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

TOMOYO Linux Cross Reference
Linux/mm/filemap.c

Version: ~ [ linux-5.19-rc8 ] ~ [ linux-5.18.14 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.57 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.133 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.207 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.253 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.289 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.324 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *      linux/mm/filemap.c
  3  *
  4  * Copyright (C) 1994-1999  Linus Torvalds
  5  */
  6 
  7 /*
  8  * This file handles the generic file mmap semantics used by
  9  * most "normal" filesystems (but you don't /have/ to use this:
 10  * the NFS filesystem used to do this differently, for example)
 11  */
 12 #include <linux/export.h>
 13 #include <linux/compiler.h>
 14 #include <linux/fs.h>
 15 #include <linux/uaccess.h>
 16 #include <linux/capability.h>
 17 #include <linux/kernel_stat.h>
 18 #include <linux/gfp.h>
 19 #include <linux/mm.h>
 20 #include <linux/swap.h>
 21 #include <linux/mman.h>
 22 #include <linux/pagemap.h>
 23 #include <linux/file.h>
 24 #include <linux/uio.h>
 25 #include <linux/hash.h>
 26 #include <linux/writeback.h>
 27 #include <linux/backing-dev.h>
 28 #include <linux/pagevec.h>
 29 #include <linux/blkdev.h>
 30 #include <linux/security.h>
 31 #include <linux/cpuset.h>
 32 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
 33 #include <linux/hugetlb.h>
 34 #include <linux/memcontrol.h>
 35 #include <linux/cleancache.h>
 36 #include <linux/rmap.h>
 37 #include "internal.h"
 38 
 39 #define CREATE_TRACE_POINTS
 40 #include <trace/events/filemap.h>
 41 
 42 /*
 43  * FIXME: remove all knowledge of the buffer layer from the core VM
 44  */
 45 #include <linux/buffer_head.h> /* for try_to_free_buffers */
 46 
 47 #include <asm/mman.h>
 48 
 49 /*
 50  * Shared mappings implemented 30.11.1994. It's not fully working yet,
 51  * though.
 52  *
 53  * Shared mappings now work. 15.8.1995  Bruno.
 54  *
 55  * finished 'unifying' the page and buffer cache and SMP-threaded the
 56  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
 57  *
 58  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
 59  */
 60 
 61 /*
 62  * Lock ordering:
 63  *
 64  *  ->i_mmap_rwsem              (truncate_pagecache)
 65  *    ->private_lock            (__free_pte->__set_page_dirty_buffers)
 66  *      ->swap_lock             (exclusive_swap_page, others)
 67  *        ->mapping->tree_lock
 68  *
 69  *  ->i_mutex
 70  *    ->i_mmap_rwsem            (truncate->unmap_mapping_range)
 71  *
 72  *  ->mmap_sem
 73  *    ->i_mmap_rwsem
 74  *      ->page_table_lock or pte_lock   (various, mainly in memory.c)
 75  *        ->mapping->tree_lock  (arch-dependent flush_dcache_mmap_lock)
 76  *
 77  *  ->mmap_sem
 78  *    ->lock_page               (access_process_vm)
 79  *
 80  *  ->i_mutex                   (generic_perform_write)
 81  *    ->mmap_sem                (fault_in_pages_readable->do_page_fault)
 82  *
 83  *  bdi->wb.list_lock
 84  *    sb_lock                   (fs/fs-writeback.c)
 85  *    ->mapping->tree_lock      (__sync_single_inode)
 86  *
 87  *  ->i_mmap_rwsem
 88  *    ->anon_vma.lock           (vma_adjust)
 89  *
 90  *  ->anon_vma.lock
 91  *    ->page_table_lock or pte_lock     (anon_vma_prepare and various)
 92  *
 93  *  ->page_table_lock or pte_lock
 94  *    ->swap_lock               (try_to_unmap_one)
 95  *    ->private_lock            (try_to_unmap_one)
 96  *    ->tree_lock               (try_to_unmap_one)
 97  *    ->zone.lru_lock           (follow_page->mark_page_accessed)
 98  *    ->zone.lru_lock           (check_pte_range->isolate_lru_page)
 99  *    ->private_lock            (page_remove_rmap->set_page_dirty)
100  *    ->tree_lock               (page_remove_rmap->set_page_dirty)
101  *    bdi.wb->list_lock         (page_remove_rmap->set_page_dirty)
102  *    ->inode->i_lock           (page_remove_rmap->set_page_dirty)
103  *    ->memcg->move_lock        (page_remove_rmap->mem_cgroup_begin_page_stat)
104  *    bdi.wb->list_lock         (zap_pte_range->set_page_dirty)
105  *    ->inode->i_lock           (zap_pte_range->set_page_dirty)
106  *    ->private_lock            (zap_pte_range->__set_page_dirty_buffers)
107  *
108  * ->i_mmap_rwsem
109  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
110  */
111 
112 static int page_cache_tree_insert(struct address_space *mapping,
113                                   struct page *page, void **shadowp)
114 {
115         struct radix_tree_node *node;
116         void **slot;
117         int error;
118 
119         error = __radix_tree_create(&mapping->page_tree, page->index,
120                                     &node, &slot);
121         if (error)
122                 return error;
123         if (*slot) {
124                 void *p;
125 
126                 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
127                 if (!radix_tree_exceptional_entry(p))
128                         return -EEXIST;
129                 if (shadowp)
130                         *shadowp = p;
131                 mapping->nrshadows--;
132                 if (node)
133                         workingset_node_shadows_dec(node);
134         }
135         radix_tree_replace_slot(slot, page);
136         mapping->nrpages++;
137         if (node) {
138                 workingset_node_pages_inc(node);
139                 /*
140                  * Don't track node that contains actual pages.
141                  *
142                  * Avoid acquiring the list_lru lock if already
143                  * untracked.  The list_empty() test is safe as
144                  * node->private_list is protected by
145                  * mapping->tree_lock.
146                  */
147                 if (!list_empty(&node->private_list))
148                         list_lru_del(&workingset_shadow_nodes,
149                                      &node->private_list);
150         }
151         return 0;
152 }
153 
154 static void page_cache_tree_delete(struct address_space *mapping,
155                                    struct page *page, void *shadow)
156 {
157         struct radix_tree_node *node;
158         unsigned long index;
159         unsigned int offset;
160         unsigned int tag;
161         void **slot;
162 
163         VM_BUG_ON(!PageLocked(page));
164 
165         __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
166 
167         if (!node) {
168                 /*
169                  * We need a node to properly account shadow
170                  * entries. Don't plant any without. XXX
171                  */
172                 shadow = NULL;
173         }
174 
175         if (shadow) {
176                 mapping->nrshadows++;
177                 /*
178                  * Make sure the nrshadows update is committed before
179                  * the nrpages update so that final truncate racing
180                  * with reclaim does not see both counters 0 at the
181                  * same time and miss a shadow entry.
182                  */
183                 smp_wmb();
184         }
185         mapping->nrpages--;
186 
187         if (!node) {
188                 /* Clear direct pointer tags in root node */
189                 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
190                 radix_tree_replace_slot(slot, shadow);
191                 return;
192         }
193 
194         /* Clear tree tags for the removed page */
195         index = page->index;
196         offset = index & RADIX_TREE_MAP_MASK;
197         for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
198                 if (test_bit(offset, node->tags[tag]))
199                         radix_tree_tag_clear(&mapping->page_tree, index, tag);
200         }
201 
202         /* Delete page, swap shadow entry */
203         radix_tree_replace_slot(slot, shadow);
204         workingset_node_pages_dec(node);
205         if (shadow)
206                 workingset_node_shadows_inc(node);
207         else
208                 if (__radix_tree_delete_node(&mapping->page_tree, node))
209                         return;
210 
211         /*
212          * Track node that only contains shadow entries.
213          *
214          * Avoid acquiring the list_lru lock if already tracked.  The
215          * list_empty() test is safe as node->private_list is
216          * protected by mapping->tree_lock.
217          */
218         if (!workingset_node_pages(node) &&
219             list_empty(&node->private_list)) {
220                 node->private_data = mapping;
221                 list_lru_add(&workingset_shadow_nodes, &node->private_list);
222         }
223 }
224 
225 /*
226  * Delete a page from the page cache and free it. Caller has to make
227  * sure the page is locked and that nobody else uses it - or that usage
228  * is safe.  The caller must hold the mapping's tree_lock and
229  * mem_cgroup_begin_page_stat().
230  */
231 void __delete_from_page_cache(struct page *page, void *shadow,
232                               struct mem_cgroup *memcg)
233 {
234         struct address_space *mapping = page->mapping;
235 
236         trace_mm_filemap_delete_from_page_cache(page);
237         /*
238          * if we're uptodate, flush out into the cleancache, otherwise
239          * invalidate any existing cleancache entries.  We can't leave
240          * stale data around in the cleancache once our page is gone
241          */
242         if (PageUptodate(page) && PageMappedToDisk(page))
243                 cleancache_put_page(page);
244         else
245                 cleancache_invalidate_page(mapping, page);
246 
247         page_cache_tree_delete(mapping, page, shadow);
248 
249         page->mapping = NULL;
250         /* Leave page->index set: truncation lookup relies upon it */
251 
252         /* hugetlb pages do not participate in page cache accounting. */
253         if (!PageHuge(page))
254                 __dec_zone_page_state(page, NR_FILE_PAGES);
255         if (PageSwapBacked(page))
256                 __dec_zone_page_state(page, NR_SHMEM);
257         BUG_ON(page_mapped(page));
258 
259         /*
260          * At this point page must be either written or cleaned by truncate.
261          * Dirty page here signals a bug and loss of unwritten data.
262          *
263          * This fixes dirty accounting after removing the page entirely but
264          * leaves PageDirty set: it has no effect for truncated page and
265          * anyway will be cleared before returning page into buddy allocator.
266          */
267         if (WARN_ON_ONCE(PageDirty(page)))
268                 account_page_cleaned(page, mapping, memcg,
269                                      inode_to_wb(mapping->host));
270 }
271 
272 /**
273  * delete_from_page_cache - delete page from page cache
274  * @page: the page which the kernel is trying to remove from page cache
275  *
276  * This must be called only on pages that have been verified to be in the page
277  * cache and locked.  It will never put the page into the free list, the caller
278  * has a reference on the page.
279  */
280 void delete_from_page_cache(struct page *page)
281 {
282         struct address_space *mapping = page->mapping;
283         struct mem_cgroup *memcg;
284         unsigned long flags;
285 
286         void (*freepage)(struct page *);
287 
288         BUG_ON(!PageLocked(page));
289 
290         freepage = mapping->a_ops->freepage;
291 
292         memcg = mem_cgroup_begin_page_stat(page);
293         spin_lock_irqsave(&mapping->tree_lock, flags);
294         __delete_from_page_cache(page, NULL, memcg);
295         spin_unlock_irqrestore(&mapping->tree_lock, flags);
296         mem_cgroup_end_page_stat(memcg);
297 
298         if (freepage)
299                 freepage(page);
300         page_cache_release(page);
301 }
302 EXPORT_SYMBOL(delete_from_page_cache);
303 
304 static int filemap_check_errors(struct address_space *mapping)
305 {
306         int ret = 0;
307         /* Check for outstanding write errors */
308         if (test_bit(AS_ENOSPC, &mapping->flags) &&
309             test_and_clear_bit(AS_ENOSPC, &mapping->flags))
310                 ret = -ENOSPC;
311         if (test_bit(AS_EIO, &mapping->flags) &&
312             test_and_clear_bit(AS_EIO, &mapping->flags))
313                 ret = -EIO;
314         return ret;
315 }
316 
317 /**
318  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
319  * @mapping:    address space structure to write
320  * @start:      offset in bytes where the range starts
321  * @end:        offset in bytes where the range ends (inclusive)
322  * @sync_mode:  enable synchronous operation
323  *
324  * Start writeback against all of a mapping's dirty pages that lie
325  * within the byte offsets <start, end> inclusive.
