]>
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> |
2f718ffc | 34 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 35 | #include <linux/hugetlb.h> |
8a9f3ccd | 36 | #include <linux/memcontrol.h> |
c515e1fd | 37 | #include <linux/cleancache.h> |
c7df8ad2 | 38 | #include <linux/shmem_fs.h> |
f1820361 | 39 | #include <linux/rmap.h> |
0f8053a5 NP |
40 | #include "internal.h" |
41 | ||
fe0bfaaf RJ |
42 | #define CREATE_TRACE_POINTS |
43 | #include <trace/events/filemap.h> | |
44 | ||
1da177e4 | 45 | /* |
1da177e4 LT |
46 | * FIXME: remove all knowledge of the buffer layer from the core VM |
47 | */ | |
148f948b | 48 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 49 | |
1da177e4 LT |
50 | #include <asm/mman.h> |
51 | ||
52 | /* | |
53 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
54 | * though. | |
55 | * | |
56 | * Shared mappings now work. 15.8.1995 Bruno. | |
57 | * | |
58 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
59 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
60 | * | |
61 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
62 | */ | |
63 | ||
64 | /* | |
65 | * Lock ordering: | |
66 | * | |
c8c06efa | 67 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 68 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
69 | * ->swap_lock (exclusive_swap_page, others) |
70 | * ->mapping->tree_lock | |
1da177e4 | 71 | * |
1b1dcc1b | 72 | * ->i_mutex |
c8c06efa | 73 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
74 | * |
75 | * ->mmap_sem | |
c8c06efa | 76 | * ->i_mmap_rwsem |
b8072f09 | 77 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
78 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
79 | * | |
80 | * ->mmap_sem | |
81 | * ->lock_page (access_process_vm) | |
82 | * | |
ccad2365 | 83 | * ->i_mutex (generic_perform_write) |
82591e6e | 84 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 85 | * |
f758eeab | 86 | * bdi->wb.list_lock |
a66979ab | 87 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
88 | * ->mapping->tree_lock (__sync_single_inode) |
89 | * | |
c8c06efa | 90 | * ->i_mmap_rwsem |
1da177e4 LT |
91 | * ->anon_vma.lock (vma_adjust) |
92 | * | |
93 | * ->anon_vma.lock | |
b8072f09 | 94 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 95 | * |
b8072f09 | 96 | * ->page_table_lock or pte_lock |
5d337b91 | 97 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
98 | * ->private_lock (try_to_unmap_one) |
99 | * ->tree_lock (try_to_unmap_one) | |
a52633d8 MG |
100 | * ->zone_lru_lock(zone) (follow_page->mark_page_accessed) |
101 | * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page) | |
1da177e4 LT |
102 | * ->private_lock (page_remove_rmap->set_page_dirty) |
103 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 104 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 105 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 106 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 107 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 108 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
109 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
110 | * | |
c8c06efa | 111 | * ->i_mmap_rwsem |
9a3c531d | 112 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
113 | */ |
114 | ||
22f2ac51 JW |
115 | static int page_cache_tree_insert(struct address_space *mapping, |
116 | struct page *page, void **shadowp) | |
117 | { | |
118 | struct radix_tree_node *node; | |
119 | void **slot; | |
120 | int error; | |
121 | ||
122 | error = __radix_tree_create(&mapping->page_tree, page->index, 0, | |
123 | &node, &slot); | |
124 | if (error) | |
125 | return error; | |
126 | if (*slot) { | |
127 | void *p; | |
128 | ||
129 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
130 | if (!radix_tree_exceptional_entry(p)) | |
131 | return -EEXIST; | |
132 | ||
133 | mapping->nrexceptional--; | |
d01ad197 RZ |
134 | if (shadowp) |
135 | *shadowp = p; | |
22f2ac51 | 136 | } |
14b46879 | 137 | __radix_tree_replace(&mapping->page_tree, node, slot, page, |
c7df8ad2 | 138 | workingset_lookup_update(mapping)); |
22f2ac51 | 139 | mapping->nrpages++; |
22f2ac51 JW |
140 | return 0; |
141 | } | |
142 | ||
91b0abe3 JW |
143 | static void page_cache_tree_delete(struct address_space *mapping, |
144 | struct page *page, void *shadow) | |
145 | { | |
c70b647d KS |
146 | int i, nr; |
147 | ||
148 | /* hugetlb pages are represented by one entry in the radix tree */ | |
149 | nr = PageHuge(page) ? 1 : hpage_nr_pages(page); | |
91b0abe3 | 150 | |
83929372 KS |
151 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
152 | VM_BUG_ON_PAGE(PageTail(page), page); | |
153 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 154 | |
83929372 | 155 | for (i = 0; i < nr; i++) { |
d3798ae8 JW |
156 | struct radix_tree_node *node; |
157 | void **slot; | |
158 | ||
159 | __radix_tree_lookup(&mapping->page_tree, page->index + i, | |
160 | &node, &slot); | |
161 | ||
dbc446b8 | 162 | VM_BUG_ON_PAGE(!node && nr != 1, page); |
449dd698 | 163 | |
14b46879 JW |
164 | radix_tree_clear_tags(&mapping->page_tree, node, slot); |
165 | __radix_tree_replace(&mapping->page_tree, node, slot, shadow, | |
c7df8ad2 | 166 | workingset_lookup_update(mapping)); |
449dd698 | 167 | } |
d3798ae8 | 168 | |
2300638b JK |
169 | page->mapping = NULL; |
170 | /* Leave page->index set: truncation lookup relies upon it */ | |
171 | ||
d3798ae8 JW |
172 | if (shadow) { |
173 | mapping->nrexceptional += nr; | |
174 | /* | |
175 | * Make sure the nrexceptional update is committed before | |
176 | * the nrpages update so that final truncate racing | |
177 | * with reclaim does not see both counters 0 at the | |
178 | * same time and miss a shadow entry. | |
179 | */ | |
180 | smp_wmb(); | |
181 | } | |
182 | mapping->nrpages -= nr; | |
91b0abe3 JW |
183 | } |
184 | ||
5ecc4d85 JK |
185 | static void unaccount_page_cache_page(struct address_space *mapping, |
186 | struct page *page) | |
1da177e4 | 187 | { |
5ecc4d85 | 188 | int nr; |
1da177e4 | 189 | |
c515e1fd DM |
190 | /* |
191 | * if we're uptodate, flush out into the cleancache, otherwise | |
192 | * invalidate any existing cleancache entries. We can't leave | |
193 | * stale data around in the cleancache once our page is gone | |
194 | */ | |
195 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
196 | cleancache_put_page(page); | |
197 | else | |
3167760f | 198 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 199 | |
83929372 | 200 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
201 | VM_BUG_ON_PAGE(page_mapped(page), page); |
202 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
203 | int mapcount; | |
204 | ||
205 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
206 | current->comm, page_to_pfn(page)); | |
207 | dump_page(page, "still mapped when deleted"); | |
208 | dump_stack(); | |
209 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
210 | ||
211 | mapcount = page_mapcount(page); | |
212 | if (mapping_exiting(mapping) && | |
213 | page_count(page) >= mapcount + 2) { | |
214 | /* | |
215 | * All vmas have already been torn down, so it's | |
216 | * a good bet that actually the page is unmapped, | |
217 | * and we'd prefer not to leak it: if we're wrong, | |
218 | * some other bad page check should catch it later. | |
219 | */ | |
220 | page_mapcount_reset(page); | |
6d061f9f | 221 | page_ref_sub(page, mapcount); |
06b241f3 HD |
222 | } |
223 | } | |
224 | ||
4165b9b4 | 225 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
226 | if (PageHuge(page)) |
227 | return; | |
09612fa6 | 228 | |
5ecc4d85 JK |
229 | nr = hpage_nr_pages(page); |
230 | ||
231 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); | |
232 | if (PageSwapBacked(page)) { | |
233 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); | |
234 | if (PageTransHuge(page)) | |
235 | __dec_node_page_state(page, NR_SHMEM_THPS); | |
236 | } else { | |
237 | VM_BUG_ON_PAGE(PageTransHuge(page), page); | |
800d8c63 | 238 | } |
5ecc4d85 JK |
239 | |
240 | /* | |
241 | * At this point page must be either written or cleaned by | |
242 | * truncate. Dirty page here signals a bug and loss of | |
243 | * unwritten data. | |
244 | * | |
245 | * This fixes dirty accounting after removing the page entirely | |
246 | * but leaves PageDirty set: it has no effect for truncated | |
247 | * page and anyway will be cleared before returning page into | |
248 | * buddy allocator. | |
249 | */ | |
250 | if (WARN_ON_ONCE(PageDirty(page))) | |
251 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
252 | } | |
253 | ||
254 | /* | |
255 | * Delete a page from the page cache and free it. Caller has to make | |
256 | * sure the page is locked and that nobody else uses it - or that usage | |
257 | * is safe. The caller must hold the mapping's tree_lock. | |
258 | */ | |
259 | void __delete_from_page_cache(struct page *page, void *shadow) | |
260 | { | |
261 | struct address_space *mapping = page->mapping; | |
262 | ||
263 | trace_mm_filemap_delete_from_page_cache(page); | |
264 | ||
265 | unaccount_page_cache_page(mapping, page); | |
76253fbc | 266 | page_cache_tree_delete(mapping, page, shadow); |
1da177e4 LT |
267 | } |
268 | ||
59c66c5f JK |
269 | static void page_cache_free_page(struct address_space *mapping, |
270 | struct page *page) | |
271 | { | |
272 | void (*freepage)(struct page *); | |
273 | ||
274 | freepage = mapping->a_ops->freepage; | |
275 | if (freepage) | |
276 | freepage(page); | |
277 | ||
278 | if (PageTransHuge(page) && !PageHuge(page)) { | |
279 | page_ref_sub(page, HPAGE_PMD_NR); | |
280 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
281 | } else { | |
282 | put_page(page); | |
283 | } | |
284 | } | |
285 | ||
702cfbf9 MK |
286 | /** |
287 | * delete_from_page_cache - delete page from page cache | |
288 | * @page: the page which the kernel is trying to remove from page cache | |
289 | * | |
290 | * This must be called only on pages that have been verified to be in the page | |
291 | * cache and locked. It will never put the page into the free list, the caller | |
292 | * has a reference on the page. | |
293 | */ | |
294 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 295 | { |
83929372 | 296 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 297 | unsigned long flags; |
1da177e4 | 298 | |
cd7619d6 | 299 | BUG_ON(!PageLocked(page)); |
c4843a75 | 300 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 301 | __delete_from_page_cache(page, NULL); |
c4843a75 | 302 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c | 303 | |
59c66c5f | 304 | page_cache_free_page(mapping, page); |
97cecb5a MK |
305 | } |
306 | EXPORT_SYMBOL(delete_from_page_cache); | |
307 | ||
aa65c29c JK |
308 | /* |
309 | * page_cache_tree_delete_batch - delete several pages from page cache | |
310 | * @mapping: the mapping to which pages belong | |
311 | * @pvec: pagevec with pages to delete | |
312 | * | |
313 | * The function walks over mapping->page_tree and removes pages passed in @pvec | |
314 | * from the radix tree. The function expects @pvec to be sorted by page index. | |
315 | * It tolerates holes in @pvec (radix tree entries at those indices are not | |
316 | * modified). The function expects only THP head pages to be present in the | |
317 | * @pvec and takes care to delete all corresponding tail pages from the radix | |
318 | * tree as well. | |
319 | * | |
320 | * The function expects mapping->tree_lock to be held. | |
321 | */ | |
322 | static void | |
323 | page_cache_tree_delete_batch(struct address_space *mapping, | |
324 | struct pagevec *pvec) | |
325 | { | |
326 | struct radix_tree_iter iter; | |
327 | void **slot; | |
328 | int total_pages = 0; | |
329 | int i = 0, tail_pages = 0; | |
330 | struct page *page; | |
331 | pgoff_t start; | |
332 | ||
333 | start = pvec->pages[0]->index; | |
334 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
335 | if (i >= pagevec_count(pvec) && !tail_pages) | |
336 | break; | |
337 | page = radix_tree_deref_slot_protected(slot, | |
338 | &mapping->tree_lock); | |
339 | if (radix_tree_exceptional_entry(page)) | |
340 | continue; | |
341 | if (!tail_pages) { | |
342 | /* | |
343 | * Some page got inserted in our range? Skip it. We | |
344 | * have our pages locked so they are protected from | |
345 | * being removed. | |
346 | */ | |
347 | if (page != pvec->pages[i]) | |
348 | continue; | |
349 | WARN_ON_ONCE(!PageLocked(page)); | |
350 | if (PageTransHuge(page) && !PageHuge(page)) | |
351 | tail_pages = HPAGE_PMD_NR - 1; | |
352 | page->mapping = NULL; | |
353 | /* | |
354 | * Leave page->index set: truncation lookup relies | |
355 | * upon it | |
356 | */ | |
357 | i++; | |
358 | } else { | |
359 | tail_pages--; | |
360 | } | |
361 | radix_tree_clear_tags(&mapping->page_tree, iter.node, slot); | |
362 | __radix_tree_replace(&mapping->page_tree, iter.node, slot, NULL, | |
c7df8ad2 | 363 | workingset_lookup_update(mapping)); |
aa65c29c JK |
364 | total_pages++; |
365 | } | |
366 | mapping->nrpages -= total_pages; | |
367 | } | |
368 | ||
369 | void delete_from_page_cache_batch(struct address_space *mapping, | |
370 | struct pagevec *pvec) | |
371 | { | |
372 | int i; | |
373 | unsigned long flags; | |
374 | ||
375 | if (!pagevec_count(pvec)) | |
376 | return; | |
377 | ||
378 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
379 | for (i = 0; i < pagevec_count(pvec); i++) { | |
380 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
381 | ||
382 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
383 | } | |
384 | page_cache_tree_delete_batch(mapping, pvec); | |
385 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
386 | ||
387 | for (i = 0; i < pagevec_count(pvec); i++) | |
388 | page_cache_free_page(mapping, pvec->pages[i]); | |
389 | } | |
390 | ||
d72d9e2a | 391 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
392 | { |
393 | int ret = 0; | |
394 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
395 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
396 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 397 | ret = -ENOSPC; |
7fcbbaf1 JA |
398 | if (test_bit(AS_EIO, &mapping->flags) && |
399 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
400 | ret = -EIO; |
401 | return ret; | |
402 | } | |
d72d9e2a | 403 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 404 | |
76341cab JL |
405 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
406 | { | |
407 | /* Check for outstanding write errors */ | |
408 | if (test_bit(AS_EIO, &mapping->flags)) | |
409 | return -EIO; | |
410 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
411 | return -ENOSPC; | |
412 | return 0; | |
413 | } | |
414 | ||
1da177e4 | 415 | /** |
485bb99b | 416 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
417 | * @mapping: address space structure to write |
418 | * @start: offset in bytes where the range starts | |
469eb4d0 | 419 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 420 | * @sync_mode: enable synchronous operation |
1da177e4 | 421 | * |
485bb99b RD |
422 | * Start writeback against all of a mapping's dirty pages that lie |
423 | * within the byte offsets <start, end> inclusive. | |
424 | * | |
1da177e4 | 425 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 426 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
427 | * these two operations is that if a dirty page/buffer is encountered, it must |
428 | * be waited upon, and not just skipped over. | |
429 | */ | |
ebcf28e1 AM |
430 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
431 | loff_t end, int sync_mode) | |
1da177e4 LT |
432 | { |
433 | int ret; | |
434 | struct writeback_control wbc = { | |
435 | .sync_mode = sync_mode, | |
05fe478d | 436 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
437 | .range_start = start, |
438 | .range_end = end, | |
1da177e4 LT |
439 | }; |
440 | ||
8b738c71 KK |
441 | if (!mapping_cap_writeback_dirty(mapping) || |
442 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
1da177e4 LT |
443 | return 0; |
444 | ||
