Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/filemap.c | |
4 | * | |
5 | * Copyright (C) 1994-1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * This file handles the generic file mmap semantics used by | |
10 | * most "normal" filesystems (but you don't /have/ to use this: | |
11 | * the NFS filesystem used to do this differently, for example) | |
12 | */ | |
b95f1b31 | 13 | #include <linux/export.h> |
1da177e4 | 14 | #include <linux/compiler.h> |
f9fe48be | 15 | #include <linux/dax.h> |
1da177e4 | 16 | #include <linux/fs.h> |
3f07c014 | 17 | #include <linux/sched/signal.h> |
c22ce143 | 18 | #include <linux/uaccess.h> |
c59ede7b | 19 | #include <linux/capability.h> |
1da177e4 | 20 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/mm.h> |
23 | #include <linux/swap.h> | |
24 | #include <linux/mman.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/uio.h> | |
cfcbfb13 | 28 | #include <linux/error-injection.h> |
1da177e4 LT |
29 | #include <linux/hash.h> |
30 | #include <linux/writeback.h> | |
53253383 | 31 | #include <linux/backing-dev.h> |
1da177e4 | 32 | #include <linux/pagevec.h> |
1da177e4 | 33 | #include <linux/security.h> |
44110fe3 | 34 | #include <linux/cpuset.h> |
00501b53 | 35 | #include <linux/hugetlb.h> |
8a9f3ccd | 36 | #include <linux/memcontrol.h> |
c515e1fd | 37 | #include <linux/cleancache.h> |
c7df8ad2 | 38 | #include <linux/shmem_fs.h> |
f1820361 | 39 | #include <linux/rmap.h> |
b1d29ba8 | 40 | #include <linux/delayacct.h> |
eb414681 | 41 | #include <linux/psi.h> |
d0e6a582 | 42 | #include <linux/ramfs.h> |
b9306a79 | 43 | #include <linux/page_idle.h> |
f9ce0be7 | 44 | #include <asm/pgalloc.h> |
de591a82 | 45 | #include <asm/tlbflush.h> |
0f8053a5 NP |
46 | #include "internal.h" |
47 | ||
fe0bfaaf RJ |
48 | #define CREATE_TRACE_POINTS |
49 | #include <trace/events/filemap.h> | |
50 | ||
1da177e4 | 51 | /* |
1da177e4 LT |
52 | * FIXME: remove all knowledge of the buffer layer from the core VM |
53 | */ | |
148f948b | 54 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 55 | |
1da177e4 LT |
56 | #include <asm/mman.h> |
57 | ||
58 | /* | |
59 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
60 | * though. | |
61 | * | |
62 | * Shared mappings now work. 15.8.1995 Bruno. | |
63 | * | |
64 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
65 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
66 | * | |
67 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
68 | */ | |
69 | ||
70 | /* | |
71 | * Lock ordering: | |
72 | * | |
c8c06efa | 73 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 74 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 75 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 76 | * ->i_pages lock |
1da177e4 | 77 | * |
9608703e | 78 | * ->i_rwsem |
730633f0 JK |
79 | * ->invalidate_lock (acquired by fs in truncate path) |
80 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) | |
1da177e4 | 81 | * |
c1e8d7c6 | 82 | * ->mmap_lock |
c8c06efa | 83 | * ->i_mmap_rwsem |
b8072f09 | 84 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 85 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 86 | * |
c1e8d7c6 | 87 | * ->mmap_lock |
730633f0 JK |
88 | * ->invalidate_lock (filemap_fault) |
89 | * ->lock_page (filemap_fault, access_process_vm) | |
1da177e4 | 90 | * |
9608703e | 91 | * ->i_rwsem (generic_perform_write) |
bb523b40 | 92 | * ->mmap_lock (fault_in_readable->do_page_fault) |
1da177e4 | 93 | * |
f758eeab | 94 | * bdi->wb.list_lock |
a66979ab | 95 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 96 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 97 | * |
c8c06efa | 98 | * ->i_mmap_rwsem |
1da177e4 LT |
99 | * ->anon_vma.lock (vma_adjust) |
100 | * | |
101 | * ->anon_vma.lock | |
b8072f09 | 102 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 103 | * |
b8072f09 | 104 | * ->page_table_lock or pte_lock |
5d337b91 | 105 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 106 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 107 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
108 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
109 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 110 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 111 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 112 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 113 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 114 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 115 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 116 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
117 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
118 | * | |
c8c06efa | 119 | * ->i_mmap_rwsem |
9a3c531d | 120 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
121 | */ |
122 | ||
5c024e6a | 123 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
124 | struct page *page, void *shadow) |
125 | { | |
5c024e6a MW |
126 | XA_STATE(xas, &mapping->i_pages, page->index); |
127 | unsigned int nr = 1; | |
c70b647d | 128 | |
5c024e6a | 129 | mapping_set_update(&xas, mapping); |
c70b647d | 130 | |
5c024e6a MW |
131 | /* hugetlb pages are represented by a single entry in the xarray */ |
132 | if (!PageHuge(page)) { | |
133 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 134 | nr = compound_nr(page); |
5c024e6a | 135 | } |
91b0abe3 | 136 | |
83929372 KS |
137 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
138 | VM_BUG_ON_PAGE(PageTail(page), page); | |
139 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 140 | |
5c024e6a MW |
141 | xas_store(&xas, shadow); |
142 | xas_init_marks(&xas); | |
d3798ae8 | 143 | |
2300638b JK |
144 | page->mapping = NULL; |
145 | /* Leave page->index set: truncation lookup relies upon it */ | |
d3798ae8 | 146 | mapping->nrpages -= nr; |
91b0abe3 JW |
147 | } |
148 | ||
5ecc4d85 JK |
149 | static void unaccount_page_cache_page(struct address_space *mapping, |
150 | struct page *page) | |
1da177e4 | 151 | { |
5ecc4d85 | 152 | int nr; |
1da177e4 | 153 | |
c515e1fd DM |
154 | /* |
155 | * if we're uptodate, flush out into the cleancache, otherwise | |
156 | * invalidate any existing cleancache entries. We can't leave | |
157 | * stale data around in the cleancache once our page is gone | |
158 | */ | |
159 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
160 | cleancache_put_page(page); | |
161 | else | |
3167760f | 162 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 163 | |
83929372 | 164 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
165 | VM_BUG_ON_PAGE(page_mapped(page), page); |
166 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
167 | int mapcount; | |
168 | ||
169 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
170 | current->comm, page_to_pfn(page)); | |
171 | dump_page(page, "still mapped when deleted"); | |
172 | dump_stack(); | |
173 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
174 | ||
175 | mapcount = page_mapcount(page); | |
176 | if (mapping_exiting(mapping) && | |
177 | page_count(page) >= mapcount + 2) { | |
178 | /* | |
179 | * All vmas have already been torn down, so it's | |
180 | * a good bet that actually the page is unmapped, | |
181 | * and we'd prefer not to leak it: if we're wrong, | |
182 | * some other bad page check should catch it later. | |
183 | */ | |
184 | page_mapcount_reset(page); | |
6d061f9f | 185 | page_ref_sub(page, mapcount); |
06b241f3 HD |
186 | } |
187 | } | |
188 | ||
4165b9b4 | 189 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
190 | if (PageHuge(page)) |
191 | return; | |
09612fa6 | 192 | |
6c357848 | 193 | nr = thp_nr_pages(page); |
5ecc4d85 | 194 | |
0d1c2072 | 195 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 196 | if (PageSwapBacked(page)) { |
0d1c2072 | 197 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 | 198 | if (PageTransHuge(page)) |
57b2847d | 199 | __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr); |
99cb0dbd | 200 | } else if (PageTransHuge(page)) { |
bf9ecead | 201 | __mod_lruvec_page_state(page, NR_FILE_THPS, -nr); |
09d91cda | 202 | filemap_nr_thps_dec(mapping); |
800d8c63 | 203 | } |
5ecc4d85 JK |
204 | |
205 | /* | |
206 | * At this point page must be either written or cleaned by | |
207 | * truncate. Dirty page here signals a bug and loss of | |
208 | * unwritten data. | |
209 | * | |
210 | * This fixes dirty accounting after removing the page entirely | |
211 | * but leaves PageDirty set: it has no effect for truncated | |
212 | * page and anyway will be cleared before returning page into | |
213 | * buddy allocator. | |
214 | */ | |
215 | if (WARN_ON_ONCE(PageDirty(page))) | |
216 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
217 | } | |
218 | ||
219 | /* | |
220 | * Delete a page from the page cache and free it. Caller has to make | |
221 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 222 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
223 | */ |
224 | void __delete_from_page_cache(struct page *page, void *shadow) | |
225 | { | |
226 | struct address_space *mapping = page->mapping; | |
227 | ||
228 | trace_mm_filemap_delete_from_page_cache(page); | |
229 | ||
230 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 231 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
232 | } |
233 | ||
59c66c5f JK |
234 | static void page_cache_free_page(struct address_space *mapping, |
235 | struct page *page) | |
236 | { | |
237 | void (*freepage)(struct page *); | |
238 | ||
239 | freepage = mapping->a_ops->freepage; | |
240 | if (freepage) | |
241 | freepage(page); | |
242 | ||
243 | if (PageTransHuge(page) && !PageHuge(page)) { | |
887b22c6 | 244 | page_ref_sub(page, thp_nr_pages(page)); |
59c66c5f JK |
245 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); |
246 | } else { | |
247 | put_page(page); | |
248 | } | |
249 | } | |
250 | ||
702cfbf9 MK |
251 | /** |
252 | * delete_from_page_cache - delete page from page cache | |
253 | * @page: the page which the kernel is trying to remove from page cache | |
254 | * | |
255 | * This must be called only on pages that have been verified to be in the page | |
256 | * cache and locked. It will never put the page into the free list, the caller | |
257 | * has a reference on the page. | |
258 | */ | |
259 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 260 | { |
83929372 | 261 | struct address_space *mapping = page_mapping(page); |
1da177e4 | 262 | |
cd7619d6 | 263 | BUG_ON(!PageLocked(page)); |
30472509 | 264 | xa_lock_irq(&mapping->i_pages); |
62cccb8c | 265 | __delete_from_page_cache(page, NULL); |
30472509 | 266 | xa_unlock_irq(&mapping->i_pages); |
6072d13c | 267 | |
59c66c5f | 268 | page_cache_free_page(mapping, page); |
97cecb5a MK |
269 | } |
270 | EXPORT_SYMBOL(delete_from_page_cache); | |
271 | ||
aa65c29c | 272 | /* |
ef8e5717 | 273 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
274 | * @mapping: the mapping to which pages belong |
275 | * @pvec: pagevec with pages to delete | |
276 | * | |
b93b0163 | 277 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
278 | * from the mapping. The function expects @pvec to be sorted by page index |
279 | * and is optimised for it to be dense. | |
b93b0163 | 280 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 281 | * modified). The function expects only THP head pages to be present in the |
4101196b | 282 | * @pvec. |
aa65c29c | 283 | * |
b93b0163 | 284 | * The function expects the i_pages lock to be held. |
aa65c29c | 285 | */ |
ef8e5717 | 286 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
287 | struct pagevec *pvec) |
288 | { | |
ef8e5717 | 289 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 290 | int total_pages = 0; |
4101196b | 291 | int i = 0; |
aa65c29c | 292 | struct page *page; |
aa65c29c | 293 | |
ef8e5717 MW |
294 | mapping_set_update(&xas, mapping); |
295 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 296 | if (i >= pagevec_count(pvec)) |
aa65c29c | 297 | break; |
4101196b MWO |
298 | |
299 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 300 | if (xa_is_value(page)) |
aa65c29c | 301 | continue; |
4101196b MWO |
302 | /* |
303 | * A page got inserted in our range? Skip it. We have our | |
304 | * pages locked so they are protected from being removed. | |
305 | * If we see a page whose index is higher than ours, it | |
306 | * means our page has been removed, which shouldn't be | |
307 | * possible because we're holding the PageLock. | |
308 | */ | |
309 | if (page != pvec->pages[i]) { | |
310 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
311 | page); | |
312 | continue; | |
313 | } | |
314 | ||
315 | WARN_ON_ONCE(!PageLocked(page)); | |
316 | ||
317 | if (page->index == xas.xa_index) | |
aa65c29c | 318 | page->mapping = NULL; |
4101196b MWO |
319 | /* Leave page->index set: truncation lookup relies on it */ |
320 | ||
321 | /* | |
322 | * Move to the next page in the vector if this is a regular | |
323 | * page or the index is of the last sub-page of this compound | |
324 | * page. | |
325 | */ | |
326 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 327 | i++; |
ef8e5717 | 328 | xas_store(&xas, NULL); |
aa65c29c JK |
329 | total_pages++; |
330 | } | |
331 | mapping->nrpages -= total_pages; | |
332 | } | |
333 | ||
334 | void delete_from_page_cache_batch(struct address_space *mapping, | |
335 | struct pagevec *pvec) | |
336 | { | |
337 | int i; | |
aa65c29c JK |
338 | |
339 | if (!pagevec_count(pvec)) | |
340 | return; | |
341 | ||
30472509 | 342 | xa_lock_irq(&mapping->i_pages); |
aa65c29c JK |
343 | for (i = 0; i < pagevec_count(pvec); i++) { |
344 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
345 | ||
346 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
347 | } | |
ef8e5717 | 348 | page_cache_delete_batch(mapping, pvec); |
30472509 | 349 | xa_unlock_irq(&mapping->i_pages); |
aa65c29c JK |
350 | |
351 | for (i = 0; i < pagevec_count(pvec); i++) | |
352 | page_cache_free_page(mapping, pvec->pages[i]); | |
353 | } | |
354 | ||
d72d9e2a | 355 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
356 | { |
357 | int ret = 0; | |
358 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
359 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
360 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 361 | ret = -ENOSPC; |
7fcbbaf1 JA |
362 | if (test_bit(AS_EIO, &mapping->flags) && |
363 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
364 | ret = -EIO; |
365 | return ret; | |
366 | } | |
d72d9e2a | 367 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 368 | |
76341cab JL |
369 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
370 | { | |
371 | /* Check for outstanding write errors */ | |
372 | if (test_bit(AS_EIO, &mapping->flags)) | |
373 | return -EIO; | |
374 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
375 | return -ENOSPC; | |
376 | return 0; | |
377 | } | |
378 | ||
5a798493 JB |
379 | /** |
380 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range | |
381 | * @mapping: address space structure to write | |
382 | * @wbc: the writeback_control controlling the writeout | |
383 | * | |
384 | * Call writepages on the mapping using the provided wbc to control the | |
385 | * writeout. | |
386 | * | |
387 | * Return: %0 on success, negative error code otherwise. | |
388 | */ | |
389 | int filemap_fdatawrite_wbc(struct address_space *mapping, | |
390 | struct writeback_control *wbc) | |
391 | { | |
392 | int ret; | |
393 | ||
394 | if (!mapping_can_writeback(mapping) || | |
395 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
396 | return 0; | |
397 | ||
398 | wbc_attach_fdatawrite_inode(wbc, mapping->host); | |
399 | ret = do_writepages(mapping, wbc); | |
400 | wbc_detach_inode(wbc); | |
401 | return ret; | |
402 | } | |
403 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); | |
404 | ||
1da177e4 | 405 | /** |
485bb99b | 406 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
407 | * @mapping: address space structure to write |
408 | * @start: offset in bytes where the range starts | |
469eb4d0 | 409 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 410 | * @sync_mode: enable synchronous operation |
1da177e4 | 411 | * |
485bb99b RD |
412 | * Start writeback against all of a mapping's dirty pages that lie |
413 | * within the byte offsets <start, end> inclusive. | |
414 | * | |
1da177e4 | 415 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 416 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
417 | * these two operations is that if a dirty page/buffer is encountered, it must |
418 | * be waited upon, and not just skipped over. | |
a862f68a MR |
419 | * |
420 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 421 | */ |
ebcf28e1 AM |
422 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
423 | loff_t end, int sync_mode) | |
1da177e4 | 424 | { |
1da177e4 LT |
425 | struct writeback_control wbc = { |
426 | .sync_mode = sync_mode, | |
05fe478d | 427 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
428 | .range_start = start, |
429 | .range_end = end, | |
1da177e4 LT |
430 | }; |
431 | ||
5a798493 | 432 | return filemap_fdatawrite_wbc(mapping, &wbc); |
1da177e4 LT |
433 | } |
434 | ||
435 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
436 | int sync_mode) | |
437 | { | |
111ebb6e | 438 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
439 | } |
440 | ||
441 | int filemap_fdatawrite(struct address_space *mapping) | |
442 | { | |
443 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
444 | } | |
445 | EXPORT_SYMBOL(filemap_fdatawrite); | |
446 | ||
f4c0a0fd | 447 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 448 | loff_t end) |
1da177e4 LT |
449 | { |
450 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
451 | } | |
f4c0a0fd | 452 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 453 | |
485bb99b RD |
454 | /** |
455 | * filemap_flush - mostly a non-blocking flush | |
456 | * @mapping: target address_space | |
457 | * | |
1da177e4 LT |
458 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
459 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
460 | * |
461 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
462 | */ |
463 | int filemap_flush(struct address_space *mapping) | |
464 | { | |
465 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
466 | } | |
467 | EXPORT_SYMBOL(filemap_flush); | |
468 | ||
7fc9e472 GR |
469 | /** |
470 | * filemap_range_has_page - check if a page exists in range. | |
471 | * @mapping: address space within which to check | |
472 | * @start_byte: offset in bytes where the range starts | |
473 | * @end_byte: offset in bytes where the range ends (inclusive) | |
474 | * | |
475 | * Find at least one page in the range supplied, usually used to check if | |
476 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
477 | * |
478 | * Return: %true if at least one page exists in the specified range, | |
479 | * %false otherwise. | |
7fc9e472 GR |
480 | */ |
481 | bool filemap_range_has_page(struct address_space *mapping, | |
482 | loff_t start_byte, loff_t end_byte) | |
483 | { | |
f7b68046 | 484 | struct page *page; |
8fa8e538 MW |
485 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
486 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
487 | |
488 | if (end_byte < start_byte) | |
489 | return false; | |
490 | ||
8fa8e538 MW |
491 | rcu_read_lock(); |
492 | for (;;) { | |
493 | page = xas_find(&xas, max); | |
494 | if (xas_retry(&xas, page)) | |
495 | continue; | |
496 | /* Shadow entries don't count */ | |
497 | if (xa_is_value(page)) | |
498 | continue; | |
499 | /* | |
500 | * We don't need to try to pin this page; we're about to | |
501 | * release the RCU lock anyway. It is enough to know that | |
502 | * there was a page here recently. | |
503 | */ | |
504 | break; | |
505 | } | |
506 | rcu_read_unlock(); | |
7fc9e472 | 507 | |
