blob: ca389394fa2a1d563097c06d34881ac0e8e42559 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/filemap.c
3 *
4 * Copyright (C) 1994-1999 Linus Torvalds
5 */
6
7/*
8 * This file handles the generic file mmap semantics used by
9 * most "normal" filesystems (but you don't /have/ to use this:
10 * the NFS filesystem used to do this differently, for example)
11 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/compiler.h>
14#include <linux/fs.h>
Hiro Yoshiokac22ce142006-06-23 02:04:16 -070015#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/aio.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080017#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/kernel_stat.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/mm.h>
21#include <linux/swap.h>
22#include <linux/mman.h>
23#include <linux/pagemap.h>
24#include <linux/file.h>
25#include <linux/uio.h>
26#include <linux/hash.h>
27#include <linux/writeback.h>
Linus Torvalds53253382007-10-18 14:47:32 -070028#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/pagevec.h>
30#include <linux/blkdev.h>
31#include <linux/security.h>
32#include <linux/syscalls.h>
Paul Jackson44110fe2006-03-24 03:16:04 -080033#include <linux/cpuset.h>
Nick Piggin2f718ff2007-10-16 01:24:59 -070034#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080035#include <linux/memcontrol.h>
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036#include <linux/mm_inline.h> /* for page_is_file_cache() */
Nick Piggin0f8053a2006-03-22 00:08:33 -080037#include "internal.h"
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 * FIXME: remove all knowledge of the buffer layer from the core VM
41 */
Jan Kara148f9482009-08-17 19:52:36 +020042#include <linux/buffer_head.h> /* for try_to_free_buffers */
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <asm/mman.h>
45
46/*
47 * Shared mappings implemented 30.11.1994. It's not fully working yet,
48 * though.
49 *
50 * Shared mappings now work. 15.8.1995 Bruno.
51 *
52 * finished 'unifying' the page and buffer cache and SMP-threaded the
53 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
54 *
55 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
56 */
57
58/*
59 * Lock ordering:
60 *
npiggin@suse.de25d9e2d2009-08-21 02:35:05 +100061 * ->i_mmap_lock (truncate_pagecache)
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 * ->private_lock (__free_pte->__set_page_dirty_buffers)
Hugh Dickins5d337b92005-09-03 15:54:41 -070063 * ->swap_lock (exclusive_swap_page, others)
64 * ->mapping->tree_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080066 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 * ->i_mmap_lock (truncate->unmap_mapping_range)
68 *
69 * ->mmap_sem
70 * ->i_mmap_lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070071 * ->page_table_lock or pte_lock (various, mainly in memory.c)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock)
73 *
74 * ->mmap_sem
75 * ->lock_page (access_process_vm)
76 *
Nick Piggin82591e62006-10-19 23:29:10 -070077 * ->i_mutex (generic_file_buffered_write)
78 * ->mmap_sem (fault_in_pages_readable->do_page_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080080 * ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * ->i_alloc_sem (various)
82 *
83 * ->inode_lock
84 * ->sb_lock (fs/fs-writeback.c)
85 * ->mapping->tree_lock (__sync_single_inode)
86 *
87 * ->i_mmap_lock
88 * ->anon_vma.lock (vma_adjust)
89 *
90 * ->anon_vma.lock
Hugh Dickinsb8072f02005-10-29 18:16:41 -070091 * ->page_table_lock or pte_lock (anon_vma_prepare and various)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -070093 * ->page_table_lock or pte_lock
Hugh Dickins5d337b92005-09-03 15:54:41 -070094 * ->swap_lock (try_to_unmap_one)
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 * ->private_lock (try_to_unmap_one)
96 * ->tree_lock (try_to_unmap_one)
97 * ->zone.lru_lock (follow_page->mark_page_accessed)
Nick Piggin053837f2006-01-18 17:42:27 -080098 * ->zone.lru_lock (check_pte_range->isolate_lru_page)
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 * ->private_lock (page_remove_rmap->set_page_dirty)
100 * ->tree_lock (page_remove_rmap->set_page_dirty)
101 * ->inode_lock (page_remove_rmap->set_page_dirty)
102 * ->inode_lock (zap_pte_range->set_page_dirty)
103 * ->private_lock (zap_pte_range->__set_page_dirty_buffers)
104 *
Andi Kleen6a460792009-09-16 11:50:15 +0200105 * (code doesn't rely on that order, so you could switch it around)
106 * ->tasklist_lock (memory_failure, collect_procs_ao)
107 * ->i_mmap_lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 */
109
110/*
111 * Remove a page from the page cache and free it. Caller has to make
112 * sure the page is locked and that nobody else uses it - or that usage
Nick Piggin19fd6232008-07-25 19:45:32 -0700113 * is safe. The caller must hold the mapping's tree_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 */
115void __remove_from_page_cache(struct page *page)
116{
117 struct address_space *mapping = page->mapping;
118
119 radix_tree_delete(&mapping->page_tree, page->index);
120 page->mapping = NULL;
121 mapping->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700122 __dec_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700123 if (PageSwapBacked(page))
124 __dec_zone_page_state(page, NR_SHMEM);
Nick Piggin45426812007-07-15 23:38:12 -0700125 BUG_ON(page_mapped(page));
Linus Torvalds3a692792007-12-19 14:05:13 -0800126
127 /*
128 * Some filesystems seem to re-dirty the page even after
129 * the VM has canceled the dirty bit (eg ext3 journaling).
