blob: 3d4df44e4221d25bb041995813823e00903161e5 [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 Karad3bccb6f2009-08-17 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 Howells385e1ca5f2009-04-03 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 Howells385e1ca5f2009-04-03 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
Randy Dunlap485bb992006-06-23 02:03:49 -0700615/**
616 * find_get_page - find and get a page reference
617 * @mapping: the address_space to search
618 * @offset: the page index
619 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700620 * Is there a pagecache struct page at the given (mapping, offset) tuple?
621 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 */
Nick Piggina60637c2008-07-25 19:45:31 -0700623struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
Nick Piggina60637c2008-07-25 19:45:31 -0700625 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 struct page *page;
627
Nick Piggina60637c2008-07-25 19:45:31 -0700628 rcu_read_lock();
629repeat:
630 page = NULL;
631 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
632 if (pagep) {
633 page = radix_tree_deref_slot(pagep);
634 if (unlikely(!page || page == RADIX_TREE_RETRY))
635 goto repeat;
636
637 if (!page_cache_get_speculative(page))
638 goto repeat;
639
640 /*
641 * Has the page moved?
642 * This is part of the lockless pagecache protocol. See
643 * include/linux/pagemap.h for details.
644 */
645 if (unlikely(page != *pagep)) {
646 page_cache_release(page);
647 goto repeat;
648 }
649 }
650 rcu_read_unlock();
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 return page;
653}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654EXPORT_SYMBOL(find_get_page);
655
Randy Dunlap485bb992006-06-23 02:03:49 -0700656/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700658 * @mapping: the address_space to search
659 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 *
661 * Locates the desired pagecache page, locks it, increments its reference
662 * count and returns its address.
663 *
664 * Returns zero if the page was not present. find_lock_page() may sleep.
665 */
Nick Piggina60637c2008-07-25 19:45:31 -0700666struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
668 struct page *page;
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700671 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700673 lock_page(page);
674 /* Has the page been truncated? */
675 if (unlikely(page->mapping != mapping)) {
676 unlock_page(page);
677 page_cache_release(page);
678 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
Nick Piggina60637c2008-07-25 19:45:31 -0700680 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 return page;
683}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684EXPORT_SYMBOL(find_lock_page);
685
686/**
687 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700688 * @mapping: the page's address_space
689 * @index: the page's index into the mapping
690 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 *
692 * Locates a page in the pagecache. If the page is not present, a new page
693 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
694 * LRU list. The returned page is locked and has its reference count
695 * incremented.
696 *
697 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
698 * allocation!
699 *
700 * find_or_create_page() returns the desired page's address, or zero on
701 * memory exhaustion.
702 */
703struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700704 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
Nick Piggineb2be182007-10-16 01:24:57 -0700706 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 int err;
708repeat:
709 page = find_lock_page(mapping, index);
710 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700711 page = __page_cache_alloc(gfp_mask);
712 if (!page)
713 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800714 /*
715 * We want a regular kernel memory (not highmem or DMA etc)
716 * allocation for the radix tree nodes, but we need to honour
717 * the context-specific requirements the caller has asked for.
718 * GFP_RECLAIM_MASK collects those requirements.
719 */
720 err = add_to_page_cache_lru(page, mapping, index,
721 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700722 if (unlikely(err)) {
723 page_cache_release(page);
724 page = NULL;
725 if (err == -EEXIST)
726 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 return page;
730}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731EXPORT_SYMBOL(find_or_create_page);
732
733/**
734 * find_get_pages - gang pagecache lookup
735 * @mapping: The address_space to search
736 * @start: The starting page index
737 * @nr_pages: The maximum number of pages
738 * @pages: Where the resulting pages are placed
739 *
740 * find_get_pages() will search for and return a group of up to
741 * @nr_pages pages in the mapping. The pages are placed at @pages.
742 * find_get_pages() takes a reference against the returned pages.
743 *
744 * The search returns a group of mapping-contiguous pages with ascending
745 * indexes. There may be holes in the indices due to not-present pages.
746 *
747 * find_get_pages() returns the number of pages which were found.
748 */
749unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
750 unsigned int nr_pages, struct page **pages)
751{
752 unsigned int i;
753 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700754 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Nick Piggina60637c2008-07-25 19:45:31 -0700756 rcu_read_lock();
757restart:
758 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
759 (void ***)pages, start, nr_pages);
760 ret = 0;
761 for (i = 0; i < nr_found; i++) {
762 struct page *page;
763repeat:
764 page = radix_tree_deref_slot((void **)pages[i]);
765 if (unlikely(!page))
766 continue;
767 /*
768 * this can only trigger if nr_found == 1, making livelock
769 * a non issue.
770 */
771 if (unlikely(page == RADIX_TREE_RETRY))
772 goto restart;
773
774 if (!page_cache_get_speculative(page))
775 goto repeat;
776
777 /* Has the page moved? */
778 if (unlikely(page != *((void **)pages[i]))) {
779 page_cache_release(page);
780 goto repeat;
781 }
782
783 pages[ret] = page;
784 ret++;
785 }
786 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 return ret;
788}
789
Jens Axboeebf43502006-04-27 08:46:01 +0200790/**
791 * find_get_pages_contig - gang contiguous pagecache lookup
792 * @mapping: The address_space to search
793 * @index: The starting page index
794 * @nr_pages: The maximum number of pages
795 * @pages: Where the resulting pages are placed
796 *
797 * find_get_pages_contig() works exactly like find_get_pages(), except
798 * that the returned number of pages are guaranteed to be contiguous.
799 *
800 * find_get_pages_contig() returns the number of pages which were found.
801 */
802unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
803 unsigned int nr_pages, struct page **pages)
804{
805 unsigned int i;
806 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700807 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200808
Nick Piggina60637c2008-07-25 19:45:31 -0700809 rcu_read_lock();
810restart:
811 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
812 (void ***)pages, index, nr_pages);
813 ret = 0;
814 for (i = 0; i < nr_found; i++) {
815 struct page *page;
816repeat:
817 page = radix_tree_deref_slot((void **)pages[i]);
818 if (unlikely(!page))
819 continue;
820 /*
821 * this can only trigger if nr_found == 1, making livelock
822 * a non issue.
