blob: e1b23fda48de62d134d3143006a64d2525816b60 [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 */
42#include <linux/buffer_head.h> /* for generic_osync_inode */
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
Adrian Bunk5ce78522005-09-10 00:26:28 -070046
Linus Torvalds1da177e2005-04-16 15:20:36 -070047/*
48 * Shared mappings implemented 30.11.1994. It's not fully working yet,
49 * though.
50 *
51 * Shared mappings now work. 15.8.1995 Bruno.
52 *
53 * finished 'unifying' the page and buffer cache and SMP-threaded the
54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
55 *
56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
57 */
58
59/*
60 * Lock ordering:
61 *
62 * ->i_mmap_lock (vmtruncate)
63 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070064 * ->swap_lock (exclusive_swap_page, others)
65 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080067 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * ->i_mmap_lock (truncate->unmap_mapping_range)
69 *
70 * ->mmap_sem
71 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070072 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
74 *
75 * ->mmap_sem
76 * ->lock_page (access_process_vm)
77 *
Nick Piggin82591e62006-10-19 23:29:10 -070078 * ->i_mutex (generic_file_buffered_write)
79 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080081 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * ->i_alloc_sem (various)
83 *
84 * ->inode_lock
85 * ->sb_lock (fs/fs-writeback.c)
86 * ->mapping->tree_lock (__sync_single_inode)
87 *
88 * ->i_mmap_lock
89 * ->anon_vma.lock (vma_adjust)
90 *
91 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070092 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070094 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070095 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * ->private_lock (try_to_unmap_one)
97 * ->tree_lock (try_to_unmap_one)
98 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080099 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 * ->private_lock (page_remove_rmap->set_page_dirty)
101 * ->tree_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (page_remove_rmap->set_page_dirty)
103 * ->inode_lock (zap_pte_range->set_page_dirty)
104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
105 *
106 * ->task->proc_lock
107 * ->dcache_lock (proc_pid_lookup)
108 */
109
110/*
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115void __remove_from_page_cache(struct page *page)
116{
117 struct address_space *mapping = page->mapping;
118
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700119 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 radix_tree_delete(&mapping->page_tree, page->index);
121 page->mapping = NULL;
122 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700123 __dec_zone_page_state(page, NR_FILE_PAGES);
Nick Piggin45426812007-07-15 23:38:12 -0700124 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800125
126 /*
127 * Some filesystems seem to re-dirty the page even after
128 * the VM has canceled the dirty bit (eg ext3 journaling).
129 *
130 * Fix it up by doing a final dirty accounting check after
131 * having removed the page entirely.
132 */
133 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
134 dec_zone_page_state(page, NR_FILE_DIRTY);
135 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137}
138
139void remove_from_page_cache(struct page *page)
140{
141 struct address_space *mapping = page->mapping;
142
Matt Mackallcd7619d2005-05-01 08:59:01 -0700143 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Nick Piggin19fd6232008-07-25 19:45:32 -0700145 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700147 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150static int sync_page(void *word)
151{
152 struct address_space *mapping;
153 struct page *page;
154
Andi Kleen07808b72005-11-05 17:25:53 +0100155 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700158 * page_mapping() is being called without PG_locked held.
159 * Some knowledge of the state and use of the page is used to
160 * reduce the requirements down to a memory barrier.
161 * The danger here is of a stale page_mapping() return value
162 * indicating a struct address_space different from the one it's
163 * associated with when it is associated with one.
164 * After smp_mb(), it's either the correct page_mapping() for
165 * the page, or an old page_mapping() and the page's own
166 * page_mapping() has gone NULL.
167 * The ->sync_page() address_space operation must tolerate
168 * page_mapping() going NULL. By an amazing coincidence,
169 * this comes about because none of the users of the page
170 * in the ->sync_page() methods make essential use of the
171 * page_mapping(), merely passing the page down to the backing
172 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700173 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700174 * of interest. When page_mapping() does go NULL, the entire
175 * call stack gracefully ignores the page and returns.
176 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 */
178 smp_mb();
179 mapping = page_mapping(page);
180 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
181 mapping->a_ops->sync_page(page);
182 io_schedule();
183 return 0;
184}
185
Matthew Wilcox2687a352007-12-06 11:18:49 -0500186static int sync_page_killable(void *word)
187{
188 sync_page(word);
189 return fatal_signal_pending(current) ? -EINTR : 0;
190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700194 * @mapping: address space structure to write
195 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800196 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700197 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700199 * Start writeback against all of a mapping's dirty pages that lie
200 * within the byte offsets <start, end> inclusive.
201 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700203 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 * these two operations is that if a dirty page/buffer is encountered, it must
205 * be waited upon, and not just skipped over.
206 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800207int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
208 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209{
210 int ret;
211 struct writeback_control wbc = {
212 .sync_mode = sync_mode,
213 .nr_to_write = mapping->nrpages * 2,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700214 .range_start = start,
215 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 };
217
218 if (!mapping_cap_writeback_dirty(mapping))
219 return 0;
220
221 ret = do_writepages(mapping, &wbc);
222 return ret;
223}
224
225static inline int __filemap_fdatawrite(struct address_space *mapping,
226 int sync_mode)
227{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229}
230
231int filemap_fdatawrite(struct address_space *mapping)
232{
233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
234}
235EXPORT_SYMBOL(filemap_fdatawrite);
236
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400237int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800238 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239{
240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
241}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400242EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Randy Dunlap485bb992006-06-23 02:03:49 -0700244/**
245 * filemap_flush - mostly a non-blocking flush
246 * @mapping: target address_space
247 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 * This is a mostly non-blocking flush. Not suitable for data-integrity
249 * purposes - I/O may not be started against all dirty pages.
250 */
251int filemap_flush(struct address_space *mapping)
252{
253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
254}
255EXPORT_SYMBOL(filemap_flush);
256
Randy Dunlap485bb992006-06-23 02:03:49 -0700257/**
258 * wait_on_page_writeback_range - wait for writeback to complete
259 * @mapping: target address_space
260 * @start: beginning page index
261 * @end: ending page index
262 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 * Wait for writeback to complete against pages indexed by start->end
264 * inclusive
265 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800266int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 pgoff_t start, pgoff_t end)
268{
269 struct pagevec pvec;
270 int nr_pages;
271 int ret = 0;
272 pgoff_t index;
273
274 if (end < start)
275 return 0;
276
277 pagevec_init(&pvec, 0);
278 index = start;
279 while ((index <= end) &&
280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
281 PAGECACHE_TAG_WRITEBACK,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
283 unsigned i;
284
285 for (i = 0; i < nr_pages; i++) {
286 struct page *page = pvec.pages[i];
287
288 /* until radix tree lookup accepts end_index */
289 if (page->index > end)
290 continue;
291
292 wait_on_page_writeback(page);
293 if (PageError(page))
294 ret = -EIO;
295 }
296 pagevec_release(&pvec);
297 cond_resched();
298 }
299
300 /* Check for outstanding write errors */
301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
302 ret = -ENOSPC;
303 if (test_and_clear_bit(AS_EIO, &mapping->flags))
304 ret = -EIO;
305
306 return ret;
307}
308
Randy Dunlap485bb992006-06-23 02:03:49 -0700309/**
310 * sync_page_range - write and wait on all pages in the passed range
311 * @inode: target inode
312 * @mapping: target address_space
313 * @pos: beginning offset in pages to write
314 * @count: number of bytes to write
315 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 * Write and wait upon all the pages in the passed range. This is a "data
317 * integrity" operation. It waits upon in-flight writeout before starting and
318 * waiting upon new writeout. If there was an IO error, return it.
319 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800320 * We need to re-take i_mutex during the generic_osync_inode list walk because
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 * it is otherwise livelockable.
322 */
323int sync_page_range(struct inode *inode, struct address_space *mapping,
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800324 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
326 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
327 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
328 int ret;
329
330 if (!mapping_cap_writeback_dirty(mapping) || !count)
331 return 0;
332 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
333 if (ret == 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800334 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800336 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 }
338 if (ret == 0)
339 ret = wait_on_page_writeback_range(mapping, start, end);
340 return ret;
341}
342EXPORT_SYMBOL(sync_page_range);
343
Randy Dunlap485bb992006-06-23 02:03:49 -0700344/**
Randy Dunlap76824862008-03-19 17:00:40 -0700345 * sync_page_range_nolock - write & wait on all pages in the passed range without locking
Randy Dunlap485bb992006-06-23 02:03:49 -0700346 * @inode: target inode
347 * @mapping: target address_space
348 * @pos: beginning offset in pages to write
349 * @count: number of bytes to write
350 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800351 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 * as it forces O_SYNC writers to different parts of the same file
353 * to be serialised right until io completion.
354 */
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800355int sync_page_range_nolock(struct inode *inode, struct address_space *mapping,
356 loff_t pos, loff_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
358 pgoff_t start = pos >> PAGE_CACHE_SHIFT;
359 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT;
360 int ret;
361
362 if (!mapping_cap_writeback_dirty(mapping) || !count)
363 return 0;
364 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1);
365 if (ret == 0)
366 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA);
367 if (ret == 0)
368 ret = wait_on_page_writeback_range(mapping, start, end);
369 return ret;
370}
OGAWA Hirofumi268fc162006-01-08 01:02:12 -0800371EXPORT_SYMBOL(sync_page_range_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
373/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700374 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700376 *
377 * Walk the list of under-writeback pages of the given address space
378 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 */
380int filemap_fdatawait(struct address_space *mapping)
381{
382 loff_t i_size = i_size_read(mapping->host);
383
384 if (i_size == 0)
385 return 0;
386
387 return wait_on_page_writeback_range(mapping, 0,
388 (i_size - 1) >> PAGE_CACHE_SHIFT);
389}
390EXPORT_SYMBOL(filemap_fdatawait);
391
392int filemap_write_and_wait(struct address_space *mapping)
393{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800394 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800397 err = filemap_fdatawrite(mapping);
398 /*
399 * Even if the above returned error, the pages may be
400 * written partially (e.g. -ENOSPC), so we wait for it.
