]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 | 13 | #include <linux/compiler.h> |
f9fe48be | 14 | #include <linux/dax.h> |
1da177e4 | 15 | #include <linux/fs.h> |
3f07c014 | 16 | #include <linux/sched/signal.h> |
c22ce143 | 17 | #include <linux/uaccess.h> |
c59ede7b | 18 | #include <linux/capability.h> |
1da177e4 | 19 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 20 | #include <linux/gfp.h> |
1da177e4 LT |
21 | #include <linux/mm.h> |
22 | #include <linux/swap.h> | |
23 | #include <linux/mman.h> | |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/file.h> | |
26 | #include <linux/uio.h> | |
27 | #include <linux/hash.h> | |
28 | #include <linux/writeback.h> | |
53253383 | 29 | #include <linux/backing-dev.h> |
1da177e4 LT |
30 | #include <linux/pagevec.h> |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/security.h> | |
44110fe3 | 33 | #include <linux/cpuset.h> |
00501b53 | 34 | #include <linux/hugetlb.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
c515e1fd | 36 | #include <linux/cleancache.h> |
c7df8ad2 | 37 | #include <linux/shmem_fs.h> |
f1820361 | 38 | #include <linux/rmap.h> |
0f8053a5 NP |
39 | #include "internal.h" |
40 | ||
fe0bfaaf RJ |
41 | #define CREATE_TRACE_POINTS |
42 | #include <trace/events/filemap.h> | |
43 | ||
1da177e4 | 44 | /* |
1da177e4 LT |
45 | * FIXME: remove all knowledge of the buffer layer from the core VM |
46 | */ | |
148f948b | 47 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 48 | |
1da177e4 LT |
49 | #include <asm/mman.h> |
50 | ||
51 | /* | |
52 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
53 | * though. | |
54 | * | |
55 | * Shared mappings now work. 15.8.1995 Bruno. | |
56 | * | |
57 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
58 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
59 | * | |
60 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
61 | */ | |
62 | ||
63 | /* | |
64 | * Lock ordering: | |
65 | * | |
c8c06efa | 66 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 67 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 68 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 69 | * ->i_pages lock |
1da177e4 | 70 | * |
1b1dcc1b | 71 | * ->i_mutex |
c8c06efa | 72 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
73 | * |
74 | * ->mmap_sem | |
c8c06efa | 75 | * ->i_mmap_rwsem |
b8072f09 | 76 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 77 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 LT |
78 | * |
79 | * ->mmap_sem | |
80 | * ->lock_page (access_process_vm) | |
81 | * | |
ccad2365 | 82 | * ->i_mutex (generic_perform_write) |
82591e6e | 83 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 84 | * |
f758eeab | 85 | * bdi->wb.list_lock |
a66979ab | 86 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 87 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 88 | * |
c8c06efa | 89 | * ->i_mmap_rwsem |
1da177e4 LT |
90 | * ->anon_vma.lock (vma_adjust) |
91 | * | |
92 | * ->anon_vma.lock | |
b8072f09 | 93 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 94 | * |
b8072f09 | 95 | * ->page_table_lock or pte_lock |
5d337b91 | 96 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 97 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 98 | * ->i_pages lock (try_to_unmap_one) |
a52633d8 MG |
99 | * ->zone_lru_lock(zone) (follow_page->mark_page_accessed) |
100 | * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page) | |
1da177e4 | 101 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 102 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 103 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 104 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 105 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 106 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 107 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
108 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
109 | * | |
c8c06efa | 110 | * ->i_mmap_rwsem |
9a3c531d | 111 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
112 | */ |
113 | ||
5c024e6a | 114 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
115 | struct page *page, void *shadow) |
116 | { | |
5c024e6a MW |
117 | XA_STATE(xas, &mapping->i_pages, page->index); |
118 | unsigned int nr = 1; | |
c70b647d | 119 | |
5c024e6a MW |
120 | mapping_set_update(&xas, mapping); |
121 | ||
122 | /* hugetlb pages are represented by a single entry in the xarray */ | |
123 | if (!PageHuge(page)) { | |
124 | xas_set_order(&xas, page->index, compound_order(page)); | |
125 | nr = 1U << compound_order(page); | |
126 | } | |
91b0abe3 | 127 | |
83929372 KS |
128 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
129 | VM_BUG_ON_PAGE(PageTail(page), page); | |
130 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 131 | |
5c024e6a MW |
132 | xas_store(&xas, shadow); |
133 | xas_init_marks(&xas); | |
d3798ae8 | 134 | |
2300638b JK |
135 | page->mapping = NULL; |
136 | /* Leave page->index set: truncation lookup relies upon it */ | |
137 | ||
d3798ae8 JW |
138 | if (shadow) { |
139 | mapping->nrexceptional += nr; | |
140 | /* | |
141 | * Make sure the nrexceptional update is committed before | |
142 | * the nrpages update so that final truncate racing | |
143 | * with reclaim does not see both counters 0 at the | |
144 | * same time and miss a shadow entry. | |
145 | */ | |
146 | smp_wmb(); | |
147 | } | |
148 | mapping->nrpages -= nr; | |
91b0abe3 JW |
149 | } |
150 | ||
5ecc4d85 JK |
151 | static void unaccount_page_cache_page(struct address_space *mapping, |
152 | struct page *page) | |
1da177e4 | 153 | { |
5ecc4d85 | 154 | int nr; |
1da177e4 | 155 | |
c515e1fd DM |
156 | /* |
157 | * if we're uptodate, flush out into the cleancache, otherwise | |
158 | * invalidate any existing cleancache entries. We can't leave | |
159 | * stale data around in the cleancache once our page is gone | |
160 | */ | |
161 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
162 | cleancache_put_page(page); | |
163 | else | |
3167760f | 164 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 165 | |
83929372 | 166 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
167 | VM_BUG_ON_PAGE(page_mapped(page), page); |
168 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
169 | int mapcount; | |
170 | ||
171 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
172 | current->comm, page_to_pfn(page)); | |
173 | dump_page(page, "still mapped when deleted"); | |
174 | dump_stack(); | |
175 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
176 | ||
177 | mapcount = page_mapcount(page); | |
178 | if (mapping_exiting(mapping) && | |
179 | page_count(page) >= mapcount + 2) { | |
180 | /* | |
181 | * All vmas have already been torn down, so it's | |
182 | * a good bet that actually the page is unmapped, | |
183 | * and we'd prefer not to leak it: if we're wrong, | |
184 | * some other bad page check should catch it later. | |
185 | */ | |
186 | page_mapcount_reset(page); | |
6d061f9f | 187 | page_ref_sub(page, mapcount); |
06b241f3 HD |
188 | } |
189 | } | |
190 | ||
4165b9b4 | 191 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
192 | if (PageHuge(page)) |
193 | return; | |
09612fa6 | 194 | |
5ecc4d85 JK |
195 | nr = hpage_nr_pages(page); |
196 | ||
197 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); | |
198 | if (PageSwapBacked(page)) { | |
199 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); | |
200 | if (PageTransHuge(page)) | |
201 | __dec_node_page_state(page, NR_SHMEM_THPS); | |
202 | } else { | |
203 | VM_BUG_ON_PAGE(PageTransHuge(page), page); | |
800d8c63 | 204 | } |
5ecc4d85 JK |
205 | |
206 | /* | |
207 | * At this point page must be either written or cleaned by | |
208 | * truncate. Dirty page here signals a bug and loss of | |
209 | * unwritten data. | |
210 | * | |
211 | * This fixes dirty accounting after removing the page entirely | |
212 | * but leaves PageDirty set: it has no effect for truncated | |
213 | * page and anyway will be cleared before returning page into | |
214 | * buddy allocator. | |
215 | */ | |
216 | if (WARN_ON_ONCE(PageDirty(page))) | |
217 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Delete a page from the page cache and free it. Caller has to make | |
222 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 223 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
224 | */ |
225 | void __delete_from_page_cache(struct page *page, void *shadow) | |
226 | { | |
227 | struct address_space *mapping = page->mapping; | |
228 | ||
229 | trace_mm_filemap_delete_from_page_cache(page); | |
230 | ||
231 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 232 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
233 | } |
234 | ||
59c66c5f JK |
235 | static void page_cache_free_page(struct address_space *mapping, |
236 | struct page *page) | |
237 | { | |
238 | void (*freepage)(struct page *); | |
239 | ||
240 | freepage = mapping->a_ops->freepage; | |
241 | if (freepage) | |
242 | freepage(page); | |
243 | ||
244 | if (PageTransHuge(page) && !PageHuge(page)) { | |
245 | page_ref_sub(page, HPAGE_PMD_NR); | |
246 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
247 | } else { | |
248 | put_page(page); | |
249 | } | |
250 | } | |
251 | ||
702cfbf9 MK |
252 | /** |
253 | * delete_from_page_cache - delete page from page cache | |
254 | * @page: the page which the kernel is trying to remove from page cache | |
255 | * | |
256 | * This must be called only on pages that have been verified to be in the page | |
257 | * cache and locked. It will never put the page into the free list, the caller | |
258 | * has a reference on the page. | |
259 | */ | |
260 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 261 | { |
83929372 | 262 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 263 | unsigned long flags; |
1da177e4 | 264 | |
cd7619d6 | 265 | BUG_ON(!PageLocked(page)); |
b93b0163 | 266 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 267 | __delete_from_page_cache(page, NULL); |
b93b0163 | 268 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 269 | |
59c66c5f | 270 | page_cache_free_page(mapping, page); |
97cecb5a MK |
271 | } |
272 | EXPORT_SYMBOL(delete_from_page_cache); | |
273 | ||
aa65c29c JK |
274 | /* |
275 | * page_cache_tree_delete_batch - delete several pages from page cache | |
276 | * @mapping: the mapping to which pages belong | |
277 | * @pvec: pagevec with pages to delete | |
278 | * | |
b93b0163 MW |
279 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
280 | * from the mapping. The function expects @pvec to be sorted by page index. | |
281 | * It tolerates holes in @pvec (mapping entries at those indices are not | |
aa65c29c | 282 | * modified). The function expects only THP head pages to be present in the |
b93b0163 MW |
283 | * @pvec and takes care to delete all corresponding tail pages from the |
284 | * mapping as well. | |
aa65c29c | 285 | * |
b93b0163 | 286 | * The function expects the i_pages lock to be held. |
aa65c29c JK |
287 | */ |
288 | static void | |
289 | page_cache_tree_delete_batch(struct address_space *mapping, | |
290 | struct pagevec *pvec) | |
291 | { | |
292 | struct radix_tree_iter iter; | |
293 | void **slot; | |
294 | int total_pages = 0; | |
295 | int i = 0, tail_pages = 0; | |
296 | struct page *page; | |
297 | pgoff_t start; | |
298 | ||
299 | start = pvec->pages[0]->index; | |
b93b0163 | 300 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
aa65c29c JK |
301 | if (i >= pagevec_count(pvec) && !tail_pages) |
302 | break; | |
303 | page = radix_tree_deref_slot_protected(slot, | |
b93b0163 | 304 | &mapping->i_pages.xa_lock); |
3159f943 | 305 | if (xa_is_value(page)) |
aa65c29c JK |
306 | continue; |
307 | if (!tail_pages) { | |
308 | /* | |
309 | * Some page got inserted in our range? Skip it. We | |
310 | * have our pages locked so they are protected from | |
311 | * being removed. | |
312 | */ | |
313 | if (page != pvec->pages[i]) | |
314 | continue; | |
315 | WARN_ON_ONCE(!PageLocked(page)); | |
316 | if (PageTransHuge(page) && !PageHuge(page)) | |
317 | tail_pages = HPAGE_PMD_NR - 1; | |
318 | page->mapping = NULL; | |
319 | /* | |
320 | * Leave page->index set: truncation lookup relies | |
321 | * upon it | |
322 | */ | |
323 | i++; | |
324 | } else { | |
325 | tail_pages--; | |
326 | } | |
b93b0163 MW |
327 | radix_tree_clear_tags(&mapping->i_pages, iter.node, slot); |
328 | __radix_tree_replace(&mapping->i_pages, iter.node, slot, NULL, | |
c7df8ad2 | 329 | workingset_lookup_update(mapping)); |
aa65c29c JK |
330 | total_pages++; |
331 | } | |
332 | mapping->nrpages -= total_pages; | |
333 | } | |
334 | ||
335 | void delete_from_page_cache_batch(struct address_space *mapping, | |
336 | struct pagevec *pvec) | |
337 | { | |
338 | int i; | |
339 | unsigned long flags; | |
340 | ||
341 | if (!pagevec_count(pvec)) | |
342 | return; | |
343 | ||
b93b0163 | 344 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
345 | for (i = 0; i < pagevec_count(pvec); i++) { |
346 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
347 | ||
348 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
349 | } | |
350 | page_cache_tree_delete_batch(mapping, pvec); | |
b93b0163 | 351 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
352 | |
353 | for (i = 0; i < pagevec_count(pvec); i++) | |
354 | page_cache_free_page(mapping, pvec->pages[i]); | |
355 | } | |
356 | ||
d72d9e2a | 357 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
358 | { |
359 | int ret = 0; | |
360 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
361 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
362 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 363 | ret = -ENOSPC; |
7fcbbaf1 JA |
364 | if (test_bit(AS_EIO, &mapping->flags) && |
365 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
366 | ret = -EIO; |
367 | return ret; | |
368 | } | |
d72d9e2a | 369 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 370 | |
76341cab JL |
371 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
372 | { | |
373 | /* Check for outstanding write errors */ | |
374 | if (test_bit(AS_EIO, &mapping->flags)) | |
375 | return -EIO; | |
376 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
377 | return -ENOSPC; | |
378 | return 0; | |
379 | } | |
380 | ||
1da177e4 | 381 | /** |
485bb99b | 382 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
383 | * @mapping: address space structure to write |
384 | * @start: offset in bytes where the range starts | |
469eb4d0 | 385 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 386 | * @sync_mode: enable synchronous operation |
1da177e4 | 387 | * |
485bb99b RD |
388 | * Start writeback against all of a mapping's dirty pages that lie |
389 | * within the byte offsets <start, end> inclusive. | |
390 | * | |
1da177e4 | 391 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 392 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
393 | * these two operations is that if a dirty page/buffer is encountered, it must |
394 | * be waited upon, and not just skipped over. | |
395 | */ | |
ebcf28e1 AM |
396 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
397 | loff_t end, int sync_mode) | |
1da177e4 LT |
398 | { |
399 | int ret; | |
400 | struct writeback_control wbc = { | |
401 | .sync_mode = sync_mode, | |
05fe478d | 402 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
403 | .range_start = start, |
404 | .range_end = end, | |
1da177e4 LT |
405 | }; |
406 | ||
407 | if (!mapping_cap_writeback_dirty(mapping)) | |
408 | return 0; | |
409 | ||
b16b1deb | 410 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 411 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 412 | wbc_detach_inode(&wbc); |
1da177e4 LT |
413 | return ret; |
414 | } | |
415 | ||
416 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
417 | int sync_mode) | |
