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> |
0f8053a5 NP |
44 | #include "internal.h" |
45 | ||
fe0bfaaf RJ |
46 | #define CREATE_TRACE_POINTS |
47 | #include <trace/events/filemap.h> | |
48 | ||
1da177e4 | 49 | /* |
1da177e4 LT |
50 | * FIXME: remove all knowledge of the buffer layer from the core VM |
51 | */ | |
148f948b | 52 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 53 | |
1da177e4 LT |
54 | #include <asm/mman.h> |
55 | ||
56 | /* | |
57 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
58 | * though. | |
59 | * | |
60 | * Shared mappings now work. 15.8.1995 Bruno. | |
61 | * | |
62 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
63 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
64 | * | |
65 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
66 | */ | |
67 | ||
68 | /* | |
69 | * Lock ordering: | |
70 | * | |
c8c06efa | 71 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 72 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 73 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 74 | * ->i_pages lock |
1da177e4 | 75 | * |
1b1dcc1b | 76 | * ->i_mutex |
c8c06efa | 77 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 | 78 | * |
c1e8d7c6 | 79 | * ->mmap_lock |
c8c06efa | 80 | * ->i_mmap_rwsem |
b8072f09 | 81 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 82 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 83 | * |
c1e8d7c6 | 84 | * ->mmap_lock |
1da177e4 LT |
85 | * ->lock_page (access_process_vm) |
86 | * | |
ccad2365 | 87 | * ->i_mutex (generic_perform_write) |
c1e8d7c6 | 88 | * ->mmap_lock (fault_in_pages_readable->do_page_fault) |
1da177e4 | 89 | * |
f758eeab | 90 | * bdi->wb.list_lock |
a66979ab | 91 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 92 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 93 | * |
c8c06efa | 94 | * ->i_mmap_rwsem |
1da177e4 LT |
95 | * ->anon_vma.lock (vma_adjust) |
96 | * | |
97 | * ->anon_vma.lock | |
b8072f09 | 98 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 99 | * |
b8072f09 | 100 | * ->page_table_lock or pte_lock |
5d337b91 | 101 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 102 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 103 | * ->i_pages lock (try_to_unmap_one) |
f4b7e272 AR |
104 | * ->pgdat->lru_lock (follow_page->mark_page_accessed) |
105 | * ->pgdat->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 106 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 107 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 108 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 109 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 110 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 111 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 112 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
113 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
114 | * | |
c8c06efa | 115 | * ->i_mmap_rwsem |
9a3c531d | 116 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
117 | */ |
118 | ||
5c024e6a | 119 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
120 | struct page *page, void *shadow) |
121 | { | |
5c024e6a MW |
122 | XA_STATE(xas, &mapping->i_pages, page->index); |
123 | unsigned int nr = 1; | |
c70b647d | 124 | |
5c024e6a | 125 | mapping_set_update(&xas, mapping); |
c70b647d | 126 | |
5c024e6a MW |
127 | /* hugetlb pages are represented by a single entry in the xarray */ |
128 | if (!PageHuge(page)) { | |
129 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 130 | nr = compound_nr(page); |
5c024e6a | 131 | } |
91b0abe3 | 132 | |
83929372 KS |
133 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
134 | VM_BUG_ON_PAGE(PageTail(page), page); | |
135 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 136 | |
5c024e6a MW |
137 | xas_store(&xas, shadow); |
138 | xas_init_marks(&xas); | |
d3798ae8 | 139 | |
2300638b JK |
140 | page->mapping = NULL; |
141 | /* Leave page->index set: truncation lookup relies upon it */ | |
142 | ||
d3798ae8 JW |
143 | if (shadow) { |
144 | mapping->nrexceptional += nr; | |
145 | /* | |
146 | * Make sure the nrexceptional update is committed before | |
147 | * the nrpages update so that final truncate racing | |
148 | * with reclaim does not see both counters 0 at the | |
149 | * same time and miss a shadow entry. | |
150 | */ | |
151 | smp_wmb(); | |
152 | } | |
153 | mapping->nrpages -= nr; | |
91b0abe3 JW |
154 | } |
155 | ||
5ecc4d85 JK |
156 | static void unaccount_page_cache_page(struct address_space *mapping, |
157 | struct page *page) | |
1da177e4 | 158 | { |
5ecc4d85 | 159 | int nr; |
1da177e4 | 160 | |
c515e1fd DM |
161 | /* |
162 | * if we're uptodate, flush out into the cleancache, otherwise | |
163 | * invalidate any existing cleancache entries. We can't leave | |
164 | * stale data around in the cleancache once our page is gone | |
165 | */ | |
166 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
167 | cleancache_put_page(page); | |
168 | else | |
3167760f | 169 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 170 | |
83929372 | 171 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
172 | VM_BUG_ON_PAGE(page_mapped(page), page); |
173 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
174 | int mapcount; | |
175 | ||
176 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
177 | current->comm, page_to_pfn(page)); | |
178 | dump_page(page, "still mapped when deleted"); | |
179 | dump_stack(); | |
180 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
181 | ||
182 | mapcount = page_mapcount(page); | |
183 | if (mapping_exiting(mapping) && | |
184 | page_count(page) >= mapcount + 2) { | |
185 | /* | |
186 | * All vmas have already been torn down, so it's | |
187 | * a good bet that actually the page is unmapped, | |
188 | * and we'd prefer not to leak it: if we're wrong, | |
189 | * some other bad page check should catch it later. | |
190 | */ | |
191 | page_mapcount_reset(page); | |
6d061f9f | 192 | page_ref_sub(page, mapcount); |
06b241f3 HD |
193 | } |
194 | } | |
195 | ||
4165b9b4 | 196 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
197 | if (PageHuge(page)) |
198 | return; | |
09612fa6 | 199 | |
5ecc4d85 JK |
200 | nr = hpage_nr_pages(page); |
201 | ||
0d1c2072 | 202 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 203 | if (PageSwapBacked(page)) { |
0d1c2072 | 204 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 JK |
205 | if (PageTransHuge(page)) |
206 | __dec_node_page_state(page, NR_SHMEM_THPS); | |
99cb0dbd SL |
207 | } else if (PageTransHuge(page)) { |
208 | __dec_node_page_state(page, NR_FILE_THPS); | |
09d91cda | 209 | filemap_nr_thps_dec(mapping); |
800d8c63 | 210 | } |
5ecc4d85 JK |
211 | |
212 | /* | |
213 | * At this point page must be either written or cleaned by | |
214 | * truncate. Dirty page here signals a bug and loss of | |
215 | * unwritten data. | |
216 | * | |
217 | * This fixes dirty accounting after removing the page entirely | |
218 | * but leaves PageDirty set: it has no effect for truncated | |
219 | * page and anyway will be cleared before returning page into | |
220 | * buddy allocator. | |
221 | */ | |
222 | if (WARN_ON_ONCE(PageDirty(page))) | |
223 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
224 | } | |
225 | ||
226 | /* | |
227 | * Delete a page from the page cache and free it. Caller has to make | |
228 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 229 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
230 | */ |
231 | void __delete_from_page_cache(struct page *page, void *shadow) | |
232 | { | |
233 | struct address_space *mapping = page->mapping; | |
234 | ||
235 | trace_mm_filemap_delete_from_page_cache(page); | |
236 | ||
237 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 238 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
239 | } |
240 | ||
59c66c5f JK |
241 | static void page_cache_free_page(struct address_space *mapping, |
242 | struct page *page) | |
243 | { | |
244 | void (*freepage)(struct page *); | |
245 | ||
246 | freepage = mapping->a_ops->freepage; | |
247 | if (freepage) | |
248 | freepage(page); | |
249 | ||
250 | if (PageTransHuge(page) && !PageHuge(page)) { | |
251 | page_ref_sub(page, HPAGE_PMD_NR); | |
252 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
253 | } else { | |
254 | put_page(page); | |
255 | } | |
256 | } | |
257 | ||
702cfbf9 MK |
258 | /** |
259 | * delete_from_page_cache - delete page from page cache | |
260 | * @page: the page which the kernel is trying to remove from page cache | |
261 | * | |
262 | * This must be called only on pages that have been verified to be in the page | |
263 | * cache and locked. It will never put the page into the free list, the caller | |
264 | * has a reference on the page. | |
265 | */ | |
266 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 267 | { |
83929372 | 268 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 269 | unsigned long flags; |
1da177e4 | 270 | |
cd7619d6 | 271 | BUG_ON(!PageLocked(page)); |
b93b0163 | 272 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 273 | __delete_from_page_cache(page, NULL); |
b93b0163 | 274 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 275 | |
59c66c5f | 276 | page_cache_free_page(mapping, page); |
97cecb5a MK |
277 | } |
278 | EXPORT_SYMBOL(delete_from_page_cache); | |
279 | ||
aa65c29c | 280 | /* |
ef8e5717 | 281 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
282 | * @mapping: the mapping to which pages belong |
283 | * @pvec: pagevec with pages to delete | |
284 | * | |
b93b0163 | 285 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
286 | * from the mapping. The function expects @pvec to be sorted by page index |
287 | * and is optimised for it to be dense. | |
b93b0163 | 288 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 289 | * modified). The function expects only THP head pages to be present in the |
4101196b | 290 | * @pvec. |
aa65c29c | 291 | * |
b93b0163 | 292 | * The function expects the i_pages lock to be held. |
aa65c29c | 293 | */ |
ef8e5717 | 294 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
295 | struct pagevec *pvec) |
296 | { | |
ef8e5717 | 297 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 298 | int total_pages = 0; |
4101196b | 299 | int i = 0; |
aa65c29c | 300 | struct page *page; |
aa65c29c | 301 | |
ef8e5717 MW |
302 | mapping_set_update(&xas, mapping); |
303 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 304 | if (i >= pagevec_count(pvec)) |
aa65c29c | 305 | break; |
4101196b MWO |
306 | |
307 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 308 | if (xa_is_value(page)) |
aa65c29c | 309 | continue; |
4101196b MWO |
310 | /* |
311 | * A page got inserted in our range? Skip it. We have our | |
312 | * pages locked so they are protected from being removed. | |
313 | * If we see a page whose index is higher than ours, it | |
314 | * means our page has been removed, which shouldn't be | |
315 | * possible because we're holding the PageLock. | |
316 | */ | |
317 | if (page != pvec->pages[i]) { | |
318 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
319 | page); | |
320 | continue; | |
321 | } | |
322 | ||
323 | WARN_ON_ONCE(!PageLocked(page)); | |
324 | ||
325 | if (page->index == xas.xa_index) | |
aa65c29c | 326 | page->mapping = NULL; |
4101196b MWO |
327 | /* Leave page->index set: truncation lookup relies on it */ |
328 | ||
329 | /* | |
330 | * Move to the next page in the vector if this is a regular | |
331 | * page or the index is of the last sub-page of this compound | |
332 | * page. | |
333 | */ | |
334 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 335 | i++; |
ef8e5717 | 336 | xas_store(&xas, NULL); |
aa65c29c JK |
337 | total_pages++; |
338 | } | |
339 | mapping->nrpages -= total_pages; | |
340 | } | |
341 | ||
342 | void delete_from_page_cache_batch(struct address_space *mapping, | |
343 | struct pagevec *pvec) | |
344 | { | |
345 | int i; | |
346 | unsigned long flags; | |
347 | ||
348 | if (!pagevec_count(pvec)) | |
349 | return; | |
350 | ||
b93b0163 | 351 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
352 | for (i = 0; i < pagevec_count(pvec); i++) { |
353 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
354 | ||
355 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
356 | } | |
ef8e5717 | 357 | page_cache_delete_batch(mapping, pvec); |
b93b0163 | 358 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
359 | |
360 | for (i = 0; i < pagevec_count(pvec); i++) | |
361 | page_cache_free_page(mapping, pvec->pages[i]); | |
362 | } | |
363 | ||
d72d9e2a | 364 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
365 | { |
366 | int ret = 0; | |
367 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
368 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
369 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 370 | ret = -ENOSPC; |
7fcbbaf1 JA |
371 | if (test_bit(AS_EIO, &mapping->flags) && |
372 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
373 | ret = -EIO; |
374 | return ret; | |
375 | } | |
d72d9e2a | 376 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 377 | |
76341cab JL |
378 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
379 | { | |
380 | /* Check for outstanding write errors */ | |
381 | if (test_bit(AS_EIO, &mapping->flags)) | |
382 | return -EIO; | |
383 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
384 | return -ENOSPC; | |
385 | return 0; | |
386 | } | |
387 | ||
1da177e4 | 388 | /** |
485bb99b | 389 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
390 | * @mapping: address space structure to write |
391 | * @start: offset in bytes where the range starts | |
469eb4d0 | 392 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 393 | * @sync_mode: enable synchronous operation |
1da177e4 | 394 | * |
485bb99b RD |
395 | * Start writeback against all of a mapping's dirty pages that lie |
396 | * within the byte offsets <start, end> inclusive. | |
397 | * | |
1da177e4 | 398 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 399 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
400 | * these two operations is that if a dirty page/buffer is encountered, it must |
401 | * be waited upon, and not just skipped over. | |
a862f68a MR |
402 | * |
403 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 404 | */ |
ebcf28e1 AM |
405 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
406 | loff_t end, int sync_mode) | |
1da177e4 LT |
407 | { |
408 | int ret; | |
409 | struct writeback_control wbc = { | |
410 | .sync_mode = sync_mode, | |
05fe478d | 411 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
412 | .range_start = start, |
413 | .range_end = end, | |
1da177e4 LT |
414 | }; |
415 | ||
c3aab9a0 KK |
416 | if (!mapping_cap_writeback_dirty(mapping) || |
417 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
1da177e4 LT |
418 | return 0; |
419 | ||
b16b1deb | 420 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 421 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 422 | wbc_detach_inode(&wbc); |
1da177e4 LT |
423 | return ret; |
424 | } | |
425 | ||
426 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
427 | int sync_mode) | |
428 | { | |
111ebb6e | 429 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
430 | } |
431 | ||
432 | int filemap_fdatawrite(struct address_space *mapping) | |
433 | { | |
434 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
435 | } | |
436 | EXPORT_SYMBOL(filemap_fdatawrite); | |
437 | ||
f4c0a0fd | 438 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 439 | loff_t end) |
1da177e4 LT |
440 | { |
441 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
442 | } | |
f4c0a0fd | 443 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 444 | |
485bb99b RD |
445 | /** |
446 | * filemap_flush - mostly a non-blocking flush | |
447 | * @mapping: target address_space | |
448 | * | |
1da177e4 LT |
449 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
450 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
451 | * |
452 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
453 | */ |
454 | int filemap_flush(struct address_space *mapping) | |
455 | { | |
456 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
457 | } | |
458 | EXPORT_SYMBOL(filemap_flush); | |
459 | ||
7fc9e472 GR |
460 | /** |
461 | * filemap_range_has_page - check if a page exists in range. | |
462 | * @mapping: address space within which to check | |
463 | * @start_byte: offset in bytes where the range starts | |
464 | * @end_byte: offset in bytes where the range ends (inclusive) | |
465 | * | |
466 | * Find at least one page in the range supplied, usually used to check if | |
467 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
468 | * |
469 | * Return: %true if at least one page exists in the specified range, | |
470 | * %false otherwise. | |
7fc9e472 GR |
471 | */ |
472 | bool filemap_range_has_page(struct address_space *mapping, | |
