]>
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 | ||
22f2ac51 JW |
113 | static int page_cache_tree_insert(struct address_space *mapping, |
114 | struct page *page, void **shadowp) | |
115 | { | |
116 | struct radix_tree_node *node; | |
117 | void **slot; | |
118 | int error; | |
119 | ||
120 | error = __radix_tree_create(&mapping->page_tree, page->index, 0, | |
121 | &node, &slot); | |
122 | if (error) | |
123 | return error; | |
124 | if (*slot) { | |
125 | void *p; | |
126 | ||
127 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
128 | if (!radix_tree_exceptional_entry(p)) | |
129 | return -EEXIST; | |
130 | ||
131 | mapping->nrexceptional--; | |
132 | if (!dax_mapping(mapping)) { | |
133 | if (shadowp) | |
134 | *shadowp = p; | |
22f2ac51 JW |
135 | } else { |
136 | /* DAX can replace empty locked entry with a hole */ | |
137 | WARN_ON_ONCE(p != | |
138 | (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY | | |
139 | RADIX_DAX_ENTRY_LOCK)); | |
22f2ac51 JW |
140 | /* Wakeup waiters for exceptional entry lock */ |
141 | dax_wake_mapping_entry_waiter(mapping, page->index, | |
142 | false); | |
143 | } | |
144 | } | |
14b46879 JW |
145 | __radix_tree_replace(&mapping->page_tree, node, slot, page, |
146 | workingset_update_node, mapping); | |
22f2ac51 | 147 | mapping->nrpages++; |
22f2ac51 JW |
148 | return 0; |
149 | } | |
150 | ||
91b0abe3 JW |
151 | static void page_cache_tree_delete(struct address_space *mapping, |
152 | struct page *page, void *shadow) | |
153 | { | |
c70b647d KS |
154 | int i, nr; |
155 | ||
156 | /* hugetlb pages are represented by one entry in the radix tree */ | |
157 | nr = PageHuge(page) ? 1 : hpage_nr_pages(page); | |
91b0abe3 | 158 | |
83929372 KS |
159 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
160 | VM_BUG_ON_PAGE(PageTail(page), page); | |
161 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 162 | |
83929372 | 163 | for (i = 0; i < nr; i++) { |
d3798ae8 JW |
164 | struct radix_tree_node *node; |
165 | void **slot; | |
166 | ||
167 | __radix_tree_lookup(&mapping->page_tree, page->index + i, | |
168 | &node, &slot); | |
169 | ||
dbc446b8 | 170 | VM_BUG_ON_PAGE(!node && nr != 1, page); |
449dd698 | 171 | |
14b46879 JW |
172 | radix_tree_clear_tags(&mapping->page_tree, node, slot); |
173 | __radix_tree_replace(&mapping->page_tree, node, slot, shadow, | |
174 | workingset_update_node, mapping); | |
449dd698 | 175 | } |
d3798ae8 JW |
176 | |
177 | if (shadow) { | |
178 | mapping->nrexceptional += nr; | |
179 | /* | |
180 | * Make sure the nrexceptional update is committed before | |
181 | * the nrpages update so that final truncate racing | |
182 | * with reclaim does not see both counters 0 at the | |
183 | * same time and miss a shadow entry. | |
184 | */ | |
185 | smp_wmb(); | |
186 | } | |
187 | mapping->nrpages -= nr; | |
91b0abe3 JW |
188 | } |
189 | ||
1da177e4 | 190 | /* |
e64a782f | 191 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 192 | * sure the page is locked and that nobody else uses it - or that usage |
fdf1cdb9 | 193 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 194 | */ |
62cccb8c | 195 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
196 | { |
197 | struct address_space *mapping = page->mapping; | |
83929372 | 198 | int nr = hpage_nr_pages(page); |
1da177e4 | 199 | |
fe0bfaaf | 200 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
201 | /* |
202 | * if we're uptodate, flush out into the cleancache, otherwise | |
203 | * invalidate any existing cleancache entries. We can't leave | |
204 | * stale data around in the cleancache once our page is gone | |
205 | */ | |
206 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
207 | cleancache_put_page(page); | |
208 | else | |
3167760f | 209 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 210 | |
83929372 | 211 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
212 | VM_BUG_ON_PAGE(page_mapped(page), page); |
213 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
214 | int mapcount; | |
215 | ||
216 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
217 | current->comm, page_to_pfn(page)); | |
218 | dump_page(page, "still mapped when deleted"); | |
219 | dump_stack(); | |
220 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
221 | ||
222 | mapcount = page_mapcount(page); | |
223 | if (mapping_exiting(mapping) && | |
224 | page_count(page) >= mapcount + 2) { | |
225 | /* | |
226 | * All vmas have already been torn down, so it's | |
227 | * a good bet that actually the page is unmapped, | |
228 | * and we'd prefer not to leak it: if we're wrong, | |
229 | * some other bad page check should catch it later. | |
230 | */ | |
231 | page_mapcount_reset(page); | |
6d061f9f | 232 | page_ref_sub(page, mapcount); |
06b241f3 HD |
233 | } |
234 | } | |
235 | ||
91b0abe3 JW |
236 | page_cache_tree_delete(mapping, page, shadow); |
237 | ||
1da177e4 | 238 | page->mapping = NULL; |
b85e0eff | 239 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 240 | |
4165b9b4 MH |
241 | /* hugetlb pages do not participate in page cache accounting. */ |
242 | if (!PageHuge(page)) | |
11fb9989 | 243 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); |
800d8c63 | 244 | if (PageSwapBacked(page)) { |
11fb9989 | 245 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); |
800d8c63 | 246 | if (PageTransHuge(page)) |
11fb9989 | 247 | __dec_node_page_state(page, NR_SHMEM_THPS); |
800d8c63 KS |
248 | } else { |
249 | VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page); | |
250 | } | |
3a692790 LT |
251 | |
252 | /* | |
b9ea2515 KK |
253 | * At this point page must be either written or cleaned by truncate. |
254 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 255 | * |
b9ea2515 KK |
256 | * This fixes dirty accounting after removing the page entirely but |
257 | * leaves PageDirty set: it has no effect for truncated page and | |
258 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 259 | */ |
b9ea2515 | 260 | if (WARN_ON_ONCE(PageDirty(page))) |
62cccb8c | 261 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); |
1da177e4 LT |
262 | } |
263 | ||
702cfbf9 MK |
264 | /** |
265 | * delete_from_page_cache - delete page from page cache | |
266 | * @page: the page which the kernel is trying to remove from page cache | |
267 | * | |
268 | * This must be called only on pages that have been verified to be in the page | |
269 | * cache and locked. It will never put the page into the free list, the caller | |
270 | * has a reference on the page. | |
271 | */ | |
272 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 273 | { |
83929372 | 274 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 275 | unsigned long flags; |
6072d13c | 276 | void (*freepage)(struct page *); |
1da177e4 | 277 | |
cd7619d6 | 278 | BUG_ON(!PageLocked(page)); |
1da177e4 | 279 | |
6072d13c | 280 | freepage = mapping->a_ops->freepage; |
c4843a75 | 281 | |
c4843a75 | 282 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 283 | __delete_from_page_cache(page, NULL); |
c4843a75 | 284 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
285 | |
286 | if (freepage) | |
287 | freepage(page); | |
83929372 KS |
288 | |
289 | if (PageTransHuge(page) && !PageHuge(page)) { | |
290 | page_ref_sub(page, HPAGE_PMD_NR); | |
291 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
292 | } else { | |
293 | put_page(page); | |
294 | } | |
97cecb5a MK |
295 | } |
296 | EXPORT_SYMBOL(delete_from_page_cache); | |
297 | ||
d72d9e2a | 298 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
299 | { |
300 | int ret = 0; | |
301 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
302 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
303 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 304 | ret = -ENOSPC; |
7fcbbaf1 JA |
305 | if (test_bit(AS_EIO, &mapping->flags) && |
306 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
307 | ret = -EIO; |
308 | return ret; | |
309 | } | |
d72d9e2a | 310 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 311 | |
1da177e4 | 312 | /** |
485bb99b | 313 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
314 | * @mapping: address space structure to write |
315 | * @start: offset in bytes where the range starts | |
469eb4d0 | 316 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 317 | * @sync_mode: enable synchronous operation |
1da177e4 | 318 | * |
485bb99b RD |
319 | * Start writeback against all of a mapping's dirty pages that lie |
320 | * within the byte offsets <start, end> inclusive. | |
321 | * | |
1da177e4 | 322 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 323 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
324 | * these two operations is that if a dirty page/buffer is encountered, it must |
325 | * be waited upon, and not just skipped over. | |
326 | */ | |
ebcf28e1 AM |
327 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
328 | loff_t end, int sync_mode) | |
1da177e4 LT |
329 | { |
330 | int ret; | |
331 | struct writeback_control wbc = { | |
332 | .sync_mode = sync_mode, | |
05fe478d | 333 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
334 | .range_start = start, |
335 | .range_end = end, | |
1da177e4 LT |
336 | }; |
337 | ||
338 | if (!mapping_cap_writeback_dirty(mapping)) | |
339 | return 0; | |
340 | ||
b16b1deb | 341 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 342 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 343 | wbc_detach_inode(&wbc); |
1da177e4 LT |
344 | return ret; |
345 | } | |
346 | ||
347 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
348 | int sync_mode) | |
349 | { | |
111ebb6e | 350 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
351 | } |
352 | ||
353 | int filemap_fdatawrite(struct address_space *mapping) | |
354 | { | |
355 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
356 | } | |
357 | EXPORT_SYMBOL(filemap_fdatawrite); | |
358 | ||
f4c0a0fd | 359 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 360 | loff_t end) |
1da177e4 LT |
361 | { |
362 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
363 | } | |
f4c0a0fd | 364 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 365 | |
485bb99b RD |
366 | /** |
367 | * filemap_flush - mostly a non-blocking flush | |
368 | * @mapping: target address_space | |
369 | * | |
1da177e4 LT |
370 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
371 | * purposes - I/O may not be started against all dirty pages. | |
372 | */ | |
373 | int filemap_flush(struct address_space *mapping) | |
374 | { | |
375 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
376 | } | |
377 | EXPORT_SYMBOL(filemap_flush); | |
378 | ||
aa750fd7 JN |
379 | static int __filemap_fdatawait_range(struct address_space *mapping, |
380 | loff_t start_byte, loff_t end_byte) | |
1da177e4 | 381 | { |
09cbfeaf KS |
382 | pgoff_t index = start_byte >> PAGE_SHIFT; |
383 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
384 | struct pagevec pvec; |
385 | int nr_pages; | |
aa750fd7 | 386 | int ret = 0; |
1da177e4 | 387 | |
94004ed7 | 388 | if (end_byte < start_byte) |
865ffef3 | 389 | goto out; |
1da177e4 LT |
390 | |
391 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
392 | while ((index <= end) && |
393 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
394 | PAGECACHE_TAG_WRITEBACK, | |
395 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
396 | unsigned i; | |
397 | ||
398 | for (i = 0; i < nr_pages; i++) { | |
399 | struct page *page = pvec.pages[i]; | |
400 | ||
401 | /* until radix tree lookup accepts end_index */ | |
402 | if (page->index > end) | |
403 | continue; | |
404 | ||
405 | wait_on_page_writeback(page); | |
212260aa | 406 | if (TestClearPageError(page)) |
1da177e4 LT |
407 | ret = -EIO; |
408 | } | |
409 | pagevec_release(&pvec); | |
410 | cond_resched(); | |
411 | } | |
865ffef3 | 412 | out: |
aa750fd7 JN |
413 | return ret; |
414 | } | |
415 | ||
416 | /** | |
417 | * filemap_fdatawait_range - wait for writeback to complete | |
418 | * @mapping: address space structure to wait for | |
419 | * @start_byte: offset in bytes where the range starts | |
420 | * @end_byte: offset in bytes where the range ends (inclusive) | |
421 | * | |
422 | * Walk the list of under-writeback pages of the given address space | |
423 | * in the given range and wait for all of them. Check error status of | |
424 | * the address space and return it. | |
425 | * | |
426 | * Since the error status of the address space is cleared by this function, | |
427 | * callers are responsible for checking the return value and handling and/or | |
428 | * reporting the error. | |
429 | */ | |
430 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
431 | loff_t end_byte) | |
432 | { | |
433 | int ret, ret2; | |
434 | ||
435 | ret = __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
865ffef3 DM |
436 | ret2 = filemap_check_errors(mapping); |
437 | if (!ret) | |
438 | ret = ret2; | |
1da177e4 LT |
439 | |
440 | return ret; | |
441 | } | |
d3bccb6f JK |
442 | EXPORT_SYMBOL(filemap_fdatawait_range); |
443 | ||
aa750fd7 JN |
444 | /** |
445 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
446 | * @mapping: address space structure to wait for | |
447 | * | |
448 | * Walk the list of under-writeback pages of the given address space | |
