1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/mm/filemap.c
4 *
5 * Copyright (C) 1994-1999 Linus Torvalds
6 */
7
8 /*
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
12 */
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
16 #include <linux/fs.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
22 #include <linux/mm.h>
23 #include <linux/swap.h>
24 #include <linux/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include "internal.h"
45
46 #define CREATE_TRACE_POINTS
47 #include <trace/events/filemap.h>
48
49 /*
50 * FIXME: remove all knowledge of the buffer layer from the core VM
51 */
52 #include <linux/buffer_head.h> /* for try_to_free_buffers */
53
54 #include <asm/mman.h>
55
56 /*
57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
58 * though.
59 *
60 * Shared mappings now work. 15.8.1995 Bruno.
61 *
62 * finished 'unifying' the page and buffer cache and SMP-threaded the
63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
64 *
65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
66 */
67
68 /*
69 * Lock ordering:
70 *
71 * ->i_mmap_rwsem (truncate_pagecache)
72 * ->private_lock (__free_pte->__set_page_dirty_buffers)
73 * ->swap_lock (exclusive_swap_page, others)
74 * ->i_pages lock
75 *
76 * ->i_mutex
77 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
78 *
79 * ->mmap_sem
80 * ->i_mmap_rwsem
81 * ->page_table_lock or pte_lock (various, mainly in memory.c)
82 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
83 *
84 * ->mmap_sem
85 * ->lock_page (access_process_vm)
86 *
87 * ->i_mutex (generic_perform_write)
88 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
89 *
90 * bdi->wb.list_lock
91 * sb_lock (fs/fs-writeback.c)
92 * ->i_pages lock (__sync_single_inode)
93 *
94 * ->i_mmap_rwsem
95 * ->anon_vma.lock (vma_adjust)
96 *
97 * ->anon_vma.lock
98 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
99 *
100 * ->page_table_lock or pte_lock
101 * ->swap_lock (try_to_unmap_one)
102 * ->private_lock (try_to_unmap_one)
103 * ->i_pages lock (try_to_unmap_one)
104 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
105 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
106 * ->private_lock (page_remove_rmap->set_page_dirty)
107 * ->i_pages lock (page_remove_rmap->set_page_dirty)
108 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
109 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
110 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
111 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
112 * ->inode->i_lock (zap_pte_range->set_page_dirty)
113 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
114 *
115 * ->i_mmap_rwsem
116 * ->tasklist_lock (memory_failure, collect_procs_ao)
117 */
118
119 static void page_cache_delete(struct address_space *mapping,
120 struct page *page, void *shadow)
121 {
122 XA_STATE(xas, &mapping->i_pages, page->index);
123 unsigned int nr = 1;
124
125 mapping_set_update(&xas, mapping);
126
127 /* hugetlb pages are represented by a single entry in the xarray */
128 if (!PageHuge(page)) {
129 xas_set_order(&xas, page->index, compound_order(page));
130 nr = compound_nr(page);
131 }
132
133 VM_BUG_ON_PAGE(!PageLocked(page), page);
134 VM_BUG_ON_PAGE(PageTail(page), page);
135 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
136
137 xas_store(&xas, shadow);
138 xas_init_marks(&xas);
139
140 page->mapping = NULL;
141 /* Leave page->index set: truncation lookup relies upon it */
142
143 if (shadow) {
144 mapping->nrexceptional += nr;
145 /*
146 * Make sure the nrexceptional update is committed before
147 * the nrpages update so that final truncate racing
148 * with reclaim does not see both counters 0 at the
149 * same time and miss a shadow entry.
150 */
151 smp_wmb();
152 }
153 mapping->nrpages -= nr;
154 }
155
156 static void unaccount_page_cache_page(struct address_space *mapping,
157 struct page *page)
158 {
159 int nr;
160
161 /*
162 * if we're uptodate, flush out into the cleancache, otherwise
163 * invalidate any existing cleancache entries. We can't leave
164 * stale data around in the cleancache once our page is gone
165 */
166 if (PageUptodate(page) && PageMappedToDisk(page))
167 cleancache_put_page(page);
168 else
169 cleancache_invalidate_page(mapping, page);
170
171 VM_BUG_ON_PAGE(PageTail(page), page);
172 VM_BUG_ON_PAGE(page_mapped(page), page);
173 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
174 int mapcount;
175
176 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
177 current->comm, page_to_pfn(page));
178 dump_page(page, "still mapped when deleted");
179 dump_stack();
180 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
181
182 mapcount = page_mapcount(page);
183 if (mapping_exiting(mapping) &&
184 page_count(page) >= mapcount + 2) {
185 /*
186 * All vmas have already been torn down, so it's
187 * a good bet that actually the page is unmapped,
188 * and we'd prefer not to leak it: if we're wrong,
189 * some other bad page check should catch it later.
190 */
191 page_mapcount_reset(page);
192 page_ref_sub(page, mapcount);
193 }
194 }
195
196 /* hugetlb pages do not participate in page cache accounting. */
197 if (PageHuge(page))
198 return;
199
200 nr = hpage_nr_pages(page);
201
202 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
203 if (PageSwapBacked(page)) {
204 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
205 if (PageTransHuge(page))
206 __dec_node_page_state(page, NR_SHMEM_THPS);
207 } else if (PageTransHuge(page)) {
208 __dec_node_page_state(page, NR_FILE_THPS);
209 filemap_nr_thps_dec(mapping);
210 }
211
212 /*
213 * At this point page must be either written or cleaned by
214 * truncate. Dirty page here signals a bug and loss of
215 * unwritten data.
216 *
217 * This fixes dirty accounting after removing the page entirely
218 * but leaves PageDirty set: it has no effect for truncated
219 * page and anyway will be cleared before returning page into
220 * buddy allocator.
221 */
222 if (WARN_ON_ONCE(PageDirty(page)))
223 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
224 }
225
226 /*
227 * Delete a page from the page cache and free it. Caller has to make
228 * sure the page is locked and that nobody else uses it - or that usage
229 * is safe. The caller must hold the i_pages lock.
230 */
231 void __delete_from_page_cache(struct page *page, void *shadow)
232 {
233 struct address_space *mapping = page->mapping;
234
235 trace_mm_filemap_delete_from_page_cache(page);
236
237 unaccount_page_cache_page(mapping, page);
238 page_cache_delete(mapping, page, shadow);
239 }
240
241 static void page_cache_free_page(struct address_space *mapping,
242 struct page *page)
243 {
244 void (*freepage)(struct page *);
245
246 freepage = mapping->a_ops->freepage;
247 if (freepage)
248 freepage(page);
249
250 if (PageTransHuge(page) && !PageHuge(page)) {
251 page_ref_sub(page, HPAGE_PMD_NR);
252 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
253 } else {
254 put_page(page);
255 }
256 }
257
258 /**
259 * delete_from_page_cache - delete page from page cache
260 * @page: the page which the kernel is trying to remove from page cache
261 *
262 * This must be called only on pages that have been verified to be in the page
263 * cache and locked. It will never put the page into the free list, the caller
264 * has a reference on the page.
265 */
266 void delete_from_page_cache(struct page *page)
267 {
268 struct address_space *mapping = page_mapping(page);
269 unsigned long flags;
270
271 BUG_ON(!PageLocked(page));
272 xa_lock_irqsave(&mapping->i_pages, flags);
273 __delete_from_page_cache(page, NULL);
274 xa_unlock_irqrestore(&mapping->i_pages, flags);
275
276 page_cache_free_page(mapping, page);
277 }
278 EXPORT_SYMBOL(delete_from_page_cache);
279
280 /*
281 * page_cache_delete_batch - delete several pages from page cache
282 * @mapping: the mapping to which pages belong
283 * @pvec: pagevec with pages to delete
284 *
285 * The function walks over mapping->i_pages and removes pages passed in @pvec
286 * from the mapping. The function expects @pvec to be sorted by page index
287 * and is optimised for it to be dense.
288 * It tolerates holes in @pvec (mapping entries at those indices are not
289 * modified). The function expects only THP head pages to be present in the
290 * @pvec.
291 *
292 * The function expects the i_pages lock to be held.
293 */
294 static void page_cache_delete_batch(struct address_space *mapping,
295 struct pagevec *pvec)
296 {
297 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
298 int total_pages = 0;
299 int i = 0;
300 struct page *page;
301
302 mapping_set_update(&xas, mapping);
303 xas_for_each(&xas, page, ULONG_MAX) {
304 if (i >= pagevec_count(pvec))
305 break;
306
307 /* A swap/dax/shadow entry got inserted? Skip it. */
308 if (xa_is_value(page))
309 continue;
310 /*
311 * A page got inserted in our range? Skip it. We have our
312 * pages locked so they are protected from being removed.
313 * If we see a page whose index is higher than ours, it
314 * means our page has been removed, which shouldn't be
315 * possible because we're holding the PageLock.
316 */
317 if (page != pvec->pages[i]) {
318 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
319 page);
320 continue;
321 }
322
323 WARN_ON_ONCE(!PageLocked(page));
324
325 if (page->index == xas.xa_index)
326 page->mapping = NULL;
327 /* Leave page->index set: truncation lookup relies on it */
328
329 /*
330 * Move to the next page in the vector if this is a regular
331 * page or the index is of the last sub-page of this compound
332 * page.
333 */
334 if (page->index + compound_nr(page) - 1 == xas.xa_index)
335 i++;
336 xas_store(&xas, NULL);
337 total_pages++;
338 }
339 mapping->nrpages -= total_pages;
340 }
341
342 void delete_from_page_cache_batch(struct address_space *mapping,
343 struct pagevec *pvec)
344 {
345 int i;
346 unsigned long flags;
347
348 if (!pagevec_count(pvec))
349 return;
350
351 xa_lock_irqsave(&mapping->i_pages, flags);
352 for (i = 0; i < pagevec_count(pvec); i++) {
353 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
354
355 unaccount_page_cache_page(mapping, pvec->pages[i]);
356 }
357 page_cache_delete_batch(mapping, pvec);
358 xa_unlock_irqrestore(&mapping->i_pages, flags);
359
360 for (i = 0; i < pagevec_count(pvec); i++)
361 page_cache_free_page(mapping, pvec->pages[i]);
362 }
363
364 int filemap_check_errors(struct address_space *mapping)
365 {
366 int ret = 0;
367 /* Check for outstanding write errors */
368 if (test_bit(AS_ENOSPC, &mapping->flags) &&
369 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
370 ret = -ENOSPC;
371 if (test_bit(AS_EIO, &mapping->flags) &&
372 test_and_clear_bit(AS_EIO, &mapping->flags))
373 ret = -EIO;
374 return ret;
375 }
376 EXPORT_SYMBOL(filemap_check_errors);
377
378 static int filemap_check_and_keep_errors(struct address_space *mapping)
379 {
380 /* Check for outstanding write errors */
381 if (test_bit(AS_EIO, &mapping->flags))
382 return -EIO;
383 if (test_bit(AS_ENOSPC, &mapping->flags))
384 return -ENOSPC;
385 return 0;
386 }
387
388 /**
389 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
390 * @mapping: address space structure to write
391 * @start: offset in bytes where the range starts
392 * @end: offset in bytes where the range ends (inclusive)
393 * @sync_mode: enable synchronous operation
394 *
395 * Start writeback against all of a mapping's dirty pages that lie
396 * within the byte offsets <start, end> inclusive.
397 *
398 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
399 * opposed to a regular memory cleansing writeback. The difference between
400 * these two operations is that if a dirty page/buffer is encountered, it must
401 * be waited upon, and not just skipped over.
402 *
403 * Return: %0 on success, negative error code otherwise.
404 */
405 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
406 loff_t end, int sync_mode)
407 {
408 int ret;
409 struct writeback_control wbc = {
410 .sync_mode = sync_mode,
411 .nr_to_write = LONG_MAX,
412 .range_start = start,
413 .range_end = end,
414 };
415
416 if (!mapping_cap_writeback_dirty(mapping) ||
417 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
418 return 0;
419
420 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
421 ret = do_writepages(mapping, &wbc);
422 wbc_detach_inode(&wbc);
423 return ret;
424 }
425
426 static inline int __filemap_fdatawrite(struct address_space *mapping,
427 int sync_mode)
428 {
429 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
430 }
431
432 int filemap_fdatawrite(struct address_space *mapping)
433 {
434 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435 }
436 EXPORT_SYMBOL(filemap_fdatawrite);
437
438 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
439 loff_t end)
440 {
441 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442 }
443 EXPORT_SYMBOL(filemap_fdatawrite_range);
444
445 /**
446 * filemap_flush - mostly a non-blocking flush
447 * @mapping: target address_space
448 *
449 * This is a mostly non-blocking flush. Not suitable for data-integrity
450 * purposes - I/O may not be started against all dirty pages.
451 *
452 * Return: %0 on success, negative error code otherwise.
453 */
454 int filemap_flush(struct address_space *mapping)
455 {
456 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457 }
458 EXPORT_SYMBOL(filemap_flush);
459
460 /**
461 * filemap_range_has_page - check if a page exists in range.
462 * @mapping: address space within which to check
463 * @start_byte: offset in bytes where the range starts
464 * @end_byte: offset in bytes where the range ends (inclusive)
465 *
466 * Find at least one page in the range supplied, usually used to check if
467 * direct writing in this range will trigger a writeback.
468 *
469 * Return: %true if at least one page exists in the specified range,
470 * %false otherwise.