418 | { | |
111ebb6e | 419 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
420 | } |
421 | ||
422 | int filemap_fdatawrite(struct address_space *mapping) | |
423 | { | |
424 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
425 | } | |
426 | EXPORT_SYMBOL(filemap_fdatawrite); | |
427 | ||
f4c0a0fd | 428 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 429 | loff_t end) |
1da177e4 LT |
430 | { |
431 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
432 | } | |
f4c0a0fd | 433 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 434 | |
485bb99b RD |
435 | /** |
436 | * filemap_flush - mostly a non-blocking flush | |
437 | * @mapping: target address_space | |
438 | * | |
1da177e4 LT |
439 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
440 | * purposes - I/O may not be started against all dirty pages. | |
441 | */ | |
442 | int filemap_flush(struct address_space *mapping) | |
443 | { | |
444 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
445 | } | |
446 | EXPORT_SYMBOL(filemap_flush); | |
447 | ||
7fc9e472 GR |
448 | /** |
449 | * filemap_range_has_page - check if a page exists in range. | |
450 | * @mapping: address space within which to check | |
451 | * @start_byte: offset in bytes where the range starts | |
452 | * @end_byte: offset in bytes where the range ends (inclusive) | |
453 | * | |
454 | * Find at least one page in the range supplied, usually used to check if | |
455 | * direct writing in this range will trigger a writeback. | |
456 | */ | |
457 | bool filemap_range_has_page(struct address_space *mapping, | |
458 | loff_t start_byte, loff_t end_byte) | |
459 | { | |
460 | pgoff_t index = start_byte >> PAGE_SHIFT; | |
461 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
f7b68046 | 462 | struct page *page; |
7fc9e472 GR |
463 | |
464 | if (end_byte < start_byte) | |
465 | return false; | |
466 | ||
467 | if (mapping->nrpages == 0) | |
468 | return false; | |
469 | ||
f7b68046 | 470 | if (!find_get_pages_range(mapping, &index, end, 1, &page)) |
7fc9e472 | 471 | return false; |
f7b68046 JK |
472 | put_page(page); |
473 | return true; | |
7fc9e472 GR |
474 | } |
475 | EXPORT_SYMBOL(filemap_range_has_page); | |
476 | ||
5e8fcc1a | 477 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 478 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 479 | { |
09cbfeaf KS |
480 | pgoff_t index = start_byte >> PAGE_SHIFT; |
481 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
482 | struct pagevec pvec; |
483 | int nr_pages; | |
1da177e4 | 484 | |
94004ed7 | 485 | if (end_byte < start_byte) |
5e8fcc1a | 486 | return; |
1da177e4 | 487 | |
86679820 | 488 | pagevec_init(&pvec); |
312e9d2f | 489 | while (index <= end) { |
1da177e4 LT |
490 | unsigned i; |
491 | ||
312e9d2f | 492 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 493 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
494 | if (!nr_pages) |
495 | break; | |
496 | ||
1da177e4 LT |
497 | for (i = 0; i < nr_pages; i++) { |
498 | struct page *page = pvec.pages[i]; | |
499 | ||
1da177e4 | 500 | wait_on_page_writeback(page); |
5e8fcc1a | 501 | ClearPageError(page); |
1da177e4 LT |
502 | } |
503 | pagevec_release(&pvec); | |
504 | cond_resched(); | |
505 | } | |
aa750fd7 JN |
506 | } |
507 | ||
508 | /** | |
509 | * filemap_fdatawait_range - wait for writeback to complete | |
510 | * @mapping: address space structure to wait for | |
511 | * @start_byte: offset in bytes where the range starts | |
512 | * @end_byte: offset in bytes where the range ends (inclusive) | |
513 | * | |
514 | * Walk the list of under-writeback pages of the given address space | |
515 | * in the given range and wait for all of them. Check error status of | |
516 | * the address space and return it. | |
517 | * | |
518 | * Since the error status of the address space is cleared by this function, | |
519 | * callers are responsible for checking the return value and handling and/or | |
520 | * reporting the error. | |
521 | */ | |
522 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
523 | loff_t end_byte) | |
524 | { | |
5e8fcc1a JL |
525 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
526 | return filemap_check_errors(mapping); | |
1da177e4 | 527 | } |
d3bccb6f JK |
528 | EXPORT_SYMBOL(filemap_fdatawait_range); |
529 | ||
a823e458 JL |
530 | /** |
531 | * file_fdatawait_range - wait for writeback to complete | |
532 | * @file: file pointing to address space structure to wait for | |
533 | * @start_byte: offset in bytes where the range starts | |
534 | * @end_byte: offset in bytes where the range ends (inclusive) | |
535 | * | |
536 | * Walk the list of under-writeback pages of the address space that file | |
537 | * refers to, in the given range and wait for all of them. Check error | |
538 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
539 | * | |
540 | * Since the error status of the file is advanced by this function, | |
541 | * callers are responsible for checking the return value and handling and/or | |
542 | * reporting the error. | |
543 | */ | |
544 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
545 | { | |
546 | struct address_space *mapping = file->f_mapping; | |
547 | ||
548 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
549 | return file_check_and_advance_wb_err(file); | |
550 | } | |
551 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 552 | |
aa750fd7 JN |
553 | /** |
554 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
555 | * @mapping: address space structure to wait for | |
556 | * | |
557 | * Walk the list of under-writeback pages of the given address space | |
558 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
559 | * does not clear error status of the address space. | |
560 | * | |
561 | * Use this function if callers don't handle errors themselves. Expected | |
562 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
563 | * fsfreeze(8) | |
564 | */ | |
76341cab | 565 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 566 | { |
ffb959bb | 567 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 568 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 569 | } |
76341cab | 570 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 571 | |
9326c9b2 | 572 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 573 | { |
9326c9b2 JL |
574 | return (!dax_mapping(mapping) && mapping->nrpages) || |
575 | (dax_mapping(mapping) && mapping->nrexceptional); | |
1da177e4 | 576 | } |
1da177e4 LT |
577 | |
578 | int filemap_write_and_wait(struct address_space *mapping) | |
579 | { | |
28fd1298 | 580 | int err = 0; |
1da177e4 | 581 | |
9326c9b2 | 582 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
583 | err = filemap_fdatawrite(mapping); |
584 | /* | |
585 | * Even if the above returned error, the pages may be | |
586 | * written partially (e.g. -ENOSPC), so we wait for it. | |
587 | * But the -EIO is special case, it may indicate the worst | |
588 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
589 | */ | |
590 | if (err != -EIO) { | |
591 | int err2 = filemap_fdatawait(mapping); | |
592 | if (!err) | |
593 | err = err2; | |
cbeaf951 JL |
594 | } else { |
595 | /* Clear any previously stored errors */ | |
596 | filemap_check_errors(mapping); | |
28fd1298 | 597 | } |
865ffef3 DM |
598 | } else { |
599 | err = filemap_check_errors(mapping); | |
1da177e4 | 600 | } |
28fd1298 | 601 | return err; |
1da177e4 | 602 | } |
28fd1298 | 603 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 604 | |
485bb99b RD |
605 | /** |
606 | * filemap_write_and_wait_range - write out & wait on a file range | |
607 | * @mapping: the address_space for the pages | |
608 | * @lstart: offset in bytes where the range starts | |
609 | * @lend: offset in bytes where the range ends (inclusive) | |
610 | * | |
469eb4d0 AM |
611 | * Write out and wait upon file offsets lstart->lend, inclusive. |
612 | * | |
0e056eb5 | 613 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 AM |
614 | * that this function can be used to write to the very end-of-file (end = -1). |
615 | */ | |
1da177e4 LT |
616 | int filemap_write_and_wait_range(struct address_space *mapping, |
617 | loff_t lstart, loff_t lend) | |
618 | { | |
28fd1298 | 619 | int err = 0; |
1da177e4 | 620 | |
9326c9b2 | 621 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
622 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
623 | WB_SYNC_ALL); | |
624 | /* See comment of filemap_write_and_wait() */ | |
625 | if (err != -EIO) { | |
94004ed7 CH |
626 | int err2 = filemap_fdatawait_range(mapping, |
627 | lstart, lend); | |
28fd1298 OH |
628 | if (!err) |
629 | err = err2; | |
cbeaf951 JL |
630 | } else { |
631 | /* Clear any previously stored errors */ | |
632 | filemap_check_errors(mapping); | |
28fd1298 | 633 | } |
865ffef3 DM |
634 | } else { |
635 | err = filemap_check_errors(mapping); | |
1da177e4 | 636 | } |
28fd1298 | 637 | return err; |
1da177e4 | 638 | } |
f6995585 | 639 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 640 | |
5660e13d JL |
641 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
642 | { | |
3acdfd28 | 643 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
644 | |
645 | trace_filemap_set_wb_err(mapping, eseq); | |
646 | } | |
647 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
648 | ||
649 | /** | |
650 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
651 | * and advance wb_err to current one | |
652 | * @file: struct file on which the error is being reported | |
653 | * | |
654 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
655 | * want to report any writeback errors that occurred since the last fsync (or | |
656 | * since the file was opened if there haven't been any). | |
657 | * | |
658 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
659 | * then just quickly return 0. The file is all caught up. | |
660 | * | |
661 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
662 | * it and try to swap it into place. If it works, or another task beat us | |
663 | * to it with the new value, then update the f_wb_err and return the error | |
664 | * portion. The error at this point must be reported via proper channels | |
665 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
666 | * | |
667 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
668 | * value is protected by the f_lock since we must ensure that it reflects | |
669 | * the latest value swapped in for this file descriptor. | |
670 | */ | |
671 | int file_check_and_advance_wb_err(struct file *file) | |
672 | { | |
673 | int err = 0; | |
674 | errseq_t old = READ_ONCE(file->f_wb_err); | |
675 | struct address_space *mapping = file->f_mapping; | |
676 | ||
677 | /* Locklessly handle the common case where nothing has changed */ | |
678 | if (errseq_check(&mapping->wb_err, old)) { | |
679 | /* Something changed, must use slow path */ | |
680 | spin_lock(&file->f_lock); | |
681 | old = file->f_wb_err; | |
682 | err = errseq_check_and_advance(&mapping->wb_err, | |
683 | &file->f_wb_err); | |
684 | trace_file_check_and_advance_wb_err(file, old); | |
685 | spin_unlock(&file->f_lock); | |
686 | } | |
f4e222c5 JL |
687 | |
688 | /* | |
689 | * We're mostly using this function as a drop in replacement for | |
690 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
691 | * that the legacy code would have had on these flags. | |
692 | */ | |
693 | clear_bit(AS_EIO, &mapping->flags); | |
694 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
695 | return err; |
696 | } | |
697 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
698 | ||
699 | /** | |
700 | * file_write_and_wait_range - write out & wait on a file range | |
701 | * @file: file pointing to address_space with pages | |
702 | * @lstart: offset in bytes where the range starts | |
703 | * @lend: offset in bytes where the range ends (inclusive) | |
704 | * | |
705 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
706 | * | |
707 | * Note that @lend is inclusive (describes the last byte to be written) so | |
708 | * that this function can be used to write to the very end-of-file (end = -1). | |
709 | * | |
710 | * After writing out and waiting on the data, we check and advance the | |
711 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
712 | */ | |
713 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
714 | { | |
715 | int err = 0, err2; | |
716 | struct address_space *mapping = file->f_mapping; | |
717 | ||
9326c9b2 | 718 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
719 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
720 | WB_SYNC_ALL); | |
721 | /* See comment of filemap_write_and_wait() */ | |
722 | if (err != -EIO) | |
723 | __filemap_fdatawait_range(mapping, lstart, lend); | |
724 | } | |
725 | err2 = file_check_and_advance_wb_err(file); | |
726 | if (!err) | |
727 | err = err2; | |
728 | return err; | |
729 | } | |
730 | EXPORT_SYMBOL(file_write_and_wait_range); | |
731 | ||
ef6a3c63 MS |
732 | /** |
733 | * replace_page_cache_page - replace a pagecache page with a new one | |
734 | * @old: page to be replaced | |
735 | * @new: page to replace with | |
736 | * @gfp_mask: allocation mode | |
737 | * | |
738 | * This function replaces a page in the pagecache with a new one. On | |
739 | * success it acquires the pagecache reference for the new page and | |
740 | * drops it for the old page. Both the old and new pages must be | |
741 | * locked. This function does not add the new page to the LRU, the | |
742 | * caller must do that. | |
743 | * | |
74d60958 | 744 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 MS |
745 | */ |
746 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
747 | { | |
74d60958 MW |
748 | struct address_space *mapping = old->mapping; |
749 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
750 | pgoff_t offset = old->index; | |
751 | XA_STATE(xas, &mapping->i_pages, offset); | |
752 | unsigned long flags; | |
ef6a3c63 | 753 | |
309381fe SL |
754 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
755 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
756 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 757 | |
74d60958 MW |
758 | get_page(new); |
759 | new->mapping = mapping; | |
760 | new->index = offset; | |
ef6a3c63 | 761 | |
74d60958 MW |
762 | xas_lock_irqsave(&xas, flags); |
763 | xas_store(&xas, new); | |
ef6a3c63 | 764 | |
74d60958 MW |
765 | old->mapping = NULL; |
766 | /* hugetlb pages do not participate in page cache accounting. */ | |
767 | if (!PageHuge(old)) | |
768 | __dec_node_page_state(new, NR_FILE_PAGES); | |
769 | if (!PageHuge(new)) | |
770 | __inc_node_page_state(new, NR_FILE_PAGES); | |
771 | if (PageSwapBacked(old)) | |
772 | __dec_node_page_state(new, NR_SHMEM); | |
773 | if (PageSwapBacked(new)) | |
774 | __inc_node_page_state(new, NR_SHMEM); | |
775 | xas_unlock_irqrestore(&xas, flags); | |
776 | mem_cgroup_migrate(old, new); | |
777 | if (freepage) | |
778 | freepage(old); | |
779 | put_page(old); | |
ef6a3c63 | 780 | |
74d60958 | 781 | return 0; |
ef6a3c63 MS |
782 | } |
783 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
784 | ||
a528910e JW |
785 | static int __add_to_page_cache_locked(struct page *page, |
786 | struct address_space *mapping, | |
787 | pgoff_t offset, gfp_t gfp_mask, | |
788 | void **shadowp) | |
1da177e4 | 789 | { |
74d60958 | 790 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 JW |
791 | int huge = PageHuge(page); |
792 | struct mem_cgroup *memcg; | |
e286781d | 793 | int error; |
74d60958 | 794 | void *old; |
e286781d | 795 | |
309381fe SL |
796 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
797 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 798 | mapping_set_update(&xas, mapping); |
e286781d | 799 | |
00501b53 JW |
800 | if (!huge) { |
801 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 802 | gfp_mask, &memcg, false); |
00501b53 JW |
