Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0-only
2 : : /*
3 : : * mm/mmap.c
4 : : *
5 : : * Written by obz.
6 : : *
7 : : * Address space accounting code <alan@lxorguk.ukuu.org.uk>
8 : : */
9 : :
10 : : #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11 : :
12 : : #include <linux/kernel.h>
13 : : #include <linux/slab.h>
14 : : #include <linux/backing-dev.h>
15 : : #include <linux/mm.h>
16 : : #include <linux/vmacache.h>
17 : : #include <linux/shm.h>
18 : : #include <linux/mman.h>
19 : : #include <linux/pagemap.h>
20 : : #include <linux/swap.h>
21 : : #include <linux/syscalls.h>
22 : : #include <linux/capability.h>
23 : : #include <linux/init.h>
24 : : #include <linux/file.h>
25 : : #include <linux/fs.h>
26 : : #include <linux/personality.h>
27 : : #include <linux/security.h>
28 : : #include <linux/hugetlb.h>
29 : : #include <linux/shmem_fs.h>
30 : : #include <linux/profile.h>
31 : : #include <linux/export.h>
32 : : #include <linux/mount.h>
33 : : #include <linux/mempolicy.h>
34 : : #include <linux/rmap.h>
35 : : #include <linux/mmu_notifier.h>
36 : : #include <linux/mmdebug.h>
37 : : #include <linux/perf_event.h>
38 : : #include <linux/audit.h>
39 : : #include <linux/khugepaged.h>
40 : : #include <linux/uprobes.h>
41 : : #include <linux/rbtree_augmented.h>
42 : : #include <linux/notifier.h>
43 : : #include <linux/memory.h>
44 : : #include <linux/printk.h>
45 : : #include <linux/userfaultfd_k.h>
46 : : #include <linux/moduleparam.h>
47 : : #include <linux/pkeys.h>
48 : : #include <linux/oom.h>
49 : : #include <linux/sched/mm.h>
50 : :
51 : : #include <linux/uaccess.h>
52 : : #include <asm/cacheflush.h>
53 : : #include <asm/tlb.h>
54 : : #include <asm/mmu_context.h>
55 : :
56 : : #include "internal.h"
57 : :
58 : : #ifndef arch_mmap_check
59 : : #define arch_mmap_check(addr, len, flags) (0)
60 : : #endif
61 : :
62 : : #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS
63 : : const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN;
64 : : const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX;
65 : : int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS;
66 : : #endif
67 : : #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS
68 : : const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN;
69 : : const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX;
70 : : int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS;
71 : : #endif
72 : :
73 : : static bool ignore_rlimit_data;
74 : : core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644);
75 : :
76 : : static void unmap_region(struct mm_struct *mm,
77 : : struct vm_area_struct *vma, struct vm_area_struct *prev,
78 : : unsigned long start, unsigned long end);
79 : :
80 : : /* description of effects of mapping type and prot in current implementation.
81 : : * this is due to the limited x86 page protection hardware. The expected
82 : : * behavior is in parens:
83 : : *
84 : : * map_type prot
85 : : * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC
86 : : * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes
87 : : * w: (no) no w: (no) no w: (yes) yes w: (no) no
88 : : * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
89 : : *
90 : : * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes
91 : : * w: (no) no w: (no) no w: (copy) copy w: (no) no
92 : : * x: (no) no x: (no) yes x: (no) yes x: (yes) yes
93 : : */
94 : : pgprot_t protection_map[16] __ro_after_init = {
95 : : __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111,
96 : : __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111
97 : : };
98 : :
99 : : #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT
100 : : static inline pgprot_t arch_filter_pgprot(pgprot_t prot)
101 : : {
102 : : return prot;
103 : : }
104 : : #endif
105 : :
106 : 3 : pgprot_t vm_get_page_prot(unsigned long vm_flags)
107 : : {
108 : 3 : pgprot_t ret = __pgprot(pgprot_val(protection_map[vm_flags &
109 : : (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) |
110 : : pgprot_val(arch_vm_get_page_prot(vm_flags)));
111 : :
112 : 3 : return arch_filter_pgprot(ret);
113 : : }
114 : : EXPORT_SYMBOL(vm_get_page_prot);
115 : :
116 : 3 : static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags)
117 : : {
118 : 3 : return pgprot_modify(oldprot, vm_get_page_prot(vm_flags));
119 : : }
120 : :
121 : : /* Update vma->vm_page_prot to reflect vma->vm_flags. */
122 : 3 : void vma_set_page_prot(struct vm_area_struct *vma)
123 : : {
124 : 3 : unsigned long vm_flags = vma->vm_flags;
125 : : pgprot_t vm_page_prot;
126 : :
127 : 3 : vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags);
128 : 3 : if (vma_wants_writenotify(vma, vm_page_prot)) {
129 : 3 : vm_flags &= ~VM_SHARED;
130 : 3 : vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags);
131 : : }
132 : : /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */
133 : : WRITE_ONCE(vma->vm_page_prot, vm_page_prot);
134 : 3 : }
135 : :
136 : : /*
137 : : * Requires inode->i_mapping->i_mmap_rwsem
138 : : */
139 : 3 : static void __remove_shared_vm_struct(struct vm_area_struct *vma,
140 : : struct file *file, struct address_space *mapping)
141 : : {
142 : 3 : if (vma->vm_flags & VM_DENYWRITE)
143 : 3 : atomic_inc(&file_inode(file)->i_writecount);
144 : 3 : if (vma->vm_flags & VM_SHARED)
145 : : mapping_unmap_writable(mapping);
146 : :
147 : : flush_dcache_mmap_lock(mapping);
148 : 3 : vma_interval_tree_remove(vma, &mapping->i_mmap);
149 : : flush_dcache_mmap_unlock(mapping);
150 : 3 : }
151 : :
152 : : /*
153 : : * Unlink a file-based vm structure from its interval tree, to hide
154 : : * vma from rmap and vmtruncate before freeing its page tables.
155 : : */
156 : 3 : void unlink_file_vma(struct vm_area_struct *vma)
157 : : {
158 : 3 : struct file *file = vma->vm_file;
159 : :
160 : 3 : if (file) {
161 : 3 : struct address_space *mapping = file->f_mapping;
162 : : i_mmap_lock_write(mapping);
163 : 3 : __remove_shared_vm_struct(vma, file, mapping);
164 : : i_mmap_unlock_write(mapping);
165 : : }
166 : 3 : }
167 : :
168 : : /*
169 : : * Close a vm structure and free it, returning the next.
170 : : */
171 : 3 : static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
172 : : {
173 : 3 : struct vm_area_struct *next = vma->vm_next;
174 : :
175 : 3 : might_sleep();
176 : 3 : if (vma->vm_ops && vma->vm_ops->close)
177 : 3 : vma->vm_ops->close(vma);
178 : 3 : if (vma->vm_file)
179 : 3 : fput(vma->vm_file);
180 : : mpol_put(vma_policy(vma));
181 : 3 : vm_area_free(vma);
182 : 3 : return next;
183 : : }
184 : :
185 : : static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags,
186 : : struct list_head *uf);
187 : 3 : SYSCALL_DEFINE1(brk, unsigned long, brk)
188 : : {
189 : : unsigned long retval;
190 : : unsigned long newbrk, oldbrk, origbrk;
191 : 3 : struct mm_struct *mm = current->mm;
192 : : struct vm_area_struct *next;
193 : : unsigned long min_brk;
194 : : bool populate;
195 : : bool downgraded = false;
196 : 3 : LIST_HEAD(uf);
197 : :
198 : 3 : if (down_write_killable(&mm->mmap_sem))
199 : : return -EINTR;
200 : :
201 : 3 : origbrk = mm->brk;
202 : :
203 : : #ifdef CONFIG_COMPAT_BRK
204 : : /*
205 : : * CONFIG_COMPAT_BRK can still be overridden by setting
206 : : * randomize_va_space to 2, which will still cause mm->start_brk
207 : : * to be arbitrarily shifted
208 : : */
209 : : if (current->brk_randomized)
210 : : min_brk = mm->start_brk;
211 : : else
212 : : min_brk = mm->end_data;
213 : : #else
214 : 3 : min_brk = mm->start_brk;
215 : : #endif
216 : 3 : if (brk < min_brk)
217 : : goto out;
218 : :
219 : : /*
220 : : * Check against rlimit here. If this check is done later after the test
221 : : * of oldbrk with newbrk then it can escape the test and let the data
222 : : * segment grow beyond its set limit the in case where the limit is
223 : : * not page aligned -Ram Gupta
224 : : */
225 : 3 : if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
226 : : mm->end_data, mm->start_data))
227 : : goto out;
228 : :
229 : 3 : newbrk = PAGE_ALIGN(brk);
230 : 3 : oldbrk = PAGE_ALIGN(mm->brk);
231 : 3 : if (oldbrk == newbrk) {
232 : 3 : mm->brk = brk;
233 : 3 : goto success;
234 : : }
235 : :
236 : : /*
237 : : * Always allow shrinking brk.
238 : : * __do_munmap() may downgrade mmap_sem to read.
239 : : */
240 : 3 : if (brk <= mm->brk) {
241 : : int ret;
242 : :
243 : : /*
244 : : * mm->brk must to be protected by write mmap_sem so update it
245 : : * before downgrading mmap_sem. When __do_munmap() fails,
246 : : * mm->brk will be restored from origbrk.
247 : : */
248 : 3 : mm->brk = brk;
249 : 3 : ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true);
250 : 3 : if (ret < 0) {
251 : 0 : mm->brk = origbrk;
252 : 0 : goto out;
253 : 3 : } else if (ret == 1) {
254 : : downgraded = true;
255 : : }
256 : : goto success;
257 : : }
258 : :
259 : : /* Check against existing mmap mappings. */
260 : 3 : next = find_vma(mm, oldbrk);
261 : 3 : if (next && newbrk + PAGE_SIZE > vm_start_gap(next))
262 : : goto out;
263 : :
264 : : /* Ok, looks good - let it rip. */
265 : 3 : if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0)
266 : : goto out;
267 : 3 : mm->brk = brk;
268 : :
269 : : success:
270 : 3 : populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0;
271 : 3 : if (downgraded)
272 : 3 : up_read(&mm->mmap_sem);
273 : : else
274 : 3 : up_write(&mm->mmap_sem);
275 : : userfaultfd_unmap_complete(mm, &uf);
276 : 3 : if (populate)
277 : 0 : mm_populate(oldbrk, newbrk - oldbrk);
278 : 3 : return brk;
279 : :
280 : : out:
281 : : retval = origbrk;
282 : 3 : up_write(&mm->mmap_sem);
283 : 3 : return retval;
284 : : }
285 : :
286 : 3 : static inline unsigned long vma_compute_gap(struct vm_area_struct *vma)
287 : : {
288 : : unsigned long gap, prev_end;
289 : :
290 : : /*
291 : : * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we
292 : : * allow two stack_guard_gaps between them here, and when choosing
293 : : * an unmapped area; whereas when expanding we only require one.
294 : : * That's a little inconsistent, but keeps the code here simpler.
295 : : */
296 : : gap = vm_start_gap(vma);
297 : 3 : if (vma->vm_prev) {
298 : : prev_end = vm_end_gap(vma->vm_prev);
299 : 3 : if (gap > prev_end)
300 : 3 : gap -= prev_end;
301 : : else
302 : : gap = 0;
303 : : }
304 : 3 : return gap;
305 : : }
306 : :
307 : : #ifdef CONFIG_DEBUG_VM_RB
308 : : static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma)
309 : : {
310 : : unsigned long max = vma_compute_gap(vma), subtree_gap;
311 : : if (vma->vm_rb.rb_left) {
312 : : subtree_gap = rb_entry(vma->vm_rb.rb_left,
313 : : struct vm_area_struct, vm_rb)->rb_subtree_gap;
314 : : if (subtree_gap > max)
315 : : max = subtree_gap;
316 : : }
317 : : if (vma->vm_rb.rb_right) {
318 : : subtree_gap = rb_entry(vma->vm_rb.rb_right,
319 : : struct vm_area_struct, vm_rb)->rb_subtree_gap;
320 : : if (subtree_gap > max)
321 : : max = subtree_gap;
322 : : }
323 : : return max;
324 : : }
325 : :
326 : : static int browse_rb(struct mm_struct *mm)
327 : : {
328 : : struct rb_root *root = &mm->mm_rb;
329 : : int i = 0, j, bug = 0;
330 : : struct rb_node *nd, *pn = NULL;
331 : : unsigned long prev = 0, pend = 0;
332 : :
333 : : for (nd = rb_first(root); nd; nd = rb_next(nd)) {
334 : : struct vm_area_struct *vma;
335 : : vma = rb_entry(nd, struct vm_area_struct, vm_rb);
336 : : if (vma->vm_start < prev) {
337 : : pr_emerg("vm_start %lx < prev %lx\n",
338 : : vma->vm_start, prev);
339 : : bug = 1;
340 : : }
341 : : if (vma->vm_start < pend) {
342 : : pr_emerg("vm_start %lx < pend %lx\n",
343 : : vma->vm_start, pend);
344 : : bug = 1;
345 : : }
346 : : if (vma->vm_start > vma->vm_end) {
347 : : pr_emerg("vm_start %lx > vm_end %lx\n",
348 : : vma->vm_start, vma->vm_end);
349 : : bug = 1;
350 : : }
351 : : spin_lock(&mm->page_table_lock);
352 : : if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) {
353 : : pr_emerg("free gap %lx, correct %lx\n",
354 : : vma->rb_subtree_gap,
355 : : vma_compute_subtree_gap(vma));
356 : : bug = 1;
357 : : }
358 : : spin_unlock(&mm->page_table_lock);
359 : : i++;
360 : : pn = nd;
361 : : prev = vma->vm_start;
362 : : pend = vma->vm_end;
363 : : }
364 : : j = 0;
365 : : for (nd = pn; nd; nd = rb_prev(nd))
366 : : j++;
367 : : if (i != j) {
368 : : pr_emerg("backwards %d, forwards %d\n", j, i);
369 : : bug = 1;
370 : : }
371 : : return bug ? -1 : i;
372 : : }
373 : :
374 : : static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore)
375 : : {
376 : : struct rb_node *nd;
377 : :
378 : : for (nd = rb_first(root); nd; nd = rb_next(nd)) {
379 : : struct vm_area_struct *vma;
380 : : vma = rb_entry(nd, struct vm_area_struct, vm_rb);
381 : : VM_BUG_ON_VMA(vma != ignore &&
382 : : vma->rb_subtree_gap != vma_compute_subtree_gap(vma),
383 : : vma);
384 : : }
385 : : }
386 : :
387 : : static void validate_mm(struct mm_struct *mm)
388 : : {
389 : : int bug = 0;
390 : : int i = 0;
391 : : unsigned long highest_address = 0;
392 : : struct vm_area_struct *vma = mm->mmap;
393 : :
394 : : while (vma) {
395 : : struct anon_vma *anon_vma = vma->anon_vma;
396 : : struct anon_vma_chain *avc;
397 : :
398 : : if (anon_vma) {
399 : : anon_vma_lock_read(anon_vma);
400 : : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
401 : : anon_vma_interval_tree_verify(avc);
402 : : anon_vma_unlock_read(anon_vma);
403 : : }
404 : :
405 : : highest_address = vm_end_gap(vma);
406 : : vma = vma->vm_next;
407 : : i++;
408 : : }
409 : : if (i != mm->map_count) {
410 : : pr_emerg("map_count %d vm_next %d\n", mm->map_count, i);
411 : : bug = 1;
412 : : }
413 : : if (highest_address != mm->highest_vm_end) {
414 : : pr_emerg("mm->highest_vm_end %lx, found %lx\n",
415 : : mm->highest_vm_end, highest_address);
416 : : bug = 1;
417 : : }
418 : : i = browse_rb(mm);
419 : : if (i != mm->map_count) {
420 : : if (i != -1)
421 : : pr_emerg("map_count %d rb %d\n", mm->map_count, i);
422 : : bug = 1;
423 : : }
424 : : VM_BUG_ON_MM(bug, mm);
425 : : }
426 : : #else
427 : : #define validate_mm_rb(root, ignore) do { } while (0)
428 : : #define validate_mm(mm) do { } while (0)
429 : : #endif
430 : :
431 : 3 : RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks,
432 : : struct vm_area_struct, vm_rb,
433 : : unsigned long, rb_subtree_gap, vma_compute_gap)
434 : :
435 : : /*
436 : : * Update augmented rbtree rb_subtree_gap values after vma->vm_start or
437 : : * vma->vm_prev->vm_end values changed, without modifying the vma's position
438 : : * in the rbtree.
439 : : */
440 : : static void vma_gap_update(struct vm_area_struct *vma)
441 : : {
442 : : /*
443 : : * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created
444 : : * a callback function that does exactly what we want.
445 : : */
446 : 3 : vma_gap_callbacks_propagate(&vma->vm_rb, NULL);
447 : : }
448 : :
449 : : static inline void vma_rb_insert(struct vm_area_struct *vma,
450 : : struct rb_root *root)
451 : : {
452 : : /* All rb_subtree_gap values must be consistent prior to insertion */
453 : : validate_mm_rb(root, NULL);
454 : :
455 : : rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
456 : : }
457 : :
458 : 3 : static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
459 : : {
460 : : /*
461 : : * Note rb_erase_augmented is a fairly large inline function,
462 : : * so make sure we instantiate it only once with our desired
463 : : * augmented rbtree callbacks.
464 : : */
465 : 3 : rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
466 : 3 : }
467 : :
468 : : static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
469 : : struct rb_root *root,
470 : : struct vm_area_struct *ignore)
471 : : {
472 : : /*
473 : : * All rb_subtree_gap values must be consistent prior to erase,
474 : : * with the possible exception of the "next" vma being erased if
475 : : * next->vm_start was reduced.
476 : : */
477 : : validate_mm_rb(root, ignore);
478 : :
479 : 3 : __vma_rb_erase(vma, root);
480 : : }
481 : :
482 : : static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
483 : : struct rb_root *root)
484 : : {
485 : : /*
486 : : * All rb_subtree_gap values must be consistent prior to erase,
487 : : * with the possible exception of the vma being erased.
488 : : */
489 : : validate_mm_rb(root, vma);
490 : :
491 : 3 : __vma_rb_erase(vma, root);
492 : : }
493 : :
494 : : /*
495 : : * vma has some anon_vma assigned, and is already inserted on that
496 : : * anon_vma's interval trees.
497 : : *
498 : : * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
499 : : * vma must be removed from the anon_vma's interval trees using
500 : : * anon_vma_interval_tree_pre_update_vma().
501 : : *
502 : : * After the update, the vma will be reinserted using
503 : : * anon_vma_interval_tree_post_update_vma().
504 : : *
505 : : * The entire update must be protected by exclusive mmap_sem and by
506 : : * the root anon_vma's mutex.
