]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - mm/filemap.c
arch,doc: Convert smp_mb__*()
[mirror_ubuntu-artful-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 */
8a9f3ccd 34#include <linux/memcontrol.h>
c515e1fd 35#include <linux/cleancache.h>
f1820361 36#include <linux/rmap.h>
0f8053a5
NP
37#include "internal.h"
38
fe0bfaaf
RJ
39#define CREATE_TRACE_POINTS
40#include <trace/events/filemap.h>
41
1da177e4 42/*
1da177e4
LT
43 * FIXME: remove all knowledge of the buffer layer from the core VM
44 */
148f948b 45#include <linux/buffer_head.h> /* for try_to_free_buffers */
1da177e4 46
1da177e4
LT
47#include <asm/mman.h>
48
49/*
50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
51 * though.
52 *
53 * Shared mappings now work. 15.8.1995 Bruno.
54 *
55 * finished 'unifying' the page and buffer cache and SMP-threaded the
56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
57 *
58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
59 */
60
61/*
62 * Lock ordering:
63 *
3d48ae45 64 * ->i_mmap_mutex (truncate_pagecache)
1da177e4 65 * ->private_lock (__free_pte->__set_page_dirty_buffers)
5d337b91
HD
66 * ->swap_lock (exclusive_swap_page, others)
67 * ->mapping->tree_lock
1da177e4 68 *
1b1dcc1b 69 * ->i_mutex
3d48ae45 70 * ->i_mmap_mutex (truncate->unmap_mapping_range)
1da177e4
LT
71 *
72 * ->mmap_sem
3d48ae45 73 * ->i_mmap_mutex
b8072f09 74 * ->page_table_lock or pte_lock (various, mainly in memory.c)
1da177e4
LT
75 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
76 *
77 * ->mmap_sem
78 * ->lock_page (access_process_vm)
79 *
ccad2365 80 * ->i_mutex (generic_perform_write)
82591e6e 81 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
1da177e4 82 *
f758eeab 83 * bdi->wb.list_lock
a66979ab 84 * sb_lock (fs/fs-writeback.c)
1da177e4
LT
85 * ->mapping->tree_lock (__sync_single_inode)
86 *
3d48ae45 87 * ->i_mmap_mutex
1da177e4
LT
88 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
b8072f09 91 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
1da177e4 92 *
b8072f09 93 * ->page_table_lock or pte_lock
5d337b91 94 * ->swap_lock (try_to_unmap_one)
1da177e4
LT
95 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
053837fc 98 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
1da177e4
LT
99 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
f758eeab 101 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
250df6ed 102 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
f758eeab 103 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
250df6ed 104 * ->inode->i_lock (zap_pte_range->set_page_dirty)
1da177e4
LT
105 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
106 *
9a3c531d
AK
107 * ->i_mmap_mutex
108 * ->tasklist_lock (memory_failure, collect_procs_ao)
1da177e4
LT
109 */
110
91b0abe3
JW
111static void page_cache_tree_delete(struct address_space *mapping,
112 struct page *page, void *shadow)
113{
449dd698
JW
114 struct radix_tree_node *node;
115 unsigned long index;
116 unsigned int offset;
117 unsigned int tag;
118 void **slot;
91b0abe3 119
449dd698
JW
120 VM_BUG_ON(!PageLocked(page));
121
122 __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
123
124 if (shadow) {
91b0abe3
JW
125 mapping->nrshadows++;
126 /*
127 * Make sure the nrshadows update is committed before
128 * the nrpages update so that final truncate racing
129 * with reclaim does not see both counters 0 at the
130 * same time and miss a shadow entry.
131 */
132 smp_wmb();
449dd698 133 }
91b0abe3 134 mapping->nrpages--;
449dd698
JW
135
136 if (!node) {
137 /* Clear direct pointer tags in root node */
138 mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
139 radix_tree_replace_slot(slot, shadow);
140 return;
141 }
142
143 /* Clear tree tags for the removed page */
144 index = page->index;
145 offset = index & RADIX_TREE_MAP_MASK;
146 for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
147 if (test_bit(offset, node->tags[tag]))
148 radix_tree_tag_clear(&mapping->page_tree, index, tag);
149 }
150
151 /* Delete page, swap shadow entry */
152 radix_tree_replace_slot(slot, shadow);
153 workingset_node_pages_dec(node);
154 if (shadow)
155 workingset_node_shadows_inc(node);
156 else
157 if (__radix_tree_delete_node(&mapping->page_tree, node))
158 return;
159
160 /*
161 * Track node that only contains shadow entries.
162 *
163 * Avoid acquiring the list_lru lock if already tracked. The
164 * list_empty() test is safe as node->private_list is
165 * protected by mapping->tree_lock.
166 */
167 if (!workingset_node_pages(node) &&
168 list_empty(&node->private_list)) {
169 node->private_data = mapping;
170 list_lru_add(&workingset_shadow_nodes, &node->private_list);
171 }
91b0abe3
JW
172}
173
1da177e4 174/*
e64a782f 175 * Delete a page from the page cache and free it. Caller has to make
1da177e4 176 * sure the page is locked and that nobody else uses it - or that usage
19fd6231 177 * is safe. The caller must hold the mapping's tree_lock.
1da177e4 178 */
91b0abe3 179void __delete_from_page_cache(struct page *page, void *shadow)
1da177e4
LT
180{
181 struct address_space *mapping = page->mapping;
182
fe0bfaaf 183 trace_mm_filemap_delete_from_page_cache(page);
c515e1fd
DM
184 /*
185 * if we're uptodate, flush out into the cleancache, otherwise
186 * invalidate any existing cleancache entries. We can't leave
187 * stale data around in the cleancache once our page is gone
188 */
189 if (PageUptodate(page) && PageMappedToDisk(page))
190 cleancache_put_page(page);
191 else
3167760f 192 cleancache_invalidate_page(mapping, page);
c515e1fd 193
91b0abe3
JW
194 page_cache_tree_delete(mapping, page, shadow);
195
1da177e4 196 page->mapping = NULL;
b85e0eff 197 /* Leave page->index set: truncation lookup relies upon it */
91b0abe3 198
347ce434 199 __dec_zone_page_state(page, NR_FILE_PAGES);
4b02108a
KM
200 if (PageSwapBacked(page))
201 __dec_zone_page_state(page, NR_SHMEM);
45426812 202 BUG_ON(page_mapped(page));
3a692790
LT
203
204 /*
205 * Some filesystems seem to re-dirty the page even after
206 * the VM has canceled the dirty bit (eg ext3 journaling).
207 *
208 * Fix it up by doing a final dirty accounting check after
209 * having removed the page entirely.
210 */
211 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
212 dec_zone_page_state(page, NR_FILE_DIRTY);
213 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
214 }
1da177e4
LT
215}
216
702cfbf9
MK
217/**
218 * delete_from_page_cache - delete page from page cache
219 * @page: the page which the kernel is trying to remove from page cache
220 *
221 * This must be called only on pages that have been verified to be in the page
222 * cache and locked. It will never put the page into the free list, the caller
223 * has a reference on the page.
224 */
225void delete_from_page_cache(struct page *page)
1da177e4
LT
226{
227 struct address_space *mapping = page->mapping;
6072d13c 228 void (*freepage)(struct page *);
1da177e4 229
cd7619d6 230 BUG_ON(!PageLocked(page));
1da177e4 231
6072d13c 232 freepage = mapping->a_ops->freepage;
19fd6231 233 spin_lock_irq(&mapping->tree_lock);
91b0abe3 234 __delete_from_page_cache(page, NULL);
19fd6231 235 spin_unlock_irq(&mapping->tree_lock);
e767e056 236 mem_cgroup_uncharge_cache_page(page);
6072d13c
LT
237
238 if (freepage)
239 freepage(page);
97cecb5a
MK
240 page_cache_release(page);
241}
242EXPORT_SYMBOL(delete_from_page_cache);
243
7eaceacc 244static int sleep_on_page(void *word)
1da177e4 245{
1da177e4
LT
246 io_schedule();
247 return 0;
248}
249
7eaceacc 250static int sleep_on_page_killable(void *word)
2687a356 251{
7eaceacc 252 sleep_on_page(word);
2687a356
MW
253 return fatal_signal_pending(current) ? -EINTR : 0;
254}
255
865ffef3
DM
256static int filemap_check_errors(struct address_space *mapping)
257{
258 int ret = 0;
259 /* Check for outstanding write errors */
260 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
261 ret = -ENOSPC;
262 if (test_and_clear_bit(AS_EIO, &mapping->flags))
263 ret = -EIO;
264 return ret;
265}
266
1da177e4 267/**
485bb99b 268 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
67be2dd1
MW
269 * @mapping: address space structure to write
270 * @start: offset in bytes where the range starts
469eb4d0 271 * @end: offset in bytes where the range ends (inclusive)
67be2dd1 272 * @sync_mode: enable synchronous operation
1da177e4 273 *
485bb99b
RD
274 * Start writeback against all of a mapping's dirty pages that lie
275 * within the byte offsets <start, end> inclusive.
276 *
1da177e4 277 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
485bb99b 278 * opposed to a regular memory cleansing writeback. The difference between
1da177e4
LT
279 * these two operations is that if a dirty page/buffer is encountered, it must
280 * be waited upon, and not just skipped over.
281 */
ebcf28e1
AM
282int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
283 loff_t end, int sync_mode)
1da177e4
LT
284{
285 int ret;
286 struct writeback_control wbc = {
287 .sync_mode = sync_mode,
05fe478d 288 .nr_to_write = LONG_MAX,
111ebb6e
OH
289 .range_start = start,
290 .range_end = end,
1da177e4
LT
291 };
292
293 if (!mapping_cap_writeback_dirty(mapping))
294 return 0;
295
296 ret = do_writepages(mapping, &wbc);
297 return ret;
298}
299
300static inline int __filemap_fdatawrite(struct address_space *mapping,
301 int sync_mode)
302{
111ebb6e 303 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
1da177e4
LT
304}
305
306int filemap_fdatawrite(struct address_space *mapping)
307{
308 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
309}
310EXPORT_SYMBOL(filemap_fdatawrite);
311
f4c0a0fd 312int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
ebcf28e1 313 loff_t end)
1da177e4
LT
314{
315 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
316}
f4c0a0fd 317EXPORT_SYMBOL(filemap_fdatawrite_range);
1da177e4 318
485bb99b
RD
319/**
320 * filemap_flush - mostly a non-blocking flush
321 * @mapping: target address_space
322 *
1da177e4
LT
323 * This is a mostly non-blocking flush. Not suitable for data-integrity
324 * purposes - I/O may not be started against all dirty pages.
325 */
326int filemap_flush(struct address_space *mapping)
327{
328 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
329}
330EXPORT_SYMBOL(filemap_flush);
331
485bb99b 332/**
94004ed7
CH
333 * filemap_fdatawait_range - wait for writeback to complete
334 * @mapping: address space structure to wait for
335 * @start_byte: offset in bytes where the range starts
336 * @end_byte: offset in bytes where the range ends (inclusive)
485bb99b 337 *
94004ed7
CH
338 * Walk the list of under-writeback pages of the given address space
339 * in the given range and wait for all of them.
1da177e4 340 */
94004ed7
CH
341int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
342 loff_t end_byte)
1da177e4 343{
94004ed7
CH
344 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
345 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
1da177e4
LT
346 struct pagevec pvec;
347 int nr_pages;
865ffef3 348 int ret2, ret = 0;
1da177e4 349
94004ed7 350 if (end_byte < start_byte)
865ffef3 351 goto out;
1da177e4
LT
352
353 pagevec_init(&pvec, 0);
1da177e4
LT
354 while ((index <= end) &&
355 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
356 PAGECACHE_TAG_WRITEBACK,
357 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
358 unsigned i;
359
360 for (i = 0; i < nr_pages; i++) {
361 struct page *page = pvec.pages[i];
362
363 /* until radix tree lookup accepts end_index */
364 if (page->index > end)
365 continue;
366
367 wait_on_page_writeback(page);
212260aa 368 if (TestClearPageError(page))
1da177e4
LT
369 ret = -EIO;
370 }
371 pagevec_release(&pvec);
372 cond_resched();
373 }
865ffef3
DM
374out:
375 ret2 = filemap_check_errors(mapping);
376 if (!ret)
377 ret = ret2;
1da177e4
LT
378
379 return ret;
380}
d3bccb6f
JK
381EXPORT_SYMBOL(filemap_fdatawait_range);
382
1da177e4 383/**
485bb99b 384 * filemap_fdatawait - wait for all under-writeback pages to complete
1da177e4 385 * @mapping: address space structure to wait for
485bb99b
RD
386 *
387 * Walk the list of under-writeback pages of the given address space
388 * and wait for all of them.