471 */
472 bool filemap_range_has_page(struct address_space *mapping,
473 loff_t start_byte, loff_t end_byte)
474 {
475 struct page *page;
476 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477 pgoff_t max = end_byte >> PAGE_SHIFT;
478
479 if (end_byte < start_byte)
480 return false;
481
482 rcu_read_lock();
483 for (;;) {
484 page = xas_find(&xas, max);
485 if (xas_retry(&xas, page))
486 continue;
487 /* Shadow entries don't count */
488 if (xa_is_value(page))
489 continue;
490 /*
491 * We don't need to try to pin this page; we're about to
492 * release the RCU lock anyway. It is enough to know that
493 * there was a page here recently.
494 */
495 break;
496 }
497 rcu_read_unlock();
498
499 return page != NULL;
500 }
501 EXPORT_SYMBOL(filemap_range_has_page);
502
503 static void __filemap_fdatawait_range(struct address_space *mapping,
504 loff_t start_byte, loff_t end_byte)
505 {
506 pgoff_t index = start_byte >> PAGE_SHIFT;
507 pgoff_t end = end_byte >> PAGE_SHIFT;
508 struct pagevec pvec;
509 int nr_pages;
510
511 if (end_byte < start_byte)
512 return;
513
514 pagevec_init(&pvec);
515 while (index <= end) {
516 unsigned i;
517
518 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
519 end, PAGECACHE_TAG_WRITEBACK);
520 if (!nr_pages)
521 break;
522
523 for (i = 0; i < nr_pages; i++) {
524 struct page *page = pvec.pages[i];
525
526 wait_on_page_writeback(page);
527 ClearPageError(page);
528 }
529 pagevec_release(&pvec);
530 cond_resched();
531 }
532 }
533
534 /**
535 * filemap_fdatawait_range - wait for writeback to complete
536 * @mapping: address space structure to wait for
537 * @start_byte: offset in bytes where the range starts
538 * @end_byte: offset in bytes where the range ends (inclusive)
539 *
540 * Walk the list of under-writeback pages of the given address space
541 * in the given range and wait for all of them. Check error status of
542 * the address space and return it.
543 *
544 * Since the error status of the address space is cleared by this function,
545 * callers are responsible for checking the return value and handling and/or
546 * reporting the error.
547 *
548 * Return: error status of the address space.
549 */
550 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
551 loff_t end_byte)
552 {
553 __filemap_fdatawait_range(mapping, start_byte, end_byte);
554 return filemap_check_errors(mapping);
555 }
556 EXPORT_SYMBOL(filemap_fdatawait_range);
557
558 /**
559 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
560 * @mapping: address space structure to wait for
561 * @start_byte: offset in bytes where the range starts
562 * @end_byte: offset in bytes where the range ends (inclusive)
563 *
564 * Walk the list of under-writeback pages of the given address space in the
565 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
566 * this function does not clear error status of the address space.
567 *
568 * Use this function if callers don't handle errors themselves. Expected
569 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
570 * fsfreeze(8)
571 */
572 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
573 loff_t start_byte, loff_t end_byte)
574 {
575 __filemap_fdatawait_range(mapping, start_byte, end_byte);
576 return filemap_check_and_keep_errors(mapping);
577 }
578 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
579
580 /**
581 * file_fdatawait_range - wait for writeback to complete
582 * @file: file pointing to address space structure to wait for
583 * @start_byte: offset in bytes where the range starts
584 * @end_byte: offset in bytes where the range ends (inclusive)
585 *
586 * Walk the list of under-writeback pages of the address space that file
587 * refers to, in the given range and wait for all of them. Check error
588 * status of the address space vs. the file->f_wb_err cursor and return it.
589 *
590 * Since the error status of the file is advanced by this function,
591 * callers are responsible for checking the return value and handling and/or
592 * reporting the error.
593 *
594 * Return: error status of the address space vs. the file->f_wb_err cursor.
595 */
596 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
597 {
598 struct address_space *mapping = file->f_mapping;
599
600 __filemap_fdatawait_range(mapping, start_byte, end_byte);
601 return file_check_and_advance_wb_err(file);
602 }
603 EXPORT_SYMBOL(file_fdatawait_range);
604
605 /**
606 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
607 * @mapping: address space structure to wait for
608 *
609 * Walk the list of under-writeback pages of the given address space
610 * and wait for all of them. Unlike filemap_fdatawait(), this function
611 * does not clear error status of the address space.
612 *
613 * Use this function if callers don't handle errors themselves. Expected
614 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
615 * fsfreeze(8)
616 *
617 * Return: error status of the address space.
618 */
619 int filemap_fdatawait_keep_errors(struct address_space *mapping)
620 {
621 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
622 return filemap_check_and_keep_errors(mapping);
623 }
624 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
625
626 /* Returns true if writeback might be needed or already in progress. */
627 static bool mapping_needs_writeback(struct address_space *mapping)
628 {
629 if (dax_mapping(mapping))
630 return mapping->nrexceptional;
631
632 return mapping->nrpages;
633 }
634
635 int filemap_write_and_wait(struct address_space *mapping)
636 {
637 int err = 0;
638
639 if (mapping_needs_writeback(mapping)) {
640 err = filemap_fdatawrite(mapping);
641 /*
642 * Even if the above returned error, the pages may be
643 * written partially (e.g. -ENOSPC), so we wait for it.
644 * But the -EIO is special case, it may indicate the worst
645 * thing (e.g. bug) happened, so we avoid waiting for it.
646 */
647 if (err != -EIO) {
648 int err2 = filemap_fdatawait(mapping);
649 if (!err)
650 err = err2;
651 } else {
652 /* Clear any previously stored errors */
653 filemap_check_errors(mapping);
654 }
655 } else {
656 err = filemap_check_errors(mapping);
657 }
658 return err;
659 }
660 EXPORT_SYMBOL(filemap_write_and_wait);
661
662 /**
663 * filemap_write_and_wait_range - write out & wait on a file range
664 * @mapping: the address_space for the pages
665 * @lstart: offset in bytes where the range starts
666 * @lend: offset in bytes where the range ends (inclusive)
667 *
668 * Write out and wait upon file offsets lstart->lend, inclusive.
669 *
670 * Note that @lend is inclusive (describes the last byte to be written) so
671 * that this function can be used to write to the very end-of-file (end = -1).
672 *
673 * Return: error status of the address space.
674 */
675 int filemap_write_and_wait_range(struct address_space *mapping,
676 loff_t lstart, loff_t lend)
677 {
678 int err = 0;
679
680 if (mapping_needs_writeback(mapping)) {
681 err = __filemap_fdatawrite_range(mapping, lstart, lend,
682 WB_SYNC_ALL);
683 /* See comment of filemap_write_and_wait() */
684 if (err != -EIO) {
685 int err2 = filemap_fdatawait_range(mapping,
686 lstart, lend);
687 if (!err)
688 err = err2;
689 } else {
690 /* Clear any previously stored errors */
691 filemap_check_errors(mapping);
692 }
693 } else {
694 err = filemap_check_errors(mapping);
695 }
696 return err;
697 }
698 EXPORT_SYMBOL(filemap_write_and_wait_range);
699
700 void __filemap_set_wb_err(struct address_space *mapping, int err)
701 {
702 errseq_t eseq = errseq_set(&mapping->wb_err, err);
703
704 trace_filemap_set_wb_err(mapping, eseq);
705 }
706 EXPORT_SYMBOL(__filemap_set_wb_err);
707
708 /**
709 * file_check_and_advance_wb_err - report wb error (if any) that was previously
710 * and advance wb_err to current one
711 * @file: struct file on which the error is being reported
712 *
713 * When userland calls fsync (or something like nfsd does the equivalent), we
714 * want to report any writeback errors that occurred since the last fsync (or
715 * since the file was opened if there haven't been any).
716 *
717 * Grab the wb_err from the mapping. If it matches what we have in the file,
718 * then just quickly return 0. The file is all caught up.
719 *
720 * If it doesn't match, then take the mapping value, set the "seen" flag in
721 * it and try to swap it into place. If it works, or another task beat us
722 * to it with the new value, then update the f_wb_err and return the error
723 * portion. The error at this point must be reported via proper channels
724 * (a'la fsync, or NFS COMMIT operation, etc.).
725 *
726 * While we handle mapping->wb_err with atomic operations, the f_wb_err
727 * value is protected by the f_lock since we must ensure that it reflects
728 * the latest value swapped in for this file descriptor.
729 *
730 * Return: %0 on success, negative error code otherwise.
731 */
732 int file_check_and_advance_wb_err(struct file *file)
733 {
734 int err = 0;
735 errseq_t old = READ_ONCE(file->f_wb_err);
736 struct address_space *mapping = file->f_mapping;
737
738 /* Locklessly handle the common case where nothing has changed */
739 if (errseq_check(&mapping->wb_err, old)) {
740 /* Something changed, must use slow path */
741 spin_lock(&file->f_lock);
742 old = file->f_wb_err;
743 err = errseq_check_and_advance(&mapping->wb_err,
744 &file->f_wb_err);
745 trace_file_check_and_advance_wb_err(file, old);
746 spin_unlock(&file->f_lock);
747 }
748
749 /*
750 * We're mostly using this function as a drop in replacement for
751 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
752 * that the legacy code would have had on these flags.
753 */
754 clear_bit(AS_EIO, &mapping->flags);
755 clear_bit(AS_ENOSPC, &mapping->flags);
756 return err;
757 }
758 EXPORT_SYMBOL(file_check_and_advance_wb_err);
759
760 /**
761 * file_write_and_wait_range - write out & wait on a file range
762 * @file: file pointing to address_space with pages
763 * @lstart: offset in bytes where the range starts
764 * @lend: offset in bytes where the range ends (inclusive)
765 *
766 * Write out and wait upon file offsets lstart->lend, inclusive.
767 *
768 * Note that @lend is inclusive (describes the last byte to be written) so
769 * that this function can be used to write to the very end-of-file (end = -1).
770 *
771 * After writing out and waiting on the data, we check and advance the
772 * f_wb_err cursor to the latest value, and return any errors detected there.
773 *
774 * Return: %0 on success, negative error code otherwise.
775 */
776 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
777 {
778 int err = 0, err2;
779 struct address_space *mapping = file->f_mapping;
780
781 if (mapping_needs_writeback(mapping)) {
782 err = __filemap_fdatawrite_range(mapping, lstart, lend,
783 WB_SYNC_ALL);
784 /* See comment of filemap_write_and_wait() */
785 if (err != -EIO)
786 __filemap_fdatawait_range(mapping, lstart, lend);
787 }
788 err2 = file_check_and_advance_wb_err(file);
789 if (!err)
790 err = err2;
791 return err;
792 }
793 EXPORT_SYMBOL(file_write_and_wait_range);
794
795 /**
796 * replace_page_cache_page - replace a pagecache page with a new one
797 * @old: page to be replaced
798 * @new: page to replace with
799 * @gfp_mask: allocation mode
800 *
801 * This function replaces a page in the pagecache with a new one. On
802 * success it acquires the pagecache reference for the new page and
803 * drops it for the old page. Both the old and new pages must be
804 * locked. This function does not add the new page to the LRU, the
805 * caller must do that.
806 *
807 * The remove + add is atomic. This function cannot fail.
808 *
809 * Return: %0
810 */
811 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
812 {
813 struct address_space *mapping = old->mapping;
814 void (*freepage)(struct page *) = mapping->a_ops->freepage;
815 pgoff_t offset = old->index;
816 XA_STATE(xas, &mapping->i_pages, offset);
817 unsigned long flags;
818
819 VM_BUG_ON_PAGE(!PageLocked(old), old);
820 VM_BUG_ON_PAGE(!PageLocked(new), new);
821 VM_BUG_ON_PAGE(new->mapping, new);
822
823 get_page(new);
824 new->mapping = mapping;
825 new->index = offset;
826
827 xas_lock_irqsave(&xas, flags);
828 xas_store(&xas, new);
829
830 old->mapping = NULL;
831 /* hugetlb pages do not participate in page cache accounting. */
832 if (!PageHuge(old))
833 __dec_node_page_state(new, NR_FILE_PAGES);
834 if (!PageHuge(new))
835 __inc_node_page_state(new, NR_FILE_PAGES);
836 if (PageSwapBacked(old))
837 __dec_node_page_state(new, NR_SHMEM);
838 if (PageSwapBacked(new))
839 __inc_node_page_state(new, NR_SHMEM);
840 xas_unlock_irqrestore(&xas, flags);
841 mem_cgroup_migrate(old, new);
842 if (freepage)
843 freepage(old);
844 put_page(old);
845
846 return 0;
847 }
848 EXPORT_SYMBOL_GPL(replace_page_cache_page);
849
850 static int __add_to_page_cache_locked(struct page *page,
851 struct address_space *mapping,
852 pgoff_t offset, gfp_t gfp_mask,
853 void **shadowp)
854 {
855 XA_STATE(xas, &mapping->i_pages, offset);
856 int huge = PageHuge(page);
857 struct mem_cgroup *memcg;
858 int error;
859 void *old;
860
861 VM_BUG_ON_PAGE(!PageLocked(page), page);
862 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
863 mapping_set_update(&xas, mapping);
864
865 if (!huge) {
866 error = mem_cgroup_try_charge(page, current->mm,
867 gfp_mask, &memcg, false);
868 if (error)
869 return error;
870 }
871
872 get_page(page);
873 page->mapping = mapping;
874 page->index = offset;
875
876 do {
877 xas_lock_irq(&xas);
878 old = xas_load(&xas);
879 if (old && !xa_is_value(old))
880 xas_set_err(&xas, -EEXIST);
881 xas_store(&xas, page);
882 if (xas_error(&xas))
883 goto unlock;
884
885 if (xa_is_value(old)) {
886 mapping->nrexceptional--;
887 if (shadowp)
888 *shadowp = old;
889 }
890 mapping->nrpages++;
891
892 /* hugetlb pages do not participate in page cache accounting */
893 if (!huge)
894 __inc_node_page_state(page, NR_FILE_PAGES);
895 unlock:
896 xas_unlock_irq(&xas);
897 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
898
899 if (xas_error(&xas))
900 goto error;
901
902 if (!huge)
903 mem_cgroup_commit_charge(page, memcg, false, false);
904 trace_mm_filemap_add_to_page_cache(page);
905 return 0;
906 error:
907 page->mapping = NULL;
908 /* Leave page->index set: truncation relies upon it */
909 if (!huge)
910 mem_cgroup_cancel_charge(page, memcg, false);
911 put_page(page);
912 return xas_error(&xas);
913 }
914 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
915
916 /**
917 * add_to_page_cache_locked - add a locked page to the pagecache
918 * @page: page to add
919 * @mapping: the page's address_space
920 * @offset: page index
921 * @gfp_mask: page allocation mode
922 *
923 * This function is used to add a page to the pagecache. It must be locked.
