• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/mm.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/kaiser.h>
62 #include <linux/delayacct.h>
63 #include <linux/taskstats_kern.h>
64 #include <linux/random.h>
65 #include <linux/tty.h>
66 #include <linux/blkdev.h>
67 #include <linux/fs_struct.h>
68 #include <linux/magic.h>
69 #include <linux/perf_event.h>
70 #include <linux/posix-timers.h>
71 #include <linux/user-return-notifier.h>
72 #include <linux/oom.h>
73 #include <linux/khugepaged.h>
74 #include <linux/signalfd.h>
75 #include <linux/uprobes.h>
76 #include <linux/aio.h>
77 #include <linux/compiler.h>
78 #include <linux/sysctl.h>
79 #include <linux/kcov.h>
80 
81 #include <asm/pgtable.h>
82 #include <asm/pgalloc.h>
83 #include <asm/uaccess.h>
84 #include <asm/mmu_context.h>
85 #include <asm/cacheflush.h>
86 #include <asm/tlbflush.h>
87 
88 #include <trace/events/sched.h>
89 
90 #define CREATE_TRACE_POINTS
91 #include <trace/events/task.h>
92 
93 /*
94  * Minimum number of threads to boot the kernel
95  */
96 #define MIN_THREADS 20
97 
98 /*
99  * Maximum number of threads
100  */
101 #define MAX_THREADS FUTEX_TID_MASK
102 
103 /*
104  * Protected counters by write_lock_irq(&tasklist_lock)
105  */
106 unsigned long total_forks;	/* Handle normal Linux uptimes. */
107 int nr_threads;			/* The idle threads do not count.. */
108 
109 int max_threads;		/* tunable limit on nr_threads */
110 
111 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
112 
113 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
114 
115 #ifdef CONFIG_PROVE_RCU
lockdep_tasklist_lock_is_held(void)116 int lockdep_tasklist_lock_is_held(void)
117 {
118 	return lockdep_is_held(&tasklist_lock);
119 }
120 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
121 #endif /* #ifdef CONFIG_PROVE_RCU */
122 
nr_processes(void)123 int nr_processes(void)
124 {
125 	int cpu;
126 	int total = 0;
127 
128 	for_each_possible_cpu(cpu)
129 		total += per_cpu(process_counts, cpu);
130 
131 	return total;
132 }
133 
arch_release_task_struct(struct task_struct * tsk)134 void __weak arch_release_task_struct(struct task_struct *tsk)
135 {
136 }
137 
138 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
139 static struct kmem_cache *task_struct_cachep;
140 
alloc_task_struct_node(int node)141 static inline struct task_struct *alloc_task_struct_node(int node)
142 {
143 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
144 }
145 
free_task_struct(struct task_struct * tsk)146 static inline void free_task_struct(struct task_struct *tsk)
147 {
148 	kmem_cache_free(task_struct_cachep, tsk);
149 }
150 #endif
151 
arch_release_thread_stack(unsigned long * stack)152 void __weak arch_release_thread_stack(unsigned long *stack)
153 {
154 }
155 
156 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
157 
158 /*
159  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
160  * kmemcache based allocator.
161  */
162 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
163 
164 #ifdef CONFIG_VMAP_STACK
165 /*
166  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
167  * flush.  Try to minimize the number of calls by caching stacks.
168  */
169 #define NR_CACHED_STACKS 2
170 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
171 #endif
172 
alloc_thread_stack_node(struct task_struct * tsk,int node)173 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
174 {
175 #ifdef CONFIG_VMAP_STACK
176 	void *stack;
177 	int i;
178 
179 	local_irq_disable();
180 	for (i = 0; i < NR_CACHED_STACKS; i++) {
181 		struct vm_struct *s = this_cpu_read(cached_stacks[i]);
182 
183 		if (!s)
184 			continue;
185 		this_cpu_write(cached_stacks[i], NULL);
186 
187 		tsk->stack_vm_area = s;
188 		local_irq_enable();
189 		return s->addr;
190 	}
191 	local_irq_enable();
192 
193 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
194 				     VMALLOC_START, VMALLOC_END,
195 				     THREADINFO_GFP | __GFP_HIGHMEM,
196 				     PAGE_KERNEL,
197 				     0, node, __builtin_return_address(0));
198 
199 	/*
200 	 * We can't call find_vm_area() in interrupt context, and
201 	 * free_thread_stack() can be called in interrupt context,
202 	 * so cache the vm_struct.
203 	 */
204 	if (stack)
205 		tsk->stack_vm_area = find_vm_area(stack);
206 	return stack;
207 #else
208 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
209 					     THREAD_SIZE_ORDER);
210 
211 	return page ? page_address(page) : NULL;
212 #endif
213 }
214 
free_thread_stack(struct task_struct * tsk)215 static inline void free_thread_stack(struct task_struct *tsk)
216 {
217 	kaiser_unmap_thread_stack(tsk->stack);
218 #ifdef CONFIG_VMAP_STACK
219 	if (task_stack_vm_area(tsk)) {
220 		unsigned long flags;
221 		int i;
222 
223 		local_irq_save(flags);
224 		for (i = 0; i < NR_CACHED_STACKS; i++) {
225 			if (this_cpu_read(cached_stacks[i]))
226 				continue;
227 
228 			this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
229 			local_irq_restore(flags);
230 			return;
231 		}
232 		local_irq_restore(flags);
233 
234 		vfree(tsk->stack);
235 		return;
236 	}
237 #endif
238 
239 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
240 }
241 # else
242 static struct kmem_cache *thread_stack_cache;
243 
alloc_thread_stack_node(struct task_struct * tsk,int node)244 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
245 						  int node)
246 {
247 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
248 }
249 
free_thread_stack(struct task_struct * tsk)250 static void free_thread_stack(struct task_struct *tsk)
251 {
252 	kmem_cache_free(thread_stack_cache, tsk->stack);
253 }
254 
thread_stack_cache_init(void)255 void thread_stack_cache_init(void)
256 {
257 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
258 					      THREAD_SIZE, 0, NULL);
259 	BUG_ON(thread_stack_cache == NULL);
260 }
261 # endif
262 #endif
263 
264 /* SLAB cache for signal_struct structures (tsk->signal) */
265 static struct kmem_cache *signal_cachep;
266 
267 /* SLAB cache for sighand_struct structures (tsk->sighand) */
268 struct kmem_cache *sighand_cachep;
269 
270 /* SLAB cache for files_struct structures (tsk->files) */
271 struct kmem_cache *files_cachep;
272 
273 /* SLAB cache for fs_struct structures (tsk->fs) */
274 struct kmem_cache *fs_cachep;
275 
276 /* SLAB cache for vm_area_struct structures */
277 struct kmem_cache *vm_area_cachep;
278 
279 /* SLAB cache for mm_struct structures (tsk->mm) */
280 static struct kmem_cache *mm_cachep;
281 
account_kernel_stack(struct task_struct * tsk,int account)282 static void account_kernel_stack(struct task_struct *tsk, int account)
283 {
284 	void *stack = task_stack_page(tsk);
285 	struct vm_struct *vm = task_stack_vm_area(tsk);
286 
287 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
288 
289 	if (vm) {
290 		int i;
291 
292 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
293 
294 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
295 			mod_zone_page_state(page_zone(vm->pages[i]),
296 					    NR_KERNEL_STACK_KB,
297 					    PAGE_SIZE / 1024 * account);
298 		}
299 
300 		/* All stack pages belong to the same memcg. */
301 		memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
302 					    account * (THREAD_SIZE / 1024));
303 	} else {
304 		/*
305 		 * All stack pages are in the same zone and belong to the
306 		 * same memcg.
307 		 */
308 		struct page *first_page = virt_to_page(stack);
309 
310 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
311 				    THREAD_SIZE / 1024 * account);
312 
313 		memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
314 					    account * (THREAD_SIZE / 1024));
315 	}
316 }
317 
release_task_stack(struct task_struct * tsk)318 static void release_task_stack(struct task_struct *tsk)
319 {
320 	if (WARN_ON(tsk->state != TASK_DEAD))
321 		return;  /* Better to leak the stack than to free prematurely */
322 
323 	account_kernel_stack(tsk, -1);
324 	arch_release_thread_stack(tsk->stack);
325 	free_thread_stack(tsk);
326 	tsk->stack = NULL;
327 #ifdef CONFIG_VMAP_STACK
328 	tsk->stack_vm_area = NULL;
329 #endif
330 }
331 
332 #ifdef CONFIG_THREAD_INFO_IN_TASK
put_task_stack(struct task_struct * tsk)333 void put_task_stack(struct task_struct *tsk)
334 {
335 	if (atomic_dec_and_test(&tsk->stack_refcount))
336 		release_task_stack(tsk);
337 }
338 #endif
339 
free_task(struct task_struct * tsk)340 void free_task(struct task_struct *tsk)
341 {
342 #ifndef CONFIG_THREAD_INFO_IN_TASK
343 	/*
344 	 * The task is finally done with both the stack and thread_info,
345 	 * so free both.
346 	 */
347 	release_task_stack(tsk);
348 #else
349 	/*
350 	 * If the task had a separate stack allocation, it should be gone
351 	 * by now.