8fa8e538 | 508 | return page != NULL; |
7fc9e472 GR |
509 | } |
510 | EXPORT_SYMBOL(filemap_range_has_page); | |
511 | ||
5e8fcc1a | 512 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 513 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 514 | { |
09cbfeaf KS |
515 | pgoff_t index = start_byte >> PAGE_SHIFT; |
516 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
517 | struct pagevec pvec; |
518 | int nr_pages; | |
1da177e4 | 519 | |
94004ed7 | 520 | if (end_byte < start_byte) |
5e8fcc1a | 521 | return; |
1da177e4 | 522 | |
86679820 | 523 | pagevec_init(&pvec); |
312e9d2f | 524 | while (index <= end) { |
1da177e4 LT |
525 | unsigned i; |
526 | ||
312e9d2f | 527 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 528 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
529 | if (!nr_pages) |
530 | break; | |
531 | ||
1da177e4 LT |
532 | for (i = 0; i < nr_pages; i++) { |
533 | struct page *page = pvec.pages[i]; | |
534 | ||
1da177e4 | 535 | wait_on_page_writeback(page); |
5e8fcc1a | 536 | ClearPageError(page); |
1da177e4 LT |
537 | } |
538 | pagevec_release(&pvec); | |
539 | cond_resched(); | |
540 | } | |
aa750fd7 JN |
541 | } |
542 | ||
543 | /** | |
544 | * filemap_fdatawait_range - wait for writeback to complete | |
545 | * @mapping: address space structure to wait for | |
546 | * @start_byte: offset in bytes where the range starts | |
547 | * @end_byte: offset in bytes where the range ends (inclusive) | |
548 | * | |
549 | * Walk the list of under-writeback pages of the given address space | |
550 | * in the given range and wait for all of them. Check error status of | |
551 | * the address space and return it. | |
552 | * | |
553 | * Since the error status of the address space is cleared by this function, | |
554 | * callers are responsible for checking the return value and handling and/or | |
555 | * reporting the error. | |
a862f68a MR |
556 | * |
557 | * Return: error status of the address space. | |
aa750fd7 JN |
558 | */ |
559 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
560 | loff_t end_byte) | |
561 | { | |
5e8fcc1a JL |
562 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
563 | return filemap_check_errors(mapping); | |
1da177e4 | 564 | } |
d3bccb6f JK |
565 | EXPORT_SYMBOL(filemap_fdatawait_range); |
566 | ||
aa0bfcd9 RZ |
567 | /** |
568 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
569 | * @mapping: address space structure to wait for | |
570 | * @start_byte: offset in bytes where the range starts | |
571 | * @end_byte: offset in bytes where the range ends (inclusive) | |
572 | * | |
573 | * Walk the list of under-writeback pages of the given address space in the | |
574 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
575 | * this function does not clear error status of the address space. | |
576 | * | |
577 | * Use this function if callers don't handle errors themselves. Expected | |
578 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
579 | * fsfreeze(8) | |
580 | */ | |
581 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
582 | loff_t start_byte, loff_t end_byte) | |
583 | { | |
584 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
585 | return filemap_check_and_keep_errors(mapping); | |
586 | } | |
587 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
588 | ||
a823e458 JL |
589 | /** |
590 | * file_fdatawait_range - wait for writeback to complete | |
591 | * @file: file pointing to address space structure to wait for | |
592 | * @start_byte: offset in bytes where the range starts | |
593 | * @end_byte: offset in bytes where the range ends (inclusive) | |
594 | * | |
595 | * Walk the list of under-writeback pages of the address space that file | |
596 | * refers to, in the given range and wait for all of them. Check error | |
597 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
598 | * | |
599 | * Since the error status of the file is advanced by this function, | |
600 | * callers are responsible for checking the return value and handling and/or | |
601 | * reporting the error. | |
a862f68a MR |
602 | * |
603 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
604 | */ |
605 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
606 | { | |
607 | struct address_space *mapping = file->f_mapping; | |
608 | ||
609 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
610 | return file_check_and_advance_wb_err(file); | |
611 | } | |
612 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 613 | |
aa750fd7 JN |
614 | /** |
615 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
616 | * @mapping: address space structure to wait for | |
617 | * | |
618 | * Walk the list of under-writeback pages of the given address space | |
619 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
620 | * does not clear error status of the address space. | |
621 | * | |
622 | * Use this function if callers don't handle errors themselves. Expected | |
623 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
624 | * fsfreeze(8) | |
a862f68a MR |
625 | * |
626 | * Return: error status of the address space. | |
aa750fd7 | 627 | */ |
76341cab | 628 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 629 | { |
ffb959bb | 630 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 631 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 632 | } |
76341cab | 633 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 634 | |
875d91b1 | 635 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 636 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 637 | { |
875d91b1 | 638 | return mapping->nrpages; |
1da177e4 | 639 | } |
1da177e4 | 640 | |
63135aa3 JA |
641 | /** |
642 | * filemap_range_needs_writeback - check if range potentially needs writeback | |
643 | * @mapping: address space within which to check | |
644 | * @start_byte: offset in bytes where the range starts | |
645 | * @end_byte: offset in bytes where the range ends (inclusive) | |
646 | * | |
647 | * Find at least one page in the range supplied, usually used to check if | |
648 | * direct writing in this range will trigger a writeback. Used by O_DIRECT | |
649 | * read/write with IOCB_NOWAIT, to see if the caller needs to do | |
650 | * filemap_write_and_wait_range() before proceeding. | |
651 | * | |
652 | * Return: %true if the caller should do filemap_write_and_wait_range() before | |
653 | * doing O_DIRECT to a page in this range, %false otherwise. | |
654 | */ | |
655 | bool filemap_range_needs_writeback(struct address_space *mapping, | |
656 | loff_t start_byte, loff_t end_byte) | |
657 | { | |
658 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
659 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
660 | struct page *page; | |
661 | ||
662 | if (!mapping_needs_writeback(mapping)) | |
663 | return false; | |
664 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && | |
665 | !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
666 | return false; | |
667 | if (end_byte < start_byte) | |
668 | return false; | |
669 | ||
670 | rcu_read_lock(); | |
671 | xas_for_each(&xas, page, max) { | |
672 | if (xas_retry(&xas, page)) | |
673 | continue; | |
674 | if (xa_is_value(page)) | |
675 | continue; | |
676 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
677 | break; | |
678 | } | |
679 | rcu_read_unlock(); | |
680 | return page != NULL; | |
681 | } | |
682 | EXPORT_SYMBOL_GPL(filemap_range_needs_writeback); | |
683 | ||
485bb99b RD |
684 | /** |
685 | * filemap_write_and_wait_range - write out & wait on a file range | |
686 | * @mapping: the address_space for the pages | |
687 | * @lstart: offset in bytes where the range starts | |
688 | * @lend: offset in bytes where the range ends (inclusive) | |
689 | * | |
469eb4d0 AM |
690 | * Write out and wait upon file offsets lstart->lend, inclusive. |
691 | * | |
0e056eb5 | 692 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 693 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
694 | * |
695 | * Return: error status of the address space. | |
469eb4d0 | 696 | */ |
1da177e4 LT |
697 | int filemap_write_and_wait_range(struct address_space *mapping, |
698 | loff_t lstart, loff_t lend) | |
699 | { | |
28fd1298 | 700 | int err = 0; |
1da177e4 | 701 | |
9326c9b2 | 702 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
703 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
704 | WB_SYNC_ALL); | |
ddf8f376 IW |
705 | /* |
706 | * Even if the above returned error, the pages may be | |
707 | * written partially (e.g. -ENOSPC), so we wait for it. | |
708 | * But the -EIO is special case, it may indicate the worst | |
709 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
710 | */ | |
28fd1298 | 711 | if (err != -EIO) { |
94004ed7 CH |
712 | int err2 = filemap_fdatawait_range(mapping, |
713 | lstart, lend); | |
28fd1298 OH |
714 | if (!err) |
715 | err = err2; | |
cbeaf951 JL |
716 | } else { |
717 | /* Clear any previously stored errors */ | |
718 | filemap_check_errors(mapping); | |
28fd1298 | 719 | } |
865ffef3 DM |
720 | } else { |
721 | err = filemap_check_errors(mapping); | |
1da177e4 | 722 | } |
28fd1298 | 723 | return err; |
1da177e4 | 724 | } |
f6995585 | 725 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 726 | |
5660e13d JL |
727 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
728 | { | |
3acdfd28 | 729 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
730 | |
731 | trace_filemap_set_wb_err(mapping, eseq); | |
732 | } | |
733 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
734 | ||
735 | /** | |
736 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
737 | * and advance wb_err to current one | |
738 | * @file: struct file on which the error is being reported | |
739 | * | |
740 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
741 | * want to report any writeback errors that occurred since the last fsync (or | |
742 | * since the file was opened if there haven't been any). | |
743 | * | |
744 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
745 | * then just quickly return 0. The file is all caught up. | |
746 | * | |
747 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
748 | * it and try to swap it into place. If it works, or another task beat us | |
749 | * to it with the new value, then update the f_wb_err and return the error | |
750 | * portion. The error at this point must be reported via proper channels | |
751 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
752 | * | |
753 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
754 | * value is protected by the f_lock since we must ensure that it reflects | |
755 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
756 | * |
757 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
758 | */ |
759 | int file_check_and_advance_wb_err(struct file *file) | |
760 | { | |
761 | int err = 0; | |
762 | errseq_t old = READ_ONCE(file->f_wb_err); | |
763 | struct address_space *mapping = file->f_mapping; | |
764 | ||
765 | /* Locklessly handle the common case where nothing has changed */ | |
766 | if (errseq_check(&mapping->wb_err, old)) { | |
767 | /* Something changed, must use slow path */ | |
768 | spin_lock(&file->f_lock); | |
769 | old = file->f_wb_err; | |
770 | err = errseq_check_and_advance(&mapping->wb_err, | |
771 | &file->f_wb_err); | |
772 | trace_file_check_and_advance_wb_err(file, old); | |
773 | spin_unlock(&file->f_lock); | |
774 | } | |
f4e222c5 JL |
775 | |
776 | /* | |
777 | * We're mostly using this function as a drop in replacement for | |
778 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
779 | * that the legacy code would have had on these flags. | |
780 | */ | |
781 | clear_bit(AS_EIO, &mapping->flags); | |
782 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
783 | return err; |
784 | } | |
785 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
786 | ||
787 | /** | |
788 | * file_write_and_wait_range - write out & wait on a file range | |
789 | * @file: file pointing to address_space with pages | |
790 | * @lstart: offset in bytes where the range starts | |
791 | * @lend: offset in bytes where the range ends (inclusive) | |
792 | * | |
793 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
794 | * | |
795 | * Note that @lend is inclusive (describes the last byte to be written) so | |
796 | * that this function can be used to write to the very end-of-file (end = -1). | |
797 | * | |
798 | * After writing out and waiting on the data, we check and advance the | |
799 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
800 | * |
801 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
802 | */ |
803 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
804 | { | |
805 | int err = 0, err2; | |
806 | struct address_space *mapping = file->f_mapping; | |
807 | ||
9326c9b2 | 808 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
809 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
810 | WB_SYNC_ALL); | |
811 | /* See comment of filemap_write_and_wait() */ | |
812 | if (err != -EIO) | |
813 | __filemap_fdatawait_range(mapping, lstart, lend); | |
814 | } | |
815 | err2 = file_check_and_advance_wb_err(file); | |
816 | if (!err) | |
817 | err = err2; | |
818 | return err; | |
819 | } | |
820 | EXPORT_SYMBOL(file_write_and_wait_range); | |
821 | ||
ef6a3c63 MS |
822 | /** |
823 | * replace_page_cache_page - replace a pagecache page with a new one | |
824 | * @old: page to be replaced | |
825 | * @new: page to replace with | |
ef6a3c63 MS |
826 | * |
827 | * This function replaces a page in the pagecache with a new one. On | |
828 | * success it acquires the pagecache reference for the new page and | |
829 | * drops it for the old page. Both the old and new pages must be | |
830 | * locked. This function does not add the new page to the LRU, the | |
831 | * caller must do that. | |
832 | * | |
74d60958 | 833 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 834 | */ |
1f7ef657 | 835 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 836 | { |
d21bba2b MWO |
837 | struct folio *fold = page_folio(old); |
838 | struct folio *fnew = page_folio(new); | |
74d60958 MW |
839 | struct address_space *mapping = old->mapping; |
840 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
841 | pgoff_t offset = old->index; | |
842 | XA_STATE(xas, &mapping->i_pages, offset); | |
ef6a3c63 | 843 | |
309381fe SL |
844 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
845 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
846 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 847 | |
74d60958 MW |
848 | get_page(new); |
849 | new->mapping = mapping; | |
850 | new->index = offset; | |
ef6a3c63 | 851 | |
d21bba2b | 852 | mem_cgroup_migrate(fold, fnew); |
0d1c2072 | 853 | |
30472509 | 854 | xas_lock_irq(&xas); |
74d60958 | 855 | xas_store(&xas, new); |
4165b9b4 | 856 | |
74d60958 MW |
857 | old->mapping = NULL; |
858 | /* hugetlb pages do not participate in page cache accounting. */ | |
859 | if (!PageHuge(old)) | |
0d1c2072 | 860 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 861 | if (!PageHuge(new)) |
0d1c2072 | 862 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 863 | if (PageSwapBacked(old)) |
0d1c2072 | 864 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 865 | if (PageSwapBacked(new)) |
0d1c2072 | 866 | __inc_lruvec_page_state(new, NR_SHMEM); |
30472509 | 867 | xas_unlock_irq(&xas); |
74d60958 MW |
868 | if (freepage) |
869 | freepage(old); | |
870 | put_page(old); | |
ef6a3c63 MS |
871 | } |
872 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
873 | ||
9dd3d069 MWO |
874 | noinline int __filemap_add_folio(struct address_space *mapping, |
875 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) | |
1da177e4 | 876 | { |
9dd3d069 MWO |
877 | XA_STATE(xas, &mapping->i_pages, index); |
878 | int huge = folio_test_hugetlb(folio); | |
e286781d | 879 | int error; |
da74240e | 880 | bool charged = false; |
e286781d | 881 | |
9dd3d069 MWO |
882 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
883 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); | |
74d60958 | 884 | mapping_set_update(&xas, mapping); |
e286781d | 885 | |
9dd3d069 MWO |
886 | folio_get(folio); |
887 | folio->mapping = mapping; | |
888 | folio->index = index; | |
66a0c8ee | 889 | |
3fea5a49 | 890 | if (!huge) { |
9dd3d069 MWO |
891 | error = mem_cgroup_charge(folio, NULL, gfp); |
892 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); | |
3fea5a49 JW |
893 | if (error) |
894 | goto error; | |
da74240e | 895 | charged = true; |
3fea5a49 JW |
896 | } |
897 | ||
198b62f8 MWO |
898 | gfp &= GFP_RECLAIM_MASK; |
899 | ||
74d60958 | 900 | do { |
198b62f8 MWO |
901 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
902 | void *entry, *old = NULL; | |
903 | ||
9dd3d069 | 904 | if (order > folio_order(folio)) |
198b62f8 MWO |
905 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
906 | order, gfp); | |
74d60958 | 907 | xas_lock_irq(&xas); |
198b62f8 MWO |
908 | xas_for_each_conflict(&xas, entry) { |
909 | old = entry; | |
910 | if (!xa_is_value(entry)) { | |
911 | xas_set_err(&xas, -EEXIST); | |
912 | goto unlock; | |
913 | } | |
914 | } | |
915 | ||
916 | if (old) { | |
917 | if (shadowp) | |
918 | *shadowp = old; | |
919 | /* entry may have been split before we acquired lock */ | |
920 | order = xa_get_order(xas.xa, xas.xa_index); | |
9dd3d069 | 921 | if (order > folio_order(folio)) { |
198b62f8 MWO |
922 | xas_split(&xas, old, order); |
923 | xas_reset(&xas); | |
924 | } | |
925 | } | |
926 | ||
9dd3d069 | 927 | xas_store(&xas, folio); |
74d60958 MW |
928 | if (xas_error(&xas)) |
929 | goto unlock; | |
930 | ||
74d60958 MW |
931 | mapping->nrpages++; |
932 | ||
933 | /* hugetlb pages do not participate in page cache accounting */ | |
934 | if (!huge) | |
9dd3d069 | 935 | __lruvec_stat_add_folio(folio, NR_FILE_PAGES); |
74d60958 MW |
936 | unlock: |
937 | xas_unlock_irq(&xas); | |
198b62f8 | 938 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 939 | |
3fea5a49 JW |
940 | if (xas_error(&xas)) { |
941 | error = xas_error(&xas); | |
da74240e | 942 | if (charged) |
9dd3d069 | 943 | mem_cgroup_uncharge(folio); |
74d60958 | 944 | goto error; |
3fea5a49 | 945 | } |
4165b9b4 | 946 | |
9dd3d069 | 947 | trace_mm_filemap_add_to_page_cache(&folio->page); |
66a0c8ee | 948 | return 0; |
74d60958 | 949 | error: |
9dd3d069 | 950 | folio->mapping = NULL; |
66a0c8ee | 951 | /* Leave page->index set: truncation relies upon it */ |
9dd3d069 | 952 | folio_put(folio); |
3fea5a49 | 953 | return error; |
1da177e4 | 954 | } |
9dd3d069 | 955 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
a528910e JW |
956 | |
957 | /** | |
958 | * add_to_page_cache_locked - add a locked page to the pagecache | |
959 | * @page: page to add | |
960 | * @mapping: the page's address_space | |
961 | * @offset: page index | |
962 | * @gfp_mask: page allocation mode | |
963 | * | |
964 | * This function is used to add a page to the pagecache. It must be locked. | |
965 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
966 | * |
967 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
968 | */ |
969 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
970 | pgoff_t offset, gfp_t gfp_mask) | |
971 | { | |
9dd3d069 | 972 | return __filemap_add_folio(mapping, page_folio(page), offset, |
a528910e JW |
973 | gfp_mask, NULL); |
974 | } | |
e286781d | 975 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 | 976 | |
9dd3d069 MWO |
977 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
978 | pgoff_t index, gfp_t gfp) | |
1da177e4 | 979 | { |
a528910e | 980 | void *shadow = NULL; |
4f98a2fe RR |
981 | int ret; |
982 | ||
9dd3d069 MWO |
983 | __folio_set_locked(folio); |
984 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); | |
a528910e | 985 | if (unlikely(ret)) |
9dd3d069 | 986 | __folio_clear_locked(folio); |
a528910e JW |
987 | else { |