924 * This function does not add the page to the LRU. The caller must do that.
925 *
926 * Return: %0 on success, negative error code otherwise.
927 */
928 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
929 pgoff_t offset, gfp_t gfp_mask)
930 {
931 return __add_to_page_cache_locked(page, mapping, offset,
932 gfp_mask, NULL);
933 }
934 EXPORT_SYMBOL(add_to_page_cache_locked);
935
936 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
937 pgoff_t offset, gfp_t gfp_mask)
938 {
939 void *shadow = NULL;
940 int ret;
941
942 __SetPageLocked(page);
943 ret = __add_to_page_cache_locked(page, mapping, offset,
944 gfp_mask, &shadow);
945 if (unlikely(ret))
946 __ClearPageLocked(page);
947 else {
948 /*
949 * The page might have been evicted from cache only
950 * recently, in which case it should be activated like
951 * any other repeatedly accessed page.
952 * The exception is pages getting rewritten; evicting other
953 * data from the working set, only to cache data that will
954 * get overwritten with something else, is a waste of memory.
955 */
956 WARN_ON_ONCE(PageActive(page));
957 if (!(gfp_mask & __GFP_WRITE) && shadow)
958 workingset_refault(page, shadow);
959 lru_cache_add(page);
960 }
961 return ret;
962 }
963 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
964
965 #ifdef CONFIG_NUMA
966 struct page *__page_cache_alloc(gfp_t gfp)
967 {
968 int n;
969 struct page *page;
970
971 if (cpuset_do_page_mem_spread()) {
972 unsigned int cpuset_mems_cookie;
973 do {
974 cpuset_mems_cookie = read_mems_allowed_begin();
975 n = cpuset_mem_spread_node();
976 page = __alloc_pages_node(n, gfp, 0);
977 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
978
979 return page;
980 }
981 return alloc_pages(gfp, 0);
982 }
983 EXPORT_SYMBOL(__page_cache_alloc);
984 #endif
985
986 /*
987 * In order to wait for pages to become available there must be
988 * waitqueues associated with pages. By using a hash table of
989 * waitqueues where the bucket discipline is to maintain all
990 * waiters on the same queue and wake all when any of the pages
991 * become available, and for the woken contexts to check to be
992 * sure the appropriate page became available, this saves space
993 * at a cost of "thundering herd" phenomena during rare hash
994 * collisions.
995 */
996 #define PAGE_WAIT_TABLE_BITS 8
997 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
998 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
999
1000 static wait_queue_head_t *page_waitqueue(struct page *page)
1001 {
1002 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1003 }
1004
1005 void __init pagecache_init(void)
1006 {
1007 int i;
1008
1009 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1010 init_waitqueue_head(&page_wait_table[i]);
1011
1012 page_writeback_init();
1013 }
1014
1015 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1016 struct wait_page_key {
1017 struct page *page;
1018 int bit_nr;
1019 int page_match;
1020 };
1021
1022 struct wait_page_queue {
1023 struct page *page;
1024 int bit_nr;
1025 wait_queue_entry_t wait;
1026 };
1027
1028 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1029 {
1030 struct wait_page_key *key = arg;
1031 struct wait_page_queue *wait_page
1032 = container_of(wait, struct wait_page_queue, wait);
1033
1034 if (wait_page->page != key->page)
1035 return 0;
1036 key->page_match = 1;
1037
1038 if (wait_page->bit_nr != key->bit_nr)
1039 return 0;
1040
1041 /*
1042 * Stop walking if it's locked.
1043 * Is this safe if put_and_wait_on_page_locked() is in use?
1044 * Yes: the waker must hold a reference to this page, and if PG_locked
1045 * has now already been set by another task, that task must also hold
1046 * a reference to the *same usage* of this page; so there is no need
1047 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1048 */
1049 if (test_bit(key->bit_nr, &key->page->flags))
1050 return -1;
1051
1052 return autoremove_wake_function(wait, mode, sync, key);
1053 }
1054
1055 static void wake_up_page_bit(struct page *page, int bit_nr)
1056 {
1057 wait_queue_head_t *q = page_waitqueue(page);
1058 struct wait_page_key key;
1059 unsigned long flags;
1060 wait_queue_entry_t bookmark;
1061
1062 key.page = page;
1063 key.bit_nr = bit_nr;
1064 key.page_match = 0;
1065
1066 bookmark.flags = 0;
1067 bookmark.private = NULL;
1068 bookmark.func = NULL;
1069 INIT_LIST_HEAD(&bookmark.entry);
1070
1071 spin_lock_irqsave(&q->lock, flags);
1072 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1073
1074 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1075 /*
1076 * Take a breather from holding the lock,
1077 * allow pages that finish wake up asynchronously
1078 * to acquire the lock and remove themselves
1079 * from wait queue
1080 */
1081 spin_unlock_irqrestore(&q->lock, flags);
1082 cpu_relax();
1083 spin_lock_irqsave(&q->lock, flags);
1084 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1085 }
1086
1087 /*
1088 * It is possible for other pages to have collided on the waitqueue
1089 * hash, so in that case check for a page match. That prevents a long-
1090 * term waiter
1091 *
1092 * It is still possible to miss a case here, when we woke page waiters
1093 * and removed them from the waitqueue, but there are still other
1094 * page waiters.
1095 */
1096 if (!waitqueue_active(q) || !key.page_match) {
1097 ClearPageWaiters(page);
1098 /*
1099 * It's possible to miss clearing Waiters here, when we woke
1100 * our page waiters, but the hashed waitqueue has waiters for
1101 * other pages on it.
1102 *
1103 * That's okay, it's a rare case. The next waker will clear it.
1104 */
1105 }
1106 spin_unlock_irqrestore(&q->lock, flags);
1107 }
1108
1109 static void wake_up_page(struct page *page, int bit)
1110 {
1111 if (!PageWaiters(page))
1112 return;
1113 wake_up_page_bit(page, bit);
1114 }
1115
1116 /*
1117 * A choice of three behaviors for wait_on_page_bit_common():
1118 */
1119 enum behavior {
1120 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1121 * __lock_page() waiting on then setting PG_locked.
1122 */
1123 SHARED, /* Hold ref to page and check the bit when woken, like
1124 * wait_on_page_writeback() waiting on PG_writeback.
1125 */
1126 DROP, /* Drop ref to page before wait, no check when woken,
1127 * like put_and_wait_on_page_locked() on PG_locked.
1128 */
1129 };
1130
1131 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1132 struct page *page, int bit_nr, int state, enum behavior behavior)
1133 {
1134 struct wait_page_queue wait_page;
1135 wait_queue_entry_t *wait = &wait_page.wait;
1136 bool bit_is_set;
1137 bool thrashing = false;
1138 bool delayacct = false;
1139 unsigned long pflags;
1140 int ret = 0;
1141
1142 if (bit_nr == PG_locked &&
1143 !PageUptodate(page) && PageWorkingset(page)) {
1144 if (!PageSwapBacked(page)) {
1145 delayacct_thrashing_start();
1146 delayacct = true;
1147 }
1148 psi_memstall_enter(&pflags);
1149 thrashing = true;
1150 }
1151
1152 init_wait(wait);
1153 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1154 wait->func = wake_page_function;
1155 wait_page.page = page;
1156 wait_page.bit_nr = bit_nr;
1157
1158 for (;;) {
1159 spin_lock_irq(&q->lock);
1160
1161 if (likely(list_empty(&wait->entry))) {
1162 __add_wait_queue_entry_tail(q, wait);
1163 SetPageWaiters(page);
1164 }
1165
1166 set_current_state(state);
1167
1168 spin_unlock_irq(&q->lock);
1169
1170 bit_is_set = test_bit(bit_nr, &page->flags);
1171 if (behavior == DROP)
1172 put_page(page);
1173
1174 if (likely(bit_is_set))
1175 io_schedule();
1176
1177 if (behavior == EXCLUSIVE) {
1178 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1179 break;
1180 } else if (behavior == SHARED) {
1181 if (!test_bit(bit_nr, &page->flags))
1182 break;
1183 }
1184
1185 if (signal_pending_state(state, current)) {
1186 ret = -EINTR;
1187 break;
1188 }
1189
1190 if (behavior == DROP) {
1191 /*
1192 * We can no longer safely access page->flags:
1193 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1194 * there is a risk of waiting forever on a page reused
1195 * for something that keeps it locked indefinitely.
1196 * But best check for -EINTR above before breaking.
1197 */
1198 break;
1199 }
1200 }
1201
1202 finish_wait(q, wait);
1203
1204 if (thrashing) {
1205 if (delayacct)
1206 delayacct_thrashing_end();
1207 psi_memstall_leave(&pflags);
1208 }
1209
1210 /*
1211 * A signal could leave PageWaiters set. Clearing it here if
1212 * !waitqueue_active would be possible (by open-coding finish_wait),
1213 * but still fail to catch it in the case of wait hash collision. We
1214 * already can fail to clear wait hash collision cases, so don't
1215 * bother with signals either.
1216 */
1217
1218 return ret;
1219 }
1220
1221 void wait_on_page_bit(struct page *page, int bit_nr)
1222 {
1223 wait_queue_head_t *q = page_waitqueue(page);
1224 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1225 }
1226 EXPORT_SYMBOL(wait_on_page_bit);
1227
1228 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1229 {
1230 wait_queue_head_t *q = page_waitqueue(page);
1231 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1232 }
1233 EXPORT_SYMBOL(wait_on_page_bit_killable);
1234
1235 /**
1236 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1237 * @page: The page to wait for.
1238 *
1239 * The caller should hold a reference on @page. They expect the page to
1240 * become unlocked relatively soon, but do not wish to hold up migration
1241 * (for example) by holding the reference while waiting for the page to
1242 * come unlocked. After this function returns, the caller should not
1243 * dereference @page.
1244 */
1245 void put_and_wait_on_page_locked(struct page *page)
1246 {
1247 wait_queue_head_t *q;
1248
1249 page = compound_head(page);
1250 q = page_waitqueue(page);
1251 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1252 }
1253
1254 /**
1255 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1256 * @page: Page defining the wait queue of interest
1257 * @waiter: Waiter to add to the queue
1258 *
1259 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1260 */
1261 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1262 {
1263 wait_queue_head_t *q = page_waitqueue(page);
1264 unsigned long flags;
1265
1266 spin_lock_irqsave(&q->lock, flags);
1267 __add_wait_queue_entry_tail(q, waiter);
1268 SetPageWaiters(page);
1269 spin_unlock_irqrestore(&q->lock, flags);
1270 }
1271 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1272
1273 #ifndef clear_bit_unlock_is_negative_byte
1274
1275 /*
1276 * PG_waiters is the high bit in the same byte as PG_lock.
1277 *
1278 * On x86 (and on many other architectures), we can clear PG_lock and
1279 * test the sign bit at the same time. But if the architecture does
1280 * not support that special operation, we just do this all by hand
1281 * instead.
1282 *
1283 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1284 * being cleared, but a memory barrier should be unneccssary since it is
1285 * in the same byte as PG_locked.
1286 */
1287 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1288 {
1289 clear_bit_unlock(nr, mem);
1290 /* smp_mb__after_atomic(); */
1291 return test_bit(PG_waiters, mem);
1292 }
1293
1294 #endif
1295
1296 /**
1297 * unlock_page - unlock a locked page
1298 * @page: the page
1299 *
1300 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1301 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1302 * mechanism between PageLocked pages and PageWriteback pages is shared.
1303 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1304 *
1305 * Note that this depends on PG_waiters being the sign bit in the byte
1306 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1307 * clear the PG_locked bit and test PG_waiters at the same time fairly
1308 * portably (architectures that do LL/SC can test any bit, while x86 can
1309 * test the sign bit).
1310 */
1311 void unlock_page(struct page *page)
1312 {
1313 BUILD_BUG_ON(PG_waiters != 7);
1314 page = compound_head(page);
1315 VM_BUG_ON_PAGE(!PageLocked(page), page);
1316 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1317 wake_up_page_bit(page, PG_locked);
1318 }
1319 EXPORT_SYMBOL(unlock_page);
1320
1321 /**
1322 * end_page_writeback - end writeback against a page
1323 * @page: the page
1324 */
1325 void end_page_writeback(struct page *page)
1326 {
1327 /*
1328 * TestClearPageReclaim could be used here but it is an atomic
1329 * operation and overkill in this particular case. Failing to
1330 * shuffle a page marked for immediate reclaim is too mild to
1331 * justify taking an atomic operation penalty at the end of
1332 * ever page writeback.
