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)); |
51b8c1fe | 264 | spin_lock(&mapping->host->i_lock); |
30472509 | 265 | xa_lock_irq(&mapping->i_pages); |
62cccb8c | 266 | __delete_from_page_cache(page, NULL); |
30472509 | 267 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
268 | if (mapping_shrinkable(mapping)) |
269 | inode_add_lru(mapping->host); | |
270 | spin_unlock(&mapping->host->i_lock); | |
6072d13c | 271 | |
59c66c5f | 272 | page_cache_free_page(mapping, page); |
97cecb5a MK |
273 | } |
274 | EXPORT_SYMBOL(delete_from_page_cache); | |
275 | ||
aa65c29c | 276 | /* |
ef8e5717 | 277 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
278 | * @mapping: the mapping to which pages belong |
279 | * @pvec: pagevec with pages to delete | |
280 | * | |
b93b0163 | 281 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
282 | * from the mapping. The function expects @pvec to be sorted by page index |
283 | * and is optimised for it to be dense. | |
b93b0163 | 284 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 285 | * modified). The function expects only THP head pages to be present in the |
4101196b | 286 | * @pvec. |
aa65c29c | 287 | * |
b93b0163 | 288 | * The function expects the i_pages lock to be held. |
aa65c29c | 289 | */ |
ef8e5717 | 290 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
291 | struct pagevec *pvec) |
292 | { | |
ef8e5717 | 293 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 294 | int total_pages = 0; |
4101196b | 295 | int i = 0; |
aa65c29c | 296 | struct page *page; |
aa65c29c | 297 | |
ef8e5717 MW |
298 | mapping_set_update(&xas, mapping); |
299 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 300 | if (i >= pagevec_count(pvec)) |
aa65c29c | 301 | break; |
4101196b MWO |
302 | |
303 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 304 | if (xa_is_value(page)) |
aa65c29c | 305 | continue; |
4101196b MWO |
306 | /* |
307 | * A page got inserted in our range? Skip it. We have our | |
308 | * pages locked so they are protected from being removed. | |
309 | * If we see a page whose index is higher than ours, it | |
310 | * means our page has been removed, which shouldn't be | |
311 | * possible because we're holding the PageLock. | |
312 | */ | |
313 | if (page != pvec->pages[i]) { | |
314 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
315 | page); | |
316 | continue; | |
317 | } | |
318 | ||
319 | WARN_ON_ONCE(!PageLocked(page)); | |
320 | ||
321 | if (page->index == xas.xa_index) | |
aa65c29c | 322 | page->mapping = NULL; |
4101196b MWO |
323 | /* Leave page->index set: truncation lookup relies on it */ |
324 | ||
325 | /* | |
326 | * Move to the next page in the vector if this is a regular | |
327 | * page or the index is of the last sub-page of this compound | |
328 | * page. | |
329 | */ | |
330 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 331 | i++; |
ef8e5717 | 332 | xas_store(&xas, NULL); |
aa65c29c JK |
333 | total_pages++; |
334 | } | |
335 | mapping->nrpages -= total_pages; | |
336 | } | |
337 | ||
338 | void delete_from_page_cache_batch(struct address_space *mapping, | |
339 | struct pagevec *pvec) | |
340 | { | |
341 | int i; | |
aa65c29c JK |
342 | |
343 | if (!pagevec_count(pvec)) | |
344 | return; | |
345 | ||
51b8c1fe | 346 | spin_lock(&mapping->host->i_lock); |
30472509 | 347 | xa_lock_irq(&mapping->i_pages); |
aa65c29c JK |
348 | for (i = 0; i < pagevec_count(pvec); i++) { |
349 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
350 | ||
351 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
352 | } | |
ef8e5717 | 353 | page_cache_delete_batch(mapping, pvec); |
30472509 | 354 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
355 | if (mapping_shrinkable(mapping)) |
356 | inode_add_lru(mapping->host); | |
357 | spin_unlock(&mapping->host->i_lock); | |
aa65c29c JK |
358 | |
359 | for (i = 0; i < pagevec_count(pvec); i++) | |
360 | page_cache_free_page(mapping, pvec->pages[i]); | |
361 | } | |
362 | ||
d72d9e2a | 363 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
364 | { |
365 | int ret = 0; | |
366 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
367 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
368 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 369 | ret = -ENOSPC; |
7fcbbaf1 JA |
370 | if (test_bit(AS_EIO, &mapping->flags) && |
371 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
372 | ret = -EIO; |
373 | return ret; | |
374 | } | |
d72d9e2a | 375 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 376 | |
76341cab JL |
377 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
378 | { | |
379 | /* Check for outstanding write errors */ | |
380 | if (test_bit(AS_EIO, &mapping->flags)) | |
381 | return -EIO; | |
382 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
383 | return -ENOSPC; | |
384 | return 0; | |
385 | } | |
386 | ||
5a798493 JB |
387 | /** |
388 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range | |
389 | * @mapping: address space structure to write | |
390 | * @wbc: the writeback_control controlling the writeout | |
391 | * | |
392 | * Call writepages on the mapping using the provided wbc to control the | |
393 | * writeout. | |
394 | * | |
395 | * Return: %0 on success, negative error code otherwise. | |
396 | */ | |
397 | int filemap_fdatawrite_wbc(struct address_space *mapping, | |
398 | struct writeback_control *wbc) | |
399 | { | |
400 | int ret; | |
401 | ||
402 | if (!mapping_can_writeback(mapping) || | |
403 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
404 | return 0; | |
405 | ||
406 | wbc_attach_fdatawrite_inode(wbc, mapping->host); | |
407 | ret = do_writepages(mapping, wbc); | |
408 | wbc_detach_inode(wbc); | |
409 | return ret; | |
410 | } | |
411 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); | |
412 | ||
1da177e4 | 413 | /** |
485bb99b | 414 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
415 | * @mapping: address space structure to write |
416 | * @start: offset in bytes where the range starts | |
469eb4d0 | 417 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 418 | * @sync_mode: enable synchronous operation |
1da177e4 | 419 | * |
485bb99b RD |
420 | * Start writeback against all of a mapping's dirty pages that lie |
421 | * within the byte offsets <start, end> inclusive. | |
422 | * | |
1da177e4 | 423 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 424 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
425 | * these two operations is that if a dirty page/buffer is encountered, it must |
426 | * be waited upon, and not just skipped over. | |
a862f68a MR |
427 | * |
428 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 429 | */ |
ebcf28e1 AM |
430 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
431 | loff_t end, int sync_mode) | |
1da177e4 | 432 | { |
1da177e4 LT |
433 | struct writeback_control wbc = { |
434 | .sync_mode = sync_mode, | |
05fe478d | 435 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
436 | .range_start = start, |
437 | .range_end = end, | |
1da177e4 LT |
438 | }; |
439 | ||
5a798493 | 440 | return filemap_fdatawrite_wbc(mapping, &wbc); |
1da177e4 LT |
441 | } |
442 | ||
443 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
444 | int sync_mode) | |
445 | { | |
111ebb6e | 446 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
447 | } |
448 | ||
449 | int filemap_fdatawrite(struct address_space *mapping) | |
450 | { | |
451 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
452 | } | |
453 | EXPORT_SYMBOL(filemap_fdatawrite); | |
454 | ||
f4c0a0fd | 455 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 456 | loff_t end) |
1da177e4 LT |
457 | { |
458 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
459 | } | |
f4c0a0fd | 460 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 461 | |
485bb99b RD |
462 | /** |
463 | * filemap_flush - mostly a non-blocking flush | |
464 | * @mapping: target address_space | |
465 | * | |
1da177e4 LT |
466 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
467 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
468 | * |
469 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
470 | */ |
471 | int filemap_flush(struct address_space *mapping) | |
472 | { | |
473 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
474 | } | |
475 | EXPORT_SYMBOL(filemap_flush); | |
476 | ||
7fc9e472 GR |
477 | /** |
478 | * filemap_range_has_page - check if a page exists in range. | |
479 | * @mapping: address space within which to check | |
480 | * @start_byte: offset in bytes where the range starts | |
481 | * @end_byte: offset in bytes where the range ends (inclusive) | |
482 | * | |
483 | * Find at least one page in the range supplied, usually used to check if | |
484 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
485 | * |
486 | * Return: %true if at least one page exists in the specified range, | |
487 | * %false otherwise. | |
7fc9e472 GR |
488 | */ |
489 | bool filemap_range_has_page(struct address_space *mapping, | |
490 | loff_t start_byte, loff_t end_byte) | |
491 | { | |
f7b68046 | 492 | struct page *page; |
8fa8e538 MW |
493 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
494 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
495 | |
496 | if (end_byte < start_byte) | |
497 | return false; | |
498 | ||
8fa8e538 MW |
499 | rcu_read_lock(); |
500 | for (;;) { | |
501 | page = xas_find(&xas, max); | |
502 | if (xas_retry(&xas, page)) | |
503 | continue; | |
504 | /* Shadow entries don't count */ | |
505 | if (xa_is_value(page)) | |
506 | continue; | |
507 | /* | |
508 | * We don't need to try to pin this page; we're about to | |
509 | * release the RCU lock anyway. It is enough to know that | |
510 | * there was a page here recently. | |
511 | */ | |
512 | break; | |
513 | } | |
514 | rcu_read_unlock(); | |
7fc9e472 | 515 | |
8fa8e538 | 516 | return page != NULL; |
7fc9e472 GR |
517 | } |
518 | EXPORT_SYMBOL(filemap_range_has_page); | |
519 | ||
5e8fcc1a | 520 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 521 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 522 | { |
09cbfeaf KS |
523 | pgoff_t index = start_byte >> PAGE_SHIFT; |
524 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
525 | struct pagevec pvec; |
526 | int nr_pages; | |
1da177e4 | 527 | |
94004ed7 | 528 | if (end_byte < start_byte) |
5e8fcc1a | 529 | return; |
1da177e4 | 530 | |
86679820 | 531 | pagevec_init(&pvec); |
312e9d2f | 532 | while (index <= end) { |
1da177e4 LT |
533 | unsigned i; |
534 | ||
312e9d2f | 535 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 536 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
537 | if (!nr_pages) |
538 | break; | |
539 | ||
1da177e4 LT |
540 | for (i = 0; i < nr_pages; i++) { |
541 | struct page *page = pvec.pages[i]; | |
542 | ||
1da177e4 | 543 | wait_on_page_writeback(page); |
5e8fcc1a | 544 | ClearPageError(page); |
1da177e4 LT |
545 | } |
546 | pagevec_release(&pvec); | |
547 | cond_resched(); | |
548 | } | |
aa750fd7 JN |
549 | } |
550 | ||
551 | /** | |
552 | * filemap_fdatawait_range - wait for writeback to complete | |
553 | * @mapping: address space structure to wait for | |
554 | * @start_byte: offset in bytes where the range starts | |
555 | * @end_byte: offset in bytes where the range ends (inclusive) | |
556 | * | |
557 | * Walk the list of under-writeback pages of the given address space | |
558 | * in the given range and wait for all of them. Check error status of | |
559 | * the address space and return it. | |
560 | * | |
561 | * Since the error status of the address space is cleared by this function, | |
562 | * callers are responsible for checking the return value and handling and/or | |
563 | * reporting the error. | |
a862f68a MR |
564 | * |
565 | * Return: error status of the address space. | |
aa750fd7 JN |
566 | */ |
567 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
568 | loff_t end_byte) | |
569 | { | |
5e8fcc1a JL |
570 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
571 | return filemap_check_errors(mapping); | |
1da177e4 | 572 | } |
d3bccb6f JK |
573 | EXPORT_SYMBOL(filemap_fdatawait_range); |
574 | ||
aa0bfcd9 RZ |
575 | /** |
576 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
577 | * @mapping: address space structure to wait for | |
578 | * @start_byte: offset in bytes where the range starts | |
579 | * @end_byte: offset in bytes where the range ends (inclusive) | |
580 | * | |
581 | * Walk the list of under-writeback pages of the given address space in the | |
582 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
583 | * this function does not clear error status of the address space. | |
584 | * | |
585 | * Use this function if callers don't handle errors themselves. Expected | |
586 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
587 | * fsfreeze(8) | |
588 | */ | |
589 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
590 | loff_t start_byte, loff_t end_byte) | |
591 | { | |
592 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
593 | return filemap_check_and_keep_errors(mapping); | |
594 | } | |
595 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
596 | ||
a823e458 JL |
597 | /** |
598 | * file_fdatawait_range - wait for writeback to complete | |
599 | * @file: file pointing to address space structure to wait for | |
600 | * @start_byte: offset in bytes where the range starts | |
601 | * @end_byte: offset in bytes where the range ends (inclusive) | |
602 | * | |
603 | * Walk the list of under-writeback pages of the address space that file | |
604 | * refers to, in the given range and wait for all of them. Check error | |
605 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
606 | * | |
607 | * Since the error status of the file is advanced by this function, | |
608 | * callers are responsible for checking the return value and handling and/or | |
609 | * reporting the error. | |
a862f68a MR |
610 | * |
611 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
612 | */ |
613 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
614 | { | |
615 | struct address_space *mapping = file->f_mapping; | |
616 | ||
617 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
618 | return file_check_and_advance_wb_err(file); | |
619 | } | |
620 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 621 | |
aa750fd7 JN |
622 | /** |
623 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
624 | * @mapping: address space structure to wait for | |
625 | * | |
626 | * Walk the list of under-writeback pages of the given address space | |
627 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
628 | * does not clear error status of the address space. | |
629 | * | |
630 | * Use this function if callers don't handle errors themselves. Expected | |
631 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
632 | * fsfreeze(8) | |
a862f68a MR |
633 | * |
634 | * Return: error status of the address space. | |
aa750fd7 | 635 | */ |
76341cab | 636 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 637 | { |
ffb959bb | 638 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 639 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 640 | } |
76341cab | 641 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 642 | |
875d91b1 | 643 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 644 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 645 | { |
875d91b1 | 646 | return mapping->nrpages; |
1da177e4 | 647 | } |
1da177e4 | 648 | |
f8ee8909 JA |
649 | static bool filemap_range_has_writeback(struct address_space *mapping, |
650 | loff_t start_byte, loff_t end_byte) | |
651 | { | |
652 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
653 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
654 | struct page *page; | |
655 | ||
656 | if (end_byte < start_byte) | |
657 | return false; | |
658 | ||
659 | rcu_read_lock(); | |
660 | xas_for_each(&xas, page, max) { | |
661 | if (xas_retry(&xas, page)) | |
662 | continue; | |
663 | if (xa_is_value(page)) | |
664 | continue; | |
665 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
666 | break; | |
667 | } | |
668 | rcu_read_unlock(); | |
669 | return page != NULL; | |
670 | ||
671 | } | |
672 | ||
63135aa3 JA |
673 | /** |
674 | * filemap_range_needs_writeback - check if range potentially needs writeback | |
675 | * @mapping: address space within which to check | |
676 | * @start_byte: offset in bytes where the range starts | |
677 | * @end_byte: offset in bytes where the range ends (inclusive) | |
678 | * | |
679 | * Find at least one page in the range supplied, usually used to check if | |
680 | * direct writing in this range will trigger a writeback. Used by O_DIRECT | |
681 | * read/write with IOCB_NOWAIT, to see if the caller needs to do | |
682 | * filemap_write_and_wait_range() before proceeding. | |
683 | * | |
684 | * Return: %true if the caller should do filemap_write_and_wait_range() before | |
685 | * doing O_DIRECT to a page in this range, %false otherwise. | |
686 | */ | |
687 | bool filemap_range_needs_writeback(struct address_space *mapping, | |
688 | loff_t start_byte, loff_t end_byte) | |
689 | { | |
63135aa3 JA |
690 | if (!mapping_needs_writeback(mapping)) |
691 | return false; | |
692 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && | |
693 | !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
694 | return false; | |
f8ee8909 | 695 | return filemap_range_has_writeback(mapping, start_byte, end_byte); |
63135aa3 JA |
696 | } |
697 | EXPORT_SYMBOL_GPL(filemap_range_needs_writeback); | |
698 | ||
485bb99b RD |
699 | /** |
700 | * filemap_write_and_wait_range - write out & wait on a file range | |
701 | * @mapping: the address_space for the pages | |
702 | * @lstart: offset in bytes where the range starts | |
703 | * @lend: offset in bytes where the range ends (inclusive) | |
704 | * | |
469eb4d0 AM |
705 | * Write out and wait upon file offsets lstart->lend, inclusive. |
706 | * | |
0e056eb5 | 707 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 708 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
709 | * |
710 | * Return: error status of the address space. | |
469eb4d0 | 711 | */ |
1da177e4 LT |
712 | int filemap_write_and_wait_range(struct address_space *mapping, |
713 | loff_t lstart, loff_t lend) | |
714 | { | |
28fd1298 | 715 | int err = 0; |
1da177e4 | 716 | |
9326c9b2 | 717 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
718 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
719 | WB_SYNC_ALL); | |
ddf8f376 IW |
720 | /* |
721 | * Even if the above returned error, the pages may be | |
722 | * written partially (e.g. -ENOSPC), so we wait for it. | |
