blob: 698ea80f21022bf709dacc1101c614327109d828 [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>
13#include <linux/slab.h>
14#include <linux/compiler.h>
15#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
20#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}
154
155static int sync_page(void *word)
156{
157 struct address_space *mapping;
158 struct page *page;
159
Andi Kleen07808b72005-11-05 17:25:53 +0100160 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700163 * page_mapping() is being called without PG_locked held.
164 * Some knowledge of the state and use of the page is used to
165 * reduce the requirements down to a memory barrier.
166 * The danger here is of a stale page_mapping() return value
167 * indicating a struct address_space different from the one it's
168 * associated with when it is associated with one.
169 * After smp_mb(), it's either the correct page_mapping() for
170 * the page, or an old page_mapping() and the page's own
171 * page_mapping() has gone NULL.
172 * The ->sync_page() address_space operation must tolerate
173 * page_mapping() going NULL. By an amazing coincidence,
174 * this comes about because none of the users of the page
175 * in the ->sync_page() methods make essential use of the
176 * page_mapping(), merely passing the page down to the backing
177 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700178 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700179 * of interest. When page_mapping() does go NULL, the entire
180 * call stack gracefully ignores the page and returns.
181 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
183 smp_mb();
184 mapping = page_mapping(page);
185 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
186 mapping->a_ops->sync_page(page);
187 io_schedule();
188 return 0;
189}
190
Matthew Wilcox2687a352007-12-06 11:18:49 -0500191static int sync_page_killable(void *word)
192{
193 sync_page(word);
194 return fatal_signal_pending(current) ? -EINTR : 0;
195}
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700198 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700199 * @mapping: address space structure to write
200 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800201 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700202 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700204 * Start writeback against all of a mapping's dirty pages that lie
205 * within the byte offsets <start, end> inclusive.
206 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700208 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * these two operations is that if a dirty page/buffer is encountered, it must
210 * be waited upon, and not just skipped over.
211 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800212int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
213 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 int ret;
216 struct writeback_control wbc = {
217 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800218 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700219 .range_start = start,
220 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 };
222
223 if (!mapping_cap_writeback_dirty(mapping))
224 return 0;
225
226 ret = do_writepages(mapping, &wbc);
227 return ret;
228}
229
230static inline int __filemap_fdatawrite(struct address_space *mapping,
231 int sync_mode)
232{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700233 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236int filemap_fdatawrite(struct address_space *mapping)
237{
238 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
239}
240EXPORT_SYMBOL(filemap_fdatawrite);
241
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400242int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800243 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
246}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400247EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Randy Dunlap485bb992006-06-23 02:03:49 -0700249/**
250 * filemap_flush - mostly a non-blocking flush
251 * @mapping: target address_space
252 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 * This is a mostly non-blocking flush. Not suitable for data-integrity
254 * purposes - I/O may not be started against all dirty pages.
255 */
256int filemap_flush(struct address_space *mapping)
257{
258 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
259}
260EXPORT_SYMBOL(filemap_flush);
261
Randy Dunlap485bb992006-06-23 02:03:49 -0700262/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200263 * filemap_fdatawait_range - wait for writeback to complete
264 * @mapping: address space structure to wait for
265 * @start_byte: offset in bytes where the range starts
266 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700267 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200268 * Walk the list of under-writeback pages of the given address space
269 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200271int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
272 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200274 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
275 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 struct pagevec pvec;
277 int nr_pages;
278 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200280 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return 0;
282
283 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 while ((index <= end) &&
285 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
286 PAGECACHE_TAG_WRITEBACK,
287 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
288 unsigned i;
289
290 for (i = 0; i < nr_pages; i++) {
291 struct page *page = pvec.pages[i];
292
293 /* until radix tree lookup accepts end_index */
294 if (page->index > end)
295 continue;
296
297 wait_on_page_writeback(page);
298 if (PageError(page))
299 ret = -EIO;
300 }
301 pagevec_release(&pvec);
302 cond_resched();
303 }
304
305 /* Check for outstanding write errors */
306 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
307 ret = -ENOSPC;
308 if (test_and_clear_bit(AS_EIO, &mapping->flags))
309 ret = -EIO;
310
311 return ret;
312}
Jan Karad3bccb6f2009-08-17 19:30:27 +0200313EXPORT_SYMBOL(filemap_fdatawait_range);
314
315/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700316 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700318 *
319 * Walk the list of under-writeback pages of the given address space
320 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 */
322int filemap_fdatawait(struct address_space *mapping)
323{
324 loff_t i_size = i_size_read(mapping->host);
325
326 if (i_size == 0)
327 return 0;
328
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200329 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331EXPORT_SYMBOL(filemap_fdatawait);
332
333int filemap_write_and_wait(struct address_space *mapping)
334{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800335 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
337 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800338 err = filemap_fdatawrite(mapping);
339 /*
340 * Even if the above returned error, the pages may be
341 * written partially (e.g. -ENOSPC), so we wait for it.
342 * But the -EIO is special case, it may indicate the worst
343 * thing (e.g. bug) happened, so we avoid waiting for it.
344 */
345 if (err != -EIO) {
346 int err2 = filemap_fdatawait(mapping);
347 if (!err)
348 err = err2;
349 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800351 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800353EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
Randy Dunlap485bb992006-06-23 02:03:49 -0700355/**
356 * filemap_write_and_wait_range - write out & wait on a file range
357 * @mapping: the address_space for the pages
358 * @lstart: offset in bytes where the range starts
359 * @lend: offset in bytes where the range ends (inclusive)
360 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800361 * Write out and wait upon file offsets lstart->lend, inclusive.
362 *
363 * Note that `lend' is inclusive (describes the last byte to be written) so
364 * that this function can be used to write to the very end-of-file (end = -1).
365 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366int filemap_write_and_wait_range(struct address_space *mapping,
367 loff_t lstart, loff_t lend)
368{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800369 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
371 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800372 err = __filemap_fdatawrite_range(mapping, lstart, lend,
373 WB_SYNC_ALL);
374 /* See comment of filemap_write_and_wait() */
375 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200376 int err2 = filemap_fdatawait_range(mapping,
377 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800378 if (!err)
379 err = err2;
380 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800382 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
Chris Masonf6995582009-04-15 13:22:37 -0400384EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
Randy Dunlap485bb992006-06-23 02:03:49 -0700386/**
Nick Piggine2867812008-07-25 19:45:30 -0700387 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700388 * @page: page to add
389 * @mapping: the page's address_space
390 * @offset: page index
391 * @gfp_mask: page allocation mode
392 *
Nick Piggine2867812008-07-25 19:45:30 -0700393 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 * This function does not add the page to the LRU. The caller must do that.
395 */
Nick Piggine2867812008-07-25 19:45:30 -0700396int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400397 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Nick Piggine2867812008-07-25 19:45:30 -0700399 int error;
400
401 VM_BUG_ON(!PageLocked(page));
402
403 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800404 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800405 if (error)
406 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
Balbir Singh35c754d2008-02-07 00:14:05 -0800408 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700410 page_cache_get(page);
411 page->mapping = mapping;
412 page->index = offset;
413
Nick Piggin19fd6232008-07-25 19:45:32 -0700414 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700416 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700418 __inc_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700419 if (PageSwapBacked(page))
420 __inc_zone_page_state(page, NR_SHMEM);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700421 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700422 } else {
423 page->mapping = NULL;
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700424 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700425 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700426 page_cache_release(page);
427 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800429 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700430 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800431out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 return error;
433}
Nick Piggine2867812008-07-25 19:45:30 -0700434EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435
436int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400437 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700439 int ret;
440
441 /*
442 * Splice_read and readahead add shmem/tmpfs pages into the page cache
443 * before shmem_readpage has a chance to mark them as SwapBacked: they
444 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
445 * (called in add_to_page_cache) needs to know where they're going too.