1da177e4
LT
389 */
390int filemap_fdatawait(struct address_space *mapping)
391{
392 loff_t i_size = i_size_read(mapping->host);
393
394 if (i_size == 0)
395 return 0;
396
94004ed7 397 return filemap_fdatawait_range(mapping, 0, i_size - 1);
1da177e4
LT
398}
399EXPORT_SYMBOL(filemap_fdatawait);
400
401int filemap_write_and_wait(struct address_space *mapping)
402{
28fd1298 403 int err = 0;
1da177e4
LT
404
405 if (mapping->nrpages) {
28fd1298
OH
406 err = filemap_fdatawrite(mapping);
407 /*
408 * Even if the above returned error, the pages may be
409 * written partially (e.g. -ENOSPC), so we wait for it.
410 * But the -EIO is special case, it may indicate the worst
411 * thing (e.g. bug) happened, so we avoid waiting for it.
412 */
413 if (err != -EIO) {
414 int err2 = filemap_fdatawait(mapping);
415 if (!err)
416 err = err2;
417 }
865ffef3
DM
418 } else {
419 err = filemap_check_errors(mapping);
1da177e4 420 }
28fd1298 421 return err;
1da177e4 422}
28fd1298 423EXPORT_SYMBOL(filemap_write_and_wait);
1da177e4 424
485bb99b
RD
425/**
426 * filemap_write_and_wait_range - write out & wait on a file range
427 * @mapping: the address_space for the pages
428 * @lstart: offset in bytes where the range starts
429 * @lend: offset in bytes where the range ends (inclusive)
430 *
469eb4d0
AM
431 * Write out and wait upon file offsets lstart->lend, inclusive.
432 *
433 * Note that `lend' is inclusive (describes the last byte to be written) so
434 * that this function can be used to write to the very end-of-file (end = -1).
435 */
1da177e4
LT
436int filemap_write_and_wait_range(struct address_space *mapping,
437 loff_t lstart, loff_t lend)
438{
28fd1298 439 int err = 0;
1da177e4
LT
440
441 if (mapping->nrpages) {
28fd1298
OH
442 err = __filemap_fdatawrite_range(mapping, lstart, lend,
443 WB_SYNC_ALL);
444 /* See comment of filemap_write_and_wait() */
445 if (err != -EIO) {
94004ed7
CH
446 int err2 = filemap_fdatawait_range(mapping,
447 lstart, lend);
28fd1298
OH
448 if (!err)
449 err = err2;
450 }
865ffef3
DM
451 } else {
452 err = filemap_check_errors(mapping);
1da177e4 453 }
28fd1298 454 return err;
1da177e4 455}
f6995585 456EXPORT_SYMBOL(filemap_write_and_wait_range);
1da177e4 457
ef6a3c63
MS
458/**
459 * replace_page_cache_page - replace a pagecache page with a new one
460 * @old: page to be replaced
461 * @new: page to replace with
462 * @gfp_mask: allocation mode
463 *
464 * This function replaces a page in the pagecache with a new one. On
465 * success it acquires the pagecache reference for the new page and
466 * drops it for the old page. Both the old and new pages must be
467 * locked. This function does not add the new page to the LRU, the
468 * caller must do that.
469 *
470 * The remove + add is atomic. The only way this function can fail is
471 * memory allocation failure.
472 */
473int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
474{
475 int error;
ef6a3c63 476
309381fe
SL
477 VM_BUG_ON_PAGE(!PageLocked(old), old);
478 VM_BUG_ON_PAGE(!PageLocked(new), new);
479 VM_BUG_ON_PAGE(new->mapping, new);
ef6a3c63 480
ef6a3c63
MS
481 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
482 if (!error) {
483 struct address_space *mapping = old->mapping;
484 void (*freepage)(struct page *);
485
486 pgoff_t offset = old->index;
487 freepage = mapping->a_ops->freepage;
488
489 page_cache_get(new);
490 new->mapping = mapping;
491 new->index = offset;
492
493 spin_lock_irq(&mapping->tree_lock);
91b0abe3 494 __delete_from_page_cache(old, NULL);
ef6a3c63
MS
495 error = radix_tree_insert(&mapping->page_tree, offset, new);
496 BUG_ON(error);
497 mapping->nrpages++;
498 __inc_zone_page_state(new, NR_FILE_PAGES);
499 if (PageSwapBacked(new))
500 __inc_zone_page_state(new, NR_SHMEM);
501 spin_unlock_irq(&mapping->tree_lock);
ab936cbc
KH
502 /* mem_cgroup codes must not be called under tree_lock */
503 mem_cgroup_replace_page_cache(old, new);
ef6a3c63
MS
504 radix_tree_preload_end();
505 if (freepage)
506 freepage(old);
507 page_cache_release(old);
ef6a3c63
MS
508 }
509
510 return error;
511}
512EXPORT_SYMBOL_GPL(replace_page_cache_page);
513
0cd6144a 514static int page_cache_tree_insert(struct address_space *mapping,
a528910e 515 struct page *page, void **shadowp)
0cd6144a 516{
449dd698 517 struct radix_tree_node *node;
0cd6144a
JW
518 void **slot;
519 int error;
520
449dd698
JW
521 error = __radix_tree_create(&mapping->page_tree, page->index,
522 &node, &slot);
523 if (error)
524 return error;
525 if (*slot) {
0cd6144a
JW
526 void *p;
527
528 p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
529 if (!radix_tree_exceptional_entry(p))
530 return -EEXIST;
a528910e
JW
531 if (shadowp)
532 *shadowp = p;
449dd698
JW
533 mapping->nrshadows--;
534 if (node)
535 workingset_node_shadows_dec(node);
0cd6144a 536 }
449dd698
JW
537 radix_tree_replace_slot(slot, page);
538 mapping->nrpages++;
539 if (node) {
540 workingset_node_pages_inc(node);
541 /*
542 * Don't track node that contains actual pages.
543 *
544 * Avoid acquiring the list_lru lock if already
545 * untracked. The list_empty() test is safe as
546 * node->private_list is protected by
547 * mapping->tree_lock.
548 */
549 if (!list_empty(&node->private_list))
550 list_lru_del(&workingset_shadow_nodes,
551 &node->private_list);
552 }
553 return 0;
0cd6144a
JW
554}
555
a528910e
JW
556static int __add_to_page_cache_locked(struct page *page,
557 struct address_space *mapping,
558 pgoff_t offset, gfp_t gfp_mask,
559 void **shadowp)
1da177e4 560{
e286781d
NP
561 int error;
562
309381fe
SL
563 VM_BUG_ON_PAGE(!PageLocked(page), page);
564 VM_BUG_ON_PAGE(PageSwapBacked(page), page);
e286781d 565
d715ae08 566 error = mem_cgroup_charge_file(page, current->mm,
2c26fdd7 567 gfp_mask & GFP_RECLAIM_MASK);
35c754d7 568 if (error)
66a0c8ee 569 return error;
1da177e4 570
5e4c0d97 571 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
66a0c8ee 572 if (error) {
69029cd5 573 mem_cgroup_uncharge_cache_page(page);
66a0c8ee
KS
574 return error;
575 }
576
577 page_cache_get(page);
578 page->mapping = mapping;
579 page->index = offset;
580
581 spin_lock_irq(&mapping->tree_lock);
a528910e 582 error = page_cache_tree_insert(mapping, page, shadowp);
66a0c8ee
KS
583 radix_tree_preload_end();
584 if (unlikely(error))
585 goto err_insert;
66a0c8ee
KS
586 __inc_zone_page_state(page, NR_FILE_PAGES);
587 spin_unlock_irq(&mapping->tree_lock);
588 trace_mm_filemap_add_to_page_cache(page);
589 return 0;
590err_insert:
591 page->mapping = NULL;
592 /* Leave page->index set: truncation relies upon it */
593 spin_unlock_irq(&mapping->tree_lock);
594 mem_cgroup_uncharge_cache_page(page);
595 page_cache_release(page);
1da177e4
LT
596 return error;
597}
a528910e
JW
598
599/**
600 * add_to_page_cache_locked - add a locked page to the pagecache
601 * @page: page to add
602 * @mapping: the page's address_space
603 * @offset: page index
604 * @gfp_mask: page allocation mode
605 *
606 * This function is used to add a page to the pagecache. It must be locked.
607 * This function does not add the page to the LRU. The caller must do that.
608 */
609int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
610 pgoff_t offset, gfp_t gfp_mask)
611{
612 return __add_to_page_cache_locked(page, mapping, offset,
613 gfp_mask, NULL);
614}
e286781d 615EXPORT_SYMBOL(add_to_page_cache_locked);
1da177e4
LT
616
617int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
6daa0e28 618 pgoff_t offset, gfp_t gfp_mask)
1da177e4 619{
a528910e 620 void *shadow = NULL;
4f98a2fe
RR
621 int ret;
622
a528910e
JW
623 __set_page_locked(page);
624 ret = __add_to_page_cache_locked(page, mapping, offset,
625 gfp_mask, &shadow);
626 if (unlikely(ret))
627 __clear_page_locked(page);
628 else {
629 /*
630 * The page might have been evicted from cache only
631 * recently, in which case it should be activated like
632 * any other repeatedly accessed page.
633 */
634 if (shadow && workingset_refault(shadow)) {
635 SetPageActive(page);
636 workingset_activation(page);
637 } else
638 ClearPageActive(page);
639 lru_cache_add(page);
640 }
1da177e4
LT
641 return ret;
642}
18bc0bbd 643EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
1da177e4 644
44110fe3 645#ifdef CONFIG_NUMA
2ae88149 646struct page *__page_cache_alloc(gfp_t gfp)
44110fe3 647{
c0ff7453
MX
648 int n;
649 struct page *page;
650
44110fe3 651 if (cpuset_do_page_mem_spread()) {
cc9a6c87
MG
652 unsigned int cpuset_mems_cookie;
653 do {
d26914d1 654 cpuset_mems_cookie = read_mems_allowed_begin();
cc9a6c87
MG
655 n = cpuset_mem_spread_node();
656 page = alloc_pages_exact_node(n, gfp, 0);
d26914d1 657 } while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
cc9a6c87 658
c0ff7453 659 return page;
44110fe3 660 }
2ae88149 661 return alloc_pages(gfp, 0);
44110fe3 662}
2ae88149 663EXPORT_SYMBOL(__page_cache_alloc);
44110fe3
PJ
664#endif
665
1da177e4
LT
666/*
667 * In order to wait for pages to become available there must be
668 * waitqueues associated with pages. By using a hash table of
669 * waitqueues where the bucket discipline is to maintain all
670 * waiters on the same queue and wake all when any of the pages
671 * become available, and for the woken contexts to check to be
672 * sure the appropriate page became available, this saves space
673 * at a cost of "thundering herd" phenomena during rare hash
674 * collisions.
675 */
676static wait_queue_head_t *page_waitqueue(struct page *page)
677{
678 const struct zone *zone = page_zone(page);
679
680 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
681}
682
683static inline void wake_up_page(struct page *page, int bit)
684{
685 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
686}
687
920c7a5d 688void wait_on_page_bit(struct page *page, int bit_nr)
1da177e4
LT
689{
690 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
691
692 if (test_bit(bit_nr, &page->flags))
7eaceacc 693 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
1da177e4
LT
694 TASK_UNINTERRUPTIBLE);
695}
696EXPORT_SYMBOL(wait_on_page_bit);
697
f62e00cc
KM
698int wait_on_page_bit_killable(struct page *page, int bit_nr)
699{
700 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
701
702 if (!test_bit(bit_nr, &page->flags))
703 return 0;
704
705 return __wait_on_bit(page_waitqueue(page), &wait,
706 sleep_on_page_killable, TASK_KILLABLE);
707}
708
385e1ca5
DH
709/**
710 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
697f619f
RD
711 * @page: Page defining the wait queue of interest
712 * @waiter: Waiter to add to the queue
385e1ca5
DH
713 *
714 * Add an arbitrary @waiter to the wait queue for the nominated @page.