723 | * But the -EIO is special case, it may indicate the worst | |
724 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
725 | */ | |
28fd1298 | 726 | if (err != -EIO) { |
94004ed7 CH |
727 | int err2 = filemap_fdatawait_range(mapping, |
728 | lstart, lend); | |
28fd1298 OH |
729 | if (!err) |
730 | err = err2; | |
cbeaf951 JL |
731 | } else { |
732 | /* Clear any previously stored errors */ | |
733 | filemap_check_errors(mapping); | |
28fd1298 | 734 | } |
865ffef3 DM |
735 | } else { |
736 | err = filemap_check_errors(mapping); | |
1da177e4 | 737 | } |
28fd1298 | 738 | return err; |
1da177e4 | 739 | } |
f6995585 | 740 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 741 | |
5660e13d JL |
742 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
743 | { | |
3acdfd28 | 744 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
745 | |
746 | trace_filemap_set_wb_err(mapping, eseq); | |
747 | } | |
748 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
749 | ||
750 | /** | |
751 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
752 | * and advance wb_err to current one | |
753 | * @file: struct file on which the error is being reported | |
754 | * | |
755 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
756 | * want to report any writeback errors that occurred since the last fsync (or | |
757 | * since the file was opened if there haven't been any). | |
758 | * | |
759 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
760 | * then just quickly return 0. The file is all caught up. | |
761 | * | |
762 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
763 | * it and try to swap it into place. If it works, or another task beat us | |
764 | * to it with the new value, then update the f_wb_err and return the error | |
765 | * portion. The error at this point must be reported via proper channels | |
766 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
767 | * | |
768 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
769 | * value is protected by the f_lock since we must ensure that it reflects | |
770 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
771 | * |
772 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
773 | */ |
774 | int file_check_and_advance_wb_err(struct file *file) | |
775 | { | |
776 | int err = 0; | |
777 | errseq_t old = READ_ONCE(file->f_wb_err); | |
778 | struct address_space *mapping = file->f_mapping; | |
779 | ||
780 | /* Locklessly handle the common case where nothing has changed */ | |
781 | if (errseq_check(&mapping->wb_err, old)) { | |
782 | /* Something changed, must use slow path */ | |
783 | spin_lock(&file->f_lock); | |
784 | old = file->f_wb_err; | |
785 | err = errseq_check_and_advance(&mapping->wb_err, | |
786 | &file->f_wb_err); | |
787 | trace_file_check_and_advance_wb_err(file, old); | |
788 | spin_unlock(&file->f_lock); | |
789 | } | |
f4e222c5 JL |
790 | |
791 | /* | |
792 | * We're mostly using this function as a drop in replacement for | |
793 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
794 | * that the legacy code would have had on these flags. | |
795 | */ | |
796 | clear_bit(AS_EIO, &mapping->flags); | |
797 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
798 | return err; |
799 | } | |
800 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
801 | ||
802 | /** | |
803 | * file_write_and_wait_range - write out & wait on a file range | |
804 | * @file: file pointing to address_space with pages | |
805 | * @lstart: offset in bytes where the range starts | |
806 | * @lend: offset in bytes where the range ends (inclusive) | |
807 | * | |
808 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
809 | * | |
810 | * Note that @lend is inclusive (describes the last byte to be written) so | |
811 | * that this function can be used to write to the very end-of-file (end = -1). | |
812 | * | |
813 | * After writing out and waiting on the data, we check and advance the | |
814 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
815 | * |
816 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
817 | */ |
818 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
819 | { | |
820 | int err = 0, err2; | |
821 | struct address_space *mapping = file->f_mapping; | |
822 | ||
9326c9b2 | 823 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
824 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
825 | WB_SYNC_ALL); | |
826 | /* See comment of filemap_write_and_wait() */ | |
827 | if (err != -EIO) | |
828 | __filemap_fdatawait_range(mapping, lstart, lend); | |
829 | } | |
830 | err2 = file_check_and_advance_wb_err(file); | |
831 | if (!err) | |
832 | err = err2; | |
833 | return err; | |
834 | } | |
835 | EXPORT_SYMBOL(file_write_and_wait_range); | |
836 | ||
ef6a3c63 MS |
837 | /** |
838 | * replace_page_cache_page - replace a pagecache page with a new one | |
839 | * @old: page to be replaced | |
840 | * @new: page to replace with | |
ef6a3c63 MS |
841 | * |
842 | * This function replaces a page in the pagecache with a new one. On | |
843 | * success it acquires the pagecache reference for the new page and | |
844 | * drops it for the old page. Both the old and new pages must be | |
845 | * locked. This function does not add the new page to the LRU, the | |
846 | * caller must do that. | |
847 | * | |
74d60958 | 848 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 849 | */ |
1f7ef657 | 850 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 851 | { |
d21bba2b MWO |
852 | struct folio *fold = page_folio(old); |
853 | struct folio *fnew = page_folio(new); | |
74d60958 MW |
854 | struct address_space *mapping = old->mapping; |
855 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
856 | pgoff_t offset = old->index; | |
857 | XA_STATE(xas, &mapping->i_pages, offset); | |
ef6a3c63 | 858 | |
309381fe SL |
859 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
860 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
861 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 862 | |
74d60958 MW |
863 | get_page(new); |
864 | new->mapping = mapping; | |
865 | new->index = offset; | |
ef6a3c63 | 866 | |
d21bba2b | 867 | mem_cgroup_migrate(fold, fnew); |
0d1c2072 | 868 | |
30472509 | 869 | xas_lock_irq(&xas); |
74d60958 | 870 | xas_store(&xas, new); |
4165b9b4 | 871 | |
74d60958 MW |
872 | old->mapping = NULL; |
873 | /* hugetlb pages do not participate in page cache accounting. */ | |
874 | if (!PageHuge(old)) | |
0d1c2072 | 875 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 876 | if (!PageHuge(new)) |
0d1c2072 | 877 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 878 | if (PageSwapBacked(old)) |
0d1c2072 | 879 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 880 | if (PageSwapBacked(new)) |
0d1c2072 | 881 | __inc_lruvec_page_state(new, NR_SHMEM); |
30472509 | 882 | xas_unlock_irq(&xas); |
74d60958 MW |
883 | if (freepage) |
884 | freepage(old); | |
885 | put_page(old); | |
ef6a3c63 MS |
886 | } |
887 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
888 | ||
9dd3d069 MWO |
889 | noinline int __filemap_add_folio(struct address_space *mapping, |
890 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) | |
1da177e4 | 891 | { |
9dd3d069 MWO |
892 | XA_STATE(xas, &mapping->i_pages, index); |
893 | int huge = folio_test_hugetlb(folio); | |
e286781d | 894 | int error; |
da74240e | 895 | bool charged = false; |
e286781d | 896 | |
9dd3d069 MWO |
897 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
898 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); | |
74d60958 | 899 | mapping_set_update(&xas, mapping); |
e286781d | 900 | |
9dd3d069 MWO |
901 | folio_get(folio); |
902 | folio->mapping = mapping; | |
903 | folio->index = index; | |
66a0c8ee | 904 | |
3fea5a49 | 905 | if (!huge) { |
9dd3d069 MWO |
906 | error = mem_cgroup_charge(folio, NULL, gfp); |
907 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); | |
3fea5a49 JW |
908 | if (error) |
909 | goto error; | |
da74240e | 910 | charged = true; |
3fea5a49 JW |
911 | } |
912 | ||
198b62f8 MWO |
913 | gfp &= GFP_RECLAIM_MASK; |
914 | ||
74d60958 | 915 | do { |
198b62f8 MWO |
916 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
917 | void *entry, *old = NULL; | |
918 | ||
9dd3d069 | 919 | if (order > folio_order(folio)) |
198b62f8 MWO |
920 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
921 | order, gfp); | |
74d60958 | 922 | xas_lock_irq(&xas); |
198b62f8 MWO |
923 | xas_for_each_conflict(&xas, entry) { |
924 | old = entry; | |
925 | if (!xa_is_value(entry)) { | |
926 | xas_set_err(&xas, -EEXIST); | |
927 | goto unlock; | |
928 | } | |
929 | } | |
930 | ||
931 | if (old) { | |
932 | if (shadowp) | |
933 | *shadowp = old; | |
934 | /* entry may have been split before we acquired lock */ | |
935 | order = xa_get_order(xas.xa, xas.xa_index); | |
9dd3d069 | 936 | if (order > folio_order(folio)) { |
198b62f8 MWO |
937 | xas_split(&xas, old, order); |
938 | xas_reset(&xas); | |
939 | } | |
940 | } | |
941 | ||
9dd3d069 | 942 | xas_store(&xas, folio); |
74d60958 MW |
943 | if (xas_error(&xas)) |
944 | goto unlock; | |
945 | ||
74d60958 MW |
946 | mapping->nrpages++; |
947 | ||
948 | /* hugetlb pages do not participate in page cache accounting */ | |
949 | if (!huge) | |
9dd3d069 | 950 | __lruvec_stat_add_folio(folio, NR_FILE_PAGES); |
74d60958 MW |
951 | unlock: |
952 | xas_unlock_irq(&xas); | |
198b62f8 | 953 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 954 | |
3fea5a49 JW |
955 | if (xas_error(&xas)) { |
956 | error = xas_error(&xas); | |
da74240e | 957 | if (charged) |
9dd3d069 | 958 | mem_cgroup_uncharge(folio); |
74d60958 | 959 | goto error; |
3fea5a49 | 960 | } |
4165b9b4 | 961 | |
9dd3d069 | 962 | trace_mm_filemap_add_to_page_cache(&folio->page); |
66a0c8ee | 963 | return 0; |
74d60958 | 964 | error: |
9dd3d069 | 965 | folio->mapping = NULL; |
66a0c8ee | 966 | /* Leave page->index set: truncation relies upon it */ |
9dd3d069 | 967 | folio_put(folio); |
3fea5a49 | 968 | return error; |
1da177e4 | 969 | } |
9dd3d069 | 970 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
a528910e JW |
971 | |
972 | /** | |
973 | * add_to_page_cache_locked - add a locked page to the pagecache | |
974 | * @page: page to add | |
975 | * @mapping: the page's address_space | |
976 | * @offset: page index | |
977 | * @gfp_mask: page allocation mode | |
978 | * | |
979 | * This function is used to add a page to the pagecache. It must be locked. | |
980 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
981 | * |
982 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
983 | */ |
984 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
985 | pgoff_t offset, gfp_t gfp_mask) | |
986 | { | |
9dd3d069 | 987 | return __filemap_add_folio(mapping, page_folio(page), offset, |
a528910e JW |
988 | gfp_mask, NULL); |
989 | } | |
e286781d | 990 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 | 991 | |
9dd3d069 MWO |
992 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
993 | pgoff_t index, gfp_t gfp) | |
1da177e4 | 994 | { |
a528910e | 995 | void *shadow = NULL; |
4f98a2fe RR |
996 | int ret; |
997 | ||
9dd3d069 MWO |
998 | __folio_set_locked(folio); |
999 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); | |
a528910e | 1000 | if (unlikely(ret)) |
9dd3d069 | 1001 | __folio_clear_locked(folio); |
a528910e JW |
1002 | else { |
1003 | /* | |
9dd3d069 | 1004 | * The folio might have been evicted from cache only |
a528910e | 1005 | * recently, in which case it should be activated like |
9dd3d069 MWO |
1006 | * any other repeatedly accessed folio. |
1007 | * The exception is folios getting rewritten; evicting other | |
f0281a00 RR |
1008 | * data from the working set, only to cache data that will |
1009 | * get overwritten with something else, is a waste of memory. | |
a528910e | 1010 | */ |
9dd3d069 MWO |
1011 | WARN_ON_ONCE(folio_test_active(folio)); |
1012 | if (!(gfp & __GFP_WRITE) && shadow) | |
1013 | workingset_refault(folio, shadow); | |
1014 | folio_add_lru(folio); | |
a528910e | 1015 | } |
1da177e4 LT |
1016 | return ret; |
1017 | } | |
9dd3d069 | 1018 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
1da177e4 | 1019 | |
44110fe3 | 1020 | #ifdef CONFIG_NUMA |
bb3c579e | 1021 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
44110fe3 | 1022 | { |
c0ff7453 | 1023 | int n; |
bb3c579e | 1024 | struct folio *folio; |
c0ff7453 | 1025 | |
44110fe3 | 1026 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
1027 | unsigned int cpuset_mems_cookie; |
1028 | do { | |
d26914d1 | 1029 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 1030 | n = cpuset_mem_spread_node(); |
bb3c579e MWO |
1031 | folio = __folio_alloc_node(gfp, order, n); |
1032 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); | |
cc9a6c87 | 1033 | |
bb3c579e | 1034 | return folio; |
44110fe3 | 1035 | } |
bb3c579e | 1036 | return folio_alloc(gfp, order); |
44110fe3 | 1037 | } |
bb3c579e | 1038 | EXPORT_SYMBOL(filemap_alloc_folio); |
44110fe3 PJ |
1039 | #endif |
1040 | ||
7506ae6a JK |
1041 | /* |
1042 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings | |
1043 | * | |
1044 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. | |
1045 | * | |
1046 | * @mapping1: the first mapping to lock | |
1047 | * @mapping2: the second mapping to lock | |
1048 | */ | |
1049 | void filemap_invalidate_lock_two(struct address_space *mapping1, | |
1050 | struct address_space *mapping2) | |
1051 | { | |
1052 | if (mapping1 > mapping2) | |
1053 | swap(mapping1, mapping2); | |
1054 | if (mapping1) | |
1055 | down_write(&mapping1->invalidate_lock); | |
1056 | if (mapping2 && mapping1 != mapping2) | |
1057 | down_write_nested(&mapping2->invalidate_lock, 1); | |
1058 | } | |
1059 | EXPORT_SYMBOL(filemap_invalidate_lock_two); | |
1060 | ||
1061 | /* | |
1062 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings | |
1063 | * | |
1064 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. | |
1065 | * | |
1066 | * @mapping1: the first mapping to unlock | |
1067 | * @mapping2: the second mapping to unlock | |
1068 | */ | |
1069 | void filemap_invalidate_unlock_two(struct address_space *mapping1, | |
1070 | struct address_space *mapping2) | |
1071 | { | |
1072 | if (mapping1) | |
1073 | up_write(&mapping1->invalidate_lock); | |
1074 | if (mapping2 && mapping1 != mapping2) | |
1075 | up_write(&mapping2->invalidate_lock); | |
1076 | } | |
1077 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); | |
1078 | ||
1da177e4 LT |
1079 | /* |
1080 | * In order to wait for pages to become available there must be | |
1081 | * waitqueues associated with pages. By using a hash table of | |
1082 | * waitqueues where the bucket discipline is to maintain all | |
1083 | * waiters on the same queue and wake all when any of the pages | |
1084 | * become available, and for the woken contexts to check to be | |
1085 | * sure the appropriate page became available, this saves space | |
1086 | * at a cost of "thundering herd" phenomena during rare hash | |
1087 | * collisions. | |
1088 | */ | |
62906027 NP |
1089 | #define PAGE_WAIT_TABLE_BITS 8 |
1090 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
df4d4f12 | 1091 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
62906027 | 1092 | |
df4d4f12 | 1093 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
1da177e4 | 1094 | { |
df4d4f12 | 1095 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1096 | } |
1da177e4 | 1097 | |
62906027 | 1098 | void __init pagecache_init(void) |
1da177e4 | 1099 | { |
62906027 | 1100 | int i; |
1da177e4 | 1101 | |
62906027 | 1102 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
df4d4f12 | 1103 | init_waitqueue_head(&folio_wait_table[i]); |
62906027 NP |
1104 | |
1105 | page_writeback_init(); | |
1da177e4 | 1106 | } |
1da177e4 | 1107 | |
5ef64cc8 LT |
1108 | /* |
1109 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1110 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1111 | * one. |
1112 | * | |
1113 | * We have: | |
1114 | * | |
1115 | * (a) no special bits set: | |
1116 | * | |
1117 | * We're just waiting for the bit to be released, and when a waker | |
1118 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1119 | * and remove it from the wait queue. | |
1120 | * | |
1121 | * Simple and straightforward. | |
1122 | * | |
1123 | * (b) WQ_FLAG_EXCLUSIVE: | |
1124 | * | |
1125 | * The waiter is waiting to get the lock, and only one waiter should | |
1126 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1127 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1128 | * | |
1129 | * This is the traditional exclusive wait. | |
1130 | * | |
5868ec26 | 1131 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1132 | * |
1133 | * The waiter is waiting to get the bit, and additionally wants the | |
1134 | * lock to be transferred to it for fair lock behavior. If the lock | |
1135 | * cannot be taken, we stop walking the wait queue without waking | |
1136 | * the waiter. | |
1137 | * | |
1138 | * This is the "fair lock handoff" case, and in addition to setting | |
1139 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1140 | * that it now has the lock. | |
1141 | */ | |
ac6424b9 | 1142 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1143 | { |
5ef64cc8 | 1144 | unsigned int flags; |
62906027 NP |
1145 | struct wait_page_key *key = arg; |
1146 | struct wait_page_queue *wait_page | |
1147 | = container_of(wait, struct wait_page_queue, wait); | |
1148 | ||
cdc8fcb4 | 1149 | if (!wake_page_match(wait_page, key)) |
62906027 | 1150 | return 0; |
3510ca20 | 1151 | |
9a1ea439 | 1152 | /* |
5ef64cc8 LT |
1153 | * If it's a lock handoff wait, we get the bit for it, and |
1154 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1155 | */ |
5ef64cc8 LT |
1156 | flags = wait->flags; |
1157 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
df4d4f12 | 1158 | if (test_bit(key->bit_nr, &key->folio->flags)) |
2a9127fc | 1159 | return -1; |
5ef64cc8 | 1160 | if (flags & WQ_FLAG_CUSTOM) { |
df4d4f12 | 1161 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
5ef64cc8 LT |
1162 | return -1; |
1163 | flags |= WQ_FLAG_DONE; | |
1164 | } | |
2a9127fc | 1165 | } |
f62e00cc | 1166 | |
5ef64cc8 LT |
1167 | /* |
1168 | * We are holding the wait-queue lock, but the waiter that | |
1169 | * is waiting for this will be checking the flags without | |
1170 | * any locking. | |
1171 | * | |
1172 | * So update the flags atomically, and wake up the waiter | |
1173 | * afterwards to avoid any races. This store-release pairs | |
101c0bf6 | 1174 | * with the load-acquire in folio_wait_bit_common(). |
5ef64cc8 LT |
1175 | */ |
