Branch data Line data Source code
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 : 2481 : static pmd_t *get_old_pmd(struct mm_struct *mm, unsigned long addr)
34 : : {
35 : 2481 : pgd_t *pgd;
36 : 2481 : p4d_t *p4d;
37 : 2481 : pud_t *pud;
38 : 2481 : pmd_t *pmd;
39 : :
40 : 2481 : pgd = pgd_offset(mm, addr);
41 [ + - ]: 2481 : if (pgd_none_or_clear_bad(pgd))
42 : : return NULL;
43 : :
44 : 2481 : p4d = p4d_offset(pgd, addr);
45 [ + - ]: 2481 : if (p4d_none_or_clear_bad(p4d))
46 : 0 : return NULL;
47 : :
48 : 2481 : pud = pud_offset(p4d, addr);
49 [ + - ]: 2481 : if (pud_none_or_clear_bad(pud))
50 : : return NULL;
51 : :
52 [ + - ]: 2481 : pmd = pmd_offset(pud, addr);
53 [ - + ]: 2481 : if (pmd_none(*pmd))
54 : 0 : 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 : 2481 : 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 : 2481 : return pte;
113 : : }
114 : :
115 : 2481 : 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 : 2481 : struct mm_struct *mm = vma->vm_mm;
121 : 2481 : pte_t *old_pte, *new_pte, pte;
122 : 2481 : spinlock_t *old_ptl, *new_ptl;
123 : 2481 : bool force_flush = false;
124 : 2481 : 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 [ - + ]: 2481 : if (need_rmap_locks)
145 : 0 : 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 [ + - ]: 4962 : old_pte = pte_offset_map_lock(mm, old_pmd, old_addr, &old_ptl);
152 [ + - ]: 2481 : new_pte = pte_offset_map(new_pmd, new_addr);
153 [ + - ]: 2481 : new_ptl = pte_lockptr(mm, new_pmd);
154 [ + + ]: 2481 : if (new_ptl != old_ptl)
155 : 2480 : spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
156 : 2481 : flush_tlb_batched_pending(vma->vm_mm);
157 : 2481 : arch_enter_lazy_mmu_mode();
158 : :
159 [ + + ]: 7443 : for (; old_addr < old_end; old_pte++, old_addr += PAGE_SIZE,
160 : 2481 : new_pte++, new_addr += PAGE_SIZE) {
161 [ - + ]: 2481 : if (pte_none(*old_pte))
162 : 0 : continue;
163 : :
164 : 2481 : 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 [ + - ]: 2481 : if (pte_present(pte))
177 : 2481 : force_flush = true;
178 : 2481 : pte = move_pte(pte, new_vma->vm_page_prot, old_addr, new_addr);
179 : 2481 : pte = move_soft_dirty_pte(pte);
180 : 2481 : set_pte_at(mm, new_addr, new_pte, pte);
181 : : }
182 : :
183 : 2481 : arch_leave_lazy_mmu_mode();
184 [ + - ]: 2481 : if (force_flush)
185 [ - + ]: 2481 : flush_tlb_range(vma, old_end - len, old_end);
186 [ + + ]: 2481 : if (new_ptl != old_ptl)
187 : 2480 : spin_unlock(new_ptl);
188 : 2481 : pte_unmap(new_pte - 1);
189 : 2481 : pte_unmap_unlock(old_pte - 1, old_ptl);
190 [ - + ]: 2481 : if (need_rmap_locks)
191 : 0 : drop_rmap_locks(vma);
192 : 2481 : }
193 : :
194 : : #ifdef CONFIG_HAVE_MOVE_PMD
195 : 0 : 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 : 0 : spinlock_t *old_ptl, *new_ptl;
