernel thread */
2107 if (!tsk->mm)
2108 return;
2109
2110 /* Skip if spawning a thread or using vfork */
2111 if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2112 return;
2113
2114 /* We need to synchronize with __set_oom_adj */
2115 mutex_lock(&oom_adj_mutex);
2116 set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2117 /* Update the values in case they were changed after copy_signal */
2118 tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2119 tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2120 mutex_unlock(&oom_adj_mutex);
2121 }
2122
2123 #ifdef CONFIG_RV
2124 static void rv_task_fork(struct task_struct *p)
2125 {
2126 int i;
2127
2128 for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2129 p->rv[i].da_mon.monitoring = false;
2130 }
2131 #else
2132 #define rv_task_fork(p) do {} while (0)
2133 #endif
2134
2135 /*
2136 * This creates a new process as a copy of the old one,
2137 * but does not actually start it yet.
2138 *
2139 * It copies the registers, and all the appropriate
2140 * parts of the process environment (as per the clone
2141 * flags). The actual kick-off is left to the caller.
2142 */
2143 __latent_entropy struct task_struct *copy_process(
2144 struct pid *pid,
2145 int trace,
2146 int node,
2147 struct kernel_clone_args *args)
2148 {
2149 int pidfd = -1, retval;
2150 struct task_struct *p;
2151 struct multiprocess_signals delayed;
2152 struct file *pidfile = NULL;
2153 const u64 clone_flags = args->flags;
2154 struct nsproxy *nsp = current->nsproxy;
2155
2156 /*
2157 * Don't allow sharing the root directory with processes in a different
2158 * namespace
2159 */
2160 if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2161 return ERR_PTR(-EINVAL);
2162
2163 if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2164 return ERR_PTR(-EINVAL);
2165
2166 /*
2167 * Thread groups must share signals as well, and detached threads
2168 * can only be started up within the thread group.
2169 */
2170 if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2171 return ERR_PTR(-EINVAL);
2172
2173 /*
2174 * Shared signal handlers imply shared VM. By way of the above,
2175 * thread groups also imply shared VM. Blocking this case allows
2176 * for various simplifications in other code.
2177 */
2178 if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2179 return ERR_PTR(-EINVAL);
2180
2181 /*
2182 * Siblings of global init remain as zombies on exit since they are
2183 * not reaped by their parent (swapper). To solve this and to avoid
2184 * multi-rooted process trees, prevent global and container-inits
2185 * from creating siblings.
2186 */
2187 if ((clone_flags & CLONE_PARENT) &&
2188 current->signal->flags & SIGNAL_UNKILLABLE)
2189 return ERR_PTR(-EINVAL);
2190
2191 /*
2192 * If the new process will be in a different pid or user namespace
2193 * do not allow it to share a thread group with the forking task.
2194 */
2195 if (clone_flags & CLONE_THREAD) {
2196 if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2197 (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2198 return ERR_PTR(-EINVAL);
2199 }
2200
2201 if (clone_flags & CLONE_PIDFD) {
2202 /*
2203 * - CLONE_DETACHED is blocked so that we can potentially
2204 * reuse it later for CLONE_PIDFD.
2205 */
2206 if (clone_flags & CLONE_DETACHED)
2207 return ERR_PTR(-EINVAL);
2208 }
2209
2210 /*
2211 * Force any signals received before this point to be delivered
2212 * before the fork happens. Collect up signals sent to multiple
2213 * processes that happen during the fork and delay them so that
2214 * they appear to happen after the fork.
2215 */
2216 sigemptyset(&delayed.signal);
2217 INIT_HLIST_NODE(&delayed.node);
2218
2219 spin_lock_irq(¤t->sighand->siglock);
2220 if (!(clone_flags & CLONE_THREAD))
2221 hlist_add_head(&delayed.node, ¤t->signal->multiprocess);
2222 recalc_sigpending();
2223 spin_unlock_irq(¤t->sighand->siglock);
2224 retval = -ERESTARTNOINTR;
2225 if (task_sigpending(current))
2226 goto fork_out;
2227
2228 retval = -ENOMEM;
2229 p = dup_task_struct(current, node);
2230 if (!p)
2231 goto fork_out;
2232 p->flags &= ~PF_KTHREAD;
2233 if (args->kthread)
2234 p->flags |= PF_KTHREAD;
2235 if (args->user_worker) {
2236 /*
2237 * Mark us a user worker, and block any signal that isn't
2238 * fatal or STOP
2239 */
2240 p->flags |= PF_USER_WORKER;
2241 siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2242 }
2243 if (args->io_thread)
2244 p->flags |= PF_IO_WORKER;
2245
2246 if (args->name)
2247 strscpy_pad(p->comm, args->name, sizeof(p->comm));
2248
2249 p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2250 /*
2251 * Clear TID on mm_release()?
