Line data Source code
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 <asm/pgalloc.h>
46 : #include <asm/tlbflush.h>
47 : #include "internal.h"
48 :
49 : #define CREATE_TRACE_POINTS
50 : #include <trace/events/filemap.h>
51 :
52 : /*
53 : * FIXME: remove all knowledge of the buffer layer from the core VM
54 : */
55 : #include <linux/buffer_head.h> /* for try_to_free_buffers */
56 :
57 : #include <asm/mman.h>
58 :
59 : /*
60 : * Shared mappings implemented 30.11.1994. It's not fully working yet,
61 : * though.
62 : *
63 : * Shared mappings now work. 15.8.1995 Bruno.
64 : *
65 : * finished 'unifying' the page and buffer cache and SMP-threaded the
66 : * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
67 : *
68 : * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
69 : */
70 :
71 : /*
72 : * Lock ordering:
73 : *
74 : * ->i_mmap_rwsem (truncate_pagecache)
75 : * ->private_lock (__free_pte->__set_page_dirty_buffers)
76 : * ->swap_lock (exclusive_swap_page, others)
77 : * ->i_pages lock
78 : *
79 : * ->i_mutex
80 : * ->i_mmap_rwsem (truncate->unmap_mapping_range)
81 : *
82 : * ->mmap_lock
83 : * ->i_mmap_rwsem
84 : * ->page_table_lock or pte_lock (various, mainly in memory.c)
85 : * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
86 : *
87 : * ->mmap_lock
88 : * ->lock_page (access_process_vm)
89 : *
90 : * ->i_mutex (generic_perform_write)
91 : * ->mmap_lock (fault_in_pages_readable->do_page_fault)
92 : *
93 : * bdi->wb.list_lock
94 : * sb_lock (fs/fs-writeback.c)
95 : * ->i_pages lock (__sync_single_inode)
96 : *
97 : * ->i_mmap_rwsem
98 : * ->anon_vma.lock (vma_adjust)
99 : *
100 : * ->anon_vma.lock
101 : * ->page_table_lock or pte_lock (anon_vma_prepare and various)
102 : *
103 : * ->page_table_lock or pte_lock
104 : * ->swap_lock (try_to_unmap_one)
105 : * ->private_lock (try_to_unmap_one)
106 : * ->i_pages lock (try_to_unmap_one)
107 : * ->lruvec->lru_lock (follow_page->mark_page_accessed)
108 : * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
109 : * ->private_lock (page_remove_rmap->set_page_dirty)
110 : * ->i_pages lock (page_remove_rmap->set_page_dirty)
111 : * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
112 : * ->inode->i_lock (page_remove_rmap->set_page_dirty)
113 : * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
114 : * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
115 : * ->inode->i_lock (zap_pte_range->set_page_dirty)
116 : * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
117 : *
118 : * ->i_mmap_rwsem
119 : * ->tasklist_lock (memory_failure, collect_procs_ao)
120 : */
121 :
122 0 : static void page_cache_delete(struct address_space *mapping,
123 : struct page *page, void *shadow)
124 : {
125 0 : XA_STATE(xas, &mapping->i_pages, page->index);
126 0 : unsigned int nr = 1;
127 :
128 0 : mapping_set_update(&xas, mapping);
129 :
130 : /* hugetlb pages are represented by a single entry in the xarray */
131 0 : if (!PageHuge(page)) {
132 0 : xas_set_order(&xas, page->index, compound_order(page));
133 0 : nr = compound_nr(page);
134 : }
135 :
136 0 : VM_BUG_ON_PAGE(!PageLocked(page), page);
137 0 : VM_BUG_ON_PAGE(PageTail(page), page);
138 0 : VM_BUG_ON_PAGE(nr != 1 && shadow, page);
139 :
140 0 : xas_store(&xas, shadow);
141 0 : xas_init_marks(&xas);
142 :
143 0 : page->mapping = NULL;
144 : /* Leave page->index set: truncation lookup relies upon it */
145 :
146 0 : if (shadow) {
147 0 : mapping->nrexceptional += nr;
148 : /*
149 : * Make sure the nrexceptional update is committed before
150 : * the nrpages update so that final truncate racing
151 : * with reclaim does not see both counters 0 at the
152 : * same time and miss a shadow entry.
153 : */
154 0 : smp_wmb();
155 : }
156 0 : mapping->nrpages -= nr;
157 0 : }
158 :
159 0 : static void unaccount_page_cache_page(struct address_space *mapping,
160 : struct page *page)
161 : {
162 0 : int nr;
163 :
164 : /*
165 : * if we're uptodate, flush out into the cleancache, otherwise
166 : * invalidate any existing cleancache entries. We can't leave
167 : * stale data around in the cleancache once our page is gone
168 : */
169 0 : if (PageUptodate(page) && PageMappedToDisk(page))
170 0 : cleancache_put_page(page);
171 : else
172 0 : cleancache_invalidate_page(mapping, page);
173 :
174 0 : VM_BUG_ON_PAGE(PageTail(page), page);
175 0 : VM_BUG_ON_PAGE(page_mapped(page), page);
176 0 : if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
177 : int mapcount;
178 :
179 : pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
180 : current->comm, page_to_pfn(page));
181 : dump_page(page, "still mapped when deleted");
182 : dump_stack();
183 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
184 :
185 : mapcount = page_mapcount(page);
186 : if (mapping_exiting(mapping) &&
187 : page_count(page) >= mapcount + 2) {
188 : /*
189 : * All vmas have already been torn down, so it's
190 : * a good bet that actually the page is unmapped,
191 : * and we'd prefer not to leak it: if we're wrong,
192 : * some other bad page check should catch it later.
193 : */
194 : page_mapcount_reset(page);
195 : page_ref_sub(page, mapcount);
196 : }
197 : }
198 :
199 : /* hugetlb pages do not participate in page cache accounting. */
200 0 : if (PageHuge(page))
201 : return;
202 :
203 0 : nr = thp_nr_pages(page);
204 :
205 0 : __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
206 0 : if (PageSwapBacked(page)) {
207 0 : __mod_lruvec_page_state(page, NR_SHMEM, -nr);
208 0 : if (PageTransHuge(page))
209 0 : __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
210 0 : } else if (PageTransHuge(page)) {
211 0 : __mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
212 0 : filemap_nr_thps_dec(mapping);
213 : }
214 :
215 : /*
216 : * At this point page must be either written or cleaned by
217 : * truncate. Dirty page here signals a bug and loss of
218 : * unwritten data.
219 : *
220 : * This fixes dirty accounting after removing the page entirely
221 : * but leaves PageDirty set: it has no effect for truncated
222 : * page and anyway will be cleared before returning page into
223 : * buddy allocator.
224 : */
225 0 : if (WARN_ON_ONCE(PageDirty(page)))
226 0 : account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
227 : }
228 :
229 : /*
230 : * Delete a page from the page cache and free it. Caller has to make
231 : * sure the page is locked and that nobody else uses it - or that usage
232 : * is safe. The caller must hold the i_pages lock.
233 : */
234 0 : void __delete_from_page_cache(struct page *page, void *shadow)
235 : {
236 0 : struct address_space *mapping = page->mapping;
237 :
238 0 : trace_mm_filemap_delete_from_page_cache(page);
239 :
240 0 : unaccount_page_cache_page(mapping, page);
241 0 : page_cache_delete(mapping, page, shadow);
242 0 : }
243 :
244 0 : static void page_cache_free_page(struct address_space *mapping,
245 : struct page *page)
246 : {
247 0 : void (*freepage)(struct page *);
248 :
249 0 : freepage = mapping->a_ops->freepage;
250 0 : if (freepage)
251 0 : freepage(page);
252 :
253 0 : if (PageTransHuge(page) && !PageHuge(page)) {
254 0 : page_ref_sub(page, thp_nr_pages(page));
255 0 : VM_BUG_ON_PAGE(page_count(page) <= 0, page);
256 : } else {
257 0 : put_page(page);
258 : }
259 0 : }
260 :
261 : /**
262 : * delete_from_page_cache - delete page from page cache
263 : * @page: the page which the kernel is trying to remove from page cache
264 : *
265 : * This must be called only on pages that have been verified to be in the page
266 : * cache and locked. It will never put the page into the free list, the caller
267 : * has a reference on the page.
268 : */
269 0 : void delete_from_page_cache(struct page *page)
270 : {
271 0 : struct address_space *mapping = page_mapping(page);
272 0 : unsigned long flags;
273 :
274 0 : BUG_ON(!PageLocked(page));
275 0 : xa_lock_irqsave(&mapping->i_pages, flags);
276 0 : __delete_from_page_cache(page, NULL);
277 0 : xa_unlock_irqrestore(&mapping->i_pages, flags);
278 :
279 0 : page_cache_free_page(mapping, page);
280 0 : }
281 : EXPORT_SYMBOL(delete_from_page_cache);
282 :
283 : /*
284 : * page_cache_delete_batch - delete several pages from page cache
285 : * @mapping: the mapping to which pages belong
286 : * @pvec: pagevec with pages to delete
287 : *
288 : * The function walks over mapping->i_pages and removes pages passed in @pvec
289 : * from the mapping. The function expects @pvec to be sorted by page index
290 : * and is optimised for it to be dense.
291 : * It tolerates holes in @pvec (mapping entries at those indices are not
292 : * modified). The function expects only THP head pages to be present in the
293 : * @pvec.
294 : *
295 : * The function expects the i_pages lock to be held.
296 : */
297 0 : static void page_cache_delete_batch(struct address_space *mapping,
298 : struct pagevec *pvec)
299 : {
300 0 : XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
301 0 : int total_pages = 0;
302 0 : int i = 0;
303 0 : struct page *page;
304 :
305 0 : mapping_set_update(&xas, mapping);
306 0 : xas_for_each(&xas, page, ULONG_MAX) {
307 0 : if (i >= pagevec_count(pvec))
308 : break;
309 :
310 : /* A swap/dax/shadow entry got inserted? Skip it. */
311 0 : if (xa_is_value(page))
312 0 : continue;
313 : /*
314 : * A page got inserted in our range? Skip it. We have our
315 : * pages locked so they are protected from being removed.
316 : * If we see a page whose index is higher than ours, it
317 : * means our page has been removed, which shouldn't be
318 : * possible because we're holding the PageLock.
319 : */
320 0 : if (page != pvec->pages[i]) {
321 0 : VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
322 : page);
323 0 : continue;
324 : }
325 :
326 0 : WARN_ON_ONCE(!PageLocked(page));
327 :
328 0 : if (page->index == xas.xa_index)
329 0 : page->mapping = NULL;
330 : /* Leave page->index set: truncation lookup relies on it */
331 :
332 : /*
333 : * Move to the next page in the vector if this is a regular
334 : * page or the index is of the last sub-page of this compound
335 : * page.
336 : */
337 0 : if (page->index + compound_nr(page) - 1 == xas.xa_index)
338 0 : i++;
339 0 : xas_store(&xas, NULL);
340 0 : total_pages++;
341 : }
342 0 : mapping->nrpages -= total_pages;
343 0 : }
344 :
345 0 : void delete_from_page_cache_batch(struct address_space *mapping,
346 : struct pagevec *pvec)
347 : {
348 0 : int i;
349 0 : unsigned long flags;
350 :
351 0 : if (!pagevec_count(pvec))
352 : return;
353 :
354 0 : xa_lock_irqsave(&mapping->i_pages, flags);
355 0 : for (i = 0; i < pagevec_count(pvec); i++) {
356 0 : trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
357 :
358 0 : unaccount_page_cache_page(mapping, pvec->pages[i]);
359 : }
360 0 : page_cache_delete_batch(mapping, pvec);
361 0 : xa_unlock_irqrestore(&mapping->i_pages, flags);
362 :
363 0 : for (i = 0; i < pagevec_count(pvec); i++)
364 0 : page_cache_free_page(mapping, pvec->pages[i]);
365 : }
366 :
367 0 : int filemap_check_errors(struct address_space *mapping)
368 : {
369 0 : int ret = 0;
370 : /* Check for outstanding write errors */
371 0 : if (test_bit(AS_ENOSPC, &mapping->flags) &&
372 0 : test_and_clear_bit(AS_ENOSPC, &mapping->flags))
373 0 : ret = -ENOSPC;
374 0 : if (test_bit(AS_EIO, &mapping->flags) &&
375 0 : test_and_clear_bit(AS_EIO, &mapping->flags))
376 0 : ret = -EIO;
377 0 : return ret;
378 : }
379 : EXPORT_SYMBOL(filemap_check_errors);
380 :
381 0 : static int filemap_check_and_keep_errors(struct address_space *mapping)
382 : {
383 : /* Check for outstanding write errors */
384 0 : if (test_bit(AS_EIO, &mapping->flags))
385 : return -EIO;
386 0 : if (test_bit(AS_ENOSPC, &mapping->flags))
387 0 : return -ENOSPC;
388 : return 0;
389 : }
390 :
391 : /**
392 : * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
393 : * @mapping: address space structure to write
394 : * @start: offset in bytes where the range starts
395 : * @end: offset in bytes where the range ends (inclusive)
396 : * @sync_mode: enable synchronous operation
397 : *
398 : * Start writeback against all of a mapping's dirty pages that lie
399 : * within the byte offsets <start, end> inclusive.
400 : *
401 : * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
402 : * opposed to a regular memory cleansing writeback. The difference between
403 : * these two operations is that if a dirty page/buffer is encountered, it must
404 : * be waited upon, and not just skipped over.
405 : *
406 : * Return: %0 on success, negative error code otherwise.
407 : */
408 0 : int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
409 : loff_t end, int sync_mode)
410 : {
411 0 : int ret;
412 0 : struct writeback_control wbc = {
413 : .sync_mode = sync_mode,
414 : .nr_to_write = LONG_MAX,
415 : .range_start = start,
416 : .range_end = end,
417 : };
418 :
419 0 : if (!mapping_can_writeback(mapping) ||
420 0 : !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
421 : return 0;
422 :
423 0 : wbc_attach_fdatawrite_inode(&wbc, mapping->host);
424 0 : ret = do_writepages(mapping, &wbc);
425 0 : wbc_detach_inode(&wbc);
426 0 : return ret;
427 : }
428 :
429 0 : static inline int __filemap_fdatawrite(struct address_space *mapping,
430 : int sync_mode)
431 : {
432 0 : return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
433 : }
434 :
435 0 : int filemap_fdatawrite(struct address_space *mapping)
436 : {
437 0 : return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
438 : }
439 : EXPORT_SYMBOL(filemap_fdatawrite);
440 :
441 0 : int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
442 : loff_t end)
443 : {
444 0 : return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
445 : }
446 : EXPORT_SYMBOL(filemap_fdatawrite_range);
447 :
448 : /**
449 : * filemap_flush - mostly a non-blocking flush
450 : * @mapping: target address_space
451 : *
452 : * This is a mostly non-blocking flush. Not suitable for data-integrity
453 : * purposes - I/O may not be started against all dirty pages.
454 : *
455 : * Return: %0 on success, negative error code otherwise.
456 : */
457 0 : int filemap_flush(struct address_space *mapping)
458 : {
459 0 : return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
460 : }
461 : EXPORT_SYMBOL(filemap_flush);
462 :
463 : /**
464 : * filemap_range_has_page - check if a page exists in range.
465 : * @mapping: address space within which to check
466 : * @start_byte: offset in bytes where the range starts
467 : * @end_byte: offset in bytes where the range ends (inclusive)
468 : *
469 : * Find at least one page in the range supplied, usually used to check if
470 : * direct writing in this range will trigger a writeback.
471 : *
472 : * Return: %true if at least one page exists in the specified range,
473 : * %false otherwise.
474 : */
475 0 : bool filemap_range_has_page(struct address_space *mapping,
476 : loff_t start_byte, loff_t end_byte)
477 : {
478 0 : struct page *page;
479 0 : XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
480 0 : pgoff_t max = end_byte >> PAGE_SHIFT;
481 :
482 0 : if (end_byte < start_byte)
483 : return false;
484 :
485 0 : rcu_read_lock();
486 0 : for (;;) {
487 0 : page = xas_find(&xas, max);
488 0 : if (xas_retry(&xas, page))
489 0 : continue;
490 : /* Shadow entries don't count */
491 0 : if (xa_is_value(page))
492 0 : continue;
493 : /*
494 : * We don't need to try to pin this page; we're about to
495 : * release the RCU lock anyway. It is enough to know that
496 : * there was a page here recently.
