blob: 75572b5f23746a4b47a42920980a22e6a5251fe7 [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
Michel Lespinassed065bd82010-10-26 14:21:57 -0700615int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
616 unsigned int flags)
617{
618 if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
619 __lock_page(page);
620 return 1;
621 } else {
622 up_read(&mm->mmap_sem);
623 wait_on_page_locked(page);
624 return 0;
625 }
626}
627
Randy Dunlap485bb992006-06-23 02:03:49 -0700628/**
629 * find_get_page - find and get a page reference
630 * @mapping: the address_space to search
631 * @offset: the page index
632 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700633 * Is there a pagecache struct page at the given (mapping, offset) tuple?
634 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 */
Nick Piggina60637c2008-07-25 19:45:31 -0700636struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
Nick Piggina60637c2008-07-25 19:45:31 -0700638 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 struct page *page;
640
Nick Piggina60637c2008-07-25 19:45:31 -0700641 rcu_read_lock();
642repeat:
643 page = NULL;
644 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
645 if (pagep) {
646 page = radix_tree_deref_slot(pagep);
647 if (unlikely(!page || page == RADIX_TREE_RETRY))
648 goto repeat;
649
650 if (!page_cache_get_speculative(page))
651 goto repeat;
652
653 /*
654 * Has the page moved?
655 * This is part of the lockless pagecache protocol. See
656 * include/linux/pagemap.h for details.
657 */
658 if (unlikely(page != *pagep)) {
659 page_cache_release(page);
660 goto repeat;
661 }
662 }
663 rcu_read_unlock();
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 return page;
666}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667EXPORT_SYMBOL(find_get_page);
668
Randy Dunlap485bb992006-06-23 02:03:49 -0700669/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700671 * @mapping: the address_space to search
672 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 *
674 * Locates the desired pagecache page, locks it, increments its reference
675 * count and returns its address.
676 *
677 * Returns zero if the page was not present. find_lock_page() may sleep.
678 */
Nick Piggina60637c2008-07-25 19:45:31 -0700679struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 struct page *page;
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700684 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700686 lock_page(page);
687 /* Has the page been truncated? */
688 if (unlikely(page->mapping != mapping)) {
689 unlock_page(page);
690 page_cache_release(page);
691 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 }
Nick Piggina60637c2008-07-25 19:45:31 -0700693 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 return page;
696}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697EXPORT_SYMBOL(find_lock_page);
698
699/**
700 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700701 * @mapping: the page's address_space
702 * @index: the page's index into the mapping
703 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 *
705 * Locates a page in the pagecache. If the page is not present, a new page
706 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
707 * LRU list. The returned page is locked and has its reference count
708 * incremented.
709 *
710 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
711 * allocation!
712 *
713 * find_or_create_page() returns the desired page's address, or zero on
714 * memory exhaustion.
715 */
716struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700717 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718{
Nick Piggineb2be182007-10-16 01:24:57 -0700719 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 int err;
721repeat:
722 page = find_lock_page(mapping, index);
723 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700724 page = __page_cache_alloc(gfp_mask);
725 if (!page)
726 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800727 /*
728 * We want a regular kernel memory (not highmem or DMA etc)
729 * allocation for the radix tree nodes, but we need to honour
730 * the context-specific requirements the caller has asked for.
731 * GFP_RECLAIM_MASK collects those requirements.
732 */
733 err = add_to_page_cache_lru(page, mapping, index,
734 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700735 if (unlikely(err)) {
736 page_cache_release(page);
737 page = NULL;
738 if (err == -EEXIST)
739 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 return page;
743}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744EXPORT_SYMBOL(find_or_create_page);
745
746/**
747 * find_get_pages - gang pagecache lookup
748 * @mapping: The address_space to search
749 * @start: The starting page index
750 * @nr_pages: The maximum number of pages
751 * @pages: Where the resulting pages are placed
752 *
753 * find_get_pages() will search for and return a group of up to
754 * @nr_pages pages in the mapping. The pages are placed at @pages.
755 * find_get_pages() takes a reference against the returned pages.
756 *
757 * The search returns a group of mapping-contiguous pages with ascending
758 * indexes. There may be holes in the indices due to not-present pages.
759 *
760 * find_get_pages() returns the number of pages which were found.
761 */
762unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
763 unsigned int nr_pages, struct page **pages)
764{
765 unsigned int i;
766 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700767 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Nick Piggina60637c2008-07-25 19:45:31 -0700769 rcu_read_lock();
770restart:
771 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
772 (void ***)pages, start, nr_pages);
773 ret = 0;
774 for (i = 0; i < nr_found; i++) {
775 struct page *page;
776repeat:
777 page = radix_tree_deref_slot((void **)pages[i]);
778 if (unlikely(!page))
779 continue;
780 /*
781 * this can only trigger if nr_found == 1, making livelock
782 * a non issue.
783 */
784 if (unlikely(page == RADIX_TREE_RETRY))
785 goto restart;
786
787 if (!page_cache_get_speculative(page))
788 goto repeat;
789
790 /* Has the page moved? */
791 if (unlikely(page != *((void **)pages[i]))) {
792 page_cache_release(page);
793 goto repeat;
794 }
795
796 pages[ret] = page;
797 ret++;
798 }
799 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 return ret;
801}
802
Jens Axboeebf43502006-04-27 08:46:01 +0200803/**
804 * find_get_pages_contig - gang contiguous pagecache lookup
805 * @mapping: The address_space to search
806 * @index: The starting page index
807 * @nr_pages: The maximum number of pages
808 * @pages: Where the resulting pages are placed
809 *
810 * find_get_pages_contig() works exactly like find_get_pages(), except
811 * that the returned number of pages are guaranteed to be contiguous.
812 *
813 * find_get_pages_contig() returns the number of pages which were found.
814 */
815unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
816 unsigned int nr_pages, struct page **pages)
817{
818 unsigned int i;
819 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700820 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200821
Nick Piggina60637c2008-07-25 19:45:31 -0700822 rcu_read_lock();
823restart:
824 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
825 (void ***)pages, index, nr_pages);
826 ret = 0;
827 for (i = 0; i < nr_found; i++) {
828 struct page *page;
829repeat:
830 page = radix_tree_deref_slot((void **)pages[i]);
831 if (unlikely(!page))
832 continue;
833 /*
834 * this can only trigger if nr_found == 1, making livelock
835 * a non issue.
