blob: ea89840fc65fa2b499ce3c19c657e9e8a680b240 [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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.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>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
Jan Kara148f9482009-08-17 19:52:36 +020042#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
46/*
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
49 *
50 * Shared mappings now work. 15.8.1995 Bruno.
51 *
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 *
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
56 */
57
58/*
59 * Lock ordering:
60 *
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +100061 * ->i_mmap_lock (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070063 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080066 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * ->i_mmap_lock (truncate->unmap_mapping_range)
68 *
69 * ->mmap_sem
70 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070071 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
73 *
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
76 *
Nick Piggin82591e62006-10-19 23:29:10 -070077 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080080 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * ->i_alloc_sem (various)
82 *
83 * ->inode_lock
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
86 *
87 * ->i_mmap_lock
88 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080098 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
104 *
105 * ->task->proc_lock
106 * ->dcache_lock (proc_pid_lookup)
Andi Kleen6a460792009-09-16 11:50:15 +0200107 *
108 * (code doesn't rely on that order, so you could switch it around)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
110 * ->i_mmap_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
112
113/*
114 * Remove a page from the page cache and free it. Caller has to make
115 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700116 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 */
118void __remove_from_page_cache(struct page *page)
119{
120 struct address_space *mapping = page->mapping;
121
122 radix_tree_delete(&mapping->page_tree, page->index);
123 page->mapping = NULL;
124 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700125 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700126 if (PageSwapBacked(page))
127 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700128 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800129
130 /*
131 * Some filesystems seem to re-dirty the page even after
132 * the VM has canceled the dirty bit (eg ext3 journaling).
133 *
134 * Fix it up by doing a final dirty accounting check after
135 * having removed the page entirely.
136 */
137 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
138 dec_zone_page_state(page, NR_FILE_DIRTY);
139 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
143void remove_from_page_cache(struct page *page)
144{
145 struct address_space *mapping = page->mapping;
146
Matt Mackallcd7619d2005-05-01 08:59:01 -0700147 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Nick Piggin19fd6232008-07-25 19:45:32 -0700149 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700151 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700152 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
Miklos Szeredia52116a2010-05-25 15:06:06 +0200154EXPORT_SYMBOL(remove_from_page_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156static int sync_page(void *word)
157{
158 struct address_space *mapping;
159 struct page *page;
160
Andi Kleen07808b72005-11-05 17:25:53 +0100161 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700164 * page_mapping() is being called without PG_locked held.
165 * Some knowledge of the state and use of the page is used to
166 * reduce the requirements down to a memory barrier.
167 * The danger here is of a stale page_mapping() return value
168 * indicating a struct address_space different from the one it's
169 * associated with when it is associated with one.
170 * After smp_mb(), it's either the correct page_mapping() for
171 * the page, or an old page_mapping() and the page's own
172 * page_mapping() has gone NULL.
173 * The ->sync_page() address_space operation must tolerate
174 * page_mapping() going NULL. By an amazing coincidence,
175 * this comes about because none of the users of the page
176 * in the ->sync_page() methods make essential use of the
177 * page_mapping(), merely passing the page down to the backing
178 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700179 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700180 * of interest. When page_mapping() does go NULL, the entire
181 * call stack gracefully ignores the page and returns.
182 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 */
184 smp_mb();
185 mapping = page_mapping(page);
186 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
187 mapping->a_ops->sync_page(page);
188 io_schedule();
189 return 0;
190}
191
Matthew Wilcox2687a352007-12-06 11:18:49 -0500192static int sync_page_killable(void *word)
193{
194 sync_page(word);
195 return fatal_signal_pending(current) ? -EINTR : 0;
196}
197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700199 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700200 * @mapping: address space structure to write
201 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800202 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700203 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700205 * Start writeback against all of a mapping's dirty pages that lie
206 * within the byte offsets <start, end> inclusive.
207 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700209 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 * these two operations is that if a dirty page/buffer is encountered, it must
211 * be waited upon, and not just skipped over.
212 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800213int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
214 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215{
216 int ret;
217 struct writeback_control wbc = {
218 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800219 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700220 .range_start = start,
221 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 };
223
224 if (!mapping_cap_writeback_dirty(mapping))
225 return 0;
226
227 ret = do_writepages(mapping, &wbc);
228 return ret;
229}
230
231static inline int __filemap_fdatawrite(struct address_space *mapping,
232 int sync_mode)
233{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700234 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235}
236
237int filemap_fdatawrite(struct address_space *mapping)
238{
239 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
240}
241EXPORT_SYMBOL(filemap_fdatawrite);
242
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400243int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800244 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
246 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
247}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400248EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Randy Dunlap485bb992006-06-23 02:03:49 -0700250/**
251 * filemap_flush - mostly a non-blocking flush
252 * @mapping: target address_space
253 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 * This is a mostly non-blocking flush. Not suitable for data-integrity
255 * purposes - I/O may not be started against all dirty pages.
256 */
257int filemap_flush(struct address_space *mapping)
258{
259 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
260}
261EXPORT_SYMBOL(filemap_flush);
262
Randy Dunlap485bb992006-06-23 02:03:49 -0700263/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200264 * filemap_fdatawait_range - wait for writeback to complete
265 * @mapping: address space structure to wait for
266 * @start_byte: offset in bytes where the range starts
267 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700268 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200269 * Walk the list of under-writeback pages of the given address space
270 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200272int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
273 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200275 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
276 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 struct pagevec pvec;
278 int nr_pages;
279 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200281 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 return 0;
283
284 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 while ((index <= end) &&
286 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
287 PAGECACHE_TAG_WRITEBACK,
288 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
289 unsigned i;
290
291 for (i = 0; i < nr_pages; i++) {
292 struct page *page = pvec.pages[i];
293
294 /* until radix tree lookup accepts end_index */
295 if (page->index > end)
296 continue;
297
298 wait_on_page_writeback(page);
299 if (PageError(page))
300 ret = -EIO;
301 }
302 pagevec_release(&pvec);
303 cond_resched();
304 }
305
306 /* Check for outstanding write errors */
307 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
308 ret = -ENOSPC;
309 if (test_and_clear_bit(AS_EIO, &mapping->flags))
310 ret = -EIO;
311
312 return ret;
313}
Jan Karad3bccb62009-08-17 19:30:27 +0200314EXPORT_SYMBOL(filemap_fdatawait_range);
315
316/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700317 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700319 *
320 * Walk the list of under-writeback pages of the given address space
321 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 */
323int filemap_fdatawait(struct address_space *mapping)
324{
325 loff_t i_size = i_size_read(mapping->host);
326
327 if (i_size == 0)
328 return 0;
329
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200330 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331}
332EXPORT_SYMBOL(filemap_fdatawait);
333
334int filemap_write_and_wait(struct address_space *mapping)
335{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800336 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800339 err = filemap_fdatawrite(mapping);
340 /*
341 * Even if the above returned error, the pages may be
342 * written partially (e.g. -ENOSPC), so we wait for it.
343 * But the -EIO is special case, it may indicate the worst
344 * thing (e.g. bug) happened, so we avoid waiting for it.
345 */
346 if (err != -EIO) {
347 int err2 = filemap_fdatawait(mapping);
348 if (!err)
349 err = err2;
350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800352 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800354EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
Randy Dunlap485bb992006-06-23 02:03:49 -0700356/**
357 * filemap_write_and_wait_range - write out & wait on a file range
358 * @mapping: the address_space for the pages
359 * @lstart: offset in bytes where the range starts
360 * @lend: offset in bytes where the range ends (inclusive)
361 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800362 * Write out and wait upon file offsets lstart->lend, inclusive.
363 *
364 * Note that `lend' is inclusive (describes the last byte to be written) so
365 * that this function can be used to write to the very end-of-file (end = -1).
366 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367int filemap_write_and_wait_range(struct address_space *mapping,
368 loff_t lstart, loff_t lend)
369{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800370 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
372 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800373 err = __filemap_fdatawrite_range(mapping, lstart, lend,
374 WB_SYNC_ALL);
375 /* See comment of filemap_write_and_wait() */
376 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200377 int err2 = filemap_fdatawait_range(mapping,
378 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800379 if (!err)
380 err = err2;
381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800383 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
Chris Masonf6995582009-04-15 13:22:37 -0400385EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
Randy Dunlap485bb992006-06-23 02:03:49 -0700387/**
Nick Piggine2867812008-07-25 19:45:30 -0700388 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700389 * @page: page to add
390 * @mapping: the page's address_space
391 * @offset: page index
392 * @gfp_mask: page allocation mode
393 *
Nick Piggine2867812008-07-25 19:45:30 -0700394 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 * This function does not add the page to the LRU. The caller must do that.