497 : */
498 0 : break;
499 : }
500 0 : rcu_read_unlock();
501 :
502 0 : return page != NULL;
503 : }
504 : EXPORT_SYMBOL(filemap_range_has_page);
505 :
506 0 : static void __filemap_fdatawait_range(struct address_space *mapping,
507 : loff_t start_byte, loff_t end_byte)
508 : {
509 0 : pgoff_t index = start_byte >> PAGE_SHIFT;
510 0 : pgoff_t end = end_byte >> PAGE_SHIFT;
511 0 : struct pagevec pvec;
512 0 : int nr_pages;
513 :
514 0 : if (end_byte < start_byte)
515 0 : return;
516 :
517 0 : pagevec_init(&pvec);
518 0 : while (index <= end) {
519 0 : unsigned i;
520 :
521 0 : nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
522 : end, PAGECACHE_TAG_WRITEBACK);
523 0 : if (!nr_pages)
524 : break;
525 :
526 0 : for (i = 0; i < nr_pages; i++) {
527 0 : struct page *page = pvec.pages[i];
528 :
529 0 : wait_on_page_writeback(page);
530 0 : ClearPageError(page);
531 : }
532 0 : pagevec_release(&pvec);
533 0 : cond_resched();
534 : }
535 : }
536 :
537 : /**
538 : * filemap_fdatawait_range - wait for writeback to complete
539 : * @mapping: address space structure to wait for
540 : * @start_byte: offset in bytes where the range starts
541 : * @end_byte: offset in bytes where the range ends (inclusive)
542 : *
543 : * Walk the list of under-writeback pages of the given address space
544 : * in the given range and wait for all of them. Check error status of
545 : * the address space and return it.
546 : *
547 : * Since the error status of the address space is cleared by this function,
548 : * callers are responsible for checking the return value and handling and/or
549 : * reporting the error.
550 : *
551 : * Return: error status of the address space.
552 : */
553 0 : int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
554 : loff_t end_byte)
555 : {
556 0 : __filemap_fdatawait_range(mapping, start_byte, end_byte);
557 0 : return filemap_check_errors(mapping);
558 : }
559 : EXPORT_SYMBOL(filemap_fdatawait_range);
560 :
561 : /**
562 : * filemap_fdatawait_range_keep_errors - wait for writeback to complete
563 : * @mapping: address space structure to wait for
564 : * @start_byte: offset in bytes where the range starts
565 : * @end_byte: offset in bytes where the range ends (inclusive)
566 : *
567 : * Walk the list of under-writeback pages of the given address space in the
568 : * given range and wait for all of them. Unlike filemap_fdatawait_range(),
569 : * this function does not clear error status of the address space.
570 : *
571 : * Use this function if callers don't handle errors themselves. Expected
572 : * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
573 : * fsfreeze(8)
574 : */
575 0 : int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
576 : loff_t start_byte, loff_t end_byte)
577 : {
578 0 : __filemap_fdatawait_range(mapping, start_byte, end_byte);
579 0 : return filemap_check_and_keep_errors(mapping);
580 : }
581 : EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
582 :
583 : /**
584 : * file_fdatawait_range - wait for writeback to complete
585 : * @file: file pointing to address space structure to wait for
586 : * @start_byte: offset in bytes where the range starts
587 : * @end_byte: offset in bytes where the range ends (inclusive)
588 : *
589 : * Walk the list of under-writeback pages of the address space that file
590 : * refers to, in the given range and wait for all of them. Check error
591 : * status of the address space vs. the file->f_wb_err cursor and return it.
592 : *
593 : * Since the error status of the file is advanced by this function,
594 : * callers are responsible for checking the return value and handling and/or
595 : * reporting the error.
596 : *
597 : * Return: error status of the address space vs. the file->f_wb_err cursor.
598 : */
599 0 : int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
600 : {
601 0 : struct address_space *mapping = file->f_mapping;
602 :
603 0 : __filemap_fdatawait_range(mapping, start_byte, end_byte);
604 0 : return file_check_and_advance_wb_err(file);
605 : }
606 : EXPORT_SYMBOL(file_fdatawait_range);
607 :
608 : /**
609 : * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
610 : * @mapping: address space structure to wait for
611 : *
612 : * Walk the list of under-writeback pages of the given address space
613 : * and wait for all of them. Unlike filemap_fdatawait(), this function
614 : * does not clear error status of the address space.
615 : *
616 : * Use this function if callers don't handle errors themselves. Expected
617 : * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
618 : * fsfreeze(8)
619 : *
620 : * Return: error status of the address space.
621 : */
622 0 : int filemap_fdatawait_keep_errors(struct address_space *mapping)
623 : {
624 0 : __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
625 0 : return filemap_check_and_keep_errors(mapping);
626 : }
627 : EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
628 :
629 : /* Returns true if writeback might be needed or already in progress. */
630 0 : static bool mapping_needs_writeback(struct address_space *mapping)
631 : {
632 0 : if (dax_mapping(mapping))
633 : return mapping->nrexceptional;
634 :
635 0 : return mapping->nrpages;
636 : }
637 :
638 : /**
639 : * filemap_write_and_wait_range - write out & wait on a file range
640 : * @mapping: the address_space for the pages
641 : * @lstart: offset in bytes where the range starts
642 : * @lend: offset in bytes where the range ends (inclusive)
643 : *
644 : * Write out and wait upon file offsets lstart->lend, inclusive.
645 : *
646 : * Note that @lend is inclusive (describes the last byte to be written) so
647 : * that this function can be used to write to the very end-of-file (end = -1).
648 : *
649 : * Return: error status of the address space.
650 : */
651 0 : int filemap_write_and_wait_range(struct address_space *mapping,
652 : loff_t lstart, loff_t lend)
653 : {
654 0 : int err = 0;
655 :
656 0 : if (mapping_needs_writeback(mapping)) {
657 0 : err = __filemap_fdatawrite_range(mapping, lstart, lend,
658 : WB_SYNC_ALL);
659 : /*
660 : * Even if the above returned error, the pages may be
661 : * written partially (e.g. -ENOSPC), so we wait for it.
662 : * But the -EIO is special case, it may indicate the worst
663 : * thing (e.g. bug) happened, so we avoid waiting for it.
664 : */
665 0 : if (err != -EIO) {
666 0 : int err2 = filemap_fdatawait_range(mapping,
667 : lstart, lend);
668 0 : if (!err)
669 0 : err = err2;
670 : } else {
671 : /* Clear any previously stored errors */
672 0 : filemap_check_errors(mapping);
673 : }
674 : } else {
675 0 : err = filemap_check_errors(mapping);
676 : }
677 0 : return err;
678 : }
679 : EXPORT_SYMBOL(filemap_write_and_wait_range);
680 :
681 0 : void __filemap_set_wb_err(struct address_space *mapping, int err)
682 : {
683 0 : errseq_t eseq = errseq_set(&mapping->wb_err, err);
684 :
685 0 : trace_filemap_set_wb_err(mapping, eseq);
686 0 : }
687 : EXPORT_SYMBOL(__filemap_set_wb_err);
688 :
689 : /**
690 : * file_check_and_advance_wb_err - report wb error (if any) that was previously
691 : * and advance wb_err to current one
692 : * @file: struct file on which the error is being reported
693 : *
694 : * When userland calls fsync (or something like nfsd does the equivalent), we
695 : * want to report any writeback errors that occurred since the last fsync (or
696 : * since the file was opened if there haven't been any).
697 : *
698 : * Grab the wb_err from the mapping. If it matches what we have in the file,
699 : * then just quickly return 0. The file is all caught up.
700 : *
701 : * If it doesn't match, then take the mapping value, set the "seen" flag in
702 : * it and try to swap it into place. If it works, or another task beat us
703 : * to it with the new value, then update the f_wb_err and return the error
704 : * portion. The error at this point must be reported via proper channels
705 : * (a'la fsync, or NFS COMMIT operation, etc.).
706 : *
707 : * While we handle mapping->wb_err with atomic operations, the f_wb_err
708 : * value is protected by the f_lock since we must ensure that it reflects
709 : * the latest value swapped in for this file descriptor.
710 : *
711 : * Return: %0 on success, negative error code otherwise.
712 : */
713 0 : int file_check_and_advance_wb_err(struct file *file)
714 : {
715 0 : int err = 0;
716 0 : errseq_t old = READ_ONCE(file->f_wb_err);
717 0 : struct address_space *mapping = file->f_mapping;
718 :
719 : /* Locklessly handle the common case where nothing has changed */
720 0 : if (errseq_check(&mapping->wb_err, old)) {
721 : /* Something changed, must use slow path */
722 0 : spin_lock(&file->f_lock);
723 0 : old = file->f_wb_err;
724 0 : err = errseq_check_and_advance(&mapping->wb_err,
725 : &file->f_wb_err);
726 0 : trace_file_check_and_advance_wb_err(file, old);
727 0 : spin_unlock(&file->f_lock);
728 : }
729 :
730 : /*
731 : * We're mostly using this function as a drop in replacement for
732 : * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
733 : * that the legacy code would have had on these flags.
734 : */
735 0 : clear_bit(AS_EIO, &mapping->flags);
736 0 : clear_bit(AS_ENOSPC, &mapping->flags);
737 0 : return err;
738 : }
739 : EXPORT_SYMBOL(file_check_and_advance_wb_err);
740 :
741 : /**
742 : * file_write_and_wait_range - write out & wait on a file range
743 : * @file: file pointing to address_space with pages
744 : * @lstart: offset in bytes where the range starts
745 : * @lend: offset in bytes where the range ends (inclusive)
746 : *
747 : * Write out and wait upon file offsets lstart->lend, inclusive.
748 : *
749 : * Note that @lend is inclusive (describes the last byte to be written) so
750 : * that this function can be used to write to the very end-of-file (end = -1).
751 : *
752 : * After writing out and waiting on the data, we check and advance the
753 : * f_wb_err cursor to the latest value, and return any errors detected there.
754 : *
755 : * Return: %0 on success, negative error code otherwise.
756 : */
757 0 : int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
758 : {
759 0 : int err = 0, err2;
760 0 : struct address_space *mapping = file->f_mapping;
761 :
762 0 : if (mapping_needs_writeback(mapping)) {
763 0 : err = __filemap_fdatawrite_range(mapping, lstart, lend,
764 : WB_SYNC_ALL);
765 : /* See comment of filemap_write_and_wait() */
766 0 : if (err != -EIO)
767 0 : __filemap_fdatawait_range(mapping, lstart, lend);
768 : }
769 0 : err2 = file_check_and_advance_wb_err(file);
770 0 : if (!err)
771 0 : err = err2;
772 0 : return err;
773 : }
774 : EXPORT_SYMBOL(file_write_and_wait_range);
775 :
776 : /**
777 : * replace_page_cache_page - replace a pagecache page with a new one
778 : * @old: page to be replaced
779 : * @new: page to replace with
780 : *
781 : * This function replaces a page in the pagecache with a new one. On
782 : * success it acquires the pagecache reference for the new page and
783 : * drops it for the old page. Both the old and new pages must be
784 : * locked. This function does not add the new page to the LRU, the
785 : * caller must do that.
786 : *
787 : * The remove + add is atomic. This function cannot fail.
788 : */
789 0 : void replace_page_cache_page(struct page *old, struct page *new)
790 : {
791 0 : struct address_space *mapping = old->mapping;
792 0 : void (*freepage)(struct page *) = mapping->a_ops->freepage;
793 0 : pgoff_t offset = old->index;
794 0 : XA_STATE(xas, &mapping->i_pages, offset);
795 0 : unsigned long flags;
796 :
797 0 : VM_BUG_ON_PAGE(!PageLocked(old), old);
798 0 : VM_BUG_ON_PAGE(!PageLocked(new), new);
799 0 : VM_BUG_ON_PAGE(new->mapping, new);
800 :
801 0 : get_page(new);
802 0 : new->mapping = mapping;
803 0 : new->index = offset;
804 :
805 0 : mem_cgroup_migrate(old, new);
806 :
807 0 : xas_lock_irqsave(&xas, flags);
808 0 : xas_store(&xas, new);
809 :
810 0 : old->mapping = NULL;
811 : /* hugetlb pages do not participate in page cache accounting. */
812 0 : if (!PageHuge(old))
813 0 : __dec_lruvec_page_state(old, NR_FILE_PAGES);
814 0 : if (!PageHuge(new))
815 0 : __inc_lruvec_page_state(new, NR_FILE_PAGES);
816 0 : if (PageSwapBacked(old))
817 0 : __dec_lruvec_page_state(old, NR_SHMEM);
818 0 : if (PageSwapBacked(new))
819 0 : __inc_lruvec_page_state(new, NR_SHMEM);
820 0 : xas_unlock_irqrestore(&xas, flags);
821 0 : if (freepage)
822 0 : freepage(old);
823 0 : put_page(old);
824 0 : }
825 : EXPORT_SYMBOL_GPL(replace_page_cache_page);
826 :
827 0 : noinline int __add_to_page_cache_locked(struct page *page,
828 : struct address_space *mapping,
829 : pgoff_t offset, gfp_t gfp,
830 : void **shadowp)
831 : {
832 0 : XA_STATE(xas, &mapping->i_pages, offset);
833 0 : int huge = PageHuge(page);
834 0 : int error;
835 0 : bool charged = false;
836 :
837 0 : VM_BUG_ON_PAGE(!PageLocked(page), page);
838 0 : VM_BUG_ON_PAGE(PageSwapBacked(page), page);
839 0 : mapping_set_update(&xas, mapping);
840 :
841 0 : get_page(page);
842 0 : page->mapping = mapping;
843 0 : page->index = offset;
844 :
845 0 : if (!huge) {
846 0 : error = mem_cgroup_charge(page, current->mm, gfp);
847 0 : if (error)
848 : goto error;
849 0 : charged = true;
850 : }
851 :
852 0 : gfp &= GFP_RECLAIM_MASK;
853 :
854 0 : do {
855 0 : unsigned int order = xa_get_order(xas.xa, xas.xa_index);
856 0 : void *entry, *old = NULL;
857 :
858 0 : if (order > thp_order(page))
859 0 : xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
860 : order, gfp);
861 0 : xas_lock_irq(&xas);
862 0 : xas_for_each_conflict(&xas, entry) {
863 0 : old = entry;
864 0 : if (!xa_is_value(entry)) {
865 0 : xas_set_err(&xas, -EEXIST);
866 0 : goto unlock;
867 : }
868 : }
869 :
870 0 : if (old) {
871 0 : if (shadowp)
872 0 : *shadowp = old;
873 : /* entry may have been split before we acquired lock */
874 0 : order = xa_get_order(xas.xa, xas.xa_index);
875 0 : if (order > thp_order(page)) {
876 0 : xas_split(&xas, old, order);
877 0 : xas_reset(&xas);
878 : }
879 : }
880 :
881 0 : xas_store(&xas, page);
882 0 : if (xas_error(&xas))
883 0 : goto unlock;
884 :
885 0 : if (old)
886 0 : mapping->nrexceptional--;
887 0 : mapping->nrpages++;
888 :
889 : /* hugetlb pages do not participate in page cache accounting */
890 0 : if (!huge)
891 0 : __inc_lruvec_page_state(page, NR_FILE_PAGES);
892 0 : unlock:
893 0 : xas_unlock_irq(&xas);
894 0 : } while (xas_nomem(&xas, gfp));
895 :
896 0 : if (xas_error(&xas)) {
897 0 : error = xas_error(&xas);
898 0 : if (charged)
899 0 : mem_cgroup_uncharge(page);
900 0 : goto error;
901 : }
902 :
903 0 : trace_mm_filemap_add_to_page_cache(page);
904 0 : return 0;
905 0 : error:
906 0 : page->mapping = NULL;
907 : /* Leave page->index set: truncation relies upon it */
908 0 : put_page(page);
909 0 : return error;
910 : }
911 : ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
912 :
913 : /**
914 : * add_to_page_cache_locked - add a locked page to the pagecache
915 : * @page: page to add
916 : * @mapping: the page's address_space
917 : * @offset: page index
918 : * @gfp_mask: page allocation mode
919 : *
920 : * This function is used to add a page to the pagecache. It must be locked.
921 : * This function does not add the page to the LRU. The caller must do that.
922 : *
923 : * Return: %0 on success, negative error code otherwise.
924 : */
925 0 : int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
926 : pgoff_t offset, gfp_t gfp_mask)
927 : {
928 0 : return __add_to_page_cache_locked(page, mapping, offset,
929 : gfp_mask, NULL);
930 : }
931 : EXPORT_SYMBOL(add_to_page_cache_locked);
932 :
933 0 : int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
934 : pgoff_t offset, gfp_t gfp_mask)
935 : {
936 0 : void *shadow = NULL;
937 0 : int ret;
938 :
939 0 : __SetPageLocked(page);
940 0 : ret = __add_to_page_cache_locked(page, mapping, offset,
941 : gfp_mask, &shadow);
942 0 : if (unlikely(ret))
943 0 : __ClearPageLocked(page);
944 : else {
945 : /*
946 : * The page might have been evicted from cache only
947 : * recently, in which case it should be activated like
948 : * any other repeatedly accessed page.
949 : * The exception is pages getting rewritten; evicting other
950 : * data from the working set, only to cache data that will
951 : * get overwritten with something else, is a waste of memory.
952 : */
953 0 : WARN_ON_ONCE(PageActive(page));
954 0 : if (!(gfp_mask & __GFP_WRITE) && shadow)
955 0 : workingset_refault(page, shadow);
956 0 : lru_cache_add(page);
957 : }
958 0 : return ret;
959 : }
960 : EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
961 :
962 : #ifdef CONFIG_NUMA
963 0 : struct page *__page_cache_alloc(gfp_t gfp)
964 : {
965 0 : int n;
966 0 : struct page *page;
967 :
968 0 : if (cpuset_do_page_mem_spread()) {
969 : unsigned int cpuset_mems_cookie;
970 : do {
971 : cpuset_mems_cookie = read_mems_allowed_begin();
972 : n = cpuset_mem_spread_node();
973 : page = __alloc_pages_node(n, gfp, 0);
974 : } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
975 :
976 : return page;
977 : }
978 0 : return alloc_pages(gfp, 0);
979 : }
980 : EXPORT_SYMBOL(__page_cache_alloc);
981 : #endif
982 :
983 : /*
984 : * In order to wait for pages to become available there must be
985 : * waitqueues associated with pages. By using a hash table of
986 : * waitqueues where the bucket discipline is to maintain all
987 : * waiters on the same queue and wake all when any of the pages
988 : * become available, and for the woken contexts to check to be
989 : * sure the appropriate page became available, this saves space
990 : * at a cost of "thundering herd" phenomena during rare hash
991 : * collisions.
