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> |
c22ce143 | 16 | #include <linux/uaccess.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
44110fe3 | 32 | #include <linux/cpuset.h> |
2f718ffc | 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 34 | #include <linux/hugetlb.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
c515e1fd | 36 | #include <linux/cleancache.h> |
f1820361 | 37 | #include <linux/rmap.h> |
0f8053a5 NP |
38 | #include "internal.h" |
39 | ||
fe0bfaaf RJ |
40 | #define CREATE_TRACE_POINTS |
41 | #include <trace/events/filemap.h> | |
42 | ||
1da177e4 | 43 | /* |
1da177e4 LT |
44 | * FIXME: remove all knowledge of the buffer layer from the core VM |
45 | */ | |
148f948b | 46 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 47 | |
1da177e4 LT |
48 | #include <asm/mman.h> |
49 | ||
50 | /* | |
51 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
52 | * though. | |
53 | * | |
54 | * Shared mappings now work. 15.8.1995 Bruno. | |
55 | * | |
56 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
57 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
58 | * | |
59 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
60 | */ | |
61 | ||
62 | /* | |
63 | * Lock ordering: | |
64 | * | |
c8c06efa | 65 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 66 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
67 | * ->swap_lock (exclusive_swap_page, others) |
68 | * ->mapping->tree_lock | |
1da177e4 | 69 | * |
1b1dcc1b | 70 | * ->i_mutex |
c8c06efa | 71 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
72 | * |
73 | * ->mmap_sem | |
c8c06efa | 74 | * ->i_mmap_rwsem |
b8072f09 | 75 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
76 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
77 | * | |
78 | * ->mmap_sem | |
79 | * ->lock_page (access_process_vm) | |
80 | * | |
ccad2365 | 81 | * ->i_mutex (generic_perform_write) |
82591e6e | 82 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 83 | * |
f758eeab | 84 | * bdi->wb.list_lock |
a66979ab | 85 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
86 | * ->mapping->tree_lock (__sync_single_inode) |
87 | * | |
c8c06efa | 88 | * ->i_mmap_rwsem |
1da177e4 LT |
89 | * ->anon_vma.lock (vma_adjust) |
90 | * | |
91 | * ->anon_vma.lock | |
b8072f09 | 92 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 93 | * |
b8072f09 | 94 | * ->page_table_lock or pte_lock |
5d337b91 | 95 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
96 | * ->private_lock (try_to_unmap_one) |
97 | * ->tree_lock (try_to_unmap_one) | |
a52633d8 MG |
98 | * ->zone_lru_lock(zone) (follow_page->mark_page_accessed) |
99 | * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page) | |
1da177e4 LT |
100 | * ->private_lock (page_remove_rmap->set_page_dirty) |
101 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 102 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 103 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 104 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 105 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 106 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
107 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
108 | * | |
c8c06efa | 109 | * ->i_mmap_rwsem |
9a3c531d | 110 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
111 | */ |
112 | ||
91b0abe3 JW |
113 | static void page_cache_tree_delete(struct address_space *mapping, |
114 | struct page *page, void *shadow) | |
115 | { | |
449dd698 | 116 | struct radix_tree_node *node; |
83929372 | 117 | int i, nr = PageHuge(page) ? 1 : hpage_nr_pages(page); |
91b0abe3 | 118 | |
83929372 KS |
119 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
120 | VM_BUG_ON_PAGE(PageTail(page), page); | |
121 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 JW |
122 | |
123 | if (shadow) { | |
83929372 | 124 | mapping->nrexceptional += nr; |
91b0abe3 | 125 | /* |
f9fe48be | 126 | * Make sure the nrexceptional update is committed before |
91b0abe3 JW |
127 | * the nrpages update so that final truncate racing |
128 | * with reclaim does not see both counters 0 at the | |
129 | * same time and miss a shadow entry. | |
130 | */ | |
131 | smp_wmb(); | |
449dd698 | 132 | } |
83929372 | 133 | mapping->nrpages -= nr; |
449dd698 | 134 | |
83929372 KS |
135 | for (i = 0; i < nr; i++) { |
136 | node = radix_tree_replace_clear_tags(&mapping->page_tree, | |
137 | page->index + i, shadow); | |
138 | if (!node) { | |
139 | VM_BUG_ON_PAGE(nr != 1, page); | |
449dd698 | 140 | return; |
83929372 | 141 | } |
449dd698 | 142 | |
83929372 KS |
143 | workingset_node_pages_dec(node); |
144 | if (shadow) | |
145 | workingset_node_shadows_inc(node); | |
146 | else | |
147 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
148 | continue; | |
149 | ||
150 | /* | |
151 | * Track node that only contains shadow entries. DAX mappings | |
152 | * contain no shadow entries and may contain other exceptional | |
153 | * entries so skip those. | |
154 | * | |
155 | * Avoid acquiring the list_lru lock if already tracked. | |
156 | * The list_empty() test is safe as node->private_list is | |
157 | * protected by mapping->tree_lock. | |
158 | */ | |
159 | if (!dax_mapping(mapping) && !workingset_node_pages(node) && | |
160 | list_empty(&node->private_list)) { | |
161 | node->private_data = mapping; | |
162 | list_lru_add(&workingset_shadow_nodes, | |
163 | &node->private_list); | |
164 | } | |
449dd698 | 165 | } |
91b0abe3 JW |
166 | } |
167 | ||
1da177e4 | 168 | /* |
e64a782f | 169 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 170 | * sure the page is locked and that nobody else uses it - or that usage |
fdf1cdb9 | 171 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 172 | */ |
62cccb8c | 173 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
174 | { |
175 | struct address_space *mapping = page->mapping; | |
83929372 | 176 | int nr = hpage_nr_pages(page); |
1da177e4 | 177 | |
fe0bfaaf | 178 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
179 | /* |
180 | * if we're uptodate, flush out into the cleancache, otherwise | |
181 | * invalidate any existing cleancache entries. We can't leave | |
182 | * stale data around in the cleancache once our page is gone | |
183 | */ | |
184 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
185 | cleancache_put_page(page); | |
186 | else | |
3167760f | 187 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 188 | |
83929372 | 189 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
190 | VM_BUG_ON_PAGE(page_mapped(page), page); |
191 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
192 | int mapcount; | |
193 | ||
194 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
195 | current->comm, page_to_pfn(page)); | |
196 | dump_page(page, "still mapped when deleted"); | |
197 | dump_stack(); | |
198 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
199 | ||
200 | mapcount = page_mapcount(page); | |
201 | if (mapping_exiting(mapping) && | |
202 | page_count(page) >= mapcount + 2) { | |
203 | /* | |
204 | * All vmas have already been torn down, so it's | |
205 | * a good bet that actually the page is unmapped, | |
206 | * and we'd prefer not to leak it: if we're wrong, | |
207 | * some other bad page check should catch it later. | |
208 | */ | |
209 | page_mapcount_reset(page); | |
6d061f9f | 210 | page_ref_sub(page, mapcount); |
06b241f3 HD |
211 | } |
212 | } | |
213 | ||
91b0abe3 JW |
214 | page_cache_tree_delete(mapping, page, shadow); |
215 | ||
1da177e4 | 216 | page->mapping = NULL; |
b85e0eff | 217 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 218 | |
4165b9b4 MH |
219 | /* hugetlb pages do not participate in page cache accounting. */ |
220 | if (!PageHuge(page)) | |
11fb9989 | 221 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); |
800d8c63 | 222 | if (PageSwapBacked(page)) { |
11fb9989 | 223 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); |
800d8c63 | 224 | if (PageTransHuge(page)) |
11fb9989 | 225 | __dec_node_page_state(page, NR_SHMEM_THPS); |
800d8c63 KS |
226 | } else { |
227 | VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page); | |
228 | } | |
3a692790 LT |
229 | |
230 | /* | |
b9ea2515 KK |
231 | * At this point page must be either written or cleaned by truncate. |
232 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 233 | * |
b9ea2515 KK |
234 | * This fixes dirty accounting after removing the page entirely but |
235 | * leaves PageDirty set: it has no effect for truncated page and | |
236 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 237 | */ |
b9ea2515 | 238 | if (WARN_ON_ONCE(PageDirty(page))) |
62cccb8c | 239 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); |
1da177e4 LT |
240 | } |
241 | ||
702cfbf9 MK |
242 | /** |
243 | * delete_from_page_cache - delete page from page cache | |
244 | * @page: the page which the kernel is trying to remove from page cache | |
245 | * | |
246 | * This must be called only on pages that have been verified to be in the page | |
247 | * cache and locked. It will never put the page into the free list, the caller | |
248 | * has a reference on the page. | |
249 | */ | |
250 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 251 | { |
83929372 | 252 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 253 | unsigned long flags; |
6072d13c | 254 | void (*freepage)(struct page *); |
1da177e4 | 255 | |
cd7619d6 | 256 | BUG_ON(!PageLocked(page)); |
1da177e4 | 257 | |
6072d13c | 258 | freepage = mapping->a_ops->freepage; |
c4843a75 | 259 | |
c4843a75 | 260 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 261 | __delete_from_page_cache(page, NULL); |
c4843a75 | 262 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
263 | |
264 | if (freepage) | |
265 | freepage(page); | |
83929372 KS |
266 | |
267 | if (PageTransHuge(page) && !PageHuge(page)) { | |
268 | page_ref_sub(page, HPAGE_PMD_NR); | |
269 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
270 | } else { | |
271 | put_page(page); | |
272 | } | |
97cecb5a MK |
273 | } |
274 | EXPORT_SYMBOL(delete_from_page_cache); | |
275 | ||
865ffef3 DM |
276 | static int filemap_check_errors(struct address_space *mapping) |
277 | { | |
278 | int ret = 0; | |
279 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
280 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
281 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 282 | ret = -ENOSPC; |
7fcbbaf1 JA |
283 | if (test_bit(AS_EIO, &mapping->flags) && |
284 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
285 | ret = -EIO; |
286 | return ret; | |
287 | } | |
288 | ||
1da177e4 | 289 | /** |
485bb99b | 290 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
291 | * @mapping: address space structure to write |
292 | * @start: offset in bytes where the range starts | |
469eb4d0 | 293 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 294 | * @sync_mode: enable synchronous operation |
1da177e4 | 295 | * |
485bb99b RD |
296 | * Start writeback against all of a mapping's dirty pages that lie |
297 | * within the byte offsets <start, end> inclusive. | |
298 | * | |
1da177e4 | 299 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 300 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
301 | * these two operations is that if a dirty page/buffer is encountered, it must |
302 | * be waited upon, and not just skipped over. | |
303 | */ | |
ebcf28e1 AM |
304 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
305 | loff_t end, int sync_mode) | |
1da177e4 LT |
306 | { |
307 | int ret; | |
308 | struct writeback_control wbc = { | |
309 | .sync_mode = sync_mode, | |
05fe478d | 310 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
311 | .range_start = start, |
312 | .range_end = end, | |
1da177e4 LT |
313 | }; |
314 | ||
315 | if (!mapping_cap_writeback_dirty(mapping)) | |
316 | return 0; | |
317 | ||
b16b1deb | 318 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 319 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 320 | wbc_detach_inode(&wbc); |
1da177e4 LT |
321 | return ret; |
322 | } | |
323 | ||
324 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
325 | int sync_mode) | |
326 | { | |
111ebb6e | 327 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
328 | } |
329 | ||
330 | int filemap_fdatawrite(struct address_space *mapping) | |
331 | { | |
332 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
333 | } | |
334 | EXPORT_SYMBOL(filemap_fdatawrite); | |
335 | ||
f4c0a0fd | 336 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 337 | loff_t end) |
1da177e4 LT |
338 | { |
339 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
340 | } | |
f4c0a0fd | 341 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 342 | |
485bb99b RD |
343 | /** |
344 | * filemap_flush - mostly a non-blocking flush | |
345 | * @mapping: target address_space | |
346 | * | |
1da177e4 LT |
347 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
348 | * purposes - I/O may not be started against all dirty pages. | |
349 | */ | |
350 | int filemap_flush(struct address_space *mapping) | |
351 | { | |
352 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
353 | } | |
354 | EXPORT_SYMBOL(filemap_flush); | |
355 | ||
aa750fd7 JN |
356 | static int __filemap_fdatawait_range(struct address_space *mapping, |
357 | loff_t start_byte, loff_t end_byte) | |
1da177e4 | 358 | { |
09cbfeaf KS |
359 | pgoff_t index = start_byte >> PAGE_SHIFT; |
360 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
361 | struct pagevec pvec; |
362 | int nr_pages; | |
aa750fd7 | 363 | int ret = 0; |
1da177e4 | 364 | |
94004ed7 | 365 | if (end_byte < start_byte) |
865ffef3 | 366 | goto out; |
1da177e4 LT |
367 | |
368 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
369 | while ((index <= end) && |