803 | if (error) |
804 | return error; | |
805 | } | |
1da177e4 | 806 | |
09cbfeaf | 807 | get_page(page); |
66a0c8ee KS |
808 | page->mapping = mapping; |
809 | page->index = offset; | |
810 | ||
74d60958 MW |
811 | do { |
812 | xas_lock_irq(&xas); | |
813 | old = xas_load(&xas); | |
814 | if (old && !xa_is_value(old)) | |
815 | xas_set_err(&xas, -EEXIST); | |
816 | xas_store(&xas, page); | |
817 | if (xas_error(&xas)) | |
818 | goto unlock; | |
819 | ||
820 | if (xa_is_value(old)) { | |
821 | mapping->nrexceptional--; | |
822 | if (shadowp) | |
823 | *shadowp = old; | |
824 | } | |
825 | mapping->nrpages++; | |
826 | ||
827 | /* hugetlb pages do not participate in page cache accounting */ | |
828 | if (!huge) | |
829 | __inc_node_page_state(page, NR_FILE_PAGES); | |
830 | unlock: | |
831 | xas_unlock_irq(&xas); | |
832 | } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK)); | |
833 | ||
834 | if (xas_error(&xas)) | |
835 | goto error; | |
4165b9b4 | 836 | |
00501b53 | 837 | if (!huge) |
f627c2f5 | 838 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
839 | trace_mm_filemap_add_to_page_cache(page); |
840 | return 0; | |
74d60958 | 841 | error: |
66a0c8ee KS |
842 | page->mapping = NULL; |
843 | /* Leave page->index set: truncation relies upon it */ | |
00501b53 | 844 | if (!huge) |
f627c2f5 | 845 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 846 | put_page(page); |
74d60958 | 847 | return xas_error(&xas); |
1da177e4 | 848 | } |
a528910e JW |
849 | |
850 | /** | |
851 | * add_to_page_cache_locked - add a locked page to the pagecache | |
852 | * @page: page to add | |
853 | * @mapping: the page's address_space | |
854 | * @offset: page index | |
855 | * @gfp_mask: page allocation mode | |
856 | * | |
857 | * This function is used to add a page to the pagecache. It must be locked. | |
858 | * This function does not add the page to the LRU. The caller must do that. | |
859 | */ | |
860 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
861 | pgoff_t offset, gfp_t gfp_mask) | |
862 | { | |
863 | return __add_to_page_cache_locked(page, mapping, offset, | |
864 | gfp_mask, NULL); | |
865 | } | |
e286781d | 866 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
867 | |
868 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 869 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 870 | { |
a528910e | 871 | void *shadow = NULL; |
4f98a2fe RR |
872 | int ret; |
873 | ||
48c935ad | 874 | __SetPageLocked(page); |
a528910e JW |
875 | ret = __add_to_page_cache_locked(page, mapping, offset, |
876 | gfp_mask, &shadow); | |
877 | if (unlikely(ret)) | |
48c935ad | 878 | __ClearPageLocked(page); |
a528910e JW |
879 | else { |
880 | /* | |
881 | * The page might have been evicted from cache only | |
882 | * recently, in which case it should be activated like | |
883 | * any other repeatedly accessed page. | |
f0281a00 RR |
884 | * The exception is pages getting rewritten; evicting other |
885 | * data from the working set, only to cache data that will | |
886 | * get overwritten with something else, is a waste of memory. | |
a528910e | 887 | */ |
f0281a00 RR |
888 | if (!(gfp_mask & __GFP_WRITE) && |
889 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
890 | SetPageActive(page); |
891 | workingset_activation(page); | |
892 | } else | |
893 | ClearPageActive(page); | |
894 | lru_cache_add(page); | |
895 | } | |
1da177e4 LT |
896 | return ret; |
897 | } | |
18bc0bbd | 898 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 899 | |
44110fe3 | 900 | #ifdef CONFIG_NUMA |
2ae88149 | 901 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 902 | { |
c0ff7453 MX |
903 | int n; |
904 | struct page *page; | |
905 | ||
44110fe3 | 906 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
907 | unsigned int cpuset_mems_cookie; |
908 | do { | |
d26914d1 | 909 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 910 | n = cpuset_mem_spread_node(); |
96db800f | 911 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 912 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 913 | |
c0ff7453 | 914 | return page; |
44110fe3 | 915 | } |
2ae88149 | 916 | return alloc_pages(gfp, 0); |
44110fe3 | 917 | } |
2ae88149 | 918 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
919 | #endif |
920 | ||
1da177e4 LT |
921 | /* |
922 | * In order to wait for pages to become available there must be | |
923 | * waitqueues associated with pages. By using a hash table of | |
924 | * waitqueues where the bucket discipline is to maintain all | |
925 | * waiters on the same queue and wake all when any of the pages | |
926 | * become available, and for the woken contexts to check to be | |
927 | * sure the appropriate page became available, this saves space | |
928 | * at a cost of "thundering herd" phenomena during rare hash | |
929 | * collisions. | |
930 | */ | |
62906027 NP |
931 | #define PAGE_WAIT_TABLE_BITS 8 |
932 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
933 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
934 | ||
935 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 936 | { |
62906027 | 937 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 938 | } |
1da177e4 | 939 | |
62906027 | 940 | void __init pagecache_init(void) |
1da177e4 | 941 | { |
62906027 | 942 | int i; |
1da177e4 | 943 | |
62906027 NP |
944 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
945 | init_waitqueue_head(&page_wait_table[i]); | |
946 | ||
947 | page_writeback_init(); | |
1da177e4 | 948 | } |
1da177e4 | 949 | |
3510ca20 | 950 | /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ |
62906027 NP |
951 | struct wait_page_key { |
952 | struct page *page; | |
953 | int bit_nr; | |
954 | int page_match; | |
955 | }; | |
956 | ||
957 | struct wait_page_queue { | |
958 | struct page *page; | |
959 | int bit_nr; | |
ac6424b9 | 960 | wait_queue_entry_t wait; |
62906027 NP |
961 | }; |
962 | ||
ac6424b9 | 963 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 964 | { |
62906027 NP |
965 | struct wait_page_key *key = arg; |
966 | struct wait_page_queue *wait_page | |
967 | = container_of(wait, struct wait_page_queue, wait); | |
968 | ||
969 | if (wait_page->page != key->page) | |
970 | return 0; | |
971 | key->page_match = 1; | |
f62e00cc | 972 | |
62906027 NP |
973 | if (wait_page->bit_nr != key->bit_nr) |
974 | return 0; | |
3510ca20 LT |
975 | |
976 | /* Stop walking if it's locked */ | |
62906027 | 977 | if (test_bit(key->bit_nr, &key->page->flags)) |
3510ca20 | 978 | return -1; |
f62e00cc | 979 | |
62906027 | 980 | return autoremove_wake_function(wait, mode, sync, key); |
f62e00cc KM |
981 | } |
982 | ||
74d81bfa | 983 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 984 | { |
62906027 NP |
985 | wait_queue_head_t *q = page_waitqueue(page); |
986 | struct wait_page_key key; | |
987 | unsigned long flags; | |
11a19c7b | 988 | wait_queue_entry_t bookmark; |
cbbce822 | 989 | |
62906027 NP |
990 | key.page = page; |
991 | key.bit_nr = bit_nr; | |
992 | key.page_match = 0; | |
993 | ||
11a19c7b TC |
994 | bookmark.flags = 0; |
995 | bookmark.private = NULL; | |
996 | bookmark.func = NULL; | |
997 | INIT_LIST_HEAD(&bookmark.entry); | |
998 | ||
62906027 | 999 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1000 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1001 | ||
1002 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1003 | /* | |
1004 | * Take a breather from holding the lock, | |
1005 | * allow pages that finish wake up asynchronously | |
1006 | * to acquire the lock and remove themselves | |
1007 | * from wait queue | |
1008 | */ | |
1009 | spin_unlock_irqrestore(&q->lock, flags); | |
1010 | cpu_relax(); | |
1011 | spin_lock_irqsave(&q->lock, flags); | |
1012 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1013 | } | |
1014 | ||
62906027 NP |
1015 | /* |
1016 | * It is possible for other pages to have collided on the waitqueue | |
1017 | * hash, so in that case check for a page match. That prevents a long- | |
1018 | * term waiter | |
1019 | * | |
1020 | * It is still possible to miss a case here, when we woke page waiters | |
1021 | * and removed them from the waitqueue, but there are still other | |
1022 | * page waiters. | |
1023 | */ | |
1024 | if (!waitqueue_active(q) || !key.page_match) { | |
1025 | ClearPageWaiters(page); | |
1026 | /* | |
1027 | * It's possible to miss clearing Waiters here, when we woke | |
1028 | * our page waiters, but the hashed waitqueue has waiters for | |
1029 | * other pages on it. | |
1030 | * | |
1031 | * That's okay, it's a rare case. The next waker will clear it. | |
1032 | */ | |
1033 | } | |
1034 | spin_unlock_irqrestore(&q->lock, flags); | |
1035 | } | |
74d81bfa NP |
1036 | |
1037 | static void wake_up_page(struct page *page, int bit) | |
1038 | { | |
1039 | if (!PageWaiters(page)) | |
1040 | return; | |
1041 | wake_up_page_bit(page, bit); | |
1042 | } | |
62906027 NP |
1043 | |
1044 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, | |
1045 | struct page *page, int bit_nr, int state, bool lock) | |
1046 | { | |
1047 | struct wait_page_queue wait_page; | |
ac6424b9 | 1048 | wait_queue_entry_t *wait = &wait_page.wait; |
62906027 NP |
1049 | int ret = 0; |
1050 | ||
1051 | init_wait(wait); | |
3510ca20 | 1052 | wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0; |
62906027 NP |
1053 | wait->func = wake_page_function; |
1054 | wait_page.page = page; | |
1055 | wait_page.bit_nr = bit_nr; | |
1056 | ||
1057 | for (;;) { | |
1058 | spin_lock_irq(&q->lock); | |
1059 | ||
2055da97 | 1060 | if (likely(list_empty(&wait->entry))) { |
3510ca20 | 1061 | __add_wait_queue_entry_tail(q, wait); |
62906027 NP |
1062 | SetPageWaiters(page); |
1063 | } | |
1064 | ||
1065 | set_current_state(state); | |
1066 | ||
1067 | spin_unlock_irq(&q->lock); | |
1068 | ||
1069 | if (likely(test_bit(bit_nr, &page->flags))) { | |
1070 | io_schedule(); | |
62906027 NP |
1071 | } |
1072 | ||
1073 | if (lock) { | |
1074 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) | |
1075 | break; | |
1076 | } else { | |
1077 | if (!test_bit(bit_nr, &page->flags)) | |
1078 | break; | |
1079 | } | |
a8b169af LT |
1080 | |
1081 | if (unlikely(signal_pending_state(state, current))) { | |
1082 | ret = -EINTR; | |
1083 | break; | |
1084 | } | |
62906027 NP |
1085 | } |
1086 | ||
1087 | finish_wait(q, wait); | |
1088 | ||
1089 | /* | |
1090 | * A signal could leave PageWaiters set. Clearing it here if | |
1091 | * !waitqueue_active would be possible (by open-coding finish_wait), | |
1092 | * but still fail to catch it in the case of wait hash collision. We | |
1093 | * already can fail to clear wait hash collision cases, so don't | |
1094 | * bother with signals either. | |
1095 | */ | |
1096 | ||
1097 | return ret; | |
1098 | } | |
1099 | ||
1100 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1101 | { | |
1102 | wait_queue_head_t *q = page_waitqueue(page); | |
1103 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false); | |
1104 | } | |
1105 | EXPORT_SYMBOL(wait_on_page_bit); | |
1106 | ||
1107 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1108 | { | |
1109 | wait_queue_head_t *q = page_waitqueue(page); | |
1110 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false); | |
cbbce822 | 1111 | } |
4343d008 | 1112 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1113 | |
385e1ca5 DH |
1114 | /** |
1115 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1116 | * @page: Page defining the wait queue of interest |
1117 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1118 | * |
1119 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1120 | */ | |
ac6424b9 | 1121 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1122 | { |
1123 | wait_queue_head_t *q = page_waitqueue(page); | |
1124 | unsigned long flags; | |
1125 | ||
1126 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1127 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1128 | SetPageWaiters(page); |
385e1ca5 DH |
1129 | spin_unlock_irqrestore(&q->lock, flags); |
1130 | } | |
1131 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1132 | ||
b91e1302 LT |
1133 | #ifndef clear_bit_unlock_is_negative_byte |
1134 | ||
1135 | /* | |
1136 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1137 | * | |
1138 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1139 | * test the sign bit at the same time. But if the architecture does | |
1140 | * not support that special operation, we just do this all by hand | |
1141 | * instead. | |
1142 | * | |
1143 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
1144 | * being cleared, but a memory barrier should be unneccssary since it is | |
1145 | * in the same byte as PG_locked. | |
1146 | */ | |
1147 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1148 | { | |
1149 | clear_bit_unlock(nr, mem); | |
1150 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1151 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1152 | } |
1153 | ||
1154 | #endif | |
1155 | ||
1da177e4 | 1156 | /** |
485bb99b | 1157 | * unlock_page - unlock a locked page |
1da177e4 LT |
1158 | * @page: the page |
1159 | * | |
1160 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1161 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1162 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1163 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1164 | * | |
b91e1302 LT |
1165 | * Note that this depends on PG_waiters being the sign bit in the byte |
1166 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1167 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1168 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1169 | * test the sign bit). | |
1da177e4 | 1170 | */ |
920c7a5d | 1171 | void unlock_page(struct page *page) |
1da177e4 | 1172 | { |
b91e1302 | 1173 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1174 | page = compound_head(page); |
309381fe | 1175 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1176 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1177 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1178 | } |
1179 | EXPORT_SYMBOL(unlock_page); | |
1180 | ||
485bb99b RD |
1181 | /** |
1182 | * end_page_writeback - end writeback against a page | |
1183 | * @page: the page | |
1da177e4 LT |
1184 | */ |
1185 | void end_page_writeback(struct page *page) | |
1186 | { | |
888cf2db MG |
1187 | /* |
1188 | * TestClearPageReclaim could be used here but it is an atomic | |
1189 | * operation and overkill in this particular case. Failing to | |
1190 | * shuffle a page marked for immediate reclaim is too mild to | |
1191 | * justify taking an atomic operation penalty at the end of | |
1192 | * ever page writeback. | |
1193 | */ | |
1194 | if (PageReclaim(page)) { | |
1195 | ClearPageReclaim(page); | |
ac6aadb2 | 1196 | rotate_reclaimable_page(page); |
888cf2db | 1197 | } |
ac6aadb2 MS |
1198 | |
1199 | if (!test_clear_page_writeback(page)) | |
1200 | BUG(); | |
1201 | ||
4e857c58 | 1202 | smp_mb__after_atomic(); |
1da177e4 LT |
1203 | wake_up_page(page, PG_writeback); |
1204 | } | |
1205 | EXPORT_SYMBOL(end_page_writeback); | |
1206 | ||
57d99845 MW |
1207 | /* |
1208 | * After completing I/O on a page, call this routine to update the page | |
1209 | * flags appropriately | |
1210 | */ | |
c11f0c0b | 1211 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1212 | { |
c11f0c0b | 1213 | if (!is_write) { |
57d99845 MW |
1214 | if (!err) { |
1215 | SetPageUptodate(page); | |
1216 | } else { | |
1217 | ClearPageUptodate(page); | |
1218 | SetPageError(page); | |
1219 | } | |
1220 | unlock_page(page); | |
abf54548 | 1221 | } else { |
57d99845 | 1222 | if (err) { |
dd8416c4 MK |
1223 | struct address_space *mapping; |
1224 | ||
57d99845 | 1225 | SetPageError(page); |
dd8416c4 MK |
1226 | mapping = page_mapping(page); |
1227 | if (mapping) | |
