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