1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * mm/mmap.c
4  *
5  * Written by obz.
6  *
7  * Address space accounting code	<alan@lxorguk.ukuu.org.uk>
8  */
9 
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
14 #include <linux/backing-dev.h>
15 #include <linux/mm.h>
16 #include <linux/mm_inline.h>
17 #include <linux/shm.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/syscalls.h>
22 #include <linux/capability.h>
23 #include <linux/init.h>
24 #include <linux/file.h>
25 #include <linux/fs.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/hugetlb.h>
29 #include <linux/shmem_fs.h>
30 #include <linux/profile.h>
31 #include <linux/export.h>
32 #include <linux/mount.h>
33 #include <linux/mempolicy.h>
34 #include <linux/rmap.h>
35 #include <linux/mmu_notifier.h>
36 #include <linux/mmdebug.h>
37 #include <linux/perf_event.h>
38 #include <linux/audit.h>
39 #include <linux/khugepaged.h>
40 #include <linux/uprobes.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 #include <linux/pkeys.h>
47 #include <linux/oom.h>
48 #include <linux/sched/mm.h>
49 #include <linux/ksm.h>
50 
51 #include <linux/uaccess.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlb.h>
54 #include <asm/mmu_context.h>
55 
56 #define CREATE_TRACE_POINTS
57 #include <trace/events/mmap.h>
58 
59 #include "internal.h"
60 
61 #ifndef arch_mmap_check
62 #define arch_mmap_check(addr, len, flags)	(0)
63 #endif
64 
65 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
66 const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
67 const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
68 int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
69 #endif
70 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
71 const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
72 const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
73 int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
74 #endif
75 
76 static bool ignore_rlimit_data;
77 core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
78 
79 static void unmap_region(struct mm_struct *mm, struct ma_state *mas,
80 		struct vm_area_struct *vma, struct vm_area_struct *prev,
81 		struct vm_area_struct *next, unsigned long start,
82 		unsigned long end, unsigned long tree_end, bool mm_wr_locked);
83 
vm_pgprot_modify(pgprot_t oldprot,unsigned long vm_flags)84 static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
85 {
86 	return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
87 }
88 
89 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
vma_set_page_prot(struct vm_area_struct * vma)90 void vma_set_page_prot(struct vm_area_struct *vma)
91 {
92 	unsigned long vm_flags = vma->vm_flags;
93 	pgprot_t vm_page_prot;
94 
95 	vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
96 	if (vma_wants_writenotify(vma, vm_page_prot)) {
97 		vm_flags &= ~VM_SHARED;
98 		vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
99 	}
100 	/* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
101 	WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
102 }
103 
104 /*
105  * Requires inode->i_mapping->i_mmap_rwsem
106  */
__remove_shared_vm_struct(struct vm_area_struct * vma,struct file * file,struct address_space * mapping)107 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
108 		struct file *file, struct address_space *mapping)
109 {
110 	if (vma->vm_flags & VM_SHARED)
111 		mapping_unmap_writable(mapping);
112 
113 	flush_dcache_mmap_lock(mapping);
114 	vma_interval_tree_remove(vma, &mapping->i_mmap);
115 	flush_dcache_mmap_unlock(mapping);
116 }
117 
118 /*
119  * Unlink a file-based vm structure from its interval tree, to hide
120  * vma from rmap and vmtruncate before freeing its page tables.
121  */
unlink_file_vma(struct vm_area_struct * vma)122 void unlink_file_vma(struct vm_area_struct *vma)
123 {
124 	struct file *file = vma->vm_file;
125 
126 	if (file) {
127 		struct address_space *mapping = file->f_mapping;
128 		i_mmap_lock_write(mapping);
129 		__remove_shared_vm_struct(vma, file, mapping);
130 		i_mmap_unlock_write(mapping);
131 	}
132 }
133 
134 /*
135  * Close a vm structure and free it.
136  */
remove_vma(struct vm_area_struct * vma,bool unreachable)137 static void remove_vma(struct vm_area_struct *vma, bool unreachable)
138 {
139 	might_sleep();
140 	if (vma->vm_ops && vma->vm_ops->close)
141 		vma->vm_ops->close(vma);
142 	if (vma->vm_file)
143 		fput(vma->vm_file);
144 	mpol_put(vma_policy(vma));
145 	if (unreachable)
146 		__vm_area_free(vma);
147 	else
148 		vm_area_free(vma);
149 }
150 
vma_prev_limit(struct vma_iterator * vmi,unsigned long min)151 static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi,
152 						    unsigned long min)
153 {
154 	return mas_prev(&vmi->mas, min);
155 }
156 
157 /*
158  * check_brk_limits() - Use platform specific check of range & verify mlock
159  * limits.
160  * @addr: The address to check
161  * @len: The size of increase.
162  *
163  * Return: 0 on success.
164  */
check_brk_limits(unsigned long addr,unsigned long len)165 static int check_brk_limits(unsigned long addr, unsigned long len)
166 {
167 	unsigned long mapped_addr;
168 
169 	mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
170 	if (IS_ERR_VALUE(mapped_addr))
171 		return mapped_addr;
172 
173 	return mlock_future_ok(current->mm, current->mm->def_flags, len)
174 		? 0 : -EAGAIN;
175 }
176 static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma,
177 		unsigned long addr, unsigned long request, unsigned long flags);
SYSCALL_DEFINE1(brk,unsigned long,brk)178 SYSCALL_DEFINE1(brk, unsigned long, brk)
179 {
180 	unsigned long newbrk, oldbrk, origbrk;
181 	struct mm_struct *mm = current->mm;
182 	struct vm_area_struct *brkvma, *next = NULL;
183 	unsigned long min_brk;
184 	bool populate = false;
185 	LIST_HEAD(uf);
186 	struct vma_iterator vmi;
187 
188 	if (mmap_write_lock_killable(mm))
189 		return -EINTR;
190 
191 	origbrk = mm->brk;
192 
193 #ifdef CONFIG_COMPAT_BRK
194 	/*
195 	 * CONFIG_COMPAT_BRK can still be overridden by setting
196 	 * randomize_va_space to 2, which will still cause mm->start_brk
197 	 * to be arbitrarily shifted
198 	 */
199 	if (current->brk_randomized)
200 		min_brk = mm->start_brk;
201 	else
202 		min_brk = mm->end_data;
203 #else
204 	min_brk = mm->start_brk;
205 #endif
206 	if (brk < min_brk)
207 		goto out;
208 
209 	/*
210 	 * Check against rlimit here. If this check is done later after the test
211 	 * of oldbrk with newbrk then it can escape the test and let the data
212 	 * segment grow beyond its set limit the in case where the limit is
213 	 * not page aligned -Ram Gupta
214 	 */
215 	if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
216 			      mm->end_data, mm->start_data))
217 		goto out;
218 
219 	newbrk = PAGE_ALIGN(brk);
220 	oldbrk = PAGE_ALIGN(mm->brk);
221 	if (oldbrk == newbrk) {
222 		mm->brk = brk;
223 		goto success;
224 	}
225 
226 	/* Always allow shrinking brk. */
227 	if (brk <= mm->brk) {
228 		/* Search one past newbrk */
229 		vma_iter_init(&vmi, mm, newbrk);
230 		brkvma = vma_find(&vmi, oldbrk);
231 		if (!brkvma || brkvma->vm_start >= oldbrk)
232 			goto out; /* mapping intersects with an existing non-brk vma. */
233 		/*
234 		 * mm->brk must be protected by write mmap_lock.
235 		 * do_vma_munmap() will drop the lock on success,  so update it
236 		 * before calling do_vma_munmap().
237 		 */
238 		mm->brk = brk;
239 		if (do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true))
240 			goto out;
241 
242 		goto success_unlocked;
243 	}
244 
245 	if (check_brk_limits(oldbrk, newbrk - oldbrk))
246 		goto out;
247 
248 	/*
249 	 * Only check if the next VMA is within the stack_guard_gap of the
250 	 * expansion area
251 	 */
252 	vma_iter_init(&vmi, mm, oldbrk);
253 	next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap);
254 	if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
255 		goto out;
256 
257 	brkvma = vma_prev_limit(&vmi, mm->start_brk);
258 	/* Ok, looks good - let it rip. */
259 	if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0)
260 		goto out;
261 
262 	mm->brk = brk;
263 	if (mm->def_flags & VM_LOCKED)
264 		populate = true;
265 
266 success:
267 	mmap_write_unlock(mm);
268 success_unlocked:
269 	userfaultfd_unmap_complete(mm, &uf);
270 	if (populate)
271 		mm_populate(oldbrk, newbrk - oldbrk);
272 	return brk;
273 
274 out:
275 	mm->brk = origbrk;
276 	mmap_write_unlock(mm);
277 	return origbrk;
278 }
279 
280 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
validate_mm(struct mm_struct * mm)281 static void validate_mm(struct mm_struct *mm)
282 {
283 	int bug = 0;
284 	int i = 0;
285 	struct vm_area_struct *vma;
286 	VMA_ITERATOR(vmi, mm, 0);
287 
288 	mt_validate(&mm->mm_mt);
289 	for_each_vma(vmi, vma) {
290 #ifdef CONFIG_DEBUG_VM_RB
291 		struct anon_vma *anon_vma = vma->anon_vma;
292 		struct anon_vma_chain *avc;
293 #endif
294 		unsigned long vmi_start, vmi_end;
295 		bool warn = 0;
296 
297 		vmi_start = vma_iter_addr(&vmi);
298 		vmi_end = vma_iter_end(&vmi);
299 		if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm))
300 			warn = 1;
301 
302 		if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm))
303 			warn = 1;
304 
305 		if (warn) {
306 			pr_emerg("issue in %s\n", current->comm);
307 			dump_stack();
308 			dump_vma(vma);
309 			pr_emerg("tree range: %px start %lx end %lx\n", vma,
310 				 vmi_start, vmi_end - 1);
311 			vma_iter_dump_tree(&vmi);
312 		}
313 
314 #ifdef CONFIG_DEBUG_VM_RB
315 		if (anon_vma) {
316 			anon_vma_lock_read(anon_vma);
317 			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
318 				anon_vma_interval_tree_verify(avc);
319 			anon_vma_unlock_read(anon_vma);
320 		}
321 #endif
322 		i++;
323 	}
324 	if (i != mm->map_count) {
325 		pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i);
326 		bug = 1;
327 	}
328 	VM_BUG_ON_MM(bug, mm);
329 }
330 
331 #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */
332 #define validate_mm(mm) do { } while (0)
333 #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */
334 
335 /*
336  * vma has some anon_vma assigned, and is already inserted on that
337  * anon_vma's interval trees.
338  *
339  * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
340  * vma must be removed from the anon_vma's interval trees using
341  * anon_vma_interval_tree_pre_update_vma().
342  *
343  * After the update, the vma will be reinserted using
344  * anon_vma_interval_tree_post_update_vma().
345  *
346  * The entire update must be protected by exclusive mmap_lock and by
347  * the root anon_vma's mutex.
348  */
349 static inline void
anon_vma_interval_tree_pre_update_vma(struct vm_area_struct * vma)350 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
351 {
352 	struct anon_vma_chain *avc;
353 
354 	list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
355 		anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
356 }
357 
358 static inline void
anon_vma_interval_tree_post_update_vma(struct vm_area_struct * vma)359 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
360 {
361 	struct anon_vma_chain *avc;
362 
363 	list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
364 		anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
365 }
366 
count_vma_pages_range(struct mm_struct * mm,unsigned long addr,unsigned long end)367 static unsigned long count_vma_pages_range(struct mm_struct *mm,
368 		unsigned long addr, unsigned long end)
369 {
370 	VMA_ITERATOR(vmi, mm, addr);
371 	struct vm_area_struct *vma;
372 	unsigned long nr_pages = 0;
373 
374 	for_each_vma_range(vmi, vma, end) {
375 		unsigned long vm_start = max(addr, vma->vm_start);
376 		unsigned long vm_end = min(end, vma->vm_end);
377 
378 		nr_pages += PHYS_PFN(vm_end - vm_start);
379 	}
380 
381 	return nr_pages;
382 }
383 
__vma_link_file(struct vm_area_struct * vma,struct address_space * mapping)384 static void __vma_link_file(struct vm_area_struct *vma,
385 			    struct address_space *mapping)
386 {
387 	if (vma->vm_flags & VM_SHARED)
388 		mapping_allow_writable(mapping);
389 
390 	flush_dcache_mmap_lock(mapping);
391 	vma_interval_tree_insert(vma, &mapping->i_mmap);
392 	flush_dcache_mmap_unlock(mapping);
393 }
394 
vma_link(struct mm_struct * mm,struct vm_area_struct * vma)395 static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma)
396 {
397 	VMA_ITERATOR(vmi, mm, 0);
398 	struct address_space *mapping = NULL;
399 
400 	vma_iter_config(&vmi, vma->vm_start, vma->vm_end);
401 	if (vma_iter_prealloc(&vmi, vma))
402 		return -ENOMEM;
403 
404 	vma_start_write(vma);
405 
406 	vma_iter_store(&vmi, vma);
407 
408 	if (vma->vm_file) {
409 		mapping = vma->vm_file->f_mapping;
410 		i_mmap_lock_write(mapping);
411 		__vma_link_file(vma, mapping);
412 		i_mmap_unlock_write(mapping);
413 	}
414 
415 	mm->map_count++;
416 	validate_mm(mm);
417 	return 0;
418 }
419 
420 /*
421  * init_multi_vma_prep() - Initializer for struct vma_prepare
422  * @vp: The vma_prepare struct
423  * @vma: The vma that will be altered once locked
424  * @next: The next vma if it is to be adjusted
425  * @remove: The first vma to be removed
426  * @remove2: The second vma to be removed
427  */
init_multi_vma_prep(struct vma_prepare * vp,struct vm_area_struct * vma,struct vm_area_struct * next,struct vm_area_struct * remove,struct vm_area_struct * remove2)428 static inline void init_multi_vma_prep(struct vma_prepare *vp,
429 		struct vm_area_struct *vma, struct vm_area_struct *next,
430 		struct vm_area_struct *remove, struct vm_area_struct *remove2)
431 {
432 	memset(vp, 0, sizeof(struct vma_prepare));
433 	vp->vma = vma;
434 	vp->anon_vma = vma->anon_vma;
435 	vp->remove = remove;
436 	vp->remove2 = remove2;
437 	vp->adj_next = next;
438 	if (!vp->anon_vma && next)
439 		vp->anon_vma = next->anon_vma;
440 
441 	vp->file = vma->vm_file;
442 	if (vp->file)
443 		vp->mapping = vma->vm_file->f_mapping;
444 
445 }
446 
447 /*
448  * init_vma_prep() - Initializer wrapper for vma_prepare struct
449  * @vp: The vma_prepare struct
450  * @vma: The vma that will be altered once locked
451  */
init_vma_prep(struct vma_prepare * vp,struct vm_area_struct * vma)452 static inline void init_vma_prep(struct vma_prepare *vp,
453 				 struct vm_area_struct *vma)
454 {
455 	init_multi_vma_prep(vp, vma, NULL, NULL, NULL);
456 }
457 
458 
459 /*
460  * vma_prepare() - Helper function for handling locking VMAs prior to altering
461  * @vp: The initialized vma_prepare struct
462  */
vma_prepare(struct vma_prepare * vp)463 static inline void vma_prepare(struct vma_prepare *vp)
464 {
465 	if (vp->file) {
466 		uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end);
467 
468 		if (vp->adj_next)
469 			uprobe_munmap(vp->adj_next, vp->adj_next->vm_start,
470 				      vp->adj_next->vm_end);
471 
472 		i_mmap_lock_write(vp->mapping);
473 		if (vp->insert && vp->insert->vm_file) {
474 			/*
475 			 * Put into interval tree now, so instantiated pages
476 			 * are visible to arm/parisc __flush_dcache_page
477 			 * throughout; but we cannot insert into address
478 			 * space until vma start or end is updated.
479 			 */
480 			__vma_link_file(vp->insert,
481 					vp->insert->vm_file->f_mapping);
482 		}
483 	}
484 
485 	if (vp->anon_vma) {
486 		anon_vma_lock_write(vp->anon_vma);
487 		anon_vma_interval_tree_pre_update_vma(vp->vma);
488 		if (vp->adj_next)
489 			anon_vma_interval_tree_pre_update_vma(vp->adj_next);
490 	}
491 
492 	if (vp->file) {
493 		flush_dcache_mmap_lock(vp->mapping);
494 		vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap);
495 		if (vp->adj_next)
496 			vma_interval_tree_remove(vp->adj_next,
497 						 &vp->mapping->i_mmap);
498 	}
499 
500 }
501 
502 /*
503  * vma_complete- Helper function for handling the unlocking after altering VMAs,
504  * or for inserting a VMA.
505  *
506  * @vp: The vma_prepare struct
507  * @vmi: The vma iterator
508  * @mm: The mm_struct
509  */
vma_complete(struct vma_prepare * vp,struct vma_iterator * vmi,struct mm_struct * mm)510 static inline void vma_complete(struct vma_prepare *vp,
511 				struct vma_iterator *vmi, struct mm_struct *mm)
512 {
513 	if (vp->file) {
514 		if (vp->adj_next)
515 			vma_interval_tree_insert(vp->adj_next,
516 						 &vp->mapping->i_mmap);
517 		vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap);
518 		flush_dcache_mmap_unlock(vp->mapping);
519 	}
520 
521 	if (vp->remove && vp->file) {
522 		__remove_shared_vm_struct(vp->remove, vp->file, vp->mapping);
523 		if (vp->remove2)
524 			__remove_shared_vm_struct(vp->remove2, vp->file,
525 						  vp->mapping);
526 	} else if (vp->insert) {
527 		/*
528 		 * split_vma has split insert from vma, and needs
529 		 * us to insert it before dropping the locks
530 		 * (it may either follow vma or precede it).
