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/vmacache.h>
18 : #include <linux/shm.h>
19 : #include <linux/mman.h>
20 : #include <linux/pagemap.h>
21 : #include <linux/swap.h>
22 : #include <linux/syscalls.h>
23 : #include <linux/capability.h>
24 : #include <linux/init.h>
25 : #include <linux/file.h>
26 : #include <linux/fs.h>
27 : #include <linux/personality.h>
28 : #include <linux/security.h>
29 : #include <linux/hugetlb.h>
30 : #include <linux/shmem_fs.h>
31 : #include <linux/profile.h>
32 : #include <linux/export.h>
33 : #include <linux/mount.h>
34 : #include <linux/mempolicy.h>
35 : #include <linux/rmap.h>
36 : #include <linux/mmu_notifier.h>
37 : #include <linux/mmdebug.h>
38 : #include <linux/perf_event.h>
39 : #include <linux/audit.h>
40 : #include <linux/khugepaged.h>
41 : #include <linux/uprobes.h>
42 : #include <linux/rbtree_augmented.h>
43 : #include <linux/notifier.h>
44 : #include <linux/memory.h>
45 : #include <linux/printk.h>
46 : #include <linux/userfaultfd_k.h>
47 : #include <linux/moduleparam.h>
48 : #include <linux/pkeys.h>
49 : #include <linux/oom.h>
50 : #include <linux/sched/mm.h>
51 :
52 : #include <linux/uaccess.h>
53 : #include <asm/cacheflush.h>
54 : #include <asm/tlb.h>
55 : #include <asm/mmu_context.h>
56 :
57 : #define CREATE_TRACE_POINTS
58 : #include <trace/events/mmap.h>
59 :
60 : #include "internal.h"
61 :
62 : #ifndef arch_mmap_check
63 : #define arch_mmap_check(addr, len, flags) (0)
64 : #endif
65 :
66 : #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
67 : const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
68 : const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
69 : int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
70 : #endif
71 : #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
72 : const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
73 : const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
74 : int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
75 : #endif
76 :
77 : static bool ignore_rlimit_data;
78 : core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
79 :
80 : static void unmap_region(struct mm_struct *mm,
81 : struct vm_area_struct *vma, struct vm_area_struct *prev,
82 : unsigned long start, unsigned long end);
83 :
84 : /* description of effects of mapping type and prot in current implementation.
85 : * this is due to the limited x86 page protection hardware. The expected
86 : * behavior is in parens:
87 : *
88 : * map_type prot
89 : * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
90 : * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 : * w: (no) no w: (no) no w: (yes) yes w: (no) no
92 : * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 : *
94 : * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
95 : * w: (no) no w: (no) no w: (copy) copy w: (no) no
96 : * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
97 : *
98 : * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and
99 : * MAP_PRIVATE (with Enhanced PAN supported):
100 : * r: (no) no
101 : * w: (no) no
102 : * x: (yes) yes
103 : */
104 : pgprot_t protection_map[16] __ro_after_init = {
105 : __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
106 : __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
107 : };
108 :
109 : #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT
110 : static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
111 : {
112 : return prot;
113 : }
114 : #endif
115 :
116 0 : pgprot_t vm_get_page_prot(unsigned long vm_flags)
117 : {
118 0 : pgprot_t ret = __pgprot(pgprot_val(protection_map[vm_flags &
119 : (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
120 : pgprot_val(arch_vm_get_page_prot(vm_flags)));
121 :
122 0 : return arch_filter_pgprot(ret);
123 : }
124 : EXPORT_SYMBOL(vm_get_page_prot);
125 :
126 : static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
127 : {
128 0 : return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
129 : }
130 :
131 : /* Update vma->vm_page_prot to reflect vma->vm_flags. */
132 0 : void vma_set_page_prot(struct vm_area_struct *vma)
133 : {
134 0 : unsigned long vm_flags = vma->vm_flags;
135 : pgprot_t vm_page_prot;
136 :
137 0 : vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
138 0 : if (vma_wants_writenotify(vma, vm_page_prot)) {
139 0 : vm_flags &= ~VM_SHARED;
140 0 : vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
141 : }
142 : /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */
143 0 : WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
144 0 : }
145 :
146 : /*
147 : * Requires inode->i_mapping->i_mmap_rwsem
148 : */
149 : static void __remove_shared_vm_struct(struct vm_area_struct *vma,
150 : struct file *file, struct address_space *mapping)
151 : {
152 0 : if (vma->vm_flags & VM_SHARED)
153 : mapping_unmap_writable(mapping);
154 :
155 0 : flush_dcache_mmap_lock(mapping);
156 0 : vma_interval_tree_remove(vma, &mapping->i_mmap);
157 0 : flush_dcache_mmap_unlock(mapping);
158 : }
159 :
160 : /*
161 : * Unlink a file-based vm structure from its interval tree, to hide
162 : * vma from rmap and vmtruncate before freeing its page tables.
163 : */
164 0 : void unlink_file_vma(struct vm_area_struct *vma)
165 : {
166 0 : struct file *file = vma->vm_file;
167 :
168 0 : if (file) {
169 0 : struct address_space *mapping = file->f_mapping;
170 0 : i_mmap_lock_write(mapping);
171 0 : __remove_shared_vm_struct(vma, file, mapping);
172 : i_mmap_unlock_write(mapping);
173 : }
174 0 : }
175 :
176 : /*
177 : * Close a vm structure and free it, returning the next.
178 : */
179 0 : static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
180 : {
181 0 : struct vm_area_struct *next = vma->vm_next;
182 :
183 : might_sleep();
184 0 : if (vma->vm_ops && vma->vm_ops->close)
185 0 : vma->vm_ops->close(vma);
186 0 : if (vma->vm_file)
187 0 : fput(vma->vm_file);
188 0 : mpol_put(vma_policy(vma));
189 0 : vm_area_free(vma);
190 0 : return next;
191 : }
192 :
193 : static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags,
194 : struct list_head *uf);
195 0 : SYSCALL_DEFINE1(brk, unsigned long, brk)
196 : {
197 : unsigned long newbrk, oldbrk, origbrk;
198 0 : struct mm_struct *mm = current->mm;
199 : struct vm_area_struct *next;
200 : unsigned long min_brk;
201 : bool populate;
202 0 : bool downgraded = false;
203 0 : LIST_HEAD(uf);
204 :
205 0 : if (mmap_write_lock_killable(mm))
206 : return -EINTR;
207 :
208 0 : origbrk = mm->brk;
209 :
210 : #ifdef CONFIG_COMPAT_BRK
211 : /*
212 : * CONFIG_COMPAT_BRK can still be overridden by setting
213 : * randomize_va_space to 2, which will still cause mm->start_brk
214 : * to be arbitrarily shifted
215 : */
216 0 : if (current->brk_randomized)
217 0 : min_brk = mm->start_brk;
218 : else
219 0 : min_brk = mm->end_data;
220 : #else
221 : min_brk = mm->start_brk;
222 : #endif
223 0 : if (brk < min_brk)
224 : goto out;
225 :
226 : /*
227 : * Check against rlimit here. If this check is done later after the test
228 : * of oldbrk with newbrk then it can escape the test and let the data
229 : * segment grow beyond its set limit the in case where the limit is
230 : * not page aligned -Ram Gupta
231 : */
232 0 : if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
233 : mm->end_data, mm->start_data))
234 : goto out;
235 :
236 0 : newbrk = PAGE_ALIGN(brk);
237 0 : oldbrk = PAGE_ALIGN(mm->brk);
238 0 : if (oldbrk == newbrk) {
239 0 : mm->brk = brk;
240 0 : goto success;
241 : }
242 :
243 : /*
244 : * Always allow shrinking brk.
245 : * __do_munmap() may downgrade mmap_lock to read.
246 : */
247 0 : if (brk <= mm->brk) {
248 : int ret;
249 :
250 : /*
251 : * mm->brk must to be protected by write mmap_lock so update it
252 : * before downgrading mmap_lock. When __do_munmap() fails,
253 : * mm->brk will be restored from origbrk.
254 : */
255 0 : mm->brk = brk;
256 0 : ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true);
257 0 : if (ret < 0) {
258 0 : mm->brk = origbrk;
259 0 : goto out;
260 0 : } else if (ret == 1) {
261 0 : downgraded = true;
262 : }
263 : goto success;
264 : }
265 :
266 : /* Check against existing mmap mappings. */
267 0 : next = find_vma(mm, oldbrk);
268 0 : if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
269 : goto out;
270 :
271 : /* Ok, looks good - let it rip. */
272 0 : if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0)
273 : goto out;
274 0 : mm->brk = brk;
275 :
276 : success:
277 0 : populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
278 0 : if (downgraded)
279 : mmap_read_unlock(mm);
280 : else
281 : mmap_write_unlock(mm);
282 0 : userfaultfd_unmap_complete(mm, &uf);
283 0 : if (populate)
284 0 : mm_populate(oldbrk, newbrk - oldbrk);
285 0 : return brk;
286 :
287 : out:
288 0 : mmap_write_unlock(mm);
289 0 : return origbrk;
290 : }
291 :
292 : static inline unsigned long vma_compute_gap(struct vm_area_struct *vma)
293 : {
294 : unsigned long gap, prev_end;
295 :
296 : /*
297 : * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
298 : * allow two stack_guard_gaps between them here, and when choosing
299 : * an unmapped area; whereas when expanding we only require one.
300 : * That's a little inconsistent, but keeps the code here simpler.
301 : */
302 0 : gap = vm_start_gap(vma);
303 0 : if (vma->vm_prev) {
304 0 : prev_end = vm_end_gap(vma->vm_prev);
305 0 : if (gap > prev_end)
306 0 : gap -= prev_end;
307 : else
308 : gap = 0;
309 : }
310 : return gap;
311 : }
312 :
313 : #ifdef CONFIG_DEBUG_VM_RB
314 : static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma)
315 : {
316 : unsigned long max = vma_compute_gap(vma), subtree_gap;
317 : if (vma->vm_rb.rb_left) {
318 : subtree_gap = rb_entry(vma->vm_rb.rb_left,
319 : struct vm_area_struct, vm_rb)->rb_subtree_gap;
320 : if (subtree_gap > max)
321 : max = subtree_gap;
322 : }
323 : if (vma->vm_rb.rb_right) {
324 : subtree_gap = rb_entry(vma->vm_rb.rb_right,
325 : struct vm_area_struct, vm_rb)->rb_subtree_gap;
326 : if (subtree_gap > max)
327 : max = subtree_gap;
328 : }
329 : return max;
330 : }
331 :
332 : static int browse_rb(struct mm_struct *mm)
333 : {
334 : struct rb_root *root = &mm->mm_rb;
335 : int i = 0, j, bug = 0;
336 : struct rb_node *nd, *pn = NULL;
337 : unsigned long prev = 0, pend = 0;
338 :
339 : for (nd = rb_first(root); nd; nd = rb_next(nd)) {
340 : struct vm_area_struct *vma;
341 : vma = rb_entry(nd, struct vm_area_struct, vm_rb);
342 : if (vma->vm_start < prev) {
343 : pr_emerg("vm_start %lx < prev %lx\n",
344 : vma->vm_start, prev);
345 : bug = 1;
346 : }
347 : if (vma->vm_start < pend) {
348 : pr_emerg("vm_start %lx < pend %lx\n",
349 : vma->vm_start, pend);
350 : bug = 1;
351 : }
352 : if (vma->vm_start > vma->vm_end) {
353 : pr_emerg("vm_start %lx > vm_end %lx\n",
354 : vma->vm_start, vma->vm_end);
355 : bug = 1;
356 : }
357 : spin_lock(&mm->page_table_lock);
358 : if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
359 : pr_emerg("free gap %lx, correct %lx\n",
360 : vma->rb_subtree_gap,
361 : vma_compute_subtree_gap(vma));
362 : bug = 1;
363 : }
364 : spin_unlock(&mm->page_table_lock);
365 : i++;
366 : pn = nd;
367 : prev = vma->vm_start;
368 : pend = vma->vm_end;
369 : }
370 : j = 0;
371 : for (nd = pn; nd; nd = rb_prev(nd))
372 : j++;
373 : if (i != j) {
374 : pr_emerg("backwards %d, forwards %d\n", j, i);
375 : bug = 1;
376 : }
377 : return bug ? -1 : i;
378 : }
379 :
380 : static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
381 : {
382 : struct rb_node *nd;
383 :
384 : for (nd = rb_first(root); nd; nd = rb_next(nd)) {
385 : struct vm_area_struct *vma;
386 : vma = rb_entry(nd, struct vm_area_struct, vm_rb);
387 : VM_BUG_ON_VMA(vma != ignore &&
388 : vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
389 : vma);
390 : }
391 : }
392 :
393 : static void validate_mm(struct mm_struct *mm)
394 : {
395 : int bug = 0;
396 : int i = 0;
397 : unsigned long highest_address = 0;
398 : struct vm_area_struct *vma = mm->mmap;
399 :
400 : while (vma) {
401 : struct anon_vma *anon_vma = vma->anon_vma;
402 : struct anon_vma_chain *avc;
403 :
404 : if (anon_vma) {
405 : anon_vma_lock_read(anon_vma);
406 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
407 : anon_vma_interval_tree_verify(avc);
408 : anon_vma_unlock_read(anon_vma);
409 : }
410 :
411 : highest_address = vm_end_gap(vma);
412 : vma = vma->vm_next;
413 : i++;
414 : }
415 : if (i != mm->map_count) {
416 : pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
417 : bug = 1;
418 : }
419 : if (highest_address != mm->highest_vm_end) {
420 : pr_emerg("mm->highest_vm_end %lx, found %lx\n",
421 : mm->highest_vm_end, highest_address);
422 : bug = 1;
423 : }
424 : i = browse_rb(mm);
425 : if (i != mm->map_count) {
426 : if (i != -1)
427 : pr_emerg("map_count %d rb %d\n", mm->map_count, i);
428 : bug = 1;
429 : }
430 : VM_BUG_ON_MM(bug, mm);
431 : }
432 : #else
433 : #define validate_mm_rb(root, ignore) do { } while (0)
434 : #define validate_mm(mm) do { } while (0)
435 : #endif
436 :
437 0 : RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks,
438 : struct vm_area_struct, vm_rb,
439 : unsigned long, rb_subtree_gap, vma_compute_gap)
440 :
441 : /*
442 : * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
443 : * vma->vm_prev->vm_end values changed, without modifying the vma's position
444 : * in the rbtree.
445 : */
446 : static void vma_gap_update(struct vm_area_struct *vma)
447 : {
448 : /*
449 : * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created
450 : * a callback function that does exactly what we want.
451 : */
452 0 : vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
453 : }
454 :
455 : static inline void vma_rb_insert(struct vm_area_struct *vma,
456 : struct rb_root *root)
457 : {
458 : /* All rb_subtree_gap values must be consistent prior to insertion */
459 : validate_mm_rb(root, NULL);
460 :
461 0 : rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
462 : }
463 :
464 0 : static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
465 : {
466 : /*
467 : * Note rb_erase_augmented is a fairly large inline function,
468 : * so make sure we instantiate it only once with our desired
469 : * augmented rbtree callbacks.
470 : */
471 0 : rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
472 0 : }
473 :
474 : static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
475 : struct rb_root *root,
476 : struct vm_area_struct *ignore)
477 : {
478 : /*
479 : * All rb_subtree_gap values must be consistent prior to erase,
480 : * with the possible exception of
481 : *
482 : * a. the "next" vma being erased if next->vm_start was reduced in
483 : * __vma_adjust() -> __vma_unlink()
484 : * b. the vma being erased in detach_vmas_to_be_unmapped() ->
485 : * vma_rb_erase()
486 : */
487 : validate_mm_rb(root, ignore);
488 :
489 0 : __vma_rb_erase(vma, root);
490 : }
491 :
492 : static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
493 : struct rb_root *root)
494 : {
495 0 : vma_rb_erase_ignore(vma, root, vma);
496 : }
497 :
498 : /*
499 : * vma has some anon_vma assigned, and is already inserted on that
500 : * anon_vma's interval trees.
501 : *
502 : * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
503 : * vma must be removed from the anon_vma's interval trees using
504 : * anon_vma_interval_tree_pre_update_vma().
505 : *
506 : * After the update, the vma will be reinserted using
507 : * anon_vma_interval_tree_post_update_vma().
508 : *
509 : * The entire update must be protected by exclusive mmap_lock and by
510 : * the root anon_vma's mutex.
