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> | |
1027e443 | 21 | |
33a709b2 | 22 | #include <asm/mmu_context.h> |
1027e443 | 23 | #include <asm/tlbflush.h> |
2667f50e | 24 | |
4bbd4c77 KS |
25 | #include "internal.h" |
26 | ||
df06b37f KB |
27 | struct follow_page_context { |
28 | struct dev_pagemap *pgmap; | |
29 | unsigned int page_mask; | |
30 | }; | |
31 | ||
b6a2619c DH |
32 | static inline void sanity_check_pinned_pages(struct page **pages, |
33 | unsigned long npages) | |
34 | { | |
35 | if (!IS_ENABLED(CONFIG_DEBUG_VM)) | |
36 | return; | |
37 | ||
38 | /* | |
39 | * We only pin anonymous pages if they are exclusive. Once pinned, we | |
40 | * can no longer turn them possibly shared and PageAnonExclusive() will | |
41 | * stick around until the page is freed. | |
42 | * | |
43 | * We'd like to verify that our pinned anonymous pages are still mapped | |
44 | * exclusively. The issue with anon THP is that we don't know how | |
45 | * they are/were mapped when pinning them. However, for anon | |
46 | * THP we can assume that either the given page (PTE-mapped THP) or | |
47 | * the head page (PMD-mapped THP) should be PageAnonExclusive(). If | |
48 | * neither is the case, there is certainly something wrong. | |
49 | */ | |
50 | for (; npages; npages--, pages++) { | |
51 | struct page *page = *pages; | |
52 | struct folio *folio = page_folio(page); | |
53 | ||
54 | if (!folio_test_anon(folio)) | |
55 | continue; | |
56 | if (!folio_test_large(folio) || folio_test_hugetlb(folio)) | |
57 | VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page), page); | |
58 | else | |
59 | /* Either a PTE-mapped or a PMD-mapped THP. */ | |
60 | VM_BUG_ON_PAGE(!PageAnonExclusive(&folio->page) && | |
61 | !PageAnonExclusive(page), page); | |
62 | } | |
63 | } | |
64 | ||
cd1adf1b | 65 | /* |
ece1ed7b | 66 | * Return the folio with ref appropriately incremented, |
cd1adf1b | 67 | * or NULL if that failed. |
a707cdd5 | 68 | */ |
ece1ed7b | 69 | static inline struct folio *try_get_folio(struct page *page, int refs) |
a707cdd5 | 70 | { |
ece1ed7b | 71 | struct folio *folio; |
a707cdd5 | 72 | |
59409373 | 73 | retry: |
ece1ed7b MWO |
74 | folio = page_folio(page); |
75 | if (WARN_ON_ONCE(folio_ref_count(folio) < 0)) | |
a707cdd5 | 76 | return NULL; |
ece1ed7b | 77 | if (unlikely(!folio_ref_try_add_rcu(folio, refs))) |
a707cdd5 | 78 | return NULL; |
c24d3732 JH |
79 | |
80 | /* | |
ece1ed7b MWO |
81 | * At this point we have a stable reference to the folio; but it |
82 | * could be that between calling page_folio() and the refcount | |
83 | * increment, the folio was split, in which case we'd end up | |
84 | * holding a reference on a folio that has nothing to do with the page | |
c24d3732 | 85 | * we were given anymore. |
ece1ed7b MWO |
86 | * So now that the folio is stable, recheck that the page still |
87 | * belongs to this folio. | |
c24d3732 | 88 | */ |
ece1ed7b MWO |
89 | if (unlikely(page_folio(page) != folio)) { |
90 | folio_put_refs(folio, refs); | |
59409373 | 91 | goto retry; |
c24d3732 JH |
92 | } |
93 | ||
ece1ed7b | 94 | return folio; |
a707cdd5 JH |
95 | } |
96 | ||
3967db22 | 97 | /** |
ece1ed7b | 98 | * try_grab_folio() - Attempt to get or pin a folio. |
3967db22 | 99 | * @page: pointer to page to be grabbed |
ece1ed7b | 100 | * @refs: the value to (effectively) add to the folio's refcount |
3967db22 JH |
101 | * @flags: gup flags: these are the FOLL_* flag values. |
102 | * | |
3faa52c0 | 103 | * "grab" names in this file mean, "look at flags to decide whether to use |
ece1ed7b | 104 | * FOLL_PIN or FOLL_GET behavior, when incrementing the folio's refcount. |
3faa52c0 JH |
105 | * |
106 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
107 | * same time. (That's true throughout the get_user_pages*() and | |
108 | * pin_user_pages*() APIs.) Cases: | |
109 | * | |
ece1ed7b | 110 | * FOLL_GET: folio's refcount will be incremented by @refs. |
3967db22 | 111 | * |
ece1ed7b MWO |
112 | * FOLL_PIN on large folios: folio's refcount will be incremented by |
113 | * @refs, and its compound_pincount will be incremented by @refs. | |
3967db22 | 114 | * |
ece1ed7b | 115 | * FOLL_PIN on single-page folios: folio's refcount will be incremented by |
5232c63f | 116 | * @refs * GUP_PIN_COUNTING_BIAS. |
3faa52c0 | 117 | * |
ece1ed7b MWO |
118 | * Return: The folio containing @page (with refcount appropriately |
119 | * incremented) for success, or NULL upon failure. If neither FOLL_GET | |
120 | * nor FOLL_PIN was set, that's considered failure, and furthermore, | |
121 | * a likely bug in the caller, so a warning is also emitted. | |
3faa52c0 | 122 | */ |
ece1ed7b | 123 | struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) |
3faa52c0 JH |
124 | { |
125 | if (flags & FOLL_GET) | |
ece1ed7b | 126 | return try_get_folio(page, refs); |
3faa52c0 | 127 | else if (flags & FOLL_PIN) { |
ece1ed7b MWO |
128 | struct folio *folio; |
129 | ||
df3a0a21 | 130 | /* |
d1e153fe PT |
131 | * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a |
132 | * right zone, so fail and let the caller fall back to the slow | |
133 | * path. | |
df3a0a21 | 134 | */ |
d1e153fe | 135 | if (unlikely((flags & FOLL_LONGTERM) && |
6077c943 | 136 | !is_longterm_pinnable_page(page))) |
df3a0a21 PL |
137 | return NULL; |
138 | ||
c24d3732 JH |
139 | /* |
140 | * CAUTION: Don't use compound_head() on the page before this | |
141 | * point, the result won't be stable. | |
142 | */ | |
ece1ed7b MWO |
143 | folio = try_get_folio(page, refs); |
144 | if (!folio) | |
c24d3732 JH |
145 | return NULL; |
146 | ||
47e29d32 | 147 | /* |
ece1ed7b | 148 | * When pinning a large folio, use an exact count to track it. |
47e29d32 | 149 | * |
ece1ed7b MWO |
150 | * However, be sure to *also* increment the normal folio |
151 | * refcount field at least once, so that the folio really | |
78d9d6ce | 152 | * is pinned. That's why the refcount from the earlier |
ece1ed7b | 153 | * try_get_folio() is left intact. |
47e29d32 | 154 | */ |
ece1ed7b MWO |
155 | if (folio_test_large(folio)) |
156 | atomic_add(refs, folio_pincount_ptr(folio)); | |
c24d3732 | 157 | else |
ece1ed7b MWO |
158 | folio_ref_add(folio, |
159 | refs * (GUP_PIN_COUNTING_BIAS - 1)); | |
160 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, refs); | |
47e29d32 | 161 | |
ece1ed7b | 162 | return folio; |
3faa52c0 JH |
163 | } |
164 | ||
165 | WARN_ON_ONCE(1); | |
166 | return NULL; | |
167 | } | |
168 | ||
d8ddc099 | 169 | static void gup_put_folio(struct folio *folio, int refs, unsigned int flags) |
4509b42c JG |
170 | { |
171 | if (flags & FOLL_PIN) { | |
d8ddc099 MWO |
172 | node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs); |
173 | if (folio_test_large(folio)) | |
174 | atomic_sub(refs, folio_pincount_ptr(folio)); | |
4509b42c JG |
175 | else |
176 | refs *= GUP_PIN_COUNTING_BIAS; | |
177 | } | |
178 | ||
d8ddc099 | 179 | folio_put_refs(folio, refs); |
4509b42c JG |
180 | } |
181 | ||
3faa52c0 JH |
182 | /** |
183 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
5fec0719 MWO |
184 | * @page: pointer to page to be grabbed |
185 | * @flags: gup flags: these are the FOLL_* flag values. | |
3faa52c0 JH |
186 | * |
187 | * This might not do anything at all, depending on the flags argument. | |
188 | * | |
189 | * "grab" names in this file mean, "look at flags to decide whether to use | |
190 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
191 | * | |
3faa52c0 | 192 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same |
ece1ed7b | 193 | * time. Cases: please see the try_grab_folio() documentation, with |
3967db22 | 194 | * "refs=1". |
3faa52c0 JH |
195 | * |
196 | * Return: true for success, or if no action was required (if neither FOLL_PIN | |
197 | * nor FOLL_GET was set, nothing is done). False for failure: FOLL_GET or | |
198 | * FOLL_PIN was set, but the page could not be grabbed. | |
199 | */ | |
200 | bool __must_check try_grab_page(struct page *page, unsigned int flags) | |
201 | { | |
5fec0719 MWO |
202 | struct folio *folio = page_folio(page); |
203 | ||
c36c04c2 | 204 | WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN)); |
5fec0719 MWO |
205 | if (WARN_ON_ONCE(folio_ref_count(folio) <= 0)) |
206 | return false; | |
3faa52c0 | 207 | |
c36c04c2 | 208 | if (flags & FOLL_GET) |
5fec0719 | 209 | folio_ref_inc(folio); |
c36c04c2 | 210 | else if (flags & FOLL_PIN) { |
c36c04c2 | 211 | /* |
5fec0719 | 212 | * Similar to try_grab_folio(): be sure to *also* |
78d9d6ce MWO |
213 | * increment the normal page refcount field at least once, |
214 | * so that the page really is pinned. | |
c36c04c2 | 215 | */ |
5fec0719 MWO |
216 | if (folio_test_large(folio)) { |
217 | folio_ref_add(folio, 1); | |
218 | atomic_add(1, folio_pincount_ptr(folio)); | |
8ea2979c | 219 | } else { |
5fec0719 | 220 | folio_ref_add(folio, GUP_PIN_COUNTING_BIAS); |
8ea2979c | 221 | } |
c36c04c2 | 222 | |
5fec0719 | 223 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, 1); |
c36c04c2 JH |
224 | } |
225 | ||
226 | return true; | |
3faa52c0 JH |
227 | } |
228 | ||
3faa52c0 JH |
229 | /** |
230 | * unpin_user_page() - release a dma-pinned page | |
231 | * @page: pointer to page to be released | |
232 | * | |
233 | * Pages that were pinned via pin_user_pages*() must be released via either | |
234 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
235 | * that such pages can be separately tracked and uniquely handled. In | |
236 | * particular, interactions with RDMA and filesystems need special handling. | |
237 | */ | |
238 | void unpin_user_page(struct page *page) | |
239 | { | |
b6a2619c | 240 | sanity_check_pinned_pages(&page, 1); |
d8ddc099 | 241 | gup_put_folio(page_folio(page), 1, FOLL_PIN); |
3faa52c0 JH |
242 | } |
243 | EXPORT_SYMBOL(unpin_user_page); | |
244 | ||
659508f9 | 245 | static inline struct folio *gup_folio_range_next(struct page *start, |
8f39f5fc | 246 | unsigned long npages, unsigned long i, unsigned int *ntails) |
458a4f78 | 247 | { |
659508f9 MWO |
248 | struct page *next = nth_page(start, i); |
249 | struct folio *folio = page_folio(next); | |
458a4f78 JM |
250 | unsigned int nr = 1; |
251 | ||
659508f9 | 252 | if (folio_test_large(folio)) |
4c654229 | 253 | nr = min_t(unsigned int, npages - i, |
659508f9 | 254 | folio_nr_pages(folio) - folio_page_idx(folio, next)); |
458a4f78 | 255 | |
458a4f78 | 256 | *ntails = nr; |
659508f9 | 257 | return folio; |
458a4f78 JM |
258 | } |
259 | ||
12521c76 | 260 | static inline struct folio *gup_folio_next(struct page **list, |
28297dbc | 261 | unsigned long npages, unsigned long i, unsigned int *ntails) |
8745d7f6 | 262 | { |
12521c76 | 263 | struct folio *folio = page_folio(list[i]); |
8745d7f6 JM |
264 | unsigned int nr; |
265 | ||
8745d7f6 | 266 | for (nr = i + 1; nr < npages; nr++) { |
12521c76 | 267 | if (page_folio(list[nr]) != folio) |
8745d7f6 JM |
268 | break; |
269 | } | |
270 | ||
8745d7f6 | 271 | *ntails = nr - i; |
12521c76 | 272 | return folio; |
8745d7f6 JM |
273 | } |
274 | ||
fc1d8e7c | 275 | /** |
f1f6a7dd | 276 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 277 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 278 | * @npages: number of pages in the @pages array. |
2d15eb31 | 279 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
280 | * |
281 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
282 | * variants called on that page. | |
283 | * | |
284 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 285 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
286 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
287 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 288 | * |
f1f6a7dd | 289 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 290 | * |
2d15eb31 | 291 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
292 | * required, then the caller should a) verify that this is really correct, | |
293 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 294 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
295 | * |
296 | */ | |
f1f6a7dd JH |
297 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
298 | bool make_dirty) | |
fc1d8e7c | 299 | { |
12521c76 MWO |
300 | unsigned long i; |
301 | struct folio *folio; | |
302 | unsigned int nr; | |
2d15eb31 | 303 | |
304 | if (!make_dirty) { | |
f1f6a7dd | 305 | unpin_user_pages(pages, npages); |
2d15eb31 | 306 | return; |
307 | } | |
308 | ||
b6a2619c | 309 | sanity_check_pinned_pages(pages, npages); |
12521c76 MWO |
310 | for (i = 0; i < npages; i += nr) { |
311 | folio = gup_folio_next(pages, npages, i, &nr); | |
2d15eb31 | 312 | /* |
313 | * Checking PageDirty at this point may race with | |
314 | * clear_page_dirty_for_io(), but that's OK. Two key | |
315 | * cases: | |
316 | * | |
317 | * 1) This code sees the page as already dirty, so it | |
318 | * skips the call to set_page_dirty(). That could happen | |
319 | * because clear_page_dirty_for_io() called | |
320 | * page_mkclean(), followed by set_page_dirty(). | |
321 | * However, now the page is going to get written back, | |
322 | * which meets the original intention of setting it | |
323 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
324 | * on to call TestClearPageDirty(), and write the page | |
325 | * back. | |
326 | * | |
327 | * 2) This code sees the page as clean, so it calls | |
328 | * set_page_dirty(). The page stays dirty, despite being | |
329 | * written back, so it gets written back again in the | |
330 | * next writeback cycle. This is harmless. | |
331 | */ | |
12521c76 MWO |
332 | if (!folio_test_dirty(folio)) { |
333 | folio_lock(folio); | |
334 | folio_mark_dirty(folio); | |
335 | folio_unlock(folio); | |
336 | } | |
337 | gup_put_folio(folio, nr, FOLL_PIN); | |
2d15eb31 | 338 | } |
fc1d8e7c | 339 | } |
f1f6a7dd | 340 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c | 341 | |
458a4f78 JM |
342 | /** |
343 | * unpin_user_page_range_dirty_lock() - release and optionally dirty | |
344 | * gup-pinned page range | |
345 | * | |
346 | * @page: the starting page of a range maybe marked dirty, and definitely released. | |
347 | * @npages: number of consecutive pages to release. | |
348 | * @make_dirty: whether to mark the pages dirty | |
349 | * | |
350 | * "gup-pinned page range" refers to a range of pages that has had one of the | |
351 | * pin_user_pages() variants called on that page. | |
352 | * | |
353 | * For the page ranges defined by [page .. page+npages], make that range (or | |
354 | * its head pages, if a compound page) dirty, if @make_dirty is true, and if the | |
355 | * page range was previously listed as clean. | |
356 | * | |
357 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is | |
358 | * required, then the caller should a) verify that this is really correct, | |
359 | * because _lock() is usually required, and b) hand code it: | |
360 | * set_page_dirty_lock(), unpin_user_page(). | |
361 | * | |
362 | */ | |
363 | void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, | |
364 | bool make_dirty) | |
365 | { | |
659508f9 MWO |
366 | unsigned long i; |
367 | struct folio *folio; | |
368 | unsigned int nr; | |
369 | ||
370 | for (i = 0; i < npages; i += nr) { | |
371 | folio = gup_folio_range_next(page, npages, i, &nr); | |
372 | if (make_dirty && !folio_test_dirty(folio)) { | |
373 | folio_lock(folio); | |
374 | folio_mark_dirty(folio); | |
375 | folio_unlock(folio); | |
376 | } | |
377 | gup_put_folio(folio, nr, FOLL_PIN); | |
458a4f78 JM |
378 | } |
379 | } | |
380 | EXPORT_SYMBOL(unpin_user_page_range_dirty_lock); | |
381 | ||
b6a2619c DH |
382 | static void unpin_user_pages_lockless(struct page **pages, unsigned long npages) |
383 | { | |
384 | unsigned long i; | |
385 | struct folio *folio; | |
386 | unsigned int nr; | |
387 | ||
388 | /* | |
