1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * mm/mremap.c |
4 | * |
5 | * (C) Copyright 1996 Linus Torvalds |
6 | * |
7 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
8 | * (C) Copyright 2002 Red Hat Inc, All Rights Reserved |
9 | */ |
10 | |
11 | #include <linux/mm.h> |
12 | #include <linux/hugetlb.h> |
13 | #include <linux/shm.h> |
14 | #include <linux/ksm.h> |
15 | #include <linux/mman.h> |
16 | #include <linux/swap.h> |
17 | #include <linux/capability.h> |
18 | #include <linux/fs.h> |
19 | #include <linux/swapops.h> |
20 | #include <linux/highmem.h> |
21 | #include <linux/security.h> |
22 | #include <linux/syscalls.h> |
23 | #include <linux/mmu_notifier.h> |
24 | #include <linux/uaccess.h> |
25 | #include <linux/mm-arch-hooks.h> |
26 | #include <linux/userfaultfd_k.h> |
27 | |
28 | #include <asm/cacheflush.h> |
29 | #include <asm/tlbflush.h> |
30 | |
31 | #include "internal.h" |
32 | |
33 | static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr) |
34 | { |
35 | pgd_t *pgd; |
36 | p4d_t *p4d; |
37 | pud_t *pud; |
38 | pmd_t *pmd; |
39 | |
40 | pgd = pgd_offset(mm, addr); |
41 | if (pgd_none_or_clear_bad(pgd)) |
42 | return NULL; |
43 | |
44 | p4d = p4d_offset(pgd, addr); |
45 | if (p4d_none_or_clear_bad(p4d)) |
46 | return NULL; |
47 | |
48 | pud = pud_offset(p4d, addr); |
49 | if (pud_none_or_clear_bad(pud)) |
50 | return NULL; |
51 | |
52 | pmd = pmd_offset(pud, addr); |
53 | if (pmd_none(*pmd)) |
54 | return NULL; |
55 | |
56 | return pmd; |
57 | } |
58 | |
59 | static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma, |
60 | unsigned long addr) |
61 | { |
62 | pgd_t *pgd; |
63 | p4d_t *p4d; |
64 | pud_t *pud; |
65 | pmd_t *pmd; |
66 | |
67 | pgd = pgd_offset(mm, addr); |
68 | p4d = p4d_alloc(mm, pgd, addr); |
69 | if (!p4d) |
70 | return NULL; |
71 | pud = pud_alloc(mm, p4d, addr); |
72 | if (!pud) |
73 | return NULL; |
74 | |
75 | pmd = pmd_alloc(mm, pud, addr); |
76 | if (!pmd) |
77 | return NULL; |
78 | |
79 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
80 | |
81 | return pmd; |
82 | } |
83 | |
84 | static void take_rmap_locks(struct vm_area_struct *vma) |
85 | { |
86 | if (vma->vm_file) |
87 | i_mmap_lock_write(vma->vm_file->f_mapping); |
88 | if (vma->anon_vma) |
89 | anon_vma_lock_write(vma->anon_vma); |
90 | } |
91 | |
92 | static void drop_rmap_locks(struct vm_area_struct *vma) |
93 | { |
94 | if (vma->anon_vma) |
95 | anon_vma_unlock_write(vma->anon_vma); |
96 | if (vma->vm_file) |
97 | i_mmap_unlock_write(vma->vm_file->f_mapping); |
98 | } |
99 | |
100 | static pte_t move_soft_dirty_pte(pte_t pte) |
101 | { |
102 | /* |
103 | * Set soft dirty bit so we can notice |
104 | * in userspace the ptes were moved. |
105 | */ |
106 | #ifdef CONFIG_MEM_SOFT_DIRTY |
107 | if (pte_present(pte)) |
108 | pte = pte_mksoft_dirty(pte); |
109 | else if (is_swap_pte(pte)) |
110 | pte = pte_swp_mksoft_dirty(pte); |
111 | #endif |
112 | return pte; |
113 | } |
114 | |
115 | static void move_ptes(struct vm_area_struct *vma, pmd_t *old_pmd, |
116 | unsigned long old_addr, unsigned long old_end, |
117 | struct vm_area_struct *new_vma, pmd_t *new_pmd, |
118 | unsigned long new_addr, bool need_rmap_locks) |
119 | { |
120 | struct mm_struct *mm = vma->vm_mm; |
121 | pte_t *old_pte, *new_pte, pte; |
122 | spinlock_t *old_ptl, *new_ptl; |
123 | bool force_flush = false; |
124 | unsigned long len = old_end - old_addr; |
125 | |
126 | /* |
127 | * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma |
128 | * locks to ensure that rmap will always observe either the old or the |
129 | * new ptes. This is the easiest way to avoid races with |
130 | * truncate_pagecache(), page migration, etc... |
131 | * |
132 | * When need_rmap_locks is false, we use other ways to avoid |
133 | * such races: |
134 | * |
135 | * - During exec() shift_arg_pages(), we use a specially tagged vma |
136 | * which rmap call sites look for using is_vma_temporary_stack(). |
137 | * |
138 | * - During mremap(), new_vma is often known to be placed after vma |
139 | * in rmap traversal order. This ensures rmap will always observe |
140 | * either the old pte, or the new pte, or both (the page table locks |
141 | * serialize access to individual ptes, but only rmap traversal |
142 | * order guarantees that we won't miss both the old and new ptes). |
143 | */ |
144 | if (need_rmap_locks) |
145 | take_rmap_locks(vma); |
146 | |
147 | /* |
148 | * We don't have to worry about the ordering of src and dst |
149 | * pte locks because exclusive mmap_sem prevents deadlock. |
150 | */ |
151 | old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl); |
152 | new_pte = pte_offset_map(new_pmd, new_addr); |
153 | new_ptl = pte_lockptr(mm, new_pmd); |
154 | if (new_ptl != old_ptl) |
155 | spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING); |
156 | flush_tlb_batched_pending(vma->vm_mm); |
157 | arch_enter_lazy_mmu_mode(); |
158 | |
159 | for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE, |
160 | new_pte++, new_addr += PAGE_SIZE) { |
161 | if (pte_none(*old_pte)) |
162 | continue; |
163 | |
164 | pte = ptep_get_and_clear(mm, old_addr, old_pte); |
165 | /* |
166 | * If we are remapping a valid PTE, make sure |
167 | * to flush TLB before we drop the PTL for the |
168 | * PTE. |
169 | * |
170 | * NOTE! Both old and new PTL matter: the old one |
171 | * for racing with page_mkclean(), the new one to |
172 | * make sure the physical page stays valid until |
173 | * the TLB entry for the old mapping has been |
174 | * flushed. |
175 | */ |
176 | if (pte_present(pte)) |
177 | force_flush = true; |
178 | pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr); |
179 | pte = move_soft_dirty_pte(pte); |
180 | set_pte_at(mm, new_addr, new_pte, pte); |
181 | } |
182 | |
183 | arch_leave_lazy_mmu_mode(); |
184 | if (force_flush) |
185 | flush_tlb_range(vma, old_end - len, old_end); |
186 | if (new_ptl != old_ptl) |
187 | spin_unlock(new_ptl); |
188 | pte_unmap(new_pte - 1); |
189 | pte_unmap_unlock(old_pte - 1, old_ptl); |
190 | if (need_rmap_locks) |
191 | drop_rmap_locks(vma); |
192 | } |
193 | |
194 | #ifdef CONFIG_HAVE_MOVE_PMD |
195 | static bool move_normal_pmd(struct vm_area_struct *vma, unsigned long old_addr, |
196 | unsigned long new_addr, unsigned long old_end, |
197 | pmd_t *old_pmd, pmd_t *new_pmd) |
198 | { |
199 | spinlock_t *old_ptl, *new_ptl; |
200 | struct mm_struct *mm = vma->vm_mm; |
201 | pmd_t pmd; |
202 | |
203 | if ((old_addr & ~PMD_MASK) || (new_addr & ~PMD_MASK) |
204 | || old_end - old_addr < PMD_SIZE) |
205 | return false; |
206 | |
207 | /* |
208 | * The destination pmd shouldn't be established, free_pgtables() |
209 | * should have release it. |
210 | */ |
211 | if (WARN_ON(!pmd_none(*new_pmd))) |
212 | return false; |
213 | |
214 | /* |
215 | * We don't have to worry about the ordering of src and dst |
216 | * ptlocks because exclusive mmap_sem prevents deadlock. |
217 | */ |