352 	 */
353 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
354 #endif
355 	rt_mutex_debug_task_free(tsk);
356 	ftrace_graph_exit_task(tsk);
357 	put_seccomp_filter(tsk);
358 	arch_release_task_struct(tsk);
359 	free_task_struct(tsk);
360 }
361 EXPORT_SYMBOL(free_task);
362 
free_signal_struct(struct signal_struct * sig)363 static inline void free_signal_struct(struct signal_struct *sig)
364 {
365 	taskstats_tgid_free(sig);
366 	sched_autogroup_exit(sig);
367 	/*
368 	 * __mmdrop is not safe to call from softirq context on x86 due to
369 	 * pgd_dtor so postpone it to the async context
370 	 */
371 	if (sig->oom_mm)
372 		mmdrop_async(sig->oom_mm);
373 	kmem_cache_free(signal_cachep, sig);
374 }
375 
put_signal_struct(struct signal_struct * sig)376 static inline void put_signal_struct(struct signal_struct *sig)
377 {
378 	if (atomic_dec_and_test(&sig->sigcnt))
379 		free_signal_struct(sig);
380 }
381 
__put_task_struct(struct task_struct * tsk)382 void __put_task_struct(struct task_struct *tsk)
383 {
384 	WARN_ON(!tsk->exit_state);
385 	WARN_ON(atomic_read(&tsk->usage));
386 	WARN_ON(tsk == current);
387 
388 	cgroup_free(tsk);
389 	task_numa_free(tsk);
390 	security_task_free(tsk);
391 	exit_creds(tsk);
392 	delayacct_tsk_free(tsk);
393 	put_signal_struct(tsk->signal);
394 
395 	if (!profile_handoff_task(tsk))
396 		free_task(tsk);
397 }
398 EXPORT_SYMBOL_GPL(__put_task_struct);
399 
arch_task_cache_init(void)400 void __init __weak arch_task_cache_init(void) { }
401 
402 /*
403  * set_max_threads
404  */
set_max_threads(unsigned int max_threads_suggested)405 static void set_max_threads(unsigned int max_threads_suggested)
406 {
407 	u64 threads;
408 
409 	/*
410 	 * The number of threads shall be limited such that the thread
411 	 * structures may only consume a small part of the available memory.
412 	 */
413 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
414 		threads = MAX_THREADS;
415 	else
416 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
417 				    (u64) THREAD_SIZE * 8UL);
418 
419 	if (threads > max_threads_suggested)
420 		threads = max_threads_suggested;
421 
422 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
423 }
424 
425 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
426 /* Initialized by the architecture: */
427 int arch_task_struct_size __read_mostly;
428 #endif
429 
fork_init(void)430 void __init fork_init(void)
431 {
432 	int i;
433 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
434 #ifndef ARCH_MIN_TASKALIGN
435 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
436 #endif
437 	/* create a slab on which task_structs can be allocated */
438 	task_struct_cachep = kmem_cache_create("task_struct",
439 			arch_task_struct_size, ARCH_MIN_TASKALIGN,
440 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
441 #endif
442 
443 	/* do the arch specific task caches init */
444 	arch_task_cache_init();
445 
446 	set_max_threads(MAX_THREADS);
447 
448 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
449 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
450 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
451 		init_task.signal->rlim[RLIMIT_NPROC];
452 
453 	for (i = 0; i < UCOUNT_COUNTS; i++) {
454 		init_user_ns.ucount_max[i] = max_threads/2;
455 	}
456 }
457 
arch_dup_task_struct(struct task_struct * dst,struct task_struct * src)458 int __weak arch_dup_task_struct(struct task_struct *dst,
459 					       struct task_struct *src)
460 {
461 	*dst = *src;
462 	return 0;
463 }
464 
set_task_stack_end_magic(struct task_struct * tsk)465 void set_task_stack_end_magic(struct task_struct *tsk)
466 {
467 	unsigned long *stackend;
468 
469 	stackend = end_of_stack(tsk);
470 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
471 }
472 
dup_task_struct(struct task_struct * orig,int node)473 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
474 {
475 	struct task_struct *tsk;
476 	unsigned long *stack;
477 	struct vm_struct *stack_vm_area;
478 	int err;
479 
480 	if (node == NUMA_NO_NODE)
481 		node = tsk_fork_get_node(orig);
482 	tsk = alloc_task_struct_node(node);
483 	if (!tsk)
484 		return NULL;
485 
486 	stack = alloc_thread_stack_node(tsk, node);
487 	if (!stack)
488 		goto free_tsk;
489 
490 	stack_vm_area = task_stack_vm_area(tsk);
491 
492 	err = arch_dup_task_struct(tsk, orig);
493 
494 	/*
495 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
496 	 * sure they're properly initialized before using any stack-related
497 	 * functions again.
498 	 */
499 	tsk->stack = stack;
500 
501 	err= kaiser_map_thread_stack(tsk->stack);
502 	if (err)
503 		goto free_stack;
504 #ifdef CONFIG_VMAP_STACK
505 	tsk->stack_vm_area = stack_vm_area;
506 #endif
507 #ifdef CONFIG_THREAD_INFO_IN_TASK
508 	atomic_set(&tsk->stack_refcount, 1);
509 #endif
510 
511 	if (err)
512 		goto free_stack;
513 
514 #ifdef CONFIG_SECCOMP
515 	/*
516 	 * We must handle setting up seccomp filters once we're under
517 	 * the sighand lock in case orig has changed between now and
518 	 * then. Until then, filter must be NULL to avoid messing up
519 	 * the usage counts on the error path calling free_task.
520 	 */
521 	tsk->seccomp.filter = NULL;
522 #endif
523 
524 	setup_thread_stack(tsk, orig);
525 	clear_user_return_notifier(tsk);
526 	clear_tsk_need_resched(tsk);
527 	set_task_stack_end_magic(tsk);
528 
529 #ifdef CONFIG_CC_STACKPROTECTOR
530 	tsk->stack_canary = get_random_long();
531 #endif
532 
533 	/*
534 	 * One for us, one for whoever does the "release_task()" (usually
535 	 * parent)
536 	 */
537 	atomic_set(&tsk->usage, 2);
538 #ifdef CONFIG_BLK_DEV_IO_TRACE
539 	tsk->btrace_seq = 0;
540 #endif
541 	tsk->splice_pipe = NULL;
542 	tsk->task_frag.page = NULL;
543 	tsk->wake_q.next = NULL;
544 
545 	account_kernel_stack(tsk, 1);
546 
547 	kcov_task_init(tsk);
548 
549 	return tsk;
550 
551 free_stack:
552 	free_thread_stack(tsk);
553 free_tsk:
554 	free_task_struct(tsk);
555 	return NULL;
556 }
557 
558 #ifdef CONFIG_MMU
dup_mmap(struct mm_struct * mm,struct mm_struct * oldmm)559 static __latent_entropy int dup_mmap(struct mm_struct *mm,
560 					struct mm_struct *oldmm)
561 {
562 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
563 	struct rb_node **rb_link, *rb_parent;
564 	int retval;
565 	unsigned long charge;
566 
567 	uprobe_start_dup_mmap();
568 	if (down_write_killable(&oldmm->mmap_sem)) {
569 		retval = -EINTR;
570 		goto fail_uprobe_end;
571 	}
572 	flush_cache_dup_mm(oldmm);
573 	uprobe_dup_mmap(oldmm, mm);
574 	/*
575 	 * Not linked in yet - no deadlock potential:
576 	 */
577 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
578 
579 	/* No ordering required: file already has been exposed. */
580 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
581 
582 	mm->total_vm = oldmm->total_vm;
583 	mm->data_vm = oldmm->data_vm;
584 	mm->exec_vm = oldmm->exec_vm;
585 	mm->stack_vm = oldmm->stack_vm;
586 
587 	rb_link = &mm->mm_rb.rb_node;
588 	rb_parent = NULL;
589 	pprev = &mm->mmap;
590 	retval = ksm_fork(mm, oldmm);
591 	if (retval)
592 		goto out;
593 	retval = khugepaged_fork(mm, oldmm);
594 	if (retval)
595 		goto out;
596 
597 	prev = NULL;
598 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
599 		struct file *file;
600 
601 		if (mpnt->vm_flags & VM_DONTCOPY) {
602 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
603 			continue;
604 		}
605 		charge = 0;
606 		if (mpnt->vm_flags & VM_ACCOUNT) {
607 			unsigned long len = vma_pages(mpnt);
608 
609 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
610 				goto fail_nomem;
611 			charge = len;
612 		}
613 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
614 		if (!tmp)
615 			goto fail_nomem;
616 		*tmp = *mpnt;
617 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
618 		retval = vma_dup_policy(mpnt, tmp);
619 		if (retval)
620 			goto fail_nomem_policy;
621 		tmp->vm_mm = mm;
622 		if (anon_vma_fork(tmp, mpnt))
623 			goto fail_nomem_anon_vma_fork;
624 		tmp->vm_flags &=
625 			~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
626 		tmp->vm_next = tmp->vm_prev = NULL;
627 		tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
628 		file = tmp->vm_file;
629 		if (file) {
630 			struct inode *inode = file_inode(file);
631 			struct address_space *mapping = file->f_mapping;
632 
633 			get_file(file);
634 			if (tmp->vm_flags & VM_DENYWRITE)
635 				atomic_dec(&inode->i_writecount);
636 			i_mmap_lock_write(mapping);
637 			if (tmp->vm_flags & VM_SHARED)
638 				atomic_inc(&mapping->i_mmap_writable);
639 			flush_dcache_mmap_lock(mapping);
640 			/* insert tmp into the share list, just after mpnt */
641 			vma_interval_tree_insert_after(tmp, mpnt,
642 					&mapping->i_mmap);
643 			flush_dcache_mmap_unlock(mapping);
644 			i_mmap_unlock_write(mapping);
645 		}
646 
647 		/*
648 		 * Clear hugetlb-related page reserves for children. This only
649 		 * affects MAP_PRIVATE mappings. Faults generated by the child
650 		 * are not guaranteed to succeed, even if read-only
651 		 */
652 		if (is_vm_hugetlb_page(tmp))
653 			reset_vma_resv_huge_pages(tmp);
654 
655 		/*
656 		 * Link in the new vma and copy the page table entries.