836 */
837 if (unlikely(page == RADIX_TREE_RETRY))
838 goto restart;
839
840 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200841 break;
842
Nick Piggina60637c2008-07-25 19:45:31 -0700843 if (!page_cache_get_speculative(page))
844 goto repeat;
845
846 /* Has the page moved? */
847 if (unlikely(page != *((void **)pages[i]))) {
848 page_cache_release(page);
849 goto repeat;
850 }
851
852 pages[ret] = page;
853 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200854 index++;
855 }
Nick Piggina60637c2008-07-25 19:45:31 -0700856 rcu_read_unlock();
857 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200858}
David Howellsef71c152007-05-09 02:33:44 -0700859EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200860
Randy Dunlap485bb992006-06-23 02:03:49 -0700861/**
862 * find_get_pages_tag - find and return pages that match @tag
863 * @mapping: the address_space to search
864 * @index: the starting page index
865 * @tag: the tag index
866 * @nr_pages: the maximum number of pages
867 * @pages: where the resulting pages are placed
868 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700870 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 */
872unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
873 int tag, unsigned int nr_pages, struct page **pages)
874{
875 unsigned int i;
876 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700877 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
Nick Piggina60637c2008-07-25 19:45:31 -0700879 rcu_read_lock();
880restart:
881 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
882 (void ***)pages, *index, nr_pages, tag);
883 ret = 0;
884 for (i = 0; i < nr_found; i++) {
885 struct page *page;
886repeat:
887 page = radix_tree_deref_slot((void **)pages[i]);
888 if (unlikely(!page))
889 continue;
890 /*
891 * this can only trigger if nr_found == 1, making livelock
892 * a non issue.
893 */
894 if (unlikely(page == RADIX_TREE_RETRY))
895 goto restart;
896
897 if (!page_cache_get_speculative(page))
898 goto repeat;
899
900 /* Has the page moved? */
901 if (unlikely(page != *((void **)pages[i]))) {
902 page_cache_release(page);
903 goto repeat;
904 }
905
906 pages[ret] = page;
907 ret++;
908 }
909 rcu_read_unlock();
910
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (ret)
912 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700913
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 return ret;
915}
David Howellsef71c152007-05-09 02:33:44 -0700916EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Randy Dunlap485bb992006-06-23 02:03:49 -0700918/**
919 * grab_cache_page_nowait - returns locked page at given index in given cache
920 * @mapping: target address_space
921 * @index: the page index
922 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800923 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 * This is intended for speculative data generators, where the data can
925 * be regenerated if the page couldn't be grabbed. This routine should
926 * be safe to call while holding the lock for another page.
927 *
928 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
929 * and deadlock against the caller's locked page.
930 */
931struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700932grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933{
934 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200937 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 return page;
939 page_cache_release(page);
940 return NULL;
941 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700942 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800943 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 page_cache_release(page);
945 page = NULL;
946 }
947 return page;
948}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949EXPORT_SYMBOL(grab_cache_page_nowait);
950
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700951/*
952 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
953 * a _large_ part of the i/o request. Imagine the worst scenario:
954 *
955 * ---R__________________________________________B__________
956 * ^ reading here ^ bad block(assume 4k)
957 *
958 * read(R) => miss => readahead(R...B) => media error => frustrating retries
959 * => failing the whole request => read(R) => read(R+1) =>
960 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
961 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
962 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
963 *
964 * It is going insane. Fix it by quickly scaling down the readahead size.
965 */
966static void shrink_readahead_size_eio(struct file *filp,
967 struct file_ra_state *ra)
968{
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700969 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700970}
971
Randy Dunlap485bb992006-06-23 02:03:49 -0700972/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800973 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700974 * @filp: the file to read
975 * @ppos: current file position
976 * @desc: read_descriptor
977 * @actor: read method
978 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700980 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 *
982 * This is really ugly. But the goto's actually try to clarify some
983 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800985static void do_generic_file_read(struct file *filp, loff_t *ppos,
986 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800988 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800990 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700991 pgoff_t index;
992 pgoff_t last_index;
993 pgoff_t prev_index;
994 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700995 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700999 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1000 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1002 offset = *ppos & ~PAGE_CACHE_MASK;
1003
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 for (;;) {
1005 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001006 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001007 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 unsigned long nr, ret;
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011find_page:
1012 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001013 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001014 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001015 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001016 index, last_index - index);
1017 page = find_get_page(mapping, index);
1018 if (unlikely(page == NULL))
1019 goto no_cached_page;
1020 }
1021 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001022 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001023 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001024 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001026 if (!PageUptodate(page)) {
1027 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1028 !mapping->a_ops->is_partially_uptodate)
1029 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001030 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001031 goto page_not_up_to_date;
1032 if (!mapping->a_ops->is_partially_uptodate(page,
1033 desc, offset))
1034 goto page_not_up_to_date_locked;
1035 unlock_page(page);
1036 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001038 /*
1039 * i_size must be checked after we know the page is Uptodate.
1040 *
1041 * Checking i_size after the check allows us to calculate
1042 * the correct value for "nr", which means the zero-filled
1043 * part of the page is not copied back to userspace (unless
1044 * another truncate extends the file - this is desired though).
1045 */
1046
1047 isize = i_size_read(inode);
1048 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1049 if (unlikely(!isize || index > end_index)) {
1050 page_cache_release(page);
1051 goto out;
1052 }
1053
1054 /* nr is the maximum number of bytes to copy from this page */
1055 nr = PAGE_CACHE_SIZE;
1056 if (index == end_index) {
1057 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1058 if (nr <= offset) {
1059 page_cache_release(page);
1060 goto out;
1061 }
1062 }
1063 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 /* If users can be writing to this page using arbitrary
1066 * virtual addresses, take care about potential aliasing
1067 * before reading the page on the kernel side.
1068 */
1069 if (mapping_writably_mapped(mapping))
1070 flush_dcache_page(page);
1071
1072 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001073 * When a sequential read accesses a page several times,
1074 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 */
Jan Karaec0f1632007-05-06 14:49:25 -07001076 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 mark_page_accessed(page);
1078 prev_index = index;
1079
1080 /*
1081 * Ok, we have the page, and it's up-to-date, so
1082 * now we can copy it to user space...
1083 *
1084 * The actor routine returns how many bytes were actually used..
1085 * NOTE! This may not be the same as how much of a user buffer
1086 * we filled up (we may be padding etc), so we can only update
1087 * "pos" here (the actor routine has to update the user buffer
1088 * pointers and the remaining count).
