blob: ef169f37156da22cc3f07d5dd9092dea00315b7f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
13#include <linux/slab.h>
14#include <linux/compiler.h>
15#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070016#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080018#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/kernel_stat.h>
20#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
Jan Kara148f9482009-08-17 19:52:36 +020042#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
46/*
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
49 *
50 * Shared mappings now work. 15.8.1995 Bruno.
51 *
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 *
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
56 */
57
58/*
59 * Lock ordering:
60 *
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +100061 * ->i_mmap_lock (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070063 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080066 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * ->i_mmap_lock (truncate->unmap_mapping_range)
68 *
69 * ->mmap_sem
70 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070071 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
73 *
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
76 *
Nick Piggin82591e62006-10-19 23:29:10 -070077 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080080 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * ->i_alloc_sem (various)
82 *
83 * ->inode_lock
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
86 *
87 * ->i_mmap_lock
88 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080098 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
104 *
105 * ->task->proc_lock
106 * ->dcache_lock (proc_pid_lookup)
Andi Kleen6a460792009-09-16 11:50:15 +0200107 *
108 * (code doesn't rely on that order, so you could switch it around)
109 * ->tasklist_lock (memory_failure, collect_procs_ao)
110 * ->i_mmap_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 */
112
113/*
114 * Remove a page from the page cache and free it. Caller has to make
115 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700116 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 */
118void __remove_from_page_cache(struct page *page)
119{
120 struct address_space *mapping = page->mapping;
121
122 radix_tree_delete(&mapping->page_tree, page->index);
123 page->mapping = NULL;
124 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700125 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700126 if (PageSwapBacked(page))
127 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700128 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800129
130 /*
131 * Some filesystems seem to re-dirty the page even after
132 * the VM has canceled the dirty bit (eg ext3 journaling).
133 *
134 * Fix it up by doing a final dirty accounting check after
135 * having removed the page entirely.
136 */
137 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
138 dec_zone_page_state(page, NR_FILE_DIRTY);
139 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
143void remove_from_page_cache(struct page *page)
144{
145 struct address_space *mapping = page->mapping;
146
Matt Mackallcd7619d2005-05-01 08:59:01 -0700147 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
Nick Piggin19fd6232008-07-25 19:45:32 -0700149 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700151 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700152 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
154
155static int sync_page(void *word)
156{
157 struct address_space *mapping;
158 struct page *page;
159
Andi Kleen07808b72005-11-05 17:25:53 +0100160 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700163 * page_mapping() is being called without PG_locked held.
164 * Some knowledge of the state and use of the page is used to
165 * reduce the requirements down to a memory barrier.
166 * The danger here is of a stale page_mapping() return value
167 * indicating a struct address_space different from the one it's
168 * associated with when it is associated with one.
169 * After smp_mb(), it's either the correct page_mapping() for
170 * the page, or an old page_mapping() and the page's own
171 * page_mapping() has gone NULL.
172 * The ->sync_page() address_space operation must tolerate
173 * page_mapping() going NULL. By an amazing coincidence,
174 * this comes about because none of the users of the page
175 * in the ->sync_page() methods make essential use of the
176 * page_mapping(), merely passing the page down to the backing
177 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700178 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700179 * of interest. When page_mapping() does go NULL, the entire
180 * call stack gracefully ignores the page and returns.
181 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
183 smp_mb();
184 mapping = page_mapping(page);
185 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
186 mapping->a_ops->sync_page(page);
187 io_schedule();
188 return 0;
189}
190
Matthew Wilcox2687a352007-12-06 11:18:49 -0500191static int sync_page_killable(void *word)
192{
193 sync_page(word);
194 return fatal_signal_pending(current) ? -EINTR : 0;
195}
196
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700198 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700199 * @mapping: address space structure to write
200 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800201 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700202 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700204 * Start writeback against all of a mapping's dirty pages that lie
205 * within the byte offsets <start, end> inclusive.
206 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700208 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 * these two operations is that if a dirty page/buffer is encountered, it must
210 * be waited upon, and not just skipped over.
211 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800212int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
213 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
215 int ret;
216 struct writeback_control wbc = {
217 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800218 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700219 .range_start = start,
220 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 };
222
223 if (!mapping_cap_writeback_dirty(mapping))
224 return 0;
225
226 ret = do_writepages(mapping, &wbc);
227 return ret;
228}
229
230static inline int __filemap_fdatawrite(struct address_space *mapping,
231 int sync_mode)
232{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700233 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236int filemap_fdatawrite(struct address_space *mapping)
237{
238 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
239}
240EXPORT_SYMBOL(filemap_fdatawrite);
241
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400242int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800243 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244{
245 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
246}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400247EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Randy Dunlap485bb992006-06-23 02:03:49 -0700249/**
250 * filemap_flush - mostly a non-blocking flush
251 * @mapping: target address_space
252 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 * This is a mostly non-blocking flush. Not suitable for data-integrity
254 * purposes - I/O may not be started against all dirty pages.
255 */
256int filemap_flush(struct address_space *mapping)
257{
258 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
259}
260EXPORT_SYMBOL(filemap_flush);
261
Randy Dunlap485bb992006-06-23 02:03:49 -0700262/**
263 * wait_on_page_writeback_range - wait for writeback to complete
264 * @mapping: target address_space
265 * @start: beginning page index
266 * @end: ending page index
267 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 * Wait for writeback to complete against pages indexed by start->end
269 * inclusive
270 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800271int wait_on_page_writeback_range(struct address_space *mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 pgoff_t start, pgoff_t end)
273{
274 struct pagevec pvec;
275 int nr_pages;
276 int ret = 0;
277 pgoff_t index;
278
279 if (end < start)
280 return 0;
281
282 pagevec_init(&pvec, 0);
283 index = start;
284 while ((index <= end) &&
285 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
286 PAGECACHE_TAG_WRITEBACK,
287 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
288 unsigned i;
289
290 for (i = 0; i < nr_pages; i++) {
291 struct page *page = pvec.pages[i];
292
293 /* until radix tree lookup accepts end_index */
294 if (page->index > end)
295 continue;
296
297 wait_on_page_writeback(page);
298 if (PageError(page))
299 ret = -EIO;
300 }
301 pagevec_release(&pvec);
302 cond_resched();
303 }
304
305 /* Check for outstanding write errors */
306 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
307 ret = -ENOSPC;
308 if (test_and_clear_bit(AS_EIO, &mapping->flags))
309 ret = -EIO;
310
311 return ret;
312}
313
Randy Dunlap485bb992006-06-23 02:03:49 -0700314/**
Jan Karad3bccb62009-08-17 19:30:27 +0200315 * filemap_fdatawait_range - wait for all under-writeback pages to complete in a given range
316 * @mapping: address space structure to wait for
317 * @start: offset in bytes where the range starts
318 * @end: offset in bytes where the range ends (inclusive)
319 *
320 * Walk the list of under-writeback pages of the given address space
321 * in the given range and wait for all of them.
322 *
323 * This is just a simple wrapper so that callers don't have to convert offsets
324 * to page indexes themselves
325 */
326int filemap_fdatawait_range(struct address_space *mapping, loff_t start,
327 loff_t end)
328{
329 return wait_on_page_writeback_range(mapping, start >> PAGE_CACHE_SHIFT,
330 end >> PAGE_CACHE_SHIFT);
331}
332EXPORT_SYMBOL(filemap_fdatawait_range);
333
334/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700335 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700337 *
338 * Walk the list of under-writeback pages of the given address space
339 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
341int filemap_fdatawait(struct address_space *mapping)
342{
343 loff_t i_size = i_size_read(mapping->host);
344
345 if (i_size == 0)
346 return 0;
347
348 return wait_on_page_writeback_range(mapping, 0,
349 (i_size - 1) >> PAGE_CACHE_SHIFT);
350}
351EXPORT_SYMBOL(filemap_fdatawait);
352
353int filemap_write_and_wait(struct address_space *mapping)
354{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800355 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356
357 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800358 err = filemap_fdatawrite(mapping);
359 /*
360 * Even if the above returned error, the pages may be
361 * written partially (e.g. -ENOSPC), so we wait for it.
362 * But the -EIO is special case, it may indicate the worst
363 * thing (e.g. bug) happened, so we avoid waiting for it.
364 */
365 if (err != -EIO) {
366 int err2 = filemap_fdatawait(mapping);
367 if (!err)
368 err = err2;
369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800371 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800373EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
Randy Dunlap485bb992006-06-23 02:03:49 -0700375/**
376 * filemap_write_and_wait_range - write out & wait on a file range
377 * @mapping: the address_space for the pages
378 * @lstart: offset in bytes where the range starts
379 * @lend: offset in bytes where the range ends (inclusive)
380 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800381 * Write out and wait upon file offsets lstart->lend, inclusive.
