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> | |
13 | ||
174cd4b1 | 14 | #include <linux/sched/signal.h> |
2667f50e | 15 | #include <linux/rwsem.h> |
f30c59e9 | 16 | #include <linux/hugetlb.h> |
9a4e9f3b AK |
17 | #include <linux/migrate.h> |
18 | #include <linux/mm_inline.h> | |
19 | #include <linux/sched/mm.h> | |
1027e443 | 20 | |
33a709b2 | 21 | #include <asm/mmu_context.h> |
1027e443 | 22 | #include <asm/tlbflush.h> |
2667f50e | 23 | |
4bbd4c77 KS |
24 | #include "internal.h" |
25 | ||
df06b37f KB |
26 | struct follow_page_context { |
27 | struct dev_pagemap *pgmap; | |
28 | unsigned int page_mask; | |
29 | }; | |
30 | ||
47e29d32 JH |
31 | static void hpage_pincount_add(struct page *page, int refs) |
32 | { | |
33 | VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); | |
34 | VM_BUG_ON_PAGE(page != compound_head(page), page); | |
35 | ||
36 | atomic_add(refs, compound_pincount_ptr(page)); | |
37 | } | |
38 | ||
39 | static void hpage_pincount_sub(struct page *page, int refs) | |
40 | { | |
41 | VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); | |
42 | VM_BUG_ON_PAGE(page != compound_head(page), page); | |
43 | ||
44 | atomic_sub(refs, compound_pincount_ptr(page)); | |
45 | } | |
46 | ||
a707cdd5 JH |
47 | /* |
48 | * Return the compound head page with ref appropriately incremented, | |
49 | * or NULL if that failed. | |
50 | */ | |
51 | static inline struct page *try_get_compound_head(struct page *page, int refs) | |
52 | { | |
53 | struct page *head = compound_head(page); | |
54 | ||
55 | if (WARN_ON_ONCE(page_ref_count(head) < 0)) | |
56 | return NULL; | |
57 | if (unlikely(!page_cache_add_speculative(head, refs))) | |
58 | return NULL; | |
59 | return head; | |
60 | } | |
61 | ||
3faa52c0 JH |
62 | /* |
63 | * try_grab_compound_head() - attempt to elevate a page's refcount, by a | |
64 | * flags-dependent amount. | |
65 | * | |
66 | * "grab" names in this file mean, "look at flags to decide whether to use | |
67 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
68 | * | |
69 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
70 | * same time. (That's true throughout the get_user_pages*() and | |
71 | * pin_user_pages*() APIs.) Cases: | |
72 | * | |
73 | * FOLL_GET: page's refcount will be incremented by 1. | |
74 | * FOLL_PIN: page's refcount will be incremented by GUP_PIN_COUNTING_BIAS. | |
75 | * | |
76 | * Return: head page (with refcount appropriately incremented) for success, or | |
77 | * NULL upon failure. If neither FOLL_GET nor FOLL_PIN was set, that's | |
78 | * considered failure, and furthermore, a likely bug in the caller, so a warning | |
79 | * is also emitted. | |
80 | */ | |
0fa5bc40 JM |
81 | __maybe_unused struct page *try_grab_compound_head(struct page *page, |
82 | int refs, unsigned int flags) | |
3faa52c0 JH |
83 | { |
84 | if (flags & FOLL_GET) | |
85 | return try_get_compound_head(page, refs); | |
86 | else if (flags & FOLL_PIN) { | |
1970dc6f JH |
87 | int orig_refs = refs; |
88 | ||
df3a0a21 PL |
89 | /* |
90 | * Can't do FOLL_LONGTERM + FOLL_PIN with CMA in the gup fast | |
91 | * path, so fail and let the caller fall back to the slow path. | |
92 | */ | |
93 | if (unlikely(flags & FOLL_LONGTERM) && | |
94 | is_migrate_cma_page(page)) | |
95 | return NULL; | |
96 | ||
47e29d32 JH |
97 | /* |
98 | * When pinning a compound page of order > 1 (which is what | |
99 | * hpage_pincount_available() checks for), use an exact count to | |
100 | * track it, via hpage_pincount_add/_sub(). | |
101 | * | |
102 | * However, be sure to *also* increment the normal page refcount | |
103 | * field at least once, so that the page really is pinned. | |
104 | */ | |
105 | if (!hpage_pincount_available(page)) | |
106 | refs *= GUP_PIN_COUNTING_BIAS; | |
107 | ||
108 | page = try_get_compound_head(page, refs); | |
109 | if (!page) | |
110 | return NULL; | |
111 | ||
112 | if (hpage_pincount_available(page)) | |
113 | hpage_pincount_add(page, refs); | |
114 | ||
1970dc6f JH |
115 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, |
116 | orig_refs); | |
117 | ||
47e29d32 | 118 | return page; |
3faa52c0 JH |
119 | } |
120 | ||
121 | WARN_ON_ONCE(1); | |
122 | return NULL; | |
123 | } | |
124 | ||
4509b42c JG |
125 | static void put_compound_head(struct page *page, int refs, unsigned int flags) |
126 | { | |
127 | if (flags & FOLL_PIN) { | |
128 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, | |
129 | refs); | |
130 | ||
131 | if (hpage_pincount_available(page)) | |
132 | hpage_pincount_sub(page, refs); | |
133 | else | |
134 | refs *= GUP_PIN_COUNTING_BIAS; | |
135 | } | |
136 | ||
137 | VM_BUG_ON_PAGE(page_ref_count(page) < refs, page); | |
138 | /* | |
139 | * Calling put_page() for each ref is unnecessarily slow. Only the last | |
140 | * ref needs a put_page(). | |
141 | */ | |
142 | if (refs > 1) | |
143 | page_ref_sub(page, refs - 1); | |
144 | put_page(page); | |
145 | } | |
146 | ||
3faa52c0 JH |
147 | /** |
148 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
149 | * | |
150 | * This might not do anything at all, depending on the flags argument. | |
151 | * | |
152 | * "grab" names in this file mean, "look at flags to decide whether to use | |
153 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
154 | * | |
155 | * @page: pointer to page to be grabbed | |
156 | * @flags: gup flags: these are the FOLL_* flag values. | |
157 | * | |
158 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same | |
159 | * time. Cases: | |
160 | * | |
161 | * FOLL_GET: page's refcount will be incremented by 1. | |
162 | * FOLL_PIN: page's refcount will be incremented by GUP_PIN_COUNTING_BIAS. | |
163 | * | |
164 | * Return: true for success, or if no action was required (if neither FOLL_PIN | |
165 | * nor FOLL_GET was set, nothing is done). False for failure: FOLL_GET or | |
166 | * FOLL_PIN was set, but the page could not be grabbed. | |
167 | */ | |
168 | bool __must_check try_grab_page(struct page *page, unsigned int flags) | |
169 | { | |
170 | WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN)); | |
171 | ||
172 | if (flags & FOLL_GET) | |
173 | return try_get_page(page); | |
174 | else if (flags & FOLL_PIN) { | |
47e29d32 JH |
175 | int refs = 1; |
176 | ||
3faa52c0 JH |
177 | page = compound_head(page); |
178 | ||
179 | if (WARN_ON_ONCE(page_ref_count(page) <= 0)) | |
180 | return false; | |
181 | ||
47e29d32 JH |
182 | if (hpage_pincount_available(page)) |
183 | hpage_pincount_add(page, 1); | |
184 | else | |
185 | refs = GUP_PIN_COUNTING_BIAS; | |
186 | ||
187 | /* | |
188 | * Similar to try_grab_compound_head(): even if using the | |
189 | * hpage_pincount_add/_sub() routines, be sure to | |
190 | * *also* increment the normal page refcount field at least | |
191 | * once, so that the page really is pinned. | |
192 | */ | |
193 | page_ref_add(page, refs); | |
1970dc6f JH |
194 | |
195 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1); | |
3faa52c0 JH |
196 | } |
197 | ||
198 | return true; | |
199 | } | |
200 | ||
3faa52c0 JH |
201 | /** |
202 | * unpin_user_page() - release a dma-pinned page | |
203 | * @page: pointer to page to be released | |
204 | * | |
205 | * Pages that were pinned via pin_user_pages*() must be released via either | |
206 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
207 | * that such pages can be separately tracked and uniquely handled. In | |
208 | * particular, interactions with RDMA and filesystems need special handling. | |
209 | */ | |
210 | void unpin_user_page(struct page *page) | |
211 | { | |
4509b42c | 212 | put_compound_head(compound_head(page), 1, FOLL_PIN); |
3faa52c0 JH |
213 | } |
214 | EXPORT_SYMBOL(unpin_user_page); | |
215 | ||
458a4f78 JM |
216 | static inline void compound_range_next(unsigned long i, unsigned long npages, |
217 | struct page **list, struct page **head, | |
218 | unsigned int *ntails) | |
219 | { | |
220 | struct page *next, *page; | |
221 | unsigned int nr = 1; | |
222 | ||
223 | if (i >= npages) | |
224 | return; | |
225 | ||
226 | next = *list + i; | |
227 | page = compound_head(next); | |
228 | if (PageCompound(page) && compound_order(page) >= 1) | |
229 | nr = min_t(unsigned int, | |
230 | page + compound_nr(page) - next, npages - i); | |
231 | ||
232 | *head = page; | |
233 | *ntails = nr; | |
234 | } | |
235 | ||
236 | #define for_each_compound_range(__i, __list, __npages, __head, __ntails) \ | |
237 | for (__i = 0, \ | |
238 | compound_range_next(__i, __npages, __list, &(__head), &(__ntails)); \ | |
239 | __i < __npages; __i += __ntails, \ | |
240 | compound_range_next(__i, __npages, __list, &(__head), &(__ntails))) | |
241 | ||
8745d7f6 JM |
242 | static inline void compound_next(unsigned long i, unsigned long npages, |
243 | struct page **list, struct page **head, | |
244 | unsigned int *ntails) | |
245 | { | |
246 | struct page *page; | |
247 | unsigned int nr; | |
248 | ||
249 | if (i >= npages) | |
250 | return; | |
251 | ||
252 | page = compound_head(list[i]); | |
253 | for (nr = i + 1; nr < npages; nr++) { | |
254 | if (compound_head(list[nr]) != page) | |
255 | break; | |
256 | } | |
257 | ||
258 | *head = page; | |
259 | *ntails = nr - i; | |
260 | } | |
261 | ||
262 | #define for_each_compound_head(__i, __list, __npages, __head, __ntails) \ | |
263 | for (__i = 0, \ | |
264 | compound_next(__i, __npages, __list, &(__head), &(__ntails)); \ | |
265 | __i < __npages; __i += __ntails, \ | |
266 | compound_next(__i, __npages, __list, &(__head), &(__ntails))) | |
267 | ||
fc1d8e7c | 268 | /** |
f1f6a7dd | 269 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 270 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 271 | * @npages: number of pages in the @pages array. |
2d15eb31 | 272 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
273 | * |
274 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
275 | * variants called on that page. | |
276 | * | |
277 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 278 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
279 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
280 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 281 | * |
f1f6a7dd | 282 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 283 | * |
2d15eb31 | 284 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
285 | * required, then the caller should a) verify that this is really correct, | |
286 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 287 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
288 | * |
289 | */ | |
f1f6a7dd JH |
290 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
291 | bool make_dirty) | |
fc1d8e7c | 292 | { |
2d15eb31 | 293 | unsigned long index; |
31b912de JM |
294 | struct page *head; |
295 | unsigned int ntails; | |
2d15eb31 | 296 | |
297 | if (!make_dirty) { | |
f1f6a7dd | 298 | unpin_user_pages(pages, npages); |
2d15eb31 | 299 | return; |
300 | } | |
301 | ||
31b912de | 302 | for_each_compound_head(index, pages, npages, head, ntails) { |
2d15eb31 | 303 | /* |
304 | * Checking PageDirty at this point may race with | |
305 | * clear_page_dirty_for_io(), but that's OK. Two key | |
306 | * cases: | |
307 | * | |
308 | * 1) This code sees the page as already dirty, so it | |
309 | * skips the call to set_page_dirty(). That could happen | |
310 | * because clear_page_dirty_for_io() called | |
311 | * page_mkclean(), followed by set_page_dirty(). | |
312 | * However, now the page is going to get written back, | |
313 | * which meets the original intention of setting it | |
314 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
315 | * on to call TestClearPageDirty(), and write the page | |
316 | * back. | |
317 | * | |
318 | * 2) This code sees the page as clean, so it calls | |
319 | * set_page_dirty(). The page stays dirty, despite being | |
320 | * written back, so it gets written back again in the | |
321 | * next writeback cycle. This is harmless. | |
322 | */ | |
31b912de JM |
323 | if (!PageDirty(head)) |
324 | set_page_dirty_lock(head); | |
325 | put_compound_head(head, ntails, FOLL_PIN); | |
2d15eb31 | 326 | } |
fc1d8e7c | 327 | } |
f1f6a7dd | 328 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c | 329 | |
458a4f78 JM |
330 | /** |
331 | * unpin_user_page_range_dirty_lock() - release and optionally dirty | |
332 | * gup-pinned page range | |
333 | * | |
334 | * @page: the starting page of a range maybe marked dirty, and definitely released. | |
335 | * @npages: number of consecutive pages to release. | |
336 | * @make_dirty: whether to mark the pages dirty | |
337 | * | |
338 | * "gup-pinned page range" refers to a range of pages that has had one of the | |
339 | * pin_user_pages() variants called on that page. | |
340 | * | |
341 | * For the page ranges defined by [page .. page+npages], make that range (or | |
342 | * its head pages, if a compound page) dirty, if @make_dirty is true, and if the | |
343 | * page range was previously listed as clean. | |
344 | * | |
345 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is | |
346 | * required, then the caller should a) verify that this is really correct, | |
347 | * because _lock() is usually required, and b) hand code it: | |
348 | * set_page_dirty_lock(), unpin_user_page(). | |
349 | * | |
350 | */ | |
351 | void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, | |
352 | bool make_dirty) | |
353 | { | |
354 | unsigned long index; | |
355 | struct page *head; | |
356 | unsigned int ntails; | |
357 | ||
358 | for_each_compound_range(index, &page, npages, head, ntails) { | |
359 | if (make_dirty && !PageDirty(head)) | |
360 | set_page_dirty_lock(head); | |
361 | put_compound_head(head, ntails, FOLL_PIN); | |
362 | } | |
363 | } | |
364 | EXPORT_SYMBOL(unpin_user_page_range_dirty_lock); | |
365 | ||
fc1d8e7c | 366 | /** |
f1f6a7dd | 367 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
368 | * @pages: array of pages to be marked dirty and released. |
369 | * @npages: number of pages in the @pages array. | |