657 		 */
658 		*pprev = tmp;
659 		pprev = &tmp->vm_next;
660 		tmp->vm_prev = prev;
661 		prev = tmp;
662 
663 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
664 		rb_link = &tmp->vm_rb.rb_right;
665 		rb_parent = &tmp->vm_rb;
666 
667 		mm->map_count++;
668 		retval = copy_page_range(mm, oldmm, mpnt);
669 
670 		if (tmp->vm_ops && tmp->vm_ops->open)
671 			tmp->vm_ops->open(tmp);
672 
673 		if (retval)
674 			goto out;
675 	}
676 	/* a new mm has just been created */
677 	arch_dup_mmap(oldmm, mm);
678 	retval = 0;
679 out:
680 	up_write(&mm->mmap_sem);
681 	flush_tlb_mm(oldmm);
682 	up_write(&oldmm->mmap_sem);
683 fail_uprobe_end:
684 	uprobe_end_dup_mmap();
685 	return retval;
686 fail_nomem_anon_vma_fork:
687 	mpol_put(vma_policy(tmp));
688 fail_nomem_policy:
689 	kmem_cache_free(vm_area_cachep, tmp);
690 fail_nomem:
691 	retval = -ENOMEM;
692 	vm_unacct_memory(charge);
693 	goto out;
694 }
695 
mm_alloc_pgd(struct mm_struct * mm)696 static inline int mm_alloc_pgd(struct mm_struct *mm)
697 {
698 	mm->pgd = pgd_alloc(mm);
699 	if (unlikely(!mm->pgd))
700 		return -ENOMEM;
701 	return 0;
702 }
703 
mm_free_pgd(struct mm_struct * mm)704 static inline void mm_free_pgd(struct mm_struct *mm)
705 {
706 	pgd_free(mm, mm->pgd);
707 }
708 #else
dup_mmap(struct mm_struct * mm,struct mm_struct * oldmm)709 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
710 {
711 	down_write(&oldmm->mmap_sem);
712 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
713 	up_write(&oldmm->mmap_sem);
714 	return 0;
715 }
716 #define mm_alloc_pgd(mm)	(0)
717 #define mm_free_pgd(mm)
718 #endif /* CONFIG_MMU */
719 
720 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
721 
722 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
723 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
724 
725 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
726 
coredump_filter_setup(char * s)727 static int __init coredump_filter_setup(char *s)
728 {
729 	default_dump_filter =
730 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
731 		MMF_DUMP_FILTER_MASK;
732 	return 1;
733 }
734 
735 __setup("coredump_filter=", coredump_filter_setup);
736 
737 #include <linux/init_task.h>
738 
mm_init_aio(struct mm_struct * mm)739 static void mm_init_aio(struct mm_struct *mm)
740 {
741 #ifdef CONFIG_AIO
742 	spin_lock_init(&mm->ioctx_lock);
743 	mm->ioctx_table = NULL;
744 #endif
745 }
746 
mm_init_owner(struct mm_struct * mm,struct task_struct * p)747 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
748 {
749 #ifdef CONFIG_MEMCG
750 	mm->owner = p;
751 #endif
752 }
753 
mm_init_uprobes_state(struct mm_struct * mm)754 static void mm_init_uprobes_state(struct mm_struct *mm)
755 {
756 #ifdef CONFIG_UPROBES
757 	mm->uprobes_state.xol_area = NULL;
758 #endif
759 }
760 
mm_init(struct mm_struct * mm,struct task_struct * p,struct user_namespace * user_ns)761 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
762 	struct user_namespace *user_ns)
763 {
764 	mm->mmap = NULL;
765 	mm->mm_rb = RB_ROOT;
766 	mm->vmacache_seqnum = 0;
767 	atomic_set(&mm->mm_users, 1);
768 	atomic_set(&mm->mm_count, 1);
769 	init_rwsem(&mm->mmap_sem);
770 	INIT_LIST_HEAD(&mm->mmlist);
771 	mm->core_state = NULL;
772 	atomic_long_set(&mm->nr_ptes, 0);
773 	mm_nr_pmds_init(mm);
774 	mm->map_count = 0;
775 	mm->locked_vm = 0;
776 	mm->pinned_vm = 0;
777 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
778 	spin_lock_init(&mm->page_table_lock);
779 	mm_init_cpumask(mm);
780 	mm_init_aio(mm);
781 	mm_init_owner(mm, p);
782 	RCU_INIT_POINTER(mm->exe_file, NULL);
783 	mmu_notifier_mm_init(mm);
784 	clear_tlb_flush_pending(mm);
785 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
786 	mm->pmd_huge_pte = NULL;
787 #endif
788 	mm_init_uprobes_state(mm);
789 
790 	if (current->mm) {
791 		mm->flags = current->mm->flags & MMF_INIT_MASK;
792 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
793 	} else {
794 		mm->flags = default_dump_filter;
795 		mm->def_flags = 0;
796 	}
797 
798 	if (mm_alloc_pgd(mm))
799 		goto fail_nopgd;
800 
801 	if (init_new_context(p, mm))
802 		goto fail_nocontext;
803 
804 	mm->user_ns = get_user_ns(user_ns);
805 	return mm;
806 
807 fail_nocontext:
808 	mm_free_pgd(mm);
809 fail_nopgd:
810 	free_mm(mm);
811 	return NULL;
812 }
813 
check_mm(struct mm_struct * mm)814 static void check_mm(struct mm_struct *mm)
815 {
816 	int i;
817 
818 	for (i = 0; i < NR_MM_COUNTERS; i++) {
819 		long x = atomic_long_read(&mm->rss_stat.count[i]);
820 
821 		if (unlikely(x))
822 			printk(KERN_ALERT "BUG: Bad rss-counter state "
823 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
824 	}
825 
826 	if (atomic_long_read(&mm->nr_ptes))
827 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
828 				atomic_long_read(&mm->nr_ptes));
829 	if (mm_nr_pmds(mm))
830 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
831 				mm_nr_pmds(mm));
832 
833 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
834 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
835 #endif
836 }
837 
838 /*
839  * Allocate and initialize an mm_struct.
840  */
mm_alloc(void)841 struct mm_struct *mm_alloc(void)
842 {
843 	struct mm_struct *mm;
844 
845 	mm = allocate_mm();
846 	if (!mm)
847 		return NULL;
848 
849 	memset(mm, 0, sizeof(*mm));
850 	return mm_init(mm, current, current_user_ns());
851 }
852 
853 /*
854  * Called when the last reference to the mm
855  * is dropped: either by a lazy thread or by
856  * mmput. Free the page directory and the mm.
857  */
__mmdrop(struct mm_struct * mm)858 void __mmdrop(struct mm_struct *mm)
859 {
860 	BUG_ON(mm == &init_mm);
861 	mm_free_pgd(mm);
862 	destroy_context(mm);
863 	mmu_notifier_mm_destroy(mm);
864 	check_mm(mm);
865 	put_user_ns(mm->user_ns);
866 	free_mm(mm);
867 }
868 EXPORT_SYMBOL_GPL(__mmdrop);
869 
__mmput(struct mm_struct * mm)870 static inline void __mmput(struct mm_struct *mm)
871 {
872 	VM_BUG_ON(atomic_read(&mm->mm_users));
873 
874 	uprobe_clear_state(mm);
875 	exit_aio(mm);
876 	ksm_exit(mm);
877 	khugepaged_exit(mm); /* must run before exit_mmap */
878 	exit_mmap(mm);
879 	mm_put_huge_zero_page(mm);
880 	set_mm_exe_file(mm, NULL);
881 	if (!list_empty(&mm->mmlist)) {
882 		spin_lock(&mmlist_lock);
883 		list_del(&mm->mmlist);
884 		spin_unlock(&mmlist_lock);
885 	}
886 	if (mm->binfmt)
887 		module_put(mm->binfmt->module);
888 	set_bit(MMF_OOM_SKIP, &mm->flags);
889 	mmdrop(mm);
890 }
891 
892 /*
893  * Decrement the use count and release all resources for an mm.
894  */
mmput(struct mm_struct * mm)895 void mmput(struct mm_struct *mm)
896 {
897 	might_sleep();
898 
899 	if (atomic_dec_and_test(&mm->mm_users))
900 		__mmput(mm);
901 }
902 EXPORT_SYMBOL_GPL(mmput);
903 
904 #ifdef CONFIG_MMU
mmput_async_fn(struct work_struct * work)905 static void mmput_async_fn(struct work_struct *work)
906 {
907 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
908 	__mmput(mm);
909 }
910 
mmput_async(struct mm_struct * mm)911 void mmput_async(struct mm_struct *mm)
912 {
913 	if (atomic_dec_and_test(&mm->mm_users)) {
914 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
915 		schedule_work(&mm->async_put_work);
916 	}
917 }
918 #endif
919 
920 /**
921  * set_mm_exe_file - change a reference to the mm's executable file
922  *
923  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
924  *
925  * Main users are mmput() and sys_execve(). Callers prevent concurrent
926  * invocations: in mmput() nobody alive left, in execve task is single
927  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
928  * mm->exe_file, but does so without using set_mm_exe_file() in order
929  * to do avoid the need for any locks.
930  */
set_mm_exe_file(struct mm_struct * mm,struct file * new_exe_file)931 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
932 {
933 	struct file *old_exe_file;
934 
935 	/*
936 	 * It is safe to dereference the exe_file without RCU as
937 	 * this function is only called if nobody else can access
938 	 * this mm -- see comment above for justification.
