blob: 4b51ac1acae7c6ae5233afeb8018dd31cadf909b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 */
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040012#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#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>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080032#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070033#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080034#include <linux/memcontrol.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060035#include <linux/cleancache.h>
Nick Piggin0f8053a2006-03-22 00:08:33 -080036#include "internal.h"
37
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -070038#define CREATE_TRACE_POINTS
39#include <trace/events/filemap.h>
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 * FIXME: remove all knowledge of the buffer layer from the core VM
43 */
Jan Kara148f9482009-08-17 19:52:36 +020044#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <asm/mman.h>
47
48/*
49 * Shared mappings implemented 30.11.1994. It's not fully working yet,
50 * though.
51 *
52 * Shared mappings now work. 15.8.1995 Bruno.
53 *
54 * finished 'unifying' the page and buffer cache and SMP-threaded the
55 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
56 *
57 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
58 */
59
60/*
61 * Lock ordering:
62 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070063 * ->i_mmap_mutex (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070065 * ->swap_lock (exclusive_swap_page, others)
66 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080068 * ->i_mutex
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070069 * ->i_mmap_mutex (truncate->unmap_mapping_range)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 *
71 * ->mmap_sem
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070072 * ->i_mmap_mutex
Hugh Dickinsb8072f02005-10-29 18:16:41 -070073 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
75 *
76 * ->mmap_sem
77 * ->lock_page (access_process_vm)
78 *
Nick Piggin82591e62006-10-19 23:29:10 -070079 * ->i_mutex (generic_file_buffered_write)
80 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 *
Christoph Hellwigf758eea2011-04-21 18:19:44 -060082 * bdi->wb.list_lock
Dave Chinnera66979a2011-03-22 22:23:41 +110083 * sb_lock (fs/fs-writeback.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * ->mapping->tree_lock (__sync_single_inode)
85 *
Peter Zijlstra3d48ae42011-05-24 17:12:06 -070086 * ->i_mmap_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * ->anon_vma.lock (vma_adjust)
88 *
89 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070090 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070093 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 * ->private_lock (try_to_unmap_one)
95 * ->tree_lock (try_to_unmap_one)
96 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080097 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 * ->private_lock (page_remove_rmap->set_page_dirty)
99 * ->tree_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600100 * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100101 * ->inode->i_lock (page_remove_rmap->set_page_dirty)
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600102 * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
Dave Chinner250df6e2011-03-22 22:23:36 +1100103 * ->inode->i_lock (zap_pte_range->set_page_dirty)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 *
Andi Kleen9a3c5312012-03-21 16:34:09 -0700106 * ->i_mmap_mutex
107 * ->tasklist_lock (memory_failure, collect_procs_ao)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 */
109
110/*
Minchan Kime64a7822011-03-22 16:32:44 -0700111 * Delete a page from the page cache and free it. Caller has to make
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
Minchan Kime64a7822011-03-22 16:32:44 -0700115void __delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116{
117 struct address_space *mapping = page->mapping;
118
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700119 trace_mm_filemap_delete_from_page_cache(page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600120 /*
121 * if we're uptodate, flush out into the cleancache, otherwise
122 * invalidate any existing cleancache entries. We can't leave
123 * stale data around in the cleancache once our page is gone
124 */
125 if (PageUptodate(page) && PageMappedToDisk(page))
126 cleancache_put_page(page);
127 else
Dan Magenheimer31677602011-09-21 11:56:28 -0400128 cleancache_invalidate_page(mapping, page);
Dan Magenheimerc515e1f2011-05-26 10:01:43 -0600129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 radix_tree_delete(&mapping->page_tree, page->index);
131 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700132 /* Leave page->index set: truncation lookup relies upon it */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700134 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700135 if (PageSwapBacked(page))
136 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700137 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800138
139 /*
140 * Some filesystems seem to re-dirty the page even after
141 * the VM has canceled the dirty bit (eg ext3 journaling).
142 *
143 * Fix it up by doing a final dirty accounting check after
144 * having removed the page entirely.
145 */
146 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
147 dec_zone_page_state(page, NR_FILE_DIRTY);
148 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150}
151
Minchan Kim702cfbf2011-03-22 16:32:43 -0700152/**
153 * delete_from_page_cache - delete page from page cache
154 * @page: the page which the kernel is trying to remove from page cache
155 *
156 * This must be called only on pages that have been verified to be in the page
157 * cache and locked. It will never put the page into the free list, the caller
158 * has a reference on the page.
159 */
160void delete_from_page_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161{
162 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500163 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Matt Mackallcd7619d2005-05-01 08:59:01 -0700165 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Linus Torvalds6072d132010-12-01 13:35:19 -0500167 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700168 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700169 __delete_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700170 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700171 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500172
173 if (freepage)
174 freepage(page);
Minchan Kim97cecb52011-03-22 16:30:53 -0700175 page_cache_release(page);
176}
177EXPORT_SYMBOL(delete_from_page_cache);
178
Jens Axboe7eaceac2011-03-10 08:52:07 +0100179static int sleep_on_page(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 io_schedule();
182 return 0;
183}
184
Jens Axboe7eaceac2011-03-10 08:52:07 +0100185static int sleep_on_page_killable(void *word)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500186{
Jens Axboe7eaceac2011-03-10 08:52:07 +0100187 sleep_on_page(word);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500188 return fatal_signal_pending(current) ? -EINTR : 0;
189}
190
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700191static int filemap_check_errors(struct address_space *mapping)
192{
193 int ret = 0;
194 /* Check for outstanding write errors */
195 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
196 ret = -ENOSPC;
197 if (test_and_clear_bit(AS_EIO, &mapping->flags))
198 ret = -EIO;
199 return ret;
200}
201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700203 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700204 * @mapping: address space structure to write
205 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800206 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700207 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700209 * Start writeback against all of a mapping's dirty pages that lie
210 * within the byte offsets <start, end> inclusive.
211 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700213 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 * these two operations is that if a dirty page/buffer is encountered, it must
215 * be waited upon, and not just skipped over.
216 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800217int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
218 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219{
220 int ret;
221 struct writeback_control wbc = {
222 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800223 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700224 .range_start = start,
225 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 };
227
228 if (!mapping_cap_writeback_dirty(mapping))
229 return 0;
230
231 ret = do_writepages(mapping, &wbc);
232 return ret;
233}
234
235static inline int __filemap_fdatawrite(struct address_space *mapping,
236 int sync_mode)
237{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700238 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
241int filemap_fdatawrite(struct address_space *mapping)
242{
243 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
244}
245EXPORT_SYMBOL(filemap_fdatawrite);
246
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400247int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800248 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249{
250 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
251}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400252EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Randy Dunlap485bb992006-06-23 02:03:49 -0700254/**
255 * filemap_flush - mostly a non-blocking flush
256 * @mapping: target address_space
257 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 * This is a mostly non-blocking flush. Not suitable for data-integrity
259 * purposes - I/O may not be started against all dirty pages.
260 */
261int filemap_flush(struct address_space *mapping)
262{
263 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
264}
265EXPORT_SYMBOL(filemap_flush);
266
Randy Dunlap485bb992006-06-23 02:03:49 -0700267/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200268 * filemap_fdatawait_range - wait for writeback to complete
269 * @mapping: address space structure to wait for
270 * @start_byte: offset in bytes where the range starts
271 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700272 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200273 * Walk the list of under-writeback pages of the given address space
274 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200276int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
277 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200279 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
280 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 struct pagevec pvec;
282 int nr_pages;
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700283 int ret2, ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200285 if (end_byte < start_byte)
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700286 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
288 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 while ((index <= end) &&
290 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
291 PAGECACHE_TAG_WRITEBACK,
292 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
293 unsigned i;
294
295 for (i = 0; i < nr_pages; i++) {
296 struct page *page = pvec.pages[i];
297
298 /* until radix tree lookup accepts end_index */
299 if (page->index > end)
300 continue;
301
302 wait_on_page_writeback(page);
Rik van Riel212260a2011-01-13 15:46:06 -0800303 if (TestClearPageError(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 ret = -EIO;
305 }
306 pagevec_release(&pvec);
307 cond_resched();
308 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700309out:
310 ret2 = filemap_check_errors(mapping);
311 if (!ret)
312 ret = ret2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
314 return ret;
315}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200316EXPORT_SYMBOL(filemap_fdatawait_range);
317
318/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700319 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700321 *
322 * Walk the list of under-writeback pages of the given address space
323 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 */
325int filemap_fdatawait(struct address_space *mapping)
326{
327 loff_t i_size = i_size_read(mapping->host);
328
329 if (i_size == 0)
330 return 0;
331
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200332 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334EXPORT_SYMBOL(filemap_fdatawait);
335
336int filemap_write_and_wait(struct address_space *mapping)
337{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800338 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800341 err = filemap_fdatawrite(mapping);
342 /*
343 * Even if the above returned error, the pages may be
344 * written partially (e.g. -ENOSPC), so we wait for it.
345 * But the -EIO is special case, it may indicate the worst
346 * thing (e.g. bug) happened, so we avoid waiting for it.