473 | loff_t start_byte, loff_t end_byte) | |
474 | { | |
f7b68046 | 475 | struct page *page; |
8fa8e538 MW |
476 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
477 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
478 | |
479 | if (end_byte < start_byte) | |
480 | return false; | |
481 | ||
8fa8e538 MW |
482 | rcu_read_lock(); |
483 | for (;;) { | |
484 | page = xas_find(&xas, max); | |
485 | if (xas_retry(&xas, page)) | |
486 | continue; | |
487 | /* Shadow entries don't count */ | |
488 | if (xa_is_value(page)) | |
489 | continue; | |
490 | /* | |
491 | * We don't need to try to pin this page; we're about to | |
492 | * release the RCU lock anyway. It is enough to know that | |
493 | * there was a page here recently. | |
494 | */ | |
495 | break; | |
496 | } | |
497 | rcu_read_unlock(); | |
7fc9e472 | 498 | |
8fa8e538 | 499 | return page != NULL; |
7fc9e472 GR |
500 | } |
501 | EXPORT_SYMBOL(filemap_range_has_page); | |
502 | ||
5e8fcc1a | 503 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 504 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 505 | { |
09cbfeaf KS |
506 | pgoff_t index = start_byte >> PAGE_SHIFT; |
507 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
508 | struct pagevec pvec; |
509 | int nr_pages; | |
1da177e4 | 510 | |
94004ed7 | 511 | if (end_byte < start_byte) |
5e8fcc1a | 512 | return; |
1da177e4 | 513 | |
86679820 | 514 | pagevec_init(&pvec); |
312e9d2f | 515 | while (index <= end) { |
1da177e4 LT |
516 | unsigned i; |
517 | ||
312e9d2f | 518 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 519 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
520 | if (!nr_pages) |
521 | break; | |
522 | ||
1da177e4 LT |
523 | for (i = 0; i < nr_pages; i++) { |
524 | struct page *page = pvec.pages[i]; | |
525 | ||
1da177e4 | 526 | wait_on_page_writeback(page); |
5e8fcc1a | 527 | ClearPageError(page); |
1da177e4 LT |
528 | } |
529 | pagevec_release(&pvec); | |
530 | cond_resched(); | |
531 | } | |
aa750fd7 JN |
532 | } |
533 | ||
534 | /** | |
535 | * filemap_fdatawait_range - wait for writeback to complete | |
536 | * @mapping: address space structure to wait for | |
537 | * @start_byte: offset in bytes where the range starts | |
538 | * @end_byte: offset in bytes where the range ends (inclusive) | |
539 | * | |
540 | * Walk the list of under-writeback pages of the given address space | |
541 | * in the given range and wait for all of them. Check error status of | |
542 | * the address space and return it. | |
543 | * | |
544 | * Since the error status of the address space is cleared by this function, | |
545 | * callers are responsible for checking the return value and handling and/or | |
546 | * reporting the error. | |
a862f68a MR |
547 | * |
548 | * Return: error status of the address space. | |
aa750fd7 JN |
549 | */ |
550 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
551 | loff_t end_byte) | |
552 | { | |
5e8fcc1a JL |
553 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
554 | return filemap_check_errors(mapping); | |
1da177e4 | 555 | } |
d3bccb6f JK |
556 | EXPORT_SYMBOL(filemap_fdatawait_range); |
557 | ||
aa0bfcd9 RZ |
558 | /** |
559 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
560 | * @mapping: address space structure to wait for | |
561 | * @start_byte: offset in bytes where the range starts | |
562 | * @end_byte: offset in bytes where the range ends (inclusive) | |
563 | * | |
564 | * Walk the list of under-writeback pages of the given address space in the | |
565 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
566 | * this function does not clear error status of the address space. | |
567 | * | |
568 | * Use this function if callers don't handle errors themselves. Expected | |
569 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
570 | * fsfreeze(8) | |
571 | */ | |
572 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
573 | loff_t start_byte, loff_t end_byte) | |
574 | { | |
575 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
576 | return filemap_check_and_keep_errors(mapping); | |
577 | } | |
578 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
579 | ||
a823e458 JL |
580 | /** |
581 | * file_fdatawait_range - wait for writeback to complete | |
582 | * @file: file pointing to address space structure to wait for | |
583 | * @start_byte: offset in bytes where the range starts | |
584 | * @end_byte: offset in bytes where the range ends (inclusive) | |
585 | * | |
586 | * Walk the list of under-writeback pages of the address space that file | |
587 | * refers to, in the given range and wait for all of them. Check error | |
588 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
589 | * | |
590 | * Since the error status of the file is advanced by this function, | |
591 | * callers are responsible for checking the return value and handling and/or | |
592 | * reporting the error. | |
a862f68a MR |
593 | * |
594 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
595 | */ |
596 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
597 | { | |
598 | struct address_space *mapping = file->f_mapping; | |
599 | ||
600 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
601 | return file_check_and_advance_wb_err(file); | |
602 | } | |
603 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 604 | |
aa750fd7 JN |
605 | /** |
606 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
607 | * @mapping: address space structure to wait for | |
608 | * | |
609 | * Walk the list of under-writeback pages of the given address space | |
610 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
611 | * does not clear error status of the address space. | |
612 | * | |
613 | * Use this function if callers don't handle errors themselves. Expected | |
614 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
615 | * fsfreeze(8) | |
a862f68a MR |
616 | * |
617 | * Return: error status of the address space. | |
aa750fd7 | 618 | */ |
76341cab | 619 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 620 | { |
ffb959bb | 621 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 622 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 623 | } |
76341cab | 624 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 625 | |
875d91b1 | 626 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 627 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 628 | { |
875d91b1 KK |
629 | if (dax_mapping(mapping)) |
630 | return mapping->nrexceptional; | |
631 | ||
632 | return mapping->nrpages; | |
1da177e4 | 633 | } |
1da177e4 | 634 | |
485bb99b RD |
635 | /** |
636 | * filemap_write_and_wait_range - write out & wait on a file range | |
637 | * @mapping: the address_space for the pages | |
638 | * @lstart: offset in bytes where the range starts | |
639 | * @lend: offset in bytes where the range ends (inclusive) | |
640 | * | |
469eb4d0 AM |
641 | * Write out and wait upon file offsets lstart->lend, inclusive. |
642 | * | |
0e056eb5 | 643 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 644 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
645 | * |
646 | * Return: error status of the address space. | |
469eb4d0 | 647 | */ |
1da177e4 LT |
648 | int filemap_write_and_wait_range(struct address_space *mapping, |
649 | loff_t lstart, loff_t lend) | |
650 | { | |
28fd1298 | 651 | int err = 0; |
1da177e4 | 652 | |
9326c9b2 | 653 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
654 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
655 | WB_SYNC_ALL); | |
ddf8f376 IW |
656 | /* |
657 | * Even if the above returned error, the pages may be | |
658 | * written partially (e.g. -ENOSPC), so we wait for it. | |
659 | * But the -EIO is special case, it may indicate the worst | |
660 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
661 | */ | |
28fd1298 | 662 | if (err != -EIO) { |
94004ed7 CH |
663 | int err2 = filemap_fdatawait_range(mapping, |
664 | lstart, lend); | |
28fd1298 OH |
665 | if (!err) |
666 | err = err2; | |
cbeaf951 JL |
667 | } else { |
668 | /* Clear any previously stored errors */ | |
669 | filemap_check_errors(mapping); | |
28fd1298 | 670 | } |
865ffef3 DM |
671 | } else { |
672 | err = filemap_check_errors(mapping); | |
1da177e4 | 673 | } |
28fd1298 | 674 | return err; |
1da177e4 | 675 | } |
f6995585 | 676 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 677 | |
5660e13d JL |
678 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
679 | { | |
3acdfd28 | 680 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
681 | |
682 | trace_filemap_set_wb_err(mapping, eseq); | |
683 | } | |
684 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
685 | ||
686 | /** | |
687 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
688 | * and advance wb_err to current one | |
689 | * @file: struct file on which the error is being reported | |
690 | * | |
691 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
692 | * want to report any writeback errors that occurred since the last fsync (or | |
693 | * since the file was opened if there haven't been any). | |
694 | * | |
695 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
696 | * then just quickly return 0. The file is all caught up. | |
697 | * | |
698 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
699 | * it and try to swap it into place. If it works, or another task beat us | |
700 | * to it with the new value, then update the f_wb_err and return the error | |
701 | * portion. The error at this point must be reported via proper channels | |
702 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
703 | * | |
704 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
705 | * value is protected by the f_lock since we must ensure that it reflects | |
706 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
707 | * |
708 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
709 | */ |
710 | int file_check_and_advance_wb_err(struct file *file) | |
711 | { | |
712 | int err = 0; | |
713 | errseq_t old = READ_ONCE(file->f_wb_err); | |
714 | struct address_space *mapping = file->f_mapping; | |
715 | ||
716 | /* Locklessly handle the common case where nothing has changed */ | |
717 | if (errseq_check(&mapping->wb_err, old)) { | |
718 | /* Something changed, must use slow path */ | |
719 | spin_lock(&file->f_lock); | |
720 | old = file->f_wb_err; | |
721 | err = errseq_check_and_advance(&mapping->wb_err, | |
722 | &file->f_wb_err); | |
723 | trace_file_check_and_advance_wb_err(file, old); | |
724 | spin_unlock(&file->f_lock); | |
725 | } | |
f4e222c5 JL |
726 | |
727 | /* | |
728 | * We're mostly using this function as a drop in replacement for | |
729 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
730 | * that the legacy code would have had on these flags. | |
731 | */ | |
732 | clear_bit(AS_EIO, &mapping->flags); | |
733 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
734 | return err; |
735 | } | |
736 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
737 | ||
738 | /** | |
739 | * file_write_and_wait_range - write out & wait on a file range | |
740 | * @file: file pointing to address_space with pages | |
741 | * @lstart: offset in bytes where the range starts | |
742 | * @lend: offset in bytes where the range ends (inclusive) | |
743 | * | |
744 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
745 | * | |
746 | * Note that @lend is inclusive (describes the last byte to be written) so | |
747 | * that this function can be used to write to the very end-of-file (end = -1). | |
748 | * | |
749 | * After writing out and waiting on the data, we check and advance the | |
750 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
751 | * |
752 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
753 | */ |
754 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
755 | { | |
756 | int err = 0, err2; | |
757 | struct address_space *mapping = file->f_mapping; | |
758 | ||
9326c9b2 | 759 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
760 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
761 | WB_SYNC_ALL); | |
762 | /* See comment of filemap_write_and_wait() */ | |
763 | if (err != -EIO) | |
764 | __filemap_fdatawait_range(mapping, lstart, lend); | |
765 | } | |
766 | err2 = file_check_and_advance_wb_err(file); | |
767 | if (!err) | |
768 | err = err2; | |
769 | return err; | |
770 | } | |
771 | EXPORT_SYMBOL(file_write_and_wait_range); | |
772 | ||
ef6a3c63 MS |
773 | /** |
774 | * replace_page_cache_page - replace a pagecache page with a new one | |
775 | * @old: page to be replaced | |
776 | * @new: page to replace with | |
777 | * @gfp_mask: allocation mode | |
778 | * | |
779 | * This function replaces a page in the pagecache with a new one. On | |
780 | * success it acquires the pagecache reference for the new page and | |
781 | * drops it for the old page. Both the old and new pages must be | |
782 | * locked. This function does not add the new page to the LRU, the | |
783 | * caller must do that. | |
784 | * | |
74d60958 | 785 | * The remove + add is atomic. This function cannot fail. |
a862f68a MR |
786 | * |
787 | * Return: %0 | |
ef6a3c63 MS |
788 | */ |
789 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
790 | { | |
74d60958 MW |
791 | struct address_space *mapping = old->mapping; |
792 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
793 | pgoff_t offset = old->index; | |
794 | XA_STATE(xas, &mapping->i_pages, offset); | |
795 | unsigned long flags; | |
ef6a3c63 | 796 | |
309381fe SL |
797 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
798 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
799 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 800 | |
74d60958 MW |
801 | get_page(new); |
802 | new->mapping = mapping; | |
803 | new->index = offset; | |
ef6a3c63 | 804 | |
0d1c2072 JW |
805 | mem_cgroup_migrate(old, new); |
806 | ||
74d60958 MW |
807 | xas_lock_irqsave(&xas, flags); |
808 | xas_store(&xas, new); | |
4165b9b4 | 809 | |
74d60958 MW |
810 | old->mapping = NULL; |
811 | /* hugetlb pages do not participate in page cache accounting. */ | |
812 | if (!PageHuge(old)) | |
0d1c2072 | 813 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 814 | if (!PageHuge(new)) |
0d1c2072 | 815 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 816 | if (PageSwapBacked(old)) |
0d1c2072 | 817 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 818 | if (PageSwapBacked(new)) |
0d1c2072 | 819 | __inc_lruvec_page_state(new, NR_SHMEM); |
74d60958 | 820 | xas_unlock_irqrestore(&xas, flags); |
74d60958 MW |
821 | if (freepage) |
822 | freepage(old); | |
823 | put_page(old); | |
ef6a3c63 | 824 | |
74d60958 | 825 | return 0; |
ef6a3c63 MS |
826 | } |
827 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
828 | ||
a528910e JW |
829 | static int __add_to_page_cache_locked(struct page *page, |
830 | struct address_space *mapping, | |
831 | pgoff_t offset, gfp_t gfp_mask, | |
832 | void **shadowp) | |
1da177e4 | 833 | { |
74d60958 | 834 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 | 835 | int huge = PageHuge(page); |
e286781d | 836 | int error; |
74d60958 | 837 | void *old; |
e286781d | 838 | |
309381fe SL |
839 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
840 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 841 | mapping_set_update(&xas, mapping); |
e286781d | 842 | |
09cbfeaf | 843 | get_page(page); |
66a0c8ee KS |
844 | page->mapping = mapping; |
845 | page->index = offset; | |
846 | ||
3fea5a49 | 847 | if (!huge) { |
d9eb1ea2 | 848 | error = mem_cgroup_charge(page, current->mm, gfp_mask); |
3fea5a49 JW |
849 | if (error) |
850 | goto error; | |
851 | } | |
852 | ||
74d60958 MW |
853 | do { |
854 | xas_lock_irq(&xas); | |
855 | old = xas_load(&xas); | |
856 | if (old && !xa_is_value(old)) | |
857 | xas_set_err(&xas, -EEXIST); | |
858 | xas_store(&xas, page); | |
859 | if (xas_error(&xas)) | |
860 | goto unlock; | |
861 | ||
862 | if (xa_is_value(old)) { | |
863 | mapping->nrexceptional--; | |
864 | if (shadowp) | |
865 | *shadowp = old; | |
866 | } | |
867 | mapping->nrpages++; | |
868 | ||
869 | /* hugetlb pages do not participate in page cache accounting */ | |
870 | if (!huge) | |
0d1c2072 | 871 | __inc_lruvec_page_state(page, NR_FILE_PAGES); |
74d60958 MW |
872 | unlock: |
873 | xas_unlock_irq(&xas); | |
874 | } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK)); | |
875 | ||
3fea5a49 JW |
876 | if (xas_error(&xas)) { |
877 | error = xas_error(&xas); | |
74d60958 | 878 | goto error; |
3fea5a49 | 879 | } |
4165b9b4 | 880 | |
66a0c8ee KS |
881 | trace_mm_filemap_add_to_page_cache(page); |
882 | return 0; | |
74d60958 | 883 | error: |
66a0c8ee KS |
884 | page->mapping = NULL; |
885 | /* Leave page->index set: truncation relies upon it */ | |
09cbfeaf | 886 | put_page(page); |
3fea5a49 | 887 | return error; |
1da177e4 | 888 | } |
cfcbfb13 | 889 | ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO); |