511 : */
512 : static inline void
513 0 : anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
514 : {
515 : struct anon_vma_chain *avc;
516 :
517 0 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
518 0 : anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
519 0 : }
520 :
521 : static inline void
522 0 : anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
523 : {
524 : struct anon_vma_chain *avc;
525 :
526 0 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
527 0 : anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
528 0 : }
529 :
530 : static int find_vma_links(struct mm_struct *mm, unsigned long addr,
531 : unsigned long end, struct vm_area_struct **pprev,
532 : struct rb_node ***rb_link, struct rb_node **rb_parent)
533 : {
534 : struct rb_node **__rb_link, *__rb_parent, *rb_prev;
535 :
536 0 : mmap_assert_locked(mm);
537 0 : __rb_link = &mm->mm_rb.rb_node;
538 0 : rb_prev = __rb_parent = NULL;
539 :
540 0 : while (*__rb_link) {
541 : struct vm_area_struct *vma_tmp;
542 :
543 0 : __rb_parent = *__rb_link;
544 0 : vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
545 :
546 0 : if (vma_tmp->vm_end > addr) {
547 : /* Fail if an existing vma overlaps the area */
548 0 : if (vma_tmp->vm_start < end)
549 : return -ENOMEM;
550 0 : __rb_link = &__rb_parent->rb_left;
551 : } else {
552 0 : rb_prev = __rb_parent;
553 0 : __rb_link = &__rb_parent->rb_right;
554 : }
555 : }
556 :
557 0 : *pprev = NULL;
558 0 : if (rb_prev)
559 0 : *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
560 0 : *rb_link = __rb_link;
561 0 : *rb_parent = __rb_parent;
562 : return 0;
563 : }
564 :
565 : /*
566 : * vma_next() - Get the next VMA.
567 : * @mm: The mm_struct.
568 : * @vma: The current vma.
569 : *
570 : * If @vma is NULL, return the first vma in the mm.
571 : *
572 : * Returns: The next VMA after @vma.
573 : */
574 : static inline struct vm_area_struct *vma_next(struct mm_struct *mm,
575 : struct vm_area_struct *vma)
576 : {
577 0 : if (!vma)
578 0 : return mm->mmap;
579 :
580 0 : return vma->vm_next;
581 : }
582 :
583 : /*
584 : * munmap_vma_range() - munmap VMAs that overlap a range.
585 : * @mm: The mm struct
586 : * @start: The start of the range.
587 : * @len: The length of the range.
588 : * @pprev: pointer to the pointer that will be set to previous vm_area_struct
589 : * @rb_link: the rb_node
590 : * @rb_parent: the parent rb_node
591 : *
592 : * Find all the vm_area_struct that overlap from @start to
593 : * @end and munmap them. Set @pprev to the previous vm_area_struct.
594 : *
595 : * Returns: -ENOMEM on munmap failure or 0 on success.
596 : */
597 : static inline int
598 0 : munmap_vma_range(struct mm_struct *mm, unsigned long start, unsigned long len,
599 : struct vm_area_struct **pprev, struct rb_node ***link,
600 : struct rb_node **parent, struct list_head *uf)
601 : {
602 :
603 0 : while (find_vma_links(mm, start, start + len, pprev, link, parent))
604 0 : if (do_munmap(mm, start, len, uf))
605 : return -ENOMEM;
606 :
607 : return 0;
608 : }
609 0 : static unsigned long count_vma_pages_range(struct mm_struct *mm,
610 : unsigned long addr, unsigned long end)
611 : {
612 0 : unsigned long nr_pages = 0;
613 : struct vm_area_struct *vma;
614 :
615 : /* Find first overlapping mapping */
616 0 : vma = find_vma_intersection(mm, addr, end);
617 0 : if (!vma)
618 : return 0;
619 :
620 0 : nr_pages = (min(end, vma->vm_end) -
621 0 : max(addr, vma->vm_start)) >> PAGE_SHIFT;
622 :
623 : /* Iterate over the rest of the overlaps */
624 0 : for (vma = vma->vm_next; vma; vma = vma->vm_next) {
625 : unsigned long overlap_len;
626 :
627 0 : if (vma->vm_start > end)
628 : break;
629 :
630 0 : overlap_len = min(end, vma->vm_end) - vma->vm_start;
631 0 : nr_pages += overlap_len >> PAGE_SHIFT;
632 : }
633 :
634 : return nr_pages;
635 : }
636 :
637 0 : void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
638 : struct rb_node **rb_link, struct rb_node *rb_parent)
639 : {
640 : /* Update tracking information for the gap following the new vma. */
641 0 : if (vma->vm_next)
642 0 : vma_gap_update(vma->vm_next);
643 : else
644 0 : mm->highest_vm_end = vm_end_gap(vma);
645 :
646 : /*
647 : * vma->vm_prev wasn't known when we followed the rbtree to find the
648 : * correct insertion point for that vma. As a result, we could not
649 : * update the vma vm_rb parents rb_subtree_gap values on the way down.
650 : * So, we first insert the vma with a zero rb_subtree_gap value
651 : * (to be consistent with what we did on the way down), and then
652 : * immediately update the gap to the correct value. Finally we
653 : * rebalance the rbtree after all augmented values have been set.
654 : */
655 0 : rb_link_node(&vma->vm_rb, rb_parent, rb_link);
656 0 : vma->rb_subtree_gap = 0;
657 0 : vma_gap_update(vma);
658 0 : vma_rb_insert(vma, &mm->mm_rb);
659 0 : }
660 :
661 0 : static void __vma_link_file(struct vm_area_struct *vma)
662 : {
663 : struct file *file;
664 :
665 0 : file = vma->vm_file;
666 0 : if (file) {
667 0 : struct address_space *mapping = file->f_mapping;
668 :
669 0 : if (vma->vm_flags & VM_SHARED)
670 : mapping_allow_writable(mapping);
671 :
672 0 : flush_dcache_mmap_lock(mapping);
673 0 : vma_interval_tree_insert(vma, &mapping->i_mmap);
674 0 : flush_dcache_mmap_unlock(mapping);
675 : }
676 0 : }
677 :
678 : static void
679 : __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
680 : struct vm_area_struct *prev, struct rb_node **rb_link,
681 : struct rb_node *rb_parent)
682 : {
683 0 : __vma_link_list(mm, vma, prev);
684 0 : __vma_link_rb(mm, vma, rb_link, rb_parent);
685 : }
686 :
687 0 : static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
688 : struct vm_area_struct *prev, struct rb_node **rb_link,
689 : struct rb_node *rb_parent)
690 : {
691 0 : struct address_space *mapping = NULL;
692 :
693 0 : if (vma->vm_file) {
694 0 : mapping = vma->vm_file->f_mapping;
695 : i_mmap_lock_write(mapping);
696 : }
697 :
698 0 : __vma_link(mm, vma, prev, rb_link, rb_parent);
699 0 : __vma_link_file(vma);
700 :
701 0 : if (mapping)
702 : i_mmap_unlock_write(mapping);
703 :
704 0 : mm->map_count++;
705 : validate_mm(mm);
706 0 : }
707 :
708 : /*
709 : * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
710 : * mm's list and rbtree. It has already been inserted into the interval tree.
711 : */
712 0 : static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
713 : {
714 : struct vm_area_struct *prev;
715 : struct rb_node **rb_link, *rb_parent;
716 :
717 0 : if (find_vma_links(mm, vma->vm_start, vma->vm_end,
718 : &prev, &rb_link, &rb_parent))
719 0 : BUG();
720 0 : __vma_link(mm, vma, prev, rb_link, rb_parent);
721 0 : mm->map_count++;
722 0 : }
723 :
724 : static __always_inline void __vma_unlink(struct mm_struct *mm,
725 : struct vm_area_struct *vma,
726 : struct vm_area_struct *ignore)
727 : {
728 0 : vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
729 0 : __vma_unlink_list(mm, vma);
730 : /* Kill the cache */
731 : vmacache_invalidate(mm);
732 : }
733 :
734 : /*
735 : * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
736 : * is already present in an i_mmap tree without adjusting the tree.
737 : * The following helper function should be used when such adjustments
738 : * are necessary. The "insert" vma (if any) is to be inserted
739 : * before we drop the necessary locks.
740 : */
741 0 : int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
742 : unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
743 : struct vm_area_struct *expand)
744 : {
745 0 : struct mm_struct *mm = vma->vm_mm;
746 0 : struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
747 0 : struct address_space *mapping = NULL;
748 0 : struct rb_root_cached *root = NULL;
749 0 : struct anon_vma *anon_vma = NULL;
750 0 : struct file *file = vma->vm_file;
751 0 : bool start_changed = false, end_changed = false;
752 0 : long adjust_next = 0;
753 0 : int remove_next = 0;
754 :
755 0 : if (next && !insert) {
756 0 : struct vm_area_struct *exporter = NULL, *importer = NULL;
757 :
758 0 : if (end >= next->vm_end) {
759 : /*
760 : * vma expands, overlapping all the next, and
761 : * perhaps the one after too (mprotect case 6).
762 : * The only other cases that gets here are
763 : * case 1, case 7 and case 8.
764 : */
765 0 : if (next == expand) {
766 : /*
767 : * The only case where we don't expand "vma"
768 : * and we expand "next" instead is case 8.
769 : */
770 : VM_WARN_ON(end != next->vm_end);
771 : /*
772 : * remove_next == 3 means we're
773 : * removing "vma" and that to do so we
774 : * swapped "vma" and "next".
775 : */
776 : remove_next = 3;
777 : VM_WARN_ON(file != next->vm_file);
778 : swap(vma, next);
779 : } else {
780 : VM_WARN_ON(expand != vma);
781 : /*
782 : * case 1, 6, 7, remove_next == 2 is case 6,
783 : * remove_next == 1 is case 1 or 7.
784 : */
785 0 : remove_next = 1 + (end > next->vm_end);
786 : VM_WARN_ON(remove_next == 2 &&
787 : end != next->vm_next->vm_end);
788 : /* trim end to next, for case 6 first pass */
789 : end = next->vm_end;
790 : }
791 :
792 0 : exporter = next;
793 0 : importer = vma;
794 :
795 : /*
796 : * If next doesn't have anon_vma, import from vma after
797 : * next, if the vma overlaps with it.
798 : */
799 0 : if (remove_next == 2 && !next->anon_vma)
800 0 : exporter = next->vm_next;
801 :
802 0 : } else if (end > next->vm_start) {
803 : /*
804 : * vma expands, overlapping part of the next:
805 : * mprotect case 5 shifting the boundary up.
806 : */
807 0 : adjust_next = (end - next->vm_start);
808 0 : exporter = next;
809 0 : importer = vma;
810 : VM_WARN_ON(expand != importer);
811 0 : } else if (end < vma->vm_end) {
812 : /*
813 : * vma shrinks, and !insert tells it's not
814 : * split_vma inserting another: so it must be
815 : * mprotect case 4 shifting the boundary down.
816 : */
817 0 : adjust_next = -(vma->vm_end - end);
818 0 : exporter = vma;
819 0 : importer = next;
820 : VM_WARN_ON(expand != importer);
821 : }
822 :
823 : /*
824 : * Easily overlooked: when mprotect shifts the boundary,
825 : * make sure the expanding vma has anon_vma set if the
826 : * shrinking vma had, to cover any anon pages imported.
827 : */
828 0 : if (exporter && exporter->anon_vma && !importer->anon_vma) {
829 : int error;
830 :
831 0 : importer->anon_vma = exporter->anon_vma;
832 0 : error = anon_vma_clone(importer, exporter);
833 0 : if (error)
834 : return error;
835 : }
836 : }
837 : again:
838 0 : vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
839 :
840 0 : if (file) {
841 0 : mapping = file->f_mapping;
842 0 : root = &mapping->i_mmap;
843 0 : uprobe_munmap(vma, vma->vm_start, vma->vm_end);
844 :
845 : if (adjust_next)
846 : uprobe_munmap(next, next->vm_start, next->vm_end);
847 :
848 0 : i_mmap_lock_write(mapping);
849 0 : if (insert) {
850 : /*
851 : * Put into interval tree now, so instantiated pages
852 : * are visible to arm/parisc __flush_dcache_page
853 : * throughout; but we cannot insert into address
854 : * space until vma start or end is updated.
855 : */
856 0 : __vma_link_file(insert);
857 : }
858 : }
859 :
860 0 : anon_vma = vma->anon_vma;
861 0 : if (!anon_vma && adjust_next)
862 0 : anon_vma = next->anon_vma;
863 0 : if (anon_vma) {
864 : VM_WARN_ON(adjust_next && next->anon_vma &&
865 : anon_vma != next->anon_vma);
866 0 : anon_vma_lock_write(anon_vma);
867 0 : anon_vma_interval_tree_pre_update_vma(vma);
868 0 : if (adjust_next)
869 0 : anon_vma_interval_tree_pre_update_vma(next);
870 : }
871 :
872 0 : if (file) {
873 0 : flush_dcache_mmap_lock(mapping);
874 0 : vma_interval_tree_remove(vma, root);
875 0 : if (adjust_next)
876 0 : vma_interval_tree_remove(next, root);
877 : }
878 :
879 0 : if (start != vma->vm_start) {
880 0 : vma->vm_start = start;
881 0 : start_changed = true;
882 : }
883 0 : if (end != vma->vm_end) {
884 0 : vma->vm_end = end;
885 0 : end_changed = true;
886 : }
887 0 : vma->vm_pgoff = pgoff;
888 0 : if (adjust_next) {
889 0 : next->vm_start += adjust_next;
890 0 : next->vm_pgoff += adjust_next >> PAGE_SHIFT;
891 : }
892 :
893 0 : if (file) {
894 0 : if (adjust_next)
895 0 : vma_interval_tree_insert(next, root);
896 0 : vma_interval_tree_insert(vma, root);
897 0 : flush_dcache_mmap_unlock(mapping);
898 : }
899 :
900 0 : if (remove_next) {
901 : /*
902 : * vma_merge has merged next into vma, and needs
903 : * us to remove next before dropping the locks.
904 : */
905 0 : if (remove_next != 3)
906 : __vma_unlink(mm, next, next);
907 : else
908 : /*
909 : * vma is not before next if they've been
910 : * swapped.
911 : *
912 : * pre-swap() next->vm_start was reduced so
913 : * tell validate_mm_rb to ignore pre-swap()
914 : * "next" (which is stored in post-swap()
915 : * "vma").
916 : */
917 : __vma_unlink(mm, next, vma);
918 0 : if (file)
919 0 : __remove_shared_vm_struct(next, file, mapping);
920 0 : } else if (insert) {
921 : /*
922 : * split_vma has split insert from vma, and needs
923 : * us to insert it before dropping the locks
924 : * (it may either follow vma or precede it).
925 : */
926 0 : __insert_vm_struct(mm, insert);
927 : } else {
928 0 : if (start_changed)
929 : vma_gap_update(vma);
930 0 : if (end_changed) {
931 0 : if (!next)
932 0 : mm->highest_vm_end = vm_end_gap(vma);
933 0 : else if (!adjust_next)
934 : vma_gap_update(next);
935 : }
936 : }
937 :
938 0 : if (anon_vma) {
939 0 : anon_vma_interval_tree_post_update_vma(vma);
940 0 : if (adjust_next)
941 0 : anon_vma_interval_tree_post_update_vma(next);
942 0 : anon_vma_unlock_write(anon_vma);
943 : }
944 :
945 0 : if (file) {
946 0 : i_mmap_unlock_write(mapping);
947 0 : uprobe_mmap(vma);
948 :
949 : if (adjust_next)
950 : uprobe_mmap(next);
951 : }
952 :
953 0 : if (remove_next) {
954 0 : if (file) {
955 0 : uprobe_munmap(next, next->vm_start, next->vm_end);
956 0 : fput(file);
957 : }
958 0 : if (next->anon_vma)
959 0 : anon_vma_merge(vma, next);
960 0 : mm->map_count--;
961 0 : mpol_put(vma_policy(next));
962 0 : vm_area_free(next);
963 : /*
964 : * In mprotect's case 6 (see comments on vma_merge),
965 : * we must remove another next too. It would clutter
966 : * up the code too much to do both in one go.
967 : */
968 0 : if (remove_next != 3) {
969 : /*
970 : * If "next" was removed and vma->vm_end was
971 : * expanded (up) over it, in turn
972 : * "next->vm_prev->vm_end" changed and the
973 : * "vma->vm_next" gap must be updated.
974 : */
975 0 : next = vma->vm_next;
976 : } else {
977 : /*
978 : * For the scope of the comment "next" and
979 : * "vma" considered pre-swap(): if "vma" was
980 : * removed, next->vm_start was expanded (down)
981 : * over it and the "next" gap must be updated.
982 : * Because of the swap() the post-swap() "vma"
983 : * actually points to pre-swap() "next"
984 : * (post-swap() "next" as opposed is now a
985 : * dangling pointer).
986 : */
987 : next = vma;
988 : }
989 0 : if (remove_next == 2) {
990 0 : remove_next = 1;
991 0 : end = next->vm_end;
992 0 : goto again;
993 : }
994 0 : else if (next)
995 : vma_gap_update(next);
996 : else {
997 : /*
998 : * If remove_next == 2 we obviously can't
999 : * reach this path.
1000 : *
1001 : * If remove_next == 3 we can't reach this
1002 : * path because pre-swap() next is always not
1003 : * NULL. pre-swap() "next" is not being
1004 : * removed and its next->vm_end is not altered
1005 : * (and furthermore "end" already matches
1006 : * next->vm_end in remove_next == 3).
