xref: /linux/kernel/fork.c (revision 2c1ed907520c50326b8f604907a8478b27881a2e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/fork.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 /*
9  *  'fork.c' contains the help-routines for the 'fork' system call
10  * (see also entry.S and others).
11  * Fork is rather simple, once you get the hang of it, but the memory
12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13  */
14 
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/user.h>
20 #include <linux/sched/numa_balancing.h>
21 #include <linux/sched/stat.h>
22 #include <linux/sched/task.h>
23 #include <linux/sched/task_stack.h>
24 #include <linux/sched/cputime.h>
25 #include <linux/sched/ext.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/kmsan.h>
41 #include <linux/binfmts.h>
42 #include <linux/mman.h>
43 #include <linux/mmu_notifier.h>
44 #include <linux/fs.h>
45 #include <linux/mm.h>
46 #include <linux/mm_inline.h>
47 #include <linux/memblock.h>
48 #include <linux/nsproxy.h>
49 #include <linux/capability.h>
50 #include <linux/cpu.h>
51 #include <linux/cgroup.h>
52 #include <linux/security.h>
53 #include <linux/hugetlb.h>
54 #include <linux/seccomp.h>
55 #include <linux/swap.h>
56 #include <linux/syscalls.h>
57 #include <linux/syscall_user_dispatch.h>
58 #include <linux/jiffies.h>
59 #include <linux/futex.h>
60 #include <linux/compat.h>
61 #include <linux/kthread.h>
62 #include <linux/task_io_accounting_ops.h>
63 #include <linux/rcupdate.h>
64 #include <linux/ptrace.h>
65 #include <linux/mount.h>
66 #include <linux/audit.h>
67 #include <linux/memcontrol.h>
68 #include <linux/ftrace.h>
69 #include <linux/proc_fs.h>
70 #include <linux/profile.h>
71 #include <linux/rmap.h>
72 #include <linux/ksm.h>
73 #include <linux/acct.h>
74 #include <linux/userfaultfd_k.h>
75 #include <linux/tsacct_kern.h>
76 #include <linux/cn_proc.h>
77 #include <linux/freezer.h>
78 #include <linux/delayacct.h>
79 #include <linux/taskstats_kern.h>
80 #include <linux/tty.h>
81 #include <linux/fs_struct.h>
82 #include <linux/magic.h>
83 #include <linux/perf_event.h>
84 #include <linux/posix-timers.h>
85 #include <linux/user-return-notifier.h>
86 #include <linux/oom.h>
87 #include <linux/khugepaged.h>
88 #include <linux/signalfd.h>
89 #include <linux/uprobes.h>
90 #include <linux/aio.h>
91 #include <linux/compiler.h>
92 #include <linux/sysctl.h>
93 #include <linux/kcov.h>
94 #include <linux/livepatch.h>
95 #include <linux/thread_info.h>
96 #include <linux/stackleak.h>
97 #include <linux/kasan.h>
98 #include <linux/scs.h>
99 #include <linux/io_uring.h>
100 #include <linux/bpf.h>
101 #include <linux/stackprotector.h>
102 #include <linux/user_events.h>
103 #include <linux/iommu.h>
104 #include <linux/rseq.h>
105 #include <uapi/linux/pidfd.h>
106 #include <linux/pidfs.h>
107 #include <linux/tick.h>
108 
109 #include <asm/pgalloc.h>
110 #include <linux/uaccess.h>
111 #include <asm/mmu_context.h>
112 #include <asm/cacheflush.h>
113 #include <asm/tlbflush.h>
114 
115 #include <trace/events/sched.h>
116 
117 #define CREATE_TRACE_POINTS
118 #include <trace/events/task.h>
119 
120 #include <kunit/visibility.h>
121 
122 /*
123  * Minimum number of threads to boot the kernel
124  */
125 #define MIN_THREADS 20
126 
127 /*
128  * Maximum number of threads
129  */
130 #define MAX_THREADS FUTEX_TID_MASK
131 
132 /*
133  * Protected counters by write_lock_irq(&tasklist_lock)
134  */
135 unsigned long total_forks;	/* Handle normal Linux uptimes. */
136 int nr_threads;			/* The idle threads do not count.. */
137 
138 static int max_threads;		/* tunable limit on nr_threads */
139 
140 #define NAMED_ARRAY_INDEX(x)	[x] = __stringify(x)
141 
142 static const char * const resident_page_types[] = {
143 	NAMED_ARRAY_INDEX(MM_FILEPAGES),
144 	NAMED_ARRAY_INDEX(MM_ANONPAGES),
145 	NAMED_ARRAY_INDEX(MM_SWAPENTS),
146 	NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
147 };
148 
149 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
150 
151 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
152 
153 #ifdef CONFIG_PROVE_RCU
lockdep_tasklist_lock_is_held(void)154 int lockdep_tasklist_lock_is_held(void)
155 {
156 	return lockdep_is_held(&tasklist_lock);
157 }
158 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
159 #endif /* #ifdef CONFIG_PROVE_RCU */
160 
nr_processes(void)161 int nr_processes(void)
162 {
163 	int cpu;
164 	int total = 0;
165 
166 	for_each_possible_cpu(cpu)
167 		total += per_cpu(process_counts, cpu);
168 
169 	return total;
170 }
171 
arch_release_task_struct(struct task_struct * tsk)172 void __weak arch_release_task_struct(struct task_struct *tsk)
173 {
174 }
175 
176 static struct kmem_cache *task_struct_cachep;
177 
alloc_task_struct_node(int node)178 static inline struct task_struct *alloc_task_struct_node(int node)
179 {
180 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
181 }
182 
free_task_struct(struct task_struct * tsk)183 static inline void free_task_struct(struct task_struct *tsk)
184 {
185 	kmem_cache_free(task_struct_cachep, tsk);
186 }
187 
188 /*
189  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
190  * kmemcache based allocator.
191  */
192 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
193 
194 #  ifdef CONFIG_VMAP_STACK
195 /*
196  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
197  * flush.  Try to minimize the number of calls by caching stacks.
198  */
199 #define NR_CACHED_STACKS 2
200 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
201 
202 struct vm_stack {
203 	struct rcu_head rcu;
204 	struct vm_struct *stack_vm_area;
205 };
206 
try_release_thread_stack_to_cache(struct vm_struct * vm)207 static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
208 {
209 	unsigned int i;
210 
211 	for (i = 0; i < NR_CACHED_STACKS; i++) {
212 		struct vm_struct *tmp = NULL;
213 
214 		if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm))
215 			return true;
216 	}
217 	return false;
218 }
219 
thread_stack_free_rcu(struct rcu_head * rh)220 static void thread_stack_free_rcu(struct rcu_head *rh)
221 {
222 	struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
223 
224 	if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
225 		return;
226 
227 	vfree(vm_stack);
228 }
229 
thread_stack_delayed_free(struct task_struct * tsk)230 static void thread_stack_delayed_free(struct task_struct *tsk)
231 {
232 	struct vm_stack *vm_stack = tsk->stack;
233 
234 	vm_stack->stack_vm_area = tsk->stack_vm_area;
235 	call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
236 }
237 
free_vm_stack_cache(unsigned int cpu)238 static int free_vm_stack_cache(unsigned int cpu)
239 {
240 	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
241 	int i;
242 
243 	for (i = 0; i < NR_CACHED_STACKS; i++) {
244 		struct vm_struct *vm_stack = cached_vm_stacks[i];
245 
246 		if (!vm_stack)
247 			continue;
248 
249 		vfree(vm_stack->addr);
250 		cached_vm_stacks[i] = NULL;
251 	}
252 
253 	return 0;
254 }
255 
memcg_charge_kernel_stack(struct vm_struct * vm)256 static int memcg_charge_kernel_stack(struct vm_struct *vm)
257 {
258 	int i;
259 	int ret;
260 	int nr_charged = 0;
261 
262 	BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
263 
264 	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
265 		ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL, 0);
266 		if (ret)
267 			goto err;
268 		nr_charged++;
269 	}
270 	return 0;
271 err:
272 	for (i = 0; i < nr_charged; i++)
273 		memcg_kmem_uncharge_page(vm->pages[i], 0);
274 	return ret;
275 }
276 
alloc_thread_stack_node(struct task_struct * tsk,int node)277 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
278 {
279 	struct vm_struct *vm;
280 	void *stack;
281 	int i;
282 
283 	for (i = 0; i < NR_CACHED_STACKS; i++) {
284 		struct vm_struct *s;
285 
286 		s = this_cpu_xchg(cached_stacks[i], NULL);
287 
288 		if (!s)
289 			continue;
290 
291 		/* Reset stack metadata. */
292 		kasan_unpoison_range(s->addr, THREAD_SIZE);
293 
294 		stack = kasan_reset_tag(s->addr);
295 
296 		/* Clear stale pointers from reused stack. */
297 		memset(stack, 0, THREAD_SIZE);
298 
299 		if (memcg_charge_kernel_stack(s)) {
300 			vfree(s->addr);
301 			return -ENOMEM;
302 		}
303 
304 		tsk->stack_vm_area = s;
305 		tsk->stack = stack;
306 		return 0;
307 	}
308 
309 	/*
310 	 * Allocated stacks are cached and later reused by new threads,
311 	 * so memcg accounting is performed manually on assigning/releasing
312 	 * stacks to tasks. Drop __GFP_ACCOUNT.
313 	 */
314 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
315 				     VMALLOC_START, VMALLOC_END,
316 				     THREADINFO_GFP & ~__GFP_ACCOUNT,
317 				     PAGE_KERNEL,
318 				     0, node, __builtin_return_address(0));
319 	if (!stack)
320 		return -ENOMEM;
321 
322 	vm = find_vm_area(stack);
323 	if (memcg_charge_kernel_stack(vm)) {
324 		vfree(stack);
325 		return -ENOMEM;
326 	}
327 	/*
328 	 * We can't call find_vm_area() in interrupt context, and
329 	 * free_thread_stack() can be called in interrupt context,
330 	 * so cache the vm_struct.
331 	 */
332 	tsk->stack_vm_area = vm;
333 	stack = kasan_reset_tag(stack);
334 	tsk->stack = stack;
335 	return 0;
336 }
337 
free_thread_stack(struct task_struct * tsk)338 static void free_thread_stack(struct task_struct *tsk)
339 {
340 	if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
341 		thread_stack_delayed_free(tsk);
342 
343 	tsk->stack = NULL;
344 	tsk->stack_vm_area = NULL;
345 }
346 
347 #  else /* !CONFIG_VMAP_STACK */
348 
thread_stack_free_rcu(struct rcu_head * rh)349 static void thread_stack_free_rcu(struct rcu_head *rh)
350 {
351 	__free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
352 }
353 
thread_stack_delayed_free(struct task_struct * tsk)354 static void thread_stack_delayed_free(struct task_struct *tsk)
355 {
356 	struct rcu_head *rh = tsk->stack;
357 
358 	call_rcu(rh, thread_stack_free_rcu);
359 }
360 
alloc_thread_stack_node(struct task_struct * tsk,int node)361 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
362 {
363 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
364 					     THREAD_SIZE_ORDER);
365 
366 	if (likely(page)) {
367 		tsk->stack = kasan_reset_tag(page_address(page));
368 		return 0;
369 	}
370 	return -ENOMEM;
371 }
372 
free_thread_stack(struct task_struct * tsk)373 static void free_thread_stack(struct task_struct *tsk)
374 {
375 	thread_stack_delayed_free(tsk);
376 	tsk->stack = NULL;
377 }
378 
379 #  endif /* CONFIG_VMAP_STACK */
380 # else /* !(THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)) */
381 
382 static struct kmem_cache *thread_stack_cache;
383 
thread_stack_free_rcu(struct rcu_head * rh)384 static void thread_stack_free_rcu(struct rcu_head *rh)
385 {
386 	kmem_cache_free(thread_stack_cache, rh);
387 }
388 
thread_stack_delayed_free(struct task_struct * tsk)389 static void thread_stack_delayed_free(struct task_struct *tsk)
390 {
391 	struct rcu_head *rh = tsk->stack;
392 
393 	call_rcu(rh, thread_stack_free_rcu);
394 }
395 
alloc_thread_stack_node(struct task_struct * tsk,int node)396 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
397 {
398 	unsigned long *stack;
399 	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
400 	stack = kasan_reset_tag(stack);
401 	tsk->stack = stack;
402 	return stack ? 0 : -ENOMEM;
403 }
404 
free_thread_stack(struct task_struct * tsk)405 static void free_thread_stack(struct task_struct *tsk)
406 {
407 	thread_stack_delayed_free(tsk);
408 	tsk->stack = NULL;
409 }
410 
thread_stack_cache_init(void)411 void thread_stack_cache_init(void)
412 {
413 	thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
414 					THREAD_SIZE, THREAD_SIZE, 0, 0,
415 					THREAD_SIZE, NULL);
416 	BUG_ON(thread_stack_cache == NULL);
417 }
418 
419 # endif /* THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK) */
420 
421 /* SLAB cache for signal_struct structures (tsk->signal) */
422 static struct kmem_cache *signal_cachep;
423 
424 /* SLAB cache for sighand_struct structures (tsk->sighand) */
425 struct kmem_cache *sighand_cachep;
426 
427 /* SLAB cache for files_struct structures (tsk->files) */
428 struct kmem_cache *files_cachep;
429 
430 /* SLAB cache for fs_struct structures (tsk->fs) */
431 struct kmem_cache *fs_cachep;
432 
433 /* SLAB cache for vm_area_struct structures */
434 static struct kmem_cache *vm_area_cachep;
435 
436 /* SLAB cache for mm_struct structures (tsk->mm) */
437 static struct kmem_cache *mm_cachep;
438 
439 #ifdef CONFIG_PER_VMA_LOCK
440 
441 /* SLAB cache for vm_area_struct.lock */
442 static struct kmem_cache *vma_lock_cachep;
443 
vma_lock_alloc(struct vm_area_struct * vma)444 static bool vma_lock_alloc(struct vm_area_struct *vma)
445 {
446 	vma->vm_lock = kmem_cache_alloc(vma_lock_cachep, GFP_KERNEL);
447 	if (!vma->vm_lock)
448 		return false;
449 
450 	init_rwsem(&vma->vm_lock->lock);
451 	vma->vm_lock_seq = UINT_MAX;
452 
453 	return true;
454 }
455 
vma_lock_free(struct vm_area_struct * vma)456 static inline void vma_lock_free(struct vm_area_struct *vma)
457 {
458 	kmem_cache_free(vma_lock_cachep, vma->vm_lock);
459 }
460 
461 #else /* CONFIG_PER_VMA_LOCK */
462 
vma_lock_alloc(struct vm_area_struct * vma)463 static inline bool vma_lock_alloc(struct vm_area_struct *vma) { return true; }
vma_lock_free(struct vm_area_struct * vma)464 static inline void vma_lock_free(struct vm_area_struct *vma) {}
465 
466 #endif /* CONFIG_PER_VMA_LOCK */
467 
vm_area_alloc(struct mm_struct * mm)468 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
469 {
470 	struct vm_area_struct *vma;
471 
472 	vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
473 	if (!vma)
474 		return NULL;
475 
476 	vma_init(vma, mm);
477 	if (!vma_lock_alloc(vma)) {
478 		kmem_cache_free(vm_area_cachep, vma);
479 		return NULL;
480 	}
481 
482 	return vma;
483 }
484 
vm_area_dup(struct vm_area_struct * orig)485 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
486 {
487 	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
488 
489 	if (!new)
490 		return NULL;
491 
492 	ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
493 	ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
494 	/*
495 	 * orig->shared.rb may be modified concurrently, but the clone
496 	 * will be reinitialized.