1228 | mapping_set_error(mapping, err); | |
57d99845 MW |
1229 | } |
1230 | end_page_writeback(page); | |
1231 | } | |
1232 | } | |
1233 | EXPORT_SYMBOL_GPL(page_endio); | |
1234 | ||
485bb99b RD |
1235 | /** |
1236 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1237 | * @__page: the page to lock |
1da177e4 | 1238 | */ |
62906027 | 1239 | void __lock_page(struct page *__page) |
1da177e4 | 1240 | { |
62906027 NP |
1241 | struct page *page = compound_head(__page); |
1242 | wait_queue_head_t *q = page_waitqueue(page); | |
1243 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true); | |
1da177e4 LT |
1244 | } |
1245 | EXPORT_SYMBOL(__lock_page); | |
1246 | ||
62906027 | 1247 | int __lock_page_killable(struct page *__page) |
2687a356 | 1248 | { |
62906027 NP |
1249 | struct page *page = compound_head(__page); |
1250 | wait_queue_head_t *q = page_waitqueue(page); | |
1251 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true); | |
2687a356 | 1252 | } |
18bc0bbd | 1253 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1254 | |
9a95f3cf PC |
1255 | /* |
1256 | * Return values: | |
1257 | * 1 - page is locked; mmap_sem is still held. | |
1258 | * 0 - page is not locked. | |
1259 | * mmap_sem has been released (up_read()), unless flags had both | |
1260 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
1261 | * which case mmap_sem is still held. | |
1262 | * | |
1263 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
1264 | * with the page locked and the mmap_sem unperturbed. | |
1265 | */ | |
d065bd81 ML |
1266 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1267 | unsigned int flags) | |
1268 | { | |
37b23e05 KM |
1269 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
1270 | /* | |
1271 | * CAUTION! In this case, mmap_sem is not released | |
1272 | * even though return 0. | |
1273 | */ | |
1274 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1275 | return 0; | |
1276 | ||
1277 | up_read(&mm->mmap_sem); | |
1278 | if (flags & FAULT_FLAG_KILLABLE) | |
1279 | wait_on_page_locked_killable(page); | |
1280 | else | |
318b275f | 1281 | wait_on_page_locked(page); |
d065bd81 | 1282 | return 0; |
37b23e05 KM |
1283 | } else { |
1284 | if (flags & FAULT_FLAG_KILLABLE) { | |
1285 | int ret; | |
1286 | ||
1287 | ret = __lock_page_killable(page); | |
1288 | if (ret) { | |
1289 | up_read(&mm->mmap_sem); | |
1290 | return 0; | |
1291 | } | |
1292 | } else | |
1293 | __lock_page(page); | |
1294 | return 1; | |
d065bd81 ML |
1295 | } |
1296 | } | |
1297 | ||
e7b563bb | 1298 | /** |
0d3f9296 MW |
1299 | * page_cache_next_miss() - Find the next gap in the page cache. |
1300 | * @mapping: Mapping. | |
1301 | * @index: Index. | |
1302 | * @max_scan: Maximum range to search. | |
e7b563bb | 1303 | * |
0d3f9296 MW |
1304 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1305 | * gap with the lowest index. | |
e7b563bb | 1306 | * |
0d3f9296 MW |
1307 | * This function may be called under the rcu_read_lock. However, this will |
1308 | * not atomically search a snapshot of the cache at a single point in time. | |
1309 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1310 | * created at index 10, page_cache_next_miss covering both indices may | |
1311 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1312 | * |
0d3f9296 MW |
1313 | * Return: The index of the gap if found, otherwise an index outside the |
1314 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1315 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1316 | */ |
0d3f9296 | 1317 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1318 | pgoff_t index, unsigned long max_scan) |
1319 | { | |
0d3f9296 | 1320 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1321 | |
0d3f9296 MW |
1322 | while (max_scan--) { |
1323 | void *entry = xas_next(&xas); | |
1324 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1325 | break; |
0d3f9296 | 1326 | if (xas.xa_index == 0) |
e7b563bb JW |
1327 | break; |
1328 | } | |
1329 | ||
0d3f9296 | 1330 | return xas.xa_index; |
e7b563bb | 1331 | } |
0d3f9296 | 1332 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1333 | |
1334 | /** | |
0d3f9296 MW |
1335 | * page_cache_prev_miss() - Find the next gap in the page cache. |
1336 | * @mapping: Mapping. | |
1337 | * @index: Index. | |
1338 | * @max_scan: Maximum range to search. | |
e7b563bb | 1339 | * |
0d3f9296 MW |
1340 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1341 | * gap with the highest index. | |
e7b563bb | 1342 | * |
0d3f9296 MW |
1343 | * This function may be called under the rcu_read_lock. However, this will |
1344 | * not atomically search a snapshot of the cache at a single point in time. | |
1345 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1346 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1347 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1348 | * |
0d3f9296 MW |
1349 | * Return: The index of the gap if found, otherwise an index outside the |
1350 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1351 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1352 | */ |
0d3f9296 | 1353 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1354 | pgoff_t index, unsigned long max_scan) |
1355 | { | |
0d3f9296 | 1356 | XA_STATE(xas, &mapping->i_pages, index); |
0cd6144a | 1357 | |
0d3f9296 MW |
1358 | while (max_scan--) { |
1359 | void *entry = xas_prev(&xas); | |
1360 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1361 | break; |
0d3f9296 | 1362 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1363 | break; |
1364 | } | |
1365 | ||
0d3f9296 | 1366 | return xas.xa_index; |
e7b563bb | 1367 | } |
0d3f9296 | 1368 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1369 | |
485bb99b | 1370 | /** |
0cd6144a | 1371 | * find_get_entry - find and get a page cache entry |
485bb99b | 1372 | * @mapping: the address_space to search |
0cd6144a JW |
1373 | * @offset: the page cache index |
1374 | * | |
1375 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1376 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1377 | * |
139b6a6f JW |
1378 | * If the slot holds a shadow entry of a previously evicted page, or a |
1379 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1380 | * |
1381 | * Otherwise, %NULL is returned. | |
1da177e4 | 1382 | */ |
0cd6144a | 1383 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1384 | { |
4c7472c0 | 1385 | XA_STATE(xas, &mapping->i_pages, offset); |
83929372 | 1386 | struct page *head, *page; |
1da177e4 | 1387 | |
a60637c8 NP |
1388 | rcu_read_lock(); |
1389 | repeat: | |
4c7472c0 MW |
1390 | xas_reset(&xas); |
1391 | page = xas_load(&xas); | |
1392 | if (xas_retry(&xas, page)) | |
1393 | goto repeat; | |
1394 | /* | |
1395 | * A shadow entry of a recently evicted page, or a swap entry from | |
1396 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1397 | */ | |
1398 | if (!page || xa_is_value(page)) | |
1399 | goto out; | |
83929372 | 1400 | |
4c7472c0 MW |
1401 | head = compound_head(page); |
1402 | if (!page_cache_get_speculative(head)) | |
1403 | goto repeat; | |
83929372 | 1404 | |
4c7472c0 MW |
1405 | /* The page was split under us? */ |
1406 | if (compound_head(page) != head) { | |
1407 | put_page(head); | |
1408 | goto repeat; | |
1409 | } | |
a60637c8 | 1410 | |
4c7472c0 MW |
1411 | /* |
1412 | * Has the page moved? | |
1413 | * This is part of the lockless pagecache protocol. See | |
1414 | * include/linux/pagemap.h for details. | |
1415 | */ | |
1416 | if (unlikely(page != xas_reload(&xas))) { | |
1417 | put_page(head); | |
1418 | goto repeat; | |
a60637c8 | 1419 | } |
27d20fdd | 1420 | out: |
a60637c8 NP |
1421 | rcu_read_unlock(); |
1422 | ||
1da177e4 LT |
1423 | return page; |
1424 | } | |
0cd6144a | 1425 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1426 | |
0cd6144a JW |
1427 | /** |
1428 | * find_lock_entry - locate, pin and lock a page cache entry | |
1429 | * @mapping: the address_space to search | |
1430 | * @offset: the page cache index | |
1431 | * | |
1432 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1433 | * page cache page, it is returned locked and with an increased | |
1434 | * refcount. | |
1435 | * | |
139b6a6f JW |
1436 | * If the slot holds a shadow entry of a previously evicted page, or a |
1437 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1438 | * |
1439 | * Otherwise, %NULL is returned. | |
1440 | * | |
1441 | * find_lock_entry() may sleep. | |
1442 | */ | |
1443 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1444 | { |
1445 | struct page *page; | |
1446 | ||
1da177e4 | 1447 | repeat: |
0cd6144a | 1448 | page = find_get_entry(mapping, offset); |
4c7472c0 | 1449 | if (page && !xa_is_value(page)) { |
a60637c8 NP |
1450 | lock_page(page); |
1451 | /* Has the page been truncated? */ | |
83929372 | 1452 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1453 | unlock_page(page); |
09cbfeaf | 1454 | put_page(page); |
a60637c8 | 1455 | goto repeat; |
1da177e4 | 1456 | } |
83929372 | 1457 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1458 | } |
1da177e4 LT |
1459 | return page; |
1460 | } | |
0cd6144a JW |
1461 | EXPORT_SYMBOL(find_lock_entry); |
1462 | ||
1463 | /** | |
2457aec6 | 1464 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1465 | * @mapping: the address_space to search |
1466 | * @offset: the page index | |
2457aec6 | 1467 | * @fgp_flags: PCG flags |
45f87de5 | 1468 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1469 | * |
2457aec6 | 1470 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1471 | * |
75325189 | 1472 | * PCG flags modify how the page is returned. |
0cd6144a | 1473 | * |
0e056eb5 MCC |
1474 | * @fgp_flags can be: |
1475 | * | |
1476 | * - FGP_ACCESSED: the page will be marked accessed | |
1477 | * - FGP_LOCK: Page is return locked | |
1478 | * - FGP_CREAT: If page is not present then a new page is allocated using | |
1479 | * @gfp_mask and added to the page cache and the VM's LRU | |
1480 | * list. The page is returned locked and with an increased | |
1481 | * refcount. Otherwise, NULL is returned. | |
1da177e4 | 1482 | * |
2457aec6 MG |
1483 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1484 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1485 | * |
2457aec6 | 1486 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1487 | */ |
2457aec6 | 1488 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1489 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1490 | { |
eb2be189 | 1491 | struct page *page; |
2457aec6 | 1492 | |
1da177e4 | 1493 | repeat: |
2457aec6 | 1494 | page = find_get_entry(mapping, offset); |
3159f943 | 1495 | if (xa_is_value(page)) |
2457aec6 MG |
1496 | page = NULL; |
1497 | if (!page) | |
1498 | goto no_page; | |
1499 | ||
1500 | if (fgp_flags & FGP_LOCK) { | |
1501 | if (fgp_flags & FGP_NOWAIT) { | |
1502 | if (!trylock_page(page)) { | |
09cbfeaf | 1503 | put_page(page); |
2457aec6 MG |
1504 | return NULL; |
1505 | } | |
1506 | } else { | |
1507 | lock_page(page); | |
1508 | } | |
1509 | ||
1510 | /* Has the page been truncated? */ | |
1511 | if (unlikely(page->mapping != mapping)) { | |
1512 | unlock_page(page); | |
09cbfeaf | 1513 | put_page(page); |
2457aec6 MG |
1514 | goto repeat; |
1515 | } | |
1516 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1517 | } | |
1518 | ||
1519 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1520 | mark_page_accessed(page); | |
1521 | ||
1522 | no_page: | |
1523 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1524 | int err; | |
1525 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1526 | gfp_mask |= __GFP_WRITE; |
1527 | if (fgp_flags & FGP_NOFS) | |
1528 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1529 | |
45f87de5 | 1530 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1531 | if (!page) |
1532 | return NULL; | |
2457aec6 MG |
1533 | |
1534 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1535 | fgp_flags |= FGP_LOCK; | |
1536 | ||
eb39d618 | 1537 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1538 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1539 | __SetPageReferenced(page); |
2457aec6 | 1540 | |
abc1be13 | 1541 | err = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
eb2be189 | 1542 | if (unlikely(err)) { |
09cbfeaf | 1543 | put_page(page); |
eb2be189 NP |
1544 | page = NULL; |
1545 | if (err == -EEXIST) | |
1546 | goto repeat; | |
1da177e4 | 1547 | } |
1da177e4 | 1548 | } |
2457aec6 | 1549 | |
1da177e4 LT |
1550 | return page; |
1551 | } | |
2457aec6 | 1552 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1553 | |
0cd6144a JW |
1554 | /** |
1555 | * find_get_entries - gang pagecache lookup | |
1556 | * @mapping: The address_space to search | |
1557 | * @start: The starting page cache index | |
1558 | * @nr_entries: The maximum number of entries | |
1559 | * @entries: Where the resulting entries are placed | |
1560 | * @indices: The cache indices corresponding to the entries in @entries | |
1561 | * | |
1562 | * find_get_entries() will search for and return a group of up to | |
1563 | * @nr_entries entries in the mapping. The entries are placed at | |
1564 | * @entries. find_get_entries() takes a reference against any actual | |
1565 | * pages it returns. | |
1566 | * | |
1567 | * The search returns a group of mapping-contiguous page cache entries | |
1568 | * with ascending indexes. There may be holes in the indices due to | |
1569 | * not-present pages. | |
1570 | * | |
139b6a6f JW |
1571 | * Any shadow entries of evicted pages, or swap entries from |
1572 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1573 | * |
1574 | * find_get_entries() returns the number of pages and shadow entries | |
1575 | * which were found. | |
1576 | */ | |
1577 | unsigned find_get_entries(struct address_space *mapping, | |
1578 | pgoff_t start, unsigned int nr_entries, | |
1579 | struct page **entries, pgoff_t *indices) | |
1580 | { | |
1581 | void **slot; | |
1582 | unsigned int ret = 0; | |
1583 | struct radix_tree_iter iter; | |
1584 | ||
1585 | if (!nr_entries) | |
1586 | return 0; | |
1587 | ||
1588 | rcu_read_lock(); | |
b93b0163 | 1589 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
83929372 | 1590 | struct page *head, *page; |
0cd6144a JW |
1591 | repeat: |
1592 | page = radix_tree_deref_slot(slot); | |
1593 | if (unlikely(!page)) | |
1594 | continue; | |
1595 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1596 | if (radix_tree_deref_retry(page)) { |
1597 | slot = radix_tree_iter_retry(&iter); | |
1598 | continue; | |
1599 | } | |
0cd6144a | 1600 | /* |
f9fe48be RZ |
1601 | * A shadow entry of a recently evicted page, a swap |
1602 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1603 | * without attempting to raise page count. | |
0cd6144a JW |
1604 | */ |
1605 | goto export; | |
1606 | } | |
83929372 KS |
1607 | |
1608 | head = compound_head(page); | |
1609 | if (!page_cache_get_speculative(head)) | |
1610 | goto repeat; | |
1611 | ||
1612 | /* The page was split under us? */ | |
1613 | if (compound_head(page) != head) { | |
1614 | put_page(head); | |
0cd6144a | 1615 | goto repeat; |
83929372 | 1616 | } |
0cd6144a JW |
1617 | |
1618 | /* Has the page moved? */ | |
1619 | if (unlikely(page != *slot)) { | |
83929372 | 1620 | put_page(head); |
0cd6144a JW |
1621 | goto repeat; |
1622 | } | |
1623 | export: | |
1624 | indices[ret] = iter.index; | |
1625 | entries[ret] = page; | |
1626 | if (++ret == nr_entries) | |
1627 | break; | |
1628 | } | |
1629 | rcu_read_unlock(); | |
1630 | return ret; | |
1631 | } | |
1632 | ||
1da177e4 | 1633 | /** |
b947cee4 | 1634 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1635 | * @mapping: The address_space to search |
1636 | * @start: The starting page index | |
b947cee4 | 1637 | * @end: The final page index (inclusive) |
1da177e4 LT |
1638 | * @nr_pages: The maximum number of pages |
1639 | * @pages: Where the resulting pages are placed | |