2252 */
2253 p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2254
2255 ftrace_graph_init_task(p);
2256
2257 rt_mutex_init_task(p);
2258
2259 lockdep_assert_irqs_enabled();
2260 #ifdef CONFIG_PROVE_LOCKING
2261 DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2262 #endif
2263 retval = copy_creds(p, clone_flags);
2264 if (retval < 0)
2265 goto bad_fork_free;
2266
2267 retval = -EAGAIN;
2268 if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2269 if (p->real_cred->user != INIT_USER &&
2270 !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2271 goto bad_fork_cleanup_count;
2272 }
2273 current->flags &= ~PF_NPROC_EXCEEDED;
2274
2275 /*
2276 * If multiple threads are within copy_process(), then this check
2277 * triggers too late. This doesn't hurt, the check is only there
2278 * to stop root fork bombs.
2279 */
2280 retval = -EAGAIN;
2281 if (data_race(nr_threads >= max_threads))
2282 goto bad_fork_cleanup_count;
2283
2284 delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
2285 p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2286 p->flags |= PF_FORKNOEXEC;
2287 INIT_LIST_HEAD(&p->children);
2288 INIT_LIST_HEAD(&p->sibling);
2289 rcu_copy_process(p);
2290 p->vfork_done = NULL;
2291 spin_lock_init(&p->alloc_lock);
2292
2293 init_sigpending(&p->pending);
2294
2295 p->utime = p->stime = p->gtime = 0;
2296 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2297 p->utimescaled = p->stimescaled = 0;
2298 #endif
2299 prev_cputime_init(&p->prev_cputime);
2300
2301 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2302 seqcount_init(&p->vtime.seqcount);
2303 p->vtime.starttime = 0;
2304 p->vtime.state = VTIME_INACTIVE;
2305 #endif
2306
2307 #ifdef CONFIG_IO_URING
2308 p->io_uring = NULL;
2309 #endif
2310
2311 p->default_timer_slack_ns = current->timer_slack_ns;
2312
2313 #ifdef CONFIG_PSI
2314 p->psi_flags = 0;
2315 #endif
2316
2317 task_io_accounting_init(&p->ioac);
2318 acct_clear_integrals(p);
2319
2320 posix_cputimers_init(&p->posix_cputimers);
2321
2322 p->io_context = NULL;
2323 audit_set_context(p, NULL);
2324 cgroup_fork(p);
2325 if (args->kthread) {
2326 if (!set_kthread_struct(p))
2327 goto bad_fork_cleanup_delayacct;
2328 }
2329 #ifdef CONFIG_NUMA
2330 p->mempolicy = mpol_dup(p->mempolicy);
2331 if (IS_ERR(p->mempolicy)) {
2332 retval = PTR_ERR(p->mempolicy);
2333 p->mempolicy = NULL;
2334 goto bad_fork_cleanup_delayacct;
2335 }
2336 #endif
2337 #ifdef CONFIG_CPUSETS
2338 p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2339 p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2340 seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2341 #endif
2342 #ifdef CONFIG_TRACE_IRQFLAGS
2343 memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2344 p->irqtrace.hardirq_disable_ip = _THIS_IP_;
2345 p->irqtrace.softirq_enable_ip = _THIS_IP_;
2346 p->softirqs_enabled = 1;
2347 p->softirq_context = 0;
2348 #endif
2349
2350 p->pagefault_disabled = 0;
2351
2352 #ifdef CONFIG_LOCKDEP
2353 lockdep_init_task(p);
2354 #endif
2355
2356 #ifdef CONFIG_DEBUG_MUTEXES
2357 p->blocked_on = NULL; /* not blocked yet */
2358 #endif
2359 #ifdef CONFIG_BCACHE
2360 p->sequential_io = 0;
2361 p->sequential_io_avg = 0;
2362 #endif
2363 #ifdef CONFIG_BPF_SYSCALL
2364 RCU_INIT_POINTER(p->bpf_storage, NULL);
2365 p->bpf_ctx = NULL;
2366 #endif
2367
2368 /* Perform scheduler related setup. Assign this task to a CPU. */
2369 retval = sched_fork(clone_flags, p);
2370 if (retval)
2371 goto bad_fork_cleanup_policy;
2372
2373 retval = perf_event_init_task(p, clone_flags);
