]>
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 | ||
21 | static int mcopy_atomic_pte(struct mm_struct *dst_mm, | |
22 | pmd_t *dst_pmd, | |
23 | struct vm_area_struct *dst_vma, | |
24 | unsigned long dst_addr, | |
b6ebaedb AA |
25 | unsigned long src_addr, |
26 | struct page **pagep) | |
c1a4de99 AA |
27 | { |
28 | struct mem_cgroup *memcg; | |
29 | pte_t _dst_pte, *dst_pte; | |
30 | spinlock_t *ptl; | |
c1a4de99 AA |
31 | void *page_kaddr; |
32 | int ret; | |
b6ebaedb | 33 | struct page *page; |
e2a50c1f AA |
34 | pgoff_t offset, max_off; |
35 | struct inode *inode; | |
c1a4de99 | 36 | |
b6ebaedb AA |
37 | if (!*pagep) { |
38 | ret = -ENOMEM; | |
39 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, dst_vma, dst_addr); | |
40 | if (!page) | |
41 | goto out; | |
42 | ||
43 | page_kaddr = kmap_atomic(page); | |
44 | ret = copy_from_user(page_kaddr, | |
45 | (const void __user *) src_addr, | |
46 | PAGE_SIZE); | |
47 | kunmap_atomic(page_kaddr); | |
48 | ||
49 | /* fallback to copy_from_user outside mmap_sem */ | |
50 | if (unlikely(ret)) { | |
9e368259 | 51 | ret = -ENOENT; |
b6ebaedb AA |
52 | *pagep = page; |
53 | /* don't free the page */ | |
54 | goto out; | |
55 | } | |
56 | } else { | |
57 | page = *pagep; | |
58 | *pagep = NULL; | |
59 | } | |
c1a4de99 AA |
60 | |
61 | /* | |
62 | * The memory barrier inside __SetPageUptodate makes sure that | |
63 | * preceeding stores to the page contents become visible before | |
64 | * the set_pte_at() write. | |
65 | */ | |
66 | __SetPageUptodate(page); | |
67 | ||
68 | ret = -ENOMEM; | |
f627c2f5 | 69 | if (mem_cgroup_try_charge(page, dst_mm, GFP_KERNEL, &memcg, false)) |
c1a4de99 AA |
70 | goto out_release; |
71 | ||
72 | _dst_pte = mk_pte(page, dst_vma->vm_page_prot); | |
73 | if (dst_vma->vm_flags & VM_WRITE) | |
74 | _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte)); | |
75 | ||
c1a4de99 | 76 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
e2a50c1f AA |
77 | if (dst_vma->vm_file) { |
78 | /* the shmem MAP_PRIVATE case requires checking the i_size */ | |
79 | inode = dst_vma->vm_file->f_inode; | |
80 | offset = linear_page_index(dst_vma, dst_addr); | |
81 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
82 | ret = -EFAULT; | |
83 | if (unlikely(offset >= max_off)) | |
84 | goto out_release_uncharge_unlock; | |
85 | } | |
86 | ret = -EEXIST; | |
c1a4de99 AA |
87 | if (!pte_none(*dst_pte)) |
88 | goto out_release_uncharge_unlock; | |
89 | ||
90 | inc_mm_counter(dst_mm, MM_ANONPAGES); | |
d281ee61 | 91 | page_add_new_anon_rmap(page, dst_vma, dst_addr, false); |
f627c2f5 | 92 | mem_cgroup_commit_charge(page, memcg, false, false); |
c1a4de99 AA |
93 | lru_cache_add_active_or_unevictable(page, dst_vma); |
94 | ||
95 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
96 | ||
97 | /* No need to invalidate - it was non-present before */ | |
98 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
99 | ||
100 | pte_unmap_unlock(dst_pte, ptl); | |
101 | ret = 0; | |
102 | out: | |
103 | return ret; | |
104 | out_release_uncharge_unlock: | |