382 *
383 * Note that `lend' is inclusive (describes the last byte to be written) so
384 * that this function can be used to write to the very end-of-file (end = -1).
385 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386int filemap_write_and_wait_range(struct address_space *mapping,
387 loff_t lstart, loff_t lend)
388{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800389 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800392 err = __filemap_fdatawrite_range(mapping, lstart, lend,
393 WB_SYNC_ALL);
394 /* See comment of filemap_write_and_wait() */
395 if (err != -EIO) {
396 int err2 = wait_on_page_writeback_range(mapping,
397 lstart >> PAGE_CACHE_SHIFT,
398 lend >> PAGE_CACHE_SHIFT);
399 if (!err)
400 err = err2;
401 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800403 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404}
Chris Masonf6995582009-04-15 13:22:37 -0400405EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Randy Dunlap485bb992006-06-23 02:03:49 -0700407/**
Nick Piggine2867812008-07-25 19:45:30 -0700408 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700409 * @page: page to add
410 * @mapping: the page's address_space
411 * @offset: page index
412 * @gfp_mask: page allocation mode
413 *
Nick Piggine2867812008-07-25 19:45:30 -0700414 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 * This function does not add the page to the LRU. The caller must do that.
416 */
Nick Piggine2867812008-07-25 19:45:30 -0700417int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400418 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419{
Nick Piggine2867812008-07-25 19:45:30 -0700420 int error;
421
422 VM_BUG_ON(!PageLocked(page));
423
424 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800425 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800426 if (error)
427 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428
Balbir Singh35c754d2008-02-07 00:14:05 -0800429 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700431 page_cache_get(page);
432 page->mapping = mapping;
433 page->index = offset;
434
Nick Piggin19fd6232008-07-25 19:45:32 -0700435 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700437 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700439 __inc_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700440 if (PageSwapBacked(page))
441 __inc_zone_page_state(page, NR_SHMEM);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700442 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700443 } else {
444 page->mapping = NULL;
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700445 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700446 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700447 page_cache_release(page);
448 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800450 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700451 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800452out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 return error;
454}
Nick Piggine2867812008-07-25 19:45:30 -0700455EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400458 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700460 int ret;
461
462 /*
463 * Splice_read and readahead add shmem/tmpfs pages into the page cache
464 * before shmem_readpage has a chance to mark them as SwapBacked: they
465 * need to go on the active_anon lru below, and mem_cgroup_cache_charge
466 * (called in add_to_page_cache) needs to know where they're going too.
467 */
468 if (mapping_cap_swap_backed(mapping))
469 SetPageSwapBacked(page);
470
471 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
472 if (ret == 0) {
473 if (page_is_file_cache(page))
474 lru_cache_add_file(page);
475 else
476 lru_cache_add_active_anon(page);
477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 return ret;
479}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300480EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Paul Jackson44110fe2006-03-24 03:16:04 -0800482#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700483struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800484{
485 if (cpuset_do_page_mem_spread()) {
486 int n = cpuset_mem_spread_node();
Mel Gorman6484eb32009-06-16 15:31:54 -0700487 return alloc_pages_exact_node(n, gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800488 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700489 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800490}
Nick Piggin2ae88142006-10-28 10:38:23 -0700491EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800492#endif
493
Nick Piggindb376482006-09-25 23:31:24 -0700494static int __sleep_on_page_lock(void *word)
495{
496 io_schedule();
497 return 0;
498}
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500/*
501 * In order to wait for pages to become available there must be
502 * waitqueues associated with pages. By using a hash table of
503 * waitqueues where the bucket discipline is to maintain all
504 * waiters on the same queue and wake all when any of the pages
505 * become available, and for the woken contexts to check to be
506 * sure the appropriate page became available, this saves space
507 * at a cost of "thundering herd" phenomena during rare hash
508 * collisions.
509 */
510static wait_queue_head_t *page_waitqueue(struct page *page)
511{
512 const struct zone *zone = page_zone(page);
513
514 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
515}
516
517static inline void wake_up_page(struct page *page, int bit)
518{
519 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
520}
521
Harvey Harrison920c7a52008-02-04 22:29:26 -0800522void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
524 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
525
526 if (test_bit(bit_nr, &page->flags))
527 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
528 TASK_UNINTERRUPTIBLE);
529}
530EXPORT_SYMBOL(wait_on_page_bit);
531
532/**
David Howells385e1ca2009-04-03 16:42:39 +0100533 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700534 * @page: Page defining the wait queue of interest
535 * @waiter: Waiter to add to the queue
David Howells385e1ca2009-04-03 16:42:39 +0100536 *
537 * Add an arbitrary @waiter to the wait queue for the nominated @page.
538 */
539void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
540{
541 wait_queue_head_t *q = page_waitqueue(page);
542 unsigned long flags;
543
544 spin_lock_irqsave(&q->lock, flags);
545 __add_wait_queue(q, waiter);
546 spin_unlock_irqrestore(&q->lock, flags);
547}
548EXPORT_SYMBOL_GPL(add_page_wait_queue);
549
550/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700551 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 * @page: the page
553 *
554 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
555 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
556 * mechananism between PageLocked pages and PageWriteback pages is shared.
557 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
558 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700559 * The mb is necessary to enforce ordering between the clear_bit and the read
560 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800562void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
Nick Piggin8413ac92008-10-18 20:26:59 -0700564 VM_BUG_ON(!PageLocked(page));
565 clear_bit_unlock(PG_locked, &page->flags);
566 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 wake_up_page(page, PG_locked);
568}
569EXPORT_SYMBOL(unlock_page);
570
Randy Dunlap485bb992006-06-23 02:03:49 -0700571/**
572 * end_page_writeback - end writeback against a page
573 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 */
575void end_page_writeback(struct page *page)
576{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700577 if (TestClearPageReclaim(page))
578 rotate_reclaimable_page(page);
579
580 if (!test_clear_page_writeback(page))
581 BUG();
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 smp_mb__after_clear_bit();
584 wake_up_page(page, PG_writeback);
585}
586EXPORT_SYMBOL(end_page_writeback);
587
Randy Dunlap485bb992006-06-23 02:03:49 -0700588/**
589 * __lock_page - get a lock on the page, assuming we need to sleep to get it
590 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700592 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
594 * chances are that on the second loop, the block layer's plug list is empty,
595 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
596 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800597void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598{
599 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
600
601 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
602 TASK_UNINTERRUPTIBLE);
603}
604EXPORT_SYMBOL(__lock_page);
605
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800606int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500607{
608 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
609
610 return __wait_on_bit_lock(page_waitqueue(page), &wait,
611 sync_page_killable, TASK_KILLABLE);
612}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300613EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500614
Randy Dunlap76824862008-03-19 17:00:40 -0700615/**
616 * __lock_page_nosync - get a lock on the page, without calling sync_page()
617 * @page: the page to lock
618 *
Nick Piggindb376482006-09-25 23:31:24 -0700619 * Variant of lock_page that does not require the caller to hold a reference
620 * on the page's mapping.
621 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800622void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700623{
624 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
625 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
626 TASK_UNINTERRUPTIBLE);
627}
628
Randy Dunlap485bb992006-06-23 02:03:49 -0700629/**
630 * find_get_page - find and get a page reference
631 * @mapping: the address_space to search
632 * @offset: the page index
633 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700634 * Is there a pagecache struct page at the given (mapping, offset) tuple?
635 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 */
Nick Piggina60637c2008-07-25 19:45:31 -0700637struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
Nick Piggina60637c2008-07-25 19:45:31 -0700639 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 struct page *page;
641
Nick Piggina60637c2008-07-25 19:45:31 -0700642 rcu_read_lock();
643repeat:
644 page = NULL;
645 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
646 if (pagep) {
647 page = radix_tree_deref_slot(pagep);
648 if (unlikely(!page || page == RADIX_TREE_RETRY))
649 goto repeat;
650
651 if (!page_cache_get_speculative(page))
652 goto repeat;
653
654 /*
655 * Has the page moved?
656 * This is part of the lockless pagecache protocol. See
657 * include/linux/pagemap.h for details.
658 */
659 if (unlikely(page != *pagep)) {
660 page_cache_release(page);
661 goto repeat;
662 }
663 }
664 rcu_read_unlock();
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 return page;
667}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668EXPORT_SYMBOL(find_get_page);
669
Randy Dunlap485bb992006-06-23 02:03:49 -0700670/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700672 * @mapping: the address_space to search
673 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 *
675 * Locates the desired pagecache page, locks it, increments its reference
676 * count and returns its address.