988 | /* | |
9dd3d069 | 989 | * The folio might have been evicted from cache only |
a528910e | 990 | * recently, in which case it should be activated like |
9dd3d069 MWO |
991 | * any other repeatedly accessed folio. |
992 | * The exception is folios getting rewritten; evicting other | |
f0281a00 RR |
993 | * data from the working set, only to cache data that will |
994 | * get overwritten with something else, is a waste of memory. | |
a528910e | 995 | */ |
9dd3d069 MWO |
996 | WARN_ON_ONCE(folio_test_active(folio)); |
997 | if (!(gfp & __GFP_WRITE) && shadow) | |
998 | workingset_refault(folio, shadow); | |
999 | folio_add_lru(folio); | |
a528910e | 1000 | } |
1da177e4 LT |
1001 | return ret; |
1002 | } | |
9dd3d069 | 1003 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
1da177e4 | 1004 | |
44110fe3 | 1005 | #ifdef CONFIG_NUMA |
bb3c579e | 1006 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
44110fe3 | 1007 | { |
c0ff7453 | 1008 | int n; |
bb3c579e | 1009 | struct folio *folio; |
c0ff7453 | 1010 | |
44110fe3 | 1011 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
1012 | unsigned int cpuset_mems_cookie; |
1013 | do { | |
d26914d1 | 1014 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 1015 | n = cpuset_mem_spread_node(); |
bb3c579e MWO |
1016 | folio = __folio_alloc_node(gfp, order, n); |
1017 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); | |
cc9a6c87 | 1018 | |
bb3c579e | 1019 | return folio; |
44110fe3 | 1020 | } |
bb3c579e | 1021 | return folio_alloc(gfp, order); |
44110fe3 | 1022 | } |
bb3c579e | 1023 | EXPORT_SYMBOL(filemap_alloc_folio); |
44110fe3 PJ |
1024 | #endif |
1025 | ||
7506ae6a JK |
1026 | /* |
1027 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings | |
1028 | * | |
1029 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. | |
1030 | * | |
1031 | * @mapping1: the first mapping to lock | |
1032 | * @mapping2: the second mapping to lock | |
1033 | */ | |
1034 | void filemap_invalidate_lock_two(struct address_space *mapping1, | |
1035 | struct address_space *mapping2) | |
1036 | { | |
1037 | if (mapping1 > mapping2) | |
1038 | swap(mapping1, mapping2); | |
1039 | if (mapping1) | |
1040 | down_write(&mapping1->invalidate_lock); | |
1041 | if (mapping2 && mapping1 != mapping2) | |
1042 | down_write_nested(&mapping2->invalidate_lock, 1); | |
1043 | } | |
1044 | EXPORT_SYMBOL(filemap_invalidate_lock_two); | |
1045 | ||
1046 | /* | |
1047 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings | |
1048 | * | |
1049 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. | |
1050 | * | |
1051 | * @mapping1: the first mapping to unlock | |
1052 | * @mapping2: the second mapping to unlock | |
1053 | */ | |
1054 | void filemap_invalidate_unlock_two(struct address_space *mapping1, | |
1055 | struct address_space *mapping2) | |
1056 | { | |
1057 | if (mapping1) | |
1058 | up_write(&mapping1->invalidate_lock); | |
1059 | if (mapping2 && mapping1 != mapping2) | |
1060 | up_write(&mapping2->invalidate_lock); | |
1061 | } | |
1062 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); | |
1063 | ||
1da177e4 LT |
1064 | /* |
1065 | * In order to wait for pages to become available there must be | |
1066 | * waitqueues associated with pages. By using a hash table of | |
1067 | * waitqueues where the bucket discipline is to maintain all | |
1068 | * waiters on the same queue and wake all when any of the pages | |
1069 | * become available, and for the woken contexts to check to be | |
1070 | * sure the appropriate page became available, this saves space | |
1071 | * at a cost of "thundering herd" phenomena during rare hash | |
1072 | * collisions. | |
1073 | */ | |
62906027 NP |
1074 | #define PAGE_WAIT_TABLE_BITS 8 |
1075 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
df4d4f12 | 1076 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
62906027 | 1077 | |
df4d4f12 | 1078 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
1da177e4 | 1079 | { |
df4d4f12 | 1080 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1081 | } |
1da177e4 | 1082 | |
62906027 | 1083 | void __init pagecache_init(void) |
1da177e4 | 1084 | { |
62906027 | 1085 | int i; |
1da177e4 | 1086 | |
62906027 | 1087 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
df4d4f12 | 1088 | init_waitqueue_head(&folio_wait_table[i]); |
62906027 NP |
1089 | |
1090 | page_writeback_init(); | |
1da177e4 | 1091 | } |
1da177e4 | 1092 | |
5ef64cc8 LT |
1093 | /* |
1094 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1095 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1096 | * one. |
1097 | * | |
1098 | * We have: | |
1099 | * | |
1100 | * (a) no special bits set: | |
1101 | * | |
1102 | * We're just waiting for the bit to be released, and when a waker | |
1103 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1104 | * and remove it from the wait queue. | |
1105 | * | |
1106 | * Simple and straightforward. | |
1107 | * | |
1108 | * (b) WQ_FLAG_EXCLUSIVE: | |
1109 | * | |
1110 | * The waiter is waiting to get the lock, and only one waiter should | |
1111 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1112 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1113 | * | |
1114 | * This is the traditional exclusive wait. | |
1115 | * | |
5868ec26 | 1116 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1117 | * |
1118 | * The waiter is waiting to get the bit, and additionally wants the | |
1119 | * lock to be transferred to it for fair lock behavior. If the lock | |
1120 | * cannot be taken, we stop walking the wait queue without waking | |
1121 | * the waiter. | |
1122 | * | |
1123 | * This is the "fair lock handoff" case, and in addition to setting | |
1124 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1125 | * that it now has the lock. | |
1126 | */ | |
ac6424b9 | 1127 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1128 | { |
5ef64cc8 | 1129 | unsigned int flags; |
62906027 NP |
1130 | struct wait_page_key *key = arg; |
1131 | struct wait_page_queue *wait_page | |
1132 | = container_of(wait, struct wait_page_queue, wait); | |
1133 | ||
cdc8fcb4 | 1134 | if (!wake_page_match(wait_page, key)) |
62906027 | 1135 | return 0; |
3510ca20 | 1136 | |
9a1ea439 | 1137 | /* |
5ef64cc8 LT |
1138 | * If it's a lock handoff wait, we get the bit for it, and |
1139 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1140 | */ |
5ef64cc8 LT |
1141 | flags = wait->flags; |
1142 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
df4d4f12 | 1143 | if (test_bit(key->bit_nr, &key->folio->flags)) |
2a9127fc | 1144 | return -1; |
5ef64cc8 | 1145 | if (flags & WQ_FLAG_CUSTOM) { |
df4d4f12 | 1146 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
5ef64cc8 LT |
1147 | return -1; |
1148 | flags |= WQ_FLAG_DONE; | |
1149 | } | |
2a9127fc | 1150 | } |
f62e00cc | 1151 | |
5ef64cc8 LT |
1152 | /* |
1153 | * We are holding the wait-queue lock, but the waiter that | |
1154 | * is waiting for this will be checking the flags without | |
1155 | * any locking. | |
1156 | * | |
1157 | * So update the flags atomically, and wake up the waiter | |
1158 | * afterwards to avoid any races. This store-release pairs | |
101c0bf6 | 1159 | * with the load-acquire in folio_wait_bit_common(). |
5ef64cc8 LT |
1160 | */ |
1161 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1162 | wake_up_state(wait->private, mode); |
1163 | ||
1164 | /* | |
1165 | * Ok, we have successfully done what we're waiting for, | |
1166 | * and we can unconditionally remove the wait entry. | |
1167 | * | |
5ef64cc8 LT |
1168 | * Note that this pairs with the "finish_wait()" in the |
1169 | * waiter, and has to be the absolute last thing we do. | |
1170 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1171 | * might be de-allocated and the process might even have |
1172 | * exited. | |
2a9127fc | 1173 | */ |
c6fe44d9 | 1174 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1175 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1176 | } |
1177 | ||
6974d7c9 | 1178 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
cbbce822 | 1179 | { |
df4d4f12 | 1180 | wait_queue_head_t *q = folio_waitqueue(folio); |
62906027 NP |
1181 | struct wait_page_key key; |
1182 | unsigned long flags; | |
11a19c7b | 1183 | wait_queue_entry_t bookmark; |
cbbce822 | 1184 | |
df4d4f12 | 1185 | key.folio = folio; |
62906027 NP |
1186 | key.bit_nr = bit_nr; |
1187 | key.page_match = 0; | |
1188 | ||
11a19c7b TC |
1189 | bookmark.flags = 0; |
1190 | bookmark.private = NULL; | |
1191 | bookmark.func = NULL; | |
1192 | INIT_LIST_HEAD(&bookmark.entry); | |
1193 | ||
62906027 | 1194 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1195 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1196 | ||
1197 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1198 | /* | |
1199 | * Take a breather from holding the lock, | |
1200 | * allow pages that finish wake up asynchronously | |
1201 | * to acquire the lock and remove themselves | |
1202 | * from wait queue | |
1203 | */ | |
1204 | spin_unlock_irqrestore(&q->lock, flags); | |
1205 | cpu_relax(); | |
1206 | spin_lock_irqsave(&q->lock, flags); | |
1207 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1208 | } | |
1209 | ||
62906027 NP |
1210 | /* |
1211 | * It is possible for other pages to have collided on the waitqueue | |
1212 | * hash, so in that case check for a page match. That prevents a long- | |
1213 | * term waiter | |
1214 | * | |
1215 | * It is still possible to miss a case here, when we woke page waiters | |
1216 | * and removed them from the waitqueue, but there are still other | |
1217 | * page waiters. | |
1218 | */ | |
1219 | if (!waitqueue_active(q) || !key.page_match) { | |
6974d7c9 | 1220 | folio_clear_waiters(folio); |
62906027 NP |
1221 | /* |
1222 | * It's possible to miss clearing Waiters here, when we woke | |
1223 | * our page waiters, but the hashed waitqueue has waiters for | |
1224 | * other pages on it. | |
1225 | * | |
1226 | * That's okay, it's a rare case. The next waker will clear it. | |
1227 | */ | |
1228 | } | |
1229 | spin_unlock_irqrestore(&q->lock, flags); | |
1230 | } | |
74d81bfa | 1231 | |
4268b480 | 1232 | static void folio_wake(struct folio *folio, int bit) |
74d81bfa | 1233 | { |
4268b480 | 1234 | if (!folio_test_waiters(folio)) |
74d81bfa | 1235 | return; |
6974d7c9 | 1236 | folio_wake_bit(folio, bit); |
74d81bfa | 1237 | } |
62906027 | 1238 | |
9a1ea439 | 1239 | /* |
101c0bf6 | 1240 | * A choice of three behaviors for folio_wait_bit_common(): |
9a1ea439 HD |
1241 | */ |
1242 | enum behavior { | |
1243 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
7c23c782 | 1244 | * __folio_lock() waiting on then setting PG_locked. |
9a1ea439 HD |
1245 | */ |
1246 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1247 | * wait_on_page_writeback() waiting on PG_writeback. | |
1248 | */ | |
1249 | DROP, /* Drop ref to page before wait, no check when woken, | |
1250 | * like put_and_wait_on_page_locked() on PG_locked. | |
1251 | */ | |
1252 | }; | |
1253 | ||
2a9127fc | 1254 | /* |
101c0bf6 | 1255 | * Attempt to check (or get) the folio flag, and mark us done |
5ef64cc8 | 1256 | * if successful. |
2a9127fc | 1257 | */ |
101c0bf6 | 1258 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
2a9127fc LT |
1259 | struct wait_queue_entry *wait) |
1260 | { | |
1261 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
101c0bf6 | 1262 | if (test_and_set_bit(bit_nr, &folio->flags)) |
2a9127fc | 1263 | return false; |
101c0bf6 | 1264 | } else if (test_bit(bit_nr, &folio->flags)) |
2a9127fc LT |
1265 | return false; |
1266 | ||
5ef64cc8 | 1267 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1268 | return true; |
1269 | } | |
1270 | ||
5ef64cc8 LT |
1271 | /* How many times do we accept lock stealing from under a waiter? */ |
1272 | int sysctl_page_lock_unfairness = 5; | |
1273 | ||
101c0bf6 MWO |
1274 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
1275 | int state, enum behavior behavior) | |
62906027 | 1276 | { |
df4d4f12 | 1277 | wait_queue_head_t *q = folio_waitqueue(folio); |
5ef64cc8 | 1278 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1279 | struct wait_page_queue wait_page; |
ac6424b9 | 1280 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1281 | bool thrashing = false; |
9a1ea439 | 1282 | bool delayacct = false; |
eb414681 | 1283 | unsigned long pflags; |
62906027 | 1284 | |
eb414681 | 1285 | if (bit_nr == PG_locked && |
101c0bf6 MWO |
1286 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
1287 | if (!folio_test_swapbacked(folio)) { | |
eb414681 | 1288 | delayacct_thrashing_start(); |
9a1ea439 HD |
1289 | delayacct = true; |
1290 | } | |
eb414681 | 1291 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1292 | thrashing = true; |
1293 | } | |
1294 | ||
62906027 NP |
1295 | init_wait(wait); |
1296 | wait->func = wake_page_function; | |
df4d4f12 | 1297 | wait_page.folio = folio; |
62906027 NP |
1298 | wait_page.bit_nr = bit_nr; |
1299 | ||
5ef64cc8 LT |
1300 | repeat: |
1301 | wait->flags = 0; | |
1302 | if (behavior == EXCLUSIVE) { | |
1303 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1304 | if (--unfairness < 0) | |
1305 | wait->flags |= WQ_FLAG_CUSTOM; | |
1306 | } | |
1307 | ||
2a9127fc LT |
1308 | /* |
1309 | * Do one last check whether we can get the | |
1310 | * page bit synchronously. | |
1311 | * | |
101c0bf6 | 1312 | * Do the folio_set_waiters() marking before that |
2a9127fc LT |
1313 | * to let any waker we _just_ missed know they |
1314 | * need to wake us up (otherwise they'll never | |
1315 | * even go to the slow case that looks at the | |
1316 | * page queue), and add ourselves to the wait | |
1317 | * queue if we need to sleep. | |
1318 | * | |
1319 | * This part needs to be done under the queue | |
1320 | * lock to avoid races. | |
1321 | */ | |
1322 | spin_lock_irq(&q->lock); | |
101c0bf6 MWO |
1323 | folio_set_waiters(folio); |
1324 | if (!folio_trylock_flag(folio, bit_nr, wait)) | |
2a9127fc LT |
1325 | __add_wait_queue_entry_tail(q, wait); |
1326 | spin_unlock_irq(&q->lock); | |
62906027 | 1327 | |
2a9127fc LT |
1328 | /* |
1329 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1330 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1331 | * see whether the page bit testing has already | |
1332 | * been done by the wake function. | |
2a9127fc | 1333 | * |
101c0bf6 | 1334 | * We can drop our reference to the folio. |
2a9127fc LT |
1335 | */ |
1336 | if (behavior == DROP) | |
101c0bf6 | 1337 | folio_put(folio); |
62906027 | 1338 | |
5ef64cc8 LT |
1339 | /* |
1340 | * Note that until the "finish_wait()", or until | |
1341 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1342 | * be very careful with the 'wait->flags', because | |
1343 | * we may race with a waker that sets them. | |
1344 | */ | |
2a9127fc | 1345 | for (;;) { |
5ef64cc8 LT |
1346 | unsigned int flags; |
1347 | ||
62906027 NP |
1348 | set_current_state(state); |
1349 | ||
5ef64cc8 LT |
1350 | /* Loop until we've been woken or interrupted */ |
1351 | flags = smp_load_acquire(&wait->flags); | |
1352 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1353 | if (signal_pending_state(state, current)) | |
1354 | break; | |
1355 | ||
1356 | io_schedule(); | |
1357 | continue; | |
1358 | } | |
1359 | ||
1360 | /* If we were non-exclusive, we're done */ | |
1361 | if (behavior != EXCLUSIVE) | |
a8b169af | 1362 | break; |
9a1ea439 | 1363 | |
5ef64cc8 LT |
1364 | /* If the waker got the lock for us, we're done */ |
1365 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1366 | break; |
2a9127fc | 1367 | |
5ef64cc8 LT |
1368 | /* |
1369 | * Otherwise, if we're getting the lock, we need to | |
1370 | * try to get it ourselves. | |
1371 | * | |
1372 | * And if that fails, we'll have to retry this all. | |
1373 | */ | |
101c0bf6 | 1374 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
5ef64cc8 LT |
1375 | goto repeat; |
1376 | ||
1377 | wait->flags |= WQ_FLAG_DONE; | |
1378 | break; | |
62906027 NP |
1379 | } |
1380 | ||
5ef64cc8 LT |
1381 | /* |
1382 | * If a signal happened, this 'finish_wait()' may remove the last | |
101c0bf6 | 1383 | * waiter from the wait-queues, but the folio waiters bit will remain |
5ef64cc8 LT |
1384 | * set. That's ok. The next wakeup will take care of it, and trying |
1385 | * to do it here would be difficult and prone to races. | |
1386 | */ | |
62906027 NP |
1387 | finish_wait(q, wait); |
1388 | ||
eb414681 | 1389 | if (thrashing) { |
9a1ea439 | 1390 | if (delayacct) |
eb414681 JW |
1391 | delayacct_thrashing_end(); |
1392 | psi_memstall_leave(&pflags); | |
1393 | } | |
b1d29ba8 | 1394 | |
62906027 | 1395 | /* |
5ef64cc8 LT |
1396 | * NOTE! The wait->flags weren't stable until we've done the |
1397 | * 'finish_wait()', and we could have exited the loop above due | |
1398 | * to a signal, and had a wakeup event happen after the signal | |
1399 | * test but before the 'finish_wait()'. | |
1400 | * | |
1401 | * So only after the finish_wait() can we reliably determine | |
1402 | * if we got woken up or not, so we can now figure out the final | |
1403 | * return value based on that state without races. | |
1404 | * | |
1405 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1406 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1407 | */ |
5ef64cc8 LT |
1408 | if (behavior == EXCLUSIVE) |
1409 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1410 | |
2a9127fc | 1411 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1412 | } |
1413 | ||
101c0bf6 | 1414 | void folio_wait_bit(struct folio *folio, int bit_nr) |
62906027 | 1415 | { |
101c0bf6 | 1416 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 | 1417 | } |
101c0bf6 | 1418 | EXPORT_SYMBOL(folio_wait_bit); |
62906027 | 1419 | |
101c0bf6 | 1420 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
62906027 | 1421 | { |
101c0bf6 | 1422 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1423 | } |
101c0bf6 | 1424 | EXPORT_SYMBOL(folio_wait_bit_killable); |
cbbce822 | 1425 | |
9a1ea439 HD |
1426 | /** |
1427 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1428 | * @page: The page to wait for. | |
48054625 | 1429 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 HD |
1430 | * |
1431 | * The caller should hold a reference on @page. They expect the page to | |
1432 | * become unlocked relatively soon, but do not wish to hold up migration | |
1433 | * (for example) by holding the reference while waiting for the page to | |
1434 | * come unlocked. After this function returns, the caller should not | |
1435 | * dereference @page. | |
48054625 MWO |
1436 | * |
1437 | * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal. | |
9a1ea439 | 1438 | */ |
48054625 | 1439 | int put_and_wait_on_page_locked(struct page *page, int state) |
9a1ea439 | 1440 | { |
101c0bf6 MWO |
1441 | return folio_wait_bit_common(page_folio(page), PG_locked, state, |
1442 | DROP); | |
9a1ea439 HD |
1443 | } |
1444 | ||
385e1ca5 | 1445 | /** |
df4d4f12 MWO |
1446 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
1447 | * @folio: Folio defining the wait queue of interest | |
697f619f | 1448 | * @waiter: Waiter to add to the queue |
385e1ca5 | 1449 | * |
df4d4f12 | 1450 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
385e1ca5 | 1451 | */ |
df4d4f12 | 1452 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
385e1ca5 | 1453 | { |
df4d4f12 | 1454 | wait_queue_head_t *q = folio_waitqueue(folio); |
385e1ca5 DH |
1455 | unsigned long flags; |
1456 | ||
1457 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1458 | __add_wait_queue_entry_tail(q, waiter); |
df4d4f12 | 1459 | folio_set_waiters(folio); |
385e1ca5 DH |
1460 | spin_unlock_irqrestore(&q->lock, flags); |
1461 | } | |
df4d4f12 | 1462 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
385e1ca5 | 1463 | |
b91e1302 LT |
1464 | #ifndef clear_bit_unlock_is_negative_byte |