1089 */
1090 ret = actor(desc, page, offset, nr);
1091 offset += ret;
1092 index += offset >> PAGE_CACHE_SHIFT;
1093 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001094 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 page_cache_release(page);
1097 if (ret == nr && desc->count)
1098 continue;
1099 goto out;
1100
1101page_not_up_to_date:
1102 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001103 error = lock_page_killable(page);
1104 if (unlikely(error))
1105 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001107page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001108 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 if (!page->mapping) {
1110 unlock_page(page);
1111 page_cache_release(page);
1112 continue;
1113 }
1114
1115 /* Did somebody else fill it already? */
1116 if (PageUptodate(page)) {
1117 unlock_page(page);
1118 goto page_ok;
1119 }
1120
1121readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001122 /*
1123 * A previous I/O error may have been due to temporary
1124 * failures, eg. multipath errors.
1125 * PG_error will be set again if readpage fails.
1126 */
1127 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 /* Start the actual read. The read will unlock the page. */
1129 error = mapping->a_ops->readpage(filp, page);
1130
Zach Brown994fc28c2005-12-15 14:28:17 -08001131 if (unlikely(error)) {
1132 if (error == AOP_TRUNCATED_PAGE) {
1133 page_cache_release(page);
1134 goto find_page;
1135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001140 error = lock_page_killable(page);
1141 if (unlikely(error))
1142 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 if (!PageUptodate(page)) {
1144 if (page->mapping == NULL) {
1145 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001146 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 */
1148 unlock_page(page);
1149 page_cache_release(page);
1150 goto find_page;
1151 }
1152 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001153 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001154 error = -EIO;
1155 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
1157 unlock_page(page);
1158 }
1159
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 goto page_ok;
1161
1162readpage_error:
1163 /* UHHUH! A synchronous read error occurred. Report it */
1164 desc->error = error;
1165 page_cache_release(page);
1166 goto out;
1167
1168no_cached_page:
1169 /*
1170 * Ok, it wasn't cached, so we need to create a new
1171 * page..
1172 */
Nick Piggineb2be182007-10-16 01:24:57 -07001173 page = page_cache_alloc_cold(mapping);
1174 if (!page) {
1175 desc->error = -ENOMEM;
1176 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
Nick Piggineb2be182007-10-16 01:24:57 -07001178 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 index, GFP_KERNEL);
1180 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001181 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (error == -EEXIST)
1183 goto find_page;
1184 desc->error = error;
1185 goto out;
1186 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 goto readpage;
1188 }
1189
1190out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001191 ra->prev_pos = prev_index;
1192 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1193 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001195 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001196 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199int file_read_actor(read_descriptor_t *desc, struct page *page,
1200 unsigned long offset, unsigned long size)
1201{
1202 char *kaddr;
1203 unsigned long left, count = desc->count;
1204
1205 if (size > count)
1206 size = count;
1207
1208 /*
1209 * Faults on the destination of a read are common, so do it before
1210 * taking the kmap.
1211 */
1212 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1213 kaddr = kmap_atomic(page, KM_USER0);
1214 left = __copy_to_user_inatomic(desc->arg.buf,
1215 kaddr + offset, size);
1216 kunmap_atomic(kaddr, KM_USER0);
1217 if (left == 0)
1218 goto success;
1219 }
1220
1221 /* Do it the slow way */
1222 kaddr = kmap(page);
1223 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1224 kunmap(page);
1225
1226 if (left) {
1227 size -= left;
1228 desc->error = -EFAULT;
1229 }
1230success:
1231 desc->count = count - size;
1232 desc->written += size;
1233 desc->arg.buf += size;
1234 return size;
1235}
1236
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001237/*
1238 * Performs necessary checks before doing a write
1239 * @iov: io vector request
1240 * @nr_segs: number of segments in the iovec
1241 * @count: number of bytes to write
1242 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1243 *
1244 * Adjust number of segments and amount of bytes to write (nr_segs should be
1245 * properly initialized first). Returns appropriate error code that caller
1246 * should return or zero in case that write should be allowed.
1247 */
1248int generic_segment_checks(const struct iovec *iov,
1249 unsigned long *nr_segs, size_t *count, int access_flags)
1250{
1251 unsigned long seg;
1252 size_t cnt = 0;
1253 for (seg = 0; seg < *nr_segs; seg++) {
1254 const struct iovec *iv = &iov[seg];
1255
1256 /*
1257 * If any segment has a negative length, or the cumulative
1258 * length ever wraps negative then return -EINVAL.
1259 */
1260 cnt += iv->iov_len;
1261 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1262 return -EINVAL;
1263 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1264 continue;
1265 if (seg == 0)
1266 return -EFAULT;
1267 *nr_segs = seg;
1268 cnt -= iv->iov_len; /* This segment is no good */
1269 break;
1270 }
1271 *count = cnt;
1272 return 0;
1273}
1274EXPORT_SYMBOL(generic_segment_checks);
1275
Randy Dunlap485bb992006-06-23 02:03:49 -07001276/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001277 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001278 * @iocb: kernel I/O control block
1279 * @iov: io vector request
1280 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001281 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001282 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 * This is the "read()" routine for all filesystems
1284 * that can use the page cache directly.
1285 */
1286ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001287generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1288 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289{
1290 struct file *filp = iocb->ki_filp;
1291 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001292 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001294 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001297 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1298 if (retval)
1299 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
1301 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1302 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001303 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 struct address_space *mapping;
1305 struct inode *inode;
1306
1307 mapping = filp->f_mapping;
1308 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (!count)
1310 goto out; /* skip atime */
1311 size = i_size_read(inode);
1312 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001313 retval = filemap_write_and_wait_range(mapping, pos,
1314 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001315 if (!retval) {
1316 retval = mapping->a_ops->direct_IO(READ, iocb,
1317 iov, pos, nr_segs);
1318 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001319 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001321 count -= retval;
1322 }
1323
1324 /*
1325 * Btrfs can have a short DIO read if we encounter
1326 * compressed extents, so if there was an error, or if
1327 * we've already read everything we wanted to, or if
1328 * there was a short read because we hit EOF, go ahead
1329 * and return. Otherwise fallthrough to buffered io for
1330 * the rest of the read.
1331 */
1332 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001333 file_accessed(filp);
1334 goto out;
1335 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001336 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
1338
Josef Bacik66f998f2010-05-23 11:00:54 -04001339 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001340 for (seg = 0; seg < nr_segs; seg++) {
1341 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001342 loff_t offset = 0;
1343
1344 /*
1345 * If we did a short DIO read we need to skip the section of the
1346 * iov that we've already read data into.
