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 | ||
c24d3732 JH |
32 | /* Equivalent to calling put_page() @refs times. */ |
33 | static void put_page_refs(struct page *page, int refs) | |
34 | { | |
35 | #ifdef CONFIG_DEBUG_VM | |
36 | if (VM_WARN_ON_ONCE_PAGE(page_ref_count(page) < refs, page)) | |
37 | return; | |
38 | #endif | |
39 | ||
40 | /* | |
41 | * Calling put_page() for each ref is unnecessarily slow. Only the last | |
42 | * ref needs a put_page(). | |
43 | */ | |
44 | if (refs > 1) | |
45 | page_ref_sub(page, refs - 1); | |
46 | put_page(page); | |
47 | } | |
48 | ||
cd1adf1b | 49 | /* |
ece1ed7b | 50 | * Return the folio with ref appropriately incremented, |
cd1adf1b | 51 | * or NULL if that failed. |
a707cdd5 | 52 | */ |
ece1ed7b | 53 | static inline struct folio *try_get_folio(struct page *page, int refs) |
a707cdd5 | 54 | { |
ece1ed7b | 55 | struct folio *folio; |
59409373 MWO |
56 | |
57 | retry: | |
ece1ed7b MWO |
58 | folio = page_folio(page); |
59 | if (WARN_ON_ONCE(folio_ref_count(folio) < 0)) | |
a707cdd5 | 60 | return NULL; |
ece1ed7b | 61 | if (unlikely(!folio_ref_try_add_rcu(folio, refs))) |
a707cdd5 | 62 | return NULL; |
c24d3732 JH |
63 | |
64 | /* | |
ece1ed7b MWO |
65 | * At this point we have a stable reference to the folio; but it |
66 | * could be that between calling page_folio() and the refcount | |
67 | * increment, the folio was split, in which case we'd end up | |
68 | * holding a reference on a folio that has nothing to do with the page | |
c24d3732 | 69 | * we were given anymore. |
ece1ed7b MWO |
70 | * So now that the folio is stable, recheck that the page still |
71 | * belongs to this folio. | |
c24d3732 | 72 | */ |
ece1ed7b MWO |
73 | if (unlikely(page_folio(page) != folio)) { |
74 | folio_put_refs(folio, refs); | |
59409373 | 75 | goto retry; |
c24d3732 JH |
76 | } |
77 | ||
ece1ed7b | 78 | return folio; |
a707cdd5 JH |
79 | } |
80 | ||
3967db22 | 81 | /** |
ece1ed7b | 82 | * try_grab_folio() - Attempt to get or pin a folio. |
3967db22 | 83 | * @page: pointer to page to be grabbed |
ece1ed7b | 84 | * @refs: the value to (effectively) add to the folio's refcount |
3967db22 JH |
85 | * @flags: gup flags: these are the FOLL_* flag values. |
86 | * | |
3faa52c0 | 87 | * "grab" names in this file mean, "look at flags to decide whether to use |
ece1ed7b | 88 | * FOLL_PIN or FOLL_GET behavior, when incrementing the folio's refcount. |
3faa52c0 JH |
89 | * |
90 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
91 | * same time. (That's true throughout the get_user_pages*() and | |
92 | * pin_user_pages*() APIs.) Cases: | |
93 | * | |
ece1ed7b | 94 | * FOLL_GET: folio's refcount will be incremented by @refs. |
3967db22 | 95 | * |
ece1ed7b MWO |
96 | * FOLL_PIN on large folios: folio's refcount will be incremented by |
97 | * @refs, and its compound_pincount will be incremented by @refs. | |
3967db22 | 98 | * |
ece1ed7b | 99 | * FOLL_PIN on single-page folios: folio's refcount will be incremented by |
5232c63f | 100 | * @refs * GUP_PIN_COUNTING_BIAS. |
3faa52c0 | 101 | * |
ece1ed7b MWO |
102 | * Return: The folio containing @page (with refcount appropriately |
103 | * incremented) for success, or NULL upon failure. If neither FOLL_GET | |
104 | * nor FOLL_PIN was set, that's considered failure, and furthermore, | |
105 | * a likely bug in the caller, so a warning is also emitted. | |
3faa52c0 | 106 | */ |
ece1ed7b | 107 | struct folio *try_grab_folio(struct page *page, int refs, unsigned int flags) |
3faa52c0 JH |
108 | { |
109 | if (flags & FOLL_GET) | |
ece1ed7b | 110 | return try_get_folio(page, refs); |
3faa52c0 | 111 | else if (flags & FOLL_PIN) { |
ece1ed7b MWO |
112 | struct folio *folio; |
113 | ||
df3a0a21 | 114 | /* |
d1e153fe PT |
115 | * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a |
116 | * right zone, so fail and let the caller fall back to the slow | |
117 | * path. | |
df3a0a21 | 118 | */ |
d1e153fe PT |
119 | if (unlikely((flags & FOLL_LONGTERM) && |
120 | !is_pinnable_page(page))) | |
df3a0a21 PL |
121 | return NULL; |
122 | ||
c24d3732 JH |
123 | /* |
124 | * CAUTION: Don't use compound_head() on the page before this | |
125 | * point, the result won't be stable. | |
126 | */ | |
ece1ed7b MWO |
127 | folio = try_get_folio(page, refs); |
128 | if (!folio) | |
c24d3732 JH |
129 | return NULL; |
130 | ||
47e29d32 | 131 | /* |
ece1ed7b | 132 | * When pinning a large folio, use an exact count to track it. |
47e29d32 | 133 | * |
ece1ed7b MWO |
134 | * However, be sure to *also* increment the normal folio |
135 | * refcount field at least once, so that the folio really | |
78d9d6ce | 136 | * is pinned. That's why the refcount from the earlier |
ece1ed7b | 137 | * try_get_folio() is left intact. |
47e29d32 | 138 | */ |
ece1ed7b MWO |
139 | if (folio_test_large(folio)) |
140 | atomic_add(refs, folio_pincount_ptr(folio)); | |
c24d3732 | 141 | else |
ece1ed7b MWO |
142 | folio_ref_add(folio, |
143 | refs * (GUP_PIN_COUNTING_BIAS - 1)); | |
144 | node_stat_mod_folio(folio, NR_FOLL_PIN_ACQUIRED, refs); | |
47e29d32 | 145 | |
ece1ed7b | 146 | return folio; |
3faa52c0 JH |
147 | } |
148 | ||
149 | WARN_ON_ONCE(1); | |
150 | return NULL; | |
151 | } | |
152 | ||
ece1ed7b MWO |
153 | struct page *try_grab_compound_head(struct page *page, |
154 | int refs, unsigned int flags) | |
155 | { | |
156 | return &try_grab_folio(page, refs, flags)->page; | |
157 | } | |
158 | ||
4509b42c JG |
159 | static void put_compound_head(struct page *page, int refs, unsigned int flags) |
160 | { | |
6315d8a2 MWO |
161 | VM_BUG_ON_PAGE(PageTail(page), page); |
162 | ||
4509b42c JG |
163 | if (flags & FOLL_PIN) { |
164 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, | |
165 | refs); | |
5232c63f | 166 | if (PageHead(page)) |
6315d8a2 | 167 | atomic_sub(refs, compound_pincount_ptr(page)); |
4509b42c JG |
168 | else |
169 | refs *= GUP_PIN_COUNTING_BIAS; | |
170 | } | |
171 | ||
c24d3732 | 172 | put_page_refs(page, refs); |
4509b42c JG |
173 | } |
174 | ||
3faa52c0 JH |
175 | /** |
176 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
177 | * | |
178 | * This might not do anything at all, depending on the flags argument. | |
179 | * | |
180 | * "grab" names in this file mean, "look at flags to decide whether to use | |
181 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
182 | * | |
183 | * @page: pointer to page to be grabbed | |
184 | * @flags: gup flags: these are the FOLL_* flag values. | |
185 | * | |
186 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same | |
ece1ed7b | 187 | * time. Cases: please see the try_grab_folio() documentation, with |
3967db22 | 188 | * "refs=1". |
3faa52c0 JH |
189 | * |
190 | * Return: true for success, or if no action was required (if neither FOLL_PIN | |
191 | * nor FOLL_GET was set, nothing is done). False for failure: FOLL_GET or | |
192 | * FOLL_PIN was set, but the page could not be grabbed. | |
193 | */ | |
194 | bool __must_check try_grab_page(struct page *page, unsigned int flags) | |
195 | { | |
c36c04c2 | 196 | WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN)); |
3faa52c0 | 197 | |
c36c04c2 JH |
198 | if (flags & FOLL_GET) |
199 | return try_get_page(page); | |
200 | else if (flags & FOLL_PIN) { | |
c36c04c2 JH |
201 | page = compound_head(page); |
202 | ||
203 | if (WARN_ON_ONCE(page_ref_count(page) <= 0)) | |
204 | return false; | |
205 | ||
c36c04c2 | 206 | /* |
78d9d6ce MWO |
207 | * Similar to try_grab_compound_head(): be sure to *also* |
208 | * increment the normal page refcount field at least once, | |
209 | * so that the page really is pinned. | |
c36c04c2 | 210 | */ |
5232c63f | 211 | if (PageHead(page)) { |
8ea2979c | 212 | page_ref_add(page, 1); |
78d9d6ce | 213 | atomic_add(1, compound_pincount_ptr(page)); |
8ea2979c MWO |
214 | } else { |
215 | page_ref_add(page, GUP_PIN_COUNTING_BIAS); | |
216 | } | |
c36c04c2 JH |
217 | |
218 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1); | |
219 | } | |
220 | ||
221 | return true; | |
3faa52c0 JH |
222 | } |
223 | ||
3faa52c0 JH |
224 | /** |
225 | * unpin_user_page() - release a dma-pinned page | |
226 | * @page: pointer to page to be released | |
227 | * | |
228 | * Pages that were pinned via pin_user_pages*() must be released via either | |
229 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
230 | * that such pages can be separately tracked and uniquely handled. In | |
231 | * particular, interactions with RDMA and filesystems need special handling. | |
232 | */ | |
233 | void unpin_user_page(struct page *page) | |
234 | { | |
4509b42c | 235 | put_compound_head(compound_head(page), 1, FOLL_PIN); |
3faa52c0 JH |
236 | } |
237 | EXPORT_SYMBOL(unpin_user_page); | |
238 | ||
8f39f5fc MWO |
239 | static inline struct page *compound_range_next(struct page *start, |
240 | unsigned long npages, unsigned long i, unsigned int *ntails) | |
458a4f78 JM |
241 | { |
242 | struct page *next, *page; | |
243 | unsigned int nr = 1; | |
244 | ||
c228afb1 | 245 | next = nth_page(start, i); |
458a4f78 | 246 | page = compound_head(next); |
0b046e12 | 247 | if (PageHead(page)) |
4c654229 MWO |
248 | nr = min_t(unsigned int, npages - i, |
249 | compound_nr(page) - page_nth(page, next)); | |
458a4f78 | 250 | |
458a4f78 | 251 | *ntails = nr; |
8f39f5fc | 252 | return page; |
458a4f78 JM |
253 | } |
254 | ||
28297dbc MWO |
255 | static inline struct page *compound_next(struct page **list, |
256 | unsigned long npages, unsigned long i, unsigned int *ntails) | |
8745d7f6 JM |
257 | { |
258 | struct page *page; | |
259 | unsigned int nr; | |
260 | ||
8745d7f6 JM |
261 | page = compound_head(list[i]); |
262 | for (nr = i + 1; nr < npages; nr++) { | |
263 | if (compound_head(list[nr]) != page) | |
264 | break; | |
265 | } | |
266 | ||
8745d7f6 | 267 | *ntails = nr - i; |
28297dbc | 268 | return page; |
8745d7f6 JM |
269 | } |
270 | ||
fc1d8e7c | 271 | /** |
f1f6a7dd | 272 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 273 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 274 | * @npages: number of pages in the @pages array. |
2d15eb31 | 275 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
276 | * |
277 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
278 | * variants called on that page. | |
279 | * | |
280 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 281 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
282 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
283 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 284 | * |
f1f6a7dd | 285 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 286 | * |
2d15eb31 | 287 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
288 | * required, then the caller should a) verify that this is really correct, | |
289 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 290 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
291 | * |
292 | */ | |
f1f6a7dd JH |
293 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
294 | bool make_dirty) | |
fc1d8e7c | 295 | { |
2d15eb31 | 296 | unsigned long index; |
31b912de JM |
297 | struct page *head; |
298 | unsigned int ntails; | |
2d15eb31 | 299 | |
300 | if (!make_dirty) { | |
f1f6a7dd | 301 | unpin_user_pages(pages, npages); |
2d15eb31 | 302 | return; |
303 | } | |
304 | ||
e7602748 | 305 | for (index = 0; index < npages; index += ntails) { |
28297dbc | 306 | head = compound_next(pages, npages, index, &ntails); |
2d15eb31 | 307 | /* |
308 | * Checking PageDirty at this point may race with | |
309 | * clear_page_dirty_for_io(), but that's OK. Two key | |
310 | * cases: | |
311 | * | |
312 | * 1) This code sees the page as already dirty, so it | |
313 | * skips the call to set_page_dirty(). That could happen | |
314 | * because clear_page_dirty_for_io() called | |
315 | * page_mkclean(), followed by set_page_dirty(). | |
316 | * However, now the page is going to get written back, | |
317 | * which meets the original intention of setting it | |
318 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
319 | * on to call TestClearPageDirty(), and write the page | |
320 | * back. | |
321 | * | |
322 | * 2) This code sees the page as clean, so it calls | |
323 | * set_page_dirty(). The page stays dirty, despite being | |
324 | * written back, so it gets written back again in the | |
325 | * next writeback cycle. This is harmless. | |
326 | */ | |
31b912de JM |
327 | if (!PageDirty(head)) |
328 | set_page_dirty_lock(head); | |
329 | put_compound_head(head, ntails, FOLL_PIN); | |
2d15eb31 | 330 | } |
fc1d8e7c | 331 | } |
f1f6a7dd | 332 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c | 333 | |
458a4f78 JM |
334 | /** |
335 | * unpin_user_page_range_dirty_lock() - release and optionally dirty | |
336 | * gup-pinned page range | |
337 | * | |
338 | * @page: the starting page of a range maybe marked dirty, and definitely released. | |
339 | * @npages: number of consecutive pages to release. | |
340 | * @make_dirty: whether to mark the pages dirty | |
341 | * | |
342 | * "gup-pinned page range" refers to a range of pages that has had one of the | |
343 | * pin_user_pages() variants called on that page. | |
344 | * | |
345 | * For the page ranges defined by [page .. page+npages], make that range (or | |
346 | * its head pages, if a compound page) dirty, if @make_dirty is true, and if the | |
347 | * page range was previously listed as clean. | |
348 | * | |
349 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is | |
350 | * required, then the caller should a) verify that this is really correct, | |
351 | * because _lock() is usually required, and b) hand code it: | |
352 | * set_page_dirty_lock(), unpin_user_page(). | |
353 | * | |
354 | */ | |
355 | void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, | |
356 | bool make_dirty) | |
357 | { | |
358 | unsigned long index; | |
359 | struct page *head; | |
360 | unsigned int ntails; | |
361 | ||
a5f100db | 362 | for (index = 0; index < npages; index += ntails) { |
8f39f5fc | 363 | head = compound_range_next(page, npages, index, &ntails); |
458a4f78 JM |
364 | if (make_dirty && !PageDirty(head)) |
365 | set_page_dirty_lock(head); | |
366 | put_compound_head(head, ntails, FOLL_PIN); | |
367 | } | |
368 | } | |
369 | EXPORT_SYMBOL(unpin_user_page_range_dirty_lock); | |
370 | ||
fc1d8e7c | 371 | /** |
f1f6a7dd | 372 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
373 | * @pages: array of pages to be marked dirty and released. |
374 | * @npages: number of pages in the @pages array. | |
375 | * | |
f1f6a7dd | 376 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 377 | * |
f1f6a7dd | 378 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 379 | */ |
f1f6a7dd | 380 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c JH |
381 | { |
382 | unsigned long index; | |
31b912de JM |
383 | struct page *head; |
384 | unsigned int ntails; | |
fc1d8e7c | 385 | |
146608bb JH |
386 | /* |
387 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
388 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
389 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
390 | */ | |
391 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
392 | return; | |
31b912de | 393 | |
e7602748 | 394 | for (index = 0; index < npages; index += ntails) { |
