Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
4bbd4c77 KS |
2 | #include <linux/kernel.h> |
3 | #include <linux/errno.h> | |
4 | #include <linux/err.h> | |
5 | #include <linux/spinlock.h> | |
6 | ||
4bbd4c77 | 7 | #include <linux/mm.h> |
3565fce3 | 8 | #include <linux/memremap.h> |
4bbd4c77 KS |
9 | #include <linux/pagemap.h> |
10 | #include <linux/rmap.h> | |
11 | #include <linux/swap.h> | |
12 | #include <linux/swapops.h> | |
13 | ||
174cd4b1 | 14 | #include <linux/sched/signal.h> |
2667f50e | 15 | #include <linux/rwsem.h> |
f30c59e9 | 16 | #include <linux/hugetlb.h> |
9a4e9f3b AK |
17 | #include <linux/migrate.h> |
18 | #include <linux/mm_inline.h> | |
19 | #include <linux/sched/mm.h> | |
1027e443 | 20 | |
33a709b2 | 21 | #include <asm/mmu_context.h> |
1027e443 | 22 | #include <asm/tlbflush.h> |
2667f50e | 23 | |
4bbd4c77 KS |
24 | #include "internal.h" |
25 | ||
df06b37f KB |
26 | struct follow_page_context { |
27 | struct dev_pagemap *pgmap; | |
28 | unsigned int page_mask; | |
29 | }; | |
30 | ||
47e29d32 JH |
31 | static void hpage_pincount_add(struct page *page, int refs) |
32 | { | |
33 | VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); | |
34 | VM_BUG_ON_PAGE(page != compound_head(page), page); | |
35 | ||
36 | atomic_add(refs, compound_pincount_ptr(page)); | |
37 | } | |
38 | ||
39 | static void hpage_pincount_sub(struct page *page, int refs) | |
40 | { | |
41 | VM_BUG_ON_PAGE(!hpage_pincount_available(page), page); | |
42 | VM_BUG_ON_PAGE(page != compound_head(page), page); | |
43 | ||
44 | atomic_sub(refs, compound_pincount_ptr(page)); | |
45 | } | |
46 | ||
a707cdd5 JH |
47 | /* |
48 | * Return the compound head page with ref appropriately incremented, | |
49 | * or NULL if that failed. | |
50 | */ | |
51 | static inline struct page *try_get_compound_head(struct page *page, int refs) | |
52 | { | |
53 | struct page *head = compound_head(page); | |
54 | ||
55 | if (WARN_ON_ONCE(page_ref_count(head) < 0)) | |
56 | return NULL; | |
57 | if (unlikely(!page_cache_add_speculative(head, refs))) | |
58 | return NULL; | |
59 | return head; | |
60 | } | |
61 | ||
3faa52c0 JH |
62 | /* |
63 | * try_grab_compound_head() - attempt to elevate a page's refcount, by a | |
64 | * flags-dependent amount. | |
65 | * | |
66 | * "grab" names in this file mean, "look at flags to decide whether to use | |
67 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
68 | * | |
69 | * Either FOLL_PIN or FOLL_GET (or neither) must be set, but not both at the | |
70 | * same time. (That's true throughout the get_user_pages*() and | |
71 | * pin_user_pages*() APIs.) Cases: | |
72 | * | |
73 | * FOLL_GET: page's refcount will be incremented by 1. | |
74 | * FOLL_PIN: page's refcount will be incremented by GUP_PIN_COUNTING_BIAS. | |
75 | * | |
76 | * Return: head page (with refcount appropriately incremented) for success, or | |
77 | * NULL upon failure. If neither FOLL_GET nor FOLL_PIN was set, that's | |
78 | * considered failure, and furthermore, a likely bug in the caller, so a warning | |
79 | * is also emitted. | |
80 | */ | |
81 | static __maybe_unused struct page *try_grab_compound_head(struct page *page, | |
82 | int refs, | |
83 | unsigned int flags) | |
84 | { | |
85 | if (flags & FOLL_GET) | |
86 | return try_get_compound_head(page, refs); | |
87 | else if (flags & FOLL_PIN) { | |
1970dc6f JH |
88 | int orig_refs = refs; |
89 | ||
df3a0a21 PL |
90 | /* |
91 | * Can't do FOLL_LONGTERM + FOLL_PIN with CMA in the gup fast | |
92 | * path, so fail and let the caller fall back to the slow path. | |
93 | */ | |
94 | if (unlikely(flags & FOLL_LONGTERM) && | |
95 | is_migrate_cma_page(page)) | |
96 | return NULL; | |
97 | ||
47e29d32 JH |
98 | /* |
99 | * When pinning a compound page of order > 1 (which is what | |
100 | * hpage_pincount_available() checks for), use an exact count to | |
101 | * track it, via hpage_pincount_add/_sub(). | |
102 | * | |
103 | * However, be sure to *also* increment the normal page refcount | |
104 | * field at least once, so that the page really is pinned. | |
105 | */ | |
106 | if (!hpage_pincount_available(page)) | |
107 | refs *= GUP_PIN_COUNTING_BIAS; | |
108 | ||
109 | page = try_get_compound_head(page, refs); | |
110 | if (!page) | |
111 | return NULL; | |
112 | ||
113 | if (hpage_pincount_available(page)) | |
114 | hpage_pincount_add(page, refs); | |
115 | ||
1970dc6f JH |
116 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, |
117 | orig_refs); | |
118 | ||
47e29d32 | 119 | return page; |
3faa52c0 JH |
120 | } |
121 | ||
122 | WARN_ON_ONCE(1); | |
123 | return NULL; | |
124 | } | |
125 | ||
4509b42c JG |
126 | static void put_compound_head(struct page *page, int refs, unsigned int flags) |
127 | { | |
128 | if (flags & FOLL_PIN) { | |
129 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_RELEASED, | |
130 | refs); | |
131 | ||
132 | if (hpage_pincount_available(page)) | |
133 | hpage_pincount_sub(page, refs); | |
134 | else | |
135 | refs *= GUP_PIN_COUNTING_BIAS; | |
136 | } | |
137 | ||
138 | VM_BUG_ON_PAGE(page_ref_count(page) < refs, page); | |
139 | /* | |
140 | * Calling put_page() for each ref is unnecessarily slow. Only the last | |
141 | * ref needs a put_page(). | |
142 | */ | |
143 | if (refs > 1) | |
144 | page_ref_sub(page, refs - 1); | |
145 | put_page(page); | |
146 | } | |
147 | ||
3faa52c0 JH |
148 | /** |
149 | * try_grab_page() - elevate a page's refcount by a flag-dependent amount | |
150 | * | |
151 | * This might not do anything at all, depending on the flags argument. | |
152 | * | |
153 | * "grab" names in this file mean, "look at flags to decide whether to use | |
154 | * FOLL_PIN or FOLL_GET behavior, when incrementing the page's refcount. | |
155 | * | |
156 | * @page: pointer to page to be grabbed | |
157 | * @flags: gup flags: these are the FOLL_* flag values. | |
158 | * | |
159 | * Either FOLL_PIN or FOLL_GET (or neither) may be set, but not both at the same | |
160 | * time. Cases: | |
161 | * | |
162 | * FOLL_GET: page's refcount will be incremented by 1. | |
163 | * FOLL_PIN: page's refcount will be incremented by GUP_PIN_COUNTING_BIAS. | |
164 | * | |
165 | * Return: true for success, or if no action was required (if neither FOLL_PIN | |
166 | * nor FOLL_GET was set, nothing is done). False for failure: FOLL_GET or | |
167 | * FOLL_PIN was set, but the page could not be grabbed. | |
168 | */ | |
169 | bool __must_check try_grab_page(struct page *page, unsigned int flags) | |
170 | { | |
171 | WARN_ON_ONCE((flags & (FOLL_GET | FOLL_PIN)) == (FOLL_GET | FOLL_PIN)); | |
172 | ||
173 | if (flags & FOLL_GET) | |
174 | return try_get_page(page); | |
175 | else if (flags & FOLL_PIN) { | |
47e29d32 JH |
176 | int refs = 1; |
177 | ||
3faa52c0 JH |
178 | page = compound_head(page); |
179 | ||
180 | if (WARN_ON_ONCE(page_ref_count(page) <= 0)) | |
181 | return false; | |
182 | ||
47e29d32 JH |
183 | if (hpage_pincount_available(page)) |
184 | hpage_pincount_add(page, 1); | |
185 | else | |
186 | refs = GUP_PIN_COUNTING_BIAS; | |
187 | ||
188 | /* | |
189 | * Similar to try_grab_compound_head(): even if using the | |
190 | * hpage_pincount_add/_sub() routines, be sure to | |
191 | * *also* increment the normal page refcount field at least | |
192 | * once, so that the page really is pinned. | |
193 | */ | |
194 | page_ref_add(page, refs); | |
1970dc6f JH |
195 | |
196 | mod_node_page_state(page_pgdat(page), NR_FOLL_PIN_ACQUIRED, 1); | |
3faa52c0 JH |
197 | } |
198 | ||
199 | return true; | |
200 | } | |
201 | ||
3faa52c0 JH |
202 | /** |
203 | * unpin_user_page() - release a dma-pinned page | |
204 | * @page: pointer to page to be released | |
205 | * | |
206 | * Pages that were pinned via pin_user_pages*() must be released via either | |
207 | * unpin_user_page(), or one of the unpin_user_pages*() routines. This is so | |
208 | * that such pages can be separately tracked and uniquely handled. In | |
209 | * particular, interactions with RDMA and filesystems need special handling. | |
210 | */ | |
211 | void unpin_user_page(struct page *page) | |
212 | { | |
4509b42c | 213 | put_compound_head(compound_head(page), 1, FOLL_PIN); |
3faa52c0 JH |
214 | } |
215 | EXPORT_SYMBOL(unpin_user_page); | |
216 | ||
fc1d8e7c | 217 | /** |
f1f6a7dd | 218 | * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages |
2d15eb31 | 219 | * @pages: array of pages to be maybe marked dirty, and definitely released. |
fc1d8e7c | 220 | * @npages: number of pages in the @pages array. |
2d15eb31 | 221 | * @make_dirty: whether to mark the pages dirty |
fc1d8e7c JH |
222 | * |
223 | * "gup-pinned page" refers to a page that has had one of the get_user_pages() | |
224 | * variants called on that page. | |
225 | * | |
226 | * For each page in the @pages array, make that page (or its head page, if a | |
2d15eb31 | 227 | * compound page) dirty, if @make_dirty is true, and if the page was previously |
f1f6a7dd JH |
228 | * listed as clean. In any case, releases all pages using unpin_user_page(), |
229 | * possibly via unpin_user_pages(), for the non-dirty case. | |
fc1d8e7c | 230 | * |
f1f6a7dd | 231 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 232 | * |
2d15eb31 | 233 | * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is |
234 | * required, then the caller should a) verify that this is really correct, | |
235 | * because _lock() is usually required, and b) hand code it: | |
f1f6a7dd | 236 | * set_page_dirty_lock(), unpin_user_page(). |
fc1d8e7c JH |
237 | * |
238 | */ | |
f1f6a7dd JH |
239 | void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages, |
240 | bool make_dirty) | |
fc1d8e7c | 241 | { |
2d15eb31 | 242 | unsigned long index; |
fc1d8e7c | 243 | |
2d15eb31 | 244 | /* |
245 | * TODO: this can be optimized for huge pages: if a series of pages is | |
246 | * physically contiguous and part of the same compound page, then a | |
247 | * single operation to the head page should suffice. | |
248 | */ | |
249 | ||
250 | if (!make_dirty) { | |
f1f6a7dd | 251 | unpin_user_pages(pages, npages); |
2d15eb31 | 252 | return; |
253 | } | |
254 | ||
255 | for (index = 0; index < npages; index++) { | |
256 | struct page *page = compound_head(pages[index]); | |
257 | /* | |
258 | * Checking PageDirty at this point may race with | |
259 | * clear_page_dirty_for_io(), but that's OK. Two key | |
260 | * cases: | |
261 | * | |
262 | * 1) This code sees the page as already dirty, so it | |
263 | * skips the call to set_page_dirty(). That could happen | |
264 | * because clear_page_dirty_for_io() called | |
265 | * page_mkclean(), followed by set_page_dirty(). | |
266 | * However, now the page is going to get written back, | |
267 | * which meets the original intention of setting it | |
268 | * dirty, so all is well: clear_page_dirty_for_io() goes | |
269 | * on to call TestClearPageDirty(), and write the page | |
270 | * back. | |
271 | * | |
272 | * 2) This code sees the page as clean, so it calls | |
273 | * set_page_dirty(). The page stays dirty, despite being | |
274 | * written back, so it gets written back again in the | |
275 | * next writeback cycle. This is harmless. | |
276 | */ | |
277 | if (!PageDirty(page)) | |
278 | set_page_dirty_lock(page); | |
f1f6a7dd | 279 | unpin_user_page(page); |
2d15eb31 | 280 | } |
fc1d8e7c | 281 | } |
f1f6a7dd | 282 | EXPORT_SYMBOL(unpin_user_pages_dirty_lock); |
fc1d8e7c JH |
283 | |
284 | /** | |
f1f6a7dd | 285 | * unpin_user_pages() - release an array of gup-pinned pages. |
fc1d8e7c JH |
286 | * @pages: array of pages to be marked dirty and released. |
287 | * @npages: number of pages in the @pages array. | |
288 | * | |
f1f6a7dd | 289 | * For each page in the @pages array, release the page using unpin_user_page(). |
fc1d8e7c | 290 | * |
f1f6a7dd | 291 | * Please see the unpin_user_page() documentation for details. |
fc1d8e7c | 292 | */ |
f1f6a7dd | 293 | void unpin_user_pages(struct page **pages, unsigned long npages) |
fc1d8e7c JH |
294 | { |
295 | unsigned long index; | |
296 | ||
146608bb JH |
297 | /* |
298 | * If this WARN_ON() fires, then the system *might* be leaking pages (by | |
299 | * leaving them pinned), but probably not. More likely, gup/pup returned | |
300 | * a hard -ERRNO error to the caller, who erroneously passed it here. | |
301 | */ | |
302 | if (WARN_ON(IS_ERR_VALUE(npages))) | |
303 | return; | |
fc1d8e7c JH |
304 | /* |
305 | * TODO: this can be optimized for huge pages: if a series of pages is | |
306 | * physically contiguous and part of the same compound page, then a | |
307 | * single operation to the head page should suffice. | |
308 | */ | |
309 | for (index = 0; index < npages; index++) | |
f1f6a7dd | 310 | unpin_user_page(pages[index]); |
fc1d8e7c | 311 | } |
f1f6a7dd | 312 | EXPORT_SYMBOL(unpin_user_pages); |
fc1d8e7c | 313 | |
050a9adc | 314 | #ifdef CONFIG_MMU |
69e68b4f KS |
315 | static struct page *no_page_table(struct vm_area_struct *vma, |
316 | unsigned int flags) | |
4bbd4c77 | 317 | { |
69e68b4f KS |
318 | /* |
319 | * When core dumping an enormous anonymous area that nobody | |
320 | * has touched so far, we don't want to allocate unnecessary pages or | |
321 | * page tables. Return error instead of NULL to skip handle_mm_fault, | |
322 | * then get_dump_page() will return NULL to leave a hole in the dump. | |
323 | * But we can only make this optimization where a hole would surely | |
324 | * be zero-filled if handle_mm_fault() actually did handle it. | |
325 | */ | |
a0137f16 AK |
326 | if ((flags & FOLL_DUMP) && |
327 | (vma_is_anonymous(vma) || !vma->vm_ops->fault)) | |
69e68b4f KS |
328 | return ERR_PTR(-EFAULT); |
329 | return NULL; | |
330 | } | |
4bbd4c77 | 331 | |
1027e443 KS |
332 | static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address, |
333 | pte_t *pte, unsigned int flags) | |
334 | { | |
335 | /* No page to get reference */ | |
336 | if (flags & FOLL_GET) | |
337 | return -EFAULT; | |
338 | ||
339 | if (flags & FOLL_TOUCH) { | |
340 | pte_t entry = *pte; | |
341 | ||
342 | if (flags & FOLL_WRITE) | |
343 | entry = pte_mkdirty(entry); | |
344 | entry = pte_mkyoung(entry); | |
345 | ||
346 | if (!pte_same(*pte, entry)) { | |
347 | set_pte_at(vma->vm_mm, address, pte, entry); | |
348 | update_mmu_cache(vma, address, pte); | |
349 | } | |
350 | } | |
351 | ||
352 | /* Proper page table entry exists, but no corresponding struct page */ | |
353 | return -EEXIST; | |
354 | } | |
355 | ||
19be0eaf | 356 | /* |
a308c71b PX |
357 | * FOLL_FORCE can write to even unwritable pte's, but only |
358 | * after we've gone through a COW cycle and they are dirty. | |
19be0eaf LT |
359 | */ |
360 | static inline bool can_follow_write_pte(pte_t pte, unsigned int flags) | |
