Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
4bbd4c77 KS |
2 | #include <linux/kernel.h> |
3 | #include <linux/errno.h> | |
4 | #include <linux/err.h> | |
5 | #include <linux/spinlock.h> | |
6 | ||
4bbd4c77 | 7 | #include <linux/mm.h> |
3565fce3 | 8 | #include <linux/memremap.h> |
4bbd4c77 KS |
9 | #include <linux/pagemap.h> |
10 | #include <linux/rmap.h> | |
11 | #include <linux/swap.h> | |
12 | #include <linux/swapops.h> | |
1507f512 | 13 | #include <linux/secretmem.h> |
4bbd4c77 | 14 | |
174cd4b1 | 15 | #include <linux/sched/signal.h> |
2667f50e | 16 | #include <linux/rwsem.h> |
f30c59e9 | 17 | #include <linux/hugetlb.h> |
9a4e9f3b AK |
18 | #include <linux/migrate.h> |
19 | #include <linux/mm_inline.h> | |
20 | #include <linux/sched/mm.h> | |
1027e443 | 21 | |
33a709b2 | 22 | #include <asm/mmu_context.h> |
1027e443 | 23 | #include <asm/tlbflush.h> |
2667f50e | 24 | |
4bbd4c77 KS |
25 | #include "internal.h" |
26 | ||
df06b37f KB |
27 | struct follow_page_context { |
28 | struct dev_pagemap *pgmap; | |
29 | unsigned int page_mask; | |
30 | }; | |
31 | ||
47e29d32 JH |
32 | static void hpage_pincount_add(struct page *page, int refs) |
33 | { | |
34 | VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); | |
35 | VM_BUG_ON_PAGE(page != compound_head(page), page); | |
36 | ||
37 | atomic_add(refs, compound_pincount_ptr(page)); | |
38 | } | |
39 | ||
40 | static void hpage_pincount_sub(struct page *page, int refs) | |
41 | { | |
42 | VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); | |
43 | VM_BUG_ON_PAGE(page != compound_head(page), page); | |
44 | ||
45 | atomic_sub(refs, compound_pincount_ptr(page)); | |
46 | } | |
47 | ||
c24d3732 JH |
48 | /* Equivalent to calling put_page() @refs times. */ |
49 | static void put_page_refs(struct page *page, int refs) | |
50 | { | |
51 | #ifdef CONFIG_DEBUG_VM | |
52 | if (VM_WARN_ON_ONCE_PAGE(page_ref_count(page) < refs, page)) | |
53 | return; | |
54 | #endif | |
55 | ||
56 | /* | |
57 | * Calling put_page() for each ref is unnecessarily slow. Only the last | |
58 | * ref needs a put_page(). | |
59 | */ | |
60 | if (refs > 1) | |
61 | page_ref_sub(page, refs - 1); | |
62 | put_page(page); | |
63 | } | |
64 | ||
cd1adf1b LT |
65 | /* |
66 | * Return the compound head page with ref appropriately incremented, | |
67 | * or NULL if that failed. | |
a707cdd5 | 68 | */ |
cd1adf1b | 69 | static inline struct page *try_get_compound_head(struct page *page, int refs) |
a707cdd5 JH |
70 | { |
71 | struct page *head = compound_head(page); | |
72 | ||
73 | if (WARN_ON_ONCE(page_ref_count(head) < 0)) | |
74 | return NULL; | |
75 | if (unlikely(!page_cache_add_speculative(head, refs))) | |
76 | return NULL; | |
c24d3732 JH |
77 | |
78 | /* | |
79 | * At this point we have a stable reference to the head page; but it | |
80 | * could be that between the compound_head() lookup and the refcount | |
81 | * increment, the compound page was split, in which case we'd end up | |
82 | * holding a reference on a page that has nothing to do with the page | |
83 | * we were given anymore. | |
84 | * So now that the head page is stable, recheck that the pages still | |
85 | * belong together. | |
86 | */ | |
87 | if (unlikely(compound_head(page) != head)) { | |
88 | put_page_refs(head, refs); | |
89 | return NULL; | |
90 | } | |
91 | ||
a707cdd5 JH |
92 | return head; |
93 | } | |
94 | ||
3967db22 | 95 | /** |
3faa52c0 JH |
96 | * try_grab_compound_head() - attempt to elevate a page's refcount, by a |
97 | * flags-dependent amount. | |
98 | * | |
3967db22 JH |
99 | * Even though the name includes "compound_head", this function is still |
100 | * appropriate for callers that have a non-compound @page to get. | |
101 | * | |
102 | * @page: pointer to page to be grabbed | |
103 | * @refs: the value to (effectively) add to the page's refcount | |
104 | * @flags: gup flags: these are the FOLL_* flag values. | |
105 | * | |
3faa52c0 JH |
106 | * "grab" names in this file mean, "look at flags to decide whether to use |
107 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
108 | * | |
109 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
110 | * same time. (That's true throughout the get_user_pages*() and | |
111 | * pin_user_pages*() APIs.) Cases: | |
112 | * | |
3967db22 JH |
113 | * FOLL_GET: page's refcount will be incremented by @refs. |
114 | * | |
115 | * FOLL_PIN on compound pages that are > two pages long: page's refcount will | |
116 | * be incremented by @refs, and page[2].hpage_pinned_refcount will be | |
117 | * incremented by @refs * GUP_PIN_COUNTING_BIAS. | |
118 | * | |
119 | * FOLL_PIN on normal pages, or compound pages that are two pages long: | |
120 | * page's refcount will be incremented by @refs * GUP_PIN_COUNTING_BIAS. | |
3faa52c0 JH |
121 | * |
122 | * Return: head page (with refcount appropriately incremented) for success, or | |
123 | * NULL upon failure. If neither FOLL_GET nor FOLL_PIN was set, that's | |
124 | * considered failure, and furthermore, a likely bug in the caller, so a warning | |
125 | * is also emitted. | |
126 | */ | |
c36c04c2 JH |
127 | __maybe_unused struct page *try_grab_compound_head(struct page *page, |
128 | int refs, unsigned int flags) | |
3faa52c0 JH |
129 | { |
130 | if (flags & FOLL_GET) | |
131 | return try_get_compound_head(page, refs); | |
132 | else if (flags & FOLL_PIN) { | |
df3a0a21 | 133 | /* |
d1e153fe PT |
134 | * Can't do FOLL_LONGTERM + FOLL_PIN gup fast path if not in a |
135 | * right zone, so fail and let the caller fall back to the slow | |
136 | * path. | |
df3a0a21 | 137 | */ |
d1e153fe PT |
138 | if (unlikely((flags & FOLL_LONGTERM) && |
139 | !is_pinnable_page(page))) | |
df3a0a21 PL |
140 | return NULL; |
141 | ||
c24d3732 JH |
142 | /* |
143 | * CAUTION: Don't use compound_head() on the page before this | |
144 | * point, the result won't be stable. | |
145 | */ | |
146 | page = try_get_compound_head(page, refs); | |
147 | if (!page) | |
148 | return NULL; | |
149 | ||
47e29d32 JH |
150 | /* |
151 | * When pinning a compound page of order > 1 (which is what | |
152 | * hpage_pincount_available() checks for), use an exact count to | |
153 | * track it, via hpage_pincount_add/_sub(). | |
154 | * | |
155 | * However, be sure to *also* increment the normal page refcount | |
156 | * field at least once, so that the page really is pinned. | |
3967db22 JH |
157 | * That's why the refcount from the earlier |
158 | * try_get_compound_head() is left intact. | |
47e29d32 | 159 | */ |
47e29d32 JH |
160 | if (hpage_pincount_available(page)) |
161 | hpage_pincount_add(page, refs); | |
c24d3732 JH |
162 | else |
163 | page_ref_add(page, refs * (GUP_PIN_COUNTING_BIAS - 1)); | |
47e29d32 | 164 | |
1970dc6f | 165 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, |
0fef147b | 166 | refs); |
1970dc6f | 167 | |
47e29d32 | 168 | return page; |
3faa52c0 JH |
169 | } |
170 | ||
171 | WARN_ON_ONCE(1); | |
172 | return NULL; | |
173 | } | |
174 | ||
4509b42c JG |
175 | static void put_compound_head(struct page *page, int refs, unsigned int flags) |
176 | { | |
177 | if (flags & FOLL_PIN) { | |
178 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, | |
179 | refs); | |
180 | ||
181 | if (hpage_pincount_available(page)) | |
182 | hpage_pincount_sub(page, refs); | |
183 | else | |
184 | refs *= GUP_PIN_COUNTING_BIAS; | |
185 | } | |
186 | ||
c24d3732 | 187 | put_page_refs(page, refs); |
4509b42c JG |
188 | } |
189 | ||
3faa52c0 JH |
190 | /** |
191 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
192 | * | |
193 | * This might not do anything at all, depending on the flags argument. | |
194 | * | |
195 | * "grab" names in this file mean, "look at flags to decide whether to use | |
196 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
197 | * | |
198 | * @page: pointer to page to be grabbed | |
199 | * @flags: gup flags: these are the FOLL_* flag values. | |
200 | * | |
201 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same | |
3967db22 JH |
202 | * time. Cases: please see the try_grab_compound_head() documentation, with |
203 | * "refs=1". | |
3faa52c0 JH |
204 | * |
205 | * Return: true for success, or if no action was required (if neither FOLL_PIN | |
206 | * nor FOLL_GET was set, nothing is done). False for failure: FOLL_GET or | |
207 | * FOLL_PIN was set, but the page could not be grabbed. | |
208 | */ | |
209 | bool __must_check try_grab_page(struct page *page, unsigned int flags) | |
210 | { | |
c36c04c2 | 211 | WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN)); |
3faa52c0 | 212 | |
c36c04c2 JH |
213 | if (flags & FOLL_GET) |
214 | return try_get_page(page); | |
215 | else if (flags & FOLL_PIN) { | |
c36c04c2 JH |
216 | page = compound_head(page); |
217 | ||
218 | if (WARN_ON_ONCE(page_ref_count(page) <= 0)) | |
219 | return false; | |
220 | ||
c36c04c2 JH |
221 | /* |
222 | * Similar to try_grab_compound_head(): even if using the | |
223 | * hpage_pincount_add/_sub() routines, be sure to | |
224 | * *also* increment the normal page refcount field at least | |
225 | * once, so that the page really is pinned. | |
226 | */ | |
8ea2979c MWO |
227 | if (hpage_pincount_available(page)) { |
228 | page_ref_add(page, 1); | |
229 | hpage_pincount_add(page, 1); | |
230 | } else { | |
231 | page_ref_add(page, GUP_PIN_COUNTING_BIAS); | |
232 | } | |
c36c04c2 JH |
233 | |
234 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1); | |
235 | } | |
236 | ||
237 | return true; | |
3faa52c0 JH |
238 | } |
239 | ||
3faa52c0 JH |
240 | /** |
241 | * unpin_user_page() - release a dma-pinned page | |
242 | * @page: pointer to page to be released | |
243 | * | |
244 | * Pages that were pinned via pin_user_pages*() must be released via either | |
245 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
246 | * that such pages can be separately tracked and uniquely handled. In | |
247 | * particular, interactions with RDMA and filesystems need special handling. | |
248 | */ | |
249 | void unpin_user_page(struct page *page) | |
250 | { | |
4509b42c | 251 | put_compound_head(compound_head(page), 1, FOLL_PIN); |
3faa52c0 JH |
252 | } |
253 | EXPORT_SYMBOL(unpin_user_page); | |
254 | ||
8f39f5fc MWO |
255 | static inline struct page *compound_range_next(struct page *start, |
256 | unsigned long npages, unsigned long i, unsigned int *ntails) | |
458a4f78 JM |
257 | { |
258 | struct page *next, *page; | |
259 | unsigned int nr = 1; | |
260 | ||
8f39f5fc | 261 | next = start + i; |
458a4f78 | 262 | page = compound_head(next); |
0b046e12 | 263 | if (PageHead(page)) |
458a4f78 JM |
264 | nr = min_t(unsigned int, |
265 | page + compound_nr(page) - next, npages - i); | |
266 | ||
458a4f78 | 267 | *ntails = nr; |
8f39f5fc | 268 | return page; |
458a4f78 JM |
269 | } |
270 | ||
8745d7f6 JM |
271 | static inline void compound_next(unsigned long i, unsigned long npages, |
272 | struct page **list, struct page **head, | |
273 | unsigned int *ntails) | |
274 | { | |
275 | struct page *page; | |
276 | unsigned int nr; | |
277 | ||
8745d7f6 JM |
278 | page = compound_head(list[i]); |
279 | for (nr = i + 1; nr < npages; nr++) { | |
280 | if (compound_head(list[nr]) != page) | |
281 | break; | |
282 | } | |
283 | ||
284 | *head = page; | |
285 | *ntails = nr - i; | |
286 | } | |
287 | ||
fc1d8e7c | 288 | /** |
f1f6a7dd | 289 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 290 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 291 | * @npages: number of pages in the @pages array. |
2d15eb31 | 292 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
293 | * |
294 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
295 | * variants called on that page. | |
296 | * | |
297 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 298 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
299 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
300 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 301 | * |
f1f6a7dd | 302 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 303 | * |
2d15eb31 | 304 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
305 | * required, then the caller should a) verify that this is really correct, | |
306 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 307 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
308 | * |
309 | */ | |
f1f6a7dd JH |
310 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
311 | bool make_dirty) | |
fc1d8e7c | 312 | { |
2d15eb31 | 313 | unsigned long index; |
31b912de JM |
314 | struct page *head; |
315 | unsigned int ntails; | |
2d15eb31 | 316 | |
317 | if (!make_dirty) { | |
f1f6a7dd | 318 | unpin_user_pages(pages, npages); |
2d15eb31 | 319 | return; |
320 | } | |
321 | ||
e7602748 MWO |
322 | for (index = 0; index < npages; index += ntails) { |
323 | compound_next(index, npages, pages, &head, &ntails); | |
2d15eb31 | 324 | /* |
325 | * Checking PageDirty at this point may race with | |
326 | * clear_page_dirty_for_io(), but that's OK. Two key | |
327 | * cases: | |
328 | * | |
329 | * 1) This code sees the page as already dirty, so it | |
330 | * skips the call to set_page_dirty(). That could happen | |
331 | * because clear_page_dirty_for_io() called | |
332 | * page_mkclean(), followed by set_page_dirty(). | |
333 | * However, now the page is going to get written back, | |
334 | * which meets the original intention of setting it | |
335 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
336 | * on to call TestClearPageDirty(), and write the page | |
337 | * back. | |
338 | * | |
339 | * 2) This code sees the page as clean, so it calls | |
340 | * set_page_dirty(). The page stays dirty, despite being | |
341 | * written back, so it gets written back again in the | |
342 | * next writeback cycle. This is harmless. | |
343 | */ | |
31b912de JM |
344 | if (!PageDirty(head)) |
345 | set_page_dirty_lock(head); | |
346 | put_compound_head(head, ntails, FOLL_PIN); | |
2d15eb31 | 347 | } |
fc1d8e7c | 348 | } |
f1f6a7dd | 349 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c | 350 | |
458a4f78 JM |
351 | /** |
352 | * unpin_user_page_range_dirty_lock() - release and optionally dirty | |
353 | * gup-pinned page range | |
354 | * | |
355 | * @page: the starting page of a range maybe marked dirty, and definitely released. | |
356 | * @npages: number of consecutive pages to release. | |
357 | * @make_dirty: whether to mark the pages dirty | |
358 | * | |
359 | * "gup-pinned page range" refers to a range of pages that has had one of the | |
360 | * pin_user_pages() variants called on that page. | |
361 | * | |
362 | * For the page ranges defined by [page .. page+npages], make that range (or | |
363 | * its head pages, if a compound page) dirty, if @make_dirty is true, and if the | |
364 | * page range was previously listed as clean. | |
365 | * | |
366 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is | |
367 | * required, then the caller should a) verify that this is really correct, | |
368 | * because _lock() is usually required, and b) hand code it: | |
369 | * set_page_dirty_lock(), unpin_user_page(). | |
370 | * | |
371 | */ | |
372 | void unpin_user_page_range_dirty_lock(struct page *page, unsigned long npages, | |
373 | bool make_dirty) | |
374 | { | |
375 | unsigned long index; | |
376 | struct page *head; | |
377 | unsigned int ntails; | |
378 | ||
a5f100db | 379 | for (index = 0; index < npages; index += ntails) { |
8f39f5fc | 380 | head = compound_range_next(page, npages, index, &ntails); |
