]> git.ipfire.org Git - thirdparty/linux.git/blame - mm/mremap.c
mm/vmstat.c: fix /proc/vmstat format for CONFIG_DEBUG_TLBFLUSH=y CONFIG_SMP=n
[thirdparty/linux.git] / mm / mremap.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
2/*
3 * mm/mremap.c
4 *
5 * (C) Copyright 1996 Linus Torvalds
6 *
046c6884 7 * Address space accounting code <alan@lxorguk.ukuu.org.uk>
1da177e4
LT
8 * (C) Copyright 2002 Red Hat Inc, All Rights Reserved
9 */
10
11#include <linux/mm.h>
12#include <linux/hugetlb.h>
1da177e4 13#include <linux/shm.h>
1ff82995 14#include <linux/ksm.h>
1da177e4
LT
15#include <linux/mman.h>
16#include <linux/swap.h>
c59ede7b 17#include <linux/capability.h>
1da177e4 18#include <linux/fs.h>
6dec97dc 19#include <linux/swapops.h>
1da177e4
LT
20#include <linux/highmem.h>
21#include <linux/security.h>
22#include <linux/syscalls.h>
cddb8a5c 23#include <linux/mmu_notifier.h>
2581d202 24#include <linux/uaccess.h>
4abad2ca 25#include <linux/mm-arch-hooks.h>
72f87654 26#include <linux/userfaultfd_k.h>
1da177e4 27
1da177e4
LT
28#include <asm/cacheflush.h>
29#include <asm/tlbflush.h>
30
ba470de4
RR
31#include "internal.h"
32
7be7a546 33static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
1da177e4
LT
34{
35 pgd_t *pgd;
c2febafc 36 p4d_t *p4d;
1da177e4
LT
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
c2febafc
KS
44 p4d = p4d_offset(pgd, addr);
45 if (p4d_none_or_clear_bad(p4d))
46 return NULL;
47
48 pud = pud_offset(p4d, addr);
1da177e4
LT
49 if (pud_none_or_clear_bad(pud))
50 return NULL;
51
52 pmd = pmd_offset(pud, addr);
37a1c49a 53 if (pmd_none(*pmd))
1da177e4
LT
54 return NULL;
55
7be7a546 56 return pmd;
1da177e4
LT
57}
58
8ac1f832
AA
59static pmd_t *alloc_new_pmd(struct mm_struct *mm, struct vm_area_struct *vma,
60 unsigned long addr)
1da177e4
LT
61{
62 pgd_t *pgd;
c2febafc 63 p4d_t *p4d;
1da177e4 64 pud_t *pud;
c74df32c 65 pmd_t *pmd;
1da177e4
LT
66
67 pgd = pgd_offset(mm, addr);
c2febafc
KS
68 p4d = p4d_alloc(mm, pgd, addr);
69 if (!p4d)
70 return NULL;
71 pud = pud_alloc(mm, p4d, addr);
1da177e4 72 if (!pud)
c74df32c 73 return NULL;
7be7a546 74
1da177e4 75 pmd = pmd_alloc(mm, pud, addr);
57a8f0cd 76 if (!pmd)
c74df32c 77 return NULL;
7be7a546 78
8ac1f832 79 VM_BUG_ON(pmd_trans_huge(*pmd));
c74df32c 80
7be7a546 81 return pmd;
1da177e4
LT
82}
83
1d069b7d
HD
84static 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
92static 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
6dec97dc
CG
100static 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);
6dec97dc
CG
111#endif
112 return pte;
113}
114
7be7a546
HD
115static 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,
eb66ae03 118 unsigned long new_addr, bool need_rmap_locks)
1da177e4 119{
1da177e4 120 struct mm_struct *mm = vma->vm_mm;
7be7a546 121 pte_t *old_pte, *new_pte, pte;
4c21e2f2 122 spinlock_t *old_ptl, *new_ptl;
5d190420
AL
123 bool force_flush = false;
124 unsigned long len = old_end - old_addr;
1da177e4 125
38a76013 126 /*
c8c06efa 127 * When need_rmap_locks is true, we take the i_mmap_rwsem and anon_vma
38a76013
ML
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 */
1d069b7d
HD
144 if (need_rmap_locks)
145 take_rmap_locks(vma);
1da177e4 146
4c21e2f2
HD
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 */
c74df32c 151 old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
ece0e2b6 152 new_pte = pte_offset_map(new_pmd, new_addr);
4c21e2f2
HD
153 new_ptl = pte_lockptr(mm, new_pmd);
154 if (new_ptl != old_ptl)
f20dc5f7 155 spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
3ea27719 156 flush_tlb_batched_pending(vma->vm_mm);
6606c3e0 157 arch_enter_lazy_mmu_mode();
7be7a546
HD
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;
5d190420 163
a2ce2666 164 pte = ptep_get_and_clear(mm, old_addr, old_pte);
5d190420 165 /*
eb66ae03 166 * If we are remapping a valid PTE, make sure
a2ce2666 167 * to flush TLB before we drop the PTL for the
eb66ae03 168 * PTE.
