]>
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 | 89 | if (unlikely(page_folio(page) != folio)) { |
f4f451a1 MS |
90 | if (!put_devmap_managed_page_refs(&folio->page, refs)) |
91 | folio_put_refs(folio, refs); | |
59409373 | 92 | goto retry; |
c24d3732 JH |
93 | } |
94 | ||
ece1ed7b | 95 | return folio; |
a707cdd5 JH |
96 | } |
97 | ||
3967db22 | 98 | /** |
ece1ed7b | 99 | * try_grab_folio() - Attempt to get or pin a folio. |
3967db22 | 100 | * @page: pointer to page to be grabbed |
ece1ed7b | 101 | * @refs: the value to (effectively) add to the folio's refcount |
3967db22 JH |
102 | * @flags: gup flags: these are the FOLL_* flag values. |
103 | * | |
3faa52c0 | 104 | * "grab" names in this file mean, "look at flags to decide whether to use |
ece1ed7b | 105 | * FOLL_PIN or FOLL_GET behavior, when incrementing the folio's refcount. |
3faa52c0 JH |
106 | * |
107 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
108 | * same time. (That's true throughout the get_user_pages*() and | |
109 | * pin_user_pages*() APIs.) Cases: | |
110 | * | |
ece1ed7b | 111 | * FOLL_GET: folio's refcount will be incremented by @refs. |
3967db22 | 112 | * |
ece1ed7b | 113 | * FOLL_PIN on large folios: folio's refcount will be incremented by |
94688e8e | 114 | * @refs, and its pincount will be incremented by @refs. |
3967db22 | 115 | * |
ece1ed7b | 116 | * FOLL_PIN on single-page folios: folio's refcount will be incremented by |
5232c63f | 117 | * @refs * GUP_PIN_COUNTING_BIAS. |
3faa52c0 | 118 | * |
ece1ed7b MWO |
119 | * Return: The folio containing @page (with refcount appropriately |
120 | * incremented) for success, or NULL upon failure. If neither FOLL_GET | |
121 | * nor FOLL_PIN was set, that's considered failure, and furthermore, | |
122 | * a likely bug in the caller, so a warning is also emitted. | |
3faa52c0 | 123 | */ |
ece1ed7b | 124 | struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) |
3faa52c0 | 125 | { |
4003f107 LG |
126 | if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page))) |
127 | return NULL; | |
128 | ||
3faa52c0 | 129 | if (flags & FOLL_GET) |
ece1ed7b | 130 | return try_get_folio(page, refs); |
3faa52c0 | 131 | else if (flags & FOLL_PIN) { |
ece1ed7b MWO |
132 | struct folio *folio; |
133 | ||
df3a0a21 | 134 | /* |
d1e153fe PT |
135 | * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a |
136 | * right zone, so fail and let the caller fall back to the slow | |
137 | * path. | |
df3a0a21 | 138 | */ |
d1e153fe | 139 | if (unlikely((flags & FOLL_LONGTERM) && |
6077c943 | 140 | !is_longterm_pinnable_page(page))) |
df3a0a21 PL |
141 | return NULL; |
142 | ||
c24d3732 JH |
143 | /* |
144 | * CAUTION: Don't use compound_head() on the page before this | |
145 | * point, the result won't be stable. | |
146 | */ | |
ece1ed7b MWO |
147 | folio = try_get_folio(page, refs); |
148 | if (!folio) | |
c24d3732 JH |
149 | return NULL; |
150 | ||
47e29d32 | 151 | /* |
ece1ed7b | 152 | * When pinning a large folio, use an exact count to track it. |
47e29d32 | 153 | * |
ece1ed7b MWO |
154 | * However, be sure to *also* increment the normal folio |
155 | * refcount field at least once, so that the folio really | |
78d9d6ce | 156 | * is pinned. That's why the refcount from the earlier |
ece1ed7b | 157 | * try_get_folio() is left intact. |
47e29d32 | 158 | */ |
ece1ed7b | 159 | if (folio_test_large(folio)) |
94688e8e | 160 | atomic_add(refs, &folio->_pincount); |
c24d3732 | 161 | else |
ece1ed7b MWO |
162 | folio_ref_add(folio, |
163 | refs * (GUP_PIN_COUNTING_BIAS - 1)); | |
088b8aa5 DH |
164 | /* |
165 | * Adjust the pincount before re-checking the PTE for changes. | |
166 | * This is essentially a smp_mb() and is paired with a memory | |
167 | * barrier in page_try_share_anon_rmap(). | |
168 | */ | |
169 | smp_mb__after_atomic(); | |
170 | ||
ece1ed7b | 171 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, refs); |
47e29d32 | 172 | |
ece1ed7b | 173 | return folio; |
3faa52c0 JH |
174 | } |
175 | ||
176 | WARN_ON_ONCE(1); | |
177 | return NULL; | |
178 | } | |
179 | ||
d8ddc099 | 180 | static void gup_put_folio(struct folio *folio, int refs, unsigned int flags) |
4509b42c JG |
181 | { |
182 | if (flags & FOLL_PIN) { | |
d8ddc099 MWO |
183 | node_stat_mod_folio(folio, NR_FOLL_PIN_RELEASED, refs); |
184 | if (folio_test_large(folio)) | |
94688e8e | 185 | atomic_sub(refs, &folio->_pincount); |
4509b42c JG |
186 | else |
187 | refs *= GUP_PIN_COUNTING_BIAS; | |
188 | } | |
189 | ||
f4f451a1 MS |
190 | if (!put_devmap_managed_page_refs(&folio->page, refs)) |
191 | folio_put_refs(folio, refs); | |
4509b42c JG |
192 | } |
193 | ||
3faa52c0 JH |
194 | /** |
195 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
5fec0719 MWO |
196 | * @page: pointer to page to be grabbed |
197 | * @flags: gup flags: these are the FOLL_* flag values. | |
3faa52c0 JH |
198 | * |
199 | * This might not do anything at all, depending on the flags argument. | |
200 | * | |
201 | * "grab" names in this file mean, "look at flags to decide whether to use | |
202 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
203 | * | |
3faa52c0 | 204 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same |
ece1ed7b | 205 | * time. Cases: please see the try_grab_folio() documentation, with |
3967db22 | 206 | * "refs=1". |
3faa52c0 | 207 | * |
0f089235 LG |
208 | * Return: 0 for success, or if no action was required (if neither FOLL_PIN |
209 | * nor FOLL_GET was set, nothing is done). A negative error code for failure: | |
210 | * | |
211 | * -ENOMEM FOLL_GET or FOLL_PIN was set, but the page could not | |
212 | * be grabbed. | |
3faa52c0 | 213 | */ |
0f089235 | 214 | int __must_check try_grab_page(struct page *page, unsigned int flags) |
3faa52c0 | 215 | { |
5fec0719 MWO |
216 | struct folio *folio = page_folio(page); |
217 | ||
5fec0719 | 218 | if (WARN_ON_ONCE(folio_ref_count(folio) <= 0)) |
0f089235 | 219 | return -ENOMEM; |
3faa52c0 | 220 | |
4003f107 LG |
221 | if (unlikely(!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page))) |
222 | return -EREMOTEIO; | |
3faa52c0 | 223 | |
c36c04c2 | 224 | if (flags & FOLL_GET) |
5fec0719 | 225 | folio_ref_inc(folio); |
c36c04c2 | 226 | else if (flags & FOLL_PIN) { |
c36c04c2 | 227 | /* |
5fec0719 | 228 | * Similar to try_grab_folio(): be sure to *also* |
78d9d6ce MWO |
229 | * increment the normal page refcount field at least once, |
230 | * so that the page really is pinned. | |
c36c04c2 | 231 | */ |
5fec0719 MWO |
232 | if (folio_test_large(folio)) { |
233 | folio_ref_add(folio, 1); | |
94688e8e | 234 | atomic_add(1, &folio->_pincount); |
8ea2979c | 235 | } else { |
5fec0719 | 236 | folio_ref_add(folio, GUP_PIN_COUNTING_BIAS); |
8ea2979c | 237 | } |
c36c04c2 | 238 | |
5fec0719 | 239 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, 1); |
c36c04c2 JH |
240 | } |
241 | ||
0f089235 | 242 | return 0; |
3faa52c0 JH |
243 | } |
244 | ||
3faa52c0 JH |
245 | /** |
246 | * unpin_user_page() - release a dma-pinned page | |
247 | * @page: pointer to page to be released | |
248 | * | |
249 | * Pages that were pinned via pin_user_pages*() must be released via either | |
250 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
251 | * that such pages can be separately tracked and uniquely handled. In | |
252 | * particular, interactions with RDMA and filesystems need special handling. | |
253 | */ | |
254 | void unpin_user_page(struct page *page) | |
255 | { | |
b6a2619c | 256 | sanity_check_pinned_pages(&page, 1); |
d8ddc099 | 257 | gup_put_folio(page_folio(page), 1, FOLL_PIN); |
3faa52c0 JH |
258 | } |
259 | EXPORT_SYMBOL(unpin_user_page); | |
260 | ||
659508f9 | 261 | static inline struct folio *gup_folio_range_next(struct page *start, |
8f39f5fc | 262 | unsigned long npages, unsigned long i, unsigned int *ntails) |
458a4f78 | 263 | { |
659508f9 MWO |
264 | struct page *next = nth_page(start, i); |
265 | struct folio *folio = page_folio(next); | |
458a4f78 JM |
266 | unsigned int nr = 1; |
267 | ||
659508f9 | 268 | if (folio_test_large(folio)) |
4c654229 | 269 | nr = min_t(unsigned int, npages - i, |
659508f9 | 270 | folio_nr_pages(folio) - folio_page_idx(folio, next)); |
458a4f78 | 271 | |
458a4f78 | 272 | *ntails = nr; |
659508f9 | 273 | return folio; |
458a4f78 JM |
274 | } |
275 | ||
12521c76 | 276 | static inline struct folio *gup_folio_next(struct page **list, |
28297dbc | 277 | unsigned long npages, unsigned long i, unsigned int *ntails) |
8745d7f6 | 278 | { |
12521c76 | 279 | struct folio *folio = page_folio(list[i]); |
8745d7f6 JM |
280 | unsigned int nr; |
281 | ||
8745d7f6 | 282 | for (nr = i + 1; nr < npages; nr++) { |
12521c76 | 283 | if (page_folio(list[nr]) != folio) |
8745d7f6 JM |
284 | break; |
285 | } | |
286 | ||
8745d7f6 | 287 | *ntails = nr - i; |
12521c76 | 288 | return folio; |
8745d7f6 JM |
289 | } |
290 | ||
fc1d8e7c | 291 | /** |
f1f6a7dd | 292 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 293 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 294 | * @npages: number of pages in the @pages array. |
2d15eb31 | 295 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
296 | * |
297 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
298 | * variants called on that page. | |
299 | * | |
300 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 301 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
302 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
303 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 304 | * |
f1f6a7dd | 305 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 306 | * |
2d15eb31 AM |
307 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
308 | * required, then the caller should a) verify that this is really correct, | |
309 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 310 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
311 | * |
312 | */ | |
f1f6a7dd JH |
313 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
314 | bool make_dirty) | |
fc1d8e7c | 315 | { |
12521c76 MWO |
316 | unsigned long i; |
317 | struct folio *folio; | |
318 | unsigned int nr; | |
2d15eb31 AM |
319 | |
320 | if (!make_dirty) { | |
f1f6a7dd | 321 | unpin_user_pages(pages, npages); |
2d15eb31 AM |
322 | return; |
323 | } | |
324 | ||
b6a2619c | 325 | sanity_check_pinned_pages(pages, npages); |
12521c76 MWO |
326 | for (i = 0; i < npages; i += nr) { |
327 | folio = gup_folio_next(pages, npages, i, &nr); | |
2d15eb31 AM |
328 | /* |
329 | * Checking PageDirty at this point may race with | |
330 | * clear_page_dirty_for_io(), but that's OK. Two key | |
331 | * cases: | |
332 | * | |
333 | * 1) This code sees the page as already dirty, so it | |
334 | * skips the call to set_page_dirty(). That could happen | |
335 | * because clear_page_dirty_for_io() called | |
336 | * page_mkclean(), followed by set_page_dirty(). | |
337 | * However, now the page is going to get written back, | |
338 | * which meets the original intention of setting it | |
339 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
340 | * on to call TestClearPageDirty(), and write the page | |
341 | * back. | |
342 | * | |
343 | * 2) This code sees the page as clean, so it calls | |
344 | * set_page_dirty(). The page stays dirty, despite being | |
345 | * written back, so it gets written back again in the | |
346 | * next writeback cycle. This is harmless. | |
347 | */ | |
12521c76 MWO |
348 | if (!folio_test_dirty(folio)) { |
349 | folio_lock(folio); | |
350 | folio_mark_dirty(folio); | |
351 | folio_unlock(folio); | |
352 | } | |
353 | gup_put_folio(folio, nr, FOLL_PIN); | |
2d15eb31 | 354 | } |
fc1d8e7c | 355 | } |
f1f6a7dd | 356 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c | 357 | |
458a4f78 JM |
358 | /** |
359 | * unpin_user_page_range_dirty_lock() - release and optionally dirty | |
360 | * gup-pinned page range | |
361 | * | |
362 | * @page: the starting page of a range maybe marked dirty, and definitely released. | |
363 | * @npages: number of consecutive pages to release. | |
364 | * @make_dirty: whether to mark the pages dirty | |
365 | * | |
366 | * "gup-pinned page range" refers to a range of pages that has had one of the | |
367 | * pin_user_pages() variants called on that page. | |
368 | * | |
369 | * For the page ranges defined by [page .. page+npages], make that range (or | |
370 | * its head pages, if a compound page) dirty, if @make_dirty is true, and if the | |
371 | * page range was previously listed as clean. | |
372 | * | |
373 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is | |
374 | * required, then the caller should a) verify that this is really correct, | |
375 | * because _lock() is usually required, and b) hand code it: | |
376 | * set_page_dirty_lock(), unpin_user_page(). | |
377 | * | |
378 | */ | |
379 | void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, | |
380 | bool make_dirty) | |
381 | { | |
659508f9 MWO |
382 | unsigned long i; |
383 | struct folio *folio; | |
384 | unsigned int nr; | |
385 | ||
386 | for (i = 0; i < npages; i += nr) { | |
387 | folio = gup_folio_range_next(page, npages, i, &nr); | |
388 | if (make_dirty && !folio_test_dirty(folio)) { | |
389 | folio_lock(folio); | |
390 | folio_mark_dirty(folio); | |
391 | folio_unlock(folio); | |
392 | } | |
393 | gup_put_folio(folio, nr, FOLL_PIN); | |
458a4f78 JM |
394 | } |
395 | } | |
396 | EXPORT_SYMBOL(unpin_user_page_range_dirty_lock); | |
397 | ||
b6a2619c DH |
398 | static void unpin_user_pages_lockless(struct page **pages, unsigned long npages) |
399 | { | |
400 | unsigned long i; | |
401 | struct folio *folio; | |
402 | unsigned int nr; | |
403 | ||
404 | /* | |
405 | * Don't perform any sanity checks because we might have raced with | |
406 | * fork() and some anonymous pages might now actually be shared -- | |
407 | * which is why we're unpinning after all. | |
408 | */ | |
409 | for (i = 0; i < npages; i += nr) { | |
410 | folio = gup_folio_next(pages, npages, i, &nr); | |
411 | gup_put_folio(folio, nr, FOLL_PIN); | |
412 | } | |
413 | } | |
414 | ||
fc1d8e7c | 415 | /** |
f1f6a7dd | 416 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
417 | * @pages: array of pages to be marked dirty and released. |
418 | * @npages: number of pages in the @pages array. | |
419 | * | |
f1f6a7dd | 420 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 421 | * |
f1f6a7dd | 422 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 423 | */ |
f1f6a7dd | 424 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c | 425 | { |
12521c76 MWO |
426 | unsigned long i; |
427 | struct folio *folio; | |
428 | unsigned int nr; | |
fc1d8e7c | 429 | |
146608bb JH |
430 | /* |
431 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
432 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
433 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
434 | */ | |
435 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
436 | return; | |
31b912de | 437 | |
b6a2619c | 438 | sanity_check_pinned_pages(pages, npages); |
12521c76 MWO |
439 | for (i = 0; i < npages; i += nr) { |
440 | folio = gup_folio_next(pages, npages, i, &nr); | |
441 | gup_put_folio(folio, nr, FOLL_PIN); | |
e7602748 | 442 | } |
fc1d8e7c | 443 | } |
f1f6a7dd | 444 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 445 | |
a458b76a AA |
446 | /* |
447 | * Set the MMF_HAS_PINNED if not set yet; after set it'll be there for the mm's | |
448 | * lifecycle. Avoid setting the bit unless necessary, or it might cause write | |
449 | * cache bouncing on large SMP machines for concurrent pinned gups. | |
450 | */ | |
451 | static inline void mm_set_has_pinned_flag(unsigned long *mm_flags) | |
