Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0-only
2 : : #include <linux/kernel.h>
3 : : #include <linux/errno.h>
4 : : #include <linux/err.h>
5 : : #include <linux/spinlock.h>
6 : :
7 : : #include <linux/mm.h>
8 : : #include <linux/memremap.h>
9 : : #include <linux/pagemap.h>
10 : : #include <linux/rmap.h>
11 : : #include <linux/swap.h>
12 : : #include <linux/swapops.h>
13 : :
14 : : #include <linux/sched/signal.h>
15 : : #include <linux/rwsem.h>
16 : : #include <linux/hugetlb.h>
17 : : #include <linux/migrate.h>
18 : : #include <linux/mm_inline.h>
19 : : #include <linux/sched/mm.h>
20 : :
21 : : #include <asm/mmu_context.h>
22 : : #include <asm/pgtable.h>
23 : : #include <asm/tlbflush.h>
24 : :
25 : : #include "internal.h"
26 : :
27 : : struct follow_page_context {
28 : : struct dev_pagemap *pgmap;
29 : : unsigned int page_mask;
30 : : };
31 : :
32 : : /*
33 : : * Return the compound head page with ref appropriately incremented,
34 : : * or NULL if that failed.
35 : : */
36 : 9 : static inline struct page *try_get_compound_head(struct page *page, int refs)
37 : : {
38 [ - + ]: 9 : struct page *head = compound_head(page);
39 : :
40 [ - + + - ]: 9 : if (WARN_ON_ONCE(page_ref_count(head) < 0))
41 : : return NULL;
42 [ - + ]: 18 : if (unlikely(!page_cache_add_speculative(head, refs)))
43 : 0 : return NULL;
44 : : return head;
45 : : }
46 : :
47 : : /**
48 : : * unpin_user_pages_dirty_lock() - release and optionally dirty gup-pinned pages
49 : : * @pages: array of pages to be maybe marked dirty, and definitely released.
50 : : * @npages: number of pages in the @pages array.
51 : : * @make_dirty: whether to mark the pages dirty
52 : : *
53 : : * "gup-pinned page" refers to a page that has had one of the get_user_pages()
54 : : * variants called on that page.
55 : : *
56 : : * For each page in the @pages array, make that page (or its head page, if a
57 : : * compound page) dirty, if @make_dirty is true, and if the page was previously
58 : : * listed as clean. In any case, releases all pages using unpin_user_page(),
59 : : * possibly via unpin_user_pages(), for the non-dirty case.
60 : : *
61 : : * Please see the unpin_user_page() documentation for details.
62 : : *
63 : : * set_page_dirty_lock() is used internally. If instead, set_page_dirty() is
64 : : * required, then the caller should a) verify that this is really correct,
65 : : * because _lock() is usually required, and b) hand code it:
66 : : * set_page_dirty_lock(), unpin_user_page().
67 : : *
68 : : */
69 : 0 : void unpin_user_pages_dirty_lock(struct page **pages, unsigned long npages,
70 : : bool make_dirty)
71 : : {
72 : 0 : unsigned long index;
73 : :
74 : : /*
75 : : * TODO: this can be optimized for huge pages: if a series of pages is
76 : : * physically contiguous and part of the same compound page, then a
77 : : * single operation to the head page should suffice.
78 : : */
79 : :
80 [ # # ]: 0 : if (!make_dirty) {
81 : : unpin_user_pages(pages, npages);
82 : : return;
83 : : }
84 : :
85 [ # # ]: 0 : for (index = 0; index < npages; index++) {
86 [ # # ]: 0 : struct page *page = compound_head(pages[index]);
87 : : /*
88 : : * Checking PageDirty at this point may race with
89 : : * clear_page_dirty_for_io(), but that's OK. Two key
90 : : * cases:
91 : : *
92 : : * 1) This code sees the page as already dirty, so it
93 : : * skips the call to set_page_dirty(). That could happen
94 : : * because clear_page_dirty_for_io() called
95 : : * page_mkclean(), followed by set_page_dirty().
96 : : * However, now the page is going to get written back,
97 : : * which meets the original intention of setting it
98 : : * dirty, so all is well: clear_page_dirty_for_io() goes
99 : : * on to call TestClearPageDirty(), and write the page
100 : : * back.
101 : : *
102 : : * 2) This code sees the page as clean, so it calls
103 : : * set_page_dirty(). The page stays dirty, despite being
104 : : * written back, so it gets written back again in the
105 : : * next writeback cycle. This is harmless.
106 : : */
107 [ # # # # ]: 0 : if (!PageDirty(page))
108 : 0 : set_page_dirty_lock(page);
109 : 0 : unpin_user_page(page);
110 : : }
111 : : }
112 : : EXPORT_SYMBOL(unpin_user_pages_dirty_lock);
113 : :
114 : : /**
115 : : * unpin_user_pages() - release an array of gup-pinned pages.
116 : : * @pages: array of pages to be marked dirty and released.
117 : : * @npages: number of pages in the @pages array.
118 : : *
119 : : * For each page in the @pages array, release the page using unpin_user_page().
120 : : *
121 : : * Please see the unpin_user_page() documentation for details.
122 : : */
123 : 0 : void unpin_user_pages(struct page **pages, unsigned long npages)
124 : : {
125 : 0 : unsigned long index;
126 : :
127 : : /*
128 : : * TODO: this can be optimized for huge pages: if a series of pages is
129 : : * physically contiguous and part of the same compound page, then a
130 : : * single operation to the head page should suffice.
131 : : */
132 [ # # # # ]: 0 : for (index = 0; index < npages; index++)
133 : 0 : unpin_user_page(pages[index]);
134 : 0 : }
135 : : EXPORT_SYMBOL(unpin_user_pages);
136 : :
137 : : #ifdef CONFIG_MMU
138 : 63558 : static struct page *no_page_table(struct vm_area_struct *vma,
139 : : unsigned int flags)
140 : : {
141 : : /*
142 : : * When core dumping an enormous anonymous area that nobody
143 : : * has touched so far, we don't want to allocate unnecessary pages or
144 : : * page tables. Return error instead of NULL to skip handle_mm_fault,
145 : : * then get_dump_page() will return NULL to leave a hole in the dump.
146 : : * But we can only make this optimization where a hole would surely
147 : : * be zero-filled if handle_mm_fault() actually did handle it.
148 : : */
149 [ # # # # : 0 : if ((flags & FOLL_DUMP) && (!vma->vm_ops || !vma->vm_ops->fault))
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # #
# # # # ]
150 : 0 : return ERR_PTR(-EFAULT);
151 : : return NULL;
152 : : }
153 : :
154 : 0 : static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
155 : : pte_t *pte, unsigned int flags)
156 : : {
157 : : /* No page to get reference */
158 : 0 : if (flags & FOLL_GET)
159 : : return -EFAULT;
160 : :
161 [ # # ]: 0 : if (flags & FOLL_TOUCH) {
162 : 0 : pte_t entry = *pte;
163 : :
164 [ # # ]: 0 : if (flags & FOLL_WRITE)
165 : 0 : entry = pte_mkdirty(entry);
166 [ # # ]: 0 : entry = pte_mkyoung(entry);
167 : :
168 [ # # ]: 0 : if (!pte_same(*pte, entry)) {
169 : 0 : set_pte_at(vma->vm_mm, address, pte, entry);
170 : 0 : update_mmu_cache(vma, address, pte);
171 : : }
172 : : }
173 : :
174 : : /* Proper page table entry exists, but no corresponding struct page */
175 : : return -EEXIST;
176 : : }
177 : :
178 : : /*
179 : : * FOLL_FORCE can write to even unwritable pte's, but only
180 : : * after we've gone through a COW cycle and they are dirty.
181 : : */
182 : 192411 : static inline bool can_follow_write_pte(pte_t pte, unsigned int flags)
183 : : {
184 [ - + ]: 192411 : return pte_write(pte) ||
185 [ # # # # ]: 0 : ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte));
186 : : }
187 : :
188 : 198594 : static struct page *follow_page_pte(struct vm_area_struct *vma,
189 : : unsigned long address, pmd_t *pmd, unsigned int flags,
190 : : struct dev_pagemap **pgmap)
191 : : {
192 : 198594 : struct mm_struct *mm = vma->vm_mm;
193 : 198594 : struct page *page;
194 : 198594 : spinlock_t *ptl;
195 : 198594 : pte_t *ptep, pte;
196 : :
197 : : /* FOLL_GET and FOLL_PIN are mutually exclusive. */
198 [ - + + - ]: 198594 : if (WARN_ON_ONCE((flags & (FOLL_PIN | FOLL_GET)) ==
199 : : (FOLL_PIN | FOLL_GET)))
200 : : return ERR_PTR(-EINVAL);
201 : 198594 : retry:
202 [ + - - + ]: 397188 : if (unlikely(pmd_bad(*pmd)))
203 [ # # ]: 0 : return no_page_table(vma, flags);
204 : :
205 [ + - ]: 397188 : ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
206 : 198594 : pte = *ptep;
207 [ - + ]: 198594 : if (!pte_present(pte)) {
208 : 0 : swp_entry_t entry;
209 : : /*
210 : : * KSM's break_ksm() relies upon recognizing a ksm page
211 : : * even while it is being migrated, so for that case we
212 : : * need migration_entry_wait().
213 : : */
214 [ # # ]: 0 : if (likely(!(flags & FOLL_MIGRATION)))
215 : 0 : goto no_page;
216 [ # # ]: 0 : if (pte_none(pte))
217 : 0 : goto no_page;
218 [ # # ]: 0 : entry = pte_to_swp_entry(pte);
219 [ # # ]: 0 : if (!is_migration_entry(entry))
220 : 0 : goto no_page;
221 : 0 : pte_unmap_unlock(ptep, ptl);
222 : 0 : migration_entry_wait(mm, pmd, address);
223 : 0 : goto retry;
224 : : }
225 : 198594 : if ((flags & FOLL_NUMA) && pte_protnone(pte))
226 : : goto no_page;
227 [ + + - + ]: 391005 : if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) {
228 : 0 : pte_unmap_unlock(ptep, ptl);
229 : 0 : return NULL;
230 : : }
231 : :
232 : 198594 : page = vm_normal_page(vma, address, pte);
233 [ - + - - : 198594 : if (!page && pte_devmap(pte) && (flags & FOLL_GET)) {
- - ]
234 : : /*
235 : : * Only return device mapping pages in the FOLL_GET case since
236 : : * they are only valid while holding the pgmap reference.
237 : : */
238 : 0 : *pgmap = get_dev_pagemap(pte_pfn(pte), *pgmap);
239 : 0 : if (*pgmap)
240 : : page = pte_page(pte);
241 : : else
242 : 0 : goto no_page;
243 [ - + ]: 198594 : } else if (unlikely(!page)) {
244 [ # # ]: 0 : if (flags & FOLL_DUMP) {
245 : : /* Avoid special (like zero) pages in core dumps */
246 : 0 : page = ERR_PTR(-EFAULT);
247 : 0 : goto out;
248 : : }
249 : :
250 [ # # # # ]: 0 : if (is_zero_pfn(pte_pfn(pte))) {
251 [ # # ]: 0 : page = pte_page(pte);
252 : : } else {
253 : 0 : int ret;
254 : :
255 [ # # ]: 0 : ret = follow_pfn_pte(vma, address, ptep, flags);
256 : 0 : page = ERR_PTR(ret);
257 : 0 : goto out;
258 : : }
259 : : }
260 : :
261 : 198594 : if (flags & FOLL_SPLIT && PageTransCompound(page)) {
262 : : int ret;
263 : : get_page(page);
264 : : pte_unmap_unlock(ptep, ptl);
265 : : lock_page(page);
266 : : ret = split_huge_page(page);
267 : : unlock_page(page);
268 : : put_page(page);
269 : : if (ret)
270 : : return ERR_PTR(ret);
271 : : goto retry;
272 : : }
273 : :
274 [ + - ]: 198594 : if (flags & FOLL_GET) {
275 [ - + ]: 198594 : if (unlikely(!try_get_page(page))) {
276 : 0 : page = ERR_PTR(-ENOMEM);
277 : 0 : goto out;
278 : : }
279 : : }
280 [ + - ]: 198594 : if (flags & FOLL_TOUCH) {
281 [ + + - + ]: 198594 : if ((flags & FOLL_WRITE) &&
282 [ # # ]: 0 : !pte_dirty(pte) && !PageDirty(page))
283 : 0 : set_page_dirty(page);
284 : : /*
285 : : * pte_mkyoung() would be more correct here, but atomic care
286 : : * is needed to avoid losing the dirty bit: it is easier to use
287 : : * mark_page_accessed().
288 : : */
289 : 198594 : mark_page_accessed(page);
290 : : }
291 [ + - - - ]: 198594 : if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
292 : : /* Do not mlock pte-mapped THP */
293 [ # # ]: 0 : if (PageTransCompound(page))
294 : : goto out;
295 : :
296 : : /*
297 : : * The preliminary mapping check is mainly to avoid the
298 : : * pointless overhead of lock_page on the ZERO_PAGE
299 : : * which might bounce very badly if there is contention.
300 : : *
301 : : * If the page is already locked, we don't need to
302 : : * handle it now - vmscan will handle it later if and
303 : : * when it attempts to reclaim the page.
304 : : */
305 [ # # # # ]: 0 : if (page->mapping && trylock_page(page)) {
306 : 0 : lru_add_drain(); /* push cached pages to LRU */
307 : : /*
308 : : * Because we lock page here, and migration is
309 : : * blocked by the pte's page reference, and we
310 : : * know the page is still mapped, we don't even
311 : : * need to check for file-cache page truncation.
312 : : */
313 : 0 : mlock_vma_page(page);
314 : 0 : unlock_page(page);
315 : : }
316 : : }
317 : 198594 : out:
318 : 198594 : pte_unmap_unlock(ptep, ptl);
319 : 198594 : return page;
320 : 0 : no_page:
321 : 0 : pte_unmap_unlock(ptep, ptl);
322 [ # # ]: 0 : if (!pte_none(pte))
323 : : return NULL;
324 [ # # ]: 0 : return no_page_table(vma, flags);
325 : : }
326 : :
327 : 198594 : static struct page *follow_pmd_mask(struct vm_area_struct *vma,
328 : : unsigned long address, pud_t *pudp,
329 : : unsigned int flags,
330 : : struct follow_page_context *ctx)
331 : : {
332 : 198594 : pmd_t *pmd, pmdval;
333 : 198594 : spinlock_t *ptl;
334 : 198594 : struct page *page;
335 : 198594 : struct mm_struct *mm = vma->vm_mm;
336 : :
337 [ + - ]: 198594 : pmd = pmd_offset(pudp, address);
338 : : /*
339 : : * The READ_ONCE() will stabilize the pmdval in a register or
340 : : * on the stack so that it will stop changing under the code.