396 */
Nick Piggine2867812008-07-25 19:45:30 -0700397int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400398 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399{
Nick Piggine2867812008-07-25 19:45:30 -0700400 int error;
401
402 VM_BUG_ON(!PageLocked(page));
403
404 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800405 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800406 if (error)
407 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Balbir Singh35c754d2008-02-07 00:14:05 -0800409 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700411 page_cache_get(page);
412 page->mapping = mapping;
413 page->index = offset;
414
Nick Piggin19fd6232008-07-25 19:45:32 -0700415 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700417 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700419 __inc_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700420 if (PageSwapBacked(page))
421 __inc_zone_page_state(page, NR_SHMEM);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700422 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700423 } else {
424 page->mapping = NULL;
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700425 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700426 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700427 page_cache_release(page);
428 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800430 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700431 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800432out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 return error;
434}
Nick Piggine2867812008-07-25 19:45:30 -0700435EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
437int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400438 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700440 int ret;
441
442 /*
443 * Splice_read and readahead add shmem/tmpfs pages into the page cache
444 * before shmem_readpage has a chance to mark them as SwapBacked: they
KOSAKI Motohiroe9d6c152010-05-24 14:31:48 -0700445 * need to go on the anon lru below, and mem_cgroup_cache_charge
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700446 * (called in add_to_page_cache) needs to know where they're going too.
447 */
448 if (mapping_cap_swap_backed(mapping))
449 SetPageSwapBacked(page);
450
451 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
452 if (ret == 0) {
453 if (page_is_file_cache(page))
454 lru_cache_add_file(page);
455 else
KOSAKI Motohiroe9d6c152010-05-24 14:31:48 -0700456 lru_cache_add_anon(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700457 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 return ret;
459}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300460EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Paul Jackson44110fe2006-03-24 03:16:04 -0800462#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700463struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800464{
Miao Xiec0ff7452010-05-24 14:32:08 -0700465 int n;
466 struct page *page;
467
Paul Jackson44110fe2006-03-24 03:16:04 -0800468 if (cpuset_do_page_mem_spread()) {
Miao Xiec0ff7452010-05-24 14:32:08 -0700469 get_mems_allowed();
470 n = cpuset_mem_spread_node();
471 page = alloc_pages_exact_node(n, gfp, 0);
472 put_mems_allowed();
473 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800474 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700475 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800476}
Nick Piggin2ae88142006-10-28 10:38:23 -0700477EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800478#endif
479
Nick Piggindb376482006-09-25 23:31:24 -0700480static int __sleep_on_page_lock(void *word)
481{
482 io_schedule();
483 return 0;
484}
485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486/*
487 * In order to wait for pages to become available there must be
488 * waitqueues associated with pages. By using a hash table of
489 * waitqueues where the bucket discipline is to maintain all
490 * waiters on the same queue and wake all when any of the pages
491 * become available, and for the woken contexts to check to be
492 * sure the appropriate page became available, this saves space
493 * at a cost of "thundering herd" phenomena during rare hash
494 * collisions.
495 */
496static wait_queue_head_t *page_waitqueue(struct page *page)
497{
498 const struct zone *zone = page_zone(page);
499
500 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
501}
502
503static inline void wake_up_page(struct page *page, int bit)
504{
505 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
506}
507
Harvey Harrison920c7a52008-02-04 22:29:26 -0800508void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
510 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
511
512 if (test_bit(bit_nr, &page->flags))
513 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
514 TASK_UNINTERRUPTIBLE);
515}
516EXPORT_SYMBOL(wait_on_page_bit);
517
518/**
David Howells385e1ca2009-04-03 16:42:39 +0100519 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700520 * @page: Page defining the wait queue of interest
521 * @waiter: Waiter to add to the queue
David Howells385e1ca2009-04-03 16:42:39 +0100522 *
523 * Add an arbitrary @waiter to the wait queue for the nominated @page.
524 */
525void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
526{
527 wait_queue_head_t *q = page_waitqueue(page);
528 unsigned long flags;
529
530 spin_lock_irqsave(&q->lock, flags);
531 __add_wait_queue(q, waiter);
532 spin_unlock_irqrestore(&q->lock, flags);
533}
534EXPORT_SYMBOL_GPL(add_page_wait_queue);
535
536/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700537 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 * @page: the page
539 *
540 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
541 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
542 * mechananism between PageLocked pages and PageWriteback pages is shared.
543 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
544 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700545 * The mb is necessary to enforce ordering between the clear_bit and the read
546 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800548void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549{
Nick Piggin8413ac92008-10-18 20:26:59 -0700550 VM_BUG_ON(!PageLocked(page));
551 clear_bit_unlock(PG_locked, &page->flags);
552 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 wake_up_page(page, PG_locked);
554}
555EXPORT_SYMBOL(unlock_page);
556
Randy Dunlap485bb992006-06-23 02:03:49 -0700557/**
558 * end_page_writeback - end writeback against a page
559 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 */
561void end_page_writeback(struct page *page)
562{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700563 if (TestClearPageReclaim(page))
564 rotate_reclaimable_page(page);
565
566 if (!test_clear_page_writeback(page))
567 BUG();
568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 smp_mb__after_clear_bit();
570 wake_up_page(page, PG_writeback);
571}
572EXPORT_SYMBOL(end_page_writeback);
573
Randy Dunlap485bb992006-06-23 02:03:49 -0700574/**
575 * __lock_page - get a lock on the page, assuming we need to sleep to get it
576 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700578 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
580 * chances are that on the second loop, the block layer's plug list is empty,
581 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
582 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800583void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584{
585 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
586
587 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
588 TASK_UNINTERRUPTIBLE);
589}
590EXPORT_SYMBOL(__lock_page);
591
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800592int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500593{
594 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
595
596 return __wait_on_bit_lock(page_waitqueue(page), &wait,
597 sync_page_killable, TASK_KILLABLE);
598}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300599EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500600
Randy Dunlap76824862008-03-19 17:00:40 -0700601/**
602 * __lock_page_nosync - get a lock on the page, without calling sync_page()
603 * @page: the page to lock
604 *
Nick Piggindb376482006-09-25 23:31:24 -0700605 * Variant of lock_page that does not require the caller to hold a reference
606 * on the page's mapping.
607 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800608void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700609{
610 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
611 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
612 TASK_UNINTERRUPTIBLE);
613}
614
Michel Lespinassed065bd82010-10-26 14:21:57 -0700615int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
616 unsigned int flags)
617{
618 if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
619 __lock_page(page);
620 return 1;
621 } else {
622 up_read(&mm->mmap_sem);
623 wait_on_page_locked(page);
624 return 0;
625 }
626}
627
Randy Dunlap485bb992006-06-23 02:03:49 -0700628/**
629 * find_get_page - find and get a page reference
630 * @mapping: the address_space to search
631 * @offset: the page index
632 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700633 * Is there a pagecache struct page at the given (mapping, offset) tuple?
634 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 */
Nick Piggina60637c2008-07-25 19:45:31 -0700636struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
Nick Piggina60637c2008-07-25 19:45:31 -0700638 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 struct page *page;
640
Nick Piggina60637c2008-07-25 19:45:31 -0700641 rcu_read_lock();
642repeat:
643 page = NULL;
644 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
645 if (pagep) {
646 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800647 if (unlikely(!page))
648 goto out;
649 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700650 goto repeat;
651
652 if (!page_cache_get_speculative(page))
653 goto repeat;
654
655 /*
656 * Has the page moved?
657 * This is part of the lockless pagecache protocol. See
658 * include/linux/pagemap.h for details.
659 */
660 if (unlikely(page != *pagep)) {
661 page_cache_release(page);
662 goto repeat;
663 }
664 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800665out:
Nick Piggina60637c2008-07-25 19:45:31 -0700666 rcu_read_unlock();
667
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 return page;
669}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670EXPORT_SYMBOL(find_get_page);
671
Randy Dunlap485bb992006-06-23 02:03:49 -0700672/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700674 * @mapping: the address_space to search
675 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 *
677 * Locates the desired pagecache page, locks it, increments its reference
678 * count and returns its address.
679 *
680 * Returns zero if the page was not present. find_lock_page() may sleep.