1465 | ||
1466 | /* | |
1467 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1468 | * | |
1469 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1470 | * test the sign bit at the same time. But if the architecture does | |
1471 | * not support that special operation, we just do this all by hand | |
1472 | * instead. | |
1473 | * | |
1474 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1475 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1476 | * in the same byte as PG_locked. |
1477 | */ | |
1478 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1479 | { | |
1480 | clear_bit_unlock(nr, mem); | |
1481 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1482 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1483 | } |
1484 | ||
1485 | #endif | |
1486 | ||
1da177e4 | 1487 | /** |
4e136428 MWO |
1488 | * folio_unlock - Unlock a locked folio. |
1489 | * @folio: The folio. | |
1490 | * | |
1491 | * Unlocks the folio and wakes up any thread sleeping on the page lock. | |
1492 | * | |
1493 | * Context: May be called from interrupt or process context. May not be | |
1494 | * called from NMI context. | |
1da177e4 | 1495 | */ |
4e136428 | 1496 | void folio_unlock(struct folio *folio) |
1da177e4 | 1497 | { |
4e136428 | 1498 | /* Bit 7 allows x86 to check the byte's sign bit */ |
b91e1302 | 1499 | BUILD_BUG_ON(PG_waiters != 7); |
4e136428 MWO |
1500 | BUILD_BUG_ON(PG_locked > 7); |
1501 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1502 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) | |
6974d7c9 | 1503 | folio_wake_bit(folio, PG_locked); |
1da177e4 | 1504 | } |
4e136428 | 1505 | EXPORT_SYMBOL(folio_unlock); |
1da177e4 | 1506 | |
73e10ded | 1507 | /** |
b47393f8 MWO |
1508 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
1509 | * @folio: The folio. | |
73e10ded | 1510 | * |
b47393f8 MWO |
1511 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
1512 | * it. The folio reference held for PG_private_2 being set is released. | |
73e10ded | 1513 | * |
b47393f8 MWO |
1514 | * This is, for example, used when a netfs folio is being written to a local |
1515 | * disk cache, thereby allowing writes to the cache for the same folio to be | |
73e10ded DH |
1516 | * serialised. |
1517 | */ | |
b47393f8 | 1518 | void folio_end_private_2(struct folio *folio) |
73e10ded | 1519 | { |
6974d7c9 MWO |
1520 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
1521 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); | |
1522 | folio_wake_bit(folio, PG_private_2); | |
1523 | folio_put(folio); | |
73e10ded | 1524 | } |
b47393f8 | 1525 | EXPORT_SYMBOL(folio_end_private_2); |
73e10ded DH |
1526 | |
1527 | /** | |
b47393f8 MWO |
1528 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
1529 | * @folio: The folio to wait on. | |
73e10ded | 1530 | * |
b47393f8 | 1531 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
73e10ded | 1532 | */ |
b47393f8 | 1533 | void folio_wait_private_2(struct folio *folio) |
73e10ded | 1534 | { |
101c0bf6 MWO |
1535 | while (folio_test_private_2(folio)) |
1536 | folio_wait_bit(folio, PG_private_2); | |
73e10ded | 1537 | } |
b47393f8 | 1538 | EXPORT_SYMBOL(folio_wait_private_2); |
73e10ded DH |
1539 | |
1540 | /** | |
b47393f8 MWO |
1541 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. |
1542 | * @folio: The folio to wait on. | |
73e10ded | 1543 | * |
b47393f8 | 1544 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a |
73e10ded DH |
1545 | * fatal signal is received by the calling task. |
1546 | * | |
1547 | * Return: | |
1548 | * - 0 if successful. | |
1549 | * - -EINTR if a fatal signal was encountered. | |
1550 | */ | |
b47393f8 | 1551 | int folio_wait_private_2_killable(struct folio *folio) |
73e10ded DH |
1552 | { |
1553 | int ret = 0; | |
1554 | ||
101c0bf6 MWO |
1555 | while (folio_test_private_2(folio)) { |
1556 | ret = folio_wait_bit_killable(folio, PG_private_2); | |
73e10ded DH |
1557 | if (ret < 0) |
1558 | break; | |
1559 | } | |
1560 | ||
1561 | return ret; | |
1562 | } | |
b47393f8 | 1563 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
73e10ded | 1564 | |
485bb99b | 1565 | /** |
4268b480 MWO |
1566 | * folio_end_writeback - End writeback against a folio. |
1567 | * @folio: The folio. | |
1da177e4 | 1568 | */ |
4268b480 | 1569 | void folio_end_writeback(struct folio *folio) |
1da177e4 | 1570 | { |
888cf2db | 1571 | /* |
4268b480 MWO |
1572 | * folio_test_clear_reclaim() could be used here but it is an |
1573 | * atomic operation and overkill in this particular case. Failing | |
1574 | * to shuffle a folio marked for immediate reclaim is too mild | |
1575 | * a gain to justify taking an atomic operation penalty at the | |
1576 | * end of every folio writeback. | |
888cf2db | 1577 | */ |
4268b480 MWO |
1578 | if (folio_test_reclaim(folio)) { |
1579 | folio_clear_reclaim(folio); | |
575ced1c | 1580 | folio_rotate_reclaimable(folio); |
888cf2db | 1581 | } |
ac6aadb2 | 1582 | |
073861ed | 1583 | /* |
4268b480 | 1584 | * Writeback does not hold a folio reference of its own, relying |
073861ed | 1585 | * on truncation to wait for the clearing of PG_writeback. |
4268b480 MWO |
1586 | * But here we must make sure that the folio is not freed and |
1587 | * reused before the folio_wake(). | |
073861ed | 1588 | */ |
4268b480 | 1589 | folio_get(folio); |
269ccca3 | 1590 | if (!__folio_end_writeback(folio)) |
ac6aadb2 MS |
1591 | BUG(); |
1592 | ||
4e857c58 | 1593 | smp_mb__after_atomic(); |
4268b480 MWO |
1594 | folio_wake(folio, PG_writeback); |
1595 | folio_put(folio); | |
1da177e4 | 1596 | } |
4268b480 | 1597 | EXPORT_SYMBOL(folio_end_writeback); |
1da177e4 | 1598 | |
57d99845 MW |
1599 | /* |
1600 | * After completing I/O on a page, call this routine to update the page | |
1601 | * flags appropriately | |
1602 | */ | |
c11f0c0b | 1603 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1604 | { |
c11f0c0b | 1605 | if (!is_write) { |
57d99845 MW |
1606 | if (!err) { |
1607 | SetPageUptodate(page); | |
1608 | } else { | |
1609 | ClearPageUptodate(page); | |
1610 | SetPageError(page); | |
1611 | } | |
1612 | unlock_page(page); | |
abf54548 | 1613 | } else { |
57d99845 | 1614 | if (err) { |
dd8416c4 MK |
1615 | struct address_space *mapping; |
1616 | ||
57d99845 | 1617 | SetPageError(page); |
dd8416c4 MK |
1618 | mapping = page_mapping(page); |
1619 | if (mapping) | |
1620 | mapping_set_error(mapping, err); | |
57d99845 MW |
1621 | } |
1622 | end_page_writeback(page); | |
1623 | } | |
1624 | } | |
1625 | EXPORT_SYMBOL_GPL(page_endio); | |
1626 | ||
485bb99b | 1627 | /** |
7c23c782 MWO |
1628 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
1629 | * @folio: The folio to lock | |
1da177e4 | 1630 | */ |
7c23c782 | 1631 | void __folio_lock(struct folio *folio) |
1da177e4 | 1632 | { |
101c0bf6 | 1633 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
9a1ea439 | 1634 | EXCLUSIVE); |
1da177e4 | 1635 | } |
7c23c782 | 1636 | EXPORT_SYMBOL(__folio_lock); |
1da177e4 | 1637 | |
af7f29d9 | 1638 | int __folio_lock_killable(struct folio *folio) |
2687a356 | 1639 | { |
101c0bf6 | 1640 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
9a1ea439 | 1641 | EXCLUSIVE); |
2687a356 | 1642 | } |
af7f29d9 | 1643 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
2687a356 | 1644 | |
ffdc8dab | 1645 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
dd3e6d50 | 1646 | { |
df4d4f12 | 1647 | struct wait_queue_head *q = folio_waitqueue(folio); |
f32b5dd7 MWO |
1648 | int ret = 0; |
1649 | ||
df4d4f12 | 1650 | wait->folio = folio; |
f32b5dd7 MWO |
1651 | wait->bit_nr = PG_locked; |
1652 | ||
1653 | spin_lock_irq(&q->lock); | |
1654 | __add_wait_queue_entry_tail(q, &wait->wait); | |
ffdc8dab MWO |
1655 | folio_set_waiters(folio); |
1656 | ret = !folio_trylock(folio); | |
f32b5dd7 MWO |
1657 | /* |
1658 | * If we were successful now, we know we're still on the | |
1659 | * waitqueue as we're still under the lock. This means it's | |
1660 | * safe to remove and return success, we know the callback | |
1661 | * isn't going to trigger. | |
1662 | */ | |
1663 | if (!ret) | |
1664 | __remove_wait_queue(q, &wait->wait); | |
1665 | else | |
1666 | ret = -EIOCBQUEUED; | |
1667 | spin_unlock_irq(&q->lock); | |
1668 | return ret; | |
dd3e6d50 JA |
1669 | } |
1670 | ||
9a95f3cf PC |
1671 | /* |
1672 | * Return values: | |
9138e47e MWO |
1673 | * true - folio is locked; mmap_lock is still held. |
1674 | * false - folio is not locked. | |
3e4e28c5 | 1675 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1676 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1677 | * which case mmap_lock is still held. |
9a95f3cf | 1678 | * |
9138e47e MWO |
1679 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
1680 | * with the folio locked and the mmap_lock unperturbed. | |
9a95f3cf | 1681 | */ |
9138e47e | 1682 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
d065bd81 ML |
1683 | unsigned int flags) |
1684 | { | |
4064b982 | 1685 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1686 | /* |
c1e8d7c6 | 1687 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1688 | * even though return 0. |
1689 | */ | |
1690 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
9138e47e | 1691 | return false; |
37b23e05 | 1692 | |
d8ed45c5 | 1693 | mmap_read_unlock(mm); |
37b23e05 | 1694 | if (flags & FAULT_FLAG_KILLABLE) |
6baa8d60 | 1695 | folio_wait_locked_killable(folio); |
37b23e05 | 1696 | else |
6baa8d60 | 1697 | folio_wait_locked(folio); |
9138e47e | 1698 | return false; |
800bca7c HL |
1699 | } |
1700 | if (flags & FAULT_FLAG_KILLABLE) { | |
9138e47e | 1701 | bool ret; |
37b23e05 | 1702 | |
af7f29d9 | 1703 | ret = __folio_lock_killable(folio); |
800bca7c HL |
1704 | if (ret) { |
1705 | mmap_read_unlock(mm); | |
9138e47e | 1706 | return false; |
800bca7c HL |
1707 | } |
1708 | } else { | |
af7f29d9 | 1709 | __folio_lock(folio); |
d065bd81 | 1710 | } |
800bca7c | 1711 | |
9138e47e | 1712 | return true; |
d065bd81 ML |
1713 | } |
1714 | ||
e7b563bb | 1715 | /** |
0d3f9296 MW |
1716 | * page_cache_next_miss() - Find the next gap in the page cache. |
1717 | * @mapping: Mapping. | |
1718 | * @index: Index. | |
1719 | * @max_scan: Maximum range to search. | |
e7b563bb | 1720 | * |
0d3f9296 MW |
1721 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1722 | * gap with the lowest index. | |
e7b563bb | 1723 | * |
0d3f9296 MW |
1724 | * This function may be called under the rcu_read_lock. However, this will |
1725 | * not atomically search a snapshot of the cache at a single point in time. | |
1726 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1727 | * created at index 10, page_cache_next_miss covering both indices may | |
1728 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1729 | * |
0d3f9296 MW |
1730 | * Return: The index of the gap if found, otherwise an index outside the |
1731 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1732 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1733 | */ |
0d3f9296 | 1734 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1735 | pgoff_t index, unsigned long max_scan) |
1736 | { | |
0d3f9296 | 1737 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1738 | |
0d3f9296 MW |
1739 | while (max_scan--) { |
1740 | void *entry = xas_next(&xas); | |
1741 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1742 | break; |
0d3f9296 | 1743 | if (xas.xa_index == 0) |
e7b563bb JW |
1744 | break; |
1745 | } | |
1746 | ||
0d3f9296 | 1747 | return xas.xa_index; |
e7b563bb | 1748 | } |
0d3f9296 | 1749 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1750 | |
1751 | /** | |
2346a560 | 1752 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1753 | * @mapping: Mapping. |
1754 | * @index: Index. | |
1755 | * @max_scan: Maximum range to search. | |
e7b563bb | 1756 | * |
0d3f9296 MW |
1757 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1758 | * gap with the highest index. | |
e7b563bb | 1759 | * |
0d3f9296 MW |
1760 | * This function may be called under the rcu_read_lock. However, this will |
1761 | * not atomically search a snapshot of the cache at a single point in time. | |
1762 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1763 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1764 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1765 | * |
0d3f9296 MW |
1766 | * Return: The index of the gap if found, otherwise an index outside the |
1767 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1768 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1769 | */ |
0d3f9296 | 1770 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1771 | pgoff_t index, unsigned long max_scan) |
1772 | { | |
0d3f9296 | 1773 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1774 | |
0d3f9296 MW |
1775 | while (max_scan--) { |
1776 | void *entry = xas_prev(&xas); | |
1777 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1778 | break; |
0d3f9296 | 1779 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1780 | break; |
1781 | } | |
1782 | ||
0d3f9296 | 1783 | return xas.xa_index; |
e7b563bb | 1784 | } |
0d3f9296 | 1785 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1786 | |
020853b6 MWO |
1787 | /* |
1788 | * Lockless page cache protocol: | |
1789 | * On the lookup side: | |
1790 | * 1. Load the folio from i_pages | |
1791 | * 2. Increment the refcount if it's not zero | |
1792 | * 3. If the folio is not found by xas_reload(), put the refcount and retry | |
1793 | * | |
1794 | * On the removal side: | |
1795 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) | |
1796 | * B. Remove the page from i_pages | |
1797 | * C. Return the page to the page allocator | |
1798 | * | |
1799 | * This means that any page may have its reference count temporarily | |
1800 | * increased by a speculative page cache (or fast GUP) lookup as it can | |
1801 | * be allocated by another user before the RCU grace period expires. | |
1802 | * Because the refcount temporarily acquired here may end up being the | |
1803 | * last refcount on the page, any page allocation must be freeable by | |
1804 | * folio_put(). | |
1805 | */ | |
1806 | ||
44835d20 | 1807 | /* |
bc5a3011 | 1808 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1809 | * @mapping: the address_space to search |
a6de4b48 | 1810 | * @index: The page cache index. |
0cd6144a | 1811 | * |
bca65eea MWO |
1812 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
1813 | * it is returned with an increased refcount. If it is a shadow entry | |
1814 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, | |
1815 | * it is returned without further action. | |
485bb99b | 1816 | * |
bca65eea | 1817 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
1da177e4 | 1818 | */ |
bca65eea | 1819 | static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1820 | { |
a6de4b48 | 1821 | XA_STATE(xas, &mapping->i_pages, index); |
bca65eea | 1822 | struct folio *folio; |
1da177e4 | 1823 | |
a60637c8 NP |
1824 | rcu_read_lock(); |
1825 | repeat: | |
4c7472c0 | 1826 | xas_reset(&xas); |
bca65eea MWO |
1827 | folio = xas_load(&xas); |
1828 | if (xas_retry(&xas, folio)) | |
4c7472c0 MW |
1829 | goto repeat; |
1830 | /* | |
1831 | * A shadow entry of a recently evicted page, or a swap entry from | |
1832 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1833 | */ | |
bca65eea | 1834 | if (!folio || xa_is_value(folio)) |
4c7472c0 | 1835 | goto out; |
83929372 | 1836 | |
bca65eea | 1837 | if (!folio_try_get_rcu(folio)) |
4c7472c0 | 1838 | goto repeat; |
83929372 | 1839 | |
bca65eea MWO |
1840 | if (unlikely(folio != xas_reload(&xas))) { |
1841 | folio_put(folio); | |
4c7472c0 | 1842 | goto repeat; |
a60637c8 | 1843 | } |
27d20fdd | 1844 | out: |
a60637c8 NP |
1845 | rcu_read_unlock(); |
1846 | ||
bca65eea | 1847 | return folio; |
1da177e4 | 1848 | } |
1da177e4 | 1849 | |
0cd6144a | 1850 | /** |
3f0c6a07 | 1851 | * __filemap_get_folio - Find and get a reference to a folio. |
2294b32e MWO |
1852 | * @mapping: The address_space to search. |
1853 | * @index: The page index. | |
3f0c6a07 MWO |
1854 | * @fgp_flags: %FGP flags modify how the folio is returned. |
1855 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1856 | * |
2294b32e | 1857 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1858 | * |
2294b32e | 1859 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1860 | * |
3f0c6a07 MWO |
1861 | * * %FGP_ACCESSED - The folio will be marked accessed. |
1862 | * * %FGP_LOCK - The folio is returned locked. | |
44835d20 | 1863 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
3f0c6a07 | 1864 | * instead of allocating a new folio to replace it. |
2294b32e | 1865 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
3f0c6a07 | 1866 | * @gfp and added to the page cache and the VM's LRU list. |
2294b32e MWO |
1867 | * The page is returned locked and with an increased refcount. |
1868 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1869 | * page is already in cache. If the page was allocated, unlock it before | |
1870 | * returning so the caller can do the same dance. | |
3f0c6a07 MWO |
1871 | * * %FGP_WRITE - The page will be written to by the caller. |
1872 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp. | |
1873 | * * %FGP_NOWAIT - Don't get blocked by page lock. | |
b27652d9 | 1874 | * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) |
1da177e4 | 1875 | * |
2294b32e MWO |
1876 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1877 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1878 | * |
2457aec6 | 1879 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1880 | * |
3f0c6a07 | 1881 | * Return: The found folio or %NULL otherwise. |
1da177e4 | 1882 | */ |
3f0c6a07 MWO |
1883 | struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, |
1884 | int fgp_flags, gfp_t gfp) | |
1da177e4 | 1885 | { |
3f0c6a07 | 1886 | struct folio *folio; |
2457aec6 | 1887 | |
1da177e4 | 1888 | repeat: |
3f0c6a07 MWO |
1889 | folio = mapping_get_entry(mapping, index); |
1890 | if (xa_is_value(folio)) { | |
44835d20 | 1891 | if (fgp_flags & FGP_ENTRY) |
3f0c6a07 MWO |
1892 | return folio; |
1893 | folio = NULL; | |
44835d20 | 1894 | } |
3f0c6a07 | 1895 | if (!folio) |
2457aec6 MG |
1896 | goto no_page; |
1897 | ||
1898 | if (fgp_flags & FGP_LOCK) { | |
1899 | if (fgp_flags & FGP_NOWAIT) { | |
3f0c6a07 MWO |
1900 | if (!folio_trylock(folio)) { |
1901 | folio_put(folio); | |
2457aec6 MG |
1902 | return NULL; |
1903 | } | |
1904 | } else { | |
3f0c6a07 | 1905 | folio_lock(folio); |
2457aec6 MG |
1906 | } |
1907 | ||
1908 | /* Has the page been truncated? */ | |
3f0c6a07 MWO |
1909 | if (unlikely(folio->mapping != mapping)) { |
1910 | folio_unlock(folio); | |
1911 | folio_put(folio); | |
2457aec6 MG |
1912 | goto repeat; |
1913 | } | |
3f0c6a07 | 1914 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
2457aec6 MG |
1915 | } |
1916 | ||
c16eb000 | 1917 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1918 | folio_mark_accessed(folio); |
b9306a79 YS |
1919 | else if (fgp_flags & FGP_WRITE) { |
1920 | /* Clear idle flag for buffer write */ | |
3f0c6a07 MWO |
1921 | if (folio_test_idle(folio)) |
1922 | folio_clear_idle(folio); | |
b9306a79 | 1923 | } |
2457aec6 | 1924 | |
b27652d9 MWO |
1925 | if (fgp_flags & FGP_STABLE) |
1926 | folio_wait_stable(folio); | |
2457aec6 | 1927 | no_page: |
3f0c6a07 | 1928 | if (!folio && (fgp_flags & FGP_CREAT)) { |
2457aec6 | 1929 | int err; |
f56753ac | 1930 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
3f0c6a07 | 1931 | gfp |= __GFP_WRITE; |
45f87de5 | 1932 | if (fgp_flags & FGP_NOFS) |
3f0c6a07 | 1933 | gfp &= ~__GFP_FS; |
2457aec6 | 1934 | |
3f0c6a07 MWO |
1935 | folio = filemap_alloc_folio(gfp, 0); |
1936 | if (!folio) | |
eb2be189 | 1937 | return NULL; |
2457aec6 | 1938 | |
a75d4c33 | 1939 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1940 | fgp_flags |= FGP_LOCK; |