1347 */
1348 if (count) {
1349 if (count > iov[seg].iov_len) {
1350 count -= iov[seg].iov_len;
1351 continue;
1352 }
1353 offset = count;
1354 count = 0;
1355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Hugh Dickins11fa9772008-07-23 21:27:34 -07001357 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001358 desc.arg.buf = iov[seg].iov_base + offset;
1359 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001360 if (desc.count == 0)
1361 continue;
1362 desc.error = 0;
1363 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1364 retval += desc.written;
1365 if (desc.error) {
1366 retval = retval ?: desc.error;
1367 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001369 if (desc.count > 0)
1370 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372out:
1373 return retval;
1374}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375EXPORT_SYMBOL(generic_file_aio_read);
1376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377static ssize_t
1378do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001379 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380{
1381 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1382 return -EINVAL;
1383
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001384 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 return 0;
1386}
1387
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001388SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
1390 ssize_t ret;
1391 struct file *file;
1392
1393 ret = -EBADF;
1394 file = fget(fd);
1395 if (file) {
1396 if (file->f_mode & FMODE_READ) {
1397 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001398 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1399 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 unsigned long len = end - start + 1;
1401 ret = do_readahead(mapping, file, start, len);
1402 }
1403 fput(file);
1404 }
1405 return ret;
1406}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001407#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1408asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1409{
1410 return SYSC_readahead((int) fd, offset, (size_t) count);
1411}
1412SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1413#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001416/**
1417 * page_cache_read - adds requested page to the page cache if not already there
1418 * @file: file to read
1419 * @offset: page index
1420 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 * This adds the requested page to the page cache if it isn't already there,
1422 * and schedules an I/O to read in its contents from disk.
1423 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001424static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 struct address_space *mapping = file->f_mapping;
1427 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001428 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Zach Brown994fc28c2005-12-15 14:28:17 -08001430 do {
1431 page = page_cache_alloc_cold(mapping);
1432 if (!page)
1433 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
Zach Brown994fc28c2005-12-15 14:28:17 -08001435 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1436 if (ret == 0)
1437 ret = mapping->a_ops->readpage(file, page);
1438 else if (ret == -EEXIST)
1439 ret = 0; /* losing race to add is OK */
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Zach Brown994fc28c2005-12-15 14:28:17 -08001443 } while (ret == AOP_TRUNCATED_PAGE);
1444
1445 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446}
1447
1448#define MMAP_LOTSAMISS (100)
1449
Linus Torvaldsef00e082009-06-16 15:31:25 -07001450/*
1451 * Synchronous readahead happens when we don't even find
1452 * a page in the page cache at all.
1453 */
1454static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1455 struct file_ra_state *ra,
1456 struct file *file,
1457 pgoff_t offset)
1458{
1459 unsigned long ra_pages;
1460 struct address_space *mapping = file->f_mapping;
1461
1462 /* If we don't want any read-ahead, don't bother */
1463 if (VM_RandomReadHint(vma))
1464 return;
1465
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001466 if (VM_SequentialReadHint(vma) ||
1467 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001468 page_cache_sync_readahead(mapping, ra, file, offset,
1469 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001470 return;
1471 }
1472
1473 if (ra->mmap_miss < INT_MAX)
1474 ra->mmap_miss++;
1475
1476 /*
1477 * Do we miss much more than hit in this file? If so,
1478 * stop bothering with read-ahead. It will only hurt.
1479 */
1480 if (ra->mmap_miss > MMAP_LOTSAMISS)
1481 return;
1482
Wu Fengguangd30a1102009-06-16 15:31:30 -07001483 /*
1484 * mmap read-around
1485 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001486 ra_pages = max_sane_readahead(ra->ra_pages);
1487 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001488 ra->start = max_t(long, 0, offset - ra_pages/2);
1489 ra->size = ra_pages;
1490 ra->async_size = 0;
1491 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001492 }
1493}
1494
1495/*
1496 * Asynchronous readahead happens when we find the page and PG_readahead,
1497 * so we want to possibly extend the readahead further..
1498 */
1499static void do_async_mmap_readahead(struct vm_area_struct *vma,
1500 struct file_ra_state *ra,
1501 struct file *file,
1502 struct page *page,
1503 pgoff_t offset)
1504{
1505 struct address_space *mapping = file->f_mapping;
1506
1507 /* If we don't want any read-ahead, don't bother */
1508 if (VM_RandomReadHint(vma))
1509 return;
1510 if (ra->mmap_miss > 0)
1511 ra->mmap_miss--;
1512 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001513 page_cache_async_readahead(mapping, ra, file,
1514 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001515}
1516
Randy Dunlap485bb992006-06-23 02:03:49 -07001517/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001518 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001519 * @vma: vma in which the fault was taken
1520 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001521 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001522 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 * mapped memory region to read in file data during a page fault.
1524 *
1525 * The goto's are kind of ugly, but this streamlines the normal case of having
1526 * it in the page cache, and handles the special cases reasonably without
1527 * having a lot of duplicated code.
1528 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001529int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
1531 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001532 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 struct address_space *mapping = file->f_mapping;
1534 struct file_ra_state *ra = &file->f_ra;
1535 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001536 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001538 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001539 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001542 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001543 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 * Do we have something in the page cache already?
1547 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001548 page = find_get_page(mapping, offset);
1549 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001551 * We found the page, so try async readahead before
1552 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001554 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001555 } else {
1556 /* No page in the page cache at all */
1557 do_sync_mmap_readahead(vma, ra, file, offset);
1558 count_vm_event(PGMAJFAULT);
1559 ret = VM_FAULT_MAJOR;
1560retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001561 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (!page)
1563 goto no_cached_page;
1564 }
1565
Michel Lespinassed065bd82010-10-26 14:21:57 -07001566 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags))
1567 return ret | VM_FAULT_RETRY;
Michel Lespinasseb522c942010-10-26 14:21:56 -07001568
1569 /* Did it get truncated? */
1570 if (unlikely(page->mapping != mapping)) {
1571 unlock_page(page);
1572 put_page(page);
1573 goto retry_find;
1574 }
1575 VM_BUG_ON(page->index != offset);
1576
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001578 * We have a locked page in the page cache, now we need to check
1579 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 */
Nick Piggind00806b2007-07-19 01:46:57 -07001581 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 goto page_not_uptodate;
1583
Linus Torvaldsef00e082009-06-16 15:31:25 -07001584 /*
1585 * Found the page and have a reference on it.
1586 * We must recheck i_size under page lock.
1587 */
Nick Piggind00806b2007-07-19 01:46:57 -07001588 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001589 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001590 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001591 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001592 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001593 }
1594
Linus Torvaldsef00e082009-06-16 15:31:25 -07001595 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001596 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001597 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599no_cached_page:
1600 /*
1601 * We're only likely to ever get here if MADV_RANDOM is in
1602 * effect.
1603 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001604 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
1606 /*
1607 * The page we want has now been added to the page cache.
1608 * In the unlikely event that someone removed it in the
1609 * meantime, we'll just come back here and read it again.