449 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
450 | * does not clear error status of the address space. | |
451 | * | |
452 | * Use this function if callers don't handle errors themselves. Expected | |
453 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
454 | * fsfreeze(8) | |
455 | */ | |
456 | void filemap_fdatawait_keep_errors(struct address_space *mapping) | |
457 | { | |
458 | loff_t i_size = i_size_read(mapping->host); | |
459 | ||
460 | if (i_size == 0) | |
461 | return; | |
462 | ||
463 | __filemap_fdatawait_range(mapping, 0, i_size - 1); | |
464 | } | |
465 | ||
1da177e4 | 466 | /** |
485bb99b | 467 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 468 | * @mapping: address space structure to wait for |
485bb99b RD |
469 | * |
470 | * Walk the list of under-writeback pages of the given address space | |
aa750fd7 JN |
471 | * and wait for all of them. Check error status of the address space |
472 | * and return it. | |
473 | * | |
474 | * Since the error status of the address space is cleared by this function, | |
475 | * callers are responsible for checking the return value and handling and/or | |
476 | * reporting the error. | |
1da177e4 LT |
477 | */ |
478 | int filemap_fdatawait(struct address_space *mapping) | |
479 | { | |
480 | loff_t i_size = i_size_read(mapping->host); | |
481 | ||
482 | if (i_size == 0) | |
483 | return 0; | |
484 | ||
94004ed7 | 485 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
486 | } |
487 | EXPORT_SYMBOL(filemap_fdatawait); | |
488 | ||
489 | int filemap_write_and_wait(struct address_space *mapping) | |
490 | { | |
28fd1298 | 491 | int err = 0; |
1da177e4 | 492 | |
7f6d5b52 RZ |
493 | if ((!dax_mapping(mapping) && mapping->nrpages) || |
494 | (dax_mapping(mapping) && mapping->nrexceptional)) { | |
28fd1298 OH |
495 | err = filemap_fdatawrite(mapping); |
496 | /* | |
497 | * Even if the above returned error, the pages may be | |
498 | * written partially (e.g. -ENOSPC), so we wait for it. | |
499 | * But the -EIO is special case, it may indicate the worst | |
500 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
501 | */ | |
502 | if (err != -EIO) { | |
503 | int err2 = filemap_fdatawait(mapping); | |
504 | if (!err) | |
505 | err = err2; | |
506 | } | |
865ffef3 DM |
507 | } else { |
508 | err = filemap_check_errors(mapping); | |
1da177e4 | 509 | } |
28fd1298 | 510 | return err; |
1da177e4 | 511 | } |
28fd1298 | 512 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 513 | |
485bb99b RD |
514 | /** |
515 | * filemap_write_and_wait_range - write out & wait on a file range | |
516 | * @mapping: the address_space for the pages | |
517 | * @lstart: offset in bytes where the range starts | |
518 | * @lend: offset in bytes where the range ends (inclusive) | |
519 | * | |
469eb4d0 AM |
520 | * Write out and wait upon file offsets lstart->lend, inclusive. |
521 | * | |
522 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
523 | * that this function can be used to write to the very end-of-file (end = -1). | |
524 | */ | |
1da177e4 LT |
525 | int filemap_write_and_wait_range(struct address_space *mapping, |
526 | loff_t lstart, loff_t lend) | |
527 | { | |
28fd1298 | 528 | int err = 0; |
1da177e4 | 529 | |
7f6d5b52 RZ |
530 | if ((!dax_mapping(mapping) && mapping->nrpages) || |
531 | (dax_mapping(mapping) && mapping->nrexceptional)) { | |
28fd1298 OH |
532 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
533 | WB_SYNC_ALL); | |
534 | /* See comment of filemap_write_and_wait() */ | |
535 | if (err != -EIO) { | |
94004ed7 CH |
536 | int err2 = filemap_fdatawait_range(mapping, |
537 | lstart, lend); | |
28fd1298 OH |
538 | if (!err) |
539 | err = err2; | |
540 | } | |
865ffef3 DM |
541 | } else { |
542 | err = filemap_check_errors(mapping); | |
1da177e4 | 543 | } |
28fd1298 | 544 | return err; |
1da177e4 | 545 | } |
f6995585 | 546 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 547 | |
ef6a3c63 MS |
548 | /** |
549 | * replace_page_cache_page - replace a pagecache page with a new one | |
550 | * @old: page to be replaced | |
551 | * @new: page to replace with | |
552 | * @gfp_mask: allocation mode | |
553 | * | |
554 | * This function replaces a page in the pagecache with a new one. On | |
555 | * success it acquires the pagecache reference for the new page and | |
556 | * drops it for the old page. Both the old and new pages must be | |
557 | * locked. This function does not add the new page to the LRU, the | |
558 | * caller must do that. | |
559 | * | |
560 | * The remove + add is atomic. The only way this function can fail is | |
561 | * memory allocation failure. | |
562 | */ | |
563 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
564 | { | |
565 | int error; | |
ef6a3c63 | 566 | |
309381fe SL |
567 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
568 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
569 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 570 | |
ef6a3c63 MS |
571 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
572 | if (!error) { | |
573 | struct address_space *mapping = old->mapping; | |
574 | void (*freepage)(struct page *); | |
c4843a75 | 575 | unsigned long flags; |
ef6a3c63 MS |
576 | |
577 | pgoff_t offset = old->index; | |
578 | freepage = mapping->a_ops->freepage; | |
579 | ||
09cbfeaf | 580 | get_page(new); |
ef6a3c63 MS |
581 | new->mapping = mapping; |
582 | new->index = offset; | |
583 | ||
c4843a75 | 584 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 585 | __delete_from_page_cache(old, NULL); |
22f2ac51 | 586 | error = page_cache_tree_insert(mapping, new, NULL); |
ef6a3c63 | 587 | BUG_ON(error); |
4165b9b4 MH |
588 | |
589 | /* | |
590 | * hugetlb pages do not participate in page cache accounting. | |
591 | */ | |
592 | if (!PageHuge(new)) | |
11fb9989 | 593 | __inc_node_page_state(new, NR_FILE_PAGES); |
ef6a3c63 | 594 | if (PageSwapBacked(new)) |
11fb9989 | 595 | __inc_node_page_state(new, NR_SHMEM); |
c4843a75 | 596 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6a93ca8f | 597 | mem_cgroup_migrate(old, new); |
ef6a3c63 MS |
598 | radix_tree_preload_end(); |
599 | if (freepage) | |
600 | freepage(old); | |
09cbfeaf | 601 | put_page(old); |
ef6a3c63 MS |
602 | } |
603 | ||
604 | return error; | |
605 | } | |
606 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
607 | ||
a528910e JW |
608 | static int __add_to_page_cache_locked(struct page *page, |
609 | struct address_space *mapping, | |
610 | pgoff_t offset, gfp_t gfp_mask, | |
611 | void **shadowp) | |
1da177e4 | 612 | { |
00501b53 JW |
613 | int huge = PageHuge(page); |
614 | struct mem_cgroup *memcg; | |
e286781d NP |
615 | int error; |
616 | ||
309381fe SL |
617 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
618 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 619 | |
00501b53 JW |
620 | if (!huge) { |
621 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 622 | gfp_mask, &memcg, false); |
00501b53 JW |
623 | if (error) |
624 | return error; | |
625 | } | |
1da177e4 | 626 | |
5e4c0d97 | 627 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 628 | if (error) { |
00501b53 | 629 | if (!huge) |
f627c2f5 | 630 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
631 | return error; |
632 | } | |
633 | ||
09cbfeaf | 634 | get_page(page); |
66a0c8ee KS |
635 | page->mapping = mapping; |
636 | page->index = offset; | |
637 | ||
638 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 639 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
640 | radix_tree_preload_end(); |
641 | if (unlikely(error)) | |
642 | goto err_insert; | |
4165b9b4 MH |
643 | |
644 | /* hugetlb pages do not participate in page cache accounting. */ | |
645 | if (!huge) | |
11fb9989 | 646 | __inc_node_page_state(page, NR_FILE_PAGES); |
66a0c8ee | 647 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 648 | if (!huge) |
f627c2f5 | 649 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
650 | trace_mm_filemap_add_to_page_cache(page); |
651 | return 0; | |
652 | err_insert: | |
653 | page->mapping = NULL; | |
654 | /* Leave page->index set: truncation relies upon it */ | |
655 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 656 | if (!huge) |
f627c2f5 | 657 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 658 | put_page(page); |
1da177e4 LT |
659 | return error; |
660 | } | |
a528910e JW |
661 | |
662 | /** | |
663 | * add_to_page_cache_locked - add a locked page to the pagecache | |
664 | * @page: page to add | |
665 | * @mapping: the page's address_space | |
666 | * @offset: page index | |
667 | * @gfp_mask: page allocation mode | |
668 | * | |
669 | * This function is used to add a page to the pagecache. It must be locked. | |
670 | * This function does not add the page to the LRU. The caller must do that. | |
671 | */ | |
672 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
673 | pgoff_t offset, gfp_t gfp_mask) | |
674 | { | |
675 | return __add_to_page_cache_locked(page, mapping, offset, | |
676 | gfp_mask, NULL); | |
677 | } | |
e286781d | 678 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
679 | |
680 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 681 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 682 | { |
a528910e | 683 | void *shadow = NULL; |
4f98a2fe RR |
684 | int ret; |
685 | ||
48c935ad | 686 | __SetPageLocked(page); |
a528910e JW |
687 | ret = __add_to_page_cache_locked(page, mapping, offset, |
688 | gfp_mask, &shadow); | |
689 | if (unlikely(ret)) | |
48c935ad | 690 | __ClearPageLocked(page); |
a528910e JW |
691 | else { |
692 | /* | |
693 | * The page might have been evicted from cache only | |
694 | * recently, in which case it should be activated like | |
695 | * any other repeatedly accessed page. | |
f0281a00 RR |
696 | * The exception is pages getting rewritten; evicting other |
697 | * data from the working set, only to cache data that will | |
698 | * get overwritten with something else, is a waste of memory. | |
a528910e | 699 | */ |
f0281a00 RR |
700 | if (!(gfp_mask & __GFP_WRITE) && |
701 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
702 | SetPageActive(page); |
703 | workingset_activation(page); | |
704 | } else | |
705 | ClearPageActive(page); | |
706 | lru_cache_add(page); | |
707 | } | |
1da177e4 LT |
708 | return ret; |
709 | } | |
18bc0bbd | 710 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 711 | |
44110fe3 | 712 | #ifdef CONFIG_NUMA |
2ae88149 | 713 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 714 | { |
c0ff7453 MX |
715 | int n; |
716 | struct page *page; | |
717 | ||
44110fe3 | 718 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
719 | unsigned int cpuset_mems_cookie; |
720 | do { | |
d26914d1 | 721 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 722 | n = cpuset_mem_spread_node(); |
96db800f | 723 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 724 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 725 | |
c0ff7453 | 726 | return page; |
44110fe3 | 727 | } |
2ae88149 | 728 | return alloc_pages(gfp, 0); |
44110fe3 | 729 | } |
2ae88149 | 730 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
731 | #endif |
732 | ||
1da177e4 LT |
733 | /* |
734 | * In order to wait for pages to become available there must be | |
735 | * waitqueues associated with pages. By using a hash table of | |
736 | * waitqueues where the bucket discipline is to maintain all | |
737 | * waiters on the same queue and wake all when any of the pages | |
738 | * become available, and for the woken contexts to check to be | |
739 | * sure the appropriate page became available, this saves space | |
740 | * at a cost of "thundering herd" phenomena during rare hash | |
741 | * collisions. | |
742 | */ | |
a4796e37 | 743 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 | 744 | { |
9dcb8b68 | 745 | return bit_waitqueue(page, 0); |
1da177e4 | 746 | } |
a4796e37 | 747 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 748 | |
920c7a5d | 749 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
750 | { |
751 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
752 | ||
753 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 754 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
755 | TASK_UNINTERRUPTIBLE); |
756 | } | |
757 | EXPORT_SYMBOL(wait_on_page_bit); | |
758 | ||
f62e00cc KM |
759 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
760 | { | |
761 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
762 | ||
763 | if (!test_bit(bit_nr, &page->flags)) | |
764 | return 0; | |
765 | ||
766 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 767 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
768 | } |
769 | ||
cbbce822 N |
770 | int wait_on_page_bit_killable_timeout(struct page *page, |
771 | int bit_nr, unsigned long timeout) | |
772 | { | |
773 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
774 | ||
775 | wait.key.timeout = jiffies + timeout; | |
776 | if (!test_bit(bit_nr, &page->flags)) | |
777 | return 0; | |
778 | return __wait_on_bit(page_waitqueue(page), &wait, | |
779 | bit_wait_io_timeout, TASK_KILLABLE); | |
780 | } | |
781 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
782 | ||
385e1ca5 DH |
783 | /** |