497 	 */
498 	data_race(memcpy(new, orig, sizeof(*new)));
499 	if (!vma_lock_alloc(new)) {
500 		kmem_cache_free(vm_area_cachep, new);
501 		return NULL;
502 	}
503 	INIT_LIST_HEAD(&new->anon_vma_chain);
504 	vma_numab_state_init(new);
505 	dup_anon_vma_name(orig, new);
506 
507 	return new;
508 }
509 
__vm_area_free(struct vm_area_struct * vma)510 void __vm_area_free(struct vm_area_struct *vma)
511 {
512 	vma_numab_state_free(vma);
513 	free_anon_vma_name(vma);
514 	vma_lock_free(vma);
515 	kmem_cache_free(vm_area_cachep, vma);
516 }
517 
518 #ifdef CONFIG_PER_VMA_LOCK
vm_area_free_rcu_cb(struct rcu_head * head)519 static void vm_area_free_rcu_cb(struct rcu_head *head)
520 {
521 	struct vm_area_struct *vma = container_of(head, struct vm_area_struct,
522 						  vm_rcu);
523 
524 	/* The vma should not be locked while being destroyed. */
525 	VM_BUG_ON_VMA(rwsem_is_locked(&vma->vm_lock->lock), vma);
526 	__vm_area_free(vma);
527 }
528 #endif
529 
vm_area_free(struct vm_area_struct * vma)530 void vm_area_free(struct vm_area_struct *vma)
531 {
532 #ifdef CONFIG_PER_VMA_LOCK
533 	call_rcu(&vma->vm_rcu, vm_area_free_rcu_cb);
534 #else
535 	__vm_area_free(vma);
536 #endif
537 }
538 
account_kernel_stack(struct task_struct * tsk,int account)539 static void account_kernel_stack(struct task_struct *tsk, int account)
540 {
541 	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
542 		struct vm_struct *vm = task_stack_vm_area(tsk);
543 		int i;
544 
545 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
546 			mod_lruvec_page_state(vm->pages[i], NR_KERNEL_STACK_KB,
547 					      account * (PAGE_SIZE / 1024));
548 	} else {
549 		void *stack = task_stack_page(tsk);
550 
551 		/* All stack pages are in the same node. */
552 		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
553 				      account * (THREAD_SIZE / 1024));
554 	}
555 }
556 
exit_task_stack_account(struct task_struct * tsk)557 void exit_task_stack_account(struct task_struct *tsk)
558 {
559 	account_kernel_stack(tsk, -1);
560 
561 	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
562 		struct vm_struct *vm;
563 		int i;
564 
565 		vm = task_stack_vm_area(tsk);
566 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
567 			memcg_kmem_uncharge_page(vm->pages[i], 0);
568 	}
569 }
570 
release_task_stack(struct task_struct * tsk)571 static void release_task_stack(struct task_struct *tsk)
572 {
573 	if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
574 		return;  /* Better to leak the stack than to free prematurely */
575 
576 	free_thread_stack(tsk);
577 }
578 
579 #ifdef CONFIG_THREAD_INFO_IN_TASK
put_task_stack(struct task_struct * tsk)580 void put_task_stack(struct task_struct *tsk)
581 {
582 	if (refcount_dec_and_test(&tsk->stack_refcount))
583 		release_task_stack(tsk);
584 }
585 #endif
586 
free_task(struct task_struct * tsk)587 void free_task(struct task_struct *tsk)
588 {
589 #ifdef CONFIG_SECCOMP
590 	WARN_ON_ONCE(tsk->seccomp.filter);
591 #endif
592 	release_user_cpus_ptr(tsk);
593 	scs_release(tsk);
594 
595 #ifndef CONFIG_THREAD_INFO_IN_TASK
596 	/*
597 	 * The task is finally done with both the stack and thread_info,
598 	 * so free both.
599 	 */
600 	release_task_stack(tsk);
601 #else
602 	/*
603 	 * If the task had a separate stack allocation, it should be gone
604 	 * by now.
605 	 */
606 	WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
607 #endif
608 	rt_mutex_debug_task_free(tsk);
609 	ftrace_graph_exit_task(tsk);
610 	arch_release_task_struct(tsk);
611 	if (tsk->flags & PF_KTHREAD)
612 		free_kthread_struct(tsk);
613 	bpf_task_storage_free(tsk);
614 	free_task_struct(tsk);
615 }
616 EXPORT_SYMBOL(free_task);
617 
dup_mm_exe_file(struct mm_struct * mm,struct mm_struct * oldmm)618 static void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
619 {
620 	struct file *exe_file;
621 
622 	exe_file = get_mm_exe_file(oldmm);
623 	RCU_INIT_POINTER(mm->exe_file, exe_file);
624 	/*
625 	 * We depend on the oldmm having properly denied write access to the
626 	 * exe_file already.
627 	 */
628 	if (exe_file && exe_file_deny_write_access(exe_file))
629 		pr_warn_once("exe_file_deny_write_access() failed in %s\n", __func__);
630 }
631 
632 #ifdef CONFIG_MMU
dup_mmap(struct mm_struct * mm,struct mm_struct * oldmm)633 static __latent_entropy int dup_mmap(struct mm_struct *mm,
634 					struct mm_struct *oldmm)
635 {
636 	struct vm_area_struct *mpnt, *tmp;
637 	int retval;
638 	unsigned long charge = 0;
639 	LIST_HEAD(uf);
640 	VMA_ITERATOR(vmi, mm, 0);
641 
642 	if (mmap_write_lock_killable(oldmm))
643 		return -EINTR;
644 	flush_cache_dup_mm(oldmm);
645 	uprobe_dup_mmap(oldmm, mm);
646 	/*
647 	 * Not linked in yet - no deadlock potential:
648 	 */
649 	mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
650 
651 	/* No ordering required: file already has been exposed. */
652 	dup_mm_exe_file(mm, oldmm);
653 
654 	mm->total_vm = oldmm->total_vm;
655 	mm->data_vm = oldmm->data_vm;
656 	mm->exec_vm = oldmm->exec_vm;
657 	mm->stack_vm = oldmm->stack_vm;
658 
659 	/* Use __mt_dup() to efficiently build an identical maple tree. */
660 	retval = __mt_dup(&oldmm->mm_mt, &mm->mm_mt, GFP_KERNEL);
661 	if (unlikely(retval))
662 		goto out;
663 
664 	mt_clear_in_rcu(vmi.mas.tree);
665 	for_each_vma(vmi, mpnt) {
666 		struct file *file;
667 
668 		vma_start_write(mpnt);
669 		if (mpnt->vm_flags & VM_DONTCOPY) {
670 			retval = vma_iter_clear_gfp(&vmi, mpnt->vm_start,
671 						    mpnt->vm_end, GFP_KERNEL);
672 			if (retval)
673 				goto loop_out;
674 
675 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
676 			continue;
677 		}
678 		charge = 0;
679 		/*
680 		 * Don't duplicate many vmas if we've been oom-killed (for
681 		 * example)
682 		 */
683 		if (fatal_signal_pending(current)) {
684 			retval = -EINTR;
685 			goto loop_out;
686 		}
687 		if (mpnt->vm_flags & VM_ACCOUNT) {
688 			unsigned long len = vma_pages(mpnt);
689 
690 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
691 				goto fail_nomem;
692 			charge = len;
693 		}
694 		tmp = vm_area_dup(mpnt);
695 		if (!tmp)
696 			goto fail_nomem;
697 		retval = vma_dup_policy(mpnt, tmp);
698 		if (retval)
699 			goto fail_nomem_policy;
700 		tmp->vm_mm = mm;
701 		retval = dup_userfaultfd(tmp, &uf);
702 		if (retval)
703 			goto fail_nomem_anon_vma_fork;
704 		if (tmp->vm_flags & VM_WIPEONFORK) {
705 			/*
706 			 * VM_WIPEONFORK gets a clean slate in the child.
707 			 * Don't prepare anon_vma until fault since we don't
708 			 * copy page for current vma.
709 			 */
710 			tmp->anon_vma = NULL;
711 		} else if (anon_vma_fork(tmp, mpnt))
712 			goto fail_nomem_anon_vma_fork;
713 		vm_flags_clear(tmp, VM_LOCKED_MASK);
714 		/*
715 		 * Copy/update hugetlb private vma information.
716 		 */
717 		if (is_vm_hugetlb_page(tmp))
718 			hugetlb_dup_vma_private(tmp);
719 
720 		/*
721 		 * Link the vma into the MT. After using __mt_dup(), memory
722 		 * allocation is not necessary here, so it cannot fail.
723 		 */
724 		vma_iter_bulk_store(&vmi, tmp);
725 
726 		mm->map_count++;
727 
728 		if (tmp->vm_ops && tmp->vm_ops->open)
729 			tmp->vm_ops->open(tmp);
730 
731 		file = tmp->vm_file;
732 		if (file) {
733 			struct address_space *mapping = file->f_mapping;
734 
735 			get_file(file);
736 			i_mmap_lock_write(mapping);
737 			if (vma_is_shared_maywrite(tmp))
738 				mapping_allow_writable(mapping);
739 			flush_dcache_mmap_lock(mapping);
740 			/* insert tmp into the share list, just after mpnt */
741 			vma_interval_tree_insert_after(tmp, mpnt,
742 					&mapping->i_mmap);
743 			flush_dcache_mmap_unlock(mapping);
744 			i_mmap_unlock_write(mapping);
745 		}
746 
747 		if (!(tmp->vm_flags & VM_WIPEONFORK))
748 			retval = copy_page_range(tmp, mpnt);
749 
750 		if (retval) {
751 			mpnt = vma_next(&vmi);
752 			goto loop_out;
753 		}
754 	}
755 	/* a new mm has just been created */
756 	retval = arch_dup_mmap(oldmm, mm);
757 loop_out:
758 	vma_iter_free(&vmi);
759 	if (!retval) {
760 		mt_set_in_rcu(vmi.mas.tree);
761 		ksm_fork(mm, oldmm);
762 		khugepaged_fork(mm, oldmm);
763 	} else {
764 
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 		if (mpnt) {
773 			mas_set_range(&vmi.mas, mpnt->vm_start, mpnt->vm_end - 1);
774 			mas_store(&vmi.mas, XA_ZERO_ENTRY);
775 			/* Avoid OOM iterating a broken tree */
776 			set_bit(MMF_OOM_SKIP, &mm->flags);
777 		}
778 		/*
779 		 * The mm_struct is going to exit, but the locks will be dropped
780 		 * first.  Set the mm_struct as unstable is advisable as it is
781 		 * not fully initialised.
782 		 */
783 		set_bit(MMF_UNSTABLE, &mm->flags);
784 	}
785 out:
786 	mmap_write_unlock(mm);
787 	flush_tlb_mm(oldmm);
788 	mmap_write_unlock(oldmm);
789 	if (!retval)
790 		dup_userfaultfd_complete(&uf);
791 	else
792 		dup_userfaultfd_fail(&uf);
793 	return retval;
794 
795 fail_nomem_anon_vma_fork:
796 	mpol_put(vma_policy(tmp));
797 fail_nomem_policy:
798 	vm_area_free(tmp);
799 fail_nomem:
800 	retval = -ENOMEM;
801 	vm_unacct_memory(charge);
802 	goto loop_out;
803 }
804 
mm_alloc_pgd(struct mm_struct * mm)805 static inline int mm_alloc_pgd(struct mm_struct *mm)
806 {
807 	mm->pgd = pgd_alloc(mm);
808 	if (unlikely(!mm->pgd))
809 		return -ENOMEM;
810 	return 0;
811 }
812 
mm_free_pgd(struct mm_struct * mm)813 static inline void mm_free_pgd(struct mm_struct *mm)
814 {
815 	pgd_free(mm, mm->pgd);
816 }
817 #else
dup_mmap(struct mm_struct * mm,struct mm_struct * oldmm)818 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
819 {
820 	mmap_write_lock(oldmm);
821 	dup_mm_exe_file(mm, oldmm);
822 	mmap_write_unlock(oldmm);
823 	return 0;
824 }
825 #define mm_alloc_pgd(mm)	(0)
826 #define mm_free_pgd(mm)
827 #endif /* CONFIG_MMU */
828 
check_mm(struct mm_struct * mm)829 static void check_mm(struct mm_struct *mm)
830 {
831 	int i;
832 
833 	BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
834 			 "Please make sure 'struct resident_page_types[]' is updated as well");
835 
836 	for (i = 0; i < NR_MM_COUNTERS; i++) {
837 		long x = percpu_counter_sum(&mm->rss_stat[i]);
838 
839 		if (unlikely(x))
840 			pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
841 				 mm, resident_page_types[i], x);
842 	}
843 
844 	if (mm_pgtables_bytes(mm))
845 		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
846 				mm_pgtables_bytes(mm));
847 
848 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
849 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
850 #endif
851 }
852 
853 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
854 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
855 
do_check_lazy_tlb(void * arg)856 static void do_check_lazy_tlb(void *arg)
857 {
858 	struct mm_struct *mm = arg;
859 
860 	WARN_ON_ONCE(current->active_mm == mm);
861 }
862 
do_shoot_lazy_tlb(void * arg)863 static void do_shoot_lazy_tlb(void *arg)
864 {
865 	struct mm_struct *mm = arg;
866 
867 	if (current->active_mm == mm) {
868 		WARN_ON_ONCE(current->mm);
869 		current->active_mm = &init_mm;
870 		switch_mm(mm, &init_mm, current);
871 	}
872 }
873 
cleanup_lazy_tlbs(struct mm_struct * mm)874 static void cleanup_lazy_tlbs(struct mm_struct *mm)
875 {
876 	if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
877 		/*
878 		 * In this case, lazy tlb mms are refounted and would not reach
879 		 * __mmdrop until all CPUs have switched away and mmdrop()ed.
880 		 */
881 		return;
882 	}
883 
884 	/*
885 	 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
886 	 * requires lazy mm users to switch to another mm when the refcount
887 	 * drops to zero, before the mm is freed. This requires IPIs here to
888 	 * switch kernel threads to init_mm.
889 	 *
890 	 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
891 	 * switch with the final userspace teardown TLB flush which leaves the
892 	 * mm lazy on this CPU but no others, reducing the need for additional
893 	 * IPIs here. There are cases where a final IPI is still required here,
894 	 * such as the final mmdrop being performed on a different CPU than the
895 	 * one exiting, or kernel threads using the mm when userspace exits.