531 		 */
532 		vma_iter_store(vmi, vp->insert);
533 		mm->map_count++;
534 	}
535 
536 	if (vp->anon_vma) {
537 		anon_vma_interval_tree_post_update_vma(vp->vma);
538 		if (vp->adj_next)
539 			anon_vma_interval_tree_post_update_vma(vp->adj_next);
540 		anon_vma_unlock_write(vp->anon_vma);
541 	}
542 
543 	if (vp->file) {
544 		i_mmap_unlock_write(vp->mapping);
545 		uprobe_mmap(vp->vma);
546 
547 		if (vp->adj_next)
548 			uprobe_mmap(vp->adj_next);
549 	}
550 
551 	if (vp->remove) {
552 again:
553 		vma_mark_detached(vp->remove, true);
554 		if (vp->file) {
555 			uprobe_munmap(vp->remove, vp->remove->vm_start,
556 				      vp->remove->vm_end);
557 			fput(vp->file);
558 		}
559 		if (vp->remove->anon_vma)
560 			anon_vma_merge(vp->vma, vp->remove);
561 		mm->map_count--;
562 		mpol_put(vma_policy(vp->remove));
563 		if (!vp->remove2)
564 			WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end);
565 		vm_area_free(vp->remove);
566 
567 		/*
568 		 * In mprotect's case 6 (see comments on vma_merge),
569 		 * we are removing both mid and next vmas
570 		 */
571 		if (vp->remove2) {
572 			vp->remove = vp->remove2;
573 			vp->remove2 = NULL;
574 			goto again;
575 		}
576 	}
577 	if (vp->insert && vp->file)
578 		uprobe_mmap(vp->insert);
579 	validate_mm(mm);
580 }
581 
582 /*
583  * dup_anon_vma() - Helper function to duplicate anon_vma
584  * @dst: The destination VMA
585  * @src: The source VMA
586  * @dup: Pointer to the destination VMA when successful.
587  *
588  * Returns: 0 on success.
589  */
dup_anon_vma(struct vm_area_struct * dst,struct vm_area_struct * src,struct vm_area_struct ** dup)590 static inline int dup_anon_vma(struct vm_area_struct *dst,
591 		struct vm_area_struct *src, struct vm_area_struct **dup)
592 {
593 	/*
594 	 * Easily overlooked: when mprotect shifts the boundary, make sure the
595 	 * expanding vma has anon_vma set if the shrinking vma had, to cover any
596 	 * anon pages imported.
597 	 */
598 	if (src->anon_vma && !dst->anon_vma) {
599 		int ret;
600 
601 		vma_assert_write_locked(dst);
602 		dst->anon_vma = src->anon_vma;
603 		ret = anon_vma_clone(dst, src);
604 		if (ret)
605 			return ret;
606 
607 		*dup = dst;
608 	}
609 
610 	return 0;
611 }
612 
613 /*
614  * vma_expand - Expand an existing VMA
615  *
616  * @vmi: The vma iterator
617  * @vma: The vma to expand
618  * @start: The start of the vma
619  * @end: The exclusive end of the vma
620  * @pgoff: The page offset of vma
621  * @next: The current of next vma.
622  *
623  * Expand @vma to @start and @end.  Can expand off the start and end.  Will
624  * expand over @next if it's different from @vma and @end == @next->vm_end.
625  * Checking if the @vma can expand and merge with @next needs to be handled by
626  * the caller.
627  *
628  * Returns: 0 on success
629  */
vma_expand(struct vma_iterator * vmi,struct vm_area_struct * vma,unsigned long start,unsigned long end,pgoff_t pgoff,struct vm_area_struct * next)630 int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma,
631 	       unsigned long start, unsigned long end, pgoff_t pgoff,
632 	       struct vm_area_struct *next)
633 {
634 	struct vm_area_struct *anon_dup = NULL;
635 	bool remove_next = false;
636 	struct vma_prepare vp;
637 
638 	vma_start_write(vma);
639 	if (next && (vma != next) && (end == next->vm_end)) {
640 		int ret;
641 
642 		remove_next = true;
643 		vma_start_write(next);
644 		ret = dup_anon_vma(vma, next, &anon_dup);
645 		if (ret)
646 			return ret;
647 	}
648 
649 	init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL);
650 	/* Not merging but overwriting any part of next is not handled. */
651 	VM_WARN_ON(next && !vp.remove &&
652 		  next != vma && end > next->vm_start);
653 	/* Only handles expanding */
654 	VM_WARN_ON(vma->vm_start < start || vma->vm_end > end);
655 
656 	/* Note: vma iterator must be pointing to 'start' */
657 	vma_iter_config(vmi, start, end);
658 	if (vma_iter_prealloc(vmi, vma))
659 		goto nomem;
660 
661 	vma_prepare(&vp);
662 	vma_adjust_trans_huge(vma, start, end, 0);
663 	vma->vm_start = start;
664 	vma->vm_end = end;
665 	vma->vm_pgoff = pgoff;
666 	vma_iter_store(vmi, vma);
667 
668 	vma_complete(&vp, vmi, vma->vm_mm);
669 	return 0;
670 
671 nomem:
672 	if (anon_dup)
673 		unlink_anon_vmas(anon_dup);
674 	return -ENOMEM;
675 }
676 
677 /*
678  * vma_shrink() - Reduce an existing VMAs memory area
679  * @vmi: The vma iterator
680  * @vma: The VMA to modify
681  * @start: The new start
682  * @end: The new end
683  *
684  * Returns: 0 on success, -ENOMEM otherwise
685  */
vma_shrink(struct vma_iterator * vmi,struct vm_area_struct * vma,unsigned long start,unsigned long end,pgoff_t pgoff)686 int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
687 	       unsigned long start, unsigned long end, pgoff_t pgoff)
688 {
689 	struct vma_prepare vp;
690 
691 	WARN_ON((vma->vm_start != start) && (vma->vm_end != end));
692 
693 	if (vma->vm_start < start)
694 		vma_iter_config(vmi, vma->vm_start, start);
695 	else
696 		vma_iter_config(vmi, end, vma->vm_end);
697 
698 	if (vma_iter_prealloc(vmi, NULL))
699 		return -ENOMEM;
700 
701 	vma_start_write(vma);
702 
703 	init_vma_prep(&vp, vma);
704 	vma_prepare(&vp);
705 	vma_adjust_trans_huge(vma, start, end, 0);
706 
707 	vma_iter_clear(vmi);
708 	vma->vm_start = start;
709 	vma->vm_end = end;
710 	vma->vm_pgoff = pgoff;
711 	vma_complete(&vp, vmi, vma->vm_mm);
712 	return 0;
713 }
714 
715 /*
716  * If the vma has a ->close operation then the driver probably needs to release
717  * per-vma resources, so we don't attempt to merge those if the caller indicates
718  * the current vma may be removed as part of the merge.
719  */
is_mergeable_vma(struct vm_area_struct * vma,struct file * file,unsigned long vm_flags,struct vm_userfaultfd_ctx vm_userfaultfd_ctx,struct anon_vma_name * anon_name,bool may_remove_vma)720 static inline bool is_mergeable_vma(struct vm_area_struct *vma,
721 		struct file *file, unsigned long vm_flags,
722 		struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
723 		struct anon_vma_name *anon_name, bool may_remove_vma)
724 {
725 	/*
726 	 * VM_SOFTDIRTY should not prevent from VMA merging, if we
727 	 * match the flags but dirty bit -- the caller should mark
728 	 * merged VMA as dirty. If dirty bit won't be excluded from
729 	 * comparison, we increase pressure on the memory system forcing
730 	 * the kernel to generate new VMAs when old one could be
731 	 * extended instead.
732 	 */
733 	if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
734 		return false;
735 	if (vma->vm_file != file)
736 		return false;
737 	if (may_remove_vma && vma->vm_ops && vma->vm_ops->close)
738 		return false;
739 	if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
740 		return false;
741 	if (!anon_vma_name_eq(anon_vma_name(vma), anon_name))
742 		return false;
743 	return true;
744 }
745 
is_mergeable_anon_vma(struct anon_vma * anon_vma1,struct anon_vma * anon_vma2,struct vm_area_struct * vma)746 static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1,
747 		 struct anon_vma *anon_vma2, struct vm_area_struct *vma)
748 {
749 	/*
750 	 * The list_is_singular() test is to avoid merging VMA cloned from
751 	 * parents. This can improve scalability caused by anon_vma lock.
752 	 */
753 	if ((!anon_vma1 || !anon_vma2) && (!vma ||
754 		list_is_singular(&vma->anon_vma_chain)))
755 		return true;
756 	return anon_vma1 == anon_vma2;
757 }
758 
759 /*
760  * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
761  * in front of (at a lower virtual address and file offset than) the vma.
762  *
763  * We cannot merge two vmas if they have differently assigned (non-NULL)
764  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
765  *
766  * We don't check here for the merged mmap wrapping around the end of pagecache
767  * indices (16TB on ia32) because do_mmap() does not permit mmap's which
768  * wrap, nor mmaps which cover the final page at index -1UL.
769  *
770  * We assume the vma may be removed as part of the merge.
771  */
772 static bool
can_vma_merge_before(struct vm_area_struct * vma,unsigned long vm_flags,struct anon_vma * anon_vma,struct file * file,pgoff_t vm_pgoff,struct vm_userfaultfd_ctx vm_userfaultfd_ctx,struct anon_vma_name * anon_name)773 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
774 		struct anon_vma *anon_vma, struct file *file,
775 		pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
776 		struct anon_vma_name *anon_name)
777 {
778 	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) &&
779 	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
780 		if (vma->vm_pgoff == vm_pgoff)
781 			return true;
782 	}
783 	return false;
784 }
785 
786 /*
787  * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
788  * beyond (at a higher virtual address and file offset than) the vma.
789  *
790  * We cannot merge two vmas if they have differently assigned (non-NULL)
791  * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
792  *
793  * We assume that vma is not removed as part of the merge.
794  */
795 static bool
can_vma_merge_after(struct vm_area_struct * vma,unsigned long vm_flags,struct anon_vma * anon_vma,struct file * file,pgoff_t vm_pgoff,struct vm_userfaultfd_ctx vm_userfaultfd_ctx,struct anon_vma_name * anon_name)796 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
797 		struct anon_vma *anon_vma, struct file *file,
798 		pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
799 		struct anon_vma_name *anon_name)
800 {
801 	if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) &&
802 	    is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
803 		pgoff_t vm_pglen;
804 		vm_pglen = vma_pages(vma);
805 		if (vma->vm_pgoff + vm_pglen == vm_pgoff)
806 			return true;
807 	}
808 	return false;
809 }
810 
811 /*
812  * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
813  * figure out whether that can be merged with its predecessor or its
814  * successor.  Or both (it neatly fills a hole).
815  *
816  * In most cases - when called for mmap, brk or mremap - [addr,end) is
817  * certain not to be mapped by the time vma_merge is called; but when
818  * called for mprotect, it is certain to be already mapped (either at
819  * an offset within prev, or at the start of next), and the flags of
820  * this area are about to be changed to vm_flags - and the no-change
821  * case has already been eliminated.
822  *
823  * The following mprotect cases have to be considered, where **** is
824  * the area passed down from mprotect_fixup, never extending beyond one
825  * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts
826  * at the same address as **** and is of the same or larger span, and
827  * NNNN the next vma after ****:
828  *
829  *     ****             ****                   ****
830  *    PPPPPPNNNNNN    PPPPPPNNNNNN       PPPPPPCCCCCC
831  *    cannot merge    might become       might become
832  *                    PPNNNNNNNNNN       PPPPPPPPPPCC
833  *    mmap, brk or    case 4 below       case 5 below
834  *    mremap move:
835  *                        ****               ****
836  *                    PPPP    NNNN       PPPPCCCCNNNN
837  *                    might become       might become
838  *                    PPPPPPPPPPPP 1 or  PPPPPPPPPPPP 6 or
839  *                    PPPPPPPPNNNN 2 or  PPPPPPPPNNNN 7 or
840  *                    PPPPNNNNNNNN 3     PPPPNNNNNNNN 8
841  *
842  * It is important for case 8 that the vma CCCC overlapping the
843  * region **** is never going to extended over NNNN. Instead NNNN must
844  * be extended in region **** and CCCC must be removed. This way in
845  * all cases where vma_merge succeeds, the moment vma_merge drops the
846  * rmap_locks, the properties of the merged vma will be already
847  * correct for the whole merged range. Some of those properties like
848  * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
849  * be correct for the whole merged range immediately after the
850  * rmap_locks are released. Otherwise if NNNN would be removed and
851  * CCCC would be extended over the NNNN range, remove_migration_ptes
852  * or other rmap walkers (if working on addresses beyond the "end"
853  * parameter) may establish ptes with the wrong permissions of CCCC
854  * instead of the right permissions of NNNN.
855  *
856  * In the code below:
857  * PPPP is represented by *prev
858  * CCCC is represented by *curr or not represented at all (NULL)
859  * NNNN is represented by *next or not represented at all (NULL)
860  * **** is not represented - it will be merged and the vma containing the
861  *      area is returned, or the function will return NULL
862  */
vma_merge(struct vma_iterator * vmi,struct mm_struct * mm,struct vm_area_struct * prev,unsigned long addr,unsigned long end,unsigned long vm_flags,struct anon_vma * anon_vma,struct file * file,pgoff_t pgoff,struct mempolicy * policy,struct vm_userfaultfd_ctx vm_userfaultfd_ctx,struct anon_vma_name * anon_name)863 struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm,
864 			struct vm_area_struct *prev, unsigned long addr,
865 			unsigned long end, unsigned long vm_flags,
866 			struct anon_vma *anon_vma, struct file *file,
867 			pgoff_t pgoff, struct mempolicy *policy,
868 			struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
869 			struct anon_vma_name *anon_name)
870 {
871 	struct vm_area_struct *curr, *next, *res;
872 	struct vm_area_struct *vma, *adjust, *remove, *remove2;
873 	struct vm_area_struct *anon_dup = NULL;
874 	struct vma_prepare vp;
875 	pgoff_t vma_pgoff;
876 	int err = 0;
877 	bool merge_prev = false;
878 	bool merge_next = false;
879 	bool vma_expanded = false;
880 	unsigned long vma_start = addr;
881 	unsigned long vma_end = end;
882 	pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
883 	long adj_start = 0;
884 
885 	/*
886 	 * We later require that vma->vm_flags == vm_flags,
887 	 * so this tests vma->vm_flags & VM_SPECIAL, too.
888 	 */
889 	if (vm_flags & VM_SPECIAL)
890 		return NULL;
891 
892 	/* Does the input range span an existing VMA? (cases 5 - 8) */
893 	curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end);
894 
895 	if (!curr ||			/* cases 1 - 4 */
896 	    end == curr->vm_end)	/* cases 6 - 8, adjacent VMA */
897 		next = vma_lookup(mm, end);
898 	else
899 		next = NULL;		/* case 5 */
900 
901 	if (prev) {
902 		vma_start = prev->vm_start;
903 		vma_pgoff = prev->vm_pgoff;
904 
905 		/* Can we merge the predecessor? */
906 		if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy)
907 		    && can_vma_merge_after(prev, vm_flags, anon_vma, file,
908 					   pgoff, vm_userfaultfd_ctx, anon_name)) {
909 			merge_prev = true;
910 			vma_prev(vmi);
911 		}
912 	}
913 
914 	/* Can we merge the successor? */
915 	if (next && mpol_equal(policy, vma_policy(next)) &&
916 	    can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen,
917 				 vm_userfaultfd_ctx, anon_name)) {
918 		merge_next = true;
919 	}
920 
921 	/* Verify some invariant that must be enforced by the caller. */
922 	VM_WARN_ON(prev && addr <= prev->vm_start);
923 	VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end));
924 	VM_WARN_ON(addr >= end);
925 
926 	if (!merge_prev && !merge_next)
927 		return NULL; /* Not mergeable. */
928 
929 	if (merge_prev)
930 		vma_start_write(prev);
931 
932 	res = vma = prev;
933 	remove = remove2 = adjust = NULL;
934 
935 	/* Can we merge both the predecessor and the successor? */
936 	if (merge_prev && merge_next &&
937 	    is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) {
938 		vma_start_write(next);
939 		remove = next;				/* case 1 */
940 		vma_end = next->vm_end;
941 		err = dup_anon_vma(prev, next, &anon_dup);
942 		if (curr) {				/* case 6 */
943 			vma_start_write(curr);
944 			remove = curr;
945 			remove2 = next;
946 			if (!next->anon_vma)
947 				err = dup_anon_vma(prev, curr, &anon_dup);
948 		}
949 	} else if (merge_prev) {			/* case 2 */
950 		if (curr) {
951 			vma_start_write(curr);
952 			err = dup_anon_vma(prev, curr, &anon_dup);
953 			if (end == curr->vm_end) {	/* case 7 */
954 				remove = curr;
955 			} else {			/* case 5 */
956 				adjust = curr;
957 				adj_start = (end - curr->vm_start);
958 			}
959 		}
960 	} else { /* merge_next */
961 		vma_start_write(next);
962 		res = next;
963 		if (prev && addr < prev->vm_end) {	/* case 4 */
964 			vma_start_write(prev);
965 			vma_end = addr;
966 			adjust = next;
967 			adj_start = -(prev->vm_end - addr);
968 			err = dup_anon_vma(next, prev, &anon_dup);
969 		} else {
970 			/*
971 			 * Note that cases 3 and 8 are the ONLY ones where prev
972 			 * is permitted to be (but is not necessarily) NULL.