784 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
785 | * @page: Page defining the wait queue of interest |
786 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
787 | * |
788 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
789 | */ | |
790 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
791 | { | |
792 | wait_queue_head_t *q = page_waitqueue(page); | |
793 | unsigned long flags; | |
794 | ||
795 | spin_lock_irqsave(&q->lock, flags); | |
796 | __add_wait_queue(q, waiter); | |
797 | spin_unlock_irqrestore(&q->lock, flags); | |
798 | } | |
799 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
800 | ||
1da177e4 | 801 | /** |
485bb99b | 802 | * unlock_page - unlock a locked page |
1da177e4 LT |
803 | * @page: the page |
804 | * | |
805 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
806 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 807 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
808 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
809 | * | |
8413ac9d NP |
810 | * The mb is necessary to enforce ordering between the clear_bit and the read |
811 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 812 | */ |
920c7a5d | 813 | void unlock_page(struct page *page) |
1da177e4 | 814 | { |
48c935ad | 815 | page = compound_head(page); |
309381fe | 816 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 817 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 818 | smp_mb__after_atomic(); |
1da177e4 LT |
819 | wake_up_page(page, PG_locked); |
820 | } | |
821 | EXPORT_SYMBOL(unlock_page); | |
822 | ||
485bb99b RD |
823 | /** |
824 | * end_page_writeback - end writeback against a page | |
825 | * @page: the page | |
1da177e4 LT |
826 | */ |
827 | void end_page_writeback(struct page *page) | |
828 | { | |
888cf2db MG |
829 | /* |
830 | * TestClearPageReclaim could be used here but it is an atomic | |
831 | * operation and overkill in this particular case. Failing to | |
832 | * shuffle a page marked for immediate reclaim is too mild to | |
833 | * justify taking an atomic operation penalty at the end of | |
834 | * ever page writeback. | |
835 | */ | |
836 | if (PageReclaim(page)) { | |
837 | ClearPageReclaim(page); | |
ac6aadb2 | 838 | rotate_reclaimable_page(page); |
888cf2db | 839 | } |
ac6aadb2 MS |
840 | |
841 | if (!test_clear_page_writeback(page)) | |
842 | BUG(); | |
843 | ||
4e857c58 | 844 | smp_mb__after_atomic(); |
1da177e4 LT |
845 | wake_up_page(page, PG_writeback); |
846 | } | |
847 | EXPORT_SYMBOL(end_page_writeback); | |
848 | ||
57d99845 MW |
849 | /* |
850 | * After completing I/O on a page, call this routine to update the page | |
851 | * flags appropriately | |
852 | */ | |
c11f0c0b | 853 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 854 | { |
c11f0c0b | 855 | if (!is_write) { |
57d99845 MW |
856 | if (!err) { |
857 | SetPageUptodate(page); | |
858 | } else { | |
859 | ClearPageUptodate(page); | |
860 | SetPageError(page); | |
861 | } | |
862 | unlock_page(page); | |
abf54548 | 863 | } else { |
57d99845 MW |
864 | if (err) { |
865 | SetPageError(page); | |
866 | if (page->mapping) | |
867 | mapping_set_error(page->mapping, err); | |
868 | } | |
869 | end_page_writeback(page); | |
870 | } | |
871 | } | |
872 | EXPORT_SYMBOL_GPL(page_endio); | |
873 | ||
485bb99b RD |
874 | /** |
875 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
876 | * @page: the page to lock | |
1da177e4 | 877 | */ |
920c7a5d | 878 | void __lock_page(struct page *page) |
1da177e4 | 879 | { |
48c935ad KS |
880 | struct page *page_head = compound_head(page); |
881 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
1da177e4 | 882 | |
48c935ad | 883 | __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io, |
1da177e4 LT |
884 | TASK_UNINTERRUPTIBLE); |
885 | } | |
886 | EXPORT_SYMBOL(__lock_page); | |
887 | ||
b5606c2d | 888 | int __lock_page_killable(struct page *page) |
2687a356 | 889 | { |
48c935ad KS |
890 | struct page *page_head = compound_head(page); |
891 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
2687a356 | 892 | |
48c935ad | 893 | return __wait_on_bit_lock(page_waitqueue(page_head), &wait, |
74316201 | 894 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 895 | } |
18bc0bbd | 896 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 897 | |
9a95f3cf PC |
898 | /* |
899 | * Return values: | |
900 | * 1 - page is locked; mmap_sem is still held. | |
901 | * 0 - page is not locked. | |
902 | * mmap_sem has been released (up_read()), unless flags had both | |
903 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
904 | * which case mmap_sem is still held. | |
905 | * | |
906 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
907 | * with the page locked and the mmap_sem unperturbed. | |
908 | */ | |
d065bd81 ML |
909 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
910 | unsigned int flags) | |
911 | { | |
37b23e05 KM |
912 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
913 | /* | |
914 | * CAUTION! In this case, mmap_sem is not released | |
915 | * even though return 0. | |
916 | */ | |
917 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
918 | return 0; | |
919 | ||
920 | up_read(&mm->mmap_sem); | |
921 | if (flags & FAULT_FLAG_KILLABLE) | |
922 | wait_on_page_locked_killable(page); | |
923 | else | |
318b275f | 924 | wait_on_page_locked(page); |
d065bd81 | 925 | return 0; |
37b23e05 KM |
926 | } else { |
927 | if (flags & FAULT_FLAG_KILLABLE) { | |
928 | int ret; | |
929 | ||
930 | ret = __lock_page_killable(page); | |
931 | if (ret) { | |
932 | up_read(&mm->mmap_sem); | |
933 | return 0; | |
934 | } | |
935 | } else | |
936 | __lock_page(page); | |
937 | return 1; | |
d065bd81 ML |
938 | } |
939 | } | |
940 | ||
e7b563bb JW |
941 | /** |
942 | * page_cache_next_hole - find the next hole (not-present entry) | |
943 | * @mapping: mapping | |
944 | * @index: index | |
945 | * @max_scan: maximum range to search | |
946 | * | |
947 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
948 | * lowest indexed hole. | |
949 | * | |
950 | * Returns: the index of the hole if found, otherwise returns an index | |
951 | * outside of the set specified (in which case 'return - index >= | |
952 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
953 | * be returned. | |
954 | * | |
955 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
956 | * like radix_tree_gang_lookup, this will not atomically search a | |
957 | * snapshot of the tree at a single point in time. For example, if a | |
958 | * hole is created at index 5, then subsequently a hole is created at | |
959 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
960 | * if called under rcu_read_lock. | |
961 | */ | |
962 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
963 | pgoff_t index, unsigned long max_scan) | |
964 | { | |
965 | unsigned long i; | |
966 | ||
967 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
968 | struct page *page; |
969 | ||
970 | page = radix_tree_lookup(&mapping->page_tree, index); | |
971 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
972 | break; |
973 | index++; | |
974 | if (index == 0) | |
975 | break; | |
976 | } | |
977 | ||
978 | return index; | |
979 | } | |
980 | EXPORT_SYMBOL(page_cache_next_hole); | |
981 | ||
982 | /** | |
983 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
984 | * @mapping: mapping | |
985 | * @index: index | |
986 | * @max_scan: maximum range to search | |
987 | * | |
988 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
989 | * the first hole. | |
990 | * | |
991 | * Returns: the index of the hole if found, otherwise returns an index | |
992 | * outside of the set specified (in which case 'index - return >= | |
993 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
994 | * will be returned. | |
995 | * | |
996 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
997 | * like radix_tree_gang_lookup, this will not atomically search a | |
998 | * snapshot of the tree at a single point in time. For example, if a | |
999 | * hole is created at index 10, then subsequently a hole is created at | |
1000 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1001 | * called under rcu_read_lock. | |
1002 | */ | |
1003 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1004 | pgoff_t index, unsigned long max_scan) | |
1005 | { | |
1006 | unsigned long i; | |
1007 | ||
1008 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1009 | struct page *page; |
1010 | ||
1011 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1012 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1013 | break; |
1014 | index--; | |
1015 | if (index == ULONG_MAX) | |
1016 | break; | |
1017 | } | |
1018 | ||
1019 | return index; | |
1020 | } | |
1021 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1022 | ||
485bb99b | 1023 | /** |
0cd6144a | 1024 | * find_get_entry - find and get a page cache entry |
485bb99b | 1025 | * @mapping: the address_space to search |
0cd6144a JW |
1026 | * @offset: the page cache index |
1027 | * | |
1028 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1029 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1030 | * |
139b6a6f JW |
1031 | * If the slot holds a shadow entry of a previously evicted page, or a |
1032 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1033 | * |
1034 | * Otherwise, %NULL is returned. | |
1da177e4 | 1035 | */ |
0cd6144a | 1036 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1037 | { |
a60637c8 | 1038 | void **pagep; |
83929372 | 1039 | struct page *head, *page; |
1da177e4 | 1040 | |
a60637c8 NP |
1041 | rcu_read_lock(); |
1042 | repeat: | |
1043 | page = NULL; | |
1044 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1045 | if (pagep) { | |
1046 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1047 | if (unlikely(!page)) |
1048 | goto out; | |
a2c16d6c | 1049 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1050 | if (radix_tree_deref_retry(page)) |
1051 | goto repeat; | |
1052 | /* | |
139b6a6f JW |
1053 | * A shadow entry of a recently evicted page, |
1054 | * or a swap entry from shmem/tmpfs. Return | |
1055 | * it without attempting to raise page count. | |
8079b1c8 HD |
1056 | */ |
1057 | goto out; | |
a2c16d6c | 1058 | } |
83929372 KS |
1059 | |
1060 | head = compound_head(page); | |
1061 | if (!page_cache_get_speculative(head)) | |
1062 | goto repeat; | |
1063 | ||
1064 | /* The page was split under us? */ | |
1065 | if (compound_head(page) != head) { | |
1066 | put_page(head); | |
a60637c8 | 1067 | goto repeat; |
83929372 | 1068 | } |
a60637c8 NP |
1069 | |
1070 | /* | |
1071 | * Has the page moved? | |
1072 | * This is part of the lockless pagecache protocol. See | |
1073 | * include/linux/pagemap.h for details. | |
1074 | */ | |
1075 | if (unlikely(page != *pagep)) { | |
83929372 | 1076 | put_page(head); |
a60637c8 NP |
1077 | goto repeat; |
1078 | } | |
1079 | } | |
27d20fdd | 1080 | out: |
a60637c8 NP |
1081 | rcu_read_unlock(); |
1082 | ||
1da177e4 LT |
1083 | return page; |
1084 | } | |
0cd6144a | 1085 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1086 | |
0cd6144a JW |
1087 | /** |
1088 | * find_lock_entry - locate, pin and lock a page cache entry | |
1089 | * @mapping: the address_space to search | |
1090 | * @offset: the page cache index | |
1091 | * | |
1092 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1093 | * page cache page, it is returned locked and with an increased | |
1094 | * refcount. | |
1095 | * | |
139b6a6f JW |
1096 | * If the slot holds a shadow entry of a previously evicted page, or a |
1097 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1098 | * |
1099 | * Otherwise, %NULL is returned. | |
1100 | * | |
1101 | * find_lock_entry() may sleep. | |
1102 | */ | |
1103 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1104 | { |
1105 | struct page *page; | |
1106 | ||
1da177e4 | 1107 | repeat: |
0cd6144a | 1108 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1109 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1110 | lock_page(page); |
1111 | /* Has the page been truncated? */ | |
83929372 | 1112 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1113 | unlock_page(page); |
09cbfeaf | 1114 | put_page(page); |
a60637c8 | 1115 | goto repeat; |
1da177e4 | 1116 | } |
83929372 | 1117 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1118 | } |
1da177e4 LT |
1119 | return page; |
1120 | } | |
0cd6144a JW |
1121 | EXPORT_SYMBOL(find_lock_entry); |
1122 | ||
1123 | /** | |
2457aec6 | 1124 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1125 | * @mapping: the address_space to search |
1126 | * @offset: the page index | |
2457aec6 | 1127 | * @fgp_flags: PCG flags |
45f87de5 | 1128 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1129 | * |
2457aec6 | 1130 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1131 | * |
75325189 | 1132 | * PCG flags modify how the page is returned. |
0cd6144a | 1133 | * |
2457aec6 MG |
1134 | * FGP_ACCESSED: the page will be marked accessed |
1135 | * FGP_LOCK: Page is return locked | |
1136 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1137 | * @gfp_mask and added to the page cache and the VM's LRU |
1138 | * list. The page is returned locked and with an increased | |
1139 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1140 | * |