452 | { | |
453 | if (!test_bit(MMF_HAS_PINNED, mm_flags)) | |
454 | set_bit(MMF_HAS_PINNED, mm_flags); | |
455 | } | |
456 | ||
050a9adc | 457 | #ifdef CONFIG_MMU |
69e68b4f KS |
458 | static struct page *no_page_table(struct vm_area_struct *vma, |
459 | unsigned int flags) | |
4bbd4c77 | 460 | { |
69e68b4f KS |
461 | /* |
462 | * When core dumping an enormous anonymous area that nobody | |
463 | * has touched so far, we don't want to allocate unnecessary pages or | |
464 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
465 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
466 | * But we can only make this optimization where a hole would surely | |
467 | * be zero-filled if handle_mm_fault() actually did handle it. | |
468 | */ | |
a0137f16 AK |
469 | if ((flags & FOLL_DUMP) && |
470 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
471 | return ERR_PTR(-EFAULT); |
472 | return NULL; | |
473 | } | |
4bbd4c77 | 474 | |
1027e443 KS |
475 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
476 | pte_t *pte, unsigned int flags) | |
477 | { | |
1027e443 KS |
478 | if (flags & FOLL_TOUCH) { |
479 | pte_t entry = *pte; | |
480 | ||
481 | if (flags & FOLL_WRITE) | |
482 | entry = pte_mkdirty(entry); | |
483 | entry = pte_mkyoung(entry); | |
484 | ||
485 | if (!pte_same(*pte, entry)) { | |
486 | set_pte_at(vma->vm_mm, address, pte, entry); | |
487 | update_mmu_cache(vma, address, pte); | |
488 | } | |
489 | } | |
490 | ||
491 | /* Proper page table entry exists, but no corresponding struct page */ | |
492 | return -EEXIST; | |
493 | } | |
494 | ||
5535be30 DH |
495 | /* FOLL_FORCE can write to even unwritable PTEs in COW mappings. */ |
496 | static inline bool can_follow_write_pte(pte_t pte, struct page *page, | |
497 | struct vm_area_struct *vma, | |
498 | unsigned int flags) | |
19be0eaf | 499 | { |
5535be30 DH |
500 | /* If the pte is writable, we can write to the page. */ |
501 | if (pte_write(pte)) | |
502 | return true; | |
503 | ||
504 | /* Maybe FOLL_FORCE is set to override it? */ | |
505 | if (!(flags & FOLL_FORCE)) | |
506 | return false; | |
507 | ||
508 | /* But FOLL_FORCE has no effect on shared mappings */ | |
509 | if (vma->vm_flags & (VM_MAYSHARE | VM_SHARED)) | |
510 | return false; | |
511 | ||
512 | /* ... or read-only private ones */ | |
513 | if (!(vma->vm_flags & VM_MAYWRITE)) | |
514 | return false; | |
515 | ||
516 | /* ... or already writable ones that just need to take a write fault */ | |
517 | if (vma->vm_flags & VM_WRITE) | |
518 | return false; | |
519 | ||
520 | /* | |
521 | * See can_change_pte_writable(): we broke COW and could map the page | |
522 | * writable if we have an exclusive anonymous page ... | |
523 | */ | |
524 | if (!page || !PageAnon(page) || !PageAnonExclusive(page)) | |
525 | return false; | |
526 | ||
527 | /* ... and a write-fault isn't required for other reasons. */ | |
528 | if (vma_soft_dirty_enabled(vma) && !pte_soft_dirty(pte)) | |
529 | return false; | |
530 | return !userfaultfd_pte_wp(vma, pte); | |
19be0eaf LT |
531 | } |
532 | ||
69e68b4f | 533 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
534 | unsigned long address, pmd_t *pmd, unsigned int flags, |
535 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
536 | { |
537 | struct mm_struct *mm = vma->vm_mm; | |
538 | struct page *page; | |
539 | spinlock_t *ptl; | |
540 | pte_t *ptep, pte; | |
f28d4363 | 541 | int ret; |
4bbd4c77 | 542 | |
eddb1c22 JH |
543 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
544 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
545 | (FOLL_PIN | FOLL_GET))) | |
546 | return ERR_PTR(-EINVAL); | |
4bbd4c77 | 547 | if (unlikely(pmd_bad(*pmd))) |
69e68b4f | 548 | return no_page_table(vma, flags); |
4bbd4c77 KS |
549 | |
550 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
4bbd4c77 | 551 | pte = *ptep; |
f7355e99 DH |
552 | if (!pte_present(pte)) |
553 | goto no_page; | |
474098ed | 554 | if (pte_protnone(pte) && !gup_can_follow_protnone(flags)) |
4bbd4c77 | 555 | goto no_page; |
4bbd4c77 KS |
556 | |
557 | page = vm_normal_page(vma, address, pte); | |
5535be30 DH |
558 | |
559 | /* | |
560 | * We only care about anon pages in can_follow_write_pte() and don't | |
561 | * have to worry about pte_devmap() because they are never anon. | |
562 | */ | |
563 | if ((flags & FOLL_WRITE) && | |
564 | !can_follow_write_pte(pte, page, vma, flags)) { | |
565 | page = NULL; | |
566 | goto out; | |
567 | } | |
568 | ||
3faa52c0 | 569 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 570 | /* |
3faa52c0 JH |
571 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
572 | * case since they are only valid while holding the pgmap | |
573 | * reference. | |
3565fce3 | 574 | */ |
df06b37f KB |
575 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
576 | if (*pgmap) | |
3565fce3 DW |
577 | page = pte_page(pte); |
578 | else | |
579 | goto no_page; | |
580 | } else if (unlikely(!page)) { | |
1027e443 KS |
581 | if (flags & FOLL_DUMP) { |
582 | /* Avoid special (like zero) pages in core dumps */ | |
583 | page = ERR_PTR(-EFAULT); | |
584 | goto out; | |
585 | } | |
586 | ||
587 | if (is_zero_pfn(pte_pfn(pte))) { | |
588 | page = pte_page(pte); | |
589 | } else { | |
1027e443 KS |
590 | ret = follow_pfn_pte(vma, address, ptep, flags); |
591 | page = ERR_PTR(ret); | |
592 | goto out; | |
593 | } | |
4bbd4c77 KS |
594 | } |
595 | ||
84209e87 | 596 | if (!pte_write(pte) && gup_must_unshare(vma, flags, page)) { |
a7f22660 DH |
597 | page = ERR_PTR(-EMLINK); |
598 | goto out; | |
599 | } | |
b6a2619c DH |
600 | |
601 | VM_BUG_ON_PAGE((flags & FOLL_PIN) && PageAnon(page) && | |
602 | !PageAnonExclusive(page), page); | |
603 | ||
3faa52c0 | 604 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
0f089235 LG |
605 | ret = try_grab_page(page, flags); |
606 | if (unlikely(ret)) { | |
607 | page = ERR_PTR(ret); | |
3faa52c0 | 608 | goto out; |
8fde12ca | 609 | } |
4003f107 | 610 | |
f28d4363 CI |
611 | /* |
612 | * We need to make the page accessible if and only if we are going | |
613 | * to access its content (the FOLL_PIN case). Please see | |
614 | * Documentation/core-api/pin_user_pages.rst for details. | |
615 | */ | |
616 | if (flags & FOLL_PIN) { | |
617 | ret = arch_make_page_accessible(page); | |
618 | if (ret) { | |
619 | unpin_user_page(page); | |
620 | page = ERR_PTR(ret); | |
621 | goto out; | |
622 | } | |
623 | } | |
4bbd4c77 KS |
624 | if (flags & FOLL_TOUCH) { |
625 | if ((flags & FOLL_WRITE) && | |
626 | !pte_dirty(pte) && !PageDirty(page)) | |
627 | set_page_dirty(page); | |
628 | /* | |
629 | * pte_mkyoung() would be more correct here, but atomic care | |
630 | * is needed to avoid losing the dirty bit: it is easier to use | |
631 | * mark_page_accessed(). | |
632 | */ | |
633 | mark_page_accessed(page); | |
634 | } | |
1027e443 | 635 | out: |
4bbd4c77 | 636 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 637 | return page; |
4bbd4c77 KS |
638 | no_page: |
639 | pte_unmap_unlock(ptep, ptl); | |
640 | if (!pte_none(pte)) | |
69e68b4f KS |
641 | return NULL; |
642 | return no_page_table(vma, flags); | |
643 | } | |
644 | ||
080dbb61 AK |
645 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
646 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
647 | unsigned int flags, |
648 | struct follow_page_context *ctx) | |
69e68b4f | 649 | { |
68827280 | 650 | pmd_t *pmd, pmdval; |
69e68b4f KS |
651 | spinlock_t *ptl; |
652 | struct page *page; | |
653 | struct mm_struct *mm = vma->vm_mm; | |
654 | ||
080dbb61 | 655 | pmd = pmd_offset(pudp, address); |
68827280 HY |
656 | /* |
657 | * The READ_ONCE() will stabilize the pmdval in a register or | |
658 | * on the stack so that it will stop changing under the code. | |
659 | */ | |
660 | pmdval = READ_ONCE(*pmd); | |
661 | if (pmd_none(pmdval)) | |
69e68b4f | 662 | return no_page_table(vma, flags); |
f7355e99 | 663 | if (!pmd_present(pmdval)) |
e66f17ff | 664 | return no_page_table(vma, flags); |
68827280 | 665 | if (pmd_devmap(pmdval)) { |
3565fce3 | 666 | ptl = pmd_lock(mm, pmd); |
df06b37f | 667 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
668 | spin_unlock(ptl); |
669 | if (page) | |
670 | return page; | |
671 | } | |
68827280 | 672 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 673 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 674 | |
474098ed | 675 | if (pmd_protnone(pmdval) && !gup_can_follow_protnone(flags)) |
db08f203 AK |
676 | return no_page_table(vma, flags); |
677 | ||
6742d293 | 678 | ptl = pmd_lock(mm, pmd); |
84c3fc4e ZY |
679 | if (unlikely(!pmd_present(*pmd))) { |
680 | spin_unlock(ptl); | |
f7355e99 | 681 | return no_page_table(vma, flags); |
84c3fc4e | 682 | } |
6742d293 KS |
683 | if (unlikely(!pmd_trans_huge(*pmd))) { |
684 | spin_unlock(ptl); | |
df06b37f | 685 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 686 | } |
4066c119 | 687 | if (flags & FOLL_SPLIT_PMD) { |
6742d293 KS |
688 | int ret; |
689 | page = pmd_page(*pmd); | |
690 | if (is_huge_zero_page(page)) { | |
691 | spin_unlock(ptl); | |
692 | ret = 0; | |
78ddc534 | 693 | split_huge_pmd(vma, pmd, address); |
337d9abf NH |
694 | if (pmd_trans_unstable(pmd)) |
695 | ret = -EBUSY; | |
4066c119 | 696 | } else { |
bfe7b00d SL |
697 | spin_unlock(ptl); |
698 | split_huge_pmd(vma, pmd, address); | |
699 | ret = pte_alloc(mm, pmd) ? -ENOMEM : 0; | |
6742d293 KS |
700 | } |
701 | ||
702 | return ret ? ERR_PTR(ret) : | |
df06b37f | 703 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 704 | } |
6742d293 KS |
705 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
706 | spin_unlock(ptl); | |
df06b37f | 707 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 708 | return page; |
4bbd4c77 KS |
709 | } |
710 | ||
080dbb61 AK |
711 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
712 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
713 | unsigned int flags, |
714 | struct follow_page_context *ctx) | |
080dbb61 AK |
715 | { |
716 | pud_t *pud; | |
717 | spinlock_t *ptl; | |
718 | struct page *page; | |
719 | struct mm_struct *mm = vma->vm_mm; | |
720 | ||
721 | pud = pud_offset(p4dp, address); | |
722 | if (pud_none(*pud)) | |
723 | return no_page_table(vma, flags); | |
080dbb61 AK |
724 | if (pud_devmap(*pud)) { |
725 | ptl = pud_lock(mm, pud); | |
df06b37f | 726 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
727 | spin_unlock(ptl); |
728 | if (page) | |
729 | return page; | |
730 | } | |
731 | if (unlikely(pud_bad(*pud))) | |
732 | return no_page_table(vma, flags); | |
733 | ||
df06b37f | 734 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
735 | } |
736 | ||
080dbb61 AK |
737 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
738 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
739 | unsigned int flags, |
740 | struct follow_page_context *ctx) | |
080dbb61 AK |
741 | { |
742 | p4d_t *p4d; | |
743 | ||
744 | p4d = p4d_offset(pgdp, address); | |
745 | if (p4d_none(*p4d)) | |
746 | return no_page_table(vma, flags); | |
747 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
748 | if (unlikely(p4d_bad(*p4d))) | |
749 | return no_page_table(vma, flags); | |
750 | ||
df06b37f | 751 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
752 | } |
753 | ||
754 | /** | |
755 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
756 | * @vma: vm_area_struct mapping @address | |
757 | * @address: virtual address to look up | |
758 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
759 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
760 | * pointer to output page_mask | |
080dbb61 AK |
761 | * |
762 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
763 | * | |
78179556 MR |
764 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
765 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
766 | * | |
a7f22660 DH |
767 | * When getting an anonymous page and the caller has to trigger unsharing |
768 | * of a shared anonymous page first, -EMLINK is returned. The caller should | |
769 | * trigger a fault with FAULT_FLAG_UNSHARE set. Note that unsharing is only | |
770 | * relevant with FOLL_PIN and !FOLL_WRITE. | |
771 | * | |
78179556 MR |
772 | * On output, the @ctx->page_mask is set according to the size of the page. |
773 | * | |
774 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
775 | * an error pointer if there is a mapping to something not represented |
776 | * by a page descriptor (see also vm_normal_page()). | |
777 | */ | |
a7030aea | 778 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 779 | unsigned long address, unsigned int flags, |
df06b37f | 780 | struct follow_page_context *ctx) |
080dbb61 AK |
781 | { |
782 | pgd_t *pgd; | |
783 | struct page *page; | |
784 | struct mm_struct *mm = vma->vm_mm; | |
785 | ||
df06b37f | 786 | ctx->page_mask = 0; |
080dbb61 | 787 | |
57a196a5 MK |
788 | /* |
789 | * Call hugetlb_follow_page_mask for hugetlb vmas as it will use | |
790 | * special hugetlb page table walking code. This eliminates the | |
791 | * need to check for hugetlb entries in the general walking code. | |
792 | * | |
793 | * hugetlb_follow_page_mask is only for follow_page() handling here. | |
794 | * Ordinary GUP uses follow_hugetlb_page for hugetlb processing. | |
795 | */ | |
796 | if (is_vm_hugetlb_page(vma)) { | |
797 | page = hugetlb_follow_page_mask(vma, address, flags); | |
798 | if (!page) | |
799 | page = no_page_table(vma, flags); | |
080dbb61 AK |
800 | return page; |
801 | } | |
802 | ||
803 | pgd = pgd_offset(mm, address); | |
804 | ||
805 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
806 | return no_page_table(vma, flags); | |
807 | ||
df06b37f KB |
808 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
809 | } | |
810 | ||
811 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
812 | unsigned int foll_flags) | |
813 | { | |
814 | struct follow_page_context ctx = { NULL }; | |
815 | struct page *page; | |
816 | ||
1507f512 MR |
817 | if (vma_is_secretmem(vma)) |
818 | return NULL; | |
819 | ||
d64e2dbc | 820 | if (WARN_ON_ONCE(foll_flags & FOLL_PIN)) |
8909691b DH |
821 | return NULL; |
822 | ||
df06b37f KB |
823 | page = follow_page_mask(vma, address, foll_flags, &ctx); |
824 | if (ctx.pgmap) | |
825 | put_dev_pagemap(ctx.pgmap); | |
826 | return page; | |
080dbb61 AK |
827 | } |
828 | ||
f2b495ca KS |
829 | static int get_gate_page(struct mm_struct *mm, unsigned long address, |
830 | unsigned int gup_flags, struct vm_area_struct **vma, | |
831 | struct page **page) | |
832 | { | |
833 | pgd_t *pgd; | |
c2febafc | 834 | p4d_t *p4d; |
f2b495ca KS |
835 | pud_t *pud; |
836 | pmd_t *pmd; | |
837 | pte_t *pte; | |
838 | int ret = -EFAULT; | |
839 | ||
840 | /* user gate pages are read-only */ | |
841 | if (gup_flags & FOLL_WRITE) | |
842 | return -EFAULT; | |
843 | if (address > TASK_SIZE) | |
844 | pgd = pgd_offset_k(address); | |
845 | else | |
846 | pgd = pgd_offset_gate(mm, address); | |
b5d1c39f AL |
847 | if (pgd_none(*pgd)) |
848 | return -EFAULT; | |
c2febafc | 849 | p4d = p4d_offset(pgd, address); |
b5d1c39f AL |
850 | if (p4d_none(*p4d)) |
851 | return -EFAULT; | |
c2febafc | 852 | pud = pud_offset(p4d, address); |
b5d1c39f AL |
853 | if (pud_none(*pud)) |
854 | return -EFAULT; | |
f2b495ca | 855 | pmd = pmd_offset(pud, address); |
84c3fc4e | 856 | if (!pmd_present(*pmd)) |
f2b495ca KS |
857 | return -EFAULT; |
858 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
859 | pte = pte_offset_map(pmd, address); | |
860 | if (pte_none(*pte)) | |
861 | goto unmap; | |
862 | *vma = get_gate_vma(mm); | |
863 | if (!page) | |
864 | goto out; | |
865 | *page = vm_normal_page(*vma, address, *pte); | |
