1/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
12#include <linux/export.h>
13#include <linux/compiler.h>
14#include <linux/dax.h>
15#include <linux/fs.h>
16#include <linux/sched/signal.h>
17#include <linux/uaccess.h>
18#include <linux/capability.h>
19#include <linux/kernel_stat.h>
20#include <linux/gfp.h>
21#include <linux/mm.h>
22#include <linux/swap.h>
23#include <linux/mman.h>
24#include <linux/pagemap.h>
25#include <linux/file.h>
26#include <linux/uio.h>
27#include <linux/hash.h>
28#include <linux/writeback.h>
29#include <linux/backing-dev.h>
30#include <linux/pagevec.h>
31#include <linux/blkdev.h>
32#include <linux/security.h>
33#include <linux/cpuset.h>
34#include <linux/hugetlb.h>
35#include <linux/memcontrol.h>
36#include <linux/cleancache.h>
37#include <linux/shmem_fs.h>
38#include <linux/rmap.h>
39#include <linux/delayacct.h>
40#include <linux/psi.h>
41#include "internal.h"
42
43#define CREATE_TRACE_POINTS
44#include <trace/events/filemap.h>
45
46/*
47 * FIXME: remove all knowledge of the buffer layer from the core VM
48 */
49#include <linux/buffer_head.h> /* for try_to_free_buffers */
50
51#include <asm/mman.h>
52
53/*
54 * Shared mappings implemented 30.11.1994. It's not fully working yet,
55 * though.
56 *
57 * Shared mappings now work. 15.8.1995 Bruno.
58 *
59 * finished 'unifying' the page and buffer cache and SMP-threaded the
60 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
61 *
62 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
63 */
64
65/*
66 * Lock ordering:
67 *
68 * ->i_mmap_rwsem (truncate_pagecache)
69 * ->private_lock (__free_pte->__set_page_dirty_buffers)
70 * ->swap_lock (exclusive_swap_page, others)
71 * ->i_pages lock
72 *
73 * ->i_mutex
74 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
75 *
76 * ->mmap_sem
77 * ->i_mmap_rwsem
78 * ->page_table_lock or pte_lock (various, mainly in memory.c)
79 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
80 *
81 * ->mmap_sem
82 * ->lock_page (access_process_vm)
83 *
84 * ->i_mutex (generic_perform_write)
85 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
86 *
87 * bdi->wb.list_lock
88 * sb_lock (fs/fs-writeback.c)
89 * ->i_pages lock (__sync_single_inode)
90 *
91 * ->i_mmap_rwsem
92 * ->anon_vma.lock (vma_adjust)
93 *
94 * ->anon_vma.lock
95 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
96 *
97 * ->page_table_lock or pte_lock
98 * ->swap_lock (try_to_unmap_one)
99 * ->private_lock (try_to_unmap_one)
100 * ->i_pages lock (try_to_unmap_one)
101 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
102 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
103 * ->private_lock (page_remove_rmap->set_page_dirty)
104 * ->i_pages lock (page_remove_rmap->set_page_dirty)
105 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
106 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
107 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
108 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
109 * ->inode->i_lock (zap_pte_range->set_page_dirty)
110 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
111 *
112 * ->i_mmap_rwsem
113 * ->tasklist_lock (memory_failure, collect_procs_ao)
114 */
115
116static void page_cache_delete(struct address_space *mapping,
117 struct page *page, void *shadow)
118{
119 XA_STATE(xas, &mapping->i_pages, page->index);
120 unsigned int nr = 1;
121
122 mapping_set_update(&xas, mapping);
123
124 /* hugetlb pages are represented by a single entry in the xarray */
125 if (!PageHuge(page)) {
126 xas_set_order(&xas, page->index, compound_order(page));
127 nr = 1U << compound_order(page);
128 }
129
130 VM_BUG_ON_PAGE(!PageLocked(page), page);
131 VM_BUG_ON_PAGE(PageTail(page), page);
132 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
133
134 xas_store(&xas, shadow);
135 xas_init_marks(&xas);
136
137 page->mapping = NULL;
138 /* Leave page->index set: truncation lookup relies upon it */
139
140 if (shadow) {
141 mapping->nrexceptional += nr;
142 /*
143 * Make sure the nrexceptional update is committed before
144 * the nrpages update so that final truncate racing
145 * with reclaim does not see both counters 0 at the
146 * same time and miss a shadow entry.
147 */
148 smp_wmb();
149 }
150 mapping->nrpages -= nr;
151}
152
153static void unaccount_page_cache_page(struct address_space *mapping,
154 struct page *page)
155{
156 int nr;
157
158 /*
159 * if we're uptodate, flush out into the cleancache, otherwise
160 * invalidate any existing cleancache entries. We can't leave
161 * stale data around in the cleancache once our page is gone
162 */
163 if (PageUptodate(page) && PageMappedToDisk(page))
164 cleancache_put_page(page);
165 else
166 cleancache_invalidate_page(mapping, page);
167
168 VM_BUG_ON_PAGE(PageTail(page), page);
169 VM_BUG_ON_PAGE(page_mapped(page), page);
170 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
171 int mapcount;
172
173 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
174 current->comm, page_to_pfn(page));
175 dump_page(page, "still mapped when deleted");
176 dump_stack();
177 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
178
179 mapcount = page_mapcount(page);
180 if (mapping_exiting(mapping) &&
181 page_count(page) >= mapcount + 2) {
182 /*
183 * All vmas have already been torn down, so it's
184 * a good bet that actually the page is unmapped,
185 * and we'd prefer not to leak it: if we're wrong,
186 * some other bad page check should catch it later.
187 */
188 page_mapcount_reset(page);
189 page_ref_sub(page, mapcount);
190 }
191 }
192
193 /* hugetlb pages do not participate in page cache accounting. */
194 if (PageHuge(page))
195 return;
196
197 nr = hpage_nr_pages(page);
198
199 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
200 if (PageSwapBacked(page)) {
201 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
202 if (PageTransHuge(page))
203 __dec_node_page_state(page, NR_SHMEM_THPS);
204 } else {
205 VM_BUG_ON_PAGE(PageTransHuge(page), page);
206 }
207
208 /*
209 * At this point page must be either written or cleaned by
210 * truncate. Dirty page here signals a bug and loss of
211 * unwritten data.
212 *
213 * This fixes dirty accounting after removing the page entirely
214 * but leaves PageDirty set: it has no effect for truncated
215 * page and anyway will be cleared before returning page into
216 * buddy allocator.
217 */
218 if (WARN_ON_ONCE(PageDirty(page)))
219 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
220}
221
222/*
223 * Delete a page from the page cache and free it. Caller has to make
224 * sure the page is locked and that nobody else uses it - or that usage
225 * is safe. The caller must hold the i_pages lock.
226 */
227void __delete_from_page_cache(struct page *page, void *shadow)
228{
229 struct address_space *mapping = page->mapping;
230
231 trace_mm_filemap_delete_from_page_cache(page);
232
233 unaccount_page_cache_page(mapping, page);
234 page_cache_delete(mapping, page, shadow);
235}
236
237static void page_cache_free_page(struct address_space *mapping,
238 struct page *page)
239{
240 void (*freepage)(struct page *);
241
242 freepage = mapping->a_ops->freepage;
243 if (freepage)
244 freepage(page);
245
246 if (PageTransHuge(page) && !PageHuge(page)) {
247 page_ref_sub(page, HPAGE_PMD_NR);
248 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
249 } else {
250 put_page(page);
251 }
252}
253
254/**
255 * delete_from_page_cache - delete page from page cache
256 * @page: the page which the kernel is trying to remove from page cache
257 *
258 * This must be called only on pages that have been verified to be in the page
259 * cache and locked. It will never put the page into the free list, the caller
260 * has a reference on the page.
261 */
262void delete_from_page_cache(struct page *page)
263{
264 struct address_space *mapping = page_mapping(page);
265 unsigned long flags;
266
267 BUG_ON(!PageLocked(page));
268 xa_lock_irqsave(&mapping->i_pages, flags);
269 __delete_from_page_cache(page, NULL);
270 xa_unlock_irqrestore(&mapping->i_pages, flags);
271
272 page_cache_free_page(mapping, page);
273}
274EXPORT_SYMBOL(delete_from_page_cache);
275
276/*
277 * page_cache_delete_batch - delete several pages from page cache
278 * @mapping: the mapping to which pages belong
279 * @pvec: pagevec with pages to delete
280 *
281 * The function walks over mapping->i_pages and removes pages passed in @pvec
282 * from the mapping. The function expects @pvec to be sorted by page index.
283 * It tolerates holes in @pvec (mapping entries at those indices are not
284 * modified). The function expects only THP head pages to be present in the
285 * @pvec and takes care to delete all corresponding tail pages from the
286 * mapping as well.
287 *
288 * The function expects the i_pages lock to be held.
289 */
290static void page_cache_delete_batch(struct address_space *mapping,
291 struct pagevec *pvec)
292{
293 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
294 int total_pages = 0;
295 int i = 0, tail_pages = 0;
296 struct page *page;
297
298 mapping_set_update(&xas, mapping);
299 xas_for_each(&xas, page, ULONG_MAX) {
300 if (i >= pagevec_count(pvec) && !tail_pages)
301 break;
302 if (xa_is_value(page))
303 continue;
304 if (!tail_pages) {
305 /*
306 * Some page got inserted in our range? Skip it. We
307 * have our pages locked so they are protected from
308 * being removed.
309 */
310 if (page != pvec->pages[i]) {
311 VM_BUG_ON_PAGE(page->index >
312 pvec->pages[i]->index, page);
313 continue;
314 }
315 WARN_ON_ONCE(!PageLocked(page));
316 if (PageTransHuge(page) && !PageHuge(page))
317 tail_pages = HPAGE_PMD_NR - 1;
318 page->mapping = NULL;
319 /*
320 * Leave page->index set: truncation lookup relies
321 * upon it
322 */
323 i++;
324 } else {
325 VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages
326 != pvec->pages[i]->index, page);
327 tail_pages--;
328 }
329 xas_store(&xas, NULL);
330 total_pages++;
331 }
332 mapping->nrpages -= total_pages;
333}
334
335void delete_from_page_cache_batch(struct address_space *mapping,
336 struct pagevec *pvec)
337{
338 int i;
339 unsigned long flags;
340
341 if (!pagevec_count(pvec))
342 return;
343
344 xa_lock_irqsave(&mapping->i_pages, flags);
345 for (i = 0; i < pagevec_count(pvec); i++) {
346 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
347
348 unaccount_page_cache_page(mapping, pvec->pages[i]);
349 }
350 page_cache_delete_batch(mapping, pvec);
351 xa_unlock_irqrestore(&mapping->i_pages, flags);
352
353 for (i = 0; i < pagevec_count(pvec); i++)
354 page_cache_free_page(mapping, pvec->pages[i]);
355}
356
357int filemap_check_errors(struct address_space *mapping)
358{
359 int ret = 0;
360 /* Check for outstanding write errors */
361 if (test_bit(AS_ENOSPC, &mapping->flags) &&
362 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
363 ret = -ENOSPC;
364 if (test_bit(AS_EIO, &mapping->flags) &&
365 test_and_clear_bit(AS_EIO, &mapping->flags))
366 ret = -EIO;
367 return ret;
368}
369EXPORT_SYMBOL(filemap_check_errors);
370
371static int filemap_check_and_keep_errors(struct address_space *mapping)
372{
373 /* Check for outstanding write errors */
374 if (test_bit(AS_EIO, &mapping->flags))
375 return -EIO;
376 if (test_bit(AS_ENOSPC, &mapping->flags))
377 return -ENOSPC;
378 return 0;
379}
380
381/**
382 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
383 * @mapping: address space structure to write
384 * @start: offset in bytes where the range starts
385 * @end: offset in bytes where the range ends (inclusive)
386 * @sync_mode: enable synchronous operation
387 *
388 * Start writeback against all of a mapping's dirty pages that lie
389 * within the byte offsets <start, end> inclusive.
390 *
391 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
392 * opposed to a regular memory cleansing writeback. The difference between
393 * these two operations is that if a dirty page/buffer is encountered, it must
394 * be waited upon, and not just skipped over.
395 *
396 * Return: %0 on success, negative error code otherwise.
397 */
398int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
399 loff_t end, int sync_mode)
400{
401 int ret;
402 struct writeback_control wbc = {
403 .sync_mode = sync_mode,
404 .nr_to_write = LONG_MAX,
405 .range_start = start,
406 .range_end = end,
407 };
408
409 if (!mapping_cap_writeback_dirty(mapping))
410 return 0;
411
412 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
413 ret = do_writepages(mapping, &wbc);
414 wbc_detach_inode(&wbc);
415 return ret;
416}
417
418static inline int __filemap_fdatawrite(struct address_space *mapping,
419 int sync_mode)
420{
421 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
422}
423
424int filemap_fdatawrite(struct address_space *mapping)
425{
426 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
427}
428EXPORT_SYMBOL(filemap_fdatawrite);
429
430int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
431 loff_t end)
432{
433 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
434}
435EXPORT_SYMBOL(filemap_fdatawrite_range);
436
437/**
438 * filemap_flush - mostly a non-blocking flush
439 * @mapping: target address_space
440 *
441 * This is a mostly non-blocking flush. Not suitable for data-integrity
442 * purposes - I/O may not be started against all dirty pages.
443 *
444 * Return: %0 on success, negative error code otherwise.
445 */
446int filemap_flush(struct address_space *mapping)
447{
448 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
449}
450EXPORT_SYMBOL(filemap_flush);
451
452/**
453 * filemap_range_has_page - check if a page exists in range.
454 * @mapping: address space within which to check
455 * @start_byte: offset in bytes where the range starts
456 * @end_byte: offset in bytes where the range ends (inclusive)
457 *
458 * Find at least one page in the range supplied, usually used to check if
459 * direct writing in this range will trigger a writeback.
