]>
Commit | Line | Data |
---|---|---|
20c8ccb1 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
c1a4de99 AA |
2 | /* |
3 | * mm/userfaultfd.c | |
4 | * | |
5 | * Copyright (C) 2015 Red Hat, Inc. | |
c1a4de99 AA |
6 | */ |
7 | ||
8 | #include <linux/mm.h> | |
174cd4b1 | 9 | #include <linux/sched/signal.h> |
c1a4de99 AA |
10 | #include <linux/pagemap.h> |
11 | #include <linux/rmap.h> | |
12 | #include <linux/swap.h> | |
13 | #include <linux/swapops.h> | |
14 | #include <linux/userfaultfd_k.h> | |
15 | #include <linux/mmu_notifier.h> | |
60d4d2d2 | 16 | #include <linux/hugetlb.h> |
26071ced | 17 | #include <linux/shmem_fs.h> |
c1a4de99 AA |
18 | #include <asm/tlbflush.h> |
19 | #include "internal.h" | |
20 | ||
643aa36e WY |
21 | static __always_inline |
22 | struct vm_area_struct *find_dst_vma(struct mm_struct *dst_mm, | |
23 | unsigned long dst_start, | |
24 | unsigned long len) | |
25 | { | |
26 | /* | |
27 | * Make sure that the dst range is both valid and fully within a | |
28 | * single existing vma. | |
29 | */ | |
30 | struct vm_area_struct *dst_vma; | |
31 | ||
32 | dst_vma = find_vma(dst_mm, dst_start); | |
33 | if (!dst_vma) | |
34 | return NULL; | |
35 | ||
36 | if (dst_start < dst_vma->vm_start || | |
37 | dst_start + len > dst_vma->vm_end) | |
38 | return NULL; | |
39 | ||
40 | /* | |
41 | * Check the vma is registered in uffd, this is required to | |
42 | * enforce the VM_MAYWRITE check done at uffd registration | |
43 | * time. | |
44 | */ | |
45 | if (!dst_vma->vm_userfaultfd_ctx.ctx) | |
46 | return NULL; | |
47 | ||
48 | return dst_vma; | |
49 | } | |
50 | ||
c1a4de99 AA |
51 | static int mcopy_atomic_pte(struct mm_struct *dst_mm, |
52 | pmd_t *dst_pmd, | |
53 | struct vm_area_struct *dst_vma, | |
54 | unsigned long dst_addr, | |
b6ebaedb AA |
55 | unsigned long src_addr, |
56 | struct page **pagep) | |
c1a4de99 AA |
57 | { |
58 | struct mem_cgroup *memcg; | |
59 | pte_t _dst_pte, *dst_pte; | |
60 | spinlock_t *ptl; | |
c1a4de99 AA |
61 | void *page_kaddr; |
62 | int ret; | |
b6ebaedb | 63 | struct page *page; |
e2a50c1f AA |
64 | pgoff_t offset, max_off; |
65 | struct inode *inode; | |
c1a4de99 | 66 | |
b6ebaedb AA |
67 | if (!*pagep) { |
68 | ret = -ENOMEM; | |
69 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, dst_vma, dst_addr); | |
70 | if (!page) | |
71 | goto out; | |
72 | ||
73 | page_kaddr = kmap_atomic(page); | |
74 | ret = copy_from_user(page_kaddr, | |
75 | (const void __user *) src_addr, | |
76 | PAGE_SIZE); | |
77 | kunmap_atomic(page_kaddr); | |
78 | ||
79 | /* fallback to copy_from_user outside mmap_sem */ | |
80 | if (unlikely(ret)) { | |
9e368259 | 81 | ret = -ENOENT; |
b6ebaedb AA |
82 | *pagep = page; |
83 | /* don't free the page */ | |
84 | goto out; | |
85 | } | |
86 | } else { | |
87 | page = *pagep; | |
88 | *pagep = NULL; | |
89 | } | |
c1a4de99 AA |
90 | |
91 | /* | |
92 | * The memory barrier inside __SetPageUptodate makes sure that | |
f4f5329d | 93 | * preceding stores to the page contents become visible before |
c1a4de99 AA |
94 | * the set_pte_at() write. |
95 | */ | |
96 | __SetPageUptodate(page); | |
97 | ||
98 | ret = -ENOMEM; | |