130 *
131 * Fix it up by doing a final dirty accounting check after
132 * having removed the page entirely.
133 */
134 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
135 dec_zone_page_state(page, NR_FILE_DIRTY);
136 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
137 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138}
139
140void remove_from_page_cache(struct page *page)
141{
142 struct address_space *mapping = page->mapping;
Linus Torvalds6072d132010-12-01 13:35:19 -0500143 void (*freepage)(struct page *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Matt Mackallcd7619d2005-05-01 08:59:01 -0700145 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Linus Torvalds6072d132010-12-01 13:35:19 -0500147 freepage = mapping->a_ops->freepage;
Nick Piggin19fd6232008-07-25 19:45:32 -0700148 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 __remove_from_page_cache(page);
Nick Piggin19fd6232008-07-25 19:45:32 -0700150 spin_unlock_irq(&mapping->tree_lock);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700151 mem_cgroup_uncharge_cache_page(page);
Linus Torvalds6072d132010-12-01 13:35:19 -0500152
153 if (freepage)
154 freepage(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155}
Miklos Szeredia52116a2010-05-25 15:06:06 +0200156EXPORT_SYMBOL(remove_from_page_cache);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158static int sync_page(void *word)
159{
160 struct address_space *mapping;
161 struct page *page;
162
Andi Kleen07808b72005-11-05 17:25:53 +0100163 page = container_of((unsigned long *)word, struct page, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 /*
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700166 * page_mapping() is being called without PG_locked held.
167 * Some knowledge of the state and use of the page is used to
168 * reduce the requirements down to a memory barrier.
169 * The danger here is of a stale page_mapping() return value
170 * indicating a struct address_space different from the one it's
171 * associated with when it is associated with one.
172 * After smp_mb(), it's either the correct page_mapping() for
173 * the page, or an old page_mapping() and the page's own
174 * page_mapping() has gone NULL.
175 * The ->sync_page() address_space operation must tolerate
176 * page_mapping() going NULL. By an amazing coincidence,
177 * this comes about because none of the users of the page
178 * in the ->sync_page() methods make essential use of the
179 * page_mapping(), merely passing the page down to the backing
180 * device's unplug functions when it's non-NULL, which in turn
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700181 * ignore it for all cases but swap, where only page_private(page) is
William Lee Irwin IIIdd1d5af2005-05-01 08:58:38 -0700182 * of interest. When page_mapping() does go NULL, the entire
183 * call stack gracefully ignores the page and returns.
184 * -- wli
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 */
186 smp_mb();
187 mapping = page_mapping(page);
188 if (mapping && mapping->a_ops && mapping->a_ops->sync_page)
189 mapping->a_ops->sync_page(page);
190 io_schedule();
191 return 0;
192}
193
Matthew Wilcox2687a352007-12-06 11:18:49 -0500194static int sync_page_killable(void *word)
195{
196 sync_page(word);
197 return fatal_signal_pending(current) ? -EINTR : 0;
198}
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700201 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
Martin Waitz67be2dd2005-05-01 08:59:26 -0700202 * @mapping: address space structure to write
203 * @start: offset in bytes where the range starts
Andrew Morton469eb4d2006-03-24 03:17:45 -0800204 * @end: offset in bytes where the range ends (inclusive)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700205 * @sync_mode: enable synchronous operation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700207 * Start writeback against all of a mapping's dirty pages that lie
208 * within the byte offsets <start, end> inclusive.
209 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
Randy Dunlap485bb992006-06-23 02:03:49 -0700211 * opposed to a regular memory cleansing writeback. The difference between
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * these two operations is that if a dirty page/buffer is encountered, it must
213 * be waited upon, and not just skipped over.
214 */
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800215int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
216 loff_t end, int sync_mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217{
218 int ret;
219 struct writeback_control wbc = {
220 .sync_mode = sync_mode,
Nick Piggin05fe4782009-01-06 14:39:08 -0800221 .nr_to_write = LONG_MAX,
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700222 .range_start = start,
223 .range_end = end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 };
225
226 if (!mapping_cap_writeback_dirty(mapping))
227 return 0;
228
229 ret = do_writepages(mapping, &wbc);
230 return ret;
231}
232
233static inline int __filemap_fdatawrite(struct address_space *mapping,
234 int sync_mode)
235{
OGAWA Hirofumi111ebb62006-06-23 02:03:26 -0700236 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
239int filemap_fdatawrite(struct address_space *mapping)
240{
241 return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
242}
243EXPORT_SYMBOL(filemap_fdatawrite);
244
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400245int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
Andrew Mortonebcf28e2006-03-24 03:18:04 -0800246 loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
249}
Jan Karaf4c0a0f2008-07-11 19:27:31 -0400250EXPORT_SYMBOL(filemap_fdatawrite_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Randy Dunlap485bb992006-06-23 02:03:49 -0700252/**
253 * filemap_flush - mostly a non-blocking flush
254 * @mapping: target address_space
255 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 * This is a mostly non-blocking flush. Not suitable for data-integrity
257 * purposes - I/O may not be started against all dirty pages.