370 | * | |
f1f6a7dd | 371 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 372 | * |
f1f6a7dd | 373 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 374 | */ |
f1f6a7dd | 375 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c JH |
376 | { |
377 | unsigned long index; | |
31b912de JM |
378 | struct page *head; |
379 | unsigned int ntails; | |
fc1d8e7c | 380 | |
146608bb JH |
381 | /* |
382 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
383 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
384 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
385 | */ | |
386 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
387 | return; | |
31b912de JM |
388 | |
389 | for_each_compound_head(index, pages, npages, head, ntails) | |
390 | put_compound_head(head, ntails, FOLL_PIN); | |
fc1d8e7c | 391 | } |
f1f6a7dd | 392 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 393 | |
050a9adc | 394 | #ifdef CONFIG_MMU |
69e68b4f KS |
395 | static struct page *no_page_table(struct vm_area_struct *vma, |
396 | unsigned int flags) | |
4bbd4c77 | 397 | { |
69e68b4f KS |
398 | /* |
399 | * When core dumping an enormous anonymous area that nobody | |
400 | * has touched so far, we don't want to allocate unnecessary pages or | |
401 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
402 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
403 | * But we can only make this optimization where a hole would surely | |
404 | * be zero-filled if handle_mm_fault() actually did handle it. | |
405 | */ | |
a0137f16 AK |
406 | if ((flags & FOLL_DUMP) && |
407 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
408 | return ERR_PTR(-EFAULT); |
409 | return NULL; | |
410 | } | |
4bbd4c77 | 411 | |
1027e443 KS |
412 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
413 | pte_t *pte, unsigned int flags) | |
414 | { | |
415 | /* No page to get reference */ | |
416 | if (flags & FOLL_GET) | |
417 | return -EFAULT; | |
418 | ||
419 | if (flags & FOLL_TOUCH) { | |
420 | pte_t entry = *pte; | |
421 | ||
422 | if (flags & FOLL_WRITE) | |
423 | entry = pte_mkdirty(entry); | |
424 | entry = pte_mkyoung(entry); | |
425 | ||
426 | if (!pte_same(*pte, entry)) { | |
427 | set_pte_at(vma->vm_mm, address, pte, entry); | |
428 | update_mmu_cache(vma, address, pte); | |
429 | } | |
430 | } | |
431 | ||
432 | /* Proper page table entry exists, but no corresponding struct page */ | |
433 | return -EEXIST; | |
434 | } | |
435 | ||
19be0eaf | 436 | /* |
a308c71b PX |
437 | * FOLL_FORCE can write to even unwritable pte's, but only |
438 | * after we've gone through a COW cycle and they are dirty. | |
19be0eaf LT |
439 | */ |
440 | static inline bool can_follow_write_pte(pte_t pte, unsigned int flags) | |
441 | { | |
a308c71b PX |
442 | return pte_write(pte) || |
443 | ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte)); | |
19be0eaf LT |
444 | } |
445 | ||
69e68b4f | 446 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
447 | unsigned long address, pmd_t *pmd, unsigned int flags, |
448 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
449 | { |
450 | struct mm_struct *mm = vma->vm_mm; | |
451 | struct page *page; | |
452 | spinlock_t *ptl; | |
453 | pte_t *ptep, pte; | |
f28d4363 | 454 | int ret; |
4bbd4c77 | 455 | |
eddb1c22 JH |
456 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
457 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
458 | (FOLL_PIN | FOLL_GET))) | |
459 | return ERR_PTR(-EINVAL); | |
69e68b4f | 460 | retry: |
4bbd4c77 | 461 | if (unlikely(pmd_bad(*pmd))) |
69e68b4f | 462 | return no_page_table(vma, flags); |
4bbd4c77 KS |
463 | |
464 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
4bbd4c77 KS |
465 | pte = *ptep; |
466 | if (!pte_present(pte)) { | |
467 | swp_entry_t entry; | |
468 | /* | |
469 | * KSM's break_ksm() relies upon recognizing a ksm page | |
470 | * even while it is being migrated, so for that case we | |
471 | * need migration_entry_wait(). | |
472 | */ | |
473 | if (likely(!(flags & FOLL_MIGRATION))) | |
474 | goto no_page; | |
0661a336 | 475 | if (pte_none(pte)) |
4bbd4c77 KS |
476 | goto no_page; |
477 | entry = pte_to_swp_entry(pte); | |
478 | if (!is_migration_entry(entry)) | |
479 | goto no_page; | |
480 | pte_unmap_unlock(ptep, ptl); | |
481 | migration_entry_wait(mm, pmd, address); | |
69e68b4f | 482 | goto retry; |
4bbd4c77 | 483 | } |
8a0516ed | 484 | if ((flags & FOLL_NUMA) && pte_protnone(pte)) |
4bbd4c77 | 485 | goto no_page; |
19be0eaf | 486 | if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) { |
69e68b4f KS |
487 | pte_unmap_unlock(ptep, ptl); |
488 | return NULL; | |
489 | } | |
4bbd4c77 KS |
490 | |
491 | page = vm_normal_page(vma, address, pte); | |
3faa52c0 | 492 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 493 | /* |
3faa52c0 JH |
494 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
495 | * case since they are only valid while holding the pgmap | |
496 | * reference. | |
3565fce3 | 497 | */ |
df06b37f KB |
498 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
499 | if (*pgmap) | |
3565fce3 DW |
500 | page = pte_page(pte); |
501 | else | |
502 | goto no_page; | |
503 | } else if (unlikely(!page)) { | |
1027e443 KS |
504 | if (flags & FOLL_DUMP) { |
505 | /* Avoid special (like zero) pages in core dumps */ | |
506 | page = ERR_PTR(-EFAULT); | |
507 | goto out; | |
508 | } | |
509 | ||
510 | if (is_zero_pfn(pte_pfn(pte))) { | |
511 | page = pte_page(pte); | |
512 | } else { | |
1027e443 KS |
513 | ret = follow_pfn_pte(vma, address, ptep, flags); |
514 | page = ERR_PTR(ret); | |
515 | goto out; | |
516 | } | |
4bbd4c77 KS |
517 | } |
518 | ||
3faa52c0 JH |
519 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
520 | if (unlikely(!try_grab_page(page, flags))) { | |
521 | page = ERR_PTR(-ENOMEM); | |
522 | goto out; | |
8fde12ca | 523 | } |
f28d4363 CI |
524 | /* |
525 | * We need to make the page accessible if and only if we are going | |
526 | * to access its content (the FOLL_PIN case). Please see | |
527 | * Documentation/core-api/pin_user_pages.rst for details. | |
528 | */ | |
529 | if (flags & FOLL_PIN) { | |
530 | ret = arch_make_page_accessible(page); | |
531 | if (ret) { | |
532 | unpin_user_page(page); | |
533 | page = ERR_PTR(ret); | |
534 | goto out; | |
535 | } | |
536 | } | |
4bbd4c77 KS |
537 | if (flags & FOLL_TOUCH) { |
538 | if ((flags & FOLL_WRITE) && | |
539 | !pte_dirty(pte) && !PageDirty(page)) | |
540 | set_page_dirty(page); | |
541 | /* | |
542 | * pte_mkyoung() would be more correct here, but atomic care | |
543 | * is needed to avoid losing the dirty bit: it is easier to use | |
544 | * mark_page_accessed(). | |
545 | */ | |
546 | mark_page_accessed(page); | |
547 | } | |
de60f5f1 | 548 | if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { |
e90309c9 KS |
549 | /* Do not mlock pte-mapped THP */ |
550 | if (PageTransCompound(page)) | |
551 | goto out; | |
552 | ||
4bbd4c77 KS |
553 | /* |
554 | * The preliminary mapping check is mainly to avoid the | |
555 | * pointless overhead of lock_page on the ZERO_PAGE | |
556 | * which might bounce very badly if there is contention. | |
557 | * | |
558 | * If the page is already locked, we don't need to | |
559 | * handle it now - vmscan will handle it later if and | |
560 | * when it attempts to reclaim the page. | |
561 | */ | |
562 | if (page->mapping && trylock_page(page)) { | |
563 | lru_add_drain(); /* push cached pages to LRU */ | |
564 | /* | |
565 | * Because we lock page here, and migration is | |
566 | * blocked by the pte's page reference, and we | |
567 | * know the page is still mapped, we don't even | |
568 | * need to check for file-cache page truncation. | |
569 | */ | |
570 | mlock_vma_page(page); | |
571 | unlock_page(page); | |
572 | } | |
573 | } | |
1027e443 | 574 | out: |
4bbd4c77 | 575 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 576 | return page; |
4bbd4c77 KS |
577 | no_page: |
578 | pte_unmap_unlock(ptep, ptl); | |
579 | if (!pte_none(pte)) | |
69e68b4f KS |
580 | return NULL; |
581 | return no_page_table(vma, flags); | |
582 | } | |
583 | ||
080dbb61 AK |
584 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
585 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
586 | unsigned int flags, |
587 | struct follow_page_context *ctx) | |
69e68b4f | 588 | { |
68827280 | 589 | pmd_t *pmd, pmdval; |
69e68b4f KS |
590 | spinlock_t *ptl; |
591 | struct page *page; | |
592 | struct mm_struct *mm = vma->vm_mm; | |
593 | ||
080dbb61 | 594 | pmd = pmd_offset(pudp, address); |
68827280 HY |
595 | /* |
596 | * The READ_ONCE() will stabilize the pmdval in a register or | |
597 | * on the stack so that it will stop changing under the code. | |
598 | */ | |
599 | pmdval = READ_ONCE(*pmd); | |
600 | if (pmd_none(pmdval)) | |
69e68b4f | 601 | return no_page_table(vma, flags); |
be9d3045 | 602 | if (pmd_huge(pmdval) && is_vm_hugetlb_page(vma)) { |
e66f17ff NH |
603 | page = follow_huge_pmd(mm, address, pmd, flags); |
604 | if (page) | |
605 | return page; | |
606 | return no_page_table(vma, flags); | |
69e68b4f | 607 | } |
68827280 | 608 | if (is_hugepd(__hugepd(pmd_val(pmdval)))) { |
4dc71451 | 609 | page = follow_huge_pd(vma, address, |
68827280 | 610 | __hugepd(pmd_val(pmdval)), flags, |
4dc71451 AK |
611 | PMD_SHIFT); |
612 | if (page) | |
613 | return page; | |
614 | return no_page_table(vma, flags); | |
615 | } | |
84c3fc4e | 616 | retry: |
68827280 | 617 | if (!pmd_present(pmdval)) { |
84c3fc4e ZY |
618 | if (likely(!(flags & FOLL_MIGRATION))) |
619 | return no_page_table(vma, flags); | |
620 | VM_BUG_ON(thp_migration_supported() && | |
68827280 HY |
621 | !is_pmd_migration_entry(pmdval)); |
622 | if (is_pmd_migration_entry(pmdval)) | |
84c3fc4e | 623 | pmd_migration_entry_wait(mm, pmd); |
68827280 HY |
624 | pmdval = READ_ONCE(*pmd); |
625 | /* | |
626 | * MADV_DONTNEED may convert the pmd to null because | |
c1e8d7c6 | 627 | * mmap_lock is held in read mode |
68827280 HY |
628 | */ |
629 | if (pmd_none(pmdval)) | |
630 | return no_page_table(vma, flags); | |
84c3fc4e ZY |
631 | goto retry; |
632 | } | |
68827280 | 633 | if (pmd_devmap(pmdval)) { |
3565fce3 | 634 | ptl = pmd_lock(mm, pmd); |
df06b37f | 635 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
636 | spin_unlock(ptl); |
637 | if (page) | |
638 | return page; | |
639 | } | |
68827280 | 640 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 641 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 642 | |
68827280 | 643 | if ((flags & FOLL_NUMA) && pmd_protnone(pmdval)) |
db08f203 AK |
644 | return no_page_table(vma, flags); |
645 | ||
84c3fc4e | 646 | retry_locked: |
6742d293 | 647 | ptl = pmd_lock(mm, pmd); |
68827280 HY |
648 | if (unlikely(pmd_none(*pmd))) { |
649 | spin_unlock(ptl); | |
650 | return no_page_table(vma, flags); | |
651 | } | |
84c3fc4e ZY |
652 | if (unlikely(!pmd_present(*pmd))) { |
653 | spin_unlock(ptl); | |
654 | if (likely(!(flags & FOLL_MIGRATION))) | |
655 | return no_page_table(vma, flags); | |
656 | pmd_migration_entry_wait(mm, pmd); | |
657 | goto retry_locked; | |
658 | } | |
6742d293 KS |
659 | if (unlikely(!pmd_trans_huge(*pmd))) { |
660 | spin_unlock(ptl); | |
df06b37f | 661 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 662 | } |
4066c119 | 663 | if (flags & FOLL_SPLIT_PMD) { |
6742d293 KS |
664 | int ret; |
665 | page = pmd_page(*pmd); | |
666 | if (is_huge_zero_page(page)) { | |
667 | spin_unlock(ptl); | |
668 | ret = 0; | |
78ddc534 | 669 | split_huge_pmd(vma, pmd, address); |
337d9abf NH |
670 | if (pmd_trans_unstable(pmd)) |
671 | ret = -EBUSY; | |
4066c119 | 672 | } else { |
bfe7b00d SL |
673 | spin_unlock(ptl); |
674 | split_huge_pmd(vma, pmd, address); | |
675 | ret = pte_alloc(mm, pmd) ? -ENOMEM : 0; | |
6742d293 KS |
676 | } |
677 | ||
678 | return ret ? ERR_PTR(ret) : | |
df06b37f | 679 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 680 | } |
6742d293 KS |
681 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
682 | spin_unlock(ptl); | |
df06b37f | 683 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 684 | return page; |
4bbd4c77 KS |
685 | } |
686 | ||
080dbb61 AK |
687 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
688 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
689 | unsigned int flags, |
690 | struct follow_page_context *ctx) | |
080dbb61 AK |
691 | { |
692 | pud_t *pud; | |
693 | spinlock_t *ptl; | |
694 | struct page *page; | |
695 | struct mm_struct *mm = vma->vm_mm; | |
696 | ||
697 | pud = pud_offset(p4dp, address); | |
698 | if (pud_none(*pud)) | |
699 | return no_page_table(vma, flags); | |
be9d3045 | 700 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) { |
080dbb61 AK |
701 | page = follow_huge_pud(mm, address, pud, flags); |
702 | if (page) | |
703 | return page; | |
704 | return no_page_table(vma, flags); | |
705 | } | |
4dc71451 AK |
706 | if (is_hugepd(__hugepd(pud_val(*pud)))) { |
707 | page = follow_huge_pd(vma, address, | |
708 | __hugepd(pud_val(*pud)), flags, | |
709 | PUD_SHIFT); | |
710 | if (page) | |
711 | return page; | |
712 | return no_page_table(vma, flags); | |
713 | } | |
080dbb61 AK |
714 | if (pud_devmap(*pud)) { |
715 | ptl = pud_lock(mm, pud); | |
df06b37f | 716 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
717 | spin_unlock(ptl); |
718 | if (page) | |
719 | return page; | |
720 | } | |
721 | if (unlikely(pud_bad(*pud))) | |
722 | return no_page_table(vma, flags); | |
723 | ||
df06b37f | 724 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
725 | } |
726 | ||
080dbb61 AK |
727 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
728 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
729 | unsigned int flags, |
730 | struct follow_page_context *ctx) | |
080dbb61 AK |
731 | { |
732 | p4d_t *p4d; | |
4dc71451 | 733 | struct page *page; |
080dbb61 AK |
734 | |
735 | p4d = p4d_offset(pgdp, address); | |
736 | if (p4d_none(*p4d)) | |
737 | return no_page_table(vma, flags); | |