370 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
371 | PAGECACHE_TAG_WRITEBACK, | |
372 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
373 | unsigned i; | |
374 | ||
375 | for (i = 0; i < nr_pages; i++) { | |
376 | struct page *page = pvec.pages[i]; | |
377 | ||
378 | /* until radix tree lookup accepts end_index */ | |
379 | if (page->index > end) | |
380 | continue; | |
381 | ||
382 | wait_on_page_writeback(page); | |
212260aa | 383 | if (TestClearPageError(page)) |
1da177e4 LT |
384 | ret = -EIO; |
385 | } | |
386 | pagevec_release(&pvec); | |
387 | cond_resched(); | |
388 | } | |
865ffef3 | 389 | out: |
aa750fd7 JN |
390 | return ret; |
391 | } | |
392 | ||
393 | /** | |
394 | * filemap_fdatawait_range - wait for writeback to complete | |
395 | * @mapping: address space structure to wait for | |
396 | * @start_byte: offset in bytes where the range starts | |
397 | * @end_byte: offset in bytes where the range ends (inclusive) | |
398 | * | |
399 | * Walk the list of under-writeback pages of the given address space | |
400 | * in the given range and wait for all of them. Check error status of | |
401 | * the address space and return it. | |
402 | * | |
403 | * Since the error status of the address space is cleared by this function, | |
404 | * callers are responsible for checking the return value and handling and/or | |
405 | * reporting the error. | |
406 | */ | |
407 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
408 | loff_t end_byte) | |
409 | { | |
410 | int ret, ret2; | |
411 | ||
412 | ret = __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
865ffef3 DM |
413 | ret2 = filemap_check_errors(mapping); |
414 | if (!ret) | |
415 | ret = ret2; | |
1da177e4 LT |
416 | |
417 | return ret; | |
418 | } | |
d3bccb6f JK |
419 | EXPORT_SYMBOL(filemap_fdatawait_range); |
420 | ||
aa750fd7 JN |
421 | /** |
422 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
423 | * @mapping: address space structure to wait for | |
424 | * | |
425 | * Walk the list of under-writeback pages of the given address space | |
426 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
427 | * does not clear error status of the address space. | |
428 | * | |
429 | * Use this function if callers don't handle errors themselves. Expected | |
430 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
431 | * fsfreeze(8) | |
432 | */ | |
433 | void filemap_fdatawait_keep_errors(struct address_space *mapping) | |
434 | { | |
435 | loff_t i_size = i_size_read(mapping->host); | |
436 | ||
437 | if (i_size == 0) | |
438 | return; | |
439 | ||
440 | __filemap_fdatawait_range(mapping, 0, i_size - 1); | |
441 | } | |
442 | ||
1da177e4 | 443 | /** |
485bb99b | 444 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 445 | * @mapping: address space structure to wait for |
485bb99b RD |
446 | * |
447 | * Walk the list of under-writeback pages of the given address space | |
aa750fd7 JN |
448 | * and wait for all of them. Check error status of the address space |
449 | * and return it. | |
450 | * | |
451 | * Since the error status of the address space is cleared by this function, | |
452 | * callers are responsible for checking the return value and handling and/or | |
453 | * reporting the error. | |
1da177e4 LT |
454 | */ |
455 | int filemap_fdatawait(struct address_space *mapping) | |
456 | { | |
457 | loff_t i_size = i_size_read(mapping->host); | |
458 | ||
459 | if (i_size == 0) | |
460 | return 0; | |
461 | ||
94004ed7 | 462 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
463 | } |
464 | EXPORT_SYMBOL(filemap_fdatawait); | |
465 | ||
466 | int filemap_write_and_wait(struct address_space *mapping) | |
467 | { | |
28fd1298 | 468 | int err = 0; |
1da177e4 | 469 | |
7f6d5b52 RZ |
470 | if ((!dax_mapping(mapping) && mapping->nrpages) || |
471 | (dax_mapping(mapping) && mapping->nrexceptional)) { | |
28fd1298 OH |
472 | err = filemap_fdatawrite(mapping); |
473 | /* | |
474 | * Even if the above returned error, the pages may be | |
475 | * written partially (e.g. -ENOSPC), so we wait for it. | |
476 | * But the -EIO is special case, it may indicate the worst | |
477 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
478 | */ | |
479 | if (err != -EIO) { | |
480 | int err2 = filemap_fdatawait(mapping); | |
481 | if (!err) | |
482 | err = err2; | |
483 | } | |
865ffef3 DM |
484 | } else { |
485 | err = filemap_check_errors(mapping); | |
1da177e4 | 486 | } |
28fd1298 | 487 | return err; |
1da177e4 | 488 | } |
28fd1298 | 489 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 490 | |
485bb99b RD |
491 | /** |
492 | * filemap_write_and_wait_range - write out & wait on a file range | |
493 | * @mapping: the address_space for the pages | |
494 | * @lstart: offset in bytes where the range starts | |
495 | * @lend: offset in bytes where the range ends (inclusive) | |
496 | * | |
469eb4d0 AM |
497 | * Write out and wait upon file offsets lstart->lend, inclusive. |
498 | * | |
499 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
500 | * that this function can be used to write to the very end-of-file (end = -1). | |
501 | */ | |
1da177e4 LT |
502 | int filemap_write_and_wait_range(struct address_space *mapping, |
503 | loff_t lstart, loff_t lend) | |
504 | { | |
28fd1298 | 505 | int err = 0; |
1da177e4 | 506 | |
7f6d5b52 RZ |
507 | if ((!dax_mapping(mapping) && mapping->nrpages) || |
508 | (dax_mapping(mapping) && mapping->nrexceptional)) { | |
28fd1298 OH |
509 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
510 | WB_SYNC_ALL); | |
511 | /* See comment of filemap_write_and_wait() */ | |
512 | if (err != -EIO) { | |
94004ed7 CH |
513 | int err2 = filemap_fdatawait_range(mapping, |
514 | lstart, lend); | |
28fd1298 OH |
515 | if (!err) |
516 | err = err2; | |
517 | } | |
865ffef3 DM |
518 | } else { |
519 | err = filemap_check_errors(mapping); | |
1da177e4 | 520 | } |
28fd1298 | 521 | return err; |
1da177e4 | 522 | } |
f6995585 | 523 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 524 | |
ef6a3c63 MS |
525 | /** |
526 | * replace_page_cache_page - replace a pagecache page with a new one | |
527 | * @old: page to be replaced | |
528 | * @new: page to replace with | |
529 | * @gfp_mask: allocation mode | |
530 | * | |
531 | * This function replaces a page in the pagecache with a new one. On | |
532 | * success it acquires the pagecache reference for the new page and | |
533 | * drops it for the old page. Both the old and new pages must be | |
534 | * locked. This function does not add the new page to the LRU, the | |
535 | * caller must do that. | |
536 | * | |
537 | * The remove + add is atomic. The only way this function can fail is | |
538 | * memory allocation failure. | |
539 | */ | |
540 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
541 | { | |
542 | int error; | |
ef6a3c63 | 543 | |
309381fe SL |
544 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
545 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
546 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 547 | |
ef6a3c63 MS |
548 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
549 | if (!error) { | |
550 | struct address_space *mapping = old->mapping; | |
551 | void (*freepage)(struct page *); | |
c4843a75 | 552 | unsigned long flags; |
ef6a3c63 MS |
553 | |
554 | pgoff_t offset = old->index; | |
555 | freepage = mapping->a_ops->freepage; | |
556 | ||
09cbfeaf | 557 | get_page(new); |
ef6a3c63 MS |
558 | new->mapping = mapping; |
559 | new->index = offset; | |
560 | ||
c4843a75 | 561 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 562 | __delete_from_page_cache(old, NULL); |
ef6a3c63 MS |
563 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
564 | BUG_ON(error); | |
565 | mapping->nrpages++; | |
4165b9b4 MH |
566 | |
567 | /* | |
568 | * hugetlb pages do not participate in page cache accounting. | |
569 | */ | |
570 | if (!PageHuge(new)) | |
11fb9989 | 571 | __inc_node_page_state(new, NR_FILE_PAGES); |
ef6a3c63 | 572 | if (PageSwapBacked(new)) |
11fb9989 | 573 | __inc_node_page_state(new, NR_SHMEM); |
c4843a75 | 574 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6a93ca8f | 575 | mem_cgroup_migrate(old, new); |
ef6a3c63 MS |
576 | radix_tree_preload_end(); |
577 | if (freepage) | |
578 | freepage(old); | |
09cbfeaf | 579 | put_page(old); |
ef6a3c63 MS |
580 | } |
581 | ||
582 | return error; | |
583 | } | |
584 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
585 | ||
0cd6144a | 586 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 587 | struct page *page, void **shadowp) |
0cd6144a | 588 | { |
449dd698 | 589 | struct radix_tree_node *node; |
0cd6144a JW |
590 | void **slot; |
591 | int error; | |
592 | ||
e6145236 | 593 | error = __radix_tree_create(&mapping->page_tree, page->index, 0, |
449dd698 JW |
594 | &node, &slot); |
595 | if (error) | |
596 | return error; | |
597 | if (*slot) { | |
0cd6144a JW |
598 | void *p; |
599 | ||
600 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
601 | if (!radix_tree_exceptional_entry(p)) | |
602 | return -EEXIST; | |
f9fe48be | 603 | |
f9fe48be | 604 | mapping->nrexceptional--; |
4f622938 JK |
605 | if (!dax_mapping(mapping)) { |
606 | if (shadowp) | |
607 | *shadowp = p; | |
608 | if (node) | |
609 | workingset_node_shadows_dec(node); | |
610 | } else { | |
611 | /* DAX can replace empty locked entry with a hole */ | |
612 | WARN_ON_ONCE(p != | |
613 | (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY | | |
614 | RADIX_DAX_ENTRY_LOCK)); | |
615 | /* DAX accounts exceptional entries as normal pages */ | |
616 | if (node) | |
617 | workingset_node_pages_dec(node); | |
ac401cc7 JK |
618 | /* Wakeup waiters for exceptional entry lock */ |
619 | dax_wake_mapping_entry_waiter(mapping, page->index, | |
620 | false); | |
4f622938 | 621 | } |
0cd6144a | 622 | } |
449dd698 JW |
623 | radix_tree_replace_slot(slot, page); |
624 | mapping->nrpages++; | |
625 | if (node) { | |
626 | workingset_node_pages_inc(node); | |
627 | /* | |
628 | * Don't track node that contains actual pages. | |
629 | * | |
630 | * Avoid acquiring the list_lru lock if already | |
631 | * untracked. The list_empty() test is safe as | |
632 | * node->private_list is protected by | |
633 | * mapping->tree_lock. | |
634 | */ | |
635 | if (!list_empty(&node->private_list)) | |
636 | list_lru_del(&workingset_shadow_nodes, | |
637 | &node->private_list); | |
638 | } | |
639 | return 0; | |
0cd6144a JW |
640 | } |
641 | ||
a528910e JW |
642 | static int __add_to_page_cache_locked(struct page *page, |
643 | struct address_space *mapping, | |
644 | pgoff_t offset, gfp_t gfp_mask, | |
645 | void **shadowp) | |
1da177e4 | 646 | { |
00501b53 JW |
647 | int huge = PageHuge(page); |
648 | struct mem_cgroup *memcg; | |
e286781d NP |
649 | int error; |
650 | ||
309381fe SL |
651 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
652 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 653 | |
00501b53 JW |
654 | if (!huge) { |
655 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 656 | gfp_mask, &memcg, false); |
00501b53 JW |
657 | if (error) |
658 | return error; | |
659 | } | |
1da177e4 | 660 | |
5e4c0d97 | 661 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 662 | if (error) { |
00501b53 | 663 | if (!huge) |
f627c2f5 | 664 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
665 | return error; |
666 | } | |
667 | ||
09cbfeaf | 668 | get_page(page); |
66a0c8ee KS |
669 | page->mapping = mapping; |
670 | page->index = offset; | |
671 | ||
672 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 673 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
674 | radix_tree_preload_end(); |
675 | if (unlikely(error)) | |
676 | goto err_insert; | |
4165b9b4 MH |
677 | |
678 | /* hugetlb pages do not participate in page cache accounting. */ | |
679 | if (!huge) | |
11fb9989 | 680 | __inc_node_page_state(page, NR_FILE_PAGES); |
66a0c8ee | 681 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 682 | if (!huge) |
f627c2f5 | 683 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
684 | trace_mm_filemap_add_to_page_cache(page); |
685 | return 0; | |
686 | err_insert: | |
687 | page->mapping = NULL; | |
688 | /* Leave page->index set: truncation relies upon it */ | |
689 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 690 | if (!huge) |
f627c2f5 | 691 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 692 | put_page(page); |
1da177e4 LT |
693 | return error; |
694 | } | |
a528910e JW |
695 | |
696 | /** | |
697 | * add_to_page_cache_locked - add a locked page to the pagecache | |
698 | * @page: page to add | |
699 | * @mapping: the page's address_space | |
700 | * @offset: page index | |
701 | * @gfp_mask: page allocation mode | |
702 | * | |
703 | * This function is used to add a page to the pagecache. It must be locked. | |
704 | * This function does not add the page to the LRU. The caller must do that. | |
705 | */ | |
706 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
707 | pgoff_t offset, gfp_t gfp_mask) | |
708 | { | |
709 | return __add_to_page_cache_locked(page, mapping, offset, | |
710 | gfp_mask, NULL); | |
711 | } | |
e286781d | 712 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
713 | |
714 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 715 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 716 | { |
a528910e | 717 | void *shadow = NULL; |
4f98a2fe RR |
718 | int ret; |
719 | ||
48c935ad | 720 | __SetPageLocked(page); |
a528910e JW |
721 | ret = __add_to_page_cache_locked(page, mapping, offset, |
722 | gfp_mask, &shadow); | |
723 | if (unlikely(ret)) | |
48c935ad | 724 | __ClearPageLocked(page); |
a528910e JW |
725 | else { |
726 | /* | |
727 | * The page might have been evicted from cache only | |
728 | * recently, in which case it should be activated like | |