2457aec6 MG |
1141 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1142 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1143 | * |
2457aec6 | 1144 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1145 | */ |
2457aec6 | 1146 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1147 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1148 | { |
eb2be189 | 1149 | struct page *page; |
2457aec6 | 1150 | |
1da177e4 | 1151 | repeat: |
2457aec6 MG |
1152 | page = find_get_entry(mapping, offset); |
1153 | if (radix_tree_exceptional_entry(page)) | |
1154 | page = NULL; | |
1155 | if (!page) | |
1156 | goto no_page; | |
1157 | ||
1158 | if (fgp_flags & FGP_LOCK) { | |
1159 | if (fgp_flags & FGP_NOWAIT) { | |
1160 | if (!trylock_page(page)) { | |
09cbfeaf | 1161 | put_page(page); |
2457aec6 MG |
1162 | return NULL; |
1163 | } | |
1164 | } else { | |
1165 | lock_page(page); | |
1166 | } | |
1167 | ||
1168 | /* Has the page been truncated? */ | |
1169 | if (unlikely(page->mapping != mapping)) { | |
1170 | unlock_page(page); | |
09cbfeaf | 1171 | put_page(page); |
2457aec6 MG |
1172 | goto repeat; |
1173 | } | |
1174 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1175 | } | |
1176 | ||
1177 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1178 | mark_page_accessed(page); | |
1179 | ||
1180 | no_page: | |
1181 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1182 | int err; | |
1183 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1184 | gfp_mask |= __GFP_WRITE; |
1185 | if (fgp_flags & FGP_NOFS) | |
1186 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1187 | |
45f87de5 | 1188 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1189 | if (!page) |
1190 | return NULL; | |
2457aec6 MG |
1191 | |
1192 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1193 | fgp_flags |= FGP_LOCK; | |
1194 | ||
eb39d618 | 1195 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1196 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1197 | __SetPageReferenced(page); |
2457aec6 | 1198 | |
45f87de5 MH |
1199 | err = add_to_page_cache_lru(page, mapping, offset, |
1200 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 | 1201 | if (unlikely(err)) { |
09cbfeaf | 1202 | put_page(page); |
eb2be189 NP |
1203 | page = NULL; |
1204 | if (err == -EEXIST) | |
1205 | goto repeat; | |
1da177e4 | 1206 | } |
1da177e4 | 1207 | } |
2457aec6 | 1208 | |
1da177e4 LT |
1209 | return page; |
1210 | } | |
2457aec6 | 1211 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1212 | |
0cd6144a JW |
1213 | /** |
1214 | * find_get_entries - gang pagecache lookup | |
1215 | * @mapping: The address_space to search | |
1216 | * @start: The starting page cache index | |
1217 | * @nr_entries: The maximum number of entries | |
1218 | * @entries: Where the resulting entries are placed | |
1219 | * @indices: The cache indices corresponding to the entries in @entries | |
1220 | * | |
1221 | * find_get_entries() will search for and return a group of up to | |
1222 | * @nr_entries entries in the mapping. The entries are placed at | |
1223 | * @entries. find_get_entries() takes a reference against any actual | |
1224 | * pages it returns. | |
1225 | * | |
1226 | * The search returns a group of mapping-contiguous page cache entries | |
1227 | * with ascending indexes. There may be holes in the indices due to | |
1228 | * not-present pages. | |
1229 | * | |
139b6a6f JW |
1230 | * Any shadow entries of evicted pages, or swap entries from |
1231 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1232 | * |
1233 | * find_get_entries() returns the number of pages and shadow entries | |
1234 | * which were found. | |
1235 | */ | |
1236 | unsigned find_get_entries(struct address_space *mapping, | |
1237 | pgoff_t start, unsigned int nr_entries, | |
1238 | struct page **entries, pgoff_t *indices) | |
1239 | { | |
1240 | void **slot; | |
1241 | unsigned int ret = 0; | |
1242 | struct radix_tree_iter iter; | |
1243 | ||
1244 | if (!nr_entries) | |
1245 | return 0; | |
1246 | ||
1247 | rcu_read_lock(); | |
0cd6144a | 1248 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
83929372 | 1249 | struct page *head, *page; |
0cd6144a JW |
1250 | repeat: |
1251 | page = radix_tree_deref_slot(slot); | |
1252 | if (unlikely(!page)) | |
1253 | continue; | |
1254 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1255 | if (radix_tree_deref_retry(page)) { |
1256 | slot = radix_tree_iter_retry(&iter); | |
1257 | continue; | |
1258 | } | |
0cd6144a | 1259 | /* |
f9fe48be RZ |
1260 | * A shadow entry of a recently evicted page, a swap |
1261 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1262 | * without attempting to raise page count. | |
0cd6144a JW |
1263 | */ |
1264 | goto export; | |
1265 | } | |
83929372 KS |
1266 | |
1267 | head = compound_head(page); | |
1268 | if (!page_cache_get_speculative(head)) | |
1269 | goto repeat; | |
1270 | ||
1271 | /* The page was split under us? */ | |
1272 | if (compound_head(page) != head) { | |
1273 | put_page(head); | |
0cd6144a | 1274 | goto repeat; |
83929372 | 1275 | } |
0cd6144a JW |
1276 | |
1277 | /* Has the page moved? */ | |
1278 | if (unlikely(page != *slot)) { | |
83929372 | 1279 | put_page(head); |
0cd6144a JW |
1280 | goto repeat; |
1281 | } | |
1282 | export: | |
1283 | indices[ret] = iter.index; | |
1284 | entries[ret] = page; | |
1285 | if (++ret == nr_entries) | |
1286 | break; | |
1287 | } | |
1288 | rcu_read_unlock(); | |
1289 | return ret; | |
1290 | } | |
1291 | ||
1da177e4 LT |
1292 | /** |
1293 | * find_get_pages - gang pagecache lookup | |
1294 | * @mapping: The address_space to search | |
1295 | * @start: The starting page index | |
1296 | * @nr_pages: The maximum number of pages | |
1297 | * @pages: Where the resulting pages are placed | |
1298 | * | |
1299 | * find_get_pages() will search for and return a group of up to | |
1300 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1301 | * find_get_pages() takes a reference against the returned pages. | |
1302 | * | |
1303 | * The search returns a group of mapping-contiguous pages with ascending | |
1304 | * indexes. There may be holes in the indices due to not-present pages. | |
1305 | * | |
1306 | * find_get_pages() returns the number of pages which were found. | |
1307 | */ | |
1308 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1309 | unsigned int nr_pages, struct page **pages) | |
1310 | { | |
0fc9d104 KK |
1311 | struct radix_tree_iter iter; |
1312 | void **slot; | |
1313 | unsigned ret = 0; | |
1314 | ||
1315 | if (unlikely(!nr_pages)) | |
1316 | return 0; | |
a60637c8 NP |
1317 | |
1318 | rcu_read_lock(); | |
0fc9d104 | 1319 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
83929372 | 1320 | struct page *head, *page; |
a60637c8 | 1321 | repeat: |
0fc9d104 | 1322 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1323 | if (unlikely(!page)) |
1324 | continue; | |
9d8aa4ea | 1325 | |
a2c16d6c | 1326 | if (radix_tree_exception(page)) { |
8079b1c8 | 1327 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1328 | slot = radix_tree_iter_retry(&iter); |
1329 | continue; | |
8079b1c8 | 1330 | } |
a2c16d6c | 1331 | /* |
139b6a6f JW |
1332 | * A shadow entry of a recently evicted page, |
1333 | * or a swap entry from shmem/tmpfs. Skip | |
1334 | * over it. | |
a2c16d6c | 1335 | */ |
8079b1c8 | 1336 | continue; |
27d20fdd | 1337 | } |
a60637c8 | 1338 | |
83929372 KS |
1339 | head = compound_head(page); |
1340 | if (!page_cache_get_speculative(head)) | |
1341 | goto repeat; | |
1342 | ||
1343 | /* The page was split under us? */ | |
1344 | if (compound_head(page) != head) { | |
1345 | put_page(head); | |
a60637c8 | 1346 | goto repeat; |
83929372 | 1347 | } |
a60637c8 NP |
1348 | |
1349 | /* Has the page moved? */ | |
0fc9d104 | 1350 | if (unlikely(page != *slot)) { |
83929372 | 1351 | put_page(head); |
a60637c8 NP |
1352 | goto repeat; |
1353 | } | |
1da177e4 | 1354 | |
a60637c8 | 1355 | pages[ret] = page; |
0fc9d104 KK |
1356 | if (++ret == nr_pages) |
1357 | break; | |
a60637c8 | 1358 | } |
5b280c0c | 1359 | |
a60637c8 | 1360 | rcu_read_unlock(); |
1da177e4 LT |
1361 | return ret; |
1362 | } | |
1363 | ||
ebf43500 JA |
1364 | /** |
1365 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1366 | * @mapping: The address_space to search | |
1367 | * @index: The starting page index | |
1368 | * @nr_pages: The maximum number of pages | |
1369 | * @pages: Where the resulting pages are placed | |
1370 | * | |
1371 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1372 | * that the returned number of pages are guaranteed to be contiguous. | |
1373 | * | |
1374 | * find_get_pages_contig() returns the number of pages which were found. | |
1375 | */ | |
1376 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1377 | unsigned int nr_pages, struct page **pages) | |
1378 | { | |
0fc9d104 KK |
1379 | struct radix_tree_iter iter; |
1380 | void **slot; | |
1381 | unsigned int ret = 0; | |
1382 | ||
1383 | if (unlikely(!nr_pages)) | |
1384 | return 0; | |
a60637c8 NP |
1385 | |
1386 | rcu_read_lock(); | |
0fc9d104 | 1387 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
83929372 | 1388 | struct page *head, *page; |
a60637c8 | 1389 | repeat: |
0fc9d104 KK |
1390 | page = radix_tree_deref_slot(slot); |
1391 | /* The hole, there no reason to continue */ | |
a60637c8 | 1392 | if (unlikely(!page)) |
0fc9d104 | 1393 | break; |
9d8aa4ea | 1394 | |
a2c16d6c | 1395 | if (radix_tree_exception(page)) { |
8079b1c8 | 1396 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1397 | slot = radix_tree_iter_retry(&iter); |
1398 | continue; | |
8079b1c8 | 1399 | } |
a2c16d6c | 1400 | /* |
139b6a6f JW |
1401 | * A shadow entry of a recently evicted page, |
1402 | * or a swap entry from shmem/tmpfs. Stop | |
1403 | * looking for contiguous pages. | |
a2c16d6c | 1404 | */ |
8079b1c8 | 1405 | break; |
a2c16d6c | 1406 | } |
ebf43500 | 1407 | |
83929372 KS |
1408 | head = compound_head(page); |
1409 | if (!page_cache_get_speculative(head)) | |
1410 | goto repeat; | |
1411 | ||
1412 | /* The page was split under us? */ | |
1413 | if (compound_head(page) != head) { | |
1414 | put_page(head); | |
a60637c8 | 1415 | goto repeat; |
83929372 | 1416 | } |
a60637c8 NP |
1417 | |
1418 | /* Has the page moved? */ | |
0fc9d104 | 1419 | if (unlikely(page != *slot)) { |
83929372 | 1420 | put_page(head); |
a60637c8 NP |
1421 | goto repeat; |
1422 | } | |
1423 | ||
9cbb4cb2 NP |
1424 | /* |
1425 | * must check mapping and index after taking the ref. | |
1426 | * otherwise we can get both false positives and false | |
1427 | * negatives, which is just confusing to the caller. | |
1428 | */ | |
83929372 | 1429 | if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { |
09cbfeaf | 1430 | put_page(page); |
9cbb4cb2 NP |
1431 | break; |
1432 | } | |
1433 | ||
a60637c8 | 1434 | pages[ret] = page; |
0fc9d104 KK |
1435 | if (++ret == nr_pages) |
1436 | break; | |
ebf43500 | 1437 | } |
a60637c8 NP |
1438 | rcu_read_unlock(); |
1439 | return ret; | |
ebf43500 | 1440 | } |
ef71c15c | 1441 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1442 | |
485bb99b RD |
1443 | /** |
1444 | * find_get_pages_tag - find and return pages that match @tag | |
1445 | * @mapping: the address_space to search | |
1446 | * @index: the starting page index | |
1447 | * @tag: the tag index | |
1448 | * @nr_pages: the maximum number of pages | |
1449 | * @pages: where the resulting pages are placed | |
1450 | * | |
1da177e4 | 1451 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1452 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1453 | */ |
1454 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1455 | int tag, unsigned int nr_pages, struct page **pages) | |
1456 | { | |
0fc9d104 KK |
1457 | struct radix_tree_iter iter; |
1458 | void **slot; | |
1459 | unsigned ret = 0; | |
1460 | ||
1461 | if (unlikely(!nr_pages)) | |
1462 | return 0; | |
a60637c8 NP |
1463 | |
1464 | rcu_read_lock(); | |
0fc9d104 KK |
1465 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1466 | &iter, *index, tag) { | |
83929372 | 1467 | struct page *head, *page; |
a60637c8 | 1468 | repeat: |
0fc9d104 | 1469 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1470 | if (unlikely(!page)) |
1471 | continue; | |
9d8aa4ea | 1472 | |
a2c16d6c | 1473 | if (radix_tree_exception(page)) { |
8079b1c8 | 1474 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1475 | slot = radix_tree_iter_retry(&iter); |
1476 | continue; | |
8079b1c8 | 1477 | } |
a2c16d6c | 1478 | /* |
139b6a6f JW |
1479 | * A shadow entry of a recently evicted page. |
1480 | * | |
1481 | * Those entries should never be tagged, but | |
1482 | * this tree walk is lockless and the tags are | |
1483 | * looked up in bulk, one radix tree node at a | |
1484 | * time, so there is a sizable window for page | |
1485 | * reclaim to evict a page we saw tagged. | |
1486 | * | |
1487 | * Skip over it. | |
a2c16d6c | 1488 | */ |
139b6a6f | 1489 | continue; |
a2c16d6c | 1490 | } |
a60637c8 | 1491 | |
83929372 KS |
1492 | head = compound_head(page); |
1493 | if (!page_cache_get_speculative(head)) | |
a60637c8 NP |
1494 | goto repeat; |
1495 | ||
83929372 KS |
1496 | /* The page was split under us? */ |
