]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - mm/mmap.c
Merge tag 'powerpc-4.7-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
[mirror_ubuntu-bionic-kernel.git] / mm / mmap.c
1 /*
2 * mm/mmap.c
3 *
4 * Written by obz.
5 *
6 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
7 */
8
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11 #include <linux/kernel.h>
12 #include <linux/slab.h>
13 #include <linux/backing-dev.h>
14 #include <linux/mm.h>
15 #include <linux/vmacache.h>
16 #include <linux/shm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/syscalls.h>
21 #include <linux/capability.h>
22 #include <linux/init.h>
23 #include <linux/file.h>
24 #include <linux/fs.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/hugetlb.h>
28 #include <linux/profile.h>
29 #include <linux/export.h>
30 #include <linux/mount.h>
31 #include <linux/mempolicy.h>
32 #include <linux/rmap.h>
33 #include <linux/mmu_notifier.h>
34 #include <linux/mmdebug.h>
35 #include <linux/perf_event.h>
36 #include <linux/audit.h>
37 #include <linux/khugepaged.h>
38 #include <linux/uprobes.h>
39 #include <linux/rbtree_augmented.h>
40 #include <linux/notifier.h>
41 #include <linux/memory.h>
42 #include <linux/printk.h>
43 #include <linux/userfaultfd_k.h>
44 #include <linux/moduleparam.h>
45 #include <linux/pkeys.h>
46
47 #include <asm/uaccess.h>
48 #include <asm/cacheflush.h>
49 #include <asm/tlb.h>
50 #include <asm/mmu_context.h>
51
52 #include "internal.h"
53
54 #ifndef arch_mmap_check
55 #define arch_mmap_check(addr, len, flags) (0)
56 #endif
57
58 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
59 const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
60 const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
61 int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
62 #endif
63 #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
64 const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
65 const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
66 int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
67 #endif
68
69 static bool ignore_rlimit_data = true;
70 core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
71
72 static void unmap_region(struct mm_struct *mm,
73 struct vm_area_struct *vma, struct vm_area_struct *prev,
74 unsigned long start, unsigned long end);
75
76 /* description of effects of mapping type and prot in current implementation.
77 * this is due to the limited x86 page protection hardware. The expected
78 * behavior is in parens:
79 *
80 * map_type prot
81 * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
82 * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
83 * w: (no) no w: (no) no w: (yes) yes w: (no) no
84 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
85 *
86 * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 * w: (no) no w: (no) no w: (copy) copy w: (no) no
88 * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
89 *
90 */
91 pgprot_t protection_map[16] = {
92 __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
93 __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
94 };
95
96 pgprot_t vm_get_page_prot(unsigned long vm_flags)
97 {
98 return __pgprot(pgprot_val(protection_map[vm_flags &
99 (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
100 pgprot_val(arch_vm_get_page_prot(vm_flags)));
101 }
102 EXPORT_SYMBOL(vm_get_page_prot);
103
104 static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
105 {
106 return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
107 }
108
109 /* Update vma->vm_page_prot to reflect vma->vm_flags. */
110 void vma_set_page_prot(struct vm_area_struct *vma)
111 {
112 unsigned long vm_flags = vma->vm_flags;
113
114 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
115 if (vma_wants_writenotify(vma)) {
116 vm_flags &= ~VM_SHARED;
117 vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot,
118 vm_flags);
119 }
120 }
121
122 /*
123 * Requires inode->i_mapping->i_mmap_rwsem
124 */
125 static void __remove_shared_vm_struct(struct vm_area_struct *vma,
126 struct file *file, struct address_space *mapping)
127 {
128 if (vma->vm_flags & VM_DENYWRITE)
129 atomic_inc(&file_inode(file)->i_writecount);
130 if (vma->vm_flags & VM_SHARED)
131 mapping_unmap_writable(mapping);
132
133 flush_dcache_mmap_lock(mapping);
134 vma_interval_tree_remove(vma, &mapping->i_mmap);
135 flush_dcache_mmap_unlock(mapping);
136 }
137
138 /*
139 * Unlink a file-based vm structure from its interval tree, to hide
140 * vma from rmap and vmtruncate before freeing its page tables.
141 */
142 void unlink_file_vma(struct vm_area_struct *vma)
143 {
144 struct file *file = vma->vm_file;
145
146 if (file) {
147 struct address_space *mapping = file->f_mapping;
148 i_mmap_lock_write(mapping);
149 __remove_shared_vm_struct(vma, file, mapping);
150 i_mmap_unlock_write(mapping);
151 }
152 }
153
154 /*
155 * Close a vm structure and free it, returning the next.
156 */
157 static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
158 {
159 struct vm_area_struct *next = vma->vm_next;
160
161 might_sleep();
162 if (vma->vm_ops && vma->vm_ops->close)
163 vma->vm_ops->close(vma);
164 if (vma->vm_file)
165 fput(vma->vm_file);
166 mpol_put(vma_policy(vma));
167 kmem_cache_free(vm_area_cachep, vma);
168 return next;
169 }
170
171 static unsigned long do_brk(unsigned long addr, unsigned long len);
172
173 SYSCALL_DEFINE1(brk, unsigned long, brk)
174 {
175 unsigned long retval;
176 unsigned long newbrk, oldbrk;
177 struct mm_struct *mm = current->mm;
178 unsigned long min_brk;
179 bool populate;
180
181 down_write(&mm->mmap_sem);
182
183 #ifdef CONFIG_COMPAT_BRK
184 /*
185 * CONFIG_COMPAT_BRK can still be overridden by setting
186 * randomize_va_space to 2, which will still cause mm->start_brk
187 * to be arbitrarily shifted
188 */
189 if (current->brk_randomized)
190 min_brk = mm->start_brk;
191 else
192 min_brk = mm->end_data;
193 #else
194 min_brk = mm->start_brk;
195 #endif
196 if (brk < min_brk)
197 goto out;
198
199 /*
200 * Check against rlimit here. If this check is done later after the test
201 * of oldbrk with newbrk then it can escape the test and let the data
202 * segment grow beyond its set limit the in case where the limit is
203 * not page aligned -Ram Gupta
204 */
205 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
206 mm->end_data, mm->start_data))
207 goto out;
208
209 newbrk = PAGE_ALIGN(brk);
210 oldbrk = PAGE_ALIGN(mm->brk);
211 if (oldbrk == newbrk)
212 goto set_brk;
213
214 /* Always allow shrinking brk. */
215 if (brk <= mm->brk) {
216 if (!do_munmap(mm, newbrk, oldbrk-newbrk))
217 goto set_brk;
218 goto out;
219 }
220
221 /* Check against existing mmap mappings. */
222 if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE))
223 goto out;
224
225 /* Ok, looks good - let it rip. */
226 if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk)
227 goto out;
228
229 set_brk:
230 mm->brk = brk;
231 populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
232 up_write(&mm->mmap_sem);
233 if (populate)
234 mm_populate(oldbrk, newbrk - oldbrk);
235 return brk;
236
237 out:
238 retval = mm->brk;
239 up_write(&mm->mmap_sem);
240 return retval;
241 }
242
243 static long vma_compute_subtree_gap(struct vm_area_struct *vma)
244 {
245 unsigned long max, subtree_gap;
246 max = vma->vm_start;
247 if (vma->vm_prev)
248 max -= vma->vm_prev->vm_end;
249 if (vma->vm_rb.rb_left) {
250 subtree_gap = rb_entry(vma->vm_rb.rb_left,
251 struct vm_area_struct, vm_rb)->rb_subtree_gap;
252 if (subtree_gap > max)
253 max = subtree_gap;
254 }
255 if (vma->vm_rb.rb_right) {
256 subtree_gap = rb_entry(vma->vm_rb.rb_right,
257 struct vm_area_struct, vm_rb)->rb_subtree_gap;
258 if (subtree_gap > max)
259 max = subtree_gap;
260 }
261 return max;
262 }
263
264 #ifdef CONFIG_DEBUG_VM_RB
265 static int browse_rb(struct mm_struct *mm)
266 {
267 struct rb_root *root = &mm->mm_rb;
268 int i = 0, j, bug = 0;
269 struct rb_node *nd, *pn = NULL;
270 unsigned long prev = 0, pend = 0;
271
272 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
273 struct vm_area_struct *vma;
274 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
275 if (vma->vm_start < prev) {
276 pr_emerg("vm_start %lx < prev %lx\n",
277 vma->vm_start, prev);
278 bug = 1;
279 }
280 if (vma->vm_start < pend) {
281 pr_emerg("vm_start %lx < pend %lx\n",
282 vma->vm_start, pend);
283 bug = 1;
284 }
285 if (vma->vm_start > vma->vm_end) {
286 pr_emerg("vm_start %lx > vm_end %lx\n",
287 vma->vm_start, vma->vm_end);
288 bug = 1;
289 }
290 spin_lock(&mm->page_table_lock);
291 if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
292 pr_emerg("free gap %lx, correct %lx\n",
293 vma->rb_subtree_gap,
294 vma_compute_subtree_gap(vma));
295 bug = 1;
296 }
297 spin_unlock(&mm->page_table_lock);
298 i++;
299 pn = nd;
300 prev = vma->vm_start;
301 pend = vma->vm_end;
302 }
303 j = 0;
304 for (nd = pn; nd; nd = rb_prev(nd))
305 j++;
306 if (i != j) {
307 pr_emerg("backwards %d, forwards %d\n", j, i);
308 bug = 1;
309 }
310 return bug ? -1 : i;
311 }
312
313 static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
314 {
315 struct rb_node *nd;
316
317 for (nd = rb_first(root); nd; nd = rb_next(nd)) {
318 struct vm_area_struct *vma;
319 vma = rb_entry(nd, struct vm_area_struct, vm_rb);
320 VM_BUG_ON_VMA(vma != ignore &&
321 vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
322 vma);
323 }
324 }
325
326 static void validate_mm(struct mm_struct *mm)
327 {
328 int bug = 0;
329 int i = 0;
330 unsigned long highest_address = 0;
331 struct vm_area_struct *vma = mm->mmap;
332
333 while (vma) {
334 struct anon_vma *anon_vma = vma->anon_vma;
335 struct anon_vma_chain *avc;
336
337 if (anon_vma) {
338 anon_vma_lock_read(anon_vma);
339 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
340 anon_vma_interval_tree_verify(avc);
341 anon_vma_unlock_read(anon_vma);
342 }
343
344 highest_address = vma->vm_end;
345 vma = vma->vm_next;
346 i++;
347 }
348 if (i != mm->map_count) {
349 pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
350 bug = 1;
351 }
352 if (highest_address != mm->highest_vm_end) {
353 pr_emerg("mm->highest_vm_end %lx, found %lx\n",
354 mm->highest_vm_end, highest_address);
355 bug = 1;
356 }
357 i = browse_rb(mm);
358 if (i != mm->map_count) {
359 if (i != -1)
360 pr_emerg("map_count %d rb %d\n", mm->map_count, i);
361 bug = 1;
362 }
363 VM_BUG_ON_MM(bug, mm);
364 }
365 #else
366 #define validate_mm_rb(root, ignore) do { } while (0)
367 #define validate_mm(mm) do { } while (0)
368 #endif
369
370 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
371 unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
372
373 /*
374 * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
375 * vma->vm_prev->vm_end values changed, without modifying the vma's position
376 * in the rbtree.
377 */
378 static void vma_gap_update(struct vm_area_struct *vma)
379 {
380 /*
381 * As it turns out, RB_DECLARE_CALLBACKS() already created a callback
382 * function that does exacltly what we want.
383 */
384 vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
385 }
386
387 static inline void vma_rb_insert(struct vm_area_struct *vma,
388 struct rb_root *root)
389 {
390 /* All rb_subtree_gap values must be consistent prior to insertion */
391 validate_mm_rb(root, NULL);
392
393 rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
394 }
395
396 static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
397 {
398 /*
399 * All rb_subtree_gap values must be consistent prior to erase,
400 * with the possible exception of the vma being erased.
401 */
402 validate_mm_rb(root, vma);
403
404 /*
405 * Note rb_erase_augmented is a fairly large inline function,
406 * so make sure we instantiate it only once with our desired
407 * augmented rbtree callbacks.
408 */
409 rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
410 }
411
412 /*
413 * vma has some anon_vma assigned, and is already inserted on that
414 * anon_vma's interval trees.
415 *
416 * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
417 * vma must be removed from the anon_vma's interval trees using
418 * anon_vma_interval_tree_pre_update_vma().
419 *
420 * After the update, the vma will be reinserted using
421 * anon_vma_interval_tree_post_update_vma().
422 *
423 * The entire update must be protected by exclusive mmap_sem and by
424 * the root anon_vma's mutex.