a528910e JW |
890 | |
891 | /** | |
892 | * add_to_page_cache_locked - add a locked page to the pagecache | |
893 | * @page: page to add | |
894 | * @mapping: the page's address_space | |
895 | * @offset: page index | |
896 | * @gfp_mask: page allocation mode | |
897 | * | |
898 | * This function is used to add a page to the pagecache. It must be locked. | |
899 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
900 | * |
901 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
902 | */ |
903 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
904 | pgoff_t offset, gfp_t gfp_mask) | |
905 | { | |
906 | return __add_to_page_cache_locked(page, mapping, offset, | |
907 | gfp_mask, NULL); | |
908 | } | |
e286781d | 909 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
910 | |
911 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 912 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 913 | { |
a528910e | 914 | void *shadow = NULL; |
4f98a2fe RR |
915 | int ret; |
916 | ||
48c935ad | 917 | __SetPageLocked(page); |
a528910e JW |
918 | ret = __add_to_page_cache_locked(page, mapping, offset, |
919 | gfp_mask, &shadow); | |
920 | if (unlikely(ret)) | |
48c935ad | 921 | __ClearPageLocked(page); |
a528910e JW |
922 | else { |
923 | /* | |
924 | * The page might have been evicted from cache only | |
925 | * recently, in which case it should be activated like | |
926 | * any other repeatedly accessed page. | |
f0281a00 RR |
927 | * The exception is pages getting rewritten; evicting other |
928 | * data from the working set, only to cache data that will | |
929 | * get overwritten with something else, is a waste of memory. | |
a528910e | 930 | */ |
1899ad18 JW |
931 | WARN_ON_ONCE(PageActive(page)); |
932 | if (!(gfp_mask & __GFP_WRITE) && shadow) | |
933 | workingset_refault(page, shadow); | |
a528910e JW |
934 | lru_cache_add(page); |
935 | } | |
1da177e4 LT |
936 | return ret; |
937 | } | |
18bc0bbd | 938 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 939 | |
44110fe3 | 940 | #ifdef CONFIG_NUMA |
2ae88149 | 941 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 942 | { |
c0ff7453 MX |
943 | int n; |
944 | struct page *page; | |
945 | ||
44110fe3 | 946 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
947 | unsigned int cpuset_mems_cookie; |
948 | do { | |
d26914d1 | 949 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 950 | n = cpuset_mem_spread_node(); |
96db800f | 951 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 952 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 953 | |
c0ff7453 | 954 | return page; |
44110fe3 | 955 | } |
2ae88149 | 956 | return alloc_pages(gfp, 0); |
44110fe3 | 957 | } |
2ae88149 | 958 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
959 | #endif |
960 | ||
1da177e4 LT |
961 | /* |
962 | * In order to wait for pages to become available there must be | |
963 | * waitqueues associated with pages. By using a hash table of | |
964 | * waitqueues where the bucket discipline is to maintain all | |
965 | * waiters on the same queue and wake all when any of the pages | |
966 | * become available, and for the woken contexts to check to be | |
967 | * sure the appropriate page became available, this saves space | |
968 | * at a cost of "thundering herd" phenomena during rare hash | |
969 | * collisions. | |
970 | */ | |
62906027 NP |
971 | #define PAGE_WAIT_TABLE_BITS 8 |
972 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
973 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
974 | ||
975 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 976 | { |
62906027 | 977 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 978 | } |
1da177e4 | 979 | |
62906027 | 980 | void __init pagecache_init(void) |
1da177e4 | 981 | { |
62906027 | 982 | int i; |
1da177e4 | 983 | |
62906027 NP |
984 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
985 | init_waitqueue_head(&page_wait_table[i]); | |
986 | ||
987 | page_writeback_init(); | |
1da177e4 | 988 | } |
1da177e4 | 989 | |
ac6424b9 | 990 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 991 | { |
62906027 NP |
992 | struct wait_page_key *key = arg; |
993 | struct wait_page_queue *wait_page | |
994 | = container_of(wait, struct wait_page_queue, wait); | |
c7510ab2 | 995 | int ret; |
62906027 | 996 | |
c7510ab2 JA |
997 | ret = wake_page_match(wait_page, key); |
998 | if (ret != 1) | |
999 | return ret; | |
62906027 | 1000 | return autoremove_wake_function(wait, mode, sync, key); |
f62e00cc KM |
1001 | } |
1002 | ||
74d81bfa | 1003 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1004 | { |
62906027 NP |
1005 | wait_queue_head_t *q = page_waitqueue(page); |
1006 | struct wait_page_key key; | |
1007 | unsigned long flags; | |
11a19c7b | 1008 | wait_queue_entry_t bookmark; |
cbbce822 | 1009 | |
62906027 NP |
1010 | key.page = page; |
1011 | key.bit_nr = bit_nr; | |
1012 | key.page_match = 0; | |
1013 | ||
11a19c7b TC |
1014 | bookmark.flags = 0; |
1015 | bookmark.private = NULL; | |
1016 | bookmark.func = NULL; | |
1017 | INIT_LIST_HEAD(&bookmark.entry); | |
1018 | ||
62906027 | 1019 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1020 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1021 | ||
1022 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1023 | /* | |
1024 | * Take a breather from holding the lock, | |
1025 | * allow pages that finish wake up asynchronously | |
1026 | * to acquire the lock and remove themselves | |
1027 | * from wait queue | |
1028 | */ | |
1029 | spin_unlock_irqrestore(&q->lock, flags); | |
1030 | cpu_relax(); | |
1031 | spin_lock_irqsave(&q->lock, flags); | |
1032 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1033 | } | |
1034 | ||
62906027 NP |
1035 | /* |
1036 | * It is possible for other pages to have collided on the waitqueue | |
1037 | * hash, so in that case check for a page match. That prevents a long- | |
1038 | * term waiter | |
1039 | * | |
1040 | * It is still possible to miss a case here, when we woke page waiters | |
1041 | * and removed them from the waitqueue, but there are still other | |
1042 | * page waiters. | |
1043 | */ | |
1044 | if (!waitqueue_active(q) || !key.page_match) { | |
1045 | ClearPageWaiters(page); | |
1046 | /* | |
1047 | * It's possible to miss clearing Waiters here, when we woke | |
1048 | * our page waiters, but the hashed waitqueue has waiters for | |
1049 | * other pages on it. | |
1050 | * | |
1051 | * That's okay, it's a rare case. The next waker will clear it. | |
1052 | */ | |
1053 | } | |
1054 | spin_unlock_irqrestore(&q->lock, flags); | |
1055 | } | |
74d81bfa NP |
1056 | |
1057 | static void wake_up_page(struct page *page, int bit) | |
1058 | { | |
1059 | if (!PageWaiters(page)) | |
1060 | return; | |
1061 | wake_up_page_bit(page, bit); | |
1062 | } | |
62906027 | 1063 | |
9a1ea439 HD |
1064 | /* |
1065 | * A choice of three behaviors for wait_on_page_bit_common(): | |
1066 | */ | |
1067 | enum behavior { | |
1068 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
1069 | * __lock_page() waiting on then setting PG_locked. | |
1070 | */ | |
1071 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1072 | * wait_on_page_writeback() waiting on PG_writeback. | |
1073 | */ | |
1074 | DROP, /* Drop ref to page before wait, no check when woken, | |
1075 | * like put_and_wait_on_page_locked() on PG_locked. | |
1076 | */ | |
1077 | }; | |
1078 | ||
62906027 | 1079 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
9a1ea439 | 1080 | struct page *page, int bit_nr, int state, enum behavior behavior) |
62906027 NP |
1081 | { |
1082 | struct wait_page_queue wait_page; | |
ac6424b9 | 1083 | wait_queue_entry_t *wait = &wait_page.wait; |
9a1ea439 | 1084 | bool bit_is_set; |
b1d29ba8 | 1085 | bool thrashing = false; |
9a1ea439 | 1086 | bool delayacct = false; |
eb414681 | 1087 | unsigned long pflags; |
62906027 NP |
1088 | int ret = 0; |
1089 | ||
eb414681 | 1090 | if (bit_nr == PG_locked && |
b1d29ba8 | 1091 | !PageUptodate(page) && PageWorkingset(page)) { |
9a1ea439 | 1092 | if (!PageSwapBacked(page)) { |
eb414681 | 1093 | delayacct_thrashing_start(); |
9a1ea439 HD |
1094 | delayacct = true; |
1095 | } | |
eb414681 | 1096 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1097 | thrashing = true; |
1098 | } | |
1099 | ||
62906027 | 1100 | init_wait(wait); |
9a1ea439 | 1101 | wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0; |
62906027 NP |
1102 | wait->func = wake_page_function; |
1103 | wait_page.page = page; | |
1104 | wait_page.bit_nr = bit_nr; | |
1105 | ||
1106 | for (;;) { | |
1107 | spin_lock_irq(&q->lock); | |
1108 | ||
2055da97 | 1109 | if (likely(list_empty(&wait->entry))) { |
3510ca20 | 1110 | __add_wait_queue_entry_tail(q, wait); |
62906027 NP |
1111 | SetPageWaiters(page); |
1112 | } | |
1113 | ||
1114 | set_current_state(state); | |
1115 | ||
1116 | spin_unlock_irq(&q->lock); | |
1117 | ||
9a1ea439 HD |
1118 | bit_is_set = test_bit(bit_nr, &page->flags); |
1119 | if (behavior == DROP) | |
1120 | put_page(page); | |
1121 | ||
1122 | if (likely(bit_is_set)) | |
62906027 | 1123 | io_schedule(); |
62906027 | 1124 | |
9a1ea439 | 1125 | if (behavior == EXCLUSIVE) { |
62906027 NP |
1126 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) |
1127 | break; | |
9a1ea439 | 1128 | } else if (behavior == SHARED) { |
62906027 NP |
1129 | if (!test_bit(bit_nr, &page->flags)) |
1130 | break; | |
1131 | } | |
a8b169af | 1132 | |
fa45f116 | 1133 | if (signal_pending_state(state, current)) { |
a8b169af LT |
1134 | ret = -EINTR; |
1135 | break; | |
1136 | } | |
9a1ea439 HD |
1137 | |
1138 | if (behavior == DROP) { | |
1139 | /* | |
1140 | * We can no longer safely access page->flags: | |
1141 | * even if CONFIG_MEMORY_HOTREMOVE is not enabled, | |
1142 | * there is a risk of waiting forever on a page reused | |
1143 | * for something that keeps it locked indefinitely. | |
1144 | * But best check for -EINTR above before breaking. | |
1145 | */ | |
1146 | break; | |
1147 | } | |
62906027 NP |
1148 | } |
1149 | ||
1150 | finish_wait(q, wait); | |
1151 | ||
eb414681 | 1152 | if (thrashing) { |
9a1ea439 | 1153 | if (delayacct) |
eb414681 JW |
1154 | delayacct_thrashing_end(); |
1155 | psi_memstall_leave(&pflags); | |
1156 | } | |
b1d29ba8 | 1157 | |
62906027 NP |
1158 | /* |
1159 | * A signal could leave PageWaiters set. Clearing it here if | |
1160 | * !waitqueue_active would be possible (by open-coding finish_wait), | |
1161 | * but still fail to catch it in the case of wait hash collision. We | |
1162 | * already can fail to clear wait hash collision cases, so don't | |
1163 | * bother with signals either. | |
1164 | */ | |
1165 | ||
1166 | return ret; | |
1167 | } | |
1168 | ||
1169 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1170 | { | |
1171 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1172 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 NP |
1173 | } |
1174 | EXPORT_SYMBOL(wait_on_page_bit); | |
1175 | ||
1176 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1177 | { | |
1178 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1179 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1180 | } |
4343d008 | 1181 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1182 | |
9a1ea439 HD |
1183 | /** |
1184 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1185 | * @page: The page to wait for. | |
1186 | * | |
1187 | * The caller should hold a reference on @page. They expect the page to | |
1188 | * become unlocked relatively soon, but do not wish to hold up migration | |
1189 | * (for example) by holding the reference while waiting for the page to | |
1190 | * come unlocked. After this function returns, the caller should not | |
1191 | * dereference @page. | |
1192 | */ | |
1193 | void put_and_wait_on_page_locked(struct page *page) | |
1194 | { | |
1195 | wait_queue_head_t *q; | |
1196 | ||
1197 | page = compound_head(page); | |
1198 | q = page_waitqueue(page); | |
1199 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP); | |
1200 | } | |
1201 | ||
385e1ca5 DH |
1202 | /** |
1203 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1204 | * @page: Page defining the wait queue of interest |
1205 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1206 | * |
1207 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1208 | */ | |
ac6424b9 | 1209 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1210 | { |
1211 | wait_queue_head_t *q = page_waitqueue(page); | |
1212 | unsigned long flags; | |
1213 | ||
1214 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1215 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1216 | SetPageWaiters(page); |
385e1ca5 DH |
1217 | spin_unlock_irqrestore(&q->lock, flags); |
1218 | } | |
1219 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1220 | ||
b91e1302 LT |
1221 | #ifndef clear_bit_unlock_is_negative_byte |
1222 | ||
1223 | /* | |
1224 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1225 | * | |
1226 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1227 | * test the sign bit at the same time. But if the architecture does | |
1228 | * not support that special operation, we just do this all by hand | |
1229 | * instead. | |
1230 | * | |
1231 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1232 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1233 | * in the same byte as PG_locked. |
1234 | */ | |
1235 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1236 | { | |
1237 | clear_bit_unlock(nr, mem); | |
1238 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1239 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1240 | } |
1241 | ||
1242 | #endif | |
1243 | ||
1da177e4 | 1244 | /** |
485bb99b | 1245 | * unlock_page - unlock a locked page |
1da177e4 LT |
1246 | * @page: the page |
1247 | * | |
1248 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1249 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1250 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1251 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1252 | * | |
b91e1302 LT |
1253 | * Note that this depends on PG_waiters being the sign bit in the byte |
1254 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1255 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1256 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1257 | * test the sign bit). | |
1da177e4 | 1258 | */ |
920c7a5d | 1259 | void unlock_page(struct page *page) |
1da177e4 | 1260 | { |
b91e1302 | 1261 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1262 | page = compound_head(page); |
309381fe | 1263 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1264 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1265 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1266 | } |
1267 | EXPORT_SYMBOL(unlock_page); | |
1268 | ||
485bb99b RD |
1269 | /** |
1270 | * end_page_writeback - end writeback against a page | |
1271 | * @page: the page | |
1da177e4 LT |
1272 | */ |
1273 | void end_page_writeback(struct page *page) | |
1274 | { | |
888cf2db MG |
1275 | /* |
1276 | * TestClearPageReclaim could be used here but it is an atomic | |
1277 | * operation and overkill in this particular case. Failing to | |
1278 | * shuffle a page marked for immediate reclaim is too mild to | |
1279 | * justify taking an atomic operation penalty at the end of | |
1280 | * ever page writeback. | |
1281 | */ | |
1282 | if (PageReclaim(page)) { | |
1283 | ClearPageReclaim(page); | |
ac6aadb2 | 1284 | rotate_reclaimable_page(page); |
888cf2db | 1285 | } |
ac6aadb2 MS |
1286 | |
1287 | if (!test_clear_page_writeback(page)) | |
1288 | BUG(); | |
1289 | ||
4e857c58 | 1290 | smp_mb__after_atomic(); |
1da177e4 LT |
1291 | wake_up_page(page, PG_writeback); |
1292 | } | |
1293 | EXPORT_SYMBOL(end_page_writeback); | |
1294 | ||
57d99845 MW |
1295 | /* |
1296 | * After completing I/O on a page, call this routine to update the page | |
1297 | * flags appropriately | |
1298 | */ | |
c11f0c0b | 1299 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1300 | { |
c11f0c0b | 1301 | if (!is_write) { |
57d99845 MW |
1302 | if (!err) { |
1303 | SetPageUptodate(page); | |
1304 | } else { | |
1305 | ClearPageUptodate(page); | |
1306 | SetPageError(page); | |
1307 | } | |
1308 | unlock_page(page); | |
abf54548 | 1309 | } else { |
57d99845 | 1310 | if (err) { |
dd8416c4 MK |
1311 | struct address_space *mapping; |
1312 | ||
57d99845 | 1313 | SetPageError(page); |
dd8416c4 MK |
1314 | mapping = page_mapping(page); |
1315 | if (mapping) | |
1316 | mapping_set_error(mapping, err); | |
57d99845 MW |
1317 | } |
1318 | end_page_writeback(page); | |
1319 | } | |
1320 | } | |
1321 | EXPORT_SYMBOL_GPL(page_endio); | |
1322 | ||
485bb99b RD |
1323 | /** |
1324 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1325 | * @__page: the page to lock |
1da177e4 | 1326 | */ |
62906027 | 1327 | void __lock_page(struct page *__page) |
1da177e4 | 1328 | { |
