1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * linux/mm/filemap.c
4 *
5 * Copyright (C) 1994-1999 Linus Torvalds
6 */
7
8 /*
9 * This file handles the generic file mmap semantics used by
10 * most "normal" filesystems (but you don't /have/ to use this:
11 * the NFS filesystem used to do this differently, for example)
12 */
13 #include <linux/export.h>
14 #include <linux/compiler.h>
15 #include <linux/dax.h>
16 #include <linux/fs.h>
17 #include <linux/sched/signal.h>
18 #include <linux/uaccess.h>
19 #include <linux/capability.h>
20 #include <linux/kernel_stat.h>
21 #include <linux/gfp.h>
22 #include <linux/mm.h>
23 #include <linux/swap.h>
24 #include <linux/mman.h>
25 #include <linux/pagemap.h>
26 #include <linux/file.h>
27 #include <linux/uio.h>
28 #include <linux/error-injection.h>
29 #include <linux/hash.h>
30 #include <linux/writeback.h>
31 #include <linux/backing-dev.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/security.h>
35 #include <linux/cpuset.h>
36 #include <linux/hugetlb.h>
37 #include <linux/memcontrol.h>
38 #include <linux/cleancache.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/rmap.h>
41 #include <linux/delayacct.h>
42 #include <linux/psi.h>
43 #include <linux/ramfs.h>
44 #include <linux/page_idle.h>
45 #include "internal.h"
46
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/filemap.h>
49
50 /*
51 * FIXME: remove all knowledge of the buffer layer from the core VM
52 */
53 #include <linux/buffer_head.h> /* for try_to_free_buffers */
54
55 #include <asm/mman.h>
56
57 /*
58 * Shared mappings implemented 30.11.1994. It's not fully working yet,
59 * though.
60 *
61 * Shared mappings now work. 15.8.1995 Bruno.
62 *
63 * finished 'unifying' the page and buffer cache and SMP-threaded the
64 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
65 *
66 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
67 */
68
69 /*
70 * Lock ordering:
71 *
72 * ->i_mmap_rwsem (truncate_pagecache)
73 * ->private_lock (__free_pte->__set_page_dirty_buffers)
74 * ->swap_lock (exclusive_swap_page, others)
75 * ->i_pages lock
76 *
77 * ->i_mutex
78 * ->i_mmap_rwsem (truncate->unmap_mapping_range)
79 *
80 * ->mmap_lock
81 * ->i_mmap_rwsem
82 * ->page_table_lock or pte_lock (various, mainly in memory.c)
83 * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
84 *
85 * ->mmap_lock
86 * ->lock_page (access_process_vm)
87 *
88 * ->i_mutex (generic_perform_write)
89 * ->mmap_lock (fault_in_pages_readable->do_page_fault)
90 *
91 * bdi->wb.list_lock
92 * sb_lock (fs/fs-writeback.c)
93 * ->i_pages lock (__sync_single_inode)
94 *
95 * ->i_mmap_rwsem
96 * ->anon_vma.lock (vma_adjust)
97 *
98 * ->anon_vma.lock
99 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
100 *
101 * ->page_table_lock or pte_lock
102 * ->swap_lock (try_to_unmap_one)
103 * ->private_lock (try_to_unmap_one)
104 * ->i_pages lock (try_to_unmap_one)
105 * ->pgdat->lru_lock (follow_page->mark_page_accessed)
106 * ->pgdat->lru_lock (check_pte_range->isolate_lru_page)
107 * ->private_lock (page_remove_rmap->set_page_dirty)
108 * ->i_pages lock (page_remove_rmap->set_page_dirty)
109 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
110 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
111 * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
112 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
113 * ->inode->i_lock (zap_pte_range->set_page_dirty)
114 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
115 *
116 * ->i_mmap_rwsem
117 * ->tasklist_lock (memory_failure, collect_procs_ao)
118 */
119
page_cache_delete(struct address_space * mapping,struct page * page,void * shadow)120 static void page_cache_delete(struct address_space *mapping,
121 struct page *page, void *shadow)
122 {
123 XA_STATE(xas, &mapping->i_pages, page->index);
124 unsigned int nr = 1;
125
126 mapping_set_update(&xas, mapping);
127
128 /* hugetlb pages are represented by a single entry in the xarray */
129 if (!PageHuge(page)) {
130 xas_set_order(&xas, page->index, compound_order(page));
131 nr = compound_nr(page);
132 }
133
134 VM_BUG_ON_PAGE(!PageLocked(page), page);
135 VM_BUG_ON_PAGE(PageTail(page), page);
136 VM_BUG_ON_PAGE(nr != 1 && shadow, page);
137
138 xas_store(&xas, shadow);
139 xas_init_marks(&xas);
140
141 page->mapping = NULL;
142 /* Leave page->index set: truncation lookup relies upon it */
143
144 if (shadow) {
145 mapping->nrexceptional += nr;
146 /*
147 * Make sure the nrexceptional update is committed before
148 * the nrpages update so that final truncate racing
149 * with reclaim does not see both counters 0 at the
150 * same time and miss a shadow entry.
151 */
152 smp_wmb();
153 }
154 mapping->nrpages -= nr;
155 }
156
unaccount_page_cache_page(struct address_space * mapping,struct page * page)157 static void unaccount_page_cache_page(struct address_space *mapping,
158 struct page *page)
159 {
160 int nr;
161
162 /*
163 * if we're uptodate, flush out into the cleancache, otherwise
164 * invalidate any existing cleancache entries. We can't leave
165 * stale data around in the cleancache once our page is gone
166 */
167 if (PageUptodate(page) && PageMappedToDisk(page))
168 cleancache_put_page(page);
169 else
170 cleancache_invalidate_page(mapping, page);
171
172 VM_BUG_ON_PAGE(PageTail(page), page);
173 VM_BUG_ON_PAGE(page_mapped(page), page);
174 if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
175 int mapcount;
176
177 pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
178 current->comm, page_to_pfn(page));
179 dump_page(page, "still mapped when deleted");
180 dump_stack();
181 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
182
183 mapcount = page_mapcount(page);
184 if (mapping_exiting(mapping) &&
185 page_count(page) >= mapcount + 2) {
186 /*
187 * All vmas have already been torn down, so it's
188 * a good bet that actually the page is unmapped,
189 * and we'd prefer not to leak it: if we're wrong,
190 * some other bad page check should catch it later.
191 */
192 page_mapcount_reset(page);
193 page_ref_sub(page, mapcount);
194 }
195 }
196
197 /* hugetlb pages do not participate in page cache accounting. */
198 if (PageHuge(page))
199 return;
200
201 nr = thp_nr_pages(page);
202
203 __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
204 if (PageSwapBacked(page)) {
205 __mod_lruvec_page_state(page, NR_SHMEM, -nr);
206 if (PageTransHuge(page))
207 __dec_node_page_state(page, NR_SHMEM_THPS);
208 } else if (PageTransHuge(page)) {
209 __dec_node_page_state(page, NR_FILE_THPS);
210 filemap_nr_thps_dec(mapping);
211 }
212
213 /*
214 * At this point page must be either written or cleaned by
215 * truncate. Dirty page here signals a bug and loss of
216 * unwritten data.
217 *
218 * This fixes dirty accounting after removing the page entirely
219 * but leaves PageDirty set: it has no effect for truncated
220 * page and anyway will be cleared before returning page into
221 * buddy allocator.
222 */
223 if (WARN_ON_ONCE(PageDirty(page)))
224 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
225 }
226
227 /*
228 * Delete a page from the page cache and free it. Caller has to make
229 * sure the page is locked and that nobody else uses it - or that usage
230 * is safe. The caller must hold the i_pages lock.
231 */
__delete_from_page_cache(struct page * page,void * shadow)232 void __delete_from_page_cache(struct page *page, void *shadow)
233 {
234 struct address_space *mapping = page->mapping;
235
236 trace_mm_filemap_delete_from_page_cache(page);
237
238 unaccount_page_cache_page(mapping, page);
239 page_cache_delete(mapping, page, shadow);
240 }
241
page_cache_free_page(struct address_space * mapping,struct page * page)242 static void page_cache_free_page(struct address_space *mapping,
243 struct page *page)
244 {
245 void (*freepage)(struct page *);
246
247 freepage = mapping->a_ops->freepage;
248 if (freepage)
249 freepage(page);
250
251 if (PageTransHuge(page) && !PageHuge(page)) {
252 page_ref_sub(page, thp_nr_pages(page));
253 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
254 } else {
255 put_page(page);
256 }
257 }
258
259 /**
260 * delete_from_page_cache - delete page from page cache
261 * @page: the page which the kernel is trying to remove from page cache
262 *
263 * This must be called only on pages that have been verified to be in the page
264 * cache and locked. It will never put the page into the free list, the caller
265 * has a reference on the page.
266 */
delete_from_page_cache(struct page * page)267 void delete_from_page_cache(struct page *page)
268 {
269 struct address_space *mapping = page_mapping(page);
270 unsigned long flags;
271
272 BUG_ON(!PageLocked(page));
273 xa_lock_irqsave(&mapping->i_pages, flags);
274 __delete_from_page_cache(page, NULL);
275 xa_unlock_irqrestore(&mapping->i_pages, flags);
276
277 page_cache_free_page(mapping, page);
278 }
279 EXPORT_SYMBOL(delete_from_page_cache);
280
281 /*
282 * page_cache_delete_batch - delete several pages from page cache
283 * @mapping: the mapping to which pages belong
284 * @pvec: pagevec with pages to delete
285 *
286 * The function walks over mapping->i_pages and removes pages passed in @pvec
287 * from the mapping. The function expects @pvec to be sorted by page index
288 * and is optimised for it to be dense.
289 * It tolerates holes in @pvec (mapping entries at those indices are not
290 * modified). The function expects only THP head pages to be present in the
291 * @pvec.
292 *
293 * The function expects the i_pages lock to be held.
294 */
page_cache_delete_batch(struct address_space * mapping,struct pagevec * pvec)295 static void page_cache_delete_batch(struct address_space *mapping,
296 struct pagevec *pvec)
297 {
298 XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
299 int total_pages = 0;
300 int i = 0;
301 struct page *page;
302
303 mapping_set_update(&xas, mapping);
304 xas_for_each(&xas, page, ULONG_MAX) {
305 if (i >= pagevec_count(pvec))
306 break;
307
308 /* A swap/dax/shadow entry got inserted? Skip it. */
309 if (xa_is_value(page))
310 continue;
311 /*
312 * A page got inserted in our range? Skip it. We have our
313 * pages locked so they are protected from being removed.
314 * If we see a page whose index is higher than ours, it
315 * means our page has been removed, which shouldn't be
316 * possible because we're holding the PageLock.
317 */
318 if (page != pvec->pages[i]) {
319 VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
320 page);
321 continue;
322 }
323
324 WARN_ON_ONCE(!PageLocked(page));
325
326 if (page->index == xas.xa_index)
327 page->mapping = NULL;
328 /* Leave page->index set: truncation lookup relies on it */
329
330 /*
331 * Move to the next page in the vector if this is a regular
332 * page or the index is of the last sub-page of this compound
333 * page.
334 */
335 if (page->index + compound_nr(page) - 1 == xas.xa_index)
336 i++;
337 xas_store(&xas, NULL);
338 total_pages++;
339 }
340 mapping->nrpages -= total_pages;
341 }
342
delete_from_page_cache_batch(struct address_space * mapping,struct pagevec * pvec)343 void delete_from_page_cache_batch(struct address_space *mapping,
344 struct pagevec *pvec)
345 {
346 int i;
347 unsigned long flags;
348
349 if (!pagevec_count(pvec))
350 return;
351
352 xa_lock_irqsave(&mapping->i_pages, flags);
353 for (i = 0; i < pagevec_count(pvec); i++) {
354 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
355
356 unaccount_page_cache_page(mapping, pvec->pages[i]);
357 }
358 page_cache_delete_batch(mapping, pvec);
359 xa_unlock_irqrestore(&mapping->i_pages, flags);
360
361 for (i = 0; i < pagevec_count(pvec); i++)
362 page_cache_free_page(mapping, pvec->pages[i]);
363 }
364
filemap_check_errors(struct address_space * mapping)365 int filemap_check_errors(struct address_space *mapping)
366 {
367 int ret = 0;
368 /* Check for outstanding write errors */
369 if (test_bit(AS_ENOSPC, &mapping->flags) &&
370 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
371 ret = -ENOSPC;
372 if (test_bit(AS_EIO, &mapping->flags) &&
373 test_and_clear_bit(AS_EIO, &mapping->flags))
374 ret = -EIO;
375 return ret;
376 }
377 EXPORT_SYMBOL(filemap_check_errors);
378
filemap_check_and_keep_errors(struct address_space * mapping)379 static int filemap_check_and_keep_errors(struct address_space *mapping)
380 {
381 /* Check for outstanding write errors */
382 if (test_bit(AS_EIO, &mapping->flags))
383 return -EIO;
384 if (test_bit(AS_ENOSPC, &mapping->flags))
385 return -ENOSPC;
386 return 0;
387 }
388
389 /**
390 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
391 * @mapping: address space structure to write
392 * @start: offset in bytes where the range starts
393 * @end: offset in bytes where the range ends (inclusive)
394 * @sync_mode: enable synchronous operation
395 *
396 * Start writeback against all of a mapping's dirty pages that lie
397 * within the byte offsets <start, end> inclusive.
398 *
399 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
400 * opposed to a regular memory cleansing writeback. The difference between
401 * these two operations is that if a dirty page/buffer is encountered, it must
402 * be waited upon, and not just skipped over.
403 *
404 * Return: %0 on success, negative error code otherwise.
405 */
__filemap_fdatawrite_range(struct address_space * mapping,loff_t start,loff_t end,int sync_mode)406 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
407 loff_t end, int sync_mode)
408 {
409 int ret;
410 struct writeback_control wbc = {
411 .sync_mode = sync_mode,
412 .nr_to_write = LONG_MAX,
413 .range_start = start,
414 .range_end = end,
415 };
416
417 if (!mapping_can_writeback(mapping) ||
418 !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
419 return 0;
420
421 wbc_attach_fdatawrite_inode(&wbc, mapping->host);
422 ret = do_writepages(mapping, &wbc);
423 wbc_detach_inode(&wbc);
424 return ret;
425 }
426
__filemap_fdatawrite(struct address_space * mapping,int sync_mode)427 static inline int __filemap_fdatawrite(struct address_space *mapping,
428 int sync_mode)
429 {
430 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
431 }
432
filemap_fdatawrite(struct address_space * mapping)433 int filemap_fdatawrite(struct address_space *mapping)
434 {
435 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
436 }
437 EXPORT_SYMBOL(filemap_fdatawrite);
438
filemap_fdatawrite_range(struct address_space * mapping,loff_t start,loff_t end)439 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
440 loff_t end)
441 {
442 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
443 }
444 EXPORT_SYMBOL(filemap_fdatawrite_range);
445
446 /**
447 * filemap_flush - mostly a non-blocking flush
448 * @mapping: target address_space
449 *
450 * This is a mostly non-blocking flush. Not suitable for data-integrity
451 * purposes - I/O may not be started against all dirty pages.