200 : 0 : struct mm_struct *mm = vma->vm_mm;
201 : 0 : pmd_t pmd;
202 : :
203 [ # # # # ]: 0 : if ((old_addr & ~PMD_MASK) || (new_addr & ~PMD_MASK)
204 [ # # ]: 0 : || 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 [ # # # # ]: 0 : 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 : 0 : old_ptl = pmd_lock(vma->vm_mm, old_pmd);
219 [ # # ]: 0 : new_ptl = pmd_lockptr(mm, new_pmd);
220 [ # # ]: 0 : if (new_ptl != old_ptl)
221 : 0 : spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
222 : :
223 : : /* Clear the pmd */
224 : 0 : pmd = *old_pmd;
225 [ # # ]: 0 : pmd_clear(old_pmd);
226 : :
227 : 0 : VM_BUG_ON(!pmd_none(*new_pmd));
228 : :
229 : : /* Set the new pmd */
230 [ # # ]: 0 : set_pmd_at(mm, new_addr, new_pmd, pmd);
231 [ # # ]: 0 : flush_tlb_range(vma, old_addr, old_addr + PMD_SIZE);
232 [ # # ]: 0 : if (new_ptl != old_ptl)
233 : 0 : spin_unlock(new_ptl);
234 : 0 : spin_unlock(old_ptl);
235 : :
236 : 0 : return true;
237 : : }
238 : : #endif
239 : :
240 : 2481 : 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 : 2481 : unsigned long extent, next, old_end;
246 : 2481 : struct mmu_notifier_range range;
247 : 2481 : pmd_t *old_pmd, *new_pmd;
248 : :
249 : 2481 : old_end = old_addr + len;
250 : 2481 : flush_cache_range(vma, old_addr, old_end);
251 : :
252 : 2481 : mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, vma, vma->vm_mm,
253 : : old_addr, old_end);
254 : 2481 : mmu_notifier_invalidate_range_start(&range);
255 : :
256 [ + + ]: 7443 : for (; old_addr < old_end; old_addr += extent, new_addr += extent) {
257 : 2481 : cond_resched();
258 : 2481 : next = (old_addr + PMD_SIZE) & PMD_MASK;
259 : : /* even if next overflowed, extent below will be ok */
260 : 2481 : extent = next - old_addr;
261 : 2481 : if (extent > old_end - old_addr)
262 : : extent = old_end - old_addr;
263 : 2481 : old_pmd = get_old_pmd(vma->vm_mm, old_addr);
264 [ - + ]: 2481 : if (!old_pmd)
265 : 0 : continue;
266 : 2481 : new_pmd = alloc_new_pmd(vma->vm_mm, vma, new_addr);
267 [ + - ]: 2481 : if (!new_pmd)
268 : : break;
269 [ - + ]: 2481 : if (is_swap_pmd(*old_pmd) || pmd_trans_huge(*old_pmd)) {
270 : : if (extent == HPAGE_PMD_SIZE) {
271 : : bool moved;
272 : : /* See comment in move_ptes() */
273 : : if (need_rmap_locks)
274 : : take_rmap_locks(vma);
275 : : moved = move_huge_pmd(vma, old_addr, new_addr,
276 : : old_end, old_pmd, new_pmd);
277 : : if (need_rmap_locks)
278 : : drop_rmap_locks(vma);
279 : : if (moved)
280 : : continue;
281 : : }
282 : : split_huge_pmd(vma, old_pmd, old_addr);
283 : : if (pmd_trans_unstable(old_pmd))
284 : : continue;
285 [ - + ]: 2481 : } else if (extent == PMD_SIZE) {
286 : : #ifdef CONFIG_HAVE_MOVE_PMD
287 : : /*
288 : : * If the extent is PMD-sized, try to speed the move by
289 : : * moving at the PMD level if possible.
