xref: /linux/mm/mmap.c (revision 607bfbd7ffc60156ae0831c917497dc91a57dd8d)
1 /*
2  * mm/mmap.c
3  *
4  * Written by obz.
5  *
6  * Address space accounting code	<alan@lxorguk.ukuu.org.uk>
7  */
8 
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/backing-dev.h>
14 #include <linux/mm.h>
15 #include <linux/vmacache.h>
16 #include <linux/shm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/syscalls.h>
21 #include <linux/capability.h>
22 #include <linux/init.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/hugetlb.h>
28 #include <linux/profile.h>
29 #include <linux/export.h>
30 #include <linux/mount.h>
31 #include <linux/mempolicy.h>
32 #include <linux/rmap.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/mmdebug.h>
35 #include <linux/perf_event.h>
36 #include <linux/audit.h>
37 #include <linux/khugepaged.h>
38 #include <linux/uprobes.h>
39 #include <linux/rbtree_augmented.h>
40 #include <linux/sched/sysctl.h>
41 #include <linux/notifier.h>
42 #include <linux/memory.h>
43 #include <linux/printk.h>
44 #include <linux/userfaultfd_k.h>
45 #include <linux/moduleparam.h>
46 
47 #include <asm/uaccess.h>
48 #include <asm/cacheflush.h>
49 #include <asm/tlb.h>
50 #include <asm/mmu_context.h>
51 
52 #include "internal.h"
53 
54 #ifndef arch_mmap_check
55 #define arch_mmap_check(addr, len, flags)	(0)
56 #endif
57 
58 #ifndef arch_rebalance_pgtables
59 #define arch_rebalance_pgtables(addr, len)		(addr)
60 #endif
61 
62 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
63 const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
64 const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
65 int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
66 #endif
67 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
68 const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
69 const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
70 int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
71 #endif
72 
73 static bool ignore_rlimit_data = true;
74 core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
75 
76 static void unmap_region(struct mm_struct *mm,
77 		struct vm_area_struct *vma, struct vm_area_struct *prev,
78 		unsigned long start, unsigned long end);
79 
80 /* description of effects of mapping type and prot in current implementation.
81  * this is due to the limited x86 page protection hardware.  The expected
82  * behavior is in parens:
83  *
84  * map_type	prot
85  *		PROT_NONE	PROT_READ	PROT_WRITE	PROT_EXEC
86  * MAP_SHARED	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
87  *		w: (no) no	w: (no) no	w: (yes) yes	w: (no) no
88  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
89  *
90  * MAP_PRIVATE	r: (no) no	r: (yes) yes	r: (no) yes	r: (no) yes
91  *		w: (no) no	w: (no) no	w: (copy) copy	w: (no) no
92  *		x: (no) no	x: (no) yes	x: (no) yes	x: (yes) yes
93  *
94  */
95 pgprot_t protection_map[16] = {
96 	__P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
97 	__S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
98 };
99 
100 pgprot_t vm_get_page_prot(unsigned long vm_flags)
101 {
102 	return __pgprot(pgprot_val(protection_map[vm_flags &
103 				(VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
104 			pgprot_val(arch_vm_get_page_prot(vm_flags)));
105 }
106 EXPORT_SYMBOL(vm_get_page_prot);
107 
108 static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
109 {
110 	return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
111 }
112 
113 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
114 void vma_set_page_prot(struct vm_area_struct *vma)
115 {
116 	unsigned long vm_flags = vma->vm_flags;
117 
118 	vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
119 	if (vma_wants_writenotify(vma)) {
120 		vm_flags &= ~VM_SHARED;
121 		vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
122 						     vm_flags);
123 	}
124 }
125 
126 
127 int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS;  /* heuristic overcommit */
128 int sysctl_overcommit_ratio __read_mostly = 50;	/* default is 50% */
129 unsigned long sysctl_overcommit_kbytes __read_mostly;
130 int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT;
131 unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */
132 unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */
133 /*
134  * Make sure vm_committed_as in one cacheline and not cacheline shared with
135  * other variables. It can be updated by several CPUs frequently.
136  */
137 struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp;
138 
139 /*
140  * The global memory commitment made in the system can be a metric
141  * that can be used to drive ballooning decisions when Linux is hosted
142  * as a guest. On Hyper-V, the host implements a policy engine for dynamically
143  * balancing memory across competing virtual machines that are hosted.
144  * Several metrics drive this policy engine including the guest reported
145  * memory commitment.
146  */
147 unsigned long vm_memory_committed(void)
148 {
149 	return percpu_counter_read_positive(&vm_committed_as);
150 }
151 EXPORT_SYMBOL_GPL(vm_memory_committed);
152 
153 /*
154  * Check that a process has enough memory to allocate a new virtual
155  * mapping. 0 means there is enough memory for the allocation to
156  * succeed and -ENOMEM implies there is not.
157  *
158  * We currently support three overcommit policies, which are set via the
159  * vm.overcommit_memory sysctl.  See Documentation/vm/overcommit-accounting
160  *
161  * Strict overcommit modes added 2002 Feb 26 by Alan Cox.
162  * Additional code 2002 Jul 20 by Robert Love.
163  *
164  * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise.
165  *
166  * Note this is a helper function intended to be used by LSMs which
167  * wish to use this logic.
168  */
169 int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin)
170 {
171 	long free, allowed, reserve;
172 
173 	VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) <
174 			-(s64)vm_committed_as_batch * num_online_cpus(),
175 			"memory commitment underflow");
176 
177 	vm_acct_memory(pages);
178 
179 	/*
180 	 * Sometimes we want to use more memory than we have
181 	 */
182 	if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS)
183 		return 0;
184 
185 	if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) {
186 		free = global_page_state(NR_FREE_PAGES);
187 		free += global_page_state(NR_FILE_PAGES);
188 
189 		/*
190 		 * shmem pages shouldn't be counted as free in this
191 		 * case, they can't be purged, only swapped out, and
192 		 * that won't affect the overall amount of available
193 		 * memory in the system.
194 		 */
195 		free -= global_page_state(NR_SHMEM);
196 
197 		free += get_nr_swap_pages();
198 
199 		/*
200 		 * Any slabs which are created with the
201 		 * SLAB_RECLAIM_ACCOUNT flag claim to have contents
202 		 * which are reclaimable, under pressure.  The dentry
203 		 * cache and most inode caches should fall into this
204 		 */
205 		free += global_page_state(NR_SLAB_RECLAIMABLE);
206 
207 		/*
208 		 * Leave reserved pages. The pages are not for anonymous pages.
209 		 */
210 		if (free <= totalreserve_pages)
211 			goto error;
212 		else
213 			free -= totalreserve_pages;
214 
215 		/*
216 		 * Reserve some for root
217 		 */
218 		if (!cap_sys_admin)
219 			free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
220 
221 		if (free > pages)
222 			return 0;
223 
224 		goto error;
225 	}
226 
227 	allowed = vm_commit_limit();
228 	/*
229 	 * Reserve some for root
230 	 */
231 	if (!cap_sys_admin)
232 		allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10);
233 
234 	/*
235 	 * Don't let a single process grow so big a user can't recover
236 	 */
237 	if (mm) {
238 		reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10);
239 		allowed -= min_t(long, mm->total_vm / 32, reserve);
240 	}
241 
242 	if (percpu_counter_read_positive(&vm_committed_as) < allowed)
243 		return 0;
244 error:
245 	vm_unacct_memory(pages);
246 
247 	return -ENOMEM;
248 }
249 
250 /*
251  * Requires inode->i_mapping->i_mmap_rwsem
252  */
253 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
254 		struct file *file, struct address_space *mapping)
255 {
256 	if (vma->vm_flags & VM_DENYWRITE)
257 		atomic_inc(&file_inode(file)->i_writecount);
258 	if (vma->vm_flags & VM_SHARED)
259 		mapping_unmap_writable(mapping);
260 
261 	flush_dcache_mmap_lock(mapping);
262 	vma_interval_tree_remove(vma, &mapping->i_mmap);
263 	flush_dcache_mmap_unlock(mapping);
264 }
265 
266 /*
267  * Unlink a file-based vm structure from its interval tree, to hide
268  * vma from rmap and vmtruncate before freeing its page tables.
269  */
270 void unlink_file_vma(struct vm_area_struct *vma)
271 {
272 	struct file *file = vma->vm_file;
273 
274 	if (file) {
275 		struct address_space *mapping = file->f_mapping;
276 		i_mmap_lock_write(mapping);
277 		__remove_shared_vm_struct(vma, file, mapping);
278 		i_mmap_unlock_write(mapping);
279 	}
280 }
281 
282 /*
283  * Close a vm structure and free it, returning the next.
284  */
285 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
286 {
287 	struct vm_area_struct *next = vma->vm_next;
288 
289 	might_sleep();
290 	if (vma->vm_ops && vma->vm_ops->close)
291 		vma->vm_ops->close(vma);
292 	if (vma->vm_file)
293 		fput(vma->vm_file);
294 	mpol_put(vma_policy(vma));
295 	kmem_cache_free(vm_area_cachep, vma);
296 	return next;
297 }
298 
299 static unsigned long do_brk(unsigned long addr, unsigned long len);
300 
301 SYSCALL_DEFINE1(brk, unsigned long, brk)
302 {
303 	unsigned long retval;
304 	unsigned long newbrk, oldbrk;
305 	struct mm_struct *mm = current->mm;
306 	unsigned long min_brk;
307 	bool populate;
308 
309 	down_write(&mm->mmap_sem);
310 
311 #ifdef CONFIG_COMPAT_BRK
312 	/*
313 	 * CONFIG_COMPAT_BRK can still be overridden by setting
314 	 * randomize_va_space to 2, which will still cause mm->start_brk
315 	 * to be arbitrarily shifted
316 	 */
317 	if (current->brk_randomized)
318 		min_brk = mm->start_brk;
319 	else
320 		min_brk = mm->end_data;
321 #else
322 	min_brk = mm->start_brk;
323 #endif
324 	if (brk < min_brk)
325 		goto out;
326 
327 	/*
328 	 * Check against rlimit here. If this check is done later after the test
329 	 * of oldbrk with newbrk then it can escape the test and let the data
330 	 * segment grow beyond its set limit the in case where the limit is
331 	 * not page aligned -Ram Gupta
332 	 */
333 	if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
334 			      mm->end_data, mm->start_data))
335 		goto out;
336 
337 	newbrk = PAGE_ALIGN(brk);
338 	oldbrk = PAGE_ALIGN(mm->brk);
339 	if (oldbrk == newbrk)
340 		goto set_brk;
341 
342 	/* Always allow shrinking brk. */
343 	if (brk <= mm->brk) {
344 		if (!do_munmap(mm, newbrk, oldbrk-newbrk))
345 			goto set_brk;
346 		goto out;
347 	}
348 
349 	/* Check against existing mmap mappings. */
350 	if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
351 		goto out;
352 
353 	/* Ok, looks good - let it rip. */
354 	if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
355 		goto out;
356 
357 set_brk:
358 	mm->brk = brk;
359 	populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
360 	up_write(&mm->mmap_sem);
361 	if (populate)
362 		mm_populate(oldbrk, newbrk - oldbrk);
363 	return brk;
364 
365 out:
366 	retval = mm->brk;
367 	up_write(&mm->mmap_sem);
368 	return retval;
369 }
370 
371 static long vma_compute_subtree_gap(struct vm_area_struct *vma)
372 {
373 	unsigned long max, subtree_gap;
374 	max = vma->vm_start;
375 	if (vma->vm_prev)
376 		max -= vma->vm_prev->vm_end;
377 	if (vma->vm_rb.rb_left) {
378 		subtree_gap = rb_entry(vma->vm_rb.rb_left,
379 				struct vm_area_struct, vm_rb)->rb_subtree_gap;
380 		if (subtree_gap > max)
381 			max = subtree_gap;
382 	}
383 	if (vma->vm_rb.rb_right) {
384 		subtree_gap = rb_entry(vma->vm_rb.rb_right,
385 				struct vm_area_struct, vm_rb)->rb_subtree_gap;
386 		if (subtree_gap > max)
387 			max = subtree_gap;
388 	}
389 	return max;
390 }
391 
392 #ifdef CONFIG_DEBUG_VM_RB
393 static int browse_rb(struct rb_root *root)
394 {
395 	int i = 0, j, bug = 0;
396 	struct rb_node *nd, *pn = NULL;
397 	unsigned long prev = 0, pend = 0;
398 
399 	for (nd = rb_first(root); nd; nd = rb_next(nd)) {
400 		struct vm_area_struct *vma;
401 		vma = rb_entry(nd, struct vm_area_struct, vm_rb);
402 		if (vma->vm_start < prev) {
403 			pr_emerg("vm_start %lx < prev %lx\n",
404 				  vma->vm_start, prev);
405 			bug = 1;
406 		}
407 		if (vma->vm_start < pend) {
408 			pr_emerg("vm_start %lx < pend %lx\n",
409 				  vma->vm_start, pend);
410 			bug = 1;
411 		}
412 		if (vma->vm_start > vma->vm_end) {
413 			pr_emerg("vm_start %lx > vm_end %lx\n",
414 				  vma->vm_start, vma->vm_end);
415 			bug = 1;
416 		}
417 		if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
418 			pr_emerg("free gap %lx, correct %lx\n",
419 			       vma->rb_subtree_gap,
420 			       vma_compute_subtree_gap(vma));
421 			bug = 1;
422 		}
423 		i++;
424 		pn = nd;
425 		prev = vma->vm_start;
426 		pend = vma->vm_end;
427 	}
428 	j = 0;
429 	for (nd = pn; nd; nd = rb_prev(nd))
430 		j++;
431 	if (i != j) {
432 		pr_emerg("backwards %d, forwards %d\n", j, i);
433 		bug = 1;
434 	}
435 	return bug ? -1 : i;
436 }
437 
438 static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
439 {
440 	struct rb_node *nd;
441 
442 	for (nd = rb_first(root); nd; nd = rb_next(nd)) {
443 		struct vm_area_struct *vma;
444 		vma = rb_entry(nd, struct vm_area_struct, vm_rb);
445 		VM_BUG_ON_VMA(vma != ignore &&
446 			vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
447 			vma);
448 	}
449 }
450 
451 static void validate_mm(struct mm_struct *mm)
452 {
453 	int bug = 0;
454 	int i = 0;
455 	unsigned long highest_address = 0;
456 	struct vm_area_struct *vma = mm->mmap;
457 
458 	while (vma) {
459 		struct anon_vma_chain *avc;
460 
461 		vma_lock_anon_vma(vma);
462 		list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
463 			anon_vma_interval_tree_verify(avc);
464 		vma_unlock_anon_vma(vma);
465 		highest_address = vma->vm_end;
466 		vma = vma->vm_next;
467 		i++;
468 	}
469 	if (i != mm->map_count) {
470 		pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
471 		bug = 1;
472 	}
473 	if (highest_address != mm->highest_vm_end) {
474 		pr_emerg("mm->highest_vm_end %lx, found %lx\n",
475 			  mm->highest_vm_end, highest_address);
476 		bug = 1;
477 	}
478 	i = browse_rb(&mm->mm_rb);
479 	if (i != mm->map_count) {
480 		if (i != -1)
481 			pr_emerg("map_count %d rb %d\n", mm->map_count, i);
482 		bug = 1;
483 	}
484 	VM_BUG_ON_MM(bug, mm);
485 }
486 #else
487 #define validate_mm_rb(root, ignore) do { } while (0)
488 #define validate_mm(mm) do { } while (0)
489 #endif
490 
491 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
492 		     unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
493 
494 /*
495  * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
496  * vma->vm_prev->vm_end values changed, without modifying the vma's position
497  * in the rbtree.