1640 | * | |
b947cee4 JK |
1641 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1642 | * pages in the mapping starting at index @start and up to index @end | |
1643 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1644 | * a reference against the returned pages. | |
1da177e4 LT |
1645 | * |
1646 | * The search returns a group of mapping-contiguous pages with ascending | |
1647 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1648 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1649 | * |
b947cee4 JK |
1650 | * find_get_pages_range() returns the number of pages which were found. If this |
1651 | * number is smaller than @nr_pages, the end of specified range has been | |
1652 | * reached. | |
1da177e4 | 1653 | */ |
b947cee4 JK |
1654 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1655 | pgoff_t end, unsigned int nr_pages, | |
1656 | struct page **pages) | |
1da177e4 | 1657 | { |
0fc9d104 KK |
1658 | struct radix_tree_iter iter; |
1659 | void **slot; | |
1660 | unsigned ret = 0; | |
1661 | ||
1662 | if (unlikely(!nr_pages)) | |
1663 | return 0; | |
a60637c8 NP |
1664 | |
1665 | rcu_read_lock(); | |
b93b0163 | 1666 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, *start) { |
83929372 | 1667 | struct page *head, *page; |
b947cee4 JK |
1668 | |
1669 | if (iter.index > end) | |
1670 | break; | |
a60637c8 | 1671 | repeat: |
0fc9d104 | 1672 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1673 | if (unlikely(!page)) |
1674 | continue; | |
9d8aa4ea | 1675 | |
a2c16d6c | 1676 | if (radix_tree_exception(page)) { |
8079b1c8 | 1677 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1678 | slot = radix_tree_iter_retry(&iter); |
1679 | continue; | |
8079b1c8 | 1680 | } |
a2c16d6c | 1681 | /* |
139b6a6f JW |
1682 | * A shadow entry of a recently evicted page, |
1683 | * or a swap entry from shmem/tmpfs. Skip | |
1684 | * over it. | |
a2c16d6c | 1685 | */ |
8079b1c8 | 1686 | continue; |
27d20fdd | 1687 | } |
a60637c8 | 1688 | |
83929372 KS |
1689 | head = compound_head(page); |
1690 | if (!page_cache_get_speculative(head)) | |
1691 | goto repeat; | |
1692 | ||
1693 | /* The page was split under us? */ | |
1694 | if (compound_head(page) != head) { | |
1695 | put_page(head); | |
a60637c8 | 1696 | goto repeat; |
83929372 | 1697 | } |
a60637c8 NP |
1698 | |
1699 | /* Has the page moved? */ | |
0fc9d104 | 1700 | if (unlikely(page != *slot)) { |
83929372 | 1701 | put_page(head); |
a60637c8 NP |
1702 | goto repeat; |
1703 | } | |
1da177e4 | 1704 | |
a60637c8 | 1705 | pages[ret] = page; |
b947cee4 JK |
1706 | if (++ret == nr_pages) { |
1707 | *start = pages[ret - 1]->index + 1; | |
1708 | goto out; | |
1709 | } | |
a60637c8 | 1710 | } |
5b280c0c | 1711 | |
b947cee4 JK |
1712 | /* |
1713 | * We come here when there is no page beyond @end. We take care to not | |
1714 | * overflow the index @start as it confuses some of the callers. This | |
1715 | * breaks the iteration when there is page at index -1 but that is | |
1716 | * already broken anyway. | |
1717 | */ | |
1718 | if (end == (pgoff_t)-1) | |
1719 | *start = (pgoff_t)-1; | |
1720 | else | |
1721 | *start = end + 1; | |
1722 | out: | |
a60637c8 | 1723 | rcu_read_unlock(); |
d72dc8a2 | 1724 | |
1da177e4 LT |
1725 | return ret; |
1726 | } | |
1727 | ||
ebf43500 JA |
1728 | /** |
1729 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1730 | * @mapping: The address_space to search | |
1731 | * @index: The starting page index | |
1732 | * @nr_pages: The maximum number of pages | |
1733 | * @pages: Where the resulting pages are placed | |
1734 | * | |
1735 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1736 | * that the returned number of pages are guaranteed to be contiguous. | |
1737 | * | |
1738 | * find_get_pages_contig() returns the number of pages which were found. | |
1739 | */ | |
1740 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1741 | unsigned int nr_pages, struct page **pages) | |
1742 | { | |
0fc9d104 KK |
1743 | struct radix_tree_iter iter; |
1744 | void **slot; | |
1745 | unsigned int ret = 0; | |
1746 | ||
1747 | if (unlikely(!nr_pages)) | |
1748 | return 0; | |
a60637c8 NP |
1749 | |
1750 | rcu_read_lock(); | |
b93b0163 | 1751 | radix_tree_for_each_contig(slot, &mapping->i_pages, &iter, index) { |
83929372 | 1752 | struct page *head, *page; |
a60637c8 | 1753 | repeat: |
0fc9d104 KK |
1754 | page = radix_tree_deref_slot(slot); |
1755 | /* The hole, there no reason to continue */ | |
a60637c8 | 1756 | if (unlikely(!page)) |
0fc9d104 | 1757 | break; |
9d8aa4ea | 1758 | |
a2c16d6c | 1759 | if (radix_tree_exception(page)) { |
8079b1c8 | 1760 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1761 | slot = radix_tree_iter_retry(&iter); |
1762 | continue; | |
8079b1c8 | 1763 | } |
a2c16d6c | 1764 | /* |
139b6a6f JW |
1765 | * A shadow entry of a recently evicted page, |
1766 | * or a swap entry from shmem/tmpfs. Stop | |
1767 | * looking for contiguous pages. | |
a2c16d6c | 1768 | */ |
8079b1c8 | 1769 | break; |
a2c16d6c | 1770 | } |
ebf43500 | 1771 | |
83929372 KS |
1772 | head = compound_head(page); |
1773 | if (!page_cache_get_speculative(head)) | |
1774 | goto repeat; | |
1775 | ||
1776 | /* The page was split under us? */ | |
1777 | if (compound_head(page) != head) { | |
1778 | put_page(head); | |
a60637c8 | 1779 | goto repeat; |
83929372 | 1780 | } |
a60637c8 NP |
1781 | |
1782 | /* Has the page moved? */ | |
0fc9d104 | 1783 | if (unlikely(page != *slot)) { |
83929372 | 1784 | put_page(head); |
a60637c8 NP |
1785 | goto repeat; |
1786 | } | |
1787 | ||
9cbb4cb2 NP |
1788 | /* |
1789 | * must check mapping and index after taking the ref. | |
1790 | * otherwise we can get both false positives and false | |
1791 | * negatives, which is just confusing to the caller. | |
1792 | */ | |
83929372 | 1793 | if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { |
09cbfeaf | 1794 | put_page(page); |
9cbb4cb2 NP |
1795 | break; |
1796 | } | |
1797 | ||
a60637c8 | 1798 | pages[ret] = page; |
0fc9d104 KK |
1799 | if (++ret == nr_pages) |
1800 | break; | |
ebf43500 | 1801 | } |
a60637c8 NP |
1802 | rcu_read_unlock(); |
1803 | return ret; | |
ebf43500 | 1804 | } |
ef71c15c | 1805 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1806 | |
485bb99b | 1807 | /** |
72b045ae | 1808 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
1809 | * @mapping: the address_space to search |
1810 | * @index: the starting page index | |
72b045ae | 1811 | * @end: The final page index (inclusive) |
485bb99b RD |
1812 | * @tag: the tag index |
1813 | * @nr_pages: the maximum number of pages | |
1814 | * @pages: where the resulting pages are placed | |
1815 | * | |
1da177e4 | 1816 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1817 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 | 1818 | */ |
72b045ae JK |
1819 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
1820 | pgoff_t end, int tag, unsigned int nr_pages, | |
1821 | struct page **pages) | |
1da177e4 | 1822 | { |
0fc9d104 KK |
1823 | struct radix_tree_iter iter; |
1824 | void **slot; | |
1825 | unsigned ret = 0; | |
1826 | ||
1827 | if (unlikely(!nr_pages)) | |
1828 | return 0; | |
a60637c8 NP |
1829 | |
1830 | rcu_read_lock(); | |
b93b0163 | 1831 | radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, *index, tag) { |
83929372 | 1832 | struct page *head, *page; |
72b045ae JK |
1833 | |
1834 | if (iter.index > end) | |
1835 | break; | |
a60637c8 | 1836 | repeat: |
0fc9d104 | 1837 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1838 | if (unlikely(!page)) |
1839 | continue; | |
9d8aa4ea | 1840 | |
a2c16d6c | 1841 | if (radix_tree_exception(page)) { |
8079b1c8 | 1842 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1843 | slot = radix_tree_iter_retry(&iter); |
1844 | continue; | |
8079b1c8 | 1845 | } |
a2c16d6c | 1846 | /* |
139b6a6f JW |
1847 | * A shadow entry of a recently evicted page. |
1848 | * | |
1849 | * Those entries should never be tagged, but | |
1850 | * this tree walk is lockless and the tags are | |
1851 | * looked up in bulk, one radix tree node at a | |
1852 | * time, so there is a sizable window for page | |
1853 | * reclaim to evict a page we saw tagged. | |
1854 | * | |
1855 | * Skip over it. | |
a2c16d6c | 1856 | */ |
139b6a6f | 1857 | continue; |
a2c16d6c | 1858 | } |
a60637c8 | 1859 | |
83929372 KS |
1860 | head = compound_head(page); |
1861 | if (!page_cache_get_speculative(head)) | |
a60637c8 NP |
1862 | goto repeat; |
1863 | ||
83929372 KS |
1864 | /* The page was split under us? */ |
1865 | if (compound_head(page) != head) { | |
1866 | put_page(head); | |
1867 | goto repeat; | |
1868 | } | |
1869 | ||
a60637c8 | 1870 | /* Has the page moved? */ |
0fc9d104 | 1871 | if (unlikely(page != *slot)) { |
83929372 | 1872 | put_page(head); |
a60637c8 NP |
1873 | goto repeat; |
1874 | } | |
1875 | ||
1876 | pages[ret] = page; | |
72b045ae JK |
1877 | if (++ret == nr_pages) { |
1878 | *index = pages[ret - 1]->index + 1; | |
1879 | goto out; | |
1880 | } | |
a60637c8 | 1881 | } |
5b280c0c | 1882 | |
72b045ae JK |
1883 | /* |
1884 | * We come here when we got at @end. We take care to not overflow the | |
1885 | * index @index as it confuses some of the callers. This breaks the | |
1886 | * iteration when there is page at index -1 but that is already broken | |
1887 | * anyway. | |
1888 | */ | |
1889 | if (end == (pgoff_t)-1) | |
1890 | *index = (pgoff_t)-1; | |
1891 | else | |
1892 | *index = end + 1; | |
1893 | out: | |
a60637c8 | 1894 | rcu_read_unlock(); |
1da177e4 | 1895 | |
1da177e4 LT |
1896 | return ret; |
1897 | } | |
72b045ae | 1898 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 1899 | |
7e7f7749 RZ |
1900 | /** |
1901 | * find_get_entries_tag - find and return entries that match @tag | |
1902 | * @mapping: the address_space to search | |
1903 | * @start: the starting page cache index | |
1904 | * @tag: the tag index | |
1905 | * @nr_entries: the maximum number of entries | |
1906 | * @entries: where the resulting entries are placed | |
1907 | * @indices: the cache indices corresponding to the entries in @entries | |
1908 | * | |
1909 | * Like find_get_entries, except we only return entries which are tagged with | |
1910 | * @tag. | |
1911 | */ | |
1912 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1913 | int tag, unsigned int nr_entries, | |
1914 | struct page **entries, pgoff_t *indices) | |
1915 | { | |
1916 | void **slot; | |
1917 | unsigned int ret = 0; | |
1918 | struct radix_tree_iter iter; | |
1919 | ||
1920 | if (!nr_entries) | |
1921 | return 0; | |
1922 | ||
1923 | rcu_read_lock(); | |
b93b0163 | 1924 | radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, start, tag) { |
83929372 | 1925 | struct page *head, *page; |
7e7f7749 RZ |
1926 | repeat: |
1927 | page = radix_tree_deref_slot(slot); | |
1928 | if (unlikely(!page)) | |
1929 | continue; | |
1930 | if (radix_tree_exception(page)) { | |
1931 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1932 | slot = radix_tree_iter_retry(&iter); |
1933 | continue; | |
7e7f7749 RZ |
1934 | } |
1935 | ||
1936 | /* | |
1937 | * A shadow entry of a recently evicted page, a swap | |
1938 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1939 | * without attempting to raise page count. | |
1940 | */ | |
1941 | goto export; | |
1942 | } | |
83929372 KS |
1943 | |
1944 | head = compound_head(page); | |
1945 | if (!page_cache_get_speculative(head)) | |
7e7f7749 RZ |
1946 | goto repeat; |
1947 | ||
83929372 KS |
1948 | /* The page was split under us? */ |
1949 | if (compound_head(page) != head) { | |
1950 | put_page(head); | |
1951 | goto repeat; | |
1952 | } | |
1953 | ||
7e7f7749 RZ |
1954 | /* Has the page moved? */ |
1955 | if (unlikely(page != *slot)) { | |
83929372 | 1956 | put_page(head); |
7e7f7749 RZ |
1957 | goto repeat; |
1958 | } | |
1959 | export: | |
1960 | indices[ret] = iter.index; | |
1961 | entries[ret] = page; | |
1962 | if (++ret == nr_entries) | |
1963 | break; | |
1964 | } | |
1965 | rcu_read_unlock(); | |
1966 | return ret; | |
1967 | } | |
1968 | EXPORT_SYMBOL(find_get_entries_tag); | |
1969 | ||
76d42bd9 WF |
1970 | /* |
1971 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1972 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1973 | * | |
1974 | * ---R__________________________________________B__________ | |
1975 | * ^ reading here ^ bad block(assume 4k) | |
1976 | * | |
1977 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1978 | * => failing the whole request => read(R) => read(R+1) => | |
1979 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1980 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1981 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1982 | * | |
1983 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1984 | */ | |
1985 | static void shrink_readahead_size_eio(struct file *filp, | |
1986 | struct file_ra_state *ra) | |
1987 | { | |
76d42bd9 | 1988 | ra->ra_pages /= 4; |
76d42bd9 WF |
1989 | } |
1990 | ||
485bb99b | 1991 | /** |
47c27bc4 CH |
1992 | * generic_file_buffered_read - generic file read routine |
1993 | * @iocb: the iocb to read | |
6e58e79d AV |
1994 | * @iter: data destination |
1995 | * @written: already copied | |
485bb99b | 1996 | * |
1da177e4 | 1997 | * This is a generic file read routine, and uses the |
485bb99b | 1998 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1999 | * |
2000 | * This is really ugly. But the goto's actually try to clarify some | |
2001 | * of the logic when it comes to error handling etc. | |
1da177e4 | 2002 | */ |
47c27bc4 | 2003 | static ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 2004 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 2005 | { |
47c27bc4 | 2006 | struct file *filp = iocb->ki_filp; |
36e78914 | 2007 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2008 | struct inode *inode = mapping->host; |
36e78914 | 2009 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 2010 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
2011 | pgoff_t index; |
2012 | pgoff_t last_index; | |
2013 | pgoff_t prev_index; | |
2014 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 2015 | unsigned int prev_offset; |
6e58e79d | 2016 | int error = 0; |
1da177e4 | 2017 | |
c2a9737f | 2018 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2019 | return 0; |
c2a9737f WF |
2020 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
2021 | ||
09cbfeaf KS |
2022 | index = *ppos >> PAGE_SHIFT; |
2023 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
2024 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
2025 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
2026 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 2027 | |
1da177e4 LT |
2028 | for (;;) { |
2029 | struct page *page; | |
57f6b96c | 2030 | pgoff_t end_index; |
a32ea1e1 | 2031 | loff_t isize; |
1da177e4 LT |
2032 | unsigned long nr, ret; |
2033 | ||
1da177e4 | 2034 | cond_resched(); |
1da177e4 | 2035 | find_page: |
5abf186a MH |
2036 | if (fatal_signal_pending(current)) { |
2037 | error = -EINTR; | |
2038 | goto out; | |
2039 | } | |
2040 | ||
1da177e4 | 2041 | page = find_get_page(mapping, index); |
3ea89ee8 | 2042 | if (!page) { |
3239d834 MT |
2043 | if (iocb->ki_flags & IOCB_NOWAIT) |
2044 | goto would_block; | |
cf914a7d | 2045 | page_cache_sync_readahead(mapping, |
7ff81078 | 2046 | ra, filp, |
3ea89ee8 FW |
2047 | index, last_index - index); |
2048 | page = find_get_page(mapping, index); | |
2049 | if (unlikely(page == NULL)) | |
2050 | goto no_cached_page; | |
2051 | } | |
2052 | if (PageReadahead(page)) { | |
cf914a7d | 2053 | page_cache_async_readahead(mapping, |
7ff81078 | 2054 | ra, filp, page, |
3ea89ee8 | 2055 | index, last_index - index); |
1da177e4 | 2056 | } |
8ab22b9a | 2057 | if (!PageUptodate(page)) { |
3239d834 MT |