1333 */
1334 if (PageReclaim(page)) {
1335 ClearPageReclaim(page);
1336 rotate_reclaimable_page(page);
1337 }
1338
1339 if (!test_clear_page_writeback(page))
1340 BUG();
1341
1342 smp_mb__after_atomic();
1343 wake_up_page(page, PG_writeback);
1344 }
1345 EXPORT_SYMBOL(end_page_writeback);
1346
1347 /*
1348 * After completing I/O on a page, call this routine to update the page
1349 * flags appropriately
1350 */
1351 void page_endio(struct page *page, bool is_write, int err)
1352 {
1353 if (!is_write) {
1354 if (!err) {
1355 SetPageUptodate(page);
1356 } else {
1357 ClearPageUptodate(page);
1358 SetPageError(page);
1359 }
1360 unlock_page(page);
1361 } else {
1362 if (err) {
1363 struct address_space *mapping;
1364
1365 SetPageError(page);
1366 mapping = page_mapping(page);
1367 if (mapping)
1368 mapping_set_error(mapping, err);
1369 }
1370 end_page_writeback(page);
1371 }
1372 }
1373 EXPORT_SYMBOL_GPL(page_endio);
1374
1375 /**
1376 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1377 * @__page: the page to lock
1378 */
1379 void __lock_page(struct page *__page)
1380 {
1381 struct page *page = compound_head(__page);
1382 wait_queue_head_t *q = page_waitqueue(page);
1383 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1384 EXCLUSIVE);
1385 }
1386 EXPORT_SYMBOL(__lock_page);
1387
1388 int __lock_page_killable(struct page *__page)
1389 {
1390 struct page *page = compound_head(__page);
1391 wait_queue_head_t *q = page_waitqueue(page);
1392 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1393 EXCLUSIVE);
1394 }
1395 EXPORT_SYMBOL_GPL(__lock_page_killable);
1396
1397 /*
1398 * Return values:
1399 * 1 - page is locked; mmap_sem is still held.
1400 * 0 - page is not locked.
1401 * mmap_sem has been released (up_read()), unless flags had both
1402 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1403 * which case mmap_sem is still held.
1404 *
1405 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1406 * with the page locked and the mmap_sem unperturbed.
1407 */
1408 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1409 unsigned int flags)
1410 {
1411 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1412 /*
1413 * CAUTION! In this case, mmap_sem is not released
1414 * even though return 0.
1415 */
1416 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1417 return 0;
1418
1419 up_read(&mm->mmap_sem);
1420 if (flags & FAULT_FLAG_KILLABLE)
1421 wait_on_page_locked_killable(page);
1422 else
1423 wait_on_page_locked(page);
1424 return 0;
1425 } else {
1426 if (flags & FAULT_FLAG_KILLABLE) {
1427 int ret;
1428
1429 ret = __lock_page_killable(page);
1430 if (ret) {
1431 up_read(&mm->mmap_sem);
1432 return 0;
1433 }
1434 } else
1435 __lock_page(page);
1436 return 1;
1437 }
1438 }
1439
1440 /**
1441 * page_cache_next_miss() - Find the next gap in the page cache.
1442 * @mapping: Mapping.
1443 * @index: Index.
1444 * @max_scan: Maximum range to search.
1445 *
1446 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1447 * gap with the lowest index.
1448 *
1449 * This function may be called under the rcu_read_lock. However, this will
1450 * not atomically search a snapshot of the cache at a single point in time.
1451 * For example, if a gap is created at index 5, then subsequently a gap is
1452 * created at index 10, page_cache_next_miss covering both indices may
1453 * return 10 if called under the rcu_read_lock.
1454 *
1455 * Return: The index of the gap if found, otherwise an index outside the
1456 * range specified (in which case 'return - index >= max_scan' will be true).
1457 * In the rare case of index wrap-around, 0 will be returned.
1458 */
1459 pgoff_t page_cache_next_miss(struct address_space *mapping,
1460 pgoff_t index, unsigned long max_scan)
1461 {
1462 XA_STATE(xas, &mapping->i_pages, index);
1463
1464 while (max_scan--) {
1465 void *entry = xas_next(&xas);
1466 if (!entry || xa_is_value(entry))
1467 break;
1468 if (xas.xa_index == 0)
1469 break;
1470 }
1471
1472 return xas.xa_index;
1473 }
1474 EXPORT_SYMBOL(page_cache_next_miss);
1475
1476 /**
1477 * page_cache_prev_miss() - Find the previous gap in the page cache.
1478 * @mapping: Mapping.
1479 * @index: Index.
1480 * @max_scan: Maximum range to search.
1481 *
1482 * Search the range [max(index - max_scan + 1, 0), index] for the
1483 * gap with the highest index.
1484 *
1485 * This function may be called under the rcu_read_lock. However, this will
1486 * not atomically search a snapshot of the cache at a single point in time.
1487 * For example, if a gap is created at index 10, then subsequently a gap is
1488 * created at index 5, page_cache_prev_miss() covering both indices may
1489 * return 5 if called under the rcu_read_lock.
1490 *
1491 * Return: The index of the gap if found, otherwise an index outside the
1492 * range specified (in which case 'index - return >= max_scan' will be true).
1493 * In the rare case of wrap-around, ULONG_MAX will be returned.
1494 */
1495 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1496 pgoff_t index, unsigned long max_scan)
1497 {
1498 XA_STATE(xas, &mapping->i_pages, index);
1499
1500 while (max_scan--) {
1501 void *entry = xas_prev(&xas);
1502 if (!entry || xa_is_value(entry))
1503 break;
1504 if (xas.xa_index == ULONG_MAX)
1505 break;
1506 }
1507
1508 return xas.xa_index;
1509 }
1510 EXPORT_SYMBOL(page_cache_prev_miss);
1511
1512 /**
1513 * find_get_entry - find and get a page cache entry
1514 * @mapping: the address_space to search
1515 * @offset: the page cache index
1516 *
1517 * Looks up the page cache slot at @mapping & @offset. If there is a
1518 * page cache page, it is returned with an increased refcount.
1519 *
1520 * If the slot holds a shadow entry of a previously evicted page, or a
1521 * swap entry from shmem/tmpfs, it is returned.
1522 *
1523 * Return: the found page or shadow entry, %NULL if nothing is found.
1524 */
1525 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1526 {
1527 XA_STATE(xas, &mapping->i_pages, offset);
1528 struct page *page;
1529
1530 rcu_read_lock();
1531 repeat:
1532 xas_reset(&xas);
1533 page = xas_load(&xas);
1534 if (xas_retry(&xas, page))
1535 goto repeat;
1536 /*
1537 * A shadow entry of a recently evicted page, or a swap entry from
1538 * shmem/tmpfs. Return it without attempting to raise page count.
1539 */
1540 if (!page || xa_is_value(page))
1541 goto out;
1542
1543 if (!page_cache_get_speculative(page))
1544 goto repeat;
1545
1546 /*
1547 * Has the page moved or been split?
1548 * This is part of the lockless pagecache protocol. See
1549 * include/linux/pagemap.h for details.
1550 */
1551 if (unlikely(page != xas_reload(&xas))) {
1552 put_page(page);
1553 goto repeat;
1554 }
1555 page = find_subpage(page, offset);
1556 out:
1557 rcu_read_unlock();
1558
1559 return page;
1560 }
1561 EXPORT_SYMBOL(find_get_entry);
1562
1563 /**
1564 * find_lock_entry - locate, pin and lock a page cache entry
1565 * @mapping: the address_space to search
1566 * @offset: the page cache index
1567 *
1568 * Looks up the page cache slot at @mapping & @offset. If there is a
1569 * page cache page, it is returned locked and with an increased
1570 * refcount.
1571 *
1572 * If the slot holds a shadow entry of a previously evicted page, or a
1573 * swap entry from shmem/tmpfs, it is returned.
1574 *
1575 * find_lock_entry() may sleep.
1576 *
1577 * Return: the found page or shadow entry, %NULL if nothing is found.
1578 */
1579 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1580 {
1581 struct page *page;
1582
1583 repeat:
1584 page = find_get_entry(mapping, offset);
1585 if (page && !xa_is_value(page)) {
1586 lock_page(page);
1587 /* Has the page been truncated? */
1588 if (unlikely(page_mapping(page) != mapping)) {
1589 unlock_page(page);
1590 put_page(page);
1591 goto repeat;
1592 }
1593 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1594 }
1595 return page;
1596 }
1597 EXPORT_SYMBOL(find_lock_entry);
1598
1599 /**
1600 * pagecache_get_page - find and get a page reference
1601 * @mapping: the address_space to search
1602 * @offset: the page index
1603 * @fgp_flags: PCG flags
1604 * @gfp_mask: gfp mask to use for the page cache data page allocation
1605 *
1606 * Looks up the page cache slot at @mapping & @offset.
1607 *
1608 * PCG flags modify how the page is returned.
1609 *
1610 * @fgp_flags can be:
1611 *
1612 * - FGP_ACCESSED: the page will be marked accessed
1613 * - FGP_LOCK: Page is return locked
1614 * - FGP_CREAT: If page is not present then a new page is allocated using
1615 * @gfp_mask and added to the page cache and the VM's LRU
1616 * list. The page is returned locked and with an increased
1617 * refcount.
1618 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1619 * its own locking dance if the page is already in cache, or unlock the page
1620 * before returning if we had to add the page to pagecache.
1621 *
1622 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1623 * if the GFP flags specified for FGP_CREAT are atomic.
1624 *
1625 * If there is a page cache page, it is returned with an increased refcount.
1626 *
1627 * Return: the found page or %NULL otherwise.
1628 */
1629 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1630 int fgp_flags, gfp_t gfp_mask)
1631 {
1632 struct page *page;
1633
1634 repeat:
1635 page = find_get_entry(mapping, offset);
1636 if (xa_is_value(page))
1637 page = NULL;
1638 if (!page)
1639 goto no_page;
1640
1641 if (fgp_flags & FGP_LOCK) {
1642 if (fgp_flags & FGP_NOWAIT) {
1643 if (!trylock_page(page)) {
1644 put_page(page);
1645 return NULL;
1646 }
1647 } else {
1648 lock_page(page);
1649 }
1650
1651 /* Has the page been truncated? */
1652 if (unlikely(compound_head(page)->mapping != mapping)) {
1653 unlock_page(page);
1654 put_page(page);
1655 goto repeat;
1656 }
1657 VM_BUG_ON_PAGE(page->index != offset, page);
1658 }
1659
1660 if (fgp_flags & FGP_ACCESSED)
1661 mark_page_accessed(page);
1662
1663 no_page:
1664 if (!page && (fgp_flags & FGP_CREAT)) {
1665 int err;
1666 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1667 gfp_mask |= __GFP_WRITE;
1668 if (fgp_flags & FGP_NOFS)
1669 gfp_mask &= ~__GFP_FS;
1670
1671 page = __page_cache_alloc(gfp_mask);
1672 if (!page)
1673 return NULL;
1674
1675 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1676 fgp_flags |= FGP_LOCK;
1677
1678 /* Init accessed so avoid atomic mark_page_accessed later */
1679 if (fgp_flags & FGP_ACCESSED)
1680 __SetPageReferenced(page);
1681
1682 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1683 if (unlikely(err)) {
1684 put_page(page);
1685 page = NULL;
1686 if (err == -EEXIST)
1687 goto repeat;
1688 }
1689
1690 /*
1691 * add_to_page_cache_lru locks the page, and for mmap we expect
1692 * an unlocked page.
1693 */
1694 if (page && (fgp_flags & FGP_FOR_MMAP))
1695 unlock_page(page);
1696 }
1697
1698 return page;
1699 }
1700 EXPORT_SYMBOL(pagecache_get_page);
1701
1702 /**
1703 * find_get_entries - gang pagecache lookup
1704 * @mapping: The address_space to search
1705 * @start: The starting page cache index
1706 * @nr_entries: The maximum number of entries
1707 * @entries: Where the resulting entries are placed
1708 * @indices: The cache indices corresponding to the entries in @entries
1709 *
1710 * find_get_entries() will search for and return a group of up to
1711 * @nr_entries entries in the mapping. The entries are placed at
1712 * @entries. find_get_entries() takes a reference against any actual
1713 * pages it returns.
1714 *
1715 * The search returns a group of mapping-contiguous page cache entries
1716 * with ascending indexes. There may be holes in the indices due to
1717 * not-present pages.
1718 *
1719 * Any shadow entries of evicted pages, or swap entries from
1720 * shmem/tmpfs, are included in the returned array.
1721 *
1722 * Return: the number of pages and shadow entries which were found.
1723 */
1724 unsigned find_get_entries(struct address_space *mapping,
1725 pgoff_t start, unsigned int nr_entries,
1726 struct page **entries, pgoff_t *indices)
1727 {
1728 XA_STATE(xas, &mapping->i_pages, start);
1729 struct page *page;
1730 unsigned int ret = 0;
1731
1732 if (!nr_entries)
1733 return 0;
1734
1735 rcu_read_lock();
1736 xas_for_each(&xas, page, ULONG_MAX) {
1737 if (xas_retry(&xas, page))
1738 continue;
1739 /*
1740 * A shadow entry of a recently evicted page, a swap
1741 * entry from shmem/tmpfs or a DAX entry. Return it
1742 * without attempting to raise page count.
1743 */
1744 if (xa_is_value(page))
1745 goto export;
1746
1747 if (!page_cache_get_speculative(page))
1748 goto retry;
1749
1750 /* Has the page moved or been split? */
1751 if (unlikely(page != xas_reload(&xas)))
1752 goto put_page;
1753 page = find_subpage(page, xas.xa_index);
1754
1755 export:
1756 indices[ret] = xas.xa_index;
1757 entries[ret] = page;
1758 if (++ret == nr_entries)
1759 break;
1760 continue;
1761 put_page:
1762 put_page(page);
1763 retry:
1764 xas_reset(&xas);
1765 }
1766 rcu_read_unlock();
1767 return ret;
1768 }
1769
1770 /**
1771 * find_get_pages_range - gang pagecache lookup
1772 * @mapping: The address_space to search
1773 * @start: The starting page index
1774 * @end: The final page index (inclusive)
1775 * @nr_pages: The maximum number of pages
1776 * @pages: Where the resulting pages are placed
1777 *
1778 * find_get_pages_range() will search for and return a group of up to @nr_pages
1779 * pages in the mapping starting at index @start and up to index @end
1780 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1781 * a reference against the returned pages.
