]>
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 | ||
91b0abe3 JW |
114 | static void page_cache_tree_delete(struct address_space *mapping, |
115 | struct page *page, void *shadow) | |
116 | { | |
c70b647d KS |
117 | int i, nr; |
118 | ||
119 | /* hugetlb pages are represented by one entry in the radix tree */ | |
120 | nr = PageHuge(page) ? 1 : hpage_nr_pages(page); | |
91b0abe3 | 121 | |
83929372 KS |
122 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
123 | VM_BUG_ON_PAGE(PageTail(page), page); | |
124 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 125 | |
83929372 | 126 | for (i = 0; i < nr; i++) { |
d3798ae8 JW |
127 | struct radix_tree_node *node; |
128 | void **slot; | |
129 | ||
b93b0163 | 130 | __radix_tree_lookup(&mapping->i_pages, page->index + i, |
d3798ae8 JW |
131 | &node, &slot); |
132 | ||
dbc446b8 | 133 | VM_BUG_ON_PAGE(!node && nr != 1, page); |
449dd698 | 134 | |
b93b0163 MW |
135 | radix_tree_clear_tags(&mapping->i_pages, node, slot); |
136 | __radix_tree_replace(&mapping->i_pages, node, slot, shadow, | |
c7df8ad2 | 137 | workingset_lookup_update(mapping)); |
449dd698 | 138 | } |
d3798ae8 | 139 | |
2300638b JK |
140 | page->mapping = NULL; |
141 | /* Leave page->index set: truncation lookup relies upon it */ | |
142 | ||
d3798ae8 JW |
143 | if (shadow) { |
144 | mapping->nrexceptional += nr; | |
145 | /* | |
146 | * Make sure the nrexceptional update is committed before | |
147 | * the nrpages update so that final truncate racing | |
148 | * with reclaim does not see both counters 0 at the | |
149 | * same time and miss a shadow entry. | |
150 | */ | |
151 | smp_wmb(); | |
152 | } | |
153 | mapping->nrpages -= nr; | |
91b0abe3 JW |
154 | } |
155 | ||
5ecc4d85 JK |
156 | static void unaccount_page_cache_page(struct address_space *mapping, |
157 | struct page *page) | |
1da177e4 | 158 | { |
5ecc4d85 | 159 | int nr; |
1da177e4 | 160 | |
c515e1fd DM |
161 | /* |
162 | * if we're uptodate, flush out into the cleancache, otherwise | |
163 | * invalidate any existing cleancache entries. We can't leave | |
164 | * stale data around in the cleancache once our page is gone | |
165 | */ | |
166 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
167 | cleancache_put_page(page); | |
168 | else | |
3167760f | 169 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 170 | |
83929372 | 171 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
172 | VM_BUG_ON_PAGE(page_mapped(page), page); |
173 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
174 | int mapcount; | |
175 | ||
176 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
177 | current->comm, page_to_pfn(page)); | |
178 | dump_page(page, "still mapped when deleted"); | |
179 | dump_stack(); | |
180 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
181 | ||
182 | mapcount = page_mapcount(page); | |
183 | if (mapping_exiting(mapping) && | |
184 | page_count(page) >= mapcount + 2) { | |
185 | /* | |
186 | * All vmas have already been torn down, so it's | |
187 | * a good bet that actually the page is unmapped, | |
188 | * and we'd prefer not to leak it: if we're wrong, | |
189 | * some other bad page check should catch it later. | |
190 | */ | |
191 | page_mapcount_reset(page); | |
6d061f9f | 192 | page_ref_sub(page, mapcount); |
06b241f3 HD |
193 | } |
194 | } | |
195 | ||
4165b9b4 | 196 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
197 | if (PageHuge(page)) |
198 | return; | |
09612fa6 | 199 | |
5ecc4d85 JK |
200 | nr = hpage_nr_pages(page); |
201 | ||
202 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); | |
203 | if (PageSwapBacked(page)) { | |
204 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); | |
205 | if (PageTransHuge(page)) | |
206 | __dec_node_page_state(page, NR_SHMEM_THPS); | |
207 | } else { | |
208 | VM_BUG_ON_PAGE(PageTransHuge(page), page); | |
800d8c63 | 209 | } |
5ecc4d85 JK |
210 | |
211 | /* | |
212 | * At this point page must be either written or cleaned by | |
213 | * truncate. Dirty page here signals a bug and loss of | |
214 | * unwritten data. | |
215 | * | |
216 | * This fixes dirty accounting after removing the page entirely | |
217 | * but leaves PageDirty set: it has no effect for truncated | |
218 | * page and anyway will be cleared before returning page into | |
219 | * buddy allocator. | |
220 | */ | |
221 | if (WARN_ON_ONCE(PageDirty(page))) | |
222 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
223 | } | |
224 | ||
225 | /* | |
226 | * Delete a page from the page cache and free it. Caller has to make | |
227 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 228 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
229 | */ |
230 | void __delete_from_page_cache(struct page *page, void *shadow) | |
231 | { | |
232 | struct address_space *mapping = page->mapping; | |
233 | ||
234 | trace_mm_filemap_delete_from_page_cache(page); | |
235 | ||
236 | unaccount_page_cache_page(mapping, page); | |
76253fbc | 237 | page_cache_tree_delete(mapping, page, shadow); |
1da177e4 LT |
238 | } |
239 | ||
59c66c5f JK |
240 | static void page_cache_free_page(struct address_space *mapping, |
241 | struct page *page) | |
242 | { | |
243 | void (*freepage)(struct page *); | |
244 | ||
245 | freepage = mapping->a_ops->freepage; | |
246 | if (freepage) | |
247 | freepage(page); | |
248 | ||
249 | if (PageTransHuge(page) && !PageHuge(page)) { | |
250 | page_ref_sub(page, HPAGE_PMD_NR); | |
251 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
252 | } else { | |
253 | put_page(page); | |
254 | } | |
255 | } | |
256 | ||
702cfbf9 MK |
257 | /** |
258 | * delete_from_page_cache - delete page from page cache | |
259 | * @page: the page which the kernel is trying to remove from page cache | |
260 | * | |
261 | * This must be called only on pages that have been verified to be in the page | |
262 | * cache and locked. It will never put the page into the free list, the caller | |
263 | * has a reference on the page. | |
264 | */ | |
265 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 266 | { |
83929372 | 267 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 268 | unsigned long flags; |
1da177e4 | 269 | |
cd7619d6 | 270 | BUG_ON(!PageLocked(page)); |
b93b0163 | 271 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 272 | __delete_from_page_cache(page, NULL); |
b93b0163 | 273 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 274 | |
59c66c5f | 275 | page_cache_free_page(mapping, page); |
97cecb5a MK |
276 | } |
277 | EXPORT_SYMBOL(delete_from_page_cache); | |
278 | ||
aa65c29c JK |
279 | /* |
280 | * page_cache_tree_delete_batch - delete several pages from page cache | |
281 | * @mapping: the mapping to which pages belong | |
282 | * @pvec: pagevec with pages to delete | |
283 | * | |
b93b0163 MW |
284 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
285 | * from the mapping. The function expects @pvec to be sorted by page index. | |
286 | * It tolerates holes in @pvec (mapping entries at those indices are not | |
aa65c29c | 287 | * modified). The function expects only THP head pages to be present in the |
b93b0163 MW |
288 | * @pvec and takes care to delete all corresponding tail pages from the |
289 | * mapping as well. | |
aa65c29c | 290 | * |
b93b0163 | 291 | * The function expects the i_pages lock to be held. |
aa65c29c JK |
292 | */ |
293 | static void | |
294 | page_cache_tree_delete_batch(struct address_space *mapping, | |
295 | struct pagevec *pvec) | |
296 | { | |
297 | struct radix_tree_iter iter; | |
298 | void **slot; | |
299 | int total_pages = 0; | |
300 | int i = 0, tail_pages = 0; | |
301 | struct page *page; | |
302 | pgoff_t start; | |
303 | ||
304 | start = pvec->pages[0]->index; | |
b93b0163 | 305 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
aa65c29c JK |
306 | if (i >= pagevec_count(pvec) && !tail_pages) |
307 | break; | |
308 | page = radix_tree_deref_slot_protected(slot, | |
b93b0163 | 309 | &mapping->i_pages.xa_lock); |
3159f943 | 310 | if (xa_is_value(page)) |
aa65c29c JK |
311 | continue; |
312 | if (!tail_pages) { | |
313 | /* | |
314 | * Some page got inserted in our range? Skip it. We | |
315 | * have our pages locked so they are protected from | |
316 | * being removed. | |
317 | */ | |
318 | if (page != pvec->pages[i]) | |
319 | continue; | |
320 | WARN_ON_ONCE(!PageLocked(page)); | |
321 | if (PageTransHuge(page) && !PageHuge(page)) | |
322 | tail_pages = HPAGE_PMD_NR - 1; | |
323 | page->mapping = NULL; | |
324 | /* | |
325 | * Leave page->index set: truncation lookup relies | |
326 | * upon it | |
327 | */ | |
328 | i++; | |
329 | } else { | |
330 | tail_pages--; | |
331 | } | |
b93b0163 MW |
332 | radix_tree_clear_tags(&mapping->i_pages, iter.node, slot); |
333 | __radix_tree_replace(&mapping->i_pages, iter.node, slot, NULL, | |
c7df8ad2 | 334 | workingset_lookup_update(mapping)); |
aa65c29c JK |
335 | total_pages++; |
336 | } | |
337 | mapping->nrpages -= total_pages; | |
338 | } | |
339 | ||
340 | void delete_from_page_cache_batch(struct address_space *mapping, | |
341 | struct pagevec *pvec) | |
342 | { | |
343 | int i; | |
344 | unsigned long flags; | |
345 | ||
346 | if (!pagevec_count(pvec)) | |
347 | return; | |
348 | ||
b93b0163 | 349 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
350 | for (i = 0; i < pagevec_count(pvec); i++) { |
351 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
352 | ||
353 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
354 | } | |
355 | page_cache_tree_delete_batch(mapping, pvec); | |
b93b0163 | 356 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
357 | |
358 | for (i = 0; i < pagevec_count(pvec); i++) | |
359 | page_cache_free_page(mapping, pvec->pages[i]); | |
360 | } | |
361 | ||
d72d9e2a | 362 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
363 | { |
364 | int ret = 0; | |
365 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
366 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
367 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 368 | ret = -ENOSPC; |
7fcbbaf1 JA |
369 | if (test_bit(AS_EIO, &mapping->flags) && |
370 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
371 | ret = -EIO; |
372 | return ret; | |
373 | } | |
d72d9e2a | 374 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 375 | |
76341cab JL |
376 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
377 | { | |
378 | /* Check for outstanding write errors */ | |
379 | if (test_bit(AS_EIO, &mapping->flags)) | |
380 | return -EIO; | |
381 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
382 | return -ENOSPC; | |
383 | return 0; | |
384 | } | |
385 | ||
1da177e4 | 386 | /** |
485bb99b | 387 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
388 | * @mapping: address space structure to write |
389 | * @start: offset in bytes where the range starts | |
469eb4d0 | 390 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 391 | * @sync_mode: enable synchronous operation |
1da177e4 | 392 | * |
485bb99b RD |
393 | * Start writeback against all of a mapping's dirty pages that lie |
394 | * within the byte offsets <start, end> inclusive. | |
395 | * | |
1da177e4 | 396 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 397 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
398 | * these two operations is that if a dirty page/buffer is encountered, it must |
399 | * be waited upon, and not just skipped over. | |
400 | */ | |
ebcf28e1 AM |
401 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
402 | loff_t end, int sync_mode) | |
1da177e4 LT |
403 | { |
404 | int ret; | |
405 | struct writeback_control wbc = { | |
406 | .sync_mode = sync_mode, | |
05fe478d | 407 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
408 | .range_start = start, |
409 | .range_end = end, | |
1da177e4 LT |
410 | }; |
411 | ||
412 | if (!mapping_cap_writeback_dirty(mapping)) | |
413 | return 0; | |
414 | ||
b16b1deb | 415 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 416 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 417 | wbc_detach_inode(&wbc); |
1da177e4 LT |
418 | return ret; |
419 | } | |
420 | ||
421 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
422 | int sync_mode) | |
423 | { | |
111ebb6e | 424 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
425 | } |
426 | ||
427 | int filemap_fdatawrite(struct address_space *mapping) | |
428 | { | |
429 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
430 | } | |
431 | EXPORT_SYMBOL(filemap_fdatawrite); | |
432 | ||
f4c0a0fd | 433 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 434 | loff_t end) |
1da177e4 LT |
435 | { |
436 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
437 | } | |
f4c0a0fd | 438 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 439 | |
485bb99b RD |
440 | /** |
441 | * filemap_flush - mostly a non-blocking flush | |
442 | * @mapping: target address_space | |
443 | * | |
1da177e4 LT |
444 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
445 | * purposes - I/O may not be started against all dirty pages. | |
446 | */ | |
447 | int filemap_flush(struct address_space *mapping) | |
448 | { | |
449 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
450 | } | |
451 | EXPORT_SYMBOL(filemap_flush); | |
452 | ||
7fc9e472 GR |
453 | /** |
454 | * filemap_range_has_page - check if a page exists in range. | |
455 | * @mapping: address space within which to check | |
456 | * @start_byte: offset in bytes where the range starts | |
457 | * @end_byte: offset in bytes where the range ends (inclusive) | |
458 | * | |
459 | * Find at least one page in the range supplied, usually used to check if | |
460 | * direct writing in this range will trigger a writeback. | |
461 | */ | |
462 | bool filemap_range_has_page(struct address_space *mapping, | |
463 | loff_t start_byte, loff_t end_byte) | |
464 | { | |
465 | pgoff_t index = start_byte >> PAGE_SHIFT; | |
466 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
f7b68046 | 467 | struct page *page; |
7fc9e472 GR |
468 | |
469 | if (end_byte < start_byte) | |
470 | return false; | |
471 | ||
472 | if (mapping->nrpages == 0) | |
473 | return false; | |
474 | ||
f7b68046 | 475 | if (!find_get_pages_range(mapping, &index, end, 1, &page)) |
7fc9e472 | 476 | return false; |
f7b68046 JK |
477 | put_page(page); |
478 | return true; | |
7fc9e472 GR |
479 | } |
480 | EXPORT_SYMBOL(filemap_range_has_page); | |
481 | ||
5e8fcc1a | 482 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 483 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 484 | { |
09cbfeaf KS |
485 | pgoff_t index = start_byte >> PAGE_SHIFT; |
486 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
487 | struct pagevec pvec; |
488 | int nr_pages; | |
1da177e4 | 489 | |
94004ed7 | 490 | if (end_byte < start_byte) |
5e8fcc1a | 491 | return; |
1da177e4 | 492 | |
86679820 | 493 | pagevec_init(&pvec); |
312e9d2f | 494 | while (index <= end) { |
1da177e4 LT |
495 | unsigned i; |
496 | ||
312e9d2f | 497 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 498 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
499 | if (!nr_pages) |
500 | break; | |
501 | ||
1da177e4 LT |
502 | for (i = 0; i < nr_pages; i++) { |
503 | struct page *page = pvec.pages[i]; | |
504 | ||
1da177e4 | 505 | wait_on_page_writeback(page); |
5e8fcc1a | 506 | ClearPageError(page); |
1da177e4 LT |
507 | } |
508 | pagevec_release(&pvec); | |
509 | cond_resched(); | |
510 | } | |
aa750fd7 JN |
511 | } |