2374 if (retval)
2375 goto bad_fork_cleanup_policy;
2376 retval = audit_alloc(p);
2377 if (retval)
2378 goto bad_fork_cleanup_perf;
2379 /* copy all the process information */
2380 shm_init_task(p);
2381 retval = security_task_alloc(p, clone_flags);
2382 if (retval)
2383 goto bad_fork_cleanup_audit;
2384 retval = copy_semundo(clone_flags, p);
2385 if (retval)
2386 goto bad_fork_cleanup_security;
2387 retval = copy_files(clone_flags, p, args->no_files);
2388 if (retval)
2389 goto bad_fork_cleanup_semundo;
2390 retval = copy_fs(clone_flags, p);
2391 if (retval)
2392 goto bad_fork_cleanup_files;
2393 retval = copy_sighand(clone_flags, p);
2394 if (retval)
2395 goto bad_fork_cleanup_fs;
2396 retval = copy_signal(clone_flags, p);
2397 if (retval)
2398 goto bad_fork_cleanup_sighand;
2399 retval = copy_mm(clone_flags, p);
2400 if (retval)
2401 goto bad_fork_cleanup_signal;
2402 retval = copy_namespaces(clone_flags, p);
2403 if (retval)
2404 goto bad_fork_cleanup_mm;
2405 retval = copy_io(clone_flags, p);
2406 if (retval)
2407 goto bad_fork_cleanup_namespaces;
2408 retval = copy_thread(p, args);
2409 if (retval)
2410 goto bad_fork_cleanup_io;
2411
2412 stackleak_task_init(p);
2413
2414 if (pid != &init_struct_pid) {
2415 pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2416 args->set_tid_size);
2417 if (IS_ERR(pid)) {
2418 retval = PTR_ERR(pid);
2419 goto bad_fork_cleanup_thread;
2420 }
2421 }
2422
2423 /*
2424 * This has to happen after we've potentially unshared the file
2425 * descriptor table (so that the pidfd doesn't leak into the child
2426 * if the fd table isn't shared).
2427 */
2428 if (clone_flags & CLONE_PIDFD) {
2429 int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2430
2431 /* Note that no task has been attached to @pid yet. */
2432 retval = __pidfd_prepare(pid, flags, &pidfile);
2433 if (retval < 0)
2434 goto bad_fork_free_pid;
2435 pidfd = retval;
2436
2437 retval = put_user(pidfd, args->pidfd);
2438 if (retval)
2439 goto bad_fork_put_pidfd;
2440 }
2441
2442 #ifdef CONFIG_BLOCK
2443 p->plug = NULL;
2444 #endif
2445 futex_init_task(p);
2446
2447 /*
2448 * sigaltstack should be cleared when sharing the same VM
2449 */
2450 if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2451 sas_ss_reset(p);
2452
2453 /*
2454 * Syscall tracing and stepping should be turned off in the
2455 * child regardless of CLONE_PTRACE.
2456 */
2457 user_disable_single_step(p);
2458 clear_task_syscall_work(p, SYSCALL_TRACE);
2459 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2460 clear_task_syscall_work(p, SYSCALL_EMU);
2461 #endif
2462 clear_tsk_latency_tracing(p);
2463
2464 /* ok, now we should be set up.. */
2465 p->pid = pid_nr(pid);
2466 if (clone_flags & CLONE_THREAD) {
2467 p->group_leader = current->group_leader;
2468 p->tgid = current->tgid;
2469 } else {
2470 p->group_leader = p;
2471 p->tgid = p->pid;
2472 }
2473
2474 p->nr_dirtied = 0;
2475 p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2476 p->dirty_paused_when = 0;
2477
2478 p->pdeath_signal = 0;
2479 p->task_works = NULL;
2480 clear_posix_cputimers_work(p);
2481
2482 #ifdef CONFIG_KRETPROBES
2483 p->kretprobe_instances.first = NULL;
2484 #endif
2485 #ifdef CONFIG_RETHOOK
2486 p->rethooks.first = NULL;
2487 #endif
2488
2489 /*
2490 * Ensure that the cgroup subsystem policies allow the new process to be
2491 * forked. It should be noted that the new process's css_set can be changed
2492 * between here and cgroup_post_fork() if an organisation operation is in
2493 * progress.