729 | * any other repeatedly accessed page. | |
f0281a00 RR |
730 | * The exception is pages getting rewritten; evicting other |
731 | * data from the working set, only to cache data that will | |
732 | * get overwritten with something else, is a waste of memory. | |
a528910e | 733 | */ |
f0281a00 RR |
734 | if (!(gfp_mask & __GFP_WRITE) && |
735 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
736 | SetPageActive(page); |
737 | workingset_activation(page); | |
738 | } else | |
739 | ClearPageActive(page); | |
740 | lru_cache_add(page); | |
741 | } | |
1da177e4 LT |
742 | return ret; |
743 | } | |
18bc0bbd | 744 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 745 | |
44110fe3 | 746 | #ifdef CONFIG_NUMA |
2ae88149 | 747 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 748 | { |
c0ff7453 MX |
749 | int n; |
750 | struct page *page; | |
751 | ||
44110fe3 | 752 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
753 | unsigned int cpuset_mems_cookie; |
754 | do { | |
d26914d1 | 755 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 756 | n = cpuset_mem_spread_node(); |
96db800f | 757 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 758 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 759 | |
c0ff7453 | 760 | return page; |
44110fe3 | 761 | } |
2ae88149 | 762 | return alloc_pages(gfp, 0); |
44110fe3 | 763 | } |
2ae88149 | 764 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
765 | #endif |
766 | ||
1da177e4 LT |
767 | /* |
768 | * In order to wait for pages to become available there must be | |
769 | * waitqueues associated with pages. By using a hash table of | |
770 | * waitqueues where the bucket discipline is to maintain all | |
771 | * waiters on the same queue and wake all when any of the pages | |
772 | * become available, and for the woken contexts to check to be | |
773 | * sure the appropriate page became available, this saves space | |
774 | * at a cost of "thundering herd" phenomena during rare hash | |
775 | * collisions. | |
776 | */ | |
a4796e37 | 777 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 LT |
778 | { |
779 | const struct zone *zone = page_zone(page); | |
780 | ||
781 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
782 | } | |
a4796e37 | 783 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 784 | |
920c7a5d | 785 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
786 | { |
787 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
788 | ||
789 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 790 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
791 | TASK_UNINTERRUPTIBLE); |
792 | } | |
793 | EXPORT_SYMBOL(wait_on_page_bit); | |
794 | ||
f62e00cc KM |
795 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
796 | { | |
797 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
798 | ||
799 | if (!test_bit(bit_nr, &page->flags)) | |
800 | return 0; | |
801 | ||
802 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 803 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
804 | } |
805 | ||
cbbce822 N |
806 | int wait_on_page_bit_killable_timeout(struct page *page, |
807 | int bit_nr, unsigned long timeout) | |
808 | { | |
809 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
810 | ||
811 | wait.key.timeout = jiffies + timeout; | |
812 | if (!test_bit(bit_nr, &page->flags)) | |
813 | return 0; | |
814 | return __wait_on_bit(page_waitqueue(page), &wait, | |
815 | bit_wait_io_timeout, TASK_KILLABLE); | |
816 | } | |
817 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
818 | ||
385e1ca5 DH |
819 | /** |
820 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
821 | * @page: Page defining the wait queue of interest |
822 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
823 | * |
824 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
825 | */ | |
826 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
827 | { | |
828 | wait_queue_head_t *q = page_waitqueue(page); | |
829 | unsigned long flags; | |
830 | ||
831 | spin_lock_irqsave(&q->lock, flags); | |
832 | __add_wait_queue(q, waiter); | |
833 | spin_unlock_irqrestore(&q->lock, flags); | |
834 | } | |
835 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
836 | ||
1da177e4 | 837 | /** |
485bb99b | 838 | * unlock_page - unlock a locked page |
1da177e4 LT |
839 | * @page: the page |
840 | * | |
841 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
842 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 843 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
844 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
845 | * | |
8413ac9d NP |
846 | * The mb is necessary to enforce ordering between the clear_bit and the read |
847 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 848 | */ |
920c7a5d | 849 | void unlock_page(struct page *page) |
1da177e4 | 850 | { |
48c935ad | 851 | page = compound_head(page); |
309381fe | 852 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 853 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 854 | smp_mb__after_atomic(); |
1da177e4 LT |
855 | wake_up_page(page, PG_locked); |
856 | } | |
857 | EXPORT_SYMBOL(unlock_page); | |
858 | ||
485bb99b RD |
859 | /** |
860 | * end_page_writeback - end writeback against a page | |
861 | * @page: the page | |
1da177e4 LT |
862 | */ |
863 | void end_page_writeback(struct page *page) | |
864 | { | |
888cf2db MG |
865 | /* |
866 | * TestClearPageReclaim could be used here but it is an atomic | |
867 | * operation and overkill in this particular case. Failing to | |
868 | * shuffle a page marked for immediate reclaim is too mild to | |
869 | * justify taking an atomic operation penalty at the end of | |
870 | * ever page writeback. | |
871 | */ | |
872 | if (PageReclaim(page)) { | |
873 | ClearPageReclaim(page); | |
ac6aadb2 | 874 | rotate_reclaimable_page(page); |
888cf2db | 875 | } |
ac6aadb2 MS |
876 | |
877 | if (!test_clear_page_writeback(page)) | |
878 | BUG(); | |
879 | ||
4e857c58 | 880 | smp_mb__after_atomic(); |
1da177e4 LT |
881 | wake_up_page(page, PG_writeback); |
882 | } | |
883 | EXPORT_SYMBOL(end_page_writeback); | |
884 | ||
57d99845 MW |
885 | /* |
886 | * After completing I/O on a page, call this routine to update the page | |
887 | * flags appropriately | |
888 | */ | |
889 | void page_endio(struct page *page, int rw, int err) | |
890 | { | |
891 | if (rw == READ) { | |
892 | if (!err) { | |
893 | SetPageUptodate(page); | |
894 | } else { | |
895 | ClearPageUptodate(page); | |
896 | SetPageError(page); | |
897 | } | |
898 | unlock_page(page); | |
899 | } else { /* rw == WRITE */ | |
900 | if (err) { | |
901 | SetPageError(page); | |
902 | if (page->mapping) | |
903 | mapping_set_error(page->mapping, err); | |
904 | } | |
905 | end_page_writeback(page); | |
906 | } | |
907 | } | |
908 | EXPORT_SYMBOL_GPL(page_endio); | |
909 | ||
485bb99b RD |
910 | /** |
911 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
912 | * @page: the page to lock | |
1da177e4 | 913 | */ |
920c7a5d | 914 | void __lock_page(struct page *page) |
1da177e4 | 915 | { |
48c935ad KS |
916 | struct page *page_head = compound_head(page); |
917 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
1da177e4 | 918 | |
48c935ad | 919 | __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io, |
1da177e4 LT |
920 | TASK_UNINTERRUPTIBLE); |
921 | } | |
922 | EXPORT_SYMBOL(__lock_page); | |
923 | ||
b5606c2d | 924 | int __lock_page_killable(struct page *page) |
2687a356 | 925 | { |
48c935ad KS |
926 | struct page *page_head = compound_head(page); |
927 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
2687a356 | 928 | |
48c935ad | 929 | return __wait_on_bit_lock(page_waitqueue(page_head), &wait, |
74316201 | 930 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 931 | } |
18bc0bbd | 932 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 933 | |
9a95f3cf PC |
934 | /* |
935 | * Return values: | |
936 | * 1 - page is locked; mmap_sem is still held. | |
937 | * 0 - page is not locked. | |
938 | * mmap_sem has been released (up_read()), unless flags had both | |
939 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
940 | * which case mmap_sem is still held. | |
941 | * | |
942 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
943 | * with the page locked and the mmap_sem unperturbed. | |
944 | */ | |
d065bd81 ML |
945 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
946 | unsigned int flags) | |
947 | { | |
37b23e05 KM |
948 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
949 | /* | |
950 | * CAUTION! In this case, mmap_sem is not released | |
951 | * even though return 0. | |
952 | */ | |
953 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
954 | return 0; | |
955 | ||
956 | up_read(&mm->mmap_sem); | |
957 | if (flags & FAULT_FLAG_KILLABLE) | |
958 | wait_on_page_locked_killable(page); | |
959 | else | |
318b275f | 960 | wait_on_page_locked(page); |
d065bd81 | 961 | return 0; |
37b23e05 KM |
962 | } else { |
963 | if (flags & FAULT_FLAG_KILLABLE) { | |
964 | int ret; | |
965 | ||
966 | ret = __lock_page_killable(page); | |
967 | if (ret) { | |
968 | up_read(&mm->mmap_sem); | |
969 | return 0; | |
970 | } | |
971 | } else | |
972 | __lock_page(page); | |
973 | return 1; | |
d065bd81 ML |
974 | } |
975 | } | |
976 | ||
e7b563bb JW |
977 | /** |
978 | * page_cache_next_hole - find the next hole (not-present entry) | |
979 | * @mapping: mapping | |
980 | * @index: index | |
981 | * @max_scan: maximum range to search | |
982 | * | |
983 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
984 | * lowest indexed hole. | |
985 | * | |
986 | * Returns: the index of the hole if found, otherwise returns an index | |
987 | * outside of the set specified (in which case 'return - index >= | |
988 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
989 | * be returned. | |
990 | * | |
991 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
992 | * like radix_tree_gang_lookup, this will not atomically search a | |
993 | * snapshot of the tree at a single point in time. For example, if a | |
994 | * hole is created at index 5, then subsequently a hole is created at | |
995 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
996 | * if called under rcu_read_lock. | |
997 | */ | |
998 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
999 | pgoff_t index, unsigned long max_scan) | |
1000 | { | |
1001 | unsigned long i; | |
1002 | ||
1003 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1004 | struct page *page; |
1005 | ||
1006 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1007 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1008 | break; |
1009 | index++; | |
1010 | if (index == 0) | |
1011 | break; | |
1012 | } | |
1013 | ||
1014 | return index; | |
1015 | } | |
1016 | EXPORT_SYMBOL(page_cache_next_hole); | |
1017 | ||
1018 | /** | |
1019 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
1020 | * @mapping: mapping | |
1021 | * @index: index | |
1022 | * @max_scan: maximum range to search | |
1023 | * | |
1024 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
1025 | * the first hole. | |
1026 | * | |
1027 | * Returns: the index of the hole if found, otherwise returns an index | |
1028 | * outside of the set specified (in which case 'index - return >= | |
1029 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1030 | * will be returned. | |
1031 | * | |
1032 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1033 | * like radix_tree_gang_lookup, this will not atomically search a | |
1034 | * snapshot of the tree at a single point in time. For example, if a | |
1035 | * hole is created at index 10, then subsequently a hole is created at | |
1036 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1037 | * called under rcu_read_lock. | |
1038 | */ | |
1039 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1040 | pgoff_t index, unsigned long max_scan) | |
1041 | { | |
1042 | unsigned long i; | |
1043 | ||
1044 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1045 | struct page *page; |
1046 | ||
1047 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1048 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1049 | break; |
1050 | index--; | |
1051 | if (index == ULONG_MAX) | |
1052 | break; | |
1053 | } | |
1054 | ||
1055 | return index; | |
1056 | } | |
1057 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1058 | ||
485bb99b | 1059 | /** |
0cd6144a | 1060 | * find_get_entry - find and get a page cache entry |
485bb99b | 1061 | * @mapping: the address_space to search |
0cd6144a JW |
1062 | * @offset: the page cache index |
1063 | * | |
1064 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1065 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1066 | * |
139b6a6f JW |
1067 | * If the slot holds a shadow entry of a previously evicted page, or a |
1068 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1069 | * |
1070 | * Otherwise, %NULL is returned. | |
1da177e4 | 1071 | */ |
0cd6144a | 1072 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1073 | { |
a60637c8 | 1074 | void **pagep; |
83929372 | 1075 | struct page *head, *page; |
1da177e4 | 1076 | |
a60637c8 NP |
1077 | rcu_read_lock(); |
1078 | repeat: | |
1079 | page = NULL; | |
1080 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1081 | if (pagep) { | |
1082 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1083 | if (unlikely(!page)) |
1084 | goto out; | |
a2c16d6c | 1085 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1086 | if (radix_tree_deref_retry(page)) |
1087 | goto repeat; | |
1088 | /* | |
139b6a6f JW |
1089 | * A shadow entry of a recently evicted page, |
1090 | * or a swap entry from shmem/tmpfs. Return | |
1091 | * it without attempting to raise page count. | |
8079b1c8 HD |
1092 | */ |
1093 | goto out; | |
a2c16d6c | 1094 | } |
