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 LT |
32 | #include <linux/pagevec.h> |
33 | #include <linux/blkdev.h> | |
34 | #include <linux/security.h> | |
44110fe3 | 35 | #include <linux/cpuset.h> |
00501b53 | 36 | #include <linux/hugetlb.h> |
8a9f3ccd | 37 | #include <linux/memcontrol.h> |
c515e1fd | 38 | #include <linux/cleancache.h> |
c7df8ad2 | 39 | #include <linux/shmem_fs.h> |
f1820361 | 40 | #include <linux/rmap.h> |
b1d29ba8 | 41 | #include <linux/delayacct.h> |
eb414681 | 42 | #include <linux/psi.h> |
d0e6a582 | 43 | #include <linux/ramfs.h> |
b9306a79 | 44 | #include <linux/page_idle.h> |
f9ce0be7 | 45 | #include <asm/pgalloc.h> |
de591a82 | 46 | #include <asm/tlbflush.h> |
0f8053a5 NP |
47 | #include "internal.h" |
48 | ||
fe0bfaaf RJ |
49 | #define CREATE_TRACE_POINTS |
50 | #include <trace/events/filemap.h> | |
51 | ||
1da177e4 | 52 | /* |
1da177e4 LT |
53 | * FIXME: remove all knowledge of the buffer layer from the core VM |
54 | */ | |
148f948b | 55 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 56 | |
1da177e4 LT |
57 | #include <asm/mman.h> |
58 | ||
59 | /* | |
60 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
61 | * though. | |
62 | * | |
63 | * Shared mappings now work. 15.8.1995 Bruno. | |
64 | * | |
65 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
66 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
67 | * | |
68 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
69 | */ | |
70 | ||
71 | /* | |
72 | * Lock ordering: | |
73 | * | |
c8c06efa | 74 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 75 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 76 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 77 | * ->i_pages lock |
1da177e4 | 78 | * |
9608703e | 79 | * ->i_rwsem |
730633f0 JK |
80 | * ->invalidate_lock (acquired by fs in truncate path) |
81 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) | |
1da177e4 | 82 | * |
c1e8d7c6 | 83 | * ->mmap_lock |
c8c06efa | 84 | * ->i_mmap_rwsem |
b8072f09 | 85 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 86 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 87 | * |
c1e8d7c6 | 88 | * ->mmap_lock |
730633f0 JK |
89 | * ->invalidate_lock (filemap_fault) |
90 | * ->lock_page (filemap_fault, access_process_vm) | |
1da177e4 | 91 | * |
9608703e | 92 | * ->i_rwsem (generic_perform_write) |
c1e8d7c6 | 93 | * ->mmap_lock (fault_in_pages_readable->do_page_fault) |
1da177e4 | 94 | * |
f758eeab | 95 | * bdi->wb.list_lock |
a66979ab | 96 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 97 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 98 | * |
c8c06efa | 99 | * ->i_mmap_rwsem |
1da177e4 LT |
100 | * ->anon_vma.lock (vma_adjust) |
101 | * | |
102 | * ->anon_vma.lock | |
b8072f09 | 103 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 104 | * |
b8072f09 | 105 | * ->page_table_lock or pte_lock |
5d337b91 | 106 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 107 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 108 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
109 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
110 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 111 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 112 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 113 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 114 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 115 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 116 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 117 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
118 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
119 | * | |
c8c06efa | 120 | * ->i_mmap_rwsem |
9a3c531d | 121 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
122 | */ |
123 | ||
5c024e6a | 124 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
125 | struct page *page, void *shadow) |
126 | { | |
5c024e6a MW |
127 | XA_STATE(xas, &mapping->i_pages, page->index); |
128 | unsigned int nr = 1; | |
c70b647d | 129 | |
5c024e6a | 130 | mapping_set_update(&xas, mapping); |
c70b647d | 131 | |
5c024e6a MW |
132 | /* hugetlb pages are represented by a single entry in the xarray */ |
133 | if (!PageHuge(page)) { | |
134 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 135 | nr = compound_nr(page); |
5c024e6a | 136 | } |
91b0abe3 | 137 | |
83929372 KS |
138 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
139 | VM_BUG_ON_PAGE(PageTail(page), page); | |
140 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 141 | |
5c024e6a MW |
142 | xas_store(&xas, shadow); |
143 | xas_init_marks(&xas); | |
d3798ae8 | 144 | |
2300638b JK |
145 | page->mapping = NULL; |
146 | /* Leave page->index set: truncation lookup relies upon it */ | |
d3798ae8 | 147 | mapping->nrpages -= nr; |
91b0abe3 JW |
148 | } |
149 | ||
5ecc4d85 JK |
150 | static void unaccount_page_cache_page(struct address_space *mapping, |
151 | struct page *page) | |
1da177e4 | 152 | { |
5ecc4d85 | 153 | int nr; |
1da177e4 | 154 | |
c515e1fd DM |
155 | /* |
156 | * if we're uptodate, flush out into the cleancache, otherwise | |
157 | * invalidate any existing cleancache entries. We can't leave | |
158 | * stale data around in the cleancache once our page is gone | |
159 | */ | |
160 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
161 | cleancache_put_page(page); | |
162 | else | |
3167760f | 163 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 164 | |
83929372 | 165 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
166 | VM_BUG_ON_PAGE(page_mapped(page), page); |
167 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
168 | int mapcount; | |
169 | ||
170 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
171 | current->comm, page_to_pfn(page)); | |
172 | dump_page(page, "still mapped when deleted"); | |
173 | dump_stack(); | |
174 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
175 | ||
176 | mapcount = page_mapcount(page); | |
177 | if (mapping_exiting(mapping) && | |
178 | page_count(page) >= mapcount + 2) { | |
179 | /* | |
180 | * All vmas have already been torn down, so it's | |
181 | * a good bet that actually the page is unmapped, | |
182 | * and we'd prefer not to leak it: if we're wrong, | |
183 | * some other bad page check should catch it later. | |
184 | */ | |
185 | page_mapcount_reset(page); | |
6d061f9f | 186 | page_ref_sub(page, mapcount); |
06b241f3 HD |
187 | } |
188 | } | |
189 | ||
4165b9b4 | 190 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
191 | if (PageHuge(page)) |
192 | return; | |
09612fa6 | 193 | |
6c357848 | 194 | nr = thp_nr_pages(page); |
5ecc4d85 | 195 | |
0d1c2072 | 196 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 197 | if (PageSwapBacked(page)) { |
0d1c2072 | 198 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 | 199 | if (PageTransHuge(page)) |
57b2847d | 200 | __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr); |
99cb0dbd | 201 | } else if (PageTransHuge(page)) { |
bf9ecead | 202 | __mod_lruvec_page_state(page, NR_FILE_THPS, -nr); |
09d91cda | 203 | filemap_nr_thps_dec(mapping); |
800d8c63 | 204 | } |
5ecc4d85 JK |
205 | |
206 | /* | |
207 | * At this point page must be either written or cleaned by | |
208 | * truncate. Dirty page here signals a bug and loss of | |
209 | * unwritten data. | |
210 | * | |
211 | * This fixes dirty accounting after removing the page entirely | |
212 | * but leaves PageDirty set: it has no effect for truncated | |
213 | * page and anyway will be cleared before returning page into | |
214 | * buddy allocator. | |
215 | */ | |
216 | if (WARN_ON_ONCE(PageDirty(page))) | |
217 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Delete a page from the page cache and free it. Caller has to make | |
222 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 223 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
224 | */ |
225 | void __delete_from_page_cache(struct page *page, void *shadow) | |
226 | { | |
227 | struct address_space *mapping = page->mapping; | |
228 | ||
229 | trace_mm_filemap_delete_from_page_cache(page); | |
230 | ||
231 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 232 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
233 | } |
234 | ||
59c66c5f JK |
235 | static void page_cache_free_page(struct address_space *mapping, |
236 | struct page *page) | |
237 | { | |
238 | void (*freepage)(struct page *); | |
239 | ||
240 | freepage = mapping->a_ops->freepage; | |
241 | if (freepage) | |
242 | freepage(page); | |
243 | ||
244 | if (PageTransHuge(page) && !PageHuge(page)) { | |
887b22c6 | 245 | page_ref_sub(page, thp_nr_pages(page)); |
59c66c5f JK |
246 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); |
247 | } else { | |
248 | put_page(page); | |
249 | } | |
250 | } | |
251 | ||
702cfbf9 MK |
252 | /** |
253 | * delete_from_page_cache - delete page from page cache | |
254 | * @page: the page which the kernel is trying to remove from page cache | |
255 | * | |
256 | * This must be called only on pages that have been verified to be in the page | |
257 | * cache and locked. It will never put the page into the free list, the caller | |
258 | * has a reference on the page. | |
259 | */ | |
260 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 261 | { |
83929372 | 262 | struct address_space *mapping = page_mapping(page); |
1da177e4 | 263 | |
cd7619d6 | 264 | BUG_ON(!PageLocked(page)); |
30472509 | 265 | xa_lock_irq(&mapping->i_pages); |
62cccb8c | 266 | __delete_from_page_cache(page, NULL); |
30472509 | 267 | xa_unlock_irq(&mapping->i_pages); |
6072d13c | 268 | |
59c66c5f | 269 | page_cache_free_page(mapping, page); |
97cecb5a MK |
270 | } |
271 | EXPORT_SYMBOL(delete_from_page_cache); | |
272 | ||
aa65c29c | 273 | /* |
ef8e5717 | 274 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
275 | * @mapping: the mapping to which pages belong |
276 | * @pvec: pagevec with pages to delete | |
277 | * | |
b93b0163 | 278 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
279 | * from the mapping. The function expects @pvec to be sorted by page index |
280 | * and is optimised for it to be dense. | |
b93b0163 | 281 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 282 | * modified). The function expects only THP head pages to be present in the |
4101196b | 283 | * @pvec. |
aa65c29c | 284 | * |
b93b0163 | 285 | * The function expects the i_pages lock to be held. |
aa65c29c | 286 | */ |
ef8e5717 | 287 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
288 | struct pagevec *pvec) |
289 | { | |
ef8e5717 | 290 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 291 | int total_pages = 0; |
4101196b | 292 | int i = 0; |
aa65c29c | 293 | struct page *page; |
aa65c29c | 294 | |
ef8e5717 MW |
295 | mapping_set_update(&xas, mapping); |
296 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 297 | if (i >= pagevec_count(pvec)) |
aa65c29c | 298 | break; |
4101196b MWO |
299 | |
300 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 301 | if (xa_is_value(page)) |
aa65c29c | 302 | continue; |
4101196b MWO |
303 | /* |
304 | * A page got inserted in our range? Skip it. We have our | |
305 | * pages locked so they are protected from being removed. | |
306 | * If we see a page whose index is higher than ours, it | |
307 | * means our page has been removed, which shouldn't be | |
308 | * possible because we're holding the PageLock. | |
309 | */ | |
310 | if (page != pvec->pages[i]) { | |
311 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
312 | page); | |
313 | continue; | |
314 | } | |
315 | ||
316 | WARN_ON_ONCE(!PageLocked(page)); | |
317 | ||
318 | if (page->index == xas.xa_index) | |
aa65c29c | 319 | page->mapping = NULL; |
4101196b MWO |
320 | /* Leave page->index set: truncation lookup relies on it */ |
321 | ||
322 | /* | |
323 | * Move to the next page in the vector if this is a regular | |
324 | * page or the index is of the last sub-page of this compound | |
325 | * page. | |
326 | */ | |
327 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 328 | i++; |
ef8e5717 | 329 | xas_store(&xas, NULL); |
aa65c29c JK |
330 | total_pages++; |
331 | } | |
332 | mapping->nrpages -= total_pages; | |
333 | } | |
334 | ||
335 | void delete_from_page_cache_batch(struct address_space *mapping, | |
336 | struct pagevec *pvec) | |
337 | { | |
338 | int i; | |
aa65c29c JK |
339 | |
340 | if (!pagevec_count(pvec)) | |
341 | return; | |
342 | ||
30472509 | 343 | xa_lock_irq(&mapping->i_pages); |
aa65c29c JK |
344 | for (i = 0; i < pagevec_count(pvec); i++) { |
345 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
346 | ||
347 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
348 | } | |
ef8e5717 | 349 | page_cache_delete_batch(mapping, pvec); |
30472509 | 350 | xa_unlock_irq(&mapping->i_pages); |
aa65c29c JK |
351 | |
352 | for (i = 0; i < pagevec_count(pvec); i++) | |
353 | page_cache_free_page(mapping, pvec->pages[i]); | |
354 | } | |
355 | ||
d72d9e2a | 356 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
357 | { |
358 | int ret = 0; | |
359 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
360 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
361 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 362 | ret = -ENOSPC; |
7fcbbaf1 JA |
363 | if (test_bit(AS_EIO, &mapping->flags) && |
364 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
365 | ret = -EIO; |
366 | return ret; | |
367 | } | |
d72d9e2a | 368 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 369 | |
76341cab JL |
370 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
371 | { | |
372 | /* Check for outstanding write errors */ | |
373 | if (test_bit(AS_EIO, &mapping->flags)) | |
374 | return -EIO; | |
375 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
376 | return -ENOSPC; | |
377 | return 0; | |
378 | } | |
379 | ||
5a798493 JB |
380 | /** |
381 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range | |
382 | * @mapping: address space structure to write | |
383 | * @wbc: the writeback_control controlling the writeout | |
384 | * | |
385 | * Call writepages on the mapping using the provided wbc to control the | |
386 | * writeout. | |
387 | * | |
388 | * Return: %0 on success, negative error code otherwise. | |
389 | */ | |
390 | int filemap_fdatawrite_wbc(struct address_space *mapping, | |
391 | struct writeback_control *wbc) | |
392 | { | |
393 | int ret; | |
394 | ||
395 | if (!mapping_can_writeback(mapping) || | |
396 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
397 | return 0; | |
398 | ||
399 | wbc_attach_fdatawrite_inode(wbc, mapping->host); | |
400 | ret = do_writepages(mapping, wbc); | |
401 | wbc_detach_inode(wbc); | |
402 | return ret; | |
403 | } | |
404 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); | |
405 | ||
1da177e4 | 406 | /** |
485bb99b | 407 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
408 | * @mapping: address space structure to write |
409 | * @start: offset in bytes where the range starts | |
469eb4d0 | 410 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 411 | * @sync_mode: enable synchronous operation |
1da177e4 | 412 | * |
485bb99b RD |
413 | * Start writeback against all of a mapping's dirty pages that lie |
414 | * within the byte offsets <start, end> inclusive. | |
415 | * | |
1da177e4 | 416 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 417 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
418 | * these two operations is that if a dirty page/buffer is encountered, it must |
419 | * be waited upon, and not just skipped over. | |
a862f68a MR |
420 | * |
421 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 422 | */ |
ebcf28e1 AM |
423 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
424 | loff_t end, int sync_mode) | |
1da177e4 | 425 | { |
1da177e4 LT |
426 | struct writeback_control wbc = { |
427 | .sync_mode = sync_mode, | |
05fe478d | 428 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
429 | .range_start = start, |
430 | .range_end = end, | |
1da177e4 LT |
431 | }; |
432 | ||
5a798493 | 433 | return filemap_fdatawrite_wbc(mapping, &wbc); |
1da177e4 LT |
434 | } |
435 | ||
436 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
437 | int sync_mode) | |
438 | { | |
111ebb6e | 439 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
440 | } |
441 | ||
442 | int filemap_fdatawrite(struct address_space *mapping) | |
443 | { | |
444 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
445 | } | |
446 | EXPORT_SYMBOL(filemap_fdatawrite); | |
447 | ||
f4c0a0fd | 448 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 449 | loff_t end) |
1da177e4 LT |
450 | { |
451 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
452 | } | |
f4c0a0fd | 453 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 454 | |
485bb99b RD |
455 | /** |
456 | * filemap_flush - mostly a non-blocking flush | |
457 | * @mapping: target address_space | |
458 | * | |
1da177e4 LT |
459 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
460 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
461 | * |
462 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
463 | */ |
464 | int filemap_flush(struct address_space *mapping) | |
465 | { | |
466 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
467 | } | |
468 | EXPORT_SYMBOL(filemap_flush); | |
469 | ||
7fc9e472 GR |
470 | /** |
471 | * filemap_range_has_page - check if a page exists in range. | |
472 | * @mapping: address space within which to check | |
473 | * @start_byte: offset in bytes where the range starts | |
474 | * @end_byte: offset in bytes where the range ends (inclusive) | |
475 | * | |
476 | * Find at least one page in the range supplied, usually used to check if | |
477 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
478 | * |
479 | * Return: %true if at least one page exists in the specified range, | |
480 | * %false otherwise. | |
7fc9e472 GR |
481 | */ |
482 | bool filemap_range_has_page(struct address_space *mapping, | |
483 | loff_t start_byte, loff_t end_byte) | |
484 | { | |
f7b68046 | 485 | struct page *page; |
8fa8e538 MW |
486 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
487 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
488 | |
489 | if (end_byte < start_byte) | |
490 | return false; | |
491 | ||
8fa8e538 MW |
492 | rcu_read_lock(); |
493 | for (;;) { | |
494 | page = xas_find(&xas, max); | |
495 | if (xas_retry(&xas, page)) | |
496 | continue; | |
497 | /* Shadow entries don't count */ | |
498 | if (xa_is_value(page)) | |
499 | continue; | |
500 | /* | |
501 | * We don't need to try to pin this page; we're about to | |
502 | * release the RCU lock anyway. It is enough to know that | |
503 | * there was a page here recently. | |
504 | */ | |
505 | break; | |
506 | } | |
507 | rcu_read_unlock(); | |
7fc9e472 | 508 | |
8fa8e538 | 509 | return page != NULL; |
7fc9e472 GR |
510 | } |
511 | EXPORT_SYMBOL(filemap_range_has_page); | |
512 | ||
5e8fcc1a | 513 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 514 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 515 | { |
09cbfeaf KS |
516 | pgoff_t index = start_byte >> PAGE_SHIFT; |
517 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
518 | struct pagevec pvec; |
519 | int nr_pages; | |
1da177e4 | 520 | |
94004ed7 | 521 | if (end_byte < start_byte) |
5e8fcc1a | 522 | return; |
1da177e4 | 523 | |
86679820 | 524 | pagevec_init(&pvec); |
312e9d2f | 525 | while (index <= end) { |
1da177e4 LT |
526 | unsigned i; |
527 | ||
312e9d2f | 528 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 529 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
530 | if (!nr_pages) |
531 | break; | |
532 | ||
1da177e4 LT |
533 | for (i = 0; i < nr_pages; i++) { |
534 | struct page *page = pvec.pages[i]; | |
535 | ||
1da177e4 | 536 | wait_on_page_writeback(page); |