973 			 */
974 			vma = next;			/* case 3 */
975 			vma_start = addr;
976 			vma_end = next->vm_end;
977 			vma_pgoff = next->vm_pgoff - pglen;
978 			if (curr) {			/* case 8 */
979 				vma_pgoff = curr->vm_pgoff;
980 				vma_start_write(curr);
981 				remove = curr;
982 				err = dup_anon_vma(next, curr, &anon_dup);
983 			}
984 		}
985 	}
986 
987 	/* Error in anon_vma clone. */
988 	if (err)
989 		goto anon_vma_fail;
990 
991 	if (vma_start < vma->vm_start || vma_end > vma->vm_end)
992 		vma_expanded = true;
993 
994 	if (vma_expanded) {
995 		vma_iter_config(vmi, vma_start, vma_end);
996 	} else {
997 		vma_iter_config(vmi, adjust->vm_start + adj_start,
998 				adjust->vm_end);
999 	}
1000 
1001 	if (vma_iter_prealloc(vmi, vma))
1002 		goto prealloc_fail;
1003 
1004 	init_multi_vma_prep(&vp, vma, adjust, remove, remove2);
1005 	VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma &&
1006 		   vp.anon_vma != adjust->anon_vma);
1007 
1008 	vma_prepare(&vp);
1009 	vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start);
1010 
1011 	vma->vm_start = vma_start;
1012 	vma->vm_end = vma_end;
1013 	vma->vm_pgoff = vma_pgoff;
1014 
1015 	if (vma_expanded)
1016 		vma_iter_store(vmi, vma);
1017 
1018 	if (adj_start) {
1019 		adjust->vm_start += adj_start;
1020 		adjust->vm_pgoff += adj_start >> PAGE_SHIFT;
1021 		if (adj_start < 0) {
1022 			WARN_ON(vma_expanded);
1023 			vma_iter_store(vmi, next);
1024 		}
1025 	}
1026 
1027 	vma_complete(&vp, vmi, mm);
1028 	khugepaged_enter_vma(res, vm_flags);
1029 	return res;
1030 
1031 prealloc_fail:
1032 	if (anon_dup)
1033 		unlink_anon_vmas(anon_dup);
1034 
1035 anon_vma_fail:
1036 	vma_iter_set(vmi, addr);
1037 	vma_iter_load(vmi);
1038 	return NULL;
1039 }
1040 
1041 /*
1042  * Rough compatibility check to quickly see if it's even worth looking
1043  * at sharing an anon_vma.
1044  *
1045  * They need to have the same vm_file, and the flags can only differ
1046  * in things that mprotect may change.
1047  *
1048  * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1049  * we can merge the two vma's. For example, we refuse to merge a vma if
1050  * there is a vm_ops->close() function, because that indicates that the
1051  * driver is doing some kind of reference counting. But that doesn't
1052  * really matter for the anon_vma sharing case.
1053  */
anon_vma_compatible(struct vm_area_struct * a,struct vm_area_struct * b)1054 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1055 {
1056 	return a->vm_end == b->vm_start &&
1057 		mpol_equal(vma_policy(a), vma_policy(b)) &&
1058 		a->vm_file == b->vm_file &&
1059 		!((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) &&
1060 		b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1061 }
1062 
1063 /*
1064  * Do some basic sanity checking to see if we can re-use the anon_vma
1065  * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1066  * the same as 'old', the other will be the new one that is trying
1067  * to share the anon_vma.
1068  *
1069  * NOTE! This runs with mmap_lock held for reading, so it is possible that
1070  * the anon_vma of 'old' is concurrently in the process of being set up
1071  * by another page fault trying to merge _that_. But that's ok: if it
1072  * is being set up, that automatically means that it will be a singleton
1073  * acceptable for merging, so we can do all of this optimistically. But
1074  * we do that READ_ONCE() to make sure that we never re-load the pointer.
1075  *
1076  * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1077  * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1078  * is to return an anon_vma that is "complex" due to having gone through
1079  * a fork).
1080  *
1081  * We also make sure that the two vma's are compatible (adjacent,
1082  * and with the same memory policies). That's all stable, even with just
1083  * a read lock on the mmap_lock.
1084  */
reusable_anon_vma(struct vm_area_struct * old,struct vm_area_struct * a,struct vm_area_struct * b)1085 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1086 {
1087 	if (anon_vma_compatible(a, b)) {
1088 		struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1089 
1090 		if (anon_vma && list_is_singular(&old->anon_vma_chain))
1091 			return anon_vma;
1092 	}
1093 	return NULL;
1094 }
1095 
1096 /*
1097  * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1098  * neighbouring vmas for a suitable anon_vma, before it goes off
1099  * to allocate a new anon_vma.  It checks because a repetitive
1100  * sequence of mprotects and faults may otherwise lead to distinct
1101  * anon_vmas being allocated, preventing vma merge in subsequent
1102  * mprotect.
1103  */
find_mergeable_anon_vma(struct vm_area_struct * vma)1104 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1105 {
1106 	MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end);
1107 	struct anon_vma *anon_vma = NULL;
1108 	struct vm_area_struct *prev, *next;
1109 
1110 	/* Try next first. */
1111 	next = mas_walk(&mas);
1112 	if (next) {
1113 		anon_vma = reusable_anon_vma(next, vma, next);
1114 		if (anon_vma)
1115 			return anon_vma;
1116 	}
1117 
1118 	prev = mas_prev(&mas, 0);
1119 	VM_BUG_ON_VMA(prev != vma, vma);
1120 	prev = mas_prev(&mas, 0);
1121 	/* Try prev next. */
1122 	if (prev)
1123 		anon_vma = reusable_anon_vma(prev, prev, vma);
1124 
1125 	/*
1126 	 * We might reach here with anon_vma == NULL if we can't find
1127 	 * any reusable anon_vma.
1128 	 * There's no absolute need to look only at touching neighbours:
1129 	 * we could search further afield for "compatible" anon_vmas.
1130 	 * But it would probably just be a waste of time searching,
1131 	 * or lead to too many vmas hanging off the same anon_vma.
1132 	 * We're trying to allow mprotect remerging later on,
1133 	 * not trying to minimize memory used for anon_vmas.
1134 	 */
1135 	return anon_vma;
1136 }
1137 
1138 /*
1139  * If a hint addr is less than mmap_min_addr change hint to be as
1140  * low as possible but still greater than mmap_min_addr
1141  */
round_hint_to_min(unsigned long hint)1142 static inline unsigned long round_hint_to_min(unsigned long hint)
1143 {
1144 	hint &= PAGE_MASK;
1145 	if (((void *)hint != NULL) &&
1146 	    (hint < mmap_min_addr))
1147 		return PAGE_ALIGN(mmap_min_addr);
1148 	return hint;
1149 }
1150 
mlock_future_ok(struct mm_struct * mm,unsigned long flags,unsigned long bytes)1151 bool mlock_future_ok(struct mm_struct *mm, unsigned long flags,
1152 			unsigned long bytes)
1153 {
1154 	unsigned long locked_pages, limit_pages;
1155 
1156 	if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK))
1157 		return true;
1158 
1159 	locked_pages = bytes >> PAGE_SHIFT;
1160 	locked_pages += mm->locked_vm;
1161 
1162 	limit_pages = rlimit(RLIMIT_MEMLOCK);
1163 	limit_pages >>= PAGE_SHIFT;
1164 
1165 	return locked_pages <= limit_pages;
1166 }
1167 
file_mmap_size_max(struct file * file,struct inode * inode)1168 static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1169 {
1170 	if (S_ISREG(inode->i_mode))
1171 		return MAX_LFS_FILESIZE;
1172 
1173 	if (S_ISBLK(inode->i_mode))
1174 		return MAX_LFS_FILESIZE;
1175 
1176 	if (S_ISSOCK(inode->i_mode))
1177 		return MAX_LFS_FILESIZE;
1178 
1179 	/* Special "we do even unsigned file positions" case */
1180 	if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1181 		return 0;
1182 
1183 	/* Yes, random drivers might want more. But I'm tired of buggy drivers */
1184 	return ULONG_MAX;
1185 }
1186 
file_mmap_ok(struct file * file,struct inode * inode,unsigned long pgoff,unsigned long len)1187 static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1188 				unsigned long pgoff, unsigned long len)
1189 {
1190 	u64 maxsize = file_mmap_size_max(file, inode);
1191 
1192 	if (maxsize && len > maxsize)
1193 		return false;
1194 	maxsize -= len;
1195 	if (pgoff > maxsize >> PAGE_SHIFT)
1196 		return false;
1197 	return true;
1198 }
1199 
1200 /*
1201  * The caller must write-lock current->mm->mmap_lock.
1202  */
do_mmap(struct file * file,unsigned long addr,unsigned long len,unsigned long prot,unsigned long flags,vm_flags_t vm_flags,unsigned long pgoff,unsigned long * populate,struct list_head * uf)1203 unsigned long do_mmap(struct file *file, unsigned long addr,
1204 			unsigned long len, unsigned long prot,
1205 			unsigned long flags, vm_flags_t vm_flags,
1206 			unsigned long pgoff, unsigned long *populate,
1207 			struct list_head *uf)
1208 {
1209 	struct mm_struct *mm = current->mm;
1210 	int pkey = 0;
1211 
1212 	*populate = 0;
1213 
1214 	if (!len)
1215 		return -EINVAL;
1216 
1217 	/*
1218 	 * Does the application expect PROT_READ to imply PROT_EXEC?
1219 	 *
1220 	 * (the exception is when the underlying filesystem is noexec
1221 	 *  mounted, in which case we dont add PROT_EXEC.)
1222 	 */
1223 	if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1224 		if (!(file && path_noexec(&file->f_path)))
1225 			prot |= PROT_EXEC;
1226 
1227 	/* force arch specific MAP_FIXED handling in get_unmapped_area */
1228 	if (flags & MAP_FIXED_NOREPLACE)
1229 		flags |= MAP_FIXED;
1230 
1231 	if (!(flags & MAP_FIXED))
1232 		addr = round_hint_to_min(addr);
1233 
1234 	/* Careful about overflows.. */
1235 	len = PAGE_ALIGN(len);
1236 	if (!len)
1237 		return -ENOMEM;
1238 
1239 	/* offset overflow? */
1240 	if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1241 		return -EOVERFLOW;
1242 
1243 	/* Too many mappings? */
1244 	if (mm->map_count > sysctl_max_map_count)
1245 		return -ENOMEM;
1246 
1247 	/* Obtain the address to map to. we verify (or select) it and ensure
1248 	 * that it represents a valid section of the address space.
1249 	 */
1250 	addr = get_unmapped_area(file, addr, len, pgoff, flags);
1251 	if (IS_ERR_VALUE(addr))
1252 		return addr;
1253 
1254 	if (flags & MAP_FIXED_NOREPLACE) {
1255 		if (find_vma_intersection(mm, addr, addr + len))
1256 			return -EEXIST;
1257 	}
1258 
1259 	if (prot == PROT_EXEC) {
1260 		pkey = execute_only_pkey(mm);
1261 		if (pkey < 0)
1262 			pkey = 0;
1263 	}
1264 
1265 	/* Do simple checking here so the lower-level routines won't have
1266 	 * to. we assume access permissions have been handled by the open
1267 	 * of the memory object, so we don't do any here.
1268 	 */
1269 	vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1270 			mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1271 
1272 	if (flags & MAP_LOCKED)
1273 		if (!can_do_mlock())
1274 			return -EPERM;
1275 
1276 	if (!mlock_future_ok(mm, vm_flags, len))
1277 		return -EAGAIN;
1278 
1279 	if (file) {
1280 		struct inode *inode = file_inode(file);
1281 		unsigned long flags_mask;
1282 
1283 		if (!file_mmap_ok(file, inode, pgoff, len))
1284 			return -EOVERFLOW;
1285 
1286 		flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1287 
1288 		switch (flags & MAP_TYPE) {
1289 		case MAP_SHARED:
1290 			/*
1291 			 * Force use of MAP_SHARED_VALIDATE with non-legacy
1292 			 * flags. E.g. MAP_SYNC is dangerous to use with
1293 			 * MAP_SHARED as you don't know which consistency model
1294 			 * you will get. We silently ignore unsupported flags
1295 			 * with MAP_SHARED to preserve backward compatibility.
1296 			 */
1297 			flags &= LEGACY_MAP_MASK;
1298 			fallthrough;
1299 		case MAP_SHARED_VALIDATE:
1300 			if (flags & ~flags_mask)
1301 				return -EOPNOTSUPP;
1302 			if (prot & PROT_WRITE) {
1303 				if (!(file->f_mode & FMODE_WRITE))
1304 					return -EACCES;
1305 				if (IS_SWAPFILE(file->f_mapping->host))
1306 					return -ETXTBSY;
1307 			}
1308 
1309 			/*
1310 			 * Make sure we don't allow writing to an append-only
1311 			 * file..
1312 			 */
1313 			if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1314 				return -EACCES;
1315 
1316 			vm_flags |= VM_SHARED | VM_MAYSHARE;
1317 			if (!(file->f_mode & FMODE_WRITE))
1318 				vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1319 			fallthrough;
1320 		case MAP_PRIVATE:
1321 			if (!(file->f_mode & FMODE_READ))
1322 				return -EACCES;
1323 			if (path_noexec(&file->f_path)) {
1324 				if (vm_flags & VM_EXEC)
1325 					return -EPERM;
1326 				vm_flags &= ~VM_MAYEXEC;
1327 			}
1328 
1329 			if (!file->f_op->mmap)
1330 				return -ENODEV;
1331 			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1332 				return -EINVAL;
1333 			break;
1334 
1335 		default:
1336 			return -EINVAL;
1337 		}
1338 	} else {
1339 		switch (flags & MAP_TYPE) {
1340 		case MAP_SHARED:
1341 			if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1342 				return -EINVAL;
1343 			/*
1344 			 * Ignore pgoff.
1345 			 */
1346 			pgoff = 0;
1347 			vm_flags |= VM_SHARED | VM_MAYSHARE;
1348 			break;
1349 		case MAP_PRIVATE:
1350 			/*
1351 			 * Set pgoff according to addr for anon_vma.
1352 			 */
1353 			pgoff = addr >> PAGE_SHIFT;
1354 			break;
1355 		default:
1356 			return -EINVAL;
1357 		}
1358 	}
1359 
1360 	/*
1361 	 * Set 'VM_NORESERVE' if we should not account for the
1362 	 * memory use of this mapping.
1363 	 */
1364 	if (flags & MAP_NORESERVE) {
1365 		/* We honor MAP_NORESERVE if allowed to overcommit */
1366 		if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1367 			vm_flags |= VM_NORESERVE;
1368 
1369 		/* hugetlb applies strict overcommit unless MAP_NORESERVE */
1370 		if (file && is_file_hugepages(file))
1371 			vm_flags |= VM_NORESERVE;
1372 	}
1373 
1374 	addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1375 	if (!IS_ERR_VALUE(addr) &&
1376 	    ((vm_flags & VM_LOCKED) ||
1377 	     (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1378 		*populate = len;
1379 	return addr;
1380 }
1381 
ksys_mmap_pgoff(unsigned long addr,unsigned long len,unsigned long prot,unsigned long flags,unsigned long fd,unsigned long pgoff)1382 unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1383 			      unsigned long prot, unsigned long flags,
1384 			      unsigned long fd, unsigned long pgoff)
1385 {
1386 	struct file *file = NULL;
1387 	unsigned long retval;
1388 
1389 	if (!(flags & MAP_ANONYMOUS)) {
1390 		audit_mmap_fd(fd, flags);
1391 		file = fget(fd);
1392 		if (!file)
1393 			return -EBADF;
1394 		if (is_file_hugepages(file)) {
1395 			len = ALIGN(len, huge_page_size(hstate_file(file)));
1396 		} else if (unlikely(flags & MAP_HUGETLB)) {
1397 			retval = -EINVAL;
1398 			goto out_fput;
1399 		}
1400 	} else if (flags & MAP_HUGETLB) {
1401 		struct hstate *hs;
1402 
1403 		hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1404 		if (!hs)
1405 			return -EINVAL;
1406 
1407 		len = ALIGN(len, huge_page_size(hs));
1408 		/*
1409 		 * VM_NORESERVE is used because the reservations will be
1410 		 * taken when vm_ops->mmap() is called
1411 		 */
1412 		file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1413 				VM_NORESERVE,
1414 				HUGETLB_ANONHUGE_INODE,
1415 				(flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1416 		if (IS_ERR(file))
1417 			return PTR_ERR(file);
1418 	}
1419 
1420 	retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1421 out_fput:
1422 	if (file)
1423 		fput(file);
1424 	return retval;
1425 }
1426 
SYSCALL_DEFINE6(mmap_pgoff,unsigned long,addr,unsigned long,len,unsigned long,prot,unsigned long,flags,unsigned long,fd,unsigned long,pgoff)1427 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1428 		unsigned long, prot, unsigned long, flags,
1429 		unsigned long, fd, unsigned long, pgoff)
1430 {
1431 	return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1432 }
1433 
1434 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1435 struct mmap_arg_struct {
1436 	unsigned long addr;
1437 	unsigned long len;
1438 	unsigned long prot;
1439 	unsigned long flags;
1440 	unsigned long fd;
1441 	unsigned long offset;
1442 };
1443 
SYSCALL_DEFINE1(old_mmap,struct mmap_arg_struct __user *,arg)1444 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1445 {
1446 	struct mmap_arg_struct a;
1447 
1448 	if (copy_from_user(&a, arg, sizeof(a)))
1449 		return -EFAULT;
1450 	if (offset_in_page(a.offset))
1451 		return -EINVAL;
1452 
1453 	return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1454 			       a.offset >> PAGE_SHIFT);
1455 }
1456 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1457 
vm_ops_needs_writenotify(const struct vm_operations_struct * vm_ops)1458 static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops)
1459 {
1460 	return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite);
1461 }
1462 
vma_is_shared_writable(struct vm_area_struct * vma)1463 static bool vma_is_shared_writable(struct vm_area_struct *vma)
1464 {
1465 	return (vma->vm_flags & (VM_WRITE | VM_SHARED)) ==
1466 		(VM_WRITE | VM_SHARED);
1467 }
1468 
vma_fs_can_writeback(struct vm_area_struct * vma)1469 static bool vma_fs_can_writeback(struct vm_area_struct *vma)
1470 {
1471 	/* No managed pages to writeback. */
1472 	if (vma->vm_flags & VM_PFNMAP)
1473 		return false;
1474 
1475 	return vma->vm_file && vma->vm_file->f_mapping &&
1476 		mapping_can_writeback(vma->vm_file->f_mapping);
1477 }
1478 
1479 /*
1480  * Does this VMA require the underlying folios to have their dirty state
1481  * tracked?
1482  */
vma_needs_dirty_tracking(struct vm_area_struct * vma)1483 bool vma_needs_dirty_tracking(struct vm_area_struct *vma)
1484 {
1485 	/* Only shared, writable VMAs require dirty tracking. */
1486 	if (!vma_is_shared_writable(vma))
1487 		return false;
1488 
1489 	/* Does the filesystem need to be notified? */
1490 	if (vm_ops_needs_writenotify(vma->vm_ops))
1491 		return true;
1492 
1493 	/*
1494 	 * Even if the filesystem doesn't indicate a need for writenotify, if it
1495 	 * can writeback, dirty tracking is still required.