1610 */
1611 if (error >= 0)
1612 goto retry_find;
1613
1614 /*
1615 * An error return from page_cache_read can result if the
1616 * system is low on memory, or a problem occurs while trying
1617 * to schedule I/O.
1618 */
1619 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001620 return VM_FAULT_OOM;
1621 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
1623page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 /*
1625 * Umm, take care of errors if the page isn't up-to-date.
1626 * Try to re-read it _once_. We do this synchronously,
1627 * because there really aren't any performance issues here
1628 * and we need to check for errors.
1629 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001631 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001632 if (!error) {
1633 wait_on_page_locked(page);
1634 if (!PageUptodate(page))
1635 error = -EIO;
1636 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001638
1639 if (!error || error == AOP_TRUNCATED_PAGE)
1640 goto retry_find;
1641
1642 /* Things didn't work out. Return zero to tell the mm layer so. */
1643 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001644 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001645}
1646EXPORT_SYMBOL(filemap_fault);
1647
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001648const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001649 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650};
1651
1652/* This is used for a general mmap of a disk file */
1653
1654int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1655{
1656 struct address_space *mapping = file->f_mapping;
1657
1658 if (!mapping->a_ops->readpage)
1659 return -ENOEXEC;
1660 file_accessed(file);
1661 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001662 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 return 0;
1664}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666/*
1667 * This is for filesystems which do not implement ->writepage.
1668 */
1669int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1670{
1671 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1672 return -EINVAL;
1673 return generic_file_mmap(file, vma);
1674}
1675#else
1676int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1677{
1678 return -ENOSYS;
1679}
1680int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1681{
1682 return -ENOSYS;
1683}
1684#endif /* CONFIG_MMU */
1685
1686EXPORT_SYMBOL(generic_file_mmap);
1687EXPORT_SYMBOL(generic_file_readonly_mmap);
1688
Nick Piggin6fe69002007-05-06 14:49:04 -07001689static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001690 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001692 void *data,
1693 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
Nick Piggineb2be182007-10-16 01:24:57 -07001695 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 int err;
1697repeat:
1698 page = find_get_page(mapping, index);
1699 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001700 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001701 if (!page)
1702 return ERR_PTR(-ENOMEM);
1703 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1704 if (unlikely(err)) {
1705 page_cache_release(page);
1706 if (err == -EEXIST)
1707 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 return ERR_PTR(err);
1710 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 err = filler(data, page);
1712 if (err < 0) {
1713 page_cache_release(page);
1714 page = ERR_PTR(err);
1715 }
1716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 return page;
1718}
1719
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001720static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001721 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001723 void *data,
1724 gfp_t gfp)
1725
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726{
1727 struct page *page;
1728 int err;
1729
1730retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001731 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001733 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 if (PageUptodate(page))
1735 goto out;
1736
1737 lock_page(page);
1738 if (!page->mapping) {
1739 unlock_page(page);
1740 page_cache_release(page);
1741 goto retry;
1742 }
1743 if (PageUptodate(page)) {
1744 unlock_page(page);
1745 goto out;
1746 }
1747 err = filler(data, page);
1748 if (err < 0) {
1749 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001750 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
David Howellsc855ff32007-05-09 13:42:20 +01001752out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001753 mark_page_accessed(page);
1754 return page;
1755}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001756
1757/**
1758 * read_cache_page_async - read into page cache, fill it if needed
1759 * @mapping: the page's address_space
1760 * @index: the page index
1761 * @filler: function to perform the read
1762 * @data: destination for read data
1763 *
1764 * Same as read_cache_page, but don't wait for page to become unlocked
1765 * after submitting it to the filler.
1766 *
1767 * Read into the page cache. If a page already exists, and PageUptodate() is
1768 * not set, try to fill the page but don't wait for it to become unlocked.
1769 *
1770 * If the page does not get brought uptodate, return -EIO.
1771 */
1772struct page *read_cache_page_async(struct address_space *mapping,
1773 pgoff_t index,
1774 int (*filler)(void *,struct page*),
1775 void *data)
1776{
1777 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1778}
Nick Piggin6fe69002007-05-06 14:49:04 -07001779EXPORT_SYMBOL(read_cache_page_async);
1780
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001781static struct page *wait_on_page_read(struct page *page)
1782{
1783 if (!IS_ERR(page)) {
1784 wait_on_page_locked(page);
1785 if (!PageUptodate(page)) {
1786 page_cache_release(page);
1787 page = ERR_PTR(-EIO);
1788 }
1789 }
1790 return page;
1791}
1792
1793/**
1794 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1795 * @mapping: the page's address_space
1796 * @index: the page index
1797 * @gfp: the page allocator flags to use if allocating
1798 *
1799 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1800 * any new page allocations done using the specified allocation flags. Note
1801 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1802 * expect to do this atomically or anything like that - but you can pass in
1803 * other page requirements.
1804 *
1805 * If the page does not get brought uptodate, return -EIO.
1806 */
1807struct page *read_cache_page_gfp(struct address_space *mapping,
1808 pgoff_t index,
1809 gfp_t gfp)
1810{
1811 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1812
1813 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1814}
1815EXPORT_SYMBOL(read_cache_page_gfp);
1816
Nick Piggin6fe69002007-05-06 14:49:04 -07001817/**
1818 * read_cache_page - read into page cache, fill it if needed
1819 * @mapping: the page's address_space
1820 * @index: the page index
1821 * @filler: function to perform the read
1822 * @data: destination for read data
1823 *
1824 * Read into the page cache. If a page already exists, and PageUptodate() is
1825 * not set, try to fill the page then wait for it to become unlocked.
1826 *
1827 * If the page does not get brought uptodate, return -EIO.
1828 */
1829struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001830 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001831 int (*filler)(void *,struct page*),
1832 void *data)
1833{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001834 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836EXPORT_SYMBOL(read_cache_page);
1837
1838/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 * The logic we want is
1840 *
1841 * if suid or (sgid and xgrp)
1842 * remove privs
1843 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001844int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845{
1846 mode_t mode = dentry->d_inode->i_mode;
1847 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848
1849 /* suid always must be killed */
1850 if (unlikely(mode & S_ISUID))
1851 kill = ATTR_KILL_SUID;
1852
1853 /*
1854 * sgid without any exec bits is just a mandatory locking mark; leave
1855 * it alone. If some exec bits are set, it's a real sgid; kill it.