1007 : *
1008 : * We reach this only in the remove_next == 1
1009 : * case if the "next" vma that was removed was
1010 : * the highest vma of the mm. However in such
1011 : * case next->vm_end == "end" and the extended
1012 : * "vma" has vma->vm_end == next->vm_end so
1013 : * mm->highest_vm_end doesn't need any update
1014 : * in remove_next == 1 case.
1015 : */
1016 : VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
1017 : }
1018 : }
1019 : if (insert && file)
1020 : uprobe_mmap(insert);
1021 :
1022 : validate_mm(mm);
1023 :
1024 0 : return 0;
1025 : }
1026 :
1027 : /*
1028 : * If the vma has a ->close operation then the driver probably needs to release
1029 : * per-vma resources, so we don't attempt to merge those.
1030 : */
1031 : static inline int is_mergeable_vma(struct vm_area_struct *vma,
1032 : struct file *file, unsigned long vm_flags,
1033 : struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1034 : struct anon_vma_name *anon_name)
1035 : {
1036 : /*
1037 : * VM_SOFTDIRTY should not prevent from VMA merging, if we
1038 : * match the flags but dirty bit -- the caller should mark
1039 : * merged VMA as dirty. If dirty bit won't be excluded from
1040 : * comparison, we increase pressure on the memory system forcing
1041 : * the kernel to generate new VMAs when old one could be
1042 : * extended instead.
1043 : */
1044 0 : if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
1045 : return 0;
1046 0 : if (vma->vm_file != file)
1047 : return 0;
1048 0 : if (vma->vm_ops && vma->vm_ops->close)
1049 : return 0;
1050 0 : if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
1051 : return 0;
1052 0 : if (!anon_vma_name_eq(anon_vma_name(vma), anon_name))
1053 : return 0;
1054 : return 1;
1055 : }
1056 :
1057 : static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
1058 : struct anon_vma *anon_vma2,
1059 : struct vm_area_struct *vma)
1060 : {
1061 : /*
1062 : * The list_is_singular() test is to avoid merging VMA cloned from
1063 : * parents. This can improve scalability caused by anon_vma lock.
1064 : */
1065 0 : if ((!anon_vma1 || !anon_vma2) && (!vma ||
1066 0 : list_is_singular(&vma->anon_vma_chain)))
1067 : return 1;
1068 0 : return anon_vma1 == anon_vma2;
1069 : }
1070 :
1071 : /*
1072 : * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1073 : * in front of (at a lower virtual address and file offset than) the vma.
1074 : *
1075 : * We cannot merge two vmas if they have differently assigned (non-NULL)
1076 : * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1077 : *
1078 : * We don't check here for the merged mmap wrapping around the end of pagecache
1079 : * indices (16TB on ia32) because do_mmap() does not permit mmap's which
1080 : * wrap, nor mmaps which cover the final page at index -1UL.
1081 : */
1082 : static int
1083 0 : can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
1084 : struct anon_vma *anon_vma, struct file *file,
1085 : pgoff_t vm_pgoff,
1086 : struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1087 : struct anon_vma_name *anon_name)
1088 : {
1089 0 : if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
1090 0 : is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1091 0 : if (vma->vm_pgoff == vm_pgoff)
1092 : return 1;
1093 : }
1094 : return 0;
1095 : }
1096 :
1097 : /*
1098 : * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1099 : * beyond (at a higher virtual address and file offset than) the vma.
1100 : *
1101 : * We cannot merge two vmas if they have differently assigned (non-NULL)
1102 : * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1103 : */
1104 : static int
1105 0 : can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
1106 : struct anon_vma *anon_vma, struct file *file,
1107 : pgoff_t vm_pgoff,
1108 : struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1109 : struct anon_vma_name *anon_name)
1110 : {
1111 0 : if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) &&
1112 0 : is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1113 : pgoff_t vm_pglen;
1114 0 : vm_pglen = vma_pages(vma);
1115 0 : if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1116 : return 1;
1117 : }
1118 : return 0;
1119 : }
1120 :
1121 : /*
1122 : * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name),
1123 : * figure out whether that can be merged with its predecessor or its
1124 : * successor. Or both (it neatly fills a hole).
1125 : *
1126 : * In most cases - when called for mmap, brk or mremap - [addr,end) is
1127 : * certain not to be mapped by the time vma_merge is called; but when
1128 : * called for mprotect, it is certain to be already mapped (either at
1129 : * an offset within prev, or at the start of next), and the flags of
1130 : * this area are about to be changed to vm_flags - and the no-change
1131 : * case has already been eliminated.
1132 : *
1133 : * The following mprotect cases have to be considered, where AAAA is
1134 : * the area passed down from mprotect_fixup, never extending beyond one
1135 : * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1136 : *
1137 : * AAAA AAAA AAAA
1138 : * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN
1139 : * cannot merge might become might become
1140 : * PPNNNNNNNNNN PPPPPPPPPPNN
1141 : * mmap, brk or case 4 below case 5 below
1142 : * mremap move:
1143 : * AAAA AAAA
1144 : * PPPP NNNN PPPPNNNNXXXX
1145 : * might become might become
1146 : * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or
1147 : * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or
1148 : * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8
1149 : *
1150 : * It is important for case 8 that the vma NNNN overlapping the
1151 : * region AAAA is never going to extended over XXXX. Instead XXXX must
1152 : * be extended in region AAAA and NNNN must be removed. This way in
1153 : * all cases where vma_merge succeeds, the moment vma_adjust drops the
1154 : * rmap_locks, the properties of the merged vma will be already
1155 : * correct for the whole merged range. Some of those properties like
1156 : * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1157 : * be correct for the whole merged range immediately after the
1158 : * rmap_locks are released. Otherwise if XXXX would be removed and
1159 : * NNNN would be extended over the XXXX range, remove_migration_ptes
1160 : * or other rmap walkers (if working on addresses beyond the "end"
1161 : * parameter) may establish ptes with the wrong permissions of NNNN
1162 : * instead of the right permissions of XXXX.
1163 : */
1164 0 : struct vm_area_struct *vma_merge(struct mm_struct *mm,
1165 : struct vm_area_struct *prev, unsigned long addr,
1166 : unsigned long end, unsigned long vm_flags,
1167 : struct anon_vma *anon_vma, struct file *file,
1168 : pgoff_t pgoff, struct mempolicy *policy,
1169 : struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
1170 : struct anon_vma_name *anon_name)
1171 : {
1172 0 : pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1173 : struct vm_area_struct *area, *next;
1174 : int err;
1175 :
1176 : /*
1177 : * We later require that vma->vm_flags == vm_flags,
1178 : * so this tests vma->vm_flags & VM_SPECIAL, too.
1179 : */
1180 0 : if (vm_flags & VM_SPECIAL)
1181 : return NULL;
1182 :
1183 0 : next = vma_next(mm, prev);
1184 0 : area = next;
1185 0 : if (area && area->vm_end == end) /* cases 6, 7, 8 */
1186 0 : next = next->vm_next;
1187 :
1188 : /* verify some invariant that must be enforced by the caller */
1189 : VM_WARN_ON(prev && addr <= prev->vm_start);
1190 : VM_WARN_ON(area && end > area->vm_end);
1191 : VM_WARN_ON(addr >= end);
1192 :
1193 : /*
1194 : * Can it merge with the predecessor?
1195 : */
1196 0 : if (prev && prev->vm_end == addr &&
1197 0 : mpol_equal(vma_policy(prev), policy) &&
1198 0 : can_vma_merge_after(prev, vm_flags,
1199 : anon_vma, file, pgoff,
1200 : vm_userfaultfd_ctx, anon_name)) {
1201 : /*
1202 : * OK, it can. Can we now merge in the successor as well?
1203 : */
1204 0 : if (next && end == next->vm_start &&
1205 0 : mpol_equal(policy, vma_policy(next)) &&
1206 0 : can_vma_merge_before(next, vm_flags,
1207 : anon_vma, file,
1208 : pgoff+pglen,
1209 0 : vm_userfaultfd_ctx, anon_name) &&
1210 0 : is_mergeable_anon_vma(prev->anon_vma,
1211 : next->anon_vma, NULL)) {
1212 : /* cases 1, 6 */
1213 0 : err = __vma_adjust(prev, prev->vm_start,
1214 : next->vm_end, prev->vm_pgoff, NULL,
1215 : prev);
1216 : } else /* cases 2, 5, 7 */
1217 0 : err = __vma_adjust(prev, prev->vm_start,
1218 : end, prev->vm_pgoff, NULL, prev);
1219 0 : if (err)
1220 : return NULL;
1221 0 : khugepaged_enter_vma_merge(prev, vm_flags);
1222 0 : return prev;
1223 : }
1224 :
1225 : /*
1226 : * Can this new request be merged in front of next?
1227 : */
1228 0 : if (next && end == next->vm_start &&
1229 0 : mpol_equal(policy, vma_policy(next)) &&
1230 0 : can_vma_merge_before(next, vm_flags,
1231 : anon_vma, file, pgoff+pglen,
1232 : vm_userfaultfd_ctx, anon_name)) {
1233 0 : if (prev && addr < prev->vm_end) /* case 4 */
1234 0 : err = __vma_adjust(prev, prev->vm_start,
1235 : addr, prev->vm_pgoff, NULL, next);
1236 : else { /* cases 3, 8 */
1237 0 : err = __vma_adjust(area, addr, next->vm_end,
1238 0 : next->vm_pgoff - pglen, NULL, next);
1239 : /*
1240 : * In case 3 area is already equal to next and
1241 : * this is a noop, but in case 8 "area" has
1242 : * been removed and next was expanded over it.
1243 : */
1244 0 : area = next;
1245 : }
1246 0 : if (err)
1247 : return NULL;
1248 0 : khugepaged_enter_vma_merge(area, vm_flags);
1249 0 : return area;
1250 : }
1251 :
1252 : return NULL;
1253 : }
1254 :
1255 : /*
1256 : * Rough compatibility check to quickly see if it's even worth looking
1257 : * at sharing an anon_vma.
1258 : *
1259 : * They need to have the same vm_file, and the flags can only differ
1260 : * in things that mprotect may change.
1261 : *
1262 : * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1263 : * we can merge the two vma's. For example, we refuse to merge a vma if
1264 : * there is a vm_ops->close() function, because that indicates that the
1265 : * driver is doing some kind of reference counting. But that doesn't
1266 : * really matter for the anon_vma sharing case.
1267 : */
1268 : static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1269 : {
1270 0 : return a->vm_end == b->vm_start &&
1271 0 : mpol_equal(vma_policy(a), vma_policy(b)) &&
1272 0 : a->vm_file == b->vm_file &&
1273 0 : !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) &&
1274 0 : b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1275 : }
1276 :
1277 : /*
1278 : * Do some basic sanity checking to see if we can re-use the anon_vma
1279 : * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1280 : * the same as 'old', the other will be the new one that is trying
1281 : * to share the anon_vma.
1282 : *
1283 : * NOTE! This runs with mm_sem held for reading, so it is possible that
1284 : * the anon_vma of 'old' is concurrently in the process of being set up
1285 : * by another page fault trying to merge _that_. But that's ok: if it
1286 : * is being set up, that automatically means that it will be a singleton
1287 : * acceptable for merging, so we can do all of this optimistically. But
1288 : * we do that READ_ONCE() to make sure that we never re-load the pointer.
1289 : *
1290 : * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1291 : * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1292 : * is to return an anon_vma that is "complex" due to having gone through
1293 : * a fork).
1294 : *
1295 : * We also make sure that the two vma's are compatible (adjacent,
1296 : * and with the same memory policies). That's all stable, even with just
1297 : * a read lock on the mm_sem.
1298 : */
1299 0 : static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1300 : {
1301 0 : if (anon_vma_compatible(a, b)) {
1302 0 : struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1303 :
1304 0 : if (anon_vma && list_is_singular(&old->anon_vma_chain))
1305 : return anon_vma;
1306 : }
1307 : return NULL;
1308 : }
1309 :
1310 : /*
1311 : * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1312 : * neighbouring vmas for a suitable anon_vma, before it goes off
1313 : * to allocate a new anon_vma. It checks because a repetitive
1314 : * sequence of mprotects and faults may otherwise lead to distinct
1315 : * anon_vmas being allocated, preventing vma merge in subsequent
1316 : * mprotect.
1317 : */
1318 0 : struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1319 : {
1320 0 : struct anon_vma *anon_vma = NULL;
1321 :
1322 : /* Try next first. */
1323 0 : if (vma->vm_next) {
1324 0 : anon_vma = reusable_anon_vma(vma->vm_next, vma, vma->vm_next);
1325 0 : if (anon_vma)
1326 : return anon_vma;
1327 : }
1328 :
1329 : /* Try prev next. */
1330 0 : if (vma->vm_prev)
1331 0 : anon_vma = reusable_anon_vma(vma->vm_prev, vma->vm_prev, vma);
1332 :
1333 : /*
1334 : * We might reach here with anon_vma == NULL if we can't find
1335 : * any reusable anon_vma.
1336 : * There's no absolute need to look only at touching neighbours:
1337 : * we could search further afield for "compatible" anon_vmas.
1338 : * But it would probably just be a waste of time searching,
1339 : * or lead to too many vmas hanging off the same anon_vma.
1340 : * We're trying to allow mprotect remerging later on,
1341 : * not trying to minimize memory used for anon_vmas.
1342 : */
1343 : return anon_vma;
1344 : }
1345 :
1346 : /*
1347 : * If a hint addr is less than mmap_min_addr change hint to be as
1348 : * low as possible but still greater than mmap_min_addr
1349 : */
1350 : static inline unsigned long round_hint_to_min(unsigned long hint)
1351 : {
1352 0 : hint &= PAGE_MASK;
1353 0 : if (((void *)hint != NULL) &&
1354 0 : (hint < mmap_min_addr))
1355 0 : return PAGE_ALIGN(mmap_min_addr);
1356 : return hint;
1357 : }
1358 :
1359 0 : int mlock_future_check(struct mm_struct *mm, unsigned long flags,
1360 : unsigned long len)
1361 : {
1362 : unsigned long locked, lock_limit;
1363 :
1364 : /* mlock MCL_FUTURE? */
1365 0 : if (flags & VM_LOCKED) {
1366 0 : locked = len >> PAGE_SHIFT;
1367 0 : locked += mm->locked_vm;
1368 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
1369 0 : lock_limit >>= PAGE_SHIFT;
1370 0 : if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1371 : return -EAGAIN;
1372 : }
1373 : return 0;
1374 : }
1375 :
1376 : static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1377 : {
1378 0 : if (S_ISREG(inode->i_mode))
1379 : return MAX_LFS_FILESIZE;
1380 :
1381 0 : if (S_ISBLK(inode->i_mode))
1382 : return MAX_LFS_FILESIZE;
1383 :
1384 0 : if (S_ISSOCK(inode->i_mode))
1385 : return MAX_LFS_FILESIZE;
1386 :
1387 : /* Special "we do even unsigned file positions" case */
1388 0 : if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1389 : return 0;
1390 :
1391 : /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1392 : return ULONG_MAX;
1393 : }
1394 :
1395 : static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1396 : unsigned long pgoff, unsigned long len)
1397 : {
1398 0 : u64 maxsize = file_mmap_size_max(file, inode);
1399 :
1400 0 : if (maxsize && len > maxsize)
1401 : return false;
1402 0 : maxsize -= len;
1403 0 : if (pgoff > maxsize >> PAGE_SHIFT)
1404 : return false;
1405 : return true;
1406 : }
1407 :
1408 : /*
1409 : * The caller must write-lock current->mm->mmap_lock.
1410 : */
1411 0 : unsigned long do_mmap(struct file *file, unsigned long addr,
1412 : unsigned long len, unsigned long prot,
1413 : unsigned long flags, unsigned long pgoff,
1414 : unsigned long *populate, struct list_head *uf)
1415 : {
1416 0 : struct mm_struct *mm = current->mm;
1417 : vm_flags_t vm_flags;
1418 0 : int pkey = 0;
1419 :
1420 0 : *populate = 0;
1421 :
1422 0 : if (!len)
1423 : return -EINVAL;
1424 :
1425 : /*
1426 : * Does the application expect PROT_READ to imply PROT_EXEC?
1427 : *
1428 : * (the exception is when the underlying filesystem is noexec
1429 : * mounted, in which case we dont add PROT_EXEC.)
1430 : */
1431 0 : if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1432 0 : if (!(file && path_noexec(&file->f_path)))
1433 0 : prot |= PROT_EXEC;
1434 :
1435 : /* force arch specific MAP_FIXED handling in get_unmapped_area */
1436 0 : if (flags & MAP_FIXED_NOREPLACE)
1437 0 : flags |= MAP_FIXED;
1438 :
1439 0 : if (!(flags & MAP_FIXED))
1440 : addr = round_hint_to_min(addr);
1441 :
1442 : /* Careful about overflows.. */
1443 0 : len = PAGE_ALIGN(len);
1444 0 : if (!len)
1445 : return -ENOMEM;
1446 :
1447 : /* offset overflow? */
1448 0 : if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1449 : return -EOVERFLOW;
1450 :
1451 : /* Too many mappings? */
1452 0 : if (mm->map_count > sysctl_max_map_count)
1453 : return -ENOMEM;
1454 :
1455 : /* Obtain the address to map to. we verify (or select) it and ensure
1456 : * that it represents a valid section of the address space.