5e8fcc1a | 537 | ClearPageError(page); |
1da177e4 LT |
538 | } |
539 | pagevec_release(&pvec); | |
540 | cond_resched(); | |
541 | } | |
aa750fd7 JN |
542 | } |
543 | ||
544 | /** | |
545 | * filemap_fdatawait_range - wait for writeback to complete | |
546 | * @mapping: address space structure to wait for | |
547 | * @start_byte: offset in bytes where the range starts | |
548 | * @end_byte: offset in bytes where the range ends (inclusive) | |
549 | * | |
550 | * Walk the list of under-writeback pages of the given address space | |
551 | * in the given range and wait for all of them. Check error status of | |
552 | * the address space and return it. | |
553 | * | |
554 | * Since the error status of the address space is cleared by this function, | |
555 | * callers are responsible for checking the return value and handling and/or | |
556 | * reporting the error. | |
a862f68a MR |
557 | * |
558 | * Return: error status of the address space. | |
aa750fd7 JN |
559 | */ |
560 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
561 | loff_t end_byte) | |
562 | { | |
5e8fcc1a JL |
563 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
564 | return filemap_check_errors(mapping); | |
1da177e4 | 565 | } |
d3bccb6f JK |
566 | EXPORT_SYMBOL(filemap_fdatawait_range); |
567 | ||
aa0bfcd9 RZ |
568 | /** |
569 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
570 | * @mapping: address space structure to wait for | |
571 | * @start_byte: offset in bytes where the range starts | |
572 | * @end_byte: offset in bytes where the range ends (inclusive) | |
573 | * | |
574 | * Walk the list of under-writeback pages of the given address space in the | |
575 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
576 | * this function does not clear error status of the address space. | |
577 | * | |
578 | * Use this function if callers don't handle errors themselves. Expected | |
579 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
580 | * fsfreeze(8) | |
581 | */ | |
582 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
583 | loff_t start_byte, loff_t end_byte) | |
584 | { | |
585 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
586 | return filemap_check_and_keep_errors(mapping); | |
587 | } | |
588 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
589 | ||
a823e458 JL |
590 | /** |
591 | * file_fdatawait_range - wait for writeback to complete | |
592 | * @file: file pointing to address space structure to wait for | |
593 | * @start_byte: offset in bytes where the range starts | |
594 | * @end_byte: offset in bytes where the range ends (inclusive) | |
595 | * | |
596 | * Walk the list of under-writeback pages of the address space that file | |
597 | * refers to, in the given range and wait for all of them. Check error | |
598 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
599 | * | |
600 | * Since the error status of the file is advanced by this function, | |
601 | * callers are responsible for checking the return value and handling and/or | |
602 | * reporting the error. | |
a862f68a MR |
603 | * |
604 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
605 | */ |
606 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
607 | { | |
608 | struct address_space *mapping = file->f_mapping; | |
609 | ||
610 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
611 | return file_check_and_advance_wb_err(file); | |
612 | } | |
613 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 614 | |
aa750fd7 JN |
615 | /** |
616 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
617 | * @mapping: address space structure to wait for | |
618 | * | |
619 | * Walk the list of under-writeback pages of the given address space | |
620 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
621 | * does not clear error status of the address space. | |
622 | * | |
623 | * Use this function if callers don't handle errors themselves. Expected | |
624 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
625 | * fsfreeze(8) | |
a862f68a MR |
626 | * |
627 | * Return: error status of the address space. | |
aa750fd7 | 628 | */ |
76341cab | 629 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 630 | { |
ffb959bb | 631 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 632 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 633 | } |
76341cab | 634 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 635 | |
875d91b1 | 636 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 637 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 638 | { |
875d91b1 | 639 | return mapping->nrpages; |
1da177e4 | 640 | } |
1da177e4 | 641 | |
63135aa3 JA |
642 | /** |
643 | * filemap_range_needs_writeback - check if range potentially needs writeback | |
644 | * @mapping: address space within which to check | |
645 | * @start_byte: offset in bytes where the range starts | |
646 | * @end_byte: offset in bytes where the range ends (inclusive) | |
647 | * | |
648 | * Find at least one page in the range supplied, usually used to check if | |
649 | * direct writing in this range will trigger a writeback. Used by O_DIRECT | |
650 | * read/write with IOCB_NOWAIT, to see if the caller needs to do | |
651 | * filemap_write_and_wait_range() before proceeding. | |
652 | * | |
653 | * Return: %true if the caller should do filemap_write_and_wait_range() before | |
654 | * doing O_DIRECT to a page in this range, %false otherwise. | |
655 | */ | |
656 | bool filemap_range_needs_writeback(struct address_space *mapping, | |
657 | loff_t start_byte, loff_t end_byte) | |
658 | { | |
659 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
660 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
661 | struct page *page; | |
662 | ||
663 | if (!mapping_needs_writeback(mapping)) | |
664 | return false; | |
665 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && | |
666 | !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
667 | return false; | |
668 | if (end_byte < start_byte) | |
669 | return false; | |
670 | ||
671 | rcu_read_lock(); | |
672 | xas_for_each(&xas, page, max) { | |
673 | if (xas_retry(&xas, page)) | |
674 | continue; | |
675 | if (xa_is_value(page)) | |
676 | continue; | |
677 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
678 | break; | |
679 | } | |
680 | rcu_read_unlock(); | |
681 | return page != NULL; | |
682 | } | |
683 | EXPORT_SYMBOL_GPL(filemap_range_needs_writeback); | |
684 | ||
485bb99b RD |
685 | /** |
686 | * filemap_write_and_wait_range - write out & wait on a file range | |
687 | * @mapping: the address_space for the pages | |
688 | * @lstart: offset in bytes where the range starts | |
689 | * @lend: offset in bytes where the range ends (inclusive) | |
690 | * | |
469eb4d0 AM |
691 | * Write out and wait upon file offsets lstart->lend, inclusive. |
692 | * | |
0e056eb5 | 693 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 694 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
695 | * |
696 | * Return: error status of the address space. | |
469eb4d0 | 697 | */ |
1da177e4 LT |
698 | int filemap_write_and_wait_range(struct address_space *mapping, |
699 | loff_t lstart, loff_t lend) | |
700 | { | |
28fd1298 | 701 | int err = 0; |
1da177e4 | 702 | |
9326c9b2 | 703 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
704 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
705 | WB_SYNC_ALL); | |
ddf8f376 IW |
706 | /* |
707 | * Even if the above returned error, the pages may be | |
708 | * written partially (e.g. -ENOSPC), so we wait for it. | |
709 | * But the -EIO is special case, it may indicate the worst | |
710 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
711 | */ | |
28fd1298 | 712 | if (err != -EIO) { |
94004ed7 CH |
713 | int err2 = filemap_fdatawait_range(mapping, |
714 | lstart, lend); | |
28fd1298 OH |
715 | if (!err) |
716 | err = err2; | |
cbeaf951 JL |
717 | } else { |
718 | /* Clear any previously stored errors */ | |
719 | filemap_check_errors(mapping); | |
28fd1298 | 720 | } |
865ffef3 DM |
721 | } else { |
722 | err = filemap_check_errors(mapping); | |
1da177e4 | 723 | } |
28fd1298 | 724 | return err; |
1da177e4 | 725 | } |
f6995585 | 726 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 727 | |
5660e13d JL |
728 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
729 | { | |
3acdfd28 | 730 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
731 | |
732 | trace_filemap_set_wb_err(mapping, eseq); | |
733 | } | |
734 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
735 | ||
736 | /** | |
737 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
738 | * and advance wb_err to current one | |
739 | * @file: struct file on which the error is being reported | |
740 | * | |
741 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
742 | * want to report any writeback errors that occurred since the last fsync (or | |
743 | * since the file was opened if there haven't been any). | |
744 | * | |
745 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
746 | * then just quickly return 0. The file is all caught up. | |
747 | * | |
748 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
749 | * it and try to swap it into place. If it works, or another task beat us | |
750 | * to it with the new value, then update the f_wb_err and return the error | |
751 | * portion. The error at this point must be reported via proper channels | |
752 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
753 | * | |
754 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
755 | * value is protected by the f_lock since we must ensure that it reflects | |
756 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
757 | * |
758 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
759 | */ |
760 | int file_check_and_advance_wb_err(struct file *file) | |
761 | { | |
762 | int err = 0; | |
763 | errseq_t old = READ_ONCE(file->f_wb_err); | |
764 | struct address_space *mapping = file->f_mapping; | |
765 | ||
766 | /* Locklessly handle the common case where nothing has changed */ | |
767 | if (errseq_check(&mapping->wb_err, old)) { | |
768 | /* Something changed, must use slow path */ | |
769 | spin_lock(&file->f_lock); | |
770 | old = file->f_wb_err; | |
771 | err = errseq_check_and_advance(&mapping->wb_err, | |
772 | &file->f_wb_err); | |
773 | trace_file_check_and_advance_wb_err(file, old); | |
774 | spin_unlock(&file->f_lock); | |
775 | } | |
f4e222c5 JL |
776 | |
777 | /* | |
778 | * We're mostly using this function as a drop in replacement for | |
779 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
780 | * that the legacy code would have had on these flags. | |
781 | */ | |
782 | clear_bit(AS_EIO, &mapping->flags); | |
783 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
784 | return err; |
785 | } | |
786 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
787 | ||
788 | /** | |
789 | * file_write_and_wait_range - write out & wait on a file range | |
790 | * @file: file pointing to address_space with pages | |
791 | * @lstart: offset in bytes where the range starts | |
792 | * @lend: offset in bytes where the range ends (inclusive) | |
793 | * | |
794 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
795 | * | |
796 | * Note that @lend is inclusive (describes the last byte to be written) so | |
797 | * that this function can be used to write to the very end-of-file (end = -1). | |
798 | * | |
799 | * After writing out and waiting on the data, we check and advance the | |
800 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
801 | * |
802 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
803 | */ |
804 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
805 | { | |
806 | int err = 0, err2; | |
807 | struct address_space *mapping = file->f_mapping; | |
808 | ||
9326c9b2 | 809 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
810 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
811 | WB_SYNC_ALL); | |
812 | /* See comment of filemap_write_and_wait() */ | |
813 | if (err != -EIO) | |
814 | __filemap_fdatawait_range(mapping, lstart, lend); | |
815 | } | |
816 | err2 = file_check_and_advance_wb_err(file); | |
817 | if (!err) | |
818 | err = err2; | |
819 | return err; | |
820 | } | |
821 | EXPORT_SYMBOL(file_write_and_wait_range); | |
822 | ||
ef6a3c63 MS |
823 | /** |
824 | * replace_page_cache_page - replace a pagecache page with a new one | |
825 | * @old: page to be replaced | |
826 | * @new: page to replace with | |
ef6a3c63 MS |
827 | * |
828 | * This function replaces a page in the pagecache with a new one. On | |
829 | * success it acquires the pagecache reference for the new page and | |
830 | * drops it for the old page. Both the old and new pages must be | |
831 | * locked. This function does not add the new page to the LRU, the | |
832 | * caller must do that. | |
833 | * | |
74d60958 | 834 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 835 | */ |
1f7ef657 | 836 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 837 | { |
d21bba2b MWO |
838 | struct folio *fold = page_folio(old); |
839 | struct folio *fnew = page_folio(new); | |
74d60958 MW |
840 | struct address_space *mapping = old->mapping; |
841 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
842 | pgoff_t offset = old->index; | |
843 | XA_STATE(xas, &mapping->i_pages, offset); | |
ef6a3c63 | 844 | |
309381fe SL |
845 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
846 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
847 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 848 | |
74d60958 MW |
849 | get_page(new); |
850 | new->mapping = mapping; | |
851 | new->index = offset; | |
ef6a3c63 | 852 | |
d21bba2b | 853 | mem_cgroup_migrate(fold, fnew); |
0d1c2072 | 854 | |
30472509 | 855 | xas_lock_irq(&xas); |
74d60958 | 856 | xas_store(&xas, new); |
4165b9b4 | 857 | |
74d60958 MW |
858 | old->mapping = NULL; |
859 | /* hugetlb pages do not participate in page cache accounting. */ | |
860 | if (!PageHuge(old)) | |
0d1c2072 | 861 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 862 | if (!PageHuge(new)) |
0d1c2072 | 863 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 864 | if (PageSwapBacked(old)) |
0d1c2072 | 865 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 866 | if (PageSwapBacked(new)) |
0d1c2072 | 867 | __inc_lruvec_page_state(new, NR_SHMEM); |
30472509 | 868 | xas_unlock_irq(&xas); |
74d60958 MW |
869 | if (freepage) |
870 | freepage(old); | |
871 | put_page(old); | |
ef6a3c63 MS |
872 | } |
873 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
874 | ||
9dd3d069 MWO |
875 | noinline int __filemap_add_folio(struct address_space *mapping, |
876 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) | |
1da177e4 | 877 | { |
9dd3d069 MWO |
878 | XA_STATE(xas, &mapping->i_pages, index); |
879 | int huge = folio_test_hugetlb(folio); | |
e286781d | 880 | int error; |
da74240e | 881 | bool charged = false; |
e286781d | 882 | |
9dd3d069 MWO |
883 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
884 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); | |
74d60958 | 885 | mapping_set_update(&xas, mapping); |
e286781d | 886 | |
9dd3d069 MWO |
887 | folio_get(folio); |
888 | folio->mapping = mapping; | |
889 | folio->index = index; | |
66a0c8ee | 890 | |
3fea5a49 | 891 | if (!huge) { |
9dd3d069 MWO |
892 | error = mem_cgroup_charge(folio, NULL, gfp); |
893 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); | |
3fea5a49 JW |
894 | if (error) |
895 | goto error; | |
da74240e | 896 | charged = true; |
3fea5a49 JW |
897 | } |
898 | ||
198b62f8 MWO |
899 | gfp &= GFP_RECLAIM_MASK; |
900 | ||
74d60958 | 901 | do { |
198b62f8 MWO |
902 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
903 | void *entry, *old = NULL; | |
904 | ||
9dd3d069 | 905 | if (order > folio_order(folio)) |
198b62f8 MWO |
906 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
907 | order, gfp); | |
74d60958 | 908 | xas_lock_irq(&xas); |
198b62f8 MWO |
909 | xas_for_each_conflict(&xas, entry) { |
910 | old = entry; | |
911 | if (!xa_is_value(entry)) { | |
912 | xas_set_err(&xas, -EEXIST); | |
913 | goto unlock; | |
914 | } | |
915 | } | |
916 | ||
917 | if (old) { | |
918 | if (shadowp) | |
919 | *shadowp = old; | |
920 | /* entry may have been split before we acquired lock */ | |
921 | order = xa_get_order(xas.xa, xas.xa_index); | |
9dd3d069 | 922 | if (order > folio_order(folio)) { |
198b62f8 MWO |
923 | xas_split(&xas, old, order); |
924 | xas_reset(&xas); | |
925 | } | |
926 | } | |
927 | ||
9dd3d069 | 928 | xas_store(&xas, folio); |
74d60958 MW |
929 | if (xas_error(&xas)) |
930 | goto unlock; | |
931 | ||
74d60958 MW |
932 | mapping->nrpages++; |
933 | ||
934 | /* hugetlb pages do not participate in page cache accounting */ | |
935 | if (!huge) | |
9dd3d069 | 936 | __lruvec_stat_add_folio(folio, NR_FILE_PAGES); |
74d60958 MW |
937 | unlock: |
938 | xas_unlock_irq(&xas); | |
198b62f8 | 939 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 940 | |
3fea5a49 JW |
941 | if (xas_error(&xas)) { |
942 | error = xas_error(&xas); | |
da74240e | 943 | if (charged) |
9dd3d069 | 944 | mem_cgroup_uncharge(folio); |
74d60958 | 945 | goto error; |
3fea5a49 | 946 | } |
4165b9b4 | 947 | |
9dd3d069 | 948 | trace_mm_filemap_add_to_page_cache(&folio->page); |
66a0c8ee | 949 | return 0; |
74d60958 | 950 | error: |
9dd3d069 | 951 | folio->mapping = NULL; |
66a0c8ee | 952 | /* Leave page->index set: truncation relies upon it */ |
9dd3d069 | 953 | folio_put(folio); |
3fea5a49 | 954 | return error; |
1da177e4 | 955 | } |
9dd3d069 | 956 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
a528910e JW |
957 | |
958 | /** | |
959 | * add_to_page_cache_locked - add a locked page to the pagecache | |
960 | * @page: page to add | |
961 | * @mapping: the page's address_space | |
962 | * @offset: page index | |
963 | * @gfp_mask: page allocation mode | |
964 | * | |
965 | * This function is used to add a page to the pagecache. It must be locked. | |
966 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
967 | * |
968 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
969 | */ |
970 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
971 | pgoff_t offset, gfp_t gfp_mask) | |
972 | { | |
9dd3d069 | 973 | return __filemap_add_folio(mapping, page_folio(page), offset, |
a528910e JW |
974 | gfp_mask, NULL); |
975 | } | |
e286781d | 976 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 | 977 | |
9dd3d069 MWO |
978 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
979 | pgoff_t index, gfp_t gfp) | |
1da177e4 | 980 | { |
a528910e | 981 | void *shadow = NULL; |
4f98a2fe RR |
982 | int ret; |
983 | ||
9dd3d069 MWO |
984 | __folio_set_locked(folio); |
985 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); | |
a528910e | 986 | if (unlikely(ret)) |
9dd3d069 | 987 | __folio_clear_locked(folio); |
a528910e JW |
988 | else { |
989 | /* | |
9dd3d069 | 990 | * The folio might have been evicted from cache only |
a528910e | 991 | * recently, in which case it should be activated like |
9dd3d069 MWO |
992 | * any other repeatedly accessed folio. |
993 | * The exception is folios getting rewritten; evicting other | |
f0281a00 RR |
994 | * data from the working set, only to cache data that will |
995 | * get overwritten with something else, is a waste of memory. | |
a528910e | 996 | */ |
9dd3d069 MWO |
997 | WARN_ON_ONCE(folio_test_active(folio)); |
998 | if (!(gfp & __GFP_WRITE) && shadow) | |
999 | workingset_refault(folio, shadow); | |
1000 | folio_add_lru(folio); | |
a528910e | 1001 | } |
1da177e4 LT |
1002 | return ret; |
1003 | } | |
9dd3d069 | 1004 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
1da177e4 | 1005 | |
44110fe3 | 1006 | #ifdef CONFIG_NUMA |
bb3c579e | 1007 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
44110fe3 | 1008 | { |