823 */
824 if (unlikely(page == RADIX_TREE_RETRY))
825 goto restart;
826
827 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200828 break;
829
Nick Piggina60637c2008-07-25 19:45:31 -0700830 if (!page_cache_get_speculative(page))
831 goto repeat;
832
833 /* Has the page moved? */
834 if (unlikely(page != *((void **)pages[i]))) {
835 page_cache_release(page);
836 goto repeat;
837 }
838
839 pages[ret] = page;
840 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200841 index++;
842 }
Nick Piggina60637c2008-07-25 19:45:31 -0700843 rcu_read_unlock();
844 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200845}
David Howellsef71c152007-05-09 02:33:44 -0700846EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200847
Randy Dunlap485bb992006-06-23 02:03:49 -0700848/**
849 * find_get_pages_tag - find and return pages that match @tag
850 * @mapping: the address_space to search
851 * @index: the starting page index
852 * @tag: the tag index
853 * @nr_pages: the maximum number of pages
854 * @pages: where the resulting pages are placed
855 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700857 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 */
859unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
860 int tag, unsigned int nr_pages, struct page **pages)
861{
862 unsigned int i;
863 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700864 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
Nick Piggina60637c2008-07-25 19:45:31 -0700866 rcu_read_lock();
867restart:
868 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
869 (void ***)pages, *index, nr_pages, tag);
870 ret = 0;
871 for (i = 0; i < nr_found; i++) {
872 struct page *page;
873repeat:
874 page = radix_tree_deref_slot((void **)pages[i]);
875 if (unlikely(!page))
876 continue;
877 /*
878 * this can only trigger if nr_found == 1, making livelock
879 * a non issue.
880 */
881 if (unlikely(page == RADIX_TREE_RETRY))
882 goto restart;
883
884 if (!page_cache_get_speculative(page))
885 goto repeat;
886
887 /* Has the page moved? */
888 if (unlikely(page != *((void **)pages[i]))) {
889 page_cache_release(page);
890 goto repeat;
891 }
892
893 pages[ret] = page;
894 ret++;
895 }
896 rcu_read_unlock();
897
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 if (ret)
899 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 return ret;
902}
David Howellsef71c152007-05-09 02:33:44 -0700903EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Randy Dunlap485bb992006-06-23 02:03:49 -0700905/**
906 * grab_cache_page_nowait - returns locked page at given index in given cache
907 * @mapping: target address_space
908 * @index: the page index
909 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800910 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 * This is intended for speculative data generators, where the data can
912 * be regenerated if the page couldn't be grabbed. This routine should
913 * be safe to call while holding the lock for another page.
914 *
915 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
916 * and deadlock against the caller's locked page.
917 */
918struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700919grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920{
921 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200924 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 return page;
926 page_cache_release(page);
927 return NULL;
928 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700929 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800930 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 page_cache_release(page);
932 page = NULL;
933 }
934 return page;
935}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936EXPORT_SYMBOL(grab_cache_page_nowait);
937
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700938/*
939 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
940 * a _large_ part of the i/o request. Imagine the worst scenario:
941 *
942 * ---R__________________________________________B__________
943 * ^ reading here ^ bad block(assume 4k)
944 *
945 * read(R) => miss => readahead(R...B) => media error => frustrating retries
946 * => failing the whole request => read(R) => read(R+1) =>
947 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
948 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
949 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
950 *
951 * It is going insane. Fix it by quickly scaling down the readahead size.
952 */
953static void shrink_readahead_size_eio(struct file *filp,
954 struct file_ra_state *ra)
955{
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700956 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700957}
958
Randy Dunlap485bb992006-06-23 02:03:49 -0700959/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800960 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700961 * @filp: the file to read
962 * @ppos: current file position
963 * @desc: read_descriptor
964 * @actor: read method
965 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700967 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 *
969 * This is really ugly. But the goto's actually try to clarify some
970 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800972static void do_generic_file_read(struct file *filp, loff_t *ppos,
973 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800975 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800977 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700978 pgoff_t index;
979 pgoff_t last_index;
980 pgoff_t prev_index;
981 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700982 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700986 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
987 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
989 offset = *ppos & ~PAGE_CACHE_MASK;
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 for (;;) {
992 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700993 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -0700994 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 unsigned long nr, ret;
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998find_page:
999 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001000 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001001 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001002 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001003 index, last_index - index);
1004 page = find_get_page(mapping, index);
1005 if (unlikely(page == NULL))
1006 goto no_cached_page;
1007 }
1008 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001009 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001010 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001011 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001013 if (!PageUptodate(page)) {
1014 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1015 !mapping->a_ops->is_partially_uptodate)
1016 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001017 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001018 goto page_not_up_to_date;
1019 if (!mapping->a_ops->is_partially_uptodate(page,
1020 desc, offset))
1021 goto page_not_up_to_date_locked;
1022 unlock_page(page);
1023 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001025 /*
1026 * i_size must be checked after we know the page is Uptodate.
1027 *
1028 * Checking i_size after the check allows us to calculate
1029 * the correct value for "nr", which means the zero-filled
1030 * part of the page is not copied back to userspace (unless
1031 * another truncate extends the file - this is desired though).
1032 */
1033
1034 isize = i_size_read(inode);
1035 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1036 if (unlikely(!isize || index > end_index)) {
1037 page_cache_release(page);
1038 goto out;
1039 }
1040
1041 /* nr is the maximum number of bytes to copy from this page */
1042 nr = PAGE_CACHE_SIZE;
1043 if (index == end_index) {
1044 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1045 if (nr <= offset) {
1046 page_cache_release(page);
1047 goto out;
1048 }
1049 }
1050 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 /* If users can be writing to this page using arbitrary
1053 * virtual addresses, take care about potential aliasing
1054 * before reading the page on the kernel side.
1055 */
1056 if (mapping_writably_mapped(mapping))
1057 flush_dcache_page(page);
1058
1059 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001060 * When a sequential read accesses a page several times,
1061 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 */
Jan Karaec0f1632007-05-06 14:49:25 -07001063 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 mark_page_accessed(page);
1065 prev_index = index;
1066
1067 /*
1068 * Ok, we have the page, and it's up-to-date, so
1069 * now we can copy it to user space...
1070 *
1071 * The actor routine returns how many bytes were actually used..
1072 * NOTE! This may not be the same as how much of a user buffer
1073 * we filled up (we may be padding etc), so we can only update
1074 * "pos" here (the actor routine has to update the user buffer
1075 * pointers and the remaining count).
1076 */
1077 ret = actor(desc, page, offset, nr);
1078 offset += ret;
1079 index += offset >> PAGE_CACHE_SHIFT;
1080 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001081 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
1083 page_cache_release(page);
1084 if (ret == nr && desc->count)
1085 continue;
1086 goto out;
1087
1088page_not_up_to_date:
1089 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001090 error = lock_page_killable(page);
1091 if (unlikely(error))
1092 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001094page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001095 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (!page->mapping) {
1097 unlock_page(page);
1098 page_cache_release(page);
1099 continue;
1100 }
1101
1102 /* Did somebody else fill it already? */
1103 if (PageUptodate(page)) {
1104 unlock_page(page);
1105 goto page_ok;
1106 }
1107
1108readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001109 /*
1110 * A previous I/O error may have been due to temporary
1111 * failures, eg. multipath errors.
