]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
4bbd4c77 KS |
2 | #include <linux/kernel.h> |
3 | #include <linux/errno.h> | |
4 | #include <linux/err.h> | |
5 | #include <linux/spinlock.h> | |
6 | ||
4bbd4c77 | 7 | #include <linux/mm.h> |
3565fce3 | 8 | #include <linux/memremap.h> |
4bbd4c77 KS |
9 | #include <linux/pagemap.h> |
10 | #include <linux/rmap.h> | |
11 | #include <linux/swap.h> | |
12 | #include <linux/swapops.h> | |
1507f512 | 13 | #include <linux/secretmem.h> |
4bbd4c77 | 14 | |
174cd4b1 | 15 | #include <linux/sched/signal.h> |
2667f50e | 16 | #include <linux/rwsem.h> |
f30c59e9 | 17 | #include <linux/hugetlb.h> |
9a4e9f3b AK |
18 | #include <linux/migrate.h> |
19 | #include <linux/mm_inline.h> | |
20 | #include <linux/sched/mm.h> | |
a6e79df9 | 21 | #include <linux/shmem_fs.h> |
1027e443 | 22 | |
33a709b2 | 23 | #include <asm/mmu_context.h> |
1027e443 | 24 | #include <asm/tlbflush.h> |
2667f50e | 25 | |
4bbd4c77 KS |
26 | #include "internal.h" |
27 | ||
df06b37f KB |
28 | struct follow_page_context { |
29 | struct dev_pagemap *pgmap; | |
30 | unsigned int page_mask; | |
31 | }; | |
32 | ||
b6a2619c DH |
33 | static inline void sanity_check_pinned_pages(struct page **pages, |
34 | unsigned long npages) | |
35 | { | |
36 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) | |
37 | return; | |
38 | ||
39 | /* | |
40 | * We only pin anonymous pages if they are exclusive. Once pinned, we | |
41 | * can no longer turn them possibly shared and PageAnonExclusive() will | |
42 | * stick around until the page is freed. | |
43 | * | |
44 | * We'd like to verify that our pinned anonymous pages are still mapped | |
45 | * exclusively. The issue with anon THP is that we don't know how | |
46 | * they are/were mapped when pinning them. However, for anon | |
47 | * THP we can assume that either the given page (PTE-mapped THP) or | |
48 | * the head page (PMD-mapped THP) should be PageAnonExclusive(). If | |
49 | * neither is the case, there is certainly something wrong. | |
50 | */ | |
51 | for (; npages; npages--, pages++) { | |
52 | struct page *page = *pages; | |
53 | struct folio *folio = page_folio(page); | |
54 | ||
c8070b78 DH |
55 | if (is_zero_page(page) || |
56 | !folio_test_anon(folio)) | |
b6a2619c DH |
57 | continue; |
58 | if (!folio_test_large(folio) || folio_test_hugetlb(folio)) | |
59 | VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page), page); | |
60 | else | |
61 | /* Either a PTE-mapped or a PMD-mapped THP. */ | |
62 | VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page) && | |
63 | !PageAnonExclusive(page), page); | |
64 | } | |
65 | } | |
66 | ||
cd1adf1b | 67 | /* |
ece1ed7b | 68 | * Return the folio with ref appropriately incremented, |
cd1adf1b | 69 | * or NULL if that failed. |
a707cdd5 | 70 | */ |
ece1ed7b | 71 | static inline struct folio *try_get_folio(struct page *page, int refs) |
a707cdd5 | 72 | { |
ece1ed7b | 73 | struct folio *folio; |
a707cdd5 | 74 | |
59409373 | 75 | retry: |
ece1ed7b MWO |
76 | folio = page_folio(page); |
77 | if (WARN_ON_ONCE(folio_ref_count(folio) < 0)) | |
a707cdd5 | 78 | return NULL; |
ece1ed7b | 79 | if (unlikely(!folio_ref_try_add_rcu(folio, refs))) |
a707cdd5 | 80 | return NULL; |
c24d3732 JH |
81 | |
82 | /* | |
ece1ed7b MWO |
83 | * At this point we have a stable reference to the folio; but it |
84 | * could be that between calling page_folio() and the refcount | |
85 | * increment, the folio was split, in which case we'd end up | |
86 | * holding a reference on a folio that has nothing to do with the page | |
c24d3732 | 87 | * we were given anymore. |
ece1ed7b MWO |
88 | * So now that the folio is stable, recheck that the page still |
89 | * belongs to this folio. | |
c24d3732 | 90 | */ |
ece1ed7b | 91 | if (unlikely(page_folio(page) != folio)) { |
f4f451a1 MS |
92 | if (!put_devmap_managed_page_refs(&folio->page, refs)) |
93 | folio_put_refs(folio, refs); | |
59409373 | 94 | goto retry; |
c24d3732 JH |
95 | } |
96 | ||
ece1ed7b | 97 | return folio; |
a707cdd5 JH |
98 | } |
99 | ||
3967db22 | 100 | /** |
ece1ed7b | 101 | * try_grab_folio() - Attempt to get or pin a folio. |
3967db22 | 102 | * @page: pointer to page to be grabbed |
ece1ed7b | 103 | * @refs: the value to (effectively) add to the folio's refcount |
3967db22 JH |
104 | * @flags: gup flags: these are the FOLL_* flag values. |
105 | * | |
3faa52c0 | 106 | * "grab" names in this file mean, "look at flags to decide whether to use |
ece1ed7b | 107 | * FOLL_PIN or FOLL_GET behavior, when incrementing the folio's refcount. |
3faa52c0 JH |
108 | * |
109 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
110 | * same time. (That's true throughout the get_user_pages*() and | |
111 | * pin_user_pages*() APIs.) Cases: | |
112 | * | |
ece1ed7b | 113 | * FOLL_GET: folio's refcount will be incremented by @refs. |
3967db22 | 114 | * |
ece1ed7b | 115 | * FOLL_PIN on large folios: folio's refcount will be incremented by |
94688e8e | 116 | * @refs, and its pincount will be incremented by @refs. |
3967db22 | 117 | * |
ece1ed7b | 118 | * FOLL_PIN on single-page folios: folio's refcount will be incremented by |
5232c63f | 119 | * @refs * GUP_PIN_COUNTING_BIAS. |
3faa52c0 | 120 | * |
ece1ed7b MWO |
121 | * Return: The folio containing @page (with refcount appropriately |
122 | * incremented) for success, or NULL upon failure. If neither FOLL_GET | |
123 | * nor FOLL_PIN was set, that's considered failure, and furthermore, | |
124 | * a likely bug in the caller, so a warning is also emitted. | |
3faa52c0 | 125 | */ |
ece1ed7b | 126 | struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) |
3faa52c0 | 127 | { |
503670ee VMO |
128 | struct folio *folio; |
129 | ||
130 | if (WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == 0)) | |
131 | return NULL; | |
132 | ||
4003f107 LG |
133 | if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page))) |
134 | return NULL; | |
135 | ||
3faa52c0 | 136 | if (flags & FOLL_GET) |
ece1ed7b | 137 | return try_get_folio(page, refs); |
ece1ed7b | 138 | |
503670ee | 139 | /* FOLL_PIN is set */ |
c8070b78 | 140 | |
6e17c6de LT |
141 | /* |
142 | * Don't take a pin on the zero page - it's not going anywhere | |
143 | * and it is used in a *lot* of places. | |
144 | */ | |
145 | if (is_zero_page(page)) | |
146 | return page_folio(page); | |
df3a0a21 | 147 | |
6e17c6de | 148 | folio = try_get_folio(page, refs); |
503670ee VMO |
149 | if (!folio) |
150 | return NULL; | |
c24d3732 | 151 | |
503670ee VMO |
152 | /* |
153 | * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a | |
154 | * right zone, so fail and let the caller fall back to the slow | |
155 | * path. | |
156 | */ | |
157 | if (unlikely((flags & FOLL_LONGTERM) && | |
158 | !folio_is_longterm_pinnable(folio))) { | |
159 | if (!put_devmap_managed_page_refs(&folio->page, refs)) | |
160 | folio_put_refs(folio, refs); | |
161 | return NULL; | |
3faa52c0 | 162 | } |
088b8aa5 | 163 | |
503670ee VMO |
164 | /* |
165 | * When pinning a large folio, use an exact count to track it. | |
166 | * | |
167 | * However, be sure to *also* increment the normal folio | |
168 | * refcount field at least once, so that the folio really | |
169 | * is pinned. That's why the refcount from the earlier | |
170 | * try_get_folio() is left intact. | |
171 | */ | |
172 | if (folio_test_large(folio)) | |
173 | atomic_add(refs, &folio->_pincount); | |
174 | else | |
175 | folio_ref_add(folio, | |
176 | refs * (GUP_PIN_COUNTING_BIAS - 1)); | |
177 | /* | |
178 | * Adjust the pincount before re-checking the PTE for changes. | |
179 | * This is essentially a smp_mb() and is paired with a memory | |
180 | * barrier in page_try_share_anon_rmap(). | |
181 | */ | |
182 | smp_mb__after_atomic(); | |
47e29d32 | 183 | |
503670ee | 184 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, refs); |
3faa52c0 | 185 | |
503670ee | 186 | return folio; |
3faa52c0 JH |
187 | } |
188 | ||
d8ddc099 | 189 | static void gup_put_folio(struct folio *folio, int refs, unsigned int flags) |
4509b42c JG |
190 | { |
191 | if (flags & FOLL_PIN) { | |
c8070b78 DH |
192 | if (is_zero_folio(folio)) |
193 | return; | |
d8ddc099 MWO |
194 | node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs); |
195 | if (folio_test_large(folio)) | |
94688e8e | 196 | atomic_sub(refs, &folio->_pincount); |
4509b42c JG |
197 | else |
198 | refs *= GUP_PIN_COUNTING_BIAS; | |
199 | } | |
200 | ||
f4f451a1 MS |
201 | if (!put_devmap_managed_page_refs(&folio->page, refs)) |
202 | folio_put_refs(folio, refs); | |
4509b42c JG |
203 | } |
204 | ||
3faa52c0 JH |
205 | /** |
206 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
5fec0719 MWO |
207 | * @page: pointer to page to be grabbed |
208 | * @flags: gup flags: these are the FOLL_* flag values. | |
3faa52c0 JH |
209 | * |
210 | * This might not do anything at all, depending on the flags argument. | |
211 | * | |
212 | * "grab" names in this file mean, "look at flags to decide whether to use | |
213 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
214 | * | |
3faa52c0 | 215 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same |
ece1ed7b | 216 | * time. Cases: please see the try_grab_folio() documentation, with |
3967db22 | 217 | * "refs=1". |
3faa52c0 | 218 | * |
0f089235 LG |
219 | * Return: 0 for success, or if no action was required (if neither FOLL_PIN |
220 | * nor FOLL_GET was set, nothing is done). A negative error code for failure: | |
221 | * | |
222 | * -ENOMEM FOLL_GET or FOLL_PIN was set, but the page could not | |
223 | * be grabbed. | |
3faa52c0 | 224 | */ |
0f089235 | 225 | int __must_check try_grab_page(struct page *page, unsigned int flags) |
3faa52c0 | 226 | { |
5fec0719 MWO |
227 | struct folio *folio = page_folio(page); |
228 | ||
5fec0719 | 229 | if (WARN_ON_ONCE(folio_ref_count(folio) <= 0)) |
0f089235 | 230 | return -ENOMEM; |
3faa52c0 | 231 | |
4003f107 LG |
232 | if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page))) |
233 | return -EREMOTEIO; | |
3faa52c0 | 234 | |
c36c04c2 | 235 | if (flags & FOLL_GET) |
5fec0719 | 236 | folio_ref_inc(folio); |
c36c04c2 | 237 | else if (flags & FOLL_PIN) { |
c8070b78 DH |
238 | /* |
239 | * Don't take a pin on the zero page - it's not going anywhere | |
240 | * and it is used in a *lot* of places. | |
241 | */ | |
242 | if (is_zero_page(page)) | |
243 | return 0; | |
244 | ||
c36c04c2 | 245 | /* |
5fec0719 | 246 | * Similar to try_grab_folio(): be sure to *also* |
78d9d6ce MWO |
247 | * increment the normal page refcount field at least once, |
248 | * so that the page really is pinned. | |
c36c04c2 | 249 | */ |
5fec0719 MWO |
250 | if (folio_test_large(folio)) { |
251 | folio_ref_add(folio, 1); | |
94688e8e | 252 | atomic_add(1, &folio->_pincount); |
8ea2979c | 253 | } else { |
5fec0719 | 254 | folio_ref_add(folio, GUP_PIN_COUNTING_BIAS); |
8ea2979c | 255 | } |
c36c04c2 | 256 | |
5fec0719 | 257 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, 1); |
c36c04c2 JH |
258 | } |
259 | ||
0f089235 | 260 | return 0; |
3faa52c0 JH |
261 | } |
262 | ||
3faa52c0 JH |
263 | /** |
264 | * unpin_user_page() - release a dma-pinned page | |
265 | * @page: pointer to page to be released | |
266 | * | |
267 | * Pages that were pinned via pin_user_pages*() must be released via either | |
268 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
269 | * that such pages can be separately tracked and uniquely handled. In | |
270 | * particular, interactions with RDMA and filesystems need special handling. | |
271 | */ | |
272 | void unpin_user_page(struct page *page) | |
273 | { | |
b6a2619c | 274 | sanity_check_pinned_pages(&page, 1); |
d8ddc099 | 275 | gup_put_folio(page_folio(page), 1, FOLL_PIN); |
3faa52c0 JH |
276 | } |
277 | EXPORT_SYMBOL(unpin_user_page); | |
278 | ||
1101fb8f DH |
279 | /** |
280 | * folio_add_pin - Try to get an additional pin on a pinned folio | |
281 | * @folio: The folio to be pinned | |
282 | * | |
283 | * Get an additional pin on a folio we already have a pin on. Makes no change | |
284 | * if the folio is a zero_page. | |
285 | */ | |
286 | void folio_add_pin(struct folio *folio) | |
287 | { | |
288 | if (is_zero_folio(folio)) | |
289 | return; | |
290 | ||
291 | /* | |
292 | * Similar to try_grab_folio(): be sure to *also* increment the normal | |
293 | * page refcount field at least once, so that the page really is | |
294 | * pinned. | |
295 | */ | |
296 | if (folio_test_large(folio)) { | |
297 | WARN_ON_ONCE(atomic_read(&folio->_pincount) < 1); | |
298 | folio_ref_inc(folio); | |
299 | atomic_inc(&folio->_pincount); | |
300 | } else { | |
301 | WARN_ON_ONCE(folio_ref_count(folio) < GUP_PIN_COUNTING_BIAS); | |
302 | folio_ref_add(folio, GUP_PIN_COUNTING_BIAS); | |
303 | } | |
304 | } | |
305 | ||
659508f9 | 306 | static inline struct folio *gup_folio_range_next(struct page *start, |
8f39f5fc | 307 | unsigned long npages, unsigned long i, unsigned int *ntails) |
458a4f78 | 308 | { |
659508f9 MWO |
309 | struct page *next = nth_page(start, i); |
310 | struct folio *folio = page_folio(next); | |
458a4f78 JM |
311 | unsigned int nr = 1; |
312 | ||
659508f9 | 313 | if (folio_test_large(folio)) |
4c654229 | 314 | nr = min_t(unsigned int, npages - i, |
659508f9 | 315 | folio_nr_pages(folio) - folio_page_idx(folio, next)); |
458a4f78 | 316 | |
458a4f78 | 317 | *ntails = nr; |
659508f9 | 318 | return folio; |
458a4f78 JM |
319 | } |
320 | ||
12521c76 | 321 | static inline struct folio *gup_folio_next(struct page **list, |
28297dbc | 322 | unsigned long npages, unsigned long i, unsigned int *ntails) |
8745d7f6 | 323 | { |
12521c76 | 324 | struct folio *folio = page_folio(list[i]); |
8745d7f6 JM |
325 | unsigned int nr; |
326 | ||
8745d7f6 | 327 | for (nr = i + 1; nr < npages; nr++) { |
12521c76 | 328 | if (page_folio(list[nr]) != folio) |
8745d7f6 JM |
329 | break; |
330 | } | |
331 | ||
8745d7f6 | 332 | *ntails = nr - i; |
12521c76 | 333 | return folio; |
8745d7f6 JM |
334 | } |
335 | ||
fc1d8e7c | 336 | /** |
f1f6a7dd | 337 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 338 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 339 | * @npages: number of pages in the @pages array. |
2d15eb31 | 340 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
341 | * |
342 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
343 | * variants called on that page. | |
344 | * | |
345 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 346 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
347 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
348 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 349 | * |
f1f6a7dd | 350 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 351 | * |
2d15eb31 AM |
352 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
353 | * required, then the caller should a) verify that this is really correct, | |
354 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 355 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
356 | * |
357 | */ | |
f1f6a7dd JH |
358 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
359 | bool make_dirty) | |
fc1d8e7c | 360 | { |
12521c76 MWO |
361 | unsigned long i; |
362 | struct folio *folio; | |
363 | unsigned int nr; | |
2d15eb31 AM |
364 | |
365 | if (!make_dirty) { | |
f1f6a7dd | 366 | unpin_user_pages(pages, npages); |
2d15eb31 AM |
367 | return; |
368 | } | |
369 | ||
b6a2619c | 370 | sanity_check_pinned_pages(pages, npages); |
12521c76 MWO |
371 | for (i = 0; i < npages; i += nr) { |
372 | folio = gup_folio_next(pages, npages, i, &nr); | |
2d15eb31 AM |
373 | /* |
374 | * Checking PageDirty at this point may race with | |
375 | * clear_page_dirty_for_io(), but that's OK. Two key | |
376 | * cases: | |
377 | * | |
378 | * 1) This code sees the page as already dirty, so it | |
379 | * skips the call to set_page_dirty(). That could happen | |
380 | * because clear_page_dirty_for_io() called | |
381 | * page_mkclean(), followed by set_page_dirty(). | |
382 | * However, now the page is going to get written back, | |
383 | * which meets the original intention of setting it | |
384 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
385 | * on to call TestClearPageDirty(), and write the page | |
386 | * back. | |
387 | * | |
388 | * 2) This code sees the page as clean, so it calls | |
389 | * set_page_dirty(). The page stays dirty, despite being | |
390 | * written back, so it gets written back again in the | |
391 | * next writeback cycle. This is harmless. | |
392 | */ | |
12521c76 MWO |
393 | if (!folio_test_dirty(folio)) { |
394 | folio_lock(folio); | |
395 | folio_mark_dirty(folio); | |
396 | folio_unlock(folio); | |
397 | } | |
398 | gup_put_folio(folio, nr, FOLL_PIN); | |
2d15eb31 | 399 | } |
fc1d8e7c | 400 | } |
f1f6a7dd | 401 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c | 402 | |