452 *
453 * Return: %0 on success, negative error code otherwise.
454 */
filemap_flush(struct address_space * mapping)455 int filemap_flush(struct address_space *mapping)
456 {
457 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
458 }
459 EXPORT_SYMBOL(filemap_flush);
460
461 /**
462 * filemap_range_has_page - check if a page exists in range.
463 * @mapping: address space within which to check
464 * @start_byte: offset in bytes where the range starts
465 * @end_byte: offset in bytes where the range ends (inclusive)
466 *
467 * Find at least one page in the range supplied, usually used to check if
468 * direct writing in this range will trigger a writeback.
469 *
470 * Return: %true if at least one page exists in the specified range,
471 * %false otherwise.
472 */
filemap_range_has_page(struct address_space * mapping,loff_t start_byte,loff_t end_byte)473 bool filemap_range_has_page(struct address_space *mapping,
474 loff_t start_byte, loff_t end_byte)
475 {
476 struct page *page;
477 XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
478 pgoff_t max = end_byte >> PAGE_SHIFT;
479
480 if (end_byte < start_byte)
481 return false;
482
483 rcu_read_lock();
484 for (;;) {
485 page = xas_find(&xas, max);
486 if (xas_retry(&xas, page))
487 continue;
488 /* Shadow entries don't count */
489 if (xa_is_value(page))
490 continue;
491 /*
492 * We don't need to try to pin this page; we're about to
493 * release the RCU lock anyway. It is enough to know that
494 * there was a page here recently.
495 */
496 break;
497 }
498 rcu_read_unlock();
499
500 return page != NULL;
501 }
502 EXPORT_SYMBOL(filemap_range_has_page);
503
__filemap_fdatawait_range(struct address_space * mapping,loff_t start_byte,loff_t end_byte)504 static void __filemap_fdatawait_range(struct address_space *mapping,
505 loff_t start_byte, loff_t end_byte)
506 {
507 pgoff_t index = start_byte >> PAGE_SHIFT;
508 pgoff_t end = end_byte >> PAGE_SHIFT;
509 struct pagevec pvec;
510 int nr_pages;
511
512 if (end_byte < start_byte)
513 return;
514
515 pagevec_init(&pvec);
516 while (index <= end) {
517 unsigned i;
518
519 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
520 end, PAGECACHE_TAG_WRITEBACK);
521 if (!nr_pages)
522 break;
523
524 for (i = 0; i < nr_pages; i++) {
525 struct page *page = pvec.pages[i];
526
527 wait_on_page_writeback(page);
528 ClearPageError(page);
529 }
530 pagevec_release(&pvec);
531 cond_resched();
532 }
533 }
534
535 /**
536 * filemap_fdatawait_range - wait for writeback to complete
537 * @mapping: address space structure to wait for
538 * @start_byte: offset in bytes where the range starts
539 * @end_byte: offset in bytes where the range ends (inclusive)
540 *
541 * Walk the list of under-writeback pages of the given address space
542 * in the given range and wait for all of them. Check error status of
543 * the address space and return it.
544 *
545 * Since the error status of the address space is cleared by this function,
546 * callers are responsible for checking the return value and handling and/or
547 * reporting the error.
548 *
549 * Return: error status of the address space.
550 */
filemap_fdatawait_range(struct address_space * mapping,loff_t start_byte,loff_t end_byte)551 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
552 loff_t end_byte)
553 {
554 __filemap_fdatawait_range(mapping, start_byte, end_byte);
555 return filemap_check_errors(mapping);
556 }
557 EXPORT_SYMBOL(filemap_fdatawait_range);
558
559 /**
560 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
561 * @mapping: address space structure to wait for
562 * @start_byte: offset in bytes where the range starts
563 * @end_byte: offset in bytes where the range ends (inclusive)
564 *
565 * Walk the list of under-writeback pages of the given address space in the
566 * given range and wait for all of them. Unlike filemap_fdatawait_range(),
567 * this function does not clear error status of the address space.
568 *
569 * Use this function if callers don't handle errors themselves. Expected
570 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
571 * fsfreeze(8)
572 */
filemap_fdatawait_range_keep_errors(struct address_space * mapping,loff_t start_byte,loff_t end_byte)573 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
574 loff_t start_byte, loff_t end_byte)
575 {
576 __filemap_fdatawait_range(mapping, start_byte, end_byte);
577 return filemap_check_and_keep_errors(mapping);
578 }
579 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
580
581 /**
582 * file_fdatawait_range - wait for writeback to complete
583 * @file: file pointing to address space structure to wait for
584 * @start_byte: offset in bytes where the range starts
585 * @end_byte: offset in bytes where the range ends (inclusive)
586 *
587 * Walk the list of under-writeback pages of the address space that file
588 * refers to, in the given range and wait for all of them. Check error
589 * status of the address space vs. the file->f_wb_err cursor and return it.
590 *
591 * Since the error status of the file is advanced by this function,
592 * callers are responsible for checking the return value and handling and/or
593 * reporting the error.
594 *
595 * Return: error status of the address space vs. the file->f_wb_err cursor.
596 */
file_fdatawait_range(struct file * file,loff_t start_byte,loff_t end_byte)597 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
598 {
599 struct address_space *mapping = file->f_mapping;
600
601 __filemap_fdatawait_range(mapping, start_byte, end_byte);
602 return file_check_and_advance_wb_err(file);
603 }
604 EXPORT_SYMBOL(file_fdatawait_range);
605
606 /**
607 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
608 * @mapping: address space structure to wait for
609 *
610 * Walk the list of under-writeback pages of the given address space
611 * and wait for all of them. Unlike filemap_fdatawait(), this function
612 * does not clear error status of the address space.
613 *
614 * Use this function if callers don't handle errors themselves. Expected
615 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
616 * fsfreeze(8)
617 *
618 * Return: error status of the address space.
619 */
filemap_fdatawait_keep_errors(struct address_space * mapping)620 int filemap_fdatawait_keep_errors(struct address_space *mapping)
621 {
622 __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
623 return filemap_check_and_keep_errors(mapping);
624 }
625 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
626
627 /* Returns true if writeback might be needed or already in progress. */
mapping_needs_writeback(struct address_space * mapping)628 static bool mapping_needs_writeback(struct address_space *mapping)
629 {
630 if (dax_mapping(mapping))
631 return mapping->nrexceptional;
632
633 return mapping->nrpages;
634 }
635
636 /**
637 * filemap_write_and_wait_range - write out & wait on a file range
638 * @mapping: the address_space for the pages
639 * @lstart: offset in bytes where the range starts
640 * @lend: offset in bytes where the range ends (inclusive)
641 *
642 * Write out and wait upon file offsets lstart->lend, inclusive.
643 *
644 * Note that @lend is inclusive (describes the last byte to be written) so
645 * that this function can be used to write to the very end-of-file (end = -1).
646 *
647 * Return: error status of the address space.
648 */
filemap_write_and_wait_range(struct address_space * mapping,loff_t lstart,loff_t lend)649 int filemap_write_and_wait_range(struct address_space *mapping,
650 loff_t lstart, loff_t lend)
651 {
652 int err = 0;
653
654 if (mapping_needs_writeback(mapping)) {
655 err = __filemap_fdatawrite_range(mapping, lstart, lend,
656 WB_SYNC_ALL);
657 /*
658 * Even if the above returned error, the pages may be
659 * written partially (e.g. -ENOSPC), so we wait for it.
660 * But the -EIO is special case, it may indicate the worst
661 * thing (e.g. bug) happened, so we avoid waiting for it.
662 */
663 if (err != -EIO) {
664 int err2 = filemap_fdatawait_range(mapping,
665 lstart, lend);
666 if (!err)
667 err = err2;
668 } else {
669 /* Clear any previously stored errors */
670 filemap_check_errors(mapping);
671 }
672 } else {
673 err = filemap_check_errors(mapping);
674 }
675 return err;
676 }
677 EXPORT_SYMBOL(filemap_write_and_wait_range);
678
__filemap_set_wb_err(struct address_space * mapping,int err)679 void __filemap_set_wb_err(struct address_space *mapping, int err)
680 {
681 errseq_t eseq = errseq_set(&mapping->wb_err, err);
682
683 trace_filemap_set_wb_err(mapping, eseq);
684 }
685 EXPORT_SYMBOL(__filemap_set_wb_err);
686
687 /**
688 * file_check_and_advance_wb_err - report wb error (if any) that was previously
689 * and advance wb_err to current one
690 * @file: struct file on which the error is being reported
691 *
692 * When userland calls fsync (or something like nfsd does the equivalent), we
693 * want to report any writeback errors that occurred since the last fsync (or
694 * since the file was opened if there haven't been any).
695 *
696 * Grab the wb_err from the mapping. If it matches what we have in the file,
697 * then just quickly return 0. The file is all caught up.
698 *
699 * If it doesn't match, then take the mapping value, set the "seen" flag in
700 * it and try to swap it into place. If it works, or another task beat us
701 * to it with the new value, then update the f_wb_err and return the error
702 * portion. The error at this point must be reported via proper channels
703 * (a'la fsync, or NFS COMMIT operation, etc.).
704 *
705 * While we handle mapping->wb_err with atomic operations, the f_wb_err
706 * value is protected by the f_lock since we must ensure that it reflects
707 * the latest value swapped in for this file descriptor.
708 *
709 * Return: %0 on success, negative error code otherwise.
710 */
file_check_and_advance_wb_err(struct file * file)711 int file_check_and_advance_wb_err(struct file *file)
712 {
713 int err = 0;
714 errseq_t old = READ_ONCE(file->f_wb_err);
715 struct address_space *mapping = file->f_mapping;
716
717 /* Locklessly handle the common case where nothing has changed */
718 if (errseq_check(&mapping->wb_err, old)) {
719 /* Something changed, must use slow path */
720 spin_lock(&file->f_lock);
721 old = file->f_wb_err;
722 err = errseq_check_and_advance(&mapping->wb_err,
723 &file->f_wb_err);
724 trace_file_check_and_advance_wb_err(file, old);
725 spin_unlock(&file->f_lock);
726 }
727
728 /*
729 * We're mostly using this function as a drop in replacement for
730 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
731 * that the legacy code would have had on these flags.
732 */
733 clear_bit(AS_EIO, &mapping->flags);
734 clear_bit(AS_ENOSPC, &mapping->flags);
735 return err;
736 }
737 EXPORT_SYMBOL(file_check_and_advance_wb_err);
738
739 /**
740 * file_write_and_wait_range - write out & wait on a file range
741 * @file: file pointing to address_space with pages
742 * @lstart: offset in bytes where the range starts
743 * @lend: offset in bytes where the range ends (inclusive)
744 *
745 * Write out and wait upon file offsets lstart->lend, inclusive.
746 *
747 * Note that @lend is inclusive (describes the last byte to be written) so
748 * that this function can be used to write to the very end-of-file (end = -1).
749 *
750 * After writing out and waiting on the data, we check and advance the
751 * f_wb_err cursor to the latest value, and return any errors detected there.
752 *
753 * Return: %0 on success, negative error code otherwise.
754 */
file_write_and_wait_range(struct file * file,loff_t lstart,loff_t lend)755 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
756 {
757 int err = 0, err2;
758 struct address_space *mapping = file->f_mapping;
759
760 if (mapping_needs_writeback(mapping)) {
761 err = __filemap_fdatawrite_range(mapping, lstart, lend,
762 WB_SYNC_ALL);
763 /* See comment of filemap_write_and_wait() */
764 if (err != -EIO)
765 __filemap_fdatawait_range(mapping, lstart, lend);
766 }
767 err2 = file_check_and_advance_wb_err(file);
768 if (!err)
769 err = err2;
770 return err;
771 }
772 EXPORT_SYMBOL(file_write_and_wait_range);
773
774 /**
775 * replace_page_cache_page - replace a pagecache page with a new one
776 * @old: page to be replaced
777 * @new: page to replace with
778 * @gfp_mask: allocation mode
779 *
780 * This function replaces a page in the pagecache with a new one. On
781 * success it acquires the pagecache reference for the new page and
782 * drops it for the old page. Both the old and new pages must be
783 * locked. This function does not add the new page to the LRU, the
784 * caller must do that.
785 *
786 * The remove + add is atomic. This function cannot fail.
787 *
788 * Return: %0
789 */
replace_page_cache_page(struct page * old,struct page * new,gfp_t gfp_mask)790 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
791 {
792 struct address_space *mapping = old->mapping;
793 void (*freepage)(struct page *) = mapping->a_ops->freepage;
794 pgoff_t offset = old->index;
795 XA_STATE(xas, &mapping->i_pages, offset);
796 unsigned long flags;
797
798 VM_BUG_ON_PAGE(!PageLocked(old), old);
799 VM_BUG_ON_PAGE(!PageLocked(new), new);
800 VM_BUG_ON_PAGE(new->mapping, new);
801
802 get_page(new);
803 new->mapping = mapping;
804 new->index = offset;
805
806 mem_cgroup_migrate(old, new);
807
808 xas_lock_irqsave(&xas, flags);
809 xas_store(&xas, new);
810
811 old->mapping = NULL;
812 /* hugetlb pages do not participate in page cache accounting. */
813 if (!PageHuge(old))
814 __dec_lruvec_page_state(old, NR_FILE_PAGES);
815 if (!PageHuge(new))
816 __inc_lruvec_page_state(new, NR_FILE_PAGES);
817 if (PageSwapBacked(old))
818 __dec_lruvec_page_state(old, NR_SHMEM);
819 if (PageSwapBacked(new))
820 __inc_lruvec_page_state(new, NR_SHMEM);
821 xas_unlock_irqrestore(&xas, flags);
822 if (freepage)
823 freepage(old);
824 put_page(old);
825
826 return 0;
827 }
828 EXPORT_SYMBOL_GPL(replace_page_cache_page);
829
__add_to_page_cache_locked(struct page * page,struct address_space * mapping,pgoff_t offset,gfp_t gfp,void ** shadowp)830 noinline int __add_to_page_cache_locked(struct page *page,
831 struct address_space *mapping,
832 pgoff_t offset, gfp_t gfp,
833 void **shadowp)
834 {
835 XA_STATE(xas, &mapping->i_pages, offset);
836 int huge = PageHuge(page);
837 int error;
838 bool charged = false;
839
840 VM_BUG_ON_PAGE(!PageLocked(page), page);
841 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
842 mapping_set_update(&xas, mapping);
843
844 get_page(page);
845 page->mapping = mapping;
846 page->index = offset;
847
848 if (!huge) {
849 error = mem_cgroup_charge(page, current->mm, gfp);
850 if (error)
851 goto error;
852 charged = true;
853 }
854
855 gfp &= GFP_RECLAIM_MASK;
856
857 do {
858 unsigned int order = xa_get_order(xas.xa, xas.xa_index);
859 void *entry, *old = NULL;
860
861 if (order > thp_order(page))
862 xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
863 order, gfp);
864 xas_lock_irq(&xas);
865 xas_for_each_conflict(&xas, entry) {
866 old = entry;
867 if (!xa_is_value(entry)) {
868 xas_set_err(&xas, -EEXIST);
869 goto unlock;
870 }
871 }
872
873 if (old) {
874 if (shadowp)
875 *shadowp = old;
876 /* entry may have been split before we acquired lock */
877 order = xa_get_order(xas.xa, xas.xa_index);
878 if (order > thp_order(page)) {
879 xas_split(&xas, old, order);
880 xas_reset(&xas);
881 }
882 }
883
884 xas_store(&xas, page);
885 if (xas_error(&xas))
886 goto unlock;
887
888 if (old)
889 mapping->nrexceptional--;
890 mapping->nrpages++;
891
892 /* hugetlb pages do not participate in page cache accounting */
893 if (!huge)
894 __inc_lruvec_page_state(page, NR_FILE_PAGES);
895 unlock:
896 xas_unlock_irq(&xas);
897 } while (xas_nomem(&xas, gfp));
898
899 if (xas_error(&xas)) {
900 error = xas_error(&xas);
901 if (charged)
902 mem_cgroup_uncharge(page);
903 goto error;
904 }
905
906 trace_mm_filemap_add_to_page_cache(page);
907 return 0;
908 error:
909 page->mapping = NULL;
910 /* Leave page->index set: truncation relies upon it */
911 put_page(page);
912 return error;
913 }
914 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
915
916 /**
917 * add_to_page_cache_locked - add a locked page to the pagecache
918 * @page: page to add
919 * @mapping: the page's address_space
920 * @offset: page index
921 * @gfp_mask: page allocation mode
922 *
923 * This function is used to add a page to the pagecache. It must be locked.