341 : : */
342 [ - + ]: 198594 : pmdval = READ_ONCE(*pmd);
343 [ - + ]: 198594 : if (pmd_none(pmdval))
344 [ # # ]: 0 : return no_page_table(vma, flags);
345 [ + - - - ]: 198594 : if (pmd_huge(pmdval) && is_vm_hugetlb_page(vma)) {
346 : 0 : page = follow_huge_pmd(mm, address, pmd, flags);
347 [ # # ]: 0 : if (page)
348 : : return page;
349 [ # # ]: 0 : return no_page_table(vma, flags);
350 : : }
351 : : if (is_hugepd(__hugepd(pmd_val(pmdval)))) {
352 : : page = follow_huge_pd(vma, address,
353 : : __hugepd(pmd_val(pmdval)), flags,
354 : : PMD_SHIFT);
355 : : if (page)
356 : : return page;
357 : : return no_page_table(vma, flags);
358 : : }
359 : : retry:
360 [ + - - + ]: 397188 : if (!pmd_present(pmdval)) {
361 [ # # ]: 0 : if (likely(!(flags & FOLL_MIGRATION)))
362 [ # # ]: 0 : return no_page_table(vma, flags);
363 : 0 : VM_BUG_ON(thp_migration_supported() &&
364 : : !is_pmd_migration_entry(pmdval));
365 [ # # ]: 0 : if (is_pmd_migration_entry(pmdval))
366 : : pmd_migration_entry_wait(mm, pmd);
367 [ # # ]: 0 : pmdval = READ_ONCE(*pmd);
368 : : /*
369 : : * MADV_DONTNEED may convert the pmd to null because
370 : : * mmap_sem is held in read mode
371 : : */
372 [ # # ]: 0 : if (pmd_none(pmdval))
373 [ # # ]: 0 : return no_page_table(vma, flags);
374 : 0 : goto retry;
375 : : }
376 : 198594 : if (pmd_devmap(pmdval)) {
377 : : ptl = pmd_lock(mm, pmd);
378 : : page = follow_devmap_pmd(vma, address, pmd, flags, &ctx->pgmap);
379 : : spin_unlock(ptl);
380 : : if (page)
381 : : return page;
382 : : }
383 : 198594 : if (likely(!pmd_trans_huge(pmdval)))
384 : 198594 : return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
385 : :
386 : : if ((flags & FOLL_NUMA) && pmd_protnone(pmdval))
387 : : return no_page_table(vma, flags);
388 : :
389 : : retry_locked:
390 : : ptl = pmd_lock(mm, pmd);
391 : : if (unlikely(pmd_none(*pmd))) {
392 : : spin_unlock(ptl);
393 : : return no_page_table(vma, flags);
394 : : }
395 : : if (unlikely(!pmd_present(*pmd))) {
396 : : spin_unlock(ptl);
397 : : if (likely(!(flags & FOLL_MIGRATION)))
398 : : return no_page_table(vma, flags);
399 : : pmd_migration_entry_wait(mm, pmd);
400 : : goto retry_locked;
401 : : }
402 : : if (unlikely(!pmd_trans_huge(*pmd))) {
403 : : spin_unlock(ptl);
404 : : return follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
405 : : }
406 : : if (flags & (FOLL_SPLIT | FOLL_SPLIT_PMD)) {
407 : : int ret;
408 : : page = pmd_page(*pmd);
409 : : if (is_huge_zero_page(page)) {
410 : : spin_unlock(ptl);
411 : : ret = 0;
412 : : split_huge_pmd(vma, pmd, address);
413 : : if (pmd_trans_unstable(pmd))
414 : : ret = -EBUSY;
415 : : } else if (flags & FOLL_SPLIT) {
416 : : if (unlikely(!try_get_page(page))) {
417 : : spin_unlock(ptl);
418 : : return ERR_PTR(-ENOMEM);
419 : : }
420 : : spin_unlock(ptl);
421 : : lock_page(page);
422 : : ret = split_huge_page(page);
423 : : unlock_page(page);
424 : : put_page(page);
425 : : if (pmd_none(*pmd))
426 : : return no_page_table(vma, flags);
427 : : } else { /* flags & FOLL_SPLIT_PMD */
428 : : spin_unlock(ptl);
429 : : split_huge_pmd(vma, pmd, address);
430 : : ret = pte_alloc(mm, pmd) ? -ENOMEM : 0;
431 : : }
432 : :
433 : : return ret ? ERR_PTR(ret) :
434 : : follow_page_pte(vma, address, pmd, flags, &ctx->pgmap);
435 : : }
436 : : page = follow_trans_huge_pmd(vma, address, pmd, flags);
437 : : spin_unlock(ptl);
438 : : ctx->page_mask = HPAGE_PMD_NR - 1;
439 : : return page;
440 : : }
441 : :
442 : 198594 : static struct page *follow_pud_mask(struct vm_area_struct *vma,
443 : : unsigned long address, p4d_t *p4dp,
444 : : unsigned int flags,
445 : : struct follow_page_context *ctx)
446 : : {
447 : 198594 : pud_t *pud;
448 : 198594 : spinlock_t *ptl;
449 : 198594 : struct page *page;
450 : 198594 : struct mm_struct *mm = vma->vm_mm;
451 : :
452 [ - + ]: 198594 : pud = pud_offset(p4dp, address);
453 [ - + ]: 198594 : if (pud_none(*pud))
454 [ # # ]: 0 : return no_page_table(vma, flags);
455 [ - + - - ]: 198594 : if (pud_huge(*pud) && is_vm_hugetlb_page(vma)) {
456 : 0 : page = follow_huge_pud(mm, address, pud, flags);
457 [ # # ]: 0 : if (page)
458 : : return page;
459 [ # # ]: 0 : return no_page_table(vma, flags);
460 : : }
461 : 198594 : if (is_hugepd(__hugepd(pud_val(*pud)))) {
462 : : page = follow_huge_pd(vma, address,
463 : : __hugepd(pud_val(*pud)), flags,
464 : : PUD_SHIFT);
465 : : if (page)
466 : : return page;
467 : : return no_page_table(vma, flags);
468 : : }
469 [ + - ]: 198594 : if (pud_devmap(*pud)) {
470 : : ptl = pud_lock(mm, pud);
471 : : page = follow_devmap_pud(vma, address, pud, flags, &ctx->pgmap);
472 : : spin_unlock(ptl);
473 : : if (page)
474 : : return page;
475 : : }
476 [ + - - + ]: 397188 : if (unlikely(pud_bad(*pud)))
477 [ # # ]: 0 : return no_page_table(vma, flags);
478 : :
479 : 198594 : return follow_pmd_mask(vma, address, pud, flags, ctx);
480 : : }
481 : :
482 : 262152 : static struct page *follow_p4d_mask(struct vm_area_struct *vma,
483 : : unsigned long address, pgd_t *pgdp,
484 : : unsigned int flags,
485 : : struct follow_page_context *ctx)
486 : : {
487 : 262152 : p4d_t *p4d;
488 : 262152 : struct page *page;
489 : :
490 : 262152 : p4d = p4d_offset(pgdp, address);
491 [ + + ]: 262152 : if (p4d_none(*p4d))
492 [ - + ]: 63558 : return no_page_table(vma, flags);
493 : 198594 : BUILD_BUG_ON(p4d_huge(*p4d));
494 [ - + ]: 198594 : if (unlikely(p4d_bad(*p4d)))
495 [ # # ]: 0 : return no_page_table(vma, flags);
496 : :
497 : 198594 : if (is_hugepd(__hugepd(p4d_val(*p4d)))) {
498 : : page = follow_huge_pd(vma, address,
499 : : __hugepd(p4d_val(*p4d)), flags,
500 : : P4D_SHIFT);
501 : : if (page)
502 : : return page;
503 : : return no_page_table(vma, flags);
504 : : }
505 : 198594 : return follow_pud_mask(vma, address, p4d, flags, ctx);
506 : : }
507 : :
508 : : /**
509 : : * follow_page_mask - look up a page descriptor from a user-virtual address
510 : : * @vma: vm_area_struct mapping @address
511 : : * @address: virtual address to look up
512 : : * @flags: flags modifying lookup behaviour
513 : : * @ctx: contains dev_pagemap for %ZONE_DEVICE memory pinning and a
514 : : * pointer to output page_mask
515 : : *
516 : : * @flags can have FOLL_ flags set, defined in <linux/mm.h>
517 : : *
518 : : * When getting pages from ZONE_DEVICE memory, the @ctx->pgmap caches
519 : : * the device's dev_pagemap metadata to avoid repeating expensive lookups.
520 : : *
521 : : * On output, the @ctx->page_mask is set according to the size of the page.
522 : : *
523 : : * Return: the mapped (struct page *), %NULL if no mapping exists, or
524 : : * an error pointer if there is a mapping to something not represented
525 : : * by a page descriptor (see also vm_normal_page()).
526 : : */
527 : 262152 : static struct page *follow_page_mask(struct vm_area_struct *vma,
528 : : unsigned long address, unsigned int flags,
529 : : struct follow_page_context *ctx)
530 : : {
531 : 262152 : pgd_t *pgd;
532 : 262152 : struct page *page;
533 : 262152 : struct mm_struct *mm = vma->vm_mm;
534 : :
535 : 262152 : ctx->page_mask = 0;
536 : :
537 : : /* make this handle hugepd */
538 : 262152 : page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
539 [ - + ]: 262152 : if (!IS_ERR(page)) {
540 [ # # ]: 0 : BUG_ON(flags & FOLL_GET);
541 : : return page;
542 : : }
543 : :
544 : 262152 : pgd = pgd_offset(mm, address);
545 : :
546 [ + - - + ]: 262152 : if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
547 [ # # ]: 0 : return no_page_table(vma, flags);
548 : :
549 : 262152 : if (pgd_huge(*pgd)) {
550 : : page = follow_huge_pgd(mm, address, pgd, flags);
551 : : if (page)
552 : : return page;
553 : : return no_page_table(vma, flags);
554 : : }
555 : 262152 : if (is_hugepd(__hugepd(pgd_val(*pgd)))) {
556 : : page = follow_huge_pd(vma, address,
557 : : __hugepd(pgd_val(*pgd)), flags,
558 : : PGDIR_SHIFT);
559 : : if (page)
560 : : return page;
561 : : return no_page_table(vma, flags);
562 : : }
563 : :
564 : 262152 : return follow_p4d_mask(vma, address, pgd, flags, ctx);
565 : : }
566 : :
567 : 0 : struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
568 : : unsigned int foll_flags)
569 : : {
570 : 0 : struct follow_page_context ctx = { NULL };
571 : 0 : struct page *page;
572 : :
573 : 0 : page = follow_page_mask(vma, address, foll_flags, &ctx);
574 [ # # ]: 0 : if (ctx.pgmap)
575 : 0 : put_dev_pagemap(ctx.pgmap);
576 : 0 : return page;
577 : : }
578 : :
579 : 0 : static int get_gate_page(struct mm_struct *mm, unsigned long address,
580 : : unsigned int gup_flags, struct vm_area_struct **vma,
581 : : struct page **page)
582 : : {
583 : 0 : pgd_t *pgd;
584 : 0 : p4d_t *p4d;
585 : 0 : pud_t *pud;
586 : 0 : pmd_t *pmd;
587 : 0 : pte_t *pte;
588 : 0 : int ret = -EFAULT;
589 : :
590 : : /* user gate pages are read-only */
591 [ # # ]: 0 : if (gup_flags & FOLL_WRITE)
592 : : return -EFAULT;
593 [ # # # # : 0 : if (address > TASK_SIZE)
# # ]
594 : 0 : pgd = pgd_offset_k(address);
595 : : else
596 : 0 : pgd = pgd_offset_gate(mm, address);
597 [ # # ]: 0 : if (pgd_none(*pgd))
598 : : return -EFAULT;
599 : 0 : p4d = p4d_offset(pgd, address);
600 [ # # ]: 0 : if (p4d_none(*p4d))
601 : : return -EFAULT;
602 [ # # ]: 0 : pud = pud_offset(p4d, address);
603 [ # # ]: 0 : if (pud_none(*pud))
604 : : return -EFAULT;
605 [ # # ]: 0 : pmd = pmd_offset(pud, address);
606 [ # # # # ]: 0 : if (!pmd_present(*pmd))
607 : : return -EFAULT;
608 : 0 : VM_BUG_ON(pmd_trans_huge(*pmd));
609 [ # # ]: 0 : pte = pte_offset_map(pmd, address);
610 [ # # ]: 0 : if (pte_none(*pte))
611 : 0 : goto unmap;
612 : 0 : *vma = get_gate_vma(mm);
613 [ # # ]: 0 : if (!page)
614 : 0 : goto out;
615 : 0 : *page = vm_normal_page(*vma, address, *pte);
616 [ # # ]: 0 : if (!*page) {
617 [ # # # # ]: 0 : if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte)))
618 : 0 : goto unmap;
619 [ # # ]: 0 : *page = pte_page(*pte);
620 : : }
621 [ # # ]: 0 : if (unlikely(!try_get_page(*page))) {
622 : 0 : ret = -ENOMEM;
623 : 0 : goto unmap;
624 : : }
625 : 0 : out:
626 : : ret = 0;
627 : : unmap:
628 : : pte_unmap(pte);
629 : : return ret;
630 : : }
631 : :
632 : : /*
633 : : * mmap_sem must be held on entry. If @nonblocking != NULL and
634 : : * *@flags does not include FOLL_NOWAIT, the mmap_sem may be released.
635 : : * If it is, *@nonblocking will be set to 0 and -EBUSY returned.