389 | * Don't perform any sanity checks because we might have raced with | |
390 | * fork() and some anonymous pages might now actually be shared -- | |
391 | * which is why we're unpinning after all. | |
392 | */ | |
393 | for (i = 0; i < npages; i += nr) { | |
394 | folio = gup_folio_next(pages, npages, i, &nr); | |
395 | gup_put_folio(folio, nr, FOLL_PIN); | |
396 | } | |
397 | } | |
398 | ||
fc1d8e7c | 399 | /** |
f1f6a7dd | 400 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
401 | * @pages: array of pages to be marked dirty and released. |
402 | * @npages: number of pages in the @pages array. | |
403 | * | |
f1f6a7dd | 404 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 405 | * |
f1f6a7dd | 406 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 407 | */ |
f1f6a7dd | 408 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c | 409 | { |
12521c76 MWO |
410 | unsigned long i; |
411 | struct folio *folio; | |
412 | unsigned int nr; | |
fc1d8e7c | 413 | |
146608bb JH |
414 | /* |
415 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
416 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
417 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
418 | */ | |
419 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
420 | return; | |
31b912de | 421 | |
b6a2619c | 422 | sanity_check_pinned_pages(pages, npages); |
12521c76 MWO |
423 | for (i = 0; i < npages; i += nr) { |
424 | folio = gup_folio_next(pages, npages, i, &nr); | |
425 | gup_put_folio(folio, nr, FOLL_PIN); | |
e7602748 | 426 | } |
fc1d8e7c | 427 | } |
f1f6a7dd | 428 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 429 | |
a458b76a AA |
430 | /* |
431 | * Set the MMF_HAS_PINNED if not set yet; after set it'll be there for the mm's | |
432 | * lifecycle. Avoid setting the bit unless necessary, or it might cause write | |
433 | * cache bouncing on large SMP machines for concurrent pinned gups. | |
434 | */ | |
435 | static inline void mm_set_has_pinned_flag(unsigned long *mm_flags) | |
436 | { | |
437 | if (!test_bit(MMF_HAS_PINNED, mm_flags)) | |
438 | set_bit(MMF_HAS_PINNED, mm_flags); | |
439 | } | |
440 | ||
050a9adc | 441 | #ifdef CONFIG_MMU |
69e68b4f KS |
442 | static struct page *no_page_table(struct vm_area_struct *vma, |
443 | unsigned int flags) | |
4bbd4c77 | 444 | { |
69e68b4f KS |
445 | /* |
446 | * When core dumping an enormous anonymous area that nobody | |
447 | * has touched so far, we don't want to allocate unnecessary pages or | |
448 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
449 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
450 | * But we can only make this optimization where a hole would surely | |
451 | * be zero-filled if handle_mm_fault() actually did handle it. | |
452 | */ | |
a0137f16 AK |
453 | if ((flags & FOLL_DUMP) && |
454 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
455 | return ERR_PTR(-EFAULT); |
456 | return NULL; | |
457 | } | |
4bbd4c77 | 458 | |
1027e443 KS |
459 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
460 | pte_t *pte, unsigned int flags) | |
461 | { | |
1027e443 KS |
462 | if (flags & FOLL_TOUCH) { |
463 | pte_t entry = *pte; | |
464 | ||
465 | if (flags & FOLL_WRITE) | |
466 | entry = pte_mkdirty(entry); | |
467 | entry = pte_mkyoung(entry); | |
468 | ||
469 | if (!pte_same(*pte, entry)) { | |
470 | set_pte_at(vma->vm_mm, address, pte, entry); | |
471 | update_mmu_cache(vma, address, pte); | |
472 | } | |
473 | } | |
474 | ||
475 | /* Proper page table entry exists, but no corresponding struct page */ | |
476 | return -EEXIST; | |
477 | } | |
478 | ||
19be0eaf | 479 | /* |
a308c71b PX |
480 | * FOLL_FORCE can write to even unwritable pte's, but only |
481 | * after we've gone through a COW cycle and they are dirty. | |
19be0eaf LT |
482 | */ |
483 | static inline bool can_follow_write_pte(pte_t pte, unsigned int flags) | |
484 | { | |
a308c71b PX |
485 | return pte_write(pte) || |
486 | ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte)); | |
19be0eaf LT |
487 | } |
488 | ||
69e68b4f | 489 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
490 | unsigned long address, pmd_t *pmd, unsigned int flags, |
491 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
492 | { |
493 | struct mm_struct *mm = vma->vm_mm; | |
494 | struct page *page; | |
495 | spinlock_t *ptl; | |
496 | pte_t *ptep, pte; | |
f28d4363 | 497 | int ret; |
4bbd4c77 | 498 | |
eddb1c22 JH |
499 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
500 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
501 | (FOLL_PIN | FOLL_GET))) | |
502 | return ERR_PTR(-EINVAL); | |
69e68b4f | 503 | retry: |
4bbd4c77 | 504 | if (unlikely(pmd_bad(*pmd))) |
69e68b4f | 505 | return no_page_table(vma, flags); |
4bbd4c77 KS |
506 | |
507 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
4bbd4c77 KS |
508 | pte = *ptep; |
509 | if (!pte_present(pte)) { | |
510 | swp_entry_t entry; | |
511 | /* | |
512 | * KSM's break_ksm() relies upon recognizing a ksm page | |
513 | * even while it is being migrated, so for that case we | |
514 | * need migration_entry_wait(). | |
515 | */ | |
516 | if (likely(!(flags & FOLL_MIGRATION))) | |
517 | goto no_page; | |
0661a336 | 518 | if (pte_none(pte)) |
4bbd4c77 KS |
519 | goto no_page; |
520 | entry = pte_to_swp_entry(pte); | |
521 | if (!is_migration_entry(entry)) | |
522 | goto no_page; | |
523 | pte_unmap_unlock(ptep, ptl); | |
524 | migration_entry_wait(mm, pmd, address); | |
69e68b4f | 525 | goto retry; |
4bbd4c77 | 526 | } |
8a0516ed | 527 | if ((flags & FOLL_NUMA) && pte_protnone(pte)) |
4bbd4c77 | 528 | goto no_page; |
19be0eaf | 529 | if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) { |
69e68b4f KS |
530 | pte_unmap_unlock(ptep, ptl); |
531 | return NULL; | |
532 | } | |
4bbd4c77 KS |
533 | |
534 | page = vm_normal_page(vma, address, pte); | |
3faa52c0 | 535 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 536 | /* |
3faa52c0 JH |
537 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
538 | * case since they are only valid while holding the pgmap | |
539 | * reference. | |
3565fce3 | 540 | */ |
df06b37f KB |
541 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
542 | if (*pgmap) | |
3565fce3 DW |
543 | page = pte_page(pte); |
544 | else | |
545 | goto no_page; | |
546 | } else if (unlikely(!page)) { | |
1027e443 KS |
547 | if (flags & FOLL_DUMP) { |
548 | /* Avoid special (like zero) pages in core dumps */ | |
549 | page = ERR_PTR(-EFAULT); | |
550 | goto out; | |
551 | } | |
552 | ||
553 | if (is_zero_pfn(pte_pfn(pte))) { | |
554 | page = pte_page(pte); | |
555 | } else { | |
1027e443 KS |
556 | ret = follow_pfn_pte(vma, address, ptep, flags); |
557 | page = ERR_PTR(ret); | |
558 | goto out; | |
559 | } | |
4bbd4c77 KS |
560 | } |
561 | ||
a7f22660 DH |
562 | if (!pte_write(pte) && gup_must_unshare(flags, page)) { |
563 | page = ERR_PTR(-EMLINK); | |
564 | goto out; | |
565 | } | |
b6a2619c DH |
566 | |
567 | VM_BUG_ON_PAGE((flags & FOLL_PIN) && PageAnon(page) && | |
568 | !PageAnonExclusive(page), page); | |
569 | ||
3faa52c0 JH |
570 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
571 | if (unlikely(!try_grab_page(page, flags))) { | |
572 | page = ERR_PTR(-ENOMEM); | |
573 | goto out; | |
8fde12ca | 574 | } |
f28d4363 CI |
575 | /* |
576 | * We need to make the page accessible if and only if we are going | |
577 | * to access its content (the FOLL_PIN case). Please see | |
578 | * Documentation/core-api/pin_user_pages.rst for details. | |
579 | */ | |
580 | if (flags & FOLL_PIN) { | |
581 | ret = arch_make_page_accessible(page); | |
582 | if (ret) { | |
583 | unpin_user_page(page); | |
584 | page = ERR_PTR(ret); | |
585 | goto out; | |
586 | } | |
587 | } | |
4bbd4c77 KS |
588 | if (flags & FOLL_TOUCH) { |
589 | if ((flags & FOLL_WRITE) && | |
590 | !pte_dirty(pte) && !PageDirty(page)) | |
591 | set_page_dirty(page); | |
592 | /* | |
593 | * pte_mkyoung() would be more correct here, but atomic care | |
594 | * is needed to avoid losing the dirty bit: it is easier to use | |
595 | * mark_page_accessed(). | |
596 | */ | |
597 | mark_page_accessed(page); | |
598 | } | |
1027e443 | 599 | out: |
4bbd4c77 | 600 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 601 | return page; |
4bbd4c77 KS |
602 | no_page: |
603 | pte_unmap_unlock(ptep, ptl); | |
604 | if (!pte_none(pte)) | |
69e68b4f KS |
605 | return NULL; |
606 | return no_page_table(vma, flags); | |
607 | } | |
608 | ||
080dbb61 AK |
609 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
610 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
611 | unsigned int flags, |
612 | struct follow_page_context *ctx) | |
69e68b4f | 613 | { |
68827280 | 614 | pmd_t *pmd, pmdval; |
69e68b4f KS |
615 | spinlock_t *ptl; |
616 | struct page *page; | |
617 | struct mm_struct *mm = vma->vm_mm; | |
618 | ||
080dbb61 | 619 | pmd = pmd_offset(pudp, address); |
68827280 HY |
620 | /* |
621 | * The READ_ONCE() will stabilize the pmdval in a register or | |
622 | * on the stack so that it will stop changing under the code. | |
623 | */ | |
624 | pmdval = READ_ONCE(*pmd); | |
625 | if (pmd_none(pmdval)) | |
69e68b4f | 626 | return no_page_table(vma, flags); |
be9d3045 | 627 | if (pmd_huge(pmdval) && is_vm_hugetlb_page(vma)) { |
e66f17ff NH |
628 | page = follow_huge_pmd(mm, address, pmd, flags); |
629 | if (page) | |
630 | return page; | |
631 | return no_page_table(vma, flags); | |
69e68b4f | 632 | } |
68827280 | 633 | if (is_hugepd(__hugepd(pmd_val(pmdval)))) { |
4dc71451 | 634 | page = follow_huge_pd(vma, address, |
68827280 | 635 | __hugepd(pmd_val(pmdval)), flags, |
4dc71451 AK |
636 | PMD_SHIFT); |
637 | if (page) | |
638 | return page; | |
639 | return no_page_table(vma, flags); | |
640 | } | |
84c3fc4e | 641 | retry: |
68827280 | 642 | if (!pmd_present(pmdval)) { |
28b0ee3f LX |
643 | /* |
644 | * Should never reach here, if thp migration is not supported; | |
645 | * Otherwise, it must be a thp migration entry. | |
646 | */ | |
647 | VM_BUG_ON(!thp_migration_supported() || | |
648 | !is_pmd_migration_entry(pmdval)); | |
649 | ||
84c3fc4e ZY |
650 | if (likely(!(flags & FOLL_MIGRATION))) |
651 | return no_page_table(vma, flags); | |
28b0ee3f LX |
652 | |
653 | pmd_migration_entry_wait(mm, pmd); | |
68827280 HY |
654 | pmdval = READ_ONCE(*pmd); |
655 | /* | |
656 | * MADV_DONTNEED may convert the pmd to null because | |
c1e8d7c6 | 657 | * mmap_lock is held in read mode |
68827280 HY |
658 | */ |
659 | if (pmd_none(pmdval)) | |
660 | return no_page_table(vma, flags); | |
84c3fc4e ZY |
661 | goto retry; |
662 | } | |
68827280 | 663 | if (pmd_devmap(pmdval)) { |
3565fce3 | 664 | ptl = pmd_lock(mm, pmd); |
df06b37f | 665 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
666 | spin_unlock(ptl); |
667 | if (page) | |
668 | return page; | |
669 | } | |
68827280 | 670 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 671 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 672 | |
68827280 | 673 | if ((flags & FOLL_NUMA) && pmd_protnone(pmdval)) |
db08f203 AK |
674 | return no_page_table(vma, flags); |
675 | ||
84c3fc4e | 676 | retry_locked: |
6742d293 | 677 | ptl = pmd_lock(mm, pmd); |
68827280 HY |
678 | if (unlikely(pmd_none(*pmd))) { |
679 | spin_unlock(ptl); | |
680 | return no_page_table(vma, flags); | |
681 | } | |
84c3fc4e ZY |
682 | if (unlikely(!pmd_present(*pmd))) { |
683 | spin_unlock(ptl); | |
684 | if (likely(!(flags & FOLL_MIGRATION))) | |
685 | return no_page_table(vma, flags); | |
686 | pmd_migration_entry_wait(mm, pmd); | |
687 | goto retry_locked; | |
688 | } | |
6742d293 KS |
689 | if (unlikely(!pmd_trans_huge(*pmd))) { |
690 | spin_unlock(ptl); | |
df06b37f | 691 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 692 | } |
4066c119 | 693 | if (flags & FOLL_SPLIT_PMD) { |
6742d293 KS |
694 | int ret; |
695 | page = pmd_page(*pmd); | |
696 | if (is_huge_zero_page(page)) { | |
697 | spin_unlock(ptl); | |
698 | ret = 0; | |
78ddc534 | 699 | split_huge_pmd(vma, pmd, address); |
337d9abf NH |
700 | if (pmd_trans_unstable(pmd)) |
701 | ret = -EBUSY; | |
4066c119 | 702 | } else { |
bfe7b00d SL |
703 | spin_unlock(ptl); |
704 | split_huge_pmd(vma, pmd, address); | |
705 | ret = pte_alloc(mm, pmd) ? -ENOMEM : 0; | |
6742d293 KS |
706 | } |
707 | ||
708 | return ret ? ERR_PTR(ret) : | |
df06b37f | 709 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 710 | } |
6742d293 KS |
711 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
712 | spin_unlock(ptl); | |
df06b37f | 713 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 714 | return page; |
4bbd4c77 KS |
715 | } |
716 | ||
080dbb61 AK |
717 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
718 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
719 | unsigned int flags, |
720 | struct follow_page_context *ctx) | |
080dbb61 AK |
721 | { |
722 | pud_t *pud; | |
723 | spinlock_t *ptl; | |
724 | struct page *page; | |
725 | struct mm_struct *mm = vma->vm_mm; | |
726 | ||
727 | pud = pud_offset(p4dp, address); | |
728 | if (pud_none(*pud)) | |
729 | return no_page_table(vma, flags); | |
be9d3045 | 730 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) { |
080dbb61 AK |
731 | page = follow_huge_pud(mm, address, pud, flags); |
732 | if (page) | |
733 | return page; | |
734 | return no_page_table(vma, flags); | |
735 | } | |
4dc71451 AK |
736 | if (is_hugepd(__hugepd(pud_val(*pud)))) { |
737 | page = follow_huge_pd(vma, address, | |
738 | __hugepd(pud_val(*pud)), flags, | |
739 | PUD_SHIFT); | |
740 | if (page) | |
741 | return page; | |
742 | return no_page_table(vma, flags); | |
743 | } | |
080dbb61 AK |
744 | if (pud_devmap(*pud)) { |
745 | ptl = pud_lock(mm, pud); | |
df06b37f | 746 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
747 | spin_unlock(ptl); |
748 | if (page) | |
749 | return page; | |
750 | } | |
751 | if (unlikely(pud_bad(*pud))) | |
752 | return no_page_table(vma, flags); | |
753 | ||
df06b37f | 754 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
755 | } |
756 | ||
080dbb61 AK |
757 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
758 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
759 | unsigned int flags, |
760 | struct follow_page_context *ctx) | |
080dbb61 AK |
761 | { |
762 | p4d_t *p4d; | |
4dc71451 | 763 | struct page *page; |
080dbb61 AK |
764 | |
765 | p4d = p4d_offset(pgdp, address); | |
766 | if (p4d_none(*p4d)) | |
767 | return no_page_table(vma, flags); | |
768 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
769 | if (unlikely(p4d_bad(*p4d))) | |
770 | return no_page_table(vma, flags); | |
771 | ||
4dc71451 AK |
772 | if (is_hugepd(__hugepd(p4d_val(*p4d)))) { |
773 | page = follow_huge_pd(vma, address, | |
774 | __hugepd(p4d_val(*p4d)), flags, | |
775 | P4D_SHIFT); | |
776 | if (page) | |
777 | return page; | |
778 | return no_page_table(vma, flags); | |
779 | } | |
df06b37f | 780 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
781 | } |
782 | ||
783 | /** | |
784 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
785 | * @vma: vm_area_struct mapping @address | |
786 | * @address: virtual address to look up | |
787 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
788 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
789 | * pointer to output page_mask | |
080dbb61 AK |
790 | * |
791 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
792 | * | |
78179556 MR |
793 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
794 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
795 | * | |
a7f22660 DH |
796 | * When getting an anonymous page and the caller has to trigger unsharing |
797 | * of a shared anonymous page first, -EMLINK is returned. The caller should | |
798 | * trigger a fault with FAULT_FLAG_UNSHARE set. Note that unsharing is only | |