924 * This function does not add the page to the LRU. The caller must do that.
925 *
926 * Return: %0 on success, negative error code otherwise.
927 */
add_to_page_cache_locked(struct page * page,struct address_space * mapping,pgoff_t offset,gfp_t gfp_mask)928 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
929 pgoff_t offset, gfp_t gfp_mask)
930 {
931 return __add_to_page_cache_locked(page, mapping, offset,
932 gfp_mask, NULL);
933 }
934 EXPORT_SYMBOL(add_to_page_cache_locked);
935
add_to_page_cache_lru(struct page * page,struct address_space * mapping,pgoff_t offset,gfp_t gfp_mask)936 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
937 pgoff_t offset, gfp_t gfp_mask)
938 {
939 void *shadow = NULL;
940 int ret;
941
942 __SetPageLocked(page);
943 ret = __add_to_page_cache_locked(page, mapping, offset,
944 gfp_mask, &shadow);
945 if (unlikely(ret))
946 __ClearPageLocked(page);
947 else {
948 /*
949 * The page might have been evicted from cache only
950 * recently, in which case it should be activated like
951 * any other repeatedly accessed page.
952 * The exception is pages getting rewritten; evicting other
953 * data from the working set, only to cache data that will
954 * get overwritten with something else, is a waste of memory.
955 */
956 WARN_ON_ONCE(PageActive(page));
957 if (!(gfp_mask & __GFP_WRITE) && shadow)
958 workingset_refault(page, shadow);
959 lru_cache_add(page);
960 }
961 return ret;
962 }
963 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
964
965 #ifdef CONFIG_NUMA
__page_cache_alloc(gfp_t gfp)966 struct page *__page_cache_alloc(gfp_t gfp)
967 {
968 int n;
969 struct page *page;
970
971 if (cpuset_do_page_mem_spread()) {
972 unsigned int cpuset_mems_cookie;
973 do {
974 cpuset_mems_cookie = read_mems_allowed_begin();
975 n = cpuset_mem_spread_node();
976 page = __alloc_pages_node(n, gfp, 0);
977 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
978
979 return page;
980 }
981 return alloc_pages(gfp, 0);
982 }
983 EXPORT_SYMBOL(__page_cache_alloc);
984 #endif
985
986 /*
987 * In order to wait for pages to become available there must be
988 * waitqueues associated with pages. By using a hash table of
989 * waitqueues where the bucket discipline is to maintain all
990 * waiters on the same queue and wake all when any of the pages
991 * become available, and for the woken contexts to check to be
992 * sure the appropriate page became available, this saves space
993 * at a cost of "thundering herd" phenomena during rare hash
994 * collisions.
995 */
996 #define PAGE_WAIT_TABLE_BITS 8
997 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
998 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
999
page_waitqueue(struct page * page)1000 static wait_queue_head_t *page_waitqueue(struct page *page)
1001 {
1002 return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1003 }
1004
pagecache_init(void)1005 void __init pagecache_init(void)
1006 {
1007 int i;
1008
1009 for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1010 init_waitqueue_head(&page_wait_table[i]);
1011
1012 page_writeback_init();
1013 }
1014
1015 /*
1016 * The page wait code treats the "wait->flags" somewhat unusually, because
1017 * we have multiple different kinds of waits, not just the usual "exclusive"
1018 * one.
1019 *
1020 * We have:
1021 *
1022 * (a) no special bits set:
1023 *
1024 * We're just waiting for the bit to be released, and when a waker
1025 * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1026 * and remove it from the wait queue.
1027 *
1028 * Simple and straightforward.
1029 *
1030 * (b) WQ_FLAG_EXCLUSIVE:
1031 *
1032 * The waiter is waiting to get the lock, and only one waiter should
1033 * be woken up to avoid any thundering herd behavior. We'll set the
1034 * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1035 *
1036 * This is the traditional exclusive wait.
1037 *
1038 * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1039 *
1040 * The waiter is waiting to get the bit, and additionally wants the
1041 * lock to be transferred to it for fair lock behavior. If the lock
1042 * cannot be taken, we stop walking the wait queue without waking
1043 * the waiter.
1044 *
1045 * This is the "fair lock handoff" case, and in addition to setting
1046 * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1047 * that it now has the lock.
1048 */
wake_page_function(wait_queue_entry_t * wait,unsigned mode,int sync,void * arg)1049 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1050 {
1051 unsigned int flags;
1052 struct wait_page_key *key = arg;
1053 struct wait_page_queue *wait_page
1054 = container_of(wait, struct wait_page_queue, wait);
1055
1056 if (!wake_page_match(wait_page, key))
1057 return 0;
1058
1059 /*
1060 * If it's a lock handoff wait, we get the bit for it, and
1061 * stop walking (and do not wake it up) if we can't.
1062 */
1063 flags = wait->flags;
1064 if (flags & WQ_FLAG_EXCLUSIVE) {
1065 if (test_bit(key->bit_nr, &key->page->flags))
1066 return -1;
1067 if (flags & WQ_FLAG_CUSTOM) {
1068 if (test_and_set_bit(key->bit_nr, &key->page->flags))
1069 return -1;
1070 flags |= WQ_FLAG_DONE;
1071 }
1072 }
1073
1074 /*
1075 * We are holding the wait-queue lock, but the waiter that
1076 * is waiting for this will be checking the flags without
1077 * any locking.
1078 *
1079 * So update the flags atomically, and wake up the waiter
1080 * afterwards to avoid any races. This store-release pairs
1081 * with the load-acquire in wait_on_page_bit_common().
1082 */
1083 smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1084 wake_up_state(wait->private, mode);
1085
1086 /*
1087 * Ok, we have successfully done what we're waiting for,
1088 * and we can unconditionally remove the wait entry.
1089 *
1090 * Note that this pairs with the "finish_wait()" in the
1091 * waiter, and has to be the absolute last thing we do.
1092 * After this list_del_init(&wait->entry) the wait entry
1093 * might be de-allocated and the process might even have
1094 * exited.
1095 */
1096 list_del_init_careful(&wait->entry);
1097 return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1098 }
1099
wake_up_page_bit(struct page * page,int bit_nr)1100 static void wake_up_page_bit(struct page *page, int bit_nr)
1101 {
1102 wait_queue_head_t *q = page_waitqueue(page);
1103 struct wait_page_key key;
1104 unsigned long flags;
1105 wait_queue_entry_t bookmark;
1106
1107 key.page = page;
1108 key.bit_nr = bit_nr;
1109 key.page_match = 0;
1110
1111 bookmark.flags = 0;
1112 bookmark.private = NULL;
1113 bookmark.func = NULL;
1114 INIT_LIST_HEAD(&bookmark.entry);
1115
1116 spin_lock_irqsave(&q->lock, flags);
1117 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1118
1119 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1120 /*
1121 * Take a breather from holding the lock,
1122 * allow pages that finish wake up asynchronously
1123 * to acquire the lock and remove themselves
1124 * from wait queue
1125 */
1126 spin_unlock_irqrestore(&q->lock, flags);
1127 cpu_relax();
1128 spin_lock_irqsave(&q->lock, flags);
1129 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1130 }
1131
1132 /*
1133 * It is possible for other pages to have collided on the waitqueue
1134 * hash, so in that case check for a page match. That prevents a long-
1135 * term waiter
1136 *
1137 * It is still possible to miss a case here, when we woke page waiters
1138 * and removed them from the waitqueue, but there are still other
1139 * page waiters.
1140 */
1141 if (!waitqueue_active(q) || !key.page_match) {
1142 ClearPageWaiters(page);
1143 /*
1144 * It's possible to miss clearing Waiters here, when we woke
1145 * our page waiters, but the hashed waitqueue has waiters for
1146 * other pages on it.
1147 *
1148 * That's okay, it's a rare case. The next waker will clear it.
1149 */
1150 }
1151 spin_unlock_irqrestore(&q->lock, flags);
1152 }
1153
wake_up_page(struct page * page,int bit)1154 static void wake_up_page(struct page *page, int bit)
1155 {
1156 if (!PageWaiters(page))
1157 return;
1158 wake_up_page_bit(page, bit);
1159 }
1160
1161 /*
1162 * A choice of three behaviors for wait_on_page_bit_common():
1163 */
1164 enum behavior {
1165 EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1166 * __lock_page() waiting on then setting PG_locked.
1167 */
1168 SHARED, /* Hold ref to page and check the bit when woken, like
1169 * wait_on_page_writeback() waiting on PG_writeback.
1170 */
1171 DROP, /* Drop ref to page before wait, no check when woken,
1172 * like put_and_wait_on_page_locked() on PG_locked.
1173 */
1174 };
1175
1176 /*
1177 * Attempt to check (or get) the page bit, and mark us done
1178 * if successful.
1179 */
trylock_page_bit_common(struct page * page,int bit_nr,struct wait_queue_entry * wait)1180 static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1181 struct wait_queue_entry *wait)
1182 {
1183 if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1184 if (test_and_set_bit(bit_nr, &page->flags))
1185 return false;
1186 } else if (test_bit(bit_nr, &page->flags))
1187 return false;
1188
1189 wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1190 return true;
1191 }
1192
1193 /* How many times do we accept lock stealing from under a waiter? */
1194 int sysctl_page_lock_unfairness = 5;
1195
wait_on_page_bit_common(wait_queue_head_t * q,struct page * page,int bit_nr,int state,enum behavior behavior)1196 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1197 struct page *page, int bit_nr, int state, enum behavior behavior)
1198 {
1199 int unfairness = sysctl_page_lock_unfairness;
1200 struct wait_page_queue wait_page;
1201 wait_queue_entry_t *wait = &wait_page.wait;
1202 bool thrashing = false;
1203 bool delayacct = false;
1204 unsigned long pflags;
1205
1206 if (bit_nr == PG_locked &&
1207 !PageUptodate(page) && PageWorkingset(page)) {
1208 if (!PageSwapBacked(page)) {
1209 delayacct_thrashing_start();
1210 delayacct = true;
1211 }
1212 psi_memstall_enter(&pflags);
1213 thrashing = true;
1214 }
1215
1216 init_wait(wait);
1217 wait->func = wake_page_function;
1218 wait_page.page = page;
1219 wait_page.bit_nr = bit_nr;
1220
1221 repeat:
1222 wait->flags = 0;
1223 if (behavior == EXCLUSIVE) {
1224 wait->flags = WQ_FLAG_EXCLUSIVE;
1225 if (--unfairness < 0)
1226 wait->flags |= WQ_FLAG_CUSTOM;
1227 }
1228
1229 /*
1230 * Do one last check whether we can get the
1231 * page bit synchronously.
1232 *
1233 * Do the SetPageWaiters() marking before that
1234 * to let any waker we _just_ missed know they
1235 * need to wake us up (otherwise they'll never
1236 * even go to the slow case that looks at the
1237 * page queue), and add ourselves to the wait
1238 * queue if we need to sleep.
1239 *
1240 * This part needs to be done under the queue
1241 * lock to avoid races.
1242 */
1243 spin_lock_irq(&q->lock);
1244 SetPageWaiters(page);
1245 if (!trylock_page_bit_common(page, bit_nr, wait))
1246 __add_wait_queue_entry_tail(q, wait);
1247 spin_unlock_irq(&q->lock);
1248
1249 /*
1250 * From now on, all the logic will be based on
1251 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1252 * see whether the page bit testing has already
1253 * been done by the wake function.
1254 *
1255 * We can drop our reference to the page.
1256 */
1257 if (behavior == DROP)
1258 put_page(page);
1259
1260 /*
1261 * Note that until the "finish_wait()", or until
1262 * we see the WQ_FLAG_WOKEN flag, we need to
1263 * be very careful with the 'wait->flags', because
1264 * we may race with a waker that sets them.
1265 */
1266 for (;;) {
1267 unsigned int flags;
1268
1269 set_current_state(state);
1270
1271 /* Loop until we've been woken or interrupted */
1272 flags = smp_load_acquire(&wait->flags);
1273 if (!(flags & WQ_FLAG_WOKEN)) {
1274 if (signal_pending_state(state, current))
1275 break;
1276
1277 io_schedule();
1278 continue;
1279 }
1280
1281 /* If we were non-exclusive, we're done */
1282 if (behavior != EXCLUSIVE)
1283 break;
1284
1285 /* If the waker got the lock for us, we're done */
1286 if (flags & WQ_FLAG_DONE)
1287 break;
1288
1289 /*
1290 * Otherwise, if we're getting the lock, we need to
1291 * try to get it ourselves.
1292 *
1293 * And if that fails, we'll have to retry this all.
1294 */
1295 if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1296 goto repeat;
1297
1298 wait->flags |= WQ_FLAG_DONE;
1299 break;
1300 }
1301
1302 /*
1303 * If a signal happened, this 'finish_wait()' may remove the last
1304 * waiter from the wait-queues, but the PageWaiters bit will remain
1305 * set. That's ok. The next wakeup will take care of it, and trying
1306 * to do it here would be difficult and prone to races.
1307 */
1308 finish_wait(q, wait);
1309
1310 if (thrashing) {
1311 if (delayacct)
1312 delayacct_thrashing_end();
1313 psi_memstall_leave(&pflags);
1314 }
1315
1316 /*
1317 * NOTE! The wait->flags weren't stable until we've done the
1318 * 'finish_wait()', and we could have exited the loop above due
1319 * to a signal, and had a wakeup event happen after the signal
1320 * test but before the 'finish_wait()'.