738 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
739 | if (unlikely(p4d_bad(*p4d))) | |
740 | return no_page_table(vma, flags); | |
741 | ||
4dc71451 AK |
742 | if (is_hugepd(__hugepd(p4d_val(*p4d)))) { |
743 | page = follow_huge_pd(vma, address, | |
744 | __hugepd(p4d_val(*p4d)), flags, | |
745 | P4D_SHIFT); | |
746 | if (page) | |
747 | return page; | |
748 | return no_page_table(vma, flags); | |
749 | } | |
df06b37f | 750 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
751 | } |
752 | ||
753 | /** | |
754 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
755 | * @vma: vm_area_struct mapping @address | |
756 | * @address: virtual address to look up | |
757 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
758 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
759 | * pointer to output page_mask | |
080dbb61 AK |
760 | * |
761 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
762 | * | |
78179556 MR |
763 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
764 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
765 | * | |
766 | * On output, the @ctx->page_mask is set according to the size of the page. | |
767 | * | |
768 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
769 | * an error pointer if there is a mapping to something not represented |
770 | * by a page descriptor (see also vm_normal_page()). | |
771 | */ | |
a7030aea | 772 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 773 | unsigned long address, unsigned int flags, |
df06b37f | 774 | struct follow_page_context *ctx) |
080dbb61 AK |
775 | { |
776 | pgd_t *pgd; | |
777 | struct page *page; | |
778 | struct mm_struct *mm = vma->vm_mm; | |
779 | ||
df06b37f | 780 | ctx->page_mask = 0; |
080dbb61 AK |
781 | |
782 | /* make this handle hugepd */ | |
783 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); | |
784 | if (!IS_ERR(page)) { | |
3faa52c0 | 785 | WARN_ON_ONCE(flags & (FOLL_GET | FOLL_PIN)); |
080dbb61 AK |
786 | return page; |
787 | } | |
788 | ||
789 | pgd = pgd_offset(mm, address); | |
790 | ||
791 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
792 | return no_page_table(vma, flags); | |
793 | ||
faaa5b62 AK |
794 | if (pgd_huge(*pgd)) { |
795 | page = follow_huge_pgd(mm, address, pgd, flags); | |
796 | if (page) | |
797 | return page; | |
798 | return no_page_table(vma, flags); | |
799 | } | |
4dc71451 AK |
800 | if (is_hugepd(__hugepd(pgd_val(*pgd)))) { |
801 | page = follow_huge_pd(vma, address, | |
802 | __hugepd(pgd_val(*pgd)), flags, | |
803 | PGDIR_SHIFT); | |
804 | if (page) | |
805 | return page; | |
806 | return no_page_table(vma, flags); | |
807 | } | |
faaa5b62 | 808 | |
df06b37f KB |
809 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
810 | } | |
811 | ||
812 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
813 | unsigned int foll_flags) | |
814 | { | |
815 | struct follow_page_context ctx = { NULL }; | |
816 | struct page *page; | |
817 | ||
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 | |
de60f5f1 EM |
888 | /* mlock all present pages, but do not fault in new pages */ |
889 | if ((*flags & (FOLL_POPULATE | FOLL_MLOCK)) == FOLL_MLOCK) | |
890 | return -ENOENT; | |
16744483 KS |
891 | if (*flags & FOLL_WRITE) |
892 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
893 | if (*flags & FOLL_REMOTE) |
894 | fault_flags |= FAULT_FLAG_REMOTE; | |
4f6da934 | 895 | if (locked) |
71335f37 | 896 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
16744483 KS |
897 | if (*flags & FOLL_NOWAIT) |
898 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 899 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
900 | /* |
901 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
902 | * can co-exist | |
903 | */ | |
234b239b ALC |
904 | fault_flags |= FAULT_FLAG_TRIED; |
905 | } | |
16744483 | 906 | |
bce617ed | 907 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
16744483 | 908 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
909 | int err = vm_fault_to_errno(ret, *flags); |
910 | ||
911 | if (err) | |
912 | return err; | |
16744483 KS |
913 | BUG(); |
914 | } | |
915 | ||
16744483 | 916 | if (ret & VM_FAULT_RETRY) { |
4f6da934 PX |
917 | if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
918 | *locked = 0; | |
16744483 KS |
919 | return -EBUSY; |
920 | } | |
921 | ||
922 | /* | |
923 | * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when | |
924 | * necessary, even if maybe_mkwrite decided not to set pte_write. We | |
925 | * can thus safely do subsequent page lookups as if they were reads. | |
926 | * But only do so when looping for pte_write is futile: in some cases | |
927 | * userspace may also be wanting to write to the gotten user page, | |
928 | * which a read fault here might prevent (a readonly page might get | |
929 | * reCOWed by userspace write). | |
930 | */ | |
931 | if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE)) | |
2923117b | 932 | *flags |= FOLL_COW; |
16744483 KS |
933 | return 0; |
934 | } | |
935 | ||
fa5bb209 KS |
936 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
937 | { | |
938 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
939 | int write = (gup_flags & FOLL_WRITE); |
940 | int foreign = (gup_flags & FOLL_REMOTE); | |
fa5bb209 KS |
941 | |
942 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
943 | return -EFAULT; | |
944 | ||
7f7ccc2c WT |
945 | if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma)) |
946 | return -EFAULT; | |
947 | ||
52650c8b JG |
948 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
949 | return -EOPNOTSUPP; | |
950 | ||
1b2ee126 | 951 | if (write) { |
fa5bb209 KS |
952 | if (!(vm_flags & VM_WRITE)) { |
953 | if (!(gup_flags & FOLL_FORCE)) | |
954 | return -EFAULT; | |
955 | /* | |
956 | * We used to let the write,force case do COW in a | |
957 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
958 | * set a breakpoint in a read-only mapping of an | |
959 | * executable, without corrupting the file (yet only | |
960 | * when that file had been opened for writing!). | |
961 | * Anon pages in shared mappings are surprising: now | |
962 | * just reject it. | |
963 | */ | |
46435364 | 964 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 965 | return -EFAULT; |
fa5bb209 KS |
966 | } |
967 | } else if (!(vm_flags & VM_READ)) { | |
968 | if (!(gup_flags & FOLL_FORCE)) | |
969 | return -EFAULT; | |
970 | /* | |
971 | * Is there actually any vma we can reach here which does not | |
972 | * have VM_MAYREAD set? | |
973 | */ | |
974 | if (!(vm_flags & VM_MAYREAD)) | |
975 | return -EFAULT; | |
976 | } | |
d61172b4 DH |
977 | /* |
978 | * gups are always data accesses, not instruction | |
979 | * fetches, so execute=false here | |
980 | */ | |
981 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 982 | return -EFAULT; |
fa5bb209 KS |
983 | return 0; |
984 | } | |
985 | ||
4bbd4c77 KS |
986 | /** |
987 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
988 | * @mm: mm_struct of target mm |
989 | * @start: starting user address | |
990 | * @nr_pages: number of pages from start to pin | |
991 | * @gup_flags: flags modifying pin behaviour | |
992 | * @pages: array that receives pointers to the pages pinned. | |
993 | * Should be at least nr_pages long. Or NULL, if caller | |
994 | * only intends to ensure the pages are faulted in. | |
995 | * @vmas: array of pointers to vmas corresponding to each page. | |
996 | * Or NULL if the caller does not require them. | |
c1e8d7c6 | 997 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 998 | * |
d2dfbe47 LX |
999 | * Returns either number of pages pinned (which may be less than the |
1000 | * number requested), or an error. Details about the return value: | |
1001 | * | |
1002 | * -- If nr_pages is 0, returns 0. | |
1003 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1004 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1005 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 1006 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
1007 | * |
1008 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1009 | * | |
c1e8d7c6 | 1010 | * @vmas are valid only as long as mmap_lock is held. |
4bbd4c77 | 1011 | * |
c1e8d7c6 | 1012 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
1013 | * |
1014 | * __get_user_pages walks a process's page tables and takes a reference to | |
1015 | * each struct page that each user address corresponds to at a given | |
1016 | * instant. That is, it takes the page that would be accessed if a user | |
1017 | * thread accesses the given user virtual address at that instant. | |
1018 | * | |
1019 | * This does not guarantee that the page exists in the user mappings when | |
1020 | * __get_user_pages returns, and there may even be a completely different | |
1021 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1022 | * and subsequently re faulted). However it does guarantee that the page | |
1023 | * won't be freed completely. And mostly callers simply care that the page | |
1024 | * contains data that was valid *at some point in time*. Typically, an IO | |
1025 | * or similar operation cannot guarantee anything stronger anyway because | |
1026 | * locks can't be held over the syscall boundary. | |
1027 | * | |
1028 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1029 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1030 | * appropriate) must be called after the page is finished with, and | |
1031 | * before put_page is called. | |
1032 | * | |
c1e8d7c6 | 1033 | * If @locked != NULL, *@locked will be set to 0 when mmap_lock is |
4f6da934 PX |
1034 | * released by an up_read(). That can happen if @gup_flags does not |
1035 | * have FOLL_NOWAIT. | |
9a95f3cf | 1036 | * |
4f6da934 | 1037 | * A caller using such a combination of @locked and @gup_flags |
c1e8d7c6 | 1038 | * must therefore hold the mmap_lock for reading only, and recognize |
9a95f3cf PC |
1039 | * when it's been released. Otherwise, it must be held for either |
1040 | * reading or writing and will not be released. | |
4bbd4c77 KS |
1041 | * |
1042 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1043 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1044 | * you need some special @gup_flags. | |
1045 | */ | |
64019a2e | 1046 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
1047 | unsigned long start, unsigned long nr_pages, |
1048 | unsigned int gup_flags, struct page **pages, | |
4f6da934 | 1049 | struct vm_area_struct **vmas, int *locked) |
4bbd4c77 | 1050 | { |
df06b37f | 1051 | long ret = 0, i = 0; |
fa5bb209 | 1052 | struct vm_area_struct *vma = NULL; |
df06b37f | 1053 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
1054 | |
1055 | if (!nr_pages) | |
1056 | return 0; | |
1057 | ||
f9652594 AK |
1058 | start = untagged_addr(start); |
1059 | ||
eddb1c22 | 1060 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 KS |
1061 | |
1062 | /* | |
1063 | * If FOLL_FORCE is set then do not force a full fault as the hinting | |
1064 | * fault information is unrelated to the reference behaviour of a task | |
1065 | * using the address space | |
1066 | */ | |
1067 | if (!(gup_flags & FOLL_FORCE)) | |
1068 | gup_flags |= FOLL_NUMA; | |
1069 | ||
4bbd4c77 | 1070 | do { |
fa5bb209 KS |
1071 | struct page *page; |
1072 | unsigned int foll_flags = gup_flags; | |
1073 | unsigned int page_increm; | |
1074 | ||
1075 | /* first iteration or cross vma bound */ | |
1076 | if (!vma || start >= vma->vm_end) { | |
1077 | vma = find_extend_vma(mm, start); | |
1078 | if (!vma && in_gate_area(mm, start)) { | |
fa5bb209 KS |
1079 | ret = get_gate_page(mm, start & PAGE_MASK, |
1080 | gup_flags, &vma, | |
1081 | pages ? &pages[i] : NULL); | |
1082 | if (ret) | |
08be37b7 | 1083 | goto out; |
df06b37f | 1084 | ctx.page_mask = 0; |
fa5bb209 KS |
1085 | goto next_page; |
1086 | } | |
4bbd4c77 | 1087 | |
52650c8b | 1088 | if (!vma) { |
df06b37f KB |
1089 | ret = -EFAULT; |
1090 | goto out; | |
1091 | } | |
52650c8b JG |
1092 | ret = check_vma_flags(vma, gup_flags); |
1093 | if (ret) | |
1094 | goto out; | |
1095 | ||
fa5bb209 KS |
1096 | if (is_vm_hugetlb_page(vma)) { |
1097 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1098 | &start, &nr_pages, i, | |
a308c71b | 1099 | gup_flags, locked); |
ad415db8 PX |
1100 | if (locked && *locked == 0) { |
1101 | /* | |
1102 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1103 | * and we've lost mmap_lock. |
ad415db8 PX |
1104 | * We must stop here. |
1105 | */ | |
1106 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
1107 | BUG_ON(ret != 0); | |
1108 | goto out; | |
1109 | } | |
fa5bb209 | 1110 | continue; |
4bbd4c77 | 1111 | } |
fa5bb209 KS |
1112 | } |
1113 | retry: | |
1114 | /* | |
1115 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1116 | * potentially allocating memory. | |
1117 | */ | |
fa45f116 | 1118 | if (fatal_signal_pending(current)) { |
d180870d | 1119 | ret = -EINTR; |
df06b37f KB |
1120 | goto out; |
1121 | } | |
fa5bb209 | 1122 | cond_resched(); |
df06b37f KB |
1123 | |
1124 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
fa5bb209 | 1125 | if (!page) { |
64019a2e | 1126 | ret = faultin_page(vma, start, &foll_flags, locked); |
fa5bb209 KS |
1127 | switch (ret) { |
1128 | case 0: | |
1129 | goto retry; | |
df06b37f KB |
1130 | case -EBUSY: |
1131 | ret = 0; | |
e4a9bc58 | 1132 | fallthrough; |
fa5bb209 KS |
1133 | case -EFAULT: |
1134 | case -ENOMEM: | |
1135 | case -EHWPOISON: | |
df06b37f | 1136 | goto out; |
fa5bb209 KS |
1137 | case -ENOENT: |
1138 | goto next_page; | |
4bbd4c77 | 1139 | } |
fa5bb209 | 1140 | BUG(); |
1027e443 KS |
1141 | } else if (PTR_ERR(page) == -EEXIST) { |
1142 | /* | |
1143 | * Proper page table entry exists, but no corresponding | |
1144 | * struct page. | |
1145 | */ | |
1146 | goto next_page; | |
1147 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1148 | ret = PTR_ERR(page); |
1149 | goto out; | |
1027e443 | 1150 | } |
fa5bb209 KS |
1151 | if (pages) { |
1152 | pages[i] = page; | |
1153 | flush_anon_page(vma, page, start); | |