636 : : */
637 : 63558 : static int faultin_page(struct task_struct *tsk, struct vm_area_struct *vma,
638 : : unsigned long address, unsigned int *flags, int *nonblocking)
639 : : {
640 : 63558 : unsigned int fault_flags = 0;
641 : 63558 : vm_fault_t ret;
642 : :
643 : : /* mlock all present pages, but do not fault in new pages */
644 [ + - ]: 63558 : if ((*flags & (FOLL_POPULATE | FOLL_MLOCK)) == FOLL_MLOCK)
645 : : return -ENOENT;
646 : 63558 : if (*flags & FOLL_WRITE)
647 : : fault_flags |= FAULT_FLAG_WRITE;
648 [ + - ]: 63558 : if (*flags & FOLL_REMOTE)
649 : 63558 : fault_flags |= FAULT_FLAG_REMOTE;
650 [ - + ]: 63558 : if (nonblocking)
651 : 0 : fault_flags |= FAULT_FLAG_ALLOW_RETRY;
652 [ - + ]: 63558 : if (*flags & FOLL_NOWAIT)
653 : 0 : fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
654 [ - + ]: 63558 : if (*flags & FOLL_TRIED) {
655 : 0 : VM_WARN_ON_ONCE(fault_flags & FAULT_FLAG_ALLOW_RETRY);
656 : 0 : fault_flags |= FAULT_FLAG_TRIED;
657 : : }
658 : :
659 : 63558 : ret = handle_mm_fault(vma, address, fault_flags);
660 [ - + ]: 63558 : if (ret & VM_FAULT_ERROR) {
661 [ # # ]: 0 : int err = vm_fault_to_errno(ret, *flags);
662 : :
663 : 0 : if (err)
664 : 0 : return err;
665 : 0 : BUG();
666 : : }
667 : :
668 [ + - ]: 63558 : if (tsk) {
669 [ - + ]: 63558 : if (ret & VM_FAULT_MAJOR)
670 : 0 : tsk->maj_flt++;
671 : : else
672 : 63558 : tsk->min_flt++;
673 : : }
674 : :
675 [ - + ]: 63558 : if (ret & VM_FAULT_RETRY) {
676 [ # # # # ]: 0 : if (nonblocking && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
677 : 0 : *nonblocking = 0;
678 : 0 : return -EBUSY;
679 : : }
680 : :
681 : : /*
682 : : * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when
683 : : * necessary, even if maybe_mkwrite decided not to set pte_write. We
684 : : * can thus safely do subsequent page lookups as if they were reads.
685 : : * But only do so when looping for pte_write is futile: in some cases
686 : : * userspace may also be wanting to write to the gotten user page,
687 : : * which a read fault here might prevent (a readonly page might get
688 : : * reCOWed by userspace write).
689 : : */
690 [ - + - - ]: 63558 : if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE))
691 : 0 : *flags |= FOLL_COW;
692 : : return 0;
693 : : }
694 : :
695 : 198594 : static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
696 : : {
697 : 198594 : vm_flags_t vm_flags = vma->vm_flags;
698 : 198594 : int write = (gup_flags & FOLL_WRITE);
699 : 198594 : int foreign = (gup_flags & FOLL_REMOTE);
700 : :
701 [ + - ]: 198594 : if (vm_flags & (VM_IO | VM_PFNMAP))
702 : : return -EFAULT;
703 : :
704 [ + + + - ]: 198594 : if (gup_flags & FOLL_ANON && !vma_is_anonymous(vma))
705 : : return -EFAULT;
706 : :
707 [ + + ]: 198594 : if (write) {
708 [ - + ]: 192411 : if (!(vm_flags & VM_WRITE)) {
709 [ # # ]: 0 : if (!(gup_flags & FOLL_FORCE))
710 : : return -EFAULT;
711 : : /*
712 : : * We used to let the write,force case do COW in a
713 : : * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could
714 : : * set a breakpoint in a read-only mapping of an
715 : : * executable, without corrupting the file (yet only
716 : : * when that file had been opened for writing!).
717 : : * Anon pages in shared mappings are surprising: now
718 : : * just reject it.
719 : : */
720 [ # # ]: 0 : if (!is_cow_mapping(vm_flags))
721 : : return -EFAULT;
722 : : }
723 [ - + ]: 6183 : } else if (!(vm_flags & VM_READ)) {
724 [ # # ]: 0 : if (!(gup_flags & FOLL_FORCE))
725 : : return -EFAULT;
726 : : /*
727 : : * Is there actually any vma we can reach here which does not
728 : : * have VM_MAYREAD set?
729 : : */
730 [ # # ]: 0 : if (!(vm_flags & VM_MAYREAD))
731 : : return -EFAULT;
732 : : }
733 : : /*
734 : : * gups are always data accesses, not instruction
735 : : * fetches, so execute=false here
736 : : */
737 [ - + ]: 198594 : if (!arch_vma_access_permitted(vma, write, false, foreign))
738 : 0 : return -EFAULT;
739 : : return 0;
740 : : }
741 : :
742 : : /**
743 : : * __get_user_pages() - pin user pages in memory
744 : : * @tsk: task_struct of target task
745 : : * @mm: mm_struct of target mm
746 : : * @start: starting user address
747 : : * @nr_pages: number of pages from start to pin
748 : : * @gup_flags: flags modifying pin behaviour
749 : : * @pages: array that receives pointers to the pages pinned.
750 : : * Should be at least nr_pages long. Or NULL, if caller
751 : : * only intends to ensure the pages are faulted in.
752 : : * @vmas: array of pointers to vmas corresponding to each page.
753 : : * Or NULL if the caller does not require them.
754 : : * @nonblocking: whether waiting for disk IO or mmap_sem contention
755 : : *
756 : : * Returns either number of pages pinned (which may be less than the
757 : : * number requested), or an error. Details about the return value:
758 : : *
759 : : * -- If nr_pages is 0, returns 0.
760 : : * -- If nr_pages is >0, but no pages were pinned, returns -errno.
761 : : * -- If nr_pages is >0, and some pages were pinned, returns the number of
762 : : * pages pinned. Again, this may be less than nr_pages.
763 : : *
764 : : * The caller is responsible for releasing returned @pages, via put_page().
765 : : *
766 : : * @vmas are valid only as long as mmap_sem is held.
767 : : *
768 : : * Must be called with mmap_sem held. It may be released. See below.
769 : : *
770 : : * __get_user_pages walks a process's page tables and takes a reference to
771 : : * each struct page that each user address corresponds to at a given
772 : : * instant. That is, it takes the page that would be accessed if a user
773 : : * thread accesses the given user virtual address at that instant.
774 : : *
775 : : * This does not guarantee that the page exists in the user mappings when
776 : : * __get_user_pages returns, and there may even be a completely different
777 : : * page there in some cases (eg. if mmapped pagecache has been invalidated
778 : : * and subsequently re faulted). However it does guarantee that the page
779 : : * won't be freed completely. And mostly callers simply care that the page
780 : : * contains data that was valid *at some point in time*. Typically, an IO
781 : : * or similar operation cannot guarantee anything stronger anyway because
782 : : * locks can't be held over the syscall boundary.
783 : : *
784 : : * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
785 : : * the page is written to, set_page_dirty (or set_page_dirty_lock, as
786 : : * appropriate) must be called after the page is finished with, and
787 : : * before put_page is called.
788 : : *
789 : : * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
790 : : * or mmap_sem contention, and if waiting is needed to pin all pages,
791 : : * *@nonblocking will be set to 0. Further, if @gup_flags does not
792 : : * include FOLL_NOWAIT, the mmap_sem will be released via up_read() in
793 : : * this case.
794 : : *
795 : : * A caller using such a combination of @nonblocking and @gup_flags
796 : : * must therefore hold the mmap_sem for reading only, and recognize
797 : : * when it's been released. Otherwise, it must be held for either
798 : : * reading or writing and will not be released.
799 : : *
800 : : * In most cases, get_user_pages or get_user_pages_fast should be used
801 : : * instead of __get_user_pages. __get_user_pages should be used only if
802 : : * you need some special @gup_flags.
803 : : */
804 : 198594 : static long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
805 : : unsigned long start, unsigned long nr_pages,
806 : : unsigned int gup_flags, struct page **pages,
807 : : struct vm_area_struct **vmas, int *nonblocking)
808 : : {
809 : 198594 : long ret = 0, i = 0;
810 : 198594 : struct vm_area_struct *vma = NULL;
811 : 198594 : struct follow_page_context ctx = { NULL };
812 : :
813 [ + - ]: 198594 : if (!nr_pages)
814 : : return 0;
815 : :
816 : 198594 : start = untagged_addr(start);
817 : :
818 : 198594 : VM_BUG_ON(!!pages != !!(gup_flags & (FOLL_GET | FOLL_PIN)));
819 : :
820 : : /*
821 : : * If FOLL_FORCE is set then do not force a full fault as the hinting
822 : : * fault information is unrelated to the reference behaviour of a task
823 : : * using the address space
824 : : */
825 [ + + ]: 198594 : if (!(gup_flags & FOLL_FORCE))
826 : 5448 : gup_flags |= FOLL_NUMA;
827 : :
828 : 198594 : do {
829 : 198594 : struct page *page;
830 : 198594 : unsigned int foll_flags = gup_flags;
831 : 198594 : unsigned int page_increm;
832 : :
833 : : /* first iteration or cross vma bound */
834 [ - + - - ]: 198594 : if (!vma || start >= vma->vm_end) {
835 : 198594 : vma = find_extend_vma(mm, start);
836 [ - + - - ]: 198594 : if (!vma && in_gate_area(mm, start)) {
837 [ # # ]: 0 : ret = get_gate_page(mm, start & PAGE_MASK,
838 : : gup_flags, &vma,
839 : 0 : pages ? &pages[i] : NULL);
840 [ # # ]: 0 : if (ret)
841 : 0 : goto out;
842 : 0 : ctx.page_mask = 0;
843 : 0 : goto next_page;
844 : : }
845 : :
846 [ + - - + ]: 198594 : if (!vma || check_vma_flags(vma, gup_flags)) {
847 : 0 : ret = -EFAULT;
848 : 0 : goto out;
849 : : }
850 [ + - ]: 198594 : if (is_vm_hugetlb_page(vma)) {
851 : 0 : i = follow_hugetlb_page(mm, vma, pages, vmas,
852 : : &start, &nr_pages, i,
853 : : gup_flags, nonblocking);
854 : 0 : continue;
855 : : }
856 : : }
857 : 198594 : retry:
858 : : /*
859 : : * If we have a pending SIGKILL, don't keep faulting pages and
860 : : * potentially allocating memory.
861 : : */
862 [ - + ]: 262152 : if (fatal_signal_pending(current)) {
863 : 0 : ret = -ERESTARTSYS;
864 : 0 : goto out;
865 : : }
866 : 262152 : cond_resched();
867 : :
868 : 262152 : page = follow_page_mask(vma, start, foll_flags, &ctx);
869 [ + + ]: 262152 : if (!page) {
870 : 63558 : ret = faultin_page(tsk, vma, start, &foll_flags,
871 : : nonblocking);
872 [ + - - - : 63558 : switch (ret) {
- ]
873 : 63558 : case 0:
874 : 63558 : goto retry;
875 : : case -EBUSY:
876 : : ret = 0;
877 : : /* FALLTHRU */
878 : 0 : case -EFAULT:
879 : : case -ENOMEM:
880 : : case -EHWPOISON:
881 : 0 : goto out;
882 : 0 : case -ENOENT:
883 : 0 : goto next_page;
884 : : }
885 : 0 : BUG();
886 [ - + ]: 198594 : } else if (PTR_ERR(page) == -EEXIST) {
887 : : /*
888 : : * Proper page table entry exists, but no corresponding
889 : : * struct page.
890 : : */
891 : 0 : goto next_page;
892 [ - + ]: 198594 : } else if (IS_ERR(page)) {
893 : 0 : ret = PTR_ERR(page);
894 : 0 : goto out;
895 : : }
896 [ - + ]: 198594 : if (pages) {
897 : 198594 : pages[i] = page;
898 : 198594 : flush_anon_page(vma, page, start);
899 : 198594 : flush_dcache_page(page);
900 : 198594 : ctx.page_mask = 0;
901 : : }
902 : 0 : next_page:
903 [ + + ]: 198594 : if (vmas) {
904 : 5448 : vmas[i] = vma;
905 : 5448 : ctx.page_mask = 0;
906 : : }
907 : 198594 : page_increm = 1 + (~(start >> PAGE_SHIFT) & ctx.page_mask);
908 [ - + ]: 198594 : if (page_increm > nr_pages)
909 : 0 : page_increm = nr_pages;
910 : 198594 : i += page_increm;
911 : 198594 : start += page_increm * PAGE_SIZE;
912 : 198594 : nr_pages -= page_increm;
913 [ - + ]: 198594 : } while (nr_pages);
914 : 198594 : out:
915 [ - + ]: 198594 : if (ctx.pgmap)
916 : 0 : put_dev_pagemap(ctx.pgmap);
917 [ - + ]: 198594 : return i ? i : ret;
918 : : }
919 : :
920 : 0 : static bool vma_permits_fault(struct vm_area_struct *vma,
921 : : unsigned int fault_flags)
922 : : {
923 : 0 : bool write = !!(fault_flags & FAULT_FLAG_WRITE);
924 : 0 : bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE);
925 : 0 : vm_flags_t vm_flags = write ? VM_WRITE : VM_READ;
926 : :
927 [ # # ]: 0 : if (!(vm_flags & vma->vm_flags))
928 : : return false;
929 : :
930 : : /*
931 : : * The architecture might have a hardware protection
932 : : * mechanism other than read/write that can deny access.
933 : : *
934 : : * gup always represents data access, not instruction
935 : : * fetches, so execute=false here:
936 : : */
937 [ # # ]: 0 : if (!arch_vma_access_permitted(vma, write, false, foreign))
938 : 0 : return false;
939 : :
940 : : return true;
941 : : }
942 : :
943 : : /*
944 : : * fixup_user_fault() - manually resolve a user page fault
945 : : * @tsk: the task_struct to use for page fault accounting, or
946 : : * NULL if faults are not to be recorded.
947 : : * @mm: mm_struct of target mm
948 : : * @address: user address
949 : : * @fault_flags:flags to pass down to handle_mm_fault()
950 : : * @unlocked: did we unlock the mmap_sem while retrying, maybe NULL if caller
951 : : * does not allow retry
952 : : *
953 : : * This is meant to be called in the specific scenario where for locking reasons
954 : : * we try to access user memory in atomic context (within a pagefault_disable()
955 : : * section), this returns -EFAULT, and we want to resolve the user fault before
956 : : * trying again.