1497 | if (compound_head(page) != head) { | |
1498 | put_page(head); | |
1499 | goto repeat; | |
1500 | } | |
1501 | ||
a60637c8 | 1502 | /* Has the page moved? */ |
0fc9d104 | 1503 | if (unlikely(page != *slot)) { |
83929372 | 1504 | put_page(head); |
a60637c8 NP |
1505 | goto repeat; |
1506 | } | |
1507 | ||
1508 | pages[ret] = page; | |
0fc9d104 KK |
1509 | if (++ret == nr_pages) |
1510 | break; | |
a60637c8 | 1511 | } |
5b280c0c | 1512 | |
a60637c8 | 1513 | rcu_read_unlock(); |
1da177e4 | 1514 | |
1da177e4 LT |
1515 | if (ret) |
1516 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1517 | |
1da177e4 LT |
1518 | return ret; |
1519 | } | |
ef71c15c | 1520 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1521 | |
7e7f7749 RZ |
1522 | /** |
1523 | * find_get_entries_tag - find and return entries that match @tag | |
1524 | * @mapping: the address_space to search | |
1525 | * @start: the starting page cache index | |
1526 | * @tag: the tag index | |
1527 | * @nr_entries: the maximum number of entries | |
1528 | * @entries: where the resulting entries are placed | |
1529 | * @indices: the cache indices corresponding to the entries in @entries | |
1530 | * | |
1531 | * Like find_get_entries, except we only return entries which are tagged with | |
1532 | * @tag. | |
1533 | */ | |
1534 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1535 | int tag, unsigned int nr_entries, | |
1536 | struct page **entries, pgoff_t *indices) | |
1537 | { | |
1538 | void **slot; | |
1539 | unsigned int ret = 0; | |
1540 | struct radix_tree_iter iter; | |
1541 | ||
1542 | if (!nr_entries) | |
1543 | return 0; | |
1544 | ||
1545 | rcu_read_lock(); | |
7e7f7749 RZ |
1546 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1547 | &iter, start, tag) { | |
83929372 | 1548 | struct page *head, *page; |
7e7f7749 RZ |
1549 | repeat: |
1550 | page = radix_tree_deref_slot(slot); | |
1551 | if (unlikely(!page)) | |
1552 | continue; | |
1553 | if (radix_tree_exception(page)) { | |
1554 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1555 | slot = radix_tree_iter_retry(&iter); |
1556 | continue; | |
7e7f7749 RZ |
1557 | } |
1558 | ||
1559 | /* | |
1560 | * A shadow entry of a recently evicted page, a swap | |
1561 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1562 | * without attempting to raise page count. | |
1563 | */ | |
1564 | goto export; | |
1565 | } | |
83929372 KS |
1566 | |
1567 | head = compound_head(page); | |
1568 | if (!page_cache_get_speculative(head)) | |
7e7f7749 RZ |
1569 | goto repeat; |
1570 | ||
83929372 KS |
1571 | /* The page was split under us? */ |
1572 | if (compound_head(page) != head) { | |
1573 | put_page(head); | |
1574 | goto repeat; | |
1575 | } | |
1576 | ||
7e7f7749 RZ |
1577 | /* Has the page moved? */ |
1578 | if (unlikely(page != *slot)) { | |
83929372 | 1579 | put_page(head); |
7e7f7749 RZ |
1580 | goto repeat; |
1581 | } | |
1582 | export: | |
1583 | indices[ret] = iter.index; | |
1584 | entries[ret] = page; | |
1585 | if (++ret == nr_entries) | |
1586 | break; | |
1587 | } | |
1588 | rcu_read_unlock(); | |
1589 | return ret; | |
1590 | } | |
1591 | EXPORT_SYMBOL(find_get_entries_tag); | |
1592 | ||
76d42bd9 WF |
1593 | /* |
1594 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1595 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1596 | * | |
1597 | * ---R__________________________________________B__________ | |
1598 | * ^ reading here ^ bad block(assume 4k) | |
1599 | * | |
1600 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1601 | * => failing the whole request => read(R) => read(R+1) => | |
1602 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1603 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1604 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1605 | * | |
1606 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1607 | */ | |
1608 | static void shrink_readahead_size_eio(struct file *filp, | |
1609 | struct file_ra_state *ra) | |
1610 | { | |
76d42bd9 | 1611 | ra->ra_pages /= 4; |
76d42bd9 WF |
1612 | } |
1613 | ||
485bb99b | 1614 | /** |
36e78914 | 1615 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1616 | * @filp: the file to read |
1617 | * @ppos: current file position | |
6e58e79d AV |
1618 | * @iter: data destination |
1619 | * @written: already copied | |
485bb99b | 1620 | * |
1da177e4 | 1621 | * This is a generic file read routine, and uses the |
485bb99b | 1622 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1623 | * |
1624 | * This is really ugly. But the goto's actually try to clarify some | |
1625 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1626 | */ |
6e58e79d AV |
1627 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1628 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1629 | { |
36e78914 | 1630 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1631 | struct inode *inode = mapping->host; |
36e78914 | 1632 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1633 | pgoff_t index; |
1634 | pgoff_t last_index; | |
1635 | pgoff_t prev_index; | |
1636 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1637 | unsigned int prev_offset; |
6e58e79d | 1638 | int error = 0; |
1da177e4 | 1639 | |
c2a9737f WF |
1640 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
1641 | return -EINVAL; | |
1642 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); | |
1643 | ||
09cbfeaf KS |
1644 | index = *ppos >> PAGE_SHIFT; |
1645 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
1646 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
1647 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
1648 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 1649 | |
1da177e4 LT |
1650 | for (;;) { |
1651 | struct page *page; | |
57f6b96c | 1652 | pgoff_t end_index; |
a32ea1e1 | 1653 | loff_t isize; |
1da177e4 LT |
1654 | unsigned long nr, ret; |
1655 | ||
1da177e4 | 1656 | cond_resched(); |
1da177e4 LT |
1657 | find_page: |
1658 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1659 | if (!page) { |
cf914a7d | 1660 | page_cache_sync_readahead(mapping, |
7ff81078 | 1661 | ra, filp, |
3ea89ee8 FW |
1662 | index, last_index - index); |
1663 | page = find_get_page(mapping, index); | |
1664 | if (unlikely(page == NULL)) | |
1665 | goto no_cached_page; | |
1666 | } | |
1667 | if (PageReadahead(page)) { | |
cf914a7d | 1668 | page_cache_async_readahead(mapping, |
7ff81078 | 1669 | ra, filp, page, |
3ea89ee8 | 1670 | index, last_index - index); |
1da177e4 | 1671 | } |
8ab22b9a | 1672 | if (!PageUptodate(page)) { |
ebded027 MG |
1673 | /* |
1674 | * See comment in do_read_cache_page on why | |
1675 | * wait_on_page_locked is used to avoid unnecessarily | |
1676 | * serialisations and why it's safe. | |
1677 | */ | |
c4b209a4 BVA |
1678 | error = wait_on_page_locked_killable(page); |
1679 | if (unlikely(error)) | |
1680 | goto readpage_error; | |
ebded027 MG |
1681 | if (PageUptodate(page)) |
1682 | goto page_ok; | |
1683 | ||
09cbfeaf | 1684 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
1685 | !mapping->a_ops->is_partially_uptodate) |
1686 | goto page_not_up_to_date; | |
6d6d36bc EG |
1687 | /* pipes can't handle partially uptodate pages */ |
1688 | if (unlikely(iter->type & ITER_PIPE)) | |
1689 | goto page_not_up_to_date; | |
529ae9aa | 1690 | if (!trylock_page(page)) |
8ab22b9a | 1691 | goto page_not_up_to_date; |
8d056cb9 DH |
1692 | /* Did it get truncated before we got the lock? */ |
1693 | if (!page->mapping) | |
1694 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1695 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1696 | offset, iter->count)) |
8ab22b9a HH |
1697 | goto page_not_up_to_date_locked; |
1698 | unlock_page(page); | |
1699 | } | |
1da177e4 | 1700 | page_ok: |
a32ea1e1 N |
1701 | /* |
1702 | * i_size must be checked after we know the page is Uptodate. | |
1703 | * | |
1704 | * Checking i_size after the check allows us to calculate | |
1705 | * the correct value for "nr", which means the zero-filled | |
1706 | * part of the page is not copied back to userspace (unless | |
1707 | * another truncate extends the file - this is desired though). | |
1708 | */ | |
1709 | ||
1710 | isize = i_size_read(inode); | |
09cbfeaf | 1711 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 1712 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 1713 | put_page(page); |
a32ea1e1 N |
1714 | goto out; |
1715 | } | |
1716 | ||
1717 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 1718 | nr = PAGE_SIZE; |
a32ea1e1 | 1719 | if (index == end_index) { |
09cbfeaf | 1720 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 1721 | if (nr <= offset) { |
09cbfeaf | 1722 | put_page(page); |
a32ea1e1 N |
1723 | goto out; |
1724 | } | |
1725 | } | |
1726 | nr = nr - offset; | |
1da177e4 LT |
1727 | |
1728 | /* If users can be writing to this page using arbitrary | |
1729 | * virtual addresses, take care about potential aliasing | |
1730 | * before reading the page on the kernel side. | |
1731 | */ | |
1732 | if (mapping_writably_mapped(mapping)) | |
1733 | flush_dcache_page(page); | |
1734 | ||
1735 | /* | |
ec0f1637 JK |
1736 | * When a sequential read accesses a page several times, |
1737 | * only mark it as accessed the first time. | |
1da177e4 | 1738 | */ |
ec0f1637 | 1739 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1740 | mark_page_accessed(page); |
1741 | prev_index = index; | |
1742 | ||
1743 | /* | |
1744 | * Ok, we have the page, and it's up-to-date, so | |
1745 | * now we can copy it to user space... | |
1da177e4 | 1746 | */ |
6e58e79d AV |
1747 | |
1748 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 1749 | offset += ret; |
09cbfeaf KS |
1750 | index += offset >> PAGE_SHIFT; |
1751 | offset &= ~PAGE_MASK; | |
6ce745ed | 1752 | prev_offset = offset; |
1da177e4 | 1753 | |
09cbfeaf | 1754 | put_page(page); |
6e58e79d AV |
1755 | written += ret; |
1756 | if (!iov_iter_count(iter)) | |
1757 | goto out; | |
1758 | if (ret < nr) { | |
1759 | error = -EFAULT; | |
1760 | goto out; | |
1761 | } | |
1762 | continue; | |
1da177e4 LT |
1763 | |
1764 | page_not_up_to_date: | |
1765 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1766 | error = lock_page_killable(page); |
1767 | if (unlikely(error)) | |
1768 | goto readpage_error; | |
1da177e4 | 1769 | |
8ab22b9a | 1770 | page_not_up_to_date_locked: |
da6052f7 | 1771 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1772 | if (!page->mapping) { |
1773 | unlock_page(page); | |
09cbfeaf | 1774 | put_page(page); |
1da177e4 LT |
1775 | continue; |
1776 | } | |
1777 | ||
1778 | /* Did somebody else fill it already? */ | |
1779 | if (PageUptodate(page)) { | |
1780 | unlock_page(page); | |
1781 | goto page_ok; | |
1782 | } | |
1783 | ||
1784 | readpage: | |
91803b49 JM |
1785 | /* |
1786 | * A previous I/O error may have been due to temporary | |
1787 | * failures, eg. multipath errors. | |
1788 | * PG_error will be set again if readpage fails. | |
1789 | */ | |
1790 | ClearPageError(page); | |
1da177e4 LT |
1791 | /* Start the actual read. The read will unlock the page. */ |
1792 | error = mapping->a_ops->readpage(filp, page); | |
1793 | ||
994fc28c ZB |
1794 | if (unlikely(error)) { |
1795 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 1796 | put_page(page); |
6e58e79d | 1797 | error = 0; |
994fc28c ZB |
1798 | goto find_page; |
1799 | } | |
1da177e4 | 1800 | goto readpage_error; |
994fc28c | 1801 | } |
1da177e4 LT |
1802 | |
1803 | if (!PageUptodate(page)) { | |
85462323 ON |
1804 | error = lock_page_killable(page); |
1805 | if (unlikely(error)) | |
1806 | goto readpage_error; | |
1da177e4 LT |
1807 | if (!PageUptodate(page)) { |
1808 | if (page->mapping == NULL) { | |
1809 | /* | |
2ecdc82e | 1810 | * invalidate_mapping_pages got it |
1da177e4 LT |
1811 | */ |
1812 | unlock_page(page); | |
09cbfeaf | 1813 | put_page(page); |
1da177e4 LT |
1814 | goto find_page; |
1815 | } | |
1816 | unlock_page(page); | |
7ff81078 | 1817 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1818 | error = -EIO; |
1819 | goto readpage_error; | |
1da177e4 LT |
1820 | } |
1821 | unlock_page(page); | |
1822 | } | |
1823 | ||
1da177e4 LT |
1824 | goto page_ok; |
1825 | ||
1826 | readpage_error: | |
1827 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 1828 | put_page(page); |
1da177e4 LT |
1829 | goto out; |
1830 | ||
1831 | no_cached_page: | |
1832 | /* | |
1833 | * Ok, it wasn't cached, so we need to create a new | |
1834 | * page.. | |
1835 | */ | |
eb2be189 NP |
1836 | page = page_cache_alloc_cold(mapping); |
1837 | if (!page) { | |
6e58e79d | 1838 | error = -ENOMEM; |
eb2be189 | 1839 | goto out; |
1da177e4 | 1840 | } |
6afdb859 | 1841 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 1842 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 1843 | if (error) { |
09cbfeaf | 1844 | put_page(page); |
6e58e79d AV |
1845 | if (error == -EEXIST) { |
1846 | error = 0; | |
1da177e4 | 1847 | goto find_page; |
6e58e79d | 1848 | } |
1da177e4 LT |
1849 | goto out; |
1850 | } | |
1da177e4 LT |
1851 | goto readpage; |
1852 | } | |
1853 | ||
1854 | out: | |
7ff81078 | 1855 | ra->prev_pos = prev_index; |
09cbfeaf | 1856 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 1857 | ra->prev_pos |= prev_offset; |
1da177e4 | 1858 | |
