blob: 1e6aec4a2d2ebae29c0fea0405a7e81f422beeb4 [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)
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700470 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
Jan Kara5e4c0d972013-09-11 14:26:05 -0700472 error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700473 if (error) {
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700474 mem_cgroup_uncharge_cache_page(page);
Kirill A. Shutemov66a0c8e2013-09-12 15:13:59 -0700475 return error;
476 }
477
478 page_cache_get(page);
479 page->mapping = mapping;
480 page->index = offset;
481
482 spin_lock_irq(&mapping->tree_lock);
483 error = radix_tree_insert(&mapping->page_tree, offset, page);
484 radix_tree_preload_end();
485 if (unlikely(error))
486 goto err_insert;
487 mapping->nrpages++;
488 __inc_zone_page_state(page, NR_FILE_PAGES);
489 spin_unlock_irq(&mapping->tree_lock);
490 trace_mm_filemap_add_to_page_cache(page);
491 return 0;
492err_insert:
493 page->mapping = NULL;
494 /* Leave page->index set: truncation relies upon it */
495 spin_unlock_irq(&mapping->tree_lock);
496 mem_cgroup_uncharge_cache_page(page);
497 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return error;
499}
Nick Piggine2867812008-07-25 19:45:30 -0700500EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400503 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700505 int ret;
506
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700507 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
Hugh Dickins31475dd2011-08-03 16:21:27 -0700508 if (ret == 0)
509 lru_cache_add_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return ret;
511}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300512EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Paul Jackson44110fe2006-03-24 03:16:04 -0800514#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700515struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800516{
Miao Xiec0ff7452010-05-24 14:32:08 -0700517 int n;
518 struct page *page;
519
Paul Jackson44110fe2006-03-24 03:16:04 -0800520 if (cpuset_do_page_mem_spread()) {
Mel Gormancc9a6c82012-03-21 16:34:11 -0700521 unsigned int cpuset_mems_cookie;
522 do {
523 cpuset_mems_cookie = get_mems_allowed();
524 n = cpuset_mem_spread_node();
525 page = alloc_pages_exact_node(n, gfp, 0);
526 } while (!put_mems_allowed(cpuset_mems_cookie) && !page);
527
Miao Xiec0ff7452010-05-24 14:32:08 -0700528 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800529 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700530 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800531}
Nick Piggin2ae88142006-10-28 10:38:23 -0700532EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800533#endif
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535/*
536 * In order to wait for pages to become available there must be
537 * waitqueues associated with pages. By using a hash table of
538 * waitqueues where the bucket discipline is to maintain all
539 * waiters on the same queue and wake all when any of the pages
540 * become available, and for the woken contexts to check to be
541 * sure the appropriate page became available, this saves space
542 * at a cost of "thundering herd" phenomena during rare hash
543 * collisions.
544 */
545static wait_queue_head_t *page_waitqueue(struct page *page)
546{
547 const struct zone *zone = page_zone(page);
548
549 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
550}
551
552static inline void wake_up_page(struct page *page, int bit)
553{
554 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
555}
556
Harvey Harrison920c7a52008-02-04 22:29:26 -0800557void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
559 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
560
561 if (test_bit(bit_nr, &page->flags))
Jens Axboe7eaceac2011-03-10 08:52:07 +0100562 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 TASK_UNINTERRUPTIBLE);
564}
565EXPORT_SYMBOL(wait_on_page_bit);
566
KOSAKI Motohirof62e00c2011-05-24 17:11:29 -0700567int wait_on_page_bit_killable(struct page *page, int bit_nr)
568{
569 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
570
571 if (!test_bit(bit_nr, &page->flags))
572 return 0;
573
574 return __wait_on_bit(page_waitqueue(page), &wait,
575 sleep_on_page_killable, TASK_KILLABLE);
576}
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100579 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700580 * @page: Page defining the wait queue of interest
581 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100582 *
583 * Add an arbitrary @waiter to the wait queue for the nominated @page.
584 */
585void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
586{
587 wait_queue_head_t *q = page_waitqueue(page);
588 unsigned long flags;
589
590 spin_lock_irqsave(&q->lock, flags);
591 __add_wait_queue(q, waiter);
592 spin_unlock_irqrestore(&q->lock, flags);
593}
594EXPORT_SYMBOL_GPL(add_page_wait_queue);
595
596/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700597 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 * @page: the page
599 *
600 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
601 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
602 * mechananism between PageLocked pages and PageWriteback pages is shared.
603 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
604 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700605 * The mb is necessary to enforce ordering between the clear_bit and the read
606 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800608void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Nick Piggin8413ac92008-10-18 20:26:59 -0700610 VM_BUG_ON(!PageLocked(page));
611 clear_bit_unlock(PG_locked, &page->flags);
612 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 wake_up_page(page, PG_locked);
614}
615EXPORT_SYMBOL(unlock_page);
616
Randy Dunlap485bb992006-06-23 02:03:49 -0700617/**
618 * end_page_writeback - end writeback against a page
619 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 */
621void end_page_writeback(struct page *page)
622{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700623 if (TestClearPageReclaim(page))
624 rotate_reclaimable_page(page);
625
626 if (!test_clear_page_writeback(page))
627 BUG();
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 smp_mb__after_clear_bit();
630 wake_up_page(page, PG_writeback);
631}
632EXPORT_SYMBOL(end_page_writeback);
633
Randy Dunlap485bb992006-06-23 02:03:49 -0700634/**
635 * __lock_page - get a lock on the page, assuming we need to sleep to get it
636 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800638void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
640 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
641
Jens Axboe7eaceac2011-03-10 08:52:07 +0100642 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 TASK_UNINTERRUPTIBLE);
644}
645EXPORT_SYMBOL(__lock_page);
646
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800647int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500648{
649 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
650
651 return __wait_on_bit_lock(page_waitqueue(page), &wait,
Jens Axboe7eaceac2011-03-10 08:52:07 +0100652 sleep_on_page_killable, TASK_KILLABLE);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500653}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300654EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500655
Michel Lespinassed065bd82010-10-26 14:21:57 -0700656int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
657 unsigned int flags)
658{
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700659 if (flags & FAULT_FLAG_ALLOW_RETRY) {
660 /*
661 * CAUTION! In this case, mmap_sem is not released
662 * even though return 0.
663 */
664 if (flags & FAULT_FLAG_RETRY_NOWAIT)
665 return 0;
666
667 up_read(&mm->mmap_sem);
668 if (flags & FAULT_FLAG_KILLABLE)
669 wait_on_page_locked_killable(page);
670 else
Gleb Natapov318b2752011-03-22 16:30:51 -0700671 wait_on_page_locked(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -0700672 return 0;
KOSAKI Motohiro37b23e02011-05-24 17:11:30 -0700673 } else {
674 if (flags & FAULT_FLAG_KILLABLE) {
675 int ret;
676
677 ret = __lock_page_killable(page);
678 if (ret) {
679 up_read(&mm->mmap_sem);
680 return 0;
681 }
682 } else
683 __lock_page(page);
684 return 1;
Michel Lespinassed065bd82010-10-26 14:21:57 -0700685 }
686}
687
Randy Dunlap485bb992006-06-23 02:03:49 -0700688/**
689 * find_get_page - find and get a page reference
690 * @mapping: the address_space to search
691 * @offset: the page index
692 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700693 * Is there a pagecache struct page at the given (mapping, offset) tuple?
694 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 */
Nick Piggina60637c2008-07-25 19:45:31 -0700696struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
Nick Piggina60637c2008-07-25 19:45:31 -0700698 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 struct page *page;
700
Nick Piggina60637c2008-07-25 19:45:31 -0700701 rcu_read_lock();
702repeat:
703 page = NULL;
704 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
705 if (pagep) {
706 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800707 if (unlikely(!page))
708 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700709 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700710 if (radix_tree_deref_retry(page))
711 goto repeat;
712 /*
713 * Otherwise, shmem/tmpfs must be storing a swap entry
714 * here as an exceptional entry: so return it without
715 * attempting to raise page count.
716 */
717 goto out;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700718 }
Nick Piggina60637c2008-07-25 19:45:31 -0700719 if (!page_cache_get_speculative(page))
720 goto repeat;
721
722 /*
723 * Has the page moved?