1457 : */
1458 0 : addr = get_unmapped_area(file, addr, len, pgoff, flags);
1459 0 : if (IS_ERR_VALUE(addr))
1460 : return addr;
1461 :
1462 0 : if (flags & MAP_FIXED_NOREPLACE) {
1463 0 : if (find_vma_intersection(mm, addr, addr + len))
1464 : return -EEXIST;
1465 : }
1466 :
1467 : if (prot == PROT_EXEC) {
1468 : pkey = execute_only_pkey(mm);
1469 : if (pkey < 0)
1470 : pkey = 0;
1471 : }
1472 :
1473 : /* Do simple checking here so the lower-level routines won't have
1474 : * to. we assume access permissions have been handled by the open
1475 : * of the memory object, so we don't do any here.
1476 : */
1477 0 : vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1478 0 : mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1479 :
1480 0 : if (flags & MAP_LOCKED)
1481 0 : if (!can_do_mlock())
1482 : return -EPERM;
1483 :
1484 0 : if (mlock_future_check(mm, vm_flags, len))
1485 : return -EAGAIN;
1486 :
1487 0 : if (file) {
1488 0 : struct inode *inode = file_inode(file);
1489 : unsigned long flags_mask;
1490 :
1491 0 : if (!file_mmap_ok(file, inode, pgoff, len))
1492 : return -EOVERFLOW;
1493 :
1494 0 : flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1495 :
1496 0 : switch (flags & MAP_TYPE) {
1497 : case MAP_SHARED:
1498 : /*
1499 : * Force use of MAP_SHARED_VALIDATE with non-legacy
1500 : * flags. E.g. MAP_SYNC is dangerous to use with
1501 : * MAP_SHARED as you don't know which consistency model
1502 : * you will get. We silently ignore unsupported flags
1503 : * with MAP_SHARED to preserve backward compatibility.
1504 : */
1505 0 : flags &= LEGACY_MAP_MASK;
1506 : fallthrough;
1507 : case MAP_SHARED_VALIDATE:
1508 0 : if (flags & ~flags_mask)
1509 : return -EOPNOTSUPP;
1510 0 : if (prot & PROT_WRITE) {
1511 0 : if (!(file->f_mode & FMODE_WRITE))
1512 : return -EACCES;
1513 0 : if (IS_SWAPFILE(file->f_mapping->host))
1514 : return -ETXTBSY;
1515 : }
1516 :
1517 : /*
1518 : * Make sure we don't allow writing to an append-only
1519 : * file..
1520 : */
1521 0 : if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1522 : return -EACCES;
1523 :
1524 0 : vm_flags |= VM_SHARED | VM_MAYSHARE;
1525 0 : if (!(file->f_mode & FMODE_WRITE))
1526 0 : vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1527 : fallthrough;
1528 : case MAP_PRIVATE:
1529 0 : if (!(file->f_mode & FMODE_READ))
1530 : return -EACCES;
1531 0 : if (path_noexec(&file->f_path)) {
1532 0 : if (vm_flags & VM_EXEC)
1533 : return -EPERM;
1534 0 : vm_flags &= ~VM_MAYEXEC;
1535 : }
1536 :
1537 0 : if (!file->f_op->mmap)
1538 : return -ENODEV;
1539 0 : if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1540 : return -EINVAL;
1541 : break;
1542 :
1543 : default:
1544 : return -EINVAL;
1545 : }
1546 : } else {
1547 0 : switch (flags & MAP_TYPE) {
1548 : case MAP_SHARED:
1549 0 : if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1550 : return -EINVAL;
1551 : /*
1552 : * Ignore pgoff.
1553 : */
1554 0 : pgoff = 0;
1555 0 : vm_flags |= VM_SHARED | VM_MAYSHARE;
1556 0 : break;
1557 : case MAP_PRIVATE:
1558 : /*
1559 : * Set pgoff according to addr for anon_vma.
1560 : */
1561 0 : pgoff = addr >> PAGE_SHIFT;
1562 0 : break;
1563 : default:
1564 : return -EINVAL;
1565 : }
1566 : }
1567 :
1568 : /*
1569 : * Set 'VM_NORESERVE' if we should not account for the
1570 : * memory use of this mapping.
1571 : */
1572 0 : if (flags & MAP_NORESERVE) {
1573 : /* We honor MAP_NORESERVE if allowed to overcommit */
1574 0 : if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1575 0 : vm_flags |= VM_NORESERVE;
1576 :
1577 : /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1578 : if (file && is_file_hugepages(file))
1579 : vm_flags |= VM_NORESERVE;
1580 : }
1581 :
1582 0 : addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1583 0 : if (!IS_ERR_VALUE(addr) &&
1584 0 : ((vm_flags & VM_LOCKED) ||
1585 0 : (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1586 0 : *populate = len;
1587 : return addr;
1588 : }
1589 :
1590 0 : unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1591 : unsigned long prot, unsigned long flags,
1592 : unsigned long fd, unsigned long pgoff)
1593 : {
1594 0 : struct file *file = NULL;
1595 : unsigned long retval;
1596 :
1597 0 : if (!(flags & MAP_ANONYMOUS)) {
1598 0 : audit_mmap_fd(fd, flags);
1599 0 : file = fget(fd);
1600 0 : if (!file)
1601 : return -EBADF;
1602 : if (is_file_hugepages(file)) {
1603 : len = ALIGN(len, huge_page_size(hstate_file(file)));
1604 0 : } else if (unlikely(flags & MAP_HUGETLB)) {
1605 : retval = -EINVAL;
1606 : goto out_fput;
1607 : }
1608 0 : } else if (flags & MAP_HUGETLB) {
1609 : struct hstate *hs;
1610 :
1611 : hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1612 : if (!hs)
1613 : return -EINVAL;
1614 :
1615 : len = ALIGN(len, huge_page_size(hs));
1616 : /*
1617 : * VM_NORESERVE is used because the reservations will be
1618 : * taken when vm_ops->mmap() is called
1619 : */
1620 : file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1621 : VM_NORESERVE,
1622 : HUGETLB_ANONHUGE_INODE,
1623 : (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1624 : if (IS_ERR(file))
1625 : return PTR_ERR(file);
1626 : }
1627 :
1628 0 : retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1629 : out_fput:
1630 0 : if (file)
1631 0 : fput(file);
1632 : return retval;
1633 : }
1634 :
1635 0 : SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1636 : unsigned long, prot, unsigned long, flags,
1637 : unsigned long, fd, unsigned long, pgoff)
1638 : {
1639 0 : return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1640 : }
1641 :
1642 : #ifdef __ARCH_WANT_SYS_OLD_MMAP
1643 : struct mmap_arg_struct {
1644 : unsigned long addr;
1645 : unsigned long len;
1646 : unsigned long prot;
1647 : unsigned long flags;
1648 : unsigned long fd;
1649 : unsigned long offset;
1650 : };
1651 :
1652 : SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1653 : {
1654 : struct mmap_arg_struct a;
1655 :
1656 : if (copy_from_user(&a, arg, sizeof(a)))
1657 : return -EFAULT;
1658 : if (offset_in_page(a.offset))
1659 : return -EINVAL;
1660 :
1661 : return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1662 : a.offset >> PAGE_SHIFT);
1663 : }
1664 : #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1665 :
1666 : /*
1667 : * Some shared mappings will want the pages marked read-only
1668 : * to track write events. If so, we'll downgrade vm_page_prot
1669 : * to the private version (using protection_map[] without the
1670 : * VM_SHARED bit).
1671 : */
1672 0 : int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
1673 : {
1674 0 : vm_flags_t vm_flags = vma->vm_flags;
1675 0 : const struct vm_operations_struct *vm_ops = vma->vm_ops;
1676 :
1677 : /* If it was private or non-writable, the write bit is already clear */
1678 0 : if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1679 : return 0;
1680 :
1681 : /* The backer wishes to know when pages are first written to? */
1682 0 : if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1683 : return 1;
1684 :
1685 : /* The open routine did something to the protections that pgprot_modify
1686 : * won't preserve? */
1687 0 : if (pgprot_val(vm_page_prot) !=
1688 0 : pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
1689 : return 0;
1690 :
1691 : /* Do we need to track softdirty? */
1692 : if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1693 : return 1;
1694 :
1695 : /* Specialty mapping? */
1696 0 : if (vm_flags & VM_PFNMAP)
1697 : return 0;
1698 :
1699 : /* Can the mapping track the dirty pages? */
1700 0 : return vma->vm_file && vma->vm_file->f_mapping &&
1701 0 : mapping_can_writeback(vma->vm_file->f_mapping);
1702 : }
1703 :
1704 : /*
1705 : * We account for memory if it's a private writeable mapping,
1706 : * not hugepages and VM_NORESERVE wasn't set.
1707 : */
1708 : static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1709 : {
1710 : /*
1711 : * hugetlb has its own accounting separate from the core VM
1712 : * VM_HUGETLB may not be set yet so we cannot check for that flag.
1713 : */
1714 : if (file && is_file_hugepages(file))
1715 : return 0;
1716 :
1717 0 : return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1718 : }
1719 :
1720 0 : unsigned long mmap_region(struct file *file, unsigned long addr,
1721 : unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1722 : struct list_head *uf)
1723 : {
1724 0 : struct mm_struct *mm = current->mm;
1725 : struct vm_area_struct *vma, *prev, *merge;
1726 : int error;
1727 : struct rb_node **rb_link, *rb_parent;
1728 0 : unsigned long charged = 0;
1729 :
1730 : /* Check against address space limit. */
1731 0 : if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
1732 : unsigned long nr_pages;
1733 :
1734 : /*
1735 : * MAP_FIXED may remove pages of mappings that intersects with
1736 : * requested mapping. Account for the pages it would unmap.
1737 : */
1738 0 : nr_pages = count_vma_pages_range(mm, addr, addr + len);
1739 :
1740 0 : if (!may_expand_vm(mm, vm_flags,
1741 : (len >> PAGE_SHIFT) - nr_pages))
1742 : return -ENOMEM;
1743 : }
1744 :
1745 : /* Clear old maps, set up prev, rb_link, rb_parent, and uf */
1746 0 : if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf))
1747 : return -ENOMEM;
1748 : /*
1749 : * Private writable mapping: check memory availability
1750 : */
1751 0 : if (accountable_mapping(file, vm_flags)) {
1752 0 : charged = len >> PAGE_SHIFT;
1753 0 : if (security_vm_enough_memory_mm(mm, charged))
1754 : return -ENOMEM;
1755 0 : vm_flags |= VM_ACCOUNT;
1756 : }
1757 :
1758 : /*
1759 : * Can we just expand an old mapping?
1760 : */
1761 0 : vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1762 : NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
1763 0 : if (vma)
1764 : goto out;
1765 :
1766 : /*
1767 : * Determine the object being mapped and call the appropriate
1768 : * specific mapper. the address has already been validated, but
1769 : * not unmapped, but the maps are removed from the list.
1770 : */
1771 0 : vma = vm_area_alloc(mm);
1772 0 : if (!vma) {
1773 : error = -ENOMEM;
1774 : goto unacct_error;
1775 : }
1776 :
1777 0 : vma->vm_start = addr;
1778 0 : vma->vm_end = addr + len;
1779 0 : vma->vm_flags = vm_flags;
1780 0 : vma->vm_page_prot = vm_get_page_prot(vm_flags);
1781 0 : vma->vm_pgoff = pgoff;
1782 :
1783 0 : if (file) {
1784 0 : if (vm_flags & VM_SHARED) {
1785 0 : error = mapping_map_writable(file->f_mapping);
1786 0 : if (error)
1787 : goto free_vma;
1788 : }
1789 :
1790 0 : vma->vm_file = get_file(file);
1791 0 : error = call_mmap(file, vma);
1792 0 : if (error)
1793 : goto unmap_and_free_vma;
1794 :
1795 : /* Can addr have changed??
1796 : *
1797 : * Answer: Yes, several device drivers can do it in their
1798 : * f_op->mmap method. -DaveM
1799 : * Bug: If addr is changed, prev, rb_link, rb_parent should
1800 : * be updated for vma_link()
1801 : */
1802 0 : WARN_ON_ONCE(addr != vma->vm_start);
1803 :
1804 0 : addr = vma->vm_start;
1805 :
1806 : /* If vm_flags changed after call_mmap(), we should try merge vma again
1807 : * as we may succeed this time.
1808 : */
1809 0 : if (unlikely(vm_flags != vma->vm_flags && prev)) {
1810 0 : merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags,
1811 : NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
1812 0 : if (merge) {
1813 : /* ->mmap() can change vma->vm_file and fput the original file. So
1814 : * fput the vma->vm_file here or we would add an extra fput for file
1815 : * and cause general protection fault ultimately.
1816 : */
1817 0 : fput(vma->vm_file);
1818 0 : vm_area_free(vma);
1819 0 : vma = merge;
1820 : /* Update vm_flags to pick up the change. */
1821 0 : vm_flags = vma->vm_flags;
1822 0 : goto unmap_writable;
1823 : }
1824 : }
1825 :
1826 0 : vm_flags = vma->vm_flags;
1827 0 : } else if (vm_flags & VM_SHARED) {
1828 0 : error = shmem_zero_setup(vma);
1829 0 : if (error)
1830 : goto free_vma;
1831 : } else {
1832 0 : vma_set_anonymous(vma);
1833 : }
1834 :
1835 : /* Allow architectures to sanity-check the vm_flags */
1836 0 : if (!arch_validate_flags(vma->vm_flags)) {
1837 : error = -EINVAL;
1838 : if (file)
1839 : goto unmap_and_free_vma;
1840 : else
1841 : goto free_vma;
1842 : }
1843 :
1844 0 : vma_link(mm, vma, prev, rb_link, rb_parent);
1845 : /* Once vma denies write, undo our temporary denial count */
1846 : unmap_writable:
1847 0 : if (file && vm_flags & VM_SHARED)
1848 0 : mapping_unmap_writable(file->f_mapping);
1849 0 : file = vma->vm_file;
1850 : out:
1851 0 : perf_event_mmap(vma);
1852 :
1853 0 : vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
1854 0 : if (vm_flags & VM_LOCKED) {
1855 0 : if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) ||
1856 0 : is_vm_hugetlb_page(vma) ||
1857 0 : vma == get_gate_vma(current->mm))
1858 0 : vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
1859 : else
1860 0 : mm->locked_vm += (len >> PAGE_SHIFT);
1861 : }
1862 :
1863 : if (file)
1864 : uprobe_mmap(vma);
1865 :
1866 : /*
1867 : * New (or expanded) vma always get soft dirty status.
1868 : * Otherwise user-space soft-dirty page tracker won't
1869 : * be able to distinguish situation when vma area unmapped,
1870 : * then new mapped in-place (which must be aimed as
1871 : * a completely new data area).