512 | ||
513 | /** | |
514 | * filemap_fdatawait_range - wait for writeback to complete | |
515 | * @mapping: address space structure to wait for | |
516 | * @start_byte: offset in bytes where the range starts | |
517 | * @end_byte: offset in bytes where the range ends (inclusive) | |
518 | * | |
519 | * Walk the list of under-writeback pages of the given address space | |
520 | * in the given range and wait for all of them. Check error status of | |
521 | * the address space and return it. | |
522 | * | |
523 | * Since the error status of the address space is cleared by this function, | |
524 | * callers are responsible for checking the return value and handling and/or | |
525 | * reporting the error. | |
526 | */ | |
527 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
528 | loff_t end_byte) | |
529 | { | |
5e8fcc1a JL |
530 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
531 | return filemap_check_errors(mapping); | |
1da177e4 | 532 | } |
d3bccb6f JK |
533 | EXPORT_SYMBOL(filemap_fdatawait_range); |
534 | ||
a823e458 JL |
535 | /** |
536 | * file_fdatawait_range - wait for writeback to complete | |
537 | * @file: file pointing to address space structure to wait for | |
538 | * @start_byte: offset in bytes where the range starts | |
539 | * @end_byte: offset in bytes where the range ends (inclusive) | |
540 | * | |
541 | * Walk the list of under-writeback pages of the address space that file | |
542 | * refers to, in the given range and wait for all of them. Check error | |
543 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
544 | * | |
545 | * Since the error status of the file is advanced by this function, | |
546 | * callers are responsible for checking the return value and handling and/or | |
547 | * reporting the error. | |
548 | */ | |
549 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
550 | { | |
551 | struct address_space *mapping = file->f_mapping; | |
552 | ||
553 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
554 | return file_check_and_advance_wb_err(file); | |
555 | } | |
556 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 557 | |
aa750fd7 JN |
558 | /** |
559 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
560 | * @mapping: address space structure to wait for | |
561 | * | |
562 | * Walk the list of under-writeback pages of the given address space | |
563 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
564 | * does not clear error status of the address space. | |
565 | * | |
566 | * Use this function if callers don't handle errors themselves. Expected | |
567 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
568 | * fsfreeze(8) | |
569 | */ | |
76341cab | 570 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 571 | { |
ffb959bb | 572 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 573 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 574 | } |
76341cab | 575 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 576 | |
9326c9b2 | 577 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 578 | { |
9326c9b2 JL |
579 | return (!dax_mapping(mapping) && mapping->nrpages) || |
580 | (dax_mapping(mapping) && mapping->nrexceptional); | |
1da177e4 | 581 | } |
1da177e4 LT |
582 | |
583 | int filemap_write_and_wait(struct address_space *mapping) | |
584 | { | |
28fd1298 | 585 | int err = 0; |
1da177e4 | 586 | |
9326c9b2 | 587 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
588 | err = filemap_fdatawrite(mapping); |
589 | /* | |
590 | * Even if the above returned error, the pages may be | |
591 | * written partially (e.g. -ENOSPC), so we wait for it. | |
592 | * But the -EIO is special case, it may indicate the worst | |
593 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
594 | */ | |
595 | if (err != -EIO) { | |
596 | int err2 = filemap_fdatawait(mapping); | |
597 | if (!err) | |
598 | err = err2; | |
cbeaf951 JL |
599 | } else { |
600 | /* Clear any previously stored errors */ | |
601 | filemap_check_errors(mapping); | |
28fd1298 | 602 | } |
865ffef3 DM |
603 | } else { |
604 | err = filemap_check_errors(mapping); | |
1da177e4 | 605 | } |
28fd1298 | 606 | return err; |
1da177e4 | 607 | } |
28fd1298 | 608 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 609 | |
485bb99b RD |
610 | /** |
611 | * filemap_write_and_wait_range - write out & wait on a file range | |
612 | * @mapping: the address_space for the pages | |
613 | * @lstart: offset in bytes where the range starts | |
614 | * @lend: offset in bytes where the range ends (inclusive) | |
615 | * | |
469eb4d0 AM |
616 | * Write out and wait upon file offsets lstart->lend, inclusive. |
617 | * | |
0e056eb5 | 618 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 AM |
619 | * that this function can be used to write to the very end-of-file (end = -1). |
620 | */ | |
1da177e4 LT |
621 | int filemap_write_and_wait_range(struct address_space *mapping, |
622 | loff_t lstart, loff_t lend) | |
623 | { | |
28fd1298 | 624 | int err = 0; |
1da177e4 | 625 | |
9326c9b2 | 626 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
627 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
628 | WB_SYNC_ALL); | |
629 | /* See comment of filemap_write_and_wait() */ | |
630 | if (err != -EIO) { | |
94004ed7 CH |
631 | int err2 = filemap_fdatawait_range(mapping, |
632 | lstart, lend); | |
28fd1298 OH |
633 | if (!err) |
634 | err = err2; | |
cbeaf951 JL |
635 | } else { |
636 | /* Clear any previously stored errors */ | |
637 | filemap_check_errors(mapping); | |
28fd1298 | 638 | } |
865ffef3 DM |
639 | } else { |
640 | err = filemap_check_errors(mapping); | |
1da177e4 | 641 | } |
28fd1298 | 642 | return err; |
1da177e4 | 643 | } |
f6995585 | 644 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 645 | |
5660e13d JL |
646 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
647 | { | |
3acdfd28 | 648 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
649 | |
650 | trace_filemap_set_wb_err(mapping, eseq); | |
651 | } | |
652 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
653 | ||
654 | /** | |
655 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
656 | * and advance wb_err to current one | |
657 | * @file: struct file on which the error is being reported | |
658 | * | |
659 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
660 | * want to report any writeback errors that occurred since the last fsync (or | |
661 | * since the file was opened if there haven't been any). | |
662 | * | |
663 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
664 | * then just quickly return 0. The file is all caught up. | |
665 | * | |
666 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
667 | * it and try to swap it into place. If it works, or another task beat us | |
668 | * to it with the new value, then update the f_wb_err and return the error | |
669 | * portion. The error at this point must be reported via proper channels | |
670 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
671 | * | |
672 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
673 | * value is protected by the f_lock since we must ensure that it reflects | |
674 | * the latest value swapped in for this file descriptor. | |
675 | */ | |
676 | int file_check_and_advance_wb_err(struct file *file) | |
677 | { | |
678 | int err = 0; | |
679 | errseq_t old = READ_ONCE(file->f_wb_err); | |
680 | struct address_space *mapping = file->f_mapping; | |
681 | ||
682 | /* Locklessly handle the common case where nothing has changed */ | |
683 | if (errseq_check(&mapping->wb_err, old)) { | |
684 | /* Something changed, must use slow path */ | |
685 | spin_lock(&file->f_lock); | |
686 | old = file->f_wb_err; | |
687 | err = errseq_check_and_advance(&mapping->wb_err, | |
688 | &file->f_wb_err); | |
689 | trace_file_check_and_advance_wb_err(file, old); | |
690 | spin_unlock(&file->f_lock); | |
691 | } | |
f4e222c5 JL |
692 | |
693 | /* | |
694 | * We're mostly using this function as a drop in replacement for | |
695 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
696 | * that the legacy code would have had on these flags. | |
697 | */ | |
698 | clear_bit(AS_EIO, &mapping->flags); | |
699 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
700 | return err; |
701 | } | |
702 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
703 | ||
704 | /** | |
705 | * file_write_and_wait_range - write out & wait on a file range | |
706 | * @file: file pointing to address_space with pages | |
707 | * @lstart: offset in bytes where the range starts | |
708 | * @lend: offset in bytes where the range ends (inclusive) | |
709 | * | |
710 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
711 | * | |
712 | * Note that @lend is inclusive (describes the last byte to be written) so | |
713 | * that this function can be used to write to the very end-of-file (end = -1). | |
714 | * | |
715 | * After writing out and waiting on the data, we check and advance the | |
716 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
717 | */ | |
718 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
719 | { | |
720 | int err = 0, err2; | |
721 | struct address_space *mapping = file->f_mapping; | |
722 | ||
9326c9b2 | 723 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
724 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
725 | WB_SYNC_ALL); | |
726 | /* See comment of filemap_write_and_wait() */ | |
727 | if (err != -EIO) | |
728 | __filemap_fdatawait_range(mapping, lstart, lend); | |
729 | } | |
730 | err2 = file_check_and_advance_wb_err(file); | |
731 | if (!err) | |
732 | err = err2; | |
733 | return err; | |
734 | } | |
735 | EXPORT_SYMBOL(file_write_and_wait_range); | |
736 | ||
ef6a3c63 MS |
737 | /** |
738 | * replace_page_cache_page - replace a pagecache page with a new one | |
739 | * @old: page to be replaced | |
740 | * @new: page to replace with | |
741 | * @gfp_mask: allocation mode | |
742 | * | |
743 | * This function replaces a page in the pagecache with a new one. On | |
744 | * success it acquires the pagecache reference for the new page and | |
745 | * drops it for the old page. Both the old and new pages must be | |
746 | * locked. This function does not add the new page to the LRU, the | |
747 | * caller must do that. | |
748 | * | |
74d60958 | 749 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 MS |
750 | */ |
751 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
752 | { | |
74d60958 MW |
753 | struct address_space *mapping = old->mapping; |
754 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
755 | pgoff_t offset = old->index; | |
756 | XA_STATE(xas, &mapping->i_pages, offset); | |
757 | unsigned long flags; | |
ef6a3c63 | 758 | |
309381fe SL |
759 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
760 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
761 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 762 | |
74d60958 MW |
763 | get_page(new); |
764 | new->mapping = mapping; | |
765 | new->index = offset; | |
ef6a3c63 | 766 | |
74d60958 MW |
767 | xas_lock_irqsave(&xas, flags); |
768 | xas_store(&xas, new); | |
ef6a3c63 | 769 | |
74d60958 MW |
770 | old->mapping = NULL; |
771 | /* hugetlb pages do not participate in page cache accounting. */ | |
772 | if (!PageHuge(old)) | |
773 | __dec_node_page_state(new, NR_FILE_PAGES); | |
774 | if (!PageHuge(new)) | |
775 | __inc_node_page_state(new, NR_FILE_PAGES); | |
776 | if (PageSwapBacked(old)) | |
777 | __dec_node_page_state(new, NR_SHMEM); | |
778 | if (PageSwapBacked(new)) | |
779 | __inc_node_page_state(new, NR_SHMEM); | |
780 | xas_unlock_irqrestore(&xas, flags); | |
781 | mem_cgroup_migrate(old, new); | |
782 | if (freepage) | |
783 | freepage(old); | |
784 | put_page(old); | |
ef6a3c63 | 785 | |
74d60958 | 786 | return 0; |
ef6a3c63 MS |
787 | } |
788 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
789 | ||
a528910e JW |
790 | static int __add_to_page_cache_locked(struct page *page, |
791 | struct address_space *mapping, | |
792 | pgoff_t offset, gfp_t gfp_mask, | |
793 | void **shadowp) | |
1da177e4 | 794 | { |
74d60958 | 795 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 JW |
796 | int huge = PageHuge(page); |
797 | struct mem_cgroup *memcg; | |
e286781d | 798 | int error; |
74d60958 | 799 | void *old; |
e286781d | 800 | |
309381fe SL |
801 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
802 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 803 | mapping_set_update(&xas, mapping); |
e286781d | 804 | |
00501b53 JW |
805 | if (!huge) { |
806 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 807 | gfp_mask, &memcg, false); |
00501b53 JW |
808 | if (error) |
809 | return error; | |
810 | } | |
1da177e4 | 811 | |
09cbfeaf | 812 | get_page(page); |
66a0c8ee KS |
813 | page->mapping = mapping; |
814 | page->index = offset; | |
815 | ||
74d60958 MW |
816 | do { |
817 | xas_lock_irq(&xas); | |
818 | old = xas_load(&xas); | |
819 | if (old && !xa_is_value(old)) | |
820 | xas_set_err(&xas, -EEXIST); | |
821 | xas_store(&xas, page); | |
822 | if (xas_error(&xas)) | |
823 | goto unlock; | |
824 | ||
825 | if (xa_is_value(old)) { | |
826 | mapping->nrexceptional--; | |
827 | if (shadowp) | |
828 | *shadowp = old; | |
829 | } | |
830 | mapping->nrpages++; | |
831 | ||
832 | /* hugetlb pages do not participate in page cache accounting */ | |
833 | if (!huge) | |
834 | __inc_node_page_state(page, NR_FILE_PAGES); | |
835 | unlock: | |
836 | xas_unlock_irq(&xas); | |
837 | } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK)); | |
838 | ||
839 | if (xas_error(&xas)) | |
840 | goto error; | |
4165b9b4 | 841 | |
00501b53 | 842 | if (!huge) |
f627c2f5 | 843 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
844 | trace_mm_filemap_add_to_page_cache(page); |
845 | return 0; | |
74d60958 | 846 | error: |
66a0c8ee KS |
847 | page->mapping = NULL; |
848 | /* Leave page->index set: truncation relies upon it */ | |
00501b53 | 849 | if (!huge) |
f627c2f5 | 850 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 851 | put_page(page); |
74d60958 | 852 | return xas_error(&xas); |
1da177e4 | 853 | } |
a528910e JW |
854 | |
855 | /** | |
856 | * add_to_page_cache_locked - add a locked page to the pagecache | |
857 | * @page: page to add | |
858 | * @mapping: the page's address_space | |
859 | * @offset: page index | |
860 | * @gfp_mask: page allocation mode | |
861 | * | |
862 | * This function is used to add a page to the pagecache. It must be locked. | |
863 | * This function does not add the page to the LRU. The caller must do that. | |
864 | */ | |
865 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
866 | pgoff_t offset, gfp_t gfp_mask) | |
867 | { | |
868 | return __add_to_page_cache_locked(page, mapping, offset, | |
869 | gfp_mask, NULL); | |
870 | } | |
e286781d | 871 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
872 | |
873 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 874 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 875 | { |
a528910e | 876 | void *shadow = NULL; |
4f98a2fe RR |
877 | int ret; |
878 | ||
48c935ad | 879 | __SetPageLocked(page); |
a528910e JW |
880 | ret = __add_to_page_cache_locked(page, mapping, offset, |
881 | gfp_mask, &shadow); | |
882 | if (unlikely(ret)) | |
48c935ad | 883 | __ClearPageLocked(page); |
a528910e JW |
884 | else { |
885 | /* | |
886 | * The page might have been evicted from cache only | |
887 | * recently, in which case it should be activated like | |