326  *
327  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
328  * opposed to a regular memory cleansing writeback.  The difference between
329  * these two operations is that if a dirty page/buffer is encountered, it must
330  * be waited upon, and not just skipped over.
331  */
332 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
333                                 loff_t end, int sync_mode)
334 {
335         int ret;
336         struct writeback_control wbc = {
337                 .sync_mode = sync_mode,
338                 .nr_to_write = LONG_MAX,
339                 .range_start = start,
340                 .range_end = end,
341         };
342 
343         if (!mapping_cap_writeback_dirty(mapping) ||
344             !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
345                 return 0;
346 
347         wbc_attach_fdatawrite_inode(&wbc, mapping->host);
348         ret = do_writepages(mapping, &wbc);
349         wbc_detach_inode(&wbc);
350         return ret;
351 }
352 
353 static inline int __filemap_fdatawrite(struct address_space *mapping,
354         int sync_mode)
355 {
356         return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
357 }
358 
359 int filemap_fdatawrite(struct address_space *mapping)
360 {
361         return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
362 }
363 EXPORT_SYMBOL(filemap_fdatawrite);
364 
365 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
366                                 loff_t end)
367 {
368         return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
369 }
370 EXPORT_SYMBOL(filemap_fdatawrite_range);
371 
372 /**
373  * filemap_flush - mostly a non-blocking flush
374  * @mapping:    target address_space
375  *
376  * This is a mostly non-blocking flush.  Not suitable for data-integrity
377  * purposes - I/O may not be started against all dirty pages.
378  */
379 int filemap_flush(struct address_space *mapping)
380 {
381         return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
382 }
383 EXPORT_SYMBOL(filemap_flush);
384 
385 static int __filemap_fdatawait_range(struct address_space *mapping,
386                                      loff_t start_byte, loff_t end_byte)
387 {
388         pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
389         pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
390         struct pagevec pvec;
391         int nr_pages;
392         int ret = 0;
393 
394         if (end_byte < start_byte)
395                 goto out;
396 
397         pagevec_init(&pvec, 0);
398         while ((index <= end) &&
399                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
400                         PAGECACHE_TAG_WRITEBACK,
401                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
402                 unsigned i;
403 
404                 for (i = 0; i < nr_pages; i++) {
405                         struct page *page = pvec.pages[i];
406 
407                         /* until radix tree lookup accepts end_index */
408                         if (page->index > end)
409                                 continue;
410 
411                         wait_on_page_writeback(page);
412                         if (TestClearPageError(page))
413                                 ret = -EIO;
414                 }
415                 pagevec_release(&pvec);
416                 cond_resched();
417         }
418 out:
419         return ret;
420 }
421 
422 /**
423  * filemap_fdatawait_range - wait for writeback to complete
424  * @mapping:            address space structure to wait for
425  * @start_byte:         offset in bytes where the range starts
426  * @end_byte:           offset in bytes where the range ends (inclusive)
427  *
428  * Walk the list of under-writeback pages of the given address space
429  * in the given range and wait for all of them.  Check error status of
430  * the address space and return it.
431  *
432  * Since the error status of the address space is cleared by this function,
433  * callers are responsible for checking the return value and handling and/or
434  * reporting the error.
435  */
436 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
437                             loff_t end_byte)
438 {
439         int ret, ret2;
440 
441         ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
442         ret2 = filemap_check_errors(mapping);
443         if (!ret)
444                 ret = ret2;
445 
446         return ret;
447 }
448 EXPORT_SYMBOL(filemap_fdatawait_range);
449 
450 /**
451  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
452  * @mapping: address space structure to wait for
453  *
454  * Walk the list of under-writeback pages of the given address space
455  * and wait for all of them.  Unlike filemap_fdatawait(), this function
456  * does not clear error status of the address space.
457  *
458  * Use this function if callers don't handle errors themselves.  Expected
459  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
460  * fsfreeze(8)
461  */
462 void filemap_fdatawait_keep_errors(struct address_space *mapping)
463 {
464         loff_t i_size = i_size_read(mapping->host);
465 
466         if (i_size == 0)
467                 return;
468 
469         __filemap_fdatawait_range(mapping, 0, i_size - 1);
470 }
471 
472 /**
473  * filemap_fdatawait - wait for all under-writeback pages to complete
474  * @mapping: address space structure to wait for
475  *
476  * Walk the list of under-writeback pages of the given address space
477  * and wait for all of them.  Check error status of the address space
478  * and return it.
479  *
480  * Since the error status of the address space is cleared by this function,
481  * callers are responsible for checking the return value and handling and/or
482  * reporting the error.
483  */
484 int filemap_fdatawait(struct address_space *mapping)
485 {
486         loff_t i_size = i_size_read(mapping->host);
487 
488         if (i_size == 0)
489                 return 0;
490 
491         return filemap_fdatawait_range(mapping, 0, i_size - 1);
492 }
493 EXPORT_SYMBOL(filemap_fdatawait);
494 
495 int filemap_write_and_wait(struct address_space *mapping)
496 {
497         int err = 0;
498 
499         if (mapping->nrpages) {
500                 err = filemap_fdatawrite(mapping);
501                 /*
502                  * Even if the above returned error, the pages may be
503                  * written partially (e.g. -ENOSPC), so we wait for it.
504                  * But the -EIO is special case, it may indicate the worst
505                  * thing (e.g. bug) happened, so we avoid waiting for it.
506                  */
507                 if (err != -EIO) {
508                         int err2 = filemap_fdatawait(mapping);
509                         if (!err)
510                                 err = err2;
511                 }
512         } else {
513                 err = filemap_check_errors(mapping);
514         }
515         return err;
516 }
517 EXPORT_SYMBOL(filemap_write_and_wait);
518 
519 /**
520  * filemap_write_and_wait_range - write out & wait on a file range
521  * @mapping:    the address_space for the pages
522  * @lstart:     offset in bytes where the range starts
523  * @lend:       offset in bytes where the range ends (inclusive)
524  *
525  * Write out and wait upon file offsets lstart->lend, inclusive.
526  *
527  * Note that `lend' is inclusive (describes the last byte to be written) so
528  * that this function can be used to write to the very end-of-file (end = -1).
529  */
530 int filemap_write_and_wait_range(struct address_space *mapping,
531                                  loff_t lstart, loff_t lend)
532 {
533         int err = 0;
534 
535         if (mapping->nrpages) {
536                 err = __filemap_fdatawrite_range(mapping, lstart, lend,
537                                                  WB_SYNC_ALL);
538                 /* See comment of filemap_write_and_wait() */
539                 if (err != -EIO) {
540                         int err2 = filemap_fdatawait_range(mapping,
541                                                 lstart, lend);
542                         if (!err)
543                                 err = err2;
544                 }
545         } else {
546                 err = filemap_check_errors(mapping);
547         }
548         return err;
549 }
550 EXPORT_SYMBOL(filemap_write_and_wait_range);
551 
552 /**
553  * replace_page_cache_page - replace a pagecache page with a new one
554  * @old:        page to be replaced
555  * @new:        page to replace with
556  * @gfp_mask:   allocation mode
557  *
558  * This function replaces a page in the pagecache with a new one.  On
559  * success it acquires the pagecache reference for the new page and
560  * drops it for the old page.  Both the old and new pages must be
561  * locked.  This function does not add the new page to the LRU, the
562  * caller must do that.
563  *
564  * The remove + add is atomic.  The only way this function can fail is
565  * memory allocation failure.
566  */
567 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
568 {
569         int error;
570 
571         VM_BUG_ON_PAGE(!PageLocked(old), old);
572         VM_BUG_ON_PAGE(!PageLocked(new), new);
573         VM_BUG_ON_PAGE(new->mapping, new);
574 
575         error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
576         if (!error) {
577                 struct address_space *mapping = old->mapping;
578                 void (*freepage)(struct page *);
579                 struct mem_cgroup *memcg;
580                 unsigned long flags;
581 
582                 pgoff_t offset = old->index;
583                 freepage = mapping->a_ops->freepage;
584 
585                 page_cache_get(new);
586                 new->mapping = mapping;
587                 new->index = offset;
588 
589                 memcg = mem_cgroup_begin_page_stat(old);
590                 spin_lock_irqsave(&mapping->tree_lock, flags);
591                 __delete_from_page_cache(old, NULL, memcg);
592                 error = page_cache_tree_insert(mapping, new, NULL);
593                 BUG_ON(error);
594 
595                 /*
596                  * hugetlb pages do not participate in page cache accounting.
597                  */
598                 if (!PageHuge(new))
599                         __inc_zone_page_state(new, NR_FILE_PAGES);
600                 if (PageSwapBacked(new))
601                         __inc_zone_page_state(new, NR_SHMEM);
602                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
603                 mem_cgroup_end_page_stat(memcg);
604                 mem_cgroup_replace_page(old, new);
605                 radix_tree_preload_end();
606                 if (freepage)
607                         freepage(old);
608                 page_cache_release(old);
609         }
610 
611         return error;
612 }
613 EXPORT_SYMBOL_GPL(replace_page_cache_page);
614 
615 static int __add_to_page_cache_locked(struct page *page,
616                                       struct address_space *mapping,
617                                       pgoff_t offset, gfp_t gfp_mask,
618                                       void **shadowp)
619 {
620         int huge = PageHuge(page);
621         struct mem_cgroup *memcg;
622         int error;
623 
624         VM_BUG_ON_PAGE(!PageLocked(page), page);
625         VM_BUG_ON_PAGE(PageSwapBacked(page), page);
626 
627         if (!huge) {
628                 error = mem_cgroup_try_charge(page, current->mm,
629                                               gfp_mask, &memcg);
630                 if (error)
631                         return error;
632         }
633 
634         error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
635         if (error) {
636                 if (!huge)
637                         mem_cgroup_cancel_charge(page, memcg);
638                 return error;
639         }
640 
641         page_cache_get(page);
642         page->mapping = mapping;
643         page->index = offset;
644 
645         spin_lock_irq(&mapping->tree_lock);
646         error = page_cache_tree_insert(mapping, page, shadowp);
647         radix_tree_preload_end();
648         if (unlikely(error))
649                 goto err_insert;
650 
651         /* hugetlb pages do not participate in page cache accounting. */
652         if (!huge)
653                 __inc_zone_page_state(page, NR_FILE_PAGES);
654         spin_unlock_irq(&mapping->tree_lock);
655         if (!huge)
656                 mem_cgroup_commit_charge(page, memcg, false);
657         trace_mm_filemap_add_to_page_cache(page);
658         return 0;
659 err_insert:
660         page->mapping = NULL;
661         /* Leave page->index set: truncation relies upon it */
662         spin_unlock_irq(&mapping->tree_lock);
663         if (!huge)
664                 mem_cgroup_cancel_charge(page, memcg);
665         page_cache_release(page);
666         return error;
667 }
668 
669 /**
670  * add_to_page_cache_locked - add a locked page to the pagecache
671  * @page:       page to add
672  * @mapping:    the page's address_space
673  * @offset:     page index
674  * @gfp_mask:   page allocation mode
675  *
676  * This function is used to add a page to the pagecache. It must be locked.