896 	 *
897 	 * IPI overheads have not found to be expensive, but they could be
898 	 * reduced in a number of possible ways, for example (roughly
899 	 * increasing order of complexity):
900 	 * - The last lazy reference created by exit_mm() could instead switch
901 	 *   to init_mm, however it's probable this will run on the same CPU
902 	 *   immediately afterwards, so this may not reduce IPIs much.
903 	 * - A batch of mms requiring IPIs could be gathered and freed at once.
904 	 * - CPUs store active_mm where it can be remotely checked without a
905 	 *   lock, to filter out false-positives in the cpumask.
906 	 * - After mm_users or mm_count reaches zero, switching away from the
907 	 *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
908 	 *   with some batching or delaying of the final IPIs.
909 	 * - A delayed freeing and RCU-like quiescing sequence based on mm
910 	 *   switching to avoid IPIs completely.
911 	 */
912 	on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
913 	if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
914 		on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
915 }
916 
917 /*
918  * Called when the last reference to the mm
919  * is dropped: either by a lazy thread or by
920  * mmput. Free the page directory and the mm.
921  */
__mmdrop(struct mm_struct * mm)922 void __mmdrop(struct mm_struct *mm)
923 {
924 	BUG_ON(mm == &init_mm);
925 	WARN_ON_ONCE(mm == current->mm);
926 
927 	/* Ensure no CPUs are using this as their lazy tlb mm */
928 	cleanup_lazy_tlbs(mm);
929 
930 	WARN_ON_ONCE(mm == current->active_mm);
931 	mm_free_pgd(mm);
932 	destroy_context(mm);
933 	mmu_notifier_subscriptions_destroy(mm);
934 	check_mm(mm);
935 	put_user_ns(mm->user_ns);
936 	mm_pasid_drop(mm);
937 	mm_destroy_cid(mm);
938 	percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
939 
940 	free_mm(mm);
941 }
942 EXPORT_SYMBOL_GPL(__mmdrop);
943 
mmdrop_async_fn(struct work_struct * work)944 static void mmdrop_async_fn(struct work_struct *work)
945 {
946 	struct mm_struct *mm;
947 
948 	mm = container_of(work, struct mm_struct, async_put_work);
949 	__mmdrop(mm);
950 }
951 
mmdrop_async(struct mm_struct * mm)952 static void mmdrop_async(struct mm_struct *mm)
953 {
954 	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
955 		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
956 		schedule_work(&mm->async_put_work);
957 	}
958 }
959 
free_signal_struct(struct signal_struct * sig)960 static inline void free_signal_struct(struct signal_struct *sig)
961 {
962 	taskstats_tgid_free(sig);
963 	sched_autogroup_exit(sig);
964 	/*
965 	 * __mmdrop is not safe to call from softirq context on x86 due to
966 	 * pgd_dtor so postpone it to the async context
967 	 */
968 	if (sig->oom_mm)
969 		mmdrop_async(sig->oom_mm);
970 	kmem_cache_free(signal_cachep, sig);
971 }
972 
put_signal_struct(struct signal_struct * sig)973 static inline void put_signal_struct(struct signal_struct *sig)
974 {
975 	if (refcount_dec_and_test(&sig->sigcnt))
976 		free_signal_struct(sig);
977 }
978 
__put_task_struct(struct task_struct * tsk)979 void __put_task_struct(struct task_struct *tsk)
980 {
981 	WARN_ON(!tsk->exit_state);
982 	WARN_ON(refcount_read(&tsk->usage));
983 	WARN_ON(tsk == current);
984 
985 	sched_ext_free(tsk);
986 	io_uring_free(tsk);
987 	cgroup_free(tsk);
988 	task_numa_free(tsk, true);
989 	security_task_free(tsk);
990 	exit_creds(tsk);
991 	delayacct_tsk_free(tsk);
992 	put_signal_struct(tsk->signal);
993 	sched_core_free(tsk);
994 	free_task(tsk);
995 }
996 EXPORT_SYMBOL_GPL(__put_task_struct);
997 
__put_task_struct_rcu_cb(struct rcu_head * rhp)998 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
999 {
1000 	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
1001 
1002 	__put_task_struct(task);
1003 }
1004 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
1005 
arch_task_cache_init(void)1006 void __init __weak arch_task_cache_init(void) { }
1007 
1008 /*
1009  * set_max_threads
1010  */
set_max_threads(unsigned int max_threads_suggested)1011 static void __init set_max_threads(unsigned int max_threads_suggested)
1012 {
1013 	u64 threads;
1014 	unsigned long nr_pages = memblock_estimated_nr_free_pages();
1015 
1016 	/*
1017 	 * The number of threads shall be limited such that the thread
1018 	 * structures may only consume a small part of the available memory.
1019 	 */
1020 	if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
1021 		threads = MAX_THREADS;
1022 	else
1023 		threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
1024 				    (u64) THREAD_SIZE * 8UL);
1025 
1026 	if (threads > max_threads_suggested)
1027 		threads = max_threads_suggested;
1028 
1029 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
1030 }
1031 
1032 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
1033 /* Initialized by the architecture: */
1034 int arch_task_struct_size __read_mostly;
1035 #endif
1036 
task_struct_whitelist(unsigned long * offset,unsigned long * size)1037 static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
1038 {
1039 	/* Fetch thread_struct whitelist for the architecture. */
1040 	arch_thread_struct_whitelist(offset, size);
1041 
1042 	/*
1043 	 * Handle zero-sized whitelist or empty thread_struct, otherwise
1044 	 * adjust offset to position of thread_struct in task_struct.
1045 	 */
1046 	if (unlikely(*size == 0))
1047 		*offset = 0;
1048 	else
1049 		*offset += offsetof(struct task_struct, thread);
1050 }
1051 
fork_init(void)1052 void __init fork_init(void)
1053 {
1054 	int i;
1055 #ifndef ARCH_MIN_TASKALIGN
1056 #define ARCH_MIN_TASKALIGN	0
1057 #endif
1058 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
1059 	unsigned long useroffset, usersize;
1060 
1061 	/* create a slab on which task_structs can be allocated */
1062 	task_struct_whitelist(&useroffset, &usersize);
1063 	task_struct_cachep = kmem_cache_create_usercopy("task_struct",
1064 			arch_task_struct_size, align,
1065 			SLAB_PANIC|SLAB_ACCOUNT,
1066 			useroffset, usersize, NULL);
1067 
1068 	/* do the arch specific task caches init */
1069 	arch_task_cache_init();
1070 
1071 	set_max_threads(MAX_THREADS);
1072 
1073 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
1074 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
1075 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
1076 		init_task.signal->rlim[RLIMIT_NPROC];
1077 
1078 	for (i = 0; i < UCOUNT_COUNTS; i++)
1079 		init_user_ns.ucount_max[i] = max_threads/2;
1080 
1081 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
1082 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
1083 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
1084 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
1085 
1086 #ifdef CONFIG_VMAP_STACK
1087 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
1088 			  NULL, free_vm_stack_cache);
1089 #endif
1090 
1091 	scs_init();
1092 
1093 	lockdep_init_task(&init_task);
1094 	uprobes_init();
1095 }
1096 
arch_dup_task_struct(struct task_struct * dst,struct task_struct * src)1097 int __weak arch_dup_task_struct(struct task_struct *dst,
1098 					       struct task_struct *src)
1099 {
1100 	*dst = *src;
1101 	return 0;
1102 }
1103 
set_task_stack_end_magic(struct task_struct * tsk)1104 void set_task_stack_end_magic(struct task_struct *tsk)
1105 {
1106 	unsigned long *stackend;
1107 
1108 	stackend = end_of_stack(tsk);
1109 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
1110 }
1111 
dup_task_struct(struct task_struct * orig,int node)1112 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
1113 {
1114 	struct task_struct *tsk;
1115 	int err;
1116 
1117 	if (node == NUMA_NO_NODE)
1118 		node = tsk_fork_get_node(orig);
1119 	tsk = alloc_task_struct_node(node);
1120 	if (!tsk)
1121 		return NULL;
1122 
1123 	err = arch_dup_task_struct(tsk, orig);
1124 	if (err)
1125 		goto free_tsk;
1126 
1127 	err = alloc_thread_stack_node(tsk, node);
1128 	if (err)
1129 		goto free_tsk;
1130 
1131 #ifdef CONFIG_THREAD_INFO_IN_TASK
1132 	refcount_set(&tsk->stack_refcount, 1);
1133 #endif
1134 	account_kernel_stack(tsk, 1);
1135 
1136 	err = scs_prepare(tsk, node);
1137 	if (err)
1138 		goto free_stack;
1139 
1140 #ifdef CONFIG_SECCOMP
1141 	/*
1142 	 * We must handle setting up seccomp filters once we're under
1143 	 * the sighand lock in case orig has changed between now and
1144 	 * then. Until then, filter must be NULL to avoid messing up
1145 	 * the usage counts on the error path calling free_task.
1146 	 */
1147 	tsk->seccomp.filter = NULL;
1148 #endif
1149 
1150 	setup_thread_stack(tsk, orig);
1151 	clear_user_return_notifier(tsk);
1152 	clear_tsk_need_resched(tsk);
1153 	set_task_stack_end_magic(tsk);
1154 	clear_syscall_work_syscall_user_dispatch(tsk);
1155 
1156 #ifdef CONFIG_STACKPROTECTOR
1157 	tsk->stack_canary = get_random_canary();
1158 #endif
1159 	if (orig->cpus_ptr == &orig->cpus_mask)
1160 		tsk->cpus_ptr = &tsk->cpus_mask;
1161 	dup_user_cpus_ptr(tsk, orig, node);
1162 
1163 	/*
1164 	 * One for the user space visible state that goes away when reaped.
1165 	 * One for the scheduler.
1166 	 */
1167 	refcount_set(&tsk->rcu_users, 2);
1168 	/* One for the rcu users */
1169 	refcount_set(&tsk->usage, 1);
1170 #ifdef CONFIG_BLK_DEV_IO_TRACE
1171 	tsk->btrace_seq = 0;
1172 #endif
1173 	tsk->splice_pipe = NULL;
1174 	tsk->task_frag.page = NULL;
1175 	tsk->wake_q.next = NULL;
1176 	tsk->worker_private = NULL;
1177 
1178 	kcov_task_init(tsk);
1179 	kmsan_task_create(tsk);
1180 	kmap_local_fork(tsk);
1181 
1182 #ifdef CONFIG_FAULT_INJECTION
1183 	tsk->fail_nth = 0;
1184 #endif
1185 
1186 #ifdef CONFIG_BLK_CGROUP
1187 	tsk->throttle_disk = NULL;
1188 	tsk->use_memdelay = 0;
1189 #endif
1190 
1191 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1192 	tsk->pasid_activated = 0;
1193 #endif
1194 
1195 #ifdef CONFIG_MEMCG
1196 	tsk->active_memcg = NULL;
1197 #endif
1198 
1199 #ifdef CONFIG_X86_BUS_LOCK_DETECT
1200 	tsk->reported_split_lock = 0;
1201 #endif
1202 
1203 #ifdef CONFIG_SCHED_MM_CID
1204 	tsk->mm_cid = -1;
1205 	tsk->last_mm_cid = -1;
1206 	tsk->mm_cid_active = 0;
1207 	tsk->migrate_from_cpu = -1;
1208 #endif
1209 	return tsk;
1210 
1211 free_stack:
1212 	exit_task_stack_account(tsk);
1213 	free_thread_stack(tsk);
1214 free_tsk:
1215 	free_task_struct(tsk);
1216 	return NULL;
1217 }
1218 
1219 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1220 
1221 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
1222 
coredump_filter_setup(char * s)1223 static int __init coredump_filter_setup(char *s)
1224 {
1225 	default_dump_filter =
1226 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
1227 		MMF_DUMP_FILTER_MASK;
1228 	return 1;
1229 }
1230 
1231 __setup("coredump_filter=", coredump_filter_setup);
1232 
1233 #include <linux/init_task.h>
1234 
mm_init_aio(struct mm_struct * mm)1235 static void mm_init_aio(struct mm_struct *mm)
1236 {
1237 #ifdef CONFIG_AIO
1238 	spin_lock_init(&mm->ioctx_lock);
1239 	mm->ioctx_table = NULL;
1240 #endif
1241 }
1242 
mm_clear_owner(struct mm_struct * mm,struct task_struct * p)1243 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1244 					   struct task_struct *p)
1245 {
1246 #ifdef CONFIG_MEMCG
1247 	if (mm->owner == p)
1248 		WRITE_ONCE(mm->owner, NULL);
1249 #endif
1250 }
1251 
mm_init_owner(struct mm_struct * mm,struct task_struct * p)1252 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1253 {
1254 #ifdef CONFIG_MEMCG
1255 	mm->owner = p;
1256 #endif
1257 }
1258 
mm_init_uprobes_state(struct mm_struct * mm)1259 static void mm_init_uprobes_state(struct mm_struct *mm)
1260 {
1261 #ifdef CONFIG_UPROBES
1262 	mm->uprobes_state.xol_area = NULL;
1263 #endif
1264 }
1265 
mm_init(struct mm_struct * mm,struct task_struct * p,struct user_namespace * user_ns)1266 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1267 	struct user_namespace *user_ns)
1268 {
1269 	mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1270 	mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1271 	atomic_set(&mm->mm_users, 1);
1272 	atomic_set(&mm->mm_count, 1);
1273 	seqcount_init(&mm->write_protect_seq);
1274 	mmap_init_lock(mm);
1275 	INIT_LIST_HEAD(&mm->mmlist);
1276 	mm_pgtables_bytes_init(mm);
1277 	mm->map_count = 0;
1278 	mm->locked_vm = 0;
1279 	atomic64_set(&mm->pinned_vm, 0);
1280 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1281 	spin_lock_init(&mm->page_table_lock);
1282 	spin_lock_init(&mm->arg_lock);
1283 	mm_init_cpumask(mm);
1284 	mm_init_aio(mm);
1285 	mm_init_owner(mm, p);
1286 	mm_pasid_init(mm);
1287 	RCU_INIT_POINTER(mm->exe_file, NULL);
1288 	mmu_notifier_subscriptions_init(mm);
1289 	init_tlb_flush_pending(mm);
1290 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1291 	mm->pmd_huge_pte = NULL;
1292 #endif
1293 	mm_init_uprobes_state(mm);
1294 	hugetlb_count_init(mm);
1295 
1296 	if (current->mm) {
1297 		mm->flags = mmf_init_flags(current->mm->flags);
1298 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1299 	} else {
1300 		mm->flags = default_dump_filter;
1301 		mm->def_flags = 0;
1302 	}
1303 
1304 	if (mm_alloc_pgd(mm))
1305 		goto fail_nopgd;
1306 
1307 	if (init_new_context(p, mm))
1308 		goto fail_nocontext;
1309 
1310 	if (mm_alloc_cid(mm, p))
1311 		goto fail_cid;
1312 
1313 	if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1314 				     NR_MM_COUNTERS))
1315 		goto fail_pcpu;
1316 
1317 	mm->user_ns = get_user_ns(user_ns);
1318 	lru_gen_init_mm(mm);
1319 	return mm;
1320 
1321 fail_pcpu:
1322 	mm_destroy_cid(mm);
1323 fail_cid:
1324 	destroy_context(mm);
1325 fail_nocontext:
1326 	mm_free_pgd(mm);
1327 fail_nopgd:
1328 	free_mm(mm);
1329 	return NULL;
1330 }
1331 
1332 /*
1333  * Allocate and initialize an mm_struct.