866 | if (!*page) { | |
867 | if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte))) | |
868 | goto unmap; | |
869 | *page = pte_page(*pte); | |
870 | } | |
0f089235 LG |
871 | ret = try_grab_page(*page, gup_flags); |
872 | if (unlikely(ret)) | |
8fde12ca | 873 | goto unmap; |
f2b495ca KS |
874 | out: |
875 | ret = 0; | |
876 | unmap: | |
877 | pte_unmap(pte); | |
878 | return ret; | |
879 | } | |
880 | ||
9a95f3cf | 881 | /* |
c1e8d7c6 ML |
882 | * mmap_lock must be held on entry. If @locked != NULL and *@flags |
883 | * does not include FOLL_NOWAIT, the mmap_lock may be released. If it | |
4f6da934 | 884 | * is, *@locked will be set to 0 and -EBUSY returned. |
9a95f3cf | 885 | */ |
64019a2e | 886 | static int faultin_page(struct vm_area_struct *vma, |
a7f22660 DH |
887 | unsigned long address, unsigned int *flags, bool unshare, |
888 | int *locked) | |
16744483 | 889 | { |
16744483 | 890 | unsigned int fault_flags = 0; |
2b740303 | 891 | vm_fault_t ret; |
16744483 | 892 | |
55b8fe70 AG |
893 | if (*flags & FOLL_NOFAULT) |
894 | return -EFAULT; | |
16744483 KS |
895 | if (*flags & FOLL_WRITE) |
896 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
897 | if (*flags & FOLL_REMOTE) |
898 | fault_flags |= FAULT_FLAG_REMOTE; | |
93c5c61d | 899 | if (locked) { |
71335f37 | 900 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
93c5c61d PX |
901 | /* |
902 | * FAULT_FLAG_INTERRUPTIBLE is opt-in. GUP callers must set | |
903 | * FOLL_INTERRUPTIBLE to enable FAULT_FLAG_INTERRUPTIBLE. | |
904 | * That's because some callers may not be prepared to | |
905 | * handle early exits caused by non-fatal signals. | |
906 | */ | |
907 | if (*flags & FOLL_INTERRUPTIBLE) | |
908 | fault_flags |= FAULT_FLAG_INTERRUPTIBLE; | |
909 | } | |
16744483 KS |
910 | if (*flags & FOLL_NOWAIT) |
911 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 912 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
913 | /* |
914 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
915 | * can co-exist | |
916 | */ | |
234b239b ALC |
917 | fault_flags |= FAULT_FLAG_TRIED; |
918 | } | |
a7f22660 DH |
919 | if (unshare) { |
920 | fault_flags |= FAULT_FLAG_UNSHARE; | |
921 | /* FAULT_FLAG_WRITE and FAULT_FLAG_UNSHARE are incompatible */ | |
922 | VM_BUG_ON(fault_flags & FAULT_FLAG_WRITE); | |
923 | } | |
16744483 | 924 | |
bce617ed | 925 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
d9272525 PX |
926 | |
927 | if (ret & VM_FAULT_COMPLETED) { | |
928 | /* | |
929 | * With FAULT_FLAG_RETRY_NOWAIT we'll never release the | |
930 | * mmap lock in the page fault handler. Sanity check this. | |
931 | */ | |
932 | WARN_ON_ONCE(fault_flags & FAULT_FLAG_RETRY_NOWAIT); | |
933 | if (locked) | |
934 | *locked = 0; | |
935 | /* | |
936 | * We should do the same as VM_FAULT_RETRY, but let's not | |
937 | * return -EBUSY since that's not reflecting the reality of | |
938 | * what has happened - we've just fully completed a page | |
939 | * fault, with the mmap lock released. Use -EAGAIN to show | |
940 | * that we want to take the mmap lock _again_. | |
941 | */ | |
942 | return -EAGAIN; | |
943 | } | |
944 | ||
16744483 | 945 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
946 | int err = vm_fault_to_errno(ret, *flags); |
947 | ||
948 | if (err) | |
949 | return err; | |
16744483 KS |
950 | BUG(); |
951 | } | |
952 | ||
16744483 | 953 | if (ret & VM_FAULT_RETRY) { |
4f6da934 PX |
954 | if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
955 | *locked = 0; | |
16744483 KS |
956 | return -EBUSY; |
957 | } | |
958 | ||
16744483 KS |
959 | return 0; |
960 | } | |
961 | ||
fa5bb209 KS |
962 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
963 | { | |
964 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
965 | int write = (gup_flags & FOLL_WRITE); |
966 | int foreign = (gup_flags & FOLL_REMOTE); | |
fa5bb209 KS |
967 | |
968 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
969 | return -EFAULT; | |
970 | ||
7f7ccc2c WT |
971 | if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma)) |
972 | return -EFAULT; | |
973 | ||
52650c8b JG |
974 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
975 | return -EOPNOTSUPP; | |
976 | ||
1507f512 MR |
977 | if (vma_is_secretmem(vma)) |
978 | return -EFAULT; | |
979 | ||
1b2ee126 | 980 | if (write) { |
fa5bb209 KS |
981 | if (!(vm_flags & VM_WRITE)) { |
982 | if (!(gup_flags & FOLL_FORCE)) | |
983 | return -EFAULT; | |
f347454d DH |
984 | /* hugetlb does not support FOLL_FORCE|FOLL_WRITE. */ |
985 | if (is_vm_hugetlb_page(vma)) | |
986 | return -EFAULT; | |
fa5bb209 KS |
987 | /* |
988 | * We used to let the write,force case do COW in a | |
989 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
990 | * set a breakpoint in a read-only mapping of an | |
991 | * executable, without corrupting the file (yet only | |
992 | * when that file had been opened for writing!). | |
993 | * Anon pages in shared mappings are surprising: now | |
994 | * just reject it. | |
995 | */ | |
46435364 | 996 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 997 | return -EFAULT; |
fa5bb209 KS |
998 | } |
999 | } else if (!(vm_flags & VM_READ)) { | |
1000 | if (!(gup_flags & FOLL_FORCE)) | |
1001 | return -EFAULT; | |
1002 | /* | |
1003 | * Is there actually any vma we can reach here which does not | |
1004 | * have VM_MAYREAD set? | |
1005 | */ | |
1006 | if (!(vm_flags & VM_MAYREAD)) | |
1007 | return -EFAULT; | |
1008 | } | |
d61172b4 DH |
1009 | /* |
1010 | * gups are always data accesses, not instruction | |
1011 | * fetches, so execute=false here | |
1012 | */ | |
1013 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 1014 | return -EFAULT; |
fa5bb209 KS |
1015 | return 0; |
1016 | } | |
1017 | ||
4bbd4c77 KS |
1018 | /** |
1019 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
1020 | * @mm: mm_struct of target mm |
1021 | * @start: starting user address | |
1022 | * @nr_pages: number of pages from start to pin | |
1023 | * @gup_flags: flags modifying pin behaviour | |
1024 | * @pages: array that receives pointers to the pages pinned. | |
1025 | * Should be at least nr_pages long. Or NULL, if caller | |
1026 | * only intends to ensure the pages are faulted in. | |
1027 | * @vmas: array of pointers to vmas corresponding to each page. | |
1028 | * Or NULL if the caller does not require them. | |
c1e8d7c6 | 1029 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 1030 | * |
d2dfbe47 LX |
1031 | * Returns either number of pages pinned (which may be less than the |
1032 | * number requested), or an error. Details about the return value: | |
1033 | * | |
1034 | * -- If nr_pages is 0, returns 0. | |
1035 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1036 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1037 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 1038 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
1039 | * |
1040 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1041 | * | |
c1e8d7c6 | 1042 | * @vmas are valid only as long as mmap_lock is held. |
4bbd4c77 | 1043 | * |
c1e8d7c6 | 1044 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
1045 | * |
1046 | * __get_user_pages walks a process's page tables and takes a reference to | |
1047 | * each struct page that each user address corresponds to at a given | |
1048 | * instant. That is, it takes the page that would be accessed if a user | |
1049 | * thread accesses the given user virtual address at that instant. | |
1050 | * | |
1051 | * This does not guarantee that the page exists in the user mappings when | |
1052 | * __get_user_pages returns, and there may even be a completely different | |
1053 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
c5acf1f6 | 1054 | * and subsequently re-faulted). However it does guarantee that the page |
4bbd4c77 KS |
1055 | * won't be freed completely. And mostly callers simply care that the page |
1056 | * contains data that was valid *at some point in time*. Typically, an IO | |
1057 | * or similar operation cannot guarantee anything stronger anyway because | |
1058 | * locks can't be held over the syscall boundary. | |
1059 | * | |
1060 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1061 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1062 | * appropriate) must be called after the page is finished with, and | |
1063 | * before put_page is called. | |
1064 | * | |
c1e8d7c6 | 1065 | * If @locked != NULL, *@locked will be set to 0 when mmap_lock is |
4f6da934 PX |
1066 | * released by an up_read(). That can happen if @gup_flags does not |
1067 | * have FOLL_NOWAIT. | |
9a95f3cf | 1068 | * |
4f6da934 | 1069 | * A caller using such a combination of @locked and @gup_flags |
c1e8d7c6 | 1070 | * must therefore hold the mmap_lock for reading only, and recognize |
9a95f3cf PC |
1071 | * when it's been released. Otherwise, it must be held for either |
1072 | * reading or writing and will not be released. | |
4bbd4c77 KS |
1073 | * |
1074 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1075 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1076 | * you need some special @gup_flags. | |
1077 | */ | |
64019a2e | 1078 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
1079 | unsigned long start, unsigned long nr_pages, |
1080 | unsigned int gup_flags, struct page **pages, | |
4f6da934 | 1081 | struct vm_area_struct **vmas, int *locked) |
4bbd4c77 | 1082 | { |
df06b37f | 1083 | long ret = 0, i = 0; |
fa5bb209 | 1084 | struct vm_area_struct *vma = NULL; |
df06b37f | 1085 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
1086 | |
1087 | if (!nr_pages) | |
1088 | return 0; | |
1089 | ||
f9652594 AK |
1090 | start = untagged_addr(start); |
1091 | ||
eddb1c22 | 1092 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 | 1093 | |
4bbd4c77 | 1094 | do { |
fa5bb209 KS |
1095 | struct page *page; |
1096 | unsigned int foll_flags = gup_flags; | |
1097 | unsigned int page_increm; | |
1098 | ||
1099 | /* first iteration or cross vma bound */ | |
1100 | if (!vma || start >= vma->vm_end) { | |
1101 | vma = find_extend_vma(mm, start); | |
1102 | if (!vma && in_gate_area(mm, start)) { | |
fa5bb209 KS |
1103 | ret = get_gate_page(mm, start & PAGE_MASK, |
1104 | gup_flags, &vma, | |
1105 | pages ? &pages[i] : NULL); | |
1106 | if (ret) | |
08be37b7 | 1107 | goto out; |
df06b37f | 1108 | ctx.page_mask = 0; |
fa5bb209 KS |
1109 | goto next_page; |
1110 | } | |
4bbd4c77 | 1111 | |
52650c8b | 1112 | if (!vma) { |
df06b37f KB |
1113 | ret = -EFAULT; |
1114 | goto out; | |
1115 | } | |
52650c8b JG |
1116 | ret = check_vma_flags(vma, gup_flags); |
1117 | if (ret) | |
1118 | goto out; | |
1119 | ||
fa5bb209 KS |
1120 | if (is_vm_hugetlb_page(vma)) { |
1121 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1122 | &start, &nr_pages, i, | |
a308c71b | 1123 | gup_flags, locked); |
ad415db8 PX |
1124 | if (locked && *locked == 0) { |
1125 | /* | |
1126 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1127 | * and we've lost mmap_lock. |
ad415db8 PX |
1128 | * We must stop here. |
1129 | */ | |
1130 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
ad415db8 PX |
1131 | goto out; |
1132 | } | |
fa5bb209 | 1133 | continue; |
4bbd4c77 | 1134 | } |
fa5bb209 KS |
1135 | } |
1136 | retry: | |
1137 | /* | |
1138 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1139 | * potentially allocating memory. | |
1140 | */ | |
fa45f116 | 1141 | if (fatal_signal_pending(current)) { |
d180870d | 1142 | ret = -EINTR; |
df06b37f KB |
1143 | goto out; |
1144 | } | |
fa5bb209 | 1145 | cond_resched(); |
df06b37f KB |
1146 | |
1147 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
a7f22660 DH |
1148 | if (!page || PTR_ERR(page) == -EMLINK) { |
1149 | ret = faultin_page(vma, start, &foll_flags, | |
1150 | PTR_ERR(page) == -EMLINK, locked); | |
fa5bb209 KS |
1151 | switch (ret) { |
1152 | case 0: | |
1153 | goto retry; | |
df06b37f | 1154 | case -EBUSY: |
d9272525 | 1155 | case -EAGAIN: |
df06b37f | 1156 | ret = 0; |
e4a9bc58 | 1157 | fallthrough; |
fa5bb209 KS |
1158 | case -EFAULT: |
1159 | case -ENOMEM: | |
1160 | case -EHWPOISON: | |
df06b37f | 1161 | goto out; |
4bbd4c77 | 1162 | } |
fa5bb209 | 1163 | BUG(); |
1027e443 KS |
1164 | } else if (PTR_ERR(page) == -EEXIST) { |
1165 | /* | |
1166 | * Proper page table entry exists, but no corresponding | |
65462462 JH |
1167 | * struct page. If the caller expects **pages to be |
1168 | * filled in, bail out now, because that can't be done | |
1169 | * for this page. | |
1027e443 | 1170 | */ |
65462462 JH |
1171 | if (pages) { |
1172 | ret = PTR_ERR(page); | |
1173 | goto out; | |
1174 | } | |
1175 | ||
1027e443 KS |
1176 | goto next_page; |
1177 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1178 | ret = PTR_ERR(page); |
1179 | goto out; | |
1027e443 | 1180 | } |
fa5bb209 KS |
1181 | if (pages) { |
1182 | pages[i] = page; | |
1183 | flush_anon_page(vma, page, start); | |
1184 | flush_dcache_page(page); | |
df06b37f | 1185 | ctx.page_mask = 0; |
4bbd4c77 | 1186 | } |
4bbd4c77 | 1187 | next_page: |
fa5bb209 KS |
1188 | if (vmas) { |
1189 | vmas[i] = vma; | |
df06b37f | 1190 | ctx.page_mask = 0; |
fa5bb209 | 1191 | } |
df06b37f | 1192 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1193 | if (page_increm > nr_pages) |
1194 | page_increm = nr_pages; | |
1195 | i += page_increm; | |
1196 | start += page_increm * PAGE_SIZE; | |
1197 | nr_pages -= page_increm; | |
4bbd4c77 | 1198 | } while (nr_pages); |
df06b37f KB |
1199 | out: |
1200 | if (ctx.pgmap) | |
1201 | put_dev_pagemap(ctx.pgmap); | |
1202 | return i ? i : ret; | |
4bbd4c77 | 1203 | } |
4bbd4c77 | 1204 | |
771ab430 TK |
1205 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1206 | unsigned int fault_flags) | |
d4925e00 | 1207 | { |
1b2ee126 DH |
1208 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1209 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1210 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1211 | |
1212 | if (!(vm_flags & vma->vm_flags)) | |
1213 | return false; | |
1214 | ||
33a709b2 DH |
1215 | /* |
1216 | * The architecture might have a hardware protection | |
1b2ee126 | 1217 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1218 | * |
1219 | * gup always represents data access, not instruction | |
1220 | * fetches, so execute=false here: | |
33a709b2 | 1221 | */ |
d61172b4 | 1222 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1223 | return false; |
1224 | ||
d4925e00 DH |
1225 | return true; |
1226 | } | |
1227 | ||
adc8cb40 | 1228 | /** |
4bbd4c77 | 1229 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1230 | * @mm: mm_struct of target mm |
1231 | * @address: user address | |
1232 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1233 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1234 | * does not allow retry. If NULL, the caller must guarantee |
1235 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1236 | * |
1237 | * This is meant to be called in the specific scenario where for locking reasons | |
1238 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1239 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1240 | * trying again. | |
1241 | * | |
1242 | * Typically this is meant to be used by the futex code. | |
1243 | * | |
1244 | * The main difference with get_user_pages() is that this function will | |
1245 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1246 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1247 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1248 | * |
1249 | * This is important for some architectures where those bits also gate the | |