888 | * any other repeatedly accessed page. | |
f0281a00 RR |
889 | * The exception is pages getting rewritten; evicting other |
890 | * data from the working set, only to cache data that will | |
891 | * get overwritten with something else, is a waste of memory. | |
a528910e | 892 | */ |
f0281a00 RR |
893 | if (!(gfp_mask & __GFP_WRITE) && |
894 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
895 | SetPageActive(page); |
896 | workingset_activation(page); | |
897 | } else | |
898 | ClearPageActive(page); | |
899 | lru_cache_add(page); | |
900 | } | |
1da177e4 LT |
901 | return ret; |
902 | } | |
18bc0bbd | 903 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 904 | |
44110fe3 | 905 | #ifdef CONFIG_NUMA |
2ae88149 | 906 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 907 | { |
c0ff7453 MX |
908 | int n; |
909 | struct page *page; | |
910 | ||
44110fe3 | 911 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
912 | unsigned int cpuset_mems_cookie; |
913 | do { | |
d26914d1 | 914 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 915 | n = cpuset_mem_spread_node(); |
96db800f | 916 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 917 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 918 | |
c0ff7453 | 919 | return page; |
44110fe3 | 920 | } |
2ae88149 | 921 | return alloc_pages(gfp, 0); |
44110fe3 | 922 | } |
2ae88149 | 923 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
924 | #endif |
925 | ||
1da177e4 LT |
926 | /* |
927 | * In order to wait for pages to become available there must be | |
928 | * waitqueues associated with pages. By using a hash table of | |
929 | * waitqueues where the bucket discipline is to maintain all | |
930 | * waiters on the same queue and wake all when any of the pages | |
931 | * become available, and for the woken contexts to check to be | |
932 | * sure the appropriate page became available, this saves space | |
933 | * at a cost of "thundering herd" phenomena during rare hash | |
934 | * collisions. | |
935 | */ | |
62906027 NP |
936 | #define PAGE_WAIT_TABLE_BITS 8 |
937 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
938 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
939 | ||
940 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 941 | { |
62906027 | 942 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 943 | } |
1da177e4 | 944 | |
62906027 | 945 | void __init pagecache_init(void) |
1da177e4 | 946 | { |
62906027 | 947 | int i; |
1da177e4 | 948 | |
62906027 NP |
949 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
950 | init_waitqueue_head(&page_wait_table[i]); | |
951 | ||
952 | page_writeback_init(); | |
1da177e4 | 953 | } |
1da177e4 | 954 | |
3510ca20 | 955 | /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ |
62906027 NP |
956 | struct wait_page_key { |
957 | struct page *page; | |
958 | int bit_nr; | |
959 | int page_match; | |
960 | }; | |
961 | ||
962 | struct wait_page_queue { | |
963 | struct page *page; | |
964 | int bit_nr; | |
ac6424b9 | 965 | wait_queue_entry_t wait; |
62906027 NP |
966 | }; |
967 | ||
ac6424b9 | 968 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 969 | { |
62906027 NP |
970 | struct wait_page_key *key = arg; |
971 | struct wait_page_queue *wait_page | |
972 | = container_of(wait, struct wait_page_queue, wait); | |
973 | ||
974 | if (wait_page->page != key->page) | |
975 | return 0; | |
976 | key->page_match = 1; | |
f62e00cc | 977 | |
62906027 NP |
978 | if (wait_page->bit_nr != key->bit_nr) |
979 | return 0; | |
3510ca20 LT |
980 | |
981 | /* Stop walking if it's locked */ | |
62906027 | 982 | if (test_bit(key->bit_nr, &key->page->flags)) |
3510ca20 | 983 | return -1; |
f62e00cc | 984 | |
62906027 | 985 | return autoremove_wake_function(wait, mode, sync, key); |
f62e00cc KM |
986 | } |
987 | ||
74d81bfa | 988 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 989 | { |
62906027 NP |
990 | wait_queue_head_t *q = page_waitqueue(page); |
991 | struct wait_page_key key; | |
992 | unsigned long flags; | |
11a19c7b | 993 | wait_queue_entry_t bookmark; |
cbbce822 | 994 | |
62906027 NP |
995 | key.page = page; |
996 | key.bit_nr = bit_nr; | |
997 | key.page_match = 0; | |
998 | ||
11a19c7b TC |
999 | bookmark.flags = 0; |
1000 | bookmark.private = NULL; | |
1001 | bookmark.func = NULL; | |
1002 | INIT_LIST_HEAD(&bookmark.entry); | |
1003 | ||
62906027 | 1004 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1005 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1006 | ||
1007 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1008 | /* | |
1009 | * Take a breather from holding the lock, | |
1010 | * allow pages that finish wake up asynchronously | |
1011 | * to acquire the lock and remove themselves | |
1012 | * from wait queue | |
1013 | */ | |
1014 | spin_unlock_irqrestore(&q->lock, flags); | |
1015 | cpu_relax(); | |
1016 | spin_lock_irqsave(&q->lock, flags); | |
1017 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1018 | } | |
1019 | ||
62906027 NP |
1020 | /* |
1021 | * It is possible for other pages to have collided on the waitqueue | |
1022 | * hash, so in that case check for a page match. That prevents a long- | |
1023 | * term waiter | |
1024 | * | |
1025 | * It is still possible to miss a case here, when we woke page waiters | |
1026 | * and removed them from the waitqueue, but there are still other | |
1027 | * page waiters. | |
1028 | */ | |
1029 | if (!waitqueue_active(q) || !key.page_match) { | |
1030 | ClearPageWaiters(page); | |
1031 | /* | |
1032 | * It's possible to miss clearing Waiters here, when we woke | |
1033 | * our page waiters, but the hashed waitqueue has waiters for | |
1034 | * other pages on it. | |
1035 | * | |
1036 | * That's okay, it's a rare case. The next waker will clear it. | |
1037 | */ | |
1038 | } | |
1039 | spin_unlock_irqrestore(&q->lock, flags); | |
1040 | } | |
74d81bfa NP |
1041 | |
1042 | static void wake_up_page(struct page *page, int bit) | |
1043 | { | |
1044 | if (!PageWaiters(page)) | |
1045 | return; | |
1046 | wake_up_page_bit(page, bit); | |
1047 | } | |
62906027 NP |
1048 | |
1049 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, | |
1050 | struct page *page, int bit_nr, int state, bool lock) | |
1051 | { | |
1052 | struct wait_page_queue wait_page; | |
ac6424b9 | 1053 | wait_queue_entry_t *wait = &wait_page.wait; |
62906027 NP |
1054 | int ret = 0; |
1055 | ||
1056 | init_wait(wait); | |
3510ca20 | 1057 | wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0; |
62906027 NP |
1058 | wait->func = wake_page_function; |
1059 | wait_page.page = page; | |
1060 | wait_page.bit_nr = bit_nr; | |
1061 | ||
1062 | for (;;) { | |
1063 | spin_lock_irq(&q->lock); | |
1064 | ||
2055da97 | 1065 | if (likely(list_empty(&wait->entry))) { |
3510ca20 | 1066 | __add_wait_queue_entry_tail(q, wait); |
62906027 NP |
1067 | SetPageWaiters(page); |
1068 | } | |
1069 | ||
1070 | set_current_state(state); | |
1071 | ||
1072 | spin_unlock_irq(&q->lock); | |
1073 | ||
1074 | if (likely(test_bit(bit_nr, &page->flags))) { | |
1075 | io_schedule(); | |
62906027 NP |
1076 | } |
1077 | ||
1078 | if (lock) { | |
1079 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) | |
1080 | break; | |
1081 | } else { | |
1082 | if (!test_bit(bit_nr, &page->flags)) | |
1083 | break; | |
1084 | } | |
a8b169af LT |
1085 | |
1086 | if (unlikely(signal_pending_state(state, current))) { | |
1087 | ret = -EINTR; | |
1088 | break; | |
1089 | } | |
62906027 NP |
1090 | } |
1091 | ||
1092 | finish_wait(q, wait); | |
1093 | ||
1094 | /* | |
1095 | * A signal could leave PageWaiters set. Clearing it here if | |
1096 | * !waitqueue_active would be possible (by open-coding finish_wait), | |
1097 | * but still fail to catch it in the case of wait hash collision. We | |
1098 | * already can fail to clear wait hash collision cases, so don't | |
1099 | * bother with signals either. | |
1100 | */ | |
1101 | ||
1102 | return ret; | |
1103 | } | |
1104 | ||
1105 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1106 | { | |
1107 | wait_queue_head_t *q = page_waitqueue(page); | |
1108 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false); | |
1109 | } | |
1110 | EXPORT_SYMBOL(wait_on_page_bit); | |
1111 | ||
1112 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1113 | { | |
1114 | wait_queue_head_t *q = page_waitqueue(page); | |
1115 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false); | |
cbbce822 | 1116 | } |
4343d008 | 1117 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1118 | |
385e1ca5 DH |
1119 | /** |
1120 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1121 | * @page: Page defining the wait queue of interest |
1122 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1123 | * |
1124 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1125 | */ | |
ac6424b9 | 1126 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1127 | { |
1128 | wait_queue_head_t *q = page_waitqueue(page); | |
1129 | unsigned long flags; | |
1130 | ||
1131 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1132 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1133 | SetPageWaiters(page); |
385e1ca5 DH |
1134 | spin_unlock_irqrestore(&q->lock, flags); |
1135 | } | |
1136 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1137 | ||
b91e1302 LT |
1138 | #ifndef clear_bit_unlock_is_negative_byte |
1139 | ||
1140 | /* | |
1141 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1142 | * | |
1143 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1144 | * test the sign bit at the same time. But if the architecture does | |
1145 | * not support that special operation, we just do this all by hand | |
1146 | * instead. | |
1147 | * | |
1148 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
1149 | * being cleared, but a memory barrier should be unneccssary since it is | |
1150 | * in the same byte as PG_locked. | |
1151 | */ | |
1152 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1153 | { | |
1154 | clear_bit_unlock(nr, mem); | |
1155 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1156 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1157 | } |
1158 | ||
1159 | #endif | |
1160 | ||
1da177e4 | 1161 | /** |
485bb99b | 1162 | * unlock_page - unlock a locked page |
1da177e4 LT |
1163 | * @page: the page |
1164 | * | |
1165 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1166 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1167 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1168 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1169 | * | |
b91e1302 LT |
1170 | * Note that this depends on PG_waiters being the sign bit in the byte |
1171 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1172 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1173 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1174 | * test the sign bit). | |
1da177e4 | 1175 | */ |
920c7a5d | 1176 | void unlock_page(struct page *page) |
1da177e4 | 1177 | { |
b91e1302 | 1178 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1179 | page = compound_head(page); |
309381fe | 1180 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1181 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1182 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1183 | } |
1184 | EXPORT_SYMBOL(unlock_page); | |
1185 | ||
485bb99b RD |
1186 | /** |
1187 | * end_page_writeback - end writeback against a page | |
1188 | * @page: the page | |
1da177e4 LT |
1189 | */ |
1190 | void end_page_writeback(struct page *page) | |
1191 | { | |
888cf2db MG |
1192 | /* |
1193 | * TestClearPageReclaim could be used here but it is an atomic | |
1194 | * operation and overkill in this particular case. Failing to | |
1195 | * shuffle a page marked for immediate reclaim is too mild to | |
1196 | * justify taking an atomic operation penalty at the end of | |
1197 | * ever page writeback. | |
1198 | */ | |
1199 | if (PageReclaim(page)) { | |
1200 | ClearPageReclaim(page); | |
ac6aadb2 | 1201 | rotate_reclaimable_page(page); |
888cf2db | 1202 | } |
ac6aadb2 MS |
1203 | |
1204 | if (!test_clear_page_writeback(page)) | |
1205 | BUG(); | |
1206 | ||
4e857c58 | 1207 | smp_mb__after_atomic(); |
1da177e4 LT |
1208 | wake_up_page(page, PG_writeback); |
1209 | } | |
1210 | EXPORT_SYMBOL(end_page_writeback); | |
1211 | ||
57d99845 MW |
1212 | /* |
1213 | * After completing I/O on a page, call this routine to update the page | |
1214 | * flags appropriately | |
1215 | */ | |
c11f0c0b | 1216 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1217 | { |
c11f0c0b | 1218 | if (!is_write) { |
57d99845 MW |
1219 | if (!err) { |
1220 | SetPageUptodate(page); | |
1221 | } else { | |
1222 | ClearPageUptodate(page); | |
1223 | SetPageError(page); | |
1224 | } | |
1225 | unlock_page(page); | |
abf54548 | 1226 | } else { |
57d99845 | 1227 | if (err) { |
dd8416c4 MK |
1228 | struct address_space *mapping; |
1229 | ||
57d99845 | 1230 | SetPageError(page); |
dd8416c4 MK |
1231 | mapping = page_mapping(page); |
1232 | if (mapping) | |
1233 | mapping_set_error(mapping, err); | |
57d99845 MW |
1234 | } |
1235 | end_page_writeback(page); | |
1236 | } | |
1237 | } | |
1238 | EXPORT_SYMBOL_GPL(page_endio); | |
1239 | ||
485bb99b RD |
1240 | /** |
1241 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1242 | * @__page: the page to lock |
1da177e4 | 1243 | */ |
62906027 | 1244 | void __lock_page(struct page *__page) |
1da177e4 | 1245 | { |
62906027 NP |
1246 | struct page *page = compound_head(__page); |
1247 | wait_queue_head_t *q = page_waitqueue(page); | |
1248 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true); | |
1da177e4 LT |
1249 | } |
1250 | EXPORT_SYMBOL(__lock_page); | |
1251 | ||
62906027 | 1252 | int __lock_page_killable(struct page *__page) |
2687a356 | 1253 | { |
62906027 NP |
1254 | struct page *page = compound_head(__page); |
1255 | wait_queue_head_t *q = page_waitqueue(page); | |
1256 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true); | |
2687a356 | 1257 | } |
18bc0bbd | 1258 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1259 | |
9a95f3cf PC |
1260 | /* |
1261 | * Return values: | |
1262 | * 1 - page is locked; mmap_sem is still held. | |
1263 | * 0 - page is not locked. | |
1264 | * mmap_sem has been released (up_read()), unless flags had both | |
1265 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
1266 | * which case mmap_sem is still held. | |
1267 | * | |
1268 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
1269 | * with the page locked and the mmap_sem unperturbed. | |
1270 | */ | |
d065bd81 ML |
1271 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1272 | unsigned int flags) | |
1273 | { | |
37b23e05 KM |
1274 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
1275 | /* | |
1276 | * CAUTION! In this case, mmap_sem is not released | |
1277 | * even though return 0. | |
1278 | */ | |
1279 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1280 | return 0; | |
1281 | ||
1282 | up_read(&mm->mmap_sem); | |
1283 | if (flags & FAULT_FLAG_KILLABLE) | |
1284 | wait_on_page_locked_killable(page); | |
1285 | else | |
318b275f | 1286 | wait_on_page_locked(page); |
d065bd81 | 1287 | return 0; |
37b23e05 KM |
1288 | } else { |
1289 | if (flags & FAULT_FLAG_KILLABLE) { | |
1290 | int ret; | |
1291 | ||
1292 | ret = __lock_page_killable(page); | |
1293 | if (ret) { | |
1294 | up_read(&mm->mmap_sem); | |
1295 | return 0; | |
1296 | } | |