f627c2f5 | 99 | if (mem_cgroup_try_charge(page, dst_mm, GFP_KERNEL, &memcg, false)) |
c1a4de99 AA |
100 | goto out_release; |
101 | ||
102 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
103 | if (dst_vma->vm_flags & VM_WRITE) | |
104 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
105 | ||
c1a4de99 | 106 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
e2a50c1f AA |
107 | if (dst_vma->vm_file) { |
108 | /* the shmem MAP_PRIVATE case requires checking the i_size */ | |
109 | inode = dst_vma->vm_file->f_inode; | |
110 | offset = linear_page_index(dst_vma, dst_addr); | |
111 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
112 | ret = -EFAULT; | |
113 | if (unlikely(offset >= max_off)) | |
114 | goto out_release_uncharge_unlock; | |
115 | } | |
116 | ret = -EEXIST; | |
c1a4de99 AA |
117 | if (!pte_none(*dst_pte)) |
118 | goto out_release_uncharge_unlock; | |
119 | ||
120 | inc_mm_counter(dst_mm, MM_ANONPAGES); | |
d281ee61 | 121 | page_add_new_anon_rmap(page, dst_vma, dst_addr, false); |
f627c2f5 | 122 | mem_cgroup_commit_charge(page, memcg, false, false); |
c1a4de99 AA |
123 | lru_cache_add_active_or_unevictable(page, dst_vma); |
124 | ||
125 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
126 | ||
127 | /* No need to invalidate - it was non-present before */ | |
128 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
129 | ||
130 | pte_unmap_unlock(dst_pte, ptl); | |
131 | ret = 0; | |
132 | out: | |
133 | return ret; | |
134 | out_release_uncharge_unlock: | |
135 | pte_unmap_unlock(dst_pte, ptl); | |
f627c2f5 | 136 | mem_cgroup_cancel_charge(page, memcg, false); |
c1a4de99 | 137 | out_release: |
09cbfeaf | 138 | put_page(page); |
c1a4de99 | 139 | goto out; |
c1a4de99 AA |
140 | } |
141 | ||
142 | static int mfill_zeropage_pte(struct mm_struct *dst_mm, | |
143 | pmd_t *dst_pmd, | |
144 | struct vm_area_struct *dst_vma, | |
145 | unsigned long dst_addr) | |
146 | { | |
147 | pte_t _dst_pte, *dst_pte; | |
148 | spinlock_t *ptl; | |
149 | int ret; | |
e2a50c1f AA |
150 | pgoff_t offset, max_off; |
151 | struct inode *inode; | |
c1a4de99 AA |
152 | |
153 | _dst_pte = pte_mkspecial(pfn_pte(my_zero_pfn(dst_addr), | |
154 | dst_vma->vm_page_prot)); | |
c1a4de99 | 155 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
e2a50c1f AA |
156 | if (dst_vma->vm_file) { |
157 | /* the shmem MAP_PRIVATE case requires checking the i_size */ | |
158 | inode = dst_vma->vm_file->f_inode; | |
159 | offset = linear_page_index(dst_vma, dst_addr); | |
160 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
161 | ret = -EFAULT; | |
162 | if (unlikely(offset >= max_off)) | |
163 | goto out_unlock; | |
164 | } | |
165 | ret = -EEXIST; | |
c1a4de99 AA |
166 | if (!pte_none(*dst_pte)) |
167 | goto out_unlock; | |
168 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
169 | /* No need to invalidate - it was non-present before */ | |
170 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
171 | ret = 0; | |
172 | out_unlock: | |
173 | pte_unmap_unlock(dst_pte, ptl); | |
174 | return ret; | |
175 | } | |
176 | ||
177 | static pmd_t *mm_alloc_pmd(struct mm_struct *mm, unsigned long address) | |
178 | { | |
179 | pgd_t *pgd; | |