458a4f78 JM |
381 | if (make_dirty && !PageDirty(head)) |
382 | set_page_dirty_lock(head); | |
383 | put_compound_head(head, ntails, FOLL_PIN); | |
384 | } | |
385 | } | |
386 | EXPORT_SYMBOL(unpin_user_page_range_dirty_lock); | |
387 | ||
fc1d8e7c | 388 | /** |
f1f6a7dd | 389 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
390 | * @pages: array of pages to be marked dirty and released. |
391 | * @npages: number of pages in the @pages array. | |
392 | * | |
f1f6a7dd | 393 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 394 | * |
f1f6a7dd | 395 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 396 | */ |
f1f6a7dd | 397 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c JH |
398 | { |
399 | unsigned long index; | |
31b912de JM |
400 | struct page *head; |
401 | unsigned int ntails; | |
fc1d8e7c | 402 | |
146608bb JH |
403 | /* |
404 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
405 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
406 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
407 | */ | |
408 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
409 | return; | |
31b912de | 410 | |
e7602748 MWO |
411 | for (index = 0; index < npages; index += ntails) { |
412 | compound_next(index, npages, pages, &head, &ntails); | |
31b912de | 413 | put_compound_head(head, ntails, FOLL_PIN); |
e7602748 | 414 | } |
fc1d8e7c | 415 | } |
f1f6a7dd | 416 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 417 | |
a458b76a AA |
418 | /* |
419 | * Set the MMF_HAS_PINNED if not set yet; after set it'll be there for the mm's | |
420 | * lifecycle. Avoid setting the bit unless necessary, or it might cause write | |
421 | * cache bouncing on large SMP machines for concurrent pinned gups. | |
422 | */ | |
423 | static inline void mm_set_has_pinned_flag(unsigned long *mm_flags) | |
424 | { | |
425 | if (!test_bit(MMF_HAS_PINNED, mm_flags)) | |
426 | set_bit(MMF_HAS_PINNED, mm_flags); | |
427 | } | |
428 | ||
050a9adc | 429 | #ifdef CONFIG_MMU |
69e68b4f KS |
430 | static struct page *no_page_table(struct vm_area_struct *vma, |
431 | unsigned int flags) | |
4bbd4c77 | 432 | { |
69e68b4f KS |
433 | /* |
434 | * When core dumping an enormous anonymous area that nobody | |
435 | * has touched so far, we don't want to allocate unnecessary pages or | |
436 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
437 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
438 | * But we can only make this optimization where a hole would surely | |
439 | * be zero-filled if handle_mm_fault() actually did handle it. | |
440 | */ | |
a0137f16 AK |
441 | if ((flags & FOLL_DUMP) && |
442 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
443 | return ERR_PTR(-EFAULT); |
444 | return NULL; | |
445 | } | |
4bbd4c77 | 446 | |
1027e443 KS |
447 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
448 | pte_t *pte, unsigned int flags) | |
449 | { | |
450 | /* No page to get reference */ | |
451 | if (flags & FOLL_GET) | |
452 | return -EFAULT; | |
453 | ||
454 | if (flags & FOLL_TOUCH) { | |
455 | pte_t entry = *pte; | |
456 | ||
457 | if (flags & FOLL_WRITE) | |
458 | entry = pte_mkdirty(entry); | |
459 | entry = pte_mkyoung(entry); | |
460 | ||
461 | if (!pte_same(*pte, entry)) { | |
462 | set_pte_at(vma->vm_mm, address, pte, entry); | |
463 | update_mmu_cache(vma, address, pte); | |
464 | } | |
465 | } | |
466 | ||
467 | /* Proper page table entry exists, but no corresponding struct page */ | |
468 | return -EEXIST; | |
469 | } | |
470 | ||
19be0eaf | 471 | /* |
a308c71b PX |
472 | * FOLL_FORCE can write to even unwritable pte's, but only |
473 | * after we've gone through a COW cycle and they are dirty. | |
19be0eaf LT |
474 | */ |
475 | static inline bool can_follow_write_pte(pte_t pte, unsigned int flags) | |
476 | { | |
a308c71b PX |
477 | return pte_write(pte) || |
478 | ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte)); | |
19be0eaf LT |
479 | } |
480 | ||
69e68b4f | 481 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
482 | unsigned long address, pmd_t *pmd, unsigned int flags, |
483 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
484 | { |
485 | struct mm_struct *mm = vma->vm_mm; | |
486 | struct page *page; | |
487 | spinlock_t *ptl; | |
488 | pte_t *ptep, pte; | |
f28d4363 | 489 | int ret; |
4bbd4c77 | 490 | |
eddb1c22 JH |
491 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
492 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
493 | (FOLL_PIN | FOLL_GET))) | |
494 | return ERR_PTR(-EINVAL); | |
69e68b4f | 495 | retry: |
4bbd4c77 | 496 | if (unlikely(pmd_bad(*pmd))) |
69e68b4f | 497 | return no_page_table(vma, flags); |
4bbd4c77 KS |
498 | |
499 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
4bbd4c77 KS |
500 | pte = *ptep; |
501 | if (!pte_present(pte)) { | |
502 | swp_entry_t entry; | |
503 | /* | |
504 | * KSM's break_ksm() relies upon recognizing a ksm page | |
505 | * even while it is being migrated, so for that case we | |
506 | * need migration_entry_wait(). | |
507 | */ | |
508 | if (likely(!(flags & FOLL_MIGRATION))) | |
509 | goto no_page; | |
0661a336 | 510 | if (pte_none(pte)) |
4bbd4c77 KS |
511 | goto no_page; |
512 | entry = pte_to_swp_entry(pte); | |
513 | if (!is_migration_entry(entry)) | |
514 | goto no_page; | |
515 | pte_unmap_unlock(ptep, ptl); | |
516 | migration_entry_wait(mm, pmd, address); | |
69e68b4f | 517 | goto retry; |
4bbd4c77 | 518 | } |
8a0516ed | 519 | if ((flags & FOLL_NUMA) && pte_protnone(pte)) |
4bbd4c77 | 520 | goto no_page; |
19be0eaf | 521 | if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) { |
69e68b4f KS |
522 | pte_unmap_unlock(ptep, ptl); |
523 | return NULL; | |
524 | } | |
4bbd4c77 KS |
525 | |
526 | page = vm_normal_page(vma, address, pte); | |
3faa52c0 | 527 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 528 | /* |
3faa52c0 JH |
529 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
530 | * case since they are only valid while holding the pgmap | |
531 | * reference. | |
3565fce3 | 532 | */ |
df06b37f KB |
533 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
534 | if (*pgmap) | |
3565fce3 DW |
535 | page = pte_page(pte); |
536 | else | |
537 | goto no_page; | |
538 | } else if (unlikely(!page)) { | |
1027e443 KS |
539 | if (flags & FOLL_DUMP) { |
540 | /* Avoid special (like zero) pages in core dumps */ | |
541 | page = ERR_PTR(-EFAULT); | |
542 | goto out; | |
543 | } | |
544 | ||
545 | if (is_zero_pfn(pte_pfn(pte))) { | |
546 | page = pte_page(pte); | |
547 | } else { | |
1027e443 KS |
548 | ret = follow_pfn_pte(vma, address, ptep, flags); |
549 | page = ERR_PTR(ret); | |
550 | goto out; | |
551 | } | |
4bbd4c77 KS |
552 | } |
553 | ||
3faa52c0 JH |
554 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
555 | if (unlikely(!try_grab_page(page, flags))) { | |
556 | page = ERR_PTR(-ENOMEM); | |
557 | goto out; | |
8fde12ca | 558 | } |
f28d4363 CI |
559 | /* |
560 | * We need to make the page accessible if and only if we are going | |
561 | * to access its content (the FOLL_PIN case). Please see | |
562 | * Documentation/core-api/pin_user_pages.rst for details. | |
563 | */ | |
564 | if (flags & FOLL_PIN) { | |
565 | ret = arch_make_page_accessible(page); | |
566 | if (ret) { | |
567 | unpin_user_page(page); | |
568 | page = ERR_PTR(ret); | |
569 | goto out; | |
570 | } | |
571 | } | |
4bbd4c77 KS |
572 | if (flags & FOLL_TOUCH) { |
573 | if ((flags & FOLL_WRITE) && | |
574 | !pte_dirty(pte) && !PageDirty(page)) | |
575 | set_page_dirty(page); | |
576 | /* | |
577 | * pte_mkyoung() would be more correct here, but atomic care | |
578 | * is needed to avoid losing the dirty bit: it is easier to use | |
579 | * mark_page_accessed(). | |
580 | */ | |
581 | mark_page_accessed(page); | |
582 | } | |
1027e443 | 583 | out: |
4bbd4c77 | 584 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 585 | return page; |
4bbd4c77 KS |
586 | no_page: |
587 | pte_unmap_unlock(ptep, ptl); | |
588 | if (!pte_none(pte)) | |
69e68b4f KS |
589 | return NULL; |
590 | return no_page_table(vma, flags); | |
591 | } | |
592 | ||
080dbb61 AK |
593 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
594 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
595 | unsigned int flags, |
596 | struct follow_page_context *ctx) | |
69e68b4f | 597 | { |
68827280 | 598 | pmd_t *pmd, pmdval; |
69e68b4f KS |
599 | spinlock_t *ptl; |
600 | struct page *page; | |
601 | struct mm_struct *mm = vma->vm_mm; | |
602 | ||
080dbb61 | 603 | pmd = pmd_offset(pudp, address); |
68827280 HY |
604 | /* |
605 | * The READ_ONCE() will stabilize the pmdval in a register or | |
606 | * on the stack so that it will stop changing under the code. | |
607 | */ | |
608 | pmdval = READ_ONCE(*pmd); | |
609 | if (pmd_none(pmdval)) | |
69e68b4f | 610 | return no_page_table(vma, flags); |
be9d3045 | 611 | if (pmd_huge(pmdval) && is_vm_hugetlb_page(vma)) { |
e66f17ff NH |
612 | page = follow_huge_pmd(mm, address, pmd, flags); |
613 | if (page) | |
614 | return page; | |
615 | return no_page_table(vma, flags); | |
69e68b4f | 616 | } |
68827280 | 617 | if (is_hugepd(__hugepd(pmd_val(pmdval)))) { |
4dc71451 | 618 | page = follow_huge_pd(vma, address, |
68827280 | 619 | __hugepd(pmd_val(pmdval)), flags, |
4dc71451 AK |
620 | PMD_SHIFT); |
621 | if (page) | |
622 | return page; | |
623 | return no_page_table(vma, flags); | |
624 | } | |
84c3fc4e | 625 | retry: |
68827280 | 626 | if (!pmd_present(pmdval)) { |
28b0ee3f LX |
627 | /* |
628 | * Should never reach here, if thp migration is not supported; | |
629 | * Otherwise, it must be a thp migration entry. | |
630 | */ | |
631 | VM_BUG_ON(!thp_migration_supported() || | |
632 | !is_pmd_migration_entry(pmdval)); | |
633 | ||
84c3fc4e ZY |
634 | if (likely(!(flags & FOLL_MIGRATION))) |
635 | return no_page_table(vma, flags); | |
28b0ee3f LX |
636 | |
637 | pmd_migration_entry_wait(mm, pmd); | |
68827280 HY |
638 | pmdval = READ_ONCE(*pmd); |
639 | /* | |
640 | * MADV_DONTNEED may convert the pmd to null because | |
c1e8d7c6 | 641 | * mmap_lock is held in read mode |
68827280 HY |
642 | */ |
643 | if (pmd_none(pmdval)) | |
644 | return no_page_table(vma, flags); | |
84c3fc4e ZY |
645 | goto retry; |
646 | } | |
68827280 | 647 | if (pmd_devmap(pmdval)) { |
3565fce3 | 648 | ptl = pmd_lock(mm, pmd); |
df06b37f | 649 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
650 | spin_unlock(ptl); |
651 | if (page) | |
652 | return page; | |
653 | } | |
68827280 | 654 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 655 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 656 | |
68827280 | 657 | if ((flags & FOLL_NUMA) && pmd_protnone(pmdval)) |
db08f203 AK |
658 | return no_page_table(vma, flags); |
659 | ||
84c3fc4e | 660 | retry_locked: |
6742d293 | 661 | ptl = pmd_lock(mm, pmd); |
68827280 HY |
662 | if (unlikely(pmd_none(*pmd))) { |
663 | spin_unlock(ptl); | |
664 | return no_page_table(vma, flags); | |
665 | } | |
84c3fc4e ZY |
666 | if (unlikely(!pmd_present(*pmd))) { |
667 | spin_unlock(ptl); | |
668 | if (likely(!(flags & FOLL_MIGRATION))) | |
669 | return no_page_table(vma, flags); | |
670 | pmd_migration_entry_wait(mm, pmd); | |
671 | goto retry_locked; | |
672 | } | |
6742d293 KS |
673 | if (unlikely(!pmd_trans_huge(*pmd))) { |
674 | spin_unlock(ptl); | |
df06b37f | 675 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 676 | } |
4066c119 | 677 | if (flags & FOLL_SPLIT_PMD) { |
6742d293 KS |
678 | int ret; |
679 | page = pmd_page(*pmd); | |
680 | if (is_huge_zero_page(page)) { | |
681 | spin_unlock(ptl); | |
682 | ret = 0; | |
78ddc534 | 683 | split_huge_pmd(vma, pmd, address); |
337d9abf NH |
684 | if (pmd_trans_unstable(pmd)) |
685 | ret = -EBUSY; | |
4066c119 | 686 | } else { |
bfe7b00d SL |
687 | spin_unlock(ptl); |
688 | split_huge_pmd(vma, pmd, address); | |
689 | ret = pte_alloc(mm, pmd) ? -ENOMEM : 0; | |
6742d293 KS |
690 | } |
691 | ||
692 | return ret ? ERR_PTR(ret) : | |
df06b37f | 693 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 694 | } |
6742d293 KS |
695 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
696 | spin_unlock(ptl); | |
df06b37f | 697 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 698 | return page; |
4bbd4c77 KS |
699 | } |
700 | ||
080dbb61 AK |
701 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
702 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
703 | unsigned int flags, |
704 | struct follow_page_context *ctx) | |
080dbb61 AK |
705 | { |
706 | pud_t *pud; | |
707 | spinlock_t *ptl; | |
708 | struct page *page; | |
709 | struct mm_struct *mm = vma->vm_mm; | |
710 | ||
711 | pud = pud_offset(p4dp, address); | |
712 | if (pud_none(*pud)) | |
713 | return no_page_table(vma, flags); | |
be9d3045 | 714 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) { |
080dbb61 AK |
715 | page = follow_huge_pud(mm, address, pud, flags); |
716 | if (page) | |
717 | return page; | |
718 | return no_page_table(vma, flags); | |
719 | } | |
4dc71451 AK |
720 | if (is_hugepd(__hugepd(pud_val(*pud)))) { |
721 | page = follow_huge_pd(vma, address, | |
722 | __hugepd(pud_val(*pud)), flags, | |
723 | PUD_SHIFT); | |
724 | if (page) | |
725 | return page; | |
726 | return no_page_table(vma, flags); | |
727 | } | |
080dbb61 AK |
728 | if (pud_devmap(*pud)) { |
729 | ptl = pud_lock(mm, pud); | |
df06b37f | 730 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
731 | spin_unlock(ptl); |
732 | if (page) | |
733 | return page; | |
734 | } | |
735 | if (unlikely(pud_bad(*pud))) | |
736 | return no_page_table(vma, flags); | |
737 | ||
df06b37f | 738 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
739 | } |
740 | ||
080dbb61 AK |
741 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
742 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
743 | unsigned int flags, |
744 | struct follow_page_context *ctx) | |
080dbb61 AK |
745 | { |
746 | p4d_t *p4d; | |
4dc71451 | 747 | struct page *page; |
080dbb61 AK |
748 | |
749 | p4d = p4d_offset(pgdp, address); | |
750 | if (p4d_none(*p4d)) | |
751 | return no_page_table(vma, flags); | |
752 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
753 | if (unlikely(p4d_bad(*p4d))) | |
754 | return no_page_table(vma, flags); | |
755 | ||
4dc71451 AK |
756 | if (is_hugepd(__hugepd(p4d_val(*p4d)))) { |
757 | page = follow_huge_pd(vma, address, | |
758 | __hugepd(p4d_val(*p4d)), flags, | |
759 | P4D_SHIFT); | |
760 | if (page) | |
761 | return page; | |
762 | return no_page_table(vma, flags); | |
763 | } | |
df06b37f | 764 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
765 | } |
766 | ||
767 | /** | |
768 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
769 | * @vma: vm_area_struct mapping @address | |
770 | * @address: virtual address to look up | |
771 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
772 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
773 | * pointer to output page_mask | |
080dbb61 AK |
774 | * |
775 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
776 | * | |
78179556 MR |
777 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
778 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
779 | * | |
780 | * On output, the @ctx->page_mask is set according to the size of the page. | |
781 | * | |
782 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
783 | * an error pointer if there is a mapping to something not represented |
784 | * by a page descriptor (see also vm_normal_page()). | |
785 | */ | |