460 *
461 * Return: %true if at least one page exists in the specified range,
462 * %false otherwise.
463 */
464bool filemap_range_has_page(struct address_space *mapping,
465 loff_t start_byte, loff_t end_byte)
466{
467 struct page *page;
468 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
469 pgoff_t max = end_byte >> PAGE_SHIFT;
470
471 if (end_byte < start_byte)
472 return false;
473
474 rcu_read_lock();
475 for (;;) {
476 page = xas_find(&xas, max);
477 if (xas_retry(&xas, page))
478 continue;
479 /* Shadow entries don't count */
480 if (xa_is_value(page))
481 continue;
482 /*
483 * We don't need to try to pin this page; we're about to
484 * release the RCU lock anyway. It is enough to know that
485 * there was a page here recently.
486 */
487 break;
488 }
489 rcu_read_unlock();
490
491 return page != NULL;
492}
493EXPORT_SYMBOL(filemap_range_has_page);
494
495static void __filemap_fdatawait_range(struct address_space *mapping,
496 loff_t start_byte, loff_t end_byte)
497{
498 pgoff_t index = start_byte >> PAGE_SHIFT;
499 pgoff_t end = end_byte >> PAGE_SHIFT;
500 struct pagevec pvec;
501 int nr_pages;
502
503 if (end_byte < start_byte)
504 return;
505
506 pagevec_init(&pvec);
507 while (index <= end) {
508 unsigned i;
509
510 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
511 end, PAGECACHE_TAG_WRITEBACK);
512 if (!nr_pages)
513 break;
514
515 for (i = 0; i < nr_pages; i++) {
516 struct page *page = pvec.pages[i];
517
518 wait_on_page_writeback(page);
519 ClearPageError(page);
520 }
521 pagevec_release(&pvec);
522 cond_resched();
523 }
524}
525
526/**
527 * filemap_fdatawait_range - wait for writeback to complete
528 * @mapping: address space structure to wait for
529 * @start_byte: offset in bytes where the range starts
530 * @end_byte: offset in bytes where the range ends (inclusive)
531 *
532 * Walk the list of under-writeback pages of the given address space
533 * in the given range and wait for all of them. Check error status of
534 * the address space and return it.
535 *
536 * Since the error status of the address space is cleared by this function,
537 * callers are responsible for checking the return value and handling and/or
538 * reporting the error.
539 *
540 * Return: error status of the address space.
541 */
542int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
543 loff_t end_byte)
544{
545 __filemap_fdatawait_range(mapping, start_byte, end_byte);
546 return filemap_check_errors(mapping);
547}
548EXPORT_SYMBOL(filemap_fdatawait_range);
549
550/**
551 * file_fdatawait_range - wait for writeback to complete
552 * @file: file pointing to address space structure to wait for
553 * @start_byte: offset in bytes where the range starts
554 * @end_byte: offset in bytes where the range ends (inclusive)
555 *
556 * Walk the list of under-writeback pages of the address space that file
557 * refers to, in the given range and wait for all of them. Check error
558 * status of the address space vs. the file->f_wb_err cursor and return it.
559 *
560 * Since the error status of the file is advanced by this function,
561 * callers are responsible for checking the return value and handling and/or
562 * reporting the error.
563 *
564 * Return: error status of the address space vs. the file->f_wb_err cursor.
565 */
566int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
567{
568 struct address_space *mapping = file->f_mapping;
569
570 __filemap_fdatawait_range(mapping, start_byte, end_byte);
571 return file_check_and_advance_wb_err(file);
572}
573EXPORT_SYMBOL(file_fdatawait_range);
574
575/**
576 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
577 * @mapping: address space structure to wait for
578 *
579 * Walk the list of under-writeback pages of the given address space
580 * and wait for all of them. Unlike filemap_fdatawait(), this function
581 * does not clear error status of the address space.
582 *
583 * Use this function if callers don't handle errors themselves. Expected
584 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
585 * fsfreeze(8)
586 *
587 * Return: error status of the address space.
588 */
589int filemap_fdatawait_keep_errors(struct address_space *mapping)
590{
591 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
592 return filemap_check_and_keep_errors(mapping);
593}
594EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
595
596static bool mapping_needs_writeback(struct address_space *mapping)
597{
598 return (!dax_mapping(mapping) && mapping->nrpages) ||
599 (dax_mapping(mapping) && mapping->nrexceptional);
600}
601
602int filemap_write_and_wait(struct address_space *mapping)
603{
604 int err = 0;
605
606 if (mapping_needs_writeback(mapping)) {
607 err = filemap_fdatawrite(mapping);
608 /*
609 * Even if the above returned error, the pages may be
610 * written partially (e.g. -ENOSPC), so we wait for it.
611 * But the -EIO is special case, it may indicate the worst
612 * thing (e.g. bug) happened, so we avoid waiting for it.
613 */
614 if (err != -EIO) {
615 int err2 = filemap_fdatawait(mapping);
616 if (!err)
617 err = err2;
618 } else {
619 /* Clear any previously stored errors */
620 filemap_check_errors(mapping);
621 }
622 } else {
623 err = filemap_check_errors(mapping);
624 }
625 return err;
626}
627EXPORT_SYMBOL(filemap_write_and_wait);
628
629/**
630 * filemap_write_and_wait_range - write out & wait on a file range
631 * @mapping: the address_space for the pages
632 * @lstart: offset in bytes where the range starts
633 * @lend: offset in bytes where the range ends (inclusive)
634 *
635 * Write out and wait upon file offsets lstart->lend, inclusive.
636 *
637 * Note that @lend is inclusive (describes the last byte to be written) so
638 * that this function can be used to write to the very end-of-file (end = -1).
639 *
640 * Return: error status of the address space.
641 */
642int filemap_write_and_wait_range(struct address_space *mapping,
643 loff_t lstart, loff_t lend)
644{
645 int err = 0;
646
647 if (mapping_needs_writeback(mapping)) {
648 err = __filemap_fdatawrite_range(mapping, lstart, lend,
649 WB_SYNC_ALL);
650 /* See comment of filemap_write_and_wait() */
651 if (err != -EIO) {
652 int err2 = filemap_fdatawait_range(mapping,
653 lstart, lend);
654 if (!err)
655 err = err2;
656 } else {
657 /* Clear any previously stored errors */
658 filemap_check_errors(mapping);
659 }
660 } else {
661 err = filemap_check_errors(mapping);
662 }
663 return err;
664}
665EXPORT_SYMBOL(filemap_write_and_wait_range);
666
667void __filemap_set_wb_err(struct address_space *mapping, int err)
668{
669 errseq_t eseq = errseq_set(&mapping->wb_err, err);
670
671 trace_filemap_set_wb_err(mapping, eseq);
672}
673EXPORT_SYMBOL(__filemap_set_wb_err);
674
675/**
676 * file_check_and_advance_wb_err - report wb error (if any) that was previously
677 * and advance wb_err to current one
678 * @file: struct file on which the error is being reported
679 *
680 * When userland calls fsync (or something like nfsd does the equivalent), we
681 * want to report any writeback errors that occurred since the last fsync (or
682 * since the file was opened if there haven't been any).
683 *
684 * Grab the wb_err from the mapping. If it matches what we have in the file,
685 * then just quickly return 0. The file is all caught up.
686 *
687 * If it doesn't match, then take the mapping value, set the "seen" flag in
688 * it and try to swap it into place. If it works, or another task beat us
689 * to it with the new value, then update the f_wb_err and return the error
690 * portion. The error at this point must be reported via proper channels
691 * (a'la fsync, or NFS COMMIT operation, etc.).
692 *
693 * While we handle mapping->wb_err with atomic operations, the f_wb_err
694 * value is protected by the f_lock since we must ensure that it reflects
695 * the latest value swapped in for this file descriptor.
696 *
697 * Return: %0 on success, negative error code otherwise.
698 */
699int file_check_and_advance_wb_err(struct file *file)
700{
701 int err = 0;
702 errseq_t old = READ_ONCE(file->f_wb_err);
703 struct address_space *mapping = file->f_mapping;
704
705 /* Locklessly handle the common case where nothing has changed */
706 if (errseq_check(&mapping->wb_err, old)) {
707 /* Something changed, must use slow path */
708 spin_lock(&file->f_lock);
709 old = file->f_wb_err;
710 err = errseq_check_and_advance(&mapping->wb_err,
711 &file->f_wb_err);
712 trace_file_check_and_advance_wb_err(file, old);
713 spin_unlock(&file->f_lock);
714 }
715
716 /*
717 * We're mostly using this function as a drop in replacement for
718 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
719 * that the legacy code would have had on these flags.
720 */
721 clear_bit(AS_EIO, &mapping->flags);
722 clear_bit(AS_ENOSPC, &mapping->flags);
723 return err;
724}
725EXPORT_SYMBOL(file_check_and_advance_wb_err);
726
727/**
728 * file_write_and_wait_range - write out & wait on a file range
729 * @file: file pointing to address_space with pages
730 * @lstart: offset in bytes where the range starts
731 * @lend: offset in bytes where the range ends (inclusive)
732 *
733 * Write out and wait upon file offsets lstart->lend, inclusive.
734 *
735 * Note that @lend is inclusive (describes the last byte to be written) so
736 * that this function can be used to write to the very end-of-file (end = -1).
737 *
738 * After writing out and waiting on the data, we check and advance the
739 * f_wb_err cursor to the latest value, and return any errors detected there.
740 *
741 * Return: %0 on success, negative error code otherwise.
742 */
743int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
744{
745 int err = 0, err2;
746 struct address_space *mapping = file->f_mapping;
747
748 if (mapping_needs_writeback(mapping)) {
749 err = __filemap_fdatawrite_range(mapping, lstart, lend,
750 WB_SYNC_ALL);
751 /* See comment of filemap_write_and_wait() */
752 if (err != -EIO)
753 __filemap_fdatawait_range(mapping, lstart, lend);
754 }
755 err2 = file_check_and_advance_wb_err(file);
756 if (!err)
757 err = err2;
758 return err;
759}
760EXPORT_SYMBOL(file_write_and_wait_range);
761
762/**
763 * replace_page_cache_page - replace a pagecache page with a new one
764 * @old: page to be replaced
765 * @new: page to replace with
766 * @gfp_mask: allocation mode
767 *
768 * This function replaces a page in the pagecache with a new one. On
769 * success it acquires the pagecache reference for the new page and
770 * drops it for the old page. Both the old and new pages must be
771 * locked. This function does not add the new page to the LRU, the
772 * caller must do that.
773 *
774 * The remove + add is atomic. This function cannot fail.
775 *
776 * Return: %0
777 */
778int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
779{
780 struct address_space *mapping = old->mapping;
781 void (*freepage)(struct page *) = mapping->a_ops->freepage;
782 pgoff_t offset = old->index;
783 XA_STATE(xas, &mapping->i_pages, offset);
784 unsigned long flags;
785
786 VM_BUG_ON_PAGE(!PageLocked(old), old);
787 VM_BUG_ON_PAGE(!PageLocked(new), new);
788 VM_BUG_ON_PAGE(new->mapping, new);
789
790 get_page(new);
791 new->mapping = mapping;
792 new->index = offset;
793
794 xas_lock_irqsave(&xas, flags);
795 xas_store(&xas, new);
796
797 old->mapping = NULL;
798 /* hugetlb pages do not participate in page cache accounting. */
799 if (!PageHuge(old))
800 __dec_node_page_state(new, NR_FILE_PAGES);
801 if (!PageHuge(new))
802 __inc_node_page_state(new, NR_FILE_PAGES);
803 if (PageSwapBacked(old))
804 __dec_node_page_state(new, NR_SHMEM);
805 if (PageSwapBacked(new))
806 __inc_node_page_state(new, NR_SHMEM);
807 xas_unlock_irqrestore(&xas, flags);
808 mem_cgroup_migrate(old, new);
809 if (freepage)
810 freepage(old);
811 put_page(old);
812
813 return 0;
814}
815EXPORT_SYMBOL_GPL(replace_page_cache_page);
816
817static int __add_to_page_cache_locked(struct page *page,
818 struct address_space *mapping,
819 pgoff_t offset, gfp_t gfp_mask,
820 void **shadowp)
821{
822 XA_STATE(xas, &mapping->i_pages, offset);
823 int huge = PageHuge(page);
824 struct mem_cgroup *memcg;
825 int error;
826 void *old;
827
828 VM_BUG_ON_PAGE(!PageLocked(page), page);
829 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
830 mapping_set_update(&xas, mapping);
831
832 if (!huge) {
833 error = mem_cgroup_try_charge(page, current->mm,
834 gfp_mask, &memcg, false);
835 if (error)
836 return error;
837 }
838
839 get_page(page);
840 page->mapping = mapping;
841 page->index = offset;
842
843 do {
844 xas_lock_irq(&xas);
845 old = xas_load(&xas);
846 if (old && !xa_is_value(old))
847 xas_set_err(&xas, -EEXIST);
848 xas_store(&xas, page);
849 if (xas_error(&xas))
850 goto unlock;
851
852 if (xa_is_value(old)) {
853 mapping->nrexceptional--;
854 if (shadowp)
855 *shadowp = old;
856 }
857 mapping->nrpages++;
858
859 /* hugetlb pages do not participate in page cache accounting */
860 if (!huge)
861 __inc_node_page_state(page, NR_FILE_PAGES);
862unlock:
863 xas_unlock_irq(&xas);
864 } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
865
866 if (xas_error(&xas))
867 goto error;
868
869 if (!huge)
870 mem_cgroup_commit_charge(page, memcg, false, false);
871 trace_mm_filemap_add_to_page_cache(page);
872 return 0;
873error:
874 page->mapping = NULL;
875 /* Leave page->index set: truncation relies upon it */
876 if (!huge)
877 mem_cgroup_cancel_charge(page, memcg, false);
878 put_page(page);
879 return xas_error(&xas);
880}
881
882/**
883 * add_to_page_cache_locked - add a locked page to the pagecache
884 * @page: page to add
885 * @mapping: the page's address_space
886 * @offset: page index
887 * @gfp_mask: page allocation mode
888 *
889 * This function is used to add a page to the pagecache. It must be locked.