c0ff7453 | 1009 | int n; |
bb3c579e | 1010 | struct folio *folio; |
c0ff7453 | 1011 | |
44110fe3 | 1012 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
1013 | unsigned int cpuset_mems_cookie; |
1014 | do { | |
d26914d1 | 1015 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 1016 | n = cpuset_mem_spread_node(); |
bb3c579e MWO |
1017 | folio = __folio_alloc_node(gfp, order, n); |
1018 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); | |
cc9a6c87 | 1019 | |
bb3c579e | 1020 | return folio; |
44110fe3 | 1021 | } |
bb3c579e | 1022 | return folio_alloc(gfp, order); |
44110fe3 | 1023 | } |
bb3c579e | 1024 | EXPORT_SYMBOL(filemap_alloc_folio); |
44110fe3 PJ |
1025 | #endif |
1026 | ||
7506ae6a JK |
1027 | /* |
1028 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings | |
1029 | * | |
1030 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. | |
1031 | * | |
1032 | * @mapping1: the first mapping to lock | |
1033 | * @mapping2: the second mapping to lock | |
1034 | */ | |
1035 | void filemap_invalidate_lock_two(struct address_space *mapping1, | |
1036 | struct address_space *mapping2) | |
1037 | { | |
1038 | if (mapping1 > mapping2) | |
1039 | swap(mapping1, mapping2); | |
1040 | if (mapping1) | |
1041 | down_write(&mapping1->invalidate_lock); | |
1042 | if (mapping2 && mapping1 != mapping2) | |
1043 | down_write_nested(&mapping2->invalidate_lock, 1); | |
1044 | } | |
1045 | EXPORT_SYMBOL(filemap_invalidate_lock_two); | |
1046 | ||
1047 | /* | |
1048 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings | |
1049 | * | |
1050 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. | |
1051 | * | |
1052 | * @mapping1: the first mapping to unlock | |
1053 | * @mapping2: the second mapping to unlock | |
1054 | */ | |
1055 | void filemap_invalidate_unlock_two(struct address_space *mapping1, | |
1056 | struct address_space *mapping2) | |
1057 | { | |
1058 | if (mapping1) | |
1059 | up_write(&mapping1->invalidate_lock); | |
1060 | if (mapping2 && mapping1 != mapping2) | |
1061 | up_write(&mapping2->invalidate_lock); | |
1062 | } | |
1063 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); | |
1064 | ||
1da177e4 LT |
1065 | /* |
1066 | * In order to wait for pages to become available there must be | |
1067 | * waitqueues associated with pages. By using a hash table of | |
1068 | * waitqueues where the bucket discipline is to maintain all | |
1069 | * waiters on the same queue and wake all when any of the pages | |
1070 | * become available, and for the woken contexts to check to be | |
1071 | * sure the appropriate page became available, this saves space | |
1072 | * at a cost of "thundering herd" phenomena during rare hash | |
1073 | * collisions. | |
1074 | */ | |
62906027 NP |
1075 | #define PAGE_WAIT_TABLE_BITS 8 |
1076 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
df4d4f12 | 1077 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
62906027 | 1078 | |
df4d4f12 | 1079 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
1da177e4 | 1080 | { |
df4d4f12 | 1081 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1082 | } |
1da177e4 | 1083 | |
62906027 | 1084 | void __init pagecache_init(void) |
1da177e4 | 1085 | { |
62906027 | 1086 | int i; |
1da177e4 | 1087 | |
62906027 | 1088 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
df4d4f12 | 1089 | init_waitqueue_head(&folio_wait_table[i]); |
62906027 NP |
1090 | |
1091 | page_writeback_init(); | |
1da177e4 | 1092 | } |
1da177e4 | 1093 | |
5ef64cc8 LT |
1094 | /* |
1095 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1096 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1097 | * one. |
1098 | * | |
1099 | * We have: | |
1100 | * | |
1101 | * (a) no special bits set: | |
1102 | * | |
1103 | * We're just waiting for the bit to be released, and when a waker | |
1104 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1105 | * and remove it from the wait queue. | |
1106 | * | |
1107 | * Simple and straightforward. | |
1108 | * | |
1109 | * (b) WQ_FLAG_EXCLUSIVE: | |
1110 | * | |
1111 | * The waiter is waiting to get the lock, and only one waiter should | |
1112 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1113 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1114 | * | |
1115 | * This is the traditional exclusive wait. | |
1116 | * | |
5868ec26 | 1117 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1118 | * |
1119 | * The waiter is waiting to get the bit, and additionally wants the | |
1120 | * lock to be transferred to it for fair lock behavior. If the lock | |
1121 | * cannot be taken, we stop walking the wait queue without waking | |
1122 | * the waiter. | |
1123 | * | |
1124 | * This is the "fair lock handoff" case, and in addition to setting | |
1125 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1126 | * that it now has the lock. | |
1127 | */ | |
ac6424b9 | 1128 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1129 | { |
5ef64cc8 | 1130 | unsigned int flags; |
62906027 NP |
1131 | struct wait_page_key *key = arg; |
1132 | struct wait_page_queue *wait_page | |
1133 | = container_of(wait, struct wait_page_queue, wait); | |
1134 | ||
cdc8fcb4 | 1135 | if (!wake_page_match(wait_page, key)) |
62906027 | 1136 | return 0; |
3510ca20 | 1137 | |
9a1ea439 | 1138 | /* |
5ef64cc8 LT |
1139 | * If it's a lock handoff wait, we get the bit for it, and |
1140 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1141 | */ |
5ef64cc8 LT |
1142 | flags = wait->flags; |
1143 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
df4d4f12 | 1144 | if (test_bit(key->bit_nr, &key->folio->flags)) |
2a9127fc | 1145 | return -1; |
5ef64cc8 | 1146 | if (flags & WQ_FLAG_CUSTOM) { |
df4d4f12 | 1147 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
5ef64cc8 LT |
1148 | return -1; |
1149 | flags |= WQ_FLAG_DONE; | |
1150 | } | |
2a9127fc | 1151 | } |
f62e00cc | 1152 | |
5ef64cc8 LT |
1153 | /* |
1154 | * We are holding the wait-queue lock, but the waiter that | |
1155 | * is waiting for this will be checking the flags without | |
1156 | * any locking. | |
1157 | * | |
1158 | * So update the flags atomically, and wake up the waiter | |
1159 | * afterwards to avoid any races. This store-release pairs | |
101c0bf6 | 1160 | * with the load-acquire in folio_wait_bit_common(). |
5ef64cc8 LT |
1161 | */ |
1162 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1163 | wake_up_state(wait->private, mode); |
1164 | ||
1165 | /* | |
1166 | * Ok, we have successfully done what we're waiting for, | |
1167 | * and we can unconditionally remove the wait entry. | |
1168 | * | |
5ef64cc8 LT |
1169 | * Note that this pairs with the "finish_wait()" in the |
1170 | * waiter, and has to be the absolute last thing we do. | |
1171 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1172 | * might be de-allocated and the process might even have |
1173 | * exited. | |
2a9127fc | 1174 | */ |
c6fe44d9 | 1175 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1176 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1177 | } |
1178 | ||
6974d7c9 | 1179 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
cbbce822 | 1180 | { |
df4d4f12 | 1181 | wait_queue_head_t *q = folio_waitqueue(folio); |
62906027 NP |
1182 | struct wait_page_key key; |
1183 | unsigned long flags; | |
11a19c7b | 1184 | wait_queue_entry_t bookmark; |
cbbce822 | 1185 | |
df4d4f12 | 1186 | key.folio = folio; |
62906027 NP |
1187 | key.bit_nr = bit_nr; |
1188 | key.page_match = 0; | |
1189 | ||
11a19c7b TC |
1190 | bookmark.flags = 0; |
1191 | bookmark.private = NULL; | |
1192 | bookmark.func = NULL; | |
1193 | INIT_LIST_HEAD(&bookmark.entry); | |
1194 | ||
62906027 | 1195 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1196 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1197 | ||
1198 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1199 | /* | |
1200 | * Take a breather from holding the lock, | |
1201 | * allow pages that finish wake up asynchronously | |
1202 | * to acquire the lock and remove themselves | |
1203 | * from wait queue | |
1204 | */ | |
1205 | spin_unlock_irqrestore(&q->lock, flags); | |
1206 | cpu_relax(); | |
1207 | spin_lock_irqsave(&q->lock, flags); | |
1208 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1209 | } | |
1210 | ||
62906027 NP |
1211 | /* |
1212 | * It is possible for other pages to have collided on the waitqueue | |
1213 | * hash, so in that case check for a page match. That prevents a long- | |
1214 | * term waiter | |
1215 | * | |
1216 | * It is still possible to miss a case here, when we woke page waiters | |
1217 | * and removed them from the waitqueue, but there are still other | |
1218 | * page waiters. | |
1219 | */ | |
1220 | if (!waitqueue_active(q) || !key.page_match) { | |
6974d7c9 | 1221 | folio_clear_waiters(folio); |
62906027 NP |
1222 | /* |
1223 | * It's possible to miss clearing Waiters here, when we woke | |
1224 | * our page waiters, but the hashed waitqueue has waiters for | |
1225 | * other pages on it. | |
1226 | * | |
1227 | * That's okay, it's a rare case. The next waker will clear it. | |
1228 | */ | |
1229 | } | |
1230 | spin_unlock_irqrestore(&q->lock, flags); | |
1231 | } | |
74d81bfa | 1232 | |
4268b480 | 1233 | static void folio_wake(struct folio *folio, int bit) |
74d81bfa | 1234 | { |
4268b480 | 1235 | if (!folio_test_waiters(folio)) |
74d81bfa | 1236 | return; |
6974d7c9 | 1237 | folio_wake_bit(folio, bit); |
74d81bfa | 1238 | } |
62906027 | 1239 | |
9a1ea439 | 1240 | /* |
101c0bf6 | 1241 | * A choice of three behaviors for folio_wait_bit_common(): |
9a1ea439 HD |
1242 | */ |
1243 | enum behavior { | |
1244 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
7c23c782 | 1245 | * __folio_lock() waiting on then setting PG_locked. |
9a1ea439 HD |
1246 | */ |
1247 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1248 | * wait_on_page_writeback() waiting on PG_writeback. | |
1249 | */ | |
1250 | DROP, /* Drop ref to page before wait, no check when woken, | |
1251 | * like put_and_wait_on_page_locked() on PG_locked. | |
1252 | */ | |
1253 | }; | |
1254 | ||
2a9127fc | 1255 | /* |
101c0bf6 | 1256 | * Attempt to check (or get) the folio flag, and mark us done |
5ef64cc8 | 1257 | * if successful. |
2a9127fc | 1258 | */ |
101c0bf6 | 1259 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
2a9127fc LT |
1260 | struct wait_queue_entry *wait) |
1261 | { | |
1262 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
101c0bf6 | 1263 | if (test_and_set_bit(bit_nr, &folio->flags)) |
2a9127fc | 1264 | return false; |
101c0bf6 | 1265 | } else if (test_bit(bit_nr, &folio->flags)) |
2a9127fc LT |
1266 | return false; |
1267 | ||
5ef64cc8 | 1268 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1269 | return true; |
1270 | } | |
1271 | ||
5ef64cc8 LT |
1272 | /* How many times do we accept lock stealing from under a waiter? */ |
1273 | int sysctl_page_lock_unfairness = 5; | |
1274 | ||
101c0bf6 MWO |
1275 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
1276 | int state, enum behavior behavior) | |
62906027 | 1277 | { |
df4d4f12 | 1278 | wait_queue_head_t *q = folio_waitqueue(folio); |
5ef64cc8 | 1279 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1280 | struct wait_page_queue wait_page; |
ac6424b9 | 1281 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1282 | bool thrashing = false; |
9a1ea439 | 1283 | bool delayacct = false; |
eb414681 | 1284 | unsigned long pflags; |
62906027 | 1285 | |
eb414681 | 1286 | if (bit_nr == PG_locked && |
101c0bf6 MWO |
1287 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
1288 | if (!folio_test_swapbacked(folio)) { | |
eb414681 | 1289 | delayacct_thrashing_start(); |
9a1ea439 HD |
1290 | delayacct = true; |
1291 | } | |
eb414681 | 1292 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1293 | thrashing = true; |
1294 | } | |
1295 | ||
62906027 NP |
1296 | init_wait(wait); |
1297 | wait->func = wake_page_function; | |
df4d4f12 | 1298 | wait_page.folio = folio; |
62906027 NP |
1299 | wait_page.bit_nr = bit_nr; |
1300 | ||
5ef64cc8 LT |
1301 | repeat: |
1302 | wait->flags = 0; | |
1303 | if (behavior == EXCLUSIVE) { | |
1304 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1305 | if (--unfairness < 0) | |
1306 | wait->flags |= WQ_FLAG_CUSTOM; | |
1307 | } | |
1308 | ||
2a9127fc LT |
1309 | /* |
1310 | * Do one last check whether we can get the | |
1311 | * page bit synchronously. | |
1312 | * | |
101c0bf6 | 1313 | * Do the folio_set_waiters() marking before that |
2a9127fc LT |
1314 | * to let any waker we _just_ missed know they |
1315 | * need to wake us up (otherwise they'll never | |
1316 | * even go to the slow case that looks at the | |
1317 | * page queue), and add ourselves to the wait | |
1318 | * queue if we need to sleep. | |
1319 | * | |
1320 | * This part needs to be done under the queue | |
1321 | * lock to avoid races. | |
1322 | */ | |
1323 | spin_lock_irq(&q->lock); | |
101c0bf6 MWO |
1324 | folio_set_waiters(folio); |
1325 | if (!folio_trylock_flag(folio, bit_nr, wait)) | |
2a9127fc LT |
1326 | __add_wait_queue_entry_tail(q, wait); |
1327 | spin_unlock_irq(&q->lock); | |
62906027 | 1328 | |
2a9127fc LT |
1329 | /* |
1330 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1331 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1332 | * see whether the page bit testing has already | |
1333 | * been done by the wake function. | |
2a9127fc | 1334 | * |
101c0bf6 | 1335 | * We can drop our reference to the folio. |
2a9127fc LT |
1336 | */ |
1337 | if (behavior == DROP) | |
101c0bf6 | 1338 | folio_put(folio); |
62906027 | 1339 | |
5ef64cc8 LT |
1340 | /* |
1341 | * Note that until the "finish_wait()", or until | |
1342 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1343 | * be very careful with the 'wait->flags', because | |
1344 | * we may race with a waker that sets them. | |
1345 | */ | |
2a9127fc | 1346 | for (;;) { |
5ef64cc8 LT |
1347 | unsigned int flags; |
1348 | ||
62906027 NP |
1349 | set_current_state(state); |
1350 | ||
5ef64cc8 LT |
1351 | /* Loop until we've been woken or interrupted */ |
1352 | flags = smp_load_acquire(&wait->flags); | |
1353 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1354 | if (signal_pending_state(state, current)) | |
1355 | break; | |
1356 | ||
1357 | io_schedule(); | |
1358 | continue; | |
1359 | } | |
1360 | ||
1361 | /* If we were non-exclusive, we're done */ | |
1362 | if (behavior != EXCLUSIVE) | |
a8b169af | 1363 | break; |
9a1ea439 | 1364 | |
5ef64cc8 LT |
1365 | /* If the waker got the lock for us, we're done */ |
1366 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1367 | break; |
2a9127fc | 1368 | |
5ef64cc8 LT |
1369 | /* |
1370 | * Otherwise, if we're getting the lock, we need to | |
1371 | * try to get it ourselves. | |
1372 | * | |
1373 | * And if that fails, we'll have to retry this all. | |
1374 | */ | |
101c0bf6 | 1375 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
5ef64cc8 LT |
1376 | goto repeat; |
1377 | ||
1378 | wait->flags |= WQ_FLAG_DONE; | |
1379 | break; | |
62906027 NP |
1380 | } |
1381 | ||
5ef64cc8 LT |
1382 | /* |
1383 | * If a signal happened, this 'finish_wait()' may remove the last | |
101c0bf6 | 1384 | * waiter from the wait-queues, but the folio waiters bit will remain |
5ef64cc8 LT |
1385 | * set. That's ok. The next wakeup will take care of it, and trying |
1386 | * to do it here would be difficult and prone to races. | |
1387 | */ | |
62906027 NP |
1388 | finish_wait(q, wait); |
1389 | ||
eb414681 | 1390 | if (thrashing) { |
9a1ea439 | 1391 | if (delayacct) |
eb414681 JW |
1392 | delayacct_thrashing_end(); |
1393 | psi_memstall_leave(&pflags); | |
1394 | } | |
b1d29ba8 | 1395 | |
62906027 | 1396 | /* |
5ef64cc8 LT |
1397 | * NOTE! The wait->flags weren't stable until we've done the |
1398 | * 'finish_wait()', and we could have exited the loop above due | |
1399 | * to a signal, and had a wakeup event happen after the signal | |
1400 | * test but before the 'finish_wait()'. | |
1401 | * | |
1402 | * So only after the finish_wait() can we reliably determine | |
1403 | * if we got woken up or not, so we can now figure out the final | |
1404 | * return value based on that state without races. | |
1405 | * | |
1406 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1407 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1408 | */ |
5ef64cc8 LT |
1409 | if (behavior == EXCLUSIVE) |
1410 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1411 | |
2a9127fc | 1412 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1413 | } |
1414 | ||
101c0bf6 | 1415 | void folio_wait_bit(struct folio *folio, int bit_nr) |
62906027 | 1416 | { |
101c0bf6 | 1417 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 | 1418 | } |
101c0bf6 | 1419 | EXPORT_SYMBOL(folio_wait_bit); |
62906027 | 1420 | |
101c0bf6 | 1421 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
62906027 | 1422 | { |
101c0bf6 | 1423 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1424 | } |
101c0bf6 | 1425 | EXPORT_SYMBOL(folio_wait_bit_killable); |
cbbce822 | 1426 | |
9a1ea439 HD |
1427 | /** |
1428 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1429 | * @page: The page to wait for. | |
48054625 | 1430 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 HD |
1431 | * |
1432 | * The caller should hold a reference on @page. They expect the page to | |
1433 | * become unlocked relatively soon, but do not wish to hold up migration | |
1434 | * (for example) by holding the reference while waiting for the page to | |
1435 | * come unlocked. After this function returns, the caller should not | |
1436 | * dereference @page. | |
48054625 MWO |
1437 | * |
1438 | * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal. | |
9a1ea439 | 1439 | */ |
48054625 | 1440 | int put_and_wait_on_page_locked(struct page *page, int state) |
9a1ea439 | 1441 | { |
101c0bf6 MWO |
1442 | return folio_wait_bit_common(page_folio(page), PG_locked, state, |
1443 | DROP); | |
9a1ea439 HD |
1444 | } |
1445 | ||
385e1ca5 | 1446 | /** |
df4d4f12 MWO |
1447 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
1448 | * @folio: Folio defining the wait queue of interest | |
697f619f | 1449 | * @waiter: Waiter to add to the queue |
385e1ca5 | 1450 | * |
df4d4f12 | 1451 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
385e1ca5 | 1452 | */ |
df4d4f12 | 1453 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
385e1ca5 | 1454 | { |
df4d4f12 | 1455 | wait_queue_head_t *q = folio_waitqueue(folio); |
385e1ca5 DH |
1456 | unsigned long flags; |
1457 | ||
1458 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1459 | __add_wait_queue_entry_tail(q, waiter); |
df4d4f12 | 1460 | folio_set_waiters(folio); |
385e1ca5 DH |
1461 | spin_unlock_irqrestore(&q->lock, flags); |
1462 | } | |
df4d4f12 | 1463 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
385e1ca5 | 1464 | |
b91e1302 LT |
1465 | #ifndef clear_bit_unlock_is_negative_byte |
1466 | ||
1467 | /* | |
1468 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1469 | * | |
1470 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1471 | * test the sign bit at the same time. But if the architecture does | |
1472 | * not support that special operation, we just do this all by hand | |
1473 | * instead. | |
1474 | * | |
1475 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1476 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1477 | * in the same byte as PG_locked. |
1478 | */ | |
1479 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1480 | { | |
1481 | clear_bit_unlock(nr, mem); | |
1482 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1483 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1484 | } |
1485 | ||
1486 | #endif | |
1487 | ||
1da177e4 | 1488 | /** |
4e136428 MWO |
1489 | * folio_unlock - Unlock a locked folio. |