a2ce2666 169 *
eb66ae03
LT
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.
5d190420 175 */
eb66ae03 176 if (pte_present(pte))
5d190420 177 force_flush = true;
7be7a546 178 pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
6dec97dc
CG
179 pte = move_soft_dirty_pte(pte);
180 set_pte_at(mm, new_addr, new_pte, pte);
1da177e4 181 }
7be7a546 182
6606c3e0 183 arch_leave_lazy_mmu_mode();
eb66ae03
LT
184 if (force_flush)
185 flush_tlb_range(vma, old_end - len, old_end);
4c21e2f2
HD
186 if (new_ptl != old_ptl)
187 spin_unlock(new_ptl);
ece0e2b6 188 pte_unmap(new_pte - 1);
c74df32c 189 pte_unmap_unlock(old_pte - 1, old_ptl);
1d069b7d
HD
190 if (need_rmap_locks)
191 drop_rmap_locks(vma);
1da177e4
LT
192}
193
2c91bd4a
JFG
194#ifdef CONFIG_HAVE_MOVE_PMD
195static 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
b6a2fea3 240unsigned long move_page_tables(struct vm_area_struct *vma,
1da177e4 241 unsigned long old_addr, struct vm_area_struct *new_vma,
38a76013
ML
242 unsigned long new_addr, unsigned long len,
243 bool need_rmap_locks)
1da177e4 244{
7be7a546 245 unsigned long extent, next, old_end;
ac46d4f3 246 struct mmu_notifier_range range;
7be7a546 247 pmd_t *old_pmd, *new_pmd;
1da177e4 248
7be7a546
HD
249 old_end = old_addr + len;
250 flush_cache_range(vma, old_addr, old_end);
1da177e4 251
ac46d4f3
JG
252 mmu_notifier_range_init(&range, vma->vm_mm, old_addr, old_end);
253 mmu_notifier_invalidate_range_start(&range);
7b6efc2b 254
7be7a546 255 for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
1da177e4 256 cond_resched();
7be7a546 257 next = (old_addr + PMD_SIZE) & PMD_MASK;
ebed4846 258 /* even if next overflowed, extent below will be ok */
7be7a546 259 extent = next - old_addr;
ebed4846
AA
260 if (extent > old_end - old_addr)
261 extent = old_end - old_addr;
7be7a546
HD
262 old_pmd = get_old_pmd(vma->vm_mm, old_addr);
263 if (!old_pmd)
264 continue;
8ac1f832 265 new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
7be7a546
HD
266 if (!new_pmd)
267 break;
84c3fc4e 268 if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) {
dd18dbc2 269 if (extent == HPAGE_PMD_SIZE) {
4b471e88 270 bool moved;
dd18dbc2
KS
271 /* See comment in move_ptes() */
272 if (need_rmap_locks)
1d069b7d 273 take_rmap_locks(vma);
bf8616d5 274 moved = move_huge_pmd(vma, old_addr, new_addr,
eb66ae03 275 old_end, old_pmd, new_pmd);
dd18dbc2 276 if (need_rmap_locks)
1d069b7d 277 drop_rmap_locks(vma);
5d190420 278 if (moved)
4b471e88 279 continue;
dd18dbc2 280 }
4b471e88 281 split_huge_pmd(vma, old_pmd, old_addr);
337d9abf 282 if (pmd_trans_unstable(old_pmd))
6b9116a6 283 continue;
2c91bd4a
JFG
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
37a1c49a 301 }
2c91bd4a 302
4cf58924 303 if (pte_alloc(new_vma->vm_mm, new_pmd))
37a1c49a 304 break;
7be7a546
HD
305 next = (new_addr + PMD_SIZE) & PMD_MASK;
306 if (extent > next - new_addr)
307 extent = next - new_addr;
5d190420 308 move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
eb66ae03 309 new_pmd, new_addr, need_rmap_locks);
1da177e4 310 }
7b6efc2b 311
ac46d4f3 312 mmu_notifier_invalidate_range_end(&range);
7be7a546
HD
313
314 return len + old_addr - old_end; /* how much done */
1da177e4
LT
315}
316
317static unsigned long move_vma(struct vm_area_struct *vma,
318 unsigned long old_addr, unsigned long old_len,
72f87654 319 unsigned long new_len, unsigned long new_addr,
897ab3e0
MR
320 bool *locked, struct vm_userfaultfd_ctx *uf,
321 struct list_head *uf_unmap)
1da177e4
LT
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;
365e9c87 329 unsigned long hiwater_vm;
1da177e4 330 int split = 0;