2058 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2059 | put_page(page); | |
2060 | goto would_block; | |
2061 | } | |
2062 | ||
ebded027 MG |
2063 | /* |
2064 | * See comment in do_read_cache_page on why | |
2065 | * wait_on_page_locked is used to avoid unnecessarily | |
2066 | * serialisations and why it's safe. | |
2067 | */ | |
c4b209a4 BVA |
2068 | error = wait_on_page_locked_killable(page); |
2069 | if (unlikely(error)) | |
2070 | goto readpage_error; | |
ebded027 MG |
2071 | if (PageUptodate(page)) |
2072 | goto page_ok; | |
2073 | ||
09cbfeaf | 2074 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2075 | !mapping->a_ops->is_partially_uptodate) |
2076 | goto page_not_up_to_date; | |
6d6d36bc EG |
2077 | /* pipes can't handle partially uptodate pages */ |
2078 | if (unlikely(iter->type & ITER_PIPE)) | |
2079 | goto page_not_up_to_date; | |
529ae9aa | 2080 | if (!trylock_page(page)) |
8ab22b9a | 2081 | goto page_not_up_to_date; |
8d056cb9 DH |
2082 | /* Did it get truncated before we got the lock? */ |
2083 | if (!page->mapping) | |
2084 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2085 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2086 | offset, iter->count)) |
8ab22b9a HH |
2087 | goto page_not_up_to_date_locked; |
2088 | unlock_page(page); | |
2089 | } | |
1da177e4 | 2090 | page_ok: |
a32ea1e1 N |
2091 | /* |
2092 | * i_size must be checked after we know the page is Uptodate. | |
2093 | * | |
2094 | * Checking i_size after the check allows us to calculate | |
2095 | * the correct value for "nr", which means the zero-filled | |
2096 | * part of the page is not copied back to userspace (unless | |
2097 | * another truncate extends the file - this is desired though). | |
2098 | */ | |
2099 | ||
2100 | isize = i_size_read(inode); | |
09cbfeaf | 2101 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2102 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2103 | put_page(page); |
a32ea1e1 N |
2104 | goto out; |
2105 | } | |
2106 | ||
2107 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2108 | nr = PAGE_SIZE; |
a32ea1e1 | 2109 | if (index == end_index) { |
09cbfeaf | 2110 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2111 | if (nr <= offset) { |
09cbfeaf | 2112 | put_page(page); |
a32ea1e1 N |
2113 | goto out; |
2114 | } | |
2115 | } | |
2116 | nr = nr - offset; | |
1da177e4 LT |
2117 | |
2118 | /* If users can be writing to this page using arbitrary | |
2119 | * virtual addresses, take care about potential aliasing | |
2120 | * before reading the page on the kernel side. | |
2121 | */ | |
2122 | if (mapping_writably_mapped(mapping)) | |
2123 | flush_dcache_page(page); | |
2124 | ||
2125 | /* | |
ec0f1637 JK |
2126 | * When a sequential read accesses a page several times, |
2127 | * only mark it as accessed the first time. | |
1da177e4 | 2128 | */ |
ec0f1637 | 2129 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2130 | mark_page_accessed(page); |
2131 | prev_index = index; | |
2132 | ||
2133 | /* | |
2134 | * Ok, we have the page, and it's up-to-date, so | |
2135 | * now we can copy it to user space... | |
1da177e4 | 2136 | */ |
6e58e79d AV |
2137 | |
2138 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2139 | offset += ret; |
09cbfeaf KS |
2140 | index += offset >> PAGE_SHIFT; |
2141 | offset &= ~PAGE_MASK; | |
6ce745ed | 2142 | prev_offset = offset; |
1da177e4 | 2143 | |
09cbfeaf | 2144 | put_page(page); |
6e58e79d AV |
2145 | written += ret; |
2146 | if (!iov_iter_count(iter)) | |
2147 | goto out; | |
2148 | if (ret < nr) { | |
2149 | error = -EFAULT; | |
2150 | goto out; | |
2151 | } | |
2152 | continue; | |
1da177e4 LT |
2153 | |
2154 | page_not_up_to_date: | |
2155 | /* Get exclusive access to the page ... */ | |
85462323 ON |
2156 | error = lock_page_killable(page); |
2157 | if (unlikely(error)) | |
2158 | goto readpage_error; | |
1da177e4 | 2159 | |
8ab22b9a | 2160 | page_not_up_to_date_locked: |
da6052f7 | 2161 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2162 | if (!page->mapping) { |
2163 | unlock_page(page); | |
09cbfeaf | 2164 | put_page(page); |
1da177e4 LT |
2165 | continue; |
2166 | } | |
2167 | ||
2168 | /* Did somebody else fill it already? */ | |
2169 | if (PageUptodate(page)) { | |
2170 | unlock_page(page); | |
2171 | goto page_ok; | |
2172 | } | |
2173 | ||
2174 | readpage: | |
91803b49 JM |
2175 | /* |
2176 | * A previous I/O error may have been due to temporary | |
2177 | * failures, eg. multipath errors. | |
2178 | * PG_error will be set again if readpage fails. | |
2179 | */ | |
2180 | ClearPageError(page); | |
1da177e4 LT |
2181 | /* Start the actual read. The read will unlock the page. */ |
2182 | error = mapping->a_ops->readpage(filp, page); | |
2183 | ||
994fc28c ZB |
2184 | if (unlikely(error)) { |
2185 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2186 | put_page(page); |
6e58e79d | 2187 | error = 0; |
994fc28c ZB |
2188 | goto find_page; |
2189 | } | |
1da177e4 | 2190 | goto readpage_error; |
994fc28c | 2191 | } |
1da177e4 LT |
2192 | |
2193 | if (!PageUptodate(page)) { | |
85462323 ON |
2194 | error = lock_page_killable(page); |
2195 | if (unlikely(error)) | |
2196 | goto readpage_error; | |
1da177e4 LT |
2197 | if (!PageUptodate(page)) { |
2198 | if (page->mapping == NULL) { | |
2199 | /* | |
2ecdc82e | 2200 | * invalidate_mapping_pages got it |
1da177e4 LT |
2201 | */ |
2202 | unlock_page(page); | |
09cbfeaf | 2203 | put_page(page); |
1da177e4 LT |
2204 | goto find_page; |
2205 | } | |
2206 | unlock_page(page); | |
7ff81078 | 2207 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
2208 | error = -EIO; |
2209 | goto readpage_error; | |
1da177e4 LT |
2210 | } |
2211 | unlock_page(page); | |
2212 | } | |
2213 | ||
1da177e4 LT |
2214 | goto page_ok; |
2215 | ||
2216 | readpage_error: | |
2217 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2218 | put_page(page); |
1da177e4 LT |
2219 | goto out; |
2220 | ||
2221 | no_cached_page: | |
2222 | /* | |
2223 | * Ok, it wasn't cached, so we need to create a new | |
2224 | * page.. | |
2225 | */ | |
453f85d4 | 2226 | page = page_cache_alloc(mapping); |
eb2be189 | 2227 | if (!page) { |
6e58e79d | 2228 | error = -ENOMEM; |
eb2be189 | 2229 | goto out; |
1da177e4 | 2230 | } |
6afdb859 | 2231 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2232 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2233 | if (error) { |
09cbfeaf | 2234 | put_page(page); |
6e58e79d AV |
2235 | if (error == -EEXIST) { |
2236 | error = 0; | |
1da177e4 | 2237 | goto find_page; |
6e58e79d | 2238 | } |
1da177e4 LT |
2239 | goto out; |
2240 | } | |
1da177e4 LT |
2241 | goto readpage; |
2242 | } | |
2243 | ||
3239d834 MT |
2244 | would_block: |
2245 | error = -EAGAIN; | |
1da177e4 | 2246 | out: |
7ff81078 | 2247 | ra->prev_pos = prev_index; |
09cbfeaf | 2248 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2249 | ra->prev_pos |= prev_offset; |
1da177e4 | 2250 | |
09cbfeaf | 2251 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2252 | file_accessed(filp); |
6e58e79d | 2253 | return written ? written : error; |
1da177e4 LT |
2254 | } |
2255 | ||
485bb99b | 2256 | /** |
6abd2322 | 2257 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2258 | * @iocb: kernel I/O control block |
6abd2322 | 2259 | * @iter: destination for the data read |
485bb99b | 2260 | * |
6abd2322 | 2261 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
2262 | * that can use the page cache directly. |
2263 | */ | |
2264 | ssize_t | |
ed978a81 | 2265 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2266 | { |
e7080a43 | 2267 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2268 | ssize_t retval = 0; |
e7080a43 NS |
2269 | |
2270 | if (!count) | |
2271 | goto out; /* skip atime */ | |
1da177e4 | 2272 | |
2ba48ce5 | 2273 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2274 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2275 | struct address_space *mapping = file->f_mapping; |
2276 | struct inode *inode = mapping->host; | |
543ade1f | 2277 | loff_t size; |
1da177e4 | 2278 | |
1da177e4 | 2279 | size = i_size_read(inode); |
6be96d3a GR |
2280 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2281 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2282 | iocb->ki_pos + count - 1)) | |
2283 | return -EAGAIN; | |
2284 | } else { | |
2285 | retval = filemap_write_and_wait_range(mapping, | |
2286 | iocb->ki_pos, | |
2287 | iocb->ki_pos + count - 1); | |
2288 | if (retval < 0) | |
2289 | goto out; | |
2290 | } | |
d8d3d94b | 2291 | |
0d5b0cf2 CH |
2292 | file_accessed(file); |
2293 | ||
5ecda137 | 2294 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2295 | if (retval >= 0) { |
c64fb5c7 | 2296 | iocb->ki_pos += retval; |
5ecda137 | 2297 | count -= retval; |
9fe55eea | 2298 | } |
5b47d59a | 2299 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2300 | |
9fe55eea SW |
2301 | /* |
2302 | * Btrfs can have a short DIO read if we encounter | |
2303 | * compressed extents, so if there was an error, or if | |
2304 | * we've already read everything we wanted to, or if | |
2305 | * there was a short read because we hit EOF, go ahead | |
2306 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2307 | * the rest of the read. Buffered reads will not work for |
2308 | * DAX files, so don't bother trying. | |
9fe55eea | 2309 | */ |
5ecda137 | 2310 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2311 | IS_DAX(inode)) |
9fe55eea | 2312 | goto out; |
1da177e4 LT |
2313 | } |
2314 | ||
47c27bc4 | 2315 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2316 | out: |
2317 | return retval; | |
2318 | } | |
ed978a81 | 2319 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2320 | |
1da177e4 | 2321 | #ifdef CONFIG_MMU |
485bb99b RD |
2322 | /** |
2323 | * page_cache_read - adds requested page to the page cache if not already there | |
2324 | * @file: file to read | |
2325 | * @offset: page index | |
62eb320a | 2326 | * @gfp_mask: memory allocation flags |
485bb99b | 2327 | * |
1da177e4 LT |
2328 | * This adds the requested page to the page cache if it isn't already there, |
2329 | * and schedules an I/O to read in its contents from disk. | |
2330 | */ | |
c20cd45e | 2331 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
2332 | { |
2333 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 2334 | struct page *page; |
994fc28c | 2335 | int ret; |
1da177e4 | 2336 | |
994fc28c | 2337 | do { |
453f85d4 | 2338 | page = __page_cache_alloc(gfp_mask); |
994fc28c ZB |
2339 | if (!page) |
2340 | return -ENOMEM; | |
2341 | ||
abc1be13 | 2342 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
994fc28c ZB |
2343 | if (ret == 0) |
2344 | ret = mapping->a_ops->readpage(file, page); | |
2345 | else if (ret == -EEXIST) | |
2346 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 2347 | |
09cbfeaf | 2348 | put_page(page); |
1da177e4 | 2349 | |
994fc28c | 2350 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 2351 | |
994fc28c | 2352 | return ret; |
1da177e4 LT |
2353 | } |
2354 | ||
2355 | #define MMAP_LOTSAMISS (100) | |
2356 | ||
ef00e08e LT |
2357 | /* |
2358 | * Synchronous readahead happens when we don't even find | |
2359 | * a page in the page cache at all. | |
2360 | */ | |
2361 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
2362 | struct file_ra_state *ra, | |
2363 | struct file *file, | |
2364 | pgoff_t offset) | |
2365 | { | |
ef00e08e LT |
2366 | struct address_space *mapping = file->f_mapping; |
2367 | ||
2368 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2369 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 2370 | return; |
275b12bf WF |
2371 | if (!ra->ra_pages) |
2372 | return; | |
ef00e08e | 2373 | |
64363aad | 2374 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
2375 | page_cache_sync_readahead(mapping, ra, file, offset, |
2376 | ra->ra_pages); | |
ef00e08e LT |
2377 | return; |
2378 | } | |
2379 | ||
207d04ba AK |
2380 | /* Avoid banging the cache line if not needed */ |
2381 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2382 | ra->mmap_miss++; |
2383 | ||
2384 | /* | |
2385 | * Do we miss much more than hit in this file? If so, | |
2386 | * stop bothering with read-ahead. It will only hurt. | |
2387 | */ | |
2388 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
2389 | return; | |
2390 | ||
d30a1100 WF |
2391 | /* |
2392 | * mmap read-around | |
2393 | */ | |
600e19af RG |
2394 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2395 | ra->size = ra->ra_pages; | |
2396 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2397 | ra_submit(ra, mapping, file); |
ef00e08e LT |
2398 | } |
2399 | ||
2400 | /* | |
2401 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2402 | * so we want to possibly extend the readahead further.. | |
2403 | */ | |
2404 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2405 | struct file_ra_state *ra, | |
2406 | struct file *file, | |
2407 | struct page *page, | |
2408 | pgoff_t offset) | |
2409 | { | |
2410 | struct address_space *mapping = file->f_mapping; | |
2411 | ||
2412 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2413 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2414 | return; |
2415 | if (ra->mmap_miss > 0) | |
2416 | ra->mmap_miss--; | |
2417 | if (PageReadahead(page)) | |
2fad6f5d WF |
2418 | page_cache_async_readahead(mapping, ra, file, |
2419 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2420 | } |
2421 | ||
485bb99b | 2422 | /** |
54cb8821 | 2423 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2424 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2425 | * |
54cb8821 | 2426 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2427 | * mapped memory region to read in file data during a page fault. |
2428 | * | |
2429 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2430 | * it in the page cache, and handles the special cases reasonably without | |
2431 | * having a lot of duplicated code. | |
9a95f3cf PC |
2432 | * |
2433 | * vma->vm_mm->mmap_sem must be held on entry. | |
2434 | * | |
2435 | * If our return value has VM_FAULT_RETRY set, it's because | |
2436 | * lock_page_or_retry() returned 0. | |
2437 | * The mmap_sem has usually been released in this case. | |
2438 | * See __lock_page_or_retry() for the exception. | |
2439 | * | |
2440 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2441 | * has not been released. | |
2442 | * | |
2443 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2444 | */ |
2bcd6454 | 2445 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2446 | { |
2447 | int error; | |
11bac800 | 2448 | struct file *file = vmf->vma->vm_file; |
1da177e4 LT |
2449 | struct address_space *mapping = file->f_mapping; |
2450 | struct file_ra_state *ra = &file->f_ra; | |
2451 | struct inode *inode = mapping->host; | |
ef00e08e | 2452 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2453 | pgoff_t max_off; |
1da177e4 | 2454 | struct page *page; |
2bcd6454 | 2455 | vm_fault_t ret = 0; |
1da177e4 | 2456 | |
9ab2594f MW |
2457 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2458 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2459 | return VM_FAULT_SIGBUS; |
1da177e4 | 2460 | |
1da177e4 | 2461 | /* |
49426420 | 2462 | * Do we have something in the page cache already? |
1da177e4 | 2463 | */ |
ef00e08e | 2464 | page = find_get_page(mapping, offset); |
45cac65b | 2465 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2466 | /* |
ef00e08e LT |
2467 | * We found the page, so try async readahead before |
2468 | * waiting for the lock. | |
1da177e4 | 2469 | */ |
11bac800 | 2470 | do_async_mmap_readahead(vmf->vma, ra, file, page, offset); |
45cac65b | 2471 | } else if (!page) { |
ef00e08e | 2472 | /* No page in the page cache at all */ |
11bac800 | 2473 | do_sync_mmap_readahead(vmf->vma, ra, file, offset); |
ef00e08e | 2474 | count_vm_event(PGMAJFAULT); |