939 	 */
940 	old_exe_file = rcu_dereference_raw(mm->exe_file);
941 
942 	if (new_exe_file)
943 		get_file(new_exe_file);
944 	rcu_assign_pointer(mm->exe_file, new_exe_file);
945 	if (old_exe_file)
946 		fput(old_exe_file);
947 }
948 
949 /**
950  * get_mm_exe_file - acquire a reference to the mm's executable file
951  *
952  * Returns %NULL if mm has no associated executable file.
953  * User must release file via fput().
954  */
get_mm_exe_file(struct mm_struct * mm)955 struct file *get_mm_exe_file(struct mm_struct *mm)
956 {
957 	struct file *exe_file;
958 
959 	rcu_read_lock();
960 	exe_file = rcu_dereference(mm->exe_file);
961 	if (exe_file && !get_file_rcu(exe_file))
962 		exe_file = NULL;
963 	rcu_read_unlock();
964 	return exe_file;
965 }
966 EXPORT_SYMBOL(get_mm_exe_file);
967 
968 /**
969  * get_task_exe_file - acquire a reference to the task's executable file
970  *
971  * Returns %NULL if task's mm (if any) has no associated executable file or
972  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
973  * User must release file via fput().
974  */
get_task_exe_file(struct task_struct * task)975 struct file *get_task_exe_file(struct task_struct *task)
976 {
977 	struct file *exe_file = NULL;
978 	struct mm_struct *mm;
979 
980 	task_lock(task);
981 	mm = task->mm;
982 	if (mm) {
983 		if (!(task->flags & PF_KTHREAD))
984 			exe_file = get_mm_exe_file(mm);
985 	}
986 	task_unlock(task);
987 	return exe_file;
988 }
989 EXPORT_SYMBOL(get_task_exe_file);
990 
991 /**
992  * get_task_mm - acquire a reference to the task's mm
993  *
994  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
995  * this kernel workthread has transiently adopted a user mm with use_mm,
996  * to do its AIO) is not set and if so returns a reference to it, after
997  * bumping up the use count.  User must release the mm via mmput()
998  * after use.  Typically used by /proc and ptrace.
999  */
get_task_mm(struct task_struct * task)1000 struct mm_struct *get_task_mm(struct task_struct *task)
1001 {
1002 	struct mm_struct *mm;
1003 
1004 	task_lock(task);
1005 	mm = task->mm;
1006 	if (mm) {
1007 		if (task->flags & PF_KTHREAD)
1008 			mm = NULL;
1009 		else
1010 			atomic_inc(&mm->mm_users);
1011 	}
1012 	task_unlock(task);
1013 	return mm;
1014 }
1015 EXPORT_SYMBOL_GPL(get_task_mm);
1016 
mm_access(struct task_struct * task,unsigned int mode)1017 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1018 {
1019 	struct mm_struct *mm;
1020 	int err;
1021 
1022 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1023 	if (err)
1024 		return ERR_PTR(err);
1025 
1026 	mm = get_task_mm(task);
1027 	if (mm && mm != current->mm &&
1028 			!ptrace_may_access(task, mode)) {
1029 		mmput(mm);
1030 		mm = ERR_PTR(-EACCES);
1031 	}
1032 	mutex_unlock(&task->signal->cred_guard_mutex);
1033 
1034 	return mm;
1035 }
1036 
complete_vfork_done(struct task_struct * tsk)1037 static void complete_vfork_done(struct task_struct *tsk)
1038 {
1039 	struct completion *vfork;
1040 
1041 	task_lock(tsk);
1042 	vfork = tsk->vfork_done;
1043 	if (likely(vfork)) {
1044 		tsk->vfork_done = NULL;
1045 		complete(vfork);
1046 	}
1047 	task_unlock(tsk);
1048 }
1049 
wait_for_vfork_done(struct task_struct * child,struct completion * vfork)1050 static int wait_for_vfork_done(struct task_struct *child,
1051 				struct completion *vfork)
1052 {
1053 	int killed;
1054 
1055 	freezer_do_not_count();
1056 	killed = wait_for_completion_killable(vfork);
1057 	freezer_count();
1058 
1059 	if (killed) {
1060 		task_lock(child);
1061 		child->vfork_done = NULL;
1062 		task_unlock(child);
1063 	}
1064 
1065 	put_task_struct(child);
1066 	return killed;
1067 }
1068 
1069 /* Please note the differences between mmput and mm_release.
1070  * mmput is called whenever we stop holding onto a mm_struct,
1071  * error success whatever.
1072  *
1073  * mm_release is called after a mm_struct has been removed
1074  * from the current process.
1075  *
1076  * This difference is important for error handling, when we
1077  * only half set up a mm_struct for a new process and need to restore
1078  * the old one.  Because we mmput the new mm_struct before
1079  * restoring the old one. . .
1080  * Eric Biederman 10 January 1998
1081  */
mm_release(struct task_struct * tsk,struct mm_struct * mm)1082 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1083 {
1084 	/* Get rid of any futexes when releasing the mm */
1085 #ifdef CONFIG_FUTEX
1086 	if (unlikely(tsk->robust_list)) {
1087 		exit_robust_list(tsk);
1088 		tsk->robust_list = NULL;
1089 	}
1090 #ifdef CONFIG_COMPAT
1091 	if (unlikely(tsk->compat_robust_list)) {
1092 		compat_exit_robust_list(tsk);
1093 		tsk->compat_robust_list = NULL;
1094 	}
1095 #endif
1096 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1097 		exit_pi_state_list(tsk);
1098 #endif
1099 
1100 	uprobe_free_utask(tsk);
1101 
1102 	/* Get rid of any cached register state */
1103 	deactivate_mm(tsk, mm);
1104 
1105 	/*
1106 	 * Signal userspace if we're not exiting with a core dump
1107 	 * because we want to leave the value intact for debugging
1108 	 * purposes.
1109 	 */
1110 	if (tsk->clear_child_tid) {
1111 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1112 		    atomic_read(&mm->mm_users) > 1) {
1113 			/*
1114 			 * We don't check the error code - if userspace has
1115 			 * not set up a proper pointer then tough luck.
1116 			 */
1117 			put_user(0, tsk->clear_child_tid);
1118 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1119 					1, NULL, NULL, 0);
1120 		}
1121 		tsk->clear_child_tid = NULL;
1122 	}
1123 
1124 	/*
1125 	 * All done, finally we can wake up parent and return this mm to him.
1126 	 * Also kthread_stop() uses this completion for synchronization.
1127 	 */
1128 	if (tsk->vfork_done)
1129 		complete_vfork_done(tsk);
1130 }
1131 
1132 /*
1133  * Allocate a new mm structure and copy contents from the
1134  * mm structure of the passed in task structure.
1135  */
dup_mm(struct task_struct * tsk)1136 static struct mm_struct *dup_mm(struct task_struct *tsk)
1137 {
1138 	struct mm_struct *mm, *oldmm = current->mm;
1139 	int err;
1140 
1141 	mm = allocate_mm();
1142 	if (!mm)
1143 		goto fail_nomem;
1144 
1145 	memcpy(mm, oldmm, sizeof(*mm));
1146 
1147 	if (!mm_init(mm, tsk, mm->user_ns))
1148 		goto fail_nomem;
1149 
1150 	err = dup_mmap(mm, oldmm);
1151 	if (err)
1152 		goto free_pt;
1153 
1154 	mm->hiwater_rss = get_mm_rss(mm);
1155 	mm->hiwater_vm = mm->total_vm;
1156 
1157 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1158 		goto free_pt;
1159 
1160 	return mm;
1161 
1162 free_pt:
1163 	/* don't put binfmt in mmput, we haven't got module yet */
1164 	mm->binfmt = NULL;
1165 	mmput(mm);
1166 
1167 fail_nomem:
1168 	return NULL;
1169 }
1170 
copy_mm(unsigned long clone_flags,struct task_struct * tsk)1171 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1172 {
1173 	struct mm_struct *mm, *oldmm;
1174 	int retval;
1175 
1176 	tsk->min_flt = tsk->maj_flt = 0;
1177 	tsk->nvcsw = tsk->nivcsw = 0;
1178 #ifdef CONFIG_DETECT_HUNG_TASK
1179 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1180 #endif
1181 
1182 	tsk->mm = NULL;
1183 	tsk->active_mm = NULL;
1184 
1185 	/*
1186 	 * Are we cloning a kernel thread?
1187 	 *
1188 	 * We need to steal a active VM for that..
1189 	 */
1190 	oldmm = current->mm;
1191 	if (!oldmm)
1192 		return 0;
1193 
1194 	/* initialize the new vmacache entries */
1195 	vmacache_flush(tsk);
1196 
1197 	if (clone_flags & CLONE_VM) {
1198 		atomic_inc(&oldmm->mm_users);
1199 		mm = oldmm;
1200 		goto good_mm;
1201 	}
1202 
1203 	retval = -ENOMEM;
1204 	mm = dup_mm(tsk);
1205 	if (!mm)
1206 		goto fail_nomem;
1207 
1208 good_mm:
1209 	tsk->mm = mm;
1210 	tsk->active_mm = mm;
1211 	return 0;
1212 
1213 fail_nomem:
1214 	return retval;
1215 }
1216 
copy_fs(unsigned long clone_flags,struct task_struct * tsk)1217 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1218 {
1219 	struct fs_struct *fs = current->fs;
1220 	if (clone_flags & CLONE_FS) {
1221 		/* tsk->fs is already what we want */
1222 		spin_lock(&fs->lock);
1223 		if (fs->in_exec) {
1224 			spin_unlock(&fs->lock);
1225 			return -EAGAIN;
1226 		}
1227 		fs->users++;
1228 		spin_unlock(&fs->lock);
1229 		return 0;
1230 	}
1231 	tsk->fs = copy_fs_struct(fs);
1232 	if (!tsk->fs)
1233 		return -ENOMEM;
1234 	return 0;
1235 }
1236 
copy_files(unsigned long clone_flags,struct task_struct * tsk)1237 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1238 {
1239 	struct files_struct *oldf, *newf;
1240 	int error = 0;
1241 
1242 	/*
1243 	 * A background process may not have any files ...