681 */
Nick Piggina60637c2008-07-25 19:45:31 -0700682struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683{
684 struct page *page;
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700687 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700689 lock_page(page);
690 /* Has the page been truncated? */
691 if (unlikely(page->mapping != mapping)) {
692 unlock_page(page);
693 page_cache_release(page);
694 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
Nick Piggina60637c2008-07-25 19:45:31 -0700696 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 return page;
699}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700EXPORT_SYMBOL(find_lock_page);
701
702/**
703 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700704 * @mapping: the page's address_space
705 * @index: the page's index into the mapping
706 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 *
708 * Locates a page in the pagecache. If the page is not present, a new page
709 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
710 * LRU list. The returned page is locked and has its reference count
711 * incremented.
712 *
713 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
714 * allocation!
715 *
716 * find_or_create_page() returns the desired page's address, or zero on
717 * memory exhaustion.
718 */
719struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700720 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
Nick Piggineb2be182007-10-16 01:24:57 -0700722 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 int err;
724repeat:
725 page = find_lock_page(mapping, index);
726 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700727 page = __page_cache_alloc(gfp_mask);
728 if (!page)
729 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800730 /*
731 * We want a regular kernel memory (not highmem or DMA etc)
732 * allocation for the radix tree nodes, but we need to honour
733 * the context-specific requirements the caller has asked for.
734 * GFP_RECLAIM_MASK collects those requirements.
735 */
736 err = add_to_page_cache_lru(page, mapping, index,
737 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700738 if (unlikely(err)) {
739 page_cache_release(page);
740 page = NULL;
741 if (err == -EEXIST)
742 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return page;
746}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747EXPORT_SYMBOL(find_or_create_page);
748
749/**
750 * find_get_pages - gang pagecache lookup
751 * @mapping: The address_space to search
752 * @start: The starting page index
753 * @nr_pages: The maximum number of pages
754 * @pages: Where the resulting pages are placed
755 *
756 * find_get_pages() will search for and return a group of up to
757 * @nr_pages pages in the mapping. The pages are placed at @pages.
758 * find_get_pages() takes a reference against the returned pages.
759 *
760 * The search returns a group of mapping-contiguous pages with ascending
761 * indexes. There may be holes in the indices due to not-present pages.
762 *
763 * find_get_pages() returns the number of pages which were found.
764 */
765unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
766 unsigned int nr_pages, struct page **pages)
767{
768 unsigned int i;
769 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700770 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Nick Piggina60637c2008-07-25 19:45:31 -0700772 rcu_read_lock();
773restart:
774 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
775 (void ***)pages, start, nr_pages);
776 ret = 0;
777 for (i = 0; i < nr_found; i++) {
778 struct page *page;
779repeat:
780 page = radix_tree_deref_slot((void **)pages[i]);
781 if (unlikely(!page))
782 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800783 if (radix_tree_deref_retry(page)) {
784 if (ret)
785 start = pages[ret-1]->index;
Nick Piggina60637c2008-07-25 19:45:31 -0700786 goto restart;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800787 }
Nick Piggina60637c2008-07-25 19:45:31 -0700788
789 if (!page_cache_get_speculative(page))
790 goto repeat;
791
792 /* Has the page moved? */
793 if (unlikely(page != *((void **)pages[i]))) {
794 page_cache_release(page);
795 goto repeat;
796 }
797
798 pages[ret] = page;
799 ret++;
800 }
801 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 return ret;
803}
804
Jens Axboeebf43502006-04-27 08:46:01 +0200805/**
806 * find_get_pages_contig - gang contiguous pagecache lookup
807 * @mapping: The address_space to search
808 * @index: The starting page index
809 * @nr_pages: The maximum number of pages
810 * @pages: Where the resulting pages are placed
811 *
812 * find_get_pages_contig() works exactly like find_get_pages(), except
813 * that the returned number of pages are guaranteed to be contiguous.
814 *
815 * find_get_pages_contig() returns the number of pages which were found.
816 */
817unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
818 unsigned int nr_pages, struct page **pages)
819{
820 unsigned int i;
821 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700822 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200823
Nick Piggina60637c2008-07-25 19:45:31 -0700824 rcu_read_lock();
825restart:
826 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
827 (void ***)pages, index, nr_pages);
828 ret = 0;
829 for (i = 0; i < nr_found; i++) {
830 struct page *page;
831repeat:
832 page = radix_tree_deref_slot((void **)pages[i]);
833 if (unlikely(!page))
834 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800835 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700836 goto restart;
837
838 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200839 break;
840
Nick Piggina60637c2008-07-25 19:45:31 -0700841 if (!page_cache_get_speculative(page))
842 goto repeat;
843
844 /* Has the page moved? */
845 if (unlikely(page != *((void **)pages[i]))) {
846 page_cache_release(page);
847 goto repeat;
848 }
849
850 pages[ret] = page;
851 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200852 index++;
853 }
Nick Piggina60637c2008-07-25 19:45:31 -0700854 rcu_read_unlock();
855 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200856}
David Howellsef71c152007-05-09 02:33:44 -0700857EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200858
Randy Dunlap485bb992006-06-23 02:03:49 -0700859/**
860 * find_get_pages_tag - find and return pages that match @tag
861 * @mapping: the address_space to search
862 * @index: the starting page index
863 * @tag: the tag index
864 * @nr_pages: the maximum number of pages
865 * @pages: where the resulting pages are placed
866 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700868 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 */
870unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
871 int tag, unsigned int nr_pages, struct page **pages)
872{
873 unsigned int i;
874 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700875 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Nick Piggina60637c2008-07-25 19:45:31 -0700877 rcu_read_lock();
878restart:
879 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
880 (void ***)pages, *index, nr_pages, tag);
881 ret = 0;
882 for (i = 0; i < nr_found; i++) {
883 struct page *page;
884repeat:
885 page = radix_tree_deref_slot((void **)pages[i]);
886 if (unlikely(!page))
887 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800888 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700889 goto restart;
890
891 if (!page_cache_get_speculative(page))
892 goto repeat;
893
894 /* Has the page moved? */
895 if (unlikely(page != *((void **)pages[i]))) {
896 page_cache_release(page);
897 goto repeat;
898 }
899
900 pages[ret] = page;
901 ret++;
902 }
903 rcu_read_unlock();
904
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 if (ret)
906 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 return ret;
909}
David Howellsef71c152007-05-09 02:33:44 -0700910EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
Randy Dunlap485bb992006-06-23 02:03:49 -0700912/**
913 * grab_cache_page_nowait - returns locked page at given index in given cache
914 * @mapping: target address_space
915 * @index: the page index
916 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800917 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 * This is intended for speculative data generators, where the data can
919 * be regenerated if the page couldn't be grabbed. This routine should
920 * be safe to call while holding the lock for another page.
921 *
922 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
923 * and deadlock against the caller's locked page.
924 */
925struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700926grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927{
928 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
930 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200931 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 return page;
933 page_cache_release(page);
934 return NULL;
935 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700936 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800937 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 page_cache_release(page);
939 page = NULL;
940 }
941 return page;
942}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943EXPORT_SYMBOL(grab_cache_page_nowait);
944
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700945/*
946 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
947 * a _large_ part of the i/o request. Imagine the worst scenario:
948 *
949 * ---R__________________________________________B__________
950 * ^ reading here ^ bad block(assume 4k)
951 *
952 * read(R) => miss => readahead(R...B) => media error => frustrating retries
953 * => failing the whole request => read(R) => read(R+1) =>
954 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
955 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
956 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
957 *
958 * It is going insane. Fix it by quickly scaling down the readahead size.
959 */
960static void shrink_readahead_size_eio(struct file *filp,
961 struct file_ra_state *ra)
962{
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700963 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700964}
965
Randy Dunlap485bb992006-06-23 02:03:49 -0700966/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800967 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700968 * @filp: the file to read
969 * @ppos: current file position
970 * @desc: read_descriptor
971 * @actor: read method
972 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700974 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 *
976 * This is really ugly. But the goto's actually try to clarify some
977 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800979static void do_generic_file_read(struct file *filp, loff_t *ppos,
980 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800982 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800984 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700985 pgoff_t index;
986 pgoff_t last_index;
987 pgoff_t prev_index;
988 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700989 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700993 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
994 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
996 offset = *ppos & ~PAGE_CACHE_MASK;
997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 for (;;) {
999 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001000 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001001 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 unsigned long nr, ret;
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005find_page:
1006 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001007 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001008 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001009 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001010 index, last_index - index);
1011 page = find_get_page(mapping, index);
1012 if (unlikely(page == NULL))
1013 goto no_cached_page;
1014 }
1015 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001016 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001017 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001018 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001020 if (!PageUptodate(page)) {
1021 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1022 !mapping->a_ops->is_partially_uptodate)
1023 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001024 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001025 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001026 /* Did it get truncated before we got the lock? */
1027 if (!page->mapping)
1028 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001029 if (!mapping->a_ops->is_partially_uptodate(page,
1030 desc, offset))
1031 goto page_not_up_to_date_locked;
1032 unlock_page(page);
1033 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001035 /*
1036 * i_size must be checked after we know the page is Uptodate.