1297 | } else | |
1298 | __lock_page(page); | |
1299 | return 1; | |
d065bd81 ML |
1300 | } |
1301 | } | |
1302 | ||
e7b563bb | 1303 | /** |
0d3f9296 MW |
1304 | * page_cache_next_miss() - Find the next gap in the page cache. |
1305 | * @mapping: Mapping. | |
1306 | * @index: Index. | |
1307 | * @max_scan: Maximum range to search. | |
e7b563bb | 1308 | * |
0d3f9296 MW |
1309 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1310 | * gap with the lowest index. | |
e7b563bb | 1311 | * |
0d3f9296 MW |
1312 | * This function may be called under the rcu_read_lock. However, this will |
1313 | * not atomically search a snapshot of the cache at a single point in time. | |
1314 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1315 | * created at index 10, page_cache_next_miss covering both indices may | |
1316 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1317 | * |
0d3f9296 MW |
1318 | * Return: The index of the gap if found, otherwise an index outside the |
1319 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1320 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1321 | */ |
0d3f9296 | 1322 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1323 | pgoff_t index, unsigned long max_scan) |
1324 | { | |
0d3f9296 | 1325 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1326 | |
0d3f9296 MW |
1327 | while (max_scan--) { |
1328 | void *entry = xas_next(&xas); | |
1329 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1330 | break; |
0d3f9296 | 1331 | if (xas.xa_index == 0) |
e7b563bb JW |
1332 | break; |
1333 | } | |
1334 | ||
0d3f9296 | 1335 | return xas.xa_index; |
e7b563bb | 1336 | } |
0d3f9296 | 1337 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1338 | |
1339 | /** | |
0d3f9296 MW |
1340 | * page_cache_prev_miss() - Find the next gap in the page cache. |
1341 | * @mapping: Mapping. | |
1342 | * @index: Index. | |
1343 | * @max_scan: Maximum range to search. | |
e7b563bb | 1344 | * |
0d3f9296 MW |
1345 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1346 | * gap with the highest index. | |
e7b563bb | 1347 | * |
0d3f9296 MW |
1348 | * This function may be called under the rcu_read_lock. However, this will |
1349 | * not atomically search a snapshot of the cache at a single point in time. | |
1350 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1351 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1352 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1353 | * |
0d3f9296 MW |
1354 | * Return: The index of the gap if found, otherwise an index outside the |
1355 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1356 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1357 | */ |
0d3f9296 | 1358 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1359 | pgoff_t index, unsigned long max_scan) |
1360 | { | |
0d3f9296 | 1361 | XA_STATE(xas, &mapping->i_pages, index); |
0cd6144a | 1362 | |
0d3f9296 MW |
1363 | while (max_scan--) { |
1364 | void *entry = xas_prev(&xas); | |
1365 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1366 | break; |
0d3f9296 | 1367 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1368 | break; |
1369 | } | |
1370 | ||
0d3f9296 | 1371 | return xas.xa_index; |
e7b563bb | 1372 | } |
0d3f9296 | 1373 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1374 | |
485bb99b | 1375 | /** |
0cd6144a | 1376 | * find_get_entry - find and get a page cache entry |
485bb99b | 1377 | * @mapping: the address_space to search |
0cd6144a JW |
1378 | * @offset: the page cache index |
1379 | * | |
1380 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1381 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1382 | * |
139b6a6f JW |
1383 | * If the slot holds a shadow entry of a previously evicted page, or a |
1384 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1385 | * |
1386 | * Otherwise, %NULL is returned. | |
1da177e4 | 1387 | */ |
0cd6144a | 1388 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1389 | { |
a60637c8 | 1390 | void **pagep; |
83929372 | 1391 | struct page *head, *page; |
1da177e4 | 1392 | |
a60637c8 NP |
1393 | rcu_read_lock(); |
1394 | repeat: | |
1395 | page = NULL; | |
b93b0163 | 1396 | pagep = radix_tree_lookup_slot(&mapping->i_pages, offset); |
a60637c8 NP |
1397 | if (pagep) { |
1398 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1399 | if (unlikely(!page)) |
1400 | goto out; | |
a2c16d6c | 1401 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1402 | if (radix_tree_deref_retry(page)) |
1403 | goto repeat; | |
1404 | /* | |
139b6a6f JW |
1405 | * A shadow entry of a recently evicted page, |
1406 | * or a swap entry from shmem/tmpfs. Return | |
1407 | * it without attempting to raise page count. | |
8079b1c8 HD |
1408 | */ |
1409 | goto out; | |
a2c16d6c | 1410 | } |
83929372 KS |
1411 | |
1412 | head = compound_head(page); | |
1413 | if (!page_cache_get_speculative(head)) | |
1414 | goto repeat; | |
1415 | ||
1416 | /* The page was split under us? */ | |
1417 | if (compound_head(page) != head) { | |
1418 | put_page(head); | |
a60637c8 | 1419 | goto repeat; |
83929372 | 1420 | } |
a60637c8 NP |
1421 | |
1422 | /* | |
1423 | * Has the page moved? | |
1424 | * This is part of the lockless pagecache protocol. See | |
1425 | * include/linux/pagemap.h for details. | |
1426 | */ | |
1427 | if (unlikely(page != *pagep)) { | |
83929372 | 1428 | put_page(head); |
a60637c8 NP |
1429 | goto repeat; |
1430 | } | |
1431 | } | |
27d20fdd | 1432 | out: |
a60637c8 NP |
1433 | rcu_read_unlock(); |
1434 | ||
1da177e4 LT |
1435 | return page; |
1436 | } | |
0cd6144a | 1437 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1438 | |
0cd6144a JW |
1439 | /** |
1440 | * find_lock_entry - locate, pin and lock a page cache entry | |
1441 | * @mapping: the address_space to search | |
1442 | * @offset: the page cache index | |
1443 | * | |
1444 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1445 | * page cache page, it is returned locked and with an increased | |
1446 | * refcount. | |
1447 | * | |
139b6a6f JW |
1448 | * If the slot holds a shadow entry of a previously evicted page, or a |
1449 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1450 | * |
1451 | * Otherwise, %NULL is returned. | |
1452 | * | |
1453 | * find_lock_entry() may sleep. | |
1454 | */ | |
1455 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1456 | { |
1457 | struct page *page; | |
1458 | ||
1da177e4 | 1459 | repeat: |
0cd6144a | 1460 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1461 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1462 | lock_page(page); |
1463 | /* Has the page been truncated? */ | |
83929372 | 1464 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1465 | unlock_page(page); |
09cbfeaf | 1466 | put_page(page); |
a60637c8 | 1467 | goto repeat; |
1da177e4 | 1468 | } |
83929372 | 1469 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1470 | } |
1da177e4 LT |
1471 | return page; |
1472 | } | |
0cd6144a JW |
1473 | EXPORT_SYMBOL(find_lock_entry); |
1474 | ||
1475 | /** | |
2457aec6 | 1476 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1477 | * @mapping: the address_space to search |
1478 | * @offset: the page index | |
2457aec6 | 1479 | * @fgp_flags: PCG flags |
45f87de5 | 1480 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1481 | * |
2457aec6 | 1482 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1483 | * |
75325189 | 1484 | * PCG flags modify how the page is returned. |
0cd6144a | 1485 | * |
0e056eb5 MCC |
1486 | * @fgp_flags can be: |
1487 | * | |
1488 | * - FGP_ACCESSED: the page will be marked accessed | |
1489 | * - FGP_LOCK: Page is return locked | |
1490 | * - FGP_CREAT: If page is not present then a new page is allocated using | |
1491 | * @gfp_mask and added to the page cache and the VM's LRU | |
1492 | * list. The page is returned locked and with an increased | |
1493 | * refcount. Otherwise, NULL is returned. | |
1da177e4 | 1494 | * |
2457aec6 MG |
1495 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1496 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1497 | * |
2457aec6 | 1498 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1499 | */ |
2457aec6 | 1500 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1501 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1502 | { |
eb2be189 | 1503 | struct page *page; |
2457aec6 | 1504 | |
1da177e4 | 1505 | repeat: |
2457aec6 | 1506 | page = find_get_entry(mapping, offset); |
3159f943 | 1507 | if (xa_is_value(page)) |
2457aec6 MG |
1508 | page = NULL; |
1509 | if (!page) | |
1510 | goto no_page; | |
1511 | ||
1512 | if (fgp_flags & FGP_LOCK) { | |
1513 | if (fgp_flags & FGP_NOWAIT) { | |
1514 | if (!trylock_page(page)) { | |
09cbfeaf | 1515 | put_page(page); |
2457aec6 MG |
1516 | return NULL; |
1517 | } | |
1518 | } else { | |
1519 | lock_page(page); | |
1520 | } | |
1521 | ||
1522 | /* Has the page been truncated? */ | |
1523 | if (unlikely(page->mapping != mapping)) { | |
1524 | unlock_page(page); | |
09cbfeaf | 1525 | put_page(page); |
2457aec6 MG |
1526 | goto repeat; |
1527 | } | |
1528 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1529 | } | |
1530 | ||
1531 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1532 | mark_page_accessed(page); | |
1533 | ||
1534 | no_page: | |
1535 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1536 | int err; | |
1537 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1538 | gfp_mask |= __GFP_WRITE; |
1539 | if (fgp_flags & FGP_NOFS) | |
1540 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1541 | |
45f87de5 | 1542 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1543 | if (!page) |
1544 | return NULL; | |
2457aec6 MG |
1545 | |
1546 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1547 | fgp_flags |= FGP_LOCK; | |
1548 | ||
eb39d618 | 1549 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1550 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1551 | __SetPageReferenced(page); |
2457aec6 | 1552 | |
abc1be13 | 1553 | err = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
eb2be189 | 1554 | if (unlikely(err)) { |
09cbfeaf | 1555 | put_page(page); |
eb2be189 NP |
1556 | page = NULL; |
1557 | if (err == -EEXIST) | |
1558 | goto repeat; | |
1da177e4 | 1559 | } |
1da177e4 | 1560 | } |
2457aec6 | 1561 | |
1da177e4 LT |
1562 | return page; |
1563 | } | |
2457aec6 | 1564 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1565 | |
0cd6144a JW |
1566 | /** |
1567 | * find_get_entries - gang pagecache lookup | |
1568 | * @mapping: The address_space to search | |
1569 | * @start: The starting page cache index | |
1570 | * @nr_entries: The maximum number of entries | |
1571 | * @entries: Where the resulting entries are placed | |
1572 | * @indices: The cache indices corresponding to the entries in @entries | |
1573 | * | |
1574 | * find_get_entries() will search for and return a group of up to | |
1575 | * @nr_entries entries in the mapping. The entries are placed at | |
1576 | * @entries. find_get_entries() takes a reference against any actual | |
1577 | * pages it returns. | |
1578 | * | |
1579 | * The search returns a group of mapping-contiguous page cache entries | |
1580 | * with ascending indexes. There may be holes in the indices due to | |
1581 | * not-present pages. | |
1582 | * | |
139b6a6f JW |
1583 | * Any shadow entries of evicted pages, or swap entries from |
1584 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1585 | * |
1586 | * find_get_entries() returns the number of pages and shadow entries | |
1587 | * which were found. | |
1588 | */ | |
1589 | unsigned find_get_entries(struct address_space *mapping, | |
1590 | pgoff_t start, unsigned int nr_entries, | |
1591 | struct page **entries, pgoff_t *indices) | |
1592 | { | |
1593 | void **slot; | |
1594 | unsigned int ret = 0; | |
1595 | struct radix_tree_iter iter; | |
1596 | ||
1597 | if (!nr_entries) | |
1598 | return 0; | |
1599 | ||
1600 | rcu_read_lock(); | |
b93b0163 | 1601 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) { |
83929372 | 1602 | struct page *head, *page; |
0cd6144a JW |
1603 | repeat: |
1604 | page = radix_tree_deref_slot(slot); | |
1605 | if (unlikely(!page)) | |
1606 | continue; | |
1607 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1608 | if (radix_tree_deref_retry(page)) { |
1609 | slot = radix_tree_iter_retry(&iter); | |
1610 | continue; | |
1611 | } | |
0cd6144a | 1612 | /* |
f9fe48be RZ |
1613 | * A shadow entry of a recently evicted page, a swap |
1614 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1615 | * without attempting to raise page count. | |
0cd6144a JW |
1616 | */ |
1617 | goto export; | |
1618 | } | |
83929372 KS |
1619 | |
1620 | head = compound_head(page); | |
1621 | if (!page_cache_get_speculative(head)) | |
1622 | goto repeat; | |
1623 | ||
1624 | /* The page was split under us? */ | |
1625 | if (compound_head(page) != head) { | |
1626 | put_page(head); | |
0cd6144a | 1627 | goto repeat; |
83929372 | 1628 | } |
0cd6144a JW |
1629 | |
1630 | /* Has the page moved? */ | |
1631 | if (unlikely(page != *slot)) { | |
83929372 | 1632 | put_page(head); |
0cd6144a JW |
1633 | goto repeat; |
1634 | } | |
1635 | export: | |
1636 | indices[ret] = iter.index; | |
1637 | entries[ret] = page; | |
1638 | if (++ret == nr_entries) | |
1639 | break; | |
1640 | } | |
1641 | rcu_read_unlock(); | |
1642 | return ret; | |
1643 | } | |
1644 | ||
1da177e4 | 1645 | /** |
b947cee4 | 1646 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1647 | * @mapping: The address_space to search |
1648 | * @start: The starting page index | |
b947cee4 | 1649 | * @end: The final page index (inclusive) |
1da177e4 LT |
1650 | * @nr_pages: The maximum number of pages |
1651 | * @pages: Where the resulting pages are placed | |
1652 | * | |
b947cee4 JK |
1653 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1654 | * pages in the mapping starting at index @start and up to index @end | |
1655 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1656 | * a reference against the returned pages. | |
1da177e4 LT |
1657 | * |
1658 | * The search returns a group of mapping-contiguous pages with ascending | |
1659 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1660 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1661 | * |
b947cee4 JK |
1662 | * find_get_pages_range() returns the number of pages which were found. If this |
1663 | * number is smaller than @nr_pages, the end of specified range has been | |
1664 | * reached. | |
1da177e4 | 1665 | */ |
b947cee4 JK |
1666 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1667 | pgoff_t end, unsigned int nr_pages, | |
1668 | struct page **pages) | |
1da177e4 | 1669 | { |
0fc9d104 KK |
1670 | struct radix_tree_iter iter; |
1671 | void **slot; | |
1672 | unsigned ret = 0; | |
1673 | ||
1674 | if (unlikely(!nr_pages)) | |
1675 | return 0; | |
a60637c8 NP |
1676 | |
1677 | rcu_read_lock(); | |
b93b0163 | 1678 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, *start) { |
83929372 | 1679 | struct page *head, *page; |
b947cee4 JK |
1680 | |
1681 | if (iter.index > end) | |
1682 | break; | |
a60637c8 | 1683 | repeat: |
0fc9d104 | 1684 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1685 | if (unlikely(!page)) |
1686 | continue; | |
9d8aa4ea | 1687 | |
a2c16d6c | 1688 | if (radix_tree_exception(page)) { |
8079b1c8 | 1689 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1690 | slot = radix_tree_iter_retry(&iter); |