258 */
259int filemap_flush(struct address_space *mapping)
260{
261 return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
262}
263EXPORT_SYMBOL(filemap_flush);
264
Randy Dunlap485bb992006-06-23 02:03:49 -0700265/**
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200266 * filemap_fdatawait_range - wait for writeback to complete
267 * @mapping: address space structure to wait for
268 * @start_byte: offset in bytes where the range starts
269 * @end_byte: offset in bytes where the range ends (inclusive)
Randy Dunlap485bb992006-06-23 02:03:49 -0700270 *
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200271 * Walk the list of under-writeback pages of the given address space
272 * in the given range and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 */
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200274int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
275 loff_t end_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276{
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200277 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
278 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 struct pagevec pvec;
280 int nr_pages;
281 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200283 if (end_byte < start_byte)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 return 0;
285
286 pagevec_init(&pvec, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 while ((index <= end) &&
288 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
289 PAGECACHE_TAG_WRITEBACK,
290 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
291 unsigned i;
292
293 for (i = 0; i < nr_pages; i++) {
294 struct page *page = pvec.pages[i];
295
296 /* until radix tree lookup accepts end_index */
297 if (page->index > end)
298 continue;
299
300 wait_on_page_writeback(page);
301 if (PageError(page))
302 ret = -EIO;
303 }
304 pagevec_release(&pvec);
305 cond_resched();
306 }
307
308 /* Check for outstanding write errors */
309 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
310 ret = -ENOSPC;
311 if (test_and_clear_bit(AS_EIO, &mapping->flags))
312 ret = -EIO;
313
314 return ret;
315}
Jan Karad3bccb62009-08-17 19:30:27 +0200316EXPORT_SYMBOL(filemap_fdatawait_range);
317
318/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700319 * filemap_fdatawait - wait for all under-writeback pages to complete
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 * @mapping: address space structure to wait for
Randy Dunlap485bb992006-06-23 02:03:49 -0700321 *
322 * Walk the list of under-writeback pages of the given address space
323 * and wait for all of them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 */
325int filemap_fdatawait(struct address_space *mapping)
326{
327 loff_t i_size = i_size_read(mapping->host);
328
329 if (i_size == 0)
330 return 0;
331
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200332 return filemap_fdatawait_range(mapping, 0, i_size - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334EXPORT_SYMBOL(filemap_fdatawait);
335
336int filemap_write_and_wait(struct address_space *mapping)
337{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800338 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339
340 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800341 err = filemap_fdatawrite(mapping);
342 /*
343 * Even if the above returned error, the pages may be
344 * written partially (e.g. -ENOSPC), so we wait for it.
345 * But the -EIO is special case, it may indicate the worst
346 * thing (e.g. bug) happened, so we avoid waiting for it.
347 */
348 if (err != -EIO) {
349 int err2 = filemap_fdatawait(mapping);
350 if (!err)
351 err = err2;
352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800354 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800356EXPORT_SYMBOL(filemap_write_and_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Randy Dunlap485bb992006-06-23 02:03:49 -0700358/**
359 * filemap_write_and_wait_range - write out & wait on a file range
360 * @mapping: the address_space for the pages
361 * @lstart: offset in bytes where the range starts
362 * @lend: offset in bytes where the range ends (inclusive)
363 *
Andrew Morton469eb4d2006-03-24 03:17:45 -0800364 * Write out and wait upon file offsets lstart->lend, inclusive.
365 *
366 * Note that `lend' is inclusive (describes the last byte to be written) so
367 * that this function can be used to write to the very end-of-file (end = -1).
368 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369int filemap_write_and_wait_range(struct address_space *mapping,
370 loff_t lstart, loff_t lend)
371{
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800372 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
374 if (mapping->nrpages) {
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800375 err = __filemap_fdatawrite_range(mapping, lstart, lend,
376 WB_SYNC_ALL);
377 /* See comment of filemap_write_and_wait() */
378 if (err != -EIO) {
Christoph Hellwig94004ed2009-09-30 22:16:33 +0200379 int err2 = filemap_fdatawait_range(mapping,
380 lstart, lend);
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800381 if (!err)
382 err = err2;
383 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800385 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
Chris Masonf6995582009-04-15 13:22:37 -0400387EXPORT_SYMBOL(filemap_write_and_wait_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
Randy Dunlap485bb992006-06-23 02:03:49 -0700389/**
Nick Piggine2867812008-07-25 19:45:30 -0700390 * add_to_page_cache_locked - add a locked page to the pagecache
Randy Dunlap485bb992006-06-23 02:03:49 -0700391 * @page: page to add
392 * @mapping: the page's address_space
393 * @offset: page index
394 * @gfp_mask: page allocation mode
395 *
Nick Piggine2867812008-07-25 19:45:30 -0700396 * This function is used to add a page to the pagecache. It must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 * This function does not add the page to the LRU. The caller must do that.