1496 	 */
1497 	return vma_fs_can_writeback(vma);
1498 }
1499 
1500 /*
1501  * Some shared mappings will want the pages marked read-only
1502  * to track write events. If so, we'll downgrade vm_page_prot
1503  * to the private version (using protection_map[] without the
1504  * VM_SHARED bit).
1505  */
vma_wants_writenotify(struct vm_area_struct * vma,pgprot_t vm_page_prot)1506 int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
1507 {
1508 	/* If it was private or non-writable, the write bit is already clear */
1509 	if (!vma_is_shared_writable(vma))
1510 		return 0;
1511 
1512 	/* The backer wishes to know when pages are first written to? */
1513 	if (vm_ops_needs_writenotify(vma->vm_ops))
1514 		return 1;
1515 
1516 	/* The open routine did something to the protections that pgprot_modify
1517 	 * won't preserve? */
1518 	if (pgprot_val(vm_page_prot) !=
1519 	    pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags)))
1520 		return 0;
1521 
1522 	/*
1523 	 * Do we need to track softdirty? hugetlb does not support softdirty
1524 	 * tracking yet.
1525 	 */
1526 	if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma))
1527 		return 1;
1528 
1529 	/* Do we need write faults for uffd-wp tracking? */
1530 	if (userfaultfd_wp(vma))
1531 		return 1;
1532 
1533 	/* Can the mapping track the dirty pages? */
1534 	return vma_fs_can_writeback(vma);
1535 }
1536 
1537 /*
1538  * We account for memory if it's a private writeable mapping,
1539  * not hugepages and VM_NORESERVE wasn't set.
1540  */
accountable_mapping(struct file * file,vm_flags_t vm_flags)1541 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1542 {
1543 	/*
1544 	 * hugetlb has its own accounting separate from the core VM
1545 	 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1546 	 */
1547 	if (file && is_file_hugepages(file))
1548 		return 0;
1549 
1550 	return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1551 }
1552 
1553 /**
1554  * unmapped_area() - Find an area between the low_limit and the high_limit with
1555  * the correct alignment and offset, all from @info. Note: current->mm is used
1556  * for the search.
1557  *
1558  * @info: The unmapped area information including the range [low_limit -
1559  * high_limit), the alignment offset and mask.
1560  *
1561  * Return: A memory address or -ENOMEM.
1562  */
unmapped_area(struct vm_unmapped_area_info * info)1563 static unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1564 {
1565 	unsigned long length, gap;
1566 	unsigned long low_limit, high_limit;
1567 	struct vm_area_struct *tmp;
1568 
1569 	MA_STATE(mas, &current->mm->mm_mt, 0, 0);
1570 
1571 	/* Adjust search length to account for worst case alignment overhead */
1572 	length = info->length + info->align_mask;
1573 	if (length < info->length)
1574 		return -ENOMEM;
1575 
1576 	low_limit = info->low_limit;
1577 	if (low_limit < mmap_min_addr)
1578 		low_limit = mmap_min_addr;
1579 	high_limit = info->high_limit;
1580 retry:
1581 	if (mas_empty_area(&mas, low_limit, high_limit - 1, length))
1582 		return -ENOMEM;
1583 
1584 	gap = mas.index;
1585 	gap += (info->align_offset - gap) & info->align_mask;
1586 	tmp = mas_next(&mas, ULONG_MAX);
1587 	if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */
1588 		if (vm_start_gap(tmp) < gap + length - 1) {
1589 			low_limit = tmp->vm_end;
1590 			mas_reset(&mas);
1591 			goto retry;
1592 		}
1593 	} else {
1594 		tmp = mas_prev(&mas, 0);
1595 		if (tmp && vm_end_gap(tmp) > gap) {
1596 			low_limit = vm_end_gap(tmp);
1597 			mas_reset(&mas);
1598 			goto retry;
1599 		}
1600 	}
1601 
1602 	return gap;
1603 }
1604 
1605 /**
1606  * unmapped_area_topdown() - Find an area between the low_limit and the
1607  * high_limit with the correct alignment and offset at the highest available
1608  * address, all from @info. Note: current->mm is used for the search.
1609  *
1610  * @info: The unmapped area information including the range [low_limit -
1611  * high_limit), the alignment offset and mask.
1612  *
1613  * Return: A memory address or -ENOMEM.
1614  */
unmapped_area_topdown(struct vm_unmapped_area_info * info)1615 static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1616 {
1617 	unsigned long length, gap, gap_end;
1618 	unsigned long low_limit, high_limit;
1619 	struct vm_area_struct *tmp;
1620 
1621 	MA_STATE(mas, &current->mm->mm_mt, 0, 0);
1622 	/* Adjust search length to account for worst case alignment overhead */
1623 	length = info->length + info->align_mask;
1624 	if (length < info->length)
1625 		return -ENOMEM;
1626 
1627 	low_limit = info->low_limit;
1628 	if (low_limit < mmap_min_addr)
1629 		low_limit = mmap_min_addr;
1630 	high_limit = info->high_limit;
1631 retry:
1632 	if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length))
1633 		return -ENOMEM;
1634 
1635 	gap = mas.last + 1 - info->length;
1636 	gap -= (gap - info->align_offset) & info->align_mask;
1637 	gap_end = mas.last;
1638 	tmp = mas_next(&mas, ULONG_MAX);
1639 	if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */
1640 		if (vm_start_gap(tmp) <= gap_end) {
1641 			high_limit = vm_start_gap(tmp);
1642 			mas_reset(&mas);
1643 			goto retry;
1644 		}
1645 	} else {
1646 		tmp = mas_prev(&mas, 0);
1647 		if (tmp && vm_end_gap(tmp) > gap) {
1648 			high_limit = tmp->vm_start;
1649 			mas_reset(&mas);
1650 			goto retry;
1651 		}
1652 	}
1653 
1654 	return gap;
1655 }
1656 
1657 /*
1658  * Search for an unmapped address range.
1659  *
1660  * We are looking for a range that:
1661  * - does not intersect with any VMA;
1662  * - is contained within the [low_limit, high_limit) interval;
1663  * - is at least the desired size.
1664  * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
1665  */
vm_unmapped_area(struct vm_unmapped_area_info * info)1666 unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info)
1667 {
1668 	unsigned long addr;
1669 
1670 	if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
1671 		addr = unmapped_area_topdown(info);
1672 	else
1673 		addr = unmapped_area(info);
1674 
1675 	trace_vm_unmapped_area(addr, info);
1676 	return addr;
1677 }
1678 
1679 /* Get an address range which is currently unmapped.
1680  * For shmat() with addr=0.
1681  *
1682  * Ugly calling convention alert:
1683  * Return value with the low bits set means error value,
1684  * ie
1685  *	if (ret & ~PAGE_MASK)
1686  *		error = ret;
1687  *
1688  * This function "knows" that -ENOMEM has the bits set.
1689  */
1690 unsigned long
generic_get_unmapped_area(struct file * filp,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)1691 generic_get_unmapped_area(struct file *filp, unsigned long addr,
1692 			  unsigned long len, unsigned long pgoff,
1693 			  unsigned long flags)
1694 {
1695 	struct mm_struct *mm = current->mm;
1696 	struct vm_area_struct *vma, *prev;
1697 	struct vm_unmapped_area_info info;
1698 	const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
1699 
1700 	if (len > mmap_end - mmap_min_addr)
1701 		return -ENOMEM;
1702 
1703 	if (flags & MAP_FIXED)
1704 		return addr;
1705 
1706 	if (addr) {
1707 		addr = PAGE_ALIGN(addr);
1708 		vma = find_vma_prev(mm, addr, &prev);
1709 		if (mmap_end - len >= addr && addr >= mmap_min_addr &&
1710 		    (!vma || addr + len <= vm_start_gap(vma)) &&
1711 		    (!prev || addr >= vm_end_gap(prev)))
1712 			return addr;
1713 	}
1714 
1715 	info.flags = 0;
1716 	info.length = len;
1717 	info.low_limit = mm->mmap_base;
1718 	info.high_limit = mmap_end;
1719 	info.align_mask = 0;
1720 	info.align_offset = 0;
1721 	return vm_unmapped_area(&info);
1722 }
1723 
1724 #ifndef HAVE_ARCH_UNMAPPED_AREA
1725 unsigned long
arch_get_unmapped_area(struct file * filp,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)1726 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1727 		       unsigned long len, unsigned long pgoff,
1728 		       unsigned long flags)
1729 {
1730 	return generic_get_unmapped_area(filp, addr, len, pgoff, flags);
1731 }
1732 #endif
1733 
1734 /*
1735  * This mmap-allocator allocates new areas top-down from below the
1736  * stack's low limit (the base):
1737  */
1738 unsigned long
generic_get_unmapped_area_topdown(struct file * filp,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)1739 generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
1740 				  unsigned long len, unsigned long pgoff,
1741 				  unsigned long flags)
1742 {
1743 	struct vm_area_struct *vma, *prev;
1744 	struct mm_struct *mm = current->mm;
1745 	struct vm_unmapped_area_info info;
1746 	const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags);
1747 
1748 	/* requested length too big for entire address space */
1749 	if (len > mmap_end - mmap_min_addr)
1750 		return -ENOMEM;
1751 
1752 	if (flags & MAP_FIXED)
1753 		return addr;
1754 
1755 	/* requesting a specific address */
1756 	if (addr) {
1757 		addr = PAGE_ALIGN(addr);
1758 		vma = find_vma_prev(mm, addr, &prev);
1759 		if (mmap_end - len >= addr && addr >= mmap_min_addr &&
1760 				(!vma || addr + len <= vm_start_gap(vma)) &&
1761 				(!prev || addr >= vm_end_gap(prev)))
1762 			return addr;
1763 	}
1764 
1765 	info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1766 	info.length = len;
1767 	info.low_limit = PAGE_SIZE;
1768 	info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
1769 	info.align_mask = 0;
1770 	info.align_offset = 0;
1771 	addr = vm_unmapped_area(&info);
1772 
1773 	/*
1774 	 * A failed mmap() very likely causes application failure,
1775 	 * so fall back to the bottom-up function here. This scenario
1776 	 * can happen with large stack limits and large mmap()
1777 	 * allocations.
1778 	 */
1779 	if (offset_in_page(addr)) {
1780 		VM_BUG_ON(addr != -ENOMEM);
1781 		info.flags = 0;
1782 		info.low_limit = TASK_UNMAPPED_BASE;
1783 		info.high_limit = mmap_end;
1784 		addr = vm_unmapped_area(&info);
1785 	}
1786 
1787 	return addr;
1788 }
1789 
1790 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1791 unsigned long
arch_get_unmapped_area_topdown(struct file * filp,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)1792 arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
1793 			       unsigned long len, unsigned long pgoff,
1794 			       unsigned long flags)
1795 {
1796 	return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags);
1797 }
1798 #endif
1799 
1800 unsigned long
get_unmapped_area(struct file * file,unsigned long addr,unsigned long len,unsigned long pgoff,unsigned long flags)1801 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1802 		unsigned long pgoff, unsigned long flags)
1803 {
1804 	unsigned long (*get_area)(struct file *, unsigned long,
1805 				  unsigned long, unsigned long, unsigned long);
1806 
1807 	unsigned long error = arch_mmap_check(addr, len, flags);
1808 	if (error)
1809 		return error;
1810 
1811 	/* Careful about overflows.. */
1812 	if (len > TASK_SIZE)
1813 		return -ENOMEM;
1814 
1815 	get_area = current->mm->get_unmapped_area;
1816 	if (file) {
1817 		if (file->f_op->get_unmapped_area)
1818 			get_area = file->f_op->get_unmapped_area;
1819 	} else if (flags & MAP_SHARED) {
1820 		/*
1821 		 * mmap_region() will call shmem_zero_setup() to create a file,
1822 		 * so use shmem's get_unmapped_area in case it can be huge.
1823 		 * do_mmap() will clear pgoff, so match alignment.
1824 		 */
1825 		pgoff = 0;
1826 		get_area = shmem_get_unmapped_area;
1827 	}
1828 
1829 	addr = get_area(file, addr, len, pgoff, flags);
1830 	if (IS_ERR_VALUE(addr))
1831 		return addr;
1832 
1833 	if (addr > TASK_SIZE - len)
1834 		return -ENOMEM;
1835 	if (offset_in_page(addr))
1836 		return -EINVAL;
1837 
1838 	error = security_mmap_addr(addr);
1839 	return error ? error : addr;
1840 }
1841 
1842 EXPORT_SYMBOL(get_unmapped_area);
1843 
1844 /**
1845  * find_vma_intersection() - Look up the first VMA which intersects the interval
1846  * @mm: The process address space.
1847  * @start_addr: The inclusive start user address.
1848  * @end_addr: The exclusive end user address.
1849  *
1850  * Returns: The first VMA within the provided range, %NULL otherwise.  Assumes
1851  * start_addr < end_addr.
1852  */
find_vma_intersection(struct mm_struct * mm,unsigned long start_addr,unsigned long end_addr)1853 struct vm_area_struct *find_vma_intersection(struct mm_struct *mm,
1854 					     unsigned long start_addr,
1855 					     unsigned long end_addr)
1856 {
1857 	unsigned long index = start_addr;
1858 
1859 	mmap_assert_locked(mm);
1860 	return mt_find(&mm->mm_mt, &index, end_addr - 1);
1861 }
1862 EXPORT_SYMBOL(find_vma_intersection);
1863 
1864 /**
1865  * find_vma() - Find the VMA for a given address, or the next VMA.
1866  * @mm: The mm_struct to check
1867  * @addr: The address
1868  *
1869  * Returns: The VMA associated with addr, or the next VMA.
1870  * May return %NULL in the case of no VMA at addr or above.
1871  */
find_vma(struct mm_struct * mm,unsigned long addr)1872 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1873 {
1874 	unsigned long index = addr;
1875 
1876 	mmap_assert_locked(mm);
1877 	return mt_find(&mm->mm_mt, &index, ULONG_MAX);
1878 }
1879 EXPORT_SYMBOL(find_vma);
1880 
1881 /**
1882  * find_vma_prev() - Find the VMA for a given address, or the next vma and
1883  * set %pprev to the previous VMA, if any.
1884  * @mm: The mm_struct to check
1885  * @addr: The address
1886  * @pprev: The pointer to set to the previous VMA
1887  *
1888  * Note that RCU lock is missing here since the external mmap_lock() is used
1889  * instead.
1890  *
1891  * Returns: The VMA associated with @addr, or the next vma.
1892  * May return %NULL in the case of no vma at addr or above.
1893  */
1894 struct vm_area_struct *
find_vma_prev(struct mm_struct * mm,unsigned long addr,struct vm_area_struct ** pprev)1895 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1896 			struct vm_area_struct **pprev)
1897 {
1898 	struct vm_area_struct *vma;
1899 	MA_STATE(mas, &mm->mm_mt, addr, addr);
1900 
1901 	vma = mas_walk(&mas);
1902 	*pprev = mas_prev(&mas, 0);
1903 	if (!vma)
1904 		vma = mas_next(&mas, ULONG_MAX);
1905 	return vma;
1906 }
1907 
1908 /*
1909  * Verify that the stack growth is acceptable and
1910  * update accounting. This is shared with both the
1911  * grow-up and grow-down cases.
1912  */
acct_stack_growth(struct vm_area_struct * vma,unsigned long size,unsigned long grow)1913 static int acct_stack_growth(struct vm_area_struct *vma,
1914 			     unsigned long size, unsigned long grow)
1915 {
1916 	struct mm_struct *mm = vma->vm_mm;
1917 	unsigned long new_start;
1918 
1919 	/* address space limit tests */
1920 	if (!may_expand_vm(mm, vma->vm_flags, grow))
1921 		return -ENOMEM;
1922 
1923 	/* Stack limit test */
1924 	if (size > rlimit(RLIMIT_STACK))
1925 		return -ENOMEM;
1926 
1927 	/* mlock limit tests */
1928 	if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT))
1929 		return -ENOMEM;
1930 
1931 	/* Check to ensure the stack will not grow into a hugetlb-only region */
1932 	new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
1933 			vma->vm_end - size;
1934 	if (is_hugepage_only_range(vma->vm_mm, new_start, size))
1935 		return -EFAULT;
1936 
1937 	/*
1938 	 * Overcommit..  This must be the final test, as it will
1939 	 * update security statistics.
1940 	 */
1941 	if (security_vm_enough_memory_mm(mm, grow))
1942 		return -ENOMEM;
1943 
1944 	return 0;
1945 }
1946 
1947 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
1948 /*
1949  * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
1950  * vma is the last one with address > vma->vm_end.  Have to extend vma.
1951  */
expand_upwards(struct vm_area_struct * vma,unsigned long address)1952 static int expand_upwards(struct vm_area_struct *vma, unsigned long address)
1953 {
1954 	struct mm_struct *mm = vma->vm_mm;
1955 	struct vm_area_struct *next;
1956 	unsigned long gap_addr;
1957 	int error = 0;
1958 	MA_STATE(mas, &mm->mm_mt, vma->vm_start, address);
1959 
1960 	if (!(vma->vm_flags & VM_GROWSUP))
1961 		return -EFAULT;
1962 
1963 	/* Guard against exceeding limits of the address space. */
1964 	address &= PAGE_MASK;
1965 	if (address >= (TASK_SIZE & PAGE_MASK))
1966 		return -ENOMEM;
1967 	address += PAGE_SIZE;
1968 
1969 	/* Enforce stack_guard_gap */
1970 	gap_addr = address + stack_guard_gap;
1971 
1972 	/* Guard against overflow */
1973 	if (gap_addr < address || gap_addr > TASK_SIZE)
1974 		gap_addr = TASK_SIZE;
1975 
1976 	next = find_vma_intersection(mm, vma->vm_end, gap_addr);
1977 	if (next && vma_is_accessible(next)) {
1978 		if (!(next->vm_flags & VM_GROWSUP))
1979 			return -ENOMEM;
1980 		/* Check that both stack segments have the same anon_vma? */
1981 	}
1982 
1983 	if (next)
1984 		mas_prev_range(&mas, address);
1985 
1986 	__mas_set_range(&mas, vma->vm_start, address - 1);
1987 	if (mas_preallocate(&mas, vma, GFP_KERNEL))
1988 		return -ENOMEM;
1989 
1990 	/* We must make sure the anon_vma is allocated. */
1991 	if (unlikely(anon_vma_prepare(vma))) {
1992 		mas_destroy(&mas);
1993 		return -ENOMEM;
1994 	}
1995 
1996 	/* Lock the VMA before expanding to prevent concurrent page faults */
1997 	vma_start_write(vma);
1998 	/*
1999 	 * vma->vm_start/vm_end cannot change under us because the caller
2000 	 * is required to hold the mmap_lock in read mode.  We need the
2001 	 * anon_vma lock to serialize against concurrent expand_stacks.