2494 */
2495 retval = cgroup_can_fork(p, args);
2496 if (retval)
2497 goto bad_fork_put_pidfd;
2498
2499 /*
2500 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2501 * the new task on the correct runqueue. All this *before* the task
2502 * becomes visible.
2503 *
2504 * This isn't part of ->can_fork() because while the re-cloning is
2505 * cgroup specific, it unconditionally needs to place the task on a
2506 * runqueue.
2507 */
2508 sched_cgroup_fork(p, args);
2509
2510 /*
2511 * From this point on we must avoid any synchronous user-space
2512 * communication until we take the tasklist-lock. In particular, we do
2513 * not want user-space to be able to predict the process start-time by
2514 * stalling fork(2) after we recorded the start_time but before it is
2515 * visible to the system.
2516 */
2517
2518 p->start_time = ktime_get_ns();
2519 p->start_boottime = ktime_get_boottime_ns();
2520
2521 /*
2522 * Make it visible to the rest of the system, but dont wake it up yet.
2523 * Need tasklist lock for parent etc handling!
2524 */
2525 write_lock_irq(&tasklist_lock);
2526
2527 /* CLONE_PARENT re-uses the old parent */
2528 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2529 p->real_parent = current->real_parent;
2530 p->parent_exec_id = current->parent_exec_id;
2531 if (clone_flags & CLONE_THREAD)
2532 p->exit_signal = -1;
2533 else
2534 p->exit_signal = current->group_leader->exit_signal;
2535 } else {
2536 p->real_parent = current;
2537 p->parent_exec_id = current->self_exec_id;
2538 p->exit_signal = args->exit_signal;
2539 }
2540
2541 klp_copy_process(p);
2542
2543 sched_core_fork(p);
2544
2545 spin_lock(¤t->sighand->siglock);
2546
2547 rv_task_fork(p);
2548
2549 rseq_fork(p, clone_flags);
2550
2551 /* Don't start children in a dying pid namespace */
2552 if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2553 retval = -ENOMEM;
2554 goto bad_fork_cancel_cgroup;
2555 }
2556
2557 /* Let kill terminate clone/fork in the middle */
2558 if (fatal_signal_pending(current)) {
2559 retval = -EINTR;
2560 goto bad_fork_cancel_cgroup;
2561 }
2562
2563 /* No more failure paths after this point. */
2564
2565 /*
2566 * Copy seccomp details explicitly here, in case they were changed
2567 * before holding sighand lock.
2568 */
2569 copy_seccomp(p);
2570
2571 init_task_pid_links(p);
2572 if (likely(p->pid)) {
2573 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2574
2575 init_task_pid(p, PIDTYPE_PID, pid);
2576 if (thread_group_leader(p)) {
2577 init_task_pid(p, PIDTYPE_TGID, pid);
2578 init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2579 init_task_pid(p, PIDTYPE_SID, task_session(current));
2580
2581 if (is_child_reaper(pid)) {
2582 ns_of_pid(pid)->child_reaper = p;
2583 p->signal->flags |= SIGNAL_UNKILLABLE;
2584 }
2585 p->signal->shared_pending.signal = delayed.signal;
2586 p->signal->tty = tty_kref_get(current->signal->tty);
2587 /*
2588 * Inherit has_child_subreaper flag under the same
2589 * tasklist_lock with adding child to the process tree
2590 * for propagate_has_child_subreaper optimization.