361 | { | |
a308c71b PX |
362 | return pte_write(pte) || |
363 | ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte)); | |
19be0eaf LT |
364 | } |
365 | ||
69e68b4f | 366 | static struct page *follow_page_pte(struct vm_area_struct *vma, |
df06b37f KB |
367 | unsigned long address, pmd_t *pmd, unsigned int flags, |
368 | struct dev_pagemap **pgmap) | |
69e68b4f KS |
369 | { |
370 | struct mm_struct *mm = vma->vm_mm; | |
371 | struct page *page; | |
372 | spinlock_t *ptl; | |
373 | pte_t *ptep, pte; | |
f28d4363 | 374 | int ret; |
4bbd4c77 | 375 | |
eddb1c22 JH |
376 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
377 | if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) == | |
378 | (FOLL_PIN | FOLL_GET))) | |
379 | return ERR_PTR(-EINVAL); | |
69e68b4f | 380 | retry: |
4bbd4c77 | 381 | if (unlikely(pmd_bad(*pmd))) |
69e68b4f | 382 | return no_page_table(vma, flags); |
4bbd4c77 KS |
383 | |
384 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); | |
4bbd4c77 KS |
385 | pte = *ptep; |
386 | if (!pte_present(pte)) { | |
387 | swp_entry_t entry; | |
388 | /* | |
389 | * KSM's break_ksm() relies upon recognizing a ksm page | |
390 | * even while it is being migrated, so for that case we | |
391 | * need migration_entry_wait(). | |
392 | */ | |
393 | if (likely(!(flags & FOLL_MIGRATION))) | |
394 | goto no_page; | |
0661a336 | 395 | if (pte_none(pte)) |
4bbd4c77 KS |
396 | goto no_page; |
397 | entry = pte_to_swp_entry(pte); | |
398 | if (!is_migration_entry(entry)) | |
399 | goto no_page; | |
400 | pte_unmap_unlock(ptep, ptl); | |
401 | migration_entry_wait(mm, pmd, address); | |
69e68b4f | 402 | goto retry; |
4bbd4c77 | 403 | } |
8a0516ed | 404 | if ((flags & FOLL_NUMA) && pte_protnone(pte)) |
4bbd4c77 | 405 | goto no_page; |
19be0eaf | 406 | if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) { |
69e68b4f KS |
407 | pte_unmap_unlock(ptep, ptl); |
408 | return NULL; | |
409 | } | |
4bbd4c77 KS |
410 | |
411 | page = vm_normal_page(vma, address, pte); | |
3faa52c0 | 412 | if (!page && pte_devmap(pte) && (flags & (FOLL_GET | FOLL_PIN))) { |
3565fce3 | 413 | /* |
3faa52c0 JH |
414 | * Only return device mapping pages in the FOLL_GET or FOLL_PIN |
415 | * case since they are only valid while holding the pgmap | |
416 | * reference. | |
3565fce3 | 417 | */ |
df06b37f KB |
418 | *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap); |
419 | if (*pgmap) | |
3565fce3 DW |
420 | page = pte_page(pte); |
421 | else | |
422 | goto no_page; | |
423 | } else if (unlikely(!page)) { | |
1027e443 KS |
424 | if (flags & FOLL_DUMP) { |
425 | /* Avoid special (like zero) pages in core dumps */ | |
426 | page = ERR_PTR(-EFAULT); | |
427 | goto out; | |
428 | } | |
429 | ||
430 | if (is_zero_pfn(pte_pfn(pte))) { | |
431 | page = pte_page(pte); | |
432 | } else { | |
1027e443 KS |
433 | ret = follow_pfn_pte(vma, address, ptep, flags); |
434 | page = ERR_PTR(ret); | |
435 | goto out; | |
436 | } | |
4bbd4c77 KS |
437 | } |
438 | ||
6742d293 | 439 | if (flags & FOLL_SPLIT && PageTransCompound(page)) { |
6742d293 KS |
440 | get_page(page); |
441 | pte_unmap_unlock(ptep, ptl); | |
442 | lock_page(page); | |
443 | ret = split_huge_page(page); | |
444 | unlock_page(page); | |
445 | put_page(page); | |
446 | if (ret) | |
447 | return ERR_PTR(ret); | |
448 | goto retry; | |
449 | } | |
450 | ||
3faa52c0 JH |
451 | /* try_grab_page() does nothing unless FOLL_GET or FOLL_PIN is set. */ |
452 | if (unlikely(!try_grab_page(page, flags))) { | |
453 | page = ERR_PTR(-ENOMEM); | |
454 | goto out; | |
8fde12ca | 455 | } |
f28d4363 CI |
456 | /* |
457 | * We need to make the page accessible if and only if we are going | |
458 | * to access its content (the FOLL_PIN case). Please see | |
459 | * Documentation/core-api/pin_user_pages.rst for details. | |
460 | */ | |
461 | if (flags & FOLL_PIN) { | |
462 | ret = arch_make_page_accessible(page); | |
463 | if (ret) { | |
464 | unpin_user_page(page); | |
465 | page = ERR_PTR(ret); | |
466 | goto out; | |
467 | } | |
468 | } | |
4bbd4c77 KS |
469 | if (flags & FOLL_TOUCH) { |
470 | if ((flags & FOLL_WRITE) && | |
471 | !pte_dirty(pte) && !PageDirty(page)) | |
472 | set_page_dirty(page); | |
473 | /* | |
474 | * pte_mkyoung() would be more correct here, but atomic care | |
475 | * is needed to avoid losing the dirty bit: it is easier to use | |
476 | * mark_page_accessed(). | |
477 | */ | |
478 | mark_page_accessed(page); | |
479 | } | |
de60f5f1 | 480 | if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) { |
e90309c9 KS |
481 | /* Do not mlock pte-mapped THP */ |
482 | if (PageTransCompound(page)) | |
483 | goto out; | |
484 | ||
4bbd4c77 KS |
485 | /* |
486 | * The preliminary mapping check is mainly to avoid the | |
487 | * pointless overhead of lock_page on the ZERO_PAGE | |
488 | * which might bounce very badly if there is contention. | |
489 | * | |
490 | * If the page is already locked, we don't need to | |
491 | * handle it now - vmscan will handle it later if and | |
492 | * when it attempts to reclaim the page. | |
493 | */ | |
494 | if (page->mapping && trylock_page(page)) { | |
495 | lru_add_drain(); /* push cached pages to LRU */ | |
496 | /* | |
497 | * Because we lock page here, and migration is | |
498 | * blocked by the pte's page reference, and we | |
499 | * know the page is still mapped, we don't even | |
500 | * need to check for file-cache page truncation. | |
501 | */ | |
502 | mlock_vma_page(page); | |
503 | unlock_page(page); | |
504 | } | |
505 | } | |
1027e443 | 506 | out: |
4bbd4c77 | 507 | pte_unmap_unlock(ptep, ptl); |
4bbd4c77 | 508 | return page; |
4bbd4c77 KS |
509 | no_page: |
510 | pte_unmap_unlock(ptep, ptl); | |
511 | if (!pte_none(pte)) | |
69e68b4f KS |
512 | return NULL; |
513 | return no_page_table(vma, flags); | |
514 | } | |
515 | ||
080dbb61 AK |
516 | static struct page *follow_pmd_mask(struct vm_area_struct *vma, |
517 | unsigned long address, pud_t *pudp, | |
df06b37f KB |
518 | unsigned int flags, |
519 | struct follow_page_context *ctx) | |
69e68b4f | 520 | { |
68827280 | 521 | pmd_t *pmd, pmdval; |
69e68b4f KS |
522 | spinlock_t *ptl; |
523 | struct page *page; | |
524 | struct mm_struct *mm = vma->vm_mm; | |
525 | ||
080dbb61 | 526 | pmd = pmd_offset(pudp, address); |
68827280 HY |
527 | /* |
528 | * The READ_ONCE() will stabilize the pmdval in a register or | |
529 | * on the stack so that it will stop changing under the code. | |
530 | */ | |
531 | pmdval = READ_ONCE(*pmd); | |
532 | if (pmd_none(pmdval)) | |
69e68b4f | 533 | return no_page_table(vma, flags); |
be9d3045 | 534 | if (pmd_huge(pmdval) && is_vm_hugetlb_page(vma)) { |
e66f17ff NH |
535 | page = follow_huge_pmd(mm, address, pmd, flags); |
536 | if (page) | |
537 | return page; | |
538 | return no_page_table(vma, flags); | |
69e68b4f | 539 | } |
68827280 | 540 | if (is_hugepd(__hugepd(pmd_val(pmdval)))) { |
4dc71451 | 541 | page = follow_huge_pd(vma, address, |
68827280 | 542 | __hugepd(pmd_val(pmdval)), flags, |
4dc71451 AK |
543 | PMD_SHIFT); |
544 | if (page) | |
545 | return page; | |
546 | return no_page_table(vma, flags); | |
547 | } | |
84c3fc4e | 548 | retry: |
68827280 | 549 | if (!pmd_present(pmdval)) { |
84c3fc4e ZY |
550 | if (likely(!(flags & FOLL_MIGRATION))) |
551 | return no_page_table(vma, flags); | |
552 | VM_BUG_ON(thp_migration_supported() && | |
68827280 HY |
553 | !is_pmd_migration_entry(pmdval)); |
554 | if (is_pmd_migration_entry(pmdval)) | |
84c3fc4e | 555 | pmd_migration_entry_wait(mm, pmd); |
68827280 HY |
556 | pmdval = READ_ONCE(*pmd); |
557 | /* | |
558 | * MADV_DONTNEED may convert the pmd to null because | |
c1e8d7c6 | 559 | * mmap_lock is held in read mode |
68827280 HY |
560 | */ |
561 | if (pmd_none(pmdval)) | |
562 | return no_page_table(vma, flags); | |
84c3fc4e ZY |
563 | goto retry; |
564 | } | |
68827280 | 565 | if (pmd_devmap(pmdval)) { |
3565fce3 | 566 | ptl = pmd_lock(mm, pmd); |
df06b37f | 567 | page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap); |
3565fce3 DW |
568 | spin_unlock(ptl); |
569 | if (page) | |
570 | return page; | |
571 | } | |
68827280 | 572 | if (likely(!pmd_trans_huge(pmdval))) |
df06b37f | 573 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 574 | |
68827280 | 575 | if ((flags & FOLL_NUMA) && pmd_protnone(pmdval)) |
db08f203 AK |
576 | return no_page_table(vma, flags); |
577 | ||
84c3fc4e | 578 | retry_locked: |
6742d293 | 579 | ptl = pmd_lock(mm, pmd); |
68827280 HY |
580 | if (unlikely(pmd_none(*pmd))) { |
581 | spin_unlock(ptl); | |
582 | return no_page_table(vma, flags); | |
583 | } | |
84c3fc4e ZY |
584 | if (unlikely(!pmd_present(*pmd))) { |
585 | spin_unlock(ptl); | |
586 | if (likely(!(flags & FOLL_MIGRATION))) | |
587 | return no_page_table(vma, flags); | |
588 | pmd_migration_entry_wait(mm, pmd); | |
589 | goto retry_locked; | |
590 | } | |
6742d293 KS |
591 | if (unlikely(!pmd_trans_huge(*pmd))) { |
592 | spin_unlock(ptl); | |
df06b37f | 593 | return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
6742d293 | 594 | } |
bfe7b00d | 595 | if (flags & (FOLL_SPLIT | FOLL_SPLIT_PMD)) { |
6742d293 KS |
596 | int ret; |
597 | page = pmd_page(*pmd); | |
598 | if (is_huge_zero_page(page)) { | |
599 | spin_unlock(ptl); | |
600 | ret = 0; | |
78ddc534 | 601 | split_huge_pmd(vma, pmd, address); |
337d9abf NH |
602 | if (pmd_trans_unstable(pmd)) |
603 | ret = -EBUSY; | |
bfe7b00d | 604 | } else if (flags & FOLL_SPLIT) { |
8fde12ca LT |
605 | if (unlikely(!try_get_page(page))) { |
606 | spin_unlock(ptl); | |
607 | return ERR_PTR(-ENOMEM); | |
608 | } | |
69e68b4f | 609 | spin_unlock(ptl); |
6742d293 KS |
610 | lock_page(page); |
611 | ret = split_huge_page(page); | |
612 | unlock_page(page); | |
613 | put_page(page); | |
baa355fd KS |
614 | if (pmd_none(*pmd)) |
615 | return no_page_table(vma, flags); | |
bfe7b00d SL |
616 | } else { /* flags & FOLL_SPLIT_PMD */ |
617 | spin_unlock(ptl); | |
618 | split_huge_pmd(vma, pmd, address); | |
619 | ret = pte_alloc(mm, pmd) ? -ENOMEM : 0; | |
6742d293 KS |
620 | } |
621 | ||
622 | return ret ? ERR_PTR(ret) : | |
df06b37f | 623 | follow_page_pte(vma, address, pmd, flags, &ctx->pgmap); |
69e68b4f | 624 | } |
6742d293 KS |
625 | page = follow_trans_huge_pmd(vma, address, pmd, flags); |
626 | spin_unlock(ptl); | |
df06b37f | 627 | ctx->page_mask = HPAGE_PMD_NR - 1; |
6742d293 | 628 | return page; |
4bbd4c77 KS |
629 | } |
630 | ||
080dbb61 AK |
631 | static struct page *follow_pud_mask(struct vm_area_struct *vma, |
632 | unsigned long address, p4d_t *p4dp, | |
df06b37f KB |
633 | unsigned int flags, |
634 | struct follow_page_context *ctx) | |
080dbb61 AK |
635 | { |
636 | pud_t *pud; | |
637 | spinlock_t *ptl; | |
638 | struct page *page; | |
639 | struct mm_struct *mm = vma->vm_mm; | |
640 | ||
641 | pud = pud_offset(p4dp, address); | |
642 | if (pud_none(*pud)) | |
643 | return no_page_table(vma, flags); | |
be9d3045 | 644 | if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) { |
080dbb61 AK |
645 | page = follow_huge_pud(mm, address, pud, flags); |
646 | if (page) | |
647 | return page; | |
648 | return no_page_table(vma, flags); | |
649 | } | |
4dc71451 AK |
650 | if (is_hugepd(__hugepd(pud_val(*pud)))) { |
651 | page = follow_huge_pd(vma, address, | |
652 | __hugepd(pud_val(*pud)), flags, | |
653 | PUD_SHIFT); | |
654 | if (page) | |
655 | return page; | |
656 | return no_page_table(vma, flags); | |
657 | } | |
080dbb61 AK |
658 | if (pud_devmap(*pud)) { |
659 | ptl = pud_lock(mm, pud); | |
df06b37f | 660 | page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap); |
080dbb61 AK |
661 | spin_unlock(ptl); |
662 | if (page) | |
663 | return page; | |
664 | } | |
665 | if (unlikely(pud_bad(*pud))) | |
666 | return no_page_table(vma, flags); | |
667 | ||
df06b37f | 668 | return follow_pmd_mask(vma, address, pud, flags, ctx); |
080dbb61 AK |
669 | } |
670 | ||
080dbb61 AK |
671 | static struct page *follow_p4d_mask(struct vm_area_struct *vma, |
672 | unsigned long address, pgd_t *pgdp, | |
df06b37f KB |
673 | unsigned int flags, |
674 | struct follow_page_context *ctx) | |
080dbb61 AK |
675 | { |
676 | p4d_t *p4d; | |
4dc71451 | 677 | struct page *page; |
080dbb61 AK |
678 | |
679 | p4d = p4d_offset(pgdp, address); | |
680 | if (p4d_none(*p4d)) | |
681 | return no_page_table(vma, flags); | |
682 | BUILD_BUG_ON(p4d_huge(*p4d)); | |
683 | if (unlikely(p4d_bad(*p4d))) | |
684 | return no_page_table(vma, flags); | |
685 | ||
4dc71451 AK |
686 | if (is_hugepd(__hugepd(p4d_val(*p4d)))) { |
687 | page = follow_huge_pd(vma, address, | |
688 | __hugepd(p4d_val(*p4d)), flags, | |
689 | P4D_SHIFT); | |
690 | if (page) | |
691 | return page; | |
692 | return no_page_table(vma, flags); | |
693 | } | |
df06b37f | 694 | return follow_pud_mask(vma, address, p4d, flags, ctx); |
080dbb61 AK |
695 | } |
696 | ||
697 | /** | |
698 | * follow_page_mask - look up a page descriptor from a user-virtual address | |
699 | * @vma: vm_area_struct mapping @address | |
700 | * @address: virtual address to look up | |
701 | * @flags: flags modifying lookup behaviour | |
78179556 MR |
702 | * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a |
703 | * pointer to output page_mask | |
080dbb61 AK |
704 | * |
705 | * @flags can have FOLL_ flags set, defined in <linux/mm.h> | |
706 | * | |
78179556 MR |
707 | * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches |
708 | * the device's dev_pagemap metadata to avoid repeating expensive lookups. | |
709 | * | |
710 | * On output, the @ctx->page_mask is set according to the size of the page. | |
711 | * | |
712 | * Return: the mapped (struct page *), %NULL if no mapping exists, or | |
080dbb61 AK |
713 | * an error pointer if there is a mapping to something not represented |
714 | * by a page descriptor (see also vm_normal_page()). | |
715 | */ | |
a7030aea | 716 | static struct page *follow_page_mask(struct vm_area_struct *vma, |
080dbb61 | 717 | unsigned long address, unsigned int flags, |
df06b37f | 718 | struct follow_page_context *ctx) |
080dbb61 AK |
719 | { |
720 | pgd_t *pgd; | |
721 | struct page *page; | |
722 | struct mm_struct *mm = vma->vm_mm; | |
723 | ||
df06b37f | 724 | ctx->page_mask = 0; |
080dbb61 AK |
725 | |
726 | /* make this handle hugepd */ | |
727 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); | |
728 | if (!IS_ERR(page)) { | |
3faa52c0 | 729 | WARN_ON_ONCE(flags & (FOLL_GET | FOLL_PIN)); |
080dbb61 AK |
730 | return page; |
731 | } | |
732 | ||
733 | pgd = pgd_offset(mm, address); | |
734 | ||
735 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) | |
736 | return no_page_table(vma, flags); | |
737 | ||
faaa5b62 AK |