507 : : */
508 : : static inline void
509 : 3 : anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma)
510 : : {
511 : : struct anon_vma_chain *avc;
512 : :
513 : 3 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
514 : 3 : anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root);
515 : 3 : }
516 : :
517 : : static inline void
518 : 3 : anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma)
519 : : {
520 : : struct anon_vma_chain *avc;
521 : :
522 : 3 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
523 : 3 : anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root);
524 : 3 : }
525 : :
526 : : static int find_vma_links(struct mm_struct *mm, unsigned long addr,
527 : : unsigned long end, struct vm_area_struct **pprev,
528 : : struct rb_node ***rb_link, struct rb_node **rb_parent)
529 : : {
530 : : struct rb_node **__rb_link, *__rb_parent, *rb_prev;
531 : :
532 : 3 : __rb_link = &mm->mm_rb.rb_node;
533 : : rb_prev = __rb_parent = NULL;
534 : :
535 : 3 : while (*__rb_link) {
536 : : struct vm_area_struct *vma_tmp;
537 : :
538 : : __rb_parent = *__rb_link;
539 : : vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb);
540 : :
541 : 3 : if (vma_tmp->vm_end > addr) {
542 : : /* Fail if an existing vma overlaps the area */
543 : 3 : if (vma_tmp->vm_start < end)
544 : : return -ENOMEM;
545 : 3 : __rb_link = &__rb_parent->rb_left;
546 : : } else {
547 : : rb_prev = __rb_parent;
548 : 3 : __rb_link = &__rb_parent->rb_right;
549 : : }
550 : : }
551 : :
552 : : *pprev = NULL;
553 : 3 : if (rb_prev)
554 : 3 : *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb);
555 : 3 : *rb_link = __rb_link;
556 : 3 : *rb_parent = __rb_parent;
557 : : return 0;
558 : : }
559 : :
560 : 0 : static unsigned long count_vma_pages_range(struct mm_struct *mm,
561 : : unsigned long addr, unsigned long end)
562 : : {
563 : : unsigned long nr_pages = 0;
564 : : struct vm_area_struct *vma;
565 : :
566 : : /* Find first overlaping mapping */
567 : : vma = find_vma_intersection(mm, addr, end);
568 : 0 : if (!vma)
569 : : return 0;
570 : :
571 : 0 : nr_pages = (min(end, vma->vm_end) -
572 : 0 : max(addr, vma->vm_start)) >> PAGE_SHIFT;
573 : :
574 : : /* Iterate over the rest of the overlaps */
575 : 0 : for (vma = vma->vm_next; vma; vma = vma->vm_next) {
576 : : unsigned long overlap_len;
577 : :
578 : 0 : if (vma->vm_start > end)
579 : : break;
580 : :
581 : 0 : overlap_len = min(end, vma->vm_end) - vma->vm_start;
582 : 0 : nr_pages += overlap_len >> PAGE_SHIFT;
583 : : }
584 : :
585 : 0 : return nr_pages;
586 : : }
587 : :
588 : 3 : void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
589 : : struct rb_node **rb_link, struct rb_node *rb_parent)
590 : : {
591 : : /* Update tracking information for the gap following the new vma. */
592 : 3 : if (vma->vm_next)
593 : : vma_gap_update(vma->vm_next);
594 : : else
595 : 3 : mm->highest_vm_end = vm_end_gap(vma);
596 : :
597 : : /*
598 : : * vma->vm_prev wasn't known when we followed the rbtree to find the
599 : : * correct insertion point for that vma. As a result, we could not
600 : : * update the vma vm_rb parents rb_subtree_gap values on the way down.
601 : : * So, we first insert the vma with a zero rb_subtree_gap value
602 : : * (to be consistent with what we did on the way down), and then
603 : : * immediately update the gap to the correct value. Finally we
604 : : * rebalance the rbtree after all augmented values have been set.
605 : : */
606 : 3 : rb_link_node(&vma->vm_rb, rb_parent, rb_link);
607 : 3 : vma->rb_subtree_gap = 0;
608 : : vma_gap_update(vma);
609 : 3 : vma_rb_insert(vma, &mm->mm_rb);
610 : 3 : }
611 : :
612 : 3 : static void __vma_link_file(struct vm_area_struct *vma)
613 : : {
614 : : struct file *file;
615 : :
616 : 3 : file = vma->vm_file;
617 : 3 : if (file) {
618 : 3 : struct address_space *mapping = file->f_mapping;
619 : :
620 : 3 : if (vma->vm_flags & VM_DENYWRITE)
621 : 3 : atomic_dec(&file_inode(file)->i_writecount);
622 : 3 : if (vma->vm_flags & VM_SHARED)
623 : 3 : atomic_inc(&mapping->i_mmap_writable);
624 : :
625 : : flush_dcache_mmap_lock(mapping);
626 : 3 : vma_interval_tree_insert(vma, &mapping->i_mmap);
627 : : flush_dcache_mmap_unlock(mapping);
628 : : }
629 : 3 : }
630 : :
631 : : static void
632 : 3 : __vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
633 : : struct vm_area_struct *prev, struct rb_node **rb_link,
634 : : struct rb_node *rb_parent)
635 : : {
636 : 3 : __vma_link_list(mm, vma, prev, rb_parent);
637 : 3 : __vma_link_rb(mm, vma, rb_link, rb_parent);
638 : 3 : }
639 : :
640 : 3 : static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma,
641 : : struct vm_area_struct *prev, struct rb_node **rb_link,
642 : : struct rb_node *rb_parent)
643 : : {
644 : : struct address_space *mapping = NULL;
645 : :
646 : 3 : if (vma->vm_file) {
647 : 3 : mapping = vma->vm_file->f_mapping;
648 : : i_mmap_lock_write(mapping);
649 : : }
650 : :
651 : 3 : __vma_link(mm, vma, prev, rb_link, rb_parent);
652 : 3 : __vma_link_file(vma);
653 : :
654 : 3 : if (mapping)
655 : : i_mmap_unlock_write(mapping);
656 : :
657 : 3 : mm->map_count++;
658 : : validate_mm(mm);
659 : 3 : }
660 : :
661 : : /*
662 : : * Helper for vma_adjust() in the split_vma insert case: insert a vma into the
663 : : * mm's list and rbtree. It has already been inserted into the interval tree.
664 : : */
665 : 3 : static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
666 : : {
667 : : struct vm_area_struct *prev;
668 : : struct rb_node **rb_link, *rb_parent;
669 : :
670 : 3 : if (find_vma_links(mm, vma->vm_start, vma->vm_end,
671 : : &prev, &rb_link, &rb_parent))
672 : 0 : BUG();
673 : 3 : __vma_link(mm, vma, prev, rb_link, rb_parent);
674 : 3 : mm->map_count++;
675 : 3 : }
676 : :
677 : : static __always_inline void __vma_unlink_common(struct mm_struct *mm,
678 : : struct vm_area_struct *vma,
679 : : struct vm_area_struct *prev,
680 : : bool has_prev,
681 : : struct vm_area_struct *ignore)
682 : : {
683 : : struct vm_area_struct *next;
684 : :
685 : 3 : vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
686 : 3 : next = vma->vm_next;
687 : : if (has_prev)
688 : 3 : prev->vm_next = next;
689 : : else {
690 : 1 : prev = vma->vm_prev;
691 : 1 : if (prev)
692 : 1 : prev->vm_next = next;
693 : : else
694 : 0 : mm->mmap = next;
695 : : }
696 : 3 : if (next)
697 : 3 : next->vm_prev = prev;
698 : :
699 : : /* Kill the cache */
700 : 3 : vmacache_invalidate(mm);
701 : : }
702 : :
703 : 3 : static inline void __vma_unlink_prev(struct mm_struct *mm,
704 : : struct vm_area_struct *vma,
705 : : struct vm_area_struct *prev)
706 : : {
707 : : __vma_unlink_common(mm, vma, prev, true, vma);
708 : 3 : }
709 : :
710 : : /*
711 : : * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that
712 : : * is already present in an i_mmap tree without adjusting the tree.
713 : : * The following helper function should be used when such adjustments
714 : : * are necessary. The "insert" vma (if any) is to be inserted
715 : : * before we drop the necessary locks.
716 : : */
717 : 3 : int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
718 : : unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
719 : : struct vm_area_struct *expand)
720 : : {
721 : 3 : struct mm_struct *mm = vma->vm_mm;
722 : 3 : struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
723 : : struct address_space *mapping = NULL;
724 : : struct rb_root_cached *root = NULL;
725 : : struct anon_vma *anon_vma = NULL;
726 : 3 : struct file *file = vma->vm_file;
727 : : bool start_changed = false, end_changed = false;
728 : : long adjust_next = 0;
729 : : int remove_next = 0;
730 : :
731 : 3 : if (next && !insert) {
732 : : struct vm_area_struct *exporter = NULL, *importer = NULL;
733 : :
734 : 3 : if (end >= next->vm_end) {
735 : : /*
736 : : * vma expands, overlapping all the next, and
737 : : * perhaps the one after too (mprotect case 6).
738 : : * The only other cases that gets here are
739 : : * case 1, case 7 and case 8.
740 : : */
741 : 3 : if (next == expand) {
742 : : /*
743 : : * The only case where we don't expand "vma"
744 : : * and we expand "next" instead is case 8.
745 : : */
746 : : VM_WARN_ON(end != next->vm_end);
747 : : /*
748 : : * remove_next == 3 means we're
749 : : * removing "vma" and that to do so we
750 : : * swapped "vma" and "next".
751 : : */
752 : : remove_next = 3;
753 : : VM_WARN_ON(file != next->vm_file);
754 : : swap(vma, next);
755 : : } else {
756 : : VM_WARN_ON(expand != vma);
757 : : /*
758 : : * case 1, 6, 7, remove_next == 2 is case 6,
759 : : * remove_next == 1 is case 1 or 7.
760 : : */
761 : 3 : remove_next = 1 + (end > next->vm_end);
762 : : VM_WARN_ON(remove_next == 2 &&
763 : : end != next->vm_next->vm_end);
764 : : VM_WARN_ON(remove_next == 1 &&
765 : : end != next->vm_end);
766 : : /* trim end to next, for case 6 first pass */
767 : : end = next->vm_end;
768 : : }
769 : :
770 : : exporter = next;
771 : : importer = vma;
772 : :
773 : : /*
774 : : * If next doesn't have anon_vma, import from vma after
775 : : * next, if the vma overlaps with it.
776 : : */
777 : 3 : if (remove_next == 2 && !next->anon_vma)
778 : 0 : exporter = next->vm_next;
779 : :
780 : 3 : } else if (end > next->vm_start) {
781 : : /*
782 : : * vma expands, overlapping part of the next:
783 : : * mprotect case 5 shifting the boundary up.
784 : : */
785 : 3 : adjust_next = (end - next->vm_start) >> PAGE_SHIFT;
786 : : exporter = next;
787 : : importer = vma;
788 : : VM_WARN_ON(expand != importer);
789 : 3 : } else if (end < vma->vm_end) {
790 : : /*
791 : : * vma shrinks, and !insert tells it's not
792 : : * split_vma inserting another: so it must be
793 : : * mprotect case 4 shifting the boundary down.
794 : : */
795 : 3 : adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT);
796 : : exporter = vma;
797 : : importer = next;
798 : : VM_WARN_ON(expand != importer);
799 : : }
800 : :
801 : : /*
802 : : * Easily overlooked: when mprotect shifts the boundary,
803 : : * make sure the expanding vma has anon_vma set if the
804 : : * shrinking vma had, to cover any anon pages imported.
805 : : */
806 : 3 : if (exporter && exporter->anon_vma && !importer->anon_vma) {
807 : : int error;
808 : :
809 : 0 : importer->anon_vma = exporter->anon_vma;
810 : 0 : error = anon_vma_clone(importer, exporter);
811 : 0 : if (error)
812 : : return error;
813 : : }
814 : : }
815 : : again:
816 : : vma_adjust_trans_huge(orig_vma, start, end, adjust_next);
817 : :
818 : 3 : if (file) {
819 : 3 : mapping = file->f_mapping;
820 : 3 : root = &mapping->i_mmap;
821 : : uprobe_munmap(vma, vma->vm_start, vma->vm_end);
822 : :
823 : : if (adjust_next)
824 : : uprobe_munmap(next, next->vm_start, next->vm_end);
825 : :
826 : : i_mmap_lock_write(mapping);
827 : 3 : if (insert) {
828 : : /*
829 : : * Put into interval tree now, so instantiated pages
830 : : * are visible to arm/parisc __flush_dcache_page
831 : : * throughout; but we cannot insert into address
832 : : * space until vma start or end is updated.
833 : : */
834 : 3 : __vma_link_file(insert);
835 : : }
836 : : }
837 : :
838 : 3 : anon_vma = vma->anon_vma;
839 : 3 : if (!anon_vma && adjust_next)
840 : 3 : anon_vma = next->anon_vma;
841 : 3 : if (anon_vma) {
842 : : VM_WARN_ON(adjust_next && next->anon_vma &&
843 : : anon_vma != next->anon_vma);
844 : : anon_vma_lock_write(anon_vma);
845 : 3 : anon_vma_interval_tree_pre_update_vma(vma);
846 : 3 : if (adjust_next)
847 : 3 : anon_vma_interval_tree_pre_update_vma(next);
848 : : }
849 : :
850 : 3 : if (root) {
851 : : flush_dcache_mmap_lock(mapping);
852 : 3 : vma_interval_tree_remove(vma, root);
853 : 3 : if (adjust_next)
854 : 0 : vma_interval_tree_remove(next, root);
855 : : }
856 : :
857 : 3 : if (start != vma->vm_start) {
858 : 3 : vma->vm_start = start;
859 : : start_changed = true;
860 : : }
861 : 3 : if (end != vma->vm_end) {
862 : 3 : vma->vm_end = end;
863 : : end_changed = true;
864 : : }
865 : 3 : vma->vm_pgoff = pgoff;
866 : 3 : if (adjust_next) {
867 : 3 : next->vm_start += adjust_next << PAGE_SHIFT;
868 : 3 : next->vm_pgoff += adjust_next;
869 : : }
870 : :
871 : 3 : if (root) {
872 : 3 : if (adjust_next)
873 : 0 : vma_interval_tree_insert(next, root);
874 : 3 : vma_interval_tree_insert(vma, root);
875 : : flush_dcache_mmap_unlock(mapping);
876 : : }
877 : :
878 : 3 : if (remove_next) {
879 : : /*
880 : : * vma_merge has merged next into vma, and needs
881 : : * us to remove next before dropping the locks.
882 : : */
883 : 3 : if (remove_next != 3)
884 : 3 : __vma_unlink_prev(mm, next, vma);
885 : : else
886 : : /*
887 : : * vma is not before next if they've been
888 : : * swapped.
889 : : *
890 : : * pre-swap() next->vm_start was reduced so
891 : : * tell validate_mm_rb to ignore pre-swap()
892 : : * "next" (which is stored in post-swap()
893 : : * "vma").
894 : : */
895 : : __vma_unlink_common(mm, next, NULL, false, vma);
896 : 3 : if (file)
897 : 1 : __remove_shared_vm_struct(next, file, mapping);
898 : 3 : } else if (insert) {
899 : : /*
900 : : * split_vma has split insert from vma, and needs
901 : : * us to insert it before dropping the locks
902 : : * (it may either follow vma or precede it).
903 : : */
904 : 3 : __insert_vm_struct(mm, insert);
905 : : } else {
906 : 3 : if (start_changed)
907 : : vma_gap_update(vma);
908 : 3 : if (end_changed) {
909 : 3 : if (!next)
910 : 3 : mm->highest_vm_end = vm_end_gap(vma);
911 : 3 : else if (!adjust_next)
912 : : vma_gap_update(next);
913 : : }
914 : : }
915 : :
916 : 3 : if (anon_vma) {
917 : 3 : anon_vma_interval_tree_post_update_vma(vma);
918 : 3 : if (adjust_next)
919 : 3 : anon_vma_interval_tree_post_update_vma(next);
920 : : anon_vma_unlock_write(anon_vma);
921 : : }
922 : 3 : if (mapping)
923 : : i_mmap_unlock_write(mapping);
924 : :
925 : : if (root) {
926 : : uprobe_mmap(vma);
927 : :
928 : : if (adjust_next)
929 : : uprobe_mmap(next);
930 : : }
931 : :
932 : 3 : if (remove_next) {
933 : 3 : if (file) {
934 : : uprobe_munmap(next, next->vm_start, next->vm_end);
935 : 1 : fput(file);
936 : : }
937 : 3 : if (next->anon_vma)
938 : : anon_vma_merge(vma, next);
939 : 3 : mm->map_count--;
940 : : mpol_put(vma_policy(next));
941 : 3 : vm_area_free(next);
942 : : /*
943 : : * In mprotect's case 6 (see comments on vma_merge),
944 : : * we must remove another next too. It would clutter
945 : : * up the code too much to do both in one go.
946 : : */
947 : 3 : if (remove_next != 3) {
948 : : /*
949 : : * If "next" was removed and vma->vm_end was
950 : : * expanded (up) over it, in turn
951 : : * "next->vm_prev->vm_end" changed and the
952 : : * "vma->vm_next" gap must be updated.
953 : : */
954 : 3 : next = vma->vm_next;
955 : : } else {
956 : : /*
957 : : * For the scope of the comment "next" and
958 : : * "vma" considered pre-swap(): if "vma" was
959 : : * removed, next->vm_start was expanded (down)
960 : : * over it and the "next" gap must be updated.
961 : : * Because of the swap() the post-swap() "vma"
962 : : * actually points to pre-swap() "next"
963 : : * (post-swap() "next" as opposed is now a
964 : : * dangling pointer).
965 : : */
966 : : next = vma;
967 : : }
968 : 3 : if (remove_next == 2) {
969 : : remove_next = 1;
970 : 3 : end = next->vm_end;
971 : 3 : goto again;
972 : : }
973 : 3 : else if (next)
974 : : vma_gap_update(next);
975 : : else {
976 : : /*
977 : : * If remove_next == 2 we obviously can't
978 : : * reach this path.
979 : : *
980 : : * If remove_next == 3 we can't reach this
981 : : * path because pre-swap() next is always not
982 : : * NULL. pre-swap() "next" is not being
983 : : * removed and its next->vm_end is not altered
984 : : * (and furthermore "end" already matches
985 : : * next->vm_end in remove_next == 3).
986 : : *
987 : : * We reach this only in the remove_next == 1
988 : : * case if the "next" vma that was removed was
989 : : * the highest vma of the mm. However in such
990 : : * case next->vm_end == "end" and the extended
991 : : * "vma" has vma->vm_end == next->vm_end so
992 : : * mm->highest_vm_end doesn't need any update
993 : : * in remove_next == 1 case.
994 : : */
995 : : VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma));
996 : : }
997 : : }
998 : : if (insert && file)
999 : : uprobe_mmap(insert);
1000 : :
1001 : : validate_mm(mm);
1002 : :
1003 : : return 0;
1004 : : }
1005 : :
1006 : : /*
1007 : : * If the vma has a ->close operation then the driver probably needs to release
1008 : : * per-vma resources, so we don't attempt to merge those.