724 * This is part of the lockless pagecache protocol. See
725 * include/linux/pagemap.h for details.
726 */
727 if (unlikely(page != *pagep)) {
728 page_cache_release(page);
729 goto repeat;
730 }
731 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800732out:
Nick Piggina60637c2008-07-25 19:45:31 -0700733 rcu_read_unlock();
734
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 return page;
736}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737EXPORT_SYMBOL(find_get_page);
738
Randy Dunlap485bb992006-06-23 02:03:49 -0700739/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700741 * @mapping: the address_space to search
742 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 *
744 * Locates the desired pagecache page, locks it, increments its reference
745 * count and returns its address.
746 *
747 * Returns zero if the page was not present. find_lock_page() may sleep.
748 */
Nick Piggina60637c2008-07-25 19:45:31 -0700749struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750{
751 struct page *page;
752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700754 page = find_get_page(mapping, offset);
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700755 if (page && !radix_tree_exception(page)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700756 lock_page(page);
757 /* Has the page been truncated? */
758 if (unlikely(page->mapping != mapping)) {
759 unlock_page(page);
760 page_cache_release(page);
761 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
Nick Piggina60637c2008-07-25 19:45:31 -0700763 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 return page;
766}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767EXPORT_SYMBOL(find_lock_page);
768
769/**
770 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700771 * @mapping: the page's address_space
772 * @index: the page's index into the mapping
773 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 *
775 * Locates a page in the pagecache. If the page is not present, a new page
776 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
777 * LRU list. The returned page is locked and has its reference count
778 * incremented.
779 *
780 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
781 * allocation!
782 *
783 * find_or_create_page() returns the desired page's address, or zero on
784 * memory exhaustion.
785 */
786struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700787 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788{
Nick Piggineb2be182007-10-16 01:24:57 -0700789 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 int err;
791repeat:
792 page = find_lock_page(mapping, index);
793 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700794 page = __page_cache_alloc(gfp_mask);
795 if (!page)
796 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800797 /*
798 * We want a regular kernel memory (not highmem or DMA etc)
799 * allocation for the radix tree nodes, but we need to honour
800 * the context-specific requirements the caller has asked for.
801 * GFP_RECLAIM_MASK collects those requirements.
802 */
803 err = add_to_page_cache_lru(page, mapping, index,
804 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700805 if (unlikely(err)) {
806 page_cache_release(page);
807 page = NULL;
808 if (err == -EEXIST)
809 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return page;
813}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814EXPORT_SYMBOL(find_or_create_page);
815
816/**
817 * find_get_pages - gang pagecache lookup
818 * @mapping: The address_space to search
819 * @start: The starting page index
820 * @nr_pages: The maximum number of pages
821 * @pages: Where the resulting pages are placed
822 *
823 * find_get_pages() will search for and return a group of up to
824 * @nr_pages pages in the mapping. The pages are placed at @pages.
825 * find_get_pages() takes a reference against the returned pages.
826 *
827 * The search returns a group of mapping-contiguous pages with ascending
828 * indexes. There may be holes in the indices due to not-present pages.
829 *
830 * find_get_pages() returns the number of pages which were found.
831 */
832unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
833 unsigned int nr_pages, struct page **pages)
834{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700835 struct radix_tree_iter iter;
836 void **slot;
837 unsigned ret = 0;
838
839 if (unlikely(!nr_pages))
840 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Nick Piggina60637c2008-07-25 19:45:31 -0700842 rcu_read_lock();
843restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700844 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Nick Piggina60637c2008-07-25 19:45:31 -0700845 struct page *page;
846repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700847 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -0700848 if (unlikely(!page))
849 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700850
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700851 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700852 if (radix_tree_deref_retry(page)) {
853 /*
854 * Transient condition which can only trigger
855 * when entry at index 0 moves out of or back
856 * to root: none yet gotten, safe to restart.
857 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700858 WARN_ON(iter.index);
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700859 goto restart;
860 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700861 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700862 * Otherwise, shmem/tmpfs must be storing a swap entry
863 * here as an exceptional entry: so skip over it -
864 * we only reach this from invalidate_mapping_pages().
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700865 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700866 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800867 }
Nick Piggina60637c2008-07-25 19:45:31 -0700868
869 if (!page_cache_get_speculative(page))
870 goto repeat;
871
872 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700873 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700874 page_cache_release(page);
875 goto repeat;
876 }
877
878 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700879 if (++ret == nr_pages)
880 break;
Nick Piggina60637c2008-07-25 19:45:31 -0700881 }
Hugh Dickins5b280c02011-03-22 16:33:07 -0700882
Nick Piggina60637c2008-07-25 19:45:31 -0700883 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 return ret;
885}
886
Jens Axboeebf43502006-04-27 08:46:01 +0200887/**
888 * find_get_pages_contig - gang contiguous pagecache lookup
889 * @mapping: The address_space to search
890 * @index: The starting page index
891 * @nr_pages: The maximum number of pages
892 * @pages: Where the resulting pages are placed
893 *
894 * find_get_pages_contig() works exactly like find_get_pages(), except
895 * that the returned number of pages are guaranteed to be contiguous.
896 *
897 * find_get_pages_contig() returns the number of pages which were found.
898 */
899unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
900 unsigned int nr_pages, struct page **pages)
901{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700902 struct radix_tree_iter iter;
903 void **slot;
904 unsigned int ret = 0;
905
906 if (unlikely(!nr_pages))
907 return 0;
Jens Axboeebf43502006-04-27 08:46:01 +0200908
Nick Piggina60637c2008-07-25 19:45:31 -0700909 rcu_read_lock();
910restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700911 radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
Nick Piggina60637c2008-07-25 19:45:31 -0700912 struct page *page;
913repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700914 page = radix_tree_deref_slot(slot);
915 /* The hole, there no reason to continue */
Nick Piggina60637c2008-07-25 19:45:31 -0700916 if (unlikely(!page))
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700917 break;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700918
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700919 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700920 if (radix_tree_deref_retry(page)) {
921 /*
922 * Transient condition which can only trigger
923 * when entry at index 0 moves out of or back
924 * to root: none yet gotten, safe to restart.
925 */
926 goto restart;
927 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700928 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700929 * Otherwise, shmem/tmpfs must be storing a swap entry
930 * here as an exceptional entry: so stop looking for
931 * contiguous pages.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700932 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700933 break;
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700934 }
Nick Piggina60637c2008-07-25 19:45:31 -0700935
Nick Piggina60637c2008-07-25 19:45:31 -0700936 if (!page_cache_get_speculative(page))
937 goto repeat;
938
939 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700940 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -0700941 page_cache_release(page);
942 goto repeat;
943 }
944
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800945 /*
946 * must check mapping and index after taking the ref.
947 * otherwise we can get both false positives and false
948 * negatives, which is just confusing to the caller.
949 */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700950 if (page->mapping == NULL || page->index != iter.index) {
Nick Piggin9cbb4cb2011-01-13 15:45:51 -0800951 page_cache_release(page);
952 break;
953 }
954
Nick Piggina60637c2008-07-25 19:45:31 -0700955 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700956 if (++ret == nr_pages)
957 break;
Jens Axboeebf43502006-04-27 08:46:01 +0200958 }
Nick Piggina60637c2008-07-25 19:45:31 -0700959 rcu_read_unlock();
960 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200961}
David Howellsef71c152007-05-09 02:33:44 -0700962EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200963
Randy Dunlap485bb992006-06-23 02:03:49 -0700964/**
965 * find_get_pages_tag - find and return pages that match @tag
966 * @mapping: the address_space to search
967 * @index: the starting page index
968 * @tag: the tag index
969 * @nr_pages: the maximum number of pages
970 * @pages: where the resulting pages are placed
971 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700973 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 */
975unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
976 int tag, unsigned int nr_pages, struct page **pages)
977{
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700978 struct radix_tree_iter iter;
979 void **slot;
980 unsigned ret = 0;
981
982 if (unlikely(!nr_pages))
983 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Nick Piggina60637c2008-07-25 19:45:31 -0700985 rcu_read_lock();
986restart:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700987 radix_tree_for_each_tagged(slot, &mapping->page_tree,
988 &iter, *index, tag) {
Nick Piggina60637c2008-07-25 19:45:31 -0700989 struct page *page;
990repeat:
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -0700991 page = radix_tree_deref_slot(slot);
Nick Piggina60637c2008-07-25 19:45:31 -0700992 if (unlikely(!page))
993 continue;
Hugh Dickins9d8aa4e2011-03-22 16:33:06 -0700994
Hugh Dickinsa2c16d62011-08-03 16:21:19 -0700995 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700996 if (radix_tree_deref_retry(page)) {
997 /*
998 * Transient condition which can only trigger
999 * when entry at index 0 moves out of or back
1000 * to root: none yet gotten, safe to restart.