715 */
716void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
717{
718 wait_queue_head_t *q = page_waitqueue(page);
719 unsigned long flags;
720
721 spin_lock_irqsave(&q->lock, flags);
722 __add_wait_queue(q, waiter);
723 spin_unlock_irqrestore(&q->lock, flags);
724}
725EXPORT_SYMBOL_GPL(add_page_wait_queue);
726
1da177e4 727/**
485bb99b 728 * unlock_page - unlock a locked page
1da177e4
LT
729 * @page: the page
730 *
731 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
732 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
733 * mechananism between PageLocked pages and PageWriteback pages is shared.
734 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
735 *
8413ac9d
NP
736 * The mb is necessary to enforce ordering between the clear_bit and the read
737 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
1da177e4 738 */
920c7a5d 739void unlock_page(struct page *page)
1da177e4 740{
309381fe 741 VM_BUG_ON_PAGE(!PageLocked(page), page);
8413ac9d
NP
742 clear_bit_unlock(PG_locked, &page->flags);
743 smp_mb__after_clear_bit();
1da177e4
LT
744 wake_up_page(page, PG_locked);
745}
746EXPORT_SYMBOL(unlock_page);
747
485bb99b
RD
748/**
749 * end_page_writeback - end writeback against a page
750 * @page: the page
1da177e4
LT
751 */
752void end_page_writeback(struct page *page)
753{
ac6aadb2
MS
754 if (TestClearPageReclaim(page))
755 rotate_reclaimable_page(page);
756
757 if (!test_clear_page_writeback(page))
758 BUG();
759
1da177e4
LT
760 smp_mb__after_clear_bit();
761 wake_up_page(page, PG_writeback);
762}
763EXPORT_SYMBOL(end_page_writeback);
764
485bb99b
RD
765/**
766 * __lock_page - get a lock on the page, assuming we need to sleep to get it
767 * @page: the page to lock
1da177e4 768 */
920c7a5d 769void __lock_page(struct page *page)
1da177e4
LT
770{
771 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
772
7eaceacc 773 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
1da177e4
LT
774 TASK_UNINTERRUPTIBLE);
775}
776EXPORT_SYMBOL(__lock_page);
777
b5606c2d 778int __lock_page_killable(struct page *page)
2687a356
MW
779{
780 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
781
782 return __wait_on_bit_lock(page_waitqueue(page), &wait,
7eaceacc 783 sleep_on_page_killable, TASK_KILLABLE);
2687a356 784}
18bc0bbd 785EXPORT_SYMBOL_GPL(__lock_page_killable);
2687a356 786
d065bd81
ML
787int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
788 unsigned int flags)
789{
37b23e05
KM
790 if (flags & FAULT_FLAG_ALLOW_RETRY) {
791 /*
792 * CAUTION! In this case, mmap_sem is not released
793 * even though return 0.
794 */
795 if (flags & FAULT_FLAG_RETRY_NOWAIT)
796 return 0;
797
798 up_read(&mm->mmap_sem);
799 if (flags & FAULT_FLAG_KILLABLE)
800 wait_on_page_locked_killable(page);
801 else
318b275f 802 wait_on_page_locked(page);
d065bd81 803 return 0;
37b23e05
KM
804 } else {
805 if (flags & FAULT_FLAG_KILLABLE) {
806 int ret;
807
808 ret = __lock_page_killable(page);
809 if (ret) {
810 up_read(&mm->mmap_sem);
811 return 0;
812 }
813 } else
814 __lock_page(page);
815 return 1;
d065bd81
ML
816 }
817}
818
e7b563bb
JW
819/**
820 * page_cache_next_hole - find the next hole (not-present entry)
821 * @mapping: mapping
822 * @index: index
823 * @max_scan: maximum range to search
824 *
825 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
826 * lowest indexed hole.
827 *
828 * Returns: the index of the hole if found, otherwise returns an index
829 * outside of the set specified (in which case 'return - index >=
830 * max_scan' will be true). In rare cases of index wrap-around, 0 will
831 * be returned.
832 *
833 * page_cache_next_hole may be called under rcu_read_lock. However,
834 * like radix_tree_gang_lookup, this will not atomically search a
835 * snapshot of the tree at a single point in time. For example, if a
836 * hole is created at index 5, then subsequently a hole is created at
837 * index 10, page_cache_next_hole covering both indexes may return 10
838 * if called under rcu_read_lock.
839 */
840pgoff_t page_cache_next_hole(struct address_space *mapping,
841 pgoff_t index, unsigned long max_scan)
842{
843 unsigned long i;
844
845 for (i = 0; i < max_scan; i++) {
0cd6144a
JW
846 struct page *page;
847
848 page = radix_tree_lookup(&mapping->page_tree, index);
849 if (!page || radix_tree_exceptional_entry(page))
e7b563bb
JW
850 break;
851 index++;
852 if (index == 0)
853 break;
854 }
855
856 return index;
857}
858EXPORT_SYMBOL(page_cache_next_hole);
859
860/**
861 * page_cache_prev_hole - find the prev hole (not-present entry)
862 * @mapping: mapping
863 * @index: index
864 * @max_scan: maximum range to search
865 *
866 * Search backwards in the range [max(index-max_scan+1, 0), index] for
867 * the first hole.
868 *
869 * Returns: the index of the hole if found, otherwise returns an index
870 * outside of the set specified (in which case 'index - return >=
871 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
872 * will be returned.
873 *
874 * page_cache_prev_hole may be called under rcu_read_lock. However,
875 * like radix_tree_gang_lookup, this will not atomically search a
876 * snapshot of the tree at a single point in time. For example, if a
877 * hole is created at index 10, then subsequently a hole is created at
878 * index 5, page_cache_prev_hole covering both indexes may return 5 if
879 * called under rcu_read_lock.
880 */
881pgoff_t page_cache_prev_hole(struct address_space *mapping,
882 pgoff_t index, unsigned long max_scan)
883{
884 unsigned long i;
885
886 for (i = 0; i < max_scan; i++) {
0cd6144a
JW
887 struct page *page;
888
889 page = radix_tree_lookup(&mapping->page_tree, index);
890 if (!page || radix_tree_exceptional_entry(page))
e7b563bb
JW
891 break;
892 index--;
893 if (index == ULONG_MAX)
894 break;
895 }
896
897 return index;
898}
899EXPORT_SYMBOL(page_cache_prev_hole);
900
485bb99b 901/**
0cd6144a 902 * find_get_entry - find and get a page cache entry
485bb99b 903 * @mapping: the address_space to search
0cd6144a
JW
904 * @offset: the page cache index
905 *
906 * Looks up the page cache slot at @mapping & @offset. If there is a
907 * page cache page, it is returned with an increased refcount.
485bb99b 908 *
0cd6144a
JW
909 * If the slot holds a shadow entry of a previously evicted page, it
910 * is returned.
911 *
912 * Otherwise, %NULL is returned.
1da177e4 913 */
0cd6144a 914struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1da177e4 915{
a60637c8 916 void **pagep;
1da177e4
LT
917 struct page *page;
918
a60637c8
NP
919 rcu_read_lock();
920repeat:
921 page = NULL;
922 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
923 if (pagep) {
924 page = radix_tree_deref_slot(pagep);
27d20fdd
NP
925 if (unlikely(!page))
926 goto out;
a2c16d6c 927 if (radix_tree_exception(page)) {
8079b1c8
HD
928 if (radix_tree_deref_retry(page))
929 goto repeat;
930 /*
931 * Otherwise, shmem/tmpfs must be storing a swap entry
932 * here as an exceptional entry: so return it without
933 * attempting to raise page count.
934 */
935 goto out;
a2c16d6c 936 }
a60637c8
NP
937 if (!page_cache_get_speculative(page))
938 goto repeat;
939
940 /*
941 * Has the page moved?
942 * This is part of the lockless pagecache protocol. See
943 * include/linux/pagemap.h for details.
944 */
945 if (unlikely(page != *pagep)) {
946 page_cache_release(page);
947 goto repeat;
948 }
949 }
27d20fdd 950out:
a60637c8
NP
951 rcu_read_unlock();
952
1da177e4
LT
953 return page;
954}
0cd6144a 955EXPORT_SYMBOL(find_get_entry);
1da177e4 956
1da177e4 957/**
0cd6144a 958 * find_get_page - find and get a page reference
67be2dd1
MW
959 * @mapping: the address_space to search
960 * @offset: the page index
1da177e4 961 *
0cd6144a
JW
962 * Looks up the page cache slot at @mapping & @offset. If there is a
963 * page cache page, it is returned with an increased refcount.
1da177e4 964 *
0cd6144a 965 * Otherwise, %NULL is returned.
1da177e4 966 */
0cd6144a
JW
967struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
968{
969 struct page *page = find_get_entry(mapping, offset);
970
971 if (radix_tree_exceptional_entry(page))
972 page = NULL;
973 return page;
974}
975EXPORT_SYMBOL(find_get_page);
976
977/**
978 * find_lock_entry - locate, pin and lock a page cache entry
979 * @mapping: the address_space to search
980 * @offset: the page cache index
981 *
982 * Looks up the page cache slot at @mapping & @offset. If there is a
983 * page cache page, it is returned locked and with an increased
984 * refcount.
985 *
986 * If the slot holds a shadow entry of a previously evicted page, it
987 * is returned.
988 *
989 * Otherwise, %NULL is returned.
990 *
991 * find_lock_entry() may sleep.
992 */
993struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1da177e4
LT
994{
995 struct page *page;
996
1da177e4 997repeat:
0cd6144a 998 page = find_get_entry(mapping, offset);
a2c16d6c 999 if (page && !radix_tree_exception(page)) {
a60637c8
NP
1000 lock_page(page);
1001 /* Has the page been truncated? */
1002 if (unlikely(page->mapping != mapping)) {
1003 unlock_page(page);
1004 page_cache_release(page);
1005 goto repeat;
1da177e4 1006 }
309381fe 1007 VM_BUG_ON_PAGE(page->index != offset, page);
1da177e4 1008 }
1da177e4
LT
1009 return page;
1010}
0cd6144a
JW
1011EXPORT_SYMBOL(find_lock_entry);
1012
1013/**
1014 * find_lock_page - locate, pin and lock a pagecache page
1015 * @mapping: the address_space to search
1016 * @offset: the page index
1017 *
1018 * Looks up the page cache slot at @mapping & @offset. If there is a
1019 * page cache page, it is returned locked and with an increased
1020 * refcount.
1021 *
1022 * Otherwise, %NULL is returned.
1023 *
1024 * find_lock_page() may sleep.
1025 */
1026struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
1027{
1028 struct page *page = find_lock_entry(mapping, offset);
1029
1030 if (radix_tree_exceptional_entry(page))
1031 page = NULL;
1032 return page;
1033}
1da177e4
LT
1034EXPORT_SYMBOL(find_lock_page);
1035
1036/**
1037 * find_or_create_page - locate or add a pagecache page
67be2dd1
MW
1038 * @mapping: the page's address_space
1039 * @index: the page's index into the mapping
1040 * @gfp_mask: page allocation mode
1da177e4 1041 *
0cd6144a
JW
1042 * Looks up the page cache slot at @mapping & @offset. If there is a
1043 * page cache page, it is returned locked and with an increased
1044 * refcount.
1045 *
1046 * If the page is not present, a new page is allocated using @gfp_mask
1047 * and added to the page cache and the VM's LRU list. The page is
1048 * returned locked and with an increased refcount.
1da177e4 1049 *
0cd6144a 1050 * On memory exhaustion, %NULL is returned.
1da177e4 1051 *
0cd6144a
JW
1052 * find_or_create_page() may sleep, even if @gfp_flags specifies an
1053 * atomic allocation!
1da177e4
LT
1054 */
1055struct page *find_or_create_page(struct address_space *mapping,
57f6b96c 1056 pgoff_t index, gfp_t gfp_mask)
1da177e4 1057{
eb2be189 1058 struct page *page;
1da177e4
LT
1059 int err;
1060repeat:
1061 page = find_lock_page(mapping, index);
1062 if (!page) {
eb2be189
NP
1063 page = __page_cache_alloc(gfp_mask);
1064 if (!page)
1065 return NULL;
67d58ac4
NP
1066 /*
1067 * We want a regular kernel memory (not highmem or DMA etc)
1068 * allocation for the radix tree nodes, but we need to honour
1069 * the context-specific requirements the caller has asked for.
1070 * GFP_RECLAIM_MASK collects those requirements.