401 * But the -EIO is special case, it may indicate the worst
402 * thing (e.g. bug) happened, so we avoid waiting for it.
403 */
404 if (err != -EIO) {
405 int err2 = filemap_fdatawait(mapping);
406 if (!err)
407 err = err2;
408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800410 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800412EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Randy Dunlap485bb992006-06-23 02:03:49 -0700414/**
415 * filemap_write_and_wait_range - write out & wait on a file range
416 * @mapping: the address_space for the pages
417 * @lstart: offset in bytes where the range starts
418 * @lend: offset in bytes where the range ends (inclusive)
419 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800420 * Write out and wait upon file offsets lstart->lend, inclusive.
421 *
422 * Note that `lend' is inclusive (describes the last byte to be written) so
423 * that this function can be used to write to the very end-of-file (end = -1).
424 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425int filemap_write_and_wait_range(struct address_space *mapping,
426 loff_t lstart, loff_t lend)
427{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800428 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800431 err = __filemap_fdatawrite_range(mapping, lstart, lend,
432 WB_SYNC_ALL);
433 /* See comment of filemap_write_and_wait() */
434 if (err != -EIO) {
435 int err2 = wait_on_page_writeback_range(mapping,
436 lstart >> PAGE_CACHE_SHIFT,
437 lend >> PAGE_CACHE_SHIFT);
438 if (!err)
439 err = err2;
440 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800442 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
Randy Dunlap485bb992006-06-23 02:03:49 -0700445/**
Nick Piggine2867812008-07-25 19:45:30 -0700446 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700447 * @page: page to add
448 * @mapping: the page's address_space
449 * @offset: page index
450 * @gfp_mask: page allocation mode
451 *
Nick Piggine2867812008-07-25 19:45:30 -0700452 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * This function does not add the page to the LRU. The caller must do that.
454 */
Nick Piggine2867812008-07-25 19:45:30 -0700455int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400456 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Nick Piggine2867812008-07-25 19:45:30 -0700458 int error;
459
460 VM_BUG_ON(!PageLocked(page));
461
462 error = mem_cgroup_cache_charge(page, current->mm,
Badari Pulavarty4c6bc8d2008-02-07 00:14:05 -0800463 gfp_mask & ~__GFP_HIGHMEM);
Balbir Singh35c754d2008-02-07 00:14:05 -0800464 if (error)
465 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466
Balbir Singh35c754d2008-02-07 00:14:05 -0800467 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700469 page_cache_get(page);
470 page->mapping = mapping;
471 page->index = offset;
472
Nick Piggin19fd6232008-07-25 19:45:32 -0700473 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700475 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700477 __inc_zone_page_state(page, NR_FILE_PAGES);
Nick Piggine2867812008-07-25 19:45:30 -0700478 } else {
479 page->mapping = NULL;
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700480 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700481 page_cache_release(page);
482 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800483
Nick Piggin19fd6232008-07-25 19:45:32 -0700484 spin_unlock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800486 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700487 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800488out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 return error;
490}
Nick Piggine2867812008-07-25 19:45:30 -0700491EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
493int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400494 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700496 int ret;
497
498 /*
499 * Splice_read and readahead add shmem/tmpfs pages into the page cache
500 * before shmem_readpage has a chance to mark them as SwapBacked: they
501 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
502 * (called in add_to_page_cache) needs to know where they're going too.
503 */
504 if (mapping_cap_swap_backed(mapping))
505 SetPageSwapBacked(page);
506
507 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
508 if (ret == 0) {
509 if (page_is_file_cache(page))
510 lru_cache_add_file(page);
511 else
512 lru_cache_add_active_anon(page);
513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return ret;
515}
516
Paul Jackson44110fe2006-03-24 03:16:04 -0800517#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700518struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800519{
520 if (cpuset_do_page_mem_spread()) {
521 int n = cpuset_mem_spread_node();
Nick Piggin2ae88142006-10-28 10:38:23 -0700522 return alloc_pages_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800523 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700524 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800525}
Nick Piggin2ae88142006-10-28 10:38:23 -0700526EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800527#endif
528
Nick Piggindb376482006-09-25 23:31:24 -0700529static int __sleep_on_page_lock(void *word)
530{
531 io_schedule();
532 return 0;
533}
534
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535/*
536 * In order to wait for pages to become available there must be
537 * waitqueues associated with pages. By using a hash table of
538 * waitqueues where the bucket discipline is to maintain all
539 * waiters on the same queue and wake all when any of the pages
540 * become available, and for the woken contexts to check to be
541 * sure the appropriate page became available, this saves space
542 * at a cost of "thundering herd" phenomena during rare hash
543 * collisions.
544 */
545static wait_queue_head_t *page_waitqueue(struct page *page)
546{
547 const struct zone *zone = page_zone(page);
548
549 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
550}
551
552static inline void wake_up_page(struct page *page, int bit)
553{
554 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
555}
556
Harvey Harrison920c7a52008-02-04 22:29:26 -0800557void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558{
559 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
560
561 if (test_bit(bit_nr, &page->flags))
562 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
563 TASK_UNINTERRUPTIBLE);
564}
565EXPORT_SYMBOL(wait_on_page_bit);
566
567/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700568 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * @page: the page
570 *
571 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
572 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
573 * mechananism between PageLocked pages and PageWriteback pages is shared.
574 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
575 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700576 * The mb is necessary to enforce ordering between the clear_bit and the read
577 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800579void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580{
Nick Piggin8413ac92008-10-18 20:26:59 -0700581 VM_BUG_ON(!PageLocked(page));
582 clear_bit_unlock(PG_locked, &page->flags);
583 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 wake_up_page(page, PG_locked);
585}
586EXPORT_SYMBOL(unlock_page);
587
Randy Dunlap485bb992006-06-23 02:03:49 -0700588/**
589 * end_page_writeback - end writeback against a page
590 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 */
592void end_page_writeback(struct page *page)
593{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700594 if (TestClearPageReclaim(page))
595 rotate_reclaimable_page(page);
596
597 if (!test_clear_page_writeback(page))
598 BUG();
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 smp_mb__after_clear_bit();
601 wake_up_page(page, PG_writeback);
602}
603EXPORT_SYMBOL(end_page_writeback);
604
Randy Dunlap485bb992006-06-23 02:03:49 -0700605/**
606 * __lock_page - get a lock on the page, assuming we need to sleep to get it
607 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700609 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
611 * chances are that on the second loop, the block layer's plug list is empty,
612 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
613 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800614void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
616 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
617
618 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
619 TASK_UNINTERRUPTIBLE);
620}
621EXPORT_SYMBOL(__lock_page);
622
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800623int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500624{
625 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
626
627 return __wait_on_bit_lock(page_waitqueue(page), &wait,
628 sync_page_killable, TASK_KILLABLE);
629}
630
Randy Dunlap76824862008-03-19 17:00:40 -0700631/**
632 * __lock_page_nosync - get a lock on the page, without calling sync_page()
633 * @page: the page to lock
634 *
Nick Piggindb376482006-09-25 23:31:24 -0700635 * Variant of lock_page that does not require the caller to hold a reference
636 * on the page's mapping.
637 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800638void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700639{
640 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
641 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
642 TASK_UNINTERRUPTIBLE);
643}
644
Randy Dunlap485bb992006-06-23 02:03:49 -0700645/**
646 * find_get_page - find and get a page reference
647 * @mapping: the address_space to search
648 * @offset: the page index
649 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700650 * Is there a pagecache struct page at the given (mapping, offset) tuple?
651 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 */
Nick Piggina60637c2008-07-25 19:45:31 -0700653struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654{
Nick Piggina60637c2008-07-25 19:45:31 -0700655 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 struct page *page;
657
Nick Piggina60637c2008-07-25 19:45:31 -0700658 rcu_read_lock();
659repeat:
660 page = NULL;
661 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
662 if (pagep) {
663 page = radix_tree_deref_slot(pagep);
664 if (unlikely(!page || page == RADIX_TREE_RETRY))
665 goto repeat;
666
667 if (!page_cache_get_speculative(page))
668 goto repeat;
669
670 /*
671 * Has the page moved?
672 * This is part of the lockless pagecache protocol. See
673 * include/linux/pagemap.h for details.
674 */
675 if (unlikely(page != *pagep)) {
676 page_cache_release(page);
677 goto repeat;
678 }
679 }
680 rcu_read_unlock();
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 return page;
683}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684EXPORT_SYMBOL(find_get_page);
685
Randy Dunlap485bb992006-06-23 02:03:49 -0700686/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700688 * @mapping: the address_space to search
689 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 *
691 * Locates the desired pagecache page, locks it, increments its reference
692 * count and returns its address.
693 *
694 * Returns zero if the page was not present. find_lock_page() may sleep.
695 */
Nick Piggina60637c2008-07-25 19:45:31 -0700696struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
698 struct page *page;
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700701 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700703 lock_page(page);
704 /* Has the page been truncated? */
705 if (unlikely(page->mapping != mapping)) {
706 unlock_page(page);
707 page_cache_release(page);
708 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 }
Nick Piggina60637c2008-07-25 19:45:31 -0700710 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 return page;
713}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714EXPORT_SYMBOL(find_lock_page);
715
716/**
717 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700718 * @mapping: the page's address_space
719 * @index: the page's index into the mapping
720 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 *
722 * Locates a page in the pagecache. If the page is not present, a new page
723 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
724 * LRU list. The returned page is locked and has its reference count
725 * incremented.
726 *
727 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
728 * allocation!
729 *
730 * find_or_create_page() returns the desired page's address, or zero on
731 * memory exhaustion.