799 | * relevant with FOLL_PIN and !FOLL_WRITE. | |
800 | * | |
78179556 MR |
801 | * On output, the @ctx->page_mask is set according to the size of the page. |
802 | * | |
803 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
804 | * an error pointer if there is a mapping to something not represented |
805 | * by a page descriptor (see also vm_normal_page()). | |
806 | */ | |
a7030aea | 807 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 808 | unsigned long address, unsigned int flags, |
df06b37f | 809 | struct follow_page_context *ctx) |
080dbb61 AK |
810 | { |
811 | pgd_t *pgd; | |
812 | struct page *page; | |
813 | struct mm_struct *mm = vma->vm_mm; | |
814 | ||
df06b37f | 815 | ctx->page_mask = 0; |
080dbb61 AK |
816 | |
817 | /* make this handle hugepd */ | |
818 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); | |
819 | if (!IS_ERR(page)) { | |
3faa52c0 | 820 | WARN_ON_ONCE(flags & (FOLL_GET | FOLL_PIN)); |
080dbb61 AK |
821 | return page; |
822 | } | |
823 | ||
824 | pgd = pgd_offset(mm, address); | |
825 | ||
826 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
827 | return no_page_table(vma, flags); | |
828 | ||
faaa5b62 AK |
829 | if (pgd_huge(*pgd)) { |
830 | page = follow_huge_pgd(mm, address, pgd, flags); | |
831 | if (page) | |
832 | return page; | |
833 | return no_page_table(vma, flags); | |
834 | } | |
4dc71451 AK |
835 | if (is_hugepd(__hugepd(pgd_val(*pgd)))) { |
836 | page = follow_huge_pd(vma, address, | |
837 | __hugepd(pgd_val(*pgd)), flags, | |
838 | PGDIR_SHIFT); | |
839 | if (page) | |
840 | return page; | |
841 | return no_page_table(vma, flags); | |
842 | } | |
faaa5b62 | 843 | |
df06b37f KB |
844 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
845 | } | |
846 | ||
847 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
848 | unsigned int foll_flags) | |
849 | { | |
850 | struct follow_page_context ctx = { NULL }; | |
851 | struct page *page; | |
852 | ||
1507f512 MR |
853 | if (vma_is_secretmem(vma)) |
854 | return NULL; | |
855 | ||
8909691b DH |
856 | if (foll_flags & FOLL_PIN) |
857 | return NULL; | |
858 | ||
df06b37f KB |
859 | page = follow_page_mask(vma, address, foll_flags, &ctx); |
860 | if (ctx.pgmap) | |
861 | put_dev_pagemap(ctx.pgmap); | |
862 | return page; | |
080dbb61 AK |
863 | } |
864 | ||
f2b495ca KS |
865 | static int get_gate_page(struct mm_struct *mm, unsigned long address, |
866 | unsigned int gup_flags, struct vm_area_struct **vma, | |
867 | struct page **page) | |
868 | { | |
869 | pgd_t *pgd; | |
c2febafc | 870 | p4d_t *p4d; |
f2b495ca KS |
871 | pud_t *pud; |
872 | pmd_t *pmd; | |
873 | pte_t *pte; | |
874 | int ret = -EFAULT; | |
875 | ||
876 | /* user gate pages are read-only */ | |
877 | if (gup_flags & FOLL_WRITE) | |
878 | return -EFAULT; | |
879 | if (address > TASK_SIZE) | |
880 | pgd = pgd_offset_k(address); | |
881 | else | |
882 | pgd = pgd_offset_gate(mm, address); | |
b5d1c39f AL |
883 | if (pgd_none(*pgd)) |
884 | return -EFAULT; | |
c2febafc | 885 | p4d = p4d_offset(pgd, address); |
b5d1c39f AL |
886 | if (p4d_none(*p4d)) |
887 | return -EFAULT; | |
c2febafc | 888 | pud = pud_offset(p4d, address); |
b5d1c39f AL |
889 | if (pud_none(*pud)) |
890 | return -EFAULT; | |
f2b495ca | 891 | pmd = pmd_offset(pud, address); |
84c3fc4e | 892 | if (!pmd_present(*pmd)) |
f2b495ca KS |
893 | return -EFAULT; |
894 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
895 | pte = pte_offset_map(pmd, address); | |
896 | if (pte_none(*pte)) | |
897 | goto unmap; | |
898 | *vma = get_gate_vma(mm); | |
899 | if (!page) | |
900 | goto out; | |
901 | *page = vm_normal_page(*vma, address, *pte); | |
902 | if (!*page) { | |
903 | if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte))) | |
904 | goto unmap; | |
905 | *page = pte_page(*pte); | |
906 | } | |
9fa2dd94 | 907 | if (unlikely(!try_grab_page(*page, gup_flags))) { |
8fde12ca LT |
908 | ret = -ENOMEM; |
909 | goto unmap; | |
910 | } | |
f2b495ca KS |
911 | out: |
912 | ret = 0; | |
913 | unmap: | |
914 | pte_unmap(pte); | |
915 | return ret; | |
916 | } | |
917 | ||
9a95f3cf | 918 | /* |
c1e8d7c6 ML |
919 | * mmap_lock must be held on entry. If @locked != NULL and *@flags |
920 | * does not include FOLL_NOWAIT, the mmap_lock may be released. If it | |
4f6da934 | 921 | * is, *@locked will be set to 0 and -EBUSY returned. |
9a95f3cf | 922 | */ |
64019a2e | 923 | static int faultin_page(struct vm_area_struct *vma, |
a7f22660 DH |
924 | unsigned long address, unsigned int *flags, bool unshare, |
925 | int *locked) | |
16744483 | 926 | { |
16744483 | 927 | unsigned int fault_flags = 0; |
2b740303 | 928 | vm_fault_t ret; |
16744483 | 929 | |
55b8fe70 AG |
930 | if (*flags & FOLL_NOFAULT) |
931 | return -EFAULT; | |
16744483 KS |
932 | if (*flags & FOLL_WRITE) |
933 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
934 | if (*flags & FOLL_REMOTE) |
935 | fault_flags |= FAULT_FLAG_REMOTE; | |
4f6da934 | 936 | if (locked) |
71335f37 | 937 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
16744483 KS |
938 | if (*flags & FOLL_NOWAIT) |
939 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 940 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
941 | /* |
942 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
943 | * can co-exist | |
944 | */ | |
234b239b ALC |
945 | fault_flags |= FAULT_FLAG_TRIED; |
946 | } | |
a7f22660 DH |
947 | if (unshare) { |
948 | fault_flags |= FAULT_FLAG_UNSHARE; | |
949 | /* FAULT_FLAG_WRITE and FAULT_FLAG_UNSHARE are incompatible */ | |
950 | VM_BUG_ON(fault_flags & FAULT_FLAG_WRITE); | |
951 | } | |
16744483 | 952 | |
bce617ed | 953 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
d9272525 PX |
954 | |
955 | if (ret & VM_FAULT_COMPLETED) { | |
956 | /* | |
957 | * With FAULT_FLAG_RETRY_NOWAIT we'll never release the | |
958 | * mmap lock in the page fault handler. Sanity check this. | |
959 | */ | |
960 | WARN_ON_ONCE(fault_flags & FAULT_FLAG_RETRY_NOWAIT); | |
961 | if (locked) | |
962 | *locked = 0; | |
963 | /* | |
964 | * We should do the same as VM_FAULT_RETRY, but let's not | |
965 | * return -EBUSY since that's not reflecting the reality of | |
966 | * what has happened - we've just fully completed a page | |
967 | * fault, with the mmap lock released. Use -EAGAIN to show | |
968 | * that we want to take the mmap lock _again_. | |
969 | */ | |
970 | return -EAGAIN; | |
971 | } | |
972 | ||
16744483 | 973 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
974 | int err = vm_fault_to_errno(ret, *flags); |
975 | ||
976 | if (err) | |
977 | return err; | |
16744483 KS |
978 | BUG(); |
979 | } | |
980 | ||
16744483 | 981 | if (ret & VM_FAULT_RETRY) { |
4f6da934 PX |
982 | if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
983 | *locked = 0; | |
16744483 KS |
984 | return -EBUSY; |
985 | } | |
986 | ||
987 | /* | |
988 | * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when | |
989 | * necessary, even if maybe_mkwrite decided not to set pte_write. We | |
990 | * can thus safely do subsequent page lookups as if they were reads. | |
991 | * But only do so when looping for pte_write is futile: in some cases | |
992 | * userspace may also be wanting to write to the gotten user page, | |
993 | * which a read fault here might prevent (a readonly page might get | |
994 | * reCOWed by userspace write). | |
995 | */ | |
996 | if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE)) | |
2923117b | 997 | *flags |= FOLL_COW; |
16744483 KS |
998 | return 0; |
999 | } | |
1000 | ||
fa5bb209 KS |
1001 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
1002 | { | |
1003 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
1004 | int write = (gup_flags & FOLL_WRITE); |
1005 | int foreign = (gup_flags & FOLL_REMOTE); | |
fa5bb209 KS |
1006 | |
1007 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
1008 | return -EFAULT; | |
1009 | ||
7f7ccc2c WT |
1010 | if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma)) |
1011 | return -EFAULT; | |
1012 | ||
52650c8b JG |
1013 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
1014 | return -EOPNOTSUPP; | |
1015 | ||
1507f512 MR |
1016 | if (vma_is_secretmem(vma)) |
1017 | return -EFAULT; | |
1018 | ||
1b2ee126 | 1019 | if (write) { |
fa5bb209 KS |
1020 | if (!(vm_flags & VM_WRITE)) { |
1021 | if (!(gup_flags & FOLL_FORCE)) | |
1022 | return -EFAULT; | |
1023 | /* | |
1024 | * We used to let the write,force case do COW in a | |
1025 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
1026 | * set a breakpoint in a read-only mapping of an | |
1027 | * executable, without corrupting the file (yet only | |
1028 | * when that file had been opened for writing!). | |
1029 | * Anon pages in shared mappings are surprising: now | |
1030 | * just reject it. | |
1031 | */ | |
46435364 | 1032 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 1033 | return -EFAULT; |
fa5bb209 KS |
1034 | } |
1035 | } else if (!(vm_flags & VM_READ)) { | |
1036 | if (!(gup_flags & FOLL_FORCE)) | |
1037 | return -EFAULT; | |
1038 | /* | |
1039 | * Is there actually any vma we can reach here which does not | |
1040 | * have VM_MAYREAD set? | |
1041 | */ | |
1042 | if (!(vm_flags & VM_MAYREAD)) | |
1043 | return -EFAULT; | |
1044 | } | |
d61172b4 DH |
1045 | /* |
1046 | * gups are always data accesses, not instruction | |
1047 | * fetches, so execute=false here | |
1048 | */ | |
1049 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 1050 | return -EFAULT; |
fa5bb209 KS |
1051 | return 0; |
1052 | } | |
1053 | ||
4bbd4c77 KS |
1054 | /** |
1055 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
1056 | * @mm: mm_struct of target mm |
1057 | * @start: starting user address | |
1058 | * @nr_pages: number of pages from start to pin | |
1059 | * @gup_flags: flags modifying pin behaviour | |
1060 | * @pages: array that receives pointers to the pages pinned. | |
1061 | * Should be at least nr_pages long. Or NULL, if caller | |
1062 | * only intends to ensure the pages are faulted in. | |
1063 | * @vmas: array of pointers to vmas corresponding to each page. | |
1064 | * Or NULL if the caller does not require them. | |
c1e8d7c6 | 1065 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 1066 | * |
d2dfbe47 LX |
1067 | * Returns either number of pages pinned (which may be less than the |
1068 | * number requested), or an error. Details about the return value: | |
1069 | * | |
1070 | * -- If nr_pages is 0, returns 0. | |
1071 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1072 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1073 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 1074 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
1075 | * |
1076 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1077 | * | |
c1e8d7c6 | 1078 | * @vmas are valid only as long as mmap_lock is held. |
4bbd4c77 | 1079 | * |
c1e8d7c6 | 1080 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
1081 | * |
1082 | * __get_user_pages walks a process's page tables and takes a reference to | |
1083 | * each struct page that each user address corresponds to at a given | |
1084 | * instant. That is, it takes the page that would be accessed if a user | |
1085 | * thread accesses the given user virtual address at that instant. | |
1086 | * | |
1087 | * This does not guarantee that the page exists in the user mappings when | |
1088 | * __get_user_pages returns, and there may even be a completely different | |
1089 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1090 | * and subsequently re faulted). However it does guarantee that the page | |
1091 | * won't be freed completely. And mostly callers simply care that the page | |
1092 | * contains data that was valid *at some point in time*. Typically, an IO | |
1093 | * or similar operation cannot guarantee anything stronger anyway because | |
1094 | * locks can't be held over the syscall boundary. | |
1095 | * | |
1096 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1097 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1098 | * appropriate) must be called after the page is finished with, and | |
1099 | * before put_page is called. | |
1100 | * | |
c1e8d7c6 | 1101 | * If @locked != NULL, *@locked will be set to 0 when mmap_lock is |
4f6da934 PX |
1102 | * released by an up_read(). That can happen if @gup_flags does not |
1103 | * have FOLL_NOWAIT. | |
9a95f3cf | 1104 | * |
4f6da934 | 1105 | * A caller using such a combination of @locked and @gup_flags |
c1e8d7c6 | 1106 | * must therefore hold the mmap_lock for reading only, and recognize |
9a95f3cf PC |
1107 | * when it's been released. Otherwise, it must be held for either |
1108 | * reading or writing and will not be released. | |
4bbd4c77 KS |
1109 | * |
1110 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1111 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1112 | * you need some special @gup_flags. | |
1113 | */ | |
64019a2e | 1114 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
1115 | unsigned long start, unsigned long nr_pages, |
1116 | unsigned int gup_flags, struct page **pages, | |
4f6da934 | 1117 | struct vm_area_struct **vmas, int *locked) |
4bbd4c77 | 1118 | { |
df06b37f | 1119 | long ret = 0, i = 0; |
fa5bb209 | 1120 | struct vm_area_struct *vma = NULL; |
df06b37f | 1121 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
1122 | |
1123 | if (!nr_pages) | |
1124 | return 0; | |
1125 | ||
f9652594 AK |
1126 | start = untagged_addr(start); |
1127 | ||
eddb1c22 | 1128 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 KS |
1129 | |
1130 | /* | |
1131 | * If FOLL_FORCE is set then do not force a full fault as the hinting | |
1132 | * fault information is unrelated to the reference behaviour of a task | |
1133 | * using the address space | |
1134 | */ | |
1135 | if (!(gup_flags & FOLL_FORCE)) | |
1136 | gup_flags |= FOLL_NUMA; | |
1137 | ||
4bbd4c77 | 1138 | do { |
fa5bb209 KS |
1139 | struct page *page; |
1140 | unsigned int foll_flags = gup_flags; | |
1141 | unsigned int page_increm; | |
1142 | ||
1143 | /* first iteration or cross vma bound */ | |
1144 | if (!vma || start >= vma->vm_end) { | |
1145 | vma = find_extend_vma(mm, start); | |
1146 | if (!vma && in_gate_area(mm, start)) { | |
fa5bb209 KS |
1147 | ret = get_gate_page(mm, start & PAGE_MASK, |
1148 | gup_flags, &vma, | |
1149 | pages ? &pages[i] : NULL); | |
1150 | if (ret) | |
08be37b7 | 1151 | goto out; |
df06b37f | 1152 | ctx.page_mask = 0; |
fa5bb209 KS |
1153 | goto next_page; |
1154 | } | |
4bbd4c77 | 1155 | |
52650c8b | 1156 | if (!vma) { |
df06b37f KB |
1157 | ret = -EFAULT; |
1158 | goto out; | |
1159 | } | |
52650c8b JG |
1160 | ret = check_vma_flags(vma, gup_flags); |
1161 | if (ret) | |
1162 | goto out; | |
1163 | ||
fa5bb209 KS |
1164 | if (is_vm_hugetlb_page(vma)) { |
1165 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1166 | &start, &nr_pages, i, | |
a308c71b | 1167 | gup_flags, locked); |
ad415db8 PX |
1168 | if (locked && *locked == 0) { |
1169 | /* | |
1170 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1171 | * and we've lost mmap_lock. |
ad415db8 PX |
1172 | * We must stop here. |
1173 | */ | |
1174 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
ad415db8 PX |
1175 | goto out; |
1176 | } | |
fa5bb209 | 1177 | continue; |
4bbd4c77 | 1178 | } |
fa5bb209 KS |
1179 | } |
1180 | retry: | |
1181 | /* | |
1182 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1183 | * potentially allocating memory. | |
1184 | */ | |
fa45f116 | 1185 | if (fatal_signal_pending(current)) { |
d180870d | 1186 | ret = -EINTR; |
df06b37f KB |
1187 | goto out; |
1188 | } | |
fa5bb209 | 1189 | cond_resched(); |
df06b37f KB |
1190 | |
1191 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
a7f22660 DH |
1192 | if (!page || PTR_ERR(page) == -EMLINK) { |