498  */
499 static void vma_gap_update(struct vm_area_struct *vma)
500 {
501 	/*
502 	 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
503 	 * function that does exacltly what we want.
504 	 */
505 	vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
506 }
507 
508 static inline void vma_rb_insert(struct vm_area_struct *vma,
509 				 struct rb_root *root)
510 {
511 	/* All rb_subtree_gap values must be consistent prior to insertion */
512 	validate_mm_rb(root, NULL);
513 
514 	rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
515 }
516 
517 static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
518 {
519 	/*
520 	 * All rb_subtree_gap values must be consistent prior to erase,
521 	 * with the possible exception of the vma being erased.
522 	 */
523 	validate_mm_rb(root, vma);
524 
525 	/*
526 	 * Note rb_erase_augmented is a fairly large inline function,
527 	 * so make sure we instantiate it only once with our desired
528 	 * augmented rbtree callbacks.
529 	 */
530 	rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
531 }
532 
533 /*
534  * vma has some anon_vma assigned, and is already inserted on that
535  * anon_vma's interval trees.
536  *
537  * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
538  * vma must be removed from the anon_vma's interval trees using
539  * anon_vma_interval_tree_pre_update_vma().
540  *
541  * After the update, the vma will be reinserted using
542  * anon_vma_interval_tree_post_update_vma().
543  *
544  * The entire update must be protected by exclusive mmap_sem and by
545  * the root anon_vma's mutex.
546  */
547 static inline void
548 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
549 {
550 	struct anon_vma_chain *avc;
551 
552 	list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
553 		anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
554 }
555 
556 static inline void
557 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
558 {
559 	struct anon_vma_chain *avc;
560 
561 	list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
562 		anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
563 }
564 
565 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
566 		unsigned long end, struct vm_area_struct **pprev,
567 		struct rb_node ***rb_link, struct rb_node **rb_parent)
568 {
569 	struct rb_node **__rb_link, *__rb_parent, *rb_prev;
570 
571 	__rb_link = &mm->mm_rb.rb_node;
572 	rb_prev = __rb_parent = NULL;
573 
574 	while (*__rb_link) {
575 		struct vm_area_struct *vma_tmp;
576 
577 		__rb_parent = *__rb_link;
578 		vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
579 
580 		if (vma_tmp->vm_end > addr) {
581 			/* Fail if an existing vma overlaps the area */
582 			if (vma_tmp->vm_start < end)
583 				return -ENOMEM;
584 			__rb_link = &__rb_parent->rb_left;
585 		} else {
586 			rb_prev = __rb_parent;
587 			__rb_link = &__rb_parent->rb_right;
588 		}
589 	}
590 
591 	*pprev = NULL;
592 	if (rb_prev)
593 		*pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
594 	*rb_link = __rb_link;
595 	*rb_parent = __rb_parent;
596 	return 0;
597 }
598 
599 static unsigned long count_vma_pages_range(struct mm_struct *mm,
600 		unsigned long addr, unsigned long end)
601 {
602 	unsigned long nr_pages = 0;
603 	struct vm_area_struct *vma;
604 
605 	/* Find first overlaping mapping */
606 	vma = find_vma_intersection(mm, addr, end);
607 	if (!vma)
608 		return 0;
609 
610 	nr_pages = (min(end, vma->vm_end) -
611 		max(addr, vma->vm_start)) >> PAGE_SHIFT;
612 
613 	/* Iterate over the rest of the overlaps */
614 	for (vma = vma->vm_next; vma; vma = vma->vm_next) {
615 		unsigned long overlap_len;
616 
617 		if (vma->vm_start > end)
618 			break;
619 
620 		overlap_len = min(end, vma->vm_end) - vma->vm_start;
621 		nr_pages += overlap_len >> PAGE_SHIFT;
622 	}
623 
624 	return nr_pages;
625 }
626 
627 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
628 		struct rb_node **rb_link, struct rb_node *rb_parent)
629 {
630 	/* Update tracking information for the gap following the new vma. */
631 	if (vma->vm_next)
632 		vma_gap_update(vma->vm_next);
633 	else
634 		mm->highest_vm_end = vma->vm_end;
635 
636 	/*
637 	 * vma->vm_prev wasn't known when we followed the rbtree to find the
638 	 * correct insertion point for that vma. As a result, we could not
639 	 * update the vma vm_rb parents rb_subtree_gap values on the way down.
640 	 * So, we first insert the vma with a zero rb_subtree_gap value
641 	 * (to be consistent with what we did on the way down), and then
642 	 * immediately update the gap to the correct value. Finally we
643 	 * rebalance the rbtree after all augmented values have been set.
644 	 */
645 	rb_link_node(&vma->vm_rb, rb_parent, rb_link);
646 	vma->rb_subtree_gap = 0;
647 	vma_gap_update(vma);
648 	vma_rb_insert(vma, &mm->mm_rb);
649 }
650 
651 static void __vma_link_file(struct vm_area_struct *vma)
652 {
653 	struct file *file;
654 
655 	file = vma->vm_file;
656 	if (file) {
657 		struct address_space *mapping = file->f_mapping;
658 
659 		if (vma->vm_flags & VM_DENYWRITE)
660 			atomic_dec(&file_inode(file)->i_writecount);
661 		if (vma->vm_flags & VM_SHARED)
662 			atomic_inc(&mapping->i_mmap_writable);
663 
664 		flush_dcache_mmap_lock(mapping);
665 		vma_interval_tree_insert(vma, &mapping->i_mmap);
666 		flush_dcache_mmap_unlock(mapping);
667 	}
668 }
669 
670 static void
671 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
672 	struct vm_area_struct *prev, struct rb_node **rb_link,
673 	struct rb_node *rb_parent)
674 {
675 	__vma_link_list(mm, vma, prev, rb_parent);
676 	__vma_link_rb(mm, vma, rb_link, rb_parent);
677 }
678 
679 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
680 			struct vm_area_struct *prev, struct rb_node **rb_link,
681 			struct rb_node *rb_parent)
682 {
683 	struct address_space *mapping = NULL;
684 
685 	if (vma->vm_file) {
686 		mapping = vma->vm_file->f_mapping;
687 		i_mmap_lock_write(mapping);
688 	}
689 
690 	__vma_link(mm, vma, prev, rb_link, rb_parent);
691 	__vma_link_file(vma);
692 
693 	if (mapping)
694 		i_mmap_unlock_write(mapping);
695 
696 	mm->map_count++;
697 	validate_mm(mm);
698 }
699 
700 /*
701  * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
702  * mm's list and rbtree.  It has already been inserted into the interval tree.
703  */
704 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
705 {
706 	struct vm_area_struct *prev;
707 	struct rb_node **rb_link, *rb_parent;
708 
709 	if (find_vma_links(mm, vma->vm_start, vma->vm_end,
710 			   &prev, &rb_link, &rb_parent))
711 		BUG();
712 	__vma_link(mm, vma, prev, rb_link, rb_parent);
713 	mm->map_count++;
714 }
715 
716 static inline void
717 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
718 		struct vm_area_struct *prev)
719 {
720 	struct vm_area_struct *next;
721 
722 	vma_rb_erase(vma, &mm->mm_rb);
723 	prev->vm_next = next = vma->vm_next;
724 	if (next)
725 		next->vm_prev = prev;
726 
727 	/* Kill the cache */
728 	vmacache_invalidate(mm);
729 }
730 
731 /*
732  * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
733  * is already present in an i_mmap tree without adjusting the tree.
734  * The following helper function should be used when such adjustments
735  * are necessary.  The "insert" vma (if any) is to be inserted
736  * before we drop the necessary locks.
737  */
738 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
739 	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
740 {
741 	struct mm_struct *mm = vma->vm_mm;
742 	struct vm_area_struct *next = vma->vm_next;
743 	struct vm_area_struct *importer = NULL;
744 	struct address_space *mapping = NULL;
745 	struct rb_root *root = NULL;
746 	struct anon_vma *anon_vma = NULL;
747 	struct file *file = vma->vm_file;
748 	bool start_changed = false, end_changed = false;
749 	long adjust_next = 0;
750 	int remove_next = 0;
751 
752 	if (next && !insert) {
753 		struct vm_area_struct *exporter = NULL;
754 
755 		if (end >= next->vm_end) {
756 			/*
757 			 * vma expands, overlapping all the next, and
758 			 * perhaps the one after too (mprotect case 6).
759 			 */
760 again:			remove_next = 1 + (end > next->vm_end);
761 			end = next->vm_end;
762 			exporter = next;
763 			importer = vma;
764 		} else if (end > next->vm_start) {
765 			/*
766 			 * vma expands, overlapping part of the next:
767 			 * mprotect case 5 shifting the boundary up.
768 			 */
769 			adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
770 			exporter = next;
771 			importer = vma;
772 		} else if (end < vma->vm_end) {
773 			/*
774 			 * vma shrinks, and !insert tells it's not
775 			 * split_vma inserting another: so it must be
776 			 * mprotect case 4 shifting the boundary down.
777 			 */
778 			adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
779 			exporter = vma;
780 			importer = next;
781 		}
782 
783 		/*
784 		 * Easily overlooked: when mprotect shifts the boundary,
785 		 * make sure the expanding vma has anon_vma set if the
786 		 * shrinking vma had, to cover any anon pages imported.
787 		 */
788 		if (exporter && exporter->anon_vma && !importer->anon_vma) {
789 			int error;
790 
791 			importer->anon_vma = exporter->anon_vma;
792 			error = anon_vma_clone(importer, exporter);
793 			if (error)
794 				return error;
795 		}
796 	}
797 
798 	if (file) {
799 		mapping = file->f_mapping;
800 		root = &mapping->i_mmap;
801 		uprobe_munmap(vma, vma->vm_start, vma->vm_end);
802 
803 		if (adjust_next)
804 			uprobe_munmap(next, next->vm_start, next->vm_end);
805 
806 		i_mmap_lock_write(mapping);
807 		if (insert) {
808 			/*
809 			 * Put into interval tree now, so instantiated pages
810 			 * are visible to arm/parisc __flush_dcache_page
811 			 * throughout; but we cannot insert into address
812 			 * space until vma start or end is updated.
813 			 */
814 			__vma_link_file(insert);
815 		}
816 	}
817 
818 	vma_adjust_trans_huge(vma, start, end, adjust_next);
819 
820 	anon_vma = vma->anon_vma;
821 	if (!anon_vma && adjust_next)
822 		anon_vma = next->anon_vma;
823 	if (anon_vma) {
824 		VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
825 			  anon_vma != next->anon_vma, next);
826 		anon_vma_lock_write(anon_vma);
827 		anon_vma_interval_tree_pre_update_vma(vma);
828 		if (adjust_next)
829 			anon_vma_interval_tree_pre_update_vma(next);
830 	}
831 
832 	if (root) {
833 		flush_dcache_mmap_lock(mapping);
834 		vma_interval_tree_remove(vma, root);
835 		if (adjust_next)
836 			vma_interval_tree_remove(next, root);
837 	}
838 
839 	if (start != vma->vm_start) {
840 		vma->vm_start = start;
841 		start_changed = true;
842 	}
843 	if (end != vma->vm_end) {
844 		vma->vm_end = end;
845 		end_changed = true;
846 	}
847 	vma->vm_pgoff = pgoff;
848 	if (adjust_next) {
849 		next->vm_start += adjust_next << PAGE_SHIFT;
850 		next->vm_pgoff += adjust_next;
851 	}
852 
853 	if (root) {
854 		if (adjust_next)
855 			vma_interval_tree_insert(next, root);
856 		vma_interval_tree_insert(vma, root);
857 		flush_dcache_mmap_unlock(mapping);
858 	}
859 
860 	if (remove_next) {
861 		/*
862 		 * vma_merge has merged next into vma, and needs
863 		 * us to remove next before dropping the locks.
864 		 */
865 		__vma_unlink(mm, next, vma);
866 		if (file)
867 			__remove_shared_vm_struct(next, file, mapping);
868 	} else if (insert) {
869 		/*
870 		 * split_vma has split insert from vma, and needs
871 		 * us to insert it before dropping the locks
872 		 * (it may either follow vma or precede it).
873 		 */
874 		__insert_vm_struct(mm, insert);
875 	} else {
876 		if (start_changed)
877 			vma_gap_update(vma);
878 		if (end_changed) {
879 			if (!next)
880 				mm->highest_vm_end = end;
881 			else if (!adjust_next)
882 				vma_gap_update(next);
883 		}
884 	}
885 
886 	if (anon_vma) {
887 		anon_vma_interval_tree_post_update_vma(vma);
888 		if (adjust_next)
889 			anon_vma_interval_tree_post_update_vma(next);
890 		anon_vma_unlock_write(anon_vma);
891 	}
892 	if (mapping)
893 		i_mmap_unlock_write(mapping);
894 
895 	if (root) {
896 		uprobe_mmap(vma);
897 
898 		if (adjust_next)
899 			uprobe_mmap(next);
900 	}
901 
902 	if (remove_next) {
903 		if (file) {
904 			uprobe_munmap(next, next->vm_start, next->vm_end);
905 			fput(file);
906 		}
907 		if (next->anon_vma)
908 			anon_vma_merge(vma, next);
909 		mm->map_count--;
910 		mpol_put(vma_policy(next));
911 		kmem_cache_free(vm_area_cachep, next);
912 		/*
913 		 * In mprotect's case 6 (see comments on vma_merge),
914 		 * we must remove another next too. It would clutter
915 		 * up the code too much to do both in one go.