446 */
447 if (mapping_cap_swap_backed(mapping))
448 SetPageSwapBacked(page);
449
450 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
451 if (ret == 0) {
452 if (page_is_file_cache(page))
453 lru_cache_add_file(page);
454 else
455 lru_cache_add_active_anon(page);
456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 return ret;
458}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300459EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Paul Jackson44110fe2006-03-24 03:16:04 -0800461#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700462struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800463{
464 if (cpuset_do_page_mem_spread()) {
465 int n = cpuset_mem_spread_node();
Mel Gorman6484eb32009-06-16 15:31:54 -0700466 return alloc_pages_exact_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800467 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700468 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800469}
Nick Piggin2ae88142006-10-28 10:38:23 -0700470EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800471#endif
472
Nick Piggindb376482006-09-25 23:31:24 -0700473static int __sleep_on_page_lock(void *word)
474{
475 io_schedule();
476 return 0;
477}
478
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479/*
480 * In order to wait for pages to become available there must be
481 * waitqueues associated with pages. By using a hash table of
482 * waitqueues where the bucket discipline is to maintain all
483 * waiters on the same queue and wake all when any of the pages
484 * become available, and for the woken contexts to check to be
485 * sure the appropriate page became available, this saves space
486 * at a cost of "thundering herd" phenomena during rare hash
487 * collisions.
488 */
489static wait_queue_head_t *page_waitqueue(struct page *page)
490{
491 const struct zone *zone = page_zone(page);
492
493 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
494}
495
496static inline void wake_up_page(struct page *page, int bit)
497{
498 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
499}
500
Harvey Harrison920c7a52008-02-04 22:29:26 -0800501void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
504
505 if (test_bit(bit_nr, &page->flags))
506 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
507 TASK_UNINTERRUPTIBLE);
508}
509EXPORT_SYMBOL(wait_on_page_bit);
510
511/**
David Howells385e1ca5f2009-04-03 16:42:39 +0100512 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700513 * @page: Page defining the wait queue of interest
514 * @waiter: Waiter to add to the queue
David Howells385e1ca5f2009-04-03 16:42:39 +0100515 *
516 * Add an arbitrary @waiter to the wait queue for the nominated @page.
517 */
518void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
519{
520 wait_queue_head_t *q = page_waitqueue(page);
521 unsigned long flags;
522
523 spin_lock_irqsave(&q->lock, flags);
524 __add_wait_queue(q, waiter);
525 spin_unlock_irqrestore(&q->lock, flags);
526}
527EXPORT_SYMBOL_GPL(add_page_wait_queue);
528
529/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700530 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 * @page: the page
532 *
533 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
534 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
535 * mechananism between PageLocked pages and PageWriteback pages is shared.
536 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
537 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700538 * The mb is necessary to enforce ordering between the clear_bit and the read
539 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800541void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
Nick Piggin8413ac92008-10-18 20:26:59 -0700543 VM_BUG_ON(!PageLocked(page));
544 clear_bit_unlock(PG_locked, &page->flags);
545 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 wake_up_page(page, PG_locked);
547}
548EXPORT_SYMBOL(unlock_page);
549
Randy Dunlap485bb992006-06-23 02:03:49 -0700550/**
551 * end_page_writeback - end writeback against a page
552 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 */
554void end_page_writeback(struct page *page)
555{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700556 if (TestClearPageReclaim(page))
557 rotate_reclaimable_page(page);
558
559 if (!test_clear_page_writeback(page))
560 BUG();
561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 smp_mb__after_clear_bit();
563 wake_up_page(page, PG_writeback);
564}
565EXPORT_SYMBOL(end_page_writeback);
566
Randy Dunlap485bb992006-06-23 02:03:49 -0700567/**
568 * __lock_page - get a lock on the page, assuming we need to sleep to get it
569 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700571 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
573 * chances are that on the second loop, the block layer's plug list is empty,
574 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
575 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800576void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577{
578 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
579
580 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
581 TASK_UNINTERRUPTIBLE);
582}
583EXPORT_SYMBOL(__lock_page);
584
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800585int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500586{
587 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
588
589 return __wait_on_bit_lock(page_waitqueue(page), &wait,
590 sync_page_killable, TASK_KILLABLE);
591}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300592EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500593
Randy Dunlap76824862008-03-19 17:00:40 -0700594/**
595 * __lock_page_nosync - get a lock on the page, without calling sync_page()
596 * @page: the page to lock
597 *
Nick Piggindb376482006-09-25 23:31:24 -0700598 * Variant of lock_page that does not require the caller to hold a reference
599 * on the page's mapping.
600 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800601void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700602{
603 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
604 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
605 TASK_UNINTERRUPTIBLE);
606}
607
Randy Dunlap485bb992006-06-23 02:03:49 -0700608/**
609 * find_get_page - find and get a page reference
610 * @mapping: the address_space to search
611 * @offset: the page index
612 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700613 * Is there a pagecache struct page at the given (mapping, offset) tuple?
614 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 */
Nick Piggina60637c2008-07-25 19:45:31 -0700616struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Nick Piggina60637c2008-07-25 19:45:31 -0700618 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 struct page *page;
620
Nick Piggina60637c2008-07-25 19:45:31 -0700621 rcu_read_lock();
622repeat:
623 page = NULL;
624 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
625 if (pagep) {
626 page = radix_tree_deref_slot(pagep);
627 if (unlikely(!page || page == RADIX_TREE_RETRY))
628 goto repeat;
629
630 if (!page_cache_get_speculative(page))
631 goto repeat;
632
633 /*
634 * Has the page moved?
635 * This is part of the lockless pagecache protocol. See
636 * include/linux/pagemap.h for details.
637 */
638 if (unlikely(page != *pagep)) {
639 page_cache_release(page);
640 goto repeat;
641 }
642 }
643 rcu_read_unlock();
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 return page;
646}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647EXPORT_SYMBOL(find_get_page);
648
Randy Dunlap485bb992006-06-23 02:03:49 -0700649/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700651 * @mapping: the address_space to search
652 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 *
654 * Locates the desired pagecache page, locks it, increments its reference
655 * count and returns its address.
656 *
657 * Returns zero if the page was not present. find_lock_page() may sleep.
658 */
Nick Piggina60637c2008-07-25 19:45:31 -0700659struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 struct page *page;
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700664 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700666 lock_page(page);
667 /* Has the page been truncated? */
668 if (unlikely(page->mapping != mapping)) {
669 unlock_page(page);
670 page_cache_release(page);
671 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
Nick Piggina60637c2008-07-25 19:45:31 -0700673 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 return page;
676}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677EXPORT_SYMBOL(find_lock_page);
678
679/**
680 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700681 * @mapping: the page's address_space
682 * @index: the page's index into the mapping
683 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 *
685 * Locates a page in the pagecache. If the page is not present, a new page
686 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
687 * LRU list. The returned page is locked and has its reference count
688 * incremented.
689 *
690 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
691 * allocation!
692 *
693 * find_or_create_page() returns the desired page's address, or zero on
694 * memory exhaustion.
695 */
696struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700697 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
Nick Piggineb2be182007-10-16 01:24:57 -0700699 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 int err;
701repeat:
702 page = find_lock_page(mapping, index);
703 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700704 page = __page_cache_alloc(gfp_mask);
705 if (!page)
706 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800707 /*
708 * We want a regular kernel memory (not highmem or DMA etc)
709 * allocation for the radix tree nodes, but we need to honour
710 * the context-specific requirements the caller has asked for.
711 * GFP_RECLAIM_MASK collects those requirements.
712 */
713 err = add_to_page_cache_lru(page, mapping, index,
714 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700715 if (unlikely(err)) {
716 page_cache_release(page);
717 page = NULL;
718 if (err == -EEXIST)
719 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 return page;
723}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724EXPORT_SYMBOL(find_or_create_page);
725
726/**
727 * find_get_pages - gang pagecache lookup
728 * @mapping: The address_space to search
729 * @start: The starting page index
730 * @nr_pages: The maximum number of pages
731 * @pages: Where the resulting pages are placed
732 *
733 * find_get_pages() will search for and return a group of up to
734 * @nr_pages pages in the mapping. The pages are placed at @pages.
735 * find_get_pages() takes a reference against the returned pages.