890 * This function does not add the page to the LRU. The caller must do that.
891 *
892 * Return: %0 on success, negative error code otherwise.
893 */
894int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
895 pgoff_t offset, gfp_t gfp_mask)
896{
897 return __add_to_page_cache_locked(page, mapping, offset,
898 gfp_mask, NULL);
899}
900EXPORT_SYMBOL(add_to_page_cache_locked);
901
902int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
903 pgoff_t offset, gfp_t gfp_mask)
904{
905 void *shadow = NULL;
906 int ret;
907
908 __SetPageLocked(page);
909 ret = __add_to_page_cache_locked(page, mapping, offset,
910 gfp_mask, &shadow);
911 if (unlikely(ret))
912 __ClearPageLocked(page);
913 else {
914 /*
915 * The page might have been evicted from cache only
916 * recently, in which case it should be activated like
917 * any other repeatedly accessed page.
918 * The exception is pages getting rewritten; evicting other
919 * data from the working set, only to cache data that will
920 * get overwritten with something else, is a waste of memory.
921 */
922 WARN_ON_ONCE(PageActive(page));
923 if (!(gfp_mask & __GFP_WRITE) && shadow)
924 workingset_refault(page, shadow);
925 lru_cache_add(page);
926 }
927 return ret;
928}
929EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
930
931#ifdef CONFIG_NUMA
932struct page *__page_cache_alloc(gfp_t gfp)
933{
934 int n;
935 struct page *page;
936
937 if (cpuset_do_page_mem_spread()) {
938 unsigned int cpuset_mems_cookie;
939 do {
940 cpuset_mems_cookie = read_mems_allowed_begin();
941 n = cpuset_mem_spread_node();
942 page = __alloc_pages_node(n, gfp, 0);
943 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
944
945 return page;
946 }
947 return alloc_pages(gfp, 0);
948}
949EXPORT_SYMBOL(__page_cache_alloc);
950#endif
951
952/*
953 * In order to wait for pages to become available there must be
954 * waitqueues associated with pages. By using a hash table of
955 * waitqueues where the bucket discipline is to maintain all
956 * waiters on the same queue and wake all when any of the pages
957 * become available, and for the woken contexts to check to be
958 * sure the appropriate page became available, this saves space
959 * at a cost of "thundering herd" phenomena during rare hash
960 * collisions.
961 */
962#define PAGE_WAIT_TABLE_BITS 8
963#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
964static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
965
966static wait_queue_head_t *page_waitqueue(struct page *page)
967{
968 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
969}
970
971void __init pagecache_init(void)
972{
973 int i;
974
975 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
976 init_waitqueue_head(&page_wait_table[i]);
977
978 page_writeback_init();
979}
980
981/* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
982struct wait_page_key {
983 struct page *page;
984 int bit_nr;
985 int page_match;
986};
987
988struct wait_page_queue {
989 struct page *page;
990 int bit_nr;
991 wait_queue_entry_t wait;
992};
993
994static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
995{
996 struct wait_page_key *key = arg;
997 struct wait_page_queue *wait_page
998 = container_of(wait, struct wait_page_queue, wait);
999
1000 if (wait_page->page != key->page)
1001 return 0;
1002 key->page_match = 1;
1003
1004 if (wait_page->bit_nr != key->bit_nr)
1005 return 0;
1006
1007 /*
1008 * Stop walking if it's locked.
1009 * Is this safe if put_and_wait_on_page_locked() is in use?
1010 * Yes: the waker must hold a reference to this page, and if PG_locked
1011 * has now already been set by another task, that task must also hold
1012 * a reference to the *same usage* of this page; so there is no need
1013 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1014 */
1015 if (test_bit(key->bit_nr, &key->page->flags))
1016 return -1;
1017
1018 return autoremove_wake_function(wait, mode, sync, key);
1019}
1020
1021static void wake_up_page_bit(struct page *page, int bit_nr)
1022{
1023 wait_queue_head_t *q = page_waitqueue(page);
1024 struct wait_page_key key;
1025 unsigned long flags;
1026 wait_queue_entry_t bookmark;
1027
1028 key.page = page;
1029 key.bit_nr = bit_nr;
1030 key.page_match = 0;
1031
1032 bookmark.flags = 0;
1033 bookmark.private = NULL;
1034 bookmark.func = NULL;
1035 INIT_LIST_HEAD(&bookmark.entry);
1036
1037 spin_lock_irqsave(&q->lock, flags);
1038 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1039
1040 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1041 /*
1042 * Take a breather from holding the lock,
1043 * allow pages that finish wake up asynchronously
1044 * to acquire the lock and remove themselves
1045 * from wait queue
1046 */
1047 spin_unlock_irqrestore(&q->lock, flags);
1048 cpu_relax();
1049 spin_lock_irqsave(&q->lock, flags);
1050 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1051 }
1052
1053 /*
1054 * It is possible for other pages to have collided on the waitqueue
1055 * hash, so in that case check for a page match. That prevents a long-
1056 * term waiter
1057 *
1058 * It is still possible to miss a case here, when we woke page waiters
1059 * and removed them from the waitqueue, but there are still other
1060 * page waiters.
1061 */
1062 if (!waitqueue_active(q) || !key.page_match) {
1063 ClearPageWaiters(page);
1064 /*
1065 * It's possible to miss clearing Waiters here, when we woke
1066 * our page waiters, but the hashed waitqueue has waiters for
1067 * other pages on it.
1068 *
1069 * That's okay, it's a rare case. The next waker will clear it.
1070 */
1071 }
1072 spin_unlock_irqrestore(&q->lock, flags);
1073}
1074
1075static void wake_up_page(struct page *page, int bit)
1076{
1077 if (!PageWaiters(page))
1078 return;
1079 wake_up_page_bit(page, bit);
1080}
1081
1082/*
1083 * A choice of three behaviors for wait_on_page_bit_common():
1084 */
1085enum behavior {
1086 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1087 * __lock_page() waiting on then setting PG_locked.
1088 */
1089 SHARED, /* Hold ref to page and check the bit when woken, like
1090 * wait_on_page_writeback() waiting on PG_writeback.
1091 */
1092 DROP, /* Drop ref to page before wait, no check when woken,
1093 * like put_and_wait_on_page_locked() on PG_locked.
1094 */
1095};
1096
1097static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1098 struct page *page, int bit_nr, int state, enum behavior behavior)
1099{
1100 struct wait_page_queue wait_page;
1101 wait_queue_entry_t *wait = &wait_page.wait;
1102 bool bit_is_set;
1103 bool thrashing = false;
1104 bool delayacct = false;
1105 unsigned long pflags;
1106 int ret = 0;
1107
1108 if (bit_nr == PG_locked &&
1109 !PageUptodate(page) && PageWorkingset(page)) {
1110 if (!PageSwapBacked(page)) {
1111 delayacct_thrashing_start();
1112 delayacct = true;
1113 }
1114 psi_memstall_enter(&pflags);
1115 thrashing = true;
1116 }
1117
1118 init_wait(wait);
1119 wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1120 wait->func = wake_page_function;
1121 wait_page.page = page;
1122 wait_page.bit_nr = bit_nr;
1123
1124 for (;;) {
1125 spin_lock_irq(&q->lock);
1126
1127 if (likely(list_empty(&wait->entry))) {
1128 __add_wait_queue_entry_tail(q, wait);
1129 SetPageWaiters(page);
1130 }
1131
1132 set_current_state(state);
1133
1134 spin_unlock_irq(&q->lock);
1135
1136 bit_is_set = test_bit(bit_nr, &page->flags);
1137 if (behavior == DROP)
1138 put_page(page);
1139
1140 if (likely(bit_is_set))
1141 io_schedule();
1142
1143 if (behavior == EXCLUSIVE) {
1144 if (!test_and_set_bit_lock(bit_nr, &page->flags))
1145 break;
1146 } else if (behavior == SHARED) {
1147 if (!test_bit(bit_nr, &page->flags))
1148 break;
1149 }
1150
1151 if (signal_pending_state(state, current)) {
1152 ret = -EINTR;
1153 break;
1154 }
1155
1156 if (behavior == DROP) {
1157 /*
1158 * We can no longer safely access page->flags:
1159 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1160 * there is a risk of waiting forever on a page reused
1161 * for something that keeps it locked indefinitely.
1162 * But best check for -EINTR above before breaking.
1163 */
1164 break;
1165 }
1166 }
1167
1168 finish_wait(q, wait);
1169
1170 if (thrashing) {
1171 if (delayacct)
1172 delayacct_thrashing_end();
1173 psi_memstall_leave(&pflags);
1174 }
1175
1176 /*
1177 * A signal could leave PageWaiters set. Clearing it here if
1178 * !waitqueue_active would be possible (by open-coding finish_wait),
1179 * but still fail to catch it in the case of wait hash collision. We
1180 * already can fail to clear wait hash collision cases, so don't
1181 * bother with signals either.
1182 */
1183
1184 return ret;
1185}
1186
1187void wait_on_page_bit(struct page *page, int bit_nr)
1188{
1189 wait_queue_head_t *q = page_waitqueue(page);
1190 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1191}
1192EXPORT_SYMBOL(wait_on_page_bit);
1193
1194int wait_on_page_bit_killable(struct page *page, int bit_nr)
1195{
1196 wait_queue_head_t *q = page_waitqueue(page);
1197 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1198}
1199EXPORT_SYMBOL(wait_on_page_bit_killable);
1200
1201/**
1202 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1203 * @page: The page to wait for.
1204 *
1205 * The caller should hold a reference on @page. They expect the page to
1206 * become unlocked relatively soon, but do not wish to hold up migration
1207 * (for example) by holding the reference while waiting for the page to
1208 * come unlocked. After this function returns, the caller should not
1209 * dereference @page.
1210 */
1211void put_and_wait_on_page_locked(struct page *page)
1212{
1213 wait_queue_head_t *q;
1214
1215 page = compound_head(page);
1216 q = page_waitqueue(page);
1217 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1218}
1219
1220/**
1221 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1222 * @page: Page defining the wait queue of interest
1223 * @waiter: Waiter to add to the queue
1224 *
1225 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1226 */
1227void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1228{
1229 wait_queue_head_t *q = page_waitqueue(page);
1230 unsigned long flags;
1231
1232 spin_lock_irqsave(&q->lock, flags);
1233 __add_wait_queue_entry_tail(q, waiter);
1234 SetPageWaiters(page);
1235 spin_unlock_irqrestore(&q->lock, flags);
1236}
1237EXPORT_SYMBOL_GPL(add_page_wait_queue);
1238
1239#ifndef clear_bit_unlock_is_negative_byte
1240
1241/*
1242 * PG_waiters is the high bit in the same byte as PG_lock.
1243 *
1244 * On x86 (and on many other architectures), we can clear PG_lock and
1245 * test the sign bit at the same time. But if the architecture does
1246 * not support that special operation, we just do this all by hand
1247 * instead.
1248 *
1249 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1250 * being cleared, but a memory barrier should be unneccssary since it is
1251 * in the same byte as PG_locked.
1252 */
1253static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1254{
1255 clear_bit_unlock(nr, mem);
1256 /* smp_mb__after_atomic(); */
1257 return test_bit(PG_waiters, mem);
1258}
1259
1260#endif
1261
1262/**
1263 * unlock_page - unlock a locked page
1264 * @page: the page
1265 *
1266 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1267 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1268 * mechanism between PageLocked pages and PageWriteback pages is shared.
1269 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1270 *
1271 * Note that this depends on PG_waiters being the sign bit in the byte
1272 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1273 * clear the PG_locked bit and test PG_waiters at the same time fairly
1274 * portably (architectures that do LL/SC can test any bit, while x86 can
1275 * test the sign bit).
1276 */
1277void unlock_page(struct page *page)
1278{
1279 BUILD_BUG_ON(PG_waiters != 7);
1280 page = compound_head(page);
1281 VM_BUG_ON_PAGE(!PageLocked(page), page);
1282 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1283 wake_up_page_bit(page, PG_locked);
1284}
1285EXPORT_SYMBOL(unlock_page);
1286
1287/**
1288 * end_page_writeback - end writeback against a page
1289 * @page: the page
1290 */
1291void end_page_writeback(struct page *page)
1292{
1293 /*
1294 * TestClearPageReclaim could be used here but it is an atomic
1295 * operation and overkill in this particular case. Failing to
1296 * shuffle a page marked for immediate reclaim is too mild to
1297 * justify taking an atomic operation penalty at the end of
1298 * ever page writeback.