732 */
733struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700734 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Nick Piggineb2be182007-10-16 01:24:57 -0700736 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 int err;
738repeat:
739 page = find_lock_page(mapping, index);
740 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700741 page = __page_cache_alloc(gfp_mask);
742 if (!page)
743 return NULL;
744 err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
745 if (unlikely(err)) {
746 page_cache_release(page);
747 page = NULL;
748 if (err == -EEXIST)
749 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 return page;
753}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754EXPORT_SYMBOL(find_or_create_page);
755
756/**
757 * find_get_pages - gang pagecache lookup
758 * @mapping: The address_space to search
759 * @start: The starting page index
760 * @nr_pages: The maximum number of pages
761 * @pages: Where the resulting pages are placed
762 *
763 * find_get_pages() will search for and return a group of up to
764 * @nr_pages pages in the mapping. The pages are placed at @pages.
765 * find_get_pages() takes a reference against the returned pages.
766 *
767 * The search returns a group of mapping-contiguous pages with ascending
768 * indexes. There may be holes in the indices due to not-present pages.
769 *
770 * find_get_pages() returns the number of pages which were found.
771 */
772unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
773 unsigned int nr_pages, struct page **pages)
774{
775 unsigned int i;
776 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700777 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Nick Piggina60637c2008-07-25 19:45:31 -0700779 rcu_read_lock();
780restart:
781 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
782 (void ***)pages, start, nr_pages);
783 ret = 0;
784 for (i = 0; i < nr_found; i++) {
785 struct page *page;
786repeat:
787 page = radix_tree_deref_slot((void **)pages[i]);
788 if (unlikely(!page))
789 continue;
790 /*
791 * this can only trigger if nr_found == 1, making livelock
792 * a non issue.
793 */
794 if (unlikely(page == RADIX_TREE_RETRY))
795 goto restart;
796
797 if (!page_cache_get_speculative(page))
798 goto repeat;
799
800 /* Has the page moved? */
801 if (unlikely(page != *((void **)pages[i]))) {
802 page_cache_release(page);
803 goto repeat;
804 }
805
806 pages[ret] = page;
807 ret++;
808 }
809 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return ret;
811}
812
Jens Axboeebf43502006-04-27 08:46:01 +0200813/**
814 * find_get_pages_contig - gang contiguous pagecache lookup
815 * @mapping: The address_space to search
816 * @index: The starting page index
817 * @nr_pages: The maximum number of pages
818 * @pages: Where the resulting pages are placed
819 *
820 * find_get_pages_contig() works exactly like find_get_pages(), except
821 * that the returned number of pages are guaranteed to be contiguous.
822 *
823 * find_get_pages_contig() returns the number of pages which were found.
824 */
825unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
826 unsigned int nr_pages, struct page **pages)
827{
828 unsigned int i;
829 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700830 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200831
Nick Piggina60637c2008-07-25 19:45:31 -0700832 rcu_read_lock();
833restart:
834 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
835 (void ***)pages, index, nr_pages);
836 ret = 0;
837 for (i = 0; i < nr_found; i++) {
838 struct page *page;
839repeat:
840 page = radix_tree_deref_slot((void **)pages[i]);
841 if (unlikely(!page))
842 continue;
843 /*
844 * this can only trigger if nr_found == 1, making livelock
845 * a non issue.
846 */
847 if (unlikely(page == RADIX_TREE_RETRY))
848 goto restart;
849
850 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200851 break;
852
Nick Piggina60637c2008-07-25 19:45:31 -0700853 if (!page_cache_get_speculative(page))
854 goto repeat;
855
856 /* Has the page moved? */
857 if (unlikely(page != *((void **)pages[i]))) {
858 page_cache_release(page);
859 goto repeat;
860 }
861
862 pages[ret] = page;
863 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200864 index++;
865 }
Nick Piggina60637c2008-07-25 19:45:31 -0700866 rcu_read_unlock();
867 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200868}
David Howellsef71c152007-05-09 02:33:44 -0700869EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200870
Randy Dunlap485bb992006-06-23 02:03:49 -0700871/**
872 * find_get_pages_tag - find and return pages that match @tag
873 * @mapping: the address_space to search
874 * @index: the starting page index
875 * @tag: the tag index
876 * @nr_pages: the maximum number of pages
877 * @pages: where the resulting pages are placed
878 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700880 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 */
882unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
883 int tag, unsigned int nr_pages, struct page **pages)
884{
885 unsigned int i;
886 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700887 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Nick Piggina60637c2008-07-25 19:45:31 -0700889 rcu_read_lock();
890restart:
891 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
892 (void ***)pages, *index, nr_pages, tag);
893 ret = 0;
894 for (i = 0; i < nr_found; i++) {
895 struct page *page;
896repeat:
897 page = radix_tree_deref_slot((void **)pages[i]);
898 if (unlikely(!page))
899 continue;
900 /*
901 * this can only trigger if nr_found == 1, making livelock
902 * a non issue.
903 */
904 if (unlikely(page == RADIX_TREE_RETRY))
905 goto restart;
906
907 if (!page_cache_get_speculative(page))
908 goto repeat;
909
910 /* Has the page moved? */
911 if (unlikely(page != *((void **)pages[i]))) {
912 page_cache_release(page);
913 goto repeat;
914 }
915
916 pages[ret] = page;
917 ret++;
918 }
919 rcu_read_unlock();
920
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (ret)
922 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700923
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 return ret;
925}
David Howellsef71c152007-05-09 02:33:44 -0700926EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927
Randy Dunlap485bb992006-06-23 02:03:49 -0700928/**
929 * grab_cache_page_nowait - returns locked page at given index in given cache
930 * @mapping: target address_space
931 * @index: the page index
932 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800933 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 * This is intended for speculative data generators, where the data can
935 * be regenerated if the page couldn't be grabbed. This routine should
936 * be safe to call while holding the lock for another page.
937 *
938 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
939 * and deadlock against the caller's locked page.
940 */
941struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700942grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943{
944 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
946 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200947 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 return page;
949 page_cache_release(page);
950 return NULL;
951 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700952 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
953 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 page_cache_release(page);
955 page = NULL;
956 }
957 return page;
958}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959EXPORT_SYMBOL(grab_cache_page_nowait);
960
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700961/*
962 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
963 * a _large_ part of the i/o request. Imagine the worst scenario:
964 *
965 * ---R__________________________________________B__________
966 * ^ reading here ^ bad block(assume 4k)
967 *
968 * read(R) => miss => readahead(R...B) => media error => frustrating retries
969 * => failing the whole request => read(R) => read(R+1) =>
970 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
971 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
972 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
973 *
974 * It is going insane. Fix it by quickly scaling down the readahead size.
975 */
976static void shrink_readahead_size_eio(struct file *filp,
977 struct file_ra_state *ra)
978{
979 if (!ra->ra_pages)
980 return;
981
982 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700983}
984
Randy Dunlap485bb992006-06-23 02:03:49 -0700985/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800986 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700987 * @filp: the file to read
988 * @ppos: current file position
989 * @desc: read_descriptor
990 * @actor: read method
991 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700993 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 *
995 * This is really ugly. But the goto's actually try to clarify some
996 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800998static void do_generic_file_read(struct file *filp, loff_t *ppos,
999 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000{
Christoph Hellwig36e78912008-02-08 04:21:24 -08001001 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -08001003 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001004 pgoff_t index;
1005 pgoff_t last_index;
1006 pgoff_t prev_index;
1007 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -07001008 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001012 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1013 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1015 offset = *ppos & ~PAGE_CACHE_MASK;
1016
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 for (;;) {
1018 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001019 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001020 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 unsigned long nr, ret;
1022
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024find_page:
1025 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001026 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001027 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001028 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001029 index, last_index - index);
1030 page = find_get_page(mapping, index);
1031 if (unlikely(page == NULL))
1032 goto no_cached_page;
1033 }
1034 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001035 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001036 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001037 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001039 if (!PageUptodate(page)) {
1040 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1041 !mapping->a_ops->is_partially_uptodate)
1042 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001043 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001044 goto page_not_up_to_date;
1045 if (!mapping->a_ops->is_partially_uptodate(page,
1046 desc, offset))
1047 goto page_not_up_to_date_locked;
1048 unlock_page(page);
1049 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001051 /*
1052 * i_size must be checked after we know the page is Uptodate.
1053 *
1054 * Checking i_size after the check allows us to calculate
1055 * the correct value for "nr", which means the zero-filled
1056 * part of the page is not copied back to userspace (unless
1057 * another truncate extends the file - this is desired though).
1058 */
1059
1060 isize = i_size_read(inode);
1061 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1062 if (unlikely(!isize || index > end_index)) {
1063 page_cache_release(page);
1064 goto out;
1065 }
1066
1067 /* nr is the maximum number of bytes to copy from this page */
1068 nr = PAGE_CACHE_SIZE;
1069 if (index == end_index) {
1070 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1071 if (nr <= offset) {
1072 page_cache_release(page);
1073 goto out;
1074 }
1075 }
1076 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077
1078 /* If users can be writing to this page using arbitrary
1079 * virtual addresses, take care about potential aliasing
1080 * before reading the page on the kernel side.
1081 */
1082 if (mapping_writably_mapped(mapping))
1083 flush_dcache_page(page);
1084
1085 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001086 * When a sequential read accesses a page several times,
1087 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 */
Jan Karaec0f1632007-05-06 14:49:25 -07001089 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 mark_page_accessed(page);
1091 prev_index = index;
1092
1093 /*
1094 * Ok, we have the page, and it's up-to-date, so
1095 * now we can copy it to user space...
1096 *
1097 * The actor routine returns how many bytes were actually used..
1098 * NOTE! This may not be the same as how much of a user buffer
1099 * we filled up (we may be padding etc), so we can only update
1100 * "pos" here (the actor routine has to update the user buffer
1101 * pointers and the remaining count).