1176 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1177 | wake_up_state(wait->private, mode); |
1178 | ||
1179 | /* | |
1180 | * Ok, we have successfully done what we're waiting for, | |
1181 | * and we can unconditionally remove the wait entry. | |
1182 | * | |
5ef64cc8 LT |
1183 | * Note that this pairs with the "finish_wait()" in the |
1184 | * waiter, and has to be the absolute last thing we do. | |
1185 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1186 | * might be de-allocated and the process might even have |
1187 | * exited. | |
2a9127fc | 1188 | */ |
c6fe44d9 | 1189 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1190 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1191 | } |
1192 | ||
6974d7c9 | 1193 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
cbbce822 | 1194 | { |
df4d4f12 | 1195 | wait_queue_head_t *q = folio_waitqueue(folio); |
62906027 NP |
1196 | struct wait_page_key key; |
1197 | unsigned long flags; | |
11a19c7b | 1198 | wait_queue_entry_t bookmark; |
cbbce822 | 1199 | |
df4d4f12 | 1200 | key.folio = folio; |
62906027 NP |
1201 | key.bit_nr = bit_nr; |
1202 | key.page_match = 0; | |
1203 | ||
11a19c7b TC |
1204 | bookmark.flags = 0; |
1205 | bookmark.private = NULL; | |
1206 | bookmark.func = NULL; | |
1207 | INIT_LIST_HEAD(&bookmark.entry); | |
1208 | ||
62906027 | 1209 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1210 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1211 | ||
1212 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1213 | /* | |
1214 | * Take a breather from holding the lock, | |
1215 | * allow pages that finish wake up asynchronously | |
1216 | * to acquire the lock and remove themselves | |
1217 | * from wait queue | |
1218 | */ | |
1219 | spin_unlock_irqrestore(&q->lock, flags); | |
1220 | cpu_relax(); | |
1221 | spin_lock_irqsave(&q->lock, flags); | |
1222 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1223 | } | |
1224 | ||
62906027 NP |
1225 | /* |
1226 | * It is possible for other pages to have collided on the waitqueue | |
1227 | * hash, so in that case check for a page match. That prevents a long- | |
1228 | * term waiter | |
1229 | * | |
1230 | * It is still possible to miss a case here, when we woke page waiters | |
1231 | * and removed them from the waitqueue, but there are still other | |
1232 | * page waiters. | |
1233 | */ | |
1234 | if (!waitqueue_active(q) || !key.page_match) { | |
6974d7c9 | 1235 | folio_clear_waiters(folio); |
62906027 NP |
1236 | /* |
1237 | * It's possible to miss clearing Waiters here, when we woke | |
1238 | * our page waiters, but the hashed waitqueue has waiters for | |
1239 | * other pages on it. | |
1240 | * | |
1241 | * That's okay, it's a rare case. The next waker will clear it. | |
1242 | */ | |
1243 | } | |
1244 | spin_unlock_irqrestore(&q->lock, flags); | |
1245 | } | |
74d81bfa | 1246 | |
4268b480 | 1247 | static void folio_wake(struct folio *folio, int bit) |
74d81bfa | 1248 | { |
4268b480 | 1249 | if (!folio_test_waiters(folio)) |
74d81bfa | 1250 | return; |
6974d7c9 | 1251 | folio_wake_bit(folio, bit); |
74d81bfa | 1252 | } |
62906027 | 1253 | |
9a1ea439 | 1254 | /* |
101c0bf6 | 1255 | * A choice of three behaviors for folio_wait_bit_common(): |
9a1ea439 HD |
1256 | */ |
1257 | enum behavior { | |
1258 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
7c23c782 | 1259 | * __folio_lock() waiting on then setting PG_locked. |
9a1ea439 HD |
1260 | */ |
1261 | SHARED, /* Hold ref to page and check the bit when woken, like | |
9f2b04a2 | 1262 | * folio_wait_writeback() waiting on PG_writeback. |
9a1ea439 HD |
1263 | */ |
1264 | DROP, /* Drop ref to page before wait, no check when woken, | |
9f2b04a2 | 1265 | * like folio_put_wait_locked() on PG_locked. |
9a1ea439 HD |
1266 | */ |
1267 | }; | |
1268 | ||
2a9127fc | 1269 | /* |
101c0bf6 | 1270 | * Attempt to check (or get) the folio flag, and mark us done |
5ef64cc8 | 1271 | * if successful. |
2a9127fc | 1272 | */ |
101c0bf6 | 1273 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
2a9127fc LT |
1274 | struct wait_queue_entry *wait) |
1275 | { | |
1276 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
101c0bf6 | 1277 | if (test_and_set_bit(bit_nr, &folio->flags)) |
2a9127fc | 1278 | return false; |
101c0bf6 | 1279 | } else if (test_bit(bit_nr, &folio->flags)) |
2a9127fc LT |
1280 | return false; |
1281 | ||
5ef64cc8 | 1282 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1283 | return true; |
1284 | } | |
1285 | ||
5ef64cc8 LT |
1286 | /* How many times do we accept lock stealing from under a waiter? */ |
1287 | int sysctl_page_lock_unfairness = 5; | |
1288 | ||
101c0bf6 MWO |
1289 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
1290 | int state, enum behavior behavior) | |
62906027 | 1291 | { |
df4d4f12 | 1292 | wait_queue_head_t *q = folio_waitqueue(folio); |
5ef64cc8 | 1293 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1294 | struct wait_page_queue wait_page; |
ac6424b9 | 1295 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1296 | bool thrashing = false; |
9a1ea439 | 1297 | bool delayacct = false; |
eb414681 | 1298 | unsigned long pflags; |
62906027 | 1299 | |
eb414681 | 1300 | if (bit_nr == PG_locked && |
101c0bf6 MWO |
1301 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
1302 | if (!folio_test_swapbacked(folio)) { | |
eb414681 | 1303 | delayacct_thrashing_start(); |
9a1ea439 HD |
1304 | delayacct = true; |
1305 | } | |
eb414681 | 1306 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1307 | thrashing = true; |
1308 | } | |
1309 | ||
62906027 NP |
1310 | init_wait(wait); |
1311 | wait->func = wake_page_function; | |
df4d4f12 | 1312 | wait_page.folio = folio; |
62906027 NP |
1313 | wait_page.bit_nr = bit_nr; |
1314 | ||
5ef64cc8 LT |
1315 | repeat: |
1316 | wait->flags = 0; | |
1317 | if (behavior == EXCLUSIVE) { | |
1318 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1319 | if (--unfairness < 0) | |
1320 | wait->flags |= WQ_FLAG_CUSTOM; | |
1321 | } | |
1322 | ||
2a9127fc LT |
1323 | /* |
1324 | * Do one last check whether we can get the | |
1325 | * page bit synchronously. | |
1326 | * | |
101c0bf6 | 1327 | * Do the folio_set_waiters() marking before that |
2a9127fc LT |
1328 | * to let any waker we _just_ missed know they |
1329 | * need to wake us up (otherwise they'll never | |
1330 | * even go to the slow case that looks at the | |
1331 | * page queue), and add ourselves to the wait | |
1332 | * queue if we need to sleep. | |
1333 | * | |
1334 | * This part needs to be done under the queue | |
1335 | * lock to avoid races. | |
1336 | */ | |
1337 | spin_lock_irq(&q->lock); | |
101c0bf6 MWO |
1338 | folio_set_waiters(folio); |
1339 | if (!folio_trylock_flag(folio, bit_nr, wait)) | |
2a9127fc LT |
1340 | __add_wait_queue_entry_tail(q, wait); |
1341 | spin_unlock_irq(&q->lock); | |
62906027 | 1342 | |
2a9127fc LT |
1343 | /* |
1344 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1345 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1346 | * see whether the page bit testing has already | |
1347 | * been done by the wake function. | |
2a9127fc | 1348 | * |
101c0bf6 | 1349 | * We can drop our reference to the folio. |
2a9127fc LT |
1350 | */ |
1351 | if (behavior == DROP) | |
101c0bf6 | 1352 | folio_put(folio); |
62906027 | 1353 | |
5ef64cc8 LT |
1354 | /* |
1355 | * Note that until the "finish_wait()", or until | |
1356 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1357 | * be very careful with the 'wait->flags', because | |
1358 | * we may race with a waker that sets them. | |
1359 | */ | |
2a9127fc | 1360 | for (;;) { |
5ef64cc8 LT |
1361 | unsigned int flags; |
1362 | ||
62906027 NP |
1363 | set_current_state(state); |
1364 | ||
5ef64cc8 LT |
1365 | /* Loop until we've been woken or interrupted */ |
1366 | flags = smp_load_acquire(&wait->flags); | |
1367 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1368 | if (signal_pending_state(state, current)) | |
1369 | break; | |
1370 | ||
1371 | io_schedule(); | |
1372 | continue; | |
1373 | } | |
1374 | ||
1375 | /* If we were non-exclusive, we're done */ | |
1376 | if (behavior != EXCLUSIVE) | |
a8b169af | 1377 | break; |
9a1ea439 | 1378 | |
5ef64cc8 LT |
1379 | /* If the waker got the lock for us, we're done */ |
1380 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1381 | break; |
2a9127fc | 1382 | |
5ef64cc8 LT |
1383 | /* |
1384 | * Otherwise, if we're getting the lock, we need to | |
1385 | * try to get it ourselves. | |
1386 | * | |
1387 | * And if that fails, we'll have to retry this all. | |
1388 | */ | |
101c0bf6 | 1389 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
5ef64cc8 LT |
1390 | goto repeat; |
1391 | ||
1392 | wait->flags |= WQ_FLAG_DONE; | |
1393 | break; | |
62906027 NP |
1394 | } |
1395 | ||
5ef64cc8 LT |
1396 | /* |
1397 | * If a signal happened, this 'finish_wait()' may remove the last | |
101c0bf6 | 1398 | * waiter from the wait-queues, but the folio waiters bit will remain |
5ef64cc8 LT |
1399 | * set. That's ok. The next wakeup will take care of it, and trying |
1400 | * to do it here would be difficult and prone to races. | |
1401 | */ | |
62906027 NP |
1402 | finish_wait(q, wait); |
1403 | ||
eb414681 | 1404 | if (thrashing) { |
9a1ea439 | 1405 | if (delayacct) |
eb414681 JW |
1406 | delayacct_thrashing_end(); |
1407 | psi_memstall_leave(&pflags); | |
1408 | } | |
b1d29ba8 | 1409 | |
62906027 | 1410 | /* |
5ef64cc8 LT |
1411 | * NOTE! The wait->flags weren't stable until we've done the |
1412 | * 'finish_wait()', and we could have exited the loop above due | |
1413 | * to a signal, and had a wakeup event happen after the signal | |
1414 | * test but before the 'finish_wait()'. | |
1415 | * | |
1416 | * So only after the finish_wait() can we reliably determine | |
1417 | * if we got woken up or not, so we can now figure out the final | |
1418 | * return value based on that state without races. | |
1419 | * | |
1420 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1421 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1422 | */ |
5ef64cc8 LT |
1423 | if (behavior == EXCLUSIVE) |
1424 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1425 | |
2a9127fc | 1426 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1427 | } |
1428 | ||
101c0bf6 | 1429 | void folio_wait_bit(struct folio *folio, int bit_nr) |
62906027 | 1430 | { |
101c0bf6 | 1431 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 | 1432 | } |
101c0bf6 | 1433 | EXPORT_SYMBOL(folio_wait_bit); |
62906027 | 1434 | |
101c0bf6 | 1435 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
62906027 | 1436 | { |
101c0bf6 | 1437 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1438 | } |
101c0bf6 | 1439 | EXPORT_SYMBOL(folio_wait_bit_killable); |
cbbce822 | 1440 | |
9a1ea439 | 1441 | /** |
9f2b04a2 MWO |
1442 | * folio_put_wait_locked - Drop a reference and wait for it to be unlocked |
1443 | * @folio: The folio to wait for. | |
48054625 | 1444 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 | 1445 | * |
9f2b04a2 | 1446 | * The caller should hold a reference on @folio. They expect the page to |
9a1ea439 | 1447 | * become unlocked relatively soon, but do not wish to hold up migration |
9f2b04a2 | 1448 | * (for example) by holding the reference while waiting for the folio to |
9a1ea439 | 1449 | * come unlocked. After this function returns, the caller should not |
9f2b04a2 | 1450 | * dereference @folio. |
48054625 | 1451 | * |
9f2b04a2 | 1452 | * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal. |
9a1ea439 | 1453 | */ |
9f2b04a2 | 1454 | int folio_put_wait_locked(struct folio *folio, int state) |
9a1ea439 | 1455 | { |
9f2b04a2 | 1456 | return folio_wait_bit_common(folio, PG_locked, state, DROP); |
9a1ea439 HD |
1457 | } |
1458 | ||
385e1ca5 | 1459 | /** |
df4d4f12 MWO |
1460 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
1461 | * @folio: Folio defining the wait queue of interest | |
697f619f | 1462 | * @waiter: Waiter to add to the queue |
385e1ca5 | 1463 | * |
df4d4f12 | 1464 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
385e1ca5 | 1465 | */ |
df4d4f12 | 1466 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
385e1ca5 | 1467 | { |
df4d4f12 | 1468 | wait_queue_head_t *q = folio_waitqueue(folio); |
385e1ca5 DH |
1469 | unsigned long flags; |
1470 | ||
1471 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1472 | __add_wait_queue_entry_tail(q, waiter); |
df4d4f12 | 1473 | folio_set_waiters(folio); |
385e1ca5 DH |
1474 | spin_unlock_irqrestore(&q->lock, flags); |
1475 | } | |
df4d4f12 | 1476 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
385e1ca5 | 1477 | |
b91e1302 LT |
1478 | #ifndef clear_bit_unlock_is_negative_byte |
1479 | ||
1480 | /* | |
1481 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1482 | * | |
1483 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1484 | * test the sign bit at the same time. But if the architecture does | |
1485 | * not support that special operation, we just do this all by hand | |
1486 | * instead. | |
1487 | * | |
1488 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1489 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1490 | * in the same byte as PG_locked. |
1491 | */ | |
1492 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1493 | { | |
1494 | clear_bit_unlock(nr, mem); | |
1495 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1496 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1497 | } |
1498 | ||
1499 | #endif | |
1500 | ||
1da177e4 | 1501 | /** |
4e136428 MWO |
1502 | * folio_unlock - Unlock a locked folio. |
1503 | * @folio: The folio. | |
1504 | * | |
1505 | * Unlocks the folio and wakes up any thread sleeping on the page lock. | |
1506 | * | |
1507 | * Context: May be called from interrupt or process context. May not be | |
1508 | * called from NMI context. | |
1da177e4 | 1509 | */ |
4e136428 | 1510 | void folio_unlock(struct folio *folio) |
1da177e4 | 1511 | { |
4e136428 | 1512 | /* Bit 7 allows x86 to check the byte's sign bit */ |
b91e1302 | 1513 | BUILD_BUG_ON(PG_waiters != 7); |
4e136428 MWO |
1514 | BUILD_BUG_ON(PG_locked > 7); |
1515 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1516 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) | |
6974d7c9 | 1517 | folio_wake_bit(folio, PG_locked); |
1da177e4 | 1518 | } |
4e136428 | 1519 | EXPORT_SYMBOL(folio_unlock); |
1da177e4 | 1520 | |
73e10ded | 1521 | /** |
b47393f8 MWO |
1522 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
1523 | * @folio: The folio. | |
73e10ded | 1524 | * |
b47393f8 MWO |
1525 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
1526 | * it. The folio reference held for PG_private_2 being set is released. | |
73e10ded | 1527 | * |
b47393f8 MWO |
1528 | * This is, for example, used when a netfs folio is being written to a local |
1529 | * disk cache, thereby allowing writes to the cache for the same folio to be | |
73e10ded DH |
1530 | * serialised. |
1531 | */ | |
b47393f8 | 1532 | void folio_end_private_2(struct folio *folio) |
73e10ded | 1533 | { |
6974d7c9 MWO |
1534 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
1535 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); | |
1536 | folio_wake_bit(folio, PG_private_2); | |
1537 | folio_put(folio); | |
73e10ded | 1538 | } |
b47393f8 | 1539 | EXPORT_SYMBOL(folio_end_private_2); |
73e10ded DH |
1540 | |
1541 | /** | |
b47393f8 MWO |
1542 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
1543 | * @folio: The folio to wait on. | |
73e10ded | 1544 | * |
b47393f8 | 1545 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
73e10ded | 1546 | */ |
b47393f8 | 1547 | void folio_wait_private_2(struct folio *folio) |
73e10ded | 1548 | { |
101c0bf6 MWO |
1549 | while (folio_test_private_2(folio)) |
1550 | folio_wait_bit(folio, PG_private_2); | |
73e10ded | 1551 | } |
b47393f8 | 1552 | EXPORT_SYMBOL(folio_wait_private_2); |
73e10ded DH |
1553 | |
1554 | /** | |
b47393f8 MWO |
1555 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. |
1556 | * @folio: The folio to wait on. | |
73e10ded | 1557 | * |
b47393f8 | 1558 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a |
73e10ded DH |
1559 | * fatal signal is received by the calling task. |
1560 | * | |
1561 | * Return: | |
1562 | * - 0 if successful. | |
1563 | * - -EINTR if a fatal signal was encountered. | |
1564 | */ | |
b47393f8 | 1565 | int folio_wait_private_2_killable(struct folio *folio) |
73e10ded DH |
1566 | { |
1567 | int ret = 0; | |
1568 | ||
101c0bf6 MWO |
1569 | while (folio_test_private_2(folio)) { |
1570 | ret = folio_wait_bit_killable(folio, PG_private_2); | |
73e10ded DH |
1571 | if (ret < 0) |
1572 | break; | |
1573 | } | |
1574 | ||
1575 | return ret; | |
1576 | } | |
b47393f8 | 1577 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
73e10ded | 1578 | |
485bb99b | 1579 | /** |
4268b480 MWO |
1580 | * folio_end_writeback - End writeback against a folio. |
1581 | * @folio: The folio. | |
1da177e4 | 1582 | */ |
4268b480 | 1583 | void folio_end_writeback(struct folio *folio) |
1da177e4 | 1584 | { |
888cf2db | 1585 | /* |
4268b480 MWO |
1586 | * folio_test_clear_reclaim() could be used here but it is an |
1587 | * atomic operation and overkill in this particular case. Failing | |
1588 | * to shuffle a folio marked for immediate reclaim is too mild | |
1589 | * a gain to justify taking an atomic operation penalty at the | |
1590 | * end of every folio writeback. | |
888cf2db | 1591 | */ |
4268b480 MWO |
1592 | if (folio_test_reclaim(folio)) { |
1593 | folio_clear_reclaim(folio); | |
575ced1c | 1594 | folio_rotate_reclaimable(folio); |
888cf2db | 1595 | } |
ac6aadb2 | 1596 | |
073861ed | 1597 | /* |
4268b480 | 1598 | * Writeback does not hold a folio reference of its own, relying |
073861ed | 1599 | * on truncation to wait for the clearing of PG_writeback. |
4268b480 MWO |
1600 | * But here we must make sure that the folio is not freed and |
1601 | * reused before the folio_wake(). | |
073861ed | 1602 | */ |
4268b480 | 1603 | folio_get(folio); |
269ccca3 | 1604 | if (!__folio_end_writeback(folio)) |
ac6aadb2 MS |
1605 | BUG(); |
1606 | ||
4e857c58 | 1607 | smp_mb__after_atomic(); |
4268b480 | 1608 | folio_wake(folio, PG_writeback); |
512b7931 | 1609 | acct_reclaim_writeback(folio); |
4268b480 | 1610 | folio_put(folio); |
1da177e4 | 1611 | } |
4268b480 | 1612 | EXPORT_SYMBOL(folio_end_writeback); |
1da177e4 | 1613 | |
57d99845 MW |
1614 | /* |
1615 | * After completing I/O on a page, call this routine to update the page | |
1616 | * flags appropriately | |
1617 | */ | |
c11f0c0b | 1618 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1619 | { |
c11f0c0b | 1620 | if (!is_write) { |