677 *
678 * Returns zero if the page was not present. find_lock_page() may sleep.
679 */
Nick Piggina60637c2008-07-25 19:45:31 -0700680struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681{
682 struct page *page;
683
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700685 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700687 lock_page(page);
688 /* Has the page been truncated? */
689 if (unlikely(page->mapping != mapping)) {
690 unlock_page(page);
691 page_cache_release(page);
692 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 }
Nick Piggina60637c2008-07-25 19:45:31 -0700694 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 return page;
697}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698EXPORT_SYMBOL(find_lock_page);
699
700/**
701 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700702 * @mapping: the page's address_space
703 * @index: the page's index into the mapping
704 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 *
706 * Locates a page in the pagecache. If the page is not present, a new page
707 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
708 * LRU list. The returned page is locked and has its reference count
709 * incremented.
710 *
711 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
712 * allocation!
713 *
714 * find_or_create_page() returns the desired page's address, or zero on
715 * memory exhaustion.
716 */
717struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700718 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Nick Piggineb2be182007-10-16 01:24:57 -0700720 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 int err;
722repeat:
723 page = find_lock_page(mapping, index);
724 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700725 page = __page_cache_alloc(gfp_mask);
726 if (!page)
727 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800728 /*
729 * We want a regular kernel memory (not highmem or DMA etc)
730 * allocation for the radix tree nodes, but we need to honour
731 * the context-specific requirements the caller has asked for.
732 * GFP_RECLAIM_MASK collects those requirements.
733 */
734 err = add_to_page_cache_lru(page, mapping, index,
735 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700736 if (unlikely(err)) {
737 page_cache_release(page);
738 page = NULL;
739 if (err == -EEXIST)
740 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 return page;
744}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745EXPORT_SYMBOL(find_or_create_page);
746
747/**
748 * find_get_pages - gang pagecache lookup
749 * @mapping: The address_space to search
750 * @start: The starting page index
751 * @nr_pages: The maximum number of pages
752 * @pages: Where the resulting pages are placed
753 *
754 * find_get_pages() will search for and return a group of up to
755 * @nr_pages pages in the mapping. The pages are placed at @pages.
756 * find_get_pages() takes a reference against the returned pages.
757 *
758 * The search returns a group of mapping-contiguous pages with ascending
759 * indexes. There may be holes in the indices due to not-present pages.
760 *
761 * find_get_pages() returns the number of pages which were found.
762 */
763unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
764 unsigned int nr_pages, struct page **pages)
765{
766 unsigned int i;
767 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700768 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Nick Piggina60637c2008-07-25 19:45:31 -0700770 rcu_read_lock();
771restart:
772 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
773 (void ***)pages, start, nr_pages);
774 ret = 0;
775 for (i = 0; i < nr_found; i++) {
776 struct page *page;
777repeat:
778 page = radix_tree_deref_slot((void **)pages[i]);
779 if (unlikely(!page))
780 continue;
781 /*
782 * this can only trigger if nr_found == 1, making livelock
783 * a non issue.
784 */
785 if (unlikely(page == RADIX_TREE_RETRY))
786 goto restart;
787
788 if (!page_cache_get_speculative(page))
789 goto repeat;
790
791 /* Has the page moved? */
792 if (unlikely(page != *((void **)pages[i]))) {
793 page_cache_release(page);
794 goto repeat;
795 }
796
797 pages[ret] = page;
798 ret++;
799 }
800 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 return ret;
802}
803
Jens Axboeebf43502006-04-27 08:46:01 +0200804/**
805 * find_get_pages_contig - gang contiguous pagecache lookup
806 * @mapping: The address_space to search
807 * @index: The starting page index
808 * @nr_pages: The maximum number of pages
809 * @pages: Where the resulting pages are placed
810 *
811 * find_get_pages_contig() works exactly like find_get_pages(), except
812 * that the returned number of pages are guaranteed to be contiguous.
813 *
814 * find_get_pages_contig() returns the number of pages which were found.
815 */
816unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
817 unsigned int nr_pages, struct page **pages)
818{
819 unsigned int i;
820 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700821 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200822
Nick Piggina60637c2008-07-25 19:45:31 -0700823 rcu_read_lock();
824restart:
825 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
826 (void ***)pages, index, nr_pages);
827 ret = 0;
828 for (i = 0; i < nr_found; i++) {
829 struct page *page;
830repeat:
831 page = radix_tree_deref_slot((void **)pages[i]);
832 if (unlikely(!page))
833 continue;
834 /*
835 * this can only trigger if nr_found == 1, making livelock
836 * a non issue.
837 */
838 if (unlikely(page == RADIX_TREE_RETRY))
839 goto restart;
840
841 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200842 break;
843
Nick Piggina60637c2008-07-25 19:45:31 -0700844 if (!page_cache_get_speculative(page))
845 goto repeat;
846
847 /* Has the page moved? */
848 if (unlikely(page != *((void **)pages[i]))) {
849 page_cache_release(page);
850 goto repeat;
851 }
852
853 pages[ret] = page;
854 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200855 index++;
856 }
Nick Piggina60637c2008-07-25 19:45:31 -0700857 rcu_read_unlock();
858 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200859}
David Howellsef71c152007-05-09 02:33:44 -0700860EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200861
Randy Dunlap485bb992006-06-23 02:03:49 -0700862/**
863 * find_get_pages_tag - find and return pages that match @tag
864 * @mapping: the address_space to search
865 * @index: the starting page index
866 * @tag: the tag index
867 * @nr_pages: the maximum number of pages
868 * @pages: where the resulting pages are placed
869 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700871 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 */
873unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
874 int tag, unsigned int nr_pages, struct page **pages)
875{
876 unsigned int i;
877 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700878 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Nick Piggina60637c2008-07-25 19:45:31 -0700880 rcu_read_lock();
881restart:
882 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
883 (void ***)pages, *index, nr_pages, tag);
884 ret = 0;
885 for (i = 0; i < nr_found; i++) {
886 struct page *page;
887repeat:
888 page = radix_tree_deref_slot((void **)pages[i]);
889 if (unlikely(!page))
890 continue;
891 /*
892 * this can only trigger if nr_found == 1, making livelock
893 * a non issue.
894 */
895 if (unlikely(page == RADIX_TREE_RETRY))
896 goto restart;
897
898 if (!page_cache_get_speculative(page))
899 goto repeat;
900
901 /* Has the page moved? */
902 if (unlikely(page != *((void **)pages[i]))) {
903 page_cache_release(page);
904 goto repeat;
905 }
906
907 pages[ret] = page;
908 ret++;
909 }
910 rcu_read_unlock();
911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (ret)
913 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700914
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 return ret;
916}
David Howellsef71c152007-05-09 02:33:44 -0700917EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
Randy Dunlap485bb992006-06-23 02:03:49 -0700919/**
920 * grab_cache_page_nowait - returns locked page at given index in given cache
921 * @mapping: target address_space
922 * @index: the page index
923 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800924 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 * This is intended for speculative data generators, where the data can
926 * be regenerated if the page couldn't be grabbed. This routine should
927 * be safe to call while holding the lock for another page.
928 *
929 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
930 * and deadlock against the caller's locked page.
931 */
932struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700933grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
935 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200938 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return page;
940 page_cache_release(page);
941 return NULL;
942 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700943 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800944 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 page_cache_release(page);
946 page = NULL;
947 }
948 return page;
949}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950EXPORT_SYMBOL(grab_cache_page_nowait);
951
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700952/*
953 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
954 * a _large_ part of the i/o request. Imagine the worst scenario:
955 *
956 * ---R__________________________________________B__________
957 * ^ reading here ^ bad block(assume 4k)
958 *
959 * read(R) => miss => readahead(R...B) => media error => frustrating retries
960 * => failing the whole request => read(R) => read(R+1) =>
961 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
962 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
963 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
964 *
965 * It is going insane. Fix it by quickly scaling down the readahead size.