290 : : */
291 : 0 : bool moved;
292 : :
293 [ # # ]: 0 : if (need_rmap_locks)
294 : 0 : take_rmap_locks(vma);
295 : 0 : moved = move_normal_pmd(vma, old_addr, new_addr,
296 : : old_end, old_pmd, new_pmd);
297 [ # # ]: 0 : if (need_rmap_locks)
298 : 0 : drop_rmap_locks(vma);
299 [ # # ]: 0 : if (moved)
300 : 0 : continue;
301 : : #endif
302 : : }
303 : :
304 [ + + + - ]: 2481 : if (pte_alloc(new_vma->vm_mm, new_pmd))
305 : : break;
306 : 2481 : next = (new_addr + PMD_SIZE) & PMD_MASK;
307 : 2481 : if (extent > next - new_addr)
308 : : extent = next - new_addr;
309 : 2481 : move_ptes(vma, old_pmd, old_addr, old_addr + extent, new_vma,
310 : : new_pmd, new_addr, need_rmap_locks);
311 : : }
312 : :
313 : 2481 : mmu_notifier_invalidate_range_end(&range);
314 : :
315 : 2481 : return len + old_addr - old_end; /* how much done */
316 : : }
317 : :
318 : : static unsigned long move_vma(struct vm_area_struct *vma,
319 : : unsigned long old_addr, unsigned long old_len,
320 : : unsigned long new_len, unsigned long new_addr,
321 : : bool *locked, struct vm_userfaultfd_ctx *uf,
322 : : struct list_head *uf_unmap)
323 : : {
324 : : struct mm_struct *mm = vma->vm_mm;
325 : : struct vm_area_struct *new_vma;
326 : : unsigned long vm_flags = vma->vm_flags;
327 : : unsigned long new_pgoff;
328 : : unsigned long moved_len;
329 : : unsigned long excess = 0;
330 : : unsigned long hiwater_vm;
331 : : int split = 0;
332 : : int err;
333 : : bool need_rmap_locks;
334 : :
335 : : /*
336 : : * We'd prefer to avoid failure later on in do_munmap:
337 : : * which may split one vma into three before unmapping.
338 : : */
339 : : if (mm->map_count >= sysctl_max_map_count - 3)
340 : : return -ENOMEM;
341 : :
342 : : /*
343 : : * Advise KSM to break any KSM pages in the area to be moved:
344 : : * it would be confusing if they were to turn up at the new
345 : : * location, where they happen to coincide with different KSM
346 : : * pages recently unmapped. But leave vma->vm_flags as it was,
347 : : * so KSM can come around to merge on vma and new_vma afterwards.
348 : : */
349 : : err = ksm_madvise(vma, old_addr, old_addr + old_len,
350 : : MADV_UNMERGEABLE, &vm_flags);
351 : : if (err)
352 : : return err;
353 : :
354 : : new_pgoff = vma->vm_pgoff + ((old_addr - vma->vm_start) >> PAGE_SHIFT);
355 : : new_vma = copy_vma(&vma, new_addr, new_len, new_pgoff,
356 : : &need_rmap_locks);
357 : : if (!new_vma)
358 : : return -ENOMEM;
359 : :
360 : : moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
361 : : need_rmap_locks);
362 : : if (moved_len < old_len) {
363 : : err = -ENOMEM;
364 : : } else if (vma->vm_ops && vma->vm_ops->mremap) {
365 : : err = vma->vm_ops->mremap(new_vma);
366 : : }
367 : :
368 : : if (unlikely(err)) {
369 : : /*
370 : : * On error, move entries back from new area to old,
371 : : * which will succeed since page tables still there,
372 : : * and then proceed to unmap new area instead of old.
373 : : */
374 : : move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
375 : : true);
376 : : vma = new_vma;
377 : : old_len = new_len;
378 : : old_addr = new_addr;
379 : : new_addr = err;
380 : : } else {
381 : : mremap_userfaultfd_prep(new_vma, uf);
382 : : arch_remap(mm, old_addr, old_addr + old_len,
383 : : new_addr, new_addr + new_len);
384 : : }
385 : :
386 : : /* Conceal VM_ACCOUNT so old reservation is not undone */
387 : : if (vm_flags & VM_ACCOUNT) {
388 : : vma->vm_flags &= ~VM_ACCOUNT;
389 : : excess = vma->vm_end - vma->vm_start - old_len;
390 : : if (old_addr > vma->vm_start &&
391 : : old_addr + old_len < vma->vm_end)
392 : : split = 1;
393 : : }
394 : :
395 : : /*
396 : : * If we failed to move page tables we still do total_vm increment
397 : : * since do_munmap() will decrement it by old_len == new_len.