677  * This function does not add the page to the LRU.  The caller must do that.
678  */
679 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
680                 pgoff_t offset, gfp_t gfp_mask)
681 {
682         return __add_to_page_cache_locked(page, mapping, offset,
683                                           gfp_mask, NULL);
684 }
685 EXPORT_SYMBOL(add_to_page_cache_locked);
686 
687 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
688                                 pgoff_t offset, gfp_t gfp_mask)
689 {
690         void *shadow = NULL;
691         int ret;
692 
693         __set_page_locked(page);
694         ret = __add_to_page_cache_locked(page, mapping, offset,
695                                          gfp_mask, &shadow);
696         if (unlikely(ret))
697                 __clear_page_locked(page);
698         else {
699                 /*
700                  * The page might have been evicted from cache only
701                  * recently, in which case it should be activated like
702                  * any other repeatedly accessed page.
703                  */
704                 if (shadow && workingset_refault(shadow)) {
705                         SetPageActive(page);
706                         workingset_activation(page);
707                 } else
708                         ClearPageActive(page);
709                 lru_cache_add(page);
710         }
711         return ret;
712 }
713 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
714 
715 #ifdef CONFIG_NUMA
716 struct page *__page_cache_alloc(gfp_t gfp)
717 {
718         int n;
719         struct page *page;
720 
721         if (cpuset_do_page_mem_spread()) {
722                 unsigned int cpuset_mems_cookie;
723                 do {
724                         cpuset_mems_cookie = read_mems_allowed_begin();
725                         n = cpuset_mem_spread_node();
726                         page = __alloc_pages_node(n, gfp, 0);
727                 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
728 
729                 return page;
730         }
731         return alloc_pages(gfp, 0);
732 }
733 EXPORT_SYMBOL(__page_cache_alloc);
734 #endif
735 
736 /*
737  * In order to wait for pages to become available there must be
738  * waitqueues associated with pages. By using a hash table of
739  * waitqueues where the bucket discipline is to maintain all
740  * waiters on the same queue and wake all when any of the pages
741  * become available, and for the woken contexts to check to be
742  * sure the appropriate page became available, this saves space
743  * at a cost of "thundering herd" phenomena during rare hash
744  * collisions.
745  */
746 wait_queue_head_t *page_waitqueue(struct page *page)
747 {
748         const struct zone *zone = page_zone(page);
749 
750         return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
751 }
752 EXPORT_SYMBOL(page_waitqueue);
753 
754 void wait_on_page_bit(struct page *page, int bit_nr)
755 {
756         DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
757 
758         if (test_bit(bit_nr, &page->flags))
759                 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
760                                                         TASK_UNINTERRUPTIBLE);
761 }
762 EXPORT_SYMBOL(wait_on_page_bit);
763 
764 int wait_on_page_bit_killable(struct page *page, int bit_nr)
765 {
766         DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
767 
768         if (!test_bit(bit_nr, &page->flags))
769                 return 0;
770 
771         return __wait_on_bit(page_waitqueue(page), &wait,
772                              bit_wait_io, TASK_KILLABLE);
773 }
774 
775 int wait_on_page_bit_killable_timeout(struct page *page,
776                                        int bit_nr, unsigned long timeout)
777 {
778         DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
779 
780         wait.key.timeout = jiffies + timeout;
781         if (!test_bit(bit_nr, &page->flags))
782                 return 0;
783         return __wait_on_bit(page_waitqueue(page), &wait,
784                              bit_wait_io_timeout, TASK_KILLABLE);
785 }
786 EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout);
787 
788 /**
789  * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
790  * @page: Page defining the wait queue of interest
791  * @waiter: Waiter to add to the queue
792  *
793  * Add an arbitrary @waiter to the wait queue for the nominated @page.
794  */
795 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
796 {
797         wait_queue_head_t *q = page_waitqueue(page);
798         unsigned long flags;
799 
800         spin_lock_irqsave(&q->lock, flags);
801         __add_wait_queue(q, waiter);
802         spin_unlock_irqrestore(&q->lock, flags);
803 }
804 EXPORT_SYMBOL_GPL(add_page_wait_queue);
805 
806 /**
807  * unlock_page - unlock a locked page
808  * @page: the page
809  *
810  * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
811  * Also wakes sleepers in wait_on_page_writeback() because the wakeup
812  * mechanism between PageLocked pages and PageWriteback pages is shared.
813  * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
814  *
815  * The mb is necessary to enforce ordering between the clear_bit and the read
816  * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
817  */
818 void unlock_page(struct page *page)
819 {
820         VM_BUG_ON_PAGE(!PageLocked(page), page);
821         clear_bit_unlock(PG_locked, &page->flags);
822         smp_mb__after_atomic();
823         wake_up_page(page, PG_locked);
824 }
825 EXPORT_SYMBOL(unlock_page);
826 
827 /**
828  * end_page_writeback - end writeback against a page
829  * @page: the page
830  */
831 void end_page_writeback(struct page *page)
832 {
833         /*
834          * TestClearPageReclaim could be used here but it is an atomic
835          * operation and overkill in this particular case. Failing to
836          * shuffle a page marked for immediate reclaim is too mild to
837          * justify taking an atomic operation penalty at the end of
838          * ever page writeback.
839          */
840         if (PageReclaim(page)) {
841                 ClearPageReclaim(page);
842                 rotate_reclaimable_page(page);
843         }
844 
845         if (!test_clear_page_writeback(page))
846                 BUG();
847 
848         smp_mb__after_atomic();
849         wake_up_page(page, PG_writeback);
850 }
851 EXPORT_SYMBOL(end_page_writeback);
852 
853 /*
854  * After completing I/O on a page, call this routine to update the page
855  * flags appropriately
856  */
857 void page_endio(struct page *page, int rw, int err)
858 {
859         if (rw == READ) {
860                 if (!err) {
861                         SetPageUptodate(page);
862                 } else {
863                         ClearPageUptodate(page);
864                         SetPageError(page);
865                 }
866                 unlock_page(page);
867         } else { /* rw == WRITE */
868                 if (err) {
869                         struct address_space *mapping;
870 
871                         SetPageError(page);
872                         mapping = page_mapping(page);
873                         if (mapping)
874                                 mapping_set_error(mapping, err);
875                 }
876                 end_page_writeback(page);
877         }
878 }
879 EXPORT_SYMBOL_GPL(page_endio);
880 
881 /**
882  * __lock_page - get a lock on the page, assuming we need to sleep to get it
883  * @page: the page to lock
884  */
885 void __lock_page(struct page *page)
886 {
887         DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
888 
889         __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
890                                                         TASK_UNINTERRUPTIBLE);
891 }
892 EXPORT_SYMBOL(__lock_page);
893 
894 int __lock_page_killable(struct page *page)
895 {
896         DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
897 
898         return __wait_on_bit_lock(page_waitqueue(page), &wait,
899                                         bit_wait_io, TASK_KILLABLE);
900 }
901 EXPORT_SYMBOL_GPL(__lock_page_killable);
902 
903 /*
904  * Return values:
905  * 1 - page is locked; mmap_sem is still held.
906  * 0 - page is not locked.
907  *     mmap_sem has been released (up_read()), unless flags had both
908  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
909  *     which case mmap_sem is still held.
910  *
911  * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
912  * with the page locked and the mmap_sem unperturbed.
913  */
914 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
915                          unsigned int flags)
916 {
917         if (flags & FAULT_FLAG_ALLOW_RETRY) {
918                 /*
919                  * CAUTION! In this case, mmap_sem is not released
920                  * even though return 0.
921                  */
922                 if (flags & FAULT_FLAG_RETRY_NOWAIT)
923                         return 0;
924 
925                 up_read(&mm->mmap_sem);
926                 if (flags & FAULT_FLAG_KILLABLE)
927                         wait_on_page_locked_killable(page);
928                 else
929                         wait_on_page_locked(page);
930                 return 0;
931         } else {
932                 if (flags & FAULT_FLAG_KILLABLE) {
933                         int ret;
934 
935                         ret = __lock_page_killable(page);
936                         if (ret) {
937                                 up_read(&mm->mmap_sem);
938                                 return 0;
939                         }
940                 } else
941                         __lock_page(page);
942                 return 1;
943         }
944 }
945 
946 /**
947  * page_cache_next_hole - find the next hole (not-present entry)
948  * @mapping: mapping
949  * @index: index
950  * @max_scan: maximum range to search
951  *
952  * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
953  * lowest indexed hole.
954  *
955  * Returns: the index of the hole if found, otherwise returns an index
956  * outside of the set specified (in which case 'return - index >=
957  * max_scan' will be true). In rare cases of index wrap-around, 0 will
958  * be returned.
959  *
960  * page_cache_next_hole may be called under rcu_read_lock. However,
961  * like radix_tree_gang_lookup, this will not atomically search a
962  * snapshot of the tree at a single point in time. For example, if a
963  * hole is created at index 5, then subsequently a hole is created at
964  * index 10, page_cache_next_hole covering both indexes may return 10
965  * if called under rcu_read_lock.
966  */
967 pgoff_t page_cache_next_hole(struct address_space *mapping,
968                              pgoff_t index, unsigned long max_scan)
969 {
970         unsigned long i;
971 
972         for (i = 0; i < max_scan; i++) {
973                 struct page *page;
974 
975                 page = radix_tree_lookup(&mapping->page_tree, index);
976                 if (!page || radix_tree_exceptional_entry(page))
977                         break;
978                 index++;
979                 if (index == 0)
980                         break;
981         }
982 
983         return index;
984 }
985 EXPORT_SYMBOL(page_cache_next_hole);
986 
987 /**
988  * page_cache_prev_hole - find the prev hole (not-present entry)
989  * @mapping: mapping
990  * @index: index
991  * @max_scan: maximum range to search
992  *
993  * Search backwards in the range [max(index-max_scan+1, 0), index] for
994  * the first hole.
995  *
996  * Returns: the index of the hole if found, otherwise returns an index
997  * outside of the set specified (in which case 'index - return >=
998  * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
999  * will be returned.
1000  *
1001  * page_cache_prev_hole may be called under rcu_read_lock. However,
1002  * like radix_tree_gang_lookup, this will not atomically search a
1003  * snapshot of the tree at a single point in time. For example, if a
1004  * hole is created at index 10, then subsequently a hole is created at
1005  * index 5, page_cache_prev_hole covering both indexes may return 5 if
1006  * called under rcu_read_lock.
1007  */
1008 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1009                              pgoff_t index, unsigned long max_scan)
1010 {
1011         unsigned long i;
1012 
1013         for (i = 0; i < max_scan; i++) {
1014                 struct page *page;
1015 
1016                 page = radix_tree_lookup(&mapping->page_tree, index);
1017                 if (!page || radix_tree_exceptional_entry(page))
1018                         break;
1019                 index--;
1020                 if (index == ULONG_MAX)
1021                         break;
1022         }
1023 
1024         return index;
1025 }
1026 EXPORT_SYMBOL(page_cache_prev_hole);
1027 
1028 /**
1029  * find_get_entry - find and get a page cache entry
1030  * @mapping: the address_space to search
1031  * @offset: the page cache index
1032  *
1033  * Looks up the page cache slot at @mapping & @offset.  If there is a
1034  * page cache page, it is returned with an increased refcount.
1035  *
1036  * If the slot holds a shadow entry of a previously evicted page, or a
1037  * swap entry from shmem/tmpfs, it is returned.
1038  *
1039  * Otherwise, %NULL is returned.