105 | pte_unmap_unlock(dst_pte, ptl); | |
f627c2f5 | 106 | mem_cgroup_cancel_charge(page, memcg, false); |
c1a4de99 | 107 | out_release: |
09cbfeaf | 108 | put_page(page); |
c1a4de99 | 109 | goto out; |
c1a4de99 AA |
110 | } |
111 | ||
112 | static int mfill_zeropage_pte(struct mm_struct *dst_mm, | |
113 | pmd_t *dst_pmd, | |
114 | struct vm_area_struct *dst_vma, | |
115 | unsigned long dst_addr) | |
116 | { | |
117 | pte_t _dst_pte, *dst_pte; | |
118 | spinlock_t *ptl; | |
119 | int ret; | |
e2a50c1f AA |
120 | pgoff_t offset, max_off; |
121 | struct inode *inode; | |
c1a4de99 AA |
122 | |
123 | _dst_pte = pte_mkspecial(pfn_pte(my_zero_pfn(dst_addr), | |
124 | dst_vma->vm_page_prot)); | |
c1a4de99 | 125 | dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl); |
e2a50c1f AA |
126 | if (dst_vma->vm_file) { |
127 | /* the shmem MAP_PRIVATE case requires checking the i_size */ | |
128 | inode = dst_vma->vm_file->f_inode; | |
129 | offset = linear_page_index(dst_vma, dst_addr); | |
130 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); | |
131 | ret = -EFAULT; | |
132 | if (unlikely(offset >= max_off)) | |
133 | goto out_unlock; | |
134 | } | |
135 | ret = -EEXIST; | |
c1a4de99 AA |
136 | if (!pte_none(*dst_pte)) |
137 | goto out_unlock; | |
138 | set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte); | |
139 | /* No need to invalidate - it was non-present before */ | |
140 | update_mmu_cache(dst_vma, dst_addr, dst_pte); | |
141 | ret = 0; | |
142 | out_unlock: | |
143 | pte_unmap_unlock(dst_pte, ptl); | |
144 | return ret; | |
145 | } | |
146 | ||
147 | static pmd_t *mm_alloc_pmd(struct mm_struct *mm, unsigned long address) | |
148 | { | |
149 | pgd_t *pgd; | |
c2febafc | 150 | p4d_t *p4d; |
c1a4de99 | 151 | pud_t *pud; |
c1a4de99 AA |
152 | |
153 | pgd = pgd_offset(mm, address); | |
c2febafc KS |
154 | p4d = p4d_alloc(mm, pgd, address); |
155 | if (!p4d) | |
156 | return NULL; | |
157 | pud = pud_alloc(mm, p4d, address); | |
158 | if (!pud) | |
159 | return NULL; | |
160 | /* | |
161 | * Note that we didn't run this because the pmd was | |
162 | * missing, the *pmd may be already established and in | |
163 | * turn it may also be a trans_huge_pmd. | |
164 | */ | |
165 | return pmd_alloc(mm, pud, address); | |
c1a4de99 AA |
166 | } |
167 | ||
60d4d2d2 MK |
168 | #ifdef CONFIG_HUGETLB_PAGE |
169 | /* | |
170 | * __mcopy_atomic processing for HUGETLB vmas. Note that this routine is | |
171 | * called with mmap_sem held, it will release mmap_sem before returning. | |
172 | */ | |
173 | static __always_inline ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm, | |
174 | struct vm_area_struct *dst_vma, | |
175 | unsigned long dst_start, | |
176 | unsigned long src_start, | |
177 | unsigned long len, | |
178 | bool zeropage) | |
179 | { | |
1c9e8def MK |
180 | int vm_alloc_shared = dst_vma->vm_flags & VM_SHARED; |
181 | int vm_shared = dst_vma->vm_flags & VM_SHARED; | |
60d4d2d2 MK |
182 | ssize_t err; |
183 | pte_t *dst_pte; | |
184 | unsigned long src_addr, dst_addr; | |
185 | long copied; | |
186 | struct page *page; | |