62906027 NP |
1329 | struct page *page = compound_head(__page); |
1330 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1331 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
1332 | EXCLUSIVE); | |
1da177e4 LT |
1333 | } |
1334 | EXPORT_SYMBOL(__lock_page); | |
1335 | ||
62906027 | 1336 | int __lock_page_killable(struct page *__page) |
2687a356 | 1337 | { |
62906027 NP |
1338 | struct page *page = compound_head(__page); |
1339 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1340 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
1341 | EXCLUSIVE); | |
2687a356 | 1342 | } |
18bc0bbd | 1343 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1344 | |
9a95f3cf PC |
1345 | /* |
1346 | * Return values: | |
c1e8d7c6 | 1347 | * 1 - page is locked; mmap_lock is still held. |
9a95f3cf | 1348 | * 0 - page is not locked. |
3e4e28c5 | 1349 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1350 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1351 | * which case mmap_lock is still held. |
9a95f3cf PC |
1352 | * |
1353 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
c1e8d7c6 | 1354 | * with the page locked and the mmap_lock unperturbed. |
9a95f3cf | 1355 | */ |
d065bd81 ML |
1356 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1357 | unsigned int flags) | |
1358 | { | |
4064b982 | 1359 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1360 | /* |
c1e8d7c6 | 1361 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1362 | * even though return 0. |
1363 | */ | |
1364 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1365 | return 0; | |
1366 | ||
d8ed45c5 | 1367 | mmap_read_unlock(mm); |
37b23e05 KM |
1368 | if (flags & FAULT_FLAG_KILLABLE) |
1369 | wait_on_page_locked_killable(page); | |
1370 | else | |
318b275f | 1371 | wait_on_page_locked(page); |
d065bd81 | 1372 | return 0; |
37b23e05 KM |
1373 | } else { |
1374 | if (flags & FAULT_FLAG_KILLABLE) { | |
1375 | int ret; | |
1376 | ||
1377 | ret = __lock_page_killable(page); | |
1378 | if (ret) { | |
d8ed45c5 | 1379 | mmap_read_unlock(mm); |
37b23e05 KM |
1380 | return 0; |
1381 | } | |
1382 | } else | |
1383 | __lock_page(page); | |
1384 | return 1; | |
d065bd81 ML |
1385 | } |
1386 | } | |
1387 | ||
e7b563bb | 1388 | /** |
0d3f9296 MW |
1389 | * page_cache_next_miss() - Find the next gap in the page cache. |
1390 | * @mapping: Mapping. | |
1391 | * @index: Index. | |
1392 | * @max_scan: Maximum range to search. | |
e7b563bb | 1393 | * |
0d3f9296 MW |
1394 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1395 | * gap with the lowest index. | |
e7b563bb | 1396 | * |
0d3f9296 MW |
1397 | * This function may be called under the rcu_read_lock. However, this will |
1398 | * not atomically search a snapshot of the cache at a single point in time. | |
1399 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1400 | * created at index 10, page_cache_next_miss covering both indices may | |
1401 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1402 | * |
0d3f9296 MW |
1403 | * Return: The index of the gap if found, otherwise an index outside the |
1404 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1405 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1406 | */ |
0d3f9296 | 1407 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1408 | pgoff_t index, unsigned long max_scan) |
1409 | { | |
0d3f9296 | 1410 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1411 | |
0d3f9296 MW |
1412 | while (max_scan--) { |
1413 | void *entry = xas_next(&xas); | |
1414 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1415 | break; |
0d3f9296 | 1416 | if (xas.xa_index == 0) |
e7b563bb JW |
1417 | break; |
1418 | } | |
1419 | ||
0d3f9296 | 1420 | return xas.xa_index; |
e7b563bb | 1421 | } |
0d3f9296 | 1422 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1423 | |
1424 | /** | |
2346a560 | 1425 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1426 | * @mapping: Mapping. |
1427 | * @index: Index. | |
1428 | * @max_scan: Maximum range to search. | |
e7b563bb | 1429 | * |
0d3f9296 MW |
1430 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1431 | * gap with the highest index. | |
e7b563bb | 1432 | * |
0d3f9296 MW |
1433 | * This function may be called under the rcu_read_lock. However, this will |
1434 | * not atomically search a snapshot of the cache at a single point in time. | |
1435 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1436 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1437 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1438 | * |
0d3f9296 MW |
1439 | * Return: The index of the gap if found, otherwise an index outside the |
1440 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1441 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1442 | */ |
0d3f9296 | 1443 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1444 | pgoff_t index, unsigned long max_scan) |
1445 | { | |
0d3f9296 | 1446 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1447 | |
0d3f9296 MW |
1448 | while (max_scan--) { |
1449 | void *entry = xas_prev(&xas); | |
1450 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1451 | break; |
0d3f9296 | 1452 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1453 | break; |
1454 | } | |
1455 | ||
0d3f9296 | 1456 | return xas.xa_index; |
e7b563bb | 1457 | } |
0d3f9296 | 1458 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1459 | |
485bb99b | 1460 | /** |
0cd6144a | 1461 | * find_get_entry - find and get a page cache entry |
485bb99b | 1462 | * @mapping: the address_space to search |
0cd6144a JW |
1463 | * @offset: the page cache index |
1464 | * | |
1465 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1466 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1467 | * |
139b6a6f JW |
1468 | * If the slot holds a shadow entry of a previously evicted page, or a |
1469 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1470 | * |
a862f68a | 1471 | * Return: the found page or shadow entry, %NULL if nothing is found. |
1da177e4 | 1472 | */ |
0cd6144a | 1473 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1474 | { |
4c7472c0 | 1475 | XA_STATE(xas, &mapping->i_pages, offset); |
4101196b | 1476 | struct page *page; |
1da177e4 | 1477 | |
a60637c8 NP |
1478 | rcu_read_lock(); |
1479 | repeat: | |
4c7472c0 MW |
1480 | xas_reset(&xas); |
1481 | page = xas_load(&xas); | |
1482 | if (xas_retry(&xas, page)) | |
1483 | goto repeat; | |
1484 | /* | |
1485 | * A shadow entry of a recently evicted page, or a swap entry from | |
1486 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1487 | */ | |
1488 | if (!page || xa_is_value(page)) | |
1489 | goto out; | |
83929372 | 1490 | |
4101196b | 1491 | if (!page_cache_get_speculative(page)) |
4c7472c0 | 1492 | goto repeat; |
83929372 | 1493 | |
4c7472c0 | 1494 | /* |
4101196b | 1495 | * Has the page moved or been split? |
4c7472c0 MW |
1496 | * This is part of the lockless pagecache protocol. See |
1497 | * include/linux/pagemap.h for details. | |
1498 | */ | |
1499 | if (unlikely(page != xas_reload(&xas))) { | |
4101196b | 1500 | put_page(page); |
4c7472c0 | 1501 | goto repeat; |
a60637c8 | 1502 | } |
4101196b | 1503 | page = find_subpage(page, offset); |
27d20fdd | 1504 | out: |
a60637c8 NP |
1505 | rcu_read_unlock(); |
1506 | ||
1da177e4 LT |
1507 | return page; |
1508 | } | |
1da177e4 | 1509 | |
0cd6144a JW |
1510 | /** |
1511 | * find_lock_entry - locate, pin and lock a page cache entry | |
1512 | * @mapping: the address_space to search | |
1513 | * @offset: the page cache index | |
1514 | * | |
1515 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1516 | * page cache page, it is returned locked and with an increased | |
1517 | * refcount. | |
1518 | * | |
139b6a6f JW |
1519 | * If the slot holds a shadow entry of a previously evicted page, or a |
1520 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1521 | * |
0cd6144a | 1522 | * find_lock_entry() may sleep. |
a862f68a MR |
1523 | * |
1524 | * Return: the found page or shadow entry, %NULL if nothing is found. | |
0cd6144a JW |
1525 | */ |
1526 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1527 | { |
1528 | struct page *page; | |
1529 | ||
1da177e4 | 1530 | repeat: |
0cd6144a | 1531 | page = find_get_entry(mapping, offset); |
4c7472c0 | 1532 | if (page && !xa_is_value(page)) { |
a60637c8 NP |
1533 | lock_page(page); |
1534 | /* Has the page been truncated? */ | |
83929372 | 1535 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1536 | unlock_page(page); |
09cbfeaf | 1537 | put_page(page); |
a60637c8 | 1538 | goto repeat; |
1da177e4 | 1539 | } |
83929372 | 1540 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1541 | } |
1da177e4 LT |
1542 | return page; |
1543 | } | |
0cd6144a JW |
1544 | EXPORT_SYMBOL(find_lock_entry); |
1545 | ||
1546 | /** | |
2294b32e MWO |
1547 | * pagecache_get_page - Find and get a reference to a page. |
1548 | * @mapping: The address_space to search. | |
1549 | * @index: The page index. | |
1550 | * @fgp_flags: %FGP flags modify how the page is returned. | |
1551 | * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1552 | * |
2294b32e | 1553 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1554 | * |
2294b32e | 1555 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1556 | * |
2294b32e MWO |
1557 | * * %FGP_ACCESSED - The page will be marked accessed. |
1558 | * * %FGP_LOCK - The page is returned locked. | |
1559 | * * %FGP_CREAT - If no page is present then a new page is allocated using | |
1560 | * @gfp_mask and added to the page cache and the VM's LRU list. | |
1561 | * The page is returned locked and with an increased refcount. | |
1562 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1563 | * page is already in cache. If the page was allocated, unlock it before | |
1564 | * returning so the caller can do the same dance. | |
1da177e4 | 1565 | * |
2294b32e MWO |
1566 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1567 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1568 | * |
2457aec6 | 1569 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1570 | * |
2294b32e | 1571 | * Return: The found page or %NULL otherwise. |
1da177e4 | 1572 | */ |
2294b32e MWO |
1573 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index, |
1574 | int fgp_flags, gfp_t gfp_mask) | |
1da177e4 | 1575 | { |
eb2be189 | 1576 | struct page *page; |
2457aec6 | 1577 | |
1da177e4 | 1578 | repeat: |
2294b32e | 1579 | page = find_get_entry(mapping, index); |
3159f943 | 1580 | if (xa_is_value(page)) |
2457aec6 MG |
1581 | page = NULL; |
1582 | if (!page) | |
1583 | goto no_page; | |
1584 | ||
1585 | if (fgp_flags & FGP_LOCK) { | |
1586 | if (fgp_flags & FGP_NOWAIT) { | |
1587 | if (!trylock_page(page)) { | |
09cbfeaf | 1588 | put_page(page); |
2457aec6 MG |
1589 | return NULL; |
1590 | } | |
1591 | } else { | |
1592 | lock_page(page); | |
1593 | } | |
1594 | ||
1595 | /* Has the page been truncated? */ | |
31895438 | 1596 | if (unlikely(compound_head(page)->mapping != mapping)) { |
2457aec6 | 1597 | unlock_page(page); |
09cbfeaf | 1598 | put_page(page); |
2457aec6 MG |
1599 | goto repeat; |
1600 | } | |
2294b32e | 1601 | VM_BUG_ON_PAGE(page->index != index, page); |
2457aec6 MG |
1602 | } |
1603 | ||
c16eb000 | 1604 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 MG |
1605 | mark_page_accessed(page); |
1606 | ||
1607 | no_page: | |
1608 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1609 | int err; | |
1610 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1611 | gfp_mask |= __GFP_WRITE; |
1612 | if (fgp_flags & FGP_NOFS) | |
1613 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1614 | |
45f87de5 | 1615 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1616 | if (!page) |
1617 | return NULL; | |
2457aec6 | 1618 | |
a75d4c33 | 1619 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1620 | fgp_flags |= FGP_LOCK; |
1621 | ||
eb39d618 | 1622 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1623 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1624 | __SetPageReferenced(page); |
2457aec6 | 1625 | |
2294b32e | 1626 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); |
eb2be189 | 1627 | if (unlikely(err)) { |
09cbfeaf | 1628 | put_page(page); |
eb2be189 NP |
1629 | page = NULL; |
1630 | if (err == -EEXIST) | |
1631 | goto repeat; | |
1da177e4 | 1632 | } |
a75d4c33 JB |
1633 | |
1634 | /* | |
1635 | * add_to_page_cache_lru locks the page, and for mmap we expect | |
1636 | * an unlocked page. | |
1637 | */ | |
1638 | if (page && (fgp_flags & FGP_FOR_MMAP)) | |
1639 | unlock_page(page); | |
1da177e4 | 1640 | } |
2457aec6 | 1641 | |
1da177e4 LT |
1642 | return page; |
1643 | } | |
2457aec6 | 1644 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1645 | |
0cd6144a JW |
1646 | /** |
1647 | * find_get_entries - gang pagecache lookup | |
1648 | * @mapping: The address_space to search | |
1649 | * @start: The starting page cache index | |
1650 | * @nr_entries: The maximum number of entries | |
1651 | * @entries: Where the resulting entries are placed | |
1652 | * @indices: The cache indices corresponding to the entries in @entries | |
1653 | * | |
1654 | * find_get_entries() will search for and return a group of up to | |
1655 | * @nr_entries entries in the mapping. The entries are placed at | |
1656 | * @entries. find_get_entries() takes a reference against any actual | |
1657 | * pages it returns. | |
1658 | * | |
1659 | * The search returns a group of mapping-contiguous page cache entries | |
1660 | * with ascending indexes. There may be holes in the indices due to | |
1661 | * not-present pages. | |
1662 | * | |
139b6a6f JW |
1663 | * Any shadow entries of evicted pages, or swap entries from |
1664 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 1665 | * |
71725ed1 HD |
1666 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
1667 | * stops at that page: the caller is likely to have a better way to handle | |
1668 | * the compound page as a whole, and then skip its extent, than repeatedly | |
1669 | * calling find_get_entries() to return all its tails. | |
1670 | * | |
a862f68a | 1671 | * Return: the number of pages and shadow entries which were found. |
0cd6144a JW |
1672 | */ |
1673 | unsigned find_get_entries(struct address_space *mapping, | |
1674 | pgoff_t start, unsigned int nr_entries, | |
1675 | struct page **entries, pgoff_t *indices) | |
1676 | { | |
f280bf09 MW |
1677 | XA_STATE(xas, &mapping->i_pages, start); |
1678 | struct page *page; | |
0cd6144a | 1679 | unsigned int ret = 0; |
0cd6144a JW |
1680 | |
1681 | if (!nr_entries) | |
1682 | return 0; | |
1683 | ||
1684 | rcu_read_lock(); | |
f280bf09 | 1685 | xas_for_each(&xas, page, ULONG_MAX) { |
f280bf09 | 1686 | if (xas_retry(&xas, page)) |
0cd6144a | 1687 | continue; |
f280bf09 MW |
1688 | /* |
1689 | * A shadow entry of a recently evicted page, a swap | |
1690 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1691 | * without attempting to raise page count. | |
1692 | */ | |
1693 | if (xa_is_value(page)) | |
0cd6144a | 1694 | goto export; |
83929372 | 1695 | |
4101196b | 1696 | if (!page_cache_get_speculative(page)) |
f280bf09 | 1697 | goto retry; |
83929372 | 1698 | |
4101196b | 1699 | /* Has the page moved or been split? */ |
f280bf09 MW |
1700 | if (unlikely(page != xas_reload(&xas))) |
1701 | goto put_page; | |
1702 | ||
71725ed1 HD |
1703 | /* |
1704 | * Terminate early on finding a THP, to allow the caller to | |
1705 | * handle it all at once; but continue if this is hugetlbfs. | |
1706 | */ | |
1707 | if (PageTransHuge(page) && !PageHuge(page)) { | |
1708 | page = find_subpage(page, xas.xa_index); | |
1709 | nr_entries = ret + 1; | |
1710 | } | |
0cd6144a | 1711 | export: |
f280bf09 | 1712 | indices[ret] = xas.xa_index; |
0cd6144a JW |
1713 | entries[ret] = page; |
1714 | if (++ret == nr_entries) | |
1715 | break; | |
f280bf09 MW |
1716 | continue; |
1717 | put_page: | |
4101196b | 1718 | put_page(page); |
f280bf09 MW |
1719 | retry: |
1720 | xas_reset(&xas); | |
0cd6144a JW |
1721 | } |
1722 | rcu_read_unlock(); | |
1723 | return ret; | |
1724 | } | |
1725 | ||
1da177e4 | 1726 | /** |
b947cee4 | 1727 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1728 | * @mapping: The address_space to search |
1729 | * @start: The starting page index | |
b947cee4 | 1730 | * @end: The final page index (inclusive) |
1da177e4 LT |
1731 | * @nr_pages: The maximum number of pages |
1732 | * @pages: Where the resulting pages are placed | |
1733 | * | |
b947cee4 JK |
1734 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1735 | * pages in the mapping starting at index @start and up to index @end | |
1736 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1737 | * a reference against the returned pages. | |
1da177e4 LT |
1738 | * |
1739 | * The search returns a group of mapping-contiguous pages with ascending | |