916 		 */
917 		next = vma->vm_next;
918 		if (remove_next == 2)
919 			goto again;
920 		else if (next)
921 			vma_gap_update(next);
922 		else
923 			mm->highest_vm_end = end;
924 	}
925 	if (insert && file)
926 		uprobe_mmap(insert);
927 
928 	validate_mm(mm);
929 
930 	return 0;
931 }
932 
933 /*
934  * If the vma has a ->close operation then the driver probably needs to release
935  * per-vma resources, so we don't attempt to merge those.
936  */
937 static inline int is_mergeable_vma(struct vm_area_struct *vma,
938 				struct file *file, unsigned long vm_flags,
939 				struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
940 {
941 	/*
942 	 * VM_SOFTDIRTY should not prevent from VMA merging, if we
943 	 * match the flags but dirty bit -- the caller should mark
944 	 * merged VMA as dirty. If dirty bit won't be excluded from
945 	 * comparison, we increase pressue on the memory system forcing
946 	 * the kernel to generate new VMAs when old one could be
947 	 * extended instead.
948 	 */
949 	if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
950 		return 0;
951 	if (vma->vm_file != file)
952 		return 0;
953 	if (vma->vm_ops && vma->vm_ops->close)
954 		return 0;
955 	if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
956 		return 0;
957 	return 1;
958 }
959 
960 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
961 					struct anon_vma *anon_vma2,
962 					struct vm_area_struct *vma)
963 {
964 	/*
965 	 * The list_is_singular() test is to avoid merging VMA cloned from
966 	 * parents. This can improve scalability caused by anon_vma lock.
967 	 */
968 	if ((!anon_vma1 || !anon_vma2) && (!vma ||
969 		list_is_singular(&vma->anon_vma_chain)))
970 		return 1;
971 	return anon_vma1 == anon_vma2;
972 }
973 
974 /*
975  * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
976  * in front of (at a lower virtual address and file offset than) the vma.
977  *
978  * We cannot merge two vmas if they have differently assigned (non-NULL)
979  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
980  *
981  * We don't check here for the merged mmap wrapping around the end of pagecache
982  * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
983  * wrap, nor mmaps which cover the final page at index -1UL.
984  */
985 static int
986 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
987 		     struct anon_vma *anon_vma, struct file *file,
988 		     pgoff_t vm_pgoff,
989 		     struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
990 {
991 	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
992 	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
993 		if (vma->vm_pgoff == vm_pgoff)
994 			return 1;
995 	}
996 	return 0;
997 }
998 
999 /*
1000  * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1001  * beyond (at a higher virtual address and file offset than) the vma.
1002  *
1003  * We cannot merge two vmas if they have differently assigned (non-NULL)
1004  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1005  */
1006 static int
1007 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
1008 		    struct anon_vma *anon_vma, struct file *file,
1009 		    pgoff_t vm_pgoff,
1010 		    struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1011 {
1012 	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1013 	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1014 		pgoff_t vm_pglen;
1015 		vm_pglen = vma_pages(vma);
1016 		if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1017 			return 1;
1018 	}
1019 	return 0;
1020 }
1021 
1022 /*
1023  * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1024  * whether that can be merged with its predecessor or its successor.
1025  * Or both (it neatly fills a hole).
1026  *
1027  * In most cases - when called for mmap, brk or mremap - [addr,end) is
1028  * certain not to be mapped by the time vma_merge is called; but when
1029  * called for mprotect, it is certain to be already mapped (either at
1030  * an offset within prev, or at the start of next), and the flags of
1031  * this area are about to be changed to vm_flags - and the no-change
1032  * case has already been eliminated.
1033  *
1034  * The following mprotect cases have to be considered, where AAAA is
1035  * the area passed down from mprotect_fixup, never extending beyond one
1036  * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1037  *
1038  *     AAAA             AAAA                AAAA          AAAA
1039  *    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPPPNNNNNN    PPPPNNNNXXXX
1040  *    cannot merge    might become    might become    might become
1041  *                    PPNNNNNNNNNN    PPPPPPPPPPNN    PPPPPPPPPPPP 6 or
1042  *    mmap, brk or    case 4 below    case 5 below    PPPPPPPPXXXX 7 or
1043  *    mremap move:                                    PPPPNNNNNNNN 8
1044  *        AAAA
1045  *    PPPP    NNNN    PPPPPPPPPPPP    PPPPPPPPNNNN    PPPPNNNNNNNN
1046  *    might become    case 1 below    case 2 below    case 3 below
1047  *
1048  * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
1049  * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
1050  */
1051 struct vm_area_struct *vma_merge(struct mm_struct *mm,
1052 			struct vm_area_struct *prev, unsigned long addr,
1053 			unsigned long end, unsigned long vm_flags,
1054 			struct anon_vma *anon_vma, struct file *file,
1055 			pgoff_t pgoff, struct mempolicy *policy,
1056 			struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1057 {
1058 	pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1059 	struct vm_area_struct *area, *next;
1060 	int err;
1061 
1062 	/*
1063 	 * We later require that vma->vm_flags == vm_flags,
1064 	 * so this tests vma->vm_flags & VM_SPECIAL, too.
1065 	 */
1066 	if (vm_flags & VM_SPECIAL)
1067 		return NULL;
1068 
1069 	if (prev)
1070 		next = prev->vm_next;
1071 	else
1072 		next = mm->mmap;
1073 	area = next;
1074 	if (next && next->vm_end == end)		/* cases 6, 7, 8 */
1075 		next = next->vm_next;
1076 
1077 	/*
1078 	 * Can it merge with the predecessor?
1079 	 */
1080 	if (prev && prev->vm_end == addr &&
1081 			mpol_equal(vma_policy(prev), policy) &&
1082 			can_vma_merge_after(prev, vm_flags,
1083 					    anon_vma, file, pgoff,
1084 					    vm_userfaultfd_ctx)) {
1085 		/*
1086 		 * OK, it can.  Can we now merge in the successor as well?
1087 		 */
1088 		if (next && end == next->vm_start &&
1089 				mpol_equal(policy, vma_policy(next)) &&
1090 				can_vma_merge_before(next, vm_flags,
1091 						     anon_vma, file,
1092 						     pgoff+pglen,
1093 						     vm_userfaultfd_ctx) &&
1094 				is_mergeable_anon_vma(prev->anon_vma,
1095 						      next->anon_vma, NULL)) {
1096 							/* cases 1, 6 */
1097 			err = vma_adjust(prev, prev->vm_start,
1098 				next->vm_end, prev->vm_pgoff, NULL);
1099 		} else					/* cases 2, 5, 7 */
1100 			err = vma_adjust(prev, prev->vm_start,
1101 				end, prev->vm_pgoff, NULL);
1102 		if (err)
1103 			return NULL;
1104 		khugepaged_enter_vma_merge(prev, vm_flags);
1105 		return prev;
1106 	}
1107 
1108 	/*
1109 	 * Can this new request be merged in front of next?
1110 	 */
1111 	if (next && end == next->vm_start &&
1112 			mpol_equal(policy, vma_policy(next)) &&
1113 			can_vma_merge_before(next, vm_flags,
1114 					     anon_vma, file, pgoff+pglen,
1115 					     vm_userfaultfd_ctx)) {
1116 		if (prev && addr < prev->vm_end)	/* case 4 */
1117 			err = vma_adjust(prev, prev->vm_start,
1118 				addr, prev->vm_pgoff, NULL);
1119 		else					/* cases 3, 8 */
1120 			err = vma_adjust(area, addr, next->vm_end,
1121 				next->vm_pgoff - pglen, NULL);
1122 		if (err)
1123 			return NULL;
1124 		khugepaged_enter_vma_merge(area, vm_flags);
1125 		return area;
1126 	}
1127 
1128 	return NULL;
1129 }
1130 
1131 /*
1132  * Rough compatbility check to quickly see if it's even worth looking
1133  * at sharing an anon_vma.
1134  *
1135  * They need to have the same vm_file, and the flags can only differ
1136  * in things that mprotect may change.
1137  *
1138  * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1139  * we can merge the two vma's. For example, we refuse to merge a vma if
1140  * there is a vm_ops->close() function, because that indicates that the
1141  * driver is doing some kind of reference counting. But that doesn't
1142  * really matter for the anon_vma sharing case.
1143  */
1144 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1145 {
1146 	return a->vm_end == b->vm_start &&
1147 		mpol_equal(vma_policy(a), vma_policy(b)) &&
1148 		a->vm_file == b->vm_file &&
1149 		!((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
1150 		b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1151 }
1152 
1153 /*
1154  * Do some basic sanity checking to see if we can re-use the anon_vma
1155  * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1156  * the same as 'old', the other will be the new one that is trying
1157  * to share the anon_vma.
1158  *
1159  * NOTE! This runs with mm_sem held for reading, so it is possible that
1160  * the anon_vma of 'old' is concurrently in the process of being set up
1161  * by another page fault trying to merge _that_. But that's ok: if it
1162  * is being set up, that automatically means that it will be a singleton
1163  * acceptable for merging, so we can do all of this optimistically. But
1164  * we do that READ_ONCE() to make sure that we never re-load the pointer.
1165  *
1166  * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1167  * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1168  * is to return an anon_vma that is "complex" due to having gone through
1169  * a fork).
1170  *
1171  * We also make sure that the two vma's are compatible (adjacent,
1172  * and with the same memory policies). That's all stable, even with just
1173  * a read lock on the mm_sem.
1174  */
1175 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1176 {
1177 	if (anon_vma_compatible(a, b)) {
1178 		struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1179 
1180 		if (anon_vma && list_is_singular(&old->anon_vma_chain))
1181 			return anon_vma;
1182 	}
1183 	return NULL;
1184 }
1185 
1186 /*
1187  * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1188  * neighbouring vmas for a suitable anon_vma, before it goes off
1189  * to allocate a new anon_vma.  It checks because a repetitive
1190  * sequence of mprotects and faults may otherwise lead to distinct
1191  * anon_vmas being allocated, preventing vma merge in subsequent
1192  * mprotect.
1193  */
1194 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1195 {
1196 	struct anon_vma *anon_vma;
1197 	struct vm_area_struct *near;
1198 
1199 	near = vma->vm_next;
1200 	if (!near)
1201 		goto try_prev;
1202 
1203 	anon_vma = reusable_anon_vma(near, vma, near);
1204 	if (anon_vma)
1205 		return anon_vma;
1206 try_prev:
1207 	near = vma->vm_prev;
1208 	if (!near)
1209 		goto none;
1210 
1211 	anon_vma = reusable_anon_vma(near, near, vma);
1212 	if (anon_vma)
1213 		return anon_vma;
1214 none:
1215 	/*
1216 	 * There's no absolute need to look only at touching neighbours:
1217 	 * we could search further afield for "compatible" anon_vmas.
1218 	 * But it would probably just be a waste of time searching,
1219 	 * or lead to too many vmas hanging off the same anon_vma.
1220 	 * We're trying to allow mprotect remerging later on,
1221 	 * not trying to minimize memory used for anon_vmas.
1222 	 */
1223 	return NULL;
1224 }
1225 
1226 /*
1227  * If a hint addr is less than mmap_min_addr change hint to be as
1228  * low as possible but still greater than mmap_min_addr
1229  */
1230 static inline unsigned long round_hint_to_min(unsigned long hint)
1231 {
1232 	hint &= PAGE_MASK;
1233 	if (((void *)hint != NULL) &&
1234 	    (hint < mmap_min_addr))
1235 		return PAGE_ALIGN(mmap_min_addr);
1236 	return hint;
1237 }
1238 
1239 static inline int mlock_future_check(struct mm_struct *mm,
1240 				     unsigned long flags,
1241 				     unsigned long len)
1242 {
1243 	unsigned long locked, lock_limit;
1244 
1245 	/*  mlock MCL_FUTURE? */
1246 	if (flags & VM_LOCKED) {
1247 		locked = len >> PAGE_SHIFT;
1248 		locked += mm->locked_vm;
1249 		lock_limit = rlimit(RLIMIT_MEMLOCK);
1250 		lock_limit >>= PAGE_SHIFT;
1251 		if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1252 			return -EAGAIN;
1253 	}
1254 	return 0;
1255 }
1256 
1257 /*
1258  * The caller must hold down_write(&current->mm->mmap_sem).
1259  */
1260 unsigned long do_mmap(struct file *file, unsigned long addr,
1261 			unsigned long len, unsigned long prot,
1262 			unsigned long flags, vm_flags_t vm_flags,
1263 			unsigned long pgoff, unsigned long *populate)
1264 {
1265 	struct mm_struct *mm = current->mm;
1266 
1267 	*populate = 0;
1268 
1269 	if (!len)
1270 		return -EINVAL;
1271 
1272 	/*
1273 	 * Does the application expect PROT_READ to imply PROT_EXEC?
1274 	 *
1275 	 * (the exception is when the underlying filesystem is noexec
1276 	 *  mounted, in which case we dont add PROT_EXEC.)
1277 	 */
1278 	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1279 		if (!(file && path_noexec(&file->f_path)))
1280 			prot |= PROT_EXEC;
1281 
1282 	if (!(flags & MAP_FIXED))
1283 		addr = round_hint_to_min(addr);
1284 
1285 	/* Careful about overflows.. */
1286 	len = PAGE_ALIGN(len);
1287 	if (!len)
1288 		return -ENOMEM;
1289 
1290 	/* offset overflow? */
1291 	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1292 		return -EOVERFLOW;
1293 
1294 	/* Too many mappings? */
1295 	if (mm->map_count > sysctl_max_map_count)
1296 		return -ENOMEM;
1297 
1298 	/* Obtain the address to map to. we verify (or select) it and ensure
1299 	 * that it represents a valid section of the address space.
1300 	 */
1301 	addr = get_unmapped_area(file, addr, len, pgoff, flags);
1302 	if (offset_in_page(addr))
1303 		return addr;
1304 
1305 	/* Do simple checking here so the lower-level routines won't have
1306 	 * to. we assume access permissions have been handled by the open
1307 	 * of the memory object, so we don't do any here.