347 */
348 if (err != -EIO) {
349 int err2 = filemap_fdatawait(mapping);
350 if (!err)
351 err = err2;
352 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700353 } else {
354 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800356 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800358EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
Randy Dunlap485bb992006-06-23 02:03:49 -0700360/**
361 * filemap_write_and_wait_range - write out & wait on a file range
362 * @mapping: the address_space for the pages
363 * @lstart: offset in bytes where the range starts
364 * @lend: offset in bytes where the range ends (inclusive)
365 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800366 * Write out and wait upon file offsets lstart->lend, inclusive.
367 *
368 * Note that `lend' is inclusive (describes the last byte to be written) so
369 * that this function can be used to write to the very end-of-file (end = -1).
370 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371int filemap_write_and_wait_range(struct address_space *mapping,
372 loff_t lstart, loff_t lend)
373{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800374 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
376 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800377 err = __filemap_fdatawrite_range(mapping, lstart, lend,
378 WB_SYNC_ALL);
379 /* See comment of filemap_write_and_wait() */
380 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200381 int err2 = filemap_fdatawait_range(mapping,
382 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800383 if (!err)
384 err = err2;
385 }
Dmitry Monakhov865ffef32013-04-29 15:08:42 -0700386 } else {
387 err = filemap_check_errors(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800389 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
Chris Masonf6995582009-04-15 13:22:37 -0400391EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
Randy Dunlap485bb992006-06-23 02:03:49 -0700393/**
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700394 * replace_page_cache_page - replace a pagecache page with a new one
395 * @old: page to be replaced
396 * @new: page to replace with
397 * @gfp_mask: allocation mode
398 *
399 * This function replaces a page in the pagecache with a new one. On
400 * success it acquires the pagecache reference for the new page and
401 * drops it for the old page. Both the old and new pages must be
402 * locked. This function does not add the new page to the LRU, the
403 * caller must do that.
404 *
405 * The remove + add is atomic. The only way this function can fail is
406 * memory allocation failure.
407 */
408int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
409{
410 int error;
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700411
412 VM_BUG_ON(!PageLocked(old));
413 VM_BUG_ON(!PageLocked(new));
414 VM_BUG_ON(new->mapping);
415
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700416 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
417 if (!error) {
418 struct address_space *mapping = old->mapping;
419 void (*freepage)(struct page *);
420
421 pgoff_t offset = old->index;
422 freepage = mapping->a_ops->freepage;
423
424 page_cache_get(new);
425 new->mapping = mapping;
426 new->index = offset;
427
428 spin_lock_irq(&mapping->tree_lock);
Minchan Kime64a7822011-03-22 16:32:44 -0700429 __delete_from_page_cache(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700430 error = radix_tree_insert(&mapping->page_tree, offset, new);
431 BUG_ON(error);
432 mapping->nrpages++;
433 __inc_zone_page_state(new, NR_FILE_PAGES);
434 if (PageSwapBacked(new))
435 __inc_zone_page_state(new, NR_SHMEM);
436 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyukiab936cb2012-01-12 17:17:44 -0800437 /* mem_cgroup codes must not be called under tree_lock */
438 mem_cgroup_replace_page_cache(old, new);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700439 radix_tree_preload_end();
440 if (freepage)
441 freepage(old);
442 page_cache_release(old);
Miklos Szeredief6a3c62011-03-22 16:30:52 -0700443 }
444
445 return error;
446}
447EXPORT_SYMBOL_GPL(replace_page_cache_page);
448
449/**
Nick Piggine2867812008-07-25 19:45:30 -0700450 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700451 * @page: page to add
452 * @mapping: the page's address_space
453 * @offset: page index
454 * @gfp_mask: page allocation mode
455 *
Nick Piggine2867812008-07-25 19:45:30 -0700456 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 * This function does not add the page to the LRU. The caller must do that.
458 */
Nick Piggine2867812008-07-25 19:45:30 -0700459int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400460 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
Nick Piggine2867812008-07-25 19:45:30 -0700462 int error;
463
464 VM_BUG_ON(!PageLocked(page));
Hugh Dickins31475dd2011-08-03 16:21:27 -0700465 VM_BUG_ON(PageSwapBacked(page));
Nick Piggine2867812008-07-25 19:45:30 -0700466
467 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800468 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800469 if (error)
470 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Balbir Singh35c754d2008-02-07 00:14:05 -0800472 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700474 page_cache_get(page);
475 page->mapping = mapping;
476 page->index = offset;
477
Nick Piggin19fd6232008-07-25 19:45:32 -0700478 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700480 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700482 __inc_zone_page_state(page, NR_FILE_PAGES);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700483 spin_unlock_irq(&mapping->tree_lock);
Robert Jarzmikfe0bfaa2013-04-29 15:06:10 -0700484 trace_mm_filemap_add_to_page_cache(page);
Nick Piggine2867812008-07-25 19:45:30 -0700485 } else {
486 page->mapping = NULL;
Hugh Dickinsb85e0ef2011-07-25 17:12:25 -0700487 /* Leave page->index set: truncation relies upon it */
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700488 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700489 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700490 page_cache_release(page);
491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800493 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700494 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800495out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return error;
497}
Nick Piggine2867812008-07-25 19:45:30 -0700498EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499
500int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400501 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700503 int ret;
504
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700505 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
Hugh Dickins31475dd2011-08-03 16:21:27 -0700506 if (ret == 0)
507 lru_cache_add_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 return ret;
509}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300510EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511
Paul Jackson44110fe2006-03-24 03:16:04 -0800512#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700513struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800514{
Miao Xiec0ff7452010-05-24 14:32:08 -0700515 int n;
516 struct page *page;
517
Paul Jackson44110fe2006-03-24 03:16:04 -0800518 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700519 unsigned int cpuset_mems_cookie;
520 do {
521 cpuset_mems_cookie = get_mems_allowed();
522 n = cpuset_mem_spread_node();
523 page = alloc_pages_exact_node(n, gfp, 0);
524 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
525
Miao Xiec0ff7452010-05-24 14:32:08 -0700526 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800527 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700528 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800529}
Nick Piggin2ae88142006-10-28 10:38:23 -0700530EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800531#endif
532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533/*
534 * In order to wait for pages to become available there must be
535 * waitqueues associated with pages. By using a hash table of
536 * waitqueues where the bucket discipline is to maintain all
537 * waiters on the same queue and wake all when any of the pages
538 * become available, and for the woken contexts to check to be
539 * sure the appropriate page became available, this saves space
540 * at a cost of "thundering herd" phenomena during rare hash
541 * collisions.
542 */
543static wait_queue_head_t *page_waitqueue(struct page *page)
544{
545 const struct zone *zone = page_zone(page);
546
547 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
548}
549
550static inline void wake_up_page(struct page *page, int bit)
551{
552 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
553}
554
Harvey Harrison920c7a52008-02-04 22:29:26 -0800555void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
557 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
558
559 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100560 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 TASK_UNINTERRUPTIBLE);
562}
563EXPORT_SYMBOL(wait_on_page_bit);
564
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700565int wait_on_page_bit_killable(struct page *page, int bit_nr)
566{
567 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
568
569 if (!test_bit(bit_nr, &page->flags))
570 return 0;
571
572 return __wait_on_bit(page_waitqueue(page), &wait,
573 sleep_on_page_killable, TASK_KILLABLE);
574}
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100577 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700578 * @page: Page defining the wait queue of interest
579 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100580 *
581 * Add an arbitrary @waiter to the wait queue for the nominated @page.
582 */
583void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
584{
585 wait_queue_head_t *q = page_waitqueue(page);
586 unsigned long flags;
587
588 spin_lock_irqsave(&q->lock, flags);
589 __add_wait_queue(q, waiter);
590 spin_unlock_irqrestore(&q->lock, flags);
591}
592EXPORT_SYMBOL_GPL(add_page_wait_queue);
593
594/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700595 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 * @page: the page
597 *
598 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
599 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
600 * mechananism between PageLocked pages and PageWriteback pages is shared.
601 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
602 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700603 * The mb is necessary to enforce ordering between the clear_bit and the read
604 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800606void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
Nick Piggin8413ac92008-10-18 20:26:59 -0700608 VM_BUG_ON(!PageLocked(page));
609 clear_bit_unlock(PG_locked, &page->flags);
610 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 wake_up_page(page, PG_locked);
612}
613EXPORT_SYMBOL(unlock_page);
614
Randy Dunlap485bb992006-06-23 02:03:49 -0700615/**
616 * end_page_writeback - end writeback against a page
617 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 */
619void end_page_writeback(struct page *page)
620{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700621 if (TestClearPageReclaim(page))
622 rotate_reclaimable_page(page);
623
624 if (!test_clear_page_writeback(page))
625 BUG();
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 smp_mb__after_clear_bit();
628 wake_up_page(page, PG_writeback);
629}
630EXPORT_SYMBOL(end_page_writeback);
631
Randy Dunlap485bb992006-06-23 02:03:49 -0700632/**
633 * __lock_page - get a lock on the page, assuming we need to sleep to get it
634 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800636void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
638 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
639
Jens Axboe7eaceac2011-03-10 08:52:07 +0100640 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 TASK_UNINTERRUPTIBLE);
642}
643EXPORT_SYMBOL(__lock_page);
644
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800645int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500646{
647 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
648
649 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100650 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500651}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300652EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500653
Michel Lespinassed065bd82010-10-26 14:21:57 -0700654int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
655 unsigned int flags)
656{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700657 if (flags & FAULT_FLAG_ALLOW_RETRY) {
658 /*
659 * CAUTION! In this case, mmap_sem is not released
660 * even though return 0.