1009 : : */
1010 : : static inline int is_mergeable_vma(struct vm_area_struct *vma,
1011 : : struct file *file, unsigned long vm_flags,
1012 : : struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1013 : : {
1014 : : /*
1015 : : * VM_SOFTDIRTY should not prevent from VMA merging, if we
1016 : : * match the flags but dirty bit -- the caller should mark
1017 : : * merged VMA as dirty. If dirty bit won't be excluded from
1018 : : * comparison, we increase pressure on the memory system forcing
1019 : : * the kernel to generate new VMAs when old one could be
1020 : : * extended instead.
1021 : : */
1022 : 3 : if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY)
1023 : : return 0;
1024 : 3 : if (vma->vm_file != file)
1025 : : return 0;
1026 : 3 : if (vma->vm_ops && vma->vm_ops->close)
1027 : : return 0;
1028 : : if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx))
1029 : : return 0;
1030 : : return 1;
1031 : : }
1032 : :
1033 : : static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1,
1034 : : struct anon_vma *anon_vma2,
1035 : : struct vm_area_struct *vma)
1036 : : {
1037 : : /*
1038 : : * The list_is_singular() test is to avoid merging VMA cloned from
1039 : : * parents. This can improve scalability caused by anon_vma lock.
1040 : : */
1041 : 3 : if ((!anon_vma1 || !anon_vma2) && (!vma ||
1042 : 3 : list_is_singular(&vma->anon_vma_chain)))
1043 : : return 1;
1044 : 3 : return anon_vma1 == anon_vma2;
1045 : : }
1046 : :
1047 : : /*
1048 : : * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1049 : : * in front of (at a lower virtual address and file offset than) the vma.
1050 : : *
1051 : : * We cannot merge two vmas if they have differently assigned (non-NULL)
1052 : : * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1053 : : *
1054 : : * We don't check here for the merged mmap wrapping around the end of pagecache
1055 : : * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which
1056 : : * wrap, nor mmaps which cover the final page at index -1UL.
1057 : : */
1058 : : static int
1059 : 3 : can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags,
1060 : : struct anon_vma *anon_vma, struct file *file,
1061 : : pgoff_t vm_pgoff,
1062 : : struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1063 : : {
1064 : 3 : if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1065 : 3 : is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1066 : 3 : if (vma->vm_pgoff == vm_pgoff)
1067 : : return 1;
1068 : : }
1069 : : return 0;
1070 : : }
1071 : :
1072 : : /*
1073 : : * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff)
1074 : : * beyond (at a higher virtual address and file offset than) the vma.
1075 : : *
1076 : : * We cannot merge two vmas if they have differently assigned (non-NULL)
1077 : : * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
1078 : : */
1079 : : static int
1080 : 3 : can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
1081 : : struct anon_vma *anon_vma, struct file *file,
1082 : : pgoff_t vm_pgoff,
1083 : : struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1084 : : {
1085 : 3 : if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) &&
1086 : 3 : is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) {
1087 : : pgoff_t vm_pglen;
1088 : : vm_pglen = vma_pages(vma);
1089 : 3 : if (vma->vm_pgoff + vm_pglen == vm_pgoff)
1090 : : return 1;
1091 : : }
1092 : : return 0;
1093 : : }
1094 : :
1095 : : /*
1096 : : * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out
1097 : : * whether that can be merged with its predecessor or its successor.
1098 : : * Or both (it neatly fills a hole).
1099 : : *
1100 : : * In most cases - when called for mmap, brk or mremap - [addr,end) is
1101 : : * certain not to be mapped by the time vma_merge is called; but when
1102 : : * called for mprotect, it is certain to be already mapped (either at
1103 : : * an offset within prev, or at the start of next), and the flags of
1104 : : * this area are about to be changed to vm_flags - and the no-change
1105 : : * case has already been eliminated.
1106 : : *
1107 : : * The following mprotect cases have to be considered, where AAAA is
1108 : : * the area passed down from mprotect_fixup, never extending beyond one
1109 : : * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after:
1110 : : *
1111 : : * AAAA AAAA AAAA AAAA
1112 : : * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX
1113 : : * cannot merge might become might become might become
1114 : : * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or
1115 : : * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or
1116 : : * mremap move: PPPPXXXXXXXX 8
1117 : : * AAAA
1118 : : * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN
1119 : : * might become case 1 below case 2 below case 3 below
1120 : : *
1121 : : * It is important for case 8 that the vma NNNN overlapping the
1122 : : * region AAAA is never going to extended over XXXX. Instead XXXX must
1123 : : * be extended in region AAAA and NNNN must be removed. This way in
1124 : : * all cases where vma_merge succeeds, the moment vma_adjust drops the
1125 : : * rmap_locks, the properties of the merged vma will be already
1126 : : * correct for the whole merged range. Some of those properties like
1127 : : * vm_page_prot/vm_flags may be accessed by rmap_walks and they must
1128 : : * be correct for the whole merged range immediately after the
1129 : : * rmap_locks are released. Otherwise if XXXX would be removed and
1130 : : * NNNN would be extended over the XXXX range, remove_migration_ptes
1131 : : * or other rmap walkers (if working on addresses beyond the "end"
1132 : : * parameter) may establish ptes with the wrong permissions of NNNN
1133 : : * instead of the right permissions of XXXX.
1134 : : */
1135 : 3 : struct vm_area_struct *vma_merge(struct mm_struct *mm,
1136 : : struct vm_area_struct *prev, unsigned long addr,
1137 : : unsigned long end, unsigned long vm_flags,
1138 : : struct anon_vma *anon_vma, struct file *file,
1139 : : pgoff_t pgoff, struct mempolicy *policy,
1140 : : struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
1141 : : {
1142 : 3 : pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
1143 : : struct vm_area_struct *area, *next;
1144 : : int err;
1145 : :
1146 : : /*
1147 : : * We later require that vma->vm_flags == vm_flags,
1148 : : * so this tests vma->vm_flags & VM_SPECIAL, too.
1149 : : */
1150 : 3 : if (vm_flags & VM_SPECIAL)
1151 : : return NULL;
1152 : :
1153 : 3 : if (prev)
1154 : 3 : next = prev->vm_next;
1155 : : else
1156 : 3 : next = mm->mmap;
1157 : : area = next;
1158 : 3 : if (area && area->vm_end == end) /* cases 6, 7, 8 */
1159 : 3 : next = next->vm_next;
1160 : :
1161 : : /* verify some invariant that must be enforced by the caller */
1162 : : VM_WARN_ON(prev && addr <= prev->vm_start);
1163 : : VM_WARN_ON(area && end > area->vm_end);
1164 : : VM_WARN_ON(addr >= end);
1165 : :
1166 : : /*
1167 : : * Can it merge with the predecessor?
1168 : : */
1169 : 3 : if (prev && prev->vm_end == addr &&
1170 : 3 : mpol_equal(vma_policy(prev), policy) &&
1171 : 3 : can_vma_merge_after(prev, vm_flags,
1172 : : anon_vma, file, pgoff,
1173 : : vm_userfaultfd_ctx)) {
1174 : : /*
1175 : : * OK, it can. Can we now merge in the successor as well?
1176 : : */
1177 : 3 : if (next && end == next->vm_start &&
1178 : 3 : mpol_equal(policy, vma_policy(next)) &&
1179 : 3 : can_vma_merge_before(next, vm_flags,
1180 : : anon_vma, file,
1181 : : pgoff+pglen,
1182 : 3 : vm_userfaultfd_ctx) &&
1183 : 3 : is_mergeable_anon_vma(prev->anon_vma,
1184 : : next->anon_vma, NULL)) {
1185 : : /* cases 1, 6 */
1186 : 3 : err = __vma_adjust(prev, prev->vm_start,
1187 : : next->vm_end, prev->vm_pgoff, NULL,
1188 : : prev);
1189 : : } else /* cases 2, 5, 7 */
1190 : 3 : err = __vma_adjust(prev, prev->vm_start,
1191 : : end, prev->vm_pgoff, NULL, prev);
1192 : 3 : if (err)
1193 : : return NULL;
1194 : : khugepaged_enter_vma_merge(prev, vm_flags);
1195 : 3 : return prev;
1196 : : }
1197 : :
1198 : : /*
1199 : : * Can this new request be merged in front of next?
1200 : : */
1201 : 3 : if (next && end == next->vm_start &&
1202 : 3 : mpol_equal(policy, vma_policy(next)) &&
1203 : 3 : can_vma_merge_before(next, vm_flags,
1204 : : anon_vma, file, pgoff+pglen,
1205 : : vm_userfaultfd_ctx)) {
1206 : 3 : if (prev && addr < prev->vm_end) /* case 4 */
1207 : 3 : err = __vma_adjust(prev, prev->vm_start,
1208 : : addr, prev->vm_pgoff, NULL, next);
1209 : : else { /* cases 3, 8 */
1210 : 3 : err = __vma_adjust(area, addr, next->vm_end,
1211 : 3 : next->vm_pgoff - pglen, NULL, next);
1212 : : /*
1213 : : * In case 3 area is already equal to next and
1214 : : * this is a noop, but in case 8 "area" has
1215 : : * been removed and next was expanded over it.
1216 : : */
1217 : : area = next;
1218 : : }
1219 : 3 : if (err)
1220 : : return NULL;
1221 : : khugepaged_enter_vma_merge(area, vm_flags);
1222 : 3 : return area;
1223 : : }
1224 : :
1225 : : return NULL;
1226 : : }
1227 : :
1228 : : /*
1229 : : * Rough compatbility check to quickly see if it's even worth looking
1230 : : * at sharing an anon_vma.
1231 : : *
1232 : : * They need to have the same vm_file, and the flags can only differ
1233 : : * in things that mprotect may change.
1234 : : *
1235 : : * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
1236 : : * we can merge the two vma's. For example, we refuse to merge a vma if
1237 : : * there is a vm_ops->close() function, because that indicates that the
1238 : : * driver is doing some kind of reference counting. But that doesn't
1239 : : * really matter for the anon_vma sharing case.
1240 : : */
1241 : 3 : static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
1242 : : {
1243 : 3 : return a->vm_end == b->vm_start &&
1244 : 3 : mpol_equal(vma_policy(a), vma_policy(b)) &&
1245 : 3 : a->vm_file == b->vm_file &&
1246 : 3 : !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) &&
1247 : 3 : b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT);
1248 : : }
1249 : :
1250 : : /*
1251 : : * Do some basic sanity checking to see if we can re-use the anon_vma
1252 : : * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
1253 : : * the same as 'old', the other will be the new one that is trying
1254 : : * to share the anon_vma.
1255 : : *
1256 : : * NOTE! This runs with mm_sem held for reading, so it is possible that
1257 : : * the anon_vma of 'old' is concurrently in the process of being set up
1258 : : * by another page fault trying to merge _that_. But that's ok: if it
1259 : : * is being set up, that automatically means that it will be a singleton
1260 : : * acceptable for merging, so we can do all of this optimistically. But
1261 : : * we do that READ_ONCE() to make sure that we never re-load the pointer.
1262 : : *
1263 : : * IOW: that the "list_is_singular()" test on the anon_vma_chain only
1264 : : * matters for the 'stable anon_vma' case (ie the thing we want to avoid
1265 : : * is to return an anon_vma that is "complex" due to having gone through
1266 : : * a fork).
1267 : : *
1268 : : * We also make sure that the two vma's are compatible (adjacent,
1269 : : * and with the same memory policies). That's all stable, even with just
1270 : : * a read lock on the mm_sem.
1271 : : */
1272 : 3 : static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b)
1273 : : {
1274 : 3 : if (anon_vma_compatible(a, b)) {
1275 : 3 : struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
1276 : :
1277 : 3 : if (anon_vma && list_is_singular(&old->anon_vma_chain))
1278 : 1 : return anon_vma;
1279 : : }
1280 : : return NULL;
1281 : : }
1282 : :
1283 : : /*
1284 : : * find_mergeable_anon_vma is used by anon_vma_prepare, to check
1285 : : * neighbouring vmas for a suitable anon_vma, before it goes off
1286 : : * to allocate a new anon_vma. It checks because a repetitive
1287 : : * sequence of mprotects and faults may otherwise lead to distinct
1288 : : * anon_vmas being allocated, preventing vma merge in subsequent
1289 : : * mprotect.
1290 : : */
1291 : 3 : struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
1292 : : {
1293 : : struct anon_vma *anon_vma;
1294 : : struct vm_area_struct *near;
1295 : :
1296 : 3 : near = vma->vm_next;
1297 : 3 : if (!near)
1298 : : goto try_prev;
1299 : :
1300 : 3 : anon_vma = reusable_anon_vma(near, vma, near);
1301 : 3 : if (anon_vma)
1302 : : return anon_vma;
1303 : : try_prev:
1304 : 3 : near = vma->vm_prev;
1305 : 3 : if (!near)
1306 : : goto none;
1307 : :
1308 : 3 : anon_vma = reusable_anon_vma(near, near, vma);
1309 : 3 : if (anon_vma)
1310 : 0 : return anon_vma;
1311 : : none:
1312 : : /*
1313 : : * There's no absolute need to look only at touching neighbours:
1314 : : * we could search further afield for "compatible" anon_vmas.
1315 : : * But it would probably just be a waste of time searching,
1316 : : * or lead to too many vmas hanging off the same anon_vma.
1317 : : * We're trying to allow mprotect remerging later on,
1318 : : * not trying to minimize memory used for anon_vmas.
1319 : : */
1320 : : return NULL;
1321 : : }
1322 : :
1323 : : /*
1324 : : * If a hint addr is less than mmap_min_addr change hint to be as
1325 : : * low as possible but still greater than mmap_min_addr
1326 : : */
1327 : : static inline unsigned long round_hint_to_min(unsigned long hint)
1328 : : {
1329 : 3 : hint &= PAGE_MASK;
1330 : 3 : if (((void *)hint != NULL) &&
1331 : 3 : (hint < mmap_min_addr))
1332 : 0 : return PAGE_ALIGN(mmap_min_addr);
1333 : : return hint;
1334 : : }
1335 : :
1336 : 3 : static inline int mlock_future_check(struct mm_struct *mm,
1337 : : unsigned long flags,
1338 : : unsigned long len)
1339 : : {
1340 : : unsigned long locked, lock_limit;
1341 : :
1342 : : /* mlock MCL_FUTURE? */
1343 : 3 : if (flags & VM_LOCKED) {
1344 : 0 : locked = len >> PAGE_SHIFT;
1345 : 0 : locked += mm->locked_vm;
1346 : : lock_limit = rlimit(RLIMIT_MEMLOCK);
1347 : 0 : lock_limit >>= PAGE_SHIFT;
1348 : 0 : if (locked > lock_limit && !capable(CAP_IPC_LOCK))
1349 : : return -EAGAIN;
1350 : : }
1351 : : return 0;
1352 : : }
1353 : :
1354 : : static inline u64 file_mmap_size_max(struct file *file, struct inode *inode)
1355 : : {
1356 : 3 : if (S_ISREG(inode->i_mode))
1357 : : return MAX_LFS_FILESIZE;
1358 : :
1359 : 3 : if (S_ISBLK(inode->i_mode))
1360 : : return MAX_LFS_FILESIZE;
1361 : :
1362 : 3 : if (S_ISSOCK(inode->i_mode))
1363 : : return MAX_LFS_FILESIZE;
1364 : :
1365 : : /* Special "we do even unsigned file positions" case */
1366 : 3 : if (file->f_mode & FMODE_UNSIGNED_OFFSET)
1367 : : return 0;
1368 : :
1369 : : /* Yes, random drivers might want more. But I'm tired of buggy drivers */
1370 : : return ULONG_MAX;
1371 : : }
1372 : :
1373 : 3 : static inline bool file_mmap_ok(struct file *file, struct inode *inode,
1374 : : unsigned long pgoff, unsigned long len)
1375 : : {
1376 : : u64 maxsize = file_mmap_size_max(file, inode);
1377 : :
1378 : 3 : if (maxsize && len > maxsize)
1379 : : return false;
1380 : 3 : maxsize -= len;
1381 : 3 : if (pgoff > maxsize >> PAGE_SHIFT)
1382 : : return false;
1383 : 3 : return true;
1384 : : }
1385 : :
1386 : : /*
1387 : : * The caller must hold down_write(¤t->mm->mmap_sem).
1388 : : */
1389 : 3 : unsigned long do_mmap(struct file *file, unsigned long addr,
1390 : : unsigned long len, unsigned long prot,
1391 : : unsigned long flags, vm_flags_t vm_flags,
1392 : : unsigned long pgoff, unsigned long *populate,
1393 : : struct list_head *uf)
1394 : : {
1395 : 3 : struct mm_struct *mm = current->mm;
1396 : : int pkey = 0;
1397 : :
1398 : 3 : *populate = 0;
1399 : :
1400 : 3 : if (!len)
1401 : : return -EINVAL;
1402 : :
1403 : : /*
1404 : : * Does the application expect PROT_READ to imply PROT_EXEC?
1405 : : *
1406 : : * (the exception is when the underlying filesystem is noexec
1407 : : * mounted, in which case we dont add PROT_EXEC.)
1408 : : */
1409 : 3 : if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC))
1410 : 0 : if (!(file && path_noexec(&file->f_path)))
1411 : 0 : prot |= PROT_EXEC;
1412 : :
1413 : : /* force arch specific MAP_FIXED handling in get_unmapped_area */
1414 : 3 : if (flags & MAP_FIXED_NOREPLACE)
1415 : 3 : flags |= MAP_FIXED;
1416 : :
1417 : 3 : if (!(flags & MAP_FIXED))
1418 : : addr = round_hint_to_min(addr);
1419 : :
1420 : : /* Careful about overflows.. */
1421 : 3 : len = PAGE_ALIGN(len);
1422 : 3 : if (!len)
1423 : : return -ENOMEM;
1424 : :
1425 : : /* offset overflow? */
1426 : 3 : if ((pgoff + (len >> PAGE_SHIFT)) < pgoff)
1427 : : return -EOVERFLOW;
1428 : :
1429 : : /* Too many mappings? */
1430 : 3 : if (mm->map_count > sysctl_max_map_count)
1431 : : return -ENOMEM;
1432 : :
1433 : : /* Obtain the address to map to. we verify (or select) it and ensure
1434 : : * that it represents a valid section of the address space.
1435 : : */
1436 : 3 : addr = get_unmapped_area(file, addr, len, pgoff, flags);
1437 : 3 : if (offset_in_page(addr))
1438 : : return addr;
1439 : :
1440 : 3 : if (flags & MAP_FIXED_NOREPLACE) {
1441 : 3 : struct vm_area_struct *vma = find_vma(mm, addr);
1442 : :
1443 : 3 : if (vma && vma->vm_start < addr + len)
1444 : : return -EEXIST;
1445 : : }
1446 : :
1447 : : if (prot == PROT_EXEC) {
1448 : : pkey = execute_only_pkey(mm);
1449 : : if (pkey < 0)
1450 : : pkey = 0;
1451 : : }
1452 : :
1453 : : /* Do simple checking here so the lower-level routines won't have
1454 : : * to. we assume access permissions have been handled by the open
1455 : : * of the memory object, so we don't do any here.