b16b1deb | 445 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 446 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 447 | wbc_detach_inode(&wbc); |
1da177e4 LT |
448 | return ret; |
449 | } | |
450 | ||
451 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
452 | int sync_mode) | |
453 | { | |
111ebb6e | 454 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
455 | } |
456 | ||
457 | int filemap_fdatawrite(struct address_space *mapping) | |
458 | { | |
459 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
460 | } | |
461 | EXPORT_SYMBOL(filemap_fdatawrite); | |
462 | ||
f4c0a0fd | 463 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 464 | loff_t end) |
1da177e4 LT |
465 | { |
466 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
467 | } | |
f4c0a0fd | 468 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 469 | |
485bb99b RD |
470 | /** |
471 | * filemap_flush - mostly a non-blocking flush | |
472 | * @mapping: target address_space | |
473 | * | |
1da177e4 LT |
474 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
475 | * purposes - I/O may not be started against all dirty pages. | |
476 | */ | |
477 | int filemap_flush(struct address_space *mapping) | |
478 | { | |
479 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
480 | } | |
481 | EXPORT_SYMBOL(filemap_flush); | |
482 | ||
7fc9e472 GR |
483 | /** |
484 | * filemap_range_has_page - check if a page exists in range. | |
485 | * @mapping: address space within which to check | |
486 | * @start_byte: offset in bytes where the range starts | |
487 | * @end_byte: offset in bytes where the range ends (inclusive) | |
488 | * | |
489 | * Find at least one page in the range supplied, usually used to check if | |
490 | * direct writing in this range will trigger a writeback. | |
491 | */ | |
492 | bool filemap_range_has_page(struct address_space *mapping, | |
493 | loff_t start_byte, loff_t end_byte) | |
494 | { | |
495 | pgoff_t index = start_byte >> PAGE_SHIFT; | |
496 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
f7b68046 | 497 | struct page *page; |
7fc9e472 GR |
498 | |
499 | if (end_byte < start_byte) | |
500 | return false; | |
501 | ||
502 | if (mapping->nrpages == 0) | |
503 | return false; | |
504 | ||
f7b68046 | 505 | if (!find_get_pages_range(mapping, &index, end, 1, &page)) |
7fc9e472 | 506 | return false; |
f7b68046 JK |
507 | put_page(page); |
508 | return true; | |
7fc9e472 GR |
509 | } |
510 | EXPORT_SYMBOL(filemap_range_has_page); | |
511 | ||
5e8fcc1a | 512 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 513 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 514 | { |
09cbfeaf KS |
515 | pgoff_t index = start_byte >> PAGE_SHIFT; |
516 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
517 | struct pagevec pvec; |
518 | int nr_pages; | |
1da177e4 | 519 | |
94004ed7 | 520 | if (end_byte < start_byte) |
5e8fcc1a | 521 | return; |
1da177e4 | 522 | |
86679820 | 523 | pagevec_init(&pvec); |
312e9d2f | 524 | while (index <= end) { |
1da177e4 LT |
525 | unsigned i; |
526 | ||
312e9d2f | 527 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 528 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
529 | if (!nr_pages) |
530 | break; | |
531 | ||
1da177e4 LT |
532 | for (i = 0; i < nr_pages; i++) { |
533 | struct page *page = pvec.pages[i]; | |
534 | ||
1da177e4 | 535 | wait_on_page_writeback(page); |
5e8fcc1a | 536 | ClearPageError(page); |
1da177e4 LT |
537 | } |
538 | pagevec_release(&pvec); | |
539 | cond_resched(); | |
540 | } | |
aa750fd7 JN |
541 | } |
542 | ||
543 | /** | |
544 | * filemap_fdatawait_range - wait for writeback to complete | |
545 | * @mapping: address space structure to wait for | |
546 | * @start_byte: offset in bytes where the range starts | |
547 | * @end_byte: offset in bytes where the range ends (inclusive) | |
548 | * | |
549 | * Walk the list of under-writeback pages of the given address space | |
550 | * in the given range and wait for all of them. Check error status of | |
551 | * the address space and return it. | |
552 | * | |
553 | * Since the error status of the address space is cleared by this function, | |
554 | * callers are responsible for checking the return value and handling and/or | |
555 | * reporting the error. | |
556 | */ | |
557 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
558 | loff_t end_byte) | |
559 | { | |
5e8fcc1a JL |
560 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
561 | return filemap_check_errors(mapping); | |
1da177e4 | 562 | } |
d3bccb6f JK |
563 | EXPORT_SYMBOL(filemap_fdatawait_range); |
564 | ||
9a6a10e6 RZ |
565 | /** |
566 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
567 | * @mapping: address space structure to wait for | |
568 | * @start_byte: offset in bytes where the range starts | |
569 | * @end_byte: offset in bytes where the range ends (inclusive) | |
570 | * | |
571 | * Walk the list of under-writeback pages of the given address space in the | |
572 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
573 | * this function does not clear error status of the address space. | |
574 | * | |
575 | * Use this function if callers don't handle errors themselves. Expected | |
576 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
577 | * fsfreeze(8) | |
578 | */ | |
579 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
580 | loff_t start_byte, loff_t end_byte) | |
581 | { | |
582 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
583 | return filemap_check_and_keep_errors(mapping); | |
584 | } | |
585 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
586 | ||
a823e458 JL |
587 | /** |
588 | * file_fdatawait_range - wait for writeback to complete | |
589 | * @file: file pointing to address space structure to wait for | |
590 | * @start_byte: offset in bytes where the range starts | |
591 | * @end_byte: offset in bytes where the range ends (inclusive) | |
592 | * | |
593 | * Walk the list of under-writeback pages of the address space that file | |
594 | * refers to, in the given range and wait for all of them. Check error | |
595 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
596 | * | |
597 | * Since the error status of the file is advanced by this function, | |
598 | * callers are responsible for checking the return value and handling and/or | |
599 | * reporting the error. | |
600 | */ | |
601 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
602 | { | |
603 | struct address_space *mapping = file->f_mapping; | |
604 | ||
605 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
606 | return file_check_and_advance_wb_err(file); | |
607 | } | |
608 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 609 | |
aa750fd7 JN |
610 | /** |
611 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
612 | * @mapping: address space structure to wait for | |
613 | * | |
614 | * Walk the list of under-writeback pages of the given address space | |
615 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
616 | * does not clear error status of the address space. | |
617 | * | |
618 | * Use this function if callers don't handle errors themselves. Expected | |
619 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
620 | * fsfreeze(8) | |
621 | */ | |
76341cab | 622 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 623 | { |
ffb959bb | 624 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 625 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 626 | } |
76341cab | 627 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 628 | |
9326c9b2 | 629 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 630 | { |
9326c9b2 JL |
631 | return (!dax_mapping(mapping) && mapping->nrpages) || |
632 | (dax_mapping(mapping) && mapping->nrexceptional); | |
1da177e4 | 633 | } |
1da177e4 LT |
634 | |
635 | int filemap_write_and_wait(struct address_space *mapping) | |
636 | { | |
28fd1298 | 637 | int err = 0; |
1da177e4 | 638 | |
9326c9b2 | 639 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
640 | err = filemap_fdatawrite(mapping); |
641 | /* | |
642 | * Even if the above returned error, the pages may be | |
643 | * written partially (e.g. -ENOSPC), so we wait for it. | |
644 | * But the -EIO is special case, it may indicate the worst | |
645 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
646 | */ | |
647 | if (err != -EIO) { | |
648 | int err2 = filemap_fdatawait(mapping); | |
649 | if (!err) | |
650 | err = err2; | |
cbeaf951 JL |
651 | } else { |
652 | /* Clear any previously stored errors */ | |
653 | filemap_check_errors(mapping); | |
28fd1298 | 654 | } |
865ffef3 DM |
655 | } else { |
656 | err = filemap_check_errors(mapping); | |
1da177e4 | 657 | } |
28fd1298 | 658 | return err; |
1da177e4 | 659 | } |
28fd1298 | 660 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 661 | |
485bb99b RD |
662 | /** |
663 | * filemap_write_and_wait_range - write out & wait on a file range | |
664 | * @mapping: the address_space for the pages | |
665 | * @lstart: offset in bytes where the range starts | |
666 | * @lend: offset in bytes where the range ends (inclusive) | |
667 | * | |
469eb4d0 AM |
668 | * Write out and wait upon file offsets lstart->lend, inclusive. |
669 | * | |
0e056eb5 | 670 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 AM |
671 | * that this function can be used to write to the very end-of-file (end = -1). |
672 | */ | |
1da177e4 LT |
673 | int filemap_write_and_wait_range(struct address_space *mapping, |
674 | loff_t lstart, loff_t lend) | |
675 | { | |
28fd1298 | 676 | int err = 0; |
1da177e4 | 677 | |
9326c9b2 | 678 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
679 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
680 | WB_SYNC_ALL); | |
681 | /* See comment of filemap_write_and_wait() */ | |
682 | if (err != -EIO) { | |
94004ed7 CH |
683 | int err2 = filemap_fdatawait_range(mapping, |
684 | lstart, lend); | |
28fd1298 OH |
685 | if (!err) |
686 | err = err2; | |
cbeaf951 JL |
687 | } else { |
688 | /* Clear any previously stored errors */ | |
689 | filemap_check_errors(mapping); | |
28fd1298 | 690 | } |
865ffef3 DM |
691 | } else { |
692 | err = filemap_check_errors(mapping); | |
1da177e4 | 693 | } |
28fd1298 | 694 | return err; |
1da177e4 | 695 | } |
f6995585 | 696 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 697 | |
5660e13d JL |
698 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
699 | { | |
3acdfd28 | 700 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
701 | |
702 | trace_filemap_set_wb_err(mapping, eseq); | |
703 | } | |
704 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
705 | ||
706 | /** | |
707 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
708 | * and advance wb_err to current one | |
709 | * @file: struct file on which the error is being reported | |
710 | * | |
711 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
712 | * want to report any writeback errors that occurred since the last fsync (or | |
713 | * since the file was opened if there haven't been any). | |
714 | * | |
715 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
716 | * then just quickly return 0. The file is all caught up. | |
717 | * | |
718 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
719 | * it and try to swap it into place. If it works, or another task beat us | |
720 | * to it with the new value, then update the f_wb_err and return the error | |
721 | * portion. The error at this point must be reported via proper channels | |
722 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
723 | * | |
724 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
725 | * value is protected by the f_lock since we must ensure that it reflects | |
726 | * the latest value swapped in for this file descriptor. | |
727 | */ | |
728 | int file_check_and_advance_wb_err(struct file *file) | |
729 | { | |
730 | int err = 0; | |
731 | errseq_t old = READ_ONCE(file->f_wb_err); | |
732 | struct address_space *mapping = file->f_mapping; | |
733 | ||
734 | /* Locklessly handle the common case where nothing has changed */ | |
735 | if (errseq_check(&mapping->wb_err, old)) { | |
736 | /* Something changed, must use slow path */ | |
737 | spin_lock(&file->f_lock); | |
738 | old = file->f_wb_err; | |
739 | err = errseq_check_and_advance(&mapping->wb_err, | |
740 | &file->f_wb_err); | |
741 | trace_file_check_and_advance_wb_err(file, old); | |
742 | spin_unlock(&file->f_lock); | |
743 | } | |
f4e222c5 JL |
744 | |
745 | /* | |
746 | * We're mostly using this function as a drop in replacement for | |
747 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
748 | * that the legacy code would have had on these flags. | |
749 | */ | |
750 | clear_bit(AS_EIO, &mapping->flags); | |
751 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
752 | return err; |
753 | } | |
754 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
755 | ||
756 | /** | |
757 | * file_write_and_wait_range - write out & wait on a file range | |
758 | * @file: file pointing to address_space with pages | |
759 | * @lstart: offset in bytes where the range starts | |
760 | * @lend: offset in bytes where the range ends (inclusive) | |
761 | * | |
762 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
763 | * | |
764 | * Note that @lend is inclusive (describes the last byte to be written) so | |
765 | * that this function can be used to write to the very end-of-file (end = -1). | |
766 | * | |
767 | * After writing out and waiting on the data, we check and advance the | |
768 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
769 | */ | |
770 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
771 | { | |
772 | int err = 0, err2; | |
773 | struct address_space *mapping = file->f_mapping; | |
774 | ||
9326c9b2 | 775 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
776 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
777 | WB_SYNC_ALL); | |
778 | /* See comment of filemap_write_and_wait() */ | |
779 | if (err != -EIO) | |
780 | __filemap_fdatawait_range(mapping, lstart, lend); | |
781 | } | |
782 | err2 = file_check_and_advance_wb_err(file); | |
783 | if (!err) | |
784 | err = err2; | |
785 | return err; | |
786 | } | |
787 | EXPORT_SYMBOL(file_write_and_wait_range); | |
788 | ||
ef6a3c63 MS |
789 | /** |
790 | * replace_page_cache_page - replace a pagecache page with a new one | |
791 | * @old: page to be replaced | |
792 | * @new: page to replace with | |
793 | * @gfp_mask: allocation mode | |
794 | * | |
795 | * This function replaces a page in the pagecache with a new one. On | |
796 | * success it acquires the pagecache reference for the new page and | |
797 | * drops it for the old page. Both the old and new pages must be | |
798 | * locked. This function does not add the new page to the LRU, the | |
799 | * caller must do that. | |
800 | * | |
801 | * The remove + add is atomic. The only way this function can fail is | |
802 | * memory allocation failure. | |
803 | */ | |
804 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
805 | { | |
806 | int error; | |
ef6a3c63 | 807 | |
309381fe SL |
808 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
809 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
810 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 811 | |
5b8748b5 | 812 | error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK); |
ef6a3c63 MS |
813 | if (!error) { |
814 | struct address_space *mapping = old->mapping; | |
815 | void (*freepage)(struct page *); | |
c4843a75 | 816 | unsigned long flags; |
ef6a3c63 MS |
817 | |
818 | pgoff_t offset = old->index; | |
819 | freepage = mapping->a_ops->freepage; | |
820 | ||
09cbfeaf | 821 | get_page(new); |
ef6a3c63 MS |
822 | new->mapping = mapping; |
823 | new->index = offset; | |
824 | ||
c4843a75 | 825 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 826 | __delete_from_page_cache(old, NULL); |
22f2ac51 | 827 | error = page_cache_tree_insert(mapping, new, NULL); |
ef6a3c63 | 828 | BUG_ON(error); |
4165b9b4 MH |
829 | |
830 | /* | |
831 | * hugetlb pages do not participate in page cache accounting. | |
832 | */ | |
833 | if (!PageHuge(new)) | |