398 */
Nick Piggine2867812008-07-25 19:45:30 -0700399int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400400 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
Nick Piggine2867812008-07-25 19:45:30 -0700402 int error;
403
404 VM_BUG_ON(!PageLocked(page));
405
406 error = mem_cgroup_cache_charge(page, current->mm,
KAMEZAWA Hiroyuki2c26fdd2009-01-07 18:08:10 -0800407 gfp_mask & GFP_RECLAIM_MASK);
Balbir Singh35c754d2008-02-07 00:14:05 -0800408 if (error)
409 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Balbir Singh35c754d2008-02-07 00:14:05 -0800411 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 if (error == 0) {
Nick Piggine2867812008-07-25 19:45:30 -0700413 page_cache_get(page);
414 page->mapping = mapping;
415 page->index = offset;
416
Nick Piggin19fd6232008-07-25 19:45:32 -0700417 spin_lock_irq(&mapping->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 error = radix_tree_insert(&mapping->page_tree, offset, page);
Nick Piggine2867812008-07-25 19:45:30 -0700419 if (likely(!error)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 mapping->nrpages++;
Christoph Lameter347ce432006-06-30 01:55:35 -0700421 __inc_zone_page_state(page, NR_FILE_PAGES);
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700422 if (PageSwapBacked(page))
423 __inc_zone_page_state(page, NR_SHMEM);
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700424 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700425 } else {
426 page->mapping = NULL;
Daisuke Nishimurae767e052009-05-28 14:34:28 -0700427 spin_unlock_irq(&mapping->tree_lock);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700428 mem_cgroup_uncharge_cache_page(page);
Nick Piggine2867812008-07-25 19:45:30 -0700429 page_cache_release(page);
430 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 radix_tree_preload_end();
Balbir Singh35c754d2008-02-07 00:14:05 -0800432 } else
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700433 mem_cgroup_uncharge_cache_page(page);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800434out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 return error;
436}
Nick Piggine2867812008-07-25 19:45:30 -0700437EXPORT_SYMBOL(add_to_page_cache_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
Al Viro6daa0e22005-10-21 03:18:50 -0400440 pgoff_t offset, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700442 int ret;
443
444 /*
445 * Splice_read and readahead add shmem/tmpfs pages into the page cache
446 * before shmem_readpage has a chance to mark them as SwapBacked: they
KOSAKI Motohiroe9d6c152010-05-24 14:31:48 -0700447 * need to go on the anon lru below, and mem_cgroup_cache_charge
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700448 * (called in add_to_page_cache) needs to know where they're going too.
449 */
450 if (mapping_cap_swap_backed(mapping))
451 SetPageSwapBacked(page);
452
453 ret = add_to_page_cache(page, mapping, offset, gfp_mask);
454 if (ret == 0) {
455 if (page_is_file_cache(page))
456 lru_cache_add_file(page);
457 else
KOSAKI Motohiroe9d6c152010-05-24 14:31:48 -0700458 lru_cache_add_anon(page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 return ret;
461}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300462EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
Paul Jackson44110fe2006-03-24 03:16:04 -0800464#ifdef CONFIG_NUMA
Nick Piggin2ae88142006-10-28 10:38:23 -0700465struct page *__page_cache_alloc(gfp_t gfp)
Paul Jackson44110fe2006-03-24 03:16:04 -0800466{
Miao Xiec0ff7452010-05-24 14:32:08 -0700467 int n;
468 struct page *page;
469
Paul Jackson44110fe2006-03-24 03:16:04 -0800470 if (cpuset_do_page_mem_spread()) {
Miao Xiec0ff7452010-05-24 14:32:08 -0700471 get_mems_allowed();
472 n = cpuset_mem_spread_node();
473 page = alloc_pages_exact_node(n, gfp, 0);
474 put_mems_allowed();
475 return page;
Paul Jackson44110fe2006-03-24 03:16:04 -0800476 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700477 return alloc_pages(gfp, 0);
Paul Jackson44110fe2006-03-24 03:16:04 -0800478}
Nick Piggin2ae88142006-10-28 10:38:23 -0700479EXPORT_SYMBOL(__page_cache_alloc);
Paul Jackson44110fe2006-03-24 03:16:04 -0800480#endif
481
Nick Piggindb376482006-09-25 23:31:24 -0700482static int __sleep_on_page_lock(void *word)
483{
484 io_schedule();
485 return 0;
486}
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488/*
489 * In order to wait for pages to become available there must be
490 * waitqueues associated with pages. By using a hash table of
491 * waitqueues where the bucket discipline is to maintain all
492 * waiters on the same queue and wake all when any of the pages
493 * become available, and for the woken contexts to check to be
494 * sure the appropriate page became available, this saves space
495 * at a cost of "thundering herd" phenomena during rare hash
496 * collisions.
497 */
498static wait_queue_head_t *page_waitqueue(struct page *page)
499{
500 const struct zone *zone = page_zone(page);
501
502 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
503}
504
505static inline void wake_up_page(struct page *page, int bit)
506{
507 __wake_up_bit(page_waitqueue(page), &page->flags, bit);
508}
509
Harvey Harrison920c7a52008-02-04 22:29:26 -0800510void wait_on_page_bit(struct page *page, int bit_nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511{
512 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
513
514 if (test_bit(bit_nr, &page->flags))
515 __wait_on_bit(page_waitqueue(page), &wait, sync_page,
516 TASK_UNINTERRUPTIBLE);
517}
518EXPORT_SYMBOL(wait_on_page_bit);
519
520/**
David Howells385e1ca2009-04-03 16:42:39 +0100521 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
Randy Dunlap697f6192009-04-13 14:39:54 -0700522 * @page: Page defining the wait queue of interest
523 * @waiter: Waiter to add to the queue
David Howells385e1ca2009-04-03 16:42:39 +0100524 *
525 * Add an arbitrary @waiter to the wait queue for the nominated @page.
526 */
527void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
528{
529 wait_queue_head_t *q = page_waitqueue(page);
530 unsigned long flags;
531
532 spin_lock_irqsave(&q->lock, flags);
533 __add_wait_queue(q, waiter);
534 spin_unlock_irqrestore(&q->lock, flags);
535}
536EXPORT_SYMBOL_GPL(add_page_wait_queue);
537
538/**
Randy Dunlap485bb992006-06-23 02:03:49 -0700539 * unlock_page - unlock a locked page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 * @page: the page
541 *
542 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
543 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
544 * mechananism between PageLocked pages and PageWriteback pages is shared.