425 */
426 static inline void
427 anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
428 {
429 struct anon_vma_chain *avc;
430
431 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
432 anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
433 }
434
435 static inline void
436 anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
437 {
438 struct anon_vma_chain *avc;
439
440 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
441 anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
442 }
443
444 static int find_vma_links(struct mm_struct *mm, unsigned long addr,
445 unsigned long end, struct vm_area_struct **pprev,
446 struct rb_node ***rb_link, struct rb_node **rb_parent)
447 {
448 struct rb_node **__rb_link, *__rb_parent, *rb_prev;
449
450 __rb_link = &mm->mm_rb.rb_node;
451 rb_prev = __rb_parent = NULL;
452
453 while (*__rb_link) {
454 struct vm_area_struct *vma_tmp;
455
456 __rb_parent = *__rb_link;
457 vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
458
459 if (vma_tmp->vm_end > addr) {
460 /* Fail if an existing vma overlaps the area */
461 if (vma_tmp->vm_start < end)
462 return -ENOMEM;
463 __rb_link = &__rb_parent->rb_left;
464 } else {
465 rb_prev = __rb_parent;
466 __rb_link = &__rb_parent->rb_right;
467 }
468 }
469
470 *pprev = NULL;
471 if (rb_prev)
472 *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
473 *rb_link = __rb_link;
474 *rb_parent = __rb_parent;
475 return 0;
476 }
477
478 static unsigned long count_vma_pages_range(struct mm_struct *mm,
479 unsigned long addr, unsigned long end)
480 {
481 unsigned long nr_pages = 0;
482 struct vm_area_struct *vma;
483
484 /* Find first overlaping mapping */
485 vma = find_vma_intersection(mm, addr, end);
486 if (!vma)
487 return 0;
488
489 nr_pages = (min(end, vma->vm_end) -
490 max(addr, vma->vm_start)) >> PAGE_SHIFT;
491
492 /* Iterate over the rest of the overlaps */
493 for (vma = vma->vm_next; vma; vma = vma->vm_next) {
494 unsigned long overlap_len;
495
496 if (vma->vm_start > end)
497 break;
498
499 overlap_len = min(end, vma->vm_end) - vma->vm_start;
500 nr_pages += overlap_len >> PAGE_SHIFT;
501 }
502
503 return nr_pages;
504 }
505
506 void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
507 struct rb_node **rb_link, struct rb_node *rb_parent)
508 {
509 /* Update tracking information for the gap following the new vma. */
510 if (vma->vm_next)
511 vma_gap_update(vma->vm_next);
512 else
513 mm->highest_vm_end = vma->vm_end;
514
515 /*
516 * vma->vm_prev wasn't known when we followed the rbtree to find the
517 * correct insertion point for that vma. As a result, we could not
518 * update the vma vm_rb parents rb_subtree_gap values on the way down.
519 * So, we first insert the vma with a zero rb_subtree_gap value
520 * (to be consistent with what we did on the way down), and then
521 * immediately update the gap to the correct value. Finally we
522 * rebalance the rbtree after all augmented values have been set.
523 */
524 rb_link_node(&vma->vm_rb, rb_parent, rb_link);
525 vma->rb_subtree_gap = 0;
526 vma_gap_update(vma);
527 vma_rb_insert(vma, &mm->mm_rb);
528 }
529
530 static void __vma_link_file(struct vm_area_struct *vma)
531 {
532 struct file *file;
533
534 file = vma->vm_file;
535 if (file) {
536 struct address_space *mapping = file->f_mapping;
537
538 if (vma->vm_flags & VM_DENYWRITE)
539 atomic_dec(&file_inode(file)->i_writecount);
540 if (vma->vm_flags & VM_SHARED)
541 atomic_inc(&mapping->i_mmap_writable);
542
543 flush_dcache_mmap_lock(mapping);
544 vma_interval_tree_insert(vma, &mapping->i_mmap);
545 flush_dcache_mmap_unlock(mapping);
546 }
547 }
548
549 static void
550 __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
551 struct vm_area_struct *prev, struct rb_node **rb_link,
552 struct rb_node *rb_parent)
553 {
554 __vma_link_list(mm, vma, prev, rb_parent);
555 __vma_link_rb(mm, vma, rb_link, rb_parent);
556 }
557
558 static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
559 struct vm_area_struct *prev, struct rb_node **rb_link,
560 struct rb_node *rb_parent)
561 {
562 struct address_space *mapping = NULL;
563
564 if (vma->vm_file) {
565 mapping = vma->vm_file->f_mapping;
566 i_mmap_lock_write(mapping);
567 }
568
569 __vma_link(mm, vma, prev, rb_link, rb_parent);
570 __vma_link_file(vma);
571
572 if (mapping)
573 i_mmap_unlock_write(mapping);
574
575 mm->map_count++;
576 validate_mm(mm);
577 }
578
579 /*
580 * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
581 * mm's list and rbtree. It has already been inserted into the interval tree.
582 */
583 static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
584 {
585 struct vm_area_struct *prev;
586 struct rb_node **rb_link, *rb_parent;
587
588 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
589 &prev, &rb_link, &rb_parent))
590 BUG();
591 __vma_link(mm, vma, prev, rb_link, rb_parent);
592 mm->map_count++;
593 }
594
595 static inline void
596 __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma,
597 struct vm_area_struct *prev)
598 {
599 struct vm_area_struct *next;
600
601 vma_rb_erase(vma, &mm->mm_rb);
602 prev->vm_next = next = vma->vm_next;
603 if (next)
604 next->vm_prev = prev;
605
606 /* Kill the cache */
607 vmacache_invalidate(mm);
608 }
609
610 /*
611 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
612 * is already present in an i_mmap tree without adjusting the tree.
613 * The following helper function should be used when such adjustments
614 * are necessary. The "insert" vma (if any) is to be inserted
615 * before we drop the necessary locks.
616 */
617 int vma_adjust(struct vm_area_struct *vma, unsigned long start,
618 unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
619 {
620 struct mm_struct *mm = vma->vm_mm;
621 struct vm_area_struct *next = vma->vm_next;
622 struct vm_area_struct *importer = NULL;
623 struct address_space *mapping = NULL;
624 struct rb_root *root = NULL;
625 struct anon_vma *anon_vma = NULL;
626 struct file *file = vma->vm_file;
627 bool start_changed = false, end_changed = false;
628 long adjust_next = 0;
629 int remove_next = 0;
630
631 if (next && !insert) {
632 struct vm_area_struct *exporter = NULL;
633
634 if (end >= next->vm_end) {
635 /*
636 * vma expands, overlapping all the next, and
637 * perhaps the one after too (mprotect case 6).
638 */
639 again: remove_next = 1 + (end > next->vm_end);
640 end = next->vm_end;
641 exporter = next;
642 importer = vma;
643 } else if (end > next->vm_start) {
644 /*
645 * vma expands, overlapping part of the next:
646 * mprotect case 5 shifting the boundary up.
647 */
648 adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
649 exporter = next;
650 importer = vma;
651 } else if (end < vma->vm_end) {
652 /*
653 * vma shrinks, and !insert tells it's not
654 * split_vma inserting another: so it must be
655 * mprotect case 4 shifting the boundary down.
656 */
657 adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
658 exporter = vma;
659 importer = next;
660 }
661
662 /*
663 * Easily overlooked: when mprotect shifts the boundary,
664 * make sure the expanding vma has anon_vma set if the
665 * shrinking vma had, to cover any anon pages imported.
666 */
667 if (exporter && exporter->anon_vma && !importer->anon_vma) {
668 int error;
669
670 importer->anon_vma = exporter->anon_vma;
671 error = anon_vma_clone(importer, exporter);
672 if (error)
673 return error;
674 }
675 }
676
677 if (file) {
678 mapping = file->f_mapping;
679 root = &mapping->i_mmap;
680 uprobe_munmap(vma, vma->vm_start, vma->vm_end);
681
682 if (adjust_next)
683 uprobe_munmap(next, next->vm_start, next->vm_end);
684
685 i_mmap_lock_write(mapping);
686 if (insert) {
687 /*
688 * Put into interval tree now, so instantiated pages
689 * are visible to arm/parisc __flush_dcache_page
690 * throughout; but we cannot insert into address
691 * space until vma start or end is updated.
692 */
693 __vma_link_file(insert);
694 }
695 }
696
697 vma_adjust_trans_huge(vma, start, end, adjust_next);
698
699 anon_vma = vma->anon_vma;
700 if (!anon_vma && adjust_next)
701 anon_vma = next->anon_vma;
702 if (anon_vma) {
703 VM_BUG_ON_VMA(adjust_next && next->anon_vma &&
704 anon_vma != next->anon_vma, next);
705 anon_vma_lock_write(anon_vma);
706 anon_vma_interval_tree_pre_update_vma(vma);
707 if (adjust_next)
708 anon_vma_interval_tree_pre_update_vma(next);
709 }
710
711 if (root) {
712 flush_dcache_mmap_lock(mapping);
713 vma_interval_tree_remove(vma, root);
714 if (adjust_next)
715 vma_interval_tree_remove(next, root);
716 }
717
718 if (start != vma->vm_start) {
719 vma->vm_start = start;
720 start_changed = true;
721 }
722 if (end != vma->vm_end) {
723 vma->vm_end = end;
724 end_changed = true;
725 }
726 vma->vm_pgoff = pgoff;
727 if (adjust_next) {
728 next->vm_start += adjust_next << PAGE_SHIFT;
729 next->vm_pgoff += adjust_next;
730 }
731
732 if (root) {
733 if (adjust_next)
734 vma_interval_tree_insert(next, root);
735 vma_interval_tree_insert(vma, root);
736 flush_dcache_mmap_unlock(mapping);
737 }
738
739 if (remove_next) {
740 /*
741 * vma_merge has merged next into vma, and needs
742 * us to remove next before dropping the locks.
743 */
744 __vma_unlink(mm, next, vma);
745 if (file)
746 __remove_shared_vm_struct(next, file, mapping);
747 } else if (insert) {
748 /*
749 * split_vma has split insert from vma, and needs
750 * us to insert it before dropping the locks
751 * (it may either follow vma or precede it).
752 */
753 __insert_vm_struct(mm, insert);
754 } else {
755 if (start_changed)
756 vma_gap_update(vma);
757 if (end_changed) {
758 if (!next)
759 mm->highest_vm_end = end;
760 else if (!adjust_next)
761 vma_gap_update(next);
762 }
763 }
764
765 if (anon_vma) {
766 anon_vma_interval_tree_post_update_vma(vma);
767 if (adjust_next)
768 anon_vma_interval_tree_post_update_vma(next);
769 anon_vma_unlock_write(anon_vma);
770 }
771 if (mapping)
772 i_mmap_unlock_write(mapping);
773
774 if (root) {
775 uprobe_mmap(vma);
776
777 if (adjust_next)
778 uprobe_mmap(next);
779 }
780
781 if (remove_next) {
782 if (file) {
783 uprobe_munmap(next, next->vm_start, next->vm_end);
784 fput(file);
785 }
786 if (next->anon_vma)
787 anon_vma_merge(vma, next);
788 mm->map_count--;
789 mpol_put(vma_policy(next));
790 kmem_cache_free(vm_area_cachep, next);
791 /*
792 * In mprotect's case 6 (see comments on vma_merge),
793 * we must remove another next too. It would clutter
794 * up the code too much to do both in one go.
795 */
796 next = vma->vm_next;
797 if (remove_next == 2)
798 goto again;
799 else if (next)
800 vma_gap_update(next);
801 else
802 mm->highest_vm_end = end;
803 }
804 if (insert && file)
805 uprobe_mmap(insert);
806
807 validate_mm(mm);
808
809 return 0;
810 }
811
812 /*
813 * If the vma has a ->close operation then the driver probably needs to release
814 * per-vma resources, so we don't attempt to merge those.
815 */
816 static inline int is_mergeable_vma(struct vm_area_struct *vma,
817 struct file *file, unsigned long vm_flags,
818 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
819 {
820 /*
821 * VM_SOFTDIRTY should not prevent from VMA merging, if we
822 * match the flags but dirty bit -- the caller should mark
823 * merged VMA as dirty. If dirty bit won't be excluded from
824 * comparison, we increase pressue on the memory system forcing
825 * the kernel to generate new VMAs when old one could be
826 * extended instead.
827 */
828 if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
829 return 0;
830 if (vma->vm_file != file)
831 return 0;
832 if (vma->vm_ops && vma->vm_ops->close)
833 return 0;
834 if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
835 return 0;
836 return 1;
837 }
838
839 static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
840 struct anon_vma *anon_vma2,
841 struct vm_area_struct *vma)
842 {
843 /*
844 * The list_is_singular() test is to avoid merging VMA cloned from
845 * parents. This can improve scalability caused by anon_vma lock.
846 */
847 if ((!anon_vma1 || !anon_vma2) && (!vma ||
848 list_is_singular(&vma->anon_vma_chain)))
849 return 1;
850 return anon_vma1 == anon_vma2;
851 }
852
853 /*
854 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
855 * in front of (at a lower virtual address and file offset than) the vma.
856 *
857 * We cannot merge two vmas if they have differently assigned (non-NULL)
858 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
859 *
860 * We don't check here for the merged mmap wrapping around the end of pagecache
861 * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
862 * wrap, nor mmaps which cover the final page at index -1UL.
863 */
864 static int
865 can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
866 struct anon_vma *anon_vma, struct file *file,
867 pgoff_t vm_pgoff,
868 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
869 {
870 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
871 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
872 if (vma->vm_pgoff == vm_pgoff)
873 return 1;
874 }
875 return 0;
876 }
877
878 /*
879 * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
880 * beyond (at a higher virtual address and file offset than) the vma.