1782 *
1783 * The search returns a group of mapping-contiguous pages with ascending
1784 * indexes. There may be holes in the indices due to not-present pages.
1785 * We also update @start to index the next page for the traversal.
1786 *
1787 * Return: the number of pages which were found. If this number is
1788 * smaller than @nr_pages, the end of specified range has been
1789 * reached.
1790 */
1791 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1792 pgoff_t end, unsigned int nr_pages,
1793 struct page **pages)
1794 {
1795 XA_STATE(xas, &mapping->i_pages, *start);
1796 struct page *page;
1797 unsigned ret = 0;
1798
1799 if (unlikely(!nr_pages))
1800 return 0;
1801
1802 rcu_read_lock();
1803 xas_for_each(&xas, page, end) {
1804 if (xas_retry(&xas, page))
1805 continue;
1806 /* Skip over shadow, swap and DAX entries */
1807 if (xa_is_value(page))
1808 continue;
1809
1810 if (!page_cache_get_speculative(page))
1811 goto retry;
1812
1813 /* Has the page moved or been split? */
1814 if (unlikely(page != xas_reload(&xas)))
1815 goto put_page;
1816
1817 pages[ret] = find_subpage(page, xas.xa_index);
1818 if (++ret == nr_pages) {
1819 *start = xas.xa_index + 1;
1820 goto out;
1821 }
1822 continue;
1823 put_page:
1824 put_page(page);
1825 retry:
1826 xas_reset(&xas);
1827 }
1828
1829 /*
1830 * We come here when there is no page beyond @end. We take care to not
1831 * overflow the index @start as it confuses some of the callers. This
1832 * breaks the iteration when there is a page at index -1 but that is
1833 * already broken anyway.
1834 */
1835 if (end == (pgoff_t)-1)
1836 *start = (pgoff_t)-1;
1837 else
1838 *start = end + 1;
1839 out:
1840 rcu_read_unlock();
1841
1842 return ret;
1843 }
1844
1845 /**
1846 * find_get_pages_contig - gang contiguous pagecache lookup
1847 * @mapping: The address_space to search
1848 * @index: The starting page index
1849 * @nr_pages: The maximum number of pages
1850 * @pages: Where the resulting pages are placed
1851 *
1852 * find_get_pages_contig() works exactly like find_get_pages(), except
1853 * that the returned number of pages are guaranteed to be contiguous.
1854 *
1855 * Return: the number of pages which were found.
1856 */
1857 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1858 unsigned int nr_pages, struct page **pages)
1859 {
1860 XA_STATE(xas, &mapping->i_pages, index);
1861 struct page *page;
1862 unsigned int ret = 0;
1863
1864 if (unlikely(!nr_pages))
1865 return 0;
1866
1867 rcu_read_lock();
1868 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1869 if (xas_retry(&xas, page))
1870 continue;
1871 /*
1872 * If the entry has been swapped out, we can stop looking.
1873 * No current caller is looking for DAX entries.
1874 */
1875 if (xa_is_value(page))
1876 break;
1877
1878 if (!page_cache_get_speculative(page))
1879 goto retry;
1880
1881 /* Has the page moved or been split? */
1882 if (unlikely(page != xas_reload(&xas)))
1883 goto put_page;
1884
1885 pages[ret] = find_subpage(page, xas.xa_index);
1886 if (++ret == nr_pages)
1887 break;
1888 continue;
1889 put_page:
1890 put_page(page);
1891 retry:
1892 xas_reset(&xas);
1893 }
1894 rcu_read_unlock();
1895 return ret;
1896 }
1897 EXPORT_SYMBOL(find_get_pages_contig);
1898
1899 /**
1900 * find_get_pages_range_tag - find and return pages in given range matching @tag
1901 * @mapping: the address_space to search
1902 * @index: the starting page index
1903 * @end: The final page index (inclusive)
1904 * @tag: the tag index
1905 * @nr_pages: the maximum number of pages
1906 * @pages: where the resulting pages are placed
1907 *
1908 * Like find_get_pages, except we only return pages which are tagged with
1909 * @tag. We update @index to index the next page for the traversal.
1910 *
1911 * Return: the number of pages which were found.
1912 */
1913 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1914 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1915 struct page **pages)
1916 {
1917 XA_STATE(xas, &mapping->i_pages, *index);
1918 struct page *page;
1919 unsigned ret = 0;
1920
1921 if (unlikely(!nr_pages))
1922 return 0;
1923
1924 rcu_read_lock();
1925 xas_for_each_marked(&xas, page, end, tag) {
1926 if (xas_retry(&xas, page))
1927 continue;
1928 /*
1929 * Shadow entries should never be tagged, but this iteration
1930 * is lockless so there is a window for page reclaim to evict
1931 * a page we saw tagged. Skip over it.
1932 */
1933 if (xa_is_value(page))
1934 continue;
1935
1936 if (!page_cache_get_speculative(page))
1937 goto retry;
1938
1939 /* Has the page moved or been split? */
1940 if (unlikely(page != xas_reload(&xas)))
1941 goto put_page;
1942
1943 pages[ret] = find_subpage(page, xas.xa_index);
1944 if (++ret == nr_pages) {
1945 *index = xas.xa_index + 1;
1946 goto out;
1947 }
1948 continue;
1949 put_page:
1950 put_page(page);
1951 retry:
1952 xas_reset(&xas);
1953 }
1954
1955 /*
1956 * We come here when we got to @end. We take care to not overflow the
1957 * index @index as it confuses some of the callers. This breaks the
1958 * iteration when there is a page at index -1 but that is already
1959 * broken anyway.
1960 */
1961 if (end == (pgoff_t)-1)
1962 *index = (pgoff_t)-1;
1963 else
1964 *index = end + 1;
1965 out:
1966 rcu_read_unlock();
1967
1968 return ret;
1969 }
1970 EXPORT_SYMBOL(find_get_pages_range_tag);
1971
1972 /*
1973 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1974 * a _large_ part of the i/o request. Imagine the worst scenario:
1975 *
1976 * ---R__________________________________________B__________
1977 * ^ reading here ^ bad block(assume 4k)
1978 *
1979 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1980 * => failing the whole request => read(R) => read(R+1) =>
1981 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1982 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1983 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1984 *
1985 * It is going insane. Fix it by quickly scaling down the readahead size.
1986 */
1987 static void shrink_readahead_size_eio(struct file *filp,
1988 struct file_ra_state *ra)
1989 {
1990 ra->ra_pages /= 4;
1991 }
1992
1993 /**
1994 * generic_file_buffered_read - generic file read routine
1995 * @iocb: the iocb to read
1996 * @iter: data destination
1997 * @written: already copied
1998 *
1999 * This is a generic file read routine, and uses the
2000 * mapping->a_ops->readpage() function for the actual low-level stuff.
2001 *
2002 * This is really ugly. But the goto's actually try to clarify some
2003 * of the logic when it comes to error handling etc.
2004 *
2005 * Return:
2006 * * total number of bytes copied, including those the were already @written
2007 * * negative error code if nothing was copied
2008 */
2009 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2010 struct iov_iter *iter, ssize_t written)
2011 {
2012 struct file *filp = iocb->ki_filp;
2013 struct address_space *mapping = filp->f_mapping;
2014 struct inode *inode = mapping->host;
2015 struct file_ra_state *ra = &filp->f_ra;
2016 loff_t *ppos = &iocb->ki_pos;
2017 pgoff_t index;
2018 pgoff_t last_index;
2019 pgoff_t prev_index;
2020 unsigned long offset; /* offset into pagecache page */
2021 unsigned int prev_offset;
2022 int error = 0;
2023
2024 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2025 return 0;
2026 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2027
2028 index = *ppos >> PAGE_SHIFT;
2029 prev_index = ra->prev_pos >> PAGE_SHIFT;
2030 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2031 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2032 offset = *ppos & ~PAGE_MASK;
2033
2034 for (;;) {
2035 struct page *page;
2036 pgoff_t end_index;
2037 loff_t isize;
2038 unsigned long nr, ret;
2039
2040 cond_resched();
2041 find_page:
2042 if (fatal_signal_pending(current)) {
2043 error = -EINTR;
2044 goto out;
2045 }
2046
2047 page = find_get_page(mapping, index);
2048 if (!page) {
2049 if (iocb->ki_flags & IOCB_NOWAIT)
2050 goto would_block;
2051 page_cache_sync_readahead(mapping,
2052 ra, filp,
2053 index, last_index - index);
2054 page = find_get_page(mapping, index);
2055 if (unlikely(page == NULL))
2056 goto no_cached_page;
2057 }
2058 if (PageReadahead(page)) {
2059 page_cache_async_readahead(mapping,
2060 ra, filp, page,
2061 index, last_index - index);
2062 }
2063 if (!PageUptodate(page)) {
2064 if (iocb->ki_flags & IOCB_NOWAIT) {
2065 put_page(page);
2066 goto would_block;
2067 }
2068
2069 /*
2070 * See comment in do_read_cache_page on why
2071 * wait_on_page_locked is used to avoid unnecessarily
2072 * serialisations and why it's safe.
2073 */
2074 error = wait_on_page_locked_killable(page);
2075 if (unlikely(error))
2076 goto readpage_error;
2077 if (PageUptodate(page))
2078 goto page_ok;
2079
2080 if (inode->i_blkbits == PAGE_SHIFT ||
2081 !mapping->a_ops->is_partially_uptodate)
2082 goto page_not_up_to_date;
2083 /* pipes can't handle partially uptodate pages */
2084 if (unlikely(iov_iter_is_pipe(iter)))
2085 goto page_not_up_to_date;
2086 if (!trylock_page(page))
2087 goto page_not_up_to_date;
2088 /* Did it get truncated before we got the lock? */
2089 if (!page->mapping)
2090 goto page_not_up_to_date_locked;
2091 if (!mapping->a_ops->is_partially_uptodate(page,
2092 offset, iter->count))
2093 goto page_not_up_to_date_locked;
2094 unlock_page(page);
2095 }
2096 page_ok:
2097 /*
2098 * i_size must be checked after we know the page is Uptodate.
2099 *
2100 * Checking i_size after the check allows us to calculate
2101 * the correct value for "nr", which means the zero-filled
2102 * part of the page is not copied back to userspace (unless
2103 * another truncate extends the file - this is desired though).
2104 */
2105
2106 isize = i_size_read(inode);
2107 end_index = (isize - 1) >> PAGE_SHIFT;
2108 if (unlikely(!isize || index > end_index)) {
2109 put_page(page);
2110 goto out;
2111 }
2112
2113 /* nr is the maximum number of bytes to copy from this page */
2114 nr = PAGE_SIZE;
2115 if (index == end_index) {
2116 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2117 if (nr <= offset) {
2118 put_page(page);
2119 goto out;
2120 }
2121 }
2122 nr = nr - offset;
2123
2124 /* If users can be writing to this page using arbitrary
2125 * virtual addresses, take care about potential aliasing
2126 * before reading the page on the kernel side.
2127 */
2128 if (mapping_writably_mapped(mapping))
2129 flush_dcache_page(page);
2130
2131 /*
2132 * When a sequential read accesses a page several times,
2133 * only mark it as accessed the first time.
2134 */
2135 if (prev_index != index || offset != prev_offset)
2136 mark_page_accessed(page);
2137 prev_index = index;
2138
2139 /*
2140 * Ok, we have the page, and it's up-to-date, so
2141 * now we can copy it to user space...
2142 */
2143
2144 ret = copy_page_to_iter(page, offset, nr, iter);
2145 offset += ret;
2146 index += offset >> PAGE_SHIFT;
2147 offset &= ~PAGE_MASK;
2148 prev_offset = offset;
2149
2150 put_page(page);
2151 written += ret;
2152 if (!iov_iter_count(iter))
2153 goto out;
2154 if (ret < nr) {
2155 error = -EFAULT;
2156 goto out;
2157 }
2158 continue;
2159
2160 page_not_up_to_date:
2161 /* Get exclusive access to the page ... */
2162 error = lock_page_killable(page);
2163 if (unlikely(error))
2164 goto readpage_error;
2165
2166 page_not_up_to_date_locked:
2167 /* Did it get truncated before we got the lock? */
2168 if (!page->mapping) {
2169 unlock_page(page);
2170 put_page(page);
2171 continue;
2172 }
2173
2174 /* Did somebody else fill it already? */
2175 if (PageUptodate(page)) {
2176 unlock_page(page);
2177 goto page_ok;
2178 }
2179
2180 readpage:
2181 /*
2182 * A previous I/O error may have been due to temporary
2183 * failures, eg. multipath errors.
2184 * PG_error will be set again if readpage fails.
2185 */
2186 ClearPageError(page);
2187 /* Start the actual read. The read will unlock the page. */
2188 error = mapping->a_ops->readpage(filp, page);
2189
2190 if (unlikely(error)) {
2191 if (error == AOP_TRUNCATED_PAGE) {
2192 put_page(page);
2193 error = 0;
2194 goto find_page;
2195 }
2196 goto readpage_error;
2197 }
2198
2199 if (!PageUptodate(page)) {
2200 error = lock_page_killable(page);
2201 if (unlikely(error))
2202 goto readpage_error;
2203 if (!PageUptodate(page)) {
2204 if (page->mapping == NULL) {
2205 /*
2206 * invalidate_mapping_pages got it
2207 */
2208 unlock_page(page);
2209 put_page(page);
2210 goto find_page;
2211 }
2212 unlock_page(page);
2213 shrink_readahead_size_eio(filp, ra);
2214 error = -EIO;
2215 goto readpage_error;
2216 }
2217 unlock_page(page);
2218 }
2219
2220 goto page_ok;
2221
2222 readpage_error:
2223 /* UHHUH! A synchronous read error occurred. Report it */
2224 put_page(page);
2225 goto out;
2226
2227 no_cached_page:
2228 /*
2229 * Ok, it wasn't cached, so we need to create a new
2230 * page..