1001 */
1002 goto restart;
1003 }
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001004 /*
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001005 * This function is never used on a shmem/tmpfs
1006 * mapping, so a swap entry won't be found here.
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001007 */
Hugh Dickins8079b1c2011-08-03 16:21:28 -07001008 BUG();
Hugh Dickinsa2c16d62011-08-03 16:21:19 -07001009 }
Nick Piggina60637c2008-07-25 19:45:31 -07001010
1011 if (!page_cache_get_speculative(page))
1012 goto repeat;
1013
1014 /* Has the page moved? */
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001015 if (unlikely(page != *slot)) {
Nick Piggina60637c2008-07-25 19:45:31 -07001016 page_cache_release(page);
1017 goto repeat;
1018 }
1019
1020 pages[ret] = page;
Konstantin Khlebnikov0fc9d102012-03-28 14:42:54 -07001021 if (++ret == nr_pages)
1022 break;
Nick Piggina60637c2008-07-25 19:45:31 -07001023 }
Hugh Dickins5b280c02011-03-22 16:33:07 -07001024
Nick Piggina60637c2008-07-25 19:45:31 -07001025 rcu_read_unlock();
1026
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (ret)
1028 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -07001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 return ret;
1031}
David Howellsef71c152007-05-09 02:33:44 -07001032EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Randy Dunlap485bb992006-06-23 02:03:49 -07001034/**
1035 * grab_cache_page_nowait - returns locked page at given index in given cache
1036 * @mapping: target address_space
1037 * @index: the page index
1038 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001039 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 * This is intended for speculative data generators, where the data can
1041 * be regenerated if the page couldn't be grabbed. This routine should
1042 * be safe to call while holding the lock for another page.
1043 *
1044 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1045 * and deadlock against the caller's locked page.
1046 */
1047struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -07001048grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049{
1050 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +02001053 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 return page;
1055 page_cache_release(page);
1056 return NULL;
1057 }
Nick Piggin2ae88142006-10-28 10:38:23 -07001058 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -08001059 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 page_cache_release(page);
1061 page = NULL;
1062 }
1063 return page;
1064}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065EXPORT_SYMBOL(grab_cache_page_nowait);
1066
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001067/*
1068 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1069 * a _large_ part of the i/o request. Imagine the worst scenario:
1070 *
1071 * ---R__________________________________________B__________
1072 * ^ reading here ^ bad block(assume 4k)
1073 *
1074 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1075 * => failing the whole request => read(R) => read(R+1) =>
1076 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1077 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1078 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1079 *
1080 * It is going insane. Fix it by quickly scaling down the readahead size.
1081 */
1082static void shrink_readahead_size_eio(struct file *filp,
1083 struct file_ra_state *ra)
1084{
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001085 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -07001086}
1087
Randy Dunlap485bb992006-06-23 02:03:49 -07001088/**
Christoph Hellwig36e78912008-02-08 04:21:24 -08001089 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001090 * @filp: the file to read
1091 * @ppos: current file position
1092 * @desc: read_descriptor
1093 * @actor: read method
1094 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -07001096 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 *
1098 * This is really ugly. But the goto's actually try to clarify some
1099 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 */
Christoph Hellwig36e78912008-02-08 04:21:24 -08001101static void do_generic_file_read(struct file *filp, loff_t *ppos,
1102 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001104 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001106 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001107 pgoff_t index;
1108 pgoff_t last_index;
1109 pgoff_t prev_index;
1110 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001111 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001115 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1116 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1118 offset = *ppos & ~PAGE_CACHE_MASK;
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 for (;;) {
1121 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001122 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001123 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 unsigned long nr, ret;
1125
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127find_page:
1128 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001129 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001130 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001131 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001132 index, last_index - index);
1133 page = find_get_page(mapping, index);
1134 if (unlikely(page == NULL))
1135 goto no_cached_page;
1136 }
1137 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001138 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001139 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001140 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001142 if (!PageUptodate(page)) {
1143 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1144 !mapping->a_ops->is_partially_uptodate)
1145 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001146 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001147 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001148 /* Did it get truncated before we got the lock? */
1149 if (!page->mapping)
1150 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001151 if (!mapping->a_ops->is_partially_uptodate(page,
1152 desc, offset))
1153 goto page_not_up_to_date_locked;
1154 unlock_page(page);
1155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001157 /*
1158 * i_size must be checked after we know the page is Uptodate.
1159 *
1160 * Checking i_size after the check allows us to calculate
1161 * the correct value for "nr", which means the zero-filled
1162 * part of the page is not copied back to userspace (unless
1163 * another truncate extends the file - this is desired though).
1164 */
1165
1166 isize = i_size_read(inode);
1167 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1168 if (unlikely(!isize || index > end_index)) {
1169 page_cache_release(page);
1170 goto out;
1171 }
1172
1173 /* nr is the maximum number of bytes to copy from this page */
1174 nr = PAGE_CACHE_SIZE;
1175 if (index == end_index) {
1176 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1177 if (nr <= offset) {
1178 page_cache_release(page);
1179 goto out;
1180 }
1181 }
1182 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184 /* If users can be writing to this page using arbitrary
1185 * virtual addresses, take care about potential aliasing
1186 * before reading the page on the kernel side.
1187 */
1188 if (mapping_writably_mapped(mapping))
1189 flush_dcache_page(page);
1190
1191 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001192 * When a sequential read accesses a page several times,
1193 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 */
Jan Karaec0f1632007-05-06 14:49:25 -07001195 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 mark_page_accessed(page);
1197 prev_index = index;
1198
1199 /*
1200 * Ok, we have the page, and it's up-to-date, so
1201 * now we can copy it to user space...
1202 *
1203 * The actor routine returns how many bytes were actually used..
1204 * NOTE! This may not be the same as how much of a user buffer
1205 * we filled up (we may be padding etc), so we can only update
1206 * "pos" here (the actor routine has to update the user buffer
1207 * pointers and the remaining count).
1208 */
1209 ret = actor(desc, page, offset, nr);
1210 offset += ret;
1211 index += offset >> PAGE_CACHE_SHIFT;
1212 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001213 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
1215 page_cache_release(page);
1216 if (ret == nr && desc->count)
1217 continue;
1218 goto out;
1219
1220page_not_up_to_date:
1221 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001222 error = lock_page_killable(page);
1223 if (unlikely(error))
1224 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001226page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001227 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (!page->mapping) {
1229 unlock_page(page);
1230 page_cache_release(page);
1231 continue;
1232 }
1233
1234 /* Did somebody else fill it already? */
1235 if (PageUptodate(page)) {
1236 unlock_page(page);
1237 goto page_ok;
1238 }
1239
1240readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001241 /*
1242 * A previous I/O error may have been due to temporary
1243 * failures, eg. multipath errors.
1244 * PG_error will be set again if readpage fails.
1245 */
1246 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 /* Start the actual read. The read will unlock the page. */
1248 error = mapping->a_ops->readpage(filp, page);
1249
Zach Brown994fc28c2005-12-15 14:28:17 -08001250 if (unlikely(error)) {
1251 if (error == AOP_TRUNCATED_PAGE) {
1252 page_cache_release(page);
1253 goto find_page;
1254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001256 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
1258 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001259 error = lock_page_killable(page);
1260 if (unlikely(error))
1261 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 if (!PageUptodate(page)) {
1263 if (page->mapping == NULL) {
1264 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001265 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 */
1267 unlock_page(page);
1268 page_cache_release(page);
1269 goto find_page;
1270 }
1271 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001272 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001273 error = -EIO;
1274 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
1276 unlock_page(page);
1277 }
1278
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 goto page_ok;
1280
1281readpage_error:
1282 /* UHHUH! A synchronous read error occurred. Report it */
1283 desc->error = error;
1284 page_cache_release(page);
1285 goto out;
1286
1287no_cached_page:
1288 /*
1289 * Ok, it wasn't cached, so we need to create a new
1290 * page..