738 | if (pgd_huge(*pgd)) { |
739 | page = follow_huge_pgd(mm, address, pgd, flags); | |
740 | if (page) | |
741 | return page; | |
742 | return no_page_table(vma, flags); | |
743 | } | |
4dc71451 AK |
744 | if (is_hugepd(__hugepd(pgd_val(*pgd)))) { |
745 | page = follow_huge_pd(vma, address, | |
746 | __hugepd(pgd_val(*pgd)), flags, | |
747 | PGDIR_SHIFT); | |
748 | if (page) | |
749 | return page; | |
750 | return no_page_table(vma, flags); | |
751 | } | |
faaa5b62 | 752 | |
df06b37f KB |
753 | return follow_p4d_mask(vma, address, pgd, flags, ctx); |
754 | } | |
755 | ||
756 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, | |
757 | unsigned int foll_flags) | |
758 | { | |
759 | struct follow_page_context ctx = { NULL }; | |
760 | struct page *page; | |
761 | ||
762 | page = follow_page_mask(vma, address, foll_flags, &ctx); | |
763 | if (ctx.pgmap) | |
764 | put_dev_pagemap(ctx.pgmap); | |
765 | return page; | |
080dbb61 AK |
766 | } |
767 | ||
f2b495ca KS |
768 | static int get_gate_page(struct mm_struct *mm, unsigned long address, |
769 | unsigned int gup_flags, struct vm_area_struct **vma, | |
770 | struct page **page) | |
771 | { | |
772 | pgd_t *pgd; | |
c2febafc | 773 | p4d_t *p4d; |
f2b495ca KS |
774 | pud_t *pud; |
775 | pmd_t *pmd; | |
776 | pte_t *pte; | |
777 | int ret = -EFAULT; | |
778 | ||
779 | /* user gate pages are read-only */ | |
780 | if (gup_flags & FOLL_WRITE) | |
781 | return -EFAULT; | |
782 | if (address > TASK_SIZE) | |
783 | pgd = pgd_offset_k(address); | |
784 | else | |
785 | pgd = pgd_offset_gate(mm, address); | |
b5d1c39f AL |
786 | if (pgd_none(*pgd)) |
787 | return -EFAULT; | |
c2febafc | 788 | p4d = p4d_offset(pgd, address); |
b5d1c39f AL |
789 | if (p4d_none(*p4d)) |
790 | return -EFAULT; | |
c2febafc | 791 | pud = pud_offset(p4d, address); |
b5d1c39f AL |
792 | if (pud_none(*pud)) |
793 | return -EFAULT; | |
f2b495ca | 794 | pmd = pmd_offset(pud, address); |
84c3fc4e | 795 | if (!pmd_present(*pmd)) |
f2b495ca KS |
796 | return -EFAULT; |
797 | VM_BUG_ON(pmd_trans_huge(*pmd)); | |
798 | pte = pte_offset_map(pmd, address); | |
799 | if (pte_none(*pte)) | |
800 | goto unmap; | |
801 | *vma = get_gate_vma(mm); | |
802 | if (!page) | |
803 | goto out; | |
804 | *page = vm_normal_page(*vma, address, *pte); | |
805 | if (!*page) { | |
806 | if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte))) | |
807 | goto unmap; | |
808 | *page = pte_page(*pte); | |
809 | } | |
9fa2dd94 | 810 | if (unlikely(!try_grab_page(*page, gup_flags))) { |
8fde12ca LT |
811 | ret = -ENOMEM; |
812 | goto unmap; | |
813 | } | |
f2b495ca KS |
814 | out: |
815 | ret = 0; | |
816 | unmap: | |
817 | pte_unmap(pte); | |
818 | return ret; | |
819 | } | |
820 | ||
9a95f3cf | 821 | /* |
c1e8d7c6 ML |
822 | * mmap_lock must be held on entry. If @locked != NULL and *@flags |
823 | * does not include FOLL_NOWAIT, the mmap_lock may be released. If it | |
4f6da934 | 824 | * is, *@locked will be set to 0 and -EBUSY returned. |
9a95f3cf | 825 | */ |
64019a2e | 826 | static int faultin_page(struct vm_area_struct *vma, |
4f6da934 | 827 | unsigned long address, unsigned int *flags, int *locked) |
16744483 | 828 | { |
16744483 | 829 | unsigned int fault_flags = 0; |
2b740303 | 830 | vm_fault_t ret; |
16744483 | 831 | |
de60f5f1 EM |
832 | /* mlock all present pages, but do not fault in new pages */ |
833 | if ((*flags & (FOLL_POPULATE | FOLL_MLOCK)) == FOLL_MLOCK) | |
834 | return -ENOENT; | |
16744483 KS |
835 | if (*flags & FOLL_WRITE) |
836 | fault_flags |= FAULT_FLAG_WRITE; | |
1b2ee126 DH |
837 | if (*flags & FOLL_REMOTE) |
838 | fault_flags |= FAULT_FLAG_REMOTE; | |
4f6da934 | 839 | if (locked) |
71335f37 | 840 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
16744483 KS |
841 | if (*flags & FOLL_NOWAIT) |
842 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT; | |
234b239b | 843 | if (*flags & FOLL_TRIED) { |
4426e945 PX |
844 | /* |
845 | * Note: FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_TRIED | |
846 | * can co-exist | |
847 | */ | |
234b239b ALC |
848 | fault_flags |= FAULT_FLAG_TRIED; |
849 | } | |
16744483 | 850 | |
bce617ed | 851 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
16744483 | 852 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
853 | int err = vm_fault_to_errno(ret, *flags); |
854 | ||
855 | if (err) | |
856 | return err; | |
16744483 KS |
857 | BUG(); |
858 | } | |
859 | ||
16744483 | 860 | if (ret & VM_FAULT_RETRY) { |
4f6da934 PX |
861 | if (locked && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT)) |
862 | *locked = 0; | |
16744483 KS |
863 | return -EBUSY; |
864 | } | |
865 | ||
866 | /* | |
867 | * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when | |
868 | * necessary, even if maybe_mkwrite decided not to set pte_write. We | |
869 | * can thus safely do subsequent page lookups as if they were reads. | |
870 | * But only do so when looping for pte_write is futile: in some cases | |
871 | * userspace may also be wanting to write to the gotten user page, | |
872 | * which a read fault here might prevent (a readonly page might get | |
873 | * reCOWed by userspace write). | |
874 | */ | |
875 | if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE)) | |
2923117b | 876 | *flags |= FOLL_COW; |
16744483 KS |
877 | return 0; |
878 | } | |
879 | ||
fa5bb209 KS |
880 | static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags) |
881 | { | |
882 | vm_flags_t vm_flags = vma->vm_flags; | |
1b2ee126 DH |
883 | int write = (gup_flags & FOLL_WRITE); |
884 | int foreign = (gup_flags & FOLL_REMOTE); | |
fa5bb209 KS |
885 | |
886 | if (vm_flags & (VM_IO | VM_PFNMAP)) | |
887 | return -EFAULT; | |
888 | ||
7f7ccc2c WT |
889 | if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma)) |
890 | return -EFAULT; | |
891 | ||
52650c8b JG |
892 | if ((gup_flags & FOLL_LONGTERM) && vma_is_fsdax(vma)) |
893 | return -EOPNOTSUPP; | |
894 | ||
1b2ee126 | 895 | if (write) { |
fa5bb209 KS |
896 | if (!(vm_flags & VM_WRITE)) { |
897 | if (!(gup_flags & FOLL_FORCE)) | |
898 | return -EFAULT; | |
899 | /* | |
900 | * We used to let the write,force case do COW in a | |
901 | * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could | |
902 | * set a breakpoint in a read-only mapping of an | |
903 | * executable, without corrupting the file (yet only | |
904 | * when that file had been opened for writing!). | |
905 | * Anon pages in shared mappings are surprising: now | |
906 | * just reject it. | |
907 | */ | |
46435364 | 908 | if (!is_cow_mapping(vm_flags)) |
fa5bb209 | 909 | return -EFAULT; |
fa5bb209 KS |
910 | } |
911 | } else if (!(vm_flags & VM_READ)) { | |
912 | if (!(gup_flags & FOLL_FORCE)) | |
913 | return -EFAULT; | |
914 | /* | |
915 | * Is there actually any vma we can reach here which does not | |
916 | * have VM_MAYREAD set? | |
917 | */ | |
918 | if (!(vm_flags & VM_MAYREAD)) | |
919 | return -EFAULT; | |
920 | } | |
d61172b4 DH |
921 | /* |
922 | * gups are always data accesses, not instruction | |
923 | * fetches, so execute=false here | |
924 | */ | |
925 | if (!arch_vma_access_permitted(vma, write, false, foreign)) | |
33a709b2 | 926 | return -EFAULT; |
fa5bb209 KS |
927 | return 0; |
928 | } | |
929 | ||
4bbd4c77 KS |
930 | /** |
931 | * __get_user_pages() - pin user pages in memory | |
4bbd4c77 KS |
932 | * @mm: mm_struct of target mm |
933 | * @start: starting user address | |
934 | * @nr_pages: number of pages from start to pin | |
935 | * @gup_flags: flags modifying pin behaviour | |
936 | * @pages: array that receives pointers to the pages pinned. | |
937 | * Should be at least nr_pages long. Or NULL, if caller | |
938 | * only intends to ensure the pages are faulted in. | |
939 | * @vmas: array of pointers to vmas corresponding to each page. | |
940 | * Or NULL if the caller does not require them. | |
c1e8d7c6 | 941 | * @locked: whether we're still with the mmap_lock held |
4bbd4c77 | 942 | * |
d2dfbe47 LX |
943 | * Returns either number of pages pinned (which may be less than the |
944 | * number requested), or an error. Details about the return value: | |
945 | * | |
946 | * -- If nr_pages is 0, returns 0. | |
947 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
948 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
949 | * pages pinned. Again, this may be less than nr_pages. | |
2d3a36a4 | 950 | * -- 0 return value is possible when the fault would need to be retried. |
d2dfbe47 LX |
951 | * |
952 | * The caller is responsible for releasing returned @pages, via put_page(). | |
953 | * | |
c1e8d7c6 | 954 | * @vmas are valid only as long as mmap_lock is held. |
4bbd4c77 | 955 | * |
c1e8d7c6 | 956 | * Must be called with mmap_lock held. It may be released. See below. |
4bbd4c77 KS |
957 | * |
958 | * __get_user_pages walks a process's page tables and takes a reference to | |
959 | * each struct page that each user address corresponds to at a given | |
960 | * instant. That is, it takes the page that would be accessed if a user | |
961 | * thread accesses the given user virtual address at that instant. | |
962 | * | |
963 | * This does not guarantee that the page exists in the user mappings when | |
964 | * __get_user_pages returns, and there may even be a completely different | |
965 | * page there in some cases (eg. if mmapped pagecache has been invalidated | |
966 | * and subsequently re faulted). However it does guarantee that the page | |
967 | * won't be freed completely. And mostly callers simply care that the page | |
968 | * contains data that was valid *at some point in time*. Typically, an IO | |
969 | * or similar operation cannot guarantee anything stronger anyway because | |
970 | * locks can't be held over the syscall boundary. | |
971 | * | |
972 | * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If | |
973 | * the page is written to, set_page_dirty (or set_page_dirty_lock, as | |
974 | * appropriate) must be called after the page is finished with, and | |
975 | * before put_page is called. | |
976 | * | |
c1e8d7c6 | 977 | * If @locked != NULL, *@locked will be set to 0 when mmap_lock is |
4f6da934 PX |
978 | * released by an up_read(). That can happen if @gup_flags does not |
979 | * have FOLL_NOWAIT. | |
9a95f3cf | 980 | * |
4f6da934 | 981 | * A caller using such a combination of @locked and @gup_flags |
c1e8d7c6 | 982 | * must therefore hold the mmap_lock for reading only, and recognize |
9a95f3cf PC |
983 | * when it's been released. Otherwise, it must be held for either |
984 | * reading or writing and will not be released. | |
4bbd4c77 KS |
985 | * |
986 | * In most cases, get_user_pages or get_user_pages_fast should be used | |
987 | * instead of __get_user_pages. __get_user_pages should be used only if | |
988 | * you need some special @gup_flags. | |
989 | */ | |
64019a2e | 990 | static long __get_user_pages(struct mm_struct *mm, |
4bbd4c77 KS |
991 | unsigned long start, unsigned long nr_pages, |
992 | unsigned int gup_flags, struct page **pages, | |
4f6da934 | 993 | struct vm_area_struct **vmas, int *locked) |
4bbd4c77 | 994 | { |
df06b37f | 995 | long ret = 0, i = 0; |
fa5bb209 | 996 | struct vm_area_struct *vma = NULL; |
df06b37f | 997 | struct follow_page_context ctx = { NULL }; |
4bbd4c77 KS |
998 | |
999 | if (!nr_pages) | |
1000 | return 0; | |
1001 | ||
f9652594 AK |
1002 | start = untagged_addr(start); |
1003 | ||
eddb1c22 | 1004 | VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN))); |
4bbd4c77 KS |
1005 | |
1006 | /* | |
1007 | * If FOLL_FORCE is set then do not force a full fault as the hinting | |
1008 | * fault information is unrelated to the reference behaviour of a task | |
1009 | * using the address space | |
1010 | */ | |
1011 | if (!(gup_flags & FOLL_FORCE)) | |
1012 | gup_flags |= FOLL_NUMA; | |
1013 | ||
4bbd4c77 | 1014 | do { |
fa5bb209 KS |
1015 | struct page *page; |
1016 | unsigned int foll_flags = gup_flags; | |
1017 | unsigned int page_increm; | |
1018 | ||
1019 | /* first iteration or cross vma bound */ | |
1020 | if (!vma || start >= vma->vm_end) { | |
1021 | vma = find_extend_vma(mm, start); | |
1022 | if (!vma && in_gate_area(mm, start)) { | |
fa5bb209 KS |
1023 | ret = get_gate_page(mm, start & PAGE_MASK, |
1024 | gup_flags, &vma, | |
1025 | pages ? &pages[i] : NULL); | |
1026 | if (ret) | |
08be37b7 | 1027 | goto out; |
df06b37f | 1028 | ctx.page_mask = 0; |
fa5bb209 KS |
1029 | goto next_page; |
1030 | } | |
4bbd4c77 | 1031 | |
52650c8b | 1032 | if (!vma) { |
df06b37f KB |
1033 | ret = -EFAULT; |
1034 | goto out; | |
1035 | } | |
52650c8b JG |
1036 | ret = check_vma_flags(vma, gup_flags); |
1037 | if (ret) | |
1038 | goto out; | |
1039 | ||
fa5bb209 KS |
1040 | if (is_vm_hugetlb_page(vma)) { |
1041 | i = follow_hugetlb_page(mm, vma, pages, vmas, | |
1042 | &start, &nr_pages, i, | |
a308c71b | 1043 | gup_flags, locked); |
ad415db8 PX |
1044 | if (locked && *locked == 0) { |
1045 | /* | |
1046 | * We've got a VM_FAULT_RETRY | |
c1e8d7c6 | 1047 | * and we've lost mmap_lock. |
ad415db8 PX |
1048 | * We must stop here. |
1049 | */ | |
1050 | BUG_ON(gup_flags & FOLL_NOWAIT); | |
1051 | BUG_ON(ret != 0); | |
1052 | goto out; | |
1053 | } | |
fa5bb209 | 1054 | continue; |
4bbd4c77 | 1055 | } |
fa5bb209 KS |
1056 | } |
1057 | retry: | |
1058 | /* | |
1059 | * If we have a pending SIGKILL, don't keep faulting pages and | |
1060 | * potentially allocating memory. | |
1061 | */ | |
fa45f116 | 1062 | if (fatal_signal_pending(current)) { |
d180870d | 1063 | ret = -EINTR; |
df06b37f KB |
1064 | goto out; |
1065 | } | |
fa5bb209 | 1066 | cond_resched(); |
df06b37f KB |
1067 | |
1068 | page = follow_page_mask(vma, start, foll_flags, &ctx); | |
fa5bb209 | 1069 | if (!page) { |
64019a2e | 1070 | ret = faultin_page(vma, start, &foll_flags, locked); |
fa5bb209 KS |
1071 | switch (ret) { |
1072 | case 0: | |
1073 | goto retry; | |
df06b37f KB |
1074 | case -EBUSY: |
1075 | ret = 0; | |
e4a9bc58 | 1076 | fallthrough; |
fa5bb209 KS |
1077 | case -EFAULT: |
1078 | case -ENOMEM: | |
1079 | case -EHWPOISON: | |
df06b37f | 1080 | goto out; |
fa5bb209 KS |
1081 | case -ENOENT: |
1082 | goto next_page; | |
4bbd4c77 | 1083 | } |
fa5bb209 | 1084 | BUG(); |
1027e443 KS |
1085 | } else if (PTR_ERR(page) == -EEXIST) { |
1086 | /* | |
1087 | * Proper page table entry exists, but no corresponding | |
1088 | * struct page. | |
1089 | */ | |
1090 | goto next_page; | |
1091 | } else if (IS_ERR(page)) { | |
df06b37f KB |
1092 | ret = PTR_ERR(page); |
1093 | goto out; | |
1027e443 | 1094 | } |
fa5bb209 KS |
1095 | if (pages) { |
1096 | pages[i] = page; | |
1097 | flush_anon_page(vma, page, start); | |
1098 | flush_dcache_page(page); | |
df06b37f | 1099 | ctx.page_mask = 0; |