881 *
882 * We cannot merge two vmas if they have differently assigned (non-NULL)
883 * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
884 */
885 static int
886 can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
887 struct anon_vma *anon_vma, struct file *file,
888 pgoff_t vm_pgoff,
889 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
890 {
891 if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
892 is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
893 pgoff_t vm_pglen;
894 vm_pglen = vma_pages(vma);
895 if (vma->vm_pgoff + vm_pglen == vm_pgoff)
896 return 1;
897 }
898 return 0;
899 }
900
901 /*
902 * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
903 * whether that can be merged with its predecessor or its successor.
904 * Or both (it neatly fills a hole).
905 *
906 * In most cases - when called for mmap, brk or mremap - [addr,end) is
907 * certain not to be mapped by the time vma_merge is called; but when
908 * called for mprotect, it is certain to be already mapped (either at
909 * an offset within prev, or at the start of next), and the flags of
910 * this area are about to be changed to vm_flags - and the no-change
911 * case has already been eliminated.
912 *
913 * The following mprotect cases have to be considered, where AAAA is
914 * the area passed down from mprotect_fixup, never extending beyond one
915 * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
916 *
917 * AAAA AAAA AAAA AAAA
918 * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
919 * cannot merge might become might become might become
920 * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
921 * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
922 * mremap move: PPPPNNNNNNNN 8
923 * AAAA
924 * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
925 * might become case 1 below case 2 below case 3 below
926 *
927 * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX:
928 * mprotect_fixup updates vm_flags & vm_page_prot on successful return.
929 */
930 struct vm_area_struct *vma_merge(struct mm_struct *mm,
931 struct vm_area_struct *prev, unsigned long addr,
932 unsigned long end, unsigned long vm_flags,
933 struct anon_vma *anon_vma, struct file *file,
934 pgoff_t pgoff, struct mempolicy *policy,
935 struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
936 {
937 pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
938 struct vm_area_struct *area, *next;
939 int err;
940
941 /*
942 * We later require that vma->vm_flags == vm_flags,
943 * so this tests vma->vm_flags & VM_SPECIAL, too.
944 */
945 if (vm_flags & VM_SPECIAL)
946 return NULL;
947
948 if (prev)
949 next = prev->vm_next;
950 else
951 next = mm->mmap;
952 area = next;
953 if (next && next->vm_end == end) /* cases 6, 7, 8 */
954 next = next->vm_next;
955
956 /*
957 * Can it merge with the predecessor?
958 */
959 if (prev && prev->vm_end == addr &&
960 mpol_equal(vma_policy(prev), policy) &&
961 can_vma_merge_after(prev, vm_flags,
962 anon_vma, file, pgoff,
963 vm_userfaultfd_ctx)) {
964 /*
965 * OK, it can. Can we now merge in the successor as well?
966 */
967 if (next && end == next->vm_start &&
968 mpol_equal(policy, vma_policy(next)) &&
969 can_vma_merge_before(next, vm_flags,
970 anon_vma, file,
971 pgoff+pglen,
972 vm_userfaultfd_ctx) &&
973 is_mergeable_anon_vma(prev->anon_vma,
974 next->anon_vma, NULL)) {
975 /* cases 1, 6 */
976 err = vma_adjust(prev, prev->vm_start,
977 next->vm_end, prev->vm_pgoff, NULL);
978 } else /* cases 2, 5, 7 */
979 err = vma_adjust(prev, prev->vm_start,
980 end, prev->vm_pgoff, NULL);
981 if (err)
982 return NULL;
983 khugepaged_enter_vma_merge(prev, vm_flags);
984 return prev;
985 }
986
987 /*
988 * Can this new request be merged in front of next?
989 */
990 if (next && end == next->vm_start &&
991 mpol_equal(policy, vma_policy(next)) &&
992 can_vma_merge_before(next, vm_flags,
993 anon_vma, file, pgoff+pglen,
994 vm_userfaultfd_ctx)) {
995 if (prev && addr < prev->vm_end) /* case 4 */
996 err = vma_adjust(prev, prev->vm_start,
997 addr, prev->vm_pgoff, NULL);
998 else /* cases 3, 8 */
999 err = vma_adjust(area, addr, next->vm_end,
1000 next->vm_pgoff - pglen, NULL);
1001 if (err)
1002 return NULL;
1003 khugepaged_enter_vma_merge(area, vm_flags);
1004 return area;
1005 }
1006
1007 return NULL;
1008 }
1009
1010 /*
1011 * Rough compatbility check to quickly see if it's even worth looking
1012 * at sharing an anon_vma.
1013 *
1014 * They need to have the same vm_file, and the flags can only differ
1015 * in things that mprotect may change.
1016 *
1017 * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1018 * we can merge the two vma's. For example, we refuse to merge a vma if
1019 * there is a vm_ops->close() function, because that indicates that the
1020 * driver is doing some kind of reference counting. But that doesn't
1021 * really matter for the anon_vma sharing case.
1022 */
1023 static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1024 {
1025 return a->vm_end == b->vm_start &&
1026 mpol_equal(vma_policy(a), vma_policy(b)) &&
1027 a->vm_file == b->vm_file &&
1028 !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
1029 b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1030 }
1031
1032 /*
1033 * Do some basic sanity checking to see if we can re-use the anon_vma
1034 * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1035 * the same as 'old', the other will be the new one that is trying
1036 * to share the anon_vma.
1037 *
1038 * NOTE! This runs with mm_sem held for reading, so it is possible that
1039 * the anon_vma of 'old' is concurrently in the process of being set up
1040 * by another page fault trying to merge _that_. But that's ok: if it
1041 * is being set up, that automatically means that it will be a singleton
1042 * acceptable for merging, so we can do all of this optimistically. But
1043 * we do that READ_ONCE() to make sure that we never re-load the pointer.
1044 *
1045 * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1046 * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1047 * is to return an anon_vma that is "complex" due to having gone through
1048 * a fork).
1049 *
1050 * We also make sure that the two vma's are compatible (adjacent,
1051 * and with the same memory policies). That's all stable, even with just
1052 * a read lock on the mm_sem.
1053 */
1054 static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1055 {
1056 if (anon_vma_compatible(a, b)) {
1057 struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1058
1059 if (anon_vma && list_is_singular(&old->anon_vma_chain))
1060 return anon_vma;
1061 }
1062 return NULL;
1063 }
1064
1065 /*
1066 * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1067 * neighbouring vmas for a suitable anon_vma, before it goes off
1068 * to allocate a new anon_vma. It checks because a repetitive
1069 * sequence of mprotects and faults may otherwise lead to distinct
1070 * anon_vmas being allocated, preventing vma merge in subsequent
1071 * mprotect.
1072 */
1073 struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1074 {
1075 struct anon_vma *anon_vma;
1076 struct vm_area_struct *near;
1077
1078 near = vma->vm_next;
1079 if (!near)
1080 goto try_prev;
1081
1082 anon_vma = reusable_anon_vma(near, vma, near);
1083 if (anon_vma)
1084 return anon_vma;
1085 try_prev:
1086 near = vma->vm_prev;
1087 if (!near)
1088 goto none;
1089
1090 anon_vma = reusable_anon_vma(near, near, vma);
1091 if (anon_vma)
1092 return anon_vma;
1093 none:
1094 /*
1095 * There's no absolute need to look only at touching neighbours:
1096 * we could search further afield for "compatible" anon_vmas.
1097 * But it would probably just be a waste of time searching,
1098 * or lead to too many vmas hanging off the same anon_vma.
1099 * We're trying to allow mprotect remerging later on,
1100 * not trying to minimize memory used for anon_vmas.
1101 */
1102 return NULL;
1103 }
1104
1105 /*
1106 * If a hint addr is less than mmap_min_addr change hint to be as
1107 * low as possible but still greater than mmap_min_addr
1108 */
1109 static inline unsigned long round_hint_to_min(unsigned long hint)
1110 {
1111 hint &= PAGE_MASK;
1112 if (((void *)hint != NULL) &&
1113 (hint < mmap_min_addr))
1114 return PAGE_ALIGN(mmap_min_addr);
1115 return hint;
1116 }
1117
1118 static inline int mlock_future_check(struct mm_struct *mm,
1119 unsigned long flags,
1120 unsigned long len)
1121 {
1122 unsigned long locked, lock_limit;
1123
1124 /* mlock MCL_FUTURE? */
1125 if (flags & VM_LOCKED) {
1126 locked = len >> PAGE_SHIFT;
1127 locked += mm->locked_vm;
1128 lock_limit = rlimit(RLIMIT_MEMLOCK);
1129 lock_limit >>= PAGE_SHIFT;
1130 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1131 return -EAGAIN;
1132 }
1133 return 0;
1134 }
1135
1136 /*
1137 * The caller must hold down_write(&current->mm->mmap_sem).
1138 */
1139 unsigned long do_mmap(struct file *file, unsigned long addr,
1140 unsigned long len, unsigned long prot,
1141 unsigned long flags, vm_flags_t vm_flags,
1142 unsigned long pgoff, unsigned long *populate)
1143 {
1144 struct mm_struct *mm = current->mm;
1145 int pkey = 0;
1146
1147 *populate = 0;
1148
1149 if (!len)
1150 return -EINVAL;
1151
1152 /*
1153 * Does the application expect PROT_READ to imply PROT_EXEC?
1154 *
1155 * (the exception is when the underlying filesystem is noexec
1156 * mounted, in which case we dont add PROT_EXEC.)
1157 */
1158 if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1159 if (!(file && path_noexec(&file->f_path)))
1160 prot |= PROT_EXEC;
1161
1162 if (!(flags & MAP_FIXED))
1163 addr = round_hint_to_min(addr);
1164
1165 /* Careful about overflows.. */
1166 len = PAGE_ALIGN(len);
1167 if (!len)
1168 return -ENOMEM;
1169
1170 /* offset overflow? */
1171 if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1172 return -EOVERFLOW;
1173
1174 /* Too many mappings? */
1175 if (mm->map_count > sysctl_max_map_count)
1176 return -ENOMEM;
1177
1178 /* Obtain the address to map to. we verify (or select) it and ensure
1179 * that it represents a valid section of the address space.
1180 */
1181 addr = get_unmapped_area(file, addr, len, pgoff, flags);
1182 if (offset_in_page(addr))
1183 return addr;
1184
1185 if (prot == PROT_EXEC) {
1186 pkey = execute_only_pkey(mm);
1187 if (pkey < 0)
1188 pkey = 0;
1189 }
1190
1191 /* Do simple checking here so the lower-level routines won't have
1192 * to. we assume access permissions have been handled by the open
1193 * of the memory object, so we don't do any here.
1194 */
1195 vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1196 mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1197
1198 if (flags & MAP_LOCKED)
1199 if (!can_do_mlock())
1200 return -EPERM;
1201
1202 if (mlock_future_check(mm, vm_flags, len))
1203 return -EAGAIN;
1204
1205 if (file) {
1206 struct inode *inode = file_inode(file);
1207
1208 switch (flags & MAP_TYPE) {
1209 case MAP_SHARED:
1210 if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE))
1211 return -EACCES;
1212
1213 /*
1214 * Make sure we don't allow writing to an append-only
1215 * file..
1216 */
1217 if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1218 return -EACCES;
1219
1220 /*
1221 * Make sure there are no mandatory locks on the file.
1222 */
1223 if (locks_verify_locked(file))
1224 return -EAGAIN;
1225
1226 vm_flags |= VM_SHARED | VM_MAYSHARE;
1227 if (!(file->f_mode & FMODE_WRITE))
1228 vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1229
1230 /* fall through */
1231 case MAP_PRIVATE:
1232 if (!(file->f_mode & FMODE_READ))
1233 return -EACCES;
1234 if (path_noexec(&file->f_path)) {
1235 if (vm_flags & VM_EXEC)
1236 return -EPERM;
1237 vm_flags &= ~VM_MAYEXEC;
1238 }
1239
1240 if (!file->f_op->mmap)
1241 return -ENODEV;
1242 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1243 return -EINVAL;
1244 break;
1245
1246 default:
1247 return -EINVAL;
1248 }
1249 } else {
1250 switch (flags & MAP_TYPE) {
1251 case MAP_SHARED:
1252 if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1253 return -EINVAL;
1254 /*
1255 * Ignore pgoff.
1256 */
1257 pgoff = 0;
1258 vm_flags |= VM_SHARED | VM_MAYSHARE;
1259 break;
1260 case MAP_PRIVATE:
1261 /*
1262 * Set pgoff according to addr for anon_vma.
1263 */
1264 pgoff = addr >> PAGE_SHIFT;
1265 break;
1266 default:
1267 return -EINVAL;
1268 }
1269 }
1270
1271 /*
1272 * Set 'VM_NORESERVE' if we should not account for the
1273 * memory use of this mapping.