60d4d2d2 MK |
187 | unsigned long vma_hpagesize; |
188 | pgoff_t idx; | |
189 | u32 hash; | |
190 | struct address_space *mapping; | |
191 | ||
192 | /* | |
193 | * There is no default zero huge page for all huge page sizes as | |
194 | * supported by hugetlb. A PMD_SIZE huge pages may exist as used | |
195 | * by THP. Since we can not reliably insert a zero page, this | |
196 | * feature is not supported. | |
197 | */ | |
198 | if (zeropage) { | |
199 | up_read(&dst_mm->mmap_sem); | |
200 | return -EINVAL; | |
201 | } | |
202 | ||
203 | src_addr = src_start; | |
204 | dst_addr = dst_start; | |
205 | copied = 0; | |
206 | page = NULL; | |
207 | vma_hpagesize = vma_kernel_pagesize(dst_vma); | |
208 | ||
209 | /* | |
210 | * Validate alignment based on huge page size | |
211 | */ | |
212 | err = -EINVAL; | |
213 | if (dst_start & (vma_hpagesize - 1) || len & (vma_hpagesize - 1)) | |
214 | goto out_unlock; | |
215 | ||
216 | retry: | |
217 | /* | |
218 | * On routine entry dst_vma is set. If we had to drop mmap_sem and | |
219 | * retry, dst_vma will be set to NULL and we must lookup again. | |
220 | */ | |
221 | if (!dst_vma) { | |
27d02568 | 222 | err = -ENOENT; |
60d4d2d2 MK |
223 | dst_vma = find_vma(dst_mm, dst_start); |
224 | if (!dst_vma || !is_vm_hugetlb_page(dst_vma)) | |
225 | goto out_unlock; | |
60d4d2d2 | 226 | /* |
29ec9066 AA |
227 | * Check the vma is registered in uffd, this is |
228 | * required to enforce the VM_MAYWRITE check done at | |
229 | * uffd registration time. | |
60d4d2d2 | 230 | */ |
27d02568 MR |
231 | if (!dst_vma->vm_userfaultfd_ctx.ctx) |
232 | goto out_unlock; | |
233 | ||
60d4d2d2 MK |
234 | if (dst_start < dst_vma->vm_start || |
235 | dst_start + len > dst_vma->vm_end) | |
236 | goto out_unlock; | |
1c9e8def | 237 | |
27d02568 MR |
238 | err = -EINVAL; |
239 | if (vma_hpagesize != vma_kernel_pagesize(dst_vma)) | |
240 | goto out_unlock; | |
241 | ||
1c9e8def | 242 | vm_shared = dst_vma->vm_flags & VM_SHARED; |
60d4d2d2 MK |
243 | } |
244 | ||
60d4d2d2 | 245 | /* |
1c9e8def | 246 | * If not shared, ensure the dst_vma has a anon_vma. |
60d4d2d2 MK |
247 | */ |
248 | err = -ENOMEM; | |
1c9e8def MK |
249 | if (!vm_shared) { |
250 | if (unlikely(anon_vma_prepare(dst_vma))) | |
251 | goto out_unlock; | |
252 | } | |
60d4d2d2 | 253 | |
60d4d2d2 MK |
254 | while (src_addr < src_start + len) { |
255 | pte_t dst_pteval; | |
256 | ||
257 | BUG_ON(dst_addr >= dst_start + len); | |
60d4d2d2 MK |
258 | |
259 | /* | |
ddeaab32 | 260 | * Serialize via hugetlb_fault_mutex |
60d4d2d2 | 261 | */ |
b43a9990 | 262 | idx = linear_page_index(dst_vma, dst_addr); |
ddeaab32 | 263 | mapping = dst_vma->vm_file->f_mapping; |
188b04a7 | 264 | hash = hugetlb_fault_mutex_hash(mapping, idx); |
60d4d2d2 MK |
265 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
266 | ||
267 | err = -ENOMEM; | |
4fb07ee6 | 268 | dst_pte = huge_pte_alloc(dst_mm, dst_addr, vma_hpagesize); |
60d4d2d2 MK |
269 | if (!dst_pte) { |
270 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
271 | goto out_unlock; | |
272 | } | |
273 | ||