1250 | * access permission to the page because they are maintained in software. On | |
1251 | * such architectures, gup() will not be enough to make a subsequent access | |
1252 | * succeed. | |
1253 | * | |
c1e8d7c6 ML |
1254 | * This function will not return with an unlocked mmap_lock. So it has not the |
1255 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1256 | */ |
64019a2e | 1257 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1258 | unsigned long address, unsigned int fault_flags, |
1259 | bool *unlocked) | |
4bbd4c77 KS |
1260 | { |
1261 | struct vm_area_struct *vma; | |
8fed2f3c | 1262 | vm_fault_t ret; |
4a9e1cda | 1263 | |
f9652594 AK |
1264 | address = untagged_addr(address); |
1265 | ||
4a9e1cda | 1266 | if (unlocked) |
71335f37 | 1267 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1268 | |
4a9e1cda | 1269 | retry: |
4bbd4c77 KS |
1270 | vma = find_extend_vma(mm, address); |
1271 | if (!vma || address < vma->vm_start) | |
1272 | return -EFAULT; | |
1273 | ||
d4925e00 | 1274 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1275 | return -EFAULT; |
1276 | ||
475f4dfc PX |
1277 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1278 | fatal_signal_pending(current)) | |
1279 | return -EINTR; | |
1280 | ||
bce617ed | 1281 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
d9272525 PX |
1282 | |
1283 | if (ret & VM_FAULT_COMPLETED) { | |
1284 | /* | |
1285 | * NOTE: it's a pity that we need to retake the lock here | |
1286 | * to pair with the unlock() in the callers. Ideally we | |
1287 | * could tell the callers so they do not need to unlock. | |
1288 | */ | |
1289 | mmap_read_lock(mm); | |
1290 | *unlocked = true; | |
1291 | return 0; | |
1292 | } | |
1293 | ||
4bbd4c77 | 1294 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1295 | int err = vm_fault_to_errno(ret, 0); |
1296 | ||
1297 | if (err) | |
1298 | return err; | |
4bbd4c77 KS |
1299 | BUG(); |
1300 | } | |
4a9e1cda DD |
1301 | |
1302 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1303 | mmap_read_lock(mm); |
475f4dfc PX |
1304 | *unlocked = true; |
1305 | fault_flags |= FAULT_FLAG_TRIED; | |
1306 | goto retry; | |
4a9e1cda DD |
1307 | } |
1308 | ||
4bbd4c77 KS |
1309 | return 0; |
1310 | } | |
add6a0cd | 1311 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1312 | |
93c5c61d PX |
1313 | /* |
1314 | * GUP always responds to fatal signals. When FOLL_INTERRUPTIBLE is | |
1315 | * specified, it'll also respond to generic signals. The caller of GUP | |
1316 | * that has FOLL_INTERRUPTIBLE should take care of the GUP interruption. | |
1317 | */ | |
1318 | static bool gup_signal_pending(unsigned int flags) | |
1319 | { | |
1320 | if (fatal_signal_pending(current)) | |
1321 | return true; | |
1322 | ||
1323 | if (!(flags & FOLL_INTERRUPTIBLE)) | |
1324 | return false; | |
1325 | ||
1326 | return signal_pending(current); | |
1327 | } | |
1328 | ||
2d3a36a4 | 1329 | /* |
b2a72dff JG |
1330 | * Locking: (*locked == 1) means that the mmap_lock has already been acquired by |
1331 | * the caller. This function may drop the mmap_lock. If it does so, then it will | |
1332 | * set (*locked = 0). | |
1333 | * | |
1334 | * (*locked == 0) means that the caller expects this function to acquire and | |
1335 | * drop the mmap_lock. Therefore, the value of *locked will still be zero when | |
1336 | * the function returns, even though it may have changed temporarily during | |
1337 | * function execution. | |
1338 | * | |
1339 | * Please note that this function, unlike __get_user_pages(), will not return 0 | |
1340 | * for nr_pages > 0, unless FOLL_NOWAIT is used. | |
2d3a36a4 | 1341 | */ |
64019a2e | 1342 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1343 | unsigned long start, |
1344 | unsigned long nr_pages, | |
f0818f47 AA |
1345 | struct page **pages, |
1346 | struct vm_area_struct **vmas, | |
e716712f | 1347 | int *locked, |
0fd71a56 | 1348 | unsigned int flags) |
f0818f47 | 1349 | { |
f0818f47 | 1350 | long ret, pages_done; |
b2a72dff | 1351 | bool must_unlock = false; |
f0818f47 | 1352 | |
b2a72dff JG |
1353 | /* |
1354 | * The internal caller expects GUP to manage the lock internally and the | |
1355 | * lock must be released when this returns. | |
1356 | */ | |
1357 | if (locked && !*locked) { | |
1358 | if (mmap_read_lock_killable(mm)) | |
1359 | return -EAGAIN; | |
1360 | must_unlock = true; | |
1361 | *locked = 1; | |
f0818f47 | 1362 | } |
961ba472 JG |
1363 | else |
1364 | mmap_assert_locked(mm); | |
f0818f47 | 1365 | |
a458b76a AA |
1366 | if (flags & FOLL_PIN) |
1367 | mm_set_has_pinned_flag(&mm->flags); | |
008cfe44 | 1368 | |
eddb1c22 JH |
1369 | /* |
1370 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1371 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1372 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1373 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1374 | * | |
1375 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1376 | * that here, as any failures will be obvious enough. | |
1377 | */ | |
1378 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1379 | flags |= FOLL_GET; |
f0818f47 AA |
1380 | |
1381 | pages_done = 0; | |
f0818f47 | 1382 | for (;;) { |
64019a2e | 1383 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
f0818f47 AA |
1384 | vmas, locked); |
1385 | if (!locked) | |
1386 | /* VM_FAULT_RETRY couldn't trigger, bypass */ | |
1387 | return ret; | |
1388 | ||
d9272525 | 1389 | /* VM_FAULT_RETRY or VM_FAULT_COMPLETED cannot return errors */ |
f0818f47 AA |
1390 | if (!*locked) { |
1391 | BUG_ON(ret < 0); | |
1392 | BUG_ON(ret >= nr_pages); | |
1393 | } | |
1394 | ||
f0818f47 AA |
1395 | if (ret > 0) { |
1396 | nr_pages -= ret; | |
1397 | pages_done += ret; | |
1398 | if (!nr_pages) | |
1399 | break; | |
1400 | } | |
1401 | if (*locked) { | |
96312e61 AA |
1402 | /* |
1403 | * VM_FAULT_RETRY didn't trigger or it was a | |
1404 | * FOLL_NOWAIT. | |
1405 | */ | |
f0818f47 AA |
1406 | if (!pages_done) |
1407 | pages_done = ret; | |
1408 | break; | |
1409 | } | |
df17277b MR |
1410 | /* |
1411 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1412 | * For the prefault case (!pages) we only update counts. | |
1413 | */ | |
1414 | if (likely(pages)) | |
1415 | pages += ret; | |
f0818f47 | 1416 | start += ret << PAGE_SHIFT; |
b2a72dff JG |
1417 | |
1418 | /* The lock was temporarily dropped, so we must unlock later */ | |
1419 | must_unlock = true; | |
f0818f47 | 1420 | |
4426e945 | 1421 | retry: |
f0818f47 AA |
1422 | /* |
1423 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1424 | * with both FAULT_FLAG_ALLOW_RETRY and |
1425 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
93c5c61d PX |
1426 | * by fatal signals of even common signals, depending on |
1427 | * the caller's request. So we need to check it before we | |
4426e945 | 1428 | * start trying again otherwise it can loop forever. |
f0818f47 | 1429 | */ |
93c5c61d | 1430 | if (gup_signal_pending(flags)) { |
ae46d2aa HD |
1431 | if (!pages_done) |
1432 | pages_done = -EINTR; | |
4426e945 | 1433 | break; |
ae46d2aa | 1434 | } |
4426e945 | 1435 | |
d8ed45c5 | 1436 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1437 | if (ret) { |
1438 | BUG_ON(ret > 0); | |
1439 | if (!pages_done) | |
1440 | pages_done = ret; | |
1441 | break; | |
1442 | } | |
4426e945 | 1443 | |
c7b6a566 | 1444 | *locked = 1; |
64019a2e | 1445 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
4426e945 PX |
1446 | pages, NULL, locked); |
1447 | if (!*locked) { | |
1448 | /* Continue to retry until we succeeded */ | |
1449 | BUG_ON(ret != 0); | |
1450 | goto retry; | |
1451 | } | |
f0818f47 AA |
1452 | if (ret != 1) { |
1453 | BUG_ON(ret > 1); | |
1454 | if (!pages_done) | |
1455 | pages_done = ret; | |
1456 | break; | |
1457 | } | |
1458 | nr_pages--; | |
1459 | pages_done++; | |
1460 | if (!nr_pages) | |
1461 | break; | |
df17277b MR |
1462 | if (likely(pages)) |
1463 | pages++; | |
f0818f47 AA |
1464 | start += PAGE_SIZE; |
1465 | } | |
b2a72dff | 1466 | if (must_unlock && *locked) { |
f0818f47 | 1467 | /* |
b2a72dff JG |
1468 | * We either temporarily dropped the lock, or the caller |
1469 | * requested that we both acquire and drop the lock. Either way, | |
1470 | * we must now unlock, and notify the caller of that state. | |
f0818f47 | 1471 | */ |
d8ed45c5 | 1472 | mmap_read_unlock(mm); |
f0818f47 AA |
1473 | *locked = 0; |
1474 | } | |
1475 | return pages_done; | |
1476 | } | |
1477 | ||
d3649f68 CH |
1478 | /** |
1479 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1480 | * @vma: target vma | |
1481 | * @start: start address | |
1482 | * @end: end address | |
c1e8d7c6 | 1483 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1484 | * |
1485 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1486 | * | |
0a36f7f8 TY |
1487 | * Return either number of pages pinned in the vma, or a negative error |
1488 | * code on error. | |
d3649f68 | 1489 | * |
c1e8d7c6 | 1490 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1491 | * |
4f6da934 | 1492 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1493 | * be unperturbed. |
1494 | * | |
4f6da934 PX |
1495 | * If @locked is non-NULL, it must held for read only and may be |
1496 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1497 | */ |
1498 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1499 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1500 | { |
1501 | struct mm_struct *mm = vma->vm_mm; | |
1502 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1503 | int gup_flags; | |
ece369c7 | 1504 | long ret; |
d3649f68 | 1505 | |
be51eb18 ML |
1506 | VM_BUG_ON(!PAGE_ALIGNED(start)); |
1507 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
d3649f68 CH |
1508 | VM_BUG_ON_VMA(start < vma->vm_start, vma); |
1509 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1510 | mmap_assert_locked(mm); |
d3649f68 | 1511 | |
b67bf49c HD |
1512 | /* |
1513 | * Rightly or wrongly, the VM_LOCKONFAULT case has never used | |
1514 | * faultin_page() to break COW, so it has no work to do here. | |
1515 | */ | |
d3649f68 | 1516 | if (vma->vm_flags & VM_LOCKONFAULT) |
b67bf49c HD |
1517 | return nr_pages; |
1518 | ||
1519 | gup_flags = FOLL_TOUCH; | |
d3649f68 CH |
1520 | /* |
1521 | * We want to touch writable mappings with a write fault in order | |
1522 | * to break COW, except for shared mappings because these don't COW | |
1523 | * and we would not want to dirty them for nothing. | |
1524 | */ | |
1525 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1526 | gup_flags |= FOLL_WRITE; | |
1527 | ||
1528 | /* | |
1529 | * We want mlock to succeed for regions that have any permissions | |
1530 | * other than PROT_NONE. | |
1531 | */ | |
3122e80e | 1532 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1533 | gup_flags |= FOLL_FORCE; |
1534 | ||
1535 | /* | |
1536 | * We made sure addr is within a VMA, so the following will | |
1537 | * not result in a stack expansion that recurses back here. | |
1538 | */ | |
ece369c7 | 1539 | ret = __get_user_pages(mm, start, nr_pages, gup_flags, |
4f6da934 | 1540 | NULL, NULL, locked); |
ece369c7 HD |
1541 | lru_add_drain(); |
1542 | return ret; | |
d3649f68 CH |
1543 | } |
1544 | ||
4ca9b385 DH |
1545 | /* |
1546 | * faultin_vma_page_range() - populate (prefault) page tables inside the | |
1547 | * given VMA range readable/writable | |
1548 | * | |
1549 | * This takes care of mlocking the pages, too, if VM_LOCKED is set. | |
1550 | * | |
1551 | * @vma: target vma | |
1552 | * @start: start address | |
1553 | * @end: end address | |
1554 | * @write: whether to prefault readable or writable | |
1555 | * @locked: whether the mmap_lock is still held | |
1556 | * | |
1557 | * Returns either number of processed pages in the vma, or a negative error | |
1558 | * code on error (see __get_user_pages()). | |
1559 | * | |
1560 | * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and | |
1561 | * covered by the VMA. | |
1562 | * | |
1563 | * If @locked is NULL, it may be held for read or write and will be unperturbed. | |
1564 | * | |
1565 | * If @locked is non-NULL, it must held for read only and may be released. If | |
1566 | * it's released, *@locked will be set to 0. | |
1567 | */ | |
1568 | long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start, | |
1569 | unsigned long end, bool write, int *locked) | |
1570 | { | |
1571 | struct mm_struct *mm = vma->vm_mm; | |
1572 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1573 | int gup_flags; | |
ece369c7 | 1574 | long ret; |
4ca9b385 DH |
1575 | |
1576 | VM_BUG_ON(!PAGE_ALIGNED(start)); | |
1577 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
1578 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1579 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
1580 | mmap_assert_locked(mm); | |
1581 | ||
1582 | /* | |
1583 | * FOLL_TOUCH: Mark page accessed and thereby young; will also mark | |
1584 | * the page dirty with FOLL_WRITE -- which doesn't make a | |
1585 | * difference with !FOLL_FORCE, because the page is writable | |
1586 | * in the page table. | |
1587 | * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit | |
1588 | * a poisoned page. | |
4ca9b385 DH |
1589 | * !FOLL_FORCE: Require proper access permissions. |
1590 | */ | |
b67bf49c | 1591 | gup_flags = FOLL_TOUCH | FOLL_HWPOISON; |
4ca9b385 DH |
1592 | if (write) |
1593 | gup_flags |= FOLL_WRITE; | |
1594 | ||
1595 | /* | |
eb2faa51 DH |
1596 | * We want to report -EINVAL instead of -EFAULT for any permission |
1597 | * problems or incompatible mappings. | |
4ca9b385 | 1598 | */ |
eb2faa51 DH |
1599 | if (check_vma_flags(vma, gup_flags)) |
1600 | return -EINVAL; | |
1601 | ||
ece369c7 | 1602 | ret = __get_user_pages(mm, start, nr_pages, gup_flags, |
4ca9b385 | 1603 | NULL, NULL, locked); |
ece369c7 HD |
1604 | lru_add_drain(); |
1605 | return ret; | |
4ca9b385 DH |
1606 | } |
1607 | ||
d3649f68 CH |
1608 | /* |
1609 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1610 | * | |
1611 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1612 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1613 | * mmap_lock must not be held. |
d3649f68 CH |
1614 | */ |
1615 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1616 | { | |
1617 | struct mm_struct *mm = current->mm; | |
1618 | unsigned long end, nstart, nend; | |
1619 | struct vm_area_struct *vma = NULL; | |
1620 | int locked = 0; | |
1621 | long ret = 0; | |
1622 | ||
1623 | end = start + len; | |
1624 | ||
1625 | for (nstart = start; nstart < end; nstart = nend) { | |
1626 | /* | |
1627 | * We want to fault in pages for [nstart; end) address range. | |
1628 | * Find first corresponding VMA. | |
1629 | */ | |
1630 | if (!locked) { | |
1631 | locked = 1; | |
d8ed45c5 | 1632 | mmap_read_lock(mm); |
c4d1a92d | 1633 | vma = find_vma_intersection(mm, nstart, end); |
d3649f68 | 1634 | } else if (nstart >= vma->vm_end) |
c4d1a92d LH |
1635 | vma = find_vma_intersection(mm, vma->vm_end, end); |
1636 | ||
1637 | if (!vma) | |
d3649f68 CH |
1638 | break; |
1639 | /* | |
1640 | * Set [nstart; nend) to intersection of desired address | |
1641 | * range with the first VMA. Also, skip undesirable VMA types. | |
1642 | */ | |
1643 | nend = min(end, vma->vm_end); | |
1644 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1645 | continue; | |
1646 | if (nstart < vma->vm_start) | |
1647 | nstart = vma->vm_start; | |
1648 | /* | |
1649 | * Now fault in a range of pages. populate_vma_page_range() | |
1650 | * double checks the vma flags, so that it won't mlock pages | |
1651 | * if the vma was already munlocked. | |
1652 | */ | |