1274 */
1275 if (flags & MAP_NORESERVE) {
1276 /* We honor MAP_NORESERVE if allowed to overcommit */
1277 if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1278 vm_flags |= VM_NORESERVE;
1279
1280 /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1281 if (file && is_file_hugepages(file))
1282 vm_flags |= VM_NORESERVE;
1283 }
1284
1285 addr = mmap_region(file, addr, len, vm_flags, pgoff);
1286 if (!IS_ERR_VALUE(addr) &&
1287 ((vm_flags & VM_LOCKED) ||
1288 (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1289 *populate = len;
1290 return addr;
1291 }
1292
1293 SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1294 unsigned long, prot, unsigned long, flags,
1295 unsigned long, fd, unsigned long, pgoff)
1296 {
1297 struct file *file = NULL;
1298 unsigned long retval;
1299
1300 if (!(flags & MAP_ANONYMOUS)) {
1301 audit_mmap_fd(fd, flags);
1302 file = fget(fd);
1303 if (!file)
1304 return -EBADF;
1305 if (is_file_hugepages(file))
1306 len = ALIGN(len, huge_page_size(hstate_file(file)));
1307 retval = -EINVAL;
1308 if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1309 goto out_fput;
1310 } else if (flags & MAP_HUGETLB) {
1311 struct user_struct *user = NULL;
1312 struct hstate *hs;
1313
1314 hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK);
1315 if (!hs)
1316 return -EINVAL;
1317
1318 len = ALIGN(len, huge_page_size(hs));
1319 /*
1320 * VM_NORESERVE is used because the reservations will be
1321 * taken when vm_ops->mmap() is called
1322 * A dummy user value is used because we are not locking
1323 * memory so no accounting is necessary
1324 */
1325 file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1326 VM_NORESERVE,
1327 &user, HUGETLB_ANONHUGE_INODE,
1328 (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1329 if (IS_ERR(file))
1330 return PTR_ERR(file);
1331 }
1332
1333 flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1334
1335 retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1336 out_fput:
1337 if (file)
1338 fput(file);
1339 return retval;
1340 }
1341
1342 #ifdef __ARCH_WANT_SYS_OLD_MMAP
1343 struct mmap_arg_struct {
1344 unsigned long addr;
1345 unsigned long len;
1346 unsigned long prot;
1347 unsigned long flags;
1348 unsigned long fd;
1349 unsigned long offset;
1350 };
1351
1352 SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1353 {
1354 struct mmap_arg_struct a;
1355
1356 if (copy_from_user(&a, arg, sizeof(a)))
1357 return -EFAULT;
1358 if (offset_in_page(a.offset))
1359 return -EINVAL;
1360
1361 return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1362 a.offset >> PAGE_SHIFT);
1363 }
1364 #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1365
1366 /*
1367 * Some shared mappigns will want the pages marked read-only
1368 * to track write events. If so, we'll downgrade vm_page_prot
1369 * to the private version (using protection_map[] without the
1370 * VM_SHARED bit).
1371 */
1372 int vma_wants_writenotify(struct vm_area_struct *vma)
1373 {
1374 vm_flags_t vm_flags = vma->vm_flags;
1375 const struct vm_operations_struct *vm_ops = vma->vm_ops;
1376
1377 /* If it was private or non-writable, the write bit is already clear */
1378 if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1379 return 0;
1380
1381 /* The backer wishes to know when pages are first written to? */
1382 if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1383 return 1;
1384
1385 /* The open routine did something to the protections that pgprot_modify
1386 * won't preserve? */
1387 if (pgprot_val(vma->vm_page_prot) !=
1388 pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags)))
1389 return 0;
1390
1391 /* Do we need to track softdirty? */
1392 if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1393 return 1;
1394
1395 /* Specialty mapping? */
1396 if (vm_flags & VM_PFNMAP)
1397 return 0;
1398
1399 /* Can the mapping track the dirty pages? */
1400 return vma->vm_file && vma->vm_file->f_mapping &&
1401 mapping_cap_account_dirty(vma->vm_file->f_mapping);
1402 }
1403
1404 /*
1405 * We account for memory if it's a private writeable mapping,
1406 * not hugepages and VM_NORESERVE wasn't set.
1407 */
1408 static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1409 {
1410 /*
1411 * hugetlb has its own accounting separate from the core VM
1412 * VM_HUGETLB may not be set yet so we cannot check for that flag.
1413 */
1414 if (file && is_file_hugepages(file))
1415 return 0;
1416
1417 return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1418 }
1419
1420 unsigned long mmap_region(struct file *file, unsigned long addr,
1421 unsigned long len, vm_flags_t vm_flags, unsigned long pgoff)
1422 {
1423 struct mm_struct *mm = current->mm;
1424 struct vm_area_struct *vma, *prev;
1425 int error;
1426 struct rb_node **rb_link, *rb_parent;
1427 unsigned long charged = 0;
1428
1429 /* Check against address space limit. */
1430 if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
1431 unsigned long nr_pages;
1432
1433 /*
1434 * MAP_FIXED may remove pages of mappings that intersects with
1435 * requested mapping. Account for the pages it would unmap.
1436 */
1437 nr_pages = count_vma_pages_range(mm, addr, addr + len);
1438
1439 if (!may_expand_vm(mm, vm_flags,
1440 (len >> PAGE_SHIFT) - nr_pages))
1441 return -ENOMEM;
1442 }
1443
1444 /* Clear old maps */
1445 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1446 &rb_parent)) {
1447 if (do_munmap(mm, addr, len))
1448 return -ENOMEM;
1449 }
1450
1451 /*
1452 * Private writable mapping: check memory availability
1453 */
1454 if (accountable_mapping(file, vm_flags)) {
1455 charged = len >> PAGE_SHIFT;
1456 if (security_vm_enough_memory_mm(mm, charged))
1457 return -ENOMEM;
1458 vm_flags |= VM_ACCOUNT;
1459 }
1460
1461 /*
1462 * Can we just expand an old mapping?
1463 */
1464 vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1465 NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
1466 if (vma)
1467 goto out;
1468
1469 /*
1470 * Determine the object being mapped and call the appropriate
1471 * specific mapper. the address has already been validated, but
1472 * not unmapped, but the maps are removed from the list.
1473 */
1474 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
1475 if (!vma) {
1476 error = -ENOMEM;
1477 goto unacct_error;
1478 }
1479
1480 vma->vm_mm = mm;
1481 vma->vm_start = addr;
1482 vma->vm_end = addr + len;
1483 vma->vm_flags = vm_flags;
1484 vma->vm_page_prot = vm_get_page_prot(vm_flags);
1485 vma->vm_pgoff = pgoff;
1486 INIT_LIST_HEAD(&vma->anon_vma_chain);
1487
1488 if (file) {
1489 if (vm_flags & VM_DENYWRITE) {
1490 error = deny_write_access(file);
1491 if (error)
1492 goto free_vma;
1493 }
1494 if (vm_flags & VM_SHARED) {
1495 error = mapping_map_writable(file->f_mapping);
1496 if (error)
1497 goto allow_write_and_free_vma;
1498 }
1499
1500 /* ->mmap() can change vma->vm_file, but must guarantee that
1501 * vma_link() below can deny write-access if VM_DENYWRITE is set
1502 * and map writably if VM_SHARED is set. This usually means the
1503 * new file must not have been exposed to user-space, yet.
1504 */
1505 vma->vm_file = get_file(file);
1506 error = file->f_op->mmap(file, vma);
1507 if (error)
1508 goto unmap_and_free_vma;
1509
1510 /* Can addr have changed??
1511 *
1512 * Answer: Yes, several device drivers can do it in their
1513 * f_op->mmap method. -DaveM
1514 * Bug: If addr is changed, prev, rb_link, rb_parent should
1515 * be updated for vma_link()
1516 */
1517 WARN_ON_ONCE(addr != vma->vm_start);
1518
1519 addr = vma->vm_start;
1520 vm_flags = vma->vm_flags;
1521 } else if (vm_flags & VM_SHARED) {
1522 error = shmem_zero_setup(vma);
1523 if (error)
1524 goto free_vma;
1525 }
1526
1527 vma_link(mm, vma, prev, rb_link, rb_parent);
1528 /* Once vma denies write, undo our temporary denial count */
1529 if (file) {
1530 if (vm_flags & VM_SHARED)
1531 mapping_unmap_writable(file->f_mapping);
1532 if (vm_flags & VM_DENYWRITE)
1533 allow_write_access(file);
1534 }
1535 file = vma->vm_file;
1536 out:
1537 perf_event_mmap(vma);
1538
1539 vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
1540 if (vm_flags & VM_LOCKED) {
1541 if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
1542 vma == get_gate_vma(current->mm)))
1543 mm->locked_vm += (len >> PAGE_SHIFT);
1544 else
1545 vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
1546 }
1547
1548 if (file)
1549 uprobe_mmap(vma);
1550
1551 /*
1552 * New (or expanded) vma always get soft dirty status.
1553 * Otherwise user-space soft-dirty page tracker won't
1554 * be able to distinguish situation when vma area unmapped,
1555 * then new mapped in-place (which must be aimed as
1556 * a completely new data area).
1557 */
1558 vma->vm_flags |= VM_SOFTDIRTY;
1559
1560 vma_set_page_prot(vma);
1561
1562 return addr;
1563
1564 unmap_and_free_vma:
1565 vma->vm_file = NULL;
1566 fput(file);
1567
1568 /* Undo any partial mapping done by a device driver. */
1569 unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1570 charged = 0;
1571 if (vm_flags & VM_SHARED)
1572 mapping_unmap_writable(file->f_mapping);
1573 allow_write_and_free_vma:
1574 if (vm_flags & VM_DENYWRITE)
1575 allow_write_access(file);
1576 free_vma:
1577 kmem_cache_free(vm_area_cachep, vma);
1578 unacct_error:
1579 if (charged)
1580 vm_unacct_memory(charged);
1581 return error;
1582 }
1583
1584 unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1585 {
1586 /*
1587 * We implement the search by looking for an rbtree node that
1588 * immediately follows a suitable gap. That is,
1589 * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1590 * - gap_end = vma->vm_start >= info->low_limit + length;
1591 * - gap_end - gap_start >= length
1592 */
1593
1594 struct mm_struct *mm = current->mm;
1595 struct vm_area_struct *vma;
1596 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1597
1598 /* Adjust search length to account for worst case alignment overhead */
1599 length = info->length + info->align_mask;
1600 if (length < info->length)
1601 return -ENOMEM;
1602
1603 /* Adjust search limits by the desired length */
1604 if (info->high_limit < length)
1605 return -ENOMEM;
1606 high_limit = info->high_limit - length;
1607
1608 if (info->low_limit > high_limit)
1609 return -ENOMEM;
1610 low_limit = info->low_limit + length;
1611
1612 /* Check if rbtree root looks promising */
1613 if (RB_EMPTY_ROOT(&mm->mm_rb))
1614 goto check_highest;
1615 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1616 if (vma->rb_subtree_gap < length)
1617 goto check_highest;
1618
1619 while (true) {
1620 /* Visit left subtree if it looks promising */
1621 gap_end = vma->vm_start;
1622 if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1623 struct vm_area_struct *left =
1624 rb_entry(vma->vm_rb.rb_left,
1625 struct vm_area_struct, vm_rb);
1626 if (left->rb_subtree_gap >= length) {
1627 vma = left;
1628 continue;
1629 }
1630 }
1631
1632 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1633 check_current:
1634 /* Check if current node has a suitable gap */
1635 if (gap_start > high_limit)
1636 return -ENOMEM;
1637 if (gap_end >= low_limit && gap_end - gap_start >= length)
1638 goto found;
1639
1640 /* Visit right subtree if it looks promising */
1641 if (vma->vm_rb.rb_right) {
1642 struct vm_area_struct *right =
1643 rb_entry(vma->vm_rb.rb_right,
1644 struct vm_area_struct, vm_rb);
1645 if (right->rb_subtree_gap >= length) {
1646 vma = right;
1647 continue;
1648 }
1649 }
1650
1651 /* Go back up the rbtree to find next candidate node */
1652 while (true) {
1653 struct rb_node *prev = &vma->vm_rb;
1654 if (!rb_parent(prev))
1655 goto check_highest;
1656 vma = rb_entry(rb_parent(prev),
1657 struct vm_area_struct, vm_rb);
1658 if (prev == vma->vm_rb.rb_left) {
1659 gap_start = vma->vm_prev->vm_end;
1660 gap_end = vma->vm_start;
1661 goto check_current;
1662 }
1663 }
1664 }
1665
1666 check_highest:
1667 /* Check highest gap, which does not precede any rbtree node */
1668 gap_start = mm->highest_vm_end;
1669 gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1670 if (gap_start > high_limit)
1671 return -ENOMEM;
1672
1673 found:
1674 /* We found a suitable gap. Clip it with the original low_limit. */
1675 if (gap_start < info->low_limit)
1676 gap_start = info->low_limit;
1677
1678 /* Adjust gap address to the desired alignment */
1679 gap_start += (info->align_offset - gap_start) & info->align_mask;
1680
1681 VM_BUG_ON(gap_start + info->length > info->high_limit);
1682 VM_BUG_ON(gap_start + info->length > gap_end);
1683 return gap_start;
1684 }
1685
1686 unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1687 {
1688 struct mm_struct *mm = current->mm;
1689 struct vm_area_struct *vma;
1690 unsigned long length, low_limit, high_limit, gap_start, gap_end;
1691
1692 /* Adjust search length to account for worst case alignment overhead */
1693 length = info->length + info->align_mask;
1694 if (length < info->length)
1695 return -ENOMEM;
1696
1697 /*
1698 * Adjust search limits by the desired length.
1699 * See implementation comment at top of unmapped_area().