218 | old_ptl = pmd_lock(vma->vm_mm, old_pmd); |
219 | new_ptl = pmd_lockptr(mm, new_pmd); |
220 | if (new_ptl != old_ptl) |
221 | spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING); |
222 | |
223 | /* Clear the pmd */ |
224 | pmd = *old_pmd; |
225 | pmd_clear(old_pmd); |
226 | |
227 | VM_BUG_ON(!pmd_none(*new_pmd)); |
228 | |
229 | /* Set the new pmd */ |
230 | set_pmd_at(mm, new_addr, new_pmd, pmd); |
231 | flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE); |
232 | if (new_ptl != old_ptl) |
233 | spin_unlock(new_ptl); |
234 | spin_unlock(old_ptl); |
235 | |
236 | return true; |
237 | } |
238 | #endif |
239 | |
240 | unsigned long move_page_tables(struct vm_area_struct *vma, |
241 | unsigned long old_addr, struct vm_area_struct *new_vma, |
242 | unsigned long new_addr, unsigned long len, |
243 | bool need_rmap_locks) |
244 | { |
245 | unsigned long extent, next, old_end; |
246 | struct mmu_notifier_range range; |
247 | pmd_t *old_pmd, *new_pmd; |
248 | |
249 | old_end = old_addr + len; |
250 | flush_cache_range(vma, old_addr, old_end); |
251 | |
252 | mmu_notifier_range_init(&range, vma->vm_mm, old_addr, old_end); |
253 | mmu_notifier_invalidate_range_start(&range); |
254 | |
255 | for (; old_addr < old_end; old_addr += extent, new_addr += extent) { |
256 | cond_resched(); |
257 | next = (old_addr + PMD_SIZE) & PMD_MASK; |
258 | /* even if next overflowed, extent below will be ok */ |
259 | extent = next - old_addr; |
260 | if (extent > old_end - old_addr) |
261 | extent = old_end - old_addr; |
262 | old_pmd = get_old_pmd(vma->vm_mm, old_addr); |
263 | if (!old_pmd) |
264 | continue; |
265 | new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr); |
266 | if (!new_pmd) |
267 | break; |
268 | if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) { |
269 | if (extent == HPAGE_PMD_SIZE) { |
270 | bool moved; |
271 | /* See comment in move_ptes() */ |
272 | if (need_rmap_locks) |
273 | take_rmap_locks(vma); |
274 | moved = move_huge_pmd(vma, old_addr, new_addr, |
275 | old_end, old_pmd, new_pmd); |
276 | if (need_rmap_locks) |
277 | drop_rmap_locks(vma); |
278 | if (moved) |
279 | continue; |
280 | } |
281 | split_huge_pmd(vma, old_pmd, old_addr); |
282 | if (pmd_trans_unstable(old_pmd)) |
283 | continue; |
284 | } else if (extent == PMD_SIZE) { |
285 | #ifdef CONFIG_HAVE_MOVE_PMD |
286 | /* |
287 | * If the extent is PMD-sized, try to speed the move by |
288 | * moving at the PMD level if possible. |
289 | */ |
290 | bool moved; |
291 | |
292 | if (need_rmap_locks) |
293 | take_rmap_locks(vma); |
294 | moved = move_normal_pmd(vma, old_addr, new_addr, |
295 | old_end, old_pmd, new_pmd); |
296 | if (need_rmap_locks) |
297 | drop_rmap_locks(vma); |
298 | if (moved) |
299 | continue; |
300 | #endif |
301 | } |
302 | |
303 | if (pte_alloc(new_vma->vm_mm, new_pmd)) |
304 | break; |
305 | next = (new_addr + PMD_SIZE) & PMD_MASK; |
306 | if (extent > next - new_addr) |
307 | extent = next - new_addr; |
308 | move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma, |
309 | new_pmd, new_addr, need_rmap_locks); |
310 | } |
311 | |
312 | mmu_notifier_invalidate_range_end(&range); |
313 | |
314 | return len + old_addr - old_end; /* how much done */ |
315 | } |
316 | |
317 | static unsigned long move_vma(struct vm_area_struct *vma, |
318 | unsigned long old_addr, unsigned long old_len, |
319 | unsigned long new_len, unsigned long new_addr, |
320 | bool *locked, struct vm_userfaultfd_ctx *uf, |