1321 *
1322 * So only after the finish_wait() can we reliably determine
1323 * if we got woken up or not, so we can now figure out the final
1324 * return value based on that state without races.
1325 *
1326 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1327 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1328 */
1329 if (behavior == EXCLUSIVE)
1330 return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1331
1332 return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1333 }
1334
wait_on_page_bit(struct page * page,int bit_nr)1335 void wait_on_page_bit(struct page *page, int bit_nr)
1336 {
1337 wait_queue_head_t *q = page_waitqueue(page);
1338 wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1339 }
1340 EXPORT_SYMBOL(wait_on_page_bit);
1341
wait_on_page_bit_killable(struct page * page,int bit_nr)1342 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1343 {
1344 wait_queue_head_t *q = page_waitqueue(page);
1345 return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1346 }
1347 EXPORT_SYMBOL(wait_on_page_bit_killable);
1348
__wait_on_page_locked_async(struct page * page,struct wait_page_queue * wait,bool set)1349 static int __wait_on_page_locked_async(struct page *page,
1350 struct wait_page_queue *wait, bool set)
1351 {
1352 struct wait_queue_head *q = page_waitqueue(page);
1353 int ret = 0;
1354
1355 wait->page = page;
1356 wait->bit_nr = PG_locked;
1357
1358 spin_lock_irq(&q->lock);
1359 __add_wait_queue_entry_tail(q, &wait->wait);
1360 SetPageWaiters(page);
1361 if (set)
1362 ret = !trylock_page(page);
1363 else
1364 ret = PageLocked(page);
1365 /*
1366 * If we were succesful now, we know we're still on the
1367 * waitqueue as we're still under the lock. This means it's
1368 * safe to remove and return success, we know the callback
1369 * isn't going to trigger.
1370 */
1371 if (!ret)
1372 __remove_wait_queue(q, &wait->wait);
1373 else
1374 ret = -EIOCBQUEUED;
1375 spin_unlock_irq(&q->lock);
1376 return ret;
1377 }
1378
wait_on_page_locked_async(struct page * page,struct wait_page_queue * wait)1379 static int wait_on_page_locked_async(struct page *page,
1380 struct wait_page_queue *wait)
1381 {
1382 if (!PageLocked(page))
1383 return 0;
1384 return __wait_on_page_locked_async(compound_head(page), wait, false);
1385 }
1386
1387 /**
1388 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1389 * @page: The page to wait for.
1390 *
1391 * The caller should hold a reference on @page. They expect the page to
1392 * become unlocked relatively soon, but do not wish to hold up migration
1393 * (for example) by holding the reference while waiting for the page to
1394 * come unlocked. After this function returns, the caller should not
1395 * dereference @page.
1396 */
put_and_wait_on_page_locked(struct page * page)1397 void put_and_wait_on_page_locked(struct page *page)
1398 {
1399 wait_queue_head_t *q;
1400
1401 page = compound_head(page);
1402 q = page_waitqueue(page);
1403 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1404 }
1405
1406 /**
1407 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1408 * @page: Page defining the wait queue of interest
1409 * @waiter: Waiter to add to the queue
1410 *
1411 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1412 */
add_page_wait_queue(struct page * page,wait_queue_entry_t * waiter)1413 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1414 {
1415 wait_queue_head_t *q = page_waitqueue(page);
1416 unsigned long flags;
1417
1418 spin_lock_irqsave(&q->lock, flags);
1419 __add_wait_queue_entry_tail(q, waiter);
1420 SetPageWaiters(page);
1421 spin_unlock_irqrestore(&q->lock, flags);
1422 }
1423 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1424
1425 #ifndef clear_bit_unlock_is_negative_byte
1426
1427 /*
1428 * PG_waiters is the high bit in the same byte as PG_lock.
1429 *
1430 * On x86 (and on many other architectures), we can clear PG_lock and
1431 * test the sign bit at the same time. But if the architecture does
1432 * not support that special operation, we just do this all by hand
1433 * instead.
1434 *
1435 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1436 * being cleared, but a memory barrier should be unnecessary since it is
1437 * in the same byte as PG_locked.
1438 */
clear_bit_unlock_is_negative_byte(long nr,volatile void * mem)1439 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1440 {
1441 clear_bit_unlock(nr, mem);
1442 /* smp_mb__after_atomic(); */
1443 return test_bit(PG_waiters, mem);
1444 }
1445
1446 #endif
1447
1448 /**
1449 * unlock_page - unlock a locked page
1450 * @page: the page
1451 *
1452 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1453 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1454 * mechanism between PageLocked pages and PageWriteback pages is shared.
1455 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1456 *
1457 * Note that this depends on PG_waiters being the sign bit in the byte
1458 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1459 * clear the PG_locked bit and test PG_waiters at the same time fairly
1460 * portably (architectures that do LL/SC can test any bit, while x86 can
1461 * test the sign bit).
1462 */
unlock_page(struct page * page)1463 void unlock_page(struct page *page)
1464 {
1465 BUILD_BUG_ON(PG_waiters != 7);
1466 page = compound_head(page);
1467 VM_BUG_ON_PAGE(!PageLocked(page), page);
1468 if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1469 wake_up_page_bit(page, PG_locked);
1470 }
1471 EXPORT_SYMBOL(unlock_page);
1472
1473 /**
1474 * end_page_writeback - end writeback against a page
1475 * @page: the page
1476 */
end_page_writeback(struct page * page)1477 void end_page_writeback(struct page *page)
1478 {
1479 /*
1480 * TestClearPageReclaim could be used here but it is an atomic
1481 * operation and overkill in this particular case. Failing to
1482 * shuffle a page marked for immediate reclaim is too mild to
1483 * justify taking an atomic operation penalty at the end of
1484 * ever page writeback.
1485 */
1486 if (PageReclaim(page)) {
1487 ClearPageReclaim(page);
1488 rotate_reclaimable_page(page);
1489 }
1490
1491 /*
1492 * Writeback does not hold a page reference of its own, relying
1493 * on truncation to wait for the clearing of PG_writeback.
1494 * But here we must make sure that the page is not freed and
1495 * reused before the wake_up_page().
1496 */
1497 get_page(page);
1498 if (!test_clear_page_writeback(page))
1499 BUG();
1500
1501 smp_mb__after_atomic();
1502 wake_up_page(page, PG_writeback);
1503 put_page(page);
1504 }
1505 EXPORT_SYMBOL(end_page_writeback);
1506
1507 /*
1508 * After completing I/O on a page, call this routine to update the page
1509 * flags appropriately
1510 */
page_endio(struct page * page,bool is_write,int err)1511 void page_endio(struct page *page, bool is_write, int err)
1512 {
1513 if (!is_write) {
1514 if (!err) {
1515 SetPageUptodate(page);
1516 } else {
1517 ClearPageUptodate(page);
1518 SetPageError(page);
1519 }
1520 unlock_page(page);
1521 } else {
1522 if (err) {
1523 struct address_space *mapping;
1524
1525 SetPageError(page);
1526 mapping = page_mapping(page);
1527 if (mapping)
1528 mapping_set_error(mapping, err);
1529 }
1530 end_page_writeback(page);
1531 }
1532 }
1533 EXPORT_SYMBOL_GPL(page_endio);
1534
1535 /**
1536 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1537 * @__page: the page to lock
1538 */
__lock_page(struct page * __page)1539 void __lock_page(struct page *__page)
1540 {
1541 struct page *page = compound_head(__page);
1542 wait_queue_head_t *q = page_waitqueue(page);
1543 wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1544 EXCLUSIVE);
1545 }
1546 EXPORT_SYMBOL(__lock_page);
1547
__lock_page_killable(struct page * __page)1548 int __lock_page_killable(struct page *__page)
1549 {
1550 struct page *page = compound_head(__page);
1551 wait_queue_head_t *q = page_waitqueue(page);
1552 return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1553 EXCLUSIVE);
1554 }
1555 EXPORT_SYMBOL_GPL(__lock_page_killable);
1556
__lock_page_async(struct page * page,struct wait_page_queue * wait)1557 int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1558 {
1559 return __wait_on_page_locked_async(page, wait, true);
1560 }
1561
1562 /*
1563 * Return values:
1564 * 1 - page is locked; mmap_lock is still held.
1565 * 0 - page is not locked.
1566 * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1567 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1568 * which case mmap_lock is still held.
1569 *
1570 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1571 * with the page locked and the mmap_lock unperturbed.
1572 */
__lock_page_or_retry(struct page * page,struct mm_struct * mm,unsigned int flags)1573 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1574 unsigned int flags)
1575 {
1576 if (fault_flag_allow_retry_first(flags)) {
1577 /*
1578 * CAUTION! In this case, mmap_lock is not released
1579 * even though return 0.
1580 */
1581 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1582 return 0;
1583
1584 mmap_read_unlock(mm);
1585 if (flags & FAULT_FLAG_KILLABLE)
1586 wait_on_page_locked_killable(page);
1587 else
1588 wait_on_page_locked(page);
1589 return 0;
1590 } else {
1591 if (flags & FAULT_FLAG_KILLABLE) {
1592 int ret;
1593
1594 ret = __lock_page_killable(page);
1595 if (ret) {
1596 mmap_read_unlock(mm);
1597 return 0;
1598 }
1599 } else
1600 __lock_page(page);
1601 return 1;
1602 }
1603 }
1604
1605 /**
1606 * page_cache_next_miss() - Find the next gap in the page cache.
1607 * @mapping: Mapping.
1608 * @index: Index.
1609 * @max_scan: Maximum range to search.
1610 *
1611 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1612 * gap with the lowest index.
1613 *
1614 * This function may be called under the rcu_read_lock. However, this will
1615 * not atomically search a snapshot of the cache at a single point in time.
1616 * For example, if a gap is created at index 5, then subsequently a gap is
1617 * created at index 10, page_cache_next_miss covering both indices may
1618 * return 10 if called under the rcu_read_lock.
1619 *
1620 * Return: The index of the gap if found, otherwise an index outside the
1621 * range specified (in which case 'return - index >= max_scan' will be true).
1622 * In the rare case of index wrap-around, 0 will be returned.
1623 */
page_cache_next_miss(struct address_space * mapping,pgoff_t index,unsigned long max_scan)1624 pgoff_t page_cache_next_miss(struct address_space *mapping,
1625 pgoff_t index, unsigned long max_scan)
1626 {
1627 XA_STATE(xas, &mapping->i_pages, index);
1628
1629 while (max_scan--) {
1630 void *entry = xas_next(&xas);
1631 if (!entry || xa_is_value(entry))
1632 break;
1633 if (xas.xa_index == 0)
1634 break;
1635 }
1636
1637 return xas.xa_index;
1638 }
1639 EXPORT_SYMBOL(page_cache_next_miss);
1640
1641 /**
1642 * page_cache_prev_miss() - Find the previous gap in the page cache.
1643 * @mapping: Mapping.
1644 * @index: Index.
1645 * @max_scan: Maximum range to search.
1646 *
1647 * Search the range [max(index - max_scan + 1, 0), index] for the
1648 * gap with the highest index.
1649 *
1650 * This function may be called under the rcu_read_lock. However, this will
1651 * not atomically search a snapshot of the cache at a single point in time.
1652 * For example, if a gap is created at index 10, then subsequently a gap is
1653 * created at index 5, page_cache_prev_miss() covering both indices may
1654 * return 5 if called under the rcu_read_lock.
1655 *
1656 * Return: The index of the gap if found, otherwise an index outside the
1657 * range specified (in which case 'index - return >= max_scan' will be true).
1658 * In the rare case of wrap-around, ULONG_MAX will be returned.
1659 */
page_cache_prev_miss(struct address_space * mapping,pgoff_t index,unsigned long max_scan)1660 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1661 pgoff_t index, unsigned long max_scan)
1662 {
1663 XA_STATE(xas, &mapping->i_pages, index);
1664
1665 while (max_scan--) {
1666 void *entry = xas_prev(&xas);
1667 if (!entry || xa_is_value(entry))
1668 break;
1669 if (xas.xa_index == ULONG_MAX)
1670 break;
1671 }
1672
1673 return xas.xa_index;
1674 }
1675 EXPORT_SYMBOL(page_cache_prev_miss);
1676
1677 /**
1678 * find_get_entry - find and get a page cache entry
1679 * @mapping: the address_space to search
1680 * @index: The page cache index.
1681 *
1682 * Looks up the page cache slot at @mapping & @offset. If there is a
1683 * page cache page, the head page is returned with an increased refcount.
1684 *
1685 * If the slot holds a shadow entry of a previously evicted page, or a
1686 * swap entry from shmem/tmpfs, it is returned.
1687 *
1688 * Return: The head page or shadow entry, %NULL if nothing is found.
1689 */
find_get_entry(struct address_space * mapping,pgoff_t index)1690 struct page *find_get_entry(struct address_space *mapping, pgoff_t index)
1691 {
1692 XA_STATE(xas, &mapping->i_pages, index);
1693 struct page *page;
1694
1695 rcu_read_lock();
1696 repeat:
1697 xas_reset(&xas);
1698 page = xas_load(&xas);
1699 if (xas_retry(&xas, page))
1700 goto repeat;
1701 /*
1702 * A shadow entry of a recently evicted page, or a swap entry from
1703 * shmem/tmpfs. Return it without attempting to raise page count.
1704 */
1705 if (!page || xa_is_value(page))
1706 goto out;
1707
1708 if (!page_cache_get_speculative(page))
1709 goto repeat;
1710
1711 /*
1712 * Has the page moved or been split?
1713 * This is part of the lockless pagecache protocol. See
1714 * include/linux/pagemap.h for details.
1715 */
1716 if (unlikely(page != xas_reload(&xas))) {
1717 put_page(page);
1718 goto repeat;
1719 }
1720 out:
1721 rcu_read_unlock();
1722
1723 return page;
1724 }
1725
1726 /**
1727 * find_lock_entry - Locate and lock a page cache entry.
1728 * @mapping: The address_space to search.
1729 * @index: The page cache index.
1730 *
1731 * Looks up the page at @mapping & @index. If there is a page in the
1732 * cache, the head page is returned locked and with an increased refcount.
1733 *
1734 * If the slot holds a shadow entry of a previously evicted page, or a
1735 * swap entry from shmem/tmpfs, it is returned.
1736 *
1737 * Context: May sleep.
1738 * Return: The head page or shadow entry, %NULL if nothing is found.
1739 */
find_lock_entry(struct address_space * mapping,pgoff_t index)1740 struct page *find_lock_entry(struct address_space *mapping, pgoff_t index)
1741 {
1742 struct page *page;
1743
1744 repeat:
1745 page = find_get_entry(mapping, index);
1746 if (page && !xa_is_value(page)) {
1747 lock_page(page);
1748 /* Has the page been truncated? */
1749 if (unlikely(page->mapping != mapping)) {
1750 unlock_page(page);
1751 put_page(page);
1752 goto repeat;
1753 }
1754 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1755 }
1756 return page;
1757 }
1758
1759 /**
1760 * pagecache_get_page - Find and get a reference to a page.