1308 	 */
1309 	vm_flags |= calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) |
1310 			mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1311 
1312 	if (flags & MAP_LOCKED)
1313 		if (!can_do_mlock())
1314 			return -EPERM;
1315 
1316 	if (mlock_future_check(mm, vm_flags, len))
1317 		return -EAGAIN;
1318 
1319 	if (file) {
1320 		struct inode *inode = file_inode(file);
1321 
1322 		switch (flags & MAP_TYPE) {
1323 		case MAP_SHARED:
1324 			if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1325 				return -EACCES;
1326 
1327 			/*
1328 			 * Make sure we don't allow writing to an append-only
1329 			 * file..
1330 			 */
1331 			if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1332 				return -EACCES;
1333 
1334 			/*
1335 			 * Make sure there are no mandatory locks on the file.
1336 			 */
1337 			if (locks_verify_locked(file))
1338 				return -EAGAIN;
1339 
1340 			vm_flags |= VM_SHARED | VM_MAYSHARE;
1341 			if (!(file->f_mode & FMODE_WRITE))
1342 				vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1343 
1344 			/* fall through */
1345 		case MAP_PRIVATE:
1346 			if (!(file->f_mode & FMODE_READ))
1347 				return -EACCES;
1348 			if (path_noexec(&file->f_path)) {
1349 				if (vm_flags & VM_EXEC)
1350 					return -EPERM;
1351 				vm_flags &= ~VM_MAYEXEC;
1352 			}
1353 
1354 			if (!file->f_op->mmap)
1355 				return -ENODEV;
1356 			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1357 				return -EINVAL;
1358 			break;
1359 
1360 		default:
1361 			return -EINVAL;
1362 		}
1363 	} else {
1364 		switch (flags & MAP_TYPE) {
1365 		case MAP_SHARED:
1366 			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1367 				return -EINVAL;
1368 			/*
1369 			 * Ignore pgoff.
1370 			 */
1371 			pgoff = 0;
1372 			vm_flags |= VM_SHARED | VM_MAYSHARE;
1373 			break;
1374 		case MAP_PRIVATE:
1375 			/*
1376 			 * Set pgoff according to addr for anon_vma.
1377 			 */
1378 			pgoff = addr >> PAGE_SHIFT;
1379 			break;
1380 		default:
1381 			return -EINVAL;
1382 		}
1383 	}
1384 
1385 	/*
1386 	 * Set 'VM_NORESERVE' if we should not account for the
1387 	 * memory use of this mapping.
1388 	 */
1389 	if (flags & MAP_NORESERVE) {
1390 		/* We honor MAP_NORESERVE if allowed to overcommit */
1391 		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1392 			vm_flags |= VM_NORESERVE;
1393 
1394 		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
1395 		if (file && is_file_hugepages(file))
1396 			vm_flags |= VM_NORESERVE;
1397 	}
1398 
1399 	addr = mmap_region(file, addr, len, vm_flags, pgoff);
1400 	if (!IS_ERR_VALUE(addr) &&
1401 	    ((vm_flags & VM_LOCKED) ||
1402 	     (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1403 		*populate = len;
1404 	return addr;
1405 }
1406 
1407 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1408 		unsigned long, prot, unsigned long, flags,
1409 		unsigned long, fd, unsigned long, pgoff)
1410 {
1411 	struct file *file = NULL;
1412 	unsigned long retval;
1413 
1414 	if (!(flags & MAP_ANONYMOUS)) {
1415 		audit_mmap_fd(fd, flags);
1416 		file = fget(fd);
1417 		if (!file)
1418 			return -EBADF;
1419 		if (is_file_hugepages(file))
1420 			len = ALIGN(len, huge_page_size(hstate_file(file)));
1421 		retval = -EINVAL;
1422 		if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1423 			goto out_fput;
1424 	} else if (flags & MAP_HUGETLB) {
1425 		struct user_struct *user = NULL;
1426 		struct hstate *hs;
1427 
1428 		hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
1429 		if (!hs)
1430 			return -EINVAL;
1431 
1432 		len = ALIGN(len, huge_page_size(hs));
1433 		/*
1434 		 * VM_NORESERVE is used because the reservations will be
1435 		 * taken when vm_ops->mmap() is called
1436 		 * A dummy user value is used because we are not locking
1437 		 * memory so no accounting is necessary
1438 		 */
1439 		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1440 				VM_NORESERVE,
1441 				&user, HUGETLB_ANONHUGE_INODE,
1442 				(flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1443 		if (IS_ERR(file))
1444 			return PTR_ERR(file);
1445 	}
1446 
1447 	flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1448 
1449 	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1450 out_fput:
1451 	if (file)
1452 		fput(file);
1453 	return retval;
1454 }
1455 
1456 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1457 struct mmap_arg_struct {
1458 	unsigned long addr;
1459 	unsigned long len;
1460 	unsigned long prot;
1461 	unsigned long flags;
1462 	unsigned long fd;
1463 	unsigned long offset;
1464 };
1465 
1466 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1467 {
1468 	struct mmap_arg_struct a;
1469 
1470 	if (copy_from_user(&a, arg, sizeof(a)))
1471 		return -EFAULT;
1472 	if (offset_in_page(a.offset))
1473 		return -EINVAL;
1474 
1475 	return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1476 			      a.offset >> PAGE_SHIFT);
1477 }
1478 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1479 
1480 /*
1481  * Some shared mappigns will want the pages marked read-only
1482  * to track write events. If so, we'll downgrade vm_page_prot
1483  * to the private version (using protection_map[] without the
1484  * VM_SHARED bit).
1485  */
1486 int vma_wants_writenotify(struct vm_area_struct *vma)
1487 {
1488 	vm_flags_t vm_flags = vma->vm_flags;
1489 	const struct vm_operations_struct *vm_ops = vma->vm_ops;
1490 
1491 	/* If it was private or non-writable, the write bit is already clear */
1492 	if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1493 		return 0;
1494 
1495 	/* The backer wishes to know when pages are first written to? */
1496 	if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1497 		return 1;
1498 
1499 	/* The open routine did something to the protections that pgprot_modify
1500 	 * won't preserve? */
1501 	if (pgprot_val(vma->vm_page_prot) !=
1502 	    pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
1503 		return 0;
1504 
1505 	/* Do we need to track softdirty? */
1506 	if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1507 		return 1;
1508 
1509 	/* Specialty mapping? */
1510 	if (vm_flags & VM_PFNMAP)
1511 		return 0;
1512 
1513 	/* Can the mapping track the dirty pages? */
1514 	return vma->vm_file && vma->vm_file->f_mapping &&
1515 		mapping_cap_account_dirty(vma->vm_file->f_mapping);
1516 }
1517 
1518 /*
1519  * We account for memory if it's a private writeable mapping,
1520  * not hugepages and VM_NORESERVE wasn't set.
1521  */
1522 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1523 {
1524 	/*
1525 	 * hugetlb has its own accounting separate from the core VM
1526 	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1527 	 */
1528 	if (file && is_file_hugepages(file))
1529 		return 0;
1530 
1531 	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1532 }
1533 
1534 unsigned long mmap_region(struct file *file, unsigned long addr,
1535 		unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
1536 {
1537 	struct mm_struct *mm = current->mm;
1538 	struct vm_area_struct *vma, *prev;
1539 	int error;
1540 	struct rb_node **rb_link, *rb_parent;
1541 	unsigned long charged = 0;
1542 
1543 	/* Check against address space limit. */
1544 	if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
1545 		unsigned long nr_pages;
1546 
1547 		/*
1548 		 * MAP_FIXED may remove pages of mappings that intersects with
1549 		 * requested mapping. Account for the pages it would unmap.
1550 		 */
1551 		nr_pages = count_vma_pages_range(mm, addr, addr + len);
1552 
1553 		if (!may_expand_vm(mm, vm_flags,
1554 					(len >> PAGE_SHIFT) - nr_pages))
1555 			return -ENOMEM;
1556 	}
1557 
1558 	/* Clear old maps */
1559 	while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1560 			      &rb_parent)) {
1561 		if (do_munmap(mm, addr, len))
1562 			return -ENOMEM;
1563 	}
1564 
1565 	/*
1566 	 * Private writable mapping: check memory availability
1567 	 */
1568 	if (accountable_mapping(file, vm_flags)) {
1569 		charged = len >> PAGE_SHIFT;
1570 		if (security_vm_enough_memory_mm(mm, charged))
1571 			return -ENOMEM;
1572 		vm_flags |= VM_ACCOUNT;
1573 	}
1574 
1575 	/*
1576 	 * Can we just expand an old mapping?
1577 	 */
1578 	vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1579 			NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
1580 	if (vma)
1581 		goto out;
1582 
1583 	/*
1584 	 * Determine the object being mapped and call the appropriate
1585 	 * specific mapper. the address has already been validated, but
1586 	 * not unmapped, but the maps are removed from the list.
1587 	 */
1588 	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1589 	if (!vma) {
1590 		error = -ENOMEM;
1591 		goto unacct_error;
1592 	}
1593 
1594 	vma->vm_mm = mm;
1595 	vma->vm_start = addr;
1596 	vma->vm_end = addr + len;
1597 	vma->vm_flags = vm_flags;
1598 	vma->vm_page_prot = vm_get_page_prot(vm_flags);
1599 	vma->vm_pgoff = pgoff;
1600 	INIT_LIST_HEAD(&vma->anon_vma_chain);
1601 
1602 	if (file) {
1603 		if (vm_flags & VM_DENYWRITE) {
1604 			error = deny_write_access(file);
1605 			if (error)
1606 				goto free_vma;
1607 		}
1608 		if (vm_flags & VM_SHARED) {
1609 			error = mapping_map_writable(file->f_mapping);
1610 			if (error)
1611 				goto allow_write_and_free_vma;
1612 		}
1613 
1614 		/* ->mmap() can change vma->vm_file, but must guarantee that
1615 		 * vma_link() below can deny write-access if VM_DENYWRITE is set
1616 		 * and map writably if VM_SHARED is set. This usually means the
1617 		 * new file must not have been exposed to user-space, yet.
1618 		 */
1619 		vma->vm_file = get_file(file);
1620 		error = file->f_op->mmap(file, vma);
1621 		if (error)
1622 			goto unmap_and_free_vma;
1623 
1624 		/* Can addr have changed??
1625 		 *
1626 		 * Answer: Yes, several device drivers can do it in their
1627 		 *         f_op->mmap method. -DaveM
1628 		 * Bug: If addr is changed, prev, rb_link, rb_parent should
1629 		 *      be updated for vma_link()
1630 		 */
1631 		WARN_ON_ONCE(addr != vma->vm_start);
1632 
1633 		addr = vma->vm_start;
1634 		vm_flags = vma->vm_flags;
1635 	} else if (vm_flags & VM_SHARED) {
1636 		error = shmem_zero_setup(vma);
1637 		if (error)
1638 			goto free_vma;
1639 	}
1640 
1641 	vma_link(mm, vma, prev, rb_link, rb_parent);
1642 	/* Once vma denies write, undo our temporary denial count */
1643 	if (file) {
1644 		if (vm_flags & VM_SHARED)
1645 			mapping_unmap_writable(file->f_mapping);
1646 		if (vm_flags & VM_DENYWRITE)
1647 			allow_write_access(file);
1648 	}
1649 	file = vma->vm_file;
1650 out:
1651 	perf_event_mmap(vma);
1652 
1653 	vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
1654 	if (vm_flags & VM_LOCKED) {
1655 		if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1656 					vma == get_gate_vma(current->mm)))
1657 			mm->locked_vm += (len >> PAGE_SHIFT);
1658 		else
1659 			vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
1660 	}
1661 
1662 	if (file)
1663 		uprobe_mmap(vma);
1664 
1665 	/*
1666 	 * New (or expanded) vma always get soft dirty status.
1667 	 * Otherwise user-space soft-dirty page tracker won't
1668 	 * be able to distinguish situation when vma area unmapped,
1669 	 * then new mapped in-place (which must be aimed as
1670 	 * a completely new data area).
1671 	 */
1672 	vma->vm_flags |= VM_SOFTDIRTY;
1673 
1674 	vma_set_page_prot(vma);
1675 
1676 	return addr;
1677 
1678 unmap_and_free_vma:
1679 	vma->vm_file = NULL;
1680 	fput(file);
1681 
1682 	/* Undo any partial mapping done by a device driver. */
1683 	unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1684 	charged = 0;
1685 	if (vm_flags & VM_SHARED)
1686 		mapping_unmap_writable(file->f_mapping);
1687 allow_write_and_free_vma:
1688 	if (vm_flags & VM_DENYWRITE)
1689 		allow_write_access(file);
1690 free_vma:
1691 	kmem_cache_free(vm_area_cachep, vma);
1692 unacct_error:
1693 	if (charged)
1694 		vm_unacct_memory(charged);
1695 	return error;
1696 }
1697 
1698 unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1699 {
1700 	/*
1701 	 * We implement the search by looking for an rbtree node that
1702 	 * immediately follows a suitable gap. That is,
1703 	 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1704 	 * - gap_end   = vma->vm_start        >= info->low_limit  + length;
1705 	 * - gap_end - gap_start >= length
1706 	 */
1707 
1708 	struct mm_struct *mm = current->mm;
1709 	struct vm_area_struct *vma;
1710 	unsigned long length, low_limit, high_limit, gap_start, gap_end;
1711 
1712 	/* Adjust search length to account for worst case alignment overhead */
1713 	length = info->length + info->align_mask;
1714 	if (length < info->length)
1715 		return -ENOMEM;
1716 
1717 	/* Adjust search limits by the desired length */
1718 	if (info->high_limit < length)
1719 		return -ENOMEM;
1720 	high_limit = info->high_limit - length;
1721 
1722 	if (info->low_limit > high_limit)
1723 		return -ENOMEM;
1724 	low_limit = info->low_limit + length;
1725 
1726 	/* Check if rbtree root looks promising */
1727 	if (RB_EMPTY_ROOT(&mm->mm_rb))
1728 		goto check_highest;
1729 	vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1730 	if (vma->rb_subtree_gap < length)
1731 		goto check_highest;
1732 
1733 	while (true) {
1734 		/* Visit left subtree if it looks promising */
1735 		gap_end = vma->vm_start;
1736 		if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1737 			struct vm_area_struct *left =
1738 				rb_entry(vma->vm_rb.rb_left,
1739 					 struct vm_area_struct, vm_rb);
1740 			if (left->rb_subtree_gap >= length) {
1741 				vma = left;
1742 				continue;
1743 			}
1744 		}
1745 
1746 		gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1747 check_current:
1748 		/* Check if current node has a suitable gap */
1749 		if (gap_start > high_limit)
1750 			return -ENOMEM;
1751 		if (gap_end >= low_limit && gap_end - gap_start >= length)
1752 			goto found;
1753 
1754 		/* Visit right subtree if it looks promising */
1755 		if (vma->vm_rb.rb_right) {
1756 			struct vm_area_struct *right =
1757 				rb_entry(vma->vm_rb.rb_right,
1758 					 struct vm_area_struct, vm_rb);
1759 			if (right->rb_subtree_gap >= length) {
1760 				vma = right;
1761 				continue;
1762 			}
1763 		}
1764 
1765 		/* Go back up the rbtree to find next candidate node */
1766 		while (true) {
1767 			struct rb_node *prev = &vma->vm_rb;
1768 			if (!rb_parent(prev))
1769 				goto check_highest;
1770 			vma = rb_entry(rb_parent(prev),
1771 				       struct vm_area_struct, vm_rb);
1772 			if (prev == vma->vm_rb.rb_left) {
1773 				gap_start = vma->vm_prev->vm_end;
1774 				gap_end = vma->vm_start;
1775 				goto check_current;
1776 			}
1777 		}
1778 	}
1779 
1780 check_highest:
1781 	/* Check highest gap, which does not precede any rbtree node */
1782 	gap_start = mm->highest_vm_end;
1783 	gap_end = ULONG_MAX;  /* Only for VM_BUG_ON below */
1784 	if (gap_start > high_limit)
1785 		return -ENOMEM;
1786 
1787 found:
1788 	/* We found a suitable gap. Clip it with the original low_limit. */
1789 	if (gap_start < info->low_limit)
1790 		gap_start = info->low_limit;
1791 
1792 	/* Adjust gap address to the desired alignment */
1793 	gap_start += (info->align_offset - gap_start) & info->align_mask;
1794 
1795 	VM_BUG_ON(gap_start + info->length > info->high_limit);
1796 	VM_BUG_ON(gap_start + info->length > gap_end);
1797 	return gap_start;
1798 }
1799 
1800 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1801 {
1802 	struct mm_struct *mm = current->mm;
1803 	struct vm_area_struct *vma;
1804 	unsigned long length, low_limit, high_limit, gap_start, gap_end;
1805 
1806 	/* Adjust search length to account for worst case alignment overhead */
1807 	length = info->length + info->align_mask;
1808 	if (length < info->length)
1809 		return -ENOMEM;
1810 
1811 	/*
1812 	 * Adjust search limits by the desired length.