a7030aea | 786 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 787 | unsigned long address, unsigned int flags, |
df06b37f | 788 | struct follow_page_context *ctx) |
080dbb61 AK |
789 | { |
790 | pgd_t *pgd; | |
791 | struct page *page; | |
792 | struct mm_struct *mm = vma->vm_mm; | |
793 | ||
df06b37f | 794 | ctx->page_mask = 0; |
080dbb61 AK |
795 | |
796 | /* make this handle hugepd */ | |
797 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); | |
798 | if (!IS_ERR(page)) { | |
3faa52c0 | 799 | WARN_ON_ONCE(flags & (FOLL_GET | FOLL_PIN)); |
080dbb61 AK |
800 | return page; |
801 | } | |
802 | ||
803 | pgd = pgd_offset(mm, address); | |
804 | ||
805 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
806 | return no_page_table(vma, flags); | |
807 | ||
faaa5b62 AK |
808 | if (pgd_huge(*pgd)) { |
809 | page = follow_huge_pgd(mm, address, pgd, flags); | |
810 | if (page) | |
811 | return page; | |
812 | return no_page_table(vma, flags); | |
813 | } | |
4dc71451 AK |
814 | if (is_hugepd(__hugepd(pgd_val(*pgd)))) { |
815 | page = follow_huge_pd(vma, address, | |
816 | __hugepd(pgd_val(*pgd)), flags, | |
817 | PGDIR_SHIFT); | |
818 | if (page) | |
819 | return page; | |
820 | return no_page_table(vma, flags); | |
821 | } | |
faaa5b62 | 822 | |
df06b37f KB |
823 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
824 | } | |
825 | ||
826 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
827 | unsigned int foll_flags) | |
828 | { | |
829 | struct follow_page_context ctx = { NULL }; | |
830 | struct page *page; | |
831 | ||
1507f512 MR |
832 | if (vma_is_secretmem(vma)) |
833 | return NULL; | |
834 | ||
df06b37f KB |
835 | page = follow_page_mask(vma, address, foll_flags, &ctx); |
836 | if (ctx.pgmap) | |
837 | put_dev_pagemap(ctx.pgmap); | |
838 | return page; | |
080dbb61 AK |
839 | } |
840 | ||
f2b495ca KS |
841 | static int get_gate_page(struct mm_struct *mm, unsigned long address, |
842 | unsigned int gup_flags, struct vm_area_struct **vma, | |
843 | struct page **page) | |
844 | { | |
845 | pgd_t *pgd; | |
c2febafc | 846 | p4d_t *p4d; |
f2b495ca KS |
847 | pud_t *pud; |
848 | pmd_t *pmd; | |
849 | pte_t *pte; | |
850 | int ret = -EFAULT; | |
851 | ||
852 | /* user gate pages are read-only */ | |
853 | if (gup_flags & FOLL_WRITE) | |
854 | return -EFAULT; | |
855 | if (address > TASK_SIZE) | |
856 | pgd = pgd_offset_k(address); | |
857 | else | |
858 | pgd = pgd_offset_gate(mm, address); | |
b5d1c39f AL |
859 | if (pgd_none(*pgd)) |
860 | return -EFAULT; | |
c2febafc | 861 | p4d = p4d_offset(pgd, address); |
b5d1c39f AL |
862 | if (p4d_none(*p4d)) |
863 | return -EFAULT; | |
c2febafc | 864 | pud = pud_offset(p4d, address); |
b5d1c39f AL |
865 | if (pud_none(*pud)) |
866 | return -EFAULT; | |
f2b495ca | 867 | pmd = pmd_offset(pud, address); |
84c3fc4e | 868 | if (!pmd_present(*pmd)) |
f2b495ca KS |
869 | return -EFAULT; |
870 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
871 | pte = pte_offset_map(pmd, address); | |
872 | if (pte_none(*pte)) | |
873 | goto unmap; | |
874 | *vma = get_gate_vma(mm); | |
875 | if (!page) | |
876 | goto out; | |
877 | *page = vm_normal_page(*vma, address, *pte); | |
878 | if (!*page) { | |
879 | if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte))) | |
880 | goto unmap; | |
881 | *page = pte_page(*pte); | |
882 | } | |
9fa2dd94 | 883 | if (unlikely(!try_grab_page(*page, gup_flags))) { |
8fde12ca LT |
884 | ret = -ENOMEM; |
885 | goto unmap; | |
886 | } | |
f2b495ca KS |
887 | out: |
888 | ret = 0; | |
889 | unmap: | |
890 | pte_unmap(pte); | |
891 | return ret; | |
892 | } | |
893 | ||
9a95f3cf | 894 | /* |
c1e8d7c6 ML |
895 | * mmap_lock must be held on entry. If @locked != NULL and *@flags |
896 | * does not include FOLL_NOWAIT, the mmap_lock may be released. If it | |
4f6da934 | 897 | * is, *@locked will be set to 0 and -EBUSY returned. |
9a95f3cf | 898 | */ |
64019a2e | 899 | static int faultin_page(struct vm_area_struct *vma, |
4f6da934 | 900 | unsigned long address, unsigned int *flags, int *locked) |
16744483 | 901 | { |
16744483 | 902 | unsigned int fault_flags = 0; |
2b740303 | 903 | vm_fault_t ret; |
16744483 | 904 | |
55b8fe70 AG |
905 | if (*flags & FOLL_NOFAULT) |
906 | return -EFAULT; | |
16744483 KS |
907 | if (*flags & FOLL_WRITE) |
908 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
909 | if (*flags & FOLL_REMOTE) |
910 | fault_flags |= FAULT_FLAG_REMOTE; | |
4f6da934 | 911 | if (locked) |
71335f37 | 912 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
16744483 KS |
913 | if (*flags & FOLL_NOWAIT) |
914 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 915 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
916 | /* |
917 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
918 | * can co-exist | |
919 | */ | |
234b239b ALC |
920 | fault_flags |= FAULT_FLAG_TRIED; |
921 | } | |
16744483 | 922 | |
bce617ed | 923 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
16744483 | 924 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
925 | int err = vm_fault_to_errno(ret, *flags); |
926 | ||
927 | if (err) | |
928 | return err; | |
16744483 KS |
929 | BUG(); |
930 | } | |
931 | ||
16744483 | 932 | if (ret & VM_FAULT_RETRY) { |
4f6da934 PX |
933 | if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
934 | *locked = 0; | |
16744483 KS |
935 | return -EBUSY; |
936 | } | |
937 | ||
938 | /* | |
939 | * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when | |
940 | * necessary, even if maybe_mkwrite decided not to set pte_write. We | |
941 | * can thus safely do subsequent page lookups as if they were reads. | |
942 | * But only do so when looping for pte_write is futile: in some cases | |
943 | * userspace may also be wanting to write to the gotten user page, | |
944 | * which a read fault here might prevent (a readonly page might get | |
945 | * reCOWed by userspace write). | |
946 | */ | |
947 | if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE)) | |
2923117b | 948 | *flags |= FOLL_COW; |
16744483 KS |
949 | return 0; |
950 | } | |
951 | ||
fa5bb209 KS |
952 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
953 | { | |
954 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
955 | int write = (gup_flags & FOLL_WRITE); |
956 | int foreign = (gup_flags & FOLL_REMOTE); | |
fa5bb209 KS |
957 | |
958 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
959 | return -EFAULT; | |
960 | ||
7f7ccc2c WT |
961 | if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma)) |
962 | return -EFAULT; | |
963 | ||
52650c8b JG |
964 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
965 | return -EOPNOTSUPP; | |
966 | ||
1507f512 MR |
967 | if (vma_is_secretmem(vma)) |
968 | return -EFAULT; | |
969 | ||
1b2ee126 | 970 | if (write) { |
fa5bb209 KS |
971 | if (!(vm_flags & VM_WRITE)) { |
972 | if (!(gup_flags & FOLL_FORCE)) | |
973 | return -EFAULT; | |
974 | /* | |
975 | * We used to let the write,force case do COW in a | |
976 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
977 | * set a breakpoint in a read-only mapping of an | |
978 | * executable, without corrupting the file (yet only | |
979 | * when that file had been opened for writing!). | |
980 | * Anon pages in shared mappings are surprising: now | |
981 | * just reject it. | |
982 | */ | |
46435364 | 983 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 984 | return -EFAULT; |
fa5bb209 KS |
985 | } |
986 | } else if (!(vm_flags & VM_READ)) { | |
987 | if (!(gup_flags & FOLL_FORCE)) | |
988 | return -EFAULT; | |
989 | /* | |
990 | * Is there actually any vma we can reach here which does not | |
991 | * have VM_MAYREAD set? | |
992 | */ | |
993 | if (!(vm_flags & VM_MAYREAD)) | |
994 | return -EFAULT; | |
995 | } | |
d61172b4 DH |
996 | /* |
997 | * gups are always data accesses, not instruction | |
998 | * fetches, so execute=false here | |
999 | */ | |
1000 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 1001 | return -EFAULT; |
fa5bb209 KS |
1002 | return 0; |
1003 | } | |
1004 | ||
4bbd4c77 KS |
1005 | /** |
1006 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
1007 | * @mm: mm_struct of target mm |
1008 | * @start: starting user address | |
1009 | * @nr_pages: number of pages from start to pin | |
1010 | * @gup_flags: flags modifying pin behaviour | |
1011 | * @pages: array that receives pointers to the pages pinned. | |
1012 | * Should be at least nr_pages long. Or NULL, if caller | |
1013 | * only intends to ensure the pages are faulted in. | |
1014 | * @vmas: array of pointers to vmas corresponding to each page. | |
1015 | * Or NULL if the caller does not require them. | |
c1e8d7c6 | 1016 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 1017 | * |
d2dfbe47 LX |
1018 | * Returns either number of pages pinned (which may be less than the |
1019 | * number requested), or an error. Details about the return value: | |
1020 | * | |
1021 | * -- If nr_pages is 0, returns 0. | |
1022 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1023 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1024 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 1025 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
1026 | * |
1027 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1028 | * | |
c1e8d7c6 | 1029 | * @vmas are valid only as long as mmap_lock is held. |
4bbd4c77 | 1030 | * |
c1e8d7c6 | 1031 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
1032 | * |
1033 | * __get_user_pages walks a process's page tables and takes a reference to | |
1034 | * each struct page that each user address corresponds to at a given | |
1035 | * instant. That is, it takes the page that would be accessed if a user | |
1036 | * thread accesses the given user virtual address at that instant. | |
1037 | * | |
1038 | * This does not guarantee that the page exists in the user mappings when | |
1039 | * __get_user_pages returns, and there may even be a completely different | |
1040 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
1041 | * and subsequently re faulted). However it does guarantee that the page | |
1042 | * won't be freed completely. And mostly callers simply care that the page | |
1043 | * contains data that was valid *at some point in time*. Typically, an IO | |
1044 | * or similar operation cannot guarantee anything stronger anyway because | |
1045 | * locks can't be held over the syscall boundary. | |
1046 | * | |
1047 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
1048 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
1049 | * appropriate) must be called after the page is finished with, and | |
1050 | * before put_page is called. | |
1051 | * | |
c1e8d7c6 | 1052 | * If @locked != NULL, *@locked will be set to 0 when mmap_lock is |
4f6da934 PX |
1053 | * released by an up_read(). That can happen if @gup_flags does not |
1054 | * have FOLL_NOWAIT. | |
9a95f3cf | 1055 | * |
4f6da934 | 1056 | * A caller using such a combination of @locked and @gup_flags |
c1e8d7c6 | 1057 | * must therefore hold the mmap_lock for reading only, and recognize |
9a95f3cf PC |
1058 | * when it's been released. Otherwise, it must be held for either |
1059 | * reading or writing and will not be released. | |
4bbd4c77 KS |
1060 | * |
1061 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
1062 | * instead of __get_user_pages. __get_user_pages should be used only if | |
1063 | * you need some special @gup_flags. | |
1064 | */ | |
64019a2e | 1065 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
1066 | unsigned long start, unsigned long nr_pages, |
1067 | unsigned int gup_flags, struct page **pages, | |
4f6da934 | 1068 | struct vm_area_struct **vmas, int *locked) |
4bbd4c77 | 1069 | { |
df06b37f | 1070 | long ret = 0, i = 0; |
fa5bb209 | 1071 | struct vm_area_struct *vma = NULL; |
df06b37f | 1072 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
1073 | |
1074 | if (!nr_pages) | |
1075 | return 0; | |
1076 | ||
f9652594 AK |
1077 | start = untagged_addr(start); |
1078 | ||
eddb1c22 | 1079 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 KS |
1080 | |
1081 | /* | |
1082 | * If FOLL_FORCE is set then do not force a full fault as the hinting | |
1083 | * fault information is unrelated to the reference behaviour of a task | |
1084 | * using the address space | |
1085 | */ | |
1086 | if (!(gup_flags & FOLL_FORCE)) | |
1087 | gup_flags |= FOLL_NUMA; | |
1088 | ||
4bbd4c77 | 1089 | do { |
fa5bb209 KS |
1090 | struct page *page; |
1091 | unsigned int foll_flags = gup_flags; | |
1092 | unsigned int page_increm; | |
1093 | ||
1094 | /* first iteration or cross vma bound */ | |
1095 | if (!vma || start >= vma->vm_end) { | |
1096 | vma = find_extend_vma(mm, start); | |
1097 | if (!vma && in_gate_area(mm, start)) { | |
fa5bb209 KS |
1098 | ret = get_gate_page(mm, start & PAGE_MASK, |
1099 | gup_flags, &vma, | |
1100 | pages ? &pages[i] : NULL); | |
1101 | if (ret) | |
08be37b7 | 1102 | goto out; |
df06b37f | 1103 | ctx.page_mask = 0; |
fa5bb209 KS |
1104 | goto next_page; |
1105 | } | |
4bbd4c77 | 1106 | |
52650c8b | 1107 | if (!vma) { |
df06b37f KB |
1108 | ret = -EFAULT; |
1109 | goto out; | |
1110 | } | |
52650c8b JG |
1111 | ret = check_vma_flags(vma, gup_flags); |
1112 | if (ret) | |
1113 | goto out; | |
1114 | ||
fa5bb209 KS |
1115 | if (is_vm_hugetlb_page(vma)) { |
1116 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1117 | &start, &nr_pages, i, | |
a308c71b | 1118 | gup_flags, locked); |
ad415db8 PX |
1119 | if (locked && *locked == 0) { |
1120 | /* | |
1121 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1122 | * and we've lost mmap_lock. |
ad415db8 PX |
1123 | * We must stop here. |
1124 | */ | |
1125 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
ad415db8 PX |
1126 | goto out; |
1127 | } | |
fa5bb209 | 1128 | continue; |
4bbd4c77 | 1129 | } |
fa5bb209 KS |
1130 | } |
1131 | retry: | |
1132 | /* | |
1133 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1134 | * potentially allocating memory. | |
1135 | */ | |
fa45f116 | 1136 | if (fatal_signal_pending(current)) { |
d180870d | 1137 | ret = -EINTR; |
df06b37f KB |
1138 | goto out; |
1139 | } | |
fa5bb209 | 1140 | cond_resched(); |
df06b37f KB |
1141 | |
1142 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
fa5bb209 | 1143 | if (!page) { |
64019a2e | 1144 | ret = faultin_page(vma, start, &foll_flags, locked); |
fa5bb209 KS |
1145 | switch (ret) { |
1146 | case 0: | |
1147 | goto retry; | |
df06b37f KB |
1148 | case -EBUSY: |
1149 | ret = 0; | |
e4a9bc58 | 1150 | fallthrough; |
fa5bb209 KS |
1151 | case -EFAULT: |
1152 | case -ENOMEM: | |
1153 | case -EHWPOISON: | |
df06b37f | 1154 | goto out; |
4bbd4c77 | 1155 | } |
fa5bb209 | 1156 | BUG(); |
1027e443 KS |
1157 | } else if (PTR_ERR(page) == -EEXIST) { |
1158 | /* | |
1159 | * Proper page table entry exists, but no corresponding | |
1160 | * struct page. | |
1161 | */ | |
1162 | goto next_page; | |
1163 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1164 | ret = PTR_ERR(page); |
1165 | goto out; | |
1027e443 | 1166 | } |
fa5bb209 KS |
1167 | if (pages) { |
1168 | pages[i] = page; | |
1169 | flush_anon_page(vma, page, start); | |
1170 | flush_dcache_page(page); | |
df06b37f | 1171 | ctx.page_mask = 0; |
4bbd4c77 | 1172 | } |
4bbd4c77 | 1173 | next_page: |
fa5bb209 KS |
1174 | if (vmas) { |
1175 | vmas[i] = vma; | |
df06b37f | 1176 | ctx.page_mask = 0; |
fa5bb209 | 1177 | } |
df06b37f | 1178 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1179 | if (page_increm > nr_pages) |