2591 */
2592 p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2593 p->real_parent->signal->is_child_subreaper;
2594 list_add_tail(&p->sibling, &p->real_parent->children);
2595 list_add_tail_rcu(&p->tasks, &init_task.tasks);
2596 attach_pid(p, PIDTYPE_TGID);
2597 attach_pid(p, PIDTYPE_PGID);
2598 attach_pid(p, PIDTYPE_SID);
2599 __this_cpu_inc(process_counts);
2600 } else {
2601 current->signal->nr_threads++;
2602 current->signal->quick_threads++;
2603 atomic_inc(¤t->signal->live);
2604 refcount_inc(¤t->signal->sigcnt);
2605 task_join_group_stop(p);
2606 list_add_tail_rcu(&p->thread_node,
2607 &p->signal->thread_head);
2608 }
2609 attach_pid(p, PIDTYPE_PID);
2610 nr_threads++;
2611 }
2612 total_forks++;
2613 hlist_del_init(&delayed.node);
2614 spin_unlock(¤t->sighand->siglock);
2615 syscall_tracepoint_update(p);
2616 write_unlock_irq(&tasklist_lock);
2617
2618 if (pidfile)
2619 fd_install(pidfd, pidfile);
2620
2621 proc_fork_connector(p);
2622 sched_post_fork(p);
2623 cgroup_post_fork(p, args);
2624 perf_event_fork(p);
2625
2626 trace_task_newtask(p, clone_flags);
2627 uprobe_copy_process(p, clone_flags);
2628 user_events_fork(p, clone_flags);
2629
2630 copy_oom_score_adj(clone_flags, p);
2631
2632 return p;
2633
2634 bad_fork_cancel_cgroup:
2635 sched_core_free(p);
2636 spin_unlock(¤t->sighand->siglock);
2637 write_unlock_irq(&tasklist_lock);
2638 cgroup_cancel_fork(p, args);
2639 bad_fork_put_pidfd:
2640 if (clone_flags & CLONE_PIDFD) {
2641 fput(pidfile);
2642 put_unused_fd(pidfd);
2643 }
2644 bad_fork_free_pid:
2645 if (pid != &init_struct_pid)
2646 free_pid(pid);
2647 bad_fork_cleanup_thread:
2648 exit_thread(p);
2649 bad_fork_cleanup_io:
2650 if (p->io_context)
2651 exit_io_context(p);
2652 bad_fork_cleanup_namespaces:
2653 exit_task_namespaces(p);
2654 bad_fork_cleanup_mm:
2655 if (p->mm) {
2656 mm_clear_owner(p->mm, p);
2657 mmput(p->mm);
2658 }
2659 bad_fork_cleanup_signal:
2660 if (!(clone_flags & CLONE_THREAD))
2661 free_signal_struct(p->signal);
2662 bad_fork_cleanup_sighand:
2663 __cleanup_sighand(p->sighand);
2664 bad_fork_cleanup_fs:
2665 exit_fs(p); /* blocking */
2666 bad_fork_cleanup_files:
2667 exit_files(p); /* blocking */
2668 bad_fork_cleanup_semundo:
2669 exit_sem(p);
2670 bad_fork_cleanup_security:
2671 security_task_free(p);
2672 bad_fork_cleanup_audit:
2673 audit_free(p);
2674 bad_fork_cleanup_perf:
2675 perf_event_free_task(p);
2676 bad_fork_cleanup_policy:
2677 lockdep_free_task(p);
2678 #ifdef CONFIG_NUMA
2679 mpol_put(p->mempolicy);
2680 #endif
2681 bad_fork_cleanup_delayacct:
2682 delayacct_tsk_free(p);
2683 bad_fork_cleanup_count:
2684 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2685 exit_creds(p);
2686 bad_fork_free:
2687 WRITE_ONCE(p->__state, TASK_DEAD);
2688 exit_task_stack_account(p);
2689 put_task_stack(p);
2690 delayed_free_task(p);
2691 fork_out:
2692 spin_lock_irq(¤t->sighand->siglock);
2693 hlist_del_init(&delayed.node);
2694 spin_unlock_irq(¤t->sighand->siglock);
2695 return ERR_PTR(retval);
2696 }
2697
2698 static inline void init_idle_pids(struct task_struct *idle)
2699 {
2700 enum pid_type type;
2701
2702 for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2703 INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2704 init_task_pid(idle, type, &init_struct_pid);
2705 }
2706 }
2707
2708 static int idle_dummy(void *dummy)
2709 {
2710 /* This function is never called */
2711 return 0;
2712 }
2713
2714 struct task_struct * __init fork_idle(int cpu)
2715 {
2716 struct task_struct *task;
2717 struct kernel_clone_args args = {
2718 .flags = CLONE_VM,
2719 .fn = &idle_dummy,
2720 .fn_arg = NULL,
2721 .kthread = 1,
2722 .idle = 1,
2723 };
2724
2725 task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2726 if (!IS_ERR(task)) {
2727 init_idle_pids(task);
2728 init_idle(task, cpu);
2729 }
2730
2731 return task;
2732 }
2733
2734 /*
2735 * This is like kernel_clone(), but shaved down and tailored to just
2736 * creating io_uring workers. It returns a created task, or an error pointer.