83929372 KS |
1095 | |
1096 | head = compound_head(page); | |
1097 | if (!page_cache_get_speculative(head)) | |
1098 | goto repeat; | |
1099 | ||
1100 | /* The page was split under us? */ | |
1101 | if (compound_head(page) != head) { | |
1102 | put_page(head); | |
a60637c8 | 1103 | goto repeat; |
83929372 | 1104 | } |
a60637c8 NP |
1105 | |
1106 | /* | |
1107 | * Has the page moved? | |
1108 | * This is part of the lockless pagecache protocol. See | |
1109 | * include/linux/pagemap.h for details. | |
1110 | */ | |
1111 | if (unlikely(page != *pagep)) { | |
83929372 | 1112 | put_page(head); |
a60637c8 NP |
1113 | goto repeat; |
1114 | } | |
1115 | } | |
27d20fdd | 1116 | out: |
a60637c8 NP |
1117 | rcu_read_unlock(); |
1118 | ||
1da177e4 LT |
1119 | return page; |
1120 | } | |
0cd6144a | 1121 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1122 | |
0cd6144a JW |
1123 | /** |
1124 | * find_lock_entry - locate, pin and lock a page cache entry | |
1125 | * @mapping: the address_space to search | |
1126 | * @offset: the page cache index | |
1127 | * | |
1128 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1129 | * page cache page, it is returned locked and with an increased | |
1130 | * refcount. | |
1131 | * | |
139b6a6f JW |
1132 | * If the slot holds a shadow entry of a previously evicted page, or a |
1133 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1134 | * |
1135 | * Otherwise, %NULL is returned. | |
1136 | * | |
1137 | * find_lock_entry() may sleep. | |
1138 | */ | |
1139 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1140 | { |
1141 | struct page *page; | |
1142 | ||
1da177e4 | 1143 | repeat: |
0cd6144a | 1144 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1145 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1146 | lock_page(page); |
1147 | /* Has the page been truncated? */ | |
83929372 | 1148 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1149 | unlock_page(page); |
09cbfeaf | 1150 | put_page(page); |
a60637c8 | 1151 | goto repeat; |
1da177e4 | 1152 | } |
83929372 | 1153 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1154 | } |
1da177e4 LT |
1155 | return page; |
1156 | } | |
0cd6144a JW |
1157 | EXPORT_SYMBOL(find_lock_entry); |
1158 | ||
1159 | /** | |
2457aec6 | 1160 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1161 | * @mapping: the address_space to search |
1162 | * @offset: the page index | |
2457aec6 | 1163 | * @fgp_flags: PCG flags |
45f87de5 | 1164 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1165 | * |
2457aec6 | 1166 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1167 | * |
75325189 | 1168 | * PCG flags modify how the page is returned. |
0cd6144a | 1169 | * |
2457aec6 MG |
1170 | * FGP_ACCESSED: the page will be marked accessed |
1171 | * FGP_LOCK: Page is return locked | |
1172 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1173 | * @gfp_mask and added to the page cache and the VM's LRU |
1174 | * list. The page is returned locked and with an increased | |
1175 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1176 | * |
2457aec6 MG |
1177 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1178 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1179 | * |
2457aec6 | 1180 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1181 | */ |
2457aec6 | 1182 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1183 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1184 | { |
eb2be189 | 1185 | struct page *page; |
2457aec6 | 1186 | |
1da177e4 | 1187 | repeat: |
2457aec6 MG |
1188 | page = find_get_entry(mapping, offset); |
1189 | if (radix_tree_exceptional_entry(page)) | |
1190 | page = NULL; | |
1191 | if (!page) | |
1192 | goto no_page; | |
1193 | ||
1194 | if (fgp_flags & FGP_LOCK) { | |
1195 | if (fgp_flags & FGP_NOWAIT) { | |
1196 | if (!trylock_page(page)) { | |
09cbfeaf | 1197 | put_page(page); |
2457aec6 MG |
1198 | return NULL; |
1199 | } | |
1200 | } else { | |
1201 | lock_page(page); | |
1202 | } | |
1203 | ||
1204 | /* Has the page been truncated? */ | |
1205 | if (unlikely(page->mapping != mapping)) { | |
1206 | unlock_page(page); | |
09cbfeaf | 1207 | put_page(page); |
2457aec6 MG |
1208 | goto repeat; |
1209 | } | |
1210 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1211 | } | |
1212 | ||
1213 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1214 | mark_page_accessed(page); | |
1215 | ||
1216 | no_page: | |
1217 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1218 | int err; | |
1219 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1220 | gfp_mask |= __GFP_WRITE; |
1221 | if (fgp_flags & FGP_NOFS) | |
1222 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1223 | |
45f87de5 | 1224 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1225 | if (!page) |
1226 | return NULL; | |
2457aec6 MG |
1227 | |
1228 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1229 | fgp_flags |= FGP_LOCK; | |
1230 | ||
eb39d618 | 1231 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1232 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1233 | __SetPageReferenced(page); |
2457aec6 | 1234 | |
45f87de5 MH |
1235 | err = add_to_page_cache_lru(page, mapping, offset, |
1236 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 | 1237 | if (unlikely(err)) { |
09cbfeaf | 1238 | put_page(page); |
eb2be189 NP |
1239 | page = NULL; |
1240 | if (err == -EEXIST) | |
1241 | goto repeat; | |
1da177e4 | 1242 | } |
1da177e4 | 1243 | } |
2457aec6 | 1244 | |
1da177e4 LT |
1245 | return page; |
1246 | } | |
2457aec6 | 1247 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1248 | |
0cd6144a JW |
1249 | /** |
1250 | * find_get_entries - gang pagecache lookup | |
1251 | * @mapping: The address_space to search | |
1252 | * @start: The starting page cache index | |
1253 | * @nr_entries: The maximum number of entries | |
1254 | * @entries: Where the resulting entries are placed | |
1255 | * @indices: The cache indices corresponding to the entries in @entries | |
1256 | * | |
1257 | * find_get_entries() will search for and return a group of up to | |
1258 | * @nr_entries entries in the mapping. The entries are placed at | |
1259 | * @entries. find_get_entries() takes a reference against any actual | |
1260 | * pages it returns. | |
1261 | * | |
1262 | * The search returns a group of mapping-contiguous page cache entries | |
1263 | * with ascending indexes. There may be holes in the indices due to | |
1264 | * not-present pages. | |
1265 | * | |
139b6a6f JW |
1266 | * Any shadow entries of evicted pages, or swap entries from |
1267 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1268 | * |
1269 | * find_get_entries() returns the number of pages and shadow entries | |
1270 | * which were found. | |
1271 | */ | |
1272 | unsigned find_get_entries(struct address_space *mapping, | |
1273 | pgoff_t start, unsigned int nr_entries, | |
1274 | struct page **entries, pgoff_t *indices) | |
1275 | { | |
1276 | void **slot; | |
1277 | unsigned int ret = 0; | |
1278 | struct radix_tree_iter iter; | |
1279 | ||
1280 | if (!nr_entries) | |
1281 | return 0; | |
1282 | ||
1283 | rcu_read_lock(); | |
0cd6144a | 1284 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
83929372 | 1285 | struct page *head, *page; |
0cd6144a JW |
1286 | repeat: |
1287 | page = radix_tree_deref_slot(slot); | |
1288 | if (unlikely(!page)) | |
1289 | continue; | |
1290 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1291 | if (radix_tree_deref_retry(page)) { |
1292 | slot = radix_tree_iter_retry(&iter); | |
1293 | continue; | |
1294 | } | |
0cd6144a | 1295 | /* |
f9fe48be RZ |
1296 | * A shadow entry of a recently evicted page, a swap |
1297 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1298 | * without attempting to raise page count. | |
0cd6144a JW |
1299 | */ |
1300 | goto export; | |
1301 | } | |
83929372 KS |
1302 | |
1303 | head = compound_head(page); | |
1304 | if (!page_cache_get_speculative(head)) | |
1305 | goto repeat; | |
1306 | ||
1307 | /* The page was split under us? */ | |
1308 | if (compound_head(page) != head) { | |
1309 | put_page(head); | |
0cd6144a | 1310 | goto repeat; |
83929372 | 1311 | } |
0cd6144a JW |
1312 | |
1313 | /* Has the page moved? */ | |
1314 | if (unlikely(page != *slot)) { | |
83929372 | 1315 | put_page(head); |
0cd6144a JW |
1316 | goto repeat; |
1317 | } | |
1318 | export: | |
1319 | indices[ret] = iter.index; | |
1320 | entries[ret] = page; | |
1321 | if (++ret == nr_entries) | |
1322 | break; | |
1323 | } | |
1324 | rcu_read_unlock(); | |
1325 | return ret; | |
1326 | } | |
1327 | ||
1da177e4 LT |
1328 | /** |
1329 | * find_get_pages - gang pagecache lookup | |
1330 | * @mapping: The address_space to search | |
1331 | * @start: The starting page index | |
1332 | * @nr_pages: The maximum number of pages | |
1333 | * @pages: Where the resulting pages are placed | |
1334 | * | |
1335 | * find_get_pages() will search for and return a group of up to | |
1336 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1337 | * find_get_pages() takes a reference against the returned pages. | |
1338 | * | |
1339 | * The search returns a group of mapping-contiguous pages with ascending | |
1340 | * indexes. There may be holes in the indices due to not-present pages. | |
1341 | * | |
1342 | * find_get_pages() returns the number of pages which were found. | |
1343 | */ | |
1344 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1345 | unsigned int nr_pages, struct page **pages) | |
1346 | { | |
0fc9d104 KK |
1347 | struct radix_tree_iter iter; |
1348 | void **slot; | |
1349 | unsigned ret = 0; | |
1350 | ||
1351 | if (unlikely(!nr_pages)) | |
1352 | return 0; | |
a60637c8 NP |
1353 | |
1354 | rcu_read_lock(); | |
0fc9d104 | 1355 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
83929372 | 1356 | struct page *head, *page; |
a60637c8 | 1357 | repeat: |
0fc9d104 | 1358 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1359 | if (unlikely(!page)) |
1360 | continue; | |
9d8aa4ea | 1361 | |
a2c16d6c | 1362 | if (radix_tree_exception(page)) { |
8079b1c8 | 1363 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1364 | slot = radix_tree_iter_retry(&iter); |
1365 | continue; | |
8079b1c8 | 1366 | } |
a2c16d6c | 1367 | /* |
139b6a6f JW |
1368 | * A shadow entry of a recently evicted page, |
1369 | * or a swap entry from shmem/tmpfs. Skip | |
1370 | * over it. | |
a2c16d6c | 1371 | */ |
8079b1c8 | 1372 | continue; |
27d20fdd | 1373 | } |
a60637c8 | 1374 | |
83929372 KS |
1375 | head = compound_head(page); |
1376 | if (!page_cache_get_speculative(head)) | |
1377 | goto repeat; | |
1378 | ||
1379 | /* The page was split under us? */ | |
1380 | if (compound_head(page) != head) { | |
1381 | put_page(head); | |
a60637c8 | 1382 | goto repeat; |
83929372 | 1383 | } |
a60637c8 NP |
1384 | |
1385 | /* Has the page moved? */ | |
0fc9d104 | 1386 | if (unlikely(page != *slot)) { |
83929372 | 1387 | put_page(head); |
a60637c8 NP |
1388 | goto repeat; |
1389 | } | |
1da177e4 | 1390 | |
a60637c8 | 1391 | pages[ret] = page; |
0fc9d104 KK |
1392 | if (++ret == nr_pages) |
1393 | break; | |
a60637c8 | 1394 | } |
5b280c0c | 1395 | |
a60637c8 | 1396 | rcu_read_unlock(); |
1da177e4 LT |
1397 | return ret; |
1398 | } | |
1399 | ||
ebf43500 JA |
1400 | /** |
1401 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1402 | * @mapping: The address_space to search | |
1403 | * @index: The starting page index | |
1404 | * @nr_pages: The maximum number of pages | |
1405 | * @pages: Where the resulting pages are placed | |
1406 | * | |
1407 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1408 | * that the returned number of pages are guaranteed to be contiguous. | |
1409 | * | |
1410 | * find_get_pages_contig() returns the number of pages which were found. | |
1411 | */ | |
1412 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1413 | unsigned int nr_pages, struct page **pages) | |
1414 | { | |
0fc9d104 KK |
1415 | struct radix_tree_iter iter; |
1416 | void **slot; | |
1417 | unsigned int ret = 0; | |
1418 | ||
1419 | if (unlikely(!nr_pages)) | |
1420 | return 0; | |
a60637c8 NP |
1421 | |
1422 | rcu_read_lock(); | |
0fc9d104 | 1423 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
83929372 | 1424 | struct page *head, *page; |
a60637c8 | 1425 | repeat: |
0fc9d104 KK |
1426 | page = radix_tree_deref_slot(slot); |
1427 | /* The hole, there no reason to continue */ | |
a60637c8 | 1428 | if (unlikely(!page)) |
0fc9d104 | 1429 | break; |
9d8aa4ea | 1430 | |
a2c16d6c | 1431 | if (radix_tree_exception(page)) { |
8079b1c8 | 1432 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1433 | slot = radix_tree_iter_retry(&iter); |
1434 | continue; | |
8079b1c8 | 1435 | } |
a2c16d6c | 1436 | /* |
139b6a6f JW |
1437 | * A shadow entry of a recently evicted page, |
1438 | * or a swap entry from shmem/tmpfs. Stop | |
1439 | * looking for contiguous pages. | |
a2c16d6c | 1440 | */ |
8079b1c8 | 1441 | break; |
a2c16d6c | 1442 | } |
ebf43500 | 1443 | |
83929372 KS |
1444 | head = compound_head(page); |
1445 | if (!page_cache_get_speculative(head)) | |
1446 | goto repeat; | |
1447 | ||
1448 | /* The page was split under us? */ | |
1449 | if (compound_head(page) != head) { | |
1450 | put_page(head); | |
a60637c8 | 1451 | goto repeat; |
83929372 | 1452 | } |
a60637c8 NP |
1453 | |
1454 | /* Has the page moved? */ | |
0fc9d104 | 1455 | if (unlikely(page != *slot)) { |
83929372 | 1456 | put_page(head); |
a60637c8 NP |
1457 | goto repeat; |
1458 | } | |
1459 | ||
9cbb4cb2 NP |
1460 | /* |
1461 | * must check mapping and index after taking the ref. | |
1462 | * otherwise we can get both false positives and false | |
1463 | * negatives, which is just confusing to the caller. | |
1464 | */ | |
83929372 | 1465 | if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { |
09cbfeaf | 1466 | put_page(page); |
9cbb4cb2 NP |
1467 | break; |
1468 | } | |
1469 | ||
a60637c8 | 1470 | pages[ret] = page; |
0fc9d104 KK |
1471 | if (++ret == nr_pages) |
1472 | break; | |
ebf43500 | 1473 | } |
a60637c8 NP |
1474 | rcu_read_unlock(); |
1475 | return ret; | |
ebf43500 | 1476 | } |
ef71c15c | 1477 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1478 | |
485bb99b RD |
1479 | /** |
1480 | * find_get_pages_tag - find and return pages that match @tag | |
1481 | * @mapping: the address_space to search | |
1482 | * @index: the starting page index | |
1483 | * @tag: the tag index | |
1484 | * @nr_pages: the maximum number of pages | |
1485 | * @pages: where the resulting pages are placed | |
1486 | * | |
1da177e4 | 1487 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1488 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1489 | */ |
1490 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1491 | int tag, unsigned int nr_pages, struct page **pages) | |
1492 | { | |
0fc9d104 KK |
1493 | struct radix_tree_iter iter; |
1494 | void **slot; | |
1495 | unsigned ret = 0; | |
1496 | ||
1497 | if (unlikely(!nr_pages)) | |
1498 | return 0; | |
a60637c8 NP |
1499 | |
1500 | rcu_read_lock(); | |
0fc9d104 KK |
1501 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1502 | &iter, *index, tag) { | |
83929372 | 1503 | struct page *head, *page; |
a60637c8 | 1504 | repeat: |
0fc9d104 | 1505 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1506 | if (unlikely(!page)) |
1507 | continue; | |
9d8aa4ea | 1508 | |
a2c16d6c | 1509 | if (radix_tree_exception(page)) { |
8079b1c8 | 1510 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1511 | slot = radix_tree_iter_retry(&iter); |
1512 | continue; | |
8079b1c8 | 1513 | } |
a2c16d6c | 1514 | /* |
139b6a6f JW |
1515 | * A shadow entry of a recently evicted page. |
1516 | * | |
1517 | * Those entries should never be tagged, but | |
1518 | * this tree walk is lockless and the tags are | |
1519 | * looked up in bulk, one radix tree node at a | |
1520 | * time, so there is a sizable window for page | |
1521 | * reclaim to evict a page we saw tagged. | |
1522 | * | |
1523 | * Skip over it. | |
a2c16d6c | 1524 | */ |
139b6a6f | 1525 | continue; |
a2c16d6c | 1526 | } |
a60637c8 | 1527 | |
83929372 KS |
1528 | head = compound_head(page); |
1529 | if (!page_cache_get_speculative(head)) | |
a60637c8 NP |
1530 | goto repeat; |
1531 | ||
83929372 KS |
1532 | /* The page was split under us? */ |
1533 | if (compound_head(page) != head) { | |
1534 | put_page(head); | |
1535 | goto repeat; | |
1536 | } | |
1537 | ||
a60637c8 | 1538 | /* Has the page moved? */ |
0fc9d104 | 1539 | if (unlikely(page != *slot)) { |
83929372 | 1540 | put_page(head); |
a60637c8 NP |
1541 | goto repeat; |
1542 | } | |
1543 | ||
1544 | pages[ret] = page; | |
0fc9d104 KK |
1545 | if (++ret == nr_pages) |
1546 | break; | |
a60637c8 | 1547 | } |
5b280c0c | 1548 | |
a60637c8 | 1549 | rcu_read_unlock(); |
1da177e4 | 1550 | |
1da177e4 LT |
1551 | if (ret) |
1552 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1553 | |
1da177e4 LT |
1554 | return ret; |
1555 | } | |
ef71c15c | 1556 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1557 | |
7e7f7749 RZ |
1558 | /** |
1559 | * find_get_entries_tag - find and return entries that match @tag | |
1560 | * @mapping: the address_space to search | |
1561 | * @start: the starting page cache index | |
1562 | * @tag: the tag index | |
1563 | * @nr_entries: the maximum number of entries | |
1564 | * @entries: where the resulting entries are placed | |
1565 | * @indices: the cache indices corresponding to the entries in @entries | |
1566 | * | |
1567 | * Like find_get_entries, except we only return entries which are tagged with | |
1568 | * @tag. | |
1569 | */ | |
1570 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1571 | int tag, unsigned int nr_entries, | |
1572 | struct page **entries, pgoff_t *indices) | |
1573 | { | |
1574 | void **slot; | |
1575 | unsigned int ret = 0; | |
1576 | struct radix_tree_iter iter; | |
1577 | ||
1578 | if (!nr_entries) | |
1579 | return 0; | |
1580 | ||
1581 | rcu_read_lock(); | |
7e7f7749 RZ |
1582 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1583 | &iter, start, tag) { | |
83929372 | 1584 | struct page *head, *page; |
7e7f7749 RZ |
1585 | repeat: |
1586 | page = radix_tree_deref_slot(slot); | |
1587 | if (unlikely(!page)) | |
1588 | continue; | |
1589 | if (radix_tree_exception(page)) { | |
1590 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1591 | slot = radix_tree_iter_retry(&iter); |
1592 | continue; | |
7e7f7749 RZ |
1593 | } |
1594 | ||
1595 | /* | |
1596 | * A shadow entry of a recently evicted page, a swap | |
1597 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1598 | * without attempting to raise page count. | |
1599 | */ | |
1600 | goto export; | |
1601 | } | |
83929372 KS |
1602 | |
1603 | head = compound_head(page); | |
1604 | if (!page_cache_get_speculative(head)) | |
7e7f7749 RZ |
1605 | goto repeat; |
1606 | ||
83929372 KS |
1607 | /* The page was split under us? */ |
1608 | if (compound_head(page) != head) { | |
1609 | put_page(head); | |
1610 | goto repeat; | |
1611 | } | |
1612 | ||
7e7f7749 RZ |
1613 | /* Has the page moved? */ |
1614 | if (unlikely(page != *slot)) { | |
83929372 | 1615 | put_page(head); |
7e7f7749 RZ |
1616 | goto repeat; |
1617 | } | |
1618 | export: | |
1619 | indices[ret] = iter.index; | |
1620 | entries[ret] = page; | |
1621 | if (++ret == nr_entries) | |
1622 | break; | |
1623 | } | |
1624 | rcu_read_unlock(); | |
1625 | return ret; | |
1626 | } | |
1627 | EXPORT_SYMBOL(find_get_entries_tag); | |
1628 | ||
76d42bd9 WF |
1629 | /* |
1630 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1631 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1632 | * | |
1633 | * ---R__________________________________________B__________ | |
1634 | * ^ reading here ^ bad block(assume 4k) | |
1635 | * | |
1636 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1637 | * => failing the whole request => read(R) => read(R+1) => | |
1638 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1639 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1640 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1641 | * | |
1642 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1643 | */ | |
1644 | static void shrink_readahead_size_eio(struct file *filp, | |
1645 | struct file_ra_state *ra) | |
1646 | { | |
76d42bd9 | 1647 | ra->ra_pages /= 4; |
76d42bd9 WF |
1648 | } |
1649 | ||
485bb99b | 1650 | /** |
36e78914 | 1651 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1652 | * @filp: the file to read |
1653 | * @ppos: current file position | |
6e58e79d AV |
1654 | * @iter: data destination |
1655 | * @written: already copied | |
485bb99b | 1656 | * |
1da177e4 | 1657 | * This is a generic file read routine, and uses the |
485bb99b | 1658 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1659 | * |
1660 | * This is really ugly. But the goto's actually try to clarify some | |
1661 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1662 | */ |
6e58e79d AV |
1663 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1664 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1665 | { |
36e78914 | 1666 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1667 | struct inode *inode = mapping->host; |
36e78914 | 1668 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1669 | pgoff_t index; |
1670 | pgoff_t last_index; | |
1671 | pgoff_t prev_index; | |
1672 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1673 | unsigned int prev_offset; |
6e58e79d | 1674 | int error = 0; |
1da177e4 | 1675 | |
09cbfeaf KS |
1676 | index = *ppos >> PAGE_SHIFT; |
1677 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
1678 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
1679 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
1680 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 1681 | |
1da177e4 LT |
1682 | for (;;) { |
1683 | struct page *page; | |
57f6b96c | 1684 | pgoff_t end_index; |
a32ea1e1 | 1685 | loff_t isize; |
1da177e4 LT |
1686 | unsigned long nr, ret; |
1687 | ||
1da177e4 | 1688 | cond_resched(); |
1da177e4 LT |
1689 | find_page: |
1690 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1691 | if (!page) { |
cf914a7d | 1692 | page_cache_sync_readahead(mapping, |
7ff81078 | 1693 | ra, filp, |
3ea89ee8 FW |
1694 | index, last_index - index); |
1695 | page = find_get_page(mapping, index); | |
1696 | if (unlikely(page == NULL)) | |
1697 | goto no_cached_page; | |
1698 | } | |
1699 | if (PageReadahead(page)) { | |
cf914a7d | 1700 | page_cache_async_readahead(mapping, |
7ff81078 | 1701 | ra, filp, page, |
3ea89ee8 | 1702 | index, last_index - index); |
1da177e4 | 1703 | } |
8ab22b9a | 1704 | if (!PageUptodate(page)) { |
ebded027 MG |
1705 | /* |
1706 | * See comment in do_read_cache_page on why | |
1707 | * wait_on_page_locked is used to avoid unnecessarily | |
1708 | * serialisations and why it's safe. | |
1709 | */ | |
1710 | wait_on_page_locked_killable(page); | |
1711 | if (PageUptodate(page)) | |
1712 | goto page_ok; | |
1713 | ||
09cbfeaf | 1714 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
1715 | !mapping->a_ops->is_partially_uptodate) |
1716 | goto page_not_up_to_date; | |
529ae9aa | 1717 | if (!trylock_page(page)) |
8ab22b9a | 1718 | goto page_not_up_to_date; |
8d056cb9 DH |
1719 | /* Did it get truncated before we got the lock? */ |
1720 | if (!page->mapping) | |
1721 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1722 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1723 | offset, iter->count)) |
8ab22b9a HH |
1724 | goto page_not_up_to_date_locked; |
1725 | unlock_page(page); | |
1726 | } | |
1da177e4 | 1727 | page_ok: |
a32ea1e1 N |
1728 | /* |
1729 | * i_size must be checked after we know the page is Uptodate. | |
1730 | * | |
1731 | * Checking i_size after the check allows us to calculate | |
1732 | * the correct value for "nr", which means the zero-filled | |
1733 | * part of the page is not copied back to userspace (unless | |
1734 | * another truncate extends the file - this is desired though). | |
1735 | */ | |
1736 | ||
1737 | isize = i_size_read(inode); | |
09cbfeaf | 1738 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 1739 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 1740 | put_page(page); |
a32ea1e1 N |
1741 | goto out; |
1742 | } | |
1743 | ||
1744 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 1745 | nr = PAGE_SIZE; |
a32ea1e1 | 1746 | if (index == end_index) { |
09cbfeaf | 1747 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 1748 | if (nr <= offset) { |
09cbfeaf | 1749 | put_page(page); |
a32ea1e1 N |
1750 | goto out; |
1751 | } | |
1752 | } | |
1753 | nr = nr - offset; | |
1da177e4 LT |
1754 | |
1755 | /* If users can be writing to this page using arbitrary | |
1756 | * virtual addresses, take care about potential aliasing | |
1757 | * before reading the page on the kernel side. | |
1758 | */ | |
1759 | if (mapping_writably_mapped(mapping)) | |
1760 | flush_dcache_page(page); | |
1761 | ||
1762 | /* | |
ec0f1637 JK |
1763 | * When a sequential read accesses a page several times, |
1764 | * only mark it as accessed the first time. | |
1da177e4 | 1765 | */ |
ec0f1637 | 1766 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1767 | mark_page_accessed(page); |
1768 | prev_index = index; | |
1769 | ||
1770 | /* | |
1771 | * Ok, we have the page, and it's up-to-date, so | |
1772 | * now we can copy it to user space... | |
1da177e4 | 1773 | */ |
6e58e79d AV |
1774 | |
1775 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 1776 | offset += ret; |
09cbfeaf KS |
1777 | index += offset >> PAGE_SHIFT; |
1778 | offset &= ~PAGE_MASK; | |
6ce745ed | 1779 | prev_offset = offset; |
1da177e4 | 1780 | |
09cbfeaf | 1781 | put_page(page); |
6e58e79d AV |
1782 | written += ret; |
1783 | if (!iov_iter_count(iter)) | |
1784 | goto out; | |
1785 | if (ret < nr) { | |
1786 | error = -EFAULT; | |
1787 | goto out; | |
1788 | } | |
1789 | continue; | |
1da177e4 LT |
1790 | |
1791 | page_not_up_to_date: | |
1792 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1793 | error = lock_page_killable(page); |
1794 | if (unlikely(error)) | |
1795 | goto readpage_error; | |
1da177e4 | 1796 | |
8ab22b9a | 1797 | page_not_up_to_date_locked: |
da6052f7 | 1798 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1799 | if (!page->mapping) { |
1800 | unlock_page(page); | |
09cbfeaf | 1801 | put_page(page); |
1da177e4 LT |
1802 | continue; |
1803 | } | |
1804 | ||
1805 | /* Did somebody else fill it already? */ | |
1806 | if (PageUptodate(page)) { | |
1807 | unlock_page(page); | |
1808 | goto page_ok; | |
1809 | } | |
1810 | ||
1811 | readpage: | |
91803b49 JM |
1812 | /* |
1813 | * A previous I/O error may have been due to temporary | |
1814 | * failures, eg. multipath errors. | |
1815 | * PG_error will be set again if readpage fails. | |
1816 | */ | |
1817 | ClearPageError(page); | |
1da177e4 LT |
1818 | /* Start the actual read. The read will unlock the page. */ |
1819 | error = mapping->a_ops->readpage(filp, page); | |
1820 | ||
994fc28c ZB |
1821 | if (unlikely(error)) { |
1822 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 1823 | put_page(page); |
6e58e79d | 1824 | error = 0; |
994fc28c ZB |
1825 | goto find_page; |
1826 | } | |
1da177e4 | 1827 | goto readpage_error; |
994fc28c | 1828 | } |
1da177e4 LT |
1829 | |
1830 | if (!PageUptodate(page)) { | |
85462323 ON |
1831 | error = lock_page_killable(page); |
1832 | if (unlikely(error)) | |
1833 | goto readpage_error; | |
1da177e4 LT |
1834 | if (!PageUptodate(page)) { |
1835 | if (page->mapping == NULL) { | |
1836 | /* | |