c2febafc | 180 | p4d_t *p4d; |
c1a4de99 | 181 | pud_t *pud; |
c1a4de99 AA |
182 | |
183 | pgd = pgd_offset(mm, address); | |
c2febafc KS |
184 | p4d = p4d_alloc(mm, pgd, address); |
185 | if (!p4d) | |
186 | return NULL; | |
187 | pud = pud_alloc(mm, p4d, address); | |
188 | if (!pud) | |
189 | return NULL; | |
190 | /* | |
191 | * Note that we didn't run this because the pmd was | |
192 | * missing, the *pmd may be already established and in | |
193 | * turn it may also be a trans_huge_pmd. | |
194 | */ | |
195 | return pmd_alloc(mm, pud, address); | |
c1a4de99 AA |
196 | } |
197 | ||
60d4d2d2 MK |
198 | #ifdef CONFIG_HUGETLB_PAGE |
199 | /* | |
200 | * __mcopy_atomic processing for HUGETLB vmas. Note that this routine is | |
201 | * called with mmap_sem held, it will release mmap_sem before returning. | |
202 | */ | |
203 | static __always_inline ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm, | |
204 | struct vm_area_struct *dst_vma, | |
205 | unsigned long dst_start, | |
206 | unsigned long src_start, | |
207 | unsigned long len, | |
208 | bool zeropage) | |
209 | { | |
1c9e8def MK |
210 | int vm_alloc_shared = dst_vma->vm_flags & VM_SHARED; |
211 | int vm_shared = dst_vma->vm_flags & VM_SHARED; | |
60d4d2d2 MK |
212 | ssize_t err; |
213 | pte_t *dst_pte; | |
214 | unsigned long src_addr, dst_addr; | |
215 | long copied; | |
216 | struct page *page; | |
60d4d2d2 MK |
217 | unsigned long vma_hpagesize; |
218 | pgoff_t idx; | |
219 | u32 hash; | |
220 | struct address_space *mapping; | |
221 | ||
222 | /* | |
223 | * There is no default zero huge page for all huge page sizes as | |
224 | * supported by hugetlb. A PMD_SIZE huge pages may exist as used | |
225 | * by THP. Since we can not reliably insert a zero page, this | |
226 | * feature is not supported. | |
227 | */ | |
228 | if (zeropage) { | |
229 | up_read(&dst_mm->mmap_sem); | |
230 | return -EINVAL; | |
231 | } | |
232 | ||
233 | src_addr = src_start; | |
234 | dst_addr = dst_start; | |
235 | copied = 0; | |
236 | page = NULL; | |
237 | vma_hpagesize = vma_kernel_pagesize(dst_vma); | |
238 | ||
239 | /* | |
240 | * Validate alignment based on huge page size | |
241 | */ | |
242 | err = -EINVAL; | |
243 | if (dst_start & (vma_hpagesize - 1) || len & (vma_hpagesize - 1)) | |
244 | goto out_unlock; | |
245 | ||
246 | retry: | |
247 | /* | |
248 | * On routine entry dst_vma is set. If we had to drop mmap_sem and | |
249 | * retry, dst_vma will be set to NULL and we must lookup again. | |
250 | */ | |
251 | if (!dst_vma) { | |
27d02568 | 252 | err = -ENOENT; |
643aa36e | 253 | dst_vma = find_dst_vma(dst_mm, dst_start, len); |
60d4d2d2 MK |
254 | if (!dst_vma || !is_vm_hugetlb_page(dst_vma)) |
255 | goto out_unlock; | |
1c9e8def | 256 | |
27d02568 MR |
257 | err = -EINVAL; |
258 | if (vma_hpagesize != vma_kernel_pagesize(dst_vma)) | |
259 | goto out_unlock; | |
260 | ||
1c9e8def | 261 | vm_shared = dst_vma->vm_flags & VM_SHARED; |
60d4d2d2 MK |
262 | } |
263 | ||
60d4d2d2 | 264 | /* |
1c9e8def | 265 | * If not shared, ensure the dst_vma has a anon_vma. |
60d4d2d2 MK |
266 | */ |
267 | err = -ENOMEM; | |
1c9e8def MK |
268 | if (!vm_shared) { |