1102 */
1103 ret = actor(desc, page, offset, nr);
1104 offset += ret;
1105 index += offset >> PAGE_CACHE_SHIFT;
1106 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001107 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
1109 page_cache_release(page);
1110 if (ret == nr && desc->count)
1111 continue;
1112 goto out;
1113
1114page_not_up_to_date:
1115 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001116 error = lock_page_killable(page);
1117 if (unlikely(error))
1118 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001120page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001121 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (!page->mapping) {
1123 unlock_page(page);
1124 page_cache_release(page);
1125 continue;
1126 }
1127
1128 /* Did somebody else fill it already? */
1129 if (PageUptodate(page)) {
1130 unlock_page(page);
1131 goto page_ok;
1132 }
1133
1134readpage:
1135 /* Start the actual read. The read will unlock the page. */
1136 error = mapping->a_ops->readpage(filp, page);
1137
Zach Brown994fc28c2005-12-15 14:28:17 -08001138 if (unlikely(error)) {
1139 if (error == AOP_TRUNCATED_PAGE) {
1140 page_cache_release(page);
1141 goto find_page;
1142 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001144 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
1146 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001147 error = lock_page_killable(page);
1148 if (unlikely(error))
1149 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (!PageUptodate(page)) {
1151 if (page->mapping == NULL) {
1152 /*
1153 * invalidate_inode_pages got it
1154 */
1155 unlock_page(page);
1156 page_cache_release(page);
1157 goto find_page;
1158 }
1159 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001160 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001161 error = -EIO;
1162 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 }
1164 unlock_page(page);
1165 }
1166
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 goto page_ok;
1168
1169readpage_error:
1170 /* UHHUH! A synchronous read error occurred. Report it */
1171 desc->error = error;
1172 page_cache_release(page);
1173 goto out;
1174
1175no_cached_page:
1176 /*
1177 * Ok, it wasn't cached, so we need to create a new
1178 * page..
1179 */
Nick Piggineb2be182007-10-16 01:24:57 -07001180 page = page_cache_alloc_cold(mapping);
1181 if (!page) {
1182 desc->error = -ENOMEM;
1183 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 }
Nick Piggineb2be182007-10-16 01:24:57 -07001185 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 index, GFP_KERNEL);
1187 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001188 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 if (error == -EEXIST)
1190 goto find_page;
1191 desc->error = error;
1192 goto out;
1193 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 goto readpage;
1195 }
1196
1197out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001198 ra->prev_pos = prev_index;
1199 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1200 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001202 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001203 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
1206int file_read_actor(read_descriptor_t *desc, struct page *page,
1207 unsigned long offset, unsigned long size)
1208{
1209 char *kaddr;
1210 unsigned long left, count = desc->count;
1211
1212 if (size > count)
1213 size = count;
1214
1215 /*
1216 * Faults on the destination of a read are common, so do it before
1217 * taking the kmap.
1218 */
1219 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1220 kaddr = kmap_atomic(page, KM_USER0);
1221 left = __copy_to_user_inatomic(desc->arg.buf,
1222 kaddr + offset, size);
1223 kunmap_atomic(kaddr, KM_USER0);
1224 if (left == 0)
1225 goto success;
1226 }
1227
1228 /* Do it the slow way */
1229 kaddr = kmap(page);
1230 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1231 kunmap(page);
1232
1233 if (left) {
1234 size -= left;
1235 desc->error = -EFAULT;
1236 }
1237success:
1238 desc->count = count - size;
1239 desc->written += size;
1240 desc->arg.buf += size;
1241 return size;
1242}
1243
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001244/*
1245 * Performs necessary checks before doing a write
1246 * @iov: io vector request
1247 * @nr_segs: number of segments in the iovec
1248 * @count: number of bytes to write
1249 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1250 *
1251 * Adjust number of segments and amount of bytes to write (nr_segs should be
1252 * properly initialized first). Returns appropriate error code that caller
1253 * should return or zero in case that write should be allowed.
1254 */
1255int generic_segment_checks(const struct iovec *iov,
1256 unsigned long *nr_segs, size_t *count, int access_flags)
1257{
1258 unsigned long seg;
1259 size_t cnt = 0;
1260 for (seg = 0; seg < *nr_segs; seg++) {
1261 const struct iovec *iv = &iov[seg];
1262
1263 /*
1264 * If any segment has a negative length, or the cumulative
1265 * length ever wraps negative then return -EINVAL.
1266 */
1267 cnt += iv->iov_len;
1268 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1269 return -EINVAL;
1270 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1271 continue;
1272 if (seg == 0)
1273 return -EFAULT;
1274 *nr_segs = seg;
1275 cnt -= iv->iov_len; /* This segment is no good */
1276 break;
1277 }
1278 *count = cnt;
1279 return 0;
1280}
1281EXPORT_SYMBOL(generic_segment_checks);
1282
Randy Dunlap485bb992006-06-23 02:03:49 -07001283/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001284 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001285 * @iocb: kernel I/O control block
1286 * @iov: io vector request
1287 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001288 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001289 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 * This is the "read()" routine for all filesystems
1291 * that can use the page cache directly.
1292 */
1293ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001294generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1295 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296{
1297 struct file *filp = iocb->ki_filp;
1298 ssize_t retval;
1299 unsigned long seg;
1300 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001301 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
1303 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001304 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1305 if (retval)
1306 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1309 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001310 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 struct address_space *mapping;
1312 struct inode *inode;
1313
1314 mapping = filp->f_mapping;
1315 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (!count)
1317 goto out; /* skip atime */
1318 size = i_size_read(inode);
1319 if (pos < size) {
Christoph Hellwiga969e902008-07-23 21:27:04 -07001320 retval = filemap_write_and_wait(mapping);
1321 if (!retval) {
1322 retval = mapping->a_ops->direct_IO(READ, iocb,
1323 iov, pos, nr_segs);
1324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 if (retval > 0)
1326 *ppos = pos + retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001327 if (retval) {
1328 file_accessed(filp);
1329 goto out;
1330 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 }
1333
Hugh Dickins11fa9772008-07-23 21:27:34 -07001334 for (seg = 0; seg < nr_segs; seg++) {
1335 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Hugh Dickins11fa9772008-07-23 21:27:34 -07001337 desc.written = 0;
1338 desc.arg.buf = iov[seg].iov_base;
1339 desc.count = iov[seg].iov_len;
1340 if (desc.count == 0)
1341 continue;
1342 desc.error = 0;
1343 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1344 retval += desc.written;
1345 if (desc.error) {
1346 retval = retval ?: desc.error;
1347 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001349 if (desc.count > 0)
1350 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352out:
1353 return retval;
1354}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355EXPORT_SYMBOL(generic_file_aio_read);
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357static ssize_t
1358do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001359 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360{
1361 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1362 return -EINVAL;
1363
1364 force_page_cache_readahead(mapping, filp, index,
1365 max_sane_readahead(nr));
1366 return 0;
1367}
1368
1369asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count)
1370{
1371 ssize_t ret;
1372 struct file *file;
1373
1374 ret = -EBADF;
1375 file = fget(fd);
1376 if (file) {
1377 if (file->f_mode & FMODE_READ) {
1378 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001379 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1380 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 unsigned long len = end - start + 1;
1382 ret = do_readahead(mapping, file, start, len);
1383 }
1384 fput(file);
1385 }
1386 return ret;
1387}
1388
1389#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001390/**
1391 * page_cache_read - adds requested page to the page cache if not already there
1392 * @file: file to read
1393 * @offset: page index
1394 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 * This adds the requested page to the page cache if it isn't already there,
1396 * and schedules an I/O to read in its contents from disk.
1397 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001398static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
1400 struct address_space *mapping = file->f_mapping;
1401 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001402 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Zach Brown994fc28c2005-12-15 14:28:17 -08001404 do {
1405 page = page_cache_alloc_cold(mapping);
1406 if (!page)
1407 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408
Zach Brown994fc28c2005-12-15 14:28:17 -08001409 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1410 if (ret == 0)
1411 ret = mapping->a_ops->readpage(file, page);
1412 else if (ret == -EEXIST)
1413 ret = 0; /* losing race to add is OK */
1414
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Zach Brown994fc28c2005-12-15 14:28:17 -08001417 } while (ret == AOP_TRUNCATED_PAGE);
1418
1419 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
1421
1422#define MMAP_LOTSAMISS (100)
1423
Randy Dunlap485bb992006-06-23 02:03:49 -07001424/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001425 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001426 * @vma: vma in which the fault was taken
1427 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001428 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001429 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 * mapped memory region to read in file data during a page fault.
1431 *
1432 * The goto's are kind of ugly, but this streamlines the normal case of having
1433 * it in the page cache, and handles the special cases reasonably without
1434 * having a lot of duplicated code.
1435 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001436int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437{
1438 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001439 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 struct address_space *mapping = file->f_mapping;
1441 struct file_ra_state *ra = &file->f_ra;
1442 struct inode *inode = mapping->host;
1443 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001444 pgoff_t size;
Nick Piggin54cb8822007-07-19 01:46:59 -07001445 int did_readaround = 0;
Nick Piggin83c54072007-07-19 01:47:05 -07001446 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001449 if (vmf->pgoff >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001450 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451
1452 /* If we don't want any read-ahead, don't bother */
Nick Piggin54cb8822007-07-19 01:46:59 -07001453 if (VM_RandomReadHint(vma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 goto no_cached_page;
1455
1456 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 * Do we have something in the page cache already?
1458 */
1459retry_find:
Nick Piggind0217ac2007-07-19 01:47:03 -07001460 page = find_lock_page(mapping, vmf->pgoff);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001461 /*
1462 * For sequential accesses, we use the generic readahead logic.