1193 | ret = faultin_page(vma, start, &foll_flags, | |
1194 | PTR_ERR(page) == -EMLINK, locked); | |
fa5bb209 KS |
1195 | switch (ret) { |
1196 | case 0: | |
1197 | goto retry; | |
df06b37f | 1198 | case -EBUSY: |
d9272525 | 1199 | case -EAGAIN: |
df06b37f | 1200 | ret = 0; |
e4a9bc58 | 1201 | fallthrough; |
fa5bb209 KS |
1202 | case -EFAULT: |
1203 | case -ENOMEM: | |
1204 | case -EHWPOISON: | |
df06b37f | 1205 | goto out; |
4bbd4c77 | 1206 | } |
fa5bb209 | 1207 | BUG(); |
1027e443 KS |
1208 | } else if (PTR_ERR(page) == -EEXIST) { |
1209 | /* | |
1210 | * Proper page table entry exists, but no corresponding | |
65462462 JH |
1211 | * struct page. If the caller expects **pages to be |
1212 | * filled in, bail out now, because that can't be done | |
1213 | * for this page. | |
1027e443 | 1214 | */ |
65462462 JH |
1215 | if (pages) { |
1216 | ret = PTR_ERR(page); | |
1217 | goto out; | |
1218 | } | |
1219 | ||
1027e443 KS |
1220 | goto next_page; |
1221 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1222 | ret = PTR_ERR(page); |
1223 | goto out; | |
1027e443 | 1224 | } |
fa5bb209 KS |
1225 | if (pages) { |
1226 | pages[i] = page; | |
1227 | flush_anon_page(vma, page, start); | |
1228 | flush_dcache_page(page); | |
df06b37f | 1229 | ctx.page_mask = 0; |
4bbd4c77 | 1230 | } |
4bbd4c77 | 1231 | next_page: |
fa5bb209 KS |
1232 | if (vmas) { |
1233 | vmas[i] = vma; | |
df06b37f | 1234 | ctx.page_mask = 0; |
fa5bb209 | 1235 | } |
df06b37f | 1236 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1237 | if (page_increm > nr_pages) |
1238 | page_increm = nr_pages; | |
1239 | i += page_increm; | |
1240 | start += page_increm * PAGE_SIZE; | |
1241 | nr_pages -= page_increm; | |
4bbd4c77 | 1242 | } while (nr_pages); |
df06b37f KB |
1243 | out: |
1244 | if (ctx.pgmap) | |
1245 | put_dev_pagemap(ctx.pgmap); | |
1246 | return i ? i : ret; | |
4bbd4c77 | 1247 | } |
4bbd4c77 | 1248 | |
771ab430 TK |
1249 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1250 | unsigned int fault_flags) | |
d4925e00 | 1251 | { |
1b2ee126 DH |
1252 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1253 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1254 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1255 | |
1256 | if (!(vm_flags & vma->vm_flags)) | |
1257 | return false; | |
1258 | ||
33a709b2 DH |
1259 | /* |
1260 | * The architecture might have a hardware protection | |
1b2ee126 | 1261 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1262 | * |
1263 | * gup always represents data access, not instruction | |
1264 | * fetches, so execute=false here: | |
33a709b2 | 1265 | */ |
d61172b4 | 1266 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1267 | return false; |
1268 | ||
d4925e00 DH |
1269 | return true; |
1270 | } | |
1271 | ||
adc8cb40 | 1272 | /** |
4bbd4c77 | 1273 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1274 | * @mm: mm_struct of target mm |
1275 | * @address: user address | |
1276 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1277 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1278 | * does not allow retry. If NULL, the caller must guarantee |
1279 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1280 | * |
1281 | * This is meant to be called in the specific scenario where for locking reasons | |
1282 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1283 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1284 | * trying again. | |
1285 | * | |
1286 | * Typically this is meant to be used by the futex code. | |
1287 | * | |
1288 | * The main difference with get_user_pages() is that this function will | |
1289 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1290 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1291 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1292 | * |
1293 | * This is important for some architectures where those bits also gate the | |
1294 | * access permission to the page because they are maintained in software. On | |
1295 | * such architectures, gup() will not be enough to make a subsequent access | |
1296 | * succeed. | |
1297 | * | |
c1e8d7c6 ML |
1298 | * This function will not return with an unlocked mmap_lock. So it has not the |
1299 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1300 | */ |
64019a2e | 1301 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1302 | unsigned long address, unsigned int fault_flags, |
1303 | bool *unlocked) | |
4bbd4c77 KS |
1304 | { |
1305 | struct vm_area_struct *vma; | |
8fed2f3c | 1306 | vm_fault_t ret; |
4a9e1cda | 1307 | |
f9652594 AK |
1308 | address = untagged_addr(address); |
1309 | ||
4a9e1cda | 1310 | if (unlocked) |
71335f37 | 1311 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1312 | |
4a9e1cda | 1313 | retry: |
4bbd4c77 KS |
1314 | vma = find_extend_vma(mm, address); |
1315 | if (!vma || address < vma->vm_start) | |
1316 | return -EFAULT; | |
1317 | ||
d4925e00 | 1318 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1319 | return -EFAULT; |
1320 | ||
475f4dfc PX |
1321 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1322 | fatal_signal_pending(current)) | |
1323 | return -EINTR; | |
1324 | ||
bce617ed | 1325 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
d9272525 PX |
1326 | |
1327 | if (ret & VM_FAULT_COMPLETED) { | |
1328 | /* | |
1329 | * NOTE: it's a pity that we need to retake the lock here | |
1330 | * to pair with the unlock() in the callers. Ideally we | |
1331 | * could tell the callers so they do not need to unlock. | |
1332 | */ | |
1333 | mmap_read_lock(mm); | |
1334 | *unlocked = true; | |
1335 | return 0; | |
1336 | } | |
1337 | ||
4bbd4c77 | 1338 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1339 | int err = vm_fault_to_errno(ret, 0); |
1340 | ||
1341 | if (err) | |
1342 | return err; | |
4bbd4c77 KS |
1343 | BUG(); |
1344 | } | |
4a9e1cda DD |
1345 | |
1346 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1347 | mmap_read_lock(mm); |
475f4dfc PX |
1348 | *unlocked = true; |
1349 | fault_flags |= FAULT_FLAG_TRIED; | |
1350 | goto retry; | |
4a9e1cda DD |
1351 | } |
1352 | ||
4bbd4c77 KS |
1353 | return 0; |
1354 | } | |
add6a0cd | 1355 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1356 | |
2d3a36a4 MH |
1357 | /* |
1358 | * Please note that this function, unlike __get_user_pages will not | |
1359 | * return 0 for nr_pages > 0 without FOLL_NOWAIT | |
1360 | */ | |
64019a2e | 1361 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1362 | unsigned long start, |
1363 | unsigned long nr_pages, | |
f0818f47 AA |
1364 | struct page **pages, |
1365 | struct vm_area_struct **vmas, | |
e716712f | 1366 | int *locked, |
0fd71a56 | 1367 | unsigned int flags) |
f0818f47 | 1368 | { |
f0818f47 AA |
1369 | long ret, pages_done; |
1370 | bool lock_dropped; | |
1371 | ||
1372 | if (locked) { | |
1373 | /* if VM_FAULT_RETRY can be returned, vmas become invalid */ | |
1374 | BUG_ON(vmas); | |
1375 | /* check caller initialized locked */ | |
1376 | BUG_ON(*locked != 1); | |
1377 | } | |
1378 | ||
a458b76a AA |
1379 | if (flags & FOLL_PIN) |
1380 | mm_set_has_pinned_flag(&mm->flags); | |
008cfe44 | 1381 | |
eddb1c22 JH |
1382 | /* |
1383 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1384 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1385 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1386 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1387 | * | |
1388 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1389 | * that here, as any failures will be obvious enough. | |
1390 | */ | |
1391 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1392 | flags |= FOLL_GET; |
f0818f47 AA |
1393 | |
1394 | pages_done = 0; | |
1395 | lock_dropped = false; | |
1396 | for (;;) { | |
64019a2e | 1397 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
f0818f47 AA |
1398 | vmas, locked); |
1399 | if (!locked) | |
1400 | /* VM_FAULT_RETRY couldn't trigger, bypass */ | |
1401 | return ret; | |
1402 | ||
d9272525 | 1403 | /* VM_FAULT_RETRY or VM_FAULT_COMPLETED cannot return errors */ |
f0818f47 AA |
1404 | if (!*locked) { |
1405 | BUG_ON(ret < 0); | |
1406 | BUG_ON(ret >= nr_pages); | |
1407 | } | |
1408 | ||
f0818f47 AA |
1409 | if (ret > 0) { |
1410 | nr_pages -= ret; | |
1411 | pages_done += ret; | |
1412 | if (!nr_pages) | |
1413 | break; | |
1414 | } | |
1415 | if (*locked) { | |
96312e61 AA |
1416 | /* |
1417 | * VM_FAULT_RETRY didn't trigger or it was a | |
1418 | * FOLL_NOWAIT. | |
1419 | */ | |
f0818f47 AA |
1420 | if (!pages_done) |
1421 | pages_done = ret; | |
1422 | break; | |
1423 | } | |
df17277b MR |
1424 | /* |
1425 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1426 | * For the prefault case (!pages) we only update counts. | |
1427 | */ | |
1428 | if (likely(pages)) | |
1429 | pages += ret; | |
f0818f47 | 1430 | start += ret << PAGE_SHIFT; |
4426e945 | 1431 | lock_dropped = true; |
f0818f47 | 1432 | |
4426e945 | 1433 | retry: |
f0818f47 AA |
1434 | /* |
1435 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1436 | * with both FAULT_FLAG_ALLOW_RETRY and |
1437 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
1438 | * by fatal signals, so we need to check it before we | |
1439 | * start trying again otherwise it can loop forever. | |
f0818f47 | 1440 | */ |
4426e945 | 1441 | |
ae46d2aa HD |
1442 | if (fatal_signal_pending(current)) { |
1443 | if (!pages_done) | |
1444 | pages_done = -EINTR; | |
4426e945 | 1445 | break; |
ae46d2aa | 1446 | } |
4426e945 | 1447 | |
d8ed45c5 | 1448 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1449 | if (ret) { |
1450 | BUG_ON(ret > 0); | |
1451 | if (!pages_done) | |
1452 | pages_done = ret; | |
1453 | break; | |
1454 | } | |
4426e945 | 1455 | |
c7b6a566 | 1456 | *locked = 1; |
64019a2e | 1457 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
4426e945 PX |
1458 | pages, NULL, locked); |
1459 | if (!*locked) { | |
1460 | /* Continue to retry until we succeeded */ | |
1461 | BUG_ON(ret != 0); | |
1462 | goto retry; | |
1463 | } | |
f0818f47 AA |
1464 | if (ret != 1) { |
1465 | BUG_ON(ret > 1); | |
1466 | if (!pages_done) | |
1467 | pages_done = ret; | |
1468 | break; | |
1469 | } | |
1470 | nr_pages--; | |
1471 | pages_done++; | |
1472 | if (!nr_pages) | |
1473 | break; | |
df17277b MR |
1474 | if (likely(pages)) |
1475 | pages++; | |
f0818f47 AA |
1476 | start += PAGE_SIZE; |
1477 | } | |
e716712f | 1478 | if (lock_dropped && *locked) { |
f0818f47 AA |
1479 | /* |
1480 | * We must let the caller know we temporarily dropped the lock | |
1481 | * and so the critical section protected by it was lost. | |
1482 | */ | |
d8ed45c5 | 1483 | mmap_read_unlock(mm); |
f0818f47 AA |
1484 | *locked = 0; |
1485 | } | |
1486 | return pages_done; | |
1487 | } | |
1488 | ||
d3649f68 CH |
1489 | /** |
1490 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1491 | * @vma: target vma | |
1492 | * @start: start address | |
1493 | * @end: end address | |
c1e8d7c6 | 1494 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1495 | * |
1496 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1497 | * | |
0a36f7f8 TY |
1498 | * Return either number of pages pinned in the vma, or a negative error |
1499 | * code on error. | |
d3649f68 | 1500 | * |
c1e8d7c6 | 1501 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1502 | * |
4f6da934 | 1503 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1504 | * be unperturbed. |
1505 | * | |
4f6da934 PX |
1506 | * If @locked is non-NULL, it must held for read only and may be |
1507 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1508 | */ |
1509 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1510 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1511 | { |
1512 | struct mm_struct *mm = vma->vm_mm; | |
1513 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1514 | int gup_flags; | |
ece369c7 | 1515 | long ret; |
d3649f68 | 1516 | |
be51eb18 ML |
1517 | VM_BUG_ON(!PAGE_ALIGNED(start)); |
1518 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
d3649f68 CH |
1519 | VM_BUG_ON_VMA(start < vma->vm_start, vma); |
1520 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1521 | mmap_assert_locked(mm); |
d3649f68 | 1522 | |
b67bf49c HD |
1523 | /* |
1524 | * Rightly or wrongly, the VM_LOCKONFAULT case has never used | |
1525 | * faultin_page() to break COW, so it has no work to do here. | |
1526 | */ | |
d3649f68 | 1527 | if (vma->vm_flags & VM_LOCKONFAULT) |
b67bf49c HD |
1528 | return nr_pages; |
1529 | ||
1530 | gup_flags = FOLL_TOUCH; | |
d3649f68 CH |
1531 | /* |
1532 | * We want to touch writable mappings with a write fault in order | |
1533 | * to break COW, except for shared mappings because these don't COW | |
1534 | * and we would not want to dirty them for nothing. | |
1535 | */ | |
1536 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1537 | gup_flags |= FOLL_WRITE; | |
1538 | ||
1539 | /* | |
1540 | * We want mlock to succeed for regions that have any permissions | |
1541 | * other than PROT_NONE. | |
1542 | */ | |
3122e80e | 1543 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1544 | gup_flags |= FOLL_FORCE; |
1545 | ||
1546 | /* | |
1547 | * We made sure addr is within a VMA, so the following will | |
1548 | * not result in a stack expansion that recurses back here. | |
1549 | */ | |
ece369c7 | 1550 | ret = __get_user_pages(mm, start, nr_pages, gup_flags, |
4f6da934 | 1551 | NULL, NULL, locked); |
ece369c7 HD |
1552 | lru_add_drain(); |
1553 | return ret; | |
d3649f68 CH |
1554 | } |
1555 | ||
4ca9b385 DH |
1556 | /* |
1557 | * faultin_vma_page_range() - populate (prefault) page tables inside the | |
1558 | * given VMA range readable/writable | |
1559 | * | |
1560 | * This takes care of mlocking the pages, too, if VM_LOCKED is set. | |
1561 | * | |
1562 | * @vma: target vma | |
1563 | * @start: start address | |
1564 | * @end: end address | |
1565 | * @write: whether to prefault readable or writable | |
1566 | * @locked: whether the mmap_lock is still held | |
1567 | * | |
1568 | * Returns either number of processed pages in the vma, or a negative error | |
1569 | * code on error (see __get_user_pages()). | |
1570 | * | |
1571 | * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and | |
1572 | * covered by the VMA. | |
1573 | * | |
1574 | * If @locked is NULL, it may be held for read or write and will be unperturbed. | |
1575 | * | |
1576 | * If @locked is non-NULL, it must held for read only and may be released. If | |
1577 | * it's released, *@locked will be set to 0. | |
1578 | */ | |
1579 | long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start, | |
1580 | unsigned long end, bool write, int *locked) | |
1581 | { | |
1582 | struct mm_struct *mm = vma->vm_mm; | |
1583 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1584 | int gup_flags; | |
ece369c7 | 1585 | long ret; |
4ca9b385 DH |
1586 | |
1587 | VM_BUG_ON(!PAGE_ALIGNED(start)); | |
1588 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
1589 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1590 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
1591 | mmap_assert_locked(mm); | |
1592 | ||
1593 | /* | |
1594 | * FOLL_TOUCH: Mark page accessed and thereby young; will also mark | |
1595 | * the page dirty with FOLL_WRITE -- which doesn't make a | |
1596 | * difference with !FOLL_FORCE, because the page is writable | |
1597 | * in the page table. | |
1598 | * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit | |
1599 | * a poisoned page. | |
4ca9b385 DH |
1600 | * !FOLL_FORCE: Require proper access permissions. |
1601 | */ | |
b67bf49c | 1602 | gup_flags = FOLL_TOUCH | FOLL_HWPOISON; |
4ca9b385 DH |
1603 | if (write) |
1604 | gup_flags |= FOLL_WRITE; | |
1605 | ||
1606 | /* | |
eb2faa51 DH |
1607 | * We want to report -EINVAL instead of -EFAULT for any permission |
1608 | * problems or incompatible mappings. | |