1490 | * @folio: The folio. | |
1491 | * | |
1492 | * Unlocks the folio and wakes up any thread sleeping on the page lock. | |
1493 | * | |
1494 | * Context: May be called from interrupt or process context. May not be | |
1495 | * called from NMI context. | |
1da177e4 | 1496 | */ |
4e136428 | 1497 | void folio_unlock(struct folio *folio) |
1da177e4 | 1498 | { |
4e136428 | 1499 | /* Bit 7 allows x86 to check the byte's sign bit */ |
b91e1302 | 1500 | BUILD_BUG_ON(PG_waiters != 7); |
4e136428 MWO |
1501 | BUILD_BUG_ON(PG_locked > 7); |
1502 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1503 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) | |
6974d7c9 | 1504 | folio_wake_bit(folio, PG_locked); |
1da177e4 | 1505 | } |
4e136428 | 1506 | EXPORT_SYMBOL(folio_unlock); |
1da177e4 | 1507 | |
73e10ded | 1508 | /** |
b47393f8 MWO |
1509 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
1510 | * @folio: The folio. | |
73e10ded | 1511 | * |
b47393f8 MWO |
1512 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
1513 | * it. The folio reference held for PG_private_2 being set is released. | |
73e10ded | 1514 | * |
b47393f8 MWO |
1515 | * This is, for example, used when a netfs folio is being written to a local |
1516 | * disk cache, thereby allowing writes to the cache for the same folio to be | |
73e10ded DH |
1517 | * serialised. |
1518 | */ | |
b47393f8 | 1519 | void folio_end_private_2(struct folio *folio) |
73e10ded | 1520 | { |
6974d7c9 MWO |
1521 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
1522 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); | |
1523 | folio_wake_bit(folio, PG_private_2); | |
1524 | folio_put(folio); | |
73e10ded | 1525 | } |
b47393f8 | 1526 | EXPORT_SYMBOL(folio_end_private_2); |
73e10ded DH |
1527 | |
1528 | /** | |
b47393f8 MWO |
1529 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
1530 | * @folio: The folio to wait on. | |
73e10ded | 1531 | * |
b47393f8 | 1532 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
73e10ded | 1533 | */ |
b47393f8 | 1534 | void folio_wait_private_2(struct folio *folio) |
73e10ded | 1535 | { |
101c0bf6 MWO |
1536 | while (folio_test_private_2(folio)) |
1537 | folio_wait_bit(folio, PG_private_2); | |
73e10ded | 1538 | } |
b47393f8 | 1539 | EXPORT_SYMBOL(folio_wait_private_2); |
73e10ded DH |
1540 | |
1541 | /** | |
b47393f8 MWO |
1542 | * folio_wait_private_2_killable - 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 or until a |
73e10ded DH |
1546 | * fatal signal is received by the calling task. |
1547 | * | |
1548 | * Return: | |
1549 | * - 0 if successful. | |
1550 | * - -EINTR if a fatal signal was encountered. | |
1551 | */ | |
b47393f8 | 1552 | int folio_wait_private_2_killable(struct folio *folio) |
73e10ded DH |
1553 | { |
1554 | int ret = 0; | |
1555 | ||
101c0bf6 MWO |
1556 | while (folio_test_private_2(folio)) { |
1557 | ret = folio_wait_bit_killable(folio, PG_private_2); | |
73e10ded DH |
1558 | if (ret < 0) |
1559 | break; | |
1560 | } | |
1561 | ||
1562 | return ret; | |
1563 | } | |
b47393f8 | 1564 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
73e10ded | 1565 | |
485bb99b | 1566 | /** |
4268b480 MWO |
1567 | * folio_end_writeback - End writeback against a folio. |
1568 | * @folio: The folio. | |
1da177e4 | 1569 | */ |
4268b480 | 1570 | void folio_end_writeback(struct folio *folio) |
1da177e4 | 1571 | { |
888cf2db | 1572 | /* |
4268b480 MWO |
1573 | * folio_test_clear_reclaim() could be used here but it is an |
1574 | * atomic operation and overkill in this particular case. Failing | |
1575 | * to shuffle a folio marked for immediate reclaim is too mild | |
1576 | * a gain to justify taking an atomic operation penalty at the | |
1577 | * end of every folio writeback. | |
888cf2db | 1578 | */ |
4268b480 MWO |
1579 | if (folio_test_reclaim(folio)) { |
1580 | folio_clear_reclaim(folio); | |
575ced1c | 1581 | folio_rotate_reclaimable(folio); |
888cf2db | 1582 | } |
ac6aadb2 | 1583 | |
073861ed | 1584 | /* |
4268b480 | 1585 | * Writeback does not hold a folio reference of its own, relying |
073861ed | 1586 | * on truncation to wait for the clearing of PG_writeback. |
4268b480 MWO |
1587 | * But here we must make sure that the folio is not freed and |
1588 | * reused before the folio_wake(). | |
073861ed | 1589 | */ |
4268b480 | 1590 | folio_get(folio); |
269ccca3 | 1591 | if (!__folio_end_writeback(folio)) |
ac6aadb2 MS |
1592 | BUG(); |
1593 | ||
4e857c58 | 1594 | smp_mb__after_atomic(); |
4268b480 MWO |
1595 | folio_wake(folio, PG_writeback); |
1596 | folio_put(folio); | |
1da177e4 | 1597 | } |
4268b480 | 1598 | EXPORT_SYMBOL(folio_end_writeback); |
1da177e4 | 1599 | |
57d99845 MW |
1600 | /* |
1601 | * After completing I/O on a page, call this routine to update the page | |
1602 | * flags appropriately | |
1603 | */ | |
c11f0c0b | 1604 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1605 | { |
c11f0c0b | 1606 | if (!is_write) { |
57d99845 MW |
1607 | if (!err) { |
1608 | SetPageUptodate(page); | |
1609 | } else { | |
1610 | ClearPageUptodate(page); | |
1611 | SetPageError(page); | |
1612 | } | |
1613 | unlock_page(page); | |
abf54548 | 1614 | } else { |
57d99845 | 1615 | if (err) { |
dd8416c4 MK |
1616 | struct address_space *mapping; |
1617 | ||
57d99845 | 1618 | SetPageError(page); |
dd8416c4 MK |
1619 | mapping = page_mapping(page); |
1620 | if (mapping) | |
1621 | mapping_set_error(mapping, err); | |
57d99845 MW |
1622 | } |
1623 | end_page_writeback(page); | |
1624 | } | |
1625 | } | |
1626 | EXPORT_SYMBOL_GPL(page_endio); | |
1627 | ||
485bb99b | 1628 | /** |
7c23c782 MWO |
1629 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
1630 | * @folio: The folio to lock | |
1da177e4 | 1631 | */ |
7c23c782 | 1632 | void __folio_lock(struct folio *folio) |
1da177e4 | 1633 | { |
101c0bf6 | 1634 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
9a1ea439 | 1635 | EXCLUSIVE); |
1da177e4 | 1636 | } |
7c23c782 | 1637 | EXPORT_SYMBOL(__folio_lock); |
1da177e4 | 1638 | |
af7f29d9 | 1639 | int __folio_lock_killable(struct folio *folio) |
2687a356 | 1640 | { |
101c0bf6 | 1641 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
9a1ea439 | 1642 | EXCLUSIVE); |
2687a356 | 1643 | } |
af7f29d9 | 1644 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
2687a356 | 1645 | |
ffdc8dab | 1646 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
dd3e6d50 | 1647 | { |
df4d4f12 | 1648 | struct wait_queue_head *q = folio_waitqueue(folio); |
f32b5dd7 MWO |
1649 | int ret = 0; |
1650 | ||
df4d4f12 | 1651 | wait->folio = folio; |
f32b5dd7 MWO |
1652 | wait->bit_nr = PG_locked; |
1653 | ||
1654 | spin_lock_irq(&q->lock); | |
1655 | __add_wait_queue_entry_tail(q, &wait->wait); | |
ffdc8dab MWO |
1656 | folio_set_waiters(folio); |
1657 | ret = !folio_trylock(folio); | |
f32b5dd7 MWO |
1658 | /* |
1659 | * If we were successful now, we know we're still on the | |
1660 | * waitqueue as we're still under the lock. This means it's | |
1661 | * safe to remove and return success, we know the callback | |
1662 | * isn't going to trigger. | |
1663 | */ | |
1664 | if (!ret) | |
1665 | __remove_wait_queue(q, &wait->wait); | |
1666 | else | |
1667 | ret = -EIOCBQUEUED; | |
1668 | spin_unlock_irq(&q->lock); | |
1669 | return ret; | |
dd3e6d50 JA |
1670 | } |
1671 | ||
9a95f3cf PC |
1672 | /* |
1673 | * Return values: | |
9138e47e MWO |
1674 | * true - folio is locked; mmap_lock is still held. |
1675 | * false - folio is not locked. | |
3e4e28c5 | 1676 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1677 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1678 | * which case mmap_lock is still held. |
9a95f3cf | 1679 | * |
9138e47e MWO |
1680 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
1681 | * with the folio locked and the mmap_lock unperturbed. | |
9a95f3cf | 1682 | */ |
9138e47e | 1683 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
d065bd81 ML |
1684 | unsigned int flags) |
1685 | { | |
4064b982 | 1686 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1687 | /* |
c1e8d7c6 | 1688 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1689 | * even though return 0. |
1690 | */ | |
1691 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
9138e47e | 1692 | return false; |
37b23e05 | 1693 | |
d8ed45c5 | 1694 | mmap_read_unlock(mm); |
37b23e05 | 1695 | if (flags & FAULT_FLAG_KILLABLE) |
6baa8d60 | 1696 | folio_wait_locked_killable(folio); |
37b23e05 | 1697 | else |
6baa8d60 | 1698 | folio_wait_locked(folio); |
9138e47e | 1699 | return false; |
800bca7c HL |
1700 | } |
1701 | if (flags & FAULT_FLAG_KILLABLE) { | |
9138e47e | 1702 | bool ret; |
37b23e05 | 1703 | |
af7f29d9 | 1704 | ret = __folio_lock_killable(folio); |
800bca7c HL |
1705 | if (ret) { |
1706 | mmap_read_unlock(mm); | |
9138e47e | 1707 | return false; |
800bca7c HL |
1708 | } |
1709 | } else { | |
af7f29d9 | 1710 | __folio_lock(folio); |
d065bd81 | 1711 | } |
800bca7c | 1712 | |
9138e47e | 1713 | return true; |
d065bd81 ML |
1714 | } |
1715 | ||
e7b563bb | 1716 | /** |
0d3f9296 MW |
1717 | * page_cache_next_miss() - Find the next gap in the page cache. |
1718 | * @mapping: Mapping. | |
1719 | * @index: Index. | |
1720 | * @max_scan: Maximum range to search. | |
e7b563bb | 1721 | * |
0d3f9296 MW |
1722 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1723 | * gap with the lowest index. | |
e7b563bb | 1724 | * |
0d3f9296 MW |
1725 | * This function may be called under the rcu_read_lock. However, this will |
1726 | * not atomically search a snapshot of the cache at a single point in time. | |
1727 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1728 | * created at index 10, page_cache_next_miss covering both indices may | |
1729 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1730 | * |
0d3f9296 MW |
1731 | * Return: The index of the gap if found, otherwise an index outside the |
1732 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1733 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1734 | */ |
0d3f9296 | 1735 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1736 | pgoff_t index, unsigned long max_scan) |
1737 | { | |
0d3f9296 | 1738 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1739 | |
0d3f9296 MW |
1740 | while (max_scan--) { |
1741 | void *entry = xas_next(&xas); | |
1742 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1743 | break; |
0d3f9296 | 1744 | if (xas.xa_index == 0) |
e7b563bb JW |
1745 | break; |
1746 | } | |
1747 | ||
0d3f9296 | 1748 | return xas.xa_index; |
e7b563bb | 1749 | } |
0d3f9296 | 1750 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1751 | |
1752 | /** | |
2346a560 | 1753 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1754 | * @mapping: Mapping. |
1755 | * @index: Index. | |
1756 | * @max_scan: Maximum range to search. | |
e7b563bb | 1757 | * |
0d3f9296 MW |
1758 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1759 | * gap with the highest index. | |
e7b563bb | 1760 | * |
0d3f9296 MW |
1761 | * This function may be called under the rcu_read_lock. However, this will |
1762 | * not atomically search a snapshot of the cache at a single point in time. | |
1763 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1764 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1765 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1766 | * |
0d3f9296 MW |
1767 | * Return: The index of the gap if found, otherwise an index outside the |
1768 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1769 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1770 | */ |
0d3f9296 | 1771 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1772 | pgoff_t index, unsigned long max_scan) |
1773 | { | |
0d3f9296 | 1774 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1775 | |
0d3f9296 MW |
1776 | while (max_scan--) { |
1777 | void *entry = xas_prev(&xas); | |
1778 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1779 | break; |
0d3f9296 | 1780 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1781 | break; |
1782 | } | |
1783 | ||
0d3f9296 | 1784 | return xas.xa_index; |
e7b563bb | 1785 | } |
0d3f9296 | 1786 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1787 | |
020853b6 MWO |
1788 | /* |
1789 | * Lockless page cache protocol: | |
1790 | * On the lookup side: | |
1791 | * 1. Load the folio from i_pages | |
1792 | * 2. Increment the refcount if it's not zero | |
1793 | * 3. If the folio is not found by xas_reload(), put the refcount and retry | |
1794 | * | |
1795 | * On the removal side: | |
1796 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) | |
1797 | * B. Remove the page from i_pages | |
1798 | * C. Return the page to the page allocator | |
1799 | * | |
1800 | * This means that any page may have its reference count temporarily | |
1801 | * increased by a speculative page cache (or fast GUP) lookup as it can | |
1802 | * be allocated by another user before the RCU grace period expires. | |
1803 | * Because the refcount temporarily acquired here may end up being the | |
1804 | * last refcount on the page, any page allocation must be freeable by | |
1805 | * folio_put(). | |
1806 | */ | |
1807 | ||
44835d20 | 1808 | /* |
bc5a3011 | 1809 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1810 | * @mapping: the address_space to search |
a6de4b48 | 1811 | * @index: The page cache index. |
0cd6144a | 1812 | * |
bca65eea MWO |
1813 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
1814 | * it is returned with an increased refcount. If it is a shadow entry | |
1815 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, | |
1816 | * it is returned without further action. | |
485bb99b | 1817 | * |
bca65eea | 1818 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
1da177e4 | 1819 | */ |
bca65eea | 1820 | static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1821 | { |
a6de4b48 | 1822 | XA_STATE(xas, &mapping->i_pages, index); |
bca65eea | 1823 | struct folio *folio; |
1da177e4 | 1824 | |
a60637c8 NP |
1825 | rcu_read_lock(); |
1826 | repeat: | |
4c7472c0 | 1827 | xas_reset(&xas); |
bca65eea MWO |
1828 | folio = xas_load(&xas); |
1829 | if (xas_retry(&xas, folio)) | |
4c7472c0 MW |
1830 | goto repeat; |
1831 | /* | |
1832 | * A shadow entry of a recently evicted page, or a swap entry from | |
1833 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1834 | */ | |
bca65eea | 1835 | if (!folio || xa_is_value(folio)) |
4c7472c0 | 1836 | goto out; |
83929372 | 1837 | |
bca65eea | 1838 | if (!folio_try_get_rcu(folio)) |
4c7472c0 | 1839 | goto repeat; |
83929372 | 1840 | |
bca65eea MWO |
1841 | if (unlikely(folio != xas_reload(&xas))) { |
1842 | folio_put(folio); | |
4c7472c0 | 1843 | goto repeat; |
a60637c8 | 1844 | } |
27d20fdd | 1845 | out: |
a60637c8 NP |
1846 | rcu_read_unlock(); |
1847 | ||
bca65eea | 1848 | return folio; |
1da177e4 | 1849 | } |
1da177e4 | 1850 | |
0cd6144a | 1851 | /** |
2294b32e MWO |
1852 | * pagecache_get_page - Find and get a reference to a page. |
1853 | * @mapping: The address_space to search. | |
1854 | * @index: The page index. | |
1855 | * @fgp_flags: %FGP flags modify how the page is returned. | |
1856 | * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1857 | * |
2294b32e | 1858 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1859 | * |
2294b32e | 1860 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1861 | * |
2294b32e MWO |
1862 | * * %FGP_ACCESSED - The page will be marked accessed. |
1863 | * * %FGP_LOCK - The page is returned locked. | |
a8cf7f27 MWO |
1864 | * * %FGP_HEAD - If the page is present and a THP, return the head page |
1865 | * rather than the exact page specified by the index. | |
44835d20 MWO |
1866 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
1867 | * instead of allocating a new page to replace it. | |
2294b32e MWO |
1868 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
1869 | * @gfp_mask and added to the page cache and the VM's LRU list. | |
1870 | * The page is returned locked and with an increased refcount. | |
1871 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1872 | * page is already in cache. If the page was allocated, unlock it before | |
1873 | * returning so the caller can do the same dance. | |
605cad83 YS |
1874 | * * %FGP_WRITE - The page will be written |
1875 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask | |
1876 | * * %FGP_NOWAIT - Don't get blocked by page lock | |
1da177e4 | 1877 | * |
2294b32e MWO |
1878 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1879 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1880 | * |
2457aec6 | 1881 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1882 | * |
2294b32e | 1883 | * Return: The found page or %NULL otherwise. |
1da177e4 | 1884 | */ |
2294b32e MWO |
1885 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index, |
1886 | int fgp_flags, gfp_t gfp_mask) | |
1da177e4 | 1887 | { |
eb2be189 | 1888 | struct page *page; |
2457aec6 | 1889 | |
1da177e4 | 1890 | repeat: |
bc5a3011 | 1891 | page = mapping_get_entry(mapping, index); |
44835d20 MWO |
1892 | if (xa_is_value(page)) { |
1893 | if (fgp_flags & FGP_ENTRY) | |
1894 | return page; | |
2457aec6 | 1895 | page = NULL; |
44835d20 | 1896 | } |
2457aec6 MG |
1897 | if (!page) |
1898 | goto no_page; | |
1899 | ||
1900 | if (fgp_flags & FGP_LOCK) { | |
1901 | if (fgp_flags & FGP_NOWAIT) { | |
1902 | if (!trylock_page(page)) { | |
09cbfeaf | 1903 | put_page(page); |
2457aec6 MG |
1904 | return NULL; |
1905 | } | |
1906 | } else { | |
1907 | lock_page(page); | |
1908 | } | |
1909 | ||
1910 | /* Has the page been truncated? */ | |
a8cf7f27 | 1911 | if (unlikely(page->mapping != mapping)) { |
2457aec6 | 1912 | unlock_page(page); |
09cbfeaf | 1913 | put_page(page); |
2457aec6 MG |
1914 | goto repeat; |
1915 | } | |
a8cf7f27 | 1916 | VM_BUG_ON_PAGE(!thp_contains(page, index), page); |
2457aec6 MG |
1917 | } |
1918 | ||
c16eb000 | 1919 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 | 1920 | mark_page_accessed(page); |
b9306a79 YS |
1921 | else if (fgp_flags & FGP_WRITE) { |
1922 | /* Clear idle flag for buffer write */ | |
1923 | if (page_is_idle(page)) | |
1924 | clear_page_idle(page); | |
1925 | } | |
a8cf7f27 MWO |
1926 | if (!(fgp_flags & FGP_HEAD)) |
1927 | page = find_subpage(page, index); | |
2457aec6 MG |
1928 | |
1929 | no_page: | |
1930 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1931 | int err; | |
f56753ac | 1932 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
45f87de5 MH |
1933 | gfp_mask |= __GFP_WRITE; |
1934 | if (fgp_flags & FGP_NOFS) | |
1935 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1936 | |
45f87de5 | 1937 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1938 | if (!page) |
1939 | return NULL; | |
2457aec6 | 1940 | |
a75d4c33 | 1941 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1942 | fgp_flags |= FGP_LOCK; |
1943 | ||
eb39d618 | 1944 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1945 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1946 | __SetPageReferenced(page); |
2457aec6 | 1947 | |
2294b32e | 1948 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); |
eb2be189 | 1949 | if (unlikely(err)) { |
09cbfeaf | 1950 | put_page(page); |
eb2be189 NP |
1951 | page = NULL; |
1952 | if (err == -EEXIST) | |
1953 | goto repeat; | |
1da177e4 | 1954 | } |
a75d4c33 JB |
1955 | |
1956 | /* | |
1957 | * add_to_page_cache_lru locks the page, and for mmap we expect | |
1958 | * an unlocked page. | |
1959 | */ | |
1960 | if (page && (fgp_flags & FGP_FOR_MMAP)) | |
1961 | unlock_page(page); | |
1da177e4 | 1962 | } |
2457aec6 | 1963 | |
1da177e4 LT |
1964 | return page; |
1965 | } | |
2457aec6 | 1966 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1967 | |
c7bad633 MWO |
1968 | static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max, |
1969 | xa_mark_t mark) | |