1653 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1654 | if (ret < 0) { | |
1655 | if (ignore_errors) { | |
1656 | ret = 0; | |
1657 | continue; /* continue at next VMA */ | |
1658 | } | |
1659 | break; | |
1660 | } | |
1661 | nend = nstart + ret * PAGE_SIZE; | |
1662 | ret = 0; | |
1663 | } | |
1664 | if (locked) | |
d8ed45c5 | 1665 | mmap_read_unlock(mm); |
d3649f68 CH |
1666 | return ret; /* 0 or negative error code */ |
1667 | } | |
050a9adc | 1668 | #else /* CONFIG_MMU */ |
64019a2e | 1669 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc CH |
1670 | unsigned long nr_pages, struct page **pages, |
1671 | struct vm_area_struct **vmas, int *locked, | |
1672 | unsigned int foll_flags) | |
1673 | { | |
1674 | struct vm_area_struct *vma; | |
b2a72dff | 1675 | bool must_unlock = false; |
050a9adc | 1676 | unsigned long vm_flags; |
24dc20c7 | 1677 | long i; |
050a9adc | 1678 | |
b2a72dff JG |
1679 | if (!nr_pages) |
1680 | return 0; | |
1681 | ||
1682 | /* | |
1683 | * The internal caller expects GUP to manage the lock internally and the | |
1684 | * lock must be released when this returns. | |
1685 | */ | |
1686 | if (locked && !*locked) { | |
1687 | if (mmap_read_lock_killable(mm)) | |
1688 | return -EAGAIN; | |
1689 | must_unlock = true; | |
1690 | *locked = 1; | |
1691 | } | |
1692 | ||
050a9adc CH |
1693 | /* calculate required read or write permissions. |
1694 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1695 | */ | |
1696 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1697 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1698 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1699 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1700 | ||
1701 | for (i = 0; i < nr_pages; i++) { | |
1702 | vma = find_vma(mm, start); | |
1703 | if (!vma) | |
b2a72dff | 1704 | break; |
050a9adc CH |
1705 | |
1706 | /* protect what we can, including chardevs */ | |
1707 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1708 | !(vm_flags & vma->vm_flags)) | |
b2a72dff | 1709 | break; |
050a9adc CH |
1710 | |
1711 | if (pages) { | |
396a400b | 1712 | pages[i] = virt_to_page((void *)start); |
050a9adc CH |
1713 | if (pages[i]) |
1714 | get_page(pages[i]); | |
1715 | } | |
1716 | if (vmas) | |
1717 | vmas[i] = vma; | |
1718 | start = (start + PAGE_SIZE) & PAGE_MASK; | |
1719 | } | |
1720 | ||
b2a72dff JG |
1721 | if (must_unlock && *locked) { |
1722 | mmap_read_unlock(mm); | |
1723 | *locked = 0; | |
1724 | } | |
050a9adc | 1725 | |
050a9adc CH |
1726 | return i ? : -EFAULT; |
1727 | } | |
1728 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1729 | |
bb523b40 AG |
1730 | /** |
1731 | * fault_in_writeable - fault in userspace address range for writing | |
1732 | * @uaddr: start of address range | |
1733 | * @size: size of address range | |
1734 | * | |
1735 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1736 | * copy_from_user()). | |
1737 | */ | |
1738 | size_t fault_in_writeable(char __user *uaddr, size_t size) | |
1739 | { | |
1740 | char __user *start = uaddr, *end; | |
1741 | ||
1742 | if (unlikely(size == 0)) | |
1743 | return 0; | |
677b2a8c CL |
1744 | if (!user_write_access_begin(uaddr, size)) |
1745 | return size; | |
bb523b40 | 1746 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1747 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1748 | uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1749 | } | |
1750 | end = (char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1751 | if (unlikely(end < start)) | |
1752 | end = NULL; | |
1753 | while (uaddr != end) { | |
677b2a8c | 1754 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1755 | uaddr += PAGE_SIZE; |
1756 | } | |
1757 | ||
1758 | out: | |
677b2a8c | 1759 | user_write_access_end(); |
bb523b40 AG |
1760 | if (size > uaddr - start) |
1761 | return size - (uaddr - start); | |
1762 | return 0; | |
1763 | } | |
1764 | EXPORT_SYMBOL(fault_in_writeable); | |
1765 | ||
da32b581 CM |
1766 | /** |
1767 | * fault_in_subpage_writeable - fault in an address range for writing | |
1768 | * @uaddr: start of address range | |
1769 | * @size: size of address range | |
1770 | * | |
1771 | * Fault in a user address range for writing while checking for permissions at | |
1772 | * sub-page granularity (e.g. arm64 MTE). This function should be used when | |
1773 | * the caller cannot guarantee forward progress of a copy_to_user() loop. | |
1774 | * | |
1775 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1776 | * copy_from_user()). | |
1777 | */ | |
1778 | size_t fault_in_subpage_writeable(char __user *uaddr, size_t size) | |
1779 | { | |
1780 | size_t faulted_in; | |
1781 | ||
1782 | /* | |
1783 | * Attempt faulting in at page granularity first for page table | |
1784 | * permission checking. The arch-specific probe_subpage_writeable() | |
1785 | * functions may not check for this. | |
1786 | */ | |
1787 | faulted_in = size - fault_in_writeable(uaddr, size); | |
1788 | if (faulted_in) | |
1789 | faulted_in -= probe_subpage_writeable(uaddr, faulted_in); | |
1790 | ||
1791 | return size - faulted_in; | |
1792 | } | |
1793 | EXPORT_SYMBOL(fault_in_subpage_writeable); | |
1794 | ||
cdd591fc AG |
1795 | /* |
1796 | * fault_in_safe_writeable - fault in an address range for writing | |
1797 | * @uaddr: start of address range | |
1798 | * @size: length of address range | |
1799 | * | |
fe673d3f LT |
1800 | * Faults in an address range for writing. This is primarily useful when we |
1801 | * already know that some or all of the pages in the address range aren't in | |
1802 | * memory. | |
cdd591fc | 1803 | * |
fe673d3f | 1804 | * Unlike fault_in_writeable(), this function is non-destructive. |
cdd591fc AG |
1805 | * |
1806 | * Note that we don't pin or otherwise hold the pages referenced that we fault | |
1807 | * in. There's no guarantee that they'll stay in memory for any duration of | |
1808 | * time. | |
1809 | * | |
1810 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
1811 | * copy_from_user(). | |
1812 | */ | |
1813 | size_t fault_in_safe_writeable(const char __user *uaddr, size_t size) | |
1814 | { | |
fe673d3f | 1815 | unsigned long start = (unsigned long)uaddr, end; |
cdd591fc | 1816 | struct mm_struct *mm = current->mm; |
fe673d3f | 1817 | bool unlocked = false; |
cdd591fc | 1818 | |
fe673d3f LT |
1819 | if (unlikely(size == 0)) |
1820 | return 0; | |
cdd591fc | 1821 | end = PAGE_ALIGN(start + size); |
fe673d3f | 1822 | if (end < start) |
cdd591fc | 1823 | end = 0; |
cdd591fc | 1824 | |
fe673d3f LT |
1825 | mmap_read_lock(mm); |
1826 | do { | |
1827 | if (fixup_user_fault(mm, start, FAULT_FLAG_WRITE, &unlocked)) | |
cdd591fc | 1828 | break; |
fe673d3f LT |
1829 | start = (start + PAGE_SIZE) & PAGE_MASK; |
1830 | } while (start != end); | |
1831 | mmap_read_unlock(mm); | |
1832 | ||
1833 | if (size > (unsigned long)uaddr - start) | |
1834 | return size - ((unsigned long)uaddr - start); | |
1835 | return 0; | |
cdd591fc AG |
1836 | } |
1837 | EXPORT_SYMBOL(fault_in_safe_writeable); | |
1838 | ||
bb523b40 AG |
1839 | /** |
1840 | * fault_in_readable - fault in userspace address range for reading | |
1841 | * @uaddr: start of user address range | |
1842 | * @size: size of user address range | |
1843 | * | |
1844 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1845 | * copy_from_user()). | |
1846 | */ | |
1847 | size_t fault_in_readable(const char __user *uaddr, size_t size) | |
1848 | { | |
1849 | const char __user *start = uaddr, *end; | |
1850 | volatile char c; | |
1851 | ||
1852 | if (unlikely(size == 0)) | |
1853 | return 0; | |
677b2a8c CL |
1854 | if (!user_read_access_begin(uaddr, size)) |
1855 | return size; | |
bb523b40 | 1856 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1857 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1858 | uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1859 | } | |
1860 | end = (const char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1861 | if (unlikely(end < start)) | |
1862 | end = NULL; | |
1863 | while (uaddr != end) { | |
677b2a8c | 1864 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1865 | uaddr += PAGE_SIZE; |
1866 | } | |
1867 | ||
1868 | out: | |
677b2a8c | 1869 | user_read_access_end(); |
bb523b40 AG |
1870 | (void)c; |
1871 | if (size > uaddr - start) | |
1872 | return size - (uaddr - start); | |
1873 | return 0; | |
1874 | } | |
1875 | EXPORT_SYMBOL(fault_in_readable); | |
1876 | ||
8f942eea JH |
1877 | /** |
1878 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1879 | * @addr: user address | |
1880 | * | |
1881 | * Returns struct page pointer of user page pinned for dump, | |
1882 | * to be freed afterwards by put_page(). | |
1883 | * | |
1884 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1885 | * the corefile, to preserve alignment with its headers; and also returns | |
1886 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
f0953a1b | 1887 | * allowing a hole to be left in the corefile to save disk space. |
8f942eea | 1888 | * |
7f3bfab5 | 1889 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1890 | */ |
1891 | #ifdef CONFIG_ELF_CORE | |
1892 | struct page *get_dump_page(unsigned long addr) | |
1893 | { | |
8f942eea | 1894 | struct page *page; |
b2a72dff | 1895 | int locked = 0; |
7f3bfab5 | 1896 | int ret; |
8f942eea | 1897 | |
b2a72dff JG |
1898 | ret = __get_user_pages_locked(current->mm, addr, 1, &page, NULL, |
1899 | &locked, | |
7f3bfab5 | 1900 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); |
7f3bfab5 | 1901 | return (ret == 1) ? page : NULL; |
8f942eea JH |
1902 | } |
1903 | #endif /* CONFIG_ELF_CORE */ | |
1904 | ||
d1e153fe | 1905 | #ifdef CONFIG_MIGRATION |
f68749ec | 1906 | /* |
67e139b0 | 1907 | * Returns the number of collected pages. Return value is always >= 0. |
f68749ec | 1908 | */ |
67e139b0 AP |
1909 | static unsigned long collect_longterm_unpinnable_pages( |
1910 | struct list_head *movable_page_list, | |
1911 | unsigned long nr_pages, | |
1912 | struct page **pages) | |
9a4e9f3b | 1913 | { |
67e139b0 | 1914 | unsigned long i, collected = 0; |
1b7f7e58 | 1915 | struct folio *prev_folio = NULL; |
67e139b0 | 1916 | bool drain_allow = true; |
9a4e9f3b | 1917 | |
83c02c23 | 1918 | for (i = 0; i < nr_pages; i++) { |
1b7f7e58 | 1919 | struct folio *folio = page_folio(pages[i]); |
f9f38f78 | 1920 | |
1b7f7e58 | 1921 | if (folio == prev_folio) |
83c02c23 | 1922 | continue; |
1b7f7e58 | 1923 | prev_folio = folio; |
f9f38f78 | 1924 | |
67e139b0 AP |
1925 | if (folio_is_longterm_pinnable(folio)) |
1926 | continue; | |
b05a79d4 | 1927 | |
67e139b0 | 1928 | collected++; |
b05a79d4 | 1929 | |
67e139b0 | 1930 | if (folio_is_device_coherent(folio)) |
f9f38f78 CH |
1931 | continue; |
1932 | ||
1b7f7e58 | 1933 | if (folio_test_hugetlb(folio)) { |
67e139b0 | 1934 | isolate_hugetlb(&folio->page, movable_page_list); |
f9f38f78 CH |
1935 | continue; |
1936 | } | |
9a4e9f3b | 1937 | |
1b7f7e58 | 1938 | if (!folio_test_lru(folio) && drain_allow) { |
f9f38f78 CH |
1939 | lru_add_drain_all(); |
1940 | drain_allow = false; | |
1941 | } | |
1942 | ||
67e139b0 | 1943 | if (!folio_isolate_lru(folio)) |
f9f38f78 | 1944 | continue; |
67e139b0 AP |
1945 | |
1946 | list_add_tail(&folio->lru, movable_page_list); | |
1b7f7e58 MWO |
1947 | node_stat_mod_folio(folio, |
1948 | NR_ISOLATED_ANON + folio_is_file_lru(folio), | |
1949 | folio_nr_pages(folio)); | |
9a4e9f3b AK |
1950 | } |
1951 | ||
67e139b0 AP |
1952 | return collected; |
1953 | } | |
1954 | ||
1955 | /* | |
1956 | * Unpins all pages and migrates device coherent pages and movable_page_list. | |
1957 | * Returns -EAGAIN if all pages were successfully migrated or -errno for failure | |
1958 | * (or partial success). | |
1959 | */ | |
1960 | static int migrate_longterm_unpinnable_pages( | |
1961 | struct list_head *movable_page_list, | |
1962 | unsigned long nr_pages, | |
1963 | struct page **pages) | |
1964 | { | |
1965 | int ret; | |
1966 | unsigned long i; | |
6e7f34eb | 1967 | |
b05a79d4 | 1968 | for (i = 0; i < nr_pages; i++) { |
67e139b0 AP |
1969 | struct folio *folio = page_folio(pages[i]); |
1970 | ||
1971 | if (folio_is_device_coherent(folio)) { | |
1972 | /* | |
1973 | * Migration will fail if the page is pinned, so convert | |
1974 | * the pin on the source page to a normal reference. | |
1975 | */ | |
1976 | pages[i] = NULL; | |
1977 | folio_get(folio); | |
1978 | gup_put_folio(folio, 1, FOLL_PIN); | |
1979 | ||
1980 | if (migrate_device_coherent_page(&folio->page)) { | |
1981 | ret = -EBUSY; | |
1982 | goto err; | |
1983 | } | |
1984 | ||
b05a79d4 | 1985 | continue; |
67e139b0 | 1986 | } |
b05a79d4 | 1987 | |
67e139b0 AP |
1988 | /* |
1989 | * We can't migrate pages with unexpected references, so drop | |
1990 | * the reference obtained by __get_user_pages_locked(). | |
1991 | * Migrating pages have been added to movable_page_list after | |
1992 | * calling folio_isolate_lru() which takes a reference so the | |
1993 | * page won't be freed if it's migrating. | |
1994 | */ | |
f6d299ec | 1995 | unpin_user_page(pages[i]); |
67e139b0 | 1996 | pages[i] = NULL; |
f68749ec | 1997 | } |
f9f38f78 | 1998 | |
67e139b0 | 1999 | if (!list_empty(movable_page_list)) { |
f9f38f78 CH |
2000 | struct migration_target_control mtc = { |
2001 | .nid = NUMA_NO_NODE, | |
2002 | .gfp_mask = GFP_USER | __GFP_NOWARN, | |
2003 | }; | |
2004 | ||
67e139b0 AP |
2005 | if (migrate_pages(movable_page_list, alloc_migration_target, |
2006 | NULL, (unsigned long)&mtc, MIGRATE_SYNC, | |
2007 | MR_LONGTERM_PIN, NULL)) { | |
f9f38f78 | 2008 | ret = -ENOMEM; |
67e139b0 AP |
2009 | goto err; |
2010 | } | |
9a4e9f3b AK |
2011 | } |
2012 | ||
67e139b0 AP |
2013 | putback_movable_pages(movable_page_list); |
2014 | ||
2015 | return -EAGAIN; | |
2016 | ||
2017 | err: | |
2018 | for (i = 0; i < nr_pages; i++) | |
2019 | if (pages[i]) | |
2020 | unpin_user_page(pages[i]); | |
2021 | putback_movable_pages(movable_page_list); | |
24a95998 | 2022 | |
67e139b0 AP |
2023 | return ret; |
2024 | } | |
2025 | ||
2026 | /* | |
2027 | * Check whether all pages are *allowed* to be pinned. Rather confusingly, all | |
2028 | * pages in the range are required to be pinned via FOLL_PIN, before calling | |
2029 | * this routine. | |
2030 | * | |
2031 | * If any pages in the range are not allowed to be pinned, then this routine | |
2032 | * will migrate those pages away, unpin all the pages in the range and return | |
2033 | * -EAGAIN. The caller should re-pin the entire range with FOLL_PIN and then | |
2034 | * call this routine again. | |
2035 | * | |
2036 | * If an error other than -EAGAIN occurs, this indicates a migration failure. | |
2037 | * The caller should give up, and propagate the error back up the call stack. | |
2038 | * | |
2039 | * If everything is OK and all pages in the range are allowed to be pinned, then | |
2040 | * this routine leaves all pages pinned and returns zero for success. | |
2041 | */ | |
2042 | static long check_and_migrate_movable_pages(unsigned long nr_pages, | |
2043 | struct page **pages) | |
2044 | { | |
2045 | unsigned long collected; | |
2046 | LIST_HEAD(movable_page_list); | |
2047 | ||
2048 | collected = collect_longterm_unpinnable_pages(&movable_page_list, | |
2049 | nr_pages, pages); | |
2050 | if (!collected) | |
2051 | return 0; | |
2052 | ||
2053 | return migrate_longterm_unpinnable_pages(&movable_page_list, nr_pages, | |
2054 | pages); | |
9a4e9f3b AK |
2055 | } |
2056 | #else | |
f68749ec | 2057 | static long check_and_migrate_movable_pages(unsigned long nr_pages, |
f6d299ec | 2058 | struct page **pages) |
9a4e9f3b | 2059 | { |
24a95998 | 2060 | return 0; |