1456 : : */
1457 : 3 : vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) |
1458 : 3 : mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC;
1459 : :
1460 : 3 : if (flags & MAP_LOCKED)
1461 : 0 : if (!can_do_mlock())
1462 : : return -EPERM;
1463 : :
1464 : 3 : if (mlock_future_check(mm, vm_flags, len))
1465 : : return -EAGAIN;
1466 : :
1467 : 3 : if (file) {
1468 : : struct inode *inode = file_inode(file);
1469 : : unsigned long flags_mask;
1470 : :
1471 : 3 : if (!file_mmap_ok(file, inode, pgoff, len))
1472 : : return -EOVERFLOW;
1473 : :
1474 : 3 : flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags;
1475 : :
1476 : 3 : switch (flags & MAP_TYPE) {
1477 : : case MAP_SHARED:
1478 : : /*
1479 : : * Force use of MAP_SHARED_VALIDATE with non-legacy
1480 : : * flags. E.g. MAP_SYNC is dangerous to use with
1481 : : * MAP_SHARED as you don't know which consistency model
1482 : : * you will get. We silently ignore unsupported flags
1483 : : * with MAP_SHARED to preserve backward compatibility.
1484 : : */
1485 : 3 : flags &= LEGACY_MAP_MASK;
1486 : : /* fall through */
1487 : : case MAP_SHARED_VALIDATE:
1488 : 3 : if (flags & ~flags_mask)
1489 : : return -EOPNOTSUPP;
1490 : 3 : if (prot & PROT_WRITE) {
1491 : 3 : if (!(file->f_mode & FMODE_WRITE))
1492 : : return -EACCES;
1493 : 3 : if (IS_SWAPFILE(file->f_mapping->host))
1494 : : return -ETXTBSY;
1495 : : }
1496 : :
1497 : : /*
1498 : : * Make sure we don't allow writing to an append-only
1499 : : * file..
1500 : : */
1501 : 3 : if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE))
1502 : : return -EACCES;
1503 : :
1504 : : /*
1505 : : * Make sure there are no mandatory locks on the file.
1506 : : */
1507 : 3 : if (locks_verify_locked(file))
1508 : : return -EAGAIN;
1509 : :
1510 : 3 : vm_flags |= VM_SHARED | VM_MAYSHARE;
1511 : 3 : if (!(file->f_mode & FMODE_WRITE))
1512 : 3 : vm_flags &= ~(VM_MAYWRITE | VM_SHARED);
1513 : :
1514 : : /* fall through */
1515 : : case MAP_PRIVATE:
1516 : 3 : if (!(file->f_mode & FMODE_READ))
1517 : : return -EACCES;
1518 : 3 : if (path_noexec(&file->f_path)) {
1519 : 0 : if (vm_flags & VM_EXEC)
1520 : : return -EPERM;
1521 : 0 : vm_flags &= ~VM_MAYEXEC;
1522 : : }
1523 : :
1524 : 3 : if (!file->f_op->mmap)
1525 : : return -ENODEV;
1526 : 3 : if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1527 : : return -EINVAL;
1528 : : break;
1529 : :
1530 : : default:
1531 : : return -EINVAL;
1532 : : }
1533 : : } else {
1534 : 3 : switch (flags & MAP_TYPE) {
1535 : : case MAP_SHARED:
1536 : 0 : if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP))
1537 : : return -EINVAL;
1538 : : /*
1539 : : * Ignore pgoff.
1540 : : */
1541 : : pgoff = 0;
1542 : 0 : vm_flags |= VM_SHARED | VM_MAYSHARE;
1543 : 0 : break;
1544 : : case MAP_PRIVATE:
1545 : : /*
1546 : : * Set pgoff according to addr for anon_vma.
1547 : : */
1548 : 3 : pgoff = addr >> PAGE_SHIFT;
1549 : 3 : break;
1550 : : default:
1551 : : return -EINVAL;
1552 : : }
1553 : : }
1554 : :
1555 : : /*
1556 : : * Set 'VM_NORESERVE' if we should not account for the
1557 : : * memory use of this mapping.
1558 : : */
1559 : 3 : if (flags & MAP_NORESERVE) {
1560 : : /* We honor MAP_NORESERVE if allowed to overcommit */
1561 : 3 : if (sysctl_overcommit_memory != OVERCOMMIT_NEVER)
1562 : 3 : vm_flags |= VM_NORESERVE;
1563 : :
1564 : : /* hugetlb applies strict overcommit unless MAP_NORESERVE */
1565 : : if (file && is_file_hugepages(file))
1566 : : vm_flags |= VM_NORESERVE;
1567 : : }
1568 : :
1569 : 3 : addr = mmap_region(file, addr, len, vm_flags, pgoff, uf);
1570 : 3 : if (!IS_ERR_VALUE(addr) &&
1571 : 3 : ((vm_flags & VM_LOCKED) ||
1572 : 3 : (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE))
1573 : 2 : *populate = len;
1574 : 3 : return addr;
1575 : : }
1576 : :
1577 : 3 : unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len,
1578 : : unsigned long prot, unsigned long flags,
1579 : : unsigned long fd, unsigned long pgoff)
1580 : : {
1581 : : struct file *file = NULL;
1582 : : unsigned long retval;
1583 : :
1584 : 3 : if (!(flags & MAP_ANONYMOUS)) {
1585 : 3 : audit_mmap_fd(fd, flags);
1586 : 3 : file = fget(fd);
1587 : 3 : if (!file)
1588 : : return -EBADF;
1589 : : if (is_file_hugepages(file))
1590 : : len = ALIGN(len, huge_page_size(hstate_file(file)));
1591 : : retval = -EINVAL;
1592 : 3 : if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file)))
1593 : : goto out_fput;
1594 : 3 : } else if (flags & MAP_HUGETLB) {
1595 : : struct user_struct *user = NULL;
1596 : : struct hstate *hs;
1597 : :
1598 : : hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1599 : : if (!hs)
1600 : : return -EINVAL;
1601 : :
1602 : : len = ALIGN(len, huge_page_size(hs));
1603 : : /*
1604 : : * VM_NORESERVE is used because the reservations will be
1605 : : * taken when vm_ops->mmap() is called
1606 : : * A dummy user value is used because we are not locking
1607 : : * memory so no accounting is necessary
1608 : : */
1609 : : file = hugetlb_file_setup(HUGETLB_ANON_FILE, len,
1610 : : VM_NORESERVE,
1611 : : &user, HUGETLB_ANONHUGE_INODE,
1612 : : (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK);
1613 : : if (IS_ERR(file))
1614 : : return PTR_ERR(file);
1615 : : }
1616 : :
1617 : 3 : flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE);
1618 : :
1619 : 3 : retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff);
1620 : : out_fput:
1621 : 3 : if (file)
1622 : 3 : fput(file);
1623 : 3 : return retval;
1624 : : }
1625 : :
1626 : 3 : SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len,
1627 : : unsigned long, prot, unsigned long, flags,
1628 : : unsigned long, fd, unsigned long, pgoff)
1629 : : {
1630 : 3 : return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
1631 : : }
1632 : :
1633 : : #ifdef __ARCH_WANT_SYS_OLD_MMAP
1634 : : struct mmap_arg_struct {
1635 : : unsigned long addr;
1636 : : unsigned long len;
1637 : : unsigned long prot;
1638 : : unsigned long flags;
1639 : : unsigned long fd;
1640 : : unsigned long offset;
1641 : : };
1642 : :
1643 : 0 : SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg)
1644 : : {
1645 : : struct mmap_arg_struct a;
1646 : :
1647 : 0 : if (copy_from_user(&a, arg, sizeof(a)))
1648 : : return -EFAULT;
1649 : 0 : if (offset_in_page(a.offset))
1650 : : return -EINVAL;
1651 : :
1652 : 0 : return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
1653 : : a.offset >> PAGE_SHIFT);
1654 : : }
1655 : : #endif /* __ARCH_WANT_SYS_OLD_MMAP */
1656 : :
1657 : : /*
1658 : : * Some shared mappings will want the pages marked read-only
1659 : : * to track write events. If so, we'll downgrade vm_page_prot
1660 : : * to the private version (using protection_map[] without the
1661 : : * VM_SHARED bit).
1662 : : */
1663 : 3 : int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
1664 : : {
1665 : 3 : vm_flags_t vm_flags = vma->vm_flags;
1666 : 3 : const struct vm_operations_struct *vm_ops = vma->vm_ops;
1667 : :
1668 : : /* If it was private or non-writable, the write bit is already clear */
1669 : 3 : if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED)))
1670 : : return 0;
1671 : :
1672 : : /* The backer wishes to know when pages are first written to? */
1673 : 3 : if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite))
1674 : : return 1;
1675 : :
1676 : : /* The open routine did something to the protections that pgprot_modify
1677 : : * won't preserve? */
1678 : 3 : if (pgprot_val(vm_page_prot) !=
1679 : 3 : pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags)))
1680 : : return 0;
1681 : :
1682 : : /* Do we need to track softdirty? */
1683 : : if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY))
1684 : : return 1;
1685 : :
1686 : : /* Specialty mapping? */
1687 : 3 : if (vm_flags & VM_PFNMAP)
1688 : : return 0;
1689 : :
1690 : : /* Can the mapping track the dirty pages? */
1691 : 3 : return vma->vm_file && vma->vm_file->f_mapping &&
1692 : : mapping_cap_account_dirty(vma->vm_file->f_mapping);
1693 : : }
1694 : :
1695 : : /*
1696 : : * We account for memory if it's a private writeable mapping,
1697 : : * not hugepages and VM_NORESERVE wasn't set.
1698 : : */
1699 : : static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
1700 : : {
1701 : : /*
1702 : : * hugetlb has its own accounting separate from the core VM
1703 : : * VM_HUGETLB may not be set yet so we cannot check for that flag.
1704 : : */
1705 : : if (file && is_file_hugepages(file))
1706 : : return 0;
1707 : :
1708 : 3 : return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
1709 : : }
1710 : :
1711 : 3 : unsigned long mmap_region(struct file *file, unsigned long addr,
1712 : : unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
1713 : : struct list_head *uf)
1714 : : {
1715 : 3 : struct mm_struct *mm = current->mm;
1716 : : struct vm_area_struct *vma, *prev;
1717 : : int error;
1718 : : struct rb_node **rb_link, *rb_parent;
1719 : : unsigned long charged = 0;
1720 : :
1721 : : /* Check against address space limit. */
1722 : 3 : if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) {
1723 : : unsigned long nr_pages;
1724 : :
1725 : : /*
1726 : : * MAP_FIXED may remove pages of mappings that intersects with
1727 : : * requested mapping. Account for the pages it would unmap.
1728 : : */
1729 : 0 : nr_pages = count_vma_pages_range(mm, addr, addr + len);
1730 : :
1731 : 0 : if (!may_expand_vm(mm, vm_flags,
1732 : : (len >> PAGE_SHIFT) - nr_pages))
1733 : : return -ENOMEM;
1734 : : }
1735 : :
1736 : : /* Clear old maps */
1737 : 3 : while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
1738 : : &rb_parent)) {
1739 : 3 : if (do_munmap(mm, addr, len, uf))
1740 : : return -ENOMEM;
1741 : : }
1742 : :
1743 : : /*
1744 : : * Private writable mapping: check memory availability
1745 : : */
1746 : 3 : if (accountable_mapping(file, vm_flags)) {
1747 : : charged = len >> PAGE_SHIFT;
1748 : 3 : if (security_vm_enough_memory_mm(mm, charged))
1749 : : return -ENOMEM;
1750 : 3 : vm_flags |= VM_ACCOUNT;
1751 : : }
1752 : :
1753 : : /*
1754 : : * Can we just expand an old mapping?
1755 : : */
1756 : 3 : vma = vma_merge(mm, prev, addr, addr + len, vm_flags,
1757 : : NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX);
1758 : 3 : if (vma)
1759 : : goto out;
1760 : :
1761 : : /*
1762 : : * Determine the object being mapped and call the appropriate
1763 : : * specific mapper. the address has already been validated, but
1764 : : * not unmapped, but the maps are removed from the list.
1765 : : */
1766 : 3 : vma = vm_area_alloc(mm);
1767 : 3 : if (!vma) {
1768 : : error = -ENOMEM;
1769 : : goto unacct_error;
1770 : : }
1771 : :
1772 : 3 : vma->vm_start = addr;
1773 : 3 : vma->vm_end = addr + len;
1774 : 3 : vma->vm_flags = vm_flags;
1775 : 3 : vma->vm_page_prot = vm_get_page_prot(vm_flags);
1776 : 3 : vma->vm_pgoff = pgoff;
1777 : :
1778 : 3 : if (file) {
1779 : 3 : if (vm_flags & VM_DENYWRITE) {
1780 : : error = deny_write_access(file);
1781 : 3 : if (error)
1782 : : goto free_vma;
1783 : : }
1784 : 3 : if (vm_flags & VM_SHARED) {
1785 : 3 : error = mapping_map_writable(file->f_mapping);
1786 : 3 : if (error)
1787 : : goto allow_write_and_free_vma;
1788 : : }
1789 : :
1790 : : /* ->mmap() can change vma->vm_file, but must guarantee that
1791 : : * vma_link() below can deny write-access if VM_DENYWRITE is set
1792 : : * and map writably if VM_SHARED is set. This usually means the
1793 : : * new file must not have been exposed to user-space, yet.
1794 : : */
1795 : 3 : vma->vm_file = get_file(file);
1796 : : error = call_mmap(file, vma);
1797 : 3 : if (error)
1798 : : goto unmap_and_free_vma;
1799 : :
1800 : : /* Can addr have changed??
1801 : : *
1802 : : * Answer: Yes, several device drivers can do it in their
1803 : : * f_op->mmap method. -DaveM
1804 : : * Bug: If addr is changed, prev, rb_link, rb_parent should
1805 : : * be updated for vma_link()
1806 : : */
1807 : 3 : WARN_ON_ONCE(addr != vma->vm_start);
1808 : :
1809 : 3 : addr = vma->vm_start;
1810 : 3 : vm_flags = vma->vm_flags;
1811 : 3 : } else if (vm_flags & VM_SHARED) {
1812 : 0 : error = shmem_zero_setup(vma);
1813 : 0 : if (error)
1814 : : goto free_vma;
1815 : : } else {
1816 : : vma_set_anonymous(vma);
1817 : : }
1818 : :
1819 : 3 : vma_link(mm, vma, prev, rb_link, rb_parent);
1820 : : /* Once vma denies write, undo our temporary denial count */
1821 : 3 : if (file) {
1822 : 3 : if (vm_flags & VM_SHARED)
1823 : 3 : mapping_unmap_writable(file->f_mapping);
1824 : 3 : if (vm_flags & VM_DENYWRITE)
1825 : 3 : allow_write_access(file);
1826 : : }
1827 : : file = vma->vm_file;
1828 : : out:
1829 : 3 : perf_event_mmap(vma);
1830 : :
1831 : 3 : vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
1832 : 3 : if (vm_flags & VM_LOCKED) {
1833 : 0 : if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) ||
1834 : 0 : is_vm_hugetlb_page(vma) ||
1835 : 0 : vma == get_gate_vma(current->mm))
1836 : 0 : vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
1837 : : else
1838 : 0 : mm->locked_vm += (len >> PAGE_SHIFT);
1839 : : }
1840 : :
1841 : : if (file)
1842 : : uprobe_mmap(vma);
1843 : :
1844 : : /*
1845 : : * New (or expanded) vma always get soft dirty status.
1846 : : * Otherwise user-space soft-dirty page tracker won't
1847 : : * be able to distinguish situation when vma area unmapped,
1848 : : * then new mapped in-place (which must be aimed as
1849 : : * a completely new data area).
1850 : : */
1851 : : vma->vm_flags |= VM_SOFTDIRTY;
1852 : :
1853 : 3 : vma_set_page_prot(vma);
1854 : :
1855 : 3 : return addr;
1856 : :
1857 : : unmap_and_free_vma:
1858 : 0 : vma->vm_file = NULL;
1859 : 0 : fput(file);
1860 : :
1861 : : /* Undo any partial mapping done by a device driver. */
1862 : 0 : unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
1863 : : charged = 0;
1864 : 0 : if (vm_flags & VM_SHARED)
1865 : 0 : mapping_unmap_writable(file->f_mapping);
1866 : : allow_write_and_free_vma:
1867 : 0 : if (vm_flags & VM_DENYWRITE)
1868 : 0 : allow_write_access(file);
1869 : : free_vma:
1870 : 1 : vm_area_free(vma);
1871 : : unacct_error:
1872 : 0 : if (charged)
1873 : 0 : vm_unacct_memory(charged);
1874 : 0 : return error;
1875 : : }
1876 : :
1877 : 0 : unsigned long unmapped_area(struct vm_unmapped_area_info *info)
1878 : : {
1879 : : /*
1880 : : * We implement the search by looking for an rbtree node that
1881 : : * immediately follows a suitable gap. That is,
1882 : : * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length;
1883 : : * - gap_end = vma->vm_start >= info->low_limit + length;
1884 : : * - gap_end - gap_start >= length
1885 : : */
1886 : :
1887 : 0 : struct mm_struct *mm = current->mm;
1888 : : struct vm_area_struct *vma;
1889 : : unsigned long length, low_limit, high_limit, gap_start, gap_end;
1890 : :
1891 : : /* Adjust search length to account for worst case alignment overhead */
1892 : 0 : length = info->length + info->align_mask;
1893 : 0 : if (length < info->length)
1894 : : return -ENOMEM;
1895 : :
1896 : : /* Adjust search limits by the desired length */
1897 : 0 : if (info->high_limit < length)
1898 : : return -ENOMEM;
1899 : 0 : high_limit = info->high_limit - length;
1900 : :
1901 : 0 : if (info->low_limit > high_limit)
1902 : : return -ENOMEM;
1903 : 0 : low_limit = info->low_limit + length;
1904 : :
1905 : : /* Check if rbtree root looks promising */
1906 : 0 : if (RB_EMPTY_ROOT(&mm->mm_rb))
1907 : : goto check_highest;
1908 : 0 : vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
1909 : 0 : if (vma->rb_subtree_gap < length)
1910 : : goto check_highest;
1911 : :
1912 : : while (true) {
1913 : : /* Visit left subtree if it looks promising */
1914 : : gap_end = vm_start_gap(vma);
1915 : 0 : if (gap_end >= low_limit && vma->vm_rb.rb_left) {
1916 : : struct vm_area_struct *left =
1917 : 0 : rb_entry(vma->vm_rb.rb_left,
1918 : : struct vm_area_struct, vm_rb);
1919 : 0 : if (left->rb_subtree_gap >= length) {
1920 : : vma = left;
1921 : 0 : continue;
1922 : : }
1923 : : }
1924 : :
1925 : 0 : gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
1926 : : check_current:
1927 : : /* Check if current node has a suitable gap */
1928 : 0 : if (gap_start > high_limit)
1929 : : return -ENOMEM;
1930 : 0 : if (gap_end >= low_limit &&
1931 : 0 : gap_end > gap_start && gap_end - gap_start >= length)
1932 : : goto found;
1933 : :
1934 : : /* Visit right subtree if it looks promising */
1935 : 0 : if (vma->vm_rb.rb_right) {
1936 : : struct vm_area_struct *right =
1937 : 0 : rb_entry(vma->vm_rb.rb_right,
1938 : : struct vm_area_struct, vm_rb);
1939 : 0 : if (right->rb_subtree_gap >= length) {
1940 : 0 : vma = right;
1941 : 0 : continue;
1942 : : }
1943 : : }
1944 : :
1945 : : /* Go back up the rbtree to find next candidate node */
1946 : : while (true) {
1947 : 0 : struct rb_node *prev = &vma->vm_rb;
1948 : 0 : if (!rb_parent(prev))
1949 : : goto check_highest;
1950 : 0 : vma = rb_entry(rb_parent(prev),
1951 : : struct vm_area_struct, vm_rb);
1952 : 0 : if (prev == vma->vm_rb.rb_left) {
1953 : 0 : gap_start = vm_end_gap(vma->vm_prev);
1954 : : gap_end = vm_start_gap(vma);
1955 : : goto check_current;
1956 : : }
1957 : : }
1958 : : }
1959 : :
1960 : : check_highest:
1961 : : /* Check highest gap, which does not precede any rbtree node */
1962 : 0 : gap_start = mm->highest_vm_end;
1963 : : gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */
1964 : 0 : if (gap_start > high_limit)
1965 : : return -ENOMEM;
1966 : :
1967 : : found:
1968 : : /* We found a suitable gap. Clip it with the original low_limit. */
1969 : 0 : if (gap_start < info->low_limit)
1970 : : gap_start = info->low_limit;
1971 : :
1972 : : /* Adjust gap address to the desired alignment */
1973 : 0 : gap_start += (info->align_offset - gap_start) & info->align_mask;
1974 : :
1975 : : VM_BUG_ON(gap_start + info->length > info->high_limit);
1976 : : VM_BUG_ON(gap_start + info->length > gap_end);
1977 : 0 : return gap_start;
1978 : : }
1979 : :
1980 : 3 : unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
1981 : : {
1982 : 3 : struct mm_struct *mm = current->mm;
1983 : : struct vm_area_struct *vma;
1984 : : unsigned long length, low_limit, high_limit, gap_start, gap_end;
1985 : :
1986 : : /* Adjust search length to account for worst case alignment overhead */
1987 : 3 : length = info->length + info->align_mask;
1988 : 3 : if (length < info->length)
1989 : : return -ENOMEM;
1990 : :
1991 : : /*
1992 : : * Adjust search limits by the desired length.