2002 	 */
2003 	anon_vma_lock_write(vma->anon_vma);
2004 
2005 	/* Somebody else might have raced and expanded it already */
2006 	if (address > vma->vm_end) {
2007 		unsigned long size, grow;
2008 
2009 		size = address - vma->vm_start;
2010 		grow = (address - vma->vm_end) >> PAGE_SHIFT;
2011 
2012 		error = -ENOMEM;
2013 		if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2014 			error = acct_stack_growth(vma, size, grow);
2015 			if (!error) {
2016 				/*
2017 				 * We only hold a shared mmap_lock lock here, so
2018 				 * we need to protect against concurrent vma
2019 				 * expansions.  anon_vma_lock_write() doesn't
2020 				 * help here, as we don't guarantee that all
2021 				 * growable vmas in a mm share the same root
2022 				 * anon vma.  So, we reuse mm->page_table_lock
2023 				 * to guard against concurrent vma expansions.
2024 				 */
2025 				spin_lock(&mm->page_table_lock);
2026 				if (vma->vm_flags & VM_LOCKED)
2027 					mm->locked_vm += grow;
2028 				vm_stat_account(mm, vma->vm_flags, grow);
2029 				anon_vma_interval_tree_pre_update_vma(vma);
2030 				vma->vm_end = address;
2031 				/* Overwrite old entry in mtree. */
2032 				mas_store_prealloc(&mas, vma);
2033 				anon_vma_interval_tree_post_update_vma(vma);
2034 				spin_unlock(&mm->page_table_lock);
2035 
2036 				perf_event_mmap(vma);
2037 			}
2038 		}
2039 	}
2040 	anon_vma_unlock_write(vma->anon_vma);
2041 	khugepaged_enter_vma(vma, vma->vm_flags);
2042 	mas_destroy(&mas);
2043 	validate_mm(mm);
2044 	return error;
2045 }
2046 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2047 
2048 /*
2049  * vma is the first one with address < vma->vm_start.  Have to extend vma.
2050  * mmap_lock held for writing.
2051  */
expand_downwards(struct vm_area_struct * vma,unsigned long address)2052 int expand_downwards(struct vm_area_struct *vma, unsigned long address)
2053 {
2054 	struct mm_struct *mm = vma->vm_mm;
2055 	MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start);
2056 	struct vm_area_struct *prev;
2057 	int error = 0;
2058 
2059 	if (!(vma->vm_flags & VM_GROWSDOWN))
2060 		return -EFAULT;
2061 
2062 	address &= PAGE_MASK;
2063 	if (address < mmap_min_addr || address < FIRST_USER_ADDRESS)
2064 		return -EPERM;
2065 
2066 	/* Enforce stack_guard_gap */
2067 	prev = mas_prev(&mas, 0);
2068 	/* Check that both stack segments have the same anon_vma? */
2069 	if (prev) {
2070 		if (!(prev->vm_flags & VM_GROWSDOWN) &&
2071 		    vma_is_accessible(prev) &&
2072 		    (address - prev->vm_end < stack_guard_gap))
2073 			return -ENOMEM;
2074 	}
2075 
2076 	if (prev)
2077 		mas_next_range(&mas, vma->vm_start);
2078 
2079 	__mas_set_range(&mas, address, vma->vm_end - 1);
2080 	if (mas_preallocate(&mas, vma, GFP_KERNEL))
2081 		return -ENOMEM;
2082 
2083 	/* We must make sure the anon_vma is allocated. */
2084 	if (unlikely(anon_vma_prepare(vma))) {
2085 		mas_destroy(&mas);
2086 		return -ENOMEM;
2087 	}
2088 
2089 	/* Lock the VMA before expanding to prevent concurrent page faults */
2090 	vma_start_write(vma);
2091 	/*
2092 	 * vma->vm_start/vm_end cannot change under us because the caller
2093 	 * is required to hold the mmap_lock in read mode.  We need the
2094 	 * anon_vma lock to serialize against concurrent expand_stacks.
2095 	 */
2096 	anon_vma_lock_write(vma->anon_vma);
2097 
2098 	/* Somebody else might have raced and expanded it already */
2099 	if (address < vma->vm_start) {
2100 		unsigned long size, grow;
2101 
2102 		size = vma->vm_end - address;
2103 		grow = (vma->vm_start - address) >> PAGE_SHIFT;
2104 
2105 		error = -ENOMEM;
2106 		if (grow <= vma->vm_pgoff) {
2107 			error = acct_stack_growth(vma, size, grow);
2108 			if (!error) {
2109 				/*
2110 				 * We only hold a shared mmap_lock lock here, so
2111 				 * we need to protect against concurrent vma
2112 				 * expansions.  anon_vma_lock_write() doesn't
2113 				 * help here, as we don't guarantee that all
2114 				 * growable vmas in a mm share the same root
2115 				 * anon vma.  So, we reuse mm->page_table_lock
2116 				 * to guard against concurrent vma expansions.
2117 				 */
2118 				spin_lock(&mm->page_table_lock);
2119 				if (vma->vm_flags & VM_LOCKED)
2120 					mm->locked_vm += grow;
2121 				vm_stat_account(mm, vma->vm_flags, grow);
2122 				anon_vma_interval_tree_pre_update_vma(vma);
2123 				vma->vm_start = address;
2124 				vma->vm_pgoff -= grow;
2125 				/* Overwrite old entry in mtree. */
2126 				mas_store_prealloc(&mas, vma);
2127 				anon_vma_interval_tree_post_update_vma(vma);
2128 				spin_unlock(&mm->page_table_lock);
2129 
2130 				perf_event_mmap(vma);
2131 			}
2132 		}
2133 	}
2134 	anon_vma_unlock_write(vma->anon_vma);
2135 	khugepaged_enter_vma(vma, vma->vm_flags);
2136 	mas_destroy(&mas);
2137 	validate_mm(mm);
2138 	return error;
2139 }
2140 
2141 /* enforced gap between the expanding stack and other mappings. */
2142 unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2143 
cmdline_parse_stack_guard_gap(char * p)2144 static int __init cmdline_parse_stack_guard_gap(char *p)
2145 {
2146 	unsigned long val;
2147 	char *endptr;
2148 
2149 	val = simple_strtoul(p, &endptr, 10);
2150 	if (!*endptr)
2151 		stack_guard_gap = val << PAGE_SHIFT;
2152 
2153 	return 1;
2154 }
2155 __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2156 
2157 #ifdef CONFIG_STACK_GROWSUP
expand_stack_locked(struct vm_area_struct * vma,unsigned long address)2158 int expand_stack_locked(struct vm_area_struct *vma, unsigned long address)
2159 {
2160 	return expand_upwards(vma, address);
2161 }
2162 
find_extend_vma_locked(struct mm_struct * mm,unsigned long addr)2163 struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr)
2164 {
2165 	struct vm_area_struct *vma, *prev;
2166 
2167 	addr &= PAGE_MASK;
2168 	vma = find_vma_prev(mm, addr, &prev);
2169 	if (vma && (vma->vm_start <= addr))
2170 		return vma;
2171 	if (!prev)
2172 		return NULL;
2173 	if (expand_stack_locked(prev, addr))
2174 		return NULL;
2175 	if (prev->vm_flags & VM_LOCKED)
2176 		populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2177 	return prev;
2178 }
2179 #else
expand_stack_locked(struct vm_area_struct * vma,unsigned long address)2180 int expand_stack_locked(struct vm_area_struct *vma, unsigned long address)
2181 {
2182 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
2183 		return -EINVAL;
2184 	return expand_downwards(vma, address);
2185 }
2186 
find_extend_vma_locked(struct mm_struct * mm,unsigned long addr)2187 struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr)
2188 {
2189 	struct vm_area_struct *vma;
2190 	unsigned long start;
2191 
2192 	addr &= PAGE_MASK;
2193 	vma = find_vma(mm, addr);
2194 	if (!vma)
2195 		return NULL;
2196 	if (vma->vm_start <= addr)
2197 		return vma;
2198 	start = vma->vm_start;
2199 	if (expand_stack_locked(vma, addr))
2200 		return NULL;
2201 	if (vma->vm_flags & VM_LOCKED)
2202 		populate_vma_page_range(vma, addr, start, NULL);
2203 	return vma;
2204 }
2205 #endif
2206 
2207 /*
2208  * IA64 has some horrid mapping rules: it can expand both up and down,
2209  * but with various special rules.
2210  *
2211  * We'll get rid of this architecture eventually, so the ugliness is
2212  * temporary.
2213  */
2214 #ifdef CONFIG_IA64
vma_expand_ok(struct vm_area_struct * vma,unsigned long addr)2215 static inline bool vma_expand_ok(struct vm_area_struct *vma, unsigned long addr)
2216 {
2217 	return REGION_NUMBER(addr) == REGION_NUMBER(vma->vm_start) &&
2218 		REGION_OFFSET(addr) < RGN_MAP_LIMIT;
2219 }
2220 
2221 /*
2222  * IA64 stacks grow down, but there's a special register backing store
2223  * that can grow up. Only sequentially, though, so the new address must
2224  * match vm_end.
2225  */
vma_expand_up(struct vm_area_struct * vma,unsigned long addr)2226 static inline int vma_expand_up(struct vm_area_struct *vma, unsigned long addr)
2227 {
2228 	if (!vma_expand_ok(vma, addr))
2229 		return -EFAULT;
2230 	if (vma->vm_end != (addr & PAGE_MASK))
2231 		return -EFAULT;
2232 	return expand_upwards(vma, addr);
2233 }
2234 
vma_expand_down(struct vm_area_struct * vma,unsigned long addr)2235 static inline bool vma_expand_down(struct vm_area_struct *vma, unsigned long addr)
2236 {
2237 	if (!vma_expand_ok(vma, addr))
2238 		return -EFAULT;
2239 	return expand_downwards(vma, addr);
2240 }
2241 
2242 #elif defined(CONFIG_STACK_GROWSUP)
2243 
2244 #define vma_expand_up(vma,addr) expand_upwards(vma, addr)
2245 #define vma_expand_down(vma, addr) (-EFAULT)
2246 
2247 #else
2248 
2249 #define vma_expand_up(vma,addr) (-EFAULT)
2250 #define vma_expand_down(vma, addr) expand_downwards(vma, addr)
2251 
2252 #endif
2253 
2254 /*
2255  * expand_stack(): legacy interface for page faulting. Don't use unless
2256  * you have to.
2257  *
2258  * This is called with the mm locked for reading, drops the lock, takes
2259  * the lock for writing, tries to look up a vma again, expands it if
2260  * necessary, and downgrades the lock to reading again.
2261  *
2262  * If no vma is found or it can't be expanded, it returns NULL and has
2263  * dropped the lock.
2264  */
expand_stack(struct mm_struct * mm,unsigned long addr)2265 struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr)
2266 {
2267 	struct vm_area_struct *vma, *prev;
2268 
2269 	mmap_read_unlock(mm);
2270 	if (mmap_write_lock_killable(mm))
2271 		return NULL;
2272 
2273 	vma = find_vma_prev(mm, addr, &prev);
2274 	if (vma && vma->vm_start <= addr)
2275 		goto success;
2276 
2277 	if (prev && !vma_expand_up(prev, addr)) {
2278 		vma = prev;
2279 		goto success;
2280 	}
2281 
2282 	if (vma && !vma_expand_down(vma, addr))
2283 		goto success;
2284 
2285 	mmap_write_unlock(mm);
2286 	return NULL;
2287 
2288 success:
2289 	mmap_write_downgrade(mm);
2290 	return vma;
2291 }
2292 
2293 /*
2294  * Ok - we have the memory areas we should free on a maple tree so release them,
2295  * and do the vma updates.
2296  *
2297  * Called with the mm semaphore held.
2298  */
remove_mt(struct mm_struct * mm,struct ma_state * mas)2299 static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas)
2300 {
2301 	unsigned long nr_accounted = 0;
2302 	struct vm_area_struct *vma;
2303 
2304 	/* Update high watermark before we lower total_vm */
2305 	update_hiwater_vm(mm);
2306 	mas_for_each(mas, vma, ULONG_MAX) {
2307 		long nrpages = vma_pages(vma);
2308 
2309 		if (vma->vm_flags & VM_ACCOUNT)
2310 			nr_accounted += nrpages;
2311 		vm_stat_account(mm, vma->vm_flags, -nrpages);
2312 		remove_vma(vma, false);
2313 	}
2314 	vm_unacct_memory(nr_accounted);
2315 }
2316 
2317 /*
2318  * Get rid of page table information in the indicated region.
2319  *
2320  * Called with the mm semaphore held.
2321  */
unmap_region(struct mm_struct * mm,struct ma_state * mas,struct vm_area_struct * vma,struct vm_area_struct * prev,struct vm_area_struct * next,unsigned long start,unsigned long end,unsigned long tree_end,bool mm_wr_locked)2322 static void unmap_region(struct mm_struct *mm, struct ma_state *mas,
2323 		struct vm_area_struct *vma, struct vm_area_struct *prev,
2324 		struct vm_area_struct *next, unsigned long start,
2325 		unsigned long end, unsigned long tree_end, bool mm_wr_locked)
2326 {
2327 	struct mmu_gather tlb;
2328 	unsigned long mt_start = mas->index;
2329 
2330 	lru_add_drain();
2331 	tlb_gather_mmu(&tlb, mm);
2332 	update_hiwater_rss(mm);
2333 	unmap_vmas(&tlb, mas, vma, start, end, tree_end, mm_wr_locked);
2334 	mas_set(mas, mt_start);
2335 	free_pgtables(&tlb, mas, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2336 				 next ? next->vm_start : USER_PGTABLES_CEILING,
2337 				 mm_wr_locked);
2338 	tlb_finish_mmu(&tlb);
2339 }
2340 
2341 /*
2342  * __split_vma() bypasses sysctl_max_map_count checking.  We use this where it
2343  * has already been checked or doesn't make sense to fail.
2344  * VMA Iterator will point to the end VMA.
2345  */
__split_vma(struct vma_iterator * vmi,struct vm_area_struct * vma,unsigned long addr,int new_below)2346 int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
2347 		unsigned long addr, int new_below)
2348 {
2349 	struct vma_prepare vp;
2350 	struct vm_area_struct *new;
2351 	int err;
2352 
2353 	WARN_ON(vma->vm_start >= addr);
2354 	WARN_ON(vma->vm_end <= addr);
2355 
2356 	if (vma->vm_ops && vma->vm_ops->may_split) {
2357 		err = vma->vm_ops->may_split(vma, addr);
2358 		if (err)
2359 			return err;
2360 	}
2361 
2362 	new = vm_area_dup(vma);
2363 	if (!new)
2364 		return -ENOMEM;
2365 
2366 	if (new_below) {
2367 		new->vm_end = addr;
2368 	} else {
2369 		new->vm_start = addr;
2370 		new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2371 	}
2372 
2373 	err = -ENOMEM;
2374 	vma_iter_config(vmi, new->vm_start, new->vm_end);
2375 	if (vma_iter_prealloc(vmi, new))
2376 		goto out_free_vma;
2377 
2378 	err = vma_dup_policy(vma, new);
2379 	if (err)
2380 		goto out_free_vmi;
2381 
2382 	err = anon_vma_clone(new, vma);
2383 	if (err)
2384 		goto out_free_mpol;
2385 
2386 	if (new->vm_file)
2387 		get_file(new->vm_file);
2388 
2389 	if (new->vm_ops && new->vm_ops->open)
2390 		new->vm_ops->open(new);
2391 
2392 	vma_start_write(vma);
2393 	vma_start_write(new);
2394 
2395 	init_vma_prep(&vp, vma);
2396 	vp.insert = new;
2397 	vma_prepare(&vp);
2398 	vma_adjust_trans_huge(vma, vma->vm_start, addr, 0);
2399 
2400 	if (new_below) {
2401 		vma->vm_start = addr;
2402 		vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT;
2403 	} else {
2404 		vma->vm_end = addr;
2405 	}
2406 
2407 	/* vma_complete stores the new vma */
2408 	vma_complete(&vp, vmi, vma->vm_mm);
2409 
2410 	/* Success. */
2411 	if (new_below)
2412 		vma_next(vmi);
2413 	return 0;
2414 
2415 out_free_mpol:
2416 	mpol_put(vma_policy(new));
2417 out_free_vmi:
2418 	vma_iter_free(vmi);
2419 out_free_vma:
2420 	vm_area_free(new);
2421 	return err;
2422 }
2423 
2424 /*
2425  * Split a vma into two pieces at address 'addr', a new vma is allocated
2426  * either for the first part or the tail.
2427  */
split_vma(struct vma_iterator * vmi,struct vm_area_struct * vma,unsigned long addr,int new_below)2428 int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
2429 	      unsigned long addr, int new_below)
2430 {
2431 	if (vma->vm_mm->map_count >= sysctl_max_map_count)
2432 		return -ENOMEM;
2433 
2434 	return __split_vma(vmi, vma, addr, new_below);
2435 }
2436 
2437 /*
2438  * do_vmi_align_munmap() - munmap the aligned region from @start to @end.
2439  * @vmi: The vma iterator
2440  * @vma: The starting vm_area_struct
2441  * @mm: The mm_struct
2442  * @start: The aligned start address to munmap.
2443  * @end: The aligned end address to munmap.
2444  * @uf: The userfaultfd list_head
2445  * @unlock: Set to true to drop the mmap_lock.  unlocking only happens on
2446  * success.
2447  *
2448  * Return: 0 on success and drops the lock if so directed, error and leaves the
2449  * lock held otherwise.
