Version:  2.0.40 2.2.26 2.4.37 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10

Linux/mm/filemap.c

  1 /*
  2  *      linux/mm/filemap.c
  3  *
  4  * Copyright (C) 1994-1999  Linus Torvalds
  5  */
  6 
  7 /*
  8  * This file handles the generic file mmap semantics used by
  9  * most "normal" filesystems (but you don't /have/ to use this:
 10  * the NFS filesystem used to do this differently, for example)
 11  */
 12 #include <linux/export.h>
 13 #include <linux/compiler.h>
 14 #include <linux/dax.h>
 15 #include <linux/fs.h>
 16 #include <linux/uaccess.h>
 17 #include <linux/capability.h>
 18 #include <linux/kernel_stat.h>
 19 #include <linux/gfp.h>
 20 #include <linux/mm.h>
 21 #include <linux/swap.h>
 22 #include <linux/mman.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/file.h>
 25 #include <linux/uio.h>
 26 #include <linux/hash.h>
 27 #include <linux/writeback.h>
 28 #include <linux/backing-dev.h>
 29 #include <linux/pagevec.h>
 30 #include <linux/blkdev.h>
 31 #include <linux/security.h>
 32 #include <linux/cpuset.h>
 33 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
 34 #include <linux/hugetlb.h>
 35 #include <linux/memcontrol.h>
 36 #include <linux/cleancache.h>
 37 #include <linux/rmap.h>
 38 #include "internal.h"
 39 
 40 #define CREATE_TRACE_POINTS
 41 #include <trace/events/filemap.h>
 42 
 43 /*
 44  * FIXME: remove all knowledge of the buffer layer from the core VM
 45  */
 46 #include <linux/buffer_head.h> /* for try_to_free_buffers */
 47 
 48 #include <asm/mman.h>
 49 
 50 /*
 51  * Shared mappings implemented 30.11.1994. It's not fully working yet,
 52  * though.
 53  *
 54  * Shared mappings now work. 15.8.1995  Bruno.
 55  *
 56  * finished 'unifying' the page and buffer cache and SMP-threaded the
 57  * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
 58  *
 59  * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
 60  */
 61 
 62 /*
 63  * Lock ordering:
 64  *
 65  *  ->i_mmap_rwsem              (truncate_pagecache)
 66  *    ->private_lock            (__free_pte->__set_page_dirty_buffers)
 67  *      ->swap_lock             (exclusive_swap_page, others)
 68  *        ->mapping->tree_lock
 69  *
 70  *  ->i_mutex
 71  *    ->i_mmap_rwsem            (truncate->unmap_mapping_range)
 72  *
 73  *  ->mmap_sem
 74  *    ->i_mmap_rwsem
 75  *      ->page_table_lock or pte_lock   (various, mainly in memory.c)
 76  *        ->mapping->tree_lock  (arch-dependent flush_dcache_mmap_lock)
 77  *
 78  *  ->mmap_sem
 79  *    ->lock_page               (access_process_vm)
 80  *
 81  *  ->i_mutex                   (generic_perform_write)
 82  *    ->mmap_sem                (fault_in_pages_readable->do_page_fault)
 83  *
 84  *  bdi->wb.list_lock
 85  *    sb_lock                   (fs/fs-writeback.c)
 86  *    ->mapping->tree_lock      (__sync_single_inode)
 87  *
 88  *  ->i_mmap_rwsem
 89  *    ->anon_vma.lock           (vma_adjust)
 90  *
 91  *  ->anon_vma.lock
 92  *    ->page_table_lock or pte_lock     (anon_vma_prepare and various)
 93  *
 94  *  ->page_table_lock or pte_lock
 95  *    ->swap_lock               (try_to_unmap_one)
 96  *    ->private_lock            (try_to_unmap_one)
 97  *    ->tree_lock               (try_to_unmap_one)
 98  *    ->zone_lru_lock(zone)     (follow_page->mark_page_accessed)
 99  *    ->zone_lru_lock(zone)     (check_pte_range->isolate_lru_page)
100  *    ->private_lock            (page_remove_rmap->set_page_dirty)
101  *    ->tree_lock               (page_remove_rmap->set_page_dirty)
102  *    bdi.wb->list_lock         (page_remove_rmap->set_page_dirty)
103  *    ->inode->i_lock           (page_remove_rmap->set_page_dirty)
104  *    ->memcg->move_lock        (page_remove_rmap->lock_page_memcg)
105  *    bdi.wb->list_lock         (zap_pte_range->set_page_dirty)
106  *    ->inode->i_lock           (zap_pte_range->set_page_dirty)
107  *    ->private_lock            (zap_pte_range->__set_page_dirty_buffers)
108  *
109  * ->i_mmap_rwsem
110  *   ->tasklist_lock            (memory_failure, collect_procs_ao)
111  */
112 
113 static int page_cache_tree_insert(struct address_space *mapping,
114                                   struct page *page, void **shadowp)
115 {
116         struct radix_tree_node *node;
117         void **slot;
118         int error;
119 
120         error = __radix_tree_create(&mapping->page_tree, page->index, 0,
121                                     &node, &slot);
122         if (error)
123                 return error;
124         if (*slot) {
125                 void *p;
126 
127                 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
128                 if (!radix_tree_exceptional_entry(p))
129                         return -EEXIST;
130 
131                 mapping->nrexceptional--;
132                 if (!dax_mapping(mapping)) {
133                         if (shadowp)
134                                 *shadowp = p;
135                 } else {
136                         /* DAX can replace empty locked entry with a hole */
137                         WARN_ON_ONCE(p !=
138                                 dax_radix_locked_entry(0, RADIX_DAX_EMPTY));
139                         /* Wakeup waiters for exceptional entry lock */
140                         dax_wake_mapping_entry_waiter(mapping, page->index, p,
141                                                       true);
142                 }
143         }
144         __radix_tree_replace(&mapping->page_tree, node, slot, page,
145                              workingset_update_node, mapping);
146         mapping->nrpages++;
147         return 0;
148 }
149 
150 static void page_cache_tree_delete(struct address_space *mapping,
151                                    struct page *page, void *shadow)
152 {
153         int i, nr;
154 
155         /* hugetlb pages are represented by one entry in the radix tree */
156         nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
157 
158         VM_BUG_ON_PAGE(!PageLocked(page), page);
159         VM_BUG_ON_PAGE(PageTail(page), page);
160         VM_BUG_ON_PAGE(nr != 1 && shadow, page);
161 
162         for (i = 0; i < nr; i++) {
163                 struct radix_tree_node *node;
164                 void **slot;
165 
166                 __radix_tree_lookup(&mapping->page_tree, page->index + i,
167                                     &node, &slot);
168 
169                 VM_BUG_ON_PAGE(!node && nr != 1, page);
170 
171                 radix_tree_clear_tags(&mapping->page_tree, node, slot);
172                 __radix_tree_replace(&mapping->page_tree, node, slot, shadow,
173                                      workingset_update_node, mapping);
174         }
175 
176         if (shadow) {
177                 mapping->nrexceptional += nr;
178                 /*
179                  * Make sure the nrexceptional update is committed before
180                  * the nrpages update so that final truncate racing
181                  * with reclaim does not see both counters 0 at the
182                  * same time and miss a shadow entry.
183                  */
184                 smp_wmb();
185         }
186         mapping->nrpages -= nr;
187 }
188 
189 /*
190  * Delete a page from the page cache and free it. Caller has to make
191  * sure the page is locked and that nobody else uses it - or that usage
192  * is safe.  The caller must hold the mapping's tree_lock.
193  */
194 void __delete_from_page_cache(struct page *page, void *shadow)
195 {
196         struct address_space *mapping = page->mapping;
197         int nr = hpage_nr_pages(page);
198 
199         trace_mm_filemap_delete_from_page_cache(page);
200         /*
201          * if we're uptodate, flush out into the cleancache, otherwise
202          * invalidate any existing cleancache entries.  We can't leave
203          * stale data around in the cleancache once our page is gone
204          */
205         if (PageUptodate(page) && PageMappedToDisk(page))
206                 cleancache_put_page(page);
207         else
208                 cleancache_invalidate_page(mapping, page);
209 
210         VM_BUG_ON_PAGE(PageTail(page), page);
211         VM_BUG_ON_PAGE(page_mapped(page), page);
212         if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
213                 int mapcount;
214 
215                 pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
216                          current->comm, page_to_pfn(page));
217                 dump_page(page, "still mapped when deleted");
218                 dump_stack();
219                 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
220 
221                 mapcount = page_mapcount(page);
222                 if (mapping_exiting(mapping) &&
223                     page_count(page) >= mapcount + 2) {
224                         /*
225                          * All vmas have already been torn down, so it's
226                          * a good bet that actually the page is unmapped,
227                          * and we'd prefer not to leak it: if we're wrong,
228                          * some other bad page check should catch it later.
229                          */
230                         page_mapcount_reset(page);
231                         page_ref_sub(page, mapcount);
232                 }
233         }
234 
235         page_cache_tree_delete(mapping, page, shadow);
236 
237         page->mapping = NULL;
238         /* Leave page->index set: truncation lookup relies upon it */
239 
240         /* hugetlb pages do not participate in page cache accounting. */
241         if (!PageHuge(page))
242                 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
243         if (PageSwapBacked(page)) {
244                 __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
245                 if (PageTransHuge(page))
246                         __dec_node_page_state(page, NR_SHMEM_THPS);
247         } else {
248                 VM_BUG_ON_PAGE(PageTransHuge(page) && !PageHuge(page), page);
249         }
250 
251         /*
252          * At this point page must be either written or cleaned by truncate.
253          * Dirty page here signals a bug and loss of unwritten data.
254          *
255          * This fixes dirty accounting after removing the page entirely but
256          * leaves PageDirty set: it has no effect for truncated page and
257          * anyway will be cleared before returning page into buddy allocator.
258          */
259         if (WARN_ON_ONCE(PageDirty(page)))
260                 account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
261 }
262 
263 /**
264  * delete_from_page_cache - delete page from page cache
265  * @page: the page which the kernel is trying to remove from page cache
266  *
267  * This must be called only on pages that have been verified to be in the page
268  * cache and locked.  It will never put the page into the free list, the caller
269  * has a reference on the page.
270  */
271 void delete_from_page_cache(struct page *page)
272 {
273         struct address_space *mapping = page_mapping(page);
274         unsigned long flags;
275         void (*freepage)(struct page *);
276 
277         BUG_ON(!PageLocked(page));
278 
279         freepage = mapping->a_ops->freepage;
280 
281         spin_lock_irqsave(&mapping->tree_lock, flags);
282         __delete_from_page_cache(page, NULL);
283         spin_unlock_irqrestore(&mapping->tree_lock, flags);
284 
285         if (freepage)
286                 freepage(page);
287 
288         if (PageTransHuge(page) && !PageHuge(page)) {
289                 page_ref_sub(page, HPAGE_PMD_NR);
290                 VM_BUG_ON_PAGE(page_count(page) <= 0, page);
291         } else {
292                 put_page(page);
293         }
294 }
295 EXPORT_SYMBOL(delete_from_page_cache);
296 
297 int filemap_check_errors(struct address_space *mapping)
298 {
299         int ret = 0;
300         /* Check for outstanding write errors */
301         if (test_bit(AS_ENOSPC, &mapping->flags) &&
302             test_and_clear_bit(AS_ENOSPC, &mapping->flags))
303                 ret = -ENOSPC;
304         if (test_bit(AS_EIO, &mapping->flags) &&
305             test_and_clear_bit(AS_EIO, &mapping->flags))
306                 ret = -EIO;
307         return ret;
308 }
309 EXPORT_SYMBOL(filemap_check_errors);
310 
311 /**
312  * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
313  * @mapping:    address space structure to write
314  * @start:      offset in bytes where the range starts
315  * @end:        offset in bytes where the range ends (inclusive)
316  * @sync_mode:  enable synchronous operation
317  *
318  * Start writeback against all of a mapping's dirty pages that lie
319  * within the byte offsets <start, end> inclusive.
320  *
321  * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
322  * opposed to a regular memory cleansing writeback.  The difference between
323  * these two operations is that if a dirty page/buffer is encountered, it must
324  * be waited upon, and not just skipped over.