1180 | page_increm = nr_pages; | |
1181 | i += page_increm; | |
1182 | start += page_increm * PAGE_SIZE; | |
1183 | nr_pages -= page_increm; | |
4bbd4c77 | 1184 | } while (nr_pages); |
df06b37f KB |
1185 | out: |
1186 | if (ctx.pgmap) | |
1187 | put_dev_pagemap(ctx.pgmap); | |
1188 | return i ? i : ret; | |
4bbd4c77 | 1189 | } |
4bbd4c77 | 1190 | |
771ab430 TK |
1191 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1192 | unsigned int fault_flags) | |
d4925e00 | 1193 | { |
1b2ee126 DH |
1194 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1195 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1196 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1197 | |
1198 | if (!(vm_flags & vma->vm_flags)) | |
1199 | return false; | |
1200 | ||
33a709b2 DH |
1201 | /* |
1202 | * The architecture might have a hardware protection | |
1b2ee126 | 1203 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1204 | * |
1205 | * gup always represents data access, not instruction | |
1206 | * fetches, so execute=false here: | |
33a709b2 | 1207 | */ |
d61172b4 | 1208 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1209 | return false; |
1210 | ||
d4925e00 DH |
1211 | return true; |
1212 | } | |
1213 | ||
adc8cb40 | 1214 | /** |
4bbd4c77 | 1215 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1216 | * @mm: mm_struct of target mm |
1217 | * @address: user address | |
1218 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1219 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1220 | * does not allow retry. If NULL, the caller must guarantee |
1221 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1222 | * |
1223 | * This is meant to be called in the specific scenario where for locking reasons | |
1224 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1225 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1226 | * trying again. | |
1227 | * | |
1228 | * Typically this is meant to be used by the futex code. | |
1229 | * | |
1230 | * The main difference with get_user_pages() is that this function will | |
1231 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1232 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1233 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1234 | * |
1235 | * This is important for some architectures where those bits also gate the | |
1236 | * access permission to the page because they are maintained in software. On | |
1237 | * such architectures, gup() will not be enough to make a subsequent access | |
1238 | * succeed. | |
1239 | * | |
c1e8d7c6 ML |
1240 | * This function will not return with an unlocked mmap_lock. So it has not the |
1241 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1242 | */ |
64019a2e | 1243 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1244 | unsigned long address, unsigned int fault_flags, |
1245 | bool *unlocked) | |
4bbd4c77 KS |
1246 | { |
1247 | struct vm_area_struct *vma; | |
8fed2f3c | 1248 | vm_fault_t ret; |
4a9e1cda | 1249 | |
f9652594 AK |
1250 | address = untagged_addr(address); |
1251 | ||
4a9e1cda | 1252 | if (unlocked) |
71335f37 | 1253 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1254 | |
4a9e1cda | 1255 | retry: |
4bbd4c77 KS |
1256 | vma = find_extend_vma(mm, address); |
1257 | if (!vma || address < vma->vm_start) | |
1258 | return -EFAULT; | |
1259 | ||
d4925e00 | 1260 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1261 | return -EFAULT; |
1262 | ||
475f4dfc PX |
1263 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1264 | fatal_signal_pending(current)) | |
1265 | return -EINTR; | |
1266 | ||
bce617ed | 1267 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
4bbd4c77 | 1268 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1269 | int err = vm_fault_to_errno(ret, 0); |
1270 | ||
1271 | if (err) | |
1272 | return err; | |
4bbd4c77 KS |
1273 | BUG(); |
1274 | } | |
4a9e1cda DD |
1275 | |
1276 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1277 | mmap_read_lock(mm); |
475f4dfc PX |
1278 | *unlocked = true; |
1279 | fault_flags |= FAULT_FLAG_TRIED; | |
1280 | goto retry; | |
4a9e1cda DD |
1281 | } |
1282 | ||
4bbd4c77 KS |
1283 | return 0; |
1284 | } | |
add6a0cd | 1285 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1286 | |
2d3a36a4 MH |
1287 | /* |
1288 | * Please note that this function, unlike __get_user_pages will not | |
1289 | * return 0 for nr_pages > 0 without FOLL_NOWAIT | |
1290 | */ | |
64019a2e | 1291 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1292 | unsigned long start, |
1293 | unsigned long nr_pages, | |
f0818f47 AA |
1294 | struct page **pages, |
1295 | struct vm_area_struct **vmas, | |
e716712f | 1296 | int *locked, |
0fd71a56 | 1297 | unsigned int flags) |
f0818f47 | 1298 | { |
f0818f47 AA |
1299 | long ret, pages_done; |
1300 | bool lock_dropped; | |
1301 | ||
1302 | if (locked) { | |
1303 | /* if VM_FAULT_RETRY can be returned, vmas become invalid */ | |
1304 | BUG_ON(vmas); | |
1305 | /* check caller initialized locked */ | |
1306 | BUG_ON(*locked != 1); | |
1307 | } | |
1308 | ||
a458b76a AA |
1309 | if (flags & FOLL_PIN) |
1310 | mm_set_has_pinned_flag(&mm->flags); | |
008cfe44 | 1311 | |
eddb1c22 JH |
1312 | /* |
1313 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1314 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1315 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1316 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1317 | * | |
1318 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1319 | * that here, as any failures will be obvious enough. | |
1320 | */ | |
1321 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1322 | flags |= FOLL_GET; |
f0818f47 AA |
1323 | |
1324 | pages_done = 0; | |
1325 | lock_dropped = false; | |
1326 | for (;;) { | |
64019a2e | 1327 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
f0818f47 AA |
1328 | vmas, locked); |
1329 | if (!locked) | |
1330 | /* VM_FAULT_RETRY couldn't trigger, bypass */ | |
1331 | return ret; | |
1332 | ||
1333 | /* VM_FAULT_RETRY cannot return errors */ | |
1334 | if (!*locked) { | |
1335 | BUG_ON(ret < 0); | |
1336 | BUG_ON(ret >= nr_pages); | |
1337 | } | |
1338 | ||
f0818f47 AA |
1339 | if (ret > 0) { |
1340 | nr_pages -= ret; | |
1341 | pages_done += ret; | |
1342 | if (!nr_pages) | |
1343 | break; | |
1344 | } | |
1345 | if (*locked) { | |
96312e61 AA |
1346 | /* |
1347 | * VM_FAULT_RETRY didn't trigger or it was a | |
1348 | * FOLL_NOWAIT. | |
1349 | */ | |
f0818f47 AA |
1350 | if (!pages_done) |
1351 | pages_done = ret; | |
1352 | break; | |
1353 | } | |
df17277b MR |
1354 | /* |
1355 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1356 | * For the prefault case (!pages) we only update counts. | |
1357 | */ | |
1358 | if (likely(pages)) | |
1359 | pages += ret; | |
f0818f47 | 1360 | start += ret << PAGE_SHIFT; |
4426e945 | 1361 | lock_dropped = true; |
f0818f47 | 1362 | |
4426e945 | 1363 | retry: |
f0818f47 AA |
1364 | /* |
1365 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1366 | * with both FAULT_FLAG_ALLOW_RETRY and |
1367 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
1368 | * by fatal signals, so we need to check it before we | |
1369 | * start trying again otherwise it can loop forever. | |
f0818f47 | 1370 | */ |
4426e945 | 1371 | |
ae46d2aa HD |
1372 | if (fatal_signal_pending(current)) { |
1373 | if (!pages_done) | |
1374 | pages_done = -EINTR; | |
4426e945 | 1375 | break; |
ae46d2aa | 1376 | } |
4426e945 | 1377 | |
d8ed45c5 | 1378 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1379 | if (ret) { |
1380 | BUG_ON(ret > 0); | |
1381 | if (!pages_done) | |
1382 | pages_done = ret; | |
1383 | break; | |
1384 | } | |
4426e945 | 1385 | |
c7b6a566 | 1386 | *locked = 1; |
64019a2e | 1387 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
4426e945 PX |
1388 | pages, NULL, locked); |
1389 | if (!*locked) { | |
1390 | /* Continue to retry until we succeeded */ | |
1391 | BUG_ON(ret != 0); | |
1392 | goto retry; | |
1393 | } | |
f0818f47 AA |
1394 | if (ret != 1) { |
1395 | BUG_ON(ret > 1); | |
1396 | if (!pages_done) | |
1397 | pages_done = ret; | |
1398 | break; | |
1399 | } | |
1400 | nr_pages--; | |
1401 | pages_done++; | |
1402 | if (!nr_pages) | |
1403 | break; | |
df17277b MR |
1404 | if (likely(pages)) |
1405 | pages++; | |
f0818f47 AA |
1406 | start += PAGE_SIZE; |
1407 | } | |
e716712f | 1408 | if (lock_dropped && *locked) { |
f0818f47 AA |
1409 | /* |
1410 | * We must let the caller know we temporarily dropped the lock | |
1411 | * and so the critical section protected by it was lost. | |
1412 | */ | |
d8ed45c5 | 1413 | mmap_read_unlock(mm); |
f0818f47 AA |
1414 | *locked = 0; |
1415 | } | |
1416 | return pages_done; | |
1417 | } | |
1418 | ||
d3649f68 CH |
1419 | /** |
1420 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1421 | * @vma: target vma | |
1422 | * @start: start address | |
1423 | * @end: end address | |
c1e8d7c6 | 1424 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1425 | * |
1426 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1427 | * | |
0a36f7f8 TY |
1428 | * Return either number of pages pinned in the vma, or a negative error |
1429 | * code on error. | |
d3649f68 | 1430 | * |
c1e8d7c6 | 1431 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1432 | * |
4f6da934 | 1433 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1434 | * be unperturbed. |
1435 | * | |
4f6da934 PX |
1436 | * If @locked is non-NULL, it must held for read only and may be |
1437 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1438 | */ |
1439 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1440 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1441 | { |
1442 | struct mm_struct *mm = vma->vm_mm; | |
1443 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1444 | int gup_flags; | |
1445 | ||
be51eb18 ML |
1446 | VM_BUG_ON(!PAGE_ALIGNED(start)); |
1447 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
d3649f68 CH |
1448 | VM_BUG_ON_VMA(start < vma->vm_start, vma); |
1449 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1450 | mmap_assert_locked(mm); |
d3649f68 | 1451 | |
b67bf49c HD |
1452 | /* |
1453 | * Rightly or wrongly, the VM_LOCKONFAULT case has never used | |
1454 | * faultin_page() to break COW, so it has no work to do here. | |
1455 | */ | |
d3649f68 | 1456 | if (vma->vm_flags & VM_LOCKONFAULT) |
b67bf49c HD |
1457 | return nr_pages; |
1458 | ||
1459 | gup_flags = FOLL_TOUCH; | |
d3649f68 CH |
1460 | /* |
1461 | * We want to touch writable mappings with a write fault in order | |
1462 | * to break COW, except for shared mappings because these don't COW | |
1463 | * and we would not want to dirty them for nothing. | |
1464 | */ | |
1465 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1466 | gup_flags |= FOLL_WRITE; | |
1467 | ||
1468 | /* | |
1469 | * We want mlock to succeed for regions that have any permissions | |
1470 | * other than PROT_NONE. | |
1471 | */ | |
3122e80e | 1472 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1473 | gup_flags |= FOLL_FORCE; |
1474 | ||
1475 | /* | |
1476 | * We made sure addr is within a VMA, so the following will | |
1477 | * not result in a stack expansion that recurses back here. | |
1478 | */ | |
64019a2e | 1479 | return __get_user_pages(mm, start, nr_pages, gup_flags, |
4f6da934 | 1480 | NULL, NULL, locked); |
d3649f68 CH |
1481 | } |
1482 | ||
4ca9b385 DH |
1483 | /* |
1484 | * faultin_vma_page_range() - populate (prefault) page tables inside the | |
1485 | * given VMA range readable/writable | |
1486 | * | |
1487 | * This takes care of mlocking the pages, too, if VM_LOCKED is set. | |
1488 | * | |
1489 | * @vma: target vma | |
1490 | * @start: start address | |
1491 | * @end: end address | |
1492 | * @write: whether to prefault readable or writable | |
1493 | * @locked: whether the mmap_lock is still held | |
1494 | * | |
1495 | * Returns either number of processed pages in the vma, or a negative error | |
1496 | * code on error (see __get_user_pages()). | |
1497 | * | |
1498 | * vma->vm_mm->mmap_lock must be held. The range must be page-aligned and | |
1499 | * covered by the VMA. | |
1500 | * | |
1501 | * If @locked is NULL, it may be held for read or write and will be unperturbed. | |
1502 | * | |
1503 | * If @locked is non-NULL, it must held for read only and may be released. If | |
1504 | * it's released, *@locked will be set to 0. | |
1505 | */ | |
1506 | long faultin_vma_page_range(struct vm_area_struct *vma, unsigned long start, | |
1507 | unsigned long end, bool write, int *locked) | |
1508 | { | |
1509 | struct mm_struct *mm = vma->vm_mm; | |
1510 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1511 | int gup_flags; | |
1512 | ||
1513 | VM_BUG_ON(!PAGE_ALIGNED(start)); | |
1514 | VM_BUG_ON(!PAGE_ALIGNED(end)); | |
1515 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1516 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
1517 | mmap_assert_locked(mm); | |
1518 | ||
1519 | /* | |
1520 | * FOLL_TOUCH: Mark page accessed and thereby young; will also mark | |
1521 | * the page dirty with FOLL_WRITE -- which doesn't make a | |
1522 | * difference with !FOLL_FORCE, because the page is writable | |
1523 | * in the page table. | |
1524 | * FOLL_HWPOISON: Return -EHWPOISON instead of -EFAULT when we hit | |
1525 | * a poisoned page. | |
4ca9b385 DH |
1526 | * !FOLL_FORCE: Require proper access permissions. |
1527 | */ | |
b67bf49c | 1528 | gup_flags = FOLL_TOUCH | FOLL_HWPOISON; |
4ca9b385 DH |
1529 | if (write) |
1530 | gup_flags |= FOLL_WRITE; | |
1531 | ||
1532 | /* | |
eb2faa51 DH |
1533 | * We want to report -EINVAL instead of -EFAULT for any permission |
1534 | * problems or incompatible mappings. | |
4ca9b385 | 1535 | */ |
eb2faa51 DH |
1536 | if (check_vma_flags(vma, gup_flags)) |
1537 | return -EINVAL; | |
1538 | ||
4ca9b385 DH |
1539 | return __get_user_pages(mm, start, nr_pages, gup_flags, |
1540 | NULL, NULL, locked); | |
1541 | } | |
1542 | ||
d3649f68 CH |
1543 | /* |
1544 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1545 | * | |
1546 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1547 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1548 | * mmap_lock must not be held. |
d3649f68 CH |
1549 | */ |
1550 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1551 | { | |
1552 | struct mm_struct *mm = current->mm; | |
1553 | unsigned long end, nstart, nend; | |
1554 | struct vm_area_struct *vma = NULL; | |
1555 | int locked = 0; | |
1556 | long ret = 0; | |
1557 | ||
1558 | end = start + len; | |
1559 | ||
1560 | for (nstart = start; nstart < end; nstart = nend) { | |
1561 | /* | |
1562 | * We want to fault in pages for [nstart; end) address range. | |
1563 | * Find first corresponding VMA. | |
1564 | */ | |
1565 | if (!locked) { | |
1566 | locked = 1; | |
d8ed45c5 | 1567 | mmap_read_lock(mm); |
d3649f68 CH |
1568 | vma = find_vma(mm, nstart); |
1569 | } else if (nstart >= vma->vm_end) | |
1570 | vma = vma->vm_next; | |
1571 | if (!vma || vma->vm_start >= end) | |
1572 | break; | |
1573 | /* | |
1574 | * Set [nstart; nend) to intersection of desired address | |
1575 | * range with the first VMA. Also, skip undesirable VMA types. | |
1576 | */ | |
1577 | nend = min(end, vma->vm_end); | |
1578 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1579 | continue; | |
1580 | if (nstart < vma->vm_start) | |
1581 | nstart = vma->vm_start; | |
1582 | /* | |
1583 | * Now fault in a range of pages. populate_vma_page_range() | |