1740 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1741 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1742 | * |
a862f68a MR |
1743 | * Return: the number of pages which were found. If this number is |
1744 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 1745 | * reached. |
1da177e4 | 1746 | */ |
b947cee4 JK |
1747 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1748 | pgoff_t end, unsigned int nr_pages, | |
1749 | struct page **pages) | |
1da177e4 | 1750 | { |
fd1b3cee MW |
1751 | XA_STATE(xas, &mapping->i_pages, *start); |
1752 | struct page *page; | |
0fc9d104 KK |
1753 | unsigned ret = 0; |
1754 | ||
1755 | if (unlikely(!nr_pages)) | |
1756 | return 0; | |
a60637c8 NP |
1757 | |
1758 | rcu_read_lock(); | |
fd1b3cee | 1759 | xas_for_each(&xas, page, end) { |
fd1b3cee | 1760 | if (xas_retry(&xas, page)) |
a60637c8 | 1761 | continue; |
fd1b3cee MW |
1762 | /* Skip over shadow, swap and DAX entries */ |
1763 | if (xa_is_value(page)) | |
8079b1c8 | 1764 | continue; |
a60637c8 | 1765 | |
4101196b | 1766 | if (!page_cache_get_speculative(page)) |
fd1b3cee | 1767 | goto retry; |
83929372 | 1768 | |
4101196b | 1769 | /* Has the page moved or been split? */ |
fd1b3cee MW |
1770 | if (unlikely(page != xas_reload(&xas))) |
1771 | goto put_page; | |
1da177e4 | 1772 | |
4101196b | 1773 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 1774 | if (++ret == nr_pages) { |
5d3ee42f | 1775 | *start = xas.xa_index + 1; |
b947cee4 JK |
1776 | goto out; |
1777 | } | |
fd1b3cee MW |
1778 | continue; |
1779 | put_page: | |
4101196b | 1780 | put_page(page); |
fd1b3cee MW |
1781 | retry: |
1782 | xas_reset(&xas); | |
a60637c8 | 1783 | } |
5b280c0c | 1784 | |
b947cee4 JK |
1785 | /* |
1786 | * We come here when there is no page beyond @end. We take care to not | |
1787 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 1788 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
1789 | * already broken anyway. |
1790 | */ | |
1791 | if (end == (pgoff_t)-1) | |
1792 | *start = (pgoff_t)-1; | |
1793 | else | |
1794 | *start = end + 1; | |
1795 | out: | |
a60637c8 | 1796 | rcu_read_unlock(); |
d72dc8a2 | 1797 | |
1da177e4 LT |
1798 | return ret; |
1799 | } | |
1800 | ||
ebf43500 JA |
1801 | /** |
1802 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1803 | * @mapping: The address_space to search | |
1804 | * @index: The starting page index | |
1805 | * @nr_pages: The maximum number of pages | |
1806 | * @pages: Where the resulting pages are placed | |
1807 | * | |
1808 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1809 | * that the returned number of pages are guaranteed to be contiguous. | |
1810 | * | |
a862f68a | 1811 | * Return: the number of pages which were found. |
ebf43500 JA |
1812 | */ |
1813 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1814 | unsigned int nr_pages, struct page **pages) | |
1815 | { | |
3ece58a2 MW |
1816 | XA_STATE(xas, &mapping->i_pages, index); |
1817 | struct page *page; | |
0fc9d104 KK |
1818 | unsigned int ret = 0; |
1819 | ||
1820 | if (unlikely(!nr_pages)) | |
1821 | return 0; | |
a60637c8 NP |
1822 | |
1823 | rcu_read_lock(); | |
3ece58a2 | 1824 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
1825 | if (xas_retry(&xas, page)) |
1826 | continue; | |
1827 | /* | |
1828 | * If the entry has been swapped out, we can stop looking. | |
1829 | * No current caller is looking for DAX entries. | |
1830 | */ | |
1831 | if (xa_is_value(page)) | |
8079b1c8 | 1832 | break; |
ebf43500 | 1833 | |
4101196b | 1834 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 1835 | goto retry; |
83929372 | 1836 | |
4101196b | 1837 | /* Has the page moved or been split? */ |
3ece58a2 MW |
1838 | if (unlikely(page != xas_reload(&xas))) |
1839 | goto put_page; | |
a60637c8 | 1840 | |
4101196b | 1841 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
1842 | if (++ret == nr_pages) |
1843 | break; | |
3ece58a2 MW |
1844 | continue; |
1845 | put_page: | |
4101196b | 1846 | put_page(page); |
3ece58a2 MW |
1847 | retry: |
1848 | xas_reset(&xas); | |
ebf43500 | 1849 | } |
a60637c8 NP |
1850 | rcu_read_unlock(); |
1851 | return ret; | |
ebf43500 | 1852 | } |
ef71c15c | 1853 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1854 | |
485bb99b | 1855 | /** |
72b045ae | 1856 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
1857 | * @mapping: the address_space to search |
1858 | * @index: the starting page index | |
72b045ae | 1859 | * @end: The final page index (inclusive) |
485bb99b RD |
1860 | * @tag: the tag index |
1861 | * @nr_pages: the maximum number of pages | |
1862 | * @pages: where the resulting pages are placed | |
1863 | * | |
1da177e4 | 1864 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1865 | * @tag. We update @index to index the next page for the traversal. |
a862f68a MR |
1866 | * |
1867 | * Return: the number of pages which were found. | |
1da177e4 | 1868 | */ |
72b045ae | 1869 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 1870 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 1871 | struct page **pages) |
1da177e4 | 1872 | { |
a6906972 MW |
1873 | XA_STATE(xas, &mapping->i_pages, *index); |
1874 | struct page *page; | |
0fc9d104 KK |
1875 | unsigned ret = 0; |
1876 | ||
1877 | if (unlikely(!nr_pages)) | |
1878 | return 0; | |
a60637c8 NP |
1879 | |
1880 | rcu_read_lock(); | |
a6906972 | 1881 | xas_for_each_marked(&xas, page, end, tag) { |
a6906972 | 1882 | if (xas_retry(&xas, page)) |
a60637c8 | 1883 | continue; |
a6906972 MW |
1884 | /* |
1885 | * Shadow entries should never be tagged, but this iteration | |
1886 | * is lockless so there is a window for page reclaim to evict | |
1887 | * a page we saw tagged. Skip over it. | |
1888 | */ | |
1889 | if (xa_is_value(page)) | |
139b6a6f | 1890 | continue; |
a60637c8 | 1891 | |
4101196b | 1892 | if (!page_cache_get_speculative(page)) |
a6906972 | 1893 | goto retry; |
a60637c8 | 1894 | |
4101196b | 1895 | /* Has the page moved or been split? */ |
a6906972 MW |
1896 | if (unlikely(page != xas_reload(&xas))) |
1897 | goto put_page; | |
a60637c8 | 1898 | |
4101196b | 1899 | pages[ret] = find_subpage(page, xas.xa_index); |
72b045ae | 1900 | if (++ret == nr_pages) { |
5d3ee42f | 1901 | *index = xas.xa_index + 1; |
72b045ae JK |
1902 | goto out; |
1903 | } | |
a6906972 MW |
1904 | continue; |
1905 | put_page: | |
4101196b | 1906 | put_page(page); |
a6906972 MW |
1907 | retry: |
1908 | xas_reset(&xas); | |
a60637c8 | 1909 | } |
5b280c0c | 1910 | |
72b045ae | 1911 | /* |
a6906972 | 1912 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 1913 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
1914 | * iteration when there is a page at index -1 but that is already |
1915 | * broken anyway. | |
72b045ae JK |
1916 | */ |
1917 | if (end == (pgoff_t)-1) | |
1918 | *index = (pgoff_t)-1; | |
1919 | else | |
1920 | *index = end + 1; | |
1921 | out: | |
a60637c8 | 1922 | rcu_read_unlock(); |
1da177e4 | 1923 | |
1da177e4 LT |
1924 | return ret; |
1925 | } | |
72b045ae | 1926 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 1927 | |
76d42bd9 WF |
1928 | /* |
1929 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1930 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1931 | * | |
1932 | * ---R__________________________________________B__________ | |
1933 | * ^ reading here ^ bad block(assume 4k) | |
1934 | * | |
1935 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1936 | * => failing the whole request => read(R) => read(R+1) => | |
1937 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1938 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1939 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1940 | * | |
1941 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1942 | */ | |
0f8e2db4 | 1943 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 1944 | { |
76d42bd9 | 1945 | ra->ra_pages /= 4; |
76d42bd9 WF |
1946 | } |
1947 | ||
485bb99b | 1948 | /** |
47c27bc4 CH |
1949 | * generic_file_buffered_read - generic file read routine |
1950 | * @iocb: the iocb to read | |
6e58e79d AV |
1951 | * @iter: data destination |
1952 | * @written: already copied | |
485bb99b | 1953 | * |
1da177e4 | 1954 | * This is a generic file read routine, and uses the |
485bb99b | 1955 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1956 | * |
1957 | * This is really ugly. But the goto's actually try to clarify some | |
1958 | * of the logic when it comes to error handling etc. | |
a862f68a MR |
1959 | * |
1960 | * Return: | |
1961 | * * total number of bytes copied, including those the were already @written | |
1962 | * * negative error code if nothing was copied | |
1da177e4 | 1963 | */ |
d85dc2e1 | 1964 | ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 1965 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 1966 | { |
47c27bc4 | 1967 | struct file *filp = iocb->ki_filp; |
36e78914 | 1968 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1969 | struct inode *inode = mapping->host; |
36e78914 | 1970 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 1971 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
1972 | pgoff_t index; |
1973 | pgoff_t last_index; | |
1974 | pgoff_t prev_index; | |
1975 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1976 | unsigned int prev_offset; |
6e58e79d | 1977 | int error = 0; |
1da177e4 | 1978 | |
c2a9737f | 1979 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 1980 | return 0; |
c2a9737f WF |
1981 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
1982 | ||
09cbfeaf KS |
1983 | index = *ppos >> PAGE_SHIFT; |
1984 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
1985 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
1986 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
1987 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 1988 | |
1da177e4 LT |
1989 | for (;;) { |
1990 | struct page *page; | |
57f6b96c | 1991 | pgoff_t end_index; |
a32ea1e1 | 1992 | loff_t isize; |
1da177e4 LT |
1993 | unsigned long nr, ret; |
1994 | ||
1da177e4 | 1995 | cond_resched(); |
1da177e4 | 1996 | find_page: |
5abf186a MH |
1997 | if (fatal_signal_pending(current)) { |
1998 | error = -EINTR; | |
1999 | goto out; | |
2000 | } | |
2001 | ||
1da177e4 | 2002 | page = find_get_page(mapping, index); |
3ea89ee8 | 2003 | if (!page) { |
cf914a7d | 2004 | page_cache_sync_readahead(mapping, |
7ff81078 | 2005 | ra, filp, |
3ea89ee8 FW |
2006 | index, last_index - index); |
2007 | page = find_get_page(mapping, index); | |
2008 | if (unlikely(page == NULL)) | |
2009 | goto no_cached_page; | |
2010 | } | |
2011 | if (PageReadahead(page)) { | |
cf914a7d | 2012 | page_cache_async_readahead(mapping, |
7ff81078 | 2013 | ra, filp, page, |
3ea89ee8 | 2014 | index, last_index - index); |
1da177e4 | 2015 | } |
8ab22b9a | 2016 | if (!PageUptodate(page)) { |
3239d834 MT |
2017 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2018 | put_page(page); | |
2019 | goto would_block; | |
2020 | } | |
2021 | ||
ebded027 MG |
2022 | /* |
2023 | * See comment in do_read_cache_page on why | |
2024 | * wait_on_page_locked is used to avoid unnecessarily | |
2025 | * serialisations and why it's safe. | |
2026 | */ | |
c4b209a4 BVA |
2027 | error = wait_on_page_locked_killable(page); |
2028 | if (unlikely(error)) | |
2029 | goto readpage_error; | |
ebded027 MG |
2030 | if (PageUptodate(page)) |
2031 | goto page_ok; | |
2032 | ||
09cbfeaf | 2033 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2034 | !mapping->a_ops->is_partially_uptodate) |
2035 | goto page_not_up_to_date; | |
6d6d36bc | 2036 | /* pipes can't handle partially uptodate pages */ |
00e23707 | 2037 | if (unlikely(iov_iter_is_pipe(iter))) |
6d6d36bc | 2038 | goto page_not_up_to_date; |
529ae9aa | 2039 | if (!trylock_page(page)) |
8ab22b9a | 2040 | goto page_not_up_to_date; |
8d056cb9 DH |
2041 | /* Did it get truncated before we got the lock? */ |
2042 | if (!page->mapping) | |
2043 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2044 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2045 | offset, iter->count)) |
8ab22b9a HH |
2046 | goto page_not_up_to_date_locked; |
2047 | unlock_page(page); | |
2048 | } | |
1da177e4 | 2049 | page_ok: |
a32ea1e1 N |
2050 | /* |
2051 | * i_size must be checked after we know the page is Uptodate. | |
2052 | * | |
2053 | * Checking i_size after the check allows us to calculate | |
2054 | * the correct value for "nr", which means the zero-filled | |
2055 | * part of the page is not copied back to userspace (unless | |
2056 | * another truncate extends the file - this is desired though). | |
2057 | */ | |
2058 | ||
2059 | isize = i_size_read(inode); | |
09cbfeaf | 2060 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2061 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2062 | put_page(page); |
a32ea1e1 N |
2063 | goto out; |
2064 | } | |
2065 | ||
2066 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2067 | nr = PAGE_SIZE; |
a32ea1e1 | 2068 | if (index == end_index) { |
09cbfeaf | 2069 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2070 | if (nr <= offset) { |
09cbfeaf | 2071 | put_page(page); |
a32ea1e1 N |
2072 | goto out; |
2073 | } | |
2074 | } | |
2075 | nr = nr - offset; | |
1da177e4 LT |
2076 | |
2077 | /* If users can be writing to this page using arbitrary | |
2078 | * virtual addresses, take care about potential aliasing | |
2079 | * before reading the page on the kernel side. | |
2080 | */ | |
2081 | if (mapping_writably_mapped(mapping)) | |
2082 | flush_dcache_page(page); | |
2083 | ||
2084 | /* | |
ec0f1637 JK |
2085 | * When a sequential read accesses a page several times, |
2086 | * only mark it as accessed the first time. | |
1da177e4 | 2087 | */ |
ec0f1637 | 2088 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2089 | mark_page_accessed(page); |
2090 | prev_index = index; | |
2091 | ||
2092 | /* | |
2093 | * Ok, we have the page, and it's up-to-date, so | |
2094 | * now we can copy it to user space... | |
1da177e4 | 2095 | */ |
6e58e79d AV |
2096 | |
2097 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2098 | offset += ret; |
09cbfeaf KS |
2099 | index += offset >> PAGE_SHIFT; |
2100 | offset &= ~PAGE_MASK; | |
6ce745ed | 2101 | prev_offset = offset; |
1da177e4 | 2102 | |
09cbfeaf | 2103 | put_page(page); |
6e58e79d AV |
2104 | written += ret; |
2105 | if (!iov_iter_count(iter)) | |
2106 | goto out; | |
2107 | if (ret < nr) { | |
2108 | error = -EFAULT; | |
2109 | goto out; | |
2110 | } | |
2111 | continue; | |
1da177e4 LT |
2112 | |
2113 | page_not_up_to_date: | |
2114 | /* Get exclusive access to the page ... */ | |
85462323 ON |
2115 | error = lock_page_killable(page); |
2116 | if (unlikely(error)) | |
2117 | goto readpage_error; | |
1da177e4 | 2118 | |
8ab22b9a | 2119 | page_not_up_to_date_locked: |
da6052f7 | 2120 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2121 | if (!page->mapping) { |
2122 | unlock_page(page); | |
09cbfeaf | 2123 | put_page(page); |
1da177e4 LT |
2124 | continue; |
2125 | } | |
2126 | ||
2127 | /* Did somebody else fill it already? */ | |
2128 | if (PageUptodate(page)) { | |
2129 | unlock_page(page); | |
2130 | goto page_ok; | |
2131 | } | |
2132 | ||
2133 | readpage: | |
91803b49 JM |
2134 | /* |
2135 | * A previous I/O error may have been due to temporary | |
2136 | * failures, eg. multipath errors. | |
2137 | * PG_error will be set again if readpage fails. | |
2138 | */ | |
2139 | ClearPageError(page); | |
1da177e4 LT |
2140 | /* Start the actual read. The read will unlock the page. */ |
2141 | error = mapping->a_ops->readpage(filp, page); | |
2142 | ||
994fc28c ZB |
2143 | if (unlikely(error)) { |
2144 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2145 | put_page(page); |
6e58e79d | 2146 | error = 0; |
994fc28c ZB |
2147 | goto find_page; |
2148 | } | |
1da177e4 | 2149 | goto readpage_error; |
994fc28c | 2150 | } |
1da177e4 LT |
2151 | |
2152 | if (!PageUptodate(page)) { | |
85462323 ON |
2153 | error = lock_page_killable(page); |
2154 | if (unlikely(error)) | |
2155 | goto readpage_error; | |
1da177e4 LT |
2156 | if (!PageUptodate(page)) { |
2157 | if (page->mapping == NULL) { | |
2158 | /* | |
2ecdc82e | 2159 | * invalidate_mapping_pages got it |
1da177e4 LT |
2160 | */ |
2161 | unlock_page(page); | |
09cbfeaf | 2162 | put_page(page); |
1da177e4 LT |
2163 | goto find_page; |
2164 | } | |
2165 | unlock_page(page); | |
0f8e2db4 | 2166 | shrink_readahead_size_eio(ra); |
85462323 ON |
2167 | error = -EIO; |
2168 | goto readpage_error; | |
1da177e4 LT |
2169 | } |
2170 | unlock_page(page); | |
2171 | } | |
2172 | ||
1da177e4 LT |
2173 | goto page_ok; |
2174 | ||
2175 | readpage_error: | |
2176 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2177 | put_page(page); |
1da177e4 LT |
2178 | goto out; |
2179 | ||
2180 | no_cached_page: | |
2181 | /* | |
2182 | * Ok, it wasn't cached, so we need to create a new | |
2183 | * page.. | |
2184 | */ | |
453f85d4 | 2185 | page = page_cache_alloc(mapping); |