2737 * The returned task is inactive, and the caller must fire it up through
2738 * wake_up_new_task(p). All signals are blocked in the created task.
2739 */
2740 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2741 {
2742 unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2743 CLONE_IO;
2744 struct kernel_clone_args args = {
2745 .flags = ((lower_32_bits(flags) | CLONE_VM |
2746 CLONE_UNTRACED) & ~CSIGNAL),
2747 .exit_signal = (lower_32_bits(flags) & CSIGNAL),
2748 .fn = fn,
2749 .fn_arg = arg,
2750 .io_thread = 1,
2751 .user_worker = 1,
2752 };
2753
2754 return copy_process(NULL, 0, node, &args);
2755 }
2756
2757 /*
2758 * Ok, this is the main fork-routine.
2759 *
2760 * It copies the process, and if successful kick-starts
2761 * it and waits for it to finish using the VM if required.
2762 *
2763 * args->exit_signal is expected to be checked for sanity by the caller.
2764 */
> 2765 #include <linux/livepatch.h>
2766 pid_t kernel_clone(struct kernel_clone_args *args)
2767 {
2768 u64 clone_flags = args->flags;
2769 struct completion vfork;
2770 struct pid *pid;
2771 struct task_struct *p;
2772 int trace = 0;
2773 pid_t nr;
2774 int *newpid;
2775 static int ctr = 0;
2776
2777 /*
2778 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2779 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2780 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2781 * field in struct clone_args and it still doesn't make sense to have
2782 * them both point at the same memory location. Performing this check
2783 * here has the advantage that we don't need to have a separate helper
2784 * to check for legacy clone().
2785 */
2786 if ((clone_flags & CLONE_PIDFD) &&
2787 (clone_flags & CLONE_PARENT_SETTID) &&
2788 (args->pidfd == args->parent_tid))
2789 return -EINVAL;
2790
2791 /*
2792 * Determine whether and which event to report to ptracer. When
2793 * called from kernel_thread or CLONE_UNTRACED is explicitly
2794 * requested, no event is reported; otherwise, report if the event
2795 * for the type of forking is enabled.
2796 */
2797 if (!(clone_flags & CLONE_UNTRACED)) {
2798 if (clone_flags & CLONE_VFORK)
2799 trace = PTRACE_EVENT_VFORK;
2800 else if (args->exit_signal != SIGCHLD)
2801 trace = PTRACE_EVENT_CLONE;
2802 else
2803 trace = PTRACE_EVENT_FORK;
2804
2805 if (likely(!ptrace_event_enabled(current, trace)))
2806 trace = 0;
2807 }
2808
2809 p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2810 add_latent_entropy();
2811
2812 if (IS_ERR(p))
2813 return PTR_ERR(p);
2814
2815 newpid = klp_shadow_get_or_alloc(p, 0, sizeof(*newpid), GFP_KERNEL,
2816 NULL, NULL);
2817 if (newpid)
2818 *newpid = ctr++;
2819
2820 /*
2821 * Do this prior waking up the new thread - the thread pointer
2822 * might get invalid after that point, if the thread exits quickly.
2823 */
2824 trace_sched_process_fork(current, p);
2825
2826 pid = get_task_pid(p, PIDTYPE_PID);
2827 nr = pid_vnr(pid);
2828
2829 if (clone_flags & CLONE_PARENT_SETTID)
2830 put_user(nr, args->parent_tid);
2831
2832 if (clone_flags & CLONE_VFORK) {
2833 p->vfork_done = &vfork;
2834 init_completion(&vfork);
2835 get_task_struct(p);
2836 }
2837
2838 if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2839 /* lock the task to synchronize with memcg migration */
2840 task_lock(p);
2841 lru_gen_add_mm(p->mm);
2842 task_unlock(p);
2843 }
2844
2845 wake_up_new_task(p);
2846
2847 /* forking complete and child started to run, tell ptracer */
2848 if (unlikely(trace))
2849 ptrace_event_pid(trace, pid);
2850
2851 if (clone_flags & CLONE_VFORK) {
2852 if (!wait_for_vfork_done(p, &vfork))
2853 ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2854 }
2855
2856 put_pid(pid);
2857 return nr;
2858 }
2859
--
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-05-27 1:48 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2024-05-25 6:40 [jpoimboe:objtool-diff 2/2] kernel/fork.c: linux/livepatch.h is included more than once kernel test robot
2024-05-27 1:47 ` Liu, Yujie
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.