957 : : *
958 : : * Typically this is meant to be used by the futex code.
959 : : *
960 : : * The main difference with get_user_pages() is that this function will
961 : : * unconditionally call handle_mm_fault() which will in turn perform all the
962 : : * necessary SW fixup of the dirty and young bits in the PTE, while
963 : : * get_user_pages() only guarantees to update these in the struct page.
964 : : *
965 : : * This is important for some architectures where those bits also gate the
966 : : * access permission to the page because they are maintained in software. On
967 : : * such architectures, gup() will not be enough to make a subsequent access
968 : : * succeed.
969 : : *
970 : : * This function will not return with an unlocked mmap_sem. So it has not the
971 : : * same semantics wrt the @mm->mmap_sem as does filemap_fault().
972 : : */
973 : 0 : int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
974 : : unsigned long address, unsigned int fault_flags,
975 : : bool *unlocked)
976 : : {
977 : 0 : struct vm_area_struct *vma;
978 : 0 : vm_fault_t ret, major = 0;
979 : :
980 : 0 : address = untagged_addr(address);
981 : :
982 [ # # ]: 0 : if (unlocked)
983 : 0 : fault_flags |= FAULT_FLAG_ALLOW_RETRY;
984 : :
985 : 0 : retry:
986 : 0 : vma = find_extend_vma(mm, address);
987 [ # # # # ]: 0 : if (!vma || address < vma->vm_start)
988 : : return -EFAULT;
989 : :
990 [ # # ]: 0 : if (!vma_permits_fault(vma, fault_flags))
991 : : return -EFAULT;
992 : :
993 : 0 : ret = handle_mm_fault(vma, address, fault_flags);
994 : 0 : major |= ret & VM_FAULT_MAJOR;
995 [ # # ]: 0 : if (ret & VM_FAULT_ERROR) {
996 [ # # ]: 0 : int err = vm_fault_to_errno(ret, 0);
997 : :
998 : 0 : if (err)
999 : 0 : return err;
1000 : 0 : BUG();
1001 : : }
1002 : :
1003 [ # # ]: 0 : if (ret & VM_FAULT_RETRY) {
1004 : 0 : down_read(&mm->mmap_sem);
1005 [ # # ]: 0 : if (!(fault_flags & FAULT_FLAG_TRIED)) {
1006 : 0 : *unlocked = true;
1007 : 0 : fault_flags &= ~FAULT_FLAG_ALLOW_RETRY;
1008 : 0 : fault_flags |= FAULT_FLAG_TRIED;
1009 : 0 : goto retry;
1010 : : }
1011 : : }
1012 : :
1013 [ # # ]: 0 : if (tsk) {
1014 [ # # ]: 0 : if (major)
1015 : 0 : tsk->maj_flt++;
1016 : : else
1017 : 0 : tsk->min_flt++;
1018 : : }
1019 : : return 0;
1020 : : }
1021 : : EXPORT_SYMBOL_GPL(fixup_user_fault);
1022 : :
1023 : 198594 : static __always_inline long __get_user_pages_locked(struct task_struct *tsk,
1024 : : struct mm_struct *mm,
1025 : : unsigned long start,
1026 : : unsigned long nr_pages,
1027 : : struct page **pages,
1028 : : struct vm_area_struct **vmas,
1029 : : int *locked,
1030 : : unsigned int flags)
1031 : : {
1032 : 198594 : long ret, pages_done;
1033 : 198594 : bool lock_dropped;
1034 : :
1035 [ # # ]: 0 : if (locked) {
1036 : : /* if VM_FAULT_RETRY can be returned, vmas become invalid */
1037 [ # # ]: 0 : BUG_ON(vmas);
1038 : : /* check caller initialized locked */
1039 [ # # # # : 0 : BUG_ON(*locked != 1);
# # ]
1040 : : }
1041 : :
1042 : : /*
1043 : : * FOLL_PIN and FOLL_GET are mutually exclusive. Traditional behavior
1044 : : * is to set FOLL_GET if the caller wants pages[] filled in (but has
1045 : : * carelessly failed to specify FOLL_GET), so keep doing that, but only
1046 : : * for FOLL_GET, not for the newer FOLL_PIN.
1047 : : *
1048 : : * FOLL_PIN always expects pages to be non-null, but no need to assert
1049 : : * that here, as any failures will be obvious enough.
1050 : : */
1051 [ - - - - : 198594 : if (pages && !(flags & FOLL_PIN))
- - - - -
- - - - -
+ - + - ]
1052 : 198594 : flags |= FOLL_GET;
1053 : :
1054 : 198594 : pages_done = 0;
1055 : 198594 : lock_dropped = false;
1056 : 198594 : for (;;) {
1057 : 198594 : ret = __get_user_pages(tsk, mm, start, nr_pages, flags, pages,
1058 : : vmas, locked);
1059 [ - - - + ]: 198594 : if (!locked)
1060 : : /* VM_FAULT_RETRY couldn't trigger, bypass */
1061 : 0 : return ret;
1062 : :
1063 : : /* VM_FAULT_RETRY cannot return errors */
1064 [ # # # # : 0 : if (!*locked) {
# # ]
1065 [ # # # # : 0 : BUG_ON(ret < 0);
# # ]
1066 [ # # # # : 0 : BUG_ON(ret >= nr_pages);
# # ]
1067 : : }
1068 : :
1069 [ # # # # : 0 : if (ret > 0) {
# # ]
1070 : 0 : nr_pages -= ret;
1071 : 0 : pages_done += ret;
1072 [ # # # # : 0 : if (!nr_pages)
# # ]
1073 : : break;
1074 : : }
1075 [ # # # # : 0 : if (*locked) {
# # ]
1076 : : /*
1077 : : * VM_FAULT_RETRY didn't trigger or it was a
1078 : : * FOLL_NOWAIT.
1079 : : */
1080 [ # # # # : 0 : if (!pages_done)
# # ]
1081 : 0 : pages_done = ret;
1082 : : break;
1083 : : }
1084 : : /*
1085 : : * VM_FAULT_RETRY triggered, so seek to the faulting offset.
1086 : : * For the prefault case (!pages) we only update counts.
1087 : : */
1088 [ # # # # : 0 : if (likely(pages))
# # ]
1089 : 0 : pages += ret;
1090 : 0 : start += ret << PAGE_SHIFT;
1091 : :
1092 : : /*
1093 : : * Repeat on the address that fired VM_FAULT_RETRY
1094 : : * without FAULT_FLAG_ALLOW_RETRY but with
1095 : : * FAULT_FLAG_TRIED.
1096 : : */
1097 : 0 : *locked = 1;
1098 : 0 : lock_dropped = true;
1099 : 0 : down_read(&mm->mmap_sem);
1100 : 0 : ret = __get_user_pages(tsk, mm, start, 1, flags | FOLL_TRIED,
1101 : : pages, NULL, NULL);
1102 [ # # # # : 0 : if (ret != 1) {
# # ]
1103 [ # # # # : 0 : BUG_ON(ret > 1);
# # ]
1104 [ # # # # : 0 : if (!pages_done)
# # ]
1105 : : pages_done = ret;
1106 : : break;
1107 : : }
1108 : 0 : nr_pages--;
1109 : 0 : pages_done++;
1110 [ # # # # : 0 : if (!nr_pages)
# # ]
1111 : : break;
1112 [ # # # # : 0 : if (likely(pages))
# # ]
1113 : 0 : pages++;
1114 : 0 : start += PAGE_SIZE;
1115 : : }
1116 [ # # # # : 0 : if (lock_dropped && *locked) {
# # # # #
# # # ]
1117 : : /*
1118 : : * We must let the caller know we temporarily dropped the lock
1119 : : * and so the critical section protected by it was lost.
1120 : : */
1121 : 0 : up_read(&mm->mmap_sem);
1122 : 0 : *locked = 0;
1123 : : }
1124 : : return pages_done;
1125 : : }
1126 : :
1127 : : /**
1128 : : * populate_vma_page_range() - populate a range of pages in the vma.
1129 : : * @vma: target vma
1130 : : * @start: start address
1131 : : * @end: end address
1132 : : * @nonblocking:
1133 : : *
1134 : : * This takes care of mlocking the pages too if VM_LOCKED is set.
1135 : : *
1136 : : * return 0 on success, negative error code on error.
1137 : : *
1138 : : * vma->vm_mm->mmap_sem must be held.
1139 : : *
1140 : : * If @nonblocking is NULL, it may be held for read or write and will
1141 : : * be unperturbed.
1142 : : *
1143 : : * If @nonblocking is non-NULL, it must held for read only and may be
1144 : : * released. If it's released, *@nonblocking will be set to 0.
1145 : : */
1146 : 0 : long populate_vma_page_range(struct vm_area_struct *vma,
1147 : : unsigned long start, unsigned long end, int *nonblocking)
1148 : : {
1149 : 0 : struct mm_struct *mm = vma->vm_mm;
1150 : 0 : unsigned long nr_pages = (end - start) / PAGE_SIZE;
1151 : 0 : int gup_flags;
1152 : :
1153 : 0 : VM_BUG_ON(start & ~PAGE_MASK);
1154 : 0 : VM_BUG_ON(end & ~PAGE_MASK);
1155 : 0 : VM_BUG_ON_VMA(start < vma->vm_start, vma);
1156 : 0 : VM_BUG_ON_VMA(end > vma->vm_end, vma);
1157 : 0 : VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
1158 : :
1159 : 0 : gup_flags = FOLL_TOUCH | FOLL_POPULATE | FOLL_MLOCK;
1160 [ # # ]: 0 : if (vma->vm_flags & VM_LOCKONFAULT)
1161 : 0 : gup_flags &= ~FOLL_POPULATE;
1162 : : /*
1163 : : * We want to touch writable mappings with a write fault in order
1164 : : * to break COW, except for shared mappings because these don't COW
1165 : : * and we would not want to dirty them for nothing.
1166 : : */
1167 [ # # ]: 0 : if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
1168 : 0 : gup_flags |= FOLL_WRITE;
1169 : :
1170 : : /*
1171 : : * We want mlock to succeed for regions that have any permissions
1172 : : * other than PROT_NONE.
1173 : : */
1174 [ # # ]: 0 : if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
1175 : 0 : gup_flags |= FOLL_FORCE;
1176 : :
1177 : : /*
1178 : : * We made sure addr is within a VMA, so the following will
1179 : : * not result in a stack expansion that recurses back here.
1180 : : */
1181 : 0 : return __get_user_pages(current, mm, start, nr_pages, gup_flags,
1182 : : NULL, NULL, nonblocking);
1183 : : }
1184 : :
1185 : : /*
1186 : : * __mm_populate - populate and/or mlock pages within a range of address space.
1187 : : *
1188 : : * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
1189 : : * flags. VMAs must be already marked with the desired vm_flags, and
1190 : : * mmap_sem must not be held.
1191 : : */
1192 : 0 : int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
1193 : : {
1194 : 0 : struct mm_struct *mm = current->mm;
1195 : 0 : unsigned long end, nstart, nend;
1196 : 0 : struct vm_area_struct *vma = NULL;
1197 : 0 : int locked = 0;
1198 : 0 : long ret = 0;
1199 : :
1200 : 0 : end = start + len;
1201 : :
1202 [ # # ]: 0 : for (nstart = start; nstart < end; nstart = nend) {
1203 : : /*
1204 : : * We want to fault in pages for [nstart; end) address range.
1205 : : * Find first corresponding VMA.
1206 : : */
1207 [ # # ]: 0 : if (!locked) {
1208 : 0 : locked = 1;
1209 : 0 : down_read(&mm->mmap_sem);
1210 : 0 : vma = find_vma(mm, nstart);
1211 [ # # ]: 0 : } else if (nstart >= vma->vm_end)
1212 : 0 : vma = vma->vm_next;
1213 [ # # # # ]: 0 : if (!vma || vma->vm_start >= end)
1214 : : break;
1215 : : /*
1216 : : * Set [nstart; nend) to intersection of desired address
1217 : : * range with the first VMA. Also, skip undesirable VMA types.
1218 : : */
1219 : 0 : nend = min(end, vma->vm_end);
1220 [ # # ]: 0 : if (vma->vm_flags & (VM_IO | VM_PFNMAP))
1221 : 0 : continue;
1222 : 0 : if (nstart < vma->vm_start)
1223 : : nstart = vma->vm_start;
1224 : : /*
1225 : : * Now fault in a range of pages. populate_vma_page_range()
1226 : : * double checks the vma flags, so that it won't mlock pages
1227 : : * if the vma was already munlocked.
1228 : : */
1229 : 0 : ret = populate_vma_page_range(vma, nstart, nend, &locked);
1230 [ # # ]: 0 : if (ret < 0) {
1231 [ # # ]: 0 : if (ignore_errors) {
1232 : 0 : ret = 0;
1233 : 0 : continue; /* continue at next VMA */
1234 : : }
1235 : : break;
1236 : : }
1237 : 0 : nend = nstart + ret * PAGE_SIZE;
1238 : 0 : ret = 0;
1239 : : }
1240 [ # # ]: 0 : if (locked)
1241 : 0 : up_read(&mm->mmap_sem);
1242 : 0 : return ret; /* 0 or negative error code */
1243 : : }
1244 : :
1245 : : /**
1246 : : * get_dump_page() - pin user page in memory while writing it to core dump
1247 : : * @addr: user address
1248 : : *
1249 : : * Returns struct page pointer of user page pinned for dump,
1250 : : * to be freed afterwards by put_page().
1251 : : *
1252 : : * Returns NULL on any kind of failure - a hole must then be inserted into
1253 : : * the corefile, to preserve alignment with its headers; and also returns
1254 : : * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1255 : : * allowing a hole to be left in the corefile to save diskspace.
1256 : : *
1257 : : * Called without mmap_sem, but after all other threads have been killed.
