~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/filemap.c

Version: ~ [ linux-5.6 ] ~ [ linux-5.5.13 ] ~ [ linux-5.4.28 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.113 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.174 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.217 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.217 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.82 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  *      linux/mm/filemap.c
  4  *
  5  * Copyright (C) 1994-1999  Linus Torvalds
  6  */
  7 
  8 /*
  9  * This file handles the generic file mmap semantics used by
 10  * most "normal" filesystems (but you don't /have/ to use this:
 11  * the NFS filesystem used to do this differently, for example)
 12  */
 13 #include <linux/export.h>
 14 #include <linux/compiler.h>
 15 #include <linux/dax.h>
 16 #include <linux/fs.h>
 17 #include <linux/sched/signal.h>
 18 #include <linux/uaccess.h>
 19 #include <linux/capability.h>
 20 #include <linux/kernel_stat.h>
 21 #include <linux/gfp.h>
 22 #include <linux/mm.h>
 23 #include <linux/swap.h>
 24 #include <linux/mman.h>
 25 #include <linux/pagemap.h>
 26 #include <linux/file.h>
 27 #include <linux/uio.h>
 28 #include <linux/error-injection.h>
 29 #include <linux/hash.h>
 30 #include <linux/writeback.h>
 31 #include <linux/backing-dev.h>
 32 #include <linux/pagevec.h>
 33 #include <linux/blkdev.h>
 34 #include <linux/security.h>
 35 #include <linux/cpuset.h>
 36 #include <linux/hugetlb.h>
 37 #include <linux/memcontrol.h>
 38 #include <linux/cleancache.h>
 39 #include <linux/shmem_fs.h>
 40 #include <linux/rmap.h>
 41 #include <linux/delayacct.h>
 42 #include <linux/psi.h>
 43 #include <linux/ramfs.h>
 44 #include "internal.h"
 45 
 46 #define CREATE_TRACE_POINTS
 47 #include <trace/events/filemap.h>
 48 
 49 /*
 50  * FIXME: remove all knowledge of the buffer layer from the core VM
 51  */
 52 #include <linux/buffer_head.h> /* for try_to_free_buffers */
 53 
 54 #include <asm/mman.h>
 55 
 56 /*
 57  * Shared mappings implemented 30.11.1994. It's not fully working yet,
 58  * though.
 59  *
 60  * Shared mappings now work. 15.8.1995  Bruno.
 61  *
 62  * finished 'unifying' the page and buffer cache and SMP-threaded the
 63  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
 64  *
 65  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
 66  */
 67 
 68 /*
 69  * Lock ordering:
 70  *
 71  *  ->i_mmap_rwsem              (truncate_pagecache)
 72  *    ->private_lock            (__free_pte->__set_page_dirty_buffers)
 73  *      ->swap_lock             (exclusive_swap_page, others)
 74  *        ->i_pages lock
 75  *
 76  *  ->i_mutex
 77  *    ->i_mmap_rwsem            (truncate->unmap_mapping_range)
 78  *
 79  *  ->mmap_sem
 80  *    ->i_mmap_rwsem
 81  *      ->page_table_lock or pte_lock   (various, mainly in memory.c)
 82  *        ->i_pages lock        (arch-dependent flush_dcache_mmap_lock)
 83  *
 84  *  ->mmap_sem
 85  *    ->lock_page               (access_process_vm)
 86  *
 87  *  ->i_mutex                   (generic_perform_write)
 88  *    ->mmap_sem                (fault_in_pages_readable->do_page_fault)
 89  *
 90  *  bdi->wb.list_lock
 91  *    sb_lock                   (fs/fs-writeback.c)
 92  *    ->i_pages lock            (__sync_single_inode)
 93  *
 94  *  ->i_mmap_rwsem
 95  *    ->anon_vma.lock           (vma_adjust)
 96  *
 97  *  ->anon_vma.lock
 98  *    ->page_table_lock or pte_lock     (anon_vma_prepare and various)
 99  *
100  *  ->page_table_lock or pte_lock
101  *    ->swap_lock               (try_to_unmap_one)
102  *    ->private_lock            (try_to_unmap_one)
103  *    ->i_pages lock            (try_to_unmap_one)
104  *    ->pgdat->lru_lock         (follow_page->mark_page_accessed)
105  *    ->pgdat->lru_lock         (check_pte_range->isolate_lru_page)
106  *    ->private_lock            (page_remove_rmap->set_page_dirty)
107  *    ->i_pages lock            (page_remove_rmap->set_page_dirty)
108  *    bdi.wb->list_lock         (page_remove_rmap->set_page_dirty)
109  *    ->inode->i_lock           (page_remove_rmap->set_page_dirty)
110  *    ->memcg->move_lock        (page_remove_rmap->lock_page_memcg)
111  *    bdi.wb->list_lock         (zap_pte_range->set_page_dirty)
112  *    ->inode->i_lock           (zap_pte_range->set_page_dirty)
113  *    ->private_lock            (zap_pte_range->__set_page_dirty_buffers)
114  *
115  * ->i_mmap_rwsem
116  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
117  */
118 
119 static void page_cache_delete(struct address_space *mapping,
120                                    struct page *page, void *shadow)
121 {
122         XA_STATE(xas, &mapping->i_pages, page->index);
123         unsigned int nr = 1;
124 
125         mapping_set_update(&xas, mapping);
126 
127         /* hugetlb pages are represented by a single entry in the xarray */
128         if (!PageHuge(page)) {
129                 xas_set_order(&xas, page->index, compound_order(page));
130                 nr = compound_nr(page);
131         }
132 
133         VM_BUG_ON_PAGE(!PageLocked(page), page);
134         VM_BUG_ON_PAGE(PageTail(page), page);
135         VM_BUG_ON_PAGE(nr != 1 && shadow, page);
136 
137         xas_store(&xas, shadow);
138         xas_init_marks(&xas);
139 
140         page->mapping = NULL;
141         /* Leave page->index set: truncation lookup relies upon it */
142 
143         if (shadow) {
144                 mapping->nrexceptional += nr;
145                 /*
146                  * Make sure the nrexceptional update is committed before
147                  * the nrpages update so that final truncate racing
148                  * with reclaim does not see both counters 0 at the
149                  * same time and miss a shadow entry.
150                  */
151                 smp_wmb();
152         }
153         mapping->nrpages -= nr;
154 }
155 
156 static void unaccount_page_cache_page(struct address_space *mapping,
157                                       struct page *page)
158 {
159         int nr;
160 
161         /*
162          * if we're uptodate, flush out into the cleancache, otherwise
163          * invalidate any existing cleancache entries.  We can't leave
164          * stale data around in the cleancache once our page is gone
165          */
166         if (PageUptodate(page) && PageMappedToDisk(page))
167                 cleancache_put_page(page);
168         else
169                 cleancache_invalidate_page(mapping, page);
170 
171         VM_BUG_ON_PAGE(PageTail(page), page);
172         VM_BUG_ON_PAGE(page_mapped(page), page);
173         if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
174                 int mapcount;
175 
176                 pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
177                          current->comm, page_to_pfn(page));
178                 dump_page(page, "still mapped when deleted");
179                 dump_stack();
180                 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
181 
182                 mapcount = page_mapcount(page);
183                 if (mapping_exiting(mapping) &&
184                     page_count(page) >= mapcount + 2) {
185                         /*
186                          * All vmas have already been torn down, so it's
187                          * a good bet that actually the page is unmapped,
188                          * and we'd prefer not to leak it: if we're wrong,
189                          * some other bad page check should catch it later.
190                          */
191                         page_mapcount_reset(page);
192                         page_ref_sub(page, mapcount);
193                 }
194         }
195 
196         /* hugetlb pages do not participate in page cache accounting. */
197         if (PageHuge(page))
198                 return;
199 
200         nr = hpage_nr_pages(page);
201 
202         __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
203         if (PageSwapBacked(page)) {
204                 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
205                 if (PageTransHuge(page))
206                         __dec_node_page_state(page, NR_SHMEM_THPS);
207         } else if (PageTransHuge(page)) {
208                 __dec_node_page_state(page, NR_FILE_THPS);
209                 filemap_nr_thps_dec(mapping);
210         }
211 
212         /*
213          * At this point page must be either written or cleaned by
214          * truncate.  Dirty page here signals a bug and loss of
215          * unwritten data.
216          *
217          * This fixes dirty accounting after removing the page entirely
218          * but leaves PageDirty set: it has no effect for truncated
219          * page and anyway will be cleared before returning page into
220          * buddy allocator.
221          */
222         if (WARN_ON_ONCE(PageDirty(page)))
223                 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
224 }
225 
226 /*
227  * Delete a page from the page cache and free it. Caller has to make
228  * sure the page is locked and that nobody else uses it - or that usage
229  * is safe.  The caller must hold the i_pages lock.
230  */
231 void __delete_from_page_cache(struct page *page, void *shadow)
232 {
233         struct address_space *mapping = page->mapping;
234 
235         trace_mm_filemap_delete_from_page_cache(page);
236 
237         unaccount_page_cache_page(mapping, page);
238         page_cache_delete(mapping, page, shadow);
239 }
240 
241 static void page_cache_free_page(struct address_space *mapping,
242                                 struct page *page)
243 {
244         void (*freepage)(struct page *);
245 
246         freepage = mapping->a_ops->freepage;
247         if (freepage)
248                 freepage(page);
249 
250         if (PageTransHuge(page) && !PageHuge(page)) {
251                 page_ref_sub(page, HPAGE_PMD_NR);
252                 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
253         } else {
254                 put_page(page);
255         }
256 }
257 
258 /**
259  * delete_from_page_cache - delete page from page cache
260  * @page: the page which the kernel is trying to remove from page cache
261  *
262  * This must be called only on pages that have been verified to be in the page
263  * cache and locked.  It will never put the page into the free list, the caller
264  * has a reference on the page.
265  */
266 void delete_from_page_cache(struct page *page)
267 {
268         struct address_space *mapping = page_mapping(page);
269         unsigned long flags;
270 
271         BUG_ON(!PageLocked(page));
272         xa_lock_irqsave(&mapping->i_pages, flags);
273         __delete_from_page_cache(page, NULL);
274         xa_unlock_irqrestore(&mapping->i_pages, flags);
275 
276         page_cache_free_page(mapping, page);
277 }
278 EXPORT_SYMBOL(delete_from_page_cache);
279 
280 /*
281  * page_cache_delete_batch - delete several pages from page cache
282  * @mapping: the mapping to which pages belong
283  * @pvec: pagevec with pages to delete
284  *
285  * The function walks over mapping->i_pages and removes pages passed in @pvec
286  * from the mapping. The function expects @pvec to be sorted by page index
287  * and is optimised for it to be dense.
288  * It tolerates holes in @pvec (mapping entries at those indices are not
289  * modified). The function expects only THP head pages to be present in the
290  * @pvec.
291  *
292  * The function expects the i_pages lock to be held.
293  */
294 static void page_cache_delete_batch(struct address_space *mapping,
295                              struct pagevec *pvec)
296 {
297         XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
298         int total_pages = 0;
299         int i = 0;
300         struct page *page;
301 
302         mapping_set_update(&xas, mapping);
303         xas_for_each(&xas, page, ULONG_MAX) {
304                 if (i >= pagevec_count(pvec))
305                         break;
306 
307                 /* A swap/dax/shadow entry got inserted? Skip it. */
308                 if (xa_is_value(page))
309                         continue;
310                 /*
311                  * A page got inserted in our range? Skip it. We have our
312                  * pages locked so they are protected from being removed.
313                  * If we see a page whose index is higher than ours, it
314                  * means our page has been removed, which shouldn't be
315                  * possible because we're holding the PageLock.
316                  */
317                 if (page != pvec->pages[i]) {
318                         VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
319                                         page);
320                         continue;
321                 }
322 
323                 WARN_ON_ONCE(!PageLocked(page));
324 
325                 if (page->index == xas.xa_index)
326                         page->mapping = NULL;
327                 /* Leave page->index set: truncation lookup relies on it */
328 
329                 /*
330                  * Move to the next page in the vector if this is a regular
331                  * page or the index is of the last sub-page of this compound
332                  * page.
333                  */
334                 if (page->index + compound_nr(page) - 1 == xas.xa_index)
335                         i++;
336                 xas_store(&xas, NULL);
337                 total_pages++;
338         }
339         mapping->nrpages -= total_pages;
340 }
341 
342 void delete_from_page_cache_batch(struct address_space *mapping,
343                                   struct pagevec *pvec)
344 {
345         int i;
346         unsigned long flags;
347 
348         if (!pagevec_count(pvec))
349                 return;
350 
351         xa_lock_irqsave(&mapping->i_pages, flags);
352         for (i = 0; i < pagevec_count(pvec); i++) {
353                 trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
354 
355                 unaccount_page_cache_page(mapping, pvec->pages[i]);
356         }
357         page_cache_delete_batch(mapping, pvec);
358         xa_unlock_irqrestore(&mapping->i_pages, flags);
359 
360         for (i = 0; i < pagevec_count(pvec); i++)
361                 page_cache_free_page(mapping, pvec->pages[i]);
362 }
363 
364 int filemap_check_errors(struct address_space *mapping)
365 {
366         int ret = 0;
367         /* Check for outstanding write errors */
368         if (test_bit(AS_ENOSPC, &mapping->flags) &&
369             test_and_clear_bit(AS_ENOSPC, &mapping->flags))
370                 ret = -ENOSPC;
371         if (test_bit(AS_EIO, &mapping->flags) &&
372             test_and_clear_bit(AS_EIO, &mapping->flags))
373                 ret = -EIO;
374         return ret;
375 }
376 EXPORT_SYMBOL(filemap_check_errors);
377 
378 static int filemap_check_and_keep_errors(struct address_space *mapping)
379 {
380         /* Check for outstanding write errors */
381         if (test_bit(AS_EIO, &mapping->flags))
382                 return -EIO;
383         if (test_bit(AS_ENOSPC, &mapping->flags))
384                 return -ENOSPC;
385         return 0;
386 }
387 
388 /**
389  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
390  * @mapping:    address space structure to write
391  * @start:      offset in bytes where the range starts
392  * @end:        offset in bytes where the range ends (inclusive)
393  * @sync_mode:  enable synchronous operation
394  *
395  * Start writeback against all of a mapping's dirty pages that lie
396  * within the byte offsets <start, end> inclusive.
397  *
398  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
399  * opposed to a regular memory cleansing writeback.  The difference between
400  * these two operations is that if a dirty page/buffer is encountered, it must
401  * be waited upon, and not just skipped over.
402  *
403  * Return: %0 on success, negative error code otherwise.