1691 | continue; | |
8079b1c8 | 1692 | } |
a2c16d6c | 1693 | /* |
139b6a6f JW |
1694 | * A shadow entry of a recently evicted page, |
1695 | * or a swap entry from shmem/tmpfs. Skip | |
1696 | * over it. | |
a2c16d6c | 1697 | */ |
8079b1c8 | 1698 | continue; |
27d20fdd | 1699 | } |
a60637c8 | 1700 | |
83929372 KS |
1701 | head = compound_head(page); |
1702 | if (!page_cache_get_speculative(head)) | |
1703 | goto repeat; | |
1704 | ||
1705 | /* The page was split under us? */ | |
1706 | if (compound_head(page) != head) { | |
1707 | put_page(head); | |
a60637c8 | 1708 | goto repeat; |
83929372 | 1709 | } |
a60637c8 NP |
1710 | |
1711 | /* Has the page moved? */ | |
0fc9d104 | 1712 | if (unlikely(page != *slot)) { |
83929372 | 1713 | put_page(head); |
a60637c8 NP |
1714 | goto repeat; |
1715 | } | |
1da177e4 | 1716 | |
a60637c8 | 1717 | pages[ret] = page; |
b947cee4 JK |
1718 | if (++ret == nr_pages) { |
1719 | *start = pages[ret - 1]->index + 1; | |
1720 | goto out; | |
1721 | } | |
a60637c8 | 1722 | } |
5b280c0c | 1723 | |
b947cee4 JK |
1724 | /* |
1725 | * We come here when there is no page beyond @end. We take care to not | |
1726 | * overflow the index @start as it confuses some of the callers. This | |
1727 | * breaks the iteration when there is page at index -1 but that is | |
1728 | * already broken anyway. | |
1729 | */ | |
1730 | if (end == (pgoff_t)-1) | |
1731 | *start = (pgoff_t)-1; | |
1732 | else | |
1733 | *start = end + 1; | |
1734 | out: | |
a60637c8 | 1735 | rcu_read_unlock(); |
d72dc8a2 | 1736 | |
1da177e4 LT |
1737 | return ret; |
1738 | } | |
1739 | ||
ebf43500 JA |
1740 | /** |
1741 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1742 | * @mapping: The address_space to search | |
1743 | * @index: The starting page index | |
1744 | * @nr_pages: The maximum number of pages | |
1745 | * @pages: Where the resulting pages are placed | |
1746 | * | |
1747 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1748 | * that the returned number of pages are guaranteed to be contiguous. | |
1749 | * | |
1750 | * find_get_pages_contig() returns the number of pages which were found. | |
1751 | */ | |
1752 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1753 | unsigned int nr_pages, struct page **pages) | |
1754 | { | |
0fc9d104 KK |
1755 | struct radix_tree_iter iter; |
1756 | void **slot; | |
1757 | unsigned int ret = 0; | |
1758 | ||
1759 | if (unlikely(!nr_pages)) | |
1760 | return 0; | |
a60637c8 NP |
1761 | |
1762 | rcu_read_lock(); | |
b93b0163 | 1763 | radix_tree_for_each_contig(slot, &mapping->i_pages, &iter, index) { |
83929372 | 1764 | struct page *head, *page; |
a60637c8 | 1765 | repeat: |
0fc9d104 KK |
1766 | page = radix_tree_deref_slot(slot); |
1767 | /* The hole, there no reason to continue */ | |
a60637c8 | 1768 | if (unlikely(!page)) |
0fc9d104 | 1769 | break; |
9d8aa4ea | 1770 | |
a2c16d6c | 1771 | if (radix_tree_exception(page)) { |
8079b1c8 | 1772 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1773 | slot = radix_tree_iter_retry(&iter); |
1774 | continue; | |
8079b1c8 | 1775 | } |
a2c16d6c | 1776 | /* |
139b6a6f JW |
1777 | * A shadow entry of a recently evicted page, |
1778 | * or a swap entry from shmem/tmpfs. Stop | |
1779 | * looking for contiguous pages. | |
a2c16d6c | 1780 | */ |
8079b1c8 | 1781 | break; |
a2c16d6c | 1782 | } |
ebf43500 | 1783 | |
83929372 KS |
1784 | head = compound_head(page); |
1785 | if (!page_cache_get_speculative(head)) | |
1786 | goto repeat; | |
1787 | ||
1788 | /* The page was split under us? */ | |
1789 | if (compound_head(page) != head) { | |
1790 | put_page(head); | |
a60637c8 | 1791 | goto repeat; |
83929372 | 1792 | } |
a60637c8 NP |
1793 | |
1794 | /* Has the page moved? */ | |
0fc9d104 | 1795 | if (unlikely(page != *slot)) { |
83929372 | 1796 | put_page(head); |
a60637c8 NP |
1797 | goto repeat; |
1798 | } | |
1799 | ||
9cbb4cb2 NP |
1800 | /* |
1801 | * must check mapping and index after taking the ref. | |
1802 | * otherwise we can get both false positives and false | |
1803 | * negatives, which is just confusing to the caller. | |
1804 | */ | |
83929372 | 1805 | if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { |
09cbfeaf | 1806 | put_page(page); |
9cbb4cb2 NP |
1807 | break; |
1808 | } | |
1809 | ||
a60637c8 | 1810 | pages[ret] = page; |
0fc9d104 KK |
1811 | if (++ret == nr_pages) |
1812 | break; | |
ebf43500 | 1813 | } |
a60637c8 NP |
1814 | rcu_read_unlock(); |
1815 | return ret; | |
ebf43500 | 1816 | } |
ef71c15c | 1817 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1818 | |
485bb99b | 1819 | /** |
72b045ae | 1820 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
1821 | * @mapping: the address_space to search |
1822 | * @index: the starting page index | |
72b045ae | 1823 | * @end: The final page index (inclusive) |
485bb99b RD |
1824 | * @tag: the tag index |
1825 | * @nr_pages: the maximum number of pages | |
1826 | * @pages: where the resulting pages are placed | |
1827 | * | |
1da177e4 | 1828 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1829 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 | 1830 | */ |
72b045ae JK |
1831 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
1832 | pgoff_t end, int tag, unsigned int nr_pages, | |
1833 | struct page **pages) | |
1da177e4 | 1834 | { |
0fc9d104 KK |
1835 | struct radix_tree_iter iter; |
1836 | void **slot; | |
1837 | unsigned ret = 0; | |
1838 | ||
1839 | if (unlikely(!nr_pages)) | |
1840 | return 0; | |
a60637c8 NP |
1841 | |
1842 | rcu_read_lock(); | |
b93b0163 | 1843 | radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, *index, tag) { |
83929372 | 1844 | struct page *head, *page; |
72b045ae JK |
1845 | |
1846 | if (iter.index > end) | |
1847 | break; | |
a60637c8 | 1848 | repeat: |
0fc9d104 | 1849 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1850 | if (unlikely(!page)) |
1851 | continue; | |
9d8aa4ea | 1852 | |
a2c16d6c | 1853 | if (radix_tree_exception(page)) { |
8079b1c8 | 1854 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1855 | slot = radix_tree_iter_retry(&iter); |
1856 | continue; | |
8079b1c8 | 1857 | } |
a2c16d6c | 1858 | /* |
139b6a6f JW |
1859 | * A shadow entry of a recently evicted page. |
1860 | * | |
1861 | * Those entries should never be tagged, but | |
1862 | * this tree walk is lockless and the tags are | |
1863 | * looked up in bulk, one radix tree node at a | |
1864 | * time, so there is a sizable window for page | |
1865 | * reclaim to evict a page we saw tagged. | |
1866 | * | |
1867 | * Skip over it. | |
a2c16d6c | 1868 | */ |
139b6a6f | 1869 | continue; |
a2c16d6c | 1870 | } |
a60637c8 | 1871 | |
83929372 KS |
1872 | head = compound_head(page); |
1873 | if (!page_cache_get_speculative(head)) | |
a60637c8 NP |
1874 | goto repeat; |
1875 | ||
83929372 KS |
1876 | /* The page was split under us? */ |
1877 | if (compound_head(page) != head) { | |
1878 | put_page(head); | |
1879 | goto repeat; | |
1880 | } | |
1881 | ||
a60637c8 | 1882 | /* Has the page moved? */ |
0fc9d104 | 1883 | if (unlikely(page != *slot)) { |
83929372 | 1884 | put_page(head); |
a60637c8 NP |
1885 | goto repeat; |
1886 | } | |
1887 | ||
1888 | pages[ret] = page; | |
72b045ae JK |
1889 | if (++ret == nr_pages) { |
1890 | *index = pages[ret - 1]->index + 1; | |
1891 | goto out; | |
1892 | } | |
a60637c8 | 1893 | } |
5b280c0c | 1894 | |
72b045ae JK |
1895 | /* |
1896 | * We come here when we got at @end. We take care to not overflow the | |
1897 | * index @index as it confuses some of the callers. This breaks the | |
1898 | * iteration when there is page at index -1 but that is already broken | |
1899 | * anyway. | |
1900 | */ | |
1901 | if (end == (pgoff_t)-1) | |
1902 | *index = (pgoff_t)-1; | |
1903 | else | |
1904 | *index = end + 1; | |
1905 | out: | |
a60637c8 | 1906 | rcu_read_unlock(); |
1da177e4 | 1907 | |
1da177e4 LT |
1908 | return ret; |
1909 | } | |
72b045ae | 1910 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 1911 | |
7e7f7749 RZ |
1912 | /** |
1913 | * find_get_entries_tag - find and return entries that match @tag | |
1914 | * @mapping: the address_space to search | |
1915 | * @start: the starting page cache index | |
1916 | * @tag: the tag index | |
1917 | * @nr_entries: the maximum number of entries | |
1918 | * @entries: where the resulting entries are placed | |
1919 | * @indices: the cache indices corresponding to the entries in @entries | |
1920 | * | |
1921 | * Like find_get_entries, except we only return entries which are tagged with | |
1922 | * @tag. | |
1923 | */ | |
1924 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1925 | int tag, unsigned int nr_entries, | |
1926 | struct page **entries, pgoff_t *indices) | |
1927 | { | |
1928 | void **slot; | |
1929 | unsigned int ret = 0; | |
1930 | struct radix_tree_iter iter; | |
1931 | ||
1932 | if (!nr_entries) | |
1933 | return 0; | |
1934 | ||
1935 | rcu_read_lock(); | |
b93b0163 | 1936 | radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, start, tag) { |
83929372 | 1937 | struct page *head, *page; |
7e7f7749 RZ |
1938 | repeat: |
1939 | page = radix_tree_deref_slot(slot); | |
1940 | if (unlikely(!page)) | |
1941 | continue; | |
1942 | if (radix_tree_exception(page)) { | |
1943 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1944 | slot = radix_tree_iter_retry(&iter); |
1945 | continue; | |
7e7f7749 RZ |
1946 | } |
1947 | ||
1948 | /* | |
1949 | * A shadow entry of a recently evicted page, a swap | |
1950 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1951 | * without attempting to raise page count. | |
1952 | */ | |
1953 | goto export; | |
1954 | } | |
83929372 KS |
1955 | |
1956 | head = compound_head(page); | |
1957 | if (!page_cache_get_speculative(head)) | |
7e7f7749 RZ |
1958 | goto repeat; |
1959 | ||
83929372 KS |
1960 | /* The page was split under us? */ |
1961 | if (compound_head(page) != head) { | |
1962 | put_page(head); | |
1963 | goto repeat; | |
1964 | } | |
1965 | ||
7e7f7749 RZ |
1966 | /* Has the page moved? */ |
1967 | if (unlikely(page != *slot)) { | |
83929372 | 1968 | put_page(head); |
7e7f7749 RZ |
1969 | goto repeat; |
1970 | } | |
1971 | export: | |
1972 | indices[ret] = iter.index; | |
1973 | entries[ret] = page; | |
1974 | if (++ret == nr_entries) | |
1975 | break; | |
1976 | } | |
1977 | rcu_read_unlock(); | |
1978 | return ret; | |
1979 | } | |
1980 | EXPORT_SYMBOL(find_get_entries_tag); | |
1981 | ||
76d42bd9 WF |
1982 | /* |
1983 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1984 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1985 | * | |
1986 | * ---R__________________________________________B__________ | |
1987 | * ^ reading here ^ bad block(assume 4k) | |
1988 | * | |
1989 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1990 | * => failing the whole request => read(R) => read(R+1) => | |
1991 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1992 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1993 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1994 | * | |
1995 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1996 | */ | |
1997 | static void shrink_readahead_size_eio(struct file *filp, | |
1998 | struct file_ra_state *ra) | |
1999 | { | |
76d42bd9 | 2000 | ra->ra_pages /= 4; |
76d42bd9 WF |
2001 | } |
2002 | ||
485bb99b | 2003 | /** |
47c27bc4 CH |
2004 | * generic_file_buffered_read - generic file read routine |
2005 | * @iocb: the iocb to read | |
6e58e79d AV |
2006 | * @iter: data destination |
2007 | * @written: already copied | |
485bb99b | 2008 | * |
1da177e4 | 2009 | * This is a generic file read routine, and uses the |
485bb99b | 2010 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
2011 | * |
2012 | * This is really ugly. But the goto's actually try to clarify some | |
2013 | * of the logic when it comes to error handling etc. | |
1da177e4 | 2014 | */ |
47c27bc4 | 2015 | static ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 2016 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 2017 | { |
47c27bc4 | 2018 | struct file *filp = iocb->ki_filp; |
36e78914 | 2019 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2020 | struct inode *inode = mapping->host; |
36e78914 | 2021 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 2022 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
2023 | pgoff_t index; |
2024 | pgoff_t last_index; | |
2025 | pgoff_t prev_index; | |
2026 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 2027 | unsigned int prev_offset; |
6e58e79d | 2028 | int error = 0; |
1da177e4 | 2029 | |
c2a9737f | 2030 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2031 | return 0; |
c2a9737f WF |
2032 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
2033 | ||
09cbfeaf KS |
2034 | index = *ppos >> PAGE_SHIFT; |
2035 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
2036 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
2037 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
2038 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 2039 | |
1da177e4 LT |
2040 | for (;;) { |
2041 | struct page *page; | |
57f6b96c | 2042 | pgoff_t end_index; |
a32ea1e1 | 2043 | loff_t isize; |
1da177e4 LT |
2044 | unsigned long nr, ret; |
2045 | ||
1da177e4 | 2046 | cond_resched(); |
1da177e4 | 2047 | find_page: |
5abf186a MH |
2048 | if (fatal_signal_pending(current)) { |
2049 | error = -EINTR; | |
2050 | goto out; | |
2051 | } | |
2052 | ||
1da177e4 | 2053 | page = find_get_page(mapping, index); |
3ea89ee8 | 2054 | if (!page) { |
3239d834 MT |
2055 | if (iocb->ki_flags & IOCB_NOWAIT) |
2056 | goto would_block; | |
cf914a7d | 2057 | page_cache_sync_readahead(mapping, |
7ff81078 | 2058 | ra, filp, |
3ea89ee8 FW |
2059 | index, last_index - index); |
2060 | page = find_get_page(mapping, index); | |
2061 | if (unlikely(page == NULL)) | |
2062 | goto no_cached_page; | |
2063 | } | |
2064 | if (PageReadahead(page)) { | |
cf914a7d | 2065 | page_cache_async_readahead(mapping, |
7ff81078 | 2066 | ra, filp, page, |
3ea89ee8 | 2067 | index, last_index - index); |
1da177e4 | 2068 | } |
8ab22b9a | 2069 | if (!PageUptodate(page)) { |
3239d834 MT |
2070 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2071 | put_page(page); | |
2072 | goto would_block; | |
2073 | } | |
2074 | ||
ebded027 MG |
2075 | /* |
2076 | * See comment in do_read_cache_page on why | |
2077 | * wait_on_page_locked is used to avoid unnecessarily | |
2078 | * serialisations and why it's safe. | |
2079 | */ | |
c4b209a4 BVA |
2080 | error = wait_on_page_locked_killable(page); |
2081 | if (unlikely(error)) | |
2082 | goto readpage_error; | |
ebded027 MG |
2083 | if (PageUptodate(page)) |
2084 | goto page_ok; | |
2085 | ||
09cbfeaf | 2086 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2087 | !mapping->a_ops->is_partially_uptodate) |
2088 | goto page_not_up_to_date; | |
6d6d36bc EG |
2089 | /* pipes can't handle partially uptodate pages */ |
2090 | if (unlikely(iter->type & ITER_PIPE)) | |
2091 | goto page_not_up_to_date; | |
529ae9aa | 2092 | if (!trylock_page(page)) |
8ab22b9a | 2093 | goto page_not_up_to_date; |
8d056cb9 DH |
2094 | /* Did it get truncated before we got the lock? */ |
2095 | if (!page->mapping) | |
2096 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2097 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2098 | offset, iter->count)) |
8ab22b9a HH |
2099 | goto page_not_up_to_date_locked; |
2100 | unlock_page(page); | |