4ca9b385 | 1609 | */ |
eb2faa51 DH |
1610 | if (check_vma_flags(vma, gup_flags)) |
1611 | return -EINVAL; | |
1612 | ||
ece369c7 | 1613 | ret = __get_user_pages(mm, start, nr_pages, gup_flags, |
4ca9b385 | 1614 | NULL, NULL, locked); |
ece369c7 HD |
1615 | lru_add_drain(); |
1616 | return ret; | |
4ca9b385 DH |
1617 | } |
1618 | ||
d3649f68 CH |
1619 | /* |
1620 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1621 | * | |
1622 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1623 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1624 | * mmap_lock must not be held. |
d3649f68 CH |
1625 | */ |
1626 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1627 | { | |
1628 | struct mm_struct *mm = current->mm; | |
1629 | unsigned long end, nstart, nend; | |
1630 | struct vm_area_struct *vma = NULL; | |
1631 | int locked = 0; | |
1632 | long ret = 0; | |
1633 | ||
1634 | end = start + len; | |
1635 | ||
1636 | for (nstart = start; nstart < end; nstart = nend) { | |
1637 | /* | |
1638 | * We want to fault in pages for [nstart; end) address range. | |
1639 | * Find first corresponding VMA. | |
1640 | */ | |
1641 | if (!locked) { | |
1642 | locked = 1; | |
d8ed45c5 | 1643 | mmap_read_lock(mm); |
d3649f68 CH |
1644 | vma = find_vma(mm, nstart); |
1645 | } else if (nstart >= vma->vm_end) | |
1646 | vma = vma->vm_next; | |
1647 | if (!vma || vma->vm_start >= end) | |
1648 | break; | |
1649 | /* | |
1650 | * Set [nstart; nend) to intersection of desired address | |
1651 | * range with the first VMA. Also, skip undesirable VMA types. | |
1652 | */ | |
1653 | nend = min(end, vma->vm_end); | |
1654 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1655 | continue; | |
1656 | if (nstart < vma->vm_start) | |
1657 | nstart = vma->vm_start; | |
1658 | /* | |
1659 | * Now fault in a range of pages. populate_vma_page_range() | |
1660 | * double checks the vma flags, so that it won't mlock pages | |
1661 | * if the vma was already munlocked. | |
1662 | */ | |
1663 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1664 | if (ret < 0) { | |
1665 | if (ignore_errors) { | |
1666 | ret = 0; | |
1667 | continue; /* continue at next VMA */ | |
1668 | } | |
1669 | break; | |
1670 | } | |
1671 | nend = nstart + ret * PAGE_SIZE; | |
1672 | ret = 0; | |
1673 | } | |
1674 | if (locked) | |
d8ed45c5 | 1675 | mmap_read_unlock(mm); |
d3649f68 CH |
1676 | return ret; /* 0 or negative error code */ |
1677 | } | |
050a9adc | 1678 | #else /* CONFIG_MMU */ |
64019a2e | 1679 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc CH |
1680 | unsigned long nr_pages, struct page **pages, |
1681 | struct vm_area_struct **vmas, int *locked, | |
1682 | unsigned int foll_flags) | |
1683 | { | |
1684 | struct vm_area_struct *vma; | |
1685 | unsigned long vm_flags; | |
24dc20c7 | 1686 | long i; |
050a9adc CH |
1687 | |
1688 | /* calculate required read or write permissions. | |
1689 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1690 | */ | |
1691 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1692 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1693 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1694 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1695 | ||
1696 | for (i = 0; i < nr_pages; i++) { | |
1697 | vma = find_vma(mm, start); | |
1698 | if (!vma) | |
1699 | goto finish_or_fault; | |
1700 | ||
1701 | /* protect what we can, including chardevs */ | |
1702 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1703 | !(vm_flags & vma->vm_flags)) | |
1704 | goto finish_or_fault; | |
1705 | ||
1706 | if (pages) { | |
1707 | pages[i] = virt_to_page(start); | |
1708 | if (pages[i]) | |
1709 | get_page(pages[i]); | |
1710 | } | |
1711 | if (vmas) | |
1712 | vmas[i] = vma; | |
1713 | start = (start + PAGE_SIZE) & PAGE_MASK; | |
1714 | } | |
1715 | ||
1716 | return i; | |
1717 | ||
1718 | finish_or_fault: | |
1719 | return i ? : -EFAULT; | |
1720 | } | |
1721 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1722 | |
bb523b40 AG |
1723 | /** |
1724 | * fault_in_writeable - fault in userspace address range for writing | |
1725 | * @uaddr: start of address range | |
1726 | * @size: size of address range | |
1727 | * | |
1728 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1729 | * copy_from_user()). | |
1730 | */ | |
1731 | size_t fault_in_writeable(char __user *uaddr, size_t size) | |
1732 | { | |
1733 | char __user *start = uaddr, *end; | |
1734 | ||
1735 | if (unlikely(size == 0)) | |
1736 | return 0; | |
677b2a8c CL |
1737 | if (!user_write_access_begin(uaddr, size)) |
1738 | return size; | |
bb523b40 | 1739 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1740 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1741 | uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1742 | } | |
1743 | end = (char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1744 | if (unlikely(end < start)) | |
1745 | end = NULL; | |
1746 | while (uaddr != end) { | |
677b2a8c | 1747 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1748 | uaddr += PAGE_SIZE; |
1749 | } | |
1750 | ||
1751 | out: | |
677b2a8c | 1752 | user_write_access_end(); |
bb523b40 AG |
1753 | if (size > uaddr - start) |
1754 | return size - (uaddr - start); | |
1755 | return 0; | |
1756 | } | |
1757 | EXPORT_SYMBOL(fault_in_writeable); | |
1758 | ||
da32b581 CM |
1759 | /** |
1760 | * fault_in_subpage_writeable - fault in an address range for writing | |
1761 | * @uaddr: start of address range | |
1762 | * @size: size of address range | |
1763 | * | |
1764 | * Fault in a user address range for writing while checking for permissions at | |
1765 | * sub-page granularity (e.g. arm64 MTE). This function should be used when | |
1766 | * the caller cannot guarantee forward progress of a copy_to_user() loop. | |
1767 | * | |
1768 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1769 | * copy_from_user()). | |
1770 | */ | |
1771 | size_t fault_in_subpage_writeable(char __user *uaddr, size_t size) | |
1772 | { | |
1773 | size_t faulted_in; | |
1774 | ||
1775 | /* | |
1776 | * Attempt faulting in at page granularity first for page table | |
1777 | * permission checking. The arch-specific probe_subpage_writeable() | |
1778 | * functions may not check for this. | |
1779 | */ | |
1780 | faulted_in = size - fault_in_writeable(uaddr, size); | |
1781 | if (faulted_in) | |
1782 | faulted_in -= probe_subpage_writeable(uaddr, faulted_in); | |
1783 | ||
1784 | return size - faulted_in; | |
1785 | } | |
1786 | EXPORT_SYMBOL(fault_in_subpage_writeable); | |
1787 | ||
cdd591fc AG |
1788 | /* |
1789 | * fault_in_safe_writeable - fault in an address range for writing | |
1790 | * @uaddr: start of address range | |
1791 | * @size: length of address range | |
1792 | * | |
fe673d3f LT |
1793 | * Faults in an address range for writing. This is primarily useful when we |
1794 | * already know that some or all of the pages in the address range aren't in | |
1795 | * memory. | |
cdd591fc | 1796 | * |
fe673d3f | 1797 | * Unlike fault_in_writeable(), this function is non-destructive. |
cdd591fc AG |
1798 | * |
1799 | * Note that we don't pin or otherwise hold the pages referenced that we fault | |
1800 | * in. There's no guarantee that they'll stay in memory for any duration of | |
1801 | * time. | |
1802 | * | |
1803 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
1804 | * copy_from_user(). | |
1805 | */ | |
1806 | size_t fault_in_safe_writeable(const char __user *uaddr, size_t size) | |
1807 | { | |
fe673d3f | 1808 | unsigned long start = (unsigned long)uaddr, end; |
cdd591fc | 1809 | struct mm_struct *mm = current->mm; |
fe673d3f | 1810 | bool unlocked = false; |
cdd591fc | 1811 | |
fe673d3f LT |
1812 | if (unlikely(size == 0)) |
1813 | return 0; | |
cdd591fc | 1814 | end = PAGE_ALIGN(start + size); |
fe673d3f | 1815 | if (end < start) |
cdd591fc | 1816 | end = 0; |
cdd591fc | 1817 | |
fe673d3f LT |
1818 | mmap_read_lock(mm); |
1819 | do { | |
1820 | if (fixup_user_fault(mm, start, FAULT_FLAG_WRITE, &unlocked)) | |
cdd591fc | 1821 | break; |
fe673d3f LT |
1822 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1823 | } while (start != end); | |
1824 | mmap_read_unlock(mm); | |
1825 | ||
1826 | if (size > (unsigned long)uaddr - start) | |
1827 | return size - ((unsigned long)uaddr - start); | |
1828 | return 0; | |
cdd591fc AG |
1829 | } |
1830 | EXPORT_SYMBOL(fault_in_safe_writeable); | |
1831 | ||
bb523b40 AG |
1832 | /** |
1833 | * fault_in_readable - fault in userspace address range for reading | |
1834 | * @uaddr: start of user address range | |
1835 | * @size: size of user address range | |
1836 | * | |
1837 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1838 | * copy_from_user()). | |
1839 | */ | |
1840 | size_t fault_in_readable(const char __user *uaddr, size_t size) | |
1841 | { | |
1842 | const char __user *start = uaddr, *end; | |
1843 | volatile char c; | |
1844 | ||
1845 | if (unlikely(size == 0)) | |
1846 | return 0; | |
677b2a8c CL |
1847 | if (!user_read_access_begin(uaddr, size)) |
1848 | return size; | |
bb523b40 | 1849 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1850 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1851 | uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1852 | } | |
1853 | end = (const char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1854 | if (unlikely(end < start)) | |
1855 | end = NULL; | |
1856 | while (uaddr != end) { | |
677b2a8c | 1857 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1858 | uaddr += PAGE_SIZE; |
1859 | } | |
1860 | ||
1861 | out: | |
677b2a8c | 1862 | user_read_access_end(); |
bb523b40 AG |
1863 | (void)c; |
1864 | if (size > uaddr - start) | |
1865 | return size - (uaddr - start); | |
1866 | return 0; | |
1867 | } | |
1868 | EXPORT_SYMBOL(fault_in_readable); | |
1869 | ||
8f942eea JH |
1870 | /** |
1871 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1872 | * @addr: user address | |
1873 | * | |
1874 | * Returns struct page pointer of user page pinned for dump, | |
1875 | * to be freed afterwards by put_page(). | |
1876 | * | |
1877 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1878 | * the corefile, to preserve alignment with its headers; and also returns | |
1879 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
f0953a1b | 1880 | * allowing a hole to be left in the corefile to save disk space. |
8f942eea | 1881 | * |
7f3bfab5 | 1882 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1883 | */ |
1884 | #ifdef CONFIG_ELF_CORE | |
1885 | struct page *get_dump_page(unsigned long addr) | |
1886 | { | |
7f3bfab5 | 1887 | struct mm_struct *mm = current->mm; |
8f942eea | 1888 | struct page *page; |
7f3bfab5 JH |
1889 | int locked = 1; |
1890 | int ret; | |
8f942eea | 1891 | |
7f3bfab5 | 1892 | if (mmap_read_lock_killable(mm)) |
8f942eea | 1893 | return NULL; |
7f3bfab5 JH |
1894 | ret = __get_user_pages_locked(mm, addr, 1, &page, NULL, &locked, |
1895 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); | |
1896 | if (locked) | |
1897 | mmap_read_unlock(mm); | |
1898 | return (ret == 1) ? page : NULL; | |
8f942eea JH |
1899 | } |
1900 | #endif /* CONFIG_ELF_CORE */ | |
1901 | ||
d1e153fe | 1902 | #ifdef CONFIG_MIGRATION |
f68749ec PT |
1903 | /* |
1904 | * Check whether all pages are pinnable, if so return number of pages. If some | |
1905 | * pages are not pinnable, migrate them, and unpin all pages. Return zero if | |
1906 | * pages were migrated, or if some pages were not successfully isolated. | |
1907 | * Return negative error if migration fails. | |
1908 | */ | |
1909 | static long check_and_migrate_movable_pages(unsigned long nr_pages, | |
d1e153fe | 1910 | struct page **pages, |
d1e153fe | 1911 | unsigned int gup_flags) |
9a4e9f3b | 1912 | { |
f9f38f78 | 1913 | unsigned long isolation_error_count = 0, i; |
1b7f7e58 | 1914 | struct folio *prev_folio = NULL; |
d1e153fe | 1915 | LIST_HEAD(movable_page_list); |
b05a79d4 | 1916 | bool drain_allow = true, coherent_pages = false; |
f9f38f78 | 1917 | int ret = 0; |
9a4e9f3b | 1918 | |
83c02c23 | 1919 | for (i = 0; i < nr_pages; i++) { |
1b7f7e58 | 1920 | struct folio *folio = page_folio(pages[i]); |
f9f38f78 | 1921 | |
1b7f7e58 | 1922 | if (folio == prev_folio) |
83c02c23 | 1923 | continue; |
1b7f7e58 | 1924 | prev_folio = folio; |
f9f38f78 | 1925 | |
b05a79d4 AP |
1926 | /* |
1927 | * Device coherent pages are managed by a driver and should not | |
1928 | * be pinned indefinitely as it prevents the driver moving the | |
1929 | * page. So when trying to pin with FOLL_LONGTERM instead try | |
1930 | * to migrate the page out of device memory. | |
1931 | */ | |
1932 | if (folio_is_device_coherent(folio)) { | |
1933 | /* | |
1934 | * We always want a new GUP lookup with device coherent | |
1935 | * pages. | |
1936 | */ | |
1937 | pages[i] = 0; | |
1938 | coherent_pages = true; | |
1939 | ||
1940 | /* | |
1941 | * Migration will fail if the page is pinned, so convert | |
1942 | * the pin on the source page to a normal reference. | |
1943 | */ | |
1944 | if (gup_flags & FOLL_PIN) { | |
1945 | get_page(&folio->page); | |
1946 | unpin_user_page(&folio->page); | |
1947 | } | |
1948 | ||
1949 | ret = migrate_device_coherent_page(&folio->page); | |
1950 | if (ret) | |
1951 | goto unpin_pages; | |
1952 | ||
f9f38f78 | 1953 | continue; |
b05a79d4 | 1954 | } |
f9f38f78 | 1955 | |
b05a79d4 AP |
1956 | if (folio_is_longterm_pinnable(folio)) |
1957 | continue; | |
9a4e9f3b | 1958 | /* |
f9f38f78 | 1959 | * Try to move out any movable page before pinning the range. |
9a4e9f3b | 1960 | */ |
1b7f7e58 | 1961 | if (folio_test_hugetlb(folio)) { |
7ce82f4c | 1962 | if (isolate_hugetlb(&folio->page, |
1b7f7e58 | 1963 | &movable_page_list)) |
f9f38f78 CH |
1964 | isolation_error_count++; |
1965 | continue; | |
1966 | } | |
9a4e9f3b | 1967 | |
1b7f7e58 | 1968 | if (!folio_test_lru(folio) && drain_allow) { |
f9f38f78 CH |
1969 | lru_add_drain_all(); |
1970 | drain_allow = false; | |
1971 | } | |
1972 | ||
1b7f7e58 | 1973 | if (folio_isolate_lru(folio)) { |
f9f38f78 CH |
1974 | isolation_error_count++; |
1975 | continue; | |
9a4e9f3b | 1976 | } |
1b7f7e58 MWO |
1977 | list_add_tail(&folio->lru, &movable_page_list); |
1978 | node_stat_mod_folio(folio, | |
1979 | NR_ISOLATED_ANON + folio_is_file_lru(folio), | |
1980 | folio_nr_pages(folio)); | |
9a4e9f3b AK |
1981 | } |
1982 | ||
b05a79d4 AP |
1983 | if (!list_empty(&movable_page_list) || isolation_error_count |
1984 | || coherent_pages) | |
f9f38f78 CH |
1985 | goto unpin_pages; |
1986 | ||
6e7f34eb PT |
1987 | /* |
1988 | * If list is empty, and no isolation errors, means that all pages are | |
1989 | * in the correct zone. | |
1990 | */ | |
f9f38f78 | 1991 | return nr_pages; |
6e7f34eb | 1992 | |
f9f38f78 | 1993 | unpin_pages: |
b05a79d4 AP |
1994 | /* |
1995 | * pages[i] might be NULL if any device coherent pages were found. | |
1996 | */ | |
1997 | for (i = 0; i < nr_pages; i++) { | |
1998 | if (!pages[i]) | |
1999 | continue; | |
2000 | ||
2001 | if (gup_flags & FOLL_PIN) | |
2002 | unpin_user_page(pages[i]); | |
2003 | else | |
f68749ec PT |
2004 | put_page(pages[i]); |
2005 | } | |
f9f38f78 | 2006 | |
d1e153fe | 2007 | if (!list_empty(&movable_page_list)) { |
f9f38f78 CH |
2008 | struct migration_target_control mtc = { |
2009 | .nid = NUMA_NO_NODE, | |
2010 | .gfp_mask = GFP_USER | __GFP_NOWARN, | |
2011 | }; | |
2012 | ||
d1e153fe | 2013 | ret = migrate_pages(&movable_page_list, alloc_migration_target, |
f0f44638 | 2014 | NULL, (unsigned long)&mtc, MIGRATE_SYNC, |
5ac95884 | 2015 | MR_LONGTERM_PIN, NULL); |
f9f38f78 CH |
2016 | if (ret > 0) /* number of pages not migrated */ |
2017 | ret = -ENOMEM; | |
9a4e9f3b AK |
2018 | } |
2019 | ||
f9f38f78 CH |
2020 | if (ret && !list_empty(&movable_page_list)) |