404  */
405 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
406                                 loff_t end, int sync_mode)
407 {
408         int ret;
409         struct writeback_control wbc = {
410                 .sync_mode = sync_mode,
411                 .nr_to_write = LONG_MAX,
412                 .range_start = start,
413                 .range_end = end,
414         };
415 
416         if (!mapping_cap_writeback_dirty(mapping) ||
417             !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
418                 return 0;
419 
420         wbc_attach_fdatawrite_inode(&wbc, mapping->host);
421         ret = do_writepages(mapping, &wbc);
422         wbc_detach_inode(&wbc);
423         return ret;
424 }
425 
426 static inline int __filemap_fdatawrite(struct address_space *mapping,
427         int sync_mode)
428 {
429         return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
430 }
431 
432 int filemap_fdatawrite(struct address_space *mapping)
433 {
434         return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435 }
436 EXPORT_SYMBOL(filemap_fdatawrite);
437 
438 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
439                                 loff_t end)
440 {
441         return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442 }
443 EXPORT_SYMBOL(filemap_fdatawrite_range);
444 
445 /**
446  * filemap_flush - mostly a non-blocking flush
447  * @mapping:    target address_space
448  *
449  * This is a mostly non-blocking flush.  Not suitable for data-integrity
450  * purposes - I/O may not be started against all dirty pages.
451  *
452  * Return: %0 on success, negative error code otherwise.
453  */
454 int filemap_flush(struct address_space *mapping)
455 {
456         return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457 }
458 EXPORT_SYMBOL(filemap_flush);
459 
460 /**
461  * filemap_range_has_page - check if a page exists in range.
462  * @mapping:           address space within which to check
463  * @start_byte:        offset in bytes where the range starts
464  * @end_byte:          offset in bytes where the range ends (inclusive)
465  *
466  * Find at least one page in the range supplied, usually used to check if
467  * direct writing in this range will trigger a writeback.
468  *
469  * Return: %true if at least one page exists in the specified range,
470  * %false otherwise.
471  */
472 bool filemap_range_has_page(struct address_space *mapping,
473                            loff_t start_byte, loff_t end_byte)
474 {
475         struct page *page;
476         XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477         pgoff_t max = end_byte >> PAGE_SHIFT;
478 
479         if (end_byte < start_byte)
480                 return false;
481 
482         rcu_read_lock();
483         for (;;) {
484                 page = xas_find(&xas, max);
485                 if (xas_retry(&xas, page))
486                         continue;
487                 /* Shadow entries don't count */
488                 if (xa_is_value(page))
489                         continue;
490                 /*
491                  * We don't need to try to pin this page; we're about to
492                  * release the RCU lock anyway.  It is enough to know that
493                  * there was a page here recently.
494                  */
495                 break;
496         }
497         rcu_read_unlock();
498 
499         return page != NULL;
500 }
501 EXPORT_SYMBOL(filemap_range_has_page);
502 
503 static void __filemap_fdatawait_range(struct address_space *mapping,
504                                      loff_t start_byte, loff_t end_byte)
505 {
506         pgoff_t index = start_byte >> PAGE_SHIFT;
507         pgoff_t end = end_byte >> PAGE_SHIFT;
508         struct pagevec pvec;
509         int nr_pages;
510 
511         if (end_byte < start_byte)
512                 return;
513 
514         pagevec_init(&pvec);
515         while (index <= end) {
516                 unsigned i;
517 
518                 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
519                                 end, PAGECACHE_TAG_WRITEBACK);
520                 if (!nr_pages)
521                         break;
522 
523                 for (i = 0; i < nr_pages; i++) {
524                         struct page *page = pvec.pages[i];
525 
526                         wait_on_page_writeback(page);
527                         ClearPageError(page);
528                 }
529                 pagevec_release(&pvec);
530                 cond_resched();
531         }
532 }
533 
534 /**
535  * filemap_fdatawait_range - wait for writeback to complete
536  * @mapping:            address space structure to wait for
537  * @start_byte:         offset in bytes where the range starts
538  * @end_byte:           offset in bytes where the range ends (inclusive)
539  *
540  * Walk the list of under-writeback pages of the given address space
541  * in the given range and wait for all of them.  Check error status of
542  * the address space and return it.
543  *
544  * Since the error status of the address space is cleared by this function,
545  * callers are responsible for checking the return value and handling and/or
546  * reporting the error.
547  *
548  * Return: error status of the address space.
549  */
550 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
551                             loff_t end_byte)
552 {
553         __filemap_fdatawait_range(mapping, start_byte, end_byte);
554         return filemap_check_errors(mapping);
555 }
556 EXPORT_SYMBOL(filemap_fdatawait_range);
557 
558 /**
559  * filemap_fdatawait_range_keep_errors - wait for writeback to complete
560  * @mapping:            address space structure to wait for
561  * @start_byte:         offset in bytes where the range starts
562  * @end_byte:           offset in bytes where the range ends (inclusive)
563  *
564  * Walk the list of under-writeback pages of the given address space in the
565  * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
566  * this function does not clear error status of the address space.
567  *
568  * Use this function if callers don't handle errors themselves.  Expected
569  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
570  * fsfreeze(8)
571  */
572 int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
573                 loff_t start_byte, loff_t end_byte)
574 {
575         __filemap_fdatawait_range(mapping, start_byte, end_byte);
576         return filemap_check_and_keep_errors(mapping);
577 }
578 EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
579 
580 /**
581  * file_fdatawait_range - wait for writeback to complete
582  * @file:               file pointing to address space structure to wait for
583  * @start_byte:         offset in bytes where the range starts
584  * @end_byte:           offset in bytes where the range ends (inclusive)
585  *
586  * Walk the list of under-writeback pages of the address space that file
587  * refers to, in the given range and wait for all of them.  Check error
588  * status of the address space vs. the file->f_wb_err cursor and return it.
589  *
590  * Since the error status of the file is advanced by this function,
591  * callers are responsible for checking the return value and handling and/or
592  * reporting the error.
593  *
594  * Return: error status of the address space vs. the file->f_wb_err cursor.
595  */
596 int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
597 {
598         struct address_space *mapping = file->f_mapping;
599 
600         __filemap_fdatawait_range(mapping, start_byte, end_byte);
601         return file_check_and_advance_wb_err(file);
602 }
603 EXPORT_SYMBOL(file_fdatawait_range);
604 
605 /**
606  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
607  * @mapping: address space structure to wait for
608  *
609  * Walk the list of under-writeback pages of the given address space
610  * and wait for all of them.  Unlike filemap_fdatawait(), this function
611  * does not clear error status of the address space.
612  *
613  * Use this function if callers don't handle errors themselves.  Expected
614  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
615  * fsfreeze(8)
616  *
617  * Return: error status of the address space.
618  */
619 int filemap_fdatawait_keep_errors(struct address_space *mapping)
620 {
621         __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
622         return filemap_check_and_keep_errors(mapping);
623 }
624 EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
625 
626 /* Returns true if writeback might be needed or already in progress. */
627 static bool mapping_needs_writeback(struct address_space *mapping)
628 {
629         if (dax_mapping(mapping))
630                 return mapping->nrexceptional;
631 
632         return mapping->nrpages;
633 }
634 
635 /**
636  * filemap_write_and_wait_range - write out & wait on a file range
637  * @mapping:    the address_space for the pages
638  * @lstart:     offset in bytes where the range starts
639  * @lend:       offset in bytes where the range ends (inclusive)
640  *
641  * Write out and wait upon file offsets lstart->lend, inclusive.
642  *
643  * Note that @lend is inclusive (describes the last byte to be written) so
644  * that this function can be used to write to the very end-of-file (end = -1).
645  *
646  * Return: error status of the address space.
647  */
648 int filemap_write_and_wait_range(struct address_space *mapping,
649                                  loff_t lstart, loff_t lend)
650 {
651         int err = 0;
652 
653         if (mapping_needs_writeback(mapping)) {
654                 err = __filemap_fdatawrite_range(mapping, lstart, lend,
655                                                  WB_SYNC_ALL);
656                 /*
657                  * Even if the above returned error, the pages may be
658                  * written partially (e.g. -ENOSPC), so we wait for it.
659                  * But the -EIO is special case, it may indicate the worst
660                  * thing (e.g. bug) happened, so we avoid waiting for it.
661                  */
662                 if (err != -EIO) {
663                         int err2 = filemap_fdatawait_range(mapping,
664                                                 lstart, lend);
665                         if (!err)
666                                 err = err2;
667                 } else {
668                         /* Clear any previously stored errors */
669                         filemap_check_errors(mapping);
670                 }
671         } else {
672                 err = filemap_check_errors(mapping);
673         }
674         return err;
675 }
676 EXPORT_SYMBOL(filemap_write_and_wait_range);
677 
678 void __filemap_set_wb_err(struct address_space *mapping, int err)
679 {
680         errseq_t eseq = errseq_set(&mapping->wb_err, err);
681 
682         trace_filemap_set_wb_err(mapping, eseq);
683 }
684 EXPORT_SYMBOL(__filemap_set_wb_err);
685 
686 /**
687  * file_check_and_advance_wb_err - report wb error (if any) that was previously
688  *                                 and advance wb_err to current one
689  * @file: struct file on which the error is being reported
690  *
691  * When userland calls fsync (or something like nfsd does the equivalent), we
692  * want to report any writeback errors that occurred since the last fsync (or
693  * since the file was opened if there haven't been any).
694  *
695  * Grab the wb_err from the mapping. If it matches what we have in the file,
696  * then just quickly return 0. The file is all caught up.
697  *
698  * If it doesn't match, then take the mapping value, set the "seen" flag in
699  * it and try to swap it into place. If it works, or another task beat us
700  * to it with the new value, then update the f_wb_err and return the error
701  * portion. The error at this point must be reported via proper channels
702  * (a'la fsync, or NFS COMMIT operation, etc.).
703  *
704  * While we handle mapping->wb_err with atomic operations, the f_wb_err
705  * value is protected by the f_lock since we must ensure that it reflects
706  * the latest value swapped in for this file descriptor.
707  *
708  * Return: %0 on success, negative error code otherwise.
709  */
710 int file_check_and_advance_wb_err(struct file *file)
711 {
712         int err = 0;
713         errseq_t old = READ_ONCE(file->f_wb_err);
714         struct address_space *mapping = file->f_mapping;
715 
716         /* Locklessly handle the common case where nothing has changed */
717         if (errseq_check(&mapping->wb_err, old)) {
718                 /* Something changed, must use slow path */
719                 spin_lock(&file->f_lock);
720                 old = file->f_wb_err;
721                 err = errseq_check_and_advance(&mapping->wb_err,
722                                                 &file->f_wb_err);
723                 trace_file_check_and_advance_wb_err(file, old);
724                 spin_unlock(&file->f_lock);
725         }
726 
727         /*
728          * We're mostly using this function as a drop in replacement for
729          * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
730          * that the legacy code would have had on these flags.
731          */
732         clear_bit(AS_EIO, &mapping->flags);
733         clear_bit(AS_ENOSPC, &mapping->flags);
734         return err;
735 }
736 EXPORT_SYMBOL(file_check_and_advance_wb_err);
737 
738 /**
739  * file_write_and_wait_range - write out & wait on a file range
740  * @file:       file pointing to address_space with pages
741  * @lstart:     offset in bytes where the range starts
742  * @lend:       offset in bytes where the range ends (inclusive)
743  *
744  * Write out and wait upon file offsets lstart->lend, inclusive.
745  *
746  * Note that @lend is inclusive (describes the last byte to be written) so
747  * that this function can be used to write to the very end-of-file (end = -1).
748  *
749  * After writing out and waiting on the data, we check and advance the
750  * f_wb_err cursor to the latest value, and return any errors detected there.
751  *
752  * Return: %0 on success, negative error code otherwise.
753  */
754 int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
755 {
756         int err = 0, err2;
757         struct address_space *mapping = file->f_mapping;
758 
759         if (mapping_needs_writeback(mapping)) {
760                 err = __filemap_fdatawrite_range(mapping, lstart, lend,
761                                                  WB_SYNC_ALL);
762                 /* See comment of filemap_write_and_wait() */
763                 if (err != -EIO)
764                         __filemap_fdatawait_range(mapping, lstart, lend);
765         }
766         err2 = file_check_and_advance_wb_err(file);
767         if (!err)
768                 err = err2;
769         return err;
770 }
771 EXPORT_SYMBOL(file_write_and_wait_range);
772 
773 /**
774  * replace_page_cache_page - replace a pagecache page with a new one
775  * @old:        page to be replaced
776  * @new:        page to replace with
777  * @gfp_mask:   allocation mode
778  *
779  * This function replaces a page in the pagecache with a new one.  On
780  * success it acquires the pagecache reference for the new page and
781  * drops it for the old page.  Both the old and new pages must be
782  * locked.  This function does not add the new page to the LRU, the
783  * caller must do that.
784  *
785  * The remove + add is atomic.  This function cannot fail.
786  *
787  * Return: %0
788  */
789 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
790 {
791         struct address_space *mapping = old->mapping;
792         void (*freepage)(struct page *) = mapping->a_ops->freepage;
793         pgoff_t offset = old->index;
794         XA_STATE(xas, &mapping->i_pages, offset);
795         unsigned long flags;
796 
797         VM_BUG_ON_PAGE(!PageLocked(old), old);
798         VM_BUG_ON_PAGE(!PageLocked(new), new);
799         VM_BUG_ON_PAGE(new->mapping, new);
800 
801         get_page(new);
802         new->mapping = mapping;
803         new->index = offset;
804 
805         xas_lock_irqsave(&xas, flags);
806         xas_store(&xas, new);
807 
808         old->mapping = NULL;
809         /* hugetlb pages do not participate in page cache accounting. */
810         if (!PageHuge(old))
811                 __dec_node_page_state(new, NR_FILE_PAGES);
812         if (!PageHuge(new))
813                 __inc_node_page_state(new, NR_FILE_PAGES);
814         if (PageSwapBacked(old))
815                 __dec_node_page_state(new, NR_SHMEM);
816         if (PageSwapBacked(new))
817                 __inc_node_page_state(new, NR_SHMEM);
818         xas_unlock_irqrestore(&xas, flags);
819         mem_cgroup_migrate(old, new);
820         if (freepage)
821                 freepage(old);
822         put_page(old);
823 
824         return 0;
825 }
826 EXPORT_SYMBOL_GPL(replace_page_cache_page);
827 
828 static int __add_to_page_cache_locked(struct page *page,
829                                       struct address_space *mapping,
830                                       pgoff_t offset, gfp_t gfp_mask,
831                                       void **shadowp)
832 {
833         XA_STATE(xas, &mapping->i_pages, offset);
834         int huge = PageHuge(page);
835         struct mem_cgroup *memcg;
836         int error;
837         void *old;
838 
839         VM_BUG_ON_PAGE(!PageLocked(page), page);
840         VM_BUG_ON_PAGE(PageSwapBacked(page), page);
841         mapping_set_update(&xas, mapping);
842 
843         if (!huge) {
844                 error = mem_cgroup_try_charge(page, current->mm,
845                                               gfp_mask, &memcg, false);
846                 if (error)
847                         return error;
848         }
849 
850         get_page(page);
851         page->mapping = mapping;
852         page->index = offset;
853 
854         do {
855                 xas_lock_irq(&xas);
856                 old = xas_load(&xas);
857                 if (old && !xa_is_value(old))
858                         xas_set_err(&xas, -EEXIST);
859                 xas_store(&xas, page);
860                 if (xas_error(&xas))
861                         goto unlock;
862 
863                 if (xa_is_value(old)) {
864                         mapping->nrexceptional--;
865                         if (shadowp)
866                                 *shadowp = old;
867                 }
868                 mapping->nrpages++;
869 
870                 /* hugetlb pages do not participate in page cache accounting */
871                 if (!huge)
872                         __inc_node_page_state(page, NR_FILE_PAGES);
873 unlock:
874                 xas_unlock_irq(&xas);
875         } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
876 
877         if (xas_error(&xas))
878                 goto error;
879 
880         if (!huge)
881                 mem_cgroup_commit_charge(page, memcg, false, false);
882         trace_mm_filemap_add_to_page_cache(page);
883         return 0;
884 error:
885         page->mapping = NULL;
886         /* Leave page->index set: truncation relies upon it */
887         if (!huge)
888                 mem_cgroup_cancel_charge(page, memcg, false);
889         put_page(page);
890         return xas_error(&xas);
891 }
892 ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
893 
894 /**
895  * add_to_page_cache_locked - add a locked page to the pagecache
896  * @page:       page to add
897  * @mapping:    the page's address_space
898  * @offset:     page index
899  * @gfp_mask:   page allocation mode
900  *
901  * This function is used to add a page to the pagecache. It must be locked.