321 | struct list_head *uf_unmap) |
322 | { |
323 | struct mm_struct *mm = vma->vm_mm; |
324 | struct vm_area_struct *new_vma; |
325 | unsigned long vm_flags = vma->vm_flags; |
326 | unsigned long new_pgoff; |
327 | unsigned long moved_len; |
328 | unsigned long excess = 0; |
329 | unsigned long hiwater_vm; |
330 | int split = 0; |
331 | int err; |
332 | bool need_rmap_locks; |
333 | |
334 | /* |
335 | * We'd prefer to avoid failure later on in do_munmap: |
336 | * which may split one vma into three before unmapping. |
337 | */ |
338 | if (mm->map_count >= sysctl_max_map_count - 3) |
339 | return -ENOMEM; |
340 | |
341 | /* |
342 | * Advise KSM to break any KSM pages in the area to be moved: |
343 | * it would be confusing if they were to turn up at the new |
344 | * location, where they happen to coincide with different KSM |
345 | * pages recently unmapped. But leave vma->vm_flags as it was, |
346 | * so KSM can come around to merge on vma and new_vma afterwards. |
347 | */ |
348 | err = ksm_madvise(vma, old_addr, old_addr + old_len, |
349 | MADV_UNMERGEABLE, &vm_flags); |
350 | if (err) |
351 | return err; |
352 | |
353 | new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT); |
354 | new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff, |
355 | &need_rmap_locks); |
356 | if (!new_vma) |
357 | return -ENOMEM; |
358 | |
359 | moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len, |
360 | need_rmap_locks); |
361 | if (moved_len < old_len) { |
362 | err = -ENOMEM; |
363 | } else if (vma->vm_ops && vma->vm_ops->mremap) { |
364 | err = vma->vm_ops->mremap(new_vma); |
365 | } |
366 | |
367 | if (unlikely(err)) { |
368 | /* |
369 | * On error, move entries back from new area to old, |
370 | * which will succeed since page tables still there, |
371 | * and then proceed to unmap new area instead of old. |
372 | */ |
373 | move_page_tables(new_vma, new_addr, vma, old_addr, moved_len, |
374 | true); |
375 | vma = new_vma; |
376 | old_len = new_len; |
377 | old_addr = new_addr; |
378 | new_addr = err; |
379 | } else { |
380 | mremap_userfaultfd_prep(new_vma, uf); |
381 | arch_remap(mm, old_addr, old_addr + old_len, |
382 | new_addr, new_addr + new_len); |
383 | } |
384 | |
385 | /* Conceal VM_ACCOUNT so old reservation is not undone */ |
386 | if (vm_flags & VM_ACCOUNT) { |
387 | vma->vm_flags &= ~VM_ACCOUNT; |
388 | excess = vma->vm_end - vma->vm_start - old_len; |
389 | if (old_addr > vma->vm_start && |
390 | old_addr + old_len < vma->vm_end) |
391 | split = 1; |
392 | } |
393 | |
394 | /* |
395 | * If we failed to move page tables we still do total_vm increment |
396 | * since do_munmap() will decrement it by old_len == new_len. |
397 | * |
398 | * Since total_vm is about to be raised artificially high for a |
399 | * moment, we need to restore high watermark afterwards: if stats |
400 | * are taken meanwhile, total_vm and hiwater_vm appear too high. |
401 | * If this were a serious issue, we'd add a flag to do_munmap(). |
402 | */ |
403 | hiwater_vm = mm->hiwater_vm; |
404 | vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT); |
405 | |
406 | /* Tell pfnmap has moved from this vma */ |
407 | if (unlikely(vma->vm_flags & VM_PFNMAP)) |
408 | untrack_pfn_moved(vma); |
409 | |
410 | if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) { |
411 | /* OOM: unable to split vma, just get accounts right */ |
412 | vm_unacct_memory(excess >> PAGE_SHIFT); |
413 | excess = 0; |
414 | } |
415 | mm->hiwater_vm = hiwater_vm; |