1299 */
1300 if (PageReclaim(page)) {
1301 ClearPageReclaim(page);
1302 rotate_reclaimable_page(page);
1303 }
1304
1305 if (!test_clear_page_writeback(page))
1306 BUG();
1307
1308 smp_mb__after_atomic();
1309 wake_up_page(page, PG_writeback);
1310}
1311EXPORT_SYMBOL(end_page_writeback);
1312
1313/*
1314 * After completing I/O on a page, call this routine to update the page
1315 * flags appropriately
1316 */
1317void page_endio(struct page *page, bool is_write, int err)
1318{
1319 if (!is_write) {
1320 if (!err) {
1321 SetPageUptodate(page);
1322 } else {
1323 ClearPageUptodate(page);
1324 SetPageError(page);
1325 }
1326 unlock_page(page);
1327 } else {
1328 if (err) {
1329 struct address_space *mapping;
1330
1331 SetPageError(page);
1332 mapping = page_mapping(page);
1333 if (mapping)
1334 mapping_set_error(mapping, err);
1335 }
1336 end_page_writeback(page);
1337 }
1338}
1339EXPORT_SYMBOL_GPL(page_endio);
1340
1341/**
1342 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1343 * @__page: the page to lock
1344 */
1345void __lock_page(struct page *__page)
1346{
1347 struct page *page = compound_head(__page);
1348 wait_queue_head_t *q = page_waitqueue(page);
1349 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1350 EXCLUSIVE);
1351}
1352EXPORT_SYMBOL(__lock_page);
1353
1354int __lock_page_killable(struct page *__page)
1355{
1356 struct page *page = compound_head(__page);
1357 wait_queue_head_t *q = page_waitqueue(page);
1358 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1359 EXCLUSIVE);
1360}
1361EXPORT_SYMBOL_GPL(__lock_page_killable);
1362
1363/*
1364 * Return values:
1365 * 1 - page is locked; mmap_sem is still held.
1366 * 0 - page is not locked.
1367 * mmap_sem has been released (up_read()), unless flags had both
1368 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1369 * which case mmap_sem is still held.
1370 *
1371 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1372 * with the page locked and the mmap_sem unperturbed.
1373 */
1374int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1375 unsigned int flags)
1376{
1377 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1378 /*
1379 * CAUTION! In this case, mmap_sem is not released
1380 * even though return 0.
1381 */
1382 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1383 return 0;
1384
1385 up_read(&mm->mmap_sem);
1386 if (flags & FAULT_FLAG_KILLABLE)
1387 wait_on_page_locked_killable(page);
1388 else
1389 wait_on_page_locked(page);
1390 return 0;
1391 } else {
1392 if (flags & FAULT_FLAG_KILLABLE) {
1393 int ret;
1394
1395 ret = __lock_page_killable(page);
1396 if (ret) {
1397 up_read(&mm->mmap_sem);
1398 return 0;
1399 }
1400 } else
1401 __lock_page(page);
1402 return 1;
1403 }
1404}
1405
1406/**
1407 * page_cache_next_miss() - Find the next gap in the page cache.
1408 * @mapping: Mapping.
1409 * @index: Index.
1410 * @max_scan: Maximum range to search.
1411 *
1412 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1413 * gap with the lowest index.
1414 *
1415 * This function may be called under the rcu_read_lock. However, this will
1416 * not atomically search a snapshot of the cache at a single point in time.
1417 * For example, if a gap is created at index 5, then subsequently a gap is
1418 * created at index 10, page_cache_next_miss covering both indices may
1419 * return 10 if called under the rcu_read_lock.
1420 *
1421 * Return: The index of the gap if found, otherwise an index outside the
1422 * range specified (in which case 'return - index >= max_scan' will be true).
1423 * In the rare case of index wrap-around, 0 will be returned.
1424 */
1425pgoff_t page_cache_next_miss(struct address_space *mapping,
1426 pgoff_t index, unsigned long max_scan)
1427{
1428 XA_STATE(xas, &mapping->i_pages, index);
1429
1430 while (max_scan--) {
1431 void *entry = xas_next(&xas);
1432 if (!entry || xa_is_value(entry))
1433 break;
1434 if (xas.xa_index == 0)
1435 break;
1436 }
1437
1438 return xas.xa_index;
1439}
1440EXPORT_SYMBOL(page_cache_next_miss);
1441
1442/**
1443 * page_cache_prev_miss() - Find the next gap in the page cache.
1444 * @mapping: Mapping.
1445 * @index: Index.
1446 * @max_scan: Maximum range to search.
1447 *
1448 * Search the range [max(index - max_scan + 1, 0), index] for the
1449 * gap with the highest index.
1450 *
1451 * This function may be called under the rcu_read_lock. However, this will
1452 * not atomically search a snapshot of the cache at a single point in time.
1453 * For example, if a gap is created at index 10, then subsequently a gap is
1454 * created at index 5, page_cache_prev_miss() covering both indices may
1455 * return 5 if called under the rcu_read_lock.
1456 *
1457 * Return: The index of the gap if found, otherwise an index outside the
1458 * range specified (in which case 'index - return >= max_scan' will be true).
1459 * In the rare case of wrap-around, ULONG_MAX will be returned.
1460 */
1461pgoff_t page_cache_prev_miss(struct address_space *mapping,
1462 pgoff_t index, unsigned long max_scan)
1463{
1464 XA_STATE(xas, &mapping->i_pages, index);
1465
1466 while (max_scan--) {
1467 void *entry = xas_prev(&xas);
1468 if (!entry || xa_is_value(entry))
1469 break;
1470 if (xas.xa_index == ULONG_MAX)
1471 break;
1472 }
1473
1474 return xas.xa_index;
1475}
1476EXPORT_SYMBOL(page_cache_prev_miss);
1477
1478/**
1479 * find_get_entry - find and get a page cache entry
1480 * @mapping: the address_space to search
1481 * @offset: the page cache index
1482 *
1483 * Looks up the page cache slot at @mapping & @offset. If there is a
1484 * page cache page, it is returned with an increased refcount.
1485 *
1486 * If the slot holds a shadow entry of a previously evicted page, or a
1487 * swap entry from shmem/tmpfs, it is returned.
1488 *
1489 * Return: the found page or shadow entry, %NULL if nothing is found.
1490 */
1491struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1492{
1493 XA_STATE(xas, &mapping->i_pages, offset);
1494 struct page *head, *page;
1495
1496 rcu_read_lock();
1497repeat:
1498 xas_reset(&xas);
1499 page = xas_load(&xas);
1500 if (xas_retry(&xas, page))
1501 goto repeat;
1502 /*
1503 * A shadow entry of a recently evicted page, or a swap entry from
1504 * shmem/tmpfs. Return it without attempting to raise page count.
1505 */
1506 if (!page || xa_is_value(page))
1507 goto out;
1508
1509 head = compound_head(page);
1510 if (!page_cache_get_speculative(head))
1511 goto repeat;
1512
1513 /* The page was split under us? */
1514 if (compound_head(page) != head) {
1515 put_page(head);
1516 goto repeat;
1517 }
1518
1519 /*
1520 * Has the page moved?
1521 * This is part of the lockless pagecache protocol. See
1522 * include/linux/pagemap.h for details.
1523 */
1524 if (unlikely(page != xas_reload(&xas))) {
1525 put_page(head);
1526 goto repeat;
1527 }
1528out:
1529 rcu_read_unlock();
1530
1531 return page;
1532}
1533EXPORT_SYMBOL(find_get_entry);
1534
1535/**
1536 * find_lock_entry - locate, pin and lock a page cache entry
1537 * @mapping: the address_space to search
1538 * @offset: the page cache index
1539 *
1540 * Looks up the page cache slot at @mapping & @offset. If there is a
1541 * page cache page, it is returned locked and with an increased
1542 * refcount.
1543 *
1544 * If the slot holds a shadow entry of a previously evicted page, or a
1545 * swap entry from shmem/tmpfs, it is returned.
1546 *
1547 * find_lock_entry() may sleep.
1548 *
1549 * Return: the found page or shadow entry, %NULL if nothing is found.
1550 */
1551struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1552{
1553 struct page *page;
1554
1555repeat:
1556 page = find_get_entry(mapping, offset);
1557 if (page && !xa_is_value(page)) {
1558 lock_page(page);
1559 /* Has the page been truncated? */
1560 if (unlikely(page_mapping(page) != mapping)) {
1561 unlock_page(page);
1562 put_page(page);
1563 goto repeat;
1564 }
1565 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1566 }
1567 return page;
1568}
1569EXPORT_SYMBOL(find_lock_entry);
1570
1571/**
1572 * pagecache_get_page - find and get a page reference
1573 * @mapping: the address_space to search
1574 * @offset: the page index
1575 * @fgp_flags: PCG flags
1576 * @gfp_mask: gfp mask to use for the page cache data page allocation
1577 *
1578 * Looks up the page cache slot at @mapping & @offset.
1579 *
1580 * PCG flags modify how the page is returned.
1581 *
1582 * @fgp_flags can be:
1583 *
1584 * - FGP_ACCESSED: the page will be marked accessed
1585 * - FGP_LOCK: Page is return locked
1586 * - FGP_CREAT: If page is not present then a new page is allocated using
1587 * @gfp_mask and added to the page cache and the VM's LRU
1588 * list. The page is returned locked and with an increased
1589 * refcount.
1590 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1591 * its own locking dance if the page is already in cache, or unlock the page
1592 * before returning if we had to add the page to pagecache.
1593 *
1594 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1595 * if the GFP flags specified for FGP_CREAT are atomic.
1596 *
1597 * If there is a page cache page, it is returned with an increased refcount.
1598 *
1599 * Return: the found page or %NULL otherwise.
1600 */
1601struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1602 int fgp_flags, gfp_t gfp_mask)
1603{
1604 struct page *page;
1605
1606repeat:
1607 page = find_get_entry(mapping, offset);
1608 if (xa_is_value(page))
1609 page = NULL;
1610 if (!page)
1611 goto no_page;
1612
1613 if (fgp_flags & FGP_LOCK) {
1614 if (fgp_flags & FGP_NOWAIT) {
1615 if (!trylock_page(page)) {
1616 put_page(page);
1617 return NULL;
1618 }
1619 } else {
1620 lock_page(page);
1621 }
1622
1623 /* Has the page been truncated? */
1624 if (unlikely(page->mapping != mapping)) {
1625 unlock_page(page);
1626 put_page(page);
1627 goto repeat;
1628 }
1629 VM_BUG_ON_PAGE(page->index != offset, page);
1630 }
1631
1632 if (fgp_flags & FGP_ACCESSED)
1633 mark_page_accessed(page);
1634
1635no_page:
1636 if (!page && (fgp_flags & FGP_CREAT)) {
1637 int err;
1638 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1639 gfp_mask |= __GFP_WRITE;
1640 if (fgp_flags & FGP_NOFS)
1641 gfp_mask &= ~__GFP_FS;
1642
1643 page = __page_cache_alloc(gfp_mask);
1644 if (!page)
1645 return NULL;
1646
1647 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1648 fgp_flags |= FGP_LOCK;
1649
1650 /* Init accessed so avoid atomic mark_page_accessed later */
1651 if (fgp_flags & FGP_ACCESSED)
1652 __SetPageReferenced(page);
1653
1654 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1655 if (unlikely(err)) {
1656 put_page(page);
1657 page = NULL;
1658 if (err == -EEXIST)
1659 goto repeat;
1660 }
1661
1662 /*
1663 * add_to_page_cache_lru locks the page, and for mmap we expect
1664 * an unlocked page.
1665 */
1666 if (page && (fgp_flags & FGP_FOR_MMAP))
1667 unlock_page(page);
1668 }
1669
1670 return page;
1671}
1672EXPORT_SYMBOL(pagecache_get_page);
1673
1674/**
1675 * find_get_entries - gang pagecache lookup
1676 * @mapping: The address_space to search
1677 * @start: The starting page cache index
1678 * @nr_entries: The maximum number of entries
1679 * @entries: Where the resulting entries are placed
1680 * @indices: The cache indices corresponding to the entries in @entries
1681 *
1682 * find_get_entries() will search for and return a group of up to
1683 * @nr_entries entries in the mapping. The entries are placed at
1684 * @entries. find_get_entries() takes a reference against any actual
1685 * pages it returns.
1686 *
1687 * The search returns a group of mapping-contiguous page cache entries
1688 * with ascending indexes. There may be holes in the indices due to
1689 * not-present pages.
1690 *
1691 * Any shadow entries of evicted pages, or swap entries from
1692 * shmem/tmpfs, are included in the returned array.
1693 *
1694 * Return: the number of pages and shadow entries which were found.
1695 */
1696unsigned find_get_entries(struct address_space *mapping,
1697 pgoff_t start, unsigned int nr_entries,
1698 struct page **entries, pgoff_t *indices)
1699{
1700 XA_STATE(xas, &mapping->i_pages, start);
1701 struct page *page;
1702 unsigned int ret = 0;
1703
1704 if (!nr_entries)
1705 return 0;
1706
1707 rcu_read_lock();
1708 xas_for_each(&xas, page, ULONG_MAX) {
1709 struct page *head;
1710 if (xas_retry(&xas, page))
1711 continue;
1712 /*
1713 * A shadow entry of a recently evicted page, a swap
1714 * entry from shmem/tmpfs or a DAX entry. Return it
1715 * without attempting to raise page count.
1716 */
1717 if (xa_is_value(page))
1718 goto export;
1719
1720 head = compound_head(page);
1721 if (!page_cache_get_speculative(head))
1722 goto retry;
1723
1724 /* The page was split under us? */
1725 if (compound_head(page) != head)
1726 goto put_page;
1727
1728 /* Has the page moved? */
1729 if (unlikely(page != xas_reload(&xas)))
1730 goto put_page;
1731
1732export:
1733 indices[ret] = xas.xa_index;
1734 entries[ret] = page;
1735 if (++ret == nr_entries)
1736 break;
1737 continue;
1738put_page:
1739 put_page(head);
1740retry:
1741 xas_reset(&xas);
1742 }
1743 rcu_read_unlock();
1744 return ret;
1745}
1746
1747/**
1748 * find_get_pages_range - gang pagecache lookup
1749 * @mapping: The address_space to search
1750 * @start: The starting page index
1751 * @end: The final page index (inclusive)
1752 * @nr_pages: The maximum number of pages
1753 * @pages: Where the resulting pages are placed
1754 *
1755 * find_get_pages_range() will search for and return a group of up to @nr_pages
1756 * pages in the mapping starting at index @start and up to index @end
1757 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1758 * a reference against the returned pages.
1759 *
1760 * The search returns a group of mapping-contiguous pages with ascending
1761 * indexes. There may be holes in the indices due to not-present pages.