992 : */
993 : #define PAGE_WAIT_TABLE_BITS 8
994 : #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
995 : static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
996 :
997 0 : static wait_queue_head_t *page_waitqueue(struct page *page)
998 : {
999 0 : return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1000 : }
1001 :
1002 0 : void __init pagecache_init(void)
1003 : {
1004 0 : int i;
1005 :
1006 0 : for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1007 0 : init_waitqueue_head(&page_wait_table[i]);
1008 :
1009 0 : page_writeback_init();
1010 0 : }
1011 :
1012 : /*
1013 : * The page wait code treats the "wait->flags" somewhat unusually, because
1014 : * we have multiple different kinds of waits, not just the usual "exclusive"
1015 : * one.
1016 : *
1017 : * We have:
1018 : *
1019 : * (a) no special bits set:
1020 : *
1021 : * We're just waiting for the bit to be released, and when a waker
1022 : * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1023 : * and remove it from the wait queue.
1024 : *
1025 : * Simple and straightforward.
1026 : *
1027 : * (b) WQ_FLAG_EXCLUSIVE:
1028 : *
1029 : * The waiter is waiting to get the lock, and only one waiter should
1030 : * be woken up to avoid any thundering herd behavior. We'll set the
1031 : * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1032 : *
1033 : * This is the traditional exclusive wait.
1034 : *
1035 : * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1036 : *
1037 : * The waiter is waiting to get the bit, and additionally wants the
1038 : * lock to be transferred to it for fair lock behavior. If the lock
1039 : * cannot be taken, we stop walking the wait queue without waking
1040 : * the waiter.
1041 : *
1042 : * This is the "fair lock handoff" case, and in addition to setting
1043 : * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1044 : * that it now has the lock.
1045 : */
1046 0 : static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1047 : {
1048 0 : unsigned int flags;
1049 0 : struct wait_page_key *key = arg;
1050 0 : struct wait_page_queue *wait_page
1051 0 : = container_of(wait, struct wait_page_queue, wait);
1052 :
1053 0 : if (!wake_page_match(wait_page, key))
1054 0 : return 0;
1055 :
1056 : /*
1057 : * If it's a lock handoff wait, we get the bit for it, and
1058 : * stop walking (and do not wake it up) if we can't.
1059 : */
1060 0 : flags = wait->flags;
1061 0 : if (flags & WQ_FLAG_EXCLUSIVE) {
1062 0 : if (test_bit(key->bit_nr, &key->page->flags))
1063 : return -1;
1064 0 : if (flags & WQ_FLAG_CUSTOM) {
1065 0 : if (test_and_set_bit(key->bit_nr, &key->page->flags))
1066 : return -1;
1067 0 : flags |= WQ_FLAG_DONE;
1068 : }
1069 : }
1070 :
1071 : /*
1072 : * We are holding the wait-queue lock, but the waiter that
1073 : * is waiting for this will be checking the flags without
1074 : * any locking.
1075 : *
1076 : * So update the flags atomically, and wake up the waiter
1077 : * afterwards to avoid any races. This store-release pairs
1078 : * with the load-acquire in wait_on_page_bit_common().
1079 : */
1080 0 : smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1081 0 : wake_up_state(wait->private, mode);
1082 :
1083 : /*
1084 : * Ok, we have successfully done what we're waiting for,
1085 : * and we can unconditionally remove the wait entry.
1086 : *
1087 : * Note that this pairs with the "finish_wait()" in the
1088 : * waiter, and has to be the absolute last thing we do.
1089 : * After this list_del_init(&wait->entry) the wait entry
1090 : * might be de-allocated and the process might even have
1091 : * exited.
1092 : */
1093 0 : list_del_init_careful(&wait->entry);
1094 0 : return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1095 : }
1096 :
1097 0 : static void wake_up_page_bit(struct page *page, int bit_nr)
1098 : {
1099 0 : wait_queue_head_t *q = page_waitqueue(page);
1100 0 : struct wait_page_key key;
1101 0 : unsigned long flags;
1102 0 : wait_queue_entry_t bookmark;
1103 :
1104 0 : key.page = page;
1105 0 : key.bit_nr = bit_nr;
1106 0 : key.page_match = 0;
1107 :
1108 0 : bookmark.flags = 0;
1109 0 : bookmark.private = NULL;
1110 0 : bookmark.func = NULL;
1111 0 : INIT_LIST_HEAD(&bookmark.entry);
1112 :
1113 0 : spin_lock_irqsave(&q->lock, flags);
1114 0 : __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1115 :
1116 0 : while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1117 : /*
1118 : * Take a breather from holding the lock,
1119 : * allow pages that finish wake up asynchronously
1120 : * to acquire the lock and remove themselves
1121 : * from wait queue
1122 : */
1123 0 : spin_unlock_irqrestore(&q->lock, flags);
1124 0 : cpu_relax();
1125 0 : spin_lock_irqsave(&q->lock, flags);
1126 0 : __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1127 : }
1128 :
1129 : /*
1130 : * It is possible for other pages to have collided on the waitqueue
1131 : * hash, so in that case check for a page match. That prevents a long-
1132 : * term waiter
1133 : *
1134 : * It is still possible to miss a case here, when we woke page waiters
1135 : * and removed them from the waitqueue, but there are still other
1136 : * page waiters.
1137 : */
1138 0 : if (!waitqueue_active(q) || !key.page_match) {
1139 0 : ClearPageWaiters(page);
1140 : /*
1141 : * It's possible to miss clearing Waiters here, when we woke
1142 : * our page waiters, but the hashed waitqueue has waiters for
1143 : * other pages on it.
1144 : *
1145 : * That's okay, it's a rare case. The next waker will clear it.
1146 : */
1147 : }
1148 0 : spin_unlock_irqrestore(&q->lock, flags);
1149 0 : }
1150 :
1151 0 : static void wake_up_page(struct page *page, int bit)
1152 : {
1153 0 : if (!PageWaiters(page))
1154 : return;
1155 0 : wake_up_page_bit(page, bit);
1156 : }
1157 :
1158 : /*
1159 : * A choice of three behaviors for wait_on_page_bit_common():
1160 : */
1161 : enum behavior {
1162 : EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1163 : * __lock_page() waiting on then setting PG_locked.
1164 : */
1165 : SHARED, /* Hold ref to page and check the bit when woken, like
1166 : * wait_on_page_writeback() waiting on PG_writeback.
1167 : */
1168 : DROP, /* Drop ref to page before wait, no check when woken,
1169 : * like put_and_wait_on_page_locked() on PG_locked.
1170 : */
1171 : };
1172 :
1173 : /*
1174 : * Attempt to check (or get) the page bit, and mark us done
1175 : * if successful.
1176 : */
1177 0 : static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1178 : struct wait_queue_entry *wait)
1179 : {
1180 0 : if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1181 0 : if (test_and_set_bit(bit_nr, &page->flags))
1182 : return false;
1183 0 : } else if (test_bit(bit_nr, &page->flags))
1184 : return false;
1185 :
1186 0 : wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1187 0 : return true;
1188 : }
1189 :
1190 : /* How many times do we accept lock stealing from under a waiter? */
1191 : int sysctl_page_lock_unfairness = 5;
1192 :
1193 0 : static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1194 : struct page *page, int bit_nr, int state, enum behavior behavior)
1195 : {
1196 0 : int unfairness = sysctl_page_lock_unfairness;
1197 0 : struct wait_page_queue wait_page;
1198 0 : wait_queue_entry_t *wait = &wait_page.wait;
1199 0 : bool thrashing = false;
1200 0 : bool delayacct = false;
1201 0 : unsigned long pflags;
1202 :
1203 0 : if (bit_nr == PG_locked &&
1204 0 : !PageUptodate(page) && PageWorkingset(page)) {
1205 0 : if (!PageSwapBacked(page)) {
1206 0 : delayacct_thrashing_start();
1207 0 : delayacct = true;
1208 : }
1209 0 : psi_memstall_enter(&pflags);
1210 : thrashing = true;
1211 : }
1212 :
1213 0 : init_wait(wait);
1214 0 : wait->func = wake_page_function;
1215 0 : wait_page.page = page;
1216 0 : wait_page.bit_nr = bit_nr;
1217 :
1218 0 : repeat:
1219 0 : wait->flags = 0;
1220 0 : if (behavior == EXCLUSIVE) {
1221 0 : wait->flags = WQ_FLAG_EXCLUSIVE;
1222 0 : if (--unfairness < 0)
1223 0 : wait->flags |= WQ_FLAG_CUSTOM;
1224 : }
1225 :
1226 : /*
1227 : * Do one last check whether we can get the
1228 : * page bit synchronously.
1229 : *
1230 : * Do the SetPageWaiters() marking before that
1231 : * to let any waker we _just_ missed know they
1232 : * need to wake us up (otherwise they'll never
1233 : * even go to the slow case that looks at the
1234 : * page queue), and add ourselves to the wait
1235 : * queue if we need to sleep.
1236 : *
1237 : * This part needs to be done under the queue
1238 : * lock to avoid races.
1239 : */
1240 0 : spin_lock_irq(&q->lock);
1241 0 : SetPageWaiters(page);
1242 0 : if (!trylock_page_bit_common(page, bit_nr, wait))
1243 0 : __add_wait_queue_entry_tail(q, wait);
1244 0 : spin_unlock_irq(&q->lock);
1245 :
1246 : /*
1247 : * From now on, all the logic will be based on
1248 : * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1249 : * see whether the page bit testing has already
1250 : * been done by the wake function.
1251 : *
1252 : * We can drop our reference to the page.
1253 : */
1254 0 : if (behavior == DROP)
1255 0 : put_page(page);
1256 :
1257 : /*
1258 : * Note that until the "finish_wait()", or until
1259 : * we see the WQ_FLAG_WOKEN flag, we need to
1260 : * be very careful with the 'wait->flags', because
1261 : * we may race with a waker that sets them.
1262 : */
1263 0 : for (;;) {
1264 0 : unsigned int flags;
1265 :
1266 0 : set_current_state(state);
1267 :
1268 : /* Loop until we've been woken or interrupted */
1269 0 : flags = smp_load_acquire(&wait->flags);
1270 0 : if (!(flags & WQ_FLAG_WOKEN)) {
1271 0 : if (signal_pending_state(state, current))
1272 : break;
1273 :
1274 0 : io_schedule();
1275 0 : continue;
1276 : }
1277 :
1278 : /* If we were non-exclusive, we're done */
1279 0 : if (behavior != EXCLUSIVE)
1280 : break;
1281 :
1282 : /* If the waker got the lock for us, we're done */
1283 0 : if (flags & WQ_FLAG_DONE)
1284 : break;
1285 :
1286 : /*
1287 : * Otherwise, if we're getting the lock, we need to
1288 : * try to get it ourselves.
1289 : *
1290 : * And if that fails, we'll have to retry this all.
1291 : */
1292 0 : if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1293 0 : goto repeat;
1294 :
1295 0 : wait->flags |= WQ_FLAG_DONE;
1296 0 : break;
1297 : }
1298 :
1299 : /*
1300 : * If a signal happened, this 'finish_wait()' may remove the last
1301 : * waiter from the wait-queues, but the PageWaiters bit will remain
1302 : * set. That's ok. The next wakeup will take care of it, and trying
1303 : * to do it here would be difficult and prone to races.
1304 : */
1305 0 : finish_wait(q, wait);
1306 :
1307 0 : if (thrashing) {
1308 0 : if (delayacct)
1309 0 : delayacct_thrashing_end();
1310 0 : psi_memstall_leave(&pflags);
1311 : }
1312 :
1313 : /*
1314 : * NOTE! The wait->flags weren't stable until we've done the
1315 : * 'finish_wait()', and we could have exited the loop above due
1316 : * to a signal, and had a wakeup event happen after the signal
1317 : * test but before the 'finish_wait()'.
1318 : *
1319 : * So only after the finish_wait() can we reliably determine
1320 : * if we got woken up or not, so we can now figure out the final
1321 : * return value based on that state without races.
1322 : *
1323 : * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1324 : * waiter, but an exclusive one requires WQ_FLAG_DONE.
1325 : */
1326 0 : if (behavior == EXCLUSIVE)
1327 0 : return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1328 :
1329 0 : return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1330 : }
1331 :
1332 0 : void wait_on_page_bit(struct page *page, int bit_nr)
1333 : {
1334 0 : wait_queue_head_t *q = page_waitqueue(page);
1335 0 : wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1336 0 : }
1337 : EXPORT_SYMBOL(wait_on_page_bit);
1338 :
1339 0 : int wait_on_page_bit_killable(struct page *page, int bit_nr)
1340 : {
1341 0 : wait_queue_head_t *q = page_waitqueue(page);
1342 0 : return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1343 : }
1344 : EXPORT_SYMBOL(wait_on_page_bit_killable);
1345 :
1346 : /**
1347 : * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1348 : * @page: The page to wait for.
1349 : * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1350 : *
1351 : * The caller should hold a reference on @page. They expect the page to
1352 : * become unlocked relatively soon, but do not wish to hold up migration
1353 : * (for example) by holding the reference while waiting for the page to
1354 : * come unlocked. After this function returns, the caller should not
1355 : * dereference @page.
1356 : *
1357 : * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1358 : */
1359 0 : int put_and_wait_on_page_locked(struct page *page, int state)
1360 : {
1361 0 : wait_queue_head_t *q;
1362 :
1363 0 : page = compound_head(page);
1364 0 : q = page_waitqueue(page);
1365 0 : return wait_on_page_bit_common(q, page, PG_locked, state, DROP);
1366 : }
1367 :
1368 : /**
1369 : * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1370 : * @page: Page defining the wait queue of interest
1371 : * @waiter: Waiter to add to the queue
1372 : *
1373 : * Add an arbitrary @waiter to the wait queue for the nominated @page.
1374 : */
1375 0 : void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1376 : {
1377 0 : wait_queue_head_t *q = page_waitqueue(page);
1378 0 : unsigned long flags;
1379 :
1380 0 : spin_lock_irqsave(&q->lock, flags);
1381 0 : __add_wait_queue_entry_tail(q, waiter);
1382 0 : SetPageWaiters(page);
1383 0 : spin_unlock_irqrestore(&q->lock, flags);
1384 0 : }
1385 : EXPORT_SYMBOL_GPL(add_page_wait_queue);
1386 :
1387 : #ifndef clear_bit_unlock_is_negative_byte
1388 :
1389 : /*
1390 : * PG_waiters is the high bit in the same byte as PG_lock.
1391 : *
1392 : * On x86 (and on many other architectures), we can clear PG_lock and
1393 : * test the sign bit at the same time. But if the architecture does
1394 : * not support that special operation, we just do this all by hand
1395 : * instead.
1396 : *
1397 : * The read of PG_waiters has to be after (or concurrently with) PG_locked
1398 : * being cleared, but a memory barrier should be unnecessary since it is
1399 : * in the same byte as PG_locked.
1400 : */
1401 : static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1402 : {
1403 : clear_bit_unlock(nr, mem);
1404 : /* smp_mb__after_atomic(); */
1405 : return test_bit(PG_waiters, mem);
1406 : }
1407 :
1408 : #endif
1409 :
1410 : /**
1411 : * unlock_page - unlock a locked page
1412 : * @page: the page
1413 : *
1414 : * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1415 : * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1416 : * mechanism between PageLocked pages and PageWriteback pages is shared.
1417 : * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1418 : *
1419 : * Note that this depends on PG_waiters being the sign bit in the byte
1420 : * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1421 : * clear the PG_locked bit and test PG_waiters at the same time fairly
1422 : * portably (architectures that do LL/SC can test any bit, while x86 can
1423 : * test the sign bit).
1424 : */
1425 0 : void unlock_page(struct page *page)
1426 : {
1427 0 : BUILD_BUG_ON(PG_waiters != 7);
1428 0 : page = compound_head(page);
1429 0 : VM_BUG_ON_PAGE(!PageLocked(page), page);
1430 0 : if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1431 0 : wake_up_page_bit(page, PG_locked);
1432 0 : }
1433 : EXPORT_SYMBOL(unlock_page);
1434 :
1435 : /**
1436 : * end_page_writeback - end writeback against a page
1437 : * @page: the page
1438 : */
1439 0 : void end_page_writeback(struct page *page)
1440 : {
1441 : /*
1442 : * TestClearPageReclaim could be used here but it is an atomic
1443 : * operation and overkill in this particular case. Failing to
1444 : * shuffle a page marked for immediate reclaim is too mild to
1445 : * justify taking an atomic operation penalty at the end of
1446 : * ever page writeback.