1700 */
1701 gap_end = info->high_limit;
1702 if (gap_end < length)
1703 return -ENOMEM;
1704 high_limit = gap_end - length;
1705
1706 if (info->low_limit > high_limit)
1707 return -ENOMEM;
1708 low_limit = info->low_limit + length;
1709
1710 /* Check highest gap, which does not precede any rbtree node */
1711 gap_start = mm->highest_vm_end;
1712 if (gap_start <= high_limit)
1713 goto found_highest;
1714
1715 /* Check if rbtree root looks promising */
1716 if (RB_EMPTY_ROOT(&mm->mm_rb))
1717 return -ENOMEM;
1718 vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1719 if (vma->rb_subtree_gap < length)
1720 return -ENOMEM;
1721
1722 while (true) {
1723 /* Visit right subtree if it looks promising */
1724 gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0;
1725 if (gap_start <= high_limit && vma->vm_rb.rb_right) {
1726 struct vm_area_struct *right =
1727 rb_entry(vma->vm_rb.rb_right,
1728 struct vm_area_struct, vm_rb);
1729 if (right->rb_subtree_gap >= length) {
1730 vma = right;
1731 continue;
1732 }
1733 }
1734
1735 check_current:
1736 /* Check if current node has a suitable gap */
1737 gap_end = vma->vm_start;
1738 if (gap_end < low_limit)
1739 return -ENOMEM;
1740 if (gap_start <= high_limit && gap_end - gap_start >= length)
1741 goto found;
1742
1743 /* Visit left subtree if it looks promising */
1744 if (vma->vm_rb.rb_left) {
1745 struct vm_area_struct *left =
1746 rb_entry(vma->vm_rb.rb_left,
1747 struct vm_area_struct, vm_rb);
1748 if (left->rb_subtree_gap >= length) {
1749 vma = left;
1750 continue;
1751 }
1752 }
1753
1754 /* Go back up the rbtree to find next candidate node */
1755 while (true) {
1756 struct rb_node *prev = &vma->vm_rb;
1757 if (!rb_parent(prev))
1758 return -ENOMEM;
1759 vma = rb_entry(rb_parent(prev),
1760 struct vm_area_struct, vm_rb);
1761 if (prev == vma->vm_rb.rb_right) {
1762 gap_start = vma->vm_prev ?
1763 vma->vm_prev->vm_end : 0;
1764 goto check_current;
1765 }
1766 }
1767 }
1768
1769 found:
1770 /* We found a suitable gap. Clip it with the original high_limit. */
1771 if (gap_end > info->high_limit)
1772 gap_end = info->high_limit;
1773
1774 found_highest:
1775 /* Compute highest gap address at the desired alignment */
1776 gap_end -= info->length;
1777 gap_end -= (gap_end - info->align_offset) & info->align_mask;
1778
1779 VM_BUG_ON(gap_end < info->low_limit);
1780 VM_BUG_ON(gap_end < gap_start);
1781 return gap_end;
1782 }
1783
1784 /* Get an address range which is currently unmapped.
1785 * For shmat() with addr=0.
1786 *
1787 * Ugly calling convention alert:
1788 * Return value with the low bits set means error value,
1789 * ie
1790 * if (ret & ~PAGE_MASK)
1791 * error = ret;
1792 *
1793 * This function "knows" that -ENOMEM has the bits set.
1794 */
1795 #ifndef HAVE_ARCH_UNMAPPED_AREA
1796 unsigned long
1797 arch_get_unmapped_area(struct file *filp, unsigned long addr,
1798 unsigned long len, unsigned long pgoff, unsigned long flags)
1799 {
1800 struct mm_struct *mm = current->mm;
1801 struct vm_area_struct *vma;
1802 struct vm_unmapped_area_info info;
1803
1804 if (len > TASK_SIZE - mmap_min_addr)
1805 return -ENOMEM;
1806
1807 if (flags & MAP_FIXED)
1808 return addr;
1809
1810 if (addr) {
1811 addr = PAGE_ALIGN(addr);
1812 vma = find_vma(mm, addr);
1813 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1814 (!vma || addr + len <= vma->vm_start))
1815 return addr;
1816 }
1817
1818 info.flags = 0;
1819 info.length = len;
1820 info.low_limit = mm->mmap_base;
1821 info.high_limit = TASK_SIZE;
1822 info.align_mask = 0;
1823 return vm_unmapped_area(&info);
1824 }
1825 #endif
1826
1827 /*
1828 * This mmap-allocator allocates new areas top-down from below the
1829 * stack's low limit (the base):
1830 */
1831 #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
1832 unsigned long
1833 arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
1834 const unsigned long len, const unsigned long pgoff,
1835 const unsigned long flags)
1836 {
1837 struct vm_area_struct *vma;
1838 struct mm_struct *mm = current->mm;
1839 unsigned long addr = addr0;
1840 struct vm_unmapped_area_info info;
1841
1842 /* requested length too big for entire address space */
1843 if (len > TASK_SIZE - mmap_min_addr)
1844 return -ENOMEM;
1845
1846 if (flags & MAP_FIXED)
1847 return addr;
1848
1849 /* requesting a specific address */
1850 if (addr) {
1851 addr = PAGE_ALIGN(addr);
1852 vma = find_vma(mm, addr);
1853 if (TASK_SIZE - len >= addr && addr >= mmap_min_addr &&
1854 (!vma || addr + len <= vma->vm_start))
1855 return addr;
1856 }
1857
1858 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
1859 info.length = len;
1860 info.low_limit = max(PAGE_SIZE, mmap_min_addr);
1861 info.high_limit = mm->mmap_base;
1862 info.align_mask = 0;
1863 addr = vm_unmapped_area(&info);
1864
1865 /*
1866 * A failed mmap() very likely causes application failure,
1867 * so fall back to the bottom-up function here. This scenario
1868 * can happen with large stack limits and large mmap()
1869 * allocations.
1870 */
1871 if (offset_in_page(addr)) {
1872 VM_BUG_ON(addr != -ENOMEM);
1873 info.flags = 0;
1874 info.low_limit = TASK_UNMAPPED_BASE;
1875 info.high_limit = TASK_SIZE;
1876 addr = vm_unmapped_area(&info);
1877 }
1878
1879 return addr;
1880 }
1881 #endif
1882
1883 unsigned long
1884 get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
1885 unsigned long pgoff, unsigned long flags)
1886 {
1887 unsigned long (*get_area)(struct file *, unsigned long,
1888 unsigned long, unsigned long, unsigned long);
1889
1890 unsigned long error = arch_mmap_check(addr, len, flags);
1891 if (error)
1892 return error;
1893
1894 /* Careful about overflows.. */
1895 if (len > TASK_SIZE)
1896 return -ENOMEM;
1897
1898 get_area = current->mm->get_unmapped_area;
1899 if (file && file->f_op->get_unmapped_area)
1900 get_area = file->f_op->get_unmapped_area;
1901 addr = get_area(file, addr, len, pgoff, flags);
1902 if (IS_ERR_VALUE(addr))
1903 return addr;
1904
1905 if (addr > TASK_SIZE - len)
1906 return -ENOMEM;
1907 if (offset_in_page(addr))
1908 return -EINVAL;
1909
1910 error = security_mmap_addr(addr);
1911 return error ? error : addr;
1912 }
1913
1914 EXPORT_SYMBOL(get_unmapped_area);
1915
1916 /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
1917 struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
1918 {
1919 struct rb_node *rb_node;
1920 struct vm_area_struct *vma;
1921
1922 /* Check the cache first. */
1923 vma = vmacache_find(mm, addr);
1924 if (likely(vma))
1925 return vma;
1926
1927 rb_node = mm->mm_rb.rb_node;
1928
1929 while (rb_node) {
1930 struct vm_area_struct *tmp;
1931
1932 tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1933
1934 if (tmp->vm_end > addr) {
1935 vma = tmp;
1936 if (tmp->vm_start <= addr)
1937 break;
1938 rb_node = rb_node->rb_left;
1939 } else
1940 rb_node = rb_node->rb_right;
1941 }
1942
1943 if (vma)
1944 vmacache_update(addr, vma);
1945 return vma;
1946 }
1947
1948 EXPORT_SYMBOL(find_vma);
1949
1950 /*
1951 * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
1952 */
1953 struct vm_area_struct *
1954 find_vma_prev(struct mm_struct *mm, unsigned long addr,
1955 struct vm_area_struct **pprev)
1956 {
1957 struct vm_area_struct *vma;
1958
1959 vma = find_vma(mm, addr);
1960 if (vma) {
1961 *pprev = vma->vm_prev;
1962 } else {
1963 struct rb_node *rb_node = mm->mm_rb.rb_node;
1964 *pprev = NULL;
1965 while (rb_node) {
1966 *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb);
1967 rb_node = rb_node->rb_right;
1968 }
1969 }
1970 return vma;
1971 }
1972
1973 /*
1974 * Verify that the stack growth is acceptable and
1975 * update accounting. This is shared with both the
1976 * grow-up and grow-down cases.
1977 */
1978 static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow)
1979 {
1980 struct mm_struct *mm = vma->vm_mm;
1981 struct rlimit *rlim = current->signal->rlim;
1982 unsigned long new_start, actual_size;
1983
1984 /* address space limit tests */
1985 if (!may_expand_vm(mm, vma->vm_flags, grow))
1986 return -ENOMEM;
1987
1988 /* Stack limit test */
1989 actual_size = size;
1990 if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN)))
1991 actual_size -= PAGE_SIZE;
1992 if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur))
1993 return -ENOMEM;
1994
1995 /* mlock limit tests */
1996 if (vma->vm_flags & VM_LOCKED) {
1997 unsigned long locked;
1998 unsigned long limit;
1999 locked = mm->locked_vm + grow;
2000 limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur);
2001 limit >>= PAGE_SHIFT;
2002 if (locked > limit && !capable(CAP_IPC_LOCK))
2003 return -ENOMEM;
2004 }
2005
2006 /* Check to ensure the stack will not grow into a hugetlb-only region */
2007 new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2008 vma->vm_end - size;
2009 if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2010 return -EFAULT;
2011
2012 /*
2013 * Overcommit.. This must be the final test, as it will
2014 * update security statistics.
2015 */
2016 if (security_vm_enough_memory_mm(mm, grow))
2017 return -ENOMEM;
2018
2019 return 0;
2020 }
2021
2022 #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2023 /*
2024 * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2025 * vma is the last one with address > vma->vm_end. Have to extend vma.
2026 */
2027 int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2028 {
2029 struct mm_struct *mm = vma->vm_mm;
2030 int error = 0;
2031
2032 if (!(vma->vm_flags & VM_GROWSUP))
2033 return -EFAULT;
2034
2035 /* Guard against wrapping around to address 0. */
2036 if (address < PAGE_ALIGN(address+4))
2037 address = PAGE_ALIGN(address+4);
2038 else
2039 return -ENOMEM;
2040
2041 /* We must make sure the anon_vma is allocated. */
2042 if (unlikely(anon_vma_prepare(vma)))
2043 return -ENOMEM;
2044
2045 /*
2046 * vma->vm_start/vm_end cannot change under us because the caller
2047 * is required to hold the mmap_sem in read mode. We need the
2048 * anon_vma lock to serialize against concurrent expand_stacks.
2049 */
2050 anon_vma_lock_write(vma->anon_vma);
2051
2052 /* Somebody else might have raced and expanded it already */
2053 if (address > vma->vm_end) {
2054 unsigned long size, grow;
2055
2056 size = address - vma->vm_start;
2057 grow = (address - vma->vm_end) >> PAGE_SHIFT;
2058
2059 error = -ENOMEM;
2060 if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2061 error = acct_stack_growth(vma, size, grow);
2062 if (!error) {
2063 /*
2064 * vma_gap_update() doesn't support concurrent
2065 * updates, but we only hold a shared mmap_sem
2066 * lock here, so we need to protect against
2067 * concurrent vma expansions.
2068 * anon_vma_lock_write() doesn't help here, as
2069 * we don't guarantee that all growable vmas
2070 * in a mm share the same root anon vma.
2071 * So, we reuse mm->page_table_lock to guard
2072 * against concurrent vma expansions.
2073 */
2074 spin_lock(&mm->page_table_lock);
2075 if (vma->vm_flags & VM_LOCKED)
2076 mm->locked_vm += grow;
2077 vm_stat_account(mm, vma->vm_flags, grow);
2078 anon_vma_interval_tree_pre_update_vma(vma);
2079 vma->vm_end = address;
2080 anon_vma_interval_tree_post_update_vma(vma);
2081 if (vma->vm_next)
2082 vma_gap_update(vma->vm_next);
2083 else
2084 mm->highest_vm_end = address;
2085 spin_unlock(&mm->page_table_lock);
2086
2087 perf_event_mmap(vma);
2088 }
2089 }
2090 }
2091 anon_vma_unlock_write(vma->anon_vma);
2092 khugepaged_enter_vma_merge(vma, vma->vm_flags);
2093 validate_mm(mm);
2094 return error;
2095 }
2096 #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2097
2098 /*
2099 * vma is the first one with address < vma->vm_start. Have to extend vma.