11fb9989 | 834 | __inc_node_page_state(new, NR_FILE_PAGES); |
ef6a3c63 | 835 | if (PageSwapBacked(new)) |
11fb9989 | 836 | __inc_node_page_state(new, NR_SHMEM); |
c4843a75 | 837 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6a93ca8f | 838 | mem_cgroup_migrate(old, new); |
ef6a3c63 MS |
839 | radix_tree_preload_end(); |
840 | if (freepage) | |
841 | freepage(old); | |
09cbfeaf | 842 | put_page(old); |
ef6a3c63 MS |
843 | } |
844 | ||
845 | return error; | |
846 | } | |
847 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
848 | ||
a528910e JW |
849 | static int __add_to_page_cache_locked(struct page *page, |
850 | struct address_space *mapping, | |
851 | pgoff_t offset, gfp_t gfp_mask, | |
852 | void **shadowp) | |
1da177e4 | 853 | { |
00501b53 JW |
854 | int huge = PageHuge(page); |
855 | struct mem_cgroup *memcg; | |
e286781d NP |
856 | int error; |
857 | ||
309381fe SL |
858 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
859 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 860 | |
00501b53 JW |
861 | if (!huge) { |
862 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 863 | gfp_mask, &memcg, false); |
00501b53 JW |
864 | if (error) |
865 | return error; | |
866 | } | |
1da177e4 | 867 | |
5b8748b5 | 868 | error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK); |
66a0c8ee | 869 | if (error) { |
00501b53 | 870 | if (!huge) |
f627c2f5 | 871 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
872 | return error; |
873 | } | |
874 | ||
09cbfeaf | 875 | get_page(page); |
66a0c8ee KS |
876 | page->mapping = mapping; |
877 | page->index = offset; | |
878 | ||
879 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 880 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
881 | radix_tree_preload_end(); |
882 | if (unlikely(error)) | |
883 | goto err_insert; | |
4165b9b4 MH |
884 | |
885 | /* hugetlb pages do not participate in page cache accounting. */ | |
886 | if (!huge) | |
11fb9989 | 887 | __inc_node_page_state(page, NR_FILE_PAGES); |
66a0c8ee | 888 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 889 | if (!huge) |
f627c2f5 | 890 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
891 | trace_mm_filemap_add_to_page_cache(page); |
892 | return 0; | |
893 | err_insert: | |
894 | page->mapping = NULL; | |
895 | /* Leave page->index set: truncation relies upon it */ | |
896 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 897 | if (!huge) |
f627c2f5 | 898 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 899 | put_page(page); |
1da177e4 LT |
900 | return error; |
901 | } | |
a528910e JW |
902 | |
903 | /** | |
904 | * add_to_page_cache_locked - add a locked page to the pagecache | |
905 | * @page: page to add | |
906 | * @mapping: the page's address_space | |
907 | * @offset: page index | |
908 | * @gfp_mask: page allocation mode | |
909 | * | |
910 | * This function is used to add a page to the pagecache. It must be locked. | |
911 | * This function does not add the page to the LRU. The caller must do that. | |
912 | */ | |
913 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
914 | pgoff_t offset, gfp_t gfp_mask) | |
915 | { | |
916 | return __add_to_page_cache_locked(page, mapping, offset, | |
917 | gfp_mask, NULL); | |
918 | } | |
e286781d | 919 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
920 | |
921 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 922 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 923 | { |
a528910e | 924 | void *shadow = NULL; |
4f98a2fe RR |
925 | int ret; |
926 | ||
48c935ad | 927 | __SetPageLocked(page); |
a528910e JW |
928 | ret = __add_to_page_cache_locked(page, mapping, offset, |
929 | gfp_mask, &shadow); | |
930 | if (unlikely(ret)) | |
48c935ad | 931 | __ClearPageLocked(page); |
a528910e JW |
932 | else { |
933 | /* | |
934 | * The page might have been evicted from cache only | |
935 | * recently, in which case it should be activated like | |
936 | * any other repeatedly accessed page. | |
f0281a00 RR |
937 | * The exception is pages getting rewritten; evicting other |
938 | * data from the working set, only to cache data that will | |
939 | * get overwritten with something else, is a waste of memory. | |
a528910e | 940 | */ |
f0281a00 RR |
941 | if (!(gfp_mask & __GFP_WRITE) && |
942 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
943 | SetPageActive(page); |
944 | workingset_activation(page); | |
945 | } else | |
946 | ClearPageActive(page); | |
947 | lru_cache_add(page); | |
948 | } | |
1da177e4 LT |
949 | return ret; |
950 | } | |
18bc0bbd | 951 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 952 | |
44110fe3 | 953 | #ifdef CONFIG_NUMA |
2ae88149 | 954 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 955 | { |
c0ff7453 MX |
956 | int n; |
957 | struct page *page; | |
958 | ||
44110fe3 | 959 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
960 | unsigned int cpuset_mems_cookie; |
961 | do { | |
d26914d1 | 962 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 963 | n = cpuset_mem_spread_node(); |
96db800f | 964 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 965 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 966 | |
c0ff7453 | 967 | return page; |
44110fe3 | 968 | } |
2ae88149 | 969 | return alloc_pages(gfp, 0); |
44110fe3 | 970 | } |
2ae88149 | 971 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
972 | #endif |
973 | ||
1da177e4 LT |
974 | /* |
975 | * In order to wait for pages to become available there must be | |
976 | * waitqueues associated with pages. By using a hash table of | |
977 | * waitqueues where the bucket discipline is to maintain all | |
978 | * waiters on the same queue and wake all when any of the pages | |
979 | * become available, and for the woken contexts to check to be | |
980 | * sure the appropriate page became available, this saves space | |
981 | * at a cost of "thundering herd" phenomena during rare hash | |
982 | * collisions. | |
983 | */ | |
62906027 NP |
984 | #define PAGE_WAIT_TABLE_BITS 8 |
985 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
986 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
987 | ||
988 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 989 | { |
62906027 | 990 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 991 | } |
1da177e4 | 992 | |
62906027 | 993 | void __init pagecache_init(void) |
1da177e4 | 994 | { |
62906027 | 995 | int i; |
1da177e4 | 996 | |
62906027 NP |
997 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
998 | init_waitqueue_head(&page_wait_table[i]); | |
999 | ||
1000 | page_writeback_init(); | |
1da177e4 | 1001 | } |
1da177e4 | 1002 | |
3510ca20 | 1003 | /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ |
62906027 NP |
1004 | struct wait_page_key { |
1005 | struct page *page; | |
1006 | int bit_nr; | |
1007 | int page_match; | |
1008 | }; | |
1009 | ||
1010 | struct wait_page_queue { | |
1011 | struct page *page; | |
1012 | int bit_nr; | |
ac6424b9 | 1013 | wait_queue_entry_t wait; |
62906027 NP |
1014 | }; |
1015 | ||
ac6424b9 | 1016 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1017 | { |
62906027 NP |
1018 | struct wait_page_key *key = arg; |
1019 | struct wait_page_queue *wait_page | |
1020 | = container_of(wait, struct wait_page_queue, wait); | |
1021 | ||
1022 | if (wait_page->page != key->page) | |
1023 | return 0; | |
1024 | key->page_match = 1; | |
f62e00cc | 1025 | |
62906027 NP |
1026 | if (wait_page->bit_nr != key->bit_nr) |
1027 | return 0; | |
3510ca20 LT |
1028 | |
1029 | /* Stop walking if it's locked */ | |
62906027 | 1030 | if (test_bit(key->bit_nr, &key->page->flags)) |
3510ca20 | 1031 | return -1; |
f62e00cc | 1032 | |
62906027 | 1033 | return autoremove_wake_function(wait, mode, sync, key); |
f62e00cc KM |
1034 | } |
1035 | ||
74d81bfa | 1036 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1037 | { |
62906027 NP |
1038 | wait_queue_head_t *q = page_waitqueue(page); |
1039 | struct wait_page_key key; | |
1040 | unsigned long flags; | |
11a19c7b | 1041 | wait_queue_entry_t bookmark; |
cbbce822 | 1042 | |
62906027 NP |
1043 | key.page = page; |
1044 | key.bit_nr = bit_nr; | |
1045 | key.page_match = 0; | |
1046 | ||
11a19c7b TC |
1047 | bookmark.flags = 0; |
1048 | bookmark.private = NULL; | |
1049 | bookmark.func = NULL; | |
1050 | INIT_LIST_HEAD(&bookmark.entry); | |
1051 | ||
62906027 | 1052 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1053 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1054 | ||
1055 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1056 | /* | |
1057 | * Take a breather from holding the lock, | |
1058 | * allow pages that finish wake up asynchronously | |
1059 | * to acquire the lock and remove themselves | |
1060 | * from wait queue | |
1061 | */ | |
1062 | spin_unlock_irqrestore(&q->lock, flags); | |
1063 | cpu_relax(); | |
1064 | spin_lock_irqsave(&q->lock, flags); | |
1065 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1066 | } | |
1067 | ||
62906027 NP |
1068 | /* |
1069 | * It is possible for other pages to have collided on the waitqueue | |
1070 | * hash, so in that case check for a page match. That prevents a long- | |
1071 | * term waiter | |
1072 | * | |
1073 | * It is still possible to miss a case here, when we woke page waiters | |
1074 | * and removed them from the waitqueue, but there are still other | |
1075 | * page waiters. | |
1076 | */ | |
1077 | if (!waitqueue_active(q) || !key.page_match) { | |
1078 | ClearPageWaiters(page); | |
1079 | /* | |
1080 | * It's possible to miss clearing Waiters here, when we woke | |
1081 | * our page waiters, but the hashed waitqueue has waiters for | |
1082 | * other pages on it. | |
1083 | * | |
1084 | * That's okay, it's a rare case. The next waker will clear it. | |
1085 | */ | |
1086 | } | |
1087 | spin_unlock_irqrestore(&q->lock, flags); | |
1088 | } | |
74d81bfa NP |
1089 | |
1090 | static void wake_up_page(struct page *page, int bit) | |
1091 | { | |
1092 | if (!PageWaiters(page)) | |
1093 | return; | |
1094 | wake_up_page_bit(page, bit); | |
1095 | } | |
62906027 NP |
1096 | |
1097 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, | |
1098 | struct page *page, int bit_nr, int state, bool lock) | |
1099 | { | |
1100 | struct wait_page_queue wait_page; | |
ac6424b9 | 1101 | wait_queue_entry_t *wait = &wait_page.wait; |
62906027 NP |
1102 | int ret = 0; |
1103 | ||
1104 | init_wait(wait); | |
3510ca20 | 1105 | wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0; |
62906027 NP |
1106 | wait->func = wake_page_function; |
1107 | wait_page.page = page; | |
1108 | wait_page.bit_nr = bit_nr; | |
1109 | ||
1110 | for (;;) { | |
1111 | spin_lock_irq(&q->lock); | |
1112 | ||
2055da97 | 1113 | if (likely(list_empty(&wait->entry))) { |
3510ca20 | 1114 | __add_wait_queue_entry_tail(q, wait); |
62906027 NP |
1115 | SetPageWaiters(page); |
1116 | } | |
1117 | ||
1118 | set_current_state(state); | |
1119 | ||
1120 | spin_unlock_irq(&q->lock); | |
1121 | ||
1122 | if (likely(test_bit(bit_nr, &page->flags))) { | |
1123 | io_schedule(); | |
62906027 NP |
1124 | } |
1125 | ||
1126 | if (lock) { | |
1127 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) | |
1128 | break; | |
1129 | } else { | |
1130 | if (!test_bit(bit_nr, &page->flags)) | |
1131 | break; | |
1132 | } | |
a8b169af LT |
1133 | |
1134 | if (unlikely(signal_pending_state(state, current))) { | |
1135 | ret = -EINTR; | |
1136 | break; | |
1137 | } | |
62906027 NP |
1138 | } |
1139 | ||
1140 | finish_wait(q, wait); | |
1141 | ||
1142 | /* | |
1143 | * A signal could leave PageWaiters set. Clearing it here if | |
1144 | * !waitqueue_active would be possible (by open-coding finish_wait), | |
1145 | * but still fail to catch it in the case of wait hash collision. We | |
1146 | * already can fail to clear wait hash collision cases, so don't | |
1147 | * bother with signals either. | |
1148 | */ | |
1149 | ||
1150 | return ret; | |
1151 | } | |
1152 | ||
1153 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1154 | { | |
1155 | wait_queue_head_t *q = page_waitqueue(page); | |
1156 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false); | |
1157 | } | |
1158 | EXPORT_SYMBOL(wait_on_page_bit); | |
1159 | ||
1160 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1161 | { | |
1162 | wait_queue_head_t *q = page_waitqueue(page); | |
1163 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false); | |
cbbce822 | 1164 | } |
4343d008 | 1165 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1166 | |
385e1ca5 DH |
1167 | /** |
1168 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1169 | * @page: Page defining the wait queue of interest |
1170 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1171 | * |
1172 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1173 | */ | |
ac6424b9 | 1174 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1175 | { |
1176 | wait_queue_head_t *q = page_waitqueue(page); | |
1177 | unsigned long flags; | |
1178 | ||
1179 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1180 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1181 | SetPageWaiters(page); |
385e1ca5 DH |
1182 | spin_unlock_irqrestore(&q->lock, flags); |
1183 | } | |
1184 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1185 | ||
b91e1302 LT |
1186 | #ifndef clear_bit_unlock_is_negative_byte |
1187 | ||
1188 | /* | |
1189 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1190 | * | |
1191 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1192 | * test the sign bit at the same time. But if the architecture does | |
1193 | * not support that special operation, we just do this all by hand | |
1194 | * instead. | |
1195 | * | |
1196 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
1197 | * being cleared, but a memory barrier should be unneccssary since it is | |
1198 | * in the same byte as PG_locked. | |
1199 | */ | |
1200 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1201 | { | |
1202 | clear_bit_unlock(nr, mem); | |
1203 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1204 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1205 | } |
1206 | ||
1207 | #endif | |
1208 | ||
1da177e4 | 1209 | /** |
485bb99b | 1210 | * unlock_page - unlock a locked page |
1da177e4 LT |
1211 | * @page: the page |
1212 | * | |
1213 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1214 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1215 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1216 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1217 | * | |
b91e1302 LT |
1218 | * Note that this depends on PG_waiters being the sign bit in the byte |
1219 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1220 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1221 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1222 | * test the sign bit). | |
1da177e4 | 1223 | */ |
920c7a5d | 1224 | void unlock_page(struct page *page) |
1da177e4 | 1225 | { |
b91e1302 | 1226 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1227 | page = compound_head(page); |
309381fe | 1228 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1229 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1230 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1231 | } |
1232 | EXPORT_SYMBOL(unlock_page); | |
1233 | ||
485bb99b RD |
1234 | /** |
1235 | * end_page_writeback - end writeback against a page | |
1236 | * @page: the page | |
1da177e4 LT |
1237 | */ |
1238 | void end_page_writeback(struct page *page) | |
1239 | { | |
888cf2db MG |
1240 | /* |
1241 | * TestClearPageReclaim could be used here but it is an atomic | |
1242 | * operation and overkill in this particular case. Failing to | |
1243 | * shuffle a page marked for immediate reclaim is too mild to | |
1244 | * justify taking an atomic operation penalty at the end of | |
1245 | * ever page writeback. | |
1246 | */ | |
1247 | if (PageReclaim(page)) { | |
1248 | ClearPageReclaim(page); | |
ac6aadb2 | 1249 | rotate_reclaimable_page(page); |
888cf2db | 1250 | } |
ac6aadb2 MS |
1251 | |