4bbd4c77 | 1100 | } |
4bbd4c77 | 1101 | next_page: |
fa5bb209 KS |
1102 | if (vmas) { |
1103 | vmas[i] = vma; | |
df06b37f | 1104 | ctx.page_mask = 0; |
fa5bb209 | 1105 | } |
df06b37f | 1106 | page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask); |
fa5bb209 KS |
1107 | if (page_increm > nr_pages) |
1108 | page_increm = nr_pages; | |
1109 | i += page_increm; | |
1110 | start += page_increm * PAGE_SIZE; | |
1111 | nr_pages -= page_increm; | |
4bbd4c77 | 1112 | } while (nr_pages); |
df06b37f KB |
1113 | out: |
1114 | if (ctx.pgmap) | |
1115 | put_dev_pagemap(ctx.pgmap); | |
1116 | return i ? i : ret; | |
4bbd4c77 | 1117 | } |
4bbd4c77 | 1118 | |
771ab430 TK |
1119 | static bool vma_permits_fault(struct vm_area_struct *vma, |
1120 | unsigned int fault_flags) | |
d4925e00 | 1121 | { |
1b2ee126 DH |
1122 | bool write = !!(fault_flags & FAULT_FLAG_WRITE); |
1123 | bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE); | |
33a709b2 | 1124 | vm_flags_t vm_flags = write ? VM_WRITE : VM_READ; |
d4925e00 DH |
1125 | |
1126 | if (!(vm_flags & vma->vm_flags)) | |
1127 | return false; | |
1128 | ||
33a709b2 DH |
1129 | /* |
1130 | * The architecture might have a hardware protection | |
1b2ee126 | 1131 | * mechanism other than read/write that can deny access. |
d61172b4 DH |
1132 | * |
1133 | * gup always represents data access, not instruction | |
1134 | * fetches, so execute=false here: | |
33a709b2 | 1135 | */ |
d61172b4 | 1136 | if (!arch_vma_access_permitted(vma, write, false, foreign)) |
33a709b2 DH |
1137 | return false; |
1138 | ||
d4925e00 DH |
1139 | return true; |
1140 | } | |
1141 | ||
adc8cb40 | 1142 | /** |
4bbd4c77 | 1143 | * fixup_user_fault() - manually resolve a user page fault |
4bbd4c77 KS |
1144 | * @mm: mm_struct of target mm |
1145 | * @address: user address | |
1146 | * @fault_flags:flags to pass down to handle_mm_fault() | |
c1e8d7c6 | 1147 | * @unlocked: did we unlock the mmap_lock while retrying, maybe NULL if caller |
548b6a1e MC |
1148 | * does not allow retry. If NULL, the caller must guarantee |
1149 | * that fault_flags does not contain FAULT_FLAG_ALLOW_RETRY. | |
4bbd4c77 KS |
1150 | * |
1151 | * This is meant to be called in the specific scenario where for locking reasons | |
1152 | * we try to access user memory in atomic context (within a pagefault_disable() | |
1153 | * section), this returns -EFAULT, and we want to resolve the user fault before | |
1154 | * trying again. | |
1155 | * | |
1156 | * Typically this is meant to be used by the futex code. | |
1157 | * | |
1158 | * The main difference with get_user_pages() is that this function will | |
1159 | * unconditionally call handle_mm_fault() which will in turn perform all the | |
1160 | * necessary SW fixup of the dirty and young bits in the PTE, while | |
4a9e1cda | 1161 | * get_user_pages() only guarantees to update these in the struct page. |
4bbd4c77 KS |
1162 | * |
1163 | * This is important for some architectures where those bits also gate the | |
1164 | * access permission to the page because they are maintained in software. On | |
1165 | * such architectures, gup() will not be enough to make a subsequent access | |
1166 | * succeed. | |
1167 | * | |
c1e8d7c6 ML |
1168 | * This function will not return with an unlocked mmap_lock. So it has not the |
1169 | * same semantics wrt the @mm->mmap_lock as does filemap_fault(). | |
4bbd4c77 | 1170 | */ |
64019a2e | 1171 | int fixup_user_fault(struct mm_struct *mm, |
4a9e1cda DD |
1172 | unsigned long address, unsigned int fault_flags, |
1173 | bool *unlocked) | |
4bbd4c77 KS |
1174 | { |
1175 | struct vm_area_struct *vma; | |
2b740303 | 1176 | vm_fault_t ret, major = 0; |
4a9e1cda | 1177 | |
f9652594 AK |
1178 | address = untagged_addr(address); |
1179 | ||
4a9e1cda | 1180 | if (unlocked) |
71335f37 | 1181 | fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; |
4bbd4c77 | 1182 | |
4a9e1cda | 1183 | retry: |
4bbd4c77 KS |
1184 | vma = find_extend_vma(mm, address); |
1185 | if (!vma || address < vma->vm_start) | |
1186 | return -EFAULT; | |
1187 | ||
d4925e00 | 1188 | if (!vma_permits_fault(vma, fault_flags)) |
4bbd4c77 KS |
1189 | return -EFAULT; |
1190 | ||
475f4dfc PX |
1191 | if ((fault_flags & FAULT_FLAG_KILLABLE) && |
1192 | fatal_signal_pending(current)) | |
1193 | return -EINTR; | |
1194 | ||
bce617ed | 1195 | ret = handle_mm_fault(vma, address, fault_flags, NULL); |
4a9e1cda | 1196 | major |= ret & VM_FAULT_MAJOR; |
4bbd4c77 | 1197 | if (ret & VM_FAULT_ERROR) { |
9a291a7c JM |
1198 | int err = vm_fault_to_errno(ret, 0); |
1199 | ||
1200 | if (err) | |
1201 | return err; | |
4bbd4c77 KS |
1202 | BUG(); |
1203 | } | |
4a9e1cda DD |
1204 | |
1205 | if (ret & VM_FAULT_RETRY) { | |
d8ed45c5 | 1206 | mmap_read_lock(mm); |
475f4dfc PX |
1207 | *unlocked = true; |
1208 | fault_flags |= FAULT_FLAG_TRIED; | |
1209 | goto retry; | |
4a9e1cda DD |
1210 | } |
1211 | ||
4bbd4c77 KS |
1212 | return 0; |
1213 | } | |
add6a0cd | 1214 | EXPORT_SYMBOL_GPL(fixup_user_fault); |
4bbd4c77 | 1215 | |
2d3a36a4 MH |
1216 | /* |
1217 | * Please note that this function, unlike __get_user_pages will not | |
1218 | * return 0 for nr_pages > 0 without FOLL_NOWAIT | |
1219 | */ | |
64019a2e | 1220 | static __always_inline long __get_user_pages_locked(struct mm_struct *mm, |
f0818f47 AA |
1221 | unsigned long start, |
1222 | unsigned long nr_pages, | |
f0818f47 AA |
1223 | struct page **pages, |
1224 | struct vm_area_struct **vmas, | |
e716712f | 1225 | int *locked, |
0fd71a56 | 1226 | unsigned int flags) |
f0818f47 | 1227 | { |
f0818f47 AA |
1228 | long ret, pages_done; |
1229 | bool lock_dropped; | |
1230 | ||
1231 | if (locked) { | |
1232 | /* if VM_FAULT_RETRY can be returned, vmas become invalid */ | |
1233 | BUG_ON(vmas); | |
1234 | /* check caller initialized locked */ | |
1235 | BUG_ON(*locked != 1); | |
1236 | } | |
1237 | ||
008cfe44 | 1238 | if (flags & FOLL_PIN) |
a4d63c37 | 1239 | atomic_set(&mm->has_pinned, 1); |
008cfe44 | 1240 | |
eddb1c22 JH |
1241 | /* |
1242 | * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior | |
1243 | * is to set FOLL_GET if the caller wants pages[] filled in (but has | |
1244 | * carelessly failed to specify FOLL_GET), so keep doing that, but only | |
1245 | * for FOLL_GET, not for the newer FOLL_PIN. | |
1246 | * | |
1247 | * FOLL_PIN always expects pages to be non-null, but no need to assert | |
1248 | * that here, as any failures will be obvious enough. | |
1249 | */ | |
1250 | if (pages && !(flags & FOLL_PIN)) | |
f0818f47 | 1251 | flags |= FOLL_GET; |
f0818f47 AA |
1252 | |
1253 | pages_done = 0; | |
1254 | lock_dropped = false; | |
1255 | for (;;) { | |
64019a2e | 1256 | ret = __get_user_pages(mm, start, nr_pages, flags, pages, |
f0818f47 AA |
1257 | vmas, locked); |
1258 | if (!locked) | |
1259 | /* VM_FAULT_RETRY couldn't trigger, bypass */ | |
1260 | return ret; | |
1261 | ||
1262 | /* VM_FAULT_RETRY cannot return errors */ | |
1263 | if (!*locked) { | |
1264 | BUG_ON(ret < 0); | |
1265 | BUG_ON(ret >= nr_pages); | |
1266 | } | |
1267 | ||
f0818f47 AA |
1268 | if (ret > 0) { |
1269 | nr_pages -= ret; | |
1270 | pages_done += ret; | |
1271 | if (!nr_pages) | |
1272 | break; | |
1273 | } | |
1274 | if (*locked) { | |
96312e61 AA |
1275 | /* |
1276 | * VM_FAULT_RETRY didn't trigger or it was a | |
1277 | * FOLL_NOWAIT. | |
1278 | */ | |
f0818f47 AA |
1279 | if (!pages_done) |
1280 | pages_done = ret; | |
1281 | break; | |
1282 | } | |
df17277b MR |
1283 | /* |
1284 | * VM_FAULT_RETRY triggered, so seek to the faulting offset. | |
1285 | * For the prefault case (!pages) we only update counts. | |
1286 | */ | |
1287 | if (likely(pages)) | |
1288 | pages += ret; | |
f0818f47 | 1289 | start += ret << PAGE_SHIFT; |
4426e945 | 1290 | lock_dropped = true; |
f0818f47 | 1291 | |
4426e945 | 1292 | retry: |
f0818f47 AA |
1293 | /* |
1294 | * Repeat on the address that fired VM_FAULT_RETRY | |
4426e945 PX |
1295 | * with both FAULT_FLAG_ALLOW_RETRY and |
1296 | * FAULT_FLAG_TRIED. Note that GUP can be interrupted | |
1297 | * by fatal signals, so we need to check it before we | |
1298 | * start trying again otherwise it can loop forever. | |
f0818f47 | 1299 | */ |
4426e945 | 1300 | |
ae46d2aa HD |
1301 | if (fatal_signal_pending(current)) { |
1302 | if (!pages_done) | |
1303 | pages_done = -EINTR; | |
4426e945 | 1304 | break; |
ae46d2aa | 1305 | } |
4426e945 | 1306 | |
d8ed45c5 | 1307 | ret = mmap_read_lock_killable(mm); |
71335f37 PX |
1308 | if (ret) { |
1309 | BUG_ON(ret > 0); | |
1310 | if (!pages_done) | |
1311 | pages_done = ret; | |
1312 | break; | |
1313 | } | |
4426e945 | 1314 | |
c7b6a566 | 1315 | *locked = 1; |
64019a2e | 1316 | ret = __get_user_pages(mm, start, 1, flags | FOLL_TRIED, |
4426e945 PX |
1317 | pages, NULL, locked); |
1318 | if (!*locked) { | |
1319 | /* Continue to retry until we succeeded */ | |
1320 | BUG_ON(ret != 0); | |
1321 | goto retry; | |
1322 | } | |
f0818f47 AA |
1323 | if (ret != 1) { |
1324 | BUG_ON(ret > 1); | |
1325 | if (!pages_done) | |
1326 | pages_done = ret; | |
1327 | break; | |
1328 | } | |
1329 | nr_pages--; | |
1330 | pages_done++; | |
1331 | if (!nr_pages) | |
1332 | break; | |
df17277b MR |
1333 | if (likely(pages)) |
1334 | pages++; | |
f0818f47 AA |
1335 | start += PAGE_SIZE; |
1336 | } | |
e716712f | 1337 | if (lock_dropped && *locked) { |
f0818f47 AA |
1338 | /* |
1339 | * We must let the caller know we temporarily dropped the lock | |
1340 | * and so the critical section protected by it was lost. | |
1341 | */ | |
d8ed45c5 | 1342 | mmap_read_unlock(mm); |
f0818f47 AA |
1343 | *locked = 0; |
1344 | } | |
1345 | return pages_done; | |
1346 | } | |
1347 | ||
d3649f68 CH |
1348 | /** |
1349 | * populate_vma_page_range() - populate a range of pages in the vma. | |
1350 | * @vma: target vma | |
1351 | * @start: start address | |
1352 | * @end: end address | |
c1e8d7c6 | 1353 | * @locked: whether the mmap_lock is still held |
d3649f68 CH |
1354 | * |
1355 | * This takes care of mlocking the pages too if VM_LOCKED is set. | |
1356 | * | |
0a36f7f8 TY |
1357 | * Return either number of pages pinned in the vma, or a negative error |
1358 | * code on error. | |
d3649f68 | 1359 | * |
c1e8d7c6 | 1360 | * vma->vm_mm->mmap_lock must be held. |
d3649f68 | 1361 | * |
4f6da934 | 1362 | * If @locked is NULL, it may be held for read or write and will |
d3649f68 CH |
1363 | * be unperturbed. |
1364 | * | |
4f6da934 PX |
1365 | * If @locked is non-NULL, it must held for read only and may be |
1366 | * released. If it's released, *@locked will be set to 0. | |
d3649f68 CH |
1367 | */ |
1368 | long populate_vma_page_range(struct vm_area_struct *vma, | |
4f6da934 | 1369 | unsigned long start, unsigned long end, int *locked) |
d3649f68 CH |
1370 | { |
1371 | struct mm_struct *mm = vma->vm_mm; | |
1372 | unsigned long nr_pages = (end - start) / PAGE_SIZE; | |
1373 | int gup_flags; | |
1374 | ||
1375 | VM_BUG_ON(start & ~PAGE_MASK); | |
1376 | VM_BUG_ON(end & ~PAGE_MASK); | |
1377 | VM_BUG_ON_VMA(start < vma->vm_start, vma); | |
1378 | VM_BUG_ON_VMA(end > vma->vm_end, vma); | |
42fc5414 | 1379 | mmap_assert_locked(mm); |
d3649f68 CH |
1380 | |
1381 | gup_flags = FOLL_TOUCH | FOLL_POPULATE | FOLL_MLOCK; | |
1382 | if (vma->vm_flags & VM_LOCKONFAULT) | |
1383 | gup_flags &= ~FOLL_POPULATE; | |
1384 | /* | |
1385 | * We want to touch writable mappings with a write fault in order | |
1386 | * to break COW, except for shared mappings because these don't COW | |
1387 | * and we would not want to dirty them for nothing. | |
1388 | */ | |
1389 | if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE) | |
1390 | gup_flags |= FOLL_WRITE; | |
1391 | ||
1392 | /* | |
1393 | * We want mlock to succeed for regions that have any permissions | |
1394 | * other than PROT_NONE. | |
1395 | */ | |
3122e80e | 1396 | if (vma_is_accessible(vma)) |
d3649f68 CH |
1397 | gup_flags |= FOLL_FORCE; |
1398 | ||
1399 | /* | |
1400 | * We made sure addr is within a VMA, so the following will | |
1401 | * not result in a stack expansion that recurses back here. | |
1402 | */ | |
64019a2e | 1403 | return __get_user_pages(mm, start, nr_pages, gup_flags, |
4f6da934 | 1404 | NULL, NULL, locked); |
d3649f68 CH |
1405 | } |
1406 | ||
1407 | /* | |
1408 | * __mm_populate - populate and/or mlock pages within a range of address space. | |
1409 | * | |
1410 | * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap | |
1411 | * flags. VMAs must be already marked with the desired vm_flags, and | |
c1e8d7c6 | 1412 | * mmap_lock must not be held. |
d3649f68 CH |
1413 | */ |
1414 | int __mm_populate(unsigned long start, unsigned long len, int ignore_errors) | |
1415 | { | |
1416 | struct mm_struct *mm = current->mm; | |
1417 | unsigned long end, nstart, nend; | |
1418 | struct vm_area_struct *vma = NULL; | |
1419 | int locked = 0; | |
1420 | long ret = 0; | |
1421 | ||
1422 | end = start + len; | |
1423 | ||
1424 | for (nstart = start; nstart < end; nstart = nend) { | |
1425 | /* | |
1426 | * We want to fault in pages for [nstart; end) address range. | |
1427 | * Find first corresponding VMA. | |
1428 | */ | |
1429 | if (!locked) { | |
1430 | locked = 1; | |
d8ed45c5 | 1431 | mmap_read_lock(mm); |
d3649f68 CH |
1432 | vma = find_vma(mm, nstart); |
1433 | } else if (nstart >= vma->vm_end) | |
1434 | vma = vma->vm_next; | |
1435 | if (!vma || vma->vm_start >= end) | |
1436 | break; | |
1437 | /* | |
1438 | * Set [nstart; nend) to intersection of desired address | |
1439 | * range with the first VMA. Also, skip undesirable VMA types. | |
1440 | */ | |
1441 | nend = min(end, vma->vm_end); | |
1442 | if (vma->vm_flags & (VM_IO | VM_PFNMAP)) | |
1443 | continue; | |
1444 | if (nstart < vma->vm_start) | |
1445 | nstart = vma->vm_start; | |
1446 | /* | |
1447 | * Now fault in a range of pages. populate_vma_page_range() | |
1448 | * double checks the vma flags, so that it won't mlock pages | |
1449 | * if the vma was already munlocked. | |
1450 | */ | |
1451 | ret = populate_vma_page_range(vma, nstart, nend, &locked); | |
1452 | if (ret < 0) { | |
1453 | if (ignore_errors) { | |
1454 | ret = 0; | |
1455 | continue; /* continue at next VMA */ | |
1456 | } | |
1457 | break; | |
1458 | } | |
1459 | nend = nstart + ret * PAGE_SIZE; | |
1460 | ret = 0; | |
1461 | } | |
1462 | if (locked) | |
d8ed45c5 | 1463 | mmap_read_unlock(mm); |
d3649f68 CH |
1464 | return ret; /* 0 or negative error code */ |
1465 | } | |
050a9adc | 1466 | #else /* CONFIG_MMU */ |
64019a2e | 1467 | static long __get_user_pages_locked(struct mm_struct *mm, unsigned long start, |
050a9adc CH |
1468 | unsigned long nr_pages, struct page **pages, |