416 | |
417 | /* Restore VM_ACCOUNT if one or two pieces of vma left */ |
418 | if (excess) { |
419 | vma->vm_flags |= VM_ACCOUNT; |
420 | if (split) |
421 | vma->vm_next->vm_flags |= VM_ACCOUNT; |
422 | } |
423 | |
424 | if (vm_flags & VM_LOCKED) { |
425 | mm->locked_vm += new_len >> PAGE_SHIFT; |
426 | *locked = true; |
427 | } |
428 | |
429 | return new_addr; |
430 | } |
431 | |
432 | static struct vm_area_struct *vma_to_resize(unsigned long addr, |
433 | unsigned long old_len, unsigned long new_len, unsigned long *p) |
434 | { |
435 | struct mm_struct *mm = current->mm; |
436 | struct vm_area_struct *vma = find_vma(mm, addr); |
437 | unsigned long pgoff; |
438 | |
439 | if (!vma || vma->vm_start > addr) |
440 | return ERR_PTR(-EFAULT); |
441 | |
442 | /* |
443 | * !old_len is a special case where an attempt is made to 'duplicate' |
444 | * a mapping. This makes no sense for private mappings as it will |
445 | * instead create a fresh/new mapping unrelated to the original. This |
446 | * is contrary to the basic idea of mremap which creates new mappings |
447 | * based on the original. There are no known use cases for this |
448 | * behavior. As a result, fail such attempts. |
449 | */ |
450 | if (!old_len && !(vma->vm_flags & (VM_SHARED | VM_MAYSHARE))) { |
451 | pr_warn_once("%s (%d): attempted to duplicate a private mapping with mremap. This is not supported.\n" , current->comm, current->pid); |
452 | return ERR_PTR(-EINVAL); |
453 | } |
454 | |
455 | if (is_vm_hugetlb_page(vma)) |
456 | return ERR_PTR(-EINVAL); |
457 | |
458 | /* We can't remap across vm area boundaries */ |
459 | if (old_len > vma->vm_end - addr) |
460 | return ERR_PTR(-EFAULT); |
461 | |
462 | if (new_len == old_len) |
463 | return vma; |
464 | |
465 | /* Need to be careful about a growing mapping */ |
466 | pgoff = (addr - vma->vm_start) >> PAGE_SHIFT; |
467 | pgoff += vma->vm_pgoff; |
468 | if (pgoff + (new_len >> PAGE_SHIFT) < pgoff) |
469 | return ERR_PTR(-EINVAL); |
470 | |
471 | if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP)) |
472 | return ERR_PTR(-EFAULT); |
473 | |
474 | if (vma->vm_flags & VM_LOCKED) { |
475 | unsigned long locked, lock_limit; |
476 | locked = mm->locked_vm << PAGE_SHIFT; |
477 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
478 | locked += new_len - old_len; |
479 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
480 | return ERR_PTR(-EAGAIN); |
481 | } |
482 | |
483 | if (!may_expand_vm(mm, vma->vm_flags, |
484 | (new_len - old_len) >> PAGE_SHIFT)) |
485 | return ERR_PTR(-ENOMEM); |
486 | |
487 | if (vma->vm_flags & VM_ACCOUNT) { |
488 | unsigned long charged = (new_len - old_len) >> PAGE_SHIFT; |
489 | if (security_vm_enough_memory_mm(mm, charged)) |
490 | return ERR_PTR(-ENOMEM); |
491 | *p = charged; |
492 | } |
493 | |
494 | return vma; |
495 | } |
496 | |
497 | static unsigned long mremap_to(unsigned long addr, unsigned long old_len, |
498 | unsigned long new_addr, unsigned long new_len, bool *locked, |
499 | struct vm_userfaultfd_ctx *uf, |
500 | struct list_head *uf_unmap_early, |
501 | struct list_head *uf_unmap) |
502 | { |
503 | struct mm_struct *mm = current->mm; |
504 | struct vm_area_struct *vma; |
505 | unsigned long ret = -EINVAL; |
506 | unsigned long charged = 0; |
507 | unsigned long map_flags; |
508 | |
509 | if (offset_in_page(new_addr)) |
510 | goto out; |
511 | |
512 | if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len) |
513 | goto out; |