458a4f78 JM |
403 | /** |
404 | * unpin_user_page_range_dirty_lock() - release and optionally dirty | |
405 | * gup-pinned page range | |
406 | * | |
407 | * @page: the starting page of a range maybe marked dirty, and definitely released. | |
408 | * @npages: number of consecutive pages to release. | |
409 | * @make_dirty: whether to mark the pages dirty | |
410 | * | |
411 | * "gup-pinned page range" refers to a range of pages that has had one of the | |
412 | * pin_user_pages() variants called on that page. | |
413 | * | |
414 | * For the page ranges defined by [page .. page+npages], make that range (or | |
415 | * its head pages, if a compound page) dirty, if @make_dirty is true, and if the | |
416 | * page range was previously listed as clean. | |
417 | * | |
418 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is | |
419 | * required, then the caller should a) verify that this is really correct, | |
420 | * because _lock() is usually required, and b) hand code it: | |
421 | * set_page_dirty_lock(), unpin_user_page(). | |
422 | * | |
423 | */ | |
424 | void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, | |
425 | bool make_dirty) | |
426 | { | |
659508f9 MWO |
427 | unsigned long i; |
428 | struct folio *folio; | |
429 | unsigned int nr; | |
430 | ||
431 | for (i = 0; i < npages; i += nr) { | |
432 | folio = gup_folio_range_next(page, npages, i, &nr); | |
433 | if (make_dirty && !folio_test_dirty(folio)) { | |
434 | folio_lock(folio); | |
435 | folio_mark_dirty(folio); | |
436 | folio_unlock(folio); | |
437 | } | |
438 | gup_put_folio(folio, nr, FOLL_PIN); | |
458a4f78 JM |
439 | } |
440 | } | |
441 | EXPORT_SYMBOL(unpin_user_page_range_dirty_lock); | |
442 | ||
b6a2619c DH |
443 | static void unpin_user_pages_lockless(struct page **pages, unsigned long npages) |
444 | { | |
445 | unsigned long i; | |
446 | struct folio *folio; | |
447 | unsigned int nr; | |
448 | ||
449 | /* | |
450 | * Don't perform any sanity checks because we might have raced with | |
451 | * fork() and some anonymous pages might now actually be shared -- | |
452 | * which is why we're unpinning after all. | |
453 | */ | |
454 | for (i = 0; i < npages; i += nr) { | |
455 | folio = gup_folio_next(pages, npages, i, &nr); | |
456 | gup_put_folio(folio, nr, FOLL_PIN); | |
457 | } | |
458 | } | |
459 | ||
fc1d8e7c | 460 | /** |
f1f6a7dd | 461 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
462 | * @pages: array of pages to be marked dirty and released. |
463 | * @npages: number of pages in the @pages array. | |
464 | * | |
f1f6a7dd | 465 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 466 | * |
f1f6a7dd | 467 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 468 | */ |
f1f6a7dd | 469 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c | 470 | { |
12521c76 MWO |
471 | unsigned long i; |
472 | struct folio *folio; | |
473 | unsigned int nr; | |
fc1d8e7c | 474 | |
146608bb JH |
475 | /* |
476 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
477 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
478 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
479 | */ | |
480 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
481 | return; | |
31b912de | 482 | |
b6a2619c | 483 | sanity_check_pinned_pages(pages, npages); |
12521c76 MWO |
484 | for (i = 0; i < npages; i += nr) { |
485 | folio = gup_folio_next(pages, npages, i, &nr); | |
486 | gup_put_folio(folio, nr, FOLL_PIN); | |
e7602748 | 487 | } |
fc1d8e7c | 488 | } |
f1f6a7dd | 489 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 490 | |
a458b76a AA |
491 | /* |
492 | * Set the MMF_HAS_PINNED if not set yet; after set it'll be there for the mm's | |
493 | * lifecycle. Avoid setting the bit unless necessary, or it might cause write | |
494 | * cache bouncing on large SMP machines for concurrent pinned gups. | |
495 | */ | |
496 | static inline void mm_set_has_pinned_flag(unsigned long *mm_flags) | |
497 | { | |
498 | if (!test_bit(MMF_HAS_PINNED, mm_flags)) | |
499 | set_bit(MMF_HAS_PINNED, mm_flags); | |
500 | } | |
501 | ||
050a9adc | 502 | #ifdef CONFIG_MMU |
69e68b4f KS |
503 | static struct page *no_page_table(struct vm_area_struct *vma, |
504 | unsigned int flags) | |
4bbd4c77 | 505 | { |
69e68b4f KS |
506 | /* |
507 | * When core dumping an enormous anonymous area that nobody | |
508 | * has touched so far, we don't want to allocate unnecessary pages or | |
509 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
510 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
511 | * But we can only make this optimization where a hole would surely | |
512 | * be zero-filled if handle_mm_fault() actually did handle it. | |
513 | */ | |
a0137f16 AK |
514 | if ((flags & FOLL_DUMP) && |
515 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
516 | return ERR_PTR(-EFAULT); |
517 | return NULL; | |
518 | } | |
4bbd4c77 | 519 | |
1027e443 KS |
520 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
521 | pte_t *pte, unsigned int flags) | |
522 | { | |
1027e443 | 523 | if (flags & FOLL_TOUCH) { |
c33c7948 RR |
524 | pte_t orig_entry = ptep_get(pte); |
525 | pte_t entry = orig_entry; | |
1027e443 KS |
526 | |
527 | if (flags & FOLL_WRITE) | |
528 | entry = pte_mkdirty(entry); | |
529 | entry = pte_mkyoung(entry); | |
530 | ||
c33c7948 | 531 | if (!pte_same(orig_entry, entry)) { |
1027e443 KS |
532 | set_pte_at(vma->vm_mm, address, pte, entry); |
533 | update_mmu_cache(vma, address, pte); | |
534 | } | |
535 | } | |
536 | ||
537 | /* Proper page table entry exists, but no corresponding struct page */ | |
538 | return -EEXIST; | |
539 | } | |
540 | ||
5535be30 DH |
541 | /* FOLL_FORCE can write to even unwritable PTEs in COW mappings. */ |
542 | static inline bool can_follow_write_pte(pte_t pte, struct page *page, | |
543 | struct vm_area_struct *vma, | |
544 | unsigned int flags) | |
19be0eaf | 545 | { |
5535be30 DH |
546 | /* If the pte is writable, we can write to the page. */ |
547 | if (pte_write(pte)) | |
548 | return true; | |
549 | ||
550 | /* Maybe FOLL_FORCE is set to override it? */ | |
551 | if (!(flags & FOLL_FORCE)) | |
552 | return false; | |
553 | ||
554 | /* But FOLL_FORCE has no effect on shared mappings */ | |
555 | if (vma->vm_flags & (VM_MAYSHARE | VM_SHARED)) | |
556 | return false; | |
557 | ||
558 | /* ... or read-only private ones */ | |
559 | if (!(vma->vm_flags & VM_MAYWRITE)) | |
560 | return false; | |
561 | ||
562 | /* ... or already writable ones that just need to take a write fault */ | |
563 | if (vma->vm_flags & VM_WRITE) | |
564 | return false; | |
565 | ||
566 | /* | |
567 | * See can_change_pte_writable(): we broke COW and could map the page | |
568 | * writable if we have an exclusive anonymous page ... | |
569 | */ | |
570 | if (!page || !PageAnon(page) || !PageAnonExclusive(page)) | |
571 | return false; | |
572 | ||
573 | /* ... and a write-fault isn't required for other reasons. */ | |
574 | if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte)) | |
575 | return false; | |
576 | return !userfaultfd_pte_wp(vma, pte); | |
19be0eaf LT |
577 | } |
578 | ||
69e68b4f | 579 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
580 | unsigned long address, pmd_t *pmd, unsigned int flags, |
581 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
582 | { |
583 | struct mm_struct *mm = vma->vm_mm; | |
584 | struct page *page; | |
585 | spinlock_t *ptl; | |
586 | pte_t *ptep, pte; | |
f28d4363 | 587 | int ret; |
4bbd4c77 | 588 | |
eddb1c22 JH |
589 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
590 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
591 | (FOLL_PIN | FOLL_GET))) | |
592 | return ERR_PTR(-EINVAL); | |
4bbd4c77 KS |
593 | |
594 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
04dee9e8 HD |
595 | if (!ptep) |
596 | return no_page_table(vma, flags); | |
c33c7948 | 597 | pte = ptep_get(ptep); |
f7355e99 DH |
598 | if (!pte_present(pte)) |
599 | goto no_page; | |
474098ed | 600 | if (pte_protnone(pte) && !gup_can_follow_protnone(flags)) |
4bbd4c77 | 601 | goto no_page; |
4bbd4c77 KS |
602 | |
603 | page = vm_normal_page(vma, address, pte); | |
5535be30 DH |
604 | |
605 | /* | |
606 | * We only care about anon pages in can_follow_write_pte() and don't | |
607 | * have to worry about pte_devmap() because they are never anon. | |
608 | */ | |
609 | if ((flags & FOLL_WRITE) && | |
610 | !can_follow_write_pte(pte, page, vma, flags)) { | |
611 | page = NULL; | |
612 | goto out; | |
613 | } | |
614 | ||
3faa52c0 | 615 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 616 | /* |
3faa52c0 JH |
617 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
618 | * case since they are only valid while holding the pgmap | |
619 | * reference. | |
3565fce3 | 620 | */ |
df06b37f KB |
621 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
622 | if (*pgmap) | |
3565fce3 DW |
623 | page = pte_page(pte); |
624 | else | |
625 | goto no_page; | |
626 | } else if (unlikely(!page)) { | |
1027e443 KS |
627 | if (flags & FOLL_DUMP) { |
628 | /* Avoid special (like zero) pages in core dumps */ | |
629 | page = ERR_PTR(-EFAULT); | |
630 | goto out; | |
631 | } | |
632 | ||
633 | if (is_zero_pfn(pte_pfn(pte))) { | |
634 | page = pte_page(pte); | |
635 | } else { | |
1027e443 KS |
636 | ret = follow_pfn_pte(vma, address, ptep, flags); |
637 | page = ERR_PTR(ret); | |
638 | goto out; | |
639 | } | |
4bbd4c77 KS |
640 | } |
641 | ||
84209e87 | 642 | if (!pte_write(pte) && gup_must_unshare(vma, flags, page)) { |
a7f22660 DH |
643 | page = ERR_PTR(-EMLINK); |
644 | goto out; | |
645 | } | |
b6a2619c DH |
646 | |
647 | VM_BUG_ON_PAGE((flags & FOLL_PIN) && PageAnon(page) && | |
648 | !PageAnonExclusive(page), page); | |
649 | ||
3faa52c0 | 650 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
0f089235 LG |
651 | ret = try_grab_page(page, flags); |
652 | if (unlikely(ret)) { | |
653 | page = ERR_PTR(ret); | |
3faa52c0 | 654 | goto out; |
8fde12ca | 655 | } |
4003f107 | 656 | |
f28d4363 CI |
657 | /* |
658 | * We need to make the page accessible if and only if we are going | |
659 | * to access its content (the FOLL_PIN case). Please see | |
660 | * Documentation/core-api/pin_user_pages.rst for details. | |
661 | */ | |
662 | if (flags & FOLL_PIN) { | |
663 | ret = arch_make_page_accessible(page); | |
664 | if (ret) { | |
665 | unpin_user_page(page); | |
666 | page = ERR_PTR(ret); | |
667 | goto out; | |
668 | } | |
669 | } | |
4bbd4c77 KS |
670 | if (flags & FOLL_TOUCH) { |
671 | if ((flags & FOLL_WRITE) && | |
672 | !pte_dirty(pte) && !PageDirty(page)) | |
673 | set_page_dirty(page); | |
674 | /* | |
675 | * pte_mkyoung() would be more correct here, but atomic care | |
676 | * is needed to avoid losing the dirty bit: it is easier to use | |
677 | * mark_page_accessed(). | |
678 | */ | |
679 | mark_page_accessed(page); | |
680 | } | |
1027e443 | 681 | out: |
4bbd4c77 | 682 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 683 | return page; |
4bbd4c77 KS |
684 | no_page: |
685 | pte_unmap_unlock(ptep, ptl); | |
686 | if (!pte_none(pte)) | |
69e68b4f KS |
687 | return NULL; |
688 | return no_page_table(vma, flags); | |
689 | } | |
690 | ||
080dbb61 AK |
691 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
692 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
693 | unsigned int flags, |
694 | struct follow_page_context *ctx) | |
69e68b4f | 695 | { |
68827280 | 696 | pmd_t *pmd, pmdval; |
69e68b4f KS |
697 | spinlock_t *ptl; |
698 | struct page *page; | |
699 | struct mm_struct *mm = vma->vm_mm; | |
700 | ||
080dbb61 | 701 | pmd = pmd_offset(pudp, address); |
26e1a0c3 | 702 | pmdval = pmdp_get_lockless(pmd); |
68827280 | 703 | if (pmd_none(pmdval)) |
69e68b4f | 704 | return no_page_table(vma, flags); |
f7355e99 | 705 | if (!pmd_present(pmdval)) |
e66f17ff | 706 | return no_page_table(vma, flags); |
68827280 | 707 | if (pmd_devmap(pmdval)) { |
3565fce3 | 708 | ptl = pmd_lock(mm, pmd); |
df06b37f | 709 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
710 | spin_unlock(ptl); |
711 | if (page) | |
712 | return page; | |
713 | } | |
68827280 | 714 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 715 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 716 | |
474098ed | 717 | if (pmd_protnone(pmdval) && !gup_can_follow_protnone(flags)) |
db08f203 AK |
718 | return no_page_table(vma, flags); |
719 | ||
6742d293 | 720 | ptl = pmd_lock(mm, pmd); |
84c3fc4e ZY |
721 | if (unlikely(!pmd_present(*pmd))) { |
722 | spin_unlock(ptl); | |
f7355e99 | 723 | return no_page_table(vma, flags); |
84c3fc4e | 724 | } |
6742d293 KS |
725 | if (unlikely(!pmd_trans_huge(*pmd))) { |
726 | spin_unlock(ptl); | |
df06b37f | 727 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 728 | } |
4066c119 | 729 | if (flags & FOLL_SPLIT_PMD) { |
2378118b HD |
730 | spin_unlock(ptl); |
731 | split_huge_pmd(vma, pmd, address); | |
732 | /* If pmd was left empty, stuff a page table in there quickly */ | |
733 | return pte_alloc(mm, pmd) ? ERR_PTR(-ENOMEM) : | |
df06b37f | 734 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 735 | } |
6742d293 KS |
736 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
737 | spin_unlock(ptl); | |
df06b37f | 738 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 739 | return page; |
4bbd4c77 KS |
740 | } |
741 | ||
080dbb61 AK |
742 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
743 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
744 | unsigned int flags, |
745 | struct follow_page_context *ctx) | |
080dbb61 AK |
746 | { |
747 | pud_t *pud; | |
748 | spinlock_t *ptl; | |
749 | struct page *page; | |
750 | struct mm_struct *mm = vma->vm_mm; | |
751 | ||
752 | pud = pud_offset(p4dp, address); | |
753 | if (pud_none(*pud)) | |
754 | return no_page_table(vma, flags); | |
080dbb61 AK |
755 | if (pud_devmap(*pud)) { |
756 | ptl = pud_lock(mm, pud); | |
df06b37f | 757 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
758 | spin_unlock(ptl); |
759 | if (page) | |
760 | return page; | |
761 | } | |
762 | if (unlikely(pud_bad(*pud))) | |
763 | return no_page_table(vma, flags); | |
764 | ||
df06b37f | 765 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
766 | } |
767 | ||
080dbb61 AK |
768 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
769 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
770 | unsigned int flags, |
771 | struct follow_page_context *ctx) | |
080dbb61 AK |
772 | { |
773 | p4d_t *p4d; | |
774 | ||
775 | p4d = p4d_offset(pgdp, address); | |
776 | if (p4d_none(*p4d)) | |
777 | return no_page_table(vma, flags); | |
778 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
779 | if (unlikely(p4d_bad(*p4d))) | |
780 | return no_page_table(vma, flags); | |
781 | ||
df06b37f | 782 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
783 | } |
784 | ||
785 | /** | |
786 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
787 | * @vma: vm_area_struct mapping @address | |
788 | * @address: virtual address to look up | |
789 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
790 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
791 | * pointer to output page_mask | |
080dbb61 AK |
792 | * |
793 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
794 | * | |
78179556 MR |
795 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
796 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
797 | * | |
a7f22660 DH |
798 | * When getting an anonymous page and the caller has to trigger unsharing |
799 | * of a shared anonymous page first, -EMLINK is returned. The caller should | |
800 | * trigger a fault with FAULT_FLAG_UNSHARE set. Note that unsharing is only | |
801 | * relevant with FOLL_PIN and !FOLL_WRITE. | |
802 | * | |
78179556 MR |
803 | * On output, the @ctx->page_mask is set according to the size of the page. |
804 | * | |
805 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
806 | * an error pointer if there is a mapping to something not represented |
807 | * by a page descriptor (see also vm_normal_page()). | |
808 | */ | |
a7030aea | 809 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 810 | unsigned long address, unsigned int flags, |
df06b37f | 811 | struct follow_page_context *ctx) |
080dbb61 AK |
812 | { |
813 | pgd_t *pgd; | |
814 | struct page *page; | |
815 | struct mm_struct *mm = vma->vm_mm; | |
816 | ||
df06b37f | 817 | ctx->page_mask = 0; |
080dbb61 | 818 | |
57a196a5 MK |
819 | /* |
820 | * Call hugetlb_follow_page_mask for hugetlb vmas as it will use | |
821 | * special hugetlb page table walking code. This eliminates the | |
822 | * need to check for hugetlb entries in the general walking code. | |
823 | * | |
824 | * hugetlb_follow_page_mask is only for follow_page() handling here. | |
825 | * Ordinary GUP uses follow_hugetlb_page for hugetlb processing. | |
826 | */ | |