1941 | ||
eb39d618 | 1942 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1943 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1944 | __folio_set_referenced(folio); |
2457aec6 | 1945 | |
3f0c6a07 | 1946 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 1947 | if (unlikely(err)) { |
3f0c6a07 MWO |
1948 | folio_put(folio); |
1949 | folio = NULL; | |
eb2be189 NP |
1950 | if (err == -EEXIST) |
1951 | goto repeat; | |
1da177e4 | 1952 | } |
a75d4c33 JB |
1953 | |
1954 | /* | |
3f0c6a07 MWO |
1955 | * filemap_add_folio locks the page, and for mmap |
1956 | * we expect an unlocked page. | |
a75d4c33 | 1957 | */ |
3f0c6a07 MWO |
1958 | if (folio && (fgp_flags & FGP_FOR_MMAP)) |
1959 | folio_unlock(folio); | |
1da177e4 | 1960 | } |
2457aec6 | 1961 | |
3f0c6a07 | 1962 | return folio; |
1da177e4 | 1963 | } |
3f0c6a07 | 1964 | EXPORT_SYMBOL(__filemap_get_folio); |
1da177e4 | 1965 | |
c7bad633 MWO |
1966 | static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max, |
1967 | xa_mark_t mark) | |
1968 | { | |
1969 | struct page *page; | |
1970 | ||
1971 | retry: | |
1972 | if (mark == XA_PRESENT) | |
1973 | page = xas_find(xas, max); | |
1974 | else | |
1975 | page = xas_find_marked(xas, max, mark); | |
1976 | ||
1977 | if (xas_retry(xas, page)) | |
1978 | goto retry; | |
1979 | /* | |
1980 | * A shadow entry of a recently evicted page, a swap | |
1981 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1982 | * without attempting to raise page count. | |
1983 | */ | |
1984 | if (!page || xa_is_value(page)) | |
1985 | return page; | |
1986 | ||
1987 | if (!page_cache_get_speculative(page)) | |
1988 | goto reset; | |
1989 | ||
1990 | /* Has the page moved or been split? */ | |
1991 | if (unlikely(page != xas_reload(xas))) { | |
1992 | put_page(page); | |
1993 | goto reset; | |
1994 | } | |
1995 | ||
1996 | return page; | |
1997 | reset: | |
1998 | xas_reset(xas); | |
1999 | goto retry; | |
2000 | } | |
2001 | ||
0cd6144a JW |
2002 | /** |
2003 | * find_get_entries - gang pagecache lookup | |
2004 | * @mapping: The address_space to search | |
2005 | * @start: The starting page cache index | |
ca122fe4 | 2006 | * @end: The final page index (inclusive). |
cf2039af | 2007 | * @pvec: Where the resulting entries are placed. |
0cd6144a JW |
2008 | * @indices: The cache indices corresponding to the entries in @entries |
2009 | * | |
cf2039af MWO |
2010 | * find_get_entries() will search for and return a batch of entries in |
2011 | * the mapping. The entries are placed in @pvec. find_get_entries() | |
2012 | * takes a reference on any actual pages it returns. | |
0cd6144a JW |
2013 | * |
2014 | * The search returns a group of mapping-contiguous page cache entries | |
2015 | * with ascending indexes. There may be holes in the indices due to | |
2016 | * not-present pages. | |
2017 | * | |
139b6a6f JW |
2018 | * Any shadow entries of evicted pages, or swap entries from |
2019 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 2020 | * |
71725ed1 HD |
2021 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
2022 | * stops at that page: the caller is likely to have a better way to handle | |
2023 | * the compound page as a whole, and then skip its extent, than repeatedly | |
2024 | * calling find_get_entries() to return all its tails. | |
2025 | * | |
a862f68a | 2026 | * Return: the number of pages and shadow entries which were found. |
0cd6144a | 2027 | */ |
ca122fe4 | 2028 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
cf2039af | 2029 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) |
0cd6144a | 2030 | { |
f280bf09 MW |
2031 | XA_STATE(xas, &mapping->i_pages, start); |
2032 | struct page *page; | |
0cd6144a | 2033 | unsigned int ret = 0; |
cf2039af | 2034 | unsigned nr_entries = PAGEVEC_SIZE; |
0cd6144a JW |
2035 | |
2036 | rcu_read_lock(); | |
ca122fe4 | 2037 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
71725ed1 HD |
2038 | /* |
2039 | * Terminate early on finding a THP, to allow the caller to | |
2040 | * handle it all at once; but continue if this is hugetlbfs. | |
2041 | */ | |
c7bad633 MWO |
2042 | if (!xa_is_value(page) && PageTransHuge(page) && |
2043 | !PageHuge(page)) { | |
71725ed1 HD |
2044 | page = find_subpage(page, xas.xa_index); |
2045 | nr_entries = ret + 1; | |
2046 | } | |
c7bad633 | 2047 | |
f280bf09 | 2048 | indices[ret] = xas.xa_index; |
cf2039af | 2049 | pvec->pages[ret] = page; |
0cd6144a JW |
2050 | if (++ret == nr_entries) |
2051 | break; | |
2052 | } | |
2053 | rcu_read_unlock(); | |
cf2039af MWO |
2054 | |
2055 | pvec->nr = ret; | |
0cd6144a JW |
2056 | return ret; |
2057 | } | |
2058 | ||
5c211ba2 MWO |
2059 | /** |
2060 | * find_lock_entries - Find a batch of pagecache entries. | |
2061 | * @mapping: The address_space to search. | |
2062 | * @start: The starting page cache index. | |
2063 | * @end: The final page index (inclusive). | |
2064 | * @pvec: Where the resulting entries are placed. | |
2065 | * @indices: The cache indices of the entries in @pvec. | |
2066 | * | |
2067 | * find_lock_entries() will return a batch of entries from @mapping. | |
2068 | * Swap, shadow and DAX entries are included. Pages are returned | |
2069 | * locked and with an incremented refcount. Pages which are locked by | |
2070 | * somebody else or under writeback are skipped. Only the head page of | |
2071 | * a THP is returned. Pages which are partially outside the range are | |
2072 | * not returned. | |
2073 | * | |
2074 | * The entries have ascending indexes. The indices may not be consecutive | |
2075 | * due to not-present entries, THP pages, pages which could not be locked | |
2076 | * or pages under writeback. | |
2077 | * | |
2078 | * Return: The number of entries which were found. | |
2079 | */ | |
2080 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
2081 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) | |
2082 | { | |
2083 | XA_STATE(xas, &mapping->i_pages, start); | |
2084 | struct page *page; | |
2085 | ||
2086 | rcu_read_lock(); | |
2087 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { | |
2088 | if (!xa_is_value(page)) { | |
2089 | if (page->index < start) | |
2090 | goto put; | |
2091 | VM_BUG_ON_PAGE(page->index != xas.xa_index, page); | |
2092 | if (page->index + thp_nr_pages(page) - 1 > end) | |
2093 | goto put; | |
2094 | if (!trylock_page(page)) | |
2095 | goto put; | |
2096 | if (page->mapping != mapping || PageWriteback(page)) | |
2097 | goto unlock; | |
2098 | VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index), | |
2099 | page); | |
2100 | } | |
2101 | indices[pvec->nr] = xas.xa_index; | |
2102 | if (!pagevec_add(pvec, page)) | |
2103 | break; | |
2104 | goto next; | |
2105 | unlock: | |
2106 | unlock_page(page); | |
2107 | put: | |
2108 | put_page(page); | |
2109 | next: | |
2d11e738 HD |
2110 | if (!xa_is_value(page) && PageTransHuge(page)) { |
2111 | unsigned int nr_pages = thp_nr_pages(page); | |
2112 | ||
2113 | /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */ | |
2114 | xas_set(&xas, page->index + nr_pages); | |
2115 | if (xas.xa_index < nr_pages) | |
2116 | break; | |
2117 | } | |
5c211ba2 MWO |
2118 | } |
2119 | rcu_read_unlock(); | |
2120 | ||
2121 | return pagevec_count(pvec); | |
2122 | } | |
2123 | ||
1da177e4 | 2124 | /** |
b947cee4 | 2125 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2126 | * @mapping: The address_space to search |
2127 | * @start: The starting page index | |
b947cee4 | 2128 | * @end: The final page index (inclusive) |
1da177e4 LT |
2129 | * @nr_pages: The maximum number of pages |
2130 | * @pages: Where the resulting pages are placed | |
2131 | * | |
b947cee4 JK |
2132 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2133 | * pages in the mapping starting at index @start and up to index @end | |
2134 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2135 | * a reference against the returned pages. | |
1da177e4 LT |
2136 | * |
2137 | * The search returns a group of mapping-contiguous pages with ascending | |
2138 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2139 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2140 | * |
a862f68a MR |
2141 | * Return: the number of pages which were found. If this number is |
2142 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2143 | * reached. |
1da177e4 | 2144 | */ |
b947cee4 JK |
2145 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2146 | pgoff_t end, unsigned int nr_pages, | |
2147 | struct page **pages) | |
1da177e4 | 2148 | { |
fd1b3cee MW |
2149 | XA_STATE(xas, &mapping->i_pages, *start); |
2150 | struct page *page; | |
0fc9d104 KK |
2151 | unsigned ret = 0; |
2152 | ||
2153 | if (unlikely(!nr_pages)) | |
2154 | return 0; | |
a60637c8 NP |
2155 | |
2156 | rcu_read_lock(); | |
c7bad633 | 2157 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee MW |
2158 | /* Skip over shadow, swap and DAX entries */ |
2159 | if (xa_is_value(page)) | |
8079b1c8 | 2160 | continue; |
a60637c8 | 2161 | |
4101196b | 2162 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 2163 | if (++ret == nr_pages) { |
5d3ee42f | 2164 | *start = xas.xa_index + 1; |
b947cee4 JK |
2165 | goto out; |
2166 | } | |
a60637c8 | 2167 | } |
5b280c0c | 2168 | |
b947cee4 JK |
2169 | /* |
2170 | * We come here when there is no page beyond @end. We take care to not | |
2171 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2172 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2173 | * already broken anyway. |
2174 | */ | |
2175 | if (end == (pgoff_t)-1) | |
2176 | *start = (pgoff_t)-1; | |
2177 | else | |
2178 | *start = end + 1; | |
2179 | out: | |
a60637c8 | 2180 | rcu_read_unlock(); |
d72dc8a2 | 2181 | |
1da177e4 LT |
2182 | return ret; |
2183 | } | |
2184 | ||
ebf43500 JA |
2185 | /** |
2186 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2187 | * @mapping: The address_space to search | |
2188 | * @index: The starting page index | |
2189 | * @nr_pages: The maximum number of pages | |
2190 | * @pages: Where the resulting pages are placed | |
2191 | * | |
2192 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2193 | * that the returned number of pages are guaranteed to be contiguous. | |
2194 | * | |
a862f68a | 2195 | * Return: the number of pages which were found. |
ebf43500 JA |
2196 | */ |
2197 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2198 | unsigned int nr_pages, struct page **pages) | |
2199 | { | |
3ece58a2 MW |
2200 | XA_STATE(xas, &mapping->i_pages, index); |
2201 | struct page *page; | |
0fc9d104 KK |
2202 | unsigned int ret = 0; |
2203 | ||
2204 | if (unlikely(!nr_pages)) | |
2205 | return 0; | |
a60637c8 NP |
2206 | |
2207 | rcu_read_lock(); | |
3ece58a2 | 2208 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2209 | if (xas_retry(&xas, page)) |
2210 | continue; | |
2211 | /* | |
2212 | * If the entry has been swapped out, we can stop looking. | |
2213 | * No current caller is looking for DAX entries. | |
2214 | */ | |
2215 | if (xa_is_value(page)) | |
8079b1c8 | 2216 | break; |
ebf43500 | 2217 | |
4101196b | 2218 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2219 | goto retry; |
83929372 | 2220 | |
4101196b | 2221 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2222 | if (unlikely(page != xas_reload(&xas))) |
2223 | goto put_page; | |
a60637c8 | 2224 | |
4101196b | 2225 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2226 | if (++ret == nr_pages) |
2227 | break; | |
3ece58a2 MW |
2228 | continue; |
2229 | put_page: | |
4101196b | 2230 | put_page(page); |
3ece58a2 MW |
2231 | retry: |
2232 | xas_reset(&xas); | |
ebf43500 | 2233 | } |
a60637c8 NP |
2234 | rcu_read_unlock(); |
2235 | return ret; | |
ebf43500 | 2236 | } |
ef71c15c | 2237 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2238 | |
485bb99b | 2239 | /** |
c49f50d1 | 2240 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2241 | * @mapping: the address_space to search |
2242 | * @index: the starting page index | |
72b045ae | 2243 | * @end: The final page index (inclusive) |
485bb99b RD |
2244 | * @tag: the tag index |
2245 | * @nr_pages: the maximum number of pages | |
2246 | * @pages: where the resulting pages are placed | |
2247 | * | |
c49f50d1 MWO |
2248 | * Like find_get_pages(), except we only return head pages which are tagged |
2249 | * with @tag. @index is updated to the index immediately after the last | |
2250 | * page we return, ready for the next iteration. | |
a862f68a MR |
2251 | * |
2252 | * Return: the number of pages which were found. | |
1da177e4 | 2253 | */ |
72b045ae | 2254 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2255 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2256 | struct page **pages) |
1da177e4 | 2257 | { |
a6906972 MW |
2258 | XA_STATE(xas, &mapping->i_pages, *index); |
2259 | struct page *page; | |
0fc9d104 KK |
2260 | unsigned ret = 0; |
2261 | ||
2262 | if (unlikely(!nr_pages)) | |
2263 | return 0; | |
a60637c8 NP |
2264 | |
2265 | rcu_read_lock(); | |
c7bad633 | 2266 | while ((page = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2267 | /* |
2268 | * Shadow entries should never be tagged, but this iteration | |
2269 | * is lockless so there is a window for page reclaim to evict | |
2270 | * a page we saw tagged. Skip over it. | |
2271 | */ | |
2272 | if (xa_is_value(page)) | |
139b6a6f | 2273 | continue; |
a60637c8 | 2274 | |
c49f50d1 | 2275 | pages[ret] = page; |
72b045ae | 2276 | if (++ret == nr_pages) { |
c49f50d1 | 2277 | *index = page->index + thp_nr_pages(page); |
72b045ae JK |
2278 | goto out; |
2279 | } | |
a60637c8 | 2280 | } |
5b280c0c | 2281 | |
72b045ae | 2282 | /* |
a6906972 | 2283 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2284 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2285 | * iteration when there is a page at index -1 but that is already |
2286 | * broken anyway. | |
72b045ae JK |
2287 | */ |
2288 | if (end == (pgoff_t)-1) | |
2289 | *index = (pgoff_t)-1; | |
2290 | else | |
2291 | *index = end + 1; | |
2292 | out: | |
a60637c8 | 2293 | rcu_read_unlock(); |
1da177e4 | 2294 | |
1da177e4 LT |
2295 | return ret; |
2296 | } | |
72b045ae | 2297 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2298 | |
76d42bd9 WF |
2299 | /* |
2300 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2301 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2302 | * | |
2303 | * ---R__________________________________________B__________ | |
2304 | * ^ reading here ^ bad block(assume 4k) | |
2305 | * | |
2306 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2307 | * => failing the whole request => read(R) => read(R+1) => | |
2308 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2309 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2310 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2311 | * | |
2312 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2313 | */ | |
0f8e2db4 | 2314 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2315 | { |
76d42bd9 | 2316 | ra->ra_pages /= 4; |
76d42bd9 WF |
2317 | } |
2318 | ||
cbd59c48 MWO |
2319 | /* |
2320 | * filemap_get_read_batch - Get a batch of pages for read | |
2321 | * | |
2322 | * Get a batch of pages which represent a contiguous range of bytes | |
2323 | * in the file. No tail pages will be returned. If @index is in the | |
2324 | * middle of a THP, the entire THP will be returned. The last page in | |
2325 | * the batch may have Readahead set or be not Uptodate so that the | |
2326 | * caller can take the appropriate action. | |
2327 | */ | |
2328 | static void filemap_get_read_batch(struct address_space *mapping, | |
2329 | pgoff_t index, pgoff_t max, struct pagevec *pvec) | |
2330 | { | |
2331 | XA_STATE(xas, &mapping->i_pages, index); | |
2332 | struct page *head; | |
2333 | ||
2334 | rcu_read_lock(); | |
2335 | for (head = xas_load(&xas); head; head = xas_next(&xas)) { | |
2336 | if (xas_retry(&xas, head)) | |
2337 | continue; | |
2338 | if (xas.xa_index > max || xa_is_value(head)) | |
2339 | break; | |
2340 | if (!page_cache_get_speculative(head)) | |
2341 | goto retry; | |
2342 | ||
2343 | /* Has the page moved or been split? */ | |
2344 | if (unlikely(head != xas_reload(&xas))) | |
2345 | goto put_page; | |
2346 | ||
2347 | if (!pagevec_add(pvec, head)) | |
2348 | break; | |
2349 | if (!PageUptodate(head)) | |
2350 | break; | |
2351 | if (PageReadahead(head)) | |
2352 | break; | |
2353 | xas.xa_index = head->index + thp_nr_pages(head) - 1; | |
2354 | xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK; | |
2355 | continue; | |
2356 | put_page: | |
2357 | put_page(head); | |
2358 | retry: | |
2359 | xas_reset(&xas); | |
2360 | } | |
2361 | rcu_read_unlock(); | |
2362 | } | |
2363 | ||
68430303 MWO |
2364 | static int filemap_read_page(struct file *file, struct address_space *mapping, |
2365 | struct page *page) | |
723ef24b | 2366 | { |
723ef24b KO |
2367 | int error; |
2368 | ||
723ef24b | 2369 | /* |
68430303 MWO |
2370 | * A previous I/O error may have been due to temporary failures, |
2371 | * eg. multipath errors. PG_error will be set again if readpage | |
2372 | * fails. | |
723ef24b KO |
2373 | */ |
2374 | ClearPageError(page); | |
2375 | /* Start the actual read. The read will unlock the page. */ | |
68430303 MWO |
2376 | error = mapping->a_ops->readpage(file, page); |
2377 | if (error) | |
2378 | return error; | |
723ef24b | 2379 | |
aa1ec2f6 | 2380 | error = wait_on_page_locked_killable(page); |
68430303 MWO |
2381 | if (error) |
2382 | return error; | |
aa1ec2f6 MWO |
2383 | if (PageUptodate(page)) |
2384 | return 0; | |
aa1ec2f6 MWO |
2385 | shrink_readahead_size_eio(&file->f_ra); |
2386 | return -EIO; | |
723ef24b KO |
2387 | } |
2388 | ||
fce70da3 MWO |
2389 | static bool filemap_range_uptodate(struct address_space *mapping, |
2390 | loff_t pos, struct iov_iter *iter, struct page *page) | |
2391 | { | |
2392 | int count; | |
2393 | ||
2394 | if (PageUptodate(page)) | |
2395 | return true; | |
2396 | /* pipes can't handle partially uptodate pages */ | |
2397 | if (iov_iter_is_pipe(iter)) | |
2398 | return false; | |
2399 | if (!mapping->a_ops->is_partially_uptodate) | |
2400 | return false; | |
2401 | if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page))) | |
2402 | return false; | |
2403 | ||
2404 | count = iter->count; | |
2405 | if (page_offset(page) > pos) { | |
2406 | count -= page_offset(page) - pos; | |
2407 | pos = 0; | |
2408 | } else { | |
2409 | pos -= page_offset(page); | |
2410 | } | |
2411 | ||
2412 | return mapping->a_ops->is_partially_uptodate(page, pos, count); | |
2413 | } | |
2414 | ||
4612aeef MWO |
2415 | static int filemap_update_page(struct kiocb *iocb, |
2416 | struct address_space *mapping, struct iov_iter *iter, | |
fce70da3 | 2417 | struct page *page) |
723ef24b | 2418 | { |
ffdc8dab | 2419 | struct folio *folio = page_folio(page); |
723ef24b KO |
2420 | int error; |
2421 | ||
730633f0 JK |
2422 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2423 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2424 | return -EAGAIN; | |
2425 | } else { | |
2426 | filemap_invalidate_lock_shared(mapping); | |
2427 | } | |
2428 | ||
ffdc8dab | 2429 | if (!folio_trylock(folio)) { |
730633f0 | 2430 | error = -EAGAIN; |
87d1d7b6 | 2431 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2432 | goto unlock_mapping; |
87d1d7b6 | 2433 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2434 | filemap_invalidate_unlock_shared(mapping); |
ffdc8dab | 2435 | put_and_wait_on_page_locked(&folio->page, TASK_KILLABLE); |
4612aeef | 2436 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2437 | } |
ffdc8dab | 2438 | error = __folio_lock_async(folio, iocb->ki_waitq); |
87d1d7b6 | 2439 | if (error) |
730633f0 | 2440 | goto unlock_mapping; |
723ef24b | 2441 | } |
723ef24b | 2442 | |
730633f0 | 2443 | error = AOP_TRUNCATED_PAGE; |
ffdc8dab | 2444 | if (!folio->mapping) |