661 */
662 if (flags & FAULT_FLAG_RETRY_NOWAIT)
663 return 0;
664
665 up_read(&mm->mmap_sem);
666 if (flags & FAULT_FLAG_KILLABLE)
667 wait_on_page_locked_killable(page);
668 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700669 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700670 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700671 } else {
672 if (flags & FAULT_FLAG_KILLABLE) {
673 int ret;
674
675 ret = __lock_page_killable(page);
676 if (ret) {
677 up_read(&mm->mmap_sem);
678 return 0;
679 }
680 } else
681 __lock_page(page);
682 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700683 }
684}
685
Randy Dunlap485bb992006-06-23 02:03:49 -0700686/**
687 * find_get_page - find and get a page reference
688 * @mapping: the address_space to search
689 * @offset: the page index
690 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700691 * Is there a pagecache struct page at the given (mapping, offset) tuple?
692 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 */
Nick Piggina60637c2008-07-25 19:45:31 -0700694struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Nick Piggina60637c2008-07-25 19:45:31 -0700696 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 struct page *page;
698
Nick Piggina60637c2008-07-25 19:45:31 -0700699 rcu_read_lock();
700repeat:
701 page = NULL;
702 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
703 if (pagep) {
704 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800705 if (unlikely(!page))
706 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700707 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700708 if (radix_tree_deref_retry(page))
709 goto repeat;
710 /*
711 * Otherwise, shmem/tmpfs must be storing a swap entry
712 * here as an exceptional entry: so return it without
713 * attempting to raise page count.
714 */
715 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700716 }
Nick Piggina60637c2008-07-25 19:45:31 -0700717 if (!page_cache_get_speculative(page))
718 goto repeat;
719
720 /*
721 * Has the page moved?
722 * This is part of the lockless pagecache protocol. See
723 * include/linux/pagemap.h for details.
724 */
725 if (unlikely(page != *pagep)) {
726 page_cache_release(page);
727 goto repeat;
728 }
729 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800730out:
Nick Piggina60637c2008-07-25 19:45:31 -0700731 rcu_read_unlock();
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return page;
734}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735EXPORT_SYMBOL(find_get_page);
736
Randy Dunlap485bb992006-06-23 02:03:49 -0700737/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700739 * @mapping: the address_space to search
740 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 *
742 * Locates the desired pagecache page, locks it, increments its reference
743 * count and returns its address.
744 *
745 * Returns zero if the page was not present. find_lock_page() may sleep.
746 */
Nick Piggina60637c2008-07-25 19:45:31 -0700747struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748{
749 struct page *page;
750
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700752 page = find_get_page(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700753 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700754 lock_page(page);
755 /* Has the page been truncated? */
756 if (unlikely(page->mapping != mapping)) {
757 unlock_page(page);
758 page_cache_release(page);
759 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Nick Piggina60637c2008-07-25 19:45:31 -0700761 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 return page;
764}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765EXPORT_SYMBOL(find_lock_page);
766
767/**
768 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700769 * @mapping: the page's address_space
770 * @index: the page's index into the mapping
771 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 *
773 * Locates a page in the pagecache. If the page is not present, a new page
774 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
775 * LRU list. The returned page is locked and has its reference count
776 * incremented.
777 *
778 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
779 * allocation!
780 *
781 * find_or_create_page() returns the desired page's address, or zero on
782 * memory exhaustion.
783 */
784struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700785 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786{
Nick Piggineb2be182007-10-16 01:24:57 -0700787 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 int err;
789repeat:
790 page = find_lock_page(mapping, index);
791 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700792 page = __page_cache_alloc(gfp_mask);
793 if (!page)
794 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800795 /*
796 * We want a regular kernel memory (not highmem or DMA etc)
797 * allocation for the radix tree nodes, but we need to honour
798 * the context-specific requirements the caller has asked for.
799 * GFP_RECLAIM_MASK collects those requirements.
800 */
801 err = add_to_page_cache_lru(page, mapping, index,
802 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700803 if (unlikely(err)) {
804 page_cache_release(page);
805 page = NULL;
806 if (err == -EEXIST)
807 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return page;
811}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812EXPORT_SYMBOL(find_or_create_page);
813
814/**
815 * find_get_pages - gang pagecache lookup
816 * @mapping: The address_space to search
817 * @start: The starting page index
818 * @nr_pages: The maximum number of pages
819 * @pages: Where the resulting pages are placed
820 *
821 * find_get_pages() will search for and return a group of up to
822 * @nr_pages pages in the mapping. The pages are placed at @pages.
823 * find_get_pages() takes a reference against the returned pages.
824 *
825 * The search returns a group of mapping-contiguous pages with ascending
826 * indexes. There may be holes in the indices due to not-present pages.
827 *
828 * find_get_pages() returns the number of pages which were found.
829 */
830unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
831 unsigned int nr_pages, struct page **pages)
832{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700833 struct radix_tree_iter iter;
834 void **slot;
835 unsigned ret = 0;
836
837 if (unlikely(!nr_pages))
838 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Nick Piggina60637c2008-07-25 19:45:31 -0700840 rcu_read_lock();
841restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700842 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -0700843 struct page *page;
844repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700845 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -0700846 if (unlikely(!page))
847 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700848
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700849 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700850 if (radix_tree_deref_retry(page)) {
851 /*
852 * Transient condition which can only trigger
853 * when entry at index 0 moves out of or back
854 * to root: none yet gotten, safe to restart.
855 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700856 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700857 goto restart;
858 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700859 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700860 * Otherwise, shmem/tmpfs must be storing a swap entry
861 * here as an exceptional entry: so skip over it -
862 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700863 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700864 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800865 }
Nick Piggina60637c2008-07-25 19:45:31 -0700866
867 if (!page_cache_get_speculative(page))
868 goto repeat;
869
870 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700871 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700872 page_cache_release(page);
873 goto repeat;
874 }
875
876 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700877 if (++ret == nr_pages)
878 break;
Nick Piggina60637c2008-07-25 19:45:31 -0700879 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700880
Nick Piggina60637c2008-07-25 19:45:31 -0700881 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 return ret;
883}
884
Jens Axboeebf43502006-04-27 08:46:01 +0200885/**
886 * find_get_pages_contig - gang contiguous pagecache lookup
887 * @mapping: The address_space to search
888 * @index: The starting page index
889 * @nr_pages: The maximum number of pages
890 * @pages: Where the resulting pages are placed
891 *
892 * find_get_pages_contig() works exactly like find_get_pages(), except
893 * that the returned number of pages are guaranteed to be contiguous.
894 *
895 * find_get_pages_contig() returns the number of pages which were found.
896 */
897unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
898 unsigned int nr_pages, struct page **pages)
899{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700900 struct radix_tree_iter iter;
901 void **slot;
902 unsigned int ret = 0;
903
904 if (unlikely(!nr_pages))
905 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +0200906
Nick Piggina60637c2008-07-25 19:45:31 -0700907 rcu_read_lock();
908restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700909 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -0700910 struct page *page;
911repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700912 page = radix_tree_deref_slot(slot);
913 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -0700914 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700915 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700916
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700917 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700918 if (radix_tree_deref_retry(page)) {
919 /*
920 * Transient condition which can only trigger
921 * when entry at index 0 moves out of or back
922 * to root: none yet gotten, safe to restart.
923 */
924 goto restart;
925 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700926 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700927 * Otherwise, shmem/tmpfs must be storing a swap entry
928 * here as an exceptional entry: so stop looking for
929 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700930 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700931 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700932 }
Nick Piggina60637c2008-07-25 19:45:31 -0700933
Nick Piggina60637c2008-07-25 19:45:31 -0700934 if (!page_cache_get_speculative(page))
935 goto repeat;
936
937 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700938 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700939 page_cache_release(page);
940 goto repeat;
941 }
942
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800943 /*
944 * must check mapping and index after taking the ref.
945 * otherwise we can get both false positives and false
946 * negatives, which is just confusing to the caller.