09cbfeaf | 1859 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 1860 | file_accessed(filp); |
6e58e79d | 1861 | return written ? written : error; |
1da177e4 LT |
1862 | } |
1863 | ||
485bb99b | 1864 | /** |
6abd2322 | 1865 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1866 | * @iocb: kernel I/O control block |
6abd2322 | 1867 | * @iter: destination for the data read |
485bb99b | 1868 | * |
6abd2322 | 1869 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1870 | * that can use the page cache directly. |
1871 | */ | |
1872 | ssize_t | |
ed978a81 | 1873 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1874 | { |
ed978a81 | 1875 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1876 | ssize_t retval = 0; |
e7080a43 NS |
1877 | size_t count = iov_iter_count(iter); |
1878 | ||
1879 | if (!count) | |
1880 | goto out; /* skip atime */ | |
1da177e4 | 1881 | |
2ba48ce5 | 1882 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1883 | struct address_space *mapping = file->f_mapping; |
1884 | struct inode *inode = mapping->host; | |
0d5b0cf2 | 1885 | struct iov_iter data = *iter; |
543ade1f | 1886 | loff_t size; |
1da177e4 | 1887 | |
1da177e4 | 1888 | size = i_size_read(inode); |
c64fb5c7 CH |
1889 | retval = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
1890 | iocb->ki_pos + count - 1); | |
0d5b0cf2 CH |
1891 | if (retval < 0) |
1892 | goto out; | |
d8d3d94b | 1893 | |
0d5b0cf2 CH |
1894 | file_accessed(file); |
1895 | ||
1896 | retval = mapping->a_ops->direct_IO(iocb, &data); | |
c3a69024 | 1897 | if (retval >= 0) { |
c64fb5c7 | 1898 | iocb->ki_pos += retval; |
ed978a81 | 1899 | iov_iter_advance(iter, retval); |
9fe55eea | 1900 | } |
66f998f6 | 1901 | |
9fe55eea SW |
1902 | /* |
1903 | * Btrfs can have a short DIO read if we encounter | |
1904 | * compressed extents, so if there was an error, or if | |
1905 | * we've already read everything we wanted to, or if | |
1906 | * there was a short read because we hit EOF, go ahead | |
1907 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1908 | * the rest of the read. Buffered reads will not work for |
1909 | * DAX files, so don't bother trying. | |
9fe55eea | 1910 | */ |
c64fb5c7 | 1911 | if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size || |
0d5b0cf2 | 1912 | IS_DAX(inode)) |
9fe55eea | 1913 | goto out; |
1da177e4 LT |
1914 | } |
1915 | ||
c64fb5c7 | 1916 | retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval); |
1da177e4 LT |
1917 | out: |
1918 | return retval; | |
1919 | } | |
ed978a81 | 1920 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1921 | |
1da177e4 | 1922 | #ifdef CONFIG_MMU |
485bb99b RD |
1923 | /** |
1924 | * page_cache_read - adds requested page to the page cache if not already there | |
1925 | * @file: file to read | |
1926 | * @offset: page index | |
62eb320a | 1927 | * @gfp_mask: memory allocation flags |
485bb99b | 1928 | * |
1da177e4 LT |
1929 | * This adds the requested page to the page cache if it isn't already there, |
1930 | * and schedules an I/O to read in its contents from disk. | |
1931 | */ | |
c20cd45e | 1932 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
1933 | { |
1934 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1935 | struct page *page; |
994fc28c | 1936 | int ret; |
1da177e4 | 1937 | |
994fc28c | 1938 | do { |
c20cd45e | 1939 | page = __page_cache_alloc(gfp_mask|__GFP_COLD); |
994fc28c ZB |
1940 | if (!page) |
1941 | return -ENOMEM; | |
1942 | ||
c20cd45e | 1943 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); |
994fc28c ZB |
1944 | if (ret == 0) |
1945 | ret = mapping->a_ops->readpage(file, page); | |
1946 | else if (ret == -EEXIST) | |
1947 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1948 | |
09cbfeaf | 1949 | put_page(page); |
1da177e4 | 1950 | |
994fc28c | 1951 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1952 | |
994fc28c | 1953 | return ret; |
1da177e4 LT |
1954 | } |
1955 | ||
1956 | #define MMAP_LOTSAMISS (100) | |
1957 | ||
ef00e08e LT |
1958 | /* |
1959 | * Synchronous readahead happens when we don't even find | |
1960 | * a page in the page cache at all. | |
1961 | */ | |
1962 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1963 | struct file_ra_state *ra, | |
1964 | struct file *file, | |
1965 | pgoff_t offset) | |
1966 | { | |
ef00e08e LT |
1967 | struct address_space *mapping = file->f_mapping; |
1968 | ||
1969 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1970 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1971 | return; |
275b12bf WF |
1972 | if (!ra->ra_pages) |
1973 | return; | |
ef00e08e | 1974 | |
64363aad | 1975 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1976 | page_cache_sync_readahead(mapping, ra, file, offset, |
1977 | ra->ra_pages); | |
ef00e08e LT |
1978 | return; |
1979 | } | |
1980 | ||
207d04ba AK |
1981 | /* Avoid banging the cache line if not needed */ |
1982 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1983 | ra->mmap_miss++; |
1984 | ||
1985 | /* | |
1986 | * Do we miss much more than hit in this file? If so, | |
1987 | * stop bothering with read-ahead. It will only hurt. | |
1988 | */ | |
1989 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1990 | return; | |
1991 | ||
d30a1100 WF |
1992 | /* |
1993 | * mmap read-around | |
1994 | */ | |
600e19af RG |
1995 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
1996 | ra->size = ra->ra_pages; | |
1997 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 1998 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1999 | } |
2000 | ||
2001 | /* | |
2002 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2003 | * so we want to possibly extend the readahead further.. | |
2004 | */ | |
2005 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2006 | struct file_ra_state *ra, | |
2007 | struct file *file, | |
2008 | struct page *page, | |
2009 | pgoff_t offset) | |
2010 | { | |
2011 | struct address_space *mapping = file->f_mapping; | |
2012 | ||
2013 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2014 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2015 | return; |
2016 | if (ra->mmap_miss > 0) | |
2017 | ra->mmap_miss--; | |
2018 | if (PageReadahead(page)) | |
2fad6f5d WF |
2019 | page_cache_async_readahead(mapping, ra, file, |
2020 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2021 | } |
2022 | ||
485bb99b | 2023 | /** |
54cb8821 | 2024 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
2025 | * @vma: vma in which the fault was taken |
2026 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 2027 | * |
54cb8821 | 2028 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2029 | * mapped memory region to read in file data during a page fault. |
2030 | * | |
2031 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2032 | * it in the page cache, and handles the special cases reasonably without | |
2033 | * having a lot of duplicated code. | |
9a95f3cf PC |
2034 | * |
2035 | * vma->vm_mm->mmap_sem must be held on entry. | |
2036 | * | |
2037 | * If our return value has VM_FAULT_RETRY set, it's because | |
2038 | * lock_page_or_retry() returned 0. | |
2039 | * The mmap_sem has usually been released in this case. | |
2040 | * See __lock_page_or_retry() for the exception. | |
2041 | * | |
2042 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2043 | * has not been released. | |
2044 | * | |
2045 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2046 | */ |
d0217ac0 | 2047 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2048 | { |
2049 | int error; | |
54cb8821 | 2050 | struct file *file = vma->vm_file; |
1da177e4 LT |
2051 | struct address_space *mapping = file->f_mapping; |
2052 | struct file_ra_state *ra = &file->f_ra; | |
2053 | struct inode *inode = mapping->host; | |
ef00e08e | 2054 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 2055 | struct page *page; |
99e3e53f | 2056 | loff_t size; |
83c54070 | 2057 | int ret = 0; |
1da177e4 | 2058 | |
09cbfeaf KS |
2059 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2060 | if (offset >= size >> PAGE_SHIFT) | |
5307cc1a | 2061 | return VM_FAULT_SIGBUS; |
1da177e4 | 2062 | |
1da177e4 | 2063 | /* |
49426420 | 2064 | * Do we have something in the page cache already? |
1da177e4 | 2065 | */ |
ef00e08e | 2066 | page = find_get_page(mapping, offset); |
45cac65b | 2067 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2068 | /* |
ef00e08e LT |
2069 | * We found the page, so try async readahead before |
2070 | * waiting for the lock. | |
1da177e4 | 2071 | */ |
ef00e08e | 2072 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 2073 | } else if (!page) { |
ef00e08e LT |
2074 | /* No page in the page cache at all */ |
2075 | do_sync_mmap_readahead(vma, ra, file, offset); | |
2076 | count_vm_event(PGMAJFAULT); | |
456f998e | 2077 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2078 | ret = VM_FAULT_MAJOR; |
2079 | retry_find: | |
b522c94d | 2080 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2081 | if (!page) |
2082 | goto no_cached_page; | |
2083 | } | |
2084 | ||
d88c0922 | 2085 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2086 | put_page(page); |
d065bd81 | 2087 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2088 | } |
b522c94d ML |
2089 | |
2090 | /* Did it get truncated? */ | |
2091 | if (unlikely(page->mapping != mapping)) { | |
2092 | unlock_page(page); | |
2093 | put_page(page); | |
2094 | goto retry_find; | |
2095 | } | |
309381fe | 2096 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2097 | |
1da177e4 | 2098 | /* |
d00806b1 NP |
2099 | * We have a locked page in the page cache, now we need to check |
2100 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2101 | */ |
d00806b1 | 2102 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2103 | goto page_not_uptodate; |
2104 | ||
ef00e08e LT |
2105 | /* |
2106 | * Found the page and have a reference on it. | |
2107 | * We must recheck i_size under page lock. | |
2108 | */ | |
09cbfeaf KS |
2109 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2110 | if (unlikely(offset >= size >> PAGE_SHIFT)) { | |
d00806b1 | 2111 | unlock_page(page); |
09cbfeaf | 2112 | put_page(page); |
5307cc1a | 2113 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2114 | } |
2115 | ||
d0217ac0 | 2116 | vmf->page = page; |
83c54070 | 2117 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2118 | |
1da177e4 LT |
2119 | no_cached_page: |
2120 | /* | |
2121 | * We're only likely to ever get here if MADV_RANDOM is in | |
2122 | * effect. | |
2123 | */ | |
c20cd45e | 2124 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2125 | |
2126 | /* | |
2127 | * The page we want has now been added to the page cache. | |
2128 | * In the unlikely event that someone removed it in the | |
2129 | * meantime, we'll just come back here and read it again. | |
2130 | */ | |
2131 | if (error >= 0) | |
2132 | goto retry_find; | |
2133 | ||
2134 | /* | |
2135 | * An error return from page_cache_read can result if the | |
2136 | * system is low on memory, or a problem occurs while trying | |
2137 | * to schedule I/O. | |
2138 | */ | |
2139 | if (error == -ENOMEM) | |
d0217ac0 NP |
2140 | return VM_FAULT_OOM; |
2141 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2142 | |
2143 | page_not_uptodate: | |
1da177e4 LT |
2144 | /* |
2145 | * Umm, take care of errors if the page isn't up-to-date. | |
2146 | * Try to re-read it _once_. We do this synchronously, | |
2147 | * because there really aren't any performance issues here | |
2148 | * and we need to check for errors. | |
2149 | */ | |
1da177e4 | 2150 | ClearPageError(page); |
994fc28c | 2151 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2152 | if (!error) { |
2153 | wait_on_page_locked(page); | |
2154 | if (!PageUptodate(page)) | |
2155 | error = -EIO; | |
2156 | } | |
09cbfeaf | 2157 | put_page(page); |
d00806b1 NP |
2158 | |
2159 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2160 | goto retry_find; |
1da177e4 | 2161 | |
d00806b1 | 2162 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2163 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2164 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2165 | } |
2166 | EXPORT_SYMBOL(filemap_fault); | |
2167 | ||
bae473a4 KS |
2168 | void filemap_map_pages(struct fault_env *fe, |
2169 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
f1820361 KS |
2170 | { |
2171 | struct radix_tree_iter iter; | |
2172 | void **slot; | |
bae473a4 | 2173 | struct file *file = fe->vma->vm_file; |
f1820361 | 2174 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2175 | pgoff_t last_pgoff = start_pgoff; |
f1820361 | 2176 | loff_t size; |
83929372 | 2177 | struct page *head, *page; |
f1820361 KS |
2178 | |
2179 | rcu_read_lock(); | |
bae473a4 KS |
2180 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, |
2181 | start_pgoff) { | |
2182 | if (iter.index > end_pgoff) | |
f1820361 KS |
2183 | break; |
2184 | repeat: | |
2185 | page = radix_tree_deref_slot(slot); | |
2186 | if (unlikely(!page)) | |
2187 | goto next; | |
2188 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2189 | if (radix_tree_deref_retry(page)) { |
2190 | slot = radix_tree_iter_retry(&iter); | |
2191 | continue; | |
2192 | } | |
2193 | goto next; | |
f1820361 KS |
2194 | } |
2195 | ||
83929372 KS |
2196 | head = compound_head(page); |