1071 */
1072 err = add_to_page_cache_lru(page, mapping, index,
1073 (gfp_mask & GFP_RECLAIM_MASK));
eb2be189
NP
1074 if (unlikely(err)) {
1075 page_cache_release(page);
1076 page = NULL;
1077 if (err == -EEXIST)
1078 goto repeat;
1da177e4 1079 }
1da177e4 1080 }
1da177e4
LT
1081 return page;
1082}
1da177e4
LT
1083EXPORT_SYMBOL(find_or_create_page);
1084
0cd6144a
JW
1085/**
1086 * find_get_entries - gang pagecache lookup
1087 * @mapping: The address_space to search
1088 * @start: The starting page cache index
1089 * @nr_entries: The maximum number of entries
1090 * @entries: Where the resulting entries are placed
1091 * @indices: The cache indices corresponding to the entries in @entries
1092 *
1093 * find_get_entries() will search for and return a group of up to
1094 * @nr_entries entries in the mapping. The entries are placed at
1095 * @entries. find_get_entries() takes a reference against any actual
1096 * pages it returns.
1097 *
1098 * The search returns a group of mapping-contiguous page cache entries
1099 * with ascending indexes. There may be holes in the indices due to
1100 * not-present pages.
1101 *
1102 * Any shadow entries of evicted pages are included in the returned
1103 * array.
1104 *
1105 * find_get_entries() returns the number of pages and shadow entries
1106 * which were found.
1107 */
1108unsigned find_get_entries(struct address_space *mapping,
1109 pgoff_t start, unsigned int nr_entries,
1110 struct page **entries, pgoff_t *indices)
1111{
1112 void **slot;
1113 unsigned int ret = 0;
1114 struct radix_tree_iter iter;
1115
1116 if (!nr_entries)
1117 return 0;
1118
1119 rcu_read_lock();
1120restart:
1121 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1122 struct page *page;
1123repeat:
1124 page = radix_tree_deref_slot(slot);
1125 if (unlikely(!page))
1126 continue;
1127 if (radix_tree_exception(page)) {
1128 if (radix_tree_deref_retry(page))
1129 goto restart;
1130 /*
1131 * Otherwise, we must be storing a swap entry
1132 * here as an exceptional entry: so return it
1133 * without attempting to raise page count.
1134 */
1135 goto export;
1136 }
1137 if (!page_cache_get_speculative(page))
1138 goto repeat;
1139
1140 /* Has the page moved? */
1141 if (unlikely(page != *slot)) {
1142 page_cache_release(page);
1143 goto repeat;
1144 }
1145export:
1146 indices[ret] = iter.index;
1147 entries[ret] = page;
1148 if (++ret == nr_entries)
1149 break;
1150 }
1151 rcu_read_unlock();
1152 return ret;
1153}
1154
1da177e4
LT
1155/**
1156 * find_get_pages - gang pagecache lookup
1157 * @mapping: The address_space to search
1158 * @start: The starting page index
1159 * @nr_pages: The maximum number of pages
1160 * @pages: Where the resulting pages are placed
1161 *
1162 * find_get_pages() will search for and return a group of up to
1163 * @nr_pages pages in the mapping. The pages are placed at @pages.
1164 * find_get_pages() takes a reference against the returned pages.
1165 *
1166 * The search returns a group of mapping-contiguous pages with ascending
1167 * indexes. There may be holes in the indices due to not-present pages.
1168 *
1169 * find_get_pages() returns the number of pages which were found.
1170 */
1171unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1172 unsigned int nr_pages, struct page **pages)
1173{
0fc9d104
KK
1174 struct radix_tree_iter iter;
1175 void **slot;
1176 unsigned ret = 0;
1177
1178 if (unlikely(!nr_pages))
1179 return 0;
a60637c8
NP
1180
1181 rcu_read_lock();
1182restart:
0fc9d104 1183 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
a60637c8
NP
1184 struct page *page;
1185repeat:
0fc9d104 1186 page = radix_tree_deref_slot(slot);
a60637c8
NP
1187 if (unlikely(!page))
1188 continue;
9d8aa4ea 1189
a2c16d6c 1190 if (radix_tree_exception(page)) {
8079b1c8
HD
1191 if (radix_tree_deref_retry(page)) {
1192 /*
1193 * Transient condition which can only trigger
1194 * when entry at index 0 moves out of or back
1195 * to root: none yet gotten, safe to restart.
1196 */
0fc9d104 1197 WARN_ON(iter.index);
8079b1c8
HD
1198 goto restart;
1199 }
a2c16d6c 1200 /*
8079b1c8
HD
1201 * Otherwise, shmem/tmpfs must be storing a swap entry
1202 * here as an exceptional entry: so skip over it -
1203 * we only reach this from invalidate_mapping_pages().
a2c16d6c 1204 */
8079b1c8 1205 continue;
27d20fdd 1206 }
a60637c8
NP
1207
1208 if (!page_cache_get_speculative(page))
1209 goto repeat;
1210
1211 /* Has the page moved? */
0fc9d104 1212 if (unlikely(page != *slot)) {
a60637c8
NP
1213 page_cache_release(page);
1214 goto repeat;
1215 }
1da177e4 1216
a60637c8 1217 pages[ret] = page;
0fc9d104
KK
1218 if (++ret == nr_pages)
1219 break;
a60637c8 1220 }
5b280c0c 1221
a60637c8 1222 rcu_read_unlock();
1da177e4
LT
1223 return ret;
1224}
1225
ebf43500
JA
1226/**
1227 * find_get_pages_contig - gang contiguous pagecache lookup
1228 * @mapping: The address_space to search
1229 * @index: The starting page index
1230 * @nr_pages: The maximum number of pages
1231 * @pages: Where the resulting pages are placed
1232 *
1233 * find_get_pages_contig() works exactly like find_get_pages(), except
1234 * that the returned number of pages are guaranteed to be contiguous.
1235 *
1236 * find_get_pages_contig() returns the number of pages which were found.
1237 */
1238unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1239 unsigned int nr_pages, struct page **pages)
1240{
0fc9d104
KK
1241 struct radix_tree_iter iter;
1242 void **slot;
1243 unsigned int ret = 0;
1244
1245 if (unlikely(!nr_pages))
1246 return 0;
a60637c8
NP
1247
1248 rcu_read_lock();
1249restart:
0fc9d104 1250 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
a60637c8
NP
1251 struct page *page;
1252repeat:
0fc9d104
KK
1253 page = radix_tree_deref_slot(slot);
1254 /* The hole, there no reason to continue */
a60637c8 1255 if (unlikely(!page))
0fc9d104 1256 break;
9d8aa4ea 1257
a2c16d6c 1258 if (radix_tree_exception(page)) {
8079b1c8
HD
1259 if (radix_tree_deref_retry(page)) {
1260 /*
1261 * Transient condition which can only trigger
1262 * when entry at index 0 moves out of or back
1263 * to root: none yet gotten, safe to restart.
1264 */
1265 goto restart;
1266 }
a2c16d6c 1267 /*
8079b1c8
HD
1268 * Otherwise, shmem/tmpfs must be storing a swap entry
1269 * here as an exceptional entry: so stop looking for
1270 * contiguous pages.
a2c16d6c 1271 */
8079b1c8 1272 break;
a2c16d6c 1273 }
ebf43500 1274
a60637c8
NP
1275 if (!page_cache_get_speculative(page))
1276 goto repeat;
1277
1278 /* Has the page moved? */
0fc9d104 1279 if (unlikely(page != *slot)) {
a60637c8
NP
1280 page_cache_release(page);
1281 goto repeat;
1282 }
1283
9cbb4cb2
NP
1284 /*
1285 * must check mapping and index after taking the ref.
1286 * otherwise we can get both false positives and false
1287 * negatives, which is just confusing to the caller.
1288 */
0fc9d104 1289 if (page->mapping == NULL || page->index != iter.index) {
9cbb4cb2
NP
1290 page_cache_release(page);
1291 break;
1292 }
1293
a60637c8 1294 pages[ret] = page;
0fc9d104
KK
1295 if (++ret == nr_pages)
1296 break;
ebf43500 1297 }
a60637c8
NP
1298 rcu_read_unlock();
1299 return ret;
ebf43500 1300}
ef71c15c 1301EXPORT_SYMBOL(find_get_pages_contig);
ebf43500 1302
485bb99b
RD
1303/**
1304 * find_get_pages_tag - find and return pages that match @tag
1305 * @mapping: the address_space to search
1306 * @index: the starting page index
1307 * @tag: the tag index
1308 * @nr_pages: the maximum number of pages
1309 * @pages: where the resulting pages are placed
1310 *
1da177e4 1311 * Like find_get_pages, except we only return pages which are tagged with
485bb99b 1312 * @tag. We update @index to index the next page for the traversal.
1da177e4
LT
1313 */
1314unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1315 int tag, unsigned int nr_pages, struct page **pages)
1316{
0fc9d104
KK
1317 struct radix_tree_iter iter;
1318 void **slot;
1319 unsigned ret = 0;
1320
1321 if (unlikely(!nr_pages))
1322 return 0;
a60637c8
NP
1323
1324 rcu_read_lock();
1325restart:
0fc9d104
KK
1326 radix_tree_for_each_tagged(slot, &mapping->page_tree,
1327 &iter, *index, tag) {
a60637c8
NP
1328 struct page *page;
1329repeat:
0fc9d104 1330 page = radix_tree_deref_slot(slot);
a60637c8
NP
1331 if (unlikely(!page))
1332 continue;
9d8aa4ea 1333
a2c16d6c 1334 if (radix_tree_exception(page)) {
8079b1c8
HD
1335 if (radix_tree_deref_retry(page)) {
1336 /*
1337 * Transient condition which can only trigger
1338 * when entry at index 0 moves out of or back
1339 * to root: none yet gotten, safe to restart.
1340 */
1341 goto restart;
1342 }
a2c16d6c 1343 /*
8079b1c8
HD
1344 * This function is never used on a shmem/tmpfs
1345 * mapping, so a swap entry won't be found here.
a2c16d6c 1346 */
8079b1c8 1347 BUG();
a2c16d6c 1348 }
a60637c8
NP
1349
1350 if (!page_cache_get_speculative(page))
1351 goto repeat;
1352
1353 /* Has the page moved? */
0fc9d104 1354 if (unlikely(page != *slot)) {
a60637c8
NP
1355 page_cache_release(page);
1356 goto repeat;
1357 }
1358
1359 pages[ret] = page;
0fc9d104
KK
1360 if (++ret == nr_pages)
1361 break;
a60637c8 1362 }
5b280c0c 1363
a60637c8 1364 rcu_read_unlock();
1da177e4 1365
1da177e4
LT
1366 if (ret)
1367 *index = pages[ret - 1]->index + 1;
a60637c8 1368
1da177e4
LT
1369 return ret;
1370}
ef71c15c 1371EXPORT_SYMBOL(find_get_pages_tag);
1da177e4 1372
485bb99b
RD
1373/**
1374 * grab_cache_page_nowait - returns locked page at given index in given cache
1375 * @mapping: target address_space
1376 * @index: the page index
1377 *
72fd4a35 1378 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1da177e4
LT
1379 * This is intended for speculative data generators, where the data can
1380 * be regenerated if the page couldn't be grabbed. This routine should
1381 * be safe to call while holding the lock for another page.
1382 *
1383 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1384 * and deadlock against the caller's locked page.
1385 */
1386struct page *
57f6b96c 1387grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1da177e4
LT
1388{
1389 struct page *page = find_get_page(mapping, index);
1da177e4
LT
1390
1391 if (page) {
529ae9aa 1392 if (trylock_page(page))
1da177e4
LT
1393 return page;
1394 page_cache_release(page);
1395 return NULL;
1396 }
2ae88149 1397 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
67d58ac4 1398 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1da177e4
LT
1399 page_cache_release(page);
1400 page = NULL;
1401 }
1402 return page;
1403}
1da177e4
LT
1404EXPORT_SYMBOL(grab_cache_page_nowait);
1405
76d42bd9
WF
1406/*
1407 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1408 * a _large_ part of the i/o request. Imagine the worst scenario:
1409 *
1410 * ---R__________________________________________B__________
1411 * ^ reading here ^ bad block(assume 4k)
1412 *
1413 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1414 * => failing the whole request => read(R) => read(R+1) =>
1415 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1416 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1417 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1418 *
1419 * It is going insane. Fix it by quickly scaling down the readahead size.
1420 */
1421static void shrink_readahead_size_eio(struct file *filp,
1422 struct file_ra_state *ra)
1423{
76d42bd9 1424 ra->ra_pages /= 4;
76d42bd9
WF
1425}
1426
485bb99b 1427/**
36e78914 1428 * do_generic_file_read - generic file read routine
485bb99b
RD
1429 * @filp: the file to read
1430 * @ppos: current file position
6e58e79d
AV
1431 * @iter: data destination
1432 * @written: already copied
485bb99b 1433 *
1da177e4 1434 * This is a generic file read routine, and uses the
485bb99b 1435 * mapping->a_ops->readpage() function for the actual low-level stuff.