7103ad32 331 int err;
38a76013 332 bool need_rmap_locks;
1da177e4
LT
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
1ff82995
HD
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 */
7103ad32
HD
348 err = ksm_madvise(vma, old_addr, old_addr + old_len,
349 MADV_UNMERGEABLE, &vm_flags);
350 if (err)
351 return err;
1ff82995 352
1da177e4 353 new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
38a76013
ML
354 new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
355 &need_rmap_locks);
1da177e4
LT
356 if (!new_vma)
357 return -ENOMEM;
358
38a76013
ML
359 moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
360 need_rmap_locks);
1da177e4 361 if (moved_len < old_len) {
df1eab30 362 err = -ENOMEM;
5477e70a
ON
363 } else if (vma->vm_ops && vma->vm_ops->mremap) {
364 err = vma->vm_ops->mremap(new_vma);
df1eab30
ON
365 }
366
367 if (unlikely(err)) {
1da177e4
LT
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 */
38a76013
ML
373 move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
374 true);
1da177e4
LT
375 vma = new_vma;
376 old_len = new_len;
377 old_addr = new_addr;
df1eab30 378 new_addr = err;
4abad2ca 379 } else {
72f87654 380 mremap_userfaultfd_prep(new_vma, uf);
4abad2ca
LD
381 arch_remap(mm, old_addr, old_addr + old_len,
382 new_addr, new_addr + new_len);
b2edffdd 383 }
1da177e4
LT
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
71799062 394 /*
365e9c87
HD
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().
71799062 402 */
365e9c87 403 hiwater_vm = mm->hiwater_vm;
84638335 404 vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
71799062 405
d9fe4fab
TK
406 /* Tell pfnmap has moved from this vma */
407 if (unlikely(vma->vm_flags & VM_PFNMAP))
408 untrack_pfn_moved(vma);
409
897ab3e0 410 if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) {
1da177e4
LT
411 /* OOM: unable to split vma, just get accounts right */
412 vm_unacct_memory(excess >> PAGE_SHIFT);
413 excess = 0;
414 }
365e9c87 415 mm->hiwater_vm = hiwater_vm;
1da177e4
LT
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
1da177e4
LT
424 if (vm_flags & VM_LOCKED) {
425 mm->locked_vm += new_len >> PAGE_SHIFT;
81909b84 426 *locked = true;
1da177e4
LT
427 }
428
429 return new_addr;
430}
431
54f5de70
AV
432static 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);
1d391686 437 unsigned long pgoff;
54f5de70
AV
438
439 if (!vma || vma->vm_start > addr)
6cd57613 440 return ERR_PTR(-EFAULT);
54f5de70 441
dba58d3b
MK
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
54f5de70 455 if (is_vm_hugetlb_page(vma))
6cd57613 456 return ERR_PTR(-EINVAL);
54f5de70
AV
457
458 /* We can't remap across vm area boundaries */
459 if (old_len > vma->vm_end - addr)
6cd57613 460 return ERR_PTR(-EFAULT);
54f5de70 461
1d391686
ON
462 if (new_len == old_len)
463 return vma;
464
982134ba 465 /* Need to be careful about a growing mapping */
1d391686
ON
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);
54f5de70
AV
473
474 if (vma->vm_flags & VM_LOCKED) {
475 unsigned long locked, lock_limit;
476 locked = mm->locked_vm << PAGE_SHIFT;
59e99e5b 477 lock_limit = rlimit(RLIMIT_MEMLOCK);
54f5de70
AV
478 locked += new_len - old_len;
479 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
6cd57613 480 return ERR_PTR(-EAGAIN);
54f5de70
AV
481 }
482
84638335
KK
483 if (!may_expand_vm(mm, vma->vm_flags,
484 (new_len - old_len) >> PAGE_SHIFT))
6cd57613 485 return ERR_PTR(-ENOMEM);
54f5de70
AV
486
487 if (vma->vm_flags & VM_ACCOUNT) {
488 unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
191c5424 489 if (security_vm_enough_memory_mm(mm, charged))