1334  */
mm_alloc(void)1335 struct mm_struct *mm_alloc(void)
1336 {
1337 	struct mm_struct *mm;
1338 
1339 	mm = allocate_mm();
1340 	if (!mm)
1341 		return NULL;
1342 
1343 	memset(mm, 0, sizeof(*mm));
1344 	return mm_init(mm, current, current_user_ns());
1345 }
1346 EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1347 
__mmput(struct mm_struct * mm)1348 static inline void __mmput(struct mm_struct *mm)
1349 {
1350 	VM_BUG_ON(atomic_read(&mm->mm_users));
1351 
1352 	uprobe_clear_state(mm);
1353 	exit_aio(mm);
1354 	ksm_exit(mm);
1355 	khugepaged_exit(mm); /* must run before exit_mmap */
1356 	exit_mmap(mm);
1357 	mm_put_huge_zero_folio(mm);
1358 	set_mm_exe_file(mm, NULL);
1359 	if (!list_empty(&mm->mmlist)) {
1360 		spin_lock(&mmlist_lock);
1361 		list_del(&mm->mmlist);
1362 		spin_unlock(&mmlist_lock);
1363 	}
1364 	if (mm->binfmt)
1365 		module_put(mm->binfmt->module);
1366 	lru_gen_del_mm(mm);
1367 	mmdrop(mm);
1368 }
1369 
1370 /*
1371  * Decrement the use count and release all resources for an mm.
1372  */
mmput(struct mm_struct * mm)1373 void mmput(struct mm_struct *mm)
1374 {
1375 	might_sleep();
1376 
1377 	if (atomic_dec_and_test(&mm->mm_users))
1378 		__mmput(mm);
1379 }
1380 EXPORT_SYMBOL_GPL(mmput);
1381 
1382 #ifdef CONFIG_MMU
mmput_async_fn(struct work_struct * work)1383 static void mmput_async_fn(struct work_struct *work)
1384 {
1385 	struct mm_struct *mm = container_of(work, struct mm_struct,
1386 					    async_put_work);
1387 
1388 	__mmput(mm);
1389 }
1390 
mmput_async(struct mm_struct * mm)1391 void mmput_async(struct mm_struct *mm)
1392 {
1393 	if (atomic_dec_and_test(&mm->mm_users)) {
1394 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
1395 		schedule_work(&mm->async_put_work);
1396 	}
1397 }
1398 EXPORT_SYMBOL_GPL(mmput_async);
1399 #endif
1400 
1401 /**
1402  * set_mm_exe_file - change a reference to the mm's executable file
1403  * @mm: The mm to change.
1404  * @new_exe_file: The new file to use.
1405  *
1406  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1407  *
1408  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1409  * invocations: in mmput() nobody alive left, in execve it happens before
1410  * the new mm is made visible to anyone.
1411  *
1412  * Can only fail if new_exe_file != NULL.
1413  */
set_mm_exe_file(struct mm_struct * mm,struct file * new_exe_file)1414 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1415 {
1416 	struct file *old_exe_file;
1417 
1418 	/*
1419 	 * It is safe to dereference the exe_file without RCU as
1420 	 * this function is only called if nobody else can access
1421 	 * this mm -- see comment above for justification.
1422 	 */
1423 	old_exe_file = rcu_dereference_raw(mm->exe_file);
1424 
1425 	if (new_exe_file) {
1426 		/*
1427 		 * We expect the caller (i.e., sys_execve) to already denied
1428 		 * write access, so this is unlikely to fail.
1429 		 */
1430 		if (unlikely(exe_file_deny_write_access(new_exe_file)))
1431 			return -EACCES;
1432 		get_file(new_exe_file);
1433 	}
1434 	rcu_assign_pointer(mm->exe_file, new_exe_file);
1435 	if (old_exe_file) {
1436 		exe_file_allow_write_access(old_exe_file);
1437 		fput(old_exe_file);
1438 	}
1439 	return 0;
1440 }
1441 
1442 /**
1443  * replace_mm_exe_file - replace a reference to the mm's executable file
1444  * @mm: The mm to change.
1445  * @new_exe_file: The new file to use.
1446  *
1447  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1448  *
1449  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1450  */
replace_mm_exe_file(struct mm_struct * mm,struct file * new_exe_file)1451 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1452 {
1453 	struct vm_area_struct *vma;
1454 	struct file *old_exe_file;
1455 	int ret = 0;
1456 
1457 	/* Forbid mm->exe_file change if old file still mapped. */
1458 	old_exe_file = get_mm_exe_file(mm);
1459 	if (old_exe_file) {
1460 		VMA_ITERATOR(vmi, mm, 0);
1461 		mmap_read_lock(mm);
1462 		for_each_vma(vmi, vma) {
1463 			if (!vma->vm_file)
1464 				continue;
1465 			if (path_equal(&vma->vm_file->f_path,
1466 				       &old_exe_file->f_path)) {
1467 				ret = -EBUSY;
1468 				break;
1469 			}
1470 		}
1471 		mmap_read_unlock(mm);
1472 		fput(old_exe_file);
1473 		if (ret)
1474 			return ret;
1475 	}
1476 
1477 	ret = exe_file_deny_write_access(new_exe_file);
1478 	if (ret)
1479 		return -EACCES;
1480 	get_file(new_exe_file);
1481 
1482 	/* set the new file */
1483 	mmap_write_lock(mm);
1484 	old_exe_file = rcu_dereference_raw(mm->exe_file);
1485 	rcu_assign_pointer(mm->exe_file, new_exe_file);
1486 	mmap_write_unlock(mm);
1487 
1488 	if (old_exe_file) {
1489 		exe_file_allow_write_access(old_exe_file);
1490 		fput(old_exe_file);
1491 	}
1492 	return 0;
1493 }
1494 
1495 /**
1496  * get_mm_exe_file - acquire a reference to the mm's executable file
1497  * @mm: The mm of interest.
1498  *
1499  * Returns %NULL if mm has no associated executable file.
1500  * User must release file via fput().
1501  */
get_mm_exe_file(struct mm_struct * mm)1502 struct file *get_mm_exe_file(struct mm_struct *mm)
1503 {
1504 	struct file *exe_file;
1505 
1506 	rcu_read_lock();
1507 	exe_file = get_file_rcu(&mm->exe_file);
1508 	rcu_read_unlock();
1509 	return exe_file;
1510 }
1511 
1512 /**
1513  * get_task_exe_file - acquire a reference to the task's executable file
1514  * @task: The task.
1515  *
1516  * Returns %NULL if task's mm (if any) has no associated executable file or
1517  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1518  * User must release file via fput().
1519  */
get_task_exe_file(struct task_struct * task)1520 struct file *get_task_exe_file(struct task_struct *task)
1521 {
1522 	struct file *exe_file = NULL;
1523 	struct mm_struct *mm;
1524 
1525 	if (task->flags & PF_KTHREAD)
1526 		return NULL;
1527 
1528 	task_lock(task);
1529 	mm = task->mm;
1530 	if (mm)
1531 		exe_file = get_mm_exe_file(mm);
1532 	task_unlock(task);
1533 	return exe_file;
1534 }
1535 
1536 /**
1537  * get_task_mm - acquire a reference to the task's mm
1538  * @task: The task.
1539  *
1540  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1541  * this kernel workthread has transiently adopted a user mm with use_mm,
1542  * to do its AIO) is not set and if so returns a reference to it, after
1543  * bumping up the use count.  User must release the mm via mmput()
1544  * after use.  Typically used by /proc and ptrace.
1545  */
get_task_mm(struct task_struct * task)1546 struct mm_struct *get_task_mm(struct task_struct *task)
1547 {
1548 	struct mm_struct *mm;
1549 
1550 	if (task->flags & PF_KTHREAD)
1551 		return NULL;
1552 
1553 	task_lock(task);
1554 	mm = task->mm;
1555 	if (mm)
1556 		mmget(mm);
1557 	task_unlock(task);
1558 	return mm;
1559 }
1560 EXPORT_SYMBOL_GPL(get_task_mm);
1561 
mm_access(struct task_struct * task,unsigned int mode)1562 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1563 {
1564 	struct mm_struct *mm;
1565 	int err;
1566 
1567 	err =  down_read_killable(&task->signal->exec_update_lock);
1568 	if (err)
1569 		return ERR_PTR(err);
1570 
1571 	mm = get_task_mm(task);
1572 	if (!mm) {
1573 		mm = ERR_PTR(-ESRCH);
1574 	} else if (mm != current->mm && !ptrace_may_access(task, mode)) {
1575 		mmput(mm);
1576 		mm = ERR_PTR(-EACCES);
1577 	}
1578 	up_read(&task->signal->exec_update_lock);
1579 
1580 	return mm;
1581 }
1582 
complete_vfork_done(struct task_struct * tsk)1583 static void complete_vfork_done(struct task_struct *tsk)
1584 {
1585 	struct completion *vfork;
1586 
1587 	task_lock(tsk);
1588 	vfork = tsk->vfork_done;
1589 	if (likely(vfork)) {
1590 		tsk->vfork_done = NULL;
1591 		complete(vfork);
1592 	}
1593 	task_unlock(tsk);
1594 }
1595 
wait_for_vfork_done(struct task_struct * child,struct completion * vfork)1596 static int wait_for_vfork_done(struct task_struct *child,
1597 				struct completion *vfork)
1598 {
1599 	unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1600 	int killed;
1601 
1602 	cgroup_enter_frozen();
1603 	killed = wait_for_completion_state(vfork, state);
1604 	cgroup_leave_frozen(false);
1605 
1606 	if (killed) {
1607 		task_lock(child);
1608 		child->vfork_done = NULL;
1609 		task_unlock(child);
1610 	}
1611 
1612 	put_task_struct(child);
1613 	return killed;
1614 }
1615 
1616 /* Please note the differences between mmput and mm_release.
1617  * mmput is called whenever we stop holding onto a mm_struct,
1618  * error success whatever.
1619  *
1620  * mm_release is called after a mm_struct has been removed
1621  * from the current process.
1622  *
1623  * This difference is important for error handling, when we
1624  * only half set up a mm_struct for a new process and need to restore
1625  * the old one.  Because we mmput the new mm_struct before
1626  * restoring the old one. . .
1627  * Eric Biederman 10 January 1998
1628  */
mm_release(struct task_struct * tsk,struct mm_struct * mm)1629 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1630 {
1631 	uprobe_free_utask(tsk);
1632 
1633 	/* Get rid of any cached register state */
1634 	deactivate_mm(tsk, mm);
1635 
1636 	/*
1637 	 * Signal userspace if we're not exiting with a core dump
1638 	 * because we want to leave the value intact for debugging
1639 	 * purposes.
1640 	 */
1641 	if (tsk->clear_child_tid) {
1642 		if (atomic_read(&mm->mm_users) > 1) {
1643 			/*
1644 			 * We don't check the error code - if userspace has
1645 			 * not set up a proper pointer then tough luck.
1646 			 */
1647 			put_user(0, tsk->clear_child_tid);
1648 			do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1649 					1, NULL, NULL, 0, 0);
1650 		}
1651 		tsk->clear_child_tid = NULL;
1652 	}
1653 
1654 	/*
1655 	 * All done, finally we can wake up parent and return this mm to him.
1656 	 * Also kthread_stop() uses this completion for synchronization.
1657 	 */
1658 	if (tsk->vfork_done)
1659 		complete_vfork_done(tsk);
1660 }
1661 
exit_mm_release(struct task_struct * tsk,struct mm_struct * mm)1662 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1663 {
1664 	futex_exit_release(tsk);
1665 	mm_release(tsk, mm);
1666 }
1667 
exec_mm_release(struct task_struct * tsk,struct mm_struct * mm)1668 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1669 {
1670 	futex_exec_release(tsk);
1671 	mm_release(tsk, mm);
1672 }
1673 
1674 /**
1675  * dup_mm() - duplicates an existing mm structure
1676  * @tsk: the task_struct with which the new mm will be associated.
1677  * @oldmm: the mm to duplicate.
1678  *
1679  * Allocates a new mm structure and duplicates the provided @oldmm structure
1680  * content into it.
1681  *
1682  * Return: the duplicated mm or NULL on failure.
1683  */
dup_mm(struct task_struct * tsk,struct mm_struct * oldmm)1684 static struct mm_struct *dup_mm(struct task_struct *tsk,
1685 				struct mm_struct *oldmm)
1686 {
1687 	struct mm_struct *mm;
1688 	int err;
1689 
1690 	mm = allocate_mm();
1691 	if (!mm)
1692 		goto fail_nomem;
1693 
1694 	memcpy(mm, oldmm, sizeof(*mm));
1695 
1696 	if (!mm_init(mm, tsk, mm->user_ns))
1697 		goto fail_nomem;
1698 
1699 	uprobe_start_dup_mmap();
1700 	err = dup_mmap(mm, oldmm);
1701 	if (err)
1702 		goto free_pt;
1703 	uprobe_end_dup_mmap();
1704 
1705 	mm->hiwater_rss = get_mm_rss(mm);
1706 	mm->hiwater_vm = mm->total_vm;
1707 
1708 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1709 		goto free_pt;
1710 
1711 	return mm;
1712 
1713 free_pt:
1714 	/* don't put binfmt in mmput, we haven't got module yet */
1715 	mm->binfmt = NULL;
1716 	mm_init_owner(mm, NULL);
1717 	mmput(mm);
1718 	if (err)
1719 		uprobe_end_dup_mmap();
1720 
1721 fail_nomem:
1722 	return NULL;
1723 }
1724 
copy_mm(unsigned long clone_flags,struct task_struct * tsk)1725 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1726 {
1727 	struct mm_struct *mm, *oldmm;
1728 
1729 	tsk->min_flt = tsk->maj_flt = 0;
1730 	tsk->nvcsw = tsk->nivcsw = 0;
1731 #ifdef CONFIG_DETECT_HUNG_TASK
1732 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1733 	tsk->last_switch_time = 0;
1734 #endif
1735 
1736 	tsk->mm = NULL;
1737 	tsk->active_mm = NULL;
1738 
1739 	/*
1740 	 * Are we cloning a kernel thread?
1741 	 *
1742 	 * We need to steal a active VM for that..