736 *
737 * The search returns a group of mapping-contiguous pages with ascending
738 * indexes. There may be holes in the indices due to not-present pages.
739 *
740 * find_get_pages() returns the number of pages which were found.
741 */
742unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
743 unsigned int nr_pages, struct page **pages)
744{
745 unsigned int i;
746 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700747 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Nick Piggina60637c2008-07-25 19:45:31 -0700749 rcu_read_lock();
750restart:
751 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
752 (void ***)pages, start, nr_pages);
753 ret = 0;
754 for (i = 0; i < nr_found; i++) {
755 struct page *page;
756repeat:
757 page = radix_tree_deref_slot((void **)pages[i]);
758 if (unlikely(!page))
759 continue;
760 /*
761 * this can only trigger if nr_found == 1, making livelock
762 * a non issue.
763 */
764 if (unlikely(page == RADIX_TREE_RETRY))
765 goto restart;
766
767 if (!page_cache_get_speculative(page))
768 goto repeat;
769
770 /* Has the page moved? */
771 if (unlikely(page != *((void **)pages[i]))) {
772 page_cache_release(page);
773 goto repeat;
774 }
775
776 pages[ret] = page;
777 ret++;
778 }
779 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 return ret;
781}
782
Jens Axboeebf43502006-04-27 08:46:01 +0200783/**
784 * find_get_pages_contig - gang contiguous pagecache lookup
785 * @mapping: The address_space to search
786 * @index: The starting page index
787 * @nr_pages: The maximum number of pages
788 * @pages: Where the resulting pages are placed
789 *
790 * find_get_pages_contig() works exactly like find_get_pages(), except
791 * that the returned number of pages are guaranteed to be contiguous.
792 *
793 * find_get_pages_contig() returns the number of pages which were found.
794 */
795unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
796 unsigned int nr_pages, struct page **pages)
797{
798 unsigned int i;
799 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700800 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200801
Nick Piggina60637c2008-07-25 19:45:31 -0700802 rcu_read_lock();
803restart:
804 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
805 (void ***)pages, index, nr_pages);
806 ret = 0;
807 for (i = 0; i < nr_found; i++) {
808 struct page *page;
809repeat:
810 page = radix_tree_deref_slot((void **)pages[i]);
811 if (unlikely(!page))
812 continue;
813 /*
814 * this can only trigger if nr_found == 1, making livelock
815 * a non issue.
816 */
817 if (unlikely(page == RADIX_TREE_RETRY))
818 goto restart;
819
820 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200821 break;
822
Nick Piggina60637c2008-07-25 19:45:31 -0700823 if (!page_cache_get_speculative(page))
824 goto repeat;
825
826 /* Has the page moved? */
827 if (unlikely(page != *((void **)pages[i]))) {
828 page_cache_release(page);
829 goto repeat;
830 }
831
832 pages[ret] = page;
833 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200834 index++;
835 }
Nick Piggina60637c2008-07-25 19:45:31 -0700836 rcu_read_unlock();
837 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200838}
David Howellsef71c152007-05-09 02:33:44 -0700839EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200840
Randy Dunlap485bb992006-06-23 02:03:49 -0700841/**
842 * find_get_pages_tag - find and return pages that match @tag
843 * @mapping: the address_space to search
844 * @index: the starting page index
845 * @tag: the tag index
846 * @nr_pages: the maximum number of pages
847 * @pages: where the resulting pages are placed
848 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700850 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 */
852unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
853 int tag, unsigned int nr_pages, struct page **pages)
854{
855 unsigned int i;
856 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700857 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Nick Piggina60637c2008-07-25 19:45:31 -0700859 rcu_read_lock();
860restart:
861 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
862 (void ***)pages, *index, nr_pages, tag);
863 ret = 0;
864 for (i = 0; i < nr_found; i++) {
865 struct page *page;
866repeat:
867 page = radix_tree_deref_slot((void **)pages[i]);
868 if (unlikely(!page))
869 continue;
870 /*
871 * this can only trigger if nr_found == 1, making livelock
872 * a non issue.
873 */
874 if (unlikely(page == RADIX_TREE_RETRY))
875 goto restart;
876
877 if (!page_cache_get_speculative(page))
878 goto repeat;
879
880 /* Has the page moved? */
881 if (unlikely(page != *((void **)pages[i]))) {
882 page_cache_release(page);
883 goto repeat;
884 }
885
886 pages[ret] = page;
887 ret++;
888 }
889 rcu_read_unlock();
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if (ret)
892 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 return ret;
895}
David Howellsef71c152007-05-09 02:33:44 -0700896EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
Randy Dunlap485bb992006-06-23 02:03:49 -0700898/**
899 * grab_cache_page_nowait - returns locked page at given index in given cache
900 * @mapping: target address_space
901 * @index: the page index
902 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800903 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 * This is intended for speculative data generators, where the data can
905 * be regenerated if the page couldn't be grabbed. This routine should
906 * be safe to call while holding the lock for another page.
907 *
908 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
909 * and deadlock against the caller's locked page.
910 */
911struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700912grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
914 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200917 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 return page;
919 page_cache_release(page);
920 return NULL;
921 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700922 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800923 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 page_cache_release(page);
925 page = NULL;
926 }
927 return page;
928}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929EXPORT_SYMBOL(grab_cache_page_nowait);
930
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700931/*
932 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
933 * a _large_ part of the i/o request. Imagine the worst scenario:
934 *
935 * ---R__________________________________________B__________
936 * ^ reading here ^ bad block(assume 4k)
937 *
938 * read(R) => miss => readahead(R...B) => media error => frustrating retries
939 * => failing the whole request => read(R) => read(R+1) =>
940 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
941 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
942 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
943 *
944 * It is going insane. Fix it by quickly scaling down the readahead size.
945 */
946static void shrink_readahead_size_eio(struct file *filp,
947 struct file_ra_state *ra)
948{
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700949 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700950}
951
Randy Dunlap485bb992006-06-23 02:03:49 -0700952/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800953 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700954 * @filp: the file to read
955 * @ppos: current file position
956 * @desc: read_descriptor
957 * @actor: read method
958 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700960 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 *
962 * This is really ugly. But the goto's actually try to clarify some
963 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800965static void do_generic_file_read(struct file *filp, loff_t *ppos,
966 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800968 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800970 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700971 pgoff_t index;
972 pgoff_t last_index;
973 pgoff_t prev_index;
974 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700975 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700979 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
980 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
982 offset = *ppos & ~PAGE_CACHE_MASK;
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 for (;;) {
985 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700986 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -0700987 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 unsigned long nr, ret;
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991find_page:
992 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700993 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -0700994 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -0700995 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -0700996 index, last_index - index);
997 page = find_get_page(mapping, index);
998 if (unlikely(page == NULL))
999 goto no_cached_page;
1000 }
1001 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001002 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001003 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001004 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001006 if (!PageUptodate(page)) {
1007 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1008 !mapping->a_ops->is_partially_uptodate)
1009 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001010 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001011 goto page_not_up_to_date;
1012 if (!mapping->a_ops->is_partially_uptodate(page,
1013 desc, offset))
1014 goto page_not_up_to_date_locked;
1015 unlock_page(page);
1016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001018 /*
1019 * i_size must be checked after we know the page is Uptodate.
1020 *
1021 * Checking i_size after the check allows us to calculate
1022 * the correct value for "nr", which means the zero-filled
1023 * part of the page is not copied back to userspace (unless
1024 * another truncate extends the file - this is desired though).
1025 */
1026
1027 isize = i_size_read(inode);
1028 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1029 if (unlikely(!isize || index > end_index)) {
1030 page_cache_release(page);
1031 goto out;
1032 }
1033
1034 /* nr is the maximum number of bytes to copy from this page */
1035 nr = PAGE_CACHE_SIZE;
1036 if (index == end_index) {
1037 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1038 if (nr <= offset) {
1039 page_cache_release(page);
1040 goto out;
1041 }
1042 }
1043 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 /* If users can be writing to this page using arbitrary
1046 * virtual addresses, take care about potential aliasing
1047 * before reading the page on the kernel side.