514 | |
515 | /* Ensure the old/new locations do not overlap */ |
516 | if (addr + old_len > new_addr && new_addr + new_len > addr) |
517 | goto out; |
518 | |
519 | /* |
520 | * move_vma() need us to stay 4 maps below the threshold, otherwise |
521 | * it will bail out at the very beginning. |
522 | * That is a problem if we have already unmaped the regions here |
523 | * (new_addr, and old_addr), because userspace will not know the |
524 | * state of the vma's after it gets -ENOMEM. |
525 | * So, to avoid such scenario we can pre-compute if the whole |
526 | * operation has high chances to success map-wise. |
527 | * Worst-scenario case is when both vma's (new_addr and old_addr) get |
528 | * split in 3 before unmaping it. |
529 | * That means 2 more maps (1 for each) to the ones we already hold. |
530 | * Check whether current map count plus 2 still leads us to 4 maps below |
531 | * the threshold, otherwise return -ENOMEM here to be more safe. |
532 | */ |
533 | if ((mm->map_count + 2) >= sysctl_max_map_count - 3) |
534 | return -ENOMEM; |
535 | |
536 | ret = do_munmap(mm, new_addr, new_len, uf_unmap_early); |
537 | if (ret) |
538 | goto out; |
539 | |
540 | if (old_len >= new_len) { |
541 | ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap); |
542 | if (ret && old_len != new_len) |
543 | goto out; |
544 | old_len = new_len; |
545 | } |
546 | |
547 | vma = vma_to_resize(addr, old_len, new_len, &charged); |
548 | if (IS_ERR(vma)) { |
549 | ret = PTR_ERR(vma); |
550 | goto out; |
551 | } |
552 | |
553 | map_flags = MAP_FIXED; |
554 | if (vma->vm_flags & VM_MAYSHARE) |
555 | map_flags |= MAP_SHARED; |
556 | |
557 | ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff + |
558 | ((addr - vma->vm_start) >> PAGE_SHIFT), |
559 | map_flags); |
560 | if (offset_in_page(ret)) |
561 | goto out1; |
562 | |
563 | ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, uf, |
564 | uf_unmap); |
565 | if (!(offset_in_page(ret))) |
566 | goto out; |
567 | out1: |
568 | vm_unacct_memory(charged); |
569 | |
570 | out: |
571 | return ret; |
572 | } |
573 | |
574 | static int vma_expandable(struct vm_area_struct *vma, unsigned long delta) |
575 | { |
576 | unsigned long end = vma->vm_end + delta; |
577 | if (end < vma->vm_end) /* overflow */ |
578 | return 0; |
579 | if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */ |
580 | return 0; |
581 | if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start, |
582 | 0, MAP_FIXED) & ~PAGE_MASK) |
583 | return 0; |
584 | return 1; |
585 | } |
586 | |
587 | /* |
588 | * Expand (or shrink) an existing mapping, potentially moving it at the |
589 | * same time (controlled by the MREMAP_MAYMOVE flag and available VM space) |
590 | * |
591 | * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise |
592 | * This option implies MREMAP_MAYMOVE. |
593 | */ |
594 | SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, |
595 | unsigned long, new_len, unsigned long, flags, |
596 | unsigned long, new_addr) |
597 | { |
598 | struct mm_struct *mm = current->mm; |
599 | struct vm_area_struct *vma; |
600 | unsigned long ret = -EINVAL; |
601 | unsigned long charged = 0; |
602 | bool locked = false; |
603 | bool downgraded = false; |
604 | struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX; |
605 | LIST_HEAD(uf_unmap_early); |
606 | LIST_HEAD(uf_unmap); |
607 | |
608 | if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE)) |
609 | return ret; |
610 | |