947 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700948 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800949 page_cache_release(page);
950 break;
951 }
952
Nick Piggina60637c2008-07-25 19:45:31 -0700953 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700954 if (++ret == nr_pages)
955 break;
Jens Axboeebf43502006-04-27 08:46:01 +0200956 }
Nick Piggina60637c2008-07-25 19:45:31 -0700957 rcu_read_unlock();
958 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200959}
David Howellsef71c152007-05-09 02:33:44 -0700960EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200961
Randy Dunlap485bb992006-06-23 02:03:49 -0700962/**
963 * find_get_pages_tag - find and return pages that match @tag
964 * @mapping: the address_space to search
965 * @index: the starting page index
966 * @tag: the tag index
967 * @nr_pages: the maximum number of pages
968 * @pages: where the resulting pages are placed
969 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700971 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 */
973unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
974 int tag, unsigned int nr_pages, struct page **pages)
975{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700976 struct radix_tree_iter iter;
977 void **slot;
978 unsigned ret = 0;
979
980 if (unlikely(!nr_pages))
981 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Nick Piggina60637c2008-07-25 19:45:31 -0700983 rcu_read_lock();
984restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700985 radix_tree_for_each_tagged(slot, &mapping->page_tree,
986 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -0700987 struct page *page;
988repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700989 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -0700990 if (unlikely(!page))
991 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700992
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700993 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700994 if (radix_tree_deref_retry(page)) {
995 /*
996 * Transient condition which can only trigger
997 * when entry at index 0 moves out of or back
998 * to root: none yet gotten, safe to restart.
999 */
1000 goto restart;
1001 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001002 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001003 * This function is never used on a shmem/tmpfs
1004 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001005 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001006 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001007 }
Nick Piggina60637c2008-07-25 19:45:31 -07001008
1009 if (!page_cache_get_speculative(page))
1010 goto repeat;
1011
1012 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001013 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001014 page_cache_release(page);
1015 goto repeat;
1016 }
1017
1018 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001019 if (++ret == nr_pages)
1020 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001021 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001022
Nick Piggina60637c2008-07-25 19:45:31 -07001023 rcu_read_unlock();
1024
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 if (ret)
1026 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 return ret;
1029}
David Howellsef71c152007-05-09 02:33:44 -07001030EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
Randy Dunlap485bb992006-06-23 02:03:49 -07001032/**
1033 * grab_cache_page_nowait - returns locked page at given index in given cache
1034 * @mapping: target address_space
1035 * @index: the page index
1036 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001037 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 * This is intended for speculative data generators, where the data can
1039 * be regenerated if the page couldn't be grabbed. This routine should
1040 * be safe to call while holding the lock for another page.
1041 *
1042 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1043 * and deadlock against the caller's locked page.
1044 */
1045struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001046grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047{
1048 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
1050 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001051 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 return page;
1053 page_cache_release(page);
1054 return NULL;
1055 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001056 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001057 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 page_cache_release(page);
1059 page = NULL;
1060 }
1061 return page;
1062}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063EXPORT_SYMBOL(grab_cache_page_nowait);
1064
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001065/*
1066 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1067 * a _large_ part of the i/o request. Imagine the worst scenario:
1068 *
1069 * ---R__________________________________________B__________
1070 * ^ reading here ^ bad block(assume 4k)
1071 *
1072 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1073 * => failing the whole request => read(R) => read(R+1) =>
1074 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1075 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1076 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1077 *
1078 * It is going insane. Fix it by quickly scaling down the readahead size.
1079 */
1080static void shrink_readahead_size_eio(struct file *filp,
1081 struct file_ra_state *ra)
1082{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001083 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001084}
1085
Randy Dunlap485bb992006-06-23 02:03:49 -07001086/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001087 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001088 * @filp: the file to read
1089 * @ppos: current file position
1090 * @desc: read_descriptor
1091 * @actor: read method
1092 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001094 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 *
1096 * This is really ugly. But the goto's actually try to clarify some
1097 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001099static void do_generic_file_read(struct file *filp, loff_t *ppos,
1100 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001102 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001104 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001105 pgoff_t index;
1106 pgoff_t last_index;
1107 pgoff_t prev_index;
1108 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001109 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001113 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1114 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1116 offset = *ppos & ~PAGE_CACHE_MASK;
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 for (;;) {
1119 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001120 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001121 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 unsigned long nr, ret;
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125find_page:
1126 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001127 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001128 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001129 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001130 index, last_index - index);
1131 page = find_get_page(mapping, index);
1132 if (unlikely(page == NULL))
1133 goto no_cached_page;
1134 }
1135 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001136 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001137 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001138 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001140 if (!PageUptodate(page)) {
1141 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1142 !mapping->a_ops->is_partially_uptodate)
1143 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001144 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001145 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001146 /* Did it get truncated before we got the lock? */
1147 if (!page->mapping)
1148 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001149 if (!mapping->a_ops->is_partially_uptodate(page,
1150 desc, offset))
1151 goto page_not_up_to_date_locked;
1152 unlock_page(page);
1153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001155 /*
1156 * i_size must be checked after we know the page is Uptodate.
1157 *
1158 * Checking i_size after the check allows us to calculate
1159 * the correct value for "nr", which means the zero-filled
1160 * part of the page is not copied back to userspace (unless
1161 * another truncate extends the file - this is desired though).
1162 */
1163
1164 isize = i_size_read(inode);
1165 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1166 if (unlikely(!isize || index > end_index)) {
1167 page_cache_release(page);
1168 goto out;
1169 }
1170
1171 /* nr is the maximum number of bytes to copy from this page */
1172 nr = PAGE_CACHE_SIZE;
1173 if (index == end_index) {
1174 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1175 if (nr <= offset) {
1176 page_cache_release(page);
1177 goto out;
1178 }
1179 }
1180 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182 /* If users can be writing to this page using arbitrary
1183 * virtual addresses, take care about potential aliasing
1184 * before reading the page on the kernel side.
1185 */
1186 if (mapping_writably_mapped(mapping))
1187 flush_dcache_page(page);
1188
1189 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001190 * When a sequential read accesses a page several times,
1191 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 */
Jan Karaec0f1632007-05-06 14:49:25 -07001193 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 mark_page_accessed(page);
1195 prev_index = index;
1196
1197 /*
1198 * Ok, we have the page, and it's up-to-date, so
1199 * now we can copy it to user space...
1200 *
1201 * The actor routine returns how many bytes were actually used..
1202 * NOTE! This may not be the same as how much of a user buffer
1203 * we filled up (we may be padding etc), so we can only update
1204 * "pos" here (the actor routine has to update the user buffer
1205 * pointers and the remaining count).
1206 */
1207 ret = actor(desc, page, offset, nr);
1208 offset += ret;
1209 index += offset >> PAGE_CACHE_SHIFT;
1210 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001211 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213 page_cache_release(page);
1214 if (ret == nr && desc->count)
1215 continue;
1216 goto out;
1217
1218page_not_up_to_date:
1219 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001220 error = lock_page_killable(page);
1221 if (unlikely(error))
1222 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001224page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001225 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 if (!page->mapping) {
1227 unlock_page(page);
1228 page_cache_release(page);
1229 continue;
1230 }
1231
1232 /* Did somebody else fill it already? */
1233 if (PageUptodate(page)) {
1234 unlock_page(page);
1235 goto page_ok;
1236 }
1237
1238readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001239 /*
1240 * A previous I/O error may have been due to temporary
1241 * failures, eg. multipath errors.
1242 * PG_error will be set again if readpage fails.
1243 */
1244 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 /* Start the actual read. The read will unlock the page. */
1246 error = mapping->a_ops->readpage(filp, page);
1247
Zach Brown994fc28c2005-12-15 14:28:17 -08001248 if (unlikely(error)) {
1249 if (error == AOP_TRUNCATED_PAGE) {
1250 page_cache_release(page);
1251 goto find_page;
1252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001257 error = lock_page_killable(page);
1258 if (unlikely(error))
1259 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 if (!PageUptodate(page)) {
1261 if (page->mapping == NULL) {
1262 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001263 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 */
1265 unlock_page(page);
1266 page_cache_release(page);
1267 goto find_page;
1268 }
1269 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001270 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001271 error = -EIO;
1272 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274 unlock_page(page);
1275 }
1276
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 goto page_ok;
1278
1279readpage_error:
1280 /* UHHUH! A synchronous read error occurred. Report it */
1281 desc->error = error;
1282 page_cache_release(page);
1283 goto out;
1284
1285no_cached_page:
1286 /*
1287 * Ok, it wasn't cached, so we need to create a new
1288 * page..
1289 */
Nick Piggineb2be182007-10-16 01:24:57 -07001290 page = page_cache_alloc_cold(mapping);
1291 if (!page) {
1292 desc->error = -ENOMEM;
1293 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
Nick Piggineb2be182007-10-16 01:24:57 -07001295 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 index, GFP_KERNEL);
1297 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001298 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 if (error == -EEXIST)
1300 goto find_page;
1301 desc->error = error;
1302 goto out;
1303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 goto readpage;
1305 }
1306
1307out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001308 ra->prev_pos = prev_index;
1309 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1310 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001312 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001313 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315
1316int file_read_actor(read_descriptor_t *desc, struct page *page,
1317 unsigned long offset, unsigned long size)
1318{
1319 char *kaddr;
1320 unsigned long left, count = desc->count;
1321
1322 if (size > count)
1323 size = count;
1324
1325 /*
1326 * Faults on the destination of a read are common, so do it before
1327 * taking the kmap.