1048 */
1049 if (mapping_writably_mapped(mapping))
1050 flush_dcache_page(page);
1051
1052 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001053 * When a sequential read accesses a page several times,
1054 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 */
Jan Karaec0f1632007-05-06 14:49:25 -07001056 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 mark_page_accessed(page);
1058 prev_index = index;
1059
1060 /*
1061 * Ok, we have the page, and it's up-to-date, so
1062 * now we can copy it to user space...
1063 *
1064 * The actor routine returns how many bytes were actually used..
1065 * NOTE! This may not be the same as how much of a user buffer
1066 * we filled up (we may be padding etc), so we can only update
1067 * "pos" here (the actor routine has to update the user buffer
1068 * pointers and the remaining count).
1069 */
1070 ret = actor(desc, page, offset, nr);
1071 offset += ret;
1072 index += offset >> PAGE_CACHE_SHIFT;
1073 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001074 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075
1076 page_cache_release(page);
1077 if (ret == nr && desc->count)
1078 continue;
1079 goto out;
1080
1081page_not_up_to_date:
1082 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001083 error = lock_page_killable(page);
1084 if (unlikely(error))
1085 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001087page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001088 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 if (!page->mapping) {
1090 unlock_page(page);
1091 page_cache_release(page);
1092 continue;
1093 }
1094
1095 /* Did somebody else fill it already? */
1096 if (PageUptodate(page)) {
1097 unlock_page(page);
1098 goto page_ok;
1099 }
1100
1101readpage:
1102 /* Start the actual read. The read will unlock the page. */
1103 error = mapping->a_ops->readpage(filp, page);
1104
Zach Brown994fc28c2005-12-15 14:28:17 -08001105 if (unlikely(error)) {
1106 if (error == AOP_TRUNCATED_PAGE) {
1107 page_cache_release(page);
1108 goto find_page;
1109 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001114 error = lock_page_killable(page);
1115 if (unlikely(error))
1116 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 if (!PageUptodate(page)) {
1118 if (page->mapping == NULL) {
1119 /*
1120 * invalidate_inode_pages got it
1121 */
1122 unlock_page(page);
1123 page_cache_release(page);
1124 goto find_page;
1125 }
1126 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001127 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001128 error = -EIO;
1129 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
1131 unlock_page(page);
1132 }
1133
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 goto page_ok;
1135
1136readpage_error:
1137 /* UHHUH! A synchronous read error occurred. Report it */
1138 desc->error = error;
1139 page_cache_release(page);
1140 goto out;
1141
1142no_cached_page:
1143 /*
1144 * Ok, it wasn't cached, so we need to create a new
1145 * page..
1146 */
Nick Piggineb2be182007-10-16 01:24:57 -07001147 page = page_cache_alloc_cold(mapping);
1148 if (!page) {
1149 desc->error = -ENOMEM;
1150 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 }
Nick Piggineb2be182007-10-16 01:24:57 -07001152 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 index, GFP_KERNEL);
1154 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001155 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (error == -EEXIST)
1157 goto find_page;
1158 desc->error = error;
1159 goto out;
1160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 goto readpage;
1162 }
1163
1164out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001165 ra->prev_pos = prev_index;
1166 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1167 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001169 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001170 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
1173int file_read_actor(read_descriptor_t *desc, struct page *page,
1174 unsigned long offset, unsigned long size)
1175{
1176 char *kaddr;
1177 unsigned long left, count = desc->count;
1178
1179 if (size > count)
1180 size = count;
1181
1182 /*
1183 * Faults on the destination of a read are common, so do it before
1184 * taking the kmap.
1185 */
1186 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1187 kaddr = kmap_atomic(page, KM_USER0);
1188 left = __copy_to_user_inatomic(desc->arg.buf,
1189 kaddr + offset, size);
1190 kunmap_atomic(kaddr, KM_USER0);
1191 if (left == 0)
1192 goto success;
1193 }
1194
1195 /* Do it the slow way */
1196 kaddr = kmap(page);
1197 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1198 kunmap(page);
1199
1200 if (left) {
1201 size -= left;
1202 desc->error = -EFAULT;
1203 }
1204success:
1205 desc->count = count - size;
1206 desc->written += size;
1207 desc->arg.buf += size;
1208 return size;
1209}
1210
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001211/*
1212 * Performs necessary checks before doing a write
1213 * @iov: io vector request
1214 * @nr_segs: number of segments in the iovec
1215 * @count: number of bytes to write
1216 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1217 *
1218 * Adjust number of segments and amount of bytes to write (nr_segs should be
1219 * properly initialized first). Returns appropriate error code that caller
1220 * should return or zero in case that write should be allowed.
1221 */
1222int generic_segment_checks(const struct iovec *iov,
1223 unsigned long *nr_segs, size_t *count, int access_flags)
1224{
1225 unsigned long seg;
1226 size_t cnt = 0;
1227 for (seg = 0; seg < *nr_segs; seg++) {
1228 const struct iovec *iv = &iov[seg];
1229
1230 /*
1231 * If any segment has a negative length, or the cumulative
1232 * length ever wraps negative then return -EINVAL.
1233 */
1234 cnt += iv->iov_len;
1235 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1236 return -EINVAL;
1237 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1238 continue;
1239 if (seg == 0)
1240 return -EFAULT;
1241 *nr_segs = seg;
1242 cnt -= iv->iov_len; /* This segment is no good */
1243 break;
1244 }
1245 *count = cnt;
1246 return 0;
1247}
1248EXPORT_SYMBOL(generic_segment_checks);
1249
Randy Dunlap485bb992006-06-23 02:03:49 -07001250/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001251 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001252 * @iocb: kernel I/O control block
1253 * @iov: io vector request
1254 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001255 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001256 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 * This is the "read()" routine for all filesystems
1258 * that can use the page cache directly.
1259 */
1260ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001261generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1262 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263{
1264 struct file *filp = iocb->ki_filp;
1265 ssize_t retval;
1266 unsigned long seg;
1267 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001268 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269
1270 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001271 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1272 if (retval)
1273 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
1275 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1276 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001277 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 struct address_space *mapping;
1279 struct inode *inode;
1280
1281 mapping = filp->f_mapping;
1282 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 if (!count)
1284 goto out; /* skip atime */
1285 size = i_size_read(inode);
1286 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001287 retval = filemap_write_and_wait_range(mapping, pos,
1288 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001289 if (!retval) {
1290 retval = mapping->a_ops->direct_IO(READ, iocb,
1291 iov, pos, nr_segs);
1292 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 if (retval > 0)
1294 *ppos = pos + retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001295 if (retval) {
1296 file_accessed(filp);
1297 goto out;
1298 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001299 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 }
1301
Hugh Dickins11fa9772008-07-23 21:27:34 -07001302 for (seg = 0; seg < nr_segs; seg++) {
1303 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Hugh Dickins11fa9772008-07-23 21:27:34 -07001305 desc.written = 0;
1306 desc.arg.buf = iov[seg].iov_base;
1307 desc.count = iov[seg].iov_len;
1308 if (desc.count == 0)
1309 continue;
1310 desc.error = 0;
1311 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1312 retval += desc.written;
1313 if (desc.error) {
1314 retval = retval ?: desc.error;
1315 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001317 if (desc.count > 0)
1318 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 }
1320out:
1321 return retval;
1322}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323EXPORT_SYMBOL(generic_file_aio_read);
1324
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325static ssize_t
1326do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001327 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
1329 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1330 return -EINVAL;
1331
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001332 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 return 0;
1334}
1335
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001336SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
1338 ssize_t ret;
1339 struct file *file;
1340
1341 ret = -EBADF;
1342 file = fget(fd);
1343 if (file) {
1344 if (file->f_mode & FMODE_READ) {
1345 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001346 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1347 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 unsigned long len = end - start + 1;
1349 ret = do_readahead(mapping, file, start, len);
1350 }
1351 fput(file);
1352 }
1353 return ret;
1354}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001355#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1356asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1357{
1358 return SYSC_readahead((int) fd, offset, (size_t) count);
1359}
1360SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1361#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
1363#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001364/**
1365 * page_cache_read - adds requested page to the page cache if not already there
1366 * @file: file to read
1367 * @offset: page index
1368 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * This adds the requested page to the page cache if it isn't already there,
1370 * and schedules an I/O to read in its contents from disk.