2021 | putback_movable_pages(&movable_page_list); | |
2022 | return ret; | |
9a4e9f3b AK |
2023 | } |
2024 | #else | |
f68749ec | 2025 | static long check_and_migrate_movable_pages(unsigned long nr_pages, |
d1e153fe | 2026 | struct page **pages, |
d1e153fe | 2027 | unsigned int gup_flags) |
9a4e9f3b AK |
2028 | { |
2029 | return nr_pages; | |
2030 | } | |
d1e153fe | 2031 | #endif /* CONFIG_MIGRATION */ |
9a4e9f3b | 2032 | |
2bb6d283 | 2033 | /* |
932f4a63 IW |
2034 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
2035 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 2036 | */ |
64019a2e | 2037 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
2038 | unsigned long start, |
2039 | unsigned long nr_pages, | |
2040 | struct page **pages, | |
2041 | struct vm_area_struct **vmas, | |
2042 | unsigned int gup_flags) | |
2bb6d283 | 2043 | { |
f68749ec | 2044 | unsigned int flags; |
52650c8b | 2045 | long rc; |
2bb6d283 | 2046 | |
f68749ec PT |
2047 | if (!(gup_flags & FOLL_LONGTERM)) |
2048 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
2049 | NULL, gup_flags); | |
2050 | flags = memalloc_pin_save(); | |
2051 | do { | |
2052 | rc = __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
2053 | NULL, gup_flags); | |
2054 | if (rc <= 0) | |
2055 | break; | |
2056 | rc = check_and_migrate_movable_pages(rc, pages, gup_flags); | |
2057 | } while (!rc); | |
2058 | memalloc_pin_restore(flags); | |
2bb6d283 | 2059 | |
2bb6d283 DW |
2060 | return rc; |
2061 | } | |
932f4a63 | 2062 | |
447f3e45 BS |
2063 | static bool is_valid_gup_flags(unsigned int gup_flags) |
2064 | { | |
2065 | /* | |
2066 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
2067 | * never directly by the caller, so enforce that with an assertion: | |
2068 | */ | |
2069 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
2070 | return false; | |
2071 | /* | |
2072 | * FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying | |
2073 | * that is, FOLL_LONGTERM is a specific case, more restrictive case of | |
2074 | * FOLL_PIN. | |
2075 | */ | |
2076 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
2077 | return false; | |
2078 | ||
2079 | return true; | |
2080 | } | |
2081 | ||
22bf29b6 | 2082 | #ifdef CONFIG_MMU |
64019a2e | 2083 | static long __get_user_pages_remote(struct mm_struct *mm, |
22bf29b6 JH |
2084 | unsigned long start, unsigned long nr_pages, |
2085 | unsigned int gup_flags, struct page **pages, | |
2086 | struct vm_area_struct **vmas, int *locked) | |
2087 | { | |
2088 | /* | |
2089 | * Parts of FOLL_LONGTERM behavior are incompatible with | |
2090 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2091 | * vmas. However, this only comes up if locked is set, and there are | |
2092 | * callers that do request FOLL_LONGTERM, but do not set locked. So, | |
2093 | * allow what we can. | |
2094 | */ | |
2095 | if (gup_flags & FOLL_LONGTERM) { | |
2096 | if (WARN_ON_ONCE(locked)) | |
2097 | return -EINVAL; | |
2098 | /* | |
2099 | * This will check the vmas (even if our vmas arg is NULL) | |
2100 | * and return -ENOTSUPP if DAX isn't allowed in this case: | |
2101 | */ | |
64019a2e | 2102 | return __gup_longterm_locked(mm, start, nr_pages, pages, |
22bf29b6 JH |
2103 | vmas, gup_flags | FOLL_TOUCH | |
2104 | FOLL_REMOTE); | |
2105 | } | |
2106 | ||
64019a2e | 2107 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, |
22bf29b6 JH |
2108 | locked, |
2109 | gup_flags | FOLL_TOUCH | FOLL_REMOTE); | |
2110 | } | |
2111 | ||
adc8cb40 | 2112 | /** |
c4237f8b | 2113 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
2114 | * @mm: mm_struct of target mm |
2115 | * @start: starting user address | |
2116 | * @nr_pages: number of pages from start to pin | |
2117 | * @gup_flags: flags modifying lookup behaviour | |
2118 | * @pages: array that receives pointers to the pages pinned. | |
2119 | * Should be at least nr_pages long. Or NULL, if caller | |
2120 | * only intends to ensure the pages are faulted in. | |
2121 | * @vmas: array of pointers to vmas corresponding to each page. | |
2122 | * Or NULL if the caller does not require them. | |
2123 | * @locked: pointer to lock flag indicating whether lock is held and | |
2124 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2125 | * utilised. Lock must initially be held. | |
2126 | * | |
2127 | * Returns either number of pages pinned (which may be less than the | |
2128 | * number requested), or an error. Details about the return value: | |
2129 | * | |
2130 | * -- If nr_pages is 0, returns 0. | |
2131 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
2132 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
2133 | * pages pinned. Again, this may be less than nr_pages. | |
2134 | * | |
2135 | * The caller is responsible for releasing returned @pages, via put_page(). | |
2136 | * | |
c1e8d7c6 | 2137 | * @vmas are valid only as long as mmap_lock is held. |
c4237f8b | 2138 | * |
c1e8d7c6 | 2139 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 2140 | * |
adc8cb40 SJ |
2141 | * get_user_pages_remote walks a process's page tables and takes a reference |
2142 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
2143 | * instant. That is, it takes the page that would be accessed if a user |
2144 | * thread accesses the given user virtual address at that instant. | |
2145 | * | |
2146 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 2147 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b JH |
2148 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
2149 | * and subsequently re faulted). However it does guarantee that the page | |
2150 | * won't be freed completely. And mostly callers simply care that the page | |
2151 | * contains data that was valid *at some point in time*. Typically, an IO | |
2152 | * or similar operation cannot guarantee anything stronger anyway because | |
2153 | * locks can't be held over the syscall boundary. | |
2154 | * | |
2155 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
2156 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
2157 | * be called after the page is finished with, and before put_page is called. | |
2158 | * | |
adc8cb40 SJ |
2159 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
2160 | * to get a handle on the memory by some means other than accesses | |
2161 | * via the user virtual addresses. The pages may be submitted for | |
2162 | * DMA to devices or accessed via their kernel linear mapping (via the | |
2163 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
2164 | * |
2165 | * See also get_user_pages_fast, for performance critical applications. | |
2166 | * | |
adc8cb40 | 2167 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 2168 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 2169 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
2170 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
2171 | */ | |
64019a2e | 2172 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
2173 | unsigned long start, unsigned long nr_pages, |
2174 | unsigned int gup_flags, struct page **pages, | |
2175 | struct vm_area_struct **vmas, int *locked) | |
2176 | { | |
447f3e45 | 2177 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2178 | return -EINVAL; |
2179 | ||
64019a2e | 2180 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
22bf29b6 | 2181 | pages, vmas, locked); |
c4237f8b JH |
2182 | } |
2183 | EXPORT_SYMBOL(get_user_pages_remote); | |
2184 | ||
eddb1c22 | 2185 | #else /* CONFIG_MMU */ |
64019a2e | 2186 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2187 | unsigned long start, unsigned long nr_pages, |
2188 | unsigned int gup_flags, struct page **pages, | |
2189 | struct vm_area_struct **vmas, int *locked) | |
2190 | { | |
2191 | return 0; | |
2192 | } | |
3faa52c0 | 2193 | |
64019a2e | 2194 | static long __get_user_pages_remote(struct mm_struct *mm, |
3faa52c0 JH |
2195 | unsigned long start, unsigned long nr_pages, |
2196 | unsigned int gup_flags, struct page **pages, | |
2197 | struct vm_area_struct **vmas, int *locked) | |
2198 | { | |
2199 | return 0; | |
2200 | } | |
eddb1c22 JH |
2201 | #endif /* !CONFIG_MMU */ |
2202 | ||
adc8cb40 SJ |
2203 | /** |
2204 | * get_user_pages() - pin user pages in memory | |
2205 | * @start: starting user address | |
2206 | * @nr_pages: number of pages from start to pin | |
2207 | * @gup_flags: flags modifying lookup behaviour | |
2208 | * @pages: array that receives pointers to the pages pinned. | |
2209 | * Should be at least nr_pages long. Or NULL, if caller | |
2210 | * only intends to ensure the pages are faulted in. | |
2211 | * @vmas: array of pointers to vmas corresponding to each page. | |
2212 | * Or NULL if the caller does not require them. | |
2213 | * | |
64019a2e PX |
2214 | * This is the same as get_user_pages_remote(), just with a less-flexible |
2215 | * calling convention where we assume that the mm being operated on belongs to | |
2216 | * the current task, and doesn't allow passing of a locked parameter. We also | |
2217 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
2218 | */ |
2219 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
2220 | unsigned int gup_flags, struct page **pages, | |
2221 | struct vm_area_struct **vmas) | |
2222 | { | |
447f3e45 | 2223 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2224 | return -EINVAL; |
2225 | ||
64019a2e | 2226 | return __gup_longterm_locked(current->mm, start, nr_pages, |
932f4a63 IW |
2227 | pages, vmas, gup_flags | FOLL_TOUCH); |
2228 | } | |
2229 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 2230 | |
acc3c8d1 | 2231 | /* |
d3649f68 | 2232 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 2233 | * |
3e4e28c5 | 2234 | * mmap_read_lock(mm); |
64019a2e | 2235 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2236 | * mmap_read_unlock(mm); |
d3649f68 CH |
2237 | * |
2238 | * with: | |
2239 | * | |
64019a2e | 2240 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
2241 | * |
2242 | * It is functionally equivalent to get_user_pages_fast so | |
2243 | * get_user_pages_fast should be used instead if specific gup_flags | |
2244 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 2245 | */ |
d3649f68 CH |
2246 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
2247 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 KS |
2248 | { |
2249 | struct mm_struct *mm = current->mm; | |
d3649f68 CH |
2250 | int locked = 1; |
2251 | long ret; | |
acc3c8d1 | 2252 | |
d3649f68 CH |
2253 | /* |
2254 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
2255 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2256 | * vmas. As there are no users of this flag in this call we simply | |
2257 | * disallow this option for now. | |
2258 | */ | |
2259 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
2260 | return -EINVAL; | |
acc3c8d1 | 2261 | |
d8ed45c5 | 2262 | mmap_read_lock(mm); |
64019a2e | 2263 | ret = __get_user_pages_locked(mm, start, nr_pages, pages, NULL, |
d3649f68 | 2264 | &locked, gup_flags | FOLL_TOUCH); |
acc3c8d1 | 2265 | if (locked) |
d8ed45c5 | 2266 | mmap_read_unlock(mm); |
d3649f68 | 2267 | return ret; |
4bbd4c77 | 2268 | } |
d3649f68 | 2269 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
2270 | |
2271 | /* | |
67a929e0 | 2272 | * Fast GUP |
2667f50e SC |
2273 | * |
2274 | * get_user_pages_fast attempts to pin user pages by walking the page | |
2275 | * tables directly and avoids taking locks. Thus the walker needs to be | |
2276 | * protected from page table pages being freed from under it, and should | |
2277 | * block any THP splits. | |
2278 | * | |
2279 | * One way to achieve this is to have the walker disable interrupts, and | |
2280 | * rely on IPIs from the TLB flushing code blocking before the page table | |
2281 | * pages are freed. This is unsuitable for architectures that do not need | |
2282 | * to broadcast an IPI when invalidating TLBs. | |
2283 | * | |
2284 | * Another way to achieve this is to batch up page table containing pages | |
2285 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
2286 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
2287 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
2288 | * (which is a relatively rare event). The code below adopts this strategy. | |
2289 | * | |
2290 | * Before activating this code, please be aware that the following assumptions | |
2291 | * are currently made: | |
2292 | * | |
ff2e6d72 | 2293 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 2294 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 2295 | * |
2667f50e SC |
2296 | * *) ptes can be read atomically by the architecture. |
2297 | * | |
2298 | * *) access_ok is sufficient to validate userspace address ranges. | |
2299 | * | |
2300 | * The last two assumptions can be relaxed by the addition of helper functions. | |
2301 | * | |
2302 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
2303 | */ | |
67a929e0 | 2304 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 2305 | |
790c7369 | 2306 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 2307 | unsigned int flags, |
790c7369 | 2308 | struct page **pages) |
b59f65fa KS |
2309 | { |
2310 | while ((*nr) - nr_start) { | |
2311 | struct page *page = pages[--(*nr)]; | |
2312 | ||
2313 | ClearPageReferenced(page); | |
3faa52c0 JH |
2314 | if (flags & FOLL_PIN) |
2315 | unpin_user_page(page); | |
2316 | else | |
2317 | put_page(page); | |
b59f65fa KS |
2318 | } |
2319 | } | |
2320 | ||
3010a5ea | 2321 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
2667f50e | 2322 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, |
b798bec4 | 2323 | unsigned int flags, struct page **pages, int *nr) |
2667f50e | 2324 | { |
b59f65fa KS |
2325 | struct dev_pagemap *pgmap = NULL; |
2326 | int nr_start = *nr, ret = 0; | |
2667f50e | 2327 | pte_t *ptep, *ptem; |
2667f50e SC |
2328 | |
2329 | ptem = ptep = pte_offset_map(&pmd, addr); | |
2330 | do { | |
2a4a06da | 2331 | pte_t pte = ptep_get_lockless(ptep); |
b0496fe4 MWO |
2332 | struct page *page; |
2333 | struct folio *folio; | |
2667f50e SC |
2334 | |
2335 | /* | |
2336 | * Similar to the PMD case below, NUMA hinting must take slow | |
8a0516ed | 2337 | * path using the pte_protnone check. |
2667f50e | 2338 | */ |
e7884f8e KS |
2339 | if (pte_protnone(pte)) |
2340 | goto pte_unmap; | |
2341 | ||
b798bec4 | 2342 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2343 | goto pte_unmap; |
2344 | ||
b59f65fa | 2345 | if (pte_devmap(pte)) { |
7af75561 IW |
2346 | if (unlikely(flags & FOLL_LONGTERM)) |
2347 | goto pte_unmap; | |
2348 | ||
b59f65fa KS |
2349 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2350 | if (unlikely(!pgmap)) { | |
3b78d834 | 2351 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2352 | goto pte_unmap; |
2353 | } | |
2354 | } else if (pte_special(pte)) | |
2667f50e SC |
2355 | goto pte_unmap; |
2356 | ||
2357 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2358 | page = pte_page(pte); | |
2359 | ||
b0496fe4 MWO |
2360 | folio = try_grab_folio(page, 1, flags); |
2361 | if (!folio) | |
2667f50e SC |
2362 | goto pte_unmap; |
2363 | ||
1507f512 | 2364 | if (unlikely(page_is_secretmem(page))) { |
b0496fe4 | 2365 | gup_put_folio(folio, 1, flags); |
1507f512 MR |
2366 | goto pte_unmap; |
2367 | } | |
2368 | ||
2667f50e | 2369 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { |
b0496fe4 | 2370 | gup_put_folio(folio, 1, flags); |
2667f50e SC |
2371 | goto pte_unmap; |
2372 | } | |
2373 | ||
a7f22660 DH |
2374 | if (!pte_write(pte) && gup_must_unshare(flags, page)) { |
2375 | gup_put_folio(folio, 1, flags); | |
2376 | goto pte_unmap; | |
2377 | } | |
2378 | ||
f28d4363 CI |
2379 | /* |
2380 | * We need to make the page accessible if and only if we are | |
2381 | * going to access its content (the FOLL_PIN case). Please | |