1244 	 */
1245 	oldf = current->files;
1246 	if (!oldf)
1247 		goto out;
1248 
1249 	if (clone_flags & CLONE_FILES) {
1250 		atomic_inc(&oldf->count);
1251 		goto out;
1252 	}
1253 
1254 	newf = dup_fd(oldf, &error);
1255 	if (!newf)
1256 		goto out;
1257 
1258 	tsk->files = newf;
1259 	error = 0;
1260 out:
1261 	return error;
1262 }
1263 
copy_io(unsigned long clone_flags,struct task_struct * tsk)1264 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1265 {
1266 #ifdef CONFIG_BLOCK
1267 	struct io_context *ioc = current->io_context;
1268 	struct io_context *new_ioc;
1269 
1270 	if (!ioc)
1271 		return 0;
1272 	/*
1273 	 * Share io context with parent, if CLONE_IO is set
1274 	 */
1275 	if (clone_flags & CLONE_IO) {
1276 		ioc_task_link(ioc);
1277 		tsk->io_context = ioc;
1278 	} else if (ioprio_valid(ioc->ioprio)) {
1279 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1280 		if (unlikely(!new_ioc))
1281 			return -ENOMEM;
1282 
1283 		new_ioc->ioprio = ioc->ioprio;
1284 		put_io_context(new_ioc);
1285 	}
1286 #endif
1287 	return 0;
1288 }
1289 
copy_sighand(unsigned long clone_flags,struct task_struct * tsk)1290 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1291 {
1292 	struct sighand_struct *sig;
1293 
1294 	if (clone_flags & CLONE_SIGHAND) {
1295 		atomic_inc(&current->sighand->count);
1296 		return 0;
1297 	}
1298 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1299 	rcu_assign_pointer(tsk->sighand, sig);
1300 	if (!sig)
1301 		return -ENOMEM;
1302 
1303 	atomic_set(&sig->count, 1);
1304 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1305 	return 0;
1306 }
1307 
__cleanup_sighand(struct sighand_struct * sighand)1308 void __cleanup_sighand(struct sighand_struct *sighand)
1309 {
1310 	if (atomic_dec_and_test(&sighand->count)) {
1311 		signalfd_cleanup(sighand);
1312 		/*
1313 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1314 		 * without an RCU grace period, see __lock_task_sighand().
1315 		 */
1316 		kmem_cache_free(sighand_cachep, sighand);
1317 	}
1318 }
1319 
1320 /*
1321  * Initialize POSIX timer handling for a thread group.
1322  */
posix_cpu_timers_init_group(struct signal_struct * sig)1323 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1324 {
1325 	unsigned long cpu_limit;
1326 
1327 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1328 	if (cpu_limit != RLIM_INFINITY) {
1329 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1330 		sig->cputimer.running = true;
1331 	}
1332 
1333 	/* The timer lists. */
1334 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1335 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1336 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1337 }
1338 
copy_signal(unsigned long clone_flags,struct task_struct * tsk)1339 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1340 {
1341 	struct signal_struct *sig;
1342 
1343 	if (clone_flags & CLONE_THREAD)
1344 		return 0;
1345 
1346 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1347 	tsk->signal = sig;
1348 	if (!sig)
1349 		return -ENOMEM;
1350 
1351 	sig->nr_threads = 1;
1352 	atomic_set(&sig->live, 1);
1353 	atomic_set(&sig->sigcnt, 1);
1354 
1355 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1356 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1357 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1358 
1359 	init_waitqueue_head(&sig->wait_chldexit);
1360 	sig->curr_target = tsk;
1361 	init_sigpending(&sig->shared_pending);
1362 	INIT_LIST_HEAD(&sig->posix_timers);
1363 	seqlock_init(&sig->stats_lock);
1364 	prev_cputime_init(&sig->prev_cputime);
1365 
1366 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1367 	sig->real_timer.function = it_real_fn;
1368 
1369 	task_lock(current->group_leader);
1370 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1371 	task_unlock(current->group_leader);
1372 
1373 	posix_cpu_timers_init_group(sig);
1374 
1375 	tty_audit_fork(sig);
1376 	sched_autogroup_fork(sig);
1377 
1378 	sig->oom_score_adj = current->signal->oom_score_adj;
1379 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1380 
1381 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1382 				   current->signal->is_child_subreaper;
1383 
1384 	mutex_init(&sig->cred_guard_mutex);
1385 
1386 	return 0;
1387 }
1388 
copy_seccomp(struct task_struct * p)1389 static void copy_seccomp(struct task_struct *p)
1390 {
1391 #ifdef CONFIG_SECCOMP
1392 	/*
1393 	 * Must be called with sighand->lock held, which is common to
1394 	 * all threads in the group. Holding cred_guard_mutex is not
1395 	 * needed because this new task is not yet running and cannot
1396 	 * be racing exec.
1397 	 */
1398 	assert_spin_locked(&current->sighand->siglock);
1399 
1400 	/* Ref-count the new filter user, and assign it. */
1401 	get_seccomp_filter(current);
1402 	p->seccomp = current->seccomp;
1403 
1404 	/*
1405 	 * Explicitly enable no_new_privs here in case it got set
1406 	 * between the task_struct being duplicated and holding the
1407 	 * sighand lock. The seccomp state and nnp must be in sync.
1408 	 */
1409 	if (task_no_new_privs(current))
1410 		task_set_no_new_privs(p);
1411 
1412 	/*
1413 	 * If the parent gained a seccomp mode after copying thread
1414 	 * flags and between before we held the sighand lock, we have
1415 	 * to manually enable the seccomp thread flag here.
1416 	 */
1417 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1418 		set_tsk_thread_flag(p, TIF_SECCOMP);
1419 #endif
1420 }
1421 
SYSCALL_DEFINE1(set_tid_address,int __user *,tidptr)1422 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1423 {
1424 	current->clear_child_tid = tidptr;
1425 
1426 	return task_pid_vnr(current);
1427 }
1428 
rt_mutex_init_task(struct task_struct * p)1429 static void rt_mutex_init_task(struct task_struct *p)
1430 {
1431 	raw_spin_lock_init(&p->pi_lock);
1432 #ifdef CONFIG_RT_MUTEXES
1433 	p->pi_waiters = RB_ROOT;
1434 	p->pi_waiters_leftmost = NULL;
1435 	p->pi_blocked_on = NULL;
1436 #endif
1437 }
1438 
1439 /*
1440  * Initialize POSIX timer handling for a single task.
1441  */
posix_cpu_timers_init(struct task_struct * tsk)1442 static void posix_cpu_timers_init(struct task_struct *tsk)
1443 {
1444 	tsk->cputime_expires.prof_exp = 0;
1445 	tsk->cputime_expires.virt_exp = 0;
1446 	tsk->cputime_expires.sched_exp = 0;
1447 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1448 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1449 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1450 }
1451 
1452 static inline void
init_task_pid(struct task_struct * task,enum pid_type type,struct pid * pid)1453 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1454 {
1455 	 task->pids[type].pid = pid;
1456 }
1457 
1458 /*
1459  * This creates a new process as a copy of the old one,
1460  * but does not actually start it yet.
1461  *
1462  * It copies the registers, and all the appropriate
1463  * parts of the process environment (as per the clone
1464  * flags). The actual kick-off is left to the caller.
1465  */
copy_process(unsigned long clone_flags,unsigned long stack_start,unsigned long stack_size,int __user * child_tidptr,struct pid * pid,int trace,unsigned long tls,int node)1466 static __latent_entropy struct task_struct *copy_process(
1467 					unsigned long clone_flags,
1468 					unsigned long stack_start,
1469 					unsigned long stack_size,
1470 					int __user *child_tidptr,
1471 					struct pid *pid,
1472 					int trace,
1473 					unsigned long tls,
1474 					int node)
1475 {
1476 	int retval;
1477 	struct task_struct *p;
1478 
1479 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1480 		return ERR_PTR(-EINVAL);
1481 
1482 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1483 		return ERR_PTR(-EINVAL);
1484 
1485 	/*
1486 	 * Thread groups must share signals as well, and detached threads
1487 	 * can only be started up within the thread group.
1488 	 */
1489 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1490 		return ERR_PTR(-EINVAL);
1491 
1492 	/*
1493 	 * Shared signal handlers imply shared VM. By way of the above,
1494 	 * thread groups also imply shared VM. Blocking this case allows
1495 	 * for various simplifications in other code.
1496 	 */
1497 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1498 		return ERR_PTR(-EINVAL);
1499 
1500 	/*
1501 	 * Siblings of global init remain as zombies on exit since they are
1502 	 * not reaped by their parent (swapper). To solve this and to avoid
1503 	 * multi-rooted process trees, prevent global and container-inits
1504 	 * from creating siblings.
1505 	 */
1506 	if ((clone_flags & CLONE_PARENT) &&
1507 				current->signal->flags & SIGNAL_UNKILLABLE)
1508 		return ERR_PTR(-EINVAL);
1509 
1510 	/*
1511 	 * If the new process will be in a different pid or user namespace
1512 	 * do not allow it to share a thread group with the forking task.