2231 */
2232 page = page_cache_alloc(mapping);
2233 if (!page) {
2234 error = -ENOMEM;
2235 goto out;
2236 }
2237 error = add_to_page_cache_lru(page, mapping, index,
2238 mapping_gfp_constraint(mapping, GFP_KERNEL));
2239 if (error) {
2240 put_page(page);
2241 if (error == -EEXIST) {
2242 error = 0;
2243 goto find_page;
2244 }
2245 goto out;
2246 }
2247 goto readpage;
2248 }
2249
2250 would_block:
2251 error = -EAGAIN;
2252 out:
2253 ra->prev_pos = prev_index;
2254 ra->prev_pos <<= PAGE_SHIFT;
2255 ra->prev_pos |= prev_offset;
2256
2257 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2258 file_accessed(filp);
2259 return written ? written : error;
2260 }
2261
2262 /**
2263 * generic_file_read_iter - generic filesystem read routine
2264 * @iocb: kernel I/O control block
2265 * @iter: destination for the data read
2266 *
2267 * This is the "read_iter()" routine for all filesystems
2268 * that can use the page cache directly.
2269 * Return:
2270 * * number of bytes copied, even for partial reads
2271 * * negative error code if nothing was read
2272 */
2273 ssize_t
2274 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2275 {
2276 size_t count = iov_iter_count(iter);
2277 ssize_t retval = 0;
2278
2279 if (!count)
2280 goto out; /* skip atime */
2281
2282 if (iocb->ki_flags & IOCB_DIRECT) {
2283 struct file *file = iocb->ki_filp;
2284 struct address_space *mapping = file->f_mapping;
2285 struct inode *inode = mapping->host;
2286 loff_t size;
2287
2288 size = i_size_read(inode);
2289 if (iocb->ki_flags & IOCB_NOWAIT) {
2290 if (filemap_range_has_page(mapping, iocb->ki_pos,
2291 iocb->ki_pos + count - 1))
2292 return -EAGAIN;
2293 } else {
2294 retval = filemap_write_and_wait_range(mapping,
2295 iocb->ki_pos,
2296 iocb->ki_pos + count - 1);
2297 if (retval < 0)
2298 goto out;
2299 }
2300
2301 file_accessed(file);
2302
2303 retval = mapping->a_ops->direct_IO(iocb, iter);
2304 if (retval >= 0) {
2305 iocb->ki_pos += retval;
2306 count -= retval;
2307 }
2308 iov_iter_revert(iter, count - iov_iter_count(iter));
2309
2310 /*
2311 * Btrfs can have a short DIO read if we encounter
2312 * compressed extents, so if there was an error, or if
2313 * we've already read everything we wanted to, or if
2314 * there was a short read because we hit EOF, go ahead
2315 * and return. Otherwise fallthrough to buffered io for
2316 * the rest of the read. Buffered reads will not work for
2317 * DAX files, so don't bother trying.
2318 */
2319 if (retval < 0 || !count || iocb->ki_pos >= size ||
2320 IS_DAX(inode))
2321 goto out;
2322 }
2323
2324 retval = generic_file_buffered_read(iocb, iter, retval);
2325 out:
2326 return retval;
2327 }
2328 EXPORT_SYMBOL(generic_file_read_iter);
2329
2330 #ifdef CONFIG_MMU
2331 #define MMAP_LOTSAMISS (100)
2332 /*
2333 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2334 * @vmf - the vm_fault for this fault.
2335 * @page - the page to lock.
2336 * @fpin - the pointer to the file we may pin (or is already pinned).
2337 *
2338 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2339 * It differs in that it actually returns the page locked if it returns 1 and 0
2340 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
2341 * will point to the pinned file and needs to be fput()'ed at a later point.
2342 */
2343 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2344 struct file **fpin)
2345 {
2346 if (trylock_page(page))
2347 return 1;
2348
2349 /*
2350 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2351 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2352 * is supposed to work. We have way too many special cases..
2353 */
2354 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2355 return 0;
2356
2357 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2358 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2359 if (__lock_page_killable(page)) {
2360 /*
2361 * We didn't have the right flags to drop the mmap_sem,
2362 * but all fault_handlers only check for fatal signals
2363 * if we return VM_FAULT_RETRY, so we need to drop the
2364 * mmap_sem here and return 0 if we don't have a fpin.
2365 */
2366 if (*fpin == NULL)
2367 up_read(&vmf->vma->vm_mm->mmap_sem);
2368 return 0;
2369 }
2370 } else
2371 __lock_page(page);
2372 return 1;
2373 }
2374
2375
2376 /*
2377 * Synchronous readahead happens when we don't even find a page in the page
2378 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2379 * to drop the mmap sem we return the file that was pinned in order for us to do
2380 * that. If we didn't pin a file then we return NULL. The file that is
2381 * returned needs to be fput()'ed when we're done with it.
2382 */
2383 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2384 {
2385 struct file *file = vmf->vma->vm_file;
2386 struct file_ra_state *ra = &file->f_ra;
2387 struct address_space *mapping = file->f_mapping;
2388 struct file *fpin = NULL;
2389 pgoff_t offset = vmf->pgoff;
2390
2391 /* If we don't want any read-ahead, don't bother */
2392 if (vmf->vma->vm_flags & VM_RAND_READ)
2393 return fpin;
2394 if (!ra->ra_pages)
2395 return fpin;
2396
2397 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2398 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2399 page_cache_sync_readahead(mapping, ra, file, offset,
2400 ra->ra_pages);
2401 return fpin;
2402 }
2403
2404 /* Avoid banging the cache line if not needed */
2405 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2406 ra->mmap_miss++;
2407
2408 /*
2409 * Do we miss much more than hit in this file? If so,
2410 * stop bothering with read-ahead. It will only hurt.
2411 */
2412 if (ra->mmap_miss > MMAP_LOTSAMISS)
2413 return fpin;
2414
2415 /*
2416 * mmap read-around
2417 */
2418 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2419 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2420 ra->size = ra->ra_pages;
2421 ra->async_size = ra->ra_pages / 4;
2422 ra_submit(ra, mapping, file);
2423 return fpin;
2424 }
2425
2426 /*
2427 * Asynchronous readahead happens when we find the page and PG_readahead,
2428 * so we want to possibly extend the readahead further. We return the file that
2429 * was pinned if we have to drop the mmap_sem in order to do IO.
2430 */
2431 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2432 struct page *page)
2433 {
2434 struct file *file = vmf->vma->vm_file;
2435 struct file_ra_state *ra = &file->f_ra;
2436 struct address_space *mapping = file->f_mapping;
2437 struct file *fpin = NULL;
2438 pgoff_t offset = vmf->pgoff;
2439
2440 /* If we don't want any read-ahead, don't bother */
2441 if (vmf->vma->vm_flags & VM_RAND_READ)
2442 return fpin;
2443 if (ra->mmap_miss > 0)
2444 ra->mmap_miss--;
2445 if (PageReadahead(page)) {
2446 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2447 page_cache_async_readahead(mapping, ra, file,
2448 page, offset, ra->ra_pages);
2449 }
2450 return fpin;
2451 }
2452
2453 /**
2454 * filemap_fault - read in file data for page fault handling
2455 * @vmf: struct vm_fault containing details of the fault
2456 *
2457 * filemap_fault() is invoked via the vma operations vector for a
2458 * mapped memory region to read in file data during a page fault.
2459 *
2460 * The goto's are kind of ugly, but this streamlines the normal case of having
2461 * it in the page cache, and handles the special cases reasonably without
2462 * having a lot of duplicated code.
2463 *
2464 * vma->vm_mm->mmap_sem must be held on entry.
2465 *
2466 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
2467 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2468 *
2469 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2470 * has not been released.
2471 *
2472 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2473 *
2474 * Return: bitwise-OR of %VM_FAULT_ codes.
2475 */
2476 vm_fault_t filemap_fault(struct vm_fault *vmf)
2477 {
2478 int error;
2479 struct file *file = vmf->vma->vm_file;
2480 struct file *fpin = NULL;
2481 struct address_space *mapping = file->f_mapping;
2482 struct file_ra_state *ra = &file->f_ra;
2483 struct inode *inode = mapping->host;
2484 pgoff_t offset = vmf->pgoff;
2485 pgoff_t max_off;
2486 struct page *page;
2487 vm_fault_t ret = 0;
2488
2489 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2490 if (unlikely(offset >= max_off))
2491 return VM_FAULT_SIGBUS;
2492
2493 /*
2494 * Do we have something in the page cache already?
2495 */
2496 page = find_get_page(mapping, offset);
2497 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2498 /*
2499 * We found the page, so try async readahead before
2500 * waiting for the lock.
2501 */
2502 fpin = do_async_mmap_readahead(vmf, page);
2503 } else if (!page) {
2504 /* No page in the page cache at all */
2505 count_vm_event(PGMAJFAULT);
2506 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2507 ret = VM_FAULT_MAJOR;
2508 fpin = do_sync_mmap_readahead(vmf);
2509 retry_find:
2510 page = pagecache_get_page(mapping, offset,
2511 FGP_CREAT|FGP_FOR_MMAP,
2512 vmf->gfp_mask);
2513 if (!page) {
2514 if (fpin)
2515 goto out_retry;
2516 return vmf_error(-ENOMEM);
2517 }
2518 }
2519
2520 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2521 goto out_retry;
2522
2523 /* Did it get truncated? */
2524 if (unlikely(compound_head(page)->mapping != mapping)) {
2525 unlock_page(page);
2526 put_page(page);
2527 goto retry_find;
2528 }
2529 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2530
2531 /*
2532 * We have a locked page in the page cache, now we need to check
2533 * that it's up-to-date. If not, it is going to be due to an error.
2534 */
2535 if (unlikely(!PageUptodate(page)))
2536 goto page_not_uptodate;
2537
2538 /*
2539 * We've made it this far and we had to drop our mmap_sem, now is the
2540 * time to return to the upper layer and have it re-find the vma and
2541 * redo the fault.
2542 */
2543 if (fpin) {
2544 unlock_page(page);
2545 goto out_retry;
2546 }
2547
2548 /*
2549 * Found the page and have a reference on it.
2550 * We must recheck i_size under page lock.
2551 */
2552 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2553 if (unlikely(offset >= max_off)) {
2554 unlock_page(page);
2555 put_page(page);
2556 return VM_FAULT_SIGBUS;
2557 }
2558
2559 vmf->page = page;
2560 return ret | VM_FAULT_LOCKED;
2561
2562 page_not_uptodate:
2563 /*
2564 * Umm, take care of errors if the page isn't up-to-date.
2565 * Try to re-read it _once_. We do this synchronously,
2566 * because there really aren't any performance issues here
2567 * and we need to check for errors.
2568 */
2569 ClearPageError(page);
2570 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2571 error = mapping->a_ops->readpage(file, page);
2572 if (!error) {
2573 wait_on_page_locked(page);
2574 if (!PageUptodate(page))
2575 error = -EIO;
2576 }
2577 if (fpin)
2578 goto out_retry;
2579 put_page(page);
2580
2581 if (!error || error == AOP_TRUNCATED_PAGE)
2582 goto retry_find;
2583
2584 /* Things didn't work out. Return zero to tell the mm layer so. */
2585 shrink_readahead_size_eio(file, ra);
2586 return VM_FAULT_SIGBUS;
2587
2588 out_retry:
2589 /*
2590 * We dropped the mmap_sem, we need to return to the fault handler to
2591 * re-find the vma and come back and find our hopefully still populated
2592 * page.
2593 */
2594 if (page)
2595 put_page(page);
2596 if (fpin)
2597 fput(fpin);
2598 return ret | VM_FAULT_RETRY;
2599 }
2600 EXPORT_SYMBOL(filemap_fault);
2601
2602 void filemap_map_pages(struct vm_fault *vmf,
2603 pgoff_t start_pgoff, pgoff_t end_pgoff)
2604 {
2605 struct file *file = vmf->vma->vm_file;
2606 struct address_space *mapping = file->f_mapping;
2607 pgoff_t last_pgoff = start_pgoff;
2608 unsigned long max_idx;
2609 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2610 struct page *page;
2611
2612 rcu_read_lock();
2613 xas_for_each(&xas, page, end_pgoff) {
2614 if (xas_retry(&xas, page))
2615 continue;
2616 if (xa_is_value(page))
2617 goto next;
2618
2619 /*
2620 * Check for a locked page first, as a speculative
2621 * reference may adversely influence page migration.