6cd57613 490 return ERR_PTR(-ENOMEM);
54f5de70
AV
491 *p = charged;
492 }
493
494 return vma;
54f5de70
AV
495}
496
81909b84 497static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
72f87654 498 unsigned long new_addr, unsigned long new_len, bool *locked,
897ab3e0 499 struct vm_userfaultfd_ctx *uf,
b2282371 500 struct list_head *uf_unmap_early,
897ab3e0 501 struct list_head *uf_unmap)
ecc1a899
AV
502{
503 struct mm_struct *mm = current->mm;
504 struct vm_area_struct *vma;
505 unsigned long ret = -EINVAL;
506 unsigned long charged = 0;
097eed10 507 unsigned long map_flags;
ecc1a899 508
f19cb115 509 if (offset_in_page(new_addr))
ecc1a899
AV
510 goto out;
511
512 if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
513 goto out;
514
9943242c
ON
515 /* Ensure the old/new locations do not overlap */
516 if (addr + old_len > new_addr && new_addr + new_len > addr)
ecc1a899
AV
517 goto out;
518
ea2c3f6f
OS
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
b2282371 536 ret = do_munmap(mm, new_addr, new_len, uf_unmap_early);
ecc1a899
AV
537 if (ret)
538 goto out;
539
540 if (old_len >= new_len) {
897ab3e0 541 ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap);
ecc1a899
AV
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
097eed10
AV
553 map_flags = MAP_FIXED;
554 if (vma->vm_flags & VM_MAYSHARE)
555 map_flags |= MAP_SHARED;
9206de95 556
097eed10
AV
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);
f19cb115 560 if (offset_in_page(ret))
097eed10
AV
561 goto out1;
562
897ab3e0
MR
563 ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, uf,
564 uf_unmap);
f19cb115 565 if (!(offset_in_page(ret)))
097eed10
AV
566 goto out;
567out1:
568 vm_unacct_memory(charged);
ecc1a899
AV
569
570out:
571 return ret;
572}
573
1a0ef85f
AV
574static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
575{
f106af4e 576 unsigned long end = vma->vm_end + delta;
9206de95 577 if (end < vma->vm_end) /* overflow */
f106af4e 578 return 0;
9206de95 579 if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
f106af4e
AV
580 return 0;
581 if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
582 0, MAP_FIXED) & ~PAGE_MASK)
1a0ef85f 583 return 0;
1a0ef85f
AV
584 return 1;
585}
586
1da177e4
LT
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 */
63a81db1
AV
594SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
595 unsigned long, new_len, unsigned long, flags,
596 unsigned long, new_addr)
1da177e4 597{
d0de32d9 598 struct mm_struct *mm = current->mm;
1da177e4
LT
599 struct vm_area_struct *vma;
600 unsigned long ret = -EINVAL;
601 unsigned long charged = 0;
81909b84 602 bool locked = false;
85a06835 603 bool downgraded = false;
72f87654 604 struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX;
b2282371 605 LIST_HEAD(uf_unmap_early);
897ab3e0 606 LIST_HEAD(uf_unmap);
1da177e4
LT
607
608 if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
9a2458a6
RV
609 return ret;
610
611 if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
612 return ret;
1da177e4 613
f19cb115 614 if (offset_in_page(addr))
9a2458a6 615 return ret;
1da177e4
LT
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)
9a2458a6
RV
626 return ret;
627
dc0ef0df
MH
628 if (down_write_killable(&current->mm->mmap_sem))
629 return -EINTR;
1da177e4 630
1da177e4 631 if (flags & MREMAP_FIXED) {
9a2458a6 632 ret = mremap_to(addr, old_len, new_addr, new_len,
b2282371 633 &locked, &uf, &uf_unmap_early, &uf_unmap);
ecc1a899 634 goto out;
1da177e4
LT
635 }
636
637 /*
638 * Always allow a shrinking remap: that just unmaps
639 * the unnecessary pages..