2ecdc82e | 1837 | * invalidate_mapping_pages got it |
1da177e4 LT |
1838 | */ |
1839 | unlock_page(page); | |
09cbfeaf | 1840 | put_page(page); |
1da177e4 LT |
1841 | goto find_page; |
1842 | } | |
1843 | unlock_page(page); | |
7ff81078 | 1844 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1845 | error = -EIO; |
1846 | goto readpage_error; | |
1da177e4 LT |
1847 | } |
1848 | unlock_page(page); | |
1849 | } | |
1850 | ||
1da177e4 LT |
1851 | goto page_ok; |
1852 | ||
1853 | readpage_error: | |
1854 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 1855 | put_page(page); |
1da177e4 LT |
1856 | goto out; |
1857 | ||
1858 | no_cached_page: | |
1859 | /* | |
1860 | * Ok, it wasn't cached, so we need to create a new | |
1861 | * page.. | |
1862 | */ | |
eb2be189 NP |
1863 | page = page_cache_alloc_cold(mapping); |
1864 | if (!page) { | |
6e58e79d | 1865 | error = -ENOMEM; |
eb2be189 | 1866 | goto out; |
1da177e4 | 1867 | } |
6afdb859 | 1868 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 1869 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 1870 | if (error) { |
09cbfeaf | 1871 | put_page(page); |
6e58e79d AV |
1872 | if (error == -EEXIST) { |
1873 | error = 0; | |
1da177e4 | 1874 | goto find_page; |
6e58e79d | 1875 | } |
1da177e4 LT |
1876 | goto out; |
1877 | } | |
1da177e4 LT |
1878 | goto readpage; |
1879 | } | |
1880 | ||
1881 | out: | |
7ff81078 | 1882 | ra->prev_pos = prev_index; |
09cbfeaf | 1883 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 1884 | ra->prev_pos |= prev_offset; |
1da177e4 | 1885 | |
09cbfeaf | 1886 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 1887 | file_accessed(filp); |
6e58e79d | 1888 | return written ? written : error; |
1da177e4 LT |
1889 | } |
1890 | ||
485bb99b | 1891 | /** |
6abd2322 | 1892 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1893 | * @iocb: kernel I/O control block |
6abd2322 | 1894 | * @iter: destination for the data read |
485bb99b | 1895 | * |
6abd2322 | 1896 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1897 | * that can use the page cache directly. |
1898 | */ | |
1899 | ssize_t | |
ed978a81 | 1900 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1901 | { |
ed978a81 | 1902 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1903 | ssize_t retval = 0; |
e7080a43 NS |
1904 | size_t count = iov_iter_count(iter); |
1905 | ||
1906 | if (!count) | |
1907 | goto out; /* skip atime */ | |
1da177e4 | 1908 | |
2ba48ce5 | 1909 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1910 | struct address_space *mapping = file->f_mapping; |
1911 | struct inode *inode = mapping->host; | |
543ade1f | 1912 | loff_t size; |
1da177e4 | 1913 | |
1da177e4 | 1914 | size = i_size_read(inode); |
c64fb5c7 CH |
1915 | retval = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
1916 | iocb->ki_pos + count - 1); | |
9fe55eea | 1917 | if (!retval) { |
ed978a81 | 1918 | struct iov_iter data = *iter; |
c8b8e32d | 1919 | retval = mapping->a_ops->direct_IO(iocb, &data); |
9fe55eea | 1920 | } |
d8d3d94b | 1921 | |
9fe55eea | 1922 | if (retval > 0) { |
c64fb5c7 | 1923 | iocb->ki_pos += retval; |
ed978a81 | 1924 | iov_iter_advance(iter, retval); |
9fe55eea | 1925 | } |
66f998f6 | 1926 | |
9fe55eea SW |
1927 | /* |
1928 | * Btrfs can have a short DIO read if we encounter | |
1929 | * compressed extents, so if there was an error, or if | |
1930 | * we've already read everything we wanted to, or if | |
1931 | * there was a short read because we hit EOF, go ahead | |
1932 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1933 | * the rest of the read. Buffered reads will not work for |
1934 | * DAX files, so don't bother trying. | |
9fe55eea | 1935 | */ |
c64fb5c7 | 1936 | if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size || |
fbbbad4b | 1937 | IS_DAX(inode)) { |
ed978a81 | 1938 | file_accessed(file); |
9fe55eea | 1939 | goto out; |
0e0bcae3 | 1940 | } |
1da177e4 LT |
1941 | } |
1942 | ||
c64fb5c7 | 1943 | retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval); |
1da177e4 LT |
1944 | out: |
1945 | return retval; | |
1946 | } | |
ed978a81 | 1947 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1948 | |
1da177e4 | 1949 | #ifdef CONFIG_MMU |
485bb99b RD |
1950 | /** |
1951 | * page_cache_read - adds requested page to the page cache if not already there | |
1952 | * @file: file to read | |
1953 | * @offset: page index | |
62eb320a | 1954 | * @gfp_mask: memory allocation flags |
485bb99b | 1955 | * |
1da177e4 LT |
1956 | * This adds the requested page to the page cache if it isn't already there, |
1957 | * and schedules an I/O to read in its contents from disk. | |
1958 | */ | |
c20cd45e | 1959 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
1960 | { |
1961 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1962 | struct page *page; |
994fc28c | 1963 | int ret; |
1da177e4 | 1964 | |
994fc28c | 1965 | do { |
c20cd45e | 1966 | page = __page_cache_alloc(gfp_mask|__GFP_COLD); |
994fc28c ZB |
1967 | if (!page) |
1968 | return -ENOMEM; | |
1969 | ||
c20cd45e | 1970 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); |
994fc28c ZB |
1971 | if (ret == 0) |
1972 | ret = mapping->a_ops->readpage(file, page); | |
1973 | else if (ret == -EEXIST) | |
1974 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1975 | |
09cbfeaf | 1976 | put_page(page); |
1da177e4 | 1977 | |
994fc28c | 1978 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1979 | |
994fc28c | 1980 | return ret; |
1da177e4 LT |
1981 | } |
1982 | ||
1983 | #define MMAP_LOTSAMISS (100) | |
1984 | ||
ef00e08e LT |
1985 | /* |
1986 | * Synchronous readahead happens when we don't even find | |
1987 | * a page in the page cache at all. | |
1988 | */ | |
1989 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1990 | struct file_ra_state *ra, | |
1991 | struct file *file, | |
1992 | pgoff_t offset) | |
1993 | { | |
ef00e08e LT |
1994 | struct address_space *mapping = file->f_mapping; |
1995 | ||
1996 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1997 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1998 | return; |
275b12bf WF |
1999 | if (!ra->ra_pages) |
2000 | return; | |
ef00e08e | 2001 | |
64363aad | 2002 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
2003 | page_cache_sync_readahead(mapping, ra, file, offset, |
2004 | ra->ra_pages); | |
ef00e08e LT |
2005 | return; |
2006 | } | |
2007 | ||
207d04ba AK |
2008 | /* Avoid banging the cache line if not needed */ |
2009 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2010 | ra->mmap_miss++; |
2011 | ||
2012 | /* | |
2013 | * Do we miss much more than hit in this file? If so, | |
2014 | * stop bothering with read-ahead. It will only hurt. | |
2015 | */ | |
2016 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
2017 | return; | |
2018 | ||
d30a1100 WF |
2019 | /* |
2020 | * mmap read-around | |
2021 | */ | |
600e19af RG |
2022 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2023 | ra->size = ra->ra_pages; | |
2024 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2025 | ra_submit(ra, mapping, file); |
ef00e08e LT |
2026 | } |
2027 | ||
2028 | /* | |
2029 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2030 | * so we want to possibly extend the readahead further.. | |
2031 | */ | |
2032 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2033 | struct file_ra_state *ra, | |
2034 | struct file *file, | |
2035 | struct page *page, | |
2036 | pgoff_t offset) | |
2037 | { | |
2038 | struct address_space *mapping = file->f_mapping; | |
2039 | ||
2040 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2041 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2042 | return; |
2043 | if (ra->mmap_miss > 0) | |
2044 | ra->mmap_miss--; | |
2045 | if (PageReadahead(page)) | |
2fad6f5d WF |
2046 | page_cache_async_readahead(mapping, ra, file, |
2047 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2048 | } |
2049 | ||
485bb99b | 2050 | /** |
54cb8821 | 2051 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
2052 | * @vma: vma in which the fault was taken |
2053 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 2054 | * |
54cb8821 | 2055 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2056 | * mapped memory region to read in file data during a page fault. |
2057 | * | |
2058 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2059 | * it in the page cache, and handles the special cases reasonably without | |
2060 | * having a lot of duplicated code. | |
9a95f3cf PC |
2061 | * |
2062 | * vma->vm_mm->mmap_sem must be held on entry. | |
2063 | * | |
2064 | * If our return value has VM_FAULT_RETRY set, it's because | |
2065 | * lock_page_or_retry() returned 0. | |
2066 | * The mmap_sem has usually been released in this case. | |
2067 | * See __lock_page_or_retry() for the exception. | |
2068 | * | |
2069 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2070 | * has not been released. | |
2071 | * | |
2072 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2073 | */ |
d0217ac0 | 2074 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2075 | { |
2076 | int error; | |
54cb8821 | 2077 | struct file *file = vma->vm_file; |
1da177e4 LT |
2078 | struct address_space *mapping = file->f_mapping; |
2079 | struct file_ra_state *ra = &file->f_ra; | |
2080 | struct inode *inode = mapping->host; | |
ef00e08e | 2081 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 2082 | struct page *page; |
99e3e53f | 2083 | loff_t size; |
83c54070 | 2084 | int ret = 0; |
1da177e4 | 2085 | |
09cbfeaf KS |
2086 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2087 | if (offset >= size >> PAGE_SHIFT) | |
5307cc1a | 2088 | return VM_FAULT_SIGBUS; |
1da177e4 | 2089 | |
1da177e4 | 2090 | /* |
49426420 | 2091 | * Do we have something in the page cache already? |
1da177e4 | 2092 | */ |
ef00e08e | 2093 | page = find_get_page(mapping, offset); |
45cac65b | 2094 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2095 | /* |
ef00e08e LT |
2096 | * We found the page, so try async readahead before |
2097 | * waiting for the lock. | |
1da177e4 | 2098 | */ |
ef00e08e | 2099 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 2100 | } else if (!page) { |
ef00e08e LT |
2101 | /* No page in the page cache at all */ |
2102 | do_sync_mmap_readahead(vma, ra, file, offset); | |
2103 | count_vm_event(PGMAJFAULT); | |
456f998e | 2104 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2105 | ret = VM_FAULT_MAJOR; |
2106 | retry_find: | |
b522c94d | 2107 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2108 | if (!page) |
2109 | goto no_cached_page; | |
2110 | } | |
2111 | ||
d88c0922 | 2112 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2113 | put_page(page); |
d065bd81 | 2114 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2115 | } |
b522c94d ML |
2116 | |
2117 | /* Did it get truncated? */ | |
2118 | if (unlikely(page->mapping != mapping)) { | |
2119 | unlock_page(page); | |
2120 | put_page(page); | |
2121 | goto retry_find; | |
2122 | } | |
309381fe | 2123 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2124 | |
1da177e4 | 2125 | /* |
d00806b1 NP |
2126 | * We have a locked page in the page cache, now we need to check |
2127 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2128 | */ |
d00806b1 | 2129 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2130 | goto page_not_uptodate; |
2131 | ||
ef00e08e LT |
2132 | /* |
2133 | * Found the page and have a reference on it. | |
2134 | * We must recheck i_size under page lock. | |
2135 | */ | |
09cbfeaf KS |
2136 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2137 | if (unlikely(offset >= size >> PAGE_SHIFT)) { | |
d00806b1 | 2138 | unlock_page(page); |
09cbfeaf | 2139 | put_page(page); |
5307cc1a | 2140 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2141 | } |
2142 | ||
d0217ac0 | 2143 | vmf->page = page; |
83c54070 | 2144 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2145 | |
1da177e4 LT |
2146 | no_cached_page: |
2147 | /* | |
2148 | * We're only likely to ever get here if MADV_RANDOM is in | |
2149 | * effect. | |
2150 | */ | |
c20cd45e | 2151 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2152 | |
2153 | /* | |
2154 | * The page we want has now been added to the page cache. | |
2155 | * In the unlikely event that someone removed it in the | |
2156 | * meantime, we'll just come back here and read it again. | |
2157 | */ | |
2158 | if (error >= 0) | |
2159 | goto retry_find; | |
2160 | ||
2161 | /* | |
2162 | * An error return from page_cache_read can result if the | |
2163 | * system is low on memory, or a problem occurs while trying | |
2164 | * to schedule I/O. | |
2165 | */ | |
2166 | if (error == -ENOMEM) | |
d0217ac0 NP |
2167 | return VM_FAULT_OOM; |
2168 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2169 | |
2170 | page_not_uptodate: | |
1da177e4 LT |
2171 | /* |
2172 | * Umm, take care of errors if the page isn't up-to-date. | |
2173 | * Try to re-read it _once_. We do this synchronously, | |
2174 | * because there really aren't any performance issues here | |
2175 | * and we need to check for errors. | |
2176 | */ | |
1da177e4 | 2177 | ClearPageError(page); |
994fc28c | 2178 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2179 | if (!error) { |
2180 | wait_on_page_locked(page); | |
2181 | if (!PageUptodate(page)) | |
2182 | error = -EIO; | |
2183 | } | |
09cbfeaf | 2184 | put_page(page); |
d00806b1 NP |
2185 | |
2186 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2187 | goto retry_find; |
1da177e4 | 2188 | |
d00806b1 | 2189 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2190 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2191 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2192 | } |
2193 | EXPORT_SYMBOL(filemap_fault); | |
2194 | ||
bae473a4 KS |