1447 : */
1448 0 : if (PageReclaim(page)) {
1449 0 : ClearPageReclaim(page);
1450 0 : rotate_reclaimable_page(page);
1451 : }
1452 :
1453 : /*
1454 : * Writeback does not hold a page reference of its own, relying
1455 : * on truncation to wait for the clearing of PG_writeback.
1456 : * But here we must make sure that the page is not freed and
1457 : * reused before the wake_up_page().
1458 : */
1459 0 : get_page(page);
1460 0 : if (!test_clear_page_writeback(page))
1461 0 : BUG();
1462 :
1463 0 : smp_mb__after_atomic();
1464 0 : wake_up_page(page, PG_writeback);
1465 0 : put_page(page);
1466 0 : }
1467 : EXPORT_SYMBOL(end_page_writeback);
1468 :
1469 : /*
1470 : * After completing I/O on a page, call this routine to update the page
1471 : * flags appropriately
1472 : */
1473 0 : void page_endio(struct page *page, bool is_write, int err)
1474 : {
1475 0 : if (!is_write) {
1476 0 : if (!err) {
1477 0 : SetPageUptodate(page);
1478 : } else {
1479 0 : ClearPageUptodate(page);
1480 0 : SetPageError(page);
1481 : }
1482 0 : unlock_page(page);
1483 : } else {
1484 0 : if (err) {
1485 0 : struct address_space *mapping;
1486 :
1487 0 : SetPageError(page);
1488 0 : mapping = page_mapping(page);
1489 0 : if (mapping)
1490 0 : mapping_set_error(mapping, err);
1491 : }
1492 0 : end_page_writeback(page);
1493 : }
1494 0 : }
1495 : EXPORT_SYMBOL_GPL(page_endio);
1496 :
1497 : /**
1498 : * __lock_page - get a lock on the page, assuming we need to sleep to get it
1499 : * @__page: the page to lock
1500 : */
1501 0 : void __lock_page(struct page *__page)
1502 : {
1503 0 : struct page *page = compound_head(__page);
1504 0 : wait_queue_head_t *q = page_waitqueue(page);
1505 0 : wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1506 : EXCLUSIVE);
1507 0 : }
1508 : EXPORT_SYMBOL(__lock_page);
1509 :
1510 0 : int __lock_page_killable(struct page *__page)
1511 : {
1512 0 : struct page *page = compound_head(__page);
1513 0 : wait_queue_head_t *q = page_waitqueue(page);
1514 0 : return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1515 : EXCLUSIVE);
1516 : }
1517 : EXPORT_SYMBOL_GPL(__lock_page_killable);
1518 :
1519 0 : int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1520 : {
1521 0 : struct wait_queue_head *q = page_waitqueue(page);
1522 0 : int ret = 0;
1523 :
1524 0 : wait->page = page;
1525 0 : wait->bit_nr = PG_locked;
1526 :
1527 0 : spin_lock_irq(&q->lock);
1528 0 : __add_wait_queue_entry_tail(q, &wait->wait);
1529 0 : SetPageWaiters(page);
1530 0 : ret = !trylock_page(page);
1531 : /*
1532 : * If we were successful now, we know we're still on the
1533 : * waitqueue as we're still under the lock. This means it's
1534 : * safe to remove and return success, we know the callback
1535 : * isn't going to trigger.
1536 : */
1537 0 : if (!ret)
1538 0 : __remove_wait_queue(q, &wait->wait);
1539 : else
1540 : ret = -EIOCBQUEUED;
1541 0 : spin_unlock_irq(&q->lock);
1542 0 : return ret;
1543 : }
1544 :
1545 : /*
1546 : * Return values:
1547 : * 1 - page is locked; mmap_lock is still held.
1548 : * 0 - page is not locked.
1549 : * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1550 : * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1551 : * which case mmap_lock is still held.
1552 : *
1553 : * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1554 : * with the page locked and the mmap_lock unperturbed.
1555 : */
1556 0 : int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1557 : unsigned int flags)
1558 : {
1559 0 : if (fault_flag_allow_retry_first(flags)) {
1560 : /*
1561 : * CAUTION! In this case, mmap_lock is not released
1562 : * even though return 0.
1563 : */
1564 0 : if (flags & FAULT_FLAG_RETRY_NOWAIT)
1565 : return 0;
1566 :
1567 0 : mmap_read_unlock(mm);
1568 0 : if (flags & FAULT_FLAG_KILLABLE)
1569 0 : wait_on_page_locked_killable(page);
1570 : else
1571 0 : wait_on_page_locked(page);
1572 0 : return 0;
1573 : }
1574 0 : if (flags & FAULT_FLAG_KILLABLE) {
1575 0 : int ret;
1576 :
1577 0 : ret = __lock_page_killable(page);
1578 0 : if (ret) {
1579 0 : mmap_read_unlock(mm);
1580 0 : return 0;
1581 : }
1582 : } else {
1583 0 : __lock_page(page);
1584 : }
1585 : return 1;
1586 :
1587 : }
1588 :
1589 : /**
1590 : * page_cache_next_miss() - Find the next gap in the page cache.
1591 : * @mapping: Mapping.
1592 : * @index: Index.
1593 : * @max_scan: Maximum range to search.
1594 : *
1595 : * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1596 : * gap with the lowest index.
1597 : *
1598 : * This function may be called under the rcu_read_lock. However, this will
1599 : * not atomically search a snapshot of the cache at a single point in time.
1600 : * For example, if a gap is created at index 5, then subsequently a gap is
1601 : * created at index 10, page_cache_next_miss covering both indices may
1602 : * return 10 if called under the rcu_read_lock.
1603 : *
1604 : * Return: The index of the gap if found, otherwise an index outside the
1605 : * range specified (in which case 'return - index >= max_scan' will be true).
1606 : * In the rare case of index wrap-around, 0 will be returned.
1607 : */
1608 0 : pgoff_t page_cache_next_miss(struct address_space *mapping,
1609 : pgoff_t index, unsigned long max_scan)
1610 : {
1611 0 : XA_STATE(xas, &mapping->i_pages, index);
1612 :
1613 0 : while (max_scan--) {
1614 0 : void *entry = xas_next(&xas);
1615 0 : if (!entry || xa_is_value(entry))
1616 : break;
1617 0 : if (xas.xa_index == 0)
1618 : break;
1619 : }
1620 :
1621 0 : return xas.xa_index;
1622 : }
1623 : EXPORT_SYMBOL(page_cache_next_miss);
1624 :
1625 : /**
1626 : * page_cache_prev_miss() - Find the previous gap in the page cache.
1627 : * @mapping: Mapping.
1628 : * @index: Index.
1629 : * @max_scan: Maximum range to search.
1630 : *
1631 : * Search the range [max(index - max_scan + 1, 0), index] for the
1632 : * gap with the highest index.
1633 : *
1634 : * This function may be called under the rcu_read_lock. However, this will
1635 : * not atomically search a snapshot of the cache at a single point in time.
1636 : * For example, if a gap is created at index 10, then subsequently a gap is
1637 : * created at index 5, page_cache_prev_miss() covering both indices may
1638 : * return 5 if called under the rcu_read_lock.
1639 : *
1640 : * Return: The index of the gap if found, otherwise an index outside the
1641 : * range specified (in which case 'index - return >= max_scan' will be true).
1642 : * In the rare case of wrap-around, ULONG_MAX will be returned.
1643 : */
1644 0 : pgoff_t page_cache_prev_miss(struct address_space *mapping,
1645 : pgoff_t index, unsigned long max_scan)
1646 : {
1647 0 : XA_STATE(xas, &mapping->i_pages, index);
1648 :
1649 0 : while (max_scan--) {
1650 0 : void *entry = xas_prev(&xas);
1651 0 : if (!entry || xa_is_value(entry))
1652 : break;
1653 0 : if (xas.xa_index == ULONG_MAX)
1654 : break;
1655 : }
1656 :
1657 0 : return xas.xa_index;
1658 : }
1659 : EXPORT_SYMBOL(page_cache_prev_miss);
1660 :
1661 : /*
1662 : * mapping_get_entry - Get a page cache entry.
1663 : * @mapping: the address_space to search
1664 : * @index: The page cache index.
1665 : *
1666 : * Looks up the page cache slot at @mapping & @offset. If there is a
1667 : * page cache page, the head page is returned with an increased refcount.
1668 : *
1669 : * If the slot holds a shadow entry of a previously evicted page, or a
1670 : * swap entry from shmem/tmpfs, it is returned.
1671 : *
1672 : * Return: The head page or shadow entry, %NULL if nothing is found.
1673 : */
1674 0 : static struct page *mapping_get_entry(struct address_space *mapping,
1675 : pgoff_t index)
1676 : {
1677 0 : XA_STATE(xas, &mapping->i_pages, index);
1678 0 : struct page *page;
1679 :
1680 0 : rcu_read_lock();
1681 : repeat:
1682 0 : xas_reset(&xas);
1683 0 : page = xas_load(&xas);
1684 0 : if (xas_retry(&xas, page))
1685 0 : goto repeat;
1686 : /*
1687 : * A shadow entry of a recently evicted page, or a swap entry from
1688 : * shmem/tmpfs. Return it without attempting to raise page count.
1689 : */
1690 0 : if (!page || xa_is_value(page))
1691 0 : goto out;
1692 :
1693 0 : if (!page_cache_get_speculative(page))
1694 0 : goto repeat;
1695 :
1696 : /*
1697 : * Has the page moved or been split?
1698 : * This is part of the lockless pagecache protocol. See
1699 : * include/linux/pagemap.h for details.
1700 : */
1701 0 : if (unlikely(page != xas_reload(&xas))) {
1702 0 : put_page(page);
1703 0 : goto repeat;
1704 : }
1705 0 : out:
1706 0 : rcu_read_unlock();
1707 :
1708 0 : return page;
1709 : }
1710 :
1711 : /**
1712 : * pagecache_get_page - Find and get a reference to a page.
1713 : * @mapping: The address_space to search.
1714 : * @index: The page index.
1715 : * @fgp_flags: %FGP flags modify how the page is returned.
1716 : * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1717 : *
1718 : * Looks up the page cache entry at @mapping & @index.
1719 : *
1720 : * @fgp_flags can be zero or more of these flags:
1721 : *
1722 : * * %FGP_ACCESSED - The page will be marked accessed.
1723 : * * %FGP_LOCK - The page is returned locked.
1724 : * * %FGP_HEAD - If the page is present and a THP, return the head page
1725 : * rather than the exact page specified by the index.
1726 : * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1727 : * instead of allocating a new page to replace it.
1728 : * * %FGP_CREAT - If no page is present then a new page is allocated using
1729 : * @gfp_mask and added to the page cache and the VM's LRU list.
1730 : * The page is returned locked and with an increased refcount.
1731 : * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1732 : * page is already in cache. If the page was allocated, unlock it before
1733 : * returning so the caller can do the same dance.
1734 : * * %FGP_WRITE - The page will be written
1735 : * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1736 : * * %FGP_NOWAIT - Don't get blocked by page lock
1737 : *
1738 : * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1739 : * if the %GFP flags specified for %FGP_CREAT are atomic.
1740 : *
1741 : * If there is a page cache page, it is returned with an increased refcount.
1742 : *
1743 : * Return: The found page or %NULL otherwise.
1744 : */
1745 0 : struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1746 : int fgp_flags, gfp_t gfp_mask)
1747 : {
1748 0 : struct page *page;
1749 :
1750 : repeat:
1751 0 : page = mapping_get_entry(mapping, index);
1752 0 : if (xa_is_value(page)) {
1753 0 : if (fgp_flags & FGP_ENTRY)
1754 0 : return page;
1755 : page = NULL;
1756 : }
1757 0 : if (!page)
1758 0 : goto no_page;
1759 :
1760 0 : if (fgp_flags & FGP_LOCK) {
1761 0 : if (fgp_flags & FGP_NOWAIT) {
1762 0 : if (!trylock_page(page)) {
1763 0 : put_page(page);
1764 0 : return NULL;
1765 : }
1766 : } else {
1767 0 : lock_page(page);
1768 : }
1769 :
1770 : /* Has the page been truncated? */
1771 0 : if (unlikely(page->mapping != mapping)) {
1772 0 : unlock_page(page);
1773 0 : put_page(page);
1774 0 : goto repeat;
1775 : }
1776 0 : VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1777 : }
1778 :
1779 0 : if (fgp_flags & FGP_ACCESSED)
1780 0 : mark_page_accessed(page);
1781 : else if (fgp_flags & FGP_WRITE) {
1782 : /* Clear idle flag for buffer write */
1783 0 : if (page_is_idle(page))
1784 0 : clear_page_idle(page);
1785 : }
1786 0 : if (!(fgp_flags & FGP_HEAD))
1787 0 : page = find_subpage(page, index);
1788 :
1789 0 : no_page:
1790 0 : if (!page && (fgp_flags & FGP_CREAT)) {
1791 0 : int err;
1792 0 : if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1793 0 : gfp_mask |= __GFP_WRITE;
1794 0 : if (fgp_flags & FGP_NOFS)
1795 0 : gfp_mask &= ~__GFP_FS;
1796 :
1797 0 : page = __page_cache_alloc(gfp_mask);
1798 0 : if (!page)
1799 : return NULL;
1800 :
1801 0 : if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1802 0 : fgp_flags |= FGP_LOCK;
1803 :
1804 : /* Init accessed so avoid atomic mark_page_accessed later */
1805 0 : if (fgp_flags & FGP_ACCESSED)
1806 0 : __SetPageReferenced(page);
1807 :
1808 0 : err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1809 0 : if (unlikely(err)) {
1810 0 : put_page(page);
1811 0 : page = NULL;
1812 0 : if (err == -EEXIST)
1813 0 : goto repeat;
1814 : }
1815 :
1816 : /*
1817 : * add_to_page_cache_lru locks the page, and for mmap we expect
1818 : * an unlocked page.
1819 : */
1820 0 : if (page && (fgp_flags & FGP_FOR_MMAP))
1821 0 : unlock_page(page);
1822 : }
1823 :
1824 : return page;
1825 : }
1826 : EXPORT_SYMBOL(pagecache_get_page);
1827 :
1828 0 : static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1829 : xa_mark_t mark)
1830 : {
1831 0 : struct page *page;
1832 :
1833 : retry:
1834 0 : if (mark == XA_PRESENT)
1835 0 : page = xas_find(xas, max);
1836 : else
1837 0 : page = xas_find_marked(xas, max, mark);
1838 :
1839 0 : if (xas_retry(xas, page))
1840 0 : goto retry;
1841 : /*
1842 : * A shadow entry of a recently evicted page, a swap
1843 : * entry from shmem/tmpfs or a DAX entry. Return it
1844 : * without attempting to raise page count.
1845 : */
1846 0 : if (!page || xa_is_value(page))
1847 0 : return page;
1848 :
1849 0 : if (!page_cache_get_speculative(page))
1850 0 : goto reset;
1851 :
1852 : /* Has the page moved or been split? */
1853 0 : if (unlikely(page != xas_reload(xas))) {
1854 0 : put_page(page);
1855 0 : goto reset;
1856 : }
1857 :
1858 : return page;
1859 0 : reset:
1860 0 : xas_reset(xas);
1861 0 : goto retry;
1862 : }
1863 :
1864 : /**
1865 : * find_get_entries - gang pagecache lookup
1866 : * @mapping: The address_space to search
1867 : * @start: The starting page cache index
1868 : * @end: The final page index (inclusive).
1869 : * @pvec: Where the resulting entries are placed.
1870 : * @indices: The cache indices corresponding to the entries in @entries
1871 : *
1872 : * find_get_entries() will search for and return a batch of entries in
1873 : * the mapping. The entries are placed in @pvec. find_get_entries()
1874 : * takes a reference on any actual pages it returns.
1875 : *
1876 : * The search returns a group of mapping-contiguous page cache entries
1877 : * with ascending indexes. There may be holes in the indices due to
1878 : * not-present pages.
1879 : *
1880 : * Any shadow entries of evicted pages, or swap entries from
1881 : * shmem/tmpfs, are included in the returned array.
1882 : *
1883 : * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1884 : * stops at that page: the caller is likely to have a better way to handle
1885 : * the compound page as a whole, and then skip its extent, than repeatedly
1886 : * calling find_get_entries() to return all its tails.
1887 : *
1888 : * Return: the number of pages and shadow entries which were found.
1889 : */
1890 0 : unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
1891 : pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
1892 : {
1893 0 : XA_STATE(xas, &mapping->i_pages, start);
1894 0 : struct page *page;
1895 0 : unsigned int ret = 0;
1896 0 : unsigned nr_entries = PAGEVEC_SIZE;
1897 :
1898 0 : rcu_read_lock();
1899 0 : while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
1900 : /*
1901 : * Terminate early on finding a THP, to allow the caller to
1902 : * handle it all at once; but continue if this is hugetlbfs.
1903 : */
1904 0 : if (!xa_is_value(page) && PageTransHuge(page) &&
1905 0 : !PageHuge(page)) {
1906 0 : page = find_subpage(page, xas.xa_index);
1907 0 : nr_entries = ret + 1;
1908 : }
1909 :
1910 0 : indices[ret] = xas.xa_index;
1911 0 : pvec->pages[ret] = page;
1912 0 : if (++ret == nr_entries)
1913 : break;
1914 : }
1915 0 : rcu_read_unlock();
1916 :
1917 0 : pvec->nr = ret;
1918 0 : return ret;
1919 : }
1920 :
1921 : /**
1922 : * find_lock_entries - Find a batch of pagecache entries.