1584 | * double checks the vma flags, so that it won't mlock pages | |
1585 | * if the vma was already munlocked. | |
1586 | */ | |
1587 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1588 | if (ret < 0) { | |
1589 | if (ignore_errors) { | |
1590 | ret = 0; | |
1591 | continue; /* continue at next VMA */ | |
1592 | } | |
1593 | break; | |
1594 | } | |
1595 | nend = nstart + ret * PAGE_SIZE; | |
1596 | ret = 0; | |
1597 | } | |
1598 | if (locked) | |
d8ed45c5 | 1599 | mmap_read_unlock(mm); |
d3649f68 CH |
1600 | return ret; /* 0 or negative error code */ |
1601 | } | |
050a9adc | 1602 | #else /* CONFIG_MMU */ |
64019a2e | 1603 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc CH |
1604 | unsigned long nr_pages, struct page **pages, |
1605 | struct vm_area_struct **vmas, int *locked, | |
1606 | unsigned int foll_flags) | |
1607 | { | |
1608 | struct vm_area_struct *vma; | |
1609 | unsigned long vm_flags; | |
24dc20c7 | 1610 | long i; |
050a9adc CH |
1611 | |
1612 | /* calculate required read or write permissions. | |
1613 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1614 | */ | |
1615 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1616 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1617 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1618 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1619 | ||
1620 | for (i = 0; i < nr_pages; i++) { | |
1621 | vma = find_vma(mm, start); | |
1622 | if (!vma) | |
1623 | goto finish_or_fault; | |
1624 | ||
1625 | /* protect what we can, including chardevs */ | |
1626 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1627 | !(vm_flags & vma->vm_flags)) | |
1628 | goto finish_or_fault; | |
1629 | ||
1630 | if (pages) { | |
1631 | pages[i] = virt_to_page(start); | |
1632 | if (pages[i]) | |
1633 | get_page(pages[i]); | |
1634 | } | |
1635 | if (vmas) | |
1636 | vmas[i] = vma; | |
1637 | start = (start + PAGE_SIZE) & PAGE_MASK; | |
1638 | } | |
1639 | ||
1640 | return i; | |
1641 | ||
1642 | finish_or_fault: | |
1643 | return i ? : -EFAULT; | |
1644 | } | |
1645 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1646 | |
bb523b40 AG |
1647 | /** |
1648 | * fault_in_writeable - fault in userspace address range for writing | |
1649 | * @uaddr: start of address range | |
1650 | * @size: size of address range | |
1651 | * | |
1652 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1653 | * copy_from_user()). | |
1654 | */ | |
1655 | size_t fault_in_writeable(char __user *uaddr, size_t size) | |
1656 | { | |
1657 | char __user *start = uaddr, *end; | |
1658 | ||
1659 | if (unlikely(size == 0)) | |
1660 | return 0; | |
677b2a8c CL |
1661 | if (!user_write_access_begin(uaddr, size)) |
1662 | return size; | |
bb523b40 | 1663 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1664 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1665 | uaddr = (char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1666 | } | |
1667 | end = (char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1668 | if (unlikely(end < start)) | |
1669 | end = NULL; | |
1670 | while (uaddr != end) { | |
677b2a8c | 1671 | unsafe_put_user(0, uaddr, out); |
bb523b40 AG |
1672 | uaddr += PAGE_SIZE; |
1673 | } | |
1674 | ||
1675 | out: | |
677b2a8c | 1676 | user_write_access_end(); |
bb523b40 AG |
1677 | if (size > uaddr - start) |
1678 | return size - (uaddr - start); | |
1679 | return 0; | |
1680 | } | |
1681 | EXPORT_SYMBOL(fault_in_writeable); | |
1682 | ||
cdd591fc AG |
1683 | /* |
1684 | * fault_in_safe_writeable - fault in an address range for writing | |
1685 | * @uaddr: start of address range | |
1686 | * @size: length of address range | |
1687 | * | |
1688 | * Faults in an address range using get_user_pages, i.e., without triggering | |
1689 | * hardware page faults. This is primarily useful when we already know that | |
1690 | * some or all of the pages in the address range aren't in memory. | |
1691 | * | |
1692 | * Other than fault_in_writeable(), this function is non-destructive. | |
1693 | * | |
1694 | * Note that we don't pin or otherwise hold the pages referenced that we fault | |
1695 | * in. There's no guarantee that they'll stay in memory for any duration of | |
1696 | * time. | |
1697 | * | |
1698 | * Returns the number of bytes not faulted in, like copy_to_user() and | |
1699 | * copy_from_user(). | |
1700 | */ | |
1701 | size_t fault_in_safe_writeable(const char __user *uaddr, size_t size) | |
1702 | { | |
1703 | unsigned long start = (unsigned long)untagged_addr(uaddr); | |
1704 | unsigned long end, nstart, nend; | |
1705 | struct mm_struct *mm = current->mm; | |
1706 | struct vm_area_struct *vma = NULL; | |
1707 | int locked = 0; | |
1708 | ||
1709 | nstart = start & PAGE_MASK; | |
1710 | end = PAGE_ALIGN(start + size); | |
1711 | if (end < nstart) | |
1712 | end = 0; | |
1713 | for (; nstart != end; nstart = nend) { | |
1714 | unsigned long nr_pages; | |
1715 | long ret; | |
1716 | ||
1717 | if (!locked) { | |
1718 | locked = 1; | |
1719 | mmap_read_lock(mm); | |
1720 | vma = find_vma(mm, nstart); | |
1721 | } else if (nstart >= vma->vm_end) | |
1722 | vma = vma->vm_next; | |
1723 | if (!vma || vma->vm_start >= end) | |
1724 | break; | |
1725 | nend = end ? min(end, vma->vm_end) : vma->vm_end; | |
1726 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1727 | continue; | |
1728 | if (nstart < vma->vm_start) | |
1729 | nstart = vma->vm_start; | |
1730 | nr_pages = (nend - nstart) / PAGE_SIZE; | |
1731 | ret = __get_user_pages_locked(mm, nstart, nr_pages, | |
1732 | NULL, NULL, &locked, | |
1733 | FOLL_TOUCH | FOLL_WRITE); | |
1734 | if (ret <= 0) | |
1735 | break; | |
1736 | nend = nstart + ret * PAGE_SIZE; | |
1737 | } | |
1738 | if (locked) | |
1739 | mmap_read_unlock(mm); | |
1740 | if (nstart == end) | |
1741 | return 0; | |
1742 | return size - min_t(size_t, nstart - start, size); | |
1743 | } | |
1744 | EXPORT_SYMBOL(fault_in_safe_writeable); | |
1745 | ||
bb523b40 AG |
1746 | /** |
1747 | * fault_in_readable - fault in userspace address range for reading | |
1748 | * @uaddr: start of user address range | |
1749 | * @size: size of user address range | |
1750 | * | |
1751 | * Returns the number of bytes not faulted in (like copy_to_user() and | |
1752 | * copy_from_user()). | |
1753 | */ | |
1754 | size_t fault_in_readable(const char __user *uaddr, size_t size) | |
1755 | { | |
1756 | const char __user *start = uaddr, *end; | |
1757 | volatile char c; | |
1758 | ||
1759 | if (unlikely(size == 0)) | |
1760 | return 0; | |
677b2a8c CL |
1761 | if (!user_read_access_begin(uaddr, size)) |
1762 | return size; | |
bb523b40 | 1763 | if (!PAGE_ALIGNED(uaddr)) { |
677b2a8c | 1764 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1765 | uaddr = (const char __user *)PAGE_ALIGN((unsigned long)uaddr); |
1766 | } | |
1767 | end = (const char __user *)PAGE_ALIGN((unsigned long)start + size); | |
1768 | if (unlikely(end < start)) | |
1769 | end = NULL; | |
1770 | while (uaddr != end) { | |
677b2a8c | 1771 | unsafe_get_user(c, uaddr, out); |
bb523b40 AG |
1772 | uaddr += PAGE_SIZE; |
1773 | } | |
1774 | ||
1775 | out: | |
677b2a8c | 1776 | user_read_access_end(); |
bb523b40 AG |
1777 | (void)c; |
1778 | if (size > uaddr - start) | |
1779 | return size - (uaddr - start); | |
1780 | return 0; | |
1781 | } | |
1782 | EXPORT_SYMBOL(fault_in_readable); | |
1783 | ||
8f942eea JH |
1784 | /** |
1785 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1786 | * @addr: user address | |
1787 | * | |
1788 | * Returns struct page pointer of user page pinned for dump, | |
1789 | * to be freed afterwards by put_page(). | |
1790 | * | |
1791 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1792 | * the corefile, to preserve alignment with its headers; and also returns | |
1793 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
f0953a1b | 1794 | * allowing a hole to be left in the corefile to save disk space. |
8f942eea | 1795 | * |
7f3bfab5 | 1796 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1797 | */ |
1798 | #ifdef CONFIG_ELF_CORE | |
1799 | struct page *get_dump_page(unsigned long addr) | |
1800 | { | |
7f3bfab5 | 1801 | struct mm_struct *mm = current->mm; |
8f942eea | 1802 | struct page *page; |
7f3bfab5 JH |
1803 | int locked = 1; |
1804 | int ret; | |
8f942eea | 1805 | |
7f3bfab5 | 1806 | if (mmap_read_lock_killable(mm)) |
8f942eea | 1807 | return NULL; |
7f3bfab5 JH |
1808 | ret = __get_user_pages_locked(mm, addr, 1, &page, NULL, &locked, |
1809 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); | |
1810 | if (locked) | |
1811 | mmap_read_unlock(mm); | |
1812 | return (ret == 1) ? page : NULL; | |
8f942eea JH |
1813 | } |
1814 | #endif /* CONFIG_ELF_CORE */ | |
1815 | ||
d1e153fe | 1816 | #ifdef CONFIG_MIGRATION |
f68749ec PT |
1817 | /* |
1818 | * Check whether all pages are pinnable, if so return number of pages. If some | |
1819 | * pages are not pinnable, migrate them, and unpin all pages. Return zero if | |
1820 | * pages were migrated, or if some pages were not successfully isolated. | |
1821 | * Return negative error if migration fails. | |
1822 | */ | |
1823 | static long check_and_migrate_movable_pages(unsigned long nr_pages, | |
d1e153fe | 1824 | struct page **pages, |
d1e153fe | 1825 | unsigned int gup_flags) |
9a4e9f3b | 1826 | { |
f9f38f78 | 1827 | unsigned long isolation_error_count = 0, i; |
f68749ec | 1828 | struct page *prev_head = NULL; |
f9f38f78 CH |
1829 | LIST_HEAD(movable_page_list); |
1830 | bool drain_allow = true; | |
1831 | int ret = 0; | |
9a4e9f3b | 1832 | |
83c02c23 | 1833 | for (i = 0; i < nr_pages; i++) { |
f9f38f78 CH |
1834 | struct page *head = compound_head(pages[i]); |
1835 | ||
83c02c23 PT |
1836 | if (head == prev_head) |
1837 | continue; | |
1838 | prev_head = head; | |
f9f38f78 CH |
1839 | |
1840 | if (is_pinnable_page(head)) | |
1841 | continue; | |
1842 | ||
9a4e9f3b | 1843 | /* |
f9f38f78 | 1844 | * Try to move out any movable page before pinning the range. |
9a4e9f3b | 1845 | */ |
f9f38f78 CH |
1846 | if (PageHuge(head)) { |
1847 | if (!isolate_huge_page(head, &movable_page_list)) | |
1848 | isolation_error_count++; | |
1849 | continue; | |
1850 | } | |
9a4e9f3b | 1851 | |
f9f38f78 CH |
1852 | if (!PageLRU(head) && drain_allow) { |
1853 | lru_add_drain_all(); | |
1854 | drain_allow = false; | |
1855 | } | |
1856 | ||
1857 | if (isolate_lru_page(head)) { | |
1858 | isolation_error_count++; | |
1859 | continue; | |
9a4e9f3b | 1860 | } |
f9f38f78 CH |
1861 | list_add_tail(&head->lru, &movable_page_list); |
1862 | mod_node_page_state(page_pgdat(head), | |
1863 | NR_ISOLATED_ANON + page_is_file_lru(head), | |
1864 | thp_nr_pages(head)); | |
9a4e9f3b AK |
1865 | } |
1866 | ||
f9f38f78 CH |
1867 | if (!list_empty(&movable_page_list) || isolation_error_count) |
1868 | goto unpin_pages; | |
1869 | ||
6e7f34eb PT |
1870 | /* |
1871 | * If list is empty, and no isolation errors, means that all pages are | |
1872 | * in the correct zone. | |
1873 | */ | |
f9f38f78 | 1874 | return nr_pages; |
6e7f34eb | 1875 | |
f9f38f78 | 1876 | unpin_pages: |
f68749ec PT |
1877 | if (gup_flags & FOLL_PIN) { |
1878 | unpin_user_pages(pages, nr_pages); | |
1879 | } else { | |
1880 | for (i = 0; i < nr_pages; i++) | |
1881 | put_page(pages[i]); | |
1882 | } | |
f9f38f78 | 1883 | |
d1e153fe | 1884 | if (!list_empty(&movable_page_list)) { |
f9f38f78 CH |
1885 | struct migration_target_control mtc = { |
1886 | .nid = NUMA_NO_NODE, | |
1887 | .gfp_mask = GFP_USER | __GFP_NOWARN, | |
1888 | }; | |
1889 | ||
d1e153fe | 1890 | ret = migrate_pages(&movable_page_list, alloc_migration_target, |
f0f44638 | 1891 | NULL, (unsigned long)&mtc, MIGRATE_SYNC, |
5ac95884 | 1892 | MR_LONGTERM_PIN, NULL); |
f9f38f78 CH |
1893 | if (ret > 0) /* number of pages not migrated */ |
1894 | ret = -ENOMEM; | |
9a4e9f3b AK |
1895 | } |
1896 | ||
f9f38f78 CH |
1897 | if (ret && !list_empty(&movable_page_list)) |
1898 | putback_movable_pages(&movable_page_list); | |
1899 | return ret; | |
9a4e9f3b AK |
1900 | } |
1901 | #else | |
f68749ec | 1902 | static long check_and_migrate_movable_pages(unsigned long nr_pages, |
d1e153fe | 1903 | struct page **pages, |
d1e153fe | 1904 | unsigned int gup_flags) |
9a4e9f3b AK |
1905 | { |
1906 | return nr_pages; | |
1907 | } | |
d1e153fe | 1908 | #endif /* CONFIG_MIGRATION */ |
9a4e9f3b | 1909 | |
2bb6d283 | 1910 | /* |
932f4a63 IW |
1911 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
1912 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 1913 | */ |
64019a2e | 1914 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
1915 | unsigned long start, |
1916 | unsigned long nr_pages, | |
1917 | struct page **pages, | |
1918 | struct vm_area_struct **vmas, | |
1919 | unsigned int gup_flags) | |
2bb6d283 | 1920 | { |
f68749ec | 1921 | unsigned int flags; |
52650c8b | 1922 | long rc; |
2bb6d283 | 1923 | |
f68749ec PT |
1924 | if (!(gup_flags & FOLL_LONGTERM)) |
1925 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
1926 | NULL, gup_flags); | |
1927 | flags = memalloc_pin_save(); | |
1928 | do { | |
1929 | rc = __get_user_pages_locked(mm, start, nr_pages, pages, vmas, | |
1930 | NULL, gup_flags); | |
1931 | if (rc <= 0) | |
1932 | break; | |
1933 | rc = check_and_migrate_movable_pages(rc, pages, gup_flags); | |
1934 | } while (!rc); | |
1935 | memalloc_pin_restore(flags); | |
2bb6d283 | 1936 | |
2bb6d283 DW |
1937 | return rc; |
1938 | } | |
932f4a63 | 1939 | |
447f3e45 BS |
1940 | static bool is_valid_gup_flags(unsigned int gup_flags) |
1941 | { | |
1942 | /* | |
1943 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
1944 | * never directly by the caller, so enforce that with an assertion: | |
1945 | */ | |
1946 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
1947 | return false; | |
1948 | /* | |
1949 | * FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying | |
1950 | * that is, FOLL_LONGTERM is a specific case, more restrictive case of | |
1951 | * FOLL_PIN. | |
1952 | */ | |
1953 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
1954 | return false; | |
1955 | ||
1956 | return true; | |
1957 | } | |
1958 | ||
22bf29b6 | 1959 | #ifdef CONFIG_MMU |
64019a2e | 1960 | static long __get_user_pages_remote(struct mm_struct *mm, |
22bf29b6 JH |
1961 | unsigned long start, unsigned long nr_pages, |
1962 | unsigned int gup_flags, struct page **pages, | |
1963 | struct vm_area_struct **vmas, int *locked) | |
1964 | { | |
1965 | /* | |
1966 | * Parts of FOLL_LONGTERM behavior are incompatible with | |
1967 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1968 | * vmas. However, this only comes up if locked is set, and there are | |
1969 | * callers that do request FOLL_LONGTERM, but do not set locked. So, | |
1970 | * allow what we can. | |
1971 | */ | |
1972 | if (gup_flags & FOLL_LONGTERM) { | |
1973 | if (WARN_ON_ONCE(locked)) | |
1974 | return -EINVAL; | |
1975 | /* | |
1976 | * This will check the vmas (even if our vmas arg is NULL) | |
1977 | * and return -ENOTSUPP if DAX isn't allowed in this case: | |
1978 | */ | |
64019a2e | 1979 | return __gup_longterm_locked(mm, start, nr_pages, pages, |
22bf29b6 JH |
1980 | vmas, gup_flags | FOLL_TOUCH | |
1981 | FOLL_REMOTE); | |
1982 | } | |
1983 | ||
64019a2e | 1984 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, |
22bf29b6 JH |
1985 | locked, |
1986 | gup_flags | FOLL_TOUCH | FOLL_REMOTE); | |
1987 | } | |
1988 | ||
adc8cb40 | 1989 | /** |
c4237f8b | 1990 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