1291 */
Nick Piggineb2be182007-10-16 01:24:57 -07001292 page = page_cache_alloc_cold(mapping);
1293 if (!page) {
1294 desc->error = -ENOMEM;
1295 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 }
Nick Piggineb2be182007-10-16 01:24:57 -07001297 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 index, GFP_KERNEL);
1299 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001300 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 if (error == -EEXIST)
1302 goto find_page;
1303 desc->error = error;
1304 goto out;
1305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 goto readpage;
1307 }
1308
1309out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001310 ra->prev_pos = prev_index;
1311 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1312 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001314 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001315 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318int file_read_actor(read_descriptor_t *desc, struct page *page,
1319 unsigned long offset, unsigned long size)
1320{
1321 char *kaddr;
1322 unsigned long left, count = desc->count;
1323
1324 if (size > count)
1325 size = count;
1326
1327 /*
1328 * Faults on the destination of a read are common, so do it before
1329 * taking the kmap.
1330 */
1331 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
Cong Wang9b04c5f2011-11-25 23:14:39 +08001332 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 left = __copy_to_user_inatomic(desc->arg.buf,
1334 kaddr + offset, size);
Cong Wang9b04c5f2011-11-25 23:14:39 +08001335 kunmap_atomic(kaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 if (left == 0)
1337 goto success;
1338 }
1339
1340 /* Do it the slow way */
1341 kaddr = kmap(page);
1342 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1343 kunmap(page);
1344
1345 if (left) {
1346 size -= left;
1347 desc->error = -EFAULT;
1348 }
1349success:
1350 desc->count = count - size;
1351 desc->written += size;
1352 desc->arg.buf += size;
1353 return size;
1354}
1355
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001356/*
1357 * Performs necessary checks before doing a write
1358 * @iov: io vector request
1359 * @nr_segs: number of segments in the iovec
1360 * @count: number of bytes to write
1361 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1362 *
1363 * Adjust number of segments and amount of bytes to write (nr_segs should be
1364 * properly initialized first). Returns appropriate error code that caller
1365 * should return or zero in case that write should be allowed.
1366 */
1367int generic_segment_checks(const struct iovec *iov,
1368 unsigned long *nr_segs, size_t *count, int access_flags)
1369{
1370 unsigned long seg;
1371 size_t cnt = 0;
1372 for (seg = 0; seg < *nr_segs; seg++) {
1373 const struct iovec *iv = &iov[seg];
1374
1375 /*
1376 * If any segment has a negative length, or the cumulative
1377 * length ever wraps negative then return -EINVAL.
1378 */
1379 cnt += iv->iov_len;
1380 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1381 return -EINVAL;
1382 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1383 continue;
1384 if (seg == 0)
1385 return -EFAULT;
1386 *nr_segs = seg;
1387 cnt -= iv->iov_len; /* This segment is no good */
1388 break;
1389 }
1390 *count = cnt;
1391 return 0;
1392}
1393EXPORT_SYMBOL(generic_segment_checks);
1394
Randy Dunlap485bb992006-06-23 02:03:49 -07001395/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001396 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001397 * @iocb: kernel I/O control block
1398 * @iov: io vector request
1399 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001400 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001401 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 * This is the "read()" routine for all filesystems
1403 * that can use the page cache directly.
1404 */
1405ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001406generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1407 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
1409 struct file *filp = iocb->ki_filp;
1410 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001411 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001413 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001416 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1417 if (retval)
1418 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1421 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001422 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 struct address_space *mapping;
1424 struct inode *inode;
1425
1426 mapping = filp->f_mapping;
1427 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 if (!count)
1429 goto out; /* skip atime */
1430 size = i_size_read(inode);
1431 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001432 retval = filemap_write_and_wait_range(mapping, pos,
1433 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001434 if (!retval) {
1435 retval = mapping->a_ops->direct_IO(READ, iocb,
1436 iov, pos, nr_segs);
1437 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001438 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001440 count -= retval;
1441 }
1442
1443 /*
1444 * Btrfs can have a short DIO read if we encounter
1445 * compressed extents, so if there was an error, or if
1446 * we've already read everything we wanted to, or if
1447 * there was a short read because we hit EOF, go ahead
1448 * and return. Otherwise fallthrough to buffered io for
1449 * the rest of the read.
1450 */
1451 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001452 file_accessed(filp);
1453 goto out;
1454 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001455 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 }
1457
Josef Bacik66f998f2010-05-23 11:00:54 -04001458 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001459 for (seg = 0; seg < nr_segs; seg++) {
1460 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001461 loff_t offset = 0;
1462
1463 /*
1464 * If we did a short DIO read we need to skip the section of the
1465 * iov that we've already read data into.
1466 */
1467 if (count) {
1468 if (count > iov[seg].iov_len) {
1469 count -= iov[seg].iov_len;
1470 continue;
1471 }
1472 offset = count;
1473 count = 0;
1474 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475
Hugh Dickins11fa9772008-07-23 21:27:34 -07001476 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001477 desc.arg.buf = iov[seg].iov_base + offset;
1478 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001479 if (desc.count == 0)
1480 continue;
1481 desc.error = 0;
1482 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1483 retval += desc.written;
1484 if (desc.error) {
1485 retval = retval ?: desc.error;
1486 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001488 if (desc.count > 0)
1489 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 }
1491out:
1492 return retval;
1493}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494EXPORT_SYMBOL(generic_file_aio_read);
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001497/**
1498 * page_cache_read - adds requested page to the page cache if not already there
1499 * @file: file to read
1500 * @offset: page index
1501 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 * This adds the requested page to the page cache if it isn't already there,
1503 * and schedules an I/O to read in its contents from disk.
1504 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001505static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
1507 struct address_space *mapping = file->f_mapping;
1508 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001509 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
Zach Brown994fc28c2005-12-15 14:28:17 -08001511 do {
1512 page = page_cache_alloc_cold(mapping);
1513 if (!page)
1514 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
Zach Brown994fc28c2005-12-15 14:28:17 -08001516 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1517 if (ret == 0)
1518 ret = mapping->a_ops->readpage(file, page);
1519 else if (ret == -EEXIST)
1520 ret = 0; /* losing race to add is OK */
1521
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Zach Brown994fc28c2005-12-15 14:28:17 -08001524 } while (ret == AOP_TRUNCATED_PAGE);
1525
1526 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
1528
1529#define MMAP_LOTSAMISS (100)
1530
Linus Torvaldsef00e082009-06-16 15:31:25 -07001531/*
1532 * Synchronous readahead happens when we don't even find
1533 * a page in the page cache at all.
1534 */
1535static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1536 struct file_ra_state *ra,
1537 struct file *file,
1538 pgoff_t offset)
1539{
1540 unsigned long ra_pages;
1541 struct address_space *mapping = file->f_mapping;
1542
1543 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001544 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001545 return;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001546 if (!ra->ra_pages)
1547 return;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001548
Joe Perches64363aa2013-07-08 16:00:18 -07001549 if (vma->vm_flags & VM_SEQ_READ) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001550 page_cache_sync_readahead(mapping, ra, file, offset,
1551 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001552 return;
1553 }
1554
Andi Kleen207d04b2011-05-24 17:12:29 -07001555 /* Avoid banging the cache line if not needed */
1556 if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001557 ra->mmap_miss++;
1558
1559 /*
1560 * Do we miss much more than hit in this file? If so,
1561 * stop bothering with read-ahead. It will only hurt.
1562 */
1563 if (ra->mmap_miss > MMAP_LOTSAMISS)
1564 return;
1565
Wu Fengguangd30a1102009-06-16 15:31:30 -07001566 /*
1567 * mmap read-around
1568 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001569 ra_pages = max_sane_readahead(ra->ra_pages);
Wu Fengguang275b12b2011-05-24 17:12:28 -07001570 ra->start = max_t(long, 0, offset - ra_pages / 2);
1571 ra->size = ra_pages;
Wu Fengguang2cbea1d2011-05-24 17:12:30 -07001572 ra->async_size = ra_pages / 4;
Wu Fengguang275b12b2011-05-24 17:12:28 -07001573 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001574}
1575
1576/*
1577 * Asynchronous readahead happens when we find the page and PG_readahead,
1578 * so we want to possibly extend the readahead further..