398 : : *
399 : : * Since total_vm is about to be raised artificially high for a
400 : : * moment, we need to restore high watermark afterwards: if stats
401 : : * are taken meanwhile, total_vm and hiwater_vm appear too high.
402 : : * If this were a serious issue, we'd add a flag to do_munmap().
403 : : */
404 : : hiwater_vm = mm->hiwater_vm;
405 : : vm_stat_account(mm, vma->vm_flags, new_len >> PAGE_SHIFT);
406 : :
407 : : /* Tell pfnmap has moved from this vma */
408 : : if (unlikely(vma->vm_flags & VM_PFNMAP))
409 : : untrack_pfn_moved(vma);
410 : :
411 : : if (do_munmap(mm, old_addr, old_len, uf_unmap) < 0) {
412 : : /* OOM: unable to split vma, just get accounts right */
413 : : vm_unacct_memory(excess >> PAGE_SHIFT);
414 : : excess = 0;
415 : : }
416 : : mm->hiwater_vm = hiwater_vm;
417 : :
418 : : /* Restore VM_ACCOUNT if one or two pieces of vma left */
419 : : if (excess) {
420 : : vma->vm_flags |= VM_ACCOUNT;
421 : : if (split)
422 : : vma->vm_next->vm_flags |= VM_ACCOUNT;
423 : : }
424 : :
425 : : if (vm_flags & VM_LOCKED) {
426 : : mm->locked_vm += new_len >> PAGE_SHIFT;
427 : : *locked = true;
428 : : }
429 : :
430 : : return new_addr;
431 : : }
432 : :
433 : 0 : static struct vm_area_struct *vma_to_resize(unsigned long addr,
434 : : unsigned long old_len, unsigned long new_len, unsigned long *p)
435 : : {
436 : 0 : struct mm_struct *mm = current->mm;
437 : 0 : struct vm_area_struct *vma = find_vma(mm, addr);
438 : 0 : unsigned long pgoff;
439 : :
440 [ # # # # ]: 0 : if (!vma || vma->vm_start > addr)
441 : : return ERR_PTR(-EFAULT);
442 : :
443 : : /*
444 : : * !old_len is a special case where an attempt is made to 'duplicate'
445 : : * a mapping. This makes no sense for private mappings as it will
446 : : * instead create a fresh/new mapping unrelated to the original. This
447 : : * is contrary to the basic idea of mremap which creates new mappings
448 : : * based on the original. There are no known use cases for this
449 : : * behavior. As a result, fail such attempts.
450 : : */
451 [ # # # # ]: 0 : if (!old_len && !(vma->vm_flags & (VM_SHARED | VM_MAYSHARE))) {
452 [ # # ]: 0 : pr_warn_once("%s (%d): attempted to duplicate a private mapping with mremap. This is not supported.\n", current->comm, current->pid);
453 : 0 : return ERR_PTR(-EINVAL);
454 : : }
455 : :
456 [ # # ]: 0 : if (is_vm_hugetlb_page(vma))
457 : : return ERR_PTR(-EINVAL);
458 : :
459 : : /* We can't remap across vm area boundaries */
460 [ # # ]: 0 : if (old_len > vma->vm_end - addr)
461 : : return ERR_PTR(-EFAULT);
462 : :
463 [ # # ]: 0 : if (new_len == old_len)
464 : : return vma;
465 : :
466 : : /* Need to be careful about a growing mapping */