1513 	 */
1514 	if (clone_flags & CLONE_THREAD) {
1515 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1516 		    (task_active_pid_ns(current) !=
1517 				current->nsproxy->pid_ns_for_children))
1518 			return ERR_PTR(-EINVAL);
1519 	}
1520 
1521 	retval = security_task_create(clone_flags);
1522 	if (retval)
1523 		goto fork_out;
1524 
1525 	retval = -ENOMEM;
1526 	p = dup_task_struct(current, node);
1527 	if (!p)
1528 		goto fork_out;
1529 
1530 	ftrace_graph_init_task(p);
1531 
1532 	rt_mutex_init_task(p);
1533 
1534 #ifdef CONFIG_PROVE_LOCKING
1535 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1536 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1537 #endif
1538 	retval = -EAGAIN;
1539 	if (atomic_read(&p->real_cred->user->processes) >=
1540 			task_rlimit(p, RLIMIT_NPROC)) {
1541 		if (p->real_cred->user != INIT_USER &&
1542 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1543 			goto bad_fork_free;
1544 	}
1545 	current->flags &= ~PF_NPROC_EXCEEDED;
1546 
1547 	retval = copy_creds(p, clone_flags);
1548 	if (retval < 0)
1549 		goto bad_fork_free;
1550 
1551 	/*
1552 	 * If multiple threads are within copy_process(), then this check
1553 	 * triggers too late. This doesn't hurt, the check is only there
1554 	 * to stop root fork bombs.
1555 	 */
1556 	retval = -EAGAIN;
1557 	if (nr_threads >= max_threads)
1558 		goto bad_fork_cleanup_count;
1559 
1560 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1561 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1562 	p->flags |= PF_FORKNOEXEC;
1563 	INIT_LIST_HEAD(&p->children);
1564 	INIT_LIST_HEAD(&p->sibling);
1565 	rcu_copy_process(p);
1566 	p->vfork_done = NULL;
1567 	spin_lock_init(&p->alloc_lock);
1568 
1569 	init_sigpending(&p->pending);
1570 
1571 	p->utime = p->stime = p->gtime = 0;
1572 	p->utimescaled = p->stimescaled = 0;
1573 	prev_cputime_init(&p->prev_cputime);
1574 
1575 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1576 	seqcount_init(&p->vtime_seqcount);
1577 	p->vtime_snap = 0;
1578 	p->vtime_snap_whence = VTIME_INACTIVE;
1579 #endif
1580 
1581 #if defined(SPLIT_RSS_COUNTING)
1582 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1583 #endif
1584 
1585 	p->default_timer_slack_ns = current->timer_slack_ns;
1586 
1587 	task_io_accounting_init(&p->ioac);
1588 	acct_clear_integrals(p);
1589 
1590 	posix_cpu_timers_init(p);
1591 
1592 	p->start_time = ktime_get_ns();
1593 	p->real_start_time = ktime_get_boot_ns();
1594 	p->io_context = NULL;
1595 	p->audit_context = NULL;
1596 	cgroup_fork(p);
1597 #ifdef CONFIG_NUMA
1598 	p->mempolicy = mpol_dup(p->mempolicy);
1599 	if (IS_ERR(p->mempolicy)) {
1600 		retval = PTR_ERR(p->mempolicy);
1601 		p->mempolicy = NULL;
1602 		goto bad_fork_cleanup_threadgroup_lock;
1603 	}
1604 #endif
1605 #ifdef CONFIG_CPUSETS
1606 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1607 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1608 	seqcount_init(&p->mems_allowed_seq);
1609 #endif
1610 #ifdef CONFIG_TRACE_IRQFLAGS
1611 	p->irq_events = 0;
1612 	p->hardirqs_enabled = 0;
1613 	p->hardirq_enable_ip = 0;
1614 	p->hardirq_enable_event = 0;
1615 	p->hardirq_disable_ip = _THIS_IP_;
1616 	p->hardirq_disable_event = 0;
1617 	p->softirqs_enabled = 1;
1618 	p->softirq_enable_ip = _THIS_IP_;
1619 	p->softirq_enable_event = 0;
1620 	p->softirq_disable_ip = 0;
1621 	p->softirq_disable_event = 0;
1622 	p->hardirq_context = 0;
1623 	p->softirq_context = 0;
1624 #endif
1625 
1626 	p->pagefault_disabled = 0;
1627 
1628 #ifdef CONFIG_LOCKDEP
1629 	p->lockdep_depth = 0; /* no locks held yet */
1630 	p->curr_chain_key = 0;
1631 	p->lockdep_recursion = 0;
1632 #endif
1633 
1634 #ifdef CONFIG_DEBUG_MUTEXES
1635 	p->blocked_on = NULL; /* not blocked yet */
1636 #endif
1637 #ifdef CONFIG_BCACHE
1638 	p->sequential_io	= 0;
1639 	p->sequential_io_avg	= 0;
1640 #endif
1641 
1642 	/* Perform scheduler related setup. Assign this task to a CPU. */
1643 	retval = sched_fork(clone_flags, p);
1644 	if (retval)
1645 		goto bad_fork_cleanup_policy;
1646 
1647 	retval = perf_event_init_task(p);
1648 	if (retval)
1649 		goto bad_fork_cleanup_policy;
1650 	retval = audit_alloc(p);
1651 	if (retval)
1652 		goto bad_fork_cleanup_perf;
1653 	/* copy all the process information */
1654 	shm_init_task(p);
1655 	retval = copy_semundo(clone_flags, p);
1656 	if (retval)
1657 		goto bad_fork_cleanup_audit;
1658 	retval = copy_files(clone_flags, p);
1659 	if (retval)
1660 		goto bad_fork_cleanup_semundo;
1661 	retval = copy_fs(clone_flags, p);
1662 	if (retval)
1663 		goto bad_fork_cleanup_files;
1664 	retval = copy_sighand(clone_flags, p);
1665 	if (retval)
1666 		goto bad_fork_cleanup_fs;
1667 	retval = copy_signal(clone_flags, p);
1668 	if (retval)
1669 		goto bad_fork_cleanup_sighand;
1670 	retval = copy_mm(clone_flags, p);
1671 	if (retval)
1672 		goto bad_fork_cleanup_signal;
1673 	retval = copy_namespaces(clone_flags, p);
1674 	if (retval)
1675 		goto bad_fork_cleanup_mm;
1676 	retval = copy_io(clone_flags, p);
1677 	if (retval)
1678 		goto bad_fork_cleanup_namespaces;
1679 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1680 	if (retval)
1681 		goto bad_fork_cleanup_io;
1682 
1683 	if (pid != &init_struct_pid) {
1684 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1685 		if (IS_ERR(pid)) {
1686 			retval = PTR_ERR(pid);
1687 			goto bad_fork_cleanup_thread;
1688 		}
1689 	}
1690 
1691 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1692 	/*
1693 	 * Clear TID on mm_release()?
1694 	 */
1695 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1696 #ifdef CONFIG_BLOCK
1697 	p->plug = NULL;
1698 #endif
1699 #ifdef CONFIG_FUTEX
1700 	p->robust_list = NULL;
1701 #ifdef CONFIG_COMPAT
1702 	p->compat_robust_list = NULL;
1703 #endif
1704 	INIT_LIST_HEAD(&p->pi_state_list);
1705 	p->pi_state_cache = NULL;
1706 #endif
1707 	/*
1708 	 * sigaltstack should be cleared when sharing the same VM
1709 	 */
1710 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1711 		sas_ss_reset(p);
1712 
1713 	/*
1714 	 * Syscall tracing and stepping should be turned off in the
1715 	 * child regardless of CLONE_PTRACE.
1716 	 */
1717 	user_disable_single_step(p);
1718 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1719 #ifdef TIF_SYSCALL_EMU
1720 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1721 #endif
1722 	clear_all_latency_tracing(p);
1723 
1724 	/* ok, now we should be set up.. */
1725 	p->pid = pid_nr(pid);
1726 	if (clone_flags & CLONE_THREAD) {
1727 		p->exit_signal = -1;
1728 		p->group_leader = current->group_leader;
1729 		p->tgid = current->tgid;
1730 	} else {
1731 		if (clone_flags & CLONE_PARENT)
1732 			p->exit_signal = current->group_leader->exit_signal;
1733 		else
1734 			p->exit_signal = (clone_flags & CSIGNAL);
1735 		p->group_leader = p;
1736 		p->tgid = p->pid;
1737 	}
1738 
1739 	p->nr_dirtied = 0;
1740 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1741 	p->dirty_paused_when = 0;
1742 
1743 	p->pdeath_signal = 0;
1744 	INIT_LIST_HEAD(&p->thread_group);
1745 	p->task_works = NULL;
1746 
1747 	threadgroup_change_begin(current);
1748 	/*
1749 	 * Ensure that the cgroup subsystem policies allow the new process to be
1750 	 * forked. It should be noted the the new process's css_set can be changed
1751 	 * between here and cgroup_post_fork() if an organisation operation is in
1752 	 * progress.
1753 	 */
1754 	retval = cgroup_can_fork(p);
1755 	if (retval)
1756 		goto bad_fork_free_pid;
1757 
1758 	/*
1759 	 * Make it visible to the rest of the system, but dont wake it up yet.
1760 	 * Need tasklist lock for parent etc handling!
1761 	 */
1762 	write_lock_irq(&tasklist_lock);
1763 
1764 	/* CLONE_PARENT re-uses the old parent */
1765 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1766 		p->real_parent = current->real_parent;
1767 		p->parent_exec_id = current->parent_exec_id;
1768 	} else {
1769 		p->real_parent = current;
1770 		p->parent_exec_id = current->self_exec_id;
1771 	}
1772 
1773 	spin_lock(&current->sighand->siglock);
1774 
1775 	/*
1776 	 * Copy seccomp details explicitly here, in case they were changed
1777 	 * before holding sighand lock.