1762 * We also update @start to index the next page for the traversal.
1763 *
1764 * Return: the number of pages which were found. If this number is
1765 * smaller than @nr_pages, the end of specified range has been
1766 * reached.
1767 */
1768unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1769 pgoff_t end, unsigned int nr_pages,
1770 struct page **pages)
1771{
1772 XA_STATE(xas, &mapping->i_pages, *start);
1773 struct page *page;
1774 unsigned ret = 0;
1775
1776 if (unlikely(!nr_pages))
1777 return 0;
1778
1779 rcu_read_lock();
1780 xas_for_each(&xas, page, end) {
1781 struct page *head;
1782 if (xas_retry(&xas, page))
1783 continue;
1784 /* Skip over shadow, swap and DAX entries */
1785 if (xa_is_value(page))
1786 continue;
1787
1788 head = compound_head(page);
1789 if (!page_cache_get_speculative(head))
1790 goto retry;
1791
1792 /* The page was split under us? */
1793 if (compound_head(page) != head)
1794 goto put_page;
1795
1796 /* Has the page moved? */
1797 if (unlikely(page != xas_reload(&xas)))
1798 goto put_page;
1799
1800 pages[ret] = page;
1801 if (++ret == nr_pages) {
1802 *start = xas.xa_index + 1;
1803 goto out;
1804 }
1805 continue;
1806put_page:
1807 put_page(head);
1808retry:
1809 xas_reset(&xas);
1810 }
1811
1812 /*
1813 * We come here when there is no page beyond @end. We take care to not
1814 * overflow the index @start as it confuses some of the callers. This
1815 * breaks the iteration when there is a page at index -1 but that is
1816 * already broken anyway.
1817 */
1818 if (end == (pgoff_t)-1)
1819 *start = (pgoff_t)-1;
1820 else
1821 *start = end + 1;
1822out:
1823 rcu_read_unlock();
1824
1825 return ret;
1826}
1827
1828/**
1829 * find_get_pages_contig - gang contiguous pagecache lookup
1830 * @mapping: The address_space to search
1831 * @index: The starting page index
1832 * @nr_pages: The maximum number of pages
1833 * @pages: Where the resulting pages are placed
1834 *
1835 * find_get_pages_contig() works exactly like find_get_pages(), except
1836 * that the returned number of pages are guaranteed to be contiguous.
1837 *
1838 * Return: the number of pages which were found.
1839 */
1840unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1841 unsigned int nr_pages, struct page **pages)
1842{
1843 XA_STATE(xas, &mapping->i_pages, index);
1844 struct page *page;
1845 unsigned int ret = 0;
1846
1847 if (unlikely(!nr_pages))
1848 return 0;
1849
1850 rcu_read_lock();
1851 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1852 struct page *head;
1853 if (xas_retry(&xas, page))
1854 continue;
1855 /*
1856 * If the entry has been swapped out, we can stop looking.
1857 * No current caller is looking for DAX entries.
1858 */
1859 if (xa_is_value(page))
1860 break;
1861
1862 head = compound_head(page);
1863 if (!page_cache_get_speculative(head))
1864 goto retry;
1865
1866 /* The page was split under us? */
1867 if (compound_head(page) != head)
1868 goto put_page;
1869
1870 /* Has the page moved? */
1871 if (unlikely(page != xas_reload(&xas)))
1872 goto put_page;
1873
1874 pages[ret] = page;
1875 if (++ret == nr_pages)
1876 break;
1877 continue;
1878put_page:
1879 put_page(head);
1880retry:
1881 xas_reset(&xas);
1882 }
1883 rcu_read_unlock();
1884 return ret;
1885}
1886EXPORT_SYMBOL(find_get_pages_contig);
1887
1888/**
1889 * find_get_pages_range_tag - find and return pages in given range matching @tag
1890 * @mapping: the address_space to search
1891 * @index: the starting page index
1892 * @end: The final page index (inclusive)
1893 * @tag: the tag index
1894 * @nr_pages: the maximum number of pages
1895 * @pages: where the resulting pages are placed
1896 *
1897 * Like find_get_pages, except we only return pages which are tagged with
1898 * @tag. We update @index to index the next page for the traversal.
1899 *
1900 * Return: the number of pages which were found.
1901 */
1902unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1903 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1904 struct page **pages)
1905{
1906 XA_STATE(xas, &mapping->i_pages, *index);
1907 struct page *page;
1908 unsigned ret = 0;
1909
1910 if (unlikely(!nr_pages))
1911 return 0;
1912
1913 rcu_read_lock();
1914 xas_for_each_marked(&xas, page, end, tag) {
1915 struct page *head;
1916 if (xas_retry(&xas, page))
1917 continue;
1918 /*
1919 * Shadow entries should never be tagged, but this iteration
1920 * is lockless so there is a window for page reclaim to evict
1921 * a page we saw tagged. Skip over it.
1922 */
1923 if (xa_is_value(page))
1924 continue;
1925
1926 head = compound_head(page);
1927 if (!page_cache_get_speculative(head))
1928 goto retry;
1929
1930 /* The page was split under us? */
1931 if (compound_head(page) != head)
1932 goto put_page;
1933
1934 /* Has the page moved? */
1935 if (unlikely(page != xas_reload(&xas)))
1936 goto put_page;
1937
1938 pages[ret] = page;
1939 if (++ret == nr_pages) {
1940 *index = xas.xa_index + 1;
1941 goto out;
1942 }
1943 continue;
1944put_page:
1945 put_page(head);
1946retry:
1947 xas_reset(&xas);
1948 }
1949
1950 /*
1951 * We come here when we got to @end. We take care to not overflow the
1952 * index @index as it confuses some of the callers. This breaks the
1953 * iteration when there is a page at index -1 but that is already
1954 * broken anyway.
1955 */
1956 if (end == (pgoff_t)-1)
1957 *index = (pgoff_t)-1;
1958 else
1959 *index = end + 1;
1960out:
1961 rcu_read_unlock();
1962
1963 return ret;
1964}
1965EXPORT_SYMBOL(find_get_pages_range_tag);
1966
1967/**
1968 * find_get_entries_tag - find and return entries that match @tag
1969 * @mapping: the address_space to search
1970 * @start: the starting page cache index
1971 * @tag: the tag index
1972 * @nr_entries: the maximum number of entries
1973 * @entries: where the resulting entries are placed
1974 * @indices: the cache indices corresponding to the entries in @entries
1975 *
1976 * Like find_get_entries, except we only return entries which are tagged with
1977 * @tag.
1978 *
1979 * Return: the number of entries which were found.
1980 */
1981unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1982 xa_mark_t tag, unsigned int nr_entries,
1983 struct page **entries, pgoff_t *indices)
1984{
1985 XA_STATE(xas, &mapping->i_pages, start);
1986 struct page *page;
1987 unsigned int ret = 0;
1988
1989 if (!nr_entries)
1990 return 0;
1991
1992 rcu_read_lock();
1993 xas_for_each_marked(&xas, page, ULONG_MAX, tag) {
1994 struct page *head;
1995 if (xas_retry(&xas, page))
1996 continue;
1997 /*
1998 * A shadow entry of a recently evicted page, a swap
1999 * entry from shmem/tmpfs or a DAX entry. Return it
2000 * without attempting to raise page count.
2001 */
2002 if (xa_is_value(page))
2003 goto export;
2004
2005 head = compound_head(page);
2006 if (!page_cache_get_speculative(head))
2007 goto retry;
2008
2009 /* The page was split under us? */
2010 if (compound_head(page) != head)
2011 goto put_page;
2012
2013 /* Has the page moved? */
2014 if (unlikely(page != xas_reload(&xas)))
2015 goto put_page;
2016
2017export:
2018 indices[ret] = xas.xa_index;
2019 entries[ret] = page;
2020 if (++ret == nr_entries)
2021 break;
2022 continue;
2023put_page:
2024 put_page(head);
2025retry:
2026 xas_reset(&xas);
2027 }
2028 rcu_read_unlock();
2029 return ret;
2030}
2031EXPORT_SYMBOL(find_get_entries_tag);
2032
2033/*
2034 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2035 * a _large_ part of the i/o request. Imagine the worst scenario:
2036 *
2037 * ---R__________________________________________B__________
2038 * ^ reading here ^ bad block(assume 4k)
2039 *
2040 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2041 * => failing the whole request => read(R) => read(R+1) =>
2042 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2043 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2044 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2045 *
2046 * It is going insane. Fix it by quickly scaling down the readahead size.
2047 */
2048static void shrink_readahead_size_eio(struct file *filp,
2049 struct file_ra_state *ra)
2050{
2051 ra->ra_pages /= 4;
2052}
2053
2054/**
2055 * generic_file_buffered_read - generic file read routine
2056 * @iocb: the iocb to read
2057 * @iter: data destination
2058 * @written: already copied
2059 *
2060 * This is a generic file read routine, and uses the
2061 * mapping->a_ops->readpage() function for the actual low-level stuff.
2062 *
2063 * This is really ugly. But the goto's actually try to clarify some
2064 * of the logic when it comes to error handling etc.
2065 *
2066 * Return:
2067 * * total number of bytes copied, including those the were already @written
2068 * * negative error code if nothing was copied
2069 */
2070static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2071 struct iov_iter *iter, ssize_t written)
2072{
2073 struct file *filp = iocb->ki_filp;
2074 struct address_space *mapping = filp->f_mapping;
2075 struct inode *inode = mapping->host;
2076 struct file_ra_state *ra = &filp->f_ra;
2077 loff_t *ppos = &iocb->ki_pos;
2078 pgoff_t index;
2079 pgoff_t last_index;
2080 pgoff_t prev_index;
2081 unsigned long offset; /* offset into pagecache page */
2082 unsigned int prev_offset;
2083 int error = 0;
2084
2085 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2086 return 0;
2087 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2088
2089 index = *ppos >> PAGE_SHIFT;
2090 prev_index = ra->prev_pos >> PAGE_SHIFT;
2091 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2092 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2093 offset = *ppos & ~PAGE_MASK;
2094
2095 for (;;) {
2096 struct page *page;
2097 pgoff_t end_index;
2098 loff_t isize;
2099 unsigned long nr, ret;
2100
2101 cond_resched();
2102find_page:
2103 if (fatal_signal_pending(current)) {
2104 error = -EINTR;
2105 goto out;
2106 }
2107
2108 page = find_get_page(mapping, index);
2109 if (!page) {
2110 if (iocb->ki_flags & IOCB_NOWAIT)
2111 goto would_block;
2112 page_cache_sync_readahead(mapping,
2113 ra, filp,
2114 index, last_index - index);
2115 page = find_get_page(mapping, index);
2116 if (unlikely(page == NULL))
2117 goto no_cached_page;
2118 }
2119 if (PageReadahead(page)) {
2120 page_cache_async_readahead(mapping,
2121 ra, filp, page,
2122 index, last_index - index);
2123 }
2124 if (!PageUptodate(page)) {
2125 if (iocb->ki_flags & IOCB_NOWAIT) {
2126 put_page(page);
2127 goto would_block;
2128 }
2129
2130 /*
2131 * See comment in do_read_cache_page on why
2132 * wait_on_page_locked is used to avoid unnecessarily
2133 * serialisations and why it's safe.
2134 */
2135 error = wait_on_page_locked_killable(page);
2136 if (unlikely(error))
2137 goto readpage_error;
2138 if (PageUptodate(page))
2139 goto page_ok;
2140
2141 if (inode->i_blkbits == PAGE_SHIFT ||
2142 !mapping->a_ops->is_partially_uptodate)
2143 goto page_not_up_to_date;
2144 /* pipes can't handle partially uptodate pages */
2145 if (unlikely(iov_iter_is_pipe(iter)))
2146 goto page_not_up_to_date;
2147 if (!trylock_page(page))
2148 goto page_not_up_to_date;
2149 /* Did it get truncated before we got the lock? */
2150 if (!page->mapping)
2151 goto page_not_up_to_date_locked;
2152 if (!mapping->a_ops->is_partially_uptodate(page,
2153 offset, iter->count))
2154 goto page_not_up_to_date_locked;
2155 unlock_page(page);
2156 }
2157page_ok:
2158 /*
2159 * i_size must be checked after we know the page is Uptodate.
2160 *
2161 * Checking i_size after the check allows us to calculate
2162 * the correct value for "nr", which means the zero-filled
2163 * part of the page is not copied back to userspace (unless
2164 * another truncate extends the file - this is desired though).
2165 */
2166
2167 isize = i_size_read(inode);
2168 end_index = (isize - 1) >> PAGE_SHIFT;
2169 if (unlikely(!isize || index > end_index)) {
2170 put_page(page);
2171 goto out;
2172 }
2173
2174 /* nr is the maximum number of bytes to copy from this page */
2175 nr = PAGE_SIZE;
2176 if (index == end_index) {
2177 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2178 if (nr <= offset) {
2179 put_page(page);
2180 goto out;
2181 }
2182 }
2183 nr = nr - offset;
2184
2185 /* If users can be writing to this page using arbitrary
2186 * virtual addresses, take care about potential aliasing
2187 * before reading the page on the kernel side.
2188 */
2189 if (mapping_writably_mapped(mapping))
2190 flush_dcache_page(page);
2191
2192 /*
2193 * When a sequential read accesses a page several times,
2194 * only mark it as accessed the first time.
2195 */
2196 if (prev_index != index || offset != prev_offset)
2197 mark_page_accessed(page);
2198 prev_index = index;
2199
2200 /*
2201 * Ok, we have the page, and it's up-to-date, so
2202 * now we can copy it to user space...