827 | if (is_vm_hugetlb_page(vma)) { | |
828 | page = hugetlb_follow_page_mask(vma, address, flags); | |
829 | if (!page) | |
830 | page = no_page_table(vma, flags); | |
080dbb61 AK |
831 | return page; |
832 | } | |
833 | ||
834 | pgd = pgd_offset(mm, address); | |
835 | ||
836 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
837 | return no_page_table(vma, flags); | |
838 | ||
df06b37f KB |
839 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
840 | } | |
841 | ||
842 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
843 | unsigned int foll_flags) | |
844 | { | |
845 | struct follow_page_context ctx = { NULL }; | |
846 | struct page *page; | |
847 | ||
1507f512 MR |
848 | if (vma_is_secretmem(vma)) |
849 | return NULL; | |
850 | ||
d64e2dbc | 851 | if (WARN_ON_ONCE(foll_flags & FOLL_PIN)) |
8909691b DH |
852 | return NULL; |
853 | ||
df06b37f KB |
854 | page = follow_page_mask(vma, address, foll_flags, &ctx); |
855 | if (ctx.pgmap) | |
856 | put_dev_pagemap(ctx.pgmap); | |
857 | return page; | |
080dbb61 AK |
858 | } |
859 | ||
f2b495ca KS |
860 | static int get_gate_page(struct mm_struct *mm, unsigned long address, |
861 | unsigned int gup_flags, struct vm_area_struct **vma, | |
862 | struct page **page) | |
863 | { | |
864 | pgd_t *pgd; | |
c2febafc | 865 | p4d_t *p4d; |
f2b495ca KS |
866 | pud_t *pud; |
867 | pmd_t *pmd; | |
868 | pte_t *pte; | |
c33c7948 | 869 | pte_t entry; |
f2b495ca KS |
870 | int ret = -EFAULT; |
871 | ||
872 | /* user gate pages are read-only */ | |
873 | if (gup_flags & FOLL_WRITE) | |
874 | return -EFAULT; | |
875 | if (address > TASK_SIZE) | |
876 | pgd = pgd_offset_k(address); | |
877 | else | |
878 | pgd = pgd_offset_gate(mm, address); | |
b5d1c39f AL |
879 | if (pgd_none(*pgd)) |
880 | return -EFAULT; | |
c2febafc | 881 | p4d = p4d_offset(pgd, address); |
b5d1c39f AL |
882 | if (p4d_none(*p4d)) |
883 | return -EFAULT; | |
c2febafc | 884 | pud = pud_offset(p4d, address); |
b5d1c39f AL |
885 | if (pud_none(*pud)) |
886 | return -EFAULT; | |
f2b495ca | 887 | pmd = pmd_offset(pud, address); |
84c3fc4e | 888 | if (!pmd_present(*pmd)) |
f2b495ca | 889 | return -EFAULT; |
f2b495ca | 890 | pte = pte_offset_map(pmd, address); |
04dee9e8 HD |
891 | if (!pte) |
892 | return -EFAULT; | |
c33c7948 RR |
893 | entry = ptep_get(pte); |
894 | if (pte_none(entry)) | |
f2b495ca KS |
895 | goto unmap; |
896 | *vma = get_gate_vma(mm); | |
897 | if (!page) | |
898 | goto out; | |
c33c7948 | 899 | *page = vm_normal_page(*vma, address, entry); |
f2b495ca | 900 | if (!*page) { |
c33c7948 | 901 | if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(entry))) |
f2b495ca | 902 | goto unmap; |
c33c7948 | 903 | *page = pte_page(entry); |
f2b495ca | 904 | } |
0f089235 LG |
905 | ret = try_grab_page(*page, gup_flags); |
906 | if (unlikely(ret)) | |
8fde12ca | 907 | goto unmap; |
f2b495ca KS |
908 | out: |
909 | ret = 0; | |
910 | unmap: | |
911 | pte_unmap(pte); | |
912 | return ret; | |
913 | } | |
914 | ||
9a95f3cf | 915 | /* |
9a863a6a JG |
916 | * mmap_lock must be held on entry. If @flags has FOLL_UNLOCKABLE but not |
917 | * FOLL_NOWAIT, the mmap_lock may be released. If it is, *@locked will be set | |
918 | * to 0 and -EBUSY returned. | |
9a95f3cf | 919 | */ |
64019a2e | 920 | static int faultin_page(struct vm_area_struct *vma, |
a7f22660 DH |
921 | unsigned long address, unsigned int *flags, bool unshare, |
922 | int *locked) | |
16744483 | 923 | { |
16744483 | 924 | unsigned int fault_flags = 0; |
2b740303 | 925 | vm_fault_t ret; |
16744483 | 926 | |
55b8fe70 AG |
927 | if (*flags & FOLL_NOFAULT) |
928 | return -EFAULT; | |
16744483 KS |
929 | if (*flags & FOLL_WRITE) |
930 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
931 | if (*flags & FOLL_REMOTE) |
932 | fault_flags |= FAULT_FLAG_REMOTE; | |
f04740f5 | 933 | if (*flags & FOLL_UNLOCKABLE) { |
71335f37 | 934 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
93c5c61d PX |
935 | /* |
936 | * FAULT_FLAG_INTERRUPTIBLE is opt-in. GUP callers must set | |
937 | * FOLL_INTERRUPTIBLE to enable FAULT_FLAG_INTERRUPTIBLE. | |
938 | * That's because some callers may not be prepared to | |
939 | * handle early exits caused by non-fatal signals. | |
940 | */ | |
941 | if (*flags & FOLL_INTERRUPTIBLE) | |
942 | fault_flags |= FAULT_FLAG_INTERRUPTIBLE; | |
943 | } | |
16744483 KS |
944 | if (*flags & FOLL_NOWAIT) |
945 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 946 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
947 | /* |
948 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
949 | * can co-exist | |
950 | */ | |
234b239b ALC |
951 | fault_flags |= FAULT_FLAG_TRIED; |
952 | } | |
a7f22660 DH |
953 | if (unshare) { |
954 | fault_flags |= FAULT_FLAG_UNSHARE; | |
955 | /* FAULT_FLAG_WRITE and FAULT_FLAG_UNSHARE are incompatible */ | |
956 | VM_BUG_ON(fault_flags & FAULT_FLAG_WRITE); | |
957 | } | |
16744483 | 958 | |
bce617ed | 959 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
d9272525 PX |
960 | |
961 | if (ret & VM_FAULT_COMPLETED) { | |
962 | /* | |
963 | * With FAULT_FLAG_RETRY_NOWAIT we'll never release the | |
964 | * mmap lock in the page fault handler. Sanity check this. | |
965 | */ | |
966 | WARN_ON_ONCE(fault_flags & FAULT_FLAG_RETRY_NOWAIT); | |
9a863a6a JG |
967 | *locked = 0; |
968 | ||
d9272525 PX |
969 | /* |
970 | * We should do the same as VM_FAULT_RETRY, but let's not | |
971 | * return -EBUSY since that's not reflecting the reality of | |
972 | * what has happened - we've just fully completed a page | |
973 | * fault, with the mmap lock released. Use -EAGAIN to show | |
974 | * that we want to take the mmap lock _again_. | |
975 | */ | |
976 | return -EAGAIN; | |
977 | } | |
978 | ||
16744483 | 979 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
980 | int err = vm_fault_to_errno(ret, *flags); |
981 | ||
982 | if (err) | |
983 | return err; | |
16744483 KS |
984 | BUG(); |
985 | } | |
986 | ||
16744483 | 987 | if (ret & VM_FAULT_RETRY) { |
9a863a6a | 988 | if (!(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
4f6da934 | 989 | *locked = 0; |
16744483 KS |
990 | return -EBUSY; |
991 | } | |
992 | ||
16744483 KS |
993 | return 0; |
994 | } | |
995 | ||
8ac26843 LS |
996 | /* |
997 | * Writing to file-backed mappings which require folio dirty tracking using GUP | |
998 | * is a fundamentally broken operation, as kernel write access to GUP mappings | |
999 | * do not adhere to the semantics expected by a file system. | |
1000 | * | |
1001 | * Consider the following scenario:- | |
1002 | * | |
1003 | * 1. A folio is written to via GUP which write-faults the memory, notifying | |
1004 | * the file system and dirtying the folio. | |
1005 | * 2. Later, writeback is triggered, resulting in the folio being cleaned and | |
1006 | * the PTE being marked read-only. | |
1007 | * 3. The GUP caller writes to the folio, as it is mapped read/write via the | |
1008 | * direct mapping. | |
1009 | * 4. The GUP caller, now done with the page, unpins it and sets it dirty | |
1010 | * (though it does not have to). | |
1011 | * | |
1012 | * This results in both data being written to a folio without writenotify, and | |
1013 | * the folio being dirtied unexpectedly (if the caller decides to do so). | |
1014 | */ | |
1015 | static bool writable_file_mapping_allowed(struct vm_area_struct *vma, | |
1016 | unsigned long gup_flags) | |
1017 | { | |
1018 | /* | |
1019 | * If we aren't pinning then no problematic write can occur. A long term | |
1020 | * pin is the most egregious case so this is the case we disallow. | |
1021 | */ | |
1022 | if ((gup_flags & (FOLL_PIN | FOLL_LONGTERM)) != | |
1023 | (FOLL_PIN | FOLL_LONGTERM)) | |
1024 | return true; | |
1025 | ||
1026 | /* | |
1027 | * If the VMA does not require dirty tracking then no problematic write | |
1028 | * can occur either. | |
1029 | */ | |
1030 | return !vma_needs_dirty_tracking(vma); | |
1031 | } | |
1032 | ||
fa5bb209 KS |
1033 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
1034 | { | |
1035 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
1036 | int write = (gup_flags & FOLL_WRITE); |
1037 | int foreign = (gup_flags & FOLL_REMOTE); | |
8ac26843 | 1038 | bool vma_anon = vma_is_anonymous(vma); |
fa5bb209 KS |
1039 | |
1040 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
1041 | return -EFAULT; | |
1042 | ||
8ac26843 | 1043 | if ((gup_flags & FOLL_ANON) && !vma_anon) |
7f7ccc2c WT |
1044 | return -EFAULT; |
1045 | ||
52650c8b JG |
1046 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
1047 | return -EOPNOTSUPP; | |
1048 | ||
1507f512 MR |
1049 | if (vma_is_secretmem(vma)) |
1050 | return -EFAULT; | |
1051 | ||
1b2ee126 | 1052 | if (write) { |
8ac26843 LS |
1053 | if (!vma_anon && |
1054 | !writable_file_mapping_allowed(vma, gup_flags)) | |
1055 | return -EFAULT; | |
1056 | ||
fa5bb209 KS |
1057 | if (!(vm_flags & VM_WRITE)) { |
1058 | if (!(gup_flags & FOLL_FORCE)) | |
1059 | return -EFAULT; | |
f347454d DH |
1060 | /* hugetlb does not support FOLL_FORCE|FOLL_WRITE. */ |
1061 | if (is_vm_hugetlb_page(vma)) | |
1062 | return -EFAULT; | |
fa5bb209 KS |
1063 | /* |
1064 | * We used to let the write,force case do COW in a | |
1065 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
1066 | * set a breakpoint in a read-only mapping of an | |
1067 | * executable, without corrupting the file (yet only | |
1068 | * when that file had been opened for writing!). | |
1069 | * Anon pages in shared mappings are surprising: now | |
1070 | * just reject it. | |
1071 | */ | |
46435364 | 1072 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 1073 | return -EFAULT; |
fa5bb209 KS |
1074 | } |
1075 | } else if (!(vm_flags & VM_READ)) { | |
1076 | if (!(gup_flags & FOLL_FORCE)) | |
1077 | return -EFAULT; | |
1078 | /* | |
1079 | * Is there actually any vma we can reach here which does not | |
1080 | * have VM_MAYREAD set? | |
1081 | */ | |
1082 | if (!(vm_flags & VM_MAYREAD)) | |
1083 | return -EFAULT; | |
1084 | } | |
d61172b4 DH |
1085 | /* |
1086 | * gups are always data accesses, not instruction | |
1087 | * fetches, so execute=false here | |
1088 | */ | |
1089 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 1090 | return -EFAULT; |
fa5bb209 KS |
1091 | return 0; |
1092 | } | |
1093 | ||
4bbd4c77 KS |
1094 | /** |
1095 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
1096 | * @mm: mm_struct of target mm |
1097 | * @start: starting user address | |
1098 | * @nr_pages: number of pages from start to pin | |
1099 | * @gup_flags: flags modifying pin behaviour | |
1100 | * @pages: array that receives pointers to the pages pinned. | |
1101 | * Should be at least nr_pages long. Or NULL, if caller | |
1102 | * only intends to ensure the pages are faulted in. | |
c1e8d7c6 | 1103 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 1104 | * |
d2dfbe47 LX |
1105 | * Returns either number of pages pinned (which may be less than the |
1106 | * number requested), or an error. Details about the return value: | |
1107 | * | |
1108 | * -- If nr_pages is 0, returns 0. | |
1109 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1110 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1111 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 1112 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
1113 | * |
1114 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1115 | * | |
c1e8d7c6 | 1116 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
1117 | * |
1118 | * __get_user_pages walks a process's page tables and takes a reference to | |
1119 | * each struct page that each user address corresponds to at a given | |
1120 | * instant. That is, it takes the page that would be accessed if a user | |
1121 | * thread accesses the given user virtual address at that instant. | |
1122 | * | |
1123 | * This does not guarantee that the page exists in the user mappings when | |
1124 | * __get_user_pages returns, and there may even be a completely different | |
1125 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
c5acf1f6 | 1126 | * and subsequently re-faulted). However it does guarantee that the page |
4bbd4c77 KS |
1127 | * won't be freed completely. And mostly callers simply care that the page |
1128 | * contains data that was valid *at some point in time*. Typically, an IO | |
1129 | * or similar operation cannot guarantee anything stronger anyway because | |
1130 | * locks can't be held over the syscall boundary. | |
1131 | * | |
1132 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1133 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1134 | * appropriate) must be called after the page is finished with, and | |
1135 | * before put_page is called. | |
1136 | * | |
9a863a6a JG |
1137 | * If FOLL_UNLOCKABLE is set without FOLL_NOWAIT then the mmap_lock may |
1138 | * be released. If this happens *@locked will be set to 0 on return. | |
9a95f3cf | 1139 | * |
9a863a6a JG |
1140 | * A caller using such a combination of @gup_flags must therefore hold the |
1141 | * mmap_lock for reading only, and recognize when it's been released. Otherwise, | |
1142 | * it must be held for either reading or writing and will not be released. | |
4bbd4c77 KS |
1143 | * |
1144 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1145 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1146 | * you need some special @gup_flags. | |
1147 | */ | |
64019a2e | 1148 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
1149 | unsigned long start, unsigned long nr_pages, |
1150 | unsigned int gup_flags, struct page **pages, | |
b2cac248 | 1151 | int *locked) |
4bbd4c77 | 1152 | { |
df06b37f | 1153 | long ret = 0, i = 0; |
fa5bb209 | 1154 | struct vm_area_struct *vma = NULL; |
df06b37f | 1155 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
1156 | |
1157 | if (!nr_pages) | |
1158 | return 0; | |
1159 | ||
428e106a | 1160 | start = untagged_addr_remote(mm, start); |
f9652594 | 1161 | |
eddb1c22 | 1162 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 | 1163 | |
4bbd4c77 | 1164 | do { |
fa5bb209 KS |
1165 | struct page *page; |
1166 | unsigned int foll_flags = gup_flags; | |
1167 | unsigned int page_increm; | |
1168 | ||
1169 | /* first iteration or cross vma bound */ | |
1170 | if (!vma || start >= vma->vm_end) { | |
a425ac53 LT |
1171 | vma = find_vma(mm, start); |
1172 | if (vma && (start < vma->vm_start)) { | |
1173 | WARN_ON_ONCE(vma->vm_flags & VM_GROWSDOWN); | |
1174 | vma = NULL; | |
1175 | } | |
fa5bb209 | 1176 | if (!vma && in_gate_area(mm, start)) { |
fa5bb209 KS |
1177 | ret = get_gate_page(mm, start & PAGE_MASK, |
1178 | gup_flags, &vma, | |
1179 | pages ? &pages[i] : NULL); | |
1180 | if (ret) | |
08be37b7 | 1181 | goto out; |
df06b37f | 1182 | ctx.page_mask = 0; |
fa5bb209 KS |
1183 | goto next_page; |
1184 | } | |
4bbd4c77 | 1185 | |
52650c8b | 1186 | if (!vma) { |
df06b37f KB |
1187 | ret = -EFAULT; |
1188 | goto out; | |
1189 | } | |
52650c8b JG |
1190 | ret = check_vma_flags(vma, gup_flags); |
1191 | if (ret) | |
1192 | goto out; | |
1193 | ||
fa5bb209 | 1194 | if (is_vm_hugetlb_page(vma)) { |
b2cac248 LS |
1195 | i = follow_hugetlb_page(mm, vma, pages, |
1196 | &start, &nr_pages, i, | |
1197 | gup_flags, locked); | |
9a863a6a | 1198 | if (!*locked) { |
ad415db8 PX |
1199 | /* |
1200 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1201 | * and we've lost mmap_lock. |
ad415db8 PX |
1202 | * We must stop here. |
1203 | */ | |
1204 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
ad415db8 PX |
1205 | goto out; |
1206 | } | |
fa5bb209 | 1207 | continue; |
4bbd4c77 | 1208 | } |
fa5bb209 KS |
1209 | } |
1210 | retry: | |
1211 | /* | |
1212 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1213 | * potentially allocating memory. | |
1214 | */ | |
fa45f116 | 1215 | if (fatal_signal_pending(current)) { |
d180870d | 1216 | ret = -EINTR; |
df06b37f KB |
1217 | goto out; |
1218 | } | |
fa5bb209 | 1219 | cond_resched(); |
df06b37f KB |
1220 | |
1221 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
a7f22660 DH |
1222 | if (!page || PTR_ERR(page) == -EMLINK) { |
1223 | ret = faultin_page(vma, start, &foll_flags, | |
1224 | PTR_ERR(page) == -EMLINK, locked); | |
fa5bb209 KS |
1225 | switch (ret) { |
1226 | case 0: | |
1227 | goto retry; | |
df06b37f | 1228 | case -EBUSY: |
d9272525 | 1229 | case -EAGAIN: |
df06b37f | 1230 | ret = 0; |
e4a9bc58 | 1231 | fallthrough; |
fa5bb209 KS |
1232 | case -EFAULT: |
1233 | case -ENOMEM: | |
1234 | case -EHWPOISON: | |
df06b37f | 1235 | goto out; |
4bbd4c77 | 1236 | } |
fa5bb209 | 1237 | BUG(); |
1027e443 KS |
1238 | } else if (PTR_ERR(page) == -EEXIST) { |
1239 | /* | |
1240 | * Proper page table entry exists, but no corresponding | |
65462462 JH |
1241 | * struct page. If the caller expects **pages to be |
1242 | * filled in, bail out now, because that can't be done | |