28297dbc | 395 | head = compound_next(pages, npages, index, &ntails); |
31b912de | 396 | put_compound_head(head, ntails, FOLL_PIN); |
e7602748 | 397 | } |
fc1d8e7c | 398 | } |
f1f6a7dd | 399 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 400 | |
a458b76a AA |
401 | /* |
402 | * Set the MMF_HAS_PINNED if not set yet; after set it'll be there for the mm's | |
403 | * lifecycle. Avoid setting the bit unless necessary, or it might cause write | |
404 | * cache bouncing on large SMP machines for concurrent pinned gups. | |
405 | */ | |
406 | static inline void mm_set_has_pinned_flag(unsigned long *mm_flags) | |
407 | { | |
408 | if (!test_bit(MMF_HAS_PINNED, mm_flags)) | |
409 | set_bit(MMF_HAS_PINNED, mm_flags); | |
410 | } | |
411 | ||
050a9adc | 412 | #ifdef CONFIG_MMU |
69e68b4f KS |
413 | static struct page *no_page_table(struct vm_area_struct *vma, |
414 | unsigned int flags) | |
4bbd4c77 | 415 | { |
69e68b4f KS |
416 | /* |
417 | * When core dumping an enormous anonymous area that nobody | |
418 | * has touched so far, we don't want to allocate unnecessary pages or | |
419 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
420 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
421 | * But we can only make this optimization where a hole would surely | |
422 | * be zero-filled if handle_mm_fault() actually did handle it. | |
423 | */ | |
a0137f16 AK |
424 | if ((flags & FOLL_DUMP) && |
425 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
426 | return ERR_PTR(-EFAULT); |
427 | return NULL; | |
428 | } | |
4bbd4c77 | 429 | |
1027e443 KS |
430 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
431 | pte_t *pte, unsigned int flags) | |
432 | { | |
433 | /* No page to get reference */ | |
434 | if (flags & FOLL_GET) | |
435 | return -EFAULT; | |
436 | ||
437 | if (flags & FOLL_TOUCH) { | |
438 | pte_t entry = *pte; | |
439 | ||
440 | if (flags & FOLL_WRITE) | |
441 | entry = pte_mkdirty(entry); | |
442 | entry = pte_mkyoung(entry); | |
443 | ||
444 | if (!pte_same(*pte, entry)) { | |
445 | set_pte_at(vma->vm_mm, address, pte, entry); | |
446 | update_mmu_cache(vma, address, pte); | |
447 | } | |
448 | } | |
449 | ||
450 | /* Proper page table entry exists, but no corresponding struct page */ | |
451 | return -EEXIST; | |
452 | } | |
453 | ||
19be0eaf | 454 | /* |
a308c71b PX |
455 | * FOLL_FORCE can write to even unwritable pte's, but only |
456 | * after we've gone through a COW cycle and they are dirty. | |
19be0eaf LT |
457 | */ |
458 | static inline bool can_follow_write_pte(pte_t pte, unsigned int flags) | |
459 | { | |
a308c71b PX |
460 | return pte_write(pte) || |
461 | ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte)); | |
19be0eaf LT |
462 | } |
463 | ||
69e68b4f | 464 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
465 | unsigned long address, pmd_t *pmd, unsigned int flags, |
466 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
467 | { |
468 | struct mm_struct *mm = vma->vm_mm; | |
469 | struct page *page; | |
470 | spinlock_t *ptl; | |
471 | pte_t *ptep, pte; | |
f28d4363 | 472 | int ret; |
4bbd4c77 | 473 | |
eddb1c22 JH |
474 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
475 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
476 | (FOLL_PIN | FOLL_GET))) | |
477 | return ERR_PTR(-EINVAL); | |
69e68b4f | 478 | retry: |
4bbd4c77 | 479 | if (unlikely(pmd_bad(*pmd))) |
69e68b4f | 480 | return no_page_table(vma, flags); |
4bbd4c77 KS |
481 | |
482 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
4bbd4c77 KS |
483 | pte = *ptep; |
484 | if (!pte_present(pte)) { | |
485 | swp_entry_t entry; | |
486 | /* | |
487 | * KSM's break_ksm() relies upon recognizing a ksm page | |
488 | * even while it is being migrated, so for that case we | |
489 | * need migration_entry_wait(). | |
490 | */ | |
491 | if (likely(!(flags & FOLL_MIGRATION))) | |
492 | goto no_page; | |
0661a336 | 493 | if (pte_none(pte)) |
4bbd4c77 KS |
494 | goto no_page; |
495 | entry = pte_to_swp_entry(pte); | |
496 | if (!is_migration_entry(entry)) | |
497 | goto no_page; | |
498 | pte_unmap_unlock(ptep, ptl); | |
499 | migration_entry_wait(mm, pmd, address); | |
69e68b4f | 500 | goto retry; |
4bbd4c77 | 501 | } |
8a0516ed | 502 | if ((flags & FOLL_NUMA) && pte_protnone(pte)) |
4bbd4c77 | 503 | goto no_page; |
19be0eaf | 504 | if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) { |
69e68b4f KS |
505 | pte_unmap_unlock(ptep, ptl); |
506 | return NULL; | |
507 | } | |
4bbd4c77 KS |
508 | |
509 | page = vm_normal_page(vma, address, pte); | |
3faa52c0 | 510 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 511 | /* |
3faa52c0 JH |
512 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
513 | * case since they are only valid while holding the pgmap | |
514 | * reference. | |
3565fce3 | 515 | */ |
df06b37f KB |
516 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
517 | if (*pgmap) | |
3565fce3 DW |
518 | page = pte_page(pte); |
519 | else | |
520 | goto no_page; | |
521 | } else if (unlikely(!page)) { | |
1027e443 KS |
522 | if (flags & FOLL_DUMP) { |
523 | /* Avoid special (like zero) pages in core dumps */ | |
524 | page = ERR_PTR(-EFAULT); | |
525 | goto out; | |
526 | } | |
527 | ||
528 | if (is_zero_pfn(pte_pfn(pte))) { | |
529 | page = pte_page(pte); | |
530 | } else { | |
1027e443 KS |
531 | ret = follow_pfn_pte(vma, address, ptep, flags); |
532 | page = ERR_PTR(ret); | |
533 | goto out; | |
534 | } | |
4bbd4c77 KS |
535 | } |
536 | ||
3faa52c0 JH |
537 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
538 | if (unlikely(!try_grab_page(page, flags))) { | |
539 | page = ERR_PTR(-ENOMEM); | |
540 | goto out; | |
8fde12ca | 541 | } |
f28d4363 CI |
542 | /* |
543 | * We need to make the page accessible if and only if we are going | |
544 | * to access its content (the FOLL_PIN case). Please see | |
545 | * Documentation/core-api/pin_user_pages.rst for details. | |
546 | */ | |
547 | if (flags & FOLL_PIN) { | |
548 | ret = arch_make_page_accessible(page); | |
549 | if (ret) { | |
550 | unpin_user_page(page); | |
551 | page = ERR_PTR(ret); | |
552 | goto out; | |
553 | } | |
554 | } | |
4bbd4c77 KS |
555 | if (flags & FOLL_TOUCH) { |
556 | if ((flags & FOLL_WRITE) && | |
557 | !pte_dirty(pte) && !PageDirty(page)) | |
558 | set_page_dirty(page); | |
559 | /* | |
560 | * pte_mkyoung() would be more correct here, but atomic care | |
561 | * is needed to avoid losing the dirty bit: it is easier to use | |
562 | * mark_page_accessed(). | |
563 | */ | |
564 | mark_page_accessed(page); | |
565 | } | |
1027e443 | 566 | out: |
4bbd4c77 | 567 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 568 | return page; |
4bbd4c77 KS |
569 | no_page: |
570 | pte_unmap_unlock(ptep, ptl); | |
571 | if (!pte_none(pte)) | |
69e68b4f KS |
572 | return NULL; |
573 | return no_page_table(vma, flags); | |
574 | } | |
575 | ||
080dbb61 AK |
576 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
577 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
578 | unsigned int flags, |
579 | struct follow_page_context *ctx) | |
69e68b4f | 580 | { |
68827280 | 581 | pmd_t *pmd, pmdval; |
69e68b4f KS |
582 | spinlock_t *ptl; |
583 | struct page *page; | |
584 | struct mm_struct *mm = vma->vm_mm; | |
585 | ||
080dbb61 | 586 | pmd = pmd_offset(pudp, address); |
68827280 HY |
587 | /* |
588 | * The READ_ONCE() will stabilize the pmdval in a register or | |
589 | * on the stack so that it will stop changing under the code. | |
590 | */ | |
591 | pmdval = READ_ONCE(*pmd); | |
592 | if (pmd_none(pmdval)) | |
69e68b4f | 593 | return no_page_table(vma, flags); |
be9d3045 | 594 | if (pmd_huge(pmdval) && is_vm_hugetlb_page(vma)) { |
e66f17ff NH |
595 | page = follow_huge_pmd(mm, address, pmd, flags); |
596 | if (page) | |
597 | return page; | |
598 | return no_page_table(vma, flags); | |
69e68b4f | 599 | } |
68827280 | 600 | if (is_hugepd(__hugepd(pmd_val(pmdval)))) { |
4dc71451 | 601 | page = follow_huge_pd(vma, address, |
68827280 | 602 | __hugepd(pmd_val(pmdval)), flags, |
4dc71451 AK |
603 | PMD_SHIFT); |
604 | if (page) | |
605 | return page; | |
606 | return no_page_table(vma, flags); | |
607 | } | |
84c3fc4e | 608 | retry: |
68827280 | 609 | if (!pmd_present(pmdval)) { |
28b0ee3f LX |
610 | /* |
611 | * Should never reach here, if thp migration is not supported; | |
612 | * Otherwise, it must be a thp migration entry. | |
613 | */ | |
614 | VM_BUG_ON(!thp_migration_supported() || | |
615 | !is_pmd_migration_entry(pmdval)); | |
616 | ||
84c3fc4e ZY |
617 | if (likely(!(flags & FOLL_MIGRATION))) |
618 | return no_page_table(vma, flags); | |
28b0ee3f LX |
619 | |
620 | pmd_migration_entry_wait(mm, pmd); | |
68827280 HY |
621 | pmdval = READ_ONCE(*pmd); |
622 | /* | |
623 | * MADV_DONTNEED may convert the pmd to null because | |
c1e8d7c6 | 624 | * mmap_lock is held in read mode |
68827280 HY |
625 | */ |
626 | if (pmd_none(pmdval)) | |
627 | return no_page_table(vma, flags); | |
84c3fc4e ZY |
628 | goto retry; |
629 | } | |
68827280 | 630 | if (pmd_devmap(pmdval)) { |
3565fce3 | 631 | ptl = pmd_lock(mm, pmd); |
df06b37f | 632 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
633 | spin_unlock(ptl); |
634 | if (page) | |
635 | return page; | |
636 | } | |
68827280 | 637 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 638 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 639 | |
68827280 | 640 | if ((flags & FOLL_NUMA) && pmd_protnone(pmdval)) |
db08f203 AK |
641 | return no_page_table(vma, flags); |
642 | ||
84c3fc4e | 643 | retry_locked: |
6742d293 | 644 | ptl = pmd_lock(mm, pmd); |
68827280 HY |
645 | if (unlikely(pmd_none(*pmd))) { |
646 | spin_unlock(ptl); | |
647 | return no_page_table(vma, flags); | |
648 | } | |
84c3fc4e ZY |
649 | if (unlikely(!pmd_present(*pmd))) { |
650 | spin_unlock(ptl); | |
651 | if (likely(!(flags & FOLL_MIGRATION))) | |
652 | return no_page_table(vma, flags); | |
653 | pmd_migration_entry_wait(mm, pmd); | |
654 | goto retry_locked; | |
655 | } | |
6742d293 KS |
656 | if (unlikely(!pmd_trans_huge(*pmd))) { |
657 | spin_unlock(ptl); | |
df06b37f | 658 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 659 | } |
4066c119 | 660 | if (flags & FOLL_SPLIT_PMD) { |
6742d293 KS |
661 | int ret; |
662 | page = pmd_page(*pmd); | |
663 | if (is_huge_zero_page(page)) { | |
664 | spin_unlock(ptl); | |
665 | ret = 0; | |
78ddc534 | 666 | split_huge_pmd(vma, pmd, address); |
337d9abf NH |
667 | if (pmd_trans_unstable(pmd)) |
668 | ret = -EBUSY; | |
4066c119 | 669 | } else { |
bfe7b00d SL |
670 | spin_unlock(ptl); |
671 | split_huge_pmd(vma, pmd, address); | |
672 | ret = pte_alloc(mm, pmd) ? -ENOMEM : 0; | |
6742d293 KS |
673 | } |
674 | ||
675 | return ret ? ERR_PTR(ret) : | |
df06b37f | 676 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 677 | } |
6742d293 KS |
678 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
679 | spin_unlock(ptl); | |
df06b37f | 680 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 681 | return page; |
4bbd4c77 KS |
682 | } |
683 | ||
080dbb61 AK |
684 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
685 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
686 | unsigned int flags, |
687 | struct follow_page_context *ctx) | |
080dbb61 AK |
688 | { |
689 | pud_t *pud; | |
690 | spinlock_t *ptl; | |
691 | struct page *page; | |
692 | struct mm_struct *mm = vma->vm_mm; | |
693 | ||
694 | pud = pud_offset(p4dp, address); | |
695 | if (pud_none(*pud)) | |
696 | return no_page_table(vma, flags); | |
be9d3045 | 697 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) { |
080dbb61 AK |
698 | page = follow_huge_pud(mm, address, pud, flags); |
699 | if (page) | |
700 | return page; | |
701 | return no_page_table(vma, flags); | |
702 | } | |
4dc71451 AK |
703 | if (is_hugepd(__hugepd(pud_val(*pud)))) { |
704 | page = follow_huge_pd(vma, address, | |
705 | __hugepd(pud_val(*pud)), flags, | |
706 | PUD_SHIFT); | |
707 | if (page) | |
708 | return page; | |
709 | return no_page_table(vma, flags); | |
710 | } | |
080dbb61 AK |
711 | if (pud_devmap(*pud)) { |
712 | ptl = pud_lock(mm, pud); | |
df06b37f | 713 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
714 | spin_unlock(ptl); |
715 | if (page) | |
716 | return page; | |
717 | } | |
718 | if (unlikely(pud_bad(*pud))) | |
719 | return no_page_table(vma, flags); | |
720 | ||
df06b37f | 721 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
722 | } |
723 | ||
080dbb61 AK |
724 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
725 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
726 | unsigned int flags, |
727 | struct follow_page_context *ctx) | |
080dbb61 AK |
728 | { |
729 | p4d_t *p4d; | |
4dc71451 | 730 | struct page *page; |
080dbb61 AK |
731 | |
732 | p4d = p4d_offset(pgdp, address); | |
733 | if (p4d_none(*p4d)) | |
734 | return no_page_table(vma, flags); | |
735 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
736 | if (unlikely(p4d_bad(*p4d))) | |
737 | return no_page_table(vma, flags); | |
738 | ||
4dc71451 AK |
739 | if (is_hugepd(__hugepd(p4d_val(*p4d)))) { |
740 | page = follow_huge_pd(vma, address, | |
741 | __hugepd(p4d_val(*p4d)), flags, | |
742 | P4D_SHIFT); | |
743 | if (page) | |
744 | return page; | |
745 | return no_page_table(vma, flags); | |
746 | } | |
df06b37f | 747 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
748 | } |
749 | ||
750 | /** | |
751 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
752 | * @vma: vm_area_struct mapping @address | |
753 | * @address: virtual address to look up | |
754 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
755 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
756 | * pointer to output page_mask | |
080dbb61 AK |
757 | * |
758 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
759 | * | |
78179556 MR |
760 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
761 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
762 | * | |
763 | * On output, the @ctx->page_mask is set according to the size of the page. | |
764 | * | |
765 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
766 | * an error pointer if there is a mapping to something not represented |
767 | * by a page descriptor (see also vm_normal_page()). | |
768 | */ | |
a7030aea | 769 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 770 | unsigned long address, unsigned int flags, |
df06b37f | 771 | struct follow_page_context *ctx) |
080dbb61 AK |
772 | { |
773 | pgd_t *pgd; | |
774 | struct page *page; | |
775 | struct mm_struct *mm = vma->vm_mm; | |
776 | ||
df06b37f | 777 | ctx->page_mask = 0; |
080dbb61 AK |
778 | |
779 | /* make this handle hugepd */ | |
780 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); | |
781 | if (!IS_ERR(page)) { | |
3faa52c0 | 782 | WARN_ON_ONCE(flags & (FOLL_GET | FOLL_PIN)); |
080dbb61 AK |
783 | return page; |
784 | } | |
785 | ||
786 | pgd = pgd_offset(mm, address); | |
787 | ||
788 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
789 | return no_page_table(vma, flags); | |
790 | ||
faaa5b62 AK |
791 | if (pgd_huge(*pgd)) { |
792 | page = follow_huge_pgd(mm, address, pgd, flags); | |
793 | if (page) | |
794 | return page; | |
795 | return no_page_table(vma, flags); | |
796 | } | |
4dc71451 AK |
797 | if (is_hugepd(__hugepd(pgd_val(*pgd)))) { |