902  * This function does not add the page to the LRU.  The caller must do that.
903  *
904  * Return: %0 on success, negative error code otherwise.
905  */
906 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
907                 pgoff_t offset, gfp_t gfp_mask)
908 {
909         return __add_to_page_cache_locked(page, mapping, offset,
910                                           gfp_mask, NULL);
911 }
912 EXPORT_SYMBOL(add_to_page_cache_locked);
913 
914 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
915                                 pgoff_t offset, gfp_t gfp_mask)
916 {
917         void *shadow = NULL;
918         int ret;
919 
920         __SetPageLocked(page);
921         ret = __add_to_page_cache_locked(page, mapping, offset,
922                                          gfp_mask, &shadow);
923         if (unlikely(ret))
924                 __ClearPageLocked(page);
925         else {
926                 /*
927                  * The page might have been evicted from cache only
928                  * recently, in which case it should be activated like
929                  * any other repeatedly accessed page.
930                  * The exception is pages getting rewritten; evicting other
931                  * data from the working set, only to cache data that will
932                  * get overwritten with something else, is a waste of memory.
933                  */
934                 WARN_ON_ONCE(PageActive(page));
935                 if (!(gfp_mask & __GFP_WRITE) && shadow)
936                         workingset_refault(page, shadow);
937                 lru_cache_add(page);
938         }
939         return ret;
940 }
941 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
942 
943 #ifdef CONFIG_NUMA
944 struct page *__page_cache_alloc(gfp_t gfp)
945 {
946         int n;
947         struct page *page;
948 
949         if (cpuset_do_page_mem_spread()) {
950                 unsigned int cpuset_mems_cookie;
951                 do {
952                         cpuset_mems_cookie = read_mems_allowed_begin();
953                         n = cpuset_mem_spread_node();
954                         page = __alloc_pages_node(n, gfp, 0);
955                 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
956 
957                 return page;
958         }
959         return alloc_pages(gfp, 0);
960 }
961 EXPORT_SYMBOL(__page_cache_alloc);
962 #endif
963 
964 /*
965  * In order to wait for pages to become available there must be
966  * waitqueues associated with pages. By using a hash table of
967  * waitqueues where the bucket discipline is to maintain all
968  * waiters on the same queue and wake all when any of the pages
969  * become available, and for the woken contexts to check to be
970  * sure the appropriate page became available, this saves space
971  * at a cost of "thundering herd" phenomena during rare hash
972  * collisions.
973  */
974 #define PAGE_WAIT_TABLE_BITS 8
975 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
976 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
977 
978 static wait_queue_head_t *page_waitqueue(struct page *page)
979 {
980         return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
981 }
982 
983 void __init pagecache_init(void)
984 {
985         int i;
986 
987         for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
988                 init_waitqueue_head(&page_wait_table[i]);
989 
990         page_writeback_init();
991 }
992 
993 /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
994 struct wait_page_key {
995         struct page *page;
996         int bit_nr;
997         int page_match;
998 };
999 
1000 struct wait_page_queue {
1001         struct page *page;
1002         int bit_nr;
1003         wait_queue_entry_t wait;
1004 };
1005 
1006 static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1007 {
1008         struct wait_page_key *key = arg;
1009         struct wait_page_queue *wait_page
1010                 = container_of(wait, struct wait_page_queue, wait);
1011 
1012         if (wait_page->page != key->page)
1013                return 0;
1014         key->page_match = 1;
1015 
1016         if (wait_page->bit_nr != key->bit_nr)
1017                 return 0;
1018 
1019         /*
1020          * Stop walking if it's locked.
1021          * Is this safe if put_and_wait_on_page_locked() is in use?
1022          * Yes: the waker must hold a reference to this page, and if PG_locked
1023          * has now already been set by another task, that task must also hold
1024          * a reference to the *same usage* of this page; so there is no need
1025          * to walk on to wake even the put_and_wait_on_page_locked() callers.
1026          */
1027         if (test_bit(key->bit_nr, &key->page->flags))
1028                 return -1;
1029 
1030         return autoremove_wake_function(wait, mode, sync, key);
1031 }
1032 
1033 static void wake_up_page_bit(struct page *page, int bit_nr)
1034 {
1035         wait_queue_head_t *q = page_waitqueue(page);
1036         struct wait_page_key key;
1037         unsigned long flags;
1038         wait_queue_entry_t bookmark;
1039 
1040         key.page = page;
1041         key.bit_nr = bit_nr;
1042         key.page_match = 0;
1043 
1044         bookmark.flags = 0;
1045         bookmark.private = NULL;
1046         bookmark.func = NULL;
1047         INIT_LIST_HEAD(&bookmark.entry);
1048 
1049         spin_lock_irqsave(&q->lock, flags);
1050         __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1051 
1052         while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1053                 /*
1054                  * Take a breather from holding the lock,
1055                  * allow pages that finish wake up asynchronously
1056                  * to acquire the lock and remove themselves
1057                  * from wait queue
1058                  */
1059                 spin_unlock_irqrestore(&q->lock, flags);
1060                 cpu_relax();
1061                 spin_lock_irqsave(&q->lock, flags);
1062                 __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1063         }
1064 
1065         /*
1066          * It is possible for other pages to have collided on the waitqueue
1067          * hash, so in that case check for a page match. That prevents a long-
1068          * term waiter
1069          *
1070          * It is still possible to miss a case here, when we woke page waiters
1071          * and removed them from the waitqueue, but there are still other
1072          * page waiters.
1073          */
1074         if (!waitqueue_active(q) || !key.page_match) {
1075                 ClearPageWaiters(page);
1076                 /*
1077                  * It's possible to miss clearing Waiters here, when we woke
1078                  * our page waiters, but the hashed waitqueue has waiters for
1079                  * other pages on it.
1080                  *
1081                  * That's okay, it's a rare case. The next waker will clear it.
1082                  */
1083         }
1084         spin_unlock_irqrestore(&q->lock, flags);
1085 }
1086 
1087 static void wake_up_page(struct page *page, int bit)
1088 {
1089         if (!PageWaiters(page))
1090                 return;
1091         wake_up_page_bit(page, bit);
1092 }
1093 
1094 /*
1095  * A choice of three behaviors for wait_on_page_bit_common():
1096  */
1097 enum behavior {
1098         EXCLUSIVE,      /* Hold ref to page and take the bit when woken, like
1099                          * __lock_page() waiting on then setting PG_locked.
1100                          */
1101         SHARED,         /* Hold ref to page and check the bit when woken, like
1102                          * wait_on_page_writeback() waiting on PG_writeback.
1103                          */
1104         DROP,           /* Drop ref to page before wait, no check when woken,
1105                          * like put_and_wait_on_page_locked() on PG_locked.
1106                          */
1107 };
1108 
1109 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1110         struct page *page, int bit_nr, int state, enum behavior behavior)
1111 {
1112         struct wait_page_queue wait_page;
1113         wait_queue_entry_t *wait = &wait_page.wait;
1114         bool bit_is_set;
1115         bool thrashing = false;
1116         bool delayacct = false;
1117         unsigned long pflags;
1118         int ret = 0;
1119 
1120         if (bit_nr == PG_locked &&
1121             !PageUptodate(page) && PageWorkingset(page)) {
1122                 if (!PageSwapBacked(page)) {
1123                         delayacct_thrashing_start();
1124                         delayacct = true;
1125                 }
1126                 psi_memstall_enter(&pflags);
1127                 thrashing = true;
1128         }
1129 
1130         init_wait(wait);
1131         wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1132         wait->func = wake_page_function;
1133         wait_page.page = page;
1134         wait_page.bit_nr = bit_nr;
1135 
1136         for (;;) {
1137                 spin_lock_irq(&q->lock);
1138 
1139                 if (likely(list_empty(&wait->entry))) {
1140                         __add_wait_queue_entry_tail(q, wait);
1141                         SetPageWaiters(page);
1142                 }
1143 
1144                 set_current_state(state);
1145 
1146                 spin_unlock_irq(&q->lock);
1147 
1148                 bit_is_set = test_bit(bit_nr, &page->flags);
1149                 if (behavior == DROP)
1150                         put_page(page);
1151 
1152                 if (likely(bit_is_set))
1153                         io_schedule();
1154 
1155                 if (behavior == EXCLUSIVE) {
1156                         if (!test_and_set_bit_lock(bit_nr, &page->flags))
1157                                 break;
1158                 } else if (behavior == SHARED) {
1159                         if (!test_bit(bit_nr, &page->flags))
1160                                 break;
1161                 }
1162 
1163                 if (signal_pending_state(state, current)) {
1164                         ret = -EINTR;
1165                         break;
1166                 }
1167 
1168                 if (behavior == DROP) {
1169                         /*
1170                          * We can no longer safely access page->flags:
1171                          * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1172                          * there is a risk of waiting forever on a page reused
1173                          * for something that keeps it locked indefinitely.
1174                          * But best check for -EINTR above before breaking.
1175                          */
1176                         break;
1177                 }
1178         }
1179 
1180         finish_wait(q, wait);
1181 
1182         if (thrashing) {
1183                 if (delayacct)
1184                         delayacct_thrashing_end();
1185                 psi_memstall_leave(&pflags);
1186         }
1187 
1188         /*
1189          * A signal could leave PageWaiters set. Clearing it here if
1190          * !waitqueue_active would be possible (by open-coding finish_wait),
1191          * but still fail to catch it in the case of wait hash collision. We
1192          * already can fail to clear wait hash collision cases, so don't
1193          * bother with signals either.
1194          */
1195 
1196         return ret;
1197 }
1198 
1199 void wait_on_page_bit(struct page *page, int bit_nr)
1200 {
1201         wait_queue_head_t *q = page_waitqueue(page);
1202         wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1203 }
1204 EXPORT_SYMBOL(wait_on_page_bit);
1205 
1206 int wait_on_page_bit_killable(struct page *page, int bit_nr)
1207 {
1208         wait_queue_head_t *q = page_waitqueue(page);
1209         return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1210 }
1211 EXPORT_SYMBOL(wait_on_page_bit_killable);
1212 
1213 /**
1214  * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1215  * @page: The page to wait for.
1216  *
1217  * The caller should hold a reference on @page.  They expect the page to
1218  * become unlocked relatively soon, but do not wish to hold up migration
1219  * (for example) by holding the reference while waiting for the page to
1220  * come unlocked.  After this function returns, the caller should not
1221  * dereference @page.
1222  */
1223 void put_and_wait_on_page_locked(struct page *page)
1224 {
1225         wait_queue_head_t *q;
1226 
1227         page = compound_head(page);
1228         q = page_waitqueue(page);
1229         wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1230 }
1231 
1232 /**
1233  * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1234  * @page: Page defining the wait queue of interest
1235  * @waiter: Waiter to add to the queue
1236  *
1237  * Add an arbitrary @waiter to the wait queue for the nominated @page.
1238  */
1239 void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1240 {
1241         wait_queue_head_t *q = page_waitqueue(page);
1242         unsigned long flags;
1243 
1244         spin_lock_irqsave(&q->lock, flags);
1245         __add_wait_queue_entry_tail(q, waiter);
1246         SetPageWaiters(page);
1247         spin_unlock_irqrestore(&q->lock, flags);
1248 }
1249 EXPORT_SYMBOL_GPL(add_page_wait_queue);
1250 
1251 #ifndef clear_bit_unlock_is_negative_byte
1252 
1253 /*
1254  * PG_waiters is the high bit in the same byte as PG_lock.
1255  *
1256  * On x86 (and on many other architectures), we can clear PG_lock and
1257  * test the sign bit at the same time. But if the architecture does
1258  * not support that special operation, we just do this all by hand
1259  * instead.
1260  *
1261  * The read of PG_waiters has to be after (or concurrently with) PG_locked
1262  * being cleared, but a memory barrier should be unneccssary since it is
1263  * in the same byte as PG_locked.
1264  */
1265 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1266 {
1267         clear_bit_unlock(nr, mem);
1268         /* smp_mb__after_atomic(); */
1269         return test_bit(PG_waiters, mem);
1270 }
1271 
1272 #endif
1273 
1274 /**
1275  * unlock_page - unlock a locked page
1276  * @page: the page
1277  *
1278  * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1279  * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1280  * mechanism between PageLocked pages and PageWriteback pages is shared.
1281  * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1282  *
1283  * Note that this depends on PG_waiters being the sign bit in the byte
1284  * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1285  * clear the PG_locked bit and test PG_waiters at the same time fairly
1286  * portably (architectures that do LL/SC can test any bit, while x86 can
1287  * test the sign bit).
1288  */
1289 void unlock_page(struct page *page)
1290 {
1291         BUILD_BUG_ON(PG_waiters != 7);
1292         page = compound_head(page);
1293         VM_BUG_ON_PAGE(!PageLocked(page), page);
1294         if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1295                 wake_up_page_bit(page, PG_locked);
1296 }
1297 EXPORT_SYMBOL(unlock_page);
1298 
1299 /**
1300  * end_page_writeback - end writeback against a page
1301  * @page: the page
1302  */
1303 void end_page_writeback(struct page *page)
1304 {
1305         /*
1306          * TestClearPageReclaim could be used here but it is an atomic
1307          * operation and overkill in this particular case. Failing to
1308          * shuffle a page marked for immediate reclaim is too mild to
1309          * justify taking an atomic operation penalty at the end of
1310          * ever page writeback.