545 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
546 *
Nick Piggin8413ac92008-10-18 20:26:59 -0700547 * The mb is necessary to enforce ordering between the clear_bit and the read
548 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800550void unlock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551{
Nick Piggin8413ac92008-10-18 20:26:59 -0700552 VM_BUG_ON(!PageLocked(page));
553 clear_bit_unlock(PG_locked, &page->flags);
554 smp_mb__after_clear_bit();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 wake_up_page(page, PG_locked);
556}
557EXPORT_SYMBOL(unlock_page);
558
Randy Dunlap485bb992006-06-23 02:03:49 -0700559/**
560 * end_page_writeback - end writeback against a page
561 * @page: the page
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 */
563void end_page_writeback(struct page *page)
564{
Miklos Szerediac6aadb2008-04-28 02:12:38 -0700565 if (TestClearPageReclaim(page))
566 rotate_reclaimable_page(page);
567
568 if (!test_clear_page_writeback(page))
569 BUG();
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 smp_mb__after_clear_bit();
572 wake_up_page(page, PG_writeback);
573}
574EXPORT_SYMBOL(end_page_writeback);
575
Randy Dunlap485bb992006-06-23 02:03:49 -0700576/**
577 * __lock_page - get a lock on the page, assuming we need to sleep to get it
578 * @page: the page to lock
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 *
Randy Dunlap485bb992006-06-23 02:03:49 -0700580 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * random driver's requestfn sets TASK_RUNNING, we could busywait. However
582 * chances are that on the second loop, the block layer's plug list is empty,
583 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE.
584 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800585void __lock_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586{
587 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
588
589 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page,
590 TASK_UNINTERRUPTIBLE);
591}
592EXPORT_SYMBOL(__lock_page);
593
Harvey Harrisonb5606c22008-02-13 15:03:16 -0800594int __lock_page_killable(struct page *page)
Matthew Wilcox2687a352007-12-06 11:18:49 -0500595{
596 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
597
598 return __wait_on_bit_lock(page_waitqueue(page), &wait,
599 sync_page_killable, TASK_KILLABLE);
600}
Evgeniy Polyakov18bc0bb2009-02-09 17:02:42 +0300601EXPORT_SYMBOL_GPL(__lock_page_killable);
Matthew Wilcox2687a352007-12-06 11:18:49 -0500602
Randy Dunlap76824862008-03-19 17:00:40 -0700603/**
604 * __lock_page_nosync - get a lock on the page, without calling sync_page()
605 * @page: the page to lock
606 *
Nick Piggindb376482006-09-25 23:31:24 -0700607 * Variant of lock_page that does not require the caller to hold a reference
608 * on the page's mapping.
609 */
Harvey Harrison920c7a52008-02-04 22:29:26 -0800610void __lock_page_nosync(struct page *page)
Nick Piggindb376482006-09-25 23:31:24 -0700611{
612 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
613 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock,
614 TASK_UNINTERRUPTIBLE);
615}
616
Michel Lespinassed065bd82010-10-26 14:21:57 -0700617int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
618 unsigned int flags)
619{
620 if (!(flags & FAULT_FLAG_ALLOW_RETRY)) {
621 __lock_page(page);
622 return 1;
623 } else {
624 up_read(&mm->mmap_sem);
625 wait_on_page_locked(page);
626 return 0;
627 }
628}
629
Randy Dunlap485bb992006-06-23 02:03:49 -0700630/**
631 * find_get_page - find and get a page reference
632 * @mapping: the address_space to search
633 * @offset: the page index
634 *
Nick Pigginda6052f2006-09-25 23:31:35 -0700635 * Is there a pagecache struct page at the given (mapping, offset) tuple?
636 * If yes, increment its refcount and return it; if no, return NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 */
Nick Piggina60637c2008-07-25 19:45:31 -0700638struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
Nick Piggina60637c2008-07-25 19:45:31 -0700640 void **pagep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 struct page *page;
642
Nick Piggina60637c2008-07-25 19:45:31 -0700643 rcu_read_lock();
644repeat:
645 page = NULL;
646 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
647 if (pagep) {
648 page = radix_tree_deref_slot(pagep);
Nick Piggin27d20fd2010-11-11 14:05:19 -0800649 if (unlikely(!page))
650 goto out;
651 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700652 goto repeat;
653
654 if (!page_cache_get_speculative(page))
655 goto repeat;
656
657 /*
658 * Has the page moved?
659 * This is part of the lockless pagecache protocol. See
660 * include/linux/pagemap.h for details.
661 */
662 if (unlikely(page != *pagep)) {
663 page_cache_release(page);
664 goto repeat;
665 }
666 }
Nick Piggin27d20fd2010-11-11 14:05:19 -0800667out:
Nick Piggina60637c2008-07-25 19:45:31 -0700668 rcu_read_unlock();
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 return page;
671}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672EXPORT_SYMBOL(find_get_page);
673
Randy Dunlap485bb992006-06-23 02:03:49 -0700674/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 * find_lock_page - locate, pin and lock a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700676 * @mapping: the address_space to search
677 * @offset: the page index
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 *
679 * Locates the desired pagecache page, locks it, increments its reference
680 * count and returns its address.
681 *
682 * Returns zero if the page was not present. find_lock_page() may sleep.