1856 */
1857 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1858 kill |= ATTR_KILL_SGID;
1859
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001860 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001861 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Jens Axboe01de85e2006-10-17 19:50:36 +02001863 return 0;
1864}
Mark Fashehd23a1472006-10-17 17:05:18 -07001865EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001866
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001867static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001868{
1869 struct iattr newattrs;
1870
1871 newattrs.ia_valid = ATTR_FORCE | kill;
1872 return notify_change(dentry, &newattrs);
1873}
1874
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001875int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001876{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001877 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001878 int killsuid = should_remove_suid(dentry);
1879 int killpriv = security_inode_need_killpriv(dentry);
1880 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001881
Serge E. Hallynb5376772007-10-16 23:31:36 -07001882 if (killpriv < 0)
1883 return killpriv;
1884 if (killpriv)
1885 error = security_inode_killpriv(dentry);
1886 if (!error && killsuid)
1887 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001888
Serge E. Hallynb5376772007-10-16 23:31:36 -07001889 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001891EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Nick Piggin2f718ff2007-10-16 01:24:59 -07001893static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 const struct iovec *iov, size_t base, size_t bytes)
1895{
Ingo Molnarf1800532009-03-02 11:00:57 +01001896 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897
1898 while (bytes) {
1899 char __user *buf = iov->iov_base + base;
1900 int copy = min(bytes, iov->iov_len - base);
1901
1902 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001903 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 copied += copy;
1905 bytes -= copy;
1906 vaddr += copy;
1907 iov++;
1908
NeilBrown01408c42006-06-25 05:47:58 -07001909 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 }
1912 return copied - left;
1913}
1914
1915/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001916 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001917 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001918 * bytes which were copied.
1919 */
1920size_t iov_iter_copy_from_user_atomic(struct page *page,
1921 struct iov_iter *i, unsigned long offset, size_t bytes)
1922{
1923 char *kaddr;
1924 size_t copied;
1925
1926 BUG_ON(!in_atomic());
1927 kaddr = kmap_atomic(page, KM_USER0);
1928 if (likely(i->nr_segs == 1)) {
1929 int left;
1930 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001931 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001932 copied = bytes - left;
1933 } else {
1934 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1935 i->iov, i->iov_offset, bytes);
1936 }
1937 kunmap_atomic(kaddr, KM_USER0);
1938
1939 return copied;
1940}
Nick Piggin89e10782007-10-16 01:25:07 -07001941EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001942
1943/*
1944 * This has the same sideeffects and return value as
1945 * iov_iter_copy_from_user_atomic().
1946 * The difference is that it attempts to resolve faults.
1947 * Page must not be locked.
1948 */
1949size_t iov_iter_copy_from_user(struct page *page,
1950 struct iov_iter *i, unsigned long offset, size_t bytes)
1951{
1952 char *kaddr;
1953 size_t copied;
1954
1955 kaddr = kmap(page);
1956 if (likely(i->nr_segs == 1)) {
1957 int left;
1958 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001959 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001960 copied = bytes - left;
1961 } else {
1962 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1963 i->iov, i->iov_offset, bytes);
1964 }
1965 kunmap(page);
1966 return copied;
1967}
Nick Piggin89e10782007-10-16 01:25:07 -07001968EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001969
Nick Pigginf7009262008-03-10 11:43:59 -07001970void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001971{
Nick Pigginf7009262008-03-10 11:43:59 -07001972 BUG_ON(i->count < bytes);
1973
Nick Piggin2f718ff2007-10-16 01:24:59 -07001974 if (likely(i->nr_segs == 1)) {
1975 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001976 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001977 } else {
1978 const struct iovec *iov = i->iov;
1979 size_t base = i->iov_offset;
1980
Nick Piggin124d3b72008-02-02 15:01:17 +01001981 /*
1982 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001983 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001984 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001985 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001986 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001987
Nick Pigginf7009262008-03-10 11:43:59 -07001988 copy = min(bytes, iov->iov_len - base);
1989 BUG_ON(!i->count || i->count < copy);
1990 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001991 bytes -= copy;
1992 base += copy;
1993 if (iov->iov_len == base) {
1994 iov++;
1995 base = 0;
1996 }
1997 }
1998 i->iov = iov;
1999 i->iov_offset = base;
2000 }
2001}
Nick Piggin89e10782007-10-16 01:25:07 -07002002EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002003
Nick Pigginafddba42007-10-16 01:25:01 -07002004/*
2005 * Fault in the first iovec of the given iov_iter, to a maximum length
2006 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2007 * accessed (ie. because it is an invalid address).
2008 *
2009 * writev-intensive code may want this to prefault several iovecs -- that
2010 * would be possible (callers must not rely on the fact that _only_ the
2011 * first iovec will be faulted with the current implementation).
2012 */
2013int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002014{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002015 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002016 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2017 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002018}
Nick Piggin89e10782007-10-16 01:25:07 -07002019EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002020
2021/*
2022 * Return the count of just the current iov_iter segment.
2023 */
2024size_t iov_iter_single_seg_count(struct iov_iter *i)
2025{
2026 const struct iovec *iov = i->iov;
2027 if (i->nr_segs == 1)
2028 return i->count;
2029 else
2030 return min(i->count, iov->iov_len - i->iov_offset);
2031}
Nick Piggin89e10782007-10-16 01:25:07 -07002032EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002033
2034/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 * Performs necessary checks before doing a write
2036 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002037 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 * Returns appropriate error code that caller should return or
2039 * zero in case that write should be allowed.
2040 */
2041inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2042{
2043 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002044 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045
2046 if (unlikely(*pos < 0))
2047 return -EINVAL;
2048
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 if (!isblk) {
2050 /* FIXME: this is for backwards compatibility with 2.4 */
2051 if (file->f_flags & O_APPEND)
2052 *pos = i_size_read(inode);
2053
2054 if (limit != RLIM_INFINITY) {
2055 if (*pos >= limit) {
2056 send_sig(SIGXFSZ, current, 0);
2057 return -EFBIG;
2058 }
2059 if (*count > limit - (typeof(limit))*pos) {
2060 *count = limit - (typeof(limit))*pos;
2061 }
2062 }
2063 }
2064
2065 /*
2066 * LFS rule
2067 */
2068 if (unlikely(*pos + *count > MAX_NON_LFS &&
2069 !(file->f_flags & O_LARGEFILE))) {
2070 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 return -EFBIG;
2072 }
2073 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2074 *count = MAX_NON_LFS - (unsigned long)*pos;
2075 }
2076 }
2077
2078 /*
2079 * Are we about to exceed the fs block limit ?
2080 *
2081 * If we have written data it becomes a short write. If we have
2082 * exceeded without writing data we send a signal and return EFBIG.
2083 * Linus frestrict idea will clean these up nicely..