2100 */
2101 int expand_downwards(struct vm_area_struct *vma,
2102 unsigned long address)
2103 {
2104 struct mm_struct *mm = vma->vm_mm;
2105 int error;
2106
2107 address &= PAGE_MASK;
2108 error = security_mmap_addr(address);
2109 if (error)
2110 return error;
2111
2112 /* We must make sure the anon_vma is allocated. */
2113 if (unlikely(anon_vma_prepare(vma)))
2114 return -ENOMEM;
2115
2116 /*
2117 * vma->vm_start/vm_end cannot change under us because the caller
2118 * is required to hold the mmap_sem in read mode. We need the
2119 * anon_vma lock to serialize against concurrent expand_stacks.
2120 */
2121 anon_vma_lock_write(vma->anon_vma);
2122
2123 /* Somebody else might have raced and expanded it already */
2124 if (address < vma->vm_start) {
2125 unsigned long size, grow;
2126
2127 size = vma->vm_end - address;
2128 grow = (vma->vm_start - address) >> PAGE_SHIFT;
2129
2130 error = -ENOMEM;
2131 if (grow <= vma->vm_pgoff) {
2132 error = acct_stack_growth(vma, size, grow);
2133 if (!error) {
2134 /*
2135 * vma_gap_update() doesn't support concurrent
2136 * updates, but we only hold a shared mmap_sem
2137 * lock here, so we need to protect against
2138 * concurrent vma expansions.
2139 * anon_vma_lock_write() doesn't help here, as
2140 * we don't guarantee that all growable vmas
2141 * in a mm share the same root anon vma.
2142 * So, we reuse mm->page_table_lock to guard
2143 * against concurrent vma expansions.
2144 */
2145 spin_lock(&mm->page_table_lock);
2146 if (vma->vm_flags & VM_LOCKED)
2147 mm->locked_vm += grow;
2148 vm_stat_account(mm, vma->vm_flags, grow);
2149 anon_vma_interval_tree_pre_update_vma(vma);
2150 vma->vm_start = address;
2151 vma->vm_pgoff -= grow;
2152 anon_vma_interval_tree_post_update_vma(vma);
2153 vma_gap_update(vma);
2154 spin_unlock(&mm->page_table_lock);
2155
2156 perf_event_mmap(vma);
2157 }
2158 }
2159 }
2160 anon_vma_unlock_write(vma->anon_vma);
2161 khugepaged_enter_vma_merge(vma, vma->vm_flags);
2162 validate_mm(mm);
2163 return error;
2164 }
2165
2166 /*
2167 * Note how expand_stack() refuses to expand the stack all the way to
2168 * abut the next virtual mapping, *unless* that mapping itself is also
2169 * a stack mapping. We want to leave room for a guard page, after all
2170 * (the guard page itself is not added here, that is done by the
2171 * actual page faulting logic)
2172 *
2173 * This matches the behavior of the guard page logic (see mm/memory.c:
2174 * check_stack_guard_page()), which only allows the guard page to be
2175 * removed under these circumstances.
2176 */
2177 #ifdef CONFIG_STACK_GROWSUP
2178 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2179 {
2180 struct vm_area_struct *next;
2181
2182 address &= PAGE_MASK;
2183 next = vma->vm_next;
2184 if (next && next->vm_start == address + PAGE_SIZE) {
2185 if (!(next->vm_flags & VM_GROWSUP))
2186 return -ENOMEM;
2187 }
2188 return expand_upwards(vma, address);
2189 }
2190
2191 struct vm_area_struct *
2192 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2193 {
2194 struct vm_area_struct *vma, *prev;
2195
2196 addr &= PAGE_MASK;
2197 vma = find_vma_prev(mm, addr, &prev);
2198 if (vma && (vma->vm_start <= addr))
2199 return vma;
2200 if (!prev || expand_stack(prev, addr))
2201 return NULL;
2202 if (prev->vm_flags & VM_LOCKED)
2203 populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2204 return prev;
2205 }
2206 #else
2207 int expand_stack(struct vm_area_struct *vma, unsigned long address)
2208 {
2209 struct vm_area_struct *prev;
2210
2211 address &= PAGE_MASK;
2212 prev = vma->vm_prev;
2213 if (prev && prev->vm_end == address) {
2214 if (!(prev->vm_flags & VM_GROWSDOWN))
2215 return -ENOMEM;
2216 }
2217 return expand_downwards(vma, address);
2218 }
2219
2220 struct vm_area_struct *
2221 find_extend_vma(struct mm_struct *mm, unsigned long addr)
2222 {
2223 struct vm_area_struct *vma;
2224 unsigned long start;
2225
2226 addr &= PAGE_MASK;
2227 vma = find_vma(mm, addr);
2228 if (!vma)
2229 return NULL;
2230 if (vma->vm_start <= addr)
2231 return vma;
2232 if (!(vma->vm_flags & VM_GROWSDOWN))
2233 return NULL;
2234 start = vma->vm_start;
2235 if (expand_stack(vma, addr))
2236 return NULL;
2237 if (vma->vm_flags & VM_LOCKED)
2238 populate_vma_page_range(vma, addr, start, NULL);
2239 return vma;
2240 }
2241 #endif
2242
2243 EXPORT_SYMBOL_GPL(find_extend_vma);
2244
2245 /*
2246 * Ok - we have the memory areas we should free on the vma list,
2247 * so release them, and do the vma updates.
2248 *
2249 * Called with the mm semaphore held.
2250 */
2251 static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2252 {
2253 unsigned long nr_accounted = 0;
2254
2255 /* Update high watermark before we lower total_vm */
2256 update_hiwater_vm(mm);
2257 do {
2258 long nrpages = vma_pages(vma);
2259
2260 if (vma->vm_flags & VM_ACCOUNT)
2261 nr_accounted += nrpages;
2262 vm_stat_account(mm, vma->vm_flags, -nrpages);
2263 vma = remove_vma(vma);
2264 } while (vma);
2265 vm_unacct_memory(nr_accounted);
2266 validate_mm(mm);
2267 }
2268
2269 /*
2270 * Get rid of page table information in the indicated region.
2271 *
2272 * Called with the mm semaphore held.
2273 */
2274 static void unmap_region(struct mm_struct *mm,
2275 struct vm_area_struct *vma, struct vm_area_struct *prev,
2276 unsigned long start, unsigned long end)
2277 {
2278 struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
2279 struct mmu_gather tlb;
2280
2281 lru_add_drain();
2282 tlb_gather_mmu(&tlb, mm, start, end);
2283 update_hiwater_rss(mm);
2284 unmap_vmas(&tlb, vma, start, end);
2285 free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2286 next ? next->vm_start : USER_PGTABLES_CEILING);
2287 tlb_finish_mmu(&tlb, start, end);
2288 }
2289
2290 /*
2291 * Create a list of vma's touched by the unmap, removing them from the mm's
2292 * vma list as we go..
2293 */
2294 static void
2295 detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2296 struct vm_area_struct *prev, unsigned long end)
2297 {
2298 struct vm_area_struct **insertion_point;
2299 struct vm_area_struct *tail_vma = NULL;
2300
2301 insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2302 vma->vm_prev = NULL;
2303 do {
2304 vma_rb_erase(vma, &mm->mm_rb);
2305 mm->map_count--;
2306 tail_vma = vma;
2307 vma = vma->vm_next;
2308 } while (vma && vma->vm_start < end);
2309 *insertion_point = vma;
2310 if (vma) {
2311 vma->vm_prev = prev;
2312 vma_gap_update(vma);
2313 } else
2314 mm->highest_vm_end = prev ? prev->vm_end : 0;
2315 tail_vma->vm_next = NULL;
2316
2317 /* Kill the cache */
2318 vmacache_invalidate(mm);
2319 }
2320
2321 /*
2322 * __split_vma() bypasses sysctl_max_map_count checking. We use this on the
2323 * munmap path where it doesn't make sense to fail.
2324 */
2325 static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2326 unsigned long addr, int new_below)
2327 {
2328 struct vm_area_struct *new;
2329 int err;
2330
2331 if (is_vm_hugetlb_page(vma) && (addr &
2332 ~(huge_page_mask(hstate_vma(vma)))))
2333 return -EINVAL;
2334
2335 new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2336 if (!new)
2337 return -ENOMEM;
2338
2339 /* most fields are the same, copy all, and then fixup */
2340 *new = *vma;
2341
2342 INIT_LIST_HEAD(&new->anon_vma_chain);
2343
2344 if (new_below)
2345 new->vm_end = addr;
2346 else {
2347 new->vm_start = addr;
2348 new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2349 }
2350
2351 err = vma_dup_policy(vma, new);
2352 if (err)
2353 goto out_free_vma;
2354
2355 err = anon_vma_clone(new, vma);
2356 if (err)
2357 goto out_free_mpol;
2358
2359 if (new->vm_file)
2360 get_file(new->vm_file);
2361
2362 if (new->vm_ops && new->vm_ops->open)
2363 new->vm_ops->open(new);
2364
2365 if (new_below)
2366 err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2367 ((addr - new->vm_start) >> PAGE_SHIFT), new);
2368 else
2369 err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2370
2371 /* Success. */
2372 if (!err)
2373 return 0;
2374
2375 /* Clean everything up if vma_adjust failed. */
2376 if (new->vm_ops && new->vm_ops->close)
2377 new->vm_ops->close(new);
2378 if (new->vm_file)
2379 fput(new->vm_file);
2380 unlink_anon_vmas(new);
2381 out_free_mpol:
2382 mpol_put(vma_policy(new));
2383 out_free_vma:
2384 kmem_cache_free(vm_area_cachep, new);
2385 return err;
2386 }
2387
2388 /*
2389 * Split a vma into two pieces at address 'addr', a new vma is allocated
2390 * either for the first part or the tail.
2391 */
2392 int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2393 unsigned long addr, int new_below)
2394 {
2395 if (mm->map_count >= sysctl_max_map_count)
2396 return -ENOMEM;
2397
2398 return __split_vma(mm, vma, addr, new_below);
2399 }
2400
2401 /* Munmap is split into 2 main parts -- this part which finds
2402 * what needs doing, and the areas themselves, which do the
2403 * work. This now handles partial unmappings.
2404 * Jeremy Fitzhardinge <jeremy@goop.org>
2405 */
2406 int do_munmap(struct mm_struct *mm, unsigned long start, size_t len)
2407 {
2408 unsigned long end;
2409 struct vm_area_struct *vma, *prev, *last;
2410
2411 if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2412 return -EINVAL;
2413
2414 len = PAGE_ALIGN(len);
2415 if (len == 0)
2416 return -EINVAL;
2417
2418 /* Find the first overlapping VMA */
2419 vma = find_vma(mm, start);
2420 if (!vma)
2421 return 0;
2422 prev = vma->vm_prev;
2423 /* we have start < vma->vm_end */
2424
2425 /* if it doesn't overlap, we have nothing.. */
2426 end = start + len;
2427 if (vma->vm_start >= end)
2428 return 0;
2429
2430 /*
2431 * If we need to split any vma, do it now to save pain later.
2432 *
2433 * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2434 * unmapped vm_area_struct will remain in use: so lower split_vma
2435 * places tmp vma above, and higher split_vma places tmp vma below.
2436 */
2437 if (start > vma->vm_start) {
2438 int error;
2439
2440 /*
2441 * Make sure that map_count on return from munmap() will
2442 * not exceed its limit; but let map_count go just above
2443 * its limit temporarily, to help free resources as expected.
2444 */
2445 if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2446 return -ENOMEM;
2447
2448 error = __split_vma(mm, vma, start, 0);
2449 if (error)
2450 return error;
2451 prev = vma;
2452 }
2453
2454 /* Does it split the last one? */
2455 last = find_vma(mm, end);
2456 if (last && end > last->vm_start) {
2457 int error = __split_vma(mm, last, end, 1);
2458 if (error)
2459 return error;
2460 }
2461 vma = prev ? prev->vm_next : mm->mmap;
2462
2463 /*
2464 * unlock any mlock()ed ranges before detaching vmas
2465 */
2466 if (mm->locked_vm) {
2467 struct vm_area_struct *tmp = vma;
2468 while (tmp && tmp->vm_start < end) {
2469 if (tmp->vm_flags & VM_LOCKED) {
2470 mm->locked_vm -= vma_pages(tmp);
2471 munlock_vma_pages_all(tmp);
2472 }
2473 tmp = tmp->vm_next;
2474 }
2475 }
2476
2477 /*
2478 * Remove the vma's, and unmap the actual pages
2479 */
2480 detach_vmas_to_be_unmapped(mm, vma, prev, end);
2481 unmap_region(mm, vma, prev, start, end);
2482
2483 arch_unmap(mm, vma, start, end);
2484
2485 /* Fix up all other VM information */
2486 remove_vma_list(mm, vma);
2487
2488 return 0;
2489 }
2490
2491 int vm_munmap(unsigned long start, size_t len)
2492 {
2493 int ret;
2494 struct mm_struct *mm = current->mm;
2495
2496 down_write(&mm->mmap_sem);
2497 ret = do_munmap(mm, start, len);
2498 up_write(&mm->mmap_sem);
2499 return ret;
2500 }
2501 EXPORT_SYMBOL(vm_munmap);
2502
2503 SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2504 {
2505 profile_munmap(addr);
2506 return vm_munmap(addr, len);
2507 }
2508
2509
2510 /*
2511 * Emulation of deprecated remap_file_pages() syscall.