2101 | } | |
1da177e4 | 2102 | page_ok: |
a32ea1e1 N |
2103 | /* |
2104 | * i_size must be checked after we know the page is Uptodate. | |
2105 | * | |
2106 | * Checking i_size after the check allows us to calculate | |
2107 | * the correct value for "nr", which means the zero-filled | |
2108 | * part of the page is not copied back to userspace (unless | |
2109 | * another truncate extends the file - this is desired though). | |
2110 | */ | |
2111 | ||
2112 | isize = i_size_read(inode); | |
09cbfeaf | 2113 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2114 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2115 | put_page(page); |
a32ea1e1 N |
2116 | goto out; |
2117 | } | |
2118 | ||
2119 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2120 | nr = PAGE_SIZE; |
a32ea1e1 | 2121 | if (index == end_index) { |
09cbfeaf | 2122 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2123 | if (nr <= offset) { |
09cbfeaf | 2124 | put_page(page); |
a32ea1e1 N |
2125 | goto out; |
2126 | } | |
2127 | } | |
2128 | nr = nr - offset; | |
1da177e4 LT |
2129 | |
2130 | /* If users can be writing to this page using arbitrary | |
2131 | * virtual addresses, take care about potential aliasing | |
2132 | * before reading the page on the kernel side. | |
2133 | */ | |
2134 | if (mapping_writably_mapped(mapping)) | |
2135 | flush_dcache_page(page); | |
2136 | ||
2137 | /* | |
ec0f1637 JK |
2138 | * When a sequential read accesses a page several times, |
2139 | * only mark it as accessed the first time. | |
1da177e4 | 2140 | */ |
ec0f1637 | 2141 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2142 | mark_page_accessed(page); |
2143 | prev_index = index; | |
2144 | ||
2145 | /* | |
2146 | * Ok, we have the page, and it's up-to-date, so | |
2147 | * now we can copy it to user space... | |
1da177e4 | 2148 | */ |
6e58e79d AV |
2149 | |
2150 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2151 | offset += ret; |
09cbfeaf KS |
2152 | index += offset >> PAGE_SHIFT; |
2153 | offset &= ~PAGE_MASK; | |
6ce745ed | 2154 | prev_offset = offset; |
1da177e4 | 2155 | |
09cbfeaf | 2156 | put_page(page); |
6e58e79d AV |
2157 | written += ret; |
2158 | if (!iov_iter_count(iter)) | |
2159 | goto out; | |
2160 | if (ret < nr) { | |
2161 | error = -EFAULT; | |
2162 | goto out; | |
2163 | } | |
2164 | continue; | |
1da177e4 LT |
2165 | |
2166 | page_not_up_to_date: | |
2167 | /* Get exclusive access to the page ... */ | |
85462323 ON |
2168 | error = lock_page_killable(page); |
2169 | if (unlikely(error)) | |
2170 | goto readpage_error; | |
1da177e4 | 2171 | |
8ab22b9a | 2172 | page_not_up_to_date_locked: |
da6052f7 | 2173 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2174 | if (!page->mapping) { |
2175 | unlock_page(page); | |
09cbfeaf | 2176 | put_page(page); |
1da177e4 LT |
2177 | continue; |
2178 | } | |
2179 | ||
2180 | /* Did somebody else fill it already? */ | |
2181 | if (PageUptodate(page)) { | |
2182 | unlock_page(page); | |
2183 | goto page_ok; | |
2184 | } | |
2185 | ||
2186 | readpage: | |
91803b49 JM |
2187 | /* |
2188 | * A previous I/O error may have been due to temporary | |
2189 | * failures, eg. multipath errors. | |
2190 | * PG_error will be set again if readpage fails. | |
2191 | */ | |
2192 | ClearPageError(page); | |
1da177e4 LT |
2193 | /* Start the actual read. The read will unlock the page. */ |
2194 | error = mapping->a_ops->readpage(filp, page); | |
2195 | ||
994fc28c ZB |
2196 | if (unlikely(error)) { |
2197 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2198 | put_page(page); |
6e58e79d | 2199 | error = 0; |
994fc28c ZB |
2200 | goto find_page; |
2201 | } | |
1da177e4 | 2202 | goto readpage_error; |
994fc28c | 2203 | } |
1da177e4 LT |
2204 | |
2205 | if (!PageUptodate(page)) { | |
85462323 ON |
2206 | error = lock_page_killable(page); |
2207 | if (unlikely(error)) | |
2208 | goto readpage_error; | |
1da177e4 LT |
2209 | if (!PageUptodate(page)) { |
2210 | if (page->mapping == NULL) { | |
2211 | /* | |
2ecdc82e | 2212 | * invalidate_mapping_pages got it |
1da177e4 LT |
2213 | */ |
2214 | unlock_page(page); | |
09cbfeaf | 2215 | put_page(page); |
1da177e4 LT |
2216 | goto find_page; |
2217 | } | |
2218 | unlock_page(page); | |
7ff81078 | 2219 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
2220 | error = -EIO; |
2221 | goto readpage_error; | |
1da177e4 LT |
2222 | } |
2223 | unlock_page(page); | |
2224 | } | |
2225 | ||
1da177e4 LT |
2226 | goto page_ok; |
2227 | ||
2228 | readpage_error: | |
2229 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2230 | put_page(page); |
1da177e4 LT |
2231 | goto out; |
2232 | ||
2233 | no_cached_page: | |
2234 | /* | |
2235 | * Ok, it wasn't cached, so we need to create a new | |
2236 | * page.. | |
2237 | */ | |
453f85d4 | 2238 | page = page_cache_alloc(mapping); |
eb2be189 | 2239 | if (!page) { |
6e58e79d | 2240 | error = -ENOMEM; |
eb2be189 | 2241 | goto out; |
1da177e4 | 2242 | } |
6afdb859 | 2243 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2244 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2245 | if (error) { |
09cbfeaf | 2246 | put_page(page); |
6e58e79d AV |
2247 | if (error == -EEXIST) { |
2248 | error = 0; | |
1da177e4 | 2249 | goto find_page; |
6e58e79d | 2250 | } |
1da177e4 LT |
2251 | goto out; |
2252 | } | |
1da177e4 LT |
2253 | goto readpage; |
2254 | } | |
2255 | ||
3239d834 MT |
2256 | would_block: |
2257 | error = -EAGAIN; | |
1da177e4 | 2258 | out: |
7ff81078 | 2259 | ra->prev_pos = prev_index; |
09cbfeaf | 2260 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2261 | ra->prev_pos |= prev_offset; |
1da177e4 | 2262 | |
09cbfeaf | 2263 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2264 | file_accessed(filp); |
6e58e79d | 2265 | return written ? written : error; |
1da177e4 LT |
2266 | } |
2267 | ||
485bb99b | 2268 | /** |
6abd2322 | 2269 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2270 | * @iocb: kernel I/O control block |
6abd2322 | 2271 | * @iter: destination for the data read |
485bb99b | 2272 | * |
6abd2322 | 2273 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
2274 | * that can use the page cache directly. |
2275 | */ | |
2276 | ssize_t | |
ed978a81 | 2277 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2278 | { |
e7080a43 | 2279 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2280 | ssize_t retval = 0; |
e7080a43 NS |
2281 | |
2282 | if (!count) | |
2283 | goto out; /* skip atime */ | |
1da177e4 | 2284 | |
2ba48ce5 | 2285 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2286 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2287 | struct address_space *mapping = file->f_mapping; |
2288 | struct inode *inode = mapping->host; | |
543ade1f | 2289 | loff_t size; |
1da177e4 | 2290 | |
1da177e4 | 2291 | size = i_size_read(inode); |
6be96d3a GR |
2292 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2293 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2294 | iocb->ki_pos + count - 1)) | |
2295 | return -EAGAIN; | |
2296 | } else { | |
2297 | retval = filemap_write_and_wait_range(mapping, | |
2298 | iocb->ki_pos, | |
2299 | iocb->ki_pos + count - 1); | |
2300 | if (retval < 0) | |
2301 | goto out; | |
2302 | } | |
d8d3d94b | 2303 | |
0d5b0cf2 CH |
2304 | file_accessed(file); |
2305 | ||
5ecda137 | 2306 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2307 | if (retval >= 0) { |
c64fb5c7 | 2308 | iocb->ki_pos += retval; |
5ecda137 | 2309 | count -= retval; |
9fe55eea | 2310 | } |
5b47d59a | 2311 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2312 | |
9fe55eea SW |
2313 | /* |
2314 | * Btrfs can have a short DIO read if we encounter | |
2315 | * compressed extents, so if there was an error, or if | |
2316 | * we've already read everything we wanted to, or if | |
2317 | * there was a short read because we hit EOF, go ahead | |
2318 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2319 | * the rest of the read. Buffered reads will not work for |
2320 | * DAX files, so don't bother trying. | |
9fe55eea | 2321 | */ |
5ecda137 | 2322 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2323 | IS_DAX(inode)) |
9fe55eea | 2324 | goto out; |
1da177e4 LT |
2325 | } |
2326 | ||
47c27bc4 | 2327 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2328 | out: |
2329 | return retval; | |
2330 | } | |
ed978a81 | 2331 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2332 | |
1da177e4 | 2333 | #ifdef CONFIG_MMU |
485bb99b RD |
2334 | /** |
2335 | * page_cache_read - adds requested page to the page cache if not already there | |
2336 | * @file: file to read | |
2337 | * @offset: page index | |
62eb320a | 2338 | * @gfp_mask: memory allocation flags |
485bb99b | 2339 | * |
1da177e4 LT |
2340 | * This adds the requested page to the page cache if it isn't already there, |
2341 | * and schedules an I/O to read in its contents from disk. | |
2342 | */ | |
c20cd45e | 2343 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
2344 | { |
2345 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 2346 | struct page *page; |
994fc28c | 2347 | int ret; |
1da177e4 | 2348 | |
994fc28c | 2349 | do { |
453f85d4 | 2350 | page = __page_cache_alloc(gfp_mask); |
994fc28c ZB |
2351 | if (!page) |
2352 | return -ENOMEM; | |
2353 | ||
abc1be13 | 2354 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
994fc28c ZB |
2355 | if (ret == 0) |
2356 | ret = mapping->a_ops->readpage(file, page); | |
2357 | else if (ret == -EEXIST) | |
2358 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 2359 | |
09cbfeaf | 2360 | put_page(page); |
1da177e4 | 2361 | |
994fc28c | 2362 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 2363 | |
994fc28c | 2364 | return ret; |
1da177e4 LT |
2365 | } |
2366 | ||
2367 | #define MMAP_LOTSAMISS (100) | |
2368 | ||
ef00e08e LT |
2369 | /* |
2370 | * Synchronous readahead happens when we don't even find | |
2371 | * a page in the page cache at all. | |
2372 | */ | |
2373 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
2374 | struct file_ra_state *ra, | |
2375 | struct file *file, | |
2376 | pgoff_t offset) | |
2377 | { | |
ef00e08e LT |
2378 | struct address_space *mapping = file->f_mapping; |
2379 | ||
2380 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2381 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 2382 | return; |
275b12bf WF |
2383 | if (!ra->ra_pages) |
2384 | return; | |
ef00e08e | 2385 | |
64363aad | 2386 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
2387 | page_cache_sync_readahead(mapping, ra, file, offset, |
2388 | ra->ra_pages); | |
ef00e08e LT |
2389 | return; |
2390 | } | |
2391 | ||
207d04ba AK |
2392 | /* Avoid banging the cache line if not needed */ |
2393 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2394 | ra->mmap_miss++; |
2395 | ||
2396 | /* | |
2397 | * Do we miss much more than hit in this file? If so, | |
2398 | * stop bothering with read-ahead. It will only hurt. | |
2399 | */ | |
2400 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
2401 | return; | |
2402 | ||
d30a1100 WF |
2403 | /* |
2404 | * mmap read-around | |
2405 | */ | |
600e19af RG |
2406 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2407 | ra->size = ra->ra_pages; | |
2408 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2409 | ra_submit(ra, mapping, file); |
ef00e08e LT |
2410 | } |
2411 | ||
2412 | /* | |
2413 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2414 | * so we want to possibly extend the readahead further.. | |
2415 | */ | |
2416 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2417 | struct file_ra_state *ra, | |
2418 | struct file *file, | |
2419 | struct page *page, | |
2420 | pgoff_t offset) | |
2421 | { | |
2422 | struct address_space *mapping = file->f_mapping; | |
2423 | ||
2424 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2425 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2426 | return; |
2427 | if (ra->mmap_miss > 0) | |
2428 | ra->mmap_miss--; | |
2429 | if (PageReadahead(page)) | |
2fad6f5d WF |
2430 | page_cache_async_readahead(mapping, ra, file, |
2431 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2432 | } |
2433 | ||
485bb99b | 2434 | /** |
54cb8821 | 2435 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2436 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2437 | * |
54cb8821 | 2438 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2439 | * mapped memory region to read in file data during a page fault. |
2440 | * | |
2441 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2442 | * it in the page cache, and handles the special cases reasonably without | |
2443 | * having a lot of duplicated code. | |
9a95f3cf PC |
2444 | * |
2445 | * vma->vm_mm->mmap_sem must be held on entry. | |
2446 | * | |
2447 | * If our return value has VM_FAULT_RETRY set, it's because | |
2448 | * lock_page_or_retry() returned 0. | |
2449 | * The mmap_sem has usually been released in this case. | |
2450 | * See __lock_page_or_retry() for the exception. | |
2451 | * | |
2452 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2453 | * has not been released. | |
2454 | * | |
2455 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2456 | */ |
2bcd6454 | 2457 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2458 | { |
2459 | int error; | |
11bac800 | 2460 | struct file *file = vmf->vma->vm_file; |
1da177e4 LT |
2461 | struct address_space *mapping = file->f_mapping; |
2462 | struct file_ra_state *ra = &file->f_ra; | |
2463 | struct inode *inode = mapping->host; | |
ef00e08e | 2464 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2465 | pgoff_t max_off; |
1da177e4 | 2466 | struct page *page; |
2bcd6454 | 2467 | vm_fault_t ret = 0; |
1da177e4 | 2468 | |
9ab2594f MW |
2469 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2470 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2471 | return VM_FAULT_SIGBUS; |
1da177e4 | 2472 | |
1da177e4 | 2473 | /* |
49426420 | 2474 | * Do we have something in the page cache already? |
1da177e4 | 2475 | */ |
ef00e08e | 2476 | page = find_get_page(mapping, offset); |
45cac65b | 2477 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2478 | /* |
ef00e08e LT |
2479 | * We found the page, so try async readahead before |
2480 | * waiting for the lock. | |
1da177e4 | 2481 | */ |
11bac800 | 2482 | do_async_mmap_readahead(vmf->vma, ra, file, page, offset); |
45cac65b | 2483 | } else if (!page) { |
ef00e08e | 2484 | /* No page in the page cache at all */ |
11bac800 | 2485 | do_sync_mmap_readahead(vmf->vma, ra, file, offset); |
ef00e08e | 2486 | count_vm_event(PGMAJFAULT); |
2262185c | 2487 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2488 | ret = VM_FAULT_MAJOR; |
2489 | retry_find: | |
b522c94d | 2490 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2491 | if (!page) |
2492 | goto no_cached_page; | |
2493 | } | |
2494 | ||
11bac800 | 2495 | if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2496 | put_page(page); |
d065bd81 | 2497 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2498 | } |
b522c94d ML |
2499 | |
2500 | /* Did it get truncated? */ | |
2501 | if (unlikely(page->mapping != mapping)) { | |
2502 | unlock_page(page); | |
2503 | put_page(page); | |
2504 | goto retry_find; | |
2505 | } | |
309381fe | 2506 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2507 | |
1da177e4 | 2508 | /* |
d00806b1 NP |
2509 | * We have a locked page in the page cache, now we need to check |
2510 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2511 | */ |