9a4e9f3b | 2061 | } |
d1e153fe | 2062 | #endif /* CONFIG_MIGRATION */ |
9a4e9f3b | 2063 | |
2bb6d283 | 2064 | /* |
932f4a63 IW |
2065 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
2066 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 2067 | */ |
64019a2e | 2068 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
2069 | unsigned long start, |
2070 | unsigned long nr_pages, | |
2071 | struct page **pages, | |
2072 | struct vm_area_struct **vmas, | |
53b2d09b | 2073 | int *locked, |
932f4a63 | 2074 | unsigned int gup_flags) |
2bb6d283 | 2075 | { |
f68749ec | 2076 | unsigned int flags; |
24a95998 | 2077 | long rc, nr_pinned_pages; |
2bb6d283 | 2078 | |
f68749ec PT |
2079 | if (!(gup_flags & FOLL_LONGTERM)) |
2080 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
53b2d09b | 2081 | locked, gup_flags); |
67e139b0 | 2082 | |
f68749ec PT |
2083 | flags = memalloc_pin_save(); |
2084 | do { | |
24a95998 | 2085 | nr_pinned_pages = __get_user_pages_locked(mm, start, nr_pages, |
53b2d09b | 2086 | pages, vmas, locked, |
24a95998 AP |
2087 | gup_flags); |
2088 | if (nr_pinned_pages <= 0) { | |
2089 | rc = nr_pinned_pages; | |
f68749ec | 2090 | break; |
24a95998 | 2091 | } |
d64e2dbc JG |
2092 | |
2093 | /* FOLL_LONGTERM implies FOLL_PIN */ | |
f6d299ec | 2094 | rc = check_and_migrate_movable_pages(nr_pinned_pages, pages); |
24a95998 | 2095 | } while (rc == -EAGAIN); |
f68749ec | 2096 | memalloc_pin_restore(flags); |
24a95998 | 2097 | return rc ? rc : nr_pinned_pages; |
2bb6d283 | 2098 | } |
932f4a63 | 2099 | |
d64e2dbc JG |
2100 | /* |
2101 | * Check that the given flags are valid for the exported gup/pup interface, and | |
2102 | * update them with the required flags that the caller must have set. | |
2103 | */ | |
2104 | static bool is_valid_gup_args(struct page **pages, struct vm_area_struct **vmas, | |
2105 | int *locked, unsigned int *gup_flags_p, | |
2106 | unsigned int to_set) | |
447f3e45 | 2107 | { |
d64e2dbc JG |
2108 | unsigned int gup_flags = *gup_flags_p; |
2109 | ||
447f3e45 | 2110 | /* |
d64e2dbc JG |
2111 | * These flags not allowed to be specified externally to the gup |
2112 | * interfaces: | |
2113 | * - FOLL_PIN/FOLL_TRIED/FOLL_FAST_ONLY are internal only | |
2114 | * - FOLL_REMOTE is internal only and used on follow_page() | |
447f3e45 | 2115 | */ |
d64e2dbc JG |
2116 | if (WARN_ON_ONCE(gup_flags & (FOLL_PIN | FOLL_TRIED | |
2117 | FOLL_REMOTE | FOLL_FAST_ONLY))) | |
2118 | return false; | |
2119 | ||
2120 | gup_flags |= to_set; | |
2121 | ||
2122 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
2123 | if (WARN_ON_ONCE((gup_flags & (FOLL_PIN | FOLL_GET)) == | |
2124 | (FOLL_PIN | FOLL_GET))) | |
2125 | return false; | |
2126 | ||
2127 | /* LONGTERM can only be specified when pinning */ | |
2128 | if (WARN_ON_ONCE(!(gup_flags & FOLL_PIN) && (gup_flags & FOLL_LONGTERM))) | |
2129 | return false; | |
2130 | ||
2131 | /* Pages input must be given if using GET/PIN */ | |
2132 | if (WARN_ON_ONCE((gup_flags & (FOLL_GET | FOLL_PIN)) && !pages)) | |
447f3e45 | 2133 | return false; |
d64e2dbc JG |
2134 | |
2135 | /* At the external interface locked must be set */ | |
2136 | if (WARN_ON_ONCE(locked && *locked != 1)) | |
2137 | return false; | |
2138 | ||
2139 | /* We want to allow the pgmap to be hot-unplugged at all times */ | |
2140 | if (WARN_ON_ONCE((gup_flags & FOLL_LONGTERM) && | |
2141 | (gup_flags & FOLL_PCI_P2PDMA))) | |
2142 | return false; | |
2143 | ||
447f3e45 | 2144 | /* |
d64e2dbc JG |
2145 | * Can't use VMAs with locked, as locked allows GUP to unlock |
2146 | * which invalidates the vmas array | |
447f3e45 | 2147 | */ |
d64e2dbc | 2148 | if (WARN_ON_ONCE(vmas && locked)) |
447f3e45 BS |
2149 | return false; |
2150 | ||
d64e2dbc | 2151 | *gup_flags_p = gup_flags; |
447f3e45 BS |
2152 | return true; |
2153 | } | |
2154 | ||
22bf29b6 | 2155 | #ifdef CONFIG_MMU |
adc8cb40 | 2156 | /** |
c4237f8b | 2157 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
2158 | * @mm: mm_struct of target mm |
2159 | * @start: starting user address | |
2160 | * @nr_pages: number of pages from start to pin | |
2161 | * @gup_flags: flags modifying lookup behaviour | |
2162 | * @pages: array that receives pointers to the pages pinned. | |
2163 | * Should be at least nr_pages long. Or NULL, if caller | |
2164 | * only intends to ensure the pages are faulted in. | |
2165 | * @vmas: array of pointers to vmas corresponding to each page. | |
2166 | * Or NULL if the caller does not require them. | |
2167 | * @locked: pointer to lock flag indicating whether lock is held and | |
2168 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2169 | * utilised. Lock must initially be held. | |
2170 | * | |
2171 | * Returns either number of pages pinned (which may be less than the | |
2172 | * number requested), or an error. Details about the return value: | |
2173 | * | |
2174 | * -- If nr_pages is 0, returns 0. | |
2175 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
2176 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
2177 | * pages pinned. Again, this may be less than nr_pages. | |
2178 | * | |
2179 | * The caller is responsible for releasing returned @pages, via put_page(). | |
2180 | * | |
c1e8d7c6 | 2181 | * @vmas are valid only as long as mmap_lock is held. |
c4237f8b | 2182 | * |
c1e8d7c6 | 2183 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 2184 | * |
adc8cb40 SJ |
2185 | * get_user_pages_remote walks a process's page tables and takes a reference |
2186 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
2187 | * instant. That is, it takes the page that would be accessed if a user |
2188 | * thread accesses the given user virtual address at that instant. | |
2189 | * | |
2190 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 2191 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b JH |
2192 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
2193 | * and subsequently re faulted). However it does guarantee that the page | |
2194 | * won't be freed completely. And mostly callers simply care that the page | |
2195 | * contains data that was valid *at some point in time*. Typically, an IO | |
2196 | * or similar operation cannot guarantee anything stronger anyway because | |
2197 | * locks can't be held over the syscall boundary. | |
2198 | * | |
2199 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
2200 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
2201 | * be called after the page is finished with, and before put_page is called. | |
2202 | * | |
adc8cb40 SJ |
2203 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
2204 | * to get a handle on the memory by some means other than accesses | |
2205 | * via the user virtual addresses. The pages may be submitted for | |
2206 | * DMA to devices or accessed via their kernel linear mapping (via the | |
2207 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
2208 | * |
2209 | * See also get_user_pages_fast, for performance critical applications. | |
2210 | * | |
adc8cb40 | 2211 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 2212 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 2213 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
2214 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
2215 | */ | |
64019a2e | 2216 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
2217 | unsigned long start, unsigned long nr_pages, |
2218 | unsigned int gup_flags, struct page **pages, | |
2219 | struct vm_area_struct **vmas, int *locked) | |
2220 | { | |
d64e2dbc JG |
2221 | if (!is_valid_gup_args(pages, vmas, locked, &gup_flags, |
2222 | FOLL_TOUCH | FOLL_REMOTE)) | |
eddb1c22 JH |
2223 | return -EINVAL; |
2224 | ||
afa3c33e | 2225 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, locked, |
d64e2dbc | 2226 | gup_flags); |
c4237f8b JH |
2227 | } |
2228 | EXPORT_SYMBOL(get_user_pages_remote); | |
2229 | ||
eddb1c22 | 2230 | #else /* CONFIG_MMU */ |
64019a2e | 2231 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2232 | unsigned long start, unsigned long nr_pages, |
2233 | unsigned int gup_flags, struct page **pages, | |
2234 | struct vm_area_struct **vmas, int *locked) | |
2235 | { | |
2236 | return 0; | |
2237 | } | |
2238 | #endif /* !CONFIG_MMU */ | |
2239 | ||
adc8cb40 SJ |
2240 | /** |
2241 | * get_user_pages() - pin user pages in memory | |
2242 | * @start: starting user address | |
2243 | * @nr_pages: number of pages from start to pin | |
2244 | * @gup_flags: flags modifying lookup behaviour | |
2245 | * @pages: array that receives pointers to the pages pinned. | |
2246 | * Should be at least nr_pages long. Or NULL, if caller | |
2247 | * only intends to ensure the pages are faulted in. | |
2248 | * @vmas: array of pointers to vmas corresponding to each page. | |
2249 | * Or NULL if the caller does not require them. | |
2250 | * | |
64019a2e PX |
2251 | * This is the same as get_user_pages_remote(), just with a less-flexible |
2252 | * calling convention where we assume that the mm being operated on belongs to | |
2253 | * the current task, and doesn't allow passing of a locked parameter. We also | |
2254 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
2255 | */ |
2256 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
2257 | unsigned int gup_flags, struct page **pages, | |
2258 | struct vm_area_struct **vmas) | |
2259 | { | |
d64e2dbc | 2260 | if (!is_valid_gup_args(pages, vmas, NULL, &gup_flags, FOLL_TOUCH)) |
eddb1c22 JH |
2261 | return -EINVAL; |
2262 | ||
afa3c33e | 2263 | return __get_user_pages_locked(current->mm, start, nr_pages, pages, |
d64e2dbc | 2264 | vmas, NULL, gup_flags); |
932f4a63 IW |
2265 | } |
2266 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 2267 | |
acc3c8d1 | 2268 | /* |
d3649f68 | 2269 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 2270 | * |
3e4e28c5 | 2271 | * mmap_read_lock(mm); |
64019a2e | 2272 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2273 | * mmap_read_unlock(mm); |
d3649f68 CH |
2274 | * |
2275 | * with: | |
2276 | * | |
64019a2e | 2277 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
2278 | * |
2279 | * It is functionally equivalent to get_user_pages_fast so | |
2280 | * get_user_pages_fast should be used instead if specific gup_flags | |
2281 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 2282 | */ |
d3649f68 CH |
2283 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
2284 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 | 2285 | { |
b2a72dff | 2286 | int locked = 0; |
acc3c8d1 | 2287 | |
d64e2dbc JG |
2288 | if (!is_valid_gup_args(pages, NULL, NULL, &gup_flags, FOLL_TOUCH)) |
2289 | return -EINVAL; | |
2290 | ||
afa3c33e | 2291 | return __get_user_pages_locked(current->mm, start, nr_pages, pages, |
d64e2dbc | 2292 | NULL, &locked, gup_flags); |
4bbd4c77 | 2293 | } |
d3649f68 | 2294 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
2295 | |
2296 | /* | |
67a929e0 | 2297 | * Fast GUP |
2667f50e SC |
2298 | * |
2299 | * get_user_pages_fast attempts to pin user pages by walking the page | |
2300 | * tables directly and avoids taking locks. Thus the walker needs to be | |
2301 | * protected from page table pages being freed from under it, and should | |
2302 | * block any THP splits. | |
2303 | * | |
2304 | * One way to achieve this is to have the walker disable interrupts, and | |
2305 | * rely on IPIs from the TLB flushing code blocking before the page table | |
2306 | * pages are freed. This is unsuitable for architectures that do not need | |
2307 | * to broadcast an IPI when invalidating TLBs. | |
2308 | * | |
2309 | * Another way to achieve this is to batch up page table containing pages | |
2310 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
2311 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
2312 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
2313 | * (which is a relatively rare event). The code below adopts this strategy. | |
2314 | * | |
2315 | * Before activating this code, please be aware that the following assumptions | |
2316 | * are currently made: | |
2317 | * | |
ff2e6d72 | 2318 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 2319 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 2320 | * |
2667f50e SC |
2321 | * *) ptes can be read atomically by the architecture. |
2322 | * | |
2323 | * *) access_ok is sufficient to validate userspace address ranges. | |
2324 | * | |
2325 | * The last two assumptions can be relaxed by the addition of helper functions. | |
2326 | * | |
2327 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
2328 | */ | |
67a929e0 | 2329 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 2330 | |
790c7369 | 2331 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 2332 | unsigned int flags, |
790c7369 | 2333 | struct page **pages) |
b59f65fa KS |
2334 | { |
2335 | while ((*nr) - nr_start) { | |
2336 | struct page *page = pages[--(*nr)]; | |
2337 | ||
2338 | ClearPageReferenced(page); | |
3faa52c0 JH |
2339 | if (flags & FOLL_PIN) |
2340 | unpin_user_page(page); | |
2341 | else | |
2342 | put_page(page); | |
b59f65fa KS |
2343 | } |
2344 | } | |
2345 | ||
3010a5ea | 2346 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
70cbc3cc YS |
2347 | /* |
2348 | * Fast-gup relies on pte change detection to avoid concurrent pgtable | |
2349 | * operations. | |
2350 | * | |
2351 | * To pin the page, fast-gup needs to do below in order: | |
2352 | * (1) pin the page (by prefetching pte), then (2) check pte not changed. | |
2353 | * | |
2354 | * For the rest of pgtable operations where pgtable updates can be racy | |
2355 | * with fast-gup, we need to do (1) clear pte, then (2) check whether page | |
2356 | * is pinned. | |
2357 | * | |
2358 | * Above will work for all pte-level operations, including THP split. | |
2359 | * | |
2360 | * For THP collapse, it's a bit more complicated because fast-gup may be | |
2361 | * walking a pgtable page that is being freed (pte is still valid but pmd | |
2362 | * can be cleared already). To avoid race in such condition, we need to | |
2363 | * also check pmd here to make sure pmd doesn't change (corresponds to | |
2364 | * pmdp_collapse_flush() in the THP collapse code path). | |
2365 | */ | |
2366 | static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr, | |
2367 | unsigned long end, unsigned int flags, | |
2368 | struct page **pages, int *nr) | |
2667f50e | 2369 | { |
b59f65fa KS |
2370 | struct dev_pagemap *pgmap = NULL; |
2371 | int nr_start = *nr, ret = 0; | |
2667f50e | 2372 | pte_t *ptep, *ptem; |
2667f50e SC |
2373 | |
2374 | ptem = ptep = pte_offset_map(&pmd, addr); | |
2375 | do { | |
2a4a06da | 2376 | pte_t pte = ptep_get_lockless(ptep); |
b0496fe4 MWO |
2377 | struct page *page; |
2378 | struct folio *folio; | |
2667f50e | 2379 | |
0cf45986 | 2380 | if (pte_protnone(pte) && !gup_can_follow_protnone(flags)) |
e7884f8e KS |
2381 | goto pte_unmap; |
2382 | ||
b798bec4 | 2383 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2384 | goto pte_unmap; |
2385 | ||
b59f65fa | 2386 | if (pte_devmap(pte)) { |
7af75561 IW |
2387 | if (unlikely(flags & FOLL_LONGTERM)) |
2388 | goto pte_unmap; | |
2389 | ||
b59f65fa KS |
2390 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2391 | if (unlikely(!pgmap)) { | |
3b78d834 | 2392 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2393 | goto pte_unmap; |
2394 | } | |
2395 | } else if (pte_special(pte)) | |
2667f50e SC |
2396 | goto pte_unmap; |
2397 | ||
2398 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2399 | page = pte_page(pte); | |
2400 | ||
b0496fe4 MWO |
2401 | folio = try_grab_folio(page, 1, flags); |