1579 */
1580static void do_async_mmap_readahead(struct vm_area_struct *vma,
1581 struct file_ra_state *ra,
1582 struct file *file,
1583 struct page *page,
1584 pgoff_t offset)
1585{
1586 struct address_space *mapping = file->f_mapping;
1587
1588 /* If we don't want any read-ahead, don't bother */
Joe Perches64363aa2013-07-08 16:00:18 -07001589 if (vma->vm_flags & VM_RAND_READ)
Linus Torvaldsef00e082009-06-16 15:31:25 -07001590 return;
1591 if (ra->mmap_miss > 0)
1592 ra->mmap_miss--;
1593 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001594 page_cache_async_readahead(mapping, ra, file,
1595 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001596}
1597
Randy Dunlap485bb992006-06-23 02:03:49 -07001598/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001599 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001600 * @vma: vma in which the fault was taken
1601 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001602 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001603 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 * mapped memory region to read in file data during a page fault.
1605 *
1606 * The goto's are kind of ugly, but this streamlines the normal case of having
1607 * it in the page cache, and handles the special cases reasonably without
1608 * having a lot of duplicated code.
1609 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001610int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
1612 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001613 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 struct address_space *mapping = file->f_mapping;
1615 struct file_ra_state *ra = &file->f_ra;
1616 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001617 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 struct page *page;
Johannes Weiner519e5242013-09-12 15:13:42 -07001619 bool memcg_oom;
Jan Kara2004dc82008-02-08 04:20:11 -08001620 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001621 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001624 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001625 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 /*
Johannes Weiner519e5242013-09-12 15:13:42 -07001628 * Do we have something in the page cache already? Either
1629 * way, try readahead, but disable the memcg OOM killer for it
1630 * as readahead is optional and no errors are propagated up
1631 * the fault stack. The OOM killer is enabled while trying to
1632 * instantiate the faulting page individually below.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001634 page = find_get_page(mapping, offset);
Shaohua Li45cac652012-10-08 16:32:19 -07001635 if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001637 * We found the page, so try async readahead before
1638 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 */
Johannes Weiner519e5242013-09-12 15:13:42 -07001640 memcg_oom = mem_cgroup_toggle_oom(false);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001641 do_async_mmap_readahead(vma, ra, file, page, offset);
Johannes Weiner519e5242013-09-12 15:13:42 -07001642 mem_cgroup_toggle_oom(memcg_oom);
Shaohua Li45cac652012-10-08 16:32:19 -07001643 } else if (!page) {
Linus Torvaldsef00e082009-06-16 15:31:25 -07001644 /* No page in the page cache at all */
Johannes Weiner519e5242013-09-12 15:13:42 -07001645 memcg_oom = mem_cgroup_toggle_oom(false);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001646 do_sync_mmap_readahead(vma, ra, file, offset);
Johannes Weiner519e5242013-09-12 15:13:42 -07001647 mem_cgroup_toggle_oom(memcg_oom);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001648 count_vm_event(PGMAJFAULT);
Ying Han456f9982011-05-26 16:25:38 -07001649 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001650 ret = VM_FAULT_MAJOR;
1651retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001652 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 if (!page)
1654 goto no_cached_page;
1655 }
1656
Michel Lespinassed88c0922010-11-02 13:05:18 -07001657 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1658 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001659 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001660 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001661
1662 /* Did it get truncated? */
1663 if (unlikely(page->mapping != mapping)) {
1664 unlock_page(page);
1665 put_page(page);
1666 goto retry_find;
1667 }
1668 VM_BUG_ON(page->index != offset);
1669
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001671 * We have a locked page in the page cache, now we need to check
1672 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 */
Nick Piggind00806b2007-07-19 01:46:57 -07001674 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 goto page_not_uptodate;
1676
Linus Torvaldsef00e082009-06-16 15:31:25 -07001677 /*
1678 * Found the page and have a reference on it.
1679 * We must recheck i_size under page lock.
1680 */
Nick Piggind00806b2007-07-19 01:46:57 -07001681 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001682 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001683 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001684 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001685 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001686 }
1687
Nick Piggind0217ac2007-07-19 01:47:03 -07001688 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001689 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691no_cached_page:
1692 /*
1693 * We're only likely to ever get here if MADV_RANDOM is in
1694 * effect.
1695 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001696 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
1698 /*
1699 * The page we want has now been added to the page cache.
1700 * In the unlikely event that someone removed it in the
1701 * meantime, we'll just come back here and read it again.
1702 */
1703 if (error >= 0)
1704 goto retry_find;
1705
1706 /*
1707 * An error return from page_cache_read can result if the
1708 * system is low on memory, or a problem occurs while trying
1709 * to schedule I/O.
1710 */
1711 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001712 return VM_FAULT_OOM;
1713 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
1715page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 /*
1717 * Umm, take care of errors if the page isn't up-to-date.
1718 * Try to re-read it _once_. We do this synchronously,
1719 * because there really aren't any performance issues here
1720 * and we need to check for errors.
1721 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001723 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001724 if (!error) {
1725 wait_on_page_locked(page);
1726 if (!PageUptodate(page))
1727 error = -EIO;
1728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001730
1731 if (!error || error == AOP_TRUNCATED_PAGE)
1732 goto retry_find;
1733
1734 /* Things didn't work out. Return zero to tell the mm layer so. */
1735 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001736 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001737}
1738EXPORT_SYMBOL(filemap_fault);
1739
Jan Kara4fcf1c62012-06-12 16:20:29 +02001740int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1741{
1742 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05001743 struct inode *inode = file_inode(vma->vm_file);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001744 int ret = VM_FAULT_LOCKED;
1745
Jan Kara14da9202012-06-12 16:20:37 +02001746 sb_start_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001747 file_update_time(vma->vm_file);
1748 lock_page(page);
1749 if (page->mapping != inode->i_mapping) {
1750 unlock_page(page);
1751 ret = VM_FAULT_NOPAGE;
1752 goto out;
1753 }
Jan Kara14da9202012-06-12 16:20:37 +02001754 /*
1755 * We mark the page dirty already here so that when freeze is in
1756 * progress, we are guaranteed that writeback during freezing will
1757 * see the dirty page and writeprotect it again.
1758 */
1759 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08001760 wait_for_stable_page(page);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001761out:
Jan Kara14da9202012-06-12 16:20:37 +02001762 sb_end_pagefault(inode->i_sb);
Jan Kara4fcf1c62012-06-12 16:20:29 +02001763 return ret;
1764}
1765EXPORT_SYMBOL(filemap_page_mkwrite);
1766
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001767const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001768 .fault = filemap_fault,
Jan Kara4fcf1c62012-06-12 16:20:29 +02001769 .page_mkwrite = filemap_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001770 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771};
1772
1773/* This is used for a general mmap of a disk file */
1774
1775int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1776{
1777 struct address_space *mapping = file->f_mapping;
1778
1779 if (!mapping->a_ops->readpage)
1780 return -ENOEXEC;
1781 file_accessed(file);
1782 vma->vm_ops = &generic_file_vm_ops;
1783 return 0;
1784}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
1786/*
1787 * This is for filesystems which do not implement ->writepage.
