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