1037 *
1038 * Checking i_size after the check allows us to calculate
1039 * the correct value for "nr", which means the zero-filled
1040 * part of the page is not copied back to userspace (unless
1041 * another truncate extends the file - this is desired though).
1042 */
1043
1044 isize = i_size_read(inode);
1045 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1046 if (unlikely(!isize || index > end_index)) {
1047 page_cache_release(page);
1048 goto out;
1049 }
1050
1051 /* nr is the maximum number of bytes to copy from this page */
1052 nr = PAGE_CACHE_SIZE;
1053 if (index == end_index) {
1054 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1055 if (nr <= offset) {
1056 page_cache_release(page);
1057 goto out;
1058 }
1059 }
1060 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
1062 /* If users can be writing to this page using arbitrary
1063 * virtual addresses, take care about potential aliasing
1064 * before reading the page on the kernel side.
1065 */
1066 if (mapping_writably_mapped(mapping))
1067 flush_dcache_page(page);
1068
1069 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001070 * When a sequential read accesses a page several times,
1071 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 */
Jan Karaec0f1632007-05-06 14:49:25 -07001073 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 mark_page_accessed(page);
1075 prev_index = index;
1076
1077 /*
1078 * Ok, we have the page, and it's up-to-date, so
1079 * now we can copy it to user space...
1080 *
1081 * The actor routine returns how many bytes were actually used..
1082 * NOTE! This may not be the same as how much of a user buffer
1083 * we filled up (we may be padding etc), so we can only update
1084 * "pos" here (the actor routine has to update the user buffer
1085 * pointers and the remaining count).
1086 */
1087 ret = actor(desc, page, offset, nr);
1088 offset += ret;
1089 index += offset >> PAGE_CACHE_SHIFT;
1090 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001091 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 page_cache_release(page);
1094 if (ret == nr && desc->count)
1095 continue;
1096 goto out;
1097
1098page_not_up_to_date:
1099 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001100 error = lock_page_killable(page);
1101 if (unlikely(error))
1102 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001104page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001105 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (!page->mapping) {
1107 unlock_page(page);
1108 page_cache_release(page);
1109 continue;
1110 }
1111
1112 /* Did somebody else fill it already? */
1113 if (PageUptodate(page)) {
1114 unlock_page(page);
1115 goto page_ok;
1116 }
1117
1118readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001119 /*
1120 * A previous I/O error may have been due to temporary
1121 * failures, eg. multipath errors.
1122 * PG_error will be set again if readpage fails.
1123 */
1124 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 /* Start the actual read. The read will unlock the page. */
1126 error = mapping->a_ops->readpage(filp, page);
1127
Zach Brown994fc28c2005-12-15 14:28:17 -08001128 if (unlikely(error)) {
1129 if (error == AOP_TRUNCATED_PAGE) {
1130 page_cache_release(page);
1131 goto find_page;
1132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001137 error = lock_page_killable(page);
1138 if (unlikely(error))
1139 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if (!PageUptodate(page)) {
1141 if (page->mapping == NULL) {
1142 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001143 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 */
1145 unlock_page(page);
1146 page_cache_release(page);
1147 goto find_page;
1148 }
1149 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001150 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001151 error = -EIO;
1152 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 }
1154 unlock_page(page);
1155 }
1156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 goto page_ok;
1158
1159readpage_error:
1160 /* UHHUH! A synchronous read error occurred. Report it */
1161 desc->error = error;
1162 page_cache_release(page);
1163 goto out;
1164
1165no_cached_page:
1166 /*
1167 * Ok, it wasn't cached, so we need to create a new
1168 * page..
1169 */
Nick Piggineb2be182007-10-16 01:24:57 -07001170 page = page_cache_alloc_cold(mapping);
1171 if (!page) {
1172 desc->error = -ENOMEM;
1173 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 }
Nick Piggineb2be182007-10-16 01:24:57 -07001175 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 index, GFP_KERNEL);
1177 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001178 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (error == -EEXIST)
1180 goto find_page;
1181 desc->error = error;
1182 goto out;
1183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 goto readpage;
1185 }
1186
1187out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001188 ra->prev_pos = prev_index;
1189 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1190 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001192 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001193 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
1196int file_read_actor(read_descriptor_t *desc, struct page *page,
1197 unsigned long offset, unsigned long size)
1198{
1199 char *kaddr;
1200 unsigned long left, count = desc->count;
1201
1202 if (size > count)
1203 size = count;
1204
1205 /*
1206 * Faults on the destination of a read are common, so do it before
1207 * taking the kmap.
1208 */
1209 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1210 kaddr = kmap_atomic(page, KM_USER0);
1211 left = __copy_to_user_inatomic(desc->arg.buf,
1212 kaddr + offset, size);
1213 kunmap_atomic(kaddr, KM_USER0);
1214 if (left == 0)
1215 goto success;
1216 }
1217
1218 /* Do it the slow way */
1219 kaddr = kmap(page);
1220 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1221 kunmap(page);
1222
1223 if (left) {
1224 size -= left;
1225 desc->error = -EFAULT;
1226 }
1227success:
1228 desc->count = count - size;
1229 desc->written += size;
1230 desc->arg.buf += size;
1231 return size;
1232}
1233
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001234/*
1235 * Performs necessary checks before doing a write
1236 * @iov: io vector request
1237 * @nr_segs: number of segments in the iovec
1238 * @count: number of bytes to write
1239 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1240 *
1241 * Adjust number of segments and amount of bytes to write (nr_segs should be
1242 * properly initialized first). Returns appropriate error code that caller
1243 * should return or zero in case that write should be allowed.
1244 */
1245int generic_segment_checks(const struct iovec *iov,
1246 unsigned long *nr_segs, size_t *count, int access_flags)
1247{
1248 unsigned long seg;
1249 size_t cnt = 0;
1250 for (seg = 0; seg < *nr_segs; seg++) {
1251 const struct iovec *iv = &iov[seg];
1252
1253 /*
1254 * If any segment has a negative length, or the cumulative
1255 * length ever wraps negative then return -EINVAL.
1256 */
1257 cnt += iv->iov_len;
1258 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1259 return -EINVAL;
1260 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1261 continue;
1262 if (seg == 0)
1263 return -EFAULT;
1264 *nr_segs = seg;
1265 cnt -= iv->iov_len; /* This segment is no good */
1266 break;
1267 }
1268 *count = cnt;
1269 return 0;
1270}
1271EXPORT_SYMBOL(generic_segment_checks);
1272
Randy Dunlap485bb992006-06-23 02:03:49 -07001273/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001274 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001275 * @iocb: kernel I/O control block
1276 * @iov: io vector request
1277 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001278 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001279 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 * This is the "read()" routine for all filesystems
1281 * that can use the page cache directly.
1282 */
1283ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001284generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1285 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286{
1287 struct file *filp = iocb->ki_filp;
1288 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001289 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001291 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
1293 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001294 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1295 if (retval)
1296 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1299 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001300 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 struct address_space *mapping;
1302 struct inode *inode;
1303
1304 mapping = filp->f_mapping;
1305 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 if (!count)
1307 goto out; /* skip atime */
1308 size = i_size_read(inode);
1309 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001310 retval = filemap_write_and_wait_range(mapping, pos,
1311 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001312 if (!retval) {
1313 retval = mapping->a_ops->direct_IO(READ, iocb,
1314 iov, pos, nr_segs);
1315 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001316 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001318 count -= retval;
1319 }
1320
1321 /*
1322 * Btrfs can have a short DIO read if we encounter
1323 * compressed extents, so if there was an error, or if
1324 * we've already read everything we wanted to, or if
1325 * there was a short read because we hit EOF, go ahead
1326 * and return. Otherwise fallthrough to buffered io for
1327 * the rest of the read.
