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