274 | err = -EEXIST; | |
275 | dst_pteval = huge_ptep_get(dst_pte); | |
276 | if (!huge_pte_none(dst_pteval)) { | |
277 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
278 | goto out_unlock; | |
279 | } | |
280 | ||
281 | err = hugetlb_mcopy_atomic_pte(dst_mm, dst_pte, dst_vma, | |
282 | dst_addr, src_addr, &page); | |
283 | ||
284 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
1c9e8def | 285 | vm_alloc_shared = vm_shared; |
60d4d2d2 MK |
286 | |
287 | cond_resched(); | |
288 | ||
9e368259 | 289 | if (unlikely(err == -ENOENT)) { |
60d4d2d2 MK |
290 | up_read(&dst_mm->mmap_sem); |
291 | BUG_ON(!page); | |
292 | ||
293 | err = copy_huge_page_from_user(page, | |
294 | (const void __user *)src_addr, | |
4fb07ee6 WY |
295 | vma_hpagesize / PAGE_SIZE, |
296 | true); | |
60d4d2d2 MK |
297 | if (unlikely(err)) { |
298 | err = -EFAULT; | |
299 | goto out; | |
300 | } | |
301 | down_read(&dst_mm->mmap_sem); | |
302 | ||
303 | dst_vma = NULL; | |
304 | goto retry; | |
305 | } else | |
306 | BUG_ON(page); | |
307 | ||
308 | if (!err) { | |
309 | dst_addr += vma_hpagesize; | |
310 | src_addr += vma_hpagesize; | |
311 | copied += vma_hpagesize; | |
312 | ||
313 | if (fatal_signal_pending(current)) | |
314 | err = -EINTR; | |
315 | } | |
316 | if (err) | |
317 | break; | |
318 | } | |
319 | ||
320 | out_unlock: | |
321 | up_read(&dst_mm->mmap_sem); | |
322 | out: | |
21205bf8 MK |
323 | if (page) { |
324 | /* | |
325 | * We encountered an error and are about to free a newly | |
1c9e8def MK |
326 | * allocated huge page. |
327 | * | |
328 | * Reservation handling is very subtle, and is different for | |
329 | * private and shared mappings. See the routine | |
330 | * restore_reserve_on_error for details. Unfortunately, we | |
331 | * can not call restore_reserve_on_error now as it would | |
332 | * require holding mmap_sem. | |
333 | * | |
334 | * If a reservation for the page existed in the reservation | |
335 | * map of a private mapping, the map was modified to indicate | |
336 | * the reservation was consumed when the page was allocated. | |
337 | * We clear the PagePrivate flag now so that the global | |
21205bf8 MK |
338 | * reserve count will not be incremented in free_huge_page. |
339 | * The reservation map will still indicate the reservation | |
340 | * was consumed and possibly prevent later page allocation. | |
1c9e8def MK |
341 | * This is better than leaking a global reservation. If no |
342 | * reservation existed, it is still safe to clear PagePrivate | |
343 | * as no adjustments to reservation counts were made during | |
344 | * allocation. | |
345 | * | |
346 | * The reservation map for shared mappings indicates which | |
347 | * pages have reservations. When a huge page is allocated | |
348 | * for an address with a reservation, no change is made to | |
349 | * the reserve map. In this case PagePrivate will be set | |
350 | * to indicate that the global reservation count should be | |
351 | * incremented when the page is freed. This is the desired | |
352 | * behavior. However, when a huge page is allocated for an | |