2512 */
2513 SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2514 unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2515 {
2516
2517 struct mm_struct *mm = current->mm;
2518 struct vm_area_struct *vma;
2519 unsigned long populate = 0;
2520 unsigned long ret = -EINVAL;
2521 struct file *file;
2522
2523 pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
2524 current->comm, current->pid);
2525
2526 if (prot)
2527 return ret;
2528 start = start & PAGE_MASK;
2529 size = size & PAGE_MASK;
2530
2531 if (start + size <= start)
2532 return ret;
2533
2534 /* Does pgoff wrap? */
2535 if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2536 return ret;
2537
2538 down_write(&mm->mmap_sem);
2539 vma = find_vma(mm, start);
2540
2541 if (!vma || !(vma->vm_flags & VM_SHARED))
2542 goto out;
2543
2544 if (start < vma->vm_start)
2545 goto out;
2546
2547 if (start + size > vma->vm_end) {
2548 struct vm_area_struct *next;
2549
2550 for (next = vma->vm_next; next; next = next->vm_next) {
2551 /* hole between vmas ? */
2552 if (next->vm_start != next->vm_prev->vm_end)
2553 goto out;
2554
2555 if (next->vm_file != vma->vm_file)
2556 goto out;
2557
2558 if (next->vm_flags != vma->vm_flags)
2559 goto out;
2560
2561 if (start + size <= next->vm_end)
2562 break;
2563 }
2564
2565 if (!next)
2566 goto out;
2567 }
2568
2569 prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2570 prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2571 prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2572
2573 flags &= MAP_NONBLOCK;
2574 flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2575 if (vma->vm_flags & VM_LOCKED) {
2576 struct vm_area_struct *tmp;
2577 flags |= MAP_LOCKED;
2578
2579 /* drop PG_Mlocked flag for over-mapped range */
2580 for (tmp = vma; tmp->vm_start >= start + size;
2581 tmp = tmp->vm_next) {
2582 munlock_vma_pages_range(tmp,
2583 max(tmp->vm_start, start),
2584 min(tmp->vm_end, start + size));
2585 }
2586 }
2587
2588 file = get_file(vma->vm_file);
2589 ret = do_mmap_pgoff(vma->vm_file, start, size,
2590 prot, flags, pgoff, &populate);
2591 fput(file);
2592 out:
2593 up_write(&mm->mmap_sem);
2594 if (populate)
2595 mm_populate(ret, populate);
2596 if (!IS_ERR_VALUE(ret))
2597 ret = 0;
2598 return ret;
2599 }
2600
2601 static inline void verify_mm_writelocked(struct mm_struct *mm)
2602 {
2603 #ifdef CONFIG_DEBUG_VM
2604 if (unlikely(down_read_trylock(&mm->mmap_sem))) {
2605 WARN_ON(1);
2606 up_read(&mm->mmap_sem);
2607 }
2608 #endif
2609 }
2610
2611 /*
2612 * this is really a simplified "do_mmap". it only handles
2613 * anonymous maps. eventually we may be able to do some
2614 * brk-specific accounting here.
2615 */
2616 static unsigned long do_brk(unsigned long addr, unsigned long len)
2617 {
2618 struct mm_struct *mm = current->mm;
2619 struct vm_area_struct *vma, *prev;
2620 unsigned long flags;
2621 struct rb_node **rb_link, *rb_parent;
2622 pgoff_t pgoff = addr >> PAGE_SHIFT;
2623 int error;
2624
2625 len = PAGE_ALIGN(len);
2626 if (!len)
2627 return addr;
2628
2629 flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
2630
2631 error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
2632 if (offset_in_page(error))
2633 return error;
2634
2635 error = mlock_future_check(mm, mm->def_flags, len);
2636 if (error)
2637 return error;
2638
2639 /*
2640 * mm->mmap_sem is required to protect against another thread
2641 * changing the mappings in case we sleep.
2642 */
2643 verify_mm_writelocked(mm);
2644
2645 /*
2646 * Clear old maps. this also does some error checking for us
2647 */
2648 while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
2649 &rb_parent)) {
2650 if (do_munmap(mm, addr, len))
2651 return -ENOMEM;
2652 }
2653
2654 /* Check against address space limits *after* clearing old maps... */
2655 if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
2656 return -ENOMEM;
2657
2658 if (mm->map_count > sysctl_max_map_count)
2659 return -ENOMEM;
2660
2661 if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
2662 return -ENOMEM;
2663
2664 /* Can we just expand an old private anonymous mapping? */
2665 vma = vma_merge(mm, prev, addr, addr + len, flags,
2666 NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
2667 if (vma)
2668 goto out;
2669
2670 /*
2671 * create a vma struct for an anonymous mapping
2672 */
2673 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2674 if (!vma) {
2675 vm_unacct_memory(len >> PAGE_SHIFT);
2676 return -ENOMEM;
2677 }
2678
2679 INIT_LIST_HEAD(&vma->anon_vma_chain);
2680 vma->vm_mm = mm;
2681 vma->vm_start = addr;
2682 vma->vm_end = addr + len;
2683 vma->vm_pgoff = pgoff;
2684 vma->vm_flags = flags;
2685 vma->vm_page_prot = vm_get_page_prot(flags);
2686 vma_link(mm, vma, prev, rb_link, rb_parent);
2687 out:
2688 perf_event_mmap(vma);
2689 mm->total_vm += len >> PAGE_SHIFT;
2690 mm->data_vm += len >> PAGE_SHIFT;
2691 if (flags & VM_LOCKED)
2692 mm->locked_vm += (len >> PAGE_SHIFT);
2693 vma->vm_flags |= VM_SOFTDIRTY;
2694 return addr;
2695 }
2696
2697 unsigned long vm_brk(unsigned long addr, unsigned long len)
2698 {
2699 struct mm_struct *mm = current->mm;
2700 unsigned long ret;
2701 bool populate;
2702
2703 down_write(&mm->mmap_sem);
2704 ret = do_brk(addr, len);
2705 populate = ((mm->def_flags & VM_LOCKED) != 0);
2706 up_write(&mm->mmap_sem);
2707 if (populate)
2708 mm_populate(addr, len);
2709 return ret;
2710 }
2711 EXPORT_SYMBOL(vm_brk);
2712
2713 /* Release all mmaps. */
2714 void exit_mmap(struct mm_struct *mm)
2715 {
2716 struct mmu_gather tlb;
2717 struct vm_area_struct *vma;
2718 unsigned long nr_accounted = 0;
2719
2720 /* mm's last user has gone, and its about to be pulled down */
2721 mmu_notifier_release(mm);
2722
2723 if (mm->locked_vm) {
2724 vma = mm->mmap;
2725 while (vma) {
2726 if (vma->vm_flags & VM_LOCKED)
2727 munlock_vma_pages_all(vma);
2728 vma = vma->vm_next;
2729 }
2730 }
2731
2732 arch_exit_mmap(mm);
2733
2734 vma = mm->mmap;
2735 if (!vma) /* Can happen if dup_mmap() received an OOM */
2736 return;
2737
2738 lru_add_drain();
2739 flush_cache_mm(mm);
2740 tlb_gather_mmu(&tlb, mm, 0, -1);
2741 /* update_hiwater_rss(mm) here? but nobody should be looking */
2742 /* Use -1 here to ensure all VMAs in the mm are unmapped */
2743 unmap_vmas(&tlb, vma, 0, -1);
2744
2745 free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
2746 tlb_finish_mmu(&tlb, 0, -1);
2747
2748 /*
2749 * Walk the list again, actually closing and freeing it,
2750 * with preemption enabled, without holding any MM locks.
2751 */
2752 while (vma) {
2753 if (vma->vm_flags & VM_ACCOUNT)
2754 nr_accounted += vma_pages(vma);
2755 vma = remove_vma(vma);
2756 }
2757 vm_unacct_memory(nr_accounted);
2758 }
2759
2760 /* Insert vm structure into process list sorted by address
2761 * and into the inode's i_mmap tree. If vm_file is non-NULL
2762 * then i_mmap_rwsem is taken here.
2763 */
2764 int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
2765 {
2766 struct vm_area_struct *prev;
2767 struct rb_node **rb_link, *rb_parent;
2768
2769 if (find_vma_links(mm, vma->vm_start, vma->vm_end,
2770 &prev, &rb_link, &rb_parent))
2771 return -ENOMEM;
2772 if ((vma->vm_flags & VM_ACCOUNT) &&
2773 security_vm_enough_memory_mm(mm, vma_pages(vma)))
2774 return -ENOMEM;
2775
2776 /*
2777 * The vm_pgoff of a purely anonymous vma should be irrelevant
2778 * until its first write fault, when page's anon_vma and index
2779 * are set. But now set the vm_pgoff it will almost certainly
2780 * end up with (unless mremap moves it elsewhere before that
2781 * first wfault), so /proc/pid/maps tells a consistent story.
2782 *
2783 * By setting it to reflect the virtual start address of the
2784 * vma, merges and splits can happen in a seamless way, just
2785 * using the existing file pgoff checks and manipulations.
2786 * Similarly in do_mmap_pgoff and in do_brk.
2787 */
2788 if (vma_is_anonymous(vma)) {
2789 BUG_ON(vma->anon_vma);
2790 vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
2791 }
2792
2793 vma_link(mm, vma, prev, rb_link, rb_parent);
2794 return 0;
2795 }
2796
2797 /*
2798 * Copy the vma structure to a new location in the same mm,
2799 * prior to moving page table entries, to effect an mremap move.
2800 */
2801 struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
2802 unsigned long addr, unsigned long len, pgoff_t pgoff,
2803 bool *need_rmap_locks)
2804 {
2805 struct vm_area_struct *vma = *vmap;
2806 unsigned long vma_start = vma->vm_start;
2807 struct mm_struct *mm = vma->vm_mm;
2808 struct vm_area_struct *new_vma, *prev;
2809 struct rb_node **rb_link, *rb_parent;
2810 bool faulted_in_anon_vma = true;
2811
2812 /*
2813 * If anonymous vma has not yet been faulted, update new pgoff
2814 * to match new location, to increase its chance of merging.
2815 */
2816 if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
2817 pgoff = addr >> PAGE_SHIFT;
2818 faulted_in_anon_vma = false;
2819 }
2820
2821 if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
2822 return NULL; /* should never get here */
2823 new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
2824 vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
2825 vma->vm_userfaultfd_ctx);
2826 if (new_vma) {
2827 /*
2828 * Source vma may have been merged into new_vma
2829 */
2830 if (unlikely(vma_start >= new_vma->vm_start &&
2831 vma_start < new_vma->vm_end)) {
2832 /*
2833 * The only way we can get a vma_merge with
2834 * self during an mremap is if the vma hasn't
2835 * been faulted in yet and we were allowed to
2836 * reset the dst vma->vm_pgoff to the
2837 * destination address of the mremap to allow
2838 * the merge to happen. mremap must change the
2839 * vm_pgoff linearity between src and dst vmas
2840 * (in turn preventing a vma_merge) to be
2841 * safe. It is only safe to keep the vm_pgoff
2842 * linear if there are no pages mapped yet.
2843 */
2844 VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
2845 *vmap = vma = new_vma;
2846 }
2847 *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
2848 } else {
2849 new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
2850 if (!new_vma)
2851 goto out;
2852 *new_vma = *vma;
2853 new_vma->vm_start = addr;
2854 new_vma->vm_end = addr + len;
2855 new_vma->vm_pgoff = pgoff;
2856 if (vma_dup_policy(vma, new_vma))
2857 goto out_free_vma;
2858 INIT_LIST_HEAD(&new_vma->anon_vma_chain);
2859 if (anon_vma_clone(new_vma, vma))
2860 goto out_free_mempol;
2861 if (new_vma->vm_file)
2862 get_file(new_vma->vm_file);
2863 if (new_vma->vm_ops && new_vma->vm_ops->open)
2864 new_vma->vm_ops->open(new_vma);
2865 vma_link(mm, new_vma, prev, rb_link, rb_parent);
2866 *need_rmap_locks = false;
2867 }
2868 return new_vma;
2869
2870 out_free_mempol:
2871 mpol_put(vma_policy(new_vma));
2872 out_free_vma:
2873 kmem_cache_free(vm_area_cachep, new_vma);
2874 out:
2875 return NULL;
2876 }
2877
2878 /*
2879 * Return true if the calling process may expand its vm space by the passed
2880 * number of pages
2881 */
2882 bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
2883 {
2884 if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
2885 return false;
2886
2887 if (is_data_mapping(flags) &&
2888 mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
2889 if (ignore_rlimit_data)
2890 pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Will be forbidden soon.\n",
2891 current->comm, current->pid,
2892 (mm->data_vm + npages) << PAGE_SHIFT,
2893 rlimit(RLIMIT_DATA));
2894 else
2895 return false;
2896 }
2897
2898 return true;
2899 }
2900
2901 void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
2902 {
2903 mm->total_vm += npages;
2904
2905 if (is_exec_mapping(flags))
2906 mm->exec_vm += npages;
2907 else if (is_stack_mapping(flags))
2908 mm->stack_vm += npages;
2909 else if (is_data_mapping(flags))
2910 mm->data_vm += npages;
2911 }
2912
2913 static int special_mapping_fault(struct vm_area_struct *vma,
2914 struct vm_fault *vmf);
2915
2916 /*
2917 * Having a close hook prevents vma merging regardless of flags.