1761 * @mapping: The address_space to search.
1762 * @index: The page index.
1763 * @fgp_flags: %FGP flags modify how the page is returned.
1764 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1765 *
1766 * Looks up the page cache entry at @mapping & @index.
1767 *
1768 * @fgp_flags can be zero or more of these flags:
1769 *
1770 * * %FGP_ACCESSED - The page will be marked accessed.
1771 * * %FGP_LOCK - The page is returned locked.
1772 * * %FGP_HEAD - If the page is present and a THP, return the head page
1773 * rather than the exact page specified by the index.
1774 * * %FGP_CREAT - If no page is present then a new page is allocated using
1775 * @gfp_mask and added to the page cache and the VM's LRU list.
1776 * The page is returned locked and with an increased refcount.
1777 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1778 * page is already in cache. If the page was allocated, unlock it before
1779 * returning so the caller can do the same dance.
1780 * * %FGP_WRITE - The page will be written
1781 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1782 * * %FGP_NOWAIT - Don't get blocked by page lock
1783 *
1784 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1785 * if the %GFP flags specified for %FGP_CREAT are atomic.
1786 *
1787 * If there is a page cache page, it is returned with an increased refcount.
1788 *
1789 * Return: The found page or %NULL otherwise.
1790 */
pagecache_get_page(struct address_space * mapping,pgoff_t index,int fgp_flags,gfp_t gfp_mask)1791 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1792 int fgp_flags, gfp_t gfp_mask)
1793 {
1794 struct page *page;
1795
1796 repeat:
1797 page = find_get_entry(mapping, index);
1798 if (xa_is_value(page))
1799 page = NULL;
1800 if (!page)
1801 goto no_page;
1802
1803 if (fgp_flags & FGP_LOCK) {
1804 if (fgp_flags & FGP_NOWAIT) {
1805 if (!trylock_page(page)) {
1806 put_page(page);
1807 return NULL;
1808 }
1809 } else {
1810 lock_page(page);
1811 }
1812
1813 /* Has the page been truncated? */
1814 if (unlikely(page->mapping != mapping)) {
1815 unlock_page(page);
1816 put_page(page);
1817 goto repeat;
1818 }
1819 VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1820 }
1821
1822 if (fgp_flags & FGP_ACCESSED)
1823 mark_page_accessed(page);
1824 else if (fgp_flags & FGP_WRITE) {
1825 /* Clear idle flag for buffer write */
1826 if (page_is_idle(page))
1827 clear_page_idle(page);
1828 }
1829 if (!(fgp_flags & FGP_HEAD))
1830 page = find_subpage(page, index);
1831
1832 no_page:
1833 if (!page && (fgp_flags & FGP_CREAT)) {
1834 int err;
1835 if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1836 gfp_mask |= __GFP_WRITE;
1837 if (fgp_flags & FGP_NOFS)
1838 gfp_mask &= ~__GFP_FS;
1839
1840 page = __page_cache_alloc(gfp_mask);
1841 if (!page)
1842 return NULL;
1843
1844 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1845 fgp_flags |= FGP_LOCK;
1846
1847 /* Init accessed so avoid atomic mark_page_accessed later */
1848 if (fgp_flags & FGP_ACCESSED)
1849 __SetPageReferenced(page);
1850
1851 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1852 if (unlikely(err)) {
1853 put_page(page);
1854 page = NULL;
1855 if (err == -EEXIST)
1856 goto repeat;
1857 }
1858
1859 /*
1860 * add_to_page_cache_lru locks the page, and for mmap we expect
1861 * an unlocked page.
1862 */
1863 if (page && (fgp_flags & FGP_FOR_MMAP))
1864 unlock_page(page);
1865 }
1866
1867 return page;
1868 }
1869 EXPORT_SYMBOL(pagecache_get_page);
1870
1871 /**
1872 * find_get_entries - gang pagecache lookup
1873 * @mapping: The address_space to search
1874 * @start: The starting page cache index
1875 * @nr_entries: The maximum number of entries
1876 * @entries: Where the resulting entries are placed
1877 * @indices: The cache indices corresponding to the entries in @entries
1878 *
1879 * find_get_entries() will search for and return a group of up to
1880 * @nr_entries entries in the mapping. The entries are placed at
1881 * @entries. find_get_entries() takes a reference against any actual
1882 * pages it returns.
1883 *
1884 * The search returns a group of mapping-contiguous page cache entries
1885 * with ascending indexes. There may be holes in the indices due to
1886 * not-present pages.
1887 *
1888 * Any shadow entries of evicted pages, or swap entries from
1889 * shmem/tmpfs, are included in the returned array.
1890 *
1891 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1892 * stops at that page: the caller is likely to have a better way to handle
1893 * the compound page as a whole, and then skip its extent, than repeatedly
1894 * calling find_get_entries() to return all its tails.
1895 *
1896 * Return: the number of pages and shadow entries which were found.
1897 */
find_get_entries(struct address_space * mapping,pgoff_t start,unsigned int nr_entries,struct page ** entries,pgoff_t * indices)1898 unsigned find_get_entries(struct address_space *mapping,
1899 pgoff_t start, unsigned int nr_entries,
1900 struct page **entries, pgoff_t *indices)
1901 {
1902 XA_STATE(xas, &mapping->i_pages, start);
1903 struct page *page;
1904 unsigned int ret = 0;
1905
1906 if (!nr_entries)
1907 return 0;
1908
1909 rcu_read_lock();
1910 xas_for_each(&xas, page, ULONG_MAX) {
1911 if (xas_retry(&xas, page))
1912 continue;
1913 /*
1914 * A shadow entry of a recently evicted page, a swap
1915 * entry from shmem/tmpfs or a DAX entry. Return it
1916 * without attempting to raise page count.
1917 */
1918 if (xa_is_value(page))
1919 goto export;
1920
1921 if (!page_cache_get_speculative(page))
1922 goto retry;
1923
1924 /* Has the page moved or been split? */
1925 if (unlikely(page != xas_reload(&xas)))
1926 goto put_page;
1927
1928 /*
1929 * Terminate early on finding a THP, to allow the caller to
1930 * handle it all at once; but continue if this is hugetlbfs.
1931 */
1932 if (PageTransHuge(page) && !PageHuge(page)) {
1933 page = find_subpage(page, xas.xa_index);
1934 nr_entries = ret + 1;
1935 }
1936 export:
1937 indices[ret] = xas.xa_index;
1938 entries[ret] = page;
1939 if (++ret == nr_entries)
1940 break;
1941 continue;
1942 put_page:
1943 put_page(page);
1944 retry:
1945 xas_reset(&xas);
1946 }
1947 rcu_read_unlock();
1948 return ret;
1949 }
1950
1951 /**
1952 * find_get_pages_range - gang pagecache lookup
1953 * @mapping: The address_space to search
1954 * @start: The starting page index
1955 * @end: The final page index (inclusive)
1956 * @nr_pages: The maximum number of pages
1957 * @pages: Where the resulting pages are placed
1958 *
1959 * find_get_pages_range() will search for and return a group of up to @nr_pages
1960 * pages in the mapping starting at index @start and up to index @end
1961 * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1962 * a reference against the returned pages.
1963 *
1964 * The search returns a group of mapping-contiguous pages with ascending
1965 * indexes. There may be holes in the indices due to not-present pages.
1966 * We also update @start to index the next page for the traversal.
1967 *
1968 * Return: the number of pages which were found. If this number is
1969 * smaller than @nr_pages, the end of specified range has been
1970 * reached.
1971 */
find_get_pages_range(struct address_space * mapping,pgoff_t * start,pgoff_t end,unsigned int nr_pages,struct page ** pages)1972 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1973 pgoff_t end, unsigned int nr_pages,
1974 struct page **pages)
1975 {
1976 XA_STATE(xas, &mapping->i_pages, *start);
1977 struct page *page;
1978 unsigned ret = 0;
1979
1980 if (unlikely(!nr_pages))
1981 return 0;
1982
1983 rcu_read_lock();
1984 xas_for_each(&xas, page, end) {
1985 if (xas_retry(&xas, page))
1986 continue;
1987 /* Skip over shadow, swap and DAX entries */
1988 if (xa_is_value(page))
1989 continue;
1990
1991 if (!page_cache_get_speculative(page))
1992 goto retry;
1993
1994 /* Has the page moved or been split? */
1995 if (unlikely(page != xas_reload(&xas)))
1996 goto put_page;
1997
1998 pages[ret] = find_subpage(page, xas.xa_index);
1999 if (++ret == nr_pages) {
2000 *start = xas.xa_index + 1;
2001 goto out;
2002 }
2003 continue;
2004 put_page:
2005 put_page(page);
2006 retry:
2007 xas_reset(&xas);
2008 }
2009
2010 /*
2011 * We come here when there is no page beyond @end. We take care to not
2012 * overflow the index @start as it confuses some of the callers. This
2013 * breaks the iteration when there is a page at index -1 but that is
2014 * already broken anyway.
2015 */
2016 if (end == (pgoff_t)-1)
2017 *start = (pgoff_t)-1;
2018 else
2019 *start = end + 1;
2020 out:
2021 rcu_read_unlock();
2022
2023 return ret;
2024 }
2025
2026 /**
2027 * find_get_pages_contig - gang contiguous pagecache lookup
2028 * @mapping: The address_space to search
2029 * @index: The starting page index
2030 * @nr_pages: The maximum number of pages
2031 * @pages: Where the resulting pages are placed
2032 *
2033 * find_get_pages_contig() works exactly like find_get_pages(), except
2034 * that the returned number of pages are guaranteed to be contiguous.
2035 *
2036 * Return: the number of pages which were found.
2037 */
find_get_pages_contig(struct address_space * mapping,pgoff_t index,unsigned int nr_pages,struct page ** pages)2038 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2039 unsigned int nr_pages, struct page **pages)
2040 {
2041 XA_STATE(xas, &mapping->i_pages, index);
2042 struct page *page;
2043 unsigned int ret = 0;
2044
2045 if (unlikely(!nr_pages))
2046 return 0;
2047
2048 rcu_read_lock();
2049 for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2050 if (xas_retry(&xas, page))
2051 continue;
2052 /*
2053 * If the entry has been swapped out, we can stop looking.
2054 * No current caller is looking for DAX entries.
2055 */
2056 if (xa_is_value(page))
2057 break;
2058
2059 if (!page_cache_get_speculative(page))
2060 goto retry;
2061
2062 /* Has the page moved or been split? */
2063 if (unlikely(page != xas_reload(&xas)))
2064 goto put_page;
2065
2066 pages[ret] = find_subpage(page, xas.xa_index);
2067 if (++ret == nr_pages)
2068 break;
2069 continue;
2070 put_page:
2071 put_page(page);
2072 retry:
2073 xas_reset(&xas);
2074 }
2075 rcu_read_unlock();
2076 return ret;
2077 }
2078 EXPORT_SYMBOL(find_get_pages_contig);
2079
2080 /**
2081 * find_get_pages_range_tag - find and return pages in given range matching @tag
2082 * @mapping: the address_space to search
2083 * @index: the starting page index
2084 * @end: The final page index (inclusive)
2085 * @tag: the tag index
2086 * @nr_pages: the maximum number of pages
2087 * @pages: where the resulting pages are placed
2088 *
2089 * Like find_get_pages, except we only return pages which are tagged with
2090 * @tag. We update @index to index the next page for the traversal.
2091 *
2092 * Return: the number of pages which were found.
2093 */
find_get_pages_range_tag(struct address_space * mapping,pgoff_t * index,pgoff_t end,xa_mark_t tag,unsigned int nr_pages,struct page ** pages)2094 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2095 pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2096 struct page **pages)
2097 {
2098 XA_STATE(xas, &mapping->i_pages, *index);
2099 struct page *page;
2100 unsigned ret = 0;
2101
2102 if (unlikely(!nr_pages))
2103 return 0;
2104
2105 rcu_read_lock();
2106 xas_for_each_marked(&xas, page, end, tag) {
2107 if (xas_retry(&xas, page))
2108 continue;
2109 /*
2110 * Shadow entries should never be tagged, but this iteration
2111 * is lockless so there is a window for page reclaim to evict
2112 * a page we saw tagged. Skip over it.
2113 */
2114 if (xa_is_value(page))
2115 continue;
2116
2117 if (!page_cache_get_speculative(page))
2118 goto retry;
2119
2120 /* Has the page moved or been split? */
2121 if (unlikely(page != xas_reload(&xas)))
2122 goto put_page;
2123
2124 pages[ret] = find_subpage(page, xas.xa_index);
2125 if (++ret == nr_pages) {
2126 *index = xas.xa_index + 1;
2127 goto out;
2128 }
2129 continue;
2130 put_page:
2131 put_page(page);
2132 retry:
2133 xas_reset(&xas);
2134 }
2135
2136 /*
2137 * We come here when we got to @end. We take care to not overflow the
2138 * index @index as it confuses some of the callers. This breaks the
2139 * iteration when there is a page at index -1 but that is already
2140 * broken anyway.
2141 */
2142 if (end == (pgoff_t)-1)
2143 *index = (pgoff_t)-1;
2144 else
2145 *index = end + 1;
2146 out:
2147 rcu_read_unlock();
2148
2149 return ret;
2150 }
2151 EXPORT_SYMBOL(find_get_pages_range_tag);
2152
2153 /*
2154 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2155 * a _large_ part of the i/o request. Imagine the worst scenario:
2156 *
2157 * ---R__________________________________________B__________
2158 * ^ reading here ^ bad block(assume 4k)
2159 *
2160 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2161 * => failing the whole request => read(R) => read(R+1) =>
2162 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2163 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2164 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2165 *
2166 * It is going insane. Fix it by quickly scaling down the readahead size.
2167 */
shrink_readahead_size_eio(struct file_ra_state * ra)2168 static void shrink_readahead_size_eio(struct file_ra_state *ra)
2169 {
2170 ra->ra_pages /= 4;
2171 }
2172
2173 /**
2174 * generic_file_buffered_read - generic file read routine
2175 * @iocb: the iocb to read
2176 * @iter: data destination
2177 * @written: already copied
2178 *
2179 * This is a generic file read routine, and uses the
2180 * mapping->a_ops->readpage() function for the actual low-level stuff.
2181 *
2182 * This is really ugly. But the goto's actually try to clarify some
2183 * of the logic when it comes to error handling etc.