d00806b1 | 2512 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2513 | goto page_not_uptodate; |
2514 | ||
ef00e08e LT |
2515 | /* |
2516 | * Found the page and have a reference on it. | |
2517 | * We must recheck i_size under page lock. | |
2518 | */ | |
9ab2594f MW |
2519 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2520 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2521 | unlock_page(page); |
09cbfeaf | 2522 | put_page(page); |
5307cc1a | 2523 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2524 | } |
2525 | ||
d0217ac0 | 2526 | vmf->page = page; |
83c54070 | 2527 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2528 | |
1da177e4 LT |
2529 | no_cached_page: |
2530 | /* | |
2531 | * We're only likely to ever get here if MADV_RANDOM is in | |
2532 | * effect. | |
2533 | */ | |
c20cd45e | 2534 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2535 | |
2536 | /* | |
2537 | * The page we want has now been added to the page cache. | |
2538 | * In the unlikely event that someone removed it in the | |
2539 | * meantime, we'll just come back here and read it again. | |
2540 | */ | |
2541 | if (error >= 0) | |
2542 | goto retry_find; | |
2543 | ||
2544 | /* | |
2545 | * An error return from page_cache_read can result if the | |
2546 | * system is low on memory, or a problem occurs while trying | |
2547 | * to schedule I/O. | |
2548 | */ | |
2549 | if (error == -ENOMEM) | |
d0217ac0 NP |
2550 | return VM_FAULT_OOM; |
2551 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2552 | |
2553 | page_not_uptodate: | |
1da177e4 LT |
2554 | /* |
2555 | * Umm, take care of errors if the page isn't up-to-date. | |
2556 | * Try to re-read it _once_. We do this synchronously, | |
2557 | * because there really aren't any performance issues here | |
2558 | * and we need to check for errors. | |
2559 | */ | |
1da177e4 | 2560 | ClearPageError(page); |
994fc28c | 2561 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2562 | if (!error) { |
2563 | wait_on_page_locked(page); | |
2564 | if (!PageUptodate(page)) | |
2565 | error = -EIO; | |
2566 | } | |
09cbfeaf | 2567 | put_page(page); |
d00806b1 NP |
2568 | |
2569 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2570 | goto retry_find; |
1da177e4 | 2571 | |
d00806b1 | 2572 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2573 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2574 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2575 | } |
2576 | EXPORT_SYMBOL(filemap_fault); | |
2577 | ||
82b0f8c3 | 2578 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2579 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 KS |
2580 | { |
2581 | struct radix_tree_iter iter; | |
2582 | void **slot; | |
82b0f8c3 | 2583 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2584 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2585 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2586 | unsigned long max_idx; |
83929372 | 2587 | struct page *head, *page; |
f1820361 KS |
2588 | |
2589 | rcu_read_lock(); | |
b93b0163 | 2590 | radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start_pgoff) { |
bae473a4 | 2591 | if (iter.index > end_pgoff) |
f1820361 KS |
2592 | break; |
2593 | repeat: | |
2594 | page = radix_tree_deref_slot(slot); | |
2595 | if (unlikely(!page)) | |
2596 | goto next; | |
2597 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2598 | if (radix_tree_deref_retry(page)) { |
2599 | slot = radix_tree_iter_retry(&iter); | |
2600 | continue; | |
2601 | } | |
2602 | goto next; | |
f1820361 KS |
2603 | } |
2604 | ||
83929372 KS |
2605 | head = compound_head(page); |
2606 | if (!page_cache_get_speculative(head)) | |
f1820361 KS |
2607 | goto repeat; |
2608 | ||
83929372 KS |
2609 | /* The page was split under us? */ |
2610 | if (compound_head(page) != head) { | |
2611 | put_page(head); | |
2612 | goto repeat; | |
2613 | } | |
2614 | ||
f1820361 KS |
2615 | /* Has the page moved? */ |
2616 | if (unlikely(page != *slot)) { | |
83929372 | 2617 | put_page(head); |
f1820361 KS |
2618 | goto repeat; |
2619 | } | |
2620 | ||
2621 | if (!PageUptodate(page) || | |
2622 | PageReadahead(page) || | |
2623 | PageHWPoison(page)) | |
2624 | goto skip; | |
2625 | if (!trylock_page(page)) | |
2626 | goto skip; | |
2627 | ||
2628 | if (page->mapping != mapping || !PageUptodate(page)) | |
2629 | goto unlock; | |
2630 | ||
9ab2594f MW |
2631 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2632 | if (page->index >= max_idx) | |
f1820361 KS |
2633 | goto unlock; |
2634 | ||
f1820361 KS |
2635 | if (file->f_ra.mmap_miss > 0) |
2636 | file->f_ra.mmap_miss--; | |
7267ec00 | 2637 | |
82b0f8c3 JK |
2638 | vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT; |
2639 | if (vmf->pte) | |
2640 | vmf->pte += iter.index - last_pgoff; | |
7267ec00 | 2641 | last_pgoff = iter.index; |
82b0f8c3 | 2642 | if (alloc_set_pte(vmf, NULL, page)) |
7267ec00 | 2643 | goto unlock; |
f1820361 KS |
2644 | unlock_page(page); |
2645 | goto next; | |
2646 | unlock: | |
2647 | unlock_page(page); | |
2648 | skip: | |
09cbfeaf | 2649 | put_page(page); |
f1820361 | 2650 | next: |
7267ec00 | 2651 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2652 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2653 | break; |
bae473a4 | 2654 | if (iter.index == end_pgoff) |
f1820361 KS |
2655 | break; |
2656 | } | |
2657 | rcu_read_unlock(); | |
2658 | } | |
2659 | EXPORT_SYMBOL(filemap_map_pages); | |
2660 | ||
2bcd6454 | 2661 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2662 | { |
2663 | struct page *page = vmf->page; | |
11bac800 | 2664 | struct inode *inode = file_inode(vmf->vma->vm_file); |
2bcd6454 | 2665 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 2666 | |
14da9200 | 2667 | sb_start_pagefault(inode->i_sb); |
11bac800 | 2668 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 JK |
2669 | lock_page(page); |
2670 | if (page->mapping != inode->i_mapping) { | |
2671 | unlock_page(page); | |
2672 | ret = VM_FAULT_NOPAGE; | |
2673 | goto out; | |
2674 | } | |
14da9200 JK |
2675 | /* |
2676 | * We mark the page dirty already here so that when freeze is in | |
2677 | * progress, we are guaranteed that writeback during freezing will | |
2678 | * see the dirty page and writeprotect it again. | |
2679 | */ | |
2680 | set_page_dirty(page); | |
1d1d1a76 | 2681 | wait_for_stable_page(page); |
4fcf1c62 | 2682 | out: |
14da9200 | 2683 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2684 | return ret; |
2685 | } | |
4fcf1c62 | 2686 | |
f0f37e2f | 2687 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2688 | .fault = filemap_fault, |
f1820361 | 2689 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2690 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2691 | }; |
2692 | ||
2693 | /* This is used for a general mmap of a disk file */ | |
2694 | ||
2695 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2696 | { | |
2697 | struct address_space *mapping = file->f_mapping; | |
2698 | ||
2699 | if (!mapping->a_ops->readpage) | |
2700 | return -ENOEXEC; | |
2701 | file_accessed(file); | |
2702 | vma->vm_ops = &generic_file_vm_ops; | |
2703 | return 0; | |
2704 | } | |
1da177e4 LT |
2705 | |
2706 | /* | |
2707 | * This is for filesystems which do not implement ->writepage. | |
2708 | */ | |
2709 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2710 | { | |
2711 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2712 | return -EINVAL; | |
2713 | return generic_file_mmap(file, vma); | |
2714 | } | |
2715 | #else | |
45397228 AB |
2716 | int filemap_page_mkwrite(struct vm_fault *vmf) |
2717 | { | |
2718 | return -ENOSYS; | |
2719 | } | |
1da177e4 LT |
2720 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2721 | { | |
2722 | return -ENOSYS; | |
2723 | } | |
2724 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2725 | { | |
2726 | return -ENOSYS; | |
2727 | } | |
2728 | #endif /* CONFIG_MMU */ | |
2729 | ||
45397228 | 2730 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
2731 | EXPORT_SYMBOL(generic_file_mmap); |
2732 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2733 | ||
67f9fd91 SL |
2734 | static struct page *wait_on_page_read(struct page *page) |
2735 | { | |
2736 | if (!IS_ERR(page)) { | |
2737 | wait_on_page_locked(page); | |
2738 | if (!PageUptodate(page)) { | |
09cbfeaf | 2739 | put_page(page); |
67f9fd91 SL |
2740 | page = ERR_PTR(-EIO); |
2741 | } | |
2742 | } | |
2743 | return page; | |
2744 | } | |
2745 | ||
32b63529 | 2746 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2747 | pgoff_t index, |
5e5358e7 | 2748 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2749 | void *data, |
2750 | gfp_t gfp) | |
1da177e4 | 2751 | { |
eb2be189 | 2752 | struct page *page; |
1da177e4 LT |
2753 | int err; |
2754 | repeat: | |
2755 | page = find_get_page(mapping, index); | |
2756 | if (!page) { | |
453f85d4 | 2757 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
2758 | if (!page) |
2759 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2760 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2761 | if (unlikely(err)) { |
09cbfeaf | 2762 | put_page(page); |
eb2be189 NP |
2763 | if (err == -EEXIST) |
2764 | goto repeat; | |
1da177e4 | 2765 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2766 | return ERR_PTR(err); |
2767 | } | |
32b63529 MG |
2768 | |
2769 | filler: | |
1da177e4 LT |
2770 | err = filler(data, page); |
2771 | if (err < 0) { | |
09cbfeaf | 2772 | put_page(page); |
32b63529 | 2773 | return ERR_PTR(err); |
1da177e4 | 2774 | } |
1da177e4 | 2775 | |
32b63529 MG |
2776 | page = wait_on_page_read(page); |
2777 | if (IS_ERR(page)) | |
2778 | return page; | |
2779 | goto out; | |
2780 | } | |
1da177e4 LT |
2781 | if (PageUptodate(page)) |
2782 | goto out; | |
2783 | ||
ebded027 MG |
2784 | /* |
2785 | * Page is not up to date and may be locked due one of the following | |
2786 | * case a: Page is being filled and the page lock is held | |
2787 | * case b: Read/write error clearing the page uptodate status | |
2788 | * case c: Truncation in progress (page locked) | |
2789 | * case d: Reclaim in progress | |
2790 | * | |
2791 | * Case a, the page will be up to date when the page is unlocked. | |
2792 | * There is no need to serialise on the page lock here as the page | |
2793 | * is pinned so the lock gives no additional protection. Even if the | |
2794 | * the page is truncated, the data is still valid if PageUptodate as | |
2795 | * it's a race vs truncate race. | |
2796 | * Case b, the page will not be up to date | |
2797 | * Case c, the page may be truncated but in itself, the data may still | |
2798 | * be valid after IO completes as it's a read vs truncate race. The | |
2799 | * operation must restart if the page is not uptodate on unlock but | |
2800 | * otherwise serialising on page lock to stabilise the mapping gives | |
2801 | * no additional guarantees to the caller as the page lock is | |
2802 | * released before return. | |
2803 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2804 | * will be a race with remove_mapping that determines if the mapping | |
2805 | * is valid on unlock but otherwise the data is valid and there is | |
2806 | * no need to serialise with page lock. | |
2807 | * | |
2808 | * As the page lock gives no additional guarantee, we optimistically | |
2809 | * wait on the page to be unlocked and check if it's up to date and | |
2810 | * use the page if it is. Otherwise, the page lock is required to | |
2811 | * distinguish between the different cases. The motivation is that we | |
2812 | * avoid spurious serialisations and wakeups when multiple processes | |
2813 | * wait on the same page for IO to complete. | |
2814 | */ | |
2815 | wait_on_page_locked(page); | |
2816 | if (PageUptodate(page)) | |
2817 | goto out; | |
2818 | ||
2819 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2820 | lock_page(page); |
ebded027 MG |
2821 | |
2822 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2823 | if (!page->mapping) { |
2824 | unlock_page(page); | |
09cbfeaf | 2825 | put_page(page); |
32b63529 | 2826 | goto repeat; |
1da177e4 | 2827 | } |
ebded027 MG |
2828 | |
2829 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2830 | if (PageUptodate(page)) { |
2831 | unlock_page(page); | |
2832 | goto out; | |
2833 | } | |
32b63529 MG |
2834 | goto filler; |
2835 | ||
c855ff37 | 2836 | out: |
6fe6900e NP |
2837 | mark_page_accessed(page); |
2838 | return page; | |
2839 | } | |
0531b2aa LT |
2840 | |
2841 | /** | |
67f9fd91 | 2842 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2843 | * @mapping: the page's address_space |
2844 | * @index: the page index | |
2845 | * @filler: function to perform the read | |
5e5358e7 | 2846 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2847 | * |
0531b2aa | 2848 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2849 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2850 | * |
2851 | * If the page does not get brought uptodate, return -EIO. | |
2852 | */ | |
67f9fd91 | 2853 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2854 | pgoff_t index, |
5e5358e7 | 2855 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2856 | void *data) |
2857 | { | |
2858 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2859 | } | |
67f9fd91 | 2860 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2861 | |
2862 | /** | |
2863 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2864 | * @mapping: the page's address_space | |
2865 | * @index: the page index | |
2866 | * @gfp: the page allocator flags to use if allocating | |
2867 | * | |
2868 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2869 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2870 | * |
2871 | * If the page does not get brought uptodate, return -EIO. | |
2872 | */ | |
2873 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2874 | pgoff_t index, | |
2875 | gfp_t gfp) | |
2876 | { | |
2877 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2878 | ||
67f9fd91 | 2879 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2880 | } |
2881 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2882 | ||
1da177e4 LT |
2883 | /* |
2884 | * Performs necessary checks before doing a write | |
2885 | * | |
485bb99b | 2886 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2887 | * Returns appropriate error code that caller should return or |
2888 | * zero in case that write should be allowed. | |
2889 | */ | |
3309dd04 | 2890 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2891 | { |
3309dd04 | 2892 | struct file *file = iocb->ki_filp; |
1da177e4 | 2893 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2894 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2895 | loff_t pos; |
1da177e4 | 2896 | |
3309dd04 AV |
2897 | if (!iov_iter_count(from)) |
2898 | return 0; | |
1da177e4 | 2899 | |
0fa6b005 | 2900 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2901 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2902 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2903 | |
3309dd04 | 2904 | pos = iocb->ki_pos; |
1da177e4 | 2905 | |
6be96d3a GR |
2906 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
2907 | return -EINVAL; | |
2908 | ||
0fa6b005 | 2909 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2910 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2911 | send_sig(SIGXFSZ, current, 0); |
2912 | return -EFBIG; | |