2450  */
2451 static int
do_vmi_align_munmap(struct vma_iterator * vmi,struct vm_area_struct * vma,struct mm_struct * mm,unsigned long start,unsigned long end,struct list_head * uf,bool unlock)2452 do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
2453 		    struct mm_struct *mm, unsigned long start,
2454 		    unsigned long end, struct list_head *uf, bool unlock)
2455 {
2456 	struct vm_area_struct *prev, *next = NULL;
2457 	struct maple_tree mt_detach;
2458 	int count = 0;
2459 	int error = -ENOMEM;
2460 	unsigned long locked_vm = 0;
2461 	MA_STATE(mas_detach, &mt_detach, 0, 0);
2462 	mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK);
2463 	mt_on_stack(mt_detach);
2464 
2465 	/*
2466 	 * If we need to split any vma, do it now to save pain later.
2467 	 *
2468 	 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2469 	 * unmapped vm_area_struct will remain in use: so lower split_vma
2470 	 * places tmp vma above, and higher split_vma places tmp vma below.
2471 	 */
2472 
2473 	/* Does it split the first one? */
2474 	if (start > vma->vm_start) {
2475 
2476 		/*
2477 		 * Make sure that map_count on return from munmap() will
2478 		 * not exceed its limit; but let map_count go just above
2479 		 * its limit temporarily, to help free resources as expected.
2480 		 */
2481 		if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2482 			goto map_count_exceeded;
2483 
2484 		error = __split_vma(vmi, vma, start, 1);
2485 		if (error)
2486 			goto start_split_failed;
2487 	}
2488 
2489 	/*
2490 	 * Detach a range of VMAs from the mm. Using next as a temp variable as
2491 	 * it is always overwritten.
2492 	 */
2493 	next = vma;
2494 	do {
2495 		/* Does it split the end? */
2496 		if (next->vm_end > end) {
2497 			error = __split_vma(vmi, next, end, 0);
2498 			if (error)
2499 				goto end_split_failed;
2500 		}
2501 		vma_start_write(next);
2502 		mas_set(&mas_detach, count);
2503 		error = mas_store_gfp(&mas_detach, next, GFP_KERNEL);
2504 		if (error)
2505 			goto munmap_gather_failed;
2506 		vma_mark_detached(next, true);
2507 		if (next->vm_flags & VM_LOCKED)
2508 			locked_vm += vma_pages(next);
2509 
2510 		count++;
2511 		if (unlikely(uf)) {
2512 			/*
2513 			 * If userfaultfd_unmap_prep returns an error the vmas
2514 			 * will remain split, but userland will get a
2515 			 * highly unexpected error anyway. This is no
2516 			 * different than the case where the first of the two
2517 			 * __split_vma fails, but we don't undo the first
2518 			 * split, despite we could. This is unlikely enough
2519 			 * failure that it's not worth optimizing it for.
2520 			 */
2521 			error = userfaultfd_unmap_prep(next, start, end, uf);
2522 
2523 			if (error)
2524 				goto userfaultfd_error;
2525 		}
2526 #ifdef CONFIG_DEBUG_VM_MAPLE_TREE
2527 		BUG_ON(next->vm_start < start);
2528 		BUG_ON(next->vm_start > end);
2529 #endif
2530 	} for_each_vma_range(*vmi, next, end);
2531 
2532 #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
2533 	/* Make sure no VMAs are about to be lost. */
2534 	{
2535 		MA_STATE(test, &mt_detach, 0, 0);
2536 		struct vm_area_struct *vma_mas, *vma_test;
2537 		int test_count = 0;
2538 
2539 		vma_iter_set(vmi, start);
2540 		rcu_read_lock();
2541 		vma_test = mas_find(&test, count - 1);
2542 		for_each_vma_range(*vmi, vma_mas, end) {
2543 			BUG_ON(vma_mas != vma_test);
2544 			test_count++;
2545 			vma_test = mas_next(&test, count - 1);
2546 		}
2547 		rcu_read_unlock();
2548 		BUG_ON(count != test_count);
2549 	}
2550 #endif
2551 
2552 	while (vma_iter_addr(vmi) > start)
2553 		vma_iter_prev_range(vmi);
2554 
2555 	error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL);
2556 	if (error)
2557 		goto clear_tree_failed;
2558 
2559 	/* Point of no return */
2560 	mm->locked_vm -= locked_vm;
2561 	mm->map_count -= count;
2562 	if (unlock)
2563 		mmap_write_downgrade(mm);
2564 
2565 	prev = vma_iter_prev_range(vmi);
2566 	next = vma_next(vmi);
2567 	if (next)
2568 		vma_iter_prev_range(vmi);
2569 
2570 	/*
2571 	 * We can free page tables without write-locking mmap_lock because VMAs
2572 	 * were isolated before we downgraded mmap_lock.
2573 	 */
2574 	mas_set(&mas_detach, 1);
2575 	unmap_region(mm, &mas_detach, vma, prev, next, start, end, count,
2576 		     !unlock);
2577 	/* Statistics and freeing VMAs */
2578 	mas_set(&mas_detach, 0);
2579 	remove_mt(mm, &mas_detach);
2580 	validate_mm(mm);
2581 	if (unlock)
2582 		mmap_read_unlock(mm);
2583 
2584 	__mt_destroy(&mt_detach);
2585 	return 0;
2586 
2587 clear_tree_failed:
2588 userfaultfd_error:
2589 munmap_gather_failed:
2590 end_split_failed:
2591 	mas_set(&mas_detach, 0);
2592 	mas_for_each(&mas_detach, next, end)
2593 		vma_mark_detached(next, false);
2594 
2595 	__mt_destroy(&mt_detach);
2596 start_split_failed:
2597 map_count_exceeded:
2598 	validate_mm(mm);
2599 	return error;
2600 }
2601 
2602 /*
2603  * do_vmi_munmap() - munmap a given range.
2604  * @vmi: The vma iterator
2605  * @mm: The mm_struct
2606  * @start: The start address to munmap
2607  * @len: The length of the range to munmap
2608  * @uf: The userfaultfd list_head
2609  * @unlock: set to true if the user wants to drop the mmap_lock on success
2610  *
2611  * This function takes a @mas that is either pointing to the previous VMA or set
2612  * to MA_START and sets it up to remove the mapping(s).  The @len will be
2613  * aligned and any arch_unmap work will be preformed.
2614  *
2615  * Return: 0 on success and drops the lock if so directed, error and leaves the
2616  * lock held otherwise.
2617  */
do_vmi_munmap(struct vma_iterator * vmi,struct mm_struct * mm,unsigned long start,size_t len,struct list_head * uf,bool unlock)2618 int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm,
2619 		  unsigned long start, size_t len, struct list_head *uf,
2620 		  bool unlock)
2621 {
2622 	unsigned long end;
2623 	struct vm_area_struct *vma;
2624 
2625 	if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2626 		return -EINVAL;
2627 
2628 	end = start + PAGE_ALIGN(len);
2629 	if (end == start)
2630 		return -EINVAL;
2631 
2632 	 /* arch_unmap() might do unmaps itself.  */
2633 	arch_unmap(mm, start, end);
2634 
2635 	/* Find the first overlapping VMA */
2636 	vma = vma_find(vmi, end);
2637 	if (!vma) {
2638 		if (unlock)
2639 			mmap_write_unlock(mm);
2640 		return 0;
2641 	}
2642 
2643 	return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock);
2644 }
2645 
2646 /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls.
2647  * @mm: The mm_struct
2648  * @start: The start address to munmap
2649  * @len: The length to be munmapped.
2650  * @uf: The userfaultfd list_head
2651  *
2652  * Return: 0 on success, error otherwise.
2653  */
do_munmap(struct mm_struct * mm,unsigned long start,size_t len,struct list_head * uf)2654 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2655 	      struct list_head *uf)
2656 {
2657 	VMA_ITERATOR(vmi, mm, start);
2658 
2659 	return do_vmi_munmap(&vmi, mm, start, len, uf, false);
2660 }
2661 
mmap_region(struct file * file,unsigned long addr,unsigned long len,vm_flags_t vm_flags,unsigned long pgoff,struct list_head * uf)2662 unsigned long mmap_region(struct file *file, unsigned long addr,
2663 		unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
2664 		struct list_head *uf)
2665 {
2666 	struct mm_struct *mm = current->mm;
2667 	struct vm_area_struct *vma = NULL;
2668 	struct vm_area_struct *next, *prev, *merge;
2669 	pgoff_t pglen = len >> PAGE_SHIFT;
2670 	unsigned long charged = 0;
2671 	unsigned long end = addr + len;
2672 	unsigned long merge_start = addr, merge_end = end;
2673 	pgoff_t vm_pgoff;
2674 	int error;
2675 	VMA_ITERATOR(vmi, mm, addr);
2676 
2677 	/* Check against address space limit. */
2678 	if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
2679 		unsigned long nr_pages;
2680 
2681 		/*
2682 		 * MAP_FIXED may remove pages of mappings that intersects with
2683 		 * requested mapping. Account for the pages it would unmap.
2684 		 */
2685 		nr_pages = count_vma_pages_range(mm, addr, end);
2686 
2687 		if (!may_expand_vm(mm, vm_flags,
2688 					(len >> PAGE_SHIFT) - nr_pages))
2689 			return -ENOMEM;
2690 	}
2691 
2692 	/* Unmap any existing mapping in the area */
2693 	if (do_vmi_munmap(&vmi, mm, addr, len, uf, false))
2694 		return -ENOMEM;
2695 
2696 	/*
2697 	 * Private writable mapping: check memory availability
2698 	 */
2699 	if (accountable_mapping(file, vm_flags)) {
2700 		charged = len >> PAGE_SHIFT;
2701 		if (security_vm_enough_memory_mm(mm, charged))
2702 			return -ENOMEM;
2703 		vm_flags |= VM_ACCOUNT;
2704 	}
2705 
2706 	next = vma_next(&vmi);
2707 	prev = vma_prev(&vmi);
2708 	if (vm_flags & VM_SPECIAL) {
2709 		if (prev)
2710 			vma_iter_next_range(&vmi);
2711 		goto cannot_expand;
2712 	}
2713 
2714 	/* Attempt to expand an old mapping */
2715 	/* Check next */
2716 	if (next && next->vm_start == end && !vma_policy(next) &&
2717 	    can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen,
2718 				 NULL_VM_UFFD_CTX, NULL)) {
2719 		merge_end = next->vm_end;
2720 		vma = next;
2721 		vm_pgoff = next->vm_pgoff - pglen;
2722 	}
2723 
2724 	/* Check prev */
2725 	if (prev && prev->vm_end == addr && !vma_policy(prev) &&
2726 	    (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file,
2727 				       pgoff, vma->vm_userfaultfd_ctx, NULL) :
2728 		   can_vma_merge_after(prev, vm_flags, NULL, file, pgoff,
2729 				       NULL_VM_UFFD_CTX, NULL))) {
2730 		merge_start = prev->vm_start;
2731 		vma = prev;
2732 		vm_pgoff = prev->vm_pgoff;
2733 	} else if (prev) {
2734 		vma_iter_next_range(&vmi);
2735 	}
2736 
2737 	/* Actually expand, if possible */
2738 	if (vma &&
2739 	    !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) {
2740 		khugepaged_enter_vma(vma, vm_flags);
2741 		goto expanded;
2742 	}
2743 
2744 	if (vma == prev)
2745 		vma_iter_set(&vmi, addr);
2746 cannot_expand:
2747 
2748 	/*
2749 	 * Determine the object being mapped and call the appropriate
2750 	 * specific mapper. the address has already been validated, but
2751 	 * not unmapped, but the maps are removed from the list.
2752 	 */
2753 	vma = vm_area_alloc(mm);
2754 	if (!vma) {
2755 		error = -ENOMEM;
2756 		goto unacct_error;
2757 	}
2758 
2759 	vma_iter_config(&vmi, addr, end);
2760 	vma->vm_start = addr;
2761 	vma->vm_end = end;
2762 	vm_flags_init(vma, vm_flags);
2763 	vma->vm_page_prot = vm_get_page_prot(vm_flags);
2764 	vma->vm_pgoff = pgoff;
2765 
2766 	if (file) {
2767 		if (vm_flags & VM_SHARED) {
2768 			error = mapping_map_writable(file->f_mapping);
2769 			if (error)
2770 				goto free_vma;
2771 		}
2772 
2773 		vma->vm_file = get_file(file);
2774 		error = call_mmap(file, vma);
2775 		if (error)
2776 			goto unmap_and_free_vma;
2777 
2778 		/*
2779 		 * Expansion is handled above, merging is handled below.
2780 		 * Drivers should not alter the address of the VMA.
2781 		 */
2782 		error = -EINVAL;
2783 		if (WARN_ON((addr != vma->vm_start)))
2784 			goto close_and_free_vma;
2785 
2786 		vma_iter_config(&vmi, addr, end);
2787 		/*
2788 		 * If vm_flags changed after call_mmap(), we should try merge
2789 		 * vma again as we may succeed this time.
2790 		 */
2791 		if (unlikely(vm_flags != vma->vm_flags && prev)) {
2792 			merge = vma_merge(&vmi, mm, prev, vma->vm_start,
2793 				    vma->vm_end, vma->vm_flags, NULL,
2794 				    vma->vm_file, vma->vm_pgoff, NULL,
2795 				    NULL_VM_UFFD_CTX, NULL);
2796 			if (merge) {
2797 				/*
2798 				 * ->mmap() can change vma->vm_file and fput
2799 				 * the original file. So fput the vma->vm_file
2800 				 * here or we would add an extra fput for file
2801 				 * and cause general protection fault
2802 				 * ultimately.
2803 				 */
2804 				fput(vma->vm_file);
2805 				vm_area_free(vma);
2806 				vma = merge;
2807 				/* Update vm_flags to pick up the change. */
2808 				vm_flags = vma->vm_flags;
2809 				goto unmap_writable;
2810 			}
2811 		}
2812 
2813 		vm_flags = vma->vm_flags;
2814 	} else if (vm_flags & VM_SHARED) {
2815 		error = shmem_zero_setup(vma);
2816 		if (error)
2817 			goto free_vma;
2818 	} else {
2819 		vma_set_anonymous(vma);
2820 	}
2821 
2822 	if (map_deny_write_exec(vma, vma->vm_flags)) {
2823 		error = -EACCES;
2824 		goto close_and_free_vma;
2825 	}
2826 
2827 	/* Allow architectures to sanity-check the vm_flags */
2828 	error = -EINVAL;
2829 	if (!arch_validate_flags(vma->vm_flags))
2830 		goto close_and_free_vma;
2831 
2832 	error = -ENOMEM;
2833 	if (vma_iter_prealloc(&vmi, vma))
2834 		goto close_and_free_vma;
2835 
2836 	/* Lock the VMA since it is modified after insertion into VMA tree */
2837 	vma_start_write(vma);
2838 	vma_iter_store(&vmi, vma);
2839 	mm->map_count++;
2840 	if (vma->vm_file) {
2841 		i_mmap_lock_write(vma->vm_file->f_mapping);
2842 		if (vma->vm_flags & VM_SHARED)
2843 			mapping_allow_writable(vma->vm_file->f_mapping);
2844 
2845 		flush_dcache_mmap_lock(vma->vm_file->f_mapping);
2846 		vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap);
2847 		flush_dcache_mmap_unlock(vma->vm_file->f_mapping);
2848 		i_mmap_unlock_write(vma->vm_file->f_mapping);
2849 	}
2850 
2851 	/*
2852 	 * vma_merge() calls khugepaged_enter_vma() either, the below
2853 	 * call covers the non-merge case.
2854 	 */
2855 	khugepaged_enter_vma(vma, vma->vm_flags);
2856 
2857 	/* Once vma denies write, undo our temporary denial count */
2858 unmap_writable:
2859 	if (file && vm_flags & VM_SHARED)
2860 		mapping_unmap_writable(file->f_mapping);
2861 	file = vma->vm_file;
2862 	ksm_add_vma(vma);
2863 expanded:
2864 	perf_event_mmap(vma);
2865 
2866 	vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
2867 	if (vm_flags & VM_LOCKED) {
2868 		if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) ||
2869 					is_vm_hugetlb_page(vma) ||
2870 					vma == get_gate_vma(current->mm))
2871 			vm_flags_clear(vma, VM_LOCKED_MASK);
2872 		else
2873 			mm->locked_vm += (len >> PAGE_SHIFT);
2874 	}
2875 
2876 	if (file)
2877 		uprobe_mmap(vma);
2878 
2879 	/*
2880 	 * New (or expanded) vma always get soft dirty status.
2881 	 * Otherwise user-space soft-dirty page tracker won't
2882 	 * be able to distinguish situation when vma area unmapped,
2883 	 * then new mapped in-place (which must be aimed as
2884 	 * a completely new data area).
2885 	 */
2886 	vm_flags_set(vma, VM_SOFTDIRTY);
2887 
2888 	vma_set_page_prot(vma);
2889 
2890 	validate_mm(mm);
2891 	return addr;
2892 
2893 close_and_free_vma:
2894 	if (file && vma->vm_ops && vma->vm_ops->close)
2895 		vma->vm_ops->close(vma);
2896 
2897 	if (file || vma->vm_file) {
2898 unmap_and_free_vma:
2899 		fput(vma->vm_file);
2900 		vma->vm_file = NULL;
2901 
2902 		vma_iter_set(&vmi, vma->vm_end);
2903 		/* Undo any partial mapping done by a device driver. */
2904 		unmap_region(mm, &vmi.mas, vma, prev, next, vma->vm_start,
2905 			     vma->vm_end, vma->vm_end, true);
2906 	}
2907 	if (file && (vm_flags & VM_SHARED))
2908 		mapping_unmap_writable(file->f_mapping);
2909 free_vma:
2910 	vm_area_free(vma);
2911 unacct_error:
2912 	if (charged)
2913 		vm_unacct_memory(charged);
2914 	validate_mm(mm);
2915 	return error;
2916 }
2917 
__vm_munmap(unsigned long start,size_t len,bool unlock)2918 static int __vm_munmap(unsigned long start, size_t len, bool unlock)
2919 {
2920 	int ret;
2921 	struct mm_struct *mm = current->mm;
2922 	LIST_HEAD(uf);
2923 	VMA_ITERATOR(vmi, mm, start);
2924 
2925 	if (mmap_write_lock_killable(mm))
2926 		return -EINTR;
2927 
2928 	ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock);
2929 	if (ret || !unlock)
2930 		mmap_write_unlock(mm);
2931 
2932 	userfaultfd_unmap_complete(mm, &uf);
2933 	return ret;
2934 }
2935 
vm_munmap(unsigned long start,size_t len)2936 int vm_munmap(unsigned long start, size_t len)
2937 {
2938 	return __vm_munmap(start, len, false);
2939 }
2940 EXPORT_SYMBOL(vm_munmap);
2941 
SYSCALL_DEFINE2(munmap,unsigned long,addr,size_t,len)2942 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2943 {
2944 	addr = untagged_addr(addr);
2945 	return __vm_munmap(addr, len, true);
2946 }
2947 
2948 
2949 /*
2950  * Emulation of deprecated remap_file_pages() syscall.