1311          */
1312         if (PageReclaim(page)) {
1313                 ClearPageReclaim(page);
1314                 rotate_reclaimable_page(page);
1315         }
1316 
1317         if (!test_clear_page_writeback(page))
1318                 BUG();
1319 
1320         smp_mb__after_atomic();
1321         wake_up_page(page, PG_writeback);
1322 }
1323 EXPORT_SYMBOL(end_page_writeback);
1324 
1325 /*
1326  * After completing I/O on a page, call this routine to update the page
1327  * flags appropriately
1328  */
1329 void page_endio(struct page *page, bool is_write, int err)
1330 {
1331         if (!is_write) {
1332                 if (!err) {
1333                         SetPageUptodate(page);
1334                 } else {
1335                         ClearPageUptodate(page);
1336                         SetPageError(page);
1337                 }
1338                 unlock_page(page);
1339         } else {
1340                 if (err) {
1341                         struct address_space *mapping;
1342 
1343                         SetPageError(page);
1344                         mapping = page_mapping(page);
1345                         if (mapping)
1346                                 mapping_set_error(mapping, err);
1347                 }
1348                 end_page_writeback(page);
1349         }
1350 }
1351 EXPORT_SYMBOL_GPL(page_endio);
1352 
1353 /**
1354  * __lock_page - get a lock on the page, assuming we need to sleep to get it
1355  * @__page: the page to lock
1356  */
1357 void __lock_page(struct page *__page)
1358 {
1359         struct page *page = compound_head(__page);
1360         wait_queue_head_t *q = page_waitqueue(page);
1361         wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1362                                 EXCLUSIVE);
1363 }
1364 EXPORT_SYMBOL(__lock_page);
1365 
1366 int __lock_page_killable(struct page *__page)
1367 {
1368         struct page *page = compound_head(__page);
1369         wait_queue_head_t *q = page_waitqueue(page);
1370         return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1371                                         EXCLUSIVE);
1372 }
1373 EXPORT_SYMBOL_GPL(__lock_page_killable);
1374 
1375 /*
1376  * Return values:
1377  * 1 - page is locked; mmap_sem is still held.
1378  * 0 - page is not locked.
1379  *     mmap_sem has been released (up_read()), unless flags had both
1380  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1381  *     which case mmap_sem is still held.
1382  *
1383  * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1384  * with the page locked and the mmap_sem unperturbed.
1385  */
1386 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1387                          unsigned int flags)
1388 {
1389         if (flags & FAULT_FLAG_ALLOW_RETRY) {
1390                 /*
1391                  * CAUTION! In this case, mmap_sem is not released
1392                  * even though return 0.
1393                  */
1394                 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1395                         return 0;
1396 
1397                 up_read(&mm->mmap_sem);
1398                 if (flags & FAULT_FLAG_KILLABLE)
1399                         wait_on_page_locked_killable(page);
1400                 else
1401                         wait_on_page_locked(page);
1402                 return 0;
1403         } else {
1404                 if (flags & FAULT_FLAG_KILLABLE) {
1405                         int ret;
1406 
1407                         ret = __lock_page_killable(page);
1408                         if (ret) {
1409                                 up_read(&mm->mmap_sem);
1410                                 return 0;
1411                         }
1412                 } else
1413                         __lock_page(page);
1414                 return 1;
1415         }
1416 }
1417 
1418 /**
1419  * page_cache_next_miss() - Find the next gap in the page cache.
1420  * @mapping: Mapping.
1421  * @index: Index.
1422  * @max_scan: Maximum range to search.
1423  *
1424  * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1425  * gap with the lowest index.
1426  *
1427  * This function may be called under the rcu_read_lock.  However, this will
1428  * not atomically search a snapshot of the cache at a single point in time.
1429  * For example, if a gap is created at index 5, then subsequently a gap is
1430  * created at index 10, page_cache_next_miss covering both indices may
1431  * return 10 if called under the rcu_read_lock.
1432  *
1433  * Return: The index of the gap if found, otherwise an index outside the
1434  * range specified (in which case 'return - index >= max_scan' will be true).
1435  * In the rare case of index wrap-around, 0 will be returned.
1436  */
1437 pgoff_t page_cache_next_miss(struct address_space *mapping,
1438                              pgoff_t index, unsigned long max_scan)
1439 {
1440         XA_STATE(xas, &mapping->i_pages, index);
1441 
1442         while (max_scan--) {
1443                 void *entry = xas_next(&xas);
1444                 if (!entry || xa_is_value(entry))
1445                         break;
1446                 if (xas.xa_index == 0)
1447                         break;
1448         }
1449 
1450         return xas.xa_index;
1451 }
1452 EXPORT_SYMBOL(page_cache_next_miss);
1453 
1454 /**
1455  * page_cache_prev_miss() - Find the previous gap in the page cache.
1456  * @mapping: Mapping.
1457  * @index: Index.
1458  * @max_scan: Maximum range to search.
1459  *
1460  * Search the range [max(index - max_scan + 1, 0), index] for the
1461  * gap with the highest index.
1462  *
1463  * This function may be called under the rcu_read_lock.  However, this will
1464  * not atomically search a snapshot of the cache at a single point in time.
1465  * For example, if a gap is created at index 10, then subsequently a gap is
1466  * created at index 5, page_cache_prev_miss() covering both indices may
1467  * return 5 if called under the rcu_read_lock.
1468  *
1469  * Return: The index of the gap if found, otherwise an index outside the
1470  * range specified (in which case 'index - return >= max_scan' will be true).
1471  * In the rare case of wrap-around, ULONG_MAX will be returned.
1472  */
1473 pgoff_t page_cache_prev_miss(struct address_space *mapping,
1474                              pgoff_t index, unsigned long max_scan)
1475 {
1476         XA_STATE(xas, &mapping->i_pages, index);
1477 
1478         while (max_scan--) {
1479                 void *entry = xas_prev(&xas);
1480                 if (!entry || xa_is_value(entry))
1481                         break;
1482                 if (xas.xa_index == ULONG_MAX)
1483                         break;
1484         }
1485 
1486         return xas.xa_index;
1487 }
1488 EXPORT_SYMBOL(page_cache_prev_miss);
1489 
1490 /**
1491  * find_get_entry - find and get a page cache entry
1492  * @mapping: the address_space to search
1493  * @offset: the page cache index
1494  *
1495  * Looks up the page cache slot at @mapping & @offset.  If there is a
1496  * page cache page, it is returned with an increased refcount.
1497  *
1498  * If the slot holds a shadow entry of a previously evicted page, or a
1499  * swap entry from shmem/tmpfs, it is returned.
1500  *
1501  * Return: the found page or shadow entry, %NULL if nothing is found.
1502  */
1503 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1504 {
1505         XA_STATE(xas, &mapping->i_pages, offset);
1506         struct page *page;
1507 
1508         rcu_read_lock();
1509 repeat:
1510         xas_reset(&xas);
1511         page = xas_load(&xas);
1512         if (xas_retry(&xas, page))
1513                 goto repeat;
1514         /*
1515          * A shadow entry of a recently evicted page, or a swap entry from
1516          * shmem/tmpfs.  Return it without attempting to raise page count.
1517          */
1518         if (!page || xa_is_value(page))
1519                 goto out;
1520 
1521         if (!page_cache_get_speculative(page))
1522                 goto repeat;
1523 
1524         /*
1525          * Has the page moved or been split?
1526          * This is part of the lockless pagecache protocol. See
1527          * include/linux/pagemap.h for details.
1528          */
1529         if (unlikely(page != xas_reload(&xas))) {
1530                 put_page(page);
1531                 goto repeat;
1532         }
1533         page = find_subpage(page, offset);
1534 out:
1535         rcu_read_unlock();
1536 
1537         return page;
1538 }
1539 EXPORT_SYMBOL(find_get_entry);
1540 
1541 /**
1542  * find_lock_entry - locate, pin and lock a page cache entry
1543  * @mapping: the address_space to search
1544  * @offset: the page cache index
1545  *
1546  * Looks up the page cache slot at @mapping & @offset.  If there is a
1547  * page cache page, it is returned locked and with an increased
1548  * refcount.
1549  *
1550  * If the slot holds a shadow entry of a previously evicted page, or a
1551  * swap entry from shmem/tmpfs, it is returned.
1552  *
1553  * find_lock_entry() may sleep.
1554  *
1555  * Return: the found page or shadow entry, %NULL if nothing is found.
1556  */
1557 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1558 {
1559         struct page *page;
1560 
1561 repeat:
1562         page = find_get_entry(mapping, offset);
1563         if (page && !xa_is_value(page)) {
1564                 lock_page(page);
1565                 /* Has the page been truncated? */
1566                 if (unlikely(page_mapping(page) != mapping)) {
1567                         unlock_page(page);
1568                         put_page(page);
1569                         goto repeat;
1570                 }
1571                 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1572         }
1573         return page;
1574 }
1575 EXPORT_SYMBOL(find_lock_entry);
1576 
1577 /**
1578  * pagecache_get_page - find and get a page reference
1579  * @mapping: the address_space to search
1580  * @offset: the page index
1581  * @fgp_flags: PCG flags
1582  * @gfp_mask: gfp mask to use for the page cache data page allocation
1583  *
1584  * Looks up the page cache slot at @mapping & @offset.
1585  *
1586  * PCG flags modify how the page is returned.
1587  *
1588  * @fgp_flags can be:
1589  *
1590  * - FGP_ACCESSED: the page will be marked accessed
1591  * - FGP_LOCK: Page is return locked
1592  * - FGP_CREAT: If page is not present then a new page is allocated using
1593  *   @gfp_mask and added to the page cache and the VM's LRU
1594  *   list. The page is returned locked and with an increased
1595  *   refcount.
1596  * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1597  *   its own locking dance if the page is already in cache, or unlock the page
1598  *   before returning if we had to add the page to pagecache.
1599  *
1600  * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1601  * if the GFP flags specified for FGP_CREAT are atomic.
1602  *
1603  * If there is a page cache page, it is returned with an increased refcount.
1604  *
1605  * Return: the found page or %NULL otherwise.
1606  */
1607 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1608         int fgp_flags, gfp_t gfp_mask)
1609 {
1610         struct page *page;
1611 
1612 repeat:
1613         page = find_get_entry(mapping, offset);
1614         if (xa_is_value(page))
1615                 page = NULL;
1616         if (!page)
1617                 goto no_page;
1618 
1619         if (fgp_flags & FGP_LOCK) {
1620                 if (fgp_flags & FGP_NOWAIT) {
1621                         if (!trylock_page(page)) {
1622                                 put_page(page);
1623                                 return NULL;
1624                         }
1625                 } else {
1626                         lock_page(page);
1627                 }
1628 
1629                 /* Has the page been truncated? */
1630                 if (unlikely(compound_head(page)->mapping != mapping)) {
1631                         unlock_page(page);
1632                         put_page(page);
1633                         goto repeat;
1634                 }
1635                 VM_BUG_ON_PAGE(page->index != offset, page);
1636         }
1637 
1638         if (fgp_flags & FGP_ACCESSED)
1639                 mark_page_accessed(page);
1640 
1641 no_page:
1642         if (!page && (fgp_flags & FGP_CREAT)) {
1643                 int err;
1644                 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1645                         gfp_mask |= __GFP_WRITE;
1646                 if (fgp_flags & FGP_NOFS)
1647                         gfp_mask &= ~__GFP_FS;
1648 
1649                 page = __page_cache_alloc(gfp_mask);
1650                 if (!page)
1651                         return NULL;
1652 
1653                 if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1654                         fgp_flags |= FGP_LOCK;
1655 
1656                 /* Init accessed so avoid atomic mark_page_accessed later */
1657                 if (fgp_flags & FGP_ACCESSED)
1658                         __SetPageReferenced(page);
1659 
1660                 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1661                 if (unlikely(err)) {
1662                         put_page(page);
1663                         page = NULL;
1664                         if (err == -EEXIST)
1665                                 goto repeat;
1666                 }
1667 
1668                 /*
1669                  * add_to_page_cache_lru locks the page, and for mmap we expect
1670                  * an unlocked page.
1671                  */
1672                 if (page && (fgp_flags & FGP_FOR_MMAP))
1673                         unlock_page(page);
1674         }
1675 
1676         return page;
1677 }
1678 EXPORT_SYMBOL(pagecache_get_page);
1679 
1680 /**
1681  * find_get_entries - gang pagecache lookup
1682  * @mapping:    The address_space to search
1683  * @start:      The starting page cache index
1684  * @nr_entries: The maximum number of entries
1685  * @entries:    Where the resulting entries are placed
1686  * @indices:    The cache indices corresponding to the entries in @entries
1687  *
1688  * find_get_entries() will search for and return a group of up to
1689  * @nr_entries entries in the mapping.  The entries are placed at
1690  * @entries.  find_get_entries() takes a reference against any actual
1691  * pages it returns.
1692  *
1693  * The search returns a group of mapping-contiguous page cache entries
1694  * with ascending indexes.  There may be holes in the indices due to
1695  * not-present pages.
1696  *
1697  * Any shadow entries of evicted pages, or swap entries from
1698  * shmem/tmpfs, are included in the returned array.
1699  *
1700  * Return: the number of pages and shadow entries which were found.
1701  */
1702 unsigned find_get_entries(struct address_space *mapping,
1703                           pgoff_t start, unsigned int nr_entries,
1704                           struct page **entries, pgoff_t *indices)
1705 {
1706         XA_STATE(xas, &mapping->i_pages, start);
1707         struct page *page;
1708         unsigned int ret = 0;
1709 
1710         if (!nr_entries)
1711                 return 0;
1712 
1713         rcu_read_lock();
1714         xas_for_each(&xas, page, ULONG_MAX) {
1715                 if (xas_retry(&xas, page))
1716                         continue;
1717                 /*
1718                  * A shadow entry of a recently evicted page, a swap
1719                  * entry from shmem/tmpfs or a DAX entry.  Return it
1720                  * without attempting to raise page count.
1721                  */
1722                 if (xa_is_value(page))
1723                         goto export;
1724 
1725                 if (!page_cache_get_speculative(page))
1726                         goto retry;
1727 
1728                 /* Has the page moved or been split? */
1729                 if (unlikely(page != xas_reload(&xas)))
1730                         goto put_page;
1731                 page = find_subpage(page, xas.xa_index);
1732 
1733 export:
1734                 indices[ret] = xas.xa_index;
1735                 entries[ret] = page;
1736                 if (++ret == nr_entries)
1737                         break;
1738                 continue;
1739 put_page:
1740                 put_page(page);
1741 retry:
1742                 xas_reset(&xas);
1743         }
1744         rcu_read_unlock();
1745         return ret;
1746 }
1747 
1748 /**
1749  * find_get_pages_range - gang pagecache lookup
1750  * @mapping:    The address_space to search
1751  * @start:      The starting page index
1752  * @end:        The final page index (inclusive)
1753  * @nr_pages:   The maximum number of pages
1754  * @pages:      Where the resulting pages are placed
1755  *
1756  * find_get_pages_range() will search for and return a group of up to @nr_pages
1757  * pages in the mapping starting at index @start and up to index @end
1758  * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
1759  * a reference against the returned pages.