1243 | * for this page. | |
1027e443 | 1244 | */ |
65462462 JH |
1245 | if (pages) { |
1246 | ret = PTR_ERR(page); | |
1247 | goto out; | |
1248 | } | |
1249 | ||
1027e443 KS |
1250 | goto next_page; |
1251 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1252 | ret = PTR_ERR(page); |
1253 | goto out; | |
1027e443 | 1254 | } |
fa5bb209 KS |
1255 | if (pages) { |
1256 | pages[i] = page; | |
1257 | flush_anon_page(vma, page, start); | |
1258 | flush_dcache_page(page); | |
df06b37f | 1259 | ctx.page_mask = 0; |
4bbd4c77 | 1260 | } |
4bbd4c77 | 1261 | next_page: |
df06b37f | 1262 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1263 | if (page_increm > nr_pages) |
1264 | page_increm = nr_pages; | |
1265 | i += page_increm; | |
1266 | start += page_increm * PAGE_SIZE; | |
1267 | nr_pages -= page_increm; | |
4bbd4c77 | 1268 | } while (nr_pages); |
df06b37f KB |
1269 | out: |
1270 | if (ctx.pgmap) | |
1271 | put_dev_pagemap(ctx.pgmap); | |
1272 | return i ? i : ret; | |
4bbd4c77 | 1273 | } |
4bbd4c77 | 1274 | |
771ab430 TK |
1275 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1276 | unsigned int fault_flags) | |
d4925e00 | 1277 | { |
1b2ee126 DH |
1278 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1279 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1280 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1281 | |
1282 | if (!(vm_flags & vma->vm_flags)) | |
1283 | return false; | |
1284 | ||
33a709b2 DH |
1285 | /* |
1286 | * The architecture might have a hardware protection | |
1b2ee126 | 1287 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1288 | * |
1289 | * gup always represents data access, not instruction | |
1290 | * fetches, so execute=false here: | |
33a709b2 | 1291 | */ |
d61172b4 | 1292 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1293 | return false; |
1294 | ||
d4925e00 DH |
1295 | return true; |
1296 | } | |
1297 | ||
adc8cb40 | 1298 | /** |
4bbd4c77 | 1299 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1300 | * @mm: mm_struct of target mm |
1301 | * @address: user address | |
1302 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1303 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1304 | * does not allow retry. If NULL, the caller must guarantee |
1305 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1306 | * |
1307 | * This is meant to be called in the specific scenario where for locking reasons | |
1308 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1309 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1310 | * trying again. | |
1311 | * | |
1312 | * Typically this is meant to be used by the futex code. | |
1313 | * | |
1314 | * The main difference with get_user_pages() is that this function will | |
1315 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1316 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1317 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1318 | * |
1319 | * This is important for some architectures where those bits also gate the | |
1320 | * access permission to the page because they are maintained in software. On | |
1321 | * such architectures, gup() will not be enough to make a subsequent access | |
1322 | * succeed. | |
1323 | * | |
c1e8d7c6 ML |
1324 | * This function will not return with an unlocked mmap_lock. So it has not the |
1325 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1326 | */ |
64019a2e | 1327 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1328 | unsigned long address, unsigned int fault_flags, |
1329 | bool *unlocked) | |
4bbd4c77 KS |
1330 | { |
1331 | struct vm_area_struct *vma; | |
8fed2f3c | 1332 | vm_fault_t ret; |
4a9e1cda | 1333 | |
428e106a | 1334 | address = untagged_addr_remote(mm, address); |
f9652594 | 1335 | |
4a9e1cda | 1336 | if (unlocked) |
71335f37 | 1337 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1338 | |
4a9e1cda | 1339 | retry: |
a425ac53 | 1340 | vma = find_vma(mm, address); |
8d7071af | 1341 | if (!vma) |
4bbd4c77 | 1342 | return -EFAULT; |
a425ac53 LT |
1343 | if (address < vma->vm_start ) { |
1344 | WARN_ON_ONCE(vma->vm_flags & VM_GROWSDOWN); | |
4bbd4c77 | 1345 | return -EFAULT; |
a425ac53 | 1346 | } |
4bbd4c77 | 1347 | |
d4925e00 | 1348 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1349 | return -EFAULT; |
1350 | ||
475f4dfc PX |
1351 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1352 | fatal_signal_pending(current)) | |
1353 | return -EINTR; | |
1354 | ||
bce617ed | 1355 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
d9272525 PX |
1356 | |
1357 | if (ret & VM_FAULT_COMPLETED) { | |
1358 | /* | |
1359 | * NOTE: it's a pity that we need to retake the lock here | |
1360 | * to pair with the unlock() in the callers. Ideally we | |
1361 | * could tell the callers so they do not need to unlock. | |
1362 | */ | |
1363 | mmap_read_lock(mm); | |
1364 | *unlocked = true; | |
1365 | return 0; | |
1366 | } | |
1367 | ||
4bbd4c77 | 1368 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1369 | int err = vm_fault_to_errno(ret, 0); |
1370 | ||
1371 | if (err) | |
1372 | return err; | |
4bbd4c77 KS |
1373 | BUG(); |
1374 | } | |
4a9e1cda DD |
1375 | |
1376 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1377 | mmap_read_lock(mm); |
475f4dfc PX |
1378 | *unlocked = true; |
1379 | fault_flags |= FAULT_FLAG_TRIED; | |
1380 | goto retry; | |
4a9e1cda DD |
1381 | } |
1382 | ||
4bbd4c77 KS |
1383 | return 0; |
1384 | } | |
add6a0cd | 1385 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1386 | |
93c5c61d PX |
1387 | /* |
1388 | * GUP always responds to fatal signals. When FOLL_INTERRUPTIBLE is | |
1389 | * specified, it'll also respond to generic signals. The caller of GUP | |
1390 | * that has FOLL_INTERRUPTIBLE should take care of the GUP interruption. | |
1391 | */ | |
1392 | static bool gup_signal_pending(unsigned int flags) | |
1393 | { | |
1394 | if (fatal_signal_pending(current)) | |
1395 | return true; | |
1396 | ||
1397 | if (!(flags & FOLL_INTERRUPTIBLE)) | |
1398 | return false; | |
1399 | ||
1400 | return signal_pending(current); | |
1401 | } | |
1402 | ||
2d3a36a4 | 1403 | /* |
b2a72dff JG |
1404 | * Locking: (*locked == 1) means that the mmap_lock has already been acquired by |
1405 | * the caller. This function may drop the mmap_lock. If it does so, then it will | |
1406 | * set (*locked = 0). | |
1407 | * | |
1408 | * (*locked == 0) means that the caller expects this function to acquire and | |
1409 | * drop the mmap_lock. Therefore, the value of *locked will still be zero when | |
1410 | * the function returns, even though it may have changed temporarily during | |
1411 | * function execution. | |
1412 | * | |
1413 | * Please note that this function, unlike __get_user_pages(), will not return 0 | |
1414 | * for nr_pages > 0, unless FOLL_NOWAIT is used. | |
2d3a36a4 | 1415 | */ |
64019a2e | 1416 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1417 | unsigned long start, |
1418 | unsigned long nr_pages, | |
f0818f47 | 1419 | struct page **pages, |
e716712f | 1420 | int *locked, |
0fd71a56 | 1421 | unsigned int flags) |
f0818f47 | 1422 | { |
f0818f47 | 1423 | long ret, pages_done; |
b2a72dff | 1424 | bool must_unlock = false; |
f0818f47 | 1425 | |
b2a72dff JG |
1426 | /* |
1427 | * The internal caller expects GUP to manage the lock internally and the | |
1428 | * lock must be released when this returns. | |
1429 | */ | |
9a863a6a | 1430 | if (!*locked) { |
b2a72dff JG |
1431 | if (mmap_read_lock_killable(mm)) |
1432 | return -EAGAIN; | |
1433 | must_unlock = true; | |
1434 | *locked = 1; | |
f0818f47 | 1435 | } |
961ba472 JG |
1436 | else |
1437 | mmap_assert_locked(mm); | |
f0818f47 | 1438 | |
a458b76a AA |
1439 | if (flags & FOLL_PIN) |
1440 | mm_set_has_pinned_flag(&mm->flags); | |
008cfe44 | 1441 | |
eddb1c22 JH |
1442 | /* |
1443 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1444 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1445 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1446 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1447 | * | |
1448 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1449 | * that here, as any failures will be obvious enough. | |
1450 | */ | |
1451 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1452 | flags |= FOLL_GET; |
f0818f47 AA |
1453 | |
1454 | pages_done = 0; | |
f0818f47 | 1455 | for (;;) { |
64019a2e | 1456 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
b2cac248 | 1457 | locked); |
f04740f5 | 1458 | if (!(flags & FOLL_UNLOCKABLE)) { |
f0818f47 | 1459 | /* VM_FAULT_RETRY couldn't trigger, bypass */ |
f04740f5 JG |
1460 | pages_done = ret; |
1461 | break; | |
1462 | } | |
f0818f47 | 1463 | |
d9272525 | 1464 | /* VM_FAULT_RETRY or VM_FAULT_COMPLETED cannot return errors */ |
f0818f47 AA |
1465 | if (!*locked) { |
1466 | BUG_ON(ret < 0); | |
1467 | BUG_ON(ret >= nr_pages); | |
1468 | } | |
1469 | ||
f0818f47 AA |
1470 | if (ret > 0) { |
1471 | nr_pages -= ret; | |
1472 | pages_done += ret; | |
1473 | if (!nr_pages) | |
1474 | break; | |
1475 | } | |
1476 | if (*locked) { | |
96312e61 AA |
1477 | /* |
1478 | * VM_FAULT_RETRY didn't trigger or it was a | |
1479 | * FOLL_NOWAIT. | |
1480 | */ | |
f0818f47 AA |
1481 | if (!pages_done) |
1482 | pages_done = ret; | |
1483 | break; | |
1484 | } | |
df17277b MR |
1485 | /* |
1486 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1487 | * For the prefault case (!pages) we only update counts. | |
1488 | */ | |
1489 | if (likely(pages)) | |
1490 | pages += ret; | |
f0818f47 | 1491 | start += ret << PAGE_SHIFT; |
b2a72dff JG |
1492 | |
1493 | /* The lock was temporarily dropped, so we must unlock later */ | |
1494 | must_unlock = true; | |
f0818f47 | 1495 | |
4426e945 | 1496 | retry: |
f0818f47 AA |
1497 | /* |
1498 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1499 | * with both FAULT_FLAG_ALLOW_RETRY and |
1500 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
93c5c61d PX |
1501 | * by fatal signals of even common signals, depending on |
1502 | * the caller's request. So we need to check it before we | |
4426e945 | 1503 | * start trying again otherwise it can loop forever. |
f0818f47 | 1504 | */ |
93c5c61d | 1505 | if (gup_signal_pending(flags)) { |
ae46d2aa HD |
1506 | if (!pages_done) |
1507 | pages_done = -EINTR; | |
4426e945 | 1508 | break; |
ae46d2aa | 1509 | } |
4426e945 | 1510 | |
d8ed45c5 | 1511 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1512 | if (ret) { |
1513 | BUG_ON(ret > 0); | |
1514 | if (!pages_done) | |
1515 | pages_done = ret; | |
1516 | break; | |
1517 | } | |
4426e945 | 1518 | |
c7b6a566 | 1519 | *locked = 1; |
64019a2e | 1520 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
b2cac248 | 1521 | pages, locked); |
4426e945 PX |
1522 | if (!*locked) { |
1523 | /* Continue to retry until we succeeded */ | |
1524 | BUG_ON(ret != 0); | |
1525 | goto retry; | |
1526 | } | |
f0818f47 AA |
1527 | if (ret != 1) { |
1528 | BUG_ON(ret > 1); | |
1529 | if (!pages_done) | |
1530 | pages_done = ret; | |
1531 | break; | |
1532 | } | |
1533 | nr_pages--; | |
1534 | pages_done++; | |
1535 | if (!nr_pages) | |
1536 | break; | |
df17277b MR |
1537 | if (likely(pages)) |
1538 | pages++; | |
f0818f47 AA |
1539 | start += PAGE_SIZE; |
1540 | } | |
b2a72dff | 1541 | if (must_unlock && *locked) { |
f0818f47 | 1542 | /* |
b2a72dff JG |
1543 | * We either temporarily dropped the lock, or the caller |
1544 | * requested that we both acquire and drop the lock. Either way, | |
1545 | * we must now unlock, and notify the caller of that state. | |
f0818f47 | 1546 | */ |
d8ed45c5 | 1547 | mmap_read_unlock(mm); |
f0818f47 AA |
1548 | *locked = 0; |
1549 | } | |
1550 | return pages_done; | |
1551 | } | |
1552 | ||
d3649f68 CH |
1553 | /** |
1554 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1555 | * @vma: target vma | |
1556 | * @start: start address | |
1557 | * @end: end address | |
c1e8d7c6 | 1558 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1559 | * |
1560 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1561 | * | |
0a36f7f8 TY |
1562 | * Return either number of pages pinned in the vma, or a negative error |
1563 | * code on error. | |
d3649f68 | 1564 | * |
c1e8d7c6 | 1565 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1566 | * |
4f6da934 | 1567 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1568 | * be unperturbed. |
1569 | * | |
4f6da934 PX |
1570 | * If @locked is non-NULL, it must held for read only and may be |
1571 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1572 | */ |
1573 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1574 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1575 | { |
1576 | struct mm_struct *mm = vma->vm_mm; | |
1577 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
9a863a6a | 1578 | int local_locked = 1; |
d3649f68 | 1579 | int gup_flags; |
ece369c7 | 1580 | long ret; |
d3649f68 | 1581 | |
be51eb18 ML |
1582 | VM_BUG_ON(!PAGE_ALIGNED(start)); |
1583 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
d3649f68 CH |
1584 | VM_BUG_ON_VMA(start < vma->vm_start, vma); |
1585 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1586 | mmap_assert_locked(mm); |
d3649f68 | 1587 | |
b67bf49c HD |
1588 | /* |
1589 | * Rightly or wrongly, the VM_LOCKONFAULT case has never used | |
1590 | * faultin_page() to break COW, so it has no work to do here. | |
1591 | */ | |
d3649f68 | 1592 | if (vma->vm_flags & VM_LOCKONFAULT) |
b67bf49c HD |
1593 | return nr_pages; |
1594 | ||
1595 | gup_flags = FOLL_TOUCH; | |
d3649f68 CH |
1596 | /* |
1597 | * We want to touch writable mappings with a write fault in order | |
1598 | * to break COW, except for shared mappings because these don't COW | |
1599 | * and we would not want to dirty them for nothing. | |
1600 | */ | |
1601 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1602 | gup_flags |= FOLL_WRITE; | |
1603 | ||
1604 | /* | |
1605 | * We want mlock to succeed for regions that have any permissions | |
1606 | * other than PROT_NONE. | |
1607 | */ | |
3122e80e | 1608 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1609 | gup_flags |= FOLL_FORCE; |
1610 | ||
f04740f5 JG |
1611 | if (locked) |
1612 | gup_flags |= FOLL_UNLOCKABLE; | |
1613 | ||
d3649f68 CH |
1614 | /* |
1615 | * We made sure addr is within a VMA, so the following will | |
1616 | * not result in a stack expansion that recurses back here. | |
1617 | */ | |
ece369c7 | 1618 | ret = __get_user_pages(mm, start, nr_pages, gup_flags, |
b2cac248 | 1619 | NULL, locked ? locked : &local_locked); |
ece369c7 HD |
1620 | lru_add_drain(); |
1621 | return ret; | |
d3649f68 CH |
1622 | } |
1623 | ||
4ca9b385 DH |
1624 | /* |
1625 | * faultin_vma_page_range() - populate (prefault) page tables inside the | |
1626 | * given VMA range readable/writable | |
1627 | * | |
1628 | * This takes care of mlocking the pages, too, if VM_LOCKED is set. | |
1629 | * | |
1630 | * @vma: target vma | |
1631 | * @start: start address | |
1632 | * @end: end address | |
1633 | * @write: whether to prefault readable or writable | |
1634 | * @locked: whether the mmap_lock is still held | |
1635 | * | |
1636 | * Returns either number of processed pages in the vma, or a negative error | |
1637 | * code on error (see __get_user_pages()). | |
1638 | * | |
1639 | * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and | |
6e4382c7 | 1640 | * covered by the VMA. If it's released, *@locked will be set to 0. |
4ca9b385 DH |
1641 | */ |
1642 | long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start, | |
1643 | unsigned long end, bool write, int *locked) | |
1644 | { | |
1645 | struct mm_struct *mm = vma->vm_mm; | |
1646 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1647 | int gup_flags; | |
ece369c7 | 1648 | long ret; |
4ca9b385 DH |
1649 | |
1650 | VM_BUG_ON(!PAGE_ALIGNED(start)); | |
1651 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
1652 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1653 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
1654 | mmap_assert_locked(mm); | |
1655 | ||
1656 | /* | |
1657 | * FOLL_TOUCH: Mark page accessed and thereby young; will also mark | |
1658 | * the page dirty with FOLL_WRITE -- which doesn't make a | |
1659 | * difference with !FOLL_FORCE, because the page is writable | |
1660 | * in the page table. | |
1661 | * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit | |
1662 | * a poisoned page. | |
4ca9b385 DH |
1663 | * !FOLL_FORCE: Require proper access permissions. |
1664 | */ | |
f04740f5 | 1665 | gup_flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_UNLOCKABLE; |
4ca9b385 DH |
1666 | if (write) |
1667 | gup_flags |= FOLL_WRITE; | |
1668 | ||
1669 | /* | |
eb2faa51 DH |
1670 | * We want to report -EINVAL instead of -EFAULT for any permission |
1671 | * problems or incompatible mappings. | |
4ca9b385 | 1672 | */ |