85a06835
YS
640 * __do_munmap does all the needed commit accounting, and
641 * downgrades mmap_sem to read if so directed.
1da177e4
LT
642 */
643 if (old_len >= new_len) {
85a06835
YS
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;
1da177e4 650 goto out;
85a06835
YS
651 /* Returning 1 indicates mmap_sem is downgraded to read. */
652 } else if (retval == 1)
653 downgraded = true;
1da177e4 654 ret = addr;
ecc1a899 655 goto out;
1da177e4
LT
656 }
657
658 /*
ecc1a899 659 * Ok, we need to grow..
1da177e4 660 */
54f5de70
AV
661 vma = vma_to_resize(addr, old_len, new_len, &charged);
662 if (IS_ERR(vma)) {
663 ret = PTR_ERR(vma);
1da177e4 664 goto out;
119f657c 665 }
1da177e4 666
1da177e4 667 /* old_len exactly to the end of the area..
1da177e4 668 */
ecc1a899 669 if (old_len == vma->vm_end - addr) {
1da177e4 670 /* can we just expand the current mapping? */
1a0ef85f 671 if (vma_expandable(vma, new_len - old_len)) {
1da177e4
LT
672 int pages = (new_len - old_len) >> PAGE_SHIFT;
673
5beb4930
RR
674 if (vma_adjust(vma, vma->vm_start, addr + new_len,
675 vma->vm_pgoff, NULL)) {
676 ret = -ENOMEM;
677 goto out;
678 }
1da177e4 679
84638335 680 vm_stat_account(mm, vma->vm_flags, pages);
1da177e4 681 if (vma->vm_flags & VM_LOCKED) {
d0de32d9 682 mm->locked_vm += pages;
81909b84
ML
683 locked = true;
684 new_addr = addr;
1da177e4
LT
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) {
ecc1a899
AV
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,
93587414
AV
702 vma->vm_pgoff +
703 ((addr - vma->vm_start) >> PAGE_SHIFT),
704 map_flags);
f19cb115 705 if (offset_in_page(new_addr)) {
ecc1a899
AV
706 ret = new_addr;
707 goto out;
1da177e4 708 }
ecc1a899 709
72f87654 710 ret = move_vma(vma, addr, old_len, new_len, new_addr,
897ab3e0 711 &locked, &uf, &uf_unmap);
1da177e4
LT
712 }
713out:
f19cb115 714 if (offset_in_page(ret)) {
1da177e4 715 vm_unacct_memory(charged);
d456fb9e
ON
716 locked = 0;
717 }
85a06835
YS
718 if (downgraded)
719 up_read(&current->mm->mmap_sem);
720 else
721 up_write(&current->mm->mmap_sem);
81909b84
ML
722 if (locked && new_len > old_len)
723 mm_populate(new_addr + old_len, new_len - old_len);
b2282371 724 userfaultfd_unmap_complete(mm, &uf_unmap_early);
90794bf1 725 mremap_userfaultfd_complete(&uf, addr, new_addr, old_len);
897ab3e0 726 userfaultfd_unmap_complete(mm, &uf_unmap);
1da177e4
LT
727 return ret;
728}