353 | * address without a reservation a reservation entry is added | |
354 | * to the reservation map, and PagePrivate will not be set. | |
355 | * When the page is freed, the global reserve count will NOT | |
356 | * be incremented and it will appear as though we have leaked | |
357 | * reserved page. In this case, set PagePrivate so that the | |
358 | * global reserve count will be incremented to match the | |
359 | * reservation map entry which was created. | |
360 | * | |
361 | * Note that vm_alloc_shared is based on the flags of the vma | |
362 | * for which the page was originally allocated. dst_vma could | |
363 | * be different or NULL on error. | |
21205bf8 | 364 | */ |
1c9e8def MK |
365 | if (vm_alloc_shared) |
366 | SetPagePrivate(page); | |
367 | else | |
368 | ClearPagePrivate(page); | |
60d4d2d2 | 369 | put_page(page); |
21205bf8 | 370 | } |
60d4d2d2 MK |
371 | BUG_ON(copied < 0); |
372 | BUG_ON(err > 0); | |
373 | BUG_ON(!copied && !err); | |
374 | return copied ? copied : err; | |
375 | } | |
376 | #else /* !CONFIG_HUGETLB_PAGE */ | |
377 | /* fail at build time if gcc attempts to use this */ | |
378 | extern ssize_t __mcopy_atomic_hugetlb(struct mm_struct *dst_mm, | |
379 | struct vm_area_struct *dst_vma, | |
380 | unsigned long dst_start, | |
381 | unsigned long src_start, | |
382 | unsigned long len, | |
383 | bool zeropage); | |
384 | #endif /* CONFIG_HUGETLB_PAGE */ | |
385 | ||
3217d3c7 MR |
386 | static __always_inline ssize_t mfill_atomic_pte(struct mm_struct *dst_mm, |
387 | pmd_t *dst_pmd, | |
388 | struct vm_area_struct *dst_vma, | |
389 | unsigned long dst_addr, | |
390 | unsigned long src_addr, | |
391 | struct page **page, | |
392 | bool zeropage) | |
393 | { | |
394 | ssize_t err; | |
395 | ||
5b51072e AA |
396 | /* |
397 | * The normal page fault path for a shmem will invoke the | |
398 | * fault, fill the hole in the file and COW it right away. The | |
399 | * result generates plain anonymous memory. So when we are | |
400 | * asked to fill an hole in a MAP_PRIVATE shmem mapping, we'll | |
401 | * generate anonymous memory directly without actually filling | |
402 | * the hole. For the MAP_PRIVATE case the robustness check | |
403 | * only happens in the pagetable (to verify it's still none) | |
404 | * and not in the radix tree. | |
405 | */ | |
406 | if (!(dst_vma->vm_flags & VM_SHARED)) { | |
3217d3c7 MR |
407 | if (!zeropage) |
408 | err = mcopy_atomic_pte(dst_mm, dst_pmd, dst_vma, | |
409 | dst_addr, src_addr, page); | |
410 | else | |
411 | err = mfill_zeropage_pte(dst_mm, dst_pmd, | |
412 | dst_vma, dst_addr); | |
413 | } else { | |
8fb44e54 | 414 | if (!zeropage) |
3217d3c7 MR |
415 | err = shmem_mcopy_atomic_pte(dst_mm, dst_pmd, |
416 | dst_vma, dst_addr, | |
417 | src_addr, page); | |
8fb44e54 MR |
418 | else |
419 | err = shmem_mfill_zeropage_pte(dst_mm, dst_pmd, | |
420 | dst_vma, dst_addr); | |
3217d3c7 MR |
421 | } |
422 | ||
423 | return err; | |
424 | } | |
425 | ||
c1a4de99 AA |
426 | static __always_inline ssize_t __mcopy_atomic(struct mm_struct *dst_mm, |