269 | if (unlikely(anon_vma_prepare(dst_vma))) | |
270 | goto out_unlock; | |
271 | } | |
60d4d2d2 | 272 | |
60d4d2d2 MK |
273 | while (src_addr < src_start + len) { |
274 | pte_t dst_pteval; | |
275 | ||
276 | BUG_ON(dst_addr >= dst_start + len); | |
60d4d2d2 MK |
277 | |
278 | /* | |
ddeaab32 | 279 | * Serialize via hugetlb_fault_mutex |
60d4d2d2 | 280 | */ |
b43a9990 | 281 | idx = linear_page_index(dst_vma, dst_addr); |
ddeaab32 | 282 | mapping = dst_vma->vm_file->f_mapping; |
188b04a7 | 283 | hash = hugetlb_fault_mutex_hash(mapping, idx); |
60d4d2d2 MK |
284 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
285 | ||
286 | err = -ENOMEM; | |
4fb07ee6 | 287 | dst_pte = huge_pte_alloc(dst_mm, dst_addr, vma_hpagesize); |
60d4d2d2 MK |
288 | if (!dst_pte) { |
289 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
290 | goto out_unlock; | |
291 | } | |
292 | ||
293 | err = -EEXIST; | |
294 | dst_pteval = huge_ptep_get(dst_pte); | |
295 | if (!huge_pte_none(dst_pteval)) { | |
296 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
297 | goto out_unlock; | |
298 | } | |
299 | ||
300 | err = hugetlb_mcopy_atomic_pte(dst_mm, dst_pte, dst_vma, | |
301 | dst_addr, src_addr, &page); | |
302 | ||
303 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
1c9e8def | 304 | vm_alloc_shared = vm_shared; |
60d4d2d2 MK |
305 | |
306 | cond_resched(); | |
307 | ||
9e368259 | 308 | if (unlikely(err == -ENOENT)) { |
60d4d2d2 MK |
309 | up_read(&dst_mm->mmap_sem); |
310 | BUG_ON(!page); | |
311 | ||
312 | err = copy_huge_page_from_user(page, | |
313 | (const void __user *)src_addr, | |
4fb07ee6 WY |
314 | vma_hpagesize / PAGE_SIZE, |
315 | true); | |
60d4d2d2 MK |
316 | if (unlikely(err)) { |
317 | err = -EFAULT; | |
318 | goto out; | |
319 | } | |
320 | down_read(&dst_mm->mmap_sem); | |
321 | ||
322 | dst_vma = NULL; | |
323 | goto retry; | |
324 | } else | |
325 | BUG_ON(page); | |
326 | ||
327 | if (!err) { | |
328 | dst_addr += vma_hpagesize; | |
329 | src_addr += vma_hpagesize; | |
330 | copied += vma_hpagesize; | |
331 | ||
332 | if (fatal_signal_pending(current)) | |
333 | err = -EINTR; | |
334 | } | |
335 | if (err) | |
336 | break; | |
337 | } | |
338 | ||
339 | out_unlock: | |
340 | up_read(&dst_mm->mmap_sem); | |
341 | out: | |
21205bf8 MK |
342 | if (page) { |
343 | /* | |
344 | * We encountered an error and are about to free a newly | |
1c9e8def MK |
345 | * allocated huge page. |
346 | * | |
347 | * Reservation handling is very subtle, and is different for | |
348 | * private and shared mappings. See the routine | |
349 | * restore_reserve_on_error for details. Unfortunately, we | |
350 | * can not call restore_reserve_on_error now as it would | |
351 | * require holding mmap_sem. | |
352 | * | |
353 | * If a reservation for the page existed in the reservation | |
354 | * map of a private mapping, the map was modified to indicate | |
355 | * the reservation was consumed when the page was allocated. | |
356 | * We clear the PagePrivate flag now so that the global | |
21205bf8 MK |
357 | * reserve count will not be incremented in free_huge_page. |
358 | * The reservation map will still indicate the reservation | |