1743 	 */
1744 	oldmm = current->mm;
1745 	if (!oldmm)
1746 		return 0;
1747 
1748 	if (clone_flags & CLONE_VM) {
1749 		mmget(oldmm);
1750 		mm = oldmm;
1751 	} else {
1752 		mm = dup_mm(tsk, current->mm);
1753 		if (!mm)
1754 			return -ENOMEM;
1755 	}
1756 
1757 	tsk->mm = mm;
1758 	tsk->active_mm = mm;
1759 	sched_mm_cid_fork(tsk);
1760 	return 0;
1761 }
1762 
copy_fs(unsigned long clone_flags,struct task_struct * tsk)1763 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1764 {
1765 	struct fs_struct *fs = current->fs;
1766 	if (clone_flags & CLONE_FS) {
1767 		/* tsk->fs is already what we want */
1768 		spin_lock(&fs->lock);
1769 		/* "users" and "in_exec" locked for check_unsafe_exec() */
1770 		if (fs->in_exec) {
1771 			spin_unlock(&fs->lock);
1772 			return -EAGAIN;
1773 		}
1774 		fs->users++;
1775 		spin_unlock(&fs->lock);
1776 		return 0;
1777 	}
1778 	tsk->fs = copy_fs_struct(fs);
1779 	if (!tsk->fs)
1780 		return -ENOMEM;
1781 	return 0;
1782 }
1783 
copy_files(unsigned long clone_flags,struct task_struct * tsk,int no_files)1784 static int copy_files(unsigned long clone_flags, struct task_struct *tsk,
1785 		      int no_files)
1786 {
1787 	struct files_struct *oldf, *newf;
1788 
1789 	/*
1790 	 * A background process may not have any files ...
1791 	 */
1792 	oldf = current->files;
1793 	if (!oldf)
1794 		return 0;
1795 
1796 	if (no_files) {
1797 		tsk->files = NULL;
1798 		return 0;
1799 	}
1800 
1801 	if (clone_flags & CLONE_FILES) {
1802 		atomic_inc(&oldf->count);
1803 		return 0;
1804 	}
1805 
1806 	newf = dup_fd(oldf, NULL);
1807 	if (IS_ERR(newf))
1808 		return PTR_ERR(newf);
1809 
1810 	tsk->files = newf;
1811 	return 0;
1812 }
1813 
copy_sighand(unsigned long clone_flags,struct task_struct * tsk)1814 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1815 {
1816 	struct sighand_struct *sig;
1817 
1818 	if (clone_flags & CLONE_SIGHAND) {
1819 		refcount_inc(&current->sighand->count);
1820 		return 0;
1821 	}
1822 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1823 	RCU_INIT_POINTER(tsk->sighand, sig);
1824 	if (!sig)
1825 		return -ENOMEM;
1826 
1827 	refcount_set(&sig->count, 1);
1828 	spin_lock_irq(&current->sighand->siglock);
1829 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1830 	spin_unlock_irq(&current->sighand->siglock);
1831 
1832 	/* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1833 	if (clone_flags & CLONE_CLEAR_SIGHAND)
1834 		flush_signal_handlers(tsk, 0);
1835 
1836 	return 0;
1837 }
1838 
__cleanup_sighand(struct sighand_struct * sighand)1839 void __cleanup_sighand(struct sighand_struct *sighand)
1840 {
1841 	if (refcount_dec_and_test(&sighand->count)) {
1842 		signalfd_cleanup(sighand);
1843 		/*
1844 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1845 		 * without an RCU grace period, see __lock_task_sighand().
1846 		 */
1847 		kmem_cache_free(sighand_cachep, sighand);
1848 	}
1849 }
1850 
1851 /*
1852  * Initialize POSIX timer handling for a thread group.
1853  */
posix_cpu_timers_init_group(struct signal_struct * sig)1854 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1855 {
1856 	struct posix_cputimers *pct = &sig->posix_cputimers;
1857 	unsigned long cpu_limit;
1858 
1859 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1860 	posix_cputimers_group_init(pct, cpu_limit);
1861 }
1862 
copy_signal(unsigned long clone_flags,struct task_struct * tsk)1863 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1864 {
1865 	struct signal_struct *sig;
1866 
1867 	if (clone_flags & CLONE_THREAD)
1868 		return 0;
1869 
1870 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1871 	tsk->signal = sig;
1872 	if (!sig)
1873 		return -ENOMEM;
1874 
1875 	sig->nr_threads = 1;
1876 	sig->quick_threads = 1;
1877 	atomic_set(&sig->live, 1);
1878 	refcount_set(&sig->sigcnt, 1);
1879 
1880 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1881 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1882 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1883 
1884 	init_waitqueue_head(&sig->wait_chldexit);
1885 	sig->curr_target = tsk;
1886 	init_sigpending(&sig->shared_pending);
1887 	INIT_HLIST_HEAD(&sig->multiprocess);
1888 	seqlock_init(&sig->stats_lock);
1889 	prev_cputime_init(&sig->prev_cputime);
1890 
1891 #ifdef CONFIG_POSIX_TIMERS
1892 	INIT_HLIST_HEAD(&sig->posix_timers);
1893 	INIT_HLIST_HEAD(&sig->ignored_posix_timers);
1894 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1895 	sig->real_timer.function = it_real_fn;
1896 #endif
1897 
1898 	task_lock(current->group_leader);
1899 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1900 	task_unlock(current->group_leader);
1901 
1902 	posix_cpu_timers_init_group(sig);
1903 
1904 	tty_audit_fork(sig);
1905 	sched_autogroup_fork(sig);
1906 
1907 	sig->oom_score_adj = current->signal->oom_score_adj;
1908 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1909 
1910 	mutex_init(&sig->cred_guard_mutex);
1911 	init_rwsem(&sig->exec_update_lock);
1912 
1913 	return 0;
1914 }
1915 
copy_seccomp(struct task_struct * p)1916 static void copy_seccomp(struct task_struct *p)
1917 {
1918 #ifdef CONFIG_SECCOMP
1919 	/*
1920 	 * Must be called with sighand->lock held, which is common to
1921 	 * all threads in the group. Holding cred_guard_mutex is not
1922 	 * needed because this new task is not yet running and cannot
1923 	 * be racing exec.
1924 	 */
1925 	assert_spin_locked(&current->sighand->siglock);
1926 
1927 	/* Ref-count the new filter user, and assign it. */
1928 	get_seccomp_filter(current);
1929 	p->seccomp = current->seccomp;
1930 
1931 	/*
1932 	 * Explicitly enable no_new_privs here in case it got set
1933 	 * between the task_struct being duplicated and holding the
1934 	 * sighand lock. The seccomp state and nnp must be in sync.
1935 	 */
1936 	if (task_no_new_privs(current))
1937 		task_set_no_new_privs(p);
1938 
1939 	/*
1940 	 * If the parent gained a seccomp mode after copying thread
1941 	 * flags and between before we held the sighand lock, we have
1942 	 * to manually enable the seccomp thread flag here.
1943 	 */
1944 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1945 		set_task_syscall_work(p, SECCOMP);
1946 #endif
1947 }
1948 
SYSCALL_DEFINE1(set_tid_address,int __user *,tidptr)1949 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1950 {
1951 	current->clear_child_tid = tidptr;
1952 
1953 	return task_pid_vnr(current);
1954 }
1955 
rt_mutex_init_task(struct task_struct * p)1956 static void rt_mutex_init_task(struct task_struct *p)
1957 {
1958 	raw_spin_lock_init(&p->pi_lock);
1959 #ifdef CONFIG_RT_MUTEXES
1960 	p->pi_waiters = RB_ROOT_CACHED;
1961 	p->pi_top_task = NULL;
1962 	p->pi_blocked_on = NULL;
1963 #endif
1964 }
1965 
init_task_pid_links(struct task_struct * task)1966 static inline void init_task_pid_links(struct task_struct *task)
1967 {
1968 	enum pid_type type;
1969 
1970 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1971 		INIT_HLIST_NODE(&task->pid_links[type]);
1972 }
1973 
1974 static inline void
init_task_pid(struct task_struct * task,enum pid_type type,struct pid * pid)1975 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1976 {
1977 	if (type == PIDTYPE_PID)
1978 		task->thread_pid = pid;
1979 	else
1980 		task->signal->pids[type] = pid;
1981 }
1982 
rcu_copy_process(struct task_struct * p)1983 static inline void rcu_copy_process(struct task_struct *p)
1984 {
1985 #ifdef CONFIG_PREEMPT_RCU
1986 	p->rcu_read_lock_nesting = 0;
1987 	p->rcu_read_unlock_special.s = 0;
1988 	p->rcu_blocked_node = NULL;
1989 	INIT_LIST_HEAD(&p->rcu_node_entry);
1990 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1991 #ifdef CONFIG_TASKS_RCU
1992 	p->rcu_tasks_holdout = false;
1993 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1994 	p->rcu_tasks_idle_cpu = -1;
1995 	INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1996 #endif /* #ifdef CONFIG_TASKS_RCU */
1997 #ifdef CONFIG_TASKS_TRACE_RCU
1998 	p->trc_reader_nesting = 0;
1999 	p->trc_reader_special.s = 0;
2000 	INIT_LIST_HEAD(&p->trc_holdout_list);
2001 	INIT_LIST_HEAD(&p->trc_blkd_node);
2002 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
2003 }
2004 
2005 /**
2006  * __pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2007  * @pid:   the struct pid for which to create a pidfd
2008  * @flags: flags of the new @pidfd
2009  * @ret: Where to return the file for the pidfd.
2010  *
2011  * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2012  * caller's file descriptor table. The pidfd is reserved but not installed yet.
2013  *
2014  * The helper doesn't perform checks on @pid which makes it useful for pidfds
2015  * created via CLONE_PIDFD where @pid has no task attached when the pidfd and
2016  * pidfd file are prepared.
2017  *
2018  * If this function returns successfully the caller is responsible to either
2019  * call fd_install() passing the returned pidfd and pidfd file as arguments in
2020  * order to install the pidfd into its file descriptor table or they must use
2021  * put_unused_fd() and fput() on the returned pidfd and pidfd file
2022  * respectively.
2023  *
2024  * This function is useful when a pidfd must already be reserved but there
2025  * might still be points of failure afterwards and the caller wants to ensure
2026  * that no pidfd is leaked into its file descriptor table.
2027  *
2028  * Return: On success, a reserved pidfd is returned from the function and a new
2029  *         pidfd file is returned in the last argument to the function. On
2030  *         error, a negative error code is returned from the function and the
2031  *         last argument remains unchanged.
2032  */
__pidfd_prepare(struct pid * pid,unsigned int flags,struct file ** ret)2033 static int __pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2034 {
2035 	int pidfd;
2036 	struct file *pidfd_file;
2037 
2038 	pidfd = get_unused_fd_flags(O_CLOEXEC);
2039 	if (pidfd < 0)
2040 		return pidfd;
2041 
2042 	pidfd_file = pidfs_alloc_file(pid, flags | O_RDWR);
2043 	if (IS_ERR(pidfd_file)) {
2044 		put_unused_fd(pidfd);
2045 		return PTR_ERR(pidfd_file);
2046 	}
2047 	/*
2048 	 * anon_inode_getfile() ignores everything outside of the
2049 	 * O_ACCMODE | O_NONBLOCK mask, set PIDFD_THREAD manually.
2050 	 */
2051 	pidfd_file->f_flags |= (flags & PIDFD_THREAD);
2052 	*ret = pidfd_file;
2053 	return pidfd;
2054 }
2055 
2056 /**
2057  * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
2058  * @pid:   the struct pid for which to create a pidfd
2059  * @flags: flags of the new @pidfd
2060  * @ret: Where to return the pidfd.
2061  *
2062  * Allocate a new file that stashes @pid and reserve a new pidfd number in the
2063  * caller's file descriptor table. The pidfd is reserved but not installed yet.
2064  *
2065  * The helper verifies that @pid is still in use, without PIDFD_THREAD the
2066  * task identified by @pid must be a thread-group leader.
2067  *
2068  * If this function returns successfully the caller is responsible to either
2069  * call fd_install() passing the returned pidfd and pidfd file as arguments in
2070  * order to install the pidfd into its file descriptor table or they must use
2071  * put_unused_fd() and fput() on the returned pidfd and pidfd file
2072  * respectively.
2073  *
2074  * This function is useful when a pidfd must already be reserved but there
2075  * might still be points of failure afterwards and the caller wants to ensure
2076  * that no pidfd is leaked into its file descriptor table.
2077  *
2078  * Return: On success, a reserved pidfd is returned from the function and a new
2079  *         pidfd file is returned in the last argument to the function. On
2080  *         error, a negative error code is returned from the function and the
2081  *         last argument remains unchanged.
2082  */
pidfd_prepare(struct pid * pid,unsigned int flags,struct file ** ret)2083 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret)
2084 {
2085 	bool thread = flags & PIDFD_THREAD;
2086 
2087 	if (!pid || !pid_has_task(pid, thread ? PIDTYPE_PID : PIDTYPE_TGID))
2088 		return -EINVAL;
2089 
2090 	return __pidfd_prepare(pid, flags, ret);
2091 }
2092 
__delayed_free_task(struct rcu_head * rhp)2093 static void __delayed_free_task(struct rcu_head *rhp)
2094 {
2095 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
2096 
2097 	free_task(tsk);
2098 }
2099 
delayed_free_task(struct task_struct * tsk)2100 static __always_inline void delayed_free_task(struct task_struct *tsk)
2101 {
2102 	if (IS_ENABLED(CONFIG_MEMCG))
2103 		call_rcu(&tsk->rcu, __delayed_free_task);
2104 	else
2105 		free_task(tsk);
2106 }
2107 
copy_oom_score_adj(u64 clone_flags,struct task_struct * tsk)2108 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
2109 {
2110 	/* Skip if kernel thread */
2111 	if (!tsk->mm)
2112 		return;
2113 
2114 	/* Skip if spawning a thread or using vfork */
2115 	if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
2116 		return;
2117 
2118 	/* We need to synchronize with __set_oom_adj */
2119 	mutex_lock(&oom_adj_mutex);
2120 	set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
2121 	/* Update the values in case they were changed after copy_signal */
2122 	tsk->signal->oom_score_adj = current->signal->oom_score_adj;
2123 	tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2124 	mutex_unlock(&oom_adj_mutex);
2125 }
2126 
2127 #ifdef CONFIG_RV
rv_task_fork(struct task_struct * p)2128 static void rv_task_fork(struct task_struct *p)
2129 {
2130 	int i;
2131 
2132 	for (i = 0; i < RV_PER_TASK_MONITORS; i++)
2133 		p->rv[i].da_mon.monitoring = false;
2134 }
2135 #else
2136 #define rv_task_fork(p) do {} while (0)
2137 #endif
2138 
2139 /*
2140  * This creates a new process as a copy of the old one,
2141  * but does not actually start it yet.
2142  *
2143  * It copies the registers, and all the appropriate
2144  * parts of the process environment (as per the clone
2145  * flags). The actual kick-off is left to the caller.
2146  */
copy_process(struct pid * pid,int trace,int node,struct kernel_clone_args * args)2147 __latent_entropy struct task_struct *copy_process(
2148 					struct pid *pid,
2149 					int trace,
2150 					int node,
2151 					struct kernel_clone_args *args)
2152 {
2153 	int pidfd = -1, retval;
2154 	struct task_struct *p;
2155 	struct multiprocess_signals delayed;
2156 	struct file *pidfile = NULL;
2157 	const u64 clone_flags = args->flags;
2158 	struct nsproxy *nsp = current->nsproxy;
2159 
2160 	/*
2161 	 * Don't allow sharing the root directory with processes in a different
2162 	 * namespace
2163 	 */
2164 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2165 		return ERR_PTR(-EINVAL);
2166 
2167 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2168 		return ERR_PTR(-EINVAL);
2169 
2170 	/*
2171 	 * Thread groups must share signals as well, and detached threads
2172 	 * can only be started up within the thread group.