1258 : : */
1259 : : #ifdef CONFIG_ELF_CORE
1260 : 0 : struct page *get_dump_page(unsigned long addr)
1261 : : {
1262 : 0 : struct vm_area_struct *vma;
1263 : 0 : struct page *page;
1264 : :
1265 [ # # ]: 0 : if (__get_user_pages(current, current->mm, addr, 1,
1266 : : FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,
1267 : : NULL) < 1)
1268 : : return NULL;
1269 : 0 : flush_cache_page(vma, addr, page_to_pfn(page));
1270 : 0 : return page;
1271 : : }
1272 : : #endif /* CONFIG_ELF_CORE */
1273 : : #else /* CONFIG_MMU */
1274 : : static long __get_user_pages_locked(struct task_struct *tsk,
1275 : : struct mm_struct *mm, unsigned long start,
1276 : : unsigned long nr_pages, struct page **pages,
1277 : : struct vm_area_struct **vmas, int *locked,
1278 : : unsigned int foll_flags)
1279 : : {
1280 : : struct vm_area_struct *vma;
1281 : : unsigned long vm_flags;
1282 : : int i;
1283 : :
1284 : : /* calculate required read or write permissions.
1285 : : * If FOLL_FORCE is set, we only require the "MAY" flags.
1286 : : */
1287 : : vm_flags = (foll_flags & FOLL_WRITE) ?
1288 : : (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1289 : : vm_flags &= (foll_flags & FOLL_FORCE) ?
1290 : : (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1291 : :
1292 : : for (i = 0; i < nr_pages; i++) {
1293 : : vma = find_vma(mm, start);
1294 : : if (!vma)
1295 : : goto finish_or_fault;
1296 : :
1297 : : /* protect what we can, including chardevs */
1298 : : if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1299 : : !(vm_flags & vma->vm_flags))
1300 : : goto finish_or_fault;
1301 : :
1302 : : if (pages) {
1303 : : pages[i] = virt_to_page(start);
1304 : : if (pages[i])
1305 : : get_page(pages[i]);
1306 : : }
1307 : : if (vmas)
1308 : : vmas[i] = vma;
1309 : : start = (start + PAGE_SIZE) & PAGE_MASK;
1310 : : }
1311 : :
1312 : : return i;
1313 : :
1314 : : finish_or_fault:
1315 : : return i ? : -EFAULT;
1316 : : }
1317 : : #endif /* !CONFIG_MMU */
1318 : :
1319 : : #if defined(CONFIG_FS_DAX) || defined (CONFIG_CMA)
1320 : : static bool check_dax_vmas(struct vm_area_struct **vmas, long nr_pages)
1321 : : {
1322 : : long i;
1323 : : struct vm_area_struct *vma_prev = NULL;
1324 : :
1325 : : for (i = 0; i < nr_pages; i++) {
1326 : : struct vm_area_struct *vma = vmas[i];
1327 : :
1328 : : if (vma == vma_prev)
1329 : : continue;
1330 : :
1331 : : vma_prev = vma;
1332 : :
1333 : : if (vma_is_fsdax(vma))
1334 : : return true;
1335 : : }
1336 : : return false;
1337 : : }
1338 : :
1339 : : #ifdef CONFIG_CMA
1340 : : static struct page *new_non_cma_page(struct page *page, unsigned long private)
1341 : : {
1342 : : /*
1343 : : * We want to make sure we allocate the new page from the same node
1344 : : * as the source page.
1345 : : */
1346 : : int nid = page_to_nid(page);
1347 : : /*
1348 : : * Trying to allocate a page for migration. Ignore allocation
1349 : : * failure warnings. We don't force __GFP_THISNODE here because
1350 : : * this node here is the node where we have CMA reservation and
1351 : : * in some case these nodes will have really less non movable
1352 : : * allocation memory.
1353 : : */
1354 : : gfp_t gfp_mask = GFP_USER | __GFP_NOWARN;
1355 : :
1356 : : if (PageHighMem(page))
1357 : : gfp_mask |= __GFP_HIGHMEM;
1358 : :
1359 : : #ifdef CONFIG_HUGETLB_PAGE
1360 : : if (PageHuge(page)) {
1361 : : struct hstate *h = page_hstate(page);
1362 : : /*
1363 : : * We don't want to dequeue from the pool because pool pages will
1364 : : * mostly be from the CMA region.
1365 : : */
1366 : : return alloc_migrate_huge_page(h, gfp_mask, nid, NULL);
1367 : : }
1368 : : #endif
1369 : : if (PageTransHuge(page)) {
1370 : : struct page *thp;
1371 : : /*
1372 : : * ignore allocation failure warnings
1373 : : */
1374 : : gfp_t thp_gfpmask = GFP_TRANSHUGE | __GFP_NOWARN;
1375 : :
1376 : : /*
1377 : : * Remove the movable mask so that we don't allocate from
1378 : : * CMA area again.
1379 : : */
1380 : : thp_gfpmask &= ~__GFP_MOVABLE;
1381 : : thp = __alloc_pages_node(nid, thp_gfpmask, HPAGE_PMD_ORDER);
1382 : : if (!thp)
1383 : : return NULL;
1384 : : prep_transhuge_page(thp);
1385 : : return thp;
1386 : : }
1387 : :
1388 : : return __alloc_pages_node(nid, gfp_mask, 0);
1389 : : }
1390 : :
1391 : : static long check_and_migrate_cma_pages(struct task_struct *tsk,
1392 : : struct mm_struct *mm,
1393 : : unsigned long start,
1394 : : unsigned long nr_pages,
1395 : : struct page **pages,
1396 : : struct vm_area_struct **vmas,
1397 : : unsigned int gup_flags)
1398 : : {
1399 : : unsigned long i;
1400 : : unsigned long step;
1401 : : bool drain_allow = true;
1402 : : bool migrate_allow = true;
1403 : : LIST_HEAD(cma_page_list);
1404 : : long ret = nr_pages;
1405 : :
1406 : : check_again:
1407 : : for (i = 0; i < nr_pages;) {
1408 : :
1409 : : struct page *head = compound_head(pages[i]);
1410 : :
1411 : : /*
1412 : : * gup may start from a tail page. Advance step by the left
1413 : : * part.
1414 : : */
1415 : : step = compound_nr(head) - (pages[i] - head);
1416 : : /*
1417 : : * If we get a page from the CMA zone, since we are going to
1418 : : * be pinning these entries, we might as well move them out
1419 : : * of the CMA zone if possible.
1420 : : */
1421 : : if (is_migrate_cma_page(head)) {
1422 : : if (PageHuge(head))
1423 : : isolate_huge_page(head, &cma_page_list);
1424 : : else {
1425 : : if (!PageLRU(head) && drain_allow) {
1426 : : lru_add_drain_all();
1427 : : drain_allow = false;
1428 : : }
1429 : :
1430 : : if (!isolate_lru_page(head)) {
1431 : : list_add_tail(&head->lru, &cma_page_list);
1432 : : mod_node_page_state(page_pgdat(head),
1433 : : NR_ISOLATED_ANON +
1434 : : page_is_file_cache(head),
1435 : : hpage_nr_pages(head));
1436 : : }
1437 : : }
1438 : : }
1439 : :
1440 : : i += step;
1441 : : }
1442 : :
1443 : : if (!list_empty(&cma_page_list)) {
1444 : : /*
1445 : : * drop the above get_user_pages reference.
1446 : : */
1447 : : for (i = 0; i < nr_pages; i++)
1448 : : put_page(pages[i]);
1449 : :
1450 : : if (migrate_pages(&cma_page_list, new_non_cma_page,
1451 : : NULL, 0, MIGRATE_SYNC, MR_CONTIG_RANGE)) {
1452 : : /*
1453 : : * some of the pages failed migration. Do get_user_pages
1454 : : * without migration.
1455 : : */
1456 : : migrate_allow = false;
1457 : :
1458 : : if (!list_empty(&cma_page_list))
1459 : : putback_movable_pages(&cma_page_list);
1460 : : }
1461 : : /*
1462 : : * We did migrate all the pages, Try to get the page references
1463 : : * again migrating any new CMA pages which we failed to isolate
1464 : : * earlier.
1465 : : */
1466 : : ret = __get_user_pages_locked(tsk, mm, start, nr_pages,
1467 : : pages, vmas, NULL,
1468 : : gup_flags);
1469 : :
1470 : : if ((ret > 0) && migrate_allow) {
1471 : : nr_pages = ret;
1472 : : drain_allow = true;
1473 : : goto check_again;
1474 : : }
1475 : : }
1476 : :
1477 : : return ret;
1478 : : }
1479 : : #else
1480 : : static long check_and_migrate_cma_pages(struct task_struct *tsk,
1481 : : struct mm_struct *mm,
1482 : : unsigned long start,
1483 : : unsigned long nr_pages,
1484 : : struct page **pages,
1485 : : struct vm_area_struct **vmas,
1486 : : unsigned int gup_flags)
1487 : : {
1488 : : return nr_pages;
1489 : : }
1490 : : #endif /* CONFIG_CMA */
1491 : :
1492 : : /*
1493 : : * __gup_longterm_locked() is a wrapper for __get_user_pages_locked which
1494 : : * allows us to process the FOLL_LONGTERM flag.
1495 : : */
1496 : : static long __gup_longterm_locked(struct task_struct *tsk,
1497 : : struct mm_struct *mm,
1498 : : unsigned long start,
1499 : : unsigned long nr_pages,
1500 : : struct page **pages,
1501 : : struct vm_area_struct **vmas,
1502 : : unsigned int gup_flags)
1503 : : {
1504 : : struct vm_area_struct **vmas_tmp = vmas;
1505 : : unsigned long flags = 0;
1506 : : long rc, i;
1507 : :
1508 : : if (gup_flags & FOLL_LONGTERM) {
1509 : : if (!pages)
1510 : : return -EINVAL;
1511 : :
1512 : : if (!vmas_tmp) {
1513 : : vmas_tmp = kcalloc(nr_pages,
1514 : : sizeof(struct vm_area_struct *),
1515 : : GFP_KERNEL);
1516 : : if (!vmas_tmp)
1517 : : return -ENOMEM;
1518 : : }
1519 : : flags = memalloc_nocma_save();
1520 : : }
1521 : :
1522 : : rc = __get_user_pages_locked(tsk, mm, start, nr_pages, pages,
1523 : : vmas_tmp, NULL, gup_flags);
1524 : :
1525 : : if (gup_flags & FOLL_LONGTERM) {
1526 : : memalloc_nocma_restore(flags);
1527 : : if (rc < 0)
1528 : : goto out;
1529 : :
1530 : : if (check_dax_vmas(vmas_tmp, rc)) {
1531 : : for (i = 0; i < rc; i++)
1532 : : put_page(pages[i]);
1533 : : rc = -EOPNOTSUPP;
1534 : : goto out;
1535 : : }
1536 : :
1537 : : rc = check_and_migrate_cma_pages(tsk, mm, start, rc, pages,
1538 : : vmas_tmp, gup_flags);
1539 : : }
1540 : :
1541 : : out:
1542 : : if (vmas_tmp != vmas)
1543 : : kfree(vmas_tmp);
1544 : : return rc;
1545 : : }
1546 : : #else /* !CONFIG_FS_DAX && !CONFIG_CMA */
1547 : 0 : static __always_inline long __gup_longterm_locked(struct task_struct *tsk,
1548 : : struct mm_struct *mm,
1549 : : unsigned long start,
1550 : : unsigned long nr_pages,
1551 : : struct page **pages,
1552 : : struct vm_area_struct **vmas,
1553 : : unsigned int flags)
1554 : : {
1555 [ # # # # ]: 0 : return __get_user_pages_locked(tsk, mm, start, nr_pages, pages, vmas,
1556 : : NULL, flags);
1557 : : }
1558 : : #endif /* CONFIG_FS_DAX || CONFIG_CMA */
1559 : :
1560 : : /*
1561 : : * get_user_pages_remote() - pin user pages in memory
1562 : : * @tsk: the task_struct to use for page fault accounting, or
1563 : : * NULL if faults are not to be recorded.
1564 : : * @mm: mm_struct of target mm
1565 : : * @start: starting user address
1566 : : * @nr_pages: number of pages from start to pin
1567 : : * @gup_flags: flags modifying lookup behaviour
1568 : : * @pages: array that receives pointers to the pages pinned.
1569 : : * Should be at least nr_pages long. Or NULL, if caller
1570 : : * only intends to ensure the pages are faulted in.
1571 : : * @vmas: array of pointers to vmas corresponding to each page.
1572 : : * Or NULL if the caller does not require them.
1573 : : * @locked: pointer to lock flag indicating whether lock is held and
1574 : : * subsequently whether VM_FAULT_RETRY functionality can be
1575 : : * utilised. Lock must initially be held.
1576 : : *
1577 : : * Returns either number of pages pinned (which may be less than the
1578 : : * number requested), or an error. Details about the return value:
1579 : : *
1580 : : * -- If nr_pages is 0, returns 0.
1581 : : * -- If nr_pages is >0, but no pages were pinned, returns -errno.
1582 : : * -- If nr_pages is >0, and some pages were pinned, returns the number of
1583 : : * pages pinned. Again, this may be less than nr_pages.
1584 : : *
1585 : : * The caller is responsible for releasing returned @pages, via put_page().
1586 : : *
1587 : : * @vmas are valid only as long as mmap_sem is held.
1588 : : *
1589 : : * Must be called with mmap_sem held for read or write.
1590 : : *
1591 : : * get_user_pages walks a process's page tables and takes a reference to
1592 : : * each struct page that each user address corresponds to at a given
1593 : : * instant. That is, it takes the page that would be accessed if a user
1594 : : * thread accesses the given user virtual address at that instant.
1595 : : *
1596 : : * This does not guarantee that the page exists in the user mappings when
1597 : : * get_user_pages returns, and there may even be a completely different
1598 : : * page there in some cases (eg. if mmapped pagecache has been invalidated
1599 : : * and subsequently re faulted). However it does guarantee that the page
1600 : : * won't be freed completely. And mostly callers simply care that the page
1601 : : * contains data that was valid *at some point in time*. Typically, an IO
1602 : : * or similar operation cannot guarantee anything stronger anyway because
1603 : : * locks can't be held over the syscall boundary.
1604 : : *
1605 : : * If gup_flags & FOLL_WRITE == 0, the page must not be written to. If the page
1606 : : * is written to, set_page_dirty (or set_page_dirty_lock, as appropriate) must
1607 : : * be called after the page is finished with, and before put_page is called.