1371 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001372static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373{
1374 struct address_space *mapping = file->f_mapping;
1375 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001376 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Zach Brown994fc28c2005-12-15 14:28:17 -08001378 do {
1379 page = page_cache_alloc_cold(mapping);
1380 if (!page)
1381 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
Zach Brown994fc28c2005-12-15 14:28:17 -08001383 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1384 if (ret == 0)
1385 ret = mapping->a_ops->readpage(file, page);
1386 else if (ret == -EEXIST)
1387 ret = 0; /* losing race to add is OK */
1388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
Zach Brown994fc28c2005-12-15 14:28:17 -08001391 } while (ret == AOP_TRUNCATED_PAGE);
1392
1393 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394}
1395
1396#define MMAP_LOTSAMISS (100)
1397
Linus Torvaldsef00e082009-06-16 15:31:25 -07001398/*
1399 * Synchronous readahead happens when we don't even find
1400 * a page in the page cache at all.
1401 */
1402static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1403 struct file_ra_state *ra,
1404 struct file *file,
1405 pgoff_t offset)
1406{
1407 unsigned long ra_pages;
1408 struct address_space *mapping = file->f_mapping;
1409
1410 /* If we don't want any read-ahead, don't bother */
1411 if (VM_RandomReadHint(vma))
1412 return;
1413
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001414 if (VM_SequentialReadHint(vma) ||
1415 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001416 page_cache_sync_readahead(mapping, ra, file, offset,
1417 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001418 return;
1419 }
1420
1421 if (ra->mmap_miss < INT_MAX)
1422 ra->mmap_miss++;
1423
1424 /*
1425 * Do we miss much more than hit in this file? If so,
1426 * stop bothering with read-ahead. It will only hurt.
1427 */
1428 if (ra->mmap_miss > MMAP_LOTSAMISS)
1429 return;
1430
Wu Fengguangd30a1102009-06-16 15:31:30 -07001431 /*
1432 * mmap read-around
1433 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001434 ra_pages = max_sane_readahead(ra->ra_pages);
1435 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001436 ra->start = max_t(long, 0, offset - ra_pages/2);
1437 ra->size = ra_pages;
1438 ra->async_size = 0;
1439 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001440 }
1441}
1442
1443/*
1444 * Asynchronous readahead happens when we find the page and PG_readahead,
1445 * so we want to possibly extend the readahead further..
1446 */
1447static void do_async_mmap_readahead(struct vm_area_struct *vma,
1448 struct file_ra_state *ra,
1449 struct file *file,
1450 struct page *page,
1451 pgoff_t offset)
1452{
1453 struct address_space *mapping = file->f_mapping;
1454
1455 /* If we don't want any read-ahead, don't bother */
1456 if (VM_RandomReadHint(vma))
1457 return;
1458 if (ra->mmap_miss > 0)
1459 ra->mmap_miss--;
1460 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001461 page_cache_async_readahead(mapping, ra, file,
1462 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001463}
1464
Randy Dunlap485bb992006-06-23 02:03:49 -07001465/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001466 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001467 * @vma: vma in which the fault was taken
1468 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001469 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001470 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 * mapped memory region to read in file data during a page fault.
1472 *
1473 * The goto's are kind of ugly, but this streamlines the normal case of having
1474 * it in the page cache, and handles the special cases reasonably without
1475 * having a lot of duplicated code.
1476 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001477int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
1479 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001480 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 struct address_space *mapping = file->f_mapping;
1482 struct file_ra_state *ra = &file->f_ra;
1483 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001484 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001486 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001487 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001490 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001491 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 * Do we have something in the page cache already?
1495 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001496 page = find_get_page(mapping, offset);
1497 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001499 * We found the page, so try async readahead before
1500 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001502 do_async_mmap_readahead(vma, ra, file, page, offset);
1503 lock_page(page);
1504
1505 /* Did it get truncated? */
1506 if (unlikely(page->mapping != mapping)) {
1507 unlock_page(page);
1508 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 goto no_cached_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 }
Linus Torvaldsef00e082009-06-16 15:31:25 -07001511 } else {
1512 /* No page in the page cache at all */
1513 do_sync_mmap_readahead(vma, ra, file, offset);
1514 count_vm_event(PGMAJFAULT);
1515 ret = VM_FAULT_MAJOR;
1516retry_find:
1517 page = find_lock_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 if (!page)
1519 goto no_cached_page;
1520 }
1521
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001523 * We have a locked page in the page cache, now we need to check
1524 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 */
Nick Piggind00806b2007-07-19 01:46:57 -07001526 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 goto page_not_uptodate;
1528
Linus Torvaldsef00e082009-06-16 15:31:25 -07001529 /*
1530 * Found the page and have a reference on it.
1531 * We must recheck i_size under page lock.
1532 */
Nick Piggind00806b2007-07-19 01:46:57 -07001533 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001534 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001535 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001536 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001537 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001538 }
1539
Linus Torvaldsef00e082009-06-16 15:31:25 -07001540 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001541 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001542 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544no_cached_page:
1545 /*
1546 * We're only likely to ever get here if MADV_RANDOM is in
1547 * effect.
1548 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001549 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
1551 /*
1552 * The page we want has now been added to the page cache.
1553 * In the unlikely event that someone removed it in the
1554 * meantime, we'll just come back here and read it again.
1555 */
1556 if (error >= 0)
1557 goto retry_find;
1558
1559 /*
1560 * An error return from page_cache_read can result if the
1561 * system is low on memory, or a problem occurs while trying
1562 * to schedule I/O.
1563 */
1564 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001565 return VM_FAULT_OOM;
1566 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
1568page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 /*
1570 * Umm, take care of errors if the page isn't up-to-date.
1571 * Try to re-read it _once_. We do this synchronously,
1572 * because there really aren't any performance issues here
1573 * and we need to check for errors.
1574 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001576 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001577 if (!error) {
1578 wait_on_page_locked(page);
1579 if (!PageUptodate(page))
1580 error = -EIO;
1581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001583
1584 if (!error || error == AOP_TRUNCATED_PAGE)
1585 goto retry_find;
1586
1587 /* Things didn't work out. Return zero to tell the mm layer so. */
1588 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001589 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001590}
1591EXPORT_SYMBOL(filemap_fault);
1592
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001593const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001594 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595};
1596
1597/* This is used for a general mmap of a disk file */
1598
1599int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1600{
1601 struct address_space *mapping = file->f_mapping;
1602
1603 if (!mapping->a_ops->readpage)
1604 return -ENOEXEC;
1605 file_accessed(file);
1606 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001607 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 return 0;
1609}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611/*
1612 * This is for filesystems which do not implement ->writepage.
1613 */
1614int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1615{
1616 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1617 return -EINVAL;
1618 return generic_file_mmap(file, vma);
1619}
1620#else
1621int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1622{
1623 return -ENOSYS;
1624}
1625int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1626{
1627 return -ENOSYS;
1628}
1629#endif /* CONFIG_MMU */
1630
1631EXPORT_SYMBOL(generic_file_mmap);
1632EXPORT_SYMBOL(generic_file_readonly_mmap);
1633
Nick Piggin6fe69002007-05-06 14:49:04 -07001634static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001635 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001637 void *data,
1638 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639{
Nick Piggineb2be182007-10-16 01:24:57 -07001640 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 int err;
1642repeat:
1643 page = find_get_page(mapping, index);
1644 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001645 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001646 if (!page)
1647 return ERR_PTR(-ENOMEM);
1648 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1649 if (unlikely(err)) {
1650 page_cache_release(page);
1651 if (err == -EEXIST)
1652 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 return ERR_PTR(err);
1655 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 err = filler(data, page);
1657 if (err < 0) {
1658 page_cache_release(page);
1659 page = ERR_PTR(err);
1660 }
1661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 return page;
1663}
1664
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001665static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001666 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001668 void *data,
1669 gfp_t gfp)
1670
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
1672 struct page *page;
1673 int err;
1674
1675retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001676 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001678 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 if (PageUptodate(page))
1680 goto out;
1681
1682 lock_page(page);
1683 if (!page->mapping) {
1684 unlock_page(page);
1685 page_cache_release(page);
1686 goto retry;
1687 }
1688 if (PageUptodate(page)) {
1689 unlock_page(page);
1690 goto out;
1691 }
1692 err = filler(data, page);
1693 if (err < 0) {
1694 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001695 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 }
David Howellsc855ff32007-05-09 13:42:20 +01001697out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001698 mark_page_accessed(page);
1699 return page;
1700}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001701
1702/**
1703 * read_cache_page_async - read into page cache, fill it if needed
1704 * @mapping: the page's address_space
1705 * @index: the page index
1706 * @filler: function to perform the read
1707 * @data: destination for read data
1708 *
1709 * Same as read_cache_page, but don't wait for page to become unlocked
1710 * after submitting it to the filler.