1252 | if (!test_clear_page_writeback(page)) | |
1253 | BUG(); | |
1254 | ||
4e857c58 | 1255 | smp_mb__after_atomic(); |
1da177e4 LT |
1256 | wake_up_page(page, PG_writeback); |
1257 | } | |
1258 | EXPORT_SYMBOL(end_page_writeback); | |
1259 | ||
57d99845 MW |
1260 | /* |
1261 | * After completing I/O on a page, call this routine to update the page | |
1262 | * flags appropriately | |
1263 | */ | |
c11f0c0b | 1264 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1265 | { |
c11f0c0b | 1266 | if (!is_write) { |
57d99845 MW |
1267 | if (!err) { |
1268 | SetPageUptodate(page); | |
1269 | } else { | |
1270 | ClearPageUptodate(page); | |
1271 | SetPageError(page); | |
1272 | } | |
1273 | unlock_page(page); | |
abf54548 | 1274 | } else { |
57d99845 | 1275 | if (err) { |
dd8416c4 MK |
1276 | struct address_space *mapping; |
1277 | ||
57d99845 | 1278 | SetPageError(page); |
dd8416c4 MK |
1279 | mapping = page_mapping(page); |
1280 | if (mapping) | |
1281 | mapping_set_error(mapping, err); | |
57d99845 MW |
1282 | } |
1283 | end_page_writeback(page); | |
1284 | } | |
1285 | } | |
1286 | EXPORT_SYMBOL_GPL(page_endio); | |
1287 | ||
485bb99b RD |
1288 | /** |
1289 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1290 | * @__page: the page to lock |
1da177e4 | 1291 | */ |
62906027 | 1292 | void __lock_page(struct page *__page) |
1da177e4 | 1293 | { |
62906027 NP |
1294 | struct page *page = compound_head(__page); |
1295 | wait_queue_head_t *q = page_waitqueue(page); | |
1296 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true); | |
1da177e4 LT |
1297 | } |
1298 | EXPORT_SYMBOL(__lock_page); | |
1299 | ||
62906027 | 1300 | int __lock_page_killable(struct page *__page) |
2687a356 | 1301 | { |
62906027 NP |
1302 | struct page *page = compound_head(__page); |
1303 | wait_queue_head_t *q = page_waitqueue(page); | |
1304 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true); | |
2687a356 | 1305 | } |
18bc0bbd | 1306 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1307 | |
9a95f3cf PC |
1308 | /* |
1309 | * Return values: | |
1310 | * 1 - page is locked; mmap_sem is still held. | |
1311 | * 0 - page is not locked. | |
1312 | * mmap_sem has been released (up_read()), unless flags had both | |
1313 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
1314 | * which case mmap_sem is still held. | |
1315 | * | |
1316 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
1317 | * with the page locked and the mmap_sem unperturbed. | |
1318 | */ | |
d065bd81 ML |
1319 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1320 | unsigned int flags) | |
1321 | { | |
37b23e05 KM |
1322 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
1323 | /* | |
1324 | * CAUTION! In this case, mmap_sem is not released | |
1325 | * even though return 0. | |
1326 | */ | |
1327 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1328 | return 0; | |
1329 | ||
1330 | up_read(&mm->mmap_sem); | |
1331 | if (flags & FAULT_FLAG_KILLABLE) | |
1332 | wait_on_page_locked_killable(page); | |
1333 | else | |
318b275f | 1334 | wait_on_page_locked(page); |
d065bd81 | 1335 | return 0; |
37b23e05 KM |
1336 | } else { |
1337 | if (flags & FAULT_FLAG_KILLABLE) { | |
1338 | int ret; | |
1339 | ||
1340 | ret = __lock_page_killable(page); | |
1341 | if (ret) { | |
1342 | up_read(&mm->mmap_sem); | |
1343 | return 0; | |
1344 | } | |
1345 | } else | |
1346 | __lock_page(page); | |
1347 | return 1; | |
d065bd81 ML |
1348 | } |
1349 | } | |
1350 | ||
e7b563bb JW |
1351 | /** |
1352 | * page_cache_next_hole - find the next hole (not-present entry) | |
1353 | * @mapping: mapping | |
1354 | * @index: index | |
1355 | * @max_scan: maximum range to search | |
1356 | * | |
1357 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
1358 | * lowest indexed hole. | |
1359 | * | |
1360 | * Returns: the index of the hole if found, otherwise returns an index | |
1361 | * outside of the set specified (in which case 'return - index >= | |
1362 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
1363 | * be returned. | |
1364 | * | |
1365 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
1366 | * like radix_tree_gang_lookup, this will not atomically search a | |
1367 | * snapshot of the tree at a single point in time. For example, if a | |
1368 | * hole is created at index 5, then subsequently a hole is created at | |
1369 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
1370 | * if called under rcu_read_lock. | |
1371 | */ | |
1372 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
1373 | pgoff_t index, unsigned long max_scan) | |
1374 | { | |
1375 | unsigned long i; | |
1376 | ||
1377 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1378 | struct page *page; |
1379 | ||
1380 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1381 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1382 | break; |
1383 | index++; | |
1384 | if (index == 0) | |
1385 | break; | |
1386 | } | |
1387 | ||
1388 | return index; | |
1389 | } | |
1390 | EXPORT_SYMBOL(page_cache_next_hole); | |
1391 | ||
1392 | /** | |
1393 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
1394 | * @mapping: mapping | |
1395 | * @index: index | |
1396 | * @max_scan: maximum range to search | |
1397 | * | |
1398 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
1399 | * the first hole. | |
1400 | * | |
1401 | * Returns: the index of the hole if found, otherwise returns an index | |
1402 | * outside of the set specified (in which case 'index - return >= | |
1403 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1404 | * will be returned. | |
1405 | * | |
1406 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1407 | * like radix_tree_gang_lookup, this will not atomically search a | |
1408 | * snapshot of the tree at a single point in time. For example, if a | |
1409 | * hole is created at index 10, then subsequently a hole is created at | |
1410 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1411 | * called under rcu_read_lock. | |
1412 | */ | |
1413 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1414 | pgoff_t index, unsigned long max_scan) | |
1415 | { | |
1416 | unsigned long i; | |
1417 | ||
1418 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1419 | struct page *page; |
1420 | ||
1421 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1422 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1423 | break; |
1424 | index--; | |
1425 | if (index == ULONG_MAX) | |
1426 | break; | |
1427 | } | |
1428 | ||
1429 | return index; | |
1430 | } | |
1431 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1432 | ||
485bb99b | 1433 | /** |
0cd6144a | 1434 | * find_get_entry - find and get a page cache entry |
485bb99b | 1435 | * @mapping: the address_space to search |
0cd6144a JW |
1436 | * @offset: the page cache index |
1437 | * | |
1438 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1439 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1440 | * |
139b6a6f JW |
1441 | * If the slot holds a shadow entry of a previously evicted page, or a |
1442 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1443 | * |
1444 | * Otherwise, %NULL is returned. | |
1da177e4 | 1445 | */ |
0cd6144a | 1446 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1447 | { |
a60637c8 | 1448 | void **pagep; |
83929372 | 1449 | struct page *head, *page; |
1da177e4 | 1450 | |
a60637c8 NP |
1451 | rcu_read_lock(); |
1452 | repeat: | |
1453 | page = NULL; | |
1454 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1455 | if (pagep) { | |
1456 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1457 | if (unlikely(!page)) |
1458 | goto out; | |
a2c16d6c | 1459 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1460 | if (radix_tree_deref_retry(page)) |
1461 | goto repeat; | |
1462 | /* | |
139b6a6f JW |
1463 | * A shadow entry of a recently evicted page, |
1464 | * or a swap entry from shmem/tmpfs. Return | |
1465 | * it without attempting to raise page count. | |
8079b1c8 HD |
1466 | */ |
1467 | goto out; | |
a2c16d6c | 1468 | } |
83929372 KS |
1469 | |
1470 | head = compound_head(page); | |
1471 | if (!page_cache_get_speculative(head)) | |
1472 | goto repeat; | |
1473 | ||
1474 | /* The page was split under us? */ | |
1475 | if (compound_head(page) != head) { | |
1476 | put_page(head); | |
a60637c8 | 1477 | goto repeat; |
83929372 | 1478 | } |
a60637c8 NP |
1479 | |
1480 | /* | |
1481 | * Has the page moved? | |
1482 | * This is part of the lockless pagecache protocol. See | |
1483 | * include/linux/pagemap.h for details. | |
1484 | */ | |
1485 | if (unlikely(page != *pagep)) { | |
83929372 | 1486 | put_page(head); |
a60637c8 NP |
1487 | goto repeat; |
1488 | } | |
1489 | } | |
27d20fdd | 1490 | out: |
a60637c8 NP |
1491 | rcu_read_unlock(); |
1492 | ||
1da177e4 LT |
1493 | return page; |
1494 | } | |
0cd6144a | 1495 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1496 | |
0cd6144a JW |
1497 | /** |
1498 | * find_lock_entry - locate, pin and lock a page cache entry | |
1499 | * @mapping: the address_space to search | |
1500 | * @offset: the page cache index | |
1501 | * | |
1502 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1503 | * page cache page, it is returned locked and with an increased | |
1504 | * refcount. | |
1505 | * | |
139b6a6f JW |
1506 | * If the slot holds a shadow entry of a previously evicted page, or a |
1507 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1508 | * |
1509 | * Otherwise, %NULL is returned. | |
1510 | * | |
1511 | * find_lock_entry() may sleep. | |
1512 | */ | |
1513 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1514 | { |
1515 | struct page *page; | |
1516 | ||
1da177e4 | 1517 | repeat: |
0cd6144a | 1518 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1519 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1520 | lock_page(page); |
1521 | /* Has the page been truncated? */ | |
83929372 | 1522 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1523 | unlock_page(page); |
09cbfeaf | 1524 | put_page(page); |
a60637c8 | 1525 | goto repeat; |
1da177e4 | 1526 | } |
83929372 | 1527 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1528 | } |
1da177e4 LT |
1529 | return page; |
1530 | } | |
0cd6144a JW |
1531 | EXPORT_SYMBOL(find_lock_entry); |
1532 | ||
1533 | /** | |
2457aec6 | 1534 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1535 | * @mapping: the address_space to search |
1536 | * @offset: the page index | |
2457aec6 | 1537 | * @fgp_flags: PCG flags |
45f87de5 | 1538 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1539 | * |
2457aec6 | 1540 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1541 | * |
75325189 | 1542 | * PCG flags modify how the page is returned. |
0cd6144a | 1543 | * |
0e056eb5 MCC |
1544 | * @fgp_flags can be: |
1545 | * | |
1546 | * - FGP_ACCESSED: the page will be marked accessed | |
1547 | * - FGP_LOCK: Page is return locked | |
1548 | * - FGP_CREAT: If page is not present then a new page is allocated using | |
1549 | * @gfp_mask and added to the page cache and the VM's LRU | |
1550 | * list. The page is returned locked and with an increased | |
1551 | * refcount. Otherwise, NULL is returned. | |
1da177e4 | 1552 | * |
2457aec6 MG |
1553 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1554 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1555 | * |
2457aec6 | 1556 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1557 | */ |
2457aec6 | 1558 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1559 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1560 | { |
eb2be189 | 1561 | struct page *page; |
2457aec6 | 1562 | |
1da177e4 | 1563 | repeat: |
2457aec6 MG |
1564 | page = find_get_entry(mapping, offset); |
1565 | if (radix_tree_exceptional_entry(page)) | |
1566 | page = NULL; | |
1567 | if (!page) | |
1568 | goto no_page; | |
1569 | ||
1570 | if (fgp_flags & FGP_LOCK) { | |
1571 | if (fgp_flags & FGP_NOWAIT) { | |
1572 | if (!trylock_page(page)) { | |
09cbfeaf | 1573 | put_page(page); |
2457aec6 MG |
1574 | return NULL; |
1575 | } | |
1576 | } else { | |
1577 | lock_page(page); | |
1578 | } | |
1579 | ||
1580 | /* Has the page been truncated? */ | |
1581 | if (unlikely(page->mapping != mapping)) { | |
1582 | unlock_page(page); | |
09cbfeaf | 1583 | put_page(page); |
2457aec6 MG |
1584 | goto repeat; |
1585 | } | |
1586 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1587 | } | |
1588 | ||
1589 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1590 | mark_page_accessed(page); | |
1591 | ||
1592 | no_page: | |
1593 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1594 | int err; | |
1595 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1596 | gfp_mask |= __GFP_WRITE; |
1597 | if (fgp_flags & FGP_NOFS) | |
1598 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1599 | |
45f87de5 | 1600 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1601 | if (!page) |
1602 | return NULL; | |
2457aec6 MG |
1603 | |
1604 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1605 | fgp_flags |= FGP_LOCK; | |
1606 | ||
eb39d618 | 1607 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1608 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1609 | __SetPageReferenced(page); |
2457aec6 | 1610 | |
5b8748b5 | 1611 | err = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
eb2be189 | 1612 | if (unlikely(err)) { |
09cbfeaf | 1613 | put_page(page); |
eb2be189 NP |
1614 | page = NULL; |
1615 | if (err == -EEXIST) | |
1616 | goto repeat; | |
1da177e4 | 1617 | } |
1da177e4 | 1618 | } |
2457aec6 | 1619 | |
1da177e4 LT |
1620 | return page; |
1621 | } | |
2457aec6 | 1622 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1623 | |
0cd6144a JW |
1624 | /** |
1625 | * find_get_entries - gang pagecache lookup | |
1626 | * @mapping: The address_space to search | |
1627 | * @start: The starting page cache index | |
1628 | * @nr_entries: The maximum number of entries | |
1629 | * @entries: Where the resulting entries are placed | |
1630 | * @indices: The cache indices corresponding to the entries in @entries | |
1631 | * | |
1632 | * find_get_entries() will search for and return a group of up to | |
1633 | * @nr_entries entries in the mapping. The entries are placed at | |
1634 | * @entries. find_get_entries() takes a reference against any actual | |
1635 | * pages it returns. | |
1636 | * | |
1637 | * The search returns a group of mapping-contiguous page cache entries | |
1638 | * with ascending indexes. There may be holes in the indices due to | |
1639 | * not-present pages. | |
1640 | * | |
139b6a6f JW |
1641 | * Any shadow entries of evicted pages, or swap entries from |
1642 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1643 | * |
1644 | * find_get_entries() returns the number of pages and shadow entries | |
1645 | * which were found. | |
1646 | */ | |
1647 | unsigned find_get_entries(struct address_space *mapping, | |
1648 | pgoff_t start, unsigned int nr_entries, | |
1649 | struct page **entries, pgoff_t *indices) | |
1650 | { | |
1651 | void **slot; | |
1652 | unsigned int ret = 0; | |
1653 | struct radix_tree_iter iter; | |
1654 | ||
1655 | if (!nr_entries) | |
1656 | return 0; | |
1657 | ||
1658 | rcu_read_lock(); | |
0cd6144a | 1659 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
83929372 | 1660 | struct page *head, *page; |
0cd6144a JW |
1661 | repeat: |
1662 | page = radix_tree_deref_slot(slot); | |
1663 | if (unlikely(!page)) | |
1664 | continue; | |
1665 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1666 | if (radix_tree_deref_retry(page)) { |
1667 | slot = radix_tree_iter_retry(&iter); | |
1668 | continue; | |
1669 | } | |
0cd6144a | 1670 | /* |
f9fe48be RZ |
1671 | * A shadow entry of a recently evicted page, a swap |
1672 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1673 | * without attempting to raise page count. | |
0cd6144a JW |
1674 | */ |
1675 | goto export; | |
1676 | } | |
83929372 KS |
1677 | |
1678 | head = compound_head(page); | |
1679 | if (!page_cache_get_speculative(head)) | |
1680 | goto repeat; | |
1681 | ||