1328 */
1329 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001330 file_accessed(filp);
1331 goto out;
1332 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
1335
Josef Bacik66f998f2010-05-23 11:00:54 -04001336 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001337 for (seg = 0; seg < nr_segs; seg++) {
1338 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001339 loff_t offset = 0;
1340
1341 /*
1342 * If we did a short DIO read we need to skip the section of the
1343 * iov that we've already read data into.
1344 */
1345 if (count) {
1346 if (count > iov[seg].iov_len) {
1347 count -= iov[seg].iov_len;
1348 continue;
1349 }
1350 offset = count;
1351 count = 0;
1352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
Hugh Dickins11fa9772008-07-23 21:27:34 -07001354 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001355 desc.arg.buf = iov[seg].iov_base + offset;
1356 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001357 if (desc.count == 0)
1358 continue;
1359 desc.error = 0;
1360 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1361 retval += desc.written;
1362 if (desc.error) {
1363 retval = retval ?: desc.error;
1364 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001366 if (desc.count > 0)
1367 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
1369out:
1370 return retval;
1371}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372EXPORT_SYMBOL(generic_file_aio_read);
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374static ssize_t
1375do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001376 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
1378 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1379 return -EINVAL;
1380
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001381 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 return 0;
1383}
1384
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001385SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386{
1387 ssize_t ret;
1388 struct file *file;
1389
1390 ret = -EBADF;
1391 file = fget(fd);
1392 if (file) {
1393 if (file->f_mode & FMODE_READ) {
1394 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001395 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1396 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 unsigned long len = end - start + 1;
1398 ret = do_readahead(mapping, file, start, len);
1399 }
1400 fput(file);
1401 }
1402 return ret;
1403}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001404#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1405asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1406{
1407 return SYSC_readahead((int) fd, offset, (size_t) count);
1408}
1409SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1410#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
1412#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001413/**
1414 * page_cache_read - adds requested page to the page cache if not already there
1415 * @file: file to read
1416 * @offset: page index
1417 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 * This adds the requested page to the page cache if it isn't already there,
1419 * and schedules an I/O to read in its contents from disk.
1420 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001421static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422{
1423 struct address_space *mapping = file->f_mapping;
1424 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001425 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Zach Brown994fc28c2005-12-15 14:28:17 -08001427 do {
1428 page = page_cache_alloc_cold(mapping);
1429 if (!page)
1430 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Zach Brown994fc28c2005-12-15 14:28:17 -08001432 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1433 if (ret == 0)
1434 ret = mapping->a_ops->readpage(file, page);
1435 else if (ret == -EEXIST)
1436 ret = 0; /* losing race to add is OK */
1437
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Zach Brown994fc28c2005-12-15 14:28:17 -08001440 } while (ret == AOP_TRUNCATED_PAGE);
1441
1442 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
1445#define MMAP_LOTSAMISS (100)
1446
Linus Torvaldsef00e082009-06-16 15:31:25 -07001447/*
1448 * Synchronous readahead happens when we don't even find
1449 * a page in the page cache at all.
1450 */
1451static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1452 struct file_ra_state *ra,
1453 struct file *file,
1454 pgoff_t offset)
1455{
1456 unsigned long ra_pages;
1457 struct address_space *mapping = file->f_mapping;
1458
1459 /* If we don't want any read-ahead, don't bother */
1460 if (VM_RandomReadHint(vma))
1461 return;
1462
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001463 if (VM_SequentialReadHint(vma) ||
1464 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001465 page_cache_sync_readahead(mapping, ra, file, offset,
1466 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001467 return;
1468 }
1469
1470 if (ra->mmap_miss < INT_MAX)
1471 ra->mmap_miss++;
1472
1473 /*
1474 * Do we miss much more than hit in this file? If so,
1475 * stop bothering with read-ahead. It will only hurt.
1476 */
1477 if (ra->mmap_miss > MMAP_LOTSAMISS)
1478 return;
1479
Wu Fengguangd30a1102009-06-16 15:31:30 -07001480 /*
1481 * mmap read-around
1482 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001483 ra_pages = max_sane_readahead(ra->ra_pages);
1484 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001485 ra->start = max_t(long, 0, offset - ra_pages/2);
1486 ra->size = ra_pages;
1487 ra->async_size = 0;
1488 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001489 }
1490}
1491
1492/*
1493 * Asynchronous readahead happens when we find the page and PG_readahead,
1494 * so we want to possibly extend the readahead further..
1495 */
1496static void do_async_mmap_readahead(struct vm_area_struct *vma,
1497 struct file_ra_state *ra,
1498 struct file *file,
1499 struct page *page,
1500 pgoff_t offset)
1501{
1502 struct address_space *mapping = file->f_mapping;
1503
1504 /* If we don't want any read-ahead, don't bother */
1505 if (VM_RandomReadHint(vma))
1506 return;
1507 if (ra->mmap_miss > 0)
1508 ra->mmap_miss--;
1509 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001510 page_cache_async_readahead(mapping, ra, file,
1511 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001512}
1513
Randy Dunlap485bb992006-06-23 02:03:49 -07001514/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001515 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001516 * @vma: vma in which the fault was taken
1517 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001518 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001519 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 * mapped memory region to read in file data during a page fault.
1521 *
1522 * The goto's are kind of ugly, but this streamlines the normal case of having
1523 * it in the page cache, and handles the special cases reasonably without
1524 * having a lot of duplicated code.
1525 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001526int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527{
1528 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001529 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 struct address_space *mapping = file->f_mapping;
1531 struct file_ra_state *ra = &file->f_ra;
1532 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001533 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001535 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001536 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001539 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001540 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 * Do we have something in the page cache already?
1544 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001545 page = find_get_page(mapping, offset);
1546 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001548 * We found the page, so try async readahead before
1549 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001551 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001552 } else {
1553 /* No page in the page cache at all */
1554 do_sync_mmap_readahead(vma, ra, file, offset);
1555 count_vm_event(PGMAJFAULT);
1556 ret = VM_FAULT_MAJOR;
1557retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001558 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 if (!page)
1560 goto no_cached_page;
1561 }
1562
Michel Lespinassed88c0922010-11-02 13:05:18 -07001563 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1564 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001565 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001566 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001567
1568 /* Did it get truncated? */
1569 if (unlikely(page->mapping != mapping)) {
1570 unlock_page(page);
1571 put_page(page);
1572 goto retry_find;
1573 }
1574 VM_BUG_ON(page->index != offset);
1575
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001577 * We have a locked page in the page cache, now we need to check
1578 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 */
Nick Piggind00806b2007-07-19 01:46:57 -07001580 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 goto page_not_uptodate;
1582
Linus Torvaldsef00e082009-06-16 15:31:25 -07001583 /*
1584 * Found the page and have a reference on it.
1585 * We must recheck i_size under page lock.
1586 */
Nick Piggind00806b2007-07-19 01:46:57 -07001587 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001588 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001589 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001590 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001591 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001592 }
1593
Linus Torvaldsef00e082009-06-16 15:31:25 -07001594 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001595 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001596 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598no_cached_page:
1599 /*
1600 * We're only likely to ever get here if MADV_RANDOM is in
1601 * effect.
1602 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001603 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605 /*
1606 * The page we want has now been added to the page cache.
1607 * In the unlikely event that someone removed it in the
1608 * meantime, we'll just come back here and read it again.
1609 */
1610 if (error >= 0)
1611 goto retry_find;
1612
1613 /*
1614 * An error return from page_cache_read can result if the
1615 * system is low on memory, or a problem occurs while trying
1616 * to schedule I/O.
1617 */
1618 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001619 return VM_FAULT_OOM;
1620 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621
1622page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 /*
1624 * Umm, take care of errors if the page isn't up-to-date.
1625 * Try to re-read it _once_. We do this synchronously,
1626 * because there really aren't any performance issues here
1627 * and we need to check for errors.
1628 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001630 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001631 if (!error) {
1632 wait_on_page_locked(page);
1633 if (!PageUptodate(page))
1634 error = -EIO;
1635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001637
1638 if (!error || error == AOP_TRUNCATED_PAGE)
1639 goto retry_find;
1640
1641 /* Things didn't work out. Return zero to tell the mm layer so. */
1642 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001643 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001644}
1645EXPORT_SYMBOL(filemap_fault);
1646
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001647const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001648 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649};
1650
1651/* This is used for a general mmap of a disk file */
1652
1653int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1654{
1655 struct address_space *mapping = file->f_mapping;
1656
1657 if (!mapping->a_ops->readpage)
1658 return -ENOEXEC;
1659 file_accessed(file);
1660 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001661 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 return 0;
1663}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665/*
1666 * This is for filesystems which do not implement ->writepage.