1469 | struct vm_area_struct **vmas, int *locked, | |
1470 | unsigned int foll_flags) | |
1471 | { | |
1472 | struct vm_area_struct *vma; | |
1473 | unsigned long vm_flags; | |
1474 | int i; | |
1475 | ||
1476 | /* calculate required read or write permissions. | |
1477 | * If FOLL_FORCE is set, we only require the "MAY" flags. | |
1478 | */ | |
1479 | vm_flags = (foll_flags & FOLL_WRITE) ? | |
1480 | (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); | |
1481 | vm_flags &= (foll_flags & FOLL_FORCE) ? | |
1482 | (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); | |
1483 | ||
1484 | for (i = 0; i < nr_pages; i++) { | |
1485 | vma = find_vma(mm, start); | |
1486 | if (!vma) | |
1487 | goto finish_or_fault; | |
1488 | ||
1489 | /* protect what we can, including chardevs */ | |
1490 | if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || | |
1491 | !(vm_flags & vma->vm_flags)) | |
1492 | goto finish_or_fault; | |
1493 | ||
1494 | if (pages) { | |
1495 | pages[i] = virt_to_page(start); | |
1496 | if (pages[i]) | |
1497 | get_page(pages[i]); | |
1498 | } | |
1499 | if (vmas) | |
1500 | vmas[i] = vma; | |
1501 | start = (start + PAGE_SIZE) & PAGE_MASK; | |
1502 | } | |
1503 | ||
1504 | return i; | |
1505 | ||
1506 | finish_or_fault: | |
1507 | return i ? : -EFAULT; | |
1508 | } | |
1509 | #endif /* !CONFIG_MMU */ | |
d3649f68 | 1510 | |
8f942eea JH |
1511 | /** |
1512 | * get_dump_page() - pin user page in memory while writing it to core dump | |
1513 | * @addr: user address | |
1514 | * | |
1515 | * Returns struct page pointer of user page pinned for dump, | |
1516 | * to be freed afterwards by put_page(). | |
1517 | * | |
1518 | * Returns NULL on any kind of failure - a hole must then be inserted into | |
1519 | * the corefile, to preserve alignment with its headers; and also returns | |
1520 | * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found - | |
1521 | * allowing a hole to be left in the corefile to save diskspace. | |
1522 | * | |
7f3bfab5 | 1523 | * Called without mmap_lock (takes and releases the mmap_lock by itself). |
8f942eea JH |
1524 | */ |
1525 | #ifdef CONFIG_ELF_CORE | |
1526 | struct page *get_dump_page(unsigned long addr) | |
1527 | { | |
7f3bfab5 | 1528 | struct mm_struct *mm = current->mm; |
8f942eea | 1529 | struct page *page; |
7f3bfab5 JH |
1530 | int locked = 1; |
1531 | int ret; | |
8f942eea | 1532 | |
7f3bfab5 | 1533 | if (mmap_read_lock_killable(mm)) |
8f942eea | 1534 | return NULL; |
7f3bfab5 JH |
1535 | ret = __get_user_pages_locked(mm, addr, 1, &page, NULL, &locked, |
1536 | FOLL_FORCE | FOLL_DUMP | FOLL_GET); | |
1537 | if (locked) | |
1538 | mmap_read_unlock(mm); | |
1539 | return (ret == 1) ? page : NULL; | |
8f942eea JH |
1540 | } |
1541 | #endif /* CONFIG_ELF_CORE */ | |
1542 | ||
9a4e9f3b | 1543 | #ifdef CONFIG_CMA |
64019a2e | 1544 | static long check_and_migrate_cma_pages(struct mm_struct *mm, |
932f4a63 IW |
1545 | unsigned long start, |
1546 | unsigned long nr_pages, | |
9a4e9f3b | 1547 | struct page **pages, |
932f4a63 IW |
1548 | struct vm_area_struct **vmas, |
1549 | unsigned int gup_flags) | |
9a4e9f3b | 1550 | { |
aa712399 PL |
1551 | unsigned long i; |
1552 | unsigned long step; | |
9a4e9f3b AK |
1553 | bool drain_allow = true; |
1554 | bool migrate_allow = true; | |
1555 | LIST_HEAD(cma_page_list); | |
b96cc655 | 1556 | long ret = nr_pages; |
ed03d924 JK |
1557 | struct migration_target_control mtc = { |
1558 | .nid = NUMA_NO_NODE, | |
1559 | .gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_NOWARN, | |
1560 | }; | |
9a4e9f3b AK |
1561 | |
1562 | check_again: | |
aa712399 PL |
1563 | for (i = 0; i < nr_pages;) { |
1564 | ||
1565 | struct page *head = compound_head(pages[i]); | |
1566 | ||
1567 | /* | |
1568 | * gup may start from a tail page. Advance step by the left | |
1569 | * part. | |
1570 | */ | |
d8c6546b | 1571 | step = compound_nr(head) - (pages[i] - head); |
9a4e9f3b AK |
1572 | /* |
1573 | * If we get a page from the CMA zone, since we are going to | |
1574 | * be pinning these entries, we might as well move them out | |
1575 | * of the CMA zone if possible. | |
1576 | */ | |
aa712399 PL |
1577 | if (is_migrate_cma_page(head)) { |
1578 | if (PageHuge(head)) | |
9a4e9f3b | 1579 | isolate_huge_page(head, &cma_page_list); |
aa712399 | 1580 | else { |
9a4e9f3b AK |
1581 | if (!PageLRU(head) && drain_allow) { |
1582 | lru_add_drain_all(); | |
1583 | drain_allow = false; | |
1584 | } | |
1585 | ||
1586 | if (!isolate_lru_page(head)) { | |
1587 | list_add_tail(&head->lru, &cma_page_list); | |
1588 | mod_node_page_state(page_pgdat(head), | |
1589 | NR_ISOLATED_ANON + | |
9de4f22a | 1590 | page_is_file_lru(head), |
6c357848 | 1591 | thp_nr_pages(head)); |
9a4e9f3b AK |
1592 | } |
1593 | } | |
1594 | } | |
aa712399 PL |
1595 | |
1596 | i += step; | |
9a4e9f3b AK |
1597 | } |
1598 | ||
1599 | if (!list_empty(&cma_page_list)) { | |
1600 | /* | |
1601 | * drop the above get_user_pages reference. | |
1602 | */ | |
96e1fac1 JG |
1603 | if (gup_flags & FOLL_PIN) |
1604 | unpin_user_pages(pages, nr_pages); | |
1605 | else | |
1606 | for (i = 0; i < nr_pages; i++) | |
1607 | put_page(pages[i]); | |
9a4e9f3b | 1608 | |
ed03d924 JK |
1609 | if (migrate_pages(&cma_page_list, alloc_migration_target, NULL, |
1610 | (unsigned long)&mtc, MIGRATE_SYNC, MR_CONTIG_RANGE)) { | |
9a4e9f3b AK |
1611 | /* |
1612 | * some of the pages failed migration. Do get_user_pages | |
1613 | * without migration. | |
1614 | */ | |
1615 | migrate_allow = false; | |
1616 | ||
1617 | if (!list_empty(&cma_page_list)) | |
1618 | putback_movable_pages(&cma_page_list); | |
1619 | } | |
1620 | /* | |
932f4a63 IW |
1621 | * We did migrate all the pages, Try to get the page references |
1622 | * again migrating any new CMA pages which we failed to isolate | |
1623 | * earlier. | |
9a4e9f3b | 1624 | */ |
64019a2e | 1625 | ret = __get_user_pages_locked(mm, start, nr_pages, |
932f4a63 IW |
1626 | pages, vmas, NULL, |
1627 | gup_flags); | |
1628 | ||
b96cc655 | 1629 | if ((ret > 0) && migrate_allow) { |
1630 | nr_pages = ret; | |
9a4e9f3b AK |
1631 | drain_allow = true; |
1632 | goto check_again; | |
1633 | } | |
1634 | } | |
1635 | ||
b96cc655 | 1636 | return ret; |
9a4e9f3b AK |
1637 | } |
1638 | #else | |
64019a2e | 1639 | static long check_and_migrate_cma_pages(struct mm_struct *mm, |
932f4a63 IW |
1640 | unsigned long start, |
1641 | unsigned long nr_pages, | |
1642 | struct page **pages, | |
1643 | struct vm_area_struct **vmas, | |
1644 | unsigned int gup_flags) | |
9a4e9f3b AK |
1645 | { |
1646 | return nr_pages; | |
1647 | } | |
050a9adc | 1648 | #endif /* CONFIG_CMA */ |
9a4e9f3b | 1649 | |
2bb6d283 | 1650 | /* |
932f4a63 IW |
1651 | * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which |
1652 | * allows us to process the FOLL_LONGTERM flag. | |
2bb6d283 | 1653 | */ |
64019a2e | 1654 | static long __gup_longterm_locked(struct mm_struct *mm, |
932f4a63 IW |
1655 | unsigned long start, |
1656 | unsigned long nr_pages, | |
1657 | struct page **pages, | |
1658 | struct vm_area_struct **vmas, | |
1659 | unsigned int gup_flags) | |
2bb6d283 | 1660 | { |
932f4a63 | 1661 | unsigned long flags = 0; |
52650c8b | 1662 | long rc; |
2bb6d283 | 1663 | |
52650c8b | 1664 | if (gup_flags & FOLL_LONGTERM) |
932f4a63 | 1665 | flags = memalloc_nocma_save(); |
2bb6d283 | 1666 | |
52650c8b JG |
1667 | rc = __get_user_pages_locked(mm, start, nr_pages, pages, vmas, NULL, |
1668 | gup_flags); | |
2bb6d283 | 1669 | |
932f4a63 | 1670 | if (gup_flags & FOLL_LONGTERM) { |
52650c8b JG |
1671 | if (rc > 0) |
1672 | rc = check_and_migrate_cma_pages(mm, start, rc, pages, | |
1673 | vmas, gup_flags); | |
41b4dc14 | 1674 | memalloc_nocma_restore(flags); |
9a4e9f3b | 1675 | } |
2bb6d283 DW |
1676 | return rc; |
1677 | } | |
932f4a63 | 1678 | |
447f3e45 BS |
1679 | static bool is_valid_gup_flags(unsigned int gup_flags) |
1680 | { | |
1681 | /* | |
1682 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
1683 | * never directly by the caller, so enforce that with an assertion: | |
1684 | */ | |
1685 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
1686 | return false; | |
1687 | /* | |
1688 | * FOLL_PIN is a prerequisite to FOLL_LONGTERM. Another way of saying | |
1689 | * that is, FOLL_LONGTERM is a specific case, more restrictive case of | |
1690 | * FOLL_PIN. | |
1691 | */ | |
1692 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
1693 | return false; | |
1694 | ||
1695 | return true; | |
1696 | } | |
1697 | ||
22bf29b6 | 1698 | #ifdef CONFIG_MMU |
64019a2e | 1699 | static long __get_user_pages_remote(struct mm_struct *mm, |
22bf29b6 JH |
1700 | unsigned long start, unsigned long nr_pages, |
1701 | unsigned int gup_flags, struct page **pages, | |
1702 | struct vm_area_struct **vmas, int *locked) | |
1703 | { | |
1704 | /* | |
1705 | * Parts of FOLL_LONGTERM behavior are incompatible with | |
1706 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1707 | * vmas. However, this only comes up if locked is set, and there are | |
1708 | * callers that do request FOLL_LONGTERM, but do not set locked. So, | |
1709 | * allow what we can. | |
1710 | */ | |
1711 | if (gup_flags & FOLL_LONGTERM) { | |
1712 | if (WARN_ON_ONCE(locked)) | |
1713 | return -EINVAL; | |
1714 | /* | |
1715 | * This will check the vmas (even if our vmas arg is NULL) | |
1716 | * and return -ENOTSUPP if DAX isn't allowed in this case: | |
1717 | */ | |
64019a2e | 1718 | return __gup_longterm_locked(mm, start, nr_pages, pages, |
22bf29b6 JH |
1719 | vmas, gup_flags | FOLL_TOUCH | |
1720 | FOLL_REMOTE); | |
1721 | } | |
1722 | ||
64019a2e | 1723 | return __get_user_pages_locked(mm, start, nr_pages, pages, vmas, |
22bf29b6 JH |
1724 | locked, |
1725 | gup_flags | FOLL_TOUCH | FOLL_REMOTE); | |
1726 | } | |
1727 | ||
adc8cb40 | 1728 | /** |
c4237f8b | 1729 | * get_user_pages_remote() - pin user pages in memory |
c4237f8b JH |
1730 | * @mm: mm_struct of target mm |
1731 | * @start: starting user address | |
1732 | * @nr_pages: number of pages from start to pin | |
1733 | * @gup_flags: flags modifying lookup behaviour | |
1734 | * @pages: array that receives pointers to the pages pinned. | |
1735 | * Should be at least nr_pages long. Or NULL, if caller | |
1736 | * only intends to ensure the pages are faulted in. | |
1737 | * @vmas: array of pointers to vmas corresponding to each page. | |
1738 | * Or NULL if the caller does not require them. | |
1739 | * @locked: pointer to lock flag indicating whether lock is held and | |
1740 | * subsequently whether VM_FAULT_RETRY functionality can be | |
1741 | * utilised. Lock must initially be held. | |
1742 | * | |
1743 | * Returns either number of pages pinned (which may be less than the | |
1744 | * number requested), or an error. Details about the return value: | |
1745 | * | |
1746 | * -- If nr_pages is 0, returns 0. | |
1747 | * -- If nr_pages is >0, but no pages were pinned, returns -errno. | |
1748 | * -- If nr_pages is >0, and some pages were pinned, returns the number of | |
1749 | * pages pinned. Again, this may be less than nr_pages. | |
1750 | * | |
1751 | * The caller is responsible for releasing returned @pages, via put_page(). | |
1752 | * | |
c1e8d7c6 | 1753 | * @vmas are valid only as long as mmap_lock is held. |
c4237f8b | 1754 | * |
c1e8d7c6 | 1755 | * Must be called with mmap_lock held for read or write. |
c4237f8b | 1756 | * |
adc8cb40 SJ |
1757 | * get_user_pages_remote walks a process's page tables and takes a reference |
1758 | * to each struct page that each user address corresponds to at a given | |
c4237f8b JH |
1759 | * instant. That is, it takes the page that would be accessed if a user |
1760 | * thread accesses the given user virtual address at that instant. | |
1761 | * | |
1762 | * This does not guarantee that the page exists in the user mappings when | |
adc8cb40 | 1763 | * get_user_pages_remote returns, and there may even be a completely different |
c4237f8b JH |
1764 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
1765 | * and subsequently re faulted). However it does guarantee that the page | |
1766 | * won't be freed completely. And mostly callers simply care that the page | |
1767 | * contains data that was valid *at some point in time*. Typically, an IO | |
1768 | * or similar operation cannot guarantee anything stronger anyway because | |
1769 | * locks can't be held over the syscall boundary. | |
1770 | * | |
1771 | * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page | |
1772 | * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must | |
1773 | * be called after the page is finished with, and before put_page is called. | |
1774 | * | |
adc8cb40 SJ |
1775 | * get_user_pages_remote is typically used for fewer-copy IO operations, |
1776 | * to get a handle on the memory by some means other than accesses | |
1777 | * via the user virtual addresses. The pages may be submitted for | |
1778 | * DMA to devices or accessed via their kernel linear mapping (via the | |
1779 | * kmap APIs). Care should be taken to use the correct cache flushing APIs. | |
c4237f8b JH |
1780 | * |
1781 | * See also get_user_pages_fast, for performance critical applications. | |
1782 | * | |
adc8cb40 | 1783 | * get_user_pages_remote should be phased out in favor of |
c4237f8b | 1784 | * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing |
adc8cb40 | 1785 | * should use get_user_pages_remote because it cannot pass |
c4237f8b JH |
1786 | * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault. |
1787 | */ | |
64019a2e | 1788 | long get_user_pages_remote(struct mm_struct *mm, |
c4237f8b JH |
1789 | unsigned long start, unsigned long nr_pages, |
1790 | unsigned int gup_flags, struct page **pages, | |
1791 | struct vm_area_struct **vmas, int *locked) | |
1792 | { | |
447f3e45 | 1793 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
1794 | return -EINVAL; |
1795 | ||
64019a2e | 1796 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
22bf29b6 | 1797 | pages, vmas, locked); |
c4237f8b JH |
1798 | } |
1799 | EXPORT_SYMBOL(get_user_pages_remote); | |
1800 | ||
eddb1c22 | 1801 | #else /* CONFIG_MMU */ |
64019a2e | 1802 | long get_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
1803 | unsigned long start, unsigned long nr_pages, |
1804 | unsigned int gup_flags, struct page **pages, | |
1805 | struct vm_area_struct **vmas, int *locked) | |
1806 | { | |
1807 | return 0; | |
1808 | } | |
3faa52c0 | 1809 | |
64019a2e | 1810 | static long __get_user_pages_remote(struct mm_struct *mm, |
3faa52c0 JH |
1811 | unsigned long start, unsigned long nr_pages, |
1812 | unsigned int gup_flags, struct page **pages, | |
1813 | struct vm_area_struct **vmas, int *locked) | |