1813 	 * See implementation comment at top of unmapped_area().
1814 	 */
1815 	gap_end = info->high_limit;
1816 	if (gap_end < length)
1817 		return -ENOMEM;
1818 	high_limit = gap_end - length;
1819 
1820 	if (info->low_limit > high_limit)
1821 		return -ENOMEM;
1822 	low_limit = info->low_limit + length;
1823 
1824 	/* Check highest gap, which does not precede any rbtree node */
1825 	gap_start = mm->highest_vm_end;
1826 	if (gap_start <= high_limit)
1827 		goto found_highest;
1828 
1829 	/* Check if rbtree root looks promising */
1830 	if (RB_EMPTY_ROOT(&mm->mm_rb))
1831 		return -ENOMEM;
1832 	vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1833 	if (vma->rb_subtree_gap < length)
1834 		return -ENOMEM;
1835 
1836 	while (true) {
1837 		/* Visit right subtree if it looks promising */
1838 		gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1839 		if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1840 			struct vm_area_struct *right =
1841 				rb_entry(vma->vm_rb.rb_right,
1842 					 struct vm_area_struct, vm_rb);
1843 			if (right->rb_subtree_gap >= length) {
1844 				vma = right;
1845 				continue;
1846 			}
1847 		}
1848 
1849 check_current:
1850 		/* Check if current node has a suitable gap */
1851 		gap_end = vma->vm_start;
1852 		if (gap_end < low_limit)
1853 			return -ENOMEM;
1854 		if (gap_start <= high_limit && gap_end - gap_start >= length)
1855 			goto found;
1856 
1857 		/* Visit left subtree if it looks promising */
1858 		if (vma->vm_rb.rb_left) {
1859 			struct vm_area_struct *left =
1860 				rb_entry(vma->vm_rb.rb_left,
1861 					 struct vm_area_struct, vm_rb);
1862 			if (left->rb_subtree_gap >= length) {
1863 				vma = left;
1864 				continue;
1865 			}
1866 		}
1867 
1868 		/* Go back up the rbtree to find next candidate node */
1869 		while (true) {
1870 			struct rb_node *prev = &vma->vm_rb;
1871 			if (!rb_parent(prev))
1872 				return -ENOMEM;
1873 			vma = rb_entry(rb_parent(prev),
1874 				       struct vm_area_struct, vm_rb);
1875 			if (prev == vma->vm_rb.rb_right) {
1876 				gap_start = vma->vm_prev ?
1877 					vma->vm_prev->vm_end : 0;
1878 				goto check_current;
1879 			}
1880 		}
1881 	}
1882 
1883 found:
1884 	/* We found a suitable gap. Clip it with the original high_limit. */
1885 	if (gap_end > info->high_limit)
1886 		gap_end = info->high_limit;
1887 
1888 found_highest:
1889 	/* Compute highest gap address at the desired alignment */
1890 	gap_end -= info->length;
1891 	gap_end -= (gap_end - info->align_offset) & info->align_mask;
1892 
1893 	VM_BUG_ON(gap_end < info->low_limit);
1894 	VM_BUG_ON(gap_end < gap_start);
1895 	return gap_end;
1896 }
1897 
1898 /* Get an address range which is currently unmapped.
1899  * For shmat() with addr=0.
1900  *
1901  * Ugly calling convention alert:
1902  * Return value with the low bits set means error value,
1903  * ie
1904  *	if (ret & ~PAGE_MASK)
1905  *		error = ret;
1906  *
1907  * This function "knows" that -ENOMEM has the bits set.
1908  */
1909 #ifndef HAVE_ARCH_UNMAPPED_AREA
1910 unsigned long
1911 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1912 		unsigned long len, unsigned long pgoff, unsigned long flags)
1913 {
1914 	struct mm_struct *mm = current->mm;
1915 	struct vm_area_struct *vma;
1916 	struct vm_unmapped_area_info info;
1917 
1918 	if (len > TASK_SIZE - mmap_min_addr)
1919 		return -ENOMEM;
1920 
1921 	if (flags & MAP_FIXED)
1922 		return addr;
1923 
1924 	if (addr) {
1925 		addr = PAGE_ALIGN(addr);
1926 		vma = find_vma(mm, addr);
1927 		if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1928 		    (!vma || addr + len <= vma->vm_start))
1929 			return addr;
1930 	}
1931 
1932 	info.flags = 0;
1933 	info.length = len;
1934 	info.low_limit = mm->mmap_base;
1935 	info.high_limit = TASK_SIZE;
1936 	info.align_mask = 0;
1937 	return vm_unmapped_area(&info);
1938 }
1939 #endif
1940 
1941 /*
1942  * This mmap-allocator allocates new areas top-down from below the
1943  * stack's low limit (the base):
1944  */
1945 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1946 unsigned long
1947 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1948 			  const unsigned long len, const unsigned long pgoff,
1949 			  const unsigned long flags)
1950 {
1951 	struct vm_area_struct *vma;
1952 	struct mm_struct *mm = current->mm;
1953 	unsigned long addr = addr0;
1954 	struct vm_unmapped_area_info info;
1955 
1956 	/* requested length too big for entire address space */
1957 	if (len > TASK_SIZE - mmap_min_addr)
1958 		return -ENOMEM;
1959 
1960 	if (flags & MAP_FIXED)
1961 		return addr;
1962 
1963 	/* requesting a specific address */
1964 	if (addr) {
1965 		addr = PAGE_ALIGN(addr);
1966 		vma = find_vma(mm, addr);
1967 		if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1968 				(!vma || addr + len <= vma->vm_start))
1969 			return addr;
1970 	}
1971 
1972 	info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1973 	info.length = len;
1974 	info.low_limit = max(PAGE_SIZE, mmap_min_addr);
1975 	info.high_limit = mm->mmap_base;
1976 	info.align_mask = 0;
1977 	addr = vm_unmapped_area(&info);
1978 
1979 	/*
1980 	 * A failed mmap() very likely causes application failure,
1981 	 * so fall back to the bottom-up function here. This scenario
1982 	 * can happen with large stack limits and large mmap()
1983 	 * allocations.
1984 	 */
1985 	if (offset_in_page(addr)) {
1986 		VM_BUG_ON(addr != -ENOMEM);
1987 		info.flags = 0;
1988 		info.low_limit = TASK_UNMAPPED_BASE;
1989 		info.high_limit = TASK_SIZE;
1990 		addr = vm_unmapped_area(&info);
1991 	}
1992 
1993 	return addr;
1994 }
1995 #endif
1996 
1997 unsigned long
1998 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1999 		unsigned long pgoff, unsigned long flags)
2000 {
2001 	unsigned long (*get_area)(struct file *, unsigned long,
2002 				  unsigned long, unsigned long, unsigned long);
2003 
2004 	unsigned long error = arch_mmap_check(addr, len, flags);
2005 	if (error)
2006 		return error;
2007 
2008 	/* Careful about overflows.. */
2009 	if (len > TASK_SIZE)
2010 		return -ENOMEM;
2011 
2012 	get_area = current->mm->get_unmapped_area;
2013 	if (file && file->f_op->get_unmapped_area)
2014 		get_area = file->f_op->get_unmapped_area;
2015 	addr = get_area(file, addr, len, pgoff, flags);
2016 	if (IS_ERR_VALUE(addr))
2017 		return addr;
2018 
2019 	if (addr > TASK_SIZE - len)
2020 		return -ENOMEM;
2021 	if (offset_in_page(addr))
2022 		return -EINVAL;
2023 
2024 	addr = arch_rebalance_pgtables(addr, len);
2025 	error = security_mmap_addr(addr);
2026 	return error ? error : addr;
2027 }
2028 
2029 EXPORT_SYMBOL(get_unmapped_area);
2030 
2031 /* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
2032 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2033 {
2034 	struct rb_node *rb_node;
2035 	struct vm_area_struct *vma;
2036 
2037 	/* Check the cache first. */
2038 	vma = vmacache_find(mm, addr);
2039 	if (likely(vma))
2040 		return vma;
2041 
2042 	rb_node = mm->mm_rb.rb_node;
2043 
2044 	while (rb_node) {
2045 		struct vm_area_struct *tmp;
2046 
2047 		tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2048 
2049 		if (tmp->vm_end > addr) {
2050 			vma = tmp;
2051 			if (tmp->vm_start <= addr)
2052 				break;
2053 			rb_node = rb_node->rb_left;
2054 		} else
2055 			rb_node = rb_node->rb_right;
2056 	}
2057 
2058 	if (vma)
2059 		vmacache_update(addr, vma);
2060 	return vma;
2061 }
2062 
2063 EXPORT_SYMBOL(find_vma);
2064 
2065 /*
2066  * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2067  */
2068 struct vm_area_struct *
2069 find_vma_prev(struct mm_struct *mm, unsigned long addr,
2070 			struct vm_area_struct **pprev)
2071 {
2072 	struct vm_area_struct *vma;
2073 
2074 	vma = find_vma(mm, addr);
2075 	if (vma) {
2076 		*pprev = vma->vm_prev;
2077 	} else {
2078 		struct rb_node *rb_node = mm->mm_rb.rb_node;
2079 		*pprev = NULL;
2080 		while (rb_node) {
2081 			*pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2082 			rb_node = rb_node->rb_right;
2083 		}
2084 	}
2085 	return vma;
2086 }
2087 
2088 /*
2089  * Verify that the stack growth is acceptable and
2090  * update accounting. This is shared with both the
2091  * grow-up and grow-down cases.
2092  */
2093 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
2094 {
2095 	struct mm_struct *mm = vma->vm_mm;
2096 	struct rlimit *rlim = current->signal->rlim;
2097 	unsigned long new_start, actual_size;
2098 
2099 	/* address space limit tests */
2100 	if (!may_expand_vm(mm, vma->vm_flags, grow))
2101 		return -ENOMEM;
2102 
2103 	/* Stack limit test */
2104 	actual_size = size;
2105 	if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
2106 		actual_size -= PAGE_SIZE;
2107 	if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
2108 		return -ENOMEM;
2109 
2110 	/* mlock limit tests */
2111 	if (vma->vm_flags & VM_LOCKED) {
2112 		unsigned long locked;
2113 		unsigned long limit;
2114 		locked = mm->locked_vm + grow;
2115 		limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2116 		limit >>= PAGE_SHIFT;
2117 		if (locked > limit && !capable(CAP_IPC_LOCK))
2118 			return -ENOMEM;
2119 	}
2120 
2121 	/* Check to ensure the stack will not grow into a hugetlb-only region */
2122 	new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2123 			vma->vm_end - size;
2124 	if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2125 		return -EFAULT;
2126 
2127 	/*
2128 	 * Overcommit..  This must be the final test, as it will
2129 	 * update security statistics.
2130 	 */
2131 	if (security_vm_enough_memory_mm(mm, grow))
2132 		return -ENOMEM;
2133 
2134 	return 0;
2135 }
2136 
2137 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2138 /*
2139  * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2140  * vma is the last one with address > vma->vm_end.  Have to extend vma.
2141  */
2142 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2143 {
2144 	struct mm_struct *mm = vma->vm_mm;
2145 	int error;
2146 
2147 	if (!(vma->vm_flags & VM_GROWSUP))
2148 		return -EFAULT;
2149 
2150 	/*
2151 	 * We must make sure the anon_vma is allocated
2152 	 * so that the anon_vma locking is not a noop.
2153 	 */
2154 	if (unlikely(anon_vma_prepare(vma)))
2155 		return -ENOMEM;
2156 	vma_lock_anon_vma(vma);
2157 
2158 	/*
2159 	 * vma->vm_start/vm_end cannot change under us because the caller
2160 	 * is required to hold the mmap_sem in read mode.  We need the
2161 	 * anon_vma lock to serialize against concurrent expand_stacks.
2162 	 * Also guard against wrapping around to address 0.