1667 */
1668int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1669{
1670 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1671 return -EINVAL;
1672 return generic_file_mmap(file, vma);
1673}
1674#else
1675int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1676{
1677 return -ENOSYS;
1678}
1679int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1680{
1681 return -ENOSYS;
1682}
1683#endif /* CONFIG_MMU */
1684
1685EXPORT_SYMBOL(generic_file_mmap);
1686EXPORT_SYMBOL(generic_file_readonly_mmap);
1687
Nick Piggin6fe69002007-05-06 14:49:04 -07001688static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001689 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001691 void *data,
1692 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693{
Nick Piggineb2be182007-10-16 01:24:57 -07001694 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 int err;
1696repeat:
1697 page = find_get_page(mapping, index);
1698 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001699 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001700 if (!page)
1701 return ERR_PTR(-ENOMEM);
1702 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1703 if (unlikely(err)) {
1704 page_cache_release(page);
1705 if (err == -EEXIST)
1706 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 return ERR_PTR(err);
1709 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 err = filler(data, page);
1711 if (err < 0) {
1712 page_cache_release(page);
1713 page = ERR_PTR(err);
1714 }
1715 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return page;
1717}
1718
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001719static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001720 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001722 void *data,
1723 gfp_t gfp)
1724
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725{
1726 struct page *page;
1727 int err;
1728
1729retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001730 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001732 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 if (PageUptodate(page))
1734 goto out;
1735
1736 lock_page(page);
1737 if (!page->mapping) {
1738 unlock_page(page);
1739 page_cache_release(page);
1740 goto retry;
1741 }
1742 if (PageUptodate(page)) {
1743 unlock_page(page);
1744 goto out;
1745 }
1746 err = filler(data, page);
1747 if (err < 0) {
1748 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001749 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 }
David Howellsc855ff32007-05-09 13:42:20 +01001751out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001752 mark_page_accessed(page);
1753 return page;
1754}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001755
1756/**
1757 * read_cache_page_async - read into page cache, fill it if needed
1758 * @mapping: the page's address_space
1759 * @index: the page index
1760 * @filler: function to perform the read
1761 * @data: destination for read data
1762 *
1763 * Same as read_cache_page, but don't wait for page to become unlocked
1764 * after submitting it to the filler.
1765 *
1766 * Read into the page cache. If a page already exists, and PageUptodate() is
1767 * not set, try to fill the page but don't wait for it to become unlocked.
1768 *
1769 * If the page does not get brought uptodate, return -EIO.
1770 */
1771struct page *read_cache_page_async(struct address_space *mapping,
1772 pgoff_t index,
1773 int (*filler)(void *,struct page*),
1774 void *data)
1775{
1776 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1777}
Nick Piggin6fe69002007-05-06 14:49:04 -07001778EXPORT_SYMBOL(read_cache_page_async);
1779
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001780static struct page *wait_on_page_read(struct page *page)
1781{
1782 if (!IS_ERR(page)) {
1783 wait_on_page_locked(page);
1784 if (!PageUptodate(page)) {
1785 page_cache_release(page);
1786 page = ERR_PTR(-EIO);
1787 }
1788 }
1789 return page;
1790}
1791
1792/**
1793 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1794 * @mapping: the page's address_space
1795 * @index: the page index
1796 * @gfp: the page allocator flags to use if allocating
1797 *
1798 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1799 * any new page allocations done using the specified allocation flags. Note
1800 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1801 * expect to do this atomically or anything like that - but you can pass in
1802 * other page requirements.
1803 *
1804 * If the page does not get brought uptodate, return -EIO.
1805 */
1806struct page *read_cache_page_gfp(struct address_space *mapping,
1807 pgoff_t index,
1808 gfp_t gfp)
1809{
1810 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1811
1812 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1813}
1814EXPORT_SYMBOL(read_cache_page_gfp);
1815
Nick Piggin6fe69002007-05-06 14:49:04 -07001816/**
1817 * read_cache_page - read into page cache, fill it if needed
1818 * @mapping: the page's address_space
1819 * @index: the page index
1820 * @filler: function to perform the read
1821 * @data: destination for read data
1822 *
1823 * Read into the page cache. If a page already exists, and PageUptodate() is
1824 * not set, try to fill the page then wait for it to become unlocked.
1825 *
1826 * If the page does not get brought uptodate, return -EIO.
1827 */
1828struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001829 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001830 int (*filler)(void *,struct page*),
1831 void *data)
1832{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001833 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835EXPORT_SYMBOL(read_cache_page);
1836
1837/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 * The logic we want is
1839 *
1840 * if suid or (sgid and xgrp)
1841 * remove privs
1842 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001843int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844{
1845 mode_t mode = dentry->d_inode->i_mode;
1846 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 /* suid always must be killed */
1849 if (unlikely(mode & S_ISUID))
1850 kill = ATTR_KILL_SUID;
1851
1852 /*
1853 * sgid without any exec bits is just a mandatory locking mark; leave
1854 * it alone. If some exec bits are set, it's a real sgid; kill it.
1855 */
1856 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1857 kill |= ATTR_KILL_SGID;
1858
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001859 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001860 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
Jens Axboe01de85e2006-10-17 19:50:36 +02001862 return 0;
1863}
Mark Fashehd23a1472006-10-17 17:05:18 -07001864EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001865
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001866static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001867{
1868 struct iattr newattrs;
1869
1870 newattrs.ia_valid = ATTR_FORCE | kill;
1871 return notify_change(dentry, &newattrs);
1872}
1873
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001874int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001875{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001876 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001877 int killsuid = should_remove_suid(dentry);
1878 int killpriv = security_inode_need_killpriv(dentry);
1879 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001880
Serge E. Hallynb5376772007-10-16 23:31:36 -07001881 if (killpriv < 0)
1882 return killpriv;
1883 if (killpriv)
1884 error = security_inode_killpriv(dentry);
1885 if (!error && killsuid)
1886 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001887
Serge E. Hallynb5376772007-10-16 23:31:36 -07001888 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001890EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
Nick Piggin2f718ff2007-10-16 01:24:59 -07001892static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 const struct iovec *iov, size_t base, size_t bytes)
1894{
Ingo Molnarf1800532009-03-02 11:00:57 +01001895 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896
1897 while (bytes) {
1898 char __user *buf = iov->iov_base + base;
1899 int copy = min(bytes, iov->iov_len - base);
1900
1901 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001902 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 copied += copy;
1904 bytes -= copy;
1905 vaddr += copy;
1906 iov++;
1907
NeilBrown01408c42006-06-25 05:47:58 -07001908 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 }
1911 return copied - left;
1912}
1913
1914/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001915 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001916 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001917 * bytes which were copied.
1918 */
1919size_t iov_iter_copy_from_user_atomic(struct page *page,
1920 struct iov_iter *i, unsigned long offset, size_t bytes)
1921{
1922 char *kaddr;
1923 size_t copied;
1924
1925 BUG_ON(!in_atomic());
1926 kaddr = kmap_atomic(page, KM_USER0);
1927 if (likely(i->nr_segs == 1)) {
1928 int left;
1929 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001930 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001931 copied = bytes - left;
1932 } else {
1933 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1934 i->iov, i->iov_offset, bytes);
1935 }
1936 kunmap_atomic(kaddr, KM_USER0);
1937
1938 return copied;
1939}
Nick Piggin89e10782007-10-16 01:25:07 -07001940EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001941
1942/*
1943 * This has the same sideeffects and return value as
1944 * iov_iter_copy_from_user_atomic().
1945 * The difference is that it attempts to resolve faults.
1946 * Page must not be locked.