966 */
967static void shrink_readahead_size_eio(struct file *filp,
968 struct file_ra_state *ra)
969{
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700970 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700971}
972
Randy Dunlap485bb992006-06-23 02:03:49 -0700973/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800974 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700975 * @filp: the file to read
976 * @ppos: current file position
977 * @desc: read_descriptor
978 * @actor: read method
979 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700981 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 *
983 * This is really ugly. But the goto's actually try to clarify some
984 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800986static void do_generic_file_read(struct file *filp, loff_t *ppos,
987 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800989 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800991 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700992 pgoff_t index;
993 pgoff_t last_index;
994 pgoff_t prev_index;
995 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700996 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -07001000 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1001 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1003 offset = *ppos & ~PAGE_CACHE_MASK;
1004
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 for (;;) {
1006 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001007 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001008 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 unsigned long nr, ret;
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012find_page:
1013 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001014 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001015 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001016 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001017 index, last_index - index);
1018 page = find_get_page(mapping, index);
1019 if (unlikely(page == NULL))
1020 goto no_cached_page;
1021 }
1022 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001023 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001024 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001025 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001027 if (!PageUptodate(page)) {
1028 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1029 !mapping->a_ops->is_partially_uptodate)
1030 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001031 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001032 goto page_not_up_to_date;
1033 if (!mapping->a_ops->is_partially_uptodate(page,
1034 desc, offset))
1035 goto page_not_up_to_date_locked;
1036 unlock_page(page);
1037 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001039 /*
1040 * i_size must be checked after we know the page is Uptodate.
1041 *
1042 * Checking i_size after the check allows us to calculate
1043 * the correct value for "nr", which means the zero-filled
1044 * part of the page is not copied back to userspace (unless
1045 * another truncate extends the file - this is desired though).
1046 */
1047
1048 isize = i_size_read(inode);
1049 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1050 if (unlikely(!isize || index > end_index)) {
1051 page_cache_release(page);
1052 goto out;
1053 }
1054
1055 /* nr is the maximum number of bytes to copy from this page */
1056 nr = PAGE_CACHE_SIZE;
1057 if (index == end_index) {
1058 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1059 if (nr <= offset) {
1060 page_cache_release(page);
1061 goto out;
1062 }
1063 }
1064 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
1066 /* If users can be writing to this page using arbitrary
1067 * virtual addresses, take care about potential aliasing
1068 * before reading the page on the kernel side.
1069 */
1070 if (mapping_writably_mapped(mapping))
1071 flush_dcache_page(page);
1072
1073 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001074 * When a sequential read accesses a page several times,
1075 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 */
Jan Karaec0f1632007-05-06 14:49:25 -07001077 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 mark_page_accessed(page);
1079 prev_index = index;
1080
1081 /*
1082 * Ok, we have the page, and it's up-to-date, so
1083 * now we can copy it to user space...
1084 *
1085 * The actor routine returns how many bytes were actually used..
1086 * NOTE! This may not be the same as how much of a user buffer
1087 * we filled up (we may be padding etc), so we can only update
1088 * "pos" here (the actor routine has to update the user buffer
1089 * pointers and the remaining count).
1090 */
1091 ret = actor(desc, page, offset, nr);
1092 offset += ret;
1093 index += offset >> PAGE_CACHE_SHIFT;
1094 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001095 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096
1097 page_cache_release(page);
1098 if (ret == nr && desc->count)
1099 continue;
1100 goto out;
1101
1102page_not_up_to_date:
1103 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001104 error = lock_page_killable(page);
1105 if (unlikely(error))
1106 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001108page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001109 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (!page->mapping) {
1111 unlock_page(page);
1112 page_cache_release(page);
1113 continue;
1114 }
1115
1116 /* Did somebody else fill it already? */
1117 if (PageUptodate(page)) {
1118 unlock_page(page);
1119 goto page_ok;
1120 }
1121
1122readpage:
1123 /* Start the actual read. The read will unlock the page. */
1124 error = mapping->a_ops->readpage(filp, page);
1125
Zach Brown994fc28c2005-12-15 14:28:17 -08001126 if (unlikely(error)) {
1127 if (error == AOP_TRUNCATED_PAGE) {
1128 page_cache_release(page);
1129 goto find_page;
1130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133
1134 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001135 error = lock_page_killable(page);
1136 if (unlikely(error))
1137 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 if (!PageUptodate(page)) {
1139 if (page->mapping == NULL) {
1140 /*
1141 * invalidate_inode_pages got it
1142 */
1143 unlock_page(page);
1144 page_cache_release(page);
1145 goto find_page;
1146 }
1147 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001148 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001149 error = -EIO;
1150 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 }
1152 unlock_page(page);
1153 }
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 goto page_ok;
1156
1157readpage_error:
1158 /* UHHUH! A synchronous read error occurred. Report it */
1159 desc->error = error;
1160 page_cache_release(page);
1161 goto out;
1162
1163no_cached_page:
1164 /*
1165 * Ok, it wasn't cached, so we need to create a new
1166 * page..
1167 */
Nick Piggineb2be182007-10-16 01:24:57 -07001168 page = page_cache_alloc_cold(mapping);
1169 if (!page) {
1170 desc->error = -ENOMEM;
1171 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 }
Nick Piggineb2be182007-10-16 01:24:57 -07001173 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 index, GFP_KERNEL);
1175 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001176 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 if (error == -EEXIST)
1178 goto find_page;
1179 desc->error = error;
1180 goto out;
1181 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 goto readpage;
1183 }
1184
1185out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001186 ra->prev_pos = prev_index;
1187 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1188 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001190 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001191 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194int file_read_actor(read_descriptor_t *desc, struct page *page,
1195 unsigned long offset, unsigned long size)
1196{
1197 char *kaddr;
1198 unsigned long left, count = desc->count;
1199
1200 if (size > count)
1201 size = count;
1202
1203 /*
1204 * Faults on the destination of a read are common, so do it before
1205 * taking the kmap.
1206 */
1207 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1208 kaddr = kmap_atomic(page, KM_USER0);
1209 left = __copy_to_user_inatomic(desc->arg.buf,
1210 kaddr + offset, size);
1211 kunmap_atomic(kaddr, KM_USER0);
1212 if (left == 0)
1213 goto success;
1214 }
1215
1216 /* Do it the slow way */
1217 kaddr = kmap(page);
1218 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1219 kunmap(page);
1220
1221 if (left) {
1222 size -= left;
1223 desc->error = -EFAULT;
1224 }
1225success:
1226 desc->count = count - size;
1227 desc->written += size;
1228 desc->arg.buf += size;
1229 return size;
1230}
1231
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001232/*
1233 * Performs necessary checks before doing a write
1234 * @iov: io vector request
1235 * @nr_segs: number of segments in the iovec
1236 * @count: number of bytes to write
1237 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1238 *
1239 * Adjust number of segments and amount of bytes to write (nr_segs should be
1240 * properly initialized first). Returns appropriate error code that caller
1241 * should return or zero in case that write should be allowed.
1242 */
1243int generic_segment_checks(const struct iovec *iov,
1244 unsigned long *nr_segs, size_t *count, int access_flags)
1245{
1246 unsigned long seg;
1247 size_t cnt = 0;
1248 for (seg = 0; seg < *nr_segs; seg++) {
1249 const struct iovec *iv = &iov[seg];
1250
1251 /*
1252 * If any segment has a negative length, or the cumulative
1253 * length ever wraps negative then return -EINVAL.
1254 */
1255 cnt += iv->iov_len;
1256 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1257 return -EINVAL;
1258 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1259 continue;
1260 if (seg == 0)
1261 return -EFAULT;
1262 *nr_segs = seg;
1263 cnt -= iv->iov_len; /* This segment is no good */
1264 break;
1265 }
1266 *count = cnt;
1267 return 0;
1268}
1269EXPORT_SYMBOL(generic_segment_checks);
1270
Randy Dunlap485bb992006-06-23 02:03:49 -07001271/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001272 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001273 * @iocb: kernel I/O control block
1274 * @iov: io vector request
1275 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001276 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001277 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 * This is the "read()" routine for all filesystems
1279 * that can use the page cache directly.