2184 *
2185 * Return:
2186 * * total number of bytes copied, including those the were already @written
2187 * * negative error code if nothing was copied
2188 */
generic_file_buffered_read(struct kiocb * iocb,struct iov_iter * iter,ssize_t written)2189 ssize_t generic_file_buffered_read(struct kiocb *iocb,
2190 struct iov_iter *iter, ssize_t written)
2191 {
2192 struct file *filp = iocb->ki_filp;
2193 struct address_space *mapping = filp->f_mapping;
2194 struct inode *inode = mapping->host;
2195 struct file_ra_state *ra = &filp->f_ra;
2196 loff_t *ppos = &iocb->ki_pos;
2197 pgoff_t index;
2198 pgoff_t last_index;
2199 pgoff_t prev_index;
2200 unsigned long offset; /* offset into pagecache page */
2201 unsigned int prev_offset;
2202 int error = 0;
2203
2204 if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2205 return 0;
2206 iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2207
2208 index = *ppos >> PAGE_SHIFT;
2209 prev_index = ra->prev_pos >> PAGE_SHIFT;
2210 prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2211 last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2212 offset = *ppos & ~PAGE_MASK;
2213
2214 /*
2215 * If we've already successfully copied some data, then we
2216 * can no longer safely return -EIOCBQUEUED. Hence mark
2217 * an async read NOWAIT at that point.
2218 */
2219 if (written && (iocb->ki_flags & IOCB_WAITQ))
2220 iocb->ki_flags |= IOCB_NOWAIT;
2221
2222 for (;;) {
2223 struct page *page;
2224 pgoff_t end_index;
2225 loff_t isize;
2226 unsigned long nr, ret;
2227
2228 cond_resched();
2229 find_page:
2230 if (fatal_signal_pending(current)) {
2231 error = -EINTR;
2232 goto out;
2233 }
2234
2235 page = find_get_page(mapping, index);
2236 if (!page) {
2237 if (iocb->ki_flags & IOCB_NOIO)
2238 goto would_block;
2239 page_cache_sync_readahead(mapping,
2240 ra, filp,
2241 index, last_index - index);
2242 page = find_get_page(mapping, index);
2243 if (unlikely(page == NULL))
2244 goto no_cached_page;
2245 }
2246 if (PageReadahead(page)) {
2247 if (iocb->ki_flags & IOCB_NOIO) {
2248 put_page(page);
2249 goto out;
2250 }
2251 page_cache_async_readahead(mapping,
2252 ra, filp, page,
2253 index, last_index - index);
2254 }
2255 if (!PageUptodate(page)) {
2256 /*
2257 * See comment in do_read_cache_page on why
2258 * wait_on_page_locked is used to avoid unnecessarily
2259 * serialisations and why it's safe.
2260 */
2261 if (iocb->ki_flags & IOCB_WAITQ) {
2262 if (written) {
2263 put_page(page);
2264 goto out;
2265 }
2266 error = wait_on_page_locked_async(page,
2267 iocb->ki_waitq);
2268 } else {
2269 if (iocb->ki_flags & IOCB_NOWAIT) {
2270 put_page(page);
2271 goto would_block;
2272 }
2273 error = wait_on_page_locked_killable(page);
2274 }
2275 if (unlikely(error))
2276 goto readpage_error;
2277 if (PageUptodate(page))
2278 goto page_ok;
2279
2280 if (inode->i_blkbits == PAGE_SHIFT ||
2281 !mapping->a_ops->is_partially_uptodate)
2282 goto page_not_up_to_date;
2283 /* pipes can't handle partially uptodate pages */
2284 if (unlikely(iov_iter_is_pipe(iter)))
2285 goto page_not_up_to_date;
2286 if (!trylock_page(page))
2287 goto page_not_up_to_date;
2288 /* Did it get truncated before we got the lock? */
2289 if (!page->mapping)
2290 goto page_not_up_to_date_locked;
2291 if (!mapping->a_ops->is_partially_uptodate(page,
2292 offset, iter->count))
2293 goto page_not_up_to_date_locked;
2294 unlock_page(page);
2295 }
2296 page_ok:
2297 /*
2298 * i_size must be checked after we know the page is Uptodate.
2299 *
2300 * Checking i_size after the check allows us to calculate
2301 * the correct value for "nr", which means the zero-filled
2302 * part of the page is not copied back to userspace (unless
2303 * another truncate extends the file - this is desired though).
2304 */
2305
2306 isize = i_size_read(inode);
2307 end_index = (isize - 1) >> PAGE_SHIFT;
2308 if (unlikely(!isize || index > end_index)) {
2309 put_page(page);
2310 goto out;
2311 }
2312
2313 /* nr is the maximum number of bytes to copy from this page */
2314 nr = PAGE_SIZE;
2315 if (index == end_index) {
2316 nr = ((isize - 1) & ~PAGE_MASK) + 1;
2317 if (nr <= offset) {
2318 put_page(page);
2319 goto out;
2320 }
2321 }
2322 nr = nr - offset;
2323
2324 /* If users can be writing to this page using arbitrary
2325 * virtual addresses, take care about potential aliasing
2326 * before reading the page on the kernel side.
2327 */
2328 if (mapping_writably_mapped(mapping))
2329 flush_dcache_page(page);
2330
2331 /*
2332 * When a sequential read accesses a page several times,
2333 * only mark it as accessed the first time.
2334 */
2335 if (prev_index != index || offset != prev_offset)
2336 mark_page_accessed(page);
2337 prev_index = index;
2338
2339 /*
2340 * Ok, we have the page, and it's up-to-date, so
2341 * now we can copy it to user space...
2342 */
2343
2344 ret = copy_page_to_iter(page, offset, nr, iter);
2345 offset += ret;
2346 index += offset >> PAGE_SHIFT;
2347 offset &= ~PAGE_MASK;
2348 prev_offset = offset;
2349
2350 put_page(page);
2351 written += ret;
2352 if (!iov_iter_count(iter))
2353 goto out;
2354 if (ret < nr) {
2355 error = -EFAULT;
2356 goto out;
2357 }
2358 continue;
2359
2360 page_not_up_to_date:
2361 /* Get exclusive access to the page ... */
2362 if (iocb->ki_flags & IOCB_WAITQ) {
2363 if (written) {
2364 put_page(page);
2365 goto out;
2366 }
2367 error = lock_page_async(page, iocb->ki_waitq);
2368 } else {
2369 error = lock_page_killable(page);
2370 }
2371 if (unlikely(error))
2372 goto readpage_error;
2373
2374 page_not_up_to_date_locked:
2375 /* Did it get truncated before we got the lock? */
2376 if (!page->mapping) {
2377 unlock_page(page);
2378 put_page(page);
2379 continue;
2380 }
2381
2382 /* Did somebody else fill it already? */
2383 if (PageUptodate(page)) {
2384 unlock_page(page);
2385 goto page_ok;
2386 }
2387
2388 readpage:
2389 if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) {
2390 unlock_page(page);
2391 put_page(page);
2392 goto would_block;
2393 }
2394 /*
2395 * A previous I/O error may have been due to temporary
2396 * failures, eg. multipath errors.
2397 * PG_error will be set again if readpage fails.
2398 */
2399 ClearPageError(page);
2400 /* Start the actual read. The read will unlock the page. */
2401 error = mapping->a_ops->readpage(filp, page);
2402
2403 if (unlikely(error)) {
2404 if (error == AOP_TRUNCATED_PAGE) {
2405 put_page(page);
2406 error = 0;
2407 goto find_page;
2408 }
2409 goto readpage_error;
2410 }
2411
2412 if (!PageUptodate(page)) {
2413 if (iocb->ki_flags & IOCB_WAITQ) {
2414 if (written) {
2415 put_page(page);
2416 goto out;
2417 }
2418 error = lock_page_async(page, iocb->ki_waitq);
2419 } else {
2420 error = lock_page_killable(page);
2421 }
2422
2423 if (unlikely(error))
2424 goto readpage_error;
2425 if (!PageUptodate(page)) {
2426 if (page->mapping == NULL) {
2427 /*
2428 * invalidate_mapping_pages got it
2429 */
2430 unlock_page(page);
2431 put_page(page);
2432 goto find_page;
2433 }
2434 unlock_page(page);
2435 shrink_readahead_size_eio(ra);
2436 error = -EIO;
2437 goto readpage_error;
2438 }
2439 unlock_page(page);
2440 }
2441
2442 goto page_ok;
2443
2444 readpage_error:
2445 /* UHHUH! A synchronous read error occurred. Report it */
2446 put_page(page);
2447 goto out;
2448
2449 no_cached_page:
2450 /*
2451 * Ok, it wasn't cached, so we need to create a new
2452 * page..
2453 */
2454 page = page_cache_alloc(mapping);
2455 if (!page) {
2456 error = -ENOMEM;
2457 goto out;
2458 }
2459 error = add_to_page_cache_lru(page, mapping, index,
2460 mapping_gfp_constraint(mapping, GFP_KERNEL));
2461 if (error) {
2462 put_page(page);
2463 if (error == -EEXIST) {
2464 error = 0;
2465 goto find_page;
2466 }
2467 goto out;
2468 }
2469 goto readpage;
2470 }
2471
2472 would_block:
2473 error = -EAGAIN;
2474 out:
2475 ra->prev_pos = prev_index;
2476 ra->prev_pos <<= PAGE_SHIFT;
2477 ra->prev_pos |= prev_offset;
2478
2479 *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2480 file_accessed(filp);
2481 return written ? written : error;
2482 }
2483 EXPORT_SYMBOL_GPL(generic_file_buffered_read);
2484
2485 /**
2486 * generic_file_read_iter - generic filesystem read routine
2487 * @iocb: kernel I/O control block
2488 * @iter: destination for the data read
2489 *
2490 * This is the "read_iter()" routine for all filesystems
2491 * that can use the page cache directly.
2492 *
2493 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2494 * be returned when no data can be read without waiting for I/O requests
2495 * to complete; it doesn't prevent readahead.
2496 *
2497 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2498 * requests shall be made for the read or for readahead. When no data
2499 * can be read, -EAGAIN shall be returned. When readahead would be
2500 * triggered, a partial, possibly empty read shall be returned.
2501 *
2502 * Return:
2503 * * number of bytes copied, even for partial reads
2504 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2505 */
2506 ssize_t
generic_file_read_iter(struct kiocb * iocb,struct iov_iter * iter)2507 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2508 {
2509 size_t count = iov_iter_count(iter);
2510 ssize_t retval = 0;
2511
2512 if (!count)
2513 goto out; /* skip atime */
2514
2515 if (iocb->ki_flags & IOCB_DIRECT) {
2516 struct file *file = iocb->ki_filp;
2517 struct address_space *mapping = file->f_mapping;
2518 struct inode *inode = mapping->host;
2519 loff_t size;
2520
2521 size = i_size_read(inode);
2522 if (iocb->ki_flags & IOCB_NOWAIT) {
2523 if (filemap_range_has_page(mapping, iocb->ki_pos,
2524 iocb->ki_pos + count - 1))
2525 return -EAGAIN;
2526 } else {
2527 retval = filemap_write_and_wait_range(mapping,
2528 iocb->ki_pos,
2529 iocb->ki_pos + count - 1);
2530 if (retval < 0)
2531 goto out;
2532 }
2533
2534 file_accessed(file);
2535
2536 retval = mapping->a_ops->direct_IO(iocb, iter);
2537 if (retval >= 0) {
2538 iocb->ki_pos += retval;
2539 count -= retval;
2540 }
2541 iov_iter_revert(iter, count - iov_iter_count(iter));
2542
2543 /*
2544 * Btrfs can have a short DIO read if we encounter
2545 * compressed extents, so if there was an error, or if
2546 * we've already read everything we wanted to, or if
2547 * there was a short read because we hit EOF, go ahead
2548 * and return. Otherwise fallthrough to buffered io for
2549 * the rest of the read. Buffered reads will not work for
2550 * DAX files, so don't bother trying.
2551 */
2552 if (retval < 0 || !count || iocb->ki_pos >= size ||
2553 IS_DAX(inode))
2554 goto out;
2555 }
2556
2557 retval = generic_file_buffered_read(iocb, iter, retval);
2558 out:
2559 return retval;
2560 }
2561 EXPORT_SYMBOL(generic_file_read_iter);
2562
2563 #ifdef CONFIG_MMU
2564 #define MMAP_LOTSAMISS (100)
2565 /*
2566 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2567 * @vmf - the vm_fault for this fault.
2568 * @page - the page to lock.
2569 * @fpin - the pointer to the file we may pin (or is already pinned).
2570 *
2571 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2572 * It differs in that it actually returns the page locked if it returns 1 and 0
2573 * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2574 * will point to the pinned file and needs to be fput()'ed at a later point.
2575 */
lock_page_maybe_drop_mmap(struct vm_fault * vmf,struct page * page,struct file ** fpin)2576 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2577 struct file **fpin)
2578 {
2579 if (trylock_page(page))
2580 return 1;
2581
2582 /*
2583 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2584 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2585 * is supposed to work. We have way too many special cases..
2586 */
2587 if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2588 return 0;
2589
2590 *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2591 if (vmf->flags & FAULT_FLAG_KILLABLE) {
2592 if (__lock_page_killable(page)) {
2593 /*
2594 * We didn't have the right flags to drop the mmap_lock,
2595 * but all fault_handlers only check for fatal signals
2596 * if we return VM_FAULT_RETRY, so we need to drop the
2597 * mmap_lock here and return 0 if we don't have a fpin.
2598 */
2599 if (*fpin == NULL)
2600 mmap_read_unlock(vmf->vma->vm_mm);
2601 return 0;
2602 }
2603 } else
2604 __lock_page(page);
2605 return 1;
2606 }
2607
2608
2609 /*
2610 * Synchronous readahead happens when we don't even find a page in the page
2611 * cache at all. We don't want to perform IO under the mmap sem, so if we have
2612 * to drop the mmap sem we return the file that was pinned in order for us to do
2613 * that. If we didn't pin a file then we return NULL. The file that is
2614 * returned needs to be fput()'ed when we're done with it.
2615 */
do_sync_mmap_readahead(struct vm_fault * vmf)2616 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2617 {
2618 struct file *file = vmf->vma->vm_file;
2619 struct file_ra_state *ra = &file->f_ra;
2620 struct address_space *mapping = file->f_mapping;
2621 DEFINE_READAHEAD(ractl, file, mapping, vmf->pgoff);
2622 struct file *fpin = NULL;
2623 unsigned int mmap_miss;
2624
2625 /* If we don't want any read-ahead, don't bother */
2626 if (vmf->vma->vm_flags & VM_RAND_READ)
2627 return fpin;
2628 if (!ra->ra_pages)
2629 return fpin;
2630
2631 if (vmf->vma->vm_flags & VM_SEQ_READ) {
2632 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2633 page_cache_sync_ra(&ractl, ra, ra->ra_pages);
2634 return fpin;
2635 }
2636
2637 /* Avoid banging the cache line if not needed */
2638 mmap_miss = READ_ONCE(ra->mmap_miss);
2639 if (mmap_miss < MMAP_LOTSAMISS * 10)
2640 WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2641
2642 /*
2643 * Do we miss much more than hit in this file? If so,
2644 * stop bothering with read-ahead. It will only hurt.