1947 */
1948size_t iov_iter_copy_from_user(struct page *page,
1949 struct iov_iter *i, unsigned long offset, size_t bytes)
1950{
1951 char *kaddr;
1952 size_t copied;
1953
1954 kaddr = kmap(page);
1955 if (likely(i->nr_segs == 1)) {
1956 int left;
1957 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001958 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001959 copied = bytes - left;
1960 } else {
1961 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1962 i->iov, i->iov_offset, bytes);
1963 }
1964 kunmap(page);
1965 return copied;
1966}
Nick Piggin89e10782007-10-16 01:25:07 -07001967EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001968
Nick Pigginf7009262008-03-10 11:43:59 -07001969void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001970{
Nick Pigginf7009262008-03-10 11:43:59 -07001971 BUG_ON(i->count < bytes);
1972
Nick Piggin2f718ff2007-10-16 01:24:59 -07001973 if (likely(i->nr_segs == 1)) {
1974 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001975 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001976 } else {
1977 const struct iovec *iov = i->iov;
1978 size_t base = i->iov_offset;
1979
Nick Piggin124d3b72008-02-02 15:01:17 +01001980 /*
1981 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001982 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001983 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001984 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001985 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001986
Nick Pigginf7009262008-03-10 11:43:59 -07001987 copy = min(bytes, iov->iov_len - base);
1988 BUG_ON(!i->count || i->count < copy);
1989 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001990 bytes -= copy;
1991 base += copy;
1992 if (iov->iov_len == base) {
1993 iov++;
1994 base = 0;
1995 }
1996 }
1997 i->iov = iov;
1998 i->iov_offset = base;
1999 }
2000}
Nick Piggin89e10782007-10-16 01:25:07 -07002001EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002002
Nick Pigginafddba42007-10-16 01:25:01 -07002003/*
2004 * Fault in the first iovec of the given iov_iter, to a maximum length
2005 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2006 * accessed (ie. because it is an invalid address).
2007 *
2008 * writev-intensive code may want this to prefault several iovecs -- that
2009 * would be possible (callers must not rely on the fact that _only_ the
2010 * first iovec will be faulted with the current implementation).
2011 */
2012int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002013{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002014 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002015 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2016 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002017}
Nick Piggin89e10782007-10-16 01:25:07 -07002018EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002019
2020/*
2021 * Return the count of just the current iov_iter segment.
2022 */
2023size_t iov_iter_single_seg_count(struct iov_iter *i)
2024{
2025 const struct iovec *iov = i->iov;
2026 if (i->nr_segs == 1)
2027 return i->count;
2028 else
2029 return min(i->count, iov->iov_len - i->iov_offset);
2030}
Nick Piggin89e10782007-10-16 01:25:07 -07002031EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002032
2033/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 * Performs necessary checks before doing a write
2035 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002036 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 * Returns appropriate error code that caller should return or
2038 * zero in case that write should be allowed.
2039 */
2040inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2041{
2042 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002043 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044
2045 if (unlikely(*pos < 0))
2046 return -EINVAL;
2047
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 if (!isblk) {
2049 /* FIXME: this is for backwards compatibility with 2.4 */
2050 if (file->f_flags & O_APPEND)
2051 *pos = i_size_read(inode);
2052
2053 if (limit != RLIM_INFINITY) {
2054 if (*pos >= limit) {
2055 send_sig(SIGXFSZ, current, 0);
2056 return -EFBIG;
2057 }
2058 if (*count > limit - (typeof(limit))*pos) {
2059 *count = limit - (typeof(limit))*pos;
2060 }
2061 }
2062 }
2063
2064 /*
2065 * LFS rule
2066 */
2067 if (unlikely(*pos + *count > MAX_NON_LFS &&
2068 !(file->f_flags & O_LARGEFILE))) {
2069 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 return -EFBIG;
2071 }
2072 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2073 *count = MAX_NON_LFS - (unsigned long)*pos;
2074 }
2075 }
2076
2077 /*
2078 * Are we about to exceed the fs block limit ?
2079 *
2080 * If we have written data it becomes a short write. If we have
2081 * exceeded without writing data we send a signal and return EFBIG.
2082 * Linus frestrict idea will clean these up nicely..
2083 */
2084 if (likely(!isblk)) {
2085 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2086 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 return -EFBIG;
2088 }
2089 /* zero-length writes at ->s_maxbytes are OK */
2090 }
2091
2092 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2093 *count = inode->i_sb->s_maxbytes - *pos;
2094 } else {
David Howells93614012006-09-30 20:45:40 +02002095#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 loff_t isize;
2097 if (bdev_read_only(I_BDEV(inode)))
2098 return -EPERM;
2099 isize = i_size_read(inode);
2100 if (*pos >= isize) {
2101 if (*count || *pos > isize)
2102 return -ENOSPC;
2103 }
2104
2105 if (*pos + *count > isize)
2106 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002107#else
2108 return -EPERM;
2109#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 }
2111 return 0;
2112}
2113EXPORT_SYMBOL(generic_write_checks);
2114
Nick Pigginafddba42007-10-16 01:25:01 -07002115int pagecache_write_begin(struct file *file, struct address_space *mapping,
2116 loff_t pos, unsigned len, unsigned flags,
2117 struct page **pagep, void **fsdata)
2118{
2119 const struct address_space_operations *aops = mapping->a_ops;
2120
Nick Piggin4e02ed42008-10-29 14:00:55 -07002121 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002122 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002123}
2124EXPORT_SYMBOL(pagecache_write_begin);
2125
2126int pagecache_write_end(struct file *file, struct address_space *mapping,
2127 loff_t pos, unsigned len, unsigned copied,
2128 struct page *page, void *fsdata)
2129{
2130 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002131
Nick Piggin4e02ed42008-10-29 14:00:55 -07002132 mark_page_accessed(page);
2133 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002134}
2135EXPORT_SYMBOL(pagecache_write_end);
2136
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137ssize_t
2138generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2139 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2140 size_t count, size_t ocount)
2141{
2142 struct file *file = iocb->ki_filp;
2143 struct address_space *mapping = file->f_mapping;
2144 struct inode *inode = mapping->host;
2145 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002146 size_t write_len;
2147 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
2149 if (count != ocount)
2150 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2151
Christoph Hellwiga969e902008-07-23 21:27:04 -07002152 write_len = iov_length(iov, *nr_segs);
2153 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002154
Nick Piggin48b47c52009-01-06 14:40:22 -08002155 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002156 if (written)
2157 goto out;
2158
2159 /*
2160 * After a write we want buffered reads to be sure to go to disk to get
2161 * the new data. We invalidate clean cached page from the region we're
2162 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002163 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002164 */
2165 if (mapping->nrpages) {
2166 written = invalidate_inode_pages2_range(mapping,
2167 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002168 /*
2169 * If a page can not be invalidated, return 0 to fall back
2170 * to buffered write.
2171 */
2172 if (written) {
2173 if (written == -EBUSY)
2174 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002175 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002176 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002177 }
2178
2179 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2180
2181 /*
2182 * Finally, try again to invalidate clean pages which might have been
2183 * cached by non-direct readahead, or faulted in by get_user_pages()
2184 * if the source of the write was an mmap'ed region of the file
2185 * we're writing. Either one is a pretty crazy thing to do,
2186 * so we don't support it 100%. If this invalidation
2187 * fails, tough, the write still worked...
2188 */
2189 if (mapping->nrpages) {
2190 invalidate_inode_pages2_range(mapping,
2191 pos >> PAGE_CACHE_SHIFT, end);
2192 }
2193
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002195 pos += written;
2196 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2197 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 mark_inode_dirty(inode);
2199 }
Namhyung Kim01166512010-10-26 14:21:58 -07002200 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002202out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 return written;
2204}
2205EXPORT_SYMBOL(generic_file_direct_write);
2206
Nick Piggineb2be182007-10-16 01:24:57 -07002207/*
2208 * Find or create a page at the given pagecache position. Return the locked
2209 * page. This function is specifically for buffered writes.
2210 */
Nick Piggin54566b22009-01-04 12:00:53 -08002211struct page *grab_cache_page_write_begin(struct address_space *mapping,
2212 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002213{
2214 int status;
2215 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002216 gfp_t gfp_notmask = 0;
2217 if (flags & AOP_FLAG_NOFS)
2218 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002219repeat:
2220 page = find_lock_page(mapping, index);
2221 if (likely(page))
2222 return page;
2223
Nick Piggin54566b22009-01-04 12:00:53 -08002224 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002225 if (!page)
2226 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002227 status = add_to_page_cache_lru(page, mapping, index,
2228 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002229 if (unlikely(status)) {
2230 page_cache_release(page);
2231 if (status == -EEXIST)
2232 goto repeat;
2233 return NULL;
2234 }
2235 return page;
2236}
Nick Piggin54566b22009-01-04 12:00:53 -08002237EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002238
Nick Pigginafddba42007-10-16 01:25:01 -07002239static ssize_t generic_perform_write(struct file *file,
2240 struct iov_iter *i, loff_t pos)
2241{
2242 struct address_space *mapping = file->f_mapping;
2243 const struct address_space_operations *a_ops = mapping->a_ops;
2244 long status = 0;
2245 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002246 unsigned int flags = 0;
2247
2248 /*
2249 * Copies from kernel address space cannot fail (NFSD is a big user).