1328 */
1329 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001330 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 left = __copy_to_user_inatomic(desc->arg.buf,
1332 kaddr + offset, size);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001333 kunmap_atomic(kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 if (left == 0)
1335 goto success;
1336 }
1337
1338 /* Do it the slow way */
1339 kaddr = kmap(page);
1340 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1341 kunmap(page);
1342
1343 if (left) {
1344 size -= left;
1345 desc->error = -EFAULT;
1346 }
1347success:
1348 desc->count = count - size;
1349 desc->written += size;
1350 desc->arg.buf += size;
1351 return size;
1352}
1353
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001354/*
1355 * Performs necessary checks before doing a write
1356 * @iov: io vector request
1357 * @nr_segs: number of segments in the iovec
1358 * @count: number of bytes to write
1359 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1360 *
1361 * Adjust number of segments and amount of bytes to write (nr_segs should be
1362 * properly initialized first). Returns appropriate error code that caller
1363 * should return or zero in case that write should be allowed.
1364 */
1365int generic_segment_checks(const struct iovec *iov,
1366 unsigned long *nr_segs, size_t *count, int access_flags)
1367{
1368 unsigned long seg;
1369 size_t cnt = 0;
1370 for (seg = 0; seg < *nr_segs; seg++) {
1371 const struct iovec *iv = &iov[seg];
1372
1373 /*
1374 * If any segment has a negative length, or the cumulative
1375 * length ever wraps negative then return -EINVAL.
1376 */
1377 cnt += iv->iov_len;
1378 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1379 return -EINVAL;
1380 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1381 continue;
1382 if (seg == 0)
1383 return -EFAULT;
1384 *nr_segs = seg;
1385 cnt -= iv->iov_len; /* This segment is no good */
1386 break;
1387 }
1388 *count = cnt;
1389 return 0;
1390}
1391EXPORT_SYMBOL(generic_segment_checks);
1392
Randy Dunlap485bb992006-06-23 02:03:49 -07001393/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001394 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001395 * @iocb: kernel I/O control block
1396 * @iov: io vector request
1397 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001398 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001399 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 * This is the "read()" routine for all filesystems
1401 * that can use the page cache directly.
1402 */
1403ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001404generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1405 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
1407 struct file *filp = iocb->ki_filp;
1408 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001409 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001411 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
1413 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001414 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1415 if (retval)
1416 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
1418 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1419 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001420 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 struct address_space *mapping;
1422 struct inode *inode;
1423
1424 mapping = filp->f_mapping;
1425 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (!count)
1427 goto out; /* skip atime */
1428 size = i_size_read(inode);
1429 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001430 retval = filemap_write_and_wait_range(mapping, pos,
1431 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001432 if (!retval) {
1433 retval = mapping->a_ops->direct_IO(READ, iocb,
1434 iov, pos, nr_segs);
1435 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001436 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001438 count -= retval;
1439 }
1440
1441 /*
1442 * Btrfs can have a short DIO read if we encounter
1443 * compressed extents, so if there was an error, or if
1444 * we've already read everything we wanted to, or if
1445 * there was a short read because we hit EOF, go ahead
1446 * and return. Otherwise fallthrough to buffered io for
1447 * the rest of the read.
1448 */
1449 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001450 file_accessed(filp);
1451 goto out;
1452 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001453 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 }
1455
Josef Bacik66f998f2010-05-23 11:00:54 -04001456 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001457 for (seg = 0; seg < nr_segs; seg++) {
1458 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001459 loff_t offset = 0;
1460
1461 /*
1462 * If we did a short DIO read we need to skip the section of the
1463 * iov that we've already read data into.
1464 */
1465 if (count) {
1466 if (count > iov[seg].iov_len) {
1467 count -= iov[seg].iov_len;
1468 continue;
1469 }
1470 offset = count;
1471 count = 0;
1472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
Hugh Dickins11fa9772008-07-23 21:27:34 -07001474 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001475 desc.arg.buf = iov[seg].iov_base + offset;
1476 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001477 if (desc.count == 0)
1478 continue;
1479 desc.error = 0;
1480 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1481 retval += desc.written;
1482 if (desc.error) {
1483 retval = retval ?: desc.error;
1484 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001486 if (desc.count > 0)
1487 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 }
1489out:
1490 return retval;
1491}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492EXPORT_SYMBOL(generic_file_aio_read);
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001495/**
1496 * page_cache_read - adds requested page to the page cache if not already there
1497 * @file: file to read
1498 * @offset: page index
1499 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 * This adds the requested page to the page cache if it isn't already there,
1501 * and schedules an I/O to read in its contents from disk.
1502 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001503static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504{
1505 struct address_space *mapping = file->f_mapping;
1506 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001507 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
Zach Brown994fc28c2005-12-15 14:28:17 -08001509 do {
1510 page = page_cache_alloc_cold(mapping);
1511 if (!page)
1512 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Zach Brown994fc28c2005-12-15 14:28:17 -08001514 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1515 if (ret == 0)
1516 ret = mapping->a_ops->readpage(file, page);
1517 else if (ret == -EEXIST)
1518 ret = 0; /* losing race to add is OK */
1519
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Zach Brown994fc28c2005-12-15 14:28:17 -08001522 } while (ret == AOP_TRUNCATED_PAGE);
1523
1524 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525}
1526
1527#define MMAP_LOTSAMISS (100)
1528
Linus Torvaldsef00e082009-06-16 15:31:25 -07001529/*
1530 * Synchronous readahead happens when we don't even find
1531 * a page in the page cache at all.
1532 */
1533static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1534 struct file_ra_state *ra,
1535 struct file *file,
1536 pgoff_t offset)
1537{
1538 unsigned long ra_pages;
1539 struct address_space *mapping = file->f_mapping;
1540
1541 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001542 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001543 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001544 if (!ra->ra_pages)
1545 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001546
Joe Perches64363aa2013-07-08 16:00:18 -07001547 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001548 page_cache_sync_readahead(mapping, ra, file, offset,
1549 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001550 return;
1551 }
1552
Andi Kleen207d04b2011-05-24 17:12:29 -07001553 /* Avoid banging the cache line if not needed */
1554 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001555 ra->mmap_miss++;
1556
1557 /*
1558 * Do we miss much more than hit in this file? If so,
1559 * stop bothering with read-ahead. It will only hurt.
1560 */
1561 if (ra->mmap_miss > MMAP_LOTSAMISS)
1562 return;
1563
Wu Fengguangd30a1102009-06-16 15:31:30 -07001564 /*
1565 * mmap read-around
1566 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001567 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001568 ra->start = max_t(long, 0, offset - ra_pages / 2);
1569 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001570 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001571 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001572}
1573
1574/*
1575 * Asynchronous readahead happens when we find the page and PG_readahead,
1576 * so we want to possibly extend the readahead further..
1577 */
1578static void do_async_mmap_readahead(struct vm_area_struct *vma,
1579 struct file_ra_state *ra,
1580 struct file *file,
1581 struct page *page,
1582 pgoff_t offset)
1583{
1584 struct address_space *mapping = file->f_mapping;
1585
1586 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001587 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001588 return;
1589 if (ra->mmap_miss > 0)
1590 ra->mmap_miss--;
1591 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001592 page_cache_async_readahead(mapping, ra, file,
1593 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001594}
1595
Randy Dunlap485bb992006-06-23 02:03:49 -07001596/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001597 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001598 * @vma: vma in which the fault was taken
1599 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001600 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001601 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 * mapped memory region to read in file data during a page fault.
1603 *
1604 * The goto's are kind of ugly, but this streamlines the normal case of having
1605 * it in the page cache, and handles the special cases reasonably without
1606 * having a lot of duplicated code.
1607 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001608int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609{
1610 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001611 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 struct address_space *mapping = file->f_mapping;
1613 struct file_ra_state *ra = &file->f_ra;
1614 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001615 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001617 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001618 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001621 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001622 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 * Do we have something in the page cache already?
1626 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001627 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001628 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001630 * We found the page, so try async readahead before
1631 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001633 do_async_mmap_readahead(vma, ra, file, page, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001634 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001635 /* No page in the page cache at all */
1636 do_sync_mmap_readahead(vma, ra, file, offset);
1637 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001638 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001639 ret = VM_FAULT_MAJOR;
1640retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001641 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 if (!page)
1643 goto no_cached_page;
1644 }
1645
Michel Lespinassed88c0922010-11-02 13:05:18 -07001646 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1647 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001648 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001649 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001650
1651 /* Did it get truncated? */
1652 if (unlikely(page->mapping != mapping)) {
1653 unlock_page(page);
1654 put_page(page);
1655 goto retry_find;
1656 }
1657 VM_BUG_ON(page->index != offset);
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001660 * We have a locked page in the page cache, now we need to check
1661 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 */
Nick Piggind00806b2007-07-19 01:46:57 -07001663 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 goto page_not_uptodate;
1665
Linus Torvaldsef00e082009-06-16 15:31:25 -07001666 /*
1667 * Found the page and have a reference on it.
1668 * We must recheck i_size under page lock.