2084 */
2085 if (likely(!isblk)) {
2086 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2087 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 return -EFBIG;
2089 }
2090 /* zero-length writes at ->s_maxbytes are OK */
2091 }
2092
2093 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2094 *count = inode->i_sb->s_maxbytes - *pos;
2095 } else {
David Howells93614012006-09-30 20:45:40 +02002096#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 loff_t isize;
2098 if (bdev_read_only(I_BDEV(inode)))
2099 return -EPERM;
2100 isize = i_size_read(inode);
2101 if (*pos >= isize) {
2102 if (*count || *pos > isize)
2103 return -ENOSPC;
2104 }
2105
2106 if (*pos + *count > isize)
2107 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002108#else
2109 return -EPERM;
2110#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
2112 return 0;
2113}
2114EXPORT_SYMBOL(generic_write_checks);
2115
Nick Pigginafddba42007-10-16 01:25:01 -07002116int pagecache_write_begin(struct file *file, struct address_space *mapping,
2117 loff_t pos, unsigned len, unsigned flags,
2118 struct page **pagep, void **fsdata)
2119{
2120 const struct address_space_operations *aops = mapping->a_ops;
2121
Nick Piggin4e02ed42008-10-29 14:00:55 -07002122 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002123 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002124}
2125EXPORT_SYMBOL(pagecache_write_begin);
2126
2127int pagecache_write_end(struct file *file, struct address_space *mapping,
2128 loff_t pos, unsigned len, unsigned copied,
2129 struct page *page, void *fsdata)
2130{
2131 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002132
Nick Piggin4e02ed42008-10-29 14:00:55 -07002133 mark_page_accessed(page);
2134 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002135}
2136EXPORT_SYMBOL(pagecache_write_end);
2137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138ssize_t
2139generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2140 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2141 size_t count, size_t ocount)
2142{
2143 struct file *file = iocb->ki_filp;
2144 struct address_space *mapping = file->f_mapping;
2145 struct inode *inode = mapping->host;
2146 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002147 size_t write_len;
2148 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
2150 if (count != ocount)
2151 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2152
Christoph Hellwiga969e902008-07-23 21:27:04 -07002153 write_len = iov_length(iov, *nr_segs);
2154 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002155
Nick Piggin48b47c52009-01-06 14:40:22 -08002156 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002157 if (written)
2158 goto out;
2159
2160 /*
2161 * After a write we want buffered reads to be sure to go to disk to get
2162 * the new data. We invalidate clean cached page from the region we're
2163 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002164 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002165 */
2166 if (mapping->nrpages) {
2167 written = invalidate_inode_pages2_range(mapping,
2168 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002169 /*
2170 * If a page can not be invalidated, return 0 to fall back
2171 * to buffered write.
2172 */
2173 if (written) {
2174 if (written == -EBUSY)
2175 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002176 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002177 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002178 }
2179
2180 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2181
2182 /*
2183 * Finally, try again to invalidate clean pages which might have been
2184 * cached by non-direct readahead, or faulted in by get_user_pages()
2185 * if the source of the write was an mmap'ed region of the file
2186 * we're writing. Either one is a pretty crazy thing to do,
2187 * so we don't support it 100%. If this invalidation
2188 * fails, tough, the write still worked...
2189 */
2190 if (mapping->nrpages) {
2191 invalidate_inode_pages2_range(mapping,
2192 pos >> PAGE_CACHE_SHIFT, end);
2193 }
2194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002196 pos += written;
2197 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2198 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 mark_inode_dirty(inode);
2200 }
Namhyung Kim01166512010-10-26 14:21:58 -07002201 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002203out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 return written;
2205}
2206EXPORT_SYMBOL(generic_file_direct_write);
2207
Nick Piggineb2be182007-10-16 01:24:57 -07002208/*
2209 * Find or create a page at the given pagecache position. Return the locked
2210 * page. This function is specifically for buffered writes.
2211 */
Nick Piggin54566b22009-01-04 12:00:53 -08002212struct page *grab_cache_page_write_begin(struct address_space *mapping,
2213 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002214{
2215 int status;
2216 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002217 gfp_t gfp_notmask = 0;
2218 if (flags & AOP_FLAG_NOFS)
2219 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002220repeat:
2221 page = find_lock_page(mapping, index);
2222 if (likely(page))
2223 return page;
2224
Nick Piggin54566b22009-01-04 12:00:53 -08002225 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002226 if (!page)
2227 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002228 status = add_to_page_cache_lru(page, mapping, index,
2229 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002230 if (unlikely(status)) {
2231 page_cache_release(page);
2232 if (status == -EEXIST)
2233 goto repeat;
2234 return NULL;
2235 }
2236 return page;
2237}
Nick Piggin54566b22009-01-04 12:00:53 -08002238EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002239
Nick Pigginafddba42007-10-16 01:25:01 -07002240static ssize_t generic_perform_write(struct file *file,
2241 struct iov_iter *i, loff_t pos)
2242{
2243 struct address_space *mapping = file->f_mapping;
2244 const struct address_space_operations *a_ops = mapping->a_ops;
2245 long status = 0;
2246 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002247 unsigned int flags = 0;
2248
2249 /*
2250 * Copies from kernel address space cannot fail (NFSD is a big user).
2251 */
2252 if (segment_eq(get_fs(), KERNEL_DS))
2253 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002254
2255 do {
2256 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002257 unsigned long offset; /* Offset into pagecache page */
2258 unsigned long bytes; /* Bytes to write to page */
2259 size_t copied; /* Bytes copied from user */
2260 void *fsdata;
2261
2262 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002263 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2264 iov_iter_count(i));
2265
2266again:
2267
2268 /*
2269 * Bring in the user page that we will copy from _first_.
2270 * Otherwise there's a nasty deadlock on copying from the
2271 * same page as we're writing to, without it being marked
2272 * up-to-date.
2273 *
2274 * Not only is this an optimisation, but it is also required
2275 * to check that the address is actually valid, when atomic
2276 * usercopies are used, below.
2277 */
2278 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2279 status = -EFAULT;
2280 break;
2281 }
2282
Nick Piggin674b8922007-10-16 01:25:03 -07002283 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002284 &page, &fsdata);
2285 if (unlikely(status))
2286 break;
2287
anfei zhou931e80e2010-02-02 13:44:02 -08002288 if (mapping_writably_mapped(mapping))
2289 flush_dcache_page(page);
2290
Nick Pigginafddba42007-10-16 01:25:01 -07002291 pagefault_disable();
2292 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2293 pagefault_enable();
2294 flush_dcache_page(page);
2295
Josef Bacikc8236db2009-07-05 12:08:18 -07002296 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002297 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2298 page, fsdata);
2299 if (unlikely(status < 0))
2300 break;
2301 copied = status;
2302
2303 cond_resched();
2304
Nick Piggin124d3b72008-02-02 15:01:17 +01002305 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002306 if (unlikely(copied == 0)) {
2307 /*
2308 * If we were unable to copy any data at all, we must
2309 * fall back to a single segment length write.