1da177e4
LT
1436 *
1437 * This is really ugly. But the goto's actually try to clarify some
1438 * of the logic when it comes to error handling etc.
1da177e4 1439 */
6e58e79d
AV
1440static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1441 struct iov_iter *iter, ssize_t written)
1da177e4 1442{
36e78914 1443 struct address_space *mapping = filp->f_mapping;
1da177e4 1444 struct inode *inode = mapping->host;
36e78914 1445 struct file_ra_state *ra = &filp->f_ra;
57f6b96c
FW
1446 pgoff_t index;
1447 pgoff_t last_index;
1448 pgoff_t prev_index;
1449 unsigned long offset; /* offset into pagecache page */
ec0f1637 1450 unsigned int prev_offset;
6e58e79d 1451 int error = 0;
1da177e4 1452
1da177e4 1453 index = *ppos >> PAGE_CACHE_SHIFT;
7ff81078
FW
1454 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1455 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
6e58e79d 1456 last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1da177e4
LT
1457 offset = *ppos & ~PAGE_CACHE_MASK;
1458
1da177e4
LT
1459 for (;;) {
1460 struct page *page;
57f6b96c 1461 pgoff_t end_index;
a32ea1e1 1462 loff_t isize;
1da177e4
LT
1463 unsigned long nr, ret;
1464
1da177e4 1465 cond_resched();
1da177e4
LT
1466find_page:
1467 page = find_get_page(mapping, index);
3ea89ee8 1468 if (!page) {
cf914a7d 1469 page_cache_sync_readahead(mapping,
7ff81078 1470 ra, filp,
3ea89ee8
FW
1471 index, last_index - index);
1472 page = find_get_page(mapping, index);
1473 if (unlikely(page == NULL))
1474 goto no_cached_page;
1475 }
1476 if (PageReadahead(page)) {
cf914a7d 1477 page_cache_async_readahead(mapping,
7ff81078 1478 ra, filp, page,
3ea89ee8 1479 index, last_index - index);
1da177e4 1480 }
8ab22b9a
HH
1481 if (!PageUptodate(page)) {
1482 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1483 !mapping->a_ops->is_partially_uptodate)
1484 goto page_not_up_to_date;
529ae9aa 1485 if (!trylock_page(page))
8ab22b9a 1486 goto page_not_up_to_date;
8d056cb9
DH
1487 /* Did it get truncated before we got the lock? */
1488 if (!page->mapping)
1489 goto page_not_up_to_date_locked;
8ab22b9a 1490 if (!mapping->a_ops->is_partially_uptodate(page,
6e58e79d 1491 offset, iter->count))
8ab22b9a
HH
1492 goto page_not_up_to_date_locked;
1493 unlock_page(page);
1494 }
1da177e4 1495page_ok:
a32ea1e1
N
1496 /*
1497 * i_size must be checked after we know the page is Uptodate.
1498 *
1499 * Checking i_size after the check allows us to calculate
1500 * the correct value for "nr", which means the zero-filled
1501 * part of the page is not copied back to userspace (unless
1502 * another truncate extends the file - this is desired though).
1503 */
1504
1505 isize = i_size_read(inode);
1506 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1507 if (unlikely(!isize || index > end_index)) {
1508 page_cache_release(page);
1509 goto out;
1510 }
1511
1512 /* nr is the maximum number of bytes to copy from this page */
1513 nr = PAGE_CACHE_SIZE;
1514 if (index == end_index) {
1515 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1516 if (nr <= offset) {
1517 page_cache_release(page);
1518 goto out;
1519 }
1520 }
1521 nr = nr - offset;
1da177e4
LT
1522
1523 /* If users can be writing to this page using arbitrary
1524 * virtual addresses, take care about potential aliasing
1525 * before reading the page on the kernel side.
1526 */
1527 if (mapping_writably_mapped(mapping))
1528 flush_dcache_page(page);
1529
1530 /*
ec0f1637
JK
1531 * When a sequential read accesses a page several times,
1532 * only mark it as accessed the first time.
1da177e4 1533 */
ec0f1637 1534 if (prev_index != index || offset != prev_offset)
1da177e4
LT
1535 mark_page_accessed(page);
1536 prev_index = index;
1537
1538 /*
1539 * Ok, we have the page, and it's up-to-date, so
1540 * now we can copy it to user space...
1da177e4 1541 */
6e58e79d
AV
1542
1543 ret = copy_page_to_iter(page, offset, nr, iter);
1da177e4
LT
1544 offset += ret;
1545 index += offset >> PAGE_CACHE_SHIFT;
1546 offset &= ~PAGE_CACHE_MASK;
6ce745ed 1547 prev_offset = offset;
1da177e4
LT
1548
1549 page_cache_release(page);
6e58e79d
AV
1550 written += ret;
1551 if (!iov_iter_count(iter))
1552 goto out;
1553 if (ret < nr) {
1554 error = -EFAULT;
1555 goto out;
1556 }
1557 continue;
1da177e4
LT
1558
1559page_not_up_to_date:
1560 /* Get exclusive access to the page ... */
85462323
ON
1561 error = lock_page_killable(page);
1562 if (unlikely(error))
1563 goto readpage_error;
1da177e4 1564
8ab22b9a 1565page_not_up_to_date_locked:
da6052f7 1566 /* Did it get truncated before we got the lock? */
1da177e4
LT
1567 if (!page->mapping) {
1568 unlock_page(page);
1569 page_cache_release(page);
1570 continue;
1571 }
1572
1573 /* Did somebody else fill it already? */
1574 if (PageUptodate(page)) {
1575 unlock_page(page);
1576 goto page_ok;
1577 }
1578
1579readpage:
91803b49
JM
1580 /*
1581 * A previous I/O error may have been due to temporary
1582 * failures, eg. multipath errors.
1583 * PG_error will be set again if readpage fails.
1584 */
1585 ClearPageError(page);
1da177e4
LT
1586 /* Start the actual read. The read will unlock the page. */
1587 error = mapping->a_ops->readpage(filp, page);
1588
994fc28c
ZB
1589 if (unlikely(error)) {
1590 if (error == AOP_TRUNCATED_PAGE) {
1591 page_cache_release(page);
6e58e79d 1592 error = 0;
994fc28c
ZB
1593 goto find_page;
1594 }
1da177e4 1595 goto readpage_error;
994fc28c 1596 }
1da177e4
LT
1597
1598 if (!PageUptodate(page)) {
85462323
ON
1599 error = lock_page_killable(page);
1600 if (unlikely(error))
1601 goto readpage_error;
1da177e4
LT
1602 if (!PageUptodate(page)) {
1603 if (page->mapping == NULL) {
1604 /*
2ecdc82e 1605 * invalidate_mapping_pages got it
1da177e4
LT
1606 */
1607 unlock_page(page);
1608 page_cache_release(page);
1609 goto find_page;
1610 }
1611 unlock_page(page);
7ff81078 1612 shrink_readahead_size_eio(filp, ra);
85462323
ON
1613 error = -EIO;
1614 goto readpage_error;
1da177e4
LT
1615 }
1616 unlock_page(page);
1617 }
1618
1da177e4
LT
1619 goto page_ok;
1620
1621readpage_error:
1622 /* UHHUH! A synchronous read error occurred. Report it */
1da177e4
LT
1623 page_cache_release(page);
1624 goto out;
1625
1626no_cached_page:
1627 /*
1628 * Ok, it wasn't cached, so we need to create a new
1629 * page..
1630 */
eb2be189
NP
1631 page = page_cache_alloc_cold(mapping);
1632 if (!page) {
6e58e79d 1633 error = -ENOMEM;
eb2be189 1634 goto out;
1da177e4 1635 }
eb2be189 1636 error = add_to_page_cache_lru(page, mapping,
1da177e4
LT
1637 index, GFP_KERNEL);
1638 if (error) {
eb2be189 1639 page_cache_release(page);
6e58e79d
AV
1640 if (error == -EEXIST) {
1641 error = 0;
1da177e4 1642 goto find_page;
6e58e79d 1643 }
1da177e4
LT
1644 goto out;
1645 }
1da177e4
LT
1646 goto readpage;
1647 }
1648
1649out:
7ff81078
FW
1650 ra->prev_pos = prev_index;
1651 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1652 ra->prev_pos |= prev_offset;
1da177e4 1653
f4e6b498 1654 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
0c6aa263 1655 file_accessed(filp);
6e58e79d 1656 return written ? written : error;
1da177e4
LT
1657}
1658
0ceb3314
DM
1659/*
1660 * Performs necessary checks before doing a write
1661 * @iov: io vector request
1662 * @nr_segs: number of segments in the iovec
1663 * @count: number of bytes to write
1664 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1665 *
1666 * Adjust number of segments and amount of bytes to write (nr_segs should be
1667 * properly initialized first). Returns appropriate error code that caller
1668 * should return or zero in case that write should be allowed.
1669 */
1670int generic_segment_checks(const struct iovec *iov,
1671 unsigned long *nr_segs, size_t *count, int access_flags)
1672{
1673 unsigned long seg;
1674 size_t cnt = 0;
1675 for (seg = 0; seg < *nr_segs; seg++) {
1676 const struct iovec *iv = &iov[seg];
1677
1678 /*
1679 * If any segment has a negative length, or the cumulative
1680 * length ever wraps negative then return -EINVAL.
1681 */
1682 cnt += iv->iov_len;
1683 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1684 return -EINVAL;
1685 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1686 continue;
1687 if (seg == 0)
1688 return -EFAULT;
1689 *nr_segs = seg;
1690 cnt -= iv->iov_len; /* This segment is no good */
1691 break;
1692 }
1693 *count = cnt;
1694 return 0;
1695}
1696EXPORT_SYMBOL(generic_segment_checks);
1697
485bb99b 1698/**
b2abacf3 1699 * generic_file_aio_read - generic filesystem read routine
485bb99b
RD
1700 * @iocb: kernel I/O control block
1701 * @iov: io vector request
1702 * @nr_segs: number of segments in the iovec
b2abacf3 1703 * @pos: current file position
485bb99b 1704 *
1da177e4
LT
1705 * This is the "read()" routine for all filesystems
1706 * that can use the page cache directly.
1707 */
1708ssize_t
543ade1f
BP
1709generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1710 unsigned long nr_segs, loff_t pos)
1da177e4
LT
1711{
1712 struct file *filp = iocb->ki_filp;
1713 ssize_t retval;
1da177e4 1714 size_t count;
543ade1f 1715 loff_t *ppos = &iocb->ki_pos;
6e58e79d 1716 struct iov_iter i;
1da177e4
LT
1717
1718 count = 0;
0ceb3314
DM
1719 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1720 if (retval)
1721 return retval;
6e58e79d 1722 iov_iter_init(&i, iov, nr_segs, count, 0);
1da177e4
LT
1723
1724 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1725 if (filp->f_flags & O_DIRECT) {
543ade1f 1726 loff_t size;
1da177e4
LT
1727 struct address_space *mapping;
1728 struct inode *inode;
1729
1730 mapping = filp->f_mapping;
1731 inode = mapping->host;
1da177e4
LT
1732 if (!count)
1733 goto out; /* skip atime */
1734 size = i_size_read(inode);
9fe55eea 1735 retval = filemap_write_and_wait_range(mapping, pos,
48b47c56 1736 pos + iov_length(iov, nr_segs) - 1);
9fe55eea
SW
1737 if (!retval) {
1738 retval = mapping->a_ops->direct_IO(READ, iocb,
1739 iov, pos, nr_segs);
1740 }
1741 if (retval > 0) {
1742 *ppos = pos + retval;
1743 count -= retval;
6e58e79d
AV
1744 /*
1745 * If we did a short DIO read we need to skip the
1746 * section of the iov that we've already read data into.
1747 */
1748 iov_iter_advance(&i, retval);
9fe55eea 1749 }
66f998f6 1750
9fe55eea
SW
1751 /*
1752 * Btrfs can have a short DIO read if we encounter
1753 * compressed extents, so if there was an error, or if
1754 * we've already read everything we wanted to, or if
1755 * there was a short read because we hit EOF, go ahead
1756 * and return. Otherwise fallthrough to buffered io for
1757 * the rest of the read.