1788 */
1789int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1790{
1791 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1792 return -EINVAL;
1793 return generic_file_mmap(file, vma);
1794}
1795#else
1796int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1797{
1798 return -ENOSYS;
1799}
1800int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1801{
1802 return -ENOSYS;
1803}
1804#endif /* CONFIG_MMU */
1805
1806EXPORT_SYMBOL(generic_file_mmap);
1807EXPORT_SYMBOL(generic_file_readonly_mmap);
1808
Nick Piggin6fe69002007-05-06 14:49:04 -07001809static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001810 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001811 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001812 void *data,
1813 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814{
Nick Piggineb2be182007-10-16 01:24:57 -07001815 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 int err;
1817repeat:
1818 page = find_get_page(mapping, index);
1819 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001820 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001821 if (!page)
1822 return ERR_PTR(-ENOMEM);
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001823 err = add_to_page_cache_lru(page, mapping, index, gfp);
Nick Piggineb2be182007-10-16 01:24:57 -07001824 if (unlikely(err)) {
1825 page_cache_release(page);
1826 if (err == -EEXIST)
1827 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 return ERR_PTR(err);
1830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 err = filler(data, page);
1832 if (err < 0) {
1833 page_cache_release(page);
1834 page = ERR_PTR(err);
1835 }
1836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 return page;
1838}
1839
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001840static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001841 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001842 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001843 void *data,
1844 gfp_t gfp)
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846{
1847 struct page *page;
1848 int err;
1849
1850retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001851 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001853 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 if (PageUptodate(page))
1855 goto out;
1856
1857 lock_page(page);
1858 if (!page->mapping) {
1859 unlock_page(page);
1860 page_cache_release(page);
1861 goto retry;
1862 }
1863 if (PageUptodate(page)) {
1864 unlock_page(page);
1865 goto out;
1866 }
1867 err = filler(data, page);
1868 if (err < 0) {
1869 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001870 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 }
David Howellsc855ff32007-05-09 13:42:20 +01001872out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001873 mark_page_accessed(page);
1874 return page;
1875}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001876
1877/**
1878 * read_cache_page_async - read into page cache, fill it if needed
1879 * @mapping: the page's address_space
1880 * @index: the page index
1881 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001882 * @data: first arg to filler(data, page) function, often left as NULL
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001883 *
1884 * Same as read_cache_page, but don't wait for page to become unlocked
1885 * after submitting it to the filler.
1886 *
1887 * Read into the page cache. If a page already exists, and PageUptodate() is
1888 * not set, try to fill the page but don't wait for it to become unlocked.
1889 *
1890 * If the page does not get brought uptodate, return -EIO.
1891 */
1892struct page *read_cache_page_async(struct address_space *mapping,
1893 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001894 int (*filler)(void *, struct page *),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001895 void *data)
1896{
1897 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1898}
Nick Piggin6fe69002007-05-06 14:49:04 -07001899EXPORT_SYMBOL(read_cache_page_async);
1900
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001901static struct page *wait_on_page_read(struct page *page)
1902{
1903 if (!IS_ERR(page)) {
1904 wait_on_page_locked(page);
1905 if (!PageUptodate(page)) {
1906 page_cache_release(page);
1907 page = ERR_PTR(-EIO);
1908 }
1909 }
1910 return page;
1911}
1912
1913/**
1914 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1915 * @mapping: the page's address_space
1916 * @index: the page index
1917 * @gfp: the page allocator flags to use if allocating
1918 *
1919 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
Dave Kleikampe6f67b82011-12-21 11:05:48 -06001920 * any new page allocations done using the specified allocation flags.
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001921 *
1922 * If the page does not get brought uptodate, return -EIO.
1923 */
1924struct page *read_cache_page_gfp(struct address_space *mapping,
1925 pgoff_t index,
1926 gfp_t gfp)
1927{
1928 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1929
1930 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1931}
1932EXPORT_SYMBOL(read_cache_page_gfp);
1933
Nick Piggin6fe69002007-05-06 14:49:04 -07001934/**
1935 * read_cache_page - read into page cache, fill it if needed
1936 * @mapping: the page's address_space
1937 * @index: the page index
1938 * @filler: function to perform the read
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001939 * @data: first arg to filler(data, page) function, often left as NULL
Nick Piggin6fe69002007-05-06 14:49:04 -07001940 *
1941 * Read into the page cache. If a page already exists, and PageUptodate() is
1942 * not set, try to fill the page then wait for it to become unlocked.
1943 *
1944 * If the page does not get brought uptodate, return -EIO.
1945 */
1946struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001947 pgoff_t index,
Hugh Dickins5e5358e2011-07-25 17:12:23 -07001948 int (*filler)(void *, struct page *),
Nick Piggin6fe69002007-05-06 14:49:04 -07001949 void *data)
1950{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001951 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953EXPORT_SYMBOL(read_cache_page);
1954
Nick Piggin2f718ff2007-10-16 01:24:59 -07001955static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 const struct iovec *iov, size_t base, size_t bytes)
1957{
Ingo Molnarf1800532009-03-02 11:00:57 +01001958 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 while (bytes) {
1961 char __user *buf = iov->iov_base + base;
1962 int copy = min(bytes, iov->iov_len - base);
1963
1964 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001965 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 copied += copy;
1967 bytes -= copy;
1968 vaddr += copy;
1969 iov++;
1970
NeilBrown01408c42006-06-25 05:47:58 -07001971 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 }
1974 return copied - left;
1975}
1976
1977/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001978 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001979 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001980 * bytes which were copied.
1981 */
1982size_t iov_iter_copy_from_user_atomic(struct page *page,
1983 struct iov_iter *i, unsigned long offset, size_t bytes)
1984{
1985 char *kaddr;
1986 size_t copied;
1987
1988 BUG_ON(!in_atomic());
Cong Wang9b04c5f2011-11-25 23:14:39 +08001989 kaddr = kmap_atomic(page);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001990 if (likely(i->nr_segs == 1)) {
1991 int left;
1992 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001993 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001994 copied = bytes - left;
1995 } else {
1996 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1997 i->iov, i->iov_offset, bytes);
1998 }
Cong Wang9b04c5f2011-11-25 23:14:39 +08001999 kunmap_atomic(kaddr);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002000
2001 return copied;
2002}
Nick Piggin89e10782007-10-16 01:25:07 -07002003EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002004
2005/*
2006 * This has the same sideeffects and return value as
2007 * iov_iter_copy_from_user_atomic().
2008 * The difference is that it attempts to resolve faults.
2009 * Page must not be locked.
2010 */
2011size_t iov_iter_copy_from_user(struct page *page,
2012 struct iov_iter *i, unsigned long offset, size_t bytes)
2013{
2014 char *kaddr;
2015 size_t copied;
2016
2017 kaddr = kmap(page);
2018 if (likely(i->nr_segs == 1)) {
2019 int left;
2020 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01002021 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002022 copied = bytes - left;
2023 } else {
2024 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
2025 i->iov, i->iov_offset, bytes);
2026 }
2027 kunmap(page);
2028 return copied;
2029}
Nick Piggin89e10782007-10-16 01:25:07 -07002030EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002031
Nick Pigginf7009262008-03-10 11:43:59 -07002032void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002033{
Nick Pigginf7009262008-03-10 11:43:59 -07002034 BUG_ON(i->count < bytes);
2035
Nick Piggin2f718ff2007-10-16 01:24:59 -07002036 if (likely(i->nr_segs == 1)) {
2037 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07002038 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002039 } else {
2040 const struct iovec *iov = i->iov;
2041 size_t base = i->iov_offset;
Jeff Layton39be79c2011-10-27 23:53:08 +02002042 unsigned long nr_segs = i->nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002043
Nick Piggin124d3b72008-02-02 15:01:17 +01002044 /*
2045 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07002046 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01002047 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07002048 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07002049 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002050
Nick Pigginf7009262008-03-10 11:43:59 -07002051 copy = min(bytes, iov->iov_len - base);
2052 BUG_ON(!i->count || i->count < copy);
2053 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002054 bytes -= copy;
2055 base += copy;
2056 if (iov->iov_len == base) {
2057 iov++;
Jeff Layton39be79c2011-10-27 23:53:08 +02002058 nr_segs--;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002059 base = 0;
2060 }
2061 }
2062 i->iov = iov;
2063 i->iov_offset = base;
Jeff Layton39be79c2011-10-27 23:53:08 +02002064 i->nr_segs = nr_segs;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002065 }
2066}
Nick Piggin89e10782007-10-16 01:25:07 -07002067EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002068
Nick Pigginafddba42007-10-16 01:25:01 -07002069/*
2070 * Fault in the first iovec of the given iov_iter, to a maximum length
2071 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2072 * accessed (ie. because it is an invalid address).
2073 *
2074 * writev-intensive code may want this to prefault several iovecs -- that
2075 * would be possible (callers must not rely on the fact that _only_ the
2076 * first iovec will be faulted with the current implementation).