1682 | /* The page was split under us? */ | |
1683 | if (compound_head(page) != head) { | |
1684 | put_page(head); | |
0cd6144a | 1685 | goto repeat; |
83929372 | 1686 | } |
0cd6144a JW |
1687 | |
1688 | /* Has the page moved? */ | |
1689 | if (unlikely(page != *slot)) { | |
83929372 | 1690 | put_page(head); |
0cd6144a JW |
1691 | goto repeat; |
1692 | } | |
1693 | export: | |
1694 | indices[ret] = iter.index; | |
1695 | entries[ret] = page; | |
1696 | if (++ret == nr_entries) | |
1697 | break; | |
1698 | } | |
1699 | rcu_read_unlock(); | |
1700 | return ret; | |
1701 | } | |
1702 | ||
1da177e4 | 1703 | /** |
b947cee4 | 1704 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1705 | * @mapping: The address_space to search |
1706 | * @start: The starting page index | |
b947cee4 | 1707 | * @end: The final page index (inclusive) |
1da177e4 LT |
1708 | * @nr_pages: The maximum number of pages |
1709 | * @pages: Where the resulting pages are placed | |
1710 | * | |
b947cee4 JK |
1711 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1712 | * pages in the mapping starting at index @start and up to index @end | |
1713 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1714 | * a reference against the returned pages. | |
1da177e4 LT |
1715 | * |
1716 | * The search returns a group of mapping-contiguous pages with ascending | |
1717 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1718 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1719 | * |
b947cee4 JK |
1720 | * find_get_pages_range() returns the number of pages which were found. If this |
1721 | * number is smaller than @nr_pages, the end of specified range has been | |
1722 | * reached. | |
1da177e4 | 1723 | */ |
b947cee4 JK |
1724 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1725 | pgoff_t end, unsigned int nr_pages, | |
1726 | struct page **pages) | |
1da177e4 | 1727 | { |
0fc9d104 KK |
1728 | struct radix_tree_iter iter; |
1729 | void **slot; | |
1730 | unsigned ret = 0; | |
1731 | ||
1732 | if (unlikely(!nr_pages)) | |
1733 | return 0; | |
a60637c8 NP |
1734 | |
1735 | rcu_read_lock(); | |
d72dc8a2 | 1736 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, *start) { |
83929372 | 1737 | struct page *head, *page; |
b947cee4 JK |
1738 | |
1739 | if (iter.index > end) | |
1740 | break; | |
a60637c8 | 1741 | repeat: |
0fc9d104 | 1742 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1743 | if (unlikely(!page)) |
1744 | continue; | |
9d8aa4ea | 1745 | |
a2c16d6c | 1746 | if (radix_tree_exception(page)) { |
8079b1c8 | 1747 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1748 | slot = radix_tree_iter_retry(&iter); |
1749 | continue; | |
8079b1c8 | 1750 | } |
a2c16d6c | 1751 | /* |
139b6a6f JW |
1752 | * A shadow entry of a recently evicted page, |
1753 | * or a swap entry from shmem/tmpfs. Skip | |
1754 | * over it. | |
a2c16d6c | 1755 | */ |
8079b1c8 | 1756 | continue; |
27d20fdd | 1757 | } |
a60637c8 | 1758 | |
83929372 KS |
1759 | head = compound_head(page); |
1760 | if (!page_cache_get_speculative(head)) | |
1761 | goto repeat; | |
1762 | ||
1763 | /* The page was split under us? */ | |
1764 | if (compound_head(page) != head) { | |
1765 | put_page(head); | |
a60637c8 | 1766 | goto repeat; |
83929372 | 1767 | } |
a60637c8 NP |
1768 | |
1769 | /* Has the page moved? */ | |
0fc9d104 | 1770 | if (unlikely(page != *slot)) { |
83929372 | 1771 | put_page(head); |
a60637c8 NP |
1772 | goto repeat; |
1773 | } | |
1da177e4 | 1774 | |
a60637c8 | 1775 | pages[ret] = page; |
b947cee4 JK |
1776 | if (++ret == nr_pages) { |
1777 | *start = pages[ret - 1]->index + 1; | |
1778 | goto out; | |
1779 | } | |
a60637c8 | 1780 | } |
5b280c0c | 1781 | |
b947cee4 JK |
1782 | /* |
1783 | * We come here when there is no page beyond @end. We take care to not | |
1784 | * overflow the index @start as it confuses some of the callers. This | |
1785 | * breaks the iteration when there is page at index -1 but that is | |
1786 | * already broken anyway. | |
1787 | */ | |
1788 | if (end == (pgoff_t)-1) | |
1789 | *start = (pgoff_t)-1; | |
1790 | else | |
1791 | *start = end + 1; | |
1792 | out: | |
a60637c8 | 1793 | rcu_read_unlock(); |
d72dc8a2 | 1794 | |
1da177e4 LT |
1795 | return ret; |
1796 | } | |
1797 | ||
ebf43500 JA |
1798 | /** |
1799 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1800 | * @mapping: The address_space to search | |
1801 | * @index: The starting page index | |
1802 | * @nr_pages: The maximum number of pages | |
1803 | * @pages: Where the resulting pages are placed | |
1804 | * | |
1805 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1806 | * that the returned number of pages are guaranteed to be contiguous. | |
1807 | * | |
1808 | * find_get_pages_contig() returns the number of pages which were found. | |
1809 | */ | |
1810 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1811 | unsigned int nr_pages, struct page **pages) | |
1812 | { | |
0fc9d104 KK |
1813 | struct radix_tree_iter iter; |
1814 | void **slot; | |
1815 | unsigned int ret = 0; | |
1816 | ||
1817 | if (unlikely(!nr_pages)) | |
1818 | return 0; | |
a60637c8 NP |
1819 | |
1820 | rcu_read_lock(); | |
0fc9d104 | 1821 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
83929372 | 1822 | struct page *head, *page; |
a60637c8 | 1823 | repeat: |
0fc9d104 KK |
1824 | page = radix_tree_deref_slot(slot); |
1825 | /* The hole, there no reason to continue */ | |
a60637c8 | 1826 | if (unlikely(!page)) |
0fc9d104 | 1827 | break; |
9d8aa4ea | 1828 | |
a2c16d6c | 1829 | if (radix_tree_exception(page)) { |
8079b1c8 | 1830 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1831 | slot = radix_tree_iter_retry(&iter); |
1832 | continue; | |
8079b1c8 | 1833 | } |
a2c16d6c | 1834 | /* |
139b6a6f JW |
1835 | * A shadow entry of a recently evicted page, |
1836 | * or a swap entry from shmem/tmpfs. Stop | |
1837 | * looking for contiguous pages. | |
a2c16d6c | 1838 | */ |
8079b1c8 | 1839 | break; |
a2c16d6c | 1840 | } |
ebf43500 | 1841 | |
83929372 KS |
1842 | head = compound_head(page); |
1843 | if (!page_cache_get_speculative(head)) | |
1844 | goto repeat; | |
1845 | ||
1846 | /* The page was split under us? */ | |
1847 | if (compound_head(page) != head) { | |
1848 | put_page(head); | |
a60637c8 | 1849 | goto repeat; |
83929372 | 1850 | } |
a60637c8 NP |
1851 | |
1852 | /* Has the page moved? */ | |
0fc9d104 | 1853 | if (unlikely(page != *slot)) { |
83929372 | 1854 | put_page(head); |
a60637c8 NP |
1855 | goto repeat; |
1856 | } | |
1857 | ||
9cbb4cb2 NP |
1858 | /* |
1859 | * must check mapping and index after taking the ref. | |
1860 | * otherwise we can get both false positives and false | |
1861 | * negatives, which is just confusing to the caller. | |
1862 | */ | |
83929372 | 1863 | if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { |
09cbfeaf | 1864 | put_page(page); |
9cbb4cb2 NP |
1865 | break; |
1866 | } | |
1867 | ||
a60637c8 | 1868 | pages[ret] = page; |
0fc9d104 KK |
1869 | if (++ret == nr_pages) |
1870 | break; | |
ebf43500 | 1871 | } |
a60637c8 NP |
1872 | rcu_read_unlock(); |
1873 | return ret; | |
ebf43500 | 1874 | } |
ef71c15c | 1875 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1876 | |
485bb99b | 1877 | /** |
72b045ae | 1878 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
1879 | * @mapping: the address_space to search |
1880 | * @index: the starting page index | |
72b045ae | 1881 | * @end: The final page index (inclusive) |
485bb99b RD |
1882 | * @tag: the tag index |
1883 | * @nr_pages: the maximum number of pages | |
1884 | * @pages: where the resulting pages are placed | |
1885 | * | |
1da177e4 | 1886 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1887 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 | 1888 | */ |
72b045ae JK |
1889 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
1890 | pgoff_t end, int tag, unsigned int nr_pages, | |
1891 | struct page **pages) | |
1da177e4 | 1892 | { |
0fc9d104 KK |
1893 | struct radix_tree_iter iter; |
1894 | void **slot; | |
1895 | unsigned ret = 0; | |
1896 | ||
1897 | if (unlikely(!nr_pages)) | |
1898 | return 0; | |
a60637c8 NP |
1899 | |
1900 | rcu_read_lock(); | |
0fc9d104 KK |
1901 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1902 | &iter, *index, tag) { | |
83929372 | 1903 | struct page *head, *page; |
72b045ae JK |
1904 | |
1905 | if (iter.index > end) | |
1906 | break; | |
a60637c8 | 1907 | repeat: |
0fc9d104 | 1908 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1909 | if (unlikely(!page)) |
1910 | continue; | |
9d8aa4ea | 1911 | |
a2c16d6c | 1912 | if (radix_tree_exception(page)) { |
8079b1c8 | 1913 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1914 | slot = radix_tree_iter_retry(&iter); |
1915 | continue; | |
8079b1c8 | 1916 | } |
a2c16d6c | 1917 | /* |
139b6a6f JW |
1918 | * A shadow entry of a recently evicted page. |
1919 | * | |
1920 | * Those entries should never be tagged, but | |
1921 | * this tree walk is lockless and the tags are | |
1922 | * looked up in bulk, one radix tree node at a | |
1923 | * time, so there is a sizable window for page | |
1924 | * reclaim to evict a page we saw tagged. | |
1925 | * | |
1926 | * Skip over it. | |
a2c16d6c | 1927 | */ |
139b6a6f | 1928 | continue; |
a2c16d6c | 1929 | } |
a60637c8 | 1930 | |
83929372 KS |
1931 | head = compound_head(page); |
1932 | if (!page_cache_get_speculative(head)) | |
a60637c8 NP |
1933 | goto repeat; |
1934 | ||
83929372 KS |
1935 | /* The page was split under us? */ |
1936 | if (compound_head(page) != head) { | |
1937 | put_page(head); | |
1938 | goto repeat; | |
1939 | } | |
1940 | ||
a60637c8 | 1941 | /* Has the page moved? */ |
0fc9d104 | 1942 | if (unlikely(page != *slot)) { |
83929372 | 1943 | put_page(head); |
a60637c8 NP |
1944 | goto repeat; |
1945 | } | |
1946 | ||
1947 | pages[ret] = page; | |
72b045ae JK |
1948 | if (++ret == nr_pages) { |
1949 | *index = pages[ret - 1]->index + 1; | |
1950 | goto out; | |
1951 | } | |
a60637c8 | 1952 | } |
5b280c0c | 1953 | |
72b045ae JK |
1954 | /* |
1955 | * We come here when we got at @end. We take care to not overflow the | |
1956 | * index @index as it confuses some of the callers. This breaks the | |
1957 | * iteration when there is page at index -1 but that is already broken | |
1958 | * anyway. | |
1959 | */ | |
1960 | if (end == (pgoff_t)-1) | |
1961 | *index = (pgoff_t)-1; | |
1962 | else | |
1963 | *index = end + 1; | |
1964 | out: | |
a60637c8 | 1965 | rcu_read_unlock(); |
1da177e4 | 1966 | |
1da177e4 LT |
1967 | return ret; |
1968 | } | |
72b045ae | 1969 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 1970 | |
7e7f7749 RZ |
1971 | /** |
1972 | * find_get_entries_tag - find and return entries that match @tag | |
1973 | * @mapping: the address_space to search | |
1974 | * @start: the starting page cache index | |
1975 | * @tag: the tag index | |
1976 | * @nr_entries: the maximum number of entries | |
1977 | * @entries: where the resulting entries are placed | |
1978 | * @indices: the cache indices corresponding to the entries in @entries | |
1979 | * | |
1980 | * Like find_get_entries, except we only return entries which are tagged with | |
1981 | * @tag. | |
1982 | */ | |
1983 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1984 | int tag, unsigned int nr_entries, | |
1985 | struct page **entries, pgoff_t *indices) | |
1986 | { | |
1987 | void **slot; | |
1988 | unsigned int ret = 0; | |
1989 | struct radix_tree_iter iter; | |
1990 | ||
1991 | if (!nr_entries) | |
1992 | return 0; | |
1993 | ||
1994 | rcu_read_lock(); | |
7e7f7749 RZ |
1995 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1996 | &iter, start, tag) { | |
83929372 | 1997 | struct page *head, *page; |
7e7f7749 RZ |
1998 | repeat: |
1999 | page = radix_tree_deref_slot(slot); | |
2000 | if (unlikely(!page)) | |
2001 | continue; | |
2002 | if (radix_tree_exception(page)) { | |
2003 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
2004 | slot = radix_tree_iter_retry(&iter); |
2005 | continue; | |
7e7f7749 RZ |
2006 | } |
2007 | ||
2008 | /* | |
2009 | * A shadow entry of a recently evicted page, a swap | |
2010 | * entry from shmem/tmpfs or a DAX entry. Return it | |
2011 | * without attempting to raise page count. | |
2012 | */ | |
2013 | goto export; | |
2014 | } | |
83929372 KS |
2015 | |
2016 | head = compound_head(page); | |
2017 | if (!page_cache_get_speculative(head)) | |
7e7f7749 RZ |
2018 | goto repeat; |
2019 | ||
83929372 KS |
2020 | /* The page was split under us? */ |
2021 | if (compound_head(page) != head) { | |
2022 | put_page(head); | |
2023 | goto repeat; | |
2024 | } | |
2025 | ||
7e7f7749 RZ |
2026 | /* Has the page moved? */ |
2027 | if (unlikely(page != *slot)) { | |
83929372 | 2028 | put_page(head); |
7e7f7749 RZ |
2029 | goto repeat; |
2030 | } | |
2031 | export: | |
2032 | indices[ret] = iter.index; | |
2033 | entries[ret] = page; | |
2034 | if (++ret == nr_entries) | |
2035 | break; | |
2036 | } | |
2037 | rcu_read_unlock(); | |
2038 | return ret; | |
2039 | } | |
2040 | EXPORT_SYMBOL(find_get_entries_tag); | |
2041 | ||
76d42bd9 WF |
2042 | /* |
2043 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2044 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2045 | * | |
2046 | * ---R__________________________________________B__________ | |
2047 | * ^ reading here ^ bad block(assume 4k) | |
2048 | * | |
2049 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2050 | * => failing the whole request => read(R) => read(R+1) => | |
2051 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2052 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2053 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2054 | * | |
2055 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2056 | */ | |
2057 | static void shrink_readahead_size_eio(struct file *filp, | |
2058 | struct file_ra_state *ra) | |
2059 | { | |
76d42bd9 | 2060 | ra->ra_pages /= 4; |
76d42bd9 WF |
2061 | } |
2062 | ||
485bb99b | 2063 | /** |
47c27bc4 CH |
2064 | * generic_file_buffered_read - generic file read routine |
2065 | * @iocb: the iocb to read | |
6e58e79d AV |
2066 | * @iter: data destination |
2067 | * @written: already copied | |
485bb99b | 2068 | * |
1da177e4 | 2069 | * This is a generic file read routine, and uses the |
485bb99b | 2070 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
2071 | * |
2072 | * This is really ugly. But the goto's actually try to clarify some | |
2073 | * of the logic when it comes to error handling etc. | |
1da177e4 | 2074 | */ |
47c27bc4 | 2075 | static ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 2076 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 2077 | { |
47c27bc4 | 2078 | struct file *filp = iocb->ki_filp; |
36e78914 | 2079 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2080 | struct inode *inode = mapping->host; |
36e78914 | 2081 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 2082 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
2083 | pgoff_t index; |
2084 | pgoff_t last_index; | |
2085 | pgoff_t prev_index; | |
2086 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 2087 | unsigned int prev_offset; |
6e58e79d | 2088 | int error = 0; |
1da177e4 | 2089 | |
c2a9737f | 2090 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2091 | return 0; |
c2a9737f WF |
2092 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
2093 | ||
09cbfeaf KS |
2094 | index = *ppos >> PAGE_SHIFT; |
2095 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
2096 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
2097 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
2098 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 2099 | |
1da177e4 LT |
2100 | for (;;) { |
2101 | struct page *page; | |
57f6b96c | 2102 | pgoff_t end_index; |
a32ea1e1 | 2103 | loff_t isize; |
1da177e4 LT |
2104 | unsigned long nr, ret; |
2105 | ||
1da177e4 | 2106 | cond_resched(); |
1da177e4 | 2107 | find_page: |
5abf186a MH |
2108 | if (fatal_signal_pending(current)) { |
2109 | error = -EINTR; | |
2110 | goto out; | |
2111 | } | |