2951  */
SYSCALL_DEFINE5(remap_file_pages,unsigned long,start,unsigned long,size,unsigned long,prot,unsigned long,pgoff,unsigned long,flags)2952 SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2953 		unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2954 {
2955 
2956 	struct mm_struct *mm = current->mm;
2957 	struct vm_area_struct *vma;
2958 	unsigned long populate = 0;
2959 	unsigned long ret = -EINVAL;
2960 	struct file *file;
2961 
2962 	pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n",
2963 		     current->comm, current->pid);
2964 
2965 	if (prot)
2966 		return ret;
2967 	start = start & PAGE_MASK;
2968 	size = size & PAGE_MASK;
2969 
2970 	if (start + size <= start)
2971 		return ret;
2972 
2973 	/* Does pgoff wrap? */
2974 	if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2975 		return ret;
2976 
2977 	if (mmap_write_lock_killable(mm))
2978 		return -EINTR;
2979 
2980 	vma = vma_lookup(mm, start);
2981 
2982 	if (!vma || !(vma->vm_flags & VM_SHARED))
2983 		goto out;
2984 
2985 	if (start + size > vma->vm_end) {
2986 		VMA_ITERATOR(vmi, mm, vma->vm_end);
2987 		struct vm_area_struct *next, *prev = vma;
2988 
2989 		for_each_vma_range(vmi, next, start + size) {
2990 			/* hole between vmas ? */
2991 			if (next->vm_start != prev->vm_end)
2992 				goto out;
2993 
2994 			if (next->vm_file != vma->vm_file)
2995 				goto out;
2996 
2997 			if (next->vm_flags != vma->vm_flags)
2998 				goto out;
2999 
3000 			if (start + size <= next->vm_end)
3001 				break;
3002 
3003 			prev = next;
3004 		}
3005 
3006 		if (!next)
3007 			goto out;
3008 	}
3009 
3010 	prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
3011 	prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
3012 	prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
3013 
3014 	flags &= MAP_NONBLOCK;
3015 	flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
3016 	if (vma->vm_flags & VM_LOCKED)
3017 		flags |= MAP_LOCKED;
3018 
3019 	file = get_file(vma->vm_file);
3020 	ret = do_mmap(vma->vm_file, start, size,
3021 			prot, flags, 0, pgoff, &populate, NULL);
3022 	fput(file);
3023 out:
3024 	mmap_write_unlock(mm);
3025 	if (populate)
3026 		mm_populate(ret, populate);
3027 	if (!IS_ERR_VALUE(ret))
3028 		ret = 0;
3029 	return ret;
3030 }
3031 
3032 /*
3033  * do_vma_munmap() - Unmap a full or partial vma.
3034  * @vmi: The vma iterator pointing at the vma
3035  * @vma: The first vma to be munmapped
3036  * @start: the start of the address to unmap
3037  * @end: The end of the address to unmap
3038  * @uf: The userfaultfd list_head
3039  * @unlock: Drop the lock on success
3040  *
3041  * unmaps a VMA mapping when the vma iterator is already in position.
3042  * Does not handle alignment.
3043  *
3044  * Return: 0 on success drops the lock of so directed, error on failure and will
3045  * still hold the lock.
3046  */
do_vma_munmap(struct vma_iterator * vmi,struct vm_area_struct * vma,unsigned long start,unsigned long end,struct list_head * uf,bool unlock)3047 int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
3048 		unsigned long start, unsigned long end, struct list_head *uf,
3049 		bool unlock)
3050 {
3051 	struct mm_struct *mm = vma->vm_mm;
3052 
3053 	arch_unmap(mm, start, end);
3054 	return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock);
3055 }
3056 
3057 /*
3058  * do_brk_flags() - Increase the brk vma if the flags match.
3059  * @vmi: The vma iterator
3060  * @addr: The start address
3061  * @len: The length of the increase
3062  * @vma: The vma,
3063  * @flags: The VMA Flags
3064  *
3065  * Extend the brk VMA from addr to addr + len.  If the VMA is NULL or the flags
3066  * do not match then create a new anonymous VMA.  Eventually we may be able to
3067  * do some brk-specific accounting here.
3068  */
do_brk_flags(struct vma_iterator * vmi,struct vm_area_struct * vma,unsigned long addr,unsigned long len,unsigned long flags)3069 static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
3070 		unsigned long addr, unsigned long len, unsigned long flags)
3071 {
3072 	struct mm_struct *mm = current->mm;
3073 	struct vma_prepare vp;
3074 
3075 	/*
3076 	 * Check against address space limits by the changed size
3077 	 * Note: This happens *after* clearing old mappings in some code paths.
3078 	 */
3079 	flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
3080 	if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
3081 		return -ENOMEM;
3082 
3083 	if (mm->map_count > sysctl_max_map_count)
3084 		return -ENOMEM;
3085 
3086 	if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
3087 		return -ENOMEM;
3088 
3089 	/*
3090 	 * Expand the existing vma if possible; Note that singular lists do not
3091 	 * occur after forking, so the expand will only happen on new VMAs.
3092 	 */
3093 	if (vma && vma->vm_end == addr && !vma_policy(vma) &&
3094 	    can_vma_merge_after(vma, flags, NULL, NULL,
3095 				addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) {
3096 		vma_iter_config(vmi, vma->vm_start, addr + len);
3097 		if (vma_iter_prealloc(vmi, vma))
3098 			goto unacct_fail;
3099 
3100 		vma_start_write(vma);
3101 
3102 		init_vma_prep(&vp, vma);
3103 		vma_prepare(&vp);
3104 		vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0);
3105 		vma->vm_end = addr + len;
3106 		vm_flags_set(vma, VM_SOFTDIRTY);
3107 		vma_iter_store(vmi, vma);
3108 
3109 		vma_complete(&vp, vmi, mm);
3110 		khugepaged_enter_vma(vma, flags);
3111 		goto out;
3112 	}
3113 
3114 	if (vma)
3115 		vma_iter_next_range(vmi);
3116 	/* create a vma struct for an anonymous mapping */
3117 	vma = vm_area_alloc(mm);
3118 	if (!vma)
3119 		goto unacct_fail;
3120 
3121 	vma_set_anonymous(vma);
3122 	vma->vm_start = addr;
3123 	vma->vm_end = addr + len;
3124 	vma->vm_pgoff = addr >> PAGE_SHIFT;
3125 	vm_flags_init(vma, flags);
3126 	vma->vm_page_prot = vm_get_page_prot(flags);
3127 	vma_start_write(vma);
3128 	if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL))
3129 		goto mas_store_fail;
3130 
3131 	mm->map_count++;
3132 	validate_mm(mm);
3133 	ksm_add_vma(vma);
3134 out:
3135 	perf_event_mmap(vma);
3136 	mm->total_vm += len >> PAGE_SHIFT;
3137 	mm->data_vm += len >> PAGE_SHIFT;
3138 	if (flags & VM_LOCKED)
3139 		mm->locked_vm += (len >> PAGE_SHIFT);
3140 	vm_flags_set(vma, VM_SOFTDIRTY);
3141 	return 0;
3142 
3143 mas_store_fail:
3144 	vm_area_free(vma);
3145 unacct_fail:
3146 	vm_unacct_memory(len >> PAGE_SHIFT);
3147 	return -ENOMEM;
3148 }
3149 
vm_brk_flags(unsigned long addr,unsigned long request,unsigned long flags)3150 int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
3151 {
3152 	struct mm_struct *mm = current->mm;
3153 	struct vm_area_struct *vma = NULL;
3154 	unsigned long len;
3155 	int ret;
3156 	bool populate;
3157 	LIST_HEAD(uf);
3158 	VMA_ITERATOR(vmi, mm, addr);
3159 
3160 	len = PAGE_ALIGN(request);
3161 	if (len < request)
3162 		return -ENOMEM;
3163 	if (!len)
3164 		return 0;
3165 
3166 	/* Until we need other flags, refuse anything except VM_EXEC. */
3167 	if ((flags & (~VM_EXEC)) != 0)
3168 		return -EINVAL;
3169 
3170 	if (mmap_write_lock_killable(mm))
3171 		return -EINTR;
3172 
3173 	ret = check_brk_limits(addr, len);
3174 	if (ret)
3175 		goto limits_failed;
3176 
3177 	ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0);
3178 	if (ret)
3179 		goto munmap_failed;
3180 
3181 	vma = vma_prev(&vmi);
3182 	ret = do_brk_flags(&vmi, vma, addr, len, flags);
3183 	populate = ((mm->def_flags & VM_LOCKED) != 0);
3184 	mmap_write_unlock(mm);
3185 	userfaultfd_unmap_complete(mm, &uf);
3186 	if (populate && !ret)
3187 		mm_populate(addr, len);
3188 	return ret;
3189 
3190 munmap_failed:
3191 limits_failed:
3192 	mmap_write_unlock(mm);
3193 	return ret;
3194 }
3195 EXPORT_SYMBOL(vm_brk_flags);
3196 
vm_brk(unsigned long addr,unsigned long len)3197 int vm_brk(unsigned long addr, unsigned long len)
3198 {
3199 	return vm_brk_flags(addr, len, 0);
3200 }
3201 EXPORT_SYMBOL(vm_brk);
3202 
3203 /* Release all mmaps. */
exit_mmap(struct mm_struct * mm)3204 void exit_mmap(struct mm_struct *mm)
3205 {
3206 	struct mmu_gather tlb;
3207 	struct vm_area_struct *vma;
3208 	unsigned long nr_accounted = 0;
3209 	MA_STATE(mas, &mm->mm_mt, 0, 0);
3210 	int count = 0;
3211 
3212 	/* mm's last user has gone, and its about to be pulled down */
3213 	mmu_notifier_release(mm);
3214 
3215 	mmap_read_lock(mm);
3216 	arch_exit_mmap(mm);
3217 
3218 	vma = mas_find(&mas, ULONG_MAX);
3219 	if (!vma) {
3220 		/* Can happen if dup_mmap() received an OOM */
3221 		mmap_read_unlock(mm);
3222 		return;
3223 	}
3224 
3225 	lru_add_drain();
3226 	flush_cache_mm(mm);
3227 	tlb_gather_mmu_fullmm(&tlb, mm);
3228 	/* update_hiwater_rss(mm) here? but nobody should be looking */
3229 	/* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */
3230 	unmap_vmas(&tlb, &mas, vma, 0, ULONG_MAX, ULONG_MAX, false);
3231 	mmap_read_unlock(mm);
3232 
3233 	/*
3234 	 * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper
3235 	 * because the memory has been already freed.
3236 	 */
3237 	set_bit(MMF_OOM_SKIP, &mm->flags);
3238 	mmap_write_lock(mm);
3239 	mt_clear_in_rcu(&mm->mm_mt);
3240 	mas_set(&mas, vma->vm_end);
3241 	free_pgtables(&tlb, &mas, vma, FIRST_USER_ADDRESS,
3242 		      USER_PGTABLES_CEILING, true);
3243 	tlb_finish_mmu(&tlb);
3244 
3245 	/*
3246 	 * Walk the list again, actually closing and freeing it, with preemption
3247 	 * enabled, without holding any MM locks besides the unreachable
3248 	 * mmap_write_lock.
3249 	 */
3250 	mas_set(&mas, vma->vm_end);
3251 	do {
3252 		if (vma->vm_flags & VM_ACCOUNT)
3253 			nr_accounted += vma_pages(vma);
3254 		remove_vma(vma, true);
3255 		count++;
3256 		cond_resched();
3257 	} while ((vma = mas_find(&mas, ULONG_MAX)) != NULL);
3258 
3259 	BUG_ON(count != mm->map_count);
3260 
3261 	trace_exit_mmap(mm);
3262 	__mt_destroy(&mm->mm_mt);
3263 	mmap_write_unlock(mm);
3264 	vm_unacct_memory(nr_accounted);
3265 }
3266 
3267 /* Insert vm structure into process list sorted by address
3268  * and into the inode's i_mmap tree.  If vm_file is non-NULL
3269  * then i_mmap_rwsem is taken here.
3270  */
insert_vm_struct(struct mm_struct * mm,struct vm_area_struct * vma)3271 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
3272 {
3273 	unsigned long charged = vma_pages(vma);
3274 
3275 
3276 	if (find_vma_intersection(mm, vma->vm_start, vma->vm_end))
3277 		return -ENOMEM;
3278 
3279 	if ((vma->vm_flags & VM_ACCOUNT) &&
3280 	     security_vm_enough_memory_mm(mm, charged))
3281 		return -ENOMEM;
3282 
3283 	/*
3284 	 * The vm_pgoff of a purely anonymous vma should be irrelevant
3285 	 * until its first write fault, when page's anon_vma and index
3286 	 * are set.  But now set the vm_pgoff it will almost certainly
3287 	 * end up with (unless mremap moves it elsewhere before that
3288 	 * first wfault), so /proc/pid/maps tells a consistent story.
3289 	 *
3290 	 * By setting it to reflect the virtual start address of the
3291 	 * vma, merges and splits can happen in a seamless way, just
3292 	 * using the existing file pgoff checks and manipulations.
3293 	 * Similarly in do_mmap and in do_brk_flags.
3294 	 */
3295 	if (vma_is_anonymous(vma)) {
3296 		BUG_ON(vma->anon_vma);
3297 		vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3298 	}
3299 
3300 	if (vma_link(mm, vma)) {
3301 		vm_unacct_memory(charged);
3302 		return -ENOMEM;
3303 	}
3304 
3305 	return 0;
3306 }
3307 
3308 /*
3309  * Copy the vma structure to a new location in the same mm,
3310  * prior to moving page table entries, to effect an mremap move.
3311  */
copy_vma(struct vm_area_struct ** vmap,unsigned long addr,unsigned long len,pgoff_t pgoff,bool * need_rmap_locks)3312 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
3313 	unsigned long addr, unsigned long len, pgoff_t pgoff,
3314 	bool *need_rmap_locks)
3315 {
3316 	struct vm_area_struct *vma = *vmap;
3317 	unsigned long vma_start = vma->vm_start;
3318 	struct mm_struct *mm = vma->vm_mm;
3319 	struct vm_area_struct *new_vma, *prev;
3320 	bool faulted_in_anon_vma = true;
3321 	VMA_ITERATOR(vmi, mm, addr);
3322 
3323 	/*
3324 	 * If anonymous vma has not yet been faulted, update new pgoff
3325 	 * to match new location, to increase its chance of merging.
3326 	 */
3327 	if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
3328 		pgoff = addr >> PAGE_SHIFT;
3329 		faulted_in_anon_vma = false;
3330 	}
3331 
3332 	new_vma = find_vma_prev(mm, addr, &prev);
3333 	if (new_vma && new_vma->vm_start < addr + len)
3334 		return NULL;	/* should never get here */
3335 
3336 	new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags,
3337 			    vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3338 			    vma->vm_userfaultfd_ctx, anon_vma_name(vma));
3339 	if (new_vma) {
3340 		/*
3341 		 * Source vma may have been merged into new_vma
3342 		 */
3343 		if (unlikely(vma_start >= new_vma->vm_start &&
3344 			     vma_start < new_vma->vm_end)) {
3345 			/*
3346 			 * The only way we can get a vma_merge with
3347 			 * self during an mremap is if the vma hasn't
3348 			 * been faulted in yet and we were allowed to
3349 			 * reset the dst vma->vm_pgoff to the
3350 			 * destination address of the mremap to allow
3351 			 * the merge to happen. mremap must change the
3352 			 * vm_pgoff linearity between src and dst vmas
3353 			 * (in turn preventing a vma_merge) to be
3354 			 * safe. It is only safe to keep the vm_pgoff
3355 			 * linear if there are no pages mapped yet.
3356 			 */
3357 			VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
3358 			*vmap = vma = new_vma;
3359 		}
3360 		*need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
3361 	} else {
3362 		new_vma = vm_area_dup(vma);
3363 		if (!new_vma)
3364 			goto out;
3365 		new_vma->vm_start = addr;
3366 		new_vma->vm_end = addr + len;
3367 		new_vma->vm_pgoff = pgoff;
3368 		if (vma_dup_policy(vma, new_vma))
3369 			goto out_free_vma;
3370 		if (anon_vma_clone(new_vma, vma))
3371 			goto out_free_mempol;
3372 		if (new_vma->vm_file)
3373 			get_file(new_vma->vm_file);
3374 		if (new_vma->vm_ops && new_vma->vm_ops->open)
3375 			new_vma->vm_ops->open(new_vma);
3376 		if (vma_link(mm, new_vma))
3377 			goto out_vma_link;
3378 		*need_rmap_locks = false;
3379 	}
3380 	return new_vma;
3381 
3382 out_vma_link:
3383 	if (new_vma->vm_ops && new_vma->vm_ops->close)
3384 		new_vma->vm_ops->close(new_vma);
3385 
3386 	if (new_vma->vm_file)
3387 		fput(new_vma->vm_file);
3388 
3389 	unlink_anon_vmas(new_vma);
3390 out_free_mempol:
3391 	mpol_put(vma_policy(new_vma));
3392 out_free_vma:
3393 	vm_area_free(new_vma);
3394 out:
3395 	return NULL;
3396 }
3397 
3398 /*
3399  * Return true if the calling process may expand its vm space by the passed
3400  * number of pages
3401  */
may_expand_vm(struct mm_struct * mm,vm_flags_t flags,unsigned long npages)3402 bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
3403 {
3404 	if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3405 		return false;
3406 
3407 	if (is_data_mapping(flags) &&
3408 	    mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3409 		/* Workaround for Valgrind */
3410 		if (rlimit(RLIMIT_DATA) == 0 &&
3411 		    mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3412 			return true;
3413 
3414 		pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3415 			     current->comm, current->pid,
3416 			     (mm->data_vm + npages) << PAGE_SHIFT,
3417 			     rlimit(RLIMIT_DATA),
3418 			     ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
3419 
3420 		if (!ignore_rlimit_data)
3421 			return false;
3422 	}
3423 
3424 	return true;
3425 }
3426 
vm_stat_account(struct mm_struct * mm,vm_flags_t flags,long npages)3427 void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3428 {
3429 	WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages);
3430 
3431 	if (is_exec_mapping(flags))
3432 		mm->exec_vm += npages;
3433 	else if (is_stack_mapping(flags))
3434 		mm->stack_vm += npages;
3435 	else if (is_data_mapping(flags))
3436 		mm->data_vm += npages;
3437 }
3438 
3439 static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
3440 
3441 /*
3442  * Having a close hook prevents vma merging regardless of flags.