eb2be189 | 2186 | if (!page) { |
6e58e79d | 2187 | error = -ENOMEM; |
eb2be189 | 2188 | goto out; |
1da177e4 | 2189 | } |
6afdb859 | 2190 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2191 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2192 | if (error) { |
09cbfeaf | 2193 | put_page(page); |
6e58e79d AV |
2194 | if (error == -EEXIST) { |
2195 | error = 0; | |
1da177e4 | 2196 | goto find_page; |
6e58e79d | 2197 | } |
1da177e4 LT |
2198 | goto out; |
2199 | } | |
1da177e4 LT |
2200 | goto readpage; |
2201 | } | |
2202 | ||
3239d834 MT |
2203 | would_block: |
2204 | error = -EAGAIN; | |
1da177e4 | 2205 | out: |
7ff81078 | 2206 | ra->prev_pos = prev_index; |
09cbfeaf | 2207 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2208 | ra->prev_pos |= prev_offset; |
1da177e4 | 2209 | |
09cbfeaf | 2210 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2211 | file_accessed(filp); |
6e58e79d | 2212 | return written ? written : error; |
1da177e4 | 2213 | } |
d85dc2e1 | 2214 | EXPORT_SYMBOL_GPL(generic_file_buffered_read); |
1da177e4 | 2215 | |
485bb99b | 2216 | /** |
6abd2322 | 2217 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2218 | * @iocb: kernel I/O control block |
6abd2322 | 2219 | * @iter: destination for the data read |
485bb99b | 2220 | * |
6abd2322 | 2221 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2222 | * that can use the page cache directly. |
a862f68a MR |
2223 | * Return: |
2224 | * * number of bytes copied, even for partial reads | |
2225 | * * negative error code if nothing was read | |
1da177e4 LT |
2226 | */ |
2227 | ssize_t | |
ed978a81 | 2228 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2229 | { |
e7080a43 | 2230 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2231 | ssize_t retval = 0; |
e7080a43 NS |
2232 | |
2233 | if (!count) | |
2234 | goto out; /* skip atime */ | |
1da177e4 | 2235 | |
2ba48ce5 | 2236 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2237 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2238 | struct address_space *mapping = file->f_mapping; |
2239 | struct inode *inode = mapping->host; | |
543ade1f | 2240 | loff_t size; |
1da177e4 | 2241 | |
1da177e4 | 2242 | size = i_size_read(inode); |
6be96d3a GR |
2243 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2244 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2245 | iocb->ki_pos + count - 1)) | |
2246 | return -EAGAIN; | |
2247 | } else { | |
2248 | retval = filemap_write_and_wait_range(mapping, | |
2249 | iocb->ki_pos, | |
2250 | iocb->ki_pos + count - 1); | |
2251 | if (retval < 0) | |
2252 | goto out; | |
2253 | } | |
d8d3d94b | 2254 | |
0d5b0cf2 CH |
2255 | file_accessed(file); |
2256 | ||
5ecda137 | 2257 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2258 | if (retval >= 0) { |
c64fb5c7 | 2259 | iocb->ki_pos += retval; |
5ecda137 | 2260 | count -= retval; |
9fe55eea | 2261 | } |
5b47d59a | 2262 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2263 | |
9fe55eea SW |
2264 | /* |
2265 | * Btrfs can have a short DIO read if we encounter | |
2266 | * compressed extents, so if there was an error, or if | |
2267 | * we've already read everything we wanted to, or if | |
2268 | * there was a short read because we hit EOF, go ahead | |
2269 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2270 | * the rest of the read. Buffered reads will not work for |
2271 | * DAX files, so don't bother trying. | |
9fe55eea | 2272 | */ |
5ecda137 | 2273 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2274 | IS_DAX(inode)) |
9fe55eea | 2275 | goto out; |
1da177e4 LT |
2276 | } |
2277 | ||
47c27bc4 | 2278 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2279 | out: |
2280 | return retval; | |
2281 | } | |
ed978a81 | 2282 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2283 | |
1da177e4 | 2284 | #ifdef CONFIG_MMU |
1da177e4 | 2285 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2286 | /* |
c1e8d7c6 | 2287 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2288 | * @vmf - the vm_fault for this fault. |
2289 | * @page - the page to lock. | |
2290 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2291 | * | |
c1e8d7c6 | 2292 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2293 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2294 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2295 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2296 | */ | |
2297 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2298 | struct file **fpin) | |
2299 | { | |
2300 | if (trylock_page(page)) | |
2301 | return 1; | |
2302 | ||
8b0f9fa2 LT |
2303 | /* |
2304 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2305 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2306 | * is supposed to work. We have way too many special cases.. |
2307 | */ | |
6b4c9f44 JB |
2308 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2309 | return 0; | |
2310 | ||
2311 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2312 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
2313 | if (__lock_page_killable(page)) { | |
2314 | /* | |
c1e8d7c6 | 2315 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2316 | * but all fault_handlers only check for fatal signals |
2317 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2318 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2319 | */ |
2320 | if (*fpin == NULL) | |
d8ed45c5 | 2321 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2322 | return 0; |
2323 | } | |
2324 | } else | |
2325 | __lock_page(page); | |
2326 | return 1; | |
2327 | } | |
2328 | ||
1da177e4 | 2329 | |
ef00e08e | 2330 | /* |
6b4c9f44 JB |
2331 | * Synchronous readahead happens when we don't even find a page in the page |
2332 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2333 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2334 | * that. If we didn't pin a file then we return NULL. The file that is | |
2335 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2336 | */ |
6b4c9f44 | 2337 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2338 | { |
2a1180f1 JB |
2339 | struct file *file = vmf->vma->vm_file; |
2340 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2341 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2342 | struct file *fpin = NULL; |
2a1180f1 | 2343 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2344 | |
2345 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2346 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2347 | return fpin; |
275b12bf | 2348 | if (!ra->ra_pages) |
6b4c9f44 | 2349 | return fpin; |
ef00e08e | 2350 | |
2a1180f1 | 2351 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2352 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
7ffc59b4 WF |
2353 | page_cache_sync_readahead(mapping, ra, file, offset, |
2354 | ra->ra_pages); | |
6b4c9f44 | 2355 | return fpin; |
ef00e08e LT |
2356 | } |
2357 | ||
207d04ba AK |
2358 | /* Avoid banging the cache line if not needed */ |
2359 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2360 | ra->mmap_miss++; |
2361 | ||
2362 | /* | |
2363 | * Do we miss much more than hit in this file? If so, | |
2364 | * stop bothering with read-ahead. It will only hurt. | |
2365 | */ | |
2366 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
6b4c9f44 | 2367 | return fpin; |
ef00e08e | 2368 | |
d30a1100 WF |
2369 | /* |
2370 | * mmap read-around | |
2371 | */ | |
6b4c9f44 | 2372 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
600e19af RG |
2373 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2374 | ra->size = ra->ra_pages; | |
2375 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2376 | ra_submit(ra, mapping, file); |
6b4c9f44 | 2377 | return fpin; |
ef00e08e LT |
2378 | } |
2379 | ||
2380 | /* | |
2381 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2382 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2383 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2384 | */ |
6b4c9f44 JB |
2385 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2386 | struct page *page) | |
ef00e08e | 2387 | { |
2a1180f1 JB |
2388 | struct file *file = vmf->vma->vm_file; |
2389 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2390 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2391 | struct file *fpin = NULL; |
2a1180f1 | 2392 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2393 | |
2394 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 2395 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 2396 | return fpin; |
ef00e08e LT |
2397 | if (ra->mmap_miss > 0) |
2398 | ra->mmap_miss--; | |
6b4c9f44 JB |
2399 | if (PageReadahead(page)) { |
2400 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
2401 | page_cache_async_readahead(mapping, ra, file, |
2402 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
2403 | } |
2404 | return fpin; | |
ef00e08e LT |
2405 | } |
2406 | ||
485bb99b | 2407 | /** |
54cb8821 | 2408 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2409 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2410 | * |
54cb8821 | 2411 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2412 | * mapped memory region to read in file data during a page fault. |
2413 | * | |
2414 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2415 | * it in the page cache, and handles the special cases reasonably without | |
2416 | * having a lot of duplicated code. | |
9a95f3cf | 2417 | * |
c1e8d7c6 | 2418 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 2419 | * |
c1e8d7c6 | 2420 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 2421 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 2422 | * |
c1e8d7c6 | 2423 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
2424 | * has not been released. |
2425 | * | |
2426 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
2427 | * |
2428 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 2429 | */ |
2bcd6454 | 2430 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2431 | { |
2432 | int error; | |
11bac800 | 2433 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 2434 | struct file *fpin = NULL; |
1da177e4 LT |
2435 | struct address_space *mapping = file->f_mapping; |
2436 | struct file_ra_state *ra = &file->f_ra; | |
2437 | struct inode *inode = mapping->host; | |
ef00e08e | 2438 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2439 | pgoff_t max_off; |
1da177e4 | 2440 | struct page *page; |
2bcd6454 | 2441 | vm_fault_t ret = 0; |
1da177e4 | 2442 | |
9ab2594f MW |
2443 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2444 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2445 | return VM_FAULT_SIGBUS; |
1da177e4 | 2446 | |
1da177e4 | 2447 | /* |
49426420 | 2448 | * Do we have something in the page cache already? |
1da177e4 | 2449 | */ |
ef00e08e | 2450 | page = find_get_page(mapping, offset); |
45cac65b | 2451 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2452 | /* |
ef00e08e LT |
2453 | * We found the page, so try async readahead before |
2454 | * waiting for the lock. | |
1da177e4 | 2455 | */ |
6b4c9f44 | 2456 | fpin = do_async_mmap_readahead(vmf, page); |
45cac65b | 2457 | } else if (!page) { |
ef00e08e | 2458 | /* No page in the page cache at all */ |
ef00e08e | 2459 | count_vm_event(PGMAJFAULT); |
2262185c | 2460 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 2461 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 2462 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 2463 | retry_find: |
a75d4c33 JB |
2464 | page = pagecache_get_page(mapping, offset, |
2465 | FGP_CREAT|FGP_FOR_MMAP, | |
2466 | vmf->gfp_mask); | |
6b4c9f44 JB |
2467 | if (!page) { |
2468 | if (fpin) | |
2469 | goto out_retry; | |
e520e932 | 2470 | return VM_FAULT_OOM; |
6b4c9f44 | 2471 | } |
1da177e4 LT |
2472 | } |
2473 | ||
6b4c9f44 JB |
2474 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
2475 | goto out_retry; | |
b522c94d ML |
2476 | |
2477 | /* Did it get truncated? */ | |
585e5a7b | 2478 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
2479 | unlock_page(page); |
2480 | put_page(page); | |
2481 | goto retry_find; | |
2482 | } | |
520e5ba4 | 2483 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 2484 | |
1da177e4 | 2485 | /* |
d00806b1 NP |
2486 | * We have a locked page in the page cache, now we need to check |
2487 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2488 | */ |
d00806b1 | 2489 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2490 | goto page_not_uptodate; |
2491 | ||
6b4c9f44 | 2492 | /* |
c1e8d7c6 | 2493 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
2494 | * time to return to the upper layer and have it re-find the vma and |
2495 | * redo the fault. | |
2496 | */ | |
2497 | if (fpin) { | |
2498 | unlock_page(page); | |
2499 | goto out_retry; | |
2500 | } | |
2501 | ||
ef00e08e LT |
2502 | /* |
2503 | * Found the page and have a reference on it. | |
2504 | * We must recheck i_size under page lock. | |
2505 | */ | |
9ab2594f MW |
2506 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2507 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2508 | unlock_page(page); |
09cbfeaf | 2509 | put_page(page); |
5307cc1a | 2510 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2511 | } |
2512 | ||
d0217ac0 | 2513 | vmf->page = page; |
83c54070 | 2514 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2515 | |
1da177e4 | 2516 | page_not_uptodate: |
1da177e4 LT |
2517 | /* |
2518 | * Umm, take care of errors if the page isn't up-to-date. | |
2519 | * Try to re-read it _once_. We do this synchronously, | |
2520 | * because there really aren't any performance issues here | |
2521 | * and we need to check for errors. | |
2522 | */ | |
1da177e4 | 2523 | ClearPageError(page); |
6b4c9f44 | 2524 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
994fc28c | 2525 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2526 | if (!error) { |
2527 | wait_on_page_locked(page); | |
2528 | if (!PageUptodate(page)) | |
2529 | error = -EIO; | |
2530 | } | |
6b4c9f44 JB |
2531 | if (fpin) |
2532 | goto out_retry; | |
09cbfeaf | 2533 | put_page(page); |
d00806b1 NP |
2534 | |
2535 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2536 | goto retry_find; |
1da177e4 | 2537 | |
0f8e2db4 | 2538 | shrink_readahead_size_eio(ra); |
d0217ac0 | 2539 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
2540 | |
2541 | out_retry: | |
2542 | /* | |
c1e8d7c6 | 2543 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
2544 | * re-find the vma and come back and find our hopefully still populated |
2545 | * page. | |
2546 | */ | |
2547 | if (page) | |
2548 | put_page(page); | |
2549 | if (fpin) | |
2550 | fput(fpin); | |
2551 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
2552 | } |
2553 | EXPORT_SYMBOL(filemap_fault); | |
2554 | ||
82b0f8c3 | 2555 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2556 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 | 2557 | { |
82b0f8c3 | 2558 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2559 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2560 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2561 | unsigned long max_idx; |
070e807c | 2562 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
4101196b | 2563 | struct page *page; |
f1820361 KS |
2564 | |
2565 | rcu_read_lock(); | |
070e807c MW |
2566 | xas_for_each(&xas, page, end_pgoff) { |
2567 | if (xas_retry(&xas, page)) | |
2568 | continue; | |
2569 | if (xa_is_value(page)) | |
2cf938aa | 2570 | goto next; |
f1820361 | 2571 | |
e0975b2a MH |
2572 | /* |
2573 | * Check for a locked page first, as a speculative | |
2574 | * reference may adversely influence page migration. | |
2575 | */ | |
4101196b | 2576 | if (PageLocked(page)) |
e0975b2a | 2577 | goto next; |
4101196b | 2578 | if (!page_cache_get_speculative(page)) |
070e807c | 2579 | goto next; |
f1820361 | 2580 | |
4101196b | 2581 | /* Has the page moved or been split? */ |
070e807c MW |
2582 | if (unlikely(page != xas_reload(&xas))) |
2583 | goto skip; | |
4101196b | 2584 | page = find_subpage(page, xas.xa_index); |
f1820361 KS |
2585 | |
2586 | if (!PageUptodate(page) || | |
2587 | PageReadahead(page) || | |
2588 | PageHWPoison(page)) | |
2589 | goto skip; | |
2590 | if (!trylock_page(page)) | |
2591 | goto skip; | |
2592 | ||
2593 | if (page->mapping != mapping || !PageUptodate(page)) | |
2594 | goto unlock; | |
2595 | ||
9ab2594f MW |
2596 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2597 | if (page->index >= max_idx) | |
f1820361 KS |
2598 | goto unlock; |
2599 | ||
f1820361 KS |
2600 | if (file->f_ra.mmap_miss > 0) |
2601 | file->f_ra.mmap_miss--; | |
7267ec00 | 2602 | |
070e807c | 2603 | vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
82b0f8c3 | 2604 | if (vmf->pte) |
070e807c MW |
2605 | vmf->pte += xas.xa_index - last_pgoff; |
2606 | last_pgoff = xas.xa_index; | |
9d82c694 | 2607 | if (alloc_set_pte(vmf, page)) |
7267ec00 | 2608 | goto unlock; |
f1820361 KS |