1711 *
1712 * Read into the page cache. If a page already exists, and PageUptodate() is
1713 * not set, try to fill the page but don't wait for it to become unlocked.
1714 *
1715 * If the page does not get brought uptodate, return -EIO.
1716 */
1717struct page *read_cache_page_async(struct address_space *mapping,
1718 pgoff_t index,
1719 int (*filler)(void *,struct page*),
1720 void *data)
1721{
1722 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1723}
Nick Piggin6fe69002007-05-06 14:49:04 -07001724EXPORT_SYMBOL(read_cache_page_async);
1725
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001726static struct page *wait_on_page_read(struct page *page)
1727{
1728 if (!IS_ERR(page)) {
1729 wait_on_page_locked(page);
1730 if (!PageUptodate(page)) {
1731 page_cache_release(page);
1732 page = ERR_PTR(-EIO);
1733 }
1734 }
1735 return page;
1736}
1737
1738/**
1739 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1740 * @mapping: the page's address_space
1741 * @index: the page index
1742 * @gfp: the page allocator flags to use if allocating
1743 *
1744 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1745 * any new page allocations done using the specified allocation flags. Note
1746 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1747 * expect to do this atomically or anything like that - but you can pass in
1748 * other page requirements.
1749 *
1750 * If the page does not get brought uptodate, return -EIO.
1751 */
1752struct page *read_cache_page_gfp(struct address_space *mapping,
1753 pgoff_t index,
1754 gfp_t gfp)
1755{
1756 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1757
1758 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1759}
1760EXPORT_SYMBOL(read_cache_page_gfp);
1761
Nick Piggin6fe69002007-05-06 14:49:04 -07001762/**
1763 * read_cache_page - read into page cache, fill it if needed
1764 * @mapping: the page's address_space
1765 * @index: the page index
1766 * @filler: function to perform the read
1767 * @data: destination for read data
1768 *
1769 * Read into the page cache. If a page already exists, and PageUptodate() is
1770 * not set, try to fill the page then wait for it to become unlocked.
1771 *
1772 * If the page does not get brought uptodate, return -EIO.
1773 */
1774struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001775 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001776 int (*filler)(void *,struct page*),
1777 void *data)
1778{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001779 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781EXPORT_SYMBOL(read_cache_page);
1782
1783/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 * The logic we want is
1785 *
1786 * if suid or (sgid and xgrp)
1787 * remove privs
1788 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001789int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790{
1791 mode_t mode = dentry->d_inode->i_mode;
1792 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794 /* suid always must be killed */
1795 if (unlikely(mode & S_ISUID))
1796 kill = ATTR_KILL_SUID;
1797
1798 /*
1799 * sgid without any exec bits is just a mandatory locking mark; leave
1800 * it alone. If some exec bits are set, it's a real sgid; kill it.
1801 */
1802 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1803 kill |= ATTR_KILL_SGID;
1804
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001805 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001806 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Jens Axboe01de85e2006-10-17 19:50:36 +02001808 return 0;
1809}
Mark Fashehd23a1472006-10-17 17:05:18 -07001810EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001811
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001812static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001813{
1814 struct iattr newattrs;
1815
1816 newattrs.ia_valid = ATTR_FORCE | kill;
1817 return notify_change(dentry, &newattrs);
1818}
1819
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001820int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001821{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001822 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001823 int killsuid = should_remove_suid(dentry);
1824 int killpriv = security_inode_need_killpriv(dentry);
1825 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001826
Serge E. Hallynb5376772007-10-16 23:31:36 -07001827 if (killpriv < 0)
1828 return killpriv;
1829 if (killpriv)
1830 error = security_inode_killpriv(dentry);
1831 if (!error && killsuid)
1832 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001833
Serge E. Hallynb5376772007-10-16 23:31:36 -07001834 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001836EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
Nick Piggin2f718ff2007-10-16 01:24:59 -07001838static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 const struct iovec *iov, size_t base, size_t bytes)
1840{
Ingo Molnarf1800532009-03-02 11:00:57 +01001841 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
1843 while (bytes) {
1844 char __user *buf = iov->iov_base + base;
1845 int copy = min(bytes, iov->iov_len - base);
1846
1847 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001848 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 copied += copy;
1850 bytes -= copy;
1851 vaddr += copy;
1852 iov++;
1853
NeilBrown01408c42006-06-25 05:47:58 -07001854 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 }
1857 return copied - left;
1858}
1859
1860/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001861 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001862 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001863 * bytes which were copied.
1864 */
1865size_t iov_iter_copy_from_user_atomic(struct page *page,
1866 struct iov_iter *i, unsigned long offset, size_t bytes)
1867{
1868 char *kaddr;
1869 size_t copied;
1870
1871 BUG_ON(!in_atomic());
1872 kaddr = kmap_atomic(page, KM_USER0);
1873 if (likely(i->nr_segs == 1)) {
1874 int left;
1875 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001876 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001877 copied = bytes - left;
1878 } else {
1879 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1880 i->iov, i->iov_offset, bytes);
1881 }
1882 kunmap_atomic(kaddr, KM_USER0);
1883
1884 return copied;
1885}
Nick Piggin89e10782007-10-16 01:25:07 -07001886EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001887
1888/*
1889 * This has the same sideeffects and return value as
1890 * iov_iter_copy_from_user_atomic().
1891 * The difference is that it attempts to resolve faults.
1892 * Page must not be locked.
1893 */
1894size_t iov_iter_copy_from_user(struct page *page,
1895 struct iov_iter *i, unsigned long offset, size_t bytes)
1896{
1897 char *kaddr;
1898 size_t copied;
1899
1900 kaddr = kmap(page);
1901 if (likely(i->nr_segs == 1)) {
1902 int left;
1903 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001904 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001905 copied = bytes - left;
1906 } else {
1907 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1908 i->iov, i->iov_offset, bytes);
1909 }
1910 kunmap(page);
1911 return copied;
1912}
Nick Piggin89e10782007-10-16 01:25:07 -07001913EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001914
Nick Pigginf7009262008-03-10 11:43:59 -07001915void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001916{
Nick Pigginf7009262008-03-10 11:43:59 -07001917 BUG_ON(i->count < bytes);
1918
Nick Piggin2f718ff2007-10-16 01:24:59 -07001919 if (likely(i->nr_segs == 1)) {
1920 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001921 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001922 } else {
1923 const struct iovec *iov = i->iov;
1924 size_t base = i->iov_offset;
1925
Nick Piggin124d3b72008-02-02 15:01:17 +01001926 /*
1927 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001928 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001929 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001930 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001931 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001932
Nick Pigginf7009262008-03-10 11:43:59 -07001933 copy = min(bytes, iov->iov_len - base);
1934 BUG_ON(!i->count || i->count < copy);
1935 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001936 bytes -= copy;
1937 base += copy;
1938 if (iov->iov_len == base) {
1939 iov++;
1940 base = 0;
1941 }
1942 }
1943 i->iov = iov;
1944 i->iov_offset = base;
1945 }
1946}
Nick Piggin89e10782007-10-16 01:25:07 -07001947EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001948
Nick Pigginafddba42007-10-16 01:25:01 -07001949/*
1950 * Fault in the first iovec of the given iov_iter, to a maximum length
1951 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1952 * accessed (ie. because it is an invalid address).