611 | if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE)) |
612 | return ret; |
613 | |
614 | if (offset_in_page(addr)) |
615 | return ret; |
616 | |
617 | old_len = PAGE_ALIGN(old_len); |
618 | new_len = PAGE_ALIGN(new_len); |
619 | |
620 | /* |
621 | * We allow a zero old-len as a special case |
622 | * for DOS-emu "duplicate shm area" thing. But |
623 | * a zero new-len is nonsensical. |
624 | */ |
625 | if (!new_len) |
626 | return ret; |
627 | |
628 | if (down_write_killable(¤t->mm->mmap_sem)) |
629 | return -EINTR; |
630 | |
631 | if (flags & MREMAP_FIXED) { |
632 | ret = mremap_to(addr, old_len, new_addr, new_len, |
633 | &locked, &uf, &uf_unmap_early, &uf_unmap); |
634 | goto out; |
635 | } |
636 | |
637 | /* |
638 | * Always allow a shrinking remap: that just unmaps |
639 | * the unnecessary pages.. |
640 | * __do_munmap does all the needed commit accounting, and |
641 | * downgrades mmap_sem to read if so directed. |
642 | */ |
643 | if (old_len >= new_len) { |
644 | int retval; |
645 | |
646 | retval = __do_munmap(mm, addr+new_len, old_len - new_len, |
647 | &uf_unmap, true); |
648 | if (retval < 0 && old_len != new_len) { |
649 | ret = retval; |
650 | goto out; |
651 | /* Returning 1 indicates mmap_sem is downgraded to read. */ |
652 | } else if (retval == 1) |
653 | downgraded = true; |
654 | ret = addr; |
655 | goto out; |
656 | } |
657 | |
658 | /* |
659 | * Ok, we need to grow.. |
660 | */ |
661 | vma = vma_to_resize(addr, old_len, new_len, &charged); |
662 | if (IS_ERR(vma)) { |
663 | ret = PTR_ERR(vma); |
664 | goto out; |
665 | } |
666 | |
667 | /* old_len exactly to the end of the area.. |
668 | */ |
669 | if (old_len == vma->vm_end - addr) { |
670 | /* can we just expand the current mapping? */ |
671 | if (vma_expandable(vma, new_len - old_len)) { |
672 | int pages = (new_len - old_len) >> PAGE_SHIFT; |
673 | |
674 | if (vma_adjust(vma, vma->vm_start, addr + new_len, |
675 | vma->vm_pgoff, NULL)) { |
676 | ret = -ENOMEM; |
677 | goto out; |
678 | } |
679 | |
680 | vm_stat_account(mm, vma->vm_flags, pages); |
681 | if (vma->vm_flags & VM_LOCKED) { |
682 | mm->locked_vm += pages; |
683 | locked = true; |
684 | new_addr = addr; |
685 | } |
686 | ret = addr; |
687 | goto out; |
688 | } |
689 | } |
690 | |
691 | /* |
692 | * We weren't able to just expand or shrink the area, |
693 | * we need to create a new one and move it.. |
694 | */ |
695 | ret = -ENOMEM; |
696 | if (flags & MREMAP_MAYMOVE) { |
697 | unsigned long map_flags = 0; |
698 | if (vma->vm_flags & VM_MAYSHARE) |
699 | map_flags |= MAP_SHARED; |
700 | |
701 | new_addr = get_unmapped_area(vma->vm_file, 0, new_len, |
702 | vma->vm_pgoff + |
703 | ((addr - vma->vm_start) >> PAGE_SHIFT), |
704 | map_flags); |
705 | if (offset_in_page(new_addr)) { |
706 | ret = new_addr; |
707 | goto out; |
708 | } |
709 | |
710 | ret = move_vma(vma, addr, old_len, new_len, new_addr, |
711 | &locked, &uf, &uf_unmap); |
712 | } |
713 | out: |
714 | if (offset_in_page(ret)) { |
715 | vm_unacct_memory(charged); |
716 | locked = 0; |
717 | } |
718 | if (downgraded) |
719 | up_read(¤t->mm->mmap_sem); |
720 | else |
721 | up_write(¤t->mm->mmap_sem); |
722 | if (locked && new_len > old_len) |
723 | mm_populate(new_addr + old_len, new_len - old_len); |
724 | userfaultfd_unmap_complete(mm, &uf_unmap_early); |
725 | mremap_userfaultfd_complete(&uf, addr, new_addr, old_len); |
726 | userfaultfd_unmap_complete(mm, &uf_unmap); |
727 | return ret; |
728 | } |
729 | |