eb2faa51 DH |
1673 | if (check_vma_flags(vma, gup_flags)) |
1674 | return -EINVAL; | |
1675 | ||
ece369c7 | 1676 | ret = __get_user_pages(mm, start, nr_pages, gup_flags, |
b2cac248 | 1677 | NULL, locked); |
ece369c7 HD |
1678 | lru_add_drain(); |
1679 | return ret; | |
4ca9b385 DH |
1680 | } |
1681 | ||
d3649f68 CH |
1682 | /* |
1683 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1684 | * | |
1685 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1686 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1687 | * mmap_lock must not be held. |
d3649f68 CH |
1688 | */ |
1689 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1690 | { | |
1691 | struct mm_struct *mm = current->mm; | |
1692 | unsigned long end, nstart, nend; | |
1693 | struct vm_area_struct *vma = NULL; | |
1694 | int locked = 0; | |
1695 | long ret = 0; | |
1696 | ||
1697 | end = start + len; | |
1698 | ||
1699 | for (nstart = start; nstart < end; nstart = nend) { | |
1700 | /* | |
1701 | * We want to fault in pages for [nstart; end) address range. | |
1702 | * Find first corresponding VMA. | |
1703 | */ | |
1704 | if (!locked) { | |
1705 | locked = 1; | |
d8ed45c5 | 1706 | mmap_read_lock(mm); |
c4d1a92d | 1707 | vma = find_vma_intersection(mm, nstart, end); |
d3649f68 | 1708 | } else if (nstart >= vma->vm_end) |
c4d1a92d LH |
1709 | vma = find_vma_intersection(mm, vma->vm_end, end); |
1710 | ||
1711 | if (!vma) | |
d3649f68 CH |
1712 | break; |
1713 | /* | |
1714 | * Set [nstart; nend) to intersection of desired address | |
1715 | * range with the first VMA. Also, skip undesirable VMA types. | |
1716 | */ | |
1717 | nend = min(end, vma->vm_end); | |
1718 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1719 | continue; | |
1720 | if (nstart < vma->vm_start) | |
1721 | nstart = vma->vm_start; | |
1722 | /* | |
1723 | * Now fault in a range of pages. populate_vma_page_range() | |
1724 | * double checks the vma flags, so that it won't mlock pages | |
1725 | * if the vma was already munlocked. | |
1726 | */ | |
1727 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1728 | if (ret < 0) { | |
1729 | if (ignore_errors) { | |
1730 | ret = 0; | |
1731 | continue; /* continue at next VMA */ | |
1732 | } | |
1733 | break; | |
1734 | } | |
1735 | nend = nstart + ret * PAGE_SIZE; | |
1736 | ret = 0; | |
1737 | } | |
1738 | if (locked) | |
d8ed45c5 | 1739 | mmap_read_unlock(mm); |
d3649f68 CH |
1740 | return ret; /* 0 or negative error code */ |
1741 | } | |
050a9adc | 1742 | #else /* CONFIG_MMU */ |
64019a2e | 1743 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc | 1744 | unsigned long nr_pages, struct page **pages, |
b2cac248 | 1745 | int *locked, unsigned int foll_flags) |
050a9adc CH |
1746 | { |
1747 | struct vm_area_struct *vma; | |
b2a72dff | 1748 | bool must_unlock = false; |
050a9adc | 1749 | unsigned long vm_flags; |
24dc20c7 | 1750 | long i; |
050a9adc | 1751 | |
b2a72dff JG |
1752 | if (!nr_pages) |
1753 | return 0; | |
1754 | ||
1755 | /* | |
1756 | * The internal caller expects GUP to manage the lock internally and the | |
1757 | * lock must be released when this returns. | |
1758 | */ | |
9a863a6a | 1759 | if (!*locked) { |
b2a72dff JG |
1760 | if (mmap_read_lock_killable(mm)) |
1761 | return -EAGAIN; | |
1762 | must_unlock = true; | |
1763 | *locked = 1; | |
1764 | } | |
1765 | ||
050a9adc CH |
1766 | /* calculate required read or write permissions. |
1767 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1768 | */ | |
1769 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1770 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1771 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1772 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1773 | ||
1774 | for (i = 0; i < nr_pages; i++) { | |
1775 | vma = find_vma(mm, start); | |
1776 | if (!vma) | |
b2a72dff | 1777 | break; |
050a9adc CH |
1778 | |
1779 | /* protect what we can, including chardevs */ | |
1780 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1781 | !(vm_flags & vma->vm_flags)) | |
b2a72dff | 1782 | break; |
050a9adc CH |
1783 | |
1784 | if (pages) { | |
396a400b | 1785 | pages[i] = virt_to_page((void *)start); |
050a9adc CH |
1786 | if (pages[i]) |
1787 | get_page(pages[i]); | |
1788 | } | |
b2cac248 | 1789 | |
050a9adc CH |
1790 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1791 | } | |
1792 | ||
b2a72dff JG |
1793 | if (must_unlock && *locked) { |
1794 | mmap_read_unlock(mm); | |
1795 | *locked = 0; | |
1796 | } | |
050a9adc | 1797 | |
050a9adc CH |
1798 | return i ? : -EFAULT; |
1799 | } | |
1800 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1801 | |
bb523b40 AG |
1802 | /** |
1803 | * fault_in_writeable - fault in userspace address range for writing | |
1804 | * @uaddr: start of address range | |
1805 | * @size: size of address range | |
1806 | * | |
1807 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1808 | * copy_from_user()). | |
1809 | */ | |
1810 | size_t fault_in_writeable(char __user *uaddr, size_t size) | |
1811 | { | |
1812 | char __user *start = uaddr, *end; | |
1813 | ||
1814 | if (unlikely(size == 0)) | |
1815 | return 0; | |
677b2a8c CL |
1816 | if (!user_write_access_begin(uaddr, size)) |
1817 | return size; | |
bb523b40 | 1818 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1819 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1820 | uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1821 | } | |
1822 | end = (char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1823 | if (unlikely(end < start)) | |
1824 | end = NULL; | |
1825 | while (uaddr != end) { | |
677b2a8c | 1826 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1827 | uaddr += PAGE_SIZE; |
1828 | } | |
1829 | ||
1830 | out: | |
677b2a8c | 1831 | user_write_access_end(); |
bb523b40 AG |
1832 | if (size > uaddr - start) |
1833 | return size - (uaddr - start); | |
1834 | return 0; | |
1835 | } | |
1836 | EXPORT_SYMBOL(fault_in_writeable); | |
1837 | ||
da32b581 CM |
1838 | /** |
1839 | * fault_in_subpage_writeable - fault in an address range for writing | |
1840 | * @uaddr: start of address range | |
1841 | * @size: size of address range | |
1842 | * | |
1843 | * Fault in a user address range for writing while checking for permissions at | |
1844 | * sub-page granularity (e.g. arm64 MTE). This function should be used when | |
1845 | * the caller cannot guarantee forward progress of a copy_to_user() loop. | |
1846 | * | |
1847 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1848 | * copy_from_user()). | |
1849 | */ | |
1850 | size_t fault_in_subpage_writeable(char __user *uaddr, size_t size) | |
1851 | { | |
1852 | size_t faulted_in; | |
1853 | ||
1854 | /* | |
1855 | * Attempt faulting in at page granularity first for page table | |
1856 | * permission checking. The arch-specific probe_subpage_writeable() | |
1857 | * functions may not check for this. | |
1858 | */ | |
1859 | faulted_in = size - fault_in_writeable(uaddr, size); | |
1860 | if (faulted_in) | |
1861 | faulted_in -= probe_subpage_writeable(uaddr, faulted_in); | |
1862 | ||
1863 | return size - faulted_in; | |
1864 | } | |
1865 | EXPORT_SYMBOL(fault_in_subpage_writeable); | |
1866 | ||
cdd591fc AG |
1867 | /* |
1868 | * fault_in_safe_writeable - fault in an address range for writing | |
1869 | * @uaddr: start of address range | |
1870 | * @size: length of address range | |
1871 | * | |
fe673d3f LT |
1872 | * Faults in an address range for writing. This is primarily useful when we |
1873 | * already know that some or all of the pages in the address range aren't in | |
1874 | * memory. | |
cdd591fc | 1875 | * |
fe673d3f | 1876 | * Unlike fault_in_writeable(), this function is non-destructive. |
cdd591fc AG |
1877 | * |
1878 | * Note that we don't pin or otherwise hold the pages referenced that we fault | |
1879 | * in. There's no guarantee that they'll stay in memory for any duration of | |
1880 | * time. | |
1881 | * | |
1882 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
1883 | * copy_from_user(). | |
1884 | */ | |
1885 | size_t fault_in_safe_writeable(const char __user *uaddr, size_t size) | |
1886 | { | |
fe673d3f | 1887 | unsigned long start = (unsigned long)uaddr, end; |
cdd591fc | 1888 | struct mm_struct *mm = current->mm; |
fe673d3f | 1889 | bool unlocked = false; |
cdd591fc | 1890 | |
fe673d3f LT |
1891 | if (unlikely(size == 0)) |
1892 | return 0; | |
cdd591fc | 1893 | end = PAGE_ALIGN(start + size); |
fe673d3f | 1894 | if (end < start) |
cdd591fc | 1895 | end = 0; |
cdd591fc | 1896 | |
fe673d3f LT |
1897 | mmap_read_lock(mm); |
1898 | do { | |
1899 | if (fixup_user_fault(mm, start, FAULT_FLAG_WRITE, &unlocked)) | |
cdd591fc | 1900 | break; |
fe673d3f LT |
1901 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1902 | } while (start != end); | |
1903 | mmap_read_unlock(mm); | |
1904 | ||
1905 | if (size > (unsigned long)uaddr - start) | |
1906 | return size - ((unsigned long)uaddr - start); | |
1907 | return 0; | |
cdd591fc AG |
1908 | } |
1909 | EXPORT_SYMBOL(fault_in_safe_writeable); | |
1910 | ||
bb523b40 AG |
1911 | /** |
1912 | * fault_in_readable - fault in userspace address range for reading | |
1913 | * @uaddr: start of user address range | |
1914 | * @size: size of user address range | |
1915 | * | |
1916 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1917 | * copy_from_user()). | |
1918 | */ | |
1919 | size_t fault_in_readable(const char __user *uaddr, size_t size) | |
1920 | { | |
1921 | const char __user *start = uaddr, *end; | |
1922 | volatile char c; | |
1923 | ||
1924 | if (unlikely(size == 0)) | |
1925 | return 0; | |
677b2a8c CL |
1926 | if (!user_read_access_begin(uaddr, size)) |
1927 | return size; | |
bb523b40 | 1928 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1929 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1930 | uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1931 | } | |
1932 | end = (const char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1933 | if (unlikely(end < start)) | |
1934 | end = NULL; | |
1935 | while (uaddr != end) { | |
677b2a8c | 1936 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1937 | uaddr += PAGE_SIZE; |
1938 | } | |
1939 | ||
1940 | out: | |
677b2a8c | 1941 | user_read_access_end(); |
bb523b40 AG |
1942 | (void)c; |
1943 | if (size > uaddr - start) | |
1944 | return size - (uaddr - start); | |
1945 | return 0; | |
1946 | } | |
1947 | EXPORT_SYMBOL(fault_in_readable); | |
1948 | ||
8f942eea JH |
1949 | /** |
1950 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1951 | * @addr: user address | |
1952 | * | |
1953 | * Returns struct page pointer of user page pinned for dump, | |
1954 | * to be freed afterwards by put_page(). | |
1955 | * | |
1956 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1957 | * the corefile, to preserve alignment with its headers; and also returns | |
1958 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
f0953a1b | 1959 | * allowing a hole to be left in the corefile to save disk space. |
8f942eea | 1960 | * |
7f3bfab5 | 1961 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1962 | */ |
1963 | #ifdef CONFIG_ELF_CORE | |
1964 | struct page *get_dump_page(unsigned long addr) | |
1965 | { | |
8f942eea | 1966 | struct page *page; |
b2a72dff | 1967 | int locked = 0; |
7f3bfab5 | 1968 | int ret; |
8f942eea | 1969 | |
b2cac248 | 1970 | ret = __get_user_pages_locked(current->mm, addr, 1, &page, &locked, |
7f3bfab5 | 1971 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); |
7f3bfab5 | 1972 | return (ret == 1) ? page : NULL; |
8f942eea JH |
1973 | } |
1974 | #endif /* CONFIG_ELF_CORE */ | |
1975 | ||
d1e153fe | 1976 | #ifdef CONFIG_MIGRATION |
f68749ec | 1977 | /* |
67e139b0 | 1978 | * Returns the number of collected pages. Return value is always >= 0. |
f68749ec | 1979 | */ |
67e139b0 AP |
1980 | static unsigned long collect_longterm_unpinnable_pages( |
1981 | struct list_head *movable_page_list, | |
1982 | unsigned long nr_pages, | |
1983 | struct page **pages) | |
9a4e9f3b | 1984 | { |
67e139b0 | 1985 | unsigned long i, collected = 0; |
1b7f7e58 | 1986 | struct folio *prev_folio = NULL; |
67e139b0 | 1987 | bool drain_allow = true; |
9a4e9f3b | 1988 | |
83c02c23 | 1989 | for (i = 0; i < nr_pages; i++) { |
1b7f7e58 | 1990 | struct folio *folio = page_folio(pages[i]); |
f9f38f78 | 1991 | |
1b7f7e58 | 1992 | if (folio == prev_folio) |
83c02c23 | 1993 | continue; |
1b7f7e58 | 1994 | prev_folio = folio; |
f9f38f78 | 1995 | |
67e139b0 AP |
1996 | if (folio_is_longterm_pinnable(folio)) |
1997 | continue; | |
b05a79d4 | 1998 | |
67e139b0 | 1999 | collected++; |
b05a79d4 | 2000 | |
67e139b0 | 2001 | if (folio_is_device_coherent(folio)) |
f9f38f78 CH |
2002 | continue; |
2003 | ||
1b7f7e58 | 2004 | if (folio_test_hugetlb(folio)) { |
6aa3a920 | 2005 | isolate_hugetlb(folio, movable_page_list); |
f9f38f78 CH |
2006 | continue; |
2007 | } | |
9a4e9f3b | 2008 | |
1b7f7e58 | 2009 | if (!folio_test_lru(folio) && drain_allow) { |
f9f38f78 CH |
2010 | lru_add_drain_all(); |
2011 | drain_allow = false; | |
2012 | } | |
2013 | ||
be2d5756 | 2014 | if (!folio_isolate_lru(folio)) |
f9f38f78 | 2015 | continue; |
67e139b0 AP |
2016 | |
2017 | list_add_tail(&folio->lru, movable_page_list); | |
1b7f7e58 MWO |
2018 | node_stat_mod_folio(folio, |
2019 | NR_ISOLATED_ANON + folio_is_file_lru(folio), | |
2020 | folio_nr_pages(folio)); | |
9a4e9f3b AK |
2021 | } |
2022 | ||
67e139b0 AP |
2023 | return collected; |
2024 | } | |
2025 | ||
2026 | /* | |
2027 | * Unpins all pages and migrates device coherent pages and movable_page_list. | |
2028 | * Returns -EAGAIN if all pages were successfully migrated or -errno for failure | |
2029 | * (or partial success). | |
2030 | */ | |
2031 | static int migrate_longterm_unpinnable_pages( | |
2032 | struct list_head *movable_page_list, | |
2033 | unsigned long nr_pages, | |
2034 | struct page **pages) | |
2035 | { | |
2036 | int ret; | |
2037 | unsigned long i; | |
6e7f34eb | 2038 | |
b05a79d4 | 2039 | for (i = 0; i < nr_pages; i++) { |
67e139b0 AP |
2040 | struct folio *folio = page_folio(pages[i]); |
2041 | ||
2042 | if (folio_is_device_coherent(folio)) { | |
2043 | /* | |
2044 | * Migration will fail if the page is pinned, so convert | |
2045 | * the pin on the source page to a normal reference. | |
2046 | */ | |
2047 | pages[i] = NULL; | |
2048 | folio_get(folio); | |
2049 | gup_put_folio(folio, 1, FOLL_PIN); | |
2050 | ||
2051 | if (migrate_device_coherent_page(&folio->page)) { | |
2052 | ret = -EBUSY; | |
2053 | goto err; | |
2054 | } | |
2055 | ||
b05a79d4 | 2056 | continue; |
67e139b0 | 2057 | } |
b05a79d4 | 2058 | |
67e139b0 AP |
2059 | /* |
2060 | * We can't migrate pages with unexpected references, so drop | |
2061 | * the reference obtained by __get_user_pages_locked(). | |
2062 | * Migrating pages have been added to movable_page_list after | |
2063 | * calling folio_isolate_lru() which takes a reference so the | |
2064 | * page won't be freed if it's migrating. | |
2065 | */ | |
f6d299ec | 2066 | unpin_user_page(pages[i]); |
67e139b0 | 2067 | pages[i] = NULL; |
f68749ec | 2068 | } |
f9f38f78 | 2069 | |
67e139b0 | 2070 | if (!list_empty(movable_page_list)) { |
f9f38f78 CH |
2071 | struct migration_target_control mtc = { |
2072 | .nid = NUMA_NO_NODE, | |
2073 | .gfp_mask = GFP_USER | __GFP_NOWARN, | |
2074 | }; | |
2075 | ||
67e139b0 AP |
2076 | if (migrate_pages(movable_page_list, alloc_migration_target, |
2077 | NULL, (unsigned long)&mtc, MIGRATE_SYNC, | |
2078 | MR_LONGTERM_PIN, NULL)) { | |
f9f38f78 | 2079 | ret = -ENOMEM; |
67e139b0 AP |
2080 | goto err; |
2081 | } | |
9a4e9f3b AK |
2082 | } |
2083 | ||
67e139b0 AP |
2084 | putback_movable_pages(movable_page_list); |
2085 | ||
2086 | return -EAGAIN; | |
2087 | ||
2088 | err: | |
2089 | for (i = 0; i < nr_pages; i++) | |
2090 | if (pages[i]) | |
2091 | unpin_user_page(pages[i]); | |
2092 | putback_movable_pages(movable_page_list); | |
24a95998 | 2093 | |
67e139b0 AP |
2094 | return ret; |
2095 | } | |
2096 | ||
2097 | /* | |
2098 | * Check whether all pages are *allowed* to be pinned. Rather confusingly, all | |
2099 | * pages in the range are required to be pinned via FOLL_PIN, before calling | |
2100 | * this routine. | |
2101 | * | |
2102 | * If any pages in the range are not allowed to be pinned, then this routine | |
2103 | * will migrate those pages away, unpin all the pages in the range and return | |
2104 | * -EAGAIN. The caller should re-pin the entire range with FOLL_PIN and then | |
2105 | * call this routine again. | |
2106 | * | |
2107 | * If an error other than -EAGAIN occurs, this indicates a migration failure. | |
2108 | * The caller should give up, and propagate the error back up the call stack. | |
2109 | * | |
2110 | * If everything is OK and all pages in the range are allowed to be pinned, then | |