2918 */
2919 static void special_mapping_close(struct vm_area_struct *vma)
2920 {
2921 }
2922
2923 static const char *special_mapping_name(struct vm_area_struct *vma)
2924 {
2925 return ((struct vm_special_mapping *)vma->vm_private_data)->name;
2926 }
2927
2928 static const struct vm_operations_struct special_mapping_vmops = {
2929 .close = special_mapping_close,
2930 .fault = special_mapping_fault,
2931 .name = special_mapping_name,
2932 };
2933
2934 static const struct vm_operations_struct legacy_special_mapping_vmops = {
2935 .close = special_mapping_close,
2936 .fault = special_mapping_fault,
2937 };
2938
2939 static int special_mapping_fault(struct vm_area_struct *vma,
2940 struct vm_fault *vmf)
2941 {
2942 pgoff_t pgoff;
2943 struct page **pages;
2944
2945 if (vma->vm_ops == &legacy_special_mapping_vmops) {
2946 pages = vma->vm_private_data;
2947 } else {
2948 struct vm_special_mapping *sm = vma->vm_private_data;
2949
2950 if (sm->fault)
2951 return sm->fault(sm, vma, vmf);
2952
2953 pages = sm->pages;
2954 }
2955
2956 for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
2957 pgoff--;
2958
2959 if (*pages) {
2960 struct page *page = *pages;
2961 get_page(page);
2962 vmf->page = page;
2963 return 0;
2964 }
2965
2966 return VM_FAULT_SIGBUS;
2967 }
2968
2969 static struct vm_area_struct *__install_special_mapping(
2970 struct mm_struct *mm,
2971 unsigned long addr, unsigned long len,
2972 unsigned long vm_flags, void *priv,
2973 const struct vm_operations_struct *ops)
2974 {
2975 int ret;
2976 struct vm_area_struct *vma;
2977
2978 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
2979 if (unlikely(vma == NULL))
2980 return ERR_PTR(-ENOMEM);
2981
2982 INIT_LIST_HEAD(&vma->anon_vma_chain);
2983 vma->vm_mm = mm;
2984 vma->vm_start = addr;
2985 vma->vm_end = addr + len;
2986
2987 vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
2988 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
2989
2990 vma->vm_ops = ops;
2991 vma->vm_private_data = priv;
2992
2993 ret = insert_vm_struct(mm, vma);
2994 if (ret)
2995 goto out;
2996
2997 vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
2998
2999 perf_event_mmap(vma);
3000
3001 return vma;
3002
3003 out:
3004 kmem_cache_free(vm_area_cachep, vma);
3005 return ERR_PTR(ret);
3006 }
3007
3008 /*
3009 * Called with mm->mmap_sem held for writing.
3010 * Insert a new vma covering the given region, with the given flags.
3011 * Its pages are supplied by the given array of struct page *.
3012 * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3013 * The region past the last page supplied will always produce SIGBUS.
3014 * The array pointer and the pages it points to are assumed to stay alive
3015 * for as long as this mapping might exist.
3016 */
3017 struct vm_area_struct *_install_special_mapping(
3018 struct mm_struct *mm,
3019 unsigned long addr, unsigned long len,
3020 unsigned long vm_flags, const struct vm_special_mapping *spec)
3021 {
3022 return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3023 &special_mapping_vmops);
3024 }
3025
3026 int install_special_mapping(struct mm_struct *mm,
3027 unsigned long addr, unsigned long len,
3028 unsigned long vm_flags, struct page **pages)
3029 {
3030 struct vm_area_struct *vma = __install_special_mapping(
3031 mm, addr, len, vm_flags, (void *)pages,
3032 &legacy_special_mapping_vmops);
3033
3034 return PTR_ERR_OR_ZERO(vma);
3035 }
3036
3037 static DEFINE_MUTEX(mm_all_locks_mutex);
3038
3039 static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3040 {
3041 if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3042 /*
3043 * The LSB of head.next can't change from under us
3044 * because we hold the mm_all_locks_mutex.
3045 */
3046 down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3047 /*
3048 * We can safely modify head.next after taking the
3049 * anon_vma->root->rwsem. If some other vma in this mm shares
3050 * the same anon_vma we won't take it again.
3051 *
3052 * No need of atomic instructions here, head.next
3053 * can't change from under us thanks to the
3054 * anon_vma->root->rwsem.
3055 */
3056 if (__test_and_set_bit(0, (unsigned long *)
3057 &anon_vma->root->rb_root.rb_node))
3058 BUG();
3059 }
3060 }
3061
3062 static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3063 {
3064 if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3065 /*
3066 * AS_MM_ALL_LOCKS can't change from under us because
3067 * we hold the mm_all_locks_mutex.
3068 *
3069 * Operations on ->flags have to be atomic because
3070 * even if AS_MM_ALL_LOCKS is stable thanks to the
3071 * mm_all_locks_mutex, there may be other cpus
3072 * changing other bitflags in parallel to us.
3073 */
3074 if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3075 BUG();
3076 down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
3077 }
3078 }
3079
3080 /*
3081 * This operation locks against the VM for all pte/vma/mm related
3082 * operations that could ever happen on a certain mm. This includes
3083 * vmtruncate, try_to_unmap, and all page faults.
3084 *
3085 * The caller must take the mmap_sem in write mode before calling
3086 * mm_take_all_locks(). The caller isn't allowed to release the
3087 * mmap_sem until mm_drop_all_locks() returns.
3088 *
3089 * mmap_sem in write mode is required in order to block all operations
3090 * that could modify pagetables and free pages without need of
3091 * altering the vma layout. It's also needed in write mode to avoid new
3092 * anon_vmas to be associated with existing vmas.
3093 *
3094 * A single task can't take more than one mm_take_all_locks() in a row
3095 * or it would deadlock.
3096 *
3097 * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3098 * mapping->flags avoid to take the same lock twice, if more than one
3099 * vma in this mm is backed by the same anon_vma or address_space.
3100 *
3101 * We take locks in following order, accordingly to comment at beginning
3102 * of mm/rmap.c:
3103 * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3104 * hugetlb mapping);
3105 * - all i_mmap_rwsem locks;
3106 * - all anon_vma->rwseml
3107 *
3108 * We can take all locks within these types randomly because the VM code
3109 * doesn't nest them and we protected from parallel mm_take_all_locks() by
3110 * mm_all_locks_mutex.
3111 *
3112 * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3113 * that may have to take thousand of locks.
3114 *
3115 * mm_take_all_locks() can fail if it's interrupted by signals.
3116 */
3117 int mm_take_all_locks(struct mm_struct *mm)
3118 {
3119 struct vm_area_struct *vma;
3120 struct anon_vma_chain *avc;
3121
3122 BUG_ON(down_read_trylock(&mm->mmap_sem));
3123
3124 mutex_lock(&mm_all_locks_mutex);
3125
3126 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3127 if (signal_pending(current))
3128 goto out_unlock;
3129 if (vma->vm_file && vma->vm_file->f_mapping &&
3130 is_vm_hugetlb_page(vma))
3131 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3132 }
3133
3134 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3135 if (signal_pending(current))
3136 goto out_unlock;
3137 if (vma->vm_file && vma->vm_file->f_mapping &&
3138 !is_vm_hugetlb_page(vma))
3139 vm_lock_mapping(mm, vma->vm_file->f_mapping);
3140 }
3141
3142 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3143 if (signal_pending(current))
3144 goto out_unlock;
3145 if (vma->anon_vma)
3146 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3147 vm_lock_anon_vma(mm, avc->anon_vma);
3148 }
3149
3150 return 0;
3151
3152 out_unlock:
3153 mm_drop_all_locks(mm);
3154 return -EINTR;
3155 }
3156
3157 static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3158 {
3159 if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) {
3160 /*
3161 * The LSB of head.next can't change to 0 from under
3162 * us because we hold the mm_all_locks_mutex.
3163 *
3164 * We must however clear the bitflag before unlocking
3165 * the vma so the users using the anon_vma->rb_root will
3166 * never see our bitflag.
3167 *
3168 * No need of atomic instructions here, head.next
3169 * can't change from under us until we release the
3170 * anon_vma->root->rwsem.
3171 */
3172 if (!__test_and_clear_bit(0, (unsigned long *)
3173 &anon_vma->root->rb_root.rb_node))
3174 BUG();
3175 anon_vma_unlock_write(anon_vma);
3176 }
3177 }
3178
3179 static void vm_unlock_mapping(struct address_space *mapping)
3180 {
3181 if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3182 /*
3183 * AS_MM_ALL_LOCKS can't change to 0 from under us
3184 * because we hold the mm_all_locks_mutex.
3185 */
3186 i_mmap_unlock_write(mapping);
3187 if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3188 &mapping->flags))
3189 BUG();
3190 }
3191 }
3192
3193 /*
3194 * The mmap_sem cannot be released by the caller until
3195 * mm_drop_all_locks() returns.
3196 */
3197 void mm_drop_all_locks(struct mm_struct *mm)
3198 {
3199 struct vm_area_struct *vma;
3200 struct anon_vma_chain *avc;
3201
3202 BUG_ON(down_read_trylock(&mm->mmap_sem));
3203 BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3204
3205 for (vma = mm->mmap; vma; vma = vma->vm_next) {
3206 if (vma->anon_vma)
3207 list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3208 vm_unlock_anon_vma(avc->anon_vma);
3209 if (vma->vm_file && vma->vm_file->f_mapping)
3210 vm_unlock_mapping(vma->vm_file->f_mapping);
3211 }
3212
3213 mutex_unlock(&mm_all_locks_mutex);
3214 }
3215
3216 /*
3217 * initialise the VMA slab
3218 */
3219 void __init mmap_init(void)
3220 {
3221 int ret;
3222
3223 ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3224 VM_BUG_ON(ret);
3225 }
3226
3227 /*
3228 * Initialise sysctl_user_reserve_kbytes.
3229 *
3230 * This is intended to prevent a user from starting a single memory hogging
3231 * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3232 * mode.
3233 *
3234 * The default value is min(3% of free memory, 128MB)
3235 * 128MB is enough to recover with sshd/login, bash, and top/kill.
3236 */
3237 static int init_user_reserve(void)
3238 {
3239 unsigned long free_kbytes;
3240
3241 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3242
3243 sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3244 return 0;
3245 }
3246 subsys_initcall(init_user_reserve);
3247
3248 /*
3249 * Initialise sysctl_admin_reserve_kbytes.
3250 *
3251 * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3252 * to log in and kill a memory hogging process.
3253 *
3254 * Systems with more than 256MB will reserve 8MB, enough to recover
3255 * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3256 * only reserve 3% of free pages by default.
3257 */
3258 static int init_admin_reserve(void)
3259 {
3260 unsigned long free_kbytes;
3261
3262 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3263
3264 sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3265 return 0;
3266 }
3267 subsys_initcall(init_admin_reserve);
3268
3269 /*
3270 * Reinititalise user and admin reserves if memory is added or removed.
3271 *
3272 * The default user reserve max is 128MB, and the default max for the
3273 * admin reserve is 8MB. These are usually, but not always, enough to
3274 * enable recovery from a memory hogging process using login/sshd, a shell,
3275 * and tools like top. It may make sense to increase or even disable the
3276 * reserve depending on the existence of swap or variations in the recovery
3277 * tools. So, the admin may have changed them.
3278 *
3279 * If memory is added and the reserves have been eliminated or increased above
3280 * the default max, then we'll trust the admin.
3281 *
3282 * If memory is removed and there isn't enough free memory, then we
3283 * need to reset the reserves.
3284 *
3285 * Otherwise keep the reserve set by the admin.
3286 */
3287 static int reserve_mem_notifier(struct notifier_block *nb,
3288 unsigned long action, void *data)
3289 {
3290 unsigned long tmp, free_kbytes;
3291
3292 switch (action) {
3293 case MEM_ONLINE:
3294 /* Default max is 128MB. Leave alone if modified by operator. */
3295 tmp = sysctl_user_reserve_kbytes;
3296 if (0 < tmp && tmp < (1UL << 17))
3297 init_user_reserve();
3298
3299 /* Default max is 8MB. Leave alone if modified by operator. */
3300 tmp = sysctl_admin_reserve_kbytes;
3301 if (0 < tmp && tmp < (1UL << 13))
3302 init_admin_reserve();
3303
3304 break;
3305 case MEM_OFFLINE:
3306 free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3307
3308 if (sysctl_user_reserve_kbytes > free_kbytes) {
3309 init_user_reserve();
3310 pr_info("vm.user_reserve_kbytes reset to %lu\n",
3311 sysctl_user_reserve_kbytes);
3312 }
3313
3314 if (sysctl_admin_reserve_kbytes > free_kbytes) {
3315 init_admin_reserve();
3316 pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3317 sysctl_admin_reserve_kbytes);
3318 }
3319 break;
3320 default:
3321 break;
3322 }
3323 return NOTIFY_OK;
3324 }
3325
3326 static struct notifier_block reserve_mem_nb = {
3327 .notifier_call = reserve_mem_notifier,
3328 };
3329
3330 static int __meminit init_reserve_notifier(void)
3331 {
3332 if (register_hotmemory_notifier(&reserve_mem_nb))
3333 pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3334
3335 return 0;
3336 }
3337 subsys_initcall(init_reserve_notifier);