1993 : : * See implementation comment at top of unmapped_area().
1994 : : */
1995 : 3 : gap_end = info->high_limit;
1996 : 3 : if (gap_end < length)
1997 : : return -ENOMEM;
1998 : 3 : high_limit = gap_end - length;
1999 : :
2000 : 3 : if (info->low_limit > high_limit)
2001 : : return -ENOMEM;
2002 : 3 : low_limit = info->low_limit + length;
2003 : :
2004 : : /* Check highest gap, which does not precede any rbtree node */
2005 : 3 : gap_start = mm->highest_vm_end;
2006 : 3 : if (gap_start <= high_limit)
2007 : : goto found_highest;
2008 : :
2009 : : /* Check if rbtree root looks promising */
2010 : 3 : if (RB_EMPTY_ROOT(&mm->mm_rb))
2011 : : return -ENOMEM;
2012 : 3 : vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb);
2013 : 3 : if (vma->rb_subtree_gap < length)
2014 : : return -ENOMEM;
2015 : :
2016 : : while (true) {
2017 : : /* Visit right subtree if it looks promising */
2018 : 3 : gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0;
2019 : 3 : if (gap_start <= high_limit && vma->vm_rb.rb_right) {
2020 : : struct vm_area_struct *right =
2021 : 3 : rb_entry(vma->vm_rb.rb_right,
2022 : : struct vm_area_struct, vm_rb);
2023 : 3 : if (right->rb_subtree_gap >= length) {
2024 : : vma = right;
2025 : 3 : continue;
2026 : : }
2027 : : }
2028 : :
2029 : : check_current:
2030 : : /* Check if current node has a suitable gap */
2031 : : gap_end = vm_start_gap(vma);
2032 : 3 : if (gap_end < low_limit)
2033 : : return -ENOMEM;
2034 : 3 : if (gap_start <= high_limit &&
2035 : 3 : gap_end > gap_start && gap_end - gap_start >= length)
2036 : : goto found;
2037 : :
2038 : : /* Visit left subtree if it looks promising */
2039 : 3 : if (vma->vm_rb.rb_left) {
2040 : : struct vm_area_struct *left =
2041 : 3 : rb_entry(vma->vm_rb.rb_left,
2042 : : struct vm_area_struct, vm_rb);
2043 : 3 : if (left->rb_subtree_gap >= length) {
2044 : 3 : vma = left;
2045 : 3 : continue;
2046 : : }
2047 : : }
2048 : :
2049 : : /* Go back up the rbtree to find next candidate node */
2050 : : while (true) {
2051 : 3 : struct rb_node *prev = &vma->vm_rb;
2052 : 3 : if (!rb_parent(prev))
2053 : : return -ENOMEM;
2054 : 3 : vma = rb_entry(rb_parent(prev),
2055 : : struct vm_area_struct, vm_rb);
2056 : 3 : if (prev == vma->vm_rb.rb_right) {
2057 : 3 : gap_start = vma->vm_prev ?
2058 : 3 : vm_end_gap(vma->vm_prev) : 0;
2059 : : goto check_current;
2060 : : }
2061 : : }
2062 : : }
2063 : :
2064 : : found:
2065 : : /* We found a suitable gap. Clip it with the original high_limit. */
2066 : 3 : if (gap_end > info->high_limit)
2067 : : gap_end = info->high_limit;
2068 : :
2069 : : found_highest:
2070 : : /* Compute highest gap address at the desired alignment */
2071 : 3 : gap_end -= info->length;
2072 : 3 : gap_end -= (gap_end - info->align_offset) & info->align_mask;
2073 : :
2074 : : VM_BUG_ON(gap_end < info->low_limit);
2075 : : VM_BUG_ON(gap_end < gap_start);
2076 : 3 : return gap_end;
2077 : : }
2078 : :
2079 : :
2080 : : #ifndef arch_get_mmap_end
2081 : : #define arch_get_mmap_end(addr) (TASK_SIZE)
2082 : : #endif
2083 : :
2084 : : #ifndef arch_get_mmap_base
2085 : : #define arch_get_mmap_base(addr, base) (base)
2086 : : #endif
2087 : :
2088 : : /* Get an address range which is currently unmapped.
2089 : : * For shmat() with addr=0.
2090 : : *
2091 : : * Ugly calling convention alert:
2092 : : * Return value with the low bits set means error value,
2093 : : * ie
2094 : : * if (ret & ~PAGE_MASK)
2095 : : * error = ret;
2096 : : *
2097 : : * This function "knows" that -ENOMEM has the bits set.
2098 : : */
2099 : : #ifndef HAVE_ARCH_UNMAPPED_AREA
2100 : : unsigned long
2101 : : arch_get_unmapped_area(struct file *filp, unsigned long addr,
2102 : : unsigned long len, unsigned long pgoff, unsigned long flags)
2103 : : {
2104 : : struct mm_struct *mm = current->mm;
2105 : : struct vm_area_struct *vma, *prev;
2106 : : struct vm_unmapped_area_info info;
2107 : : const unsigned long mmap_end = arch_get_mmap_end(addr);
2108 : :
2109 : : if (len > mmap_end - mmap_min_addr)
2110 : : return -ENOMEM;
2111 : :
2112 : : if (flags & MAP_FIXED)
2113 : : return addr;
2114 : :
2115 : : if (addr) {
2116 : : addr = PAGE_ALIGN(addr);
2117 : : vma = find_vma_prev(mm, addr, &prev);
2118 : : if (mmap_end - len >= addr && addr >= mmap_min_addr &&
2119 : : (!vma || addr + len <= vm_start_gap(vma)) &&
2120 : : (!prev || addr >= vm_end_gap(prev)))
2121 : : return addr;
2122 : : }
2123 : :
2124 : : info.flags = 0;
2125 : : info.length = len;
2126 : : info.low_limit = mm->mmap_base;
2127 : : info.high_limit = mmap_end;
2128 : : info.align_mask = 0;
2129 : : return vm_unmapped_area(&info);
2130 : : }
2131 : : #endif
2132 : :
2133 : : /*
2134 : : * This mmap-allocator allocates new areas top-down from below the
2135 : : * stack's low limit (the base):
2136 : : */
2137 : : #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
2138 : : unsigned long
2139 : : arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr,
2140 : : unsigned long len, unsigned long pgoff,
2141 : : unsigned long flags)
2142 : : {
2143 : : struct vm_area_struct *vma, *prev;
2144 : : struct mm_struct *mm = current->mm;
2145 : : struct vm_unmapped_area_info info;
2146 : : const unsigned long mmap_end = arch_get_mmap_end(addr);
2147 : :
2148 : : /* requested length too big for entire address space */
2149 : : if (len > mmap_end - mmap_min_addr)
2150 : : return -ENOMEM;
2151 : :
2152 : : if (flags & MAP_FIXED)
2153 : : return addr;
2154 : :
2155 : : /* requesting a specific address */
2156 : : if (addr) {
2157 : : addr = PAGE_ALIGN(addr);
2158 : : vma = find_vma_prev(mm, addr, &prev);
2159 : : if (mmap_end - len >= addr && addr >= mmap_min_addr &&
2160 : : (!vma || addr + len <= vm_start_gap(vma)) &&
2161 : : (!prev || addr >= vm_end_gap(prev)))
2162 : : return addr;
2163 : : }
2164 : :
2165 : : info.flags = VM_UNMAPPED_AREA_TOPDOWN;
2166 : : info.length = len;
2167 : : info.low_limit = max(PAGE_SIZE, mmap_min_addr);
2168 : : info.high_limit = arch_get_mmap_base(addr, mm->mmap_base);
2169 : : info.align_mask = 0;
2170 : : addr = vm_unmapped_area(&info);
2171 : :
2172 : : /*
2173 : : * A failed mmap() very likely causes application failure,
2174 : : * so fall back to the bottom-up function here. This scenario
2175 : : * can happen with large stack limits and large mmap()
2176 : : * allocations.
2177 : : */
2178 : : if (offset_in_page(addr)) {
2179 : : VM_BUG_ON(addr != -ENOMEM);
2180 : : info.flags = 0;
2181 : : info.low_limit = TASK_UNMAPPED_BASE;
2182 : : info.high_limit = mmap_end;
2183 : : addr = vm_unmapped_area(&info);
2184 : : }
2185 : :
2186 : : return addr;
2187 : : }
2188 : : #endif
2189 : :
2190 : : unsigned long
2191 : 3 : get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
2192 : : unsigned long pgoff, unsigned long flags)
2193 : : {
2194 : : unsigned long (*get_area)(struct file *, unsigned long,
2195 : : unsigned long, unsigned long, unsigned long);
2196 : :
2197 : 3 : unsigned long error = arch_mmap_check(addr, len, flags);
2198 : 3 : if (error)
2199 : : return error;
2200 : :
2201 : : /* Careful about overflows.. */
2202 : 3 : if (len > TASK_SIZE)
2203 : : return -ENOMEM;
2204 : :
2205 : 3 : get_area = current->mm->get_unmapped_area;
2206 : 3 : if (file) {
2207 : 3 : if (file->f_op->get_unmapped_area)
2208 : : get_area = file->f_op->get_unmapped_area;
2209 : 3 : } else if (flags & MAP_SHARED) {
2210 : : /*
2211 : : * mmap_region() will call shmem_zero_setup() to create a file,
2212 : : * so use shmem's get_unmapped_area in case it can be huge.
2213 : : * do_mmap_pgoff() will clear pgoff, so match alignment.
2214 : : */
2215 : : pgoff = 0;
2216 : : get_area = shmem_get_unmapped_area;
2217 : : }
2218 : :
2219 : 3 : addr = get_area(file, addr, len, pgoff, flags);
2220 : 3 : if (IS_ERR_VALUE(addr))
2221 : : return addr;
2222 : :
2223 : 3 : if (addr > TASK_SIZE - len)
2224 : : return -ENOMEM;
2225 : 3 : if (offset_in_page(addr))
2226 : : return -EINVAL;
2227 : :
2228 : 3 : error = security_mmap_addr(addr);
2229 : 3 : return error ? error : addr;
2230 : : }
2231 : :
2232 : : EXPORT_SYMBOL(get_unmapped_area);
2233 : :
2234 : : /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */
2235 : 3 : struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
2236 : : {
2237 : : struct rb_node *rb_node;
2238 : : struct vm_area_struct *vma;
2239 : :
2240 : : /* Check the cache first. */
2241 : 3 : vma = vmacache_find(mm, addr);
2242 : 3 : if (likely(vma))
2243 : : return vma;
2244 : :
2245 : 3 : rb_node = mm->mm_rb.rb_node;
2246 : :
2247 : 3 : while (rb_node) {
2248 : : struct vm_area_struct *tmp;
2249 : :
2250 : 3 : tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb);
2251 : :
2252 : 3 : if (tmp->vm_end > addr) {
2253 : : vma = tmp;
2254 : 3 : if (tmp->vm_start <= addr)
2255 : : break;
2256 : 3 : rb_node = rb_node->rb_left;
2257 : : } else
2258 : 3 : rb_node = rb_node->rb_right;
2259 : : }
2260 : :
2261 : 3 : if (vma)
2262 : 3 : vmacache_update(addr, vma);
2263 : 3 : return vma;
2264 : : }
2265 : :
2266 : : EXPORT_SYMBOL(find_vma);
2267 : :
2268 : : /*
2269 : : * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
2270 : : */
2271 : : struct vm_area_struct *
2272 : 3 : find_vma_prev(struct mm_struct *mm, unsigned long addr,
2273 : : struct vm_area_struct **pprev)
2274 : : {
2275 : : struct vm_area_struct *vma;
2276 : :
2277 : 3 : vma = find_vma(mm, addr);
2278 : 3 : if (vma) {
2279 : 3 : *pprev = vma->vm_prev;
2280 : : } else {
2281 : 0 : struct rb_node *rb_node = rb_last(&mm->mm_rb);
2282 : :
2283 : 0 : *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL;
2284 : : }
2285 : 3 : return vma;
2286 : : }
2287 : :
2288 : : /*
2289 : : * Verify that the stack growth is acceptable and
2290 : : * update accounting. This is shared with both the
2291 : : * grow-up and grow-down cases.
2292 : : */
2293 : 3 : static int acct_stack_growth(struct vm_area_struct *vma,
2294 : : unsigned long size, unsigned long grow)
2295 : : {
2296 : 3 : struct mm_struct *mm = vma->vm_mm;
2297 : : unsigned long new_start;
2298 : :
2299 : : /* address space limit tests */
2300 : 3 : if (!may_expand_vm(mm, vma->vm_flags, grow))
2301 : : return -ENOMEM;
2302 : :
2303 : : /* Stack limit test */
2304 : 3 : if (size > rlimit(RLIMIT_STACK))
2305 : : return -ENOMEM;
2306 : :
2307 : : /* mlock limit tests */
2308 : 3 : if (vma->vm_flags & VM_LOCKED) {
2309 : : unsigned long locked;
2310 : : unsigned long limit;
2311 : 0 : locked = mm->locked_vm + grow;
2312 : : limit = rlimit(RLIMIT_MEMLOCK);
2313 : 0 : limit >>= PAGE_SHIFT;
2314 : 0 : if (locked > limit && !capable(CAP_IPC_LOCK))
2315 : : return -ENOMEM;
2316 : : }
2317 : :
2318 : : /* Check to ensure the stack will not grow into a hugetlb-only region */
2319 : : new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start :
2320 : : vma->vm_end - size;
2321 : : if (is_hugepage_only_range(vma->vm_mm, new_start, size))
2322 : : return -EFAULT;
2323 : :
2324 : : /*
2325 : : * Overcommit.. This must be the final test, as it will
2326 : : * update security statistics.
2327 : : */
2328 : 3 : if (security_vm_enough_memory_mm(mm, grow))
2329 : : return -ENOMEM;
2330 : :
2331 : 3 : return 0;
2332 : : }
2333 : :
2334 : : #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64)
2335 : : /*
2336 : : * PA-RISC uses this for its stack; IA64 for its Register Backing Store.
2337 : : * vma is the last one with address > vma->vm_end. Have to extend vma.
2338 : : */
2339 : : int expand_upwards(struct vm_area_struct *vma, unsigned long address)
2340 : : {
2341 : : struct mm_struct *mm = vma->vm_mm;
2342 : : struct vm_area_struct *next;
2343 : : unsigned long gap_addr;
2344 : : int error = 0;
2345 : :
2346 : : if (!(vma->vm_flags & VM_GROWSUP))
2347 : : return -EFAULT;
2348 : :
2349 : : /* Guard against exceeding limits of the address space. */
2350 : : address &= PAGE_MASK;
2351 : : if (address >= (TASK_SIZE & PAGE_MASK))
2352 : : return -ENOMEM;
2353 : : address += PAGE_SIZE;
2354 : :
2355 : : /* Enforce stack_guard_gap */
2356 : : gap_addr = address + stack_guard_gap;
2357 : :
2358 : : /* Guard against overflow */
2359 : : if (gap_addr < address || gap_addr > TASK_SIZE)
2360 : : gap_addr = TASK_SIZE;
2361 : :
2362 : : next = vma->vm_next;
2363 : : if (next && next->vm_start < gap_addr &&
2364 : : (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
2365 : : if (!(next->vm_flags & VM_GROWSUP))
2366 : : return -ENOMEM;
2367 : : /* Check that both stack segments have the same anon_vma? */
2368 : : }
2369 : :
2370 : : /* We must make sure the anon_vma is allocated. */
2371 : : if (unlikely(anon_vma_prepare(vma)))
2372 : : return -ENOMEM;
2373 : :
2374 : : /*
2375 : : * vma->vm_start/vm_end cannot change under us because the caller
2376 : : * is required to hold the mmap_sem in read mode. We need the
2377 : : * anon_vma lock to serialize against concurrent expand_stacks.