2402 | if (!folio) | |
2667f50e SC |
2403 | goto pte_unmap; |
2404 | ||
1507f512 | 2405 | if (unlikely(page_is_secretmem(page))) { |
b0496fe4 | 2406 | gup_put_folio(folio, 1, flags); |
1507f512 MR |
2407 | goto pte_unmap; |
2408 | } | |
2409 | ||
70cbc3cc YS |
2410 | if (unlikely(pmd_val(pmd) != pmd_val(*pmdp)) || |
2411 | unlikely(pte_val(pte) != pte_val(*ptep))) { | |
b0496fe4 | 2412 | gup_put_folio(folio, 1, flags); |
2667f50e SC |
2413 | goto pte_unmap; |
2414 | } | |
2415 | ||
84209e87 | 2416 | if (!pte_write(pte) && gup_must_unshare(NULL, flags, page)) { |
a7f22660 DH |
2417 | gup_put_folio(folio, 1, flags); |
2418 | goto pte_unmap; | |
2419 | } | |
2420 | ||
f28d4363 CI |
2421 | /* |
2422 | * We need to make the page accessible if and only if we are | |
2423 | * going to access its content (the FOLL_PIN case). Please | |
2424 | * see Documentation/core-api/pin_user_pages.rst for | |
2425 | * details. | |
2426 | */ | |
2427 | if (flags & FOLL_PIN) { | |
2428 | ret = arch_make_page_accessible(page); | |
2429 | if (ret) { | |
b0496fe4 | 2430 | gup_put_folio(folio, 1, flags); |
f28d4363 CI |
2431 | goto pte_unmap; |
2432 | } | |
2433 | } | |
b0496fe4 | 2434 | folio_set_referenced(folio); |
2667f50e SC |
2435 | pages[*nr] = page; |
2436 | (*nr)++; | |
2667f50e SC |
2437 | } while (ptep++, addr += PAGE_SIZE, addr != end); |
2438 | ||
2439 | ret = 1; | |
2440 | ||
2441 | pte_unmap: | |
832d7aa0 CH |
2442 | if (pgmap) |
2443 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2444 | pte_unmap(ptem); |
2445 | return ret; | |
2446 | } | |
2447 | #else | |
2448 | ||
2449 | /* | |
2450 | * If we can't determine whether or not a pte is special, then fail immediately | |
2451 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2452 | * to be special. | |
2453 | * | |
2454 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2455 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2456 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2457 | */ | |
70cbc3cc YS |
2458 | static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr, |
2459 | unsigned long end, unsigned int flags, | |
2460 | struct page **pages, int *nr) | |
2667f50e SC |
2461 | { |
2462 | return 0; | |
2463 | } | |
3010a5ea | 2464 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2465 | |
17596731 | 2466 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2467 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2468 | unsigned long end, unsigned int flags, |
2469 | struct page **pages, int *nr) | |
b59f65fa KS |
2470 | { |
2471 | int nr_start = *nr; | |
2472 | struct dev_pagemap *pgmap = NULL; | |
2473 | ||
2474 | do { | |
2475 | struct page *page = pfn_to_page(pfn); | |
2476 | ||
2477 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2478 | if (unlikely(!pgmap)) { | |
3b78d834 | 2479 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2480 | break; |
b59f65fa | 2481 | } |
4003f107 LG |
2482 | |
2483 | if (!(flags & FOLL_PCI_P2PDMA) && is_pci_p2pdma_page(page)) { | |
2484 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
2485 | break; | |
2486 | } | |
2487 | ||
b59f65fa KS |
2488 | SetPageReferenced(page); |
2489 | pages[*nr] = page; | |
0f089235 | 2490 | if (unlikely(try_grab_page(page, flags))) { |
3faa52c0 | 2491 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2492 | break; |
3faa52c0 | 2493 | } |
b59f65fa KS |
2494 | (*nr)++; |
2495 | pfn++; | |
2496 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 | 2497 | |
6401c4eb | 2498 | put_dev_pagemap(pgmap); |
20b7fee7 | 2499 | return addr == end; |
b59f65fa KS |
2500 | } |
2501 | ||
a9b6de77 | 2502 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2503 | unsigned long end, unsigned int flags, |
2504 | struct page **pages, int *nr) | |
b59f65fa KS |
2505 | { |
2506 | unsigned long fault_pfn; | |
a9b6de77 DW |
2507 | int nr_start = *nr; |
2508 | ||
2509 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2510 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2511 | return 0; |
b59f65fa | 2512 | |
a9b6de77 | 2513 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2514 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2515 | return 0; |
2516 | } | |
2517 | return 1; | |
b59f65fa KS |
2518 | } |
2519 | ||
a9b6de77 | 2520 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2521 | unsigned long end, unsigned int flags, |
2522 | struct page **pages, int *nr) | |
b59f65fa KS |
2523 | { |
2524 | unsigned long fault_pfn; | |
a9b6de77 DW |
2525 | int nr_start = *nr; |
2526 | ||
2527 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2528 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2529 | return 0; |
b59f65fa | 2530 | |
a9b6de77 | 2531 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2532 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2533 | return 0; |
2534 | } | |
2535 | return 1; | |
b59f65fa KS |
2536 | } |
2537 | #else | |
a9b6de77 | 2538 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2539 | unsigned long end, unsigned int flags, |
2540 | struct page **pages, int *nr) | |
b59f65fa KS |
2541 | { |
2542 | BUILD_BUG(); | |
2543 | return 0; | |
2544 | } | |
2545 | ||
a9b6de77 | 2546 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2547 | unsigned long end, unsigned int flags, |
2548 | struct page **pages, int *nr) | |
b59f65fa KS |
2549 | { |
2550 | BUILD_BUG(); | |
2551 | return 0; | |
2552 | } | |
2553 | #endif | |
2554 | ||
a43e9820 JH |
2555 | static int record_subpages(struct page *page, unsigned long addr, |
2556 | unsigned long end, struct page **pages) | |
2557 | { | |
2558 | int nr; | |
2559 | ||
c228afb1 MWO |
2560 | for (nr = 0; addr != end; nr++, addr += PAGE_SIZE) |
2561 | pages[nr] = nth_page(page, nr); | |
a43e9820 JH |
2562 | |
2563 | return nr; | |
2564 | } | |
2565 | ||
cbd34da7 CH |
2566 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2567 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2568 | unsigned long sz) | |
2569 | { | |
2570 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2571 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2572 | } | |
2573 | ||
2574 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2575 | unsigned long end, unsigned int flags, |
2576 | struct page **pages, int *nr) | |
cbd34da7 CH |
2577 | { |
2578 | unsigned long pte_end; | |
09a1626e MWO |
2579 | struct page *page; |
2580 | struct folio *folio; | |
cbd34da7 CH |
2581 | pte_t pte; |
2582 | int refs; | |
2583 | ||
2584 | pte_end = (addr + sz) & ~(sz-1); | |
2585 | if (pte_end < end) | |
2586 | end = pte_end; | |
2587 | ||
55ca2263 | 2588 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2589 | |
0cd22afd | 2590 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2591 | return 0; |
2592 | ||
2593 | /* hugepages are never "special" */ | |
2594 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2595 | ||
09a1626e | 2596 | page = nth_page(pte_page(pte), (addr & (sz - 1)) >> PAGE_SHIFT); |
a43e9820 | 2597 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2598 | |
09a1626e MWO |
2599 | folio = try_grab_folio(page, refs, flags); |
2600 | if (!folio) | |
cbd34da7 | 2601 | return 0; |
cbd34da7 CH |
2602 | |
2603 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
09a1626e | 2604 | gup_put_folio(folio, refs, flags); |
cbd34da7 CH |
2605 | return 0; |
2606 | } | |
2607 | ||
84209e87 | 2608 | if (!pte_write(pte) && gup_must_unshare(NULL, flags, &folio->page)) { |
a7f22660 DH |
2609 | gup_put_folio(folio, refs, flags); |
2610 | return 0; | |
2611 | } | |
2612 | ||
a43e9820 | 2613 | *nr += refs; |
09a1626e | 2614 | folio_set_referenced(folio); |
cbd34da7 CH |
2615 | return 1; |
2616 | } | |
2617 | ||
2618 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2619 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2620 | struct page **pages, int *nr) |
2621 | { | |
2622 | pte_t *ptep; | |
2623 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2624 | unsigned long next; | |
2625 | ||
2626 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2627 | do { | |
2628 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2629 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2630 | return 0; |
2631 | } while (ptep++, addr = next, addr != end); | |
2632 | ||
2633 | return 1; | |
2634 | } | |
2635 | #else | |
2636 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2637 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2638 | struct page **pages, int *nr) |
2639 | { | |
2640 | return 0; | |
2641 | } | |
2642 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2643 | ||
2667f50e | 2644 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2645 | unsigned long end, unsigned int flags, |
2646 | struct page **pages, int *nr) | |
2667f50e | 2647 | { |
667ed1f7 MWO |
2648 | struct page *page; |
2649 | struct folio *folio; | |
2667f50e SC |
2650 | int refs; |
2651 | ||
b798bec4 | 2652 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2653 | return 0; |
2654 | ||
7af75561 IW |
2655 | if (pmd_devmap(orig)) { |
2656 | if (unlikely(flags & FOLL_LONGTERM)) | |
2657 | return 0; | |
86dfbed4 JH |
2658 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2659 | pages, nr); | |
7af75561 | 2660 | } |
b59f65fa | 2661 | |
c228afb1 | 2662 | page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2663 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2664 | |
667ed1f7 MWO |
2665 | folio = try_grab_folio(page, refs, flags); |
2666 | if (!folio) | |
2667f50e | 2667 | return 0; |
2667f50e SC |
2668 | |
2669 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
667ed1f7 | 2670 | gup_put_folio(folio, refs, flags); |
2667f50e SC |
2671 | return 0; |
2672 | } | |
2673 | ||
84209e87 | 2674 | if (!pmd_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) { |
a7f22660 DH |
2675 | gup_put_folio(folio, refs, flags); |
2676 | return 0; | |
2677 | } | |
2678 | ||
a43e9820 | 2679 | *nr += refs; |
667ed1f7 | 2680 | folio_set_referenced(folio); |
2667f50e SC |
2681 | return 1; |
2682 | } | |
2683 | ||
2684 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2685 | unsigned long end, unsigned int flags, |
2686 | struct page **pages, int *nr) | |
2667f50e | 2687 | { |
83afb52e MWO |
2688 | struct page *page; |
2689 | struct folio *folio; | |
2667f50e SC |
2690 | int refs; |
2691 | ||
b798bec4 | 2692 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2693 | return 0; |
2694 | ||
7af75561 IW |
2695 | if (pud_devmap(orig)) { |
2696 | if (unlikely(flags & FOLL_LONGTERM)) | |
2697 | return 0; | |
86dfbed4 JH |
2698 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2699 | pages, nr); | |
7af75561 | 2700 | } |
b59f65fa | 2701 | |
c228afb1 | 2702 | page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2703 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2704 | |
83afb52e MWO |
2705 | folio = try_grab_folio(page, refs, flags); |
2706 | if (!folio) | |
2667f50e | 2707 | return 0; |
2667f50e SC |
2708 | |
2709 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
83afb52e | 2710 | gup_put_folio(folio, refs, flags); |
2667f50e SC |
2711 | return 0; |
2712 | } | |
2713 | ||
84209e87 | 2714 | if (!pud_write(orig) && gup_must_unshare(NULL, flags, &folio->page)) { |
a7f22660 DH |
2715 | gup_put_folio(folio, refs, flags); |
2716 | return 0; | |
2717 | } | |
2718 | ||
a43e9820 | 2719 | *nr += refs; |
83afb52e | 2720 | folio_set_referenced(folio); |
2667f50e SC |
2721 | return 1; |
2722 | } | |
2723 | ||
f30c59e9 | 2724 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2725 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2726 | struct page **pages, int *nr) |
2727 | { | |
2728 | int refs; | |
2d7919a2 MWO |
2729 | struct page *page; |
2730 | struct folio *folio; | |
f30c59e9 | 2731 | |
b798bec4 | 2732 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2733 | return 0; |
2734 | ||
b59f65fa | 2735 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2736 | |
c228afb1 | 2737 | page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2738 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2739 | |
2d7919a2 MWO |
2740 | folio = try_grab_folio(page, refs, flags); |
2741 | if (!folio) | |
f30c59e9 | 2742 | return 0; |
f30c59e9 AK |
2743 | |
2744 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
2d7919a2 | 2745 | gup_put_folio(folio, refs, flags); |
f30c59e9 AK |
2746 | return 0; |
2747 | } | |
2748 | ||
a43e9820 | 2749 | *nr += refs; |
2d7919a2 | 2750 | folio_set_referenced(folio); |
f30c59e9 AK |
2751 | return 1; |
2752 | } | |
2753 | ||
d3f7b1bb | 2754 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2755 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2756 | { |
2757 | unsigned long next; | |
2758 | pmd_t *pmdp; | |
2759 | ||
d3f7b1bb | 2760 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2761 | do { |
1180e732 | 2762 | pmd_t pmd = pmdp_get_lockless(pmdp); |
2667f50e SC |
2763 | |
2764 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2765 | if (!pmd_present(pmd)) |
2667f50e SC |
2766 | return 0; |
2767 | ||
414fd080 YZ |
2768 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2769 | pmd_devmap(pmd))) { | |
0cf45986 DH |
2770 | if (pmd_protnone(pmd) && |
2771 | !gup_can_follow_protnone(flags)) | |
2667f50e SC |
2772 | return 0; |
2773 | ||
b798bec4 | 2774 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2775 | pages, nr)) |
2776 | return 0; | |
2777 | ||
f30c59e9 AK |
2778 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2779 | /* | |
2780 | * architecture have different format for hugetlbfs | |
2781 | * pmd format and THP pmd format | |
2782 | */ | |
2783 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2784 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2785 | return 0; |
70cbc3cc | 2786 | } else if (!gup_pte_range(pmd, pmdp, addr, next, flags, pages, nr)) |
2923117b | 2787 | return 0; |
2667f50e SC |
2788 | } while (pmdp++, addr = next, addr != end); |
2789 | ||
2790 | return 1; | |
2791 | } | |
2792 | ||
d3f7b1bb | 2793 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2794 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2795 | { |
2796 | unsigned long next; | |
2797 | pud_t *pudp; | |
2798 | ||
d3f7b1bb | 2799 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2800 | do { |
e37c6982 | 2801 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2802 | |
2803 | next = pud_addr_end(addr, end); | |
15494520 | 2804 | if (unlikely(!pud_present(pud))) |
2667f50e | 2805 | return 0; |
fcd0ccd8 | 2806 | if (unlikely(pud_huge(pud) || pud_devmap(pud))) { |
b798bec4 | 2807 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2808 | pages, nr)) |
2809 | return 0; | |
2810 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2811 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2812 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2813 | return 0; |
d3f7b1bb | 2814 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2815 | return 0; |
2816 | } while (pudp++, addr = next, addr != end); | |
2817 | ||
2818 | return 1; | |
2819 | } | |
2820 | ||
d3f7b1bb | 2821 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2822 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2823 | { |
2824 | unsigned long next; | |
2825 | p4d_t *p4dp; | |
2826 | ||
d3f7b1bb | 2827 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
2828 | do { |
2829 | p4d_t p4d = READ_ONCE(*p4dp); | |
2830 | ||
2831 | next = p4d_addr_end(addr, end); | |
2832 | if (p4d_none(p4d)) | |
2833 | return 0; | |
2834 | BUILD_BUG_ON(p4d_huge(p4d)); | |
2835 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
2836 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 2837 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 2838 | return 0; |