1758 */
1759 if (retval < 0 || !count || *ppos >= size) {
1760 file_accessed(filp);
1761 goto out;
0e0bcae3 1762 }
1da177e4
LT
1763 }
1764
6e58e79d 1765 retval = do_generic_file_read(filp, ppos, &i, retval);
1da177e4
LT
1766out:
1767 return retval;
1768}
1da177e4
LT
1769EXPORT_SYMBOL(generic_file_aio_read);
1770
1da177e4 1771#ifdef CONFIG_MMU
485bb99b
RD
1772/**
1773 * page_cache_read - adds requested page to the page cache if not already there
1774 * @file: file to read
1775 * @offset: page index
1776 *
1da177e4
LT
1777 * This adds the requested page to the page cache if it isn't already there,
1778 * and schedules an I/O to read in its contents from disk.
1779 */
920c7a5d 1780static int page_cache_read(struct file *file, pgoff_t offset)
1da177e4
LT
1781{
1782 struct address_space *mapping = file->f_mapping;
1783 struct page *page;
994fc28c 1784 int ret;
1da177e4 1785
994fc28c
ZB
1786 do {
1787 page = page_cache_alloc_cold(mapping);
1788 if (!page)
1789 return -ENOMEM;
1790
1791 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1792 if (ret == 0)
1793 ret = mapping->a_ops->readpage(file, page);
1794 else if (ret == -EEXIST)
1795 ret = 0; /* losing race to add is OK */
1da177e4 1796
1da177e4 1797 page_cache_release(page);
1da177e4 1798
994fc28c
ZB
1799 } while (ret == AOP_TRUNCATED_PAGE);
1800
1801 return ret;
1da177e4
LT
1802}
1803
1804#define MMAP_LOTSAMISS (100)
1805
ef00e08e
LT
1806/*
1807 * Synchronous readahead happens when we don't even find
1808 * a page in the page cache at all.
1809 */
1810static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1811 struct file_ra_state *ra,
1812 struct file *file,
1813 pgoff_t offset)
1814{
1815 unsigned long ra_pages;
1816 struct address_space *mapping = file->f_mapping;
1817
1818 /* If we don't want any read-ahead, don't bother */
64363aad 1819 if (vma->vm_flags & VM_RAND_READ)
ef00e08e 1820 return;
275b12bf
WF
1821 if (!ra->ra_pages)
1822 return;
ef00e08e 1823
64363aad 1824 if (vma->vm_flags & VM_SEQ_READ) {
7ffc59b4
WF
1825 page_cache_sync_readahead(mapping, ra, file, offset,
1826 ra->ra_pages);
ef00e08e
LT
1827 return;
1828 }
1829
207d04ba
AK
1830 /* Avoid banging the cache line if not needed */
1831 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
ef00e08e
LT
1832 ra->mmap_miss++;
1833
1834 /*
1835 * Do we miss much more than hit in this file? If so,
1836 * stop bothering with read-ahead. It will only hurt.
1837 */
1838 if (ra->mmap_miss > MMAP_LOTSAMISS)
1839 return;
1840
d30a1100
WF
1841 /*
1842 * mmap read-around
1843 */
ef00e08e 1844 ra_pages = max_sane_readahead(ra->ra_pages);
275b12bf
WF
1845 ra->start = max_t(long, 0, offset - ra_pages / 2);
1846 ra->size = ra_pages;
2cbea1d3 1847 ra->async_size = ra_pages / 4;
275b12bf 1848 ra_submit(ra, mapping, file);
ef00e08e
LT
1849}
1850
1851/*
1852 * Asynchronous readahead happens when we find the page and PG_readahead,
1853 * so we want to possibly extend the readahead further..
1854 */
1855static void do_async_mmap_readahead(struct vm_area_struct *vma,
1856 struct file_ra_state *ra,
1857 struct file *file,
1858 struct page *page,
1859 pgoff_t offset)
1860{
1861 struct address_space *mapping = file->f_mapping;
1862
1863 /* If we don't want any read-ahead, don't bother */
64363aad 1864 if (vma->vm_flags & VM_RAND_READ)
ef00e08e
LT
1865 return;
1866 if (ra->mmap_miss > 0)
1867 ra->mmap_miss--;
1868 if (PageReadahead(page))
2fad6f5d
WF
1869 page_cache_async_readahead(mapping, ra, file,
1870 page, offset, ra->ra_pages);
ef00e08e
LT
1871}
1872
485bb99b 1873/**
54cb8821 1874 * filemap_fault - read in file data for page fault handling
d0217ac0
NP
1875 * @vma: vma in which the fault was taken
1876 * @vmf: struct vm_fault containing details of the fault
485bb99b 1877 *
54cb8821 1878 * filemap_fault() is invoked via the vma operations vector for a
1da177e4
LT
1879 * mapped memory region to read in file data during a page fault.
1880 *
1881 * The goto's are kind of ugly, but this streamlines the normal case of having
1882 * it in the page cache, and handles the special cases reasonably without
1883 * having a lot of duplicated code.
1884 */
d0217ac0 1885int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1da177e4
LT
1886{
1887 int error;
54cb8821 1888 struct file *file = vma->vm_file;
1da177e4
LT
1889 struct address_space *mapping = file->f_mapping;
1890 struct file_ra_state *ra = &file->f_ra;
1891 struct inode *inode = mapping->host;
ef00e08e 1892 pgoff_t offset = vmf->pgoff;
1da177e4 1893 struct page *page;
99e3e53f 1894 loff_t size;
83c54070 1895 int ret = 0;
1da177e4 1896
99e3e53f
KS
1897 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1898 if (offset >= size >> PAGE_CACHE_SHIFT)
5307cc1a 1899 return VM_FAULT_SIGBUS;
1da177e4 1900
1da177e4 1901 /*
49426420 1902 * Do we have something in the page cache already?
1da177e4 1903 */
ef00e08e 1904 page = find_get_page(mapping, offset);
45cac65b 1905 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1da177e4 1906 /*
ef00e08e
LT
1907 * We found the page, so try async readahead before
1908 * waiting for the lock.
1da177e4 1909 */
ef00e08e 1910 do_async_mmap_readahead(vma, ra, file, page, offset);
45cac65b 1911 } else if (!page) {
ef00e08e
LT
1912 /* No page in the page cache at all */
1913 do_sync_mmap_readahead(vma, ra, file, offset);
1914 count_vm_event(PGMAJFAULT);
456f998e 1915 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
ef00e08e
LT
1916 ret = VM_FAULT_MAJOR;
1917retry_find:
b522c94d 1918 page = find_get_page(mapping, offset);
1da177e4
LT
1919 if (!page)
1920 goto no_cached_page;
1921 }
1922
d88c0922
ML
1923 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1924 page_cache_release(page);
d065bd81 1925 return ret | VM_FAULT_RETRY;
d88c0922 1926 }
b522c94d
ML
1927
1928 /* Did it get truncated? */
1929 if (unlikely(page->mapping != mapping)) {
1930 unlock_page(page);
1931 put_page(page);
1932 goto retry_find;
1933 }
309381fe 1934 VM_BUG_ON_PAGE(page->index != offset, page);
b522c94d 1935
1da177e4 1936 /*
d00806b1
NP
1937 * We have a locked page in the page cache, now we need to check
1938 * that it's up-to-date. If not, it is going to be due to an error.
1da177e4 1939 */
d00806b1 1940 if (unlikely(!PageUptodate(page)))
1da177e4
LT
1941 goto page_not_uptodate;
1942
ef00e08e
LT
1943 /*
1944 * Found the page and have a reference on it.
1945 * We must recheck i_size under page lock.
1946 */
99e3e53f
KS
1947 size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1948 if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
d00806b1 1949 unlock_page(page);
745ad48e 1950 page_cache_release(page);
5307cc1a 1951 return VM_FAULT_SIGBUS;
d00806b1
NP
1952 }
1953
d0217ac0 1954 vmf->page = page;
83c54070 1955 return ret | VM_FAULT_LOCKED;
1da177e4 1956
1da177e4
LT
1957no_cached_page:
1958 /*
1959 * We're only likely to ever get here if MADV_RANDOM is in
1960 * effect.
1961 */
ef00e08e 1962 error = page_cache_read(file, offset);
1da177e4
LT
1963
1964 /*
1965 * The page we want has now been added to the page cache.
1966 * In the unlikely event that someone removed it in the
1967 * meantime, we'll just come back here and read it again.
1968 */
1969 if (error >= 0)
1970 goto retry_find;
1971
1972 /*
1973 * An error return from page_cache_read can result if the
1974 * system is low on memory, or a problem occurs while trying
1975 * to schedule I/O.
1976 */
1977 if (error == -ENOMEM)
d0217ac0
NP
1978 return VM_FAULT_OOM;
1979 return VM_FAULT_SIGBUS;
1da177e4
LT
1980
1981page_not_uptodate:
1da177e4
LT
1982 /*
1983 * Umm, take care of errors if the page isn't up-to-date.
1984 * Try to re-read it _once_. We do this synchronously,
1985 * because there really aren't any performance issues here
1986 * and we need to check for errors.
1987 */
1da177e4 1988 ClearPageError(page);
994fc28c 1989 error = mapping->a_ops->readpage(file, page);
3ef0f720
MS
1990 if (!error) {
1991 wait_on_page_locked(page);
1992 if (!PageUptodate(page))
1993 error = -EIO;
1994 }
d00806b1
NP
1995 page_cache_release(page);
1996
1997 if (!error || error == AOP_TRUNCATED_PAGE)
994fc28c 1998 goto retry_find;
1da177e4 1999
d00806b1 2000 /* Things didn't work out. Return zero to tell the mm layer so. */
76d42bd9 2001 shrink_readahead_size_eio(file, ra);
d0217ac0 2002 return VM_FAULT_SIGBUS;
54cb8821
NP
2003}
2004EXPORT_SYMBOL(filemap_fault);
2005
f1820361
KS
2006void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
2007{
2008 struct radix_tree_iter iter;
2009 void **slot;
2010 struct file *file = vma->vm_file;
2011 struct address_space *mapping = file->f_mapping;
2012 loff_t size;
2013 struct page *page;
2014 unsigned long address = (unsigned long) vmf->virtual_address;
2015 unsigned long addr;
2016 pte_t *pte;
2017
2018 rcu_read_lock();
2019 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2020 if (iter.index > vmf->max_pgoff)
2021 break;
2022repeat:
2023 page = radix_tree_deref_slot(slot);
2024 if (unlikely(!page))
2025 goto next;
2026 if (radix_tree_exception(page)) {
2027 if (radix_tree_deref_retry(page))
2028 break;
2029 else
2030 goto next;
2031 }
2032
2033 if (!page_cache_get_speculative(page))
2034 goto repeat;
2035
2036 /* Has the page moved? */
2037 if (unlikely(page != *slot)) {
2038 page_cache_release(page);
2039 goto repeat;
2040 }
2041
2042 if (!PageUptodate(page) ||
2043 PageReadahead(page) ||
2044 PageHWPoison(page))
2045 goto skip;
2046 if (!trylock_page(page))
2047 goto skip;
2048
2049 if (page->mapping != mapping || !PageUptodate(page))
2050 goto unlock;
2051
99e3e53f
KS
2052 size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2053 if (page->index >= size >> PAGE_CACHE_SHIFT)
f1820361
KS
2054 goto unlock;
2055
2056 pte = vmf->pte + page->index - vmf->pgoff;
2057 if (!pte_none(*pte))
2058 goto unlock;
2059
2060 if (file->f_ra.mmap_miss > 0)
2061 file->f_ra.mmap_miss--;
2062 addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2063 do_set_pte(vma, addr, page, pte, false, false);
2064 unlock_page(page);
2065 goto next;
2066unlock:
2067 unlock_page(page);
2068skip:
2069 page_cache_release(page);
2070next:
2071 if (iter.index == vmf->max_pgoff)
2072 break;
2073 }
2074 rcu_read_unlock();
2075}
2076EXPORT_SYMBOL(filemap_map_pages);
2077
4fcf1c62
JK
2078int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2079{
2080 struct page *page = vmf->page;
496ad9aa 2081 struct inode *inode = file_inode(vma->vm_file);
4fcf1c62
JK
2082 int ret = VM_FAULT_LOCKED;
2083
14da9200 2084 sb_start_pagefault(inode->i_sb);
4fcf1c62
JK
2085 file_update_time(vma->vm_file);
2086 lock_page(page);
2087 if (page->mapping != inode->i_mapping) {
2088 unlock_page(page);
2089 ret = VM_FAULT_NOPAGE;
2090 goto out;
2091 }
14da9200
JK
2092 /*
2093 * We mark the page dirty already here so that when freeze is in
2094 * progress, we are guaranteed that writeback during freezing will
2095 * see the dirty page and writeprotect it again.