798 | page = follow_huge_pd(vma, address, | |
799 | __hugepd(pgd_val(*pgd)), flags, | |
800 | PGDIR_SHIFT); | |
801 | if (page) | |
802 | return page; | |
803 | return no_page_table(vma, flags); | |
804 | } | |
faaa5b62 | 805 | |
df06b37f KB |
806 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
807 | } | |
808 | ||
809 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
810 | unsigned int foll_flags) | |
811 | { | |
812 | struct follow_page_context ctx = { NULL }; | |
813 | struct page *page; | |
814 | ||
1507f512 MR |
815 | if (vma_is_secretmem(vma)) |
816 | return NULL; | |
817 | ||
df06b37f KB |
818 | page = follow_page_mask(vma, address, foll_flags, &ctx); |
819 | if (ctx.pgmap) | |
820 | put_dev_pagemap(ctx.pgmap); | |
821 | return page; | |
080dbb61 AK |
822 | } |
823 | ||
f2b495ca KS |
824 | static int get_gate_page(struct mm_struct *mm, unsigned long address, |
825 | unsigned int gup_flags, struct vm_area_struct **vma, | |
826 | struct page **page) | |
827 | { | |
828 | pgd_t *pgd; | |
c2febafc | 829 | p4d_t *p4d; |
f2b495ca KS |
830 | pud_t *pud; |
831 | pmd_t *pmd; | |
832 | pte_t *pte; | |
833 | int ret = -EFAULT; | |
834 | ||
835 | /* user gate pages are read-only */ | |
836 | if (gup_flags & FOLL_WRITE) | |
837 | return -EFAULT; | |
838 | if (address > TASK_SIZE) | |
839 | pgd = pgd_offset_k(address); | |
840 | else | |
841 | pgd = pgd_offset_gate(mm, address); | |
b5d1c39f AL |
842 | if (pgd_none(*pgd)) |
843 | return -EFAULT; | |
c2febafc | 844 | p4d = p4d_offset(pgd, address); |
b5d1c39f AL |
845 | if (p4d_none(*p4d)) |
846 | return -EFAULT; | |
c2febafc | 847 | pud = pud_offset(p4d, address); |
b5d1c39f AL |
848 | if (pud_none(*pud)) |
849 | return -EFAULT; | |
f2b495ca | 850 | pmd = pmd_offset(pud, address); |
84c3fc4e | 851 | if (!pmd_present(*pmd)) |
f2b495ca KS |
852 | return -EFAULT; |
853 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
854 | pte = pte_offset_map(pmd, address); | |
855 | if (pte_none(*pte)) | |
856 | goto unmap; | |
857 | *vma = get_gate_vma(mm); | |
858 | if (!page) | |
859 | goto out; | |
860 | *page = vm_normal_page(*vma, address, *pte); | |
861 | if (!*page) { | |
862 | if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte))) | |
863 | goto unmap; | |
864 | *page = pte_page(*pte); | |
865 | } | |
9fa2dd94 | 866 | if (unlikely(!try_grab_page(*page, gup_flags))) { |
8fde12ca LT |
867 | ret = -ENOMEM; |
868 | goto unmap; | |
869 | } | |
f2b495ca KS |
870 | out: |
871 | ret = 0; | |
872 | unmap: | |
873 | pte_unmap(pte); | |
874 | return ret; | |
875 | } | |
876 | ||
9a95f3cf | 877 | /* |
c1e8d7c6 ML |
878 | * mmap_lock must be held on entry. If @locked != NULL and *@flags |
879 | * does not include FOLL_NOWAIT, the mmap_lock may be released. If it | |
4f6da934 | 880 | * is, *@locked will be set to 0 and -EBUSY returned. |
9a95f3cf | 881 | */ |
64019a2e | 882 | static int faultin_page(struct vm_area_struct *vma, |
4f6da934 | 883 | unsigned long address, unsigned int *flags, int *locked) |
16744483 | 884 | { |
16744483 | 885 | unsigned int fault_flags = 0; |
2b740303 | 886 | vm_fault_t ret; |
16744483 | 887 | |
55b8fe70 AG |
888 | if (*flags & FOLL_NOFAULT) |
889 | return -EFAULT; | |
16744483 KS |
890 | if (*flags & FOLL_WRITE) |
891 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
892 | if (*flags & FOLL_REMOTE) |
893 | fault_flags |= FAULT_FLAG_REMOTE; | |
4f6da934 | 894 | if (locked) |
71335f37 | 895 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
16744483 KS |
896 | if (*flags & FOLL_NOWAIT) |
897 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 898 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
899 | /* |
900 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
901 | * can co-exist | |
902 | */ | |
234b239b ALC |
903 | fault_flags |= FAULT_FLAG_TRIED; |
904 | } | |
16744483 | 905 | |
bce617ed | 906 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
16744483 | 907 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
908 | int err = vm_fault_to_errno(ret, *flags); |
909 | ||
910 | if (err) | |
911 | return err; | |
16744483 KS |
912 | BUG(); |
913 | } | |
914 | ||
16744483 | 915 | if (ret & VM_FAULT_RETRY) { |
4f6da934 PX |
916 | if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
917 | *locked = 0; | |
16744483 KS |
918 | return -EBUSY; |
919 | } | |
920 | ||
921 | /* | |
922 | * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when | |
923 | * necessary, even if maybe_mkwrite decided not to set pte_write. We | |
924 | * can thus safely do subsequent page lookups as if they were reads. | |
925 | * But only do so when looping for pte_write is futile: in some cases | |
926 | * userspace may also be wanting to write to the gotten user page, | |
927 | * which a read fault here might prevent (a readonly page might get | |
928 | * reCOWed by userspace write). | |
929 | */ | |
930 | if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE)) | |
2923117b | 931 | *flags |= FOLL_COW; |
16744483 KS |
932 | return 0; |
933 | } | |
934 | ||
fa5bb209 KS |
935 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
936 | { | |
937 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
938 | int write = (gup_flags & FOLL_WRITE); |
939 | int foreign = (gup_flags & FOLL_REMOTE); | |
fa5bb209 KS |
940 | |
941 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
942 | return -EFAULT; | |
943 | ||
7f7ccc2c WT |
944 | if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma)) |
945 | return -EFAULT; | |
946 | ||
52650c8b JG |
947 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
948 | return -EOPNOTSUPP; | |
949 | ||
1507f512 MR |
950 | if (vma_is_secretmem(vma)) |
951 | return -EFAULT; | |
952 | ||
1b2ee126 | 953 | if (write) { |
fa5bb209 KS |
954 | if (!(vm_flags & VM_WRITE)) { |
955 | if (!(gup_flags & FOLL_FORCE)) | |
956 | return -EFAULT; | |
957 | /* | |
958 | * We used to let the write,force case do COW in a | |
959 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
960 | * set a breakpoint in a read-only mapping of an | |
961 | * executable, without corrupting the file (yet only | |
962 | * when that file had been opened for writing!). | |
963 | * Anon pages in shared mappings are surprising: now | |
964 | * just reject it. | |
965 | */ | |
46435364 | 966 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 967 | return -EFAULT; |
fa5bb209 KS |
968 | } |
969 | } else if (!(vm_flags & VM_READ)) { | |
970 | if (!(gup_flags & FOLL_FORCE)) | |
971 | return -EFAULT; | |
972 | /* | |
973 | * Is there actually any vma we can reach here which does not | |
974 | * have VM_MAYREAD set? | |
975 | */ | |
976 | if (!(vm_flags & VM_MAYREAD)) | |
977 | return -EFAULT; | |
978 | } | |
d61172b4 DH |
979 | /* |
980 | * gups are always data accesses, not instruction | |
981 | * fetches, so execute=false here | |
982 | */ | |
983 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 984 | return -EFAULT; |
fa5bb209 KS |
985 | return 0; |
986 | } | |
987 | ||
4bbd4c77 KS |
988 | /** |
989 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
990 | * @mm: mm_struct of target mm |
991 | * @start: starting user address | |
992 | * @nr_pages: number of pages from start to pin | |
993 | * @gup_flags: flags modifying pin behaviour | |
994 | * @pages: array that receives pointers to the pages pinned. | |
995 | * Should be at least nr_pages long. Or NULL, if caller | |
996 | * only intends to ensure the pages are faulted in. | |
997 | * @vmas: array of pointers to vmas corresponding to each page. | |
998 | * Or NULL if the caller does not require them. | |
c1e8d7c6 | 999 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 1000 | * |
d2dfbe47 LX |
1001 | * Returns either number of pages pinned (which may be less than the |
1002 | * number requested), or an error. Details about the return value: | |
1003 | * | |
1004 | * -- If nr_pages is 0, returns 0. | |
1005 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1006 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1007 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 1008 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
1009 | * |
1010 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1011 | * | |
c1e8d7c6 | 1012 | * @vmas are valid only as long as mmap_lock is held. |
4bbd4c77 | 1013 | * |
c1e8d7c6 | 1014 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
1015 | * |
1016 | * __get_user_pages walks a process's page tables and takes a reference to | |
1017 | * each struct page that each user address corresponds to at a given | |
1018 | * instant. That is, it takes the page that would be accessed if a user | |
1019 | * thread accesses the given user virtual address at that instant. | |
1020 | * | |
1021 | * This does not guarantee that the page exists in the user mappings when | |
1022 | * __get_user_pages returns, and there may even be a completely different | |
1023 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1024 | * and subsequently re faulted). However it does guarantee that the page | |
1025 | * won't be freed completely. And mostly callers simply care that the page | |
1026 | * contains data that was valid *at some point in time*. Typically, an IO | |
1027 | * or similar operation cannot guarantee anything stronger anyway because | |
1028 | * locks can't be held over the syscall boundary. | |
1029 | * | |
1030 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1031 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1032 | * appropriate) must be called after the page is finished with, and | |
1033 | * before put_page is called. | |
1034 | * | |
c1e8d7c6 | 1035 | * If @locked != NULL, *@locked will be set to 0 when mmap_lock is |
4f6da934 PX |
1036 | * released by an up_read(). That can happen if @gup_flags does not |
1037 | * have FOLL_NOWAIT. | |
9a95f3cf | 1038 | * |
4f6da934 | 1039 | * A caller using such a combination of @locked and @gup_flags |
c1e8d7c6 | 1040 | * must therefore hold the mmap_lock for reading only, and recognize |
9a95f3cf PC |
1041 | * when it's been released. Otherwise, it must be held for either |
1042 | * reading or writing and will not be released. | |
4bbd4c77 KS |
1043 | * |
1044 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1045 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1046 | * you need some special @gup_flags. | |
1047 | */ | |
64019a2e | 1048 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
1049 | unsigned long start, unsigned long nr_pages, |
1050 | unsigned int gup_flags, struct page **pages, | |
4f6da934 | 1051 | struct vm_area_struct **vmas, int *locked) |
4bbd4c77 | 1052 | { |
df06b37f | 1053 | long ret = 0, i = 0; |
fa5bb209 | 1054 | struct vm_area_struct *vma = NULL; |
df06b37f | 1055 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
1056 | |
1057 | if (!nr_pages) | |
1058 | return 0; | |
1059 | ||
f9652594 AK |
1060 | start = untagged_addr(start); |
1061 | ||
eddb1c22 | 1062 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 KS |
1063 | |
1064 | /* | |
1065 | * If FOLL_FORCE is set then do not force a full fault as the hinting | |
1066 | * fault information is unrelated to the reference behaviour of a task | |
1067 | * using the address space | |
1068 | */ | |
1069 | if (!(gup_flags & FOLL_FORCE)) | |
1070 | gup_flags |= FOLL_NUMA; | |
1071 | ||
4bbd4c77 | 1072 | do { |
fa5bb209 KS |
1073 | struct page *page; |
1074 | unsigned int foll_flags = gup_flags; | |
1075 | unsigned int page_increm; | |
1076 | ||
1077 | /* first iteration or cross vma bound */ | |
1078 | if (!vma || start >= vma->vm_end) { | |
1079 | vma = find_extend_vma(mm, start); | |
1080 | if (!vma && in_gate_area(mm, start)) { | |
fa5bb209 KS |
1081 | ret = get_gate_page(mm, start & PAGE_MASK, |
1082 | gup_flags, &vma, | |
1083 | pages ? &pages[i] : NULL); | |
1084 | if (ret) | |
08be37b7 | 1085 | goto out; |
df06b37f | 1086 | ctx.page_mask = 0; |
fa5bb209 KS |
1087 | goto next_page; |
1088 | } | |
4bbd4c77 | 1089 | |
52650c8b | 1090 | if (!vma) { |
df06b37f KB |
1091 | ret = -EFAULT; |
1092 | goto out; | |
1093 | } | |
52650c8b JG |
1094 | ret = check_vma_flags(vma, gup_flags); |
1095 | if (ret) | |
1096 | goto out; | |
1097 | ||
fa5bb209 KS |
1098 | if (is_vm_hugetlb_page(vma)) { |
1099 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1100 | &start, &nr_pages, i, | |
a308c71b | 1101 | gup_flags, locked); |
ad415db8 PX |
1102 | if (locked && *locked == 0) { |
1103 | /* | |
1104 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1105 | * and we've lost mmap_lock. |
ad415db8 PX |
1106 | * We must stop here. |
1107 | */ | |
1108 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
ad415db8 PX |
1109 | goto out; |
1110 | } | |
fa5bb209 | 1111 | continue; |
4bbd4c77 | 1112 | } |
fa5bb209 KS |
1113 | } |
1114 | retry: | |
1115 | /* | |
1116 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1117 | * potentially allocating memory. | |
1118 | */ | |
fa45f116 | 1119 | if (fatal_signal_pending(current)) { |
d180870d | 1120 | ret = -EINTR; |
df06b37f KB |
1121 | goto out; |
1122 | } | |
fa5bb209 | 1123 | cond_resched(); |
df06b37f KB |
1124 | |
1125 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
fa5bb209 | 1126 | if (!page) { |
64019a2e | 1127 | ret = faultin_page(vma, start, &foll_flags, locked); |
fa5bb209 KS |
1128 | switch (ret) { |
1129 | case 0: | |
1130 | goto retry; | |
df06b37f KB |
1131 | case -EBUSY: |
1132 | ret = 0; | |
e4a9bc58 | 1133 | fallthrough; |
fa5bb209 KS |
1134 | case -EFAULT: |
1135 | case -ENOMEM: | |
1136 | case -EHWPOISON: | |
df06b37f | 1137 | goto out; |
4bbd4c77 | 1138 | } |
fa5bb209 | 1139 | BUG(); |
1027e443 KS |
1140 | } else if (PTR_ERR(page) == -EEXIST) { |
1141 | /* | |
1142 | * Proper page table entry exists, but no corresponding | |
1143 | * struct page. | |
1144 | */ | |
1145 | goto next_page; | |
1146 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1147 | ret = PTR_ERR(page); |
1148 | goto out; | |
1027e443 | 1149 | } |
fa5bb209 KS |
1150 | if (pages) { |
1151 | pages[i] = page; | |
1152 | flush_anon_page(vma, page, start); | |
1153 | flush_dcache_page(page); | |
df06b37f | 1154 | ctx.page_mask = 0; |
4bbd4c77 | 1155 | } |
4bbd4c77 | 1156 | next_page: |
fa5bb209 KS |
1157 | if (vmas) { |
1158 | vmas[i] = vma; | |
df06b37f | 1159 | ctx.page_mask = 0; |
fa5bb209 | 1160 | } |
df06b37f | 1161 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1162 | if (page_increm > nr_pages) |
1163 | page_increm = nr_pages; | |
1164 | i += page_increm; | |
1165 | start += page_increm * PAGE_SIZE; | |
1166 | nr_pages -= page_increm; | |
4bbd4c77 | 1167 | } while (nr_pages); |
df06b37f KB |
1168 | out: |
1169 | if (ctx.pgmap) | |
1170 | put_dev_pagemap(ctx.pgmap); | |
1171 | return i ? i : ret; | |
4bbd4c77 | 1172 | } |
4bbd4c77 | 1173 | |
771ab430 TK |
1174 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1175 | unsigned int fault_flags) | |
d4925e00 | 1176 | { |
1b2ee126 DH |
1177 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1178 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1179 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1180 | |
1181 | if (!(vm_flags & vma->vm_flags)) | |
1182 | return false; | |
1183 | ||
33a709b2 DH |
1184 | /* |
1185 | * The architecture might have a hardware protection | |
1b2ee126 | 1186 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1187 | * |