1da177e4 | 2913 | } |
3309dd04 | 2914 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2915 | } |
2916 | ||
2917 | /* | |
2918 | * LFS rule | |
2919 | */ | |
3309dd04 | 2920 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2921 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2922 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2923 | return -EFBIG; |
3309dd04 | 2924 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2925 | } |
2926 | ||
2927 | /* | |
2928 | * Are we about to exceed the fs block limit ? | |
2929 | * | |
2930 | * If we have written data it becomes a short write. If we have | |
2931 | * exceeded without writing data we send a signal and return EFBIG. | |
2932 | * Linus frestrict idea will clean these up nicely.. | |
2933 | */ | |
3309dd04 AV |
2934 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2935 | return -EFBIG; | |
1da177e4 | 2936 | |
3309dd04 AV |
2937 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2938 | return iov_iter_count(from); | |
1da177e4 LT |
2939 | } |
2940 | EXPORT_SYMBOL(generic_write_checks); | |
2941 | ||
afddba49 NP |
2942 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2943 | loff_t pos, unsigned len, unsigned flags, | |
2944 | struct page **pagep, void **fsdata) | |
2945 | { | |
2946 | const struct address_space_operations *aops = mapping->a_ops; | |
2947 | ||
4e02ed4b | 2948 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2949 | pagep, fsdata); |
afddba49 NP |
2950 | } |
2951 | EXPORT_SYMBOL(pagecache_write_begin); | |
2952 | ||
2953 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2954 | loff_t pos, unsigned len, unsigned copied, | |
2955 | struct page *page, void *fsdata) | |
2956 | { | |
2957 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2958 | |
4e02ed4b | 2959 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2960 | } |
2961 | EXPORT_SYMBOL(pagecache_write_end); | |
2962 | ||
1da177e4 | 2963 | ssize_t |
1af5bb49 | 2964 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2965 | { |
2966 | struct file *file = iocb->ki_filp; | |
2967 | struct address_space *mapping = file->f_mapping; | |
2968 | struct inode *inode = mapping->host; | |
1af5bb49 | 2969 | loff_t pos = iocb->ki_pos; |
1da177e4 | 2970 | ssize_t written; |
a969e903 CH |
2971 | size_t write_len; |
2972 | pgoff_t end; | |
1da177e4 | 2973 | |
0c949334 | 2974 | write_len = iov_iter_count(from); |
09cbfeaf | 2975 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2976 | |
6be96d3a GR |
2977 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2978 | /* If there are pages to writeback, return */ | |
2979 | if (filemap_range_has_page(inode->i_mapping, pos, | |
2980 | pos + iov_iter_count(from))) | |
2981 | return -EAGAIN; | |
2982 | } else { | |
2983 | written = filemap_write_and_wait_range(mapping, pos, | |
2984 | pos + write_len - 1); | |
2985 | if (written) | |
2986 | goto out; | |
2987 | } | |
a969e903 CH |
2988 | |
2989 | /* | |
2990 | * After a write we want buffered reads to be sure to go to disk to get | |
2991 | * the new data. We invalidate clean cached page from the region we're | |
2992 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2993 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 2994 | */ |
55635ba7 | 2995 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 2996 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
2997 | /* |
2998 | * If a page can not be invalidated, return 0 to fall back | |
2999 | * to buffered write. | |
3000 | */ | |
3001 | if (written) { | |
3002 | if (written == -EBUSY) | |
3003 | return 0; | |
3004 | goto out; | |
a969e903 CH |
3005 | } |
3006 | ||
639a93a5 | 3007 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3008 | |
3009 | /* | |
3010 | * Finally, try again to invalidate clean pages which might have been | |
3011 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3012 | * if the source of the write was an mmap'ed region of the file | |
3013 | * we're writing. Either one is a pretty crazy thing to do, | |
3014 | * so we don't support it 100%. If this invalidation | |
3015 | * fails, tough, the write still worked... | |
332391a9 LC |
3016 | * |
3017 | * Most of the time we do not need this since dio_complete() will do | |
3018 | * the invalidation for us. However there are some file systems that | |
3019 | * do not end up with dio_complete() being called, so let's not break | |
3020 | * them by removing it completely | |
a969e903 | 3021 | */ |
332391a9 LC |
3022 | if (mapping->nrpages) |
3023 | invalidate_inode_pages2_range(mapping, | |
3024 | pos >> PAGE_SHIFT, end); | |
a969e903 | 3025 | |
1da177e4 | 3026 | if (written > 0) { |
0116651c | 3027 | pos += written; |
639a93a5 | 3028 | write_len -= written; |
0116651c NK |
3029 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3030 | i_size_write(inode, pos); | |
1da177e4 LT |
3031 | mark_inode_dirty(inode); |
3032 | } | |
5cb6c6c7 | 3033 | iocb->ki_pos = pos; |
1da177e4 | 3034 | } |
639a93a5 | 3035 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3036 | out: |
1da177e4 LT |
3037 | return written; |
3038 | } | |
3039 | EXPORT_SYMBOL(generic_file_direct_write); | |
3040 | ||
eb2be189 NP |
3041 | /* |
3042 | * Find or create a page at the given pagecache position. Return the locked | |
3043 | * page. This function is specifically for buffered writes. | |
3044 | */ | |
54566b2c NP |
3045 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3046 | pgoff_t index, unsigned flags) | |
eb2be189 | 3047 | { |
eb2be189 | 3048 | struct page *page; |
bbddabe2 | 3049 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3050 | |
54566b2c | 3051 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3052 | fgp_flags |= FGP_NOFS; |
3053 | ||
3054 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3055 | mapping_gfp_mask(mapping)); |
c585a267 | 3056 | if (page) |
2457aec6 | 3057 | wait_for_stable_page(page); |
eb2be189 | 3058 | |
eb2be189 NP |
3059 | return page; |
3060 | } | |
54566b2c | 3061 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3062 | |
3b93f911 | 3063 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3064 | struct iov_iter *i, loff_t pos) |
3065 | { | |
3066 | struct address_space *mapping = file->f_mapping; | |
3067 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3068 | long status = 0; | |
3069 | ssize_t written = 0; | |
674b892e NP |
3070 | unsigned int flags = 0; |
3071 | ||
afddba49 NP |
3072 | do { |
3073 | struct page *page; | |
afddba49 NP |
3074 | unsigned long offset; /* Offset into pagecache page */ |
3075 | unsigned long bytes; /* Bytes to write to page */ | |
3076 | size_t copied; /* Bytes copied from user */ | |
3077 | void *fsdata; | |
3078 | ||
09cbfeaf KS |
3079 | offset = (pos & (PAGE_SIZE - 1)); |
3080 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3081 | iov_iter_count(i)); |
3082 | ||
3083 | again: | |
00a3d660 LT |
3084 | /* |
3085 | * Bring in the user page that we will copy from _first_. | |
3086 | * Otherwise there's a nasty deadlock on copying from the | |
3087 | * same page as we're writing to, without it being marked | |
3088 | * up-to-date. | |
3089 | * | |
3090 | * Not only is this an optimisation, but it is also required | |
3091 | * to check that the address is actually valid, when atomic | |
3092 | * usercopies are used, below. | |
3093 | */ | |
3094 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3095 | status = -EFAULT; | |
3096 | break; | |
3097 | } | |
3098 | ||
296291cd JK |
3099 | if (fatal_signal_pending(current)) { |
3100 | status = -EINTR; | |
3101 | break; | |
3102 | } | |
3103 | ||
674b892e | 3104 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3105 | &page, &fsdata); |
2457aec6 | 3106 | if (unlikely(status < 0)) |
afddba49 NP |
3107 | break; |
3108 | ||
931e80e4 | 3109 | if (mapping_writably_mapped(mapping)) |
3110 | flush_dcache_page(page); | |
00a3d660 | 3111 | |
afddba49 | 3112 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3113 | flush_dcache_page(page); |
3114 | ||
3115 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3116 | page, fsdata); | |
3117 | if (unlikely(status < 0)) | |
3118 | break; | |
3119 | copied = status; | |
3120 | ||
3121 | cond_resched(); | |
3122 | ||
124d3b70 | 3123 | iov_iter_advance(i, copied); |
afddba49 NP |
3124 | if (unlikely(copied == 0)) { |
3125 | /* | |
3126 | * If we were unable to copy any data at all, we must | |
3127 | * fall back to a single segment length write. | |
3128 | * | |
3129 | * If we didn't fallback here, we could livelock | |
3130 | * because not all segments in the iov can be copied at | |
3131 | * once without a pagefault. | |
3132 | */ | |
09cbfeaf | 3133 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3134 | iov_iter_single_seg_count(i)); |
3135 | goto again; | |
3136 | } | |
afddba49 NP |
3137 | pos += copied; |
3138 | written += copied; | |
3139 | ||
3140 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3141 | } while (iov_iter_count(i)); |
3142 | ||
3143 | return written ? written : status; | |
3144 | } | |
3b93f911 | 3145 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3146 | |
e4dd9de3 | 3147 | /** |
8174202b | 3148 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3149 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3150 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3151 | * |
3152 | * This function does all the work needed for actually writing data to a | |
3153 | * file. It does all basic checks, removes SUID from the file, updates | |
3154 | * modification times and calls proper subroutines depending on whether we | |
3155 | * do direct IO or a standard buffered write. | |
3156 | * | |
3157 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3158 | * object which does not need locking at all. | |
3159 | * | |
3160 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3161 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3162 | * avoid syncing under i_mutex. | |
3163 | */ | |
8174202b | 3164 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3165 | { |
3166 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3167 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3168 | struct inode *inode = mapping->host; |
3b93f911 | 3169 | ssize_t written = 0; |
1da177e4 | 3170 | ssize_t err; |
3b93f911 | 3171 | ssize_t status; |
1da177e4 | 3172 | |
1da177e4 | 3173 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3174 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3175 | err = file_remove_privs(file); |
1da177e4 LT |
3176 | if (err) |
3177 | goto out; | |
3178 | ||
c3b2da31 JB |
3179 | err = file_update_time(file); |
3180 | if (err) | |
3181 | goto out; | |
1da177e4 | 3182 | |
2ba48ce5 | 3183 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3184 | loff_t pos, endbyte; |
fb5527e6 | 3185 | |
1af5bb49 | 3186 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3187 | /* |
fbbbad4b MW |
3188 | * If the write stopped short of completing, fall back to |
3189 | * buffered writes. Some filesystems do this for writes to | |
3190 | * holes, for example. For DAX files, a buffered write will | |
3191 | * not succeed (even if it did, DAX does not handle dirty | |
3192 | * page-cache pages correctly). | |
1da177e4 | 3193 | */ |
0b8def9d | 3194 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3195 | goto out; |
3196 | ||
0b8def9d | 3197 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3198 | /* |
3b93f911 | 3199 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3200 | * then we want to return the number of bytes which were |
3201 | * direct-written, or the error code if that was zero. Note | |
3202 | * that this differs from normal direct-io semantics, which | |
3203 | * will return -EFOO even if some bytes were written. | |
3204 | */ | |
60bb4529 | 3205 | if (unlikely(status < 0)) { |
3b93f911 | 3206 | err = status; |
fb5527e6 JM |
3207 | goto out; |
3208 | } | |
fb5527e6 JM |
3209 | /* |
3210 | * We need to ensure that the page cache pages are written to | |
3211 | * disk and invalidated to preserve the expected O_DIRECT | |
3212 | * semantics. | |
3213 | */ | |
3b93f911 | 3214 | endbyte = pos + status - 1; |
0b8def9d | 3215 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3216 | if (err == 0) { |
0b8def9d | 3217 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3218 | written += status; |
fb5527e6 | 3219 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3220 | pos >> PAGE_SHIFT, |
3221 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3222 | } else { |
3223 | /* | |
3224 | * We don't know how much we wrote, so just return | |
3225 | * the number of bytes which were direct-written | |
3226 | */ | |
3227 | } | |
3228 | } else { | |
0b8def9d AV |
3229 | written = generic_perform_write(file, from, iocb->ki_pos); |
3230 | if (likely(written > 0)) | |
3231 | iocb->ki_pos += written; | |
fb5527e6 | 3232 | } |
1da177e4 LT |
3233 | out: |
3234 | current->backing_dev_info = NULL; | |
3235 | return written ? written : err; | |
3236 | } | |
8174202b | 3237 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3238 | |
e4dd9de3 | 3239 | /** |
8174202b | 3240 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3241 | * @iocb: IO state structure |
8174202b | 3242 | * @from: iov_iter with data to write |
e4dd9de3 | 3243 | * |
8174202b | 3244 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3245 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3246 | * and acquires i_mutex as needed. | |
3247 | */ | |
8174202b | 3248 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3249 | { |
3250 | struct file *file = iocb->ki_filp; | |
148f948b | 3251 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3252 | ssize_t ret; |
1da177e4 | 3253 | |
5955102c | 3254 | inode_lock(inode); |
3309dd04 AV |
3255 | ret = generic_write_checks(iocb, from); |
3256 | if (ret > 0) | |
5f380c7f | 3257 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3258 | inode_unlock(inode); |
1da177e4 | 3259 | |
e2592217 CH |
3260 | if (ret > 0) |
3261 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3262 | return ret; |
3263 | } | |
8174202b | 3264 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3265 | |
cf9a2ae8 DH |
3266 | /** |
3267 | * try_to_release_page() - release old fs-specific metadata on a page | |
3268 | * | |
3269 | * @page: the page which the kernel is trying to free | |
3270 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3271 | * | |
3272 | * The address_space is to try to release any data against the page | |
0e056eb5 | 3273 | * (presumably at page->private). If the release was successful, return '1'. |
cf9a2ae8 DH |
3274 | * Otherwise return zero. |
3275 | * | |
266cf658 DH |
3276 | * This may also be called if PG_fscache is set on a page, indicating that the |
3277 | * page is known to the local caching routines. | |
3278 | * | |
cf9a2ae8 | 3279 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3280 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3281 | * |
cf9a2ae8 DH |
3282 | */ |
3283 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3284 | { | |
3285 | struct address_space * const mapping = page->mapping; | |
3286 | ||
3287 | BUG_ON(!PageLocked(page)); | |
3288 | if (PageWriteback(page)) | |
3289 | return 0; | |
3290 | ||
3291 | if (mapping && mapping->a_ops->releasepage) | |
3292 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3293 | return try_to_free_buffers(page); | |
3294 | } | |
3295 | ||
3296 | EXPORT_SYMBOL(try_to_release_page); |