1669 */
Nick Piggind00806b2007-07-19 01:46:57 -07001670 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001671 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001672 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001673 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001674 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001675 }
1676
Nick Piggind0217ac2007-07-19 01:47:03 -07001677 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001678 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680no_cached_page:
1681 /*
1682 * We're only likely to ever get here if MADV_RANDOM is in
1683 * effect.
1684 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001685 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686
1687 /*
1688 * The page we want has now been added to the page cache.
1689 * In the unlikely event that someone removed it in the
1690 * meantime, we'll just come back here and read it again.
1691 */
1692 if (error >= 0)
1693 goto retry_find;
1694
1695 /*
1696 * An error return from page_cache_read can result if the
1697 * system is low on memory, or a problem occurs while trying
1698 * to schedule I/O.
1699 */
1700 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001701 return VM_FAULT_OOM;
1702 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 /*
1706 * Umm, take care of errors if the page isn't up-to-date.
1707 * Try to re-read it _once_. We do this synchronously,
1708 * because there really aren't any performance issues here
1709 * and we need to check for errors.
1710 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001712 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001713 if (!error) {
1714 wait_on_page_locked(page);
1715 if (!PageUptodate(page))
1716 error = -EIO;
1717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001719
1720 if (!error || error == AOP_TRUNCATED_PAGE)
1721 goto retry_find;
1722
1723 /* Things didn't work out. Return zero to tell the mm layer so. */
1724 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001725 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001726}
1727EXPORT_SYMBOL(filemap_fault);
1728
Jan Kara4fcf1c62012-06-12 16:20:29 +02001729int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1730{
1731 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05001732 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001733 int ret = VM_FAULT_LOCKED;
1734
Jan Kara14da9202012-06-12 16:20:37 +02001735 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001736 file_update_time(vma->vm_file);
1737 lock_page(page);
1738 if (page->mapping != inode->i_mapping) {
1739 unlock_page(page);
1740 ret = VM_FAULT_NOPAGE;
1741 goto out;
1742 }
Jan Kara14da9202012-06-12 16:20:37 +02001743 /*
1744 * We mark the page dirty already here so that when freeze is in
1745 * progress, we are guaranteed that writeback during freezing will
1746 * see the dirty page and writeprotect it again.
1747 */
1748 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08001749 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001750out:
Jan Kara14da9202012-06-12 16:20:37 +02001751 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001752 return ret;
1753}
1754EXPORT_SYMBOL(filemap_page_mkwrite);
1755
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001756const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001757 .fault = filemap_fault,
Jan Kara4fcf1c62012-06-12 16:20:29 +02001758 .page_mkwrite = filemap_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001759 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760};
1761
1762/* This is used for a general mmap of a disk file */
1763
1764int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1765{
1766 struct address_space *mapping = file->f_mapping;
1767
1768 if (!mapping->a_ops->readpage)
1769 return -ENOEXEC;
1770 file_accessed(file);
1771 vma->vm_ops = &generic_file_vm_ops;
1772 return 0;
1773}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775/*
1776 * This is for filesystems which do not implement ->writepage.
1777 */
1778int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1779{
1780 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1781 return -EINVAL;
1782 return generic_file_mmap(file, vma);
1783}
1784#else
1785int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1786{
1787 return -ENOSYS;
1788}
1789int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1790{
1791 return -ENOSYS;
1792}
1793#endif /* CONFIG_MMU */
1794
1795EXPORT_SYMBOL(generic_file_mmap);
1796EXPORT_SYMBOL(generic_file_readonly_mmap);
1797
Nick Piggin6fe69002007-05-06 14:49:04 -07001798static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001799 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001800 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001801 void *data,
1802 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
Nick Piggineb2be182007-10-16 01:24:57 -07001804 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 int err;
1806repeat:
1807 page = find_get_page(mapping, index);
1808 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001809 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001810 if (!page)
1811 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001812 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07001813 if (unlikely(err)) {
1814 page_cache_release(page);
1815 if (err == -EEXIST)
1816 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 return ERR_PTR(err);
1819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 err = filler(data, page);
1821 if (err < 0) {
1822 page_cache_release(page);
1823 page = ERR_PTR(err);
1824 }
1825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 return page;
1827}
1828
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001829static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001830 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001831 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001832 void *data,
1833 gfp_t gfp)
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835{
1836 struct page *page;
1837 int err;
1838
1839retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001840 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001842 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 if (PageUptodate(page))
1844 goto out;
1845
1846 lock_page(page);
1847 if (!page->mapping) {
1848 unlock_page(page);
1849 page_cache_release(page);
1850 goto retry;
1851 }
1852 if (PageUptodate(page)) {
1853 unlock_page(page);
1854 goto out;
1855 }
1856 err = filler(data, page);
1857 if (err < 0) {
1858 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001859 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 }
David Howellsc855ff32007-05-09 13:42:20 +01001861out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001862 mark_page_accessed(page);
1863 return page;
1864}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001865
1866/**
1867 * read_cache_page_async - read into page cache, fill it if needed
1868 * @mapping: the page's address_space
1869 * @index: the page index
1870 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001871 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001872 *
1873 * Same as read_cache_page, but don't wait for page to become unlocked
1874 * after submitting it to the filler.
1875 *
1876 * Read into the page cache. If a page already exists, and PageUptodate() is
1877 * not set, try to fill the page but don't wait for it to become unlocked.
1878 *
1879 * If the page does not get brought uptodate, return -EIO.
1880 */
1881struct page *read_cache_page_async(struct address_space *mapping,
1882 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001883 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001884 void *data)
1885{
1886 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1887}
Nick Piggin6fe69002007-05-06 14:49:04 -07001888EXPORT_SYMBOL(read_cache_page_async);
1889
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001890static struct page *wait_on_page_read(struct page *page)
1891{
1892 if (!IS_ERR(page)) {
1893 wait_on_page_locked(page);
1894 if (!PageUptodate(page)) {
1895 page_cache_release(page);
1896 page = ERR_PTR(-EIO);
1897 }
1898 }
1899 return page;
1900}
1901
1902/**
1903 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1904 * @mapping: the page's address_space
1905 * @index: the page index
1906 * @gfp: the page allocator flags to use if allocating
1907 *
1908 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001909 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001910 *
1911 * If the page does not get brought uptodate, return -EIO.
1912 */
1913struct page *read_cache_page_gfp(struct address_space *mapping,
1914 pgoff_t index,
1915 gfp_t gfp)
1916{
1917 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1918
1919 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1920}
1921EXPORT_SYMBOL(read_cache_page_gfp);
1922
Nick Piggin6fe69002007-05-06 14:49:04 -07001923/**
1924 * read_cache_page - read into page cache, fill it if needed
1925 * @mapping: the page's address_space
1926 * @index: the page index
1927 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001928 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07001929 *
1930 * Read into the page cache. If a page already exists, and PageUptodate() is
1931 * not set, try to fill the page then wait for it to become unlocked.
1932 *
1933 * If the page does not get brought uptodate, return -EIO.
1934 */
1935struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001936 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001937 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07001938 void *data)
1939{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001940 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942EXPORT_SYMBOL(read_cache_page);
1943
Nick Piggin2f718ff2007-10-16 01:24:59 -07001944static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 const struct iovec *iov, size_t base, size_t bytes)
1946{
Ingo Molnarf1800532009-03-02 11:00:57 +01001947 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948
1949 while (bytes) {
1950 char __user *buf = iov->iov_base + base;
1951 int copy = min(bytes, iov->iov_len - base);
1952
1953 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001954 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 copied += copy;
1956 bytes -= copy;
1957 vaddr += copy;
1958 iov++;
1959
NeilBrown01408c42006-06-25 05:47:58 -07001960 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 }
1963 return copied - left;
1964}
1965
1966/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001967 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001968 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001969 * bytes which were copied.
1970 */
1971size_t iov_iter_copy_from_user_atomic(struct page *page,
1972 struct iov_iter *i, unsigned long offset, size_t bytes)
1973{
1974 char *kaddr;
1975 size_t copied;
1976
1977 BUG_ON(!in_atomic());
Cong Wang9b04c5f2011-11-25 23:14:39 +08001978 kaddr = kmap_atomic(page);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001979 if (likely(i->nr_segs == 1)) {
1980 int left;
1981 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001982 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001983 copied = bytes - left;
1984 } else {
1985 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1986 i->iov, i->iov_offset, bytes);
1987 }
Cong Wang9b04c5f2011-11-25 23:14:39 +08001988 kunmap_atomic(kaddr);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001989
1990 return copied;
1991}
Nick Piggin89e10782007-10-16 01:25:07 -07001992EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001993
1994/*
1995 * This has the same sideeffects and return value as
1996 * iov_iter_copy_from_user_atomic().
1997 * The difference is that it attempts to resolve faults.
1998 * Page must not be locked.