325  */
326 int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
327                                 loff_t end, int sync_mode)
328 {
329         int ret;
330         struct writeback_control wbc = {
331                 .sync_mode = sync_mode,
332                 .nr_to_write = LONG_MAX,
333                 .range_start = start,
334                 .range_end = end,
335         };
336 
337         if (!mapping_cap_writeback_dirty(mapping))
338                 return 0;
339 
340         wbc_attach_fdatawrite_inode(&wbc, mapping->host);
341         ret = do_writepages(mapping, &wbc);
342         wbc_detach_inode(&wbc);
343         return ret;
344 }
345 
346 static inline int __filemap_fdatawrite(struct address_space *mapping,
347         int sync_mode)
348 {
349         return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
350 }
351 
352 int filemap_fdatawrite(struct address_space *mapping)
353 {
354         return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
355 }
356 EXPORT_SYMBOL(filemap_fdatawrite);
357 
358 int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
359                                 loff_t end)
360 {
361         return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
362 }
363 EXPORT_SYMBOL(filemap_fdatawrite_range);
364 
365 /**
366  * filemap_flush - mostly a non-blocking flush
367  * @mapping:    target address_space
368  *
369  * This is a mostly non-blocking flush.  Not suitable for data-integrity
370  * purposes - I/O may not be started against all dirty pages.
371  */
372 int filemap_flush(struct address_space *mapping)
373 {
374         return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
375 }
376 EXPORT_SYMBOL(filemap_flush);
377 
378 static int __filemap_fdatawait_range(struct address_space *mapping,
379                                      loff_t start_byte, loff_t end_byte)
380 {
381         pgoff_t index = start_byte >> PAGE_SHIFT;
382         pgoff_t end = end_byte >> PAGE_SHIFT;
383         struct pagevec pvec;
384         int nr_pages;
385         int ret = 0;
386 
387         if (end_byte < start_byte)
388                 goto out;
389 
390         pagevec_init(&pvec, 0);
391         while ((index <= end) &&
392                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
393                         PAGECACHE_TAG_WRITEBACK,
394                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
395                 unsigned i;
396 
397                 for (i = 0; i < nr_pages; i++) {
398                         struct page *page = pvec.pages[i];
399 
400                         /* until radix tree lookup accepts end_index */
401                         if (page->index > end)
402                                 continue;
403 
404                         wait_on_page_writeback(page);
405                         if (TestClearPageError(page))
406                                 ret = -EIO;
407                 }
408                 pagevec_release(&pvec);
409                 cond_resched();
410         }
411 out:
412         return ret;
413 }
414 
415 /**
416  * filemap_fdatawait_range - wait for writeback to complete
417  * @mapping:            address space structure to wait for
418  * @start_byte:         offset in bytes where the range starts
419  * @end_byte:           offset in bytes where the range ends (inclusive)
420  *
421  * Walk the list of under-writeback pages of the given address space
422  * in the given range and wait for all of them.  Check error status of
423  * the address space and return it.
424  *
425  * Since the error status of the address space is cleared by this function,
426  * callers are responsible for checking the return value and handling and/or
427  * reporting the error.
428  */
429 int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
430                             loff_t end_byte)
431 {
432         int ret, ret2;
433 
434         ret = __filemap_fdatawait_range(mapping, start_byte, end_byte);
435         ret2 = filemap_check_errors(mapping);
436         if (!ret)
437                 ret = ret2;
438 
439         return ret;
440 }
441 EXPORT_SYMBOL(filemap_fdatawait_range);
442 
443 /**
444  * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
445  * @mapping: address space structure to wait for
446  *
447  * Walk the list of under-writeback pages of the given address space
448  * and wait for all of them.  Unlike filemap_fdatawait(), this function
449  * does not clear error status of the address space.
450  *
451  * Use this function if callers don't handle errors themselves.  Expected
452  * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
453  * fsfreeze(8)
454  */
455 void filemap_fdatawait_keep_errors(struct address_space *mapping)
456 {
457         loff_t i_size = i_size_read(mapping->host);
458 
459         if (i_size == 0)
460                 return;
461 
462         __filemap_fdatawait_range(mapping, 0, i_size - 1);
463 }
464 
465 /**
466  * filemap_fdatawait - wait for all under-writeback pages to complete
467  * @mapping: address space structure to wait for
468  *
469  * Walk the list of under-writeback pages of the given address space
470  * and wait for all of them.  Check error status of the address space
471  * and return it.
472  *
473  * Since the error status of the address space is cleared by this function,
474  * callers are responsible for checking the return value and handling and/or
475  * reporting the error.
476  */
477 int filemap_fdatawait(struct address_space *mapping)
478 {
479         loff_t i_size = i_size_read(mapping->host);
480 
481         if (i_size == 0)
482                 return 0;
483 
484         return filemap_fdatawait_range(mapping, 0, i_size - 1);
485 }
486 EXPORT_SYMBOL(filemap_fdatawait);
487 
488 int filemap_write_and_wait(struct address_space *mapping)
489 {
490         int err = 0;
491 
492         if ((!dax_mapping(mapping) && mapping->nrpages) ||
493             (dax_mapping(mapping) && mapping->nrexceptional)) {
494                 err = filemap_fdatawrite(mapping);
495                 /*
496                  * Even if the above returned error, the pages may be
497                  * written partially (e.g. -ENOSPC), so we wait for it.
498                  * But the -EIO is special case, it may indicate the worst
499                  * thing (e.g. bug) happened, so we avoid waiting for it.
500                  */
501                 if (err != -EIO) {
502                         int err2 = filemap_fdatawait(mapping);
503                         if (!err)
504                                 err = err2;
505                 }
506         } else {
507                 err = filemap_check_errors(mapping);
508         }
509         return err;
510 }
511 EXPORT_SYMBOL(filemap_write_and_wait);
512 
513 /**
514  * filemap_write_and_wait_range - write out & wait on a file range
515  * @mapping:    the address_space for the pages
516  * @lstart:     offset in bytes where the range starts
517  * @lend:       offset in bytes where the range ends (inclusive)
518  *
519  * Write out and wait upon file offsets lstart->lend, inclusive.
520  *
521  * Note that `lend' is inclusive (describes the last byte to be written) so
522  * that this function can be used to write to the very end-of-file (end = -1).
523  */
524 int filemap_write_and_wait_range(struct address_space *mapping,
525                                  loff_t lstart, loff_t lend)
526 {
527         int err = 0;
528 
529         if ((!dax_mapping(mapping) && mapping->nrpages) ||
530             (dax_mapping(mapping) && mapping->nrexceptional)) {
531                 err = __filemap_fdatawrite_range(mapping, lstart, lend,
532                                                  WB_SYNC_ALL);
533                 /* See comment of filemap_write_and_wait() */
534                 if (err != -EIO) {
535                         int err2 = filemap_fdatawait_range(mapping,
536                                                 lstart, lend);
537                         if (!err)
538                                 err = err2;
539                 }
540         } else {
541                 err = filemap_check_errors(mapping);
542         }
543         return err;
544 }
545 EXPORT_SYMBOL(filemap_write_and_wait_range);
546 
547 /**
548  * replace_page_cache_page - replace a pagecache page with a new one
549  * @old:        page to be replaced
550  * @new:        page to replace with
551  * @gfp_mask:   allocation mode
552  *
553  * This function replaces a page in the pagecache with a new one.  On
554  * success it acquires the pagecache reference for the new page and
555  * drops it for the old page.  Both the old and new pages must be
556  * locked.  This function does not add the new page to the LRU, the
557  * caller must do that.
558  *
559  * The remove + add is atomic.  The only way this function can fail is
560  * memory allocation failure.
561  */
562 int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
563 {
564         int error;
565 
566         VM_BUG_ON_PAGE(!PageLocked(old), old);
567         VM_BUG_ON_PAGE(!PageLocked(new), new);
568         VM_BUG_ON_PAGE(new->mapping, new);
569 
570         error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
571         if (!error) {
572                 struct address_space *mapping = old->mapping;
573                 void (*freepage)(struct page *);
574                 unsigned long flags;
575 
576                 pgoff_t offset = old->index;
577                 freepage = mapping->a_ops->freepage;
578 
579                 get_page(new);
580                 new->mapping = mapping;
581                 new->index = offset;
582 
583                 spin_lock_irqsave(&mapping->tree_lock, flags);
584                 __delete_from_page_cache(old, NULL);
585                 error = page_cache_tree_insert(mapping, new, NULL);
586                 BUG_ON(error);
587 
588                 /*
589                  * hugetlb pages do not participate in page cache accounting.
590                  */
591                 if (!PageHuge(new))
592                         __inc_node_page_state(new, NR_FILE_PAGES);
593                 if (PageSwapBacked(new))
594                         __inc_node_page_state(new, NR_SHMEM);
595                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
596                 mem_cgroup_migrate(old, new);
597                 radix_tree_preload_end();
598                 if (freepage)
599                         freepage(old);
600                 put_page(old);
601         }
602 
603         return error;
604 }
605 EXPORT_SYMBOL_GPL(replace_page_cache_page);
606 
607 static int __add_to_page_cache_locked(struct page *page,
608                                       struct address_space *mapping,
609                                       pgoff_t offset, gfp_t gfp_mask,
610                                       void **shadowp)
611 {
612         int huge = PageHuge(page);
613         struct mem_cgroup *memcg;
614         int error;
615 
616         VM_BUG_ON_PAGE(!PageLocked(page), page);
617         VM_BUG_ON_PAGE(PageSwapBacked(page), page);
618 
619         if (!huge) {
620                 error = mem_cgroup_try_charge(page, current->mm,
621                                               gfp_mask, &memcg, false);
622                 if (error)
623                         return error;
624         }
625 
626         error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
627         if (error) {
628                 if (!huge)
629                         mem_cgroup_cancel_charge(page, memcg, false);
630                 return error;
631         }
632 
633         get_page(page);
634         page->mapping = mapping;
635         page->index = offset;
636 
637         spin_lock_irq(&mapping->tree_lock);
638         error = page_cache_tree_insert(mapping, page, shadowp);
639         radix_tree_preload_end();
640         if (unlikely(error))
641                 goto err_insert;
642 
643         /* hugetlb pages do not participate in page cache accounting. */
644         if (!huge)
645                 __inc_node_page_state(page, NR_FILE_PAGES);
646         spin_unlock_irq(&mapping->tree_lock);
647         if (!huge)
648                 mem_cgroup_commit_charge(page, memcg, false, false);
649         trace_mm_filemap_add_to_page_cache(page);
650         return 0;
651 err_insert:
652         page->mapping = NULL;
653         /* Leave page->index set: truncation relies upon it */
654         spin_unlock_irq(&mapping->tree_lock);
655         if (!huge)
656                 mem_cgroup_cancel_charge(page, memcg, false);
657         put_page(page);
658         return error;
659 }
660 
661 /**
662  * add_to_page_cache_locked - add a locked page to the pagecache
663  * @page:       page to add
664  * @mapping:    the page's address_space
665  * @offset:     page index
666  * @gfp_mask:   page allocation mode
667  *
668  * This function is used to add a page to the pagecache. It must be locked.
669  * This function does not add the page to the LRU.  The caller must do that.
670  */
671 int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
672                 pgoff_t offset, gfp_t gfp_mask)
673 {
674         return __add_to_page_cache_locked(page, mapping, offset,
675                                           gfp_mask, NULL);
676 }
677 EXPORT_SYMBOL(add_to_page_cache_locked);
678 
679 int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
680                                 pgoff_t offset, gfp_t gfp_mask)
681 {
682         void *shadow = NULL;
683         int ret;
684 
685         __SetPageLocked(page);
686         ret = __add_to_page_cache_locked(page, mapping, offset,
687                                          gfp_mask, &shadow);
688         if (unlikely(ret))
689                 __ClearPageLocked(page);
690         else {
691                 /*
692                  * The page might have been evicted from cache only
693                  * recently, in which case it should be activated like
694                  * any other repeatedly accessed page.
695                  * The exception is pages getting rewritten; evicting other
696                  * data from the working set, only to cache data that will
697                  * get overwritten with something else, is a waste of memory.
698                  */
699                 if (!(gfp_mask & __GFP_WRITE) &&
700                     shadow && workingset_refault(shadow)) {
701                         SetPageActive(page);
702                         workingset_activation(page);
703                 } else
704                         ClearPageActive(page);
705                 lru_cache_add(page);
706         }
707         return ret;
708 }
709 EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
710 
711 #ifdef CONFIG_NUMA
712 struct page *__page_cache_alloc(gfp_t gfp)
713 {
714         int n;
715         struct page *page;
716 
717         if (cpuset_do_page_mem_spread()) {
718                 unsigned int cpuset_mems_cookie;
719                 do {
720                         cpuset_mems_cookie = read_mems_allowed_begin();
721                         n = cpuset_mem_spread_node();
722                         page = __alloc_pages_node(n, gfp, 0);
723                 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
724 
725                 return page;
726         }
727         return alloc_pages(gfp, 0);
728 }
729 EXPORT_SYMBOL(__page_cache_alloc);
730 #endif
731 
732 /*
733  * In order to wait for pages to become available there must be
734  * waitqueues associated with pages. By using a hash table of
735  * waitqueues where the bucket discipline is to maintain all
736  * waiters on the same queue and wake all when any of the pages
737  * become available, and for the woken contexts to check to be
738  * sure the appropriate page became available, this saves space
739  * at a cost of "thundering herd" phenomena during rare hash
740  * collisions.
741  */
742 #define PAGE_WAIT_TABLE_BITS 8
743 #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
744 static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
745 
746 static wait_queue_head_t *page_waitqueue(struct page *page)
747 {
748         return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
749 }
750 
751 void __init pagecache_init(void)
752 {
753         int i;
754 
755         for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
756                 init_waitqueue_head(&page_wait_table[i]);
757 
758         page_writeback_init();
759 }
760 
761 struct wait_page_key {
762         struct page *page;
763         int bit_nr;
764         int page_match;
765 };
766 
767 struct wait_page_queue {
768         struct page *page;
769         int bit_nr;
770         wait_queue_t wait;
771 };
772 
773 static int wake_page_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
774 {
775         struct wait_page_key *key = arg;
776         struct wait_page_queue *wait_page
777                 = container_of(wait, struct wait_page_queue, wait);
778 
779         if (wait_page->page != key->page)
780                return 0;
781         key->page_match = 1;
782 
783         if (wait_page->bit_nr != key->bit_nr)
784                 return 0;
785         if (test_bit(key->bit_nr, &key->page->flags))
786                 return 0;
787 
788         return autoremove_wake_function(wait, mode, sync, key);
789 }
790 
791 void wake_up_page_bit(struct page *page, int bit_nr)
792 {
793         wait_queue_head_t *q = page_waitqueue(page);
794         struct wait_page_key key;
795         unsigned long flags;
796 
797         key.page = page;
798         key.bit_nr = bit_nr;
799         key.page_match = 0;
800 
801         spin_lock_irqsave(&q->lock, flags);
802         __wake_up_locked_key(q, TASK_NORMAL, &key);
803         /*
804          * It is possible for other pages to have collided on the waitqueue
805          * hash, so in that case check for a page match. That prevents a long-
806          * term waiter
807          *
808          * It is still possible to miss a case here, when we woke page waiters
809          * and removed them from the waitqueue, but there are still other
810          * page waiters.
811          */
812         if (!waitqueue_active(q) || !key.page_match) {
813                 ClearPageWaiters(page);
814                 /*
815                  * It's possible to miss clearing Waiters here, when we woke
816                  * our page waiters, but the hashed waitqueue has waiters for
817                  * other pages on it.
818                  *
819                  * That's okay, it's a rare case. The next waker will clear it.
820                  */
821         }
822         spin_unlock_irqrestore(&q->lock, flags);
823 }
824 EXPORT_SYMBOL(wake_up_page_bit);
825 
826 static inline int wait_on_page_bit_common(wait_queue_head_t *q,
827                 struct page *page, int bit_nr, int state, bool lock)
828 {
829         struct wait_page_queue wait_page;
830         wait_queue_t *wait = &wait_page.wait;
831         int ret = 0;
832 
833         init_wait(wait);
834         wait->func = wake_page_function;
835         wait_page.page = page;
836         wait_page.bit_nr = bit_nr;
837 
838         for (;;) {
839                 spin_lock_irq(&q->lock);
840 
841                 if (likely(list_empty(&wait->task_list))) {
842                         if (lock)
843                                 __add_wait_queue_tail_exclusive(q, wait);
844                         else
845                                 __add_wait_queue(q, wait);
846                         SetPageWaiters(page);
847                 }
848 
849                 set_current_state(state);
850 
851                 spin_unlock_irq(&q->lock);
852 
853                 if (likely(test_bit(bit_nr, &page->flags))) {
854                         io_schedule();
855                         if (unlikely(signal_pending_state(state, current))) {
856                                 ret = -EINTR;
857                                 break;
858                         }
859                 }
860 
861                 if (lock) {
862                         if (!test_and_set_bit_lock(bit_nr, &page->flags))
863                                 break;
864                 } else {
865                         if (!test_bit(bit_nr, &page->flags))
866                                 break;
867                 }
868         }
869 
870         finish_wait(q, wait);
871 
872         /*
873          * A signal could leave PageWaiters set. Clearing it here if
874          * !waitqueue_active would be possible (by open-coding finish_wait),
875          * but still fail to catch it in the case of wait hash collision. We
876          * already can fail to clear wait hash collision cases, so don't
877          * bother with signals either.
878          */
879 
880         return ret;
881 }
882 
883 void wait_on_page_bit(struct page *page, int bit_nr)
884 {
885         wait_queue_head_t *q = page_waitqueue(page);
886         wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
887 }
888 EXPORT_SYMBOL(wait_on_page_bit);
889 
890 int wait_on_page_bit_killable(struct page *page, int bit_nr)
891 {
892         wait_queue_head_t *q = page_waitqueue(page);
893         return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
894 }
895 
896 /**
897  * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
898  * @page: Page defining the wait queue of interest
899  * @waiter: Waiter to add to the queue
900  *
901  * Add an arbitrary @waiter to the wait queue for the nominated @page.
902  */
903 void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
904 {
905         wait_queue_head_t *q = page_waitqueue(page);
906         unsigned long flags;
907 
908         spin_lock_irqsave(&q->lock, flags);
909         __add_wait_queue(q, waiter);
910         SetPageWaiters(page);
911         spin_unlock_irqrestore(&q->lock, flags);
912 }
913 EXPORT_SYMBOL_GPL(add_page_wait_queue);
914 
915 #ifndef clear_bit_unlock_is_negative_byte
916 
917 /*
918  * PG_waiters is the high bit in the same byte as PG_lock.
919  *
920  * On x86 (and on many other architectures), we can clear PG_lock and
921  * test the sign bit at the same time. But if the architecture does
922  * not support that special operation, we just do this all by hand
923  * instead.
924  *
925  * The read of PG_waiters has to be after (or concurrently with) PG_locked
926  * being cleared, but a memory barrier should be unneccssary since it is
927  * in the same byte as PG_locked.
928  */
929 static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
930 {
931         clear_bit_unlock(nr, mem);
932         /* smp_mb__after_atomic(); */
933         return test_bit(PG_waiters, mem);
934 }
935 
936 #endif
937 
938 /**
939  * unlock_page - unlock a locked page
940  * @page: the page
941  *
942  * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
943  * Also wakes sleepers in wait_on_page_writeback() because the wakeup
944  * mechanism between PageLocked pages and PageWriteback pages is shared.
945  * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
946  *
947  * Note that this depends on PG_waiters being the sign bit in the byte
948  * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
949  * clear the PG_locked bit and test PG_waiters at the same time fairly
950  * portably (architectures that do LL/SC can test any bit, while x86 can
951  * test the sign bit).
952  */
953 void unlock_page(struct page *page)
954 {
955         BUILD_BUG_ON(PG_waiters != 7);
956         page = compound_head(page);
957         VM_BUG_ON_PAGE(!PageLocked(page), page);
958         if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
959                 wake_up_page_bit(page, PG_locked);
960 }
961 EXPORT_SYMBOL(unlock_page);
962 
963 /**
964  * end_page_writeback - end writeback against a page
965  * @page: the page
966  */
967 void end_page_writeback(struct page *page)
968 {
969         /*
970          * TestClearPageReclaim could be used here but it is an atomic
971          * operation and overkill in this particular case. Failing to
972          * shuffle a page marked for immediate reclaim is too mild to
973          * justify taking an atomic operation penalty at the end of
974          * ever page writeback.
975          */
976         if (PageReclaim(page)) {
977                 ClearPageReclaim(page);
978                 rotate_reclaimable_page(page);
979         }
980 
981         if (!test_clear_page_writeback(page))
982                 BUG();
983 
984         smp_mb__after_atomic();
985         wake_up_page(page, PG_writeback);
986 }
987 EXPORT_SYMBOL(end_page_writeback);
988 
989 /*
990  * After completing I/O on a page, call this routine to update the page
991  * flags appropriately
992  */
993 void page_endio(struct page *page, bool is_write, int err)
994 {
995         if (!is_write) {
996                 if (!err) {
997                         SetPageUptodate(page);
998                 } else {
999                         ClearPageUptodate(page);
1000                         SetPageError(page);
1001                 }
1002                 unlock_page(page);
1003         } else {
1004                 if (err) {
1005                         SetPageError(page);
1006                         if (page->mapping)
1007                                 mapping_set_error(page->mapping, err);
1008                 }
1009                 end_page_writeback(page);
1010         }
1011 }
1012 EXPORT_SYMBOL_GPL(page_endio);
1013 
1014 /**
1015  * __lock_page - get a lock on the page, assuming we need to sleep to get it
1016  * @page: the page to lock
1017  */
1018 void __lock_page(struct page *__page)
1019 {
1020         struct page *page = compound_head(__page);
1021         wait_queue_head_t *q = page_waitqueue(page);
1022         wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
1023 }
1024 EXPORT_SYMBOL(__lock_page);
1025 
1026 int __lock_page_killable(struct page *__page)
1027 {
1028         struct page *page = compound_head(__page);
1029         wait_queue_head_t *q = page_waitqueue(page);
1030         return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
1031 }
1032 EXPORT_SYMBOL_GPL(__lock_page_killable);
1033 
1034 /*
1035  * Return values:
1036  * 1 - page is locked; mmap_sem is still held.
1037  * 0 - page is not locked.
1038  *     mmap_sem has been released (up_read()), unless flags had both
1039  *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1040  *     which case mmap_sem is still held.
1041  *
1042  * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1043  * with the page locked and the mmap_sem unperturbed.
1044  */
1045 int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1046                          unsigned int flags)
1047 {
1048         if (flags & FAULT_FLAG_ALLOW_RETRY) {
1049                 /*
1050                  * CAUTION! In this case, mmap_sem is not released
1051                  * even though return 0.
1052                  */
1053                 if (flags & FAULT_FLAG_RETRY_NOWAIT)
1054                         return 0;
1055 
1056                 up_read(&mm->mmap_sem);
1057                 if (flags & FAULT_FLAG_KILLABLE)
1058                         wait_on_page_locked_killable(page);
1059                 else
1060                         wait_on_page_locked(page);
1061                 return 0;
1062         } else {
1063                 if (flags & FAULT_FLAG_KILLABLE) {
1064                         int ret;
1065 
1066                         ret = __lock_page_killable(page);
1067                         if (ret) {
1068                                 up_read(&mm->mmap_sem);
1069                                 return 0;
1070                         }
1071                 } else
1072                         __lock_page(page);
1073                 return 1;
1074         }
1075 }
1076 
1077 /**
1078  * page_cache_next_hole - find the next hole (not-present entry)
1079  * @mapping: mapping
1080  * @index: index
1081  * @max_scan: maximum range to search
1082  *
1083  * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1084  * lowest indexed hole.
1085  *
1086  * Returns: the index of the hole if found, otherwise returns an index
1087  * outside of the set specified (in which case 'return - index >=
1088  * max_scan' will be true). In rare cases of index wrap-around, 0 will
1089  * be returned.
1090  *
1091  * page_cache_next_hole may be called under rcu_read_lock. However,
1092  * like radix_tree_gang_lookup, this will not atomically search a
1093  * snapshot of the tree at a single point in time. For example, if a
1094  * hole is created at index 5, then subsequently a hole is created at
1095  * index 10, page_cache_next_hole covering both indexes may return 10
1096  * if called under rcu_read_lock.
1097  */
1098 pgoff_t page_cache_next_hole(struct address_space *mapping,
1099                              pgoff_t index, unsigned long max_scan)
1100 {
1101         unsigned long i;
1102 
1103         for (i = 0; i < max_scan; i++) {
1104                 struct page *page;
1105 
1106                 page = radix_tree_lookup(&mapping->page_tree, index);
1107                 if (!page || radix_tree_exceptional_entry(page))
1108                         break;
1109                 index++;
1110                 if (index == 0)
1111                         break;
1112         }
1113 
1114         return index;
1115 }
1116 EXPORT_SYMBOL(page_cache_next_hole);
1117 
1118 /**
1119  * page_cache_prev_hole - find the prev hole (not-present entry)
1120  * @mapping: mapping
1121  * @index: index
1122  * @max_scan: maximum range to search
1123  *
1124  * Search backwards in the range [max(index-max_scan+1, 0), index] for
1125  * the first hole.
1126  *
1127  * Returns: the index of the hole if found, otherwise returns an index
1128  * outside of the set specified (in which case 'index - return >=
1129  * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1130  * will be returned.
1131  *
1132  * page_cache_prev_hole may be called under rcu_read_lock. However,
1133  * like radix_tree_gang_lookup, this will not atomically search a
1134  * snapshot of the tree at a single point in time. For example, if a
1135  * hole is created at index 10, then subsequently a hole is created at
1136  * index 5, page_cache_prev_hole covering both indexes may return 5 if
1137  * called under rcu_read_lock.
1138  */
1139 pgoff_t page_cache_prev_hole(struct address_space *mapping,
1140                              pgoff_t index, unsigned long max_scan)
1141 {
1142         unsigned long i;
1143 
1144         for (i = 0; i < max_scan; i++) {
1145                 struct page *page;
1146 
1147                 page = radix_tree_lookup(&mapping->page_tree, index);
1148                 if (!page || radix_tree_exceptional_entry(page))
1149                         break;
1150                 index--;
1151                 if (index == ULONG_MAX)
1152                         break;
1153         }
1154 
1155         return index;
1156 }
1157 EXPORT_SYMBOL(page_cache_prev_hole);
1158 
1159 /**
1160  * find_get_entry - find and get a page cache entry
1161  * @mapping: the address_space to search
1162  * @offset: the page cache index
1163  *
1164  * Looks up the page cache slot at @mapping & @offset.  If there is a
1165  * page cache page, it is returned with an increased refcount.
1166  *
1167  * If the slot holds a shadow entry of a previously evicted page, or a
1168  * swap entry from shmem/tmpfs, it is returned.
1169  *
1170  * Otherwise, %NULL is returned.
1171  */
1172 struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1173 {
1174         void **pagep;
1175         struct page *head, *page;
1176 
1177         rcu_read_lock();
1178 repeat:
1179         page = NULL;
1180         pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
1181         if (pagep) {
1182                 page = radix_tree_deref_slot(pagep);
1183                 if (unlikely(!page))
1184                         goto out;
1185                 if (radix_tree_exception(page)) {
1186                         if (radix_tree_deref_retry(page))
1187                                 goto repeat;
1188                         /*
1189                          * A shadow entry of a recently evicted page,
1190                          * or a swap entry from shmem/tmpfs.  Return
1191                          * it without attempting to raise page count.
1192                          */
1193                         goto out;
1194                 }
1195 
1196                 head = compound_head(page);
1197                 if (!page_cache_get_speculative(head))
1198                         goto repeat;
1199 
1200                 /* The page was split under us? */
1201                 if (compound_head(page) != head) {
1202                         put_page(head);
1203                         goto repeat;
1204                 }
1205 
1206                 /*
1207                  * Has the page moved?
1208                  * This is part of the lockless pagecache protocol. See
1209                  * include/linux/pagemap.h for details.
1210                  */
1211                 if (unlikely(page != *pagep)) {
1212                         put_page(head);
1213                         goto repeat;
1214                 }
1215         }
1216 out:
1217         rcu_read_unlock();
1218 
1219         return page;
1220 }
1221 EXPORT_SYMBOL(find_get_entry);
1222 
1223 /**
1224  * find_lock_entry - locate, pin and lock a page cache entry
1225  * @mapping: the address_space to search
1226  * @offset: the page cache index
1227  *
1228  * Looks up the page cache slot at @mapping & @offset.  If there is a
1229  * page cache page, it is returned locked and with an increased
1230  * refcount.
1231  *
1232  * If the slot holds a shadow entry of a previously evicted page, or a
1233  * swap entry from shmem/tmpfs, it is returned.
1234  *
1235  * Otherwise, %NULL is returned.
1236  *
1237  * find_lock_entry() may sleep.
1238  */
1239 struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1240 {
1241         struct page *page;
1242 
1243 repeat:
1244         page = find_get_entry(mapping, offset);
1245         if (page && !radix_tree_exception(page)) {
1246                 lock_page(page);
1247                 /* Has the page been truncated? */
1248                 if (unlikely(page_mapping(page) != mapping)) {
1249                         unlock_page(page);
1250                         put_page(page);
1251                         goto repeat;
1252                 }
1253                 VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1254         }
1255         return page;
1256 }
1257 EXPORT_SYMBOL(find_lock_entry);
1258 
1259 /**
1260  * pagecache_get_page - find and get a page reference
1261  * @mapping: the address_space to search
1262  * @offset: the page index
1263  * @fgp_flags: PCG flags
1264  * @gfp_mask: gfp mask to use for the page cache data page allocation
1265  *
1266  * Looks up the page cache slot at @mapping & @offset.
1267  *
1268  * PCG flags modify how the page is returned.
1269  *
1270  * FGP_ACCESSED: the page will be marked accessed
1271  * FGP_LOCK: Page is return locked
1272  * FGP_CREAT: If page is not present then a new page is allocated using
1273  *              @gfp_mask and added to the page cache and the VM's LRU
1274  *              list. The page is returned locked and with an increased
1275  *              refcount. Otherwise, %NULL is returned.
1276  *
1277  * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1278  * if the GFP flags specified for FGP_CREAT are atomic.
1279  *
1280  * If there is a page cache page, it is returned with an increased refcount.
1281  */
1282 struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1283         int fgp_flags, gfp_t gfp_mask)
1284 {
1285         struct page *page;
1286 
1287 repeat:
1288         page = find_get_entry(mapping, offset);
1289         if (radix_tree_exceptional_entry(page))
1290                 page = NULL;
1291         if (!page)
1292                 goto no_page;
1293 
1294         if (fgp_flags & FGP_LOCK) {
1295                 if (fgp_flags & FGP_NOWAIT) {
1296                         if (!trylock_page(page)) {
1297                                 put_page(page);
1298                                 return NULL;
1299                         }
1300                 } else {
1301                         lock_page(page);
1302                 }
1303 
1304                 /* Has the page been truncated? */
1305                 if (unlikely(page->mapping != mapping)) {
1306                         unlock_page(page);
1307                         put_page(page);
1308                         goto repeat;
1309                 }
1310                 VM_BUG_ON_PAGE(page->index != offset, page);
1311         }
1312 
1313         if (page && (fgp_flags & FGP_ACCESSED))
1314                 mark_page_accessed(page);
1315 
1316 no_page:
1317         if (!page && (fgp_flags & FGP_CREAT)) {
1318                 int err;
1319                 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1320                         gfp_mask |= __GFP_WRITE;
1321                 if (fgp_flags & FGP_NOFS)
1322                         gfp_mask &= ~__GFP_FS;
1323 
1324                 page = __page_cache_alloc(gfp_mask);
1325                 if (!page)
1326                         return NULL;
1327 
1328                 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1329                         fgp_flags |= FGP_LOCK;
1330 
1331                 /* Init accessed so avoid atomic mark_page_accessed later */
1332                 if (fgp_flags & FGP_ACCESSED)
1333                         __SetPageReferenced(page);
1334 
1335                 err = add_to_page_cache_lru(page, mapping, offset,
1336                                 gfp_mask & GFP_RECLAIM_MASK);
1337                 if (unlikely(err)) {
1338                         put_page(page);
1339                         page = NULL;
1340                         if (err == -EEXIST)
1341                                 goto repeat;
1342                 }
1343         }
1344 
1345         return page;
1346 }
1347 EXPORT_SYMBOL(pagecache_get_page);
1348 
1349 /**
1350  * find_get_entries - gang pagecache lookup
1351  * @mapping:    The address_space to search
1352  * @start:      The starting page cache index
1353  * @nr_entries: The maximum number of entries
1354  * @entries:    Where the resulting entries are placed
1355  * @indices:    The cache indices corresponding to the entries in @entries
1356  *
1357  * find_get_entries() will search for and return a group of up to
1358  * @nr_entries entries in the mapping.  The entries are placed at
1359  * @entries.  find_get_entries() takes a reference against any actual
1360  * pages it returns.
1361  *
1362  * The search returns a group of mapping-contiguous page cache entries
1363  * with ascending indexes.  There may be holes in the indices due to
1364  * not-present pages.
1365  *
1366  * Any shadow entries of evicted pages, or swap entries from
1367  * shmem/tmpfs, are included in the returned array.
1368  *
1369  * find_get_entries() returns the number of pages and shadow entries
1370  * which were found.
1371  */
1372 unsigned find_get_entries(struct address_space *mapping,
1373                           pgoff_t start, unsigned int nr_entries,
1374                           struct page **entries, pgoff_t *indices)
1375 {
1376         void **slot;
1377         unsigned int ret = 0;
1378         struct radix_tree_iter iter;
1379 
1380         if (!nr_entries)
1381                 return 0;
1382 
1383         rcu_read_lock();
1384         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1385                 struct page *head, *page;
1386 repeat:
1387                 page = radix_tree_deref_slot(slot);
1388                 if (unlikely(!page))
1389                         continue;
1390                 if (radix_tree_exception(page)) {
1391                         if (radix_tree_deref_retry(page)) {
1392                                 slot = radix_tree_iter_retry(&iter);
1393                                 continue;
1394                         }
1395                         /*
1396                          * A shadow entry of a recently evicted page, a swap
1397                          * entry from shmem/tmpfs or a DAX entry.  Return it
1398                          * without attempting to raise page count.
1399                          */
1400                         goto export;
1401                 }
1402 
1403                 head = compound_head(page);
1404                 if (!page_cache_get_speculative(head))
1405                         goto repeat;
1406 
1407                 /* The page was split under us? */
1408                 if (compound_head(page) != head) {
1409                         put_page(head);
1410                         goto repeat;
1411                 }
1412 
1413                 /* Has the page moved? */
1414                 if (unlikely(page != *slot)) {
1415                         put_page(head);
1416                         goto repeat;
1417                 }
1418 export:
1419                 indices[ret] = iter.index;
1420                 entries[ret] = page;
1421                 if (++ret == nr_entries)
1422                         break;
1423         }
1424         rcu_read_unlock();
1425         return ret;
1426 }
1427 
1428 /**
1429  * find_get_pages - gang pagecache lookup
1430  * @mapping:    The address_space to search
1431  * @start:      The starting page index
1432  * @nr_pages:   The maximum number of pages
1433  * @pages:      Where the resulting pages are placed
1434  *
1435  * find_get_pages() will search for and return a group of up to
1436  * @nr_pages pages in the mapping.  The pages are placed at @pages.
1437  * find_get_pages() takes a reference against the returned pages.
1438  *
1439  * The search returns a group of mapping-contiguous pages with ascending
1440  * indexes.  There may be holes in the indices due to not-present pages.
1441  *
1442  * find_get_pages() returns the number of pages which were found.
1443  */
1444 unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1445                             unsigned int nr_pages, struct page **pages)
1446 {
1447         struct radix_tree_iter iter;
1448         void **slot;
1449         unsigned ret = 0;
1450 
1451         if (unlikely(!nr_pages))
1452                 return 0;
1453 
1454         rcu_read_lock();
1455         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1456                 struct page *head, *page;
1457 repeat:
1458                 page = radix_tree_deref_slot(slot);
1459                 if (unlikely(!page))
1460                         continue;
1461 
1462                 if (radix_tree_exception(page)) {
1463                         if (radix_tree_deref_retry(page)) {
1464                                 slot = radix_tree_iter_retry(&iter);
1465                                 continue;
1466                         }
1467                         /*
1468                          * A shadow entry of a recently evicted page,
1469                          * or a swap entry from shmem/tmpfs.  Skip
1470                          * over it.
1471                          */
1472                         continue;
1473                 }
1474 
1475                 head = compound_head(page);
1476                 if (!page_cache_get_speculative(head))
1477                         goto repeat;
1478 
1479                 /* The page was split under us? */
1480                 if (compound_head(page) != head) {
1481                         put_page(head);
1482                         goto repeat;
1483                 }
1484 
1485                 /* Has the page moved? */
1486                 if (unlikely(page != *slot)) {
1487                         put_page(head);
1488                         goto repeat;
1489                 }
1490 
1491                 pages[ret] = page;
1492                 if (++ret == nr_pages)
1493                         break;
1494         }
1495 
1496         rcu_read_unlock();
1497         return ret;
1498 }
1499 
1500 /**
1501  * find_get_pages_contig - gang contiguous pagecache lookup
1502  * @mapping:    The address_space to search
1503  * @index:      The starting page index
1504  * @nr_pages:   The maximum number of pages
1505  * @pages:      Where the resulting pages are placed
1506  *
1507  * find_get_pages_contig() works exactly like find_get_pages(), except
1508  * that the returned number of pages are guaranteed to be contiguous.
1509  *
1510  * find_get_pages_contig() returns the number of pages which were found.
1511  */
1512 unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1513                                unsigned int nr_pages, struct page **pages)
1514 {
1515         struct radix_tree_iter iter;
1516         void **slot;
1517         unsigned int ret = 0;
1518 
1519         if (unlikely(!nr_pages))
1520                 return 0;
1521 
1522         rcu_read_lock();
1523         radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1524                 struct page *head, *page;
1525 repeat:
1526                 page = radix_tree_deref_slot(slot);
1527                 /* The hole, there no reason to continue */
1528                 if (unlikely(!page))
1529                         break;
1530 
1531                 if (radix_tree_exception(page)) {
1532                         if (radix_tree_deref_retry(page)) {
1533                                 slot = radix_tree_iter_retry(&iter);
1534                                 continue;
1535                         }
1536                         /*
1537                          * A shadow entry of a recently evicted page,
1538                          * or a swap entry from shmem/tmpfs.  Stop
1539                          * looking for contiguous pages.
1540                          */
1541                         break;
1542                 }
1543 
1544                 head = compound_head(page);
1545                 if (!page_cache_get_speculative(head))
1546                         goto repeat;
1547 
1548                 /* The page was split under us? */
1549                 if (compound_head(page) != head) {
1550                         put_page(head);
1551                         goto repeat;
1552                 }
1553 
1554                 /* Has the page moved? */
1555                 if (unlikely(page != *slot)) {
1556                         put_page(head);
1557                         goto repeat;
1558                 }
1559 
1560                 /*
1561                  * must check mapping and index after taking the ref.
1562                  * otherwise we can get both false positives and false
1563                  * negatives, which is just confusing to the caller.
1564                  */
1565                 if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1566                         put_page(page);
1567                         break;
1568                 }
1569 
1570                 pages[ret] = page;
1571                 if (++ret == nr_pages)
1572                         break;
1573         }
1574         rcu_read_unlock();
1575         return ret;
1576 }
1577 EXPORT_SYMBOL(find_get_pages_contig);
1578 
1579 /**
1580  * find_get_pages_tag - find and return pages that match @tag
1581  * @mapping:    the address_space to search
1582  * @index:      the starting page index
1583  * @tag:        the tag index
1584  * @nr_pages:   the maximum number of pages
1585  * @pages:      where the resulting pages are placed
1586  *
1587  * Like find_get_pages, except we only return pages which are tagged with
1588  * @tag.   We update @index to index the next page for the traversal.
1589  */
1590 unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1591                         int tag, unsigned int nr_pages, struct page **pages)
1592 {
1593         struct radix_tree_iter iter;
1594         void **slot;
1595         unsigned ret = 0;
1596 
1597         if (unlikely(!nr_pages))
1598                 return 0;
1599 
1600         rcu_read_lock();
1601         radix_tree_for_each_tagged(slot, &mapping->page_tree,
1602                                    &iter, *index, tag) {
1603                 struct page *head, *page;
1604 repeat:
1605                 page = radix_tree_deref_slot(slot);
1606                 if (unlikely(!page))
1607                         continue;
1608 
1609                 if (radix_tree_exception(page)) {
1610                         if (radix_tree_deref_retry(page)) {
1611                                 slot = radix_tree_iter_retry(&iter);
1612                                 continue;
1613                         }
1614                         /*
1615                          * A shadow entry of a recently evicted page.
1616                          *
1617                          * Those entries should never be tagged, but
1618                          * this tree walk is lockless and the tags are
1619                          * looked up in bulk, one radix tree node at a
1620                          * time, so there is a sizable window for page
1621                          * reclaim to evict a page we saw tagged.
1622                          *
1623                          * Skip over it.
1624                          */
1625                         continue;
1626                 }
1627 
1628                 head = compound_head(page);
1629                 if (!page_cache_get_speculative(head))
1630                         goto repeat;
1631 
1632                 /* The page was split under us? */
1633                 if (compound_head(page) != head) {
1634                         put_page(head);
1635                         goto repeat;
1636                 }
1637 
1638                 /* Has the page moved? */
1639                 if (unlikely(page != *slot)) {
1640                         put_page(head);
1641                         goto repeat;
1642                 }
1643 
1644                 pages[ret] = page;
1645                 if (++ret == nr_pages)
1646                         break;
1647         }
1648 
1649         rcu_read_unlock();
1650 
1651         if (ret)
1652                 *index = pages[ret - 1]->index + 1;
1653 
1654         return ret;
1655 }
1656 EXPORT_SYMBOL(find_get_pages_tag);
1657 
1658 /**
1659  * find_get_entries_tag - find and return entries that match @tag
1660  * @mapping:    the address_space to search
1661  * @start:      the starting page cache index
1662  * @tag:        the tag index
1663  * @nr_entries: the maximum number of entries
1664  * @entries:    where the resulting entries are placed
1665  * @indices:    the cache indices corresponding to the entries in @entries
1666  *
1667  * Like find_get_entries, except we only return entries which are tagged with
1668  * @tag.
1669  */
1670 unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1671                         int tag, unsigned int nr_entries,
1672                         struct page **entries, pgoff_t *indices)
1673 {
1674         void **slot;
1675         unsigned int ret = 0;
1676         struct radix_tree_iter iter;
1677 
1678         if (!nr_entries)
1679                 return 0;
1680 
1681         rcu_read_lock();
1682         radix_tree_for_each_tagged(slot, &mapping->page_tree,
1683                                    &iter, start, tag) {
1684                 struct page *head, *page;
1685 repeat:
1686                 page = radix_tree_deref_slot(slot);
1687                 if (unlikely(!page))
1688                         continue;
1689                 if (radix_tree_exception(page)) {
1690                         if (radix_tree_deref_retry(page)) {
1691                                 slot = radix_tree_iter_retry(&iter);
1692                                 continue;
1693                         }
1694 
1695                         /*
1696                          * A shadow entry of a recently evicted page, a swap
1697                          * entry from shmem/tmpfs or a DAX entry.  Return it
1698                          * without attempting to raise page count.
1699                          */
1700                         goto export;
1701                 }
1702 
1703                 head = compound_head(page);
1704                 if (!page_cache_get_speculative(head))
1705                         goto repeat;
1706 
1707                 /* The page was split under us? */
1708                 if (compound_head(page) != head) {
1709                         put_page(head);
1710                         goto repeat;
1711                 }
1712 
1713                 /* Has the page moved? */
1714                 if (unlikely(page != *slot)) {
1715                         put_page(head);
1716                         goto repeat;
1717                 }
1718 export:
1719                 indices[ret] = iter.index;
1720                 entries[ret] = page;
1721                 if (++ret == nr_entries)
1722                         break;
1723         }
1724         rcu_read_unlock();
1725         return ret;
1726 }
1727 EXPORT_SYMBOL(find_get_entries_tag);
1728 
1729 /*
1730  * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1731  * a _large_ part of the i/o request. Imagine the worst scenario:
1732  *
1733  *      ---R__________________________________________B__________
1734  *         ^ reading here                             ^ bad block(assume 4k)
1735  *
1736  * read(R) => miss => readahead(R...B) => media error => frustrating retries
1737  * => failing the whole request => read(R) => read(R+1) =>
1738  * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1739  * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1740  * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1741  *
1742  * It is going insane. Fix it by quickly scaling down the readahead size.
1743  */
1744 static void shrink_readahead_size_eio(struct file *filp,
1745                                         struct file_ra_state *ra)
1746 {
1747         ra->ra_pages /= 4;
1748 }
1749 
1750 /**
1751  * do_generic_file_read - generic file read routine
1752  * @filp:       the file to read
1753  * @ppos:       current file position
1754  * @iter:       data destination
1755  * @written:    already copied
1756  *
1757  * This is a generic file read routine, and uses the
1758  * mapping->a_ops->readpage() function for the actual low-level stuff.
1759  *
1760  * This is really ugly. But the goto's actually try to clarify some
1761  * of the logic when it comes to error handling etc.
1762  */
1763 static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1764                 struct iov_iter *iter, ssize_t written)
1765 {
1766         struct address_space *mapping = filp->f_mapping;
1767         struct inode *inode = mapping->host;
1768         struct file_ra_state *ra = &filp->f_ra;
1769         pgoff_t index;
1770         pgoff_t last_index;
1771         pgoff_t prev_index;
1772         unsigned long offset;      /* offset into pagecache page */
1773         unsigned int prev_offset;
1774         int error = 0;
1775 
1776         if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
1777                 return 0;
1778         iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
1779 
1780         index = *ppos >> PAGE_SHIFT;
1781         prev_index = ra->prev_pos >> PAGE_SHIFT;
1782         prev_offset = ra->prev_pos & (PAGE_SIZE-1);
1783         last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
1784         offset = *ppos & ~PAGE_MASK;
1785 
1786         for (;;) {
1787                 struct page *page;
1788                 pgoff_t end_index;
1789                 loff_t isize;
1790                 unsigned long nr, ret;
1791 
1792                 cond_resched();
1793 find_page:
1794                 if (fatal_signal_pending(current)) {
1795                         error = -EINTR;
1796                         goto out;
1797                 }
1798 
1799                 page = find_get_page(mapping, index);
1800                 if (!page) {
1801                         page_cache_sync_readahead(mapping,
1802                                         ra, filp,
1803                                         index, last_index - index);
1804                         page = find_get_page(mapping, index);
1805                         if (unlikely(page == NULL))
1806                                 goto no_cached_page;
1807                 }
1808                 if (PageReadahead(page)) {
1809                         page_cache_async_readahead(mapping,
1810                                         ra, filp, page,
1811                                         index, last_index - index);
1812                 }
1813                 if (!PageUptodate(page)) {
1814                         /*
1815                          * See comment in do_read_cache_page on why
1816                          * wait_on_page_locked is used to avoid unnecessarily
1817                          * serialisations and why it's safe.
1818                          */
1819                         error = wait_on_page_locked_killable(page);
1820                         if (unlikely(error))
1821                                 goto readpage_error;
1822                         if (PageUptodate(page))
1823                                 goto page_ok;
1824 
1825                         if (inode->i_blkbits == PAGE_SHIFT ||
1826                                         !mapping->a_ops->is_partially_uptodate)
1827                                 goto page_not_up_to_date;
1828                         /* pipes can't handle partially uptodate pages */
1829                         if (unlikely(iter->type & ITER_PIPE))
1830                                 goto page_not_up_to_date;
1831                         if (!trylock_page(page))
1832                                 goto page_not_up_to_date;
1833                         /* Did it get truncated before we got the lock? */
1834                         if (!page->mapping)
1835                                 goto page_not_up_to_date_locked;
1836                         if (!mapping->a_ops->is_partially_uptodate(page,
1837                                                         offset, iter->count))
1838                                 goto page_not_up_to_date_locked;
1839                         unlock_page(page);
1840                 }
1841 page_ok:
1842                 /*
1843                  * i_size must be checked after we know the page is Uptodate.
1844                  *
1845                  * Checking i_size after the check allows us to calculate
1846                  * the correct value for "nr", which means the zero-filled
1847                  * part of the page is not copied back to userspace (unless
1848                  * another truncate extends the file - this is desired though).
1849                  */
1850 
1851                 isize = i_size_read(inode);
1852                 end_index = (isize - 1) >> PAGE_SHIFT;
1853                 if (unlikely(!isize || index > end_index)) {
1854                         put_page(page);
1855                         goto out;
1856                 }
1857 
1858                 /* nr is the maximum number of bytes to copy from this page */
1859                 nr = PAGE_SIZE;
1860                 if (index == end_index) {
1861                         nr = ((isize - 1) & ~PAGE_MASK) + 1;
1862                         if (nr <= offset) {
1863                                 put_page(page);
1864                                 goto out;
1865                         }
1866                 }
1867                 nr = nr - offset;
1868 
1869                 /* If users can be writing to this page using arbitrary
1870                  * virtual addresses, take care about potential aliasing
1871                  * before reading the page on the kernel side.
1872                  */
1873                 if (mapping_writably_mapped(mapping))
1874                         flush_dcache_page(page);
1875 
1876                 /*
1877                  * When a sequential read accesses a page several times,
1878                  * only mark it as accessed the first time.
1879                  */
1880                 if (prev_index != index || offset != prev_offset)
1881                         mark_page_accessed(page);
1882                 prev_index = index;
1883 
1884                 /*
1885                  * Ok, we have the page, and it's up-to-date, so
1886                  * now we can copy it to user space...
1887                  */
1888 
1889                 ret = copy_page_to_iter(page, offset, nr, iter);
1890                 offset += ret;
1891                 index += offset >> PAGE_SHIFT;
1892                 offset &= ~PAGE_MASK;
1893                 prev_offset = offset;
1894 
1895                 put_page(page);
1896                 written += ret;
1897                 if (!iov_iter_count(iter))
1898                         goto out;
1899                 if (ret < nr) {
1900                         error = -EFAULT;
1901                         goto out;
1902                 }
1903                 continue;
1904 
1905 page_not_up_to_date:
1906                 /* Get exclusive access to the page ... */
1907                 error = lock_page_killable(page);
1908                 if (unlikely(error))
1909                         goto readpage_error;
1910 
1911 page_not_up_to_date_locked:
1912                 /* Did it get truncated before we got the lock? */
1913                 if (!page->mapping) {
1914                         unlock_page(page);
1915                         put_page(page);
1916                         continue;
1917                 }
1918 
1919                 /* Did somebody else fill it already? */
1920                 if (PageUptodate(page)) {
1921                         unlock_page(page);
1922                         goto page_ok;
1923                 }
1924 
1925 readpage:
1926                 /*
1927                  * A previous I/O error may have been due to temporary
1928                  * failures, eg. multipath errors.
1929                  * PG_error will be set again if readpage fails.
1930                  */
1931                 ClearPageError(page);
1932                 /* Start the actual read. The read will unlock the page. */
1933                 error = mapping->a_ops->readpage(filp, page);
1934 
1935                 if (unlikely(error)) {
1936                         if (error == AOP_TRUNCATED_PAGE) {
1937                                 put_page(page);
1938                                 error = 0;
1939                                 goto find_page;
1940                         }
1941                         goto readpage_error;
1942                 }
1943 
1944                 if (!PageUptodate(page)) {
1945                         error = lock_page_killable(page);
1946                         if (unlikely(error))
1947                                 goto readpage_error;
1948                         if (!PageUptodate(page)) {
1949                                 if (page->mapping == NULL) {
1950                                         /*
1951                                          * invalidate_mapping_pages got it
1952                                          */
1953                                         unlock_page(page);
1954                                         put_page(page);
1955                                         goto find_page;
1956                                 }
1957                                 unlock_page(page);
1958                                 shrink_readahead_size_eio(filp, ra);
1959                                 error = -EIO;
1960                                 goto readpage_error;
1961                         }
1962                         unlock_page(page);
1963                 }
1964 
1965                 goto page_ok;
1966 
1967 readpage_error:
1968                 /* UHHUH! A synchronous read error occurred. Report it */
1969                 put_page(page);
1970                 goto out;
1971 
1972 no_cached_page:
1973                 /*
1974                  * Ok, it wasn't cached, so we need to create a new
1975                  * page..
1976                  */
1977                 page = page_cache_alloc_cold(mapping);
1978                 if (!page) {
1979                         error = -ENOMEM;
1980                         goto out;
1981                 }
1982                 error = add_to_page_cache_lru(page, mapping, index,
1983                                 mapping_gfp_constraint(mapping, GFP_KERNEL));
1984                 if (error) {
1985                         put_page(page);
1986                         if (error == -EEXIST) {
1987                                 error = 0;
1988                                 goto find_page;
1989                         }
1990                         goto out;
1991                 }
1992                 goto readpage;
1993         }
1994 
1995 out:
1996         ra->prev_pos = prev_index;
1997         ra->prev_pos <<= PAGE_SHIFT;
1998         ra->prev_pos |= prev_offset;
1999 
2000         *ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2001         file_accessed(filp);
2002         return written ? written : error;
2003 }
2004 
2005 /**
2006  * generic_file_read_iter - generic filesystem read routine
2007  * @iocb:       kernel I/O control block
2008  * @iter:       destination for the data read
2009  *
2010  * This is the "read_iter()" routine for all filesystems
2011  * that can use the page cache directly.
2012  */
2013 ssize_t
2014 generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2015 {
2016         struct file *file = iocb->ki_filp;
2017         ssize_t retval = 0;
2018         size_t count = iov_iter_count(iter);
2019 
2020         if (!count)
2021                 goto out; /* skip atime */
2022 
2023         if (iocb->ki_flags & IOCB_DIRECT) {
2024                 struct address_space *mapping = file->f_mapping;
2025                 struct inode *inode = mapping->host;
2026                 struct iov_iter data = *iter;
2027                 loff_t size;
2028 
2029                 size = i_size_read(inode);
2030                 retval = filemap_write_and_wait_range(mapping, iocb->ki_pos,
2031                                         iocb->ki_pos + count - 1);
2032                 if (retval < 0)
2033                         goto out;
2034 
2035                 file_accessed(file);
2036 
2037                 retval = mapping->a_ops->direct_IO(iocb, &data);
2038                 if (retval >= 0) {
2039                         iocb->ki_pos += retval;
2040                         iov_iter_advance(iter, retval);
2041                 }
2042 
2043                 /*
2044                  * Btrfs can have a short DIO read if we encounter
2045                  * compressed extents, so if there was an error, or if
2046                  * we've already read everything we wanted to, or if
2047                  * there was a short read because we hit EOF, go ahead
2048                  * and return.  Otherwise fallthrough to buffered io for
2049                  * the rest of the read.  Buffered reads will not work for
2050                  * DAX files, so don't bother trying.
2051                  */
2052                 if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size ||
2053                     IS_DAX(inode))
2054                         goto out;
2055         }
2056 
2057         retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval);
2058 out:
2059         return retval;
2060 }
2061 EXPORT_SYMBOL(generic_file_read_iter);
2062 
2063 #ifdef CONFIG_MMU
2064 /**
2065  * page_cache_read - adds requested page to the page cache if not already there
2066  * @file:       file to read
2067  * @offset:     page index
2068  * @gfp_mask:   memory allocation flags
2069  *
2070  * This adds the requested page to the page cache if it isn't already there,
2071  * and schedules an I/O to read in its contents from disk.
2072  */
2073 static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2074 {
2075         struct address_space *mapping = file->f_mapping;
2076         struct page *page;
2077         int ret;
2078 
2079         do {
2080                 page = __page_cache_alloc(gfp_mask|__GFP_COLD);
2081                 if (!page)
2082                         return -ENOMEM;
2083 
2084                 ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL);
2085                 if (ret == 0)
2086                         ret = mapping->a_ops->readpage(file, page);
2087                 else if (ret == -EEXIST)
2088                         ret = 0; /* losing race to add is OK */
2089 
2090                 put_page(page);
2091 
2092         } while (ret == AOP_TRUNCATED_PAGE);
2093 
2094         return ret;
2095 }
2096 
2097 #define MMAP_LOTSAMISS  (100)
2098 
2099 /*
2100  * Synchronous readahead happens when we don't even find
2101  * a page in the page cache at all.
2102  */
2103 static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2104                                    struct file_ra_state *ra,
2105                                    struct file *file,
2106                                    pgoff_t offset)
2107 {
2108         struct address_space *mapping = file->f_mapping;
2109 
2110         /* If we don't want any read-ahead, don't bother */
2111         if (vma->vm_flags & VM_RAND_READ)
2112                 return;
2113         if (!ra->ra_pages)
2114                 return;
2115 
2116         if (vma->vm_flags & VM_SEQ_READ) {
2117                 page_cache_sync_readahead(mapping, ra, file, offset,
2118                                           ra->ra_pages);
2119                 return;
2120         }
2121 
2122         /* Avoid banging the cache line if not needed */
2123         if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2124                 ra->mmap_miss++;
2125 
2126         /*
2127          * Do we miss much more than hit in this file? If so,
2128          * stop bothering with read-ahead. It will only hurt.
2129          */
2130         if (ra->mmap_miss > MMAP_LOTSAMISS)
2131                 return;
2132 
2133         /*
2134          * mmap read-around
2135          */
2136         ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2137         ra->size = ra->ra_pages;
2138         ra->async_size = ra->ra_pages / 4;
2139         ra_submit(ra, mapping, file);
2140 }
2141 
2142 /*
2143  * Asynchronous readahead happens when we find the page and PG_readahead,
2144  * so we want to possibly extend the readahead further..
2145  */
2146 static void do_async_mmap_readahead(struct vm_area_struct *vma,
2147                                     struct file_ra_state *ra,
2148                                     struct file *file,
2149                                     struct page *page,
2150                                     pgoff_t offset)
2151 {
2152         struct address_space *mapping = file->f_mapping;
2153 
2154         /* If we don't want any read-ahead, don't bother */
2155         if (vma->vm_flags & VM_RAND_READ)
2156                 return;
2157         if (ra->mmap_miss > 0)
2158                 ra->mmap_miss--;
2159         if (PageReadahead(page))
2160                 page_cache_async_readahead(mapping, ra, file,
2161                                            page, offset, ra->ra_pages);
2162 }
2163 
2164 /**
2165  * filemap_fault - read in file data for page fault handling
2166  * @vma:        vma in which the fault was taken
2167  * @vmf:        struct vm_fault containing details of the fault
2168  *
2169  * filemap_fault() is invoked via the vma operations vector for a
2170  * mapped memory region to read in file data during a page fault.
2171  *
2172  * The goto's are kind of ugly, but this streamlines the normal case of having
2173  * it in the page cache, and handles the special cases reasonably without
2174  * having a lot of duplicated code.
2175  *
2176  * vma->vm_mm->mmap_sem must be held on entry.
2177  *
2178  * If our return value has VM_FAULT_RETRY set, it's because
2179  * lock_page_or_retry() returned 0.
2180  * The mmap_sem has usually been released in this case.
2181  * See __lock_page_or_retry() for the exception.
2182  *
2183  * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2184  * has not been released.
2185  *
2186  * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2187  */
2188 int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2189 {
2190         int error;
2191         struct file *file = vma->vm_file;
2192         struct address_space *mapping = file->f_mapping;
2193         struct file_ra_state *ra = &file->f_ra;
2194         struct inode *inode = mapping->host;
2195         pgoff_t offset = vmf->pgoff;
2196         struct page *page;
2197         loff_t size;
2198         int ret = 0;
2199 
2200         size = round_up(i_size_read(inode), PAGE_SIZE);
2201         if (offset >= size >> PAGE_SHIFT)
2202                 return VM_FAULT_SIGBUS;
2203 
2204         /*
2205          * Do we have something in the page cache already?
2206          */
2207         page = find_get_page(mapping, offset);
2208         if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2209                 /*
2210                  * We found the page, so try async readahead before
2211                  * waiting for the lock.
2212                  */
2213                 do_async_mmap_readahead(vma, ra, file, page, offset);
2214         } else if (!page) {
2215                 /* No page in the page cache at all */
2216                 do_sync_mmap_readahead(vma, ra, file, offset);
2217                 count_vm_event(PGMAJFAULT);
2218                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
2219                 ret = VM_FAULT_MAJOR;
2220 retry_find:
2221                 page = find_get_page(mapping, offset);
2222                 if (!page)
2223                         goto no_cached_page;
2224         }
2225 
2226         if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
2227                 put_page(page);
2228                 return ret | VM_FAULT_RETRY;
2229         }
2230 
2231         /* Did it get truncated? */
2232         if (unlikely(page->mapping != mapping)) {
2233                 unlock_page(page);
2234                 put_page(page);
2235                 goto retry_find;
2236         }
2237         VM_BUG_ON_PAGE(page->index != offset, page);
2238 
2239         /*
2240          * We have a locked page in the page cache, now we need to check
2241          * that it's up-to-date. If not, it is going to be due to an error.
2242          */
2243         if (unlikely(!PageUptodate(page)))
2244                 goto page_not_uptodate;
2245 
2246         /*
2247          * Found the page and have a reference on it.
2248          * We must recheck i_size under page lock.
2249          */
2250         size = round_up(i_size_read(inode), PAGE_SIZE);
2251         if (unlikely(offset >= size >> PAGE_SHIFT)) {
2252                 unlock_page(page);
2253                 put_page(page);
2254                 return VM_FAULT_SIGBUS;
2255         }
2256 
2257         vmf->page = page;
2258         return ret | VM_FAULT_LOCKED;
2259 
2260 no_cached_page:
2261         /*
2262          * We're only likely to ever get here if MADV_RANDOM is in
2263          * effect.
2264          */
2265         error = page_cache_read(file, offset, vmf->gfp_mask);
2266 
2267         /*
2268          * The page we want has now been added to the page cache.
2269          * In the unlikely event that someone removed it in the
2270          * meantime, we'll just come back here and read it again.
2271          */
2272         if (error >= 0)
2273                 goto retry_find;
2274 
2275         /*
2276          * An error return from page_cache_read can result if the
2277          * system is low on memory, or a problem occurs while trying
2278          * to schedule I/O.
2279          */
2280         if (error == -ENOMEM)
2281                 return VM_FAULT_OOM;
2282         return VM_FAULT_SIGBUS;
2283 
2284 page_not_uptodate:
2285         /*
2286          * Umm, take care of errors if the page isn't up-to-date.
2287          * Try to re-read it _once_. We do this synchronously,
2288          * because there really aren't any performance issues here
2289          * and we need to check for errors.
2290          */
2291         ClearPageError(page);
2292         error = mapping->a_ops->readpage(file, page);
2293         if (!error) {
2294                 wait_on_page_locked(page);
2295                 if (!PageUptodate(page))
2296                         error = -EIO;
2297         }
2298         put_page(page);
2299 
2300         if (!error || error == AOP_TRUNCATED_PAGE)
2301                 goto retry_find;
2302 
2303         /* Things didn't work out. Return zero to tell the mm layer so. */
2304         shrink_readahead_size_eio(file, ra);
2305         return VM_FAULT_SIGBUS;
2306 }
2307 EXPORT_SYMBOL(filemap_fault);
2308 
2309 void filemap_map_pages(struct vm_fault *vmf,
2310                 pgoff_t start_pgoff, pgoff_t end_pgoff)
2311 {
2312         struct radix_tree_iter iter;
2313         void **slot;
2314         struct file *file = vmf->vma->vm_file;
2315         struct address_space *mapping = file->f_mapping;
2316         pgoff_t last_pgoff = start_pgoff;
2317         loff_t size;
2318         struct page *head, *page;
2319 
2320         rcu_read_lock();
2321         radix_tree_for_each_slot(slot, &mapping->page_tree, &iter,
2322                         start_pgoff) {
2323                 if (iter.index > end_pgoff)
2324                         break;
2325 repeat:
2326                 page = radix_tree_deref_slot(slot);
2327                 if (unlikely(!page))
2328                         goto next;
2329                 if (radix_tree_exception(page)) {
2330                         if (radix_tree_deref_retry(page)) {
2331                                 slot = radix_tree_iter_retry(&iter);
2332                                 continue;
2333                         }
2334                         goto next;
2335                 }
2336 
2337                 head = compound_head(page);
2338                 if (!page_cache_get_speculative(head))
2339                         goto repeat;
2340 
2341                 /* The page was split under us? */
2342                 if (compound_head(page) != head) {
2343                         put_page(head);
2344                         goto repeat;
2345                 }
2346 
2347                 /* Has the page moved? */
2348                 if (unlikely(page != *slot)) {
2349                         put_page(head);
2350                         goto repeat;
2351                 }
2352 
2353                 if (!PageUptodate(page) ||
2354                                 PageReadahead(page) ||
2355                                 PageHWPoison(page))
2356                         goto skip;
2357                 if (!trylock_page(page))
2358                         goto skip;
2359 
2360                 if (page->mapping != mapping || !PageUptodate(page))
2361                         goto unlock;
2362 
2363                 size = round_up(i_size_read(mapping->host), PAGE_SIZE);
2364                 if (page->index >= size >> PAGE_SHIFT)
2365                         goto unlock;
2366 
2367                 if (file->f_ra.mmap_miss > 0)
2368                         file->f_ra.mmap_miss--;
2369 
2370                 vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2371                 if (vmf->pte)
2372                         vmf->pte += iter.index - last_pgoff;
2373                 last_pgoff = iter.index;
2374                 if (alloc_set_pte(vmf, NULL, page))
2375                         goto unlock;
2376                 unlock_page(page);
2377                 goto next;
2378 unlock:
2379                 unlock_page(page);
2380 skip:
2381                 put_page(page);
2382 next:
2383                 /* Huge page is mapped? No need to proceed. */
2384                 if (pmd_trans_huge(*vmf->pmd))
2385                         break;
2386                 if (iter.index == end_pgoff)
2387                         break;
2388         }
2389         rcu_read_unlock();
2390 }
2391 EXPORT_SYMBOL(filemap_map_pages);
2392 
2393 int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2394 {
2395         struct page *page = vmf->page;
2396         struct inode *inode = file_inode(vma->vm_file);
2397         int ret = VM_FAULT_LOCKED;
2398 
2399         sb_start_pagefault(inode->i_sb);
2400         file_update_time(vma->vm_file);
2401         lock_page(page);
2402         if (page->mapping != inode->i_mapping) {
2403                 unlock_page(page);
2404                 ret = VM_FAULT_NOPAGE;
2405                 goto out;
2406         }
2407         /*
2408          * We mark the page dirty already here so that when freeze is in
2409          * progress, we are guaranteed that writeback during freezing will
2410          * see the dirty page and writeprotect it again.
2411          */
2412         set_page_dirty(page);
2413         wait_for_stable_page(page);
2414 out:
2415         sb_end_pagefault(inode->i_sb);
2416         return ret;
2417 }
2418 EXPORT_SYMBOL(filemap_page_mkwrite);
2419 
2420 const struct vm_operations_struct generic_file_vm_ops = {
2421         .fault          = filemap_fault,
2422         .map_pages      = filemap_map_pages,
2423         .page_mkwrite   = filemap_page_mkwrite,
2424 };
2425 
2426 /* This is used for a general mmap of a disk file */
2427 
2428 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2429 {
2430         struct address_space *mapping = file->f_mapping;
2431 
2432         if (!mapping->a_ops->readpage)
2433                 return -ENOEXEC;
2434         file_accessed(file);
2435         vma->vm_ops = &generic_file_vm_ops;
2436         return 0;
2437 }
2438 
2439 /*
2440  * This is for filesystems which do not implement ->writepage.
2441  */
2442 int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2443 {
2444         if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2445                 return -EINVAL;
2446         return generic_file_mmap(file, vma);
2447 }
2448 #else
2449 int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2450 {
2451         return -ENOSYS;
2452 }
2453 int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2454 {
2455         return -ENOSYS;
2456 }
2457 #endif /* CONFIG_MMU */
2458 
2459 EXPORT_SYMBOL(generic_file_mmap);
2460 EXPORT_SYMBOL(generic_file_readonly_mmap);
2461 
2462 static struct page *wait_on_page_read(struct page *page)
2463 {
2464         if (!IS_ERR(page)) {
2465                 wait_on_page_locked(page);
2466                 if (!PageUptodate(page)) {
2467                         put_page(page);
2468                         page = ERR_PTR(-EIO);
2469                 }
2470         }
2471         return page;
2472 }
2473 
2474 static struct page *do_read_cache_page(struct address_space *mapping,
2475                                 pgoff_t index,
2476                                 int (*filler)(void *, struct page *),
2477                                 void *data,
2478                                 gfp_t gfp)
2479 {
2480         struct page *page;
2481         int err;
2482 repeat:
2483         page = find_get_page(mapping, index);
2484         if (!page) {
2485                 page = __page_cache_alloc(gfp | __GFP_COLD);
2486                 if (!page)
2487                         return ERR_PTR(-ENOMEM);
2488                 err = add_to_page_cache_lru(page, mapping, index, gfp);
2489                 if (unlikely(err)) {
2490                         put_page(page);
2491                         if (err == -EEXIST)
2492                                 goto repeat;
2493                         /* Presumably ENOMEM for radix tree node */
2494                         return ERR_PTR(err);
2495                 }
2496 
2497 filler:
2498                 err = filler(data, page);
2499                 if (err < 0) {
2500                         put_page(page);
2501                         return ERR_PTR(err);
2502                 }
2503 
2504                 page = wait_on_page_read(page);
2505                 if (IS_ERR(page))
2506                         return page;
2507                 goto out;
2508         }
2509         if (PageUptodate(page))
2510                 goto out;
2511 
2512         /*
2513          * Page is not up to date and may be locked due one of the following
2514          * case a: Page is being filled and the page lock is held
2515          * case b: Read/write error clearing the page uptodate status
2516          * case c: Truncation in progress (page locked)
2517          * case d: Reclaim in progress
2518          *
2519          * Case a, the page will be up to date when the page is unlocked.
2520          *    There is no need to serialise on the page lock here as the page
2521          *    is pinned so the lock gives no additional protection. Even if the
2522          *    the page is truncated, the data is still valid if PageUptodate as
2523          *    it's a race vs truncate race.
2524          * Case b, the page will not be up to date
2525          * Case c, the page may be truncated but in itself, the data may still
2526          *    be valid after IO completes as it's a read vs truncate race. The
2527          *    operation must restart if the page is not uptodate on unlock but
2528          *    otherwise serialising on page lock to stabilise the mapping gives
2529          *    no additional guarantees to the caller as the page lock is
2530          *    released before return.
2531          * Case d, similar to truncation. If reclaim holds the page lock, it
2532          *    will be a race with remove_mapping that determines if the mapping
2533          *    is valid on unlock but otherwise the data is valid and there is
2534          *    no need to serialise with page lock.
2535          *
2536          * As the page lock gives no additional guarantee, we optimistically
2537          * wait on the page to be unlocked and check if it's up to date and
2538          * use the page if it is. Otherwise, the page lock is required to
2539          * distinguish between the different cases. The motivation is that we
2540          * avoid spurious serialisations and wakeups when multiple processes
2541          * wait on the same page for IO to complete.
2542          */
2543         wait_on_page_locked(page);
2544         if (PageUptodate(page))
2545                 goto out;
2546 
2547         /* Distinguish between all the cases under the safety of the lock */
2548         lock_page(page);
2549 
2550         /* Case c or d, restart the operation */
2551         if (!page->mapping) {
2552                 unlock_page(page);
2553                 put_page(page);
2554                 goto repeat;
2555         }
2556 
2557         /* Someone else locked and filled the page in a very small window */
2558         if (PageUptodate(page)) {
2559                 unlock_page(page);
2560                 goto out;
2561         }
2562         goto filler;
2563 
2564 out:
2565         mark_page_accessed(page);
2566         return page;
2567 }
2568 
2569 /**
2570  * read_cache_page - read into page cache, fill it if needed
2571  * @mapping:    the page's address_space
2572  * @index:      the page index
2573  * @filler:     function to perform the read
2574  * @data:       first arg to filler(data, page) function, often left as NULL
2575  *
2576  * Read into the page cache. If a page already exists, and PageUptodate() is
2577  * not set, try to fill the page and wait for it to become unlocked.
2578  *
2579  * If the page does not get brought uptodate, return -EIO.
2580  */
2581 struct page *read_cache_page(struct address_space *mapping,
2582                                 pgoff_t index,
2583                                 int (*filler)(void *, struct page *),
2584                                 void *data)
2585 {
2586         return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2587 }
2588 EXPORT_SYMBOL(read_cache_page);
2589 
2590 /**
2591  * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2592  * @mapping:    the page's address_space
2593  * @index:      the page index
2594  * @gfp:        the page allocator flags to use if allocating
2595  *
2596  * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2597  * any new page allocations done using the specified allocation flags.
2598  *
2599  * If the page does not get brought uptodate, return -EIO.
2600  */
2601 struct page *read_cache_page_gfp(struct address_space *mapping,
2602                                 pgoff_t index,
2603                                 gfp_t gfp)
2604 {
2605         filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2606 
2607         return do_read_cache_page(mapping, index, filler, NULL, gfp);
2608 }
2609 EXPORT_SYMBOL(read_cache_page_gfp);
2610 
2611 /*
2612  * Performs necessary checks before doing a write
2613  *
2614  * Can adjust writing position or amount of bytes to write.
2615  * Returns appropriate error code that caller should return or
2616  * zero in case that write should be allowed.
2617  */
2618 inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2619 {
2620         struct file *file = iocb->ki_filp;
2621         struct inode *inode = file->f_mapping->host;
2622         unsigned long limit = rlimit(RLIMIT_FSIZE);
2623         loff_t pos;
2624 
2625         if (!iov_iter_count(from))
2626                 return 0;
2627 
2628         /* FIXME: this is for backwards compatibility with 2.4 */
2629         if (iocb->ki_flags & IOCB_APPEND)
2630                 iocb->ki_pos = i_size_read(inode);
2631 
2632         pos = iocb->ki_pos;
2633 
2634         if (limit != RLIM_INFINITY) {
2635                 if (iocb->ki_pos >= limit) {
2636                         send_sig(SIGXFSZ, current, 0);
2637                         return -EFBIG;
2638                 }
2639                 iov_iter_truncate(from, limit - (unsigned long)pos);
2640         }
2641 
2642         /*
2643          * LFS rule
2644          */
2645         if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2646                                 !(file->f_flags & O_LARGEFILE))) {
2647                 if (pos >= MAX_NON_LFS)
2648                         return -EFBIG;
2649                 iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
2650         }
2651 
2652         /*
2653          * Are we about to exceed the fs block limit ?
2654          *
2655          * If we have written data it becomes a short write.  If we have
2656          * exceeded without writing data we send a signal and return EFBIG.
2657          * Linus frestrict idea will clean these up nicely..
2658          */
2659         if (unlikely(pos >= inode->i_sb->s_maxbytes))
2660                 return -EFBIG;
2661 
2662         iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2663         return iov_iter_count(from);
2664 }
2665 EXPORT_SYMBOL(generic_write_checks);
2666 
2667 int pagecache_write_begin(struct file *file, struct address_space *mapping,
2668                                 loff_t pos, unsigned len, unsigned flags,
2669                                 struct page **pagep, void **fsdata)
2670 {
2671         const struct address_space_operations *aops = mapping->a_ops;
2672 
2673         return aops->write_begin(file, mapping, pos, len, flags,
2674                                                         pagep, fsdata);
2675 }
2676 EXPORT_SYMBOL(pagecache_write_begin);
2677 
2678 int pagecache_write_end(struct file *file, struct address_space *mapping,
2679                                 loff_t pos, unsigned len, unsigned copied,
2680                                 struct page *page, void *fsdata)
2681 {
2682         const struct address_space_operations *aops = mapping->a_ops;
2683 
2684         return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2685 }
2686 EXPORT_SYMBOL(pagecache_write_end);
2687 
2688 ssize_t
2689 generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
2690 {
2691         struct file     *file = iocb->ki_filp;
2692         struct address_space *mapping = file->f_mapping;
2693         struct inode    *inode = mapping->host;
2694         loff_t          pos = iocb->ki_pos;
2695         ssize_t         written;
2696         size_t          write_len;
2697         pgoff_t         end;
2698         struct iov_iter data;
2699 
2700         write_len = iov_iter_count(from);
2701         end = (pos + write_len - 1) >> PAGE_SHIFT;
2702 
2703         written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2704         if (written)
2705                 goto out;
2706 
2707         /*
2708          * After a write we want buffered reads to be sure to go to disk to get
2709          * the new data.  We invalidate clean cached page from the region we're
2710          * about to write.  We do this *before* the write so that we can return
2711          * without clobbering -EIOCBQUEUED from ->direct_IO().
2712          */
2713         if (mapping->nrpages) {
2714                 written = invalidate_inode_pages2_range(mapping,
2715                                         pos >> PAGE_SHIFT, end);
2716                 /*
2717                  * If a page can not be invalidated, return 0 to fall back
2718                  * to buffered write.
2719                  */
2720                 if (written) {
2721                         if (written == -EBUSY)
2722                                 return 0;
2723                         goto out;
2724                 }
2725         }
2726 
2727         data = *from;
2728         written = mapping->a_ops->direct_IO(iocb, &data);
2729 
2730         /*
2731          * Finally, try again to invalidate clean pages which might have been
2732          * cached by non-direct readahead, or faulted in by get_user_pages()
2733          * if the source of the write was an mmap'ed region of the file
2734          * we're writing.  Either one is a pretty crazy thing to do,
2735          * so we don't support it 100%.  If this invalidation
2736          * fails, tough, the write still worked...
2737          */
2738         if (mapping->nrpages) {
2739                 invalidate_inode_pages2_range(mapping,
2740                                               pos >> PAGE_SHIFT, end);
2741         }
2742 
2743         if (written > 0) {
2744                 pos += written;
2745                 iov_iter_advance(from, written);
2746                 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2747                         i_size_write(inode, pos);
2748                         mark_inode_dirty(inode);
2749                 }
2750                 iocb->ki_pos = pos;
2751         }
2752 out:
2753         return written;
2754 }
2755 EXPORT_SYMBOL(generic_file_direct_write);
2756 
2757 /*
2758  * Find or create a page at the given pagecache position. Return the locked
2759  * page. This function is specifically for buffered writes.
2760  */
2761 struct page *grab_cache_page_write_begin(struct address_space *mapping,
2762                                         pgoff_t index, unsigned flags)
2763 {
2764         struct page *page;
2765         int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
2766 
2767         if (flags & AOP_FLAG_NOFS)
2768                 fgp_flags |= FGP_NOFS;
2769 
2770         page = pagecache_get_page(mapping, index, fgp_flags,
2771                         mapping_gfp_mask(mapping));
2772         if (page)
2773                 wait_for_stable_page(page);
2774 
2775         return page;
2776 }
2777 EXPORT_SYMBOL(grab_cache_page_write_begin);
2778 
2779 ssize_t generic_perform_write(struct file *file,
2780                                 struct iov_iter *i, loff_t pos)
2781 {
2782         struct address_space *mapping = file->f_mapping;
2783         const struct address_space_operations *a_ops = mapping->a_ops;
2784         long status = 0;
2785         ssize_t written = 0;
2786         unsigned int flags = 0;
2787 
2788         /*
2789          * Copies from kernel address space cannot fail (NFSD is a big user).
2790          */
2791         if (!iter_is_iovec(i))
2792                 flags |= AOP_FLAG_UNINTERRUPTIBLE;
2793 
2794         do {
2795                 struct page *page;
2796                 unsigned long offset;   /* Offset into pagecache page */
2797                 unsigned long bytes;    /* Bytes to write to page */
2798                 size_t copied;          /* Bytes copied from user */
2799                 void *fsdata;
2800 
2801                 offset = (pos & (PAGE_SIZE - 1));
2802                 bytes = min_t(unsigned long, PAGE_SIZE - offset,
2803                                                 iov_iter_count(i));
2804 
2805 again:
2806                 /*
2807                  * Bring in the user page that we will copy from _first_.
2808                  * Otherwise there's a nasty deadlock on copying from the
2809                  * same page as we're writing to, without it being marked
2810                  * up-to-date.
2811                  *
2812                  * Not only is this an optimisation, but it is also required
2813                  * to check that the address is actually valid, when atomic
2814                  * usercopies are used, below.
2815                  */
2816                 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2817                         status = -EFAULT;
2818                         break;
2819                 }
2820 
2821                 if (fatal_signal_pending(current)) {
2822                         status = -EINTR;
2823                         break;
2824                 }
2825 
2826                 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2827                                                 &page, &fsdata);
2828                 if (unlikely(status < 0))
2829                         break;
2830 
2831                 if (mapping_writably_mapped(mapping))
2832                         flush_dcache_page(page);
2833 
2834                 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2835                 flush_dcache_page(page);
2836 
2837                 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2838                                                 page, fsdata);
2839                 if (unlikely(status < 0))
2840                         break;
2841                 copied = status;
2842 
2843                 cond_resched();
2844 
2845                 iov_iter_advance(i, copied);
2846                 if (unlikely(copied == 0)) {
2847                         /*
2848                          * If we were unable to copy any data at all, we must
2849                          * fall back to a single segment length write.
2850                          *
2851                          * If we didn't fallback here, we could livelock
2852                          * because not all segments in the iov can be copied at
2853                          * once without a pagefault.
2854                          */
2855                         bytes = min_t(unsigned long, PAGE_SIZE - offset,
2856                                                 iov_iter_single_seg_count(i));
2857                         goto again;
2858                 }
2859                 pos += copied;
2860                 written += copied;
2861 
2862                 balance_dirty_pages_ratelimited(mapping);
2863         } while (iov_iter_count(i));
2864 
2865         return written ? written : status;
2866 }
2867 EXPORT_SYMBOL(generic_perform_write);
2868 
2869 /**
2870  * __generic_file_write_iter - write data to a file
2871  * @iocb:       IO state structure (file, offset, etc.)
2872  * @from:       iov_iter with data to write
2873  *
2874  * This function does all the work needed for actually writing data to a
2875  * file. It does all basic checks, removes SUID from the file, updates
2876  * modification times and calls proper subroutines depending on whether we
2877  * do direct IO or a standard buffered write.
2878  *
2879  * It expects i_mutex to be grabbed unless we work on a block device or similar
2880  * object which does not need locking at all.
2881  *
2882  * This function does *not* take care of syncing data in case of O_SYNC write.
2883  * A caller has to handle it. This is mainly due to the fact that we want to
2884  * avoid syncing under i_mutex.
2885  */
2886 ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2887 {
2888         struct file *file = iocb->ki_filp;
2889         struct address_space * mapping = file->f_mapping;
2890         struct inode    *inode = mapping->host;
2891         ssize_t         written = 0;
2892         ssize_t         err;
2893         ssize_t         status;
2894 
2895         /* We can write back this queue in page reclaim */
2896         current->backing_dev_info = inode_to_bdi(inode);
2897         err = file_remove_privs(file);
2898         if (err)
2899                 goto out;
2900 
2901         err = file_update_time(file);
2902         if (err)
2903                 goto out;
2904 
2905         if (iocb->ki_flags & IOCB_DIRECT) {
2906                 loff_t pos, endbyte;
2907 
2908                 written = generic_file_direct_write(iocb, from);
2909                 /*
2910                  * If the write stopped short of completing, fall back to
2911                  * buffered writes.  Some filesystems do this for writes to
2912                  * holes, for example.  For DAX files, a buffered write will
2913                  * not succeed (even if it did, DAX does not handle dirty
2914                  * page-cache pages correctly).
2915                  */
2916                 if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
2917                         goto out;
2918 
2919                 status = generic_perform_write(file, from, pos = iocb->ki_pos);
2920                 /*
2921                  * If generic_perform_write() returned a synchronous error
2922                  * then we want to return the number of bytes which were
2923                  * direct-written, or the error code if that was zero.  Note
2924                  * that this differs from normal direct-io semantics, which
2925                  * will return -EFOO even if some bytes were written.
2926                  */
2927                 if (unlikely(status < 0)) {
2928                         err = status;
2929                         goto out;
2930                 }
2931                 /*
2932                  * We need to ensure that the page cache pages are written to
2933                  * disk and invalidated to preserve the expected O_DIRECT
2934                  * semantics.
2935                  */
2936                 endbyte = pos + status - 1;
2937                 err = filemap_write_and_wait_range(mapping, pos, endbyte);
2938                 if (err == 0) {
2939                         iocb->ki_pos = endbyte + 1;
2940                         written += status;
2941                         invalidate_mapping_pages(mapping,
2942                                                  pos >> PAGE_SHIFT,
2943                                                  endbyte >> PAGE_SHIFT);
2944                 } else {
2945                         /*
2946                          * We don't know how much we wrote, so just return
2947                          * the number of bytes which were direct-written
2948                          */
2949                 }
2950         } else {
2951                 written = generic_perform_write(file, from, iocb->ki_pos);
2952                 if (likely(written > 0))
2953                         iocb->ki_pos += written;
2954         }
2955 out:
2956         current->backing_dev_info = NULL;
2957         return written ? written : err;
2958 }
2959 EXPORT_SYMBOL(__generic_file_write_iter);
2960 
2961 /**
2962  * generic_file_write_iter - write data to a file
2963  * @iocb:       IO state structure
2964  * @from:       iov_iter with data to write
2965  *
2966  * This is a wrapper around __generic_file_write_iter() to be used by most
2967  * filesystems. It takes care of syncing the file in case of O_SYNC file
2968  * and acquires i_mutex as needed.
2969  */
2970 ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
2971 {
2972         struct file *file = iocb->ki_filp;
2973         struct inode *inode = file->f_mapping->host;
2974         ssize_t ret;
2975 
2976         inode_lock(inode);
2977         ret = generic_write_checks(iocb, from);
2978         if (ret > 0)
2979                 ret = __generic_file_write_iter(iocb, from);
2980         inode_unlock(inode);
2981 
2982         if (ret > 0)
2983                 ret = generic_write_sync(iocb, ret);
2984         return ret;
2985 }
2986 EXPORT_SYMBOL(generic_file_write_iter);
2987 
2988 /**
2989  * try_to_release_page() - release old fs-specific metadata on a page
2990  *
2991  * @page: the page which the kernel is trying to free
2992  * @gfp_mask: memory allocation flags (and I/O mode)
2993  *
2994  * The address_space is to try to release any data against the page
2995  * (presumably at page->private).  If the release was successful, return `1'.
2996  * Otherwise return zero.
2997  *
2998  * This may also be called if PG_fscache is set on a page, indicating that the
2999  * page is known to the local caching routines.
3000  *
3001  * The @gfp_mask argument specifies whether I/O may be performed to release
3002  * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3003  *
3004  */
3005 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3006 {
3007         struct address_space * const mapping = page->mapping;
3008 
3009         BUG_ON(!PageLocked(page));
3010         if (PageWriteback(page))
3011                 return 0;
3012 
3013         if (mapping && mapping->a_ops->releasepage)
3014                 return mapping->a_ops->releasepage(page, gfp_mask);
3015         return try_to_free_buffers(page);
3016 }
3017 
3018 EXPORT_SYMBOL(try_to_release_page);
3019 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us