1970 | { | |
1971 | struct page *page; | |
1972 | ||
1973 | retry: | |
1974 | if (mark == XA_PRESENT) | |
1975 | page = xas_find(xas, max); | |
1976 | else | |
1977 | page = xas_find_marked(xas, max, mark); | |
1978 | ||
1979 | if (xas_retry(xas, page)) | |
1980 | goto retry; | |
1981 | /* | |
1982 | * A shadow entry of a recently evicted page, a swap | |
1983 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1984 | * without attempting to raise page count. | |
1985 | */ | |
1986 | if (!page || xa_is_value(page)) | |
1987 | return page; | |
1988 | ||
1989 | if (!page_cache_get_speculative(page)) | |
1990 | goto reset; | |
1991 | ||
1992 | /* Has the page moved or been split? */ | |
1993 | if (unlikely(page != xas_reload(xas))) { | |
1994 | put_page(page); | |
1995 | goto reset; | |
1996 | } | |
1997 | ||
1998 | return page; | |
1999 | reset: | |
2000 | xas_reset(xas); | |
2001 | goto retry; | |
2002 | } | |
2003 | ||
0cd6144a JW |
2004 | /** |
2005 | * find_get_entries - gang pagecache lookup | |
2006 | * @mapping: The address_space to search | |
2007 | * @start: The starting page cache index | |
ca122fe4 | 2008 | * @end: The final page index (inclusive). |
cf2039af | 2009 | * @pvec: Where the resulting entries are placed. |
0cd6144a JW |
2010 | * @indices: The cache indices corresponding to the entries in @entries |
2011 | * | |
cf2039af MWO |
2012 | * find_get_entries() will search for and return a batch of entries in |
2013 | * the mapping. The entries are placed in @pvec. find_get_entries() | |
2014 | * takes a reference on any actual pages it returns. | |
0cd6144a JW |
2015 | * |
2016 | * The search returns a group of mapping-contiguous page cache entries | |
2017 | * with ascending indexes. There may be holes in the indices due to | |
2018 | * not-present pages. | |
2019 | * | |
139b6a6f JW |
2020 | * Any shadow entries of evicted pages, or swap entries from |
2021 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 2022 | * |
71725ed1 HD |
2023 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
2024 | * stops at that page: the caller is likely to have a better way to handle | |
2025 | * the compound page as a whole, and then skip its extent, than repeatedly | |
2026 | * calling find_get_entries() to return all its tails. | |
2027 | * | |
a862f68a | 2028 | * Return: the number of pages and shadow entries which were found. |
0cd6144a | 2029 | */ |
ca122fe4 | 2030 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
cf2039af | 2031 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) |
0cd6144a | 2032 | { |
f280bf09 MW |
2033 | XA_STATE(xas, &mapping->i_pages, start); |
2034 | struct page *page; | |
0cd6144a | 2035 | unsigned int ret = 0; |
cf2039af | 2036 | unsigned nr_entries = PAGEVEC_SIZE; |
0cd6144a JW |
2037 | |
2038 | rcu_read_lock(); | |
ca122fe4 | 2039 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
71725ed1 HD |
2040 | /* |
2041 | * Terminate early on finding a THP, to allow the caller to | |
2042 | * handle it all at once; but continue if this is hugetlbfs. | |
2043 | */ | |
c7bad633 MWO |
2044 | if (!xa_is_value(page) && PageTransHuge(page) && |
2045 | !PageHuge(page)) { | |
71725ed1 HD |
2046 | page = find_subpage(page, xas.xa_index); |
2047 | nr_entries = ret + 1; | |
2048 | } | |
c7bad633 | 2049 | |
f280bf09 | 2050 | indices[ret] = xas.xa_index; |
cf2039af | 2051 | pvec->pages[ret] = page; |
0cd6144a JW |
2052 | if (++ret == nr_entries) |
2053 | break; | |
2054 | } | |
2055 | rcu_read_unlock(); | |
cf2039af MWO |
2056 | |
2057 | pvec->nr = ret; | |
0cd6144a JW |
2058 | return ret; |
2059 | } | |
2060 | ||
5c211ba2 MWO |
2061 | /** |
2062 | * find_lock_entries - Find a batch of pagecache entries. | |
2063 | * @mapping: The address_space to search. | |
2064 | * @start: The starting page cache index. | |
2065 | * @end: The final page index (inclusive). | |
2066 | * @pvec: Where the resulting entries are placed. | |
2067 | * @indices: The cache indices of the entries in @pvec. | |
2068 | * | |
2069 | * find_lock_entries() will return a batch of entries from @mapping. | |
2070 | * Swap, shadow and DAX entries are included. Pages are returned | |
2071 | * locked and with an incremented refcount. Pages which are locked by | |
2072 | * somebody else or under writeback are skipped. Only the head page of | |
2073 | * a THP is returned. Pages which are partially outside the range are | |
2074 | * not returned. | |
2075 | * | |
2076 | * The entries have ascending indexes. The indices may not be consecutive | |
2077 | * due to not-present entries, THP pages, pages which could not be locked | |
2078 | * or pages under writeback. | |
2079 | * | |
2080 | * Return: The number of entries which were found. | |
2081 | */ | |
2082 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
2083 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) | |
2084 | { | |
2085 | XA_STATE(xas, &mapping->i_pages, start); | |
2086 | struct page *page; | |
2087 | ||
2088 | rcu_read_lock(); | |
2089 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { | |
2090 | if (!xa_is_value(page)) { | |
2091 | if (page->index < start) | |
2092 | goto put; | |
2093 | VM_BUG_ON_PAGE(page->index != xas.xa_index, page); | |
2094 | if (page->index + thp_nr_pages(page) - 1 > end) | |
2095 | goto put; | |
2096 | if (!trylock_page(page)) | |
2097 | goto put; | |
2098 | if (page->mapping != mapping || PageWriteback(page)) | |
2099 | goto unlock; | |
2100 | VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index), | |
2101 | page); | |
2102 | } | |
2103 | indices[pvec->nr] = xas.xa_index; | |
2104 | if (!pagevec_add(pvec, page)) | |
2105 | break; | |
2106 | goto next; | |
2107 | unlock: | |
2108 | unlock_page(page); | |
2109 | put: | |
2110 | put_page(page); | |
2111 | next: | |
2d11e738 HD |
2112 | if (!xa_is_value(page) && PageTransHuge(page)) { |
2113 | unsigned int nr_pages = thp_nr_pages(page); | |
2114 | ||
2115 | /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */ | |
2116 | xas_set(&xas, page->index + nr_pages); | |
2117 | if (xas.xa_index < nr_pages) | |
2118 | break; | |
2119 | } | |
5c211ba2 MWO |
2120 | } |
2121 | rcu_read_unlock(); | |
2122 | ||
2123 | return pagevec_count(pvec); | |
2124 | } | |
2125 | ||
1da177e4 | 2126 | /** |
b947cee4 | 2127 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2128 | * @mapping: The address_space to search |
2129 | * @start: The starting page index | |
b947cee4 | 2130 | * @end: The final page index (inclusive) |
1da177e4 LT |
2131 | * @nr_pages: The maximum number of pages |
2132 | * @pages: Where the resulting pages are placed | |
2133 | * | |
b947cee4 JK |
2134 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2135 | * pages in the mapping starting at index @start and up to index @end | |
2136 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2137 | * a reference against the returned pages. | |
1da177e4 LT |
2138 | * |
2139 | * The search returns a group of mapping-contiguous pages with ascending | |
2140 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2141 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2142 | * |
a862f68a MR |
2143 | * Return: the number of pages which were found. If this number is |
2144 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2145 | * reached. |
1da177e4 | 2146 | */ |
b947cee4 JK |
2147 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2148 | pgoff_t end, unsigned int nr_pages, | |
2149 | struct page **pages) | |
1da177e4 | 2150 | { |
fd1b3cee MW |
2151 | XA_STATE(xas, &mapping->i_pages, *start); |
2152 | struct page *page; | |
0fc9d104 KK |
2153 | unsigned ret = 0; |
2154 | ||
2155 | if (unlikely(!nr_pages)) | |
2156 | return 0; | |
a60637c8 NP |
2157 | |
2158 | rcu_read_lock(); | |
c7bad633 | 2159 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee MW |
2160 | /* Skip over shadow, swap and DAX entries */ |
2161 | if (xa_is_value(page)) | |
8079b1c8 | 2162 | continue; |
a60637c8 | 2163 | |
4101196b | 2164 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 2165 | if (++ret == nr_pages) { |
5d3ee42f | 2166 | *start = xas.xa_index + 1; |
b947cee4 JK |
2167 | goto out; |
2168 | } | |
a60637c8 | 2169 | } |
5b280c0c | 2170 | |
b947cee4 JK |
2171 | /* |
2172 | * We come here when there is no page beyond @end. We take care to not | |
2173 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2174 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2175 | * already broken anyway. |
2176 | */ | |
2177 | if (end == (pgoff_t)-1) | |
2178 | *start = (pgoff_t)-1; | |
2179 | else | |
2180 | *start = end + 1; | |
2181 | out: | |
a60637c8 | 2182 | rcu_read_unlock(); |
d72dc8a2 | 2183 | |
1da177e4 LT |
2184 | return ret; |
2185 | } | |
2186 | ||
ebf43500 JA |
2187 | /** |
2188 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2189 | * @mapping: The address_space to search | |
2190 | * @index: The starting page index | |
2191 | * @nr_pages: The maximum number of pages | |
2192 | * @pages: Where the resulting pages are placed | |
2193 | * | |
2194 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2195 | * that the returned number of pages are guaranteed to be contiguous. | |
2196 | * | |
a862f68a | 2197 | * Return: the number of pages which were found. |
ebf43500 JA |
2198 | */ |
2199 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2200 | unsigned int nr_pages, struct page **pages) | |
2201 | { | |
3ece58a2 MW |
2202 | XA_STATE(xas, &mapping->i_pages, index); |
2203 | struct page *page; | |
0fc9d104 KK |
2204 | unsigned int ret = 0; |
2205 | ||
2206 | if (unlikely(!nr_pages)) | |
2207 | return 0; | |
a60637c8 NP |
2208 | |
2209 | rcu_read_lock(); | |
3ece58a2 | 2210 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2211 | if (xas_retry(&xas, page)) |
2212 | continue; | |
2213 | /* | |
2214 | * If the entry has been swapped out, we can stop looking. | |
2215 | * No current caller is looking for DAX entries. | |
2216 | */ | |
2217 | if (xa_is_value(page)) | |
8079b1c8 | 2218 | break; |
ebf43500 | 2219 | |
4101196b | 2220 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2221 | goto retry; |
83929372 | 2222 | |
4101196b | 2223 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2224 | if (unlikely(page != xas_reload(&xas))) |
2225 | goto put_page; | |
a60637c8 | 2226 | |
4101196b | 2227 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2228 | if (++ret == nr_pages) |
2229 | break; | |
3ece58a2 MW |
2230 | continue; |
2231 | put_page: | |
4101196b | 2232 | put_page(page); |
3ece58a2 MW |
2233 | retry: |
2234 | xas_reset(&xas); | |
ebf43500 | 2235 | } |
a60637c8 NP |
2236 | rcu_read_unlock(); |
2237 | return ret; | |
ebf43500 | 2238 | } |
ef71c15c | 2239 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2240 | |
485bb99b | 2241 | /** |
c49f50d1 | 2242 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2243 | * @mapping: the address_space to search |
2244 | * @index: the starting page index | |
72b045ae | 2245 | * @end: The final page index (inclusive) |
485bb99b RD |
2246 | * @tag: the tag index |
2247 | * @nr_pages: the maximum number of pages | |
2248 | * @pages: where the resulting pages are placed | |
2249 | * | |
c49f50d1 MWO |
2250 | * Like find_get_pages(), except we only return head pages which are tagged |
2251 | * with @tag. @index is updated to the index immediately after the last | |
2252 | * page we return, ready for the next iteration. | |
a862f68a MR |
2253 | * |
2254 | * Return: the number of pages which were found. | |
1da177e4 | 2255 | */ |
72b045ae | 2256 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2257 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2258 | struct page **pages) |
1da177e4 | 2259 | { |
a6906972 MW |
2260 | XA_STATE(xas, &mapping->i_pages, *index); |
2261 | struct page *page; | |
0fc9d104 KK |
2262 | unsigned ret = 0; |
2263 | ||
2264 | if (unlikely(!nr_pages)) | |
2265 | return 0; | |
a60637c8 NP |
2266 | |
2267 | rcu_read_lock(); | |
c7bad633 | 2268 | while ((page = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2269 | /* |
2270 | * Shadow entries should never be tagged, but this iteration | |
2271 | * is lockless so there is a window for page reclaim to evict | |
2272 | * a page we saw tagged. Skip over it. | |
2273 | */ | |
2274 | if (xa_is_value(page)) | |
139b6a6f | 2275 | continue; |
a60637c8 | 2276 | |
c49f50d1 | 2277 | pages[ret] = page; |
72b045ae | 2278 | if (++ret == nr_pages) { |
c49f50d1 | 2279 | *index = page->index + thp_nr_pages(page); |
72b045ae JK |
2280 | goto out; |
2281 | } | |
a60637c8 | 2282 | } |
5b280c0c | 2283 | |
72b045ae | 2284 | /* |
a6906972 | 2285 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2286 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2287 | * iteration when there is a page at index -1 but that is already |
2288 | * broken anyway. | |
72b045ae JK |
2289 | */ |
2290 | if (end == (pgoff_t)-1) | |
2291 | *index = (pgoff_t)-1; | |
2292 | else | |
2293 | *index = end + 1; | |
2294 | out: | |
a60637c8 | 2295 | rcu_read_unlock(); |
1da177e4 | 2296 | |
1da177e4 LT |
2297 | return ret; |
2298 | } | |
72b045ae | 2299 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2300 | |
76d42bd9 WF |
2301 | /* |
2302 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2303 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2304 | * | |
2305 | * ---R__________________________________________B__________ | |
2306 | * ^ reading here ^ bad block(assume 4k) | |
2307 | * | |
2308 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2309 | * => failing the whole request => read(R) => read(R+1) => | |
2310 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2311 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2312 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2313 | * | |
2314 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2315 | */ | |
0f8e2db4 | 2316 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2317 | { |
76d42bd9 | 2318 | ra->ra_pages /= 4; |
76d42bd9 WF |
2319 | } |
2320 | ||
cbd59c48 MWO |
2321 | /* |
2322 | * filemap_get_read_batch - Get a batch of pages for read | |
2323 | * | |
2324 | * Get a batch of pages which represent a contiguous range of bytes | |
2325 | * in the file. No tail pages will be returned. If @index is in the | |
2326 | * middle of a THP, the entire THP will be returned. The last page in | |
2327 | * the batch may have Readahead set or be not Uptodate so that the | |
2328 | * caller can take the appropriate action. | |
2329 | */ | |
2330 | static void filemap_get_read_batch(struct address_space *mapping, | |
2331 | pgoff_t index, pgoff_t max, struct pagevec *pvec) | |
2332 | { | |
2333 | XA_STATE(xas, &mapping->i_pages, index); | |
2334 | struct page *head; | |
2335 | ||
2336 | rcu_read_lock(); | |
2337 | for (head = xas_load(&xas); head; head = xas_next(&xas)) { | |
2338 | if (xas_retry(&xas, head)) | |
2339 | continue; | |
2340 | if (xas.xa_index > max || xa_is_value(head)) | |
2341 | break; | |
2342 | if (!page_cache_get_speculative(head)) | |
2343 | goto retry; | |
2344 | ||
2345 | /* Has the page moved or been split? */ | |
2346 | if (unlikely(head != xas_reload(&xas))) | |
2347 | goto put_page; | |
2348 | ||
2349 | if (!pagevec_add(pvec, head)) | |
2350 | break; | |
2351 | if (!PageUptodate(head)) | |
2352 | break; | |
2353 | if (PageReadahead(head)) | |
2354 | break; | |
2355 | xas.xa_index = head->index + thp_nr_pages(head) - 1; | |
2356 | xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK; | |
2357 | continue; | |
2358 | put_page: | |
2359 | put_page(head); | |
2360 | retry: | |
2361 | xas_reset(&xas); | |
2362 | } | |
2363 | rcu_read_unlock(); | |
2364 | } | |
2365 | ||
68430303 MWO |
2366 | static int filemap_read_page(struct file *file, struct address_space *mapping, |
2367 | struct page *page) | |
723ef24b | 2368 | { |
723ef24b KO |
2369 | int error; |
2370 | ||
723ef24b | 2371 | /* |
68430303 MWO |
2372 | * A previous I/O error may have been due to temporary failures, |
2373 | * eg. multipath errors. PG_error will be set again if readpage | |
2374 | * fails. | |
723ef24b KO |
2375 | */ |
2376 | ClearPageError(page); | |
2377 | /* Start the actual read. The read will unlock the page. */ | |
68430303 MWO |
2378 | error = mapping->a_ops->readpage(file, page); |
2379 | if (error) | |
2380 | return error; | |
723ef24b | 2381 | |
aa1ec2f6 | 2382 | error = wait_on_page_locked_killable(page); |
68430303 MWO |
2383 | if (error) |
2384 | return error; | |
aa1ec2f6 MWO |
2385 | if (PageUptodate(page)) |
2386 | return 0; | |
aa1ec2f6 MWO |
2387 | shrink_readahead_size_eio(&file->f_ra); |
2388 | return -EIO; | |
723ef24b KO |
2389 | } |
2390 | ||
fce70da3 MWO |
2391 | static bool filemap_range_uptodate(struct address_space *mapping, |
2392 | loff_t pos, struct iov_iter *iter, struct page *page) | |
2393 | { | |
2394 | int count; | |
2395 | ||
2396 | if (PageUptodate(page)) | |
2397 | return true; | |
2398 | /* pipes can't handle partially uptodate pages */ | |
2399 | if (iov_iter_is_pipe(iter)) | |
2400 | return false; | |
2401 | if (!mapping->a_ops->is_partially_uptodate) | |
2402 | return false; | |
2403 | if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page))) | |
2404 | return false; | |
2405 | ||
2406 | count = iter->count; | |
2407 | if (page_offset(page) > pos) { | |
2408 | count -= page_offset(page) - pos; | |
2409 | pos = 0; | |
2410 | } else { | |
2411 | pos -= page_offset(page); | |
2412 | } | |
2413 | ||
2414 | return mapping->a_ops->is_partially_uptodate(page, pos, count); | |
2415 | } | |
2416 | ||
4612aeef MWO |
2417 | static int filemap_update_page(struct kiocb *iocb, |
2418 | struct address_space *mapping, struct iov_iter *iter, | |
fce70da3 | 2419 | struct page *page) |
723ef24b | 2420 | { |
ffdc8dab | 2421 | struct folio *folio = page_folio(page); |
723ef24b KO |
2422 | int error; |
2423 | ||
730633f0 JK |
2424 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2425 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2426 | return -EAGAIN; | |
2427 | } else { | |
2428 | filemap_invalidate_lock_shared(mapping); | |
2429 | } | |
2430 | ||
ffdc8dab | 2431 | if (!folio_trylock(folio)) { |
730633f0 | 2432 | error = -EAGAIN; |
87d1d7b6 | 2433 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2434 | goto unlock_mapping; |
87d1d7b6 | 2435 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2436 | filemap_invalidate_unlock_shared(mapping); |
ffdc8dab | 2437 | put_and_wait_on_page_locked(&folio->page, TASK_KILLABLE); |
4612aeef | 2438 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2439 | } |
ffdc8dab | 2440 | error = __folio_lock_async(folio, iocb->ki_waitq); |
87d1d7b6 | 2441 | if (error) |
730633f0 | 2442 | goto unlock_mapping; |
723ef24b | 2443 | } |
723ef24b | 2444 | |
730633f0 | 2445 | error = AOP_TRUNCATED_PAGE; |
ffdc8dab | 2446 | if (!folio->mapping) |
730633f0 | 2447 | goto unlock; |
723ef24b | 2448 | |
fce70da3 | 2449 | error = 0; |
ffdc8dab | 2450 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, &folio->page)) |
fce70da3 MWO |
2451 | goto unlock; |
2452 | ||
2453 | error = -EAGAIN; | |
2454 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2455 | goto unlock; | |
2456 | ||
ffdc8dab | 2457 | error = filemap_read_page(iocb->ki_filp, mapping, &folio->page); |
730633f0 | 2458 | goto unlock_mapping; |
fce70da3 | 2459 | unlock: |
ffdc8dab | 2460 | folio_unlock(folio); |
730633f0 JK |
2461 | unlock_mapping: |
2462 | filemap_invalidate_unlock_shared(mapping); | |
2463 | if (error == AOP_TRUNCATED_PAGE) | |
ffdc8dab | 2464 | folio_put(folio); |
fce70da3 | 2465 | return error; |
723ef24b KO |
2466 | } |
2467 | ||
f253e185 MWO |
2468 | static int filemap_create_page(struct file *file, |
2469 | struct address_space *mapping, pgoff_t index, | |
2470 | struct pagevec *pvec) | |
723ef24b | 2471 | { |
723ef24b KO |