1188 | * gup always represents data access, not instruction | |
1189 | * fetches, so execute=false here: | |
33a709b2 | 1190 | */ |
d61172b4 | 1191 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1192 | return false; |
1193 | ||
d4925e00 DH |
1194 | return true; |
1195 | } | |
1196 | ||
adc8cb40 | 1197 | /** |
4bbd4c77 | 1198 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1199 | * @mm: mm_struct of target mm |
1200 | * @address: user address | |
1201 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1202 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1203 | * does not allow retry. If NULL, the caller must guarantee |
1204 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1205 | * |
1206 | * This is meant to be called in the specific scenario where for locking reasons | |
1207 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1208 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1209 | * trying again. | |
1210 | * | |
1211 | * Typically this is meant to be used by the futex code. | |
1212 | * | |
1213 | * The main difference with get_user_pages() is that this function will | |
1214 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1215 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1216 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1217 | * |
1218 | * This is important for some architectures where those bits also gate the | |
1219 | * access permission to the page because they are maintained in software. On | |
1220 | * such architectures, gup() will not be enough to make a subsequent access | |
1221 | * succeed. | |
1222 | * | |
c1e8d7c6 ML |
1223 | * This function will not return with an unlocked mmap_lock. So it has not the |
1224 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1225 | */ |
64019a2e | 1226 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1227 | unsigned long address, unsigned int fault_flags, |
1228 | bool *unlocked) | |
4bbd4c77 KS |
1229 | { |
1230 | struct vm_area_struct *vma; | |
8fed2f3c | 1231 | vm_fault_t ret; |
4a9e1cda | 1232 | |
f9652594 AK |
1233 | address = untagged_addr(address); |
1234 | ||
4a9e1cda | 1235 | if (unlocked) |
71335f37 | 1236 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1237 | |
4a9e1cda | 1238 | retry: |
4bbd4c77 KS |
1239 | vma = find_extend_vma(mm, address); |
1240 | if (!vma || address < vma->vm_start) | |
1241 | return -EFAULT; | |
1242 | ||
d4925e00 | 1243 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1244 | return -EFAULT; |
1245 | ||
475f4dfc PX |
1246 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1247 | fatal_signal_pending(current)) | |
1248 | return -EINTR; | |
1249 | ||
bce617ed | 1250 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
4bbd4c77 | 1251 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1252 | int err = vm_fault_to_errno(ret, 0); |
1253 | ||
1254 | if (err) | |
1255 | return err; | |
4bbd4c77 KS |
1256 | BUG(); |
1257 | } | |
4a9e1cda DD |
1258 | |
1259 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1260 | mmap_read_lock(mm); |
475f4dfc PX |
1261 | *unlocked = true; |
1262 | fault_flags |= FAULT_FLAG_TRIED; | |
1263 | goto retry; | |
4a9e1cda DD |
1264 | } |
1265 | ||
4bbd4c77 KS |
1266 | return 0; |
1267 | } | |
add6a0cd | 1268 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1269 | |
2d3a36a4 MH |
1270 | /* |
1271 | * Please note that this function, unlike __get_user_pages will not | |
1272 | * return 0 for nr_pages > 0 without FOLL_NOWAIT | |
1273 | */ | |
64019a2e | 1274 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1275 | unsigned long start, |
1276 | unsigned long nr_pages, | |
f0818f47 AA |
1277 | struct page **pages, |
1278 | struct vm_area_struct **vmas, | |
e716712f | 1279 | int *locked, |
0fd71a56 | 1280 | unsigned int flags) |
f0818f47 | 1281 | { |
f0818f47 AA |
1282 | long ret, pages_done; |
1283 | bool lock_dropped; | |
1284 | ||
1285 | if (locked) { | |
1286 | /* if VM_FAULT_RETRY can be returned, vmas become invalid */ | |
1287 | BUG_ON(vmas); | |
1288 | /* check caller initialized locked */ | |
1289 | BUG_ON(*locked != 1); | |
1290 | } | |
1291 | ||
a458b76a AA |
1292 | if (flags & FOLL_PIN) |
1293 | mm_set_has_pinned_flag(&mm->flags); | |
008cfe44 | 1294 | |
eddb1c22 JH |
1295 | /* |
1296 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1297 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1298 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1299 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1300 | * | |
1301 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1302 | * that here, as any failures will be obvious enough. | |
1303 | */ | |
1304 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1305 | flags |= FOLL_GET; |
f0818f47 AA |
1306 | |
1307 | pages_done = 0; | |
1308 | lock_dropped = false; | |
1309 | for (;;) { | |
64019a2e | 1310 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
f0818f47 AA |
1311 | vmas, locked); |
1312 | if (!locked) | |
1313 | /* VM_FAULT_RETRY couldn't trigger, bypass */ | |
1314 | return ret; | |
1315 | ||
1316 | /* VM_FAULT_RETRY cannot return errors */ | |
1317 | if (!*locked) { | |
1318 | BUG_ON(ret < 0); | |
1319 | BUG_ON(ret >= nr_pages); | |
1320 | } | |
1321 | ||
f0818f47 AA |
1322 | if (ret > 0) { |
1323 | nr_pages -= ret; | |
1324 | pages_done += ret; | |
1325 | if (!nr_pages) | |
1326 | break; | |
1327 | } | |
1328 | if (*locked) { | |
96312e61 AA |
1329 | /* |
1330 | * VM_FAULT_RETRY didn't trigger or it was a | |
1331 | * FOLL_NOWAIT. | |
1332 | */ | |
f0818f47 AA |
1333 | if (!pages_done) |
1334 | pages_done = ret; | |
1335 | break; | |
1336 | } | |
df17277b MR |
1337 | /* |
1338 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1339 | * For the prefault case (!pages) we only update counts. | |
1340 | */ | |
1341 | if (likely(pages)) | |
1342 | pages += ret; | |
f0818f47 | 1343 | start += ret << PAGE_SHIFT; |
4426e945 | 1344 | lock_dropped = true; |
f0818f47 | 1345 | |
4426e945 | 1346 | retry: |
f0818f47 AA |
1347 | /* |
1348 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1349 | * with both FAULT_FLAG_ALLOW_RETRY and |
1350 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
1351 | * by fatal signals, so we need to check it before we | |
1352 | * start trying again otherwise it can loop forever. | |
f0818f47 | 1353 | */ |
4426e945 | 1354 | |
ae46d2aa HD |
1355 | if (fatal_signal_pending(current)) { |
1356 | if (!pages_done) | |
1357 | pages_done = -EINTR; | |
4426e945 | 1358 | break; |
ae46d2aa | 1359 | } |
4426e945 | 1360 | |
d8ed45c5 | 1361 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1362 | if (ret) { |
1363 | BUG_ON(ret > 0); | |
1364 | if (!pages_done) | |
1365 | pages_done = ret; | |
1366 | break; | |
1367 | } | |
4426e945 | 1368 | |
c7b6a566 | 1369 | *locked = 1; |
64019a2e | 1370 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
4426e945 PX |
1371 | pages, NULL, locked); |
1372 | if (!*locked) { | |
1373 | /* Continue to retry until we succeeded */ | |
1374 | BUG_ON(ret != 0); | |
1375 | goto retry; | |
1376 | } | |
f0818f47 AA |
1377 | if (ret != 1) { |
1378 | BUG_ON(ret > 1); | |
1379 | if (!pages_done) | |
1380 | pages_done = ret; | |
1381 | break; | |
1382 | } | |
1383 | nr_pages--; | |
1384 | pages_done++; | |
1385 | if (!nr_pages) | |
1386 | break; | |
df17277b MR |
1387 | if (likely(pages)) |
1388 | pages++; | |
f0818f47 AA |
1389 | start += PAGE_SIZE; |
1390 | } | |
e716712f | 1391 | if (lock_dropped && *locked) { |
f0818f47 AA |
1392 | /* |
1393 | * We must let the caller know we temporarily dropped the lock | |
1394 | * and so the critical section protected by it was lost. | |
1395 | */ | |
d8ed45c5 | 1396 | mmap_read_unlock(mm); |
f0818f47 AA |
1397 | *locked = 0; |
1398 | } | |
1399 | return pages_done; | |
1400 | } | |
1401 | ||
d3649f68 CH |
1402 | /** |
1403 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1404 | * @vma: target vma | |
1405 | * @start: start address | |
1406 | * @end: end address | |
c1e8d7c6 | 1407 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1408 | * |
1409 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1410 | * | |
0a36f7f8 TY |
1411 | * Return either number of pages pinned in the vma, or a negative error |
1412 | * code on error. | |
d3649f68 | 1413 | * |
c1e8d7c6 | 1414 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1415 | * |
4f6da934 | 1416 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1417 | * be unperturbed. |
1418 | * | |
4f6da934 PX |
1419 | * If @locked is non-NULL, it must held for read only and may be |
1420 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1421 | */ |
1422 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1423 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1424 | { |
1425 | struct mm_struct *mm = vma->vm_mm; | |
1426 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1427 | int gup_flags; | |
1428 | ||
be51eb18 ML |
1429 | VM_BUG_ON(!PAGE_ALIGNED(start)); |
1430 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
d3649f68 CH |
1431 | VM_BUG_ON_VMA(start < vma->vm_start, vma); |
1432 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1433 | mmap_assert_locked(mm); |
d3649f68 | 1434 | |
b67bf49c HD |
1435 | /* |
1436 | * Rightly or wrongly, the VM_LOCKONFAULT case has never used | |
1437 | * faultin_page() to break COW, so it has no work to do here. | |
1438 | */ | |
d3649f68 | 1439 | if (vma->vm_flags & VM_LOCKONFAULT) |
b67bf49c HD |
1440 | return nr_pages; |
1441 | ||
1442 | gup_flags = FOLL_TOUCH; | |
d3649f68 CH |
1443 | /* |
1444 | * We want to touch writable mappings with a write fault in order | |
1445 | * to break COW, except for shared mappings because these don't COW | |
1446 | * and we would not want to dirty them for nothing. | |
1447 | */ | |
1448 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1449 | gup_flags |= FOLL_WRITE; | |
1450 | ||
1451 | /* | |
1452 | * We want mlock to succeed for regions that have any permissions | |
1453 | * other than PROT_NONE. | |
1454 | */ | |
3122e80e | 1455 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1456 | gup_flags |= FOLL_FORCE; |
1457 | ||
1458 | /* | |
1459 | * We made sure addr is within a VMA, so the following will | |
1460 | * not result in a stack expansion that recurses back here. | |
1461 | */ | |
64019a2e | 1462 | return __get_user_pages(mm, start, nr_pages, gup_flags, |
4f6da934 | 1463 | NULL, NULL, locked); |
d3649f68 CH |
1464 | } |
1465 | ||
4ca9b385 DH |
1466 | /* |
1467 | * faultin_vma_page_range() - populate (prefault) page tables inside the | |
1468 | * given VMA range readable/writable | |
1469 | * | |
1470 | * This takes care of mlocking the pages, too, if VM_LOCKED is set. | |
1471 | * | |
1472 | * @vma: target vma | |
1473 | * @start: start address | |
1474 | * @end: end address | |
1475 | * @write: whether to prefault readable or writable | |
1476 | * @locked: whether the mmap_lock is still held | |
1477 | * | |
1478 | * Returns either number of processed pages in the vma, or a negative error | |
1479 | * code on error (see __get_user_pages()). | |
1480 | * | |
1481 | * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and | |
1482 | * covered by the VMA. | |
1483 | * | |
1484 | * If @locked is NULL, it may be held for read or write and will be unperturbed. | |
1485 | * | |
1486 | * If @locked is non-NULL, it must held for read only and may be released. If | |
1487 | * it's released, *@locked will be set to 0. | |
1488 | */ | |
1489 | long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start, | |
1490 | unsigned long end, bool write, int *locked) | |
1491 | { | |
1492 | struct mm_struct *mm = vma->vm_mm; | |
1493 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1494 | int gup_flags; | |
1495 | ||
1496 | VM_BUG_ON(!PAGE_ALIGNED(start)); | |
1497 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
1498 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1499 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
1500 | mmap_assert_locked(mm); | |
1501 | ||
1502 | /* | |
1503 | * FOLL_TOUCH: Mark page accessed and thereby young; will also mark | |
1504 | * the page dirty with FOLL_WRITE -- which doesn't make a | |
1505 | * difference with !FOLL_FORCE, because the page is writable | |
1506 | * in the page table. | |
1507 | * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit | |
1508 | * a poisoned page. | |
4ca9b385 DH |
1509 | * !FOLL_FORCE: Require proper access permissions. |
1510 | */ | |
b67bf49c | 1511 | gup_flags = FOLL_TOUCH | FOLL_HWPOISON; |
4ca9b385 DH |
1512 | if (write) |
1513 | gup_flags |= FOLL_WRITE; | |
1514 | ||
1515 | /* | |
eb2faa51 DH |
1516 | * We want to report -EINVAL instead of -EFAULT for any permission |
1517 | * problems or incompatible mappings. | |
4ca9b385 | 1518 | */ |
eb2faa51 DH |
1519 | if (check_vma_flags(vma, gup_flags)) |
1520 | return -EINVAL; | |
1521 | ||
4ca9b385 DH |
1522 | return __get_user_pages(mm, start, nr_pages, gup_flags, |
1523 | NULL, NULL, locked); | |
1524 | } | |
1525 | ||
d3649f68 CH |
1526 | /* |
1527 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1528 | * | |
1529 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1530 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1531 | * mmap_lock must not be held. |
d3649f68 CH |
1532 | */ |
1533 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1534 | { | |
1535 | struct mm_struct *mm = current->mm; | |
1536 | unsigned long end, nstart, nend; | |
1537 | struct vm_area_struct *vma = NULL; | |
1538 | int locked = 0; | |
1539 | long ret = 0; | |
1540 | ||
1541 | end = start + len; | |
1542 | ||
1543 | for (nstart = start; nstart < end; nstart = nend) { | |
1544 | /* | |
1545 | * We want to fault in pages for [nstart; end) address range. | |
1546 | * Find first corresponding VMA. | |
1547 | */ | |
1548 | if (!locked) { | |
1549 | locked = 1; | |
d8ed45c5 | 1550 | mmap_read_lock(mm); |
d3649f68 CH |
1551 | vma = find_vma(mm, nstart); |
1552 | } else if (nstart >= vma->vm_end) | |
1553 | vma = vma->vm_next; | |
1554 | if (!vma || vma->vm_start >= end) | |
1555 | break; | |
1556 | /* | |
1557 | * Set [nstart; nend) to intersection of desired address | |
1558 | * range with the first VMA. Also, skip undesirable VMA types. | |
1559 | */ | |
1560 | nend = min(end, vma->vm_end); | |
1561 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1562 | continue; | |
1563 | if (nstart < vma->vm_start) | |
1564 | nstart = vma->vm_start; | |
1565 | /* | |
1566 | * Now fault in a range of pages. populate_vma_page_range() | |
1567 | * double checks the vma flags, so that it won't mlock pages | |
1568 | * if the vma was already munlocked. | |
1569 | */ | |
1570 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1571 | if (ret < 0) { | |
1572 | if (ignore_errors) { | |
1573 | ret = 0; | |
1574 | continue; /* continue at next VMA */ | |
1575 | } | |
1576 | break; | |
1577 | } | |
1578 | nend = nstart + ret * PAGE_SIZE; | |
1579 | ret = 0; | |
1580 | } | |
1581 | if (locked) | |
d8ed45c5 | 1582 | mmap_read_unlock(mm); |
d3649f68 CH |
1583 | return ret; /* 0 or negative error code */ |
1584 | } | |
050a9adc | 1585 | #else /* CONFIG_MMU */ |
64019a2e | 1586 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc CH |
1587 | unsigned long nr_pages, struct page **pages, |