2111 | * this routine leaves all pages pinned and returns zero for success. | |
2112 | */ | |
2113 | static long check_and_migrate_movable_pages(unsigned long nr_pages, | |
2114 | struct page **pages) | |
2115 | { | |
2116 | unsigned long collected; | |
2117 | LIST_HEAD(movable_page_list); | |
2118 | ||
2119 | collected = collect_longterm_unpinnable_pages(&movable_page_list, | |
2120 | nr_pages, pages); | |
2121 | if (!collected) | |
2122 | return 0; | |
2123 | ||
2124 | return migrate_longterm_unpinnable_pages(&movable_page_list, nr_pages, | |
2125 | pages); | |
9a4e9f3b AK |
2126 | } |
2127 | #else | |
f68749ec | 2128 | static long check_and_migrate_movable_pages(unsigned long nr_pages, |
f6d299ec | 2129 | struct page **pages) |
9a4e9f3b | 2130 | { |
24a95998 | 2131 | return 0; |
9a4e9f3b | 2132 | } |
d1e153fe | 2133 | #endif /* CONFIG_MIGRATION */ |
9a4e9f3b | 2134 | |
2bb6d283 | 2135 | /* |
932f4a63 IW |
2136 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
2137 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 2138 | */ |
64019a2e | 2139 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
2140 | unsigned long start, |
2141 | unsigned long nr_pages, | |
2142 | struct page **pages, | |
53b2d09b | 2143 | int *locked, |
932f4a63 | 2144 | unsigned int gup_flags) |
2bb6d283 | 2145 | { |
f68749ec | 2146 | unsigned int flags; |
24a95998 | 2147 | long rc, nr_pinned_pages; |
2bb6d283 | 2148 | |
f68749ec | 2149 | if (!(gup_flags & FOLL_LONGTERM)) |
b2cac248 | 2150 | return __get_user_pages_locked(mm, start, nr_pages, pages, |
53b2d09b | 2151 | locked, gup_flags); |
67e139b0 | 2152 | |
f68749ec PT |
2153 | flags = memalloc_pin_save(); |
2154 | do { | |
24a95998 | 2155 | nr_pinned_pages = __get_user_pages_locked(mm, start, nr_pages, |
b2cac248 | 2156 | pages, locked, |
24a95998 AP |
2157 | gup_flags); |
2158 | if (nr_pinned_pages <= 0) { | |
2159 | rc = nr_pinned_pages; | |
f68749ec | 2160 | break; |
24a95998 | 2161 | } |
d64e2dbc JG |
2162 | |
2163 | /* FOLL_LONGTERM implies FOLL_PIN */ | |
f6d299ec | 2164 | rc = check_and_migrate_movable_pages(nr_pinned_pages, pages); |
24a95998 | 2165 | } while (rc == -EAGAIN); |
f68749ec | 2166 | memalloc_pin_restore(flags); |
24a95998 | 2167 | return rc ? rc : nr_pinned_pages; |
2bb6d283 | 2168 | } |
932f4a63 | 2169 | |
d64e2dbc JG |
2170 | /* |
2171 | * Check that the given flags are valid for the exported gup/pup interface, and | |
2172 | * update them with the required flags that the caller must have set. | |
2173 | */ | |
b2cac248 LS |
2174 | static bool is_valid_gup_args(struct page **pages, int *locked, |
2175 | unsigned int *gup_flags_p, unsigned int to_set) | |
447f3e45 | 2176 | { |
d64e2dbc JG |
2177 | unsigned int gup_flags = *gup_flags_p; |
2178 | ||
447f3e45 | 2179 | /* |
d64e2dbc JG |
2180 | * These flags not allowed to be specified externally to the gup |
2181 | * interfaces: | |
2182 | * - FOLL_PIN/FOLL_TRIED/FOLL_FAST_ONLY are internal only | |
2183 | * - FOLL_REMOTE is internal only and used on follow_page() | |
f04740f5 | 2184 | * - FOLL_UNLOCKABLE is internal only and used if locked is !NULL |
447f3e45 | 2185 | */ |
f04740f5 | 2186 | if (WARN_ON_ONCE(gup_flags & (FOLL_PIN | FOLL_TRIED | FOLL_UNLOCKABLE | |
d64e2dbc JG |
2187 | FOLL_REMOTE | FOLL_FAST_ONLY))) |
2188 | return false; | |
2189 | ||
2190 | gup_flags |= to_set; | |
f04740f5 JG |
2191 | if (locked) { |
2192 | /* At the external interface locked must be set */ | |
2193 | if (WARN_ON_ONCE(*locked != 1)) | |
2194 | return false; | |
2195 | ||
2196 | gup_flags |= FOLL_UNLOCKABLE; | |
2197 | } | |
d64e2dbc JG |
2198 | |
2199 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
2200 | if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) == | |
2201 | (FOLL_PIN | FOLL_GET))) | |
2202 | return false; | |
2203 | ||
2204 | /* LONGTERM can only be specified when pinning */ | |
2205 | if (WARN_ON_ONCE(!(gup_flags & FOLL_PIN) && (gup_flags & FOLL_LONGTERM))) | |
2206 | return false; | |
2207 | ||
2208 | /* Pages input must be given if using GET/PIN */ | |
2209 | if (WARN_ON_ONCE((gup_flags & (FOLL_GET | FOLL_PIN)) && !pages)) | |
447f3e45 | 2210 | return false; |
d64e2dbc | 2211 | |
d64e2dbc JG |
2212 | /* We want to allow the pgmap to be hot-unplugged at all times */ |
2213 | if (WARN_ON_ONCE((gup_flags & FOLL_LONGTERM) && | |
2214 | (gup_flags & FOLL_PCI_P2PDMA))) | |
2215 | return false; | |
2216 | ||
d64e2dbc | 2217 | *gup_flags_p = gup_flags; |
447f3e45 BS |
2218 | return true; |
2219 | } | |
2220 | ||
22bf29b6 | 2221 | #ifdef CONFIG_MMU |
adc8cb40 | 2222 | /** |
c4237f8b | 2223 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
2224 | * @mm: mm_struct of target mm |
2225 | * @start: starting user address | |
2226 | * @nr_pages: number of pages from start to pin | |
2227 | * @gup_flags: flags modifying lookup behaviour | |
2228 | * @pages: array that receives pointers to the pages pinned. | |
2229 | * Should be at least nr_pages long. Or NULL, if caller | |
2230 | * only intends to ensure the pages are faulted in. | |
c4237f8b JH |
2231 | * @locked: pointer to lock flag indicating whether lock is held and |
2232 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2233 | * utilised. Lock must initially be held. | |
2234 | * | |
2235 | * Returns either number of pages pinned (which may be less than the | |
2236 | * number requested), or an error. Details about the return value: | |
2237 | * | |
2238 | * -- If nr_pages is 0, returns 0. | |
2239 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
2240 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
2241 | * pages pinned. Again, this may be less than nr_pages. | |
2242 | * | |
2243 | * The caller is responsible for releasing returned @pages, via put_page(). | |
2244 | * | |
c1e8d7c6 | 2245 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 2246 | * |
adc8cb40 SJ |
2247 | * get_user_pages_remote walks a process's page tables and takes a reference |
2248 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
2249 | * instant. That is, it takes the page that would be accessed if a user |
2250 | * thread accesses the given user virtual address at that instant. | |
2251 | * | |
2252 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 2253 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b | 2254 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
5da1a868 | 2255 | * and subsequently re-faulted). However it does guarantee that the page |
c4237f8b JH |
2256 | * won't be freed completely. And mostly callers simply care that the page |
2257 | * contains data that was valid *at some point in time*. Typically, an IO | |
2258 | * or similar operation cannot guarantee anything stronger anyway because | |
2259 | * locks can't be held over the syscall boundary. | |
2260 | * | |
2261 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
2262 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
2263 | * be called after the page is finished with, and before put_page is called. | |
2264 | * | |
adc8cb40 SJ |
2265 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
2266 | * to get a handle on the memory by some means other than accesses | |
2267 | * via the user virtual addresses. The pages may be submitted for | |
2268 | * DMA to devices or accessed via their kernel linear mapping (via the | |
2269 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
2270 | * |
2271 | * See also get_user_pages_fast, for performance critical applications. | |
2272 | * | |
adc8cb40 | 2273 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 2274 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 2275 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
2276 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
2277 | */ | |
64019a2e | 2278 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
2279 | unsigned long start, unsigned long nr_pages, |
2280 | unsigned int gup_flags, struct page **pages, | |
ca5e8632 | 2281 | int *locked) |
c4237f8b | 2282 | { |
9a863a6a JG |
2283 | int local_locked = 1; |
2284 | ||
b2cac248 | 2285 | if (!is_valid_gup_args(pages, locked, &gup_flags, |
d64e2dbc | 2286 | FOLL_TOUCH | FOLL_REMOTE)) |
eddb1c22 JH |
2287 | return -EINVAL; |
2288 | ||
b2cac248 | 2289 | return __get_user_pages_locked(mm, start, nr_pages, pages, |
9a863a6a | 2290 | locked ? locked : &local_locked, |
d64e2dbc | 2291 | gup_flags); |
c4237f8b JH |
2292 | } |
2293 | EXPORT_SYMBOL(get_user_pages_remote); | |
2294 | ||
eddb1c22 | 2295 | #else /* CONFIG_MMU */ |
64019a2e | 2296 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2297 | unsigned long start, unsigned long nr_pages, |
2298 | unsigned int gup_flags, struct page **pages, | |
ca5e8632 | 2299 | int *locked) |
eddb1c22 JH |
2300 | { |
2301 | return 0; | |
2302 | } | |
2303 | #endif /* !CONFIG_MMU */ | |
2304 | ||
adc8cb40 SJ |
2305 | /** |
2306 | * get_user_pages() - pin user pages in memory | |
2307 | * @start: starting user address | |
2308 | * @nr_pages: number of pages from start to pin | |
2309 | * @gup_flags: flags modifying lookup behaviour | |
2310 | * @pages: array that receives pointers to the pages pinned. | |
2311 | * Should be at least nr_pages long. Or NULL, if caller | |
2312 | * only intends to ensure the pages are faulted in. | |
adc8cb40 | 2313 | * |
64019a2e PX |
2314 | * This is the same as get_user_pages_remote(), just with a less-flexible |
2315 | * calling convention where we assume that the mm being operated on belongs to | |
2316 | * the current task, and doesn't allow passing of a locked parameter. We also | |
2317 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
2318 | */ |
2319 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
54d02069 | 2320 | unsigned int gup_flags, struct page **pages) |
932f4a63 | 2321 | { |
9a863a6a JG |
2322 | int locked = 1; |
2323 | ||
b2cac248 | 2324 | if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_TOUCH)) |
eddb1c22 JH |
2325 | return -EINVAL; |
2326 | ||
afa3c33e | 2327 | return __get_user_pages_locked(current->mm, start, nr_pages, pages, |
b2cac248 | 2328 | &locked, gup_flags); |
932f4a63 IW |
2329 | } |
2330 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 2331 | |
acc3c8d1 | 2332 | /* |
d3649f68 | 2333 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 2334 | * |
3e4e28c5 | 2335 | * mmap_read_lock(mm); |
64019a2e | 2336 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2337 | * mmap_read_unlock(mm); |
d3649f68 CH |
2338 | * |
2339 | * with: | |
2340 | * | |
64019a2e | 2341 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
2342 | * |
2343 | * It is functionally equivalent to get_user_pages_fast so | |
2344 | * get_user_pages_fast should be used instead if specific gup_flags | |
2345 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 2346 | */ |
d3649f68 CH |
2347 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
2348 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 | 2349 | { |
b2a72dff | 2350 | int locked = 0; |
acc3c8d1 | 2351 | |
b2cac248 | 2352 | if (!is_valid_gup_args(pages, NULL, &gup_flags, |
f04740f5 | 2353 | FOLL_TOUCH | FOLL_UNLOCKABLE)) |
d64e2dbc JG |
2354 | return -EINVAL; |
2355 | ||
afa3c33e | 2356 | return __get_user_pages_locked(current->mm, start, nr_pages, pages, |
b2cac248 | 2357 | &locked, gup_flags); |
4bbd4c77 | 2358 | } |
d3649f68 | 2359 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
2360 | |
2361 | /* | |
67a929e0 | 2362 | * Fast GUP |
2667f50e SC |
2363 | * |
2364 | * get_user_pages_fast attempts to pin user pages by walking the page | |
2365 | * tables directly and avoids taking locks. Thus the walker needs to be | |
2366 | * protected from page table pages being freed from under it, and should | |
2367 | * block any THP splits. | |
2368 | * | |
2369 | * One way to achieve this is to have the walker disable interrupts, and | |
2370 | * rely on IPIs from the TLB flushing code blocking before the page table | |
2371 | * pages are freed. This is unsuitable for architectures that do not need | |
2372 | * to broadcast an IPI when invalidating TLBs. | |
2373 | * | |
2374 | * Another way to achieve this is to batch up page table containing pages | |
2375 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
2376 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
2377 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
2378 | * (which is a relatively rare event). The code below adopts this strategy. | |
2379 | * | |
2380 | * Before activating this code, please be aware that the following assumptions | |
2381 | * are currently made: | |
2382 | * | |
ff2e6d72 | 2383 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 2384 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 2385 | * |
2667f50e SC |
2386 | * *) ptes can be read atomically by the architecture. |
2387 | * | |
2388 | * *) access_ok is sufficient to validate userspace address ranges. | |
2389 | * | |
2390 | * The last two assumptions can be relaxed by the addition of helper functions. | |
2391 | * | |
2392 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
2393 | */ | |
67a929e0 | 2394 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 2395 | |
a6e79df9 LS |
2396 | /* |
2397 | * Used in the GUP-fast path to determine whether a pin is permitted for a | |
2398 | * specific folio. | |
2399 | * | |
2400 | * This call assumes the caller has pinned the folio, that the lowest page table | |
2401 | * level still points to this folio, and that interrupts have been disabled. | |
2402 | * | |
2403 | * Writing to pinned file-backed dirty tracked folios is inherently problematic | |
2404 | * (see comment describing the writable_file_mapping_allowed() function). We | |
2405 | * therefore try to avoid the most egregious case of a long-term mapping doing | |
2406 | * so. | |
2407 | * | |
2408 | * This function cannot be as thorough as that one as the VMA is not available | |
2409 | * in the fast path, so instead we whitelist known good cases and if in doubt, | |
2410 | * fall back to the slow path. | |
2411 | */ | |
2412 | static bool folio_fast_pin_allowed(struct folio *folio, unsigned int flags) | |
2413 | { | |
2414 | struct address_space *mapping; | |
2415 | unsigned long mapping_flags; | |
2416 | ||
2417 | /* | |
2418 | * If we aren't pinning then no problematic write can occur. A long term | |
2419 | * pin is the most egregious case so this is the one we disallow. | |
2420 | */ | |
2421 | if ((flags & (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE)) != | |
2422 | (FOLL_PIN | FOLL_LONGTERM | FOLL_WRITE)) | |
2423 | return true; | |
2424 | ||
2425 | /* The folio is pinned, so we can safely access folio fields. */ | |
2426 | ||
2427 | if (WARN_ON_ONCE(folio_test_slab(folio))) | |
2428 | return false; | |
2429 | ||
2430 | /* hugetlb mappings do not require dirty-tracking. */ | |
2431 | if (folio_test_hugetlb(folio)) | |
2432 | return true; | |
2433 | ||
2434 | /* | |
2435 | * GUP-fast disables IRQs. When IRQS are disabled, RCU grace periods | |
2436 | * cannot proceed, which means no actions performed under RCU can | |
2437 | * proceed either. | |
2438 | * | |
2439 | * inodes and thus their mappings are freed under RCU, which means the | |
2440 | * mapping cannot be freed beneath us and thus we can safely dereference | |
2441 | * it. | |
2442 | */ | |
2443 | lockdep_assert_irqs_disabled(); | |
2444 | ||
2445 | /* | |
2446 | * However, there may be operations which _alter_ the mapping, so ensure | |
2447 | * we read it once and only once. | |
2448 | */ | |
2449 | mapping = READ_ONCE(folio->mapping); | |
2450 | ||
2451 | /* | |
2452 | * The mapping may have been truncated, in any case we cannot determine | |
2453 | * if this mapping is safe - fall back to slow path to determine how to | |
2454 | * proceed. | |
2455 | */ | |
2456 | if (!mapping) | |
2457 | return false; | |
2458 | ||
2459 | /* Anonymous folios pose no problem. */ | |
2460 | mapping_flags = (unsigned long)mapping & PAGE_MAPPING_FLAGS; | |
2461 | if (mapping_flags) | |
2462 | return mapping_flags & PAGE_MAPPING_ANON; | |
2463 | ||
2464 | /* | |
2465 | * At this point, we know the mapping is non-null and points to an | |
2466 | * address_space object. The only remaining whitelisted file system is | |
2467 | * shmem. | |
2468 | */ | |
2469 | return shmem_mapping(mapping); | |
2470 | } | |
2471 | ||
790c7369 | 2472 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 2473 | unsigned int flags, |
790c7369 | 2474 | struct page **pages) |
b59f65fa KS |
2475 | { |
2476 | while ((*nr) - nr_start) { | |
2477 | struct page *page = pages[--(*nr)]; | |
2478 | ||
2479 | ClearPageReferenced(page); | |
3faa52c0 JH |
2480 | if (flags & FOLL_PIN) |
2481 | unpin_user_page(page); | |
2482 | else | |
2483 | put_page(page); | |
b59f65fa KS |
2484 | } |
2485 | } | |
2486 | ||