1923 : * @mapping: The address_space to search.
1924 : * @start: The starting page cache index.
1925 : * @end: The final page index (inclusive).
1926 : * @pvec: Where the resulting entries are placed.
1927 : * @indices: The cache indices of the entries in @pvec.
1928 : *
1929 : * find_lock_entries() will return a batch of entries from @mapping.
1930 : * Swap, shadow and DAX entries are included. Pages are returned
1931 : * locked and with an incremented refcount. Pages which are locked by
1932 : * somebody else or under writeback are skipped. Only the head page of
1933 : * a THP is returned. Pages which are partially outside the range are
1934 : * not returned.
1935 : *
1936 : * The entries have ascending indexes. The indices may not be consecutive
1937 : * due to not-present entries, THP pages, pages which could not be locked
1938 : * or pages under writeback.
1939 : *
1940 : * Return: The number of entries which were found.
1941 : */
1942 0 : unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
1943 : pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
1944 : {
1945 0 : XA_STATE(xas, &mapping->i_pages, start);
1946 0 : struct page *page;
1947 :
1948 0 : rcu_read_lock();
1949 0 : while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
1950 0 : if (!xa_is_value(page)) {
1951 0 : if (page->index < start)
1952 0 : goto put;
1953 0 : VM_BUG_ON_PAGE(page->index != xas.xa_index, page);
1954 0 : if (page->index + thp_nr_pages(page) - 1 > end)
1955 0 : goto put;
1956 0 : if (!trylock_page(page))
1957 0 : goto put;
1958 0 : if (page->mapping != mapping || PageWriteback(page))
1959 0 : goto unlock;
1960 0 : VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
1961 : page);
1962 : }
1963 0 : indices[pvec->nr] = xas.xa_index;
1964 0 : if (!pagevec_add(pvec, page))
1965 : break;
1966 0 : goto next;
1967 0 : unlock:
1968 0 : unlock_page(page);
1969 0 : put:
1970 0 : put_page(page);
1971 0 : next:
1972 0 : if (!xa_is_value(page) && PageTransHuge(page))
1973 0 : xas_set(&xas, page->index + thp_nr_pages(page));
1974 : }
1975 0 : rcu_read_unlock();
1976 :
1977 0 : return pagevec_count(pvec);
1978 : }
1979 :
1980 : /**
1981 : * find_get_pages_range - gang pagecache lookup
1982 : * @mapping: The address_space to search
1983 : * @start: The starting page index
1984 : * @end: The final page index (inclusive)
1985 : * @nr_pages: The maximum number of pages
1986 : * @pages: Where the resulting pages are placed
1987 : *
1988 : * find_get_pages_range() will search for and return a group of up to @nr_pages
1989 : * pages in the mapping starting at index @start and up to index @end
1990 : * (inclusive). The pages are placed at @pages. find_get_pages_range() takes
1991 : * a reference against the returned pages.
1992 : *
1993 : * The search returns a group of mapping-contiguous pages with ascending
1994 : * indexes. There may be holes in the indices due to not-present pages.
1995 : * We also update @start to index the next page for the traversal.
1996 : *
1997 : * Return: the number of pages which were found. If this number is
1998 : * smaller than @nr_pages, the end of specified range has been
1999 : * reached.
2000 : */
2001 0 : unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2002 : pgoff_t end, unsigned int nr_pages,
2003 : struct page **pages)
2004 : {
2005 0 : XA_STATE(xas, &mapping->i_pages, *start);
2006 0 : struct page *page;
2007 0 : unsigned ret = 0;
2008 :
2009 0 : if (unlikely(!nr_pages))
2010 : return 0;
2011 :
2012 0 : rcu_read_lock();
2013 0 : while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2014 : /* Skip over shadow, swap and DAX entries */
2015 0 : if (xa_is_value(page))
2016 0 : continue;
2017 :
2018 0 : pages[ret] = find_subpage(page, xas.xa_index);
2019 0 : if (++ret == nr_pages) {
2020 0 : *start = xas.xa_index + 1;
2021 0 : goto out;
2022 : }
2023 : }
2024 :
2025 : /*
2026 : * We come here when there is no page beyond @end. We take care to not
2027 : * overflow the index @start as it confuses some of the callers. This
2028 : * breaks the iteration when there is a page at index -1 but that is
2029 : * already broken anyway.
2030 : */
2031 0 : if (end == (pgoff_t)-1)
2032 0 : *start = (pgoff_t)-1;
2033 : else
2034 0 : *start = end + 1;
2035 0 : out:
2036 0 : rcu_read_unlock();
2037 :
2038 0 : return ret;
2039 : }
2040 :
2041 : /**
2042 : * find_get_pages_contig - gang contiguous pagecache lookup
2043 : * @mapping: The address_space to search
2044 : * @index: The starting page index
2045 : * @nr_pages: The maximum number of pages
2046 : * @pages: Where the resulting pages are placed
2047 : *
2048 : * find_get_pages_contig() works exactly like find_get_pages(), except
2049 : * that the returned number of pages are guaranteed to be contiguous.
2050 : *
2051 : * Return: the number of pages which were found.
2052 : */
2053 0 : unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2054 : unsigned int nr_pages, struct page **pages)
2055 : {
2056 0 : XA_STATE(xas, &mapping->i_pages, index);
2057 0 : struct page *page;
2058 0 : unsigned int ret = 0;
2059 :
2060 0 : if (unlikely(!nr_pages))
2061 : return 0;
2062 :
2063 0 : rcu_read_lock();
2064 0 : for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2065 0 : if (xas_retry(&xas, page))
2066 0 : continue;
2067 : /*
2068 : * If the entry has been swapped out, we can stop looking.
2069 : * No current caller is looking for DAX entries.
2070 : */
2071 0 : if (xa_is_value(page))
2072 : break;
2073 :
2074 0 : if (!page_cache_get_speculative(page))
2075 0 : goto retry;
2076 :
2077 : /* Has the page moved or been split? */
2078 0 : if (unlikely(page != xas_reload(&xas)))
2079 0 : goto put_page;
2080 :
2081 0 : pages[ret] = find_subpage(page, xas.xa_index);
2082 0 : if (++ret == nr_pages)
2083 : break;
2084 0 : continue;
2085 0 : put_page:
2086 0 : put_page(page);
2087 0 : retry:
2088 0 : xas_reset(&xas);
2089 : }
2090 0 : rcu_read_unlock();
2091 0 : return ret;
2092 : }
2093 : EXPORT_SYMBOL(find_get_pages_contig);
2094 :
2095 : /**
2096 : * find_get_pages_range_tag - Find and return head pages matching @tag.
2097 : * @mapping: the address_space to search
2098 : * @index: the starting page index
2099 : * @end: The final page index (inclusive)
2100 : * @tag: the tag index
2101 : * @nr_pages: the maximum number of pages
2102 : * @pages: where the resulting pages are placed
2103 : *
2104 : * Like find_get_pages(), except we only return head pages which are tagged
2105 : * with @tag. @index is updated to the index immediately after the last
2106 : * page we return, ready for the next iteration.
2107 : *
2108 : * Return: the number of pages which were found.
2109 : */
2110 0 : unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2111 : pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2112 : struct page **pages)
2113 : {
2114 0 : XA_STATE(xas, &mapping->i_pages, *index);
2115 0 : struct page *page;
2116 0 : unsigned ret = 0;
2117 :
2118 0 : if (unlikely(!nr_pages))
2119 : return 0;
2120 :
2121 0 : rcu_read_lock();
2122 0 : while ((page = find_get_entry(&xas, end, tag))) {
2123 : /*
2124 : * Shadow entries should never be tagged, but this iteration
2125 : * is lockless so there is a window for page reclaim to evict
2126 : * a page we saw tagged. Skip over it.
2127 : */
2128 0 : if (xa_is_value(page))
2129 0 : continue;
2130 :
2131 0 : pages[ret] = page;
2132 0 : if (++ret == nr_pages) {
2133 0 : *index = page->index + thp_nr_pages(page);
2134 0 : goto out;
2135 : }
2136 : }
2137 :
2138 : /*
2139 : * We come here when we got to @end. We take care to not overflow the
2140 : * index @index as it confuses some of the callers. This breaks the
2141 : * iteration when there is a page at index -1 but that is already
2142 : * broken anyway.
2143 : */
2144 0 : if (end == (pgoff_t)-1)
2145 0 : *index = (pgoff_t)-1;
2146 : else
2147 0 : *index = end + 1;
2148 0 : out:
2149 0 : rcu_read_unlock();
2150 :
2151 0 : return ret;
2152 : }
2153 : EXPORT_SYMBOL(find_get_pages_range_tag);
2154 :
2155 : /*
2156 : * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2157 : * a _large_ part of the i/o request. Imagine the worst scenario:
2158 : *
2159 : * ---R__________________________________________B__________
2160 : * ^ reading here ^ bad block(assume 4k)
2161 : *
2162 : * read(R) => miss => readahead(R...B) => media error => frustrating retries
2163 : * => failing the whole request => read(R) => read(R+1) =>
2164 : * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2165 : * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2166 : * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2167 : *
2168 : * It is going insane. Fix it by quickly scaling down the readahead size.
2169 : */
2170 0 : static void shrink_readahead_size_eio(struct file_ra_state *ra)
2171 : {
2172 0 : ra->ra_pages /= 4;
2173 : }
2174 :
2175 : /*
2176 : * filemap_get_read_batch - Get a batch of pages for read
2177 : *
2178 : * Get a batch of pages which represent a contiguous range of bytes
2179 : * in the file. No tail pages will be returned. If @index is in the
2180 : * middle of a THP, the entire THP will be returned. The last page in
2181 : * the batch may have Readahead set or be not Uptodate so that the
2182 : * caller can take the appropriate action.
2183 : */
2184 0 : static void filemap_get_read_batch(struct address_space *mapping,
2185 : pgoff_t index, pgoff_t max, struct pagevec *pvec)
2186 : {
2187 0 : XA_STATE(xas, &mapping->i_pages, index);
2188 0 : struct page *head;
2189 :
2190 0 : rcu_read_lock();
2191 0 : for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2192 0 : if (xas_retry(&xas, head))
2193 0 : continue;
2194 0 : if (xas.xa_index > max || xa_is_value(head))
2195 : break;
2196 0 : if (!page_cache_get_speculative(head))
2197 0 : goto retry;
2198 :
2199 : /* Has the page moved or been split? */
2200 0 : if (unlikely(head != xas_reload(&xas)))
2201 0 : goto put_page;
2202 :
2203 0 : if (!pagevec_add(pvec, head))
2204 : break;
2205 0 : if (!PageUptodate(head))
2206 : break;
2207 0 : if (PageReadahead(head))
2208 : break;
2209 0 : xas.xa_index = head->index + thp_nr_pages(head) - 1;
2210 0 : xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2211 0 : continue;
2212 0 : put_page:
2213 0 : put_page(head);
2214 0 : retry:
2215 0 : xas_reset(&xas);
2216 : }
2217 0 : rcu_read_unlock();
2218 0 : }
2219 :
2220 0 : static int filemap_read_page(struct file *file, struct address_space *mapping,
2221 : struct page *page)
2222 : {
2223 0 : int error;
2224 :
2225 : /*
2226 : * A previous I/O error may have been due to temporary failures,
2227 : * eg. multipath errors. PG_error will be set again if readpage
2228 : * fails.
2229 : */
2230 0 : ClearPageError(page);
2231 : /* Start the actual read. The read will unlock the page. */
2232 0 : error = mapping->a_ops->readpage(file, page);
2233 0 : if (error)
2234 : return error;
2235 :
2236 0 : error = wait_on_page_locked_killable(page);
2237 0 : if (error)
2238 : return error;
2239 0 : if (PageUptodate(page))
2240 : return 0;
2241 0 : if (!page->mapping) /* page truncated */
2242 : return AOP_TRUNCATED_PAGE;
2243 0 : shrink_readahead_size_eio(&file->f_ra);
2244 0 : return -EIO;
2245 : }
2246 :
2247 0 : static bool filemap_range_uptodate(struct address_space *mapping,
2248 : loff_t pos, struct iov_iter *iter, struct page *page)
2249 : {
2250 0 : int count;
2251 :
2252 0 : if (PageUptodate(page))
2253 : return true;
2254 : /* pipes can't handle partially uptodate pages */
2255 0 : if (iov_iter_is_pipe(iter))
2256 : return false;
2257 0 : if (!mapping->a_ops->is_partially_uptodate)
2258 : return false;
2259 0 : if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2260 : return false;
2261 :
2262 0 : count = iter->count;
2263 0 : if (page_offset(page) > pos) {
2264 0 : count -= page_offset(page) - pos;
2265 0 : pos = 0;
2266 : } else {
2267 0 : pos -= page_offset(page);
2268 : }
2269 :
2270 0 : return mapping->a_ops->is_partially_uptodate(page, pos, count);
2271 : }
2272 :
2273 0 : static int filemap_update_page(struct kiocb *iocb,
2274 : struct address_space *mapping, struct iov_iter *iter,
2275 : struct page *page)
2276 : {
2277 0 : int error;
2278 :
2279 0 : if (!trylock_page(page)) {
2280 0 : if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2281 : return -EAGAIN;
2282 0 : if (!(iocb->ki_flags & IOCB_WAITQ)) {
2283 0 : put_and_wait_on_page_locked(page, TASK_KILLABLE);
2284 0 : return AOP_TRUNCATED_PAGE;
2285 : }
2286 0 : error = __lock_page_async(page, iocb->ki_waitq);
2287 0 : if (error)
2288 : return error;
2289 : }
2290 :
2291 0 : if (!page->mapping)
2292 0 : goto truncated;
2293 :
2294 0 : error = 0;
2295 0 : if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
2296 0 : goto unlock;
2297 :
2298 0 : error = -EAGAIN;
2299 0 : if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2300 0 : goto unlock;
2301 :
2302 0 : error = filemap_read_page(iocb->ki_filp, mapping, page);
2303 0 : if (error == AOP_TRUNCATED_PAGE)
2304 0 : put_page(page);
2305 : return error;
2306 0 : truncated:
2307 0 : unlock_page(page);
2308 0 : put_page(page);
2309 0 : return AOP_TRUNCATED_PAGE;
2310 0 : unlock:
2311 0 : unlock_page(page);
2312 0 : return error;
2313 : }
2314 :
2315 0 : static int filemap_create_page(struct file *file,
2316 : struct address_space *mapping, pgoff_t index,
2317 : struct pagevec *pvec)
2318 : {
2319 0 : struct page *page;
2320 0 : int error;
2321 :
2322 0 : page = page_cache_alloc(mapping);
2323 0 : if (!page)
2324 : return -ENOMEM;
2325 :
2326 0 : error = add_to_page_cache_lru(page, mapping, index,
2327 : mapping_gfp_constraint(mapping, GFP_KERNEL));
2328 0 : if (error == -EEXIST)
2329 : error = AOP_TRUNCATED_PAGE;
2330 0 : if (error)
2331 0 : goto error;
2332 :
2333 0 : error = filemap_read_page(file, mapping, page);
2334 0 : if (error)
2335 0 : goto error;
2336 :
2337 0 : pagevec_add(pvec, page);
2338 0 : return 0;
2339 0 : error:
2340 0 : put_page(page);
2341 0 : return error;
2342 : }
2343 :
2344 0 : static int filemap_readahead(struct kiocb *iocb, struct file *file,
2345 : struct address_space *mapping, struct page *page,
2346 : pgoff_t last_index)
2347 : {
2348 0 : if (iocb->ki_flags & IOCB_NOIO)
2349 : return -EAGAIN;
2350 0 : page_cache_async_readahead(mapping, &file->f_ra, file, page,
2351 0 : page->index, last_index - page->index);
2352 0 : return 0;
2353 : }
2354 :
2355 0 : static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2356 : struct pagevec *pvec)
2357 : {
2358 0 : struct file *filp = iocb->ki_filp;
2359 0 : struct address_space *mapping = filp->f_mapping;
2360 0 : struct file_ra_state *ra = &filp->f_ra;
2361 0 : pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2362 0 : pgoff_t last_index;
2363 0 : struct page *page;
2364 0 : int err = 0;
2365 :
2366 0 : last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2367 : retry:
2368 0 : if (fatal_signal_pending(current))
2369 : return -EINTR;
2370 :
2371 0 : filemap_get_read_batch(mapping, index, last_index, pvec);
2372 0 : if (!pagevec_count(pvec)) {
2373 0 : if (iocb->ki_flags & IOCB_NOIO)
2374 : return -EAGAIN;
2375 0 : page_cache_sync_readahead(mapping, ra, filp, index,
2376 : last_index - index);
2377 0 : filemap_get_read_batch(mapping, index, last_index, pvec);
2378 : }
2379 0 : if (!pagevec_count(pvec)) {
2380 0 : if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2381 : return -EAGAIN;
2382 0 : err = filemap_create_page(filp, mapping,
2383 0 : iocb->ki_pos >> PAGE_SHIFT, pvec);
2384 0 : if (err == AOP_TRUNCATED_PAGE)
2385 0 : goto retry;
2386 0 : return err;
2387 : }
2388 :
2389 0 : page = pvec->pages[pagevec_count(pvec) - 1];
2390 0 : if (PageReadahead(page)) {
2391 0 : err = filemap_readahead(iocb, filp, mapping, page, last_index);
2392 0 : if (err)
2393 0 : goto err;
2394 : }
2395 0 : if (!PageUptodate(page)) {
2396 0 : if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2397 0 : iocb->ki_flags |= IOCB_NOWAIT;
2398 0 : err = filemap_update_page(iocb, mapping, iter, page);
2399 0 : if (err)
2400 0 : goto err;
2401 : }
2402 :
2403 : return 0;
2404 0 : err:
2405 0 : if (err < 0)
2406 0 : put_page(page);
2407 0 : if (likely(--pvec->nr))
2408 : return 0;
2409 0 : if (err == AOP_TRUNCATED_PAGE)
2410 0 : goto retry;
2411 : return err;
2412 : }
2413 :
2414 : /**
2415 : * filemap_read - Read data from the page cache.