1588 | struct vm_area_struct **vmas, int *locked, | |
1589 | unsigned int foll_flags) | |
1590 | { | |
1591 | struct vm_area_struct *vma; | |
1592 | unsigned long vm_flags; | |
24dc20c7 | 1593 | long i; |
050a9adc CH |
1594 | |
1595 | /* calculate required read or write permissions. | |
1596 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1597 | */ | |
1598 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1599 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1600 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1601 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1602 | ||
1603 | for (i = 0; i < nr_pages; i++) { | |
1604 | vma = find_vma(mm, start); | |
1605 | if (!vma) | |
1606 | goto finish_or_fault; | |
1607 | ||
1608 | /* protect what we can, including chardevs */ | |
1609 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1610 | !(vm_flags & vma->vm_flags)) | |
1611 | goto finish_or_fault; | |
1612 | ||
1613 | if (pages) { | |
1614 | pages[i] = virt_to_page(start); | |
1615 | if (pages[i]) | |
1616 | get_page(pages[i]); | |
1617 | } | |
1618 | if (vmas) | |
1619 | vmas[i] = vma; | |
1620 | start = (start + PAGE_SIZE) & PAGE_MASK; | |
1621 | } | |
1622 | ||
1623 | return i; | |
1624 | ||
1625 | finish_or_fault: | |
1626 | return i ? : -EFAULT; | |
1627 | } | |
1628 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1629 | |
bb523b40 AG |
1630 | /** |
1631 | * fault_in_writeable - fault in userspace address range for writing | |
1632 | * @uaddr: start of address range | |
1633 | * @size: size of address range | |
1634 | * | |
1635 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1636 | * copy_from_user()). | |
1637 | */ | |
1638 | size_t fault_in_writeable(char __user *uaddr, size_t size) | |
1639 | { | |
1640 | char __user *start = uaddr, *end; | |
1641 | ||
1642 | if (unlikely(size == 0)) | |
1643 | return 0; | |
677b2a8c CL |
1644 | if (!user_write_access_begin(uaddr, size)) |
1645 | return size; | |
bb523b40 | 1646 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1647 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1648 | uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1649 | } | |
1650 | end = (char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1651 | if (unlikely(end < start)) | |
1652 | end = NULL; | |
1653 | while (uaddr != end) { | |
677b2a8c | 1654 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1655 | uaddr += PAGE_SIZE; |
1656 | } | |
1657 | ||
1658 | out: | |
677b2a8c | 1659 | user_write_access_end(); |
bb523b40 AG |
1660 | if (size > uaddr - start) |
1661 | return size - (uaddr - start); | |
1662 | return 0; | |
1663 | } | |
1664 | EXPORT_SYMBOL(fault_in_writeable); | |
1665 | ||
cdd591fc AG |
1666 | /* |
1667 | * fault_in_safe_writeable - fault in an address range for writing | |
1668 | * @uaddr: start of address range | |
1669 | * @size: length of address range | |
1670 | * | |
1671 | * Faults in an address range using get_user_pages, i.e., without triggering | |
1672 | * hardware page faults. This is primarily useful when we already know that | |
1673 | * some or all of the pages in the address range aren't in memory. | |
1674 | * | |
1675 | * Other than fault_in_writeable(), this function is non-destructive. | |
1676 | * | |
1677 | * Note that we don't pin or otherwise hold the pages referenced that we fault | |
1678 | * in. There's no guarantee that they'll stay in memory for any duration of | |
1679 | * time. | |
1680 | * | |
1681 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
1682 | * copy_from_user(). | |
1683 | */ | |
1684 | size_t fault_in_safe_writeable(const char __user *uaddr, size_t size) | |
1685 | { | |
1686 | unsigned long start = (unsigned long)untagged_addr(uaddr); | |
1687 | unsigned long end, nstart, nend; | |
1688 | struct mm_struct *mm = current->mm; | |
1689 | struct vm_area_struct *vma = NULL; | |
1690 | int locked = 0; | |
1691 | ||
1692 | nstart = start & PAGE_MASK; | |
1693 | end = PAGE_ALIGN(start + size); | |
1694 | if (end < nstart) | |
1695 | end = 0; | |
1696 | for (; nstart != end; nstart = nend) { | |
1697 | unsigned long nr_pages; | |
1698 | long ret; | |
1699 | ||
1700 | if (!locked) { | |
1701 | locked = 1; | |
1702 | mmap_read_lock(mm); | |
1703 | vma = find_vma(mm, nstart); | |
1704 | } else if (nstart >= vma->vm_end) | |
1705 | vma = vma->vm_next; | |
1706 | if (!vma || vma->vm_start >= end) | |
1707 | break; | |
1708 | nend = end ? min(end, vma->vm_end) : vma->vm_end; | |
1709 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1710 | continue; | |
1711 | if (nstart < vma->vm_start) | |
1712 | nstart = vma->vm_start; | |
1713 | nr_pages = (nend - nstart) / PAGE_SIZE; | |
1714 | ret = __get_user_pages_locked(mm, nstart, nr_pages, | |
1715 | NULL, NULL, &locked, | |
1716 | FOLL_TOUCH | FOLL_WRITE); | |
1717 | if (ret <= 0) | |
1718 | break; | |
1719 | nend = nstart + ret * PAGE_SIZE; | |
1720 | } | |
1721 | if (locked) | |
1722 | mmap_read_unlock(mm); | |
1723 | if (nstart == end) | |
1724 | return 0; | |
1725 | return size - min_t(size_t, nstart - start, size); | |
1726 | } | |
1727 | EXPORT_SYMBOL(fault_in_safe_writeable); | |
1728 | ||
bb523b40 AG |
1729 | /** |
1730 | * fault_in_readable - fault in userspace address range for reading | |
1731 | * @uaddr: start of user address range | |
1732 | * @size: size of user address range | |
1733 | * | |
1734 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1735 | * copy_from_user()). | |
1736 | */ | |
1737 | size_t fault_in_readable(const char __user *uaddr, size_t size) | |
1738 | { | |
1739 | const char __user *start = uaddr, *end; | |
1740 | volatile char c; | |
1741 | ||
1742 | if (unlikely(size == 0)) | |
1743 | return 0; | |
677b2a8c CL |
1744 | if (!user_read_access_begin(uaddr, size)) |
1745 | return size; | |
bb523b40 | 1746 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1747 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1748 | uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1749 | } | |
1750 | end = (const char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1751 | if (unlikely(end < start)) | |
1752 | end = NULL; | |
1753 | while (uaddr != end) { | |
677b2a8c | 1754 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1755 | uaddr += PAGE_SIZE; |
1756 | } | |
1757 | ||
1758 | out: | |
677b2a8c | 1759 | user_read_access_end(); |
bb523b40 AG |
1760 | (void)c; |
1761 | if (size > uaddr - start) | |
1762 | return size - (uaddr - start); | |
1763 | return 0; | |
1764 | } | |
1765 | EXPORT_SYMBOL(fault_in_readable); | |
1766 | ||
8f942eea JH |
1767 | /** |
1768 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1769 | * @addr: user address | |
1770 | * | |
1771 | * Returns struct page pointer of user page pinned for dump, | |
1772 | * to be freed afterwards by put_page(). | |
1773 | * | |
1774 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1775 | * the corefile, to preserve alignment with its headers; and also returns | |
1776 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
f0953a1b | 1777 | * allowing a hole to be left in the corefile to save disk space. |
8f942eea | 1778 | * |
7f3bfab5 | 1779 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1780 | */ |
1781 | #ifdef CONFIG_ELF_CORE | |
1782 | struct page *get_dump_page(unsigned long addr) | |
1783 | { | |
7f3bfab5 | 1784 | struct mm_struct *mm = current->mm; |
8f942eea | 1785 | struct page *page; |
7f3bfab5 JH |
1786 | int locked = 1; |
1787 | int ret; | |
8f942eea | 1788 | |
7f3bfab5 | 1789 | if (mmap_read_lock_killable(mm)) |
8f942eea | 1790 | return NULL; |
7f3bfab5 JH |
1791 | ret = __get_user_pages_locked(mm, addr, 1, &page, NULL, &locked, |
1792 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); | |
1793 | if (locked) | |
1794 | mmap_read_unlock(mm); | |
1795 | return (ret == 1) ? page : NULL; | |
8f942eea JH |
1796 | } |
1797 | #endif /* CONFIG_ELF_CORE */ | |
1798 | ||
d1e153fe | 1799 | #ifdef CONFIG_MIGRATION |
f68749ec PT |
1800 | /* |
1801 | * Check whether all pages are pinnable, if so return number of pages. If some | |
1802 | * pages are not pinnable, migrate them, and unpin all pages. Return zero if | |
1803 | * pages were migrated, or if some pages were not successfully isolated. | |
1804 | * Return negative error if migration fails. | |
1805 | */ | |
1806 | static long check_and_migrate_movable_pages(unsigned long nr_pages, | |
d1e153fe | 1807 | struct page **pages, |
d1e153fe | 1808 | unsigned int gup_flags) |
9a4e9f3b | 1809 | { |
f9f38f78 | 1810 | unsigned long isolation_error_count = 0, i; |
f68749ec | 1811 | struct page *prev_head = NULL; |
f9f38f78 CH |
1812 | LIST_HEAD(movable_page_list); |
1813 | bool drain_allow = true; | |
1814 | int ret = 0; | |
9a4e9f3b | 1815 | |
83c02c23 | 1816 | for (i = 0; i < nr_pages; i++) { |
f9f38f78 CH |
1817 | struct page *head = compound_head(pages[i]); |
1818 | ||
83c02c23 PT |
1819 | if (head == prev_head) |
1820 | continue; | |
1821 | prev_head = head; | |
f9f38f78 CH |
1822 | |
1823 | if (is_pinnable_page(head)) | |
1824 | continue; | |
1825 | ||
9a4e9f3b | 1826 | /* |
f9f38f78 | 1827 | * Try to move out any movable page before pinning the range. |
9a4e9f3b | 1828 | */ |
f9f38f78 CH |
1829 | if (PageHuge(head)) { |
1830 | if (!isolate_huge_page(head, &movable_page_list)) | |
1831 | isolation_error_count++; | |
1832 | continue; | |
1833 | } | |
9a4e9f3b | 1834 | |
f9f38f78 CH |
1835 | if (!PageLRU(head) && drain_allow) { |
1836 | lru_add_drain_all(); | |
1837 | drain_allow = false; | |
1838 | } | |
1839 | ||
1840 | if (isolate_lru_page(head)) { | |
1841 | isolation_error_count++; | |
1842 | continue; | |
9a4e9f3b | 1843 | } |
f9f38f78 CH |
1844 | list_add_tail(&head->lru, &movable_page_list); |
1845 | mod_node_page_state(page_pgdat(head), | |
1846 | NR_ISOLATED_ANON + page_is_file_lru(head), | |
1847 | thp_nr_pages(head)); | |
9a4e9f3b AK |
1848 | } |
1849 | ||
f9f38f78 CH |
1850 | if (!list_empty(&movable_page_list) || isolation_error_count) |
1851 | goto unpin_pages; | |
1852 | ||
6e7f34eb PT |
1853 | /* |
1854 | * If list is empty, and no isolation errors, means that all pages are | |
1855 | * in the correct zone. | |
1856 | */ | |
f9f38f78 | 1857 | return nr_pages; |
6e7f34eb | 1858 | |
f9f38f78 | 1859 | unpin_pages: |
f68749ec PT |
1860 | if (gup_flags & FOLL_PIN) { |
1861 | unpin_user_pages(pages, nr_pages); | |
1862 | } else { | |
1863 | for (i = 0; i < nr_pages; i++) | |
1864 | put_page(pages[i]); | |
1865 | } | |
f9f38f78 | 1866 | |
d1e153fe | 1867 | if (!list_empty(&movable_page_list)) { |
f9f38f78 CH |
1868 | struct migration_target_control mtc = { |
1869 | .nid = NUMA_NO_NODE, | |
1870 | .gfp_mask = GFP_USER | __GFP_NOWARN, | |
1871 | }; | |
1872 | ||
d1e153fe | 1873 | ret = migrate_pages(&movable_page_list, alloc_migration_target, |
f0f44638 | 1874 | NULL, (unsigned long)&mtc, MIGRATE_SYNC, |
5ac95884 | 1875 | MR_LONGTERM_PIN, NULL); |
f9f38f78 CH |
1876 | if (ret > 0) /* number of pages not migrated */ |
1877 | ret = -ENOMEM; | |
9a4e9f3b AK |
1878 | } |
1879 | ||
f9f38f78 CH |
1880 | if (ret && !list_empty(&movable_page_list)) |
1881 | putback_movable_pages(&movable_page_list); | |
1882 | return ret; | |
9a4e9f3b AK |
1883 | } |
1884 | #else | |
f68749ec | 1885 | static long check_and_migrate_movable_pages(unsigned long nr_pages, |
d1e153fe | 1886 | struct page **pages, |
d1e153fe | 1887 | unsigned int gup_flags) |
9a4e9f3b AK |
1888 | { |
1889 | return nr_pages; | |
1890 | } | |
d1e153fe | 1891 | #endif /* CONFIG_MIGRATION */ |
9a4e9f3b | 1892 | |
2bb6d283 | 1893 | /* |
932f4a63 IW |
1894 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
1895 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 1896 | */ |
64019a2e | 1897 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
1898 | unsigned long start, |
1899 | unsigned long nr_pages, | |
1900 | struct page **pages, | |
1901 | struct vm_area_struct **vmas, | |
1902 | unsigned int gup_flags) | |
2bb6d283 | 1903 | { |
f68749ec | 1904 | unsigned int flags; |
52650c8b | 1905 | long rc; |
2bb6d283 | 1906 | |
f68749ec PT |
1907 | if (!(gup_flags & FOLL_LONGTERM)) |
1908 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
1909 | NULL, gup_flags); | |
1910 | flags = memalloc_pin_save(); | |
1911 | do { | |
1912 | rc = __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
1913 | NULL, gup_flags); | |
1914 | if (rc <= 0) | |
1915 | break; | |
1916 | rc = check_and_migrate_movable_pages(rc, pages, gup_flags); | |
1917 | } while (!rc); | |
1918 | memalloc_pin_restore(flags); | |
2bb6d283 | 1919 | |
2bb6d283 DW |
1920 | return rc; |
1921 | } | |
932f4a63 | 1922 | |
447f3e45 BS |
1923 | static bool is_valid_gup_flags(unsigned int gup_flags) |
1924 | { | |
1925 | /* | |
1926 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
1927 | * never directly by the caller, so enforce that with an assertion: | |
1928 | */ | |
1929 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
1930 | return false; | |
1931 | /* | |
1932 | * FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying | |
1933 | * that is, FOLL_LONGTERM is a specific case, more restrictive case of | |
1934 | * FOLL_PIN. | |
1935 | */ | |
1936 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
1937 | return false; | |
1938 | ||
1939 | return true; | |
1940 | } | |
1941 | ||
22bf29b6 | 1942 | #ifdef CONFIG_MMU |
64019a2e | 1943 | static long __get_user_pages_remote(struct mm_struct *mm, |
22bf29b6 JH |
1944 | unsigned long start, unsigned long nr_pages, |
1945 | unsigned int gup_flags, struct page **pages, | |
1946 | struct vm_area_struct **vmas, int *locked) | |
1947 | { | |
1948 | /* | |
1949 | * Parts of FOLL_LONGTERM behavior are incompatible with | |
1950 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1951 | * vmas. However, this only comes up if locked is set, and there are | |
1952 | * callers that do request FOLL_LONGTERM, but do not set locked. So, | |
1953 | * allow what we can. | |
1954 | */ | |
1955 | if (gup_flags & FOLL_LONGTERM) { | |
1956 | if (WARN_ON_ONCE(locked)) | |
1957 | return -EINVAL; | |
1958 | /* | |
1959 | * This will check the vmas (even if our vmas arg is NULL) | |
1960 | * and return -ENOTSUPP if DAX isn't allowed in this case: | |
1961 | */ | |
64019a2e | 1962 | return __gup_longterm_locked(mm, start, nr_pages, pages, |
22bf29b6 JH |
1963 | vmas, gup_flags | FOLL_TOUCH | |
1964 | FOLL_REMOTE); | |
1965 | } | |
1966 | ||
64019a2e | 1967 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, |
22bf29b6 JH |
1968 | locked, |
1969 | gup_flags | FOLL_TOUCH | FOLL_REMOTE); | |
1970 | } | |
1971 | ||
adc8cb40 | 1972 | /** |
c4237f8b | 1973 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
1974 | * @mm: mm_struct of target mm |
1975 | * @start: starting user address | |
1976 | * @nr_pages: number of pages from start to pin | |
1977 | * @gup_flags: flags modifying lookup behaviour | |
1978 | * @pages: array that receives pointers to the pages pinned. | |
1979 | * Should be at least nr_pages long. Or NULL, if caller | |
1980 | * only intends to ensure the pages are faulted in. | |
1981 | * @vmas: array of pointers to vmas corresponding to each page. | |
1982 | * Or NULL if the caller does not require them. | |
1983 | * @locked: pointer to lock flag indicating whether lock is held and | |