683 */
Nick Piggina60637c2008-07-25 19:45:31 -0700684struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685{
686 struct page *page;
687
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688repeat:
Nick Piggina60637c2008-07-25 19:45:31 -0700689 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (page) {
Nick Piggina60637c2008-07-25 19:45:31 -0700691 lock_page(page);
692 /* Has the page been truncated? */
693 if (unlikely(page->mapping != mapping)) {
694 unlock_page(page);
695 page_cache_release(page);
696 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 }
Nick Piggina60637c2008-07-25 19:45:31 -0700698 VM_BUG_ON(page->index != offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 return page;
701}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702EXPORT_SYMBOL(find_lock_page);
703
704/**
705 * find_or_create_page - locate or add a pagecache page
Martin Waitz67be2dd2005-05-01 08:59:26 -0700706 * @mapping: the page's address_space
707 * @index: the page's index into the mapping
708 * @gfp_mask: page allocation mode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 *
710 * Locates a page in the pagecache. If the page is not present, a new page
711 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
712 * LRU list. The returned page is locked and has its reference count
713 * incremented.
714 *
715 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
716 * allocation!
717 *
718 * find_or_create_page() returns the desired page's address, or zero on
719 * memory exhaustion.
720 */
721struct page *find_or_create_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -0700722 pgoff_t index, gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723{
Nick Piggineb2be182007-10-16 01:24:57 -0700724 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 int err;
726repeat:
727 page = find_lock_page(mapping, index);
728 if (!page) {
Nick Piggineb2be182007-10-16 01:24:57 -0700729 page = __page_cache_alloc(gfp_mask);
730 if (!page)
731 return NULL;
Nick Piggin67d58ac2009-01-06 14:40:28 -0800732 /*
733 * We want a regular kernel memory (not highmem or DMA etc)
734 * allocation for the radix tree nodes, but we need to honour
735 * the context-specific requirements the caller has asked for.
736 * GFP_RECLAIM_MASK collects those requirements.
737 */
738 err = add_to_page_cache_lru(page, mapping, index,
739 (gfp_mask & GFP_RECLAIM_MASK));
Nick Piggineb2be182007-10-16 01:24:57 -0700740 if (unlikely(err)) {
741 page_cache_release(page);
742 page = NULL;
743 if (err == -EEXIST)
744 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 return page;
748}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749EXPORT_SYMBOL(find_or_create_page);
750
751/**
752 * find_get_pages - gang pagecache lookup
753 * @mapping: The address_space to search
754 * @start: The starting page index
755 * @nr_pages: The maximum number of pages
756 * @pages: Where the resulting pages are placed
757 *
758 * find_get_pages() will search for and return a group of up to
759 * @nr_pages pages in the mapping. The pages are placed at @pages.
760 * find_get_pages() takes a reference against the returned pages.
761 *
762 * The search returns a group of mapping-contiguous pages with ascending
763 * indexes. There may be holes in the indices due to not-present pages.
764 *
765 * find_get_pages() returns the number of pages which were found.
766 */
767unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
768 unsigned int nr_pages, struct page **pages)
769{
770 unsigned int i;
771 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700772 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Nick Piggina60637c2008-07-25 19:45:31 -0700774 rcu_read_lock();
775restart:
776 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
777 (void ***)pages, start, nr_pages);
778 ret = 0;
779 for (i = 0; i < nr_found; i++) {
780 struct page *page;
781repeat:
782 page = radix_tree_deref_slot((void **)pages[i]);
783 if (unlikely(!page))
784 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800785 if (radix_tree_deref_retry(page)) {
786 if (ret)
787 start = pages[ret-1]->index;
Nick Piggina60637c2008-07-25 19:45:31 -0700788 goto restart;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800789 }
Nick Piggina60637c2008-07-25 19:45:31 -0700790
791 if (!page_cache_get_speculative(page))
792 goto repeat;
793
794 /* Has the page moved? */
795 if (unlikely(page != *((void **)pages[i]))) {
796 page_cache_release(page);
797 goto repeat;
798 }
799
800 pages[ret] = page;
801 ret++;
802 }
803 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 return ret;
805}
806
Jens Axboeebf43502006-04-27 08:46:01 +0200807/**
808 * find_get_pages_contig - gang contiguous pagecache lookup
809 * @mapping: The address_space to search
810 * @index: The starting page index
811 * @nr_pages: The maximum number of pages
812 * @pages: Where the resulting pages are placed
813 *
814 * find_get_pages_contig() works exactly like find_get_pages(), except
815 * that the returned number of pages are guaranteed to be contiguous.
816 *
817 * find_get_pages_contig() returns the number of pages which were found.