57d99845 MW |
1621 | if (!err) { |
1622 | SetPageUptodate(page); | |
1623 | } else { | |
1624 | ClearPageUptodate(page); | |
1625 | SetPageError(page); | |
1626 | } | |
1627 | unlock_page(page); | |
abf54548 | 1628 | } else { |
57d99845 | 1629 | if (err) { |
dd8416c4 MK |
1630 | struct address_space *mapping; |
1631 | ||
57d99845 | 1632 | SetPageError(page); |
dd8416c4 MK |
1633 | mapping = page_mapping(page); |
1634 | if (mapping) | |
1635 | mapping_set_error(mapping, err); | |
57d99845 MW |
1636 | } |
1637 | end_page_writeback(page); | |
1638 | } | |
1639 | } | |
1640 | EXPORT_SYMBOL_GPL(page_endio); | |
1641 | ||
485bb99b | 1642 | /** |
7c23c782 MWO |
1643 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
1644 | * @folio: The folio to lock | |
1da177e4 | 1645 | */ |
7c23c782 | 1646 | void __folio_lock(struct folio *folio) |
1da177e4 | 1647 | { |
101c0bf6 | 1648 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
9a1ea439 | 1649 | EXCLUSIVE); |
1da177e4 | 1650 | } |
7c23c782 | 1651 | EXPORT_SYMBOL(__folio_lock); |
1da177e4 | 1652 | |
af7f29d9 | 1653 | int __folio_lock_killable(struct folio *folio) |
2687a356 | 1654 | { |
101c0bf6 | 1655 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
9a1ea439 | 1656 | EXCLUSIVE); |
2687a356 | 1657 | } |
af7f29d9 | 1658 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
2687a356 | 1659 | |
ffdc8dab | 1660 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
dd3e6d50 | 1661 | { |
df4d4f12 | 1662 | struct wait_queue_head *q = folio_waitqueue(folio); |
f32b5dd7 MWO |
1663 | int ret = 0; |
1664 | ||
df4d4f12 | 1665 | wait->folio = folio; |
f32b5dd7 MWO |
1666 | wait->bit_nr = PG_locked; |
1667 | ||
1668 | spin_lock_irq(&q->lock); | |
1669 | __add_wait_queue_entry_tail(q, &wait->wait); | |
ffdc8dab MWO |
1670 | folio_set_waiters(folio); |
1671 | ret = !folio_trylock(folio); | |
f32b5dd7 MWO |
1672 | /* |
1673 | * If we were successful now, we know we're still on the | |
1674 | * waitqueue as we're still under the lock. This means it's | |
1675 | * safe to remove and return success, we know the callback | |
1676 | * isn't going to trigger. | |
1677 | */ | |
1678 | if (!ret) | |
1679 | __remove_wait_queue(q, &wait->wait); | |
1680 | else | |
1681 | ret = -EIOCBQUEUED; | |
1682 | spin_unlock_irq(&q->lock); | |
1683 | return ret; | |
dd3e6d50 JA |
1684 | } |
1685 | ||
9a95f3cf PC |
1686 | /* |
1687 | * Return values: | |
9138e47e MWO |
1688 | * true - folio is locked; mmap_lock is still held. |
1689 | * false - folio is not locked. | |
3e4e28c5 | 1690 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1691 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1692 | * which case mmap_lock is still held. |
9a95f3cf | 1693 | * |
9138e47e MWO |
1694 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
1695 | * with the folio locked and the mmap_lock unperturbed. | |
9a95f3cf | 1696 | */ |
9138e47e | 1697 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
d065bd81 ML |
1698 | unsigned int flags) |
1699 | { | |
4064b982 | 1700 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1701 | /* |
c1e8d7c6 | 1702 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1703 | * even though return 0. |
1704 | */ | |
1705 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
9138e47e | 1706 | return false; |
37b23e05 | 1707 | |
d8ed45c5 | 1708 | mmap_read_unlock(mm); |
37b23e05 | 1709 | if (flags & FAULT_FLAG_KILLABLE) |
6baa8d60 | 1710 | folio_wait_locked_killable(folio); |
37b23e05 | 1711 | else |
6baa8d60 | 1712 | folio_wait_locked(folio); |
9138e47e | 1713 | return false; |
800bca7c HL |
1714 | } |
1715 | if (flags & FAULT_FLAG_KILLABLE) { | |
9138e47e | 1716 | bool ret; |
37b23e05 | 1717 | |
af7f29d9 | 1718 | ret = __folio_lock_killable(folio); |
800bca7c HL |
1719 | if (ret) { |
1720 | mmap_read_unlock(mm); | |
9138e47e | 1721 | return false; |
800bca7c HL |
1722 | } |
1723 | } else { | |
af7f29d9 | 1724 | __folio_lock(folio); |
d065bd81 | 1725 | } |
800bca7c | 1726 | |
9138e47e | 1727 | return true; |
d065bd81 ML |
1728 | } |
1729 | ||
e7b563bb | 1730 | /** |
0d3f9296 MW |
1731 | * page_cache_next_miss() - Find the next gap in the page cache. |
1732 | * @mapping: Mapping. | |
1733 | * @index: Index. | |
1734 | * @max_scan: Maximum range to search. | |
e7b563bb | 1735 | * |
0d3f9296 MW |
1736 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1737 | * gap with the lowest index. | |
e7b563bb | 1738 | * |
0d3f9296 MW |
1739 | * This function may be called under the rcu_read_lock. However, this will |
1740 | * not atomically search a snapshot of the cache at a single point in time. | |
1741 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1742 | * created at index 10, page_cache_next_miss covering both indices may | |
1743 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1744 | * |
0d3f9296 MW |
1745 | * Return: The index of the gap if found, otherwise an index outside the |
1746 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1747 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1748 | */ |
0d3f9296 | 1749 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1750 | pgoff_t index, unsigned long max_scan) |
1751 | { | |
0d3f9296 | 1752 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1753 | |
0d3f9296 MW |
1754 | while (max_scan--) { |
1755 | void *entry = xas_next(&xas); | |
1756 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1757 | break; |
0d3f9296 | 1758 | if (xas.xa_index == 0) |
e7b563bb JW |
1759 | break; |
1760 | } | |
1761 | ||
0d3f9296 | 1762 | return xas.xa_index; |
e7b563bb | 1763 | } |
0d3f9296 | 1764 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1765 | |
1766 | /** | |
2346a560 | 1767 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1768 | * @mapping: Mapping. |
1769 | * @index: Index. | |
1770 | * @max_scan: Maximum range to search. | |
e7b563bb | 1771 | * |
0d3f9296 MW |
1772 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1773 | * gap with the highest index. | |
e7b563bb | 1774 | * |
0d3f9296 MW |
1775 | * This function may be called under the rcu_read_lock. However, this will |
1776 | * not atomically search a snapshot of the cache at a single point in time. | |
1777 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1778 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1779 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1780 | * |
0d3f9296 MW |
1781 | * Return: The index of the gap if found, otherwise an index outside the |
1782 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1783 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1784 | */ |
0d3f9296 | 1785 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1786 | pgoff_t index, unsigned long max_scan) |
1787 | { | |
0d3f9296 | 1788 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1789 | |
0d3f9296 MW |
1790 | while (max_scan--) { |
1791 | void *entry = xas_prev(&xas); | |
1792 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1793 | break; |
0d3f9296 | 1794 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1795 | break; |
1796 | } | |
1797 | ||
0d3f9296 | 1798 | return xas.xa_index; |
e7b563bb | 1799 | } |
0d3f9296 | 1800 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1801 | |
020853b6 MWO |
1802 | /* |
1803 | * Lockless page cache protocol: | |
1804 | * On the lookup side: | |
1805 | * 1. Load the folio from i_pages | |
1806 | * 2. Increment the refcount if it's not zero | |
1807 | * 3. If the folio is not found by xas_reload(), put the refcount and retry | |
1808 | * | |
1809 | * On the removal side: | |
1810 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) | |
1811 | * B. Remove the page from i_pages | |
1812 | * C. Return the page to the page allocator | |
1813 | * | |
1814 | * This means that any page may have its reference count temporarily | |
1815 | * increased by a speculative page cache (or fast GUP) lookup as it can | |
1816 | * be allocated by another user before the RCU grace period expires. | |
1817 | * Because the refcount temporarily acquired here may end up being the | |
1818 | * last refcount on the page, any page allocation must be freeable by | |
1819 | * folio_put(). | |
1820 | */ | |
1821 | ||
44835d20 | 1822 | /* |
bc5a3011 | 1823 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1824 | * @mapping: the address_space to search |
a6de4b48 | 1825 | * @index: The page cache index. |
0cd6144a | 1826 | * |
bca65eea MWO |
1827 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
1828 | * it is returned with an increased refcount. If it is a shadow entry | |
1829 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, | |
1830 | * it is returned without further action. | |
485bb99b | 1831 | * |
bca65eea | 1832 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
1da177e4 | 1833 | */ |
bca65eea | 1834 | static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1835 | { |
a6de4b48 | 1836 | XA_STATE(xas, &mapping->i_pages, index); |
bca65eea | 1837 | struct folio *folio; |
1da177e4 | 1838 | |
a60637c8 NP |
1839 | rcu_read_lock(); |
1840 | repeat: | |
4c7472c0 | 1841 | xas_reset(&xas); |
bca65eea MWO |
1842 | folio = xas_load(&xas); |
1843 | if (xas_retry(&xas, folio)) | |
4c7472c0 MW |
1844 | goto repeat; |
1845 | /* | |
1846 | * A shadow entry of a recently evicted page, or a swap entry from | |
1847 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1848 | */ | |
bca65eea | 1849 | if (!folio || xa_is_value(folio)) |
4c7472c0 | 1850 | goto out; |
83929372 | 1851 | |
bca65eea | 1852 | if (!folio_try_get_rcu(folio)) |
4c7472c0 | 1853 | goto repeat; |
83929372 | 1854 | |
bca65eea MWO |
1855 | if (unlikely(folio != xas_reload(&xas))) { |
1856 | folio_put(folio); | |
4c7472c0 | 1857 | goto repeat; |
a60637c8 | 1858 | } |
27d20fdd | 1859 | out: |
a60637c8 NP |
1860 | rcu_read_unlock(); |
1861 | ||
bca65eea | 1862 | return folio; |
1da177e4 | 1863 | } |
1da177e4 | 1864 | |
0cd6144a | 1865 | /** |
3f0c6a07 | 1866 | * __filemap_get_folio - Find and get a reference to a folio. |
2294b32e MWO |
1867 | * @mapping: The address_space to search. |
1868 | * @index: The page index. | |
3f0c6a07 MWO |
1869 | * @fgp_flags: %FGP flags modify how the folio is returned. |
1870 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1871 | * |
2294b32e | 1872 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1873 | * |
2294b32e | 1874 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1875 | * |
3f0c6a07 MWO |
1876 | * * %FGP_ACCESSED - The folio will be marked accessed. |
1877 | * * %FGP_LOCK - The folio is returned locked. | |
44835d20 | 1878 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
3f0c6a07 | 1879 | * instead of allocating a new folio to replace it. |
2294b32e | 1880 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
3f0c6a07 | 1881 | * @gfp and added to the page cache and the VM's LRU list. |
2294b32e MWO |
1882 | * The page is returned locked and with an increased refcount. |
1883 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1884 | * page is already in cache. If the page was allocated, unlock it before | |
1885 | * returning so the caller can do the same dance. | |
3f0c6a07 MWO |
1886 | * * %FGP_WRITE - The page will be written to by the caller. |
1887 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp. | |
1888 | * * %FGP_NOWAIT - Don't get blocked by page lock. | |
b27652d9 | 1889 | * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) |
1da177e4 | 1890 | * |
2294b32e MWO |
1891 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1892 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1893 | * |
2457aec6 | 1894 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1895 | * |
3f0c6a07 | 1896 | * Return: The found folio or %NULL otherwise. |
1da177e4 | 1897 | */ |
3f0c6a07 MWO |
1898 | struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, |
1899 | int fgp_flags, gfp_t gfp) | |
1da177e4 | 1900 | { |
3f0c6a07 | 1901 | struct folio *folio; |
2457aec6 | 1902 | |
1da177e4 | 1903 | repeat: |
3f0c6a07 MWO |
1904 | folio = mapping_get_entry(mapping, index); |
1905 | if (xa_is_value(folio)) { | |
44835d20 | 1906 | if (fgp_flags & FGP_ENTRY) |
3f0c6a07 MWO |
1907 | return folio; |
1908 | folio = NULL; | |
44835d20 | 1909 | } |
3f0c6a07 | 1910 | if (!folio) |
2457aec6 MG |
1911 | goto no_page; |
1912 | ||
1913 | if (fgp_flags & FGP_LOCK) { | |
1914 | if (fgp_flags & FGP_NOWAIT) { | |
3f0c6a07 MWO |
1915 | if (!folio_trylock(folio)) { |
1916 | folio_put(folio); | |
2457aec6 MG |
1917 | return NULL; |
1918 | } | |
1919 | } else { | |
3f0c6a07 | 1920 | folio_lock(folio); |
2457aec6 MG |
1921 | } |
1922 | ||
1923 | /* Has the page been truncated? */ | |
3f0c6a07 MWO |
1924 | if (unlikely(folio->mapping != mapping)) { |
1925 | folio_unlock(folio); | |
1926 | folio_put(folio); | |
2457aec6 MG |
1927 | goto repeat; |
1928 | } | |
3f0c6a07 | 1929 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
2457aec6 MG |
1930 | } |
1931 | ||
c16eb000 | 1932 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1933 | folio_mark_accessed(folio); |
b9306a79 YS |
1934 | else if (fgp_flags & FGP_WRITE) { |
1935 | /* Clear idle flag for buffer write */ | |
3f0c6a07 MWO |
1936 | if (folio_test_idle(folio)) |
1937 | folio_clear_idle(folio); | |
b9306a79 | 1938 | } |
2457aec6 | 1939 | |
b27652d9 MWO |
1940 | if (fgp_flags & FGP_STABLE) |
1941 | folio_wait_stable(folio); | |
2457aec6 | 1942 | no_page: |
3f0c6a07 | 1943 | if (!folio && (fgp_flags & FGP_CREAT)) { |
2457aec6 | 1944 | int err; |
f56753ac | 1945 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
3f0c6a07 | 1946 | gfp |= __GFP_WRITE; |
45f87de5 | 1947 | if (fgp_flags & FGP_NOFS) |
3f0c6a07 | 1948 | gfp &= ~__GFP_FS; |
2457aec6 | 1949 | |
3f0c6a07 MWO |
1950 | folio = filemap_alloc_folio(gfp, 0); |
1951 | if (!folio) | |
eb2be189 | 1952 | return NULL; |
2457aec6 | 1953 | |
a75d4c33 | 1954 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1955 | fgp_flags |= FGP_LOCK; |
1956 | ||
eb39d618 | 1957 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1958 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1959 | __folio_set_referenced(folio); |
2457aec6 | 1960 | |
3f0c6a07 | 1961 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 1962 | if (unlikely(err)) { |
3f0c6a07 MWO |
1963 | folio_put(folio); |
1964 | folio = NULL; | |
eb2be189 NP |
1965 | if (err == -EEXIST) |
1966 | goto repeat; | |
1da177e4 | 1967 | } |
a75d4c33 JB |
1968 | |
1969 | /* | |
3f0c6a07 MWO |
1970 | * filemap_add_folio locks the page, and for mmap |
1971 | * we expect an unlocked page. | |
a75d4c33 | 1972 | */ |
3f0c6a07 MWO |
1973 | if (folio && (fgp_flags & FGP_FOR_MMAP)) |
1974 | folio_unlock(folio); | |
1da177e4 | 1975 | } |
2457aec6 | 1976 | |
3f0c6a07 | 1977 | return folio; |
1da177e4 | 1978 | } |
3f0c6a07 | 1979 | EXPORT_SYMBOL(__filemap_get_folio); |
1da177e4 | 1980 | |
c7bad633 MWO |
1981 | static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max, |
1982 | xa_mark_t mark) | |
1983 | { | |
1984 | struct page *page; | |
1985 | ||
1986 | retry: | |
1987 | if (mark == XA_PRESENT) | |
1988 | page = xas_find(xas, max); | |
1989 | else | |
1990 | page = xas_find_marked(xas, max, mark); | |
1991 | ||
1992 | if (xas_retry(xas, page)) | |
1993 | goto retry; | |
1994 | /* | |
1995 | * A shadow entry of a recently evicted page, a swap | |
1996 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1997 | * without attempting to raise page count. | |
1998 | */ | |
1999 | if (!page || xa_is_value(page)) | |
2000 | return page; | |
2001 | ||
2002 | if (!page_cache_get_speculative(page)) | |
2003 | goto reset; | |
2004 | ||
2005 | /* Has the page moved or been split? */ | |
2006 | if (unlikely(page != xas_reload(xas))) { | |
2007 | put_page(page); | |
2008 | goto reset; | |
2009 | } | |
2010 | ||
2011 | return page; | |
2012 | reset: | |
2013 | xas_reset(xas); | |
2014 | goto retry; | |
2015 | } | |
2016 | ||
0cd6144a JW |
2017 | /** |
2018 | * find_get_entries - gang pagecache lookup | |
2019 | * @mapping: The address_space to search | |
2020 | * @start: The starting page cache index | |
ca122fe4 | 2021 | * @end: The final page index (inclusive). |
cf2039af | 2022 | * @pvec: Where the resulting entries are placed. |
0cd6144a JW |
2023 | * @indices: The cache indices corresponding to the entries in @entries |
2024 | * | |
cf2039af MWO |
2025 | * find_get_entries() will search for and return a batch of entries in |
2026 | * the mapping. The entries are placed in @pvec. find_get_entries() | |
2027 | * takes a reference on any actual pages it returns. | |
0cd6144a JW |
2028 | * |
2029 | * The search returns a group of mapping-contiguous page cache entries | |
2030 | * with ascending indexes. There may be holes in the indices due to | |
2031 | * not-present pages. | |
2032 | * | |
139b6a6f JW |
2033 | * Any shadow entries of evicted pages, or swap entries from |
2034 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 2035 | * |
71725ed1 HD |
2036 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
2037 | * stops at that page: the caller is likely to have a better way to handle | |
2038 | * the compound page as a whole, and then skip its extent, than repeatedly | |
2039 | * calling find_get_entries() to return all its tails. | |
2040 | * | |
a862f68a | 2041 | * Return: the number of pages and shadow entries which were found. |
0cd6144a | 2042 | */ |
ca122fe4 | 2043 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
cf2039af | 2044 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) |
0cd6144a | 2045 | { |
f280bf09 MW |
2046 | XA_STATE(xas, &mapping->i_pages, start); |
2047 | struct page *page; | |
0cd6144a | 2048 | unsigned int ret = 0; |
cf2039af | 2049 | unsigned nr_entries = PAGEVEC_SIZE; |
0cd6144a JW |
2050 | |
2051 | rcu_read_lock(); | |
ca122fe4 | 2052 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
71725ed1 HD |
2053 | /* |
2054 | * Terminate early on finding a THP, to allow the caller to | |
2055 | * handle it all at once; but continue if this is hugetlbfs. | |
2056 | */ | |
c7bad633 MWO |
2057 | if (!xa_is_value(page) && PageTransHuge(page) && |
2058 | !PageHuge(page)) { | |
71725ed1 HD |
2059 | page = find_subpage(page, xas.xa_index); |
2060 | nr_entries = ret + 1; | |
2061 | } | |
c7bad633 | 2062 | |
f280bf09 | 2063 | indices[ret] = xas.xa_index; |
cf2039af | 2064 | pvec->pages[ret] = page; |
0cd6144a JW |
2065 | if (++ret == nr_entries) |
2066 | break; | |
2067 | } | |
2068 | rcu_read_unlock(); | |
cf2039af MWO |
2069 | |
2070 | pvec->nr = ret; | |
0cd6144a JW |
2071 | return ret; |
2072 | } | |
2073 | ||
5c211ba2 MWO |
2074 | /** |
2075 | * find_lock_entries - Find a batch of pagecache entries. | |