1814 | { | |
1815 | return 0; | |
1816 | } | |
eddb1c22 JH |
1817 | #endif /* !CONFIG_MMU */ |
1818 | ||
adc8cb40 SJ |
1819 | /** |
1820 | * get_user_pages() - pin user pages in memory | |
1821 | * @start: starting user address | |
1822 | * @nr_pages: number of pages from start to pin | |
1823 | * @gup_flags: flags modifying lookup behaviour | |
1824 | * @pages: array that receives pointers to the pages pinned. | |
1825 | * Should be at least nr_pages long. Or NULL, if caller | |
1826 | * only intends to ensure the pages are faulted in. | |
1827 | * @vmas: array of pointers to vmas corresponding to each page. | |
1828 | * Or NULL if the caller does not require them. | |
1829 | * | |
64019a2e PX |
1830 | * This is the same as get_user_pages_remote(), just with a less-flexible |
1831 | * calling convention where we assume that the mm being operated on belongs to | |
1832 | * the current task, and doesn't allow passing of a locked parameter. We also | |
1833 | * obviously don't pass FOLL_REMOTE in here. | |
932f4a63 IW |
1834 | */ |
1835 | long get_user_pages(unsigned long start, unsigned long nr_pages, | |
1836 | unsigned int gup_flags, struct page **pages, | |
1837 | struct vm_area_struct **vmas) | |
1838 | { | |
447f3e45 | 1839 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
1840 | return -EINVAL; |
1841 | ||
64019a2e | 1842 | return __gup_longterm_locked(current->mm, start, nr_pages, |
932f4a63 IW |
1843 | pages, vmas, gup_flags | FOLL_TOUCH); |
1844 | } | |
1845 | EXPORT_SYMBOL(get_user_pages); | |
2bb6d283 | 1846 | |
adc8cb40 | 1847 | /** |
a00cda3f MCC |
1848 | * get_user_pages_locked() - variant of get_user_pages() |
1849 | * | |
1850 | * @start: starting user address | |
1851 | * @nr_pages: number of pages from start to pin | |
1852 | * @gup_flags: flags modifying lookup behaviour | |
1853 | * @pages: array that receives pointers to the pages pinned. | |
1854 | * Should be at least nr_pages long. Or NULL, if caller | |
1855 | * only intends to ensure the pages are faulted in. | |
1856 | * @locked: pointer to lock flag indicating whether lock is held and | |
1857 | * subsequently whether VM_FAULT_RETRY functionality can be | |
1858 | * utilised. Lock must initially be held. | |
1859 | * | |
1860 | * It is suitable to replace the form: | |
acc3c8d1 | 1861 | * |
3e4e28c5 | 1862 | * mmap_read_lock(mm); |
d3649f68 | 1863 | * do_something() |
64019a2e | 1864 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 1865 | * mmap_read_unlock(mm); |
acc3c8d1 | 1866 | * |
d3649f68 | 1867 | * to: |
acc3c8d1 | 1868 | * |
d3649f68 | 1869 | * int locked = 1; |
3e4e28c5 | 1870 | * mmap_read_lock(mm); |
d3649f68 | 1871 | * do_something() |
64019a2e | 1872 | * get_user_pages_locked(mm, ..., pages, &locked); |
d3649f68 | 1873 | * if (locked) |
3e4e28c5 | 1874 | * mmap_read_unlock(mm); |
adc8cb40 | 1875 | * |
adc8cb40 SJ |
1876 | * We can leverage the VM_FAULT_RETRY functionality in the page fault |
1877 | * paths better by using either get_user_pages_locked() or | |
1878 | * get_user_pages_unlocked(). | |
1879 | * | |
acc3c8d1 | 1880 | */ |
d3649f68 CH |
1881 | long get_user_pages_locked(unsigned long start, unsigned long nr_pages, |
1882 | unsigned int gup_flags, struct page **pages, | |
1883 | int *locked) | |
acc3c8d1 | 1884 | { |
acc3c8d1 | 1885 | /* |
d3649f68 CH |
1886 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with |
1887 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1888 | * vmas. As there are no users of this flag in this call we simply | |
1889 | * disallow this option for now. | |
acc3c8d1 | 1890 | */ |
d3649f68 CH |
1891 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) |
1892 | return -EINVAL; | |
420c2091 JH |
1893 | /* |
1894 | * FOLL_PIN must only be set internally by the pin_user_pages*() APIs, | |
1895 | * never directly by the caller, so enforce that: | |
1896 | */ | |
1897 | if (WARN_ON_ONCE(gup_flags & FOLL_PIN)) | |
1898 | return -EINVAL; | |
acc3c8d1 | 1899 | |
64019a2e | 1900 | return __get_user_pages_locked(current->mm, start, nr_pages, |
d3649f68 CH |
1901 | pages, NULL, locked, |
1902 | gup_flags | FOLL_TOUCH); | |
acc3c8d1 | 1903 | } |
d3649f68 | 1904 | EXPORT_SYMBOL(get_user_pages_locked); |
acc3c8d1 KS |
1905 | |
1906 | /* | |
d3649f68 | 1907 | * get_user_pages_unlocked() is suitable to replace the form: |
acc3c8d1 | 1908 | * |
3e4e28c5 | 1909 | * mmap_read_lock(mm); |
64019a2e | 1910 | * get_user_pages(mm, ..., pages, NULL); |
3e4e28c5 | 1911 | * mmap_read_unlock(mm); |
d3649f68 CH |
1912 | * |
1913 | * with: | |
1914 | * | |
64019a2e | 1915 | * get_user_pages_unlocked(mm, ..., pages); |
d3649f68 CH |
1916 | * |
1917 | * It is functionally equivalent to get_user_pages_fast so | |
1918 | * get_user_pages_fast should be used instead if specific gup_flags | |
1919 | * (e.g. FOLL_FORCE) are not required. | |
acc3c8d1 | 1920 | */ |
d3649f68 CH |
1921 | long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages, |
1922 | struct page **pages, unsigned int gup_flags) | |
acc3c8d1 KS |
1923 | { |
1924 | struct mm_struct *mm = current->mm; | |
d3649f68 CH |
1925 | int locked = 1; |
1926 | long ret; | |
acc3c8d1 | 1927 | |
d3649f68 CH |
1928 | /* |
1929 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
1930 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
1931 | * vmas. As there are no users of this flag in this call we simply | |
1932 | * disallow this option for now. | |
1933 | */ | |
1934 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
1935 | return -EINVAL; | |
acc3c8d1 | 1936 | |
d8ed45c5 | 1937 | mmap_read_lock(mm); |
64019a2e | 1938 | ret = __get_user_pages_locked(mm, start, nr_pages, pages, NULL, |
d3649f68 | 1939 | &locked, gup_flags | FOLL_TOUCH); |
acc3c8d1 | 1940 | if (locked) |
d8ed45c5 | 1941 | mmap_read_unlock(mm); |
d3649f68 | 1942 | return ret; |
4bbd4c77 | 1943 | } |
d3649f68 | 1944 | EXPORT_SYMBOL(get_user_pages_unlocked); |
2667f50e SC |
1945 | |
1946 | /* | |
67a929e0 | 1947 | * Fast GUP |
2667f50e SC |
1948 | * |
1949 | * get_user_pages_fast attempts to pin user pages by walking the page | |
1950 | * tables directly and avoids taking locks. Thus the walker needs to be | |
1951 | * protected from page table pages being freed from under it, and should | |
1952 | * block any THP splits. | |
1953 | * | |
1954 | * One way to achieve this is to have the walker disable interrupts, and | |
1955 | * rely on IPIs from the TLB flushing code blocking before the page table | |
1956 | * pages are freed. This is unsuitable for architectures that do not need | |
1957 | * to broadcast an IPI when invalidating TLBs. | |
1958 | * | |
1959 | * Another way to achieve this is to batch up page table containing pages | |
1960 | * belonging to more than one mm_user, then rcu_sched a callback to free those | |
1961 | * pages. Disabling interrupts will allow the fast_gup walker to both block | |
1962 | * the rcu_sched callback, and an IPI that we broadcast for splitting THPs | |
1963 | * (which is a relatively rare event). The code below adopts this strategy. | |
1964 | * | |
1965 | * Before activating this code, please be aware that the following assumptions | |
1966 | * are currently made: | |
1967 | * | |
ff2e6d72 | 1968 | * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to |
e585513b | 1969 | * free pages containing page tables or TLB flushing requires IPI broadcast. |
2667f50e | 1970 | * |
2667f50e SC |
1971 | * *) ptes can be read atomically by the architecture. |
1972 | * | |
1973 | * *) access_ok is sufficient to validate userspace address ranges. | |
1974 | * | |
1975 | * The last two assumptions can be relaxed by the addition of helper functions. | |
1976 | * | |
1977 | * This code is based heavily on the PowerPC implementation by Nick Piggin. | |
1978 | */ | |
67a929e0 | 1979 | #ifdef CONFIG_HAVE_FAST_GUP |
3faa52c0 | 1980 | |
790c7369 | 1981 | static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start, |
3b78d834 | 1982 | unsigned int flags, |
790c7369 | 1983 | struct page **pages) |
b59f65fa KS |
1984 | { |
1985 | while ((*nr) - nr_start) { | |
1986 | struct page *page = pages[--(*nr)]; | |
1987 | ||
1988 | ClearPageReferenced(page); | |
3faa52c0 JH |
1989 | if (flags & FOLL_PIN) |
1990 | unpin_user_page(page); | |
1991 | else | |
1992 | put_page(page); | |
b59f65fa KS |
1993 | } |
1994 | } | |
1995 | ||
3010a5ea | 1996 | #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL |
2667f50e | 1997 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, |
b798bec4 | 1998 | unsigned int flags, struct page **pages, int *nr) |
2667f50e | 1999 | { |
b59f65fa KS |
2000 | struct dev_pagemap *pgmap = NULL; |
2001 | int nr_start = *nr, ret = 0; | |
2667f50e | 2002 | pte_t *ptep, *ptem; |
2667f50e SC |
2003 | |
2004 | ptem = ptep = pte_offset_map(&pmd, addr); | |
2005 | do { | |
2a4a06da | 2006 | pte_t pte = ptep_get_lockless(ptep); |
7aef4172 | 2007 | struct page *head, *page; |
2667f50e SC |
2008 | |
2009 | /* | |
2010 | * Similar to the PMD case below, NUMA hinting must take slow | |
8a0516ed | 2011 | * path using the pte_protnone check. |
2667f50e | 2012 | */ |
e7884f8e KS |
2013 | if (pte_protnone(pte)) |
2014 | goto pte_unmap; | |
2015 | ||
b798bec4 | 2016 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
e7884f8e KS |
2017 | goto pte_unmap; |
2018 | ||
b59f65fa | 2019 | if (pte_devmap(pte)) { |
7af75561 IW |
2020 | if (unlikely(flags & FOLL_LONGTERM)) |
2021 | goto pte_unmap; | |
2022 | ||
b59f65fa KS |
2023 | pgmap = get_dev_pagemap(pte_pfn(pte), pgmap); |
2024 | if (unlikely(!pgmap)) { | |
3b78d834 | 2025 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2026 | goto pte_unmap; |
2027 | } | |
2028 | } else if (pte_special(pte)) | |
2667f50e SC |
2029 | goto pte_unmap; |
2030 | ||
2031 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2032 | page = pte_page(pte); | |
2033 | ||
3faa52c0 | 2034 | head = try_grab_compound_head(page, 1, flags); |
8fde12ca | 2035 | if (!head) |
2667f50e SC |
2036 | goto pte_unmap; |
2037 | ||
2038 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
3faa52c0 | 2039 | put_compound_head(head, 1, flags); |
2667f50e SC |
2040 | goto pte_unmap; |
2041 | } | |
2042 | ||
7aef4172 | 2043 | VM_BUG_ON_PAGE(compound_head(page) != head, page); |
e9348053 | 2044 | |
f28d4363 CI |
2045 | /* |
2046 | * We need to make the page accessible if and only if we are | |
2047 | * going to access its content (the FOLL_PIN case). Please | |
2048 | * see Documentation/core-api/pin_user_pages.rst for | |
2049 | * details. | |
2050 | */ | |
2051 | if (flags & FOLL_PIN) { | |
2052 | ret = arch_make_page_accessible(page); | |
2053 | if (ret) { | |
2054 | unpin_user_page(page); | |
2055 | goto pte_unmap; | |
2056 | } | |
2057 | } | |
e9348053 | 2058 | SetPageReferenced(page); |
2667f50e SC |
2059 | pages[*nr] = page; |
2060 | (*nr)++; | |
2061 | ||
2062 | } while (ptep++, addr += PAGE_SIZE, addr != end); | |
2063 | ||
2064 | ret = 1; | |
2065 | ||
2066 | pte_unmap: | |
832d7aa0 CH |
2067 | if (pgmap) |
2068 | put_dev_pagemap(pgmap); | |
2667f50e SC |
2069 | pte_unmap(ptem); |
2070 | return ret; | |
2071 | } | |
2072 | #else | |
2073 | ||
2074 | /* | |
2075 | * If we can't determine whether or not a pte is special, then fail immediately | |
2076 | * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not | |
2077 | * to be special. | |
2078 | * | |
2079 | * For a futex to be placed on a THP tail page, get_futex_key requires a | |
dadbb612 | 2080 | * get_user_pages_fast_only implementation that can pin pages. Thus it's still |
2667f50e SC |
2081 | * useful to have gup_huge_pmd even if we can't operate on ptes. |
2082 | */ | |
2083 | static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end, | |
b798bec4 | 2084 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2085 | { |
2086 | return 0; | |
2087 | } | |
3010a5ea | 2088 | #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */ |
2667f50e | 2089 | |
17596731 | 2090 | #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
b59f65fa | 2091 | static int __gup_device_huge(unsigned long pfn, unsigned long addr, |
86dfbed4 JH |
2092 | unsigned long end, unsigned int flags, |
2093 | struct page **pages, int *nr) | |
b59f65fa KS |
2094 | { |
2095 | int nr_start = *nr; | |
2096 | struct dev_pagemap *pgmap = NULL; | |
2097 | ||
2098 | do { | |
2099 | struct page *page = pfn_to_page(pfn); | |
2100 | ||
2101 | pgmap = get_dev_pagemap(pfn, pgmap); | |
2102 | if (unlikely(!pgmap)) { | |
3b78d834 | 2103 | undo_dev_pagemap(nr, nr_start, flags, pages); |
b59f65fa KS |
2104 | return 0; |
2105 | } | |
2106 | SetPageReferenced(page); | |
2107 | pages[*nr] = page; | |
3faa52c0 JH |
2108 | if (unlikely(!try_grab_page(page, flags))) { |
2109 | undo_dev_pagemap(nr, nr_start, flags, pages); | |
2110 | return 0; | |
2111 | } | |
b59f65fa KS |
2112 | (*nr)++; |
2113 | pfn++; | |
2114 | } while (addr += PAGE_SIZE, addr != end); | |
832d7aa0 CH |
2115 | |
2116 | if (pgmap) | |
2117 | put_dev_pagemap(pgmap); | |
b59f65fa KS |
2118 | return 1; |
2119 | } | |
2120 | ||
a9b6de77 | 2121 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2122 | unsigned long end, unsigned int flags, |
2123 | struct page **pages, int *nr) | |
b59f65fa KS |
2124 | { |
2125 | unsigned long fault_pfn; | |
a9b6de77 DW |
2126 | int nr_start = *nr; |
2127 | ||
2128 | fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2129 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2130 | return 0; |
b59f65fa | 2131 | |
a9b6de77 | 2132 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { |
3b78d834 | 2133 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2134 | return 0; |
2135 | } | |
2136 | return 1; | |
b59f65fa KS |
2137 | } |
2138 | ||
a9b6de77 | 2139 | static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2140 | unsigned long end, unsigned int flags, |
2141 | struct page **pages, int *nr) | |
b59f65fa KS |
2142 | { |
2143 | unsigned long fault_pfn; | |
a9b6de77 DW |
2144 | int nr_start = *nr; |
2145 | ||
2146 | fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); | |
86dfbed4 | 2147 | if (!__gup_device_huge(fault_pfn, addr, end, flags, pages, nr)) |
a9b6de77 | 2148 | return 0; |
b59f65fa | 2149 | |
a9b6de77 | 2150 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { |
3b78d834 | 2151 | undo_dev_pagemap(nr, nr_start, flags, pages); |
a9b6de77 DW |
2152 | return 0; |
2153 | } | |
2154 | return 1; | |
b59f65fa KS |
2155 | } |
2156 | #else | |
a9b6de77 | 2157 | static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
86dfbed4 JH |
2158 | unsigned long end, unsigned int flags, |
2159 | struct page **pages, int *nr) | |
b59f65fa KS |
2160 | { |
2161 | BUILD_BUG(); | |
2162 | return 0; | |