1112 * PG_error will be set again if readpage fails.
1113 */
1114 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /* Start the actual read. The read will unlock the page. */
1116 error = mapping->a_ops->readpage(filp, page);
1117
Zach Brown994fc28c2005-12-15 14:28:17 -08001118 if (unlikely(error)) {
1119 if (error == AOP_TRUNCATED_PAGE) {
1120 page_cache_release(page);
1121 goto find_page;
1122 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001124 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125
1126 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001127 error = lock_page_killable(page);
1128 if (unlikely(error))
1129 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 if (!PageUptodate(page)) {
1131 if (page->mapping == NULL) {
1132 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001133 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 */
1135 unlock_page(page);
1136 page_cache_release(page);
1137 goto find_page;
1138 }
1139 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001140 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001141 error = -EIO;
1142 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 }
1144 unlock_page(page);
1145 }
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 goto page_ok;
1148
1149readpage_error:
1150 /* UHHUH! A synchronous read error occurred. Report it */
1151 desc->error = error;
1152 page_cache_release(page);
1153 goto out;
1154
1155no_cached_page:
1156 /*
1157 * Ok, it wasn't cached, so we need to create a new
1158 * page..
1159 */
Nick Piggineb2be182007-10-16 01:24:57 -07001160 page = page_cache_alloc_cold(mapping);
1161 if (!page) {
1162 desc->error = -ENOMEM;
1163 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 }
Nick Piggineb2be182007-10-16 01:24:57 -07001165 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 index, GFP_KERNEL);
1167 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001168 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 if (error == -EEXIST)
1170 goto find_page;
1171 desc->error = error;
1172 goto out;
1173 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 goto readpage;
1175 }
1176
1177out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001178 ra->prev_pos = prev_index;
1179 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1180 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001182 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001183 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
1186int file_read_actor(read_descriptor_t *desc, struct page *page,
1187 unsigned long offset, unsigned long size)
1188{
1189 char *kaddr;
1190 unsigned long left, count = desc->count;
1191
1192 if (size > count)
1193 size = count;
1194
1195 /*
1196 * Faults on the destination of a read are common, so do it before
1197 * taking the kmap.
1198 */
1199 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1200 kaddr = kmap_atomic(page, KM_USER0);
1201 left = __copy_to_user_inatomic(desc->arg.buf,
1202 kaddr + offset, size);
1203 kunmap_atomic(kaddr, KM_USER0);
1204 if (left == 0)
1205 goto success;
1206 }
1207
1208 /* Do it the slow way */
1209 kaddr = kmap(page);
1210 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1211 kunmap(page);
1212
1213 if (left) {
1214 size -= left;
1215 desc->error = -EFAULT;
1216 }
1217success:
1218 desc->count = count - size;
1219 desc->written += size;
1220 desc->arg.buf += size;
1221 return size;
1222}
1223
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001224/*
1225 * Performs necessary checks before doing a write
1226 * @iov: io vector request
1227 * @nr_segs: number of segments in the iovec
1228 * @count: number of bytes to write
1229 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1230 *
1231 * Adjust number of segments and amount of bytes to write (nr_segs should be
1232 * properly initialized first). Returns appropriate error code that caller
1233 * should return or zero in case that write should be allowed.
1234 */
1235int generic_segment_checks(const struct iovec *iov,
1236 unsigned long *nr_segs, size_t *count, int access_flags)
1237{
1238 unsigned long seg;
1239 size_t cnt = 0;
1240 for (seg = 0; seg < *nr_segs; seg++) {
1241 const struct iovec *iv = &iov[seg];
1242
1243 /*
1244 * If any segment has a negative length, or the cumulative
1245 * length ever wraps negative then return -EINVAL.
1246 */
1247 cnt += iv->iov_len;
1248 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1249 return -EINVAL;
1250 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1251 continue;
1252 if (seg == 0)
1253 return -EFAULT;
1254 *nr_segs = seg;
1255 cnt -= iv->iov_len; /* This segment is no good */
1256 break;
1257 }
1258 *count = cnt;
1259 return 0;
1260}
1261EXPORT_SYMBOL(generic_segment_checks);
1262
Randy Dunlap485bb992006-06-23 02:03:49 -07001263/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001264 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001265 * @iocb: kernel I/O control block
1266 * @iov: io vector request
1267 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001268 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001269 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 * This is the "read()" routine for all filesystems
1271 * that can use the page cache directly.
1272 */
1273ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001274generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1275 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276{
1277 struct file *filp = iocb->ki_filp;
1278 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001279 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001281 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001284 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1285 if (retval)
1286 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1289 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001290 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 struct address_space *mapping;
1292 struct inode *inode;
1293
1294 mapping = filp->f_mapping;
1295 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 if (!count)
1297 goto out; /* skip atime */
1298 size = i_size_read(inode);
1299 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001300 retval = filemap_write_and_wait_range(mapping, pos,
1301 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001302 if (!retval) {
1303 retval = mapping->a_ops->direct_IO(READ, iocb,
1304 iov, pos, nr_segs);
1305 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001306 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001308 count -= retval;
1309 }
1310
1311 /*
1312 * Btrfs can have a short DIO read if we encounter
1313 * compressed extents, so if there was an error, or if
1314 * we've already read everything we wanted to, or if
1315 * there was a short read because we hit EOF, go ahead
1316 * and return. Otherwise fallthrough to buffered io for
1317 * the rest of the read.
1318 */
1319 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001320 file_accessed(filp);
1321 goto out;
1322 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001323 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 }
1325
Josef Bacik66f998f2010-05-23 11:00:54 -04001326 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001327 for (seg = 0; seg < nr_segs; seg++) {
1328 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001329 loff_t offset = 0;
1330
1331 /*
1332 * If we did a short DIO read we need to skip the section of the
1333 * iov that we've already read data into.