2076 | * @mapping: The address_space to search. | |
2077 | * @start: The starting page cache index. | |
2078 | * @end: The final page index (inclusive). | |
2079 | * @pvec: Where the resulting entries are placed. | |
2080 | * @indices: The cache indices of the entries in @pvec. | |
2081 | * | |
2082 | * find_lock_entries() will return a batch of entries from @mapping. | |
2083 | * Swap, shadow and DAX entries are included. Pages are returned | |
2084 | * locked and with an incremented refcount. Pages which are locked by | |
2085 | * somebody else or under writeback are skipped. Only the head page of | |
2086 | * a THP is returned. Pages which are partially outside the range are | |
2087 | * not returned. | |
2088 | * | |
2089 | * The entries have ascending indexes. The indices may not be consecutive | |
2090 | * due to not-present entries, THP pages, pages which could not be locked | |
2091 | * or pages under writeback. | |
2092 | * | |
2093 | * Return: The number of entries which were found. | |
2094 | */ | |
2095 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
2096 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) | |
2097 | { | |
2098 | XA_STATE(xas, &mapping->i_pages, start); | |
2099 | struct page *page; | |
2100 | ||
2101 | rcu_read_lock(); | |
2102 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { | |
2103 | if (!xa_is_value(page)) { | |
2104 | if (page->index < start) | |
2105 | goto put; | |
5c211ba2 MWO |
2106 | if (page->index + thp_nr_pages(page) - 1 > end) |
2107 | goto put; | |
2108 | if (!trylock_page(page)) | |
2109 | goto put; | |
2110 | if (page->mapping != mapping || PageWriteback(page)) | |
2111 | goto unlock; | |
2112 | VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index), | |
2113 | page); | |
2114 | } | |
2115 | indices[pvec->nr] = xas.xa_index; | |
2116 | if (!pagevec_add(pvec, page)) | |
2117 | break; | |
2118 | goto next; | |
2119 | unlock: | |
2120 | unlock_page(page); | |
2121 | put: | |
2122 | put_page(page); | |
2123 | next: | |
2d11e738 HD |
2124 | if (!xa_is_value(page) && PageTransHuge(page)) { |
2125 | unsigned int nr_pages = thp_nr_pages(page); | |
2126 | ||
2127 | /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */ | |
2128 | xas_set(&xas, page->index + nr_pages); | |
2129 | if (xas.xa_index < nr_pages) | |
2130 | break; | |
2131 | } | |
5c211ba2 MWO |
2132 | } |
2133 | rcu_read_unlock(); | |
2134 | ||
2135 | return pagevec_count(pvec); | |
2136 | } | |
2137 | ||
1da177e4 | 2138 | /** |
b947cee4 | 2139 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2140 | * @mapping: The address_space to search |
2141 | * @start: The starting page index | |
b947cee4 | 2142 | * @end: The final page index (inclusive) |
1da177e4 LT |
2143 | * @nr_pages: The maximum number of pages |
2144 | * @pages: Where the resulting pages are placed | |
2145 | * | |
b947cee4 JK |
2146 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2147 | * pages in the mapping starting at index @start and up to index @end | |
2148 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2149 | * a reference against the returned pages. | |
1da177e4 LT |
2150 | * |
2151 | * The search returns a group of mapping-contiguous pages with ascending | |
2152 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2153 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2154 | * |
a862f68a MR |
2155 | * Return: the number of pages which were found. If this number is |
2156 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2157 | * reached. |
1da177e4 | 2158 | */ |
b947cee4 JK |
2159 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2160 | pgoff_t end, unsigned int nr_pages, | |
2161 | struct page **pages) | |
1da177e4 | 2162 | { |
fd1b3cee MW |
2163 | XA_STATE(xas, &mapping->i_pages, *start); |
2164 | struct page *page; | |
0fc9d104 KK |
2165 | unsigned ret = 0; |
2166 | ||
2167 | if (unlikely(!nr_pages)) | |
2168 | return 0; | |
a60637c8 NP |
2169 | |
2170 | rcu_read_lock(); | |
c7bad633 | 2171 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee MW |
2172 | /* Skip over shadow, swap and DAX entries */ |
2173 | if (xa_is_value(page)) | |
8079b1c8 | 2174 | continue; |
a60637c8 | 2175 | |
4101196b | 2176 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 2177 | if (++ret == nr_pages) { |
5d3ee42f | 2178 | *start = xas.xa_index + 1; |
b947cee4 JK |
2179 | goto out; |
2180 | } | |
a60637c8 | 2181 | } |
5b280c0c | 2182 | |
b947cee4 JK |
2183 | /* |
2184 | * We come here when there is no page beyond @end. We take care to not | |
2185 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2186 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2187 | * already broken anyway. |
2188 | */ | |
2189 | if (end == (pgoff_t)-1) | |
2190 | *start = (pgoff_t)-1; | |
2191 | else | |
2192 | *start = end + 1; | |
2193 | out: | |
a60637c8 | 2194 | rcu_read_unlock(); |
d72dc8a2 | 2195 | |
1da177e4 LT |
2196 | return ret; |
2197 | } | |
2198 | ||
ebf43500 JA |
2199 | /** |
2200 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2201 | * @mapping: The address_space to search | |
2202 | * @index: The starting page index | |
2203 | * @nr_pages: The maximum number of pages | |
2204 | * @pages: Where the resulting pages are placed | |
2205 | * | |
2206 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2207 | * that the returned number of pages are guaranteed to be contiguous. | |
2208 | * | |
a862f68a | 2209 | * Return: the number of pages which were found. |
ebf43500 JA |
2210 | */ |
2211 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2212 | unsigned int nr_pages, struct page **pages) | |
2213 | { | |
3ece58a2 MW |
2214 | XA_STATE(xas, &mapping->i_pages, index); |
2215 | struct page *page; | |
0fc9d104 KK |
2216 | unsigned int ret = 0; |
2217 | ||
2218 | if (unlikely(!nr_pages)) | |
2219 | return 0; | |
a60637c8 NP |
2220 | |
2221 | rcu_read_lock(); | |
3ece58a2 | 2222 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2223 | if (xas_retry(&xas, page)) |
2224 | continue; | |
2225 | /* | |
2226 | * If the entry has been swapped out, we can stop looking. | |
2227 | * No current caller is looking for DAX entries. | |
2228 | */ | |
2229 | if (xa_is_value(page)) | |
8079b1c8 | 2230 | break; |
ebf43500 | 2231 | |
4101196b | 2232 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2233 | goto retry; |
83929372 | 2234 | |
4101196b | 2235 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2236 | if (unlikely(page != xas_reload(&xas))) |
2237 | goto put_page; | |
a60637c8 | 2238 | |
4101196b | 2239 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2240 | if (++ret == nr_pages) |
2241 | break; | |
3ece58a2 MW |
2242 | continue; |
2243 | put_page: | |
4101196b | 2244 | put_page(page); |
3ece58a2 MW |
2245 | retry: |
2246 | xas_reset(&xas); | |
ebf43500 | 2247 | } |
a60637c8 NP |
2248 | rcu_read_unlock(); |
2249 | return ret; | |
ebf43500 | 2250 | } |
ef71c15c | 2251 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2252 | |
485bb99b | 2253 | /** |
c49f50d1 | 2254 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2255 | * @mapping: the address_space to search |
2256 | * @index: the starting page index | |
72b045ae | 2257 | * @end: The final page index (inclusive) |
485bb99b RD |
2258 | * @tag: the tag index |
2259 | * @nr_pages: the maximum number of pages | |
2260 | * @pages: where the resulting pages are placed | |
2261 | * | |
c49f50d1 MWO |
2262 | * Like find_get_pages(), except we only return head pages which are tagged |
2263 | * with @tag. @index is updated to the index immediately after the last | |
2264 | * page we return, ready for the next iteration. | |
a862f68a MR |
2265 | * |
2266 | * Return: the number of pages which were found. | |
1da177e4 | 2267 | */ |
72b045ae | 2268 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2269 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2270 | struct page **pages) |
1da177e4 | 2271 | { |
a6906972 MW |
2272 | XA_STATE(xas, &mapping->i_pages, *index); |
2273 | struct page *page; | |
0fc9d104 KK |
2274 | unsigned ret = 0; |
2275 | ||
2276 | if (unlikely(!nr_pages)) | |
2277 | return 0; | |
a60637c8 NP |
2278 | |
2279 | rcu_read_lock(); | |
c7bad633 | 2280 | while ((page = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2281 | /* |
2282 | * Shadow entries should never be tagged, but this iteration | |
2283 | * is lockless so there is a window for page reclaim to evict | |
2284 | * a page we saw tagged. Skip over it. | |
2285 | */ | |
2286 | if (xa_is_value(page)) | |
139b6a6f | 2287 | continue; |
a60637c8 | 2288 | |
c49f50d1 | 2289 | pages[ret] = page; |
72b045ae | 2290 | if (++ret == nr_pages) { |
c49f50d1 | 2291 | *index = page->index + thp_nr_pages(page); |
72b045ae JK |
2292 | goto out; |
2293 | } | |
a60637c8 | 2294 | } |
5b280c0c | 2295 | |
72b045ae | 2296 | /* |
a6906972 | 2297 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2298 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2299 | * iteration when there is a page at index -1 but that is already |
2300 | * broken anyway. | |
72b045ae JK |
2301 | */ |
2302 | if (end == (pgoff_t)-1) | |
2303 | *index = (pgoff_t)-1; | |
2304 | else | |
2305 | *index = end + 1; | |
2306 | out: | |
a60637c8 | 2307 | rcu_read_unlock(); |
1da177e4 | 2308 | |
1da177e4 LT |
2309 | return ret; |
2310 | } | |
72b045ae | 2311 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2312 | |
76d42bd9 WF |
2313 | /* |
2314 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2315 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2316 | * | |
2317 | * ---R__________________________________________B__________ | |
2318 | * ^ reading here ^ bad block(assume 4k) | |
2319 | * | |
2320 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2321 | * => failing the whole request => read(R) => read(R+1) => | |
2322 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2323 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2324 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2325 | * | |
2326 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2327 | */ | |
0f8e2db4 | 2328 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2329 | { |
76d42bd9 | 2330 | ra->ra_pages /= 4; |
76d42bd9 WF |
2331 | } |
2332 | ||
cbd59c48 MWO |
2333 | /* |
2334 | * filemap_get_read_batch - Get a batch of pages for read | |
2335 | * | |
2336 | * Get a batch of pages which represent a contiguous range of bytes | |
2337 | * in the file. No tail pages will be returned. If @index is in the | |
2338 | * middle of a THP, the entire THP will be returned. The last page in | |
2339 | * the batch may have Readahead set or be not Uptodate so that the | |
2340 | * caller can take the appropriate action. | |
2341 | */ | |
2342 | static void filemap_get_read_batch(struct address_space *mapping, | |
2343 | pgoff_t index, pgoff_t max, struct pagevec *pvec) | |
2344 | { | |
2345 | XA_STATE(xas, &mapping->i_pages, index); | |
2346 | struct page *head; | |
2347 | ||
2348 | rcu_read_lock(); | |
2349 | for (head = xas_load(&xas); head; head = xas_next(&xas)) { | |
2350 | if (xas_retry(&xas, head)) | |
2351 | continue; | |
2352 | if (xas.xa_index > max || xa_is_value(head)) | |
2353 | break; | |
2354 | if (!page_cache_get_speculative(head)) | |
2355 | goto retry; | |
2356 | ||
2357 | /* Has the page moved or been split? */ | |
2358 | if (unlikely(head != xas_reload(&xas))) | |
2359 | goto put_page; | |
2360 | ||
2361 | if (!pagevec_add(pvec, head)) | |
2362 | break; | |
2363 | if (!PageUptodate(head)) | |
2364 | break; | |
2365 | if (PageReadahead(head)) | |
2366 | break; | |
2367 | xas.xa_index = head->index + thp_nr_pages(head) - 1; | |
2368 | xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK; | |
2369 | continue; | |
2370 | put_page: | |
2371 | put_page(head); | |
2372 | retry: | |
2373 | xas_reset(&xas); | |
2374 | } | |
2375 | rcu_read_unlock(); | |
2376 | } | |
2377 | ||
68430303 MWO |
2378 | static int filemap_read_page(struct file *file, struct address_space *mapping, |
2379 | struct page *page) | |
723ef24b | 2380 | { |
723ef24b KO |
2381 | int error; |
2382 | ||
723ef24b | 2383 | /* |
68430303 MWO |
2384 | * A previous I/O error may have been due to temporary failures, |
2385 | * eg. multipath errors. PG_error will be set again if readpage | |
2386 | * fails. | |
723ef24b KO |
2387 | */ |
2388 | ClearPageError(page); | |
2389 | /* Start the actual read. The read will unlock the page. */ | |
68430303 MWO |
2390 | error = mapping->a_ops->readpage(file, page); |
2391 | if (error) | |
2392 | return error; | |
723ef24b | 2393 | |
aa1ec2f6 | 2394 | error = wait_on_page_locked_killable(page); |
68430303 MWO |
2395 | if (error) |
2396 | return error; | |
aa1ec2f6 MWO |
2397 | if (PageUptodate(page)) |
2398 | return 0; | |
aa1ec2f6 MWO |
2399 | shrink_readahead_size_eio(&file->f_ra); |
2400 | return -EIO; | |
723ef24b KO |
2401 | } |
2402 | ||
fce70da3 MWO |
2403 | static bool filemap_range_uptodate(struct address_space *mapping, |
2404 | loff_t pos, struct iov_iter *iter, struct page *page) | |
2405 | { | |
2406 | int count; | |
2407 | ||
2408 | if (PageUptodate(page)) | |
2409 | return true; | |
2410 | /* pipes can't handle partially uptodate pages */ | |
2411 | if (iov_iter_is_pipe(iter)) | |
2412 | return false; | |
2413 | if (!mapping->a_ops->is_partially_uptodate) | |
2414 | return false; | |
2415 | if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page))) | |
2416 | return false; | |
2417 | ||
2418 | count = iter->count; | |
2419 | if (page_offset(page) > pos) { | |
2420 | count -= page_offset(page) - pos; | |
2421 | pos = 0; | |
2422 | } else { | |
2423 | pos -= page_offset(page); | |
2424 | } | |
2425 | ||
2426 | return mapping->a_ops->is_partially_uptodate(page, pos, count); | |
2427 | } | |
2428 | ||
4612aeef MWO |
2429 | static int filemap_update_page(struct kiocb *iocb, |
2430 | struct address_space *mapping, struct iov_iter *iter, | |
fce70da3 | 2431 | struct page *page) |
723ef24b | 2432 | { |
ffdc8dab | 2433 | struct folio *folio = page_folio(page); |
723ef24b KO |
2434 | int error; |
2435 | ||
730633f0 JK |
2436 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2437 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2438 | return -EAGAIN; | |
2439 | } else { | |
2440 | filemap_invalidate_lock_shared(mapping); | |
2441 | } | |
2442 | ||
ffdc8dab | 2443 | if (!folio_trylock(folio)) { |
730633f0 | 2444 | error = -EAGAIN; |
87d1d7b6 | 2445 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2446 | goto unlock_mapping; |
87d1d7b6 | 2447 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2448 | filemap_invalidate_unlock_shared(mapping); |
9f2b04a2 MWO |
2449 | /* |
2450 | * This is where we usually end up waiting for a | |
2451 | * previously submitted readahead to finish. | |
2452 | */ | |
2453 | folio_put_wait_locked(folio, TASK_KILLABLE); | |
4612aeef | 2454 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2455 | } |
ffdc8dab | 2456 | error = __folio_lock_async(folio, iocb->ki_waitq); |
87d1d7b6 | 2457 | if (error) |
730633f0 | 2458 | goto unlock_mapping; |
723ef24b | 2459 | } |
723ef24b | 2460 | |
730633f0 | 2461 | error = AOP_TRUNCATED_PAGE; |
ffdc8dab | 2462 | if (!folio->mapping) |
730633f0 | 2463 | goto unlock; |
723ef24b | 2464 | |
fce70da3 | 2465 | error = 0; |
ffdc8dab | 2466 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, &folio->page)) |
fce70da3 MWO |
2467 | goto unlock; |
2468 | ||
2469 | error = -EAGAIN; | |
2470 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2471 | goto unlock; | |
2472 | ||
ffdc8dab | 2473 | error = filemap_read_page(iocb->ki_filp, mapping, &folio->page); |
730633f0 | 2474 | goto unlock_mapping; |
fce70da3 | 2475 | unlock: |
ffdc8dab | 2476 | folio_unlock(folio); |
730633f0 JK |
2477 | unlock_mapping: |
2478 | filemap_invalidate_unlock_shared(mapping); | |
2479 | if (error == AOP_TRUNCATED_PAGE) | |
ffdc8dab | 2480 | folio_put(folio); |
fce70da3 | 2481 | return error; |
723ef24b KO |
2482 | } |
2483 | ||
f253e185 MWO |
2484 | static int filemap_create_page(struct file *file, |
2485 | struct address_space *mapping, pgoff_t index, | |
2486 | struct pagevec *pvec) | |
723ef24b | 2487 | { |
723ef24b KO |
2488 | struct page *page; |
2489 | int error; | |
2490 | ||
723ef24b KO |
2491 | page = page_cache_alloc(mapping); |
2492 | if (!page) | |
f253e185 | 2493 | return -ENOMEM; |
723ef24b | 2494 | |
730633f0 JK |
2495 | /* |
2496 | * Protect against truncate / hole punch. Grabbing invalidate_lock here | |
2497 | * assures we cannot instantiate and bring uptodate new pagecache pages | |
2498 | * after evicting page cache during truncate and before actually | |
2499 | * freeing blocks. Note that we could release invalidate_lock after | |
2500 | * inserting the page into page cache as the locked page would then be | |
2501 | * enough to synchronize with hole punching. But there are code paths | |
2502 | * such as filemap_update_page() filling in partially uptodate pages or | |
2503 | * ->readpages() that need to hold invalidate_lock while mapping blocks | |
2504 | * for IO so let's hold the lock here as well to keep locking rules | |
2505 | * simple. | |
2506 | */ | |
2507 | filemap_invalidate_lock_shared(mapping); | |
723ef24b | 2508 | error = add_to_page_cache_lru(page, mapping, index, |
f253e185 MWO |
2509 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2510 | if (error == -EEXIST) | |
2511 | error = AOP_TRUNCATED_PAGE; | |
2512 | if (error) | |
2513 | goto error; | |
2514 | ||
2515 | error = filemap_read_page(file, mapping, page); | |
2516 | if (error) | |
2517 | goto error; | |
2518 | ||
730633f0 | 2519 | filemap_invalidate_unlock_shared(mapping); |
f253e185 MWO |
2520 | pagevec_add(pvec, page); |
2521 | return 0; | |
2522 | error: | |
730633f0 | 2523 | filemap_invalidate_unlock_shared(mapping); |
68430303 | 2524 | put_page(page); |
f253e185 | 2525 | return error; |
723ef24b KO |
2526 | } |
2527 | ||
5963fe03 MWO |
2528 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
2529 | struct address_space *mapping, struct page *page, | |
2530 | pgoff_t last_index) | |
2531 | { | |
2532 | if (iocb->ki_flags & IOCB_NOIO) | |
2533 | return -EAGAIN; | |
2534 | page_cache_async_readahead(mapping, &file->f_ra, file, page, | |
2535 | page->index, last_index - page->index); | |
2536 | return 0; | |
2537 | } | |
2538 | ||
3a6bae48 | 2539 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
ff993ba1 | 2540 | struct pagevec *pvec) |
06c04442 KO |
2541 | { |
2542 | struct file *filp = iocb->ki_filp; | |
2543 | struct address_space *mapping = filp->f_mapping; | |
2544 | struct file_ra_state *ra = &filp->f_ra; | |