2622 */
2623 if (PageLocked(page))
2624 goto next;
2625 if (!page_cache_get_speculative(page))
2626 goto next;
2627
2628 /* Has the page moved or been split? */
2629 if (unlikely(page != xas_reload(&xas)))
2630 goto skip;
2631 page = find_subpage(page, xas.xa_index);
2632
2633 if (!PageUptodate(page) ||
2634 PageReadahead(page) ||
2635 PageHWPoison(page))
2636 goto skip;
2637 if (!trylock_page(page))
2638 goto skip;
2639
2640 if (page->mapping != mapping || !PageUptodate(page))
2641 goto unlock;
2642
2643 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2644 if (page->index >= max_idx)
2645 goto unlock;
2646
2647 if (file->f_ra.mmap_miss > 0)
2648 file->f_ra.mmap_miss--;
2649
2650 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2651 if (vmf->pte)
2652 vmf->pte += xas.xa_index - last_pgoff;
2653 last_pgoff = xas.xa_index;
2654 if (alloc_set_pte(vmf, NULL, page))
2655 goto unlock;
2656 unlock_page(page);
2657 goto next;
2658 unlock:
2659 unlock_page(page);
2660 skip:
2661 put_page(page);
2662 next:
2663 /* Huge page is mapped? No need to proceed. */
2664 if (pmd_trans_huge(*vmf->pmd))
2665 break;
2666 }
2667 rcu_read_unlock();
2668 }
2669 EXPORT_SYMBOL(filemap_map_pages);
2670
2671 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2672 {
2673 struct page *page = vmf->page;
2674 struct inode *inode = file_inode(vmf->vma->vm_file);
2675 vm_fault_t ret = VM_FAULT_LOCKED;
2676
2677 sb_start_pagefault(inode->i_sb);
2678 file_update_time(vmf->vma->vm_file);
2679 lock_page(page);
2680 if (page->mapping != inode->i_mapping) {
2681 unlock_page(page);
2682 ret = VM_FAULT_NOPAGE;
2683 goto out;
2684 }
2685 /*
2686 * We mark the page dirty already here so that when freeze is in
2687 * progress, we are guaranteed that writeback during freezing will
2688 * see the dirty page and writeprotect it again.
2689 */
2690 set_page_dirty(page);
2691 wait_for_stable_page(page);
2692 out:
2693 sb_end_pagefault(inode->i_sb);
2694 return ret;
2695 }
2696
2697 const struct vm_operations_struct generic_file_vm_ops = {
2698 .fault = filemap_fault,
2699 .map_pages = filemap_map_pages,
2700 .page_mkwrite = filemap_page_mkwrite,
2701 };
2702
2703 /* This is used for a general mmap of a disk file */
2704
2705 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2706 {
2707 struct address_space *mapping = file->f_mapping;
2708
2709 if (!mapping->a_ops->readpage)
2710 return -ENOEXEC;
2711 file_accessed(file);
2712 vma->vm_ops = &generic_file_vm_ops;
2713 return 0;
2714 }
2715
2716 /*
2717 * This is for filesystems which do not implement ->writepage.
2718 */
2719 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2720 {
2721 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2722 return -EINVAL;
2723 return generic_file_mmap(file, vma);
2724 }
2725 #else
2726 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2727 {
2728 return VM_FAULT_SIGBUS;
2729 }
2730 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2731 {
2732 return -ENOSYS;
2733 }
2734 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2735 {
2736 return -ENOSYS;
2737 }
2738 #endif /* CONFIG_MMU */
2739
2740 EXPORT_SYMBOL(filemap_page_mkwrite);
2741 EXPORT_SYMBOL(generic_file_mmap);
2742 EXPORT_SYMBOL(generic_file_readonly_mmap);
2743
2744 static struct page *wait_on_page_read(struct page *page)
2745 {
2746 if (!IS_ERR(page)) {
2747 wait_on_page_locked(page);
2748 if (!PageUptodate(page)) {
2749 put_page(page);
2750 page = ERR_PTR(-EIO);
2751 }
2752 }
2753 return page;
2754 }
2755
2756 static struct page *do_read_cache_page(struct address_space *mapping,
2757 pgoff_t index,
2758 int (*filler)(void *, struct page *),
2759 void *data,
2760 gfp_t gfp)
2761 {
2762 struct page *page;
2763 int err;
2764 repeat:
2765 page = find_get_page(mapping, index);
2766 if (!page) {
2767 page = __page_cache_alloc(gfp);
2768 if (!page)
2769 return ERR_PTR(-ENOMEM);
2770 err = add_to_page_cache_lru(page, mapping, index, gfp);
2771 if (unlikely(err)) {
2772 put_page(page);
2773 if (err == -EEXIST)
2774 goto repeat;
2775 /* Presumably ENOMEM for xarray node */
2776 return ERR_PTR(err);
2777 }
2778
2779 filler:
2780 if (filler)
2781 err = filler(data, page);
2782 else
2783 err = mapping->a_ops->readpage(data, page);
2784
2785 if (err < 0) {
2786 put_page(page);
2787 return ERR_PTR(err);
2788 }
2789
2790 page = wait_on_page_read(page);
2791 if (IS_ERR(page))
2792 return page;
2793 goto out;
2794 }
2795 if (PageUptodate(page))
2796 goto out;
2797
2798 /*
2799 * Page is not up to date and may be locked due one of the following
2800 * case a: Page is being filled and the page lock is held
2801 * case b: Read/write error clearing the page uptodate status
2802 * case c: Truncation in progress (page locked)
2803 * case d: Reclaim in progress
2804 *
2805 * Case a, the page will be up to date when the page is unlocked.
2806 * There is no need to serialise on the page lock here as the page
2807 * is pinned so the lock gives no additional protection. Even if the
2808 * the page is truncated, the data is still valid if PageUptodate as
2809 * it's a race vs truncate race.
2810 * Case b, the page will not be up to date
2811 * Case c, the page may be truncated but in itself, the data may still
2812 * be valid after IO completes as it's a read vs truncate race. The
2813 * operation must restart if the page is not uptodate on unlock but
2814 * otherwise serialising on page lock to stabilise the mapping gives
2815 * no additional guarantees to the caller as the page lock is
2816 * released before return.
2817 * Case d, similar to truncation. If reclaim holds the page lock, it
2818 * will be a race with remove_mapping that determines if the mapping
2819 * is valid on unlock but otherwise the data is valid and there is
2820 * no need to serialise with page lock.
2821 *
2822 * As the page lock gives no additional guarantee, we optimistically
2823 * wait on the page to be unlocked and check if it's up to date and
2824 * use the page if it is. Otherwise, the page lock is required to
2825 * distinguish between the different cases. The motivation is that we
2826 * avoid spurious serialisations and wakeups when multiple processes
2827 * wait on the same page for IO to complete.
2828 */
2829 wait_on_page_locked(page);
2830 if (PageUptodate(page))
2831 goto out;
2832
2833 /* Distinguish between all the cases under the safety of the lock */
2834 lock_page(page);
2835
2836 /* Case c or d, restart the operation */
2837 if (!page->mapping) {
2838 unlock_page(page);
2839 put_page(page);
2840 goto repeat;
2841 }
2842
2843 /* Someone else locked and filled the page in a very small window */
2844 if (PageUptodate(page)) {
2845 unlock_page(page);
2846 goto out;
2847 }
2848 goto filler;
2849
2850 out:
2851 mark_page_accessed(page);
2852 return page;
2853 }
2854
2855 /**
2856 * read_cache_page - read into page cache, fill it if needed
2857 * @mapping: the page's address_space
2858 * @index: the page index
2859 * @filler: function to perform the read
2860 * @data: first arg to filler(data, page) function, often left as NULL
2861 *
2862 * Read into the page cache. If a page already exists, and PageUptodate() is
2863 * not set, try to fill the page and wait for it to become unlocked.
2864 *
2865 * If the page does not get brought uptodate, return -EIO.
2866 *
2867 * Return: up to date page on success, ERR_PTR() on failure.
2868 */
2869 struct page *read_cache_page(struct address_space *mapping,
2870 pgoff_t index,
2871 int (*filler)(void *, struct page *),
2872 void *data)
2873 {
2874 return do_read_cache_page(mapping, index, filler, data,
2875 mapping_gfp_mask(mapping));
2876 }
2877 EXPORT_SYMBOL(read_cache_page);
2878
2879 /**
2880 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2881 * @mapping: the page's address_space
2882 * @index: the page index
2883 * @gfp: the page allocator flags to use if allocating
2884 *
2885 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2886 * any new page allocations done using the specified allocation flags.
2887 *
2888 * If the page does not get brought uptodate, return -EIO.
2889 *
2890 * Return: up to date page on success, ERR_PTR() on failure.
2891 */
2892 struct page *read_cache_page_gfp(struct address_space *mapping,
2893 pgoff_t index,
2894 gfp_t gfp)
2895 {
2896 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
2897 }
2898 EXPORT_SYMBOL(read_cache_page_gfp);
2899
2900 /*
2901 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2902 * LFS limits. If pos is under the limit it becomes a short access. If it
2903 * exceeds the limit we return -EFBIG.
2904 */
2905 static int generic_write_check_limits(struct file *file, loff_t pos,
2906 loff_t *count)
2907 {
2908 struct inode *inode = file->f_mapping->host;
2909 loff_t max_size = inode->i_sb->s_maxbytes;
2910 loff_t limit = rlimit(RLIMIT_FSIZE);
2911
2912 if (limit != RLIM_INFINITY) {
2913 if (pos >= limit) {
2914 send_sig(SIGXFSZ, current, 0);
2915 return -EFBIG;
2916 }
2917 *count = min(*count, limit - pos);
2918 }
2919
2920 if (!(file->f_flags & O_LARGEFILE))
2921 max_size = MAX_NON_LFS;
2922
2923 if (unlikely(pos >= max_size))
2924 return -EFBIG;
2925
2926 *count = min(*count, max_size - pos);
2927
2928 return 0;
2929 }
2930
2931 /*
2932 * Performs necessary checks before doing a write
2933 *
2934 * Can adjust writing position or amount of bytes to write.
2935 * Returns appropriate error code that caller should return or
2936 * zero in case that write should be allowed.
2937 */
2938 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2939 {
2940 struct file *file = iocb->ki_filp;
2941 struct inode *inode = file->f_mapping->host;
2942 loff_t count;
2943 int ret;
2944
2945 if (IS_SWAPFILE(inode))
2946 return -ETXTBSY;
2947
2948 if (!iov_iter_count(from))
2949 return 0;
2950
2951 /* FIXME: this is for backwards compatibility with 2.4 */
2952 if (iocb->ki_flags & IOCB_APPEND)
2953 iocb->ki_pos = i_size_read(inode);
2954
2955 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2956 return -EINVAL;
2957
2958 count = iov_iter_count(from);
2959 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
2960 if (ret)
2961 return ret;
2962
2963 iov_iter_truncate(from, count);
2964 return iov_iter_count(from);
2965 }
2966 EXPORT_SYMBOL(generic_write_checks);
2967
2968 /*
2969 * Performs necessary checks before doing a clone.
2970 *
2971 * Can adjust amount of bytes to clone via @req_count argument.
2972 * Returns appropriate error code that caller should return or
2973 * zero in case the clone should be allowed.
2974 */
2975 int generic_remap_checks(struct file *file_in, loff_t pos_in,
2976 struct file *file_out, loff_t pos_out,
2977 loff_t *req_count, unsigned int remap_flags)
2978 {
2979 struct inode *inode_in = file_in->f_mapping->host;
2980 struct inode *inode_out = file_out->f_mapping->host;
2981 uint64_t count = *req_count;
2982 uint64_t bcount;
2983 loff_t size_in, size_out;
2984 loff_t bs = inode_out->i_sb->s_blocksize;
2985 int ret;
2986
2987 /* The start of both ranges must be aligned to an fs block. */
2988 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
2989 return -EINVAL;
2990
2991 /* Ensure offsets don't wrap. */
2992 if (pos_in + count < pos_in || pos_out + count < pos_out)
2993 return -EINVAL;
2994
2995 size_in = i_size_read(inode_in);
2996 size_out = i_size_read(inode_out);
2997
2998 /* Dedupe requires both ranges to be within EOF. */
2999 if ((remap_flags & REMAP_FILE_DEDUP) &&
3000 (pos_in >= size_in || pos_in + count > size_in ||
3001 pos_out >= size_out || pos_out + count > size_out))
3002 return -EINVAL;
3003
3004 /* Ensure the infile range is within the infile. */
3005 if (pos_in >= size_in)
3006 return -EINVAL;
3007 count = min(count, size_in - (uint64_t)pos_in);
3008
3009 ret = generic_write_check_limits(file_out, pos_out, &count);
3010 if (ret)
3011 return ret;
3012
3013 /*
3014 * If the user wanted us to link to the infile's EOF, round up to the
3015 * next block boundary for this check.
3016 *
3017 * Otherwise, make sure the count is also block-aligned, having
3018 * already confirmed the starting offsets' block alignment.
3019 */
3020 if (pos_in + count == size_in) {
3021 bcount = ALIGN(size_in, bs) - pos_in;
3022 } else {
3023 if (!IS_ALIGNED(count, bs))
3024 count = ALIGN_DOWN(count, bs);
3025 bcount = count;
3026 }
3027
3028 /* Don't allow overlapped cloning within the same file. */
3029 if (inode_in == inode_out &&
3030 pos_out + bcount > pos_in &&
3031 pos_out < pos_in + bcount)
3032 return -EINVAL;
3033
3034 /*
3035 * We shortened the request but the caller can't deal with that, so
3036 * bounce the request back to userspace.
3037 */
3038 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3039 return -EINVAL;
3040
3041 *req_count = count;
3042 return 0;
3043 }
3044
3045
3046 /*
3047 * Performs common checks before doing a file copy/clone
3048 * from @file_in to @file_out.
3049 */
3050 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3051 {
3052 struct inode *inode_in = file_inode(file_in);
3053 struct inode *inode_out = file_inode(file_out);
3054
3055 /* Don't copy dirs, pipes, sockets... */
3056 if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3057 return -EISDIR;
3058 if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3059 return -EINVAL;
3060
3061 if (!(file_in->f_mode & FMODE_READ) ||
3062 !(file_out->f_mode & FMODE_WRITE) ||
3063 (file_out->f_flags & O_APPEND))
3064 return -EBADF;
3065
3066 return 0;
3067 }
3068
3069 /*
3070 * Performs necessary checks before doing a file copy
3071 *
3072 * Can adjust amount of bytes to copy via @req_count argument.