2378 : : */
2379 : : anon_vma_lock_write(vma->anon_vma);
2380 : :
2381 : : /* Somebody else might have raced and expanded it already */
2382 : : if (address > vma->vm_end) {
2383 : : unsigned long size, grow;
2384 : :
2385 : : size = address - vma->vm_start;
2386 : : grow = (address - vma->vm_end) >> PAGE_SHIFT;
2387 : :
2388 : : error = -ENOMEM;
2389 : : if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) {
2390 : : error = acct_stack_growth(vma, size, grow);
2391 : : if (!error) {
2392 : : /*
2393 : : * vma_gap_update() doesn't support concurrent
2394 : : * updates, but we only hold a shared mmap_sem
2395 : : * lock here, so we need to protect against
2396 : : * concurrent vma expansions.
2397 : : * anon_vma_lock_write() doesn't help here, as
2398 : : * we don't guarantee that all growable vmas
2399 : : * in a mm share the same root anon vma.
2400 : : * So, we reuse mm->page_table_lock to guard
2401 : : * against concurrent vma expansions.
2402 : : */
2403 : : spin_lock(&mm->page_table_lock);
2404 : : if (vma->vm_flags & VM_LOCKED)
2405 : : mm->locked_vm += grow;
2406 : : vm_stat_account(mm, vma->vm_flags, grow);
2407 : : anon_vma_interval_tree_pre_update_vma(vma);
2408 : : vma->vm_end = address;
2409 : : anon_vma_interval_tree_post_update_vma(vma);
2410 : : if (vma->vm_next)
2411 : : vma_gap_update(vma->vm_next);
2412 : : else
2413 : : mm->highest_vm_end = vm_end_gap(vma);
2414 : : spin_unlock(&mm->page_table_lock);
2415 : :
2416 : : perf_event_mmap(vma);
2417 : : }
2418 : : }
2419 : : }
2420 : : anon_vma_unlock_write(vma->anon_vma);
2421 : : khugepaged_enter_vma_merge(vma, vma->vm_flags);
2422 : : validate_mm(mm);
2423 : : return error;
2424 : : }
2425 : : #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */
2426 : :
2427 : : /*
2428 : : * vma is the first one with address < vma->vm_start. Have to extend vma.
2429 : : */
2430 : 3 : int expand_downwards(struct vm_area_struct *vma,
2431 : : unsigned long address)
2432 : : {
2433 : 3 : struct mm_struct *mm = vma->vm_mm;
2434 : : struct vm_area_struct *prev;
2435 : : int error = 0;
2436 : :
2437 : 3 : address &= PAGE_MASK;
2438 : 3 : if (address < mmap_min_addr)
2439 : : return -EPERM;
2440 : :
2441 : : /* Enforce stack_guard_gap */
2442 : 3 : prev = vma->vm_prev;
2443 : : /* Check that both stack segments have the same anon_vma? */
2444 : 3 : if (prev && !(prev->vm_flags & VM_GROWSDOWN) &&
2445 : 0 : (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) {
2446 : 0 : if (address - prev->vm_end < stack_guard_gap)
2447 : : return -ENOMEM;
2448 : : }
2449 : :
2450 : : /* We must make sure the anon_vma is allocated. */
2451 : 3 : if (unlikely(anon_vma_prepare(vma)))
2452 : : return -ENOMEM;
2453 : :
2454 : : /*
2455 : : * vma->vm_start/vm_end cannot change under us because the caller
2456 : : * is required to hold the mmap_sem in read mode. We need the
2457 : : * anon_vma lock to serialize against concurrent expand_stacks.
2458 : : */
2459 : 3 : anon_vma_lock_write(vma->anon_vma);
2460 : :
2461 : : /* Somebody else might have raced and expanded it already */
2462 : 3 : if (address < vma->vm_start) {
2463 : : unsigned long size, grow;
2464 : :
2465 : 3 : size = vma->vm_end - address;
2466 : 3 : grow = (vma->vm_start - address) >> PAGE_SHIFT;
2467 : :
2468 : : error = -ENOMEM;
2469 : 3 : if (grow <= vma->vm_pgoff) {
2470 : 3 : error = acct_stack_growth(vma, size, grow);
2471 : 3 : if (!error) {
2472 : : /*
2473 : : * vma_gap_update() doesn't support concurrent
2474 : : * updates, but we only hold a shared mmap_sem
2475 : : * lock here, so we need to protect against
2476 : : * concurrent vma expansions.
2477 : : * anon_vma_lock_write() doesn't help here, as
2478 : : * we don't guarantee that all growable vmas
2479 : : * in a mm share the same root anon vma.
2480 : : * So, we reuse mm->page_table_lock to guard
2481 : : * against concurrent vma expansions.
2482 : : */
2483 : : spin_lock(&mm->page_table_lock);
2484 : 3 : if (vma->vm_flags & VM_LOCKED)
2485 : 0 : mm->locked_vm += grow;
2486 : 3 : vm_stat_account(mm, vma->vm_flags, grow);
2487 : 3 : anon_vma_interval_tree_pre_update_vma(vma);
2488 : 3 : vma->vm_start = address;
2489 : 3 : vma->vm_pgoff -= grow;
2490 : 3 : anon_vma_interval_tree_post_update_vma(vma);
2491 : : vma_gap_update(vma);
2492 : : spin_unlock(&mm->page_table_lock);
2493 : :
2494 : 3 : perf_event_mmap(vma);
2495 : : }
2496 : : }
2497 : : }
2498 : 3 : anon_vma_unlock_write(vma->anon_vma);
2499 : : khugepaged_enter_vma_merge(vma, vma->vm_flags);
2500 : : validate_mm(mm);
2501 : 3 : return error;
2502 : : }
2503 : :
2504 : : /* enforced gap between the expanding stack and other mappings. */
2505 : : unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT;
2506 : :
2507 : 0 : static int __init cmdline_parse_stack_guard_gap(char *p)
2508 : : {
2509 : : unsigned long val;
2510 : : char *endptr;
2511 : :
2512 : 0 : val = simple_strtoul(p, &endptr, 10);
2513 : 0 : if (!*endptr)
2514 : 0 : stack_guard_gap = val << PAGE_SHIFT;
2515 : :
2516 : 0 : return 0;
2517 : : }
2518 : : __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap);
2519 : :
2520 : : #ifdef CONFIG_STACK_GROWSUP
2521 : : int expand_stack(struct vm_area_struct *vma, unsigned long address)
2522 : : {
2523 : : return expand_upwards(vma, address);
2524 : : }
2525 : :
2526 : : struct vm_area_struct *
2527 : : find_extend_vma(struct mm_struct *mm, unsigned long addr)
2528 : : {
2529 : : struct vm_area_struct *vma, *prev;
2530 : :
2531 : : addr &= PAGE_MASK;
2532 : : vma = find_vma_prev(mm, addr, &prev);
2533 : : if (vma && (vma->vm_start <= addr))
2534 : : return vma;
2535 : : /* don't alter vm_end if the coredump is running */
2536 : : if (!prev || !mmget_still_valid(mm) || expand_stack(prev, addr))
2537 : : return NULL;
2538 : : if (prev->vm_flags & VM_LOCKED)
2539 : : populate_vma_page_range(prev, addr, prev->vm_end, NULL);
2540 : : return prev;
2541 : : }
2542 : : #else
2543 : 3 : int expand_stack(struct vm_area_struct *vma, unsigned long address)
2544 : : {
2545 : 3 : return expand_downwards(vma, address);
2546 : : }
2547 : :
2548 : : struct vm_area_struct *
2549 : 3 : find_extend_vma(struct mm_struct *mm, unsigned long addr)
2550 : : {
2551 : : struct vm_area_struct *vma;
2552 : : unsigned long start;
2553 : :
2554 : 3 : addr &= PAGE_MASK;
2555 : 3 : vma = find_vma(mm, addr);
2556 : 3 : if (!vma)
2557 : : return NULL;
2558 : 3 : if (vma->vm_start <= addr)
2559 : : return vma;
2560 : 3 : if (!(vma->vm_flags & VM_GROWSDOWN))
2561 : : return NULL;
2562 : : /* don't alter vm_start if the coredump is running */
2563 : 3 : if (!mmget_still_valid(mm))
2564 : : return NULL;
2565 : : start = vma->vm_start;
2566 : 3 : if (expand_stack(vma, addr))
2567 : : return NULL;
2568 : 3 : if (vma->vm_flags & VM_LOCKED)
2569 : 0 : populate_vma_page_range(vma, addr, start, NULL);
2570 : 3 : return vma;
2571 : : }
2572 : : #endif
2573 : :
2574 : : EXPORT_SYMBOL_GPL(find_extend_vma);
2575 : :
2576 : : /*
2577 : : * Ok - we have the memory areas we should free on the vma list,
2578 : : * so release them, and do the vma updates.
2579 : : *
2580 : : * Called with the mm semaphore held.
2581 : : */
2582 : 3 : static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma)
2583 : : {
2584 : : unsigned long nr_accounted = 0;
2585 : :
2586 : : /* Update high watermark before we lower total_vm */
2587 : : update_hiwater_vm(mm);
2588 : : do {
2589 : 3 : long nrpages = vma_pages(vma);
2590 : :
2591 : 3 : if (vma->vm_flags & VM_ACCOUNT)
2592 : 3 : nr_accounted += nrpages;
2593 : 3 : vm_stat_account(mm, vma->vm_flags, -nrpages);
2594 : 3 : vma = remove_vma(vma);
2595 : 3 : } while (vma);
2596 : 3 : vm_unacct_memory(nr_accounted);
2597 : : validate_mm(mm);
2598 : 3 : }
2599 : :
2600 : : /*
2601 : : * Get rid of page table information in the indicated region.
2602 : : *
2603 : : * Called with the mm semaphore held.
2604 : : */
2605 : 3 : static void unmap_region(struct mm_struct *mm,
2606 : : struct vm_area_struct *vma, struct vm_area_struct *prev,
2607 : : unsigned long start, unsigned long end)
2608 : : {
2609 : 3 : struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap;
2610 : : struct mmu_gather tlb;
2611 : :
2612 : 3 : lru_add_drain();
2613 : 3 : tlb_gather_mmu(&tlb, mm, start, end);
2614 : : update_hiwater_rss(mm);
2615 : 3 : unmap_vmas(&tlb, vma, start, end);
2616 : 3 : free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS,
2617 : : next ? next->vm_start : USER_PGTABLES_CEILING);
2618 : 3 : tlb_finish_mmu(&tlb, start, end);
2619 : 3 : }
2620 : :
2621 : : /*
2622 : : * Create a list of vma's touched by the unmap, removing them from the mm's
2623 : : * vma list as we go..
2624 : : */
2625 : : static bool
2626 : 3 : detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
2627 : : struct vm_area_struct *prev, unsigned long end)
2628 : : {
2629 : : struct vm_area_struct **insertion_point;
2630 : : struct vm_area_struct *tail_vma = NULL;
2631 : :
2632 : 3 : insertion_point = (prev ? &prev->vm_next : &mm->mmap);
2633 : 3 : vma->vm_prev = NULL;
2634 : : do {
2635 : 3 : vma_rb_erase(vma, &mm->mm_rb);
2636 : 3 : mm->map_count--;
2637 : : tail_vma = vma;
2638 : 3 : vma = vma->vm_next;
2639 : 3 : } while (vma && vma->vm_start < end);
2640 : 3 : *insertion_point = vma;
2641 : 3 : if (vma) {
2642 : 3 : vma->vm_prev = prev;
2643 : : vma_gap_update(vma);
2644 : : } else
2645 : 0 : mm->highest_vm_end = prev ? vm_end_gap(prev) : 0;
2646 : 3 : tail_vma->vm_next = NULL;
2647 : :
2648 : : /* Kill the cache */
2649 : : vmacache_invalidate(mm);
2650 : :
2651 : : /*
2652 : : * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or
2653 : : * VM_GROWSUP VMA. Such VMAs can change their size under
2654 : : * down_read(mmap_lock) and collide with the VMA we are about to unmap.
2655 : : */
2656 : 3 : if (vma && (vma->vm_flags & VM_GROWSDOWN))
2657 : : return false;
2658 : : if (prev && (prev->vm_flags & VM_GROWSUP))
2659 : : return false;
2660 : 3 : return true;
2661 : : }
2662 : :
2663 : : /*
2664 : : * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
2665 : : * has already been checked or doesn't make sense to fail.
2666 : : */
2667 : 3 : int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2668 : : unsigned long addr, int new_below)
2669 : : {
2670 : : struct vm_area_struct *new;
2671 : : int err;
2672 : :
2673 : 3 : if (vma->vm_ops && vma->vm_ops->split) {
2674 : 0 : err = vma->vm_ops->split(vma, addr);
2675 : 0 : if (err)
2676 : : return err;
2677 : : }
2678 : :
2679 : 3 : new = vm_area_dup(vma);
2680 : 3 : if (!new)
2681 : : return -ENOMEM;
2682 : :
2683 : 3 : if (new_below)
2684 : 3 : new->vm_end = addr;
2685 : : else {
2686 : 3 : new->vm_start = addr;
2687 : 3 : new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT);
2688 : : }
2689 : :
2690 : : err = vma_dup_policy(vma, new);
2691 : : if (err)
2692 : : goto out_free_vma;
2693 : :
2694 : 3 : err = anon_vma_clone(new, vma);
2695 : 3 : if (err)
2696 : : goto out_free_mpol;
2697 : :
2698 : 3 : if (new->vm_file)
2699 : : get_file(new->vm_file);
2700 : :
2701 : 3 : if (new->vm_ops && new->vm_ops->open)
2702 : 0 : new->vm_ops->open(new);
2703 : :
2704 : 3 : if (new_below)
2705 : 3 : err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff +
2706 : 3 : ((addr - new->vm_start) >> PAGE_SHIFT), new);
2707 : : else
2708 : 3 : err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new);
2709 : :
2710 : : /* Success. */
2711 : 3 : if (!err)
2712 : : return 0;
2713 : :
2714 : : /* Clean everything up if vma_adjust failed. */
2715 : 0 : if (new->vm_ops && new->vm_ops->close)
2716 : 0 : new->vm_ops->close(new);
2717 : 0 : if (new->vm_file)
2718 : 0 : fput(new->vm_file);
2719 : 0 : unlink_anon_vmas(new);
2720 : : out_free_mpol:
2721 : : mpol_put(vma_policy(new));
2722 : : out_free_vma:
2723 : 3 : vm_area_free(new);
2724 : 0 : return err;
2725 : : }
2726 : :
2727 : : /*
2728 : : * Split a vma into two pieces at address 'addr', a new vma is allocated
2729 : : * either for the first part or the tail.
2730 : : */
2731 : 3 : int split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
2732 : : unsigned long addr, int new_below)
2733 : : {
2734 : 3 : if (mm->map_count >= sysctl_max_map_count)
2735 : : return -ENOMEM;
2736 : :
2737 : 3 : return __split_vma(mm, vma, addr, new_below);
2738 : : }
2739 : :
2740 : : /* Munmap is split into 2 main parts -- this part which finds
2741 : : * what needs doing, and the areas themselves, which do the
2742 : : * work. This now handles partial unmappings.
2743 : : * Jeremy Fitzhardinge <jeremy@goop.org>
2744 : : */
2745 : 3 : int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2746 : : struct list_head *uf, bool downgrade)
2747 : : {
2748 : : unsigned long end;
2749 : : struct vm_area_struct *vma, *prev, *last;
2750 : :
2751 : 3 : if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
2752 : : return -EINVAL;
2753 : :
2754 : 3 : len = PAGE_ALIGN(len);
2755 : 3 : end = start + len;
2756 : 3 : if (len == 0)
2757 : : return -EINVAL;
2758 : :
2759 : : /*
2760 : : * arch_unmap() might do unmaps itself. It must be called
2761 : : * and finish any rbtree manipulation before this code
2762 : : * runs and also starts to manipulate the rbtree.
2763 : : */
2764 : : arch_unmap(mm, start, end);
2765 : :
2766 : : /* Find the first overlapping VMA */
2767 : 3 : vma = find_vma(mm, start);
2768 : 3 : if (!vma)
2769 : : return 0;
2770 : 3 : prev = vma->vm_prev;
2771 : : /* we have start < vma->vm_end */
2772 : :
2773 : : /* if it doesn't overlap, we have nothing.. */
2774 : 3 : if (vma->vm_start >= end)
2775 : : return 0;
2776 : :
2777 : : /*
2778 : : * If we need to split any vma, do it now to save pain later.
2779 : : *
2780 : : * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially
2781 : : * unmapped vm_area_struct will remain in use: so lower split_vma
2782 : : * places tmp vma above, and higher split_vma places tmp vma below.
2783 : : */
2784 : 3 : if (start > vma->vm_start) {
2785 : : int error;
2786 : :
2787 : : /*
2788 : : * Make sure that map_count on return from munmap() will
2789 : : * not exceed its limit; but let map_count go just above
2790 : : * its limit temporarily, to help free resources as expected.
2791 : : */
2792 : 3 : if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count)
2793 : : return -ENOMEM;
2794 : :
2795 : 3 : error = __split_vma(mm, vma, start, 0);
2796 : 3 : if (error)
2797 : : return error;
2798 : : prev = vma;
2799 : : }
2800 : :
2801 : : /* Does it split the last one? */
2802 : 3 : last = find_vma(mm, end);
2803 : 3 : if (last && end > last->vm_start) {
2804 : 3 : int error = __split_vma(mm, last, end, 1);
2805 : 3 : if (error)
2806 : : return error;
2807 : : }
2808 : 3 : vma = prev ? prev->vm_next : mm->mmap;
2809 : :
2810 : : if (unlikely(uf)) {
2811 : : /*
2812 : : * If userfaultfd_unmap_prep returns an error the vmas
2813 : : * will remain splitted, but userland will get a
2814 : : * highly unexpected error anyway. This is no
2815 : : * different than the case where the first of the two
2816 : : * __split_vma fails, but we don't undo the first
2817 : : * split, despite we could. This is unlikely enough
2818 : : * failure that it's not worth optimizing it for.
2819 : : */
2820 : : int error = userfaultfd_unmap_prep(vma, start, end, uf);
2821 : : if (error)
2822 : : return error;
2823 : : }
2824 : :
2825 : : /*
2826 : : * unlock any mlock()ed ranges before detaching vmas
2827 : : */
2828 : 3 : if (mm->locked_vm) {
2829 : : struct vm_area_struct *tmp = vma;
2830 : 3 : while (tmp && tmp->vm_start < end) {
2831 : 3 : if (tmp->vm_flags & VM_LOCKED) {
2832 : 3 : mm->locked_vm -= vma_pages(tmp);
2833 : : munlock_vma_pages_all(tmp);
2834 : : }
2835 : :
2836 : 3 : tmp = tmp->vm_next;
2837 : : }
2838 : : }
2839 : :
2840 : : /* Detach vmas from rbtree */
2841 : 3 : if (!detach_vmas_to_be_unmapped(mm, vma, prev, end))
2842 : : downgrade = false;
2843 : :
2844 : 3 : if (downgrade)
2845 : 3 : downgrade_write(&mm->mmap_sem);
2846 : :
2847 : 3 : unmap_region(mm, vma, prev, start, end);
2848 : :
2849 : : /* Fix up all other VM information */
2850 : 3 : remove_vma_list(mm, vma);
2851 : :
2852 : 3 : return downgrade ? 1 : 0;
2853 : : }
2854 : :
2855 : 3 : int do_munmap(struct mm_struct *mm, unsigned long start, size_t len,
2856 : : struct list_head *uf)
2857 : : {
2858 : 3 : return __do_munmap(mm, start, len, uf, false);
2859 : : }
2860 : :
2861 : 3 : static int __vm_munmap(unsigned long start, size_t len, bool downgrade)
2862 : : {
2863 : : int ret;
2864 : 3 : struct mm_struct *mm = current->mm;
2865 : 3 : LIST_HEAD(uf);
2866 : :
2867 : 3 : if (down_write_killable(&mm->mmap_sem))
2868 : : return -EINTR;
2869 : :
2870 : 3 : ret = __do_munmap(mm, start, len, &uf, downgrade);
2871 : : /*
2872 : : * Returning 1 indicates mmap_sem is downgraded.