1991 | * @mm: mm_struct of target mm |
1992 | * @start: starting user address | |
1993 | * @nr_pages: number of pages from start to pin | |
1994 | * @gup_flags: flags modifying lookup behaviour | |
1995 | * @pages: array that receives pointers to the pages pinned. | |
1996 | * Should be at least nr_pages long. Or NULL, if caller | |
1997 | * only intends to ensure the pages are faulted in. | |
1998 | * @vmas: array of pointers to vmas corresponding to each page. | |
1999 | * Or NULL if the caller does not require them. | |
2000 | * @locked: pointer to lock flag indicating whether lock is held and | |
2001 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2002 | * utilised. Lock must initially be held. | |
2003 | * | |
2004 | * Returns either number of pages pinned (which may be less than the | |
2005 | * number requested), or an error. Details about the return value: | |
2006 | * | |
2007 | * -- If nr_pages is 0, returns 0. | |
2008 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
2009 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
2010 | * pages pinned. Again, this may be less than nr_pages. | |
2011 | * | |
2012 | * The caller is responsible for releasing returned @pages, via put_page(). | |
2013 | * | |
c1e8d7c6 | 2014 | * @vmas are valid only as long as mmap_lock is held. |
c4237f8b | 2015 | * |
c1e8d7c6 | 2016 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 2017 | * |
adc8cb40 SJ |
2018 | * get_user_pages_remote walks a process's page tables and takes a reference |
2019 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
2020 | * instant. That is, it takes the page that would be accessed if a user |
2021 | * thread accesses the given user virtual address at that instant. | |
2022 | * | |
2023 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 2024 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b JH |
2025 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
2026 | * and subsequently re faulted). However it does guarantee that the page | |
2027 | * won't be freed completely. And mostly callers simply care that the page | |
2028 | * contains data that was valid *at some point in time*. Typically, an IO | |
2029 | * or similar operation cannot guarantee anything stronger anyway because | |
2030 | * locks can't be held over the syscall boundary. | |
2031 | * | |
2032 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
2033 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
2034 | * be called after the page is finished with, and before put_page is called. | |
2035 | * | |
adc8cb40 SJ |
2036 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
2037 | * to get a handle on the memory by some means other than accesses | |
2038 | * via the user virtual addresses. The pages may be submitted for | |
2039 | * DMA to devices or accessed via their kernel linear mapping (via the | |
2040 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
2041 | * |
2042 | * See also get_user_pages_fast, for performance critical applications. | |
2043 | * | |
adc8cb40 | 2044 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 2045 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 2046 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
2047 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
2048 | */ | |
64019a2e | 2049 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
2050 | unsigned long start, unsigned long nr_pages, |
2051 | unsigned int gup_flags, struct page **pages, | |
2052 | struct vm_area_struct **vmas, int *locked) | |
2053 | { | |
447f3e45 | 2054 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2055 | return -EINVAL; |
2056 | ||
64019a2e | 2057 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
22bf29b6 | 2058 | pages, vmas, locked); |
c4237f8b JH |
2059 | } |
2060 | EXPORT_SYMBOL(get_user_pages_remote); | |
2061 | ||
eddb1c22 | 2062 | #else /* CONFIG_MMU */ |
64019a2e | 2063 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2064 | unsigned long start, unsigned long nr_pages, |
2065 | unsigned int gup_flags, struct page **pages, | |
2066 | struct vm_area_struct **vmas, int *locked) | |
2067 | { | |
2068 | return 0; | |
2069 | } | |
3faa52c0 | 2070 | |
64019a2e | 2071 | static long __get_user_pages_remote(struct mm_struct *mm, |
3faa52c0 JH |
2072 | unsigned long start, unsigned long nr_pages, |
2073 | unsigned int gup_flags, struct page **pages, | |
2074 | struct vm_area_struct **vmas, int *locked) | |
2075 | { | |
2076 | return 0; | |
2077 | } | |
eddb1c22 JH |
2078 | #endif /* !CONFIG_MMU */ |
2079 | ||
adc8cb40 SJ |
2080 | /** |
2081 | * get_user_pages() - pin user pages in memory | |
2082 | * @start: starting user address | |
2083 | * @nr_pages: number of pages from start to pin | |
2084 | * @gup_flags: flags modifying lookup behaviour | |
2085 | * @pages: array that receives pointers to the pages pinned. | |
2086 | * Should be at least nr_pages long. Or NULL, if caller | |
2087 | * only intends to ensure the pages are faulted in. | |
2088 | * @vmas: array of pointers to vmas corresponding to each page. | |
2089 | * Or NULL if the caller does not require them. | |
2090 | * | |
64019a2e PX |
2091 | * This is the same as get_user_pages_remote(), just with a less-flexible |
2092 | * calling convention where we assume that the mm being operated on belongs to | |
2093 | * the current task, and doesn't allow passing of a locked parameter. We also | |
2094 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
2095 | */ |
2096 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
2097 | unsigned int gup_flags, struct page **pages, | |
2098 | struct vm_area_struct **vmas) | |
2099 | { | |
447f3e45 | 2100 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2101 | return -EINVAL; |
2102 | ||
64019a2e | 2103 | return __gup_longterm_locked(current->mm, start, nr_pages, |
932f4a63 IW |
2104 | pages, vmas, gup_flags | FOLL_TOUCH); |
2105 | } | |
2106 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 2107 | |
adc8cb40 | 2108 | /** |
a00cda3f MCC |
2109 | * get_user_pages_locked() - variant of get_user_pages() |
2110 | * | |
2111 | * @start: starting user address | |
2112 | * @nr_pages: number of pages from start to pin | |
2113 | * @gup_flags: flags modifying lookup behaviour | |
2114 | * @pages: array that receives pointers to the pages pinned. | |
2115 | * Should be at least nr_pages long. Or NULL, if caller | |
2116 | * only intends to ensure the pages are faulted in. | |
2117 | * @locked: pointer to lock flag indicating whether lock is held and | |
2118 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2119 | * utilised. Lock must initially be held. | |
2120 | * | |
2121 | * It is suitable to replace the form: | |
acc3c8d1 | 2122 | * |
3e4e28c5 | 2123 | * mmap_read_lock(mm); |
d3649f68 | 2124 | * do_something() |
64019a2e | 2125 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2126 | * mmap_read_unlock(mm); |
acc3c8d1 | 2127 | * |
d3649f68 | 2128 | * to: |
acc3c8d1 | 2129 | * |
d3649f68 | 2130 | * int locked = 1; |
3e4e28c5 | 2131 | * mmap_read_lock(mm); |
d3649f68 | 2132 | * do_something() |
64019a2e | 2133 | * get_user_pages_locked(mm, ..., pages, &locked); |
d3649f68 | 2134 | * if (locked) |
3e4e28c5 | 2135 | * mmap_read_unlock(mm); |
adc8cb40 | 2136 | * |
adc8cb40 SJ |
2137 | * We can leverage the VM_FAULT_RETRY functionality in the page fault |
2138 | * paths better by using either get_user_pages_locked() or | |
2139 | * get_user_pages_unlocked(). | |
2140 | * | |
acc3c8d1 | 2141 | */ |
d3649f68 CH |
2142 | long get_user_pages_locked(unsigned long start, unsigned long nr_pages, |
2143 | unsigned int gup_flags, struct page **pages, | |
2144 | int *locked) | |
acc3c8d1 | 2145 | { |
acc3c8d1 | 2146 | /* |
d3649f68 CH |
2147 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with |
2148 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2149 | * vmas. As there are no users of this flag in this call we simply | |
2150 | * disallow this option for now. | |
acc3c8d1 | 2151 | */ |
d3649f68 CH |
2152 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) |
2153 | return -EINVAL; | |
420c2091 JH |
2154 | /* |
2155 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
2156 | * never directly by the caller, so enforce that: | |
2157 | */ | |
2158 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
2159 | return -EINVAL; | |
acc3c8d1 | 2160 | |
64019a2e | 2161 | return __get_user_pages_locked(current->mm, start, nr_pages, |
d3649f68 CH |
2162 | pages, NULL, locked, |
2163 | gup_flags | FOLL_TOUCH); | |
acc3c8d1 | 2164 | } |
d3649f68 | 2165 | EXPORT_SYMBOL(get_user_pages_locked); |
acc3c8d1 KS |
2166 | |
2167 | /* | |
d3649f68 | 2168 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 2169 | * |
3e4e28c5 | 2170 | * mmap_read_lock(mm); |
64019a2e | 2171 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 2172 | * mmap_read_unlock(mm); |
d3649f68 CH |
2173 | * |
2174 | * with: | |
2175 | * | |
64019a2e | 2176 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
2177 | * |
2178 | * It is functionally equivalent to get_user_pages_fast so | |
2179 | * get_user_pages_fast should be used instead if specific gup_flags | |
2180 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 2181 | */ |
d3649f68 CH |
2182 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
2183 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 KS |
2184 | { |
2185 | struct mm_struct *mm = current->mm; | |
d3649f68 CH |
2186 | int locked = 1; |
2187 | long ret; | |
acc3c8d1 | 2188 | |
d3649f68 CH |
2189 | /* |
2190 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
2191 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2192 | * vmas. As there are no users of this flag in this call we simply | |
2193 | * disallow this option for now. | |
2194 | */ | |
2195 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
2196 | return -EINVAL; | |
acc3c8d1 | 2197 | |
d8ed45c5 | 2198 | mmap_read_lock(mm); |
64019a2e | 2199 | ret = __get_user_pages_locked(mm, start, nr_pages, pages, NULL, |
d3649f68 | 2200 | &locked, gup_flags | FOLL_TOUCH); |
acc3c8d1 | 2201 | if (locked) |
d8ed45c5 | 2202 | mmap_read_unlock(mm); |
d3649f68 | 2203 | return ret; |
4bbd4c77 | 2204 | } |
d3649f68 | 2205 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
2206 | |
2207 | /* | |
67a929e0 | 2208 | * Fast GUP |
2667f50e SC |
2209 | * |
2210 | * get_user_pages_fast attempts to pin user pages by walking the page | |
2211 | * tables directly and avoids taking locks. Thus the walker needs to be | |
2212 | * protected from page table pages being freed from under it, and should | |
2213 | * block any THP splits. | |
2214 | * | |
2215 | * One way to achieve this is to have the walker disable interrupts, and | |
2216 | * rely on IPIs from the TLB flushing code blocking before the page table | |
2217 | * pages are freed. This is unsuitable for architectures that do not need | |
2218 | * to broadcast an IPI when invalidating TLBs. | |
2219 | * | |
2220 | * Another way to achieve this is to batch up page table containing pages | |
2221 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
2222 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
2223 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
2224 | * (which is a relatively rare event). The code below adopts this strategy. | |
2225 | * | |
2226 | * Before activating this code, please be aware that the following assumptions | |
2227 | * are currently made: | |
2228 | * | |
ff2e6d72 | 2229 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 2230 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 2231 | * |
2667f50e SC |
2232 | * *) ptes can be read atomically by the architecture. |
2233 | * | |
2234 | * *) access_ok is sufficient to validate userspace address ranges. | |
2235 | * | |
2236 | * The last two assumptions can be relaxed by the addition of helper functions. | |
2237 | * | |
2238 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
2239 | */ | |
67a929e0 | 2240 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 2241 | |
790c7369 | 2242 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 2243 | unsigned int flags, |
790c7369 | 2244 | struct page **pages) |
b59f65fa KS |
2245 | { |
2246 | while ((*nr) - nr_start) { | |
2247 | struct page *page = pages[--(*nr)]; | |
2248 | ||
2249 | ClearPageReferenced(page); | |
3faa52c0 JH |
2250 | if (flags & FOLL_PIN) |
2251 | unpin_user_page(page); | |
2252 | else | |
2253 | put_page(page); | |
b59f65fa KS |
2254 | } |
2255 | } | |
2256 | ||
3010a5ea | 2257 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
2667f50e | 2258 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, |
b798bec4 | 2259 | unsigned int flags, struct page **pages, int *nr) |
2667f50e | 2260 | { |
b59f65fa KS |
2261 | struct dev_pagemap *pgmap = NULL; |
2262 | int nr_start = *nr, ret = 0; | |
2667f50e | 2263 | pte_t *ptep, *ptem; |
2667f50e SC |
2264 | |
2265 | ptem = ptep = pte_offset_map(&pmd, addr); | |
2266 | do { | |
2a4a06da | 2267 | pte_t pte = ptep_get_lockless(ptep); |
7aef4172 | 2268 | struct page *head, *page; |
2667f50e SC |
2269 | |
2270 | /* | |
2271 | * Similar to the PMD case below, NUMA hinting must take slow | |
8a0516ed | 2272 | * path using the pte_protnone check. |
2667f50e | 2273 | */ |
e7884f8e KS |
2274 | if (pte_protnone(pte)) |
2275 | goto pte_unmap; | |
2276 | ||
b798bec4 | 2277 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2278 | goto pte_unmap; |
2279 | ||
b59f65fa | 2280 | if (pte_devmap(pte)) { |
7af75561 IW |
2281 | if (unlikely(flags & FOLL_LONGTERM)) |
2282 | goto pte_unmap; | |
2283 | ||
b59f65fa KS |
2284 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2285 | if (unlikely(!pgmap)) { | |
3b78d834 | 2286 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2287 | goto pte_unmap; |
2288 | } | |
2289 | } else if (pte_special(pte)) | |
2667f50e SC |
2290 | goto pte_unmap; |
2291 | ||
2292 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2293 | page = pte_page(pte); | |
2294 | ||
3faa52c0 | 2295 | head = try_grab_compound_head(page, 1, flags); |
8fde12ca | 2296 | if (!head) |
2667f50e SC |
2297 | goto pte_unmap; |
2298 | ||
1507f512 MR |
2299 | if (unlikely(page_is_secretmem(page))) { |
2300 | put_compound_head(head, 1, flags); | |
2301 | goto pte_unmap; | |
2302 | } | |
2303 | ||
2667f50e | 2304 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { |
3faa52c0 | 2305 | put_compound_head(head, 1, flags); |
2667f50e SC |
2306 | goto pte_unmap; |
2307 | } | |
2308 | ||
7aef4172 | 2309 | VM_BUG_ON_PAGE(compound_head(page) != head, page); |
e9348053 | 2310 | |
f28d4363 CI |
2311 | /* |
2312 | * We need to make the page accessible if and only if we are | |
2313 | * going to access its content (the FOLL_PIN case). Please | |
2314 | * see Documentation/core-api/pin_user_pages.rst for | |
2315 | * details. | |
2316 | */ | |
2317 | if (flags & FOLL_PIN) { | |
2318 | ret = arch_make_page_accessible(page); | |
2319 | if (ret) { | |
2320 | unpin_user_page(page); | |
2321 | goto pte_unmap; | |
2322 | } | |
2323 | } | |
e9348053 | 2324 | SetPageReferenced(page); |
2667f50e SC |
2325 | pages[*nr] = page; |
2326 | (*nr)++; | |
2327 | ||
2328 | } while (ptep++, addr += PAGE_SIZE, addr != end); | |
2329 | ||
2330 | ret = 1; | |
2331 | ||
2332 | pte_unmap: | |
832d7aa0 CH |
2333 | if (pgmap) |
2334 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2335 | pte_unmap(ptem); |
2336 | return ret; | |
2337 | } | |
2338 | #else | |