1778 	 */
1779 	copy_seccomp(p);
1780 
1781 	/*
1782 	 * Process group and session signals need to be delivered to just the
1783 	 * parent before the fork or both the parent and the child after the
1784 	 * fork. Restart if a signal comes in before we add the new process to
1785 	 * it's process group.
1786 	 * A fatal signal pending means that current will exit, so the new
1787 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1788 	*/
1789 	recalc_sigpending();
1790 	if (signal_pending(current)) {
1791 		retval = -ERESTARTNOINTR;
1792 		goto bad_fork_cancel_cgroup;
1793 	}
1794 	if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1795 		retval = -ENOMEM;
1796 		goto bad_fork_cancel_cgroup;
1797 	}
1798 
1799 	if (likely(p->pid)) {
1800 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1801 
1802 		init_task_pid(p, PIDTYPE_PID, pid);
1803 		if (thread_group_leader(p)) {
1804 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1805 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1806 
1807 			if (is_child_reaper(pid)) {
1808 				ns_of_pid(pid)->child_reaper = p;
1809 				p->signal->flags |= SIGNAL_UNKILLABLE;
1810 			}
1811 
1812 			p->signal->leader_pid = pid;
1813 			p->signal->tty = tty_kref_get(current->signal->tty);
1814 			list_add_tail(&p->sibling, &p->real_parent->children);
1815 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1816 			attach_pid(p, PIDTYPE_PGID);
1817 			attach_pid(p, PIDTYPE_SID);
1818 			__this_cpu_inc(process_counts);
1819 		} else {
1820 			current->signal->nr_threads++;
1821 			atomic_inc(&current->signal->live);
1822 			atomic_inc(&current->signal->sigcnt);
1823 			list_add_tail_rcu(&p->thread_group,
1824 					  &p->group_leader->thread_group);
1825 			list_add_tail_rcu(&p->thread_node,
1826 					  &p->signal->thread_head);
1827 		}
1828 		attach_pid(p, PIDTYPE_PID);
1829 		nr_threads++;
1830 	}
1831 
1832 	total_forks++;
1833 	spin_unlock(&current->sighand->siglock);
1834 	syscall_tracepoint_update(p);
1835 	write_unlock_irq(&tasklist_lock);
1836 
1837 	proc_fork_connector(p);
1838 	cgroup_post_fork(p);
1839 	threadgroup_change_end(current);
1840 	perf_event_fork(p);
1841 
1842 	trace_task_newtask(p, clone_flags);
1843 	uprobe_copy_process(p, clone_flags);
1844 
1845 	return p;
1846 
1847 bad_fork_cancel_cgroup:
1848 	spin_unlock(&current->sighand->siglock);
1849 	write_unlock_irq(&tasklist_lock);
1850 	cgroup_cancel_fork(p);
1851 bad_fork_free_pid:
1852 	threadgroup_change_end(current);
1853 	if (pid != &init_struct_pid)
1854 		free_pid(pid);
1855 bad_fork_cleanup_thread:
1856 	exit_thread(p);
1857 bad_fork_cleanup_io:
1858 	if (p->io_context)
1859 		exit_io_context(p);
1860 bad_fork_cleanup_namespaces:
1861 	exit_task_namespaces(p);
1862 bad_fork_cleanup_mm:
1863 	if (p->mm)
1864 		mmput(p->mm);
1865 bad_fork_cleanup_signal:
1866 	if (!(clone_flags & CLONE_THREAD))
1867 		free_signal_struct(p->signal);
1868 bad_fork_cleanup_sighand:
1869 	__cleanup_sighand(p->sighand);
1870 bad_fork_cleanup_fs:
1871 	exit_fs(p); /* blocking */
1872 bad_fork_cleanup_files:
1873 	exit_files(p); /* blocking */
1874 bad_fork_cleanup_semundo:
1875 	exit_sem(p);
1876 bad_fork_cleanup_audit:
1877 	audit_free(p);
1878 bad_fork_cleanup_perf:
1879 	perf_event_free_task(p);
1880 bad_fork_cleanup_policy:
1881 #ifdef CONFIG_NUMA
1882 	mpol_put(p->mempolicy);
1883 bad_fork_cleanup_threadgroup_lock:
1884 #endif
1885 	delayacct_tsk_free(p);
1886 bad_fork_cleanup_count:
1887 	atomic_dec(&p->cred->user->processes);
1888 	exit_creds(p);
1889 bad_fork_free:
1890 	p->state = TASK_DEAD;
1891 	put_task_stack(p);
1892 	free_task(p);
1893 fork_out:
1894 	return ERR_PTR(retval);
1895 }
1896 
init_idle_pids(struct pid_link * links)1897 static inline void init_idle_pids(struct pid_link *links)
1898 {
1899 	enum pid_type type;
1900 
1901 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1902 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1903 		links[type].pid = &init_struct_pid;
1904 	}
1905 }
1906 
fork_idle(int cpu)1907 struct task_struct *fork_idle(int cpu)
1908 {
1909 	struct task_struct *task;
1910 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1911 			    cpu_to_node(cpu));
1912 	if (!IS_ERR(task)) {
1913 		init_idle_pids(task->pids);
1914 		init_idle(task, cpu);
1915 	}
1916 
1917 	return task;
1918 }
1919 
1920 /*
1921  *  Ok, this is the main fork-routine.
1922  *
1923  * It copies the process, and if successful kick-starts
1924  * it and waits for it to finish using the VM if required.
1925  */
_do_fork(unsigned long clone_flags,unsigned long stack_start,unsigned long stack_size,int __user * parent_tidptr,int __user * child_tidptr,unsigned long tls)1926 long _do_fork(unsigned long clone_flags,
1927 	      unsigned long stack_start,
1928 	      unsigned long stack_size,
1929 	      int __user *parent_tidptr,
1930 	      int __user *child_tidptr,
1931 	      unsigned long tls)
1932 {
1933 	struct task_struct *p;
1934 	int trace = 0;
1935 	long nr;
1936 
1937 	/*
1938 	 * Determine whether and which event to report to ptracer.  When
1939 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1940 	 * requested, no event is reported; otherwise, report if the event
1941 	 * for the type of forking is enabled.
1942 	 */
1943 	if (!(clone_flags & CLONE_UNTRACED)) {
1944 		if (clone_flags & CLONE_VFORK)
1945 			trace = PTRACE_EVENT_VFORK;
1946 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1947 			trace = PTRACE_EVENT_CLONE;
1948 		else
1949 			trace = PTRACE_EVENT_FORK;
1950 
1951 		if (likely(!ptrace_event_enabled(current, trace)))
1952 			trace = 0;
1953 	}
1954 
1955 	p = copy_process(clone_flags, stack_start, stack_size,
1956 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1957 	add_latent_entropy();
1958 	/*
1959 	 * Do this prior waking up the new thread - the thread pointer
1960 	 * might get invalid after that point, if the thread exits quickly.
1961 	 */
1962 	if (!IS_ERR(p)) {
1963 		struct completion vfork;
1964 		struct pid *pid;
1965 
1966 		trace_sched_process_fork(current, p);
1967 
1968 		pid = get_task_pid(p, PIDTYPE_PID);
1969 		nr = pid_vnr(pid);
1970 
1971 		if (clone_flags & CLONE_PARENT_SETTID)
1972 			put_user(nr, parent_tidptr);
1973 
1974 		if (clone_flags & CLONE_VFORK) {
1975 			p->vfork_done = &vfork;
1976 			init_completion(&vfork);
1977 			get_task_struct(p);
1978 		}
1979 
1980 		wake_up_new_task(p);
1981 
1982 		/* forking complete and child started to run, tell ptracer */
1983 		if (unlikely(trace))
1984 			ptrace_event_pid(trace, pid);
1985 
1986 		if (clone_flags & CLONE_VFORK) {
1987 			if (!wait_for_vfork_done(p, &vfork))
1988 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1989 		}
1990 
1991 		put_pid(pid);
1992 	} else {
1993 		nr = PTR_ERR(p);
1994 	}
1995 	return nr;
1996 }
1997 
1998 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1999 /* For compatibility with architectures that call do_fork directly rather than
2000  * 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)2001 long do_fork(unsigned long clone_flags,
2002 	      unsigned long stack_start,
2003 	      unsigned long stack_size,
2004 	      int __user *parent_tidptr,
2005 	      int __user *child_tidptr)
2006 {
2007 	return _do_fork(clone_flags, stack_start, stack_size,
2008 			parent_tidptr, child_tidptr, 0);
2009 }
2010 #endif
2011 
2012 /*
2013  * Create a kernel thread.