2262185c | 2475 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2476 | ret = VM_FAULT_MAJOR; |
2477 | retry_find: | |
b522c94d | 2478 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2479 | if (!page) |
2480 | goto no_cached_page; | |
2481 | } | |
2482 | ||
11bac800 | 2483 | if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2484 | put_page(page); |
d065bd81 | 2485 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2486 | } |
b522c94d ML |
2487 | |
2488 | /* Did it get truncated? */ | |
2489 | if (unlikely(page->mapping != mapping)) { | |
2490 | unlock_page(page); | |
2491 | put_page(page); | |
2492 | goto retry_find; | |
2493 | } | |
309381fe | 2494 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2495 | |
1da177e4 | 2496 | /* |
d00806b1 NP |
2497 | * We have a locked page in the page cache, now we need to check |
2498 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2499 | */ |
d00806b1 | 2500 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2501 | goto page_not_uptodate; |
2502 | ||
ef00e08e LT |
2503 | /* |
2504 | * Found the page and have a reference on it. | |
2505 | * We must recheck i_size under page lock. | |
2506 | */ | |
9ab2594f MW |
2507 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2508 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2509 | unlock_page(page); |
09cbfeaf | 2510 | put_page(page); |
5307cc1a | 2511 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2512 | } |
2513 | ||
d0217ac0 | 2514 | vmf->page = page; |
83c54070 | 2515 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2516 | |
1da177e4 LT |
2517 | no_cached_page: |
2518 | /* | |
2519 | * We're only likely to ever get here if MADV_RANDOM is in | |
2520 | * effect. | |
2521 | */ | |
c20cd45e | 2522 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2523 | |
2524 | /* | |
2525 | * The page we want has now been added to the page cache. | |
2526 | * In the unlikely event that someone removed it in the | |
2527 | * meantime, we'll just come back here and read it again. | |
2528 | */ | |
2529 | if (error >= 0) | |
2530 | goto retry_find; | |
2531 | ||
2532 | /* | |
2533 | * An error return from page_cache_read can result if the | |
2534 | * system is low on memory, or a problem occurs while trying | |
2535 | * to schedule I/O. | |
2536 | */ | |
2537 | if (error == -ENOMEM) | |
d0217ac0 NP |
2538 | return VM_FAULT_OOM; |
2539 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2540 | |
2541 | page_not_uptodate: | |
1da177e4 LT |
2542 | /* |
2543 | * Umm, take care of errors if the page isn't up-to-date. | |
2544 | * Try to re-read it _once_. We do this synchronously, | |
2545 | * because there really aren't any performance issues here | |
2546 | * and we need to check for errors. | |
2547 | */ | |
1da177e4 | 2548 | ClearPageError(page); |
994fc28c | 2549 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2550 | if (!error) { |
2551 | wait_on_page_locked(page); | |
2552 | if (!PageUptodate(page)) | |
2553 | error = -EIO; | |
2554 | } | |
09cbfeaf | 2555 | put_page(page); |
d00806b1 NP |
2556 | |
2557 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2558 | goto retry_find; |
1da177e4 | 2559 | |
d00806b1 | 2560 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2561 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2562 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2563 | } |
2564 | EXPORT_SYMBOL(filemap_fault); | |
2565 | ||
82b0f8c3 | 2566 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2567 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 KS |
2568 | { |
2569 | struct radix_tree_iter iter; | |
2570 | void **slot; | |
82b0f8c3 | 2571 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2572 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2573 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2574 | unsigned long max_idx; |
83929372 | 2575 | struct page *head, *page; |
f1820361 KS |
2576 | |
2577 | rcu_read_lock(); | |
b93b0163 | 2578 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start_pgoff) { |
bae473a4 | 2579 | if (iter.index > end_pgoff) |
f1820361 KS |
2580 | break; |
2581 | repeat: | |
2582 | page = radix_tree_deref_slot(slot); | |
2583 | if (unlikely(!page)) | |
2584 | goto next; | |
2585 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2586 | if (radix_tree_deref_retry(page)) { |
2587 | slot = radix_tree_iter_retry(&iter); | |
2588 | continue; | |
2589 | } | |
2590 | goto next; | |
f1820361 KS |
2591 | } |
2592 | ||
83929372 KS |
2593 | head = compound_head(page); |
2594 | if (!page_cache_get_speculative(head)) | |
f1820361 KS |
2595 | goto repeat; |
2596 | ||
83929372 KS |
2597 | /* The page was split under us? */ |
2598 | if (compound_head(page) != head) { | |
2599 | put_page(head); | |
2600 | goto repeat; | |
2601 | } | |
2602 | ||
f1820361 KS |
2603 | /* Has the page moved? */ |
2604 | if (unlikely(page != *slot)) { | |
83929372 | 2605 | put_page(head); |
f1820361 KS |
2606 | goto repeat; |
2607 | } | |
2608 | ||
2609 | if (!PageUptodate(page) || | |
2610 | PageReadahead(page) || | |
2611 | PageHWPoison(page)) | |
2612 | goto skip; | |
2613 | if (!trylock_page(page)) | |
2614 | goto skip; | |
2615 | ||
2616 | if (page->mapping != mapping || !PageUptodate(page)) | |
2617 | goto unlock; | |
2618 | ||
9ab2594f MW |
2619 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2620 | if (page->index >= max_idx) | |
f1820361 KS |
2621 | goto unlock; |
2622 | ||
f1820361 KS |
2623 | if (file->f_ra.mmap_miss > 0) |
2624 | file->f_ra.mmap_miss--; | |
7267ec00 | 2625 | |
82b0f8c3 JK |
2626 | vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT; |
2627 | if (vmf->pte) | |
2628 | vmf->pte += iter.index - last_pgoff; | |
7267ec00 | 2629 | last_pgoff = iter.index; |
82b0f8c3 | 2630 | if (alloc_set_pte(vmf, NULL, page)) |
7267ec00 | 2631 | goto unlock; |
f1820361 KS |
2632 | unlock_page(page); |
2633 | goto next; | |
2634 | unlock: | |
2635 | unlock_page(page); | |
2636 | skip: | |
09cbfeaf | 2637 | put_page(page); |
f1820361 | 2638 | next: |
7267ec00 | 2639 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2640 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2641 | break; |
bae473a4 | 2642 | if (iter.index == end_pgoff) |
f1820361 KS |
2643 | break; |
2644 | } | |
2645 | rcu_read_unlock(); | |
2646 | } | |
2647 | EXPORT_SYMBOL(filemap_map_pages); | |
2648 | ||
2bcd6454 | 2649 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2650 | { |
2651 | struct page *page = vmf->page; | |
11bac800 | 2652 | struct inode *inode = file_inode(vmf->vma->vm_file); |
2bcd6454 | 2653 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 2654 | |
14da9200 | 2655 | sb_start_pagefault(inode->i_sb); |
11bac800 | 2656 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 JK |
2657 | lock_page(page); |
2658 | if (page->mapping != inode->i_mapping) { | |
2659 | unlock_page(page); | |
2660 | ret = VM_FAULT_NOPAGE; | |
2661 | goto out; | |
2662 | } | |
14da9200 JK |
2663 | /* |
2664 | * We mark the page dirty already here so that when freeze is in | |
2665 | * progress, we are guaranteed that writeback during freezing will | |
2666 | * see the dirty page and writeprotect it again. | |
2667 | */ | |
2668 | set_page_dirty(page); | |
1d1d1a76 | 2669 | wait_for_stable_page(page); |
4fcf1c62 | 2670 | out: |
14da9200 | 2671 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2672 | return ret; |
2673 | } | |
4fcf1c62 | 2674 | |
f0f37e2f | 2675 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2676 | .fault = filemap_fault, |
f1820361 | 2677 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2678 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2679 | }; |
2680 | ||
2681 | /* This is used for a general mmap of a disk file */ | |
2682 | ||
2683 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2684 | { | |
2685 | struct address_space *mapping = file->f_mapping; | |
2686 | ||
2687 | if (!mapping->a_ops->readpage) | |
2688 | return -ENOEXEC; | |
2689 | file_accessed(file); | |
2690 | vma->vm_ops = &generic_file_vm_ops; | |
2691 | return 0; | |
2692 | } | |
1da177e4 LT |
2693 | |
2694 | /* | |
2695 | * This is for filesystems which do not implement ->writepage. | |
2696 | */ | |
2697 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2698 | { | |
2699 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2700 | return -EINVAL; | |
2701 | return generic_file_mmap(file, vma); | |
2702 | } | |
2703 | #else | |
45397228 AB |
2704 | int filemap_page_mkwrite(struct vm_fault *vmf) |
2705 | { | |
2706 | return -ENOSYS; | |
2707 | } | |
1da177e4 LT |
2708 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2709 | { | |
2710 | return -ENOSYS; | |
2711 | } | |
2712 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2713 | { | |
2714 | return -ENOSYS; | |
2715 | } | |
2716 | #endif /* CONFIG_MMU */ | |
2717 | ||
45397228 | 2718 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
2719 | EXPORT_SYMBOL(generic_file_mmap); |
2720 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2721 | ||
67f9fd91 SL |
2722 | static struct page *wait_on_page_read(struct page *page) |
2723 | { | |
2724 | if (!IS_ERR(page)) { | |
2725 | wait_on_page_locked(page); | |
2726 | if (!PageUptodate(page)) { | |
09cbfeaf | 2727 | put_page(page); |
67f9fd91 SL |
2728 | page = ERR_PTR(-EIO); |
2729 | } | |
2730 | } | |
2731 | return page; | |
2732 | } | |
2733 | ||
32b63529 | 2734 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2735 | pgoff_t index, |
5e5358e7 | 2736 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2737 | void *data, |
2738 | gfp_t gfp) | |
1da177e4 | 2739 | { |
eb2be189 | 2740 | struct page *page; |
1da177e4 LT |
2741 | int err; |
2742 | repeat: | |
2743 | page = find_get_page(mapping, index); | |
2744 | if (!page) { | |
453f85d4 | 2745 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
2746 | if (!page) |
2747 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2748 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2749 | if (unlikely(err)) { |
09cbfeaf | 2750 | put_page(page); |
eb2be189 NP |
2751 | if (err == -EEXIST) |
2752 | goto repeat; | |
1da177e4 | 2753 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2754 | return ERR_PTR(err); |
2755 | } | |
32b63529 MG |
2756 | |
2757 | filler: | |
1da177e4 LT |
2758 | err = filler(data, page); |
2759 | if (err < 0) { | |
09cbfeaf | 2760 | put_page(page); |
32b63529 | 2761 | return ERR_PTR(err); |
1da177e4 | 2762 | } |
1da177e4 | 2763 | |
32b63529 MG |
2764 | page = wait_on_page_read(page); |
2765 | if (IS_ERR(page)) | |
2766 | return page; | |
2767 | goto out; | |
2768 | } | |
1da177e4 LT |
2769 | if (PageUptodate(page)) |
2770 | goto out; | |
2771 | ||
ebded027 MG |
2772 | /* |
2773 | * Page is not up to date and may be locked due one of the following | |
2774 | * case a: Page is being filled and the page lock is held | |
2775 | * case b: Read/write error clearing the page uptodate status | |
2776 | * case c: Truncation in progress (page locked) | |
2777 | * case d: Reclaim in progress | |
2778 | * | |
2779 | * Case a, the page will be up to date when the page is unlocked. | |
2780 | * There is no need to serialise on the page lock here as the page | |
2781 | * is pinned so the lock gives no additional protection. Even if the | |
2782 | * the page is truncated, the data is still valid if PageUptodate as | |
2783 | * it's a race vs truncate race. | |
2784 | * Case b, the page will not be up to date | |
2785 | * Case c, the page may be truncated but in itself, the data may still | |
2786 | * be valid after IO completes as it's a read vs truncate race. The | |
2787 | * operation must restart if the page is not uptodate on unlock but | |
2788 | * otherwise serialising on page lock to stabilise the mapping gives | |
2789 | * no additional guarantees to the caller as the page lock is | |
2790 | * released before return. | |
2791 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2792 | * will be a race with remove_mapping that determines if the mapping | |
2793 | * is valid on unlock but otherwise the data is valid and there is | |
2794 | * no need to serialise with page lock. | |
2795 | * | |
2796 | * As the page lock gives no additional guarantee, we optimistically | |
2797 | * wait on the page to be unlocked and check if it's up to date and | |
2798 | * use the page if it is. Otherwise, the page lock is required to | |
2799 | * distinguish between the different cases. The motivation is that we | |
2800 | * avoid spurious serialisations and wakeups when multiple processes | |
2801 | * wait on the same page for IO to complete. | |
2802 | */ | |
2803 | wait_on_page_locked(page); | |
2804 | if (PageUptodate(page)) | |
2805 | goto out; | |
2806 | ||
2807 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2808 | lock_page(page); |
ebded027 MG |
2809 | |
2810 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2811 | if (!page->mapping) { |
2812 | unlock_page(page); | |
09cbfeaf | 2813 | put_page(page); |
32b63529 | 2814 | goto repeat; |
1da177e4 | 2815 | } |
ebded027 MG |
2816 | |
2817 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2818 | if (PageUptodate(page)) { |
2819 | unlock_page(page); | |
2820 | goto out; | |
2821 | } | |
32b63529 MG |
2822 | goto filler; |
2823 | ||
c855ff37 | 2824 | out: |
6fe6900e NP |
2825 | mark_page_accessed(page); |
2826 | return page; | |
2827 | } | |
0531b2aa LT |
2828 | |
2829 | /** | |
67f9fd91 | 2830 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2831 | * @mapping: the page's address_space |
2832 | * @index: the page index | |
2833 | * @filler: function to perform the read | |
5e5358e7 | 2834 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2835 | * |
0531b2aa | 2836 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2837 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2838 | * |
2839 | * If the page does not get brought uptodate, return -EIO. | |
2840 | */ | |
67f9fd91 | 2841 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2842 | pgoff_t index, |
5e5358e7 | 2843 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2844 | void *data) |
2845 | { | |
2846 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2847 | } | |
67f9fd91 | 2848 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2849 | |
2850 | /** | |
2851 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2852 | * @mapping: the page's address_space | |
2853 | * @index: the page index | |
2854 | * @gfp: the page allocator flags to use if allocating | |
2855 | * | |
2856 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2857 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2858 | * |
2859 | * If the page does not get brought uptodate, return -EIO. | |
2860 | */ | |
2861 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2862 | pgoff_t index, | |
2863 | gfp_t gfp) | |
2864 | { | |
2865 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2866 | ||
67f9fd91 | 2867 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2868 | } |
2869 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2870 | ||
1da177e4 LT |
2871 | /* |
2872 | * Performs necessary checks before doing a write | |
2873 | * | |
485bb99b | 2874 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2875 | * Returns appropriate error code that caller should return or |
2876 | * zero in case that write should be allowed. | |
2877 | */ | |
3309dd04 | 2878 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2879 | { |
3309dd04 | 2880 | struct file *file = iocb->ki_filp; |
1da177e4 | 2881 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2882 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2883 | loff_t pos; |
1da177e4 | 2884 | |
3309dd04 AV |
2885 | if (!iov_iter_count(from)) |
2886 | return 0; | |
1da177e4 | 2887 | |
0fa6b005 | 2888 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2889 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2890 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2891 | |