2112 | ||
1da177e4 | 2113 | page = find_get_page(mapping, index); |
3ea89ee8 | 2114 | if (!page) { |
3239d834 MT |
2115 | if (iocb->ki_flags & IOCB_NOWAIT) |
2116 | goto would_block; | |
cf914a7d | 2117 | page_cache_sync_readahead(mapping, |
7ff81078 | 2118 | ra, filp, |
3ea89ee8 FW |
2119 | index, last_index - index); |
2120 | page = find_get_page(mapping, index); | |
2121 | if (unlikely(page == NULL)) | |
2122 | goto no_cached_page; | |
2123 | } | |
2124 | if (PageReadahead(page)) { | |
cf914a7d | 2125 | page_cache_async_readahead(mapping, |
7ff81078 | 2126 | ra, filp, page, |
3ea89ee8 | 2127 | index, last_index - index); |
1da177e4 | 2128 | } |
8ab22b9a | 2129 | if (!PageUptodate(page)) { |
3239d834 MT |
2130 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2131 | put_page(page); | |
2132 | goto would_block; | |
2133 | } | |
2134 | ||
ebded027 MG |
2135 | /* |
2136 | * See comment in do_read_cache_page on why | |
2137 | * wait_on_page_locked is used to avoid unnecessarily | |
2138 | * serialisations and why it's safe. | |
2139 | */ | |
c4b209a4 BVA |
2140 | error = wait_on_page_locked_killable(page); |
2141 | if (unlikely(error)) | |
2142 | goto readpage_error; | |
ebded027 MG |
2143 | if (PageUptodate(page)) |
2144 | goto page_ok; | |
2145 | ||
09cbfeaf | 2146 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2147 | !mapping->a_ops->is_partially_uptodate) |
2148 | goto page_not_up_to_date; | |
6d6d36bc EG |
2149 | /* pipes can't handle partially uptodate pages */ |
2150 | if (unlikely(iter->type & ITER_PIPE)) | |
2151 | goto page_not_up_to_date; | |
529ae9aa | 2152 | if (!trylock_page(page)) |
8ab22b9a | 2153 | goto page_not_up_to_date; |
8d056cb9 DH |
2154 | /* Did it get truncated before we got the lock? */ |
2155 | if (!page->mapping) | |
2156 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2157 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2158 | offset, iter->count)) |
8ab22b9a HH |
2159 | goto page_not_up_to_date_locked; |
2160 | unlock_page(page); | |
2161 | } | |
1da177e4 | 2162 | page_ok: |
a32ea1e1 N |
2163 | /* |
2164 | * i_size must be checked after we know the page is Uptodate. | |
2165 | * | |
2166 | * Checking i_size after the check allows us to calculate | |
2167 | * the correct value for "nr", which means the zero-filled | |
2168 | * part of the page is not copied back to userspace (unless | |
2169 | * another truncate extends the file - this is desired though). | |
2170 | */ | |
2171 | ||
2172 | isize = i_size_read(inode); | |
09cbfeaf | 2173 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2174 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2175 | put_page(page); |
a32ea1e1 N |
2176 | goto out; |
2177 | } | |
2178 | ||
2179 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2180 | nr = PAGE_SIZE; |
a32ea1e1 | 2181 | if (index == end_index) { |
09cbfeaf | 2182 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2183 | if (nr <= offset) { |
09cbfeaf | 2184 | put_page(page); |
a32ea1e1 N |
2185 | goto out; |
2186 | } | |
2187 | } | |
2188 | nr = nr - offset; | |
1da177e4 LT |
2189 | |
2190 | /* If users can be writing to this page using arbitrary | |
2191 | * virtual addresses, take care about potential aliasing | |
2192 | * before reading the page on the kernel side. | |
2193 | */ | |
2194 | if (mapping_writably_mapped(mapping)) | |
2195 | flush_dcache_page(page); | |
2196 | ||
2197 | /* | |
ec0f1637 JK |
2198 | * When a sequential read accesses a page several times, |
2199 | * only mark it as accessed the first time. | |
1da177e4 | 2200 | */ |
ec0f1637 | 2201 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2202 | mark_page_accessed(page); |
2203 | prev_index = index; | |
2204 | ||
2205 | /* | |
2206 | * Ok, we have the page, and it's up-to-date, so | |
2207 | * now we can copy it to user space... | |
1da177e4 | 2208 | */ |
6e58e79d AV |
2209 | |
2210 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2211 | offset += ret; |
09cbfeaf KS |
2212 | index += offset >> PAGE_SHIFT; |
2213 | offset &= ~PAGE_MASK; | |
6ce745ed | 2214 | prev_offset = offset; |
1da177e4 | 2215 | |
09cbfeaf | 2216 | put_page(page); |
6e58e79d AV |
2217 | written += ret; |
2218 | if (!iov_iter_count(iter)) | |
2219 | goto out; | |
2220 | if (ret < nr) { | |
2221 | error = -EFAULT; | |
2222 | goto out; | |
2223 | } | |
2224 | continue; | |
1da177e4 LT |
2225 | |
2226 | page_not_up_to_date: | |
2227 | /* Get exclusive access to the page ... */ | |
85462323 ON |
2228 | error = lock_page_killable(page); |
2229 | if (unlikely(error)) | |
2230 | goto readpage_error; | |
1da177e4 | 2231 | |
8ab22b9a | 2232 | page_not_up_to_date_locked: |
da6052f7 | 2233 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2234 | if (!page->mapping) { |
2235 | unlock_page(page); | |
09cbfeaf | 2236 | put_page(page); |
1da177e4 LT |
2237 | continue; |
2238 | } | |
2239 | ||
2240 | /* Did somebody else fill it already? */ | |
2241 | if (PageUptodate(page)) { | |
2242 | unlock_page(page); | |
2243 | goto page_ok; | |
2244 | } | |
2245 | ||
2246 | readpage: | |
91803b49 JM |
2247 | /* |
2248 | * A previous I/O error may have been due to temporary | |
2249 | * failures, eg. multipath errors. | |
2250 | * PG_error will be set again if readpage fails. | |
2251 | */ | |
2252 | ClearPageError(page); | |
1da177e4 LT |
2253 | /* Start the actual read. The read will unlock the page. */ |
2254 | error = mapping->a_ops->readpage(filp, page); | |
2255 | ||
994fc28c ZB |
2256 | if (unlikely(error)) { |
2257 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2258 | put_page(page); |
6e58e79d | 2259 | error = 0; |
994fc28c ZB |
2260 | goto find_page; |
2261 | } | |
1da177e4 | 2262 | goto readpage_error; |
994fc28c | 2263 | } |
1da177e4 LT |
2264 | |
2265 | if (!PageUptodate(page)) { | |
85462323 ON |
2266 | error = lock_page_killable(page); |
2267 | if (unlikely(error)) | |
2268 | goto readpage_error; | |
1da177e4 LT |
2269 | if (!PageUptodate(page)) { |
2270 | if (page->mapping == NULL) { | |
2271 | /* | |
2ecdc82e | 2272 | * invalidate_mapping_pages got it |
1da177e4 LT |
2273 | */ |
2274 | unlock_page(page); | |
09cbfeaf | 2275 | put_page(page); |
1da177e4 LT |
2276 | goto find_page; |
2277 | } | |
2278 | unlock_page(page); | |
7ff81078 | 2279 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
2280 | error = -EIO; |
2281 | goto readpage_error; | |
1da177e4 LT |
2282 | } |
2283 | unlock_page(page); | |
2284 | } | |
2285 | ||
1da177e4 LT |
2286 | goto page_ok; |
2287 | ||
2288 | readpage_error: | |
2289 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2290 | put_page(page); |
1da177e4 LT |
2291 | goto out; |
2292 | ||
2293 | no_cached_page: | |
2294 | /* | |
2295 | * Ok, it wasn't cached, so we need to create a new | |
2296 | * page.. | |
2297 | */ | |
453f85d4 | 2298 | page = page_cache_alloc(mapping); |
eb2be189 | 2299 | if (!page) { |
6e58e79d | 2300 | error = -ENOMEM; |
eb2be189 | 2301 | goto out; |
1da177e4 | 2302 | } |
6afdb859 | 2303 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2304 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2305 | if (error) { |
09cbfeaf | 2306 | put_page(page); |
6e58e79d AV |
2307 | if (error == -EEXIST) { |
2308 | error = 0; | |
1da177e4 | 2309 | goto find_page; |
6e58e79d | 2310 | } |
1da177e4 LT |
2311 | goto out; |
2312 | } | |
1da177e4 LT |
2313 | goto readpage; |
2314 | } | |
2315 | ||
3239d834 MT |
2316 | would_block: |
2317 | error = -EAGAIN; | |
1da177e4 | 2318 | out: |
7ff81078 | 2319 | ra->prev_pos = prev_index; |
09cbfeaf | 2320 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2321 | ra->prev_pos |= prev_offset; |
1da177e4 | 2322 | |
09cbfeaf | 2323 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2324 | file_accessed(filp); |
6e58e79d | 2325 | return written ? written : error; |
1da177e4 LT |
2326 | } |
2327 | ||
485bb99b | 2328 | /** |
6abd2322 | 2329 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2330 | * @iocb: kernel I/O control block |
6abd2322 | 2331 | * @iter: destination for the data read |
485bb99b | 2332 | * |
6abd2322 | 2333 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
2334 | * that can use the page cache directly. |
2335 | */ | |
2336 | ssize_t | |
ed978a81 | 2337 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2338 | { |
e7080a43 | 2339 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2340 | ssize_t retval = 0; |
e7080a43 NS |
2341 | |
2342 | if (!count) | |
2343 | goto out; /* skip atime */ | |
1da177e4 | 2344 | |
2ba48ce5 | 2345 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2346 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2347 | struct address_space *mapping = file->f_mapping; |
2348 | struct inode *inode = mapping->host; | |
543ade1f | 2349 | loff_t size; |
1da177e4 | 2350 | |
1da177e4 | 2351 | size = i_size_read(inode); |
6be96d3a GR |
2352 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2353 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2354 | iocb->ki_pos + count - 1)) | |
2355 | return -EAGAIN; | |
2356 | } else { | |
2357 | retval = filemap_write_and_wait_range(mapping, | |
2358 | iocb->ki_pos, | |
2359 | iocb->ki_pos + count - 1); | |
2360 | if (retval < 0) | |
2361 | goto out; | |
2362 | } | |
d8d3d94b | 2363 | |
0d5b0cf2 CH |
2364 | file_accessed(file); |
2365 | ||
5ecda137 | 2366 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2367 | if (retval >= 0) { |
c64fb5c7 | 2368 | iocb->ki_pos += retval; |
5ecda137 | 2369 | count -= retval; |
9fe55eea | 2370 | } |
5b47d59a | 2371 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2372 | |
9fe55eea SW |
2373 | /* |
2374 | * Btrfs can have a short DIO read if we encounter | |
2375 | * compressed extents, so if there was an error, or if | |
2376 | * we've already read everything we wanted to, or if | |
2377 | * there was a short read because we hit EOF, go ahead | |
2378 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2379 | * the rest of the read. Buffered reads will not work for |
2380 | * DAX files, so don't bother trying. | |
9fe55eea | 2381 | */ |
5ecda137 | 2382 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2383 | IS_DAX(inode)) |
9fe55eea | 2384 | goto out; |
1da177e4 LT |
2385 | } |
2386 | ||
47c27bc4 | 2387 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2388 | out: |
2389 | return retval; | |
2390 | } | |
ed978a81 | 2391 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2392 | |
1da177e4 | 2393 | #ifdef CONFIG_MMU |
485bb99b RD |
2394 | /** |
2395 | * page_cache_read - adds requested page to the page cache if not already there | |
2396 | * @file: file to read | |
2397 | * @offset: page index | |
62eb320a | 2398 | * @gfp_mask: memory allocation flags |
485bb99b | 2399 | * |
1da177e4 LT |
2400 | * This adds the requested page to the page cache if it isn't already there, |
2401 | * and schedules an I/O to read in its contents from disk. | |
2402 | */ | |
c20cd45e | 2403 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
2404 | { |
2405 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 2406 | struct page *page; |
994fc28c | 2407 | int ret; |
1da177e4 | 2408 | |
994fc28c | 2409 | do { |
453f85d4 | 2410 | page = __page_cache_alloc(gfp_mask); |
994fc28c ZB |
2411 | if (!page) |
2412 | return -ENOMEM; | |
2413 | ||
5b8748b5 | 2414 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
994fc28c ZB |
2415 | if (ret == 0) |
2416 | ret = mapping->a_ops->readpage(file, page); | |
2417 | else if (ret == -EEXIST) | |
2418 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 2419 | |
09cbfeaf | 2420 | put_page(page); |
1da177e4 | 2421 | |
994fc28c | 2422 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 2423 | |
994fc28c | 2424 | return ret; |
1da177e4 LT |
2425 | } |
2426 | ||
2427 | #define MMAP_LOTSAMISS (100) | |
2428 | ||
ef00e08e LT |
2429 | /* |
2430 | * Synchronous readahead happens when we don't even find | |
2431 | * a page in the page cache at all. | |
2432 | */ | |
2433 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
2434 | struct file_ra_state *ra, | |
2435 | struct file *file, | |
2436 | pgoff_t offset) | |
2437 | { | |
ef00e08e LT |
2438 | struct address_space *mapping = file->f_mapping; |
2439 | ||
2440 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2441 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 2442 | return; |
275b12bf WF |
2443 | if (!ra->ra_pages) |
2444 | return; | |
ef00e08e | 2445 | |
64363aad | 2446 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
2447 | page_cache_sync_readahead(mapping, ra, file, offset, |
2448 | ra->ra_pages); | |
ef00e08e LT |
2449 | return; |
2450 | } | |
2451 | ||
207d04ba AK |
2452 | /* Avoid banging the cache line if not needed */ |
2453 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2454 | ra->mmap_miss++; |
2455 | ||
2456 | /* | |
2457 | * Do we miss much more than hit in this file? If so, | |
2458 | * stop bothering with read-ahead. It will only hurt. | |
2459 | */ | |
2460 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
2461 | return; | |
2462 | ||
d30a1100 WF |
2463 | /* |
2464 | * mmap read-around | |
2465 | */ | |
600e19af RG |
2466 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2467 | ra->size = ra->ra_pages; | |
2468 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2469 | ra_submit(ra, mapping, file); |
ef00e08e LT |
2470 | } |
2471 | ||
2472 | /* | |
2473 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2474 | * so we want to possibly extend the readahead further.. | |
2475 | */ | |
2476 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2477 | struct file_ra_state *ra, | |
2478 | struct file *file, | |
2479 | struct page *page, | |
2480 | pgoff_t offset) | |
2481 | { | |
2482 | struct address_space *mapping = file->f_mapping; | |
2483 | ||
2484 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2485 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2486 | return; |
2487 | if (ra->mmap_miss > 0) | |
2488 | ra->mmap_miss--; | |
2489 | if (PageReadahead(page)) | |
2fad6f5d WF |
2490 | page_cache_async_readahead(mapping, ra, file, |
2491 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2492 | } |
2493 | ||
485bb99b | 2494 | /** |
54cb8821 | 2495 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2496 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2497 | * |
54cb8821 | 2498 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2499 | * mapped memory region to read in file data during a page fault. |
2500 | * | |
2501 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2502 | * it in the page cache, and handles the special cases reasonably without | |
2503 | * having a lot of duplicated code. | |
9a95f3cf PC |
2504 | * |
2505 | * vma->vm_mm->mmap_sem must be held on entry. | |
2506 | * | |
2507 | * If our return value has VM_FAULT_RETRY set, it's because | |
2508 | * lock_page_or_retry() returned 0. | |
2509 | * The mmap_sem has usually been released in this case. | |
2510 | * See __lock_page_or_retry() for the exception. | |
2511 | * | |
2512 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2513 | * has not been released. | |
2514 | * | |
2515 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2516 | */ |
11bac800 | 2517 | int filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2518 | { |
2519 | int error; | |
11bac800 | 2520 | struct file *file = vmf->vma->vm_file; |
1da177e4 LT |
2521 | struct address_space *mapping = file->f_mapping; |
2522 | struct file_ra_state *ra = &file->f_ra; | |
2523 | struct inode *inode = mapping->host; | |
ef00e08e | 2524 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2525 | pgoff_t max_off; |
1da177e4 | 2526 | struct page *page; |
83c54070 | 2527 | int ret = 0; |
1da177e4 | 2528 | |
9ab2594f MW |
2529 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2530 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2531 | return VM_FAULT_SIGBUS; |
1da177e4 | 2532 | |
1da177e4 | 2533 | /* |
49426420 | 2534 | * Do we have something in the page cache already? |
1da177e4 | 2535 | */ |
ef00e08e | 2536 | page = find_get_page(mapping, offset); |