2416 : * @iocb: The iocb to read.
2417 : * @iter: Destination for the data.
2418 : * @already_read: Number of bytes already read by the caller.
2419 : *
2420 : * Copies data from the page cache. If the data is not currently present,
2421 : * uses the readahead and readpage address_space operations to fetch it.
2422 : *
2423 : * Return: Total number of bytes copied, including those already read by
2424 : * the caller. If an error happens before any bytes are copied, returns
2425 : * a negative error number.
2426 : */
2427 0 : ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2428 : ssize_t already_read)
2429 : {
2430 0 : struct file *filp = iocb->ki_filp;
2431 0 : struct file_ra_state *ra = &filp->f_ra;
2432 0 : struct address_space *mapping = filp->f_mapping;
2433 0 : struct inode *inode = mapping->host;
2434 0 : struct pagevec pvec;
2435 0 : int i, error = 0;
2436 0 : bool writably_mapped;
2437 0 : loff_t isize, end_offset;
2438 :
2439 0 : if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2440 : return 0;
2441 0 : if (unlikely(!iov_iter_count(iter)))
2442 : return 0;
2443 :
2444 0 : iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2445 0 : pagevec_init(&pvec);
2446 :
2447 0 : do {
2448 0 : cond_resched();
2449 :
2450 : /*
2451 : * If we've already successfully copied some data, then we
2452 : * can no longer safely return -EIOCBQUEUED. Hence mark
2453 : * an async read NOWAIT at that point.
2454 : */
2455 0 : if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2456 0 : iocb->ki_flags |= IOCB_NOWAIT;
2457 :
2458 0 : error = filemap_get_pages(iocb, iter, &pvec);
2459 0 : if (error < 0)
2460 : break;
2461 :
2462 : /*
2463 : * i_size must be checked after we know the pages are Uptodate.
2464 : *
2465 : * Checking i_size after the check allows us to calculate
2466 : * the correct value for "nr", which means the zero-filled
2467 : * part of the page is not copied back to userspace (unless
2468 : * another truncate extends the file - this is desired though).
2469 : */
2470 0 : isize = i_size_read(inode);
2471 0 : if (unlikely(iocb->ki_pos >= isize))
2472 0 : goto put_pages;
2473 0 : end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2474 :
2475 : /*
2476 : * Once we start copying data, we don't want to be touching any
2477 : * cachelines that might be contended:
2478 : */
2479 0 : writably_mapped = mapping_writably_mapped(mapping);
2480 :
2481 : /*
2482 : * When a sequential read accesses a page several times, only
2483 : * mark it as accessed the first time.
2484 : */
2485 0 : if (iocb->ki_pos >> PAGE_SHIFT !=
2486 0 : ra->prev_pos >> PAGE_SHIFT)
2487 0 : mark_page_accessed(pvec.pages[0]);
2488 :
2489 0 : for (i = 0; i < pagevec_count(&pvec); i++) {
2490 0 : struct page *page = pvec.pages[i];
2491 0 : size_t page_size = thp_size(page);
2492 0 : size_t offset = iocb->ki_pos & (page_size - 1);
2493 0 : size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2494 : page_size - offset);
2495 0 : size_t copied;
2496 :
2497 0 : if (end_offset < page_offset(page))
2498 : break;
2499 0 : if (i > 0)
2500 0 : mark_page_accessed(page);
2501 : /*
2502 : * If users can be writing to this page using arbitrary
2503 : * virtual addresses, take care about potential aliasing
2504 : * before reading the page on the kernel side.
2505 : */
2506 0 : if (writably_mapped) {
2507 : int j;
2508 :
2509 0 : for (j = 0; j < thp_nr_pages(page); j++)
2510 0 : flush_dcache_page(page + j);
2511 : }
2512 :
2513 0 : copied = copy_page_to_iter(page, offset, bytes, iter);
2514 :
2515 0 : already_read += copied;
2516 0 : iocb->ki_pos += copied;
2517 0 : ra->prev_pos = iocb->ki_pos;
2518 :
2519 0 : if (copied < bytes) {
2520 : error = -EFAULT;
2521 : break;
2522 : }
2523 : }
2524 0 : put_pages:
2525 0 : for (i = 0; i < pagevec_count(&pvec); i++)
2526 0 : put_page(pvec.pages[i]);
2527 0 : pagevec_reinit(&pvec);
2528 0 : } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2529 :
2530 0 : file_accessed(filp);
2531 :
2532 0 : return already_read ? already_read : error;
2533 : }
2534 : EXPORT_SYMBOL_GPL(filemap_read);
2535 :
2536 : /**
2537 : * generic_file_read_iter - generic filesystem read routine
2538 : * @iocb: kernel I/O control block
2539 : * @iter: destination for the data read
2540 : *
2541 : * This is the "read_iter()" routine for all filesystems
2542 : * that can use the page cache directly.
2543 : *
2544 : * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2545 : * be returned when no data can be read without waiting for I/O requests
2546 : * to complete; it doesn't prevent readahead.
2547 : *
2548 : * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2549 : * requests shall be made for the read or for readahead. When no data
2550 : * can be read, -EAGAIN shall be returned. When readahead would be
2551 : * triggered, a partial, possibly empty read shall be returned.
2552 : *
2553 : * Return:
2554 : * * number of bytes copied, even for partial reads
2555 : * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2556 : */
2557 : ssize_t
2558 0 : generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2559 : {
2560 0 : size_t count = iov_iter_count(iter);
2561 0 : ssize_t retval = 0;
2562 :
2563 0 : if (!count)
2564 : return 0; /* skip atime */
2565 :
2566 0 : if (iocb->ki_flags & IOCB_DIRECT) {
2567 0 : struct file *file = iocb->ki_filp;
2568 0 : struct address_space *mapping = file->f_mapping;
2569 0 : struct inode *inode = mapping->host;
2570 0 : loff_t size;
2571 :
2572 0 : size = i_size_read(inode);
2573 0 : if (iocb->ki_flags & IOCB_NOWAIT) {
2574 0 : if (filemap_range_has_page(mapping, iocb->ki_pos,
2575 0 : iocb->ki_pos + count - 1))
2576 : return -EAGAIN;
2577 : } else {
2578 0 : retval = filemap_write_and_wait_range(mapping,
2579 : iocb->ki_pos,
2580 0 : iocb->ki_pos + count - 1);
2581 0 : if (retval < 0)
2582 : return retval;
2583 : }
2584 :
2585 0 : file_accessed(file);
2586 :
2587 0 : retval = mapping->a_ops->direct_IO(iocb, iter);
2588 0 : if (retval >= 0) {
2589 0 : iocb->ki_pos += retval;
2590 0 : count -= retval;
2591 : }
2592 0 : if (retval != -EIOCBQUEUED)
2593 0 : iov_iter_revert(iter, count - iov_iter_count(iter));
2594 :
2595 : /*
2596 : * Btrfs can have a short DIO read if we encounter
2597 : * compressed extents, so if there was an error, or if
2598 : * we've already read everything we wanted to, or if
2599 : * there was a short read because we hit EOF, go ahead
2600 : * and return. Otherwise fallthrough to buffered io for
2601 : * the rest of the read. Buffered reads will not work for
2602 : * DAX files, so don't bother trying.
2603 : */
2604 0 : if (retval < 0 || !count || iocb->ki_pos >= size ||
2605 : IS_DAX(inode))
2606 : return retval;
2607 : }
2608 :
2609 0 : return filemap_read(iocb, iter, retval);
2610 : }
2611 : EXPORT_SYMBOL(generic_file_read_iter);
2612 :
2613 0 : static inline loff_t page_seek_hole_data(struct xa_state *xas,
2614 : struct address_space *mapping, struct page *page,
2615 : loff_t start, loff_t end, bool seek_data)
2616 : {
2617 0 : const struct address_space_operations *ops = mapping->a_ops;
2618 0 : size_t offset, bsz = i_blocksize(mapping->host);
2619 :
2620 0 : if (xa_is_value(page) || PageUptodate(page))
2621 0 : return seek_data ? start : end;
2622 0 : if (!ops->is_partially_uptodate)
2623 0 : return seek_data ? end : start;
2624 :
2625 0 : xas_pause(xas);
2626 0 : rcu_read_unlock();
2627 0 : lock_page(page);
2628 0 : if (unlikely(page->mapping != mapping))
2629 0 : goto unlock;
2630 :
2631 0 : offset = offset_in_thp(page, start) & ~(bsz - 1);
2632 :
2633 0 : do {
2634 0 : if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
2635 : break;
2636 0 : start = (start + bsz) & ~(bsz - 1);
2637 0 : offset += bsz;
2638 0 : } while (offset < thp_size(page));
2639 0 : unlock:
2640 0 : unlock_page(page);
2641 0 : rcu_read_lock();
2642 : return start;
2643 : }
2644 :
2645 : static inline
2646 0 : unsigned int seek_page_size(struct xa_state *xas, struct page *page)
2647 : {
2648 0 : if (xa_is_value(page))
2649 0 : return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2650 0 : return thp_size(page);
2651 : }
2652 :
2653 : /**
2654 : * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2655 : * @mapping: Address space to search.
2656 : * @start: First byte to consider.
2657 : * @end: Limit of search (exclusive).
2658 : * @whence: Either SEEK_HOLE or SEEK_DATA.
2659 : *
2660 : * If the page cache knows which blocks contain holes and which blocks
2661 : * contain data, your filesystem can use this function to implement
2662 : * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
2663 : * entirely memory-based such as tmpfs, and filesystems which support
2664 : * unwritten extents.
2665 : *
2666 : * Return: The requested offset on successs, or -ENXIO if @whence specifies
2667 : * SEEK_DATA and there is no data after @start. There is an implicit hole
2668 : * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2669 : * and @end contain data.
2670 : */
2671 0 : loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2672 : loff_t end, int whence)
2673 : {
2674 0 : XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2675 0 : pgoff_t max = (end - 1) / PAGE_SIZE;
2676 0 : bool seek_data = (whence == SEEK_DATA);
2677 0 : struct page *page;
2678 :
2679 0 : if (end <= start)
2680 : return -ENXIO;
2681 :
2682 0 : rcu_read_lock();
2683 0 : while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
2684 0 : loff_t pos = xas.xa_index * PAGE_SIZE;
2685 :
2686 0 : if (start < pos) {
2687 0 : if (!seek_data)
2688 0 : goto unlock;
2689 : start = pos;
2690 : }
2691 :
2692 0 : pos += seek_page_size(&xas, page);
2693 0 : start = page_seek_hole_data(&xas, mapping, page, start, pos,
2694 : seek_data);
2695 0 : if (start < pos)
2696 0 : goto unlock;
2697 0 : if (!xa_is_value(page))
2698 0 : put_page(page);
2699 : }
2700 0 : rcu_read_unlock();
2701 :
2702 0 : if (seek_data)
2703 : return -ENXIO;
2704 0 : goto out;
2705 :
2706 0 : unlock:
2707 0 : rcu_read_unlock();
2708 0 : if (!xa_is_value(page))
2709 0 : put_page(page);
2710 0 : out:
2711 0 : if (start > end)
2712 : return end;
2713 : return start;
2714 : }
2715 :
2716 : #ifdef CONFIG_MMU
2717 : #define MMAP_LOTSAMISS (100)
2718 : /*
2719 : * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2720 : * @vmf - the vm_fault for this fault.
2721 : * @page - the page to lock.
2722 : * @fpin - the pointer to the file we may pin (or is already pinned).
2723 : *
2724 : * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2725 : * It differs in that it actually returns the page locked if it returns 1 and 0
2726 : * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin
2727 : * will point to the pinned file and needs to be fput()'ed at a later point.
2728 : */
2729 0 : static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2730 : struct file **fpin)
2731 : {
2732 0 : if (trylock_page(page))
2733 : return 1;
2734 :
2735 : /*
2736 : * NOTE! This will make us return with VM_FAULT_RETRY, but with
2737 : * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2738 : * is supposed to work. We have way too many special cases..
2739 : */
2740 0 : if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2741 : return 0;
2742 :
2743 0 : *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2744 0 : if (vmf->flags & FAULT_FLAG_KILLABLE) {
2745 0 : if (__lock_page_killable(page)) {
2746 : /*
2747 : * We didn't have the right flags to drop the mmap_lock,
2748 : * but all fault_handlers only check for fatal signals
2749 : * if we return VM_FAULT_RETRY, so we need to drop the
2750 : * mmap_lock here and return 0 if we don't have a fpin.
2751 : */
2752 0 : if (*fpin == NULL)
2753 0 : mmap_read_unlock(vmf->vma->vm_mm);
2754 0 : return 0;
2755 : }
2756 : } else
2757 0 : __lock_page(page);
2758 : return 1;
2759 : }
2760 :
2761 :
2762 : /*
2763 : * Synchronous readahead happens when we don't even find a page in the page
2764 : * cache at all. We don't want to perform IO under the mmap sem, so if we have
2765 : * to drop the mmap sem we return the file that was pinned in order for us to do
2766 : * that. If we didn't pin a file then we return NULL. The file that is
2767 : * returned needs to be fput()'ed when we're done with it.
2768 : */
2769 0 : static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2770 : {
2771 0 : struct file *file = vmf->vma->vm_file;
2772 0 : struct file_ra_state *ra = &file->f_ra;
2773 0 : struct address_space *mapping = file->f_mapping;
2774 0 : DEFINE_READAHEAD(ractl, file, mapping, vmf->pgoff);
2775 0 : struct file *fpin = NULL;
2776 0 : unsigned int mmap_miss;
2777 :
2778 : /* If we don't want any read-ahead, don't bother */
2779 0 : if (vmf->vma->vm_flags & VM_RAND_READ)
2780 : return fpin;
2781 0 : if (!ra->ra_pages)
2782 : return fpin;
2783 :
2784 0 : if (vmf->vma->vm_flags & VM_SEQ_READ) {
2785 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2786 0 : page_cache_sync_ra(&ractl, ra, ra->ra_pages);
2787 0 : return fpin;
2788 : }
2789 :
2790 : /* Avoid banging the cache line if not needed */
2791 0 : mmap_miss = READ_ONCE(ra->mmap_miss);
2792 0 : if (mmap_miss < MMAP_LOTSAMISS * 10)
2793 0 : WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2794 :
2795 : /*
2796 : * Do we miss much more than hit in this file? If so,
2797 : * stop bothering with read-ahead. It will only hurt.
2798 : */
2799 0 : if (mmap_miss > MMAP_LOTSAMISS)
2800 : return fpin;
2801 :
2802 : /*
2803 : * mmap read-around
2804 : */
2805 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2806 0 : ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2807 0 : ra->size = ra->ra_pages;
2808 0 : ra->async_size = ra->ra_pages / 4;
2809 0 : ractl._index = ra->start;
2810 0 : do_page_cache_ra(&ractl, ra->size, ra->async_size);
2811 0 : return fpin;
2812 : }
2813 :
2814 : /*
2815 : * Asynchronous readahead happens when we find the page and PG_readahead,
2816 : * so we want to possibly extend the readahead further. We return the file that
2817 : * was pinned if we have to drop the mmap_lock in order to do IO.
2818 : */
2819 0 : static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2820 : struct page *page)
2821 : {
2822 0 : struct file *file = vmf->vma->vm_file;
2823 0 : struct file_ra_state *ra = &file->f_ra;
2824 0 : struct address_space *mapping = file->f_mapping;
2825 0 : struct file *fpin = NULL;
2826 0 : unsigned int mmap_miss;
2827 0 : pgoff_t offset = vmf->pgoff;
2828 :
2829 : /* If we don't want any read-ahead, don't bother */
2830 0 : if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2831 : return fpin;
2832 0 : mmap_miss = READ_ONCE(ra->mmap_miss);
2833 0 : if (mmap_miss)
2834 0 : WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2835 0 : if (PageReadahead(page)) {
2836 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2837 0 : page_cache_async_readahead(mapping, ra, file,
2838 0 : page, offset, ra->ra_pages);
2839 : }
2840 : return fpin;
2841 : }
2842 :
2843 : /**
2844 : * filemap_fault - read in file data for page fault handling
2845 : * @vmf: struct vm_fault containing details of the fault
2846 : *
2847 : * filemap_fault() is invoked via the vma operations vector for a
2848 : * mapped memory region to read in file data during a page fault.