2873 : : * But 1 is not legal return value of vm_munmap() and munmap(), reset
2874 : : * it to 0 before return.
2875 : : */
2876 : 3 : if (ret == 1) {
2877 : 3 : up_read(&mm->mmap_sem);
2878 : : ret = 0;
2879 : : } else
2880 : 3 : up_write(&mm->mmap_sem);
2881 : :
2882 : : userfaultfd_unmap_complete(mm, &uf);
2883 : 3 : return ret;
2884 : : }
2885 : :
2886 : 3 : int vm_munmap(unsigned long start, size_t len)
2887 : : {
2888 : 3 : return __vm_munmap(start, len, false);
2889 : : }
2890 : : EXPORT_SYMBOL(vm_munmap);
2891 : :
2892 : 3 : SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len)
2893 : : {
2894 : : addr = untagged_addr(addr);
2895 : 3 : profile_munmap(addr);
2896 : 3 : return __vm_munmap(addr, len, true);
2897 : : }
2898 : :
2899 : :
2900 : : /*
2901 : : * Emulation of deprecated remap_file_pages() syscall.
2902 : : */
2903 : 0 : SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
2904 : : unsigned long, prot, unsigned long, pgoff, unsigned long, flags)
2905 : : {
2906 : :
2907 : 0 : struct mm_struct *mm = current->mm;
2908 : : struct vm_area_struct *vma;
2909 : 0 : unsigned long populate = 0;
2910 : : unsigned long ret = -EINVAL;
2911 : : struct file *file;
2912 : :
2913 : 0 : pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n",
2914 : : current->comm, current->pid);
2915 : :
2916 : 0 : if (prot)
2917 : : return ret;
2918 : 0 : start = start & PAGE_MASK;
2919 : 0 : size = size & PAGE_MASK;
2920 : :
2921 : 0 : if (start + size <= start)
2922 : : return ret;
2923 : :
2924 : : /* Does pgoff wrap? */
2925 : 0 : if (pgoff + (size >> PAGE_SHIFT) < pgoff)
2926 : : return ret;
2927 : :
2928 : 0 : if (down_write_killable(&mm->mmap_sem))
2929 : : return -EINTR;
2930 : :
2931 : 0 : vma = find_vma(mm, start);
2932 : :
2933 : 0 : if (!vma || !(vma->vm_flags & VM_SHARED))
2934 : : goto out;
2935 : :
2936 : 0 : if (start < vma->vm_start)
2937 : : goto out;
2938 : :
2939 : 0 : if (start + size > vma->vm_end) {
2940 : : struct vm_area_struct *next;
2941 : :
2942 : 0 : for (next = vma->vm_next; next; next = next->vm_next) {
2943 : : /* hole between vmas ? */
2944 : 0 : if (next->vm_start != next->vm_prev->vm_end)
2945 : : goto out;
2946 : :
2947 : 0 : if (next->vm_file != vma->vm_file)
2948 : : goto out;
2949 : :
2950 : 0 : if (next->vm_flags != vma->vm_flags)
2951 : : goto out;
2952 : :
2953 : 0 : if (start + size <= next->vm_end)
2954 : : break;
2955 : : }
2956 : :
2957 : 0 : if (!next)
2958 : : goto out;
2959 : : }
2960 : :
2961 : 0 : prot |= vma->vm_flags & VM_READ ? PROT_READ : 0;
2962 : 0 : prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0;
2963 : 0 : prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0;
2964 : :
2965 : 0 : flags &= MAP_NONBLOCK;
2966 : 0 : flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE;
2967 : 0 : if (vma->vm_flags & VM_LOCKED) {
2968 : : struct vm_area_struct *tmp;
2969 : 0 : flags |= MAP_LOCKED;
2970 : :
2971 : : /* drop PG_Mlocked flag for over-mapped range */
2972 : 0 : for (tmp = vma; tmp->vm_start >= start + size;
2973 : 0 : tmp = tmp->vm_next) {
2974 : : /*
2975 : : * Split pmd and munlock page on the border
2976 : : * of the range.
2977 : : */
2978 : : vma_adjust_trans_huge(tmp, start, start + size, 0);
2979 : :
2980 : 0 : munlock_vma_pages_range(tmp,
2981 : 0 : max(tmp->vm_start, start),
2982 : 0 : min(tmp->vm_end, start + size));
2983 : : }
2984 : : }
2985 : :
2986 : 0 : file = get_file(vma->vm_file);
2987 : 0 : ret = do_mmap_pgoff(vma->vm_file, start, size,
2988 : : prot, flags, pgoff, &populate, NULL);
2989 : 0 : fput(file);
2990 : : out:
2991 : 0 : up_write(&mm->mmap_sem);
2992 : 0 : if (populate)
2993 : : mm_populate(ret, populate);
2994 : 0 : if (!IS_ERR_VALUE(ret))
2995 : : ret = 0;
2996 : 0 : return ret;
2997 : : }
2998 : :
2999 : : /*
3000 : : * this is really a simplified "do_mmap". it only handles
3001 : : * anonymous maps. eventually we may be able to do some
3002 : : * brk-specific accounting here.
3003 : : */
3004 : 3 : static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf)
3005 : : {
3006 : 3 : struct mm_struct *mm = current->mm;
3007 : : struct vm_area_struct *vma, *prev;
3008 : : struct rb_node **rb_link, *rb_parent;
3009 : 3 : pgoff_t pgoff = addr >> PAGE_SHIFT;
3010 : : int error;
3011 : :
3012 : : /* Until we need other flags, refuse anything except VM_EXEC. */
3013 : 3 : if ((flags & (~VM_EXEC)) != 0)
3014 : : return -EINVAL;
3015 : 3 : flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags;
3016 : :
3017 : 3 : error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED);
3018 : 3 : if (offset_in_page(error))
3019 : : return error;
3020 : :
3021 : 3 : error = mlock_future_check(mm, mm->def_flags, len);
3022 : 3 : if (error)
3023 : : return error;
3024 : :
3025 : : /*
3026 : : * Clear old maps. this also does some error checking for us
3027 : : */
3028 : 3 : while (find_vma_links(mm, addr, addr + len, &prev, &rb_link,
3029 : : &rb_parent)) {
3030 : 0 : if (do_munmap(mm, addr, len, uf))
3031 : : return -ENOMEM;
3032 : : }
3033 : :
3034 : : /* Check against address space limits *after* clearing old maps... */
3035 : 3 : if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT))
3036 : : return -ENOMEM;
3037 : :
3038 : 3 : if (mm->map_count > sysctl_max_map_count)
3039 : : return -ENOMEM;
3040 : :
3041 : 3 : if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
3042 : : return -ENOMEM;
3043 : :
3044 : : /* Can we just expand an old private anonymous mapping? */
3045 : 3 : vma = vma_merge(mm, prev, addr, addr + len, flags,
3046 : : NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX);
3047 : 3 : if (vma)
3048 : : goto out;
3049 : :
3050 : : /*
3051 : : * create a vma struct for an anonymous mapping
3052 : : */
3053 : 3 : vma = vm_area_alloc(mm);
3054 : 3 : if (!vma) {
3055 : : vm_unacct_memory(len >> PAGE_SHIFT);
3056 : 0 : return -ENOMEM;
3057 : : }
3058 : :
3059 : : vma_set_anonymous(vma);
3060 : 3 : vma->vm_start = addr;
3061 : 3 : vma->vm_end = addr + len;
3062 : 3 : vma->vm_pgoff = pgoff;
3063 : 3 : vma->vm_flags = flags;
3064 : 3 : vma->vm_page_prot = vm_get_page_prot(flags);
3065 : 3 : vma_link(mm, vma, prev, rb_link, rb_parent);
3066 : : out:
3067 : 3 : perf_event_mmap(vma);
3068 : 3 : mm->total_vm += len >> PAGE_SHIFT;
3069 : 3 : mm->data_vm += len >> PAGE_SHIFT;
3070 : 3 : if (flags & VM_LOCKED)
3071 : 0 : mm->locked_vm += (len >> PAGE_SHIFT);
3072 : : vma->vm_flags |= VM_SOFTDIRTY;
3073 : : return 0;
3074 : : }
3075 : :
3076 : 3 : int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags)
3077 : : {
3078 : 3 : struct mm_struct *mm = current->mm;
3079 : : unsigned long len;
3080 : : int ret;
3081 : : bool populate;
3082 : 3 : LIST_HEAD(uf);
3083 : :
3084 : 3 : len = PAGE_ALIGN(request);
3085 : 3 : if (len < request)
3086 : : return -ENOMEM;
3087 : 3 : if (!len)
3088 : : return 0;
3089 : :
3090 : 3 : if (down_write_killable(&mm->mmap_sem))
3091 : : return -EINTR;
3092 : :
3093 : 3 : ret = do_brk_flags(addr, len, flags, &uf);
3094 : 3 : populate = ((mm->def_flags & VM_LOCKED) != 0);
3095 : 3 : up_write(&mm->mmap_sem);
3096 : : userfaultfd_unmap_complete(mm, &uf);
3097 : 3 : if (populate && !ret)
3098 : : mm_populate(addr, len);
3099 : 3 : return ret;
3100 : : }
3101 : : EXPORT_SYMBOL(vm_brk_flags);
3102 : :
3103 : 0 : int vm_brk(unsigned long addr, unsigned long len)
3104 : : {
3105 : 0 : return vm_brk_flags(addr, len, 0);
3106 : : }
3107 : : EXPORT_SYMBOL(vm_brk);
3108 : :
3109 : : /* Release all mmaps. */
3110 : 3 : void exit_mmap(struct mm_struct *mm)
3111 : : {
3112 : : struct mmu_gather tlb;
3113 : : struct vm_area_struct *vma;
3114 : : unsigned long nr_accounted = 0;
3115 : :
3116 : : /* mm's last user has gone, and its about to be pulled down */
3117 : : mmu_notifier_release(mm);
3118 : :
3119 : 3 : if (unlikely(mm_is_oom_victim(mm))) {
3120 : : /*
3121 : : * Manually reap the mm to free as much memory as possible.
3122 : : * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard
3123 : : * this mm from further consideration. Taking mm->mmap_sem for
3124 : : * write after setting MMF_OOM_SKIP will guarantee that the oom
3125 : : * reaper will not run on this mm again after mmap_sem is
3126 : : * dropped.
3127 : : *
3128 : : * Nothing can be holding mm->mmap_sem here and the above call
3129 : : * to mmu_notifier_release(mm) ensures mmu notifier callbacks in
3130 : : * __oom_reap_task_mm() will not block.
3131 : : *
3132 : : * This needs to be done before calling munlock_vma_pages_all(),
3133 : : * which clears VM_LOCKED, otherwise the oom reaper cannot
3134 : : * reliably test it.
3135 : : */
3136 : 0 : (void)__oom_reap_task_mm(mm);
3137 : :
3138 : 0 : set_bit(MMF_OOM_SKIP, &mm->flags);
3139 : 0 : down_write(&mm->mmap_sem);
3140 : 0 : up_write(&mm->mmap_sem);
3141 : : }
3142 : :
3143 : 3 : if (mm->locked_vm) {
3144 : 0 : vma = mm->mmap;
3145 : 0 : while (vma) {
3146 : 0 : if (vma->vm_flags & VM_LOCKED)
3147 : : munlock_vma_pages_all(vma);
3148 : 0 : vma = vma->vm_next;
3149 : : }
3150 : : }
3151 : :
3152 : : arch_exit_mmap(mm);
3153 : :
3154 : 3 : vma = mm->mmap;
3155 : 3 : if (!vma) /* Can happen if dup_mmap() received an OOM */
3156 : 0 : return;
3157 : :
3158 : 3 : lru_add_drain();
3159 : 3 : flush_cache_mm(mm);
3160 : 3 : tlb_gather_mmu(&tlb, mm, 0, -1);
3161 : : /* update_hiwater_rss(mm) here? but nobody should be looking */
3162 : : /* Use -1 here to ensure all VMAs in the mm are unmapped */
3163 : 3 : unmap_vmas(&tlb, vma, 0, -1);
3164 : 3 : free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING);
3165 : 3 : tlb_finish_mmu(&tlb, 0, -1);
3166 : :
3167 : : /*
3168 : : * Walk the list again, actually closing and freeing it,
3169 : : * with preemption enabled, without holding any MM locks.
3170 : : */
3171 : 3 : while (vma) {
3172 : 3 : if (vma->vm_flags & VM_ACCOUNT)
3173 : 3 : nr_accounted += vma_pages(vma);
3174 : 3 : vma = remove_vma(vma);
3175 : 3 : cond_resched();
3176 : : }
3177 : 3 : vm_unacct_memory(nr_accounted);
3178 : : }
3179 : :
3180 : : /* Insert vm structure into process list sorted by address
3181 : : * and into the inode's i_mmap tree. If vm_file is non-NULL
3182 : : * then i_mmap_rwsem is taken here.
3183 : : */
3184 : 3 : int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
3185 : : {
3186 : : struct vm_area_struct *prev;
3187 : : struct rb_node **rb_link, *rb_parent;
3188 : :
3189 : 3 : if (find_vma_links(mm, vma->vm_start, vma->vm_end,
3190 : : &prev, &rb_link, &rb_parent))
3191 : : return -ENOMEM;
3192 : 3 : if ((vma->vm_flags & VM_ACCOUNT) &&
3193 : 3 : security_vm_enough_memory_mm(mm, vma_pages(vma)))
3194 : : return -ENOMEM;
3195 : :
3196 : : /*
3197 : : * The vm_pgoff of a purely anonymous vma should be irrelevant
3198 : : * until its first write fault, when page's anon_vma and index
3199 : : * are set. But now set the vm_pgoff it will almost certainly
3200 : : * end up with (unless mremap moves it elsewhere before that
3201 : : * first wfault), so /proc/pid/maps tells a consistent story.
3202 : : *
3203 : : * By setting it to reflect the virtual start address of the
3204 : : * vma, merges and splits can happen in a seamless way, just
3205 : : * using the existing file pgoff checks and manipulations.
3206 : : * Similarly in do_mmap_pgoff and in do_brk.
3207 : : */
3208 : 3 : if (vma_is_anonymous(vma)) {
3209 : 3 : BUG_ON(vma->anon_vma);
3210 : 3 : vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT;
3211 : : }
3212 : :
3213 : 3 : vma_link(mm, vma, prev, rb_link, rb_parent);
3214 : 3 : return 0;
3215 : : }
3216 : :
3217 : : /*
3218 : : * Copy the vma structure to a new location in the same mm,
3219 : : * prior to moving page table entries, to effect an mremap move.
3220 : : */
3221 : 3 : struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
3222 : : unsigned long addr, unsigned long len, pgoff_t pgoff,
3223 : : bool *need_rmap_locks)
3224 : : {
3225 : 3 : struct vm_area_struct *vma = *vmap;
3226 : 3 : unsigned long vma_start = vma->vm_start;
3227 : 3 : struct mm_struct *mm = vma->vm_mm;
3228 : : struct vm_area_struct *new_vma, *prev;
3229 : : struct rb_node **rb_link, *rb_parent;
3230 : : bool faulted_in_anon_vma = true;
3231 : :
3232 : : /*
3233 : : * If anonymous vma has not yet been faulted, update new pgoff
3234 : : * to match new location, to increase its chance of merging.
3235 : : */
3236 : 3 : if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) {
3237 : 0 : pgoff = addr >> PAGE_SHIFT;
3238 : : faulted_in_anon_vma = false;
3239 : : }
3240 : :
3241 : 3 : if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
3242 : : return NULL; /* should never get here */
3243 : 3 : new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
3244 : : vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
3245 : : vma->vm_userfaultfd_ctx);
3246 : 3 : if (new_vma) {
3247 : : /*
3248 : : * Source vma may have been merged into new_vma
3249 : : */
3250 : 0 : if (unlikely(vma_start >= new_vma->vm_start &&
3251 : : vma_start < new_vma->vm_end)) {
3252 : : /*
3253 : : * The only way we can get a vma_merge with
3254 : : * self during an mremap is if the vma hasn't
3255 : : * been faulted in yet and we were allowed to
3256 : : * reset the dst vma->vm_pgoff to the
3257 : : * destination address of the mremap to allow
3258 : : * the merge to happen. mremap must change the
3259 : : * vm_pgoff linearity between src and dst vmas
3260 : : * (in turn preventing a vma_merge) to be
3261 : : * safe. It is only safe to keep the vm_pgoff
3262 : : * linear if there are no pages mapped yet.