1953 *
1954 * writev-intensive code may want this to prefault several iovecs -- that
1955 * would be possible (callers must not rely on the fact that _only_ the
1956 * first iovec will be faulted with the current implementation).
1957 */
1958int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001959{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001960 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001961 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1962 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001963}
Nick Piggin89e10782007-10-16 01:25:07 -07001964EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001965
1966/*
1967 * Return the count of just the current iov_iter segment.
1968 */
1969size_t iov_iter_single_seg_count(struct iov_iter *i)
1970{
1971 const struct iovec *iov = i->iov;
1972 if (i->nr_segs == 1)
1973 return i->count;
1974 else
1975 return min(i->count, iov->iov_len - i->iov_offset);
1976}
Nick Piggin89e10782007-10-16 01:25:07 -07001977EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001978
1979/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 * Performs necessary checks before doing a write
1981 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001982 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 * Returns appropriate error code that caller should return or
1984 * zero in case that write should be allowed.
1985 */
1986inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1987{
1988 struct inode *inode = file->f_mapping->host;
1989 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1990
1991 if (unlikely(*pos < 0))
1992 return -EINVAL;
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 if (!isblk) {
1995 /* FIXME: this is for backwards compatibility with 2.4 */
1996 if (file->f_flags & O_APPEND)
1997 *pos = i_size_read(inode);
1998
1999 if (limit != RLIM_INFINITY) {
2000 if (*pos >= limit) {
2001 send_sig(SIGXFSZ, current, 0);
2002 return -EFBIG;
2003 }
2004 if (*count > limit - (typeof(limit))*pos) {
2005 *count = limit - (typeof(limit))*pos;
2006 }
2007 }
2008 }
2009
2010 /*
2011 * LFS rule
2012 */
2013 if (unlikely(*pos + *count > MAX_NON_LFS &&
2014 !(file->f_flags & O_LARGEFILE))) {
2015 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 return -EFBIG;
2017 }
2018 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2019 *count = MAX_NON_LFS - (unsigned long)*pos;
2020 }
2021 }
2022
2023 /*
2024 * Are we about to exceed the fs block limit ?
2025 *
2026 * If we have written data it becomes a short write. If we have
2027 * exceeded without writing data we send a signal and return EFBIG.
2028 * Linus frestrict idea will clean these up nicely..
2029 */
2030 if (likely(!isblk)) {
2031 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2032 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 return -EFBIG;
2034 }
2035 /* zero-length writes at ->s_maxbytes are OK */
2036 }
2037
2038 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2039 *count = inode->i_sb->s_maxbytes - *pos;
2040 } else {
David Howells93614012006-09-30 20:45:40 +02002041#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 loff_t isize;
2043 if (bdev_read_only(I_BDEV(inode)))
2044 return -EPERM;
2045 isize = i_size_read(inode);
2046 if (*pos >= isize) {
2047 if (*count || *pos > isize)
2048 return -ENOSPC;
2049 }
2050
2051 if (*pos + *count > isize)
2052 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002053#else
2054 return -EPERM;
2055#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 }
2057 return 0;
2058}
2059EXPORT_SYMBOL(generic_write_checks);
2060
Nick Pigginafddba42007-10-16 01:25:01 -07002061int pagecache_write_begin(struct file *file, struct address_space *mapping,
2062 loff_t pos, unsigned len, unsigned flags,
2063 struct page **pagep, void **fsdata)
2064{
2065 const struct address_space_operations *aops = mapping->a_ops;
2066
Nick Piggin4e02ed42008-10-29 14:00:55 -07002067 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002068 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002069}
2070EXPORT_SYMBOL(pagecache_write_begin);
2071
2072int pagecache_write_end(struct file *file, struct address_space *mapping,
2073 loff_t pos, unsigned len, unsigned copied,
2074 struct page *page, void *fsdata)
2075{
2076 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002077
Nick Piggin4e02ed42008-10-29 14:00:55 -07002078 mark_page_accessed(page);
2079 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002080}
2081EXPORT_SYMBOL(pagecache_write_end);
2082
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083ssize_t
2084generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2085 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2086 size_t count, size_t ocount)
2087{
2088 struct file *file = iocb->ki_filp;
2089 struct address_space *mapping = file->f_mapping;
2090 struct inode *inode = mapping->host;
2091 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002092 size_t write_len;
2093 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
2095 if (count != ocount)
2096 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2097
Christoph Hellwiga969e902008-07-23 21:27:04 -07002098 write_len = iov_length(iov, *nr_segs);
2099 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002100
Nick Piggin48b47c52009-01-06 14:40:22 -08002101 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002102 if (written)
2103 goto out;
2104
2105 /*
2106 * After a write we want buffered reads to be sure to go to disk to get
2107 * the new data. We invalidate clean cached page from the region we're
2108 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002109 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002110 */
2111 if (mapping->nrpages) {
2112 written = invalidate_inode_pages2_range(mapping,
2113 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002114 /*
2115 * If a page can not be invalidated, return 0 to fall back
2116 * to buffered write.
2117 */
2118 if (written) {
2119 if (written == -EBUSY)
2120 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002121 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002122 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002123 }
2124
2125 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2126
2127 /*
2128 * Finally, try again to invalidate clean pages which might have been
2129 * cached by non-direct readahead, or faulted in by get_user_pages()
2130 * if the source of the write was an mmap'ed region of the file
2131 * we're writing. Either one is a pretty crazy thing to do,
2132 * so we don't support it 100%. If this invalidation
2133 * fails, tough, the write still worked...
2134 */
2135 if (mapping->nrpages) {
2136 invalidate_inode_pages2_range(mapping,
2137 pos >> PAGE_CACHE_SHIFT, end);
2138 }
2139
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 if (written > 0) {
2141 loff_t end = pos + written;
2142 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2143 i_size_write(inode, end);
2144 mark_inode_dirty(inode);
2145 }
2146 *ppos = end;
2147 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002148out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 return written;
2150}
2151EXPORT_SYMBOL(generic_file_direct_write);
2152
Nick Piggineb2be182007-10-16 01:24:57 -07002153/*
2154 * Find or create a page at the given pagecache position. Return the locked
2155 * page. This function is specifically for buffered writes.
2156 */
Nick Piggin54566b22009-01-04 12:00:53 -08002157struct page *grab_cache_page_write_begin(struct address_space *mapping,
2158 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002159{
2160 int status;
2161 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002162 gfp_t gfp_notmask = 0;
2163 if (flags & AOP_FLAG_NOFS)
2164 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002165repeat:
2166 page = find_lock_page(mapping, index);
2167 if (likely(page))
2168 return page;
2169
Nick Piggin54566b22009-01-04 12:00:53 -08002170 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002171 if (!page)
2172 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002173 status = add_to_page_cache_lru(page, mapping, index,
2174 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002175 if (unlikely(status)) {
2176 page_cache_release(page);
2177 if (status == -EEXIST)
2178 goto repeat;
2179 return NULL;
2180 }
2181 return page;
2182}
Nick Piggin54566b22009-01-04 12:00:53 -08002183EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002184
Nick Pigginafddba42007-10-16 01:25:01 -07002185static ssize_t generic_perform_write(struct file *file,
2186 struct iov_iter *i, loff_t pos)
2187{
2188 struct address_space *mapping = file->f_mapping;
2189 const struct address_space_operations *a_ops = mapping->a_ops;
2190 long status = 0;
2191 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002192 unsigned int flags = 0;
2193
2194 /*
2195 * Copies from kernel address space cannot fail (NFSD is a big user).