1280 */
1281ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001282generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1283 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284{
1285 struct file *filp = iocb->ki_filp;
1286 ssize_t retval;
1287 unsigned long seg;
1288 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001289 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
1291 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001292 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1293 if (retval)
1294 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1297 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001298 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 struct address_space *mapping;
1300 struct inode *inode;
1301
1302 mapping = filp->f_mapping;
1303 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 if (!count)
1305 goto out; /* skip atime */
1306 size = i_size_read(inode);
1307 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001308 retval = filemap_write_and_wait_range(mapping, pos,
1309 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001310 if (!retval) {
1311 retval = mapping->a_ops->direct_IO(READ, iocb,
1312 iov, pos, nr_segs);
1313 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (retval > 0)
1315 *ppos = pos + retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001316 if (retval) {
1317 file_accessed(filp);
1318 goto out;
1319 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001320 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 }
1322
Hugh Dickins11fa9772008-07-23 21:27:34 -07001323 for (seg = 0; seg < nr_segs; seg++) {
1324 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325
Hugh Dickins11fa9772008-07-23 21:27:34 -07001326 desc.written = 0;
1327 desc.arg.buf = iov[seg].iov_base;
1328 desc.count = iov[seg].iov_len;
1329 if (desc.count == 0)
1330 continue;
1331 desc.error = 0;
1332 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1333 retval += desc.written;
1334 if (desc.error) {
1335 retval = retval ?: desc.error;
1336 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001338 if (desc.count > 0)
1339 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 }
1341out:
1342 return retval;
1343}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344EXPORT_SYMBOL(generic_file_aio_read);
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346static ssize_t
1347do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001348 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
1350 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1351 return -EINVAL;
1352
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001353 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 return 0;
1355}
1356
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001357SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358{
1359 ssize_t ret;
1360 struct file *file;
1361
1362 ret = -EBADF;
1363 file = fget(fd);
1364 if (file) {
1365 if (file->f_mode & FMODE_READ) {
1366 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001367 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1368 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 unsigned long len = end - start + 1;
1370 ret = do_readahead(mapping, file, start, len);
1371 }
1372 fput(file);
1373 }
1374 return ret;
1375}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001376#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1377asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1378{
1379 return SYSC_readahead((int) fd, offset, (size_t) count);
1380}
1381SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1382#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
1384#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001385/**
1386 * page_cache_read - adds requested page to the page cache if not already there
1387 * @file: file to read
1388 * @offset: page index
1389 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 * This adds the requested page to the page cache if it isn't already there,
1391 * and schedules an I/O to read in its contents from disk.
1392 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001393static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394{
1395 struct address_space *mapping = file->f_mapping;
1396 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001397 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Zach Brown994fc28c2005-12-15 14:28:17 -08001399 do {
1400 page = page_cache_alloc_cold(mapping);
1401 if (!page)
1402 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Zach Brown994fc28c2005-12-15 14:28:17 -08001404 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1405 if (ret == 0)
1406 ret = mapping->a_ops->readpage(file, page);
1407 else if (ret == -EEXIST)
1408 ret = 0; /* losing race to add is OK */
1409
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
Zach Brown994fc28c2005-12-15 14:28:17 -08001412 } while (ret == AOP_TRUNCATED_PAGE);
1413
1414 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415}
1416
1417#define MMAP_LOTSAMISS (100)
1418
Linus Torvaldsef00e082009-06-16 15:31:25 -07001419/*
1420 * Synchronous readahead happens when we don't even find
1421 * a page in the page cache at all.
1422 */
1423static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1424 struct file_ra_state *ra,
1425 struct file *file,
1426 pgoff_t offset)
1427{
1428 unsigned long ra_pages;
1429 struct address_space *mapping = file->f_mapping;
1430
1431 /* If we don't want any read-ahead, don't bother */
1432 if (VM_RandomReadHint(vma))
1433 return;
1434
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001435 if (VM_SequentialReadHint(vma) ||
1436 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001437 page_cache_sync_readahead(mapping, ra, file, offset,
1438 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001439 return;
1440 }
1441
1442 if (ra->mmap_miss < INT_MAX)
1443 ra->mmap_miss++;
1444
1445 /*
1446 * Do we miss much more than hit in this file? If so,
1447 * stop bothering with read-ahead. It will only hurt.
1448 */
1449 if (ra->mmap_miss > MMAP_LOTSAMISS)
1450 return;
1451
Wu Fengguangd30a1102009-06-16 15:31:30 -07001452 /*
1453 * mmap read-around
1454 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001455 ra_pages = max_sane_readahead(ra->ra_pages);
1456 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001457 ra->start = max_t(long, 0, offset - ra_pages/2);
1458 ra->size = ra_pages;
1459 ra->async_size = 0;
1460 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001461 }
1462}
1463
1464/*
1465 * Asynchronous readahead happens when we find the page and PG_readahead,
1466 * so we want to possibly extend the readahead further..
1467 */
1468static void do_async_mmap_readahead(struct vm_area_struct *vma,
1469 struct file_ra_state *ra,
1470 struct file *file,
1471 struct page *page,
1472 pgoff_t offset)
1473{
1474 struct address_space *mapping = file->f_mapping;
1475
1476 /* If we don't want any read-ahead, don't bother */
1477 if (VM_RandomReadHint(vma))
1478 return;
1479 if (ra->mmap_miss > 0)
1480 ra->mmap_miss--;
1481 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001482 page_cache_async_readahead(mapping, ra, file,
1483 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001484}
1485
Randy Dunlap485bb992006-06-23 02:03:49 -07001486/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001487 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001488 * @vma: vma in which the fault was taken
1489 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001490 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001491 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 * mapped memory region to read in file data during a page fault.
1493 *
1494 * The goto's are kind of ugly, but this streamlines the normal case of having
1495 * it in the page cache, and handles the special cases reasonably without
1496 * having a lot of duplicated code.
1497 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001498int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
1500 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001501 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 struct address_space *mapping = file->f_mapping;
1503 struct file_ra_state *ra = &file->f_ra;
1504 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001505 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001507 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001508 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001511 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001512 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 * Do we have something in the page cache already?
1516 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001517 page = find_get_page(mapping, offset);
1518 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001520 * We found the page, so try async readahead before
1521 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001523 do_async_mmap_readahead(vma, ra, file, page, offset);
1524 lock_page(page);
1525
1526 /* Did it get truncated? */
1527 if (unlikely(page->mapping != mapping)) {
1528 unlock_page(page);
1529 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 goto no_cached_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 }
Linus Torvaldsef00e082009-06-16 15:31:25 -07001532 } else {
1533 /* No page in the page cache at all */
1534 do_sync_mmap_readahead(vma, ra, file, offset);
1535 count_vm_event(PGMAJFAULT);
1536 ret = VM_FAULT_MAJOR;
1537retry_find:
1538 page = find_lock_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (!page)
1540 goto no_cached_page;
1541 }
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001544 * We have a locked page in the page cache, now we need to check
1545 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 */
Nick Piggind00806b2007-07-19 01:46:57 -07001547 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 goto page_not_uptodate;
1549
Linus Torvaldsef00e082009-06-16 15:31:25 -07001550 /*
1551 * Found the page and have a reference on it.
1552 * We must recheck i_size under page lock.
1553 */
Nick Piggind00806b2007-07-19 01:46:57 -07001554 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001555 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001556 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001557 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001558 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001559 }
1560
Linus Torvaldsef00e082009-06-16 15:31:25 -07001561 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001562 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001563 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565no_cached_page:
1566 /*
1567 * We're only likely to ever get here if MADV_RANDOM is in
1568 * effect.
1569 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001570 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572 /*
1573 * The page we want has now been added to the page cache.
1574 * In the unlikely event that someone removed it in the
1575 * meantime, we'll just come back here and read it again.
1576 */
1577 if (error >= 0)
1578 goto retry_find;
1579
1580 /*
1581 * An error return from page_cache_read can result if the
1582 * system is low on memory, or a problem occurs while trying
1583 * to schedule I/O.
1584 */
1585 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001586 return VM_FAULT_OOM;
1587 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 /*
1591 * Umm, take care of errors if the page isn't up-to-date.
1592 * Try to re-read it _once_. We do this synchronously,
1593 * because there really aren't any performance issues here
1594 * and we need to check for errors.
1595 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001597 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001598 if (!error) {
1599 wait_on_page_locked(page);
1600 if (!PageUptodate(page))
1601 error = -EIO;
1602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001604
1605 if (!error || error == AOP_TRUNCATED_PAGE)
1606 goto retry_find;
1607
1608 /* Things didn't work out. Return zero to tell the mm layer so. */
1609 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001610 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001611}
1612EXPORT_SYMBOL(filemap_fault);
1613
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001614const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001615 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616};
1617
1618/* This is used for a general mmap of a disk file */
1619
1620int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1621{
1622 struct address_space *mapping = file->f_mapping;
1623
1624 if (!mapping->a_ops->readpage)
1625 return -ENOEXEC;
1626 file_accessed(file);
1627 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001628 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 return 0;
1630}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632/*
1633 * This is for filesystems which do not implement ->writepage.