1154 | flush_dcache_page(page); | |
df06b37f | 1155 | ctx.page_mask = 0; |
4bbd4c77 | 1156 | } |
4bbd4c77 | 1157 | next_page: |
fa5bb209 KS |
1158 | if (vmas) { |
1159 | vmas[i] = vma; | |
df06b37f | 1160 | ctx.page_mask = 0; |
fa5bb209 | 1161 | } |
df06b37f | 1162 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1163 | if (page_increm > nr_pages) |
1164 | page_increm = nr_pages; | |
1165 | i += page_increm; | |
1166 | start += page_increm * PAGE_SIZE; | |
1167 | nr_pages -= page_increm; | |
4bbd4c77 | 1168 | } while (nr_pages); |
df06b37f KB |
1169 | out: |
1170 | if (ctx.pgmap) | |
1171 | put_dev_pagemap(ctx.pgmap); | |
1172 | return i ? i : ret; | |
4bbd4c77 | 1173 | } |
4bbd4c77 | 1174 | |
771ab430 TK |
1175 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1176 | unsigned int fault_flags) | |
d4925e00 | 1177 | { |
1b2ee126 DH |
1178 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1179 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1180 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1181 | |
1182 | if (!(vm_flags & vma->vm_flags)) | |
1183 | return false; | |
1184 | ||
33a709b2 DH |
1185 | /* |
1186 | * The architecture might have a hardware protection | |
1b2ee126 | 1187 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1188 | * |
1189 | * gup always represents data access, not instruction | |
1190 | * fetches, so execute=false here: | |
33a709b2 | 1191 | */ |
d61172b4 | 1192 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1193 | return false; |
1194 | ||
d4925e00 DH |
1195 | return true; |
1196 | } | |
1197 | ||
adc8cb40 | 1198 | /** |
4bbd4c77 | 1199 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1200 | * @mm: mm_struct of target mm |
1201 | * @address: user address | |
1202 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1203 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1204 | * does not allow retry. If NULL, the caller must guarantee |
1205 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1206 | * |
1207 | * This is meant to be called in the specific scenario where for locking reasons | |
1208 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1209 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1210 | * trying again. | |
1211 | * | |
1212 | * Typically this is meant to be used by the futex code. | |
1213 | * | |
1214 | * The main difference with get_user_pages() is that this function will | |
1215 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1216 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1217 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1218 | * |
1219 | * This is important for some architectures where those bits also gate the | |
1220 | * access permission to the page because they are maintained in software. On | |
1221 | * such architectures, gup() will not be enough to make a subsequent access | |
1222 | * succeed. | |
1223 | * | |
c1e8d7c6 ML |
1224 | * This function will not return with an unlocked mmap_lock. So it has not the |
1225 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1226 | */ |
64019a2e | 1227 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1228 | unsigned long address, unsigned int fault_flags, |
1229 | bool *unlocked) | |
4bbd4c77 KS |
1230 | { |
1231 | struct vm_area_struct *vma; | |
2b740303 | 1232 | vm_fault_t ret, major = 0; |
4a9e1cda | 1233 | |
f9652594 AK |
1234 | address = untagged_addr(address); |
1235 | ||
4a9e1cda | 1236 | if (unlocked) |
71335f37 | 1237 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1238 | |
4a9e1cda | 1239 | retry: |
4bbd4c77 KS |
1240 | vma = find_extend_vma(mm, address); |
1241 | if (!vma || address < vma->vm_start) | |
1242 | return -EFAULT; | |
1243 | ||
d4925e00 | 1244 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1245 | return -EFAULT; |
1246 | ||
475f4dfc PX |
1247 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1248 | fatal_signal_pending(current)) | |
1249 | return -EINTR; | |
1250 | ||
bce617ed | 1251 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
4a9e1cda | 1252 | major |= ret & VM_FAULT_MAJOR; |
4bbd4c77 | 1253 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1254 | int err = vm_fault_to_errno(ret, 0); |
1255 | ||
1256 | if (err) | |
1257 | return err; | |
4bbd4c77 KS |
1258 | BUG(); |
1259 | } | |
4a9e1cda DD |
1260 | |
1261 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1262 | mmap_read_lock(mm); |
475f4dfc PX |
1263 | *unlocked = true; |
1264 | fault_flags |= FAULT_FLAG_TRIED; | |
1265 | goto retry; | |
4a9e1cda DD |
1266 | } |
1267 | ||
4bbd4c77 KS |
1268 | return 0; |
1269 | } | |
add6a0cd | 1270 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1271 | |
2d3a36a4 MH |
1272 | /* |
1273 | * Please note that this function, unlike __get_user_pages will not | |
1274 | * return 0 for nr_pages > 0 without FOLL_NOWAIT | |
1275 | */ | |
64019a2e | 1276 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1277 | unsigned long start, |
1278 | unsigned long nr_pages, | |
f0818f47 AA |
1279 | struct page **pages, |
1280 | struct vm_area_struct **vmas, | |
e716712f | 1281 | int *locked, |
0fd71a56 | 1282 | unsigned int flags) |
f0818f47 | 1283 | { |
f0818f47 AA |
1284 | long ret, pages_done; |
1285 | bool lock_dropped; | |
1286 | ||
1287 | if (locked) { | |
1288 | /* if VM_FAULT_RETRY can be returned, vmas become invalid */ | |
1289 | BUG_ON(vmas); | |
1290 | /* check caller initialized locked */ | |
1291 | BUG_ON(*locked != 1); | |
1292 | } | |
1293 | ||
008cfe44 | 1294 | if (flags & FOLL_PIN) |
a4d63c37 | 1295 | atomic_set(&mm->has_pinned, 1); |
008cfe44 | 1296 | |
eddb1c22 JH |
1297 | /* |
1298 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1299 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1300 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1301 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1302 | * | |
1303 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1304 | * that here, as any failures will be obvious enough. | |
1305 | */ | |
1306 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1307 | flags |= FOLL_GET; |
f0818f47 AA |
1308 | |
1309 | pages_done = 0; | |
1310 | lock_dropped = false; | |
1311 | for (;;) { | |
64019a2e | 1312 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
f0818f47 AA |
1313 | vmas, locked); |
1314 | if (!locked) | |
1315 | /* VM_FAULT_RETRY couldn't trigger, bypass */ | |
1316 | return ret; | |
1317 | ||
1318 | /* VM_FAULT_RETRY cannot return errors */ | |
1319 | if (!*locked) { | |
1320 | BUG_ON(ret < 0); | |
1321 | BUG_ON(ret >= nr_pages); | |
1322 | } | |
1323 | ||
f0818f47 AA |
1324 | if (ret > 0) { |
1325 | nr_pages -= ret; | |
1326 | pages_done += ret; | |
1327 | if (!nr_pages) | |
1328 | break; | |
1329 | } | |
1330 | if (*locked) { | |
96312e61 AA |
1331 | /* |
1332 | * VM_FAULT_RETRY didn't trigger or it was a | |
1333 | * FOLL_NOWAIT. | |
1334 | */ | |
f0818f47 AA |
1335 | if (!pages_done) |
1336 | pages_done = ret; | |
1337 | break; | |
1338 | } | |
df17277b MR |
1339 | /* |
1340 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1341 | * For the prefault case (!pages) we only update counts. | |
1342 | */ | |
1343 | if (likely(pages)) | |
1344 | pages += ret; | |
f0818f47 | 1345 | start += ret << PAGE_SHIFT; |
4426e945 | 1346 | lock_dropped = true; |
f0818f47 | 1347 | |
4426e945 | 1348 | retry: |
f0818f47 AA |
1349 | /* |
1350 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1351 | * with both FAULT_FLAG_ALLOW_RETRY and |
1352 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
1353 | * by fatal signals, so we need to check it before we | |
1354 | * start trying again otherwise it can loop forever. | |
f0818f47 | 1355 | */ |
4426e945 | 1356 | |
ae46d2aa HD |
1357 | if (fatal_signal_pending(current)) { |
1358 | if (!pages_done) | |
1359 | pages_done = -EINTR; | |
4426e945 | 1360 | break; |
ae46d2aa | 1361 | } |
4426e945 | 1362 | |
d8ed45c5 | 1363 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1364 | if (ret) { |
1365 | BUG_ON(ret > 0); | |
1366 | if (!pages_done) | |
1367 | pages_done = ret; | |
1368 | break; | |
1369 | } | |
4426e945 | 1370 | |
c7b6a566 | 1371 | *locked = 1; |
64019a2e | 1372 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
4426e945 PX |
1373 | pages, NULL, locked); |
1374 | if (!*locked) { | |
1375 | /* Continue to retry until we succeeded */ | |
1376 | BUG_ON(ret != 0); | |
1377 | goto retry; | |
1378 | } | |
f0818f47 AA |
1379 | if (ret != 1) { |
1380 | BUG_ON(ret > 1); | |
1381 | if (!pages_done) | |
1382 | pages_done = ret; | |
1383 | break; | |
1384 | } | |
1385 | nr_pages--; | |
1386 | pages_done++; | |
1387 | if (!nr_pages) | |
1388 | break; | |
df17277b MR |
1389 | if (likely(pages)) |
1390 | pages++; | |
f0818f47 AA |
1391 | start += PAGE_SIZE; |
1392 | } | |
e716712f | 1393 | if (lock_dropped && *locked) { |
f0818f47 AA |
1394 | /* |
1395 | * We must let the caller know we temporarily dropped the lock | |
1396 | * and so the critical section protected by it was lost. | |
1397 | */ | |
d8ed45c5 | 1398 | mmap_read_unlock(mm); |
f0818f47 AA |
1399 | *locked = 0; |
1400 | } | |
1401 | return pages_done; | |
1402 | } | |
1403 | ||
d3649f68 CH |
1404 | /** |
1405 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1406 | * @vma: target vma | |
1407 | * @start: start address | |
1408 | * @end: end address | |
c1e8d7c6 | 1409 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1410 | * |
1411 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1412 | * | |
0a36f7f8 TY |
1413 | * Return either number of pages pinned in the vma, or a negative error |
1414 | * code on error. | |
d3649f68 | 1415 | * |
c1e8d7c6 | 1416 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1417 | * |
4f6da934 | 1418 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1419 | * be unperturbed. |
1420 | * | |
4f6da934 PX |
1421 | * If @locked is non-NULL, it must held for read only and may be |
1422 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1423 | */ |
1424 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1425 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1426 | { |
1427 | struct mm_struct *mm = vma->vm_mm; | |
1428 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1429 | int gup_flags; | |
1430 | ||
1431 | VM_BUG_ON(start & ~PAGE_MASK); | |
1432 | VM_BUG_ON(end & ~PAGE_MASK); | |
1433 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1434 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1435 | mmap_assert_locked(mm); |
d3649f68 CH |
1436 | |
1437 | gup_flags = FOLL_TOUCH | FOLL_POPULATE | FOLL_MLOCK; | |
1438 | if (vma->vm_flags & VM_LOCKONFAULT) | |
1439 | gup_flags &= ~FOLL_POPULATE; | |
1440 | /* | |
1441 | * We want to touch writable mappings with a write fault in order | |
1442 | * to break COW, except for shared mappings because these don't COW | |
1443 | * and we would not want to dirty them for nothing. | |
1444 | */ | |
1445 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1446 | gup_flags |= FOLL_WRITE; | |
1447 | ||
1448 | /* | |
1449 | * We want mlock to succeed for regions that have any permissions | |
1450 | * other than PROT_NONE. | |
1451 | */ | |
3122e80e | 1452 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1453 | gup_flags |= FOLL_FORCE; |
1454 | ||
1455 | /* | |
1456 | * We made sure addr is within a VMA, so the following will | |
1457 | * not result in a stack expansion that recurses back here. | |
1458 | */ | |
64019a2e | 1459 | return __get_user_pages(mm, start, nr_pages, gup_flags, |
4f6da934 | 1460 | NULL, NULL, locked); |
d3649f68 CH |
1461 | } |
1462 | ||
1463 | /* | |
1464 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1465 | * | |
1466 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1467 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1468 | * mmap_lock must not be held. |
d3649f68 CH |
1469 | */ |
1470 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1471 | { | |
1472 | struct mm_struct *mm = current->mm; | |
1473 | unsigned long end, nstart, nend; | |
1474 | struct vm_area_struct *vma = NULL; | |
1475 | int locked = 0; | |
1476 | long ret = 0; | |
1477 | ||
1478 | end = start + len; | |
1479 | ||
1480 | for (nstart = start; nstart < end; nstart = nend) { | |
1481 | /* | |
1482 | * We want to fault in pages for [nstart; end) address range. | |
1483 | * Find first corresponding VMA. | |
1484 | */ | |
1485 | if (!locked) { | |
1486 | locked = 1; | |
d8ed45c5 | 1487 | mmap_read_lock(mm); |
d3649f68 CH |
1488 | vma = find_vma(mm, nstart); |
1489 | } else if (nstart >= vma->vm_end) | |
1490 | vma = vma->vm_next; | |
1491 | if (!vma || vma->vm_start >= end) | |
1492 | break; | |
1493 | /* | |
1494 | * Set [nstart; nend) to intersection of desired address | |
1495 | * range with the first VMA. Also, skip undesirable VMA types. | |
1496 | */ | |
1497 | nend = min(end, vma->vm_end); | |
1498 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1499 | continue; | |
1500 | if (nstart < vma->vm_start) | |
1501 | nstart = vma->vm_start; | |
1502 | /* | |
1503 | * Now fault in a range of pages. populate_vma_page_range() | |
1504 | * double checks the vma flags, so that it won't mlock pages | |
1505 | * if the vma was already munlocked. | |
1506 | */ | |
1507 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1508 | if (ret < 0) { | |
1509 | if (ignore_errors) { | |
1510 | ret = 0; | |
1511 | continue; /* continue at next VMA */ | |
1512 | } | |
1513 | break; | |
1514 | } | |
1515 | nend = nstart + ret * PAGE_SIZE; | |
1516 | ret = 0; | |
1517 | } | |
1518 | if (locked) | |
d8ed45c5 | 1519 | mmap_read_unlock(mm); |
d3649f68 CH |
1520 | return ret; /* 0 or negative error code */ |
1521 | } | |
050a9adc | 1522 | #else /* CONFIG_MMU */ |
64019a2e | 1523 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc CH |
1524 | unsigned long nr_pages, struct page **pages, |
1525 | struct vm_area_struct **vmas, int *locked, | |
1526 | unsigned int foll_flags) | |
1527 | { | |
1528 | struct vm_area_struct *vma; | |