2545 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2546 | pgoff_t last_index; |
2642fca6 | 2547 | struct page *page; |
cbd59c48 | 2548 | int err = 0; |
06c04442 | 2549 | |
cbd59c48 | 2550 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2551 | retry: |
06c04442 KO |
2552 | if (fatal_signal_pending(current)) |
2553 | return -EINTR; | |
2554 | ||
cbd59c48 | 2555 | filemap_get_read_batch(mapping, index, last_index, pvec); |
2642fca6 MWO |
2556 | if (!pagevec_count(pvec)) { |
2557 | if (iocb->ki_flags & IOCB_NOIO) | |
2558 | return -EAGAIN; | |
2559 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2560 | last_index - index); | |
2561 | filemap_get_read_batch(mapping, index, last_index, pvec); | |
2562 | } | |
f253e185 MWO |
2563 | if (!pagevec_count(pvec)) { |
2564 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) | |
2565 | return -EAGAIN; | |
2566 | err = filemap_create_page(filp, mapping, | |
2567 | iocb->ki_pos >> PAGE_SHIFT, pvec); | |
2568 | if (err == AOP_TRUNCATED_PAGE) | |
2642fca6 | 2569 | goto retry; |
f253e185 MWO |
2570 | return err; |
2571 | } | |
06c04442 | 2572 | |
2642fca6 MWO |
2573 | page = pvec->pages[pagevec_count(pvec) - 1]; |
2574 | if (PageReadahead(page)) { | |
2575 | err = filemap_readahead(iocb, filp, mapping, page, last_index); | |
2576 | if (err) | |
2577 | goto err; | |
2578 | } | |
2579 | if (!PageUptodate(page)) { | |
2580 | if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1) | |
2581 | iocb->ki_flags |= IOCB_NOWAIT; | |
2582 | err = filemap_update_page(iocb, mapping, iter, page); | |
2583 | if (err) | |
2584 | goto err; | |
06c04442 KO |
2585 | } |
2586 | ||
2642fca6 | 2587 | return 0; |
cbd59c48 | 2588 | err: |
2642fca6 MWO |
2589 | if (err < 0) |
2590 | put_page(page); | |
2591 | if (likely(--pvec->nr)) | |
ff993ba1 | 2592 | return 0; |
4612aeef | 2593 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2594 | goto retry; |
2595 | return err; | |
06c04442 KO |
2596 | } |
2597 | ||
485bb99b | 2598 | /** |
87fa0f3e CH |
2599 | * filemap_read - Read data from the page cache. |
2600 | * @iocb: The iocb to read. | |
2601 | * @iter: Destination for the data. | |
2602 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2603 | * |
87fa0f3e CH |
2604 | * Copies data from the page cache. If the data is not currently present, |
2605 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2606 | * |
87fa0f3e CH |
2607 | * Return: Total number of bytes copied, including those already read by |
2608 | * the caller. If an error happens before any bytes are copied, returns | |
2609 | * a negative error number. | |
1da177e4 | 2610 | */ |
87fa0f3e CH |
2611 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2612 | ssize_t already_read) | |
1da177e4 | 2613 | { |
47c27bc4 | 2614 | struct file *filp = iocb->ki_filp; |
06c04442 | 2615 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2616 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2617 | struct inode *inode = mapping->host; |
ff993ba1 MWO |
2618 | struct pagevec pvec; |
2619 | int i, error = 0; | |
06c04442 KO |
2620 | bool writably_mapped; |
2621 | loff_t isize, end_offset; | |
1da177e4 | 2622 | |
723ef24b | 2623 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2624 | return 0; |
3644e2d2 KO |
2625 | if (unlikely(!iov_iter_count(iter))) |
2626 | return 0; | |
2627 | ||
c2a9737f | 2628 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
cbd59c48 | 2629 | pagevec_init(&pvec); |
c2a9737f | 2630 | |
06c04442 | 2631 | do { |
1da177e4 | 2632 | cond_resched(); |
5abf186a | 2633 | |
723ef24b | 2634 | /* |
06c04442 KO |
2635 | * If we've already successfully copied some data, then we |
2636 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2637 | * an async read NOWAIT at that point. | |
723ef24b | 2638 | */ |
87fa0f3e | 2639 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2640 | iocb->ki_flags |= IOCB_NOWAIT; |
2641 | ||
8c8387ee DH |
2642 | if (unlikely(iocb->ki_pos >= i_size_read(inode))) |
2643 | break; | |
2644 | ||
ff993ba1 MWO |
2645 | error = filemap_get_pages(iocb, iter, &pvec); |
2646 | if (error < 0) | |
06c04442 | 2647 | break; |
1da177e4 | 2648 | |
06c04442 KO |
2649 | /* |
2650 | * i_size must be checked after we know the pages are Uptodate. | |
2651 | * | |
2652 | * Checking i_size after the check allows us to calculate | |
2653 | * the correct value for "nr", which means the zero-filled | |
2654 | * part of the page is not copied back to userspace (unless | |
2655 | * another truncate extends the file - this is desired though). | |
2656 | */ | |
2657 | isize = i_size_read(inode); | |
2658 | if (unlikely(iocb->ki_pos >= isize)) | |
2659 | goto put_pages; | |
06c04442 KO |
2660 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2661 | ||
06c04442 KO |
2662 | /* |
2663 | * Once we start copying data, we don't want to be touching any | |
2664 | * cachelines that might be contended: | |
2665 | */ | |
2666 | writably_mapped = mapping_writably_mapped(mapping); | |
2667 | ||
2668 | /* | |
2669 | * When a sequential read accesses a page several times, only | |
2670 | * mark it as accessed the first time. | |
2671 | */ | |
2672 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2673 | ra->prev_pos >> PAGE_SHIFT) | |
ff993ba1 | 2674 | mark_page_accessed(pvec.pages[0]); |
06c04442 | 2675 | |
ff993ba1 | 2676 | for (i = 0; i < pagevec_count(&pvec); i++) { |
cbd59c48 MWO |
2677 | struct page *page = pvec.pages[i]; |
2678 | size_t page_size = thp_size(page); | |
2679 | size_t offset = iocb->ki_pos & (page_size - 1); | |
2680 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, | |
2681 | page_size - offset); | |
2682 | size_t copied; | |
06c04442 | 2683 | |
cbd59c48 MWO |
2684 | if (end_offset < page_offset(page)) |
2685 | break; | |
2686 | if (i > 0) | |
2687 | mark_page_accessed(page); | |
06c04442 KO |
2688 | /* |
2689 | * If users can be writing to this page using arbitrary | |
2690 | * virtual addresses, take care about potential aliasing | |
2691 | * before reading the page on the kernel side. | |
2692 | */ | |
cbd59c48 MWO |
2693 | if (writably_mapped) { |
2694 | int j; | |
2695 | ||
2696 | for (j = 0; j < thp_nr_pages(page); j++) | |
2697 | flush_dcache_page(page + j); | |
2698 | } | |
06c04442 | 2699 | |
cbd59c48 | 2700 | copied = copy_page_to_iter(page, offset, bytes, iter); |
06c04442 | 2701 | |
87fa0f3e | 2702 | already_read += copied; |
06c04442 KO |
2703 | iocb->ki_pos += copied; |
2704 | ra->prev_pos = iocb->ki_pos; | |
2705 | ||
2706 | if (copied < bytes) { | |
2707 | error = -EFAULT; | |
2708 | break; | |
2709 | } | |
1da177e4 | 2710 | } |
06c04442 | 2711 | put_pages: |
ff993ba1 MWO |
2712 | for (i = 0; i < pagevec_count(&pvec); i++) |
2713 | put_page(pvec.pages[i]); | |
cbd59c48 | 2714 | pagevec_reinit(&pvec); |
06c04442 | 2715 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2716 | |
0c6aa263 | 2717 | file_accessed(filp); |
06c04442 | 2718 | |
87fa0f3e | 2719 | return already_read ? already_read : error; |
1da177e4 | 2720 | } |
87fa0f3e | 2721 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2722 | |
485bb99b | 2723 | /** |
6abd2322 | 2724 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2725 | * @iocb: kernel I/O control block |
6abd2322 | 2726 | * @iter: destination for the data read |
485bb99b | 2727 | * |
6abd2322 | 2728 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2729 | * that can use the page cache directly. |
41da51bc AG |
2730 | * |
2731 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2732 | * be returned when no data can be read without waiting for I/O requests | |
2733 | * to complete; it doesn't prevent readahead. | |
2734 | * | |
2735 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2736 | * requests shall be made for the read or for readahead. When no data | |
2737 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2738 | * triggered, a partial, possibly empty read shall be returned. | |
2739 | * | |
a862f68a MR |
2740 | * Return: |
2741 | * * number of bytes copied, even for partial reads | |
41da51bc | 2742 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2743 | */ |
2744 | ssize_t | |
ed978a81 | 2745 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2746 | { |
e7080a43 | 2747 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2748 | ssize_t retval = 0; |
e7080a43 NS |
2749 | |
2750 | if (!count) | |
826ea860 | 2751 | return 0; /* skip atime */ |
1da177e4 | 2752 | |
2ba48ce5 | 2753 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2754 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2755 | struct address_space *mapping = file->f_mapping; |
2756 | struct inode *inode = mapping->host; | |
1da177e4 | 2757 | |
6be96d3a | 2758 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2759 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2760 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2761 | return -EAGAIN; |
2762 | } else { | |
2763 | retval = filemap_write_and_wait_range(mapping, | |
2764 | iocb->ki_pos, | |
2765 | iocb->ki_pos + count - 1); | |
2766 | if (retval < 0) | |
826ea860 | 2767 | return retval; |
6be96d3a | 2768 | } |
d8d3d94b | 2769 | |
0d5b0cf2 CH |
2770 | file_accessed(file); |
2771 | ||
5ecda137 | 2772 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2773 | if (retval >= 0) { |
c64fb5c7 | 2774 | iocb->ki_pos += retval; |
5ecda137 | 2775 | count -= retval; |
9fe55eea | 2776 | } |
ab2125df PB |
2777 | if (retval != -EIOCBQUEUED) |
2778 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2779 | |
9fe55eea SW |
2780 | /* |
2781 | * Btrfs can have a short DIO read if we encounter | |
2782 | * compressed extents, so if there was an error, or if | |
2783 | * we've already read everything we wanted to, or if | |
2784 | * there was a short read because we hit EOF, go ahead | |
2785 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2786 | * the rest of the read. Buffered reads will not work for |
2787 | * DAX files, so don't bother trying. | |
9fe55eea | 2788 | */ |
61d0017e JA |
2789 | if (retval < 0 || !count || IS_DAX(inode)) |
2790 | return retval; | |
2791 | if (iocb->ki_pos >= i_size_read(inode)) | |
826ea860 | 2792 | return retval; |
1da177e4 LT |
2793 | } |
2794 | ||
826ea860 | 2795 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2796 | } |
ed978a81 | 2797 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2798 | |
54fa39ac MWO |
2799 | static inline loff_t page_seek_hole_data(struct xa_state *xas, |
2800 | struct address_space *mapping, struct page *page, | |
2801 | loff_t start, loff_t end, bool seek_data) | |
41139aa4 | 2802 | { |
54fa39ac MWO |
2803 | const struct address_space_operations *ops = mapping->a_ops; |
2804 | size_t offset, bsz = i_blocksize(mapping->host); | |
2805 | ||
41139aa4 | 2806 | if (xa_is_value(page) || PageUptodate(page)) |
54fa39ac MWO |
2807 | return seek_data ? start : end; |
2808 | if (!ops->is_partially_uptodate) | |
2809 | return seek_data ? end : start; | |
2810 | ||
2811 | xas_pause(xas); | |
2812 | rcu_read_unlock(); | |
2813 | lock_page(page); | |
2814 | if (unlikely(page->mapping != mapping)) | |
2815 | goto unlock; | |
2816 | ||
2817 | offset = offset_in_thp(page, start) & ~(bsz - 1); | |
2818 | ||
2819 | do { | |
2820 | if (ops->is_partially_uptodate(page, offset, bsz) == seek_data) | |
2821 | break; | |
2822 | start = (start + bsz) & ~(bsz - 1); | |
2823 | offset += bsz; | |
2824 | } while (offset < thp_size(page)); | |
2825 | unlock: | |
2826 | unlock_page(page); | |
2827 | rcu_read_lock(); | |
2828 | return start; | |
41139aa4 MWO |
2829 | } |
2830 | ||
2831 | static inline | |
2832 | unsigned int seek_page_size(struct xa_state *xas, struct page *page) | |
2833 | { | |
2834 | if (xa_is_value(page)) | |
2835 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); | |
2836 | return thp_size(page); | |
2837 | } | |
2838 | ||
2839 | /** | |
2840 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2841 | * @mapping: Address space to search. | |
2842 | * @start: First byte to consider. | |
2843 | * @end: Limit of search (exclusive). | |
2844 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2845 | * | |
2846 | * If the page cache knows which blocks contain holes and which blocks | |
2847 | * contain data, your filesystem can use this function to implement | |
2848 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2849 | * entirely memory-based such as tmpfs, and filesystems which support | |
2850 | * unwritten extents. | |
2851 | * | |
f0953a1b | 2852 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2853 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2854 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2855 | * and @end contain data. | |
2856 | */ | |
2857 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2858 | loff_t end, int whence) | |
2859 | { | |
2860 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2861 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 MWO |
2862 | bool seek_data = (whence == SEEK_DATA); |
2863 | struct page *page; | |
2864 | ||
2865 | if (end <= start) | |
2866 | return -ENXIO; | |
2867 | ||
2868 | rcu_read_lock(); | |
2869 | while ((page = find_get_entry(&xas, max, XA_PRESENT))) { | |
ed98b015 HD |
2870 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
2871 | unsigned int seek_size; | |
41139aa4 MWO |
2872 | |
2873 | if (start < pos) { | |
2874 | if (!seek_data) | |
2875 | goto unlock; | |
2876 | start = pos; | |
2877 | } | |
2878 | ||
ed98b015 HD |
2879 | seek_size = seek_page_size(&xas, page); |
2880 | pos = round_up(pos + 1, seek_size); | |
54fa39ac MWO |
2881 | start = page_seek_hole_data(&xas, mapping, page, start, pos, |
2882 | seek_data); | |
2883 | if (start < pos) | |
41139aa4 | 2884 | goto unlock; |
ed98b015 HD |
2885 | if (start >= end) |
2886 | break; | |
2887 | if (seek_size > PAGE_SIZE) | |
2888 | xas_set(&xas, pos >> PAGE_SHIFT); | |
41139aa4 MWO |
2889 | if (!xa_is_value(page)) |
2890 | put_page(page); | |
2891 | } | |
41139aa4 | 2892 | if (seek_data) |
ed98b015 | 2893 | start = -ENXIO; |
41139aa4 MWO |
2894 | unlock: |
2895 | rcu_read_unlock(); | |
ed98b015 | 2896 | if (page && !xa_is_value(page)) |
41139aa4 | 2897 | put_page(page); |
41139aa4 MWO |
2898 | if (start > end) |
2899 | return end; | |
2900 | return start; | |
2901 | } | |
2902 | ||
1da177e4 | 2903 | #ifdef CONFIG_MMU |
1da177e4 | 2904 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2905 | /* |
c1e8d7c6 | 2906 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2907 | * @vmf - the vm_fault for this fault. |
2908 | * @page - the page to lock. | |
2909 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2910 | * | |
c1e8d7c6 | 2911 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2912 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2913 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2914 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2915 | */ | |
2916 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2917 | struct file **fpin) | |
2918 | { | |
7c23c782 MWO |
2919 | struct folio *folio = page_folio(page); |
2920 | ||
2921 | if (folio_trylock(folio)) | |
6b4c9f44 JB |
2922 | return 1; |
2923 | ||
8b0f9fa2 LT |
2924 | /* |
2925 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2926 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2927 | * is supposed to work. We have way too many special cases.. |
2928 | */ | |
6b4c9f44 JB |
2929 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2930 | return 0; | |
2931 | ||
2932 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2933 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
af7f29d9 | 2934 | if (__folio_lock_killable(folio)) { |
6b4c9f44 | 2935 | /* |
c1e8d7c6 | 2936 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2937 | * but all fault_handlers only check for fatal signals |
2938 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2939 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2940 | */ |
2941 | if (*fpin == NULL) | |
d8ed45c5 | 2942 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2943 | return 0; |
2944 | } | |
2945 | } else | |
7c23c782 MWO |
2946 | __folio_lock(folio); |
2947 | ||
6b4c9f44 JB |
2948 | return 1; |
2949 | } | |
2950 | ||
ef00e08e | 2951 | /* |
6b4c9f44 JB |
2952 | * Synchronous readahead happens when we don't even find a page in the page |
2953 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2954 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2955 | * that. If we didn't pin a file then we return NULL. The file that is | |
2956 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2957 | */ |
6b4c9f44 | 2958 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2959 | { |
2a1180f1 JB |
2960 | struct file *file = vmf->vma->vm_file; |
2961 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2962 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2963 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2964 | struct file *fpin = NULL; |
e630bfac | 2965 | unsigned int mmap_miss; |
ef00e08e LT |
2966 | |
2967 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2968 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2969 | return fpin; |
275b12bf | 2970 | if (!ra->ra_pages) |
6b4c9f44 | 2971 | return fpin; |
ef00e08e | 2972 | |
2a1180f1 | 2973 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2974 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 2975 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 2976 | return fpin; |
ef00e08e LT |
2977 | } |
2978 | ||
207d04ba | 2979 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2980 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2981 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2982 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2983 | |
2984 | /* | |
2985 | * Do we miss much more than hit in this file? If so, | |
2986 | * stop bothering with read-ahead. It will only hurt. | |
2987 | */ | |
e630bfac | 2988 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2989 | return fpin; |
ef00e08e | 2990 | |
d30a1100 WF |
2991 | /* |
2992 | * mmap read-around | |
2993 | */ | |
6b4c9f44 | 2994 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2995 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2996 | ra->size = ra->ra_pages; |
2997 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2998 | ractl._index = ra->start; |
2999 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 3000 | return fpin; |
ef00e08e LT |
3001 | } |
3002 | ||
3003 | /* | |
3004 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 3005 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 3006 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 3007 | */ |
6b4c9f44 JB |