2382 | * see Documentation/core-api/pin_user_pages.rst for | |
2383 | * details. | |
2384 | */ | |
2385 | if (flags & FOLL_PIN) { | |
2386 | ret = arch_make_page_accessible(page); | |
2387 | if (ret) { | |
b0496fe4 | 2388 | gup_put_folio(folio, 1, flags); |
f28d4363 CI |
2389 | goto pte_unmap; |
2390 | } | |
2391 | } | |
b0496fe4 | 2392 | folio_set_referenced(folio); |
2667f50e SC |
2393 | pages[*nr] = page; |
2394 | (*nr)++; | |
2667f50e SC |
2395 | } while (ptep++, addr += PAGE_SIZE, addr != end); |
2396 | ||
2397 | ret = 1; | |
2398 | ||
2399 | pte_unmap: | |
832d7aa0 CH |
2400 | if (pgmap) |
2401 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2402 | pte_unmap(ptem); |
2403 | return ret; | |
2404 | } | |
2405 | #else | |
2406 | ||
2407 | /* | |
2408 | * If we can't determine whether or not a pte is special, then fail immediately | |
2409 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2410 | * to be special. | |
2411 | * | |
2412 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2413 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2414 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2415 | */ | |
2416 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, | |
b798bec4 | 2417 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2418 | { |
2419 | return 0; | |
2420 | } | |
3010a5ea | 2421 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2422 | |
17596731 | 2423 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2424 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2425 | unsigned long end, unsigned int flags, |
2426 | struct page **pages, int *nr) | |
b59f65fa KS |
2427 | { |
2428 | int nr_start = *nr; | |
2429 | struct dev_pagemap *pgmap = NULL; | |
2430 | ||
2431 | do { | |
2432 | struct page *page = pfn_to_page(pfn); | |
2433 | ||
2434 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2435 | if (unlikely(!pgmap)) { | |
3b78d834 | 2436 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2437 | break; |
b59f65fa KS |
2438 | } |
2439 | SetPageReferenced(page); | |
2440 | pages[*nr] = page; | |
3faa52c0 JH |
2441 | if (unlikely(!try_grab_page(page, flags))) { |
2442 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
6401c4eb | 2443 | break; |
3faa52c0 | 2444 | } |
b59f65fa KS |
2445 | (*nr)++; |
2446 | pfn++; | |
2447 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 | 2448 | |
6401c4eb | 2449 | put_dev_pagemap(pgmap); |
20b7fee7 | 2450 | return addr == end; |
b59f65fa KS |
2451 | } |
2452 | ||
a9b6de77 | 2453 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2454 | unsigned long end, unsigned int flags, |
2455 | struct page **pages, int *nr) | |
b59f65fa KS |
2456 | { |
2457 | unsigned long fault_pfn; | |
a9b6de77 DW |
2458 | int nr_start = *nr; |
2459 | ||
2460 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2461 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2462 | return 0; |
b59f65fa | 2463 | |
a9b6de77 | 2464 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2465 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2466 | return 0; |
2467 | } | |
2468 | return 1; | |
b59f65fa KS |
2469 | } |
2470 | ||
a9b6de77 | 2471 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2472 | unsigned long end, unsigned int flags, |
2473 | struct page **pages, int *nr) | |
b59f65fa KS |
2474 | { |
2475 | unsigned long fault_pfn; | |
a9b6de77 DW |
2476 | int nr_start = *nr; |
2477 | ||
2478 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2479 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2480 | return 0; |
b59f65fa | 2481 | |
a9b6de77 | 2482 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2483 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2484 | return 0; |
2485 | } | |
2486 | return 1; | |
b59f65fa KS |
2487 | } |
2488 | #else | |
a9b6de77 | 2489 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2490 | unsigned long end, unsigned int flags, |
2491 | struct page **pages, int *nr) | |
b59f65fa KS |
2492 | { |
2493 | BUILD_BUG(); | |
2494 | return 0; | |
2495 | } | |
2496 | ||
a9b6de77 | 2497 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2498 | unsigned long end, unsigned int flags, |
2499 | struct page **pages, int *nr) | |
b59f65fa KS |
2500 | { |
2501 | BUILD_BUG(); | |
2502 | return 0; | |
2503 | } | |
2504 | #endif | |
2505 | ||
a43e9820 JH |
2506 | static int record_subpages(struct page *page, unsigned long addr, |
2507 | unsigned long end, struct page **pages) | |
2508 | { | |
2509 | int nr; | |
2510 | ||
c228afb1 MWO |
2511 | for (nr = 0; addr != end; nr++, addr += PAGE_SIZE) |
2512 | pages[nr] = nth_page(page, nr); | |
a43e9820 JH |
2513 | |
2514 | return nr; | |
2515 | } | |
2516 | ||
cbd34da7 CH |
2517 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2518 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2519 | unsigned long sz) | |
2520 | { | |
2521 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2522 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2523 | } | |
2524 | ||
2525 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2526 | unsigned long end, unsigned int flags, |
2527 | struct page **pages, int *nr) | |
cbd34da7 CH |
2528 | { |
2529 | unsigned long pte_end; | |
09a1626e MWO |
2530 | struct page *page; |
2531 | struct folio *folio; | |
cbd34da7 CH |
2532 | pte_t pte; |
2533 | int refs; | |
2534 | ||
2535 | pte_end = (addr + sz) & ~(sz-1); | |
2536 | if (pte_end < end) | |
2537 | end = pte_end; | |
2538 | ||
55ca2263 | 2539 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2540 | |
0cd22afd | 2541 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2542 | return 0; |
2543 | ||
2544 | /* hugepages are never "special" */ | |
2545 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2546 | ||
09a1626e | 2547 | page = nth_page(pte_page(pte), (addr & (sz - 1)) >> PAGE_SHIFT); |
a43e9820 | 2548 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2549 | |
09a1626e MWO |
2550 | folio = try_grab_folio(page, refs, flags); |
2551 | if (!folio) | |
cbd34da7 | 2552 | return 0; |
cbd34da7 CH |
2553 | |
2554 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
09a1626e | 2555 | gup_put_folio(folio, refs, flags); |
cbd34da7 CH |
2556 | return 0; |
2557 | } | |
2558 | ||
a7f22660 DH |
2559 | if (!pte_write(pte) && gup_must_unshare(flags, &folio->page)) { |
2560 | gup_put_folio(folio, refs, flags); | |
2561 | return 0; | |
2562 | } | |
2563 | ||
a43e9820 | 2564 | *nr += refs; |
09a1626e | 2565 | folio_set_referenced(folio); |
cbd34da7 CH |
2566 | return 1; |
2567 | } | |
2568 | ||
2569 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2570 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2571 | struct page **pages, int *nr) |
2572 | { | |
2573 | pte_t *ptep; | |
2574 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2575 | unsigned long next; | |
2576 | ||
2577 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2578 | do { | |
2579 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2580 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2581 | return 0; |
2582 | } while (ptep++, addr = next, addr != end); | |
2583 | ||
2584 | return 1; | |
2585 | } | |
2586 | #else | |
2587 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2588 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2589 | struct page **pages, int *nr) |
2590 | { | |
2591 | return 0; | |
2592 | } | |
2593 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2594 | ||
2667f50e | 2595 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2596 | unsigned long end, unsigned int flags, |
2597 | struct page **pages, int *nr) | |
2667f50e | 2598 | { |
667ed1f7 MWO |
2599 | struct page *page; |
2600 | struct folio *folio; | |
2667f50e SC |
2601 | int refs; |
2602 | ||
b798bec4 | 2603 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2604 | return 0; |
2605 | ||
7af75561 IW |
2606 | if (pmd_devmap(orig)) { |
2607 | if (unlikely(flags & FOLL_LONGTERM)) | |
2608 | return 0; | |
86dfbed4 JH |
2609 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2610 | pages, nr); | |
7af75561 | 2611 | } |
b59f65fa | 2612 | |
c228afb1 | 2613 | page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2614 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2615 | |
667ed1f7 MWO |
2616 | folio = try_grab_folio(page, refs, flags); |
2617 | if (!folio) | |
2667f50e | 2618 | return 0; |
2667f50e SC |
2619 | |
2620 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
667ed1f7 | 2621 | gup_put_folio(folio, refs, flags); |
2667f50e SC |
2622 | return 0; |
2623 | } | |
2624 | ||
a7f22660 DH |
2625 | if (!pmd_write(orig) && gup_must_unshare(flags, &folio->page)) { |
2626 | gup_put_folio(folio, refs, flags); | |
2627 | return 0; | |
2628 | } | |
2629 | ||
a43e9820 | 2630 | *nr += refs; |
667ed1f7 | 2631 | folio_set_referenced(folio); |
2667f50e SC |
2632 | return 1; |
2633 | } | |
2634 | ||
2635 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2636 | unsigned long end, unsigned int flags, |
2637 | struct page **pages, int *nr) | |
2667f50e | 2638 | { |
83afb52e MWO |
2639 | struct page *page; |
2640 | struct folio *folio; | |
2667f50e SC |
2641 | int refs; |
2642 | ||
b798bec4 | 2643 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2644 | return 0; |
2645 | ||
7af75561 IW |
2646 | if (pud_devmap(orig)) { |
2647 | if (unlikely(flags & FOLL_LONGTERM)) | |
2648 | return 0; | |
86dfbed4 JH |
2649 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2650 | pages, nr); | |
7af75561 | 2651 | } |
b59f65fa | 2652 | |
c228afb1 | 2653 | page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2654 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2655 | |
83afb52e MWO |
2656 | folio = try_grab_folio(page, refs, flags); |
2657 | if (!folio) | |
2667f50e | 2658 | return 0; |
2667f50e SC |
2659 | |
2660 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
83afb52e | 2661 | gup_put_folio(folio, refs, flags); |
2667f50e SC |
2662 | return 0; |
2663 | } | |
2664 | ||
a7f22660 DH |
2665 | if (!pud_write(orig) && gup_must_unshare(flags, &folio->page)) { |
2666 | gup_put_folio(folio, refs, flags); | |
2667 | return 0; | |
2668 | } | |
2669 | ||
a43e9820 | 2670 | *nr += refs; |
83afb52e | 2671 | folio_set_referenced(folio); |
2667f50e SC |
2672 | return 1; |
2673 | } | |
2674 | ||
f30c59e9 | 2675 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2676 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2677 | struct page **pages, int *nr) |
2678 | { | |
2679 | int refs; | |
2d7919a2 MWO |
2680 | struct page *page; |
2681 | struct folio *folio; | |
f30c59e9 | 2682 | |
b798bec4 | 2683 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2684 | return 0; |
2685 | ||
b59f65fa | 2686 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2687 | |
c228afb1 | 2688 | page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2689 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2690 | |
2d7919a2 MWO |
2691 | folio = try_grab_folio(page, refs, flags); |
2692 | if (!folio) | |
f30c59e9 | 2693 | return 0; |
f30c59e9 AK |
2694 | |
2695 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
2d7919a2 | 2696 | gup_put_folio(folio, refs, flags); |
f30c59e9 AK |
2697 | return 0; |
2698 | } | |
2699 | ||
a43e9820 | 2700 | *nr += refs; |
2d7919a2 | 2701 | folio_set_referenced(folio); |
f30c59e9 AK |
2702 | return 1; |
2703 | } | |
2704 | ||
d3f7b1bb | 2705 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2706 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2707 | { |
2708 | unsigned long next; | |
2709 | pmd_t *pmdp; | |
2710 | ||
d3f7b1bb | 2711 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2712 | do { |
38c5ce93 | 2713 | pmd_t pmd = READ_ONCE(*pmdp); |
2667f50e SC |
2714 | |
2715 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2716 | if (!pmd_present(pmd)) |
2667f50e SC |
2717 | return 0; |
2718 | ||
414fd080 YZ |
2719 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2720 | pmd_devmap(pmd))) { | |
2667f50e SC |
2721 | /* |
2722 | * NUMA hinting faults need to be handled in the GUP | |
2723 | * slowpath for accounting purposes and so that they | |
2724 | * can be serialised against THP migration. | |
2725 | */ | |
8a0516ed | 2726 | if (pmd_protnone(pmd)) |
2667f50e SC |
2727 | return 0; |
2728 | ||
b798bec4 | 2729 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2730 | pages, nr)) |
2731 | return 0; | |
2732 | ||
f30c59e9 AK |
2733 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2734 | /* | |
2735 | * architecture have different format for hugetlbfs | |
2736 | * pmd format and THP pmd format | |
2737 | */ | |
2738 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2739 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2740 | return 0; |
b798bec4 | 2741 | } else if (!gup_pte_range(pmd, addr, next, flags, pages, nr)) |
2923117b | 2742 | return 0; |
2667f50e SC |
2743 | } while (pmdp++, addr = next, addr != end); |
2744 | ||
2745 | return 1; | |
2746 | } | |
2747 | ||
d3f7b1bb | 2748 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2749 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2750 | { |
2751 | unsigned long next; | |
2752 | pud_t *pudp; | |
2753 | ||
d3f7b1bb | 2754 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2755 | do { |
e37c6982 | 2756 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2757 | |
2758 | next = pud_addr_end(addr, end); | |
15494520 | 2759 | if (unlikely(!pud_present(pud))) |
2667f50e | 2760 | return 0; |
f30c59e9 | 2761 | if (unlikely(pud_huge(pud))) { |
b798bec4 | 2762 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2763 | pages, nr)) |
2764 | return 0; | |
2765 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2766 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2767 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2768 | return 0; |
d3f7b1bb | 2769 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2770 | return 0; |
2771 | } while (pudp++, addr = next, addr != end); | |
2772 | ||
2773 | return 1; | |
2774 | } | |
2775 | ||
d3f7b1bb | 2776 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2777 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2778 | { |
2779 | unsigned long next; | |
2780 | p4d_t *p4dp; | |
2781 | ||
d3f7b1bb | 2782 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
2783 | do { |
2784 | p4d_t p4d = READ_ONCE(*p4dp); | |
2785 | ||
2786 | next = p4d_addr_end(addr, end); | |
2787 | if (p4d_none(p4d)) | |
2788 | return 0; | |
2789 | BUILD_BUG_ON(p4d_huge(p4d)); | |
2790 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
2791 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 2792 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 2793 | return 0; |
d3f7b1bb | 2794 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
2795 | return 0; |
2796 | } while (p4dp++, addr = next, addr != end); | |
2797 | ||
2798 | return 1; | |
2799 | } | |
2800 | ||
5b65c467 | 2801 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 2802 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
2803 | { |
2804 | unsigned long next; | |
2805 | pgd_t *pgdp; | |
2806 | ||
2807 | pgdp = pgd_offset(current->mm, addr); | |
2808 | do { | |
2809 | pgd_t pgd = READ_ONCE(*pgdp); | |
2810 | ||
2811 | next = pgd_addr_end(addr, end); | |
2812 | if (pgd_none(pgd)) | |
2813 | return; | |
2814 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 2815 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
2816 | pages, nr)) |
2817 | return; | |
2818 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
2819 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 2820 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 2821 | return; |
d3f7b1bb | 2822 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
2823 | return; |
2824 | } while (pgdp++, addr = next, addr != end); | |
2825 | } | |
050a9adc CH |
2826 | #else |