2196 */
2197 if (segment_eq(get_fs(), KERNEL_DS))
2198 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002199
2200 do {
2201 struct page *page;
2202 pgoff_t index; /* Pagecache index for current page */
2203 unsigned long offset; /* Offset into pagecache page */
2204 unsigned long bytes; /* Bytes to write to page */
2205 size_t copied; /* Bytes copied from user */
2206 void *fsdata;
2207
2208 offset = (pos & (PAGE_CACHE_SIZE - 1));
2209 index = pos >> PAGE_CACHE_SHIFT;
2210 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2211 iov_iter_count(i));
2212
2213again:
2214
2215 /*
2216 * Bring in the user page that we will copy from _first_.
2217 * Otherwise there's a nasty deadlock on copying from the
2218 * same page as we're writing to, without it being marked
2219 * up-to-date.
2220 *
2221 * Not only is this an optimisation, but it is also required
2222 * to check that the address is actually valid, when atomic
2223 * usercopies are used, below.
2224 */
2225 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2226 status = -EFAULT;
2227 break;
2228 }
2229
Nick Piggin674b8922007-10-16 01:25:03 -07002230 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002231 &page, &fsdata);
2232 if (unlikely(status))
2233 break;
2234
anfei zhou931e80e2010-02-02 13:44:02 -08002235 if (mapping_writably_mapped(mapping))
2236 flush_dcache_page(page);
2237
Nick Pigginafddba42007-10-16 01:25:01 -07002238 pagefault_disable();
2239 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2240 pagefault_enable();
2241 flush_dcache_page(page);
2242
Josef Bacikc8236db2009-07-05 12:08:18 -07002243 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002244 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2245 page, fsdata);
2246 if (unlikely(status < 0))
2247 break;
2248 copied = status;
2249
2250 cond_resched();
2251
Nick Piggin124d3b72008-02-02 15:01:17 +01002252 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002253 if (unlikely(copied == 0)) {
2254 /*
2255 * If we were unable to copy any data at all, we must
2256 * fall back to a single segment length write.
2257 *
2258 * If we didn't fallback here, we could livelock
2259 * because not all segments in the iov can be copied at
2260 * once without a pagefault.
2261 */
2262 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2263 iov_iter_single_seg_count(i));
2264 goto again;
2265 }
Nick Pigginafddba42007-10-16 01:25:01 -07002266 pos += copied;
2267 written += copied;
2268
2269 balance_dirty_pages_ratelimited(mapping);
2270
2271 } while (iov_iter_count(i));
2272
2273 return written ? written : status;
2274}
2275
2276ssize_t
2277generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2278 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2279 size_t count, ssize_t written)
2280{
2281 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002282 ssize_t status;
2283 struct iov_iter i;
2284
2285 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002286 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002287
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002289 written += status;
2290 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 }
2292
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293 return written ? written : status;
2294}
2295EXPORT_SYMBOL(generic_file_buffered_write);
2296
Jan Karae4dd9de2009-08-17 18:10:06 +02002297/**
2298 * __generic_file_aio_write - write data to a file
2299 * @iocb: IO state structure (file, offset, etc.)
2300 * @iov: vector with data to write
2301 * @nr_segs: number of segments in the vector
2302 * @ppos: position where to write
2303 *
2304 * This function does all the work needed for actually writing data to a
2305 * file. It does all basic checks, removes SUID from the file, updates
2306 * modification times and calls proper subroutines depending on whether we
2307 * do direct IO or a standard buffered write.
2308 *
2309 * It expects i_mutex to be grabbed unless we work on a block device or similar
2310 * object which does not need locking at all.
2311 *
2312 * This function does *not* take care of syncing data in case of O_SYNC write.
2313 * A caller has to handle it. This is mainly due to the fact that we want to
2314 * avoid syncing under i_mutex.
2315 */
2316ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2317 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318{
2319 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002320 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 size_t ocount; /* original count */
2322 size_t count; /* after file limit checks */
2323 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 loff_t pos;
2325 ssize_t written;
2326 ssize_t err;
2327
2328 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002329 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2330 if (err)
2331 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332
2333 count = ocount;
2334 pos = *ppos;
2335
2336 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2337
2338 /* We can write back this queue in page reclaim */
2339 current->backing_dev_info = mapping->backing_dev_info;
2340 written = 0;
2341
2342 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2343 if (err)
2344 goto out;
2345
2346 if (count == 0)
2347 goto out;
2348
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002349 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 if (err)
2351 goto out;
2352
Christoph Hellwig870f4812006-01-09 20:52:01 -08002353 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354
2355 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2356 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002357 loff_t endbyte;
2358 ssize_t written_buffered;
2359
2360 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2361 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 if (written < 0 || written == count)
2363 goto out;
2364 /*
2365 * direct-io write to a hole: fall through to buffered I/O
2366 * for completing the rest of the request.
2367 */
2368 pos += written;
2369 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002370 written_buffered = generic_file_buffered_write(iocb, iov,
2371 nr_segs, pos, ppos, count,
2372 written);
2373 /*
2374 * If generic_file_buffered_write() retuned a synchronous error
2375 * then we want to return the number of bytes which were
2376 * direct-written, or the error code if that was zero. Note
2377 * that this differs from normal direct-io semantics, which
2378 * will return -EFOO even if some bytes were written.
2379 */
2380 if (written_buffered < 0) {
2381 err = written_buffered;
2382 goto out;
2383 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002385 /*
2386 * We need to ensure that the page cache pages are written to
2387 * disk and invalidated to preserve the expected O_DIRECT
2388 * semantics.
2389 */
2390 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002391 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002392 if (err == 0) {
2393 written = written_buffered;
2394 invalidate_mapping_pages(mapping,
2395 pos >> PAGE_CACHE_SHIFT,
2396 endbyte >> PAGE_CACHE_SHIFT);
2397 } else {
2398 /*
2399 * We don't know how much we wrote, so just return
2400 * the number of bytes which were direct-written
2401 */
2402 }
2403 } else {
2404 written = generic_file_buffered_write(iocb, iov, nr_segs,
2405 pos, ppos, count, written);
2406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407out:
2408 current->backing_dev_info = NULL;
2409 return written ? written : err;
2410}
Jan Karae4dd9de2009-08-17 18:10:06 +02002411EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412
Jan Karae4dd9de2009-08-17 18:10:06 +02002413/**
2414 * generic_file_aio_write - write data to a file
2415 * @iocb: IO state structure
2416 * @iov: vector with data to write
2417 * @nr_segs: number of segments in the vector
2418 * @pos: position in file where to write
2419 *
2420 * This is a wrapper around __generic_file_aio_write() to be used by most
2421 * filesystems. It takes care of syncing the file in case of O_SYNC file
2422 * and acquires i_mutex as needed.
2423 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002424ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2425 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426{
2427 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002428 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430
2431 BUG_ON(iocb->ki_pos != pos);
2432
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002433 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002434 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002435 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436
Jan Kara148f9482009-08-17 19:52:36 +02002437 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438 ssize_t err;
2439
Jan Kara148f9482009-08-17 19:52:36 +02002440 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002441 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 ret = err;
2443 }
2444 return ret;
2445}
2446EXPORT_SYMBOL(generic_file_aio_write);
2447
David Howellscf9a2ae2006-08-29 19:05:54 +01002448/**
2449 * try_to_release_page() - release old fs-specific metadata on a page
2450 *
2451 * @page: the page which the kernel is trying to free
2452 * @gfp_mask: memory allocation flags (and I/O mode)
2453 *
2454 * The address_space is to try to release any data against the page
2455 * (presumably at page->private). If the release was successful, return `1'.
2456 * Otherwise return zero.
2457 *
David Howells266cf652009-04-03 16:42:36 +01002458 * This may also be called if PG_fscache is set on a page, indicating that the
2459 * page is known to the local caching routines.
2460 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002461 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002462 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002463 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002464 */
2465int try_to_release_page(struct page *page, gfp_t gfp_mask)
2466{
2467 struct address_space * const mapping = page->mapping;
2468
2469 BUG_ON(!PageLocked(page));
2470 if (PageWriteback(page))
2471 return 0;
2472
2473 if (mapping && mapping->a_ops->releasepage)
2474 return mapping->a_ops->releasepage(page, gfp_mask);
2475 return try_to_free_buffers(page);
2476}
2477
2478EXPORT_SYMBOL(try_to_release_page);