3008 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
3009 | struct page *page) | |
ef00e08e | 3010 | { |
2a1180f1 JB |
3011 | struct file *file = vmf->vma->vm_file; |
3012 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 3013 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 3014 | struct file *fpin = NULL; |
e630bfac | 3015 | unsigned int mmap_miss; |
2a1180f1 | 3016 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
3017 | |
3018 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 3019 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 3020 | return fpin; |
e630bfac KS |
3021 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3022 | if (mmap_miss) | |
3023 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
3024 | if (PageReadahead(page)) { |
3025 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
3026 | page_cache_async_readahead(mapping, ra, file, |
3027 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
3028 | } |
3029 | return fpin; | |
ef00e08e LT |
3030 | } |
3031 | ||
485bb99b | 3032 | /** |
54cb8821 | 3033 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 3034 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 3035 | * |
54cb8821 | 3036 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
3037 | * mapped memory region to read in file data during a page fault. |
3038 | * | |
3039 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
3040 | * it in the page cache, and handles the special cases reasonably without | |
3041 | * having a lot of duplicated code. | |
9a95f3cf | 3042 | * |
c1e8d7c6 | 3043 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 3044 | * |
c1e8d7c6 | 3045 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 3046 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 3047 | * |
c1e8d7c6 | 3048 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
3049 | * has not been released. |
3050 | * | |
3051 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
3052 | * |
3053 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 3054 | */ |
2bcd6454 | 3055 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
3056 | { |
3057 | int error; | |
11bac800 | 3058 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 3059 | struct file *fpin = NULL; |
1da177e4 | 3060 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3061 | struct inode *inode = mapping->host; |
ef00e08e | 3062 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 3063 | pgoff_t max_off; |
1da177e4 | 3064 | struct page *page; |
2bcd6454 | 3065 | vm_fault_t ret = 0; |
730633f0 | 3066 | bool mapping_locked = false; |
1da177e4 | 3067 | |
9ab2594f MW |
3068 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3069 | if (unlikely(offset >= max_off)) | |
5307cc1a | 3070 | return VM_FAULT_SIGBUS; |
1da177e4 | 3071 | |
1da177e4 | 3072 | /* |
49426420 | 3073 | * Do we have something in the page cache already? |
1da177e4 | 3074 | */ |
ef00e08e | 3075 | page = find_get_page(mapping, offset); |
730633f0 | 3076 | if (likely(page)) { |
1da177e4 | 3077 | /* |
730633f0 JK |
3078 | * We found the page, so try async readahead before waiting for |
3079 | * the lock. | |
1da177e4 | 3080 | */ |
730633f0 JK |
3081 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
3082 | fpin = do_async_mmap_readahead(vmf, page); | |
3083 | if (unlikely(!PageUptodate(page))) { | |
3084 | filemap_invalidate_lock_shared(mapping); | |
3085 | mapping_locked = true; | |
3086 | } | |
3087 | } else { | |
ef00e08e | 3088 | /* No page in the page cache at all */ |
ef00e08e | 3089 | count_vm_event(PGMAJFAULT); |
2262185c | 3090 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3091 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3092 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3093 | retry_find: |
730633f0 JK |
3094 | /* |
3095 | * See comment in filemap_create_page() why we need | |
3096 | * invalidate_lock | |
3097 | */ | |
3098 | if (!mapping_locked) { | |
3099 | filemap_invalidate_lock_shared(mapping); | |
3100 | mapping_locked = true; | |
3101 | } | |
a75d4c33 JB |
3102 | page = pagecache_get_page(mapping, offset, |
3103 | FGP_CREAT|FGP_FOR_MMAP, | |
3104 | vmf->gfp_mask); | |
6b4c9f44 JB |
3105 | if (!page) { |
3106 | if (fpin) | |
3107 | goto out_retry; | |
730633f0 | 3108 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3109 | return VM_FAULT_OOM; |
6b4c9f44 | 3110 | } |
1da177e4 LT |
3111 | } |
3112 | ||
6b4c9f44 JB |
3113 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
3114 | goto out_retry; | |
b522c94d ML |
3115 | |
3116 | /* Did it get truncated? */ | |
585e5a7b | 3117 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
3118 | unlock_page(page); |
3119 | put_page(page); | |
3120 | goto retry_find; | |
3121 | } | |
520e5ba4 | 3122 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 3123 | |
1da177e4 | 3124 | /* |
d00806b1 NP |
3125 | * We have a locked page in the page cache, now we need to check |
3126 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3127 | */ |
730633f0 JK |
3128 | if (unlikely(!PageUptodate(page))) { |
3129 | /* | |
3130 | * The page was in cache and uptodate and now it is not. | |
3131 | * Strange but possible since we didn't hold the page lock all | |
3132 | * the time. Let's drop everything get the invalidate lock and | |
3133 | * try again. | |
3134 | */ | |
3135 | if (!mapping_locked) { | |
3136 | unlock_page(page); | |
3137 | put_page(page); | |
3138 | goto retry_find; | |
3139 | } | |
1da177e4 | 3140 | goto page_not_uptodate; |
730633f0 | 3141 | } |
1da177e4 | 3142 | |
6b4c9f44 | 3143 | /* |
c1e8d7c6 | 3144 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3145 | * time to return to the upper layer and have it re-find the vma and |
3146 | * redo the fault. | |
3147 | */ | |
3148 | if (fpin) { | |
3149 | unlock_page(page); | |
3150 | goto out_retry; | |
3151 | } | |
730633f0 JK |
3152 | if (mapping_locked) |
3153 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3154 | |
ef00e08e LT |
3155 | /* |
3156 | * Found the page and have a reference on it. | |
3157 | * We must recheck i_size under page lock. | |
3158 | */ | |
9ab2594f MW |
3159 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3160 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 3161 | unlock_page(page); |
09cbfeaf | 3162 | put_page(page); |
5307cc1a | 3163 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3164 | } |
3165 | ||
d0217ac0 | 3166 | vmf->page = page; |
83c54070 | 3167 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3168 | |
1da177e4 | 3169 | page_not_uptodate: |
1da177e4 LT |
3170 | /* |
3171 | * Umm, take care of errors if the page isn't up-to-date. | |
3172 | * Try to re-read it _once_. We do this synchronously, | |
3173 | * because there really aren't any performance issues here | |
3174 | * and we need to check for errors. | |
3175 | */ | |
6b4c9f44 | 3176 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
d31fa86a | 3177 | error = filemap_read_page(file, mapping, page); |
6b4c9f44 JB |
3178 | if (fpin) |
3179 | goto out_retry; | |
09cbfeaf | 3180 | put_page(page); |
d00806b1 NP |
3181 | |
3182 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3183 | goto retry_find; |
730633f0 | 3184 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3185 | |
d0217ac0 | 3186 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3187 | |
3188 | out_retry: | |
3189 | /* | |
c1e8d7c6 | 3190 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3191 | * re-find the vma and come back and find our hopefully still populated |
3192 | * page. | |
3193 | */ | |
3194 | if (page) | |
3195 | put_page(page); | |
730633f0 JK |
3196 | if (mapping_locked) |
3197 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3198 | if (fpin) |
3199 | fput(fpin); | |
3200 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3201 | } |
3202 | EXPORT_SYMBOL(filemap_fault); | |
3203 | ||
f9ce0be7 | 3204 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3205 | { |
f9ce0be7 KS |
3206 | struct mm_struct *mm = vmf->vma->vm_mm; |
3207 | ||
3208 | /* Huge page is mapped? No need to proceed. */ | |
3209 | if (pmd_trans_huge(*vmf->pmd)) { | |
3210 | unlock_page(page); | |
3211 | put_page(page); | |
3212 | return true; | |
3213 | } | |
3214 | ||
3215 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
e0f43fa5 YS |
3216 | vm_fault_t ret = do_set_pmd(vmf, page); |
3217 | if (!ret) { | |
3218 | /* The page is mapped successfully, reference consumed. */ | |
3219 | unlock_page(page); | |
3220 | return true; | |
f9ce0be7 | 3221 | } |
f9ce0be7 KS |
3222 | } |
3223 | ||
03c4f204 QZ |
3224 | if (pmd_none(*vmf->pmd)) |
3225 | pmd_install(mm, vmf->pmd, &vmf->prealloc_pte); | |
f9ce0be7 KS |
3226 | |
3227 | /* See comment in handle_pte_fault() */ | |
3228 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3229 | unlock_page(page); | |
3230 | put_page(page); | |
3231 | return true; | |
3232 | } | |
3233 | ||
3234 | return false; | |
3235 | } | |
3236 | ||
3237 | static struct page *next_uptodate_page(struct page *page, | |
3238 | struct address_space *mapping, | |
3239 | struct xa_state *xas, pgoff_t end_pgoff) | |
3240 | { | |
3241 | unsigned long max_idx; | |
3242 | ||
3243 | do { | |
3244 | if (!page) | |
3245 | return NULL; | |
3246 | if (xas_retry(xas, page)) | |
3247 | continue; | |
3248 | if (xa_is_value(page)) | |
3249 | continue; | |
3250 | if (PageLocked(page)) | |
3251 | continue; | |
3252 | if (!page_cache_get_speculative(page)) | |
3253 | continue; | |
3254 | /* Has the page moved or been split? */ | |
3255 | if (unlikely(page != xas_reload(xas))) | |
3256 | goto skip; | |
3257 | if (!PageUptodate(page) || PageReadahead(page)) | |
3258 | goto skip; | |
f9ce0be7 KS |
3259 | if (!trylock_page(page)) |
3260 | goto skip; | |
3261 | if (page->mapping != mapping) | |
3262 | goto unlock; | |
3263 | if (!PageUptodate(page)) | |
3264 | goto unlock; | |
3265 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3266 | if (xas->xa_index >= max_idx) | |
3267 | goto unlock; | |
3268 | return page; | |
3269 | unlock: | |
3270 | unlock_page(page); | |
3271 | skip: | |
3272 | put_page(page); | |
3273 | } while ((page = xas_next_entry(xas, end_pgoff)) != NULL); | |
3274 | ||
3275 | return NULL; | |
3276 | } | |
3277 | ||
3278 | static inline struct page *first_map_page(struct address_space *mapping, | |
3279 | struct xa_state *xas, | |
3280 | pgoff_t end_pgoff) | |
3281 | { | |
3282 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3283 | mapping, xas, end_pgoff); | |
3284 | } | |
3285 | ||
3286 | static inline struct page *next_map_page(struct address_space *mapping, | |
3287 | struct xa_state *xas, | |
3288 | pgoff_t end_pgoff) | |
3289 | { | |
3290 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3291 | mapping, xas, end_pgoff); | |
3292 | } | |
3293 | ||
3294 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3295 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3296 | { | |
3297 | struct vm_area_struct *vma = vmf->vma; | |
3298 | struct file *file = vma->vm_file; | |
f1820361 | 3299 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3300 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3301 | unsigned long addr; |
070e807c | 3302 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
27a83a60 | 3303 | struct page *head, *page; |
e630bfac | 3304 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3305 | vm_fault_t ret = 0; |
f1820361 KS |
3306 | |
3307 | rcu_read_lock(); | |
f9ce0be7 KS |
3308 | head = first_map_page(mapping, &xas, end_pgoff); |
3309 | if (!head) | |
3310 | goto out; | |
f1820361 | 3311 | |
f9ce0be7 KS |
3312 | if (filemap_map_pmd(vmf, head)) { |
3313 | ret = VM_FAULT_NOPAGE; | |
3314 | goto out; | |
3315 | } | |
f1820361 | 3316 | |
9d3af4b4 WD |
3317 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3318 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3319 | do { |
27a83a60 | 3320 | page = find_subpage(head, xas.xa_index); |
f9ce0be7 | 3321 | if (PageHWPoison(page)) |
f1820361 KS |
3322 | goto unlock; |
3323 | ||
e630bfac KS |
3324 | if (mmap_miss > 0) |
3325 | mmap_miss--; | |
7267ec00 | 3326 | |
9d3af4b4 | 3327 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3328 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3329 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3330 | |
3331 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3332 | goto unlock; |
f9ce0be7 | 3333 | |
46bdb427 | 3334 | /* We're about to handle the fault */ |
9d3af4b4 | 3335 | if (vmf->address == addr) |
46bdb427 | 3336 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3337 | |
9d3af4b4 | 3338 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3339 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3340 | update_mmu_cache(vma, addr, vmf->pte); |
27a83a60 | 3341 | unlock_page(head); |
f9ce0be7 | 3342 | continue; |
f1820361 | 3343 | unlock: |
27a83a60 | 3344 | unlock_page(head); |
27a83a60 | 3345 | put_page(head); |
f9ce0be7 KS |
3346 | } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
3347 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3348 | out: | |
f1820361 | 3349 | rcu_read_unlock(); |
e630bfac | 3350 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3351 | return ret; |
f1820361 KS |
3352 | } |
3353 | EXPORT_SYMBOL(filemap_map_pages); | |
3354 | ||
2bcd6454 | 3355 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3356 | { |
5df1a672 | 3357 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
4fcf1c62 | 3358 | struct page *page = vmf->page; |
2bcd6454 | 3359 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3360 | |
5df1a672 | 3361 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3362 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 | 3363 | lock_page(page); |
5df1a672 | 3364 | if (page->mapping != mapping) { |
4fcf1c62 JK |
3365 | unlock_page(page); |
3366 | ret = VM_FAULT_NOPAGE; | |
3367 | goto out; | |
3368 | } | |
14da9200 JK |
3369 | /* |
3370 | * We mark the page dirty already here so that when freeze is in | |
3371 | * progress, we are guaranteed that writeback during freezing will | |
3372 | * see the dirty page and writeprotect it again. | |
3373 | */ | |
3374 | set_page_dirty(page); | |
1d1d1a76 | 3375 | wait_for_stable_page(page); |
4fcf1c62 | 3376 | out: |
5df1a672 | 3377 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3378 | return ret; |
3379 | } | |
4fcf1c62 | 3380 | |
f0f37e2f | 3381 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3382 | .fault = filemap_fault, |
f1820361 | 3383 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3384 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3385 | }; |
3386 | ||
3387 | /* This is used for a general mmap of a disk file */ | |
3388 | ||
68d68ff6 | 3389 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3390 | { |
3391 | struct address_space *mapping = file->f_mapping; | |
3392 | ||
3393 | if (!mapping->a_ops->readpage) | |
3394 | return -ENOEXEC; | |
3395 | file_accessed(file); | |
3396 | vma->vm_ops = &generic_file_vm_ops; | |
3397 | return 0; | |
3398 | } | |
1da177e4 LT |
3399 | |
3400 | /* | |
3401 | * This is for filesystems which do not implement ->writepage. | |
3402 | */ | |
3403 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3404 | { | |
3405 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3406 | return -EINVAL; | |
3407 | return generic_file_mmap(file, vma); | |
3408 | } | |
3409 | #else | |
4b96a37d | 3410 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3411 | { |
4b96a37d | 3412 | return VM_FAULT_SIGBUS; |
45397228 | 3413 | } |
68d68ff6 | 3414 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3415 | { |
3416 | return -ENOSYS; | |
3417 | } | |
68d68ff6 | 3418 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3419 | { |
3420 | return -ENOSYS; | |
3421 | } | |
3422 | #endif /* CONFIG_MMU */ | |
3423 | ||
45397228 | 3424 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3425 | EXPORT_SYMBOL(generic_file_mmap); |
3426 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3427 | ||
67f9fd91 SL |
3428 | static struct page *wait_on_page_read(struct page *page) |
3429 | { | |
3430 | if (!IS_ERR(page)) { | |
3431 | wait_on_page_locked(page); | |
3432 | if (!PageUptodate(page)) { | |
09cbfeaf | 3433 | put_page(page); |
67f9fd91 SL |
3434 | page = ERR_PTR(-EIO); |
3435 | } | |
3436 | } | |
3437 | return page; | |
3438 | } | |
3439 | ||
32b63529 | 3440 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 3441 | pgoff_t index, |
5e5358e7 | 3442 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3443 | void *data, |
3444 | gfp_t gfp) | |
1da177e4 | 3445 | { |
eb2be189 | 3446 | struct page *page; |
1da177e4 LT |
3447 | int err; |
3448 | repeat: | |
3449 | page = find_get_page(mapping, index); | |
3450 | if (!page) { | |
453f85d4 | 3451 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
3452 | if (!page) |
3453 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 3454 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 3455 | if (unlikely(err)) { |
09cbfeaf | 3456 | put_page(page); |
eb2be189 NP |
3457 | if (err == -EEXIST) |
3458 | goto repeat; | |
22ecdb4f | 3459 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3460 | return ERR_PTR(err); |
3461 | } | |
32b63529 MG |
3462 | |
3463 | filler: | |
6c45b454 CH |
3464 | if (filler) |
3465 | err = filler(data, page); | |
3466 | else | |
3467 | err = mapping->a_ops->readpage(data, page); | |
3468 | ||
1da177e4 | 3469 | if (err < 0) { |
09cbfeaf | 3470 | put_page(page); |
32b63529 | 3471 | return ERR_PTR(err); |
1da177e4 | 3472 | } |
1da177e4 | 3473 | |
32b63529 MG |
3474 | page = wait_on_page_read(page); |
3475 | if (IS_ERR(page)) | |
3476 | return page; | |
3477 | goto out; | |
3478 | } | |
1da177e4 LT |
3479 | if (PageUptodate(page)) |
3480 | goto out; | |
3481 | ||
ebded027 | 3482 | /* |
0e9aa675 | 3483 | * Page is not up to date and may be locked due to one of the following |
ebded027 MG |
3484 | * case a: Page is being filled and the page lock is held |
3485 | * case b: Read/write error clearing the page uptodate status | |
3486 | * case c: Truncation in progress (page locked) | |
3487 | * case d: Reclaim in progress | |
3488 | * | |
3489 | * Case a, the page will be up to date when the page is unlocked. | |
3490 | * There is no need to serialise on the page lock here as the page | |
3491 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 3492 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
3493 | * it's a race vs truncate race. |
3494 | * Case b, the page will not be up to date | |
3495 | * Case c, the page may be truncated but in itself, the data may still | |
3496 | * be valid after IO completes as it's a read vs truncate race. The | |