1872 : */
1873 : vma->vm_flags |= VM_SOFTDIRTY;
1874 :
1875 0 : vma_set_page_prot(vma);
1876 :
1877 0 : return addr;
1878 :
1879 : unmap_and_free_vma:
1880 0 : fput(vma->vm_file);
1881 0 : vma->vm_file = NULL;
1882 :
1883 : /* Undo any partial mapping done by a device driver. */
1884 0 : unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1885 0 : charged = 0;
1886 0 : if (vm_flags & VM_SHARED)
1887 0 : mapping_unmap_writable(file->f_mapping);
1888 : free_vma:
1889 0 : vm_area_free(vma);
1890 : unacct_error:
1891 0 : if (charged)
1892 0 : vm_unacct_memory(charged);
1893 0 : return error;
1894 : }
1895 :
1896 0 : static unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1897 : {
1898 : /*
1899 : * We implement the search by looking for an rbtree node that
1900 : * immediately follows a suitable gap. That is,
1901 : * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1902 : * - gap_end = vma->vm_start >= info->low_limit + length;
1903 : * - gap_end - gap_start >= length
1904 : */
1905 :
1906 0 : struct mm_struct *mm = current->mm;
1907 : struct vm_area_struct *vma;
1908 : unsigned long length, low_limit, high_limit, gap_start, gap_end;
1909 :
1910 : /* Adjust search length to account for worst case alignment overhead */
1911 0 : length = info->length + info->align_mask;
1912 0 : if (length < info->length)
1913 : return -ENOMEM;
1914 :
1915 : /* Adjust search limits by the desired length */
1916 0 : if (info->high_limit < length)
1917 : return -ENOMEM;
1918 0 : high_limit = info->high_limit - length;
1919 :
1920 0 : if (info->low_limit > high_limit)
1921 : return -ENOMEM;
1922 0 : low_limit = info->low_limit + length;
1923 :
1924 : /* Check if rbtree root looks promising */
1925 0 : if (RB_EMPTY_ROOT(&mm->mm_rb))
1926 : goto check_highest;
1927 0 : vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1928 0 : if (vma->rb_subtree_gap < length)
1929 : goto check_highest;
1930 :
1931 : while (true) {
1932 : /* Visit left subtree if it looks promising */
1933 0 : gap_end = vm_start_gap(vma);
1934 0 : if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1935 0 : struct vm_area_struct *left =
1936 0 : rb_entry(vma->vm_rb.rb_left,
1937 : struct vm_area_struct, vm_rb);
1938 0 : if (left->rb_subtree_gap >= length) {
1939 0 : vma = left;
1940 0 : continue;
1941 : }
1942 : }
1943 :
1944 0 : gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
1945 : check_current:
1946 : /* Check if current node has a suitable gap */
1947 0 : if (gap_start > high_limit)
1948 : return -ENOMEM;
1949 0 : if (gap_end >= low_limit &&
1950 0 : gap_end > gap_start && gap_end - gap_start >= length)
1951 : goto found;
1952 :
1953 : /* Visit right subtree if it looks promising */
1954 0 : if (vma->vm_rb.rb_right) {
1955 0 : struct vm_area_struct *right =
1956 0 : rb_entry(vma->vm_rb.rb_right,
1957 : struct vm_area_struct, vm_rb);
1958 0 : if (right->rb_subtree_gap >= length) {
1959 0 : vma = right;
1960 0 : continue;
1961 : }
1962 : }
1963 :
1964 : /* Go back up the rbtree to find next candidate node */
1965 : while (true) {
1966 0 : struct rb_node *prev = &vma->vm_rb;
1967 0 : if (!rb_parent(prev))
1968 : goto check_highest;
1969 0 : vma = rb_entry(rb_parent(prev),
1970 : struct vm_area_struct, vm_rb);
1971 0 : if (prev == vma->vm_rb.rb_left) {
1972 0 : gap_start = vm_end_gap(vma->vm_prev);
1973 0 : gap_end = vm_start_gap(vma);
1974 : goto check_current;
1975 : }
1976 : }
1977 : }
1978 :
1979 : check_highest:
1980 : /* Check highest gap, which does not precede any rbtree node */
1981 0 : gap_start = mm->highest_vm_end;
1982 0 : gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1983 0 : if (gap_start > high_limit)
1984 : return -ENOMEM;
1985 :
1986 : found:
1987 : /* We found a suitable gap. Clip it with the original low_limit. */
1988 0 : if (gap_start < info->low_limit)
1989 0 : gap_start = info->low_limit;
1990 :
1991 : /* Adjust gap address to the desired alignment */
1992 0 : gap_start += (info->align_offset - gap_start) & info->align_mask;
1993 :
1994 : VM_BUG_ON(gap_start + info->length > info->high_limit);
1995 : VM_BUG_ON(gap_start + info->length > gap_end);
1996 0 : return gap_start;
1997 : }
1998 :
1999 0 : static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
2000 : {
2001 0 : struct mm_struct *mm = current->mm;
2002 : struct vm_area_struct *vma;
2003 : unsigned long length, low_limit, high_limit, gap_start, gap_end;
2004 :
2005 : /* Adjust search length to account for worst case alignment overhead */
2006 0 : length = info->length + info->align_mask;
2007 0 : if (length < info->length)
2008 : return -ENOMEM;
2009 :
2010 : /*
2011 : * Adjust search limits by the desired length.
2012 : * See implementation comment at top of unmapped_area().
2013 : */
2014 0 : gap_end = info->high_limit;
2015 0 : if (gap_end < length)
2016 : return -ENOMEM;
2017 0 : high_limit = gap_end - length;
2018 :
2019 0 : if (info->low_limit > high_limit)
2020 : return -ENOMEM;
2021 0 : low_limit = info->low_limit + length;
2022 :
2023 : /* Check highest gap, which does not precede any rbtree node */
2024 0 : gap_start = mm->highest_vm_end;
2025 0 : if (gap_start <= high_limit)
2026 : goto found_highest;
2027 :
2028 : /* Check if rbtree root looks promising */
2029 0 : if (RB_EMPTY_ROOT(&mm->mm_rb))
2030 : return -ENOMEM;
2031 0 : vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
2032 0 : if (vma->rb_subtree_gap < length)
2033 : return -ENOMEM;
2034 :
2035 : while (true) {
2036 : /* Visit right subtree if it looks promising */
2037 0 : gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
2038 0 : if (gap_start <= high_limit && vma->vm_rb.rb_right) {
2039 0 : struct vm_area_struct *right =
2040 0 : rb_entry(vma->vm_rb.rb_right,
2041 : struct vm_area_struct, vm_rb);
2042 0 : if (right->rb_subtree_gap >= length) {
2043 0 : vma = right;
2044 0 : continue;
2045 : }
2046 : }
2047 :
2048 : check_current:
2049 : /* Check if current node has a suitable gap */
2050 0 : gap_end = vm_start_gap(vma);
2051 0 : if (gap_end < low_limit)
2052 : return -ENOMEM;
2053 0 : if (gap_start <= high_limit &&
2054 0 : gap_end > gap_start && gap_end - gap_start >= length)
2055 : goto found;
2056 :
2057 : /* Visit left subtree if it looks promising */
2058 0 : if (vma->vm_rb.rb_left) {
2059 0 : struct vm_area_struct *left =
2060 0 : rb_entry(vma->vm_rb.rb_left,
2061 : struct vm_area_struct, vm_rb);
2062 0 : if (left->rb_subtree_gap >= length) {
2063 0 : vma = left;
2064 0 : continue;
2065 : }
2066 : }
2067 :
2068 : /* Go back up the rbtree to find next candidate node */
2069 : while (true) {
2070 0 : struct rb_node *prev = &vma->vm_rb;
2071 0 : if (!rb_parent(prev))
2072 : return -ENOMEM;
2073 0 : vma = rb_entry(rb_parent(prev),
2074 : struct vm_area_struct, vm_rb);
2075 0 : if (prev == vma->vm_rb.rb_right) {
2076 0 : gap_start = vma->vm_prev ?
2077 0 : vm_end_gap(vma->vm_prev) : 0;
2078 : goto check_current;
2079 : }
2080 : }
2081 : }
2082 :
2083 : found:
2084 : /* We found a suitable gap. Clip it with the original high_limit. */
2085 0 : if (gap_end > info->high_limit)
2086 0 : gap_end = info->high_limit;
2087 :
2088 : found_highest:
2089 : /* Compute highest gap address at the desired alignment */
2090 0 : gap_end -= info->length;
2091 0 : gap_end -= (gap_end - info->align_offset) & info->align_mask;
2092 :
2093 : VM_BUG_ON(gap_end < info->low_limit);
2094 : VM_BUG_ON(gap_end < gap_start);
2095 0 : return gap_end;
2096 : }
2097 :
2098 : /*
2099 : * Search for an unmapped address range.
2100 : *
2101 : * We are looking for a range that:
2102 : * - does not intersect with any VMA;
2103 : * - is contained within the [low_limit, high_limit) interval;
2104 : * - is at least the desired size.
2105 : * - satisfies (begin_addr & align_mask) == (align_offset & align_mask)
2106 : */
2107 0 : unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info)
2108 : {
2109 : unsigned long addr;
2110 :
2111 0 : if (info->flags & VM_UNMAPPED_AREA_TOPDOWN)
2112 0 : addr = unmapped_area_topdown(info);
2113 : else
2114 0 : addr = unmapped_area(info);
2115 :
2116 0 : trace_vm_unmapped_area(addr, info);
2117 0 : return addr;
2118 : }
2119 :
2120 : /* Get an address range which is currently unmapped.
2121 : * For shmat() with addr=0.
2122 : *
2123 : * Ugly calling convention alert:
2124 : * Return value with the low bits set means error value,
2125 : * ie
2126 : * if (ret & ~PAGE_MASK)
2127 : * error = ret;
2128 : *
2129 : * This function "knows" that -ENOMEM has the bits set.
2130 : */
2131 : #ifndef HAVE_ARCH_UNMAPPED_AREA
2132 : unsigned long
2133 0 : arch_get_unmapped_area(struct file *filp, unsigned long addr,
2134 : unsigned long len, unsigned long pgoff, unsigned long flags)
2135 : {
2136 0 : struct mm_struct *mm = current->mm;
2137 : struct vm_area_struct *vma, *prev;
2138 : struct vm_unmapped_area_info info;
2139 0 : const unsigned long mmap_end = arch_get_mmap_end(addr);
2140 :
2141 0 : if (len > mmap_end - mmap_min_addr)
2142 : return -ENOMEM;
2143 :
2144 0 : if (flags & MAP_FIXED)
2145 : return addr;
2146 :
2147 0 : if (addr) {
2148 0 : addr = PAGE_ALIGN(addr);
2149 0 : vma = find_vma_prev(mm, addr, &prev);
2150 0 : if (mmap_end - len >= addr && addr >= mmap_min_addr &&
2151 0 : (!vma || addr + len <= vm_start_gap(vma)) &&
2152 0 : (!prev || addr >= vm_end_gap(prev)))
2153 : return addr;
2154 : }
2155 :
2156 0 : info.flags = 0;
2157 0 : info.length = len;
2158 0 : info.low_limit = mm->mmap_base;
2159 0 : info.high_limit = mmap_end;
2160 0 : info.align_mask = 0;
2161 0 : info.align_offset = 0;
2162 0 : return vm_unmapped_area(&info);
2163 : }
2164 : #endif
2165 :
2166 : /*
2167 : * This mmap-allocator allocates new areas top-down from below the
2168 : * stack's low limit (the base):
2169 : */
2170 : #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2171 : unsigned long
2172 0 : arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
2173 : unsigned long len, unsigned long pgoff,
2174 : unsigned long flags)
2175 : {
2176 : struct vm_area_struct *vma, *prev;
2177 0 : struct mm_struct *mm = current->mm;
2178 : struct vm_unmapped_area_info info;
2179 0 : const unsigned long mmap_end = arch_get_mmap_end(addr);
2180 :
2181 : /* requested length too big for entire address space */
2182 0 : if (len > mmap_end - mmap_min_addr)
2183 : return -ENOMEM;
2184 :
2185 0 : if (flags & MAP_FIXED)
2186 : return addr;
2187 :
2188 : /* requesting a specific address */
2189 0 : if (addr) {
2190 0 : addr = PAGE_ALIGN(addr);
2191 0 : vma = find_vma_prev(mm, addr, &prev);
2192 0 : if (mmap_end - len >= addr && addr >= mmap_min_addr &&
2193 0 : (!vma || addr + len <= vm_start_gap(vma)) &&
2194 0 : (!prev || addr >= vm_end_gap(prev)))
2195 : return addr;
2196 : }
2197 :
2198 0 : info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2199 0 : info.length = len;
2200 0 : info.low_limit = max(PAGE_SIZE, mmap_min_addr);
2201 0 : info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
2202 0 : info.align_mask = 0;
2203 0 : info.align_offset = 0;
2204 0 : addr = vm_unmapped_area(&info);
2205 :
2206 : /*
2207 : * A failed mmap() very likely causes application failure,
2208 : * so fall back to the bottom-up function here. This scenario
2209 : * can happen with large stack limits and large mmap()
2210 : * allocations.
2211 : */
2212 0 : if (offset_in_page(addr)) {
2213 : VM_BUG_ON(addr != -ENOMEM);
2214 0 : info.flags = 0;
2215 0 : info.low_limit = TASK_UNMAPPED_BASE;
2216 0 : info.high_limit = mmap_end;
2217 0 : addr = vm_unmapped_area(&info);
2218 : }
2219 :
2220 : return addr;
2221 : }
2222 : #endif
2223 :
2224 : unsigned long
2225 0 : get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2226 : unsigned long pgoff, unsigned long flags)
2227 : {
2228 : unsigned long (*get_area)(struct file *, unsigned long,
2229 : unsigned long, unsigned long, unsigned long);
2230 :
2231 0 : unsigned long error = arch_mmap_check(addr, len, flags);
2232 : if (error)
2233 : return error;
2234 :
2235 : /* Careful about overflows.. */
2236 0 : if (len > TASK_SIZE)
2237 : return -ENOMEM;
2238 :
2239 0 : get_area = current->mm->get_unmapped_area;
2240 0 : if (file) {
2241 0 : if (file->f_op->get_unmapped_area)
2242 0 : get_area = file->f_op->get_unmapped_area;
2243 0 : } else if (flags & MAP_SHARED) {
2244 : /*
2245 : * mmap_region() will call shmem_zero_setup() to create a file,
2246 : * so use shmem's get_unmapped_area in case it can be huge.
2247 : * do_mmap() will clear pgoff, so match alignment.
2248 : */
2249 0 : pgoff = 0;
2250 0 : get_area = shmem_get_unmapped_area;
2251 : }
2252 :
2253 0 : addr = get_area(file, addr, len, pgoff, flags);
2254 0 : if (IS_ERR_VALUE(addr))
2255 : return addr;
2256 :
2257 0 : if (addr > TASK_SIZE - len)
2258 : return -ENOMEM;
2259 0 : if (offset_in_page(addr))
2260 : return -EINVAL;
2261 :
2262 0 : error = security_mmap_addr(addr);
2263 0 : return error ? error : addr;
2264 : }
2265 :
2266 : EXPORT_SYMBOL(get_unmapped_area);
2267 :
2268 : /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2269 0 : struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2270 : {
2271 : struct rb_node *rb_node;
2272 : struct vm_area_struct *vma;
2273 :
2274 0 : mmap_assert_locked(mm);
2275 : /* Check the cache first. */
2276 0 : vma = vmacache_find(mm, addr);
2277 0 : if (likely(vma))
2278 : return vma;
2279 :
2280 0 : rb_node = mm->mm_rb.rb_node;
2281 :
2282 0 : while (rb_node) {
2283 : struct vm_area_struct *tmp;
2284 :
2285 0 : tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2286 :
2287 0 : if (tmp->vm_end > addr) {
2288 0 : vma = tmp;
2289 0 : if (tmp->vm_start <= addr)
2290 : break;
2291 0 : rb_node = rb_node->rb_left;
2292 : } else
2293 0 : rb_node = rb_node->rb_right;
2294 : }
2295 :
2296 0 : if (vma)
2297 0 : vmacache_update(addr, vma);
2298 : return vma;
2299 : }
2300 :
2301 : EXPORT_SYMBOL(find_vma);
2302 :
2303 : /*
2304 : * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2305 : */
2306 : struct vm_area_struct *
2307 0 : find_vma_prev(struct mm_struct *mm, unsigned long addr,
2308 : struct vm_area_struct **pprev)
2309 : {
2310 : struct vm_area_struct *vma;
2311 :
2312 0 : vma = find_vma(mm, addr);
2313 0 : if (vma) {
2314 0 : *pprev = vma->vm_prev;
2315 : } else {
2316 0 : struct rb_node *rb_node = rb_last(&mm->mm_rb);
2317 :
2318 0 : *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL;
2319 : }
2320 0 : return vma;
2321 : }
2322 :
2323 : /*
2324 : * Verify that the stack growth is acceptable and
2325 : * update accounting. This is shared with both the
2326 : * grow-up and grow-down cases.
2327 : */
2328 0 : static int acct_stack_growth(struct vm_area_struct *vma,
2329 : unsigned long size, unsigned long grow)
2330 : {
2331 0 : struct mm_struct *mm = vma->vm_mm;
2332 : unsigned long new_start;
2333 :
2334 : /* address space limit tests */
2335 0 : if (!may_expand_vm(mm, vma->vm_flags, grow))
2336 : return -ENOMEM;
2337 :
2338 : /* Stack limit test */
2339 0 : if (size > rlimit(RLIMIT_STACK))
2340 : return -ENOMEM;
2341 :
2342 : /* mlock limit tests */
2343 0 : if (vma->vm_flags & VM_LOCKED) {
2344 : unsigned long locked;
2345 : unsigned long limit;
2346 0 : locked = mm->locked_vm + grow;
2347 0 : limit = rlimit(RLIMIT_MEMLOCK);
2348 0 : limit >>= PAGE_SHIFT;
2349 0 : if (locked > limit && !capable(CAP_IPC_LOCK))
2350 : return -ENOMEM;
2351 : }
2352 :
2353 : /* Check to ensure the stack will not grow into a hugetlb-only region */
2354 0 : new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2355 0 : vma->vm_end - size;
2356 0 : if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2357 : return -EFAULT;
2358 :
2359 : /*
2360 : * Overcommit.. This must be the final test, as it will
2361 : * update security statistics.
2362 : */
2363 0 : if (security_vm_enough_memory_mm(mm, grow))
2364 : return -ENOMEM;
2365 :
2366 : return 0;
2367 : }
2368 :
2369 : #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2370 : /*
2371 : * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2372 : * vma is the last one with address > vma->vm_end. Have to extend vma.