2203 */
2204
2205 ret = copy_page_to_iter(page, offset, nr, iter);
2206 offset += ret;
2207 index += offset >> PAGE_SHIFT;
2208 offset &= ~PAGE_MASK;
2209 prev_offset = offset;
2210
2211 put_page(page);
2212 written += ret;
2213 if (!iov_iter_count(iter))
2214 goto out;
2215 if (ret < nr) {
2216 error = -EFAULT;
2217 goto out;
2218 }
2219 continue;
2220
2221page_not_up_to_date:
2222 /* Get exclusive access to the page ... */
2223 error = lock_page_killable(page);
2224 if (unlikely(error))
2225 goto readpage_error;
2226
2227page_not_up_to_date_locked:
2228 /* Did it get truncated before we got the lock? */
2229 if (!page->mapping) {
2230 unlock_page(page);
2231 put_page(page);
2232 continue;
2233 }
2234
2235 /* Did somebody else fill it already? */
2236 if (PageUptodate(page)) {
2237 unlock_page(page);
2238 goto page_ok;
2239 }
2240
2241readpage:
2242 /*
2243 * A previous I/O error may have been due to temporary
2244 * failures, eg. multipath errors.
2245 * PG_error will be set again if readpage fails.
2246 */
2247 ClearPageError(page);
2248 /* Start the actual read. The read will unlock the page. */
2249 error = mapping->a_ops->readpage(filp, page);
2250
2251 if (unlikely(error)) {
2252 if (error == AOP_TRUNCATED_PAGE) {
2253 put_page(page);
2254 error = 0;
2255 goto find_page;
2256 }
2257 goto readpage_error;
2258 }
2259
2260 if (!PageUptodate(page)) {
2261 error = lock_page_killable(page);
2262 if (unlikely(error))
2263 goto readpage_error;
2264 if (!PageUptodate(page)) {
2265 if (page->mapping == NULL) {
2266 /*
2267 * invalidate_mapping_pages got it
2268 */
2269 unlock_page(page);
2270 put_page(page);
2271 goto find_page;
2272 }
2273 unlock_page(page);
2274 shrink_readahead_size_eio(filp, ra);
2275 error = -EIO;
2276 goto readpage_error;
2277 }
2278 unlock_page(page);
2279 }
2280
2281 goto page_ok;
2282
2283readpage_error:
2284 /* UHHUH! A synchronous read error occurred. Report it */
2285 put_page(page);
2286 goto out;
2287
2288no_cached_page:
2289 /*
2290 * Ok, it wasn't cached, so we need to create a new
2291 * page..
2292 */
2293 page = page_cache_alloc(mapping);
2294 if (!page) {
2295 error = -ENOMEM;
2296 goto out;
2297 }
2298 error = add_to_page_cache_lru(page, mapping, index,
2299 mapping_gfp_constraint(mapping, GFP_KERNEL));
2300 if (error) {
2301 put_page(page);
2302 if (error == -EEXIST) {
2303 error = 0;
2304 goto find_page;
2305 }
2306 goto out;
2307 }
2308 goto readpage;
2309 }
2310
2311would_block:
2312 error = -EAGAIN;
2313out:
2314 ra->prev_pos = prev_index;
2315 ra->prev_pos <<= PAGE_SHIFT;
2316 ra->prev_pos |= prev_offset;
2317
2318 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2319 file_accessed(filp);
2320 return written ? written : error;
2321}
2322
2323/**
2324 * generic_file_read_iter - generic filesystem read routine
2325 * @iocb: kernel I/O control block
2326 * @iter: destination for the data read
2327 *
2328 * This is the "read_iter()" routine for all filesystems
2329 * that can use the page cache directly.
2330 * Return:
2331 * * number of bytes copied, even for partial reads
2332 * * negative error code if nothing was read
2333 */
2334ssize_t
2335generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2336{
2337 size_t count = iov_iter_count(iter);
2338 ssize_t retval = 0;
2339
2340 if (!count)
2341 goto out; /* skip atime */
2342
2343 if (iocb->ki_flags & IOCB_DIRECT) {
2344 struct file *file = iocb->ki_filp;
2345 struct address_space *mapping = file->f_mapping;
2346 struct inode *inode = mapping->host;
2347 loff_t size;
2348
2349 size = i_size_read(inode);
2350 if (iocb->ki_flags & IOCB_NOWAIT) {
2351 if (filemap_range_has_page(mapping, iocb->ki_pos,
2352 iocb->ki_pos + count - 1))
2353 return -EAGAIN;
2354 } else {
2355 retval = filemap_write_and_wait_range(mapping,
2356 iocb->ki_pos,
2357 iocb->ki_pos + count - 1);
2358 if (retval < 0)
2359 goto out;
2360 }
2361
2362 file_accessed(file);
2363
2364 retval = mapping->a_ops->direct_IO(iocb, iter);
2365 if (retval >= 0) {
2366 iocb->ki_pos += retval;
2367 count -= retval;
2368 }
2369 iov_iter_revert(iter, count - iov_iter_count(iter));
2370
2371 /*
2372 * Btrfs can have a short DIO read if we encounter
2373 * compressed extents, so if there was an error, or if
2374 * we've already read everything we wanted to, or if
2375 * there was a short read because we hit EOF, go ahead
2376 * and return. Otherwise fallthrough to buffered io for
2377 * the rest of the read. Buffered reads will not work for
2378 * DAX files, so don't bother trying.
2379 */
2380 if (retval < 0 || !count || iocb->ki_pos >= size ||
2381 IS_DAX(inode))
2382 goto out;
2383 }
2384
2385 retval = generic_file_buffered_read(iocb, iter, retval);
2386out:
2387 return retval;
2388}
2389EXPORT_SYMBOL(generic_file_read_iter);
2390
2391#ifdef CONFIG_MMU
2392#define MMAP_LOTSAMISS (100)
2393static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
2394 struct file *fpin)
2395{
2396 int flags = vmf->flags;
2397
2398 if (fpin)
2399 return fpin;
2400
2401 /*
2402 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
2403 * anything, so we only pin the file and drop the mmap_sem if only
2404 * FAULT_FLAG_ALLOW_RETRY is set.
2405 */
2406 if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
2407 FAULT_FLAG_ALLOW_RETRY) {
2408 fpin = get_file(vmf->vma->vm_file);
2409 up_read(&vmf->vma->vm_mm->mmap_sem);
2410 }
2411 return fpin;
2412}
2413
2414/*
2415 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2416 * @vmf - the vm_fault for this fault.
2417 * @page - the page to lock.
2418 * @fpin - the pointer to the file we may pin (or is already pinned).
2419 *
2420 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2421 * It differs in that it actually returns the page locked if it returns 1 and 0
2422 * if it couldn't lock the page. If we did have to drop the mmap_sem then fpin
2423 * will point to the pinned file and needs to be fput()'ed at a later point.
2424 */
2425static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2426 struct file **fpin)
2427{
2428 if (trylock_page(page))
2429 return 1;
2430
2431 /*
2432 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2433 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2434 * is supposed to work. We have way too many special cases..
2435 */
2436 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2437 return 0;
2438
2439 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2440 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2441 if (__lock_page_killable(page)) {
2442 /*
2443 * We didn't have the right flags to drop the mmap_sem,
2444 * but all fault_handlers only check for fatal signals
2445 * if we return VM_FAULT_RETRY, so we need to drop the
2446 * mmap_sem here and return 0 if we don't have a fpin.
2447 */
2448 if (*fpin == NULL)
2449 up_read(&vmf->vma->vm_mm->mmap_sem);
2450 return 0;
2451 }
2452 } else
2453 __lock_page(page);
2454 return 1;
2455}
2456
2457
2458/*
2459 * Synchronous readahead happens when we don't even find a page in the page
2460 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2461 * to drop the mmap sem we return the file that was pinned in order for us to do
2462 * that. If we didn't pin a file then we return NULL. The file that is
2463 * returned needs to be fput()'ed when we're done with it.
2464 */
2465static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2466{
2467 struct file *file = vmf->vma->vm_file;
2468 struct file_ra_state *ra = &file->f_ra;
2469 struct address_space *mapping = file->f_mapping;
2470 struct file *fpin = NULL;
2471 pgoff_t offset = vmf->pgoff;
2472
2473 /* If we don't want any read-ahead, don't bother */
2474 if (vmf->vma->vm_flags & VM_RAND_READ)
2475 return fpin;
2476 if (!ra->ra_pages)
2477 return fpin;
2478
2479 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2480 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2481 page_cache_sync_readahead(mapping, ra, file, offset,
2482 ra->ra_pages);
2483 return fpin;
2484 }
2485
2486 /* Avoid banging the cache line if not needed */
2487 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2488 ra->mmap_miss++;
2489
2490 /*
2491 * Do we miss much more than hit in this file? If so,
2492 * stop bothering with read-ahead. It will only hurt.
2493 */
2494 if (ra->mmap_miss > MMAP_LOTSAMISS)
2495 return fpin;
2496
2497 /*
2498 * mmap read-around
2499 */
2500 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2501 ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2502 ra->size = ra->ra_pages;
2503 ra->async_size = ra->ra_pages / 4;
2504 ra_submit(ra, mapping, file);
2505 return fpin;
2506}
2507
2508/*
2509 * Asynchronous readahead happens when we find the page and PG_readahead,
2510 * so we want to possibly extend the readahead further. We return the file that
2511 * was pinned if we have to drop the mmap_sem in order to do IO.
2512 */
2513static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2514 struct page *page)
2515{
2516 struct file *file = vmf->vma->vm_file;
2517 struct file_ra_state *ra = &file->f_ra;
2518 struct address_space *mapping = file->f_mapping;
2519 struct file *fpin = NULL;
2520 pgoff_t offset = vmf->pgoff;
2521
2522 /* If we don't want any read-ahead, don't bother */
2523 if (vmf->vma->vm_flags & VM_RAND_READ)
2524 return fpin;
2525 if (ra->mmap_miss > 0)
2526 ra->mmap_miss--;
2527 if (PageReadahead(page)) {
2528 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2529 page_cache_async_readahead(mapping, ra, file,
2530 page, offset, ra->ra_pages);
2531 }
2532 return fpin;
2533}
2534
2535/**
2536 * filemap_fault - read in file data for page fault handling
2537 * @vmf: struct vm_fault containing details of the fault
2538 *
2539 * filemap_fault() is invoked via the vma operations vector for a
2540 * mapped memory region to read in file data during a page fault.
2541 *
2542 * The goto's are kind of ugly, but this streamlines the normal case of having
2543 * it in the page cache, and handles the special cases reasonably without
2544 * having a lot of duplicated code.
2545 *
2546 * vma->vm_mm->mmap_sem must be held on entry.
2547 *
2548 * If our return value has VM_FAULT_RETRY set, it's because
2549 * lock_page_or_retry() returned 0.
2550 * The mmap_sem has usually been released in this case.
2551 * See __lock_page_or_retry() for the exception.
2552 *
2553 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2554 * has not been released.
2555 *
2556 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2557 *
2558 * Return: bitwise-OR of %VM_FAULT_ codes.
2559 */
2560vm_fault_t filemap_fault(struct vm_fault *vmf)
2561{
2562 int error;
2563 struct file *file = vmf->vma->vm_file;
2564 struct file *fpin = NULL;
2565 struct address_space *mapping = file->f_mapping;
2566 struct file_ra_state *ra = &file->f_ra;
2567 struct inode *inode = mapping->host;
2568 pgoff_t offset = vmf->pgoff;
2569 pgoff_t max_off;
2570 struct page *page;
2571 vm_fault_t ret = 0;
2572
2573 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2574 if (unlikely(offset >= max_off))
2575 return VM_FAULT_SIGBUS;
2576
2577 /*
2578 * Do we have something in the page cache already?
2579 */
2580 page = find_get_page(mapping, offset);
2581 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2582 /*
2583 * We found the page, so try async readahead before
2584 * waiting for the lock.
2585 */
2586 fpin = do_async_mmap_readahead(vmf, page);
2587 } else if (!page) {
2588 /* No page in the page cache at all */
2589 count_vm_event(PGMAJFAULT);
2590 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2591 ret = VM_FAULT_MAJOR;
2592 fpin = do_sync_mmap_readahead(vmf);
2593retry_find:
2594 page = pagecache_get_page(mapping, offset,
2595 FGP_CREAT|FGP_FOR_MMAP,
2596 vmf->gfp_mask);
2597 if (!page) {
2598 if (fpin)
2599 goto out_retry;
2600 return vmf_error(-ENOMEM);
2601 }
2602 }
2603
2604 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2605 goto out_retry;
2606
2607 /* Did it get truncated? */
2608 if (unlikely(page->mapping != mapping)) {
2609 unlock_page(page);
2610 put_page(page);
2611 goto retry_find;
2612 }
2613 VM_BUG_ON_PAGE(page->index != offset, page);
2614
2615 /*
2616 * We have a locked page in the page cache, now we need to check
2617 * that it's up-to-date. If not, it is going to be due to an error.
2618 */
2619 if (unlikely(!PageUptodate(page)))
2620 goto page_not_uptodate;
2621
2622 /*
2623 * We've made it this far and we had to drop our mmap_sem, now is the
2624 * time to return to the upper layer and have it re-find the vma and
2625 * redo the fault.
2626 */
2627 if (fpin) {
2628 unlock_page(page);
2629 goto out_retry;
2630 }
2631
2632 /*
2633 * Found the page and have a reference on it.
2634 * We must recheck i_size under page lock.
2635 */
2636 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2637 if (unlikely(offset >= max_off)) {
2638 unlock_page(page);
2639 put_page(page);
2640 return VM_FAULT_SIGBUS;
2641 }
2642
2643 vmf->page = page;
2644 return ret | VM_FAULT_LOCKED;
2645
2646page_not_uptodate:
2647 /*
2648 * Umm, take care of errors if the page isn't up-to-date.
2649 * Try to re-read it _once_. We do this synchronously,
2650 * because there really aren't any performance issues here
2651 * and we need to check for errors.