730633f0 | 2445 | goto unlock; |
723ef24b | 2446 | |
fce70da3 | 2447 | error = 0; |
ffdc8dab | 2448 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, &folio->page)) |
fce70da3 MWO |
2449 | goto unlock; |
2450 | ||
2451 | error = -EAGAIN; | |
2452 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2453 | goto unlock; | |
2454 | ||
ffdc8dab | 2455 | error = filemap_read_page(iocb->ki_filp, mapping, &folio->page); |
730633f0 | 2456 | goto unlock_mapping; |
fce70da3 | 2457 | unlock: |
ffdc8dab | 2458 | folio_unlock(folio); |
730633f0 JK |
2459 | unlock_mapping: |
2460 | filemap_invalidate_unlock_shared(mapping); | |
2461 | if (error == AOP_TRUNCATED_PAGE) | |
ffdc8dab | 2462 | folio_put(folio); |
fce70da3 | 2463 | return error; |
723ef24b KO |
2464 | } |
2465 | ||
f253e185 MWO |
2466 | static int filemap_create_page(struct file *file, |
2467 | struct address_space *mapping, pgoff_t index, | |
2468 | struct pagevec *pvec) | |
723ef24b | 2469 | { |
723ef24b KO |
2470 | struct page *page; |
2471 | int error; | |
2472 | ||
723ef24b KO |
2473 | page = page_cache_alloc(mapping); |
2474 | if (!page) | |
f253e185 | 2475 | return -ENOMEM; |
723ef24b | 2476 | |
730633f0 JK |
2477 | /* |
2478 | * Protect against truncate / hole punch. Grabbing invalidate_lock here | |
2479 | * assures we cannot instantiate and bring uptodate new pagecache pages | |
2480 | * after evicting page cache during truncate and before actually | |
2481 | * freeing blocks. Note that we could release invalidate_lock after | |
2482 | * inserting the page into page cache as the locked page would then be | |
2483 | * enough to synchronize with hole punching. But there are code paths | |
2484 | * such as filemap_update_page() filling in partially uptodate pages or | |
2485 | * ->readpages() that need to hold invalidate_lock while mapping blocks | |
2486 | * for IO so let's hold the lock here as well to keep locking rules | |
2487 | * simple. | |
2488 | */ | |
2489 | filemap_invalidate_lock_shared(mapping); | |
723ef24b | 2490 | error = add_to_page_cache_lru(page, mapping, index, |
f253e185 MWO |
2491 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2492 | if (error == -EEXIST) | |
2493 | error = AOP_TRUNCATED_PAGE; | |
2494 | if (error) | |
2495 | goto error; | |
2496 | ||
2497 | error = filemap_read_page(file, mapping, page); | |
2498 | if (error) | |
2499 | goto error; | |
2500 | ||
730633f0 | 2501 | filemap_invalidate_unlock_shared(mapping); |
f253e185 MWO |
2502 | pagevec_add(pvec, page); |
2503 | return 0; | |
2504 | error: | |
730633f0 | 2505 | filemap_invalidate_unlock_shared(mapping); |
68430303 | 2506 | put_page(page); |
f253e185 | 2507 | return error; |
723ef24b KO |
2508 | } |
2509 | ||
5963fe03 MWO |
2510 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
2511 | struct address_space *mapping, struct page *page, | |
2512 | pgoff_t last_index) | |
2513 | { | |
2514 | if (iocb->ki_flags & IOCB_NOIO) | |
2515 | return -EAGAIN; | |
2516 | page_cache_async_readahead(mapping, &file->f_ra, file, page, | |
2517 | page->index, last_index - page->index); | |
2518 | return 0; | |
2519 | } | |
2520 | ||
3a6bae48 | 2521 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
ff993ba1 | 2522 | struct pagevec *pvec) |
06c04442 KO |
2523 | { |
2524 | struct file *filp = iocb->ki_filp; | |
2525 | struct address_space *mapping = filp->f_mapping; | |
2526 | struct file_ra_state *ra = &filp->f_ra; | |
2527 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2528 | pgoff_t last_index; |
2642fca6 | 2529 | struct page *page; |
cbd59c48 | 2530 | int err = 0; |
06c04442 | 2531 | |
cbd59c48 | 2532 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2533 | retry: |
06c04442 KO |
2534 | if (fatal_signal_pending(current)) |
2535 | return -EINTR; | |
2536 | ||
cbd59c48 | 2537 | filemap_get_read_batch(mapping, index, last_index, pvec); |
2642fca6 MWO |
2538 | if (!pagevec_count(pvec)) { |
2539 | if (iocb->ki_flags & IOCB_NOIO) | |
2540 | return -EAGAIN; | |
2541 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2542 | last_index - index); | |
2543 | filemap_get_read_batch(mapping, index, last_index, pvec); | |
2544 | } | |
f253e185 MWO |
2545 | if (!pagevec_count(pvec)) { |
2546 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) | |
2547 | return -EAGAIN; | |
2548 | err = filemap_create_page(filp, mapping, | |
2549 | iocb->ki_pos >> PAGE_SHIFT, pvec); | |
2550 | if (err == AOP_TRUNCATED_PAGE) | |
2642fca6 | 2551 | goto retry; |
f253e185 MWO |
2552 | return err; |
2553 | } | |
06c04442 | 2554 | |
2642fca6 MWO |
2555 | page = pvec->pages[pagevec_count(pvec) - 1]; |
2556 | if (PageReadahead(page)) { | |
2557 | err = filemap_readahead(iocb, filp, mapping, page, last_index); | |
2558 | if (err) | |
2559 | goto err; | |
2560 | } | |
2561 | if (!PageUptodate(page)) { | |
2562 | if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1) | |
2563 | iocb->ki_flags |= IOCB_NOWAIT; | |
2564 | err = filemap_update_page(iocb, mapping, iter, page); | |
2565 | if (err) | |
2566 | goto err; | |
06c04442 KO |
2567 | } |
2568 | ||
2642fca6 | 2569 | return 0; |
cbd59c48 | 2570 | err: |
2642fca6 MWO |
2571 | if (err < 0) |
2572 | put_page(page); | |
2573 | if (likely(--pvec->nr)) | |
ff993ba1 | 2574 | return 0; |
4612aeef | 2575 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2576 | goto retry; |
2577 | return err; | |
06c04442 KO |
2578 | } |
2579 | ||
485bb99b | 2580 | /** |
87fa0f3e CH |
2581 | * filemap_read - Read data from the page cache. |
2582 | * @iocb: The iocb to read. | |
2583 | * @iter: Destination for the data. | |
2584 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2585 | * |
87fa0f3e CH |
2586 | * Copies data from the page cache. If the data is not currently present, |
2587 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2588 | * |
87fa0f3e CH |
2589 | * Return: Total number of bytes copied, including those already read by |
2590 | * the caller. If an error happens before any bytes are copied, returns | |
2591 | * a negative error number. | |
1da177e4 | 2592 | */ |
87fa0f3e CH |
2593 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2594 | ssize_t already_read) | |
1da177e4 | 2595 | { |
47c27bc4 | 2596 | struct file *filp = iocb->ki_filp; |
06c04442 | 2597 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2598 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2599 | struct inode *inode = mapping->host; |
ff993ba1 MWO |
2600 | struct pagevec pvec; |
2601 | int i, error = 0; | |
06c04442 KO |
2602 | bool writably_mapped; |
2603 | loff_t isize, end_offset; | |
1da177e4 | 2604 | |
723ef24b | 2605 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2606 | return 0; |
3644e2d2 KO |
2607 | if (unlikely(!iov_iter_count(iter))) |
2608 | return 0; | |
2609 | ||
c2a9737f | 2610 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
cbd59c48 | 2611 | pagevec_init(&pvec); |
c2a9737f | 2612 | |
06c04442 | 2613 | do { |
1da177e4 | 2614 | cond_resched(); |
5abf186a | 2615 | |
723ef24b | 2616 | /* |
06c04442 KO |
2617 | * If we've already successfully copied some data, then we |
2618 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2619 | * an async read NOWAIT at that point. | |
723ef24b | 2620 | */ |
87fa0f3e | 2621 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2622 | iocb->ki_flags |= IOCB_NOWAIT; |
2623 | ||
ff993ba1 MWO |
2624 | error = filemap_get_pages(iocb, iter, &pvec); |
2625 | if (error < 0) | |
06c04442 | 2626 | break; |
1da177e4 | 2627 | |
06c04442 KO |
2628 | /* |
2629 | * i_size must be checked after we know the pages are Uptodate. | |
2630 | * | |
2631 | * Checking i_size after the check allows us to calculate | |
2632 | * the correct value for "nr", which means the zero-filled | |
2633 | * part of the page is not copied back to userspace (unless | |
2634 | * another truncate extends the file - this is desired though). | |
2635 | */ | |
2636 | isize = i_size_read(inode); | |
2637 | if (unlikely(iocb->ki_pos >= isize)) | |
2638 | goto put_pages; | |
06c04442 KO |
2639 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2640 | ||
06c04442 KO |
2641 | /* |
2642 | * Once we start copying data, we don't want to be touching any | |
2643 | * cachelines that might be contended: | |
2644 | */ | |
2645 | writably_mapped = mapping_writably_mapped(mapping); | |
2646 | ||
2647 | /* | |
2648 | * When a sequential read accesses a page several times, only | |
2649 | * mark it as accessed the first time. | |
2650 | */ | |
2651 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2652 | ra->prev_pos >> PAGE_SHIFT) | |
ff993ba1 | 2653 | mark_page_accessed(pvec.pages[0]); |
06c04442 | 2654 | |
ff993ba1 | 2655 | for (i = 0; i < pagevec_count(&pvec); i++) { |
cbd59c48 MWO |
2656 | struct page *page = pvec.pages[i]; |
2657 | size_t page_size = thp_size(page); | |
2658 | size_t offset = iocb->ki_pos & (page_size - 1); | |
2659 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, | |
2660 | page_size - offset); | |
2661 | size_t copied; | |
06c04442 | 2662 | |
cbd59c48 MWO |
2663 | if (end_offset < page_offset(page)) |
2664 | break; | |
2665 | if (i > 0) | |
2666 | mark_page_accessed(page); | |
06c04442 KO |
2667 | /* |
2668 | * If users can be writing to this page using arbitrary | |
2669 | * virtual addresses, take care about potential aliasing | |
2670 | * before reading the page on the kernel side. | |
2671 | */ | |
cbd59c48 MWO |
2672 | if (writably_mapped) { |
2673 | int j; | |
2674 | ||
2675 | for (j = 0; j < thp_nr_pages(page); j++) | |
2676 | flush_dcache_page(page + j); | |
2677 | } | |
06c04442 | 2678 | |
cbd59c48 | 2679 | copied = copy_page_to_iter(page, offset, bytes, iter); |
06c04442 | 2680 | |
87fa0f3e | 2681 | already_read += copied; |
06c04442 KO |
2682 | iocb->ki_pos += copied; |
2683 | ra->prev_pos = iocb->ki_pos; | |
2684 | ||
2685 | if (copied < bytes) { | |
2686 | error = -EFAULT; | |
2687 | break; | |
2688 | } | |
1da177e4 | 2689 | } |
06c04442 | 2690 | put_pages: |
ff993ba1 MWO |
2691 | for (i = 0; i < pagevec_count(&pvec); i++) |
2692 | put_page(pvec.pages[i]); | |
cbd59c48 | 2693 | pagevec_reinit(&pvec); |
06c04442 | 2694 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2695 | |
0c6aa263 | 2696 | file_accessed(filp); |
06c04442 | 2697 | |
87fa0f3e | 2698 | return already_read ? already_read : error; |
1da177e4 | 2699 | } |
87fa0f3e | 2700 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2701 | |
485bb99b | 2702 | /** |
6abd2322 | 2703 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2704 | * @iocb: kernel I/O control block |
6abd2322 | 2705 | * @iter: destination for the data read |
485bb99b | 2706 | * |
6abd2322 | 2707 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2708 | * that can use the page cache directly. |
41da51bc AG |
2709 | * |
2710 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2711 | * be returned when no data can be read without waiting for I/O requests | |
2712 | * to complete; it doesn't prevent readahead. | |
2713 | * | |
2714 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2715 | * requests shall be made for the read or for readahead. When no data | |
2716 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2717 | * triggered, a partial, possibly empty read shall be returned. | |
2718 | * | |
a862f68a MR |
2719 | * Return: |
2720 | * * number of bytes copied, even for partial reads | |
41da51bc | 2721 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2722 | */ |
2723 | ssize_t | |
ed978a81 | 2724 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2725 | { |
e7080a43 | 2726 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2727 | ssize_t retval = 0; |
e7080a43 NS |
2728 | |
2729 | if (!count) | |
826ea860 | 2730 | return 0; /* skip atime */ |
1da177e4 | 2731 | |
2ba48ce5 | 2732 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2733 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2734 | struct address_space *mapping = file->f_mapping; |
2735 | struct inode *inode = mapping->host; | |
543ade1f | 2736 | loff_t size; |
1da177e4 | 2737 | |
1da177e4 | 2738 | size = i_size_read(inode); |
6be96d3a | 2739 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2740 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2741 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2742 | return -EAGAIN; |
2743 | } else { | |
2744 | retval = filemap_write_and_wait_range(mapping, | |
2745 | iocb->ki_pos, | |
2746 | iocb->ki_pos + count - 1); | |
2747 | if (retval < 0) | |
826ea860 | 2748 | return retval; |
6be96d3a | 2749 | } |
d8d3d94b | 2750 | |
0d5b0cf2 CH |
2751 | file_accessed(file); |
2752 | ||
5ecda137 | 2753 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2754 | if (retval >= 0) { |
c64fb5c7 | 2755 | iocb->ki_pos += retval; |
5ecda137 | 2756 | count -= retval; |
9fe55eea | 2757 | } |
ab2125df PB |
2758 | if (retval != -EIOCBQUEUED) |
2759 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2760 | |
9fe55eea SW |
2761 | /* |
2762 | * Btrfs can have a short DIO read if we encounter | |
2763 | * compressed extents, so if there was an error, or if | |
2764 | * we've already read everything we wanted to, or if | |
2765 | * there was a short read because we hit EOF, go ahead | |
2766 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2767 | * the rest of the read. Buffered reads will not work for |
2768 | * DAX files, so don't bother trying. | |
9fe55eea | 2769 | */ |
5ecda137 | 2770 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2771 | IS_DAX(inode)) |
826ea860 | 2772 | return retval; |
1da177e4 LT |
2773 | } |
2774 | ||
826ea860 | 2775 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2776 | } |
ed978a81 | 2777 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2778 | |
54fa39ac MWO |
2779 | static inline loff_t page_seek_hole_data(struct xa_state *xas, |
2780 | struct address_space *mapping, struct page *page, | |
2781 | loff_t start, loff_t end, bool seek_data) | |
41139aa4 | 2782 | { |
54fa39ac MWO |
2783 | const struct address_space_operations *ops = mapping->a_ops; |
2784 | size_t offset, bsz = i_blocksize(mapping->host); | |
2785 | ||
41139aa4 | 2786 | if (xa_is_value(page) || PageUptodate(page)) |
54fa39ac MWO |
2787 | return seek_data ? start : end; |
2788 | if (!ops->is_partially_uptodate) | |
2789 | return seek_data ? end : start; | |
2790 | ||
2791 | xas_pause(xas); | |
2792 | rcu_read_unlock(); | |
2793 | lock_page(page); | |
2794 | if (unlikely(page->mapping != mapping)) | |
2795 | goto unlock; | |
2796 | ||
2797 | offset = offset_in_thp(page, start) & ~(bsz - 1); | |
2798 | ||
2799 | do { | |
2800 | if (ops->is_partially_uptodate(page, offset, bsz) == seek_data) | |
2801 | break; | |
2802 | start = (start + bsz) & ~(bsz - 1); | |
2803 | offset += bsz; | |
2804 | } while (offset < thp_size(page)); | |
2805 | unlock: | |
2806 | unlock_page(page); | |
2807 | rcu_read_lock(); | |
2808 | return start; | |
41139aa4 MWO |
2809 | } |
2810 | ||
2811 | static inline | |
2812 | unsigned int seek_page_size(struct xa_state *xas, struct page *page) | |
2813 | { | |
2814 | if (xa_is_value(page)) | |
2815 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); | |
2816 | return thp_size(page); | |
2817 | } | |
2818 | ||
2819 | /** | |
2820 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2821 | * @mapping: Address space to search. | |
2822 | * @start: First byte to consider. | |
2823 | * @end: Limit of search (exclusive). | |
2824 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2825 | * | |
2826 | * If the page cache knows which blocks contain holes and which blocks | |
2827 | * contain data, your filesystem can use this function to implement | |
2828 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2829 | * entirely memory-based such as tmpfs, and filesystems which support | |
2830 | * unwritten extents. | |
2831 | * | |
f0953a1b | 2832 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2833 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2834 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2835 | * and @end contain data. | |
2836 | */ | |
2837 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2838 | loff_t end, int whence) | |
2839 | { | |
2840 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2841 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 MWO |
2842 | bool seek_data = (whence == SEEK_DATA); |
2843 | struct page *page; | |
2844 | ||
2845 | if (end <= start) | |
2846 | return -ENXIO; | |
2847 | ||
2848 | rcu_read_lock(); | |
2849 | while ((page = find_get_entry(&xas, max, XA_PRESENT))) { | |
ed98b015 HD |
2850 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
2851 | unsigned int seek_size; | |
41139aa4 MWO |
2852 | |
2853 | if (start < pos) { | |
2854 | if (!seek_data) | |
2855 | goto unlock; | |
2856 | start = pos; | |
2857 | } | |
2858 | ||
ed98b015 HD |
2859 | seek_size = seek_page_size(&xas, page); |
2860 | pos = round_up(pos + 1, seek_size); | |
54fa39ac MWO |
2861 | start = page_seek_hole_data(&xas, mapping, page, start, pos, |
2862 | seek_data); | |
2863 | if (start < pos) | |
41139aa4 | 2864 | goto unlock; |
ed98b015 HD |
2865 | if (start >= end) |
2866 | break; | |
2867 | if (seek_size > PAGE_SIZE) | |
2868 | xas_set(&xas, pos >> PAGE_SHIFT); | |
41139aa4 MWO |
2869 | if (!xa_is_value(page)) |
2870 | put_page(page); | |
2871 | } | |
41139aa4 | 2872 | if (seek_data) |
ed98b015 | 2873 | start = -ENXIO; |
41139aa4 MWO |
2874 | unlock: |
2875 | rcu_read_unlock(); | |
ed98b015 | 2876 | if (page && !xa_is_value(page)) |
41139aa4 | 2877 | put_page(page); |
41139aa4 MWO |
2878 | if (start > end) |
2879 | return end; | |
2880 | return start; | |
2881 | } | |
2882 | ||
1da177e4 | 2883 | #ifdef CONFIG_MMU |
1da177e4 | 2884 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2885 | /* |
c1e8d7c6 | 2886 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2887 | * @vmf - the vm_fault for this fault. |
2888 | * @page - the page to lock. | |
2889 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2890 | * | |
c1e8d7c6 | 2891 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2892 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2893 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2894 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2895 | */ | |
2896 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2897 | struct file **fpin) | |
2898 | { | |
7c23c782 MWO |
2899 | struct folio *folio = page_folio(page); |
2900 | ||
2901 | if (folio_trylock(folio)) | |
6b4c9f44 JB |
2902 | return 1; |
2903 | ||
8b0f9fa2 LT |
2904 | /* |
2905 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2906 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2907 | * is supposed to work. We have way too many special cases.. |
2908 | */ | |
6b4c9f44 JB |
2909 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2910 | return 0; | |
2911 | ||
2912 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2913 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
af7f29d9 | 2914 | if (__folio_lock_killable(folio)) { |
6b4c9f44 | 2915 | /* |
c1e8d7c6 | 2916 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2917 | * but all fault_handlers only check for fatal signals |
2918 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2919 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2920 | */ |
2921 | if (*fpin == NULL) | |
d8ed45c5 | 2922 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2923 | return 0; |
2924 | } | |
2925 | } else | |
7c23c782 MWO |
2926 | __folio_lock(folio); |
2927 | ||
6b4c9f44 JB |
2928 | return 1; |
2929 | } | |
2930 | ||
ef00e08e | 2931 | /* |
6b4c9f44 JB |
2932 | * Synchronous readahead happens when we don't even find a page in the page |
2933 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2934 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2935 | * that. If we didn't pin a file then we return NULL. The file that is | |