2014  */
kernel_thread(int (* fn)(void *),void * arg,unsigned long flags)2015 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2016 {
2017 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2018 		(unsigned long)arg, NULL, NULL, 0);
2019 }
2020 
2021 #ifdef __ARCH_WANT_SYS_FORK
SYSCALL_DEFINE0(fork)2022 SYSCALL_DEFINE0(fork)
2023 {
2024 #ifdef CONFIG_MMU
2025 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2026 #else
2027 	/* can not support in nommu mode */
2028 	return -EINVAL;
2029 #endif
2030 }
2031 #endif
2032 
2033 #ifdef __ARCH_WANT_SYS_VFORK
SYSCALL_DEFINE0(vfork)2034 SYSCALL_DEFINE0(vfork)
2035 {
2036 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2037 			0, NULL, NULL, 0);
2038 }
2039 #endif
2040 
2041 #ifdef __ARCH_WANT_SYS_CLONE
2042 #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)2043 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2044 		 int __user *, parent_tidptr,
2045 		 unsigned long, tls,
2046 		 int __user *, child_tidptr)
2047 #elif defined(CONFIG_CLONE_BACKWARDS2)
2048 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2049 		 int __user *, parent_tidptr,
2050 		 int __user *, child_tidptr,
2051 		 unsigned long, tls)
2052 #elif defined(CONFIG_CLONE_BACKWARDS3)
2053 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2054 		int, stack_size,
2055 		int __user *, parent_tidptr,
2056 		int __user *, child_tidptr,
2057 		unsigned long, tls)
2058 #else
2059 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2060 		 int __user *, parent_tidptr,
2061 		 int __user *, child_tidptr,
2062 		 unsigned long, tls)
2063 #endif
2064 {
2065 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2066 }
2067 #endif
2068 
2069 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2070 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2071 #endif
2072 
sighand_ctor(void * data)2073 static void sighand_ctor(void *data)
2074 {
2075 	struct sighand_struct *sighand = data;
2076 
2077 	spin_lock_init(&sighand->siglock);
2078 	init_waitqueue_head(&sighand->signalfd_wqh);
2079 }
2080 
proc_caches_init(void)2081 void __init proc_caches_init(void)
2082 {
2083 	sighand_cachep = kmem_cache_create("sighand_cache",
2084 			sizeof(struct sighand_struct), 0,
2085 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2086 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2087 	signal_cachep = kmem_cache_create("signal_cache",
2088 			sizeof(struct signal_struct), 0,
2089 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2090 			NULL);
2091 	files_cachep = kmem_cache_create("files_cache",
2092 			sizeof(struct files_struct), 0,
2093 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2094 			NULL);
2095 	fs_cachep = kmem_cache_create("fs_cache",
2096 			sizeof(struct fs_struct), 0,
2097 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2098 			NULL);
2099 	/*
2100 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2101 	 * whole struct cpumask for the OFFSTACK case. We could change
2102 	 * this to *only* allocate as much of it as required by the
2103 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2104 	 * is at the end of the structure, exactly for that reason.
2105 	 */
2106 	mm_cachep = kmem_cache_create("mm_struct",
2107 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2108 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2109 			NULL);
2110 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2111 	mmap_init();
2112 	nsproxy_cache_init();
2113 }
2114 
2115 /*
2116  * Check constraints on flags passed to the unshare system call.
2117  */
check_unshare_flags(unsigned long unshare_flags)2118 static int check_unshare_flags(unsigned long unshare_flags)
2119 {
2120 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2121 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2122 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2123 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2124 		return -EINVAL;
2125 	/*
2126 	 * Not implemented, but pretend it works if there is nothing
2127 	 * to unshare.  Note that unsharing the address space or the
2128 	 * signal handlers also need to unshare the signal queues (aka
2129 	 * CLONE_THREAD).
2130 	 */
2131 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2132 		if (!thread_group_empty(current))
2133 			return -EINVAL;
2134 	}
2135 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2136 		if (atomic_read(&current->sighand->count) > 1)
2137 			return -EINVAL;
2138 	}
2139 	if (unshare_flags & CLONE_VM) {
2140 		if (!current_is_single_threaded())
2141 			return -EINVAL;
2142 	}
2143 
2144 	return 0;
2145 }
2146 
2147 /*
2148  * Unshare the filesystem structure if it is being shared
2149  */
unshare_fs(unsigned long unshare_flags,struct fs_struct ** new_fsp)2150 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2151 {
2152 	struct fs_struct *fs = current->fs;
2153 
2154 	if (!(unshare_flags & CLONE_FS) || !fs)
2155 		return 0;
2156 
2157 	/* don't need lock here; in the worst case we'll do useless copy */
2158 	if (fs->users == 1)
2159 		return 0;
2160 
2161 	*new_fsp = copy_fs_struct(fs);
2162 	if (!*new_fsp)
2163 		return -ENOMEM;
2164 
2165 	return 0;
2166 }
2167 
2168 /*
2169  * Unshare file descriptor table if it is being shared
2170  */
unshare_fd(unsigned long unshare_flags,struct files_struct ** new_fdp)2171 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2172 {
2173 	struct files_struct *fd = current->files;
2174 	int error = 0;
2175 
2176 	if ((unshare_flags & CLONE_FILES) &&
2177 	    (fd && atomic_read(&fd->count) > 1)) {
2178 		*new_fdp = dup_fd(fd, &error);
2179 		if (!*new_fdp)
2180 			return error;
2181 	}
2182 
2183 	return 0;
2184 }
2185 
2186 /*
2187  * unshare allows a process to 'unshare' part of the process
2188  * context which was originally shared using clone.  copy_*
2189  * functions used by do_fork() cannot be used here directly
2190  * because they modify an inactive task_struct that is being
2191  * constructed. Here we are modifying the current, active,
2192  * task_struct.
2193  */
SYSCALL_DEFINE1(unshare,unsigned long,unshare_flags)2194 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2195 {
2196 	struct fs_struct *fs, *new_fs = NULL;
2197 	struct files_struct *fd, *new_fd = NULL;
2198 	struct cred *new_cred = NULL;
2199 	struct nsproxy *new_nsproxy = NULL;
2200 	int do_sysvsem = 0;
2201 	int err;
2202 
2203 	/*
2204 	 * If unsharing a user namespace must also unshare the thread group
2205 	 * and unshare the filesystem root and working directories.
2206 	 */
2207 	if (unshare_flags & CLONE_NEWUSER)
2208 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2209 	/*
2210 	 * If unsharing vm, must also unshare signal handlers.
2211 	 */
2212 	if (unshare_flags & CLONE_VM)
2213 		unshare_flags |= CLONE_SIGHAND;
2214 	/*
2215 	 * If unsharing a signal handlers, must also unshare the signal queues.
2216 	 */
2217 	if (unshare_flags & CLONE_SIGHAND)
2218 		unshare_flags |= CLONE_THREAD;
2219 	/*
2220 	 * If unsharing namespace, must also unshare filesystem information.
2221 	 */
2222 	if (unshare_flags & CLONE_NEWNS)
2223 		unshare_flags |= CLONE_FS;
2224 
2225 	err = check_unshare_flags(unshare_flags);
2226 	if (err)
2227 		goto bad_unshare_out;
2228 	/*
2229 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2230 	 * to a new ipc namespace, the semaphore arrays from the old
2231 	 * namespace are unreachable.
2232 	 */
2233 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2234 		do_sysvsem = 1;
2235 	err = unshare_fs(unshare_flags, &new_fs);
2236 	if (err)
2237 		goto bad_unshare_out;
2238 	err = unshare_fd(unshare_flags, &new_fd);
2239 	if (err)
2240 		goto bad_unshare_cleanup_fs;
2241 	err = unshare_userns(unshare_flags, &new_cred);
2242 	if (err)
2243 		goto bad_unshare_cleanup_fd;
2244 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2245 					 new_cred, new_fs);
2246 	if (err)
2247 		goto bad_unshare_cleanup_cred;
2248 
2249 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2250 		if (do_sysvsem) {
2251 			/*
2252 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2253 			 */
2254 			exit_sem(current);
2255 		}
2256 		if (unshare_flags & CLONE_NEWIPC) {
2257 			/* Orphan segments in old ns (see sem above). */
2258 			exit_shm(current);
2259 			shm_init_task(current);
2260 		}
2261 
2262 		if (new_nsproxy)
2263 			switch_task_namespaces(current, new_nsproxy);
2264 
2265 		task_lock(current);
2266 
2267 		if (new_fs) {
2268 			fs = current->fs;
2269 			spin_lock(&fs->lock);
2270 			current->fs = new_fs;
2271 			if (--fs->users)
2272 				new_fs = NULL;
2273 			else
2274 				new_fs = fs;
2275 			spin_unlock(&fs->lock);
2276 		}
2277 
2278 		if (new_fd) {
2279 			fd = current->files;
2280 			current->files = new_fd;
2281 			new_fd = fd;
2282 		}
2283 
2284 		task_unlock(current);
2285 
2286 		if (new_cred) {
2287 			/* Install the new user namespace */
2288 			commit_creds(new_cred);
2289 			new_cred = NULL;
2290 		}
2291 	}
2292 
2293 bad_unshare_cleanup_cred:
2294 	if (new_cred)
2295 		put_cred(new_cred);
2296 bad_unshare_cleanup_fd:
2297 	if (new_fd)
2298 		put_files_struct(new_fd);
2299 
2300 bad_unshare_cleanup_fs:
2301 	if (new_fs)
2302 		free_fs_struct(new_fs);
2303 
2304 bad_unshare_out:
2305 	return err;
2306 }
2307 
2308 /*
2309  *	Helper to unshare the files of the current task.
2310  *	We don't want to expose copy_files internals to
2311  *	the exec layer of the kernel.
2312  */
2313 
unshare_files(struct files_struct ** displaced)2314 int unshare_files(struct files_struct **displaced)
2315 {
2316 	struct task_struct *task = current;
2317 	struct files_struct *copy = NULL;
2318 	int error;
2319 
2320 	error = unshare_fd(CLONE_FILES, &copy);
2321 	if (error || !copy) {
2322 		*displaced = NULL;
2323 		return error;
2324 	}
2325 	*displaced = task->files;
2326 	task_lock(task);
2327 	task->files = copy;
2328 	task_unlock(task);
2329 	return 0;
2330 }
2331 
sysctl_max_threads(struct ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)2332 int sysctl_max_threads(struct ctl_table *table, int write,
2333 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2334 {
2335 	struct ctl_table t;
2336 	int ret;
2337 	int threads = max_threads;
2338 	int min = MIN_THREADS;
2339 	int max = MAX_THREADS;
2340 
2341 	t = *table;
2342 	t.data = &threads;
2343 	t.extra1 = &min;
2344 	t.extra2 = &max;
2345 
2346 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2347 	if (ret || !write)
2348 		return ret;
2349 
2350 	set_max_threads(threads);
2351 
2352 	return 0;
2353 }
2354