2163 | } | |
2164 | ||
a9b6de77 | 2165 | static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr, |
86dfbed4 JH |
2166 | unsigned long end, unsigned int flags, |
2167 | struct page **pages, int *nr) | |
b59f65fa KS |
2168 | { |
2169 | BUILD_BUG(); | |
2170 | return 0; | |
2171 | } | |
2172 | #endif | |
2173 | ||
a43e9820 JH |
2174 | static int record_subpages(struct page *page, unsigned long addr, |
2175 | unsigned long end, struct page **pages) | |
2176 | { | |
2177 | int nr; | |
2178 | ||
2179 | for (nr = 0; addr != end; addr += PAGE_SIZE) | |
2180 | pages[nr++] = page++; | |
2181 | ||
2182 | return nr; | |
2183 | } | |
2184 | ||
cbd34da7 CH |
2185 | #ifdef CONFIG_ARCH_HAS_HUGEPD |
2186 | static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end, | |
2187 | unsigned long sz) | |
2188 | { | |
2189 | unsigned long __boundary = (addr + sz) & ~(sz-1); | |
2190 | return (__boundary - 1 < end - 1) ? __boundary : end; | |
2191 | } | |
2192 | ||
2193 | static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr, | |
0cd22afd JH |
2194 | unsigned long end, unsigned int flags, |
2195 | struct page **pages, int *nr) | |
cbd34da7 CH |
2196 | { |
2197 | unsigned long pte_end; | |
2198 | struct page *head, *page; | |
2199 | pte_t pte; | |
2200 | int refs; | |
2201 | ||
2202 | pte_end = (addr + sz) & ~(sz-1); | |
2203 | if (pte_end < end) | |
2204 | end = pte_end; | |
2205 | ||
55ca2263 | 2206 | pte = huge_ptep_get(ptep); |
cbd34da7 | 2207 | |
0cd22afd | 2208 | if (!pte_access_permitted(pte, flags & FOLL_WRITE)) |
cbd34da7 CH |
2209 | return 0; |
2210 | ||
2211 | /* hugepages are never "special" */ | |
2212 | VM_BUG_ON(!pfn_valid(pte_pfn(pte))); | |
2213 | ||
cbd34da7 | 2214 | head = pte_page(pte); |
cbd34da7 | 2215 | page = head + ((addr & (sz-1)) >> PAGE_SHIFT); |
a43e9820 | 2216 | refs = record_subpages(page, addr, end, pages + *nr); |
cbd34da7 | 2217 | |
3faa52c0 | 2218 | head = try_grab_compound_head(head, refs, flags); |
a43e9820 | 2219 | if (!head) |
cbd34da7 | 2220 | return 0; |
cbd34da7 CH |
2221 | |
2222 | if (unlikely(pte_val(pte) != pte_val(*ptep))) { | |
3b78d834 | 2223 | put_compound_head(head, refs, flags); |
cbd34da7 CH |
2224 | return 0; |
2225 | } | |
2226 | ||
a43e9820 | 2227 | *nr += refs; |
520b4a44 | 2228 | SetPageReferenced(head); |
cbd34da7 CH |
2229 | return 1; |
2230 | } | |
2231 | ||
2232 | static int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2233 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2234 | struct page **pages, int *nr) |
2235 | { | |
2236 | pte_t *ptep; | |
2237 | unsigned long sz = 1UL << hugepd_shift(hugepd); | |
2238 | unsigned long next; | |
2239 | ||
2240 | ptep = hugepte_offset(hugepd, addr, pdshift); | |
2241 | do { | |
2242 | next = hugepte_addr_end(addr, end, sz); | |
0cd22afd | 2243 | if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr)) |
cbd34da7 CH |
2244 | return 0; |
2245 | } while (ptep++, addr = next, addr != end); | |
2246 | ||
2247 | return 1; | |
2248 | } | |
2249 | #else | |
2250 | static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr, | |
0cd22afd | 2251 | unsigned int pdshift, unsigned long end, unsigned int flags, |
cbd34da7 CH |
2252 | struct page **pages, int *nr) |
2253 | { | |
2254 | return 0; | |
2255 | } | |
2256 | #endif /* CONFIG_ARCH_HAS_HUGEPD */ | |
2257 | ||
2667f50e | 2258 | static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr, |
0cd22afd JH |
2259 | unsigned long end, unsigned int flags, |
2260 | struct page **pages, int *nr) | |
2667f50e | 2261 | { |
ddc58f27 | 2262 | struct page *head, *page; |
2667f50e SC |
2263 | int refs; |
2264 | ||
b798bec4 | 2265 | if (!pmd_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2266 | return 0; |
2267 | ||
7af75561 IW |
2268 | if (pmd_devmap(orig)) { |
2269 | if (unlikely(flags & FOLL_LONGTERM)) | |
2270 | return 0; | |
86dfbed4 JH |
2271 | return __gup_device_huge_pmd(orig, pmdp, addr, end, flags, |
2272 | pages, nr); | |
7af75561 | 2273 | } |
b59f65fa | 2274 | |
d63206ee | 2275 | page = pmd_page(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2276 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2277 | |
3faa52c0 | 2278 | head = try_grab_compound_head(pmd_page(orig), refs, flags); |
a43e9820 | 2279 | if (!head) |
2667f50e | 2280 | return 0; |
2667f50e SC |
2281 | |
2282 | if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) { | |
3b78d834 | 2283 | put_compound_head(head, refs, flags); |
2667f50e SC |
2284 | return 0; |
2285 | } | |
2286 | ||
a43e9820 | 2287 | *nr += refs; |
e9348053 | 2288 | SetPageReferenced(head); |
2667f50e SC |
2289 | return 1; |
2290 | } | |
2291 | ||
2292 | static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr, | |
86dfbed4 JH |
2293 | unsigned long end, unsigned int flags, |
2294 | struct page **pages, int *nr) | |
2667f50e | 2295 | { |
ddc58f27 | 2296 | struct page *head, *page; |
2667f50e SC |
2297 | int refs; |
2298 | ||
b798bec4 | 2299 | if (!pud_access_permitted(orig, flags & FOLL_WRITE)) |
2667f50e SC |
2300 | return 0; |
2301 | ||
7af75561 IW |
2302 | if (pud_devmap(orig)) { |
2303 | if (unlikely(flags & FOLL_LONGTERM)) | |
2304 | return 0; | |
86dfbed4 JH |
2305 | return __gup_device_huge_pud(orig, pudp, addr, end, flags, |
2306 | pages, nr); | |
7af75561 | 2307 | } |
b59f65fa | 2308 | |
d63206ee | 2309 | page = pud_page(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT); |
a43e9820 | 2310 | refs = record_subpages(page, addr, end, pages + *nr); |
2667f50e | 2311 | |
3faa52c0 | 2312 | head = try_grab_compound_head(pud_page(orig), refs, flags); |
a43e9820 | 2313 | if (!head) |
2667f50e | 2314 | return 0; |
2667f50e SC |
2315 | |
2316 | if (unlikely(pud_val(orig) != pud_val(*pudp))) { | |
3b78d834 | 2317 | put_compound_head(head, refs, flags); |
2667f50e SC |
2318 | return 0; |
2319 | } | |
2320 | ||
a43e9820 | 2321 | *nr += refs; |
e9348053 | 2322 | SetPageReferenced(head); |
2667f50e SC |
2323 | return 1; |
2324 | } | |
2325 | ||
f30c59e9 | 2326 | static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr, |
b798bec4 | 2327 | unsigned long end, unsigned int flags, |
f30c59e9 AK |
2328 | struct page **pages, int *nr) |
2329 | { | |
2330 | int refs; | |
ddc58f27 | 2331 | struct page *head, *page; |
f30c59e9 | 2332 | |
b798bec4 | 2333 | if (!pgd_access_permitted(orig, flags & FOLL_WRITE)) |
f30c59e9 AK |
2334 | return 0; |
2335 | ||
b59f65fa | 2336 | BUILD_BUG_ON(pgd_devmap(orig)); |
a43e9820 | 2337 | |
d63206ee | 2338 | page = pgd_page(orig) + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT); |
a43e9820 | 2339 | refs = record_subpages(page, addr, end, pages + *nr); |
f30c59e9 | 2340 | |
3faa52c0 | 2341 | head = try_grab_compound_head(pgd_page(orig), refs, flags); |
a43e9820 | 2342 | if (!head) |
f30c59e9 | 2343 | return 0; |
f30c59e9 AK |
2344 | |
2345 | if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) { | |
3b78d834 | 2346 | put_compound_head(head, refs, flags); |
f30c59e9 AK |
2347 | return 0; |
2348 | } | |
2349 | ||
a43e9820 | 2350 | *nr += refs; |
e9348053 | 2351 | SetPageReferenced(head); |
f30c59e9 AK |
2352 | return 1; |
2353 | } | |
2354 | ||
d3f7b1bb | 2355 | static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned long end, |
b798bec4 | 2356 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2357 | { |
2358 | unsigned long next; | |
2359 | pmd_t *pmdp; | |
2360 | ||
d3f7b1bb | 2361 | pmdp = pmd_offset_lockless(pudp, pud, addr); |
2667f50e | 2362 | do { |
38c5ce93 | 2363 | pmd_t pmd = READ_ONCE(*pmdp); |
2667f50e SC |
2364 | |
2365 | next = pmd_addr_end(addr, end); | |
84c3fc4e | 2366 | if (!pmd_present(pmd)) |
2667f50e SC |
2367 | return 0; |
2368 | ||
414fd080 YZ |
2369 | if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) || |
2370 | pmd_devmap(pmd))) { | |
2667f50e SC |
2371 | /* |
2372 | * NUMA hinting faults need to be handled in the GUP | |
2373 | * slowpath for accounting purposes and so that they | |
2374 | * can be serialised against THP migration. | |
2375 | */ | |
8a0516ed | 2376 | if (pmd_protnone(pmd)) |
2667f50e SC |
2377 | return 0; |
2378 | ||
b798bec4 | 2379 | if (!gup_huge_pmd(pmd, pmdp, addr, next, flags, |
2667f50e SC |
2380 | pages, nr)) |
2381 | return 0; | |
2382 | ||
f30c59e9 AK |
2383 | } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) { |
2384 | /* | |
2385 | * architecture have different format for hugetlbfs | |
2386 | * pmd format and THP pmd format | |
2387 | */ | |
2388 | if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr, | |
b798bec4 | 2389 | PMD_SHIFT, next, flags, pages, nr)) |
f30c59e9 | 2390 | return 0; |
b798bec4 | 2391 | } else if (!gup_pte_range(pmd, addr, next, flags, pages, nr)) |
2923117b | 2392 | return 0; |
2667f50e SC |
2393 | } while (pmdp++, addr = next, addr != end); |
2394 | ||
2395 | return 1; | |
2396 | } | |
2397 | ||
d3f7b1bb | 2398 | static int gup_pud_range(p4d_t *p4dp, p4d_t p4d, unsigned long addr, unsigned long end, |
b798bec4 | 2399 | unsigned int flags, struct page **pages, int *nr) |
2667f50e SC |
2400 | { |
2401 | unsigned long next; | |
2402 | pud_t *pudp; | |
2403 | ||
d3f7b1bb | 2404 | pudp = pud_offset_lockless(p4dp, p4d, addr); |
2667f50e | 2405 | do { |
e37c6982 | 2406 | pud_t pud = READ_ONCE(*pudp); |
2667f50e SC |
2407 | |
2408 | next = pud_addr_end(addr, end); | |
15494520 | 2409 | if (unlikely(!pud_present(pud))) |
2667f50e | 2410 | return 0; |
f30c59e9 | 2411 | if (unlikely(pud_huge(pud))) { |
b798bec4 | 2412 | if (!gup_huge_pud(pud, pudp, addr, next, flags, |
f30c59e9 AK |
2413 | pages, nr)) |
2414 | return 0; | |
2415 | } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) { | |
2416 | if (!gup_huge_pd(__hugepd(pud_val(pud)), addr, | |
b798bec4 | 2417 | PUD_SHIFT, next, flags, pages, nr)) |
2667f50e | 2418 | return 0; |
d3f7b1bb | 2419 | } else if (!gup_pmd_range(pudp, pud, addr, next, flags, pages, nr)) |
2667f50e SC |
2420 | return 0; |
2421 | } while (pudp++, addr = next, addr != end); | |
2422 | ||
2423 | return 1; | |
2424 | } | |
2425 | ||
d3f7b1bb | 2426 | static int gup_p4d_range(pgd_t *pgdp, pgd_t pgd, unsigned long addr, unsigned long end, |
b798bec4 | 2427 | unsigned int flags, struct page **pages, int *nr) |
c2febafc KS |
2428 | { |
2429 | unsigned long next; | |
2430 | p4d_t *p4dp; | |
2431 | ||
d3f7b1bb | 2432 | p4dp = p4d_offset_lockless(pgdp, pgd, addr); |
c2febafc KS |
2433 | do { |
2434 | p4d_t p4d = READ_ONCE(*p4dp); | |
2435 | ||
2436 | next = p4d_addr_end(addr, end); | |
2437 | if (p4d_none(p4d)) | |
2438 | return 0; | |
2439 | BUILD_BUG_ON(p4d_huge(p4d)); | |
2440 | if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) { | |
2441 | if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr, | |
b798bec4 | 2442 | P4D_SHIFT, next, flags, pages, nr)) |
c2febafc | 2443 | return 0; |
d3f7b1bb | 2444 | } else if (!gup_pud_range(p4dp, p4d, addr, next, flags, pages, nr)) |
c2febafc KS |
2445 | return 0; |
2446 | } while (p4dp++, addr = next, addr != end); | |
2447 | ||
2448 | return 1; | |
2449 | } | |
2450 | ||
5b65c467 | 2451 | static void gup_pgd_range(unsigned long addr, unsigned long end, |
b798bec4 | 2452 | unsigned int flags, struct page **pages, int *nr) |
5b65c467 KS |
2453 | { |
2454 | unsigned long next; | |
2455 | pgd_t *pgdp; | |
2456 | ||
2457 | pgdp = pgd_offset(current->mm, addr); | |
2458 | do { | |
2459 | pgd_t pgd = READ_ONCE(*pgdp); | |
2460 | ||
2461 | next = pgd_addr_end(addr, end); | |
2462 | if (pgd_none(pgd)) | |
2463 | return; | |
2464 | if (unlikely(pgd_huge(pgd))) { | |
b798bec4 | 2465 | if (!gup_huge_pgd(pgd, pgdp, addr, next, flags, |
5b65c467 KS |
2466 | pages, nr)) |
2467 | return; | |
2468 | } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) { | |
2469 | if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr, | |
b798bec4 | 2470 | PGDIR_SHIFT, next, flags, pages, nr)) |
5b65c467 | 2471 | return; |
d3f7b1bb | 2472 | } else if (!gup_p4d_range(pgdp, pgd, addr, next, flags, pages, nr)) |
5b65c467 KS |
2473 | return; |
2474 | } while (pgdp++, addr = next, addr != end); | |
2475 | } | |
050a9adc CH |
2476 | #else |
2477 | static inline void gup_pgd_range(unsigned long addr, unsigned long end, | |
2478 | unsigned int flags, struct page **pages, int *nr) | |
2479 | { | |
2480 | } | |
2481 | #endif /* CONFIG_HAVE_FAST_GUP */ | |
5b65c467 KS |
2482 | |
2483 | #ifndef gup_fast_permitted | |
2484 | /* | |
dadbb612 | 2485 | * Check if it's allowed to use get_user_pages_fast_only() for the range, or |
5b65c467 KS |
2486 | * we need to fall back to the slow version: |
2487 | */ | |
26f4c328 | 2488 | static bool gup_fast_permitted(unsigned long start, unsigned long end) |
5b65c467 | 2489 | { |
26f4c328 | 2490 | return true; |
5b65c467 KS |
2491 | } |
2492 | #endif | |
2493 | ||
7af75561 IW |
2494 | static int __gup_longterm_unlocked(unsigned long start, int nr_pages, |
2495 | unsigned int gup_flags, struct page **pages) | |
2496 | { | |
2497 | int ret; | |
2498 | ||
2499 | /* | |
2500 | * FIXME: FOLL_LONGTERM does not work with | |
2501 | * get_user_pages_unlocked() (see comments in that function) | |
2502 | */ | |
2503 | if (gup_flags & FOLL_LONGTERM) { | |
d8ed45c5 | 2504 | mmap_read_lock(current->mm); |
64019a2e | 2505 | ret = __gup_longterm_locked(current->mm, |
7af75561 IW |
2506 | start, nr_pages, |
2507 | pages, NULL, gup_flags); | |
d8ed45c5 | 2508 | mmap_read_unlock(current->mm); |
7af75561 IW |
2509 | } else { |
2510 | ret = get_user_pages_unlocked(start, nr_pages, | |
2511 | pages, gup_flags); | |
2512 | } | |
2513 | ||
2514 | return ret; | |
2515 | } | |
2516 | ||
c28b1fc7 JG |
2517 | static unsigned long lockless_pages_from_mm(unsigned long start, |
2518 | unsigned long end, | |
2519 | unsigned int gup_flags, | |
2520 | struct page **pages) | |
2521 | { | |
2522 | unsigned long flags; | |
2523 | int nr_pinned = 0; | |
57efa1fe | 2524 | unsigned seq; |
c28b1fc7 JG |
2525 | |
2526 | if (!IS_ENABLED(CONFIG_HAVE_FAST_GUP) || | |
2527 | !gup_fast_permitted(start, end)) | |
2528 | return 0; | |
2529 | ||
57efa1fe JG |
2530 | if (gup_flags & FOLL_PIN) { |
2531 | seq = raw_read_seqcount(¤t->mm->write_protect_seq); | |
2532 | if (seq & 1) | |
2533 | return 0; | |
2534 | } | |
2535 | ||
c28b1fc7 JG |
2536 | /* |
2537 | * Disable interrupts. The nested form is used, in order to allow full, | |
2538 | * general purpose use of this routine. | |
2539 | * | |
2540 | * With interrupts disabled, we block page table pages from being freed | |
2541 | * from under us. See struct mmu_table_batch comments in | |
2542 | * include/asm-generic/tlb.h for more details. | |
2543 | * | |
2544 | * We do not adopt an rcu_read_lock() here as we also want to block IPIs | |
2545 | * that come from THPs splitting. | |