2827 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
2828 | unsigned int flags, struct page **pages, int *nr) | |
2829 | { | |
2830 | } | |
2831 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
2832 | |
2833 | #ifndef gup_fast_permitted | |
2834 | /* | |
dadbb612 | 2835 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
2836 | * we need to fall back to the slow version: |
2837 | */ | |
26f4c328 | 2838 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 2839 | { |
26f4c328 | 2840 | return true; |
5b65c467 KS |
2841 | } |
2842 | #endif | |
2843 | ||
7af75561 IW |
2844 | static int __gup_longterm_unlocked(unsigned long start, int nr_pages, |
2845 | unsigned int gup_flags, struct page **pages) | |
2846 | { | |
2847 | int ret; | |
2848 | ||
2849 | /* | |
2850 | * FIXME: FOLL_LONGTERM does not work with | |
2851 | * get_user_pages_unlocked() (see comments in that function) | |
2852 | */ | |
2853 | if (gup_flags & FOLL_LONGTERM) { | |
d8ed45c5 | 2854 | mmap_read_lock(current->mm); |
64019a2e | 2855 | ret = __gup_longterm_locked(current->mm, |
7af75561 IW |
2856 | start, nr_pages, |
2857 | pages, NULL, gup_flags); | |
d8ed45c5 | 2858 | mmap_read_unlock(current->mm); |
7af75561 IW |
2859 | } else { |
2860 | ret = get_user_pages_unlocked(start, nr_pages, | |
2861 | pages, gup_flags); | |
2862 | } | |
2863 | ||
2864 | return ret; | |
2865 | } | |
2866 | ||
c28b1fc7 JG |
2867 | static unsigned long lockless_pages_from_mm(unsigned long start, |
2868 | unsigned long end, | |
2869 | unsigned int gup_flags, | |
2870 | struct page **pages) | |
2871 | { | |
2872 | unsigned long flags; | |
2873 | int nr_pinned = 0; | |
57efa1fe | 2874 | unsigned seq; |
c28b1fc7 JG |
2875 | |
2876 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
2877 | !gup_fast_permitted(start, end)) | |
2878 | return 0; | |
2879 | ||
57efa1fe JG |
2880 | if (gup_flags & FOLL_PIN) { |
2881 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
2882 | if (seq & 1) | |
2883 | return 0; | |
2884 | } | |
2885 | ||
c28b1fc7 JG |
2886 | /* |
2887 | * Disable interrupts. The nested form is used, in order to allow full, | |
2888 | * general purpose use of this routine. | |
2889 | * | |
2890 | * With interrupts disabled, we block page table pages from being freed | |
2891 | * from under us. See struct mmu_table_batch comments in | |
2892 | * include/asm-generic/tlb.h for more details. | |
2893 | * | |
2894 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
2895 | * that come from THPs splitting. | |
2896 | */ | |
2897 | local_irq_save(flags); | |
2898 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
2899 | local_irq_restore(flags); | |
57efa1fe JG |
2900 | |
2901 | /* | |
2902 | * When pinning pages for DMA there could be a concurrent write protect | |
2903 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
2904 | */ | |
2905 | if (gup_flags & FOLL_PIN) { | |
2906 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
b6a2619c | 2907 | unpin_user_pages_lockless(pages, nr_pinned); |
57efa1fe | 2908 | return 0; |
b6a2619c DH |
2909 | } else { |
2910 | sanity_check_pinned_pages(pages, nr_pinned); | |
57efa1fe JG |
2911 | } |
2912 | } | |
c28b1fc7 JG |
2913 | return nr_pinned; |
2914 | } | |
2915 | ||
2916 | static int internal_get_user_pages_fast(unsigned long start, | |
2917 | unsigned long nr_pages, | |
eddb1c22 JH |
2918 | unsigned int gup_flags, |
2919 | struct page **pages) | |
2667f50e | 2920 | { |
c28b1fc7 JG |
2921 | unsigned long len, end; |
2922 | unsigned long nr_pinned; | |
2923 | int ret; | |
2667f50e | 2924 | |
f4000fdf | 2925 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef | 2926 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
55b8fe70 | 2927 | FOLL_FAST_ONLY | FOLL_NOFAULT))) |
817be129 CH |
2928 | return -EINVAL; |
2929 | ||
a458b76a AA |
2930 | if (gup_flags & FOLL_PIN) |
2931 | mm_set_has_pinned_flag(¤t->mm->flags); | |
008cfe44 | 2932 | |
f81cd178 | 2933 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 2934 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 2935 | |
f455c854 | 2936 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
2937 | len = nr_pages << PAGE_SHIFT; |
2938 | if (check_add_overflow(start, len, &end)) | |
c61611f7 | 2939 | return 0; |
96d4f267 | 2940 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 2941 | return -EFAULT; |
73e10a61 | 2942 | |
c28b1fc7 JG |
2943 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
2944 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
2945 | return nr_pinned; | |
2667f50e | 2946 | |
c28b1fc7 JG |
2947 | /* Slow path: try to get the remaining pages with get_user_pages */ |
2948 | start += nr_pinned << PAGE_SHIFT; | |
2949 | pages += nr_pinned; | |
2950 | ret = __gup_longterm_unlocked(start, nr_pages - nr_pinned, gup_flags, | |
2951 | pages); | |
2952 | if (ret < 0) { | |
2953 | /* | |
2954 | * The caller has to unpin the pages we already pinned so | |
2955 | * returning -errno is not an option | |
2956 | */ | |
2957 | if (nr_pinned) | |
2958 | return nr_pinned; | |
2959 | return ret; | |
2667f50e | 2960 | } |
c28b1fc7 | 2961 | return ret + nr_pinned; |
2667f50e | 2962 | } |
c28b1fc7 | 2963 | |
dadbb612 SJ |
2964 | /** |
2965 | * get_user_pages_fast_only() - pin user pages in memory | |
2966 | * @start: starting user address | |
2967 | * @nr_pages: number of pages from start to pin | |
2968 | * @gup_flags: flags modifying pin behaviour | |
2969 | * @pages: array that receives pointers to the pages pinned. | |
2970 | * Should be at least nr_pages long. | |
2971 | * | |
9e1f0580 JH |
2972 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
2973 | * the regular GUP. | |
2974 | * Note a difference with get_user_pages_fast: this always returns the | |
2975 | * number of pages pinned, 0 if no pages were pinned. | |
2976 | * | |
2977 | * If the architecture does not support this function, simply return with no | |
2978 | * pages pinned. | |
2979 | * | |
2980 | * Careful, careful! COW breaking can go either way, so a non-write | |
2981 | * access can get ambiguous page results. If you call this function without | |
2982 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
2983 | */ | |
dadbb612 SJ |
2984 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
2985 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 2986 | { |
376a34ef | 2987 | int nr_pinned; |
9e1f0580 JH |
2988 | /* |
2989 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
2990 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
2991 | * |
2992 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2993 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 2994 | */ |
dadbb612 | 2995 | gup_flags |= FOLL_GET | FOLL_FAST_ONLY; |
9e1f0580 | 2996 | |
376a34ef JH |
2997 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
2998 | pages); | |
9e1f0580 JH |
2999 | |
3000 | /* | |
376a34ef JH |
3001 | * As specified in the API description above, this routine is not |
3002 | * allowed to return negative values. However, the common core | |
3003 | * routine internal_get_user_pages_fast() *can* return -errno. | |
3004 | * Therefore, correct for that here: | |
9e1f0580 | 3005 | */ |
376a34ef JH |
3006 | if (nr_pinned < 0) |
3007 | nr_pinned = 0; | |
9e1f0580 JH |
3008 | |
3009 | return nr_pinned; | |
3010 | } | |
dadbb612 | 3011 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 3012 | |
eddb1c22 JH |
3013 | /** |
3014 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
3015 | * @start: starting user address |
3016 | * @nr_pages: number of pages from start to pin | |
3017 | * @gup_flags: flags modifying pin behaviour | |
3018 | * @pages: array that receives pointers to the pages pinned. | |
3019 | * Should be at least nr_pages long. | |
eddb1c22 | 3020 | * |
c1e8d7c6 | 3021 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
3022 | * If not successful, it will fall back to taking the lock and |
3023 | * calling get_user_pages(). | |
3024 | * | |
3025 | * Returns number of pages pinned. This may be fewer than the number requested. | |
3026 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
3027 | * -errno. | |
3028 | */ | |
3029 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
3030 | unsigned int gup_flags, struct page **pages) | |
3031 | { | |
447f3e45 | 3032 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
3033 | return -EINVAL; |
3034 | ||
94202f12 JH |
3035 | /* |
3036 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
3037 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
3038 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
3039 | * request. | |
3040 | */ | |
3041 | gup_flags |= FOLL_GET; | |
eddb1c22 JH |
3042 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
3043 | } | |
050a9adc | 3044 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
3045 | |
3046 | /** | |
3047 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
3048 | * | |
3faa52c0 JH |
3049 | * @start: starting user address |
3050 | * @nr_pages: number of pages from start to pin | |
3051 | * @gup_flags: flags modifying pin behaviour | |
3052 | * @pages: array that receives pointers to the pages pinned. | |
3053 | * Should be at least nr_pages long. | |
3054 | * | |
3055 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
3056 | * get_user_pages_fast() for documentation on the function arguments, because | |
3057 | * the arguments here are identical. | |
3058 | * | |
3059 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3060 | * see Documentation/core-api/pin_user_pages.rst for further details. |
eddb1c22 JH |
3061 | */ |
3062 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
3063 | unsigned int gup_flags, struct page **pages) | |
3064 | { | |
3faa52c0 JH |
3065 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3066 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3067 | return -EINVAL; | |
3068 | ||
0768c8de YN |
3069 | if (WARN_ON_ONCE(!pages)) |
3070 | return -EINVAL; | |
3071 | ||
3faa52c0 JH |
3072 | gup_flags |= FOLL_PIN; |
3073 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); | |
eddb1c22 JH |
3074 | } |
3075 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
3076 | ||
104acc32 | 3077 | /* |
dadbb612 SJ |
3078 | * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior |
3079 | * is the same, except that this one sets FOLL_PIN instead of FOLL_GET. | |
104acc32 JH |
3080 | * |
3081 | * The API rules are the same, too: no negative values may be returned. | |
3082 | */ | |
3083 | int pin_user_pages_fast_only(unsigned long start, int nr_pages, | |
3084 | unsigned int gup_flags, struct page **pages) | |
3085 | { | |
3086 | int nr_pinned; | |
3087 | ||
3088 | /* | |
3089 | * FOLL_GET and FOLL_PIN are mutually exclusive. Note that the API | |
3090 | * rules require returning 0, rather than -errno: | |
3091 | */ | |
3092 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3093 | return 0; | |
0768c8de YN |
3094 | |
3095 | if (WARN_ON_ONCE(!pages)) | |
3096 | return 0; | |
104acc32 JH |
3097 | /* |
3098 | * FOLL_FAST_ONLY is required in order to match the API description of | |
3099 | * this routine: no fall back to regular ("slow") GUP. | |
3100 | */ | |
3101 | gup_flags |= (FOLL_PIN | FOLL_FAST_ONLY); | |
3102 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, | |
3103 | pages); | |
3104 | /* | |
3105 | * This routine is not allowed to return negative values. However, | |
3106 | * internal_get_user_pages_fast() *can* return -errno. Therefore, | |
3107 | * correct for that here: | |
3108 | */ | |
3109 | if (nr_pinned < 0) | |
3110 | nr_pinned = 0; | |
3111 | ||
3112 | return nr_pinned; | |
3113 | } | |
3114 | EXPORT_SYMBOL_GPL(pin_user_pages_fast_only); | |
3115 | ||
eddb1c22 | 3116 | /** |
64019a2e | 3117 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 3118 | * |
3faa52c0 JH |
3119 | * @mm: mm_struct of target mm |
3120 | * @start: starting user address | |
3121 | * @nr_pages: number of pages from start to pin | |
3122 | * @gup_flags: flags modifying lookup behaviour | |
3123 | * @pages: array that receives pointers to the pages pinned. | |
0768c8de | 3124 | * Should be at least nr_pages long. |
3faa52c0 JH |
3125 | * @vmas: array of pointers to vmas corresponding to each page. |
3126 | * Or NULL if the caller does not require them. | |
3127 | * @locked: pointer to lock flag indicating whether lock is held and | |
3128 | * subsequently whether VM_FAULT_RETRY functionality can be | |
3129 | * utilised. Lock must initially be held. | |
3130 | * | |
3131 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
3132 | * get_user_pages_remote() for documentation on the function arguments, because | |
3133 | * the arguments here are identical. | |
3134 | * | |
3135 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3136 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 | 3137 | */ |
64019a2e | 3138 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
3139 | unsigned long start, unsigned long nr_pages, |
3140 | unsigned int gup_flags, struct page **pages, | |
3141 | struct vm_area_struct **vmas, int *locked) | |
3142 | { | |
3faa52c0 JH |
3143 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3144 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3145 | return -EINVAL; | |
3146 | ||
0768c8de YN |
3147 | if (WARN_ON_ONCE(!pages)) |
3148 | return -EINVAL; | |
3149 | ||
3faa52c0 | 3150 | gup_flags |= FOLL_PIN; |
64019a2e | 3151 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
3faa52c0 | 3152 | pages, vmas, locked); |
eddb1c22 JH |
3153 | } |
3154 | EXPORT_SYMBOL(pin_user_pages_remote); | |
3155 | ||
3156 | /** | |
3157 | * pin_user_pages() - pin user pages in memory for use by other devices | |
3158 | * | |
3faa52c0 JH |
3159 | * @start: starting user address |
3160 | * @nr_pages: number of pages from start to pin | |
3161 | * @gup_flags: flags modifying lookup behaviour | |
3162 | * @pages: array that receives pointers to the pages pinned. | |
0768c8de | 3163 | * Should be at least nr_pages long. |
3faa52c0 JH |
3164 | * @vmas: array of pointers to vmas corresponding to each page. |
3165 | * Or NULL if the caller does not require them. | |
3166 | * | |
3167 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
3168 | * FOLL_PIN is set. | |
3169 | * | |
3170 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3171 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 JH |
3172 | */ |
3173 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
3174 | unsigned int gup_flags, struct page **pages, | |
3175 | struct vm_area_struct **vmas) | |
3176 | { | |
3faa52c0 JH |
3177 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3178 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3179 | return -EINVAL; | |
3180 | ||
0768c8de YN |
3181 | if (WARN_ON_ONCE(!pages)) |
3182 | return -EINVAL; | |
3183 | ||
3faa52c0 | 3184 | gup_flags |= FOLL_PIN; |
64019a2e | 3185 | return __gup_longterm_locked(current->mm, start, nr_pages, |
3faa52c0 | 3186 | pages, vmas, gup_flags); |
eddb1c22 JH |
3187 | } |
3188 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
3189 | |
3190 | /* | |
3191 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
3192 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
3193 | * FOLL_PIN and rejects FOLL_GET. | |
3194 | */ | |
3195 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
3196 | struct page **pages, unsigned int gup_flags) | |
3197 | { | |
3198 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
3199 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3200 | return -EINVAL; | |
3201 | ||
0768c8de YN |
3202 | if (WARN_ON_ONCE(!pages)) |
3203 | return -EINVAL; | |
3204 | ||
91429023 JH |
3205 | gup_flags |= FOLL_PIN; |
3206 | return get_user_pages_unlocked(start, nr_pages, pages, gup_flags); | |
3207 | } | |
3208 | EXPORT_SYMBOL(pin_user_pages_unlocked); |