2339 | ||
2340 | /* | |
2341 | * If we can't determine whether or not a pte is special, then fail immediately | |
2342 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2343 | * to be special. | |
2344 | * | |
2345 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2346 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2347 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2348 | */ | |
2349 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, | |
b798bec4 | 2350 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2351 | { |
2352 | return 0; | |
2353 | } | |
3010a5ea | 2354 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2355 | |
17596731 | 2356 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2357 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2358 | unsigned long end, unsigned int flags, |
2359 | struct page **pages, int *nr) | |
b59f65fa KS |
2360 | { |
2361 | int nr_start = *nr; | |
2362 | struct dev_pagemap *pgmap = NULL; | |
2363 | ||
2364 | do { | |
2365 | struct page *page = pfn_to_page(pfn); | |
2366 | ||
2367 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2368 | if (unlikely(!pgmap)) { | |
3b78d834 | 2369 | undo_dev_pagemap(nr, nr_start, flags, pages); |
6401c4eb | 2370 | break; |
b59f65fa KS |
2371 | } |
2372 | SetPageReferenced(page); | |
2373 | pages[*nr] = page; | |
3faa52c0 JH |
2374 | if (unlikely(!try_grab_page(page, flags))) { |
2375 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
6401c4eb | 2376 | break; |
3faa52c0 | 2377 | } |
b59f65fa KS |
2378 | (*nr)++; |
2379 | pfn++; | |
2380 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 | 2381 | |
6401c4eb | 2382 | put_dev_pagemap(pgmap); |
20b7fee7 | 2383 | return addr == end; |
b59f65fa KS |
2384 | } |
2385 | ||
a9b6de77 | 2386 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2387 | unsigned long end, unsigned int flags, |
2388 | struct page **pages, int *nr) | |
b59f65fa KS |
2389 | { |
2390 | unsigned long fault_pfn; | |
a9b6de77 DW |
2391 | int nr_start = *nr; |
2392 | ||
2393 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2394 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2395 | return 0; |
b59f65fa | 2396 | |
a9b6de77 | 2397 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2398 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2399 | return 0; |
2400 | } | |
2401 | return 1; | |
b59f65fa KS |
2402 | } |
2403 | ||
a9b6de77 | 2404 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2405 | unsigned long end, unsigned int flags, |
2406 | struct page **pages, int *nr) | |
b59f65fa KS |
2407 | { |
2408 | unsigned long fault_pfn; | |
a9b6de77 DW |
2409 | int nr_start = *nr; |
2410 | ||
2411 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2412 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2413 | return 0; |
b59f65fa | 2414 | |
a9b6de77 | 2415 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2416 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2417 | return 0; |
2418 | } | |
2419 | return 1; | |
b59f65fa KS |
2420 | } |
2421 | #else | |
a9b6de77 | 2422 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2423 | unsigned long end, unsigned int flags, |
2424 | struct page **pages, int *nr) | |
b59f65fa KS |
2425 | { |
2426 | BUILD_BUG(); | |
2427 | return 0; | |
2428 | } | |
2429 | ||
a9b6de77 | 2430 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2431 | unsigned long end, unsigned int flags, |
2432 | struct page **pages, int *nr) | |
b59f65fa KS |
2433 | { |
2434 | BUILD_BUG(); | |
2435 | return 0; | |
2436 | } | |
2437 | #endif | |
2438 | ||
a43e9820 JH |
2439 | static int record_subpages(struct page *page, unsigned long addr, |
2440 | unsigned long end, struct page **pages) | |
2441 | { | |
2442 | int nr; | |
2443 | ||
2444 | for (nr = 0; addr != end; addr += PAGE_SIZE) | |
2445 | pages[nr++] = page++; | |
2446 | ||
2447 | return nr; | |
2448 | } | |
2449 | ||
cbd34da7 CH |
2450 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2451 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2452 | unsigned long sz) | |
2453 | { | |
2454 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2455 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2456 | } | |
2457 | ||
2458 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2459 | unsigned long end, unsigned int flags, |
2460 | struct page **pages, int *nr) | |
cbd34da7 CH |
2461 | { |
2462 | unsigned long pte_end; | |
2463 | struct page *head, *page; | |
2464 | pte_t pte; | |
2465 | int refs; | |
2466 | ||
2467 | pte_end = (addr + sz) & ~(sz-1); | |
2468 | if (pte_end < end) | |
2469 | end = pte_end; | |
2470 | ||
55ca2263 | 2471 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2472 | |
0cd22afd | 2473 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2474 | return 0; |
2475 | ||
2476 | /* hugepages are never "special" */ | |
2477 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2478 | ||
cbd34da7 | 2479 | head = pte_page(pte); |
cbd34da7 | 2480 | page = head + ((addr & (sz-1)) >> PAGE_SHIFT); |
a43e9820 | 2481 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2482 | |
3faa52c0 | 2483 | head = try_grab_compound_head(head, refs, flags); |
a43e9820 | 2484 | if (!head) |
cbd34da7 | 2485 | return 0; |
cbd34da7 CH |
2486 | |
2487 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
3b78d834 | 2488 | put_compound_head(head, refs, flags); |
cbd34da7 CH |
2489 | return 0; |
2490 | } | |
2491 | ||
a43e9820 | 2492 | *nr += refs; |
520b4a44 | 2493 | SetPageReferenced(head); |
cbd34da7 CH |
2494 | return 1; |
2495 | } | |
2496 | ||
2497 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2498 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2499 | struct page **pages, int *nr) |
2500 | { | |
2501 | pte_t *ptep; | |
2502 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2503 | unsigned long next; | |
2504 | ||
2505 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2506 | do { | |
2507 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2508 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2509 | return 0; |
2510 | } while (ptep++, addr = next, addr != end); | |
2511 | ||
2512 | return 1; | |
2513 | } | |
2514 | #else | |
2515 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2516 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2517 | struct page **pages, int *nr) |
2518 | { | |
2519 | return 0; | |
2520 | } | |
2521 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2522 | ||
2667f50e | 2523 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2524 | unsigned long end, unsigned int flags, |
2525 | struct page **pages, int *nr) | |
2667f50e | 2526 | { |
ddc58f27 | 2527 | struct page *head, *page; |
2667f50e SC |
2528 | int refs; |
2529 | ||
b798bec4 | 2530 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2531 | return 0; |
2532 | ||
7af75561 IW |
2533 | if (pmd_devmap(orig)) { |
2534 | if (unlikely(flags & FOLL_LONGTERM)) | |
2535 | return 0; | |
86dfbed4 JH |
2536 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2537 | pages, nr); | |
7af75561 | 2538 | } |
b59f65fa | 2539 | |
d63206ee | 2540 | page = pmd_page(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2541 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2542 | |
3faa52c0 | 2543 | head = try_grab_compound_head(pmd_page(orig), refs, flags); |
a43e9820 | 2544 | if (!head) |
2667f50e | 2545 | return 0; |
2667f50e SC |
2546 | |
2547 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
3b78d834 | 2548 | put_compound_head(head, refs, flags); |
2667f50e SC |
2549 | return 0; |
2550 | } | |
2551 | ||
a43e9820 | 2552 | *nr += refs; |
e9348053 | 2553 | SetPageReferenced(head); |
2667f50e SC |
2554 | return 1; |
2555 | } | |
2556 | ||
2557 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2558 | unsigned long end, unsigned int flags, |
2559 | struct page **pages, int *nr) | |
2667f50e | 2560 | { |
ddc58f27 | 2561 | struct page *head, *page; |
2667f50e SC |
2562 | int refs; |
2563 | ||
b798bec4 | 2564 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2565 | return 0; |
2566 | ||
7af75561 IW |
2567 | if (pud_devmap(orig)) { |
2568 | if (unlikely(flags & FOLL_LONGTERM)) | |
2569 | return 0; | |
86dfbed4 JH |
2570 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2571 | pages, nr); | |
7af75561 | 2572 | } |
b59f65fa | 2573 | |
d63206ee | 2574 | page = pud_page(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2575 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2576 | |
3faa52c0 | 2577 | head = try_grab_compound_head(pud_page(orig), refs, flags); |
a43e9820 | 2578 | if (!head) |
2667f50e | 2579 | return 0; |
2667f50e SC |
2580 | |
2581 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
3b78d834 | 2582 | put_compound_head(head, refs, flags); |
2667f50e SC |
2583 | return 0; |
2584 | } | |
2585 | ||
a43e9820 | 2586 | *nr += refs; |
e9348053 | 2587 | SetPageReferenced(head); |
2667f50e SC |
2588 | return 1; |
2589 | } | |
2590 | ||
f30c59e9 | 2591 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2592 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2593 | struct page **pages, int *nr) |
2594 | { | |
2595 | int refs; | |
ddc58f27 | 2596 | struct page *head, *page; |
f30c59e9 | 2597 | |
b798bec4 | 2598 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2599 | return 0; |
2600 | ||
b59f65fa | 2601 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2602 | |
d63206ee | 2603 | page = pgd_page(orig) + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2604 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2605 | |
3faa52c0 | 2606 | head = try_grab_compound_head(pgd_page(orig), refs, flags); |
a43e9820 | 2607 | if (!head) |
f30c59e9 | 2608 | return 0; |
f30c59e9 AK |
2609 | |
2610 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
3b78d834 | 2611 | put_compound_head(head, refs, flags); |
f30c59e9 AK |
2612 | return 0; |
2613 | } | |
2614 | ||
a43e9820 | 2615 | *nr += refs; |
e9348053 | 2616 | SetPageReferenced(head); |
f30c59e9 AK |
2617 | return 1; |
2618 | } | |
2619 | ||
d3f7b1bb | 2620 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2621 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2622 | { |
2623 | unsigned long next; | |
2624 | pmd_t *pmdp; | |
2625 | ||
d3f7b1bb | 2626 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2627 | do { |
38c5ce93 | 2628 | pmd_t pmd = READ_ONCE(*pmdp); |
2667f50e SC |
2629 | |
2630 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2631 | if (!pmd_present(pmd)) |
2667f50e SC |
2632 | return 0; |
2633 | ||
414fd080 YZ |
2634 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2635 | pmd_devmap(pmd))) { | |
2667f50e SC |
2636 | /* |
2637 | * NUMA hinting faults need to be handled in the GUP | |
2638 | * slowpath for accounting purposes and so that they | |
2639 | * can be serialised against THP migration. | |
2640 | */ | |
8a0516ed | 2641 | if (pmd_protnone(pmd)) |
2667f50e SC |
2642 | return 0; |
2643 | ||
b798bec4 | 2644 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2645 | pages, nr)) |
2646 | return 0; | |
2647 | ||
f30c59e9 AK |
2648 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2649 | /* | |
2650 | * architecture have different format for hugetlbfs | |
2651 | * pmd format and THP pmd format | |
2652 | */ | |
2653 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2654 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2655 | return 0; |
b798bec4 | 2656 | } else if (!gup_pte_range(pmd, addr, next, flags, pages, nr)) |
2923117b | 2657 | return 0; |
2667f50e SC |
2658 | } while (pmdp++, addr = next, addr != end); |
2659 | ||
2660 | return 1; | |
2661 | } | |
2662 | ||
d3f7b1bb | 2663 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2664 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2665 | { |
2666 | unsigned long next; | |
2667 | pud_t *pudp; | |
2668 | ||
d3f7b1bb | 2669 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2670 | do { |
e37c6982 | 2671 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2672 | |
2673 | next = pud_addr_end(addr, end); | |
15494520 | 2674 | if (unlikely(!pud_present(pud))) |
2667f50e | 2675 | return 0; |
f30c59e9 | 2676 | if (unlikely(pud_huge(pud))) { |
b798bec4 | 2677 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2678 | pages, nr)) |
2679 | return 0; | |
2680 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2681 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2682 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2683 | return 0; |
d3f7b1bb | 2684 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2685 | return 0; |
2686 | } while (pudp++, addr = next, addr != end); | |
2687 | ||
2688 | return 1; | |
2689 | } | |
2690 | ||
d3f7b1bb | 2691 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2692 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2693 | { |
2694 | unsigned long next; | |
2695 | p4d_t *p4dp; | |
2696 | ||
d3f7b1bb | 2697 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
2698 | do { |
2699 | p4d_t p4d = READ_ONCE(*p4dp); | |
2700 | ||
2701 | next = p4d_addr_end(addr, end); | |
2702 | if (p4d_none(p4d)) | |
2703 | return 0; | |
2704 | BUILD_BUG_ON(p4d_huge(p4d)); | |
2705 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
2706 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 2707 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 2708 | return 0; |
d3f7b1bb | 2709 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
2710 | return 0; |
2711 | } while (p4dp++, addr = next, addr != end); | |
2712 | ||
2713 | return 1; | |
2714 | } | |
2715 | ||
5b65c467 | 2716 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 2717 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
2718 | { |
2719 | unsigned long next; | |
2720 | pgd_t *pgdp; | |
2721 | ||
2722 | pgdp = pgd_offset(current->mm, addr); | |
2723 | do { | |
2724 | pgd_t pgd = READ_ONCE(*pgdp); | |
2725 | ||
2726 | next = pgd_addr_end(addr, end); | |
2727 | if (pgd_none(pgd)) | |
2728 | return; | |
2729 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 2730 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
2731 | pages, nr)) |
2732 | return; | |
2733 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
2734 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 2735 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 2736 | return; |
d3f7b1bb | 2737 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
2738 | return; |
2739 | } while (pgdp++, addr = next, addr != end); | |
2740 | } | |
050a9adc CH |
2741 | #else |
2742 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
2743 | unsigned int flags, struct page **pages, int *nr) | |
2744 | { | |
2745 | } | |
2746 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
2747 | |
2748 | #ifndef gup_fast_permitted | |
2749 | /* | |
dadbb612 | 2750 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
2751 | * we need to fall back to the slow version: |
2752 | */ | |
26f4c328 | 2753 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 2754 | { |
26f4c328 | 2755 | return true; |
5b65c467 KS |
2756 | } |
2757 | #endif | |
2758 | ||
7af75561 IW |
2759 | static int __gup_longterm_unlocked(unsigned long start, int nr_pages, |
2760 | unsigned int gup_flags, struct page **pages) | |
2761 | { | |
2762 | int ret; | |
2763 | ||
2764 | /* | |
2765 | * FIXME: FOLL_LONGTERM does not work with | |
2766 | * get_user_pages_unlocked() (see comments in that function) | |
2767 | */ | |
2768 | if (gup_flags & FOLL_LONGTERM) { | |
d8ed45c5 | 2769 | mmap_read_lock(current->mm); |