1984 | * subsequently whether VM_FAULT_RETRY functionality can be | |
1985 | * utilised. Lock must initially be held. | |
1986 | * | |
1987 | * Returns either number of pages pinned (which may be less than the | |
1988 | * number requested), or an error. Details about the return value: | |
1989 | * | |
1990 | * -- If nr_pages is 0, returns 0. | |
1991 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1992 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1993 | * pages pinned. Again, this may be less than nr_pages. | |
1994 | * | |
1995 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1996 | * | |
c1e8d7c6 | 1997 | * @vmas are valid only as long as mmap_lock is held. |
c4237f8b | 1998 | * |
c1e8d7c6 | 1999 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 2000 | * |
adc8cb40 SJ |
2001 | * get_user_pages_remote walks a process's page tables and takes a reference |
2002 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
2003 | * instant. That is, it takes the page that would be accessed if a user |
2004 | * thread accesses the given user virtual address at that instant. | |
2005 | * | |
2006 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 2007 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b JH |
2008 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
2009 | * and subsequently re faulted). However it does guarantee that the page | |
2010 | * won't be freed completely. And mostly callers simply care that the page | |
2011 | * contains data that was valid *at some point in time*. Typically, an IO | |
2012 | * or similar operation cannot guarantee anything stronger anyway because | |
2013 | * locks can't be held over the syscall boundary. | |
2014 | * | |
2015 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
2016 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
2017 | * be called after the page is finished with, and before put_page is called. | |
2018 | * | |
adc8cb40 SJ |
2019 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
2020 | * to get a handle on the memory by some means other than accesses | |
2021 | * via the user virtual addresses. The pages may be submitted for | |
2022 | * DMA to devices or accessed via their kernel linear mapping (via the | |
2023 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
2024 | * |
2025 | * See also get_user_pages_fast, for performance critical applications. | |
2026 | * | |
adc8cb40 | 2027 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 2028 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 2029 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
2030 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
2031 | */ | |
64019a2e | 2032 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
2033 | unsigned long start, unsigned long nr_pages, |
2034 | unsigned int gup_flags, struct page **pages, | |
2035 | struct vm_area_struct **vmas, int *locked) | |
2036 | { | |
447f3e45 | 2037 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2038 | return -EINVAL; |
2039 | ||
64019a2e | 2040 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
22bf29b6 | 2041 | pages, vmas, locked); |
c4237f8b JH |
2042 | } |
2043 | EXPORT_SYMBOL(get_user_pages_remote); | |
2044 | ||
eddb1c22 | 2045 | #else /* CONFIG_MMU */ |
64019a2e | 2046 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2047 | unsigned long start, unsigned long nr_pages, |
2048 | unsigned int gup_flags, struct page **pages, | |
2049 | struct vm_area_struct **vmas, int *locked) | |
2050 | { | |
2051 | return 0; | |
2052 | } | |
3faa52c0 | 2053 | |
64019a2e | 2054 | static long __get_user_pages_remote(struct mm_struct *mm, |
3faa52c0 JH |
2055 | unsigned long start, unsigned long nr_pages, |
2056 | unsigned int gup_flags, struct page **pages, | |
2057 | struct vm_area_struct **vmas, int *locked) | |
2058 | { | |
2059 | return 0; | |
2060 | } | |
eddb1c22 JH |
2061 | #endif /* !CONFIG_MMU */ |
2062 | ||
adc8cb40 SJ |
2063 | /** |
2064 | * get_user_pages() - pin user pages in memory | |
2065 | * @start: starting user address | |
2066 | * @nr_pages: number of pages from start to pin | |
2067 | * @gup_flags: flags modifying lookup behaviour | |
2068 | * @pages: array that receives pointers to the pages pinned. | |
2069 | * Should be at least nr_pages long. Or NULL, if caller | |
2070 | * only intends to ensure the pages are faulted in. | |
2071 | * @vmas: array of pointers to vmas corresponding to each page. | |
2072 | * Or NULL if the caller does not require them. | |
2073 | * | |
64019a2e PX |
2074 | * This is the same as get_user_pages_remote(), just with a less-flexible |
2075 | * calling convention where we assume that the mm being operated on belongs to | |
2076 | * the current task, and doesn't allow passing of a locked parameter. We also | |
2077 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
2078 | */ |
2079 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
2080 | unsigned int gup_flags, struct page **pages, | |
2081 | struct vm_area_struct **vmas) | |
2082 | { | |
447f3e45 | 2083 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2084 | return -EINVAL; |
2085 | ||
64019a2e | 2086 | return __gup_longterm_locked(current->mm, start, nr_pages, |
932f4a63 IW |
2087 | pages, vmas, gup_flags | FOLL_TOUCH); |
2088 | } | |
2089 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 2090 | |
adc8cb40 | 2091 | /** |
a00cda3f MCC |
2092 | * get_user_pages_locked() - variant of get_user_pages() |
2093 | * | |
2094 | * @start: starting user address | |
2095 | * @nr_pages: number of pages from start to pin | |
2096 | * @gup_flags: flags modifying lookup behaviour | |
2097 | * @pages: array that receives pointers to the pages pinned. | |
2098 | * Should be at least nr_pages long. Or NULL, if caller | |
2099 | * only intends to ensure the pages are faulted in. | |
2100 | * @locked: pointer to lock flag indicating whether lock is held and | |
2101 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2102 | * utilised. Lock must initially be held. | |
2103 | * | |
2104 | * It is suitable to replace the form: | |
acc3c8d1 | 2105 | * |
3e4e28c5 | 2106 | * mmap_read_lock(mm); |
d3649f68 | 2107 | * do_something() |
64019a2e | 2108 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2109 | * mmap_read_unlock(mm); |
acc3c8d1 | 2110 | * |
d3649f68 | 2111 | * to: |
acc3c8d1 | 2112 | * |
d3649f68 | 2113 | * int locked = 1; |
3e4e28c5 | 2114 | * mmap_read_lock(mm); |
d3649f68 | 2115 | * do_something() |
64019a2e | 2116 | * get_user_pages_locked(mm, ..., pages, &locked); |
d3649f68 | 2117 | * if (locked) |
3e4e28c5 | 2118 | * mmap_read_unlock(mm); |
adc8cb40 | 2119 | * |
adc8cb40 SJ |
2120 | * We can leverage the VM_FAULT_RETRY functionality in the page fault |
2121 | * paths better by using either get_user_pages_locked() or | |
2122 | * get_user_pages_unlocked(). | |
2123 | * | |
acc3c8d1 | 2124 | */ |
d3649f68 CH |
2125 | long get_user_pages_locked(unsigned long start, unsigned long nr_pages, |
2126 | unsigned int gup_flags, struct page **pages, | |
2127 | int *locked) | |
acc3c8d1 | 2128 | { |
acc3c8d1 | 2129 | /* |
d3649f68 CH |
2130 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with |
2131 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2132 | * vmas. As there are no users of this flag in this call we simply | |
2133 | * disallow this option for now. | |
acc3c8d1 | 2134 | */ |
d3649f68 CH |
2135 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) |
2136 | return -EINVAL; | |
420c2091 JH |
2137 | /* |
2138 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
2139 | * never directly by the caller, so enforce that: | |
2140 | */ | |
2141 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
2142 | return -EINVAL; | |
acc3c8d1 | 2143 | |
64019a2e | 2144 | return __get_user_pages_locked(current->mm, start, nr_pages, |
d3649f68 CH |
2145 | pages, NULL, locked, |
2146 | gup_flags | FOLL_TOUCH); | |
acc3c8d1 | 2147 | } |
d3649f68 | 2148 | EXPORT_SYMBOL(get_user_pages_locked); |
acc3c8d1 KS |
2149 | |
2150 | /* | |
d3649f68 | 2151 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 2152 | * |
3e4e28c5 | 2153 | * mmap_read_lock(mm); |
64019a2e | 2154 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2155 | * mmap_read_unlock(mm); |
d3649f68 CH |
2156 | * |
2157 | * with: | |
2158 | * | |
64019a2e | 2159 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
2160 | * |
2161 | * It is functionally equivalent to get_user_pages_fast so | |
2162 | * get_user_pages_fast should be used instead if specific gup_flags | |
2163 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 2164 | */ |
d3649f68 CH |
2165 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
2166 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 KS |
2167 | { |
2168 | struct mm_struct *mm = current->mm; | |
d3649f68 CH |
2169 | int locked = 1; |
2170 | long ret; | |
acc3c8d1 | 2171 | |
d3649f68 CH |
2172 | /* |
2173 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
2174 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2175 | * vmas. As there are no users of this flag in this call we simply | |
2176 | * disallow this option for now. | |
2177 | */ | |
2178 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
2179 | return -EINVAL; | |
acc3c8d1 | 2180 | |
d8ed45c5 | 2181 | mmap_read_lock(mm); |
64019a2e | 2182 | ret = __get_user_pages_locked(mm, start, nr_pages, pages, NULL, |
d3649f68 | 2183 | &locked, gup_flags | FOLL_TOUCH); |
acc3c8d1 | 2184 | if (locked) |
d8ed45c5 | 2185 | mmap_read_unlock(mm); |
d3649f68 | 2186 | return ret; |
4bbd4c77 | 2187 | } |
d3649f68 | 2188 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
2189 | |
2190 | /* | |
67a929e0 | 2191 | * Fast GUP |
2667f50e SC |
2192 | * |
2193 | * get_user_pages_fast attempts to pin user pages by walking the page | |
2194 | * tables directly and avoids taking locks. Thus the walker needs to be | |
2195 | * protected from page table pages being freed from under it, and should | |
2196 | * block any THP splits. | |
2197 | * | |
2198 | * One way to achieve this is to have the walker disable interrupts, and | |
2199 | * rely on IPIs from the TLB flushing code blocking before the page table | |
2200 | * pages are freed. This is unsuitable for architectures that do not need | |
2201 | * to broadcast an IPI when invalidating TLBs. | |
2202 | * | |
2203 | * Another way to achieve this is to batch up page table containing pages | |
2204 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
2205 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
2206 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
2207 | * (which is a relatively rare event). The code below adopts this strategy. | |
2208 | * | |
2209 | * Before activating this code, please be aware that the following assumptions | |
2210 | * are currently made: | |
2211 | * | |
ff2e6d72 | 2212 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 2213 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 2214 | * |
2667f50e SC |
2215 | * *) ptes can be read atomically by the architecture. |
2216 | * | |
2217 | * *) access_ok is sufficient to validate userspace address ranges. | |
2218 | * | |
2219 | * The last two assumptions can be relaxed by the addition of helper functions. | |
2220 | * | |
2221 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
2222 | */ | |
67a929e0 | 2223 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 2224 | |
790c7369 | 2225 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 2226 | unsigned int flags, |
790c7369 | 2227 | struct page **pages) |
b59f65fa KS |
2228 | { |
2229 | while ((*nr) - nr_start) { | |
2230 | struct page *page = pages[--(*nr)]; | |
2231 | ||
2232 | ClearPageReferenced(page); | |
3faa52c0 JH |
2233 | if (flags & FOLL_PIN) |
2234 | unpin_user_page(page); | |
2235 | else | |
2236 | put_page(page); | |
b59f65fa KS |
2237 | } |
2238 | } | |
2239 | ||
3010a5ea | 2240 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
2667f50e | 2241 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, |
b798bec4 | 2242 | unsigned int flags, struct page **pages, int *nr) |
2667f50e | 2243 | { |
b59f65fa KS |
2244 | struct dev_pagemap *pgmap = NULL; |
2245 | int nr_start = *nr, ret = 0; | |
2667f50e | 2246 | pte_t *ptep, *ptem; |
2667f50e SC |
2247 | |
2248 | ptem = ptep = pte_offset_map(&pmd, addr); | |
2249 | do { | |
2a4a06da | 2250 | pte_t pte = ptep_get_lockless(ptep); |
7aef4172 | 2251 | struct page *head, *page; |
2667f50e SC |
2252 | |
2253 | /* | |
2254 | * Similar to the PMD case below, NUMA hinting must take slow | |
8a0516ed | 2255 | * path using the pte_protnone check. |
2667f50e | 2256 | */ |
e7884f8e KS |
2257 | if (pte_protnone(pte)) |
2258 | goto pte_unmap; | |
2259 | ||
b798bec4 | 2260 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2261 | goto pte_unmap; |
2262 | ||
b59f65fa | 2263 | if (pte_devmap(pte)) { |
7af75561 IW |
2264 | if (unlikely(flags & FOLL_LONGTERM)) |
2265 | goto pte_unmap; | |
2266 | ||
b59f65fa KS |
2267 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2268 | if (unlikely(!pgmap)) { | |
3b78d834 | 2269 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2270 | goto pte_unmap; |
2271 | } | |
2272 | } else if (pte_special(pte)) | |
2667f50e SC |
2273 | goto pte_unmap; |
2274 | ||
2275 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2276 | page = pte_page(pte); | |
2277 | ||
3faa52c0 | 2278 | head = try_grab_compound_head(page, 1, flags); |
8fde12ca | 2279 | if (!head) |
2667f50e SC |
2280 | goto pte_unmap; |
2281 | ||
1507f512 MR |
2282 | if (unlikely(page_is_secretmem(page))) { |
2283 | put_compound_head(head, 1, flags); | |
2284 | goto pte_unmap; | |
2285 | } | |
2286 | ||
2667f50e | 2287 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { |
3faa52c0 | 2288 | put_compound_head(head, 1, flags); |
2667f50e SC |
2289 | goto pte_unmap; |
2290 | } | |
2291 | ||
7aef4172 | 2292 | VM_BUG_ON_PAGE(compound_head(page) != head, page); |
e9348053 | 2293 | |
f28d4363 CI |
2294 | /* |
2295 | * We need to make the page accessible if and only if we are | |
2296 | * going to access its content (the FOLL_PIN case). Please | |
2297 | * see Documentation/core-api/pin_user_pages.rst for | |
2298 | * details. | |
2299 | */ | |
2300 | if (flags & FOLL_PIN) { | |
2301 | ret = arch_make_page_accessible(page); | |
2302 | if (ret) { | |
2303 | unpin_user_page(page); | |
2304 | goto pte_unmap; | |
2305 | } | |
2306 | } | |
e9348053 | 2307 | SetPageReferenced(page); |
2667f50e SC |
2308 | pages[*nr] = page; |
2309 | (*nr)++; | |
2310 | ||
2311 | } while (ptep++, addr += PAGE_SIZE, addr != end); | |
2312 | ||
2313 | ret = 1; | |
2314 | ||
2315 | pte_unmap: | |
832d7aa0 CH |
2316 | if (pgmap) |
2317 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2318 | pte_unmap(ptem); |
2319 | return ret; | |
2320 | } | |
2321 | #else | |
2322 | ||
2323 | /* | |
2324 | * If we can't determine whether or not a pte is special, then fail immediately | |
2325 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2326 | * to be special. | |
2327 | * | |
2328 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2329 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2330 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2331 | */ | |
2332 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, | |