1463 */
1464 if (VM_SequentialReadHint(vma)) {
1465 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001466 page_cache_sync_readahead(mapping, ra, file,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001467 vmf->pgoff, 1);
1468 page = find_lock_page(mapping, vmf->pgoff);
1469 if (!page)
1470 goto no_cached_page;
1471 }
1472 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001473 page_cache_async_readahead(mapping, ra, file, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001474 vmf->pgoff, 1);
1475 }
1476 }
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 if (!page) {
1479 unsigned long ra_pages;
1480
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 ra->mmap_miss++;
1482
1483 /*
1484 * Do we miss much more than hit in this file? If so,
1485 * stop bothering with read-ahead. It will only hurt.
1486 */
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001487 if (ra->mmap_miss > MMAP_LOTSAMISS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 goto no_cached_page;
1489
1490 /*
1491 * To keep the pgmajfault counter straight, we need to
1492 * check did_readaround, as this is an inner loop.
1493 */
1494 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001495 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001496 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 }
1498 did_readaround = 1;
1499 ra_pages = max_sane_readahead(file->f_ra.ra_pages);
1500 if (ra_pages) {
1501 pgoff_t start = 0;
1502
Nick Piggind0217ac2007-07-19 01:47:03 -07001503 if (vmf->pgoff > ra_pages / 2)
1504 start = vmf->pgoff - ra_pages / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 do_page_cache_readahead(mapping, file, start, ra_pages);
1506 }
Nick Piggind0217ac2007-07-19 01:47:03 -07001507 page = find_lock_page(mapping, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 if (!page)
1509 goto no_cached_page;
1510 }
1511
1512 if (!did_readaround)
Fengguang Wu0bb7ba62007-10-16 01:24:32 -07001513 ra->mmap_miss--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
1515 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001516 * We have a locked page in the page cache, now we need to check
1517 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 */
Nick Piggind00806b2007-07-19 01:46:57 -07001519 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 goto page_not_uptodate;
1521
Nick Piggind00806b2007-07-19 01:46:57 -07001522 /* Must recheck i_size under page lock */
1523 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001524 if (unlikely(vmf->pgoff >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001525 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001526 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001527 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001528 }
1529
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 /*
1531 * Found the page and have a reference on it.
1532 */
1533 mark_page_accessed(page);
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001534 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001535 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001536 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538no_cached_page:
1539 /*
1540 * We're only likely to ever get here if MADV_RANDOM is in
1541 * effect.
1542 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001543 error = page_cache_read(file, vmf->pgoff);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 /*
1546 * The page we want has now been added to the page cache.
1547 * In the unlikely event that someone removed it in the
1548 * meantime, we'll just come back here and read it again.
1549 */
1550 if (error >= 0)
1551 goto retry_find;
1552
1553 /*
1554 * An error return from page_cache_read can result if the
1555 * system is low on memory, or a problem occurs while trying
1556 * to schedule I/O.
1557 */
1558 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001559 return VM_FAULT_OOM;
1560 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561
1562page_not_uptodate:
Nick Piggind00806b2007-07-19 01:46:57 -07001563 /* IO error path */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 if (!did_readaround) {
Nick Piggind0217ac2007-07-19 01:47:03 -07001565 ret = VM_FAULT_MAJOR;
Christoph Lameterf8891e52006-06-30 01:55:45 -07001566 count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
1569 /*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593struct 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*),
1637 void *data)
1638{
Nick Piggineb2be182007-10-16 01:24:57 -07001639 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 int err;
1641repeat:
1642 page = find_get_page(mapping, index);
1643 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001644 page = page_cache_alloc_cold(mapping);
1645 if (!page)
1646 return ERR_PTR(-ENOMEM);
1647 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1648 if (unlikely(err)) {
1649 page_cache_release(page);
1650 if (err == -EEXIST)
1651 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 return ERR_PTR(err);
1654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 err = filler(data, page);
1656 if (err < 0) {
1657 page_cache_release(page);
1658 page = ERR_PTR(err);
1659 }
1660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return page;
1662}
1663
Randy Dunlap76824862008-03-19 17:00:40 -07001664/**
1665 * read_cache_page_async - read into page cache, fill it if needed
1666 * @mapping: the page's address_space
1667 * @index: the page index
1668 * @filler: function to perform the read
1669 * @data: destination for read data
1670 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001671 * Same as read_cache_page, but don't wait for page to become unlocked
1672 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001673 *
1674 * Read into the page cache. If a page already exists, and PageUptodate() is
1675 * not set, try to fill the page but don't wait for it to become unlocked.
1676 *
1677 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001679struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001680 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 int (*filler)(void *,struct page*),
1682 void *data)
1683{
1684 struct page *page;
1685 int err;
1686
1687retry:
1688 page = __read_cache_page(mapping, index, filler, data);
1689 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001690 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 if (PageUptodate(page))
1692 goto out;
1693
1694 lock_page(page);
1695 if (!page->mapping) {
1696 unlock_page(page);
1697 page_cache_release(page);
1698 goto retry;
1699 }
1700 if (PageUptodate(page)) {
1701 unlock_page(page);
1702 goto out;
1703 }
1704 err = filler(data, page);
1705 if (err < 0) {
1706 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001707 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
David Howellsc855ff32007-05-09 13:42:20 +01001709out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001710 mark_page_accessed(page);
1711 return page;
1712}
1713EXPORT_SYMBOL(read_cache_page_async);
1714
1715/**
1716 * read_cache_page - read into page cache, fill it if needed
1717 * @mapping: the page's address_space
1718 * @index: the page index
1719 * @filler: function to perform the read
1720 * @data: destination for read data
1721 *
1722 * Read into the page cache. If a page already exists, and PageUptodate() is
1723 * not set, try to fill the page then wait for it to become unlocked.
1724 *
1725 * If the page does not get brought uptodate, return -EIO.
1726 */
1727struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001728 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001729 int (*filler)(void *,struct page*),
1730 void *data)
1731{
1732 struct page *page;
1733
1734 page = read_cache_page_async(mapping, index, filler, data);
1735 if (IS_ERR(page))
1736 goto out;
1737 wait_on_page_locked(page);
1738 if (!PageUptodate(page)) {
1739 page_cache_release(page);
1740 page = ERR_PTR(-EIO);
1741 }
1742 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 return page;
1744}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745EXPORT_SYMBOL(read_cache_page);
1746
1747/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 * The logic we want is
1749 *
1750 * if suid or (sgid and xgrp)
1751 * remove privs
1752 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001753int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754{
1755 mode_t mode = dentry->d_inode->i_mode;
1756 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
1758 /* suid always must be killed */
1759 if (unlikely(mode & S_ISUID))
1760 kill = ATTR_KILL_SUID;
1761
1762 /*
1763 * sgid without any exec bits is just a mandatory locking mark; leave
1764 * it alone. If some exec bits are set, it's a real sgid; kill it.
1765 */
1766 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1767 kill |= ATTR_KILL_SGID;
1768
Jens Axboe01de85e2006-10-17 19:50:36 +02001769 if (unlikely(kill && !capable(CAP_FSETID)))
1770 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
Jens Axboe01de85e2006-10-17 19:50:36 +02001772 return 0;
1773}
Mark Fashehd23a1472006-10-17 17:05:18 -07001774EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001775
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001776static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001777{
1778 struct iattr newattrs;
1779
1780 newattrs.ia_valid = ATTR_FORCE | kill;
1781 return notify_change(dentry, &newattrs);
1782}
1783
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001784int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001785{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001786 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001787 int killsuid = should_remove_suid(dentry);
1788 int killpriv = security_inode_need_killpriv(dentry);
1789 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001790
Serge E. Hallynb5376772007-10-16 23:31:36 -07001791 if (killpriv < 0)
1792 return killpriv;
1793 if (killpriv)
1794 error = security_inode_killpriv(dentry);
1795 if (!error && killsuid)
1796 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001797
Serge E. Hallynb5376772007-10-16 23:31:36 -07001798 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001800EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Nick Piggin2f718ff2007-10-16 01:24:59 -07001802static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 const struct iovec *iov, size_t base, size_t bytes)
1804{
1805 size_t copied = 0, left = 0;
1806
1807 while (bytes) {
1808 char __user *buf = iov->iov_base + base;
1809 int copy = min(bytes, iov->iov_len - base);
1810
1811 base = 0;
Hiro Yoshiokac22ce142006-06-23 02:04:16 -07001812 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 copied += copy;
1814 bytes -= copy;
1815 vaddr += copy;
1816 iov++;
1817
NeilBrown01408c42006-06-25 05:47:58 -07001818 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 }
1821 return copied - left;
1822}
1823
1824/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001825 * Copy as much as we can into the page and return the number of bytes which
1826 * were sucessfully copied. If a fault is encountered then return the number of
1827 * bytes which were copied.
1828 */
1829size_t iov_iter_copy_from_user_atomic(struct page *page,
1830 struct iov_iter *i, unsigned long offset, size_t bytes)
1831{
1832 char *kaddr;
1833 size_t copied;
1834
1835 BUG_ON(!in_atomic());
1836 kaddr = kmap_atomic(page, KM_USER0);
1837 if (likely(i->nr_segs == 1)) {
1838 int left;
1839 char __user *buf = i->iov->iov_base + i->iov_offset;
1840 left = __copy_from_user_inatomic_nocache(kaddr + offset,
1841 buf, bytes);
1842 copied = bytes - left;
1843 } else {
1844 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1845 i->iov, i->iov_offset, bytes);
1846 }
1847 kunmap_atomic(kaddr, KM_USER0);
1848
1849 return copied;
1850}
Nick Piggin89e10782007-10-16 01:25:07 -07001851EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001852
1853/*
1854 * This has the same sideeffects and return value as
1855 * iov_iter_copy_from_user_atomic().
1856 * The difference is that it attempts to resolve faults.
1857 * Page must not be locked.