2195 | void filemap_map_pages(struct fault_env *fe, |
2196 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
f1820361 KS |
2197 | { |
2198 | struct radix_tree_iter iter; | |
2199 | void **slot; | |
bae473a4 | 2200 | struct file *file = fe->vma->vm_file; |
f1820361 | 2201 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2202 | pgoff_t last_pgoff = start_pgoff; |
f1820361 | 2203 | loff_t size; |
83929372 | 2204 | struct page *head, *page; |
f1820361 KS |
2205 | |
2206 | rcu_read_lock(); | |
bae473a4 KS |
2207 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, |
2208 | start_pgoff) { | |
2209 | if (iter.index > end_pgoff) | |
f1820361 KS |
2210 | break; |
2211 | repeat: | |
2212 | page = radix_tree_deref_slot(slot); | |
2213 | if (unlikely(!page)) | |
2214 | goto next; | |
2215 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2216 | if (radix_tree_deref_retry(page)) { |
2217 | slot = radix_tree_iter_retry(&iter); | |
2218 | continue; | |
2219 | } | |
2220 | goto next; | |
f1820361 KS |
2221 | } |
2222 | ||
83929372 KS |
2223 | head = compound_head(page); |
2224 | if (!page_cache_get_speculative(head)) | |
f1820361 KS |
2225 | goto repeat; |
2226 | ||
83929372 KS |
2227 | /* The page was split under us? */ |
2228 | if (compound_head(page) != head) { | |
2229 | put_page(head); | |
2230 | goto repeat; | |
2231 | } | |
2232 | ||
f1820361 KS |
2233 | /* Has the page moved? */ |
2234 | if (unlikely(page != *slot)) { | |
83929372 | 2235 | put_page(head); |
f1820361 KS |
2236 | goto repeat; |
2237 | } | |
2238 | ||
2239 | if (!PageUptodate(page) || | |
2240 | PageReadahead(page) || | |
2241 | PageHWPoison(page)) | |
2242 | goto skip; | |
2243 | if (!trylock_page(page)) | |
2244 | goto skip; | |
2245 | ||
2246 | if (page->mapping != mapping || !PageUptodate(page)) | |
2247 | goto unlock; | |
2248 | ||
09cbfeaf KS |
2249 | size = round_up(i_size_read(mapping->host), PAGE_SIZE); |
2250 | if (page->index >= size >> PAGE_SHIFT) | |
f1820361 KS |
2251 | goto unlock; |
2252 | ||
f1820361 KS |
2253 | if (file->f_ra.mmap_miss > 0) |
2254 | file->f_ra.mmap_miss--; | |
7267ec00 KS |
2255 | |
2256 | fe->address += (iter.index - last_pgoff) << PAGE_SHIFT; | |
2257 | if (fe->pte) | |
2258 | fe->pte += iter.index - last_pgoff; | |
2259 | last_pgoff = iter.index; | |
2260 | if (alloc_set_pte(fe, NULL, page)) | |
2261 | goto unlock; | |
f1820361 KS |
2262 | unlock_page(page); |
2263 | goto next; | |
2264 | unlock: | |
2265 | unlock_page(page); | |
2266 | skip: | |
09cbfeaf | 2267 | put_page(page); |
f1820361 | 2268 | next: |
7267ec00 KS |
2269 | /* Huge page is mapped? No need to proceed. */ |
2270 | if (pmd_trans_huge(*fe->pmd)) | |
2271 | break; | |
bae473a4 | 2272 | if (iter.index == end_pgoff) |
f1820361 KS |
2273 | break; |
2274 | } | |
2275 | rcu_read_unlock(); | |
2276 | } | |
2277 | EXPORT_SYMBOL(filemap_map_pages); | |
2278 | ||
4fcf1c62 JK |
2279 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2280 | { | |
2281 | struct page *page = vmf->page; | |
496ad9aa | 2282 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2283 | int ret = VM_FAULT_LOCKED; |
2284 | ||
14da9200 | 2285 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2286 | file_update_time(vma->vm_file); |
2287 | lock_page(page); | |
2288 | if (page->mapping != inode->i_mapping) { | |
2289 | unlock_page(page); | |
2290 | ret = VM_FAULT_NOPAGE; | |
2291 | goto out; | |
2292 | } | |
14da9200 JK |
2293 | /* |
2294 | * We mark the page dirty already here so that when freeze is in | |
2295 | * progress, we are guaranteed that writeback during freezing will | |
2296 | * see the dirty page and writeprotect it again. | |
2297 | */ | |
2298 | set_page_dirty(page); | |
1d1d1a76 | 2299 | wait_for_stable_page(page); |
4fcf1c62 | 2300 | out: |
14da9200 | 2301 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2302 | return ret; |
2303 | } | |
2304 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2305 | ||
f0f37e2f | 2306 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2307 | .fault = filemap_fault, |
f1820361 | 2308 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2309 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2310 | }; |
2311 | ||
2312 | /* This is used for a general mmap of a disk file */ | |
2313 | ||
2314 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2315 | { | |
2316 | struct address_space *mapping = file->f_mapping; | |
2317 | ||
2318 | if (!mapping->a_ops->readpage) | |
2319 | return -ENOEXEC; | |
2320 | file_accessed(file); | |
2321 | vma->vm_ops = &generic_file_vm_ops; | |
2322 | return 0; | |
2323 | } | |
1da177e4 LT |
2324 | |
2325 | /* | |
2326 | * This is for filesystems which do not implement ->writepage. | |
2327 | */ | |
2328 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2329 | { | |
2330 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2331 | return -EINVAL; | |
2332 | return generic_file_mmap(file, vma); | |
2333 | } | |
2334 | #else | |
2335 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2336 | { | |
2337 | return -ENOSYS; | |
2338 | } | |
2339 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2340 | { | |
2341 | return -ENOSYS; | |
2342 | } | |
2343 | #endif /* CONFIG_MMU */ | |
2344 | ||
2345 | EXPORT_SYMBOL(generic_file_mmap); | |
2346 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2347 | ||
67f9fd91 SL |
2348 | static struct page *wait_on_page_read(struct page *page) |
2349 | { | |
2350 | if (!IS_ERR(page)) { | |
2351 | wait_on_page_locked(page); | |
2352 | if (!PageUptodate(page)) { | |
09cbfeaf | 2353 | put_page(page); |
67f9fd91 SL |
2354 | page = ERR_PTR(-EIO); |
2355 | } | |
2356 | } | |
2357 | return page; | |
2358 | } | |
2359 | ||
32b63529 | 2360 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2361 | pgoff_t index, |
5e5358e7 | 2362 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2363 | void *data, |
2364 | gfp_t gfp) | |
1da177e4 | 2365 | { |
eb2be189 | 2366 | struct page *page; |
1da177e4 LT |
2367 | int err; |
2368 | repeat: | |
2369 | page = find_get_page(mapping, index); | |
2370 | if (!page) { | |
0531b2aa | 2371 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2372 | if (!page) |
2373 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2374 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2375 | if (unlikely(err)) { |
09cbfeaf | 2376 | put_page(page); |
eb2be189 NP |
2377 | if (err == -EEXIST) |
2378 | goto repeat; | |
1da177e4 | 2379 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2380 | return ERR_PTR(err); |
2381 | } | |
32b63529 MG |
2382 | |
2383 | filler: | |
1da177e4 LT |
2384 | err = filler(data, page); |
2385 | if (err < 0) { | |
09cbfeaf | 2386 | put_page(page); |
32b63529 | 2387 | return ERR_PTR(err); |
1da177e4 | 2388 | } |
1da177e4 | 2389 | |
32b63529 MG |
2390 | page = wait_on_page_read(page); |
2391 | if (IS_ERR(page)) | |
2392 | return page; | |
2393 | goto out; | |
2394 | } | |
1da177e4 LT |
2395 | if (PageUptodate(page)) |
2396 | goto out; | |
2397 | ||
ebded027 MG |
2398 | /* |
2399 | * Page is not up to date and may be locked due one of the following | |
2400 | * case a: Page is being filled and the page lock is held | |
2401 | * case b: Read/write error clearing the page uptodate status | |
2402 | * case c: Truncation in progress (page locked) | |
2403 | * case d: Reclaim in progress | |
2404 | * | |
2405 | * Case a, the page will be up to date when the page is unlocked. | |
2406 | * There is no need to serialise on the page lock here as the page | |
2407 | * is pinned so the lock gives no additional protection. Even if the | |
2408 | * the page is truncated, the data is still valid if PageUptodate as | |
2409 | * it's a race vs truncate race. | |
2410 | * Case b, the page will not be up to date | |
2411 | * Case c, the page may be truncated but in itself, the data may still | |
2412 | * be valid after IO completes as it's a read vs truncate race. The | |
2413 | * operation must restart if the page is not uptodate on unlock but | |
2414 | * otherwise serialising on page lock to stabilise the mapping gives | |
2415 | * no additional guarantees to the caller as the page lock is | |
2416 | * released before return. | |
2417 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2418 | * will be a race with remove_mapping that determines if the mapping | |
2419 | * is valid on unlock but otherwise the data is valid and there is | |
2420 | * no need to serialise with page lock. | |
2421 | * | |
2422 | * As the page lock gives no additional guarantee, we optimistically | |
2423 | * wait on the page to be unlocked and check if it's up to date and | |
2424 | * use the page if it is. Otherwise, the page lock is required to | |
2425 | * distinguish between the different cases. The motivation is that we | |
2426 | * avoid spurious serialisations and wakeups when multiple processes | |
2427 | * wait on the same page for IO to complete. | |
2428 | */ | |
2429 | wait_on_page_locked(page); | |
2430 | if (PageUptodate(page)) | |
2431 | goto out; | |
2432 | ||
2433 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2434 | lock_page(page); |
ebded027 MG |
2435 | |
2436 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2437 | if (!page->mapping) { |
2438 | unlock_page(page); | |
09cbfeaf | 2439 | put_page(page); |
32b63529 | 2440 | goto repeat; |
1da177e4 | 2441 | } |
ebded027 MG |
2442 | |
2443 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2444 | if (PageUptodate(page)) { |
2445 | unlock_page(page); | |
2446 | goto out; | |
2447 | } | |
32b63529 MG |
2448 | goto filler; |
2449 | ||
c855ff37 | 2450 | out: |
6fe6900e NP |
2451 | mark_page_accessed(page); |
2452 | return page; | |
2453 | } | |
0531b2aa LT |
2454 | |
2455 | /** | |
67f9fd91 | 2456 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2457 | * @mapping: the page's address_space |
2458 | * @index: the page index | |
2459 | * @filler: function to perform the read | |
5e5358e7 | 2460 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2461 | * |
0531b2aa | 2462 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2463 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2464 | * |
2465 | * If the page does not get brought uptodate, return -EIO. | |
2466 | */ | |
67f9fd91 | 2467 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2468 | pgoff_t index, |
5e5358e7 | 2469 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2470 | void *data) |
2471 | { | |
2472 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2473 | } | |
67f9fd91 | 2474 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2475 | |
2476 | /** | |
2477 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2478 | * @mapping: the page's address_space | |
2479 | * @index: the page index | |
2480 | * @gfp: the page allocator flags to use if allocating | |
2481 | * | |
2482 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2483 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2484 | * |
2485 | * If the page does not get brought uptodate, return -EIO. | |
2486 | */ | |
2487 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2488 | pgoff_t index, | |
2489 | gfp_t gfp) | |
2490 | { | |
2491 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2492 | ||
67f9fd91 | 2493 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2494 | } |
2495 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2496 | ||
1da177e4 LT |
2497 | /* |
2498 | * Performs necessary checks before doing a write | |
2499 | * | |
485bb99b | 2500 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2501 | * Returns appropriate error code that caller should return or |
2502 | * zero in case that write should be allowed. | |
2503 | */ | |
3309dd04 | 2504 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2505 | { |
3309dd04 | 2506 | struct file *file = iocb->ki_filp; |
1da177e4 | 2507 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2508 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2509 | loff_t pos; |
1da177e4 | 2510 | |
3309dd04 AV |
2511 | if (!iov_iter_count(from)) |
2512 | return 0; | |
1da177e4 | 2513 | |
0fa6b005 | 2514 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2515 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2516 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2517 | |
3309dd04 | 2518 | pos = iocb->ki_pos; |
1da177e4 | 2519 | |
0fa6b005 | 2520 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2521 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2522 | send_sig(SIGXFSZ, current, 0); |
2523 | return -EFBIG; | |
1da177e4 | 2524 | } |
3309dd04 | 2525 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2526 | } |
2527 | ||
2528 | /* | |
2529 | * LFS rule | |
2530 | */ | |
3309dd04 | 2531 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2532 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2533 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2534 | return -EFBIG; |
3309dd04 | 2535 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2536 | } |
2537 | ||
2538 | /* | |
2539 | * Are we about to exceed the fs block limit ? | |
2540 | * | |
2541 | * If we have written data it becomes a short write. If we have | |
2542 | * exceeded without writing data we send a signal and return EFBIG. | |
2543 | * Linus frestrict idea will clean these up nicely.. | |
2544 | */ | |
3309dd04 AV |
2545 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2546 | return -EFBIG; | |
1da177e4 | 2547 | |
3309dd04 AV |
2548 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2549 | return iov_iter_count(from); | |
1da177e4 LT |
2550 | } |
2551 | EXPORT_SYMBOL(generic_write_checks); | |
2552 | ||
afddba49 NP |
2553 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2554 | loff_t pos, unsigned len, unsigned flags, | |
2555 | struct page **pagep, void **fsdata) | |
2556 | { | |