1529 | unsigned long vm_flags; | |
1530 | int i; | |
1531 | ||
1532 | /* calculate required read or write permissions. | |
1533 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1534 | */ | |
1535 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1536 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1537 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1538 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1539 | ||
1540 | for (i = 0; i < nr_pages; i++) { | |
1541 | vma = find_vma(mm, start); | |
1542 | if (!vma) | |
1543 | goto finish_or_fault; | |
1544 | ||
1545 | /* protect what we can, including chardevs */ | |
1546 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1547 | !(vm_flags & vma->vm_flags)) | |
1548 | goto finish_or_fault; | |
1549 | ||
1550 | if (pages) { | |
1551 | pages[i] = virt_to_page(start); | |
1552 | if (pages[i]) | |
1553 | get_page(pages[i]); | |
1554 | } | |
1555 | if (vmas) | |
1556 | vmas[i] = vma; | |
1557 | start = (start + PAGE_SIZE) & PAGE_MASK; | |
1558 | } | |
1559 | ||
1560 | return i; | |
1561 | ||
1562 | finish_or_fault: | |
1563 | return i ? : -EFAULT; | |
1564 | } | |
1565 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1566 | |
8f942eea JH |
1567 | /** |
1568 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1569 | * @addr: user address | |
1570 | * | |
1571 | * Returns struct page pointer of user page pinned for dump, | |
1572 | * to be freed afterwards by put_page(). | |
1573 | * | |
1574 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1575 | * the corefile, to preserve alignment with its headers; and also returns | |
1576 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
1577 | * allowing a hole to be left in the corefile to save diskspace. | |
1578 | * | |
7f3bfab5 | 1579 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1580 | */ |
1581 | #ifdef CONFIG_ELF_CORE | |
1582 | struct page *get_dump_page(unsigned long addr) | |
1583 | { | |
7f3bfab5 | 1584 | struct mm_struct *mm = current->mm; |
8f942eea | 1585 | struct page *page; |
7f3bfab5 JH |
1586 | int locked = 1; |
1587 | int ret; | |
8f942eea | 1588 | |
7f3bfab5 | 1589 | if (mmap_read_lock_killable(mm)) |
8f942eea | 1590 | return NULL; |
7f3bfab5 JH |
1591 | ret = __get_user_pages_locked(mm, addr, 1, &page, NULL, &locked, |
1592 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); | |
1593 | if (locked) | |
1594 | mmap_read_unlock(mm); | |
d3378e86 AY |
1595 | |
1596 | if (ret == 1 && is_page_poisoned(page)) | |
1597 | return NULL; | |
1598 | ||
7f3bfab5 | 1599 | return (ret == 1) ? page : NULL; |
8f942eea JH |
1600 | } |
1601 | #endif /* CONFIG_ELF_CORE */ | |
1602 | ||
9a4e9f3b | 1603 | #ifdef CONFIG_CMA |
64019a2e | 1604 | static long check_and_migrate_cma_pages(struct mm_struct *mm, |
932f4a63 IW |
1605 | unsigned long start, |
1606 | unsigned long nr_pages, | |
9a4e9f3b | 1607 | struct page **pages, |
932f4a63 IW |
1608 | struct vm_area_struct **vmas, |
1609 | unsigned int gup_flags) | |
9a4e9f3b | 1610 | { |
6e7f34eb PT |
1611 | unsigned long i, isolation_error_count; |
1612 | bool drain_allow; | |
9a4e9f3b | 1613 | LIST_HEAD(cma_page_list); |
b96cc655 | 1614 | long ret = nr_pages; |
83c02c23 | 1615 | struct page *prev_head, *head; |
ed03d924 JK |
1616 | struct migration_target_control mtc = { |
1617 | .nid = NUMA_NO_NODE, | |
c991ffef | 1618 | .gfp_mask = GFP_USER | __GFP_NOWARN, |
ed03d924 | 1619 | }; |
9a4e9f3b AK |
1620 | |
1621 | check_again: | |
83c02c23 | 1622 | prev_head = NULL; |
6e7f34eb PT |
1623 | isolation_error_count = 0; |
1624 | drain_allow = true; | |
83c02c23 PT |
1625 | for (i = 0; i < nr_pages; i++) { |
1626 | head = compound_head(pages[i]); | |
1627 | if (head == prev_head) | |
1628 | continue; | |
1629 | prev_head = head; | |
9a4e9f3b AK |
1630 | /* |
1631 | * If we get a page from the CMA zone, since we are going to | |
1632 | * be pinning these entries, we might as well move them out | |
1633 | * of the CMA zone if possible. | |
1634 | */ | |
aa712399 | 1635 | if (is_migrate_cma_page(head)) { |
6e7f34eb PT |
1636 | if (PageHuge(head)) { |
1637 | if (!isolate_huge_page(head, &cma_page_list)) | |
1638 | isolation_error_count++; | |
1639 | } else { | |
9a4e9f3b AK |
1640 | if (!PageLRU(head) && drain_allow) { |
1641 | lru_add_drain_all(); | |
1642 | drain_allow = false; | |
1643 | } | |
1644 | ||
6e7f34eb PT |
1645 | if (isolate_lru_page(head)) { |
1646 | isolation_error_count++; | |
1647 | continue; | |
9a4e9f3b | 1648 | } |
6e7f34eb PT |
1649 | list_add_tail(&head->lru, &cma_page_list); |
1650 | mod_node_page_state(page_pgdat(head), | |
1651 | NR_ISOLATED_ANON + | |
1652 | page_is_file_lru(head), | |
1653 | thp_nr_pages(head)); | |
9a4e9f3b AK |
1654 | } |
1655 | } | |
1656 | } | |
1657 | ||
6e7f34eb PT |
1658 | /* |
1659 | * If list is empty, and no isolation errors, means that all pages are | |
1660 | * in the correct zone. | |
1661 | */ | |
1662 | if (list_empty(&cma_page_list) && !isolation_error_count) | |
1663 | return ret; | |
1664 | ||
9a4e9f3b AK |
1665 | if (!list_empty(&cma_page_list)) { |
1666 | /* | |
1667 | * drop the above get_user_pages reference. | |
1668 | */ | |
96e1fac1 JG |
1669 | if (gup_flags & FOLL_PIN) |
1670 | unpin_user_pages(pages, nr_pages); | |
1671 | else | |
1672 | for (i = 0; i < nr_pages; i++) | |
1673 | put_page(pages[i]); | |
9a4e9f3b | 1674 | |
f0f44638 PT |
1675 | ret = migrate_pages(&cma_page_list, alloc_migration_target, |
1676 | NULL, (unsigned long)&mtc, MIGRATE_SYNC, | |
1677 | MR_CONTIG_RANGE); | |
1678 | if (ret) { | |
9a4e9f3b AK |
1679 | if (!list_empty(&cma_page_list)) |
1680 | putback_movable_pages(&cma_page_list); | |
f0f44638 | 1681 | return ret > 0 ? -ENOMEM : ret; |
9a4e9f3b | 1682 | } |
f0f44638 | 1683 | |
6e7f34eb PT |
1684 | /* We unpinned pages before migration, pin them again */ |
1685 | ret = __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
1686 | NULL, gup_flags); | |
1687 | if (ret <= 0) | |
1688 | return ret; | |
1689 | nr_pages = ret; | |
9a4e9f3b AK |
1690 | } |
1691 | ||
6e7f34eb PT |
1692 | /* |
1693 | * check again because pages were unpinned, and we also might have | |
1694 | * had isolation errors and need more pages to migrate. | |
1695 | */ | |
1696 | goto check_again; | |
9a4e9f3b AK |
1697 | } |
1698 | #else | |
64019a2e | 1699 | static long check_and_migrate_cma_pages(struct mm_struct *mm, |
932f4a63 IW |
1700 | unsigned long start, |
1701 | unsigned long nr_pages, | |
1702 | struct page **pages, | |
1703 | struct vm_area_struct **vmas, | |
1704 | unsigned int gup_flags) | |
9a4e9f3b AK |
1705 | { |
1706 | return nr_pages; | |
1707 | } | |
050a9adc | 1708 | #endif /* CONFIG_CMA */ |
9a4e9f3b | 1709 | |
2bb6d283 | 1710 | /* |
932f4a63 IW |
1711 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
1712 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 1713 | */ |
64019a2e | 1714 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
1715 | unsigned long start, |
1716 | unsigned long nr_pages, | |
1717 | struct page **pages, | |
1718 | struct vm_area_struct **vmas, | |
1719 | unsigned int gup_flags) | |
2bb6d283 | 1720 | { |
932f4a63 | 1721 | unsigned long flags = 0; |
52650c8b | 1722 | long rc; |
2bb6d283 | 1723 | |
52650c8b | 1724 | if (gup_flags & FOLL_LONGTERM) |
1a08ae36 | 1725 | flags = memalloc_pin_save(); |
2bb6d283 | 1726 | |
52650c8b JG |
1727 | rc = __get_user_pages_locked(mm, start, nr_pages, pages, vmas, NULL, |
1728 | gup_flags); | |
2bb6d283 | 1729 | |
932f4a63 | 1730 | if (gup_flags & FOLL_LONGTERM) { |
52650c8b JG |
1731 | if (rc > 0) |
1732 | rc = check_and_migrate_cma_pages(mm, start, rc, pages, | |
1733 | vmas, gup_flags); | |
1a08ae36 | 1734 | memalloc_pin_restore(flags); |
9a4e9f3b | 1735 | } |
2bb6d283 DW |
1736 | return rc; |
1737 | } | |
932f4a63 | 1738 | |
447f3e45 BS |
1739 | static bool is_valid_gup_flags(unsigned int gup_flags) |
1740 | { | |
1741 | /* | |
1742 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
1743 | * never directly by the caller, so enforce that with an assertion: | |
1744 | */ | |
1745 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
1746 | return false; | |
1747 | /* | |
1748 | * FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying | |
1749 | * that is, FOLL_LONGTERM is a specific case, more restrictive case of | |
1750 | * FOLL_PIN. | |
1751 | */ | |
1752 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
1753 | return false; | |
1754 | ||
1755 | return true; | |
1756 | } | |
1757 | ||
22bf29b6 | 1758 | #ifdef CONFIG_MMU |
64019a2e | 1759 | static long __get_user_pages_remote(struct mm_struct *mm, |
22bf29b6 JH |
1760 | unsigned long start, unsigned long nr_pages, |
1761 | unsigned int gup_flags, struct page **pages, | |
1762 | struct vm_area_struct **vmas, int *locked) | |
1763 | { | |
1764 | /* | |
1765 | * Parts of FOLL_LONGTERM behavior are incompatible with | |
1766 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1767 | * vmas. However, this only comes up if locked is set, and there are | |
1768 | * callers that do request FOLL_LONGTERM, but do not set locked. So, | |
1769 | * allow what we can. | |
1770 | */ | |
1771 | if (gup_flags & FOLL_LONGTERM) { | |
1772 | if (WARN_ON_ONCE(locked)) | |
1773 | return -EINVAL; | |
1774 | /* | |
1775 | * This will check the vmas (even if our vmas arg is NULL) | |
1776 | * and return -ENOTSUPP if DAX isn't allowed in this case: | |
1777 | */ | |
64019a2e | 1778 | return __gup_longterm_locked(mm, start, nr_pages, pages, |
22bf29b6 JH |
1779 | vmas, gup_flags | FOLL_TOUCH | |
1780 | FOLL_REMOTE); | |
1781 | } | |
1782 | ||
64019a2e | 1783 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, |
22bf29b6 JH |
1784 | locked, |
1785 | gup_flags | FOLL_TOUCH | FOLL_REMOTE); | |
1786 | } | |
1787 | ||
adc8cb40 | 1788 | /** |
c4237f8b | 1789 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
1790 | * @mm: mm_struct of target mm |
1791 | * @start: starting user address | |
1792 | * @nr_pages: number of pages from start to pin | |
1793 | * @gup_flags: flags modifying lookup behaviour | |
1794 | * @pages: array that receives pointers to the pages pinned. | |
1795 | * Should be at least nr_pages long. Or NULL, if caller | |
1796 | * only intends to ensure the pages are faulted in. | |
1797 | * @vmas: array of pointers to vmas corresponding to each page. | |
1798 | * Or NULL if the caller does not require them. | |
1799 | * @locked: pointer to lock flag indicating whether lock is held and | |
1800 | * subsequently whether VM_FAULT_RETRY functionality can be | |
1801 | * utilised. Lock must initially be held. | |
1802 | * | |
1803 | * Returns either number of pages pinned (which may be less than the | |
1804 | * number requested), or an error. Details about the return value: | |
1805 | * | |
1806 | * -- If nr_pages is 0, returns 0. | |
1807 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1808 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1809 | * pages pinned. Again, this may be less than nr_pages. | |
1810 | * | |
1811 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1812 | * | |
c1e8d7c6 | 1813 | * @vmas are valid only as long as mmap_lock is held. |
c4237f8b | 1814 | * |
c1e8d7c6 | 1815 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 1816 | * |
adc8cb40 SJ |
1817 | * get_user_pages_remote walks a process's page tables and takes a reference |
1818 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
1819 | * instant. That is, it takes the page that would be accessed if a user |
1820 | * thread accesses the given user virtual address at that instant. | |
1821 | * | |
1822 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 1823 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b JH |
1824 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
1825 | * and subsequently re faulted). However it does guarantee that the page | |
1826 | * won't be freed completely. And mostly callers simply care that the page | |
1827 | * contains data that was valid *at some point in time*. Typically, an IO | |
1828 | * or similar operation cannot guarantee anything stronger anyway because | |
1829 | * locks can't be held over the syscall boundary. | |
1830 | * | |
1831 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
1832 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
1833 | * be called after the page is finished with, and before put_page is called. | |
1834 | * | |
adc8cb40 SJ |
1835 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
1836 | * to get a handle on the memory by some means other than accesses | |
1837 | * via the user virtual addresses. The pages may be submitted for | |
1838 | * DMA to devices or accessed via their kernel linear mapping (via the | |
1839 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
1840 | * |
1841 | * See also get_user_pages_fast, for performance critical applications. | |
1842 | * | |
adc8cb40 | 1843 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 1844 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 1845 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
1846 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
1847 | */ | |
64019a2e | 1848 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
1849 | unsigned long start, unsigned long nr_pages, |
1850 | unsigned int gup_flags, struct page **pages, | |
1851 | struct vm_area_struct **vmas, int *locked) | |
1852 | { | |
447f3e45 | 1853 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
1854 | return -EINVAL; |
1855 | ||
64019a2e | 1856 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
22bf29b6 | 1857 | pages, vmas, locked); |
c4237f8b JH |
1858 | } |
1859 | EXPORT_SYMBOL(get_user_pages_remote); | |
1860 | ||
eddb1c22 | 1861 | #else /* CONFIG_MMU */ |
64019a2e | 1862 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
1863 | unsigned long start, unsigned long nr_pages, |
1864 | unsigned int gup_flags, struct page **pages, | |
1865 | struct vm_area_struct **vmas, int *locked) | |
1866 | { | |
1867 | return 0; | |
1868 | } | |