359 | * was consumed and possibly prevent later page allocation. | |
1c9e8def MK |
360 | * This is better than leaking a global reservation. If no |
361 | * reservation existed, it is still safe to clear PagePrivate | |
362 | * as no adjustments to reservation counts were made during | |
363 | * allocation. | |
364 | * | |
365 | * The reservation map for shared mappings indicates which | |
366 | * pages have reservations. When a huge page is allocated | |
367 | * for an address with a reservation, no change is made to | |
368 | * the reserve map. In this case PagePrivate will be set | |
369 | * to indicate that the global reservation count should be | |
370 | * incremented when the page is freed. This is the desired | |
371 | * behavior. However, when a huge page is allocated for an | |
372 | * address without a reservation a reservation entry is added | |
373 | * to the reservation map, and PagePrivate will not be set. | |
374 | * When the page is freed, the global reserve count will NOT | |
375 | * be incremented and it will appear as though we have leaked | |
376 | * reserved page. In this case, set PagePrivate so that the | |
377 | * global reserve count will be incremented to match the | |
378 | * reservation map entry which was created. | |
379 | * | |
380 | * Note that vm_alloc_shared is based on the flags of the vma | |
381 | * for which the page was originally allocated. dst_vma could | |
382 | * be different or NULL on error. | |
21205bf8 | 383 | */ |
1c9e8def MK |
384 | if (vm_alloc_shared) |
385 | SetPagePrivate(page); | |
386 | else | |
387 | ClearPagePrivate(page); | |
60d4d2d2 | 388 | put_page(page); |
21205bf8 | 389 | } |
60d4d2d2 MK |
390 | BUG_ON(copied < 0); |
391 | BUG_ON(err > 0); | |
392 | BUG_ON(!copied && !err); | |
393 | return copied ? copied : err; | |
394 | } | |
395 | #else /* !CONFIG_HUGETLB_PAGE */ | |
396 | /* fail at build time if gcc attempts to use this */ | |
397 | extern ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm, | |
398 | struct vm_area_struct *dst_vma, | |
399 | unsigned long dst_start, | |
400 | unsigned long src_start, | |
401 | unsigned long len, | |
402 | bool zeropage); | |
403 | #endif /* CONFIG_HUGETLB_PAGE */ | |
404 | ||
3217d3c7 MR |
405 | static __always_inline ssize_t mfill_atomic_pte(struct mm_struct *dst_mm, |
406 | pmd_t *dst_pmd, | |
407 | struct vm_area_struct *dst_vma, | |
408 | unsigned long dst_addr, | |
409 | unsigned long src_addr, | |
410 | struct page **page, | |
411 | bool zeropage) | |
412 | { | |
413 | ssize_t err; | |
414 | ||
5b51072e AA |
415 | /* |
416 | * The normal page fault path for a shmem will invoke the | |
417 | * fault, fill the hole in the file and COW it right away. The | |
418 | * result generates plain anonymous memory. So when we are | |
419 | * asked to fill an hole in a MAP_PRIVATE shmem mapping, we'll | |
420 | * generate anonymous memory directly without actually filling | |
421 | * the hole. For the MAP_PRIVATE case the robustness check | |
422 | * only happens in the pagetable (to verify it's still none) | |
423 | * and not in the radix tree. | |
424 | */ | |
425 | if (!(dst_vma->vm_flags & VM_SHARED)) { | |
3217d3c7 MR |
426 | if (!zeropage) |