3263 : : */
3264 : : VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma);
3265 : 0 : *vmap = vma = new_vma;
3266 : : }
3267 : 0 : *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff);
3268 : : } else {
3269 : 3 : new_vma = vm_area_dup(vma);
3270 : 3 : if (!new_vma)
3271 : : goto out;
3272 : 3 : new_vma->vm_start = addr;
3273 : 3 : new_vma->vm_end = addr + len;
3274 : 3 : new_vma->vm_pgoff = pgoff;
3275 : : if (vma_dup_policy(vma, new_vma))
3276 : : goto out_free_vma;
3277 : 3 : if (anon_vma_clone(new_vma, vma))
3278 : : goto out_free_mempol;
3279 : 3 : if (new_vma->vm_file)
3280 : : get_file(new_vma->vm_file);
3281 : 3 : if (new_vma->vm_ops && new_vma->vm_ops->open)
3282 : 0 : new_vma->vm_ops->open(new_vma);
3283 : 3 : vma_link(mm, new_vma, prev, rb_link, rb_parent);
3284 : 3 : *need_rmap_locks = false;
3285 : : }
3286 : 3 : return new_vma;
3287 : :
3288 : : out_free_mempol:
3289 : : mpol_put(vma_policy(new_vma));
3290 : : out_free_vma:
3291 : 0 : vm_area_free(new_vma);
3292 : : out:
3293 : : return NULL;
3294 : : }
3295 : :
3296 : : /*
3297 : : * Return true if the calling process may expand its vm space by the passed
3298 : : * number of pages
3299 : : */
3300 : 3 : bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages)
3301 : : {
3302 : 3 : if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT)
3303 : : return false;
3304 : :
3305 : 3 : if (is_data_mapping(flags) &&
3306 : 3 : mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) {
3307 : : /* Workaround for Valgrind */
3308 : 0 : if (rlimit(RLIMIT_DATA) == 0 &&
3309 : 0 : mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT)
3310 : : return true;
3311 : :
3312 : 0 : pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n",
3313 : : current->comm, current->pid,
3314 : : (mm->data_vm + npages) << PAGE_SHIFT,
3315 : : rlimit(RLIMIT_DATA),
3316 : : ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data");
3317 : :
3318 : 0 : if (!ignore_rlimit_data)
3319 : : return false;
3320 : : }
3321 : :
3322 : : return true;
3323 : : }
3324 : :
3325 : 3 : void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages)
3326 : : {
3327 : 3 : mm->total_vm += npages;
3328 : :
3329 : 3 : if (is_exec_mapping(flags))
3330 : 3 : mm->exec_vm += npages;
3331 : 3 : else if (is_stack_mapping(flags))
3332 : 3 : mm->stack_vm += npages;
3333 : 3 : else if (is_data_mapping(flags))
3334 : 3 : mm->data_vm += npages;
3335 : 3 : }
3336 : :
3337 : : static vm_fault_t special_mapping_fault(struct vm_fault *vmf);
3338 : :
3339 : : /*
3340 : : * Having a close hook prevents vma merging regardless of flags.
3341 : : */
3342 : 3 : static void special_mapping_close(struct vm_area_struct *vma)
3343 : : {
3344 : 3 : }
3345 : :
3346 : 1 : static const char *special_mapping_name(struct vm_area_struct *vma)
3347 : : {
3348 : 1 : return ((struct vm_special_mapping *)vma->vm_private_data)->name;
3349 : : }
3350 : :
3351 : 0 : static int special_mapping_mremap(struct vm_area_struct *new_vma)
3352 : : {
3353 : 0 : struct vm_special_mapping *sm = new_vma->vm_private_data;
3354 : :
3355 : 0 : if (WARN_ON_ONCE(current->mm != new_vma->vm_mm))
3356 : : return -EFAULT;
3357 : :
3358 : 0 : if (sm->mremap)
3359 : 0 : return sm->mremap(sm, new_vma);
3360 : :
3361 : : return 0;
3362 : : }
3363 : :
3364 : : static const struct vm_operations_struct special_mapping_vmops = {
3365 : : .close = special_mapping_close,
3366 : : .fault = special_mapping_fault,
3367 : : .mremap = special_mapping_mremap,
3368 : : .name = special_mapping_name,
3369 : : };
3370 : :
3371 : : static const struct vm_operations_struct legacy_special_mapping_vmops = {
3372 : : .close = special_mapping_close,
3373 : : .fault = special_mapping_fault,
3374 : : };
3375 : :
3376 : 3 : static vm_fault_t special_mapping_fault(struct vm_fault *vmf)
3377 : : {
3378 : 3 : struct vm_area_struct *vma = vmf->vma;
3379 : : pgoff_t pgoff;
3380 : : struct page **pages;
3381 : :
3382 : 3 : if (vma->vm_ops == &legacy_special_mapping_vmops) {
3383 : 0 : pages = vma->vm_private_data;
3384 : : } else {
3385 : 3 : struct vm_special_mapping *sm = vma->vm_private_data;
3386 : :
3387 : 3 : if (sm->fault)
3388 : 0 : return sm->fault(sm, vmf->vma, vmf);
3389 : :
3390 : 3 : pages = sm->pages;
3391 : : }
3392 : :
3393 : 3 : for (pgoff = vmf->pgoff; pgoff && *pages; ++pages)
3394 : 0 : pgoff--;
3395 : :
3396 : 3 : if (*pages) {
3397 : : struct page *page = *pages;
3398 : 3 : get_page(page);
3399 : 3 : vmf->page = page;
3400 : 3 : return 0;
3401 : : }
3402 : :
3403 : : return VM_FAULT_SIGBUS;
3404 : : }
3405 : :
3406 : 3 : static struct vm_area_struct *__install_special_mapping(
3407 : : struct mm_struct *mm,
3408 : : unsigned long addr, unsigned long len,
3409 : : unsigned long vm_flags, void *priv,
3410 : : const struct vm_operations_struct *ops)
3411 : : {
3412 : : int ret;
3413 : : struct vm_area_struct *vma;
3414 : :
3415 : 3 : vma = vm_area_alloc(mm);
3416 : 3 : if (unlikely(vma == NULL))
3417 : : return ERR_PTR(-ENOMEM);
3418 : :
3419 : 3 : vma->vm_start = addr;
3420 : 3 : vma->vm_end = addr + len;
3421 : :
3422 : 3 : vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY;
3423 : 3 : vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
3424 : :
3425 : 3 : vma->vm_ops = ops;
3426 : 3 : vma->vm_private_data = priv;
3427 : :
3428 : 3 : ret = insert_vm_struct(mm, vma);
3429 : 3 : if (ret)
3430 : : goto out;
3431 : :
3432 : 3 : vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
3433 : :
3434 : 3 : perf_event_mmap(vma);
3435 : :
3436 : 3 : return vma;
3437 : :
3438 : : out:
3439 : 0 : vm_area_free(vma);
3440 : 0 : return ERR_PTR(ret);
3441 : : }
3442 : :
3443 : 0 : bool vma_is_special_mapping(const struct vm_area_struct *vma,
3444 : : const struct vm_special_mapping *sm)
3445 : : {
3446 : 0 : return vma->vm_private_data == sm &&
3447 : 0 : (vma->vm_ops == &special_mapping_vmops ||
3448 : : vma->vm_ops == &legacy_special_mapping_vmops);
3449 : : }
3450 : :
3451 : : /*
3452 : : * Called with mm->mmap_sem held for writing.
3453 : : * Insert a new vma covering the given region, with the given flags.
3454 : : * Its pages are supplied by the given array of struct page *.
3455 : : * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated.
3456 : : * The region past the last page supplied will always produce SIGBUS.
3457 : : * The array pointer and the pages it points to are assumed to stay alive
3458 : : * for as long as this mapping might exist.
3459 : : */
3460 : 3 : struct vm_area_struct *_install_special_mapping(
3461 : : struct mm_struct *mm,
3462 : : unsigned long addr, unsigned long len,
3463 : : unsigned long vm_flags, const struct vm_special_mapping *spec)
3464 : : {
3465 : 3 : return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec,
3466 : : &special_mapping_vmops);
3467 : : }
3468 : :
3469 : 0 : int install_special_mapping(struct mm_struct *mm,
3470 : : unsigned long addr, unsigned long len,
3471 : : unsigned long vm_flags, struct page **pages)
3472 : : {
3473 : 0 : struct vm_area_struct *vma = __install_special_mapping(
3474 : : mm, addr, len, vm_flags, (void *)pages,
3475 : : &legacy_special_mapping_vmops);
3476 : :
3477 : 0 : return PTR_ERR_OR_ZERO(vma);
3478 : : }
3479 : :
3480 : : static DEFINE_MUTEX(mm_all_locks_mutex);
3481 : :
3482 : 0 : static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
3483 : : {
3484 : 0 : if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3485 : : /*
3486 : : * The LSB of head.next can't change from under us
3487 : : * because we hold the mm_all_locks_mutex.
3488 : : */
3489 : 0 : down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem);
3490 : : /*
3491 : : * We can safely modify head.next after taking the
3492 : : * anon_vma->root->rwsem. If some other vma in this mm shares
3493 : : * the same anon_vma we won't take it again.
3494 : : *
3495 : : * No need of atomic instructions here, head.next
3496 : : * can't change from under us thanks to the
3497 : : * anon_vma->root->rwsem.
3498 : : */
3499 : 0 : if (__test_and_set_bit(0, (unsigned long *)
3500 : 0 : &anon_vma->root->rb_root.rb_root.rb_node))
3501 : 0 : BUG();
3502 : : }
3503 : 0 : }
3504 : :
3505 : 0 : static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
3506 : : {
3507 : 0 : if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3508 : : /*
3509 : : * AS_MM_ALL_LOCKS can't change from under us because
3510 : : * we hold the mm_all_locks_mutex.
3511 : : *
3512 : : * Operations on ->flags have to be atomic because
3513 : : * even if AS_MM_ALL_LOCKS is stable thanks to the
3514 : : * mm_all_locks_mutex, there may be other cpus
3515 : : * changing other bitflags in parallel to us.
3516 : : */
3517 : 0 : if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
3518 : 0 : BUG();
3519 : 0 : down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem);
3520 : : }
3521 : 0 : }
3522 : :
3523 : : /*
3524 : : * This operation locks against the VM for all pte/vma/mm related
3525 : : * operations that could ever happen on a certain mm. This includes
3526 : : * vmtruncate, try_to_unmap, and all page faults.
3527 : : *
3528 : : * The caller must take the mmap_sem in write mode before calling
3529 : : * mm_take_all_locks(). The caller isn't allowed to release the
3530 : : * mmap_sem until mm_drop_all_locks() returns.
3531 : : *
3532 : : * mmap_sem in write mode is required in order to block all operations
3533 : : * that could modify pagetables and free pages without need of
3534 : : * altering the vma layout. It's also needed in write mode to avoid new
3535 : : * anon_vmas to be associated with existing vmas.
3536 : : *
3537 : : * A single task can't take more than one mm_take_all_locks() in a row
3538 : : * or it would deadlock.
3539 : : *
3540 : : * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
3541 : : * mapping->flags avoid to take the same lock twice, if more than one
3542 : : * vma in this mm is backed by the same anon_vma or address_space.
3543 : : *
3544 : : * We take locks in following order, accordingly to comment at beginning
3545 : : * of mm/rmap.c:
3546 : : * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
3547 : : * hugetlb mapping);
3548 : : * - all i_mmap_rwsem locks;
3549 : : * - all anon_vma->rwseml
3550 : : *
3551 : : * We can take all locks within these types randomly because the VM code
3552 : : * doesn't nest them and we protected from parallel mm_take_all_locks() by
3553 : : * mm_all_locks_mutex.
3554 : : *
3555 : : * mm_take_all_locks() and mm_drop_all_locks are expensive operations
3556 : : * that may have to take thousand of locks.
3557 : : *
3558 : : * mm_take_all_locks() can fail if it's interrupted by signals.
3559 : : */
3560 : 0 : int mm_take_all_locks(struct mm_struct *mm)
3561 : : {
3562 : : struct vm_area_struct *vma;
3563 : : struct anon_vma_chain *avc;
3564 : :
3565 : 0 : BUG_ON(down_read_trylock(&mm->mmap_sem));
3566 : :
3567 : 0 : mutex_lock(&mm_all_locks_mutex);
3568 : :
3569 : 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3570 : 0 : if (signal_pending(current))
3571 : : goto out_unlock;
3572 : : if (vma->vm_file && vma->vm_file->f_mapping &&
3573 : : is_vm_hugetlb_page(vma))
3574 : : vm_lock_mapping(mm, vma->vm_file->f_mapping);
3575 : : }
3576 : :
3577 : 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3578 : 0 : if (signal_pending(current))
3579 : : goto out_unlock;
3580 : 0 : if (vma->vm_file && vma->vm_file->f_mapping &&
3581 : : !is_vm_hugetlb_page(vma))
3582 : 0 : vm_lock_mapping(mm, vma->vm_file->f_mapping);
3583 : : }
3584 : :
3585 : 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3586 : 0 : if (signal_pending(current))
3587 : : goto out_unlock;
3588 : 0 : if (vma->anon_vma)
3589 : 0 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3590 : 0 : vm_lock_anon_vma(mm, avc->anon_vma);
3591 : : }
3592 : :
3593 : : return 0;
3594 : :
3595 : : out_unlock:
3596 : 0 : mm_drop_all_locks(mm);
3597 : 0 : return -EINTR;
3598 : : }
3599 : :
3600 : 0 : static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
3601 : : {
3602 : 0 : if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
3603 : : /*
3604 : : * The LSB of head.next can't change to 0 from under
3605 : : * us because we hold the mm_all_locks_mutex.
3606 : : *
3607 : : * We must however clear the bitflag before unlocking
3608 : : * the vma so the users using the anon_vma->rb_root will
3609 : : * never see our bitflag.
3610 : : *
3611 : : * No need of atomic instructions here, head.next
3612 : : * can't change from under us until we release the
3613 : : * anon_vma->root->rwsem.
3614 : : */
3615 : 0 : if (!__test_and_clear_bit(0, (unsigned long *)
3616 : : &anon_vma->root->rb_root.rb_root.rb_node))
3617 : 0 : BUG();
3618 : : anon_vma_unlock_write(anon_vma);
3619 : : }
3620 : 0 : }
3621 : :
3622 : 0 : static void vm_unlock_mapping(struct address_space *mapping)
3623 : : {
3624 : 0 : if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
3625 : : /*
3626 : : * AS_MM_ALL_LOCKS can't change to 0 from under us
3627 : : * because we hold the mm_all_locks_mutex.
3628 : : */
3629 : : i_mmap_unlock_write(mapping);
3630 : 0 : if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
3631 : : &mapping->flags))
3632 : 0 : BUG();
3633 : : }
3634 : 0 : }
3635 : :
3636 : : /*
3637 : : * The mmap_sem cannot be released by the caller until
3638 : : * mm_drop_all_locks() returns.
3639 : : */
3640 : 0 : void mm_drop_all_locks(struct mm_struct *mm)
3641 : : {
3642 : : struct vm_area_struct *vma;
3643 : : struct anon_vma_chain *avc;
3644 : :
3645 : 0 : BUG_ON(down_read_trylock(&mm->mmap_sem));
3646 : 0 : BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
3647 : :
3648 : 0 : for (vma = mm->mmap; vma; vma = vma->vm_next) {
3649 : 0 : if (vma->anon_vma)
3650 : 0 : list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
3651 : 0 : vm_unlock_anon_vma(avc->anon_vma);
3652 : 0 : if (vma->vm_file && vma->vm_file->f_mapping)
3653 : 0 : vm_unlock_mapping(vma->vm_file->f_mapping);
3654 : : }
3655 : :
3656 : 0 : mutex_unlock(&mm_all_locks_mutex);
3657 : 0 : }
3658 : :
3659 : : /*
3660 : : * initialise the percpu counter for VM
3661 : : */
3662 : 3 : void __init mmap_init(void)
3663 : : {
3664 : : int ret;
3665 : :
3666 : 3 : ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL);
3667 : : VM_BUG_ON(ret);
3668 : 3 : }
3669 : :
3670 : : /*
3671 : : * Initialise sysctl_user_reserve_kbytes.
3672 : : *
3673 : : * This is intended to prevent a user from starting a single memory hogging
3674 : : * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER
3675 : : * mode.
3676 : : *
3677 : : * The default value is min(3% of free memory, 128MB)
3678 : : * 128MB is enough to recover with sshd/login, bash, and top/kill.
3679 : : */
3680 : 3 : static int init_user_reserve(void)
3681 : : {
3682 : : unsigned long free_kbytes;
3683 : :
3684 : 3 : free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3685 : :
3686 : 3 : sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17);
3687 : 3 : return 0;
3688 : : }
3689 : : subsys_initcall(init_user_reserve);
3690 : :
3691 : : /*
3692 : : * Initialise sysctl_admin_reserve_kbytes.
3693 : : *
3694 : : * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin
3695 : : * to log in and kill a memory hogging process.
3696 : : *
3697 : : * Systems with more than 256MB will reserve 8MB, enough to recover
3698 : : * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will
3699 : : * only reserve 3% of free pages by default.
3700 : : */
3701 : 3 : static int init_admin_reserve(void)
3702 : : {
3703 : : unsigned long free_kbytes;
3704 : :
3705 : 3 : free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3706 : :
3707 : 3 : sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13);
3708 : 3 : return 0;
3709 : : }
3710 : : subsys_initcall(init_admin_reserve);
3711 : :
3712 : : /*
3713 : : * Reinititalise user and admin reserves if memory is added or removed.
3714 : : *
3715 : : * The default user reserve max is 128MB, and the default max for the
3716 : : * admin reserve is 8MB. These are usually, but not always, enough to
3717 : : * enable recovery from a memory hogging process using login/sshd, a shell,
3718 : : * and tools like top. It may make sense to increase or even disable the
3719 : : * reserve depending on the existence of swap or variations in the recovery
3720 : : * tools. So, the admin may have changed them.
3721 : : *
3722 : : * If memory is added and the reserves have been eliminated or increased above
3723 : : * the default max, then we'll trust the admin.
3724 : : *
3725 : : * If memory is removed and there isn't enough free memory, then we
3726 : : * need to reset the reserves.
3727 : : *
3728 : : * Otherwise keep the reserve set by the admin.
3729 : : */
3730 : : static int reserve_mem_notifier(struct notifier_block *nb,
3731 : : unsigned long action, void *data)
3732 : : {
3733 : : unsigned long tmp, free_kbytes;
3734 : :
3735 : : switch (action) {
3736 : : case MEM_ONLINE:
3737 : : /* Default max is 128MB. Leave alone if modified by operator. */
3738 : : tmp = sysctl_user_reserve_kbytes;
3739 : : if (0 < tmp && tmp < (1UL << 17))
3740 : : init_user_reserve();
3741 : :
3742 : : /* Default max is 8MB. Leave alone if modified by operator. */
3743 : : tmp = sysctl_admin_reserve_kbytes;
3744 : : if (0 < tmp && tmp < (1UL << 13))
3745 : : init_admin_reserve();
3746 : :
3747 : : break;
3748 : : case MEM_OFFLINE:
3749 : : free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10);
3750 : :
3751 : : if (sysctl_user_reserve_kbytes > free_kbytes) {
3752 : : init_user_reserve();
3753 : : pr_info("vm.user_reserve_kbytes reset to %lu\n",
3754 : : sysctl_user_reserve_kbytes);
3755 : : }
3756 : :
3757 : : if (sysctl_admin_reserve_kbytes > free_kbytes) {
3758 : : init_admin_reserve();
3759 : : pr_info("vm.admin_reserve_kbytes reset to %lu\n",
3760 : : sysctl_admin_reserve_kbytes);
3761 : : }
3762 : : break;
3763 : : default:
3764 : : break;
3765 : : }
3766 : : return NOTIFY_OK;
3767 : : }
3768 : :
3769 : : static struct notifier_block reserve_mem_nb = {
3770 : : .notifier_call = reserve_mem_notifier,
3771 : : };
3772 : :
3773 : 3 : static int __meminit init_reserve_notifier(void)
3774 : : {
3775 : : if (register_hotmemory_notifier(&reserve_mem_nb))
3776 : : pr_err("Failed registering memory add/remove notifier for admin reserve\n");
3777 : :
3778 : 3 : return 0;
3779 : : }
3780 : : subsys_initcall(init_reserve_notifier);
|