2163 	 */
2164 	if (address < PAGE_ALIGN(address+4))
2165 		address = PAGE_ALIGN(address+4);
2166 	else {
2167 		vma_unlock_anon_vma(vma);
2168 		return -ENOMEM;
2169 	}
2170 	error = 0;
2171 
2172 	/* Somebody else might have raced and expanded it already */
2173 	if (address > vma->vm_end) {
2174 		unsigned long size, grow;
2175 
2176 		size = address - vma->vm_start;
2177 		grow = (address - vma->vm_end) >> PAGE_SHIFT;
2178 
2179 		error = -ENOMEM;
2180 		if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2181 			error = acct_stack_growth(vma, size, grow);
2182 			if (!error) {
2183 				/*
2184 				 * vma_gap_update() doesn't support concurrent
2185 				 * updates, but we only hold a shared mmap_sem
2186 				 * lock here, so we need to protect against
2187 				 * concurrent vma expansions.
2188 				 * vma_lock_anon_vma() doesn't help here, as
2189 				 * we don't guarantee that all growable vmas
2190 				 * in a mm share the same root anon vma.
2191 				 * So, we reuse mm->page_table_lock to guard
2192 				 * against concurrent vma expansions.
2193 				 */
2194 				spin_lock(&mm->page_table_lock);
2195 				if (vma->vm_flags & VM_LOCKED)
2196 					mm->locked_vm += grow;
2197 				vm_stat_account(mm, vma->vm_flags, grow);
2198 				anon_vma_interval_tree_pre_update_vma(vma);
2199 				vma->vm_end = address;
2200 				anon_vma_interval_tree_post_update_vma(vma);
2201 				if (vma->vm_next)
2202 					vma_gap_update(vma->vm_next);
2203 				else
2204 					mm->highest_vm_end = address;
2205 				spin_unlock(&mm->page_table_lock);
2206 
2207 				perf_event_mmap(vma);
2208 			}
2209 		}
2210 	}
2211 	vma_unlock_anon_vma(vma);
2212 	khugepaged_enter_vma_merge(vma, vma->vm_flags);
2213 	validate_mm(mm);
2214 	return error;
2215 }
2216 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2217 
2218 /*
2219  * vma is the first one with address < vma->vm_start.  Have to extend vma.
2220  */
2221 int expand_downwards(struct vm_area_struct *vma,
2222 				   unsigned long address)
2223 {
2224 	struct mm_struct *mm = vma->vm_mm;
2225 	int error;
2226 
2227 	/*
2228 	 * We must make sure the anon_vma is allocated
2229 	 * so that the anon_vma locking is not a noop.
2230 	 */
2231 	if (unlikely(anon_vma_prepare(vma)))
2232 		return -ENOMEM;
2233 
2234 	address &= PAGE_MASK;
2235 	error = security_mmap_addr(address);
2236 	if (error)
2237 		return error;
2238 
2239 	vma_lock_anon_vma(vma);
2240 
2241 	/*
2242 	 * vma->vm_start/vm_end cannot change under us because the caller
2243 	 * is required to hold the mmap_sem in read mode.  We need the
2244 	 * anon_vma lock to serialize against concurrent expand_stacks.
2245 	 */
2246 
2247 	/* Somebody else might have raced and expanded it already */
2248 	if (address < vma->vm_start) {
2249 		unsigned long size, grow;
2250 
2251 		size = vma->vm_end - address;
2252 		grow = (vma->vm_start - address) >> PAGE_SHIFT;
2253 
2254 		error = -ENOMEM;
2255 		if (grow <= vma->vm_pgoff) {
2256 			error = acct_stack_growth(vma, size, grow);
2257 			if (!error) {
2258 				/*
2259 				 * vma_gap_update() doesn't support concurrent
2260 				 * updates, but we only hold a shared mmap_sem
2261 				 * lock here, so we need to protect against
2262 				 * concurrent vma expansions.
2263 				 * vma_lock_anon_vma() doesn't help here, as
2264 				 * we don't guarantee that all growable vmas
2265 				 * in a mm share the same root anon vma.
2266 				 * So, we reuse mm->page_table_lock to guard
2267 				 * against concurrent vma expansions.
2268 				 */
2269 				spin_lock(&mm->page_table_lock);
2270 				if (vma->vm_flags & VM_LOCKED)
2271 					mm->locked_vm += grow;
2272 				vm_stat_account(mm, vma->vm_flags, grow);
2273 				anon_vma_interval_tree_pre_update_vma(vma);
2274 				vma->vm_start = address;
2275 				vma->vm_pgoff -= grow;
2276 				anon_vma_interval_tree_post_update_vma(vma);
2277 				vma_gap_update(vma);
2278 				spin_unlock(&mm->page_table_lock);
2279 
2280 				perf_event_mmap(vma);
2281 			}
2282 		}
2283 	}
2284 	vma_unlock_anon_vma(vma);
2285 	khugepaged_enter_vma_merge(vma, vma->vm_flags);
2286 	validate_mm(mm);
2287 	return error;
2288 }
2289 
2290 /*
2291  * Note how expand_stack() refuses to expand the stack all the way to
2292  * abut the next virtual mapping, *unless* that mapping itself is also
2293  * a stack mapping. We want to leave room for a guard page, after all
2294  * (the guard page itself is not added here, that is done by the
2295  * actual page faulting logic)
2296  *
2297  * This matches the behavior of the guard page logic (see mm/memory.c:
2298  * check_stack_guard_page()), which only allows the guard page to be
2299  * removed under these circumstances.
2300  */
2301 #ifdef CONFIG_STACK_GROWSUP
2302 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2303 {
2304 	struct vm_area_struct *next;
2305 
2306 	address &= PAGE_MASK;
2307 	next = vma->vm_next;
2308 	if (next && next->vm_start == address + PAGE_SIZE) {
2309 		if (!(next->vm_flags & VM_GROWSUP))
2310 			return -ENOMEM;
2311 	}
2312 	return expand_upwards(vma, address);
2313 }
2314 
2315 struct vm_area_struct *
2316 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2317 {
2318 	struct vm_area_struct *vma, *prev;
2319 
2320 	addr &= PAGE_MASK;
2321 	vma = find_vma_prev(mm, addr, &prev);
2322 	if (vma && (vma->vm_start <= addr))
2323 		return vma;
2324 	if (!prev || expand_stack(prev, addr))
2325 		return NULL;
2326 	if (prev->vm_flags & VM_LOCKED)
2327 		populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2328 	return prev;
2329 }
2330 #else
2331 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2332 {
2333 	struct vm_area_struct *prev;
2334 
2335 	address &= PAGE_MASK;
2336 	prev = vma->vm_prev;
2337 	if (prev && prev->vm_end == address) {
2338 		if (!(prev->vm_flags & VM_GROWSDOWN))
2339 			return -ENOMEM;
2340 	}
2341 	return expand_downwards(vma, address);
2342 }
2343 
2344 struct vm_area_struct *
2345 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2346 {
2347 	struct vm_area_struct *vma;
2348 	unsigned long start;
2349 
2350 	addr &= PAGE_MASK;
2351 	vma = find_vma(mm, addr);
2352 	if (!vma)
2353 		return NULL;
2354 	if (vma->vm_start <= addr)
2355 		return vma;
2356 	if (!(vma->vm_flags & VM_GROWSDOWN))
2357 		return NULL;
2358 	start = vma->vm_start;
2359 	if (expand_stack(vma, addr))
2360 		return NULL;
2361 	if (vma->vm_flags & VM_LOCKED)
2362 		populate_vma_page_range(vma, addr, start, NULL);
2363 	return vma;
2364 }
2365 #endif
2366 
2367 EXPORT_SYMBOL_GPL(find_extend_vma);
2368 
2369 /*
2370  * Ok - we have the memory areas we should free on the vma list,
2371  * so release them, and do the vma updates.
2372  *
2373  * Called with the mm semaphore held.
2374  */
2375 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2376 {
2377 	unsigned long nr_accounted = 0;
2378 
2379 	/* Update high watermark before we lower total_vm */
2380 	update_hiwater_vm(mm);
2381 	do {
2382 		long nrpages = vma_pages(vma);
2383 
2384 		if (vma->vm_flags & VM_ACCOUNT)
2385 			nr_accounted += nrpages;
2386 		vm_stat_account(mm, vma->vm_flags, -nrpages);
2387 		vma = remove_vma(vma);
2388 	} while (vma);
2389 	vm_unacct_memory(nr_accounted);
2390 	validate_mm(mm);
2391 }
2392 
2393 /*
2394  * Get rid of page table information in the indicated region.
2395  *
2396  * Called with the mm semaphore held.
2397  */
2398 static void unmap_region(struct mm_struct *mm,
2399 		struct vm_area_struct *vma, struct vm_area_struct *prev,
2400 		unsigned long start, unsigned long end)
2401 {
2402 	struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
2403 	struct mmu_gather tlb;
2404 
2405 	lru_add_drain();
2406 	tlb_gather_mmu(&tlb, mm, start, end);
2407 	update_hiwater_rss(mm);
2408 	unmap_vmas(&tlb, vma, start, end);
2409 	free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2410 				 next ? next->vm_start : USER_PGTABLES_CEILING);
2411 	tlb_finish_mmu(&tlb, start, end);
2412 }
2413 
2414 /*
2415  * Create a list of vma's touched by the unmap, removing them from the mm's
2416  * vma list as we go..
2417  */
2418 static void
2419 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2420 	struct vm_area_struct *prev, unsigned long end)
2421 {
2422 	struct vm_area_struct **insertion_point;
2423 	struct vm_area_struct *tail_vma = NULL;
2424 
2425 	insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2426 	vma->vm_prev = NULL;
2427 	do {
2428 		vma_rb_erase(vma, &mm->mm_rb);
2429 		mm->map_count--;
2430 		tail_vma = vma;
2431 		vma = vma->vm_next;
2432 	} while (vma && vma->vm_start < end);
2433 	*insertion_point = vma;
2434 	if (vma) {
2435 		vma->vm_prev = prev;
2436 		vma_gap_update(vma);
2437 	} else
2438 		mm->highest_vm_end = prev ? prev->vm_end : 0;
2439 	tail_vma->vm_next = NULL;
2440 
2441 	/* Kill the cache */
2442 	vmacache_invalidate(mm);
2443 }
2444 
2445 /*
2446  * __split_vma() bypasses sysctl_max_map_count checking.  We use this on the
2447  * munmap path where it doesn't make sense to fail.
2448  */
2449 static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2450 	      unsigned long addr, int new_below)
2451 {
2452 	struct vm_area_struct *new;
2453 	int err;
2454 
2455 	if (is_vm_hugetlb_page(vma) && (addr &
2456 					~(huge_page_mask(hstate_vma(vma)))))
2457 		return -EINVAL;
2458 
2459 	new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2460 	if (!new)
2461 		return -ENOMEM;
2462 
2463 	/* most fields are the same, copy all, and then fixup */
2464 	*new = *vma;
2465 
2466 	INIT_LIST_HEAD(&new->anon_vma_chain);
2467 
2468 	if (new_below)
2469 		new->vm_end = addr;
2470 	else {
2471 		new->vm_start = addr;
2472 		new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2473 	}
2474 
2475 	err = vma_dup_policy(vma, new);
2476 	if (err)
2477 		goto out_free_vma;
2478 
2479 	err = anon_vma_clone(new, vma);
2480 	if (err)
2481 		goto out_free_mpol;
2482 
2483 	if (new->vm_file)
2484 		get_file(new->vm_file);
2485 
2486 	if (new->vm_ops && new->vm_ops->open)
2487 		new->vm_ops->open(new);
2488 
2489 	if (new_below)
2490 		err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2491 			((addr - new->vm_start) >> PAGE_SHIFT), new);
2492 	else
2493 		err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2494 
2495 	/* Success. */
2496 	if (!err)
2497 		return 0;
2498 
2499 	/* Clean everything up if vma_adjust failed. */
2500 	if (new->vm_ops && new->vm_ops->close)
2501 		new->vm_ops->close(new);
2502 	if (new->vm_file)
2503 		fput(new->vm_file);
2504 	unlink_anon_vmas(new);
2505  out_free_mpol:
2506 	mpol_put(vma_policy(new));
2507  out_free_vma:
2508 	kmem_cache_free(vm_area_cachep, new);
2509 	return err;
2510 }
2511 
2512 /*
2513  * Split a vma into two pieces at address 'addr', a new vma is allocated
2514  * either for the first part or the tail.
2515  */
2516 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2517 	      unsigned long addr, int new_below)
2518 {
2519 	if (mm->map_count >= sysctl_max_map_count)
2520 		return -ENOMEM;
2521 
2522 	return __split_vma(mm, vma, addr, new_below);
2523 }
2524 
2525 /* Munmap is split into 2 main parts -- this part which finds
2526  * what needs doing, and the areas themselves, which do the
2527  * work.  This now handles partial unmappings.
2528  * Jeremy Fitzhardinge <jeremy@goop.org>
2529  */
2530 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2531 {
2532 	unsigned long end;
2533 	struct vm_area_struct *vma, *prev, *last;
2534 
2535 	if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2536 		return -EINVAL;
2537 
2538 	len = PAGE_ALIGN(len);
2539 	if (len == 0)
2540 		return -EINVAL;
2541 
2542 	/* Find the first overlapping VMA */
2543 	vma = find_vma(mm, start);
2544 	if (!vma)
2545 		return 0;
2546 	prev = vma->vm_prev;
2547 	/* we have  start < vma->vm_end  */
2548 
2549 	/* if it doesn't overlap, we have nothing.. */
2550 	end = start + len;
2551 	if (vma->vm_start >= end)
2552 		return 0;
2553 
2554 	/*
2555 	 * If we need to split any vma, do it now to save pain later.
2556 	 *
2557 	 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2558 	 * unmapped vm_area_struct will remain in use: so lower split_vma
2559 	 * places tmp vma above, and higher split_vma places tmp vma below.
2560 	 */
2561 	if (start > vma->vm_start) {
2562 		int error;
2563 
2564 		/*
2565 		 * Make sure that map_count on return from munmap() will
2566 		 * not exceed its limit; but let map_count go just above
2567 		 * its limit temporarily, to help free resources as expected.