1858 */
1859size_t iov_iter_copy_from_user(struct page *page,
1860 struct iov_iter *i, unsigned long offset, size_t bytes)
1861{
1862 char *kaddr;
1863 size_t copied;
1864
1865 kaddr = kmap(page);
1866 if (likely(i->nr_segs == 1)) {
1867 int left;
1868 char __user *buf = i->iov->iov_base + i->iov_offset;
1869 left = __copy_from_user_nocache(kaddr + offset, buf, bytes);
1870 copied = bytes - left;
1871 } else {
1872 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1873 i->iov, i->iov_offset, bytes);
1874 }
1875 kunmap(page);
1876 return copied;
1877}
Nick Piggin89e10782007-10-16 01:25:07 -07001878EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001879
Nick Pigginf7009262008-03-10 11:43:59 -07001880void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001881{
Nick Pigginf7009262008-03-10 11:43:59 -07001882 BUG_ON(i->count < bytes);
1883
Nick Piggin2f718ff2007-10-16 01:24:59 -07001884 if (likely(i->nr_segs == 1)) {
1885 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001886 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001887 } else {
1888 const struct iovec *iov = i->iov;
1889 size_t base = i->iov_offset;
1890
Nick Piggin124d3b72008-02-02 15:01:17 +01001891 /*
1892 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001893 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001894 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001895 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001896 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001897
Nick Pigginf7009262008-03-10 11:43:59 -07001898 copy = min(bytes, iov->iov_len - base);
1899 BUG_ON(!i->count || i->count < copy);
1900 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001901 bytes -= copy;
1902 base += copy;
1903 if (iov->iov_len == base) {
1904 iov++;
1905 base = 0;
1906 }
1907 }
1908 i->iov = iov;
1909 i->iov_offset = base;
1910 }
1911}
Nick Piggin89e10782007-10-16 01:25:07 -07001912EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001913
Nick Pigginafddba42007-10-16 01:25:01 -07001914/*
1915 * Fault in the first iovec of the given iov_iter, to a maximum length
1916 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1917 * accessed (ie. because it is an invalid address).
1918 *
1919 * writev-intensive code may want this to prefault several iovecs -- that
1920 * would be possible (callers must not rely on the fact that _only_ the
1921 * first iovec will be faulted with the current implementation).
1922 */
1923int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001924{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001925 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001926 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1927 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001928}
Nick Piggin89e10782007-10-16 01:25:07 -07001929EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001930
1931/*
1932 * Return the count of just the current iov_iter segment.
1933 */
1934size_t iov_iter_single_seg_count(struct iov_iter *i)
1935{
1936 const struct iovec *iov = i->iov;
1937 if (i->nr_segs == 1)
1938 return i->count;
1939 else
1940 return min(i->count, iov->iov_len - i->iov_offset);
1941}
Nick Piggin89e10782007-10-16 01:25:07 -07001942EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001943
1944/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 * Performs necessary checks before doing a write
1946 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001947 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 * Returns appropriate error code that caller should return or
1949 * zero in case that write should be allowed.
1950 */
1951inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1952{
1953 struct inode *inode = file->f_mapping->host;
1954 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1955
1956 if (unlikely(*pos < 0))
1957 return -EINVAL;
1958
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 if (!isblk) {
1960 /* FIXME: this is for backwards compatibility with 2.4 */
1961 if (file->f_flags & O_APPEND)
1962 *pos = i_size_read(inode);
1963
1964 if (limit != RLIM_INFINITY) {
1965 if (*pos >= limit) {
1966 send_sig(SIGXFSZ, current, 0);
1967 return -EFBIG;
1968 }
1969 if (*count > limit - (typeof(limit))*pos) {
1970 *count = limit - (typeof(limit))*pos;
1971 }
1972 }
1973 }
1974
1975 /*
1976 * LFS rule
1977 */
1978 if (unlikely(*pos + *count > MAX_NON_LFS &&
1979 !(file->f_flags & O_LARGEFILE))) {
1980 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 return -EFBIG;
1982 }
1983 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
1984 *count = MAX_NON_LFS - (unsigned long)*pos;
1985 }
1986 }
1987
1988 /*
1989 * Are we about to exceed the fs block limit ?
1990 *
1991 * If we have written data it becomes a short write. If we have
1992 * exceeded without writing data we send a signal and return EFBIG.
1993 * Linus frestrict idea will clean these up nicely..
1994 */
1995 if (likely(!isblk)) {
1996 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
1997 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 return -EFBIG;
1999 }
2000 /* zero-length writes at ->s_maxbytes are OK */
2001 }
2002
2003 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2004 *count = inode->i_sb->s_maxbytes - *pos;
2005 } else {
David Howells93614012006-09-30 20:45:40 +02002006#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 loff_t isize;
2008 if (bdev_read_only(I_BDEV(inode)))
2009 return -EPERM;
2010 isize = i_size_read(inode);
2011 if (*pos >= isize) {
2012 if (*count || *pos > isize)
2013 return -ENOSPC;
2014 }
2015
2016 if (*pos + *count > isize)
2017 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002018#else
2019 return -EPERM;
2020#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 }
2022 return 0;
2023}
2024EXPORT_SYMBOL(generic_write_checks);
2025
Nick Pigginafddba42007-10-16 01:25:01 -07002026int pagecache_write_begin(struct file *file, struct address_space *mapping,
2027 loff_t pos, unsigned len, unsigned flags,
2028 struct page **pagep, void **fsdata)
2029{
2030 const struct address_space_operations *aops = mapping->a_ops;
2031
2032 if (aops->write_begin) {
2033 return aops->write_begin(file, mapping, pos, len, flags,
2034 pagep, fsdata);
2035 } else {
2036 int ret;
2037 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2038 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2039 struct inode *inode = mapping->host;
2040 struct page *page;
2041again:
2042 page = __grab_cache_page(mapping, index);
2043 *pagep = page;
2044 if (!page)
2045 return -ENOMEM;
2046
2047 if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) {
2048 /*
2049 * There is no way to resolve a short write situation
2050 * for a !Uptodate page (except by double copying in
2051 * the caller done by generic_perform_write_2copy).
2052 *
2053 * Instead, we have to bring it uptodate here.
2054 */
2055 ret = aops->readpage(file, page);
2056 page_cache_release(page);
2057 if (ret) {
2058 if (ret == AOP_TRUNCATED_PAGE)
2059 goto again;
2060 return ret;
2061 }
2062 goto again;
2063 }
2064
2065 ret = aops->prepare_write(file, page, offset, offset+len);
2066 if (ret) {
Nick Piggin55144762007-10-16 01:25:26 -07002067 unlock_page(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002068 page_cache_release(page);
2069 if (pos + len > inode->i_size)
2070 vmtruncate(inode, inode->i_size);
Nick Pigginafddba42007-10-16 01:25:01 -07002071 }
2072 return ret;
2073 }
2074}
2075EXPORT_SYMBOL(pagecache_write_begin);
2076
2077int pagecache_write_end(struct file *file, struct address_space *mapping,
2078 loff_t pos, unsigned len, unsigned copied,
2079 struct page *page, void *fsdata)
2080{
2081 const struct address_space_operations *aops = mapping->a_ops;
2082 int ret;
2083
2084 if (aops->write_end) {
2085 mark_page_accessed(page);
2086 ret = aops->write_end(file, mapping, pos, len, copied,
2087 page, fsdata);
2088 } else {
2089 unsigned offset = pos & (PAGE_CACHE_SIZE - 1);
2090 struct inode *inode = mapping->host;
2091
2092 flush_dcache_page(page);
2093 ret = aops->commit_write(file, page, offset, offset+len);
2094 unlock_page(page);
2095 mark_page_accessed(page);
2096 page_cache_release(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002097
2098 if (ret < 0) {
2099 if (pos + len > inode->i_size)
2100 vmtruncate(inode, inode->i_size);
2101 } else if (ret > 0)
2102 ret = min_t(size_t, copied, ret);
2103 else
2104 ret = copied;
2105 }
2106
2107 return ret;
2108}
2109EXPORT_SYMBOL(pagecache_write_end);
2110
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111ssize_t
2112generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2113 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2114 size_t count, size_t ocount)
2115{
2116 struct file *file = iocb->ki_filp;
2117 struct address_space *mapping = file->f_mapping;
2118 struct inode *inode = mapping->host;
2119 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002120 size_t write_len;
2121 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122
2123 if (count != ocount)
2124 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2125
Christoph Hellwiga969e902008-07-23 21:27:04 -07002126 /*
2127 * Unmap all mmappings of the file up-front.
2128 *
2129 * This will cause any pte dirty bits to be propagated into the
2130 * pageframes for the subsequent filemap_write_and_wait().
2131 */
2132 write_len = iov_length(iov, *nr_segs);
2133 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2134 if (mapping_mapped(mapping))
2135 unmap_mapping_range(mapping, pos, write_len, 0);
2136
2137 written = filemap_write_and_wait(mapping);
2138 if (written)
2139 goto out;
2140
2141 /*
2142 * After a write we want buffered reads to be sure to go to disk to get
2143 * the new data. We invalidate clean cached page from the region we're
2144 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002145 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002146 */
2147 if (mapping->nrpages) {
2148 written = invalidate_inode_pages2_range(mapping,
2149 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002150 /*
2151 * If a page can not be invalidated, return 0 to fall back
2152 * to buffered write.
2153 */
2154 if (written) {
2155 if (written == -EBUSY)
2156 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002157 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002158 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002159 }
2160
2161 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2162
2163 /*
2164 * Finally, try again to invalidate clean pages which might have been
2165 * cached by non-direct readahead, or faulted in by get_user_pages()
2166 * if the source of the write was an mmap'ed region of the file
2167 * we're writing. Either one is a pretty crazy thing to do,
2168 * so we don't support it 100%. If this invalidation
2169 * fails, tough, the write still worked...