1760  *
1761  * The search returns a group of mapping-contiguous pages with ascending
1762  * indexes.  There may be holes in the indices due to not-present pages.
1763  * We also update @start to index the next page for the traversal.
1764  *
1765  * Return: the number of pages which were found. If this number is
1766  * smaller than @nr_pages, the end of specified range has been
1767  * reached.
1768  */
1769 unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1770                               pgoff_t end, unsigned int nr_pages,
1771                               struct page **pages)
1772 {
1773         XA_STATE(xas, &mapping->i_pages, *start);
1774         struct page *page;
1775         unsigned ret = 0;
1776 
1777         if (unlikely(!nr_pages))
1778                 return 0;
1779 
1780         rcu_read_lock();
1781         xas_for_each(&xas, page, end) {
1782                 if (xas_retry(&xas, page))
1783                         continue;
1784                 /* Skip over shadow, swap and DAX entries */
1785                 if (xa_is_value(page))
1786                         continue;
1787 
1788                 if (!page_cache_get_speculative(page))
1789                         goto retry;
1790 
1791                 /* Has the page moved or been split? */
1792                 if (unlikely(page != xas_reload(&xas)))
1793                         goto put_page;
1794 
1795                 pages[ret] = find_subpage(page, xas.xa_index);
1796                 if (++ret == nr_pages) {
1797                         *start = xas.xa_index + 1;
1798                         goto out;
1799                 }
1800                 continue;
1801 put_page:
1802                 put_page(page);
1803 retry:
1804                 xas_reset(&xas);
1805         }
1806 
1807         /*
1808          * We come here when there is no page beyond @end. We take care to not
1809          * overflow the index @start as it confuses some of the callers. This
1810          * breaks the iteration when there is a page at index -1 but that is
1811          * already broken anyway.
1812          */
1813         if (end == (pgoff_t)-1)
1814                 *start = (pgoff_t)-1;
1815         else
1816                 *start = end + 1;
1817 out:
1818         rcu_read_unlock();
1819 
1820         return ret;
1821 }
1822 
1823 /**
1824  * find_get_pages_contig - gang contiguous pagecache lookup
1825  * @mapping:    The address_space to search
1826  * @index:      The starting page index
1827  * @nr_pages:   The maximum number of pages
1828  * @pages:      Where the resulting pages are placed
1829  *
1830  * find_get_pages_contig() works exactly like find_get_pages(), except
1831  * that the returned number of pages are guaranteed to be contiguous.
1832  *
1833  * Return: the number of pages which were found.
1834  */
1835 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1836                                unsigned int nr_pages, struct page **pages)
1837 {
1838         XA_STATE(xas, &mapping->i_pages, index);
1839         struct page *page;
1840         unsigned int ret = 0;
1841 
1842         if (unlikely(!nr_pages))
1843                 return 0;
1844 
1845         rcu_read_lock();
1846         for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1847                 if (xas_retry(&xas, page))
1848                         continue;
1849                 /*
1850                  * If the entry has been swapped out, we can stop looking.
1851                  * No current caller is looking for DAX entries.
1852                  */
1853                 if (xa_is_value(page))
1854                         break;
1855 
1856                 if (!page_cache_get_speculative(page))
1857                         goto retry;
1858 
1859                 /* Has the page moved or been split? */
1860                 if (unlikely(page != xas_reload(&xas)))
1861                         goto put_page;
1862 
1863                 pages[ret] = find_subpage(page, xas.xa_index);
1864                 if (++ret == nr_pages)
1865                         break;
1866                 continue;
1867 put_page:
1868                 put_page(page);
1869 retry:
1870                 xas_reset(&xas);
1871         }
1872         rcu_read_unlock();
1873         return ret;
1874 }
1875 EXPORT_SYMBOL(find_get_pages_contig);
1876 
1877 /**
1878  * find_get_pages_range_tag - find and return pages in given range matching @tag
1879  * @mapping:    the address_space to search
1880  * @index:      the starting page index
1881  * @end:        The final page index (inclusive)
1882  * @tag:        the tag index
1883  * @nr_pages:   the maximum number of pages
1884  * @pages:      where the resulting pages are placed
1885  *
1886  * Like find_get_pages, except we only return pages which are tagged with
1887  * @tag.   We update @index to index the next page for the traversal.
1888  *
1889  * Return: the number of pages which were found.
1890  */
1891 unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1892                         pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1893                         struct page **pages)
1894 {
1895         XA_STATE(xas, &mapping->i_pages, *index);
1896         struct page *page;
1897         unsigned ret = 0;
1898 
1899         if (unlikely(!nr_pages))
1900                 return 0;
1901 
1902         rcu_read_lock();
1903         xas_for_each_marked(&xas, page, end, tag) {
1904                 if (xas_retry(&xas, page))
1905                         continue;
1906                 /*
1907                  * Shadow entries should never be tagged, but this iteration
1908                  * is lockless so there is a window for page reclaim to evict
1909                  * a page we saw tagged.  Skip over it.
1910                  */
1911                 if (xa_is_value(page))
1912                         continue;
1913 
1914                 if (!page_cache_get_speculative(page))
1915                         goto retry;
1916 
1917                 /* Has the page moved or been split? */
1918                 if (unlikely(page != xas_reload(&xas)))
1919                         goto put_page;
1920 
1921                 pages[ret] = find_subpage(page, xas.xa_index);
1922                 if (++ret == nr_pages) {
1923                         *index = xas.xa_index + 1;
1924                         goto out;
1925                 }
1926                 continue;
1927 put_page:
1928                 put_page(page);
1929 retry:
1930                 xas_reset(&xas);
1931         }
1932 
1933         /*
1934          * We come here when we got to @end. We take care to not overflow the
1935          * index @index as it confuses some of the callers. This breaks the
1936          * iteration when there is a page at index -1 but that is already
1937          * broken anyway.
1938          */
1939         if (end == (pgoff_t)-1)
1940                 *index = (pgoff_t)-1;
1941         else
1942                 *index = end + 1;
1943 out:
1944         rcu_read_unlock();
1945 
1946         return ret;
1947 }
1948 EXPORT_SYMBOL(find_get_pages_range_tag);
1949 
1950 /*
1951  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1952  * a _large_ part of the i/o request. Imagine the worst scenario:
1953  *
1954  *      ---R__________________________________________B__________
1955  *         ^ reading here                             ^ bad block(assume 4k)
1956  *
1957  * read(R) => miss => readahead(R...B) => media error => frustrating retries
1958  * => failing the whole request => read(R) => read(R+1) =>
1959  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1960  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1961  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1962  *
1963  * It is going insane. Fix it by quickly scaling down the readahead size.
1964  */
1965 static void shrink_readahead_size_eio(struct file *filp,
1966                                         struct file_ra_state *ra)
1967 {
1968         ra->ra_pages /= 4;
1969 }
1970 
1971 /**
1972  * generic_file_buffered_read - generic file read routine
1973  * @iocb:       the iocb to read
1974  * @iter:       data destination
1975  * @written:    already copied
1976  *
1977  * This is a generic file read routine, and uses the
1978  * mapping->a_ops->readpage() function for the actual low-level stuff.
1979  *
1980  * This is really ugly. But the goto's actually try to clarify some
1981  * of the logic when it comes to error handling etc.
1982  *
1983  * Return:
1984  * * total number of bytes copied, including those the were already @written
1985  * * negative error code if nothing was copied
1986  */
1987 static ssize_t generic_file_buffered_read(struct kiocb *iocb,
1988                 struct iov_iter *iter, ssize_t written)
1989 {
1990         struct file *filp = iocb->ki_filp;
1991         struct address_space *mapping = filp->f_mapping;
1992         struct inode *inode = mapping->host;
1993         struct file_ra_state *ra = &filp->f_ra;
1994         loff_t *ppos = &iocb->ki_pos;
1995         pgoff_t index;
1996         pgoff_t last_index;
1997         pgoff_t prev_index;
1998         unsigned long offset;      /* offset into pagecache page */
1999         unsigned int prev_offset;
2000         int error = 0;
2001 
2002         if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2003                 return 0;
2004         iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2005 
2006         index = *ppos >> PAGE_SHIFT;
2007         prev_index = ra->prev_pos >> PAGE_SHIFT;
2008         prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2009         last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2010         offset = *ppos & ~PAGE_MASK;
2011 
2012         for (;;) {
2013                 struct page *page;
2014                 pgoff_t end_index;
2015                 loff_t isize;
2016                 unsigned long nr, ret;
2017 
2018                 cond_resched();
2019 find_page:
2020                 if (fatal_signal_pending(current)) {
2021                         error = -EINTR;
2022                         goto out;
2023                 }
2024 
2025                 page = find_get_page(mapping, index);
2026                 if (!page) {
2027                         if (iocb->ki_flags & IOCB_NOWAIT)
2028                                 goto would_block;
2029                         page_cache_sync_readahead(mapping,
2030                                         ra, filp,
2031                                         index, last_index - index);
2032                         page = find_get_page(mapping, index);
2033                         if (unlikely(page == NULL))
2034                                 goto no_cached_page;
2035                 }
2036                 if (PageReadahead(page)) {
2037                         page_cache_async_readahead(mapping,
2038                                         ra, filp, page,
2039                                         index, last_index - index);
2040                 }
2041                 if (!PageUptodate(page)) {
2042                         if (iocb->ki_flags & IOCB_NOWAIT) {
2043                                 put_page(page);
2044                                 goto would_block;
2045                         }
2046 
2047                         /*
2048                          * See comment in do_read_cache_page on why
2049                          * wait_on_page_locked is used to avoid unnecessarily
2050                          * serialisations and why it's safe.
2051                          */
2052                         error = wait_on_page_locked_killable(page);
2053                         if (unlikely(error))
2054                                 goto readpage_error;
2055                         if (PageUptodate(page))
2056                                 goto page_ok;
2057 
2058                         if (inode->i_blkbits == PAGE_SHIFT ||
2059                                         !mapping->a_ops->is_partially_uptodate)
2060                                 goto page_not_up_to_date;
2061                         /* pipes can't handle partially uptodate pages */
2062                         if (unlikely(iov_iter_is_pipe(iter)))
2063                                 goto page_not_up_to_date;
2064                         if (!trylock_page(page))
2065                                 goto page_not_up_to_date;
2066                         /* Did it get truncated before we got the lock? */
2067                         if (!page->mapping)
2068                                 goto page_not_up_to_date_locked;
2069                         if (!mapping->a_ops->is_partially_uptodate(page,
2070                                                         offset, iter->count))
2071                                 goto page_not_up_to_date_locked;
2072                         unlock_page(page);
2073                 }
2074 page_ok:
2075                 /*
2076                  * i_size must be checked after we know the page is Uptodate.
2077                  *
2078                  * Checking i_size after the check allows us to calculate
2079                  * the correct value for "nr", which means the zero-filled
2080                  * part of the page is not copied back to userspace (unless
2081                  * another truncate extends the file - this is desired though).
2082                  */
2083 
2084                 isize = i_size_read(inode);
2085                 end_index = (isize - 1) >> PAGE_SHIFT;
2086                 if (unlikely(!isize || index > end_index)) {
2087                         put_page(page);
2088                         goto out;
2089                 }
2090 
2091                 /* nr is the maximum number of bytes to copy from this page */
2092                 nr = PAGE_SIZE;
2093                 if (index == end_index) {
2094                         nr = ((isize - 1) & ~PAGE_MASK) + 1;
2095                         if (nr <= offset) {
2096                                 put_page(page);
2097                                 goto out;
2098                         }
2099                 }
2100                 nr = nr - offset;
2101 
2102                 /* If users can be writing to this page using arbitrary
2103                  * virtual addresses, take care about potential aliasing
2104                  * before reading the page on the kernel side.
2105                  */
2106                 if (mapping_writably_mapped(mapping))
2107                         flush_dcache_page(page);
2108 
2109                 /*
2110                  * When a sequential read accesses a page several times,
2111                  * only mark it as accessed the first time.
2112                  */
2113                 if (prev_index != index || offset != prev_offset)
2114                         mark_page_accessed(page);
2115                 prev_index = index;
2116 
2117                 /*
2118                  * Ok, we have the page, and it's up-to-date, so
2119                  * now we can copy it to user space...
2120                  */
2121 
2122                 ret = copy_page_to_iter(page, offset, nr, iter);
2123                 offset += ret;
2124                 index += offset >> PAGE_SHIFT;
2125                 offset &= ~PAGE_MASK;
2126                 prev_offset = offset;
2127 
2128                 put_page(page);
2129                 written += ret;
2130                 if (!iov_iter_count(iter))
2131                         goto out;
2132                 if (ret < nr) {
2133                         error = -EFAULT;
2134                         goto out;
2135                 }
2136                 continue;
2137 
2138 page_not_up_to_date:
2139                 /* Get exclusive access to the page ... */
2140                 error = lock_page_killable(page);
2141                 if (unlikely(error))
2142                         goto readpage_error;
2143 
2144 page_not_up_to_date_locked:
2145                 /* Did it get truncated before we got the lock? */
2146                 if (!page->mapping) {
2147                         unlock_page(page);
2148                         put_page(page);
2149                         continue;
2150                 }
2151 
2152                 /* Did somebody else fill it already? */
2153                 if (PageUptodate(page)) {
2154                         unlock_page(page);
2155                         goto page_ok;
2156                 }
2157 
2158 readpage:
2159                 /*
2160                  * A previous I/O error may have been due to temporary
2161                  * failures, eg. multipath errors.
2162                  * PG_error will be set again if readpage fails.
2163                  */
2164                 ClearPageError(page);
2165                 /* Start the actual read. The read will unlock the page. */
2166                 error = mapping->a_ops->readpage(filp, page);
2167 
2168                 if (unlikely(error)) {
2169                         if (error == AOP_TRUNCATED_PAGE) {
2170                                 put_page(page);
2171                                 error = 0;
2172                                 goto find_page;
2173                         }
2174                         goto readpage_error;
2175                 }
2176 
2177                 if (!PageUptodate(page)) {
2178                         error = lock_page_killable(page);
2179                         if (unlikely(error))
2180                                 goto readpage_error;
2181                         if (!PageUptodate(page)) {
2182                                 if (page->mapping == NULL) {
2183                                         /*
2184                                          * invalidate_mapping_pages got it
2185                                          */
2186                                         unlock_page(page);
2187                                         put_page(page);
2188                                         goto find_page;
2189                                 }
2190                                 unlock_page(page);
2191                                 shrink_readahead_size_eio(filp, ra);
2192                                 error = -EIO;
2193                                 goto readpage_error;
2194                         }
2195                         unlock_page(page);
2196                 }
2197 
2198                 goto page_ok;
2199 
2200 readpage_error:
2201                 /* UHHUH! A synchronous read error occurred. Report it */
2202                 put_page(page);
2203                 goto out;
2204 
2205 no_cached_page:
2206                 /*
2207                  * Ok, it wasn't cached, so we need to create a new
2208                  * page..