45cac65b | 2537 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2538 | /* |
ef00e08e LT |
2539 | * We found the page, so try async readahead before |
2540 | * waiting for the lock. | |
1da177e4 | 2541 | */ |
11bac800 | 2542 | do_async_mmap_readahead(vmf->vma, ra, file, page, offset); |
45cac65b | 2543 | } else if (!page) { |
ef00e08e | 2544 | /* No page in the page cache at all */ |
11bac800 | 2545 | do_sync_mmap_readahead(vmf->vma, ra, file, offset); |
ef00e08e | 2546 | count_vm_event(PGMAJFAULT); |
2262185c | 2547 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2548 | ret = VM_FAULT_MAJOR; |
2549 | retry_find: | |
b522c94d | 2550 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2551 | if (!page) |
2552 | goto no_cached_page; | |
2553 | } | |
2554 | ||
11bac800 | 2555 | if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2556 | put_page(page); |
d065bd81 | 2557 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2558 | } |
b522c94d ML |
2559 | |
2560 | /* Did it get truncated? */ | |
2561 | if (unlikely(page->mapping != mapping)) { | |
2562 | unlock_page(page); | |
2563 | put_page(page); | |
2564 | goto retry_find; | |
2565 | } | |
309381fe | 2566 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2567 | |
1da177e4 | 2568 | /* |
d00806b1 NP |
2569 | * We have a locked page in the page cache, now we need to check |
2570 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2571 | */ |
d00806b1 | 2572 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2573 | goto page_not_uptodate; |
2574 | ||
ef00e08e LT |
2575 | /* |
2576 | * Found the page and have a reference on it. | |
2577 | * We must recheck i_size under page lock. | |
2578 | */ | |
9ab2594f MW |
2579 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2580 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2581 | unlock_page(page); |
09cbfeaf | 2582 | put_page(page); |
5307cc1a | 2583 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2584 | } |
2585 | ||
d0217ac0 | 2586 | vmf->page = page; |
83c54070 | 2587 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2588 | |
1da177e4 LT |
2589 | no_cached_page: |
2590 | /* | |
2591 | * We're only likely to ever get here if MADV_RANDOM is in | |
2592 | * effect. | |
2593 | */ | |
c20cd45e | 2594 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2595 | |
2596 | /* | |
2597 | * The page we want has now been added to the page cache. | |
2598 | * In the unlikely event that someone removed it in the | |
2599 | * meantime, we'll just come back here and read it again. | |
2600 | */ | |
2601 | if (error >= 0) | |
2602 | goto retry_find; | |
2603 | ||
2604 | /* | |
2605 | * An error return from page_cache_read can result if the | |
2606 | * system is low on memory, or a problem occurs while trying | |
2607 | * to schedule I/O. | |
2608 | */ | |
2609 | if (error == -ENOMEM) | |
d0217ac0 NP |
2610 | return VM_FAULT_OOM; |
2611 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2612 | |
2613 | page_not_uptodate: | |
1da177e4 LT |
2614 | /* |
2615 | * Umm, take care of errors if the page isn't up-to-date. | |
2616 | * Try to re-read it _once_. We do this synchronously, | |
2617 | * because there really aren't any performance issues here | |
2618 | * and we need to check for errors. | |
2619 | */ | |
1da177e4 | 2620 | ClearPageError(page); |
994fc28c | 2621 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2622 | if (!error) { |
2623 | wait_on_page_locked(page); | |
2624 | if (!PageUptodate(page)) | |
2625 | error = -EIO; | |
2626 | } | |
09cbfeaf | 2627 | put_page(page); |
d00806b1 NP |
2628 | |
2629 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2630 | goto retry_find; |
1da177e4 | 2631 | |
d00806b1 | 2632 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2633 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2634 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2635 | } |
2636 | EXPORT_SYMBOL(filemap_fault); | |
2637 | ||
82b0f8c3 | 2638 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2639 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 KS |
2640 | { |
2641 | struct radix_tree_iter iter; | |
2642 | void **slot; | |
82b0f8c3 | 2643 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2644 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2645 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2646 | unsigned long max_idx; |
83929372 | 2647 | struct page *head, *page; |
f1820361 KS |
2648 | |
2649 | rcu_read_lock(); | |
bae473a4 KS |
2650 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, |
2651 | start_pgoff) { | |
2652 | if (iter.index > end_pgoff) | |
f1820361 KS |
2653 | break; |
2654 | repeat: | |
2655 | page = radix_tree_deref_slot(slot); | |
2656 | if (unlikely(!page)) | |
2657 | goto next; | |
2658 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2659 | if (radix_tree_deref_retry(page)) { |
2660 | slot = radix_tree_iter_retry(&iter); | |
2661 | continue; | |
2662 | } | |
2663 | goto next; | |
f1820361 KS |
2664 | } |
2665 | ||
83929372 KS |
2666 | head = compound_head(page); |
2667 | if (!page_cache_get_speculative(head)) | |
f1820361 KS |
2668 | goto repeat; |
2669 | ||
83929372 KS |
2670 | /* The page was split under us? */ |
2671 | if (compound_head(page) != head) { | |
2672 | put_page(head); | |
2673 | goto repeat; | |
2674 | } | |
2675 | ||
f1820361 KS |
2676 | /* Has the page moved? */ |
2677 | if (unlikely(page != *slot)) { | |
83929372 | 2678 | put_page(head); |
f1820361 KS |
2679 | goto repeat; |
2680 | } | |
2681 | ||
2682 | if (!PageUptodate(page) || | |
2683 | PageReadahead(page) || | |
2684 | PageHWPoison(page)) | |
2685 | goto skip; | |
2686 | if (!trylock_page(page)) | |
2687 | goto skip; | |
2688 | ||
2689 | if (page->mapping != mapping || !PageUptodate(page)) | |
2690 | goto unlock; | |
2691 | ||
9ab2594f MW |
2692 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2693 | if (page->index >= max_idx) | |
f1820361 KS |
2694 | goto unlock; |
2695 | ||
f1820361 KS |
2696 | if (file->f_ra.mmap_miss > 0) |
2697 | file->f_ra.mmap_miss--; | |
7267ec00 | 2698 | |
82b0f8c3 JK |
2699 | vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT; |
2700 | if (vmf->pte) | |
2701 | vmf->pte += iter.index - last_pgoff; | |
7267ec00 | 2702 | last_pgoff = iter.index; |
82b0f8c3 | 2703 | if (alloc_set_pte(vmf, NULL, page)) |
7267ec00 | 2704 | goto unlock; |
f1820361 KS |
2705 | unlock_page(page); |
2706 | goto next; | |
2707 | unlock: | |
2708 | unlock_page(page); | |
2709 | skip: | |
09cbfeaf | 2710 | put_page(page); |
f1820361 | 2711 | next: |
7267ec00 | 2712 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2713 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2714 | break; |
bae473a4 | 2715 | if (iter.index == end_pgoff) |
f1820361 KS |
2716 | break; |
2717 | } | |
2718 | rcu_read_unlock(); | |
2719 | } | |
2720 | EXPORT_SYMBOL(filemap_map_pages); | |
2721 | ||
11bac800 | 2722 | int filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2723 | { |
2724 | struct page *page = vmf->page; | |
11bac800 | 2725 | struct inode *inode = file_inode(vmf->vma->vm_file); |
4fcf1c62 JK |
2726 | int ret = VM_FAULT_LOCKED; |
2727 | ||
14da9200 | 2728 | sb_start_pagefault(inode->i_sb); |
c088e31d | 2729 | vma_file_update_time(vmf->vma); |
4fcf1c62 JK |
2730 | lock_page(page); |
2731 | if (page->mapping != inode->i_mapping) { | |
2732 | unlock_page(page); | |
2733 | ret = VM_FAULT_NOPAGE; | |
2734 | goto out; | |
2735 | } | |
14da9200 JK |
2736 | /* |
2737 | * We mark the page dirty already here so that when freeze is in | |
2738 | * progress, we are guaranteed that writeback during freezing will | |
2739 | * see the dirty page and writeprotect it again. | |
2740 | */ | |
2741 | set_page_dirty(page); | |
1d1d1a76 | 2742 | wait_for_stable_page(page); |
4fcf1c62 | 2743 | out: |
14da9200 | 2744 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2745 | return ret; |
2746 | } | |
2747 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2748 | ||
f0f37e2f | 2749 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2750 | .fault = filemap_fault, |
f1820361 | 2751 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2752 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2753 | }; |
2754 | ||
2755 | /* This is used for a general mmap of a disk file */ | |
2756 | ||
2757 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2758 | { | |
2759 | struct address_space *mapping = file->f_mapping; | |
2760 | ||
2761 | if (!mapping->a_ops->readpage) | |
2762 | return -ENOEXEC; | |
2763 | file_accessed(file); | |
2764 | vma->vm_ops = &generic_file_vm_ops; | |
2765 | return 0; | |
2766 | } | |
1da177e4 LT |
2767 | |
2768 | /* | |
2769 | * This is for filesystems which do not implement ->writepage. | |
2770 | */ | |
2771 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2772 | { | |
2773 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2774 | return -EINVAL; | |
2775 | return generic_file_mmap(file, vma); | |
2776 | } | |
2777 | #else | |
2778 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2779 | { | |
2780 | return -ENOSYS; | |
2781 | } | |
2782 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2783 | { | |
2784 | return -ENOSYS; | |
2785 | } | |
2786 | #endif /* CONFIG_MMU */ | |
2787 | ||
2788 | EXPORT_SYMBOL(generic_file_mmap); | |
2789 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2790 | ||
67f9fd91 SL |
2791 | static struct page *wait_on_page_read(struct page *page) |
2792 | { | |
2793 | if (!IS_ERR(page)) { | |
2794 | wait_on_page_locked(page); | |
2795 | if (!PageUptodate(page)) { | |
09cbfeaf | 2796 | put_page(page); |
67f9fd91 SL |
2797 | page = ERR_PTR(-EIO); |
2798 | } | |
2799 | } | |
2800 | return page; | |
2801 | } | |
2802 | ||
32b63529 | 2803 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2804 | pgoff_t index, |
5e5358e7 | 2805 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2806 | void *data, |
2807 | gfp_t gfp) | |
1da177e4 | 2808 | { |
eb2be189 | 2809 | struct page *page; |
1da177e4 LT |
2810 | int err; |
2811 | repeat: | |
2812 | page = find_get_page(mapping, index); | |
2813 | if (!page) { | |
453f85d4 | 2814 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
2815 | if (!page) |
2816 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2817 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2818 | if (unlikely(err)) { |
09cbfeaf | 2819 | put_page(page); |
eb2be189 NP |
2820 | if (err == -EEXIST) |
2821 | goto repeat; | |
1da177e4 | 2822 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2823 | return ERR_PTR(err); |
2824 | } | |
32b63529 MG |
2825 | |
2826 | filler: | |
1da177e4 LT |
2827 | err = filler(data, page); |
2828 | if (err < 0) { | |
09cbfeaf | 2829 | put_page(page); |
32b63529 | 2830 | return ERR_PTR(err); |
1da177e4 | 2831 | } |
1da177e4 | 2832 | |
32b63529 MG |
2833 | page = wait_on_page_read(page); |
2834 | if (IS_ERR(page)) | |
2835 | return page; | |
2836 | goto out; | |
2837 | } | |
1da177e4 LT |
2838 | if (PageUptodate(page)) |
2839 | goto out; | |
2840 | ||
ebded027 MG |
2841 | /* |
2842 | * Page is not up to date and may be locked due one of the following | |
2843 | * case a: Page is being filled and the page lock is held | |
2844 | * case b: Read/write error clearing the page uptodate status | |
2845 | * case c: Truncation in progress (page locked) | |
2846 | * case d: Reclaim in progress | |
2847 | * | |
2848 | * Case a, the page will be up to date when the page is unlocked. | |
2849 | * There is no need to serialise on the page lock here as the page | |
2850 | * is pinned so the lock gives no additional protection. Even if the | |
2851 | * the page is truncated, the data is still valid if PageUptodate as | |
2852 | * it's a race vs truncate race. | |
2853 | * Case b, the page will not be up to date | |
2854 | * Case c, the page may be truncated but in itself, the data may still | |
2855 | * be valid after IO completes as it's a read vs truncate race. The | |
2856 | * operation must restart if the page is not uptodate on unlock but | |
2857 | * otherwise serialising on page lock to stabilise the mapping gives | |
2858 | * no additional guarantees to the caller as the page lock is | |
2859 | * released before return. | |
2860 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2861 | * will be a race with remove_mapping that determines if the mapping | |
2862 | * is valid on unlock but otherwise the data is valid and there is | |
2863 | * no need to serialise with page lock. | |
2864 | * | |
2865 | * As the page lock gives no additional guarantee, we optimistically | |
2866 | * wait on the page to be unlocked and check if it's up to date and | |
2867 | * use the page if it is. Otherwise, the page lock is required to | |
2868 | * distinguish between the different cases. The motivation is that we | |
2869 | * avoid spurious serialisations and wakeups when multiple processes | |
2870 | * wait on the same page for IO to complete. | |
2871 | */ | |
2872 | wait_on_page_locked(page); | |
2873 | if (PageUptodate(page)) | |
2874 | goto out; | |
2875 | ||
2876 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2877 | lock_page(page); |
ebded027 MG |
2878 | |
2879 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2880 | if (!page->mapping) { |
2881 | unlock_page(page); | |
09cbfeaf | 2882 | put_page(page); |
32b63529 | 2883 | goto repeat; |
1da177e4 | 2884 | } |
ebded027 MG |
2885 | |
2886 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2887 | if (PageUptodate(page)) { |
2888 | unlock_page(page); | |
2889 | goto out; | |
2890 | } | |
32b63529 MG |
2891 | goto filler; |
2892 | ||
c855ff37 | 2893 | out: |
6fe6900e NP |
2894 | mark_page_accessed(page); |
2895 | return page; | |
2896 | } | |
0531b2aa LT |
2897 | |
2898 | /** | |
67f9fd91 | 2899 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2900 | * @mapping: the page's address_space |
2901 | * @index: the page index | |
2902 | * @filler: function to perform the read | |
5e5358e7 | 2903 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2904 | * |
0531b2aa | 2905 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2906 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2907 | * |
2908 | * If the page does not get brought uptodate, return -EIO. | |
2909 | */ | |
67f9fd91 | 2910 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2911 | pgoff_t index, |
5e5358e7 | 2912 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2913 | void *data) |
2914 | { | |
2915 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2916 | } | |
67f9fd91 | 2917 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2918 | |
2919 | /** | |
2920 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2921 | * @mapping: the page's address_space | |
2922 | * @index: the page index | |
2923 | * @gfp: the page allocator flags to use if allocating | |
2924 | * | |
2925 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2926 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2927 | * |
2928 | * If the page does not get brought uptodate, return -EIO. | |
2929 | */ | |
2930 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2931 | pgoff_t index, | |
2932 | gfp_t gfp) | |
2933 | { | |
2934 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2935 | ||
67f9fd91 | 2936 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2937 | } |
2938 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2939 | ||
1da177e4 LT |
2940 | /* |
2941 | * Performs necessary checks before doing a write | |
2942 | * | |
485bb99b | 2943 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2944 | * Returns appropriate error code that caller should return or |
2945 | * zero in case that write should be allowed. | |
2946 | */ | |
3309dd04 | 2947 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2948 | { |
3309dd04 | 2949 | struct file *file = iocb->ki_filp; |
1da177e4 | 2950 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2951 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2952 | loff_t pos; |
1da177e4 | 2953 | |
3309dd04 AV |
2954 | if (!iov_iter_count(from)) |