b798bec4 | 2333 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2334 | { |
2335 | return 0; | |
2336 | } | |
3010a5ea | 2337 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2338 | |
17596731 | 2339 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2340 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2341 | unsigned long end, unsigned int flags, |
2342 | struct page **pages, int *nr) | |
b59f65fa KS |
2343 | { |
2344 | int nr_start = *nr; | |
2345 | struct dev_pagemap *pgmap = NULL; | |
2346 | ||
2347 | do { | |
2348 | struct page *page = pfn_to_page(pfn); | |
2349 | ||
2350 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2351 | if (unlikely(!pgmap)) { | |
3b78d834 | 2352 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2353 | break; |
b59f65fa KS |
2354 | } |
2355 | SetPageReferenced(page); | |
2356 | pages[*nr] = page; | |
3faa52c0 JH |
2357 | if (unlikely(!try_grab_page(page, flags))) { |
2358 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
6401c4eb | 2359 | break; |
3faa52c0 | 2360 | } |
b59f65fa KS |
2361 | (*nr)++; |
2362 | pfn++; | |
2363 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 | 2364 | |
6401c4eb | 2365 | put_dev_pagemap(pgmap); |
20b7fee7 | 2366 | return addr == end; |
b59f65fa KS |
2367 | } |
2368 | ||
a9b6de77 | 2369 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2370 | unsigned long end, unsigned int flags, |
2371 | struct page **pages, int *nr) | |
b59f65fa KS |
2372 | { |
2373 | unsigned long fault_pfn; | |
a9b6de77 DW |
2374 | int nr_start = *nr; |
2375 | ||
2376 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2377 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2378 | return 0; |
b59f65fa | 2379 | |
a9b6de77 | 2380 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2381 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2382 | return 0; |
2383 | } | |
2384 | return 1; | |
b59f65fa KS |
2385 | } |
2386 | ||
a9b6de77 | 2387 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2388 | unsigned long end, unsigned int flags, |
2389 | struct page **pages, int *nr) | |
b59f65fa KS |
2390 | { |
2391 | unsigned long fault_pfn; | |
a9b6de77 DW |
2392 | int nr_start = *nr; |
2393 | ||
2394 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2395 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2396 | return 0; |
b59f65fa | 2397 | |
a9b6de77 | 2398 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2399 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2400 | return 0; |
2401 | } | |
2402 | return 1; | |
b59f65fa KS |
2403 | } |
2404 | #else | |
a9b6de77 | 2405 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2406 | unsigned long end, unsigned int flags, |
2407 | struct page **pages, int *nr) | |
b59f65fa KS |
2408 | { |
2409 | BUILD_BUG(); | |
2410 | return 0; | |
2411 | } | |
2412 | ||
a9b6de77 | 2413 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2414 | unsigned long end, unsigned int flags, |
2415 | struct page **pages, int *nr) | |
b59f65fa KS |
2416 | { |
2417 | BUILD_BUG(); | |
2418 | return 0; | |
2419 | } | |
2420 | #endif | |
2421 | ||
a43e9820 JH |
2422 | static int record_subpages(struct page *page, unsigned long addr, |
2423 | unsigned long end, struct page **pages) | |
2424 | { | |
2425 | int nr; | |
2426 | ||
c228afb1 MWO |
2427 | for (nr = 0; addr != end; nr++, addr += PAGE_SIZE) |
2428 | pages[nr] = nth_page(page, nr); | |
a43e9820 JH |
2429 | |
2430 | return nr; | |
2431 | } | |
2432 | ||
cbd34da7 CH |
2433 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2434 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2435 | unsigned long sz) | |
2436 | { | |
2437 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2438 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2439 | } | |
2440 | ||
2441 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2442 | unsigned long end, unsigned int flags, |
2443 | struct page **pages, int *nr) | |
cbd34da7 CH |
2444 | { |
2445 | unsigned long pte_end; | |
2446 | struct page *head, *page; | |
2447 | pte_t pte; | |
2448 | int refs; | |
2449 | ||
2450 | pte_end = (addr + sz) & ~(sz-1); | |
2451 | if (pte_end < end) | |
2452 | end = pte_end; | |
2453 | ||
55ca2263 | 2454 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2455 | |
0cd22afd | 2456 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2457 | return 0; |
2458 | ||
2459 | /* hugepages are never "special" */ | |
2460 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2461 | ||
cbd34da7 | 2462 | head = pte_page(pte); |
c228afb1 | 2463 | page = nth_page(head, (addr & (sz - 1)) >> PAGE_SHIFT); |
a43e9820 | 2464 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2465 | |
3faa52c0 | 2466 | head = try_grab_compound_head(head, refs, flags); |
a43e9820 | 2467 | if (!head) |
cbd34da7 | 2468 | return 0; |
cbd34da7 CH |
2469 | |
2470 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
3b78d834 | 2471 | put_compound_head(head, refs, flags); |
cbd34da7 CH |
2472 | return 0; |
2473 | } | |
2474 | ||
a43e9820 | 2475 | *nr += refs; |
520b4a44 | 2476 | SetPageReferenced(head); |
cbd34da7 CH |
2477 | return 1; |
2478 | } | |
2479 | ||
2480 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2481 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2482 | struct page **pages, int *nr) |
2483 | { | |
2484 | pte_t *ptep; | |
2485 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2486 | unsigned long next; | |
2487 | ||
2488 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2489 | do { | |
2490 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2491 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2492 | return 0; |
2493 | } while (ptep++, addr = next, addr != end); | |
2494 | ||
2495 | return 1; | |
2496 | } | |
2497 | #else | |
2498 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2499 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2500 | struct page **pages, int *nr) |
2501 | { | |
2502 | return 0; | |
2503 | } | |
2504 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2505 | ||
2667f50e | 2506 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2507 | unsigned long end, unsigned int flags, |
2508 | struct page **pages, int *nr) | |
2667f50e | 2509 | { |
ddc58f27 | 2510 | struct page *head, *page; |
2667f50e SC |
2511 | int refs; |
2512 | ||
b798bec4 | 2513 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2514 | return 0; |
2515 | ||
7af75561 IW |
2516 | if (pmd_devmap(orig)) { |
2517 | if (unlikely(flags & FOLL_LONGTERM)) | |
2518 | return 0; | |
86dfbed4 JH |
2519 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2520 | pages, nr); | |
7af75561 | 2521 | } |
b59f65fa | 2522 | |
c228afb1 | 2523 | page = nth_page(pmd_page(orig), (addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2524 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2525 | |
3faa52c0 | 2526 | head = try_grab_compound_head(pmd_page(orig), refs, flags); |
a43e9820 | 2527 | if (!head) |
2667f50e | 2528 | return 0; |
2667f50e SC |
2529 | |
2530 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
3b78d834 | 2531 | put_compound_head(head, refs, flags); |
2667f50e SC |
2532 | return 0; |
2533 | } | |
2534 | ||
a43e9820 | 2535 | *nr += refs; |
e9348053 | 2536 | SetPageReferenced(head); |
2667f50e SC |
2537 | return 1; |
2538 | } | |
2539 | ||
2540 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2541 | unsigned long end, unsigned int flags, |
2542 | struct page **pages, int *nr) | |
2667f50e | 2543 | { |
ddc58f27 | 2544 | struct page *head, *page; |
2667f50e SC |
2545 | int refs; |
2546 | ||
b798bec4 | 2547 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2548 | return 0; |
2549 | ||
7af75561 IW |
2550 | if (pud_devmap(orig)) { |
2551 | if (unlikely(flags & FOLL_LONGTERM)) | |
2552 | return 0; | |
86dfbed4 JH |
2553 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2554 | pages, nr); | |
7af75561 | 2555 | } |
b59f65fa | 2556 | |
c228afb1 | 2557 | page = nth_page(pud_page(orig), (addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2558 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2559 | |
3faa52c0 | 2560 | head = try_grab_compound_head(pud_page(orig), refs, flags); |
a43e9820 | 2561 | if (!head) |
2667f50e | 2562 | return 0; |
2667f50e SC |
2563 | |
2564 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
3b78d834 | 2565 | put_compound_head(head, refs, flags); |
2667f50e SC |
2566 | return 0; |
2567 | } | |
2568 | ||
a43e9820 | 2569 | *nr += refs; |
e9348053 | 2570 | SetPageReferenced(head); |
2667f50e SC |
2571 | return 1; |
2572 | } | |
2573 | ||
f30c59e9 | 2574 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2575 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2576 | struct page **pages, int *nr) |
2577 | { | |
2578 | int refs; | |
ddc58f27 | 2579 | struct page *head, *page; |
f30c59e9 | 2580 | |
b798bec4 | 2581 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2582 | return 0; |
2583 | ||
b59f65fa | 2584 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2585 | |
c228afb1 | 2586 | page = nth_page(pgd_page(orig), (addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2587 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2588 | |
3faa52c0 | 2589 | head = try_grab_compound_head(pgd_page(orig), refs, flags); |
a43e9820 | 2590 | if (!head) |
f30c59e9 | 2591 | return 0; |
f30c59e9 AK |
2592 | |
2593 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
3b78d834 | 2594 | put_compound_head(head, refs, flags); |
f30c59e9 AK |
2595 | return 0; |
2596 | } | |
2597 | ||
a43e9820 | 2598 | *nr += refs; |
e9348053 | 2599 | SetPageReferenced(head); |
f30c59e9 AK |
2600 | return 1; |
2601 | } | |
2602 | ||
d3f7b1bb | 2603 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2604 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2605 | { |
2606 | unsigned long next; | |
2607 | pmd_t *pmdp; | |
2608 | ||
d3f7b1bb | 2609 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2610 | do { |
38c5ce93 | 2611 | pmd_t pmd = READ_ONCE(*pmdp); |
2667f50e SC |
2612 | |
2613 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2614 | if (!pmd_present(pmd)) |
2667f50e SC |
2615 | return 0; |
2616 | ||
414fd080 YZ |
2617 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2618 | pmd_devmap(pmd))) { | |
2667f50e SC |
2619 | /* |
2620 | * NUMA hinting faults need to be handled in the GUP | |
2621 | * slowpath for accounting purposes and so that they | |
2622 | * can be serialised against THP migration. | |
2623 | */ | |
8a0516ed | 2624 | if (pmd_protnone(pmd)) |
2667f50e SC |
2625 | return 0; |
2626 | ||
b798bec4 | 2627 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2628 | pages, nr)) |
2629 | return 0; | |
2630 | ||
f30c59e9 AK |
2631 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2632 | /* | |
2633 | * architecture have different format for hugetlbfs | |
2634 | * pmd format and THP pmd format | |
2635 | */ | |
2636 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2637 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2638 | return 0; |
b798bec4 | 2639 | } else if (!gup_pte_range(pmd, addr, next, flags, pages, nr)) |
2923117b | 2640 | return 0; |
2667f50e SC |
2641 | } while (pmdp++, addr = next, addr != end); |
2642 | ||
2643 | return 1; | |
2644 | } | |
2645 | ||
d3f7b1bb | 2646 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2647 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2648 | { |
2649 | unsigned long next; | |
2650 | pud_t *pudp; | |
2651 | ||
d3f7b1bb | 2652 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2653 | do { |
e37c6982 | 2654 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2655 | |
2656 | next = pud_addr_end(addr, end); | |
15494520 | 2657 | if (unlikely(!pud_present(pud))) |
2667f50e | 2658 | return 0; |
f30c59e9 | 2659 | if (unlikely(pud_huge(pud))) { |
b798bec4 | 2660 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2661 | pages, nr)) |
2662 | return 0; | |
2663 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2664 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2665 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2666 | return 0; |
d3f7b1bb | 2667 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2668 | return 0; |
2669 | } while (pudp++, addr = next, addr != end); | |
2670 | ||
2671 | return 1; | |
2672 | } | |
2673 | ||
d3f7b1bb | 2674 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2675 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2676 | { |
2677 | unsigned long next; | |
2678 | p4d_t *p4dp; | |
2679 | ||
d3f7b1bb | 2680 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
2681 | do { |
2682 | p4d_t p4d = READ_ONCE(*p4dp); | |
2683 | ||
2684 | next = p4d_addr_end(addr, end); | |
2685 | if (p4d_none(p4d)) | |
2686 | return 0; | |
2687 | BUILD_BUG_ON(p4d_huge(p4d)); | |
2688 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
2689 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 2690 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 2691 | return 0; |
d3f7b1bb | 2692 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
2693 | return 0; |
2694 | } while (p4dp++, addr = next, addr != end); | |
2695 | ||
2696 | return 1; | |
2697 | } | |
2698 | ||
5b65c467 | 2699 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 2700 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
2701 | { |
2702 | unsigned long next; | |
2703 | pgd_t *pgdp; | |
2704 | ||
2705 | pgdp = pgd_offset(current->mm, addr); | |
2706 | do { | |
2707 | pgd_t pgd = READ_ONCE(*pgdp); | |
2708 | ||
2709 | next = pgd_addr_end(addr, end); | |
2710 | if (pgd_none(pgd)) | |
2711 | return; | |
2712 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 2713 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
2714 | pages, nr)) |
2715 | return; | |
2716 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
2717 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 2718 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 2719 | return; |
d3f7b1bb | 2720 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
2721 | return; |
2722 | } while (pgdp++, addr = next, addr != end); | |
2723 | } | |
050a9adc CH |
2724 | #else |
2725 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
2726 | unsigned int flags, struct page **pages, int *nr) | |
2727 | { | |
2728 | } | |
2729 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
2730 | |
2731 | #ifndef gup_fast_permitted | |
2732 | /* | |
dadbb612 | 2733 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
2734 | * we need to fall back to the slow version: |
2735 | */ | |
26f4c328 | 2736 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 2737 | { |
26f4c328 | 2738 | return true; |
5b65c467 KS |
2739 | } |
2740 | #endif | |
2741 | ||
7af75561 IW |
2742 | static int __gup_longterm_unlocked(unsigned long start, int nr_pages, |
2743 | unsigned int gup_flags, struct page **pages) | |
2744 | { | |
2745 | int ret; | |
2746 | ||
2747 | /* | |
2748 | * FIXME: FOLL_LONGTERM does not work with | |
2749 | * get_user_pages_unlocked() (see comments in that function) | |
2750 | */ | |
2751 | if (gup_flags & FOLL_LONGTERM) { | |
d8ed45c5 | 2752 | mmap_read_lock(current->mm); |
64019a2e | 2753 | ret = __gup_longterm_locked(current->mm, |
7af75561 IW |
2754 | start, nr_pages, |
2755 | pages, NULL, gup_flags); | |
d8ed45c5 | 2756 | mmap_read_unlock(current->mm); |