2209                  */
2210                 page = page_cache_alloc(mapping);
2211                 if (!page) {
2212                         error = -ENOMEM;
2213                         goto out;
2214                 }
2215                 error = add_to_page_cache_lru(page, mapping, index,
2216                                 mapping_gfp_constraint(mapping, GFP_KERNEL));
2217                 if (error) {
2218                         put_page(page);
2219                         if (error == -EEXIST) {
2220                                 error = 0;
2221                                 goto find_page;
2222                         }
2223                         goto out;
2224                 }
2225                 goto readpage;
2226         }
2227 
2228 would_block:
2229         error = -EAGAIN;
2230 out:
2231         ra->prev_pos = prev_index;
2232         ra->prev_pos <<= PAGE_SHIFT;
2233         ra->prev_pos |= prev_offset;
2234 
2235         *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2236         file_accessed(filp);
2237         return written ? written : error;
2238 }
2239 
2240 /**
2241  * generic_file_read_iter - generic filesystem read routine
2242  * @iocb:       kernel I/O control block
2243  * @iter:       destination for the data read
2244  *
2245  * This is the "read_iter()" routine for all filesystems
2246  * that can use the page cache directly.
2247  * Return:
2248  * * number of bytes copied, even for partial reads
2249  * * negative error code if nothing was read
2250  */
2251 ssize_t
2252 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2253 {
2254         size_t count = iov_iter_count(iter);
2255         ssize_t retval = 0;
2256 
2257         if (!count)
2258                 goto out; /* skip atime */
2259 
2260         if (iocb->ki_flags & IOCB_DIRECT) {
2261                 struct file *file = iocb->ki_filp;
2262                 struct address_space *mapping = file->f_mapping;
2263                 struct inode *inode = mapping->host;
2264                 loff_t size;
2265 
2266                 size = i_size_read(inode);
2267                 if (iocb->ki_flags & IOCB_NOWAIT) {
2268                         if (filemap_range_has_page(mapping, iocb->ki_pos,
2269                                                    iocb->ki_pos + count - 1))
2270                                 return -EAGAIN;
2271                 } else {
2272                         retval = filemap_write_and_wait_range(mapping,
2273                                                 iocb->ki_pos,
2274                                                 iocb->ki_pos + count - 1);
2275                         if (retval < 0)
2276                                 goto out;
2277                 }
2278 
2279                 file_accessed(file);
2280 
2281                 retval = mapping->a_ops->direct_IO(iocb, iter);
2282                 if (retval >= 0) {
2283                         iocb->ki_pos += retval;
2284                         count -= retval;
2285                 }
2286                 iov_iter_revert(iter, count - iov_iter_count(iter));
2287 
2288                 /*
2289                  * Btrfs can have a short DIO read if we encounter
2290                  * compressed extents, so if there was an error, or if
2291                  * we've already read everything we wanted to, or if
2292                  * there was a short read because we hit EOF, go ahead
2293                  * and return.  Otherwise fallthrough to buffered io for
2294                  * the rest of the read.  Buffered reads will not work for
2295                  * DAX files, so don't bother trying.
2296                  */
2297                 if (retval < 0 || !count || iocb->ki_pos >= size ||
2298                     IS_DAX(inode))
2299                         goto out;
2300         }
2301 
2302         retval = generic_file_buffered_read(iocb, iter, retval);
2303 out:
2304         return retval;
2305 }
2306 EXPORT_SYMBOL(generic_file_read_iter);
2307 
2308 #ifdef CONFIG_MMU
2309 #define MMAP_LOTSAMISS  (100)
2310 /*
2311  * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2312  * @vmf - the vm_fault for this fault.
2313  * @page - the page to lock.
2314  * @fpin - the pointer to the file we may pin (or is already pinned).
2315  *
2316  * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2317  * It differs in that it actually returns the page locked if it returns 1 and 0
2318  * if it couldn't lock the page.  If we did have to drop the mmap_sem then fpin
2319  * will point to the pinned file and needs to be fput()'ed at a later point.
2320  */
2321 static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2322                                      struct file **fpin)
2323 {
2324         if (trylock_page(page))
2325                 return 1;
2326 
2327         /*
2328          * NOTE! This will make us return with VM_FAULT_RETRY, but with
2329          * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2330          * is supposed to work. We have way too many special cases..
2331          */
2332         if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2333                 return 0;
2334 
2335         *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2336         if (vmf->flags & FAULT_FLAG_KILLABLE) {
2337                 if (__lock_page_killable(page)) {
2338                         /*
2339                          * We didn't have the right flags to drop the mmap_sem,
2340                          * but all fault_handlers only check for fatal signals
2341                          * if we return VM_FAULT_RETRY, so we need to drop the
2342                          * mmap_sem here and return 0 if we don't have a fpin.
2343                          */
2344                         if (*fpin == NULL)
2345                                 up_read(&vmf->vma->vm_mm->mmap_sem);
2346                         return 0;
2347                 }
2348         } else
2349                 __lock_page(page);
2350         return 1;
2351 }
2352 
2353 
2354 /*
2355  * Synchronous readahead happens when we don't even find a page in the page
2356  * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2357  * to drop the mmap sem we return the file that was pinned in order for us to do
2358  * that.  If we didn't pin a file then we return NULL.  The file that is
2359  * returned needs to be fput()'ed when we're done with it.
2360  */
2361 static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2362 {
2363         struct file *file = vmf->vma->vm_file;
2364         struct file_ra_state *ra = &file->f_ra;
2365         struct address_space *mapping = file->f_mapping;
2366         struct file *fpin = NULL;
2367         pgoff_t offset = vmf->pgoff;
2368 
2369         /* If we don't want any read-ahead, don't bother */
2370         if (vmf->vma->vm_flags & VM_RAND_READ)
2371                 return fpin;
2372         if (!ra->ra_pages)
2373                 return fpin;
2374 
2375         if (vmf->vma->vm_flags & VM_SEQ_READ) {
2376                 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2377                 page_cache_sync_readahead(mapping, ra, file, offset,
2378                                           ra->ra_pages);
2379                 return fpin;
2380         }
2381 
2382         /* Avoid banging the cache line if not needed */
2383         if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2384                 ra->mmap_miss++;
2385 
2386         /*
2387          * Do we miss much more than hit in this file? If so,
2388          * stop bothering with read-ahead. It will only hurt.
2389          */
2390         if (ra->mmap_miss > MMAP_LOTSAMISS)
2391                 return fpin;
2392 
2393         /*
2394          * mmap read-around
2395          */
2396         fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2397         ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2398         ra->size = ra->ra_pages;
2399         ra->async_size = ra->ra_pages / 4;
2400         ra_submit(ra, mapping, file);
2401         return fpin;
2402 }
2403 
2404 /*
2405  * Asynchronous readahead happens when we find the page and PG_readahead,
2406  * so we want to possibly extend the readahead further.  We return the file that
2407  * was pinned if we have to drop the mmap_sem in order to do IO.
2408  */
2409 static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2410                                             struct page *page)
2411 {
2412         struct file *file = vmf->vma->vm_file;
2413         struct file_ra_state *ra = &file->f_ra;
2414         struct address_space *mapping = file->f_mapping;
2415         struct file *fpin = NULL;
2416         pgoff_t offset = vmf->pgoff;
2417 
2418         /* If we don't want any read-ahead, don't bother */
2419         if (vmf->vma->vm_flags & VM_RAND_READ)
2420                 return fpin;
2421         if (ra->mmap_miss > 0)
2422                 ra->mmap_miss--;
2423         if (PageReadahead(page)) {
2424                 fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2425                 page_cache_async_readahead(mapping, ra, file,
2426                                            page, offset, ra->ra_pages);
2427         }
2428         return fpin;
2429 }
2430 
2431 /**
2432  * filemap_fault - read in file data for page fault handling
2433  * @vmf:        struct vm_fault containing details of the fault
2434  *
2435  * filemap_fault() is invoked via the vma operations vector for a
2436  * mapped memory region to read in file data during a page fault.
2437  *
2438  * The goto's are kind of ugly, but this streamlines the normal case of having
2439  * it in the page cache, and handles the special cases reasonably without
2440  * having a lot of duplicated code.
2441  *
2442  * vma->vm_mm->mmap_sem must be held on entry.
2443  *
2444  * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
2445  * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2446  *
2447  * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2448  * has not been released.
2449  *
2450  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2451  *
2452  * Return: bitwise-OR of %VM_FAULT_ codes.
2453  */
2454 vm_fault_t filemap_fault(struct vm_fault *vmf)
2455 {
2456         int error;
2457         struct file *file = vmf->vma->vm_file;
2458         struct file *fpin = NULL;
2459         struct address_space *mapping = file->f_mapping;
2460         struct file_ra_state *ra = &file->f_ra;
2461         struct inode *inode = mapping->host;
2462         pgoff_t offset = vmf->pgoff;
2463         pgoff_t max_off;
2464         struct page *page;
2465         vm_fault_t ret = 0;
2466 
2467         max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2468         if (unlikely(offset >= max_off))
2469                 return VM_FAULT_SIGBUS;
2470 
2471         /*
2472          * Do we have something in the page cache already?
2473          */
2474         page = find_get_page(mapping, offset);
2475         if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2476                 /*
2477                  * We found the page, so try async readahead before
2478                  * waiting for the lock.
2479                  */
2480                 fpin = do_async_mmap_readahead(vmf, page);
2481         } else if (!page) {
2482                 /* No page in the page cache at all */
2483                 count_vm_event(PGMAJFAULT);
2484                 count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2485                 ret = VM_FAULT_MAJOR;
2486                 fpin = do_sync_mmap_readahead(vmf);
2487 retry_find:
2488                 page = pagecache_get_page(mapping, offset,
2489                                           FGP_CREAT|FGP_FOR_MMAP,
2490                                           vmf->gfp_mask);
2491                 if (!page) {
2492                         if (fpin)
2493                                 goto out_retry;
2494                         return vmf_error(-ENOMEM);
2495                 }
2496         }
2497 
2498         if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2499                 goto out_retry;
2500 
2501         /* Did it get truncated? */
2502         if (unlikely(compound_head(page)->mapping != mapping)) {
2503                 unlock_page(page);
2504                 put_page(page);
2505                 goto retry_find;
2506         }
2507         VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2508 
2509         /*
2510          * We have a locked page in the page cache, now we need to check
2511          * that it's up-to-date. If not, it is going to be due to an error.
2512          */
2513         if (unlikely(!PageUptodate(page)))
2514                 goto page_not_uptodate;
2515 
2516         /*
2517          * We've made it this far and we had to drop our mmap_sem, now is the
2518          * time to return to the upper layer and have it re-find the vma and
2519          * redo the fault.
2520          */
2521         if (fpin) {
2522                 unlock_page(page);
2523                 goto out_retry;
2524         }
2525 
2526         /*
2527          * Found the page and have a reference on it.
2528          * We must recheck i_size under page lock.
2529          */
2530         max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2531         if (unlikely(offset >= max_off)) {
2532                 unlock_page(page);
2533                 put_page(page);
2534                 return VM_FAULT_SIGBUS;
2535         }
2536 
2537         vmf->page = page;
2538         return ret | VM_FAULT_LOCKED;
2539 
2540 page_not_uptodate:
2541         /*
2542          * Umm, take care of errors if the page isn't up-to-date.
2543          * Try to re-read it _once_. We do this synchronously,
2544          * because there really aren't any performance issues here
2545          * and we need to check for errors.
2546          */
2547         ClearPageError(page);
2548         fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2549         error = mapping->a_ops->readpage(file, page);
2550         if (!error) {
2551                 wait_on_page_locked(page);
2552                 if (!PageUptodate(page))
2553                         error = -EIO;
2554         }
2555         if (fpin)
2556                 goto out_retry;
2557         put_page(page);
2558 
2559         if (!error || error == AOP_TRUNCATED_PAGE)
2560                 goto retry_find;
2561 
2562         /* Things didn't work out. Return zero to tell the mm layer so. */
2563         shrink_readahead_size_eio(file, ra);
2564         return VM_FAULT_SIGBUS;
2565 
2566 out_retry:
2567         /*
2568          * We dropped the mmap_sem, we need to return to the fault handler to
2569          * re-find the vma and come back and find our hopefully still populated
2570          * page.
2571          */
2572         if (page)
2573                 put_page(page);
2574         if (fpin)
2575                 fput(fpin);
2576         return ret | VM_FAULT_RETRY;
2577 }
2578 EXPORT_SYMBOL(filemap_fault);
2579 
2580 void filemap_map_pages(struct vm_fault *vmf,
2581                 pgoff_t start_pgoff, pgoff_t end_pgoff)
2582 {
2583         struct file *file = vmf->vma->vm_file;
2584         struct address_space *mapping = file->f_mapping;
2585         pgoff_t last_pgoff = start_pgoff;
2586         unsigned long max_idx;
2587         XA_STATE(xas, &mapping->i_pages, start_pgoff);
2588         struct page *page;
2589 
2590         rcu_read_lock();
2591         xas_for_each(&xas, page, end_pgoff) {
2592                 if (xas_retry(&xas, page))
2593                         continue;
2594                 if (xa_is_value(page))
2595                         goto next;
2596 
2597                 /*
2598                  * Check for a locked page first, as a speculative
2599                  * reference may adversely influence page migration.