3497 | * operation must restart if the page is not uptodate on unlock but | |
3498 | * otherwise serialising on page lock to stabilise the mapping gives | |
3499 | * no additional guarantees to the caller as the page lock is | |
3500 | * released before return. | |
3501 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3502 | * will be a race with remove_mapping that determines if the mapping | |
3503 | * is valid on unlock but otherwise the data is valid and there is | |
3504 | * no need to serialise with page lock. | |
3505 | * | |
3506 | * As the page lock gives no additional guarantee, we optimistically | |
3507 | * wait on the page to be unlocked and check if it's up to date and | |
3508 | * use the page if it is. Otherwise, the page lock is required to | |
3509 | * distinguish between the different cases. The motivation is that we | |
3510 | * avoid spurious serialisations and wakeups when multiple processes | |
3511 | * wait on the same page for IO to complete. | |
3512 | */ | |
3513 | wait_on_page_locked(page); | |
3514 | if (PageUptodate(page)) | |
3515 | goto out; | |
3516 | ||
3517 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3518 | lock_page(page); |
ebded027 MG |
3519 | |
3520 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3521 | if (!page->mapping) { |
3522 | unlock_page(page); | |
09cbfeaf | 3523 | put_page(page); |
32b63529 | 3524 | goto repeat; |
1da177e4 | 3525 | } |
ebded027 MG |
3526 | |
3527 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3528 | if (PageUptodate(page)) { |
3529 | unlock_page(page); | |
3530 | goto out; | |
3531 | } | |
faffdfa0 XT |
3532 | |
3533 | /* | |
3534 | * A previous I/O error may have been due to temporary | |
3535 | * failures. | |
3536 | * Clear page error before actual read, PG_error will be | |
3537 | * set again if read page fails. | |
3538 | */ | |
3539 | ClearPageError(page); | |
32b63529 MG |
3540 | goto filler; |
3541 | ||
c855ff37 | 3542 | out: |
6fe6900e NP |
3543 | mark_page_accessed(page); |
3544 | return page; | |
3545 | } | |
0531b2aa LT |
3546 | |
3547 | /** | |
67f9fd91 | 3548 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3549 | * @mapping: the page's address_space |
3550 | * @index: the page index | |
3551 | * @filler: function to perform the read | |
5e5358e7 | 3552 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3553 | * |
0531b2aa | 3554 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3555 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3556 | * |
3557 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3558 | * |
730633f0 JK |
3559 | * The function expects mapping->invalidate_lock to be already held. |
3560 | * | |
a862f68a | 3561 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa | 3562 | */ |
67f9fd91 | 3563 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3564 | pgoff_t index, |
5e5358e7 | 3565 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3566 | void *data) |
3567 | { | |
d322a8e5 CH |
3568 | return do_read_cache_page(mapping, index, filler, data, |
3569 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3570 | } |
67f9fd91 | 3571 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3572 | |
3573 | /** | |
3574 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3575 | * @mapping: the page's address_space | |
3576 | * @index: the page index | |
3577 | * @gfp: the page allocator flags to use if allocating | |
3578 | * | |
3579 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3580 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3581 | * |
3582 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3583 | * |
730633f0 JK |
3584 | * The function expects mapping->invalidate_lock to be already held. |
3585 | * | |
a862f68a | 3586 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3587 | */ |
3588 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3589 | pgoff_t index, | |
3590 | gfp_t gfp) | |
3591 | { | |
6c45b454 | 3592 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3593 | } |
3594 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3595 | ||
afddba49 NP |
3596 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3597 | loff_t pos, unsigned len, unsigned flags, | |
3598 | struct page **pagep, void **fsdata) | |
3599 | { | |
3600 | const struct address_space_operations *aops = mapping->a_ops; | |
3601 | ||
4e02ed4b | 3602 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3603 | pagep, fsdata); |
afddba49 NP |
3604 | } |
3605 | EXPORT_SYMBOL(pagecache_write_begin); | |
3606 | ||
3607 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3608 | loff_t pos, unsigned len, unsigned copied, | |
3609 | struct page *page, void *fsdata) | |
3610 | { | |
3611 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3612 | |
4e02ed4b | 3613 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3614 | } |
3615 | EXPORT_SYMBOL(pagecache_write_end); | |
3616 | ||
a92853b6 KK |
3617 | /* |
3618 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3619 | */ | |
3620 | void dio_warn_stale_pagecache(struct file *filp) | |
3621 | { | |
3622 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3623 | char pathname[128]; | |
a92853b6 KK |
3624 | char *path; |
3625 | ||
5df1a672 | 3626 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3627 | if (__ratelimit(&_rs)) { |
3628 | path = file_path(filp, pathname, sizeof(pathname)); | |
3629 | if (IS_ERR(path)) | |
3630 | path = "(unknown)"; | |
3631 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3632 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3633 | current->comm); | |
3634 | } | |
3635 | } | |
3636 | ||
1da177e4 | 3637 | ssize_t |
1af5bb49 | 3638 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3639 | { |
3640 | struct file *file = iocb->ki_filp; | |
3641 | struct address_space *mapping = file->f_mapping; | |
3642 | struct inode *inode = mapping->host; | |
1af5bb49 | 3643 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3644 | ssize_t written; |
a969e903 CH |
3645 | size_t write_len; |
3646 | pgoff_t end; | |
1da177e4 | 3647 | |
0c949334 | 3648 | write_len = iov_iter_count(from); |
09cbfeaf | 3649 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3650 | |
6be96d3a GR |
3651 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3652 | /* If there are pages to writeback, return */ | |
5df1a672 | 3653 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3654 | pos + write_len - 1)) |
6be96d3a GR |
3655 | return -EAGAIN; |
3656 | } else { | |
3657 | written = filemap_write_and_wait_range(mapping, pos, | |
3658 | pos + write_len - 1); | |
3659 | if (written) | |
3660 | goto out; | |
3661 | } | |
a969e903 CH |
3662 | |
3663 | /* | |
3664 | * After a write we want buffered reads to be sure to go to disk to get | |
3665 | * the new data. We invalidate clean cached page from the region we're | |
3666 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3667 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3668 | */ |
55635ba7 | 3669 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3670 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3671 | /* |
3672 | * If a page can not be invalidated, return 0 to fall back | |
3673 | * to buffered write. | |
3674 | */ | |
3675 | if (written) { | |
3676 | if (written == -EBUSY) | |
3677 | return 0; | |
3678 | goto out; | |
a969e903 CH |
3679 | } |
3680 | ||
639a93a5 | 3681 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3682 | |
3683 | /* | |
3684 | * Finally, try again to invalidate clean pages which might have been | |
3685 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3686 | * if the source of the write was an mmap'ed region of the file | |
3687 | * we're writing. Either one is a pretty crazy thing to do, | |
3688 | * so we don't support it 100%. If this invalidation | |
3689 | * fails, tough, the write still worked... | |
332391a9 LC |
3690 | * |
3691 | * Most of the time we do not need this since dio_complete() will do | |
3692 | * the invalidation for us. However there are some file systems that | |
3693 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3694 | * them by removing it completely. |
3695 | * | |
9266a140 KK |
3696 | * Noticeable example is a blkdev_direct_IO(). |
3697 | * | |
80c1fe90 | 3698 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3699 | */ |
9266a140 KK |
3700 | if (written > 0 && mapping->nrpages && |
3701 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3702 | dio_warn_stale_pagecache(file); | |
a969e903 | 3703 | |
1da177e4 | 3704 | if (written > 0) { |
0116651c | 3705 | pos += written; |
639a93a5 | 3706 | write_len -= written; |
0116651c NK |
3707 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3708 | i_size_write(inode, pos); | |
1da177e4 LT |
3709 | mark_inode_dirty(inode); |
3710 | } | |
5cb6c6c7 | 3711 | iocb->ki_pos = pos; |
1da177e4 | 3712 | } |
ab2125df PB |
3713 | if (written != -EIOCBQUEUED) |
3714 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3715 | out: |
1da177e4 LT |
3716 | return written; |
3717 | } | |
3718 | EXPORT_SYMBOL(generic_file_direct_write); | |
3719 | ||
3b93f911 | 3720 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3721 | struct iov_iter *i, loff_t pos) |
3722 | { | |
3723 | struct address_space *mapping = file->f_mapping; | |
3724 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3725 | long status = 0; | |
3726 | ssize_t written = 0; | |
674b892e NP |
3727 | unsigned int flags = 0; |
3728 | ||
afddba49 NP |
3729 | do { |
3730 | struct page *page; | |
afddba49 NP |
3731 | unsigned long offset; /* Offset into pagecache page */ |
3732 | unsigned long bytes; /* Bytes to write to page */ | |
3733 | size_t copied; /* Bytes copied from user */ | |
3734 | void *fsdata; | |
3735 | ||
09cbfeaf KS |
3736 | offset = (pos & (PAGE_SIZE - 1)); |
3737 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3738 | iov_iter_count(i)); |
3739 | ||
3740 | again: | |
00a3d660 LT |
3741 | /* |
3742 | * Bring in the user page that we will copy from _first_. | |
3743 | * Otherwise there's a nasty deadlock on copying from the | |
3744 | * same page as we're writing to, without it being marked | |
3745 | * up-to-date. | |
00a3d660 | 3746 | */ |
a6294593 | 3747 | if (unlikely(fault_in_iov_iter_readable(i, bytes))) { |
00a3d660 LT |
3748 | status = -EFAULT; |
3749 | break; | |
3750 | } | |
3751 | ||
296291cd JK |
3752 | if (fatal_signal_pending(current)) { |
3753 | status = -EINTR; | |
3754 | break; | |
3755 | } | |
3756 | ||
674b892e | 3757 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3758 | &page, &fsdata); |
2457aec6 | 3759 | if (unlikely(status < 0)) |
afddba49 NP |
3760 | break; |
3761 | ||
931e80e4 | 3762 | if (mapping_writably_mapped(mapping)) |
3763 | flush_dcache_page(page); | |
00a3d660 | 3764 | |
f0b65f39 | 3765 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3766 | flush_dcache_page(page); |
3767 | ||
3768 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3769 | page, fsdata); | |
f0b65f39 AV |
3770 | if (unlikely(status != copied)) { |
3771 | iov_iter_revert(i, copied - max(status, 0L)); | |
3772 | if (unlikely(status < 0)) | |
3773 | break; | |
3774 | } | |
afddba49 NP |
3775 | cond_resched(); |
3776 | ||
bc1bb416 | 3777 | if (unlikely(status == 0)) { |
afddba49 | 3778 | /* |
bc1bb416 AV |
3779 | * A short copy made ->write_end() reject the |
3780 | * thing entirely. Might be memory poisoning | |
3781 | * halfway through, might be a race with munmap, | |
3782 | * might be severe memory pressure. | |
afddba49 | 3783 | */ |
bc1bb416 AV |
3784 | if (copied) |
3785 | bytes = copied; | |
afddba49 NP |
3786 | goto again; |
3787 | } | |
f0b65f39 AV |
3788 | pos += status; |
3789 | written += status; | |
afddba49 NP |
3790 | |
3791 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3792 | } while (iov_iter_count(i)); |
3793 | ||
3794 | return written ? written : status; | |
3795 | } | |
3b93f911 | 3796 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3797 | |
e4dd9de3 | 3798 | /** |
8174202b | 3799 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3800 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3801 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3802 | * |
3803 | * This function does all the work needed for actually writing data to a | |
3804 | * file. It does all basic checks, removes SUID from the file, updates | |
3805 | * modification times and calls proper subroutines depending on whether we | |
3806 | * do direct IO or a standard buffered write. | |
3807 | * | |
9608703e | 3808 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3809 | * object which does not need locking at all. |
3810 | * | |
3811 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3812 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3813 | * avoid syncing under i_rwsem. |
a862f68a MR |
3814 | * |
3815 | * Return: | |
3816 | * * number of bytes written, even for truncated writes | |
3817 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3818 | */ |
8174202b | 3819 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3820 | { |
3821 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3822 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3823 | struct inode *inode = mapping->host; |
3b93f911 | 3824 | ssize_t written = 0; |
1da177e4 | 3825 | ssize_t err; |
3b93f911 | 3826 | ssize_t status; |
1da177e4 | 3827 | |
1da177e4 | 3828 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3829 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3830 | err = file_remove_privs(file); |
1da177e4 LT |
3831 | if (err) |
3832 | goto out; | |
3833 | ||
c3b2da31 JB |
3834 | err = file_update_time(file); |
3835 | if (err) | |
3836 | goto out; | |
1da177e4 | 3837 | |
2ba48ce5 | 3838 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3839 | loff_t pos, endbyte; |
fb5527e6 | 3840 | |
1af5bb49 | 3841 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3842 | /* |
fbbbad4b MW |
3843 | * If the write stopped short of completing, fall back to |
3844 | * buffered writes. Some filesystems do this for writes to | |
3845 | * holes, for example. For DAX files, a buffered write will | |
3846 | * not succeed (even if it did, DAX does not handle dirty | |
3847 | * page-cache pages correctly). | |
1da177e4 | 3848 | */ |
0b8def9d | 3849 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3850 | goto out; |
3851 | ||
0b8def9d | 3852 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3853 | /* |
3b93f911 | 3854 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3855 | * then we want to return the number of bytes which were |
3856 | * direct-written, or the error code if that was zero. Note | |
3857 | * that this differs from normal direct-io semantics, which | |
3858 | * will return -EFOO even if some bytes were written. | |
3859 | */ | |
60bb4529 | 3860 | if (unlikely(status < 0)) { |
3b93f911 | 3861 | err = status; |
fb5527e6 JM |
3862 | goto out; |
3863 | } | |
fb5527e6 JM |
3864 | /* |
3865 | * We need to ensure that the page cache pages are written to | |
3866 | * disk and invalidated to preserve the expected O_DIRECT | |
3867 | * semantics. | |
3868 | */ | |
3b93f911 | 3869 | endbyte = pos + status - 1; |
0b8def9d | 3870 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3871 | if (err == 0) { |
0b8def9d | 3872 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3873 | written += status; |
fb5527e6 | 3874 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3875 | pos >> PAGE_SHIFT, |
3876 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3877 | } else { |
3878 | /* | |
3879 | * We don't know how much we wrote, so just return | |
3880 | * the number of bytes which were direct-written | |
3881 | */ | |
3882 | } | |
3883 | } else { | |
0b8def9d AV |
3884 | written = generic_perform_write(file, from, iocb->ki_pos); |
3885 | if (likely(written > 0)) | |
3886 | iocb->ki_pos += written; | |
fb5527e6 | 3887 | } |
1da177e4 LT |
3888 | out: |
3889 | current->backing_dev_info = NULL; | |
3890 | return written ? written : err; | |
3891 | } | |
8174202b | 3892 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3893 | |
e4dd9de3 | 3894 | /** |
8174202b | 3895 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3896 | * @iocb: IO state structure |
8174202b | 3897 | * @from: iov_iter with data to write |
e4dd9de3 | 3898 | * |
8174202b | 3899 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 3900 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 3901 | * and acquires i_rwsem as needed. |
a862f68a MR |
3902 | * Return: |
3903 | * * negative error code if no data has been written at all of | |
3904 | * vfs_fsync_range() failed for a synchronous write | |
3905 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3906 | */ |
8174202b | 3907 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3908 | { |
3909 | struct file *file = iocb->ki_filp; | |
148f948b | 3910 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3911 | ssize_t ret; |
1da177e4 | 3912 | |
5955102c | 3913 | inode_lock(inode); |
3309dd04 AV |
3914 | ret = generic_write_checks(iocb, from); |
3915 | if (ret > 0) | |
5f380c7f | 3916 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3917 | inode_unlock(inode); |
1da177e4 | 3918 | |
e2592217 CH |
3919 | if (ret > 0) |
3920 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3921 | return ret; |
3922 | } | |
8174202b | 3923 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3924 | |
cf9a2ae8 DH |
3925 | /** |
3926 | * try_to_release_page() - release old fs-specific metadata on a page | |
3927 | * | |
3928 | * @page: the page which the kernel is trying to free | |
3929 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3930 | * | |
3931 | * The address_space is to try to release any data against the page | |
a862f68a | 3932 | * (presumably at page->private). |
cf9a2ae8 | 3933 | * |
266cf658 DH |
3934 | * This may also be called if PG_fscache is set on a page, indicating that the |
3935 | * page is known to the local caching routines. | |
3936 | * | |
cf9a2ae8 | 3937 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3938 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3939 | * |
a862f68a | 3940 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3941 | */ |
3942 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3943 | { | |
3944 | struct address_space * const mapping = page->mapping; | |
3945 | ||
3946 | BUG_ON(!PageLocked(page)); | |
3947 | if (PageWriteback(page)) | |
3948 | return 0; | |
3949 | ||
3950 | if (mapping && mapping->a_ops->releasepage) | |
3951 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3952 | return try_to_free_buffers(page); | |
3953 | } | |
3954 | ||
3955 | EXPORT_SYMBOL(try_to_release_page); |