2096 */
2097 set_page_dirty(page);
1d1d1a76 2098 wait_for_stable_page(page);
4fcf1c62 2099out:
14da9200 2100 sb_end_pagefault(inode->i_sb);
4fcf1c62
JK
2101 return ret;
2102}
2103EXPORT_SYMBOL(filemap_page_mkwrite);
2104
f0f37e2f 2105const struct vm_operations_struct generic_file_vm_ops = {
54cb8821 2106 .fault = filemap_fault,
f1820361 2107 .map_pages = filemap_map_pages,
4fcf1c62 2108 .page_mkwrite = filemap_page_mkwrite,
0b173bc4 2109 .remap_pages = generic_file_remap_pages,
1da177e4
LT
2110};
2111
2112/* This is used for a general mmap of a disk file */
2113
2114int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2115{
2116 struct address_space *mapping = file->f_mapping;
2117
2118 if (!mapping->a_ops->readpage)
2119 return -ENOEXEC;
2120 file_accessed(file);
2121 vma->vm_ops = &generic_file_vm_ops;
2122 return 0;
2123}
1da177e4
LT
2124
2125/*
2126 * This is for filesystems which do not implement ->writepage.
2127 */
2128int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2129{
2130 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2131 return -EINVAL;
2132 return generic_file_mmap(file, vma);
2133}
2134#else
2135int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2136{
2137 return -ENOSYS;
2138}
2139int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2140{
2141 return -ENOSYS;
2142}
2143#endif /* CONFIG_MMU */
2144
2145EXPORT_SYMBOL(generic_file_mmap);
2146EXPORT_SYMBOL(generic_file_readonly_mmap);
2147
67f9fd91
SL
2148static struct page *wait_on_page_read(struct page *page)
2149{
2150 if (!IS_ERR(page)) {
2151 wait_on_page_locked(page);
2152 if (!PageUptodate(page)) {
2153 page_cache_release(page);
2154 page = ERR_PTR(-EIO);
2155 }
2156 }
2157 return page;
2158}
2159
6fe6900e 2160static struct page *__read_cache_page(struct address_space *mapping,
57f6b96c 2161 pgoff_t index,
5e5358e7 2162 int (*filler)(void *, struct page *),
0531b2aa
LT
2163 void *data,
2164 gfp_t gfp)
1da177e4 2165{
eb2be189 2166 struct page *page;
1da177e4
LT
2167 int err;
2168repeat:
2169 page = find_get_page(mapping, index);
2170 if (!page) {
0531b2aa 2171 page = __page_cache_alloc(gfp | __GFP_COLD);
eb2be189
NP
2172 if (!page)
2173 return ERR_PTR(-ENOMEM);
e6f67b8c 2174 err = add_to_page_cache_lru(page, mapping, index, gfp);
eb2be189
NP
2175 if (unlikely(err)) {
2176 page_cache_release(page);
2177 if (err == -EEXIST)
2178 goto repeat;
1da177e4 2179 /* Presumably ENOMEM for radix tree node */
1da177e4
LT
2180 return ERR_PTR(err);
2181 }
1da177e4
LT
2182 err = filler(data, page);
2183 if (err < 0) {
2184 page_cache_release(page);
2185 page = ERR_PTR(err);
67f9fd91
SL
2186 } else {
2187 page = wait_on_page_read(page);
1da177e4
LT
2188 }
2189 }
1da177e4
LT
2190 return page;
2191}
2192
0531b2aa 2193static struct page *do_read_cache_page(struct address_space *mapping,
57f6b96c 2194 pgoff_t index,
5e5358e7 2195 int (*filler)(void *, struct page *),
0531b2aa
LT
2196 void *data,
2197 gfp_t gfp)
2198
1da177e4
LT
2199{
2200 struct page *page;
2201 int err;
2202
2203retry:
0531b2aa 2204 page = __read_cache_page(mapping, index, filler, data, gfp);
1da177e4 2205 if (IS_ERR(page))
c855ff37 2206 return page;
1da177e4
LT
2207 if (PageUptodate(page))
2208 goto out;
2209
2210 lock_page(page);
2211 if (!page->mapping) {
2212 unlock_page(page);
2213 page_cache_release(page);
2214 goto retry;
2215 }
2216 if (PageUptodate(page)) {
2217 unlock_page(page);
2218 goto out;
2219 }
2220 err = filler(data, page);
2221 if (err < 0) {
2222 page_cache_release(page);
c855ff37 2223 return ERR_PTR(err);
67f9fd91
SL
2224 } else {
2225 page = wait_on_page_read(page);
2226 if (IS_ERR(page))
2227 return page;
1da177e4 2228 }
c855ff37 2229out:
6fe6900e
NP
2230 mark_page_accessed(page);
2231 return page;
2232}
0531b2aa
LT
2233
2234/**
67f9fd91 2235 * read_cache_page - read into page cache, fill it if needed
0531b2aa
LT
2236 * @mapping: the page's address_space
2237 * @index: the page index
2238 * @filler: function to perform the read
5e5358e7 2239 * @data: first arg to filler(data, page) function, often left as NULL
0531b2aa 2240 *
0531b2aa 2241 * Read into the page cache. If a page already exists, and PageUptodate() is
67f9fd91 2242 * not set, try to fill the page and wait for it to become unlocked.
0531b2aa
LT
2243 *
2244 * If the page does not get brought uptodate, return -EIO.
2245 */
67f9fd91 2246struct page *read_cache_page(struct address_space *mapping,
0531b2aa 2247 pgoff_t index,
5e5358e7 2248 int (*filler)(void *, struct page *),
0531b2aa
LT
2249 void *data)
2250{
2251 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2252}
67f9fd91 2253EXPORT_SYMBOL(read_cache_page);
0531b2aa
LT
2254
2255/**
2256 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2257 * @mapping: the page's address_space
2258 * @index: the page index
2259 * @gfp: the page allocator flags to use if allocating
2260 *
2261 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
e6f67b8c 2262 * any new page allocations done using the specified allocation flags.
0531b2aa
LT
2263 *
2264 * If the page does not get brought uptodate, return -EIO.
2265 */
2266struct page *read_cache_page_gfp(struct address_space *mapping,
2267 pgoff_t index,
2268 gfp_t gfp)
2269{
2270 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2271
67f9fd91 2272 return do_read_cache_page(mapping, index, filler, NULL, gfp);
0531b2aa
LT
2273}
2274EXPORT_SYMBOL(read_cache_page_gfp);
2275
1da177e4
LT
2276/*
2277 * Performs necessary checks before doing a write
2278 *
485bb99b 2279 * Can adjust writing position or amount of bytes to write.
1da177e4
LT
2280 * Returns appropriate error code that caller should return or
2281 * zero in case that write should be allowed.
2282 */
2283inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2284{
2285 struct inode *inode = file->f_mapping->host;
59e99e5b 2286 unsigned long limit = rlimit(RLIMIT_FSIZE);
1da177e4
LT
2287
2288 if (unlikely(*pos < 0))
2289 return -EINVAL;
2290
1da177e4
LT
2291 if (!isblk) {
2292 /* FIXME: this is for backwards compatibility with 2.4 */
2293 if (file->f_flags & O_APPEND)
2294 *pos = i_size_read(inode);
2295
2296 if (limit != RLIM_INFINITY) {
2297 if (*pos >= limit) {
2298 send_sig(SIGXFSZ, current, 0);
2299 return -EFBIG;
2300 }
2301 if (*count > limit - (typeof(limit))*pos) {
2302 *count = limit - (typeof(limit))*pos;
2303 }
2304 }
2305 }
2306
2307 /*
2308 * LFS rule
2309 */
2310 if (unlikely(*pos + *count > MAX_NON_LFS &&
2311 !(file->f_flags & O_LARGEFILE))) {
2312 if (*pos >= MAX_NON_LFS) {
1da177e4
LT
2313 return -EFBIG;
2314 }
2315 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2316 *count = MAX_NON_LFS - (unsigned long)*pos;
2317 }
2318 }
2319
2320 /*
2321 * Are we about to exceed the fs block limit ?
2322 *
2323 * If we have written data it becomes a short write. If we have
2324 * exceeded without writing data we send a signal and return EFBIG.
2325 * Linus frestrict idea will clean these up nicely..
2326 */
2327 if (likely(!isblk)) {
2328 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2329 if (*count || *pos > inode->i_sb->s_maxbytes) {
1da177e4
LT
2330 return -EFBIG;
2331 }
2332 /* zero-length writes at ->s_maxbytes are OK */
2333 }
2334
2335 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2336 *count = inode->i_sb->s_maxbytes - *pos;
2337 } else {
9361401e 2338#ifdef CONFIG_BLOCK
1da177e4
LT
2339 loff_t isize;
2340 if (bdev_read_only(I_BDEV(inode)))
2341 return -EPERM;
2342 isize = i_size_read(inode);
2343 if (*pos >= isize) {
2344 if (*count || *pos > isize)
2345 return -ENOSPC;
2346 }
2347
2348 if (*pos + *count > isize)
2349 *count = isize - *pos;
9361401e
DH
2350#else
2351 return -EPERM;
2352#endif
1da177e4
LT
2353 }
2354 return 0;
2355}
2356EXPORT_SYMBOL(generic_write_checks);
2357
afddba49
NP
2358int pagecache_write_begin(struct file *file, struct address_space *mapping,
2359 loff_t pos, unsigned len, unsigned flags,
2360 struct page **pagep, void **fsdata)
2361{
2362 const struct address_space_operations *aops = mapping->a_ops;
2363
4e02ed4b 2364 return aops->write_begin(file, mapping, pos, len, flags,
afddba49 2365 pagep, fsdata);
afddba49
NP
2366}
2367EXPORT_SYMBOL(pagecache_write_begin);
2368
2369int pagecache_write_end(struct file *file, struct address_space *mapping,
2370 loff_t pos, unsigned len, unsigned copied,
2371 struct page *page, void *fsdata)
2372{
2373 const struct address_space_operations *aops = mapping->a_ops;
afddba49 2374
4e02ed4b
NP
2375 mark_page_accessed(page);
2376 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
afddba49
NP
2377}
2378EXPORT_SYMBOL(pagecache_write_end);
2379
1da177e4
LT
2380ssize_t
2381generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
5cb6c6c7 2382 unsigned long *nr_segs, loff_t pos,
1da177e4
LT
2383 size_t count, size_t ocount)
2384{
2385 struct file *file = iocb->ki_filp;
2386 struct address_space *mapping = file->f_mapping;
2387 struct inode *inode = mapping->host;
2388 ssize_t written;
a969e903
CH
2389 size_t write_len;
2390 pgoff_t end;
1da177e4
LT
2391
2392 if (count != ocount)
2393 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2394
a969e903
CH
2395 write_len = iov_length(iov, *nr_segs);
2396 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
a969e903 2397
48b47c56 2398 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
a969e903
CH
2399 if (written)
2400 goto out;
2401
2402 /*
2403 * After a write we want buffered reads to be sure to go to disk to get
2404 * the new data. We invalidate clean cached page from the region we're
2405 * about to write. We do this *before* the write so that we can return
6ccfa806 2406 * without clobbering -EIOCBQUEUED from ->direct_IO().
a969e903
CH
2407 */
2408 if (mapping->nrpages) {
2409 written = invalidate_inode_pages2_range(mapping,
2410 pos >> PAGE_CACHE_SHIFT, end);
6ccfa806
HH
2411 /*
2412 * If a page can not be invalidated, return 0 to fall back
2413 * to buffered write.
2414 */
2415 if (written) {
2416 if (written == -EBUSY)
2417 return 0;
a969e903 2418 goto out;
6ccfa806 2419 }
a969e903
CH
2420 }
2421
2422 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2423
2424 /*
2425 * Finally, try again to invalidate clean pages which might have been
2426 * cached by non-direct readahead, or faulted in by get_user_pages()
2427 * if the source of the write was an mmap'ed region of the file
2428 * we're writing. Either one is a pretty crazy thing to do,
2429 * so we don't support it 100%. If this invalidation
2430 * fails, tough, the write still worked...