1999 */
2000size_t iov_iter_copy_from_user(struct page *page,
2001 struct iov_iter *i, unsigned long offset, size_t bytes)
2002{
2003 char *kaddr;
2004 size_t copied;
2005
2006 kaddr = kmap(page);
2007 if (likely(i->nr_segs == 1)) {
2008 int left;
2009 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002010 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002011 copied = bytes - left;
2012 } else {
2013 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2014 i->iov, i->iov_offset, bytes);
2015 }
2016 kunmap(page);
2017 return copied;
2018}
Nick Piggin89e10782007-10-16 01:25:07 -07002019EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002020
Nick Pigginf7009262008-03-10 11:43:59 -07002021void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002022{
Nick Pigginf7009262008-03-10 11:43:59 -07002023 BUG_ON(i->count < bytes);
2024
Nick Piggin2f718ff2007-10-16 01:24:59 -07002025 if (likely(i->nr_segs == 1)) {
2026 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07002027 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002028 } else {
2029 const struct iovec *iov = i->iov;
2030 size_t base = i->iov_offset;
Jeff Layton39be79c2011-10-27 23:53:08 +02002031 unsigned long nr_segs = i->nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002032
Nick Piggin124d3b72008-02-02 15:01:17 +01002033 /*
2034 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07002035 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01002036 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07002037 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07002038 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002039
Nick Pigginf7009262008-03-10 11:43:59 -07002040 copy = min(bytes, iov->iov_len - base);
2041 BUG_ON(!i->count || i->count < copy);
2042 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002043 bytes -= copy;
2044 base += copy;
2045 if (iov->iov_len == base) {
2046 iov++;
Jeff Layton39be79c2011-10-27 23:53:08 +02002047 nr_segs--;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002048 base = 0;
2049 }
2050 }
2051 i->iov = iov;
2052 i->iov_offset = base;
Jeff Layton39be79c2011-10-27 23:53:08 +02002053 i->nr_segs = nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002054 }
2055}
Nick Piggin89e10782007-10-16 01:25:07 -07002056EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002057
Nick Pigginafddba42007-10-16 01:25:01 -07002058/*
2059 * Fault in the first iovec of the given iov_iter, to a maximum length
2060 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2061 * accessed (ie. because it is an invalid address).
2062 *
2063 * writev-intensive code may want this to prefault several iovecs -- that
2064 * would be possible (callers must not rely on the fact that _only_ the
2065 * first iovec will be faulted with the current implementation).
2066 */
2067int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002068{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002069 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002070 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2071 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002072}
Nick Piggin89e10782007-10-16 01:25:07 -07002073EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002074
2075/*
2076 * Return the count of just the current iov_iter segment.
2077 */
Maxim Patlasovd28574e2012-10-26 19:50:04 +04002078size_t iov_iter_single_seg_count(const struct iov_iter *i)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002079{
2080 const struct iovec *iov = i->iov;
2081 if (i->nr_segs == 1)
2082 return i->count;
2083 else
2084 return min(i->count, iov->iov_len - i->iov_offset);
2085}
Nick Piggin89e10782007-10-16 01:25:07 -07002086EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002087
2088/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 * Performs necessary checks before doing a write
2090 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002091 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 * Returns appropriate error code that caller should return or
2093 * zero in case that write should be allowed.
2094 */
2095inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2096{
2097 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002098 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099
2100 if (unlikely(*pos < 0))
2101 return -EINVAL;
2102
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 if (!isblk) {
2104 /* FIXME: this is for backwards compatibility with 2.4 */
2105 if (file->f_flags & O_APPEND)
2106 *pos = i_size_read(inode);
2107
2108 if (limit != RLIM_INFINITY) {
2109 if (*pos >= limit) {
2110 send_sig(SIGXFSZ, current, 0);
2111 return -EFBIG;
2112 }
2113 if (*count > limit - (typeof(limit))*pos) {
2114 *count = limit - (typeof(limit))*pos;
2115 }
2116 }
2117 }
2118
2119 /*
2120 * LFS rule
2121 */
2122 if (unlikely(*pos + *count > MAX_NON_LFS &&
2123 !(file->f_flags & O_LARGEFILE))) {
2124 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 return -EFBIG;
2126 }
2127 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2128 *count = MAX_NON_LFS - (unsigned long)*pos;
2129 }
2130 }
2131
2132 /*
2133 * Are we about to exceed the fs block limit ?
2134 *
2135 * If we have written data it becomes a short write. If we have
2136 * exceeded without writing data we send a signal and return EFBIG.
2137 * Linus frestrict idea will clean these up nicely..
2138 */
2139 if (likely(!isblk)) {
2140 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2141 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 return -EFBIG;
2143 }
2144 /* zero-length writes at ->s_maxbytes are OK */
2145 }
2146
2147 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2148 *count = inode->i_sb->s_maxbytes - *pos;
2149 } else {
David Howells93614012006-09-30 20:45:40 +02002150#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 loff_t isize;
2152 if (bdev_read_only(I_BDEV(inode)))
2153 return -EPERM;
2154 isize = i_size_read(inode);
2155 if (*pos >= isize) {
2156 if (*count || *pos > isize)
2157 return -ENOSPC;
2158 }
2159
2160 if (*pos + *count > isize)
2161 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002162#else
2163 return -EPERM;
2164#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 }
2166 return 0;
2167}
2168EXPORT_SYMBOL(generic_write_checks);
2169
Nick Pigginafddba42007-10-16 01:25:01 -07002170int pagecache_write_begin(struct file *file, struct address_space *mapping,
2171 loff_t pos, unsigned len, unsigned flags,
2172 struct page **pagep, void **fsdata)
2173{
2174 const struct address_space_operations *aops = mapping->a_ops;
2175
Nick Piggin4e02ed42008-10-29 14:00:55 -07002176 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002177 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002178}
2179EXPORT_SYMBOL(pagecache_write_begin);
2180
2181int pagecache_write_end(struct file *file, struct address_space *mapping,
2182 loff_t pos, unsigned len, unsigned copied,
2183 struct page *page, void *fsdata)
2184{
2185 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002186
Nick Piggin4e02ed42008-10-29 14:00:55 -07002187 mark_page_accessed(page);
2188 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002189}
2190EXPORT_SYMBOL(pagecache_write_end);
2191
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192ssize_t
2193generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2194 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2195 size_t count, size_t ocount)
2196{
2197 struct file *file = iocb->ki_filp;
2198 struct address_space *mapping = file->f_mapping;
2199 struct inode *inode = mapping->host;
2200 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002201 size_t write_len;
2202 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203
2204 if (count != ocount)
2205 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2206
Christoph Hellwiga969e902008-07-23 21:27:04 -07002207 write_len = iov_length(iov, *nr_segs);
2208 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002209
Nick Piggin48b47c52009-01-06 14:40:22 -08002210 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002211 if (written)
2212 goto out;
2213
2214 /*
2215 * After a write we want buffered reads to be sure to go to disk to get
2216 * the new data. We invalidate clean cached page from the region we're
2217 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002218 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002219 */
2220 if (mapping->nrpages) {
2221 written = invalidate_inode_pages2_range(mapping,
2222 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002223 /*
2224 * If a page can not be invalidated, return 0 to fall back
2225 * to buffered write.
2226 */
2227 if (written) {
2228 if (written == -EBUSY)
2229 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002230 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002231 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002232 }
2233
2234 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2235
2236 /*
2237 * Finally, try again to invalidate clean pages which might have been
2238 * cached by non-direct readahead, or faulted in by get_user_pages()
2239 * if the source of the write was an mmap'ed region of the file
2240 * we're writing. Either one is a pretty crazy thing to do,
2241 * so we don't support it 100%. If this invalidation
2242 * fails, tough, the write still worked...
2243 */
2244 if (mapping->nrpages) {
2245 invalidate_inode_pages2_range(mapping,
2246 pos >> PAGE_CACHE_SHIFT, end);
2247 }
2248
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002250 pos += written;
2251 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2252 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 mark_inode_dirty(inode);
2254 }
Namhyung Kim01166512010-10-26 14:21:58 -07002255 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002257out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 return written;
2259}
2260EXPORT_SYMBOL(generic_file_direct_write);
2261
Nick Piggineb2be182007-10-16 01:24:57 -07002262/*
2263 * Find or create a page at the given pagecache position. Return the locked
2264 * page. This function is specifically for buffered writes.
2265 */
Nick Piggin54566b22009-01-04 12:00:53 -08002266struct page *grab_cache_page_write_begin(struct address_space *mapping,
2267 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002268{
2269 int status;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002270 gfp_t gfp_mask;
Nick Piggineb2be182007-10-16 01:24:57 -07002271 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002272 gfp_t gfp_notmask = 0;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002273
Fengguang Wu1010bb12012-03-21 16:34:08 -07002274 gfp_mask = mapping_gfp_mask(mapping);
2275 if (mapping_cap_account_dirty(mapping))
2276 gfp_mask |= __GFP_WRITE;
Nick Piggin54566b22009-01-04 12:00:53 -08002277 if (flags & AOP_FLAG_NOFS)
2278 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002279repeat:
2280 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002281 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002282 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002283
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002284 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002285 if (!page)
2286 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002287 status = add_to_page_cache_lru(page, mapping, index,
2288 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002289 if (unlikely(status)) {
2290 page_cache_release(page);
2291 if (status == -EEXIST)
2292 goto repeat;
2293 return NULL;
2294 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002295found:
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002296 wait_for_stable_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002297 return page;
2298}
Nick Piggin54566b22009-01-04 12:00:53 -08002299EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002300
Nick Pigginafddba42007-10-16 01:25:01 -07002301static ssize_t generic_perform_write(struct file *file,
2302 struct iov_iter *i, loff_t pos)
2303{
2304 struct address_space *mapping = file->f_mapping;
2305 const struct address_space_operations *a_ops = mapping->a_ops;
2306 long status = 0;
2307 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002308 unsigned int flags = 0;
2309
2310 /*
2311 * Copies from kernel address space cannot fail (NFSD is a big user).