7af75561 IW |
2757 | } else { |
2758 | ret = get_user_pages_unlocked(start, nr_pages, | |
2759 | pages, gup_flags); | |
2760 | } | |
2761 | ||
2762 | return ret; | |
2763 | } | |
2764 | ||
c28b1fc7 JG |
2765 | static unsigned long lockless_pages_from_mm(unsigned long start, |
2766 | unsigned long end, | |
2767 | unsigned int gup_flags, | |
2768 | struct page **pages) | |
2769 | { | |
2770 | unsigned long flags; | |
2771 | int nr_pinned = 0; | |
57efa1fe | 2772 | unsigned seq; |
c28b1fc7 JG |
2773 | |
2774 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
2775 | !gup_fast_permitted(start, end)) | |
2776 | return 0; | |
2777 | ||
57efa1fe JG |
2778 | if (gup_flags & FOLL_PIN) { |
2779 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
2780 | if (seq & 1) | |
2781 | return 0; | |
2782 | } | |
2783 | ||
c28b1fc7 JG |
2784 | /* |
2785 | * Disable interrupts. The nested form is used, in order to allow full, | |
2786 | * general purpose use of this routine. | |
2787 | * | |
2788 | * With interrupts disabled, we block page table pages from being freed | |
2789 | * from under us. See struct mmu_table_batch comments in | |
2790 | * include/asm-generic/tlb.h for more details. | |
2791 | * | |
2792 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
2793 | * that come from THPs splitting. | |
2794 | */ | |
2795 | local_irq_save(flags); | |
2796 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
2797 | local_irq_restore(flags); | |
57efa1fe JG |
2798 | |
2799 | /* | |
2800 | * When pinning pages for DMA there could be a concurrent write protect | |
2801 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
2802 | */ | |
2803 | if (gup_flags & FOLL_PIN) { | |
2804 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
2805 | unpin_user_pages(pages, nr_pinned); | |
2806 | return 0; | |
2807 | } | |
2808 | } | |
c28b1fc7 JG |
2809 | return nr_pinned; |
2810 | } | |
2811 | ||
2812 | static int internal_get_user_pages_fast(unsigned long start, | |
2813 | unsigned long nr_pages, | |
eddb1c22 JH |
2814 | unsigned int gup_flags, |
2815 | struct page **pages) | |
2667f50e | 2816 | { |
c28b1fc7 JG |
2817 | unsigned long len, end; |
2818 | unsigned long nr_pinned; | |
2819 | int ret; | |
2667f50e | 2820 | |
f4000fdf | 2821 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef | 2822 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
55b8fe70 | 2823 | FOLL_FAST_ONLY | FOLL_NOFAULT))) |
817be129 CH |
2824 | return -EINVAL; |
2825 | ||
a458b76a AA |
2826 | if (gup_flags & FOLL_PIN) |
2827 | mm_set_has_pinned_flag(¤t->mm->flags); | |
008cfe44 | 2828 | |
f81cd178 | 2829 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 2830 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 2831 | |
f455c854 | 2832 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
2833 | len = nr_pages << PAGE_SHIFT; |
2834 | if (check_add_overflow(start, len, &end)) | |
c61611f7 | 2835 | return 0; |
96d4f267 | 2836 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 2837 | return -EFAULT; |
73e10a61 | 2838 | |
c28b1fc7 JG |
2839 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
2840 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
2841 | return nr_pinned; | |
2667f50e | 2842 | |
c28b1fc7 JG |
2843 | /* Slow path: try to get the remaining pages with get_user_pages */ |
2844 | start += nr_pinned << PAGE_SHIFT; | |
2845 | pages += nr_pinned; | |
2846 | ret = __gup_longterm_unlocked(start, nr_pages - nr_pinned, gup_flags, | |
2847 | pages); | |
2848 | if (ret < 0) { | |
2849 | /* | |
2850 | * The caller has to unpin the pages we already pinned so | |
2851 | * returning -errno is not an option | |
2852 | */ | |
2853 | if (nr_pinned) | |
2854 | return nr_pinned; | |
2855 | return ret; | |
2667f50e | 2856 | } |
c28b1fc7 | 2857 | return ret + nr_pinned; |
2667f50e | 2858 | } |
c28b1fc7 | 2859 | |
dadbb612 SJ |
2860 | /** |
2861 | * get_user_pages_fast_only() - pin user pages in memory | |
2862 | * @start: starting user address | |
2863 | * @nr_pages: number of pages from start to pin | |
2864 | * @gup_flags: flags modifying pin behaviour | |
2865 | * @pages: array that receives pointers to the pages pinned. | |
2866 | * Should be at least nr_pages long. | |
2867 | * | |
9e1f0580 JH |
2868 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
2869 | * the regular GUP. | |
2870 | * Note a difference with get_user_pages_fast: this always returns the | |
2871 | * number of pages pinned, 0 if no pages were pinned. | |
2872 | * | |
2873 | * If the architecture does not support this function, simply return with no | |
2874 | * pages pinned. | |
2875 | * | |
2876 | * Careful, careful! COW breaking can go either way, so a non-write | |
2877 | * access can get ambiguous page results. If you call this function without | |
2878 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
2879 | */ | |
dadbb612 SJ |
2880 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
2881 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 2882 | { |
376a34ef | 2883 | int nr_pinned; |
9e1f0580 JH |
2884 | /* |
2885 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
2886 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
2887 | * |
2888 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2889 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 2890 | */ |
dadbb612 | 2891 | gup_flags |= FOLL_GET | FOLL_FAST_ONLY; |
9e1f0580 | 2892 | |
376a34ef JH |
2893 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
2894 | pages); | |
9e1f0580 JH |
2895 | |
2896 | /* | |
376a34ef JH |
2897 | * As specified in the API description above, this routine is not |
2898 | * allowed to return negative values. However, the common core | |
2899 | * routine internal_get_user_pages_fast() *can* return -errno. | |
2900 | * Therefore, correct for that here: | |
9e1f0580 | 2901 | */ |
376a34ef JH |
2902 | if (nr_pinned < 0) |
2903 | nr_pinned = 0; | |
9e1f0580 JH |
2904 | |
2905 | return nr_pinned; | |
2906 | } | |
dadbb612 | 2907 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 2908 | |
eddb1c22 JH |
2909 | /** |
2910 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
2911 | * @start: starting user address |
2912 | * @nr_pages: number of pages from start to pin | |
2913 | * @gup_flags: flags modifying pin behaviour | |
2914 | * @pages: array that receives pointers to the pages pinned. | |
2915 | * Should be at least nr_pages long. | |
eddb1c22 | 2916 | * |
c1e8d7c6 | 2917 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
2918 | * If not successful, it will fall back to taking the lock and |
2919 | * calling get_user_pages(). | |
2920 | * | |
2921 | * Returns number of pages pinned. This may be fewer than the number requested. | |
2922 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
2923 | * -errno. | |
2924 | */ | |
2925 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
2926 | unsigned int gup_flags, struct page **pages) | |
2927 | { | |
447f3e45 | 2928 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2929 | return -EINVAL; |
2930 | ||
94202f12 JH |
2931 | /* |
2932 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
2933 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
2934 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
2935 | * request. | |
2936 | */ | |
2937 | gup_flags |= FOLL_GET; | |
eddb1c22 JH |
2938 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
2939 | } | |
050a9adc | 2940 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
2941 | |
2942 | /** | |
2943 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
2944 | * | |
3faa52c0 JH |
2945 | * @start: starting user address |
2946 | * @nr_pages: number of pages from start to pin | |
2947 | * @gup_flags: flags modifying pin behaviour | |
2948 | * @pages: array that receives pointers to the pages pinned. | |
2949 | * Should be at least nr_pages long. | |
2950 | * | |
2951 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
2952 | * get_user_pages_fast() for documentation on the function arguments, because | |
2953 | * the arguments here are identical. | |
2954 | * | |
2955 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2956 | * see Documentation/core-api/pin_user_pages.rst for further details. |
eddb1c22 JH |
2957 | */ |
2958 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
2959 | unsigned int gup_flags, struct page **pages) | |
2960 | { | |
3faa52c0 JH |
2961 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2962 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2963 | return -EINVAL; | |
2964 | ||
2965 | gup_flags |= FOLL_PIN; | |
2966 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); | |
eddb1c22 JH |
2967 | } |
2968 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
2969 | ||
104acc32 | 2970 | /* |
dadbb612 SJ |
2971 | * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior |
2972 | * is the same, except that this one sets FOLL_PIN instead of FOLL_GET. | |
104acc32 JH |
2973 | * |
2974 | * The API rules are the same, too: no negative values may be returned. | |
2975 | */ | |
2976 | int pin_user_pages_fast_only(unsigned long start, int nr_pages, | |
2977 | unsigned int gup_flags, struct page **pages) | |
2978 | { | |
2979 | int nr_pinned; | |
2980 | ||
2981 | /* | |
2982 | * FOLL_GET and FOLL_PIN are mutually exclusive. Note that the API | |
2983 | * rules require returning 0, rather than -errno: | |
2984 | */ | |
2985 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2986 | return 0; | |
2987 | /* | |
2988 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2989 | * this routine: no fall back to regular ("slow") GUP. | |
2990 | */ | |
2991 | gup_flags |= (FOLL_PIN | FOLL_FAST_ONLY); | |
2992 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, | |
2993 | pages); | |
2994 | /* | |
2995 | * This routine is not allowed to return negative values. However, | |
2996 | * internal_get_user_pages_fast() *can* return -errno. Therefore, | |
2997 | * correct for that here: | |
2998 | */ | |
2999 | if (nr_pinned < 0) | |
3000 | nr_pinned = 0; | |
3001 | ||
3002 | return nr_pinned; | |
3003 | } | |
3004 | EXPORT_SYMBOL_GPL(pin_user_pages_fast_only); | |
3005 | ||
eddb1c22 | 3006 | /** |
64019a2e | 3007 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 3008 | * |
3faa52c0 JH |
3009 | * @mm: mm_struct of target mm |
3010 | * @start: starting user address | |
3011 | * @nr_pages: number of pages from start to pin | |
3012 | * @gup_flags: flags modifying lookup behaviour | |
3013 | * @pages: array that receives pointers to the pages pinned. | |
3014 | * Should be at least nr_pages long. Or NULL, if caller | |
3015 | * only intends to ensure the pages are faulted in. | |
3016 | * @vmas: array of pointers to vmas corresponding to each page. | |
3017 | * Or NULL if the caller does not require them. | |
3018 | * @locked: pointer to lock flag indicating whether lock is held and | |
3019 | * subsequently whether VM_FAULT_RETRY functionality can be | |
3020 | * utilised. Lock must initially be held. | |
3021 | * | |
3022 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
3023 | * get_user_pages_remote() for documentation on the function arguments, because | |
3024 | * the arguments here are identical. | |
3025 | * | |
3026 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3027 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 | 3028 | */ |
64019a2e | 3029 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
3030 | unsigned long start, unsigned long nr_pages, |
3031 | unsigned int gup_flags, struct page **pages, | |
3032 | struct vm_area_struct **vmas, int *locked) | |
3033 | { | |
3faa52c0 JH |
3034 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3035 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3036 | return -EINVAL; | |
3037 | ||
3038 | gup_flags |= FOLL_PIN; | |
64019a2e | 3039 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
3faa52c0 | 3040 | pages, vmas, locked); |
eddb1c22 JH |
3041 | } |
3042 | EXPORT_SYMBOL(pin_user_pages_remote); | |
3043 | ||
3044 | /** | |
3045 | * pin_user_pages() - pin user pages in memory for use by other devices | |
3046 | * | |
3faa52c0 JH |
3047 | * @start: starting user address |
3048 | * @nr_pages: number of pages from start to pin | |
3049 | * @gup_flags: flags modifying lookup behaviour | |
3050 | * @pages: array that receives pointers to the pages pinned. | |
3051 | * Should be at least nr_pages long. Or NULL, if caller | |
3052 | * only intends to ensure the pages are faulted in. | |
3053 | * @vmas: array of pointers to vmas corresponding to each page. | |
3054 | * Or NULL if the caller does not require them. | |
3055 | * | |
3056 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
3057 | * FOLL_PIN is set. | |
3058 | * | |
3059 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3060 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 JH |
3061 | */ |
3062 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
3063 | unsigned int gup_flags, struct page **pages, | |
3064 | struct vm_area_struct **vmas) | |
3065 | { | |
3faa52c0 JH |
3066 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3067 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3068 | return -EINVAL; | |
3069 | ||
3070 | gup_flags |= FOLL_PIN; | |
64019a2e | 3071 | return __gup_longterm_locked(current->mm, start, nr_pages, |
3faa52c0 | 3072 | pages, vmas, gup_flags); |
eddb1c22 JH |
3073 | } |
3074 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
3075 | |
3076 | /* | |
3077 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
3078 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
3079 | * FOLL_PIN and rejects FOLL_GET. | |
3080 | */ | |
3081 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
3082 | struct page **pages, unsigned int gup_flags) | |
3083 | { | |
3084 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
3085 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3086 | return -EINVAL; | |
3087 | ||
3088 | gup_flags |= FOLL_PIN; | |
3089 | return get_user_pages_unlocked(start, nr_pages, pages, gup_flags); | |
3090 | } | |
3091 | EXPORT_SYMBOL(pin_user_pages_unlocked); | |
420c2091 JH |
3092 | |
3093 | /* | |
3094 | * pin_user_pages_locked() is the FOLL_PIN variant of get_user_pages_locked(). | |
3095 | * Behavior is the same, except that this one sets FOLL_PIN and rejects | |
3096 | * FOLL_GET. | |
3097 | */ | |
3098 | long pin_user_pages_locked(unsigned long start, unsigned long nr_pages, | |
3099 | unsigned int gup_flags, struct page **pages, | |
3100 | int *locked) | |
3101 | { | |
3102 | /* | |
3103 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
3104 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
3105 | * vmas. As there are no users of this flag in this call we simply | |
3106 | * disallow this option for now. | |
3107 | */ | |
3108 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
3109 | return -EINVAL; | |
3110 | ||
3111 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
3112 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3113 | return -EINVAL; | |
3114 | ||
3115 | gup_flags |= FOLL_PIN; | |
64019a2e | 3116 | return __get_user_pages_locked(current->mm, start, nr_pages, |
420c2091 JH |
3117 | pages, NULL, locked, |
3118 | gup_flags | FOLL_TOUCH); | |
3119 | } | |
3120 | EXPORT_SYMBOL(pin_user_pages_locked); |