2077 */
2078int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002079{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002080 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002081 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2082 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002083}
Nick Piggin89e10782007-10-16 01:25:07 -07002084EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002085
2086/*
2087 * Return the count of just the current iov_iter segment.
2088 */
Maxim Patlasovd28574e2012-10-26 19:50:04 +04002089size_t iov_iter_single_seg_count(const struct iov_iter *i)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002090{
2091 const struct iovec *iov = i->iov;
2092 if (i->nr_segs == 1)
2093 return i->count;
2094 else
2095 return min(i->count, iov->iov_len - i->iov_offset);
2096}
Nick Piggin89e10782007-10-16 01:25:07 -07002097EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002098
2099/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 * Performs necessary checks before doing a write
2101 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002102 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 * Returns appropriate error code that caller should return or
2104 * zero in case that write should be allowed.
2105 */
2106inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2107{
2108 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002109 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
2111 if (unlikely(*pos < 0))
2112 return -EINVAL;
2113
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 if (!isblk) {
2115 /* FIXME: this is for backwards compatibility with 2.4 */
2116 if (file->f_flags & O_APPEND)
2117 *pos = i_size_read(inode);
2118
2119 if (limit != RLIM_INFINITY) {
2120 if (*pos >= limit) {
2121 send_sig(SIGXFSZ, current, 0);
2122 return -EFBIG;
2123 }
2124 if (*count > limit - (typeof(limit))*pos) {
2125 *count = limit - (typeof(limit))*pos;
2126 }
2127 }
2128 }
2129
2130 /*
2131 * LFS rule
2132 */
2133 if (unlikely(*pos + *count > MAX_NON_LFS &&
2134 !(file->f_flags & O_LARGEFILE))) {
2135 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 return -EFBIG;
2137 }
2138 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2139 *count = MAX_NON_LFS - (unsigned long)*pos;
2140 }
2141 }
2142
2143 /*
2144 * Are we about to exceed the fs block limit ?
2145 *
2146 * If we have written data it becomes a short write. If we have
2147 * exceeded without writing data we send a signal and return EFBIG.
2148 * Linus frestrict idea will clean these up nicely..
2149 */
2150 if (likely(!isblk)) {
2151 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2152 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 return -EFBIG;
2154 }
2155 /* zero-length writes at ->s_maxbytes are OK */
2156 }
2157
2158 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2159 *count = inode->i_sb->s_maxbytes - *pos;
2160 } else {
David Howells93614012006-09-30 20:45:40 +02002161#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 loff_t isize;
2163 if (bdev_read_only(I_BDEV(inode)))
2164 return -EPERM;
2165 isize = i_size_read(inode);
2166 if (*pos >= isize) {
2167 if (*count || *pos > isize)
2168 return -ENOSPC;
2169 }
2170
2171 if (*pos + *count > isize)
2172 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002173#else
2174 return -EPERM;
2175#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 }
2177 return 0;
2178}
2179EXPORT_SYMBOL(generic_write_checks);
2180
Nick Pigginafddba42007-10-16 01:25:01 -07002181int pagecache_write_begin(struct file *file, struct address_space *mapping,
2182 loff_t pos, unsigned len, unsigned flags,
2183 struct page **pagep, void **fsdata)
2184{
2185 const struct address_space_operations *aops = mapping->a_ops;
2186
Nick Piggin4e02ed42008-10-29 14:00:55 -07002187 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002188 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002189}
2190EXPORT_SYMBOL(pagecache_write_begin);
2191
2192int pagecache_write_end(struct file *file, struct address_space *mapping,
2193 loff_t pos, unsigned len, unsigned copied,
2194 struct page *page, void *fsdata)
2195{
2196 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002197
Nick Piggin4e02ed42008-10-29 14:00:55 -07002198 mark_page_accessed(page);
2199 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002200}
2201EXPORT_SYMBOL(pagecache_write_end);
2202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203ssize_t
2204generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2205 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2206 size_t count, size_t ocount)
2207{
2208 struct file *file = iocb->ki_filp;
2209 struct address_space *mapping = file->f_mapping;
2210 struct inode *inode = mapping->host;
2211 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002212 size_t write_len;
2213 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214
2215 if (count != ocount)
2216 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2217
Christoph Hellwiga969e902008-07-23 21:27:04 -07002218 write_len = iov_length(iov, *nr_segs);
2219 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002220
Nick Piggin48b47c52009-01-06 14:40:22 -08002221 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002222 if (written)
2223 goto out;
2224
2225 /*
2226 * After a write we want buffered reads to be sure to go to disk to get
2227 * the new data. We invalidate clean cached page from the region we're
2228 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002229 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002230 */
2231 if (mapping->nrpages) {
2232 written = invalidate_inode_pages2_range(mapping,
2233 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002234 /*
2235 * If a page can not be invalidated, return 0 to fall back
2236 * to buffered write.
2237 */
2238 if (written) {
2239 if (written == -EBUSY)
2240 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002241 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002242 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002243 }
2244
2245 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2246
2247 /*
2248 * Finally, try again to invalidate clean pages which might have been
2249 * cached by non-direct readahead, or faulted in by get_user_pages()
2250 * if the source of the write was an mmap'ed region of the file
2251 * we're writing. Either one is a pretty crazy thing to do,
2252 * so we don't support it 100%. If this invalidation
2253 * fails, tough, the write still worked...
2254 */
2255 if (mapping->nrpages) {
2256 invalidate_inode_pages2_range(mapping,
2257 pos >> PAGE_CACHE_SHIFT, end);
2258 }
2259
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002261 pos += written;
2262 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2263 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 mark_inode_dirty(inode);
2265 }
Namhyung Kim01166512010-10-26 14:21:58 -07002266 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002268out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 return written;
2270}
2271EXPORT_SYMBOL(generic_file_direct_write);
2272
Nick Piggineb2be182007-10-16 01:24:57 -07002273/*
2274 * Find or create a page at the given pagecache position. Return the locked
2275 * page. This function is specifically for buffered writes.
2276 */
Nick Piggin54566b22009-01-04 12:00:53 -08002277struct page *grab_cache_page_write_begin(struct address_space *mapping,
2278 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002279{
2280 int status;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002281 gfp_t gfp_mask;
Nick Piggineb2be182007-10-16 01:24:57 -07002282 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002283 gfp_t gfp_notmask = 0;
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002284
Fengguang Wu1010bb12012-03-21 16:34:08 -07002285 gfp_mask = mapping_gfp_mask(mapping);
2286 if (mapping_cap_account_dirty(mapping))
2287 gfp_mask |= __GFP_WRITE;
Nick Piggin54566b22009-01-04 12:00:53 -08002288 if (flags & AOP_FLAG_NOFS)
2289 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002290repeat:
2291 page = find_lock_page(mapping, index);
Steven Rostedtc585a262011-01-13 15:46:18 -08002292 if (page)
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002293 goto found;
Nick Piggineb2be182007-10-16 01:24:57 -07002294
Johannes Weiner0faa70c2012-01-10 15:07:53 -08002295 page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002296 if (!page)
2297 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002298 status = add_to_page_cache_lru(page, mapping, index,
2299 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002300 if (unlikely(status)) {
2301 page_cache_release(page);
2302 if (status == -EEXIST)
2303 goto repeat;
2304 return NULL;
2305 }
Darrick J. Wong3d08bcc2011-05-27 12:23:34 -07002306found:
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002307 wait_for_stable_page(page);
Nick Piggineb2be182007-10-16 01:24:57 -07002308 return page;
2309}
Nick Piggin54566b22009-01-04 12:00:53 -08002310EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002311
Nick Pigginafddba42007-10-16 01:25:01 -07002312static ssize_t generic_perform_write(struct file *file,
2313 struct iov_iter *i, loff_t pos)
2314{
2315 struct address_space *mapping = file->f_mapping;
2316 const struct address_space_operations *a_ops = mapping->a_ops;
2317 long status = 0;
2318 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002319 unsigned int flags = 0;
2320
2321 /*
2322 * Copies from kernel address space cannot fail (NFSD is a big user).