2609 | unlock_page(page); |
2610 | goto next; | |
2611 | unlock: | |
2612 | unlock_page(page); | |
2613 | skip: | |
09cbfeaf | 2614 | put_page(page); |
f1820361 | 2615 | next: |
7267ec00 | 2616 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2617 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2618 | break; |
f1820361 KS |
2619 | } |
2620 | rcu_read_unlock(); | |
2621 | } | |
2622 | EXPORT_SYMBOL(filemap_map_pages); | |
2623 | ||
2bcd6454 | 2624 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2625 | { |
2626 | struct page *page = vmf->page; | |
11bac800 | 2627 | struct inode *inode = file_inode(vmf->vma->vm_file); |
2bcd6454 | 2628 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 2629 | |
14da9200 | 2630 | sb_start_pagefault(inode->i_sb); |
11bac800 | 2631 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 JK |
2632 | lock_page(page); |
2633 | if (page->mapping != inode->i_mapping) { | |
2634 | unlock_page(page); | |
2635 | ret = VM_FAULT_NOPAGE; | |
2636 | goto out; | |
2637 | } | |
14da9200 JK |
2638 | /* |
2639 | * We mark the page dirty already here so that when freeze is in | |
2640 | * progress, we are guaranteed that writeback during freezing will | |
2641 | * see the dirty page and writeprotect it again. | |
2642 | */ | |
2643 | set_page_dirty(page); | |
1d1d1a76 | 2644 | wait_for_stable_page(page); |
4fcf1c62 | 2645 | out: |
14da9200 | 2646 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2647 | return ret; |
2648 | } | |
4fcf1c62 | 2649 | |
f0f37e2f | 2650 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2651 | .fault = filemap_fault, |
f1820361 | 2652 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2653 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2654 | }; |
2655 | ||
2656 | /* This is used for a general mmap of a disk file */ | |
2657 | ||
2658 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2659 | { | |
2660 | struct address_space *mapping = file->f_mapping; | |
2661 | ||
2662 | if (!mapping->a_ops->readpage) | |
2663 | return -ENOEXEC; | |
2664 | file_accessed(file); | |
2665 | vma->vm_ops = &generic_file_vm_ops; | |
2666 | return 0; | |
2667 | } | |
1da177e4 LT |
2668 | |
2669 | /* | |
2670 | * This is for filesystems which do not implement ->writepage. | |
2671 | */ | |
2672 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2673 | { | |
2674 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2675 | return -EINVAL; | |
2676 | return generic_file_mmap(file, vma); | |
2677 | } | |
2678 | #else | |
4b96a37d | 2679 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 2680 | { |
4b96a37d | 2681 | return VM_FAULT_SIGBUS; |
45397228 | 2682 | } |
1da177e4 LT |
2683 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2684 | { | |
2685 | return -ENOSYS; | |
2686 | } | |
2687 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2688 | { | |
2689 | return -ENOSYS; | |
2690 | } | |
2691 | #endif /* CONFIG_MMU */ | |
2692 | ||
45397228 | 2693 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
2694 | EXPORT_SYMBOL(generic_file_mmap); |
2695 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2696 | ||
67f9fd91 SL |
2697 | static struct page *wait_on_page_read(struct page *page) |
2698 | { | |
2699 | if (!IS_ERR(page)) { | |
2700 | wait_on_page_locked(page); | |
2701 | if (!PageUptodate(page)) { | |
09cbfeaf | 2702 | put_page(page); |
67f9fd91 SL |
2703 | page = ERR_PTR(-EIO); |
2704 | } | |
2705 | } | |
2706 | return page; | |
2707 | } | |
2708 | ||
32b63529 | 2709 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2710 | pgoff_t index, |
5e5358e7 | 2711 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2712 | void *data, |
2713 | gfp_t gfp) | |
1da177e4 | 2714 | { |
eb2be189 | 2715 | struct page *page; |
1da177e4 LT |
2716 | int err; |
2717 | repeat: | |
2718 | page = find_get_page(mapping, index); | |
2719 | if (!page) { | |
453f85d4 | 2720 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
2721 | if (!page) |
2722 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2723 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2724 | if (unlikely(err)) { |
09cbfeaf | 2725 | put_page(page); |
eb2be189 NP |
2726 | if (err == -EEXIST) |
2727 | goto repeat; | |
22ecdb4f | 2728 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
2729 | return ERR_PTR(err); |
2730 | } | |
32b63529 MG |
2731 | |
2732 | filler: | |
6c45b454 CH |
2733 | if (filler) |
2734 | err = filler(data, page); | |
2735 | else | |
2736 | err = mapping->a_ops->readpage(data, page); | |
2737 | ||
1da177e4 | 2738 | if (err < 0) { |
09cbfeaf | 2739 | put_page(page); |
32b63529 | 2740 | return ERR_PTR(err); |
1da177e4 | 2741 | } |
1da177e4 | 2742 | |
32b63529 MG |
2743 | page = wait_on_page_read(page); |
2744 | if (IS_ERR(page)) | |
2745 | return page; | |
2746 | goto out; | |
2747 | } | |
1da177e4 LT |
2748 | if (PageUptodate(page)) |
2749 | goto out; | |
2750 | ||
ebded027 MG |
2751 | /* |
2752 | * Page is not up to date and may be locked due one of the following | |
2753 | * case a: Page is being filled and the page lock is held | |
2754 | * case b: Read/write error clearing the page uptodate status | |
2755 | * case c: Truncation in progress (page locked) | |
2756 | * case d: Reclaim in progress | |
2757 | * | |
2758 | * Case a, the page will be up to date when the page is unlocked. | |
2759 | * There is no need to serialise on the page lock here as the page | |
2760 | * is pinned so the lock gives no additional protection. Even if the | |
2761 | * the page is truncated, the data is still valid if PageUptodate as | |
2762 | * it's a race vs truncate race. | |
2763 | * Case b, the page will not be up to date | |
2764 | * Case c, the page may be truncated but in itself, the data may still | |
2765 | * be valid after IO completes as it's a read vs truncate race. The | |
2766 | * operation must restart if the page is not uptodate on unlock but | |
2767 | * otherwise serialising on page lock to stabilise the mapping gives | |
2768 | * no additional guarantees to the caller as the page lock is | |
2769 | * released before return. | |
2770 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2771 | * will be a race with remove_mapping that determines if the mapping | |
2772 | * is valid on unlock but otherwise the data is valid and there is | |
2773 | * no need to serialise with page lock. | |
2774 | * | |
2775 | * As the page lock gives no additional guarantee, we optimistically | |
2776 | * wait on the page to be unlocked and check if it's up to date and | |
2777 | * use the page if it is. Otherwise, the page lock is required to | |
2778 | * distinguish between the different cases. The motivation is that we | |
2779 | * avoid spurious serialisations and wakeups when multiple processes | |
2780 | * wait on the same page for IO to complete. | |
2781 | */ | |
2782 | wait_on_page_locked(page); | |
2783 | if (PageUptodate(page)) | |
2784 | goto out; | |
2785 | ||
2786 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2787 | lock_page(page); |
ebded027 MG |
2788 | |
2789 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2790 | if (!page->mapping) { |
2791 | unlock_page(page); | |
09cbfeaf | 2792 | put_page(page); |
32b63529 | 2793 | goto repeat; |
1da177e4 | 2794 | } |
ebded027 MG |
2795 | |
2796 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2797 | if (PageUptodate(page)) { |
2798 | unlock_page(page); | |
2799 | goto out; | |
2800 | } | |
faffdfa0 XT |
2801 | |
2802 | /* | |
2803 | * A previous I/O error may have been due to temporary | |
2804 | * failures. | |
2805 | * Clear page error before actual read, PG_error will be | |
2806 | * set again if read page fails. | |
2807 | */ | |
2808 | ClearPageError(page); | |
32b63529 MG |
2809 | goto filler; |
2810 | ||
c855ff37 | 2811 | out: |
6fe6900e NP |
2812 | mark_page_accessed(page); |
2813 | return page; | |
2814 | } | |
0531b2aa LT |
2815 | |
2816 | /** | |
67f9fd91 | 2817 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2818 | * @mapping: the page's address_space |
2819 | * @index: the page index | |
2820 | * @filler: function to perform the read | |
5e5358e7 | 2821 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2822 | * |
0531b2aa | 2823 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2824 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2825 | * |
2826 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
2827 | * |
2828 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa | 2829 | */ |
67f9fd91 | 2830 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2831 | pgoff_t index, |
5e5358e7 | 2832 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2833 | void *data) |
2834 | { | |
d322a8e5 CH |
2835 | return do_read_cache_page(mapping, index, filler, data, |
2836 | mapping_gfp_mask(mapping)); | |
0531b2aa | 2837 | } |
67f9fd91 | 2838 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2839 | |
2840 | /** | |
2841 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2842 | * @mapping: the page's address_space | |
2843 | * @index: the page index | |
2844 | * @gfp: the page allocator flags to use if allocating | |
2845 | * | |
2846 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2847 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2848 | * |
2849 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
2850 | * |
2851 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa LT |
2852 | */ |
2853 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2854 | pgoff_t index, | |
2855 | gfp_t gfp) | |
2856 | { | |
6c45b454 | 2857 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
2858 | } |
2859 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2860 | ||
9fd91a90 DW |
2861 | /* |
2862 | * Don't operate on ranges the page cache doesn't support, and don't exceed the | |
2863 | * LFS limits. If pos is under the limit it becomes a short access. If it | |
2864 | * exceeds the limit we return -EFBIG. | |
2865 | */ | |
9fd91a90 DW |
2866 | static int generic_write_check_limits(struct file *file, loff_t pos, |
2867 | loff_t *count) | |
2868 | { | |
646955cd AG |
2869 | struct inode *inode = file->f_mapping->host; |
2870 | loff_t max_size = inode->i_sb->s_maxbytes; | |
9fd91a90 DW |
2871 | loff_t limit = rlimit(RLIMIT_FSIZE); |
2872 | ||
2873 | if (limit != RLIM_INFINITY) { | |
2874 | if (pos >= limit) { | |
2875 | send_sig(SIGXFSZ, current, 0); | |
2876 | return -EFBIG; | |
2877 | } | |
2878 | *count = min(*count, limit - pos); | |
2879 | } | |
2880 | ||
646955cd AG |
2881 | if (!(file->f_flags & O_LARGEFILE)) |
2882 | max_size = MAX_NON_LFS; | |
2883 | ||
2884 | if (unlikely(pos >= max_size)) | |
2885 | return -EFBIG; | |
2886 | ||
2887 | *count = min(*count, max_size - pos); | |
2888 | ||
2889 | return 0; | |
9fd91a90 DW |
2890 | } |
2891 | ||
1da177e4 LT |
2892 | /* |
2893 | * Performs necessary checks before doing a write | |
2894 | * | |
485bb99b | 2895 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2896 | * Returns appropriate error code that caller should return or |
2897 | * zero in case that write should be allowed. | |
2898 | */ | |
3309dd04 | 2899 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2900 | { |
3309dd04 | 2901 | struct file *file = iocb->ki_filp; |
1da177e4 | 2902 | struct inode *inode = file->f_mapping->host; |
9fd91a90 DW |
2903 | loff_t count; |
2904 | int ret; | |
1da177e4 | 2905 | |
dc617f29 DW |
2906 | if (IS_SWAPFILE(inode)) |
2907 | return -ETXTBSY; | |
2908 | ||
3309dd04 AV |
2909 | if (!iov_iter_count(from)) |
2910 | return 0; | |
1da177e4 | 2911 | |
0fa6b005 | 2912 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2913 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2914 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2915 | |
6be96d3a GR |
2916 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
2917 | return -EINVAL; | |
2918 | ||
9fd91a90 DW |
2919 | count = iov_iter_count(from); |
2920 | ret = generic_write_check_limits(file, iocb->ki_pos, &count); | |
2921 | if (ret) | |
2922 | return ret; | |
1da177e4 | 2923 | |
9fd91a90 | 2924 | iov_iter_truncate(from, count); |
3309dd04 | 2925 | return iov_iter_count(from); |
1da177e4 LT |
2926 | } |
2927 | EXPORT_SYMBOL(generic_write_checks); | |
2928 | ||
1383a7ed DW |
2929 | /* |
2930 | * Performs necessary checks before doing a clone. | |
2931 | * | |
646955cd | 2932 | * Can adjust amount of bytes to clone via @req_count argument. |
1383a7ed DW |
2933 | * Returns appropriate error code that caller should return or |
2934 | * zero in case the clone should be allowed. | |
2935 | */ | |
2936 | int generic_remap_checks(struct file *file_in, loff_t pos_in, | |
2937 | struct file *file_out, loff_t pos_out, | |
42ec3d4c | 2938 | loff_t *req_count, unsigned int remap_flags) |
1383a7ed DW |
2939 | { |
2940 | struct inode *inode_in = file_in->f_mapping->host; | |
2941 | struct inode *inode_out = file_out->f_mapping->host; | |
2942 | uint64_t count = *req_count; | |
2943 | uint64_t bcount; | |
2944 | loff_t size_in, size_out; | |
2945 | loff_t bs = inode_out->i_sb->s_blocksize; | |
9fd91a90 | 2946 | int ret; |
1383a7ed DW |
2947 | |
2948 | /* The start of both ranges must be aligned to an fs block. */ | |
2949 | if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs)) | |
2950 | return -EINVAL; | |
2951 | ||
2952 | /* Ensure offsets don't wrap. */ | |
2953 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
2954 | return -EINVAL; | |
2955 | ||
2956 | size_in = i_size_read(inode_in); | |
2957 | size_out = i_size_read(inode_out); | |
2958 | ||
2959 | /* Dedupe requires both ranges to be within EOF. */ | |
3d28193e | 2960 | if ((remap_flags & REMAP_FILE_DEDUP) && |
1383a7ed DW |
2961 | (pos_in >= size_in || pos_in + count > size_in || |
2962 | pos_out >= size_out || pos_out + count > size_out)) | |
2963 | return -EINVAL; | |
2964 | ||
2965 | /* Ensure the infile range is within the infile. */ | |
2966 | if (pos_in >= size_in) | |
2967 | return -EINVAL; | |
2968 | count = min(count, size_in - (uint64_t)pos_in); | |
2969 | ||
9fd91a90 DW |
2970 | ret = generic_write_check_limits(file_out, pos_out, &count); |
2971 | if (ret) | |
2972 | return ret; | |
1da177e4 LT |
2973 | |
2974 | /* | |
1383a7ed DW |
2975 | * If the user wanted us to link to the infile's EOF, round up to the |
2976 | * next block boundary for this check. | |
2977 | * | |
2978 | * Otherwise, make sure the count is also block-aligned, having | |
2979 | * already confirmed the starting offsets' block alignment. | |
1da177e4 | 2980 | */ |
1383a7ed DW |
2981 | if (pos_in + count == size_in) { |
2982 | bcount = ALIGN(size_in, bs) - pos_in; | |
2983 | } else { | |
2984 | if (!IS_ALIGNED(count, bs)) | |
eca3654e | 2985 | count = ALIGN_DOWN(count, bs); |
1383a7ed | 2986 | bcount = count; |
1da177e4 LT |
2987 | } |
2988 | ||
1383a7ed DW |
2989 | /* Don't allow overlapped cloning within the same file. */ |
2990 | if (inode_in == inode_out && | |
2991 | pos_out + bcount > pos_in && | |
2992 | pos_out < pos_in + bcount) | |
2993 | return -EINVAL; | |
2994 | ||
1da177e4 | 2995 | /* |
eca3654e DW |
2996 | * We shortened the request but the caller can't deal with that, so |
2997 | * bounce the request back to userspace. | |
1da177e4 | 2998 | */ |
eca3654e | 2999 | if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) |
1383a7ed | 3000 | return -EINVAL; |
1da177e4 | 3001 | |
eca3654e | 3002 | *req_count = count; |
1383a7ed | 3003 | return 0; |
1da177e4 | 3004 | } |
1da177e4 | 3005 | |
a3171351 AG |
3006 | |
3007 | /* | |
3008 | * Performs common checks before doing a file copy/clone | |
3009 | * from @file_in to @file_out. | |
3010 | */ | |
3011 | int generic_file_rw_checks(struct file *file_in, struct file *file_out) | |
3012 | { | |
3013 | struct inode *inode_in = file_inode(file_in); | |
3014 | struct inode *inode_out = file_inode(file_out); | |
3015 | ||
3016 | /* Don't copy dirs, pipes, sockets... */ | |
3017 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) | |
3018 | return -EISDIR; | |
3019 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) | |
3020 | return -EINVAL; | |
3021 | ||
3022 | if (!(file_in->f_mode & FMODE_READ) || | |
3023 | !(file_out->f_mode & FMODE_WRITE) || | |
3024 | (file_out->f_flags & O_APPEND)) | |
3025 | return -EBADF; | |
3026 | ||
3027 | return 0; | |
3028 | } | |
3029 | ||
96e6e8f4 AG |
3030 | /* |
3031 | * Performs necessary checks before doing a file copy | |
3032 | * | |
3033 | * Can adjust amount of bytes to copy via @req_count argument. | |
3034 | * Returns appropriate error code that caller should return or | |
3035 | * zero in case the copy should be allowed. | |
3036 | */ | |
3037 | int generic_copy_file_checks(struct file *file_in, loff_t pos_in, | |
3038 | struct file *file_out, loff_t pos_out, | |
3039 | size_t *req_count, unsigned int flags) | |
3040 | { | |
3041 | struct inode *inode_in = file_inode(file_in); | |
3042 | struct inode *inode_out = file_inode(file_out); | |