d3f7b1bb | 2839 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
2840 | return 0; |
2841 | } while (p4dp++, addr = next, addr != end); | |
2842 | ||
2843 | return 1; | |
2844 | } | |
2845 | ||
5b65c467 | 2846 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 2847 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
2848 | { |
2849 | unsigned long next; | |
2850 | pgd_t *pgdp; | |
2851 | ||
2852 | pgdp = pgd_offset(current->mm, addr); | |
2853 | do { | |
2854 | pgd_t pgd = READ_ONCE(*pgdp); | |
2855 | ||
2856 | next = pgd_addr_end(addr, end); | |
2857 | if (pgd_none(pgd)) | |
2858 | return; | |
2859 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 2860 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
2861 | pages, nr)) |
2862 | return; | |
2863 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
2864 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 2865 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 2866 | return; |
d3f7b1bb | 2867 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
2868 | return; |
2869 | } while (pgdp++, addr = next, addr != end); | |
2870 | } | |
050a9adc CH |
2871 | #else |
2872 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
2873 | unsigned int flags, struct page **pages, int *nr) | |
2874 | { | |
2875 | } | |
2876 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
2877 | |
2878 | #ifndef gup_fast_permitted | |
2879 | /* | |
dadbb612 | 2880 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
2881 | * we need to fall back to the slow version: |
2882 | */ | |
26f4c328 | 2883 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 2884 | { |
26f4c328 | 2885 | return true; |
5b65c467 KS |
2886 | } |
2887 | #endif | |
2888 | ||
c28b1fc7 JG |
2889 | static unsigned long lockless_pages_from_mm(unsigned long start, |
2890 | unsigned long end, | |
2891 | unsigned int gup_flags, | |
2892 | struct page **pages) | |
2893 | { | |
2894 | unsigned long flags; | |
2895 | int nr_pinned = 0; | |
57efa1fe | 2896 | unsigned seq; |
c28b1fc7 JG |
2897 | |
2898 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
2899 | !gup_fast_permitted(start, end)) | |
2900 | return 0; | |
2901 | ||
57efa1fe JG |
2902 | if (gup_flags & FOLL_PIN) { |
2903 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
2904 | if (seq & 1) | |
2905 | return 0; | |
2906 | } | |
2907 | ||
c28b1fc7 JG |
2908 | /* |
2909 | * Disable interrupts. The nested form is used, in order to allow full, | |
2910 | * general purpose use of this routine. | |
2911 | * | |
2912 | * With interrupts disabled, we block page table pages from being freed | |
2913 | * from under us. See struct mmu_table_batch comments in | |
2914 | * include/asm-generic/tlb.h for more details. | |
2915 | * | |
2916 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
2917 | * that come from THPs splitting. | |
2918 | */ | |
2919 | local_irq_save(flags); | |
2920 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
2921 | local_irq_restore(flags); | |
57efa1fe JG |
2922 | |
2923 | /* | |
2924 | * When pinning pages for DMA there could be a concurrent write protect | |
2925 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
2926 | */ | |
2927 | if (gup_flags & FOLL_PIN) { | |
2928 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
b6a2619c | 2929 | unpin_user_pages_lockless(pages, nr_pinned); |
57efa1fe | 2930 | return 0; |
b6a2619c DH |
2931 | } else { |
2932 | sanity_check_pinned_pages(pages, nr_pinned); | |
57efa1fe JG |
2933 | } |
2934 | } | |
c28b1fc7 JG |
2935 | return nr_pinned; |
2936 | } | |
2937 | ||
2938 | static int internal_get_user_pages_fast(unsigned long start, | |
2939 | unsigned long nr_pages, | |
eddb1c22 JH |
2940 | unsigned int gup_flags, |
2941 | struct page **pages) | |
2667f50e | 2942 | { |
c28b1fc7 JG |
2943 | unsigned long len, end; |
2944 | unsigned long nr_pinned; | |
b2a72dff | 2945 | int locked = 0; |
c28b1fc7 | 2946 | int ret; |
2667f50e | 2947 | |
f4000fdf | 2948 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef | 2949 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
4003f107 LG |
2950 | FOLL_FAST_ONLY | FOLL_NOFAULT | |
2951 | FOLL_PCI_P2PDMA))) | |
817be129 CH |
2952 | return -EINVAL; |
2953 | ||
a458b76a AA |
2954 | if (gup_flags & FOLL_PIN) |
2955 | mm_set_has_pinned_flag(¤t->mm->flags); | |
008cfe44 | 2956 | |
f81cd178 | 2957 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 2958 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 2959 | |
f455c854 | 2960 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
2961 | len = nr_pages << PAGE_SHIFT; |
2962 | if (check_add_overflow(start, len, &end)) | |
c61611f7 | 2963 | return 0; |
96d4f267 | 2964 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 2965 | return -EFAULT; |
73e10a61 | 2966 | |
c28b1fc7 JG |
2967 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
2968 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
2969 | return nr_pinned; | |
2667f50e | 2970 | |
c28b1fc7 JG |
2971 | /* Slow path: try to get the remaining pages with get_user_pages */ |
2972 | start += nr_pinned << PAGE_SHIFT; | |
2973 | pages += nr_pinned; | |
b2a72dff JG |
2974 | ret = __gup_longterm_locked(current->mm, start, nr_pages - nr_pinned, |
2975 | pages, NULL, &locked, | |
2976 | gup_flags | FOLL_TOUCH); | |
c28b1fc7 JG |
2977 | if (ret < 0) { |
2978 | /* | |
2979 | * The caller has to unpin the pages we already pinned so | |
2980 | * returning -errno is not an option | |
2981 | */ | |
2982 | if (nr_pinned) | |
2983 | return nr_pinned; | |
2984 | return ret; | |
2667f50e | 2985 | } |
c28b1fc7 | 2986 | return ret + nr_pinned; |
2667f50e | 2987 | } |
c28b1fc7 | 2988 | |
dadbb612 SJ |
2989 | /** |
2990 | * get_user_pages_fast_only() - pin user pages in memory | |
2991 | * @start: starting user address | |
2992 | * @nr_pages: number of pages from start to pin | |
2993 | * @gup_flags: flags modifying pin behaviour | |
2994 | * @pages: array that receives pointers to the pages pinned. | |
2995 | * Should be at least nr_pages long. | |
2996 | * | |
9e1f0580 JH |
2997 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
2998 | * the regular GUP. | |
2999 | * Note a difference with get_user_pages_fast: this always returns the | |
3000 | * number of pages pinned, 0 if no pages were pinned. | |
3001 | * | |
3002 | * If the architecture does not support this function, simply return with no | |
3003 | * pages pinned. | |
3004 | * | |
3005 | * Careful, careful! COW breaking can go either way, so a non-write | |
3006 | * access can get ambiguous page results. If you call this function without | |
3007 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
3008 | */ | |
dadbb612 SJ |
3009 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
3010 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 3011 | { |
376a34ef | 3012 | int nr_pinned; |
9e1f0580 JH |
3013 | /* |
3014 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
3015 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
3016 | * |
3017 | * FOLL_FAST_ONLY is required in order to match the API description of | |
3018 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 3019 | */ |
d64e2dbc JG |
3020 | if (!is_valid_gup_args(pages, NULL, NULL, &gup_flags, |
3021 | FOLL_GET | FOLL_FAST_ONLY)) | |
3022 | return -EINVAL; | |
9e1f0580 | 3023 | |
376a34ef JH |
3024 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
3025 | pages); | |
9e1f0580 JH |
3026 | |
3027 | /* | |
376a34ef JH |
3028 | * As specified in the API description above, this routine is not |
3029 | * allowed to return negative values. However, the common core | |
3030 | * routine internal_get_user_pages_fast() *can* return -errno. | |
3031 | * Therefore, correct for that here: | |
9e1f0580 | 3032 | */ |
376a34ef JH |
3033 | if (nr_pinned < 0) |
3034 | nr_pinned = 0; | |
9e1f0580 JH |
3035 | |
3036 | return nr_pinned; | |
3037 | } | |
dadbb612 | 3038 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 3039 | |
eddb1c22 JH |
3040 | /** |
3041 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
3042 | * @start: starting user address |
3043 | * @nr_pages: number of pages from start to pin | |
3044 | * @gup_flags: flags modifying pin behaviour | |
3045 | * @pages: array that receives pointers to the pages pinned. | |
3046 | * Should be at least nr_pages long. | |
eddb1c22 | 3047 | * |
c1e8d7c6 | 3048 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
3049 | * If not successful, it will fall back to taking the lock and |
3050 | * calling get_user_pages(). | |
3051 | * | |
3052 | * Returns number of pages pinned. This may be fewer than the number requested. | |
3053 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
3054 | * -errno. | |
3055 | */ | |
3056 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
3057 | unsigned int gup_flags, struct page **pages) | |
3058 | { | |
94202f12 JH |
3059 | /* |
3060 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
3061 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
3062 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
3063 | * request. | |
3064 | */ | |
d64e2dbc JG |
3065 | if (!is_valid_gup_args(pages, NULL, NULL, &gup_flags, FOLL_GET)) |
3066 | return -EINVAL; | |
eddb1c22 JH |
3067 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
3068 | } | |
050a9adc | 3069 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
3070 | |
3071 | /** | |
3072 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
3073 | * | |
3faa52c0 JH |
3074 | * @start: starting user address |
3075 | * @nr_pages: number of pages from start to pin | |
3076 | * @gup_flags: flags modifying pin behaviour | |
3077 | * @pages: array that receives pointers to the pages pinned. | |
3078 | * Should be at least nr_pages long. | |
3079 | * | |
3080 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
3081 | * get_user_pages_fast() for documentation on the function arguments, because | |
3082 | * the arguments here are identical. | |
3083 | * | |
3084 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3085 | * see Documentation/core-api/pin_user_pages.rst for further details. |
eddb1c22 JH |
3086 | */ |
3087 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
3088 | unsigned int gup_flags, struct page **pages) | |
3089 | { | |
d64e2dbc | 3090 | if (!is_valid_gup_args(pages, NULL, NULL, &gup_flags, FOLL_PIN)) |
3faa52c0 | 3091 | return -EINVAL; |
3faa52c0 | 3092 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
eddb1c22 JH |
3093 | } |
3094 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
3095 | ||
104acc32 | 3096 | /* |
dadbb612 SJ |
3097 | * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior |
3098 | * is the same, except that this one sets FOLL_PIN instead of FOLL_GET. | |
104acc32 JH |
3099 | * |
3100 | * The API rules are the same, too: no negative values may be returned. | |
3101 | */ | |
3102 | int pin_user_pages_fast_only(unsigned long start, int nr_pages, | |
3103 | unsigned int gup_flags, struct page **pages) | |
3104 | { | |
3105 | int nr_pinned; | |
3106 | ||
104acc32 JH |
3107 | /* |
3108 | * FOLL_FAST_ONLY is required in order to match the API description of | |
3109 | * this routine: no fall back to regular ("slow") GUP. | |
3110 | */ | |
d64e2dbc JG |
3111 | if (!is_valid_gup_args(pages, NULL, NULL, &gup_flags, |
3112 | FOLL_PIN | FOLL_FAST_ONLY)) | |
3113 | return 0; | |
3114 | ||
104acc32 JH |
3115 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
3116 | pages); | |
3117 | /* | |
3118 | * This routine is not allowed to return negative values. However, | |
3119 | * internal_get_user_pages_fast() *can* return -errno. Therefore, | |
3120 | * correct for that here: | |
3121 | */ | |
3122 | if (nr_pinned < 0) | |
3123 | nr_pinned = 0; | |
3124 | ||
3125 | return nr_pinned; | |
3126 | } | |
3127 | EXPORT_SYMBOL_GPL(pin_user_pages_fast_only); | |
3128 | ||
eddb1c22 | 3129 | /** |
64019a2e | 3130 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 3131 | * |
3faa52c0 JH |
3132 | * @mm: mm_struct of target mm |
3133 | * @start: starting user address | |
3134 | * @nr_pages: number of pages from start to pin | |
3135 | * @gup_flags: flags modifying lookup behaviour | |
3136 | * @pages: array that receives pointers to the pages pinned. | |
0768c8de | 3137 | * Should be at least nr_pages long. |
3faa52c0 JH |
3138 | * @vmas: array of pointers to vmas corresponding to each page. |
3139 | * Or NULL if the caller does not require them. | |
3140 | * @locked: pointer to lock flag indicating whether lock is held and | |
3141 | * subsequently whether VM_FAULT_RETRY functionality can be | |
3142 | * utilised. Lock must initially be held. | |
3143 | * | |
3144 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
3145 | * get_user_pages_remote() for documentation on the function arguments, because | |
3146 | * the arguments here are identical. | |
3147 | * | |
3148 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3149 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 | 3150 | */ |
64019a2e | 3151 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
3152 | unsigned long start, unsigned long nr_pages, |
3153 | unsigned int gup_flags, struct page **pages, | |
3154 | struct vm_area_struct **vmas, int *locked) | |
3155 | { | |
d64e2dbc JG |
3156 | if (!is_valid_gup_args(pages, vmas, locked, &gup_flags, |
3157 | FOLL_PIN | FOLL_TOUCH | FOLL_REMOTE)) | |
3158 | return 0; | |
53b2d09b | 3159 | return __gup_longterm_locked(mm, start, nr_pages, pages, vmas, locked, |
d64e2dbc | 3160 | gup_flags); |
eddb1c22 JH |
3161 | } |
3162 | EXPORT_SYMBOL(pin_user_pages_remote); | |
3163 | ||
3164 | /** | |
3165 | * pin_user_pages() - pin user pages in memory for use by other devices | |
3166 | * | |
3faa52c0 JH |
3167 | * @start: starting user address |
3168 | * @nr_pages: number of pages from start to pin | |
3169 | * @gup_flags: flags modifying lookup behaviour | |
3170 | * @pages: array that receives pointers to the pages pinned. | |
0768c8de | 3171 | * Should be at least nr_pages long. |
3faa52c0 JH |
3172 | * @vmas: array of pointers to vmas corresponding to each page. |
3173 | * Or NULL if the caller does not require them. | |
3174 | * | |
3175 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
3176 | * FOLL_PIN is set. | |
3177 | * | |
3178 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3179 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 JH |
3180 | */ |
3181 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
3182 | unsigned int gup_flags, struct page **pages, | |
3183 | struct vm_area_struct **vmas) | |
3184 | { | |
d64e2dbc JG |
3185 | if (!is_valid_gup_args(pages, vmas, NULL, &gup_flags, FOLL_PIN)) |
3186 | return 0; | |
64019a2e | 3187 | return __gup_longterm_locked(current->mm, start, nr_pages, |
53b2d09b | 3188 | pages, vmas, NULL, gup_flags); |
eddb1c22 JH |
3189 | } |
3190 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
3191 | |
3192 | /* | |
3193 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
3194 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
3195 | * FOLL_PIN and rejects FOLL_GET. | |
3196 | */ | |
3197 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
3198 | struct page **pages, unsigned int gup_flags) | |
3199 | { | |
b2a72dff | 3200 | int locked = 0; |
91429023 | 3201 | |
d64e2dbc JG |
3202 | if (!is_valid_gup_args(pages, NULL, NULL, &gup_flags, |
3203 | FOLL_PIN | FOLL_TOUCH)) | |
3204 | return 0; | |
0768c8de | 3205 | |
b2a72dff JG |
3206 | return __gup_longterm_locked(current->mm, start, nr_pages, pages, NULL, |
3207 | &locked, gup_flags); | |
91429023 JH |
3208 | } |
3209 | EXPORT_SYMBOL(pin_user_pages_unlocked); |