1634 */
1635int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1636{
1637 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1638 return -EINVAL;
1639 return generic_file_mmap(file, vma);
1640}
1641#else
1642int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1643{
1644 return -ENOSYS;
1645}
1646int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1647{
1648 return -ENOSYS;
1649}
1650#endif /* CONFIG_MMU */
1651
1652EXPORT_SYMBOL(generic_file_mmap);
1653EXPORT_SYMBOL(generic_file_readonly_mmap);
1654
Nick Piggin6fe69002007-05-06 14:49:04 -07001655static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001656 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 int (*filler)(void *,struct page*),
1658 void *data)
1659{
Nick Piggineb2be182007-10-16 01:24:57 -07001660 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 int err;
1662repeat:
1663 page = find_get_page(mapping, index);
1664 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -07001665 page = page_cache_alloc_cold(mapping);
1666 if (!page)
1667 return ERR_PTR(-ENOMEM);
1668 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1669 if (unlikely(err)) {
1670 page_cache_release(page);
1671 if (err == -EEXIST)
1672 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 return ERR_PTR(err);
1675 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 err = filler(data, page);
1677 if (err < 0) {
1678 page_cache_release(page);
1679 page = ERR_PTR(err);
1680 }
1681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 return page;
1683}
1684
Randy Dunlap76824862008-03-19 17:00:40 -07001685/**
1686 * read_cache_page_async - read into page cache, fill it if needed
1687 * @mapping: the page's address_space
1688 * @index: the page index
1689 * @filler: function to perform the read
1690 * @data: destination for read data
1691 *
Nick Piggin6fe69002007-05-06 14:49:04 -07001692 * Same as read_cache_page, but don't wait for page to become unlocked
1693 * after submitting it to the filler.
Randy Dunlap76824862008-03-19 17:00:40 -07001694 *
1695 * Read into the page cache. If a page already exists, and PageUptodate() is
1696 * not set, try to fill the page but don't wait for it to become unlocked.
1697 *
1698 * If the page does not get brought uptodate, return -EIO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 */
Nick Piggin6fe69002007-05-06 14:49:04 -07001700struct page *read_cache_page_async(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001701 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 int (*filler)(void *,struct page*),
1703 void *data)
1704{
1705 struct page *page;
1706 int err;
1707
1708retry:
1709 page = __read_cache_page(mapping, index, filler, data);
1710 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001711 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (PageUptodate(page))
1713 goto out;
1714
1715 lock_page(page);
1716 if (!page->mapping) {
1717 unlock_page(page);
1718 page_cache_release(page);
1719 goto retry;
1720 }
1721 if (PageUptodate(page)) {
1722 unlock_page(page);
1723 goto out;
1724 }
1725 err = filler(data, page);
1726 if (err < 0) {
1727 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001728 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 }
David Howellsc855ff32007-05-09 13:42:20 +01001730out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001731 mark_page_accessed(page);
1732 return page;
1733}
1734EXPORT_SYMBOL(read_cache_page_async);
1735
1736/**
1737 * read_cache_page - read into page cache, fill it if needed
1738 * @mapping: the page's address_space
1739 * @index: the page index
1740 * @filler: function to perform the read
1741 * @data: destination for read data
1742 *
1743 * Read into the page cache. If a page already exists, and PageUptodate() is
1744 * not set, try to fill the page then wait for it to become unlocked.
1745 *
1746 * If the page does not get brought uptodate, return -EIO.
1747 */
1748struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001749 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001750 int (*filler)(void *,struct page*),
1751 void *data)
1752{
1753 struct page *page;
1754
1755 page = read_cache_page_async(mapping, index, filler, data);
1756 if (IS_ERR(page))
1757 goto out;
1758 wait_on_page_locked(page);
1759 if (!PageUptodate(page)) {
1760 page_cache_release(page);
1761 page = ERR_PTR(-EIO);
1762 }
1763 out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 return page;
1765}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766EXPORT_SYMBOL(read_cache_page);
1767
1768/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 * The logic we want is
1770 *
1771 * if suid or (sgid and xgrp)
1772 * remove privs
1773 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001774int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775{
1776 mode_t mode = dentry->d_inode->i_mode;
1777 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
1779 /* suid always must be killed */
1780 if (unlikely(mode & S_ISUID))
1781 kill = ATTR_KILL_SUID;
1782
1783 /*
1784 * sgid without any exec bits is just a mandatory locking mark; leave
1785 * it alone. If some exec bits are set, it's a real sgid; kill it.
1786 */
1787 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1788 kill |= ATTR_KILL_SGID;
1789
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001790 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001791 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792
Jens Axboe01de85e2006-10-17 19:50:36 +02001793 return 0;
1794}
Mark Fashehd23a1472006-10-17 17:05:18 -07001795EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001796
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001797static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001798{
1799 struct iattr newattrs;
1800
1801 newattrs.ia_valid = ATTR_FORCE | kill;
1802 return notify_change(dentry, &newattrs);
1803}
1804
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001805int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001806{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001807 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001808 int killsuid = should_remove_suid(dentry);
1809 int killpriv = security_inode_need_killpriv(dentry);
1810 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001811
Serge E. Hallynb5376772007-10-16 23:31:36 -07001812 if (killpriv < 0)
1813 return killpriv;
1814 if (killpriv)
1815 error = security_inode_killpriv(dentry);
1816 if (!error && killsuid)
1817 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001818
Serge E. Hallynb5376772007-10-16 23:31:36 -07001819 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001821EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822
Nick Piggin2f718ff2007-10-16 01:24:59 -07001823static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 const struct iovec *iov, size_t base, size_t bytes)
1825{
Ingo Molnarf1800532009-03-02 11:00:57 +01001826 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
1828 while (bytes) {
1829 char __user *buf = iov->iov_base + base;
1830 int copy = min(bytes, iov->iov_len - base);
1831
1832 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001833 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 copied += copy;
1835 bytes -= copy;
1836 vaddr += copy;
1837 iov++;
1838
NeilBrown01408c42006-06-25 05:47:58 -07001839 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 }
1842 return copied - left;
1843}
1844
1845/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001846 * Copy as much as we can into the page and return the number of bytes which
1847 * were sucessfully copied. If a fault is encountered then return the number of
1848 * bytes which were copied.
1849 */
1850size_t iov_iter_copy_from_user_atomic(struct page *page,
1851 struct iov_iter *i, unsigned long offset, size_t bytes)
1852{
1853 char *kaddr;
1854 size_t copied;
1855
1856 BUG_ON(!in_atomic());
1857 kaddr = kmap_atomic(page, KM_USER0);
1858 if (likely(i->nr_segs == 1)) {
1859 int left;
1860 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001861 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001862 copied = bytes - left;
1863 } else {
1864 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1865 i->iov, i->iov_offset, bytes);
1866 }
1867 kunmap_atomic(kaddr, KM_USER0);
1868
1869 return copied;
1870}
Nick Piggin89e10782007-10-16 01:25:07 -07001871EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001872
1873/*
1874 * This has the same sideeffects and return value as
1875 * iov_iter_copy_from_user_atomic().
1876 * The difference is that it attempts to resolve faults.
1877 * Page must not be locked.
1878 */
1879size_t iov_iter_copy_from_user(struct page *page,
1880 struct iov_iter *i, unsigned long offset, size_t bytes)
1881{
1882 char *kaddr;
1883 size_t copied;
1884
1885 kaddr = kmap(page);
1886 if (likely(i->nr_segs == 1)) {
1887 int left;
1888 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001889 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001890 copied = bytes - left;
1891 } else {
1892 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1893 i->iov, i->iov_offset, bytes);
1894 }
1895 kunmap(page);
1896 return copied;
1897}
Nick Piggin89e10782007-10-16 01:25:07 -07001898EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001899
Nick Pigginf7009262008-03-10 11:43:59 -07001900void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001901{
Nick Pigginf7009262008-03-10 11:43:59 -07001902 BUG_ON(i->count < bytes);
1903
Nick Piggin2f718ff2007-10-16 01:24:59 -07001904 if (likely(i->nr_segs == 1)) {
1905 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001906 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001907 } else {
1908 const struct iovec *iov = i->iov;
1909 size_t base = i->iov_offset;
1910
Nick Piggin124d3b72008-02-02 15:01:17 +01001911 /*
1912 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001913 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001914 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001915 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001916 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001917
Nick Pigginf7009262008-03-10 11:43:59 -07001918 copy = min(bytes, iov->iov_len - base);
1919 BUG_ON(!i->count || i->count < copy);
1920 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001921 bytes -= copy;
1922 base += copy;
1923 if (iov->iov_len == base) {
1924 iov++;
1925 base = 0;
1926 }
1927 }
1928 i->iov = iov;
1929 i->iov_offset = base;
1930 }
1931}
Nick Piggin89e10782007-10-16 01:25:07 -07001932EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001933
Nick Pigginafddba42007-10-16 01:25:01 -07001934/*
1935 * Fault in the first iovec of the given iov_iter, to a maximum length
1936 * of bytes. Returns 0 on success, or non-zero if the memory could not be
1937 * accessed (ie. because it is an invalid address).