2431 */
2432 if (mapping->nrpages) {
2433 invalidate_inode_pages2_range(mapping,
2434 pos >> PAGE_CACHE_SHIFT, end);
2435 }
2436
1da177e4 2437 if (written > 0) {
0116651c
NK
2438 pos += written;
2439 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2440 i_size_write(inode, pos);
1da177e4
LT
2441 mark_inode_dirty(inode);
2442 }
5cb6c6c7 2443 iocb->ki_pos = pos;
1da177e4 2444 }
a969e903 2445out:
1da177e4
LT
2446 return written;
2447}
2448EXPORT_SYMBOL(generic_file_direct_write);
2449
eb2be189
NP
2450/*
2451 * Find or create a page at the given pagecache position. Return the locked
2452 * page. This function is specifically for buffered writes.
2453 */
54566b2c
NP
2454struct page *grab_cache_page_write_begin(struct address_space *mapping,
2455 pgoff_t index, unsigned flags)
eb2be189
NP
2456{
2457 int status;
0faa70cb 2458 gfp_t gfp_mask;
eb2be189 2459 struct page *page;
54566b2c 2460 gfp_t gfp_notmask = 0;
0faa70cb 2461
1010bb1b
FW
2462 gfp_mask = mapping_gfp_mask(mapping);
2463 if (mapping_cap_account_dirty(mapping))
2464 gfp_mask |= __GFP_WRITE;
54566b2c
NP
2465 if (flags & AOP_FLAG_NOFS)
2466 gfp_notmask = __GFP_FS;
eb2be189
NP
2467repeat:
2468 page = find_lock_page(mapping, index);
c585a267 2469 if (page)
3d08bcc8 2470 goto found;
eb2be189 2471
0faa70cb 2472 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
eb2be189
NP
2473 if (!page)
2474 return NULL;
54566b2c
NP
2475 status = add_to_page_cache_lru(page, mapping, index,
2476 GFP_KERNEL & ~gfp_notmask);
eb2be189
NP
2477 if (unlikely(status)) {
2478 page_cache_release(page);
2479 if (status == -EEXIST)
2480 goto repeat;
2481 return NULL;
2482 }
3d08bcc8 2483found:
1d1d1a76 2484 wait_for_stable_page(page);
eb2be189
NP
2485 return page;
2486}
54566b2c 2487EXPORT_SYMBOL(grab_cache_page_write_begin);
eb2be189 2488
3b93f911 2489ssize_t generic_perform_write(struct file *file,
afddba49
NP
2490 struct iov_iter *i, loff_t pos)
2491{
2492 struct address_space *mapping = file->f_mapping;
2493 const struct address_space_operations *a_ops = mapping->a_ops;
2494 long status = 0;
2495 ssize_t written = 0;
674b892e
NP
2496 unsigned int flags = 0;
2497
2498 /*
2499 * Copies from kernel address space cannot fail (NFSD is a big user).
2500 */
2501 if (segment_eq(get_fs(), KERNEL_DS))
2502 flags |= AOP_FLAG_UNINTERRUPTIBLE;
afddba49
NP
2503
2504 do {
2505 struct page *page;
afddba49
NP
2506 unsigned long offset; /* Offset into pagecache page */
2507 unsigned long bytes; /* Bytes to write to page */
2508 size_t copied; /* Bytes copied from user */
2509 void *fsdata;
2510
2511 offset = (pos & (PAGE_CACHE_SIZE - 1));
afddba49
NP
2512 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2513 iov_iter_count(i));
2514
2515again:
afddba49
NP
2516 /*
2517 * Bring in the user page that we will copy from _first_.
2518 * Otherwise there's a nasty deadlock on copying from the
2519 * same page as we're writing to, without it being marked
2520 * up-to-date.
2521 *
2522 * Not only is this an optimisation, but it is also required
2523 * to check that the address is actually valid, when atomic
2524 * usercopies are used, below.
2525 */
2526 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2527 status = -EFAULT;
2528 break;
2529 }
2530
674b892e 2531 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
afddba49
NP
2532 &page, &fsdata);
2533 if (unlikely(status))
2534 break;
2535
931e80e4 2536 if (mapping_writably_mapped(mapping))
2537 flush_dcache_page(page);
2538
afddba49 2539 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
afddba49
NP
2540 flush_dcache_page(page);
2541
c8236db9 2542 mark_page_accessed(page);
afddba49
NP
2543 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2544 page, fsdata);
2545 if (unlikely(status < 0))
2546 break;
2547 copied = status;
2548
2549 cond_resched();
2550
124d3b70 2551 iov_iter_advance(i, copied);
afddba49
NP
2552 if (unlikely(copied == 0)) {
2553 /*
2554 * If we were unable to copy any data at all, we must
2555 * fall back to a single segment length write.
2556 *
2557 * If we didn't fallback here, we could livelock
2558 * because not all segments in the iov can be copied at
2559 * once without a pagefault.
2560 */
2561 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2562 iov_iter_single_seg_count(i));
2563 goto again;
2564 }
afddba49
NP
2565 pos += copied;
2566 written += copied;
2567
2568 balance_dirty_pages_ratelimited(mapping);
a50527b1
JK
2569 if (fatal_signal_pending(current)) {
2570 status = -EINTR;
2571 break;
2572 }
afddba49
NP
2573 } while (iov_iter_count(i));
2574
2575 return written ? written : status;
2576}
3b93f911 2577EXPORT_SYMBOL(generic_perform_write);
1da177e4 2578
e4dd9de3
JK
2579/**
2580 * __generic_file_aio_write - write data to a file
2581 * @iocb: IO state structure (file, offset, etc.)
2582 * @iov: vector with data to write
2583 * @nr_segs: number of segments in the vector
2584 * @ppos: position where to write
2585 *
2586 * This function does all the work needed for actually writing data to a
2587 * file. It does all basic checks, removes SUID from the file, updates
2588 * modification times and calls proper subroutines depending on whether we
2589 * do direct IO or a standard buffered write.
2590 *
2591 * It expects i_mutex to be grabbed unless we work on a block device or similar
2592 * object which does not need locking at all.
2593 *
2594 * This function does *not* take care of syncing data in case of O_SYNC write.
2595 * A caller has to handle it. This is mainly due to the fact that we want to
2596 * avoid syncing under i_mutex.
2597 */
2598ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
41fc56d5 2599 unsigned long nr_segs)
1da177e4
LT
2600{
2601 struct file *file = iocb->ki_filp;
fb5527e6 2602 struct address_space * mapping = file->f_mapping;
1da177e4
LT
2603 size_t ocount; /* original count */
2604 size_t count; /* after file limit checks */
2605 struct inode *inode = mapping->host;
41fc56d5 2606 loff_t pos = iocb->ki_pos;
3b93f911 2607 ssize_t written = 0;
1da177e4 2608 ssize_t err;
3b93f911
AV
2609 ssize_t status;
2610 struct iov_iter from;
1da177e4
LT
2611
2612 ocount = 0;
0ceb3314
DM
2613 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2614 if (err)
2615 return err;
1da177e4
LT
2616
2617 count = ocount;
1da177e4 2618
1da177e4
LT
2619 /* We can write back this queue in page reclaim */
2620 current->backing_dev_info = mapping->backing_dev_info;
1da177e4
LT
2621 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2622 if (err)
2623 goto out;
2624
2625 if (count == 0)
2626 goto out;
2627
2f1936b8 2628 err = file_remove_suid(file);
1da177e4
LT
2629 if (err)
2630 goto out;
2631
c3b2da31
JB
2632 err = file_update_time(file);
2633 if (err)
2634 goto out;
1da177e4 2635
3b93f911
AV
2636 iov_iter_init(&from, iov, nr_segs, count, 0);
2637
1da177e4
LT
2638 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2639 if (unlikely(file->f_flags & O_DIRECT)) {
fb5527e6 2640 loff_t endbyte;
fb5527e6 2641
3b93f911 2642 written = generic_file_direct_write(iocb, iov, &from.nr_segs, pos,
5cb6c6c7 2643 count, ocount);
1da177e4
LT
2644 if (written < 0 || written == count)
2645 goto out;
3b93f911
AV
2646 iov_iter_advance(&from, written);
2647
1da177e4
LT
2648 /*
2649 * direct-io write to a hole: fall through to buffered I/O
2650 * for completing the rest of the request.
2651 */
2652 pos += written;
2653 count -= written;
3b93f911
AV
2654
2655 status = generic_perform_write(file, &from, pos);
fb5527e6 2656 /*
3b93f911 2657 * If generic_perform_write() returned a synchronous error
fb5527e6
JM
2658 * then we want to return the number of bytes which were
2659 * direct-written, or the error code if that was zero. Note
2660 * that this differs from normal direct-io semantics, which
2661 * will return -EFOO even if some bytes were written.
2662 */
3b93f911
AV
2663 if (unlikely(status < 0) && !written) {
2664 err = status;
fb5527e6
JM
2665 goto out;
2666 }
3b93f911 2667 iocb->ki_pos = pos + status;
fb5527e6
JM
2668 /*
2669 * We need to ensure that the page cache pages are written to
2670 * disk and invalidated to preserve the expected O_DIRECT
2671 * semantics.
2672 */
3b93f911 2673 endbyte = pos + status - 1;
c05c4edd 2674 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
fb5527e6 2675 if (err == 0) {
3b93f911 2676 written += status;
fb5527e6
JM
2677 invalidate_mapping_pages(mapping,
2678 pos >> PAGE_CACHE_SHIFT,
2679 endbyte >> PAGE_CACHE_SHIFT);
2680 } else {
2681 /*
2682 * We don't know how much we wrote, so just return
2683 * the number of bytes which were direct-written
2684 */
2685 }
2686 } else {
3b93f911
AV
2687 written = generic_perform_write(file, &from, pos);
2688 if (likely(written >= 0))
2689 iocb->ki_pos = pos + written;
fb5527e6 2690 }
1da177e4
LT
2691out:
2692 current->backing_dev_info = NULL;
2693 return written ? written : err;
2694}
e4dd9de3
JK
2695EXPORT_SYMBOL(__generic_file_aio_write);
2696
e4dd9de3
JK
2697/**
2698 * generic_file_aio_write - write data to a file
2699 * @iocb: IO state structure
2700 * @iov: vector with data to write
2701 * @nr_segs: number of segments in the vector
2702 * @pos: position in file where to write
2703 *
2704 * This is a wrapper around __generic_file_aio_write() to be used by most
2705 * filesystems. It takes care of syncing the file in case of O_SYNC file
2706 * and acquires i_mutex as needed.
2707 */
027445c3
BP
2708ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2709 unsigned long nr_segs, loff_t pos)
1da177e4
LT
2710{
2711 struct file *file = iocb->ki_filp;
148f948b 2712 struct inode *inode = file->f_mapping->host;
1da177e4 2713 ssize_t ret;
1da177e4
LT
2714
2715 BUG_ON(iocb->ki_pos != pos);
2716
1b1dcc1b 2717 mutex_lock(&inode->i_mutex);
41fc56d5 2718 ret = __generic_file_aio_write(iocb, iov, nr_segs);
1b1dcc1b 2719 mutex_unlock(&inode->i_mutex);
1da177e4 2720
02afc27f 2721 if (ret > 0) {
1da177e4
LT
2722 ssize_t err;
2723
d311d79d
AV
2724 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2725 if (err < 0)
1da177e4
LT
2726 ret = err;
2727 }
2728 return ret;
2729}
2730EXPORT_SYMBOL(generic_file_aio_write);
2731
cf9a2ae8
DH
2732/**
2733 * try_to_release_page() - release old fs-specific metadata on a page
2734 *
2735 * @page: the page which the kernel is trying to free
2736 * @gfp_mask: memory allocation flags (and I/O mode)
2737 *
2738 * The address_space is to try to release any data against the page
2739 * (presumably at page->private). If the release was successful, return `1'.
2740 * Otherwise return zero.
2741 *
266cf658
DH
2742 * This may also be called if PG_fscache is set on a page, indicating that the
2743 * page is known to the local caching routines.
2744 *
cf9a2ae8 2745 * The @gfp_mask argument specifies whether I/O may be performed to release
3f31fddf 2746 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
cf9a2ae8 2747 *
cf9a2ae8
DH
2748 */
2749int try_to_release_page(struct page *page, gfp_t gfp_mask)
2750{
2751 struct address_space * const mapping = page->mapping;
2752
2753 BUG_ON(!PageLocked(page));
2754 if (PageWriteback(page))
2755 return 0;
2756
2757 if (mapping && mapping->a_ops->releasepage)
2758 return mapping->a_ops->releasepage(page, gfp_mask);
2759 return try_to_free_buffers(page);
2760}
2761
2762EXPORT_SYMBOL(try_to_release_page);