1040  */
1041 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1042 {
1043         void **pagep;
1044         struct page *page;
1045 
1046         rcu_read_lock();
1047 repeat:
1048         page = NULL;
1049         pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1050         if (pagep) {
1051                 page = radix_tree_deref_slot(pagep);
1052                 if (unlikely(!page))
1053                         goto out;
1054                 if (radix_tree_exception(page)) {
1055                         if (radix_tree_deref_retry(page))
1056                                 goto repeat;
1057                         /*
1058                          * A shadow entry of a recently evicted page,
1059                          * or a swap entry from shmem/tmpfs.  Return
1060                          * it without attempting to raise page count.
1061                          */
1062                         goto out;
1063                 }
1064                 if (!page_cache_get_speculative(page))
1065                         goto repeat;
1066 
1067                 /*
1068                  * Has the page moved?
1069                  * This is part of the lockless pagecache protocol. See
1070                  * include/linux/pagemap.h for details.
1071                  */
1072                 if (unlikely(page != *pagep)) {
1073                         page_cache_release(page);
1074                         goto repeat;
1075                 }
1076         }
1077 out:
1078         rcu_read_unlock();
1079 
1080         return page;
1081 }
1082 EXPORT_SYMBOL(find_get_entry);
1083 
1084 /**
1085  * find_lock_entry - locate, pin and lock a page cache entry
1086  * @mapping: the address_space to search
1087  * @offset: the page cache index
1088  *
1089  * Looks up the page cache slot at @mapping & @offset.  If there is a
1090  * page cache page, it is returned locked and with an increased
1091  * refcount.
1092  *
1093  * If the slot holds a shadow entry of a previously evicted page, or a
1094  * swap entry from shmem/tmpfs, it is returned.
1095  *
1096  * Otherwise, %NULL is returned.
1097  *
1098  * find_lock_entry() may sleep.
1099  */
1100 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1101 {
1102         struct page *page;
1103 
1104 repeat:
1105         page = find_get_entry(mapping, offset);
1106         if (page && !radix_tree_exception(page)) {
1107                 lock_page(page);
1108                 /* Has the page been truncated? */
1109                 if (unlikely(page->mapping != mapping)) {
1110                         unlock_page(page);
1111                         page_cache_release(page);
1112                         goto repeat;
1113                 }
1114                 VM_BUG_ON_PAGE(page->index != offset, page);
1115         }
1116         return page;
1117 }
1118 EXPORT_SYMBOL(find_lock_entry);
1119 
1120 /**
1121  * pagecache_get_page - find and get a page reference
1122  * @mapping: the address_space to search
1123  * @offset: the page index
1124  * @fgp_flags: PCG flags
1125  * @gfp_mask: gfp mask to use for the page cache data page allocation
1126  *
1127  * Looks up the page cache slot at @mapping & @offset.
1128  *
1129  * PCG flags modify how the page is returned.
1130  *
1131  * FGP_ACCESSED: the page will be marked accessed
1132  * FGP_LOCK: Page is return locked
1133  * FGP_CREAT: If page is not present then a new page is allocated using
1134  *              @gfp_mask and added to the page cache and the VM's LRU
1135  *              list. The page is returned locked and with an increased
1136  *              refcount. Otherwise, %NULL is returned.
1137  *
1138  * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1139  * if the GFP flags specified for FGP_CREAT are atomic.
1140  *
1141  * If there is a page cache page, it is returned with an increased refcount.
1142  */
1143 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1144         int fgp_flags, gfp_t gfp_mask)
1145 {
1146         struct page *page;
1147 
1148 repeat:
1149         page = find_get_entry(mapping, offset);
1150         if (radix_tree_exceptional_entry(page))
1151                 page = NULL;
1152         if (!page)
1153                 goto no_page;
1154 
1155         if (fgp_flags & FGP_LOCK) {
1156                 if (fgp_flags & FGP_NOWAIT) {
1157                         if (!trylock_page(page)) {
1158                                 page_cache_release(page);
1159                                 return NULL;
1160                         }
1161                 } else {
1162                         lock_page(page);
1163                 }
1164 
1165                 /* Has the page been truncated? */
1166                 if (unlikely(page->mapping != mapping)) {
1167                         unlock_page(page);
1168                         page_cache_release(page);
1169                         goto repeat;
1170                 }
1171                 VM_BUG_ON_PAGE(page->index != offset, page);
1172         }
1173 
1174         if (page && (fgp_flags & FGP_ACCESSED))
1175                 mark_page_accessed(page);
1176 
1177 no_page:
1178         if (!page && (fgp_flags & FGP_CREAT)) {
1179                 int err;
1180                 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1181                         gfp_mask |= __GFP_WRITE;
1182                 if (fgp_flags & FGP_NOFS)
1183                         gfp_mask &= ~__GFP_FS;
1184 
1185                 page = __page_cache_alloc(gfp_mask);
1186                 if (!page)
1187                         return NULL;
1188 
1189                 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1190                         fgp_flags |= FGP_LOCK;
1191 
1192                 /* Init accessed so avoid atomic mark_page_accessed later */
1193                 if (fgp_flags & FGP_ACCESSED)
1194                         __SetPageReferenced(page);
1195 
1196                 err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1197                 if (unlikely(err)) {
1198                         page_cache_release(page);
1199                         page = NULL;
1200                         if (err == -EEXIST)
1201                                 goto repeat;
1202                 }
1203         }
1204 
1205         return page;
1206 }
1207 EXPORT_SYMBOL(pagecache_get_page);
1208 
1209 /**
1210  * find_get_entries - gang pagecache lookup
1211  * @mapping:    The address_space to search
1212  * @start:      The starting page cache index
1213  * @nr_entries: The maximum number of entries
1214  * @entries:    Where the resulting entries are placed
1215  * @indices:    The cache indices corresponding to the entries in @entries
1216  *
1217  * find_get_entries() will search for and return a group of up to
1218  * @nr_entries entries in the mapping.  The entries are placed at
1219  * @entries.  find_get_entries() takes a reference against any actual
1220  * pages it returns.
1221  *
1222  * The search returns a group of mapping-contiguous page cache entries
1223  * with ascending indexes.  There may be holes in the indices due to
1224  * not-present pages.
1225  *
1226  * Any shadow entries of evicted pages, or swap entries from
1227  * shmem/tmpfs, are included in the returned array.
1228  *
1229  * find_get_entries() returns the number of pages and shadow entries
1230  * which were found.
1231  */
1232 unsigned find_get_entries(struct address_space *mapping,
1233                           pgoff_t start, unsigned int nr_entries,
1234                           struct page **entries, pgoff_t *indices)
1235 {
1236         void **slot;
1237         unsigned int ret = 0;
1238         struct radix_tree_iter iter;
1239 
1240         if (!nr_entries)
1241                 return 0;
1242 
1243         rcu_read_lock();
1244 restart:
1245         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1246                 struct page *page;
1247 repeat:
1248                 page = radix_tree_deref_slot(slot);
1249                 if (unlikely(!page))
1250                         continue;
1251                 if (radix_tree_exception(page)) {
1252                         if (radix_tree_deref_retry(page))
1253                                 goto restart;
1254                         /*
1255                          * A shadow entry of a recently evicted page,
1256                          * or a swap entry from shmem/tmpfs.  Return
1257                          * it without attempting to raise page count.
1258                          */
1259                         goto export;
1260                 }
1261                 if (!page_cache_get_speculative(page))
1262                         goto repeat;
1263 
1264                 /* Has the page moved? */
1265                 if (unlikely(page != *slot)) {
1266                         page_cache_release(page);
1267                         goto repeat;
1268                 }
1269 export:
1270                 indices[ret] = iter.index;
1271                 entries[ret] = page;
1272                 if (++ret == nr_entries)
1273                         break;
1274         }
1275         rcu_read_unlock();
1276         return ret;
1277 }
1278 
1279 /**
1280  * find_get_pages - gang pagecache lookup
1281  * @mapping:    The address_space to search
1282  * @start:      The starting page index
1283  * @nr_pages:   The maximum number of pages
1284  * @pages:      Where the resulting pages are placed
1285  *
1286  * find_get_pages() will search for and return a group of up to
1287  * @nr_pages pages in the mapping.  The pages are placed at @pages.
1288  * find_get_pages() takes a reference against the returned pages.
1289  *
1290  * The search returns a group of mapping-contiguous pages with ascending
1291  * indexes.  There may be holes in the indices due to not-present pages.
1292  *
1293  * find_get_pages() returns the number of pages which were found.
1294  */
1295 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1296                             unsigned int nr_pages, struct page **pages)
1297 {
1298         struct radix_tree_iter iter;
1299         void **slot;
1300         unsigned ret = 0;
1301 
1302         if (unlikely(!nr_pages))
1303                 return 0;
1304 
1305         rcu_read_lock();
1306 restart:
1307         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1308                 struct page *page;
1309 repeat:
1310                 page = radix_tree_deref_slot(slot);
1311                 if (unlikely(!page))
1312                         continue;
1313 
1314                 if (radix_tree_exception(page)) {
1315                         if (radix_tree_deref_retry(page)) {
1316                                 /*
1317                                  * Transient condition which can only trigger
1318                                  * when entry at index 0 moves out of or back
1319                                  * to root: none yet gotten, safe to restart.
1320                                  */
1321                                 WARN_ON(iter.index);
1322                                 goto restart;
1323                         }
1324                         /*
1325                          * A shadow entry of a recently evicted page,
1326                          * or a swap entry from shmem/tmpfs.  Skip
1327                          * over it.
1328                          */
1329                         continue;
1330                 }
1331 
1332                 if (!page_cache_get_speculative(page))
1333                         goto repeat;
1334 
1335                 /* Has the page moved? */
1336                 if (unlikely(page != *slot)) {
1337                         page_cache_release(page);
1338                         goto repeat;
1339                 }
1340 
1341                 pages[ret] = page;
1342                 if (++ret == nr_pages)
1343                         break;
1344         }
1345 
1346         rcu_read_unlock();
1347         return ret;
1348 }
1349 
1350 /**
1351  * find_get_pages_contig - gang contiguous pagecache lookup
1352  * @mapping:    The address_space to search
1353  * @index:      The starting page index
1354  * @nr_pages:   The maximum number of pages
1355  * @pages:      Where the resulting pages are placed
1356  *
1357  * find_get_pages_contig() works exactly like find_get_pages(), except
1358  * that the returned number of pages are guaranteed to be contiguous.
1359  *
1360  * find_get_pages_contig() returns the number of pages which were found.
1361  */
1362 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1363                                unsigned int nr_pages, struct page **pages)
1364 {
1365         struct radix_tree_iter iter;
1366         void **slot;
1367         unsigned int ret = 0;
1368 
1369         if (unlikely(!nr_pages))
1370                 return 0;
1371 
1372         rcu_read_lock();
1373 restart:
1374         radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1375                 struct page *page;
1376 repeat:
1377                 page = radix_tree_deref_slot(slot);
1378                 /* The hole, there no reason to continue */
1379                 if (unlikely(!page))
1380                         break;
1381 
1382                 if (radix_tree_exception(page)) {
1383                         if (radix_tree_deref_retry(page)) {
1384                                 /*
1385                                  * Transient condition which can only trigger
1386                                  * when entry at index 0 moves out of or back
1387                                  * to root: none yet gotten, safe to restart.
1388                                  */
1389                                 goto restart;
1390                         }
1391                         /*
1392                          * A shadow entry of a recently evicted page,
1393                          * or a swap entry from shmem/tmpfs.  Stop
1394                          * looking for contiguous pages.
1395                          */
1396                         break;
1397                 }
1398 
1399                 if (!page_cache_get_speculative(page))
1400                         goto repeat;
1401 
1402                 /* Has the page moved? */
1403                 if (unlikely(page != *slot)) {
1404                         page_cache_release(page);
1405                         goto repeat;
1406                 }
1407 
1408                 /*
1409                  * must check mapping and index after taking the ref.
1410                  * otherwise we can get both false positives and false
1411                  * negatives, which is just confusing to the caller.
1412                  */
1413                 if (page->mapping == NULL || page->index != iter.index) {
1414                         page_cache_release(page);
1415                         break;
1416                 }
1417 
1418                 pages[ret] = page;
1419                 if (++ret == nr_pages)
1420                         break;
1421         }
1422         rcu_read_unlock();
1423         return ret;
1424 }
1425 EXPORT_SYMBOL(find_get_pages_contig);
1426 
1427 /**
1428  * find_get_pages_tag - find and return pages that match @tag
1429  * @mapping:    the address_space to search
1430  * @index:      the starting page index
1431  * @tag:        the tag index
1432  * @nr_pages:   the maximum number of pages
1433  * @pages:      where the resulting pages are placed
1434  *
1435  * Like find_get_pages, except we only return pages which are tagged with
1436  * @tag.   We update @index to index the next page for the traversal.
1437  */
1438 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1439                         int tag, unsigned int nr_pages, struct page **pages)
1440 {
1441         struct radix_tree_iter iter;
1442         void **slot;
1443         unsigned ret = 0;
1444 
1445         if (unlikely(!nr_pages))
1446                 return 0;
1447 
1448         rcu_read_lock();
1449 restart:
1450         radix_tree_for_each_tagged(slot, &mapping->page_tree,
1451                                    &iter, *index, tag) {
1452                 struct page *page;
1453 repeat:
1454                 page = radix_tree_deref_slot(slot);
1455                 if (unlikely(!page))
1456                         continue;
1457 
1458                 if (radix_tree_exception(page)) {
1459                         if (radix_tree_deref_retry(page)) {
1460                                 /*
1461                                  * Transient condition which can only trigger
1462                                  * when entry at index 0 moves out of or back
1463                                  * to root: none yet gotten, safe to restart.
1464                                  */
1465                                 goto restart;
1466                         }
1467                         /*
1468                          * A shadow entry of a recently evicted page.
1469                          *
1470                          * Those entries should never be tagged, but
1471                          * this tree walk is lockless and the tags are
1472                          * looked up in bulk, one radix tree node at a
1473                          * time, so there is a sizable window for page
1474                          * reclaim to evict a page we saw tagged.
1475                          *
1476                          * Skip over it.
1477                          */
1478                         continue;
1479                 }
1480 
1481                 if (!page_cache_get_speculative(page))
1482                         goto repeat;
1483 
1484                 /* Has the page moved? */
1485                 if (unlikely(page != *slot)) {
1486                         page_cache_release(page);
1487                         goto repeat;
1488                 }
1489 
1490                 pages[ret] = page;
1491                 if (++ret == nr_pages)
1492                         break;
1493         }
1494 
1495         rcu_read_unlock();
1496 
1497         if (ret)
1498                 *index = pages[ret - 1]->index + 1;
1499 
1500         return ret;
1501 }
1502 EXPORT_SYMBOL(find_get_pages_tag);
1503 
1504 /*
1505  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1506  * a _large_ part of the i/o request. Imagine the worst scenario:
1507  *
1508  *      ---R__________________________________________B__________
1509  *         ^ reading here                             ^ bad block(assume 4k)
1510  *
1511  * read(R) => miss => readahead(R...B) => media error => frustrating retries
1512  * => failing the whole request => read(R) => read(R+1) =>
1513  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1514  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1515  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1516  *
1517  * It is going insane. Fix it by quickly scaling down the readahead size.
1518  */
1519 static void shrink_readahead_size_eio(struct file *filp,
1520                                         struct file_ra_state *ra)
1521 {
1522         ra->ra_pages /= 4;
1523 }
1524 
1525 /**
1526  * do_generic_file_read - generic file read routine
1527  * @filp:       the file to read
1528  * @ppos:       current file position
1529  * @iter:       data destination
1530  * @written:    already copied
1531  *
1532  * This is a generic file read routine, and uses the
1533  * mapping->a_ops->readpage() function for the actual low-level stuff.
1534  *
1535  * This is really ugly. But the goto's actually try to clarify some
1536  * of the logic when it comes to error handling etc.
1537  */
1538 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1539                 struct iov_iter *iter, ssize_t written)
1540 {
1541         struct address_space *mapping = filp->f_mapping;
1542         struct inode *inode = mapping->host;
1543         struct file_ra_state *ra = &filp->f_ra;
1544         pgoff_t index;
1545         pgoff_t last_index;
1546         pgoff_t prev_index;
1547         unsigned long offset;      /* offset into pagecache page */
1548         unsigned int prev_offset;
1549         int error = 0;
1550 
1551         index = *ppos >> PAGE_CACHE_SHIFT;
1552         prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1553         prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1554         last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1555         offset = *ppos & ~PAGE_CACHE_MASK;
1556 
1557         for (;;) {
1558                 struct page *page;
1559                 pgoff_t end_index;
1560                 loff_t isize;
1561                 unsigned long nr, ret;
1562 
1563                 cond_resched();
1564 find_page:
1565                 if (fatal_signal_pending(current)) {
1566                         error = -EINTR;
1567                         goto out;
1568                 }
1569 
1570                 page = find_get_page(mapping, index);
1571                 if (!page) {
1572                         page_cache_sync_readahead(mapping,
1573                                         ra, filp,
1574                                         index, last_index - index);
1575                         page = find_get_page(mapping, index);
1576                         if (unlikely(page == NULL))
1577                                 goto no_cached_page;
1578                 }
1579                 if (PageReadahead(page)) {
1580                         page_cache_async_readahead(mapping,
1581                                         ra, filp, page,
1582                                         index, last_index - index);
1583                 }
1584                 if (!PageUptodate(page)) {
1585                         /*
1586                          * See comment in do_read_cache_page on why
1587                          * wait_on_page_locked is used to avoid unnecessarily
1588                          * serialisations and why it's safe.
1589                          */
1590                         wait_on_page_locked_killable(page);
1591                         if (PageUptodate(page))
1592                                 goto page_ok;
1593 
1594                         if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1595                                         !mapping->a_ops->is_partially_uptodate)
1596                                 goto page_not_up_to_date;
1597                         if (!trylock_page(page))
1598                                 goto page_not_up_to_date;
1599                         /* Did it get truncated before we got the lock? */
1600                         if (!page->mapping)
1601                                 goto page_not_up_to_date_locked;
1602                         if (!mapping->a_ops->is_partially_uptodate(page,
1603                                                         offset, iter->count))
1604                                 goto page_not_up_to_date_locked;
1605                         unlock_page(page);
1606                 }
1607 page_ok:
1608                 /*
1609                  * i_size must be checked after we know the page is Uptodate.
1610                  *
1611                  * Checking i_size after the check allows us to calculate
1612                  * the correct value for "nr", which means the zero-filled
1613                  * part of the page is not copied back to userspace (unless
1614                  * another truncate extends the file - this is desired though).
1615                  */
1616 
1617                 isize = i_size_read(inode);
1618                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1619                 if (unlikely(!isize || index > end_index)) {
1620                         page_cache_release(page);
1621                         goto out;
1622                 }
1623 
1624                 /* nr is the maximum number of bytes to copy from this page */
1625                 nr = PAGE_CACHE_SIZE;
1626                 if (index == end_index) {
1627                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1628                         if (nr <= offset) {
1629                                 page_cache_release(page);
1630                                 goto out;
1631                         }
1632                 }
1633                 nr = nr - offset;
1634 
1635                 /* If users can be writing to this page using arbitrary
1636                  * virtual addresses, take care about potential aliasing
1637                  * before reading the page on the kernel side.
1638                  */
1639                 if (mapping_writably_mapped(mapping))
1640                         flush_dcache_page(page);
1641 
1642                 /*
1643                  * When a sequential read accesses a page several times,
1644                  * only mark it as accessed the first time.
1645                  */
1646                 if (prev_index != index || offset != prev_offset)
1647                         mark_page_accessed(page);
1648                 prev_index = index;
1649 
1650                 /*
1651                  * Ok, we have the page, and it's up-to-date, so
1652                  * now we can copy it to user space...
1653                  */
1654 
1655                 ret = copy_page_to_iter(page, offset, nr, iter);
1656                 offset += ret;
1657                 index += offset >> PAGE_CACHE_SHIFT;
1658                 offset &= ~PAGE_CACHE_MASK;
1659                 prev_offset = offset;
1660 
1661                 page_cache_release(page);
1662                 written += ret;
1663                 if (!iov_iter_count(iter))
1664                         goto out;
1665                 if (ret < nr) {
1666                         error = -EFAULT;
1667                         goto out;
1668                 }
1669                 continue;
1670 
1671 page_not_up_to_date:
1672                 /* Get exclusive access to the page ... */
1673                 error = lock_page_killable(page);
1674                 if (unlikely(error))
1675                         goto readpage_error;
1676 
1677 page_not_up_to_date_locked:
1678                 /* Did it get truncated before we got the lock? */
1679                 if (!page->mapping) {
1680                         unlock_page(page);
1681                         page_cache_release(page);
1682                         continue;
1683                 }
1684 
1685                 /* Did somebody else fill it already? */
1686                 if (PageUptodate(page)) {
1687                         unlock_page(page);
1688                         goto page_ok;
1689                 }
1690 
1691 readpage:
1692                 /*
1693                  * A previous I/O error may have been due to temporary
1694                  * failures, eg. multipath errors.
1695                  * PG_error will be set again if readpage fails.
1696                  */
1697                 ClearPageError(page);
1698                 /* Start the actual read. The read will unlock the page. */
1699                 error = mapping->a_ops->readpage(filp, page);
1700 
1701                 if (unlikely(error)) {
1702                         if (error == AOP_TRUNCATED_PAGE) {
1703                                 page_cache_release(page);
1704                                 error = 0;
1705                                 goto find_page;
1706                         }
1707                         goto readpage_error;
1708                 }
1709 
1710                 if (!PageUptodate(page)) {
1711                         error = lock_page_killable(page);
1712                         if (unlikely(error))
1713                                 goto readpage_error;
1714                         if (!PageUptodate(page)) {
1715                                 if (page->mapping == NULL) {
1716                                         /*
1717                                          * invalidate_mapping_pages got it
1718                                          */
1719                                         unlock_page(page);
1720                                         page_cache_release(page);
1721                                         goto find_page;
1722                                 }
1723                                 unlock_page(page);
1724                                 shrink_readahead_size_eio(filp, ra);
1725                                 error = -EIO;
1726                                 goto readpage_error;
1727                         }
1728                         unlock_page(page);
1729                 }
1730 
1731                 goto page_ok;
1732 
1733 readpage_error:
1734                 /* UHHUH! A synchronous read error occurred. Report it */
1735                 page_cache_release(page);
1736                 goto out;
1737 
1738 no_cached_page:
1739                 /*
1740                  * Ok, it wasn't cached, so we need to create a new
1741                  * page..
1742                  */
1743                 page = page_cache_alloc_cold(mapping);
1744                 if (!page) {
1745                         error = -ENOMEM;
1746                         goto out;
1747                 }
1748                 error = add_to_page_cache_lru(page, mapping, index,
1749                                 mapping_gfp_constraint(mapping, GFP_KERNEL));
1750                 if (error) {
1751                         page_cache_release(page);
1752                         if (error == -EEXIST) {
1753                                 error = 0;
1754                                 goto find_page;
1755                         }
1756                         goto out;
1757                 }
1758                 goto readpage;
1759         }
1760 
1761 out:
1762         ra->prev_pos = prev_index;
1763         ra->prev_pos <<= PAGE_CACHE_SHIFT;
1764         ra->prev_pos |= prev_offset;
1765 
1766         *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1767         file_accessed(filp);
1768         return written ? written : error;
1769 }
1770 
1771 /**
1772  * generic_file_read_iter - generic filesystem read routine
1773  * @iocb:       kernel I/O control block
1774  * @iter:       destination for the data read
1775  *
1776  * This is the "read_iter()" routine for all filesystems
1777  * that can use the page cache directly.
1778  */
1779 ssize_t
1780 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1781 {
1782         struct file *file = iocb->ki_filp;
1783         ssize_t retval = 0;
1784         loff_t *ppos = &iocb->ki_pos;
1785         loff_t pos = *ppos;
1786 
1787         if (iocb->ki_flags & IOCB_DIRECT) {
1788                 struct address_space *mapping = file->f_mapping;
1789                 struct inode *inode = mapping->host;
1790                 size_t count = iov_iter_count(iter);
1791                 loff_t size;
1792 
1793                 if (!count)
1794                         goto out; /* skip atime */
1795                 size = i_size_read(inode);
1796                 retval = filemap_write_and_wait_range(mapping, pos,
1797                                         pos + count - 1);
1798                 if (!retval) {
1799                         struct iov_iter data = *iter;
1800                         retval = mapping->a_ops->direct_IO(iocb, &data, pos);
1801                 }
1802 
1803                 if (retval > 0) {
1804                         *ppos = pos + retval;
1805                         iov_iter_advance(iter, retval);
1806                 }
1807 
1808                 /*
1809                  * Btrfs can have a short DIO read if we encounter
1810                  * compressed extents, so if there was an error, or if
1811                  * we've already read everything we wanted to, or if
1812                  * there was a short read because we hit EOF, go ahead
1813                  * and return.  Otherwise fallthrough to buffered io for
1814                  * the rest of the read.  Buffered reads will not work for
1815                  * DAX files, so don't bother trying.
1816                  */
1817                 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size ||
1818                     IS_DAX(inode)) {
1819                         file_accessed(file);
1820                         goto out;
1821                 }
1822         }
1823 
1824         retval = do_generic_file_read(file, ppos, iter, retval);
1825 out:
1826         return retval;
1827 }
1828 EXPORT_SYMBOL(generic_file_read_iter);
1829 
1830 #ifdef CONFIG_MMU
1831 /**
1832  * page_cache_read - adds requested page to the page cache if not already there
1833  * @file:       file to read
1834  * @offset:     page index
1835  *
1836  * This adds the requested page to the page cache if it isn't already there,
1837  * and schedules an I/O to read in its contents from disk.
1838  */
1839 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
1840 {
1841         struct address_space *mapping = file->f_mapping;
1842         struct page *page;
1843         int ret;
1844 
1845         do {
1846                 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
1847                 if (!page)
1848                         return -ENOMEM;
1849 
1850                 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1851                 if (ret == 0)
1852                         ret = mapping->a_ops->readpage(file, page);
1853                 else if (ret == -EEXIST)
1854                         ret = 0; /* losing race to add is OK */
1855 
1856                 page_cache_release(page);
1857 
1858         } while (ret == AOP_TRUNCATED_PAGE);
1859 
1860         return ret;
1861 }
1862 
1863 #define MMAP_LOTSAMISS  (100)
1864 
1865 /*
1866  * Synchronous readahead happens when we don't even find
1867  * a page in the page cache at all.
1868  */
1869 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1870                                    struct file_ra_state *ra,
1871                                    struct file *file,
1872                                    pgoff_t offset)
1873 {
1874         struct address_space *mapping = file->f_mapping;
1875 
1876         /* If we don't want any read-ahead, don't bother */
1877         if (vma->vm_flags & VM_RAND_READ)
1878                 return;
1879         if (!ra->ra_pages)
1880                 return;
1881 
1882         if (vma->vm_flags & VM_SEQ_READ) {
1883                 page_cache_sync_readahead(mapping, ra, file, offset,
1884                                           ra->ra_pages);
1885                 return;
1886         }
1887 
1888         /* Avoid banging the cache line if not needed */
1889         if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1890                 ra->mmap_miss++;
1891 
1892         /*
1893          * Do we miss much more than hit in this file? If so,
1894          * stop bothering with read-ahead. It will only hurt.
1895          */
1896         if (ra->mmap_miss > MMAP_LOTSAMISS)
1897                 return;
1898 
1899         /*
1900          * mmap read-around
1901          */
1902         ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
1903         ra->size = ra->ra_pages;
1904         ra->async_size = ra->ra_pages / 4;
1905         ra_submit(ra, mapping, file);
1906 }
1907 
1908 /*
1909  * Asynchronous readahead happens when we find the page and PG_readahead,
1910  * so we want to possibly extend the readahead further..
1911  */
1912 static void do_async_mmap_readahead(struct vm_area_struct *vma,
1913                                     struct file_ra_state *ra,
1914                                     struct file *file,
1915                                     struct page *page,
1916                                     pgoff_t offset)
1917 {
1918         struct address_space *mapping = file->f_mapping;
1919 
1920         /* If we don't want any read-ahead, don't bother */
1921         if (vma->vm_flags & VM_RAND_READ)
1922                 return;
1923         if (ra->mmap_miss > 0)
1924                 ra->mmap_miss--;
1925         if (PageReadahead(page))
1926                 page_cache_async_readahead(mapping, ra, file,
1927                                            page, offset, ra->ra_pages);
1928 }
1929 
1930 /**
1931  * filemap_fault - read in file data for page fault handling
1932  * @vma:        vma in which the fault was taken
1933  * @vmf:        struct vm_fault containing details of the fault
1934  *
1935  * filemap_fault() is invoked via the vma operations vector for a
1936  * mapped memory region to read in file data during a page fault.
1937  *
1938  * The goto's are kind of ugly, but this streamlines the normal case of having
1939  * it in the page cache, and handles the special cases reasonably without
1940  * having a lot of duplicated code.
1941  *
1942  * vma->vm_mm->mmap_sem must be held on entry.
1943  *
1944  * If our return value has VM_FAULT_RETRY set, it's because
1945  * lock_page_or_retry() returned 0.
1946  * The mmap_sem has usually been released in this case.
1947  * See __lock_page_or_retry() for the exception.
1948  *
1949  * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1950  * has not been released.
1951  *
1952  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1953  */
1954 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1955 {
1956         int error;
1957         struct file *file = vma->vm_file;
1958         struct address_space *mapping = file->f_mapping;
1959         struct file_ra_state *ra = &file->f_ra;
1960         struct inode *inode = mapping->host;
1961         pgoff_t offset = vmf->pgoff;
1962         struct page *page;
1963         loff_t size;
1964         int ret = 0;
1965 
1966         size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1967         if (offset >= size >> PAGE_CACHE_SHIFT)
1968                 return VM_FAULT_SIGBUS;
1969 
1970         /*
1971          * Do we have something in the page cache already?
1972          */
1973         page = find_get_page(mapping, offset);
1974         if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1975                 /*
1976                  * We found the page, so try async readahead before
1977                  * waiting for the lock.
1978                  */
1979                 do_async_mmap_readahead(vma, ra, file, page, offset);
1980         } else if (!page) {
1981                 /* No page in the page cache at all */
1982                 do_sync_mmap_readahead(vma, ra, file, offset);
1983                 count_vm_event(PGMAJFAULT);
1984                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1985                 ret = VM_FAULT_MAJOR;
1986 retry_find:
1987                 page = find_get_page(mapping, offset);
1988                 if (!page)
1989                         goto no_cached_page;
1990         }
1991 
1992         if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1993                 page_cache_release(page);
1994                 return ret | VM_FAULT_RETRY;
1995         }
1996 
1997         /* Did it get truncated? */
1998         if (unlikely(page->mapping != mapping)) {
1999                 unlock_page(page);
2000                 put_page(page);
2001                 goto retry_find;
2002         }
2003         VM_BUG_ON_PAGE(page->index != offset, page);
2004 
2005         /*
2006          * We have a locked page in the page cache, now we need to check
2007          * that it's up-to-date. If not, it is going to be due to an error.
2008          */
2009         if (unlikely(!PageUptodate(page)))
2010                 goto page_not_uptodate;
2011 
2012         /*
2013          * Found the page and have a reference on it.
2014          * We must recheck i_size under page lock.
2015          */
2016         size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
2017         if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
2018                 unlock_page(page);
2019                 page_cache_release(page);
2020                 return VM_FAULT_SIGBUS;
2021         }
2022 
2023         vmf->page = page;
2024         return ret | VM_FAULT_LOCKED;
2025 
2026 no_cached_page:
2027         /*
2028          * We're only likely to ever get here if MADV_RANDOM is in
2029          * effect.
2030          */
2031         error = page_cache_read(file, offset, vmf->gfp_mask);
2032 
2033         /*
2034          * The page we want has now been added to the page cache.
2035          * In the unlikely event that someone removed it in the
2036          * meantime, we'll just come back here and read it again.
2037          */
2038         if (error >= 0)
2039                 goto retry_find;
2040 
2041         /*
2042          * An error return from page_cache_read can result if the
2043          * system is low on memory, or a problem occurs while trying
2044          * to schedule I/O.
2045          */
2046         if (error == -ENOMEM)
2047                 return VM_FAULT_OOM;
2048         return VM_FAULT_SIGBUS;
2049 
2050 page_not_uptodate:
2051         /*
2052          * Umm, take care of errors if the page isn't up-to-date.
2053          * Try to re-read it _once_. We do this synchronously,
2054          * because there really aren't any performance issues here
2055          * and we need to check for errors.