2173 	 */
2174 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2175 		return ERR_PTR(-EINVAL);
2176 
2177 	/*
2178 	 * Shared signal handlers imply shared VM. By way of the above,
2179 	 * thread groups also imply shared VM. Blocking this case allows
2180 	 * for various simplifications in other code.
2181 	 */
2182 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2183 		return ERR_PTR(-EINVAL);
2184 
2185 	/*
2186 	 * Siblings of global init remain as zombies on exit since they are
2187 	 * not reaped by their parent (swapper). To solve this and to avoid
2188 	 * multi-rooted process trees, prevent global and container-inits
2189 	 * from creating siblings.
2190 	 */
2191 	if ((clone_flags & CLONE_PARENT) &&
2192 				current->signal->flags & SIGNAL_UNKILLABLE)
2193 		return ERR_PTR(-EINVAL);
2194 
2195 	/*
2196 	 * If the new process will be in a different pid or user namespace
2197 	 * do not allow it to share a thread group with the forking task.
2198 	 */
2199 	if (clone_flags & CLONE_THREAD) {
2200 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2201 		    (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2202 			return ERR_PTR(-EINVAL);
2203 	}
2204 
2205 	if (clone_flags & CLONE_PIDFD) {
2206 		/*
2207 		 * - CLONE_DETACHED is blocked so that we can potentially
2208 		 *   reuse it later for CLONE_PIDFD.
2209 		 */
2210 		if (clone_flags & CLONE_DETACHED)
2211 			return ERR_PTR(-EINVAL);
2212 	}
2213 
2214 	/*
2215 	 * Force any signals received before this point to be delivered
2216 	 * before the fork happens.  Collect up signals sent to multiple
2217 	 * processes that happen during the fork and delay them so that
2218 	 * they appear to happen after the fork.
2219 	 */
2220 	sigemptyset(&delayed.signal);
2221 	INIT_HLIST_NODE(&delayed.node);
2222 
2223 	spin_lock_irq(&current->sighand->siglock);
2224 	if (!(clone_flags & CLONE_THREAD))
2225 		hlist_add_head(&delayed.node, &current->signal->multiprocess);
2226 	recalc_sigpending();
2227 	spin_unlock_irq(&current->sighand->siglock);
2228 	retval = -ERESTARTNOINTR;
2229 	if (task_sigpending(current))
2230 		goto fork_out;
2231 
2232 	retval = -ENOMEM;
2233 	p = dup_task_struct(current, node);
2234 	if (!p)
2235 		goto fork_out;
2236 	p->flags &= ~PF_KTHREAD;
2237 	if (args->kthread)
2238 		p->flags |= PF_KTHREAD;
2239 	if (args->user_worker) {
2240 		/*
2241 		 * Mark us a user worker, and block any signal that isn't
2242 		 * fatal or STOP
2243 		 */
2244 		p->flags |= PF_USER_WORKER;
2245 		siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
2246 	}
2247 	if (args->io_thread)
2248 		p->flags |= PF_IO_WORKER;
2249 
2250 	if (args->name)
2251 		strscpy_pad(p->comm, args->name, sizeof(p->comm));
2252 
2253 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2254 	/*
2255 	 * Clear TID on mm_release()?
2256 	 */
2257 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2258 
2259 	ftrace_graph_init_task(p);
2260 
2261 	rt_mutex_init_task(p);
2262 
2263 	lockdep_assert_irqs_enabled();
2264 #ifdef CONFIG_PROVE_LOCKING
2265 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2266 #endif
2267 	retval = copy_creds(p, clone_flags);
2268 	if (retval < 0)
2269 		goto bad_fork_free;
2270 
2271 	retval = -EAGAIN;
2272 	if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2273 		if (p->real_cred->user != INIT_USER &&
2274 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2275 			goto bad_fork_cleanup_count;
2276 	}
2277 	current->flags &= ~PF_NPROC_EXCEEDED;
2278 
2279 	/*
2280 	 * If multiple threads are within copy_process(), then this check
2281 	 * triggers too late. This doesn't hurt, the check is only there
2282 	 * to stop root fork bombs.
2283 	 */
2284 	retval = -EAGAIN;
2285 	if (data_race(nr_threads >= max_threads))
2286 		goto bad_fork_cleanup_count;
2287 
2288 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
2289 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2290 	p->flags |= PF_FORKNOEXEC;
2291 	INIT_LIST_HEAD(&p->children);
2292 	INIT_LIST_HEAD(&p->sibling);
2293 	rcu_copy_process(p);
2294 	p->vfork_done = NULL;
2295 	spin_lock_init(&p->alloc_lock);
2296 
2297 	init_sigpending(&p->pending);
2298 
2299 	p->utime = p->stime = p->gtime = 0;
2300 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2301 	p->utimescaled = p->stimescaled = 0;
2302 #endif
2303 	prev_cputime_init(&p->prev_cputime);
2304 
2305 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2306 	seqcount_init(&p->vtime.seqcount);
2307 	p->vtime.starttime = 0;
2308 	p->vtime.state = VTIME_INACTIVE;
2309 #endif
2310 
2311 #ifdef CONFIG_IO_URING
2312 	p->io_uring = NULL;
2313 #endif
2314 
2315 	p->default_timer_slack_ns = current->timer_slack_ns;
2316 
2317 #ifdef CONFIG_PSI
2318 	p->psi_flags = 0;
2319 #endif
2320 
2321 	task_io_accounting_init(&p->ioac);
2322 	acct_clear_integrals(p);
2323 
2324 	posix_cputimers_init(&p->posix_cputimers);
2325 	tick_dep_init_task(p);
2326 
2327 	p->io_context = NULL;
2328 	audit_set_context(p, NULL);
2329 	cgroup_fork(p);
2330 	if (args->kthread) {
2331 		if (!set_kthread_struct(p))
2332 			goto bad_fork_cleanup_delayacct;
2333 	}
2334 #ifdef CONFIG_NUMA
2335 	p->mempolicy = mpol_dup(p->mempolicy);
2336 	if (IS_ERR(p->mempolicy)) {
2337 		retval = PTR_ERR(p->mempolicy);
2338 		p->mempolicy = NULL;
2339 		goto bad_fork_cleanup_delayacct;
2340 	}
2341 #endif
2342 #ifdef CONFIG_CPUSETS
2343 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2344 	seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2345 #endif
2346 #ifdef CONFIG_TRACE_IRQFLAGS
2347 	memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2348 	p->irqtrace.hardirq_disable_ip	= _THIS_IP_;
2349 	p->irqtrace.softirq_enable_ip	= _THIS_IP_;
2350 	p->softirqs_enabled		= 1;
2351 	p->softirq_context		= 0;
2352 #endif
2353 
2354 	p->pagefault_disabled = 0;
2355 
2356 #ifdef CONFIG_LOCKDEP
2357 	lockdep_init_task(p);
2358 #endif
2359 
2360 #ifdef CONFIG_DEBUG_MUTEXES
2361 	p->blocked_on = NULL; /* not blocked yet */
2362 #endif
2363 #ifdef CONFIG_BCACHE
2364 	p->sequential_io	= 0;
2365 	p->sequential_io_avg	= 0;
2366 #endif
2367 #ifdef CONFIG_BPF_SYSCALL
2368 	RCU_INIT_POINTER(p->bpf_storage, NULL);
2369 	p->bpf_ctx = NULL;
2370 #endif
2371 
2372 	/* Perform scheduler related setup. Assign this task to a CPU. */
2373 	retval = sched_fork(clone_flags, p);
2374 	if (retval)
2375 		goto bad_fork_cleanup_policy;
2376 
2377 	retval = perf_event_init_task(p, clone_flags);
2378 	if (retval)
2379 		goto bad_fork_sched_cancel_fork;
2380 	retval = audit_alloc(p);
2381 	if (retval)
2382 		goto bad_fork_cleanup_perf;
2383 	/* copy all the process information */
2384 	shm_init_task(p);
2385 	retval = security_task_alloc(p, clone_flags);
2386 	if (retval)
2387 		goto bad_fork_cleanup_audit;
2388 	retval = copy_semundo(clone_flags, p);
2389 	if (retval)
2390 		goto bad_fork_cleanup_security;
2391 	retval = copy_files(clone_flags, p, args->no_files);
2392 	if (retval)
2393 		goto bad_fork_cleanup_semundo;
2394 	retval = copy_fs(clone_flags, p);
2395 	if (retval)
2396 		goto bad_fork_cleanup_files;
2397 	retval = copy_sighand(clone_flags, p);
2398 	if (retval)
2399 		goto bad_fork_cleanup_fs;
2400 	retval = copy_signal(clone_flags, p);
2401 	if (retval)
2402 		goto bad_fork_cleanup_sighand;
2403 	retval = copy_mm(clone_flags, p);
2404 	if (retval)
2405 		goto bad_fork_cleanup_signal;
2406 	retval = copy_namespaces(clone_flags, p);
2407 	if (retval)
2408 		goto bad_fork_cleanup_mm;
2409 	retval = copy_io(clone_flags, p);
2410 	if (retval)
2411 		goto bad_fork_cleanup_namespaces;
2412 	retval = copy_thread(p, args);
2413 	if (retval)
2414 		goto bad_fork_cleanup_io;
2415 
2416 	stackleak_task_init(p);
2417 
2418 	if (pid != &init_struct_pid) {
2419 		pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2420 				args->set_tid_size);
2421 		if (IS_ERR(pid)) {
2422 			retval = PTR_ERR(pid);
2423 			goto bad_fork_cleanup_thread;
2424 		}
2425 	}
2426 
2427 	/*
2428 	 * This has to happen after we've potentially unshared the file
2429 	 * descriptor table (so that the pidfd doesn't leak into the child
2430 	 * if the fd table isn't shared).
2431 	 */
2432 	if (clone_flags & CLONE_PIDFD) {
2433 		int flags = (clone_flags & CLONE_THREAD) ? PIDFD_THREAD : 0;
2434 
2435 		/* Note that no task has been attached to @pid yet. */
2436 		retval = __pidfd_prepare(pid, flags, &pidfile);
2437 		if (retval < 0)
2438 			goto bad_fork_free_pid;
2439 		pidfd = retval;
2440 
2441 		retval = put_user(pidfd, args->pidfd);
2442 		if (retval)
2443 			goto bad_fork_put_pidfd;
2444 	}
2445 
2446 #ifdef CONFIG_BLOCK
2447 	p->plug = NULL;
2448 #endif
2449 	futex_init_task(p);
2450 
2451 	/*
2452 	 * sigaltstack should be cleared when sharing the same VM
2453 	 */
2454 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2455 		sas_ss_reset(p);
2456 
2457 	/*
2458 	 * Syscall tracing and stepping should be turned off in the
2459 	 * child regardless of CLONE_PTRACE.
2460 	 */
2461 	user_disable_single_step(p);
2462 	clear_task_syscall_work(p, SYSCALL_TRACE);
2463 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2464 	clear_task_syscall_work(p, SYSCALL_EMU);
2465 #endif
2466 	clear_tsk_latency_tracing(p);
2467 
2468 	/* ok, now we should be set up.. */
2469 	p->pid = pid_nr(pid);
2470 	if (clone_flags & CLONE_THREAD) {
2471 		p->group_leader = current->group_leader;
2472 		p->tgid = current->tgid;
2473 	} else {
2474 		p->group_leader = p;
2475 		p->tgid = p->pid;
2476 	}
2477 
2478 	p->nr_dirtied = 0;
2479 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2480 	p->dirty_paused_when = 0;
2481 
2482 	p->pdeath_signal = 0;
2483 	p->task_works = NULL;
2484 	clear_posix_cputimers_work(p);
2485 
2486 #ifdef CONFIG_KRETPROBES
2487 	p->kretprobe_instances.first = NULL;
2488 #endif
2489 #ifdef CONFIG_RETHOOK
2490 	p->rethooks.first = NULL;
2491 #endif
2492 
2493 	/*
2494 	 * Ensure that the cgroup subsystem policies allow the new process to be
2495 	 * forked. It should be noted that the new process's css_set can be changed
2496 	 * between here and cgroup_post_fork() if an organisation operation is in
2497 	 * progress.
2498 	 */
2499 	retval = cgroup_can_fork(p, args);
2500 	if (retval)
2501 		goto bad_fork_put_pidfd;
2502 
2503 	/*
2504 	 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2505 	 * the new task on the correct runqueue. All this *before* the task
2506 	 * becomes visible.
2507 	 *
2508 	 * This isn't part of ->can_fork() because while the re-cloning is
2509 	 * cgroup specific, it unconditionally needs to place the task on a
2510 	 * runqueue.
2511 	 */
2512 	retval = sched_cgroup_fork(p, args);
2513 	if (retval)
2514 		goto bad_fork_cancel_cgroup;
2515 
2516 	/*
2517 	 * From this point on we must avoid any synchronous user-space
2518 	 * communication until we take the tasklist-lock. In particular, we do
2519 	 * not want user-space to be able to predict the process start-time by
2520 	 * stalling fork(2) after we recorded the start_time but before it is
2521 	 * visible to the system.
2522 	 */
2523 
2524 	p->start_time = ktime_get_ns();
2525 	p->start_boottime = ktime_get_boottime_ns();
2526 
2527 	/*
2528 	 * Make it visible to the rest of the system, but dont wake it up yet.
2529 	 * Need tasklist lock for parent etc handling!
2530 	 */
2531 	write_lock_irq(&tasklist_lock);
2532 
2533 	/* CLONE_PARENT re-uses the old parent */
2534 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2535 		p->real_parent = current->real_parent;
2536 		p->parent_exec_id = current->parent_exec_id;
2537 		if (clone_flags & CLONE_THREAD)
2538 			p->exit_signal = -1;
2539 		else
2540 			p->exit_signal = current->group_leader->exit_signal;
2541 	} else {
2542 		p->real_parent = current;
2543 		p->parent_exec_id = current->self_exec_id;
2544 		p->exit_signal = args->exit_signal;
2545 	}
2546 
2547 	klp_copy_process(p);
2548 
2549 	sched_core_fork(p);
2550 
2551 	spin_lock(&current->sighand->siglock);
2552 
2553 	rv_task_fork(p);
2554 
2555 	rseq_fork(p, clone_flags);
2556 
2557 	/* Don't start children in a dying pid namespace */
2558 	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2559 		retval = -ENOMEM;
2560 		goto bad_fork_core_free;
2561 	}
2562 
2563 	/* Let kill terminate clone/fork in the middle */
2564 	if (fatal_signal_pending(current)) {
2565 		retval = -EINTR;
2566 		goto bad_fork_core_free;
2567 	}
2568 
2569 	/* No more failure paths after this point. */
2570 
2571 	/*
2572 	 * Copy seccomp details explicitly here, in case they were changed
2573 	 * before holding sighand lock.