2546 | */ | |
2547 | local_irq_save(flags); | |
2548 | gup_pgd_range(start, end, gup_flags, pages, &nr_pinned); | |
2549 | local_irq_restore(flags); | |
57efa1fe JG |
2550 | |
2551 | /* | |
2552 | * When pinning pages for DMA there could be a concurrent write protect | |
2553 | * from fork() via copy_page_range(), in this case always fail fast GUP. | |
2554 | */ | |
2555 | if (gup_flags & FOLL_PIN) { | |
2556 | if (read_seqcount_retry(¤t->mm->write_protect_seq, seq)) { | |
2557 | unpin_user_pages(pages, nr_pinned); | |
2558 | return 0; | |
2559 | } | |
2560 | } | |
c28b1fc7 JG |
2561 | return nr_pinned; |
2562 | } | |
2563 | ||
2564 | static int internal_get_user_pages_fast(unsigned long start, | |
2565 | unsigned long nr_pages, | |
eddb1c22 JH |
2566 | unsigned int gup_flags, |
2567 | struct page **pages) | |
2667f50e | 2568 | { |
c28b1fc7 JG |
2569 | unsigned long len, end; |
2570 | unsigned long nr_pinned; | |
2571 | int ret; | |
2667f50e | 2572 | |
f4000fdf | 2573 | if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM | |
376a34ef JH |
2574 | FOLL_FORCE | FOLL_PIN | FOLL_GET | |
2575 | FOLL_FAST_ONLY))) | |
817be129 CH |
2576 | return -EINVAL; |
2577 | ||
008cfe44 PX |
2578 | if (gup_flags & FOLL_PIN) |
2579 | atomic_set(¤t->mm->has_pinned, 1); | |
2580 | ||
f81cd178 | 2581 | if (!(gup_flags & FOLL_FAST_ONLY)) |
da1c55f1 | 2582 | might_lock_read(¤t->mm->mmap_lock); |
f81cd178 | 2583 | |
f455c854 | 2584 | start = untagged_addr(start) & PAGE_MASK; |
c28b1fc7 JG |
2585 | len = nr_pages << PAGE_SHIFT; |
2586 | if (check_add_overflow(start, len, &end)) | |
c61611f7 | 2587 | return 0; |
96d4f267 | 2588 | if (unlikely(!access_ok((void __user *)start, len))) |
c61611f7 | 2589 | return -EFAULT; |
73e10a61 | 2590 | |
c28b1fc7 JG |
2591 | nr_pinned = lockless_pages_from_mm(start, end, gup_flags, pages); |
2592 | if (nr_pinned == nr_pages || gup_flags & FOLL_FAST_ONLY) | |
2593 | return nr_pinned; | |
2667f50e | 2594 | |
c28b1fc7 JG |
2595 | /* Slow path: try to get the remaining pages with get_user_pages */ |
2596 | start += nr_pinned << PAGE_SHIFT; | |
2597 | pages += nr_pinned; | |
2598 | ret = __gup_longterm_unlocked(start, nr_pages - nr_pinned, gup_flags, | |
2599 | pages); | |
2600 | if (ret < 0) { | |
2601 | /* | |
2602 | * The caller has to unpin the pages we already pinned so | |
2603 | * returning -errno is not an option | |
2604 | */ | |
2605 | if (nr_pinned) | |
2606 | return nr_pinned; | |
2607 | return ret; | |
2667f50e | 2608 | } |
c28b1fc7 | 2609 | return ret + nr_pinned; |
2667f50e | 2610 | } |
c28b1fc7 | 2611 | |
dadbb612 SJ |
2612 | /** |
2613 | * get_user_pages_fast_only() - pin user pages in memory | |
2614 | * @start: starting user address | |
2615 | * @nr_pages: number of pages from start to pin | |
2616 | * @gup_flags: flags modifying pin behaviour | |
2617 | * @pages: array that receives pointers to the pages pinned. | |
2618 | * Should be at least nr_pages long. | |
2619 | * | |
9e1f0580 JH |
2620 | * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to |
2621 | * the regular GUP. | |
2622 | * Note a difference with get_user_pages_fast: this always returns the | |
2623 | * number of pages pinned, 0 if no pages were pinned. | |
2624 | * | |
2625 | * If the architecture does not support this function, simply return with no | |
2626 | * pages pinned. | |
2627 | * | |
2628 | * Careful, careful! COW breaking can go either way, so a non-write | |
2629 | * access can get ambiguous page results. If you call this function without | |
2630 | * 'write' set, you'd better be sure that you're ok with that ambiguity. | |
2631 | */ | |
dadbb612 SJ |
2632 | int get_user_pages_fast_only(unsigned long start, int nr_pages, |
2633 | unsigned int gup_flags, struct page **pages) | |
9e1f0580 | 2634 | { |
376a34ef | 2635 | int nr_pinned; |
9e1f0580 JH |
2636 | /* |
2637 | * Internally (within mm/gup.c), gup fast variants must set FOLL_GET, | |
2638 | * because gup fast is always a "pin with a +1 page refcount" request. | |
376a34ef JH |
2639 | * |
2640 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2641 | * this routine: no fall back to regular ("slow") GUP. | |
9e1f0580 | 2642 | */ |
dadbb612 | 2643 | gup_flags |= FOLL_GET | FOLL_FAST_ONLY; |
9e1f0580 | 2644 | |
376a34ef JH |
2645 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, |
2646 | pages); | |
9e1f0580 JH |
2647 | |
2648 | /* | |
376a34ef JH |
2649 | * As specified in the API description above, this routine is not |
2650 | * allowed to return negative values. However, the common core | |
2651 | * routine internal_get_user_pages_fast() *can* return -errno. | |
2652 | * Therefore, correct for that here: | |
9e1f0580 | 2653 | */ |
376a34ef JH |
2654 | if (nr_pinned < 0) |
2655 | nr_pinned = 0; | |
9e1f0580 JH |
2656 | |
2657 | return nr_pinned; | |
2658 | } | |
dadbb612 | 2659 | EXPORT_SYMBOL_GPL(get_user_pages_fast_only); |
9e1f0580 | 2660 | |
eddb1c22 JH |
2661 | /** |
2662 | * get_user_pages_fast() - pin user pages in memory | |
3faa52c0 JH |
2663 | * @start: starting user address |
2664 | * @nr_pages: number of pages from start to pin | |
2665 | * @gup_flags: flags modifying pin behaviour | |
2666 | * @pages: array that receives pointers to the pages pinned. | |
2667 | * Should be at least nr_pages long. | |
eddb1c22 | 2668 | * |
c1e8d7c6 | 2669 | * Attempt to pin user pages in memory without taking mm->mmap_lock. |
eddb1c22 JH |
2670 | * If not successful, it will fall back to taking the lock and |
2671 | * calling get_user_pages(). | |
2672 | * | |
2673 | * Returns number of pages pinned. This may be fewer than the number requested. | |
2674 | * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns | |
2675 | * -errno. | |
2676 | */ | |
2677 | int get_user_pages_fast(unsigned long start, int nr_pages, | |
2678 | unsigned int gup_flags, struct page **pages) | |
2679 | { | |
447f3e45 | 2680 | if (!is_valid_gup_flags(gup_flags)) |
eddb1c22 JH |
2681 | return -EINVAL; |
2682 | ||
94202f12 JH |
2683 | /* |
2684 | * The caller may or may not have explicitly set FOLL_GET; either way is | |
2685 | * OK. However, internally (within mm/gup.c), gup fast variants must set | |
2686 | * FOLL_GET, because gup fast is always a "pin with a +1 page refcount" | |
2687 | * request. | |
2688 | */ | |
2689 | gup_flags |= FOLL_GET; | |
eddb1c22 JH |
2690 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); |
2691 | } | |
050a9adc | 2692 | EXPORT_SYMBOL_GPL(get_user_pages_fast); |
eddb1c22 JH |
2693 | |
2694 | /** | |
2695 | * pin_user_pages_fast() - pin user pages in memory without taking locks | |
2696 | * | |
3faa52c0 JH |
2697 | * @start: starting user address |
2698 | * @nr_pages: number of pages from start to pin | |
2699 | * @gup_flags: flags modifying pin behaviour | |
2700 | * @pages: array that receives pointers to the pages pinned. | |
2701 | * Should be at least nr_pages long. | |
2702 | * | |
2703 | * Nearly the same as get_user_pages_fast(), except that FOLL_PIN is set. See | |
2704 | * get_user_pages_fast() for documentation on the function arguments, because | |
2705 | * the arguments here are identical. | |
2706 | * | |
2707 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2708 | * see Documentation/core-api/pin_user_pages.rst for further details. |
eddb1c22 JH |
2709 | */ |
2710 | int pin_user_pages_fast(unsigned long start, int nr_pages, | |
2711 | unsigned int gup_flags, struct page **pages) | |
2712 | { | |
3faa52c0 JH |
2713 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2714 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2715 | return -EINVAL; | |
2716 | ||
2717 | gup_flags |= FOLL_PIN; | |
2718 | return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages); | |
eddb1c22 JH |
2719 | } |
2720 | EXPORT_SYMBOL_GPL(pin_user_pages_fast); | |
2721 | ||
104acc32 | 2722 | /* |
dadbb612 SJ |
2723 | * This is the FOLL_PIN equivalent of get_user_pages_fast_only(). Behavior |
2724 | * is the same, except that this one sets FOLL_PIN instead of FOLL_GET. | |
104acc32 JH |
2725 | * |
2726 | * The API rules are the same, too: no negative values may be returned. | |
2727 | */ | |
2728 | int pin_user_pages_fast_only(unsigned long start, int nr_pages, | |
2729 | unsigned int gup_flags, struct page **pages) | |
2730 | { | |
2731 | int nr_pinned; | |
2732 | ||
2733 | /* | |
2734 | * FOLL_GET and FOLL_PIN are mutually exclusive. Note that the API | |
2735 | * rules require returning 0, rather than -errno: | |
2736 | */ | |
2737 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2738 | return 0; | |
2739 | /* | |
2740 | * FOLL_FAST_ONLY is required in order to match the API description of | |
2741 | * this routine: no fall back to regular ("slow") GUP. | |
2742 | */ | |
2743 | gup_flags |= (FOLL_PIN | FOLL_FAST_ONLY); | |
2744 | nr_pinned = internal_get_user_pages_fast(start, nr_pages, gup_flags, | |
2745 | pages); | |
2746 | /* | |
2747 | * This routine is not allowed to return negative values. However, | |
2748 | * internal_get_user_pages_fast() *can* return -errno. Therefore, | |
2749 | * correct for that here: | |
2750 | */ | |
2751 | if (nr_pinned < 0) | |
2752 | nr_pinned = 0; | |
2753 | ||
2754 | return nr_pinned; | |
2755 | } | |
2756 | EXPORT_SYMBOL_GPL(pin_user_pages_fast_only); | |
2757 | ||
eddb1c22 | 2758 | /** |
64019a2e | 2759 | * pin_user_pages_remote() - pin pages of a remote process |
eddb1c22 | 2760 | * |
3faa52c0 JH |
2761 | * @mm: mm_struct of target mm |
2762 | * @start: starting user address | |
2763 | * @nr_pages: number of pages from start to pin | |
2764 | * @gup_flags: flags modifying lookup behaviour | |
2765 | * @pages: array that receives pointers to the pages pinned. | |
2766 | * Should be at least nr_pages long. Or NULL, if caller | |
2767 | * only intends to ensure the pages are faulted in. | |
2768 | * @vmas: array of pointers to vmas corresponding to each page. | |
2769 | * Or NULL if the caller does not require them. | |
2770 | * @locked: pointer to lock flag indicating whether lock is held and | |
2771 | * subsequently whether VM_FAULT_RETRY functionality can be | |
2772 | * utilised. Lock must initially be held. | |
2773 | * | |
2774 | * Nearly the same as get_user_pages_remote(), except that FOLL_PIN is set. See | |
2775 | * get_user_pages_remote() for documentation on the function arguments, because | |
2776 | * the arguments here are identical. | |
2777 | * | |
2778 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2779 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 | 2780 | */ |
64019a2e | 2781 | long pin_user_pages_remote(struct mm_struct *mm, |
eddb1c22 JH |
2782 | unsigned long start, unsigned long nr_pages, |
2783 | unsigned int gup_flags, struct page **pages, | |
2784 | struct vm_area_struct **vmas, int *locked) | |
2785 | { | |
3faa52c0 JH |
2786 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2787 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2788 | return -EINVAL; | |
2789 | ||
2790 | gup_flags |= FOLL_PIN; | |
64019a2e | 2791 | return __get_user_pages_remote(mm, start, nr_pages, gup_flags, |
3faa52c0 | 2792 | pages, vmas, locked); |
eddb1c22 JH |
2793 | } |
2794 | EXPORT_SYMBOL(pin_user_pages_remote); | |
2795 | ||
2796 | /** | |
2797 | * pin_user_pages() - pin user pages in memory for use by other devices | |
2798 | * | |
3faa52c0 JH |
2799 | * @start: starting user address |
2800 | * @nr_pages: number of pages from start to pin | |
2801 | * @gup_flags: flags modifying lookup behaviour | |
2802 | * @pages: array that receives pointers to the pages pinned. | |
2803 | * Should be at least nr_pages long. Or NULL, if caller | |
2804 | * only intends to ensure the pages are faulted in. | |
2805 | * @vmas: array of pointers to vmas corresponding to each page. | |
2806 | * Or NULL if the caller does not require them. | |
2807 | * | |
2808 | * Nearly the same as get_user_pages(), except that FOLL_TOUCH is not set, and | |
2809 | * FOLL_PIN is set. | |
2810 | * | |
2811 | * FOLL_PIN means that the pages must be released via unpin_user_page(). Please | |
72ef5e52 | 2812 | * see Documentation/core-api/pin_user_pages.rst for details. |
eddb1c22 JH |
2813 | */ |
2814 | long pin_user_pages(unsigned long start, unsigned long nr_pages, | |
2815 | unsigned int gup_flags, struct page **pages, | |
2816 | struct vm_area_struct **vmas) | |
2817 | { | |
3faa52c0 JH |
2818 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ |
2819 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2820 | return -EINVAL; | |
2821 | ||
2822 | gup_flags |= FOLL_PIN; | |
64019a2e | 2823 | return __gup_longterm_locked(current->mm, start, nr_pages, |
3faa52c0 | 2824 | pages, vmas, gup_flags); |
eddb1c22 JH |
2825 | } |
2826 | EXPORT_SYMBOL(pin_user_pages); | |
91429023 JH |
2827 | |
2828 | /* | |
2829 | * pin_user_pages_unlocked() is the FOLL_PIN variant of | |
2830 | * get_user_pages_unlocked(). Behavior is the same, except that this one sets | |
2831 | * FOLL_PIN and rejects FOLL_GET. | |
2832 | */ | |
2833 | long pin_user_pages_unlocked(unsigned long start, unsigned long nr_pages, | |
2834 | struct page **pages, unsigned int gup_flags) | |
2835 | { | |
2836 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
2837 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2838 | return -EINVAL; | |
2839 | ||
2840 | gup_flags |= FOLL_PIN; | |
2841 | return get_user_pages_unlocked(start, nr_pages, pages, gup_flags); | |
2842 | } | |
2843 | EXPORT_SYMBOL(pin_user_pages_unlocked); | |
420c2091 JH |
2844 | |
2845 | /* | |
2846 | * pin_user_pages_locked() is the FOLL_PIN variant of get_user_pages_locked(). | |
2847 | * Behavior is the same, except that this one sets FOLL_PIN and rejects | |
2848 | * FOLL_GET. | |
2849 | */ | |
2850 | long pin_user_pages_locked(unsigned long start, unsigned long nr_pages, | |
2851 | unsigned int gup_flags, struct page **pages, | |
2852 | int *locked) | |
2853 | { | |
2854 | /* | |
2855 | * FIXME: Current FOLL_LONGTERM behavior is incompatible with | |
2856 | * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on | |
2857 | * vmas. As there are no users of this flag in this call we simply | |
2858 | * disallow this option for now. | |
2859 | */ | |
2860 | if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM)) | |
2861 | return -EINVAL; | |
2862 | ||
2863 | /* FOLL_GET and FOLL_PIN are mutually exclusive. */ | |
2864 | if (WARN_ON_ONCE(gup_flags & FOLL_GET)) | |
2865 | return -EINVAL; | |
2866 | ||
2867 | gup_flags |= FOLL_PIN; | |
64019a2e | 2868 | return __get_user_pages_locked(current->mm, start, nr_pages, |
420c2091 JH |
2869 | pages, NULL, locked, |
2870 | gup_flags | FOLL_TOUCH); | |
2871 | } | |
2872 | EXPORT_SYMBOL(pin_user_pages_locked); |