3443  */
special_mapping_close(struct vm_area_struct * vma)3444 static void special_mapping_close(struct vm_area_struct *vma)
3445 {
3446 }
3447 
special_mapping_name(struct vm_area_struct * vma)3448 static const char *special_mapping_name(struct vm_area_struct *vma)
3449 {
3450 	return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3451 }
3452 
special_mapping_mremap(struct vm_area_struct * new_vma)3453 static int special_mapping_mremap(struct vm_area_struct *new_vma)
3454 {
3455 	struct vm_special_mapping *sm = new_vma->vm_private_data;
3456 
3457 	if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3458 		return -EFAULT;
3459 
3460 	if (sm->mremap)
3461 		return sm->mremap(sm, new_vma);
3462 
3463 	return 0;
3464 }
3465 
special_mapping_split(struct vm_area_struct * vma,unsigned long addr)3466 static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr)
3467 {
3468 	/*
3469 	 * Forbid splitting special mappings - kernel has expectations over
3470 	 * the number of pages in mapping. Together with VM_DONTEXPAND
3471 	 * the size of vma should stay the same over the special mapping's
3472 	 * lifetime.
3473 	 */
3474 	return -EINVAL;
3475 }
3476 
3477 static const struct vm_operations_struct special_mapping_vmops = {
3478 	.close = special_mapping_close,
3479 	.fault = special_mapping_fault,
3480 	.mremap = special_mapping_mremap,
3481 	.name = special_mapping_name,
3482 	/* vDSO code relies that VVAR can't be accessed remotely */
3483 	.access = NULL,
3484 	.may_split = special_mapping_split,
3485 };
3486 
3487 static const struct vm_operations_struct legacy_special_mapping_vmops = {
3488 	.close = special_mapping_close,
3489 	.fault = special_mapping_fault,
3490 };
3491 
special_mapping_fault(struct vm_fault * vmf)3492 static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
3493 {
3494 	struct vm_area_struct *vma = vmf->vma;
3495 	pgoff_t pgoff;
3496 	struct page **pages;
3497 
3498 	if (vma->vm_ops == &legacy_special_mapping_vmops) {
3499 		pages = vma->vm_private_data;
3500 	} else {
3501 		struct vm_special_mapping *sm = vma->vm_private_data;
3502 
3503 		if (sm->fault)
3504 			return sm->fault(sm, vmf->vma, vmf);
3505 
3506 		pages = sm->pages;
3507 	}
3508 
3509 	for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3510 		pgoff--;
3511 
3512 	if (*pages) {
3513 		struct page *page = *pages;
3514 		get_page(page);
3515 		vmf->page = page;
3516 		return 0;
3517 	}
3518 
3519 	return VM_FAULT_SIGBUS;
3520 }
3521 
__install_special_mapping(struct mm_struct * mm,unsigned long addr,unsigned long len,unsigned long vm_flags,void * priv,const struct vm_operations_struct * ops)3522 static struct vm_area_struct *__install_special_mapping(
3523 	struct mm_struct *mm,
3524 	unsigned long addr, unsigned long len,
3525 	unsigned long vm_flags, void *priv,
3526 	const struct vm_operations_struct *ops)
3527 {
3528 	int ret;
3529 	struct vm_area_struct *vma;
3530 
3531 	vma = vm_area_alloc(mm);
3532 	if (unlikely(vma == NULL))
3533 		return ERR_PTR(-ENOMEM);
3534 
3535 	vma->vm_start = addr;
3536 	vma->vm_end = addr + len;
3537 
3538 	vm_flags_init(vma, (vm_flags | mm->def_flags |
3539 		      VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK);
3540 	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3541 
3542 	vma->vm_ops = ops;
3543 	vma->vm_private_data = priv;
3544 
3545 	ret = insert_vm_struct(mm, vma);
3546 	if (ret)
3547 		goto out;
3548 
3549 	vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3550 
3551 	perf_event_mmap(vma);
3552 
3553 	return vma;
3554 
3555 out:
3556 	vm_area_free(vma);
3557 	return ERR_PTR(ret);
3558 }
3559 
vma_is_special_mapping(const struct vm_area_struct * vma,const struct vm_special_mapping * sm)3560 bool vma_is_special_mapping(const struct vm_area_struct *vma,
3561 	const struct vm_special_mapping *sm)
3562 {
3563 	return vma->vm_private_data == sm &&
3564 		(vma->vm_ops == &special_mapping_vmops ||
3565 		 vma->vm_ops == &legacy_special_mapping_vmops);
3566 }
3567 
3568 /*
3569  * Called with mm->mmap_lock held for writing.
3570  * Insert a new vma covering the given region, with the given flags.
3571  * Its pages are supplied by the given array of struct page *.
3572  * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3573  * The region past the last page supplied will always produce SIGBUS.
3574  * The array pointer and the pages it points to are assumed to stay alive
3575  * for as long as this mapping might exist.
3576  */
_install_special_mapping(struct mm_struct * mm,unsigned long addr,unsigned long len,unsigned long vm_flags,const struct vm_special_mapping * spec)3577 struct vm_area_struct *_install_special_mapping(
3578 	struct mm_struct *mm,
3579 	unsigned long addr, unsigned long len,
3580 	unsigned long vm_flags, const struct vm_special_mapping *spec)
3581 {
3582 	return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3583 					&special_mapping_vmops);
3584 }
3585 
install_special_mapping(struct mm_struct * mm,unsigned long addr,unsigned long len,unsigned long vm_flags,struct page ** pages)3586 int install_special_mapping(struct mm_struct *mm,
3587 			    unsigned long addr, unsigned long len,
3588 			    unsigned long vm_flags, struct page **pages)
3589 {
3590 	struct vm_area_struct *vma = __install_special_mapping(
3591 		mm, addr, len, vm_flags, (void *)pages,
3592 		&legacy_special_mapping_vmops);
3593 
3594 	return PTR_ERR_OR_ZERO(vma);
3595 }
3596 
3597 static DEFINE_MUTEX(mm_all_locks_mutex);
3598 
vm_lock_anon_vma(struct mm_struct * mm,struct anon_vma * anon_vma)3599 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3600 {
3601 	if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3602 		/*
3603 		 * The LSB of head.next can't change from under us
3604 		 * because we hold the mm_all_locks_mutex.
3605 		 */
3606 		down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock);
3607 		/*
3608 		 * We can safely modify head.next after taking the
3609 		 * anon_vma->root->rwsem. If some other vma in this mm shares
3610 		 * the same anon_vma we won't take it again.
3611 		 *
3612 		 * No need of atomic instructions here, head.next
3613 		 * can't change from under us thanks to the
3614 		 * anon_vma->root->rwsem.
3615 		 */
3616 		if (__test_and_set_bit(0, (unsigned long *)
3617 				       &anon_vma->root->rb_root.rb_root.rb_node))
3618 			BUG();
3619 	}
3620 }
3621 
vm_lock_mapping(struct mm_struct * mm,struct address_space * mapping)3622 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3623 {
3624 	if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3625 		/*
3626 		 * AS_MM_ALL_LOCKS can't change from under us because
3627 		 * we hold the mm_all_locks_mutex.
3628 		 *
3629 		 * Operations on ->flags have to be atomic because
3630 		 * even if AS_MM_ALL_LOCKS is stable thanks to the
3631 		 * mm_all_locks_mutex, there may be other cpus
3632 		 * changing other bitflags in parallel to us.
3633 		 */
3634 		if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3635 			BUG();
3636 		down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock);
3637 	}
3638 }
3639 
3640 /*
3641  * This operation locks against the VM for all pte/vma/mm related
3642  * operations that could ever happen on a certain mm. This includes
3643  * vmtruncate, try_to_unmap, and all page faults.
3644  *
3645  * The caller must take the mmap_lock in write mode before calling
3646  * mm_take_all_locks(). The caller isn't allowed to release the
3647  * mmap_lock until mm_drop_all_locks() returns.
3648  *
3649  * mmap_lock in write mode is required in order to block all operations
3650  * that could modify pagetables and free pages without need of
3651  * altering the vma layout. It's also needed in write mode to avoid new
3652  * anon_vmas to be associated with existing vmas.
3653  *
3654  * A single task can't take more than one mm_take_all_locks() in a row
3655  * or it would deadlock.
3656  *
3657  * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3658  * mapping->flags avoid to take the same lock twice, if more than one
3659  * vma in this mm is backed by the same anon_vma or address_space.
3660  *
3661  * We take locks in following order, accordingly to comment at beginning
3662  * of mm/rmap.c:
3663  *   - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3664  *     hugetlb mapping);
3665  *   - all vmas marked locked
3666  *   - all i_mmap_rwsem locks;
3667  *   - all anon_vma->rwseml
3668  *
3669  * We can take all locks within these types randomly because the VM code
3670  * doesn't nest them and we protected from parallel mm_take_all_locks() by
3671  * mm_all_locks_mutex.
3672  *
3673  * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3674  * that may have to take thousand of locks.
3675  *
3676  * mm_take_all_locks() can fail if it's interrupted by signals.
3677  */
mm_take_all_locks(struct mm_struct * mm)3678 int mm_take_all_locks(struct mm_struct *mm)
3679 {
3680 	struct vm_area_struct *vma;
3681 	struct anon_vma_chain *avc;
3682 	MA_STATE(mas, &mm->mm_mt, 0, 0);
3683 
3684 	mmap_assert_write_locked(mm);
3685 
3686 	mutex_lock(&mm_all_locks_mutex);
3687 
3688 	/*
3689 	 * vma_start_write() does not have a complement in mm_drop_all_locks()
3690 	 * because vma_start_write() is always asymmetrical; it marks a VMA as
3691 	 * being written to until mmap_write_unlock() or mmap_write_downgrade()
3692 	 * is reached.
3693 	 */
3694 	mas_for_each(&mas, vma, ULONG_MAX) {
3695 		if (signal_pending(current))
3696 			goto out_unlock;
3697 		vma_start_write(vma);
3698 	}
3699 
3700 	mas_set(&mas, 0);
3701 	mas_for_each(&mas, vma, ULONG_MAX) {
3702 		if (signal_pending(current))
3703 			goto out_unlock;
3704 		if (vma->vm_file && vma->vm_file->f_mapping &&
3705 				is_vm_hugetlb_page(vma))
3706 			vm_lock_mapping(mm, vma->vm_file->f_mapping);
3707 	}
3708 
3709 	mas_set(&mas, 0);
3710 	mas_for_each(&mas, vma, ULONG_MAX) {
3711 		if (signal_pending(current))
3712 			goto out_unlock;
3713 		if (vma->vm_file && vma->vm_file->f_mapping &&
3714 				!is_vm_hugetlb_page(vma))
3715 			vm_lock_mapping(mm, vma->vm_file->f_mapping);
3716 	}
3717 
3718 	mas_set(&mas, 0);
3719 	mas_for_each(&mas, vma, ULONG_MAX) {
3720 		if (signal_pending(current))
3721 			goto out_unlock;
3722 		if (vma->anon_vma)
3723 			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3724 				vm_lock_anon_vma(mm, avc->anon_vma);
3725 	}
3726 
3727 	return 0;
3728 
3729 out_unlock:
3730 	mm_drop_all_locks(mm);
3731 	return -EINTR;
3732 }
3733 
vm_unlock_anon_vma(struct anon_vma * anon_vma)3734 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3735 {
3736 	if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3737 		/*
3738 		 * The LSB of head.next can't change to 0 from under
3739 		 * us because we hold the mm_all_locks_mutex.
3740 		 *
3741 		 * We must however clear the bitflag before unlocking
3742 		 * the vma so the users using the anon_vma->rb_root will
3743 		 * never see our bitflag.
3744 		 *
3745 		 * No need of atomic instructions here, head.next
3746 		 * can't change from under us until we release the
3747 		 * anon_vma->root->rwsem.
3748 		 */
3749 		if (!__test_and_clear_bit(0, (unsigned long *)
3750 					  &anon_vma->root->rb_root.rb_root.rb_node))
3751 			BUG();
3752 		anon_vma_unlock_write(anon_vma);
3753 	}
3754 }
3755 
vm_unlock_mapping(struct address_space * mapping)3756 static void vm_unlock_mapping(struct address_space *mapping)
3757 {
3758 	if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3759 		/*
3760 		 * AS_MM_ALL_LOCKS can't change to 0 from under us
3761 		 * because we hold the mm_all_locks_mutex.
3762 		 */
3763 		i_mmap_unlock_write(mapping);
3764 		if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3765 					&mapping->flags))
3766 			BUG();
3767 	}
3768 }
3769 
3770 /*
3771  * The mmap_lock cannot be released by the caller until
3772  * mm_drop_all_locks() returns.
3773  */
mm_drop_all_locks(struct mm_struct * mm)3774 void mm_drop_all_locks(struct mm_struct *mm)
3775 {
3776 	struct vm_area_struct *vma;
3777 	struct anon_vma_chain *avc;
3778 	MA_STATE(mas, &mm->mm_mt, 0, 0);
3779 
3780 	mmap_assert_write_locked(mm);
3781 	BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3782 
3783 	mas_for_each(&mas, vma, ULONG_MAX) {
3784 		if (vma->anon_vma)
3785 			list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3786 				vm_unlock_anon_vma(avc->anon_vma);
3787 		if (vma->vm_file && vma->vm_file->f_mapping)
3788 			vm_unlock_mapping(vma->vm_file->f_mapping);
3789 	}
3790 
3791 	mutex_unlock(&mm_all_locks_mutex);
3792 }
3793 
3794 /*
3795  * initialise the percpu counter for VM
3796  */
mmap_init(void)3797 void __init mmap_init(void)
3798 {
3799 	int ret;
3800 
3801 	ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3802 	VM_BUG_ON(ret);
3803 }
3804 
3805 /*
3806  * Initialise sysctl_user_reserve_kbytes.
3807  *
3808  * This is intended to prevent a user from starting a single memory hogging
3809  * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3810  * mode.
3811  *
3812  * The default value is min(3% of free memory, 128MB)
3813  * 128MB is enough to recover with sshd/login, bash, and top/kill.
3814  */
init_user_reserve(void)3815 static int init_user_reserve(void)
3816 {
3817 	unsigned long free_kbytes;
3818 
3819 	free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
3820 
3821 	sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3822 	return 0;
3823 }
3824 subsys_initcall(init_user_reserve);
3825 
3826 /*
3827  * Initialise sysctl_admin_reserve_kbytes.
3828  *
3829  * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3830  * to log in and kill a memory hogging process.
3831  *
3832  * Systems with more than 256MB will reserve 8MB, enough to recover
3833  * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3834  * only reserve 3% of free pages by default.
3835  */
init_admin_reserve(void)3836 static int init_admin_reserve(void)
3837 {
3838 	unsigned long free_kbytes;
3839 
3840 	free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
3841 
3842 	sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3843 	return 0;
3844 }
3845 subsys_initcall(init_admin_reserve);
3846 
3847 /*
3848  * Reinititalise user and admin reserves if memory is added or removed.
3849  *
3850  * The default user reserve max is 128MB, and the default max for the
3851  * admin reserve is 8MB. These are usually, but not always, enough to
3852  * enable recovery from a memory hogging process using login/sshd, a shell,
3853  * and tools like top. It may make sense to increase or even disable the
3854  * reserve depending on the existence of swap or variations in the recovery
3855  * tools. So, the admin may have changed them.
3856  *
3857  * If memory is added and the reserves have been eliminated or increased above
3858  * the default max, then we'll trust the admin.
3859  *
3860  * If memory is removed and there isn't enough free memory, then we
3861  * need to reset the reserves.
3862  *
3863  * Otherwise keep the reserve set by the admin.
3864  */
reserve_mem_notifier(struct notifier_block * nb,unsigned long action,void * data)3865 static int reserve_mem_notifier(struct notifier_block *nb,
3866 			     unsigned long action, void *data)
3867 {
3868 	unsigned long tmp, free_kbytes;
3869 
3870 	switch (action) {
3871 	case MEM_ONLINE:
3872 		/* Default max is 128MB. Leave alone if modified by operator. */
3873 		tmp = sysctl_user_reserve_kbytes;
3874 		if (0 < tmp && tmp < (1UL << 17))
3875 			init_user_reserve();
3876 
3877 		/* Default max is 8MB.  Leave alone if modified by operator. */
3878 		tmp = sysctl_admin_reserve_kbytes;
3879 		if (0 < tmp && tmp < (1UL << 13))
3880 			init_admin_reserve();
3881 
3882 		break;
3883 	case MEM_OFFLINE:
3884 		free_kbytes = K(global_zone_page_state(NR_FREE_PAGES));
3885 
3886 		if (sysctl_user_reserve_kbytes > free_kbytes) {
3887 			init_user_reserve();
3888 			pr_info("vm.user_reserve_kbytes reset to %lu\n",
3889 				sysctl_user_reserve_kbytes);
3890 		}
3891 
3892 		if (sysctl_admin_reserve_kbytes > free_kbytes) {
3893 			init_admin_reserve();
3894 			pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3895 				sysctl_admin_reserve_kbytes);
3896 		}
3897 		break;
3898 	default:
3899 		break;
3900 	}
3901 	return NOTIFY_OK;
3902 }
3903 
init_reserve_notifier(void)3904 static int __meminit init_reserve_notifier(void)
3905 {
3906 	if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI))
3907 		pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3908 
3909 	return 0;
3910 }
3911 subsys_initcall(init_reserve_notifier);
3912