2472 | struct page *page; |
2473 | int error; | |
2474 | ||
723ef24b KO |
2475 | page = page_cache_alloc(mapping); |
2476 | if (!page) | |
f253e185 | 2477 | return -ENOMEM; |
723ef24b | 2478 | |
730633f0 JK |
2479 | /* |
2480 | * Protect against truncate / hole punch. Grabbing invalidate_lock here | |
2481 | * assures we cannot instantiate and bring uptodate new pagecache pages | |
2482 | * after evicting page cache during truncate and before actually | |
2483 | * freeing blocks. Note that we could release invalidate_lock after | |
2484 | * inserting the page into page cache as the locked page would then be | |
2485 | * enough to synchronize with hole punching. But there are code paths | |
2486 | * such as filemap_update_page() filling in partially uptodate pages or | |
2487 | * ->readpages() that need to hold invalidate_lock while mapping blocks | |
2488 | * for IO so let's hold the lock here as well to keep locking rules | |
2489 | * simple. | |
2490 | */ | |
2491 | filemap_invalidate_lock_shared(mapping); | |
723ef24b | 2492 | error = add_to_page_cache_lru(page, mapping, index, |
f253e185 MWO |
2493 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2494 | if (error == -EEXIST) | |
2495 | error = AOP_TRUNCATED_PAGE; | |
2496 | if (error) | |
2497 | goto error; | |
2498 | ||
2499 | error = filemap_read_page(file, mapping, page); | |
2500 | if (error) | |
2501 | goto error; | |
2502 | ||
730633f0 | 2503 | filemap_invalidate_unlock_shared(mapping); |
f253e185 MWO |
2504 | pagevec_add(pvec, page); |
2505 | return 0; | |
2506 | error: | |
730633f0 | 2507 | filemap_invalidate_unlock_shared(mapping); |
68430303 | 2508 | put_page(page); |
f253e185 | 2509 | return error; |
723ef24b KO |
2510 | } |
2511 | ||
5963fe03 MWO |
2512 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
2513 | struct address_space *mapping, struct page *page, | |
2514 | pgoff_t last_index) | |
2515 | { | |
2516 | if (iocb->ki_flags & IOCB_NOIO) | |
2517 | return -EAGAIN; | |
2518 | page_cache_async_readahead(mapping, &file->f_ra, file, page, | |
2519 | page->index, last_index - page->index); | |
2520 | return 0; | |
2521 | } | |
2522 | ||
3a6bae48 | 2523 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
ff993ba1 | 2524 | struct pagevec *pvec) |
06c04442 KO |
2525 | { |
2526 | struct file *filp = iocb->ki_filp; | |
2527 | struct address_space *mapping = filp->f_mapping; | |
2528 | struct file_ra_state *ra = &filp->f_ra; | |
2529 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2530 | pgoff_t last_index; |
2642fca6 | 2531 | struct page *page; |
cbd59c48 | 2532 | int err = 0; |
06c04442 | 2533 | |
cbd59c48 | 2534 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2535 | retry: |
06c04442 KO |
2536 | if (fatal_signal_pending(current)) |
2537 | return -EINTR; | |
2538 | ||
cbd59c48 | 2539 | filemap_get_read_batch(mapping, index, last_index, pvec); |
2642fca6 MWO |
2540 | if (!pagevec_count(pvec)) { |
2541 | if (iocb->ki_flags & IOCB_NOIO) | |
2542 | return -EAGAIN; | |
2543 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2544 | last_index - index); | |
2545 | filemap_get_read_batch(mapping, index, last_index, pvec); | |
2546 | } | |
f253e185 MWO |
2547 | if (!pagevec_count(pvec)) { |
2548 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) | |
2549 | return -EAGAIN; | |
2550 | err = filemap_create_page(filp, mapping, | |
2551 | iocb->ki_pos >> PAGE_SHIFT, pvec); | |
2552 | if (err == AOP_TRUNCATED_PAGE) | |
2642fca6 | 2553 | goto retry; |
f253e185 MWO |
2554 | return err; |
2555 | } | |
06c04442 | 2556 | |
2642fca6 MWO |
2557 | page = pvec->pages[pagevec_count(pvec) - 1]; |
2558 | if (PageReadahead(page)) { | |
2559 | err = filemap_readahead(iocb, filp, mapping, page, last_index); | |
2560 | if (err) | |
2561 | goto err; | |
2562 | } | |
2563 | if (!PageUptodate(page)) { | |
2564 | if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1) | |
2565 | iocb->ki_flags |= IOCB_NOWAIT; | |
2566 | err = filemap_update_page(iocb, mapping, iter, page); | |
2567 | if (err) | |
2568 | goto err; | |
06c04442 KO |
2569 | } |
2570 | ||
2642fca6 | 2571 | return 0; |
cbd59c48 | 2572 | err: |
2642fca6 MWO |
2573 | if (err < 0) |
2574 | put_page(page); | |
2575 | if (likely(--pvec->nr)) | |
ff993ba1 | 2576 | return 0; |
4612aeef | 2577 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2578 | goto retry; |
2579 | return err; | |
06c04442 KO |
2580 | } |
2581 | ||
485bb99b | 2582 | /** |
87fa0f3e CH |
2583 | * filemap_read - Read data from the page cache. |
2584 | * @iocb: The iocb to read. | |
2585 | * @iter: Destination for the data. | |
2586 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2587 | * |
87fa0f3e CH |
2588 | * Copies data from the page cache. If the data is not currently present, |
2589 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2590 | * |
87fa0f3e CH |
2591 | * Return: Total number of bytes copied, including those already read by |
2592 | * the caller. If an error happens before any bytes are copied, returns | |
2593 | * a negative error number. | |
1da177e4 | 2594 | */ |
87fa0f3e CH |
2595 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2596 | ssize_t already_read) | |
1da177e4 | 2597 | { |
47c27bc4 | 2598 | struct file *filp = iocb->ki_filp; |
06c04442 | 2599 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2600 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2601 | struct inode *inode = mapping->host; |
ff993ba1 MWO |
2602 | struct pagevec pvec; |
2603 | int i, error = 0; | |
06c04442 KO |
2604 | bool writably_mapped; |
2605 | loff_t isize, end_offset; | |
1da177e4 | 2606 | |
723ef24b | 2607 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2608 | return 0; |
3644e2d2 KO |
2609 | if (unlikely(!iov_iter_count(iter))) |
2610 | return 0; | |
2611 | ||
c2a9737f | 2612 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
cbd59c48 | 2613 | pagevec_init(&pvec); |
c2a9737f | 2614 | |
06c04442 | 2615 | do { |
1da177e4 | 2616 | cond_resched(); |
5abf186a | 2617 | |
723ef24b | 2618 | /* |
06c04442 KO |
2619 | * If we've already successfully copied some data, then we |
2620 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2621 | * an async read NOWAIT at that point. | |
723ef24b | 2622 | */ |
87fa0f3e | 2623 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2624 | iocb->ki_flags |= IOCB_NOWAIT; |
2625 | ||
ff993ba1 MWO |
2626 | error = filemap_get_pages(iocb, iter, &pvec); |
2627 | if (error < 0) | |
06c04442 | 2628 | break; |
1da177e4 | 2629 | |
06c04442 KO |
2630 | /* |
2631 | * i_size must be checked after we know the pages are Uptodate. | |
2632 | * | |
2633 | * Checking i_size after the check allows us to calculate | |
2634 | * the correct value for "nr", which means the zero-filled | |
2635 | * part of the page is not copied back to userspace (unless | |
2636 | * another truncate extends the file - this is desired though). | |
2637 | */ | |
2638 | isize = i_size_read(inode); | |
2639 | if (unlikely(iocb->ki_pos >= isize)) | |
2640 | goto put_pages; | |
06c04442 KO |
2641 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2642 | ||
06c04442 KO |
2643 | /* |
2644 | * Once we start copying data, we don't want to be touching any | |
2645 | * cachelines that might be contended: | |
2646 | */ | |
2647 | writably_mapped = mapping_writably_mapped(mapping); | |
2648 | ||
2649 | /* | |
2650 | * When a sequential read accesses a page several times, only | |
2651 | * mark it as accessed the first time. | |
2652 | */ | |
2653 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2654 | ra->prev_pos >> PAGE_SHIFT) | |
ff993ba1 | 2655 | mark_page_accessed(pvec.pages[0]); |
06c04442 | 2656 | |
ff993ba1 | 2657 | for (i = 0; i < pagevec_count(&pvec); i++) { |
cbd59c48 MWO |
2658 | struct page *page = pvec.pages[i]; |
2659 | size_t page_size = thp_size(page); | |
2660 | size_t offset = iocb->ki_pos & (page_size - 1); | |
2661 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, | |
2662 | page_size - offset); | |
2663 | size_t copied; | |
06c04442 | 2664 | |
cbd59c48 MWO |
2665 | if (end_offset < page_offset(page)) |
2666 | break; | |
2667 | if (i > 0) | |
2668 | mark_page_accessed(page); | |
06c04442 KO |
2669 | /* |
2670 | * If users can be writing to this page using arbitrary | |
2671 | * virtual addresses, take care about potential aliasing | |
2672 | * before reading the page on the kernel side. | |
2673 | */ | |
cbd59c48 MWO |
2674 | if (writably_mapped) { |
2675 | int j; | |
2676 | ||
2677 | for (j = 0; j < thp_nr_pages(page); j++) | |
2678 | flush_dcache_page(page + j); | |
2679 | } | |
06c04442 | 2680 | |
cbd59c48 | 2681 | copied = copy_page_to_iter(page, offset, bytes, iter); |
06c04442 | 2682 | |
87fa0f3e | 2683 | already_read += copied; |
06c04442 KO |
2684 | iocb->ki_pos += copied; |
2685 | ra->prev_pos = iocb->ki_pos; | |
2686 | ||
2687 | if (copied < bytes) { | |
2688 | error = -EFAULT; | |
2689 | break; | |
2690 | } | |
1da177e4 | 2691 | } |
06c04442 | 2692 | put_pages: |
ff993ba1 MWO |
2693 | for (i = 0; i < pagevec_count(&pvec); i++) |
2694 | put_page(pvec.pages[i]); | |
cbd59c48 | 2695 | pagevec_reinit(&pvec); |
06c04442 | 2696 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2697 | |
0c6aa263 | 2698 | file_accessed(filp); |
06c04442 | 2699 | |
87fa0f3e | 2700 | return already_read ? already_read : error; |
1da177e4 | 2701 | } |
87fa0f3e | 2702 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2703 | |
485bb99b | 2704 | /** |
6abd2322 | 2705 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2706 | * @iocb: kernel I/O control block |
6abd2322 | 2707 | * @iter: destination for the data read |
485bb99b | 2708 | * |
6abd2322 | 2709 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2710 | * that can use the page cache directly. |
41da51bc AG |
2711 | * |
2712 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2713 | * be returned when no data can be read without waiting for I/O requests | |
2714 | * to complete; it doesn't prevent readahead. | |
2715 | * | |
2716 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2717 | * requests shall be made for the read or for readahead. When no data | |
2718 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2719 | * triggered, a partial, possibly empty read shall be returned. | |
2720 | * | |
a862f68a MR |
2721 | * Return: |
2722 | * * number of bytes copied, even for partial reads | |
41da51bc | 2723 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2724 | */ |
2725 | ssize_t | |
ed978a81 | 2726 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2727 | { |
e7080a43 | 2728 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2729 | ssize_t retval = 0; |
e7080a43 NS |
2730 | |
2731 | if (!count) | |
826ea860 | 2732 | return 0; /* skip atime */ |
1da177e4 | 2733 | |
2ba48ce5 | 2734 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2735 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2736 | struct address_space *mapping = file->f_mapping; |
2737 | struct inode *inode = mapping->host; | |
543ade1f | 2738 | loff_t size; |
1da177e4 | 2739 | |
1da177e4 | 2740 | size = i_size_read(inode); |
6be96d3a | 2741 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2742 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2743 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2744 | return -EAGAIN; |
2745 | } else { | |
2746 | retval = filemap_write_and_wait_range(mapping, | |
2747 | iocb->ki_pos, | |
2748 | iocb->ki_pos + count - 1); | |
2749 | if (retval < 0) | |
826ea860 | 2750 | return retval; |
6be96d3a | 2751 | } |
d8d3d94b | 2752 | |
0d5b0cf2 CH |
2753 | file_accessed(file); |
2754 | ||
5ecda137 | 2755 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2756 | if (retval >= 0) { |
c64fb5c7 | 2757 | iocb->ki_pos += retval; |
5ecda137 | 2758 | count -= retval; |
9fe55eea | 2759 | } |
ab2125df PB |
2760 | if (retval != -EIOCBQUEUED) |
2761 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2762 | |
9fe55eea SW |
2763 | /* |
2764 | * Btrfs can have a short DIO read if we encounter | |
2765 | * compressed extents, so if there was an error, or if | |
2766 | * we've already read everything we wanted to, or if | |
2767 | * there was a short read because we hit EOF, go ahead | |
2768 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2769 | * the rest of the read. Buffered reads will not work for |
2770 | * DAX files, so don't bother trying. | |
9fe55eea | 2771 | */ |
5ecda137 | 2772 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2773 | IS_DAX(inode)) |
826ea860 | 2774 | return retval; |
1da177e4 LT |
2775 | } |
2776 | ||
826ea860 | 2777 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2778 | } |
ed978a81 | 2779 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2780 | |
54fa39ac MWO |
2781 | static inline loff_t page_seek_hole_data(struct xa_state *xas, |
2782 | struct address_space *mapping, struct page *page, | |
2783 | loff_t start, loff_t end, bool seek_data) | |
41139aa4 | 2784 | { |
54fa39ac MWO |
2785 | const struct address_space_operations *ops = mapping->a_ops; |
2786 | size_t offset, bsz = i_blocksize(mapping->host); | |
2787 | ||
41139aa4 | 2788 | if (xa_is_value(page) || PageUptodate(page)) |
54fa39ac MWO |
2789 | return seek_data ? start : end; |
2790 | if (!ops->is_partially_uptodate) | |
2791 | return seek_data ? end : start; | |
2792 | ||
2793 | xas_pause(xas); | |
2794 | rcu_read_unlock(); | |
2795 | lock_page(page); | |
2796 | if (unlikely(page->mapping != mapping)) | |
2797 | goto unlock; | |
2798 | ||
2799 | offset = offset_in_thp(page, start) & ~(bsz - 1); | |
2800 | ||
2801 | do { | |
2802 | if (ops->is_partially_uptodate(page, offset, bsz) == seek_data) | |
2803 | break; | |
2804 | start = (start + bsz) & ~(bsz - 1); | |
2805 | offset += bsz; | |
2806 | } while (offset < thp_size(page)); | |
2807 | unlock: | |
2808 | unlock_page(page); | |
2809 | rcu_read_lock(); | |
2810 | return start; | |
41139aa4 MWO |
2811 | } |
2812 | ||
2813 | static inline | |
2814 | unsigned int seek_page_size(struct xa_state *xas, struct page *page) | |
2815 | { | |
2816 | if (xa_is_value(page)) | |
2817 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); | |
2818 | return thp_size(page); | |
2819 | } | |
2820 | ||
2821 | /** | |
2822 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2823 | * @mapping: Address space to search. | |
2824 | * @start: First byte to consider. | |
2825 | * @end: Limit of search (exclusive). | |
2826 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2827 | * | |
2828 | * If the page cache knows which blocks contain holes and which blocks | |
2829 | * contain data, your filesystem can use this function to implement | |
2830 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2831 | * entirely memory-based such as tmpfs, and filesystems which support | |
2832 | * unwritten extents. | |
2833 | * | |
f0953a1b | 2834 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2835 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2836 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2837 | * and @end contain data. | |
2838 | */ | |
2839 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2840 | loff_t end, int whence) | |
2841 | { | |
2842 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2843 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 MWO |
2844 | bool seek_data = (whence == SEEK_DATA); |
2845 | struct page *page; | |
2846 | ||
2847 | if (end <= start) | |
2848 | return -ENXIO; | |
2849 | ||
2850 | rcu_read_lock(); | |
2851 | while ((page = find_get_entry(&xas, max, XA_PRESENT))) { | |
ed98b015 HD |
2852 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
2853 | unsigned int seek_size; | |
41139aa4 MWO |
2854 | |
2855 | if (start < pos) { | |
2856 | if (!seek_data) | |
2857 | goto unlock; | |
2858 | start = pos; | |
2859 | } | |
2860 | ||
ed98b015 HD |
2861 | seek_size = seek_page_size(&xas, page); |
2862 | pos = round_up(pos + 1, seek_size); | |
54fa39ac MWO |
2863 | start = page_seek_hole_data(&xas, mapping, page, start, pos, |
2864 | seek_data); | |
2865 | if (start < pos) | |
41139aa4 | 2866 | goto unlock; |
ed98b015 HD |
2867 | if (start >= end) |
2868 | break; | |
2869 | if (seek_size > PAGE_SIZE) | |
2870 | xas_set(&xas, pos >> PAGE_SHIFT); | |
41139aa4 MWO |
2871 | if (!xa_is_value(page)) |
2872 | put_page(page); | |
2873 | } | |
41139aa4 | 2874 | if (seek_data) |
ed98b015 | 2875 | start = -ENXIO; |
41139aa4 MWO |
2876 | unlock: |
2877 | rcu_read_unlock(); | |
ed98b015 | 2878 | if (page && !xa_is_value(page)) |
41139aa4 | 2879 | put_page(page); |
41139aa4 MWO |
2880 | if (start > end) |
2881 | return end; | |
2882 | return start; | |
2883 | } | |
2884 | ||
1da177e4 | 2885 | #ifdef CONFIG_MMU |
1da177e4 | 2886 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2887 | /* |
c1e8d7c6 | 2888 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2889 | * @vmf - the vm_fault for this fault. |
2890 | * @page - the page to lock. | |
2891 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2892 | * | |
c1e8d7c6 | 2893 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2894 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2895 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2896 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2897 | */ | |
2898 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2899 | struct file **fpin) | |
2900 | { | |
7c23c782 MWO |
2901 | struct folio *folio = page_folio(page); |
2902 | ||
2903 | if (folio_trylock(folio)) | |
6b4c9f44 JB |
2904 | return 1; |
2905 | ||
8b0f9fa2 LT |
2906 | /* |
2907 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2908 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2909 | * is supposed to work. We have way too many special cases.. |
2910 | */ | |
6b4c9f44 JB |
2911 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2912 | return 0; | |
2913 | ||
2914 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2915 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
af7f29d9 | 2916 | if (__folio_lock_killable(folio)) { |
6b4c9f44 | 2917 | /* |
c1e8d7c6 | 2918 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2919 | * but all fault_handlers only check for fatal signals |
2920 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2921 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2922 | */ |
2923 | if (*fpin == NULL) | |
d8ed45c5 | 2924 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2925 | return 0; |
2926 | } | |
2927 | } else | |
7c23c782 MWO |
2928 | __folio_lock(folio); |
2929 | ||
6b4c9f44 JB |
2930 | return 1; |
2931 | } | |
2932 | ||
ef00e08e | 2933 | /* |
6b4c9f44 JB |
2934 | * Synchronous readahead happens when we don't even find a page in the page |
2935 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2936 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2937 | * that. If we didn't pin a file then we return NULL. The file that is | |
2938 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2939 | */ |
6b4c9f44 | 2940 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2941 | { |
2a1180f1 JB |
2942 | struct file *file = vmf->vma->vm_file; |
2943 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2944 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2945 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2946 | struct file *fpin = NULL; |
e630bfac | 2947 | unsigned int mmap_miss; |
ef00e08e LT |
2948 | |
2949 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2950 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2951 | return fpin; |
275b12bf | 2952 | if (!ra->ra_pages) |
6b4c9f44 | 2953 | return fpin; |
ef00e08e | 2954 | |
2a1180f1 | 2955 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2956 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 2957 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 2958 | return fpin; |