818 */
819unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
820 unsigned int nr_pages, struct page **pages)
821{
822 unsigned int i;
823 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700824 unsigned int nr_found;
Jens Axboeebf43502006-04-27 08:46:01 +0200825
Nick Piggina60637c2008-07-25 19:45:31 -0700826 rcu_read_lock();
827restart:
828 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
829 (void ***)pages, index, nr_pages);
830 ret = 0;
831 for (i = 0; i < nr_found; i++) {
832 struct page *page;
833repeat:
834 page = radix_tree_deref_slot((void **)pages[i]);
835 if (unlikely(!page))
836 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800837 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700838 goto restart;
839
840 if (page->mapping == NULL || page->index != index)
Jens Axboeebf43502006-04-27 08:46:01 +0200841 break;
842
Nick Piggina60637c2008-07-25 19:45:31 -0700843 if (!page_cache_get_speculative(page))
844 goto repeat;
845
846 /* Has the page moved? */
847 if (unlikely(page != *((void **)pages[i]))) {
848 page_cache_release(page);
849 goto repeat;
850 }
851
852 pages[ret] = page;
853 ret++;
Jens Axboeebf43502006-04-27 08:46:01 +0200854 index++;
855 }
Nick Piggina60637c2008-07-25 19:45:31 -0700856 rcu_read_unlock();
857 return ret;
Jens Axboeebf43502006-04-27 08:46:01 +0200858}
David Howellsef71c152007-05-09 02:33:44 -0700859EXPORT_SYMBOL(find_get_pages_contig);
Jens Axboeebf43502006-04-27 08:46:01 +0200860
Randy Dunlap485bb992006-06-23 02:03:49 -0700861/**
862 * find_get_pages_tag - find and return pages that match @tag
863 * @mapping: the address_space to search
864 * @index: the starting page index
865 * @tag: the tag index
866 * @nr_pages: the maximum number of pages
867 * @pages: where the resulting pages are placed
868 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 * Like find_get_pages, except we only return pages which are tagged with
Randy Dunlap485bb992006-06-23 02:03:49 -0700870 * @tag. We update @index to index the next page for the traversal.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 */
872unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
873 int tag, unsigned int nr_pages, struct page **pages)
874{
875 unsigned int i;
876 unsigned int ret;
Nick Piggina60637c2008-07-25 19:45:31 -0700877 unsigned int nr_found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
Nick Piggina60637c2008-07-25 19:45:31 -0700879 rcu_read_lock();
880restart:
881 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree,
882 (void ***)pages, *index, nr_pages, tag);
883 ret = 0;
884 for (i = 0; i < nr_found; i++) {
885 struct page *page;
886repeat:
887 page = radix_tree_deref_slot((void **)pages[i]);
888 if (unlikely(!page))
889 continue;
Nick Piggin27d20fd2010-11-11 14:05:19 -0800890 if (radix_tree_deref_retry(page))
Nick Piggina60637c2008-07-25 19:45:31 -0700891 goto restart;
892
893 if (!page_cache_get_speculative(page))
894 goto repeat;
895
896 /* Has the page moved? */
897 if (unlikely(page != *((void **)pages[i]))) {
898 page_cache_release(page);
899 goto repeat;
900 }
901
902 pages[ret] = page;
903 ret++;
904 }
905 rcu_read_unlock();
906
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 if (ret)
908 *index = pages[ret - 1]->index + 1;
Nick Piggina60637c2008-07-25 19:45:31 -0700909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 return ret;
911}
David Howellsef71c152007-05-09 02:33:44 -0700912EXPORT_SYMBOL(find_get_pages_tag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913
Randy Dunlap485bb992006-06-23 02:03:49 -0700914/**
915 * grab_cache_page_nowait - returns locked page at given index in given cache
916 * @mapping: target address_space
917 * @index: the page index
918 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800919 * Same as grab_cache_page(), but do not wait if the page is unavailable.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 * This is intended for speculative data generators, where the data can
921 * be regenerated if the page couldn't be grabbed. This routine should
922 * be safe to call while holding the lock for another page.
923 *
924 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
925 * and deadlock against the caller's locked page.
926 */
927struct page *
Fengguang Wu57f6b962007-10-16 01:24:37 -0700928grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929{
930 struct page *page = find_get_page(mapping, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931
932 if (page) {
Nick Piggin529ae9a2008-08-02 12:01:03 +0200933 if (trylock_page(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 return page;
935 page_cache_release(page);
936 return NULL;
937 }
Nick Piggin2ae88142006-10-28 10:38:23 -0700938 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
Nick Piggin67d58ac2009-01-06 14:40:28 -0800939 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 page_cache_release(page);
941 page = NULL;
942 }
943 return page;
944}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945EXPORT_SYMBOL(grab_cache_page_nowait);
946
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700947/*
948 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
949 * a _large_ part of the i/o request. Imagine the worst scenario:
950 *
951 * ---R__________________________________________B__________
952 * ^ reading here ^ bad block(assume 4k)
953 *
954 * read(R) => miss => readahead(R...B) => media error => frustrating retries
955 * => failing the whole request => read(R) => read(R+1) =>
956 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
957 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
958 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
959 *
960 * It is going insane. Fix it by quickly scaling down the readahead size.