467 : 0 : pgoff = (addr - vma->vm_start) >> PAGE_SHIFT;
468 : 0 : pgoff += vma->vm_pgoff;
469 [ # # ]: 0 : if (pgoff + (new_len >> PAGE_SHIFT) < pgoff)
470 : : return ERR_PTR(-EINVAL);
471 : :
472 [ # # ]: 0 : if (vma->vm_flags & (VM_DONTEXPAND | VM_PFNMAP))
473 : : return ERR_PTR(-EFAULT);
474 : :
475 [ # # ]: 0 : if (vma->vm_flags & VM_LOCKED) {
476 : 0 : unsigned long locked, lock_limit;
477 : 0 : locked = mm->locked_vm << PAGE_SHIFT;
478 [ # # ]: 0 : lock_limit = rlimit(RLIMIT_MEMLOCK);
479 : 0 : locked += new_len - old_len;
480 [ # # # # ]: 0 : if (locked > lock_limit && !capable(CAP_IPC_LOCK))
481 : : return ERR_PTR(-EAGAIN);
482 : : }
483 : :
484 [ # # ]: 0 : if (!may_expand_vm(mm, vma->vm_flags,
485 : 0 : (new_len - old_len) >> PAGE_SHIFT))
486 : : return ERR_PTR(-ENOMEM);
487 : :
488 [ # # ]: 0 : if (vma->vm_flags & VM_ACCOUNT) {
489 : 0 : unsigned long charged = (new_len - old_len) >> PAGE_SHIFT;
490 [ # # ]: 0 : if (security_vm_enough_memory_mm(mm, charged))
491 : : return ERR_PTR(-ENOMEM);
492 : 0 : *p = charged;
493 : : }
494 : :
495 : : return vma;
496 : : }
497 : :
498 : : static unsigned long mremap_to(unsigned long addr, unsigned long old_len,
499 : : unsigned long new_addr, unsigned long new_len, bool *locked,
500 : : struct vm_userfaultfd_ctx *uf,
501 : : struct list_head *uf_unmap_early,
502 : : struct list_head *uf_unmap)
503 : : {
504 : : struct mm_struct *mm = current->mm;
505 : : struct vm_area_struct *vma;
506 : : unsigned long ret = -EINVAL;
507 : : unsigned long charged = 0;
508 : : unsigned long map_flags;
509 : :
510 : : if (offset_in_page(new_addr))
511 : : goto out;
512 : :
513 : : if (new_len > TASK_SIZE || new_addr > TASK_SIZE - new_len)
514 : : goto out;
515 : :
516 : : /* Ensure the old/new locations do not overlap */
517 : : if (addr + old_len > new_addr && new_addr + new_len > addr)
518 : : goto out;
519 : :
520 : : /*
521 : : * move_vma() need us to stay 4 maps below the threshold, otherwise
522 : : * it will bail out at the very beginning.
523 : : * That is a problem if we have already unmaped the regions here
524 : : * (new_addr, and old_addr), because userspace will not know the
525 : : * state of the vma's after it gets -ENOMEM.
526 : : * So, to avoid such scenario we can pre-compute if the whole
527 : : * operation has high chances to success map-wise.
528 : : * Worst-scenario case is when both vma's (new_addr and old_addr) get
529 : : * split in 3 before unmaping it.
530 : : * That means 2 more maps (1 for each) to the ones we already hold.
531 : : * Check whether current map count plus 2 still leads us to 4 maps below
532 : : * the threshold, otherwise return -ENOMEM here to be more safe.