2936 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2937 | */ |
6b4c9f44 | 2938 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2939 | { |
2a1180f1 JB |
2940 | struct file *file = vmf->vma->vm_file; |
2941 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2942 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2943 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2944 | struct file *fpin = NULL; |
e630bfac | 2945 | unsigned int mmap_miss; |
ef00e08e LT |
2946 | |
2947 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2948 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2949 | return fpin; |
275b12bf | 2950 | if (!ra->ra_pages) |
6b4c9f44 | 2951 | return fpin; |
ef00e08e | 2952 | |
2a1180f1 | 2953 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2954 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 2955 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 2956 | return fpin; |
ef00e08e LT |
2957 | } |
2958 | ||
207d04ba | 2959 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2960 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2961 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2962 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2963 | |
2964 | /* | |
2965 | * Do we miss much more than hit in this file? If so, | |
2966 | * stop bothering with read-ahead. It will only hurt. | |
2967 | */ | |
e630bfac | 2968 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2969 | return fpin; |
ef00e08e | 2970 | |
d30a1100 WF |
2971 | /* |
2972 | * mmap read-around | |
2973 | */ | |
6b4c9f44 | 2974 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2975 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2976 | ra->size = ra->ra_pages; |
2977 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2978 | ractl._index = ra->start; |
2979 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 2980 | return fpin; |
ef00e08e LT |
2981 | } |
2982 | ||
2983 | /* | |
2984 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2985 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2986 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2987 | */ |
6b4c9f44 JB |
2988 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2989 | struct page *page) | |
ef00e08e | 2990 | { |
2a1180f1 JB |
2991 | struct file *file = vmf->vma->vm_file; |
2992 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2993 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2994 | struct file *fpin = NULL; |
e630bfac | 2995 | unsigned int mmap_miss; |
2a1180f1 | 2996 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2997 | |
2998 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 2999 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 3000 | return fpin; |
e630bfac KS |
3001 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3002 | if (mmap_miss) | |
3003 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
3004 | if (PageReadahead(page)) { |
3005 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
3006 | page_cache_async_readahead(mapping, ra, file, |
3007 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
3008 | } |
3009 | return fpin; | |
ef00e08e LT |
3010 | } |
3011 | ||
485bb99b | 3012 | /** |
54cb8821 | 3013 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 3014 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 3015 | * |
54cb8821 | 3016 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
3017 | * mapped memory region to read in file data during a page fault. |
3018 | * | |
3019 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
3020 | * it in the page cache, and handles the special cases reasonably without | |
3021 | * having a lot of duplicated code. | |
9a95f3cf | 3022 | * |
c1e8d7c6 | 3023 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 3024 | * |
c1e8d7c6 | 3025 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 3026 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 3027 | * |
c1e8d7c6 | 3028 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
3029 | * has not been released. |
3030 | * | |
3031 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
3032 | * |
3033 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 3034 | */ |
2bcd6454 | 3035 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
3036 | { |
3037 | int error; | |
11bac800 | 3038 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 3039 | struct file *fpin = NULL; |
1da177e4 | 3040 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3041 | struct inode *inode = mapping->host; |
ef00e08e | 3042 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 3043 | pgoff_t max_off; |
1da177e4 | 3044 | struct page *page; |
2bcd6454 | 3045 | vm_fault_t ret = 0; |
730633f0 | 3046 | bool mapping_locked = false; |
1da177e4 | 3047 | |
9ab2594f MW |
3048 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3049 | if (unlikely(offset >= max_off)) | |
5307cc1a | 3050 | return VM_FAULT_SIGBUS; |
1da177e4 | 3051 | |
1da177e4 | 3052 | /* |
49426420 | 3053 | * Do we have something in the page cache already? |
1da177e4 | 3054 | */ |
ef00e08e | 3055 | page = find_get_page(mapping, offset); |
730633f0 | 3056 | if (likely(page)) { |
1da177e4 | 3057 | /* |
730633f0 JK |
3058 | * We found the page, so try async readahead before waiting for |
3059 | * the lock. | |
1da177e4 | 3060 | */ |
730633f0 JK |
3061 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
3062 | fpin = do_async_mmap_readahead(vmf, page); | |
3063 | if (unlikely(!PageUptodate(page))) { | |
3064 | filemap_invalidate_lock_shared(mapping); | |
3065 | mapping_locked = true; | |
3066 | } | |
3067 | } else { | |
ef00e08e | 3068 | /* No page in the page cache at all */ |
ef00e08e | 3069 | count_vm_event(PGMAJFAULT); |
2262185c | 3070 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3071 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3072 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3073 | retry_find: |
730633f0 JK |
3074 | /* |
3075 | * See comment in filemap_create_page() why we need | |
3076 | * invalidate_lock | |
3077 | */ | |
3078 | if (!mapping_locked) { | |
3079 | filemap_invalidate_lock_shared(mapping); | |
3080 | mapping_locked = true; | |
3081 | } | |
a75d4c33 JB |
3082 | page = pagecache_get_page(mapping, offset, |
3083 | FGP_CREAT|FGP_FOR_MMAP, | |
3084 | vmf->gfp_mask); | |
6b4c9f44 JB |
3085 | if (!page) { |
3086 | if (fpin) | |
3087 | goto out_retry; | |
730633f0 | 3088 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3089 | return VM_FAULT_OOM; |
6b4c9f44 | 3090 | } |
1da177e4 LT |
3091 | } |
3092 | ||
6b4c9f44 JB |
3093 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
3094 | goto out_retry; | |
b522c94d ML |
3095 | |
3096 | /* Did it get truncated? */ | |
585e5a7b | 3097 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
3098 | unlock_page(page); |
3099 | put_page(page); | |
3100 | goto retry_find; | |
3101 | } | |
520e5ba4 | 3102 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 3103 | |
1da177e4 | 3104 | /* |
d00806b1 NP |
3105 | * We have a locked page in the page cache, now we need to check |
3106 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3107 | */ |
730633f0 JK |
3108 | if (unlikely(!PageUptodate(page))) { |
3109 | /* | |
3110 | * The page was in cache and uptodate and now it is not. | |
3111 | * Strange but possible since we didn't hold the page lock all | |
3112 | * the time. Let's drop everything get the invalidate lock and | |
3113 | * try again. | |
3114 | */ | |
3115 | if (!mapping_locked) { | |
3116 | unlock_page(page); | |
3117 | put_page(page); | |
3118 | goto retry_find; | |
3119 | } | |
1da177e4 | 3120 | goto page_not_uptodate; |
730633f0 | 3121 | } |
1da177e4 | 3122 | |
6b4c9f44 | 3123 | /* |
c1e8d7c6 | 3124 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3125 | * time to return to the upper layer and have it re-find the vma and |
3126 | * redo the fault. | |
3127 | */ | |
3128 | if (fpin) { | |
3129 | unlock_page(page); | |
3130 | goto out_retry; | |
3131 | } | |
730633f0 JK |
3132 | if (mapping_locked) |
3133 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3134 | |
ef00e08e LT |
3135 | /* |
3136 | * Found the page and have a reference on it. | |
3137 | * We must recheck i_size under page lock. | |
3138 | */ | |
9ab2594f MW |
3139 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3140 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 3141 | unlock_page(page); |
09cbfeaf | 3142 | put_page(page); |
5307cc1a | 3143 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3144 | } |
3145 | ||
d0217ac0 | 3146 | vmf->page = page; |
83c54070 | 3147 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3148 | |
1da177e4 | 3149 | page_not_uptodate: |
1da177e4 LT |
3150 | /* |
3151 | * Umm, take care of errors if the page isn't up-to-date. | |
3152 | * Try to re-read it _once_. We do this synchronously, | |
3153 | * because there really aren't any performance issues here | |
3154 | * and we need to check for errors. | |
3155 | */ | |
6b4c9f44 | 3156 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
d31fa86a | 3157 | error = filemap_read_page(file, mapping, page); |
6b4c9f44 JB |
3158 | if (fpin) |
3159 | goto out_retry; | |
09cbfeaf | 3160 | put_page(page); |
d00806b1 NP |
3161 | |
3162 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3163 | goto retry_find; |
730633f0 | 3164 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3165 | |
d0217ac0 | 3166 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3167 | |
3168 | out_retry: | |
3169 | /* | |
c1e8d7c6 | 3170 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3171 | * re-find the vma and come back and find our hopefully still populated |
3172 | * page. | |
3173 | */ | |
3174 | if (page) | |
3175 | put_page(page); | |
730633f0 JK |
3176 | if (mapping_locked) |
3177 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3178 | if (fpin) |
3179 | fput(fpin); | |
3180 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3181 | } |
3182 | EXPORT_SYMBOL(filemap_fault); | |
3183 | ||
f9ce0be7 | 3184 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3185 | { |
f9ce0be7 KS |
3186 | struct mm_struct *mm = vmf->vma->vm_mm; |
3187 | ||
3188 | /* Huge page is mapped? No need to proceed. */ | |
3189 | if (pmd_trans_huge(*vmf->pmd)) { | |
3190 | unlock_page(page); | |
3191 | put_page(page); | |
3192 | return true; | |
3193 | } | |
3194 | ||
3195 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
3196 | vm_fault_t ret = do_set_pmd(vmf, page); | |
3197 | if (!ret) { | |
3198 | /* The page is mapped successfully, reference consumed. */ | |
3199 | unlock_page(page); | |
3200 | return true; | |
3201 | } | |
3202 | } | |
3203 | ||
3204 | if (pmd_none(*vmf->pmd)) { | |
3205 | vmf->ptl = pmd_lock(mm, vmf->pmd); | |
3206 | if (likely(pmd_none(*vmf->pmd))) { | |
3207 | mm_inc_nr_ptes(mm); | |
3208 | pmd_populate(mm, vmf->pmd, vmf->prealloc_pte); | |
3209 | vmf->prealloc_pte = NULL; | |
3210 | } | |
3211 | spin_unlock(vmf->ptl); | |
3212 | } | |
3213 | ||
3214 | /* See comment in handle_pte_fault() */ | |
3215 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3216 | unlock_page(page); | |
3217 | put_page(page); | |
3218 | return true; | |
3219 | } | |
3220 | ||
3221 | return false; | |
3222 | } | |
3223 | ||
3224 | static struct page *next_uptodate_page(struct page *page, | |
3225 | struct address_space *mapping, | |
3226 | struct xa_state *xas, pgoff_t end_pgoff) | |
3227 | { | |
3228 | unsigned long max_idx; | |
3229 | ||
3230 | do { | |
3231 | if (!page) | |
3232 | return NULL; | |
3233 | if (xas_retry(xas, page)) | |
3234 | continue; | |
3235 | if (xa_is_value(page)) | |
3236 | continue; | |
3237 | if (PageLocked(page)) | |
3238 | continue; | |
3239 | if (!page_cache_get_speculative(page)) | |
3240 | continue; | |
3241 | /* Has the page moved or been split? */ | |
3242 | if (unlikely(page != xas_reload(xas))) | |
3243 | goto skip; | |
3244 | if (!PageUptodate(page) || PageReadahead(page)) | |
3245 | goto skip; | |
3246 | if (PageHWPoison(page)) | |
3247 | goto skip; | |
3248 | if (!trylock_page(page)) | |
3249 | goto skip; | |
3250 | if (page->mapping != mapping) | |
3251 | goto unlock; | |
3252 | if (!PageUptodate(page)) | |
3253 | goto unlock; | |
3254 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3255 | if (xas->xa_index >= max_idx) | |
3256 | goto unlock; | |
3257 | return page; | |
3258 | unlock: | |
3259 | unlock_page(page); | |
3260 | skip: | |
3261 | put_page(page); | |
3262 | } while ((page = xas_next_entry(xas, end_pgoff)) != NULL); | |
3263 | ||
3264 | return NULL; | |
3265 | } | |
3266 | ||
3267 | static inline struct page *first_map_page(struct address_space *mapping, | |
3268 | struct xa_state *xas, | |
3269 | pgoff_t end_pgoff) | |
3270 | { | |
3271 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3272 | mapping, xas, end_pgoff); | |
3273 | } | |
3274 | ||
3275 | static inline struct page *next_map_page(struct address_space *mapping, | |
3276 | struct xa_state *xas, | |
3277 | pgoff_t end_pgoff) | |
3278 | { | |
3279 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3280 | mapping, xas, end_pgoff); | |
3281 | } | |
3282 | ||
3283 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3284 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3285 | { | |
3286 | struct vm_area_struct *vma = vmf->vma; | |
3287 | struct file *file = vma->vm_file; | |
f1820361 | 3288 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3289 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3290 | unsigned long addr; |
070e807c | 3291 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
27a83a60 | 3292 | struct page *head, *page; |
e630bfac | 3293 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3294 | vm_fault_t ret = 0; |
f1820361 KS |
3295 | |
3296 | rcu_read_lock(); | |
f9ce0be7 KS |
3297 | head = first_map_page(mapping, &xas, end_pgoff); |
3298 | if (!head) | |
3299 | goto out; | |
f1820361 | 3300 | |
f9ce0be7 KS |
3301 | if (filemap_map_pmd(vmf, head)) { |
3302 | ret = VM_FAULT_NOPAGE; | |
3303 | goto out; | |
3304 | } | |
f1820361 | 3305 | |
9d3af4b4 WD |
3306 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3307 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3308 | do { |
27a83a60 | 3309 | page = find_subpage(head, xas.xa_index); |
f9ce0be7 | 3310 | if (PageHWPoison(page)) |
f1820361 KS |
3311 | goto unlock; |
3312 | ||
e630bfac KS |
3313 | if (mmap_miss > 0) |
3314 | mmap_miss--; | |
7267ec00 | 3315 | |
9d3af4b4 | 3316 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3317 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3318 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3319 | |
3320 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3321 | goto unlock; |
f9ce0be7 | 3322 | |
46bdb427 | 3323 | /* We're about to handle the fault */ |
9d3af4b4 | 3324 | if (vmf->address == addr) |
46bdb427 | 3325 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3326 | |
9d3af4b4 | 3327 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3328 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3329 | update_mmu_cache(vma, addr, vmf->pte); |
27a83a60 | 3330 | unlock_page(head); |
f9ce0be7 | 3331 | continue; |
f1820361 | 3332 | unlock: |
27a83a60 | 3333 | unlock_page(head); |
27a83a60 | 3334 | put_page(head); |
f9ce0be7 KS |
3335 | } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
3336 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3337 | out: | |
f1820361 | 3338 | rcu_read_unlock(); |
e630bfac | 3339 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3340 | return ret; |
f1820361 KS |
3341 | } |
3342 | EXPORT_SYMBOL(filemap_map_pages); | |
3343 | ||
2bcd6454 | 3344 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3345 | { |
5df1a672 | 3346 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
4fcf1c62 | 3347 | struct page *page = vmf->page; |
2bcd6454 | 3348 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3349 | |
5df1a672 | 3350 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3351 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 | 3352 | lock_page(page); |
5df1a672 | 3353 | if (page->mapping != mapping) { |
4fcf1c62 JK |
3354 | unlock_page(page); |
3355 | ret = VM_FAULT_NOPAGE; | |
3356 | goto out; | |
3357 | } | |
14da9200 JK |
3358 | /* |
3359 | * We mark the page dirty already here so that when freeze is in | |
3360 | * progress, we are guaranteed that writeback during freezing will | |
3361 | * see the dirty page and writeprotect it again. | |
3362 | */ | |
3363 | set_page_dirty(page); | |
1d1d1a76 | 3364 | wait_for_stable_page(page); |
4fcf1c62 | 3365 | out: |
5df1a672 | 3366 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3367 | return ret; |
3368 | } | |
4fcf1c62 | 3369 | |
f0f37e2f | 3370 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3371 | .fault = filemap_fault, |
f1820361 | 3372 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3373 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3374 | }; |
3375 | ||
3376 | /* This is used for a general mmap of a disk file */ | |
3377 | ||
68d68ff6 | 3378 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3379 | { |
3380 | struct address_space *mapping = file->f_mapping; | |
3381 | ||
3382 | if (!mapping->a_ops->readpage) | |
3383 | return -ENOEXEC; | |
3384 | file_accessed(file); | |
3385 | vma->vm_ops = &generic_file_vm_ops; | |
3386 | return 0; | |
3387 | } | |
1da177e4 LT |
3388 | |
3389 | /* | |
3390 | * This is for filesystems which do not implement ->writepage. | |
3391 | */ | |
3392 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3393 | { | |
3394 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3395 | return -EINVAL; | |
3396 | return generic_file_mmap(file, vma); | |
3397 | } | |
3398 | #else | |
4b96a37d | 3399 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3400 | { |
4b96a37d | 3401 | return VM_FAULT_SIGBUS; |
45397228 | 3402 | } |
68d68ff6 | 3403 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3404 | { |
3405 | return -ENOSYS; | |
3406 | } | |
68d68ff6 | 3407 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3408 | { |
3409 | return -ENOSYS; | |
3410 | } | |
3411 | #endif /* CONFIG_MMU */ | |
3412 | ||
45397228 | 3413 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3414 | EXPORT_SYMBOL(generic_file_mmap); |
3415 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3416 | ||
67f9fd91 SL |
3417 | static struct page *wait_on_page_read(struct page *page) |
3418 | { | |
3419 | if (!IS_ERR(page)) { | |
3420 | wait_on_page_locked(page); | |
3421 | if (!PageUptodate(page)) { | |
09cbfeaf | 3422 | put_page(page); |
67f9fd91 SL |
3423 | page = ERR_PTR(-EIO); |
3424 | } | |
3425 | } | |
3426 | return page; | |
3427 | } | |
3428 | ||
32b63529 | 3429 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 3430 | pgoff_t index, |
5e5358e7 | 3431 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3432 | void *data, |
3433 | gfp_t gfp) | |
1da177e4 | 3434 | { |
eb2be189 | 3435 | struct page *page; |
1da177e4 LT |
3436 | int err; |
3437 | repeat: | |
3438 | page = find_get_page(mapping, index); | |
3439 | if (!page) { | |
453f85d4 | 3440 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
3441 | if (!page) |
3442 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 3443 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 3444 | if (unlikely(err)) { |
09cbfeaf | 3445 | put_page(page); |
eb2be189 NP |
3446 | if (err == -EEXIST) |
3447 | goto repeat; | |
22ecdb4f | 3448 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3449 | return ERR_PTR(err); |
3450 | } | |
32b63529 MG |
3451 | |
3452 | filler: | |
6c45b454 CH |
3453 | if (filler) |
3454 | err = filler(data, page); | |
3455 | else | |
3456 | err = mapping->a_ops->readpage(data, page); | |
3457 | ||
1da177e4 | 3458 | if (err < 0) { |
09cbfeaf | 3459 | put_page(page); |
32b63529 | 3460 | return ERR_PTR(err); |