3010a5ea | 2487 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
70cbc3cc YS |
2488 | /* |
2489 | * Fast-gup relies on pte change detection to avoid concurrent pgtable | |
2490 | * operations. | |
2491 | * | |
2492 | * To pin the page, fast-gup needs to do below in order: | |
2493 | * (1) pin the page (by prefetching pte), then (2) check pte not changed. | |
2494 | * | |
2495 | * For the rest of pgtable operations where pgtable updates can be racy | |
2496 | * with fast-gup, we need to do (1) clear pte, then (2) check whether page | |
2497 | * is pinned. | |
2498 | * | |
2499 | * Above will work for all pte-level operations, including THP split. | |
2500 | * | |
2501 | * For THP collapse, it's a bit more complicated because fast-gup may be | |
2502 | * walking a pgtable page that is being freed (pte is still valid but pmd | |
2503 | * can be cleared already). To avoid race in such condition, we need to | |
2504 | * also check pmd here to make sure pmd doesn't change (corresponds to | |
2505 | * pmdp_collapse_flush() in the THP collapse code path). | |
2506 | */ | |
2507 | static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr, | |
2508 | unsigned long end, unsigned int flags, | |
2509 | struct page **pages, int *nr) | |
2667f50e | 2510 | { |
b59f65fa KS |
2511 | struct dev_pagemap *pgmap = NULL; |
2512 | int nr_start = *nr, ret = 0; | |
2667f50e | 2513 | pte_t *ptep, *ptem; |
2667f50e SC |
2514 | |
2515 | ptem = ptep = pte_offset_map(&pmd, addr); | |
04dee9e8 HD |
2516 | if (!ptep) |
2517 | return 0; | |
2667f50e | 2518 | do { |
2a4a06da | 2519 | pte_t pte = ptep_get_lockless(ptep); |
b0496fe4 MWO |
2520 | struct page *page; |
2521 | struct folio *folio; | |
2667f50e | 2522 | |
0cf45986 | 2523 | if (pte_protnone(pte) && !gup_can_follow_protnone(flags)) |
e7884f8e KS |
2524 | goto pte_unmap; |
2525 | ||
b798bec4 | 2526 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2527 | goto pte_unmap; |
2528 | ||
b59f65fa | 2529 | if (pte_devmap(pte)) { |
7af75561 IW |
2530 | if (unlikely(flags & FOLL_LONGTERM)) |
2531 | goto pte_unmap; | |
2532 | ||
b59f65fa KS |
2533 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2534 | if (unlikely(!pgmap)) { | |
3b78d834 | 2535 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2536 | goto pte_unmap; |
2537 | } | |
2538 | } else if (pte_special(pte)) | |
2667f50e SC |
2539 | goto pte_unmap; |
2540 | ||
2541 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2542 | page = pte_page(pte); | |
2543 | ||
b0496fe4 MWO |
2544 | folio = try_grab_folio(page, 1, flags); |
2545 | if (!folio) | |
2667f50e SC |
2546 | goto pte_unmap; |
2547 | ||
1507f512 | 2548 | if (unlikely(page_is_secretmem(page))) { |
b0496fe4 | 2549 | gup_put_folio(folio, 1, flags); |
1507f512 MR |
2550 | goto pte_unmap; |
2551 | } | |
2552 | ||
70cbc3cc | 2553 | if (unlikely(pmd_val(pmd) != pmd_val(*pmdp)) || |
c33c7948 | 2554 | unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) { |
b0496fe4 | 2555 | gup_put_folio(folio, 1, flags); |
2667f50e SC |
2556 | goto pte_unmap; |
2557 | } | |
2558 | ||
a6e79df9 | 2559 | if (!folio_fast_pin_allowed(folio, flags)) { |
b0496fe4 | 2560 | gup_put_folio(folio, 1, flags); |
2667f50e SC |
2561 | goto pte_unmap; |
2562 | } | |
2563 | ||
84209e87 | 2564 | if (!pte_write(pte) && gup_must_unshare(NULL, flags, page)) { |
a7f22660 DH |
2565 | gup_put_folio(folio, 1, flags); |
2566 | goto pte_unmap; | |
2567 | } | |
2568 | ||
f28d4363 CI |
2569 | /* |
2570 | * We need to make the page accessible if and only if we are | |
2571 | * going to access its content (the FOLL_PIN case). Please | |
2572 | * see Documentation/core-api/pin_user_pages.rst for | |
2573 | * details. | |
2574 | */ | |
2575 | if (flags & FOLL_PIN) { | |
2576 | ret = arch_make_page_accessible(page); | |
2577 | if (ret) { | |
b0496fe4 | 2578 | gup_put_folio(folio, 1, flags); |
f28d4363 CI |
2579 | goto pte_unmap; |
2580 | } | |
2581 | } | |
b0496fe4 | 2582 | folio_set_referenced(folio); |
2667f50e SC |
2583 | pages[*nr] = page; |
2584 | (*nr)++; | |
2667f50e SC |
2585 | } while (ptep++, addr += PAGE_SIZE, addr != end); |
2586 | ||
2587 | ret = 1; | |
2588 | ||
2589 | pte_unmap: | |
832d7aa0 CH |
2590 | if (pgmap) |
2591 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2592 | pte_unmap(ptem); |
2593 | return ret; | |
2594 | } | |
2595 | #else | |
2596 | ||
2597 | /* | |
2598 | * If we can't determine whether or not a pte is special, then fail immediately | |
2599 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2600 | * to be special. | |
2601 | * | |
2602 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2603 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2604 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2605 | */ | |
70cbc3cc YS |
2606 | static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr, |
2607 | unsigned long end, unsigned int flags, | |
2608 | struct page **pages, int *nr) | |
2667f50e SC |
2609 | { |
2610 | return 0; | |
2611 | } | |
3010a5ea | 2612 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2613 | |
17596731 | 2614 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2615 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2616 | unsigned long end, unsigned int flags, |
2617 | struct page **pages, int *nr) | |
b59f65fa KS |
2618 | { |
2619 | int nr_start = *nr; | |
2620 | struct dev_pagemap *pgmap = NULL; | |
2621 | ||
2622 | do { | |
2623 | struct page *page = pfn_to_page(pfn); | |
2624 | ||
2625 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2626 | if (unlikely(!pgmap)) { | |
3b78d834 | 2627 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2628 | break; |
b59f65fa | 2629 | } |
4003f107 LG |
2630 | |
2631 | if (!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)) { | |
2632 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
2633 | break; | |
2634 | } | |
2635 | ||
b59f65fa KS |
2636 | SetPageReferenced(page); |
2637 | pages[*nr] = page; | |
0f089235 | 2638 | if (unlikely(try_grab_page(page, flags))) { |
3faa52c0 | 2639 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2640 | break; |
3faa52c0 | 2641 | } |
b59f65fa KS |
2642 | (*nr)++; |
2643 | pfn++; | |
2644 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 | 2645 | |
6401c4eb | 2646 | put_dev_pagemap(pgmap); |
20b7fee7 | 2647 | return addr == end; |
b59f65fa KS |
2648 | } |
2649 | ||
a9b6de77 | 2650 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2651 | unsigned long end, unsigned int flags, |
2652 | struct page **pages, int *nr) | |
b59f65fa KS |
2653 | { |
2654 | unsigned long fault_pfn; | |
a9b6de77 DW |
2655 | int nr_start = *nr; |
2656 | ||
2657 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2658 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2659 | return 0; |
b59f65fa | 2660 | |
a9b6de77 | 2661 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2662 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2663 | return 0; |
2664 | } | |
2665 | return 1; | |
b59f65fa KS |
2666 | } |
2667 | ||
a9b6de77 | 2668 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2669 | unsigned long end, unsigned int flags, |
2670 | struct page **pages, int *nr) | |
b59f65fa KS |
2671 | { |
2672 | unsigned long fault_pfn; | |
a9b6de77 DW |
2673 | int nr_start = *nr; |
2674 | ||
2675 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2676 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2677 | return 0; |
b59f65fa | 2678 | |
a9b6de77 | 2679 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2680 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2681 | return 0; |
2682 | } | |
2683 | return 1; | |
b59f65fa KS |
2684 | } |
2685 | #else | |
a9b6de77 | 2686 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2687 | unsigned long end, unsigned int flags, |
2688 | struct page **pages, int *nr) | |
b59f65fa KS |
2689 | { |
2690 | BUILD_BUG(); | |
2691 | return 0; | |
2692 | } | |
2693 | ||
a9b6de77 | 2694 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2695 | unsigned long end, unsigned int flags, |
2696 | struct page **pages, int *nr) | |
b59f65fa KS |
2697 | { |
2698 | BUILD_BUG(); | |
2699 | return 0; | |
2700 | } | |
2701 | #endif | |
2702 | ||
a43e9820 JH |
2703 | static int record_subpages(struct page *page, unsigned long addr, |
2704 | unsigned long end, struct page **pages) | |
2705 | { | |
2706 | int nr; | |
2707 | ||
c228afb1 MWO |
2708 | for (nr = 0; addr != end; nr++, addr += PAGE_SIZE) |
2709 | pages[nr] = nth_page(page, nr); | |
a43e9820 JH |
2710 | |
2711 | return nr; | |
2712 | } | |
2713 | ||
cbd34da7 CH |
2714 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2715 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2716 | unsigned long sz) | |
2717 | { | |
2718 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2719 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2720 | } | |
2721 | ||
2722 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2723 | unsigned long end, unsigned int flags, |
2724 | struct page **pages, int *nr) | |
cbd34da7 CH |
2725 | { |
2726 | unsigned long pte_end; | |
09a1626e MWO |
2727 | struct page *page; |
2728 | struct folio *folio; | |
cbd34da7 CH |
2729 | pte_t pte; |
2730 | int refs; | |
2731 | ||
2732 | pte_end = (addr + sz) & ~(sz-1); | |
2733 | if (pte_end < end) | |
2734 | end = pte_end; | |
2735 | ||
55ca2263 | 2736 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2737 | |
0cd22afd | 2738 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2739 | return 0; |
2740 | ||
2741 | /* hugepages are never "special" */ | |
2742 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2743 | ||
09a1626e | 2744 | page = nth_page(pte_page(pte), (addr & (sz - 1)) >> PAGE_SHIFT); |
a43e9820 | 2745 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2746 | |
09a1626e MWO |
2747 | folio = try_grab_folio(page, refs, flags); |
2748 | if (!folio) | |
cbd34da7 | 2749 | return 0; |
cbd34da7 | 2750 | |
c33c7948 | 2751 | if (unlikely(pte_val(pte) != pte_val(ptep_get(ptep)))) { |
09a1626e | 2752 | gup_put_folio(folio, refs, flags); |
cbd34da7 CH |
2753 | return 0; |
2754 | } | |
2755 | ||
a6e79df9 | 2756 | if (!folio_fast_pin_allowed(folio, flags)) { |
09a1626e | 2757 | gup_put_folio(folio, refs, flags); |
cbd34da7 CH |
2758 | return 0; |
2759 | } | |
2760 | ||
84209e87 | 2761 | if (!pte_write(pte) && gup_must_unshare(NULL, flags, &folio->page)) { |
a7f22660 DH |
2762 | gup_put_folio(folio, refs, flags); |
2763 | return 0; | |
2764 | } | |
2765 | ||
a43e9820 | 2766 | *nr += refs; |
09a1626e | 2767 | folio_set_referenced(folio); |
cbd34da7 CH |
2768 | return 1; |
2769 | } | |
2770 | ||
2771 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2772 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2773 | struct page **pages, int *nr) |
2774 | { | |
2775 | pte_t *ptep; | |
2776 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2777 | unsigned long next; | |
2778 | ||
2779 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2780 | do { | |
2781 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2782 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2783 | return 0; |
2784 | } while (ptep++, addr = next, addr != end); | |
2785 | ||
2786 | return 1; | |
2787 | } | |
2788 | #else | |
2789 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2790 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2791 | struct page **pages, int *nr) |
2792 | { | |
2793 | return 0; | |
2794 | } | |
2795 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2796 | ||
2667f50e | 2797 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2798 | unsigned long end, unsigned int flags, |
2799 | struct page **pages, int *nr) | |
2667f50e | 2800 | { |
667ed1f7 MWO |
2801 | struct page *page; |
2802 | struct folio *folio; | |
2667f50e SC |
2803 | int refs; |
2804 | ||
b798bec4 | 2805 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2806 | return 0; |
2807 | ||
7af75561 IW |
2808 | if (pmd_devmap(orig)) { |
2809 | if (unlikely(flags & FOLL_LONGTERM)) | |
2810 | return 0; | |
86dfbed4 JH |
2811 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2812 | pages, nr); | |
7af75561 | 2813 | } |
b59f65fa | 2814 | |
c228afb1 | 2815 | page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2816 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2817 | |
667ed1f7 MWO |
2818 | folio = try_grab_folio(page, refs, flags); |
2819 | if (!folio) | |
2667f50e | 2820 | return 0; |
2667f50e SC |
2821 | |
2822 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
667ed1f7 | 2823 | gup_put_folio(folio, refs, flags); |
2667f50e SC |
2824 | return 0; |
2825 | } | |
2826 | ||
a6e79df9 LS |
2827 | if (!folio_fast_pin_allowed(folio, flags)) { |
2828 | gup_put_folio(folio, refs, flags); | |
2829 | return 0; | |
2830 | } | |
84209e87 | 2831 | if (!pmd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) { |
a7f22660 DH |
2832 | gup_put_folio(folio, refs, flags); |
2833 | return 0; | |
2834 | } | |
2835 | ||
a43e9820 | 2836 | *nr += refs; |
667ed1f7 | 2837 | folio_set_referenced(folio); |
2667f50e SC |
2838 | return 1; |
2839 | } | |
2840 | ||
2841 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2842 | unsigned long end, unsigned int flags, |
2843 | struct page **pages, int *nr) | |
2667f50e | 2844 | { |
83afb52e MWO |
2845 | struct page *page; |
2846 | struct folio *folio; | |
2667f50e SC |
2847 | int refs; |
2848 | ||
b798bec4 | 2849 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2850 | return 0; |
2851 | ||
7af75561 IW |
2852 | if (pud_devmap(orig)) { |
2853 | if (unlikely(flags & FOLL_LONGTERM)) | |
2854 | return 0; | |
86dfbed4 JH |
2855 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2856 | pages, nr); | |
7af75561 | 2857 | } |
b59f65fa | 2858 | |
c228afb1 | 2859 | page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2860 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2861 | |
83afb52e MWO |
2862 | folio = try_grab_folio(page, refs, flags); |
2863 | if (!folio) | |
2667f50e | 2864 | return 0; |
2667f50e SC |
2865 | |
2866 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
83afb52e | 2867 | gup_put_folio(folio, refs, flags); |
2667f50e SC |
2868 | return 0; |
2869 | } | |
2870 | ||
a6e79df9 LS |
2871 | if (!folio_fast_pin_allowed(folio, flags)) { |
2872 | gup_put_folio(folio, refs, flags); | |
2873 | return 0; | |
2874 | } | |
2875 | ||
84209e87 | 2876 | if (!pud_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) { |
a7f22660 DH |
2877 | gup_put_folio(folio, refs, flags); |
2878 | return 0; | |
2879 | } | |
2880 | ||
a43e9820 | 2881 | *nr += refs; |
83afb52e | 2882 | folio_set_referenced(folio); |
2667f50e SC |
2883 | return 1; |
2884 | } | |
2885 | ||
f30c59e9 | 2886 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2887 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2888 | struct page **pages, int *nr) |
2889 | { | |
2890 | int refs; | |
2d7919a2 MWO |
2891 | struct page *page; |
2892 | struct folio *folio; | |
f30c59e9 | 2893 | |
b798bec4 | 2894 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2895 | return 0; |
2896 | ||
b59f65fa | 2897 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2898 | |
c228afb1 | 2899 | page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2900 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2901 | |
2d7919a2 MWO |
2902 | folio = try_grab_folio(page, refs, flags); |
2903 | if (!folio) | |
f30c59e9 | 2904 | return 0; |
f30c59e9 AK |
2905 | |
2906 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
2d7919a2 | 2907 | gup_put_folio(folio, refs, flags); |
f30c59e9 AK |
2908 | return 0; |
2909 | } | |
2910 | ||
31115034 LS |
2911 | if (!pgd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) { |
2912 | gup_put_folio(folio, refs, flags); | |
2913 | return 0; | |
2914 | } | |
2915 | ||
a6e79df9 LS |
2916 | if (!folio_fast_pin_allowed(folio, flags)) { |
2917 | gup_put_folio(folio, refs, flags); | |
2918 | return 0; | |
2919 | } | |
2920 | ||
a43e9820 | 2921 | *nr += refs; |
2d7919a2 | 2922 | folio_set_referenced(folio); |
f30c59e9 AK |
2923 | return 1; |
2924 | } | |
2925 | ||
d3f7b1bb | 2926 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2927 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2928 | { |
2929 | unsigned long next; | |
2930 | pmd_t *pmdp; | |
2931 | ||
d3f7b1bb | 2932 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2933 | do { |
1180e732 | 2934 | pmd_t pmd = pmdp_get_lockless(pmdp); |
2667f50e SC |
2935 | |
2936 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2937 | if (!pmd_present(pmd)) |
2667f50e SC |
2938 | return 0; |
2939 | ||
414fd080 YZ |
2940 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2941 | pmd_devmap(pmd))) { | |
0cf45986 DH |
2942 | if (pmd_protnone(pmd) && |
2943 | !gup_can_follow_protnone(flags)) | |
2667f50e SC |
2944 | return 0; |
2945 | ||
b798bec4 | 2946 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2947 | pages, nr)) |