2574 	 */
2575 	copy_seccomp(p);
2576 
2577 	init_task_pid_links(p);
2578 	if (likely(p->pid)) {
2579 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2580 
2581 		init_task_pid(p, PIDTYPE_PID, pid);
2582 		if (thread_group_leader(p)) {
2583 			init_task_pid(p, PIDTYPE_TGID, pid);
2584 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2585 			init_task_pid(p, PIDTYPE_SID, task_session(current));
2586 
2587 			if (is_child_reaper(pid)) {
2588 				ns_of_pid(pid)->child_reaper = p;
2589 				p->signal->flags |= SIGNAL_UNKILLABLE;
2590 			}
2591 			p->signal->shared_pending.signal = delayed.signal;
2592 			p->signal->tty = tty_kref_get(current->signal->tty);
2593 			/*
2594 			 * Inherit has_child_subreaper flag under the same
2595 			 * tasklist_lock with adding child to the process tree
2596 			 * for propagate_has_child_subreaper optimization.
2597 			 */
2598 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2599 							 p->real_parent->signal->is_child_subreaper;
2600 			list_add_tail(&p->sibling, &p->real_parent->children);
2601 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
2602 			attach_pid(p, PIDTYPE_TGID);
2603 			attach_pid(p, PIDTYPE_PGID);
2604 			attach_pid(p, PIDTYPE_SID);
2605 			__this_cpu_inc(process_counts);
2606 		} else {
2607 			current->signal->nr_threads++;
2608 			current->signal->quick_threads++;
2609 			atomic_inc(&current->signal->live);
2610 			refcount_inc(&current->signal->sigcnt);
2611 			task_join_group_stop(p);
2612 			list_add_tail_rcu(&p->thread_node,
2613 					  &p->signal->thread_head);
2614 		}
2615 		attach_pid(p, PIDTYPE_PID);
2616 		nr_threads++;
2617 	}
2618 	total_forks++;
2619 	hlist_del_init(&delayed.node);
2620 	spin_unlock(&current->sighand->siglock);
2621 	syscall_tracepoint_update(p);
2622 	write_unlock_irq(&tasklist_lock);
2623 
2624 	if (pidfile)
2625 		fd_install(pidfd, pidfile);
2626 
2627 	proc_fork_connector(p);
2628 	sched_post_fork(p);
2629 	cgroup_post_fork(p, args);
2630 	perf_event_fork(p);
2631 
2632 	trace_task_newtask(p, clone_flags);
2633 	uprobe_copy_process(p, clone_flags);
2634 	user_events_fork(p, clone_flags);
2635 
2636 	copy_oom_score_adj(clone_flags, p);
2637 
2638 	return p;
2639 
2640 bad_fork_core_free:
2641 	sched_core_free(p);
2642 	spin_unlock(&current->sighand->siglock);
2643 	write_unlock_irq(&tasklist_lock);
2644 bad_fork_cancel_cgroup:
2645 	cgroup_cancel_fork(p, args);
2646 bad_fork_put_pidfd:
2647 	if (clone_flags & CLONE_PIDFD) {
2648 		fput(pidfile);
2649 		put_unused_fd(pidfd);
2650 	}
2651 bad_fork_free_pid:
2652 	if (pid != &init_struct_pid)
2653 		free_pid(pid);
2654 bad_fork_cleanup_thread:
2655 	exit_thread(p);
2656 bad_fork_cleanup_io:
2657 	if (p->io_context)
2658 		exit_io_context(p);
2659 bad_fork_cleanup_namespaces:
2660 	exit_task_namespaces(p);
2661 bad_fork_cleanup_mm:
2662 	if (p->mm) {
2663 		mm_clear_owner(p->mm, p);
2664 		mmput(p->mm);
2665 	}
2666 bad_fork_cleanup_signal:
2667 	if (!(clone_flags & CLONE_THREAD))
2668 		free_signal_struct(p->signal);
2669 bad_fork_cleanup_sighand:
2670 	__cleanup_sighand(p->sighand);
2671 bad_fork_cleanup_fs:
2672 	exit_fs(p); /* blocking */
2673 bad_fork_cleanup_files:
2674 	exit_files(p); /* blocking */
2675 bad_fork_cleanup_semundo:
2676 	exit_sem(p);
2677 bad_fork_cleanup_security:
2678 	security_task_free(p);
2679 bad_fork_cleanup_audit:
2680 	audit_free(p);
2681 bad_fork_cleanup_perf:
2682 	perf_event_free_task(p);
2683 bad_fork_sched_cancel_fork:
2684 	sched_cancel_fork(p);
2685 bad_fork_cleanup_policy:
2686 	lockdep_free_task(p);
2687 #ifdef CONFIG_NUMA
2688 	mpol_put(p->mempolicy);
2689 #endif
2690 bad_fork_cleanup_delayacct:
2691 	delayacct_tsk_free(p);
2692 bad_fork_cleanup_count:
2693 	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2694 	exit_creds(p);
2695 bad_fork_free:
2696 	WRITE_ONCE(p->__state, TASK_DEAD);
2697 	exit_task_stack_account(p);
2698 	put_task_stack(p);
2699 	delayed_free_task(p);
2700 fork_out:
2701 	spin_lock_irq(&current->sighand->siglock);
2702 	hlist_del_init(&delayed.node);
2703 	spin_unlock_irq(&current->sighand->siglock);
2704 	return ERR_PTR(retval);
2705 }
2706 
init_idle_pids(struct task_struct * idle)2707 static inline void init_idle_pids(struct task_struct *idle)
2708 {
2709 	enum pid_type type;
2710 
2711 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2712 		INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2713 		init_task_pid(idle, type, &init_struct_pid);
2714 	}
2715 }
2716 
idle_dummy(void * dummy)2717 static int idle_dummy(void *dummy)
2718 {
2719 	/* This function is never called */
2720 	return 0;
2721 }
2722 
fork_idle(int cpu)2723 struct task_struct * __init fork_idle(int cpu)
2724 {
2725 	struct task_struct *task;
2726 	struct kernel_clone_args args = {
2727 		.flags		= CLONE_VM,
2728 		.fn		= &idle_dummy,
2729 		.fn_arg		= NULL,
2730 		.kthread	= 1,
2731 		.idle		= 1,
2732 	};
2733 
2734 	task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2735 	if (!IS_ERR(task)) {
2736 		init_idle_pids(task);
2737 		init_idle(task, cpu);
2738 	}
2739 
2740 	return task;
2741 }
2742 
2743 /*
2744  * This is like kernel_clone(), but shaved down and tailored to just
2745  * creating io_uring workers. It returns a created task, or an error pointer.
2746  * The returned task is inactive, and the caller must fire it up through
2747  * wake_up_new_task(p). All signals are blocked in the created task.
2748  */
create_io_thread(int (* fn)(void *),void * arg,int node)2749 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2750 {
2751 	unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2752 				CLONE_IO;
2753 	struct kernel_clone_args args = {
2754 		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2755 				    CLONE_UNTRACED) & ~CSIGNAL),
2756 		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2757 		.fn		= fn,
2758 		.fn_arg		= arg,
2759 		.io_thread	= 1,
2760 		.user_worker	= 1,
2761 	};
2762 
2763 	return copy_process(NULL, 0, node, &args);
2764 }
2765 
2766 /*
2767  *  Ok, this is the main fork-routine.
2768  *
2769  * It copies the process, and if successful kick-starts
2770  * it and waits for it to finish using the VM if required.
2771  *
2772  * args->exit_signal is expected to be checked for sanity by the caller.
2773  */
kernel_clone(struct kernel_clone_args * args)2774 pid_t kernel_clone(struct kernel_clone_args *args)
2775 {
2776 	u64 clone_flags = args->flags;
2777 	struct completion vfork;
2778 	struct pid *pid;
2779 	struct task_struct *p;
2780 	int trace = 0;
2781 	pid_t nr;
2782 
2783 	/*
2784 	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2785 	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2786 	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2787 	 * field in struct clone_args and it still doesn't make sense to have
2788 	 * them both point at the same memory location. Performing this check
2789 	 * here has the advantage that we don't need to have a separate helper
2790 	 * to check for legacy clone().
2791 	 */
2792 	if ((clone_flags & CLONE_PIDFD) &&
2793 	    (clone_flags & CLONE_PARENT_SETTID) &&
2794 	    (args->pidfd == args->parent_tid))
2795 		return -EINVAL;
2796 
2797 	/*
2798 	 * Determine whether and which event to report to ptracer.  When
2799 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2800 	 * requested, no event is reported; otherwise, report if the event
2801 	 * for the type of forking is enabled.
2802 	 */
2803 	if (!(clone_flags & CLONE_UNTRACED)) {
2804 		if (clone_flags & CLONE_VFORK)
2805 			trace = PTRACE_EVENT_VFORK;
2806 		else if (args->exit_signal != SIGCHLD)
2807 			trace = PTRACE_EVENT_CLONE;
2808 		else
2809 			trace = PTRACE_EVENT_FORK;
2810 
2811 		if (likely(!ptrace_event_enabled(current, trace)))
2812 			trace = 0;
2813 	}
2814 
2815 	p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2816 	add_latent_entropy();
2817 
2818 	if (IS_ERR(p))
2819 		return PTR_ERR(p);
2820 
2821 	/*
2822 	 * Do this prior waking up the new thread - the thread pointer
2823 	 * might get invalid after that point, if the thread exits quickly.
2824 	 */
2825 	trace_sched_process_fork(current, p);
2826 
2827 	pid = get_task_pid(p, PIDTYPE_PID);
2828 	nr = pid_vnr(pid);
2829 
2830 	if (clone_flags & CLONE_PARENT_SETTID)
2831 		put_user(nr, args->parent_tid);
2832 
2833 	if (clone_flags & CLONE_VFORK) {
2834 		p->vfork_done = &vfork;
2835 		init_completion(&vfork);
2836 		get_task_struct(p);
2837 	}
2838 
2839 	if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2840 		/* lock the task to synchronize with memcg migration */
2841 		task_lock(p);
2842 		lru_gen_add_mm(p->mm);
2843 		task_unlock(p);
2844 	}
2845 
2846 	wake_up_new_task(p);
2847 
2848 	/* forking complete and child started to run, tell ptracer */
2849 	if (unlikely(trace))
2850 		ptrace_event_pid(trace, pid);
2851 
2852 	if (clone_flags & CLONE_VFORK) {
2853 		if (!wait_for_vfork_done(p, &vfork))
2854 			ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2855 	}
2856 
2857 	put_pid(pid);
2858 	return nr;
2859 }
2860 
2861 /*
2862  * Create a kernel thread.
2863  */
kernel_thread(int (* fn)(void *),void * arg,const char * name,unsigned long flags)2864 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2865 		    unsigned long flags)
2866 {
2867 	struct kernel_clone_args args = {
2868 		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2869 				    CLONE_UNTRACED) & ~CSIGNAL),
2870 		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2871 		.fn		= fn,
2872 		.fn_arg		= arg,
2873 		.name		= name,
2874 		.kthread	= 1,
2875 	};
2876 
2877 	return kernel_clone(&args);
2878 }
2879 
2880 /*
2881  * Create a user mode thread.
2882  */
user_mode_thread(int (* fn)(void *),void * arg,unsigned long flags)2883 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2884 {
2885 	struct kernel_clone_args args = {
2886 		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2887 				    CLONE_UNTRACED) & ~CSIGNAL),
2888 		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2889 		.fn		= fn,
2890 		.fn_arg		= arg,
2891 	};
2892 
2893 	return kernel_clone(&args);
2894 }
2895 
2896 #ifdef __ARCH_WANT_SYS_FORK
SYSCALL_DEFINE0(fork)2897 SYSCALL_DEFINE0(fork)
2898 {
2899 #ifdef CONFIG_MMU
2900 	struct kernel_clone_args args = {
2901 		.exit_signal = SIGCHLD,
2902 	};
2903 
2904 	return kernel_clone(&args);
2905 #else
2906 	/* can not support in nommu mode */
2907 	return -EINVAL;
2908 #endif
2909 }
2910 #endif
2911 
2912 #ifdef __ARCH_WANT_SYS_VFORK
SYSCALL_DEFINE0(vfork)2913 SYSCALL_DEFINE0(vfork)
2914 {
2915 	struct kernel_clone_args args = {
2916 		.flags		= CLONE_VFORK | CLONE_VM,
2917 		.exit_signal	= SIGCHLD,
2918 	};
2919 
2920 	return kernel_clone(&args);
2921 }
2922 #endif
2923 
2924 #ifdef __ARCH_WANT_SYS_CLONE
2925 #ifdef CONFIG_CLONE_BACKWARDS
SYSCALL_DEFINE5(clone,unsigned long,clone_flags,unsigned long,newsp,int __user *,parent_tidptr,unsigned long,tls,int __user *,child_tidptr)2926 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2927 		 int __user *, parent_tidptr,
2928 		 unsigned long, tls,
2929 		 int __user *, child_tidptr)
2930 #elif defined(CONFIG_CLONE_BACKWARDS2)
2931 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2932 		 int __user *, parent_tidptr,
2933 		 int __user *, child_tidptr,
2934 		 unsigned long, tls)
2935 #elif defined(CONFIG_CLONE_BACKWARDS3)
2936 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2937 		int, stack_size,
2938 		int __user *, parent_tidptr,
2939 		int __user *, child_tidptr,
2940 		unsigned long, tls)
2941 #else
2942 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2943 		 int __user *, parent_tidptr,
2944 		 int __user *, child_tidptr,
2945 		 unsigned long, tls)
2946 #endif
2947 {
2948 	struct kernel_clone_args args = {
2949 		.flags		= (lower_32_bits(clone_flags) & ~CSIGNAL),
2950 		.pidfd		= parent_tidptr,
2951 		.child_tid	= child_tidptr,
2952 		.parent_tid	= parent_tidptr,
2953 		.exit_signal	= (lower_32_bits(clone_flags) & CSIGNAL),
2954 		.stack		= newsp,
2955 		.tls		= tls,
2956 	};
2957 
2958 	return kernel_clone(&args);
2959 }
2960 #endif
2961 
copy_clone_args_from_user(struct kernel_clone_args * kargs,struct clone_args __user * uargs,size_t usize)2962 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2963 					      struct clone_args __user *uargs,
2964 					      size_t usize)
2965 {
2966 	int err;
2967 	struct clone_args args;
2968 	pid_t *kset_tid = kargs->set_tid;
2969 
2970 	BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2971 		     CLONE_ARGS_SIZE_VER0);
2972 	BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2973 		     CLONE_ARGS_SIZE_VER1);
2974 	BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2975 		     CLONE_ARGS_SIZE_VER2);
2976 	BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2977 
2978 	if (unlikely(usize > PAGE_SIZE))
2979 		return -E2BIG;
2980 	if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2981 		return -EINVAL;
2982 
2983 	err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2984 	if (err)
2985 		return err;
2986 
2987 	if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2988 		return -EINVAL;
2989 
2990 	if (unlikely(!args.set_tid && args.set_tid_size > 0))
2991 		return -EINVAL;
2992 
2993 	if (unlikely(args.set_tid && args.set_tid_size == 0))
2994 		return -EINVAL;
2995 
2996 	/*
2997 	 * Verify that higher 32bits of exit_signal are unset and that
2998 	 * it is a valid signal
2999 	 */
3000 	if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
3001 		     !valid_signal(args.exit_signal)))
3002 		return -EINVAL;
3003 
3004 	if ((args.flags & CLONE_INTO_CGROUP) &&
3005 	    (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
3006 		return -EINVAL;
3007 
3008 	*kargs = (struct kernel_clone_args){
3009 		.flags		= args.flags,
3010 		.pidfd		= u64_to_user_ptr(args.pidfd),
3011 		.child_tid	= u64_to_user_ptr(args.child_tid),
3012 		.parent_tid	= u64_to_user_ptr(args.parent_tid),
3013 		.exit_signal	= args.exit_signal,
3014 		.stack		= args.stack,
3015 		.stack_size	= args.stack_size,
3016 		.tls		= args.tls,
3017 		.set_tid_size	= args.set_tid_size,
3018 		.cgroup		= args.cgroup,
3019 	};
3020 
3021 	if (args.set_tid &&
3022 		copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3023 			(kargs->set_tid_size * sizeof(pid_t))))
3024 		return -EFAULT;
3025 
3026 	kargs->set_tid = kset_tid;
3027 
3028 	return 0;
3029 }
3030 
3031 /**
3032  * clone3_stack_valid - check and prepare stack
3033  * @kargs: kernel clone args
3034  *
3035  * Verify that the stack arguments userspace gave us are sane.