1334 */
1335 if (count) {
1336 if (count > iov[seg].iov_len) {
1337 count -= iov[seg].iov_len;
1338 continue;
1339 }
1340 offset = count;
1341 count = 0;
1342 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Hugh Dickins11fa9772008-07-23 21:27:34 -07001344 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001345 desc.arg.buf = iov[seg].iov_base + offset;
1346 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001347 if (desc.count == 0)
1348 continue;
1349 desc.error = 0;
1350 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1351 retval += desc.written;
1352 if (desc.error) {
1353 retval = retval ?: desc.error;
1354 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001356 if (desc.count > 0)
1357 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 }
1359out:
1360 return retval;
1361}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362EXPORT_SYMBOL(generic_file_aio_read);
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364static ssize_t
1365do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001366 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367{
1368 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1369 return -EINVAL;
1370
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001371 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return 0;
1373}
1374
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001375SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376{
1377 ssize_t ret;
1378 struct file *file;
1379
1380 ret = -EBADF;
1381 file = fget(fd);
1382 if (file) {
1383 if (file->f_mode & FMODE_READ) {
1384 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001385 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1386 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 unsigned long len = end - start + 1;
1388 ret = do_readahead(mapping, file, start, len);
1389 }
1390 fput(file);
1391 }
1392 return ret;
1393}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001394#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1395asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1396{
1397 return SYSC_readahead((int) fd, offset, (size_t) count);
1398}
1399SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1400#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001403/**
1404 * page_cache_read - adds requested page to the page cache if not already there
1405 * @file: file to read
1406 * @offset: page index
1407 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 * This adds the requested page to the page cache if it isn't already there,
1409 * and schedules an I/O to read in its contents from disk.
1410 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001411static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412{
1413 struct address_space *mapping = file->f_mapping;
1414 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001415 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Zach Brown994fc28c2005-12-15 14:28:17 -08001417 do {
1418 page = page_cache_alloc_cold(mapping);
1419 if (!page)
1420 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Zach Brown994fc28c2005-12-15 14:28:17 -08001422 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1423 if (ret == 0)
1424 ret = mapping->a_ops->readpage(file, page);
1425 else if (ret == -EEXIST)
1426 ret = 0; /* losing race to add is OK */
1427
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Zach Brown994fc28c2005-12-15 14:28:17 -08001430 } while (ret == AOP_TRUNCATED_PAGE);
1431
1432 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433}
1434
1435#define MMAP_LOTSAMISS (100)
1436
Linus Torvaldsef00e082009-06-16 15:31:25 -07001437/*
1438 * Synchronous readahead happens when we don't even find
1439 * a page in the page cache at all.
1440 */
1441static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1442 struct file_ra_state *ra,
1443 struct file *file,
1444 pgoff_t offset)
1445{
1446 unsigned long ra_pages;
1447 struct address_space *mapping = file->f_mapping;
1448
1449 /* If we don't want any read-ahead, don't bother */
1450 if (VM_RandomReadHint(vma))
1451 return;
1452
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001453 if (VM_SequentialReadHint(vma) ||
1454 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001455 page_cache_sync_readahead(mapping, ra, file, offset,
1456 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001457 return;
1458 }
1459
1460 if (ra->mmap_miss < INT_MAX)
1461 ra->mmap_miss++;
1462
1463 /*
1464 * Do we miss much more than hit in this file? If so,
1465 * stop bothering with read-ahead. It will only hurt.
1466 */
1467 if (ra->mmap_miss > MMAP_LOTSAMISS)
1468 return;
1469
Wu Fengguangd30a1102009-06-16 15:31:30 -07001470 /*
1471 * mmap read-around
1472 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001473 ra_pages = max_sane_readahead(ra->ra_pages);
1474 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001475 ra->start = max_t(long, 0, offset - ra_pages/2);
1476 ra->size = ra_pages;
1477 ra->async_size = 0;
1478 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001479 }
1480}
1481
1482/*
1483 * Asynchronous readahead happens when we find the page and PG_readahead,
1484 * so we want to possibly extend the readahead further..
1485 */
1486static void do_async_mmap_readahead(struct vm_area_struct *vma,
1487 struct file_ra_state *ra,
1488 struct file *file,
1489 struct page *page,
1490 pgoff_t offset)
1491{
1492 struct address_space *mapping = file->f_mapping;
1493
1494 /* If we don't want any read-ahead, don't bother */
1495 if (VM_RandomReadHint(vma))
1496 return;
1497 if (ra->mmap_miss > 0)
1498 ra->mmap_miss--;
1499 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001500 page_cache_async_readahead(mapping, ra, file,
1501 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001502}
1503
Randy Dunlap485bb992006-06-23 02:03:49 -07001504/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001505 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001506 * @vma: vma in which the fault was taken
1507 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001508 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001509 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 * mapped memory region to read in file data during a page fault.
1511 *
1512 * The goto's are kind of ugly, but this streamlines the normal case of having
1513 * it in the page cache, and handles the special cases reasonably without
1514 * having a lot of duplicated code.
1515 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001516int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
1518 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001519 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 struct address_space *mapping = file->f_mapping;
1521 struct file_ra_state *ra = &file->f_ra;
1522 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001523 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001525 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001526 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001529 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001530 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 * Do we have something in the page cache already?
1534 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001535 page = find_get_page(mapping, offset);
1536 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001538 * We found the page, so try async readahead before
1539 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001541 do_async_mmap_readahead(vma, ra, file, page, offset);
1542 lock_page(page);
1543
1544 /* Did it get truncated? */
1545 if (unlikely(page->mapping != mapping)) {
1546 unlock_page(page);
1547 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 goto no_cached_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 }
Linus Torvaldsef00e082009-06-16 15:31:25 -07001550 } else {
1551 /* No page in the page cache at all */
1552 do_sync_mmap_readahead(vma, ra, file, offset);
1553 count_vm_event(PGMAJFAULT);
1554 ret = VM_FAULT_MAJOR;
1555retry_find:
1556 page = find_lock_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (!page)
1558 goto no_cached_page;
1559 }
1560
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001562 * We have a locked page in the page cache, now we need to check
1563 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 */
Nick Piggind00806b2007-07-19 01:46:57 -07001565 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 goto page_not_uptodate;
1567
Linus Torvaldsef00e082009-06-16 15:31:25 -07001568 /*
1569 * Found the page and have a reference on it.
1570 * We must recheck i_size under page lock.
1571 */
Nick Piggind00806b2007-07-19 01:46:57 -07001572 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001573 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001574 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001575 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001576 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001577 }
1578
Linus Torvaldsef00e082009-06-16 15:31:25 -07001579 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001580 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001581 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583no_cached_page:
1584 /*
1585 * We're only likely to ever get here if MADV_RANDOM is in
1586 * effect.
1587 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001588 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589
1590 /*
1591 * The page we want has now been added to the page cache.
1592 * In the unlikely event that someone removed it in the
1593 * meantime, we'll just come back here and read it again.
1594 */
1595 if (error >= 0)
1596 goto retry_find;
1597
1598 /*
1599 * An error return from page_cache_read can result if the
1600 * system is low on memory, or a problem occurs while trying
1601 * to schedule I/O.