2645 */
2646 if (mmap_miss > MMAP_LOTSAMISS)
2647 return fpin;
2648
2649 /*
2650 * mmap read-around
2651 */
2652 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2653 ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2654 ra->size = ra->ra_pages;
2655 ra->async_size = ra->ra_pages / 4;
2656 ractl._index = ra->start;
2657 do_page_cache_ra(&ractl, ra->size, ra->async_size);
2658 return fpin;
2659 }
2660
2661 /*
2662 * Asynchronous readahead happens when we find the page and PG_readahead,
2663 * so we want to possibly extend the readahead further. We return the file that
2664 * was pinned if we have to drop the mmap_lock in order to do IO.
2665 */
do_async_mmap_readahead(struct vm_fault * vmf,struct page * page)2666 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2667 struct page *page)
2668 {
2669 struct file *file = vmf->vma->vm_file;
2670 struct file_ra_state *ra = &file->f_ra;
2671 struct address_space *mapping = file->f_mapping;
2672 struct file *fpin = NULL;
2673 unsigned int mmap_miss;
2674 pgoff_t offset = vmf->pgoff;
2675
2676 /* If we don't want any read-ahead, don't bother */
2677 if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2678 return fpin;
2679 mmap_miss = READ_ONCE(ra->mmap_miss);
2680 if (mmap_miss)
2681 WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2682 if (PageReadahead(page)) {
2683 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2684 page_cache_async_readahead(mapping, ra, file,
2685 page, offset, ra->ra_pages);
2686 }
2687 return fpin;
2688 }
2689
2690 /**
2691 * filemap_fault - read in file data for page fault handling
2692 * @vmf: struct vm_fault containing details of the fault
2693 *
2694 * filemap_fault() is invoked via the vma operations vector for a
2695 * mapped memory region to read in file data during a page fault.
2696 *
2697 * The goto's are kind of ugly, but this streamlines the normal case of having
2698 * it in the page cache, and handles the special cases reasonably without
2699 * having a lot of duplicated code.
2700 *
2701 * vma->vm_mm->mmap_lock must be held on entry.
2702 *
2703 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2704 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2705 *
2706 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2707 * has not been released.
2708 *
2709 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2710 *
2711 * Return: bitwise-OR of %VM_FAULT_ codes.
2712 */
filemap_fault(struct vm_fault * vmf)2713 vm_fault_t filemap_fault(struct vm_fault *vmf)
2714 {
2715 int error;
2716 struct file *file = vmf->vma->vm_file;
2717 struct file *fpin = NULL;
2718 struct address_space *mapping = file->f_mapping;
2719 struct file_ra_state *ra = &file->f_ra;
2720 struct inode *inode = mapping->host;
2721 pgoff_t offset = vmf->pgoff;
2722 pgoff_t max_off;
2723 struct page *page;
2724 vm_fault_t ret = 0;
2725
2726 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2727 if (unlikely(offset >= max_off))
2728 return VM_FAULT_SIGBUS;
2729
2730 /*
2731 * Do we have something in the page cache already?
2732 */
2733 page = find_get_page(mapping, offset);
2734 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2735 /*
2736 * We found the page, so try async readahead before
2737 * waiting for the lock.
2738 */
2739 fpin = do_async_mmap_readahead(vmf, page);
2740 } else if (!page) {
2741 /* No page in the page cache at all */
2742 count_vm_event(PGMAJFAULT);
2743 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2744 ret = VM_FAULT_MAJOR;
2745 fpin = do_sync_mmap_readahead(vmf);
2746 retry_find:
2747 page = pagecache_get_page(mapping, offset,
2748 FGP_CREAT|FGP_FOR_MMAP,
2749 vmf->gfp_mask);
2750 if (!page) {
2751 if (fpin)
2752 goto out_retry;
2753 return VM_FAULT_OOM;
2754 }
2755 }
2756
2757 if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2758 goto out_retry;
2759
2760 /* Did it get truncated? */
2761 if (unlikely(compound_head(page)->mapping != mapping)) {
2762 unlock_page(page);
2763 put_page(page);
2764 goto retry_find;
2765 }
2766 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2767
2768 /*
2769 * We have a locked page in the page cache, now we need to check
2770 * that it's up-to-date. If not, it is going to be due to an error.
2771 */
2772 if (unlikely(!PageUptodate(page)))
2773 goto page_not_uptodate;
2774
2775 /*
2776 * We've made it this far and we had to drop our mmap_lock, now is the
2777 * time to return to the upper layer and have it re-find the vma and
2778 * redo the fault.
2779 */
2780 if (fpin) {
2781 unlock_page(page);
2782 goto out_retry;
2783 }
2784
2785 /*
2786 * Found the page and have a reference on it.
2787 * We must recheck i_size under page lock.
2788 */
2789 max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2790 if (unlikely(offset >= max_off)) {
2791 unlock_page(page);
2792 put_page(page);
2793 return VM_FAULT_SIGBUS;
2794 }
2795
2796 vmf->page = page;
2797 return ret | VM_FAULT_LOCKED;
2798
2799 page_not_uptodate:
2800 /*
2801 * Umm, take care of errors if the page isn't up-to-date.
2802 * Try to re-read it _once_. We do this synchronously,
2803 * because there really aren't any performance issues here
2804 * and we need to check for errors.
2805 */
2806 ClearPageError(page);
2807 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2808 error = mapping->a_ops->readpage(file, page);
2809 if (!error) {
2810 wait_on_page_locked(page);
2811 if (!PageUptodate(page))
2812 error = -EIO;
2813 }
2814 if (fpin)
2815 goto out_retry;
2816 put_page(page);
2817
2818 if (!error || error == AOP_TRUNCATED_PAGE)
2819 goto retry_find;
2820
2821 shrink_readahead_size_eio(ra);
2822 return VM_FAULT_SIGBUS;
2823
2824 out_retry:
2825 /*
2826 * We dropped the mmap_lock, we need to return to the fault handler to
2827 * re-find the vma and come back and find our hopefully still populated
2828 * page.
2829 */
2830 if (page)
2831 put_page(page);
2832 if (fpin)
2833 fput(fpin);
2834 return ret | VM_FAULT_RETRY;
2835 }
2836 EXPORT_SYMBOL(filemap_fault);
2837
filemap_map_pages(struct vm_fault * vmf,pgoff_t start_pgoff,pgoff_t end_pgoff)2838 void filemap_map_pages(struct vm_fault *vmf,
2839 pgoff_t start_pgoff, pgoff_t end_pgoff)
2840 {
2841 struct file *file = vmf->vma->vm_file;
2842 struct address_space *mapping = file->f_mapping;
2843 pgoff_t last_pgoff = start_pgoff;
2844 unsigned long max_idx;
2845 XA_STATE(xas, &mapping->i_pages, start_pgoff);
2846 struct page *head, *page;
2847 unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
2848
2849 rcu_read_lock();
2850 xas_for_each(&xas, head, end_pgoff) {
2851 if (xas_retry(&xas, head))
2852 continue;
2853 if (xa_is_value(head))
2854 goto next;
2855
2856 /*
2857 * Check for a locked page first, as a speculative
2858 * reference may adversely influence page migration.
2859 */
2860 if (PageLocked(head))
2861 goto next;
2862 if (!page_cache_get_speculative(head))
2863 goto next;
2864
2865 /* Has the page moved or been split? */
2866 if (unlikely(head != xas_reload(&xas)))
2867 goto skip;
2868 page = find_subpage(head, xas.xa_index);
2869
2870 if (!PageUptodate(head) ||
2871 PageReadahead(page) ||
2872 PageHWPoison(page))
2873 goto skip;
2874 if (!trylock_page(head))
2875 goto skip;
2876
2877 if (head->mapping != mapping || !PageUptodate(head))
2878 goto unlock;
2879
2880 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2881 if (xas.xa_index >= max_idx)
2882 goto unlock;
2883
2884 if (mmap_miss > 0)
2885 mmap_miss--;
2886
2887 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2888 if (vmf->pte)
2889 vmf->pte += xas.xa_index - last_pgoff;
2890 last_pgoff = xas.xa_index;
2891 if (alloc_set_pte(vmf, page))
2892 goto unlock;
2893 unlock_page(head);
2894 goto next;
2895 unlock:
2896 unlock_page(head);
2897 skip:
2898 put_page(head);
2899 next:
2900 /* Huge page is mapped? No need to proceed. */
2901 if (pmd_trans_huge(*vmf->pmd))
2902 break;
2903 }
2904 rcu_read_unlock();
2905 WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
2906 }
2907 EXPORT_SYMBOL(filemap_map_pages);
2908
filemap_page_mkwrite(struct vm_fault * vmf)2909 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2910 {
2911 struct page *page = vmf->page;
2912 struct inode *inode = file_inode(vmf->vma->vm_file);
2913 vm_fault_t ret = VM_FAULT_LOCKED;
2914
2915 sb_start_pagefault(inode->i_sb);
2916 file_update_time(vmf->vma->vm_file);
2917 lock_page(page);
2918 if (page->mapping != inode->i_mapping) {
2919 unlock_page(page);
2920 ret = VM_FAULT_NOPAGE;
2921 goto out;
2922 }
2923 /*
2924 * We mark the page dirty already here so that when freeze is in
2925 * progress, we are guaranteed that writeback during freezing will
2926 * see the dirty page and writeprotect it again.
2927 */
2928 set_page_dirty(page);
2929 wait_for_stable_page(page);
2930 out:
2931 sb_end_pagefault(inode->i_sb);
2932 return ret;
2933 }
2934
2935 const struct vm_operations_struct generic_file_vm_ops = {
2936 .fault = filemap_fault,
2937 .map_pages = filemap_map_pages,
2938 .page_mkwrite = filemap_page_mkwrite,
2939 };
2940
2941 /* This is used for a general mmap of a disk file */
2942
generic_file_mmap(struct file * file,struct vm_area_struct * vma)2943 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2944 {
2945 struct address_space *mapping = file->f_mapping;
2946
2947 if (!mapping->a_ops->readpage)
2948 return -ENOEXEC;
2949 file_accessed(file);
2950 vma->vm_ops = &generic_file_vm_ops;
2951 return 0;
2952 }
2953
2954 /*
2955 * This is for filesystems which do not implement ->writepage.
2956 */
generic_file_readonly_mmap(struct file * file,struct vm_area_struct * vma)2957 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2958 {
2959 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2960 return -EINVAL;
2961 return generic_file_mmap(file, vma);
2962 }
2963 #else
filemap_page_mkwrite(struct vm_fault * vmf)2964 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2965 {
2966 return VM_FAULT_SIGBUS;
2967 }
generic_file_mmap(struct file * file,struct vm_area_struct * vma)2968 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2969 {
2970 return -ENOSYS;
2971 }
generic_file_readonly_mmap(struct file * file,struct vm_area_struct * vma)2972 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2973 {
2974 return -ENOSYS;
2975 }
2976 #endif /* CONFIG_MMU */
2977
2978 EXPORT_SYMBOL(filemap_page_mkwrite);
2979 EXPORT_SYMBOL(generic_file_mmap);
2980 EXPORT_SYMBOL(generic_file_readonly_mmap);
2981
wait_on_page_read(struct page * page)2982 static struct page *wait_on_page_read(struct page *page)
2983 {
2984 if (!IS_ERR(page)) {
2985 wait_on_page_locked(page);
2986 if (!PageUptodate(page)) {
2987 put_page(page);
2988 page = ERR_PTR(-EIO);
2989 }
2990 }
2991 return page;
2992 }
2993
do_read_cache_page(struct address_space * mapping,pgoff_t index,int (* filler)(void *,struct page *),void * data,gfp_t gfp)2994 static struct page *do_read_cache_page(struct address_space *mapping,
2995 pgoff_t index,
2996 int (*filler)(void *, struct page *),
2997 void *data,
2998 gfp_t gfp)
2999 {
3000 struct page *page;
3001 int err;
3002 repeat:
3003 page = find_get_page(mapping, index);
3004 if (!page) {
3005 page = __page_cache_alloc(gfp);
3006 if (!page)
3007 return ERR_PTR(-ENOMEM);
3008 err = add_to_page_cache_lru(page, mapping, index, gfp);
3009 if (unlikely(err)) {
3010 put_page(page);
3011 if (err == -EEXIST)
3012 goto repeat;
3013 /* Presumably ENOMEM for xarray node */
3014 return ERR_PTR(err);
3015 }
3016
3017 filler:
3018 if (filler)
3019 err = filler(data, page);
3020 else
3021 err = mapping->a_ops->readpage(data, page);
3022
3023 if (err < 0) {
3024 put_page(page);
3025 return ERR_PTR(err);
3026 }
3027
3028 page = wait_on_page_read(page);
3029 if (IS_ERR(page))
3030 return page;
3031 goto out;
3032 }
3033 if (PageUptodate(page))
3034 goto out;
3035
3036 /*
3037 * Page is not up to date and may be locked due to one of the following
3038 * case a: Page is being filled and the page lock is held
3039 * case b: Read/write error clearing the page uptodate status
3040 * case c: Truncation in progress (page locked)
3041 * case d: Reclaim in progress
3042 *
3043 * Case a, the page will be up to date when the page is unlocked.
3044 * There is no need to serialise on the page lock here as the page
3045 * is pinned so the lock gives no additional protection. Even if the
3046 * page is truncated, the data is still valid if PageUptodate as
3047 * it's a race vs truncate race.
3048 * Case b, the page will not be up to date
3049 * Case c, the page may be truncated but in itself, the data may still
3050 * be valid after IO completes as it's a read vs truncate race. The
3051 * operation must restart if the page is not uptodate on unlock but
3052 * otherwise serialising on page lock to stabilise the mapping gives
3053 * no additional guarantees to the caller as the page lock is
3054 * released before return.
3055 * Case d, similar to truncation. If reclaim holds the page lock, it
3056 * will be a race with remove_mapping that determines if the mapping
3057 * is valid on unlock but otherwise the data is valid and there is
3058 * no need to serialise with page lock.
3059 *
3060 * As the page lock gives no additional guarantee, we optimistically
3061 * wait on the page to be unlocked and check if it's up to date and
3062 * use the page if it is. Otherwise, the page lock is required to
3063 * distinguish between the different cases. The motivation is that we
3064 * avoid spurious serialisations and wakeups when multiple processes
3065 * wait on the same page for IO to complete.
3066 */
3067 wait_on_page_locked(page);
3068 if (PageUptodate(page))
3069 goto out;
3070
3071 /* Distinguish between all the cases under the safety of the lock */
3072 lock_page(page);
3073
3074 /* Case c or d, restart the operation */
3075 if (!page->mapping) {
3076 unlock_page(page);
3077 put_page(page);
3078 goto repeat;
3079 }
3080
3081 /* Someone else locked and filled the page in a very small window */
3082 if (PageUptodate(page)) {
3083 unlock_page(page);
3084 goto out;
3085 }
3086
3087 /*
3088 * A previous I/O error may have been due to temporary
3089 * failures.
3090 * Clear page error before actual read, PG_error will be
3091 * set again if read page fails.