427 | err = mcopy_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
428 | dst_addr, src_addr, page); | |
429 | else | |
430 | err = mfill_zeropage_pte(dst_mm, dst_pmd, | |
431 | dst_vma, dst_addr); | |
432 | } else { | |
8fb44e54 | 433 | if (!zeropage) |
3217d3c7 MR |
434 | err = shmem_mcopy_atomic_pte(dst_mm, dst_pmd, |
435 | dst_vma, dst_addr, | |
436 | src_addr, page); | |
8fb44e54 MR |
437 | else |
438 | err = shmem_mfill_zeropage_pte(dst_mm, dst_pmd, | |
439 | dst_vma, dst_addr); | |
3217d3c7 MR |
440 | } |
441 | ||
442 | return err; | |
443 | } | |
444 | ||
c1a4de99 AA |
445 | static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm, |
446 | unsigned long dst_start, | |
447 | unsigned long src_start, | |
448 | unsigned long len, | |
df2cc96e MR |
449 | bool zeropage, |
450 | bool *mmap_changing) | |
c1a4de99 AA |
451 | { |
452 | struct vm_area_struct *dst_vma; | |
453 | ssize_t err; | |
454 | pmd_t *dst_pmd; | |
455 | unsigned long src_addr, dst_addr; | |
b6ebaedb AA |
456 | long copied; |
457 | struct page *page; | |
c1a4de99 AA |
458 | |
459 | /* | |
460 | * Sanitize the command parameters: | |
461 | */ | |
462 | BUG_ON(dst_start & ~PAGE_MASK); | |
463 | BUG_ON(len & ~PAGE_MASK); | |
464 | ||
465 | /* Does the address range wrap, or is the span zero-sized? */ | |
466 | BUG_ON(src_start + len <= src_start); | |
467 | BUG_ON(dst_start + len <= dst_start); | |
468 | ||
b6ebaedb AA |
469 | src_addr = src_start; |
470 | dst_addr = dst_start; | |
471 | copied = 0; | |
472 | page = NULL; | |
473 | retry: | |
c1a4de99 AA |
474 | down_read(&dst_mm->mmap_sem); |
475 | ||
df2cc96e MR |
476 | /* |
477 | * If memory mappings are changing because of non-cooperative | |
478 | * operation (e.g. mremap) running in parallel, bail out and | |
479 | * request the user to retry later | |
480 | */ | |
481 | err = -EAGAIN; | |
482 | if (mmap_changing && READ_ONCE(*mmap_changing)) | |
483 | goto out_unlock; | |
484 | ||
c1a4de99 AA |
485 | /* |
486 | * Make sure the vma is not shared, that the dst range is | |
487 | * both valid and fully within a single existing vma. | |
488 | */ | |
27d02568 | 489 | err = -ENOENT; |
643aa36e | 490 | dst_vma = find_dst_vma(dst_mm, dst_start, len); |
26071ced MR |
491 | if (!dst_vma) |
492 | goto out_unlock; | |
c1a4de99 | 493 | |
27d02568 MR |
494 | err = -EINVAL; |
495 | /* | |
496 | * shmem_zero_setup is invoked in mmap for MAP_ANONYMOUS|MAP_SHARED but | |
497 | * it will overwrite vm_ops, so vma_is_anonymous must return false. | |
498 | */ | |
499 | if (WARN_ON_ONCE(vma_is_anonymous(dst_vma) && | |
500 | dst_vma->vm_flags & VM_SHARED)) | |
501 | goto out_unlock; | |
502 | ||
60d4d2d2 MK |
503 | /* |
504 | * If this is a HUGETLB vma, pass off to appropriate routine | |
505 | */ | |
506 | if (is_vm_hugetlb_page(dst_vma)) | |
507 | return __mcopy_atomic_hugetlb(dst_mm, dst_vma, dst_start, | |
508 | src_start, len, zeropage); | |
509 | ||
26071ced | 510 | if (!vma_is_anonymous(dst_vma) && !vma_is_shmem(dst_vma)) |
b6ebaedb | 511 | goto out_unlock; |
c1a4de99 AA |
512 | |
513 | /* | |
514 | * Ensure the dst_vma has a anon_vma or this page | |
515 | * would get a NULL anon_vma when moved in the | |