64019a2e | 2770 | ret = __gup_longterm_locked(current->mm, |
7af75561 IW |
2771 | start, nr_pages, |
2772 | pages, NULL, gup_flags); | |
d8ed45c5 | 2773 | mmap_read_unlock(current->mm); |
7af75561 IW |
2774 | } else { |
2775 | ret = get_user_pages_unlocked(start, nr_pages, | |
2776 | pages, gup_flags); | |
2777 | } | |
2778 | ||
2779 | return ret; | |
2780 | } | |
2781 | ||
c28b1fc7 JG |
2782 | static unsigned long lockless_pages_from_mm(unsigned long start, |
2783 | unsigned long end, | |
2784 | unsigned int gup_flags, | |
2785 | struct page **pages) | |
2786 | { | |
2787 | unsigned long flags; | |
2788 | int nr_pinned = 0; | |
57efa1fe | 2789 | unsigned seq; |
c28b1fc7 JG |
2790 | |
2791 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
2792 | !gup_fast_permitted(start, end)) | |
2793 | return 0; | |
2794 | ||
57efa1fe JG |
2795 | if (gup_flags & FOLL_PIN) { |
2796 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
2797 | if (seq & 1) | |
2798 | return 0; | |
2799 | } | |
2800 | ||
c28b1fc7 JG |
2801 | /* |
2802 | * Disable interrupts. The nested form is used, in order to allow full, | |
2803 | * general purpose use of this routine. | |
2804 | * | |
2805 | * With interrupts disabled, we block page table pages from being freed | |
2806 | * from under us. See struct mmu_table_batch comments in | |
2807 | * include/asm-generic/tlb.h for more details. | |
2808 | * | |
2809 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
2810 | * that come from THPs splitting. | |
2811 | */ | |
2812 | local_irq_save(flags); | |
2813 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
2814 | local_irq_restore(flags); | |
57efa1fe JG |
2815 | |
2816 | /* | |
2817 | * When pinning pages for DMA there could be a concurrent write protect | |
2818 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
2819 | */ | |
2820 | if (gup_flags & FOLL_PIN) { | |
2821 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
2822 | unpin_user_pages(pages, nr_pinned); | |
2823 | return 0; | |
2824 | } | |
2825 | } | |
c28b1fc7 JG |
2826 | return nr_pinned; |
2827 | } | |
2828 | ||
2829 | static int internal_get_user_pages_fast(unsigned long start, | |
2830 | unsigned long nr_pages, | |
eddb1c22 JH |
2831 | unsigned int gup_flags, |
2832 | struct page **pages) | |
2667f50e | 2833 | { |
c28b1fc7 JG |
2834 | unsigned long len, end; |
2835 | unsigned long nr_pinned; | |
2836 | int ret; | |
2667f50e | 2837 | |
f4000fdf | 2838 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef | 2839 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
55b8fe70 | 2840 | FOLL_FAST_ONLY | FOLL_NOFAULT))) |
817be129 CH |
2841 | return -EINVAL; |
2842 | ||
a458b76a AA |
2843 | if (gup_flags & FOLL_PIN) |
2844 | mm_set_has_pinned_flag(¤t->mm->flags); | |
008cfe44 | 2845 | |
f81cd178 | 2846 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 2847 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 2848 | |
f455c854 | 2849 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
2850 | len = nr_pages << PAGE_SHIFT; |
2851 | if (check_add_overflow(start, len, &end)) | |
c61611f7 | 2852 | return 0; |
96d4f267 | 2853 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 2854 | return -EFAULT; |
73e10a61 | 2855 | |
c28b1fc7 JG |
2856 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
2857 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
2858 | return nr_pinned; | |
2667f50e | 2859 | |
c28b1fc7 JG |
2860 | /* Slow path: try to get the remaining pages with get_user_pages */ |
2861 | start += nr_pinned << PAGE_SHIFT; | |
2862 | pages += nr_pinned; | |
2863 | ret = __gup_longterm_unlocked(start, nr_pages - nr_pinned, gup_flags, | |
2864 | pages); | |
2865 | if (ret < 0) { | |
2866 | /* | |
2867 | * The caller has to unpin the pages we already pinned so | |
2868 | * returning -errno is not an option | |
2869 | */ | |
2870 | if (nr_pinned) | |
2871 | return nr_pinned; | |
2872 | return ret; | |
2667f50e | 2873 | } |
c28b1fc7 | 2874 | return ret + nr_pinned; |
2667f50e | 2875 | } |
c28b1fc7 | 2876 | |
dadbb612 SJ |
2877 | /** |
2878 | * get_user_pages_fast_only() - pin user pages in memory | |
2879 | * @start: starting user address | |
2880 | * @nr_pages: number of pages from start to pin | |
2881 | * @gup_flags: flags modifying pin behaviour | |
2882 | * @pages: array that receives pointers to the pages pinned. | |
2883 | * Should be at least nr_pages long. | |
2884 | * | |
9e1f0580 JH |
2885 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
2886 | * the regular GUP. | |
2887 | * Note a difference with get_user_pages_fast: this always returns the | |
2888 | * number of pages pinned, 0 if no pages were pinned. | |
2889 | * | |
2890 | * If the architecture does not support this function, simply return with no | |
2891 | * pages pinned. | |
2892 | * | |
2893 | * Careful, careful! COW breaking can go either way, so a non-write | |
2894 | * access can get ambiguous page results. If you call this function without | |
2895 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
2896 | */ | |
dadbb612 SJ |
2897 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
2898 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 2899 | { |
376a34ef | 2900 | int nr_pinned; |
9e1f0580 JH |
2901 | /* |
2902 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
2903 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
2904 | * |
2905 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2906 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 2907 | */ |
dadbb612 | 2908 | gup_flags |= FOLL_GET | FOLL_FAST_ONLY; |
9e1f0580 | 2909 | |
376a34ef JH |
2910 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
2911 | pages); | |
9e1f0580 JH |
2912 | |
2913 | /* | |
376a34ef JH |
2914 | * As specified in the API description above, this routine is not |
2915 | * allowed to return negative values. However, the common core | |
2916 | * routine internal_get_user_pages_fast() *can* return -errno. | |
2917 | * Therefore, correct for that here: | |
9e1f0580 | 2918 | */ |
376a34ef JH |
2919 | if (nr_pinned < 0) |
2920 | nr_pinned = 0; | |
9e1f0580 JH |
2921 | |
2922 | return nr_pinned; | |
2923 | } | |
dadbb612 | 2924 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 2925 | |
eddb1c22 JH |
2926 | /** |
2927 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
2928 | * @start: starting user address |
2929 | * @nr_pages: number of pages from start to pin | |
2930 | * @gup_flags: flags modifying pin behaviour | |
2931 | * @pages: array that receives pointers to the pages pinned. | |
2932 | * Should be at least nr_pages long. | |
eddb1c22 | 2933 | * |
c1e8d7c6 | 2934 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
2935 | * If not successful, it will fall back to taking the lock and |
2936 | * calling get_user_pages(). | |
2937 | * | |
2938 | * Returns number of pages pinned. This may be fewer than the number requested. | |
2939 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
2940 | * -errno. | |
2941 | */ | |
2942 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
2943 | unsigned int gup_flags, struct page **pages) | |
2944 | { | |
447f3e45 | 2945 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2946 | return -EINVAL; |
2947 | ||
94202f12 JH |
2948 | /* |
2949 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
2950 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
2951 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
2952 | * request. | |
2953 | */ | |
2954 | gup_flags |= FOLL_GET; | |
eddb1c22 JH |
2955 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
2956 | } | |
050a9adc | 2957 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
2958 | |
2959 | /** | |
2960 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
2961 | * | |
3faa52c0 JH |
2962 | * @start: starting user address |
2963 | * @nr_pages: number of pages from start to pin | |
2964 | * @gup_flags: flags modifying pin behaviour | |
2965 | * @pages: array that receives pointers to the pages pinned. | |
2966 | * Should be at least nr_pages long. | |
2967 | * | |
2968 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
2969 | * get_user_pages_fast() for documentation on the function arguments, because | |
2970 | * the arguments here are identical. | |
2971 | * | |
2972 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2973 | * see Documentation/core-api/pin_user_pages.rst for further details. |
eddb1c22 JH |
2974 | */ |
2975 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
2976 | unsigned int gup_flags, struct page **pages) | |
2977 | { | |
3faa52c0 JH |
2978 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2979 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2980 | return -EINVAL; | |
2981 | ||
2982 | gup_flags |= FOLL_PIN; | |
2983 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); | |
eddb1c22 JH |
2984 | } |
2985 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
2986 | ||
104acc32 | 2987 | /* |
dadbb612 SJ |
2988 | * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior |
2989 | * is the same, except that this one sets FOLL_PIN instead of FOLL_GET. | |
104acc32 JH |
2990 | * |
2991 | * The API rules are the same, too: no negative values may be returned. | |
2992 | */ | |
2993 | int pin_user_pages_fast_only(unsigned long start, int nr_pages, | |
2994 | unsigned int gup_flags, struct page **pages) | |
2995 | { | |
2996 | int nr_pinned; | |
2997 | ||
2998 | /* | |
2999 | * FOLL_GET and FOLL_PIN are mutually exclusive. Note that the API | |
3000 | * rules require returning 0, rather than -errno: | |
3001 | */ | |
3002 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3003 | return 0; | |
3004 | /* | |
3005 | * FOLL_FAST_ONLY is required in order to match the API description of | |
3006 | * this routine: no fall back to regular ("slow") GUP. | |
3007 | */ | |
3008 | gup_flags |= (FOLL_PIN | FOLL_FAST_ONLY); | |
3009 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, | |
3010 | pages); | |
3011 | /* | |
3012 | * This routine is not allowed to return negative values. However, | |
3013 | * internal_get_user_pages_fast() *can* return -errno. Therefore, | |
3014 | * correct for that here: | |
3015 | */ | |
3016 | if (nr_pinned < 0) | |
3017 | nr_pinned = 0; | |
3018 | ||
3019 | return nr_pinned; | |
3020 | } | |
3021 | EXPORT_SYMBOL_GPL(pin_user_pages_fast_only); | |
3022 | ||
eddb1c22 | 3023 | /** |
64019a2e | 3024 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 3025 | * |
3faa52c0 JH |
3026 | * @mm: mm_struct of target mm |
3027 | * @start: starting user address | |
3028 | * @nr_pages: number of pages from start to pin | |
3029 | * @gup_flags: flags modifying lookup behaviour | |
3030 | * @pages: array that receives pointers to the pages pinned. | |
3031 | * Should be at least nr_pages long. Or NULL, if caller | |
3032 | * only intends to ensure the pages are faulted in. | |
3033 | * @vmas: array of pointers to vmas corresponding to each page. | |
3034 | * Or NULL if the caller does not require them. | |
3035 | * @locked: pointer to lock flag indicating whether lock is held and | |
3036 | * subsequently whether VM_FAULT_RETRY functionality can be | |
3037 | * utilised. Lock must initially be held. | |
3038 | * | |
3039 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
3040 | * get_user_pages_remote() for documentation on the function arguments, because | |
3041 | * the arguments here are identical. | |
3042 | * | |
3043 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3044 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 | 3045 | */ |
64019a2e | 3046 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
3047 | unsigned long start, unsigned long nr_pages, |
3048 | unsigned int gup_flags, struct page **pages, | |
3049 | struct vm_area_struct **vmas, int *locked) | |
3050 | { | |
3faa52c0 JH |
3051 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3052 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3053 | return -EINVAL; | |
3054 | ||
3055 | gup_flags |= FOLL_PIN; | |
64019a2e | 3056 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
3faa52c0 | 3057 | pages, vmas, locked); |
eddb1c22 JH |
3058 | } |
3059 | EXPORT_SYMBOL(pin_user_pages_remote); | |
3060 | ||
3061 | /** | |
3062 | * pin_user_pages() - pin user pages in memory for use by other devices | |
3063 | * | |
3faa52c0 JH |
3064 | * @start: starting user address |
3065 | * @nr_pages: number of pages from start to pin | |
3066 | * @gup_flags: flags modifying lookup behaviour | |
3067 | * @pages: array that receives pointers to the pages pinned. | |
3068 | * Should be at least nr_pages long. Or NULL, if caller | |
3069 | * only intends to ensure the pages are faulted in. | |
3070 | * @vmas: array of pointers to vmas corresponding to each page. | |
3071 | * Or NULL if the caller does not require them. | |
3072 | * | |
3073 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
3074 | * FOLL_PIN is set. | |
3075 | * | |
3076 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 3077 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 JH |
3078 | */ |
3079 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
3080 | unsigned int gup_flags, struct page **pages, | |
3081 | struct vm_area_struct **vmas) | |
3082 | { | |
3faa52c0 JH |
3083 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
3084 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3085 | return -EINVAL; | |
3086 | ||
3087 | gup_flags |= FOLL_PIN; | |
64019a2e | 3088 | return __gup_longterm_locked(current->mm, start, nr_pages, |
3faa52c0 | 3089 | pages, vmas, gup_flags); |
eddb1c22 JH |
3090 | } |
3091 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
3092 | |
3093 | /* | |
3094 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
3095 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
3096 | * FOLL_PIN and rejects FOLL_GET. | |
3097 | */ | |
3098 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
3099 | struct page **pages, unsigned int gup_flags) | |
3100 | { | |
3101 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
3102 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3103 | return -EINVAL; | |
3104 | ||
3105 | gup_flags |= FOLL_PIN; | |
3106 | return get_user_pages_unlocked(start, nr_pages, pages, gup_flags); | |
3107 | } | |
3108 | EXPORT_SYMBOL(pin_user_pages_unlocked); | |
420c2091 JH |
3109 | |
3110 | /* | |
3111 | * pin_user_pages_locked() is the FOLL_PIN variant of get_user_pages_locked(). | |
3112 | * Behavior is the same, except that this one sets FOLL_PIN and rejects | |
3113 | * FOLL_GET. | |
3114 | */ | |
3115 | long pin_user_pages_locked(unsigned long start, unsigned long nr_pages, | |
3116 | unsigned int gup_flags, struct page **pages, | |
3117 | int *locked) | |
3118 | { | |
3119 | /* | |
3120 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
3121 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
3122 | * vmas. As there are no users of this flag in this call we simply | |
3123 | * disallow this option for now. | |
3124 | */ | |
3125 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
3126 | return -EINVAL; | |
3127 | ||
3128 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
3129 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
3130 | return -EINVAL; | |
3131 | ||
3132 | gup_flags |= FOLL_PIN; | |
64019a2e | 3133 | return __get_user_pages_locked(current->mm, start, nr_pages, |
420c2091 JH |
3134 | pages, NULL, locked, |
3135 | gup_flags | FOLL_TOUCH); | |
3136 | } | |
3137 | EXPORT_SYMBOL(pin_user_pages_locked); |