2955 | return 0; | |
1da177e4 | 2956 | |
0fa6b005 | 2957 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2958 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2959 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2960 | |
3309dd04 | 2961 | pos = iocb->ki_pos; |
1da177e4 | 2962 | |
6be96d3a GR |
2963 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
2964 | return -EINVAL; | |
2965 | ||
0fa6b005 | 2966 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2967 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2968 | send_sig(SIGXFSZ, current, 0); |
2969 | return -EFBIG; | |
1da177e4 | 2970 | } |
3309dd04 | 2971 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2972 | } |
2973 | ||
2974 | /* | |
2975 | * LFS rule | |
2976 | */ | |
3309dd04 | 2977 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2978 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2979 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2980 | return -EFBIG; |
3309dd04 | 2981 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2982 | } |
2983 | ||
2984 | /* | |
2985 | * Are we about to exceed the fs block limit ? | |
2986 | * | |
2987 | * If we have written data it becomes a short write. If we have | |
2988 | * exceeded without writing data we send a signal and return EFBIG. | |
2989 | * Linus frestrict idea will clean these up nicely.. | |
2990 | */ | |
3309dd04 AV |
2991 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2992 | return -EFBIG; | |
1da177e4 | 2993 | |
3309dd04 AV |
2994 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2995 | return iov_iter_count(from); | |
1da177e4 LT |
2996 | } |
2997 | EXPORT_SYMBOL(generic_write_checks); | |
2998 | ||
afddba49 NP |
2999 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3000 | loff_t pos, unsigned len, unsigned flags, | |
3001 | struct page **pagep, void **fsdata) | |
3002 | { | |
3003 | const struct address_space_operations *aops = mapping->a_ops; | |
3004 | ||
4e02ed4b | 3005 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3006 | pagep, fsdata); |
afddba49 NP |
3007 | } |
3008 | EXPORT_SYMBOL(pagecache_write_begin); | |
3009 | ||
3010 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3011 | loff_t pos, unsigned len, unsigned copied, | |
3012 | struct page *page, void *fsdata) | |
3013 | { | |
3014 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3015 | |
4e02ed4b | 3016 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3017 | } |
3018 | EXPORT_SYMBOL(pagecache_write_end); | |
3019 | ||
1da177e4 | 3020 | ssize_t |
1af5bb49 | 3021 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3022 | { |
3023 | struct file *file = iocb->ki_filp; | |
3024 | struct address_space *mapping = file->f_mapping; | |
3025 | struct inode *inode = mapping->host; | |
1af5bb49 | 3026 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3027 | ssize_t written; |
a969e903 CH |
3028 | size_t write_len; |
3029 | pgoff_t end; | |
1da177e4 | 3030 | |
0c949334 | 3031 | write_len = iov_iter_count(from); |
09cbfeaf | 3032 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3033 | |
6be96d3a GR |
3034 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3035 | /* If there are pages to writeback, return */ | |
3036 | if (filemap_range_has_page(inode->i_mapping, pos, | |
3037 | pos + iov_iter_count(from))) | |
3038 | return -EAGAIN; | |
3039 | } else { | |
3040 | written = filemap_write_and_wait_range(mapping, pos, | |
3041 | pos + write_len - 1); | |
3042 | if (written) | |
3043 | goto out; | |
3044 | } | |
a969e903 CH |
3045 | |
3046 | /* | |
3047 | * After a write we want buffered reads to be sure to go to disk to get | |
3048 | * the new data. We invalidate clean cached page from the region we're | |
3049 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3050 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3051 | */ |
55635ba7 | 3052 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3053 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3054 | /* |
3055 | * If a page can not be invalidated, return 0 to fall back | |
3056 | * to buffered write. | |
3057 | */ | |
3058 | if (written) { | |
3059 | if (written == -EBUSY) | |
3060 | return 0; | |
3061 | goto out; | |
a969e903 CH |
3062 | } |
3063 | ||
639a93a5 | 3064 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3065 | |
3066 | /* | |
3067 | * Finally, try again to invalidate clean pages which might have been | |
3068 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3069 | * if the source of the write was an mmap'ed region of the file | |
3070 | * we're writing. Either one is a pretty crazy thing to do, | |
3071 | * so we don't support it 100%. If this invalidation | |
3072 | * fails, tough, the write still worked... | |
332391a9 LC |
3073 | * |
3074 | * Most of the time we do not need this since dio_complete() will do | |
3075 | * the invalidation for us. However there are some file systems that | |
3076 | * do not end up with dio_complete() being called, so let's not break | |
3077 | * them by removing it completely | |
a969e903 | 3078 | */ |
332391a9 LC |
3079 | if (mapping->nrpages) |
3080 | invalidate_inode_pages2_range(mapping, | |
3081 | pos >> PAGE_SHIFT, end); | |
a969e903 | 3082 | |
1da177e4 | 3083 | if (written > 0) { |
0116651c | 3084 | pos += written; |
639a93a5 | 3085 | write_len -= written; |
0116651c NK |
3086 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3087 | i_size_write(inode, pos); | |
1da177e4 LT |
3088 | mark_inode_dirty(inode); |
3089 | } | |
5cb6c6c7 | 3090 | iocb->ki_pos = pos; |
1da177e4 | 3091 | } |
639a93a5 | 3092 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3093 | out: |
1da177e4 LT |
3094 | return written; |
3095 | } | |
3096 | EXPORT_SYMBOL(generic_file_direct_write); | |
3097 | ||
eb2be189 NP |
3098 | /* |
3099 | * Find or create a page at the given pagecache position. Return the locked | |
3100 | * page. This function is specifically for buffered writes. | |
3101 | */ | |
54566b2c NP |
3102 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3103 | pgoff_t index, unsigned flags) | |
eb2be189 | 3104 | { |
eb2be189 | 3105 | struct page *page; |
bbddabe2 | 3106 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3107 | |
54566b2c | 3108 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3109 | fgp_flags |= FGP_NOFS; |
3110 | ||
3111 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3112 | mapping_gfp_mask(mapping)); |
c585a267 | 3113 | if (page) |
2457aec6 | 3114 | wait_for_stable_page(page); |
eb2be189 | 3115 | |
eb2be189 NP |
3116 | return page; |
3117 | } | |
54566b2c | 3118 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3119 | |
3b93f911 | 3120 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3121 | struct iov_iter *i, loff_t pos) |
3122 | { | |
3123 | struct address_space *mapping = file->f_mapping; | |
3124 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3125 | long status = 0; | |
3126 | ssize_t written = 0; | |
674b892e NP |
3127 | unsigned int flags = 0; |
3128 | ||
afddba49 NP |
3129 | do { |
3130 | struct page *page; | |
afddba49 NP |
3131 | unsigned long offset; /* Offset into pagecache page */ |
3132 | unsigned long bytes; /* Bytes to write to page */ | |
3133 | size_t copied; /* Bytes copied from user */ | |
3134 | void *fsdata; | |
3135 | ||
09cbfeaf KS |
3136 | offset = (pos & (PAGE_SIZE - 1)); |
3137 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3138 | iov_iter_count(i)); |
3139 | ||
3140 | again: | |
00a3d660 LT |
3141 | /* |
3142 | * Bring in the user page that we will copy from _first_. | |
3143 | * Otherwise there's a nasty deadlock on copying from the | |
3144 | * same page as we're writing to, without it being marked | |
3145 | * up-to-date. | |
3146 | * | |
3147 | * Not only is this an optimisation, but it is also required | |
3148 | * to check that the address is actually valid, when atomic | |
3149 | * usercopies are used, below. | |
3150 | */ | |
3151 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3152 | status = -EFAULT; | |
3153 | break; | |
3154 | } | |
3155 | ||
296291cd JK |
3156 | if (fatal_signal_pending(current)) { |
3157 | status = -EINTR; | |
3158 | break; | |
3159 | } | |
3160 | ||
674b892e | 3161 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3162 | &page, &fsdata); |
2457aec6 | 3163 | if (unlikely(status < 0)) |
afddba49 NP |
3164 | break; |
3165 | ||
931e80e4 | 3166 | if (mapping_writably_mapped(mapping)) |
3167 | flush_dcache_page(page); | |
00a3d660 | 3168 | |
afddba49 | 3169 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3170 | flush_dcache_page(page); |
3171 | ||
3172 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3173 | page, fsdata); | |
3174 | if (unlikely(status < 0)) | |
3175 | break; | |
3176 | copied = status; | |
3177 | ||
3178 | cond_resched(); | |
3179 | ||
124d3b70 | 3180 | iov_iter_advance(i, copied); |
afddba49 NP |
3181 | if (unlikely(copied == 0)) { |
3182 | /* | |
3183 | * If we were unable to copy any data at all, we must | |
3184 | * fall back to a single segment length write. | |
3185 | * | |
3186 | * If we didn't fallback here, we could livelock | |
3187 | * because not all segments in the iov can be copied at | |
3188 | * once without a pagefault. | |
3189 | */ | |
09cbfeaf | 3190 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3191 | iov_iter_single_seg_count(i)); |
3192 | goto again; | |
3193 | } | |
afddba49 NP |
3194 | pos += copied; |
3195 | written += copied; | |
3196 | ||
3197 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3198 | } while (iov_iter_count(i)); |
3199 | ||
3200 | return written ? written : status; | |
3201 | } | |
3b93f911 | 3202 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3203 | |
e4dd9de3 | 3204 | /** |
8174202b | 3205 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3206 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3207 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3208 | * |
3209 | * This function does all the work needed for actually writing data to a | |
3210 | * file. It does all basic checks, removes SUID from the file, updates | |
3211 | * modification times and calls proper subroutines depending on whether we | |
3212 | * do direct IO or a standard buffered write. | |
3213 | * | |
3214 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3215 | * object which does not need locking at all. | |
3216 | * | |
3217 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3218 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3219 | * avoid syncing under i_mutex. | |
3220 | */ | |
8174202b | 3221 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3222 | { |
3223 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3224 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3225 | struct inode *inode = mapping->host; |
3b93f911 | 3226 | ssize_t written = 0; |
1da177e4 | 3227 | ssize_t err; |
3b93f911 | 3228 | ssize_t status; |
1da177e4 | 3229 | |
1da177e4 | 3230 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3231 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3232 | err = file_remove_privs(file); |
1da177e4 LT |
3233 | if (err) |
3234 | goto out; | |
3235 | ||
c3b2da31 JB |
3236 | err = file_update_time(file); |
3237 | if (err) | |
3238 | goto out; | |
1da177e4 | 3239 | |
2ba48ce5 | 3240 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3241 | loff_t pos, endbyte; |
fb5527e6 | 3242 | |
1af5bb49 | 3243 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3244 | /* |
fbbbad4b MW |
3245 | * If the write stopped short of completing, fall back to |
3246 | * buffered writes. Some filesystems do this for writes to | |
3247 | * holes, for example. For DAX files, a buffered write will | |
3248 | * not succeed (even if it did, DAX does not handle dirty | |
3249 | * page-cache pages correctly). | |
1da177e4 | 3250 | */ |
0b8def9d | 3251 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3252 | goto out; |
3253 | ||
0b8def9d | 3254 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3255 | /* |
3b93f911 | 3256 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3257 | * then we want to return the number of bytes which were |
3258 | * direct-written, or the error code if that was zero. Note | |
3259 | * that this differs from normal direct-io semantics, which | |
3260 | * will return -EFOO even if some bytes were written. | |
3261 | */ | |
60bb4529 | 3262 | if (unlikely(status < 0)) { |
3b93f911 | 3263 | err = status; |
fb5527e6 JM |
3264 | goto out; |
3265 | } | |
fb5527e6 JM |
3266 | /* |
3267 | * We need to ensure that the page cache pages are written to | |
3268 | * disk and invalidated to preserve the expected O_DIRECT | |
3269 | * semantics. | |
3270 | */ | |
3b93f911 | 3271 | endbyte = pos + status - 1; |
0b8def9d | 3272 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3273 | if (err == 0) { |
0b8def9d | 3274 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3275 | written += status; |
fb5527e6 | 3276 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3277 | pos >> PAGE_SHIFT, |
3278 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3279 | } else { |
3280 | /* | |
3281 | * We don't know how much we wrote, so just return | |
3282 | * the number of bytes which were direct-written | |
3283 | */ | |
3284 | } | |
3285 | } else { | |
0b8def9d AV |
3286 | written = generic_perform_write(file, from, iocb->ki_pos); |
3287 | if (likely(written > 0)) | |
3288 | iocb->ki_pos += written; | |
fb5527e6 | 3289 | } |
1da177e4 LT |
3290 | out: |
3291 | current->backing_dev_info = NULL; | |
3292 | return written ? written : err; | |
3293 | } | |
8174202b | 3294 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3295 | |
e4dd9de3 | 3296 | /** |
8174202b | 3297 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3298 | * @iocb: IO state structure |
8174202b | 3299 | * @from: iov_iter with data to write |
e4dd9de3 | 3300 | * |
8174202b | 3301 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3302 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3303 | * and acquires i_mutex as needed. | |
3304 | */ | |
8174202b | 3305 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3306 | { |
3307 | struct file *file = iocb->ki_filp; | |
148f948b | 3308 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3309 | ssize_t ret; |
1da177e4 | 3310 | |
5955102c | 3311 | inode_lock(inode); |
3309dd04 AV |
3312 | ret = generic_write_checks(iocb, from); |
3313 | if (ret > 0) | |
5f380c7f | 3314 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3315 | inode_unlock(inode); |
1da177e4 | 3316 | |
e2592217 CH |
3317 | if (ret > 0) |
3318 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3319 | return ret; |
3320 | } | |
8174202b | 3321 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3322 | |
cf9a2ae8 DH |
3323 | /** |
3324 | * try_to_release_page() - release old fs-specific metadata on a page | |
3325 | * | |
3326 | * @page: the page which the kernel is trying to free | |
3327 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3328 | * | |
3329 | * The address_space is to try to release any data against the page | |
0e056eb5 | 3330 | * (presumably at page->private). If the release was successful, return '1'. |
cf9a2ae8 DH |
3331 | * Otherwise return zero. |
3332 | * | |
266cf658 DH |
3333 | * This may also be called if PG_fscache is set on a page, indicating that the |
3334 | * page is known to the local caching routines. | |
3335 | * | |
cf9a2ae8 | 3336 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3337 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3338 | * |
cf9a2ae8 DH |
3339 | */ |
3340 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3341 | { | |
3342 | struct address_space * const mapping = page->mapping; | |
3343 | ||
3344 | BUG_ON(!PageLocked(page)); | |
3345 | if (PageWriteback(page)) | |
3346 | return 0; | |
3347 | ||
3348 | if (mapping && mapping->a_ops->releasepage) | |
3349 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3350 | return try_to_free_buffers(page); | |
3351 | } | |
3352 | ||
3353 | EXPORT_SYMBOL(try_to_release_page); |