427 | unsigned long dst_start, | |
428 | unsigned long src_start, | |
429 | unsigned long len, | |
df2cc96e MR |
430 | bool zeropage, |
431 | bool *mmap_changing) | |
c1a4de99 AA |
432 | { |
433 | struct vm_area_struct *dst_vma; | |
434 | ssize_t err; | |
435 | pmd_t *dst_pmd; | |
436 | unsigned long src_addr, dst_addr; | |
b6ebaedb AA |
437 | long copied; |
438 | struct page *page; | |
c1a4de99 AA |
439 | |
440 | /* | |
441 | * Sanitize the command parameters: | |
442 | */ | |
443 | BUG_ON(dst_start & ~PAGE_MASK); | |
444 | BUG_ON(len & ~PAGE_MASK); | |
445 | ||
446 | /* Does the address range wrap, or is the span zero-sized? */ | |
447 | BUG_ON(src_start + len <= src_start); | |
448 | BUG_ON(dst_start + len <= dst_start); | |
449 | ||
b6ebaedb AA |
450 | src_addr = src_start; |
451 | dst_addr = dst_start; | |
452 | copied = 0; | |
453 | page = NULL; | |
454 | retry: | |
c1a4de99 AA |
455 | down_read(&dst_mm->mmap_sem); |
456 | ||
df2cc96e MR |
457 | /* |
458 | * If memory mappings are changing because of non-cooperative | |
459 | * operation (e.g. mremap) running in parallel, bail out and | |
460 | * request the user to retry later | |
461 | */ | |
462 | err = -EAGAIN; | |
463 | if (mmap_changing && READ_ONCE(*mmap_changing)) | |
464 | goto out_unlock; | |
465 | ||
c1a4de99 AA |
466 | /* |
467 | * Make sure the vma is not shared, that the dst range is | |
468 | * both valid and fully within a single existing vma. | |
469 | */ | |
27d02568 | 470 | err = -ENOENT; |
c1a4de99 | 471 | dst_vma = find_vma(dst_mm, dst_start); |
26071ced MR |
472 | if (!dst_vma) |
473 | goto out_unlock; | |
1c9e8def | 474 | /* |
29ec9066 AA |
475 | * Check the vma is registered in uffd, this is required to |
476 | * enforce the VM_MAYWRITE check done at uffd registration | |
477 | * time. | |
1c9e8def | 478 | */ |
27d02568 | 479 | if (!dst_vma->vm_userfaultfd_ctx.ctx) |
b6ebaedb | 480 | goto out_unlock; |
1c9e8def | 481 | |
c1a4de99 AA |
482 | if (dst_start < dst_vma->vm_start || |
483 | dst_start + len > dst_vma->vm_end) | |
b6ebaedb | 484 | goto out_unlock; |
c1a4de99 | 485 | |
27d02568 MR |
486 | err = -EINVAL; |
487 | /* | |
488 | * shmem_zero_setup is invoked in mmap for MAP_ANONYMOUS|MAP_SHARED but | |
489 | * it will overwrite vm_ops, so vma_is_anonymous must return false. | |
490 | */ | |
491 | if (WARN_ON_ONCE(vma_is_anonymous(dst_vma) && | |
492 | dst_vma->vm_flags & VM_SHARED)) | |
493 | goto out_unlock; | |
494 | ||
60d4d2d2 MK |
495 | /* |
496 | * If this is a HUGETLB vma, pass off to appropriate routine | |
497 | */ | |
498 | if (is_vm_hugetlb_page(dst_vma)) | |
499 | return __mcopy_atomic_hugetlb(dst_mm, dst_vma, dst_start, | |
500 | src_start, len, zeropage); | |
501 | ||
26071ced | 502 | if (!vma_is_anonymous(dst_vma) && !vma_is_shmem(dst_vma)) |
b6ebaedb | 503 | goto out_unlock; |
c1a4de99 AA |
504 | |
505 | /* | |
506 | * Ensure the dst_vma has a anon_vma or this page | |
507 | * would get a NULL anon_vma when moved in the | |
508 | * dst_vma. | |
509 | */ | |
510 | err = -ENOMEM; | |
5b51072e AA |