2600                  */
2601                 if (PageLocked(page))
2602                         goto next;
2603                 if (!page_cache_get_speculative(page))
2604                         goto next;
2605 
2606                 /* Has the page moved or been split? */
2607                 if (unlikely(page != xas_reload(&xas)))
2608                         goto skip;
2609                 page = find_subpage(page, xas.xa_index);
2610 
2611                 if (!PageUptodate(page) ||
2612                                 PageReadahead(page) ||
2613                                 PageHWPoison(page))
2614                         goto skip;
2615                 if (!trylock_page(page))
2616                         goto skip;
2617 
2618                 if (page->mapping != mapping || !PageUptodate(page))
2619                         goto unlock;
2620 
2621                 max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2622                 if (page->index >= max_idx)
2623                         goto unlock;
2624 
2625                 if (file->f_ra.mmap_miss > 0)
2626                         file->f_ra.mmap_miss--;
2627 
2628                 vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2629                 if (vmf->pte)
2630                         vmf->pte += xas.xa_index - last_pgoff;
2631                 last_pgoff = xas.xa_index;
2632                 if (alloc_set_pte(vmf, NULL, page))
2633                         goto unlock;
2634                 unlock_page(page);
2635                 goto next;
2636 unlock:
2637                 unlock_page(page);
2638 skip:
2639                 put_page(page);
2640 next:
2641                 /* Huge page is mapped? No need to proceed. */
2642                 if (pmd_trans_huge(*vmf->pmd))
2643                         break;
2644         }
2645         rcu_read_unlock();
2646 }
2647 EXPORT_SYMBOL(filemap_map_pages);
2648 
2649 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2650 {
2651         struct page *page = vmf->page;
2652         struct inode *inode = file_inode(vmf->vma->vm_file);
2653         vm_fault_t ret = VM_FAULT_LOCKED;
2654 
2655         sb_start_pagefault(inode->i_sb);
2656         file_update_time(vmf->vma->vm_file);
2657         lock_page(page);
2658         if (page->mapping != inode->i_mapping) {
2659                 unlock_page(page);
2660                 ret = VM_FAULT_NOPAGE;
2661                 goto out;
2662         }
2663         /*
2664          * We mark the page dirty already here so that when freeze is in
2665          * progress, we are guaranteed that writeback during freezing will
2666          * see the dirty page and writeprotect it again.
2667          */
2668         set_page_dirty(page);
2669         wait_for_stable_page(page);
2670 out:
2671         sb_end_pagefault(inode->i_sb);
2672         return ret;
2673 }
2674 
2675 const struct vm_operations_struct generic_file_vm_ops = {
2676         .fault          = filemap_fault,
2677         .map_pages      = filemap_map_pages,
2678         .page_mkwrite   = filemap_page_mkwrite,
2679 };
2680 
2681 /* This is used for a general mmap of a disk file */
2682 
2683 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2684 {
2685         struct address_space *mapping = file->f_mapping;
2686 
2687         if (!mapping->a_ops->readpage)
2688                 return -ENOEXEC;
2689         file_accessed(file);
2690         vma->vm_ops = &generic_file_vm_ops;
2691         return 0;
2692 }
2693 
2694 /*
2695  * This is for filesystems which do not implement ->writepage.
2696  */
2697 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2698 {
2699         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2700                 return -EINVAL;
2701         return generic_file_mmap(file, vma);
2702 }
2703 #else
2704 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2705 {
2706         return VM_FAULT_SIGBUS;
2707 }
2708 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2709 {
2710         return -ENOSYS;
2711 }
2712 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2713 {
2714         return -ENOSYS;
2715 }
2716 #endif /* CONFIG_MMU */
2717 
2718 EXPORT_SYMBOL(filemap_page_mkwrite);
2719 EXPORT_SYMBOL(generic_file_mmap);
2720 EXPORT_SYMBOL(generic_file_readonly_mmap);
2721 
2722 static struct page *wait_on_page_read(struct page *page)
2723 {
2724         if (!IS_ERR(page)) {
2725                 wait_on_page_locked(page);
2726                 if (!PageUptodate(page)) {
2727                         put_page(page);
2728                         page = ERR_PTR(-EIO);
2729                 }
2730         }
2731         return page;
2732 }
2733 
2734 static struct page *do_read_cache_page(struct address_space *mapping,
2735                                 pgoff_t index,
2736                                 int (*filler)(void *, struct page *),
2737                                 void *data,
2738                                 gfp_t gfp)
2739 {
2740         struct page *page;
2741         int err;
2742 repeat:
2743         page = find_get_page(mapping, index);
2744         if (!page) {
2745                 page = __page_cache_alloc(gfp);
2746                 if (!page)
2747                         return ERR_PTR(-ENOMEM);
2748                 err = add_to_page_cache_lru(page, mapping, index, gfp);
2749                 if (unlikely(err)) {
2750                         put_page(page);
2751                         if (err == -EEXIST)
2752                                 goto repeat;
2753                         /* Presumably ENOMEM for xarray node */
2754                         return ERR_PTR(err);
2755                 }
2756 
2757 filler:
2758                 if (filler)
2759                         err = filler(data, page);
2760                 else
2761                         err = mapping->a_ops->readpage(data, page);
2762 
2763                 if (err < 0) {
2764                         put_page(page);
2765                         return ERR_PTR(err);
2766                 }
2767 
2768                 page = wait_on_page_read(page);
2769                 if (IS_ERR(page))
2770                         return page;
2771                 goto out;
2772         }
2773         if (PageUptodate(page))
2774                 goto out;
2775 
2776         /*
2777          * Page is not up to date and may be locked due one of the following
2778          * case a: Page is being filled and the page lock is held
2779          * case b: Read/write error clearing the page uptodate status
2780          * case c: Truncation in progress (page locked)
2781          * case d: Reclaim in progress
2782          *
2783          * Case a, the page will be up to date when the page is unlocked.
2784          *    There is no need to serialise on the page lock here as the page
2785          *    is pinned so the lock gives no additional protection. Even if the
2786          *    the page is truncated, the data is still valid if PageUptodate as
2787          *    it's a race vs truncate race.
2788          * Case b, the page will not be up to date
2789          * Case c, the page may be truncated but in itself, the data may still
2790          *    be valid after IO completes as it's a read vs truncate race. The
2791          *    operation must restart if the page is not uptodate on unlock but
2792          *    otherwise serialising on page lock to stabilise the mapping gives
2793          *    no additional guarantees to the caller as the page lock is
2794          *    released before return.
2795          * Case d, similar to truncation. If reclaim holds the page lock, it
2796          *    will be a race with remove_mapping that determines if the mapping
2797          *    is valid on unlock but otherwise the data is valid and there is
2798          *    no need to serialise with page lock.
2799          *
2800          * As the page lock gives no additional guarantee, we optimistically
2801          * wait on the page to be unlocked and check if it's up to date and
2802          * use the page if it is. Otherwise, the page lock is required to
2803          * distinguish between the different cases. The motivation is that we
2804          * avoid spurious serialisations and wakeups when multiple processes
2805          * wait on the same page for IO to complete.
2806          */
2807         wait_on_page_locked(page);
2808         if (PageUptodate(page))
2809                 goto out;
2810 
2811         /* Distinguish between all the cases under the safety of the lock */
2812         lock_page(page);
2813 
2814         /* Case c or d, restart the operation */
2815         if (!page->mapping) {
2816                 unlock_page(page);
2817                 put_page(page);
2818                 goto repeat;
2819         }
2820 
2821         /* Someone else locked and filled the page in a very small window */
2822         if (PageUptodate(page)) {
2823                 unlock_page(page);
2824                 goto out;
2825         }
2826         goto filler;
2827 
2828 out:
2829         mark_page_accessed(page);
2830         return page;
2831 }
2832 
2833 /**
2834  * read_cache_page - read into page cache, fill it if needed
2835  * @mapping:    the page's address_space
2836  * @index:      the page index
2837  * @filler:     function to perform the read
2838  * @data:       first arg to filler(data, page) function, often left as NULL
2839  *
2840  * Read into the page cache. If a page already exists, and PageUptodate() is
2841  * not set, try to fill the page and wait for it to become unlocked.
2842  *
2843  * If the page does not get brought uptodate, return -EIO.
2844  *
2845  * Return: up to date page on success, ERR_PTR() on failure.
2846  */
2847 struct page *read_cache_page(struct address_space *mapping,
2848                                 pgoff_t index,
2849                                 int (*filler)(void *, struct page *),
2850                                 void *data)
2851 {
2852         return do_read_cache_page(mapping, index, filler, data,
2853                         mapping_gfp_mask(mapping));
2854 }
2855 EXPORT_SYMBOL(read_cache_page);
2856 
2857 /**
2858  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2859  * @mapping:    the page's address_space
2860  * @index:      the page index
2861  * @gfp:        the page allocator flags to use if allocating
2862  *
2863  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2864  * any new page allocations done using the specified allocation flags.
2865  *
2866  * If the page does not get brought uptodate, return -EIO.
2867  *
2868  * Return: up to date page on success, ERR_PTR() on failure.
2869  */
2870 struct page *read_cache_page_gfp(struct address_space *mapping,
2871                                 pgoff_t index,
2872                                 gfp_t gfp)
2873 {
2874         return do_read_cache_page(mapping, index, NULL, NULL, gfp);
2875 }
2876 EXPORT_SYMBOL(read_cache_page_gfp);
2877 
2878 /*
2879  * Don't operate on ranges the page cache doesn't support, and don't exceed the
2880  * LFS limits.  If pos is under the limit it becomes a short access.  If it
2881  * exceeds the limit we return -EFBIG.
2882  */
2883 static int generic_write_check_limits(struct file *file, loff_t pos,
2884                                       loff_t *count)
2885 {
2886         struct inode *inode = file->f_mapping->host;
2887         loff_t max_size = inode->i_sb->s_maxbytes;
2888         loff_t limit = rlimit(RLIMIT_FSIZE);
2889 
2890         if (limit != RLIM_INFINITY) {
2891                 if (pos >= limit) {
2892                         send_sig(SIGXFSZ, current, 0);
2893                         return -EFBIG;
2894                 }
2895                 *count = min(*count, limit - pos);
2896         }
2897 
2898         if (!(file->f_flags & O_LARGEFILE))
2899                 max_size = MAX_NON_LFS;
2900 
2901         if (unlikely(pos >= max_size))
2902                 return -EFBIG;
2903 
2904         *count = min(*count, max_size - pos);
2905 
2906         return 0;
2907 }
2908 
2909 /*
2910  * Performs necessary checks before doing a write
2911  *
2912  * Can adjust writing position or amount of bytes to write.
2913  * Returns appropriate error code that caller should return or
2914  * zero in case that write should be allowed.
2915  */
2916 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2917 {
2918         struct file *file = iocb->ki_filp;
2919         struct inode *inode = file->f_mapping->host;
2920         loff_t count;
2921         int ret;
2922 
2923         if (IS_SWAPFILE(inode))
2924                 return -ETXTBSY;
2925 
2926         if (!iov_iter_count(from))
2927                 return 0;
2928 
2929         /* FIXME: this is for backwards compatibility with 2.4 */
2930         if (iocb->ki_flags & IOCB_APPEND)
2931                 iocb->ki_pos = i_size_read(inode);
2932 
2933         if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2934                 return -EINVAL;
2935 
2936         count = iov_iter_count(from);
2937         ret = generic_write_check_limits(file, iocb->ki_pos, &count);
2938         if (ret)
2939                 return ret;
2940 
2941         iov_iter_truncate(from, count);
2942         return iov_iter_count(from);
2943 }
2944 EXPORT_SYMBOL(generic_write_checks);
2945 
2946 /*
2947  * Performs necessary checks before doing a clone.
2948  *
2949  * Can adjust amount of bytes to clone via @req_count argument.
2950  * Returns appropriate error code that caller should return or
2951  * zero in case the clone should be allowed.
2952  */
2953 int generic_remap_checks(struct file *file_in, loff_t pos_in,
2954                          struct file *file_out, loff_t pos_out,
2955                          loff_t *req_count, unsigned int remap_flags)
2956 {
2957         struct inode *inode_in = file_in->f_mapping->host;
2958         struct inode *inode_out = file_out->f_mapping->host;
2959         uint64_t count = *req_count;
2960         uint64_t bcount;
2961         loff_t size_in, size_out;
2962         loff_t bs = inode_out->i_sb->s_blocksize;
2963         int ret;
2964 
2965         /* The start of both ranges must be aligned to an fs block. */
2966         if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
2967                 return -EINVAL;
2968 
2969         /* Ensure offsets don't wrap. */
2970         if (pos_in + count < pos_in || pos_out + count < pos_out)
2971                 return -EINVAL;
2972 
2973         size_in = i_size_read(inode_in);
2974         size_out = i_size_read(inode_out);
2975 
2976         /* Dedupe requires both ranges to be within EOF. */
2977         if ((remap_flags & REMAP_FILE_DEDUP) &&
2978             (pos_in >= size_in || pos_in + count > size_in ||
2979              pos_out >= size_out || pos_out + count > size_out))
2980                 return -EINVAL;
2981 
2982         /* Ensure the infile range is within the infile. */
2983         if (pos_in >= size_in)
2984                 return -EINVAL;
2985         count = min(count, size_in - (uint64_t)pos_in);
2986 
2987         ret = generic_write_check_limits(file_out, pos_out, &count);
2988         if (ret)
2989                 return ret;
2990 
2991         /*
2992          * If the user wanted us to link to the infile's EOF, round up to the
2993          * next block boundary for this check.
2994          *
2995          * Otherwise, make sure the count is also block-aligned, having
2996          * already confirmed the starting offsets' block alignment.
2997          */
2998         if (pos_in + count == size_in) {
2999                 bcount = ALIGN(size_in, bs) - pos_in;
3000         } else {
3001                 if (!IS_ALIGNED(count, bs))
3002                         count = ALIGN_DOWN(count, bs);
3003                 bcount = count;
3004         }
3005 
3006         /* Don't allow overlapped cloning within the same file. */
3007         if (inode_in == inode_out &&
3008             pos_out + bcount > pos_in &&
3009             pos_out < pos_in + bcount)
3010                 return -EINVAL;
3011 
3012         /*
3013          * We shortened the request but the caller can't deal with that, so
3014          * bounce the request back to userspace.
3015          */
3016         if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3017                 return -EINVAL;
3018 
3019         *req_count = count;
3020         return 0;
3021 }
3022 
3023 
3024 /*
3025  * Performs common checks before doing a file copy/clone
3026  * from @file_in to @file_out.
3027  */
3028 int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3029 {
3030         struct inode *inode_in = file_inode(file_in);
3031         struct inode *inode_out = file_inode(file_out);
3032 
3033         /* Don't copy dirs, pipes, sockets... */
3034         if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3035                 return -EISDIR;
3036         if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3037                 return -EINVAL;
3038 
3039         if (!(file_in->f_mode & FMODE_READ) ||
3040             !(file_out->f_mode & FMODE_WRITE) ||
3041             (file_out->f_flags & O_APPEND))
3042                 return -EBADF;
3043 
3044         return 0;
3045 }
3046 
3047 /*
3048  * Performs necessary checks before doing a file copy
3049  *
3050  * Can adjust amount of bytes to copy via @req_count argument.
3051  * Returns appropriate error code that caller should return or
3052  * zero in case the copy should be allowed.