1602 */
1603 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001604 return VM_FAULT_OOM;
1605 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
1607page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 /*
1609 * Umm, take care of errors if the page isn't up-to-date.
1610 * Try to re-read it _once_. We do this synchronously,
1611 * because there really aren't any performance issues here
1612 * and we need to check for errors.
1613 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001615 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001616 if (!error) {
1617 wait_on_page_locked(page);
1618 if (!PageUptodate(page))
1619 error = -EIO;
1620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001622
1623 if (!error || error == AOP_TRUNCATED_PAGE)
1624 goto retry_find;
1625
1626 /* Things didn't work out. Return zero to tell the mm layer so. */
1627 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001628 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001629}
1630EXPORT_SYMBOL(filemap_fault);
1631
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001632const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001633 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634};
1635
1636/* This is used for a general mmap of a disk file */
1637
1638int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1639{
1640 struct address_space *mapping = file->f_mapping;
1641
1642 if (!mapping->a_ops->readpage)
1643 return -ENOEXEC;
1644 file_accessed(file);
1645 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001646 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 return 0;
1648}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649
1650/*
1651 * This is for filesystems which do not implement ->writepage.
1652 */
1653int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1654{
1655 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1656 return -EINVAL;
1657 return generic_file_mmap(file, vma);
1658}
1659#else
1660int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1661{
1662 return -ENOSYS;
1663}
1664int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1665{
1666 return -ENOSYS;
1667}
1668#endif /* CONFIG_MMU */
1669
1670EXPORT_SYMBOL(generic_file_mmap);
1671EXPORT_SYMBOL(generic_file_readonly_mmap);
1672
Nick Piggin6fe69002007-05-06 14:49:04 -07001673static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001674 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001676 void *data,
1677 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678{
Nick Piggineb2be182007-10-16 01:24:57 -07001679 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 int err;
1681repeat:
1682 page = find_get_page(mapping, index);
1683 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001684 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001685 if (!page)
1686 return ERR_PTR(-ENOMEM);
1687 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1688 if (unlikely(err)) {
1689 page_cache_release(page);
1690 if (err == -EEXIST)
1691 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 return ERR_PTR(err);
1694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 err = filler(data, page);
1696 if (err < 0) {
1697 page_cache_release(page);
1698 page = ERR_PTR(err);
1699 }
1700 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 return page;
1702}
1703
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001704static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001705 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001707 void *data,
1708 gfp_t gfp)
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710{
1711 struct page *page;
1712 int err;
1713
1714retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001715 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001717 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 if (PageUptodate(page))
1719 goto out;
1720
1721 lock_page(page);
1722 if (!page->mapping) {
1723 unlock_page(page);
1724 page_cache_release(page);
1725 goto retry;
1726 }
1727 if (PageUptodate(page)) {
1728 unlock_page(page);
1729 goto out;
1730 }
1731 err = filler(data, page);
1732 if (err < 0) {
1733 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001734 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
David Howellsc855ff32007-05-09 13:42:20 +01001736out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001737 mark_page_accessed(page);
1738 return page;
1739}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001740
1741/**
1742 * read_cache_page_async - read into page cache, fill it if needed
1743 * @mapping: the page's address_space
1744 * @index: the page index
1745 * @filler: function to perform the read
1746 * @data: destination for read data
1747 *
1748 * Same as read_cache_page, but don't wait for page to become unlocked
1749 * after submitting it to the filler.
1750 *
1751 * Read into the page cache. If a page already exists, and PageUptodate() is
1752 * not set, try to fill the page but don't wait for it to become unlocked.
1753 *
1754 * If the page does not get brought uptodate, return -EIO.
1755 */
1756struct page *read_cache_page_async(struct address_space *mapping,
1757 pgoff_t index,
1758 int (*filler)(void *,struct page*),
1759 void *data)
1760{
1761 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1762}
Nick Piggin6fe69002007-05-06 14:49:04 -07001763EXPORT_SYMBOL(read_cache_page_async);
1764
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001765static struct page *wait_on_page_read(struct page *page)
1766{
1767 if (!IS_ERR(page)) {
1768 wait_on_page_locked(page);
1769 if (!PageUptodate(page)) {
1770 page_cache_release(page);
1771 page = ERR_PTR(-EIO);
1772 }
1773 }
1774 return page;
1775}
1776
1777/**
1778 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1779 * @mapping: the page's address_space
1780 * @index: the page index
1781 * @gfp: the page allocator flags to use if allocating
1782 *
1783 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1784 * any new page allocations done using the specified allocation flags. Note
1785 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1786 * expect to do this atomically or anything like that - but you can pass in
1787 * other page requirements.
1788 *
1789 * If the page does not get brought uptodate, return -EIO.
1790 */
1791struct page *read_cache_page_gfp(struct address_space *mapping,
1792 pgoff_t index,
1793 gfp_t gfp)
1794{
1795 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1796
1797 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1798}
1799EXPORT_SYMBOL(read_cache_page_gfp);
1800
Nick Piggin6fe69002007-05-06 14:49:04 -07001801/**
1802 * read_cache_page - read into page cache, fill it if needed
1803 * @mapping: the page's address_space
1804 * @index: the page index
1805 * @filler: function to perform the read
1806 * @data: destination for read data
1807 *
1808 * Read into the page cache. If a page already exists, and PageUptodate() is
1809 * not set, try to fill the page then wait for it to become unlocked.
1810 *
1811 * If the page does not get brought uptodate, return -EIO.
1812 */
1813struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001814 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001815 int (*filler)(void *,struct page*),
1816 void *data)
1817{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001818 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820EXPORT_SYMBOL(read_cache_page);
1821
1822/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 * The logic we want is
1824 *
1825 * if suid or (sgid and xgrp)
1826 * remove privs
1827 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001828int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829{
1830 mode_t mode = dentry->d_inode->i_mode;
1831 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
1833 /* suid always must be killed */
1834 if (unlikely(mode & S_ISUID))
1835 kill = ATTR_KILL_SUID;
1836
1837 /*
1838 * sgid without any exec bits is just a mandatory locking mark; leave
1839 * it alone. If some exec bits are set, it's a real sgid; kill it.