3faa52c0 | 1869 | |
64019a2e | 1870 | static long __get_user_pages_remote(struct mm_struct *mm, |
3faa52c0 JH |
1871 | unsigned long start, unsigned long nr_pages, |
1872 | unsigned int gup_flags, struct page **pages, | |
1873 | struct vm_area_struct **vmas, int *locked) | |
1874 | { | |
1875 | return 0; | |
1876 | } | |
eddb1c22 JH |
1877 | #endif /* !CONFIG_MMU */ |
1878 | ||
adc8cb40 SJ |
1879 | /** |
1880 | * get_user_pages() - pin user pages in memory | |
1881 | * @start: starting user address | |
1882 | * @nr_pages: number of pages from start to pin | |
1883 | * @gup_flags: flags modifying lookup behaviour | |
1884 | * @pages: array that receives pointers to the pages pinned. | |
1885 | * Should be at least nr_pages long. Or NULL, if caller | |
1886 | * only intends to ensure the pages are faulted in. | |
1887 | * @vmas: array of pointers to vmas corresponding to each page. | |
1888 | * Or NULL if the caller does not require them. | |
1889 | * | |
64019a2e PX |
1890 | * This is the same as get_user_pages_remote(), just with a less-flexible |
1891 | * calling convention where we assume that the mm being operated on belongs to | |
1892 | * the current task, and doesn't allow passing of a locked parameter. We also | |
1893 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
1894 | */ |
1895 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
1896 | unsigned int gup_flags, struct page **pages, | |
1897 | struct vm_area_struct **vmas) | |
1898 | { | |
447f3e45 | 1899 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
1900 | return -EINVAL; |
1901 | ||
64019a2e | 1902 | return __gup_longterm_locked(current->mm, start, nr_pages, |
932f4a63 IW |
1903 | pages, vmas, gup_flags | FOLL_TOUCH); |
1904 | } | |
1905 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 1906 | |
adc8cb40 | 1907 | /** |
a00cda3f MCC |
1908 | * get_user_pages_locked() - variant of get_user_pages() |
1909 | * | |
1910 | * @start: starting user address | |
1911 | * @nr_pages: number of pages from start to pin | |
1912 | * @gup_flags: flags modifying lookup behaviour | |
1913 | * @pages: array that receives pointers to the pages pinned. | |
1914 | * Should be at least nr_pages long. Or NULL, if caller | |
1915 | * only intends to ensure the pages are faulted in. | |
1916 | * @locked: pointer to lock flag indicating whether lock is held and | |
1917 | * subsequently whether VM_FAULT_RETRY functionality can be | |
1918 | * utilised. Lock must initially be held. | |
1919 | * | |
1920 | * It is suitable to replace the form: | |
acc3c8d1 | 1921 | * |
3e4e28c5 | 1922 | * mmap_read_lock(mm); |
d3649f68 | 1923 | * do_something() |
64019a2e | 1924 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 1925 | * mmap_read_unlock(mm); |
acc3c8d1 | 1926 | * |
d3649f68 | 1927 | * to: |
acc3c8d1 | 1928 | * |
d3649f68 | 1929 | * int locked = 1; |
3e4e28c5 | 1930 | * mmap_read_lock(mm); |
d3649f68 | 1931 | * do_something() |
64019a2e | 1932 | * get_user_pages_locked(mm, ..., pages, &locked); |
d3649f68 | 1933 | * if (locked) |
3e4e28c5 | 1934 | * mmap_read_unlock(mm); |
adc8cb40 | 1935 | * |
adc8cb40 SJ |
1936 | * We can leverage the VM_FAULT_RETRY functionality in the page fault |
1937 | * paths better by using either get_user_pages_locked() or | |
1938 | * get_user_pages_unlocked(). | |
1939 | * | |
acc3c8d1 | 1940 | */ |
d3649f68 CH |
1941 | long get_user_pages_locked(unsigned long start, unsigned long nr_pages, |
1942 | unsigned int gup_flags, struct page **pages, | |
1943 | int *locked) | |
acc3c8d1 | 1944 | { |
acc3c8d1 | 1945 | /* |
d3649f68 CH |
1946 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with |
1947 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1948 | * vmas. As there are no users of this flag in this call we simply | |
1949 | * disallow this option for now. | |
acc3c8d1 | 1950 | */ |
d3649f68 CH |
1951 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) |
1952 | return -EINVAL; | |
420c2091 JH |
1953 | /* |
1954 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
1955 | * never directly by the caller, so enforce that: | |
1956 | */ | |
1957 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
1958 | return -EINVAL; | |
acc3c8d1 | 1959 | |
64019a2e | 1960 | return __get_user_pages_locked(current->mm, start, nr_pages, |
d3649f68 CH |
1961 | pages, NULL, locked, |
1962 | gup_flags | FOLL_TOUCH); | |
acc3c8d1 | 1963 | } |
d3649f68 | 1964 | EXPORT_SYMBOL(get_user_pages_locked); |
acc3c8d1 KS |
1965 | |
1966 | /* | |
d3649f68 | 1967 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 1968 | * |
3e4e28c5 | 1969 | * mmap_read_lock(mm); |
64019a2e | 1970 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 1971 | * mmap_read_unlock(mm); |
d3649f68 CH |
1972 | * |
1973 | * with: | |
1974 | * | |
64019a2e | 1975 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
1976 | * |
1977 | * It is functionally equivalent to get_user_pages_fast so | |
1978 | * get_user_pages_fast should be used instead if specific gup_flags | |
1979 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 1980 | */ |
d3649f68 CH |
1981 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
1982 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 KS |
1983 | { |
1984 | struct mm_struct *mm = current->mm; | |
d3649f68 CH |
1985 | int locked = 1; |
1986 | long ret; | |
acc3c8d1 | 1987 | |
d3649f68 CH |
1988 | /* |
1989 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
1990 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1991 | * vmas. As there are no users of this flag in this call we simply | |
1992 | * disallow this option for now. | |
1993 | */ | |
1994 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
1995 | return -EINVAL; | |
acc3c8d1 | 1996 | |
d8ed45c5 | 1997 | mmap_read_lock(mm); |
64019a2e | 1998 | ret = __get_user_pages_locked(mm, start, nr_pages, pages, NULL, |
d3649f68 | 1999 | &locked, gup_flags | FOLL_TOUCH); |
acc3c8d1 | 2000 | if (locked) |
d8ed45c5 | 2001 | mmap_read_unlock(mm); |
d3649f68 | 2002 | return ret; |
4bbd4c77 | 2003 | } |
d3649f68 | 2004 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
2005 | |
2006 | /* | |
67a929e0 | 2007 | * Fast GUP |
2667f50e SC |
2008 | * |
2009 | * get_user_pages_fast attempts to pin user pages by walking the page | |
2010 | * tables directly and avoids taking locks. Thus the walker needs to be | |
2011 | * protected from page table pages being freed from under it, and should | |
2012 | * block any THP splits. | |
2013 | * | |
2014 | * One way to achieve this is to have the walker disable interrupts, and | |
2015 | * rely on IPIs from the TLB flushing code blocking before the page table | |
2016 | * pages are freed. This is unsuitable for architectures that do not need | |
2017 | * to broadcast an IPI when invalidating TLBs. | |
2018 | * | |
2019 | * Another way to achieve this is to batch up page table containing pages | |
2020 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
2021 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
2022 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
2023 | * (which is a relatively rare event). The code below adopts this strategy. | |
2024 | * | |
2025 | * Before activating this code, please be aware that the following assumptions | |
2026 | * are currently made: | |
2027 | * | |
ff2e6d72 | 2028 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 2029 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 2030 | * |
2667f50e SC |
2031 | * *) ptes can be read atomically by the architecture. |
2032 | * | |
2033 | * *) access_ok is sufficient to validate userspace address ranges. | |
2034 | * | |
2035 | * The last two assumptions can be relaxed by the addition of helper functions. | |
2036 | * | |
2037 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
2038 | */ | |
67a929e0 | 2039 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 2040 | |
790c7369 | 2041 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 2042 | unsigned int flags, |
790c7369 | 2043 | struct page **pages) |
b59f65fa KS |
2044 | { |
2045 | while ((*nr) - nr_start) { | |
2046 | struct page *page = pages[--(*nr)]; | |
2047 | ||
2048 | ClearPageReferenced(page); | |
3faa52c0 JH |
2049 | if (flags & FOLL_PIN) |
2050 | unpin_user_page(page); | |
2051 | else | |
2052 | put_page(page); | |
b59f65fa KS |
2053 | } |
2054 | } | |
2055 | ||
3010a5ea | 2056 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
2667f50e | 2057 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, |
b798bec4 | 2058 | unsigned int flags, struct page **pages, int *nr) |
2667f50e | 2059 | { |
b59f65fa KS |
2060 | struct dev_pagemap *pgmap = NULL; |
2061 | int nr_start = *nr, ret = 0; | |
2667f50e | 2062 | pte_t *ptep, *ptem; |
2667f50e SC |
2063 | |
2064 | ptem = ptep = pte_offset_map(&pmd, addr); | |
2065 | do { | |
2a4a06da | 2066 | pte_t pte = ptep_get_lockless(ptep); |
7aef4172 | 2067 | struct page *head, *page; |
2667f50e SC |
2068 | |
2069 | /* | |
2070 | * Similar to the PMD case below, NUMA hinting must take slow | |
8a0516ed | 2071 | * path using the pte_protnone check. |
2667f50e | 2072 | */ |
e7884f8e KS |
2073 | if (pte_protnone(pte)) |
2074 | goto pte_unmap; | |
2075 | ||
b798bec4 | 2076 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2077 | goto pte_unmap; |
2078 | ||
b59f65fa | 2079 | if (pte_devmap(pte)) { |
7af75561 IW |
2080 | if (unlikely(flags & FOLL_LONGTERM)) |
2081 | goto pte_unmap; | |
2082 | ||
b59f65fa KS |
2083 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2084 | if (unlikely(!pgmap)) { | |
3b78d834 | 2085 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2086 | goto pte_unmap; |
2087 | } | |
2088 | } else if (pte_special(pte)) | |
2667f50e SC |
2089 | goto pte_unmap; |
2090 | ||
2091 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2092 | page = pte_page(pte); | |
2093 | ||
3faa52c0 | 2094 | head = try_grab_compound_head(page, 1, flags); |
8fde12ca | 2095 | if (!head) |
2667f50e SC |
2096 | goto pte_unmap; |
2097 | ||
2098 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
3faa52c0 | 2099 | put_compound_head(head, 1, flags); |
2667f50e SC |
2100 | goto pte_unmap; |
2101 | } | |
2102 | ||
7aef4172 | 2103 | VM_BUG_ON_PAGE(compound_head(page) != head, page); |
e9348053 | 2104 | |
f28d4363 CI |
2105 | /* |
2106 | * We need to make the page accessible if and only if we are | |
2107 | * going to access its content (the FOLL_PIN case). Please | |
2108 | * see Documentation/core-api/pin_user_pages.rst for | |
2109 | * details. | |
2110 | */ | |
2111 | if (flags & FOLL_PIN) { | |
2112 | ret = arch_make_page_accessible(page); | |
2113 | if (ret) { | |
2114 | unpin_user_page(page); | |
2115 | goto pte_unmap; | |
2116 | } | |
2117 | } | |
e9348053 | 2118 | SetPageReferenced(page); |
2667f50e SC |
2119 | pages[*nr] = page; |
2120 | (*nr)++; | |
2121 | ||
2122 | } while (ptep++, addr += PAGE_SIZE, addr != end); | |
2123 | ||
2124 | ret = 1; | |
2125 | ||
2126 | pte_unmap: | |
832d7aa0 CH |
2127 | if (pgmap) |
2128 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2129 | pte_unmap(ptem); |
2130 | return ret; | |
2131 | } | |
2132 | #else | |
2133 | ||
2134 | /* | |
2135 | * If we can't determine whether or not a pte is special, then fail immediately | |
2136 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2137 | * to be special. | |
2138 | * | |
2139 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2140 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2141 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2142 | */ | |
2143 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, | |
b798bec4 | 2144 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2145 | { |
2146 | return 0; | |
2147 | } | |
3010a5ea | 2148 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2149 | |
17596731 | 2150 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2151 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2152 | unsigned long end, unsigned int flags, |
2153 | struct page **pages, int *nr) | |
b59f65fa KS |
2154 | { |
2155 | int nr_start = *nr; | |
2156 | struct dev_pagemap *pgmap = NULL; | |
2157 | ||
2158 | do { | |
2159 | struct page *page = pfn_to_page(pfn); | |
2160 | ||
2161 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2162 | if (unlikely(!pgmap)) { | |
3b78d834 | 2163 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2164 | return 0; |
2165 | } | |
2166 | SetPageReferenced(page); | |
2167 | pages[*nr] = page; | |
3faa52c0 JH |
2168 | if (unlikely(!try_grab_page(page, flags))) { |
2169 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
2170 | return 0; | |
2171 | } | |
b59f65fa KS |
2172 | (*nr)++; |
2173 | pfn++; | |
2174 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 CH |
2175 | |
2176 | if (pgmap) | |
2177 | put_dev_pagemap(pgmap); | |
b59f65fa KS |
2178 | return 1; |
2179 | } | |
2180 | ||
a9b6de77 | 2181 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2182 | unsigned long end, unsigned int flags, |
2183 | struct page **pages, int *nr) | |
b59f65fa KS |
2184 | { |
2185 | unsigned long fault_pfn; | |
a9b6de77 DW |
2186 | int nr_start = *nr; |
2187 | ||
2188 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2189 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2190 | return 0; |
b59f65fa | 2191 | |
a9b6de77 | 2192 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2193 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2194 | return 0; |
2195 | } | |
2196 | return 1; | |
b59f65fa KS |
2197 | } |
2198 | ||
a9b6de77 | 2199 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2200 | unsigned long end, unsigned int flags, |
2201 | struct page **pages, int *nr) | |
b59f65fa KS |
2202 | { |
2203 | unsigned long fault_pfn; | |
a9b6de77 DW |
2204 | int nr_start = *nr; |
2205 | ||
2206 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2207 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2208 | return 0; |
b59f65fa | 2209 | |