2170 */
2171 if (mapping->nrpages) {
2172 invalidate_inode_pages2_range(mapping,
2173 pos >> PAGE_CACHE_SHIFT, end);
2174 }
2175
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 if (written > 0) {
2177 loff_t end = pos + written;
2178 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2179 i_size_write(inode, end);
2180 mark_inode_dirty(inode);
2181 }
2182 *ppos = end;
2183 }
2184
2185 /*
2186 * Sync the fs metadata but not the minor inode changes and
2187 * of course not the data as we did direct DMA for the IO.
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002188 * i_mutex is held, which protects generic_osync_inode() from
Zach Brown8459d862006-12-10 02:21:05 -08002189 * livelocking. AIO O_DIRECT ops attempt to sync metadata here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 */
Christoph Hellwiga969e902008-07-23 21:27:04 -07002191out:
Zach Brown8459d862006-12-10 02:21:05 -08002192 if ((written >= 0 || written == -EIOCBQUEUED) &&
2193 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Hifumi Hisashi1e8a81c2005-06-25 14:54:32 -07002194 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA);
2195 if (err < 0)
2196 written = err;
2197 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 return written;
2199}
2200EXPORT_SYMBOL(generic_file_direct_write);
2201
Nick Piggineb2be182007-10-16 01:24:57 -07002202/*
2203 * Find or create a page at the given pagecache position. Return the locked
2204 * page. This function is specifically for buffered writes.
2205 */
Nick Pigginafddba42007-10-16 01:25:01 -07002206struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index)
Nick Piggineb2be182007-10-16 01:24:57 -07002207{
2208 int status;
2209 struct page *page;
2210repeat:
2211 page = find_lock_page(mapping, index);
2212 if (likely(page))
2213 return page;
2214
2215 page = page_cache_alloc(mapping);
2216 if (!page)
2217 return NULL;
2218 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
2219 if (unlikely(status)) {
2220 page_cache_release(page);
2221 if (status == -EEXIST)
2222 goto repeat;
2223 return NULL;
2224 }
2225 return page;
2226}
Nick Pigginafddba42007-10-16 01:25:01 -07002227EXPORT_SYMBOL(__grab_cache_page);
Nick Piggineb2be182007-10-16 01:24:57 -07002228
Nick Pigginafddba42007-10-16 01:25:01 -07002229static ssize_t generic_perform_write_2copy(struct file *file,
2230 struct iov_iter *i, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231{
Andrew Mortonae374612007-10-16 01:24:55 -07002232 struct address_space *mapping = file->f_mapping;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002233 const struct address_space_operations *a_ops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002234 struct inode *inode = mapping->host;
2235 long status = 0;
2236 ssize_t written = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238 do {
Nick Piggin08291422007-10-16 01:24:59 -07002239 struct page *src_page;
Nick Piggineb2be182007-10-16 01:24:57 -07002240 struct page *page;
Andrew Mortonae374612007-10-16 01:24:55 -07002241 pgoff_t index; /* Pagecache index for current page */
2242 unsigned long offset; /* Offset into pagecache page */
Nick Piggin08291422007-10-16 01:24:59 -07002243 unsigned long bytes; /* Bytes to write to page */
Andrew Mortonae374612007-10-16 01:24:55 -07002244 size_t copied; /* Bytes copied from user */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
Andrew Mortonae374612007-10-16 01:24:55 -07002246 offset = (pos & (PAGE_CACHE_SIZE - 1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin2f718ff2007-10-16 01:24:59 -07002248 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
Nick Pigginafddba42007-10-16 01:25:01 -07002249 iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250
Nick Piggin08291422007-10-16 01:24:59 -07002251 /*
2252 * a non-NULL src_page indicates that we're doing the
2253 * copy via get_user_pages and kmap.
2254 */
2255 src_page = NULL;
Andrew Mortonae374612007-10-16 01:24:55 -07002256
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002257 /*
2258 * Bring in the user page that we will copy from _first_.
2259 * Otherwise there's a nasty deadlock on copying from the
2260 * same page as we're writing to, without it being marked
2261 * up-to-date.
Nick Piggin08291422007-10-16 01:24:59 -07002262 *
2263 * Not only is this an optimisation, but it is also required
2264 * to check that the address is actually valid, when atomic
2265 * usercopies are used, below.
Nick Piggin41cb8ac2007-10-16 01:24:53 -07002266 */
Nick Pigginafddba42007-10-16 01:25:01 -07002267 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
Nick Piggin08291422007-10-16 01:24:59 -07002268 status = -EFAULT;
2269 break;
2270 }
Nick Piggineb2be182007-10-16 01:24:57 -07002271
2272 page = __grab_cache_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 if (!page) {
2274 status = -ENOMEM;
2275 break;
2276 }
2277
Nick Piggin08291422007-10-16 01:24:59 -07002278 /*
2279 * non-uptodate pages cannot cope with short copies, and we
2280 * cannot take a pagefault with the destination page locked.
2281 * So pin the source page to copy it.
2282 */
Nick Piggin674b8922007-10-16 01:25:03 -07002283 if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) {
Nick Piggin08291422007-10-16 01:24:59 -07002284 unlock_page(page);
2285
2286 src_page = alloc_page(GFP_KERNEL);
2287 if (!src_page) {
2288 page_cache_release(page);
2289 status = -ENOMEM;
2290 break;
2291 }
2292
2293 /*
2294 * Cannot get_user_pages with a page locked for the
2295 * same reason as we can't take a page fault with a
2296 * page locked (as explained below).
2297 */
Nick Pigginafddba42007-10-16 01:25:01 -07002298 copied = iov_iter_copy_from_user(src_page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002299 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002300 if (unlikely(copied == 0)) {
2301 status = -EFAULT;
2302 page_cache_release(page);
2303 page_cache_release(src_page);
2304 break;
2305 }
2306 bytes = copied;
2307
2308 lock_page(page);
2309 /*
2310 * Can't handle the page going uptodate here, because
2311 * that means we would use non-atomic usercopies, which
2312 * zero out the tail of the page, which can cause
2313 * zeroes to become transiently visible. We could just
2314 * use a non-zeroing copy, but the APIs aren't too
2315 * consistent.
2316 */
2317 if (unlikely(!page->mapping || PageUptodate(page))) {
2318 unlock_page(page);
2319 page_cache_release(page);
2320 page_cache_release(src_page);
2321 continue;
2322 }
Nick Piggin08291422007-10-16 01:24:59 -07002323 }
2324
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 status = a_ops->prepare_write(file, page, offset, offset+bytes);
Nick Piggin64649a52007-10-16 01:24:56 -07002326 if (unlikely(status))
2327 goto fs_write_aop_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08002328
Nick Piggin08291422007-10-16 01:24:59 -07002329 if (!src_page) {
2330 /*
2331 * Must not enter the pagefault handler here, because
2332 * we hold the page lock, so we might recursively
2333 * deadlock on the same lock, or get an ABBA deadlock
2334 * against a different lock, or against the mmap_sem
2335 * (which nests outside the page lock). So increment
2336 * preempt count, and use _atomic usercopies.
2337 *
2338 * The page is uptodate so we are OK to encounter a
2339 * short copy: if unmodified parts of the page are
2340 * marked dirty and written out to disk, it doesn't
2341 * really matter.
2342 */
2343 pagefault_disable();
Nick Pigginafddba42007-10-16 01:25:01 -07002344 copied = iov_iter_copy_from_user_atomic(page, i,
Nick Piggin2f718ff2007-10-16 01:24:59 -07002345 offset, bytes);
Nick Piggin08291422007-10-16 01:24:59 -07002346 pagefault_enable();
2347 } else {
2348 void *src, *dst;
2349 src = kmap_atomic(src_page, KM_USER0);
2350 dst = kmap_atomic(page, KM_USER1);
2351 memcpy(dst + offset, src + offset, bytes);
2352 kunmap_atomic(dst, KM_USER1);
2353 kunmap_atomic(src, KM_USER0);
2354 copied = bytes;
2355 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 flush_dcache_page(page);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002357
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 status = a_ops->commit_write(file, page, offset, offset+bytes);
Nick Piggin55144762007-10-16 01:25:26 -07002359 if (unlikely(status < 0))
Nick Piggin64649a52007-10-16 01:24:56 -07002360 goto fs_write_aop_error;
Nick Piggin64649a52007-10-16 01:24:56 -07002361 if (unlikely(status > 0)) /* filesystem did partial write */
Nick Piggin08291422007-10-16 01:24:59 -07002362 copied = min_t(size_t, copied, status);
2363
2364 unlock_page(page);
2365 mark_page_accessed(page);
2366 page_cache_release(page);
2367 if (src_page)
2368 page_cache_release(src_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369
Nick Pigginafddba42007-10-16 01:25:01 -07002370 iov_iter_advance(i, copied);
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002371 pos += copied;
Nick Pigginafddba42007-10-16 01:25:01 -07002372 written += copied;
Nick Piggin4a9e5ef2007-10-16 01:24:58 -07002373
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 balance_dirty_pages_ratelimited(mapping);
2375 cond_resched();
Nick Piggin64649a52007-10-16 01:24:56 -07002376 continue;
2377
2378fs_write_aop_error:
Nick Piggin55144762007-10-16 01:25:26 -07002379 unlock_page(page);
Nick Piggin64649a52007-10-16 01:24:56 -07002380 page_cache_release(page);
Nick Piggin08291422007-10-16 01:24:59 -07002381 if (src_page)
2382 page_cache_release(src_page);
Nick Piggin64649a52007-10-16 01:24:56 -07002383
2384 /*
2385 * prepare_write() may have instantiated a few blocks
2386 * outside i_size. Trim these off again. Don't need
2387 * i_size_read because we hold i_mutex.