511 | if (!(dst_vma->vm_flags & VM_SHARED) && |
512 | unlikely(anon_vma_prepare(dst_vma))) | |
b6ebaedb | 513 | goto out_unlock; |
c1a4de99 | 514 | |
b6ebaedb | 515 | while (src_addr < src_start + len) { |
c1a4de99 | 516 | pmd_t dst_pmdval; |
b6ebaedb | 517 | |
c1a4de99 | 518 | BUG_ON(dst_addr >= dst_start + len); |
b6ebaedb | 519 | |
c1a4de99 AA |
520 | dst_pmd = mm_alloc_pmd(dst_mm, dst_addr); |
521 | if (unlikely(!dst_pmd)) { | |
522 | err = -ENOMEM; | |
523 | break; | |
524 | } | |
525 | ||
526 | dst_pmdval = pmd_read_atomic(dst_pmd); | |
527 | /* | |
528 | * If the dst_pmd is mapped as THP don't | |
529 | * override it and just be strict. | |
530 | */ | |
531 | if (unlikely(pmd_trans_huge(dst_pmdval))) { | |
532 | err = -EEXIST; | |
533 | break; | |
534 | } | |
535 | if (unlikely(pmd_none(dst_pmdval)) && | |
4cf58924 | 536 | unlikely(__pte_alloc(dst_mm, dst_pmd))) { |
c1a4de99 AA |
537 | err = -ENOMEM; |
538 | break; | |
539 | } | |
540 | /* If an huge pmd materialized from under us fail */ | |
541 | if (unlikely(pmd_trans_huge(*dst_pmd))) { | |
542 | err = -EFAULT; | |
543 | break; | |
544 | } | |
545 | ||
546 | BUG_ON(pmd_none(*dst_pmd)); | |
547 | BUG_ON(pmd_trans_huge(*dst_pmd)); | |
548 | ||
3217d3c7 MR |
549 | err = mfill_atomic_pte(dst_mm, dst_pmd, dst_vma, dst_addr, |
550 | src_addr, &page, zeropage); | |
c1a4de99 AA |
551 | cond_resched(); |
552 | ||
9e368259 | 553 | if (unlikely(err == -ENOENT)) { |
b6ebaedb AA |
554 | void *page_kaddr; |
555 | ||
556 | up_read(&dst_mm->mmap_sem); | |
557 | BUG_ON(!page); | |
558 | ||
559 | page_kaddr = kmap(page); | |
560 | err = copy_from_user(page_kaddr, | |
561 | (const void __user *) src_addr, | |
562 | PAGE_SIZE); | |
563 | kunmap(page); | |
564 | if (unlikely(err)) { | |
565 | err = -EFAULT; | |
566 | goto out; | |
567 | } | |
568 | goto retry; | |
569 | } else | |
570 | BUG_ON(page); | |
571 | ||
c1a4de99 AA |
572 | if (!err) { |
573 | dst_addr += PAGE_SIZE; | |
574 | src_addr += PAGE_SIZE; | |
575 | copied += PAGE_SIZE; | |
576 | ||
577 | if (fatal_signal_pending(current)) | |
578 | err = -EINTR; | |
579 | } | |
580 | if (err) | |
581 | break; | |
582 | } | |
583 | ||
b6ebaedb | 584 | out_unlock: |
c1a4de99 | 585 | up_read(&dst_mm->mmap_sem); |
b6ebaedb AA |
586 | out: |
587 | if (page) | |
09cbfeaf | 588 | put_page(page); |
c1a4de99 AA |
589 | BUG_ON(copied < 0); |
590 | BUG_ON(err > 0); | |
591 | BUG_ON(!copied && !err); | |
592 | return copied ? copied : err; | |
593 | } | |
594 | ||
595 | ssize_t mcopy_atomic(struct mm_struct *dst_mm, unsigned long dst_start, | |
df2cc96e MR |
596 | unsigned long src_start, unsigned long len, |
597 | bool *mmap_changing) | |
c1a4de99 | 598 | { |
df2cc96e MR |
599 | return __mcopy_atomic(dst_mm, dst_start, src_start, len, false, |
600 | mmap_changing); | |
c1a4de99 AA |
601 | } |
602 | ||
603 | ssize_t mfill_zeropage(struct mm_struct *dst_mm, unsigned long start, | |
df2cc96e | 604 | unsigned long len, bool *mmap_changing) |
c1a4de99 | 605 | { |
df2cc96e | 606 | return __mcopy_atomic(dst_mm, start, 0, len, true, mmap_changing); |
c1a4de99 | 607 | } |