1608 : : *
1609 : : * get_user_pages is typically used for fewer-copy IO operations, to get a
1610 : : * handle on the memory by some means other than accesses via the user virtual
1611 : : * addresses. The pages may be submitted for DMA to devices or accessed via
1612 : : * their kernel linear mapping (via the kmap APIs). Care should be taken to
1613 : : * use the correct cache flushing APIs.
1614 : : *
1615 : : * See also get_user_pages_fast, for performance critical applications.
1616 : : *
1617 : : * get_user_pages should be phased out in favor of
1618 : : * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing
1619 : : * should use get_user_pages because it cannot pass
1620 : : * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault.
1621 : : */
1622 : : #ifdef CONFIG_MMU
1623 : 198594 : long get_user_pages_remote(struct task_struct *tsk, struct mm_struct *mm,
1624 : : unsigned long start, unsigned long nr_pages,
1625 : : unsigned int gup_flags, struct page **pages,
1626 : : struct vm_area_struct **vmas, int *locked)
1627 : : {
1628 : : /*
1629 : : * FOLL_PIN must only be set internally by the pin_user_pages*() APIs,
1630 : : * never directly by the caller, so enforce that with an assertion:
1631 : : */
1632 [ - + + - ]: 198594 : if (WARN_ON_ONCE(gup_flags & FOLL_PIN))
1633 : : return -EINVAL;
1634 : :
1635 : : /*
1636 : : * Parts of FOLL_LONGTERM behavior are incompatible with
1637 : : * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on
1638 : : * vmas. However, this only comes up if locked is set, and there are
1639 : : * callers that do request FOLL_LONGTERM, but do not set locked. So,
1640 : : * allow what we can.
1641 : : */
1642 [ - + ]: 198594 : if (gup_flags & FOLL_LONGTERM) {
1643 [ # # # # ]: 0 : if (WARN_ON_ONCE(locked))
1644 : : return -EINVAL;
1645 : : /*
1646 : : * This will check the vmas (even if our vmas arg is NULL)
1647 : : * and return -ENOTSUPP if DAX isn't allowed in this case:
1648 : : */
1649 [ # # ]: 0 : return __gup_longterm_locked(tsk, mm, start, nr_pages, pages,
1650 : : vmas, gup_flags | FOLL_TOUCH |
1651 : : FOLL_REMOTE);
1652 : : }
1653 : :
1654 [ - + ]: 198594 : return __get_user_pages_locked(tsk, mm, start, nr_pages, pages, vmas,
1655 : : locked,
1656 : : gup_flags | FOLL_TOUCH | FOLL_REMOTE);
1657 : : }
1658 : : EXPORT_SYMBOL(get_user_pages_remote);
1659 : :
1660 : : #else /* CONFIG_MMU */
1661 : : long get_user_pages_remote(struct task_struct *tsk, struct mm_struct *mm,
1662 : : unsigned long start, unsigned long nr_pages,
1663 : : unsigned int gup_flags, struct page **pages,
1664 : : struct vm_area_struct **vmas, int *locked)
1665 : : {
1666 : : return 0;
1667 : : }
1668 : : #endif /* !CONFIG_MMU */
1669 : :
1670 : : /*
1671 : : * This is the same as get_user_pages_remote(), just with a
1672 : : * less-flexible calling convention where we assume that the task
1673 : : * and mm being operated on are the current task's and don't allow
1674 : : * passing of a locked parameter. We also obviously don't pass
1675 : : * FOLL_REMOTE in here.
1676 : : */
1677 : 0 : long get_user_pages(unsigned long start, unsigned long nr_pages,
1678 : : unsigned int gup_flags, struct page **pages,
1679 : : struct vm_area_struct **vmas)
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 [ # # # # ]: 0 : if (WARN_ON_ONCE(gup_flags & FOLL_PIN))
1686 : : return -EINVAL;
1687 : :
1688 [ # # ]: 0 : return __gup_longterm_locked(current, current->mm, start, nr_pages,
1689 : : pages, vmas, gup_flags | FOLL_TOUCH);
1690 : : }
1691 : : EXPORT_SYMBOL(get_user_pages);
1692 : :
1693 : : /*
1694 : : * We can leverage the VM_FAULT_RETRY functionality in the page fault
1695 : : * paths better by using either get_user_pages_locked() or
1696 : : * get_user_pages_unlocked().
1697 : : *
1698 : : * get_user_pages_locked() is suitable to replace the form:
1699 : : *
1700 : : * down_read(&mm->mmap_sem);
1701 : : * do_something()
1702 : : * get_user_pages(tsk, mm, ..., pages, NULL);
1703 : : * up_read(&mm->mmap_sem);
1704 : : *
1705 : : * to:
1706 : : *
1707 : : * int locked = 1;
1708 : : * down_read(&mm->mmap_sem);
1709 : : * do_something()
1710 : : * get_user_pages_locked(tsk, mm, ..., pages, &locked);
1711 : : * if (locked)
1712 : : * up_read(&mm->mmap_sem);
1713 : : */
1714 : 0 : long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
1715 : : unsigned int gup_flags, struct page **pages,
1716 : : int *locked)
1717 : : {
1718 : : /*
1719 : : * FIXME: Current FOLL_LONGTERM behavior is incompatible with
1720 : : * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on
1721 : : * vmas. As there are no users of this flag in this call we simply
1722 : : * disallow this option for now.
1723 : : */
1724 [ # # # # ]: 0 : if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM))
1725 : : return -EINVAL;
1726 : :
1727 [ # # ]: 0 : return __get_user_pages_locked(current, current->mm, start, nr_pages,
1728 : : pages, NULL, locked,
1729 : : gup_flags | FOLL_TOUCH);
1730 : : }
1731 : : EXPORT_SYMBOL(get_user_pages_locked);
1732 : :
1733 : : /*
1734 : : * get_user_pages_unlocked() is suitable to replace the form:
1735 : : *
1736 : : * down_read(&mm->mmap_sem);
1737 : : * get_user_pages(tsk, mm, ..., pages, NULL);
1738 : : * up_read(&mm->mmap_sem);
1739 : : *
1740 : : * with:
1741 : : *
1742 : : * get_user_pages_unlocked(tsk, mm, ..., pages);
1743 : : *
1744 : : * It is functionally equivalent to get_user_pages_fast so
1745 : : * get_user_pages_fast should be used instead if specific gup_flags
1746 : : * (e.g. FOLL_FORCE) are not required.
1747 : : */
1748 : 0 : long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
1749 : : struct page **pages, unsigned int gup_flags)
1750 : : {
1751 [ # # ]: 0 : struct mm_struct *mm = current->mm;
1752 : 0 : int locked = 1;
1753 : 0 : long ret;
1754 : :
1755 : : /*
1756 : : * FIXME: Current FOLL_LONGTERM behavior is incompatible with
1757 : : * FAULT_FLAG_ALLOW_RETRY because of the FS DAX check requirement on
1758 : : * vmas. As there are no users of this flag in this call we simply
1759 : : * disallow this option for now.
1760 : : */
1761 [ # # # # ]: 0 : if (WARN_ON_ONCE(gup_flags & FOLL_LONGTERM))
1762 : : return -EINVAL;
1763 : :
1764 : 0 : down_read(&mm->mmap_sem);
1765 [ # # ]: 0 : ret = __get_user_pages_locked(current, mm, start, nr_pages, pages, NULL,
1766 : : &locked, gup_flags | FOLL_TOUCH);
1767 [ # # ]: 0 : if (locked)
1768 : 0 : up_read(&mm->mmap_sem);
1769 : : return ret;
1770 : : }
1771 : : EXPORT_SYMBOL(get_user_pages_unlocked);
1772 : :
1773 : : /*
1774 : : * Fast GUP
1775 : : *
1776 : : * get_user_pages_fast attempts to pin user pages by walking the page
1777 : : * tables directly and avoids taking locks. Thus the walker needs to be
1778 : : * protected from page table pages being freed from under it, and should
1779 : : * block any THP splits.
1780 : : *
1781 : : * One way to achieve this is to have the walker disable interrupts, and
1782 : : * rely on IPIs from the TLB flushing code blocking before the page table
1783 : : * pages are freed. This is unsuitable for architectures that do not need
1784 : : * to broadcast an IPI when invalidating TLBs.
1785 : : *
1786 : : * Another way to achieve this is to batch up page table containing pages
1787 : : * belonging to more than one mm_user, then rcu_sched a callback to free those
1788 : : * pages. Disabling interrupts will allow the fast_gup walker to both block
1789 : : * the rcu_sched callback, and an IPI that we broadcast for splitting THPs
1790 : : * (which is a relatively rare event). The code below adopts this strategy.
1791 : : *
1792 : : * Before activating this code, please be aware that the following assumptions
1793 : : * are currently made:
1794 : : *
1795 : : * *) Either MMU_GATHER_RCU_TABLE_FREE is enabled, and tlb_remove_table() is used to
1796 : : * free pages containing page tables or TLB flushing requires IPI broadcast.
1797 : : *
1798 : : * *) ptes can be read atomically by the architecture.
1799 : : *
1800 : : * *) access_ok is sufficient to validate userspace address ranges.
1801 : : *
1802 : : * The last two assumptions can be relaxed by the addition of helper functions.
1803 : : *
1804 : : * This code is based heavily on the PowerPC implementation by Nick Piggin.
1805 : : */
1806 : : #ifdef CONFIG_HAVE_FAST_GUP
1807 : : #ifdef CONFIG_GUP_GET_PTE_LOW_HIGH
1808 : : /*
1809 : : * WARNING: only to be used in the get_user_pages_fast() implementation.
1810 : : *
1811 : : * With get_user_pages_fast(), we walk down the pagetables without taking any
1812 : : * locks. For this we would like to load the pointers atomically, but sometimes
1813 : : * that is not possible (e.g. without expensive cmpxchg8b on x86_32 PAE). What
1814 : : * we do have is the guarantee that a PTE will only either go from not present
1815 : : * to present, or present to not present or both -- it will not switch to a
1816 : : * completely different present page without a TLB flush in between; something
1817 : : * that we are blocking by holding interrupts off.
1818 : : *
1819 : : * Setting ptes from not present to present goes:
1820 : : *
1821 : : * ptep->pte_high = h;
1822 : : * smp_wmb();
1823 : : * ptep->pte_low = l;
1824 : : *
1825 : : * And present to not present goes:
1826 : : *
1827 : : * ptep->pte_low = 0;
1828 : : * smp_wmb();
1829 : : * ptep->pte_high = 0;
1830 : : *
1831 : : * We must ensure here that the load of pte_low sees 'l' IFF pte_high sees 'h'.
1832 : : * We load pte_high *after* loading pte_low, which ensures we don't see an older
1833 : : * value of pte_high. *Then* we recheck pte_low, which ensures that we haven't
1834 : : * picked up a changed pte high. We might have gotten rubbish values from
1835 : : * pte_low and pte_high, but we are guaranteed that pte_low will not have the
1836 : : * present bit set *unless* it is 'l'. Because get_user_pages_fast() only
1837 : : * operates on present ptes we're safe.
1838 : : */
1839 : : static inline pte_t gup_get_pte(pte_t *ptep)
1840 : : {
1841 : : pte_t pte;
1842 : :
1843 : : do {
1844 : : pte.pte_low = ptep->pte_low;
1845 : : smp_rmb();
1846 : : pte.pte_high = ptep->pte_high;
1847 : : smp_rmb();
1848 : : } while (unlikely(pte.pte_low != ptep->pte_low));
1849 : :
1850 : : return pte;
1851 : : }
1852 : : #else /* CONFIG_GUP_GET_PTE_LOW_HIGH */
1853 : : /*
1854 : : * We require that the PTE can be read atomically.
1855 : : */
1856 : 9 : static inline pte_t gup_get_pte(pte_t *ptep)
1857 : : {
1858 : 9 : return READ_ONCE(*ptep);
1859 : : }
1860 : : #endif /* CONFIG_GUP_GET_PTE_LOW_HIGH */
1861 : :
1862 : 0 : static void __maybe_unused undo_dev_pagemap(int *nr, int nr_start,
1863 : : struct page **pages)
1864 : : {
1865 [ # # ]: 0 : while ((*nr) - nr_start) {
1866 : 0 : struct page *page = pages[--(*nr)];
1867 : :
1868 [ # # ]: 0 : ClearPageReferenced(page);
1869 : 0 : put_page(page);
1870 : : }
1871 : 0 : }
1872 : :
1873 : : #ifdef CONFIG_ARCH_HAS_PTE_SPECIAL
1874 : 9 : static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1875 : : unsigned int flags, struct page **pages, int *nr)
1876 : : {
1877 : 9 : struct dev_pagemap *pgmap = NULL;
1878 : 9 : int nr_start = *nr, ret = 0;
1879 : 9 : pte_t *ptep, *ptem;
1880 : :
1881 [ + - ]: 9 : ptem = ptep = pte_offset_map(&pmd, addr);
1882 : 9 : do {
1883 [ + - ]: 9 : pte_t pte = gup_get_pte(ptep);
1884 : 9 : struct page *head, *page;
1885 : :
1886 : : /*
1887 : : * Similar to the PMD case below, NUMA hinting must take slow
1888 : : * path using the pte_protnone check.
1889 : : */
1890 [ + - ]: 9 : if (pte_protnone(pte))
1891 : : goto pte_unmap;
1892 : :
1893 [ + - - + ]: 18 : if (!pte_access_permitted(pte, flags & FOLL_WRITE))
1894 : 0 : goto pte_unmap;
1895 : :
1896 [ - + ]: 9 : if (pte_devmap(pte)) {
1897 [ # # ]: 0 : if (unlikely(flags & FOLL_LONGTERM))
1898 : 0 : goto pte_unmap;
1899 : :
1900 : 0 : pgmap = get_dev_pagemap(pte_pfn(pte), pgmap);
1901 : 0 : if (unlikely(!pgmap)) {
1902 : 0 : undo_dev_pagemap(nr, nr_start, pages);
1903 : 0 : goto pte_unmap;
1904 : : }
1905 [ - + ]: 9 : } else if (pte_special(pte))
1906 : 0 : goto pte_unmap;
1907 : :
1908 : 9 : VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
1909 [ + - ]: 9 : page = pte_page(pte);
1910 : :
1911 : 9 : head = try_get_compound_head(page, 1);
1912 [ - + ]: 9 : if (!head)
1913 : 0 : goto pte_unmap;
1914 : :
1915 [ - + ]: 9 : if (unlikely(pte_val(pte) != pte_val(*ptep))) {
1916 : 0 : put_page(head);
1917 : 0 : goto pte_unmap;
1918 : : }
1919 : :
1920 : 9 : VM_BUG_ON_PAGE(compound_head(page) != head, page);
1921 : :
1922 [ - + ]: 9 : SetPageReferenced(page);
1923 : 9 : pages[*nr] = page;
1924 : 9 : (*nr)++;
1925 : :
1926 [ - + ]: 9 : } while (ptep++, addr += PAGE_SIZE, addr != end);
1927 : :
1928 : : ret = 1;
1929 : :
1930 : 9 : pte_unmap:
1931 : 9 : if (pgmap)
1932 : : put_dev_pagemap(pgmap);
1933 : 9 : pte_unmap(ptem);
1934 : 9 : return ret;
1935 : : }
1936 : : #else
1937 : :
1938 : : /*
1939 : : * If we can't determine whether or not a pte is special, then fail immediately
1940 : : * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not
1941 : : * to be special.