1938 *
1939 * writev-intensive code may want this to prefault several iovecs -- that
1940 * would be possible (callers must not rely on the fact that _only_ the
1941 * first iovec will be faulted with the current implementation).
1942 */
1943int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001944{
Nick Piggin2f718ff2007-10-16 01:24:59 -07001945 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07001946 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
1947 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001948}
Nick Piggin89e10782007-10-16 01:25:07 -07001949EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001950
1951/*
1952 * Return the count of just the current iov_iter segment.
1953 */
1954size_t iov_iter_single_seg_count(struct iov_iter *i)
1955{
1956 const struct iovec *iov = i->iov;
1957 if (i->nr_segs == 1)
1958 return i->count;
1959 else
1960 return min(i->count, iov->iov_len - i->iov_offset);
1961}
Nick Piggin89e10782007-10-16 01:25:07 -07001962EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001963
1964/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 * Performs necessary checks before doing a write
1966 *
Randy Dunlap485bb992006-06-23 02:03:49 -07001967 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 * Returns appropriate error code that caller should return or
1969 * zero in case that write should be allowed.
1970 */
1971inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
1972{
1973 struct inode *inode = file->f_mapping->host;
1974 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1975
1976 if (unlikely(*pos < 0))
1977 return -EINVAL;
1978
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 if (!isblk) {
1980 /* FIXME: this is for backwards compatibility with 2.4 */
1981 if (file->f_flags & O_APPEND)
1982 *pos = i_size_read(inode);
1983
1984 if (limit != RLIM_INFINITY) {
1985 if (*pos >= limit) {
1986 send_sig(SIGXFSZ, current, 0);
1987 return -EFBIG;
1988 }
1989 if (*count > limit - (typeof(limit))*pos) {
1990 *count = limit - (typeof(limit))*pos;
1991 }
1992 }
1993 }
1994
1995 /*
1996 * LFS rule
1997 */
1998 if (unlikely(*pos + *count > MAX_NON_LFS &&
1999 !(file->f_flags & O_LARGEFILE))) {
2000 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 return -EFBIG;
2002 }
2003 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2004 *count = MAX_NON_LFS - (unsigned long)*pos;
2005 }
2006 }
2007
2008 /*
2009 * Are we about to exceed the fs block limit ?
2010 *
2011 * If we have written data it becomes a short write. If we have
2012 * exceeded without writing data we send a signal and return EFBIG.
2013 * Linus frestrict idea will clean these up nicely..
2014 */
2015 if (likely(!isblk)) {
2016 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2017 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 return -EFBIG;
2019 }
2020 /* zero-length writes at ->s_maxbytes are OK */
2021 }
2022
2023 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2024 *count = inode->i_sb->s_maxbytes - *pos;
2025 } else {
David Howells93614012006-09-30 20:45:40 +02002026#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 loff_t isize;
2028 if (bdev_read_only(I_BDEV(inode)))
2029 return -EPERM;
2030 isize = i_size_read(inode);
2031 if (*pos >= isize) {
2032 if (*count || *pos > isize)
2033 return -ENOSPC;
2034 }
2035
2036 if (*pos + *count > isize)
2037 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002038#else
2039 return -EPERM;
2040#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 }
2042 return 0;
2043}
2044EXPORT_SYMBOL(generic_write_checks);
2045
Nick Pigginafddba42007-10-16 01:25:01 -07002046int pagecache_write_begin(struct file *file, struct address_space *mapping,
2047 loff_t pos, unsigned len, unsigned flags,
2048 struct page **pagep, void **fsdata)
2049{
2050 const struct address_space_operations *aops = mapping->a_ops;
2051
Nick Piggin4e02ed42008-10-29 14:00:55 -07002052 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002053 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002054}
2055EXPORT_SYMBOL(pagecache_write_begin);
2056
2057int pagecache_write_end(struct file *file, struct address_space *mapping,
2058 loff_t pos, unsigned len, unsigned copied,
2059 struct page *page, void *fsdata)
2060{
2061 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002062
Nick Piggin4e02ed42008-10-29 14:00:55 -07002063 mark_page_accessed(page);
2064 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002065}
2066EXPORT_SYMBOL(pagecache_write_end);
2067
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068ssize_t
2069generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2070 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2071 size_t count, size_t ocount)
2072{
2073 struct file *file = iocb->ki_filp;
2074 struct address_space *mapping = file->f_mapping;
2075 struct inode *inode = mapping->host;
2076 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002077 size_t write_len;
2078 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079
2080 if (count != ocount)
2081 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2082
Christoph Hellwiga969e902008-07-23 21:27:04 -07002083 write_len = iov_length(iov, *nr_segs);
2084 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002085
Nick Piggin48b47c52009-01-06 14:40:22 -08002086 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002087 if (written)
2088 goto out;
2089
2090 /*
2091 * After a write we want buffered reads to be sure to go to disk to get
2092 * the new data. We invalidate clean cached page from the region we're
2093 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002094 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002095 */
2096 if (mapping->nrpages) {
2097 written = invalidate_inode_pages2_range(mapping,
2098 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002099 /*
2100 * If a page can not be invalidated, return 0 to fall back
2101 * to buffered write.
2102 */
2103 if (written) {
2104 if (written == -EBUSY)
2105 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002106 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002107 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002108 }
2109
2110 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2111
2112 /*
2113 * Finally, try again to invalidate clean pages which might have been
2114 * cached by non-direct readahead, or faulted in by get_user_pages()
2115 * if the source of the write was an mmap'ed region of the file
2116 * we're writing. Either one is a pretty crazy thing to do,
2117 * so we don't support it 100%. If this invalidation
2118 * fails, tough, the write still worked...
2119 */
2120 if (mapping->nrpages) {
2121 invalidate_inode_pages2_range(mapping,
2122 pos >> PAGE_CACHE_SHIFT, end);
2123 }
2124
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 if (written > 0) {
2126 loff_t end = pos + written;
2127 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2128 i_size_write(inode, end);
2129 mark_inode_dirty(inode);
2130 }
2131 *ppos = end;
2132 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002133out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 return written;
2135}
2136EXPORT_SYMBOL(generic_file_direct_write);
2137
Nick Piggineb2be182007-10-16 01:24:57 -07002138/*
2139 * Find or create a page at the given pagecache position. Return the locked
2140 * page. This function is specifically for buffered writes.
2141 */
Nick Piggin54566b22009-01-04 12:00:53 -08002142struct page *grab_cache_page_write_begin(struct address_space *mapping,
2143 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002144{
2145 int status;
2146 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002147 gfp_t gfp_notmask = 0;
2148 if (flags & AOP_FLAG_NOFS)
2149 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002150repeat:
2151 page = find_lock_page(mapping, index);
2152 if (likely(page))
2153 return page;
2154
Nick Piggin54566b22009-01-04 12:00:53 -08002155 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002156 if (!page)
2157 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002158 status = add_to_page_cache_lru(page, mapping, index,
2159 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002160 if (unlikely(status)) {
2161 page_cache_release(page);
2162 if (status == -EEXIST)
2163 goto repeat;
2164 return NULL;
2165 }
2166 return page;
2167}
Nick Piggin54566b22009-01-04 12:00:53 -08002168EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002169
Nick Pigginafddba42007-10-16 01:25:01 -07002170static ssize_t generic_perform_write(struct file *file,
2171 struct iov_iter *i, loff_t pos)
2172{
2173 struct address_space *mapping = file->f_mapping;
2174 const struct address_space_operations *a_ops = mapping->a_ops;
2175 long status = 0;
2176 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002177 unsigned int flags = 0;
2178
2179 /*
2180 * Copies from kernel address space cannot fail (NFSD is a big user).
2181 */
2182 if (segment_eq(get_fs(), KERNEL_DS))
2183 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002184
2185 do {
2186 struct page *page;
2187 pgoff_t index; /* Pagecache index for current page */
2188 unsigned long offset; /* Offset into pagecache page */
2189 unsigned long bytes; /* Bytes to write to page */
2190 size_t copied; /* Bytes copied from user */
2191 void *fsdata;
2192
2193 offset = (pos & (PAGE_CACHE_SIZE - 1));
2194 index = pos >> PAGE_CACHE_SHIFT;
2195 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2196 iov_iter_count(i));
2197
2198again:
2199
2200 /*
2201 * Bring in the user page that we will copy from _first_.