3053  */
3054 int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3055                              struct file *file_out, loff_t pos_out,
3056                              size_t *req_count, unsigned int flags)
3057 {
3058         struct inode *inode_in = file_inode(file_in);
3059         struct inode *inode_out = file_inode(file_out);
3060         uint64_t count = *req_count;
3061         loff_t size_in;
3062         int ret;
3063 
3064         ret = generic_file_rw_checks(file_in, file_out);
3065         if (ret)
3066                 return ret;
3067 
3068         /* Don't touch certain kinds of inodes */
3069         if (IS_IMMUTABLE(inode_out))
3070                 return -EPERM;
3071 
3072         if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3073                 return -ETXTBSY;
3074 
3075         /* Ensure offsets don't wrap. */
3076         if (pos_in + count < pos_in || pos_out + count < pos_out)
3077                 return -EOVERFLOW;
3078 
3079         /* Shorten the copy to EOF */
3080         size_in = i_size_read(inode_in);
3081         if (pos_in >= size_in)
3082                 count = 0;
3083         else
3084                 count = min(count, size_in - (uint64_t)pos_in);
3085 
3086         ret = generic_write_check_limits(file_out, pos_out, &count);
3087         if (ret)
3088                 return ret;
3089 
3090         /* Don't allow overlapped copying within the same file. */
3091         if (inode_in == inode_out &&
3092             pos_out + count > pos_in &&
3093             pos_out < pos_in + count)
3094                 return -EINVAL;
3095 
3096         *req_count = count;
3097         return 0;
3098 }
3099 
3100 int pagecache_write_begin(struct file *file, struct address_space *mapping,
3101                                 loff_t pos, unsigned len, unsigned flags,
3102                                 struct page **pagep, void **fsdata)
3103 {
3104         const struct address_space_operations *aops = mapping->a_ops;
3105 
3106         return aops->write_begin(file, mapping, pos, len, flags,
3107                                                         pagep, fsdata);
3108 }
3109 EXPORT_SYMBOL(pagecache_write_begin);
3110 
3111 int pagecache_write_end(struct file *file, struct address_space *mapping,
3112                                 loff_t pos, unsigned len, unsigned copied,
3113                                 struct page *page, void *fsdata)
3114 {
3115         const struct address_space_operations *aops = mapping->a_ops;
3116 
3117         return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3118 }
3119 EXPORT_SYMBOL(pagecache_write_end);
3120 
3121 /*
3122  * Warn about a page cache invalidation failure during a direct I/O write.
3123  */
3124 void dio_warn_stale_pagecache(struct file *filp)
3125 {
3126         static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3127         char pathname[128];
3128         struct inode *inode = file_inode(filp);
3129         char *path;
3130 
3131         errseq_set(&inode->i_mapping->wb_err, -EIO);
3132         if (__ratelimit(&_rs)) {
3133                 path = file_path(filp, pathname, sizeof(pathname));
3134                 if (IS_ERR(path))
3135                         path = "(unknown)";
3136                 pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3137                 pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3138                         current->comm);
3139         }
3140 }
3141 
3142 ssize_t
3143 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3144 {
3145         struct file     *file = iocb->ki_filp;
3146         struct address_space *mapping = file->f_mapping;
3147         struct inode    *inode = mapping->host;
3148         loff_t          pos = iocb->ki_pos;
3149         ssize_t         written;
3150         size_t          write_len;
3151         pgoff_t         end;
3152 
3153         write_len = iov_iter_count(from);
3154         end = (pos + write_len - 1) >> PAGE_SHIFT;
3155 
3156         if (iocb->ki_flags & IOCB_NOWAIT) {
3157                 /* If there are pages to writeback, return */
3158                 if (filemap_range_has_page(inode->i_mapping, pos,
3159                                            pos + write_len - 1))
3160                         return -EAGAIN;
3161         } else {
3162                 written = filemap_write_and_wait_range(mapping, pos,
3163                                                         pos + write_len - 1);
3164                 if (written)
3165                         goto out;
3166         }
3167 
3168         /*
3169          * After a write we want buffered reads to be sure to go to disk to get
3170          * the new data.  We invalidate clean cached page from the region we're
3171          * about to write.  We do this *before* the write so that we can return
3172          * without clobbering -EIOCBQUEUED from ->direct_IO().
3173          */
3174         written = invalidate_inode_pages2_range(mapping,
3175                                         pos >> PAGE_SHIFT, end);
3176         /*
3177          * If a page can not be invalidated, return 0 to fall back
3178          * to buffered write.
3179          */
3180         if (written) {
3181                 if (written == -EBUSY)
3182                         return 0;
3183                 goto out;
3184         }
3185 
3186         written = mapping->a_ops->direct_IO(iocb, from);
3187 
3188         /*
3189          * Finally, try again to invalidate clean pages which might have been
3190          * cached by non-direct readahead, or faulted in by get_user_pages()
3191          * if the source of the write was an mmap'ed region of the file
3192          * we're writing.  Either one is a pretty crazy thing to do,
3193          * so we don't support it 100%.  If this invalidation
3194          * fails, tough, the write still worked...
3195          *
3196          * Most of the time we do not need this since dio_complete() will do
3197          * the invalidation for us. However there are some file systems that
3198          * do not end up with dio_complete() being called, so let's not break
3199          * them by removing it completely.
3200          *
3201          * Noticeable example is a blkdev_direct_IO().
3202          *
3203          * Skip invalidation for async writes or if mapping has no pages.
3204          */
3205         if (written > 0 && mapping->nrpages &&
3206             invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3207                 dio_warn_stale_pagecache(file);
3208 
3209         if (written > 0) {
3210                 pos += written;
3211                 write_len -= written;
3212                 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3213                         i_size_write(inode, pos);
3214                         mark_inode_dirty(inode);
3215                 }
3216                 iocb->ki_pos = pos;
3217         }
3218         iov_iter_revert(from, write_len - iov_iter_count(from));
3219 out:
3220         return written;
3221 }
3222 EXPORT_SYMBOL(generic_file_direct_write);
3223 
3224 /*
3225  * Find or create a page at the given pagecache position. Return the locked
3226  * page. This function is specifically for buffered writes.
3227  */
3228 struct page *grab_cache_page_write_begin(struct address_space *mapping,
3229                                         pgoff_t index, unsigned flags)
3230 {
3231         struct page *page;
3232         int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3233 
3234         if (flags & AOP_FLAG_NOFS)
3235                 fgp_flags |= FGP_NOFS;
3236 
3237         page = pagecache_get_page(mapping, index, fgp_flags,
3238                         mapping_gfp_mask(mapping));
3239         if (page)
3240                 wait_for_stable_page(page);
3241 
3242         return page;
3243 }
3244 EXPORT_SYMBOL(grab_cache_page_write_begin);
3245 
3246 ssize_t generic_perform_write(struct file *file,
3247                                 struct iov_iter *i, loff_t pos)
3248 {
3249         struct address_space *mapping = file->f_mapping;
3250         const struct address_space_operations *a_ops = mapping->a_ops;
3251         long status = 0;
3252         ssize_t written = 0;
3253         unsigned int flags = 0;
3254 
3255         do {
3256                 struct page *page;
3257                 unsigned long offset;   /* Offset into pagecache page */
3258                 unsigned long bytes;    /* Bytes to write to page */
3259                 size_t copied;          /* Bytes copied from user */
3260                 void *fsdata;
3261 
3262                 offset = (pos & (PAGE_SIZE - 1));
3263                 bytes = min_t(unsigned long, PAGE_SIZE - offset,
3264                                                 iov_iter_count(i));
3265 
3266 again:
3267                 /*
3268                  * Bring in the user page that we will copy from _first_.
3269                  * Otherwise there's a nasty deadlock on copying from the
3270                  * same page as we're writing to, without it being marked
3271                  * up-to-date.
3272                  *
3273                  * Not only is this an optimisation, but it is also required
3274                  * to check that the address is actually valid, when atomic
3275                  * usercopies are used, below.
3276                  */
3277                 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3278                         status = -EFAULT;
3279                         break;
3280                 }
3281 
3282                 if (fatal_signal_pending(current)) {
3283                         status = -EINTR;
3284                         break;
3285                 }
3286 
3287                 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3288                                                 &page, &fsdata);
3289                 if (unlikely(status < 0))
3290                         break;
3291 
3292                 if (mapping_writably_mapped(mapping))
3293                         flush_dcache_page(page);
3294 
3295                 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3296                 flush_dcache_page(page);
3297 
3298                 status = a_ops->write_end(file, mapping, pos, bytes, copied,
3299                                                 page, fsdata);
3300                 if (unlikely(status < 0))
3301                         break;
3302                 copied = status;
3303 
3304                 cond_resched();
3305 
3306                 iov_iter_advance(i, copied);
3307                 if (unlikely(copied == 0)) {
3308                         /*
3309                          * If we were unable to copy any data at all, we must
3310                          * fall back to a single segment length write.
3311                          *
3312                          * If we didn't fallback here, we could livelock
3313                          * because not all segments in the iov can be copied at
3314                          * once without a pagefault.
3315                          */
3316                         bytes = min_t(unsigned long, PAGE_SIZE - offset,
3317                                                 iov_iter_single_seg_count(i));
3318                         goto again;
3319                 }
3320                 pos += copied;
3321                 written += copied;
3322 
3323                 balance_dirty_pages_ratelimited(mapping);
3324         } while (iov_iter_count(i));
3325 
3326         return written ? written : status;
3327 }
3328 EXPORT_SYMBOL(generic_perform_write);
3329 
3330 /**
3331  * __generic_file_write_iter - write data to a file
3332  * @iocb:       IO state structure (file, offset, etc.)
3333  * @from:       iov_iter with data to write
3334  *
3335  * This function does all the work needed for actually writing data to a
3336  * file. It does all basic checks, removes SUID from the file, updates
3337  * modification times and calls proper subroutines depending on whether we
3338  * do direct IO or a standard buffered write.
3339  *
3340  * It expects i_mutex to be grabbed unless we work on a block device or similar
3341  * object which does not need locking at all.
3342  *
3343  * This function does *not* take care of syncing data in case of O_SYNC write.
3344  * A caller has to handle it. This is mainly due to the fact that we want to
3345  * avoid syncing under i_mutex.
3346  *
3347  * Return:
3348  * * number of bytes written, even for truncated writes
3349  * * negative error code if no data has been written at all
3350  */
3351 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3352 {
3353         struct file *file = iocb->ki_filp;
3354         struct address_space * mapping = file->f_mapping;
3355         struct inode    *inode = mapping->host;
3356         ssize_t         written = 0;
3357         ssize_t         err;
3358         ssize_t         status;
3359 
3360         /* We can write back this queue in page reclaim */
3361         current->backing_dev_info = inode_to_bdi(inode);
3362         err = file_remove_privs(file);
3363         if (err)
3364                 goto out;
3365 
3366         err = file_update_time(file);
3367         if (err)
3368                 goto out;
3369 
3370         if (iocb->ki_flags & IOCB_DIRECT) {
3371                 loff_t pos, endbyte;
3372 
3373                 written = generic_file_direct_write(iocb, from);
3374                 /*
3375                  * If the write stopped short of completing, fall back to
3376                  * buffered writes.  Some filesystems do this for writes to
3377                  * holes, for example.  For DAX files, a buffered write will
3378                  * not succeed (even if it did, DAX does not handle dirty
3379                  * page-cache pages correctly).
3380                  */
3381                 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3382                         goto out;
3383 
3384                 status = generic_perform_write(file, from, pos = iocb->ki_pos);
3385                 /*
3386                  * If generic_perform_write() returned a synchronous error
3387                  * then we want to return the number of bytes which were
3388                  * direct-written, or the error code if that was zero.  Note
3389                  * that this differs from normal direct-io semantics, which
3390                  * will return -EFOO even if some bytes were written.
3391                  */
3392                 if (unlikely(status < 0)) {
3393                         err = status;
3394                         goto out;
3395                 }
3396                 /*
3397                  * We need to ensure that the page cache pages are written to
3398                  * disk and invalidated to preserve the expected O_DIRECT
3399                  * semantics.
3400                  */
3401                 endbyte = pos + status - 1;
3402                 err = filemap_write_and_wait_range(mapping, pos, endbyte);
3403                 if (err == 0) {
3404                         iocb->ki_pos = endbyte + 1;
3405                         written += status;
3406                         invalidate_mapping_pages(mapping,
3407                                                  pos >> PAGE_SHIFT,
3408                                                  endbyte >> PAGE_SHIFT);
3409                 } else {
3410                         /*
3411                          * We don't know how much we wrote, so just return
3412                          * the number of bytes which were direct-written
3413                          */
3414                 }
3415         } else {
3416                 written = generic_perform_write(file, from, iocb->ki_pos);
3417                 if (likely(written > 0))
3418                         iocb->ki_pos += written;
3419         }
3420 out:
3421         current->backing_dev_info = NULL;
3422         return written ? written : err;
3423 }
3424 EXPORT_SYMBOL(__generic_file_write_iter);
3425 
3426 /**
3427  * generic_file_write_iter - write data to a file
3428  * @iocb:       IO state structure
3429  * @from:       iov_iter with data to write
3430  *
3431  * This is a wrapper around __generic_file_write_iter() to be used by most
3432  * filesystems. It takes care of syncing the file in case of O_SYNC file
3433  * and acquires i_mutex as needed.
3434  * Return:
3435  * * negative error code if no data has been written at all of
3436  *   vfs_fsync_range() failed for a synchronous write
3437  * * number of bytes written, even for truncated writes
3438  */
3439 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3440 {
3441         struct file *file = iocb->ki_filp;
3442         struct inode *inode = file->f_mapping->host;
3443         ssize_t ret;
3444 
3445         inode_lock(inode);
3446         ret = generic_write_checks(iocb, from);
3447         if (ret > 0)
3448                 ret = __generic_file_write_iter(iocb, from);
3449         inode_unlock(inode);
3450 
3451         if (ret > 0)
3452                 ret = generic_write_sync(iocb, ret);
3453         return ret;
3454 }
3455 EXPORT_SYMBOL(generic_file_write_iter);
3456 
3457 /**
3458  * try_to_release_page() - release old fs-specific metadata on a page
3459  *
3460  * @page: the page which the kernel is trying to free
3461  * @gfp_mask: memory allocation flags (and I/O mode)
3462  *
3463  * The address_space is to try to release any data against the page
3464  * (presumably at page->private).
3465  *
3466  * This may also be called if PG_fscache is set on a page, indicating that the
3467  * page is known to the local caching routines.
3468  *
3469  * The @gfp_mask argument specifies whether I/O may be performed to release
3470  * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3471  *
3472  * Return: %1 if the release was successful, otherwise return zero.
3473  */
3474 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3475 {
3476         struct address_space * const mapping = page->mapping;
3477 
3478         BUG_ON(!PageLocked(page));
3479         if (PageWriteback(page))
3480                 return 0;
3481 
3482         if (mapping && mapping->a_ops->releasepage)
3483                 return mapping->a_ops->releasepage(page, gfp_mask);
3484         return try_to_free_buffers(page);
3485 }
3486 
3487 EXPORT_SYMBOL(try_to_release_page);
3488 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp