]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - mm/filemap.c
hwmon: (k10temp) Add support for F15h M60h
[mirror_ubuntu-bionic-kernel.git] / mm / filemap.c
CommitLineData
1da177e4
LT
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 */
b95f1b31 12#include <linux/export.h>
1da177e4
LT
13#include <linux/compiler.h>
14#include <linux/fs.h>
c22ce143 15#include <linux/uaccess.h>
1da177e4 16#include <linux/aio.h>
c59ede7b 17#include <linux/capability.h>
1da177e4 18#include <linux/kernel_stat.h>
5a0e3ad6 19#include <linux/gfp.h>
1da177e4
LT
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>
53253383 28#include <linux/backing-dev.h>
1da177e4
LT
29#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
44110fe3 32#include <linux/cpuset.h>
2f718ffc 33#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
00501b53 34#include <linux/hugetlb.h>
8a9f3ccd 35#include <linux/memcontrol.h>
c515e1fd 36#include <linux/cleancache.h>
f1820361 37#include <linux/rmap.h>
0f8053a5
NP
38#include "internal.h"
39
fe0bfaaf
RJ
40#define CREATE_TRACE_POINTS
41#include <trace/events/filemap.h>
42
1da177e4 43/*
1da177e4
LT
44 * FIXME: remove all knowledge of the buffer layer from the core VM
45 */
148f948b 46#include <linux/buffer_head.h> /* for try_to_free_buffers */
1da177e4 47
1da177e4
LT
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 *
3d48ae45 65 * ->i_mmap_mutex (truncate_pagecache)
1da177e4 66 * ->private_lock (__free_pte->__set_page_dirty_buffers)
5d337b91
HD
67 * ->swap_lock (exclusive_swap_page, others)
68 * ->mapping->tree_lock
1da177e4 69 *
1b1dcc1b 70 * ->i_mutex
3d48ae45 71 * ->i_mmap_mutex (truncate->unmap_mapping_range)
1da177e4
LT
72 *
73 * ->mmap_sem
3d48ae45 74 * ->i_mmap_mutex
b8072f09 75 * ->page_table_lock or pte_lock (various, mainly in memory.c)
1da177e4
LT
76 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
77 *
78 * ->mmap_sem
79 * ->lock_page (access_process_vm)
80 *
ccad2365 81 * ->i_mutex (generic_perform_write)
82591e6e 82 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
1da177e4 83 *
f758eeab 84 * bdi->wb.list_lock
a66979ab 85 * sb_lock (fs/fs-writeback.c)
1da177e4
LT
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
3d48ae45 88 * ->i_mmap_mutex
1da177e4
LT
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
b8072f09 92 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
1da177e4 93 *
b8072f09 94 * ->page_table_lock or pte_lock
5d337b91 95 * ->swap_lock (try_to_unmap_one)
1da177e4
LT
96 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
053837fc 99 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
1da177e4
LT
100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
f758eeab 102 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
250df6ed 103 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
f758eeab 104 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
250df6ed 105 * ->inode->i_lock (zap_pte_range->set_page_dirty)
1da177e4
LT
106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
107 *
9a3c531d
AK
108 * ->i_mmap_mutex
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
1da177e4
LT
110 */
111
91b0abe3
JW
112static void page_cache_tree_delete(struct address_space *mapping,
113 struct page *page, void *shadow)
114{
449dd698
JW
115 struct radix_tree_node *node;
116 unsigned long index;
117 unsigned int offset;
118 unsigned int tag;
119 void **slot;
91b0abe3 120
449dd698
JW
121 VM_BUG_ON(!PageLocked(page));
122
123 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
124
125 if (shadow) {
91b0abe3
JW
126 mapping->nrshadows++;
127 /*
128 * Make sure the nrshadows update is committed before
129 * the nrpages update so that final truncate racing
130 * with reclaim does not see both counters 0 at the
131 * same time and miss a shadow entry.
132 */
133 smp_wmb();
449dd698 134 }
91b0abe3 135 mapping->nrpages--;
449dd698
JW
136
137 if (!node) {
138 /* Clear direct pointer tags in root node */
139 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
140 radix_tree_replace_slot(slot, shadow);
141 return;
142 }
143
144 /* Clear tree tags for the removed page */
145 index = page->index;
146 offset = index & RADIX_TREE_MAP_MASK;
147 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
148 if (test_bit(offset, node->tags[tag]))
149 radix_tree_tag_clear(&mapping->page_tree, index, tag);
150 }
151
152 /* Delete page, swap shadow entry */
153 radix_tree_replace_slot(slot, shadow);
154 workingset_node_pages_dec(node);
155 if (shadow)
156 workingset_node_shadows_inc(node);
157 else
158 if (__radix_tree_delete_node(&mapping->page_tree, node))
159 return;
160
161 /*
162 * Track node that only contains shadow entries.
163 *
164 * Avoid acquiring the list_lru lock if already tracked. The
165 * list_empty() test is safe as node->private_list is
166 * protected by mapping->tree_lock.
167 */
168 if (!workingset_node_pages(node) &&
169 list_empty(&node->private_list)) {
170 node->private_data = mapping;
171 list_lru_add(&workingset_shadow_nodes, &node->private_list);
172 }
91b0abe3
JW
173}
174
1da177e4 175/*
e64a782f 176 * Delete a page from the page cache and free it. Caller has to make
1da177e4 177 * sure the page is locked and that nobody else uses it - or that usage
19fd6231 178 * is safe. The caller must hold the mapping's tree_lock.
1da177e4 179 */
91b0abe3 180void __delete_from_page_cache(struct page *page, void *shadow)
1da177e4
LT
181{
182 struct address_space *mapping = page->mapping;
183
fe0bfaaf 184 trace_mm_filemap_delete_from_page_cache(page);
c515e1fd
DM
185 /*
186 * if we're uptodate, flush out into the cleancache, otherwise
187 * invalidate any existing cleancache entries. We can't leave
188 * stale data around in the cleancache once our page is gone
189 */
190 if (PageUptodate(page) && PageMappedToDisk(page))
191 cleancache_put_page(page);
192 else
3167760f 193 cleancache_invalidate_page(mapping, page);
c515e1fd 194
91b0abe3
JW
195 page_cache_tree_delete(mapping, page, shadow);
196
1da177e4 197 page->mapping = NULL;
b85e0eff 198 /* Leave page->index set: truncation lookup relies upon it */
91b0abe3 199
347ce434 200 __dec_zone_page_state(page, NR_FILE_PAGES);
4b02108a
KM
201 if (PageSwapBacked(page))
202 __dec_zone_page_state(page, NR_SHMEM);
45426812 203 BUG_ON(page_mapped(page));
3a692790
LT
204
205 /*
206 * Some filesystems seem to re-dirty the page even after
207 * the VM has canceled the dirty bit (eg ext3 journaling).
208 *
209 * Fix it up by doing a final dirty accounting check after
210 * having removed the page entirely.
211 */
212 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
213 dec_zone_page_state(page, NR_FILE_DIRTY);
214 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
215 }
1da177e4
LT
216}
217
702cfbf9
MK
218/**
219 * delete_from_page_cache - delete page from page cache
220 * @page: the page which the kernel is trying to remove from page cache
221 *
222 * This must be called only on pages that have been verified to be in the page
223 * cache and locked. It will never put the page into the free list, the caller
224 * has a reference on the page.
225 */
226void delete_from_page_cache(struct page *page)
1da177e4
LT
227{
228 struct address_space *mapping = page->mapping;
6072d13c 229 void (*freepage)(struct page *);
1da177e4 230
cd7619d6 231 BUG_ON(!PageLocked(page));
1da177e4 232
6072d13c 233 freepage = mapping->a_ops->freepage;
19fd6231 234 spin_lock_irq(&mapping->tree_lock);
91b0abe3 235 __delete_from_page_cache(page, NULL);
19fd6231 236 spin_unlock_irq(&mapping->tree_lock);
6072d13c
LT
237
238 if (freepage)
239 freepage(page);
97cecb5a
MK
240 page_cache_release(page);
241}
242EXPORT_SYMBOL(delete_from_page_cache);
243
865ffef3
DM
244static int filemap_check_errors(struct address_space *mapping)
245{
246 int ret = 0;
247 /* Check for outstanding write errors */
7fcbbaf1
JA
248 if (test_bit(AS_ENOSPC, &mapping->flags) &&
249 test_and_clear_bit(AS_ENOSPC, &mapping->flags))
865ffef3 250 ret = -ENOSPC;
7fcbbaf1
JA
251 if (test_bit(AS_EIO, &mapping->flags) &&
252 test_and_clear_bit(AS_EIO, &mapping->flags))
865ffef3
DM
253 ret = -EIO;
254 return ret;
255}
256
1da177e4 257/**
485bb99b 258 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
67be2dd1
MW
259 * @mapping: address space structure to write
260 * @start: offset in bytes where the range starts
469eb4d0 261 * @end: offset in bytes where the range ends (inclusive)
67be2dd1 262 * @sync_mode: enable synchronous operation
1da177e4 263 *
485bb99b
RD
264 * Start writeback against all of a mapping's dirty pages that lie
265 * within the byte offsets <start, end> inclusive.
266 *
1da177e4 267 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
485bb99b 268 * opposed to a regular memory cleansing writeback. The difference between
1da177e4
LT
269 * these two operations is that if a dirty page/buffer is encountered, it must
270 * be waited upon, and not just skipped over.
271 */
ebcf28e1
AM
272int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
273 loff_t end, int sync_mode)
1da177e4
LT
274{
275 int ret;
276 struct writeback_control wbc = {
277 .sync_mode = sync_mode,
05fe478d 278 .nr_to_write = LONG_MAX,
111ebb6e
OH
279 .range_start = start,
280 .range_end = end,
1da177e4
LT
281 };
282
283 if (!mapping_cap_writeback_dirty(mapping))
284 return 0;
285
286 ret = do_writepages(mapping, &wbc);
287 return ret;
288}
289
290static inline int __filemap_fdatawrite(struct address_space *mapping,
291 int sync_mode)
292{
111ebb6e 293 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
1da177e4
LT
294}
295
296int filemap_fdatawrite(struct address_space *mapping)
297{
298 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
299}
300EXPORT_SYMBOL(filemap_fdatawrite);
301
f4c0a0fd 302int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
ebcf28e1 303 loff_t end)
1da177e4
LT
304{
305 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
306}
f4c0a0fd 307EXPORT_SYMBOL(filemap_fdatawrite_range);
1da177e4 308
485bb99b
RD
309/**
310 * filemap_flush - mostly a non-blocking flush
311 * @mapping: target address_space
312 *
1da177e4
LT
313 * This is a mostly non-blocking flush. Not suitable for data-integrity
314 * purposes - I/O may not be started against all dirty pages.
315 */
316int filemap_flush(struct address_space *mapping)
317{
318 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
319}
320EXPORT_SYMBOL(filemap_flush);
321
485bb99b 322/**
94004ed7
CH
323 * filemap_fdatawait_range - wait for writeback to complete
324 * @mapping: address space structure to wait for
325 * @start_byte: offset in bytes where the range starts
326 * @end_byte: offset in bytes where the range ends (inclusive)
485bb99b 327 *
94004ed7
CH
328 * Walk the list of under-writeback pages of the given address space
329 * in the given range and wait for all of them.
1da177e4 330 */
94004ed7
CH
331int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
332 loff_t end_byte)
1da177e4 333{
94004ed7
CH
334 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
335 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
1da177e4
LT
336 struct pagevec pvec;
337 int nr_pages;
865ffef3 338 int ret2, ret = 0;
1da177e4 339
94004ed7 340 if (end_byte < start_byte)
865ffef3 341 goto out;
1da177e4
LT
342
343 pagevec_init(&pvec, 0);
1da177e4
LT
344 while ((index <= end) &&
345 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
346 PAGECACHE_TAG_WRITEBACK,
347 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
348 unsigned i;
349
350 for (i = 0; i < nr_pages; i++) {
351 struct page *page = pvec.pages[i];
352
353 /* until radix tree lookup accepts end_index */
354 if (page->index > end)
355 continue;
356
357 wait_on_page_writeback(page);
212260aa 358 if (TestClearPageError(page))
1da177e4
LT
359 ret = -EIO;
360 }
361 pagevec_release(&pvec);
362 cond_resched();
363 }
865ffef3
DM
364out:
365 ret2 = filemap_check_errors(mapping);
366 if (!ret)
367 ret = ret2;
1da177e4
LT
368
369 return ret;
370}
d3bccb6f
JK
371EXPORT_SYMBOL(filemap_fdatawait_range);
372
1da177e4 373/**
485bb99b 374 * filemap_fdatawait - wait for all under-writeback pages to complete
1da177e4 375 * @mapping: address space structure to wait for
485bb99b
RD
376 *
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
1da177e4
LT
379 */
380int filemap_fdatawait(struct address_space *mapping)
381{
382 loff_t i_size = i_size_read(mapping->host);
383
384 if (i_size == 0)
385 return 0;
386
94004ed7 387 return filemap_fdatawait_range(mapping, 0, i_size - 1);
1da177e4
LT
388}
389EXPORT_SYMBOL(filemap_fdatawait);
390
391int filemap_write_and_wait(struct address_space *mapping)
392{
28fd1298 393 int err = 0;
1da177e4
LT
394
395 if (mapping->nrpages) {
28fd1298
OH
396 err = filemap_fdatawrite(mapping);
397 /*
398 * Even if the above returned error, the pages may be
399 * written partially (e.g. -ENOSPC), so we wait for it.
400 * But the -EIO is special case, it may indicate the worst
401 * thing (e.g. bug) happened, so we avoid waiting for it.
402 */
403 if (err != -EIO) {
404 int err2 = filemap_fdatawait(mapping);
405 if (!err)
406 err = err2;
407 }
865ffef3
DM
408 } else {
409 err = filemap_check_errors(mapping);
1da177e4 410 }
28fd1298 411 return err;
1da177e4 412}
28fd1298 413EXPORT_SYMBOL(filemap_write_and_wait);
1da177e4 414
485bb99b
RD
415/**
416 * filemap_write_and_wait_range - write out & wait on a file range
417 * @mapping: the address_space for the pages
418 * @lstart: offset in bytes where the range starts
419 * @lend: offset in bytes where the range ends (inclusive)
420 *
469eb4d0
AM
421 * Write out and wait upon file offsets lstart->lend, inclusive.
422 *
423 * Note that `lend' is inclusive (describes the last byte to be written) so
424 * that this function can be used to write to the very end-of-file (end = -1).
425 */
1da177e4
LT
426int filemap_write_and_wait_range(struct address_space *mapping,
427 loff_t lstart, loff_t lend)
428{
28fd1298 429 int err = 0;
1da177e4
LT
430
431 if (mapping->nrpages) {
28fd1298
OH
432 err = __filemap_fdatawrite_range(mapping, lstart, lend,
433 WB_SYNC_ALL);
434 /* See comment of filemap_write_and_wait() */
435 if (err != -EIO) {
94004ed7
CH
436 int err2 = filemap_fdatawait_range(mapping,
437 lstart, lend);
28fd1298
OH
438 if (!err)
439 err = err2;
440 }
865ffef3
DM
441 } else {
442 err = filemap_check_errors(mapping);
1da177e4 443 }
28fd1298 444 return err;
1da177e4 445}
f6995585 446EXPORT_SYMBOL(filemap_write_and_wait_range);
1da177e4 447
ef6a3c63
MS
448/**
449 * replace_page_cache_page - replace a pagecache page with a new one
450 * @old: page to be replaced
451 * @new: page to replace with
452 * @gfp_mask: allocation mode
453 *
454 * This function replaces a page in the pagecache with a new one. On
455 * success it acquires the pagecache reference for the new page and
456 * drops it for the old page. Both the old and new pages must be
457 * locked. This function does not add the new page to the LRU, the
458 * caller must do that.
459 *
460 * The remove + add is atomic. The only way this function can fail is
461 * memory allocation failure.
462 */
463int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
464{
465 int error;
ef6a3c63 466
309381fe
SL
467 VM_BUG_ON_PAGE(!PageLocked(old), old);
468 VM_BUG_ON_PAGE(!PageLocked(new), new);
469 VM_BUG_ON_PAGE(new->mapping, new);
ef6a3c63 470
ef6a3c63
MS
471 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
472 if (!error) {
473 struct address_space *mapping = old->mapping;
474 void (*freepage)(struct page *);
475
476 pgoff_t offset = old->index;
477 freepage = mapping->a_ops->freepage;
478
479 page_cache_get(new);
480 new->mapping = mapping;
481 new->index = offset;
482
483 spin_lock_irq(&mapping->tree_lock);
91b0abe3 484 __delete_from_page_cache(old, NULL);
ef6a3c63
MS
485 error = radix_tree_insert(&mapping->page_tree, offset, new);
486 BUG_ON(error);
487 mapping->nrpages++;
488 __inc_zone_page_state(new, NR_FILE_PAGES);
489 if (PageSwapBacked(new))
490 __inc_zone_page_state(new, NR_SHMEM);
491 spin_unlock_irq(&mapping->tree_lock);
0a31bc97 492 mem_cgroup_migrate(old, new, true);
ef6a3c63
MS
493 radix_tree_preload_end();
494 if (freepage)
495 freepage(old);
496 page_cache_release(old);
ef6a3c63
MS
497 }
498
499 return error;
500}
501EXPORT_SYMBOL_GPL(replace_page_cache_page);
502
0cd6144a 503static int page_cache_tree_insert(struct address_space *mapping,
a528910e 504 struct page *page, void **shadowp)
0cd6144a 505{
449dd698 506 struct radix_tree_node *node;
0cd6144a
JW
507 void **slot;
508 int error;
509
449dd698
JW
510 error = __radix_tree_create(&mapping->page_tree, page->index,
511 &node, &slot);
512 if (error)
513 return error;
514 if (*slot) {
0cd6144a
JW
515 void *p;
516
517 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
518 if (!radix_tree_exceptional_entry(p))
519 return -EEXIST;
a528910e
JW
520 if (shadowp)
521 *shadowp = p;
449dd698
JW
522 mapping->nrshadows--;
523 if (node)
524 workingset_node_shadows_dec(node);
0cd6144a 525 }
449dd698
JW
526 radix_tree_replace_slot(slot, page);
527 mapping->nrpages++;
528 if (node) {
529 workingset_node_pages_inc(node);
530 /*
531 * Don't track node that contains actual pages.
532 *
533 * Avoid acquiring the list_lru lock if already
534 * untracked. The list_empty() test is safe as
535 * node->private_list is protected by
536 * mapping->tree_lock.
537 */
538 if (!list_empty(&node->private_list))
539 list_lru_del(&workingset_shadow_nodes,
540 &node->private_list);
541 }
542 return 0;
0cd6144a
JW
543}
544
a528910e
JW
545static int __add_to_page_cache_locked(struct page *page,
546 struct address_space *mapping,
547 pgoff_t offset, gfp_t gfp_mask,
548 void **shadowp)
1da177e4 549{
00501b53
JW
550 int huge = PageHuge(page);
551 struct mem_cgroup *memcg;
e286781d
NP
552 int error;
553
309381fe
SL
554 VM_BUG_ON_PAGE(!PageLocked(page), page);
555 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
e286781d 556
00501b53
JW
557 if (!huge) {
558 error = mem_cgroup_try_charge(page, current->mm,
559 gfp_mask, &memcg);
560 if (error)
561 return error;
562 }
1da177e4 563
5e4c0d97 564 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
66a0c8ee 565 if (error) {
00501b53
JW
566 if (!huge)
567 mem_cgroup_cancel_charge(page, memcg);
66a0c8ee
KS
568 return error;
569 }
570
571 page_cache_get(page);
572 page->mapping = mapping;
573 page->index = offset;
574
575 spin_lock_irq(&mapping->tree_lock);
a528910e 576 error = page_cache_tree_insert(mapping, page, shadowp);
66a0c8ee
KS
577 radix_tree_preload_end();
578 if (unlikely(error))
579 goto err_insert;
66a0c8ee
KS
580 __inc_zone_page_state(page, NR_FILE_PAGES);
581 spin_unlock_irq(&mapping->tree_lock);
00501b53
JW
582 if (!huge)
583 mem_cgroup_commit_charge(page, memcg, false);
66a0c8ee
KS
584 trace_mm_filemap_add_to_page_cache(page);
585 return 0;
586err_insert:
587 page->mapping = NULL;
588 /* Leave page->index set: truncation relies upon it */
589 spin_unlock_irq(&mapping->tree_lock);
00501b53
JW
590 if (!huge)
591 mem_cgroup_cancel_charge(page, memcg);
66a0c8ee 592 page_cache_release(page);
1da177e4
LT
593 return error;
594}
a528910e
JW
595
596/**
597 * add_to_page_cache_locked - add a locked page to the pagecache
598 * @page: page to add
599 * @mapping: the page's address_space
600 * @offset: page index
601 * @gfp_mask: page allocation mode
602 *
603 * This function is used to add a page to the pagecache. It must be locked.
604 * This function does not add the page to the LRU. The caller must do that.
605 */
606int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
607 pgoff_t offset, gfp_t gfp_mask)
608{
609 return __add_to_page_cache_locked(page, mapping, offset,
610 gfp_mask, NULL);
611}
e286781d 612EXPORT_SYMBOL(add_to_page_cache_locked);
1da177e4
LT
613
614int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
6daa0e28 615 pgoff_t offset, gfp_t gfp_mask)
1da177e4 616{
a528910e 617 void *shadow = NULL;
4f98a2fe
RR
618 int ret;
619
a528910e
JW
620 __set_page_locked(page);
621 ret = __add_to_page_cache_locked(page, mapping, offset,
622 gfp_mask, &shadow);
623 if (unlikely(ret))
624 __clear_page_locked(page);
625 else {
626 /*
627 * The page might have been evicted from cache only
628 * recently, in which case it should be activated like
629 * any other repeatedly accessed page.
630 */
631 if (shadow && workingset_refault(shadow)) {
632 SetPageActive(page);
633 workingset_activation(page);
634 } else
635 ClearPageActive(page);
636 lru_cache_add(page);
637 }
1da177e4
LT
638 return ret;
639}
18bc0bbd 640EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
1da177e4 641
44110fe3 642#ifdef CONFIG_NUMA
2ae88149 643struct page *__page_cache_alloc(gfp_t gfp)
44110fe3 644{
c0ff7453
MX
645 int n;
646 struct page *page;
647
44110fe3 648 if (cpuset_do_page_mem_spread()) {
cc9a6c87
MG
649 unsigned int cpuset_mems_cookie;
650 do {
d26914d1 651 cpuset_mems_cookie = read_mems_allowed_begin();
cc9a6c87
MG
652 n = cpuset_mem_spread_node();
653 page = alloc_pages_exact_node(n, gfp, 0);
d26914d1 654 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
cc9a6c87 655
c0ff7453 656 return page;
44110fe3 657 }
2ae88149 658 return alloc_pages(gfp, 0);
44110fe3 659}
2ae88149 660EXPORT_SYMBOL(__page_cache_alloc);
44110fe3
PJ
661#endif
662
1da177e4
LT
663/*
664 * In order to wait for pages to become available there must be
665 * waitqueues associated with pages. By using a hash table of
666 * waitqueues where the bucket discipline is to maintain all
667 * waiters on the same queue and wake all when any of the pages
668 * become available, and for the woken contexts to check to be
669 * sure the appropriate page became available, this saves space
670 * at a cost of "thundering herd" phenomena during rare hash
671 * collisions.
672 */
673static wait_queue_head_t *page_waitqueue(struct page *page)
674{
675 const struct zone *zone = page_zone(page);
676
677 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
678}
679
680static inline void wake_up_page(struct page *page, int bit)
681{
682 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
683}
684
920c7a5d 685void wait_on_page_bit(struct page *page, int bit_nr)
1da177e4
LT
686{
687 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
688
689 if (test_bit(bit_nr, &page->flags))
74316201 690 __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io,
1da177e4
LT
691 TASK_UNINTERRUPTIBLE);
692}
693EXPORT_SYMBOL(wait_on_page_bit);
694
f62e00cc
KM
695int wait_on_page_bit_killable(struct page *page, int bit_nr)
696{
697 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
698
699 if (!test_bit(bit_nr, &page->flags))
700 return 0;
701
702 return __wait_on_bit(page_waitqueue(page), &wait,
74316201 703 bit_wait_io, TASK_KILLABLE);
f62e00cc
KM
704}
705
385e1ca5
DH
706/**
707 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
697f619f
RD
708 * @page: Page defining the wait queue of interest
709 * @waiter: Waiter to add to the queue
385e1ca5
DH
710 *
711 * Add an arbitrary @waiter to the wait queue for the nominated @page.
712 */
713void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
714{
715 wait_queue_head_t *q = page_waitqueue(page);
716 unsigned long flags;
717
718 spin_lock_irqsave(&q->lock, flags);
719 __add_wait_queue(q, waiter);
720 spin_unlock_irqrestore(&q->lock, flags);
721}
722EXPORT_SYMBOL_GPL(add_page_wait_queue);
723
1da177e4 724/**
485bb99b 725 * unlock_page - unlock a locked page
1da177e4
LT
726 * @page: the page
727 *
728 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
729 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
730 * mechananism between PageLocked pages and PageWriteback pages is shared.
731 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
732 *
8413ac9d
NP
733 * The mb is necessary to enforce ordering between the clear_bit and the read
734 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
1da177e4 735 */
920c7a5d 736void unlock_page(struct page *page)
1da177e4 737{
309381fe 738 VM_BUG_ON_PAGE(!PageLocked(page), page);
8413ac9d 739 clear_bit_unlock(PG_locked, &page->flags);
4e857c58 740 smp_mb__after_atomic();
1da177e4
LT
741 wake_up_page(page, PG_locked);
742}
743EXPORT_SYMBOL(unlock_page);
744
485bb99b
RD
745/**
746 * end_page_writeback - end writeback against a page
747 * @page: the page
1da177e4
LT
748 */
749void end_page_writeback(struct page *page)
750{
888cf2db
MG
751 /*
752 * TestClearPageReclaim could be used here but it is an atomic
753 * operation and overkill in this particular case. Failing to
754 * shuffle a page marked for immediate reclaim is too mild to
755 * justify taking an atomic operation penalty at the end of
756 * ever page writeback.
757 */
758 if (PageReclaim(page)) {
759 ClearPageReclaim(page);
ac6aadb2 760 rotate_reclaimable_page(page);
888cf2db 761 }
ac6aadb2
MS
762
763 if (!test_clear_page_writeback(page))
764 BUG();
765
4e857c58 766 smp_mb__after_atomic();
1da177e4
LT
767 wake_up_page(page, PG_writeback);
768}
769EXPORT_SYMBOL(end_page_writeback);
770
57d99845
MW
771/*
772 * After completing I/O on a page, call this routine to update the page
773 * flags appropriately
774 */
775void page_endio(struct page *page, int rw, int err)
776{
777 if (rw == READ) {
778 if (!err) {
779 SetPageUptodate(page);
780 } else {
781 ClearPageUptodate(page);
782 SetPageError(page);
783 }
784 unlock_page(page);
785 } else { /* rw == WRITE */
786 if (err) {
787 SetPageError(page);
788 if (page->mapping)
789 mapping_set_error(page->mapping, err);
790 }
791 end_page_writeback(page);
792 }
793}
794EXPORT_SYMBOL_GPL(page_endio);
795
485bb99b
RD
796/**
797 * __lock_page - get a lock on the page, assuming we need to sleep to get it
798 * @page: the page to lock
1da177e4 799 */
920c7a5d 800void __lock_page(struct page *page)
1da177e4
LT
801{
802 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
803
74316201 804 __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io,
1da177e4
LT
805 TASK_UNINTERRUPTIBLE);
806}
807EXPORT_SYMBOL(__lock_page);
808
b5606c2d 809int __lock_page_killable(struct page *page)
2687a356
MW
810{
811 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
812
813 return __wait_on_bit_lock(page_waitqueue(page), &wait,
74316201 814 bit_wait_io, TASK_KILLABLE);
2687a356 815}
18bc0bbd 816EXPORT_SYMBOL_GPL(__lock_page_killable);
2687a356 817
9a95f3cf
PC
818/*
819 * Return values:
820 * 1 - page is locked; mmap_sem is still held.
821 * 0 - page is not locked.
822 * mmap_sem has been released (up_read()), unless flags had both
823 * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
824 * which case mmap_sem is still held.
825 *
826 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
827 * with the page locked and the mmap_sem unperturbed.
828 */
d065bd81
ML
829int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
830 unsigned int flags)
831{
37b23e05
KM
832 if (flags & FAULT_FLAG_ALLOW_RETRY) {
833 /*
834 * CAUTION! In this case, mmap_sem is not released
835 * even though return 0.
836 */
837 if (flags & FAULT_FLAG_RETRY_NOWAIT)
838 return 0;
839
840 up_read(&mm->mmap_sem);
841 if (flags & FAULT_FLAG_KILLABLE)
842 wait_on_page_locked_killable(page);
843 else
318b275f 844 wait_on_page_locked(page);
d065bd81 845 return 0;
37b23e05
KM
846 } else {
847 if (flags & FAULT_FLAG_KILLABLE) {
848 int ret;
849
850 ret = __lock_page_killable(page);
851 if (ret) {
852 up_read(&mm->mmap_sem);
853 return 0;
854 }
855 } else
856 __lock_page(page);
857 return 1;
d065bd81
ML
858 }
859}
860
e7b563bb
JW
861/**
862 * page_cache_next_hole - find the next hole (not-present entry)
863 * @mapping: mapping
864 * @index: index
865 * @max_scan: maximum range to search
866 *
867 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
868 * lowest indexed hole.
869 *
870 * Returns: the index of the hole if found, otherwise returns an index
871 * outside of the set specified (in which case 'return - index >=
872 * max_scan' will be true). In rare cases of index wrap-around, 0 will
873 * be returned.
874 *
875 * page_cache_next_hole may be called under rcu_read_lock. However,
876 * like radix_tree_gang_lookup, this will not atomically search a
877 * snapshot of the tree at a single point in time. For example, if a
878 * hole is created at index 5, then subsequently a hole is created at
879 * index 10, page_cache_next_hole covering both indexes may return 10
880 * if called under rcu_read_lock.
881 */
882pgoff_t page_cache_next_hole(struct address_space *mapping,
883 pgoff_t index, unsigned long max_scan)
884{
885 unsigned long i;
886
887 for (i = 0; i < max_scan; i++) {
0cd6144a
JW
888 struct page *page;
889
890 page = radix_tree_lookup(&mapping->page_tree, index);
891 if (!page || radix_tree_exceptional_entry(page))
e7b563bb
JW
892 break;
893 index++;
894 if (index == 0)
895 break;
896 }
897
898 return index;
899}
900EXPORT_SYMBOL(page_cache_next_hole);
901
902/**
903 * page_cache_prev_hole - find the prev hole (not-present entry)
904 * @mapping: mapping
905 * @index: index
906 * @max_scan: maximum range to search
907 *
908 * Search backwards in the range [max(index-max_scan+1, 0), index] for
909 * the first hole.
910 *
911 * Returns: the index of the hole if found, otherwise returns an index
912 * outside of the set specified (in which case 'index - return >=
913 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
914 * will be returned.
915 *
916 * page_cache_prev_hole may be called under rcu_read_lock. However,
917 * like radix_tree_gang_lookup, this will not atomically search a
918 * snapshot of the tree at a single point in time. For example, if a
919 * hole is created at index 10, then subsequently a hole is created at
920 * index 5, page_cache_prev_hole covering both indexes may return 5 if
921 * called under rcu_read_lock.
922 */
923pgoff_t page_cache_prev_hole(struct address_space *mapping,
924 pgoff_t index, unsigned long max_scan)
925{
926 unsigned long i;
927
928 for (i = 0; i < max_scan; i++) {
0cd6144a
JW
929 struct page *page;
930
931 page = radix_tree_lookup(&mapping->page_tree, index);
932 if (!page || radix_tree_exceptional_entry(page))
e7b563bb
JW
933 break;
934 index--;
935 if (index == ULONG_MAX)
936 break;
937 }
938
939 return index;
940}
941EXPORT_SYMBOL(page_cache_prev_hole);
942
485bb99b 943/**
0cd6144a 944 * find_get_entry - find and get a page cache entry
485bb99b 945 * @mapping: the address_space to search
0cd6144a
JW
946 * @offset: the page cache index
947 *
948 * Looks up the page cache slot at @mapping & @offset. If there is a
949 * page cache page, it is returned with an increased refcount.
485bb99b 950 *
139b6a6f
JW
951 * If the slot holds a shadow entry of a previously evicted page, or a
952 * swap entry from shmem/tmpfs, it is returned.
0cd6144a
JW
953 *
954 * Otherwise, %NULL is returned.
1da177e4 955 */
0cd6144a 956struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1da177e4 957{
a60637c8 958 void **pagep;
1da177e4
LT
959 struct page *page;
960
a60637c8
NP
961 rcu_read_lock();
962repeat:
963 page = NULL;
964 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
965 if (pagep) {
966 page = radix_tree_deref_slot(pagep);
27d20fdd
NP
967 if (unlikely(!page))
968 goto out;
a2c16d6c 969 if (radix_tree_exception(page)) {
8079b1c8
HD
970 if (radix_tree_deref_retry(page))
971 goto repeat;
972 /*
139b6a6f
JW
973 * A shadow entry of a recently evicted page,
974 * or a swap entry from shmem/tmpfs. Return
975 * it without attempting to raise page count.
8079b1c8
HD
976 */
977 goto out;
a2c16d6c 978 }
a60637c8
NP
979 if (!page_cache_get_speculative(page))
980 goto repeat;
981
982 /*
983 * Has the page moved?
984 * This is part of the lockless pagecache protocol. See
985 * include/linux/pagemap.h for details.
986 */
987 if (unlikely(page != *pagep)) {
988 page_cache_release(page);
989 goto repeat;
990 }
991 }
27d20fdd 992out:
a60637c8
NP
993 rcu_read_unlock();
994
1da177e4
LT
995 return page;
996}
0cd6144a 997EXPORT_SYMBOL(find_get_entry);
1da177e4 998
0cd6144a
JW
999/**
1000 * find_lock_entry - locate, pin and lock a page cache entry
1001 * @mapping: the address_space to search
1002 * @offset: the page cache index
1003 *
1004 * Looks up the page cache slot at @mapping & @offset. If there is a
1005 * page cache page, it is returned locked and with an increased
1006 * refcount.
1007 *
139b6a6f
JW
1008 * If the slot holds a shadow entry of a previously evicted page, or a
1009 * swap entry from shmem/tmpfs, it is returned.
0cd6144a
JW
1010 *
1011 * Otherwise, %NULL is returned.
1012 *
1013 * find_lock_entry() may sleep.
1014 */
1015struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1da177e4
LT
1016{
1017 struct page *page;
1018
1da177e4 1019repeat:
0cd6144a 1020 page = find_get_entry(mapping, offset);
a2c16d6c 1021 if (page && !radix_tree_exception(page)) {
a60637c8
NP
1022 lock_page(page);
1023 /* Has the page been truncated? */
1024 if (unlikely(page->mapping != mapping)) {
1025 unlock_page(page);
1026 page_cache_release(page);
1027 goto repeat;
1da177e4 1028 }
309381fe 1029 VM_BUG_ON_PAGE(page->index != offset, page);
1da177e4 1030 }
1da177e4
LT
1031 return page;
1032}
0cd6144a
JW
1033EXPORT_SYMBOL(find_lock_entry);
1034
1035/**
2457aec6 1036 * pagecache_get_page - find and get a page reference
0cd6144a
JW
1037 * @mapping: the address_space to search
1038 * @offset: the page index
2457aec6 1039 * @fgp_flags: PCG flags
75325189
RD
1040 * @cache_gfp_mask: gfp mask to use for the page cache data page allocation
1041 * @radix_gfp_mask: gfp mask to use for radix tree node allocation
0cd6144a 1042 *
2457aec6 1043 * Looks up the page cache slot at @mapping & @offset.
1da177e4 1044 *
75325189 1045 * PCG flags modify how the page is returned.
0cd6144a 1046 *
2457aec6
MG
1047 * FGP_ACCESSED: the page will be marked accessed
1048 * FGP_LOCK: Page is return locked
1049 * FGP_CREAT: If page is not present then a new page is allocated using
75325189
RD
1050 * @cache_gfp_mask and added to the page cache and the VM's LRU
1051 * list. If radix tree nodes are allocated during page cache
1052 * insertion then @radix_gfp_mask is used. The page is returned
1053 * locked and with an increased refcount. Otherwise, %NULL is
1054 * returned.
1da177e4 1055 *
2457aec6
MG
1056 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1057 * if the GFP flags specified for FGP_CREAT are atomic.
1da177e4 1058 *
2457aec6 1059 * If there is a page cache page, it is returned with an increased refcount.
1da177e4 1060 */
2457aec6
MG
1061struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1062 int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask)
1da177e4 1063{
eb2be189 1064 struct page *page;
2457aec6 1065
1da177e4 1066repeat:
2457aec6
MG
1067 page = find_get_entry(mapping, offset);
1068 if (radix_tree_exceptional_entry(page))
1069 page = NULL;
1070 if (!page)
1071 goto no_page;
1072
1073 if (fgp_flags & FGP_LOCK) {
1074 if (fgp_flags & FGP_NOWAIT) {
1075 if (!trylock_page(page)) {
1076 page_cache_release(page);
1077 return NULL;
1078 }
1079 } else {
1080 lock_page(page);
1081 }
1082
1083 /* Has the page been truncated? */
1084 if (unlikely(page->mapping != mapping)) {
1085 unlock_page(page);
1086 page_cache_release(page);
1087 goto repeat;
1088 }
1089 VM_BUG_ON_PAGE(page->index != offset, page);
1090 }
1091
1092 if (page && (fgp_flags & FGP_ACCESSED))
1093 mark_page_accessed(page);
1094
1095no_page:
1096 if (!page && (fgp_flags & FGP_CREAT)) {
1097 int err;
1098 if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1099 cache_gfp_mask |= __GFP_WRITE;
1100 if (fgp_flags & FGP_NOFS) {
1101 cache_gfp_mask &= ~__GFP_FS;
1102 radix_gfp_mask &= ~__GFP_FS;
1103 }
1104
1105 page = __page_cache_alloc(cache_gfp_mask);
eb2be189
NP
1106 if (!page)
1107 return NULL;
2457aec6
MG
1108
1109 if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1110 fgp_flags |= FGP_LOCK;
1111
eb39d618 1112 /* Init accessed so avoid atomic mark_page_accessed later */
2457aec6 1113 if (fgp_flags & FGP_ACCESSED)
eb39d618 1114 __SetPageReferenced(page);
2457aec6
MG
1115
1116 err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask);
eb2be189
NP
1117 if (unlikely(err)) {
1118 page_cache_release(page);
1119 page = NULL;
1120 if (err == -EEXIST)
1121 goto repeat;
1da177e4 1122 }
1da177e4 1123 }
2457aec6 1124
1da177e4
LT
1125 return page;
1126}
2457aec6 1127EXPORT_SYMBOL(pagecache_get_page);
1da177e4 1128
0cd6144a
JW
1129/**
1130 * find_get_entries - gang pagecache lookup
1131 * @mapping: The address_space to search
1132 * @start: The starting page cache index
1133 * @nr_entries: The maximum number of entries
1134 * @entries: Where the resulting entries are placed
1135 * @indices: The cache indices corresponding to the entries in @entries
1136 *
1137 * find_get_entries() will search for and return a group of up to
1138 * @nr_entries entries in the mapping. The entries are placed at
1139 * @entries. find_get_entries() takes a reference against any actual
1140 * pages it returns.
1141 *
1142 * The search returns a group of mapping-contiguous page cache entries
1143 * with ascending indexes. There may be holes in the indices due to
1144 * not-present pages.
1145 *
139b6a6f
JW
1146 * Any shadow entries of evicted pages, or swap entries from
1147 * shmem/tmpfs, are included in the returned array.
0cd6144a
JW
1148 *
1149 * find_get_entries() returns the number of pages and shadow entries
1150 * which were found.
1151 */
1152unsigned find_get_entries(struct address_space *mapping,
1153 pgoff_t start, unsigned int nr_entries,
1154 struct page **entries, pgoff_t *indices)
1155{
1156 void **slot;
1157 unsigned int ret = 0;
1158 struct radix_tree_iter iter;
1159
1160 if (!nr_entries)
1161 return 0;
1162
1163 rcu_read_lock();
1164restart:
1165 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1166 struct page *page;
1167repeat:
1168 page = radix_tree_deref_slot(slot);
1169 if (unlikely(!page))
1170 continue;
1171 if (radix_tree_exception(page)) {
1172 if (radix_tree_deref_retry(page))
1173 goto restart;
1174 /*
139b6a6f
JW
1175 * A shadow entry of a recently evicted page,
1176 * or a swap entry from shmem/tmpfs. Return
1177 * it without attempting to raise page count.
0cd6144a
JW
1178 */
1179 goto export;
1180 }
1181 if (!page_cache_get_speculative(page))
1182 goto repeat;
1183
1184 /* Has the page moved? */
1185 if (unlikely(page != *slot)) {
1186 page_cache_release(page);
1187 goto repeat;
1188 }
1189export:
1190 indices[ret] = iter.index;
1191 entries[ret] = page;
1192 if (++ret == nr_entries)
1193 break;
1194 }
1195 rcu_read_unlock();
1196 return ret;
1197}
1198
1da177e4
LT
1199/**
1200 * find_get_pages - gang pagecache lookup
1201 * @mapping: The address_space to search
1202 * @start: The starting page index
1203 * @nr_pages: The maximum number of pages
1204 * @pages: Where the resulting pages are placed
1205 *
1206 * find_get_pages() will search for and return a group of up to
1207 * @nr_pages pages in the mapping. The pages are placed at @pages.
1208 * find_get_pages() takes a reference against the returned pages.
1209 *
1210 * The search returns a group of mapping-contiguous pages with ascending
1211 * indexes. There may be holes in the indices due to not-present pages.
1212 *
1213 * find_get_pages() returns the number of pages which were found.
1214 */
1215unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1216 unsigned int nr_pages, struct page **pages)
1217{
0fc9d104
KK
1218 struct radix_tree_iter iter;
1219 void **slot;
1220 unsigned ret = 0;
1221
1222 if (unlikely(!nr_pages))
1223 return 0;
a60637c8
NP
1224
1225 rcu_read_lock();
1226restart:
0fc9d104 1227 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
a60637c8
NP
1228 struct page *page;
1229repeat:
0fc9d104 1230 page = radix_tree_deref_slot(slot);
a60637c8
NP
1231 if (unlikely(!page))
1232 continue;
9d8aa4ea 1233
a2c16d6c 1234 if (radix_tree_exception(page)) {
8079b1c8
HD
1235 if (radix_tree_deref_retry(page)) {
1236 /*
1237 * Transient condition which can only trigger
1238 * when entry at index 0 moves out of or back
1239 * to root: none yet gotten, safe to restart.
1240 */
0fc9d104 1241 WARN_ON(iter.index);
8079b1c8
HD
1242 goto restart;
1243 }
a2c16d6c 1244 /*
139b6a6f
JW
1245 * A shadow entry of a recently evicted page,
1246 * or a swap entry from shmem/tmpfs. Skip
1247 * over it.
a2c16d6c 1248 */
8079b1c8 1249 continue;
27d20fdd 1250 }
a60637c8
NP
1251
1252 if (!page_cache_get_speculative(page))
1253 goto repeat;
1254
1255 /* Has the page moved? */
0fc9d104 1256 if (unlikely(page != *slot)) {
a60637c8
NP
1257 page_cache_release(page);
1258 goto repeat;
1259 }
1da177e4 1260
a60637c8 1261 pages[ret] = page;
0fc9d104
KK
1262 if (++ret == nr_pages)
1263 break;
a60637c8 1264 }
5b280c0c 1265
a60637c8 1266 rcu_read_unlock();
1da177e4
LT
1267 return ret;
1268}
1269
ebf43500
JA
1270/**
1271 * find_get_pages_contig - gang contiguous pagecache lookup
1272 * @mapping: The address_space to search
1273 * @index: The starting page index
1274 * @nr_pages: The maximum number of pages
1275 * @pages: Where the resulting pages are placed
1276 *
1277 * find_get_pages_contig() works exactly like find_get_pages(), except
1278 * that the returned number of pages are guaranteed to be contiguous.
1279 *
1280 * find_get_pages_contig() returns the number of pages which were found.
1281 */
1282unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1283 unsigned int nr_pages, struct page **pages)
1284{
0fc9d104
KK
1285 struct radix_tree_iter iter;
1286 void **slot;
1287 unsigned int ret = 0;
1288
1289 if (unlikely(!nr_pages))
1290 return 0;
a60637c8
NP
1291
1292 rcu_read_lock();
1293restart:
0fc9d104 1294 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
a60637c8
NP
1295 struct page *page;
1296repeat:
0fc9d104
KK
1297 page = radix_tree_deref_slot(slot);
1298 /* The hole, there no reason to continue */
a60637c8 1299 if (unlikely(!page))
0fc9d104 1300 break;
9d8aa4ea 1301
a2c16d6c 1302 if (radix_tree_exception(page)) {
8079b1c8
HD
1303 if (radix_tree_deref_retry(page)) {
1304 /*
1305 * Transient condition which can only trigger
1306 * when entry at index 0 moves out of or back
1307 * to root: none yet gotten, safe to restart.
1308 */
1309 goto restart;
1310 }
a2c16d6c 1311 /*
139b6a6f
JW
1312 * A shadow entry of a recently evicted page,
1313 * or a swap entry from shmem/tmpfs. Stop
1314 * looking for contiguous pages.
a2c16d6c 1315 */
8079b1c8 1316 break;
a2c16d6c 1317 }
ebf43500 1318
a60637c8
NP
1319 if (!page_cache_get_speculative(page))
1320 goto repeat;
1321
1322 /* Has the page moved? */
0fc9d104 1323 if (unlikely(page != *slot)) {
a60637c8
NP
1324 page_cache_release(page);
1325 goto repeat;
1326 }
1327
9cbb4cb2
NP
1328 /*
1329 * must check mapping and index after taking the ref.
1330 * otherwise we can get both false positives and false
1331 * negatives, which is just confusing to the caller.
1332 */
0fc9d104 1333 if (page->mapping == NULL || page->index != iter.index) {
9cbb4cb2
NP
1334 page_cache_release(page);
1335 break;
1336 }
1337
a60637c8 1338 pages[ret] = page;
0fc9d104
KK
1339 if (++ret == nr_pages)
1340 break;
ebf43500 1341 }
a60637c8
NP
1342 rcu_read_unlock();
1343 return ret;
ebf43500 1344}
ef71c15c 1345EXPORT_SYMBOL(find_get_pages_contig);
ebf43500 1346
485bb99b
RD
1347/**
1348 * find_get_pages_tag - find and return pages that match @tag
1349 * @mapping: the address_space to search
1350 * @index: the starting page index
1351 * @tag: the tag index
1352 * @nr_pages: the maximum number of pages
1353 * @pages: where the resulting pages are placed
1354 *
1da177e4 1355 * Like find_get_pages, except we only return pages which are tagged with
485bb99b 1356 * @tag. We update @index to index the next page for the traversal.
1da177e4
LT
1357 */
1358unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1359 int tag, unsigned int nr_pages, struct page **pages)
1360{
0fc9d104
KK
1361 struct radix_tree_iter iter;
1362 void **slot;
1363 unsigned ret = 0;
1364
1365 if (unlikely(!nr_pages))
1366 return 0;
a60637c8
NP
1367
1368 rcu_read_lock();
1369restart:
0fc9d104
KK
1370 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1371 &iter, *index, tag) {
a60637c8
NP
1372 struct page *page;
1373repeat:
0fc9d104 1374 page = radix_tree_deref_slot(slot);
a60637c8
NP
1375 if (unlikely(!page))
1376 continue;
9d8aa4ea 1377
a2c16d6c 1378 if (radix_tree_exception(page)) {
8079b1c8
HD
1379 if (radix_tree_deref_retry(page)) {
1380 /*
1381 * Transient condition which can only trigger
1382 * when entry at index 0 moves out of or back
1383 * to root: none yet gotten, safe to restart.
1384 */
1385 goto restart;
1386 }
a2c16d6c 1387 /*
139b6a6f
JW
1388 * A shadow entry of a recently evicted page.
1389 *
1390 * Those entries should never be tagged, but
1391 * this tree walk is lockless and the tags are
1392 * looked up in bulk, one radix tree node at a
1393 * time, so there is a sizable window for page
1394 * reclaim to evict a page we saw tagged.
1395 *
1396 * Skip over it.
a2c16d6c 1397 */
139b6a6f 1398 continue;
a2c16d6c 1399 }
a60637c8
NP
1400
1401 if (!page_cache_get_speculative(page))
1402 goto repeat;
1403
1404 /* Has the page moved? */
0fc9d104 1405 if (unlikely(page != *slot)) {
a60637c8
NP
1406 page_cache_release(page);
1407 goto repeat;
1408 }
1409
1410 pages[ret] = page;
0fc9d104
KK
1411 if (++ret == nr_pages)
1412 break;
a60637c8 1413 }
5b280c0c 1414
a60637c8 1415 rcu_read_unlock();
1da177e4 1416
1da177e4
LT
1417 if (ret)
1418 *index = pages[ret - 1]->index + 1;
a60637c8 1419
1da177e4
LT
1420 return ret;
1421}
ef71c15c 1422EXPORT_SYMBOL(find_get_pages_tag);
1da177e4 1423
76d42bd9
WF
1424/*
1425 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1426 * a _large_ part of the i/o request. Imagine the worst scenario:
1427 *
1428 * ---R__________________________________________B__________
1429 * ^ reading here ^ bad block(assume 4k)
1430 *
1431 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1432 * => failing the whole request => read(R) => read(R+1) =>
1433 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1434 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1435 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1436 *
1437 * It is going insane. Fix it by quickly scaling down the readahead size.
1438 */
1439static void shrink_readahead_size_eio(struct file *filp,
1440 struct file_ra_state *ra)
1441{
76d42bd9 1442 ra->ra_pages /= 4;
76d42bd9
WF
1443}
1444
485bb99b 1445/**
36e78914 1446 * do_generic_file_read - generic file read routine
485bb99b
RD
1447 * @filp: the file to read
1448 * @ppos: current file position
6e58e79d
AV
1449 * @iter: data destination
1450 * @written: already copied
485bb99b 1451 *
1da177e4 1452 * This is a generic file read routine, and uses the
485bb99b 1453 * mapping->a_ops->readpage() function for the actual low-level stuff.
1da177e4
LT
1454 *
1455 * This is really ugly. But the goto's actually try to clarify some
1456 * of the logic when it comes to error handling etc.
1da177e4 1457 */
6e58e79d
AV
1458static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1459 struct iov_iter *iter, ssize_t written)
1da177e4 1460{
36e78914 1461 struct address_space *mapping = filp->f_mapping;
1da177e4 1462 struct inode *inode = mapping->host;
36e78914 1463 struct file_ra_state *ra = &filp->f_ra;
57f6b96c
FW
1464 pgoff_t index;
1465 pgoff_t last_index;
1466 pgoff_t prev_index;
1467 unsigned long offset; /* offset into pagecache page */
ec0f1637 1468 unsigned int prev_offset;
6e58e79d 1469 int error = 0;
1da177e4 1470
1da177e4 1471 index = *ppos >> PAGE_CACHE_SHIFT;
7ff81078
FW
1472 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1473 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
6e58e79d 1474 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1da177e4
LT
1475 offset = *ppos & ~PAGE_CACHE_MASK;
1476
1da177e4
LT
1477 for (;;) {
1478 struct page *page;
57f6b96c 1479 pgoff_t end_index;
a32ea1e1 1480 loff_t isize;
1da177e4
LT
1481 unsigned long nr, ret;
1482
1da177e4 1483 cond_resched();
1da177e4
LT
1484find_page:
1485 page = find_get_page(mapping, index);
3ea89ee8 1486 if (!page) {
cf914a7d 1487 page_cache_sync_readahead(mapping,
7ff81078 1488 ra, filp,
3ea89ee8
FW
1489 index, last_index - index);
1490 page = find_get_page(mapping, index);
1491 if (unlikely(page == NULL))
1492 goto no_cached_page;
1493 }
1494 if (PageReadahead(page)) {
cf914a7d 1495 page_cache_async_readahead(mapping,
7ff81078 1496 ra, filp, page,
3ea89ee8 1497 index, last_index - index);
1da177e4 1498 }
8ab22b9a
HH
1499 if (!PageUptodate(page)) {
1500 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1501 !mapping->a_ops->is_partially_uptodate)
1502 goto page_not_up_to_date;
529ae9aa 1503 if (!trylock_page(page))
8ab22b9a 1504 goto page_not_up_to_date;
8d056cb9
DH
1505 /* Did it get truncated before we got the lock? */
1506 if (!page->mapping)
1507 goto page_not_up_to_date_locked;
8ab22b9a 1508 if (!mapping->a_ops->is_partially_uptodate(page,
6e58e79d 1509 offset, iter->count))
8ab22b9a
HH
1510 goto page_not_up_to_date_locked;
1511 unlock_page(page);
1512 }
1da177e4 1513page_ok:
a32ea1e1
N
1514 /*
1515 * i_size must be checked after we know the page is Uptodate.
1516 *
1517 * Checking i_size after the check allows us to calculate
1518 * the correct value for "nr", which means the zero-filled
1519 * part of the page is not copied back to userspace (unless
1520 * another truncate extends the file - this is desired though).
1521 */
1522
1523 isize = i_size_read(inode);
1524 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1525 if (unlikely(!isize || index > end_index)) {
1526 page_cache_release(page);
1527 goto out;
1528 }
1529
1530 /* nr is the maximum number of bytes to copy from this page */
1531 nr = PAGE_CACHE_SIZE;
1532 if (index == end_index) {
1533 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1534 if (nr <= offset) {
1535 page_cache_release(page);
1536 goto out;
1537 }
1538 }
1539 nr = nr - offset;
1da177e4
LT
1540
1541 /* If users can be writing to this page using arbitrary
1542 * virtual addresses, take care about potential aliasing
1543 * before reading the page on the kernel side.
1544 */
1545 if (mapping_writably_mapped(mapping))
1546 flush_dcache_page(page);
1547
1548 /*
ec0f1637
JK
1549 * When a sequential read accesses a page several times,
1550 * only mark it as accessed the first time.
1da177e4 1551 */
ec0f1637 1552 if (prev_index != index || offset != prev_offset)
1da177e4
LT
1553 mark_page_accessed(page);
1554 prev_index = index;
1555
1556 /*
1557 * Ok, we have the page, and it's up-to-date, so
1558 * now we can copy it to user space...
1da177e4 1559 */
6e58e79d
AV
1560
1561 ret = copy_page_to_iter(page, offset, nr, iter);
1da177e4
LT
1562 offset += ret;
1563 index += offset >> PAGE_CACHE_SHIFT;
1564 offset &= ~PAGE_CACHE_MASK;
6ce745ed 1565 prev_offset = offset;
1da177e4
LT
1566
1567 page_cache_release(page);
6e58e79d
AV
1568 written += ret;
1569 if (!iov_iter_count(iter))
1570 goto out;
1571 if (ret < nr) {
1572 error = -EFAULT;
1573 goto out;
1574 }
1575 continue;
1da177e4
LT
1576
1577page_not_up_to_date:
1578 /* Get exclusive access to the page ... */
85462323
ON
1579 error = lock_page_killable(page);
1580 if (unlikely(error))
1581 goto readpage_error;
1da177e4 1582
8ab22b9a 1583page_not_up_to_date_locked:
da6052f7 1584 /* Did it get truncated before we got the lock? */
1da177e4
LT
1585 if (!page->mapping) {
1586 unlock_page(page);
1587 page_cache_release(page);
1588 continue;
1589 }
1590
1591 /* Did somebody else fill it already? */
1592 if (PageUptodate(page)) {
1593 unlock_page(page);
1594 goto page_ok;
1595 }
1596
1597readpage:
91803b49
JM
1598 /*
1599 * A previous I/O error may have been due to temporary
1600 * failures, eg. multipath errors.
1601 * PG_error will be set again if readpage fails.
1602 */
1603 ClearPageError(page);
1da177e4
LT
1604 /* Start the actual read. The read will unlock the page. */
1605 error = mapping->a_ops->readpage(filp, page);
1606
994fc28c
ZB
1607 if (unlikely(error)) {
1608 if (error == AOP_TRUNCATED_PAGE) {
1609 page_cache_release(page);
6e58e79d 1610 error = 0;
994fc28c
ZB
1611 goto find_page;
1612 }
1da177e4 1613 goto readpage_error;
994fc28c 1614 }
1da177e4
LT
1615
1616 if (!PageUptodate(page)) {
85462323
ON
1617 error = lock_page_killable(page);
1618 if (unlikely(error))
1619 goto readpage_error;
1da177e4
LT
1620 if (!PageUptodate(page)) {
1621 if (page->mapping == NULL) {
1622 /*
2ecdc82e 1623 * invalidate_mapping_pages got it
1da177e4
LT
1624 */
1625 unlock_page(page);
1626 page_cache_release(page);
1627 goto find_page;
1628 }
1629 unlock_page(page);
7ff81078 1630 shrink_readahead_size_eio(filp, ra);
85462323
ON
1631 error = -EIO;
1632 goto readpage_error;
1da177e4
LT
1633 }
1634 unlock_page(page);
1635 }
1636
1da177e4
LT
1637 goto page_ok;
1638
1639readpage_error:
1640 /* UHHUH! A synchronous read error occurred. Report it */
1da177e4
LT
1641 page_cache_release(page);
1642 goto out;
1643
1644no_cached_page:
1645 /*
1646 * Ok, it wasn't cached, so we need to create a new
1647 * page..
1648 */
eb2be189
NP
1649 page = page_cache_alloc_cold(mapping);
1650 if (!page) {
6e58e79d 1651 error = -ENOMEM;
eb2be189 1652 goto out;
1da177e4 1653 }
eb2be189 1654 error = add_to_page_cache_lru(page, mapping,
1da177e4
LT
1655 index, GFP_KERNEL);
1656 if (error) {
eb2be189 1657 page_cache_release(page);
6e58e79d
AV
1658 if (error == -EEXIST) {
1659 error = 0;
1da177e4 1660 goto find_page;
6e58e79d 1661 }
1da177e4
LT
1662 goto out;
1663 }
1da177e4
LT
1664 goto readpage;
1665 }
1666
1667out:
7ff81078
FW
1668 ra->prev_pos = prev_index;
1669 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1670 ra->prev_pos |= prev_offset;
1da177e4 1671
f4e6b498 1672 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
0c6aa263 1673 file_accessed(filp);
6e58e79d 1674 return written ? written : error;
1da177e4
LT
1675}
1676
485bb99b 1677/**
6abd2322 1678 * generic_file_read_iter - generic filesystem read routine
485bb99b 1679 * @iocb: kernel I/O control block
6abd2322 1680 * @iter: destination for the data read
485bb99b 1681 *
6abd2322 1682 * This is the "read_iter()" routine for all filesystems
1da177e4
LT
1683 * that can use the page cache directly.
1684 */
1685ssize_t
ed978a81 1686generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
1da177e4 1687{
ed978a81 1688 struct file *file = iocb->ki_filp;
cb66a7a1 1689 ssize_t retval = 0;
543ade1f 1690 loff_t *ppos = &iocb->ki_pos;
ed978a81 1691 loff_t pos = *ppos;
1da177e4
LT
1692
1693 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
ed978a81
AV
1694 if (file->f_flags & O_DIRECT) {
1695 struct address_space *mapping = file->f_mapping;
1696 struct inode *inode = mapping->host;
1697 size_t count = iov_iter_count(iter);
543ade1f 1698 loff_t size;
1da177e4 1699
1da177e4
LT
1700 if (!count)
1701 goto out; /* skip atime */
1702 size = i_size_read(inode);
9fe55eea 1703 retval = filemap_write_and_wait_range(mapping, pos,
a6cbcd4a 1704 pos + count - 1);
9fe55eea 1705 if (!retval) {
ed978a81 1706 struct iov_iter data = *iter;
26978b8b 1707 retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos);
9fe55eea 1708 }
d8d3d94b 1709
9fe55eea
SW
1710 if (retval > 0) {
1711 *ppos = pos + retval;
ed978a81 1712 iov_iter_advance(iter, retval);
9fe55eea 1713 }
66f998f6 1714
9fe55eea
SW
1715 /*
1716 * Btrfs can have a short DIO read if we encounter
1717 * compressed extents, so if there was an error, or if
1718 * we've already read everything we wanted to, or if
1719 * there was a short read because we hit EOF, go ahead
1720 * and return. Otherwise fallthrough to buffered io for
1721 * the rest of the read.
1722 */
ed978a81
AV
1723 if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) {
1724 file_accessed(file);
9fe55eea 1725 goto out;
0e0bcae3 1726 }
1da177e4
LT
1727 }
1728
ed978a81 1729 retval = do_generic_file_read(file, ppos, iter, retval);
1da177e4
LT
1730out:
1731 return retval;
1732}
ed978a81 1733EXPORT_SYMBOL(generic_file_read_iter);
1da177e4 1734
1da177e4 1735#ifdef CONFIG_MMU
485bb99b
RD
1736/**
1737 * page_cache_read - adds requested page to the page cache if not already there
1738 * @file: file to read
1739 * @offset: page index
1740 *
1da177e4
LT
1741 * This adds the requested page to the page cache if it isn't already there,
1742 * and schedules an I/O to read in its contents from disk.
1743 */
920c7a5d 1744static int page_cache_read(struct file *file, pgoff_t offset)
1da177e4
LT
1745{
1746 struct address_space *mapping = file->f_mapping;
1747 struct page *page;
994fc28c 1748 int ret;
1da177e4 1749
994fc28c
ZB
1750 do {
1751 page = page_cache_alloc_cold(mapping);
1752 if (!page)
1753 return -ENOMEM;
1754
1755 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1756 if (ret == 0)
1757 ret = mapping->a_ops->readpage(file, page);
1758 else if (ret == -EEXIST)
1759 ret = 0; /* losing race to add is OK */
1da177e4 1760
1da177e4 1761 page_cache_release(page);
1da177e4 1762
994fc28c
ZB
1763 } while (ret == AOP_TRUNCATED_PAGE);
1764
1765 return ret;
1da177e4
LT
1766}
1767
1768#define MMAP_LOTSAMISS (100)
1769
ef00e08e
LT
1770/*
1771 * Synchronous readahead happens when we don't even find
1772 * a page in the page cache at all.
1773 */
1774static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1775 struct file_ra_state *ra,
1776 struct file *file,
1777 pgoff_t offset)
1778{
1779 unsigned long ra_pages;
1780 struct address_space *mapping = file->f_mapping;
1781
1782 /* If we don't want any read-ahead, don't bother */
64363aad 1783 if (vma->vm_flags & VM_RAND_READ)
ef00e08e 1784 return;
275b12bf
WF
1785 if (!ra->ra_pages)
1786 return;
ef00e08e 1787
64363aad 1788 if (vma->vm_flags & VM_SEQ_READ) {
7ffc59b4
WF
1789 page_cache_sync_readahead(mapping, ra, file, offset,
1790 ra->ra_pages);
ef00e08e
LT
1791 return;
1792 }
1793
207d04ba
AK
1794 /* Avoid banging the cache line if not needed */
1795 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
ef00e08e
LT
1796 ra->mmap_miss++;
1797
1798 /*
1799 * Do we miss much more than hit in this file? If so,
1800 * stop bothering with read-ahead. It will only hurt.
1801 */
1802 if (ra->mmap_miss > MMAP_LOTSAMISS)
1803 return;
1804
d30a1100
WF
1805 /*
1806 * mmap read-around
1807 */
ef00e08e 1808 ra_pages = max_sane_readahead(ra->ra_pages);
275b12bf
WF
1809 ra->start = max_t(long, 0, offset - ra_pages / 2);
1810 ra->size = ra_pages;
2cbea1d3 1811 ra->async_size = ra_pages / 4;
275b12bf 1812 ra_submit(ra, mapping, file);
ef00e08e
LT
1813}
1814
1815/*
1816 * Asynchronous readahead happens when we find the page and PG_readahead,
1817 * so we want to possibly extend the readahead further..
1818 */
1819static void do_async_mmap_readahead(struct vm_area_struct *vma,
1820 struct file_ra_state *ra,
1821 struct file *file,
1822 struct page *page,
1823 pgoff_t offset)
1824{
1825 struct address_space *mapping = file->f_mapping;
1826
1827 /* If we don't want any read-ahead, don't bother */
64363aad 1828 if (vma->vm_flags & VM_RAND_READ)
ef00e08e
LT
1829 return;
1830 if (ra->mmap_miss > 0)
1831 ra->mmap_miss--;
1832 if (PageReadahead(page))
2fad6f5d
WF
1833 page_cache_async_readahead(mapping, ra, file,
1834 page, offset, ra->ra_pages);
ef00e08e
LT
1835}
1836
485bb99b 1837/**
54cb8821 1838 * filemap_fault - read in file data for page fault handling
d0217ac0
NP
1839 * @vma: vma in which the fault was taken
1840 * @vmf: struct vm_fault containing details of the fault
485bb99b 1841 *
54cb8821 1842 * filemap_fault() is invoked via the vma operations vector for a
1da177e4
LT
1843 * mapped memory region to read in file data during a page fault.
1844 *
1845 * The goto's are kind of ugly, but this streamlines the normal case of having
1846 * it in the page cache, and handles the special cases reasonably without
1847 * having a lot of duplicated code.
9a95f3cf
PC
1848 *
1849 * vma->vm_mm->mmap_sem must be held on entry.
1850 *
1851 * If our return value has VM_FAULT_RETRY set, it's because
1852 * lock_page_or_retry() returned 0.
1853 * The mmap_sem has usually been released in this case.
1854 * See __lock_page_or_retry() for the exception.
1855 *
1856 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
1857 * has not been released.
1858 *
1859 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
1da177e4 1860 */
d0217ac0 1861int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1da177e4
LT
1862{
1863 int error;
54cb8821 1864 struct file *file = vma->vm_file;
1da177e4
LT
1865 struct address_space *mapping = file->f_mapping;
1866 struct file_ra_state *ra = &file->f_ra;
1867 struct inode *inode = mapping->host;
ef00e08e 1868 pgoff_t offset = vmf->pgoff;
1da177e4 1869 struct page *page;
99e3e53f 1870 loff_t size;
83c54070 1871 int ret = 0;
1da177e4 1872
99e3e53f
KS
1873 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1874 if (offset >= size >> PAGE_CACHE_SHIFT)
5307cc1a 1875 return VM_FAULT_SIGBUS;
1da177e4 1876
1da177e4 1877 /*
49426420 1878 * Do we have something in the page cache already?
1da177e4 1879 */
ef00e08e 1880 page = find_get_page(mapping, offset);
45cac65b 1881 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1da177e4 1882 /*
ef00e08e
LT
1883 * We found the page, so try async readahead before
1884 * waiting for the lock.
1da177e4 1885 */
ef00e08e 1886 do_async_mmap_readahead(vma, ra, file, page, offset);
45cac65b 1887 } else if (!page) {
ef00e08e
LT
1888 /* No page in the page cache at all */
1889 do_sync_mmap_readahead(vma, ra, file, offset);
1890 count_vm_event(PGMAJFAULT);
456f998e 1891 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
ef00e08e
LT
1892 ret = VM_FAULT_MAJOR;
1893retry_find:
b522c94d 1894 page = find_get_page(mapping, offset);
1da177e4
LT
1895 if (!page)
1896 goto no_cached_page;
1897 }
1898
d88c0922
ML
1899 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1900 page_cache_release(page);
d065bd81 1901 return ret | VM_FAULT_RETRY;
d88c0922 1902 }
b522c94d
ML
1903
1904 /* Did it get truncated? */
1905 if (unlikely(page->mapping != mapping)) {
1906 unlock_page(page);
1907 put_page(page);
1908 goto retry_find;
1909 }
309381fe 1910 VM_BUG_ON_PAGE(page->index != offset, page);
b522c94d 1911
1da177e4 1912 /*
d00806b1
NP
1913 * We have a locked page in the page cache, now we need to check
1914 * that it's up-to-date. If not, it is going to be due to an error.
1da177e4 1915 */
d00806b1 1916 if (unlikely(!PageUptodate(page)))
1da177e4
LT
1917 goto page_not_uptodate;
1918
ef00e08e
LT
1919 /*
1920 * Found the page and have a reference on it.
1921 * We must recheck i_size under page lock.
1922 */
99e3e53f
KS
1923 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1924 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
d00806b1 1925 unlock_page(page);
745ad48e 1926 page_cache_release(page);
5307cc1a 1927 return VM_FAULT_SIGBUS;
d00806b1
NP
1928 }
1929
d0217ac0 1930 vmf->page = page;
83c54070 1931 return ret | VM_FAULT_LOCKED;
1da177e4 1932
1da177e4
LT
1933no_cached_page:
1934 /*
1935 * We're only likely to ever get here if MADV_RANDOM is in
1936 * effect.
1937 */
ef00e08e 1938 error = page_cache_read(file, offset);
1da177e4
LT
1939
1940 /*
1941 * The page we want has now been added to the page cache.
1942 * In the unlikely event that someone removed it in the
1943 * meantime, we'll just come back here and read it again.
1944 */
1945 if (error >= 0)
1946 goto retry_find;
1947
1948 /*
1949 * An error return from page_cache_read can result if the
1950 * system is low on memory, or a problem occurs while trying
1951 * to schedule I/O.
1952 */
1953 if (error == -ENOMEM)
d0217ac0
NP
1954 return VM_FAULT_OOM;
1955 return VM_FAULT_SIGBUS;
1da177e4
LT
1956
1957page_not_uptodate:
1da177e4
LT
1958 /*
1959 * Umm, take care of errors if the page isn't up-to-date.
1960 * Try to re-read it _once_. We do this synchronously,
1961 * because there really aren't any performance issues here
1962 * and we need to check for errors.
1963 */
1da177e4 1964 ClearPageError(page);
994fc28c 1965 error = mapping->a_ops->readpage(file, page);
3ef0f720
MS
1966 if (!error) {
1967 wait_on_page_locked(page);
1968 if (!PageUptodate(page))
1969 error = -EIO;
1970 }
d00806b1
NP
1971 page_cache_release(page);
1972
1973 if (!error || error == AOP_TRUNCATED_PAGE)
994fc28c 1974 goto retry_find;
1da177e4 1975
d00806b1 1976 /* Things didn't work out. Return zero to tell the mm layer so. */
76d42bd9 1977 shrink_readahead_size_eio(file, ra);
d0217ac0 1978 return VM_FAULT_SIGBUS;
54cb8821
NP
1979}
1980EXPORT_SYMBOL(filemap_fault);
1981
f1820361
KS
1982void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
1983{
1984 struct radix_tree_iter iter;
1985 void **slot;
1986 struct file *file = vma->vm_file;
1987 struct address_space *mapping = file->f_mapping;
1988 loff_t size;
1989 struct page *page;
1990 unsigned long address = (unsigned long) vmf->virtual_address;
1991 unsigned long addr;
1992 pte_t *pte;
1993
1994 rcu_read_lock();
1995 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
1996 if (iter.index > vmf->max_pgoff)
1997 break;
1998repeat:
1999 page = radix_tree_deref_slot(slot);
2000 if (unlikely(!page))
2001 goto next;
2002 if (radix_tree_exception(page)) {
2003 if (radix_tree_deref_retry(page))
2004 break;
2005 else
2006 goto next;
2007 }
2008
2009 if (!page_cache_get_speculative(page))
2010 goto repeat;
2011
2012 /* Has the page moved? */
2013 if (unlikely(page != *slot)) {
2014 page_cache_release(page);
2015 goto repeat;
2016 }
2017
2018 if (!PageUptodate(page) ||
2019 PageReadahead(page) ||
2020 PageHWPoison(page))
2021 goto skip;
2022 if (!trylock_page(page))
2023 goto skip;
2024
2025 if (page->mapping != mapping || !PageUptodate(page))
2026 goto unlock;
2027
99e3e53f
KS
2028 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2029 if (page->index >= size >> PAGE_CACHE_SHIFT)
f1820361
KS
2030 goto unlock;
2031
2032 pte = vmf->pte + page->index - vmf->pgoff;
2033 if (!pte_none(*pte))
2034 goto unlock;
2035
2036 if (file->f_ra.mmap_miss > 0)
2037 file->f_ra.mmap_miss--;
2038 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2039 do_set_pte(vma, addr, page, pte, false, false);
2040 unlock_page(page);
2041 goto next;
2042unlock:
2043 unlock_page(page);
2044skip:
2045 page_cache_release(page);
2046next:
2047 if (iter.index == vmf->max_pgoff)
2048 break;
2049 }
2050 rcu_read_unlock();
2051}
2052EXPORT_SYMBOL(filemap_map_pages);
2053
4fcf1c62
JK
2054int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2055{
2056 struct page *page = vmf->page;
496ad9aa 2057 struct inode *inode = file_inode(vma->vm_file);
4fcf1c62
JK
2058 int ret = VM_FAULT_LOCKED;
2059
14da9200 2060 sb_start_pagefault(inode->i_sb);
4fcf1c62
JK
2061 file_update_time(vma->vm_file);
2062 lock_page(page);
2063 if (page->mapping != inode->i_mapping) {
2064 unlock_page(page);
2065 ret = VM_FAULT_NOPAGE;
2066 goto out;
2067 }
14da9200
JK
2068 /*
2069 * We mark the page dirty already here so that when freeze is in
2070 * progress, we are guaranteed that writeback during freezing will
2071 * see the dirty page and writeprotect it again.
2072 */
2073 set_page_dirty(page);
1d1d1a76 2074 wait_for_stable_page(page);
4fcf1c62 2075out:
14da9200 2076 sb_end_pagefault(inode->i_sb);
4fcf1c62
JK
2077 return ret;
2078}
2079EXPORT_SYMBOL(filemap_page_mkwrite);
2080
f0f37e2f 2081const struct vm_operations_struct generic_file_vm_ops = {
54cb8821 2082 .fault = filemap_fault,
f1820361 2083 .map_pages = filemap_map_pages,
4fcf1c62 2084 .page_mkwrite = filemap_page_mkwrite,
0b173bc4 2085 .remap_pages = generic_file_remap_pages,
1da177e4
LT
2086};
2087
2088/* This is used for a general mmap of a disk file */
2089
2090int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2091{
2092 struct address_space *mapping = file->f_mapping;
2093
2094 if (!mapping->a_ops->readpage)
2095 return -ENOEXEC;
2096 file_accessed(file);
2097 vma->vm_ops = &generic_file_vm_ops;
2098 return 0;
2099}
1da177e4
LT
2100
2101/*
2102 * This is for filesystems which do not implement ->writepage.
2103 */
2104int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2105{
2106 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2107 return -EINVAL;
2108 return generic_file_mmap(file, vma);
2109}
2110#else
2111int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2112{
2113 return -ENOSYS;
2114}
2115int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2116{
2117 return -ENOSYS;
2118}
2119#endif /* CONFIG_MMU */
2120
2121EXPORT_SYMBOL(generic_file_mmap);
2122EXPORT_SYMBOL(generic_file_readonly_mmap);
2123
67f9fd91
SL
2124static struct page *wait_on_page_read(struct page *page)
2125{
2126 if (!IS_ERR(page)) {
2127 wait_on_page_locked(page);
2128 if (!PageUptodate(page)) {
2129 page_cache_release(page);
2130 page = ERR_PTR(-EIO);
2131 }
2132 }
2133 return page;
2134}
2135
6fe6900e 2136static struct page *__read_cache_page(struct address_space *mapping,
57f6b96c 2137 pgoff_t index,
5e5358e7 2138 int (*filler)(void *, struct page *),
0531b2aa
LT
2139 void *data,
2140 gfp_t gfp)
1da177e4 2141{
eb2be189 2142 struct page *page;
1da177e4
LT
2143 int err;
2144repeat:
2145 page = find_get_page(mapping, index);
2146 if (!page) {
0531b2aa 2147 page = __page_cache_alloc(gfp | __GFP_COLD);
eb2be189
NP
2148 if (!page)
2149 return ERR_PTR(-ENOMEM);
e6f67b8c 2150 err = add_to_page_cache_lru(page, mapping, index, gfp);
eb2be189
NP
2151 if (unlikely(err)) {
2152 page_cache_release(page);
2153 if (err == -EEXIST)
2154 goto repeat;
1da177e4 2155 /* Presumably ENOMEM for radix tree node */
1da177e4
LT
2156 return ERR_PTR(err);
2157 }
1da177e4
LT
2158 err = filler(data, page);
2159 if (err < 0) {
2160 page_cache_release(page);
2161 page = ERR_PTR(err);
67f9fd91
SL
2162 } else {
2163 page = wait_on_page_read(page);
1da177e4
LT
2164 }
2165 }
1da177e4
LT
2166 return page;
2167}
2168
0531b2aa 2169static struct page *do_read_cache_page(struct address_space *mapping,
57f6b96c 2170 pgoff_t index,
5e5358e7 2171 int (*filler)(void *, struct page *),
0531b2aa
LT
2172 void *data,
2173 gfp_t gfp)
2174
1da177e4
LT
2175{
2176 struct page *page;
2177 int err;
2178
2179retry:
0531b2aa 2180 page = __read_cache_page(mapping, index, filler, data, gfp);
1da177e4 2181 if (IS_ERR(page))
c855ff37 2182 return page;
1da177e4
LT
2183 if (PageUptodate(page))
2184 goto out;
2185
2186 lock_page(page);
2187 if (!page->mapping) {
2188 unlock_page(page);
2189 page_cache_release(page);
2190 goto retry;
2191 }
2192 if (PageUptodate(page)) {
2193 unlock_page(page);
2194 goto out;
2195 }
2196 err = filler(data, page);
2197 if (err < 0) {
2198 page_cache_release(page);
c855ff37 2199 return ERR_PTR(err);
67f9fd91
SL
2200 } else {
2201 page = wait_on_page_read(page);
2202 if (IS_ERR(page))
2203 return page;
1da177e4 2204 }
c855ff37 2205out:
6fe6900e
NP
2206 mark_page_accessed(page);
2207 return page;
2208}
0531b2aa
LT
2209
2210/**
67f9fd91 2211 * read_cache_page - read into page cache, fill it if needed
0531b2aa
LT
2212 * @mapping: the page's address_space
2213 * @index: the page index
2214 * @filler: function to perform the read
5e5358e7 2215 * @data: first arg to filler(data, page) function, often left as NULL
0531b2aa 2216 *
0531b2aa 2217 * Read into the page cache. If a page already exists, and PageUptodate() is
67f9fd91 2218 * not set, try to fill the page and wait for it to become unlocked.
0531b2aa
LT
2219 *
2220 * If the page does not get brought uptodate, return -EIO.
2221 */
67f9fd91 2222struct page *read_cache_page(struct address_space *mapping,
0531b2aa 2223 pgoff_t index,
5e5358e7 2224 int (*filler)(void *, struct page *),
0531b2aa
LT
2225 void *data)
2226{
2227 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2228}
67f9fd91 2229EXPORT_SYMBOL(read_cache_page);
0531b2aa
LT
2230
2231/**
2232 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2233 * @mapping: the page's address_space
2234 * @index: the page index
2235 * @gfp: the page allocator flags to use if allocating
2236 *
2237 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
e6f67b8c 2238 * any new page allocations done using the specified allocation flags.
0531b2aa
LT
2239 *
2240 * If the page does not get brought uptodate, return -EIO.
2241 */
2242struct page *read_cache_page_gfp(struct address_space *mapping,
2243 pgoff_t index,
2244 gfp_t gfp)
2245{
2246 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2247
67f9fd91 2248 return do_read_cache_page(mapping, index, filler, NULL, gfp);
0531b2aa
LT
2249}
2250EXPORT_SYMBOL(read_cache_page_gfp);
2251
1da177e4
LT
2252/*
2253 * Performs necessary checks before doing a write
2254 *
485bb99b 2255 * Can adjust writing position or amount of bytes to write.
1da177e4
LT
2256 * Returns appropriate error code that caller should return or
2257 * zero in case that write should be allowed.
2258 */
2259inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2260{
2261 struct inode *inode = file->f_mapping->host;
59e99e5b 2262 unsigned long limit = rlimit(RLIMIT_FSIZE);
1da177e4
LT
2263
2264 if (unlikely(*pos < 0))
2265 return -EINVAL;
2266
1da177e4
LT
2267 if (!isblk) {
2268 /* FIXME: this is for backwards compatibility with 2.4 */
2269 if (file->f_flags & O_APPEND)
2270 *pos = i_size_read(inode);
2271
2272 if (limit != RLIM_INFINITY) {
2273 if (*pos >= limit) {
2274 send_sig(SIGXFSZ, current, 0);
2275 return -EFBIG;
2276 }
2277 if (*count > limit - (typeof(limit))*pos) {
2278 *count = limit - (typeof(limit))*pos;
2279 }
2280 }
2281 }
2282
2283 /*
2284 * LFS rule
2285 */
2286 if (unlikely(*pos + *count > MAX_NON_LFS &&
2287 !(file->f_flags & O_LARGEFILE))) {
2288 if (*pos >= MAX_NON_LFS) {
1da177e4
LT
2289 return -EFBIG;
2290 }
2291 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2292 *count = MAX_NON_LFS - (unsigned long)*pos;
2293 }
2294 }
2295
2296 /*
2297 * Are we about to exceed the fs block limit ?
2298 *
2299 * If we have written data it becomes a short write. If we have
2300 * exceeded without writing data we send a signal and return EFBIG.
2301 * Linus frestrict idea will clean these up nicely..
2302 */
2303 if (likely(!isblk)) {
2304 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2305 if (*count || *pos > inode->i_sb->s_maxbytes) {
1da177e4
LT
2306 return -EFBIG;
2307 }
2308 /* zero-length writes at ->s_maxbytes are OK */
2309 }
2310
2311 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2312 *count = inode->i_sb->s_maxbytes - *pos;
2313 } else {
9361401e 2314#ifdef CONFIG_BLOCK
1da177e4
LT
2315 loff_t isize;
2316 if (bdev_read_only(I_BDEV(inode)))
2317 return -EPERM;
2318 isize = i_size_read(inode);
2319 if (*pos >= isize) {
2320 if (*count || *pos > isize)
2321 return -ENOSPC;
2322 }
2323
2324 if (*pos + *count > isize)
2325 *count = isize - *pos;
9361401e
DH
2326#else
2327 return -EPERM;
2328#endif
1da177e4
LT
2329 }
2330 return 0;
2331}
2332EXPORT_SYMBOL(generic_write_checks);
2333
afddba49
NP
2334int pagecache_write_begin(struct file *file, struct address_space *mapping,
2335 loff_t pos, unsigned len, unsigned flags,
2336 struct page **pagep, void **fsdata)
2337{
2338 const struct address_space_operations *aops = mapping->a_ops;
2339
4e02ed4b 2340 return aops->write_begin(file, mapping, pos, len, flags,
afddba49 2341 pagep, fsdata);
afddba49
NP
2342}
2343EXPORT_SYMBOL(pagecache_write_begin);
2344
2345int pagecache_write_end(struct file *file, struct address_space *mapping,
2346 loff_t pos, unsigned len, unsigned copied,
2347 struct page *page, void *fsdata)
2348{
2349 const struct address_space_operations *aops = mapping->a_ops;
afddba49 2350
4e02ed4b 2351 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
afddba49
NP
2352}
2353EXPORT_SYMBOL(pagecache_write_end);
2354
1da177e4 2355ssize_t
0c949334 2356generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
1da177e4
LT
2357{
2358 struct file *file = iocb->ki_filp;
2359 struct address_space *mapping = file->f_mapping;
2360 struct inode *inode = mapping->host;
2361 ssize_t written;
a969e903
CH
2362 size_t write_len;
2363 pgoff_t end;
26978b8b 2364 struct iov_iter data;
1da177e4 2365
0c949334 2366 write_len = iov_iter_count(from);
a969e903 2367 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
a969e903 2368
48b47c56 2369 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
a969e903
CH
2370 if (written)
2371 goto out;
2372
2373 /*
2374 * After a write we want buffered reads to be sure to go to disk to get
2375 * the new data. We invalidate clean cached page from the region we're
2376 * about to write. We do this *before* the write so that we can return
6ccfa806 2377 * without clobbering -EIOCBQUEUED from ->direct_IO().
a969e903
CH
2378 */
2379 if (mapping->nrpages) {
2380 written = invalidate_inode_pages2_range(mapping,
2381 pos >> PAGE_CACHE_SHIFT, end);
6ccfa806
HH
2382 /*
2383 * If a page can not be invalidated, return 0 to fall back
2384 * to buffered write.