2373 : */
2374 : int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2375 : {
2376 : struct mm_struct *mm = vma->vm_mm;
2377 : struct vm_area_struct *next;
2378 : unsigned long gap_addr;
2379 : int error = 0;
2380 :
2381 : if (!(vma->vm_flags & VM_GROWSUP))
2382 : return -EFAULT;
2383 :
2384 : /* Guard against exceeding limits of the address space. */
2385 : address &= PAGE_MASK;
2386 : if (address >= (TASK_SIZE & PAGE_MASK))
2387 : return -ENOMEM;
2388 : address += PAGE_SIZE;
2389 :
2390 : /* Enforce stack_guard_gap */
2391 : gap_addr = address + stack_guard_gap;
2392 :
2393 : /* Guard against overflow */
2394 : if (gap_addr < address || gap_addr > TASK_SIZE)
2395 : gap_addr = TASK_SIZE;
2396 :
2397 : next = vma->vm_next;
2398 : if (next && next->vm_start < gap_addr && vma_is_accessible(next)) {
2399 : if (!(next->vm_flags & VM_GROWSUP))
2400 : return -ENOMEM;
2401 : /* Check that both stack segments have the same anon_vma? */
2402 : }
2403 :
2404 : /* We must make sure the anon_vma is allocated. */
2405 : if (unlikely(anon_vma_prepare(vma)))
2406 : return -ENOMEM;
2407 :
2408 : /*
2409 : * vma->vm_start/vm_end cannot change under us because the caller
2410 : * is required to hold the mmap_lock in read mode. We need the
2411 : * anon_vma lock to serialize against concurrent expand_stacks.
2412 : */
2413 : anon_vma_lock_write(vma->anon_vma);
2414 :
2415 : /* Somebody else might have raced and expanded it already */
2416 : if (address > vma->vm_end) {
2417 : unsigned long size, grow;
2418 :
2419 : size = address - vma->vm_start;
2420 : grow = (address - vma->vm_end) >> PAGE_SHIFT;
2421 :
2422 : error = -ENOMEM;
2423 : if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2424 : error = acct_stack_growth(vma, size, grow);
2425 : if (!error) {
2426 : /*
2427 : * vma_gap_update() doesn't support concurrent
2428 : * updates, but we only hold a shared mmap_lock
2429 : * lock here, so we need to protect against
2430 : * concurrent vma expansions.
2431 : * anon_vma_lock_write() doesn't help here, as
2432 : * we don't guarantee that all growable vmas
2433 : * in a mm share the same root anon vma.
2434 : * So, we reuse mm->page_table_lock to guard
2435 : * against concurrent vma expansions.
2436 : */
2437 : spin_lock(&mm->page_table_lock);
2438 : if (vma->vm_flags & VM_LOCKED)
2439 : mm->locked_vm += grow;
2440 : vm_stat_account(mm, vma->vm_flags, grow);
2441 : anon_vma_interval_tree_pre_update_vma(vma);
2442 : vma->vm_end = address;
2443 : anon_vma_interval_tree_post_update_vma(vma);
2444 : if (vma->vm_next)
2445 : vma_gap_update(vma->vm_next);
2446 : else
2447 : mm->highest_vm_end = vm_end_gap(vma);
2448 : spin_unlock(&mm->page_table_lock);
2449 :
2450 : perf_event_mmap(vma);
2451 : }
2452 : }
2453 : }
2454 : anon_vma_unlock_write(vma->anon_vma);
2455 : khugepaged_enter_vma_merge(vma, vma->vm_flags);
2456 : validate_mm(mm);
2457 : return error;
2458 : }
2459 : #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2460 :
2461 : /*
2462 : * vma is the first one with address < vma->vm_start. Have to extend vma.
2463 : */
2464 0 : int expand_downwards(struct vm_area_struct *vma,
2465 : unsigned long address)
2466 : {
2467 0 : struct mm_struct *mm = vma->vm_mm;
2468 : struct vm_area_struct *prev;
2469 0 : int error = 0;
2470 :
2471 0 : address &= PAGE_MASK;
2472 0 : if (address < mmap_min_addr)
2473 : return -EPERM;
2474 :
2475 : /* Enforce stack_guard_gap */
2476 0 : prev = vma->vm_prev;
2477 : /* Check that both stack segments have the same anon_vma? */
2478 0 : if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
2479 0 : vma_is_accessible(prev)) {
2480 0 : if (address - prev->vm_end < stack_guard_gap)
2481 : return -ENOMEM;
2482 : }
2483 :
2484 : /* We must make sure the anon_vma is allocated. */
2485 0 : if (unlikely(anon_vma_prepare(vma)))
2486 : return -ENOMEM;
2487 :
2488 : /*
2489 : * vma->vm_start/vm_end cannot change under us because the caller
2490 : * is required to hold the mmap_lock in read mode. We need the
2491 : * anon_vma lock to serialize against concurrent expand_stacks.
2492 : */
2493 0 : anon_vma_lock_write(vma->anon_vma);
2494 :
2495 : /* Somebody else might have raced and expanded it already */
2496 0 : if (address < vma->vm_start) {
2497 : unsigned long size, grow;
2498 :
2499 0 : size = vma->vm_end - address;
2500 0 : grow = (vma->vm_start - address) >> PAGE_SHIFT;
2501 :
2502 0 : error = -ENOMEM;
2503 0 : if (grow <= vma->vm_pgoff) {
2504 0 : error = acct_stack_growth(vma, size, grow);
2505 0 : if (!error) {
2506 : /*
2507 : * vma_gap_update() doesn't support concurrent
2508 : * updates, but we only hold a shared mmap_lock
2509 : * lock here, so we need to protect against
2510 : * concurrent vma expansions.
2511 : * anon_vma_lock_write() doesn't help here, as
2512 : * we don't guarantee that all growable vmas
2513 : * in a mm share the same root anon vma.
2514 : * So, we reuse mm->page_table_lock to guard
2515 : * against concurrent vma expansions.
2516 : */
2517 0 : spin_lock(&mm->page_table_lock);
2518 0 : if (vma->vm_flags & VM_LOCKED)
2519 0 : mm->locked_vm += grow;
2520 0 : vm_stat_account(mm, vma->vm_flags, grow);
2521 0 : anon_vma_interval_tree_pre_update_vma(vma);
2522 0 : vma->vm_start = address;
2523 0 : vma->vm_pgoff -= grow;
2524 0 : anon_vma_interval_tree_post_update_vma(vma);
2525 0 : vma_gap_update(vma);
2526 0 : spin_unlock(&mm->page_table_lock);
2527 :
2528 : perf_event_mmap(vma);
2529 : }
2530 : }
2531 : }
2532 0 : anon_vma_unlock_write(vma->anon_vma);
2533 0 : khugepaged_enter_vma_merge(vma, vma->vm_flags);
2534 : validate_mm(mm);
2535 0 : return error;
2536 : }
2537 :
2538 : /* enforced gap between the expanding stack and other mappings. */
2539 : unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2540 :
2541 0 : static int __init cmdline_parse_stack_guard_gap(char *p)
2542 : {
2543 : unsigned long val;
2544 : char *endptr;
2545 :
2546 0 : val = simple_strtoul(p, &endptr, 10);
2547 0 : if (!*endptr)
2548 0 : stack_guard_gap = val << PAGE_SHIFT;
2549 :
2550 0 : return 1;
2551 : }
2552 : __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2553 :
2554 : #ifdef CONFIG_STACK_GROWSUP
2555 : int expand_stack(struct vm_area_struct *vma, unsigned long address)
2556 : {
2557 : return expand_upwards(vma, address);
2558 : }
2559 :
2560 : struct vm_area_struct *
2561 : find_extend_vma(struct mm_struct *mm, unsigned long addr)
2562 : {
2563 : struct vm_area_struct *vma, *prev;
2564 :
2565 : addr &= PAGE_MASK;
2566 : vma = find_vma_prev(mm, addr, &prev);
2567 : if (vma && (vma->vm_start <= addr))
2568 : return vma;
2569 : /* don't alter vm_end if the coredump is running */
2570 : if (!prev || expand_stack(prev, addr))
2571 : return NULL;
2572 : if (prev->vm_flags & VM_LOCKED)
2573 : populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2574 : return prev;
2575 : }
2576 : #else
2577 0 : int expand_stack(struct vm_area_struct *vma, unsigned long address)
2578 : {
2579 0 : return expand_downwards(vma, address);
2580 : }
2581 :
2582 : struct vm_area_struct *
2583 0 : find_extend_vma(struct mm_struct *mm, unsigned long addr)
2584 : {
2585 : struct vm_area_struct *vma;
2586 : unsigned long start;
2587 :
2588 0 : addr &= PAGE_MASK;
2589 0 : vma = find_vma(mm, addr);
2590 0 : if (!vma)
2591 : return NULL;
2592 0 : if (vma->vm_start <= addr)
2593 : return vma;
2594 0 : if (!(vma->vm_flags & VM_GROWSDOWN))
2595 : return NULL;
2596 0 : start = vma->vm_start;
2597 0 : if (expand_stack(vma, addr))
2598 : return NULL;
2599 0 : if (vma->vm_flags & VM_LOCKED)
2600 0 : populate_vma_page_range(vma, addr, start, NULL);
2601 : return vma;
2602 : }
2603 : #endif
2604 :
2605 : EXPORT_SYMBOL_GPL(find_extend_vma);
2606 :
2607 : /*
2608 : * Ok - we have the memory areas we should free on the vma list,
2609 : * so release them, and do the vma updates.
2610 : *
2611 : * Called with the mm semaphore held.
2612 : */
2613 0 : static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2614 : {
2615 0 : unsigned long nr_accounted = 0;
2616 :
2617 : /* Update high watermark before we lower total_vm */
2618 : update_hiwater_vm(mm);
2619 : do {
2620 0 : long nrpages = vma_pages(vma);
2621 :
2622 0 : if (vma->vm_flags & VM_ACCOUNT)
2623 0 : nr_accounted += nrpages;
2624 0 : vm_stat_account(mm, vma->vm_flags, -nrpages);
2625 0 : vma = remove_vma(vma);
2626 0 : } while (vma);
2627 0 : vm_unacct_memory(nr_accounted);
2628 : validate_mm(mm);
2629 0 : }
2630 :
2631 : /*
2632 : * Get rid of page table information in the indicated region.
2633 : *
2634 : * Called with the mm semaphore held.
2635 : */
2636 0 : static void unmap_region(struct mm_struct *mm,
2637 : struct vm_area_struct *vma, struct vm_area_struct *prev,
2638 : unsigned long start, unsigned long end)
2639 : {
2640 0 : struct vm_area_struct *next = vma_next(mm, prev);
2641 : struct mmu_gather tlb;
2642 :
2643 0 : lru_add_drain();
2644 0 : tlb_gather_mmu(&tlb, mm);
2645 0 : update_hiwater_rss(mm);
2646 0 : unmap_vmas(&tlb, vma, start, end);
2647 0 : free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2648 : next ? next->vm_start : USER_PGTABLES_CEILING);
2649 0 : tlb_finish_mmu(&tlb);
2650 0 : }
2651 :
2652 : /*
2653 : * Create a list of vma's touched by the unmap, removing them from the mm's
2654 : * vma list as we go..
2655 : */
2656 : static bool
2657 0 : detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2658 : struct vm_area_struct *prev, unsigned long end)
2659 : {
2660 : struct vm_area_struct **insertion_point;
2661 0 : struct vm_area_struct *tail_vma = NULL;
2662 :
2663 0 : insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2664 0 : vma->vm_prev = NULL;
2665 : do {
2666 0 : vma_rb_erase(vma, &mm->mm_rb);
2667 0 : if (vma->vm_flags & VM_LOCKED)
2668 0 : mm->locked_vm -= vma_pages(vma);
2669 0 : mm->map_count--;
2670 0 : tail_vma = vma;
2671 0 : vma = vma->vm_next;
2672 0 : } while (vma && vma->vm_start < end);
2673 0 : *insertion_point = vma;
2674 0 : if (vma) {
2675 0 : vma->vm_prev = prev;
2676 : vma_gap_update(vma);
2677 : } else
2678 0 : mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
2679 0 : tail_vma->vm_next = NULL;
2680 :
2681 : /* Kill the cache */
2682 0 : vmacache_invalidate(mm);
2683 :
2684 : /*
2685 : * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or
2686 : * VM_GROWSUP VMA. Such VMAs can change their size under
2687 : * down_read(mmap_lock) and collide with the VMA we are about to unmap.
2688 : */
2689 0 : if (vma && (vma->vm_flags & VM_GROWSDOWN))
2690 : return false;
2691 : if (prev && (prev->vm_flags & VM_GROWSUP))
2692 : return false;
2693 0 : return true;
2694 : }
2695 :
2696 : /*
2697 : * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2698 : * has already been checked or doesn't make sense to fail.
2699 : */
2700 0 : int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2701 : unsigned long addr, int new_below)
2702 : {
2703 : struct vm_area_struct *new;
2704 : int err;
2705 :
2706 0 : if (vma->vm_ops && vma->vm_ops->may_split) {
2707 0 : err = vma->vm_ops->may_split(vma, addr);
2708 0 : if (err)
2709 : return err;
2710 : }
2711 :
2712 0 : new = vm_area_dup(vma);
2713 0 : if (!new)
2714 : return -ENOMEM;
2715 :
2716 0 : if (new_below)
2717 0 : new->vm_end = addr;
2718 : else {
2719 0 : new->vm_start = addr;
2720 0 : new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2721 : }
2722 :
2723 0 : err = vma_dup_policy(vma, new);
2724 : if (err)
2725 : goto out_free_vma;
2726 :
2727 0 : err = anon_vma_clone(new, vma);
2728 0 : if (err)
2729 : goto out_free_mpol;
2730 :
2731 0 : if (new->vm_file)
2732 0 : get_file(new->vm_file);
2733 :
2734 0 : if (new->vm_ops && new->vm_ops->open)
2735 0 : new->vm_ops->open(new);
2736 :
2737 0 : if (new_below)
2738 0 : err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2739 0 : ((addr - new->vm_start) >> PAGE_SHIFT), new);
2740 : else
2741 0 : err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2742 :
2743 : /* Success. */
2744 0 : if (!err)
2745 : return 0;
2746 :
2747 : /* Clean everything up if vma_adjust failed. */
2748 0 : if (new->vm_ops && new->vm_ops->close)
2749 0 : new->vm_ops->close(new);
2750 0 : if (new->vm_file)
2751 0 : fput(new->vm_file);
2752 0 : unlink_anon_vmas(new);
2753 : out_free_mpol:
2754 0 : mpol_put(vma_policy(new));
2755 : out_free_vma:
2756 0 : vm_area_free(new);
2757 0 : return err;
2758 : }
2759 :
2760 : /*
2761 : * Split a vma into two pieces at address 'addr', a new vma is allocated
2762 : * either for the first part or the tail.
2763 : */
2764 0 : int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2765 : unsigned long addr, int new_below)
2766 : {
2767 0 : if (mm->map_count >= sysctl_max_map_count)
2768 : return -ENOMEM;
2769 :
2770 0 : return __split_vma(mm, vma, addr, new_below);
2771 : }
2772 :
2773 : /* Munmap is split into 2 main parts -- this part which finds
2774 : * what needs doing, and the areas themselves, which do the
2775 : * work. This now handles partial unmappings.
2776 : * Jeremy Fitzhardinge <jeremy@goop.org>
2777 : */
2778 0 : int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2779 : struct list_head *uf, bool downgrade)
2780 : {
2781 : unsigned long end;
2782 : struct vm_area_struct *vma, *prev, *last;
2783 :
2784 0 : if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2785 : return -EINVAL;
2786 :
2787 0 : len = PAGE_ALIGN(len);
2788 0 : end = start + len;
2789 0 : if (len == 0)
2790 : return -EINVAL;
2791 :
2792 : /*
2793 : * arch_unmap() might do unmaps itself. It must be called
2794 : * and finish any rbtree manipulation before this code
2795 : * runs and also starts to manipulate the rbtree.
2796 : */
2797 0 : arch_unmap(mm, start, end);
2798 :
2799 : /* Find the first overlapping VMA where start < vma->vm_end */
2800 0 : vma = find_vma_intersection(mm, start, end);
2801 0 : if (!vma)
2802 : return 0;
2803 0 : prev = vma->vm_prev;
2804 :
2805 : /*
2806 : * If we need to split any vma, do it now to save pain later.
2807 : *
2808 : * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2809 : * unmapped vm_area_struct will remain in use: so lower split_vma
2810 : * places tmp vma above, and higher split_vma places tmp vma below.
2811 : */
2812 0 : if (start > vma->vm_start) {
2813 : int error;
2814 :
2815 : /*
2816 : * Make sure that map_count on return from munmap() will
2817 : * not exceed its limit; but let map_count go just above
2818 : * its limit temporarily, to help free resources as expected.