2557 | const struct address_space_operations *aops = mapping->a_ops; | |
2558 | ||
4e02ed4b | 2559 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2560 | pagep, fsdata); |
afddba49 NP |
2561 | } |
2562 | EXPORT_SYMBOL(pagecache_write_begin); | |
2563 | ||
2564 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2565 | loff_t pos, unsigned len, unsigned copied, | |
2566 | struct page *page, void *fsdata) | |
2567 | { | |
2568 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2569 | |
4e02ed4b | 2570 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2571 | } |
2572 | EXPORT_SYMBOL(pagecache_write_end); | |
2573 | ||
1da177e4 | 2574 | ssize_t |
1af5bb49 | 2575 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2576 | { |
2577 | struct file *file = iocb->ki_filp; | |
2578 | struct address_space *mapping = file->f_mapping; | |
2579 | struct inode *inode = mapping->host; | |
1af5bb49 | 2580 | loff_t pos = iocb->ki_pos; |
1da177e4 | 2581 | ssize_t written; |
a969e903 CH |
2582 | size_t write_len; |
2583 | pgoff_t end; | |
26978b8b | 2584 | struct iov_iter data; |
1da177e4 | 2585 | |
0c949334 | 2586 | write_len = iov_iter_count(from); |
09cbfeaf | 2587 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2588 | |
48b47c56 | 2589 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2590 | if (written) |
2591 | goto out; | |
2592 | ||
2593 | /* | |
2594 | * After a write we want buffered reads to be sure to go to disk to get | |
2595 | * the new data. We invalidate clean cached page from the region we're | |
2596 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2597 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2598 | */ |
2599 | if (mapping->nrpages) { | |
2600 | written = invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2601 | pos >> PAGE_SHIFT, end); |
6ccfa806 HH |
2602 | /* |
2603 | * If a page can not be invalidated, return 0 to fall back | |
2604 | * to buffered write. | |
2605 | */ | |
2606 | if (written) { | |
2607 | if (written == -EBUSY) | |
2608 | return 0; | |
a969e903 | 2609 | goto out; |
6ccfa806 | 2610 | } |
a969e903 CH |
2611 | } |
2612 | ||
26978b8b | 2613 | data = *from; |
c8b8e32d | 2614 | written = mapping->a_ops->direct_IO(iocb, &data); |
a969e903 CH |
2615 | |
2616 | /* | |
2617 | * Finally, try again to invalidate clean pages which might have been | |
2618 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2619 | * if the source of the write was an mmap'ed region of the file | |
2620 | * we're writing. Either one is a pretty crazy thing to do, | |
2621 | * so we don't support it 100%. If this invalidation | |
2622 | * fails, tough, the write still worked... | |
2623 | */ | |
2624 | if (mapping->nrpages) { | |
2625 | invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2626 | pos >> PAGE_SHIFT, end); |
a969e903 CH |
2627 | } |
2628 | ||
1da177e4 | 2629 | if (written > 0) { |
0116651c | 2630 | pos += written; |
f8579f86 | 2631 | iov_iter_advance(from, written); |
0116651c NK |
2632 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2633 | i_size_write(inode, pos); | |
1da177e4 LT |
2634 | mark_inode_dirty(inode); |
2635 | } | |
5cb6c6c7 | 2636 | iocb->ki_pos = pos; |
1da177e4 | 2637 | } |
a969e903 | 2638 | out: |
1da177e4 LT |
2639 | return written; |
2640 | } | |
2641 | EXPORT_SYMBOL(generic_file_direct_write); | |
2642 | ||
eb2be189 NP |
2643 | /* |
2644 | * Find or create a page at the given pagecache position. Return the locked | |
2645 | * page. This function is specifically for buffered writes. | |
2646 | */ | |
54566b2c NP |
2647 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2648 | pgoff_t index, unsigned flags) | |
eb2be189 | 2649 | { |
eb2be189 | 2650 | struct page *page; |
bbddabe2 | 2651 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2652 | |
54566b2c | 2653 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2654 | fgp_flags |= FGP_NOFS; |
2655 | ||
2656 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2657 | mapping_gfp_mask(mapping)); |
c585a267 | 2658 | if (page) |
2457aec6 | 2659 | wait_for_stable_page(page); |
eb2be189 | 2660 | |
eb2be189 NP |
2661 | return page; |
2662 | } | |
54566b2c | 2663 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2664 | |
3b93f911 | 2665 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2666 | struct iov_iter *i, loff_t pos) |
2667 | { | |
2668 | struct address_space *mapping = file->f_mapping; | |
2669 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2670 | long status = 0; | |
2671 | ssize_t written = 0; | |
674b892e NP |
2672 | unsigned int flags = 0; |
2673 | ||
2674 | /* | |
2675 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2676 | */ | |
777eda2c | 2677 | if (!iter_is_iovec(i)) |
674b892e | 2678 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2679 | |
2680 | do { | |
2681 | struct page *page; | |
afddba49 NP |
2682 | unsigned long offset; /* Offset into pagecache page */ |
2683 | unsigned long bytes; /* Bytes to write to page */ | |
2684 | size_t copied; /* Bytes copied from user */ | |
2685 | void *fsdata; | |
2686 | ||
09cbfeaf KS |
2687 | offset = (pos & (PAGE_SIZE - 1)); |
2688 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
2689 | iov_iter_count(i)); |
2690 | ||
2691 | again: | |
00a3d660 LT |
2692 | /* |
2693 | * Bring in the user page that we will copy from _first_. | |
2694 | * Otherwise there's a nasty deadlock on copying from the | |
2695 | * same page as we're writing to, without it being marked | |
2696 | * up-to-date. | |
2697 | * | |
2698 | * Not only is this an optimisation, but it is also required | |
2699 | * to check that the address is actually valid, when atomic | |
2700 | * usercopies are used, below. | |
2701 | */ | |
2702 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2703 | status = -EFAULT; | |
2704 | break; | |
2705 | } | |
2706 | ||
296291cd JK |
2707 | if (fatal_signal_pending(current)) { |
2708 | status = -EINTR; | |
2709 | break; | |
2710 | } | |
2711 | ||
674b892e | 2712 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2713 | &page, &fsdata); |
2457aec6 | 2714 | if (unlikely(status < 0)) |
afddba49 NP |
2715 | break; |
2716 | ||
931e80e4 | 2717 | if (mapping_writably_mapped(mapping)) |
2718 | flush_dcache_page(page); | |
00a3d660 | 2719 | |
afddba49 | 2720 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2721 | flush_dcache_page(page); |
2722 | ||
2723 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2724 | page, fsdata); | |
2725 | if (unlikely(status < 0)) | |
2726 | break; | |
2727 | copied = status; | |
2728 | ||
2729 | cond_resched(); | |
2730 | ||
124d3b70 | 2731 | iov_iter_advance(i, copied); |
afddba49 NP |
2732 | if (unlikely(copied == 0)) { |
2733 | /* | |
2734 | * If we were unable to copy any data at all, we must | |
2735 | * fall back to a single segment length write. | |
2736 | * | |
2737 | * If we didn't fallback here, we could livelock | |
2738 | * because not all segments in the iov can be copied at | |
2739 | * once without a pagefault. | |
2740 | */ | |
09cbfeaf | 2741 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
2742 | iov_iter_single_seg_count(i)); |
2743 | goto again; | |
2744 | } | |
afddba49 NP |
2745 | pos += copied; |
2746 | written += copied; | |
2747 | ||
2748 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
2749 | } while (iov_iter_count(i)); |
2750 | ||
2751 | return written ? written : status; | |
2752 | } | |
3b93f911 | 2753 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2754 | |
e4dd9de3 | 2755 | /** |
8174202b | 2756 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2757 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2758 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2759 | * |
2760 | * This function does all the work needed for actually writing data to a | |
2761 | * file. It does all basic checks, removes SUID from the file, updates | |
2762 | * modification times and calls proper subroutines depending on whether we | |
2763 | * do direct IO or a standard buffered write. | |
2764 | * | |
2765 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2766 | * object which does not need locking at all. | |
2767 | * | |
2768 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2769 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2770 | * avoid syncing under i_mutex. | |
2771 | */ | |
8174202b | 2772 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2773 | { |
2774 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2775 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2776 | struct inode *inode = mapping->host; |
3b93f911 | 2777 | ssize_t written = 0; |
1da177e4 | 2778 | ssize_t err; |
3b93f911 | 2779 | ssize_t status; |
1da177e4 | 2780 | |
1da177e4 | 2781 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2782 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 2783 | err = file_remove_privs(file); |
1da177e4 LT |
2784 | if (err) |
2785 | goto out; | |
2786 | ||
c3b2da31 JB |
2787 | err = file_update_time(file); |
2788 | if (err) | |
2789 | goto out; | |
1da177e4 | 2790 | |
2ba48ce5 | 2791 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2792 | loff_t pos, endbyte; |
fb5527e6 | 2793 | |
1af5bb49 | 2794 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 2795 | /* |
fbbbad4b MW |
2796 | * If the write stopped short of completing, fall back to |
2797 | * buffered writes. Some filesystems do this for writes to | |
2798 | * holes, for example. For DAX files, a buffered write will | |
2799 | * not succeed (even if it did, DAX does not handle dirty | |
2800 | * page-cache pages correctly). | |
1da177e4 | 2801 | */ |
0b8def9d | 2802 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2803 | goto out; |
2804 | ||
0b8def9d | 2805 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2806 | /* |
3b93f911 | 2807 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2808 | * then we want to return the number of bytes which were |
2809 | * direct-written, or the error code if that was zero. Note | |
2810 | * that this differs from normal direct-io semantics, which | |
2811 | * will return -EFOO even if some bytes were written. | |
2812 | */ | |
60bb4529 | 2813 | if (unlikely(status < 0)) { |
3b93f911 | 2814 | err = status; |
fb5527e6 JM |
2815 | goto out; |
2816 | } | |
fb5527e6 JM |
2817 | /* |
2818 | * We need to ensure that the page cache pages are written to | |
2819 | * disk and invalidated to preserve the expected O_DIRECT | |
2820 | * semantics. | |
2821 | */ | |
3b93f911 | 2822 | endbyte = pos + status - 1; |
0b8def9d | 2823 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2824 | if (err == 0) { |
0b8def9d | 2825 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2826 | written += status; |
fb5527e6 | 2827 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
2828 | pos >> PAGE_SHIFT, |
2829 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
2830 | } else { |
2831 | /* | |
2832 | * We don't know how much we wrote, so just return | |
2833 | * the number of bytes which were direct-written | |
2834 | */ | |
2835 | } | |
2836 | } else { | |
0b8def9d AV |
2837 | written = generic_perform_write(file, from, iocb->ki_pos); |
2838 | if (likely(written > 0)) | |
2839 | iocb->ki_pos += written; | |
fb5527e6 | 2840 | } |
1da177e4 LT |
2841 | out: |
2842 | current->backing_dev_info = NULL; | |
2843 | return written ? written : err; | |
2844 | } | |
8174202b | 2845 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2846 | |
e4dd9de3 | 2847 | /** |
8174202b | 2848 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2849 | * @iocb: IO state structure |
8174202b | 2850 | * @from: iov_iter with data to write |
e4dd9de3 | 2851 | * |
8174202b | 2852 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2853 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2854 | * and acquires i_mutex as needed. | |
2855 | */ | |
8174202b | 2856 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2857 | { |
2858 | struct file *file = iocb->ki_filp; | |
148f948b | 2859 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2860 | ssize_t ret; |
1da177e4 | 2861 | |
5955102c | 2862 | inode_lock(inode); |
3309dd04 AV |
2863 | ret = generic_write_checks(iocb, from); |
2864 | if (ret > 0) | |
5f380c7f | 2865 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 2866 | inode_unlock(inode); |
1da177e4 | 2867 | |
e2592217 CH |
2868 | if (ret > 0) |
2869 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
2870 | return ret; |
2871 | } | |
8174202b | 2872 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2873 | |
cf9a2ae8 DH |
2874 | /** |
2875 | * try_to_release_page() - release old fs-specific metadata on a page | |
2876 | * | |
2877 | * @page: the page which the kernel is trying to free | |
2878 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2879 | * | |
2880 | * The address_space is to try to release any data against the page | |
2881 | * (presumably at page->private). If the release was successful, return `1'. | |
2882 | * Otherwise return zero. | |
2883 | * | |
266cf658 DH |
2884 | * This may also be called if PG_fscache is set on a page, indicating that the |
2885 | * page is known to the local caching routines. | |
2886 | * | |
cf9a2ae8 | 2887 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 2888 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 2889 | * |
cf9a2ae8 DH |
2890 | */ |
2891 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2892 | { | |
2893 | struct address_space * const mapping = page->mapping; | |
2894 | ||
2895 | BUG_ON(!PageLocked(page)); | |
2896 | if (PageWriteback(page)) | |
2897 | return 0; | |
2898 | ||
2899 | if (mapping && mapping->a_ops->releasepage) | |
2900 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2901 | return try_to_free_buffers(page); | |
2902 | } | |
2903 | ||
2904 | EXPORT_SYMBOL(try_to_release_page); |