516 | * dst_vma. | |
517 | */ | |
518 | err = -ENOMEM; | |
5b51072e AA |
519 | if (!(dst_vma->vm_flags & VM_SHARED) && |
520 | unlikely(anon_vma_prepare(dst_vma))) | |
b6ebaedb | 521 | goto out_unlock; |
c1a4de99 | 522 | |
b6ebaedb | 523 | while (src_addr < src_start + len) { |
c1a4de99 | 524 | pmd_t dst_pmdval; |
b6ebaedb | 525 | |
c1a4de99 | 526 | BUG_ON(dst_addr >= dst_start + len); |
b6ebaedb | 527 | |
c1a4de99 AA |
528 | dst_pmd = mm_alloc_pmd(dst_mm, dst_addr); |
529 | if (unlikely(!dst_pmd)) { | |
530 | err = -ENOMEM; | |
531 | break; | |
532 | } | |
533 | ||
534 | dst_pmdval = pmd_read_atomic(dst_pmd); | |
535 | /* | |
536 | * If the dst_pmd is mapped as THP don't | |
537 | * override it and just be strict. | |
538 | */ | |
539 | if (unlikely(pmd_trans_huge(dst_pmdval))) { | |
540 | err = -EEXIST; | |
541 | break; | |
542 | } | |
543 | if (unlikely(pmd_none(dst_pmdval)) && | |
4cf58924 | 544 | unlikely(__pte_alloc(dst_mm, dst_pmd))) { |
c1a4de99 AA |
545 | err = -ENOMEM; |
546 | break; | |
547 | } | |
548 | /* If an huge pmd materialized from under us fail */ | |
549 | if (unlikely(pmd_trans_huge(*dst_pmd))) { | |
550 | err = -EFAULT; | |
551 | break; | |
552 | } | |
553 | ||
554 | BUG_ON(pmd_none(*dst_pmd)); | |
555 | BUG_ON(pmd_trans_huge(*dst_pmd)); | |
556 | ||
3217d3c7 MR |
557 | err = mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, dst_addr, |
558 | src_addr, &page, zeropage); | |
c1a4de99 AA |
559 | cond_resched(); |
560 | ||
9e368259 | 561 | if (unlikely(err == -ENOENT)) { |
b6ebaedb AA |
562 | void *page_kaddr; |
563 | ||
564 | up_read(&dst_mm->mmap_sem); | |
565 | BUG_ON(!page); | |
566 | ||
567 | page_kaddr = kmap(page); | |
568 | err = copy_from_user(page_kaddr, | |
569 | (const void __user *) src_addr, | |
570 | PAGE_SIZE); | |
571 | kunmap(page); | |
572 | if (unlikely(err)) { | |
573 | err = -EFAULT; | |
574 | goto out; | |
575 | } | |
576 | goto retry; | |
577 | } else | |
578 | BUG_ON(page); | |
579 | ||
c1a4de99 AA |
580 | if (!err) { |
581 | dst_addr += PAGE_SIZE; | |
582 | src_addr += PAGE_SIZE; | |
583 | copied += PAGE_SIZE; | |
584 | ||
585 | if (fatal_signal_pending(current)) | |
586 | err = -EINTR; | |
587 | } | |
588 | if (err) | |
589 | break; | |
590 | } | |
591 | ||
b6ebaedb | 592 | out_unlock: |
c1a4de99 | 593 | up_read(&dst_mm->mmap_sem); |
b6ebaedb AA |
594 | out: |
595 | if (page) | |
09cbfeaf | 596 | put_page(page); |
c1a4de99 AA |
597 | BUG_ON(copied < 0); |
598 | BUG_ON(err > 0); | |
599 | BUG_ON(!copied && !err); | |
600 | return copied ? copied : err; | |
601 | } | |
602 | ||
603 | ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start, | |
df2cc96e MR |
604 | unsigned long src_start, unsigned long len, |
605 | bool *mmap_changing) | |
c1a4de99 | 606 | { |
df2cc96e MR |
607 | return __mcopy_atomic(dst_mm, dst_start, src_start, len, false, |
608 | mmap_changing); | |
c1a4de99 AA |
609 | } |
610 | ||
611 | ssize_t mfill_zeropage(struct mm_struct *dst_mm, unsigned long start, | |
df2cc96e | 612 | unsigned long len, bool *mmap_changing) |
c1a4de99 | 613 | { |
df2cc96e | 614 | return __mcopy_atomic(dst_mm, start, 0, len, true, mmap_changing); |
c1a4de99 | 615 | } |