a9b6de77 | 2210 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2211 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2212 | return 0; |
2213 | } | |
2214 | return 1; | |
b59f65fa KS |
2215 | } |
2216 | #else | |
a9b6de77 | 2217 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2218 | unsigned long end, unsigned int flags, |
2219 | struct page **pages, int *nr) | |
b59f65fa KS |
2220 | { |
2221 | BUILD_BUG(); | |
2222 | return 0; | |
2223 | } | |
2224 | ||
a9b6de77 | 2225 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2226 | unsigned long end, unsigned int flags, |
2227 | struct page **pages, int *nr) | |
b59f65fa KS |
2228 | { |
2229 | BUILD_BUG(); | |
2230 | return 0; | |
2231 | } | |
2232 | #endif | |
2233 | ||
a43e9820 JH |
2234 | static int record_subpages(struct page *page, unsigned long addr, |
2235 | unsigned long end, struct page **pages) | |
2236 | { | |
2237 | int nr; | |
2238 | ||
2239 | for (nr = 0; addr != end; addr += PAGE_SIZE) | |
2240 | pages[nr++] = page++; | |
2241 | ||
2242 | return nr; | |
2243 | } | |
2244 | ||
cbd34da7 CH |
2245 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2246 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2247 | unsigned long sz) | |
2248 | { | |
2249 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2250 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2251 | } | |
2252 | ||
2253 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2254 | unsigned long end, unsigned int flags, |
2255 | struct page **pages, int *nr) | |
cbd34da7 CH |
2256 | { |
2257 | unsigned long pte_end; | |
2258 | struct page *head, *page; | |
2259 | pte_t pte; | |
2260 | int refs; | |
2261 | ||
2262 | pte_end = (addr + sz) & ~(sz-1); | |
2263 | if (pte_end < end) | |
2264 | end = pte_end; | |
2265 | ||
55ca2263 | 2266 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2267 | |
0cd22afd | 2268 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2269 | return 0; |
2270 | ||
2271 | /* hugepages are never "special" */ | |
2272 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2273 | ||
cbd34da7 | 2274 | head = pte_page(pte); |
cbd34da7 | 2275 | page = head + ((addr & (sz-1)) >> PAGE_SHIFT); |
a43e9820 | 2276 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2277 | |
3faa52c0 | 2278 | head = try_grab_compound_head(head, refs, flags); |
a43e9820 | 2279 | if (!head) |
cbd34da7 | 2280 | return 0; |
cbd34da7 CH |
2281 | |
2282 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
3b78d834 | 2283 | put_compound_head(head, refs, flags); |
cbd34da7 CH |
2284 | return 0; |
2285 | } | |
2286 | ||
a43e9820 | 2287 | *nr += refs; |
520b4a44 | 2288 | SetPageReferenced(head); |
cbd34da7 CH |
2289 | return 1; |
2290 | } | |
2291 | ||
2292 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2293 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2294 | struct page **pages, int *nr) |
2295 | { | |
2296 | pte_t *ptep; | |
2297 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2298 | unsigned long next; | |
2299 | ||
2300 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2301 | do { | |
2302 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2303 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2304 | return 0; |
2305 | } while (ptep++, addr = next, addr != end); | |
2306 | ||
2307 | return 1; | |
2308 | } | |
2309 | #else | |
2310 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2311 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2312 | struct page **pages, int *nr) |
2313 | { | |
2314 | return 0; | |
2315 | } | |
2316 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2317 | ||
2667f50e | 2318 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2319 | unsigned long end, unsigned int flags, |
2320 | struct page **pages, int *nr) | |
2667f50e | 2321 | { |
ddc58f27 | 2322 | struct page *head, *page; |
2667f50e SC |
2323 | int refs; |
2324 | ||
b798bec4 | 2325 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2326 | return 0; |
2327 | ||
7af75561 IW |
2328 | if (pmd_devmap(orig)) { |
2329 | if (unlikely(flags & FOLL_LONGTERM)) | |
2330 | return 0; | |
86dfbed4 JH |
2331 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2332 | pages, nr); | |
7af75561 | 2333 | } |
b59f65fa | 2334 | |
d63206ee | 2335 | page = pmd_page(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2336 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2337 | |
3faa52c0 | 2338 | head = try_grab_compound_head(pmd_page(orig), refs, flags); |
a43e9820 | 2339 | if (!head) |
2667f50e | 2340 | return 0; |
2667f50e SC |
2341 | |
2342 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
3b78d834 | 2343 | put_compound_head(head, refs, flags); |
2667f50e SC |
2344 | return 0; |
2345 | } | |
2346 | ||
a43e9820 | 2347 | *nr += refs; |
e9348053 | 2348 | SetPageReferenced(head); |
2667f50e SC |
2349 | return 1; |
2350 | } | |
2351 | ||
2352 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2353 | unsigned long end, unsigned int flags, |
2354 | struct page **pages, int *nr) | |
2667f50e | 2355 | { |
ddc58f27 | 2356 | struct page *head, *page; |
2667f50e SC |
2357 | int refs; |
2358 | ||
b798bec4 | 2359 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2360 | return 0; |
2361 | ||
7af75561 IW |
2362 | if (pud_devmap(orig)) { |
2363 | if (unlikely(flags & FOLL_LONGTERM)) | |
2364 | return 0; | |
86dfbed4 JH |
2365 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2366 | pages, nr); | |
7af75561 | 2367 | } |
b59f65fa | 2368 | |
d63206ee | 2369 | page = pud_page(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2370 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2371 | |
3faa52c0 | 2372 | head = try_grab_compound_head(pud_page(orig), refs, flags); |
a43e9820 | 2373 | if (!head) |
2667f50e | 2374 | return 0; |
2667f50e SC |
2375 | |
2376 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
3b78d834 | 2377 | put_compound_head(head, refs, flags); |
2667f50e SC |
2378 | return 0; |
2379 | } | |
2380 | ||
a43e9820 | 2381 | *nr += refs; |
e9348053 | 2382 | SetPageReferenced(head); |
2667f50e SC |
2383 | return 1; |
2384 | } | |
2385 | ||
f30c59e9 | 2386 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2387 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2388 | struct page **pages, int *nr) |
2389 | { | |
2390 | int refs; | |
ddc58f27 | 2391 | struct page *head, *page; |
f30c59e9 | 2392 | |
b798bec4 | 2393 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2394 | return 0; |
2395 | ||
b59f65fa | 2396 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2397 | |
d63206ee | 2398 | page = pgd_page(orig) + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2399 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2400 | |
3faa52c0 | 2401 | head = try_grab_compound_head(pgd_page(orig), refs, flags); |
a43e9820 | 2402 | if (!head) |
f30c59e9 | 2403 | return 0; |
f30c59e9 AK |
2404 | |
2405 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
3b78d834 | 2406 | put_compound_head(head, refs, flags); |
f30c59e9 AK |
2407 | return 0; |
2408 | } | |
2409 | ||
a43e9820 | 2410 | *nr += refs; |
e9348053 | 2411 | SetPageReferenced(head); |
f30c59e9 AK |
2412 | return 1; |
2413 | } | |
2414 | ||
d3f7b1bb | 2415 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2416 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2417 | { |
2418 | unsigned long next; | |
2419 | pmd_t *pmdp; | |
2420 | ||
d3f7b1bb | 2421 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2422 | do { |
38c5ce93 | 2423 | pmd_t pmd = READ_ONCE(*pmdp); |
2667f50e SC |
2424 | |
2425 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2426 | if (!pmd_present(pmd)) |
2667f50e SC |
2427 | return 0; |
2428 | ||
414fd080 YZ |
2429 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2430 | pmd_devmap(pmd))) { | |
2667f50e SC |
2431 | /* |
2432 | * NUMA hinting faults need to be handled in the GUP | |
2433 | * slowpath for accounting purposes and so that they | |
2434 | * can be serialised against THP migration. | |
2435 | */ | |
8a0516ed | 2436 | if (pmd_protnone(pmd)) |
2667f50e SC |
2437 | return 0; |
2438 | ||
b798bec4 | 2439 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2440 | pages, nr)) |
2441 | return 0; | |
2442 | ||
f30c59e9 AK |
2443 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2444 | /* | |
2445 | * architecture have different format for hugetlbfs | |
2446 | * pmd format and THP pmd format | |
2447 | */ | |
2448 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2449 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2450 | return 0; |
b798bec4 | 2451 | } else if (!gup_pte_range(pmd, addr, next, flags, pages, nr)) |
2923117b | 2452 | return 0; |
2667f50e SC |
2453 | } while (pmdp++, addr = next, addr != end); |
2454 | ||
2455 | return 1; | |
2456 | } | |
2457 | ||
d3f7b1bb | 2458 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2459 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2460 | { |
2461 | unsigned long next; | |
2462 | pud_t *pudp; | |
2463 | ||
d3f7b1bb | 2464 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2465 | do { |
e37c6982 | 2466 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2467 | |
2468 | next = pud_addr_end(addr, end); | |
15494520 | 2469 | if (unlikely(!pud_present(pud))) |
2667f50e | 2470 | return 0; |
f30c59e9 | 2471 | if (unlikely(pud_huge(pud))) { |
b798bec4 | 2472 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2473 | pages, nr)) |
2474 | return 0; | |
2475 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2476 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2477 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2478 | return 0; |
d3f7b1bb | 2479 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2480 | return 0; |
2481 | } while (pudp++, addr = next, addr != end); | |
2482 | ||
2483 | return 1; | |
2484 | } | |
2485 | ||
d3f7b1bb | 2486 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2487 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2488 | { |
2489 | unsigned long next; | |
2490 | p4d_t *p4dp; | |
2491 | ||
d3f7b1bb | 2492 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
2493 | do { |
2494 | p4d_t p4d = READ_ONCE(*p4dp); | |
2495 | ||
2496 | next = p4d_addr_end(addr, end); | |
2497 | if (p4d_none(p4d)) | |
2498 | return 0; | |
2499 | BUILD_BUG_ON(p4d_huge(p4d)); | |
2500 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
2501 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 2502 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 2503 | return 0; |
d3f7b1bb | 2504 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
2505 | return 0; |
2506 | } while (p4dp++, addr = next, addr != end); | |
2507 | ||
2508 | return 1; | |
2509 | } | |
2510 | ||
5b65c467 | 2511 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 2512 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
2513 | { |
2514 | unsigned long next; | |
2515 | pgd_t *pgdp; | |
2516 | ||
2517 | pgdp = pgd_offset(current->mm, addr); | |
2518 | do { | |
2519 | pgd_t pgd = READ_ONCE(*pgdp); | |
2520 | ||
2521 | next = pgd_addr_end(addr, end); | |
2522 | if (pgd_none(pgd)) | |
2523 | return; | |
2524 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 2525 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
2526 | pages, nr)) |
2527 | return; | |
2528 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
2529 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 2530 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 2531 | return; |
d3f7b1bb | 2532 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
2533 | return; |
2534 | } while (pgdp++, addr = next, addr != end); | |
2535 | } | |
050a9adc CH |
2536 | #else |
2537 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
2538 | unsigned int flags, struct page **pages, int *nr) | |
2539 | { | |
2540 | } | |
2541 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
2542 | |
2543 | #ifndef gup_fast_permitted | |
2544 | /* | |
dadbb612 | 2545 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
2546 | * we need to fall back to the slow version: |
2547 | */ | |
26f4c328 | 2548 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 2549 | { |
26f4c328 | 2550 | return true; |
5b65c467 KS |
2551 | } |
2552 | #endif | |
2553 | ||
7af75561 IW |
2554 | static int __gup_longterm_unlocked(unsigned long start, int nr_pages, |
2555 | unsigned int gup_flags, struct page **pages) | |
2556 | { | |
2557 | int ret; | |
2558 | ||
2559 | /* | |
2560 | * FIXME: FOLL_LONGTERM does not work with | |
2561 | * get_user_pages_unlocked() (see comments in that function) | |
2562 | */ | |
2563 | if (gup_flags & FOLL_LONGTERM) { | |
d8ed45c5 | 2564 | mmap_read_lock(current->mm); |
64019a2e | 2565 | ret = __gup_longterm_locked(current->mm, |
7af75561 IW |
2566 | start, nr_pages, |
2567 | pages, NULL, gup_flags); | |
d8ed45c5 | 2568 | mmap_read_unlock(current->mm); |
7af75561 IW |
2569 | } else { |
2570 | ret = get_user_pages_unlocked(start, nr_pages, | |
2571 | pages, gup_flags); | |
2572 | } | |
2573 | ||
2574 | return ret; | |
2575 | } | |
2576 | ||
c28b1fc7 JG |
2577 | static unsigned long lockless_pages_from_mm(unsigned long start, |
2578 | unsigned long end, | |
2579 | unsigned int gup_flags, | |
2580 | struct page **pages) | |
2581 | { | |
2582 | unsigned long flags; | |
2583 | int nr_pinned = 0; | |
57efa1fe | 2584 | unsigned seq; |
c28b1fc7 JG |
2585 | |
2586 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
2587 | !gup_fast_permitted(start, end)) | |
2588 | return 0; | |
2589 | ||
57efa1fe JG |
2590 | if (gup_flags & FOLL_PIN) { |
2591 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
2592 | if (seq & 1) | |
2593 | return 0; | |
2594 | } | |
2595 | ||
c28b1fc7 JG |
2596 | /* |
2597 | * Disable interrupts. The nested form is used, in order to allow full, | |
2598 | * general purpose use of this routine. | |
2599 | * | |
2600 | * With interrupts disabled, we block page table pages from being freed | |
2601 | * from under us. See struct mmu_table_batch comments in | |
2602 | * include/asm-generic/tlb.h for more details. | |
2603 | * | |