1840 */
1841 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1842 kill |= ATTR_KILL_SGID;
1843
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001844 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001845 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846
Jens Axboe01de85e2006-10-17 19:50:36 +02001847 return 0;
1848}
Mark Fashehd23a1472006-10-17 17:05:18 -07001849EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001850
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001851static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001852{
1853 struct iattr newattrs;
1854
1855 newattrs.ia_valid = ATTR_FORCE | kill;
1856 return notify_change(dentry, &newattrs);
1857}
1858
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001859int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001860{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001861 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001862 int killsuid = should_remove_suid(dentry);
1863 int killpriv = security_inode_need_killpriv(dentry);
1864 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001865
Serge E. Hallynb5376772007-10-16 23:31:36 -07001866 if (killpriv < 0)
1867 return killpriv;
1868 if (killpriv)
1869 error = security_inode_killpriv(dentry);
1870 if (!error && killsuid)
1871 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001872
Serge E. Hallynb5376772007-10-16 23:31:36 -07001873 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001875EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
Nick Piggin2f718ff2007-10-16 01:24:59 -07001877static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 const struct iovec *iov, size_t base, size_t bytes)
1879{
Ingo Molnarf1800532009-03-02 11:00:57 +01001880 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882 while (bytes) {
1883 char __user *buf = iov->iov_base + base;
1884 int copy = min(bytes, iov->iov_len - base);
1885
1886 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001887 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 copied += copy;
1889 bytes -= copy;
1890 vaddr += copy;
1891 iov++;
1892
NeilBrown01408c42006-06-25 05:47:58 -07001893 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 }
1896 return copied - left;
1897}
1898
1899/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001900 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001901 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001902 * bytes which were copied.
1903 */
1904size_t iov_iter_copy_from_user_atomic(struct page *page,
1905 struct iov_iter *i, unsigned long offset, size_t bytes)
1906{
1907 char *kaddr;
1908 size_t copied;
1909
1910 BUG_ON(!in_atomic());
1911 kaddr = kmap_atomic(page, KM_USER0);
1912 if (likely(i->nr_segs == 1)) {
1913 int left;
1914 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001915 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001916 copied = bytes - left;
1917 } else {
1918 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1919 i->iov, i->iov_offset, bytes);
1920 }
1921 kunmap_atomic(kaddr, KM_USER0);
1922
1923 return copied;
1924}
Nick Piggin89e10782007-10-16 01:25:07 -07001925EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001926
1927/*
1928 * This has the same sideeffects and return value as
1929 * iov_iter_copy_from_user_atomic().
1930 * The difference is that it attempts to resolve faults.
1931 * Page must not be locked.
1932 */
1933size_t iov_iter_copy_from_user(struct page *page,
1934 struct iov_iter *i, unsigned long offset, size_t bytes)
1935{
1936 char *kaddr;
1937 size_t copied;
1938
1939 kaddr = kmap(page);
1940 if (likely(i->nr_segs == 1)) {
1941 int left;
1942 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001943 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001944 copied = bytes - left;
1945 } else {
1946 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1947 i->iov, i->iov_offset, bytes);
1948 }
1949 kunmap(page);
1950 return copied;
1951}
Nick Piggin89e10782007-10-16 01:25:07 -07001952EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001953
Nick Pigginf7009262008-03-10 11:43:59 -07001954void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001955{
Nick Pigginf7009262008-03-10 11:43:59 -07001956 BUG_ON(i->count < bytes);
1957
Nick Piggin2f718ff2007-10-16 01:24:59 -07001958 if (likely(i->nr_segs == 1)) {
1959 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001960 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001961 } else {
1962 const struct iovec *iov = i->iov;
1963 size_t base = i->iov_offset;
1964
Nick Piggin124d3b72008-02-02 15:01:17 +01001965 /*
1966 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001967 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001968 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001969 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001970 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001971
Nick Pigginf7009262008-03-10 11:43:59 -07001972 copy = min(bytes, iov->iov_len - base);
1973 BUG_ON(!i->count || i->count < copy);
1974 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001975 bytes -= copy;
1976 base += copy;
1977 if (iov->iov_len == base) {
1978 iov++;
1979 base = 0;
1980 }
1981 }
1982 i->iov = iov;
1983 i->iov_offset = base;
1984 }
1985}
Nick Piggin89e10782007-10-16 01:25:07 -07001986EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001987
Nick Pigginafddba42007-10-16 01:25:01 -07001988/*
1989 * Fault in the first iovec of the given iov_iter, to a maximum length
1990 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1991 * accessed (ie. because it is an invalid address).
1992 *
1993 * writev-intensive code may want this to prefault several iovecs -- that
1994 * would be possible (callers must not rely on the fact that _only_ the
1995 * first iovec will be faulted with the current implementation).
1996 */
1997int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001998{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001999 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002000 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2001 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002002}
Nick Piggin89e10782007-10-16 01:25:07 -07002003EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002004
2005/*
2006 * Return the count of just the current iov_iter segment.
2007 */
2008size_t iov_iter_single_seg_count(struct iov_iter *i)
2009{
2010 const struct iovec *iov = i->iov;
2011 if (i->nr_segs == 1)
2012 return i->count;
2013 else
2014 return min(i->count, iov->iov_len - i->iov_offset);
2015}
Nick Piggin89e10782007-10-16 01:25:07 -07002016EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002017
2018/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 * Performs necessary checks before doing a write
2020 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002021 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 * Returns appropriate error code that caller should return or
2023 * zero in case that write should be allowed.
2024 */
2025inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2026{
2027 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002028 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
2030 if (unlikely(*pos < 0))
2031 return -EINVAL;
2032
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 if (!isblk) {
2034 /* FIXME: this is for backwards compatibility with 2.4 */
2035 if (file->f_flags & O_APPEND)
2036 *pos = i_size_read(inode);
2037
2038 if (limit != RLIM_INFINITY) {
2039 if (*pos >= limit) {
2040 send_sig(SIGXFSZ, current, 0);
2041 return -EFBIG;
2042 }
2043 if (*count > limit - (typeof(limit))*pos) {
2044 *count = limit - (typeof(limit))*pos;
2045 }
2046 }
2047 }
2048
2049 /*
2050 * LFS rule
2051 */
2052 if (unlikely(*pos + *count > MAX_NON_LFS &&
2053 !(file->f_flags & O_LARGEFILE))) {
2054 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 return -EFBIG;
2056 }
2057 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2058 *count = MAX_NON_LFS - (unsigned long)*pos;
2059 }
2060 }
2061
2062 /*
2063 * Are we about to exceed the fs block limit ?
2064 *
2065 * If we have written data it becomes a short write. If we have
2066 * exceeded without writing data we send a signal and return EFBIG.
2067 * Linus frestrict idea will clean these up nicely..