2056          */
2057         ClearPageError(page);
2058         error = mapping->a_ops->readpage(file, page);
2059         if (!error) {
2060                 wait_on_page_locked(page);
2061                 if (!PageUptodate(page))
2062                         error = -EIO;
2063         }
2064         page_cache_release(page);
2065 
2066         if (!error || error == AOP_TRUNCATED_PAGE)
2067                 goto retry_find;
2068 
2069         /* Things didn't work out. Return zero to tell the mm layer so. */
2070         shrink_readahead_size_eio(file, ra);
2071         return VM_FAULT_SIGBUS;
2072 }
2073 EXPORT_SYMBOL(filemap_fault);
2074 
2075 void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2076 {
2077         struct radix_tree_iter iter;
2078         void **slot;
2079         struct file *file = vma->vm_file;
2080         struct address_space *mapping = file->f_mapping;
2081         loff_t size;
2082         struct page *page;
2083         unsigned long address = (unsigned long) vmf->virtual_address;
2084         unsigned long addr;
2085         pte_t *pte;
2086 
2087         rcu_read_lock();
2088         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2089                 if (iter.index > vmf->max_pgoff)
2090                         break;
2091 repeat:
2092                 page = radix_tree_deref_slot(slot);
2093                 if (unlikely(!page))
2094                         goto next;
2095                 if (radix_tree_exception(page)) {
2096                         if (radix_tree_deref_retry(page))
2097                                 break;
2098                         else
2099                                 goto next;
2100                 }
2101 
2102                 if (!page_cache_get_speculative(page))
2103                         goto repeat;
2104 
2105                 /* Has the page moved? */
2106                 if (unlikely(page != *slot)) {
2107                         page_cache_release(page);
2108                         goto repeat;
2109                 }
2110 
2111                 if (!PageUptodate(page) ||
2112                                 PageReadahead(page) ||
2113                                 PageHWPoison(page))
2114                         goto skip;
2115                 if (!trylock_page(page))
2116                         goto skip;
2117 
2118                 if (page->mapping != mapping || !PageUptodate(page))
2119                         goto unlock;
2120 
2121                 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2122                 if (page->index >= size >> PAGE_CACHE_SHIFT)
2123                         goto unlock;
2124 
2125                 pte = vmf->pte + page->index - vmf->pgoff;
2126                 if (!pte_none(*pte))
2127                         goto unlock;
2128 
2129                 if (file->f_ra.mmap_miss > 0)
2130                         file->f_ra.mmap_miss--;
2131                 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2132                 do_set_pte(vma, addr, page, pte, false, false);
2133                 unlock_page(page);
2134                 goto next;
2135 unlock:
2136                 unlock_page(page);
2137 skip:
2138                 page_cache_release(page);
2139 next:
2140                 if (iter.index == vmf->max_pgoff)
2141                         break;
2142         }
2143         rcu_read_unlock();
2144 }
2145 EXPORT_SYMBOL(filemap_map_pages);
2146 
2147 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2148 {
2149         struct page *page = vmf->page;
2150         struct inode *inode = file_inode(vma->vm_file);
2151         int ret = VM_FAULT_LOCKED;
2152 
2153         sb_start_pagefault(inode->i_sb);
2154         file_update_time(vma->vm_file);
2155         lock_page(page);
2156         if (page->mapping != inode->i_mapping) {
2157                 unlock_page(page);
2158                 ret = VM_FAULT_NOPAGE;
2159                 goto out;
2160         }
2161         /*
2162          * We mark the page dirty already here so that when freeze is in
2163          * progress, we are guaranteed that writeback during freezing will
2164          * see the dirty page and writeprotect it again.
2165          */
2166         set_page_dirty(page);
2167         wait_for_stable_page(page);
2168 out:
2169         sb_end_pagefault(inode->i_sb);
2170         return ret;
2171 }
2172 EXPORT_SYMBOL(filemap_page_mkwrite);
2173 
2174 const struct vm_operations_struct generic_file_vm_ops = {
2175         .fault          = filemap_fault,
2176         .map_pages      = filemap_map_pages,
2177         .page_mkwrite   = filemap_page_mkwrite,
2178 };
2179 
2180 /* This is used for a general mmap of a disk file */
2181 
2182 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2183 {
2184         struct address_space *mapping = file->f_mapping;
2185 
2186         if (!mapping->a_ops->readpage)
2187                 return -ENOEXEC;
2188         file_accessed(file);
2189         vma->vm_ops = &generic_file_vm_ops;
2190         return 0;
2191 }
2192 
2193 /*
2194  * This is for filesystems which do not implement ->writepage.
2195  */
2196 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2197 {
2198         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2199                 return -EINVAL;
2200         return generic_file_mmap(file, vma);
2201 }
2202 #else
2203 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2204 {
2205         return -ENOSYS;
2206 }
2207 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2208 {
2209         return -ENOSYS;
2210 }
2211 #endif /* CONFIG_MMU */
2212 
2213 EXPORT_SYMBOL(generic_file_mmap);
2214 EXPORT_SYMBOL(generic_file_readonly_mmap);
2215 
2216 static struct page *wait_on_page_read(struct page *page)
2217 {
2218         if (!IS_ERR(page)) {
2219                 wait_on_page_locked(page);
2220                 if (!PageUptodate(page)) {
2221                         page_cache_release(page);
2222                         page = ERR_PTR(-EIO);
2223                 }
2224         }
2225         return page;
2226 }
2227 
2228 static struct page *do_read_cache_page(struct address_space *mapping,
2229                                 pgoff_t index,
2230                                 int (*filler)(void *, struct page *),
2231                                 void *data,
2232                                 gfp_t gfp)
2233 {
2234         struct page *page;
2235         int err;
2236 repeat:
2237         page = find_get_page(mapping, index);
2238         if (!page) {
2239                 page = __page_cache_alloc(gfp | __GFP_COLD);
2240                 if (!page)
2241                         return ERR_PTR(-ENOMEM);
2242                 err = add_to_page_cache_lru(page, mapping, index, gfp);
2243                 if (unlikely(err)) {
2244                         page_cache_release(page);
2245                         if (err == -EEXIST)
2246                                 goto repeat;
2247                         /* Presumably ENOMEM for radix tree node */
2248                         return ERR_PTR(err);
2249                 }
2250 
2251 filler:
2252                 err = filler(data, page);
2253                 if (err < 0) {
2254                         page_cache_release(page);
2255                         return ERR_PTR(err);
2256                 }
2257 
2258                 page = wait_on_page_read(page);
2259                 if (IS_ERR(page))
2260                         return page;
2261                 goto out;
2262         }
2263         if (PageUptodate(page))
2264                 goto out;
2265 
2266         /*
2267          * Page is not up to date and may be locked due one of the following
2268          * case a: Page is being filled and the page lock is held
2269          * case b: Read/write error clearing the page uptodate status
2270          * case c: Truncation in progress (page locked)
2271          * case d: Reclaim in progress
2272          *
2273          * Case a, the page will be up to date when the page is unlocked.
2274          *    There is no need to serialise on the page lock here as the page
2275          *    is pinned so the lock gives no additional protection. Even if the
2276          *    the page is truncated, the data is still valid if PageUptodate as
2277          *    it's a race vs truncate race.
2278          * Case b, the page will not be up to date
2279          * Case c, the page may be truncated but in itself, the data may still
2280          *    be valid after IO completes as it's a read vs truncate race. The
2281          *    operation must restart if the page is not uptodate on unlock but
2282          *    otherwise serialising on page lock to stabilise the mapping gives
2283          *    no additional guarantees to the caller as the page lock is
2284          *    released before return.
2285          * Case d, similar to truncation. If reclaim holds the page lock, it
2286          *    will be a race with remove_mapping that determines if the mapping
2287          *    is valid on unlock but otherwise the data is valid and there is
2288          *    no need to serialise with page lock.
2289          *
2290          * As the page lock gives no additional guarantee, we optimistically
2291          * wait on the page to be unlocked and check if it's up to date and
2292          * use the page if it is. Otherwise, the page lock is required to
2293          * distinguish between the different cases. The motivation is that we
2294          * avoid spurious serialisations and wakeups when multiple processes
2295          * wait on the same page for IO to complete.
2296          */
2297         wait_on_page_locked(page);
2298         if (PageUptodate(page))
2299                 goto out;
2300 
2301         /* Distinguish between all the cases under the safety of the lock */
2302         lock_page(page);
2303 
2304         /* Case c or d, restart the operation */
2305         if (!page->mapping) {
2306                 unlock_page(page);
2307                 page_cache_release(page);
2308                 goto repeat;
2309         }
2310 
2311         /* Someone else locked and filled the page in a very small window */
2312         if (PageUptodate(page)) {
2313                 unlock_page(page);
2314                 goto out;
2315         }
2316 
2317         /*
2318          * A previous I/O error may have been due to temporary
2319          * failures.
2320          * Clear page error before actual read, PG_error will be
2321          * set again if read page fails.
2322          */
2323         ClearPageError(page);
2324         goto filler;
2325 
2326 out:
2327         mark_page_accessed(page);
2328         return page;
2329 }
2330 
2331 /**
2332  * read_cache_page - read into page cache, fill it if needed
2333  * @mapping:    the page's address_space
2334  * @index:      the page index
2335  * @filler:     function to perform the read
2336  * @data:       first arg to filler(data, page) function, often left as NULL
2337  *
2338  * Read into the page cache. If a page already exists, and PageUptodate() is
2339  * not set, try to fill the page and wait for it to become unlocked.
2340  *
2341  * If the page does not get brought uptodate, return -EIO.
2342  */
2343 struct page *read_cache_page(struct address_space *mapping,
2344                                 pgoff_t index,
2345                                 int (*filler)(void *, struct page *),
2346                                 void *data)
2347 {
2348         return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2349 }
2350 EXPORT_SYMBOL(read_cache_page);
2351 
2352 /**
2353  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2354  * @mapping:    the page's address_space
2355  * @index:      the page index
2356  * @gfp:        the page allocator flags to use if allocating
2357  *
2358  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2359  * any new page allocations done using the specified allocation flags.
2360  *
2361  * If the page does not get brought uptodate, return -EIO.
2362  */
2363 struct page *read_cache_page_gfp(struct address_space *mapping,
2364                                 pgoff_t index,
2365                                 gfp_t gfp)
2366 {
2367         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2368 
2369         return do_read_cache_page(mapping, index, filler, NULL, gfp);
2370 }
2371 EXPORT_SYMBOL(read_cache_page_gfp);
2372 
2373 /*
2374  * Performs necessary checks before doing a write
2375  *
2376  * Can adjust writing position or amount of bytes to write.
2377  * Returns appropriate error code that caller should return or
2378  * zero in case that write should be allowed.
2379  */
2380 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2381 {
2382         struct file *file = iocb->ki_filp;
2383         struct inode *inode = file->f_mapping->host;
2384         unsigned long limit = rlimit(RLIMIT_FSIZE);
2385         loff_t pos;
2386 
2387         if (!iov_iter_count(from))
2388                 return 0;
2389 
2390         /* FIXME: this is for backwards compatibility with 2.4 */
2391         if (iocb->ki_flags & IOCB_APPEND)
2392                 iocb->ki_pos = i_size_read(inode);
2393 
2394         pos = iocb->ki_pos;
2395 
2396         if (limit != RLIM_INFINITY) {
2397                 if (iocb->ki_pos >= limit) {
2398                         send_sig(SIGXFSZ, current, 0);
2399                         return -EFBIG;
2400                 }
2401                 iov_iter_truncate(from, limit - (unsigned long)pos);
2402         }
2403 
2404         /*
2405          * LFS rule
2406          */
2407         if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2408                                 !(file->f_flags & O_LARGEFILE))) {
2409                 if (pos >= MAX_NON_LFS)
2410                         return -EFBIG;
2411                 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2412         }
2413 
2414         /*
2415          * Are we about to exceed the fs block limit ?
2416          *
2417          * If we have written data it becomes a short write.  If we have
2418          * exceeded without writing data we send a signal and return EFBIG.
2419          * Linus frestrict idea will clean these up nicely..