2312 */
2313 if (segment_eq(get_fs(), KERNEL_DS))
2314 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002315
2316 do {
2317 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002318 unsigned long offset; /* Offset into pagecache page */
2319 unsigned long bytes; /* Bytes to write to page */
2320 size_t copied; /* Bytes copied from user */
2321 void *fsdata;
2322
2323 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002324 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2325 iov_iter_count(i));
2326
2327again:
Nick Pigginafddba42007-10-16 01:25:01 -07002328 /*
2329 * Bring in the user page that we will copy from _first_.
2330 * Otherwise there's a nasty deadlock on copying from the
2331 * same page as we're writing to, without it being marked
2332 * up-to-date.
2333 *
2334 * Not only is this an optimisation, but it is also required
2335 * to check that the address is actually valid, when atomic
2336 * usercopies are used, below.
2337 */
2338 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2339 status = -EFAULT;
2340 break;
2341 }
2342
Nick Piggin674b8922007-10-16 01:25:03 -07002343 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002344 &page, &fsdata);
2345 if (unlikely(status))
2346 break;
2347
anfei zhou931e80e2010-02-02 13:44:02 -08002348 if (mapping_writably_mapped(mapping))
2349 flush_dcache_page(page);
2350
Nick Pigginafddba42007-10-16 01:25:01 -07002351 pagefault_disable();
2352 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2353 pagefault_enable();
2354 flush_dcache_page(page);
2355
Josef Bacikc8236db2009-07-05 12:08:18 -07002356 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002357 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2358 page, fsdata);
2359 if (unlikely(status < 0))
2360 break;
2361 copied = status;
2362
2363 cond_resched();
2364
Nick Piggin124d3b72008-02-02 15:01:17 +01002365 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002366 if (unlikely(copied == 0)) {
2367 /*
2368 * If we were unable to copy any data at all, we must
2369 * fall back to a single segment length write.
2370 *
2371 * If we didn't fallback here, we could livelock
2372 * because not all segments in the iov can be copied at
2373 * once without a pagefault.
2374 */
2375 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2376 iov_iter_single_seg_count(i));
2377 goto again;
2378 }
Nick Pigginafddba42007-10-16 01:25:01 -07002379 pos += copied;
2380 written += copied;
2381
2382 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002383 if (fatal_signal_pending(current)) {
2384 status = -EINTR;
2385 break;
2386 }
Nick Pigginafddba42007-10-16 01:25:01 -07002387 } while (iov_iter_count(i));
2388
2389 return written ? written : status;
2390}
2391
2392ssize_t
2393generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2394 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2395 size_t count, ssize_t written)
2396{
2397 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002398 ssize_t status;
2399 struct iov_iter i;
2400
2401 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002402 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002403
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002405 written += status;
2406 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 }
2408
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 return written ? written : status;
2410}
2411EXPORT_SYMBOL(generic_file_buffered_write);
2412
Jan Karae4dd9de2009-08-17 18:10:06 +02002413/**
2414 * __generic_file_aio_write - write data to a file
2415 * @iocb: IO state structure (file, offset, etc.)
2416 * @iov: vector with data to write
2417 * @nr_segs: number of segments in the vector
2418 * @ppos: position where to write
2419 *
2420 * This function does all the work needed for actually writing data to a
2421 * file. It does all basic checks, removes SUID from the file, updates
2422 * modification times and calls proper subroutines depending on whether we
2423 * do direct IO or a standard buffered write.
2424 *
2425 * It expects i_mutex to be grabbed unless we work on a block device or similar
2426 * object which does not need locking at all.
2427 *
2428 * This function does *not* take care of syncing data in case of O_SYNC write.
2429 * A caller has to handle it. This is mainly due to the fact that we want to
2430 * avoid syncing under i_mutex.
2431 */
2432ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2433 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434{
2435 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002436 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 size_t ocount; /* original count */
2438 size_t count; /* after file limit checks */
2439 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 loff_t pos;
2441 ssize_t written;
2442 ssize_t err;
2443
2444 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002445 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2446 if (err)
2447 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448
2449 count = ocount;
2450 pos = *ppos;
2451
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452 /* We can write back this queue in page reclaim */
2453 current->backing_dev_info = mapping->backing_dev_info;
2454 written = 0;
2455
2456 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2457 if (err)
2458 goto out;
2459
2460 if (count == 0)
2461 goto out;
2462
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002463 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 if (err)
2465 goto out;
2466
Josef Bacikc3b2da32012-03-26 09:59:21 -04002467 err = file_update_time(file);
2468 if (err)
2469 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
2471 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2472 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002473 loff_t endbyte;
2474 ssize_t written_buffered;
2475
2476 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2477 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 if (written < 0 || written == count)
2479 goto out;
2480 /*
2481 * direct-io write to a hole: fall through to buffered I/O
2482 * for completing the rest of the request.
2483 */
2484 pos += written;
2485 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002486 written_buffered = generic_file_buffered_write(iocb, iov,
2487 nr_segs, pos, ppos, count,
2488 written);
2489 /*
2490 * If generic_file_buffered_write() retuned a synchronous error
2491 * then we want to return the number of bytes which were
2492 * direct-written, or the error code if that was zero. Note
2493 * that this differs from normal direct-io semantics, which
2494 * will return -EFOO even if some bytes were written.
2495 */
2496 if (written_buffered < 0) {
2497 err = written_buffered;
2498 goto out;
2499 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002501 /*
2502 * We need to ensure that the page cache pages are written to
2503 * disk and invalidated to preserve the expected O_DIRECT
2504 * semantics.
2505 */
2506 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002507 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002508 if (err == 0) {
2509 written = written_buffered;
2510 invalidate_mapping_pages(mapping,
2511 pos >> PAGE_CACHE_SHIFT,
2512 endbyte >> PAGE_CACHE_SHIFT);
2513 } else {
2514 /*
2515 * We don't know how much we wrote, so just return
2516 * the number of bytes which were direct-written
2517 */
2518 }
2519 } else {
2520 written = generic_file_buffered_write(iocb, iov, nr_segs,
2521 pos, ppos, count, written);
2522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523out:
2524 current->backing_dev_info = NULL;
2525 return written ? written : err;
2526}
Jan Karae4dd9de2009-08-17 18:10:06 +02002527EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528
Jan Karae4dd9de2009-08-17 18:10:06 +02002529/**
2530 * generic_file_aio_write - write data to a file
2531 * @iocb: IO state structure
2532 * @iov: vector with data to write
2533 * @nr_segs: number of segments in the vector
2534 * @pos: position in file where to write
2535 *
2536 * This is a wrapper around __generic_file_aio_write() to be used by most
2537 * filesystems. It takes care of syncing the file in case of O_SYNC file
2538 * and acquires i_mutex as needed.
2539 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002540ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2541 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542{
2543 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002544 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
2547 BUG_ON(iocb->ki_pos != pos);
2548
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002549 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002550 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002551 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552
Jan Kara148f9482009-08-17 19:52:36 +02002553 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 ssize_t err;
2555
Jan Kara148f9482009-08-17 19:52:36 +02002556 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002557 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 ret = err;
2559 }
2560 return ret;
2561}
2562EXPORT_SYMBOL(generic_file_aio_write);
2563
David Howellscf9a2ae2006-08-29 19:05:54 +01002564/**
2565 * try_to_release_page() - release old fs-specific metadata on a page
2566 *
2567 * @page: the page which the kernel is trying to free
2568 * @gfp_mask: memory allocation flags (and I/O mode)
2569 *
2570 * The address_space is to try to release any data against the page
2571 * (presumably at page->private). If the release was successful, return `1'.
2572 * Otherwise return zero.
2573 *
David Howells266cf652009-04-03 16:42:36 +01002574 * This may also be called if PG_fscache is set on a page, indicating that the
2575 * page is known to the local caching routines.
2576 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002577 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002578 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002579 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002580 */
2581int try_to_release_page(struct page *page, gfp_t gfp_mask)
2582{
2583 struct address_space * const mapping = page->mapping;
2584
2585 BUG_ON(!PageLocked(page));
2586 if (PageWriteback(page))
2587 return 0;
2588
2589 if (mapping && mapping->a_ops->releasepage)
2590 return mapping->a_ops->releasepage(page, gfp_mask);
2591 return try_to_free_buffers(page);
2592}
2593
2594EXPORT_SYMBOL(try_to_release_page);