2068 */
2069 if (likely(!isblk)) {
2070 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2071 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return -EFBIG;
2073 }
2074 /* zero-length writes at ->s_maxbytes are OK */
2075 }
2076
2077 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2078 *count = inode->i_sb->s_maxbytes - *pos;
2079 } else {
David Howells93614012006-09-30 20:45:40 +02002080#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 loff_t isize;
2082 if (bdev_read_only(I_BDEV(inode)))
2083 return -EPERM;
2084 isize = i_size_read(inode);
2085 if (*pos >= isize) {
2086 if (*count || *pos > isize)
2087 return -ENOSPC;
2088 }
2089
2090 if (*pos + *count > isize)
2091 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002092#else
2093 return -EPERM;
2094#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 }
2096 return 0;
2097}
2098EXPORT_SYMBOL(generic_write_checks);
2099
Nick Pigginafddba42007-10-16 01:25:01 -07002100int pagecache_write_begin(struct file *file, struct address_space *mapping,
2101 loff_t pos, unsigned len, unsigned flags,
2102 struct page **pagep, void **fsdata)
2103{
2104 const struct address_space_operations *aops = mapping->a_ops;
2105
Nick Piggin4e02ed42008-10-29 14:00:55 -07002106 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002107 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002108}
2109EXPORT_SYMBOL(pagecache_write_begin);
2110
2111int pagecache_write_end(struct file *file, struct address_space *mapping,
2112 loff_t pos, unsigned len, unsigned copied,
2113 struct page *page, void *fsdata)
2114{
2115 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002116
Nick Piggin4e02ed42008-10-29 14:00:55 -07002117 mark_page_accessed(page);
2118 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002119}
2120EXPORT_SYMBOL(pagecache_write_end);
2121
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122ssize_t
2123generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2124 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2125 size_t count, size_t ocount)
2126{
2127 struct file *file = iocb->ki_filp;
2128 struct address_space *mapping = file->f_mapping;
2129 struct inode *inode = mapping->host;
2130 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002131 size_t write_len;
2132 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
2134 if (count != ocount)
2135 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2136
Christoph Hellwiga969e902008-07-23 21:27:04 -07002137 write_len = iov_length(iov, *nr_segs);
2138 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002139
Nick Piggin48b47c52009-01-06 14:40:22 -08002140 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002141 if (written)
2142 goto out;
2143
2144 /*
2145 * After a write we want buffered reads to be sure to go to disk to get
2146 * the new data. We invalidate clean cached page from the region we're
2147 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002148 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002149 */
2150 if (mapping->nrpages) {
2151 written = invalidate_inode_pages2_range(mapping,
2152 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002153 /*
2154 * If a page can not be invalidated, return 0 to fall back
2155 * to buffered write.
2156 */
2157 if (written) {
2158 if (written == -EBUSY)
2159 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002160 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002161 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002162 }
2163
2164 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2165
2166 /*
2167 * Finally, try again to invalidate clean pages which might have been
2168 * cached by non-direct readahead, or faulted in by get_user_pages()
2169 * if the source of the write was an mmap'ed region of the file
2170 * we're writing. Either one is a pretty crazy thing to do,
2171 * so we don't support it 100%. If this invalidation
2172 * fails, tough, the write still worked...
2173 */
2174 if (mapping->nrpages) {
2175 invalidate_inode_pages2_range(mapping,
2176 pos >> PAGE_CACHE_SHIFT, end);
2177 }
2178
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 if (written > 0) {
2180 loff_t end = pos + written;
2181 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2182 i_size_write(inode, end);
2183 mark_inode_dirty(inode);
2184 }
2185 *ppos = end;
2186 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002187out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 return written;
2189}
2190EXPORT_SYMBOL(generic_file_direct_write);
2191
Nick Piggineb2be182007-10-16 01:24:57 -07002192/*
2193 * Find or create a page at the given pagecache position. Return the locked
2194 * page. This function is specifically for buffered writes.
2195 */
Nick Piggin54566b22009-01-04 12:00:53 -08002196struct page *grab_cache_page_write_begin(struct address_space *mapping,
2197 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002198{
2199 int status;
2200 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002201 gfp_t gfp_notmask = 0;
2202 if (flags & AOP_FLAG_NOFS)
2203 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002204repeat:
2205 page = find_lock_page(mapping, index);
2206 if (likely(page))
2207 return page;
2208
Nick Piggin54566b22009-01-04 12:00:53 -08002209 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002210 if (!page)
2211 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002212 status = add_to_page_cache_lru(page, mapping, index,
2213 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002214 if (unlikely(status)) {
2215 page_cache_release(page);
2216 if (status == -EEXIST)
2217 goto repeat;
2218 return NULL;
2219 }
2220 return page;
2221}
Nick Piggin54566b22009-01-04 12:00:53 -08002222EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002223
Nick Pigginafddba42007-10-16 01:25:01 -07002224static ssize_t generic_perform_write(struct file *file,
2225 struct iov_iter *i, loff_t pos)
2226{
2227 struct address_space *mapping = file->f_mapping;
2228 const struct address_space_operations *a_ops = mapping->a_ops;
2229 long status = 0;
2230 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002231 unsigned int flags = 0;
2232
2233 /*
2234 * Copies from kernel address space cannot fail (NFSD is a big user).
2235 */
2236 if (segment_eq(get_fs(), KERNEL_DS))
2237 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002238
2239 do {
2240 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002241 unsigned long offset; /* Offset into pagecache page */
2242 unsigned long bytes; /* Bytes to write to page */
2243 size_t copied; /* Bytes copied from user */
2244 void *fsdata;
2245
2246 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002247 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2248 iov_iter_count(i));
2249
2250again:
2251
2252 /*
2253 * Bring in the user page that we will copy from _first_.
2254 * Otherwise there's a nasty deadlock on copying from the
2255 * same page as we're writing to, without it being marked
2256 * up-to-date.
2257 *
2258 * Not only is this an optimisation, but it is also required
2259 * to check that the address is actually valid, when atomic
2260 * usercopies are used, below.
2261 */
2262 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2263 status = -EFAULT;
2264 break;
2265 }
2266
Nick Piggin674b8922007-10-16 01:25:03 -07002267 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002268 &page, &fsdata);
2269 if (unlikely(status))
2270 break;
2271
anfei zhou931e80e2010-02-02 13:44:02 -08002272 if (mapping_writably_mapped(mapping))
2273 flush_dcache_page(page);
2274
Nick Pigginafddba42007-10-16 01:25:01 -07002275 pagefault_disable();
2276 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2277 pagefault_enable();
2278 flush_dcache_page(page);
2279
Josef Bacikc8236db2009-07-05 12:08:18 -07002280 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002281 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2282 page, fsdata);
2283 if (unlikely(status < 0))
2284 break;
2285 copied = status;
2286
2287 cond_resched();
2288
Nick Piggin124d3b72008-02-02 15:01:17 +01002289 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002290 if (unlikely(copied == 0)) {
2291 /*
2292 * If we were unable to copy any data at all, we must
2293 * fall back to a single segment length write.