2604 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
2605 | * that come from THPs splitting. | |
2606 | */ | |
2607 | local_irq_save(flags); | |
2608 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
2609 | local_irq_restore(flags); | |
57efa1fe JG |
2610 | |
2611 | /* | |
2612 | * When pinning pages for DMA there could be a concurrent write protect | |
2613 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
2614 | */ | |
2615 | if (gup_flags & FOLL_PIN) { | |
2616 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
2617 | unpin_user_pages(pages, nr_pinned); | |
2618 | return 0; | |
2619 | } | |
2620 | } | |
c28b1fc7 JG |
2621 | return nr_pinned; |
2622 | } | |
2623 | ||
2624 | static int internal_get_user_pages_fast(unsigned long start, | |
2625 | unsigned long nr_pages, | |
eddb1c22 JH |
2626 | unsigned int gup_flags, |
2627 | struct page **pages) | |
2667f50e | 2628 | { |
c28b1fc7 JG |
2629 | unsigned long len, end; |
2630 | unsigned long nr_pinned; | |
2631 | int ret; | |
2667f50e | 2632 | |
f4000fdf | 2633 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef JH |
2634 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
2635 | FOLL_FAST_ONLY))) | |
817be129 CH |
2636 | return -EINVAL; |
2637 | ||
008cfe44 PX |
2638 | if (gup_flags & FOLL_PIN) |
2639 | atomic_set(¤t->mm->has_pinned, 1); | |
2640 | ||
f81cd178 | 2641 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 2642 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 2643 | |
f455c854 | 2644 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
2645 | len = nr_pages << PAGE_SHIFT; |
2646 | if (check_add_overflow(start, len, &end)) | |
c61611f7 | 2647 | return 0; |
96d4f267 | 2648 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 2649 | return -EFAULT; |
73e10a61 | 2650 | |
c28b1fc7 JG |
2651 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
2652 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
2653 | return nr_pinned; | |
2667f50e | 2654 | |
c28b1fc7 JG |
2655 | /* Slow path: try to get the remaining pages with get_user_pages */ |
2656 | start += nr_pinned << PAGE_SHIFT; | |
2657 | pages += nr_pinned; | |
2658 | ret = __gup_longterm_unlocked(start, nr_pages - nr_pinned, gup_flags, | |
2659 | pages); | |
2660 | if (ret < 0) { | |
2661 | /* | |
2662 | * The caller has to unpin the pages we already pinned so | |
2663 | * returning -errno is not an option | |
2664 | */ | |
2665 | if (nr_pinned) | |
2666 | return nr_pinned; | |
2667 | return ret; | |
2667f50e | 2668 | } |
c28b1fc7 | 2669 | return ret + nr_pinned; |
2667f50e | 2670 | } |
c28b1fc7 | 2671 | |
dadbb612 SJ |
2672 | /** |
2673 | * get_user_pages_fast_only() - pin user pages in memory | |
2674 | * @start: starting user address | |
2675 | * @nr_pages: number of pages from start to pin | |
2676 | * @gup_flags: flags modifying pin behaviour | |
2677 | * @pages: array that receives pointers to the pages pinned. | |
2678 | * Should be at least nr_pages long. | |
2679 | * | |
9e1f0580 JH |
2680 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
2681 | * the regular GUP. | |
2682 | * Note a difference with get_user_pages_fast: this always returns the | |
2683 | * number of pages pinned, 0 if no pages were pinned. | |
2684 | * | |
2685 | * If the architecture does not support this function, simply return with no | |
2686 | * pages pinned. | |
2687 | * | |
2688 | * Careful, careful! COW breaking can go either way, so a non-write | |
2689 | * access can get ambiguous page results. If you call this function without | |
2690 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
2691 | */ | |
dadbb612 SJ |
2692 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
2693 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 2694 | { |
376a34ef | 2695 | int nr_pinned; |
9e1f0580 JH |
2696 | /* |
2697 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
2698 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
2699 | * |
2700 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2701 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 2702 | */ |
dadbb612 | 2703 | gup_flags |= FOLL_GET | FOLL_FAST_ONLY; |
9e1f0580 | 2704 | |
376a34ef JH |
2705 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
2706 | pages); | |
9e1f0580 JH |
2707 | |
2708 | /* | |
376a34ef JH |
2709 | * As specified in the API description above, this routine is not |
2710 | * allowed to return negative values. However, the common core | |
2711 | * routine internal_get_user_pages_fast() *can* return -errno. | |
2712 | * Therefore, correct for that here: | |
9e1f0580 | 2713 | */ |
376a34ef JH |
2714 | if (nr_pinned < 0) |
2715 | nr_pinned = 0; | |
9e1f0580 JH |
2716 | |
2717 | return nr_pinned; | |
2718 | } | |
dadbb612 | 2719 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 2720 | |
eddb1c22 JH |
2721 | /** |
2722 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
2723 | * @start: starting user address |
2724 | * @nr_pages: number of pages from start to pin | |
2725 | * @gup_flags: flags modifying pin behaviour | |
2726 | * @pages: array that receives pointers to the pages pinned. | |
2727 | * Should be at least nr_pages long. | |
eddb1c22 | 2728 | * |
c1e8d7c6 | 2729 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
2730 | * If not successful, it will fall back to taking the lock and |
2731 | * calling get_user_pages(). | |
2732 | * | |
2733 | * Returns number of pages pinned. This may be fewer than the number requested. | |
2734 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
2735 | * -errno. | |
2736 | */ | |
2737 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
2738 | unsigned int gup_flags, struct page **pages) | |
2739 | { | |
447f3e45 | 2740 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2741 | return -EINVAL; |
2742 | ||
94202f12 JH |
2743 | /* |
2744 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
2745 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
2746 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
2747 | * request. | |
2748 | */ | |
2749 | gup_flags |= FOLL_GET; | |
eddb1c22 JH |
2750 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
2751 | } | |
050a9adc | 2752 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
2753 | |
2754 | /** | |
2755 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
2756 | * | |
3faa52c0 JH |
2757 | * @start: starting user address |
2758 | * @nr_pages: number of pages from start to pin | |
2759 | * @gup_flags: flags modifying pin behaviour | |
2760 | * @pages: array that receives pointers to the pages pinned. | |
2761 | * Should be at least nr_pages long. | |
2762 | * | |
2763 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
2764 | * get_user_pages_fast() for documentation on the function arguments, because | |
2765 | * the arguments here are identical. | |
2766 | * | |
2767 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2768 | * see Documentation/core-api/pin_user_pages.rst for further details. |
eddb1c22 JH |
2769 | */ |
2770 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
2771 | unsigned int gup_flags, struct page **pages) | |
2772 | { | |
3faa52c0 JH |
2773 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2774 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2775 | return -EINVAL; | |
2776 | ||
2777 | gup_flags |= FOLL_PIN; | |
2778 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); | |
eddb1c22 JH |
2779 | } |
2780 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
2781 | ||
104acc32 | 2782 | /* |
dadbb612 SJ |
2783 | * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior |
2784 | * is the same, except that this one sets FOLL_PIN instead of FOLL_GET. | |
104acc32 JH |
2785 | * |
2786 | * The API rules are the same, too: no negative values may be returned. | |
2787 | */ | |
2788 | int pin_user_pages_fast_only(unsigned long start, int nr_pages, | |
2789 | unsigned int gup_flags, struct page **pages) | |
2790 | { | |
2791 | int nr_pinned; | |
2792 | ||
2793 | /* | |
2794 | * FOLL_GET and FOLL_PIN are mutually exclusive. Note that the API | |
2795 | * rules require returning 0, rather than -errno: | |
2796 | */ | |
2797 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2798 | return 0; | |
2799 | /* | |
2800 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2801 | * this routine: no fall back to regular ("slow") GUP. | |
2802 | */ | |
2803 | gup_flags |= (FOLL_PIN | FOLL_FAST_ONLY); | |
2804 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, | |
2805 | pages); | |
2806 | /* | |
2807 | * This routine is not allowed to return negative values. However, | |
2808 | * internal_get_user_pages_fast() *can* return -errno. Therefore, | |
2809 | * correct for that here: | |
2810 | */ | |
2811 | if (nr_pinned < 0) | |
2812 | nr_pinned = 0; | |
2813 | ||
2814 | return nr_pinned; | |
2815 | } | |
2816 | EXPORT_SYMBOL_GPL(pin_user_pages_fast_only); | |
2817 | ||
eddb1c22 | 2818 | /** |
64019a2e | 2819 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 2820 | * |
3faa52c0 JH |
2821 | * @mm: mm_struct of target mm |
2822 | * @start: starting user address | |
2823 | * @nr_pages: number of pages from start to pin | |
2824 | * @gup_flags: flags modifying lookup behaviour | |
2825 | * @pages: array that receives pointers to the pages pinned. | |
2826 | * Should be at least nr_pages long. Or NULL, if caller | |
2827 | * only intends to ensure the pages are faulted in. | |
2828 | * @vmas: array of pointers to vmas corresponding to each page. | |
2829 | * Or NULL if the caller does not require them. | |
2830 | * @locked: pointer to lock flag indicating whether lock is held and | |
2831 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2832 | * utilised. Lock must initially be held. | |
2833 | * | |
2834 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
2835 | * get_user_pages_remote() for documentation on the function arguments, because | |
2836 | * the arguments here are identical. | |
2837 | * | |
2838 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2839 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 | 2840 | */ |
64019a2e | 2841 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2842 | unsigned long start, unsigned long nr_pages, |
2843 | unsigned int gup_flags, struct page **pages, | |
2844 | struct vm_area_struct **vmas, int *locked) | |
2845 | { | |
3faa52c0 JH |
2846 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2847 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2848 | return -EINVAL; | |
2849 | ||
2850 | gup_flags |= FOLL_PIN; | |
64019a2e | 2851 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
3faa52c0 | 2852 | pages, vmas, locked); |
eddb1c22 JH |
2853 | } |
2854 | EXPORT_SYMBOL(pin_user_pages_remote); | |
2855 | ||
2856 | /** | |
2857 | * pin_user_pages() - pin user pages in memory for use by other devices | |
2858 | * | |
3faa52c0 JH |
2859 | * @start: starting user address |
2860 | * @nr_pages: number of pages from start to pin | |
2861 | * @gup_flags: flags modifying lookup behaviour | |
2862 | * @pages: array that receives pointers to the pages pinned. | |
2863 | * Should be at least nr_pages long. Or NULL, if caller | |
2864 | * only intends to ensure the pages are faulted in. | |
2865 | * @vmas: array of pointers to vmas corresponding to each page. | |
2866 | * Or NULL if the caller does not require them. | |
2867 | * | |
2868 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
2869 | * FOLL_PIN is set. | |
2870 | * | |
2871 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2872 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 JH |
2873 | */ |
2874 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
2875 | unsigned int gup_flags, struct page **pages, | |
2876 | struct vm_area_struct **vmas) | |
2877 | { | |
3faa52c0 JH |
2878 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2879 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2880 | return -EINVAL; | |
2881 | ||
2882 | gup_flags |= FOLL_PIN; | |
64019a2e | 2883 | return __gup_longterm_locked(current->mm, start, nr_pages, |
3faa52c0 | 2884 | pages, vmas, gup_flags); |
eddb1c22 JH |
2885 | } |
2886 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
2887 | |
2888 | /* | |
2889 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
2890 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
2891 | * FOLL_PIN and rejects FOLL_GET. | |
2892 | */ | |
2893 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
2894 | struct page **pages, unsigned int gup_flags) | |
2895 | { | |
2896 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
2897 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2898 | return -EINVAL; | |
2899 | ||
2900 | gup_flags |= FOLL_PIN; | |
2901 | return get_user_pages_unlocked(start, nr_pages, pages, gup_flags); | |
2902 | } | |
2903 | EXPORT_SYMBOL(pin_user_pages_unlocked); | |
420c2091 JH |
2904 | |
2905 | /* | |
2906 | * pin_user_pages_locked() is the FOLL_PIN variant of get_user_pages_locked(). | |
2907 | * Behavior is the same, except that this one sets FOLL_PIN and rejects | |
2908 | * FOLL_GET. | |
2909 | */ | |
2910 | long pin_user_pages_locked(unsigned long start, unsigned long nr_pages, | |
2911 | unsigned int gup_flags, struct page **pages, | |
2912 | int *locked) | |
2913 | { | |
2914 | /* | |
2915 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
2916 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2917 | * vmas. As there are no users of this flag in this call we simply | |
2918 | * disallow this option for now. | |
2919 | */ | |
2920 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
2921 | return -EINVAL; | |
2922 | ||
2923 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
2924 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2925 | return -EINVAL; | |
2926 | ||
2927 | gup_flags |= FOLL_PIN; | |
64019a2e | 2928 | return __get_user_pages_locked(current->mm, start, nr_pages, |
420c2091 JH |
2929 | pages, NULL, locked, |
2930 | gup_flags | FOLL_TOUCH); | |
2931 | } | |
2932 | EXPORT_SYMBOL(pin_user_pages_locked); |