2420          */
2421         if (unlikely(pos >= inode->i_sb->s_maxbytes))
2422                 return -EFBIG;
2423 
2424         iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2425         return iov_iter_count(from);
2426 }
2427 EXPORT_SYMBOL(generic_write_checks);
2428 
2429 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2430                                 loff_t pos, unsigned len, unsigned flags,
2431                                 struct page **pagep, void **fsdata)
2432 {
2433         const struct address_space_operations *aops = mapping->a_ops;
2434 
2435         return aops->write_begin(file, mapping, pos, len, flags,
2436                                                         pagep, fsdata);
2437 }
2438 EXPORT_SYMBOL(pagecache_write_begin);
2439 
2440 int pagecache_write_end(struct file *file, struct address_space *mapping,
2441                                 loff_t pos, unsigned len, unsigned copied,
2442                                 struct page *page, void *fsdata)
2443 {
2444         const struct address_space_operations *aops = mapping->a_ops;
2445 
2446         return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2447 }
2448 EXPORT_SYMBOL(pagecache_write_end);
2449 
2450 ssize_t
2451 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
2452 {
2453         struct file     *file = iocb->ki_filp;
2454         struct address_space *mapping = file->f_mapping;
2455         struct inode    *inode = mapping->host;
2456         ssize_t         written;
2457         size_t          write_len;
2458         pgoff_t         end;
2459         struct iov_iter data;
2460 
2461         write_len = iov_iter_count(from);
2462         end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2463 
2464         written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2465         if (written)
2466                 goto out;
2467 
2468         /*
2469          * After a write we want buffered reads to be sure to go to disk to get
2470          * the new data.  We invalidate clean cached page from the region we're
2471          * about to write.  We do this *before* the write so that we can return
2472          * without clobbering -EIOCBQUEUED from ->direct_IO().
2473          */
2474         if (mapping->nrpages) {
2475                 written = invalidate_inode_pages2_range(mapping,
2476                                         pos >> PAGE_CACHE_SHIFT, end);
2477                 /*
2478                  * If a page can not be invalidated, return 0 to fall back
2479                  * to buffered write.
2480                  */
2481                 if (written) {
2482                         if (written == -EBUSY)
2483                                 return 0;
2484                         goto out;
2485                 }
2486         }
2487 
2488         data = *from;
2489         written = mapping->a_ops->direct_IO(iocb, &data, pos);
2490 
2491         /*
2492          * Finally, try again to invalidate clean pages which might have been
2493          * cached by non-direct readahead, or faulted in by get_user_pages()
2494          * if the source of the write was an mmap'ed region of the file
2495          * we're writing.  Either one is a pretty crazy thing to do,
2496          * so we don't support it 100%.  If this invalidation
2497          * fails, tough, the write still worked...
2498          */
2499         if (mapping->nrpages) {
2500                 invalidate_inode_pages2_range(mapping,
2501                                               pos >> PAGE_CACHE_SHIFT, end);
2502         }
2503 
2504         if (written > 0) {
2505                 pos += written;
2506                 iov_iter_advance(from, written);
2507                 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2508                         i_size_write(inode, pos);
2509                         mark_inode_dirty(inode);
2510                 }
2511                 iocb->ki_pos = pos;
2512         }
2513 out:
2514         return written;
2515 }
2516 EXPORT_SYMBOL(generic_file_direct_write);
2517 
2518 /*
2519  * Find or create a page at the given pagecache position. Return the locked
2520  * page. This function is specifically for buffered writes.
2521  */
2522 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2523                                         pgoff_t index, unsigned flags)
2524 {
2525         struct page *page;
2526         int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
2527 
2528         if (flags & AOP_FLAG_NOFS)
2529                 fgp_flags |= FGP_NOFS;
2530 
2531         page = pagecache_get_page(mapping, index, fgp_flags,
2532                         mapping_gfp_mask(mapping));
2533         if (page)
2534                 wait_for_stable_page(page);
2535 
2536         return page;
2537 }
2538 EXPORT_SYMBOL(grab_cache_page_write_begin);
2539 
2540 ssize_t generic_perform_write(struct file *file,
2541                                 struct iov_iter *i, loff_t pos)
2542 {
2543         struct address_space *mapping = file->f_mapping;
2544         const struct address_space_operations *a_ops = mapping->a_ops;
2545         long status = 0;
2546         ssize_t written = 0;
2547         unsigned int flags = 0;
2548 
2549         /*
2550          * Copies from kernel address space cannot fail (NFSD is a big user).
2551          */
2552         if (!iter_is_iovec(i))
2553                 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2554 
2555         do {
2556                 struct page *page;
2557                 unsigned long offset;   /* Offset into pagecache page */
2558                 unsigned long bytes;    /* Bytes to write to page */
2559                 size_t copied;          /* Bytes copied from user */
2560                 void *fsdata;
2561 
2562                 offset = (pos & (PAGE_CACHE_SIZE - 1));
2563                 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2564                                                 iov_iter_count(i));
2565 
2566 again:
2567                 /*
2568                  * Bring in the user page that we will copy from _first_.
2569                  * Otherwise there's a nasty deadlock on copying from the
2570                  * same page as we're writing to, without it being marked
2571                  * up-to-date.
2572                  *
2573                  * Not only is this an optimisation, but it is also required
2574                  * to check that the address is actually valid, when atomic
2575                  * usercopies are used, below.
2576                  */
2577                 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2578                         status = -EFAULT;
2579                         break;
2580                 }
2581 
2582                 if (fatal_signal_pending(current)) {
2583                         status = -EINTR;
2584                         break;
2585                 }
2586 
2587                 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2588                                                 &page, &fsdata);
2589                 if (unlikely(status < 0))
2590                         break;
2591 
2592                 if (mapping_writably_mapped(mapping))
2593                         flush_dcache_page(page);
2594 
2595                 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2596                 flush_dcache_page(page);
2597 
2598                 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2599                                                 page, fsdata);
2600                 if (unlikely(status < 0))
2601                         break;
2602                 copied = status;
2603 
2604                 cond_resched();
2605 
2606                 iov_iter_advance(i, copied);
2607                 if (unlikely(copied == 0)) {
2608                         /*
2609                          * If we were unable to copy any data at all, we must
2610                          * fall back to a single segment length write.
2611                          *
2612                          * If we didn't fallback here, we could livelock
2613                          * because not all segments in the iov can be copied at
2614                          * once without a pagefault.
2615                          */
2616                         bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2617                                                 iov_iter_single_seg_count(i));
2618                         goto again;
2619                 }
2620                 pos += copied;
2621                 written += copied;
2622 
2623                 balance_dirty_pages_ratelimited(mapping);
2624         } while (iov_iter_count(i));
2625 
2626         return written ? written : status;
2627 }
2628 EXPORT_SYMBOL(generic_perform_write);
2629 
2630 /**
2631  * __generic_file_write_iter - write data to a file
2632  * @iocb:       IO state structure (file, offset, etc.)
2633  * @from:       iov_iter with data to write
2634  *
2635  * This function does all the work needed for actually writing data to a
2636  * file. It does all basic checks, removes SUID from the file, updates
2637  * modification times and calls proper subroutines depending on whether we
2638  * do direct IO or a standard buffered write.
2639  *
2640  * It expects i_mutex to be grabbed unless we work on a block device or similar
2641  * object which does not need locking at all.
2642  *
2643  * This function does *not* take care of syncing data in case of O_SYNC write.
2644  * A caller has to handle it. This is mainly due to the fact that we want to
2645  * avoid syncing under i_mutex.
2646  */
2647 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2648 {
2649         struct file *file = iocb->ki_filp;
2650         struct address_space * mapping = file->f_mapping;
2651         struct inode    *inode = mapping->host;
2652         ssize_t         written = 0;
2653         ssize_t         err;
2654         ssize_t         status;
2655 
2656         /* We can write back this queue in page reclaim */
2657         current->backing_dev_info = inode_to_bdi(inode);
2658         err = file_remove_privs(file);
2659         if (err)
2660                 goto out;
2661 
2662         err = file_update_time(file);
2663         if (err)
2664                 goto out;
2665 
2666         if (iocb->ki_flags & IOCB_DIRECT) {
2667                 loff_t pos, endbyte;
2668 
2669                 written = generic_file_direct_write(iocb, from, iocb->ki_pos);
2670                 /*
2671                  * If the write stopped short of completing, fall back to
2672                  * buffered writes.  Some filesystems do this for writes to
2673                  * holes, for example.  For DAX files, a buffered write will
2674                  * not succeed (even if it did, DAX does not handle dirty
2675                  * page-cache pages correctly).
2676                  */
2677                 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2678                         goto out;
2679 
2680                 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2681                 /*
2682                  * If generic_perform_write() returned a synchronous error
2683                  * then we want to return the number of bytes which were
2684                  * direct-written, or the error code if that was zero.  Note
2685                  * that this differs from normal direct-io semantics, which
2686                  * will return -EFOO even if some bytes were written.
2687                  */
2688                 if (unlikely(status < 0)) {
2689                         err = status;
2690                         goto out;
2691                 }
2692                 /*
2693                  * We need to ensure that the page cache pages are written to
2694                  * disk and invalidated to preserve the expected O_DIRECT
2695                  * semantics.
2696                  */
2697                 endbyte = pos + status - 1;
2698                 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2699                 if (err == 0) {
2700                         iocb->ki_pos = endbyte + 1;
2701                         written += status;
2702                         invalidate_mapping_pages(mapping,
2703                                                  pos >> PAGE_CACHE_SHIFT,
2704                                                  endbyte >> PAGE_CACHE_SHIFT);
2705                 } else {
2706                         /*
2707                          * We don't know how much we wrote, so just return
2708                          * the number of bytes which were direct-written
2709                          */
2710                 }
2711         } else {
2712                 written = generic_perform_write(file, from, iocb->ki_pos);
2713                 if (likely(written > 0))
2714                         iocb->ki_pos += written;
2715         }
2716 out:
2717         current->backing_dev_info = NULL;
2718         return written ? written : err;
2719 }
2720 EXPORT_SYMBOL(__generic_file_write_iter);
2721 
2722 /**
2723  * generic_file_write_iter - write data to a file
2724  * @iocb:       IO state structure
2725  * @from:       iov_iter with data to write
2726  *
2727  * This is a wrapper around __generic_file_write_iter() to be used by most
2728  * filesystems. It takes care of syncing the file in case of O_SYNC file
2729  * and acquires i_mutex as needed.
2730  */
2731 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2732 {
2733         struct file *file = iocb->ki_filp;
2734         struct inode *inode = file->f_mapping->host;
2735         ssize_t ret;
2736 
2737         mutex_lock(&inode->i_mutex);
2738         ret = generic_write_checks(iocb, from);
2739         if (ret > 0)
2740                 ret = __generic_file_write_iter(iocb, from);
2741         mutex_unlock(&inode->i_mutex);
2742 
2743         if (ret > 0) {
2744                 ssize_t err;
2745 
2746                 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2747                 if (err < 0)
2748                         ret = err;
2749         }
2750         return ret;
2751 }
2752 EXPORT_SYMBOL(generic_file_write_iter);
2753 
2754 /**
2755  * try_to_release_page() - release old fs-specific metadata on a page
2756  *
2757  * @page: the page which the kernel is trying to free
2758  * @gfp_mask: memory allocation flags (and I/O mode)
2759  *
2760  * The address_space is to try to release any data against the page
2761  * (presumably at page->private).  If the release was successful, return `1'.
2762  * Otherwise return zero.
2763  *
2764  * This may also be called if PG_fscache is set on a page, indicating that the
2765  * page is known to the local caching routines.
2766  *
2767  * The @gfp_mask argument specifies whether I/O may be performed to release
2768  * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
2769  *
2770  */
2771 int try_to_release_page(struct page *page, gfp_t gfp_mask)
2772 {
2773         struct address_space * const mapping = page->mapping;
2774 
2775         BUG_ON(!PageLocked(page));
2776         if (PageWriteback(page))
2777                 return 0;
2778 
2779         if (mapping && mapping->a_ops->releasepage)
2780                 return mapping->a_ops->releasepage(page, gfp_mask);
2781         return try_to_free_buffers(page);
2782 }
2783 
2784 EXPORT_SYMBOL(try_to_release_page);
2785 

~ [ 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