2294 *
2295 * If we didn't fallback here, we could livelock
2296 * because not all segments in the iov can be copied at
2297 * once without a pagefault.
2298 */
2299 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2300 iov_iter_single_seg_count(i));
2301 goto again;
2302 }
Nick Pigginafddba42007-10-16 01:25:01 -07002303 pos += copied;
2304 written += copied;
2305
2306 balance_dirty_pages_ratelimited(mapping);
2307
2308 } while (iov_iter_count(i));
2309
2310 return written ? written : status;
2311}
2312
2313ssize_t
2314generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2315 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2316 size_t count, ssize_t written)
2317{
2318 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002319 ssize_t status;
2320 struct iov_iter i;
2321
2322 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002323 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002324
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002326 written += status;
2327 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 }
2329
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 return written ? written : status;
2331}
2332EXPORT_SYMBOL(generic_file_buffered_write);
2333
Jan Karae4dd9de2009-08-17 18:10:06 +02002334/**
2335 * __generic_file_aio_write - write data to a file
2336 * @iocb: IO state structure (file, offset, etc.)
2337 * @iov: vector with data to write
2338 * @nr_segs: number of segments in the vector
2339 * @ppos: position where to write
2340 *
2341 * This function does all the work needed for actually writing data to a
2342 * file. It does all basic checks, removes SUID from the file, updates
2343 * modification times and calls proper subroutines depending on whether we
2344 * do direct IO or a standard buffered write.
2345 *
2346 * It expects i_mutex to be grabbed unless we work on a block device or similar
2347 * object which does not need locking at all.
2348 *
2349 * This function does *not* take care of syncing data in case of O_SYNC write.
2350 * A caller has to handle it. This is mainly due to the fact that we want to
2351 * avoid syncing under i_mutex.
2352 */
2353ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2354 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355{
2356 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002357 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 size_t ocount; /* original count */
2359 size_t count; /* after file limit checks */
2360 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 loff_t pos;
2362 ssize_t written;
2363 ssize_t err;
2364
2365 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002366 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2367 if (err)
2368 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
2370 count = ocount;
2371 pos = *ppos;
2372
2373 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2374
2375 /* We can write back this queue in page reclaim */
2376 current->backing_dev_info = mapping->backing_dev_info;
2377 written = 0;
2378
2379 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2380 if (err)
2381 goto out;
2382
2383 if (count == 0)
2384 goto out;
2385
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002386 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 if (err)
2388 goto out;
2389
Christoph Hellwig870f4812006-01-09 20:52:01 -08002390 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
2392 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2393 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002394 loff_t endbyte;
2395 ssize_t written_buffered;
2396
2397 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2398 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (written < 0 || written == count)
2400 goto out;
2401 /*
2402 * direct-io write to a hole: fall through to buffered I/O
2403 * for completing the rest of the request.
2404 */
2405 pos += written;
2406 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002407 written_buffered = generic_file_buffered_write(iocb, iov,
2408 nr_segs, pos, ppos, count,
2409 written);
2410 /*
2411 * If generic_file_buffered_write() retuned a synchronous error
2412 * then we want to return the number of bytes which were
2413 * direct-written, or the error code if that was zero. Note
2414 * that this differs from normal direct-io semantics, which
2415 * will return -EFOO even if some bytes were written.
2416 */
2417 if (written_buffered < 0) {
2418 err = written_buffered;
2419 goto out;
2420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002422 /*
2423 * We need to ensure that the page cache pages are written to
2424 * disk and invalidated to preserve the expected O_DIRECT
2425 * semantics.
2426 */
2427 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002428 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002429 if (err == 0) {
2430 written = written_buffered;
2431 invalidate_mapping_pages(mapping,
2432 pos >> PAGE_CACHE_SHIFT,
2433 endbyte >> PAGE_CACHE_SHIFT);
2434 } else {
2435 /*
2436 * We don't know how much we wrote, so just return
2437 * the number of bytes which were direct-written
2438 */
2439 }
2440 } else {
2441 written = generic_file_buffered_write(iocb, iov, nr_segs,
2442 pos, ppos, count, written);
2443 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444out:
2445 current->backing_dev_info = NULL;
2446 return written ? written : err;
2447}
Jan Karae4dd9de2009-08-17 18:10:06 +02002448EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449
Jan Karae4dd9de2009-08-17 18:10:06 +02002450/**
2451 * generic_file_aio_write - write data to a file
2452 * @iocb: IO state structure
2453 * @iov: vector with data to write
2454 * @nr_segs: number of segments in the vector
2455 * @pos: position in file where to write
2456 *
2457 * This is a wrapper around __generic_file_aio_write() to be used by most
2458 * filesystems. It takes care of syncing the file in case of O_SYNC file
2459 * and acquires i_mutex as needed.
2460 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002461ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2462 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463{
2464 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002465 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467
2468 BUG_ON(iocb->ki_pos != pos);
2469
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002470 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002471 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002472 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473
Jan Kara148f9482009-08-17 19:52:36 +02002474 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 ssize_t err;
2476
Jan Kara148f9482009-08-17 19:52:36 +02002477 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002478 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 ret = err;
2480 }
2481 return ret;
2482}
2483EXPORT_SYMBOL(generic_file_aio_write);
2484
David Howellscf9a2ae2006-08-29 19:05:54 +01002485/**
2486 * try_to_release_page() - release old fs-specific metadata on a page
2487 *
2488 * @page: the page which the kernel is trying to free
2489 * @gfp_mask: memory allocation flags (and I/O mode)
2490 *
2491 * The address_space is to try to release any data against the page
2492 * (presumably at page->private). If the release was successful, return `1'.
2493 * Otherwise return zero.
2494 *
David Howells266cf652009-04-03 16:42:36 +01002495 * This may also be called if PG_fscache is set on a page, indicating that the
2496 * page is known to the local caching routines.
2497 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002498 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002499 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002500 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002501 */
2502int try_to_release_page(struct page *page, gfp_t gfp_mask)
2503{
2504 struct address_space * const mapping = page->mapping;
2505
2506 BUG_ON(!PageLocked(page));
2507 if (PageWriteback(page))
2508 return 0;
2509
2510 if (mapping && mapping->a_ops->releasepage)
2511 return mapping->a_ops->releasepage(page, gfp_mask);
2512 return try_to_free_buffers(page);
2513}
2514
2515EXPORT_SYMBOL(try_to_release_page);