3043 | uint64_t count = *req_count; | |
3044 | loff_t size_in; | |
3045 | int ret; | |
3046 | ||
3047 | ret = generic_file_rw_checks(file_in, file_out); | |
3048 | if (ret) | |
3049 | return ret; | |
3050 | ||
3051 | /* Don't touch certain kinds of inodes */ | |
3052 | if (IS_IMMUTABLE(inode_out)) | |
3053 | return -EPERM; | |
3054 | ||
3055 | if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) | |
3056 | return -ETXTBSY; | |
3057 | ||
3058 | /* Ensure offsets don't wrap. */ | |
3059 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
3060 | return -EOVERFLOW; | |
3061 | ||
3062 | /* Shorten the copy to EOF */ | |
3063 | size_in = i_size_read(inode_in); | |
3064 | if (pos_in >= size_in) | |
3065 | count = 0; | |
3066 | else | |
3067 | count = min(count, size_in - (uint64_t)pos_in); | |
3068 | ||
3069 | ret = generic_write_check_limits(file_out, pos_out, &count); | |
3070 | if (ret) | |
3071 | return ret; | |
3072 | ||
3073 | /* Don't allow overlapped copying within the same file. */ | |
3074 | if (inode_in == inode_out && | |
3075 | pos_out + count > pos_in && | |
3076 | pos_out < pos_in + count) | |
3077 | return -EINVAL; | |
3078 | ||
3079 | *req_count = count; | |
3080 | return 0; | |
3081 | } | |
3082 | ||
afddba49 NP |
3083 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3084 | loff_t pos, unsigned len, unsigned flags, | |
3085 | struct page **pagep, void **fsdata) | |
3086 | { | |
3087 | const struct address_space_operations *aops = mapping->a_ops; | |
3088 | ||
4e02ed4b | 3089 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3090 | pagep, fsdata); |
afddba49 NP |
3091 | } |
3092 | EXPORT_SYMBOL(pagecache_write_begin); | |
3093 | ||
3094 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3095 | loff_t pos, unsigned len, unsigned copied, | |
3096 | struct page *page, void *fsdata) | |
3097 | { | |
3098 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3099 | |
4e02ed4b | 3100 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3101 | } |
3102 | EXPORT_SYMBOL(pagecache_write_end); | |
3103 | ||
a92853b6 KK |
3104 | /* |
3105 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3106 | */ | |
3107 | void dio_warn_stale_pagecache(struct file *filp) | |
3108 | { | |
3109 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3110 | char pathname[128]; | |
3111 | struct inode *inode = file_inode(filp); | |
3112 | char *path; | |
3113 | ||
3114 | errseq_set(&inode->i_mapping->wb_err, -EIO); | |
3115 | if (__ratelimit(&_rs)) { | |
3116 | path = file_path(filp, pathname, sizeof(pathname)); | |
3117 | if (IS_ERR(path)) | |
3118 | path = "(unknown)"; | |
3119 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3120 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3121 | current->comm); | |
3122 | } | |
3123 | } | |
3124 | ||
1da177e4 | 3125 | ssize_t |
1af5bb49 | 3126 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3127 | { |
3128 | struct file *file = iocb->ki_filp; | |
3129 | struct address_space *mapping = file->f_mapping; | |
3130 | struct inode *inode = mapping->host; | |
1af5bb49 | 3131 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3132 | ssize_t written; |
a969e903 CH |
3133 | size_t write_len; |
3134 | pgoff_t end; | |
1da177e4 | 3135 | |
0c949334 | 3136 | write_len = iov_iter_count(from); |
09cbfeaf | 3137 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3138 | |
6be96d3a GR |
3139 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3140 | /* If there are pages to writeback, return */ | |
3141 | if (filemap_range_has_page(inode->i_mapping, pos, | |
35f12f0f | 3142 | pos + write_len - 1)) |
6be96d3a GR |
3143 | return -EAGAIN; |
3144 | } else { | |
3145 | written = filemap_write_and_wait_range(mapping, pos, | |
3146 | pos + write_len - 1); | |
3147 | if (written) | |
3148 | goto out; | |
3149 | } | |
a969e903 CH |
3150 | |
3151 | /* | |
3152 | * After a write we want buffered reads to be sure to go to disk to get | |
3153 | * the new data. We invalidate clean cached page from the region we're | |
3154 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3155 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3156 | */ |
55635ba7 | 3157 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3158 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3159 | /* |
3160 | * If a page can not be invalidated, return 0 to fall back | |
3161 | * to buffered write. | |
3162 | */ | |
3163 | if (written) { | |
3164 | if (written == -EBUSY) | |
3165 | return 0; | |
3166 | goto out; | |
a969e903 CH |
3167 | } |
3168 | ||
639a93a5 | 3169 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3170 | |
3171 | /* | |
3172 | * Finally, try again to invalidate clean pages which might have been | |
3173 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3174 | * if the source of the write was an mmap'ed region of the file | |
3175 | * we're writing. Either one is a pretty crazy thing to do, | |
3176 | * so we don't support it 100%. If this invalidation | |
3177 | * fails, tough, the write still worked... | |
332391a9 LC |
3178 | * |
3179 | * Most of the time we do not need this since dio_complete() will do | |
3180 | * the invalidation for us. However there are some file systems that | |
3181 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3182 | * them by removing it completely. |
3183 | * | |
9266a140 KK |
3184 | * Noticeable example is a blkdev_direct_IO(). |
3185 | * | |
80c1fe90 | 3186 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3187 | */ |
9266a140 KK |
3188 | if (written > 0 && mapping->nrpages && |
3189 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3190 | dio_warn_stale_pagecache(file); | |
a969e903 | 3191 | |
1da177e4 | 3192 | if (written > 0) { |
0116651c | 3193 | pos += written; |
639a93a5 | 3194 | write_len -= written; |
0116651c NK |
3195 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3196 | i_size_write(inode, pos); | |
1da177e4 LT |
3197 | mark_inode_dirty(inode); |
3198 | } | |
5cb6c6c7 | 3199 | iocb->ki_pos = pos; |
1da177e4 | 3200 | } |
639a93a5 | 3201 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3202 | out: |
1da177e4 LT |
3203 | return written; |
3204 | } | |
3205 | EXPORT_SYMBOL(generic_file_direct_write); | |
3206 | ||
eb2be189 NP |
3207 | /* |
3208 | * Find or create a page at the given pagecache position. Return the locked | |
3209 | * page. This function is specifically for buffered writes. | |
3210 | */ | |
54566b2c NP |
3211 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3212 | pgoff_t index, unsigned flags) | |
eb2be189 | 3213 | { |
eb2be189 | 3214 | struct page *page; |
bbddabe2 | 3215 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3216 | |
54566b2c | 3217 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3218 | fgp_flags |= FGP_NOFS; |
3219 | ||
3220 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3221 | mapping_gfp_mask(mapping)); |
c585a267 | 3222 | if (page) |
2457aec6 | 3223 | wait_for_stable_page(page); |
eb2be189 | 3224 | |
eb2be189 NP |
3225 | return page; |
3226 | } | |
54566b2c | 3227 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3228 | |
3b93f911 | 3229 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3230 | struct iov_iter *i, loff_t pos) |
3231 | { | |
3232 | struct address_space *mapping = file->f_mapping; | |
3233 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3234 | long status = 0; | |
3235 | ssize_t written = 0; | |
674b892e NP |
3236 | unsigned int flags = 0; |
3237 | ||
afddba49 NP |
3238 | do { |
3239 | struct page *page; | |
afddba49 NP |
3240 | unsigned long offset; /* Offset into pagecache page */ |
3241 | unsigned long bytes; /* Bytes to write to page */ | |
3242 | size_t copied; /* Bytes copied from user */ | |
3243 | void *fsdata; | |
3244 | ||
09cbfeaf KS |
3245 | offset = (pos & (PAGE_SIZE - 1)); |
3246 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3247 | iov_iter_count(i)); |
3248 | ||
3249 | again: | |
00a3d660 LT |
3250 | /* |
3251 | * Bring in the user page that we will copy from _first_. | |
3252 | * Otherwise there's a nasty deadlock on copying from the | |
3253 | * same page as we're writing to, without it being marked | |
3254 | * up-to-date. | |
3255 | * | |
3256 | * Not only is this an optimisation, but it is also required | |
3257 | * to check that the address is actually valid, when atomic | |
3258 | * usercopies are used, below. | |
3259 | */ | |
3260 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3261 | status = -EFAULT; | |
3262 | break; | |
3263 | } | |
3264 | ||
296291cd JK |
3265 | if (fatal_signal_pending(current)) { |
3266 | status = -EINTR; | |
3267 | break; | |
3268 | } | |
3269 | ||
674b892e | 3270 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3271 | &page, &fsdata); |
2457aec6 | 3272 | if (unlikely(status < 0)) |
afddba49 NP |
3273 | break; |
3274 | ||
931e80e4 | 3275 | if (mapping_writably_mapped(mapping)) |
3276 | flush_dcache_page(page); | |
00a3d660 | 3277 | |
afddba49 | 3278 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3279 | flush_dcache_page(page); |
3280 | ||
3281 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3282 | page, fsdata); | |
3283 | if (unlikely(status < 0)) | |
3284 | break; | |
3285 | copied = status; | |
3286 | ||
3287 | cond_resched(); | |
3288 | ||
124d3b70 | 3289 | iov_iter_advance(i, copied); |
afddba49 NP |
3290 | if (unlikely(copied == 0)) { |
3291 | /* | |
3292 | * If we were unable to copy any data at all, we must | |
3293 | * fall back to a single segment length write. | |
3294 | * | |
3295 | * If we didn't fallback here, we could livelock | |
3296 | * because not all segments in the iov can be copied at | |
3297 | * once without a pagefault. | |
3298 | */ | |
09cbfeaf | 3299 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3300 | iov_iter_single_seg_count(i)); |
3301 | goto again; | |
3302 | } | |
afddba49 NP |
3303 | pos += copied; |
3304 | written += copied; | |
3305 | ||
3306 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3307 | } while (iov_iter_count(i)); |
3308 | ||
3309 | return written ? written : status; | |
3310 | } | |
3b93f911 | 3311 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3312 | |
e4dd9de3 | 3313 | /** |
8174202b | 3314 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3315 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3316 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3317 | * |
3318 | * This function does all the work needed for actually writing data to a | |
3319 | * file. It does all basic checks, removes SUID from the file, updates | |
3320 | * modification times and calls proper subroutines depending on whether we | |
3321 | * do direct IO or a standard buffered write. | |
3322 | * | |
3323 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3324 | * object which does not need locking at all. | |
3325 | * | |
3326 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3327 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3328 | * avoid syncing under i_mutex. | |
a862f68a MR |
3329 | * |
3330 | * Return: | |
3331 | * * number of bytes written, even for truncated writes | |
3332 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3333 | */ |
8174202b | 3334 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3335 | { |
3336 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3337 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3338 | struct inode *inode = mapping->host; |
3b93f911 | 3339 | ssize_t written = 0; |
1da177e4 | 3340 | ssize_t err; |
3b93f911 | 3341 | ssize_t status; |
1da177e4 | 3342 | |
1da177e4 | 3343 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3344 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3345 | err = file_remove_privs(file); |
1da177e4 LT |
3346 | if (err) |
3347 | goto out; | |
3348 | ||
c3b2da31 JB |
3349 | err = file_update_time(file); |
3350 | if (err) | |
3351 | goto out; | |
1da177e4 | 3352 | |
2ba48ce5 | 3353 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3354 | loff_t pos, endbyte; |
fb5527e6 | 3355 | |
1af5bb49 | 3356 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3357 | /* |
fbbbad4b MW |
3358 | * If the write stopped short of completing, fall back to |
3359 | * buffered writes. Some filesystems do this for writes to | |
3360 | * holes, for example. For DAX files, a buffered write will | |
3361 | * not succeed (even if it did, DAX does not handle dirty | |
3362 | * page-cache pages correctly). | |
1da177e4 | 3363 | */ |
0b8def9d | 3364 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3365 | goto out; |
3366 | ||
0b8def9d | 3367 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3368 | /* |
3b93f911 | 3369 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3370 | * then we want to return the number of bytes which were |
3371 | * direct-written, or the error code if that was zero. Note | |
3372 | * that this differs from normal direct-io semantics, which | |
3373 | * will return -EFOO even if some bytes were written. | |
3374 | */ | |
60bb4529 | 3375 | if (unlikely(status < 0)) { |
3b93f911 | 3376 | err = status; |
fb5527e6 JM |
3377 | goto out; |
3378 | } | |
fb5527e6 JM |
3379 | /* |
3380 | * We need to ensure that the page cache pages are written to | |
3381 | * disk and invalidated to preserve the expected O_DIRECT | |
3382 | * semantics. | |
3383 | */ | |
3b93f911 | 3384 | endbyte = pos + status - 1; |
0b8def9d | 3385 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3386 | if (err == 0) { |
0b8def9d | 3387 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3388 | written += status; |
fb5527e6 | 3389 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3390 | pos >> PAGE_SHIFT, |
3391 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3392 | } else { |
3393 | /* | |
3394 | * We don't know how much we wrote, so just return | |
3395 | * the number of bytes which were direct-written | |
3396 | */ | |
3397 | } | |
3398 | } else { | |
0b8def9d AV |
3399 | written = generic_perform_write(file, from, iocb->ki_pos); |
3400 | if (likely(written > 0)) | |
3401 | iocb->ki_pos += written; | |
fb5527e6 | 3402 | } |
1da177e4 LT |
3403 | out: |
3404 | current->backing_dev_info = NULL; | |
3405 | return written ? written : err; | |
3406 | } | |
8174202b | 3407 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3408 | |
e4dd9de3 | 3409 | /** |
8174202b | 3410 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3411 | * @iocb: IO state structure |
8174202b | 3412 | * @from: iov_iter with data to write |
e4dd9de3 | 3413 | * |
8174202b | 3414 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3415 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3416 | * and acquires i_mutex as needed. | |
a862f68a MR |
3417 | * Return: |
3418 | * * negative error code if no data has been written at all of | |
3419 | * vfs_fsync_range() failed for a synchronous write | |
3420 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3421 | */ |
8174202b | 3422 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3423 | { |
3424 | struct file *file = iocb->ki_filp; | |
148f948b | 3425 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3426 | ssize_t ret; |
1da177e4 | 3427 | |
5955102c | 3428 | inode_lock(inode); |
3309dd04 AV |
3429 | ret = generic_write_checks(iocb, from); |
3430 | if (ret > 0) | |
5f380c7f | 3431 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3432 | inode_unlock(inode); |
1da177e4 | 3433 | |
e2592217 CH |
3434 | if (ret > 0) |
3435 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3436 | return ret; |
3437 | } | |
8174202b | 3438 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3439 | |
cf9a2ae8 DH |
3440 | /** |
3441 | * try_to_release_page() - release old fs-specific metadata on a page | |
3442 | * | |
3443 | * @page: the page which the kernel is trying to free | |
3444 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3445 | * | |
3446 | * The address_space is to try to release any data against the page | |
a862f68a | 3447 | * (presumably at page->private). |
cf9a2ae8 | 3448 | * |
266cf658 DH |
3449 | * This may also be called if PG_fscache is set on a page, indicating that the |
3450 | * page is known to the local caching routines. | |
3451 | * | |
cf9a2ae8 | 3452 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3453 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3454 | * |
a862f68a | 3455 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3456 | */ |
3457 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3458 | { | |
3459 | struct address_space * const mapping = page->mapping; | |
3460 | ||
3461 | BUG_ON(!PageLocked(page)); | |
3462 | if (PageWriteback(page)) | |
3463 | return 0; | |
3464 | ||
3465 | if (mapping && mapping->a_ops->releasepage) | |
3466 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3467 | return try_to_free_buffers(page); | |
3468 | } | |
3469 | ||
3470 | EXPORT_SYMBOL(try_to_release_page); |