ef00e08e LT |
2959 | } |
2960 | ||
207d04ba | 2961 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2962 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2963 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2964 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2965 | |
2966 | /* | |
2967 | * Do we miss much more than hit in this file? If so, | |
2968 | * stop bothering with read-ahead. It will only hurt. | |
2969 | */ | |
e630bfac | 2970 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2971 | return fpin; |
ef00e08e | 2972 | |
d30a1100 WF |
2973 | /* |
2974 | * mmap read-around | |
2975 | */ | |
6b4c9f44 | 2976 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2977 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2978 | ra->size = ra->ra_pages; |
2979 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2980 | ractl._index = ra->start; |
2981 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 2982 | return fpin; |
ef00e08e LT |
2983 | } |
2984 | ||
2985 | /* | |
2986 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2987 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2988 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2989 | */ |
6b4c9f44 JB |
2990 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2991 | struct page *page) | |
ef00e08e | 2992 | { |
2a1180f1 JB |
2993 | struct file *file = vmf->vma->vm_file; |
2994 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2995 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2996 | struct file *fpin = NULL; |
e630bfac | 2997 | unsigned int mmap_miss; |
2a1180f1 | 2998 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2999 | |
3000 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 3001 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 3002 | return fpin; |
e630bfac KS |
3003 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3004 | if (mmap_miss) | |
3005 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
3006 | if (PageReadahead(page)) { |
3007 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
3008 | page_cache_async_readahead(mapping, ra, file, |
3009 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
3010 | } |
3011 | return fpin; | |
ef00e08e LT |
3012 | } |
3013 | ||
485bb99b | 3014 | /** |
54cb8821 | 3015 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 3016 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 3017 | * |
54cb8821 | 3018 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
3019 | * mapped memory region to read in file data during a page fault. |
3020 | * | |
3021 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
3022 | * it in the page cache, and handles the special cases reasonably without | |
3023 | * having a lot of duplicated code. | |
9a95f3cf | 3024 | * |
c1e8d7c6 | 3025 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 3026 | * |
c1e8d7c6 | 3027 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 3028 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 3029 | * |
c1e8d7c6 | 3030 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
3031 | * has not been released. |
3032 | * | |
3033 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
3034 | * |
3035 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 3036 | */ |
2bcd6454 | 3037 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
3038 | { |
3039 | int error; | |
11bac800 | 3040 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 3041 | struct file *fpin = NULL; |
1da177e4 | 3042 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3043 | struct inode *inode = mapping->host; |
ef00e08e | 3044 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 3045 | pgoff_t max_off; |
1da177e4 | 3046 | struct page *page; |
2bcd6454 | 3047 | vm_fault_t ret = 0; |
730633f0 | 3048 | bool mapping_locked = false; |
1da177e4 | 3049 | |
9ab2594f MW |
3050 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3051 | if (unlikely(offset >= max_off)) | |
5307cc1a | 3052 | return VM_FAULT_SIGBUS; |
1da177e4 | 3053 | |
1da177e4 | 3054 | /* |
49426420 | 3055 | * Do we have something in the page cache already? |
1da177e4 | 3056 | */ |
ef00e08e | 3057 | page = find_get_page(mapping, offset); |
730633f0 | 3058 | if (likely(page)) { |
1da177e4 | 3059 | /* |
730633f0 JK |
3060 | * We found the page, so try async readahead before waiting for |
3061 | * the lock. | |
1da177e4 | 3062 | */ |
730633f0 JK |
3063 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
3064 | fpin = do_async_mmap_readahead(vmf, page); | |
3065 | if (unlikely(!PageUptodate(page))) { | |
3066 | filemap_invalidate_lock_shared(mapping); | |
3067 | mapping_locked = true; | |
3068 | } | |
3069 | } else { | |
ef00e08e | 3070 | /* No page in the page cache at all */ |
ef00e08e | 3071 | count_vm_event(PGMAJFAULT); |
2262185c | 3072 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3073 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3074 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3075 | retry_find: |
730633f0 JK |
3076 | /* |
3077 | * See comment in filemap_create_page() why we need | |
3078 | * invalidate_lock | |
3079 | */ | |
3080 | if (!mapping_locked) { | |
3081 | filemap_invalidate_lock_shared(mapping); | |
3082 | mapping_locked = true; | |
3083 | } | |
a75d4c33 JB |
3084 | page = pagecache_get_page(mapping, offset, |
3085 | FGP_CREAT|FGP_FOR_MMAP, | |
3086 | vmf->gfp_mask); | |
6b4c9f44 JB |
3087 | if (!page) { |
3088 | if (fpin) | |
3089 | goto out_retry; | |
730633f0 | 3090 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3091 | return VM_FAULT_OOM; |
6b4c9f44 | 3092 | } |
1da177e4 LT |
3093 | } |
3094 | ||
6b4c9f44 JB |
3095 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
3096 | goto out_retry; | |
b522c94d ML |
3097 | |
3098 | /* Did it get truncated? */ | |
585e5a7b | 3099 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
3100 | unlock_page(page); |
3101 | put_page(page); | |
3102 | goto retry_find; | |
3103 | } | |
520e5ba4 | 3104 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 3105 | |
1da177e4 | 3106 | /* |
d00806b1 NP |
3107 | * We have a locked page in the page cache, now we need to check |
3108 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3109 | */ |
730633f0 JK |
3110 | if (unlikely(!PageUptodate(page))) { |
3111 | /* | |
3112 | * The page was in cache and uptodate and now it is not. | |
3113 | * Strange but possible since we didn't hold the page lock all | |
3114 | * the time. Let's drop everything get the invalidate lock and | |
3115 | * try again. | |
3116 | */ | |
3117 | if (!mapping_locked) { | |
3118 | unlock_page(page); | |
3119 | put_page(page); | |
3120 | goto retry_find; | |
3121 | } | |
1da177e4 | 3122 | goto page_not_uptodate; |
730633f0 | 3123 | } |
1da177e4 | 3124 | |
6b4c9f44 | 3125 | /* |
c1e8d7c6 | 3126 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3127 | * time to return to the upper layer and have it re-find the vma and |
3128 | * redo the fault. | |
3129 | */ | |
3130 | if (fpin) { | |
3131 | unlock_page(page); | |
3132 | goto out_retry; | |
3133 | } | |
730633f0 JK |
3134 | if (mapping_locked) |
3135 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3136 | |
ef00e08e LT |
3137 | /* |
3138 | * Found the page and have a reference on it. | |
3139 | * We must recheck i_size under page lock. | |
3140 | */ | |
9ab2594f MW |
3141 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3142 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 3143 | unlock_page(page); |
09cbfeaf | 3144 | put_page(page); |
5307cc1a | 3145 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3146 | } |
3147 | ||
d0217ac0 | 3148 | vmf->page = page; |
83c54070 | 3149 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3150 | |
1da177e4 | 3151 | page_not_uptodate: |
1da177e4 LT |
3152 | /* |
3153 | * Umm, take care of errors if the page isn't up-to-date. | |
3154 | * Try to re-read it _once_. We do this synchronously, | |
3155 | * because there really aren't any performance issues here | |
3156 | * and we need to check for errors. | |
3157 | */ | |
6b4c9f44 | 3158 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
d31fa86a | 3159 | error = filemap_read_page(file, mapping, page); |
6b4c9f44 JB |
3160 | if (fpin) |
3161 | goto out_retry; | |
09cbfeaf | 3162 | put_page(page); |
d00806b1 NP |
3163 | |
3164 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3165 | goto retry_find; |
730633f0 | 3166 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3167 | |
d0217ac0 | 3168 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3169 | |
3170 | out_retry: | |
3171 | /* | |
c1e8d7c6 | 3172 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3173 | * re-find the vma and come back and find our hopefully still populated |
3174 | * page. | |
3175 | */ | |
3176 | if (page) | |
3177 | put_page(page); | |
730633f0 JK |
3178 | if (mapping_locked) |
3179 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3180 | if (fpin) |
3181 | fput(fpin); | |
3182 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3183 | } |
3184 | EXPORT_SYMBOL(filemap_fault); | |
3185 | ||
f9ce0be7 | 3186 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3187 | { |
f9ce0be7 KS |
3188 | struct mm_struct *mm = vmf->vma->vm_mm; |
3189 | ||
3190 | /* Huge page is mapped? No need to proceed. */ | |
3191 | if (pmd_trans_huge(*vmf->pmd)) { | |
3192 | unlock_page(page); | |
3193 | put_page(page); | |
3194 | return true; | |
3195 | } | |
3196 | ||
3197 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
3198 | vm_fault_t ret = do_set_pmd(vmf, page); | |
3199 | if (!ret) { | |
3200 | /* The page is mapped successfully, reference consumed. */ | |
3201 | unlock_page(page); | |
3202 | return true; | |
3203 | } | |
3204 | } | |
3205 | ||
3206 | if (pmd_none(*vmf->pmd)) { | |
3207 | vmf->ptl = pmd_lock(mm, vmf->pmd); | |
3208 | if (likely(pmd_none(*vmf->pmd))) { | |
3209 | mm_inc_nr_ptes(mm); | |
3210 | pmd_populate(mm, vmf->pmd, vmf->prealloc_pte); | |
3211 | vmf->prealloc_pte = NULL; | |
3212 | } | |
3213 | spin_unlock(vmf->ptl); | |
3214 | } | |
3215 | ||
3216 | /* See comment in handle_pte_fault() */ | |
3217 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3218 | unlock_page(page); | |
3219 | put_page(page); | |
3220 | return true; | |
3221 | } | |
3222 | ||
3223 | return false; | |
3224 | } | |
3225 | ||
3226 | static struct page *next_uptodate_page(struct page *page, | |
3227 | struct address_space *mapping, | |
3228 | struct xa_state *xas, pgoff_t end_pgoff) | |
3229 | { | |
3230 | unsigned long max_idx; | |
3231 | ||
3232 | do { | |
3233 | if (!page) | |
3234 | return NULL; | |
3235 | if (xas_retry(xas, page)) | |
3236 | continue; | |
3237 | if (xa_is_value(page)) | |
3238 | continue; | |
3239 | if (PageLocked(page)) | |
3240 | continue; | |
3241 | if (!page_cache_get_speculative(page)) | |
3242 | continue; | |
3243 | /* Has the page moved or been split? */ | |
3244 | if (unlikely(page != xas_reload(xas))) | |
3245 | goto skip; | |
3246 | if (!PageUptodate(page) || PageReadahead(page)) | |
3247 | goto skip; | |
3248 | if (PageHWPoison(page)) | |
3249 | goto skip; | |
3250 | if (!trylock_page(page)) | |
3251 | goto skip; | |
3252 | if (page->mapping != mapping) | |
3253 | goto unlock; | |
3254 | if (!PageUptodate(page)) | |
3255 | goto unlock; | |
3256 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3257 | if (xas->xa_index >= max_idx) | |
3258 | goto unlock; | |
3259 | return page; | |
3260 | unlock: | |
3261 | unlock_page(page); | |
3262 | skip: | |
3263 | put_page(page); | |
3264 | } while ((page = xas_next_entry(xas, end_pgoff)) != NULL); | |
3265 | ||
3266 | return NULL; | |
3267 | } | |
3268 | ||
3269 | static inline struct page *first_map_page(struct address_space *mapping, | |
3270 | struct xa_state *xas, | |
3271 | pgoff_t end_pgoff) | |
3272 | { | |
3273 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3274 | mapping, xas, end_pgoff); | |
3275 | } | |
3276 | ||
3277 | static inline struct page *next_map_page(struct address_space *mapping, | |
3278 | struct xa_state *xas, | |
3279 | pgoff_t end_pgoff) | |
3280 | { | |
3281 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3282 | mapping, xas, end_pgoff); | |
3283 | } | |
3284 | ||
3285 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3286 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3287 | { | |
3288 | struct vm_area_struct *vma = vmf->vma; | |
3289 | struct file *file = vma->vm_file; | |
f1820361 | 3290 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3291 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3292 | unsigned long addr; |
070e807c | 3293 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
27a83a60 | 3294 | struct page *head, *page; |
e630bfac | 3295 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3296 | vm_fault_t ret = 0; |
f1820361 KS |
3297 | |
3298 | rcu_read_lock(); | |
f9ce0be7 KS |
3299 | head = first_map_page(mapping, &xas, end_pgoff); |
3300 | if (!head) | |
3301 | goto out; | |
f1820361 | 3302 | |
f9ce0be7 KS |
3303 | if (filemap_map_pmd(vmf, head)) { |
3304 | ret = VM_FAULT_NOPAGE; | |
3305 | goto out; | |
3306 | } | |
f1820361 | 3307 | |
9d3af4b4 WD |
3308 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3309 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3310 | do { |
27a83a60 | 3311 | page = find_subpage(head, xas.xa_index); |
f9ce0be7 | 3312 | if (PageHWPoison(page)) |
f1820361 KS |
3313 | goto unlock; |
3314 | ||
e630bfac KS |
3315 | if (mmap_miss > 0) |
3316 | mmap_miss--; | |
7267ec00 | 3317 | |
9d3af4b4 | 3318 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3319 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3320 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3321 | |
3322 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3323 | goto unlock; |
f9ce0be7 | 3324 | |
46bdb427 | 3325 | /* We're about to handle the fault */ |
9d3af4b4 | 3326 | if (vmf->address == addr) |
46bdb427 | 3327 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3328 | |
9d3af4b4 | 3329 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3330 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3331 | update_mmu_cache(vma, addr, vmf->pte); |
27a83a60 | 3332 | unlock_page(head); |
f9ce0be7 | 3333 | continue; |
f1820361 | 3334 | unlock: |
27a83a60 | 3335 | unlock_page(head); |
27a83a60 | 3336 | put_page(head); |
f9ce0be7 KS |
3337 | } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
3338 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3339 | out: | |
f1820361 | 3340 | rcu_read_unlock(); |
e630bfac | 3341 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3342 | return ret; |
f1820361 KS |
3343 | } |
3344 | EXPORT_SYMBOL(filemap_map_pages); | |
3345 | ||
2bcd6454 | 3346 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3347 | { |
5df1a672 | 3348 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
4fcf1c62 | 3349 | struct page *page = vmf->page; |
2bcd6454 | 3350 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3351 | |
5df1a672 | 3352 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3353 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 | 3354 | lock_page(page); |
5df1a672 | 3355 | if (page->mapping != mapping) { |
4fcf1c62 JK |
3356 | unlock_page(page); |
3357 | ret = VM_FAULT_NOPAGE; | |
3358 | goto out; | |
3359 | } | |
14da9200 JK |
3360 | /* |
3361 | * We mark the page dirty already here so that when freeze is in | |
3362 | * progress, we are guaranteed that writeback during freezing will | |
3363 | * see the dirty page and writeprotect it again. | |
3364 | */ | |
3365 | set_page_dirty(page); | |
1d1d1a76 | 3366 | wait_for_stable_page(page); |
4fcf1c62 | 3367 | out: |
5df1a672 | 3368 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3369 | return ret; |
3370 | } | |
4fcf1c62 | 3371 | |
f0f37e2f | 3372 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3373 | .fault = filemap_fault, |
f1820361 | 3374 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3375 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3376 | }; |
3377 | ||
3378 | /* This is used for a general mmap of a disk file */ | |
3379 | ||
68d68ff6 | 3380 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3381 | { |
3382 | struct address_space *mapping = file->f_mapping; | |
3383 | ||
3384 | if (!mapping->a_ops->readpage) | |
3385 | return -ENOEXEC; | |
3386 | file_accessed(file); | |
3387 | vma->vm_ops = &generic_file_vm_ops; | |
3388 | return 0; | |
3389 | } | |
1da177e4 LT |
3390 | |
3391 | /* | |
3392 | * This is for filesystems which do not implement ->writepage. | |
3393 | */ | |
3394 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3395 | { | |
3396 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3397 | return -EINVAL; | |
3398 | return generic_file_mmap(file, vma); | |
3399 | } | |
3400 | #else | |
4b96a37d | 3401 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3402 | { |
4b96a37d | 3403 | return VM_FAULT_SIGBUS; |
45397228 | 3404 | } |
68d68ff6 | 3405 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3406 | { |
3407 | return -ENOSYS; | |
3408 | } | |
68d68ff6 | 3409 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3410 | { |
3411 | return -ENOSYS; | |
3412 | } | |
3413 | #endif /* CONFIG_MMU */ | |
3414 | ||
45397228 | 3415 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3416 | EXPORT_SYMBOL(generic_file_mmap); |
3417 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3418 | ||
67f9fd91 SL |
3419 | static struct page *wait_on_page_read(struct page *page) |
3420 | { | |
3421 | if (!IS_ERR(page)) { | |
3422 | wait_on_page_locked(page); | |
3423 | if (!PageUptodate(page)) { | |
09cbfeaf | 3424 | put_page(page); |
67f9fd91 SL |
3425 | page = ERR_PTR(-EIO); |
3426 | } | |
3427 | } | |
3428 | return page; | |
3429 | } | |
3430 | ||
32b63529 | 3431 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 3432 | pgoff_t index, |
5e5358e7 | 3433 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3434 | void *data, |
3435 | gfp_t gfp) | |
1da177e4 | 3436 | { |
eb2be189 | 3437 | struct page *page; |
1da177e4 LT |
3438 | int err; |
3439 | repeat: | |
3440 | page = find_get_page(mapping, index); | |
3441 | if (!page) { | |
453f85d4 | 3442 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
3443 | if (!page) |
3444 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 3445 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 3446 | if (unlikely(err)) { |
09cbfeaf | 3447 | put_page(page); |
eb2be189 NP |
3448 | if (err == -EEXIST) |
3449 | goto repeat; | |
22ecdb4f | 3450 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3451 | return ERR_PTR(err); |
3452 | } | |
32b63529 MG |
3453 | |
3454 | filler: | |
6c45b454 CH |
3455 | if (filler) |
3456 | err = filler(data, page); | |
3457 | else | |
3458 | err = mapping->a_ops->readpage(data, page); | |
3459 | ||
1da177e4 | 3460 | if (err < 0) { |
09cbfeaf | 3461 | put_page(page); |
32b63529 | 3462 | return ERR_PTR(err); |
1da177e4 | 3463 | } |
1da177e4 | 3464 | |
32b63529 MG |
3465 | page = wait_on_page_read(page); |
3466 | if (IS_ERR(page)) | |
3467 | return page; | |
3468 | goto out; | |
3469 | } | |
1da177e4 LT |
3470 | if (PageUptodate(page)) |
3471 | goto out; | |
3472 | ||
ebded027 | 3473 | /* |
0e9aa675 | 3474 | * Page is not up to date and may be locked due to one of the following |
ebded027 MG |
3475 | * case a: Page is being filled and the page lock is held |
3476 | * case b: Read/write error clearing the page uptodate status | |
3477 | * case c: Truncation in progress (page locked) | |
3478 | * case d: Reclaim in progress | |
3479 | * | |
3480 | * Case a, the page will be up to date when the page is unlocked. | |
3481 | * There is no need to serialise on the page lock here as the page | |
3482 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 3483 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