2652 */
2653 ClearPageError(page);
2654 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2655 error = mapping->a_ops->readpage(file, page);
2656 if (!error) {
2657 wait_on_page_locked(page);
2658 if (!PageUptodate(page))
2659 error = -EIO;
2660 }
2661 if (fpin)
2662 goto out_retry;
2663 put_page(page);
2664
2665 if (!error || error == AOP_TRUNCATED_PAGE)
2666 goto retry_find;
2667
2668 /* Things didn't work out. Return zero to tell the mm layer so. */
2669 shrink_readahead_size_eio(file, ra);
2670 return VM_FAULT_SIGBUS;
2671
2672out_retry:
2673 /*
2674 * We dropped the mmap_sem, we need to return to the fault handler to
2675 * re-find the vma and come back and find our hopefully still populated
2676 * page.
2677 */
2678 if (page)
2679 put_page(page);
2680 if (fpin)
2681 fput(fpin);
2682 return ret | VM_FAULT_RETRY;
2683}
2684EXPORT_SYMBOL(filemap_fault);
2685
2686void filemap_map_pages(struct vm_fault *vmf,
2687 pgoff_t start_pgoff, pgoff_t end_pgoff)
2688{
2689 struct file *file = vmf->vma->vm_file;
2690 struct address_space *mapping = file->f_mapping;
2691 pgoff_t last_pgoff = start_pgoff;
2692 unsigned long max_idx;
2693 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2694 struct page *head, *page;
2695
2696 rcu_read_lock();
2697 xas_for_each(&xas, page, end_pgoff) {
2698 if (xas_retry(&xas, page))
2699 continue;
2700 if (xa_is_value(page))
2701 goto next;
2702
2703 head = compound_head(page);
2704
2705 /*
2706 * Check for a locked page first, as a speculative
2707 * reference may adversely influence page migration.
2708 */
2709 if (PageLocked(head))
2710 goto next;
2711 if (!page_cache_get_speculative(head))
2712 goto next;
2713
2714 /* The page was split under us? */
2715 if (compound_head(page) != head)
2716 goto skip;
2717
2718 /* Has the page moved? */
2719 if (unlikely(page != xas_reload(&xas)))
2720 goto skip;
2721
2722 if (!PageUptodate(page) ||
2723 PageReadahead(page) ||
2724 PageHWPoison(page))
2725 goto skip;
2726 if (!trylock_page(page))
2727 goto skip;
2728
2729 if (page->mapping != mapping || !PageUptodate(page))
2730 goto unlock;
2731
2732 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2733 if (page->index >= max_idx)
2734 goto unlock;
2735
2736 if (file->f_ra.mmap_miss > 0)
2737 file->f_ra.mmap_miss--;
2738
2739 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2740 if (vmf->pte)
2741 vmf->pte += xas.xa_index - last_pgoff;
2742 last_pgoff = xas.xa_index;
2743 if (alloc_set_pte(vmf, NULL, page))
2744 goto unlock;
2745 unlock_page(page);
2746 goto next;
2747unlock:
2748 unlock_page(page);
2749skip:
2750 put_page(page);
2751next:
2752 /* Huge page is mapped? No need to proceed. */
2753 if (pmd_trans_huge(*vmf->pmd))
2754 break;
2755 }
2756 rcu_read_unlock();
2757}
2758EXPORT_SYMBOL(filemap_map_pages);
2759
2760vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2761{
2762 struct page *page = vmf->page;
2763 struct inode *inode = file_inode(vmf->vma->vm_file);
2764 vm_fault_t ret = VM_FAULT_LOCKED;
2765
2766 sb_start_pagefault(inode->i_sb);
2767 file_update_time(vmf->vma->vm_file);
2768 lock_page(page);
2769 if (page->mapping != inode->i_mapping) {
2770 unlock_page(page);
2771 ret = VM_FAULT_NOPAGE;
2772 goto out;
2773 }
2774 /*
2775 * We mark the page dirty already here so that when freeze is in
2776 * progress, we are guaranteed that writeback during freezing will
2777 * see the dirty page and writeprotect it again.
2778 */
2779 set_page_dirty(page);
2780 wait_for_stable_page(page);
2781out:
2782 sb_end_pagefault(inode->i_sb);
2783 return ret;
2784}
2785
2786const struct vm_operations_struct generic_file_vm_ops = {
2787 .fault = filemap_fault,
2788 .map_pages = filemap_map_pages,
2789 .page_mkwrite = filemap_page_mkwrite,
2790};
2791
2792/* This is used for a general mmap of a disk file */
2793
2794int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2795{
2796 struct address_space *mapping = file->f_mapping;
2797
2798 if (!mapping->a_ops->readpage)
2799 return -ENOEXEC;
2800 file_accessed(file);
2801 vma->vm_ops = &generic_file_vm_ops;
2802 return 0;
2803}
2804
2805/*
2806 * This is for filesystems which do not implement ->writepage.
2807 */
2808int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2809{
2810 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2811 return -EINVAL;
2812 return generic_file_mmap(file, vma);
2813}
2814#else
2815vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2816{
2817 return VM_FAULT_SIGBUS;
2818}
2819int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2820{
2821 return -ENOSYS;
2822}
2823int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2824{
2825 return -ENOSYS;
2826}
2827#endif /* CONFIG_MMU */
2828
2829EXPORT_SYMBOL(filemap_page_mkwrite);
2830EXPORT_SYMBOL(generic_file_mmap);
2831EXPORT_SYMBOL(generic_file_readonly_mmap);
2832
2833static struct page *wait_on_page_read(struct page *page)
2834{
2835 if (!IS_ERR(page)) {
2836 wait_on_page_locked(page);
2837 if (!PageUptodate(page)) {
2838 put_page(page);
2839 page = ERR_PTR(-EIO);
2840 }
2841 }
2842 return page;
2843}
2844
2845static struct page *do_read_cache_page(struct address_space *mapping,
2846 pgoff_t index,
2847 int (*filler)(void *, struct page *),
2848 void *data,
2849 gfp_t gfp)
2850{
2851 struct page *page;
2852 int err;
2853repeat:
2854 page = find_get_page(mapping, index);
2855 if (!page) {
2856 page = __page_cache_alloc(gfp);
2857 if (!page)
2858 return ERR_PTR(-ENOMEM);
2859 err = add_to_page_cache_lru(page, mapping, index, gfp);
2860 if (unlikely(err)) {
2861 put_page(page);
2862 if (err == -EEXIST)
2863 goto repeat;
2864 /* Presumably ENOMEM for xarray node */
2865 return ERR_PTR(err);
2866 }
2867
2868filler:
2869 err = filler(data, page);
2870 if (err < 0) {
2871 put_page(page);
2872 return ERR_PTR(err);
2873 }
2874
2875 page = wait_on_page_read(page);
2876 if (IS_ERR(page))
2877 return page;
2878 goto out;
2879 }
2880 if (PageUptodate(page))
2881 goto out;
2882
2883 /*
2884 * Page is not up to date and may be locked due one of the following
2885 * case a: Page is being filled and the page lock is held
2886 * case b: Read/write error clearing the page uptodate status
2887 * case c: Truncation in progress (page locked)
2888 * case d: Reclaim in progress
2889 *
2890 * Case a, the page will be up to date when the page is unlocked.
2891 * There is no need to serialise on the page lock here as the page
2892 * is pinned so the lock gives no additional protection. Even if the
2893 * the page is truncated, the data is still valid if PageUptodate as
2894 * it's a race vs truncate race.
2895 * Case b, the page will not be up to date
2896 * Case c, the page may be truncated but in itself, the data may still
2897 * be valid after IO completes as it's a read vs truncate race. The
2898 * operation must restart if the page is not uptodate on unlock but
2899 * otherwise serialising on page lock to stabilise the mapping gives
2900 * no additional guarantees to the caller as the page lock is
2901 * released before return.
2902 * Case d, similar to truncation. If reclaim holds the page lock, it
2903 * will be a race with remove_mapping that determines if the mapping
2904 * is valid on unlock but otherwise the data is valid and there is
2905 * no need to serialise with page lock.
2906 *
2907 * As the page lock gives no additional guarantee, we optimistically
2908 * wait on the page to be unlocked and check if it's up to date and
2909 * use the page if it is. Otherwise, the page lock is required to
2910 * distinguish between the different cases. The motivation is that we
2911 * avoid spurious serialisations and wakeups when multiple processes
2912 * wait on the same page for IO to complete.
2913 */
2914 wait_on_page_locked(page);
2915 if (PageUptodate(page))
2916 goto out;
2917
2918 /* Distinguish between all the cases under the safety of the lock */
2919 lock_page(page);
2920
2921 /* Case c or d, restart the operation */
2922 if (!page->mapping) {
2923 unlock_page(page);
2924 put_page(page);
2925 goto repeat;
2926 }
2927
2928 /* Someone else locked and filled the page in a very small window */
2929 if (PageUptodate(page)) {
2930 unlock_page(page);
2931 goto out;
2932 }
2933 goto filler;
2934
2935out:
2936 mark_page_accessed(page);
2937 return page;
2938}
2939
2940/**
2941 * read_cache_page - read into page cache, fill it if needed
2942 * @mapping: the page's address_space
2943 * @index: the page index
2944 * @filler: function to perform the read
2945 * @data: first arg to filler(data, page) function, often left as NULL
2946 *
2947 * Read into the page cache. If a page already exists, and PageUptodate() is
2948 * not set, try to fill the page and wait for it to become unlocked.
2949 *
2950 * If the page does not get brought uptodate, return -EIO.
2951 *
2952 * Return: up to date page on success, ERR_PTR() on failure.
2953 */
2954struct page *read_cache_page(struct address_space *mapping,
2955 pgoff_t index,
2956 int (*filler)(void *, struct page *),
2957 void *data)
2958{
2959 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2960}
2961EXPORT_SYMBOL(read_cache_page);
2962
2963/**
2964 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2965 * @mapping: the page's address_space
2966 * @index: the page index
2967 * @gfp: the page allocator flags to use if allocating
2968 *
2969 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2970 * any new page allocations done using the specified allocation flags.
2971 *
2972 * If the page does not get brought uptodate, return -EIO.
2973 *
2974 * Return: up to date page on success, ERR_PTR() on failure.
2975 */
2976struct page *read_cache_page_gfp(struct address_space *mapping,
2977 pgoff_t index,
2978 gfp_t gfp)
2979{
2980 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2981
2982 return do_read_cache_page(mapping, index, filler, NULL, gfp);
2983}
2984EXPORT_SYMBOL(read_cache_page_gfp);
2985
2986/*
2987 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2988 * LFS limits. If pos is under the limit it becomes a short access. If it
2989 * exceeds the limit we return -EFBIG.
2990 */
2991static int generic_access_check_limits(struct file *file, loff_t pos,
2992 loff_t *count)
2993{
2994 struct inode *inode = file->f_mapping->host;
2995 loff_t max_size = inode->i_sb->s_maxbytes;
2996
2997 if (!(file->f_flags & O_LARGEFILE))
2998 max_size = MAX_NON_LFS;
2999
3000 if (unlikely(pos >= max_size))
3001 return -EFBIG;
3002 *count = min(*count, max_size - pos);
3003 return 0;
3004}
3005
3006static int generic_write_check_limits(struct file *file, loff_t pos,
3007 loff_t *count)
3008{
3009 loff_t limit = rlimit(RLIMIT_FSIZE);
3010
3011 if (limit != RLIM_INFINITY) {
3012 if (pos >= limit) {
3013 send_sig(SIGXFSZ, current, 0);
3014 return -EFBIG;
3015 }
3016 *count = min(*count, limit - pos);
3017 }
3018
3019 return generic_access_check_limits(file, pos, count);
3020}
3021
3022/*
3023 * Performs necessary checks before doing a write
3024 *
3025 * Can adjust writing position or amount of bytes to write.
3026 * Returns appropriate error code that caller should return or
3027 * zero in case that write should be allowed.
3028 */
3029inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
3030{
3031 struct file *file = iocb->ki_filp;
3032 struct inode *inode = file->f_mapping->host;
3033 loff_t count;
3034 int ret;
3035
3036 if (!iov_iter_count(from))
3037 return 0;
3038
3039 /* FIXME: this is for backwards compatibility with 2.4 */
3040 if (iocb->ki_flags & IOCB_APPEND)
3041 iocb->ki_pos = i_size_read(inode);
3042
3043 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
3044 return -EINVAL;
3045
3046 count = iov_iter_count(from);
3047 ret = generic_write_check_limits(file, iocb->ki_pos, &count);
3048 if (ret)
3049 return ret;
3050
3051 iov_iter_truncate(from, count);
3052 return iov_iter_count(from);
3053}
3054EXPORT_SYMBOL(generic_write_checks);
3055
3056/*
3057 * Performs necessary checks before doing a clone.
3058 *
3059 * Can adjust amount of bytes to clone.
3060 * Returns appropriate error code that caller should return or
3061 * zero in case the clone should be allowed.