1942 : : *
1943 : : * For a futex to be placed on a THP tail page, get_futex_key requires a
1944 : : * __get_user_pages_fast implementation that can pin pages. Thus it's still
1945 : : * useful to have gup_huge_pmd even if we can't operate on ptes.
1946 : : */
1947 : : static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1948 : : unsigned int flags, struct page **pages, int *nr)
1949 : : {
1950 : : return 0;
1951 : : }
1952 : : #endif /* CONFIG_ARCH_HAS_PTE_SPECIAL */
1953 : :
1954 : : #if defined(CONFIG_ARCH_HAS_PTE_DEVMAP) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1955 : : static int __gup_device_huge(unsigned long pfn, unsigned long addr,
1956 : : unsigned long end, struct page **pages, int *nr)
1957 : : {
1958 : : int nr_start = *nr;
1959 : : struct dev_pagemap *pgmap = NULL;
1960 : :
1961 : : do {
1962 : : struct page *page = pfn_to_page(pfn);
1963 : :
1964 : : pgmap = get_dev_pagemap(pfn, pgmap);
1965 : : if (unlikely(!pgmap)) {
1966 : : undo_dev_pagemap(nr, nr_start, pages);
1967 : : return 0;
1968 : : }
1969 : : SetPageReferenced(page);
1970 : : pages[*nr] = page;
1971 : : get_page(page);
1972 : : (*nr)++;
1973 : : pfn++;
1974 : : } while (addr += PAGE_SIZE, addr != end);
1975 : :
1976 : : if (pgmap)
1977 : : put_dev_pagemap(pgmap);
1978 : : return 1;
1979 : : }
1980 : :
1981 : : static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
1982 : : unsigned long end, struct page **pages, int *nr)
1983 : : {
1984 : : unsigned long fault_pfn;
1985 : : int nr_start = *nr;
1986 : :
1987 : : fault_pfn = pmd_pfn(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1988 : : if (!__gup_device_huge(fault_pfn, addr, end, pages, nr))
1989 : : return 0;
1990 : :
1991 : : if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
1992 : : undo_dev_pagemap(nr, nr_start, pages);
1993 : : return 0;
1994 : : }
1995 : : return 1;
1996 : : }
1997 : :
1998 : : static int __gup_device_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
1999 : : unsigned long end, struct page **pages, int *nr)
2000 : : {
2001 : : unsigned long fault_pfn;
2002 : : int nr_start = *nr;
2003 : :
2004 : : fault_pfn = pud_pfn(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
2005 : : if (!__gup_device_huge(fault_pfn, addr, end, pages, nr))
2006 : : return 0;
2007 : :
2008 : : if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2009 : : undo_dev_pagemap(nr, nr_start, pages);
2010 : : return 0;
2011 : : }
2012 : : return 1;
2013 : : }
2014 : : #else
2015 : : static int __gup_device_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2016 : : unsigned long end, struct page **pages, int *nr)
2017 : : {
2018 : : BUILD_BUG();
2019 : : return 0;
2020 : : }
2021 : :
2022 : : static int __gup_device_huge_pud(pud_t pud, pud_t *pudp, unsigned long addr,
2023 : : unsigned long end, struct page **pages, int *nr)
2024 : : {
2025 : : BUILD_BUG();
2026 : : return 0;
2027 : : }
2028 : : #endif
2029 : :
2030 : 0 : static int record_subpages(struct page *page, unsigned long addr,
2031 : : unsigned long end, struct page **pages)
2032 : : {
2033 : : int nr;
2034 : :
2035 [ # # # # ]: 0 : for (nr = 0; addr != end; addr += PAGE_SIZE)
2036 : 0 : pages[nr++] = page++;
2037 : :
2038 : 0 : return nr;
2039 : : }
2040 : :
2041 : 0 : static void put_compound_head(struct page *page, int refs)
2042 : : {
2043 : 0 : VM_BUG_ON_PAGE(page_ref_count(page) < refs, page);
2044 : : /*
2045 : : * Calling put_page() for each ref is unnecessarily slow. Only the last
2046 : : * ref needs a put_page().
2047 : : */
2048 [ # # ]: 0 : if (refs > 1)
2049 : 0 : page_ref_sub(page, refs - 1);
2050 : 0 : put_page(page);
2051 : 0 : }
2052 : :
2053 : : #ifdef CONFIG_ARCH_HAS_HUGEPD
2054 : : static unsigned long hugepte_addr_end(unsigned long addr, unsigned long end,
2055 : : unsigned long sz)
2056 : : {
2057 : : unsigned long __boundary = (addr + sz) & ~(sz-1);
2058 : : return (__boundary - 1 < end - 1) ? __boundary : end;
2059 : : }
2060 : :
2061 : : static int gup_hugepte(pte_t *ptep, unsigned long sz, unsigned long addr,
2062 : : unsigned long end, unsigned int flags,
2063 : : struct page **pages, int *nr)
2064 : : {
2065 : : unsigned long pte_end;
2066 : : struct page *head, *page;
2067 : : pte_t pte;
2068 : : int refs;
2069 : :
2070 : : pte_end = (addr + sz) & ~(sz-1);
2071 : : if (pte_end < end)
2072 : : end = pte_end;
2073 : :
2074 : : pte = READ_ONCE(*ptep);
2075 : :
2076 : : if (!pte_access_permitted(pte, flags & FOLL_WRITE))
2077 : : return 0;
2078 : :
2079 : : /* hugepages are never "special" */
2080 : : VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
2081 : :
2082 : : head = pte_page(pte);
2083 : : page = head + ((addr & (sz-1)) >> PAGE_SHIFT);
2084 : : refs = record_subpages(page, addr, end, pages + *nr);
2085 : :
2086 : : head = try_get_compound_head(head, refs);
2087 : : if (!head)
2088 : : return 0;
2089 : :
2090 : : if (unlikely(pte_val(pte) != pte_val(*ptep))) {
2091 : : put_compound_head(head, refs);
2092 : : return 0;
2093 : : }
2094 : :
2095 : : *nr += refs;
2096 : : SetPageReferenced(head);
2097 : : return 1;
2098 : : }
2099 : :
2100 : : static int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2101 : : unsigned int pdshift, unsigned long end, unsigned int flags,
2102 : : struct page **pages, int *nr)
2103 : : {
2104 : : pte_t *ptep;
2105 : : unsigned long sz = 1UL << hugepd_shift(hugepd);
2106 : : unsigned long next;
2107 : :
2108 : : ptep = hugepte_offset(hugepd, addr, pdshift);
2109 : : do {
2110 : : next = hugepte_addr_end(addr, end, sz);
2111 : : if (!gup_hugepte(ptep, sz, addr, end, flags, pages, nr))
2112 : : return 0;
2113 : : } while (ptep++, addr = next, addr != end);
2114 : :
2115 : : return 1;
2116 : : }
2117 : : #else
2118 : : static inline int gup_huge_pd(hugepd_t hugepd, unsigned long addr,
2119 : : unsigned int pdshift, unsigned long end, unsigned int flags,
2120 : : struct page **pages, int *nr)
2121 : : {
2122 : : return 0;
2123 : : }
2124 : : #endif /* CONFIG_ARCH_HAS_HUGEPD */
2125 : :
2126 : 0 : static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
2127 : : unsigned long end, unsigned int flags,
2128 : : struct page **pages, int *nr)
2129 : : {
2130 : 0 : struct page *head, *page;
2131 : 0 : int refs;
2132 : :
2133 [ # # # # ]: 0 : if (!pmd_access_permitted(orig, flags & FOLL_WRITE))
2134 : 0 : return 0;
2135 : :
2136 [ # # ]: 0 : if (pmd_devmap(orig)) {
2137 : : if (unlikely(flags & FOLL_LONGTERM))
2138 : : return 0;
2139 : : return __gup_device_huge_pmd(orig, pmdp, addr, end, pages, nr);
2140 : : }
2141 : :
2142 [ # # ]: 0 : page = pmd_page(orig) + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
2143 : 0 : refs = record_subpages(page, addr, end, pages + *nr);
2144 : :
2145 [ # # ]: 0 : head = try_get_compound_head(pmd_page(orig), refs);
2146 [ # # ]: 0 : if (!head)
2147 : : return 0;
2148 : :
2149 [ # # ]: 0 : if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
2150 : 0 : put_compound_head(head, refs);
2151 : 0 : return 0;
2152 : : }
2153 : :
2154 : 0 : *nr += refs;
2155 [ # # ]: 0 : SetPageReferenced(head);
2156 : 0 : return 1;
2157 : : }
2158 : :
2159 : 0 : static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
2160 : : unsigned long end, unsigned int flags, struct page **pages, int *nr)
2161 : : {
2162 : 0 : struct page *head, *page;
2163 : 0 : int refs;
2164 : :
2165 [ # # # # ]: 0 : if (!pud_access_permitted(orig, flags & FOLL_WRITE))
2166 : 0 : return 0;
2167 : :
2168 [ # # ]: 0 : if (pud_devmap(orig)) {
2169 : : if (unlikely(flags & FOLL_LONGTERM))
2170 : : return 0;
2171 : : return __gup_device_huge_pud(orig, pudp, addr, end, pages, nr);
2172 : : }
2173 : :
2174 [ # # ]: 0 : page = pud_page(orig) + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
2175 : 0 : refs = record_subpages(page, addr, end, pages + *nr);
2176 : :
2177 [ # # ]: 0 : head = try_get_compound_head(pud_page(orig), refs);
2178 [ # # ]: 0 : if (!head)
2179 : : return 0;
2180 : :
2181 [ # # ]: 0 : if (unlikely(pud_val(orig) != pud_val(*pudp))) {
2182 : 0 : put_compound_head(head, refs);
2183 : 0 : return 0;
2184 : : }
2185 : :
2186 : 0 : *nr += refs;
2187 [ # # ]: 0 : SetPageReferenced(head);
2188 : 0 : return 1;
2189 : : }
2190 : :
2191 : : static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
2192 : : unsigned long end, unsigned int flags,
2193 : : struct page **pages, int *nr)
2194 : : {
2195 : : int refs;
2196 : : struct page *head, *page;
2197 : :
2198 : : if (!pgd_access_permitted(orig, flags & FOLL_WRITE))
2199 : : return 0;
2200 : :
2201 : : BUILD_BUG_ON(pgd_devmap(orig));
2202 : :
2203 : : page = pgd_page(orig) + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT);
2204 : : refs = record_subpages(page, addr, end, pages + *nr);
2205 : :
2206 : : head = try_get_compound_head(pgd_page(orig), refs);
2207 : : if (!head)
2208 : : return 0;
2209 : :
2210 : : if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) {
2211 : : put_compound_head(head, refs);
2212 : : return 0;
2213 : : }
2214 : :
2215 : : *nr += refs;
2216 : : SetPageReferenced(head);
2217 : : return 1;
2218 : : }
2219 : :
2220 : 9 : static int gup_pmd_range(pud_t pud, unsigned long addr, unsigned long end,
2221 : : unsigned int flags, struct page **pages, int *nr)
2222 : : {
2223 : 9 : unsigned long next;
2224 : 9 : pmd_t *pmdp;
2225 : :
2226 [ + - ]: 9 : pmdp = pmd_offset(&pud, addr);
2227 : 9 : do {
2228 [ - + ]: 9 : pmd_t pmd = READ_ONCE(*pmdp);
2229 : :
2230 [ - + ]: 9 : next = pmd_addr_end(addr, end);
2231 [ + - + - ]: 18 : if (!pmd_present(pmd))
2232 : : return 0;
2233 : :
2234 [ - + ]: 9 : if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
2235 : : pmd_devmap(pmd))) {
2236 : : /*
2237 : : * NUMA hinting faults need to be handled in the GUP
2238 : : * slowpath for accounting purposes and so that they
2239 : : * can be serialised against THP migration.