2849 : *
2850 : * The goto's are kind of ugly, but this streamlines the normal case of having
2851 : * it in the page cache, and handles the special cases reasonably without
2852 : * having a lot of duplicated code.
2853 : *
2854 : * vma->vm_mm->mmap_lock must be held on entry.
2855 : *
2856 : * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2857 : * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2858 : *
2859 : * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2860 : * has not been released.
2861 : *
2862 : * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2863 : *
2864 : * Return: bitwise-OR of %VM_FAULT_ codes.
2865 : */
2866 0 : vm_fault_t filemap_fault(struct vm_fault *vmf)
2867 : {
2868 0 : int error;
2869 0 : struct file *file = vmf->vma->vm_file;
2870 0 : struct file *fpin = NULL;
2871 0 : struct address_space *mapping = file->f_mapping;
2872 0 : struct file_ra_state *ra = &file->f_ra;
2873 0 : struct inode *inode = mapping->host;
2874 0 : pgoff_t offset = vmf->pgoff;
2875 0 : pgoff_t max_off;
2876 0 : struct page *page;
2877 0 : vm_fault_t ret = 0;
2878 :
2879 0 : max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2880 0 : if (unlikely(offset >= max_off))
2881 : return VM_FAULT_SIGBUS;
2882 :
2883 : /*
2884 : * Do we have something in the page cache already?
2885 : */
2886 0 : page = find_get_page(mapping, offset);
2887 0 : if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2888 : /*
2889 : * We found the page, so try async readahead before
2890 : * waiting for the lock.
2891 : */
2892 0 : fpin = do_async_mmap_readahead(vmf, page);
2893 0 : } else if (!page) {
2894 : /* No page in the page cache at all */
2895 0 : count_vm_event(PGMAJFAULT);
2896 0 : count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2897 0 : ret = VM_FAULT_MAJOR;
2898 0 : fpin = do_sync_mmap_readahead(vmf);
2899 0 : retry_find:
2900 0 : page = pagecache_get_page(mapping, offset,
2901 : FGP_CREAT|FGP_FOR_MMAP,
2902 : vmf->gfp_mask);
2903 0 : if (!page) {
2904 0 : if (fpin)
2905 0 : goto out_retry;
2906 : return VM_FAULT_OOM;
2907 : }
2908 : }
2909 :
2910 0 : if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2911 0 : goto out_retry;
2912 :
2913 : /* Did it get truncated? */
2914 0 : if (unlikely(compound_head(page)->mapping != mapping)) {
2915 0 : unlock_page(page);
2916 0 : put_page(page);
2917 0 : goto retry_find;
2918 : }
2919 0 : VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2920 :
2921 : /*
2922 : * We have a locked page in the page cache, now we need to check
2923 : * that it's up-to-date. If not, it is going to be due to an error.
2924 : */
2925 0 : if (unlikely(!PageUptodate(page)))
2926 0 : goto page_not_uptodate;
2927 :
2928 : /*
2929 : * We've made it this far and we had to drop our mmap_lock, now is the
2930 : * time to return to the upper layer and have it re-find the vma and
2931 : * redo the fault.
2932 : */
2933 0 : if (fpin) {
2934 0 : unlock_page(page);
2935 0 : goto out_retry;
2936 : }
2937 :
2938 : /*
2939 : * Found the page and have a reference on it.
2940 : * We must recheck i_size under page lock.
2941 : */
2942 0 : max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2943 0 : if (unlikely(offset >= max_off)) {
2944 0 : unlock_page(page);
2945 0 : put_page(page);
2946 0 : return VM_FAULT_SIGBUS;
2947 : }
2948 :
2949 0 : vmf->page = page;
2950 0 : return ret | VM_FAULT_LOCKED;
2951 :
2952 0 : page_not_uptodate:
2953 : /*
2954 : * Umm, take care of errors if the page isn't up-to-date.
2955 : * Try to re-read it _once_. We do this synchronously,
2956 : * because there really aren't any performance issues here
2957 : * and we need to check for errors.
2958 : */
2959 0 : ClearPageError(page);
2960 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2961 0 : error = mapping->a_ops->readpage(file, page);
2962 0 : if (!error) {
2963 0 : wait_on_page_locked(page);
2964 0 : if (!PageUptodate(page))
2965 0 : error = -EIO;
2966 : }
2967 0 : if (fpin)
2968 0 : goto out_retry;
2969 0 : put_page(page);
2970 :
2971 0 : if (!error || error == AOP_TRUNCATED_PAGE)
2972 0 : goto retry_find;
2973 :
2974 0 : shrink_readahead_size_eio(ra);
2975 0 : return VM_FAULT_SIGBUS;
2976 :
2977 0 : out_retry:
2978 : /*
2979 : * We dropped the mmap_lock, we need to return to the fault handler to
2980 : * re-find the vma and come back and find our hopefully still populated
2981 : * page.
2982 : */
2983 0 : if (page)
2984 0 : put_page(page);
2985 0 : if (fpin)
2986 0 : fput(fpin);
2987 0 : return ret | VM_FAULT_RETRY;
2988 : }
2989 : EXPORT_SYMBOL(filemap_fault);
2990 :
2991 0 : static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
2992 : {
2993 0 : struct mm_struct *mm = vmf->vma->vm_mm;
2994 :
2995 : /* Huge page is mapped? No need to proceed. */
2996 0 : if (pmd_trans_huge(*vmf->pmd)) {
2997 0 : unlock_page(page);
2998 0 : put_page(page);
2999 0 : return true;
3000 : }
3001 :
3002 0 : if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3003 0 : vm_fault_t ret = do_set_pmd(vmf, page);
3004 0 : if (!ret) {
3005 : /* The page is mapped successfully, reference consumed. */
3006 0 : unlock_page(page);
3007 0 : return true;
3008 : }
3009 : }
3010 :
3011 0 : if (pmd_none(*vmf->pmd)) {
3012 0 : vmf->ptl = pmd_lock(mm, vmf->pmd);
3013 0 : if (likely(pmd_none(*vmf->pmd))) {
3014 0 : mm_inc_nr_ptes(mm);
3015 0 : pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
3016 0 : vmf->prealloc_pte = NULL;
3017 : }
3018 0 : spin_unlock(vmf->ptl);
3019 : }
3020 :
3021 : /* See comment in handle_pte_fault() */
3022 0 : if (pmd_devmap_trans_unstable(vmf->pmd)) {
3023 0 : unlock_page(page);
3024 0 : put_page(page);
3025 0 : return true;
3026 : }
3027 :
3028 : return false;
3029 : }
3030 :
3031 0 : static struct page *next_uptodate_page(struct page *page,
3032 : struct address_space *mapping,
3033 : struct xa_state *xas, pgoff_t end_pgoff)
3034 : {
3035 0 : unsigned long max_idx;
3036 :
3037 0 : do {
3038 0 : if (!page)
3039 : return NULL;
3040 0 : if (xas_retry(xas, page))
3041 0 : continue;
3042 0 : if (xa_is_value(page))
3043 0 : continue;
3044 0 : if (PageLocked(page))
3045 0 : continue;
3046 0 : if (!page_cache_get_speculative(page))
3047 0 : continue;
3048 : /* Has the page moved or been split? */
3049 0 : if (unlikely(page != xas_reload(xas)))
3050 0 : goto skip;
3051 0 : if (!PageUptodate(page) || PageReadahead(page))
3052 0 : goto skip;
3053 0 : if (PageHWPoison(page))
3054 : goto skip;
3055 0 : if (!trylock_page(page))
3056 0 : goto skip;
3057 0 : if (page->mapping != mapping)
3058 0 : goto unlock;
3059 0 : if (!PageUptodate(page))
3060 0 : goto unlock;
3061 0 : max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3062 0 : if (xas->xa_index >= max_idx)
3063 0 : goto unlock;
3064 : return page;
3065 0 : unlock:
3066 0 : unlock_page(page);
3067 0 : skip:
3068 0 : put_page(page);
3069 0 : } while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3070 :
3071 : return NULL;
3072 : }
3073 :
3074 0 : static inline struct page *first_map_page(struct address_space *mapping,
3075 : struct xa_state *xas,
3076 : pgoff_t end_pgoff)
3077 : {
3078 0 : return next_uptodate_page(xas_find(xas, end_pgoff),
3079 : mapping, xas, end_pgoff);
3080 : }
3081 :
3082 0 : static inline struct page *next_map_page(struct address_space *mapping,
3083 : struct xa_state *xas,
3084 : pgoff_t end_pgoff)
3085 : {
3086 0 : return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3087 : mapping, xas, end_pgoff);
3088 : }
3089 :
3090 0 : vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3091 : pgoff_t start_pgoff, pgoff_t end_pgoff)
3092 : {
3093 0 : struct vm_area_struct *vma = vmf->vma;
3094 0 : struct file *file = vma->vm_file;
3095 0 : struct address_space *mapping = file->f_mapping;
3096 0 : pgoff_t last_pgoff = start_pgoff;
3097 0 : unsigned long addr;
3098 0 : XA_STATE(xas, &mapping->i_pages, start_pgoff);
3099 0 : struct page *head, *page;
3100 0 : unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3101 0 : vm_fault_t ret = 0;
3102 :
3103 0 : rcu_read_lock();
3104 0 : head = first_map_page(mapping, &xas, end_pgoff);
3105 0 : if (!head)
3106 0 : goto out;
3107 :
3108 0 : if (filemap_map_pmd(vmf, head)) {
3109 0 : ret = VM_FAULT_NOPAGE;
3110 0 : goto out;
3111 : }
3112 :
3113 0 : addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3114 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3115 0 : do {
3116 0 : page = find_subpage(head, xas.xa_index);
3117 0 : if (PageHWPoison(page))
3118 : goto unlock;
3119 :
3120 0 : if (mmap_miss > 0)
3121 0 : mmap_miss--;
3122 :
3123 0 : addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3124 0 : vmf->pte += xas.xa_index - last_pgoff;
3125 0 : last_pgoff = xas.xa_index;
3126 :
3127 0 : if (!pte_none(*vmf->pte))
3128 0 : goto unlock;
3129 :
3130 : /* We're about to handle the fault */
3131 0 : if (vmf->address == addr)
3132 0 : ret = VM_FAULT_NOPAGE;
3133 :
3134 0 : do_set_pte(vmf, page, addr);
3135 : /* no need to invalidate: a not-present page won't be cached */
3136 0 : update_mmu_cache(vma, addr, vmf->pte);
3137 0 : unlock_page(head);
3138 0 : continue;
3139 0 : unlock:
3140 0 : unlock_page(head);
3141 0 : put_page(head);
3142 0 : } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3143 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3144 0 : out:
3145 0 : rcu_read_unlock();
3146 0 : WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3147 0 : return ret;
3148 : }
3149 : EXPORT_SYMBOL(filemap_map_pages);
3150 :
3151 0 : vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3152 : {
3153 0 : struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3154 0 : struct page *page = vmf->page;
3155 0 : vm_fault_t ret = VM_FAULT_LOCKED;
3156 :
3157 0 : sb_start_pagefault(mapping->host->i_sb);
3158 0 : file_update_time(vmf->vma->vm_file);
3159 0 : lock_page(page);
3160 0 : if (page->mapping != mapping) {
3161 0 : unlock_page(page);
3162 0 : ret = VM_FAULT_NOPAGE;
3163 0 : goto out;
3164 : }
3165 : /*
3166 : * We mark the page dirty already here so that when freeze is in
3167 : * progress, we are guaranteed that writeback during freezing will
3168 : * see the dirty page and writeprotect it again.
3169 : */
3170 0 : set_page_dirty(page);
3171 0 : wait_for_stable_page(page);
3172 0 : out:
3173 0 : sb_end_pagefault(mapping->host->i_sb);
3174 0 : return ret;
3175 : }
3176 :
3177 : const struct vm_operations_struct generic_file_vm_ops = {
3178 : .fault = filemap_fault,
3179 : .map_pages = filemap_map_pages,
3180 : .page_mkwrite = filemap_page_mkwrite,
3181 : };
3182 :
3183 : /* This is used for a general mmap of a disk file */
3184 :
3185 0 : int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3186 : {
3187 0 : struct address_space *mapping = file->f_mapping;
3188 :
3189 0 : if (!mapping->a_ops->readpage)
3190 : return -ENOEXEC;
3191 0 : file_accessed(file);
3192 0 : vma->vm_ops = &generic_file_vm_ops;
3193 0 : return 0;
3194 : }
3195 :
3196 : /*
3197 : * This is for filesystems which do not implement ->writepage.
3198 : */
3199 0 : int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3200 : {
3201 0 : if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3202 : return -EINVAL;
3203 0 : return generic_file_mmap(file, vma);
3204 : }
3205 : #else
3206 : vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3207 : {
3208 : return VM_FAULT_SIGBUS;
3209 : }
3210 : int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
3211 : {
3212 : return -ENOSYS;
3213 : }
3214 : int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
3215 : {
3216 : return -ENOSYS;
3217 : }
3218 : #endif /* CONFIG_MMU */
3219 :
3220 : EXPORT_SYMBOL(filemap_page_mkwrite);
3221 : EXPORT_SYMBOL(generic_file_mmap);
3222 : EXPORT_SYMBOL(generic_file_readonly_mmap);
3223 :
3224 0 : static struct page *wait_on_page_read(struct page *page)
3225 : {
3226 0 : if (!IS_ERR(page)) {
3227 0 : wait_on_page_locked(page);
3228 0 : if (!PageUptodate(page)) {
3229 0 : put_page(page);
3230 0 : page = ERR_PTR(-EIO);
3231 : }
3232 : }
3233 0 : return page;
3234 : }
3235 :
3236 0 : static struct page *do_read_cache_page(struct address_space *mapping,
3237 : pgoff_t index,
3238 : int (*filler)(void *, struct page *),
3239 : void *data,
3240 : gfp_t gfp)
3241 : {
3242 0 : struct page *page;
3243 0 : int err;
3244 : repeat:
3245 0 : page = find_get_page(mapping, index);
3246 0 : if (!page) {
3247 0 : page = __page_cache_alloc(gfp);
3248 0 : if (!page)
3249 0 : return ERR_PTR(-ENOMEM);
3250 0 : err = add_to_page_cache_lru(page, mapping, index, gfp);
3251 0 : if (unlikely(err)) {
3252 0 : put_page(page);
3253 0 : if (err == -EEXIST)
3254 0 : goto repeat;
3255 : /* Presumably ENOMEM for xarray node */
3256 0 : return ERR_PTR(err);
3257 : }
3258 :
3259 0 : filler:
3260 0 : if (filler)
3261 0 : err = filler(data, page);
3262 : else
3263 0 : err = mapping->a_ops->readpage(data, page);
3264 :
3265 0 : if (err < 0) {
3266 0 : put_page(page);
3267 0 : return ERR_PTR(err);
3268 : }
3269 :
3270 0 : page = wait_on_page_read(page);
3271 0 : if (IS_ERR(page))
3272 : return page;
3273 0 : goto out;
3274 : }
3275 0 : if (PageUptodate(page))
3276 0 : goto out;
3277 :
3278 : /*
3279 : * Page is not up to date and may be locked due to one of the following
3280 : * case a: Page is being filled and the page lock is held
3281 : * case b: Read/write error clearing the page uptodate status
3282 : * case c: Truncation in progress (page locked)
3283 : * case d: Reclaim in progress
3284 : *
3285 : * Case a, the page will be up to date when the page is unlocked.
3286 : * There is no need to serialise on the page lock here as the page
3287 : * is pinned so the lock gives no additional protection. Even if the
3288 : * page is truncated, the data is still valid if PageUptodate as
3289 : * it's a race vs truncate race.
3290 : * Case b, the page will not be up to date
3291 : * Case c, the page may be truncated but in itself, the data may still
3292 : * be valid after IO completes as it's a read vs truncate race. The
3293 : * operation must restart if the page is not uptodate on unlock but
3294 : * otherwise serialising on page lock to stabilise the mapping gives
3295 : * no additional guarantees to the caller as the page lock is
3296 : * released before return.
3297 : * Case d, similar to truncation. If reclaim holds the page lock, it
3298 : * will be a race with remove_mapping that determines if the mapping
3299 : * is valid on unlock but otherwise the data is valid and there is
3300 : * no need to serialise with page lock.
3301 : *
3302 : * As the page lock gives no additional guarantee, we optimistically
3303 : * wait on the page to be unlocked and check if it's up to date and
3304 : * use the page if it is. Otherwise, the page lock is required to
3305 : * distinguish between the different cases. The motivation is that we
3306 : * avoid spurious serialisations and wakeups when multiple processes
3307 : * wait on the same page for IO to complete.
3308 : */
3309 0 : wait_on_page_locked(page);
3310 0 : if (PageUptodate(page))
3311 0 : goto out;
3312 :
3313 : /* Distinguish between all the cases under the safety of the lock */
3314 0 : lock_page(page);
3315 :
3316 : /* Case c or d, restart the operation */
3317 0 : if (!page->mapping) {
3318 0 : unlock_page(page);
3319 0 : put_page(page);
3320 0 : goto repeat;
3321 : }
3322 :
3323 : /* Someone else locked and filled the page in a very small window */
3324 0 : if (PageUptodate(page)) {
3325 0 : unlock_page(page);
3326 0 : goto out;
3327 : }
3328 :
3329 : /*
3330 : * A previous I/O error may have been due to temporary
3331 : * failures.