3309dd04 | 2892 | pos = iocb->ki_pos; |
1da177e4 | 2893 | |
6be96d3a GR |
2894 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
2895 | return -EINVAL; | |
2896 | ||
0fa6b005 | 2897 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2898 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2899 | send_sig(SIGXFSZ, current, 0); |
2900 | return -EFBIG; | |
1da177e4 | 2901 | } |
3309dd04 | 2902 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2903 | } |
2904 | ||
2905 | /* | |
2906 | * LFS rule | |
2907 | */ | |
3309dd04 | 2908 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2909 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2910 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2911 | return -EFBIG; |
3309dd04 | 2912 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2913 | } |
2914 | ||
2915 | /* | |
2916 | * Are we about to exceed the fs block limit ? | |
2917 | * | |
2918 | * If we have written data it becomes a short write. If we have | |
2919 | * exceeded without writing data we send a signal and return EFBIG. | |
2920 | * Linus frestrict idea will clean these up nicely.. | |
2921 | */ | |
3309dd04 AV |
2922 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2923 | return -EFBIG; | |
1da177e4 | 2924 | |
3309dd04 AV |
2925 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2926 | return iov_iter_count(from); | |
1da177e4 LT |
2927 | } |
2928 | EXPORT_SYMBOL(generic_write_checks); | |
2929 | ||
afddba49 NP |
2930 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2931 | loff_t pos, unsigned len, unsigned flags, | |
2932 | struct page **pagep, void **fsdata) | |
2933 | { | |
2934 | const struct address_space_operations *aops = mapping->a_ops; | |
2935 | ||
4e02ed4b | 2936 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2937 | pagep, fsdata); |
afddba49 NP |
2938 | } |
2939 | EXPORT_SYMBOL(pagecache_write_begin); | |
2940 | ||
2941 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2942 | loff_t pos, unsigned len, unsigned copied, | |
2943 | struct page *page, void *fsdata) | |
2944 | { | |
2945 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2946 | |
4e02ed4b | 2947 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2948 | } |
2949 | EXPORT_SYMBOL(pagecache_write_end); | |
2950 | ||
1da177e4 | 2951 | ssize_t |
1af5bb49 | 2952 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2953 | { |
2954 | struct file *file = iocb->ki_filp; | |
2955 | struct address_space *mapping = file->f_mapping; | |
2956 | struct inode *inode = mapping->host; | |
1af5bb49 | 2957 | loff_t pos = iocb->ki_pos; |
1da177e4 | 2958 | ssize_t written; |
a969e903 CH |
2959 | size_t write_len; |
2960 | pgoff_t end; | |
1da177e4 | 2961 | |
0c949334 | 2962 | write_len = iov_iter_count(from); |
09cbfeaf | 2963 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2964 | |
6be96d3a GR |
2965 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2966 | /* If there are pages to writeback, return */ | |
2967 | if (filemap_range_has_page(inode->i_mapping, pos, | |
2968 | pos + iov_iter_count(from))) | |
2969 | return -EAGAIN; | |
2970 | } else { | |
2971 | written = filemap_write_and_wait_range(mapping, pos, | |
2972 | pos + write_len - 1); | |
2973 | if (written) | |
2974 | goto out; | |
2975 | } | |
a969e903 CH |
2976 | |
2977 | /* | |
2978 | * After a write we want buffered reads to be sure to go to disk to get | |
2979 | * the new data. We invalidate clean cached page from the region we're | |
2980 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2981 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 2982 | */ |
55635ba7 | 2983 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 2984 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
2985 | /* |
2986 | * If a page can not be invalidated, return 0 to fall back | |
2987 | * to buffered write. | |
2988 | */ | |
2989 | if (written) { | |
2990 | if (written == -EBUSY) | |
2991 | return 0; | |
2992 | goto out; | |
a969e903 CH |
2993 | } |
2994 | ||
639a93a5 | 2995 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
2996 | |
2997 | /* | |
2998 | * Finally, try again to invalidate clean pages which might have been | |
2999 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3000 | * if the source of the write was an mmap'ed region of the file | |
3001 | * we're writing. Either one is a pretty crazy thing to do, | |
3002 | * so we don't support it 100%. If this invalidation | |
3003 | * fails, tough, the write still worked... | |
332391a9 LC |
3004 | * |
3005 | * Most of the time we do not need this since dio_complete() will do | |
3006 | * the invalidation for us. However there are some file systems that | |
3007 | * do not end up with dio_complete() being called, so let's not break | |
3008 | * them by removing it completely | |
a969e903 | 3009 | */ |
332391a9 LC |
3010 | if (mapping->nrpages) |
3011 | invalidate_inode_pages2_range(mapping, | |
3012 | pos >> PAGE_SHIFT, end); | |
a969e903 | 3013 | |
1da177e4 | 3014 | if (written > 0) { |
0116651c | 3015 | pos += written; |
639a93a5 | 3016 | write_len -= written; |
0116651c NK |
3017 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3018 | i_size_write(inode, pos); | |
1da177e4 LT |
3019 | mark_inode_dirty(inode); |
3020 | } | |
5cb6c6c7 | 3021 | iocb->ki_pos = pos; |
1da177e4 | 3022 | } |
639a93a5 | 3023 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3024 | out: |
1da177e4 LT |
3025 | return written; |
3026 | } | |
3027 | EXPORT_SYMBOL(generic_file_direct_write); | |
3028 | ||
eb2be189 NP |
3029 | /* |
3030 | * Find or create a page at the given pagecache position. Return the locked | |
3031 | * page. This function is specifically for buffered writes. | |
3032 | */ | |
54566b2c NP |
3033 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3034 | pgoff_t index, unsigned flags) | |
eb2be189 | 3035 | { |
eb2be189 | 3036 | struct page *page; |
bbddabe2 | 3037 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3038 | |
54566b2c | 3039 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3040 | fgp_flags |= FGP_NOFS; |
3041 | ||
3042 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3043 | mapping_gfp_mask(mapping)); |
c585a267 | 3044 | if (page) |
2457aec6 | 3045 | wait_for_stable_page(page); |
eb2be189 | 3046 | |
eb2be189 NP |
3047 | return page; |
3048 | } | |
54566b2c | 3049 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3050 | |
3b93f911 | 3051 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3052 | struct iov_iter *i, loff_t pos) |
3053 | { | |
3054 | struct address_space *mapping = file->f_mapping; | |
3055 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3056 | long status = 0; | |
3057 | ssize_t written = 0; | |
674b892e NP |
3058 | unsigned int flags = 0; |
3059 | ||
afddba49 NP |
3060 | do { |
3061 | struct page *page; | |
afddba49 NP |
3062 | unsigned long offset; /* Offset into pagecache page */ |
3063 | unsigned long bytes; /* Bytes to write to page */ | |
3064 | size_t copied; /* Bytes copied from user */ | |
3065 | void *fsdata; | |
3066 | ||
09cbfeaf KS |
3067 | offset = (pos & (PAGE_SIZE - 1)); |
3068 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3069 | iov_iter_count(i)); |
3070 | ||
3071 | again: | |
00a3d660 LT |
3072 | /* |
3073 | * Bring in the user page that we will copy from _first_. | |
3074 | * Otherwise there's a nasty deadlock on copying from the | |
3075 | * same page as we're writing to, without it being marked | |
3076 | * up-to-date. | |
3077 | * | |
3078 | * Not only is this an optimisation, but it is also required | |
3079 | * to check that the address is actually valid, when atomic | |
3080 | * usercopies are used, below. | |
3081 | */ | |
3082 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3083 | status = -EFAULT; | |
3084 | break; | |
3085 | } | |
3086 | ||
296291cd JK |
3087 | if (fatal_signal_pending(current)) { |
3088 | status = -EINTR; | |
3089 | break; | |
3090 | } | |
3091 | ||
674b892e | 3092 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3093 | &page, &fsdata); |
2457aec6 | 3094 | if (unlikely(status < 0)) |
afddba49 NP |
3095 | break; |
3096 | ||
931e80e4 | 3097 | if (mapping_writably_mapped(mapping)) |
3098 | flush_dcache_page(page); | |
00a3d660 | 3099 | |
afddba49 | 3100 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3101 | flush_dcache_page(page); |
3102 | ||
3103 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3104 | page, fsdata); | |
3105 | if (unlikely(status < 0)) | |
3106 | break; | |
3107 | copied = status; | |
3108 | ||
3109 | cond_resched(); | |
3110 | ||
124d3b70 | 3111 | iov_iter_advance(i, copied); |
afddba49 NP |
3112 | if (unlikely(copied == 0)) { |
3113 | /* | |
3114 | * If we were unable to copy any data at all, we must | |
3115 | * fall back to a single segment length write. | |
3116 | * | |
3117 | * If we didn't fallback here, we could livelock | |
3118 | * because not all segments in the iov can be copied at | |
3119 | * once without a pagefault. | |
3120 | */ | |
09cbfeaf | 3121 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3122 | iov_iter_single_seg_count(i)); |
3123 | goto again; | |
3124 | } | |
afddba49 NP |
3125 | pos += copied; |
3126 | written += copied; | |
3127 | ||
3128 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3129 | } while (iov_iter_count(i)); |
3130 | ||
3131 | return written ? written : status; | |
3132 | } | |
3b93f911 | 3133 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3134 | |
e4dd9de3 | 3135 | /** |
8174202b | 3136 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3137 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3138 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3139 | * |
3140 | * This function does all the work needed for actually writing data to a | |
3141 | * file. It does all basic checks, removes SUID from the file, updates | |
3142 | * modification times and calls proper subroutines depending on whether we | |
3143 | * do direct IO or a standard buffered write. | |
3144 | * | |
3145 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3146 | * object which does not need locking at all. | |
3147 | * | |
3148 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3149 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3150 | * avoid syncing under i_mutex. | |
3151 | */ | |
8174202b | 3152 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3153 | { |
3154 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3155 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3156 | struct inode *inode = mapping->host; |
3b93f911 | 3157 | ssize_t written = 0; |
1da177e4 | 3158 | ssize_t err; |
3b93f911 | 3159 | ssize_t status; |
1da177e4 | 3160 | |
1da177e4 | 3161 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3162 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3163 | err = file_remove_privs(file); |
1da177e4 LT |
3164 | if (err) |
3165 | goto out; | |
3166 | ||
c3b2da31 JB |
3167 | err = file_update_time(file); |
3168 | if (err) | |
3169 | goto out; | |
1da177e4 | 3170 | |
2ba48ce5 | 3171 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3172 | loff_t pos, endbyte; |
fb5527e6 | 3173 | |
1af5bb49 | 3174 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3175 | /* |
fbbbad4b MW |
3176 | * If the write stopped short of completing, fall back to |
3177 | * buffered writes. Some filesystems do this for writes to | |
3178 | * holes, for example. For DAX files, a buffered write will | |
3179 | * not succeed (even if it did, DAX does not handle dirty | |
3180 | * page-cache pages correctly). | |
1da177e4 | 3181 | */ |
0b8def9d | 3182 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3183 | goto out; |
3184 | ||
0b8def9d | 3185 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3186 | /* |
3b93f911 | 3187 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3188 | * then we want to return the number of bytes which were |
3189 | * direct-written, or the error code if that was zero. Note | |
3190 | * that this differs from normal direct-io semantics, which | |
3191 | * will return -EFOO even if some bytes were written. | |
3192 | */ | |
60bb4529 | 3193 | if (unlikely(status < 0)) { |
3b93f911 | 3194 | err = status; |
fb5527e6 JM |
3195 | goto out; |
3196 | } | |
fb5527e6 JM |
3197 | /* |
3198 | * We need to ensure that the page cache pages are written to | |
3199 | * disk and invalidated to preserve the expected O_DIRECT | |
3200 | * semantics. | |
3201 | */ | |
3b93f911 | 3202 | endbyte = pos + status - 1; |
0b8def9d | 3203 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3204 | if (err == 0) { |
0b8def9d | 3205 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3206 | written += status; |
fb5527e6 | 3207 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3208 | pos >> PAGE_SHIFT, |
3209 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3210 | } else { |
3211 | /* | |
3212 | * We don't know how much we wrote, so just return | |
3213 | * the number of bytes which were direct-written | |
3214 | */ | |
3215 | } | |
3216 | } else { | |
0b8def9d AV |
3217 | written = generic_perform_write(file, from, iocb->ki_pos); |
3218 | if (likely(written > 0)) | |
3219 | iocb->ki_pos += written; | |
fb5527e6 | 3220 | } |
1da177e4 LT |
3221 | out: |
3222 | current->backing_dev_info = NULL; | |
3223 | return written ? written : err; | |
3224 | } | |
8174202b | 3225 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3226 | |
e4dd9de3 | 3227 | /** |
8174202b | 3228 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3229 | * @iocb: IO state structure |
8174202b | 3230 | * @from: iov_iter with data to write |
e4dd9de3 | 3231 | * |
8174202b | 3232 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3233 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3234 | * and acquires i_mutex as needed. | |
3235 | */ | |
8174202b | 3236 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3237 | { |
3238 | struct file *file = iocb->ki_filp; | |
148f948b | 3239 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3240 | ssize_t ret; |
1da177e4 | 3241 | |
5955102c | 3242 | inode_lock(inode); |
3309dd04 AV |
3243 | ret = generic_write_checks(iocb, from); |
3244 | if (ret > 0) | |
5f380c7f | 3245 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3246 | inode_unlock(inode); |
1da177e4 | 3247 | |
e2592217 CH |
3248 | if (ret > 0) |
3249 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3250 | return ret; |
3251 | } | |
8174202b | 3252 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3253 | |
cf9a2ae8 DH |
3254 | /** |
3255 | * try_to_release_page() - release old fs-specific metadata on a page | |
3256 | * | |
3257 | * @page: the page which the kernel is trying to free | |
3258 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3259 | * | |
3260 | * The address_space is to try to release any data against the page | |
0e056eb5 | 3261 | * (presumably at page->private). If the release was successful, return '1'. |
cf9a2ae8 DH |
3262 | * Otherwise return zero. |
3263 | * | |
266cf658 DH |
3264 | * This may also be called if PG_fscache is set on a page, indicating that the |
3265 | * page is known to the local caching routines. | |
3266 | * | |
cf9a2ae8 | 3267 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3268 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3269 | * |
cf9a2ae8 DH |
3270 | */ |
3271 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3272 | { | |
3273 | struct address_space * const mapping = page->mapping; | |
3274 | ||
3275 | BUG_ON(!PageLocked(page)); | |
3276 | if (PageWriteback(page)) | |
3277 | return 0; | |
3278 | ||
3279 | if (mapping && mapping->a_ops->releasepage) | |
3280 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3281 | return try_to_free_buffers(page); | |
3282 | } | |
3283 | ||
3284 | EXPORT_SYMBOL(try_to_release_page); |