2250 */
2251 if (segment_eq(get_fs(), KERNEL_DS))
2252 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002253
2254 do {
2255 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002256 unsigned long offset; /* Offset into pagecache page */
2257 unsigned long bytes; /* Bytes to write to page */
2258 size_t copied; /* Bytes copied from user */
2259 void *fsdata;
2260
2261 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002262 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2263 iov_iter_count(i));
2264
2265again:
2266
2267 /*
2268 * Bring in the user page that we will copy from _first_.
2269 * Otherwise there's a nasty deadlock on copying from the
2270 * same page as we're writing to, without it being marked
2271 * up-to-date.
2272 *
2273 * Not only is this an optimisation, but it is also required
2274 * to check that the address is actually valid, when atomic
2275 * usercopies are used, below.
2276 */
2277 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2278 status = -EFAULT;
2279 break;
2280 }
2281
Nick Piggin674b8922007-10-16 01:25:03 -07002282 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002283 &page, &fsdata);
2284 if (unlikely(status))
2285 break;
2286
anfei zhou931e80e2010-02-02 13:44:02 -08002287 if (mapping_writably_mapped(mapping))
2288 flush_dcache_page(page);
2289
Nick Pigginafddba42007-10-16 01:25:01 -07002290 pagefault_disable();
2291 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2292 pagefault_enable();
2293 flush_dcache_page(page);
2294
Josef Bacikc8236db2009-07-05 12:08:18 -07002295 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002296 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2297 page, fsdata);
2298 if (unlikely(status < 0))
2299 break;
2300 copied = status;
2301
2302 cond_resched();
2303
Nick Piggin124d3b72008-02-02 15:01:17 +01002304 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002305 if (unlikely(copied == 0)) {
2306 /*
2307 * If we were unable to copy any data at all, we must
2308 * fall back to a single segment length write.
2309 *
2310 * If we didn't fallback here, we could livelock
2311 * because not all segments in the iov can be copied at
2312 * once without a pagefault.
2313 */
2314 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2315 iov_iter_single_seg_count(i));
2316 goto again;
2317 }
Nick Pigginafddba42007-10-16 01:25:01 -07002318 pos += copied;
2319 written += copied;
2320
2321 balance_dirty_pages_ratelimited(mapping);
2322
2323 } while (iov_iter_count(i));
2324
2325 return written ? written : status;
2326}
2327
2328ssize_t
2329generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2330 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2331 size_t count, ssize_t written)
2332{
2333 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002334 ssize_t status;
2335 struct iov_iter i;
2336
2337 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002338 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002339
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002341 written += status;
2342 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 }
2344
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 return written ? written : status;
2346}
2347EXPORT_SYMBOL(generic_file_buffered_write);
2348
Jan Karae4dd9de2009-08-17 18:10:06 +02002349/**
2350 * __generic_file_aio_write - write data to a file
2351 * @iocb: IO state structure (file, offset, etc.)
2352 * @iov: vector with data to write
2353 * @nr_segs: number of segments in the vector
2354 * @ppos: position where to write
2355 *
2356 * This function does all the work needed for actually writing data to a
2357 * file. It does all basic checks, removes SUID from the file, updates
2358 * modification times and calls proper subroutines depending on whether we
2359 * do direct IO or a standard buffered write.
2360 *
2361 * It expects i_mutex to be grabbed unless we work on a block device or similar
2362 * object which does not need locking at all.
2363 *
2364 * This function does *not* take care of syncing data in case of O_SYNC write.
2365 * A caller has to handle it. This is mainly due to the fact that we want to
2366 * avoid syncing under i_mutex.
2367 */
2368ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2369 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370{
2371 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002372 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 size_t ocount; /* original count */
2374 size_t count; /* after file limit checks */
2375 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 loff_t pos;
2377 ssize_t written;
2378 ssize_t err;
2379
2380 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002381 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2382 if (err)
2383 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
2385 count = ocount;
2386 pos = *ppos;
2387
2388 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2389
2390 /* We can write back this queue in page reclaim */
2391 current->backing_dev_info = mapping->backing_dev_info;
2392 written = 0;
2393
2394 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2395 if (err)
2396 goto out;
2397
2398 if (count == 0)
2399 goto out;
2400
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002401 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 if (err)
2403 goto out;
2404
Christoph Hellwig870f4812006-01-09 20:52:01 -08002405 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
2407 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2408 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002409 loff_t endbyte;
2410 ssize_t written_buffered;
2411
2412 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2413 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 if (written < 0 || written == count)
2415 goto out;
2416 /*
2417 * direct-io write to a hole: fall through to buffered I/O
2418 * for completing the rest of the request.
2419 */
2420 pos += written;
2421 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002422 written_buffered = generic_file_buffered_write(iocb, iov,
2423 nr_segs, pos, ppos, count,
2424 written);
2425 /*
2426 * If generic_file_buffered_write() retuned a synchronous error
2427 * then we want to return the number of bytes which were
2428 * direct-written, or the error code if that was zero. Note
2429 * that this differs from normal direct-io semantics, which
2430 * will return -EFOO even if some bytes were written.
2431 */
2432 if (written_buffered < 0) {
2433 err = written_buffered;
2434 goto out;
2435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002437 /*
2438 * We need to ensure that the page cache pages are written to
2439 * disk and invalidated to preserve the expected O_DIRECT
2440 * semantics.
2441 */
2442 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002443 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002444 if (err == 0) {
2445 written = written_buffered;
2446 invalidate_mapping_pages(mapping,
2447 pos >> PAGE_CACHE_SHIFT,
2448 endbyte >> PAGE_CACHE_SHIFT);
2449 } else {
2450 /*
2451 * We don't know how much we wrote, so just return
2452 * the number of bytes which were direct-written
2453 */
2454 }
2455 } else {
2456 written = generic_file_buffered_write(iocb, iov, nr_segs,
2457 pos, ppos, count, written);
2458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459out:
2460 current->backing_dev_info = NULL;
2461 return written ? written : err;
2462}
Jan Karae4dd9de2009-08-17 18:10:06 +02002463EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464
Jan Karae4dd9de2009-08-17 18:10:06 +02002465/**
2466 * generic_file_aio_write - write data to a file
2467 * @iocb: IO state structure
2468 * @iov: vector with data to write
2469 * @nr_segs: number of segments in the vector
2470 * @pos: position in file where to write
2471 *
2472 * This is a wrapper around __generic_file_aio_write() to be used by most
2473 * filesystems. It takes care of syncing the file in case of O_SYNC file
2474 * and acquires i_mutex as needed.
2475 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002476ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2477 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478{
2479 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002480 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
2483 BUG_ON(iocb->ki_pos != pos);
2484
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002485 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002486 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002487 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
Jan Kara148f9482009-08-17 19:52:36 +02002489 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 ssize_t err;
2491
Jan Kara148f9482009-08-17 19:52:36 +02002492 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002493 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 ret = err;
2495 }
2496 return ret;
2497}
2498EXPORT_SYMBOL(generic_file_aio_write);
2499
David Howellscf9a2ae2006-08-29 19:05:54 +01002500/**
2501 * try_to_release_page() - release old fs-specific metadata on a page
2502 *
2503 * @page: the page which the kernel is trying to free
2504 * @gfp_mask: memory allocation flags (and I/O mode)
2505 *
2506 * The address_space is to try to release any data against the page
2507 * (presumably at page->private). If the release was successful, return `1'.
2508 * Otherwise return zero.
2509 *
David Howells266cf652009-04-03 16:42:36 +01002510 * This may also be called if PG_fscache is set on a page, indicating that the
2511 * page is known to the local caching routines.
2512 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002513 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002514 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002515 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002516 */
2517int try_to_release_page(struct page *page, gfp_t gfp_mask)
2518{
2519 struct address_space * const mapping = page->mapping;
2520
2521 BUG_ON(!PageLocked(page));
2522 if (PageWriteback(page))
2523 return 0;
2524
2525 if (mapping && mapping->a_ops->releasepage)
2526 return mapping->a_ops->releasepage(page, gfp_mask);
2527 return try_to_free_buffers(page);
2528}
2529
2530EXPORT_SYMBOL(try_to_release_page);