2197 | if (!page_cache_get_speculative(head)) | |
f1820361 KS |
2198 | goto repeat; |
2199 | ||
83929372 KS |
2200 | /* The page was split under us? */ |
2201 | if (compound_head(page) != head) { | |
2202 | put_page(head); | |
2203 | goto repeat; | |
2204 | } | |
2205 | ||
f1820361 KS |
2206 | /* Has the page moved? */ |
2207 | if (unlikely(page != *slot)) { | |
83929372 | 2208 | put_page(head); |
f1820361 KS |
2209 | goto repeat; |
2210 | } | |
2211 | ||
2212 | if (!PageUptodate(page) || | |
2213 | PageReadahead(page) || | |
2214 | PageHWPoison(page)) | |
2215 | goto skip; | |
2216 | if (!trylock_page(page)) | |
2217 | goto skip; | |
2218 | ||
2219 | if (page->mapping != mapping || !PageUptodate(page)) | |
2220 | goto unlock; | |
2221 | ||
09cbfeaf KS |
2222 | size = round_up(i_size_read(mapping->host), PAGE_SIZE); |
2223 | if (page->index >= size >> PAGE_SHIFT) | |
f1820361 KS |
2224 | goto unlock; |
2225 | ||
f1820361 KS |
2226 | if (file->f_ra.mmap_miss > 0) |
2227 | file->f_ra.mmap_miss--; | |
7267ec00 KS |
2228 | |
2229 | fe->address += (iter.index - last_pgoff) << PAGE_SHIFT; | |
2230 | if (fe->pte) | |
2231 | fe->pte += iter.index - last_pgoff; | |
2232 | last_pgoff = iter.index; | |
2233 | if (alloc_set_pte(fe, NULL, page)) | |
2234 | goto unlock; | |
f1820361 KS |
2235 | unlock_page(page); |
2236 | goto next; | |
2237 | unlock: | |
2238 | unlock_page(page); | |
2239 | skip: | |
09cbfeaf | 2240 | put_page(page); |
f1820361 | 2241 | next: |
7267ec00 KS |
2242 | /* Huge page is mapped? No need to proceed. */ |
2243 | if (pmd_trans_huge(*fe->pmd)) | |
2244 | break; | |
bae473a4 | 2245 | if (iter.index == end_pgoff) |
f1820361 KS |
2246 | break; |
2247 | } | |
2248 | rcu_read_unlock(); | |
2249 | } | |
2250 | EXPORT_SYMBOL(filemap_map_pages); | |
2251 | ||
4fcf1c62 JK |
2252 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2253 | { | |
2254 | struct page *page = vmf->page; | |
496ad9aa | 2255 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2256 | int ret = VM_FAULT_LOCKED; |
2257 | ||
14da9200 | 2258 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2259 | file_update_time(vma->vm_file); |
2260 | lock_page(page); | |
2261 | if (page->mapping != inode->i_mapping) { | |
2262 | unlock_page(page); | |
2263 | ret = VM_FAULT_NOPAGE; | |
2264 | goto out; | |
2265 | } | |
14da9200 JK |
2266 | /* |
2267 | * We mark the page dirty already here so that when freeze is in | |
2268 | * progress, we are guaranteed that writeback during freezing will | |
2269 | * see the dirty page and writeprotect it again. | |
2270 | */ | |
2271 | set_page_dirty(page); | |
1d1d1a76 | 2272 | wait_for_stable_page(page); |
4fcf1c62 | 2273 | out: |
14da9200 | 2274 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2275 | return ret; |
2276 | } | |
2277 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2278 | ||
f0f37e2f | 2279 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2280 | .fault = filemap_fault, |
f1820361 | 2281 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2282 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2283 | }; |
2284 | ||
2285 | /* This is used for a general mmap of a disk file */ | |
2286 | ||
2287 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2288 | { | |
2289 | struct address_space *mapping = file->f_mapping; | |
2290 | ||
2291 | if (!mapping->a_ops->readpage) | |
2292 | return -ENOEXEC; | |
2293 | file_accessed(file); | |
2294 | vma->vm_ops = &generic_file_vm_ops; | |
2295 | return 0; | |
2296 | } | |
1da177e4 LT |
2297 | |
2298 | /* | |
2299 | * This is for filesystems which do not implement ->writepage. | |
2300 | */ | |
2301 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2302 | { | |
2303 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2304 | return -EINVAL; | |
2305 | return generic_file_mmap(file, vma); | |
2306 | } | |
2307 | #else | |
2308 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2309 | { | |
2310 | return -ENOSYS; | |
2311 | } | |
2312 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2313 | { | |
2314 | return -ENOSYS; | |
2315 | } | |
2316 | #endif /* CONFIG_MMU */ | |
2317 | ||
2318 | EXPORT_SYMBOL(generic_file_mmap); | |
2319 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2320 | ||
67f9fd91 SL |
2321 | static struct page *wait_on_page_read(struct page *page) |
2322 | { | |
2323 | if (!IS_ERR(page)) { | |
2324 | wait_on_page_locked(page); | |
2325 | if (!PageUptodate(page)) { | |
09cbfeaf | 2326 | put_page(page); |
67f9fd91 SL |
2327 | page = ERR_PTR(-EIO); |
2328 | } | |
2329 | } | |
2330 | return page; | |
2331 | } | |
2332 | ||
32b63529 | 2333 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2334 | pgoff_t index, |
5e5358e7 | 2335 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2336 | void *data, |
2337 | gfp_t gfp) | |
1da177e4 | 2338 | { |
eb2be189 | 2339 | struct page *page; |
1da177e4 LT |
2340 | int err; |
2341 | repeat: | |
2342 | page = find_get_page(mapping, index); | |
2343 | if (!page) { | |
0531b2aa | 2344 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2345 | if (!page) |
2346 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2347 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2348 | if (unlikely(err)) { |
09cbfeaf | 2349 | put_page(page); |
eb2be189 NP |
2350 | if (err == -EEXIST) |
2351 | goto repeat; | |
1da177e4 | 2352 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2353 | return ERR_PTR(err); |
2354 | } | |
32b63529 MG |
2355 | |
2356 | filler: | |
1da177e4 LT |
2357 | err = filler(data, page); |
2358 | if (err < 0) { | |
09cbfeaf | 2359 | put_page(page); |
32b63529 | 2360 | return ERR_PTR(err); |
1da177e4 | 2361 | } |
1da177e4 | 2362 | |
32b63529 MG |
2363 | page = wait_on_page_read(page); |
2364 | if (IS_ERR(page)) | |
2365 | return page; | |
2366 | goto out; | |
2367 | } | |
1da177e4 LT |
2368 | if (PageUptodate(page)) |
2369 | goto out; | |
2370 | ||
ebded027 MG |
2371 | /* |
2372 | * Page is not up to date and may be locked due one of the following | |
2373 | * case a: Page is being filled and the page lock is held | |
2374 | * case b: Read/write error clearing the page uptodate status | |
2375 | * case c: Truncation in progress (page locked) | |
2376 | * case d: Reclaim in progress | |
2377 | * | |
2378 | * Case a, the page will be up to date when the page is unlocked. | |
2379 | * There is no need to serialise on the page lock here as the page | |
2380 | * is pinned so the lock gives no additional protection. Even if the | |
2381 | * the page is truncated, the data is still valid if PageUptodate as | |
2382 | * it's a race vs truncate race. | |
2383 | * Case b, the page will not be up to date | |
2384 | * Case c, the page may be truncated but in itself, the data may still | |
2385 | * be valid after IO completes as it's a read vs truncate race. The | |
2386 | * operation must restart if the page is not uptodate on unlock but | |
2387 | * otherwise serialising on page lock to stabilise the mapping gives | |
2388 | * no additional guarantees to the caller as the page lock is | |
2389 | * released before return. | |
2390 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2391 | * will be a race with remove_mapping that determines if the mapping | |
2392 | * is valid on unlock but otherwise the data is valid and there is | |
2393 | * no need to serialise with page lock. | |
2394 | * | |
2395 | * As the page lock gives no additional guarantee, we optimistically | |
2396 | * wait on the page to be unlocked and check if it's up to date and | |
2397 | * use the page if it is. Otherwise, the page lock is required to | |
2398 | * distinguish between the different cases. The motivation is that we | |
2399 | * avoid spurious serialisations and wakeups when multiple processes | |
2400 | * wait on the same page for IO to complete. | |
2401 | */ | |
2402 | wait_on_page_locked(page); | |
2403 | if (PageUptodate(page)) | |
2404 | goto out; | |
2405 | ||
2406 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2407 | lock_page(page); |
ebded027 MG |
2408 | |
2409 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2410 | if (!page->mapping) { |
2411 | unlock_page(page); | |
09cbfeaf | 2412 | put_page(page); |
32b63529 | 2413 | goto repeat; |
1da177e4 | 2414 | } |
ebded027 MG |
2415 | |
2416 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2417 | if (PageUptodate(page)) { |
2418 | unlock_page(page); | |
2419 | goto out; | |
2420 | } | |
32b63529 MG |
2421 | goto filler; |
2422 | ||
c855ff37 | 2423 | out: |
6fe6900e NP |
2424 | mark_page_accessed(page); |
2425 | return page; | |
2426 | } | |
0531b2aa LT |
2427 | |
2428 | /** | |
67f9fd91 | 2429 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2430 | * @mapping: the page's address_space |
2431 | * @index: the page index | |
2432 | * @filler: function to perform the read | |
5e5358e7 | 2433 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2434 | * |
0531b2aa | 2435 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2436 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2437 | * |
2438 | * If the page does not get brought uptodate, return -EIO. | |
2439 | */ | |
67f9fd91 | 2440 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2441 | pgoff_t index, |
5e5358e7 | 2442 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2443 | void *data) |
2444 | { | |
2445 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2446 | } | |
67f9fd91 | 2447 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2448 | |
2449 | /** | |
2450 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2451 | * @mapping: the page's address_space | |
2452 | * @index: the page index | |
2453 | * @gfp: the page allocator flags to use if allocating | |
2454 | * | |
2455 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2456 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2457 | * |
2458 | * If the page does not get brought uptodate, return -EIO. | |
2459 | */ | |
2460 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2461 | pgoff_t index, | |
2462 | gfp_t gfp) | |
2463 | { | |
2464 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2465 | ||
67f9fd91 | 2466 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2467 | } |
2468 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2469 | ||
1da177e4 LT |
2470 | /* |
2471 | * Performs necessary checks before doing a write | |
2472 | * | |
485bb99b | 2473 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2474 | * Returns appropriate error code that caller should return or |
2475 | * zero in case that write should be allowed. | |
2476 | */ | |
3309dd04 | 2477 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2478 | { |
3309dd04 | 2479 | struct file *file = iocb->ki_filp; |
1da177e4 | 2480 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2481 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2482 | loff_t pos; |
1da177e4 | 2483 | |
3309dd04 AV |
2484 | if (!iov_iter_count(from)) |
2485 | return 0; | |
1da177e4 | 2486 | |
0fa6b005 | 2487 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2488 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2489 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2490 | |
3309dd04 | 2491 | pos = iocb->ki_pos; |
1da177e4 | 2492 | |
0fa6b005 | 2493 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2494 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2495 | send_sig(SIGXFSZ, current, 0); |
2496 | return -EFBIG; | |
1da177e4 | 2497 | } |
3309dd04 | 2498 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2499 | } |
2500 | ||
2501 | /* | |
2502 | * LFS rule | |
2503 | */ | |
3309dd04 | 2504 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2505 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2506 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2507 | return -EFBIG; |
3309dd04 | 2508 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2509 | } |
2510 | ||
2511 | /* | |
2512 | * Are we about to exceed the fs block limit ? | |
2513 | * | |
2514 | * If we have written data it becomes a short write. If we have | |
2515 | * exceeded without writing data we send a signal and return EFBIG. | |
2516 | * Linus frestrict idea will clean these up nicely.. | |
2517 | */ | |
3309dd04 AV |
2518 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2519 | return -EFBIG; | |
1da177e4 | 2520 | |
3309dd04 AV |
2521 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2522 | return iov_iter_count(from); | |
1da177e4 LT |
2523 | } |
2524 | EXPORT_SYMBOL(generic_write_checks); | |
2525 | ||
afddba49 NP |
2526 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2527 | loff_t pos, unsigned len, unsigned flags, | |
2528 | struct page **pagep, void **fsdata) | |
2529 | { | |
2530 | const struct address_space_operations *aops = mapping->a_ops; | |
2531 | ||
4e02ed4b | 2532 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2533 | pagep, fsdata); |
afddba49 NP |
2534 | } |
2535 | EXPORT_SYMBOL(pagecache_write_begin); | |
2536 | ||
2537 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2538 | loff_t pos, unsigned len, unsigned copied, | |
2539 | struct page *page, void *fsdata) | |
2540 | { | |
2541 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2542 | |
4e02ed4b | 2543 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2544 | } |