2310 *
2311 * If we didn't fallback here, we could livelock
2312 * because not all segments in the iov can be copied at
2313 * once without a pagefault.
2314 */
2315 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2316 iov_iter_single_seg_count(i));
2317 goto again;
2318 }
Nick Pigginafddba42007-10-16 01:25:01 -07002319 pos += copied;
2320 written += copied;
2321
2322 balance_dirty_pages_ratelimited(mapping);
2323
2324 } while (iov_iter_count(i));
2325
2326 return written ? written : status;
2327}
2328
2329ssize_t
2330generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2331 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2332 size_t count, ssize_t written)
2333{
2334 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002335 ssize_t status;
2336 struct iov_iter i;
2337
2338 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002339 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002340
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002342 written += status;
2343 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 }
2345
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 return written ? written : status;
2347}
2348EXPORT_SYMBOL(generic_file_buffered_write);
2349
Jan Karae4dd9de2009-08-17 18:10:06 +02002350/**
2351 * __generic_file_aio_write - write data to a file
2352 * @iocb: IO state structure (file, offset, etc.)
2353 * @iov: vector with data to write
2354 * @nr_segs: number of segments in the vector
2355 * @ppos: position where to write
2356 *
2357 * This function does all the work needed for actually writing data to a
2358 * file. It does all basic checks, removes SUID from the file, updates
2359 * modification times and calls proper subroutines depending on whether we
2360 * do direct IO or a standard buffered write.
2361 *
2362 * It expects i_mutex to be grabbed unless we work on a block device or similar
2363 * object which does not need locking at all.
2364 *
2365 * This function does *not* take care of syncing data in case of O_SYNC write.
2366 * A caller has to handle it. This is mainly due to the fact that we want to
2367 * avoid syncing under i_mutex.
2368 */
2369ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2370 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371{
2372 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002373 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 size_t ocount; /* original count */
2375 size_t count; /* after file limit checks */
2376 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 loff_t pos;
2378 ssize_t written;
2379 ssize_t err;
2380
2381 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002382 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2383 if (err)
2384 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385
2386 count = ocount;
2387 pos = *ppos;
2388
2389 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2390
2391 /* We can write back this queue in page reclaim */
2392 current->backing_dev_info = mapping->backing_dev_info;
2393 written = 0;
2394
2395 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2396 if (err)
2397 goto out;
2398
2399 if (count == 0)
2400 goto out;
2401
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002402 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 if (err)
2404 goto out;
2405
Christoph Hellwig870f4812006-01-09 20:52:01 -08002406 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
2408 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2409 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002410 loff_t endbyte;
2411 ssize_t written_buffered;
2412
2413 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2414 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (written < 0 || written == count)
2416 goto out;
2417 /*
2418 * direct-io write to a hole: fall through to buffered I/O
2419 * for completing the rest of the request.
2420 */
2421 pos += written;
2422 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002423 written_buffered = generic_file_buffered_write(iocb, iov,
2424 nr_segs, pos, ppos, count,
2425 written);
2426 /*
2427 * If generic_file_buffered_write() retuned a synchronous error
2428 * then we want to return the number of bytes which were
2429 * direct-written, or the error code if that was zero. Note
2430 * that this differs from normal direct-io semantics, which
2431 * will return -EFOO even if some bytes were written.
2432 */
2433 if (written_buffered < 0) {
2434 err = written_buffered;
2435 goto out;
2436 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002438 /*
2439 * We need to ensure that the page cache pages are written to
2440 * disk and invalidated to preserve the expected O_DIRECT
2441 * semantics.
2442 */
2443 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002444 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002445 if (err == 0) {
2446 written = written_buffered;
2447 invalidate_mapping_pages(mapping,
2448 pos >> PAGE_CACHE_SHIFT,
2449 endbyte >> PAGE_CACHE_SHIFT);
2450 } else {
2451 /*
2452 * We don't know how much we wrote, so just return
2453 * the number of bytes which were direct-written
2454 */
2455 }
2456 } else {
2457 written = generic_file_buffered_write(iocb, iov, nr_segs,
2458 pos, ppos, count, written);
2459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460out:
2461 current->backing_dev_info = NULL;
2462 return written ? written : err;
2463}
Jan Karae4dd9de2009-08-17 18:10:06 +02002464EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465
Jan Karae4dd9de2009-08-17 18:10:06 +02002466/**
2467 * generic_file_aio_write - write data to a file
2468 * @iocb: IO state structure
2469 * @iov: vector with data to write
2470 * @nr_segs: number of segments in the vector
2471 * @pos: position in file where to write
2472 *
2473 * This is a wrapper around __generic_file_aio_write() to be used by most
2474 * filesystems. It takes care of syncing the file in case of O_SYNC file
2475 * and acquires i_mutex as needed.
2476 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002477ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2478 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
2480 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002481 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
2484 BUG_ON(iocb->ki_pos != pos);
2485
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002486 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002487 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002488 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489
Jan Kara148f9482009-08-17 19:52:36 +02002490 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 ssize_t err;
2492
Jan Kara148f9482009-08-17 19:52:36 +02002493 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002494 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 ret = err;
2496 }
2497 return ret;
2498}
2499EXPORT_SYMBOL(generic_file_aio_write);
2500
David Howellscf9a2ae2006-08-29 19:05:54 +01002501/**
2502 * try_to_release_page() - release old fs-specific metadata on a page
2503 *
2504 * @page: the page which the kernel is trying to free
2505 * @gfp_mask: memory allocation flags (and I/O mode)
2506 *
2507 * The address_space is to try to release any data against the page
2508 * (presumably at page->private). If the release was successful, return `1'.
2509 * Otherwise return zero.
2510 *
David Howells266cf652009-04-03 16:42:36 +01002511 * This may also be called if PG_fscache is set on a page, indicating that the
2512 * page is known to the local caching routines.
2513 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002514 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002515 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002516 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002517 */
2518int try_to_release_page(struct page *page, gfp_t gfp_mask)
2519{
2520 struct address_space * const mapping = page->mapping;
2521
2522 BUG_ON(!PageLocked(page));
2523 if (PageWriteback(page))
2524 return 0;
2525
2526 if (mapping && mapping->a_ops->releasepage)
2527 return mapping->a_ops->releasepage(page, gfp_mask);
2528 return try_to_free_buffers(page);
2529}
2530
2531EXPORT_SYMBOL(try_to_release_page);