3484 | * it's a race vs truncate race. |
3485 | * Case b, the page will not be up to date | |
3486 | * Case c, the page may be truncated but in itself, the data may still | |
3487 | * be valid after IO completes as it's a read vs truncate race. The | |
3488 | * operation must restart if the page is not uptodate on unlock but | |
3489 | * otherwise serialising on page lock to stabilise the mapping gives | |
3490 | * no additional guarantees to the caller as the page lock is | |
3491 | * released before return. | |
3492 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3493 | * will be a race with remove_mapping that determines if the mapping | |
3494 | * is valid on unlock but otherwise the data is valid and there is | |
3495 | * no need to serialise with page lock. | |
3496 | * | |
3497 | * As the page lock gives no additional guarantee, we optimistically | |
3498 | * wait on the page to be unlocked and check if it's up to date and | |
3499 | * use the page if it is. Otherwise, the page lock is required to | |
3500 | * distinguish between the different cases. The motivation is that we | |
3501 | * avoid spurious serialisations and wakeups when multiple processes | |
3502 | * wait on the same page for IO to complete. | |
3503 | */ | |
3504 | wait_on_page_locked(page); | |
3505 | if (PageUptodate(page)) | |
3506 | goto out; | |
3507 | ||
3508 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3509 | lock_page(page); |
ebded027 MG |
3510 | |
3511 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3512 | if (!page->mapping) { |
3513 | unlock_page(page); | |
09cbfeaf | 3514 | put_page(page); |
32b63529 | 3515 | goto repeat; |
1da177e4 | 3516 | } |
ebded027 MG |
3517 | |
3518 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3519 | if (PageUptodate(page)) { |
3520 | unlock_page(page); | |
3521 | goto out; | |
3522 | } | |
faffdfa0 XT |
3523 | |
3524 | /* | |
3525 | * A previous I/O error may have been due to temporary | |
3526 | * failures. | |
3527 | * Clear page error before actual read, PG_error will be | |
3528 | * set again if read page fails. | |
3529 | */ | |
3530 | ClearPageError(page); | |
32b63529 MG |
3531 | goto filler; |
3532 | ||
c855ff37 | 3533 | out: |
6fe6900e NP |
3534 | mark_page_accessed(page); |
3535 | return page; | |
3536 | } | |
0531b2aa LT |
3537 | |
3538 | /** | |
67f9fd91 | 3539 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3540 | * @mapping: the page's address_space |
3541 | * @index: the page index | |
3542 | * @filler: function to perform the read | |
5e5358e7 | 3543 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3544 | * |
0531b2aa | 3545 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3546 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3547 | * |
3548 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3549 | * |
730633f0 JK |
3550 | * The function expects mapping->invalidate_lock to be already held. |
3551 | * | |
a862f68a | 3552 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa | 3553 | */ |
67f9fd91 | 3554 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3555 | pgoff_t index, |
5e5358e7 | 3556 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3557 | void *data) |
3558 | { | |
d322a8e5 CH |
3559 | return do_read_cache_page(mapping, index, filler, data, |
3560 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3561 | } |
67f9fd91 | 3562 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3563 | |
3564 | /** | |
3565 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3566 | * @mapping: the page's address_space | |
3567 | * @index: the page index | |
3568 | * @gfp: the page allocator flags to use if allocating | |
3569 | * | |
3570 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3571 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3572 | * |
3573 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3574 | * |
730633f0 JK |
3575 | * The function expects mapping->invalidate_lock to be already held. |
3576 | * | |
a862f68a | 3577 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3578 | */ |
3579 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3580 | pgoff_t index, | |
3581 | gfp_t gfp) | |
3582 | { | |
6c45b454 | 3583 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3584 | } |
3585 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3586 | ||
afddba49 NP |
3587 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3588 | loff_t pos, unsigned len, unsigned flags, | |
3589 | struct page **pagep, void **fsdata) | |
3590 | { | |
3591 | const struct address_space_operations *aops = mapping->a_ops; | |
3592 | ||
4e02ed4b | 3593 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3594 | pagep, fsdata); |
afddba49 NP |
3595 | } |
3596 | EXPORT_SYMBOL(pagecache_write_begin); | |
3597 | ||
3598 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3599 | loff_t pos, unsigned len, unsigned copied, | |
3600 | struct page *page, void *fsdata) | |
3601 | { | |
3602 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3603 | |
4e02ed4b | 3604 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3605 | } |
3606 | EXPORT_SYMBOL(pagecache_write_end); | |
3607 | ||
a92853b6 KK |
3608 | /* |
3609 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3610 | */ | |
3611 | void dio_warn_stale_pagecache(struct file *filp) | |
3612 | { | |
3613 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3614 | char pathname[128]; | |
a92853b6 KK |
3615 | char *path; |
3616 | ||
5df1a672 | 3617 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3618 | if (__ratelimit(&_rs)) { |
3619 | path = file_path(filp, pathname, sizeof(pathname)); | |
3620 | if (IS_ERR(path)) | |
3621 | path = "(unknown)"; | |
3622 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3623 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3624 | current->comm); | |
3625 | } | |
3626 | } | |
3627 | ||
1da177e4 | 3628 | ssize_t |
1af5bb49 | 3629 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3630 | { |
3631 | struct file *file = iocb->ki_filp; | |
3632 | struct address_space *mapping = file->f_mapping; | |
3633 | struct inode *inode = mapping->host; | |
1af5bb49 | 3634 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3635 | ssize_t written; |
a969e903 CH |
3636 | size_t write_len; |
3637 | pgoff_t end; | |
1da177e4 | 3638 | |
0c949334 | 3639 | write_len = iov_iter_count(from); |
09cbfeaf | 3640 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3641 | |
6be96d3a GR |
3642 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3643 | /* If there are pages to writeback, return */ | |
5df1a672 | 3644 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3645 | pos + write_len - 1)) |
6be96d3a GR |
3646 | return -EAGAIN; |
3647 | } else { | |
3648 | written = filemap_write_and_wait_range(mapping, pos, | |
3649 | pos + write_len - 1); | |
3650 | if (written) | |
3651 | goto out; | |
3652 | } | |
a969e903 CH |
3653 | |
3654 | /* | |
3655 | * After a write we want buffered reads to be sure to go to disk to get | |
3656 | * the new data. We invalidate clean cached page from the region we're | |
3657 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3658 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3659 | */ |
55635ba7 | 3660 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3661 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3662 | /* |
3663 | * If a page can not be invalidated, return 0 to fall back | |
3664 | * to buffered write. | |
3665 | */ | |
3666 | if (written) { | |
3667 | if (written == -EBUSY) | |
3668 | return 0; | |
3669 | goto out; | |
a969e903 CH |
3670 | } |
3671 | ||
639a93a5 | 3672 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3673 | |
3674 | /* | |
3675 | * Finally, try again to invalidate clean pages which might have been | |
3676 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3677 | * if the source of the write was an mmap'ed region of the file | |
3678 | * we're writing. Either one is a pretty crazy thing to do, | |
3679 | * so we don't support it 100%. If this invalidation | |
3680 | * fails, tough, the write still worked... | |
332391a9 LC |
3681 | * |
3682 | * Most of the time we do not need this since dio_complete() will do | |
3683 | * the invalidation for us. However there are some file systems that | |
3684 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3685 | * them by removing it completely. |
3686 | * | |
9266a140 KK |
3687 | * Noticeable example is a blkdev_direct_IO(). |
3688 | * | |
80c1fe90 | 3689 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3690 | */ |
9266a140 KK |
3691 | if (written > 0 && mapping->nrpages && |
3692 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3693 | dio_warn_stale_pagecache(file); | |
a969e903 | 3694 | |
1da177e4 | 3695 | if (written > 0) { |
0116651c | 3696 | pos += written; |
639a93a5 | 3697 | write_len -= written; |
0116651c NK |
3698 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3699 | i_size_write(inode, pos); | |
1da177e4 LT |
3700 | mark_inode_dirty(inode); |
3701 | } | |
5cb6c6c7 | 3702 | iocb->ki_pos = pos; |
1da177e4 | 3703 | } |
ab2125df PB |
3704 | if (written != -EIOCBQUEUED) |
3705 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3706 | out: |
1da177e4 LT |
3707 | return written; |
3708 | } | |
3709 | EXPORT_SYMBOL(generic_file_direct_write); | |
3710 | ||
eb2be189 NP |
3711 | /* |
3712 | * Find or create a page at the given pagecache position. Return the locked | |
3713 | * page. This function is specifically for buffered writes. | |
3714 | */ | |
54566b2c NP |
3715 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3716 | pgoff_t index, unsigned flags) | |
eb2be189 | 3717 | { |
eb2be189 | 3718 | struct page *page; |
bbddabe2 | 3719 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3720 | |
54566b2c | 3721 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3722 | fgp_flags |= FGP_NOFS; |
3723 | ||
3724 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3725 | mapping_gfp_mask(mapping)); |
c585a267 | 3726 | if (page) |
2457aec6 | 3727 | wait_for_stable_page(page); |
eb2be189 | 3728 | |
eb2be189 NP |
3729 | return page; |
3730 | } | |
54566b2c | 3731 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3732 | |
3b93f911 | 3733 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3734 | struct iov_iter *i, loff_t pos) |
3735 | { | |
3736 | struct address_space *mapping = file->f_mapping; | |
3737 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3738 | long status = 0; | |
3739 | ssize_t written = 0; | |
674b892e NP |
3740 | unsigned int flags = 0; |
3741 | ||
afddba49 NP |
3742 | do { |
3743 | struct page *page; | |
afddba49 NP |
3744 | unsigned long offset; /* Offset into pagecache page */ |
3745 | unsigned long bytes; /* Bytes to write to page */ | |
3746 | size_t copied; /* Bytes copied from user */ | |
3747 | void *fsdata; | |
3748 | ||
09cbfeaf KS |
3749 | offset = (pos & (PAGE_SIZE - 1)); |
3750 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3751 | iov_iter_count(i)); |
3752 | ||
3753 | again: | |
00a3d660 LT |
3754 | /* |
3755 | * Bring in the user page that we will copy from _first_. | |
3756 | * Otherwise there's a nasty deadlock on copying from the | |
3757 | * same page as we're writing to, without it being marked | |
3758 | * up-to-date. | |
00a3d660 LT |
3759 | */ |
3760 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3761 | status = -EFAULT; | |
3762 | break; | |
3763 | } | |
3764 | ||
296291cd JK |
3765 | if (fatal_signal_pending(current)) { |
3766 | status = -EINTR; | |
3767 | break; | |
3768 | } | |
3769 | ||
674b892e | 3770 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3771 | &page, &fsdata); |
2457aec6 | 3772 | if (unlikely(status < 0)) |
afddba49 NP |
3773 | break; |
3774 | ||
931e80e4 | 3775 | if (mapping_writably_mapped(mapping)) |
3776 | flush_dcache_page(page); | |
00a3d660 | 3777 | |
f0b65f39 | 3778 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3779 | flush_dcache_page(page); |
3780 | ||
3781 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3782 | page, fsdata); | |
f0b65f39 AV |
3783 | if (unlikely(status != copied)) { |
3784 | iov_iter_revert(i, copied - max(status, 0L)); | |
3785 | if (unlikely(status < 0)) | |
3786 | break; | |
3787 | } | |
afddba49 NP |
3788 | cond_resched(); |
3789 | ||
bc1bb416 | 3790 | if (unlikely(status == 0)) { |
afddba49 | 3791 | /* |
bc1bb416 AV |
3792 | * A short copy made ->write_end() reject the |
3793 | * thing entirely. Might be memory poisoning | |
3794 | * halfway through, might be a race with munmap, | |
3795 | * might be severe memory pressure. | |
afddba49 | 3796 | */ |
bc1bb416 AV |
3797 | if (copied) |
3798 | bytes = copied; | |
afddba49 NP |
3799 | goto again; |
3800 | } | |
f0b65f39 AV |
3801 | pos += status; |
3802 | written += status; | |
afddba49 NP |
3803 | |
3804 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3805 | } while (iov_iter_count(i)); |
3806 | ||
3807 | return written ? written : status; | |
3808 | } | |
3b93f911 | 3809 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3810 | |
e4dd9de3 | 3811 | /** |
8174202b | 3812 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3813 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3814 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3815 | * |
3816 | * This function does all the work needed for actually writing data to a | |
3817 | * file. It does all basic checks, removes SUID from the file, updates | |
3818 | * modification times and calls proper subroutines depending on whether we | |
3819 | * do direct IO or a standard buffered write. | |
3820 | * | |
9608703e | 3821 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3822 | * object which does not need locking at all. |
3823 | * | |
3824 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3825 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3826 | * avoid syncing under i_rwsem. |
a862f68a MR |
3827 | * |
3828 | * Return: | |
3829 | * * number of bytes written, even for truncated writes | |
3830 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3831 | */ |
8174202b | 3832 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3833 | { |
3834 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3835 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3836 | struct inode *inode = mapping->host; |
3b93f911 | 3837 | ssize_t written = 0; |
1da177e4 | 3838 | ssize_t err; |
3b93f911 | 3839 | ssize_t status; |
1da177e4 | 3840 | |
1da177e4 | 3841 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3842 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3843 | err = file_remove_privs(file); |
1da177e4 LT |
3844 | if (err) |
3845 | goto out; | |
3846 | ||
c3b2da31 JB |
3847 | err = file_update_time(file); |
3848 | if (err) | |
3849 | goto out; | |
1da177e4 | 3850 | |
2ba48ce5 | 3851 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3852 | loff_t pos, endbyte; |
fb5527e6 | 3853 | |
1af5bb49 | 3854 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3855 | /* |
fbbbad4b MW |
3856 | * If the write stopped short of completing, fall back to |
3857 | * buffered writes. Some filesystems do this for writes to | |
3858 | * holes, for example. For DAX files, a buffered write will | |
3859 | * not succeed (even if it did, DAX does not handle dirty | |
3860 | * page-cache pages correctly). | |
1da177e4 | 3861 | */ |
0b8def9d | 3862 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3863 | goto out; |
3864 | ||
0b8def9d | 3865 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3866 | /* |
3b93f911 | 3867 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3868 | * then we want to return the number of bytes which were |
3869 | * direct-written, or the error code if that was zero. Note | |
3870 | * that this differs from normal direct-io semantics, which | |
3871 | * will return -EFOO even if some bytes were written. | |
3872 | */ | |
60bb4529 | 3873 | if (unlikely(status < 0)) { |
3b93f911 | 3874 | err = status; |
fb5527e6 JM |
3875 | goto out; |
3876 | } | |
fb5527e6 JM |
3877 | /* |
3878 | * We need to ensure that the page cache pages are written to | |
3879 | * disk and invalidated to preserve the expected O_DIRECT | |
3880 | * semantics. | |
3881 | */ | |
3b93f911 | 3882 | endbyte = pos + status - 1; |
0b8def9d | 3883 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3884 | if (err == 0) { |
0b8def9d | 3885 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3886 | written += status; |
fb5527e6 | 3887 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3888 | pos >> PAGE_SHIFT, |
3889 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3890 | } else { |
3891 | /* | |
3892 | * We don't know how much we wrote, so just return | |
3893 | * the number of bytes which were direct-written | |
3894 | */ | |
3895 | } | |
3896 | } else { | |
0b8def9d AV |
3897 | written = generic_perform_write(file, from, iocb->ki_pos); |
3898 | if (likely(written > 0)) | |
3899 | iocb->ki_pos += written; | |
fb5527e6 | 3900 | } |
1da177e4 LT |
3901 | out: |
3902 | current->backing_dev_info = NULL; | |
3903 | return written ? written : err; | |
3904 | } | |
8174202b | 3905 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3906 | |
e4dd9de3 | 3907 | /** |
8174202b | 3908 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3909 | * @iocb: IO state structure |
8174202b | 3910 | * @from: iov_iter with data to write |
e4dd9de3 | 3911 | * |
8174202b | 3912 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 3913 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 3914 | * and acquires i_rwsem as needed. |
a862f68a MR |
3915 | * Return: |
3916 | * * negative error code if no data has been written at all of | |
3917 | * vfs_fsync_range() failed for a synchronous write | |
3918 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3919 | */ |
8174202b | 3920 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3921 | { |
3922 | struct file *file = iocb->ki_filp; | |
148f948b | 3923 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3924 | ssize_t ret; |
1da177e4 | 3925 | |
5955102c | 3926 | inode_lock(inode); |
3309dd04 AV |
3927 | ret = generic_write_checks(iocb, from); |
3928 | if (ret > 0) | |
5f380c7f | 3929 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3930 | inode_unlock(inode); |
1da177e4 | 3931 | |
e2592217 CH |
3932 | if (ret > 0) |
3933 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3934 | return ret; |
3935 | } | |
8174202b | 3936 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3937 | |
cf9a2ae8 DH |
3938 | /** |
3939 | * try_to_release_page() - release old fs-specific metadata on a page | |
3940 | * | |
3941 | * @page: the page which the kernel is trying to free | |
3942 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3943 | * | |
3944 | * The address_space is to try to release any data against the page | |
a862f68a | 3945 | * (presumably at page->private). |
cf9a2ae8 | 3946 | * |
266cf658 DH |
3947 | * This may also be called if PG_fscache is set on a page, indicating that the |
3948 | * page is known to the local caching routines. | |
3949 | * | |
cf9a2ae8 | 3950 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3951 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3952 | * |
a862f68a | 3953 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3954 | */ |
3955 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3956 | { | |
3957 | struct address_space * const mapping = page->mapping; | |
3958 | ||
3959 | BUG_ON(!PageLocked(page)); | |
3960 | if (PageWriteback(page)) | |
3961 | return 0; | |
3962 | ||
3963 | if (mapping && mapping->a_ops->releasepage) | |
3964 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3965 | return try_to_free_buffers(page); | |
3966 | } | |
3967 | ||
3968 | EXPORT_SYMBOL(try_to_release_page); |