3036  * In addition, set the stack direction for userspace since it's easy for us to
3037  * determine.
3038  */
clone3_stack_valid(struct kernel_clone_args * kargs)3039 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3040 {
3041 	if (kargs->stack == 0) {
3042 		if (kargs->stack_size > 0)
3043 			return false;
3044 	} else {
3045 		if (kargs->stack_size == 0)
3046 			return false;
3047 
3048 		if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3049 			return false;
3050 
3051 #if !defined(CONFIG_STACK_GROWSUP)
3052 		kargs->stack += kargs->stack_size;
3053 #endif
3054 	}
3055 
3056 	return true;
3057 }
3058 
clone3_args_valid(struct kernel_clone_args * kargs)3059 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3060 {
3061 	/* Verify that no unknown flags are passed along. */
3062 	if (kargs->flags &
3063 	    ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
3064 		return false;
3065 
3066 	/*
3067 	 * - make the CLONE_DETACHED bit reusable for clone3
3068 	 * - make the CSIGNAL bits reusable for clone3
3069 	 */
3070 	if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3071 		return false;
3072 
3073 	if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3074 	    (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3075 		return false;
3076 
3077 	if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3078 	    kargs->exit_signal)
3079 		return false;
3080 
3081 	if (!clone3_stack_valid(kargs))
3082 		return false;
3083 
3084 	return true;
3085 }
3086 
3087 /**
3088  * sys_clone3 - create a new process with specific properties
3089  * @uargs: argument structure
3090  * @size:  size of @uargs
3091  *
3092  * clone3() is the extensible successor to clone()/clone2().
3093  * It takes a struct as argument that is versioned by its size.
3094  *
3095  * Return: On success, a positive PID for the child process.
3096  *         On error, a negative errno number.
3097  */
SYSCALL_DEFINE2(clone3,struct clone_args __user *,uargs,size_t,size)3098 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3099 {
3100 	int err;
3101 
3102 	struct kernel_clone_args kargs;
3103 	pid_t set_tid[MAX_PID_NS_LEVEL];
3104 
3105 #ifdef __ARCH_BROKEN_SYS_CLONE3
3106 #warning clone3() entry point is missing, please fix
3107 	return -ENOSYS;
3108 #endif
3109 
3110 	kargs.set_tid = set_tid;
3111 
3112 	err = copy_clone_args_from_user(&kargs, uargs, size);
3113 	if (err)
3114 		return err;
3115 
3116 	if (!clone3_args_valid(&kargs))
3117 		return -EINVAL;
3118 
3119 	return kernel_clone(&kargs);
3120 }
3121 
walk_process_tree(struct task_struct * top,proc_visitor visitor,void * data)3122 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3123 {
3124 	struct task_struct *leader, *parent, *child;
3125 	int res;
3126 
3127 	read_lock(&tasklist_lock);
3128 	leader = top = top->group_leader;
3129 down:
3130 	for_each_thread(leader, parent) {
3131 		list_for_each_entry(child, &parent->children, sibling) {
3132 			res = visitor(child, data);
3133 			if (res) {
3134 				if (res < 0)
3135 					goto out;
3136 				leader = child;
3137 				goto down;
3138 			}
3139 up:
3140 			;
3141 		}
3142 	}
3143 
3144 	if (leader != top) {
3145 		child = leader;
3146 		parent = child->real_parent;
3147 		leader = parent->group_leader;
3148 		goto up;
3149 	}
3150 out:
3151 	read_unlock(&tasklist_lock);
3152 }
3153 
3154 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3155 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3156 #endif
3157 
sighand_ctor(void * data)3158 static void sighand_ctor(void *data)
3159 {
3160 	struct sighand_struct *sighand = data;
3161 
3162 	spin_lock_init(&sighand->siglock);
3163 	init_waitqueue_head(&sighand->signalfd_wqh);
3164 }
3165 
mm_cache_init(void)3166 void __init mm_cache_init(void)
3167 {
3168 	unsigned int mm_size;
3169 
3170 	/*
3171 	 * The mm_cpumask is located at the end of mm_struct, and is
3172 	 * dynamically sized based on the maximum CPU number this system
3173 	 * can have, taking hotplug into account (nr_cpu_ids).
3174 	 */
3175 	mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3176 
3177 	mm_cachep = kmem_cache_create_usercopy("mm_struct",
3178 			mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3179 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3180 			offsetof(struct mm_struct, saved_auxv),
3181 			sizeof_field(struct mm_struct, saved_auxv),
3182 			NULL);
3183 }
3184 
proc_caches_init(void)3185 void __init proc_caches_init(void)
3186 {
3187 	sighand_cachep = kmem_cache_create("sighand_cache",
3188 			sizeof(struct sighand_struct), 0,
3189 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3190 			SLAB_ACCOUNT, sighand_ctor);
3191 	signal_cachep = kmem_cache_create("signal_cache",
3192 			sizeof(struct signal_struct), 0,
3193 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3194 			NULL);
3195 	files_cachep = kmem_cache_create("files_cache",
3196 			sizeof(struct files_struct), 0,
3197 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3198 			NULL);
3199 	fs_cachep = kmem_cache_create("fs_cache",
3200 			sizeof(struct fs_struct), 0,
3201 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3202 			NULL);
3203 
3204 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
3205 #ifdef CONFIG_PER_VMA_LOCK
3206 	vma_lock_cachep = KMEM_CACHE(vma_lock, SLAB_PANIC|SLAB_ACCOUNT);
3207 #endif
3208 	mmap_init();
3209 	nsproxy_cache_init();
3210 }
3211 
3212 /*
3213  * Check constraints on flags passed to the unshare system call.
3214  */
check_unshare_flags(unsigned long unshare_flags)3215 static int check_unshare_flags(unsigned long unshare_flags)
3216 {
3217 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
3218 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3219 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
3220 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
3221 				CLONE_NEWTIME))
3222 		return -EINVAL;
3223 	/*
3224 	 * Not implemented, but pretend it works if there is nothing
3225 	 * to unshare.  Note that unsharing the address space or the
3226 	 * signal handlers also need to unshare the signal queues (aka
3227 	 * CLONE_THREAD).
3228 	 */
3229 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3230 		if (!thread_group_empty(current))
3231 			return -EINVAL;
3232 	}
3233 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3234 		if (refcount_read(&current->sighand->count) > 1)
3235 			return -EINVAL;
3236 	}
3237 	if (unshare_flags & CLONE_VM) {
3238 		if (!current_is_single_threaded())
3239 			return -EINVAL;
3240 	}
3241 
3242 	return 0;
3243 }
3244 
3245 /*
3246  * Unshare the filesystem structure if it is being shared
3247  */
unshare_fs(unsigned long unshare_flags,struct fs_struct ** new_fsp)3248 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3249 {
3250 	struct fs_struct *fs = current->fs;
3251 
3252 	if (!(unshare_flags & CLONE_FS) || !fs)
3253 		return 0;
3254 
3255 	/* don't need lock here; in the worst case we'll do useless copy */
3256 	if (fs->users == 1)
3257 		return 0;
3258 
3259 	*new_fsp = copy_fs_struct(fs);
3260 	if (!*new_fsp)
3261 		return -ENOMEM;
3262 
3263 	return 0;
3264 }
3265 
3266 /*
3267  * Unshare file descriptor table if it is being shared
3268  */
unshare_fd(unsigned long unshare_flags,struct files_struct ** new_fdp)3269 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
3270 {
3271 	struct files_struct *fd = current->files;
3272 
3273 	if ((unshare_flags & CLONE_FILES) &&
3274 	    (fd && atomic_read(&fd->count) > 1)) {
3275 		fd = dup_fd(fd, NULL);
3276 		if (IS_ERR(fd))
3277 			return PTR_ERR(fd);
3278 		*new_fdp = fd;
3279 	}
3280 
3281 	return 0;
3282 }
3283 
3284 /*
3285  * unshare allows a process to 'unshare' part of the process
3286  * context which was originally shared using clone.  copy_*
3287  * functions used by kernel_clone() cannot be used here directly
3288  * because they modify an inactive task_struct that is being
3289  * constructed. Here we are modifying the current, active,
3290  * task_struct.
3291  */
ksys_unshare(unsigned long unshare_flags)3292 int ksys_unshare(unsigned long unshare_flags)
3293 {
3294 	struct fs_struct *fs, *new_fs = NULL;
3295 	struct files_struct *new_fd = NULL;
3296 	struct cred *new_cred = NULL;
3297 	struct nsproxy *new_nsproxy = NULL;
3298 	int do_sysvsem = 0;
3299 	int err;
3300 
3301 	/*
3302 	 * If unsharing a user namespace must also unshare the thread group
3303 	 * and unshare the filesystem root and working directories.
3304 	 */
3305 	if (unshare_flags & CLONE_NEWUSER)
3306 		unshare_flags |= CLONE_THREAD | CLONE_FS;
3307 	/*
3308 	 * If unsharing vm, must also unshare signal handlers.
3309 	 */
3310 	if (unshare_flags & CLONE_VM)
3311 		unshare_flags |= CLONE_SIGHAND;
3312 	/*
3313 	 * If unsharing a signal handlers, must also unshare the signal queues.
3314 	 */
3315 	if (unshare_flags & CLONE_SIGHAND)
3316 		unshare_flags |= CLONE_THREAD;
3317 	/*
3318 	 * If unsharing namespace, must also unshare filesystem information.
3319 	 */
3320 	if (unshare_flags & CLONE_NEWNS)
3321 		unshare_flags |= CLONE_FS;
3322 
3323 	err = check_unshare_flags(unshare_flags);
3324 	if (err)
3325 		goto bad_unshare_out;
3326 	/*
3327 	 * CLONE_NEWIPC must also detach from the undolist: after switching
3328 	 * to a new ipc namespace, the semaphore arrays from the old
3329 	 * namespace are unreachable.
3330 	 */
3331 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3332 		do_sysvsem = 1;
3333 	err = unshare_fs(unshare_flags, &new_fs);
3334 	if (err)
3335 		goto bad_unshare_out;
3336 	err = unshare_fd(unshare_flags, &new_fd);
3337 	if (err)
3338 		goto bad_unshare_cleanup_fs;
3339 	err = unshare_userns(unshare_flags, &new_cred);
3340 	if (err)
3341 		goto bad_unshare_cleanup_fd;
3342 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3343 					 new_cred, new_fs);
3344 	if (err)
3345 		goto bad_unshare_cleanup_cred;
3346 
3347 	if (new_cred) {
3348 		err = set_cred_ucounts(new_cred);
3349 		if (err)
3350 			goto bad_unshare_cleanup_cred;
3351 	}
3352 
3353 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3354 		if (do_sysvsem) {
3355 			/*
3356 			 * CLONE_SYSVSEM is equivalent to sys_exit().
3357 			 */
3358 			exit_sem(current);
3359 		}
3360 		if (unshare_flags & CLONE_NEWIPC) {
3361 			/* Orphan segments in old ns (see sem above). */
3362 			exit_shm(current);
3363 			shm_init_task(current);
3364 		}
3365 
3366 		if (new_nsproxy)
3367 			switch_task_namespaces(current, new_nsproxy);
3368 
3369 		task_lock(current);
3370 
3371 		if (new_fs) {
3372 			fs = current->fs;
3373 			spin_lock(&fs->lock);
3374 			current->fs = new_fs;
3375 			if (--fs->users)
3376 				new_fs = NULL;
3377 			else
3378 				new_fs = fs;
3379 			spin_unlock(&fs->lock);
3380 		}
3381 
3382 		if (new_fd)
3383 			swap(current->files, new_fd);
3384 
3385 		task_unlock(current);
3386 
3387 		if (new_cred) {
3388 			/* Install the new user namespace */
3389 			commit_creds(new_cred);
3390 			new_cred = NULL;
3391 		}
3392 	}
3393 
3394 	perf_event_namespaces(current);
3395 
3396 bad_unshare_cleanup_cred:
3397 	if (new_cred)
3398 		put_cred(new_cred);
3399 bad_unshare_cleanup_fd:
3400 	if (new_fd)
3401 		put_files_struct(new_fd);
3402 
3403 bad_unshare_cleanup_fs:
3404 	if (new_fs)
3405 		free_fs_struct(new_fs);
3406 
3407 bad_unshare_out:
3408 	return err;
3409 }
3410 
SYSCALL_DEFINE1(unshare,unsigned long,unshare_flags)3411 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3412 {
3413 	return ksys_unshare(unshare_flags);
3414 }
3415 
3416 /*
3417  *	Helper to unshare the files of the current task.
3418  *	We don't want to expose copy_files internals to
3419  *	the exec layer of the kernel.
3420  */
3421 
unshare_files(void)3422 int unshare_files(void)
3423 {
3424 	struct task_struct *task = current;
3425 	struct files_struct *old, *copy = NULL;
3426 	int error;
3427 
3428 	error = unshare_fd(CLONE_FILES, &copy);
3429 	if (error || !copy)
3430 		return error;
3431 
3432 	old = task->files;
3433 	task_lock(task);
3434 	task->files = copy;
3435 	task_unlock(task);
3436 	put_files_struct(old);
3437 	return 0;
3438 }
3439 
sysctl_max_threads(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)3440 int sysctl_max_threads(const struct ctl_table *table, int write,
3441 		       void *buffer, size_t *lenp, loff_t *ppos)
3442 {
3443 	struct ctl_table t;
3444 	int ret;
3445 	int threads = max_threads;
3446 	int min = 1;
3447 	int max = MAX_THREADS;
3448 
3449 	t = *table;
3450 	t.data = &threads;
3451 	t.extra1 = &min;
3452 	t.extra2 = &max;
3453 
3454 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3455 	if (ret || !write)
3456 		return ret;
3457 
3458 	max_threads = threads;
3459 
3460 	return 0;
3461 }
3462