2385 */
2386 if (written) {
2387 if (written == -EBUSY)
2388 return 0;
a969e903 2389 goto out;
6ccfa806 2390 }
a969e903
CH
2391 }
2392
26978b8b
AV
2393 data = *from;
2394 written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos);
a969e903
CH
2395
2396 /*
2397 * Finally, try again to invalidate clean pages which might have been
2398 * cached by non-direct readahead, or faulted in by get_user_pages()
2399 * if the source of the write was an mmap'ed region of the file
2400 * we're writing. Either one is a pretty crazy thing to do,
2401 * so we don't support it 100%. If this invalidation
2402 * fails, tough, the write still worked...
2403 */
2404 if (mapping->nrpages) {
2405 invalidate_inode_pages2_range(mapping,
2406 pos >> PAGE_CACHE_SHIFT, end);
2407 }
2408
1da177e4 2409 if (written > 0) {
0116651c 2410 pos += written;
f8579f86 2411 iov_iter_advance(from, written);
0116651c
NK
2412 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2413 i_size_write(inode, pos);
1da177e4
LT
2414 mark_inode_dirty(inode);
2415 }
5cb6c6c7 2416 iocb->ki_pos = pos;
1da177e4 2417 }
a969e903 2418out:
1da177e4
LT
2419 return written;
2420}
2421EXPORT_SYMBOL(generic_file_direct_write);
2422
eb2be189
NP
2423/*
2424 * Find or create a page at the given pagecache position. Return the locked
2425 * page. This function is specifically for buffered writes.
2426 */
54566b2c
NP
2427struct page *grab_cache_page_write_begin(struct address_space *mapping,
2428 pgoff_t index, unsigned flags)
eb2be189 2429{
eb2be189 2430 struct page *page;
2457aec6 2431 int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT;
0faa70cb 2432
54566b2c 2433 if (flags & AOP_FLAG_NOFS)
2457aec6
MG
2434 fgp_flags |= FGP_NOFS;
2435
2436 page = pagecache_get_page(mapping, index, fgp_flags,
2437 mapping_gfp_mask(mapping),
2438 GFP_KERNEL);
c585a267 2439 if (page)
2457aec6 2440 wait_for_stable_page(page);
eb2be189 2441
eb2be189
NP
2442 return page;
2443}
54566b2c 2444EXPORT_SYMBOL(grab_cache_page_write_begin);
eb2be189 2445
3b93f911 2446ssize_t generic_perform_write(struct file *file,
afddba49
NP
2447 struct iov_iter *i, loff_t pos)
2448{
2449 struct address_space *mapping = file->f_mapping;
2450 const struct address_space_operations *a_ops = mapping->a_ops;
2451 long status = 0;
2452 ssize_t written = 0;
674b892e
NP
2453 unsigned int flags = 0;
2454
2455 /*
2456 * Copies from kernel address space cannot fail (NFSD is a big user).
2457 */
2458 if (segment_eq(get_fs(), KERNEL_DS))
2459 flags |= AOP_FLAG_UNINTERRUPTIBLE;
afddba49
NP
2460
2461 do {
2462 struct page *page;
afddba49
NP
2463 unsigned long offset; /* Offset into pagecache page */
2464 unsigned long bytes; /* Bytes to write to page */
2465 size_t copied; /* Bytes copied from user */
2466 void *fsdata;
2467
2468 offset = (pos & (PAGE_CACHE_SIZE - 1));
afddba49
NP
2469 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2470 iov_iter_count(i));
2471
2472again:
afddba49
NP
2473 /*
2474 * Bring in the user page that we will copy from _first_.
2475 * Otherwise there's a nasty deadlock on copying from the
2476 * same page as we're writing to, without it being marked
2477 * up-to-date.
2478 *
2479 * Not only is this an optimisation, but it is also required
2480 * to check that the address is actually valid, when atomic
2481 * usercopies are used, below.
2482 */
2483 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2484 status = -EFAULT;
2485 break;
2486 }
2487
674b892e 2488 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
afddba49 2489 &page, &fsdata);
2457aec6 2490 if (unlikely(status < 0))
afddba49
NP
2491 break;
2492
931e80e4 2493 if (mapping_writably_mapped(mapping))
2494 flush_dcache_page(page);
2495
afddba49 2496 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
afddba49
NP
2497 flush_dcache_page(page);
2498
2499 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2500 page, fsdata);
2501 if (unlikely(status < 0))
2502 break;
2503 copied = status;
2504
2505 cond_resched();
2506
124d3b70 2507 iov_iter_advance(i, copied);
afddba49
NP
2508 if (unlikely(copied == 0)) {
2509 /*
2510 * If we were unable to copy any data at all, we must
2511 * fall back to a single segment length write.
2512 *
2513 * If we didn't fallback here, we could livelock
2514 * because not all segments in the iov can be copied at
2515 * once without a pagefault.
2516 */
2517 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2518 iov_iter_single_seg_count(i));
2519 goto again;
2520 }
afddba49
NP
2521 pos += copied;
2522 written += copied;
2523
2524 balance_dirty_pages_ratelimited(mapping);
a50527b1
JK
2525 if (fatal_signal_pending(current)) {
2526 status = -EINTR;
2527 break;
2528 }
afddba49
NP
2529 } while (iov_iter_count(i));
2530
2531 return written ? written : status;
2532}
3b93f911 2533EXPORT_SYMBOL(generic_perform_write);
1da177e4 2534
e4dd9de3 2535/**
8174202b 2536 * __generic_file_write_iter - write data to a file
e4dd9de3 2537 * @iocb: IO state structure (file, offset, etc.)
8174202b 2538 * @from: iov_iter with data to write
e4dd9de3
JK
2539 *
2540 * This function does all the work needed for actually writing data to a
2541 * file. It does all basic checks, removes SUID from the file, updates
2542 * modification times and calls proper subroutines depending on whether we
2543 * do direct IO or a standard buffered write.
2544 *
2545 * It expects i_mutex to be grabbed unless we work on a block device or similar
2546 * object which does not need locking at all.
2547 *
2548 * This function does *not* take care of syncing data in case of O_SYNC write.
2549 * A caller has to handle it. This is mainly due to the fact that we want to
2550 * avoid syncing under i_mutex.
2551 */
8174202b 2552ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1da177e4
LT
2553{
2554 struct file *file = iocb->ki_filp;
fb5527e6 2555 struct address_space * mapping = file->f_mapping;
1da177e4 2556 struct inode *inode = mapping->host;
41fc56d5 2557 loff_t pos = iocb->ki_pos;
3b93f911 2558 ssize_t written = 0;
1da177e4 2559 ssize_t err;
3b93f911 2560 ssize_t status;
8174202b 2561 size_t count = iov_iter_count(from);
1da177e4 2562
1da177e4
LT
2563 /* We can write back this queue in page reclaim */
2564 current->backing_dev_info = mapping->backing_dev_info;
1da177e4
LT
2565 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2566 if (err)
2567 goto out;
2568
2569 if (count == 0)
2570 goto out;
2571
8174202b 2572 iov_iter_truncate(from, count);
0c949334 2573
2f1936b8 2574 err = file_remove_suid(file);
1da177e4
LT
2575 if (err)
2576 goto out;
2577
c3b2da31
JB
2578 err = file_update_time(file);
2579 if (err)
2580 goto out;
1da177e4
LT
2581
2582 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2583 if (unlikely(file->f_flags & O_DIRECT)) {
fb5527e6 2584 loff_t endbyte;
fb5527e6 2585
8174202b 2586 written = generic_file_direct_write(iocb, from, pos);
1da177e4
LT
2587 if (written < 0 || written == count)
2588 goto out;
3b93f911 2589
1da177e4
LT
2590 /*
2591 * direct-io write to a hole: fall through to buffered I/O
2592 * for completing the rest of the request.
2593 */
2594 pos += written;
2595 count -= written;
3b93f911 2596
8174202b 2597 status = generic_perform_write(file, from, pos);
fb5527e6 2598 /*
3b93f911 2599 * If generic_perform_write() returned a synchronous error
fb5527e6
JM
2600 * then we want to return the number of bytes which were
2601 * direct-written, or the error code if that was zero. Note
2602 * that this differs from normal direct-io semantics, which
2603 * will return -EFOO even if some bytes were written.
2604 */
60bb4529 2605 if (unlikely(status < 0)) {
3b93f911 2606 err = status;
fb5527e6
JM
2607 goto out;
2608 }
3b93f911 2609 iocb->ki_pos = pos + status;
fb5527e6
JM
2610 /*
2611 * We need to ensure that the page cache pages are written to
2612 * disk and invalidated to preserve the expected O_DIRECT
2613 * semantics.
2614 */
3b93f911 2615 endbyte = pos + status - 1;
c05c4edd 2616 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
fb5527e6 2617 if (err == 0) {
3b93f911 2618 written += status;
fb5527e6
JM
2619 invalidate_mapping_pages(mapping,
2620 pos >> PAGE_CACHE_SHIFT,
2621 endbyte >> PAGE_CACHE_SHIFT);
2622 } else {
2623 /*
2624 * We don't know how much we wrote, so just return
2625 * the number of bytes which were direct-written
2626 */
2627 }
2628 } else {
8174202b 2629 written = generic_perform_write(file, from, pos);
3b93f911
AV
2630 if (likely(written >= 0))
2631 iocb->ki_pos = pos + written;
fb5527e6 2632 }
1da177e4
LT
2633out:
2634 current->backing_dev_info = NULL;
2635 return written ? written : err;
2636}
8174202b 2637EXPORT_SYMBOL(__generic_file_write_iter);
e4dd9de3 2638
e4dd9de3 2639/**
8174202b 2640 * generic_file_write_iter - write data to a file
e4dd9de3 2641 * @iocb: IO state structure
8174202b 2642 * @from: iov_iter with data to write
e4dd9de3 2643 *
8174202b 2644 * This is a wrapper around __generic_file_write_iter() to be used by most
e4dd9de3
JK
2645 * filesystems. It takes care of syncing the file in case of O_SYNC file
2646 * and acquires i_mutex as needed.
2647 */
8174202b 2648ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1da177e4
LT
2649{
2650 struct file *file = iocb->ki_filp;
148f948b 2651 struct inode *inode = file->f_mapping->host;
1da177e4 2652 ssize_t ret;
1da177e4 2653
1b1dcc1b 2654 mutex_lock(&inode->i_mutex);
8174202b 2655 ret = __generic_file_write_iter(iocb, from);
1b1dcc1b 2656 mutex_unlock(&inode->i_mutex);
1da177e4 2657
02afc27f 2658 if (ret > 0) {
1da177e4
LT
2659 ssize_t err;
2660
d311d79d
AV
2661 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2662 if (err < 0)
1da177e4
LT
2663 ret = err;
2664 }
2665 return ret;
2666}
8174202b 2667EXPORT_SYMBOL(generic_file_write_iter);
1da177e4 2668
cf9a2ae8
DH
2669/**
2670 * try_to_release_page() - release old fs-specific metadata on a page
2671 *
2672 * @page: the page which the kernel is trying to free
2673 * @gfp_mask: memory allocation flags (and I/O mode)
2674 *
2675 * The address_space is to try to release any data against the page
2676 * (presumably at page->private). If the release was successful, return `1'.
2677 * Otherwise return zero.
2678 *
266cf658
DH
2679 * This may also be called if PG_fscache is set on a page, indicating that the
2680 * page is known to the local caching routines.
2681 *
cf9a2ae8 2682 * The @gfp_mask argument specifies whether I/O may be performed to release
3f31fddf 2683 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
cf9a2ae8 2684 *
cf9a2ae8
DH
2685 */
2686int try_to_release_page(struct page *page, gfp_t gfp_mask)
2687{
2688 struct address_space * const mapping = page->mapping;
2689
2690 BUG_ON(!PageLocked(page));
2691 if (PageWriteback(page))
2692 return 0;
2693
2694 if (mapping && mapping->a_ops->releasepage)
2695 return mapping->a_ops->releasepage(page, gfp_mask);
2696 return try_to_free_buffers(page);
2697}
2698
2699EXPORT_SYMBOL(try_to_release_page);