533 : : */
534 : : if ((mm->map_count + 2) >= sysctl_max_map_count - 3)
535 : : return -ENOMEM;
536 : :
537 : : ret = do_munmap(mm, new_addr, new_len, uf_unmap_early);
538 : : if (ret)
539 : : goto out;
540 : :
541 : : if (old_len >= new_len) {
542 : : ret = do_munmap(mm, addr+new_len, old_len - new_len, uf_unmap);
543 : : if (ret && old_len != new_len)
544 : : goto out;
545 : : old_len = new_len;
546 : : }
547 : :
548 : : vma = vma_to_resize(addr, old_len, new_len, &charged);
549 : : if (IS_ERR(vma)) {
550 : : ret = PTR_ERR(vma);
551 : : goto out;
552 : : }
553 : :
554 : : map_flags = MAP_FIXED;
555 : : if (vma->vm_flags & VM_MAYSHARE)
556 : : map_flags |= MAP_SHARED;
557 : :
558 : : ret = get_unmapped_area(vma->vm_file, new_addr, new_len, vma->vm_pgoff +
559 : : ((addr - vma->vm_start) >> PAGE_SHIFT),
560 : : map_flags);
561 : : if (IS_ERR_VALUE(ret))
562 : : goto out1;
563 : :
564 : : ret = move_vma(vma, addr, old_len, new_len, new_addr, locked, uf,
565 : : uf_unmap);
566 : : if (!(offset_in_page(ret)))
567 : : goto out;
568 : : out1:
569 : : vm_unacct_memory(charged);
570 : :
571 : : out:
572 : : return ret;
573 : : }
574 : :
575 : 0 : static int vma_expandable(struct vm_area_struct *vma, unsigned long delta)
576 : : {
577 : 0 : unsigned long end = vma->vm_end + delta;
578 [ # # ]: 0 : if (end < vma->vm_end) /* overflow */
579 : : return 0;
580 [ # # # # ]: 0 : if (vma->vm_next && vma->vm_next->vm_start < end) /* intersection */
581 : : return 0;
582 : 0 : if (get_unmapped_area(NULL, vma->vm_start, end - vma->vm_start,
583 [ # # ]: 0 : 0, MAP_FIXED) & ~PAGE_MASK)
584 : 0 : return 0;
585 : : return 1;
586 : : }
587 : :
588 : : /*
589 : : * Expand (or shrink) an existing mapping, potentially moving it at the
590 : : * same time (controlled by the MREMAP_MAYMOVE flag and available VM space)
591 : : *
592 : : * MREMAP_FIXED option added 5-Dec-1999 by Benjamin LaHaise
593 : : * This option implies MREMAP_MAYMOVE.
594 : : */
595 : 78 : SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len,
596 : : unsigned long, new_len, unsigned long, flags,
597 : : unsigned long, new_addr)
598 : : {
599 [ + - ]: 39 : struct mm_struct *mm = current->mm;
600 : 39 : struct vm_area_struct *vma;
601 : 39 : unsigned long ret = -EINVAL;
602 : 39 : unsigned long charged = 0;
603 : 39 : bool locked = false;
604 : 39 : bool downgraded = false;
605 : 39 : struct vm_userfaultfd_ctx uf = NULL_VM_UFFD_CTX;
606 : 39 : LIST_HEAD(uf_unmap_early);
607 : 39 : LIST_HEAD(uf_unmap);
608 : :
609 : 39 : addr = untagged_addr(addr);
610 : :
611 [ + - ]: 39 : if (flags & ~(MREMAP_FIXED | MREMAP_MAYMOVE))
612 : : return ret;
613 : :
614 [ + - ]: 39 : if (flags & MREMAP_FIXED && !(flags & MREMAP_MAYMOVE))
615 : : return ret;
616 : :
617 [ + - ]: 39 : if (offset_in_page(addr))
618 : : return ret;
619 : :
620 : 39 : old_len = PAGE_ALIGN(old_len);
621 : 39 : new_len = PAGE_ALIGN(new_len);
622 : :
623 : : /*
624 : : * We allow a zero old-len as a special case
625 : : * for DOS-emu "duplicate shm area" thing. But
626 : : * a zero new-len is nonsensical.
627 : : */
628 [ + - ]: 39 : if (!new_len)
629 : : return ret;
630 : :
631 [ + - ]: 39 : if (down_write_killable(¤t->mm->mmap_sem))
632 : : return -EINTR;
633 : :
634 [ - + ]: 39 : if (flags & MREMAP_FIXED) {
635 : 0 : ret = mremap_to(addr, old_len, new_addr, new_len,
636 : : &locked, &uf, &uf_unmap_early, &uf_unmap);
637 : 0 : goto out;
638 : : }
639 : :
640 : : /*
641 : : * Always allow a shrinking remap: that just unmaps
642 : : * the unnecessary pages..
643 : : * __do_munmap does all the needed commit accounting, and
644 : : * downgrades mmap_sem to read if so directed.