2948 | return 0; | |
2949 | ||
f30c59e9 AK |
2950 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2951 | /* | |
2952 | * architecture have different format for hugetlbfs | |
2953 | * pmd format and THP pmd format | |
2954 | */ | |
2955 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2956 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2957 | return 0; |
70cbc3cc | 2958 | } else if (!gup_pte_range(pmd, pmdp, addr, next, flags, pages, nr)) |
2923117b | 2959 | return 0; |
2667f50e SC |
2960 | } while (pmdp++, addr = next, addr != end); |
2961 | ||
2962 | return 1; | |
2963 | } | |
2964 | ||
d3f7b1bb | 2965 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2966 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2967 | { |
2968 | unsigned long next; | |
2969 | pud_t *pudp; | |
2970 | ||
d3f7b1bb | 2971 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2972 | do { |
e37c6982 | 2973 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2974 | |
2975 | next = pud_addr_end(addr, end); | |
15494520 | 2976 | if (unlikely(!pud_present(pud))) |
2667f50e | 2977 | return 0; |
fcd0ccd8 | 2978 | if (unlikely(pud_huge(pud) || pud_devmap(pud))) { |
b798bec4 | 2979 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2980 | pages, nr)) |
2981 | return 0; | |
2982 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2983 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2984 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2985 | return 0; |
d3f7b1bb | 2986 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2987 | return 0; |
2988 | } while (pudp++, addr = next, addr != end); | |
2989 | ||
2990 | return 1; | |
2991 | } | |
2992 | ||
d3f7b1bb | 2993 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2994 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2995 | { |
2996 | unsigned long next; | |
2997 | p4d_t *p4dp; | |
2998 | ||
d3f7b1bb | 2999 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
3000 | do { |
3001 | p4d_t p4d = READ_ONCE(*p4dp); | |
3002 | ||
3003 | next = p4d_addr_end(addr, end); | |
3004 | if (p4d_none(p4d)) | |
3005 | return 0; | |
3006 | BUILD_BUG_ON(p4d_huge(p4d)); | |
3007 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
3008 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 3009 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 3010 | return 0; |
d3f7b1bb | 3011 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
3012 | return 0; |
3013 | } while (p4dp++, addr = next, addr != end); | |
3014 | ||
3015 | return 1; | |
3016 | } | |
3017 | ||
5b65c467 | 3018 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 3019 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
3020 | { |
3021 | unsigned long next; | |
3022 | pgd_t *pgdp; | |
3023 | ||
3024 | pgdp = pgd_offset(current->mm, addr); | |
3025 | do { | |
3026 | pgd_t pgd = READ_ONCE(*pgdp); | |
3027 | ||
3028 | next = pgd_addr_end(addr, end); | |
3029 | if (pgd_none(pgd)) | |
3030 | return; | |
3031 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 3032 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
3033 | pages, nr)) |
3034 | return; | |
3035 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
3036 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 3037 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 3038 | return; |
d3f7b1bb | 3039 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
3040 | return; |
3041 | } while (pgdp++, addr = next, addr != end); | |
3042 | } | |
050a9adc CH |
3043 | #else |
3044 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
3045 | unsigned int flags, struct page **pages, int *nr) | |
3046 | { | |
3047 | } | |
3048 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
3049 | |
3050 | #ifndef gup_fast_permitted | |
3051 | /* | |
dadbb612 | 3052 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
3053 | * we need to fall back to the slow version: |
3054 | */ | |
26f4c328 | 3055 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 3056 | { |
26f4c328 | 3057 | return true; |
5b65c467 KS |
3058 | } |
3059 | #endif | |
3060 | ||
c28b1fc7 JG |
3061 | static unsigned long lockless_pages_from_mm(unsigned long start, |
3062 | unsigned long end, | |
3063 | unsigned int gup_flags, | |
3064 | struct page **pages) | |
3065 | { | |
3066 | unsigned long flags; | |
3067 | int nr_pinned = 0; | |
57efa1fe | 3068 | unsigned seq; |
c28b1fc7 JG |
3069 | |
3070 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
3071 | !gup_fast_permitted(start, end)) | |
3072 | return 0; | |
3073 | ||
57efa1fe JG |
3074 | if (gup_flags & FOLL_PIN) { |
3075 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
3076 | if (seq & 1) | |
3077 | return 0; | |
3078 | } | |
3079 | ||
c28b1fc7 JG |
3080 | /* |
3081 | * Disable interrupts. The nested form is used, in order to allow full, | |
3082 | * general purpose use of this routine. | |
3083 | * | |
3084 | * With interrupts disabled, we block page table pages from being freed | |
3085 | * from under us. See struct mmu_table_batch comments in | |
3086 | * include/asm-generic/tlb.h for more details. | |
3087 | * | |
3088 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
3089 | * that come from THPs splitting. | |
3090 | */ | |
3091 | local_irq_save(flags); | |
3092 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
3093 | local_irq_restore(flags); | |
57efa1fe JG |
3094 | |
3095 | /* | |
3096 | * When pinning pages for DMA there could be a concurrent write protect | |
3097 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
3098 | */ | |
3099 | if (gup_flags & FOLL_PIN) { | |
3100 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
b6a2619c | 3101 | unpin_user_pages_lockless(pages, nr_pinned); |
57efa1fe | 3102 | return 0; |
b6a2619c DH |
3103 | } else { |
3104 | sanity_check_pinned_pages(pages, nr_pinned); | |
57efa1fe JG |
3105 | } |
3106 | } | |
c28b1fc7 JG |
3107 | return nr_pinned; |
3108 | } | |
3109 | ||
3110 | static int internal_get_user_pages_fast(unsigned long start, | |
3111 | unsigned long nr_pages, | |
eddb1c22 JH |
3112 | unsigned int gup_flags, |
3113 | struct page **pages) | |
2667f50e | 3114 | { |
c28b1fc7 JG |
3115 | unsigned long len, end; |
3116 | unsigned long nr_pinned; | |
b2a72dff | 3117 | int locked = 0; |
c28b1fc7 | 3118 | int ret; |
2667f50e | 3119 | |
f4000fdf | 3120 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef | 3121 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
4003f107 LG |
3122 | FOLL_FAST_ONLY | FOLL_NOFAULT | |
3123 | FOLL_PCI_P2PDMA))) | |
817be129 CH |
3124 | return -EINVAL; |
3125 | ||
a458b76a AA |
3126 | if (gup_flags & FOLL_PIN) |
3127 | mm_set_has_pinned_flag(¤t->mm->flags); | |
008cfe44 | 3128 | |
f81cd178 | 3129 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 3130 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 3131 | |
f455c854 | 3132 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
3133 | len = nr_pages << PAGE_SHIFT; |
3134 | if (check_add_overflow(start, len, &end)) | |
9883c7f8 | 3135 | return -EOVERFLOW; |
6014bc27 LT |
3136 | if (end > TASK_SIZE_MAX) |
3137 | return -EFAULT; | |
96d4f267 | 3138 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 3139 | return -EFAULT; |
73e10a61 | 3140 | |
c28b1fc7 JG |
3141 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
3142 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
3143 | return nr_pinned; | |
2667f50e | 3144 | |
c28b1fc7 JG |
3145 | /* Slow path: try to get the remaining pages with get_user_pages */ |
3146 | start += nr_pinned << PAGE_SHIFT; | |
3147 | pages += nr_pinned; | |
b2a72dff | 3148 | ret = __gup_longterm_locked(current->mm, start, nr_pages - nr_pinned, |
b2cac248 | 3149 | pages, &locked, |
f04740f5 | 3150 | gup_flags | FOLL_TOUCH | FOLL_UNLOCKABLE); |
c28b1fc7 JG |
3151 | if (ret < 0) { |
3152 | /* | |
3153 | * The caller has to unpin the pages we already pinned so | |
3154 | * returning -errno is not an option | |
3155 | */ | |
3156 | if (nr_pinned) | |
3157 | return nr_pinned; | |
3158 | return ret; | |
2667f50e | 3159 | } |
c28b1fc7 | 3160 | return ret + nr_pinned; |
2667f50e | 3161 | } |
c28b1fc7 | 3162 | |
dadbb612 SJ |
3163 | /** |
3164 | * get_user_pages_fast_only() - pin user pages in memory | |
3165 | * @start: starting user address | |
3166 | * @nr_pages: number of pages from start to pin | |
3167 | * @gup_flags: flags modifying pin behaviour | |
3168 | * @pages: array that receives pointers to the pages pinned. | |
3169 | * Should be at least nr_pages long. | |
3170 | * | |
9e1f0580 JH |
3171 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
3172 | * the regular GUP. | |
9e1f0580 JH |
3173 | * |
3174 | * If the architecture does not support this function, simply return with no | |
3175 | * pages pinned. | |
3176 | * | |
3177 | * Careful, careful! COW breaking can go either way, so a non-write | |
3178 | * access can get ambiguous page results. If you call this function without | |
3179 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
3180 | */ | |
dadbb612 SJ |
3181 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
3182 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 3183 | { |
9e1f0580 JH |
3184 | /* |
3185 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
3186 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
3187 | * |
3188 | * FOLL_FAST_ONLY is required in order to match the API description of | |
3189 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 3190 | */ |
b2cac248 | 3191 | if (!is_valid_gup_args(pages, NULL, &gup_flags, |
d64e2dbc JG |
3192 | FOLL_GET | FOLL_FAST_ONLY)) |
3193 | return -EINVAL; | |
9e1f0580 | 3194 | |
9198a919 | 3195 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
9e1f0580 | 3196 | } |
dadbb612 | 3197 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 3198 | |
eddb1c22 JH |
3199 | /** |
3200 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
3201 | * @start: starting user address |
3202 | * @nr_pages: number of pages from start to pin | |
3203 | * @gup_flags: flags modifying pin behaviour | |
3204 | * @pages: array that receives pointers to the pages pinned. | |
3205 | * Should be at least nr_pages long. | |
eddb1c22 | 3206 | * |
c1e8d7c6 | 3207 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
3208 | * If not successful, it will fall back to taking the lock and |
3209 | * calling get_user_pages(). | |
3210 | * | |
3211 | * Returns number of pages pinned. This may be fewer than the number requested. | |
3212 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
3213 | * -errno. | |
3214 | */ | |
3215 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
3216 | unsigned int gup_flags, struct page **pages) | |
3217 | { | |
94202f12 JH |
3218 | /* |
3219 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
3220 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
3221 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
3222 | * request. | |
3223 | */ | |
b2cac248 | 3224 | if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_GET)) |
d64e2dbc | 3225 | return -EINVAL; |
eddb1c22 JH |
3226 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
3227 | } | |
050a9adc | 3228 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
3229 | |
3230 | /** | |
3231 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
3232 | * | |
3faa52c0 JH |
3233 | * @start: starting user address |
3234 | * @nr_pages: number of pages from start to pin | |
3235 | * @gup_flags: flags modifying pin behaviour | |
3236 | * @pages: array that receives pointers to the pages pinned. | |
3237 | * Should be at least nr_pages long. | |
3238 | * | |
3239 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
3240 | * get_user_pages_fast() for documentation on the function arguments, because | |
3241 | * the arguments here are identical. | |
3242 | * | |
3243 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3244 | * see Documentation/core-api/pin_user_pages.rst for further details. |
c8070b78 DH |
3245 | * |
3246 | * Note that if a zero_page is amongst the returned pages, it will not have | |
3247 | * pins in it and unpin_user_page() will not remove pins from it. | |
eddb1c22 JH |
3248 | */ |
3249 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
3250 | unsigned int gup_flags, struct page **pages) | |
3251 | { | |
b2cac248 | 3252 | if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN)) |
3faa52c0 | 3253 | return -EINVAL; |
3faa52c0 | 3254 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
eddb1c22 JH |
3255 | } |
3256 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
3257 | ||
3258 | /** | |
64019a2e | 3259 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 3260 | * |
3faa52c0 JH |
3261 | * @mm: mm_struct of target mm |
3262 | * @start: starting user address | |
3263 | * @nr_pages: number of pages from start to pin | |
3264 | * @gup_flags: flags modifying lookup behaviour | |
3265 | * @pages: array that receives pointers to the pages pinned. | |
0768c8de | 3266 | * Should be at least nr_pages long. |
3faa52c0 JH |
3267 | * @locked: pointer to lock flag indicating whether lock is held and |
3268 | * subsequently whether VM_FAULT_RETRY functionality can be | |
3269 | * utilised. Lock must initially be held. | |
3270 | * | |
3271 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
3272 | * get_user_pages_remote() for documentation on the function arguments, because | |
3273 | * the arguments here are identical. | |
3274 | * | |
3275 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3276 | * see Documentation/core-api/pin_user_pages.rst for details. |
c8070b78 DH |
3277 | * |
3278 | * Note that if a zero_page is amongst the returned pages, it will not have | |
3279 | * pins in it and unpin_user_page*() will not remove pins from it. | |
eddb1c22 | 3280 | */ |
64019a2e | 3281 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
3282 | unsigned long start, unsigned long nr_pages, |
3283 | unsigned int gup_flags, struct page **pages, | |
0b295316 | 3284 | int *locked) |
eddb1c22 | 3285 | { |
9a863a6a JG |
3286 | int local_locked = 1; |
3287 | ||
b2cac248 | 3288 | if (!is_valid_gup_args(pages, locked, &gup_flags, |
d64e2dbc JG |
3289 | FOLL_PIN | FOLL_TOUCH | FOLL_REMOTE)) |
3290 | return 0; | |
b2cac248 | 3291 | return __gup_longterm_locked(mm, start, nr_pages, pages, |
9a863a6a | 3292 | locked ? locked : &local_locked, |
d64e2dbc | 3293 | gup_flags); |
eddb1c22 JH |
3294 | } |
3295 | EXPORT_SYMBOL(pin_user_pages_remote); | |
3296 | ||
3297 | /** | |
3298 | * pin_user_pages() - pin user pages in memory for use by other devices | |
3299 | * | |
3faa52c0 JH |
3300 | * @start: starting user address |
3301 | * @nr_pages: number of pages from start to pin | |
3302 | * @gup_flags: flags modifying lookup behaviour | |
3303 | * @pages: array that receives pointers to the pages pinned. | |
0768c8de | 3304 | * Should be at least nr_pages long. |
3faa52c0 JH |
3305 | * |
3306 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
3307 | * FOLL_PIN is set. | |
3308 | * | |
3309 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3310 | * see Documentation/core-api/pin_user_pages.rst for details. |
c8070b78 DH |
3311 | * |
3312 | * Note that if a zero_page is amongst the returned pages, it will not have | |
3313 | * pins in it and unpin_user_page*() will not remove pins from it. | |
eddb1c22 JH |
3314 | */ |
3315 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
4c630f30 | 3316 | unsigned int gup_flags, struct page **pages) |
eddb1c22 | 3317 | { |
9a863a6a JG |
3318 | int locked = 1; |
3319 | ||
b2cac248 | 3320 | if (!is_valid_gup_args(pages, NULL, &gup_flags, FOLL_PIN)) |
d64e2dbc | 3321 | return 0; |
64019a2e | 3322 | return __gup_longterm_locked(current->mm, start, nr_pages, |
b2cac248 | 3323 | pages, &locked, gup_flags); |
eddb1c22 JH |
3324 | } |
3325 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
3326 | |
3327 | /* | |
3328 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
3329 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
3330 | * FOLL_PIN and rejects FOLL_GET. | |
c8070b78 DH |
3331 | * |
3332 | * Note that if a zero_page is amongst the returned pages, it will not have | |
3333 | * pins in it and unpin_user_page*() will not remove pins from it. | |
91429023 JH |
3334 | */ |
3335 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
3336 | struct page **pages, unsigned int gup_flags) | |
3337 | { | |
b2a72dff | 3338 | int locked = 0; |
91429023 | 3339 | |
b2cac248 | 3340 | if (!is_valid_gup_args(pages, NULL, &gup_flags, |
f04740f5 | 3341 | FOLL_PIN | FOLL_TOUCH | FOLL_UNLOCKABLE)) |
d64e2dbc | 3342 | return 0; |
0768c8de | 3343 | |
b2cac248 | 3344 | return __gup_longterm_locked(current->mm, start, nr_pages, pages, |
b2a72dff | 3345 | &locked, gup_flags); |
91429023 JH |
3346 | } |
3347 | EXPORT_SYMBOL(pin_user_pages_unlocked); |