2545 | EXPORT_SYMBOL(pagecache_write_end); | |
2546 | ||
1da177e4 | 2547 | ssize_t |
1af5bb49 | 2548 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2549 | { |
2550 | struct file *file = iocb->ki_filp; | |
2551 | struct address_space *mapping = file->f_mapping; | |
2552 | struct inode *inode = mapping->host; | |
1af5bb49 | 2553 | loff_t pos = iocb->ki_pos; |
1da177e4 | 2554 | ssize_t written; |
a969e903 CH |
2555 | size_t write_len; |
2556 | pgoff_t end; | |
26978b8b | 2557 | struct iov_iter data; |
1da177e4 | 2558 | |
0c949334 | 2559 | write_len = iov_iter_count(from); |
09cbfeaf | 2560 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2561 | |
48b47c56 | 2562 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2563 | if (written) |
2564 | goto out; | |
2565 | ||
2566 | /* | |
2567 | * After a write we want buffered reads to be sure to go to disk to get | |
2568 | * the new data. We invalidate clean cached page from the region we're | |
2569 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2570 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2571 | */ |
2572 | if (mapping->nrpages) { | |
2573 | written = invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2574 | pos >> PAGE_SHIFT, end); |
6ccfa806 HH |
2575 | /* |
2576 | * If a page can not be invalidated, return 0 to fall back | |
2577 | * to buffered write. | |
2578 | */ | |
2579 | if (written) { | |
2580 | if (written == -EBUSY) | |
2581 | return 0; | |
a969e903 | 2582 | goto out; |
6ccfa806 | 2583 | } |
a969e903 CH |
2584 | } |
2585 | ||
26978b8b | 2586 | data = *from; |
c8b8e32d | 2587 | written = mapping->a_ops->direct_IO(iocb, &data); |
a969e903 CH |
2588 | |
2589 | /* | |
2590 | * Finally, try again to invalidate clean pages which might have been | |
2591 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2592 | * if the source of the write was an mmap'ed region of the file | |
2593 | * we're writing. Either one is a pretty crazy thing to do, | |
2594 | * so we don't support it 100%. If this invalidation | |
2595 | * fails, tough, the write still worked... | |
2596 | */ | |
2597 | if (mapping->nrpages) { | |
2598 | invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2599 | pos >> PAGE_SHIFT, end); |
a969e903 CH |
2600 | } |
2601 | ||
1da177e4 | 2602 | if (written > 0) { |
0116651c | 2603 | pos += written; |
f8579f86 | 2604 | iov_iter_advance(from, written); |
0116651c NK |
2605 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2606 | i_size_write(inode, pos); | |
1da177e4 LT |
2607 | mark_inode_dirty(inode); |
2608 | } | |
5cb6c6c7 | 2609 | iocb->ki_pos = pos; |
1da177e4 | 2610 | } |
a969e903 | 2611 | out: |
1da177e4 LT |
2612 | return written; |
2613 | } | |
2614 | EXPORT_SYMBOL(generic_file_direct_write); | |
2615 | ||
eb2be189 NP |
2616 | /* |
2617 | * Find or create a page at the given pagecache position. Return the locked | |
2618 | * page. This function is specifically for buffered writes. | |
2619 | */ | |
54566b2c NP |
2620 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2621 | pgoff_t index, unsigned flags) | |
eb2be189 | 2622 | { |
eb2be189 | 2623 | struct page *page; |
bbddabe2 | 2624 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2625 | |
54566b2c | 2626 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2627 | fgp_flags |= FGP_NOFS; |
2628 | ||
2629 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2630 | mapping_gfp_mask(mapping)); |
c585a267 | 2631 | if (page) |
2457aec6 | 2632 | wait_for_stable_page(page); |
eb2be189 | 2633 | |
eb2be189 NP |
2634 | return page; |
2635 | } | |
54566b2c | 2636 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2637 | |
3b93f911 | 2638 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2639 | struct iov_iter *i, loff_t pos) |
2640 | { | |
2641 | struct address_space *mapping = file->f_mapping; | |
2642 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2643 | long status = 0; | |
2644 | ssize_t written = 0; | |
674b892e NP |
2645 | unsigned int flags = 0; |
2646 | ||
2647 | /* | |
2648 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2649 | */ | |
777eda2c | 2650 | if (!iter_is_iovec(i)) |
674b892e | 2651 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2652 | |
2653 | do { | |
2654 | struct page *page; | |
afddba49 NP |
2655 | unsigned long offset; /* Offset into pagecache page */ |
2656 | unsigned long bytes; /* Bytes to write to page */ | |
2657 | size_t copied; /* Bytes copied from user */ | |
2658 | void *fsdata; | |
2659 | ||
09cbfeaf KS |
2660 | offset = (pos & (PAGE_SIZE - 1)); |
2661 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
2662 | iov_iter_count(i)); |
2663 | ||
2664 | again: | |
00a3d660 LT |
2665 | /* |
2666 | * Bring in the user page that we will copy from _first_. | |
2667 | * Otherwise there's a nasty deadlock on copying from the | |
2668 | * same page as we're writing to, without it being marked | |
2669 | * up-to-date. | |
2670 | * | |
2671 | * Not only is this an optimisation, but it is also required | |
2672 | * to check that the address is actually valid, when atomic | |
2673 | * usercopies are used, below. | |
2674 | */ | |
2675 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2676 | status = -EFAULT; | |
2677 | break; | |
2678 | } | |
2679 | ||
296291cd JK |
2680 | if (fatal_signal_pending(current)) { |
2681 | status = -EINTR; | |
2682 | break; | |
2683 | } | |
2684 | ||
674b892e | 2685 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2686 | &page, &fsdata); |
2457aec6 | 2687 | if (unlikely(status < 0)) |
afddba49 NP |
2688 | break; |
2689 | ||
931e80e4 | 2690 | if (mapping_writably_mapped(mapping)) |
2691 | flush_dcache_page(page); | |
00a3d660 | 2692 | |
afddba49 | 2693 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2694 | flush_dcache_page(page); |
2695 | ||
2696 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2697 | page, fsdata); | |
2698 | if (unlikely(status < 0)) | |
2699 | break; | |
2700 | copied = status; | |
2701 | ||
2702 | cond_resched(); | |
2703 | ||
124d3b70 | 2704 | iov_iter_advance(i, copied); |
afddba49 NP |
2705 | if (unlikely(copied == 0)) { |
2706 | /* | |
2707 | * If we were unable to copy any data at all, we must | |
2708 | * fall back to a single segment length write. | |
2709 | * | |
2710 | * If we didn't fallback here, we could livelock | |
2711 | * because not all segments in the iov can be copied at | |
2712 | * once without a pagefault. | |
2713 | */ | |
09cbfeaf | 2714 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
2715 | iov_iter_single_seg_count(i)); |
2716 | goto again; | |
2717 | } | |
afddba49 NP |
2718 | pos += copied; |
2719 | written += copied; | |
2720 | ||
2721 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
2722 | } while (iov_iter_count(i)); |
2723 | ||
2724 | return written ? written : status; | |
2725 | } | |
3b93f911 | 2726 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2727 | |
e4dd9de3 | 2728 | /** |
8174202b | 2729 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2730 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2731 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2732 | * |
2733 | * This function does all the work needed for actually writing data to a | |
2734 | * file. It does all basic checks, removes SUID from the file, updates | |
2735 | * modification times and calls proper subroutines depending on whether we | |
2736 | * do direct IO or a standard buffered write. | |
2737 | * | |
2738 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2739 | * object which does not need locking at all. | |
2740 | * | |
2741 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2742 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2743 | * avoid syncing under i_mutex. | |
2744 | */ | |
8174202b | 2745 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2746 | { |
2747 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2748 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2749 | struct inode *inode = mapping->host; |
3b93f911 | 2750 | ssize_t written = 0; |
1da177e4 | 2751 | ssize_t err; |
3b93f911 | 2752 | ssize_t status; |
1da177e4 | 2753 | |
1da177e4 | 2754 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2755 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 2756 | err = file_remove_privs(file); |
1da177e4 LT |
2757 | if (err) |
2758 | goto out; | |
2759 | ||
c3b2da31 JB |
2760 | err = file_update_time(file); |
2761 | if (err) | |
2762 | goto out; | |
1da177e4 | 2763 | |
2ba48ce5 | 2764 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2765 | loff_t pos, endbyte; |
fb5527e6 | 2766 | |
1af5bb49 | 2767 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 2768 | /* |
fbbbad4b MW |
2769 | * If the write stopped short of completing, fall back to |
2770 | * buffered writes. Some filesystems do this for writes to | |
2771 | * holes, for example. For DAX files, a buffered write will | |
2772 | * not succeed (even if it did, DAX does not handle dirty | |
2773 | * page-cache pages correctly). | |
1da177e4 | 2774 | */ |
0b8def9d | 2775 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2776 | goto out; |
2777 | ||
0b8def9d | 2778 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2779 | /* |
3b93f911 | 2780 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2781 | * then we want to return the number of bytes which were |
2782 | * direct-written, or the error code if that was zero. Note | |
2783 | * that this differs from normal direct-io semantics, which | |
2784 | * will return -EFOO even if some bytes were written. | |
2785 | */ | |
60bb4529 | 2786 | if (unlikely(status < 0)) { |
3b93f911 | 2787 | err = status; |
fb5527e6 JM |
2788 | goto out; |
2789 | } | |
fb5527e6 JM |
2790 | /* |
2791 | * We need to ensure that the page cache pages are written to | |
2792 | * disk and invalidated to preserve the expected O_DIRECT | |
2793 | * semantics. | |
2794 | */ | |
3b93f911 | 2795 | endbyte = pos + status - 1; |
0b8def9d | 2796 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2797 | if (err == 0) { |
0b8def9d | 2798 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2799 | written += status; |
fb5527e6 | 2800 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
2801 | pos >> PAGE_SHIFT, |
2802 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
2803 | } else { |
2804 | /* | |
2805 | * We don't know how much we wrote, so just return | |
2806 | * the number of bytes which were direct-written | |
2807 | */ | |
2808 | } | |
2809 | } else { | |
0b8def9d AV |
2810 | written = generic_perform_write(file, from, iocb->ki_pos); |
2811 | if (likely(written > 0)) | |
2812 | iocb->ki_pos += written; | |
fb5527e6 | 2813 | } |
1da177e4 LT |
2814 | out: |
2815 | current->backing_dev_info = NULL; | |
2816 | return written ? written : err; | |
2817 | } | |
8174202b | 2818 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2819 | |
e4dd9de3 | 2820 | /** |
8174202b | 2821 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2822 | * @iocb: IO state structure |
8174202b | 2823 | * @from: iov_iter with data to write |
e4dd9de3 | 2824 | * |
8174202b | 2825 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2826 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2827 | * and acquires i_mutex as needed. | |
2828 | */ | |
8174202b | 2829 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2830 | { |
2831 | struct file *file = iocb->ki_filp; | |
148f948b | 2832 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2833 | ssize_t ret; |
1da177e4 | 2834 | |
5955102c | 2835 | inode_lock(inode); |
3309dd04 AV |
2836 | ret = generic_write_checks(iocb, from); |
2837 | if (ret > 0) | |
5f380c7f | 2838 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 2839 | inode_unlock(inode); |
1da177e4 | 2840 | |
e2592217 CH |
2841 | if (ret > 0) |
2842 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
2843 | return ret; |
2844 | } | |
8174202b | 2845 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2846 | |
cf9a2ae8 DH |
2847 | /** |
2848 | * try_to_release_page() - release old fs-specific metadata on a page | |
2849 | * | |
2850 | * @page: the page which the kernel is trying to free | |
2851 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2852 | * | |
2853 | * The address_space is to try to release any data against the page | |
2854 | * (presumably at page->private). If the release was successful, return `1'. | |
2855 | * Otherwise return zero. | |
2856 | * | |
266cf658 DH |
2857 | * This may also be called if PG_fscache is set on a page, indicating that the |
2858 | * page is known to the local caching routines. | |
2859 | * | |
cf9a2ae8 | 2860 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 2861 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 2862 | * |
cf9a2ae8 DH |
2863 | */ |
2864 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2865 | { | |
2866 | struct address_space * const mapping = page->mapping; | |
2867 | ||
2868 | BUG_ON(!PageLocked(page)); | |
2869 | if (PageWriteback(page)) | |
2870 | return 0; | |
2871 | ||
2872 | if (mapping && mapping->a_ops->releasepage) | |
2873 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2874 | return try_to_free_buffers(page); | |
2875 | } | |
2876 | ||
2877 | EXPORT_SYMBOL(try_to_release_page); |