2819 : */
2820 0 : if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2821 : return -ENOMEM;
2822 :
2823 0 : error = __split_vma(mm, vma, start, 0);
2824 0 : if (error)
2825 : return error;
2826 : prev = vma;
2827 : }
2828 :
2829 : /* Does it split the last one? */
2830 0 : last = find_vma(mm, end);
2831 0 : if (last && end > last->vm_start) {
2832 0 : int error = __split_vma(mm, last, end, 1);
2833 0 : if (error)
2834 : return error;
2835 : }
2836 0 : vma = vma_next(mm, prev);
2837 :
2838 : if (unlikely(uf)) {
2839 : /*
2840 : * If userfaultfd_unmap_prep returns an error the vmas
2841 : * will remain split, but userland will get a
2842 : * highly unexpected error anyway. This is no
2843 : * different than the case where the first of the two
2844 : * __split_vma fails, but we don't undo the first
2845 : * split, despite we could. This is unlikely enough
2846 : * failure that it's not worth optimizing it for.
2847 : */
2848 : int error = userfaultfd_unmap_prep(vma, start, end, uf);
2849 : if (error)
2850 : return error;
2851 : }
2852 :
2853 : /* Detach vmas from rbtree */
2854 0 : if (!detach_vmas_to_be_unmapped(mm, vma, prev, end))
2855 0 : downgrade = false;
2856 :
2857 0 : if (downgrade)
2858 : mmap_write_downgrade(mm);
2859 :
2860 0 : unmap_region(mm, vma, prev, start, end);
2861 :
2862 : /* Fix up all other VM information */
2863 0 : remove_vma_list(mm, vma);
2864 :
2865 0 : return downgrade ? 1 : 0;
2866 : }
2867 :
2868 0 : int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2869 : struct list_head *uf)
2870 : {
2871 0 : return __do_munmap(mm, start, len, uf, false);
2872 : }
2873 :
2874 0 : static int __vm_munmap(unsigned long start, size_t len, bool downgrade)
2875 : {
2876 : int ret;
2877 0 : struct mm_struct *mm = current->mm;
2878 0 : LIST_HEAD(uf);
2879 :
2880 0 : if (mmap_write_lock_killable(mm))
2881 : return -EINTR;
2882 :
2883 0 : ret = __do_munmap(mm, start, len, &uf, downgrade);
2884 : /*
2885 : * Returning 1 indicates mmap_lock is downgraded.
2886 : * But 1 is not legal return value of vm_munmap() and munmap(), reset
2887 : * it to 0 before return.
2888 : */
2889 0 : if (ret == 1) {
2890 0 : mmap_read_unlock(mm);
2891 0 : ret = 0;
2892 : } else
2893 : mmap_write_unlock(mm);
2894 :
2895 : userfaultfd_unmap_complete(mm, &uf);
2896 : return ret;
2897 : }
2898 :
2899 0 : int vm_munmap(unsigned long start, size_t len)
2900 : {
2901 0 : return __vm_munmap(start, len, false);
2902 : }
2903 : EXPORT_SYMBOL(vm_munmap);
2904 :
2905 0 : SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2906 : {
2907 0 : addr = untagged_addr(addr);
2908 0 : return __vm_munmap(addr, len, true);
2909 : }
2910 :
2911 :
2912 : /*
2913 : * Emulation of deprecated remap_file_pages() syscall.
2914 : */
2915 0 : SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2916 : unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2917 : {
2918 :
2919 0 : struct mm_struct *mm = current->mm;
2920 : struct vm_area_struct *vma;
2921 0 : unsigned long populate = 0;
2922 0 : unsigned long ret = -EINVAL;
2923 : struct file *file;
2924 :
2925 0 : pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n",
2926 : current->comm, current->pid);
2927 :
2928 0 : if (prot)
2929 : return ret;
2930 0 : start = start & PAGE_MASK;
2931 0 : size = size & PAGE_MASK;
2932 :
2933 0 : if (start + size <= start)
2934 : return ret;
2935 :
2936 : /* Does pgoff wrap? */
2937 0 : if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2938 : return ret;
2939 :
2940 0 : if (mmap_write_lock_killable(mm))
2941 : return -EINTR;
2942 :
2943 0 : vma = vma_lookup(mm, start);
2944 :
2945 0 : if (!vma || !(vma->vm_flags & VM_SHARED))
2946 : goto out;
2947 :
2948 0 : if (start + size > vma->vm_end) {
2949 : struct vm_area_struct *next;
2950 :
2951 0 : for (next = vma->vm_next; next; next = next->vm_next) {
2952 : /* hole between vmas ? */
2953 0 : if (next->vm_start != next->vm_prev->vm_end)
2954 : goto out;
2955 :
2956 0 : if (next->vm_file != vma->vm_file)
2957 : goto out;
2958 :
2959 0 : if (next->vm_flags != vma->vm_flags)
2960 : goto out;
2961 :
2962 0 : if (start + size <= next->vm_end)
2963 : break;
2964 : }
2965 :
2966 0 : if (!next)
2967 : goto out;
2968 : }
2969 :
2970 0 : prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2971 0 : prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2972 0 : prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2973 :
2974 0 : flags &= MAP_NONBLOCK;
2975 0 : flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2976 0 : if (vma->vm_flags & VM_LOCKED)
2977 0 : flags |= MAP_LOCKED;
2978 :
2979 0 : file = get_file(vma->vm_file);
2980 0 : ret = do_mmap(vma->vm_file, start, size,
2981 : prot, flags, pgoff, &populate, NULL);
2982 0 : fput(file);
2983 : out:
2984 0 : mmap_write_unlock(mm);
2985 0 : if (populate)
2986 0 : mm_populate(ret, populate);
2987 0 : if (!IS_ERR_VALUE(ret))
2988 0 : ret = 0;
2989 0 : return ret;
2990 : }
2991 :
2992 : /*
2993 : * this is really a simplified "do_mmap". it only handles
2994 : * anonymous maps. eventually we may be able to do some
2995 : * brk-specific accounting here.
2996 : */
2997 0 : static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf)
2998 : {
2999 0 : struct mm_struct *mm = current->mm;
3000 : struct vm_area_struct *vma, *prev;
3001 : struct rb_node **rb_link, *rb_parent;
3002 0 : pgoff_t pgoff = addr >> PAGE_SHIFT;
3003 : int error;
3004 : unsigned long mapped_addr;
3005 :
3006 : /* Until we need other flags, refuse anything except VM_EXEC. */
3007 0 : if ((flags & (~VM_EXEC)) != 0)
3008 : return -EINVAL;
3009 0 : flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
3010 :
3011 0 : mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
3012 0 : if (IS_ERR_VALUE(mapped_addr))
3013 0 : return mapped_addr;
3014 :
3015 0 : error = mlock_future_check(mm, mm->def_flags, len);
3016 0 : if (error)
3017 : return error;
3018 :
3019 : /* Clear old maps, set up prev, rb_link, rb_parent, and uf */
3020 0 : if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf))
3021 : return -ENOMEM;
3022 :
3023 : /* Check against address space limits *after* clearing old maps... */
3024 0 : if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
3025 : return -ENOMEM;
3026 :
3027 0 : if (mm->map_count > sysctl_max_map_count)
3028 : return -ENOMEM;
3029 :
3030 0 : if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
3031 : return -ENOMEM;
3032 :
3033 : /* Can we just expand an old private anonymous mapping? */
3034 0 : vma = vma_merge(mm, prev, addr, addr + len, flags,
3035 : NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL);
3036 0 : if (vma)
3037 : goto out;
3038 :
3039 : /*
3040 : * create a vma struct for an anonymous mapping
3041 : */
3042 0 : vma = vm_area_alloc(mm);
3043 0 : if (!vma) {
3044 0 : vm_unacct_memory(len >> PAGE_SHIFT);
3045 0 : return -ENOMEM;
3046 : }
3047 :
3048 0 : vma_set_anonymous(vma);
3049 0 : vma->vm_start = addr;
3050 0 : vma->vm_end = addr + len;
3051 0 : vma->vm_pgoff = pgoff;
3052 0 : vma->vm_flags = flags;
3053 0 : vma->vm_page_prot = vm_get_page_prot(flags);
3054 0 : vma_link(mm, vma, prev, rb_link, rb_parent);
3055 : out:
3056 0 : perf_event_mmap(vma);
3057 0 : mm->total_vm += len >> PAGE_SHIFT;
3058 0 : mm->data_vm += len >> PAGE_SHIFT;
3059 0 : if (flags & VM_LOCKED)
3060 0 : mm->locked_vm += (len >> PAGE_SHIFT);
3061 : vma->vm_flags |= VM_SOFTDIRTY;
3062 0 : return 0;
3063 : }
3064 :
3065 0 : int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
3066 : {
3067 0 : struct mm_struct *mm = current->mm;
3068 : unsigned long len;
3069 : int ret;
3070 : bool populate;
3071 0 : LIST_HEAD(uf);
3072 :
3073 0 : len = PAGE_ALIGN(request);
3074 0 : if (len < request)
3075 : return -ENOMEM;
3076 0 : if (!len)
3077 : return 0;
3078 :
3079 0 : if (mmap_write_lock_killable(mm))
3080 : return -EINTR;
3081 :
3082 0 : ret = do_brk_flags(addr, len, flags, &uf);
3083 0 : populate = ((mm->def_flags & VM_LOCKED) != 0);
3084 0 : mmap_write_unlock(mm);
3085 0 : userfaultfd_unmap_complete(mm, &uf);
3086 0 : if (populate && !ret)
3087 : mm_populate(addr, len);
3088 : return ret;
3089 : }
3090 : EXPORT_SYMBOL(vm_brk_flags);
3091 :
3092 0 : int vm_brk(unsigned long addr, unsigned long len)
3093 : {
3094 0 : return vm_brk_flags(addr, len, 0);
3095 : }
3096 : EXPORT_SYMBOL(vm_brk);
3097 :
3098 : /* Release all mmaps. */
3099 0 : void exit_mmap(struct mm_struct *mm)
3100 : {
3101 : struct mmu_gather tlb;
3102 : struct vm_area_struct *vma;
3103 0 : unsigned long nr_accounted = 0;
3104 :
3105 : /* mm's last user has gone, and its about to be pulled down */
3106 0 : mmu_notifier_release(mm);
3107 :
3108 0 : if (unlikely(mm_is_oom_victim(mm))) {
3109 : /*
3110 : * Manually reap the mm to free as much memory as possible.
3111 : * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3112 : * this mm from further consideration. Taking mm->mmap_lock for
3113 : * write after setting MMF_OOM_SKIP will guarantee that the oom
3114 : * reaper will not run on this mm again after mmap_lock is
3115 : * dropped.
3116 : *
3117 : * Nothing can be holding mm->mmap_lock here and the above call
3118 : * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3119 : * __oom_reap_task_mm() will not block.
3120 : */
3121 0 : (void)__oom_reap_task_mm(mm);
3122 0 : set_bit(MMF_OOM_SKIP, &mm->flags);
3123 : }
3124 :
3125 0 : mmap_write_lock(mm);
3126 0 : arch_exit_mmap(mm);
3127 :
3128 0 : vma = mm->mmap;
3129 0 : if (!vma) {
3130 : /* Can happen if dup_mmap() received an OOM */
3131 0 : mmap_write_unlock(mm);
3132 0 : return;
3133 : }
3134 :
3135 0 : lru_add_drain();
3136 0 : flush_cache_mm(mm);
3137 0 : tlb_gather_mmu_fullmm(&tlb, mm);
3138 : /* update_hiwater_rss(mm) here? but nobody should be looking */
3139 : /* Use -1 here to ensure all VMAs in the mm are unmapped */
3140 0 : unmap_vmas(&tlb, vma, 0, -1);
3141 0 : free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
3142 0 : tlb_finish_mmu(&tlb);
3143 :
3144 : /* Walk the list again, actually closing and freeing it. */
3145 0 : while (vma) {
3146 0 : if (vma->vm_flags & VM_ACCOUNT)
3147 0 : nr_accounted += vma_pages(vma);
3148 0 : vma = remove_vma(vma);
3149 0 : cond_resched();
3150 : }
3151 0 : mm->mmap = NULL;
3152 0 : mmap_write_unlock(mm);
3153 0 : vm_unacct_memory(nr_accounted);
3154 : }
3155 :
3156 : /* Insert vm structure into process list sorted by address
3157 : * and into the inode's i_mmap tree. If vm_file is non-NULL
3158 : * then i_mmap_rwsem is taken here.
3159 : */
3160 0 : int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
3161 : {
3162 : struct vm_area_struct *prev;
3163 : struct rb_node **rb_link, *rb_parent;
3164 :
3165 0 : if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3166 : &prev, &rb_link, &rb_parent))
3167 : return -ENOMEM;
3168 0 : if ((vma->vm_flags & VM_ACCOUNT) &&
3169 0 : security_vm_enough_memory_mm(mm, vma_pages(vma)))
3170 : return -ENOMEM;
3171 :
3172 : /*
3173 : * The vm_pgoff of a purely anonymous vma should be irrelevant
3174 : * until its first write fault, when page's anon_vma and index
3175 : * are set. But now set the vm_pgoff it will almost certainly
3176 : * end up with (unless mremap moves it elsewhere before that
3177 : * first wfault), so /proc/pid/maps tells a consistent story.
3178 : *
3179 : * By setting it to reflect the virtual start address of the
3180 : * vma, merges and splits can happen in a seamless way, just
3181 : * using the existing file pgoff checks and manipulations.
3182 : * Similarly in do_mmap and in do_brk_flags.
3183 : */
3184 0 : if (vma_is_anonymous(vma)) {
3185 0 : BUG_ON(vma->anon_vma);
3186 0 : vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3187 : }
3188 :
3189 0 : vma_link(mm, vma, prev, rb_link, rb_parent);
3190 0 : return 0;
3191 : }
3192 :
3193 : /*
3194 : * Copy the vma structure to a new location in the same mm,
3195 : * prior to moving page table entries, to effect an mremap move.
3196 : */
3197 0 : struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
3198 : unsigned long addr, unsigned long len, pgoff_t pgoff,
3199 : bool *need_rmap_locks)
3200 : {
3201 0 : struct vm_area_struct *vma = *vmap;
3202 0 : unsigned long vma_start = vma->vm_start;
3203 0 : struct mm_struct *mm = vma->vm_mm;
3204 : struct vm_area_struct *new_vma, *prev;
3205 : struct rb_node **rb_link, *rb_parent;
3206 0 : bool faulted_in_anon_vma = true;
3207 :
3208 : /*
3209 : * If anonymous vma has not yet been faulted, update new pgoff
3210 : * to match new location, to increase its chance of merging.
3211 : */
3212 0 : if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
3213 0 : pgoff = addr >> PAGE_SHIFT;
3214 0 : faulted_in_anon_vma = false;
3215 : }
3216 :
3217 0 : if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3218 : return NULL; /* should never get here */
3219 0 : new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
3220 : vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3221 : vma->vm_userfaultfd_ctx, anon_vma_name(vma));
3222 0 : if (new_vma) {
3223 : /*
3224 : * Source vma may have been merged into new_vma
3225 : */
3226 0 : if (unlikely(vma_start >= new_vma->vm_start &&
3227 : vma_start < new_vma->vm_end)) {
3228 : /*
3229 : * The only way we can get a vma_merge with
3230 : * self during an mremap is if the vma hasn't
3231 : * been faulted in yet and we were allowed to
3232 : * reset the dst vma->vm_pgoff to the
3233 : * destination address of the mremap to allow
3234 : * the merge to happen. mremap must change the
3235 : * vm_pgoff linearity between src and dst vmas
3236 : * (in turn preventing a vma_merge) to be
3237 : * safe. It is only safe to keep the vm_pgoff
3238 : * linear if there are no pages mapped yet.
3239 : */
3240 : VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
3241 0 : *vmap = vma = new_vma;
3242 : }
3243 0 : *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
3244 : } else {
3245 0 : new_vma = vm_area_dup(vma);
3246 0 : if (!new_vma)
3247 : goto out;
3248 0 : new_vma->vm_start = addr;
3249 0 : new_vma->vm_end = addr + len;
3250 0 : new_vma->vm_pgoff = pgoff;
3251 0 : if (vma_dup_policy(vma, new_vma))
3252 : goto out_free_vma;
3253 0 : if (anon_vma_clone(new_vma, vma))
3254 : goto out_free_mempol;
3255 0 : if (new_vma->vm_file)
3256 0 : get_file(new_vma->vm_file);
3257 0 : if (new_vma->vm_ops && new_vma->vm_ops->open)
3258 0 : new_vma->vm_ops->open(new_vma);
3259 0 : vma_link(mm, new_vma, prev, rb_link, rb_parent);
3260 0 : *need_rmap_locks = false;
3261 : }
3262 : return new_vma;
3263 :
3264 : out_free_mempol:
3265 : mpol_put(vma_policy(new_vma));
3266 : out_free_vma:
3267 0 : vm_area_free(new_vma);
3268 : out:
3269 : return NULL;
3270 : }
3271 :
3272 : /*
3273 : * Return true if the calling process may expand its vm space by the passed
3274 : * number of pages
3275 : */
3276 0 : bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
3277 : {
3278 0 : if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3279 : return false;
3280 :
3281 0 : if (is_data_mapping(flags) &&
3282 0 : mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3283 : /* Workaround for Valgrind */
3284 0 : if (rlimit(RLIMIT_DATA) == 0 &&
3285 0 : mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3286 : return true;
3287 :
3288 0 : pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3289 : current->comm, current->pid,
3290 : (mm->data_vm + npages) << PAGE_SHIFT,
3291 : rlimit(RLIMIT_DATA),
3292 : ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
3293 :
3294 0 : if (!ignore_rlimit_data)
3295 : return false;
3296 : }
3297 :
3298 : return true;
3299 : }
3300 :
3301 0 : void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3302 : {
3303 0 : WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages);
3304 :
3305 0 : if (is_exec_mapping(flags))
3306 0 : mm->exec_vm += npages;
3307 0 : else if (is_stack_mapping(flags))
3308 0 : mm->stack_vm += npages;
3309 0 : else if (is_data_mapping(flags))
3310 0 : mm->data_vm += npages;
3311 0 : }
3312 :
3313 : static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
3314 :
3315 : /*
3316 : * Having a close hook prevents vma merging regardless of flags.