3332 : * Clear page error before actual read, PG_error will be
3333 : * set again if read page fails.
3334 : */
3335 0 : ClearPageError(page);
3336 0 : goto filler;
3337 :
3338 0 : out:
3339 0 : mark_page_accessed(page);
3340 0 : return page;
3341 : }
3342 :
3343 : /**
3344 : * read_cache_page - read into page cache, fill it if needed
3345 : * @mapping: the page's address_space
3346 : * @index: the page index
3347 : * @filler: function to perform the read
3348 : * @data: first arg to filler(data, page) function, often left as NULL
3349 : *
3350 : * Read into the page cache. If a page already exists, and PageUptodate() is
3351 : * not set, try to fill the page and wait for it to become unlocked.
3352 : *
3353 : * If the page does not get brought uptodate, return -EIO.
3354 : *
3355 : * Return: up to date page on success, ERR_PTR() on failure.
3356 : */
3357 0 : struct page *read_cache_page(struct address_space *mapping,
3358 : pgoff_t index,
3359 : int (*filler)(void *, struct page *),
3360 : void *data)
3361 : {
3362 0 : return do_read_cache_page(mapping, index, filler, data,
3363 : mapping_gfp_mask(mapping));
3364 : }
3365 : EXPORT_SYMBOL(read_cache_page);
3366 :
3367 : /**
3368 : * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3369 : * @mapping: the page's address_space
3370 : * @index: the page index
3371 : * @gfp: the page allocator flags to use if allocating
3372 : *
3373 : * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3374 : * any new page allocations done using the specified allocation flags.
3375 : *
3376 : * If the page does not get brought uptodate, return -EIO.
3377 : *
3378 : * Return: up to date page on success, ERR_PTR() on failure.
3379 : */
3380 0 : struct page *read_cache_page_gfp(struct address_space *mapping,
3381 : pgoff_t index,
3382 : gfp_t gfp)
3383 : {
3384 0 : return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3385 : }
3386 : EXPORT_SYMBOL(read_cache_page_gfp);
3387 :
3388 0 : int pagecache_write_begin(struct file *file, struct address_space *mapping,
3389 : loff_t pos, unsigned len, unsigned flags,
3390 : struct page **pagep, void **fsdata)
3391 : {
3392 0 : const struct address_space_operations *aops = mapping->a_ops;
3393 :
3394 0 : return aops->write_begin(file, mapping, pos, len, flags,
3395 : pagep, fsdata);
3396 : }
3397 : EXPORT_SYMBOL(pagecache_write_begin);
3398 :
3399 0 : int pagecache_write_end(struct file *file, struct address_space *mapping,
3400 : loff_t pos, unsigned len, unsigned copied,
3401 : struct page *page, void *fsdata)
3402 : {
3403 0 : const struct address_space_operations *aops = mapping->a_ops;
3404 :
3405 0 : return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3406 : }
3407 : EXPORT_SYMBOL(pagecache_write_end);
3408 :
3409 : /*
3410 : * Warn about a page cache invalidation failure during a direct I/O write.
3411 : */
3412 0 : void dio_warn_stale_pagecache(struct file *filp)
3413 : {
3414 0 : static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3415 0 : char pathname[128];
3416 0 : char *path;
3417 :
3418 0 : errseq_set(&filp->f_mapping->wb_err, -EIO);
3419 0 : if (__ratelimit(&_rs)) {
3420 0 : path = file_path(filp, pathname, sizeof(pathname));
3421 0 : if (IS_ERR(path))
3422 0 : path = "(unknown)";
3423 0 : pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3424 0 : pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3425 : current->comm);
3426 : }
3427 0 : }
3428 :
3429 : ssize_t
3430 0 : generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3431 : {
3432 0 : struct file *file = iocb->ki_filp;
3433 0 : struct address_space *mapping = file->f_mapping;
3434 0 : struct inode *inode = mapping->host;
3435 0 : loff_t pos = iocb->ki_pos;
3436 0 : ssize_t written;
3437 0 : size_t write_len;
3438 0 : pgoff_t end;
3439 :
3440 0 : write_len = iov_iter_count(from);
3441 0 : end = (pos + write_len - 1) >> PAGE_SHIFT;
3442 :
3443 0 : if (iocb->ki_flags & IOCB_NOWAIT) {
3444 : /* If there are pages to writeback, return */
3445 0 : if (filemap_range_has_page(file->f_mapping, pos,
3446 : pos + write_len - 1))
3447 : return -EAGAIN;
3448 : } else {
3449 0 : written = filemap_write_and_wait_range(mapping, pos,
3450 : pos + write_len - 1);
3451 0 : if (written)
3452 0 : goto out;
3453 : }
3454 :
3455 : /*
3456 : * After a write we want buffered reads to be sure to go to disk to get
3457 : * the new data. We invalidate clean cached page from the region we're
3458 : * about to write. We do this *before* the write so that we can return
3459 : * without clobbering -EIOCBQUEUED from ->direct_IO().
3460 : */
3461 0 : written = invalidate_inode_pages2_range(mapping,
3462 0 : pos >> PAGE_SHIFT, end);
3463 : /*
3464 : * If a page can not be invalidated, return 0 to fall back
3465 : * to buffered write.
3466 : */
3467 0 : if (written) {
3468 0 : if (written == -EBUSY)
3469 : return 0;
3470 0 : goto out;
3471 : }
3472 :
3473 0 : written = mapping->a_ops->direct_IO(iocb, from);
3474 :
3475 : /*
3476 : * Finally, try again to invalidate clean pages which might have been
3477 : * cached by non-direct readahead, or faulted in by get_user_pages()
3478 : * if the source of the write was an mmap'ed region of the file
3479 : * we're writing. Either one is a pretty crazy thing to do,
3480 : * so we don't support it 100%. If this invalidation
3481 : * fails, tough, the write still worked...
3482 : *
3483 : * Most of the time we do not need this since dio_complete() will do
3484 : * the invalidation for us. However there are some file systems that
3485 : * do not end up with dio_complete() being called, so let's not break
3486 : * them by removing it completely.
3487 : *
3488 : * Noticeable example is a blkdev_direct_IO().
3489 : *
3490 : * Skip invalidation for async writes or if mapping has no pages.
3491 : */
3492 0 : if (written > 0 && mapping->nrpages &&
3493 0 : invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3494 0 : dio_warn_stale_pagecache(file);
3495 :
3496 0 : if (written > 0) {
3497 0 : pos += written;
3498 0 : write_len -= written;
3499 0 : if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3500 0 : i_size_write(inode, pos);
3501 0 : mark_inode_dirty(inode);
3502 : }
3503 0 : iocb->ki_pos = pos;
3504 : }
3505 0 : if (written != -EIOCBQUEUED)
3506 0 : iov_iter_revert(from, write_len - iov_iter_count(from));
3507 0 : out:
3508 : return written;
3509 : }
3510 : EXPORT_SYMBOL(generic_file_direct_write);
3511 :
3512 : /*
3513 : * Find or create a page at the given pagecache position. Return the locked
3514 : * page. This function is specifically for buffered writes.
3515 : */
3516 0 : struct page *grab_cache_page_write_begin(struct address_space *mapping,
3517 : pgoff_t index, unsigned flags)
3518 : {
3519 0 : struct page *page;
3520 0 : int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3521 :
3522 0 : if (flags & AOP_FLAG_NOFS)
3523 0 : fgp_flags |= FGP_NOFS;
3524 :
3525 0 : page = pagecache_get_page(mapping, index, fgp_flags,
3526 : mapping_gfp_mask(mapping));
3527 0 : if (page)
3528 0 : wait_for_stable_page(page);
3529 :
3530 0 : return page;
3531 : }
3532 : EXPORT_SYMBOL(grab_cache_page_write_begin);
3533 :
3534 0 : ssize_t generic_perform_write(struct file *file,
3535 : struct iov_iter *i, loff_t pos)
3536 : {
3537 0 : struct address_space *mapping = file->f_mapping;
3538 0 : const struct address_space_operations *a_ops = mapping->a_ops;
3539 0 : long status = 0;
3540 0 : ssize_t written = 0;
3541 0 : unsigned int flags = 0;
3542 :
3543 0 : do {
3544 0 : struct page *page;
3545 0 : unsigned long offset; /* Offset into pagecache page */
3546 0 : unsigned long bytes; /* Bytes to write to page */
3547 0 : size_t copied; /* Bytes copied from user */
3548 0 : void *fsdata;
3549 :
3550 0 : offset = (pos & (PAGE_SIZE - 1));
3551 0 : bytes = min_t(unsigned long, PAGE_SIZE - offset,
3552 : iov_iter_count(i));
3553 :
3554 0 : again:
3555 : /*
3556 : * Bring in the user page that we will copy from _first_.
3557 : * Otherwise there's a nasty deadlock on copying from the
3558 : * same page as we're writing to, without it being marked
3559 : * up-to-date.
3560 : *
3561 : * Not only is this an optimisation, but it is also required
3562 : * to check that the address is actually valid, when atomic
3563 : * usercopies are used, below.
3564 : */
3565 0 : if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3566 : status = -EFAULT;
3567 0 : break;
3568 : }
3569 :
3570 0 : if (fatal_signal_pending(current)) {
3571 : status = -EINTR;
3572 : break;
3573 : }
3574 :
3575 0 : status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3576 : &page, &fsdata);
3577 0 : if (unlikely(status < 0))
3578 : break;
3579 :
3580 0 : if (mapping_writably_mapped(mapping))
3581 0 : flush_dcache_page(page);
3582 :
3583 0 : copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3584 0 : flush_dcache_page(page);
3585 :
3586 0 : status = a_ops->write_end(file, mapping, pos, bytes, copied,
3587 : page, fsdata);
3588 0 : if (unlikely(status < 0))
3589 : break;
3590 0 : copied = status;
3591 :
3592 0 : cond_resched();
3593 :
3594 0 : iov_iter_advance(i, copied);
3595 0 : if (unlikely(copied == 0)) {
3596 : /*
3597 : * If we were unable to copy any data at all, we must
3598 : * fall back to a single segment length write.
3599 : *
3600 : * If we didn't fallback here, we could livelock
3601 : * because not all segments in the iov can be copied at
3602 : * once without a pagefault.
3603 : */
3604 0 : bytes = min_t(unsigned long, PAGE_SIZE - offset,
3605 : iov_iter_single_seg_count(i));
3606 0 : goto again;
3607 : }
3608 0 : pos += copied;
3609 0 : written += copied;
3610 :
3611 0 : balance_dirty_pages_ratelimited(mapping);
3612 0 : } while (iov_iter_count(i));
3613 :
3614 0 : return written ? written : status;
3615 : }
3616 : EXPORT_SYMBOL(generic_perform_write);
3617 :
3618 : /**
3619 : * __generic_file_write_iter - write data to a file
3620 : * @iocb: IO state structure (file, offset, etc.)
3621 : * @from: iov_iter with data to write
3622 : *
3623 : * This function does all the work needed for actually writing data to a
3624 : * file. It does all basic checks, removes SUID from the file, updates
3625 : * modification times and calls proper subroutines depending on whether we
3626 : * do direct IO or a standard buffered write.
3627 : *
3628 : * It expects i_mutex to be grabbed unless we work on a block device or similar
3629 : * object which does not need locking at all.
3630 : *
3631 : * This function does *not* take care of syncing data in case of O_SYNC write.
3632 : * A caller has to handle it. This is mainly due to the fact that we want to
3633 : * avoid syncing under i_mutex.
3634 : *
3635 : * Return:
3636 : * * number of bytes written, even for truncated writes
3637 : * * negative error code if no data has been written at all
3638 : */
3639 0 : ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3640 : {
3641 0 : struct file *file = iocb->ki_filp;
3642 0 : struct address_space * mapping = file->f_mapping;
3643 0 : struct inode *inode = mapping->host;
3644 0 : ssize_t written = 0;
3645 0 : ssize_t err;
3646 0 : ssize_t status;
3647 :
3648 : /* We can write back this queue in page reclaim */
3649 0 : current->backing_dev_info = inode_to_bdi(inode);
3650 0 : err = file_remove_privs(file);
3651 0 : if (err)
3652 0 : goto out;
3653 :
3654 0 : err = file_update_time(file);
3655 0 : if (err)
3656 0 : goto out;
3657 :
3658 0 : if (iocb->ki_flags & IOCB_DIRECT) {
3659 0 : loff_t pos, endbyte;
3660 :
3661 0 : written = generic_file_direct_write(iocb, from);
3662 : /*
3663 : * If the write stopped short of completing, fall back to
3664 : * buffered writes. Some filesystems do this for writes to
3665 : * holes, for example. For DAX files, a buffered write will
3666 : * not succeed (even if it did, DAX does not handle dirty
3667 : * page-cache pages correctly).
3668 : */
3669 0 : if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3670 0 : goto out;
3671 :
3672 0 : status = generic_perform_write(file, from, pos = iocb->ki_pos);
3673 : /*
3674 : * If generic_perform_write() returned a synchronous error
3675 : * then we want to return the number of bytes which were
3676 : * direct-written, or the error code if that was zero. Note
3677 : * that this differs from normal direct-io semantics, which
3678 : * will return -EFOO even if some bytes were written.
3679 : */
3680 0 : if (unlikely(status < 0)) {
3681 0 : err = status;
3682 0 : goto out;
3683 : }
3684 : /*
3685 : * We need to ensure that the page cache pages are written to
3686 : * disk and invalidated to preserve the expected O_DIRECT
3687 : * semantics.
3688 : */
3689 0 : endbyte = pos + status - 1;
3690 0 : err = filemap_write_and_wait_range(mapping, pos, endbyte);
3691 0 : if (err == 0) {
3692 0 : iocb->ki_pos = endbyte + 1;
3693 0 : written += status;
3694 0 : invalidate_mapping_pages(mapping,
3695 0 : pos >> PAGE_SHIFT,
3696 0 : endbyte >> PAGE_SHIFT);
3697 : } else {
3698 : /*
3699 : * We don't know how much we wrote, so just return
3700 : * the number of bytes which were direct-written
3701 : */
3702 : }
3703 : } else {
3704 0 : written = generic_perform_write(file, from, iocb->ki_pos);
3705 0 : if (likely(written > 0))
3706 0 : iocb->ki_pos += written;
3707 : }
3708 0 : out:
3709 0 : current->backing_dev_info = NULL;
3710 0 : return written ? written : err;
3711 : }
3712 : EXPORT_SYMBOL(__generic_file_write_iter);
3713 :
3714 : /**
3715 : * generic_file_write_iter - write data to a file
3716 : * @iocb: IO state structure
3717 : * @from: iov_iter with data to write
3718 : *
3719 : * This is a wrapper around __generic_file_write_iter() to be used by most
3720 : * filesystems. It takes care of syncing the file in case of O_SYNC file
3721 : * and acquires i_mutex as needed.
3722 : * Return:
3723 : * * negative error code if no data has been written at all of
3724 : * vfs_fsync_range() failed for a synchronous write
3725 : * * number of bytes written, even for truncated writes
3726 : */
3727 0 : ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3728 : {
3729 0 : struct file *file = iocb->ki_filp;
3730 0 : struct inode *inode = file->f_mapping->host;
3731 0 : ssize_t ret;
3732 :
3733 0 : inode_lock(inode);
3734 0 : ret = generic_write_checks(iocb, from);
3735 0 : if (ret > 0)
3736 0 : ret = __generic_file_write_iter(iocb, from);
3737 0 : inode_unlock(inode);
3738 :
3739 0 : if (ret > 0)
3740 0 : ret = generic_write_sync(iocb, ret);
3741 0 : return ret;
3742 : }
3743 : EXPORT_SYMBOL(generic_file_write_iter);
3744 :
3745 : /**
3746 : * try_to_release_page() - release old fs-specific metadata on a page
3747 : *
3748 : * @page: the page which the kernel is trying to free
3749 : * @gfp_mask: memory allocation flags (and I/O mode)
3750 : *
3751 : * The address_space is to try to release any data against the page
3752 : * (presumably at page->private).
3753 : *
3754 : * This may also be called if PG_fscache is set on a page, indicating that the
3755 : * page is known to the local caching routines.
3756 : *
3757 : * The @gfp_mask argument specifies whether I/O may be performed to release
3758 : * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3759 : *
3760 : * Return: %1 if the release was successful, otherwise return zero.
3761 : */
3762 0 : int try_to_release_page(struct page *page, gfp_t gfp_mask)
3763 : {
3764 0 : struct address_space * const mapping = page->mapping;
3765 :
3766 0 : BUG_ON(!PageLocked(page));
3767 0 : if (PageWriteback(page))
3768 : return 0;
3769 :
3770 0 : if (mapping && mapping->a_ops->releasepage)
3771 0 : return mapping->a_ops->releasepage(page, gfp_mask);
3772 0 : return try_to_free_buffers(page);
3773 : }
3774 :
3775 : EXPORT_SYMBOL(try_to_release_page);
|