2323 */
2324 if (segment_eq(get_fs(), KERNEL_DS))
2325 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002326
2327 do {
2328 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002329 unsigned long offset; /* Offset into pagecache page */
2330 unsigned long bytes; /* Bytes to write to page */
2331 size_t copied; /* Bytes copied from user */
2332 void *fsdata;
2333
2334 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002335 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2336 iov_iter_count(i));
2337
2338again:
Nick Pigginafddba42007-10-16 01:25:01 -07002339 /*
2340 * Bring in the user page that we will copy from _first_.
2341 * Otherwise there's a nasty deadlock on copying from the
2342 * same page as we're writing to, without it being marked
2343 * up-to-date.
2344 *
2345 * Not only is this an optimisation, but it is also required
2346 * to check that the address is actually valid, when atomic
2347 * usercopies are used, below.
2348 */
2349 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2350 status = -EFAULT;
2351 break;
2352 }
2353
Nick Piggin674b8922007-10-16 01:25:03 -07002354 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002355 &page, &fsdata);
2356 if (unlikely(status))
2357 break;
2358
anfei zhou931e80e2010-02-02 13:44:02 -08002359 if (mapping_writably_mapped(mapping))
2360 flush_dcache_page(page);
2361
Nick Pigginafddba42007-10-16 01:25:01 -07002362 pagefault_disable();
2363 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2364 pagefault_enable();
2365 flush_dcache_page(page);
2366
Josef Bacikc8236db2009-07-05 12:08:18 -07002367 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002368 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2369 page, fsdata);
2370 if (unlikely(status < 0))
2371 break;
2372 copied = status;
2373
2374 cond_resched();
2375
Nick Piggin124d3b72008-02-02 15:01:17 +01002376 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002377 if (unlikely(copied == 0)) {
2378 /*
2379 * If we were unable to copy any data at all, we must
2380 * fall back to a single segment length write.
2381 *
2382 * If we didn't fallback here, we could livelock
2383 * because not all segments in the iov can be copied at
2384 * once without a pagefault.
2385 */
2386 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2387 iov_iter_single_seg_count(i));
2388 goto again;
2389 }
Nick Pigginafddba42007-10-16 01:25:01 -07002390 pos += copied;
2391 written += copied;
2392
2393 balance_dirty_pages_ratelimited(mapping);
Jan Karaa50527b2011-12-02 09:17:02 +08002394 if (fatal_signal_pending(current)) {
2395 status = -EINTR;
2396 break;
2397 }
Nick Pigginafddba42007-10-16 01:25:01 -07002398 } while (iov_iter_count(i));
2399
2400 return written ? written : status;
2401}
2402
2403ssize_t
2404generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2405 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2406 size_t count, ssize_t written)
2407{
2408 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002409 ssize_t status;
2410 struct iov_iter i;
2411
2412 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002413 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002416 written += status;
2417 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 }
2419
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 return written ? written : status;
2421}
2422EXPORT_SYMBOL(generic_file_buffered_write);
2423
Jan Karae4dd9de2009-08-17 18:10:06 +02002424/**
2425 * __generic_file_aio_write - write data to a file
2426 * @iocb: IO state structure (file, offset, etc.)
2427 * @iov: vector with data to write
2428 * @nr_segs: number of segments in the vector
2429 * @ppos: position where to write
2430 *
2431 * This function does all the work needed for actually writing data to a
2432 * file. It does all basic checks, removes SUID from the file, updates
2433 * modification times and calls proper subroutines depending on whether we
2434 * do direct IO or a standard buffered write.
2435 *
2436 * It expects i_mutex to be grabbed unless we work on a block device or similar
2437 * object which does not need locking at all.
2438 *
2439 * This function does *not* take care of syncing data in case of O_SYNC write.
2440 * A caller has to handle it. This is mainly due to the fact that we want to
2441 * avoid syncing under i_mutex.
2442 */
2443ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2444 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445{
2446 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002447 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 size_t ocount; /* original count */
2449 size_t count; /* after file limit checks */
2450 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451 loff_t pos;
2452 ssize_t written;
2453 ssize_t err;
2454
2455 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002456 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2457 if (err)
2458 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
2460 count = ocount;
2461 pos = *ppos;
2462
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 /* We can write back this queue in page reclaim */
2464 current->backing_dev_info = mapping->backing_dev_info;
2465 written = 0;
2466
2467 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2468 if (err)
2469 goto out;
2470
2471 if (count == 0)
2472 goto out;
2473
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002474 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 if (err)
2476 goto out;
2477
Josef Bacikc3b2da32012-03-26 09:59:21 -04002478 err = file_update_time(file);
2479 if (err)
2480 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481
2482 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2483 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002484 loff_t endbyte;
2485 ssize_t written_buffered;
2486
2487 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2488 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 if (written < 0 || written == count)
2490 goto out;
2491 /*
2492 * direct-io write to a hole: fall through to buffered I/O
2493 * for completing the rest of the request.
2494 */
2495 pos += written;
2496 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002497 written_buffered = generic_file_buffered_write(iocb, iov,
2498 nr_segs, pos, ppos, count,
2499 written);
2500 /*
2501 * If generic_file_buffered_write() retuned a synchronous error
2502 * then we want to return the number of bytes which were
2503 * direct-written, or the error code if that was zero. Note
2504 * that this differs from normal direct-io semantics, which
2505 * will return -EFOO even if some bytes were written.
2506 */
2507 if (written_buffered < 0) {
2508 err = written_buffered;
2509 goto out;
2510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002512 /*
2513 * We need to ensure that the page cache pages are written to
2514 * disk and invalidated to preserve the expected O_DIRECT
2515 * semantics.
2516 */
2517 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002518 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002519 if (err == 0) {
2520 written = written_buffered;
2521 invalidate_mapping_pages(mapping,
2522 pos >> PAGE_CACHE_SHIFT,
2523 endbyte >> PAGE_CACHE_SHIFT);
2524 } else {
2525 /*
2526 * We don't know how much we wrote, so just return
2527 * the number of bytes which were direct-written
2528 */
2529 }
2530 } else {
2531 written = generic_file_buffered_write(iocb, iov, nr_segs,
2532 pos, ppos, count, written);
2533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534out:
2535 current->backing_dev_info = NULL;
2536 return written ? written : err;
2537}
Jan Karae4dd9de2009-08-17 18:10:06 +02002538EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Jan Karae4dd9de2009-08-17 18:10:06 +02002540/**
2541 * generic_file_aio_write - write data to a file
2542 * @iocb: IO state structure
2543 * @iov: vector with data to write
2544 * @nr_segs: number of segments in the vector
2545 * @pos: position in file where to write
2546 *
2547 * This is a wrapper around __generic_file_aio_write() to be used by most
2548 * filesystems. It takes care of syncing the file in case of O_SYNC file
2549 * and acquires i_mutex as needed.
2550 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002551ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2552 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553{
2554 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002555 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557
2558 BUG_ON(iocb->ki_pos != pos);
2559
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002560 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002561 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002562 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
Christoph Hellwig02afc272013-09-04 15:04:40 +02002564 if (ret > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 ssize_t err;
2566
Jan Kara148f9482009-08-17 19:52:36 +02002567 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002568 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569 ret = err;
2570 }
2571 return ret;
2572}
2573EXPORT_SYMBOL(generic_file_aio_write);
2574
David Howellscf9a2ae2006-08-29 19:05:54 +01002575/**
2576 * try_to_release_page() - release old fs-specific metadata on a page
2577 *
2578 * @page: the page which the kernel is trying to free
2579 * @gfp_mask: memory allocation flags (and I/O mode)
2580 *
2581 * The address_space is to try to release any data against the page
2582 * (presumably at page->private). If the release was successful, return `1'.
2583 * Otherwise return zero.
2584 *
David Howells266cf652009-04-03 16:42:36 +01002585 * This may also be called if PG_fscache is set on a page, indicating that the
2586 * page is known to the local caching routines.
2587 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002588 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002589 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002590 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002591 */
2592int try_to_release_page(struct page *page, gfp_t gfp_mask)
2593{
2594 struct address_space * const mapping = page->mapping;
2595
2596 BUG_ON(!PageLocked(page));
2597 if (PageWriteback(page))
2598 return 0;
2599
2600 if (mapping && mapping->a_ops->releasepage)
2601 return mapping->a_ops->releasepage(page, gfp_mask);
2602 return try_to_free_buffers(page);
2603}
2604
2605EXPORT_SYMBOL(try_to_release_page);