3092 */
3093 ClearPageError(page);
3094 goto filler;
3095
3096 out:
3097 mark_page_accessed(page);
3098 return page;
3099 }
3100
3101 /**
3102 * read_cache_page - read into page cache, fill it if needed
3103 * @mapping: the page's address_space
3104 * @index: the page index
3105 * @filler: function to perform the read
3106 * @data: first arg to filler(data, page) function, often left as NULL
3107 *
3108 * Read into the page cache. If a page already exists, and PageUptodate() is
3109 * not set, try to fill the page and wait for it to become unlocked.
3110 *
3111 * If the page does not get brought uptodate, return -EIO.
3112 *
3113 * Return: up to date page on success, ERR_PTR() on failure.
3114 */
read_cache_page(struct address_space * mapping,pgoff_t index,int (* filler)(void *,struct page *),void * data)3115 struct page *read_cache_page(struct address_space *mapping,
3116 pgoff_t index,
3117 int (*filler)(void *, struct page *),
3118 void *data)
3119 {
3120 return do_read_cache_page(mapping, index, filler, data,
3121 mapping_gfp_mask(mapping));
3122 }
3123 EXPORT_SYMBOL(read_cache_page);
3124
3125 /**
3126 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3127 * @mapping: the page's address_space
3128 * @index: the page index
3129 * @gfp: the page allocator flags to use if allocating
3130 *
3131 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3132 * any new page allocations done using the specified allocation flags.
3133 *
3134 * If the page does not get brought uptodate, return -EIO.
3135 *
3136 * Return: up to date page on success, ERR_PTR() on failure.
3137 */
read_cache_page_gfp(struct address_space * mapping,pgoff_t index,gfp_t gfp)3138 struct page *read_cache_page_gfp(struct address_space *mapping,
3139 pgoff_t index,
3140 gfp_t gfp)
3141 {
3142 return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3143 }
3144 EXPORT_SYMBOL(read_cache_page_gfp);
3145
pagecache_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** pagep,void ** fsdata)3146 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3147 loff_t pos, unsigned len, unsigned flags,
3148 struct page **pagep, void **fsdata)
3149 {
3150 const struct address_space_operations *aops = mapping->a_ops;
3151
3152 return aops->write_begin(file, mapping, pos, len, flags,
3153 pagep, fsdata);
3154 }
3155 EXPORT_SYMBOL(pagecache_write_begin);
3156
pagecache_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)3157 int pagecache_write_end(struct file *file, struct address_space *mapping,
3158 loff_t pos, unsigned len, unsigned copied,
3159 struct page *page, void *fsdata)
3160 {
3161 const struct address_space_operations *aops = mapping->a_ops;
3162
3163 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3164 }
3165 EXPORT_SYMBOL(pagecache_write_end);
3166
3167 /*
3168 * Warn about a page cache invalidation failure during a direct I/O write.
3169 */
dio_warn_stale_pagecache(struct file * filp)3170 void dio_warn_stale_pagecache(struct file *filp)
3171 {
3172 static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3173 char pathname[128];
3174 struct inode *inode = file_inode(filp);
3175 char *path;
3176
3177 errseq_set(&inode->i_mapping->wb_err, -EIO);
3178 if (__ratelimit(&_rs)) {
3179 path = file_path(filp, pathname, sizeof(pathname));
3180 if (IS_ERR(path))
3181 path = "(unknown)";
3182 pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3183 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3184 current->comm);
3185 }
3186 }
3187
3188 ssize_t
generic_file_direct_write(struct kiocb * iocb,struct iov_iter * from)3189 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3190 {
3191 struct file *file = iocb->ki_filp;
3192 struct address_space *mapping = file->f_mapping;
3193 struct inode *inode = mapping->host;
3194 loff_t pos = iocb->ki_pos;
3195 ssize_t written;
3196 size_t write_len;
3197 pgoff_t end;
3198
3199 write_len = iov_iter_count(from);
3200 end = (pos + write_len - 1) >> PAGE_SHIFT;
3201
3202 if (iocb->ki_flags & IOCB_NOWAIT) {
3203 /* If there are pages to writeback, return */
3204 if (filemap_range_has_page(inode->i_mapping, pos,
3205 pos + write_len - 1))
3206 return -EAGAIN;
3207 } else {
3208 written = filemap_write_and_wait_range(mapping, pos,
3209 pos + write_len - 1);
3210 if (written)
3211 goto out;
3212 }
3213
3214 /*
3215 * After a write we want buffered reads to be sure to go to disk to get
3216 * the new data. We invalidate clean cached page from the region we're
3217 * about to write. We do this *before* the write so that we can return
3218 * without clobbering -EIOCBQUEUED from ->direct_IO().
3219 */
3220 written = invalidate_inode_pages2_range(mapping,
3221 pos >> PAGE_SHIFT, end);
3222 /*
3223 * If a page can not be invalidated, return 0 to fall back
3224 * to buffered write.
3225 */
3226 if (written) {
3227 if (written == -EBUSY)
3228 return 0;
3229 goto out;
3230 }
3231
3232 written = mapping->a_ops->direct_IO(iocb, from);
3233
3234 /*
3235 * Finally, try again to invalidate clean pages which might have been
3236 * cached by non-direct readahead, or faulted in by get_user_pages()
3237 * if the source of the write was an mmap'ed region of the file
3238 * we're writing. Either one is a pretty crazy thing to do,
3239 * so we don't support it 100%. If this invalidation
3240 * fails, tough, the write still worked...
3241 *
3242 * Most of the time we do not need this since dio_complete() will do
3243 * the invalidation for us. However there are some file systems that
3244 * do not end up with dio_complete() being called, so let's not break
3245 * them by removing it completely.
3246 *
3247 * Noticeable example is a blkdev_direct_IO().
3248 *
3249 * Skip invalidation for async writes or if mapping has no pages.
3250 */
3251 if (written > 0 && mapping->nrpages &&
3252 invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3253 dio_warn_stale_pagecache(file);
3254
3255 if (written > 0) {
3256 pos += written;
3257 write_len -= written;
3258 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3259 i_size_write(inode, pos);
3260 mark_inode_dirty(inode);
3261 }
3262 iocb->ki_pos = pos;
3263 }
3264 iov_iter_revert(from, write_len - iov_iter_count(from));
3265 out:
3266 return written;
3267 }
3268 EXPORT_SYMBOL(generic_file_direct_write);
3269
3270 /*
3271 * Find or create a page at the given pagecache position. Return the locked
3272 * page. This function is specifically for buffered writes.
3273 */
grab_cache_page_write_begin(struct address_space * mapping,pgoff_t index,unsigned flags)3274 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3275 pgoff_t index, unsigned flags)
3276 {
3277 struct page *page;
3278 int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3279
3280 if (flags & AOP_FLAG_NOFS)
3281 fgp_flags |= FGP_NOFS;
3282
3283 page = pagecache_get_page(mapping, index, fgp_flags,
3284 mapping_gfp_mask(mapping));
3285 if (page)
3286 wait_for_stable_page(page);
3287
3288 return page;
3289 }
3290 EXPORT_SYMBOL(grab_cache_page_write_begin);
3291
generic_perform_write(struct file * file,struct iov_iter * i,loff_t pos)3292 ssize_t generic_perform_write(struct file *file,
3293 struct iov_iter *i, loff_t pos)
3294 {
3295 struct address_space *mapping = file->f_mapping;
3296 const struct address_space_operations *a_ops = mapping->a_ops;
3297 long status = 0;
3298 ssize_t written = 0;
3299 unsigned int flags = 0;
3300
3301 do {
3302 struct page *page;
3303 unsigned long offset; /* Offset into pagecache page */
3304 unsigned long bytes; /* Bytes to write to page */
3305 size_t copied; /* Bytes copied from user */
3306 void *fsdata;
3307
3308 offset = (pos & (PAGE_SIZE - 1));
3309 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3310 iov_iter_count(i));
3311
3312 again:
3313 /*
3314 * Bring in the user page that we will copy from _first_.
3315 * Otherwise there's a nasty deadlock on copying from the
3316 * same page as we're writing to, without it being marked
3317 * up-to-date.
3318 *
3319 * Not only is this an optimisation, but it is also required
3320 * to check that the address is actually valid, when atomic
3321 * usercopies are used, below.
3322 */
3323 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3324 status = -EFAULT;
3325 break;
3326 }
3327
3328 if (fatal_signal_pending(current)) {
3329 status = -EINTR;
3330 break;
3331 }
3332
3333 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3334 &page, &fsdata);
3335 if (unlikely(status < 0))
3336 break;
3337
3338 if (mapping_writably_mapped(mapping))
3339 flush_dcache_page(page);
3340
3341 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3342 flush_dcache_page(page);
3343
3344 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3345 page, fsdata);
3346 if (unlikely(status < 0))
3347 break;
3348 copied = status;
3349
3350 cond_resched();
3351
3352 iov_iter_advance(i, copied);
3353 if (unlikely(copied == 0)) {
3354 /*
3355 * If we were unable to copy any data at all, we must
3356 * fall back to a single segment length write.
3357 *
3358 * If we didn't fallback here, we could livelock
3359 * because not all segments in the iov can be copied at
3360 * once without a pagefault.
3361 */
3362 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3363 iov_iter_single_seg_count(i));
3364 goto again;
3365 }
3366 pos += copied;
3367 written += copied;
3368
3369 balance_dirty_pages_ratelimited(mapping);
3370 } while (iov_iter_count(i));
3371
3372 return written ? written : status;
3373 }
3374 EXPORT_SYMBOL(generic_perform_write);
3375
3376 /**
3377 * __generic_file_write_iter - write data to a file
3378 * @iocb: IO state structure (file, offset, etc.)
3379 * @from: iov_iter with data to write
3380 *
3381 * This function does all the work needed for actually writing data to a
3382 * file. It does all basic checks, removes SUID from the file, updates
3383 * modification times and calls proper subroutines depending on whether we
3384 * do direct IO or a standard buffered write.
3385 *
3386 * It expects i_mutex to be grabbed unless we work on a block device or similar
3387 * object which does not need locking at all.
3388 *
3389 * This function does *not* take care of syncing data in case of O_SYNC write.
3390 * A caller has to handle it. This is mainly due to the fact that we want to
3391 * avoid syncing under i_mutex.
3392 *
3393 * Return:
3394 * * number of bytes written, even for truncated writes
3395 * * negative error code if no data has been written at all
3396 */
__generic_file_write_iter(struct kiocb * iocb,struct iov_iter * from)3397 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3398 {
3399 struct file *file = iocb->ki_filp;
3400 struct address_space * mapping = file->f_mapping;
3401 struct inode *inode = mapping->host;
3402 ssize_t written = 0;
3403 ssize_t err;
3404 ssize_t status;
3405
3406 /* We can write back this queue in page reclaim */
3407 current->backing_dev_info = inode_to_bdi(inode);
3408 err = file_remove_privs(file);
3409 if (err)
3410 goto out;
3411
3412 err = file_update_time(file);
3413 if (err)
3414 goto out;
3415
3416 if (iocb->ki_flags & IOCB_DIRECT) {
3417 loff_t pos, endbyte;
3418
3419 written = generic_file_direct_write(iocb, from);
3420 /*
3421 * If the write stopped short of completing, fall back to
3422 * buffered writes. Some filesystems do this for writes to
3423 * holes, for example. For DAX files, a buffered write will
3424 * not succeed (even if it did, DAX does not handle dirty
3425 * page-cache pages correctly).
3426 */
3427 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3428 goto out;
3429
3430 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3431 /*
3432 * If generic_perform_write() returned a synchronous error
3433 * then we want to return the number of bytes which were
3434 * direct-written, or the error code if that was zero. Note
3435 * that this differs from normal direct-io semantics, which
3436 * will return -EFOO even if some bytes were written.
3437 */
3438 if (unlikely(status < 0)) {
3439 err = status;
3440 goto out;
3441 }
3442 /*
3443 * We need to ensure that the page cache pages are written to
3444 * disk and invalidated to preserve the expected O_DIRECT
3445 * semantics.
3446 */
3447 endbyte = pos + status - 1;
3448 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3449 if (err == 0) {
3450 iocb->ki_pos = endbyte + 1;
3451 written += status;
3452 invalidate_mapping_pages(mapping,
3453 pos >> PAGE_SHIFT,
3454 endbyte >> PAGE_SHIFT);
3455 } else {
3456 /*
3457 * We don't know how much we wrote, so just return
3458 * the number of bytes which were direct-written
3459 */
3460 }
3461 } else {
3462 written = generic_perform_write(file, from, iocb->ki_pos);
3463 if (likely(written > 0))
3464 iocb->ki_pos += written;
3465 }
3466 out:
3467 current->backing_dev_info = NULL;
3468 return written ? written : err;
3469 }
3470 EXPORT_SYMBOL(__generic_file_write_iter);
3471
3472 /**
3473 * generic_file_write_iter - write data to a file
3474 * @iocb: IO state structure
3475 * @from: iov_iter with data to write
3476 *
3477 * This is a wrapper around __generic_file_write_iter() to be used by most
3478 * filesystems. It takes care of syncing the file in case of O_SYNC file
3479 * and acquires i_mutex as needed.
3480 * Return:
3481 * * negative error code if no data has been written at all of
3482 * vfs_fsync_range() failed for a synchronous write
3483 * * number of bytes written, even for truncated writes
3484 */
generic_file_write_iter(struct kiocb * iocb,struct iov_iter * from)3485 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3486 {
3487 struct file *file = iocb->ki_filp;
3488 struct inode *inode = file->f_mapping->host;
3489 ssize_t ret;
3490
3491 inode_lock(inode);
3492 ret = generic_write_checks(iocb, from);
3493 if (ret > 0)
3494 ret = __generic_file_write_iter(iocb, from);
3495 inode_unlock(inode);
3496
3497 if (ret > 0)
3498 ret = generic_write_sync(iocb, ret);
3499 return ret;
3500 }
3501 EXPORT_SYMBOL(generic_file_write_iter);
3502
3503 /**
3504 * try_to_release_page() - release old fs-specific metadata on a page
3505 *
3506 * @page: the page which the kernel is trying to free
3507 * @gfp_mask: memory allocation flags (and I/O mode)
3508 *
3509 * The address_space is to try to release any data against the page
3510 * (presumably at page->private).
3511 *
3512 * This may also be called if PG_fscache is set on a page, indicating that the
3513 * page is known to the local caching routines.
3514 *
3515 * The @gfp_mask argument specifies whether I/O may be performed to release
3516 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3517 *
3518 * Return: %1 if the release was successful, otherwise return zero.
3519 */
try_to_release_page(struct page * page,gfp_t gfp_mask)3520 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3521 {
3522 struct address_space * const mapping = page->mapping;
3523
3524 BUG_ON(!PageLocked(page));
3525 if (PageWriteback(page))
3526 return 0;
3527
3528 if (mapping && mapping->a_ops->releasepage)
3529 return mapping->a_ops->releasepage(page, gfp_mask);
3530 return try_to_free_buffers(page);
3531 }
3532
3533 EXPORT_SYMBOL(try_to_release_page);
3534