2202 * Otherwise there's a nasty deadlock on copying from the
2203 * same page as we're writing to, without it being marked
2204 * up-to-date.
2205 *
2206 * Not only is this an optimisation, but it is also required
2207 * to check that the address is actually valid, when atomic
2208 * usercopies are used, below.
2209 */
2210 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2211 status = -EFAULT;
2212 break;
2213 }
2214
Nick Piggin674b8922007-10-16 01:25:03 -07002215 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002216 &page, &fsdata);
2217 if (unlikely(status))
2218 break;
2219
2220 pagefault_disable();
2221 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2222 pagefault_enable();
2223 flush_dcache_page(page);
2224
Josef Bacikc8236db2009-07-05 12:08:18 -07002225 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002226 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2227 page, fsdata);
2228 if (unlikely(status < 0))
2229 break;
2230 copied = status;
2231
2232 cond_resched();
2233
Nick Piggin124d3b72008-02-02 15:01:17 +01002234 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002235 if (unlikely(copied == 0)) {
2236 /*
2237 * If we were unable to copy any data at all, we must
2238 * fall back to a single segment length write.
2239 *
2240 * If we didn't fallback here, we could livelock
2241 * because not all segments in the iov can be copied at
2242 * once without a pagefault.
2243 */
2244 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2245 iov_iter_single_seg_count(i));
2246 goto again;
2247 }
Nick Pigginafddba42007-10-16 01:25:01 -07002248 pos += copied;
2249 written += copied;
2250
2251 balance_dirty_pages_ratelimited(mapping);
2252
2253 } while (iov_iter_count(i));
2254
2255 return written ? written : status;
2256}
2257
2258ssize_t
2259generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2260 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2261 size_t count, ssize_t written)
2262{
2263 struct file *file = iocb->ki_filp;
2264 struct address_space *mapping = file->f_mapping;
Nick Pigginafddba42007-10-16 01:25:01 -07002265 ssize_t status;
2266 struct iov_iter i;
2267
2268 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002269 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002270
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002272 written += status;
2273 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 }
2275
2276 /*
2277 * If we get here for O_DIRECT writes then we must have fallen through
2278 * to buffered writes (block instantiation inside i_size). So we sync
2279 * the file data here, to try to honour O_DIRECT expectations.
2280 */
2281 if (unlikely(file->f_flags & O_DIRECT) && written)
Nick Piggin48b47c52009-01-06 14:40:22 -08002282 status = filemap_write_and_wait_range(mapping,
2283 pos, pos + written - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 return written ? written : status;
2286}
2287EXPORT_SYMBOL(generic_file_buffered_write);
2288
Jan Karae4dd9de2009-08-17 18:10:06 +02002289/**
2290 * __generic_file_aio_write - write data to a file
2291 * @iocb: IO state structure (file, offset, etc.)
2292 * @iov: vector with data to write
2293 * @nr_segs: number of segments in the vector
2294 * @ppos: position where to write
2295 *
2296 * This function does all the work needed for actually writing data to a
2297 * file. It does all basic checks, removes SUID from the file, updates
2298 * modification times and calls proper subroutines depending on whether we
2299 * do direct IO or a standard buffered write.
2300 *
2301 * It expects i_mutex to be grabbed unless we work on a block device or similar
2302 * object which does not need locking at all.
2303 *
2304 * This function does *not* take care of syncing data in case of O_SYNC write.
2305 * A caller has to handle it. This is mainly due to the fact that we want to
2306 * avoid syncing under i_mutex.
2307 */
2308ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2309 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
2311 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002312 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 size_t ocount; /* original count */
2314 size_t count; /* after file limit checks */
2315 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 loff_t pos;
2317 ssize_t written;
2318 ssize_t err;
2319
2320 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002321 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2322 if (err)
2323 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324
2325 count = ocount;
2326 pos = *ppos;
2327
2328 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2329
2330 /* We can write back this queue in page reclaim */
2331 current->backing_dev_info = mapping->backing_dev_info;
2332 written = 0;
2333
2334 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2335 if (err)
2336 goto out;
2337
2338 if (count == 0)
2339 goto out;
2340
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002341 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (err)
2343 goto out;
2344
Christoph Hellwig870f4812006-01-09 20:52:01 -08002345 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
2347 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2348 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002349 loff_t endbyte;
2350 ssize_t written_buffered;
2351
2352 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2353 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 if (written < 0 || written == count)
2355 goto out;
2356 /*
2357 * direct-io write to a hole: fall through to buffered I/O
2358 * for completing the rest of the request.
2359 */
2360 pos += written;
2361 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002362 written_buffered = generic_file_buffered_write(iocb, iov,
2363 nr_segs, pos, ppos, count,
2364 written);
2365 /*
2366 * If generic_file_buffered_write() retuned a synchronous error
2367 * then we want to return the number of bytes which were
2368 * direct-written, or the error code if that was zero. Note
2369 * that this differs from normal direct-io semantics, which
2370 * will return -EFOO even if some bytes were written.
2371 */
2372 if (written_buffered < 0) {
2373 err = written_buffered;
2374 goto out;
2375 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002377 /*
2378 * We need to ensure that the page cache pages are written to
2379 * disk and invalidated to preserve the expected O_DIRECT
2380 * semantics.
2381 */
2382 endbyte = pos + written_buffered - written - 1;
Mark Fashehef51c972007-05-08 00:27:10 -07002383 err = do_sync_mapping_range(file->f_mapping, pos, endbyte,
2384 SYNC_FILE_RANGE_WAIT_BEFORE|
2385 SYNC_FILE_RANGE_WRITE|
2386 SYNC_FILE_RANGE_WAIT_AFTER);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002387 if (err == 0) {
2388 written = written_buffered;
2389 invalidate_mapping_pages(mapping,
2390 pos >> PAGE_CACHE_SHIFT,
2391 endbyte >> PAGE_CACHE_SHIFT);
2392 } else {
2393 /*
2394 * We don't know how much we wrote, so just return
2395 * the number of bytes which were direct-written
2396 */
2397 }
2398 } else {
2399 written = generic_file_buffered_write(iocb, iov, nr_segs,
2400 pos, ppos, count, written);
2401 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402out:
2403 current->backing_dev_info = NULL;
2404 return written ? written : err;
2405}
Jan Karae4dd9de2009-08-17 18:10:06 +02002406EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
Jan Karae4dd9de2009-08-17 18:10:06 +02002408/**
2409 * generic_file_aio_write - write data to a file
2410 * @iocb: IO state structure
2411 * @iov: vector with data to write
2412 * @nr_segs: number of segments in the vector
2413 * @pos: position in file where to write
2414 *
2415 * This is a wrapper around __generic_file_aio_write() to be used by most
2416 * filesystems. It takes care of syncing the file in case of O_SYNC file
2417 * and acquires i_mutex as needed.
2418 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002419ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2420 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421{
2422 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002423 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
2426 BUG_ON(iocb->ki_pos != pos);
2427
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002428 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002429 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002430 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
Jan Kara148f9482009-08-17 19:52:36 +02002432 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 ssize_t err;
2434
Jan Kara148f9482009-08-17 19:52:36 +02002435 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002436 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 ret = err;
2438 }
2439 return ret;
2440}
2441EXPORT_SYMBOL(generic_file_aio_write);
2442
David Howellscf9a2ae2006-08-29 19:05:54 +01002443/**
2444 * try_to_release_page() - release old fs-specific metadata on a page
2445 *
2446 * @page: the page which the kernel is trying to free
2447 * @gfp_mask: memory allocation flags (and I/O mode)
2448 *
2449 * The address_space is to try to release any data against the page
2450 * (presumably at page->private). If the release was successful, return `1'.
2451 * Otherwise return zero.
2452 *
David Howells266cf652009-04-03 16:42:36 +01002453 * This may also be called if PG_fscache is set on a page, indicating that the
2454 * page is known to the local caching routines.
2455 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002456 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002457 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002458 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002459 */
2460int try_to_release_page(struct page *page, gfp_t gfp_mask)
2461{
2462 struct address_space * const mapping = page->mapping;
2463
2464 BUG_ON(!PageLocked(page));
2465 if (PageWriteback(page))
2466 return 0;
2467
2468 if (mapping && mapping->a_ops->releasepage)
2469 return mapping->a_ops->releasepage(page, gfp_mask);
2470 return try_to_free_buffers(page);
2471}
2472
2473EXPORT_SYMBOL(try_to_release_page);