3062 */
3063int generic_remap_checks(struct file *file_in, loff_t pos_in,
3064 struct file *file_out, loff_t pos_out,
3065 loff_t *req_count, unsigned int remap_flags)
3066{
3067 struct inode *inode_in = file_in->f_mapping->host;
3068 struct inode *inode_out = file_out->f_mapping->host;
3069 uint64_t count = *req_count;
3070 uint64_t bcount;
3071 loff_t size_in, size_out;
3072 loff_t bs = inode_out->i_sb->s_blocksize;
3073 int ret;
3074
3075 /* The start of both ranges must be aligned to an fs block. */
3076 if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3077 return -EINVAL;
3078
3079 /* Ensure offsets don't wrap. */
3080 if (pos_in + count < pos_in || pos_out + count < pos_out)
3081 return -EINVAL;
3082
3083 size_in = i_size_read(inode_in);
3084 size_out = i_size_read(inode_out);
3085
3086 /* Dedupe requires both ranges to be within EOF. */
3087 if ((remap_flags & REMAP_FILE_DEDUP) &&
3088 (pos_in >= size_in || pos_in + count > size_in ||
3089 pos_out >= size_out || pos_out + count > size_out))
3090 return -EINVAL;
3091
3092 /* Ensure the infile range is within the infile. */
3093 if (pos_in >= size_in)
3094 return -EINVAL;
3095 count = min(count, size_in - (uint64_t)pos_in);
3096
3097 ret = generic_access_check_limits(file_in, pos_in, &count);
3098 if (ret)
3099 return ret;
3100
3101 ret = generic_write_check_limits(file_out, pos_out, &count);
3102 if (ret)
3103 return ret;
3104
3105 /*
3106 * If the user wanted us to link to the infile's EOF, round up to the
3107 * next block boundary for this check.
3108 *
3109 * Otherwise, make sure the count is also block-aligned, having
3110 * already confirmed the starting offsets' block alignment.
3111 */
3112 if (pos_in + count == size_in) {
3113 bcount = ALIGN(size_in, bs) - pos_in;
3114 } else {
3115 if (!IS_ALIGNED(count, bs))
3116 count = ALIGN_DOWN(count, bs);
3117 bcount = count;
3118 }
3119
3120 /* Don't allow overlapped cloning within the same file. */
3121 if (inode_in == inode_out &&
3122 pos_out + bcount > pos_in &&
3123 pos_out < pos_in + bcount)
3124 return -EINVAL;
3125
3126 /*
3127 * We shortened the request but the caller can't deal with that, so
3128 * bounce the request back to userspace.
3129 */
3130 if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3131 return -EINVAL;
3132
3133 *req_count = count;
3134 return 0;
3135}
3136
3137int pagecache_write_begin(struct file *file, struct address_space *mapping,
3138 loff_t pos, unsigned len, unsigned flags,
3139 struct page **pagep, void **fsdata)
3140{
3141 const struct address_space_operations *aops = mapping->a_ops;
3142
3143 return aops->write_begin(file, mapping, pos, len, flags,
3144 pagep, fsdata);
3145}
3146EXPORT_SYMBOL(pagecache_write_begin);
3147
3148int pagecache_write_end(struct file *file, struct address_space *mapping,
3149 loff_t pos, unsigned len, unsigned copied,
3150 struct page *page, void *fsdata)
3151{
3152 const struct address_space_operations *aops = mapping->a_ops;
3153
3154 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3155}
3156EXPORT_SYMBOL(pagecache_write_end);
3157
3158ssize_t
3159generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3160{
3161 struct file *file = iocb->ki_filp;
3162 struct address_space *mapping = file->f_mapping;
3163 struct inode *inode = mapping->host;
3164 loff_t pos = iocb->ki_pos;
3165 ssize_t written;
3166 size_t write_len;
3167 pgoff_t end;
3168
3169 write_len = iov_iter_count(from);
3170 end = (pos + write_len - 1) >> PAGE_SHIFT;
3171
3172 if (iocb->ki_flags & IOCB_NOWAIT) {
3173 /* If there are pages to writeback, return */
3174 if (filemap_range_has_page(inode->i_mapping, pos,
3175 pos + write_len - 1))
3176 return -EAGAIN;
3177 } else {
3178 written = filemap_write_and_wait_range(mapping, pos,
3179 pos + write_len - 1);
3180 if (written)
3181 goto out;
3182 }
3183
3184 /*
3185 * After a write we want buffered reads to be sure to go to disk to get
3186 * the new data. We invalidate clean cached page from the region we're
3187 * about to write. We do this *before* the write so that we can return
3188 * without clobbering -EIOCBQUEUED from ->direct_IO().
3189 */
3190 written = invalidate_inode_pages2_range(mapping,
3191 pos >> PAGE_SHIFT, end);
3192 /*
3193 * If a page can not be invalidated, return 0 to fall back
3194 * to buffered write.
3195 */
3196 if (written) {
3197 if (written == -EBUSY)
3198 return 0;
3199 goto out;
3200 }
3201
3202 written = mapping->a_ops->direct_IO(iocb, from);
3203
3204 /*
3205 * Finally, try again to invalidate clean pages which might have been
3206 * cached by non-direct readahead, or faulted in by get_user_pages()
3207 * if the source of the write was an mmap'ed region of the file
3208 * we're writing. Either one is a pretty crazy thing to do,
3209 * so we don't support it 100%. If this invalidation
3210 * fails, tough, the write still worked...
3211 *
3212 * Most of the time we do not need this since dio_complete() will do
3213 * the invalidation for us. However there are some file systems that
3214 * do not end up with dio_complete() being called, so let's not break
3215 * them by removing it completely
3216 */
3217 if (mapping->nrpages)
3218 invalidate_inode_pages2_range(mapping,
3219 pos >> PAGE_SHIFT, end);
3220
3221 if (written > 0) {
3222 pos += written;
3223 write_len -= written;
3224 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3225 i_size_write(inode, pos);
3226 mark_inode_dirty(inode);
3227 }
3228 iocb->ki_pos = pos;
3229 }
3230 iov_iter_revert(from, write_len - iov_iter_count(from));
3231out:
3232 return written;
3233}
3234EXPORT_SYMBOL(generic_file_direct_write);
3235
3236/*
3237 * Find or create a page at the given pagecache position. Return the locked
3238 * page. This function is specifically for buffered writes.
3239 */
3240struct page *grab_cache_page_write_begin(struct address_space *mapping,
3241 pgoff_t index, unsigned flags)
3242{
3243 struct page *page;
3244 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3245
3246 if (flags & AOP_FLAG_NOFS)
3247 fgp_flags |= FGP_NOFS;
3248
3249 page = pagecache_get_page(mapping, index, fgp_flags,
3250 mapping_gfp_mask(mapping));
3251 if (page)
3252 wait_for_stable_page(page);
3253
3254 return page;
3255}
3256EXPORT_SYMBOL(grab_cache_page_write_begin);
3257
3258ssize_t generic_perform_write(struct file *file,
3259 struct iov_iter *i, loff_t pos)
3260{
3261 struct address_space *mapping = file->f_mapping;
3262 const struct address_space_operations *a_ops = mapping->a_ops;
3263 long status = 0;
3264 ssize_t written = 0;
3265 unsigned int flags = 0;
3266
3267 do {
3268 struct page *page;
3269 unsigned long offset; /* Offset into pagecache page */
3270 unsigned long bytes; /* Bytes to write to page */
3271 size_t copied; /* Bytes copied from user */
3272 void *fsdata;
3273
3274 offset = (pos & (PAGE_SIZE - 1));
3275 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3276 iov_iter_count(i));
3277
3278again:
3279 /*
3280 * Bring in the user page that we will copy from _first_.
3281 * Otherwise there's a nasty deadlock on copying from the
3282 * same page as we're writing to, without it being marked
3283 * up-to-date.
3284 *
3285 * Not only is this an optimisation, but it is also required
3286 * to check that the address is actually valid, when atomic
3287 * usercopies are used, below.
3288 */
3289 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3290 status = -EFAULT;
3291 break;
3292 }
3293
3294 if (fatal_signal_pending(current)) {
3295 status = -EINTR;
3296 break;
3297 }
3298
3299 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3300 &page, &fsdata);
3301 if (unlikely(status < 0))
3302 break;
3303
3304 if (mapping_writably_mapped(mapping))
3305 flush_dcache_page(page);
3306
3307 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3308 flush_dcache_page(page);
3309
3310 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3311 page, fsdata);
3312 if (unlikely(status < 0))
3313 break;
3314 copied = status;
3315
3316 cond_resched();
3317
3318 iov_iter_advance(i, copied);
3319 if (unlikely(copied == 0)) {
3320 /*
3321 * If we were unable to copy any data at all, we must
3322 * fall back to a single segment length write.
3323 *
3324 * If we didn't fallback here, we could livelock
3325 * because not all segments in the iov can be copied at
3326 * once without a pagefault.
3327 */
3328 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3329 iov_iter_single_seg_count(i));
3330 goto again;
3331 }
3332 pos += copied;
3333 written += copied;
3334
3335 balance_dirty_pages_ratelimited(mapping);
3336 } while (iov_iter_count(i));
3337
3338 return written ? written : status;
3339}
3340EXPORT_SYMBOL(generic_perform_write);
3341
3342/**
3343 * __generic_file_write_iter - write data to a file
3344 * @iocb: IO state structure (file, offset, etc.)
3345 * @from: iov_iter with data to write
3346 *
3347 * This function does all the work needed for actually writing data to a
3348 * file. It does all basic checks, removes SUID from the file, updates
3349 * modification times and calls proper subroutines depending on whether we
3350 * do direct IO or a standard buffered write.
3351 *
3352 * It expects i_mutex to be grabbed unless we work on a block device or similar
3353 * object which does not need locking at all.
3354 *
3355 * This function does *not* take care of syncing data in case of O_SYNC write.
3356 * A caller has to handle it. This is mainly due to the fact that we want to
3357 * avoid syncing under i_mutex.
3358 *
3359 * Return:
3360 * * number of bytes written, even for truncated writes
3361 * * negative error code if no data has been written at all
3362 */
3363ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3364{
3365 struct file *file = iocb->ki_filp;
3366 struct address_space * mapping = file->f_mapping;
3367 struct inode *inode = mapping->host;
3368 ssize_t written = 0;
3369 ssize_t err;
3370 ssize_t status;
3371
3372 /* We can write back this queue in page reclaim */
3373 current->backing_dev_info = inode_to_bdi(inode);
3374 err = file_remove_privs(file);
3375 if (err)
3376 goto out;
3377
3378 err = file_update_time(file);
3379 if (err)
3380 goto out;
3381
3382 if (iocb->ki_flags & IOCB_DIRECT) {
3383 loff_t pos, endbyte;
3384
3385 written = generic_file_direct_write(iocb, from);
3386 /*
3387 * If the write stopped short of completing, fall back to
3388 * buffered writes. Some filesystems do this for writes to
3389 * holes, for example. For DAX files, a buffered write will
3390 * not succeed (even if it did, DAX does not handle dirty
3391 * page-cache pages correctly).
3392 */
3393 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3394 goto out;
3395
3396 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3397 /*
3398 * If generic_perform_write() returned a synchronous error
3399 * then we want to return the number of bytes which were
3400 * direct-written, or the error code if that was zero. Note
3401 * that this differs from normal direct-io semantics, which
3402 * will return -EFOO even if some bytes were written.
3403 */
3404 if (unlikely(status < 0)) {
3405 err = status;
3406 goto out;
3407 }
3408 /*
3409 * We need to ensure that the page cache pages are written to
3410 * disk and invalidated to preserve the expected O_DIRECT
3411 * semantics.
3412 */
3413 endbyte = pos + status - 1;
3414 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3415 if (err == 0) {
3416 iocb->ki_pos = endbyte + 1;
3417 written += status;
3418 invalidate_mapping_pages(mapping,
3419 pos >> PAGE_SHIFT,
3420 endbyte >> PAGE_SHIFT);
3421 } else {
3422 /*
3423 * We don't know how much we wrote, so just return
3424 * the number of bytes which were direct-written
3425 */
3426 }
3427 } else {
3428 written = generic_perform_write(file, from, iocb->ki_pos);
3429 if (likely(written > 0))
3430 iocb->ki_pos += written;
3431 }
3432out:
3433 current->backing_dev_info = NULL;
3434 return written ? written : err;
3435}
3436EXPORT_SYMBOL(__generic_file_write_iter);
3437
3438/**
3439 * generic_file_write_iter - write data to a file
3440 * @iocb: IO state structure
3441 * @from: iov_iter with data to write
3442 *
3443 * This is a wrapper around __generic_file_write_iter() to be used by most
3444 * filesystems. It takes care of syncing the file in case of O_SYNC file
3445 * and acquires i_mutex as needed.
3446 * Return:
3447 * * negative error code if no data has been written at all of
3448 * vfs_fsync_range() failed for a synchronous write
3449 * * number of bytes written, even for truncated writes
3450 */
3451ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3452{
3453 struct file *file = iocb->ki_filp;
3454 struct inode *inode = file->f_mapping->host;
3455 ssize_t ret;
3456
3457 inode_lock(inode);
3458 ret = generic_write_checks(iocb, from);
3459 if (ret > 0)
3460 ret = __generic_file_write_iter(iocb, from);
3461 inode_unlock(inode);
3462
3463 if (ret > 0)
3464 ret = generic_write_sync(iocb, ret);
3465 return ret;
3466}
3467EXPORT_SYMBOL(generic_file_write_iter);
3468
3469/**
3470 * try_to_release_page() - release old fs-specific metadata on a page
3471 *
3472 * @page: the page which the kernel is trying to free
3473 * @gfp_mask: memory allocation flags (and I/O mode)
3474 *
3475 * The address_space is to try to release any data against the page
3476 * (presumably at page->private).
3477 *
3478 * This may also be called if PG_fscache is set on a page, indicating that the
3479 * page is known to the local caching routines.
3480 *
3481 * The @gfp_mask argument specifies whether I/O may be performed to release
3482 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3483 *
3484 * Return: %1 if the release was successful, otherwise return zero.
3485 */
3486int try_to_release_page(struct page *page, gfp_t gfp_mask)
3487{
3488 struct address_space * const mapping = page->mapping;
3489
3490 BUG_ON(!PageLocked(page));
3491 if (PageWriteback(page))
3492 return 0;
3493
3494 if (mapping && mapping->a_ops->releasepage)
3495 return mapping->a_ops->releasepage(page, gfp_mask);
3496 return try_to_free_buffers(page);
3497}
3498
3499EXPORT_SYMBOL(try_to_release_page);
3500