1da177e4 | 3461 | } |
1da177e4 | 3462 | |
32b63529 MG |
3463 | page = wait_on_page_read(page); |
3464 | if (IS_ERR(page)) | |
3465 | return page; | |
3466 | goto out; | |
3467 | } | |
1da177e4 LT |
3468 | if (PageUptodate(page)) |
3469 | goto out; | |
3470 | ||
ebded027 | 3471 | /* |
0e9aa675 | 3472 | * Page is not up to date and may be locked due to one of the following |
ebded027 MG |
3473 | * case a: Page is being filled and the page lock is held |
3474 | * case b: Read/write error clearing the page uptodate status | |
3475 | * case c: Truncation in progress (page locked) | |
3476 | * case d: Reclaim in progress | |
3477 | * | |
3478 | * Case a, the page will be up to date when the page is unlocked. | |
3479 | * There is no need to serialise on the page lock here as the page | |
3480 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 3481 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
3482 | * it's a race vs truncate race. |
3483 | * Case b, the page will not be up to date | |
3484 | * Case c, the page may be truncated but in itself, the data may still | |
3485 | * be valid after IO completes as it's a read vs truncate race. The | |
3486 | * operation must restart if the page is not uptodate on unlock but | |
3487 | * otherwise serialising on page lock to stabilise the mapping gives | |
3488 | * no additional guarantees to the caller as the page lock is | |
3489 | * released before return. | |
3490 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3491 | * will be a race with remove_mapping that determines if the mapping | |
3492 | * is valid on unlock but otherwise the data is valid and there is | |
3493 | * no need to serialise with page lock. | |
3494 | * | |
3495 | * As the page lock gives no additional guarantee, we optimistically | |
3496 | * wait on the page to be unlocked and check if it's up to date and | |
3497 | * use the page if it is. Otherwise, the page lock is required to | |
3498 | * distinguish between the different cases. The motivation is that we | |
3499 | * avoid spurious serialisations and wakeups when multiple processes | |
3500 | * wait on the same page for IO to complete. | |
3501 | */ | |
3502 | wait_on_page_locked(page); | |
3503 | if (PageUptodate(page)) | |
3504 | goto out; | |
3505 | ||
3506 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3507 | lock_page(page); |
ebded027 MG |
3508 | |
3509 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3510 | if (!page->mapping) { |
3511 | unlock_page(page); | |
09cbfeaf | 3512 | put_page(page); |
32b63529 | 3513 | goto repeat; |
1da177e4 | 3514 | } |
ebded027 MG |
3515 | |
3516 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3517 | if (PageUptodate(page)) { |
3518 | unlock_page(page); | |
3519 | goto out; | |
3520 | } | |
faffdfa0 XT |
3521 | |
3522 | /* | |
3523 | * A previous I/O error may have been due to temporary | |
3524 | * failures. | |
3525 | * Clear page error before actual read, PG_error will be | |
3526 | * set again if read page fails. | |
3527 | */ | |
3528 | ClearPageError(page); | |
32b63529 MG |
3529 | goto filler; |
3530 | ||
c855ff37 | 3531 | out: |
6fe6900e NP |
3532 | mark_page_accessed(page); |
3533 | return page; | |
3534 | } | |
0531b2aa LT |
3535 | |
3536 | /** | |
67f9fd91 | 3537 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3538 | * @mapping: the page's address_space |
3539 | * @index: the page index | |
3540 | * @filler: function to perform the read | |
5e5358e7 | 3541 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3542 | * |
0531b2aa | 3543 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3544 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3545 | * |
3546 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3547 | * |
730633f0 JK |
3548 | * The function expects mapping->invalidate_lock to be already held. |
3549 | * | |
a862f68a | 3550 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa | 3551 | */ |
67f9fd91 | 3552 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3553 | pgoff_t index, |
5e5358e7 | 3554 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3555 | void *data) |
3556 | { | |
d322a8e5 CH |
3557 | return do_read_cache_page(mapping, index, filler, data, |
3558 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3559 | } |
67f9fd91 | 3560 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3561 | |
3562 | /** | |
3563 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3564 | * @mapping: the page's address_space | |
3565 | * @index: the page index | |
3566 | * @gfp: the page allocator flags to use if allocating | |
3567 | * | |
3568 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3569 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3570 | * |
3571 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3572 | * |
730633f0 JK |
3573 | * The function expects mapping->invalidate_lock to be already held. |
3574 | * | |
a862f68a | 3575 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3576 | */ |
3577 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3578 | pgoff_t index, | |
3579 | gfp_t gfp) | |
3580 | { | |
6c45b454 | 3581 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3582 | } |
3583 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3584 | ||
afddba49 NP |
3585 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3586 | loff_t pos, unsigned len, unsigned flags, | |
3587 | struct page **pagep, void **fsdata) | |
3588 | { | |
3589 | const struct address_space_operations *aops = mapping->a_ops; | |
3590 | ||
4e02ed4b | 3591 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3592 | pagep, fsdata); |
afddba49 NP |
3593 | } |
3594 | EXPORT_SYMBOL(pagecache_write_begin); | |
3595 | ||
3596 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3597 | loff_t pos, unsigned len, unsigned copied, | |
3598 | struct page *page, void *fsdata) | |
3599 | { | |
3600 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3601 | |
4e02ed4b | 3602 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3603 | } |
3604 | EXPORT_SYMBOL(pagecache_write_end); | |
3605 | ||
a92853b6 KK |
3606 | /* |
3607 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3608 | */ | |
3609 | void dio_warn_stale_pagecache(struct file *filp) | |
3610 | { | |
3611 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3612 | char pathname[128]; | |
a92853b6 KK |
3613 | char *path; |
3614 | ||
5df1a672 | 3615 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3616 | if (__ratelimit(&_rs)) { |
3617 | path = file_path(filp, pathname, sizeof(pathname)); | |
3618 | if (IS_ERR(path)) | |
3619 | path = "(unknown)"; | |
3620 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3621 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3622 | current->comm); | |
3623 | } | |
3624 | } | |
3625 | ||
1da177e4 | 3626 | ssize_t |
1af5bb49 | 3627 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3628 | { |
3629 | struct file *file = iocb->ki_filp; | |
3630 | struct address_space *mapping = file->f_mapping; | |
3631 | struct inode *inode = mapping->host; | |
1af5bb49 | 3632 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3633 | ssize_t written; |
a969e903 CH |
3634 | size_t write_len; |
3635 | pgoff_t end; | |
1da177e4 | 3636 | |
0c949334 | 3637 | write_len = iov_iter_count(from); |
09cbfeaf | 3638 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3639 | |
6be96d3a GR |
3640 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3641 | /* If there are pages to writeback, return */ | |
5df1a672 | 3642 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3643 | pos + write_len - 1)) |
6be96d3a GR |
3644 | return -EAGAIN; |
3645 | } else { | |
3646 | written = filemap_write_and_wait_range(mapping, pos, | |
3647 | pos + write_len - 1); | |
3648 | if (written) | |
3649 | goto out; | |
3650 | } | |
a969e903 CH |
3651 | |
3652 | /* | |
3653 | * After a write we want buffered reads to be sure to go to disk to get | |
3654 | * the new data. We invalidate clean cached page from the region we're | |
3655 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3656 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3657 | */ |
55635ba7 | 3658 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3659 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3660 | /* |
3661 | * If a page can not be invalidated, return 0 to fall back | |
3662 | * to buffered write. | |
3663 | */ | |
3664 | if (written) { | |
3665 | if (written == -EBUSY) | |
3666 | return 0; | |
3667 | goto out; | |
a969e903 CH |
3668 | } |
3669 | ||
639a93a5 | 3670 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3671 | |
3672 | /* | |
3673 | * Finally, try again to invalidate clean pages which might have been | |
3674 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3675 | * if the source of the write was an mmap'ed region of the file | |
3676 | * we're writing. Either one is a pretty crazy thing to do, | |
3677 | * so we don't support it 100%. If this invalidation | |
3678 | * fails, tough, the write still worked... | |
332391a9 LC |
3679 | * |
3680 | * Most of the time we do not need this since dio_complete() will do | |
3681 | * the invalidation for us. However there are some file systems that | |
3682 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3683 | * them by removing it completely. |
3684 | * | |
9266a140 KK |
3685 | * Noticeable example is a blkdev_direct_IO(). |
3686 | * | |
80c1fe90 | 3687 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3688 | */ |
9266a140 KK |
3689 | if (written > 0 && mapping->nrpages && |
3690 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3691 | dio_warn_stale_pagecache(file); | |
a969e903 | 3692 | |
1da177e4 | 3693 | if (written > 0) { |
0116651c | 3694 | pos += written; |
639a93a5 | 3695 | write_len -= written; |
0116651c NK |
3696 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3697 | i_size_write(inode, pos); | |
1da177e4 LT |
3698 | mark_inode_dirty(inode); |
3699 | } | |
5cb6c6c7 | 3700 | iocb->ki_pos = pos; |
1da177e4 | 3701 | } |
ab2125df PB |
3702 | if (written != -EIOCBQUEUED) |
3703 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3704 | out: |
1da177e4 LT |
3705 | return written; |
3706 | } | |
3707 | EXPORT_SYMBOL(generic_file_direct_write); | |
3708 | ||
3b93f911 | 3709 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3710 | struct iov_iter *i, loff_t pos) |
3711 | { | |
3712 | struct address_space *mapping = file->f_mapping; | |
3713 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3714 | long status = 0; | |
3715 | ssize_t written = 0; | |
674b892e NP |
3716 | unsigned int flags = 0; |
3717 | ||
afddba49 NP |
3718 | do { |
3719 | struct page *page; | |
afddba49 NP |
3720 | unsigned long offset; /* Offset into pagecache page */ |
3721 | unsigned long bytes; /* Bytes to write to page */ | |
3722 | size_t copied; /* Bytes copied from user */ | |
3723 | void *fsdata; | |
3724 | ||
09cbfeaf KS |
3725 | offset = (pos & (PAGE_SIZE - 1)); |
3726 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3727 | iov_iter_count(i)); |
3728 | ||
3729 | again: | |
00a3d660 LT |
3730 | /* |
3731 | * Bring in the user page that we will copy from _first_. | |
3732 | * Otherwise there's a nasty deadlock on copying from the | |
3733 | * same page as we're writing to, without it being marked | |
3734 | * up-to-date. | |
00a3d660 | 3735 | */ |
a6294593 | 3736 | if (unlikely(fault_in_iov_iter_readable(i, bytes))) { |
00a3d660 LT |
3737 | status = -EFAULT; |
3738 | break; | |
3739 | } | |
3740 | ||
296291cd JK |
3741 | if (fatal_signal_pending(current)) { |
3742 | status = -EINTR; | |
3743 | break; | |
3744 | } | |
3745 | ||
674b892e | 3746 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3747 | &page, &fsdata); |
2457aec6 | 3748 | if (unlikely(status < 0)) |
afddba49 NP |
3749 | break; |
3750 | ||
931e80e4 | 3751 | if (mapping_writably_mapped(mapping)) |
3752 | flush_dcache_page(page); | |
00a3d660 | 3753 | |
f0b65f39 | 3754 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3755 | flush_dcache_page(page); |
3756 | ||
3757 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3758 | page, fsdata); | |
f0b65f39 AV |
3759 | if (unlikely(status != copied)) { |
3760 | iov_iter_revert(i, copied - max(status, 0L)); | |
3761 | if (unlikely(status < 0)) | |
3762 | break; | |
3763 | } | |
afddba49 NP |
3764 | cond_resched(); |
3765 | ||
bc1bb416 | 3766 | if (unlikely(status == 0)) { |
afddba49 | 3767 | /* |
bc1bb416 AV |
3768 | * A short copy made ->write_end() reject the |
3769 | * thing entirely. Might be memory poisoning | |
3770 | * halfway through, might be a race with munmap, | |
3771 | * might be severe memory pressure. | |
afddba49 | 3772 | */ |
bc1bb416 AV |
3773 | if (copied) |
3774 | bytes = copied; | |
afddba49 NP |
3775 | goto again; |
3776 | } | |
f0b65f39 AV |
3777 | pos += status; |
3778 | written += status; | |
afddba49 NP |
3779 | |
3780 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3781 | } while (iov_iter_count(i)); |
3782 | ||
3783 | return written ? written : status; | |
3784 | } | |
3b93f911 | 3785 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3786 | |
e4dd9de3 | 3787 | /** |
8174202b | 3788 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3789 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3790 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3791 | * |
3792 | * This function does all the work needed for actually writing data to a | |
3793 | * file. It does all basic checks, removes SUID from the file, updates | |
3794 | * modification times and calls proper subroutines depending on whether we | |
3795 | * do direct IO or a standard buffered write. | |
3796 | * | |
9608703e | 3797 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3798 | * object which does not need locking at all. |
3799 | * | |
3800 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3801 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3802 | * avoid syncing under i_rwsem. |
a862f68a MR |
3803 | * |
3804 | * Return: | |
3805 | * * number of bytes written, even for truncated writes | |
3806 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3807 | */ |
8174202b | 3808 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3809 | { |
3810 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3811 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3812 | struct inode *inode = mapping->host; |
3b93f911 | 3813 | ssize_t written = 0; |
1da177e4 | 3814 | ssize_t err; |
3b93f911 | 3815 | ssize_t status; |
1da177e4 | 3816 | |
1da177e4 | 3817 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3818 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3819 | err = file_remove_privs(file); |
1da177e4 LT |
3820 | if (err) |
3821 | goto out; | |
3822 | ||
c3b2da31 JB |
3823 | err = file_update_time(file); |
3824 | if (err) | |
3825 | goto out; | |
1da177e4 | 3826 | |
2ba48ce5 | 3827 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3828 | loff_t pos, endbyte; |
fb5527e6 | 3829 | |
1af5bb49 | 3830 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3831 | /* |
fbbbad4b MW |
3832 | * If the write stopped short of completing, fall back to |
3833 | * buffered writes. Some filesystems do this for writes to | |
3834 | * holes, for example. For DAX files, a buffered write will | |
3835 | * not succeed (even if it did, DAX does not handle dirty | |
3836 | * page-cache pages correctly). | |
1da177e4 | 3837 | */ |
0b8def9d | 3838 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3839 | goto out; |
3840 | ||
0b8def9d | 3841 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3842 | /* |
3b93f911 | 3843 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3844 | * then we want to return the number of bytes which were |
3845 | * direct-written, or the error code if that was zero. Note | |
3846 | * that this differs from normal direct-io semantics, which | |
3847 | * will return -EFOO even if some bytes were written. | |
3848 | */ | |
60bb4529 | 3849 | if (unlikely(status < 0)) { |
3b93f911 | 3850 | err = status; |
fb5527e6 JM |
3851 | goto out; |
3852 | } | |
fb5527e6 JM |
3853 | /* |
3854 | * We need to ensure that the page cache pages are written to | |
3855 | * disk and invalidated to preserve the expected O_DIRECT | |
3856 | * semantics. | |
3857 | */ | |
3b93f911 | 3858 | endbyte = pos + status - 1; |
0b8def9d | 3859 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3860 | if (err == 0) { |
0b8def9d | 3861 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3862 | written += status; |
fb5527e6 | 3863 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3864 | pos >> PAGE_SHIFT, |
3865 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3866 | } else { |
3867 | /* | |
3868 | * We don't know how much we wrote, so just return | |
3869 | * the number of bytes which were direct-written | |
3870 | */ | |
3871 | } | |
3872 | } else { | |
0b8def9d AV |
3873 | written = generic_perform_write(file, from, iocb->ki_pos); |
3874 | if (likely(written > 0)) | |
3875 | iocb->ki_pos += written; | |
fb5527e6 | 3876 | } |
1da177e4 LT |
3877 | out: |
3878 | current->backing_dev_info = NULL; | |
3879 | return written ? written : err; | |
3880 | } | |
8174202b | 3881 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3882 | |
e4dd9de3 | 3883 | /** |
8174202b | 3884 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3885 | * @iocb: IO state structure |
8174202b | 3886 | * @from: iov_iter with data to write |
e4dd9de3 | 3887 | * |
8174202b | 3888 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 3889 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 3890 | * and acquires i_rwsem as needed. |
a862f68a MR |
3891 | * Return: |
3892 | * * negative error code if no data has been written at all of | |
3893 | * vfs_fsync_range() failed for a synchronous write | |
3894 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3895 | */ |
8174202b | 3896 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3897 | { |
3898 | struct file *file = iocb->ki_filp; | |
148f948b | 3899 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3900 | ssize_t ret; |
1da177e4 | 3901 | |
5955102c | 3902 | inode_lock(inode); |
3309dd04 AV |
3903 | ret = generic_write_checks(iocb, from); |
3904 | if (ret > 0) | |
5f380c7f | 3905 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3906 | inode_unlock(inode); |
1da177e4 | 3907 | |
e2592217 CH |
3908 | if (ret > 0) |
3909 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3910 | return ret; |
3911 | } | |
8174202b | 3912 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3913 | |
cf9a2ae8 DH |
3914 | /** |
3915 | * try_to_release_page() - release old fs-specific metadata on a page | |
3916 | * | |
3917 | * @page: the page which the kernel is trying to free | |
3918 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3919 | * | |
3920 | * The address_space is to try to release any data against the page | |
a862f68a | 3921 | * (presumably at page->private). |
cf9a2ae8 | 3922 | * |
266cf658 DH |
3923 | * This may also be called if PG_fscache is set on a page, indicating that the |
3924 | * page is known to the local caching routines. | |
3925 | * | |
cf9a2ae8 | 3926 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3927 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3928 | * |
a862f68a | 3929 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3930 | */ |
3931 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3932 | { | |
3933 | struct address_space * const mapping = page->mapping; | |
3934 | ||
3935 | BUG_ON(!PageLocked(page)); | |
3936 | if (PageWriteback(page)) | |
3937 | return 0; | |
3938 | ||
3939 | if (mapping && mapping->a_ops->releasepage) | |
3940 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3941 | return try_to_free_buffers(page); | |
3942 | } | |
3943 | ||
3944 | EXPORT_SYMBOL(try_to_release_page); |