3073 * Returns appropriate error code that caller should return or
3074 * zero in case the copy should be allowed.
3075 */
3076 int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3077 struct file *file_out, loff_t pos_out,
3078 size_t *req_count, unsigned int flags)
3079 {
3080 struct inode *inode_in = file_inode(file_in);
3081 struct inode *inode_out = file_inode(file_out);
3082 uint64_t count = *req_count;
3083 loff_t size_in;
3084 int ret;
3085
3086 ret = generic_file_rw_checks(file_in, file_out);
3087 if (ret)
3088 return ret;
3089
3090 /* Don't touch certain kinds of inodes */
3091 if (IS_IMMUTABLE(inode_out))
3092 return -EPERM;
3093
3094 if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3095 return -ETXTBSY;
3096
3097 /* Ensure offsets don't wrap. */
3098 if (pos_in + count < pos_in || pos_out + count < pos_out)
3099 return -EOVERFLOW;
3100
3101 /* Shorten the copy to EOF */
3102 size_in = i_size_read(inode_in);
3103 if (pos_in >= size_in)
3104 count = 0;
3105 else
3106 count = min(count, size_in - (uint64_t)pos_in);
3107
3108 ret = generic_write_check_limits(file_out, pos_out, &count);
3109 if (ret)
3110 return ret;
3111
3112 /* Don't allow overlapped copying within the same file. */
3113 if (inode_in == inode_out &&
3114 pos_out + count > pos_in &&
3115 pos_out < pos_in + count)
3116 return -EINVAL;
3117
3118 *req_count = count;
3119 return 0;
3120 }
3121
3122 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3123 loff_t pos, unsigned len, unsigned flags,
3124 struct page **pagep, void **fsdata)
3125 {
3126 const struct address_space_operations *aops = mapping->a_ops;
3127
3128 return aops->write_begin(file, mapping, pos, len, flags,
3129 pagep, fsdata);
3130 }
3131 EXPORT_SYMBOL(pagecache_write_begin);
3132
3133 int pagecache_write_end(struct file *file, struct address_space *mapping,
3134 loff_t pos, unsigned len, unsigned copied,
3135 struct page *page, void *fsdata)
3136 {
3137 const struct address_space_operations *aops = mapping->a_ops;
3138
3139 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3140 }
3141 EXPORT_SYMBOL(pagecache_write_end);
3142
3143 ssize_t
3144 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3145 {
3146 struct file *file = iocb->ki_filp;
3147 struct address_space *mapping = file->f_mapping;
3148 struct inode *inode = mapping->host;
3149 loff_t pos = iocb->ki_pos;
3150 ssize_t written;
3151 size_t write_len;
3152 pgoff_t end;
3153
3154 write_len = iov_iter_count(from);
3155 end = (pos + write_len - 1) >> PAGE_SHIFT;
3156
3157 if (iocb->ki_flags & IOCB_NOWAIT) {
3158 /* If there are pages to writeback, return */
3159 if (filemap_range_has_page(inode->i_mapping, pos,
3160 pos + write_len - 1))
3161 return -EAGAIN;
3162 } else {
3163 written = filemap_write_and_wait_range(mapping, pos,
3164 pos + write_len - 1);
3165 if (written)
3166 goto out;
3167 }
3168
3169 /*
3170 * After a write we want buffered reads to be sure to go to disk to get
3171 * the new data. We invalidate clean cached page from the region we're
3172 * about to write. We do this *before* the write so that we can return
3173 * without clobbering -EIOCBQUEUED from ->direct_IO().
3174 */
3175 written = invalidate_inode_pages2_range(mapping,
3176 pos >> PAGE_SHIFT, end);
3177 /*
3178 * If a page can not be invalidated, return 0 to fall back
3179 * to buffered write.
3180 */
3181 if (written) {
3182 if (written == -EBUSY)
3183 return 0;
3184 goto out;
3185 }
3186
3187 written = mapping->a_ops->direct_IO(iocb, from);
3188
3189 /*
3190 * Finally, try again to invalidate clean pages which might have been
3191 * cached by non-direct readahead, or faulted in by get_user_pages()
3192 * if the source of the write was an mmap'ed region of the file
3193 * we're writing. Either one is a pretty crazy thing to do,
3194 * so we don't support it 100%. If this invalidation
3195 * fails, tough, the write still worked...
3196 *
3197 * Most of the time we do not need this since dio_complete() will do
3198 * the invalidation for us. However there are some file systems that
3199 * do not end up with dio_complete() being called, so let's not break
3200 * them by removing it completely
3201 */
3202 if (mapping->nrpages)
3203 invalidate_inode_pages2_range(mapping,
3204 pos >> PAGE_SHIFT, end);
3205
3206 if (written > 0) {
3207 pos += written;
3208 write_len -= written;
3209 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3210 i_size_write(inode, pos);
3211 mark_inode_dirty(inode);
3212 }
3213 iocb->ki_pos = pos;
3214 }
3215 iov_iter_revert(from, write_len - iov_iter_count(from));
3216 out:
3217 return written;
3218 }
3219 EXPORT_SYMBOL(generic_file_direct_write);
3220
3221 /*
3222 * Find or create a page at the given pagecache position. Return the locked
3223 * page. This function is specifically for buffered writes.
3224 */
3225 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3226 pgoff_t index, unsigned flags)
3227 {
3228 struct page *page;
3229 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3230
3231 if (flags & AOP_FLAG_NOFS)
3232 fgp_flags |= FGP_NOFS;
3233
3234 page = pagecache_get_page(mapping, index, fgp_flags,
3235 mapping_gfp_mask(mapping));
3236 if (page)
3237 wait_for_stable_page(page);
3238
3239 return page;
3240 }
3241 EXPORT_SYMBOL(grab_cache_page_write_begin);
3242
3243 ssize_t generic_perform_write(struct file *file,
3244 struct iov_iter *i, loff_t pos)
3245 {
3246 struct address_space *mapping = file->f_mapping;
3247 const struct address_space_operations *a_ops = mapping->a_ops;
3248 long status = 0;
3249 ssize_t written = 0;
3250 unsigned int flags = 0;
3251
3252 do {
3253 struct page *page;
3254 unsigned long offset; /* Offset into pagecache page */
3255 unsigned long bytes; /* Bytes to write to page */
3256 size_t copied; /* Bytes copied from user */
3257 void *fsdata;
3258
3259 offset = (pos & (PAGE_SIZE - 1));
3260 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3261 iov_iter_count(i));
3262
3263 again:
3264 /*
3265 * Bring in the user page that we will copy from _first_.
3266 * Otherwise there's a nasty deadlock on copying from the
3267 * same page as we're writing to, without it being marked
3268 * up-to-date.
3269 *
3270 * Not only is this an optimisation, but it is also required
3271 * to check that the address is actually valid, when atomic
3272 * usercopies are used, below.
3273 */
3274 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3275 status = -EFAULT;
3276 break;
3277 }
3278
3279 if (fatal_signal_pending(current)) {
3280 status = -EINTR;
3281 break;
3282 }
3283
3284 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3285 &page, &fsdata);
3286 if (unlikely(status < 0))
3287 break;
3288
3289 if (mapping_writably_mapped(mapping))
3290 flush_dcache_page(page);
3291
3292 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3293 flush_dcache_page(page);
3294
3295 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3296 page, fsdata);
3297 if (unlikely(status < 0))
3298 break;
3299 copied = status;
3300
3301 cond_resched();
3302
3303 iov_iter_advance(i, copied);
3304 if (unlikely(copied == 0)) {
3305 /*
3306 * If we were unable to copy any data at all, we must
3307 * fall back to a single segment length write.
3308 *
3309 * If we didn't fallback here, we could livelock
3310 * because not all segments in the iov can be copied at
3311 * once without a pagefault.
3312 */
3313 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3314 iov_iter_single_seg_count(i));
3315 goto again;
3316 }
3317 pos += copied;
3318 written += copied;
3319
3320 balance_dirty_pages_ratelimited(mapping);
3321 } while (iov_iter_count(i));
3322
3323 return written ? written : status;
3324 }
3325 EXPORT_SYMBOL(generic_perform_write);
3326
3327 /**
3328 * __generic_file_write_iter - write data to a file
3329 * @iocb: IO state structure (file, offset, etc.)
3330 * @from: iov_iter with data to write
3331 *
3332 * This function does all the work needed for actually writing data to a
3333 * file. It does all basic checks, removes SUID from the file, updates
3334 * modification times and calls proper subroutines depending on whether we
3335 * do direct IO or a standard buffered write.
3336 *
3337 * It expects i_mutex to be grabbed unless we work on a block device or similar
3338 * object which does not need locking at all.
3339 *
3340 * This function does *not* take care of syncing data in case of O_SYNC write.
3341 * A caller has to handle it. This is mainly due to the fact that we want to
3342 * avoid syncing under i_mutex.
3343 *
3344 * Return:
3345 * * number of bytes written, even for truncated writes
3346 * * negative error code if no data has been written at all
3347 */
3348 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3349 {
3350 struct file *file = iocb->ki_filp;
3351 struct address_space * mapping = file->f_mapping;
3352 struct inode *inode = mapping->host;
3353 ssize_t written = 0;
3354 ssize_t err;
3355 ssize_t status;
3356
3357 /* We can write back this queue in page reclaim */
3358 current->backing_dev_info = inode_to_bdi(inode);
3359 err = file_remove_privs(file);
3360 if (err)
3361 goto out;
3362
3363 err = file_update_time(file);
3364 if (err)
3365 goto out;
3366
3367 if (iocb->ki_flags & IOCB_DIRECT) {
3368 loff_t pos, endbyte;
3369
3370 written = generic_file_direct_write(iocb, from);
3371 /*
3372 * If the write stopped short of completing, fall back to
3373 * buffered writes. Some filesystems do this for writes to
3374 * holes, for example. For DAX files, a buffered write will
3375 * not succeed (even if it did, DAX does not handle dirty
3376 * page-cache pages correctly).
3377 */
3378 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3379 goto out;
3380
3381 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3382 /*
3383 * If generic_perform_write() returned a synchronous error
3384 * then we want to return the number of bytes which were
3385 * direct-written, or the error code if that was zero. Note
3386 * that this differs from normal direct-io semantics, which
3387 * will return -EFOO even if some bytes were written.
3388 */
3389 if (unlikely(status < 0)) {
3390 err = status;
3391 goto out;
3392 }
3393 /*
3394 * We need to ensure that the page cache pages are written to
3395 * disk and invalidated to preserve the expected O_DIRECT
3396 * semantics.
3397 */
3398 endbyte = pos + status - 1;
3399 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3400 if (err == 0) {
3401 iocb->ki_pos = endbyte + 1;
3402 written += status;
3403 invalidate_mapping_pages(mapping,
3404 pos >> PAGE_SHIFT,
3405 endbyte >> PAGE_SHIFT);
3406 } else {
3407 /*
3408 * We don't know how much we wrote, so just return
3409 * the number of bytes which were direct-written
3410 */
3411 }
3412 } else {
3413 written = generic_perform_write(file, from, iocb->ki_pos);
3414 if (likely(written > 0))
3415 iocb->ki_pos += written;
3416 }
3417 out:
3418 current->backing_dev_info = NULL;
3419 return written ? written : err;
3420 }
3421 EXPORT_SYMBOL(__generic_file_write_iter);
3422
3423 /**
3424 * generic_file_write_iter - write data to a file
3425 * @iocb: IO state structure
3426 * @from: iov_iter with data to write
3427 *
3428 * This is a wrapper around __generic_file_write_iter() to be used by most
3429 * filesystems. It takes care of syncing the file in case of O_SYNC file
3430 * and acquires i_mutex as needed.
3431 * Return:
3432 * * negative error code if no data has been written at all of
3433 * vfs_fsync_range() failed for a synchronous write
3434 * * number of bytes written, even for truncated writes
3435 */
3436 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3437 {
3438 struct file *file = iocb->ki_filp;
3439 struct inode *inode = file->f_mapping->host;
3440 ssize_t ret;
3441
3442 inode_lock(inode);
3443 ret = generic_write_checks(iocb, from);
3444 if (ret > 0)
3445 ret = __generic_file_write_iter(iocb, from);
3446 inode_unlock(inode);
3447
3448 if (ret > 0)
3449 ret = generic_write_sync(iocb, ret);
3450 return ret;
3451 }
3452 EXPORT_SYMBOL(generic_file_write_iter);
3453
3454 /**
3455 * try_to_release_page() - release old fs-specific metadata on a page
3456 *
3457 * @page: the page which the kernel is trying to free
3458 * @gfp_mask: memory allocation flags (and I/O mode)
3459 *
3460 * The address_space is to try to release any data against the page
3461 * (presumably at page->private).
3462 *
3463 * This may also be called if PG_fscache is set on a page, indicating that the
3464 * page is known to the local caching routines.
3465 *
3466 * The @gfp_mask argument specifies whether I/O may be performed to release
3467 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3468 *
3469 * Return: %1 if the release was successful, otherwise return zero.
3470 */
3471 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3472 {
3473 struct address_space * const mapping = page->mapping;
3474
3475 BUG_ON(!PageLocked(page));
3476 if (PageWriteback(page))
3477 return 0;
3478
3479 if (mapping && mapping->a_ops->releasepage)
3480 return mapping->a_ops->releasepage(page, gfp_mask);
3481 return try_to_free_buffers(page);
3482 }
3483
3484 EXPORT_SYMBOL(try_to_release_page);