645 : : */
646 [ + - ]: 39 : if (old_len >= new_len) {
647 : 39 : int retval;
648 : :
649 : 39 : retval = __do_munmap(mm, addr+new_len, old_len - new_len,
650 : : &uf_unmap, true);
651 [ - + ]: 39 : if (retval < 0 && old_len != new_len) {
652 : 0 : ret = retval;
653 : 0 : goto out;
654 : : /* Returning 1 indicates mmap_sem is downgraded to read. */
655 [ + - ]: 39 : } else if (retval == 1)
656 : 39 : downgraded = true;
657 : 39 : ret = addr;
658 : 39 : goto out;
659 : : }
660 : :
661 : : /*
662 : : * Ok, we need to grow..
663 : : */
664 : 0 : vma = vma_to_resize(addr, old_len, new_len, &charged);
665 [ # # ]: 0 : if (IS_ERR(vma)) {
666 : 0 : ret = PTR_ERR(vma);
667 : 0 : goto out;
668 : : }
669 : :
670 : : /* old_len exactly to the end of the area..
671 : : */
672 [ # # ]: 0 : if (old_len == vma->vm_end - addr) {
673 : : /* can we just expand the current mapping? */
674 [ # # ]: 0 : if (vma_expandable(vma, new_len - old_len)) {
675 : 0 : int pages = (new_len - old_len) >> PAGE_SHIFT;
676 : :
677 [ # # ]: 0 : if (vma_adjust(vma, vma->vm_start, addr + new_len,
678 : : vma->vm_pgoff, NULL)) {
679 : 0 : ret = -ENOMEM;
680 : 0 : goto out;
681 : : }
682 : :
683 : 0 : vm_stat_account(mm, vma->vm_flags, pages);
684 [ # # ]: 0 : if (vma->vm_flags & VM_LOCKED) {
685 : 0 : mm->locked_vm += pages;
686 : 0 : locked = true;
687 : 0 : new_addr = addr;
688 : : }
689 : 0 : ret = addr;
690 : 0 : goto out;
691 : : }
692 : : }
693 : :
694 : : /*
695 : : * We weren't able to just expand or shrink the area,
696 : : * we need to create a new one and move it..
697 : : */
698 : 0 : ret = -ENOMEM;
699 [ # # ]: 0 : if (flags & MREMAP_MAYMOVE) {
700 : 0 : unsigned long map_flags = 0;
701 [ # # ]: 0 : if (vma->vm_flags & VM_MAYSHARE)
702 : 0 : map_flags |= MAP_SHARED;
703 : :
704 : 0 : new_addr = get_unmapped_area(vma->vm_file, 0, new_len,
705 : 0 : vma->vm_pgoff +
706 : 0 : ((addr - vma->vm_start) >> PAGE_SHIFT),
707 : : map_flags);
708 [ # # ]: 0 : if (IS_ERR_VALUE(new_addr)) {
709 : 0 : ret = new_addr;
710 : 0 : goto out;
711 : : }
712 : :
713 : 0 : ret = move_vma(vma, addr, old_len, new_len, new_addr,
714 : : &locked, &uf, &uf_unmap);
715 : : }
716 : 0 : out:
717 [ - + ]: 39 : if (offset_in_page(ret)) {
718 : 0 : vm_unacct_memory(charged);
719 : 0 : locked = 0;
720 : : }
721 [ + - ]: 39 : if (downgraded)
722 : 39 : up_read(¤t->mm->mmap_sem);
723 : : else
724 : 0 : up_write(¤t->mm->mmap_sem);
725 [ - + - - ]: 39 : if (locked && new_len > old_len)
726 : 0 : mm_populate(new_addr + old_len, new_len - old_len);
727 : 39 : userfaultfd_unmap_complete(mm, &uf_unmap_early);
728 : 39 : mremap_userfaultfd_complete(&uf, addr, new_addr, old_len);
729 : 39 : userfaultfd_unmap_complete(mm, &uf_unmap);
730 : 39 : return ret;
731 : : }
|