2568 		 */
2569 		if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2570 			return -ENOMEM;
2571 
2572 		error = __split_vma(mm, vma, start, 0);
2573 		if (error)
2574 			return error;
2575 		prev = vma;
2576 	}
2577 
2578 	/* Does it split the last one? */
2579 	last = find_vma(mm, end);
2580 	if (last && end > last->vm_start) {
2581 		int error = __split_vma(mm, last, end, 1);
2582 		if (error)
2583 			return error;
2584 	}
2585 	vma = prev ? prev->vm_next : mm->mmap;
2586 
2587 	/*
2588 	 * unlock any mlock()ed ranges before detaching vmas
2589 	 */
2590 	if (mm->locked_vm) {
2591 		struct vm_area_struct *tmp = vma;
2592 		while (tmp && tmp->vm_start < end) {
2593 			if (tmp->vm_flags & VM_LOCKED) {
2594 				mm->locked_vm -= vma_pages(tmp);
2595 				munlock_vma_pages_all(tmp);
2596 			}
2597 			tmp = tmp->vm_next;
2598 		}
2599 	}
2600 
2601 	/*
2602 	 * Remove the vma's, and unmap the actual pages
2603 	 */
2604 	detach_vmas_to_be_unmapped(mm, vma, prev, end);
2605 	unmap_region(mm, vma, prev, start, end);
2606 
2607 	arch_unmap(mm, vma, start, end);
2608 
2609 	/* Fix up all other VM information */
2610 	remove_vma_list(mm, vma);
2611 
2612 	return 0;
2613 }
2614 
2615 int vm_munmap(unsigned long start, size_t len)
2616 {
2617 	int ret;
2618 	struct mm_struct *mm = current->mm;
2619 
2620 	down_write(&mm->mmap_sem);
2621 	ret = do_munmap(mm, start, len);
2622 	up_write(&mm->mmap_sem);
2623 	return ret;
2624 }
2625 EXPORT_SYMBOL(vm_munmap);
2626 
2627 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2628 {
2629 	profile_munmap(addr);
2630 	return vm_munmap(addr, len);
2631 }
2632 
2633 
2634 /*
2635  * Emulation of deprecated remap_file_pages() syscall.
2636  */
2637 SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2638 		unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2639 {
2640 
2641 	struct mm_struct *mm = current->mm;
2642 	struct vm_area_struct *vma;
2643 	unsigned long populate = 0;
2644 	unsigned long ret = -EINVAL;
2645 	struct file *file;
2646 
2647 	pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. "
2648 			"See Documentation/vm/remap_file_pages.txt.\n",
2649 			current->comm, current->pid);
2650 
2651 	if (prot)
2652 		return ret;
2653 	start = start & PAGE_MASK;
2654 	size = size & PAGE_MASK;
2655 
2656 	if (start + size <= start)
2657 		return ret;
2658 
2659 	/* Does pgoff wrap? */
2660 	if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2661 		return ret;
2662 
2663 	down_write(&mm->mmap_sem);
2664 	vma = find_vma(mm, start);
2665 
2666 	if (!vma || !(vma->vm_flags & VM_SHARED))
2667 		goto out;
2668 
2669 	if (start < vma->vm_start || start + size > vma->vm_end)
2670 		goto out;
2671 
2672 	if (pgoff == linear_page_index(vma, start)) {
2673 		ret = 0;
2674 		goto out;
2675 	}
2676 
2677 	prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2678 	prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2679 	prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2680 
2681 	flags &= MAP_NONBLOCK;
2682 	flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2683 	if (vma->vm_flags & VM_LOCKED) {
2684 		flags |= MAP_LOCKED;
2685 		/* drop PG_Mlocked flag for over-mapped range */
2686 		munlock_vma_pages_range(vma, start, start + size);
2687 	}
2688 
2689 	file = get_file(vma->vm_file);
2690 	ret = do_mmap_pgoff(vma->vm_file, start, size,
2691 			prot, flags, pgoff, &populate);
2692 	fput(file);
2693 out:
2694 	up_write(&mm->mmap_sem);
2695 	if (populate)
2696 		mm_populate(ret, populate);
2697 	if (!IS_ERR_VALUE(ret))
2698 		ret = 0;
2699 	return ret;
2700 }
2701 
2702 static inline void verify_mm_writelocked(struct mm_struct *mm)
2703 {
2704 #ifdef CONFIG_DEBUG_VM
2705 	if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2706 		WARN_ON(1);
2707 		up_read(&mm->mmap_sem);
2708 	}
2709 #endif
2710 }
2711 
2712 /*
2713  *  this is really a simplified "do_mmap".  it only handles
2714  *  anonymous maps.  eventually we may be able to do some
2715  *  brk-specific accounting here.
2716  */
2717 static unsigned long do_brk(unsigned long addr, unsigned long len)
2718 {
2719 	struct mm_struct *mm = current->mm;
2720 	struct vm_area_struct *vma, *prev;
2721 	unsigned long flags;
2722 	struct rb_node **rb_link, *rb_parent;
2723 	pgoff_t pgoff = addr >> PAGE_SHIFT;
2724 	int error;
2725 
2726 	len = PAGE_ALIGN(len);
2727 	if (!len)
2728 		return addr;
2729 
2730 	flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2731 
2732 	error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2733 	if (offset_in_page(error))
2734 		return error;
2735 
2736 	error = mlock_future_check(mm, mm->def_flags, len);
2737 	if (error)
2738 		return error;
2739 
2740 	/*
2741 	 * mm->mmap_sem is required to protect against another thread
2742 	 * changing the mappings in case we sleep.
2743 	 */
2744 	verify_mm_writelocked(mm);
2745 
2746 	/*
2747 	 * Clear old maps.  this also does some error checking for us
2748 	 */
2749 	while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2750 			      &rb_parent)) {
2751 		if (do_munmap(mm, addr, len))
2752 			return -ENOMEM;
2753 	}
2754 
2755 	/* Check against address space limits *after* clearing old maps... */
2756 	if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
2757 		return -ENOMEM;
2758 
2759 	if (mm->map_count > sysctl_max_map_count)
2760 		return -ENOMEM;
2761 
2762 	if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2763 		return -ENOMEM;
2764 
2765 	/* Can we just expand an old private anonymous mapping? */
2766 	vma = vma_merge(mm, prev, addr, addr + len, flags,
2767 			NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
2768 	if (vma)
2769 		goto out;
2770 
2771 	/*
2772 	 * create a vma struct for an anonymous mapping
2773 	 */
2774 	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2775 	if (!vma) {
2776 		vm_unacct_memory(len >> PAGE_SHIFT);
2777 		return -ENOMEM;
2778 	}
2779 
2780 	INIT_LIST_HEAD(&vma->anon_vma_chain);
2781 	vma->vm_mm = mm;
2782 	vma->vm_start = addr;
2783 	vma->vm_end = addr + len;
2784 	vma->vm_pgoff = pgoff;
2785 	vma->vm_flags = flags;
2786 	vma->vm_page_prot = vm_get_page_prot(flags);
2787 	vma_link(mm, vma, prev, rb_link, rb_parent);
2788 out:
2789 	perf_event_mmap(vma);
2790 	mm->total_vm += len >> PAGE_SHIFT;
2791 	mm->data_vm += len >> PAGE_SHIFT;
2792 	if (flags & VM_LOCKED)
2793 		mm->locked_vm += (len >> PAGE_SHIFT);
2794 	vma->vm_flags |= VM_SOFTDIRTY;
2795 	return addr;
2796 }
2797 
2798 unsigned long vm_brk(unsigned long addr, unsigned long len)
2799 {
2800 	struct mm_struct *mm = current->mm;
2801 	unsigned long ret;
2802 	bool populate;
2803 
2804 	down_write(&mm->mmap_sem);
2805 	ret = do_brk(addr, len);
2806 	populate = ((mm->def_flags & VM_LOCKED) != 0);
2807 	up_write(&mm->mmap_sem);
2808 	if (populate)
2809 		mm_populate(addr, len);
2810 	return ret;
2811 }
2812 EXPORT_SYMBOL(vm_brk);
2813 
2814 /* Release all mmaps. */
2815 void exit_mmap(struct mm_struct *mm)
2816 {
2817 	struct mmu_gather tlb;
2818 	struct vm_area_struct *vma;
2819 	unsigned long nr_accounted = 0;
2820 
2821 	/* mm's last user has gone, and its about to be pulled down */
2822 	mmu_notifier_release(mm);
2823 
2824 	if (mm->locked_vm) {
2825 		vma = mm->mmap;
2826 		while (vma) {
2827 			if (vma->vm_flags & VM_LOCKED)
2828 				munlock_vma_pages_all(vma);
2829 			vma = vma->vm_next;
2830 		}
2831 	}
2832 
2833 	arch_exit_mmap(mm);
2834 
2835 	vma = mm->mmap;
2836 	if (!vma)	/* Can happen if dup_mmap() received an OOM */
2837 		return;
2838 
2839 	lru_add_drain();
2840 	flush_cache_mm(mm);
2841 	tlb_gather_mmu(&tlb, mm, 0, -1);
2842 	/* update_hiwater_rss(mm) here? but nobody should be looking */
2843 	/* Use -1 here to ensure all VMAs in the mm are unmapped */
2844 	unmap_vmas(&tlb, vma, 0, -1);
2845 
2846 	free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
2847 	tlb_finish_mmu(&tlb, 0, -1);
2848 
2849 	/*
2850 	 * Walk the list again, actually closing and freeing it,
2851 	 * with preemption enabled, without holding any MM locks.
2852 	 */
2853 	while (vma) {
2854 		if (vma->vm_flags & VM_ACCOUNT)
2855 			nr_accounted += vma_pages(vma);
2856 		vma = remove_vma(vma);
2857 	}
2858 	vm_unacct_memory(nr_accounted);
2859 }
2860 
2861 /* Insert vm structure into process list sorted by address
2862  * and into the inode's i_mmap tree.  If vm_file is non-NULL
2863  * then i_mmap_rwsem is taken here.
2864  */
2865 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2866 {
2867 	struct vm_area_struct *prev;
2868 	struct rb_node **rb_link, *rb_parent;
2869 
2870 	if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2871 			   &prev, &rb_link, &rb_parent))
2872 		return -ENOMEM;
2873 	if ((vma->vm_flags & VM_ACCOUNT) &&
2874 	     security_vm_enough_memory_mm(mm, vma_pages(vma)))
2875 		return -ENOMEM;
2876 
2877 	/*
2878 	 * The vm_pgoff of a purely anonymous vma should be irrelevant
2879 	 * until its first write fault, when page's anon_vma and index
2880 	 * are set.  But now set the vm_pgoff it will almost certainly
2881 	 * end up with (unless mremap moves it elsewhere before that
2882 	 * first wfault), so /proc/pid/maps tells a consistent story.
2883 	 *
2884 	 * By setting it to reflect the virtual start address of the
2885 	 * vma, merges and splits can happen in a seamless way, just
2886 	 * using the existing file pgoff checks and manipulations.
2887 	 * Similarly in do_mmap_pgoff and in do_brk.
2888 	 */
2889 	if (vma_is_anonymous(vma)) {
2890 		BUG_ON(vma->anon_vma);
2891 		vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2892 	}
2893 
2894 	vma_link(mm, vma, prev, rb_link, rb_parent);
2895 	return 0;
2896 }
2897 
2898 /*
2899  * Copy the vma structure to a new location in the same mm,
2900  * prior to moving page table entries, to effect an mremap move.
2901  */
2902 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2903 	unsigned long addr, unsigned long len, pgoff_t pgoff,
2904 	bool *need_rmap_locks)
2905 {
2906 	struct vm_area_struct *vma = *vmap;
2907 	unsigned long vma_start = vma->vm_start;
2908 	struct mm_struct *mm = vma->vm_mm;
2909 	struct vm_area_struct *new_vma, *prev;
2910 	struct rb_node **rb_link, *rb_parent;
2911 	bool faulted_in_anon_vma = true;
2912 
2913 	/*
2914 	 * If anonymous vma has not yet been faulted, update new pgoff
2915 	 * to match new location, to increase its chance of merging.
2916 	 */
2917 	if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
2918 		pgoff = addr >> PAGE_SHIFT;
2919 		faulted_in_anon_vma = false;
2920 	}
2921 
2922 	if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2923 		return NULL;	/* should never get here */
2924 	new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2925 			    vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2926 			    vma->vm_userfaultfd_ctx);
2927 	if (new_vma) {
2928 		/*
2929 		 * Source vma may have been merged into new_vma
2930 		 */
2931 		if (unlikely(vma_start >= new_vma->vm_start &&
2932 			     vma_start < new_vma->vm_end)) {
2933 			/*
2934 			 * The only way we can get a vma_merge with
2935 			 * self during an mremap is if the vma hasn't
2936 			 * been faulted in yet and we were allowed to
2937 			 * reset the dst vma->vm_pgoff to the
2938 			 * destination address of the mremap to allow
2939 			 * the merge to happen. mremap must change the
2940 			 * vm_pgoff linearity between src and dst vmas
2941 			 * (in turn preventing a vma_merge) to be
2942 			 * safe. It is only safe to keep the vm_pgoff
2943 			 * linear if there are no pages mapped yet.
2944 			 */
2945 			VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
2946 			*vmap = vma = new_vma;
2947 		}
2948 		*need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2949 	} else {
2950 		new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2951 		if (!new_vma)
2952 			goto out;
2953 		*new_vma = *vma;
2954 		new_vma->vm_start = addr;
2955 		new_vma->vm_end = addr + len;
2956 		new_vma->vm_pgoff = pgoff;
2957 		if (vma_dup_policy(vma, new_vma))
2958 			goto out_free_vma;
2959 		INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2960 		if (anon_vma_clone(new_vma, vma))
2961 			goto out_free_mempol;
2962 		if (new_vma->vm_file)
2963 			get_file(new_vma->vm_file);
2964 		if (new_vma->vm_ops && new_vma->vm_ops->open)
2965 			new_vma->vm_ops->open(new_vma);
2966 		vma_link(mm, new_vma, prev, rb_link, rb_parent);
2967 		*need_rmap_locks = false;
2968 	}
2969 	return new_vma;
2970 
2971 out_free_mempol:
2972 	mpol_put(vma_policy(new_vma));
2973 out_free_vma:
2974 	kmem_cache_free(vm_area_cachep, new_vma);
2975 out:
2976 	return NULL;
2977 }
2978 
2979 /*
2980  * Return true if the calling process may expand its vm space by the passed
2981  * number of pages
2982  */
2983 bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
2984 {
2985 	if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
2986 		return false;
2987 
2988 	if (is_data_mapping(flags) &&
2989 	    mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
2990 		if (ignore_rlimit_data)
2991 			pr_warn_once("%s (%d): VmData %lu exceed data ulimit "
2992 				     "%lu. Will be forbidden soon.\n",
2993 				     current->comm, current->pid,
2994 				     (mm->data_vm + npages) << PAGE_SHIFT,
2995 				     rlimit(RLIMIT_DATA));
2996 		else
2997 			return false;
2998 	}
2999 
3000 	return true;
3001 }
3002 
3003 void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3004 {
3005 	mm->total_vm += npages;
3006 
3007 	if (is_exec_mapping(flags))
3008 		mm->exec_vm += npages;
3009 	else if (is_stack_mapping(flags))
3010 		mm->stack_vm += npages;
3011 	else if (is_data_mapping(flags))
3012 		mm->data_vm += npages;
3013 }
3014 
3015 static int special_mapping_fault(struct vm_area_struct *vma,
3016 				 struct vm_fault *vmf);
3017 
3018 /*
3019  * Having a close hook prevents vma merging regardless of flags.