2388 */
2389 if (pos + bytes > inode->i_size)
2390 vmtruncate(inode, inode->i_size);
Nick Piggin55144762007-10-16 01:25:26 -07002391 break;
Nick Pigginafddba42007-10-16 01:25:01 -07002392 } while (iov_iter_count(i));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Nick Pigginafddba42007-10-16 01:25:01 -07002394 return written ? written : status;
2395}
2396
2397static ssize_t generic_perform_write(struct file *file,
2398 struct iov_iter *i, loff_t pos)
2399{
2400 struct address_space *mapping = file->f_mapping;
2401 const struct address_space_operations *a_ops = mapping->a_ops;
2402 long status = 0;
2403 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002404 unsigned int flags = 0;
2405
2406 /*
2407 * Copies from kernel address space cannot fail (NFSD is a big user).
2408 */
2409 if (segment_eq(get_fs(), KERNEL_DS))
2410 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002411
2412 do {
2413 struct page *page;
2414 pgoff_t index; /* Pagecache index for current page */
2415 unsigned long offset; /* Offset into pagecache page */
2416 unsigned long bytes; /* Bytes to write to page */
2417 size_t copied; /* Bytes copied from user */
2418 void *fsdata;
2419
2420 offset = (pos & (PAGE_CACHE_SIZE - 1));
2421 index = pos >> PAGE_CACHE_SHIFT;
2422 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2423 iov_iter_count(i));
2424
2425again:
2426
2427 /*
2428 * Bring in the user page that we will copy from _first_.
2429 * Otherwise there's a nasty deadlock on copying from the
2430 * same page as we're writing to, without it being marked
2431 * up-to-date.
2432 *
2433 * Not only is this an optimisation, but it is also required
2434 * to check that the address is actually valid, when atomic
2435 * usercopies are used, below.
2436 */
2437 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2438 status = -EFAULT;
2439 break;
2440 }
2441
Nick Piggin674b8922007-10-16 01:25:03 -07002442 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002443 &page, &fsdata);
2444 if (unlikely(status))
2445 break;
2446
2447 pagefault_disable();
2448 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2449 pagefault_enable();
2450 flush_dcache_page(page);
2451
2452 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2453 page, fsdata);
2454 if (unlikely(status < 0))
2455 break;
2456 copied = status;
2457
2458 cond_resched();
2459
Nick Piggin124d3b72008-02-02 15:01:17 +01002460 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002461 if (unlikely(copied == 0)) {
2462 /*
2463 * If we were unable to copy any data at all, we must
2464 * fall back to a single segment length write.
2465 *
2466 * If we didn't fallback here, we could livelock
2467 * because not all segments in the iov can be copied at
2468 * once without a pagefault.
2469 */
2470 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2471 iov_iter_single_seg_count(i));
2472 goto again;
2473 }
Nick Pigginafddba42007-10-16 01:25:01 -07002474 pos += copied;
2475 written += copied;
2476
2477 balance_dirty_pages_ratelimited(mapping);
2478
2479 } while (iov_iter_count(i));
2480
2481 return written ? written : status;
2482}
2483
2484ssize_t
2485generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2486 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2487 size_t count, ssize_t written)
2488{
2489 struct file *file = iocb->ki_filp;
2490 struct address_space *mapping = file->f_mapping;
2491 const struct address_space_operations *a_ops = mapping->a_ops;
2492 struct inode *inode = mapping->host;
2493 ssize_t status;
2494 struct iov_iter i;
2495
2496 iov_iter_init(&i, iov, nr_segs, count, written);
2497 if (a_ops->write_begin)
2498 status = generic_perform_write(file, &i, pos);
2499 else
2500 status = generic_perform_write_2copy(file, &i, pos);
2501
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002503 written += status;
2504 *ppos = pos + status;
2505
2506 /*
2507 * For now, when the user asks for O_SYNC, we'll actually give
2508 * O_DSYNC
2509 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2511 if (!a_ops->writepage || !is_sync_kiocb(iocb))
2512 status = generic_osync_inode(inode, mapping,
2513 OSYNC_METADATA|OSYNC_DATA);
2514 }
2515 }
2516
2517 /*
2518 * If we get here for O_DIRECT writes then we must have fallen through
2519 * to buffered writes (block instantiation inside i_size). So we sync
2520 * the file data here, to try to honour O_DIRECT expectations.
2521 */
2522 if (unlikely(file->f_flags & O_DIRECT) && written)
2523 status = filemap_write_and_wait(mapping);
2524
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 return written ? written : status;
2526}
2527EXPORT_SYMBOL(generic_file_buffered_write);
2528
Adrian Bunk5ce78522005-09-10 00:26:28 -07002529static ssize_t
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov,
2531 unsigned long nr_segs, loff_t *ppos)
2532{
2533 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002534 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 size_t ocount; /* original count */
2536 size_t count; /* after file limit checks */
2537 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 loff_t pos;
2539 ssize_t written;
2540 ssize_t err;
2541
2542 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002543 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2544 if (err)
2545 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546
2547 count = ocount;
2548 pos = *ppos;
2549
2550 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2551
2552 /* We can write back this queue in page reclaim */
2553 current->backing_dev_info = mapping->backing_dev_info;
2554 written = 0;
2555
2556 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2557 if (err)
2558 goto out;
2559
2560 if (count == 0)
2561 goto out;
2562
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002563 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 if (err)
2565 goto out;
2566
Christoph Hellwig870f4812006-01-09 20:52:01 -08002567 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
2569 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2570 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002571 loff_t endbyte;
2572 ssize_t written_buffered;
2573
2574 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2575 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 if (written < 0 || written == count)
2577 goto out;
2578 /*
2579 * direct-io write to a hole: fall through to buffered I/O
2580 * for completing the rest of the request.
2581 */
2582 pos += written;
2583 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002584 written_buffered = generic_file_buffered_write(iocb, iov,
2585 nr_segs, pos, ppos, count,
2586 written);
2587 /*
2588 * If generic_file_buffered_write() retuned a synchronous error
2589 * then we want to return the number of bytes which were
2590 * direct-written, or the error code if that was zero. Note
2591 * that this differs from normal direct-io semantics, which
2592 * will return -EFOO even if some bytes were written.
2593 */
2594 if (written_buffered < 0) {
2595 err = written_buffered;
2596 goto out;
2597 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002599 /*
2600 * We need to ensure that the page cache pages are written to
2601 * disk and invalidated to preserve the expected O_DIRECT
2602 * semantics.
2603 */
2604 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002605 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2606 SYNC_FILE_RANGE_WAIT_BEFORE|
2607 SYNC_FILE_RANGE_WRITE|
2608 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002609 if (err == 0) {
2610 written = written_buffered;
2611 invalidate_mapping_pages(mapping,
2612 pos >> PAGE_CACHE_SHIFT,
2613 endbyte >> PAGE_CACHE_SHIFT);
2614 } else {
2615 /*
2616 * We don't know how much we wrote, so just return
2617 * the number of bytes which were direct-written
2618 */
2619 }
2620 } else {
2621 written = generic_file_buffered_write(iocb, iov, nr_segs,
2622 pos, ppos, count, written);
2623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624out:
2625 current->backing_dev_info = NULL;
2626 return written ? written : err;
2627}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628
Badari Pulavarty027445c2006-09-30 23:28:46 -07002629ssize_t generic_file_aio_write_nolock(struct kiocb *iocb,
2630 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631{
2632 struct file *file = iocb->ki_filp;
2633 struct address_space *mapping = file->f_mapping;
2634 struct inode *inode = mapping->host;
2635 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
Badari Pulavarty027445c2006-09-30 23:28:46 -07002637 BUG_ON(iocb->ki_pos != pos);
2638
2639 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2640 &iocb->ki_pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
2642 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
Badari Pulavarty027445c2006-09-30 23:28:46 -07002643 ssize_t err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644
2645 err = sync_page_range_nolock(inode, mapping, pos, ret);
2646 if (err < 0)
2647 ret = err;
2648 }
2649 return ret;
2650}
Badari Pulavarty027445c2006-09-30 23:28:46 -07002651EXPORT_SYMBOL(generic_file_aio_write_nolock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Badari Pulavarty027445c2006-09-30 23:28:46 -07002653ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2654 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655{
2656 struct file *file = iocb->ki_filp;
2657 struct address_space *mapping = file->f_mapping;
2658 struct inode *inode = mapping->host;
2659 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
2661 BUG_ON(iocb->ki_pos != pos);
2662
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002663 mutex_lock(&inode->i_mutex);
Badari Pulavarty027445c2006-09-30 23:28:46 -07002664 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs,
2665 &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002666 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667
2668 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) {
2669 ssize_t err;
2670
2671 err = sync_page_range(inode, mapping, pos, ret);
2672 if (err < 0)
2673 ret = err;
2674 }
2675 return ret;
2676}
2677EXPORT_SYMBOL(generic_file_aio_write);
2678
David Howellscf9a2ae2006-08-29 19:05:54 +01002679/**
2680 * try_to_release_page() - release old fs-specific metadata on a page
2681 *
2682 * @page: the page which the kernel is trying to free
2683 * @gfp_mask: memory allocation flags (and I/O mode)
2684 *
2685 * The address_space is to try to release any data against the page
2686 * (presumably at page->private). If the release was successful, return `1'.
2687 * Otherwise return zero.
2688 *
2689 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002690 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002691 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002692 */
2693int try_to_release_page(struct page *page, gfp_t gfp_mask)
2694{
2695 struct address_space * const mapping = page->mapping;
2696
2697 BUG_ON(!PageLocked(page));
2698 if (PageWriteback(page))
2699 return 0;
2700
2701 if (mapping && mapping->a_ops->releasepage)
2702 return mapping->a_ops->releasepage(page, gfp_mask);
2703 return try_to_free_buffers(page);
2704}
2705
2706EXPORT_SYMBOL(try_to_release_page);