3317 : */
3318 0 : static void special_mapping_close(struct vm_area_struct *vma)
3319 : {
3320 0 : }
3321 :
3322 0 : static const char *special_mapping_name(struct vm_area_struct *vma)
3323 : {
3324 0 : return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3325 : }
3326 :
3327 0 : static int special_mapping_mremap(struct vm_area_struct *new_vma)
3328 : {
3329 0 : struct vm_special_mapping *sm = new_vma->vm_private_data;
3330 :
3331 0 : if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3332 : return -EFAULT;
3333 :
3334 0 : if (sm->mremap)
3335 0 : return sm->mremap(sm, new_vma);
3336 :
3337 : return 0;
3338 : }
3339 :
3340 0 : static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr)
3341 : {
3342 : /*
3343 : * Forbid splitting special mappings - kernel has expectations over
3344 : * the number of pages in mapping. Together with VM_DONTEXPAND
3345 : * the size of vma should stay the same over the special mapping's
3346 : * lifetime.
3347 : */
3348 0 : return -EINVAL;
3349 : }
3350 :
3351 : static const struct vm_operations_struct special_mapping_vmops = {
3352 : .close = special_mapping_close,
3353 : .fault = special_mapping_fault,
3354 : .mremap = special_mapping_mremap,
3355 : .name = special_mapping_name,
3356 : /* vDSO code relies that VVAR can't be accessed remotely */
3357 : .access = NULL,
3358 : .may_split = special_mapping_split,
3359 : };
3360 :
3361 : static const struct vm_operations_struct legacy_special_mapping_vmops = {
3362 : .close = special_mapping_close,
3363 : .fault = special_mapping_fault,
3364 : };
3365 :
3366 0 : static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
3367 : {
3368 0 : struct vm_area_struct *vma = vmf->vma;
3369 : pgoff_t pgoff;
3370 : struct page **pages;
3371 :
3372 0 : if (vma->vm_ops == &legacy_special_mapping_vmops) {
3373 0 : pages = vma->vm_private_data;
3374 : } else {
3375 0 : struct vm_special_mapping *sm = vma->vm_private_data;
3376 :
3377 0 : if (sm->fault)
3378 0 : return sm->fault(sm, vmf->vma, vmf);
3379 :
3380 0 : pages = sm->pages;
3381 : }
3382 :
3383 0 : for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3384 0 : pgoff--;
3385 :
3386 0 : if (*pages) {
3387 0 : struct page *page = *pages;
3388 0 : get_page(page);
3389 0 : vmf->page = page;
3390 0 : return 0;
3391 : }
3392 :
3393 : return VM_FAULT_SIGBUS;
3394 : }
3395 :
3396 0 : static struct vm_area_struct *__install_special_mapping(
3397 : struct mm_struct *mm,
3398 : unsigned long addr, unsigned long len,
3399 : unsigned long vm_flags, void *priv,
3400 : const struct vm_operations_struct *ops)
3401 : {
3402 : int ret;
3403 : struct vm_area_struct *vma;
3404 :
3405 0 : vma = vm_area_alloc(mm);
3406 0 : if (unlikely(vma == NULL))
3407 : return ERR_PTR(-ENOMEM);
3408 :
3409 0 : vma->vm_start = addr;
3410 0 : vma->vm_end = addr + len;
3411 :
3412 0 : vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
3413 0 : vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
3414 0 : vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3415 :
3416 0 : vma->vm_ops = ops;
3417 0 : vma->vm_private_data = priv;
3418 :
3419 0 : ret = insert_vm_struct(mm, vma);
3420 0 : if (ret)
3421 : goto out;
3422 :
3423 0 : vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3424 :
3425 : perf_event_mmap(vma);
3426 :
3427 : return vma;
3428 :
3429 : out:
3430 0 : vm_area_free(vma);
3431 0 : return ERR_PTR(ret);
3432 : }
3433 :
3434 0 : bool vma_is_special_mapping(const struct vm_area_struct *vma,
3435 : const struct vm_special_mapping *sm)
3436 : {
3437 0 : return vma->vm_private_data == sm &&
3438 0 : (vma->vm_ops == &special_mapping_vmops ||
3439 : vma->vm_ops == &legacy_special_mapping_vmops);
3440 : }
3441 :
3442 : /*
3443 : * Called with mm->mmap_lock held for writing.
3444 : * Insert a new vma covering the given region, with the given flags.
3445 : * Its pages are supplied by the given array of struct page *.
3446 : * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3447 : * The region past the last page supplied will always produce SIGBUS.
3448 : * The array pointer and the pages it points to are assumed to stay alive
3449 : * for as long as this mapping might exist.
3450 : */
3451 0 : struct vm_area_struct *_install_special_mapping(
3452 : struct mm_struct *mm,
3453 : unsigned long addr, unsigned long len,
3454 : unsigned long vm_flags, const struct vm_special_mapping *spec)
3455 : {
3456 0 : return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3457 : &special_mapping_vmops);
3458 : }
3459 :
3460 0 : int install_special_mapping(struct mm_struct *mm,
3461 : unsigned long addr, unsigned long len,
3462 : unsigned long vm_flags, struct page **pages)
3463 : {
3464 0 : struct vm_area_struct *vma = __install_special_mapping(
3465 : mm, addr, len, vm_flags, (void *)pages,
3466 : &legacy_special_mapping_vmops);
3467 :
3468 0 : return PTR_ERR_OR_ZERO(vma);
3469 : }
3470 :
3471 : static DEFINE_MUTEX(mm_all_locks_mutex);
3472 :
3473 0 : static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3474 : {
3475 0 : if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3476 : /*
3477 : * The LSB of head.next can't change from under us
3478 : * because we hold the mm_all_locks_mutex.
3479 : */
3480 0 : down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock);
3481 : /*
3482 : * We can safely modify head.next after taking the
3483 : * anon_vma->root->rwsem. If some other vma in this mm shares
3484 : * the same anon_vma we won't take it again.
3485 : *
3486 : * No need of atomic instructions here, head.next
3487 : * can't change from under us thanks to the
3488 : * anon_vma->root->rwsem.
3489 : */
3490 0 : if (__test_and_set_bit(0, (unsigned long *)
3491 0 : &anon_vma->root->rb_root.rb_root.rb_node))
3492 0 : BUG();
3493 : }
3494 0 : }
3495 :
3496 0 : static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3497 : {
3498 0 : if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3499 : /*
3500 : * AS_MM_ALL_LOCKS can't change from under us because
3501 : * we hold the mm_all_locks_mutex.
3502 : *
3503 : * Operations on ->flags have to be atomic because
3504 : * even if AS_MM_ALL_LOCKS is stable thanks to the
3505 : * mm_all_locks_mutex, there may be other cpus
3506 : * changing other bitflags in parallel to us.
3507 : */
3508 0 : if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3509 0 : BUG();
3510 0 : down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock);
3511 : }
3512 0 : }
3513 :
3514 : /*
3515 : * This operation locks against the VM for all pte/vma/mm related
3516 : * operations that could ever happen on a certain mm. This includes
3517 : * vmtruncate, try_to_unmap, and all page faults.
3518 : *
3519 : * The caller must take the mmap_lock in write mode before calling
3520 : * mm_take_all_locks(). The caller isn't allowed to release the
3521 : * mmap_lock until mm_drop_all_locks() returns.
3522 : *
3523 : * mmap_lock in write mode is required in order to block all operations
3524 : * that could modify pagetables and free pages without need of
3525 : * altering the vma layout. It's also needed in write mode to avoid new
3526 : * anon_vmas to be associated with existing vmas.
3527 : *
3528 : * A single task can't take more than one mm_take_all_locks() in a row
3529 : * or it would deadlock.
3530 : *
3531 : * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3532 : * mapping->flags avoid to take the same lock twice, if more than one
3533 : * vma in this mm is backed by the same anon_vma or address_space.
3534 : *
3535 : * We take locks in following order, accordingly to comment at beginning
3536 : * of mm/rmap.c:
3537 : * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3538 : * hugetlb mapping);
3539 : * - all i_mmap_rwsem locks;
3540 : * - all anon_vma->rwseml
3541 : *
3542 : * We can take all locks within these types randomly because the VM code
3543 : * doesn't nest them and we protected from parallel mm_take_all_locks() by
3544 : * mm_all_locks_mutex.
3545 : *
3546 : * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3547 : * that may have to take thousand of locks.
3548 : *
3549 : * mm_take_all_locks() can fail if it's interrupted by signals.
3550 : */
3551 0 : int mm_take_all_locks(struct mm_struct *mm)
3552 : {
3553 : struct vm_area_struct *vma;
3554 : struct anon_vma_chain *avc;
3555 :
3556 0 : BUG_ON(mmap_read_trylock(mm));
3557 :
3558 0 : mutex_lock(&mm_all_locks_mutex);
3559 :
3560 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3561 0 : if (signal_pending(current))
3562 : goto out_unlock;
3563 : if (vma->vm_file && vma->vm_file->f_mapping &&
3564 : is_vm_hugetlb_page(vma))
3565 : vm_lock_mapping(mm, vma->vm_file->f_mapping);
3566 : }
3567 :
3568 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3569 0 : if (signal_pending(current))
3570 : goto out_unlock;
3571 0 : if (vma->vm_file && vma->vm_file->f_mapping &&
3572 0 : !is_vm_hugetlb_page(vma))
3573 0 : vm_lock_mapping(mm, vma->vm_file->f_mapping);
3574 : }
3575 :
3576 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3577 0 : if (signal_pending(current))
3578 : goto out_unlock;
3579 0 : if (vma->anon_vma)
3580 0 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3581 0 : vm_lock_anon_vma(mm, avc->anon_vma);
3582 : }
3583 :
3584 : return 0;
3585 :
3586 : out_unlock:
3587 0 : mm_drop_all_locks(mm);
3588 0 : return -EINTR;
3589 : }
3590 :
3591 0 : static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3592 : {
3593 0 : if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3594 : /*
3595 : * The LSB of head.next can't change to 0 from under
3596 : * us because we hold the mm_all_locks_mutex.
3597 : *
3598 : * We must however clear the bitflag before unlocking
3599 : * the vma so the users using the anon_vma->rb_root will
3600 : * never see our bitflag.
3601 : *
3602 : * No need of atomic instructions here, head.next
3603 : * can't change from under us until we release the
3604 : * anon_vma->root->rwsem.
3605 : */
3606 0 : if (!__test_and_clear_bit(0, (unsigned long *)
3607 0 : &anon_vma->root->rb_root.rb_root.rb_node))
3608 0 : BUG();
3609 0 : anon_vma_unlock_write(anon_vma);
3610 : }
3611 0 : }
3612 :
3613 0 : static void vm_unlock_mapping(struct address_space *mapping)
3614 : {
3615 0 : if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3616 : /*
3617 : * AS_MM_ALL_LOCKS can't change to 0 from under us
3618 : * because we hold the mm_all_locks_mutex.
3619 : */
3620 0 : i_mmap_unlock_write(mapping);
3621 0 : if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3622 0 : &mapping->flags))
3623 0 : BUG();
3624 : }
3625 0 : }
3626 :
3627 : /*
3628 : * The mmap_lock cannot be released by the caller until
3629 : * mm_drop_all_locks() returns.
3630 : */
3631 0 : void mm_drop_all_locks(struct mm_struct *mm)
3632 : {
3633 : struct vm_area_struct *vma;
3634 : struct anon_vma_chain *avc;
3635 :
3636 0 : BUG_ON(mmap_read_trylock(mm));
3637 0 : BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3638 :
3639 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3640 0 : if (vma->anon_vma)
3641 0 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3642 0 : vm_unlock_anon_vma(avc->anon_vma);
3643 0 : if (vma->vm_file && vma->vm_file->f_mapping)
3644 0 : vm_unlock_mapping(vma->vm_file->f_mapping);
3645 : }
3646 :
3647 0 : mutex_unlock(&mm_all_locks_mutex);
3648 0 : }
3649 :
3650 : /*
3651 : * initialise the percpu counter for VM
3652 : */
3653 1 : void __init mmap_init(void)
3654 : {
3655 : int ret;
3656 :
3657 2 : ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3658 : VM_BUG_ON(ret);
3659 1 : }
3660 :
3661 : /*
3662 : * Initialise sysctl_user_reserve_kbytes.
3663 : *
3664 : * This is intended to prevent a user from starting a single memory hogging
3665 : * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3666 : * mode.
3667 : *
3668 : * The default value is min(3% of free memory, 128MB)
3669 : * 128MB is enough to recover with sshd/login, bash, and top/kill.
3670 : */
3671 1 : static int init_user_reserve(void)
3672 : {
3673 : unsigned long free_kbytes;
3674 :
3675 1 : free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3676 :
3677 1 : sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3678 1 : return 0;
3679 : }
3680 : subsys_initcall(init_user_reserve);
3681 :
3682 : /*
3683 : * Initialise sysctl_admin_reserve_kbytes.
3684 : *
3685 : * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3686 : * to log in and kill a memory hogging process.
3687 : *
3688 : * Systems with more than 256MB will reserve 8MB, enough to recover
3689 : * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3690 : * only reserve 3% of free pages by default.
3691 : */
3692 1 : static int init_admin_reserve(void)
3693 : {
3694 : unsigned long free_kbytes;
3695 :
3696 1 : free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3697 :
3698 1 : sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3699 1 : return 0;
3700 : }
3701 : subsys_initcall(init_admin_reserve);
3702 :
3703 : /*
3704 : * Reinititalise user and admin reserves if memory is added or removed.
3705 : *
3706 : * The default user reserve max is 128MB, and the default max for the
3707 : * admin reserve is 8MB. These are usually, but not always, enough to
3708 : * enable recovery from a memory hogging process using login/sshd, a shell,
3709 : * and tools like top. It may make sense to increase or even disable the
3710 : * reserve depending on the existence of swap or variations in the recovery
3711 : * tools. So, the admin may have changed them.
3712 : *
3713 : * If memory is added and the reserves have been eliminated or increased above
3714 : * the default max, then we'll trust the admin.
3715 : *
3716 : * If memory is removed and there isn't enough free memory, then we
3717 : * need to reset the reserves.
3718 : *
3719 : * Otherwise keep the reserve set by the admin.
3720 : */
3721 : static int reserve_mem_notifier(struct notifier_block *nb,
3722 : unsigned long action, void *data)
3723 : {
3724 : unsigned long tmp, free_kbytes;
3725 :
3726 : switch (action) {
3727 : case MEM_ONLINE:
3728 : /* Default max is 128MB. Leave alone if modified by operator. */
3729 : tmp = sysctl_user_reserve_kbytes;
3730 : if (0 < tmp && tmp < (1UL << 17))
3731 : init_user_reserve();
3732 :
3733 : /* Default max is 8MB. Leave alone if modified by operator. */
3734 : tmp = sysctl_admin_reserve_kbytes;
3735 : if (0 < tmp && tmp < (1UL << 13))
3736 : init_admin_reserve();
3737 :
3738 : break;
3739 : case MEM_OFFLINE:
3740 : free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3741 :
3742 : if (sysctl_user_reserve_kbytes > free_kbytes) {
3743 : init_user_reserve();
3744 : pr_info("vm.user_reserve_kbytes reset to %lu\n",
3745 : sysctl_user_reserve_kbytes);
3746 : }
3747 :
3748 : if (sysctl_admin_reserve_kbytes > free_kbytes) {
3749 : init_admin_reserve();
3750 : pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3751 : sysctl_admin_reserve_kbytes);
3752 : }
3753 : break;
3754 : default:
3755 : break;
3756 : }
3757 : return NOTIFY_OK;
3758 : }
3759 :
3760 : static struct notifier_block reserve_mem_nb = {
3761 : .notifier_call = reserve_mem_notifier,
3762 : };
3763 :
3764 1 : static int __meminit init_reserve_notifier(void)
3765 : {
3766 : if (register_hotmemory_notifier(&reserve_mem_nb))
3767 : pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3768 :
3769 1 : return 0;
3770 : }
3771 : subsys_initcall(init_reserve_notifier);
|