3020  */
3021 static void special_mapping_close(struct vm_area_struct *vma)
3022 {
3023 }
3024 
3025 static const char *special_mapping_name(struct vm_area_struct *vma)
3026 {
3027 	return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3028 }
3029 
3030 static const struct vm_operations_struct special_mapping_vmops = {
3031 	.close = special_mapping_close,
3032 	.fault = special_mapping_fault,
3033 	.name = special_mapping_name,
3034 };
3035 
3036 static const struct vm_operations_struct legacy_special_mapping_vmops = {
3037 	.close = special_mapping_close,
3038 	.fault = special_mapping_fault,
3039 };
3040 
3041 static int special_mapping_fault(struct vm_area_struct *vma,
3042 				struct vm_fault *vmf)
3043 {
3044 	pgoff_t pgoff;
3045 	struct page **pages;
3046 
3047 	if (vma->vm_ops == &legacy_special_mapping_vmops)
3048 		pages = vma->vm_private_data;
3049 	else
3050 		pages = ((struct vm_special_mapping *)vma->vm_private_data)->
3051 			pages;
3052 
3053 	for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3054 		pgoff--;
3055 
3056 	if (*pages) {
3057 		struct page *page = *pages;
3058 		get_page(page);
3059 		vmf->page = page;
3060 		return 0;
3061 	}
3062 
3063 	return VM_FAULT_SIGBUS;
3064 }
3065 
3066 static struct vm_area_struct *__install_special_mapping(
3067 	struct mm_struct *mm,
3068 	unsigned long addr, unsigned long len,
3069 	unsigned long vm_flags, void *priv,
3070 	const struct vm_operations_struct *ops)
3071 {
3072 	int ret;
3073 	struct vm_area_struct *vma;
3074 
3075 	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
3076 	if (unlikely(vma == NULL))
3077 		return ERR_PTR(-ENOMEM);
3078 
3079 	INIT_LIST_HEAD(&vma->anon_vma_chain);
3080 	vma->vm_mm = mm;
3081 	vma->vm_start = addr;
3082 	vma->vm_end = addr + len;
3083 
3084 	vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
3085 	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3086 
3087 	vma->vm_ops = ops;
3088 	vma->vm_private_data = priv;
3089 
3090 	ret = insert_vm_struct(mm, vma);
3091 	if (ret)
3092 		goto out;
3093 
3094 	vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3095 
3096 	perf_event_mmap(vma);
3097 
3098 	return vma;
3099 
3100 out:
3101 	kmem_cache_free(vm_area_cachep, vma);
3102 	return ERR_PTR(ret);
3103 }
3104 
3105 /*
3106  * Called with mm->mmap_sem held for writing.
3107  * Insert a new vma covering the given region, with the given flags.
3108  * Its pages are supplied by the given array of struct page *.
3109  * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3110  * The region past the last page supplied will always produce SIGBUS.
3111  * The array pointer and the pages it points to are assumed to stay alive
3112  * for as long as this mapping might exist.
3113  */
3114 struct vm_area_struct *_install_special_mapping(
3115 	struct mm_struct *mm,
3116 	unsigned long addr, unsigned long len,
3117 	unsigned long vm_flags, const struct vm_special_mapping *spec)
3118 {
3119 	return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3120 					&special_mapping_vmops);
3121 }
3122 
3123 int install_special_mapping(struct mm_struct *mm,
3124 			    unsigned long addr, unsigned long len,
3125 			    unsigned long vm_flags, struct page **pages)
3126 {
3127 	struct vm_area_struct *vma = __install_special_mapping(
3128 		mm, addr, len, vm_flags, (void *)pages,
3129 		&legacy_special_mapping_vmops);
3130 
3131 	return PTR_ERR_OR_ZERO(vma);
3132 }
3133 
3134 static DEFINE_MUTEX(mm_all_locks_mutex);
3135 
3136 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3137 {
3138 	if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3139 		/*
3140 		 * The LSB of head.next can't change from under us
3141 		 * because we hold the mm_all_locks_mutex.
3142 		 */
3143 		down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3144 		/*
3145 		 * We can safely modify head.next after taking the
3146 		 * anon_vma->root->rwsem. If some other vma in this mm shares
3147 		 * the same anon_vma we won't take it again.
3148 		 *
3149 		 * No need of atomic instructions here, head.next
3150 		 * can't change from under us thanks to the
3151 		 * anon_vma->root->rwsem.
3152 		 */
3153 		if (__test_and_set_bit(0, (unsigned long *)
3154 				       &anon_vma->root->rb_root.rb_node))
3155 			BUG();
3156 	}
3157 }
3158 
3159 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3160 {
3161 	if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3162 		/*
3163 		 * AS_MM_ALL_LOCKS can't change from under us because
3164 		 * we hold the mm_all_locks_mutex.
3165 		 *
3166 		 * Operations on ->flags have to be atomic because
3167 		 * even if AS_MM_ALL_LOCKS is stable thanks to the
3168 		 * mm_all_locks_mutex, there may be other cpus
3169 		 * changing other bitflags in parallel to us.
3170 		 */
3171 		if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3172 			BUG();
3173 		down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
3174 	}
3175 }
3176 
3177 /*
3178  * This operation locks against the VM for all pte/vma/mm related
3179  * operations that could ever happen on a certain mm. This includes
3180  * vmtruncate, try_to_unmap, and all page faults.
3181  *
3182  * The caller must take the mmap_sem in write mode before calling
3183  * mm_take_all_locks(). The caller isn't allowed to release the
3184  * mmap_sem until mm_drop_all_locks() returns.
3185  *
3186  * mmap_sem in write mode is required in order to block all operations
3187  * that could modify pagetables and free pages without need of
3188  * altering the vma layout. It's also needed in write mode to avoid new
3189  * anon_vmas to be associated with existing vmas.
3190  *
3191  * A single task can't take more than one mm_take_all_locks() in a row
3192  * or it would deadlock.
3193  *
3194  * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3195  * mapping->flags avoid to take the same lock twice, if more than one
3196  * vma in this mm is backed by the same anon_vma or address_space.
3197  *
3198  * We take locks in following order, accordingly to comment at beginning
3199  * of mm/rmap.c:
3200  *   - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3201  *     hugetlb mapping);
3202  *   - all i_mmap_rwsem locks;
3203  *   - all anon_vma->rwseml
3204  *
3205  * We can take all locks within these types randomly because the VM code
3206  * doesn't nest them and we protected from parallel mm_take_all_locks() by
3207  * mm_all_locks_mutex.
3208  *
3209  * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3210  * that may have to take thousand of locks.
3211  *
3212  * mm_take_all_locks() can fail if it's interrupted by signals.
3213  */
3214 int mm_take_all_locks(struct mm_struct *mm)
3215 {
3216 	struct vm_area_struct *vma;
3217 	struct anon_vma_chain *avc;
3218 
3219 	BUG_ON(down_read_trylock(&mm->mmap_sem));
3220 
3221 	mutex_lock(&mm_all_locks_mutex);
3222 
3223 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
3224 		if (signal_pending(current))
3225 			goto out_unlock;
3226 		if (vma->vm_file && vma->vm_file->f_mapping &&
3227 				is_vm_hugetlb_page(vma))
3228 			vm_lock_mapping(mm, vma->vm_file->f_mapping);
3229 	}
3230 
3231 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
3232 		if (signal_pending(current))
3233 			goto out_unlock;
3234 		if (vma->vm_file && vma->vm_file->f_mapping &&
3235 				!is_vm_hugetlb_page(vma))
3236 			vm_lock_mapping(mm, vma->vm_file->f_mapping);
3237 	}
3238 
3239 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
3240 		if (signal_pending(current))
3241 			goto out_unlock;
3242 		if (vma->anon_vma)
3243 			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3244 				vm_lock_anon_vma(mm, avc->anon_vma);
3245 	}
3246 
3247 	return 0;
3248 
3249 out_unlock:
3250 	mm_drop_all_locks(mm);
3251 	return -EINTR;
3252 }
3253 
3254 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3255 {
3256 	if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3257 		/*
3258 		 * The LSB of head.next can't change to 0 from under
3259 		 * us because we hold the mm_all_locks_mutex.
3260 		 *
3261 		 * We must however clear the bitflag before unlocking
3262 		 * the vma so the users using the anon_vma->rb_root will
3263 		 * never see our bitflag.
3264 		 *
3265 		 * No need of atomic instructions here, head.next
3266 		 * can't change from under us until we release the
3267 		 * anon_vma->root->rwsem.
3268 		 */
3269 		if (!__test_and_clear_bit(0, (unsigned long *)
3270 					  &anon_vma->root->rb_root.rb_node))
3271 			BUG();
3272 		anon_vma_unlock_write(anon_vma);
3273 	}
3274 }
3275 
3276 static void vm_unlock_mapping(struct address_space *mapping)
3277 {
3278 	if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3279 		/*
3280 		 * AS_MM_ALL_LOCKS can't change to 0 from under us
3281 		 * because we hold the mm_all_locks_mutex.
3282 		 */
3283 		i_mmap_unlock_write(mapping);
3284 		if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3285 					&mapping->flags))
3286 			BUG();
3287 	}
3288 }
3289 
3290 /*
3291  * The mmap_sem cannot be released by the caller until
3292  * mm_drop_all_locks() returns.
3293  */
3294 void mm_drop_all_locks(struct mm_struct *mm)
3295 {
3296 	struct vm_area_struct *vma;
3297 	struct anon_vma_chain *avc;
3298 
3299 	BUG_ON(down_read_trylock(&mm->mmap_sem));
3300 	BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3301 
3302 	for (vma = mm->mmap; vma; vma = vma->vm_next) {
3303 		if (vma->anon_vma)
3304 			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3305 				vm_unlock_anon_vma(avc->anon_vma);
3306 		if (vma->vm_file && vma->vm_file->f_mapping)
3307 			vm_unlock_mapping(vma->vm_file->f_mapping);
3308 	}
3309 
3310 	mutex_unlock(&mm_all_locks_mutex);
3311 }
3312 
3313 /*
3314  * initialise the VMA slab
3315  */
3316 void __init mmap_init(void)
3317 {
3318 	int ret;
3319 
3320 	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3321 	VM_BUG_ON(ret);
3322 }
3323 
3324 /*
3325  * Initialise sysctl_user_reserve_kbytes.
3326  *
3327  * This is intended to prevent a user from starting a single memory hogging
3328  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3329  * mode.
3330  *
3331  * The default value is min(3% of free memory, 128MB)
3332  * 128MB is enough to recover with sshd/login, bash, and top/kill.
3333  */
3334 static int init_user_reserve(void)
3335 {
3336 	unsigned long free_kbytes;
3337 
3338 	free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3339 
3340 	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3341 	return 0;
3342 }
3343 subsys_initcall(init_user_reserve);
3344 
3345 /*
3346  * Initialise sysctl_admin_reserve_kbytes.
3347  *
3348  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3349  * to log in and kill a memory hogging process.
3350  *
3351  * Systems with more than 256MB will reserve 8MB, enough to recover
3352  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3353  * only reserve 3% of free pages by default.
3354  */
3355 static int init_admin_reserve(void)
3356 {
3357 	unsigned long free_kbytes;
3358 
3359 	free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3360 
3361 	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3362 	return 0;
3363 }
3364 subsys_initcall(init_admin_reserve);
3365 
3366 /*
3367  * Reinititalise user and admin reserves if memory is added or removed.
3368  *
3369  * The default user reserve max is 128MB, and the default max for the
3370  * admin reserve is 8MB. These are usually, but not always, enough to
3371  * enable recovery from a memory hogging process using login/sshd, a shell,
3372  * and tools like top. It may make sense to increase or even disable the
3373  * reserve depending on the existence of swap or variations in the recovery
3374  * tools. So, the admin may have changed them.
3375  *
3376  * If memory is added and the reserves have been eliminated or increased above
3377  * the default max, then we'll trust the admin.
3378  *
3379  * If memory is removed and there isn't enough free memory, then we
3380  * need to reset the reserves.
3381  *
3382  * Otherwise keep the reserve set by the admin.
3383  */
3384 static int reserve_mem_notifier(struct notifier_block *nb,
3385 			     unsigned long action, void *data)
3386 {
3387 	unsigned long tmp, free_kbytes;
3388 
3389 	switch (action) {
3390 	case MEM_ONLINE:
3391 		/* Default max is 128MB. Leave alone if modified by operator. */
3392 		tmp = sysctl_user_reserve_kbytes;
3393 		if (0 < tmp && tmp < (1UL << 17))
3394 			init_user_reserve();
3395 
3396 		/* Default max is 8MB.  Leave alone if modified by operator. */
3397 		tmp = sysctl_admin_reserve_kbytes;
3398 		if (0 < tmp && tmp < (1UL << 13))
3399 			init_admin_reserve();
3400 
3401 		break;
3402 	case MEM_OFFLINE:
3403 		free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3404 
3405 		if (sysctl_user_reserve_kbytes > free_kbytes) {
3406 			init_user_reserve();
3407 			pr_info("vm.user_reserve_kbytes reset to %lu\n",
3408 				sysctl_user_reserve_kbytes);
3409 		}
3410 
3411 		if (sysctl_admin_reserve_kbytes > free_kbytes) {
3412 			init_admin_reserve();
3413 			pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3414 				sysctl_admin_reserve_kbytes);
3415 		}
3416 		break;
3417 	default:
3418 		break;
3419 	}
3420 	return NOTIFY_OK;
3421 }
3422 
3423 static struct notifier_block reserve_mem_nb = {
3424 	.notifier_call = reserve_mem_notifier,
3425 };
3426 
3427 static int __meminit init_reserve_notifier(void)
3428 {
3429 	if (register_hotmemory_notifier(&reserve_mem_nb))
3430 		pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3431 
3432 	return 0;
3433 }
3434 subsys_initcall(init_reserve_notifier);
3435