2240 : : */
2241 : 0 : if (pmd_protnone(pmd))
2242 : : return 0;
2243 : :
2244 [ # # ]: 0 : if (!gup_huge_pmd(pmd, pmdp, addr, next, flags,
2245 : : pages, nr))
2246 : : return 0;
2247 : :
2248 : 9 : } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) {
2249 : : /*
2250 : : * architecture have different format for hugetlbfs
2251 : : * pmd format and THP pmd format
2252 : : */
2253 : : if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr,
2254 : : PMD_SHIFT, next, flags, pages, nr))
2255 : : return 0;
2256 [ + - ]: 9 : } else if (!gup_pte_range(pmd, addr, next, flags, pages, nr))
2257 : : return 0;
2258 [ - + ]: 9 : } while (pmdp++, addr = next, addr != end);
2259 : :
2260 : : return 1;
2261 : : }
2262 : :
2263 : 9 : static int gup_pud_range(p4d_t p4d, unsigned long addr, unsigned long end,
2264 : : unsigned int flags, struct page **pages, int *nr)
2265 : : {
2266 : 9 : unsigned long next;
2267 : 9 : pud_t *pudp;
2268 : :
2269 : 9 : pudp = pud_offset(&p4d, addr);
2270 : 9 : do {
2271 [ - + ]: 9 : pud_t pud = READ_ONCE(*pudp);
2272 : :
2273 [ - + ]: 9 : next = pud_addr_end(addr, end);
2274 [ + - + - ]: 18 : if (unlikely(!pud_present(pud)))
2275 : : return 0;
2276 [ - + ]: 9 : if (unlikely(pud_huge(pud))) {
2277 [ # # ]: 0 : if (!gup_huge_pud(pud, pudp, addr, next, flags,
2278 : : pages, nr))
2279 : : return 0;
2280 : 9 : } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) {
2281 : : if (!gup_huge_pd(__hugepd(pud_val(pud)), addr,
2282 : : PUD_SHIFT, next, flags, pages, nr))
2283 : : return 0;
2284 [ + - ]: 9 : } else if (!gup_pmd_range(pud, addr, next, flags, pages, nr))
2285 : : return 0;
2286 [ - + ]: 9 : } while (pudp++, addr = next, addr != end);
2287 : :
2288 : : return 1;
2289 : : }
2290 : :
2291 : 9 : static int gup_p4d_range(pgd_t pgd, unsigned long addr, unsigned long end,
2292 : : unsigned int flags, struct page **pages, int *nr)
2293 : : {
2294 : 9 : unsigned long next;
2295 : 9 : p4d_t *p4dp;
2296 : :
2297 : 9 : p4dp = p4d_offset(&pgd, addr);
2298 : 9 : do {
2299 [ - + ]: 9 : p4d_t p4d = READ_ONCE(*p4dp);
2300 : :
2301 [ - + ]: 9 : next = p4d_addr_end(addr, end);
2302 [ + - ]: 9 : if (p4d_none(p4d))
2303 : : return 0;
2304 : 9 : BUILD_BUG_ON(p4d_huge(p4d));
2305 : 9 : if (unlikely(is_hugepd(__hugepd(p4d_val(p4d))))) {
2306 : : if (!gup_huge_pd(__hugepd(p4d_val(p4d)), addr,
2307 : : P4D_SHIFT, next, flags, pages, nr))
2308 : : return 0;
2309 [ + - ]: 9 : } else if (!gup_pud_range(p4d, addr, next, flags, pages, nr))
2310 : : return 0;
2311 [ - + ]: 9 : } while (p4dp++, addr = next, addr != end);
2312 : :
2313 : : return 1;
2314 : : }
2315 : :
2316 : 9 : static void gup_pgd_range(unsigned long addr, unsigned long end,
2317 : : unsigned int flags, struct page **pages, int *nr)
2318 : : {
2319 : 9 : unsigned long next;
2320 : 9 : pgd_t *pgdp;
2321 : :
2322 : 9 : pgdp = pgd_offset(current->mm, addr);
2323 : 9 : do {
2324 [ + - ]: 9 : pgd_t pgd = READ_ONCE(*pgdp);
2325 : :
2326 [ + - ]: 9 : next = pgd_addr_end(addr, end);
2327 [ + - ]: 9 : if (pgd_none(pgd))
2328 : 0 : return;
2329 : 9 : if (unlikely(pgd_huge(pgd))) {
2330 : : if (!gup_huge_pgd(pgd, pgdp, addr, next, flags,
2331 : : pages, nr))
2332 : : return;
2333 : 9 : } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) {
2334 : : if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr,
2335 : : PGDIR_SHIFT, next, flags, pages, nr))
2336 : : return;
2337 [ + - ]: 9 : } else if (!gup_p4d_range(pgd, addr, next, flags, pages, nr))
2338 : : return;
2339 [ - + ]: 9 : } while (pgdp++, addr = next, addr != end);
2340 : : }
2341 : : #else
2342 : : static inline void gup_pgd_range(unsigned long addr, unsigned long end,
2343 : : unsigned int flags, struct page **pages, int *nr)
2344 : : {
2345 : : }
2346 : : #endif /* CONFIG_HAVE_FAST_GUP */
2347 : :
2348 : : #ifndef gup_fast_permitted
2349 : : /*
2350 : : * Check if it's allowed to use __get_user_pages_fast() for the range, or
2351 : : * we need to fall back to the slow version:
2352 : : */
2353 : : static bool gup_fast_permitted(unsigned long start, unsigned long end)
2354 : : {
2355 : : return true;
2356 : : }
2357 : : #endif
2358 : :
2359 : : /*
2360 : : * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to
2361 : : * the regular GUP.
2362 : : * Note a difference with get_user_pages_fast: this always returns the
2363 : : * number of pages pinned, 0 if no pages were pinned.
2364 : : *
2365 : : * If the architecture does not support this function, simply return with no
2366 : : * pages pinned.
2367 : : */
2368 : 0 : int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
2369 : : struct page **pages)
2370 : : {
2371 : 0 : unsigned long len, end;
2372 : 0 : unsigned long flags;
2373 : 0 : int nr = 0;
2374 : :
2375 : 0 : start = untagged_addr(start) & PAGE_MASK;
2376 : 0 : len = (unsigned long) nr_pages << PAGE_SHIFT;
2377 : 0 : end = start + len;
2378 : :
2379 [ # # ]: 0 : if (end <= start)
2380 : : return 0;
2381 [ # # # # ]: 0 : if (unlikely(!access_ok((void __user *)start, len)))
2382 : : return 0;
2383 : :
2384 : : /*
2385 : : * Disable interrupts. We use the nested form as we can already have
2386 : : * interrupts disabled by get_futex_key.
2387 : : *
2388 : : * With interrupts disabled, we block page table pages from being
2389 : : * freed from under us. See struct mmu_table_batch comments in
2390 : : * include/asm-generic/tlb.h for more details.
2391 : : *
2392 : : * We do not adopt an rcu_read_lock(.) here as we also want to
2393 : : * block IPIs that come from THPs splitting.
2394 : : */
2395 : :
2396 [ # # ]: 0 : if (IS_ENABLED(CONFIG_HAVE_FAST_GUP) &&
2397 : 0 : gup_fast_permitted(start, end)) {
2398 : 0 : local_irq_save(flags);
2399 : 0 : gup_pgd_range(start, end, write ? FOLL_WRITE : 0, pages, &nr);
2400 : 0 : local_irq_restore(flags);
2401 : : }
2402 : :
2403 : 0 : return nr;
2404 : : }
2405 : : EXPORT_SYMBOL_GPL(__get_user_pages_fast);
2406 : :
2407 : 0 : static int __gup_longterm_unlocked(unsigned long start, int nr_pages,
2408 : : unsigned int gup_flags, struct page **pages)
2409 : : {
2410 : 0 : int ret;
2411 : :
2412 : : /*
2413 : : * FIXME: FOLL_LONGTERM does not work with
2414 : : * get_user_pages_unlocked() (see comments in that function)
2415 : : */
2416 [ # # ]: 0 : if (gup_flags & FOLL_LONGTERM) {
2417 : 0 : down_read(¤t->mm->mmap_sem);
2418 [ # # ]: 0 : ret = __gup_longterm_locked(current, current->mm,
2419 : : start, nr_pages,
2420 : : pages, NULL, gup_flags);
2421 : 0 : up_read(¤t->mm->mmap_sem);
2422 : : } else {
2423 : 0 : ret = get_user_pages_unlocked(start, nr_pages,
2424 : : pages, gup_flags);
2425 : : }
2426 : :
2427 : 0 : return ret;
2428 : : }
2429 : :
2430 : 9 : static int internal_get_user_pages_fast(unsigned long start, int nr_pages,
2431 : : unsigned int gup_flags,
2432 : : struct page **pages)
2433 : : {
2434 : 9 : unsigned long addr, len, end;
2435 : 9 : int nr = 0, ret = 0;
2436 : :
2437 [ - + + - ]: 9 : if (WARN_ON_ONCE(gup_flags & ~(FOLL_WRITE | FOLL_LONGTERM |
2438 : : FOLL_FORCE | FOLL_PIN)))
2439 : : return -EINVAL;
2440 : :
2441 : 9 : start = untagged_addr(start) & PAGE_MASK;
2442 : 9 : addr = start;
2443 : 9 : len = (unsigned long) nr_pages << PAGE_SHIFT;
2444 : 9 : end = start + len;
2445 : :
2446 [ + - ]: 9 : if (end <= start)
2447 : : return 0;
2448 [ - + + - ]: 18 : if (unlikely(!access_ok((void __user *)start, len)))
2449 : : return -EFAULT;
2450 : :
2451 [ + - ]: 9 : if (IS_ENABLED(CONFIG_HAVE_FAST_GUP) &&
2452 : 9 : gup_fast_permitted(start, end)) {
2453 : 9 : local_irq_disable();
2454 : 9 : gup_pgd_range(addr, end, gup_flags, pages, &nr);
2455 : 9 : local_irq_enable();
2456 : 9 : ret = nr;
2457 : : }
2458 : :
2459 [ - + ]: 9 : if (nr < nr_pages) {
2460 : : /* Try to get the remaining pages with get_user_pages */
2461 : 0 : start += nr << PAGE_SHIFT;
2462 : 0 : pages += nr;
2463 : :
2464 : 0 : ret = __gup_longterm_unlocked(start, nr_pages - nr,
2465 : : gup_flags, pages);
2466 : :
2467 : : /* Have to be a bit careful with return values */
2468 [ # # ]: 0 : if (nr > 0) {
2469 [ # # ]: 0 : if (ret < 0)
2470 : : ret = nr;
2471 : : else
2472 : 0 : ret += nr;
2473 : : }
2474 : : }
2475 : :
2476 : : return ret;
2477 : : }
2478 : :
2479 : : /**
2480 : : * get_user_pages_fast() - pin user pages in memory
2481 : : * @start: starting user address
2482 : : * @nr_pages: number of pages from start to pin
2483 : : * @gup_flags: flags modifying pin behaviour
2484 : : * @pages: array that receives pointers to the pages pinned.
2485 : : * Should be at least nr_pages long.
2486 : : *
2487 : : * Attempt to pin user pages in memory without taking mm->mmap_sem.
2488 : : * If not successful, it will fall back to taking the lock and
2489 : : * calling get_user_pages().
2490 : : *
2491 : : * Returns number of pages pinned. This may be fewer than the number requested.
2492 : : * If nr_pages is 0 or negative, returns 0. If no pages were pinned, returns
2493 : : * -errno.
2494 : : */
2495 : 9 : int get_user_pages_fast(unsigned long start, int nr_pages,
2496 : : unsigned int gup_flags, struct page **pages)
2497 : : {
2498 : : /*
2499 : : * FOLL_PIN must only be set internally by the pin_user_pages*() APIs,
2500 : : * never directly by the caller, so enforce that:
2501 : : */
2502 [ - + + - ]: 9 : if (WARN_ON_ONCE(gup_flags & FOLL_PIN))
2503 : : return -EINVAL;
2504 : :
2505 : 9 : return internal_get_user_pages_fast(start, nr_pages, gup_flags, pages);
2506 : : }
2507 : : EXPORT_SYMBOL_GPL(get_user_pages_fast);
2508 : :
2509 : : /**
2510 : : * pin_user_pages_fast() - pin user pages in memory without taking locks
2511 : : *
2512 : : * For now, this is a placeholder function, until various call sites are
2513 : : * converted to use the correct get_user_pages*() or pin_user_pages*() API. So,
2514 : : * this is identical to get_user_pages_fast().
2515 : : *
2516 : : * This is intended for Case 1 (DIO) in Documentation/vm/pin_user_pages.rst. It
2517 : : * is NOT intended for Case 2 (RDMA: long-term pins).
2518 : : */
2519 : 0 : int pin_user_pages_fast(unsigned long start, int nr_pages,
2520 : : unsigned int gup_flags, struct page **pages)
2521 : : {
2522 : : /*
2523 : : * This is a placeholder, until the pin functionality is activated.
2524 : : * Until then, just behave like the corresponding get_user_pages*()
2525 : : * routine.
2526 : : */
2527 : 0 : return get_user_pages_fast(start, nr_pages, gup_flags, pages);
2528 : : }
2529 : : EXPORT_SYMBOL_GPL(pin_user_pages_fast);
2530 : :
2531 : : /**
2532 : : * pin_user_pages_remote() - pin pages of a remote process (task != current)
2533 : : *
2534 : : * For now, this is a placeholder function, until various call sites are
2535 : : * converted to use the correct get_user_pages*() or pin_user_pages*() API. So,
2536 : : * this is identical to get_user_pages_remote().
2537 : : *
2538 : : * This is intended for Case 1 (DIO) in Documentation/vm/pin_user_pages.rst. It
2539 : : * is NOT intended for Case 2 (RDMA: long-term pins).
2540 : : */
2541 : 0 : long pin_user_pages_remote(struct task_struct *tsk, struct mm_struct *mm,
2542 : : unsigned long start, unsigned long nr_pages,
2543 : : unsigned int gup_flags, struct page **pages,
2544 : : struct vm_area_struct **vmas, int *locked)
2545 : : {
2546 : : /*
2547 : : * This is a placeholder, until the pin functionality is activated.
2548 : : * Until then, just behave like the corresponding get_user_pages*()
2549 : : * routine.
2550 : : */
2551 : 0 : return get_user_pages_remote(tsk, mm, start, nr_pages, gup_flags, pages,
2552 : : vmas, locked);
2553 : : }
2554 : : EXPORT_SYMBOL(pin_user_pages_remote);
2555 : :
2556 : : /**
2557 : : * pin_user_pages() - pin user pages in memory for use by other devices
2558 : : *
2559 : : * For now, this is a placeholder function, until various call sites are
2560 : : * converted to use the correct get_user_pages*() or pin_user_pages*() API. So,
2561 : : * this is identical to get_user_pages().
2562 : : *
2563 : : * This is intended for Case 1 (DIO) in Documentation/vm/pin_user_pages.rst. It
2564 : : * is NOT intended for Case 2 (RDMA: long-term pins).
2565 : : */
2566 : 0 : long pin_user_pages(unsigned long start, unsigned long nr_pages,
2567 : : unsigned int gup_flags, struct page **pages,
2568 : : struct vm_area_struct **vmas)
2569 : : {
2570 : : /*
2571 : : * This is a placeholder, until the pin functionality is activated.
2572 : : * Until then, just behave like the corresponding get_user_pages*()
2573 : : * routine.
2574 : : */
2575 : 0 : return get_user_pages(start, nr_pages, gup_flags, pages, vmas);
2576 : : }
2577 : : EXPORT_SYMBOL(pin_user_pages);
|