961 */
962static void shrink_readahead_size_eio(struct file *filp,
963 struct file_ra_state *ra)
964{
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700965 ra->ra_pages /= 4;
Wu Fengguang76d42bd2006-06-25 05:48:43 -0700966}
967
Randy Dunlap485bb992006-06-23 02:03:49 -0700968/**
Christoph Hellwig36e78912008-02-08 04:21:24 -0800969 * do_generic_file_read - generic file read routine
Randy Dunlap485bb992006-06-23 02:03:49 -0700970 * @filp: the file to read
971 * @ppos: current file position
972 * @desc: read_descriptor
973 * @actor: read method
974 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 * This is a generic file read routine, and uses the
Randy Dunlap485bb992006-06-23 02:03:49 -0700976 * mapping->a_ops->readpage() function for the actual low-level stuff.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 *
978 * This is really ugly. But the goto's actually try to clarify some
979 * of the logic when it comes to error handling etc.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 */
Christoph Hellwig36e78912008-02-08 04:21:24 -0800981static void do_generic_file_read(struct file *filp, loff_t *ppos,
982 read_descriptor_t *desc, read_actor_t actor)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983{
Christoph Hellwig36e78912008-02-08 04:21:24 -0800984 struct address_space *mapping = filp->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 struct inode *inode = mapping->host;
Christoph Hellwig36e78912008-02-08 04:21:24 -0800986 struct file_ra_state *ra = &filp->f_ra;
Fengguang Wu57f6b962007-10-16 01:24:37 -0700987 pgoff_t index;
988 pgoff_t last_index;
989 pgoff_t prev_index;
990 unsigned long offset; /* offset into pagecache page */
Jan Karaec0f1632007-05-06 14:49:25 -0700991 unsigned int prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 index = *ppos >> PAGE_CACHE_SHIFT;
Fengguang Wu7ff81072007-10-16 01:24:35 -0700995 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
996 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
998 offset = *ppos & ~PAGE_CACHE_MASK;
999
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 for (;;) {
1001 struct page *page;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001002 pgoff_t end_index;
NeilBrowna32ea1e2007-07-17 04:03:04 -07001003 loff_t isize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 unsigned long nr, ret;
1005
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007find_page:
1008 page = find_get_page(mapping, index);
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001009 if (!page) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001010 page_cache_sync_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001011 ra, filp,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001012 index, last_index - index);
1013 page = find_get_page(mapping, index);
1014 if (unlikely(page == NULL))
1015 goto no_cached_page;
1016 }
1017 if (PageReadahead(page)) {
Rusty Russellcf914a72007-07-19 01:48:08 -07001018 page_cache_async_readahead(mapping,
Fengguang Wu7ff81072007-10-16 01:24:35 -07001019 ra, filp, page,
Fengguang Wu3ea89ee2007-07-19 01:48:02 -07001020 index, last_index - index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001022 if (!PageUptodate(page)) {
1023 if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1024 !mapping->a_ops->is_partially_uptodate)
1025 goto page_not_up_to_date;
Nick Piggin529ae9a2008-08-02 12:01:03 +02001026 if (!trylock_page(page))
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001027 goto page_not_up_to_date;
Dave Hansen8d056cb2010-11-11 14:05:15 -08001028 /* Did it get truncated before we got the lock? */
1029 if (!page->mapping)
1030 goto page_not_up_to_date_locked;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001031 if (!mapping->a_ops->is_partially_uptodate(page,
1032 desc, offset))
1033 goto page_not_up_to_date_locked;
1034 unlock_page(page);
1035 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036page_ok:
NeilBrowna32ea1e2007-07-17 04:03:04 -07001037 /*
1038 * i_size must be checked after we know the page is Uptodate.
1039 *
1040 * Checking i_size after the check allows us to calculate
1041 * the correct value for "nr", which means the zero-filled
1042 * part of the page is not copied back to userspace (unless
1043 * another truncate extends the file - this is desired though).
1044 */
1045
1046 isize = i_size_read(inode);
1047 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1048 if (unlikely(!isize || index > end_index)) {
1049 page_cache_release(page);
1050 goto out;
1051 }
1052
1053 /* nr is the maximum number of bytes to copy from this page */
1054 nr = PAGE_CACHE_SIZE;
1055 if (index == end_index) {
1056 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1057 if (nr <= offset) {
1058 page_cache_release(page);
1059 goto out;
1060 }
1061 }
1062 nr = nr - offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
1064 /* If users can be writing to this page using arbitrary
1065 * virtual addresses, take care about potential aliasing
1066 * before reading the page on the kernel side.
1067 */
1068 if (mapping_writably_mapped(mapping))
1069 flush_dcache_page(page);
1070
1071 /*
Jan Karaec0f1632007-05-06 14:49:25 -07001072 * When a sequential read accesses a page several times,
1073 * only mark it as accessed the first time.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 */
Jan Karaec0f1632007-05-06 14:49:25 -07001075 if (prev_index != index || offset != prev_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 mark_page_accessed(page);
1077 prev_index = index;
1078
1079 /*
1080 * Ok, we have the page, and it's up-to-date, so
1081 * now we can copy it to user space...
1082 *
1083 * The actor routine returns how many bytes were actually used..
1084 * NOTE! This may not be the same as how much of a user buffer
1085 * we filled up (we may be padding etc), so we can only update
1086 * "pos" here (the actor routine has to update the user buffer
1087 * pointers and the remaining count).
1088 */
1089 ret = actor(desc, page, offset, nr);
1090 offset += ret;
1091 index += offset >> PAGE_CACHE_SHIFT;
1092 offset &= ~PAGE_CACHE_MASK;
Jan Kara6ce745e2007-05-06 14:49:26 -07001093 prev_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095 page_cache_release(page);
1096 if (ret == nr && desc->count)
1097 continue;
1098 goto out;
1099
1100page_not_up_to_date:
1101 /* Get exclusive access to the page ... */
Oleg Nesterov85462322008-06-08 21:20:43 +04001102 error = lock_page_killable(page);
1103 if (unlikely(error))
1104 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07001106page_not_up_to_date_locked:
Nick Pigginda6052f2006-09-25 23:31:35 -07001107 /* Did it get truncated before we got the lock? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 if (!page->mapping) {
1109 unlock_page(page);
1110 page_cache_release(page);
1111 continue;
1112 }
1113
1114 /* Did somebody else fill it already? */
1115 if (PageUptodate(page)) {
1116 unlock_page(page);
1117 goto page_ok;
1118 }
1119
1120readpage:
Jeff Moyer91803b42010-05-26 11:49:40 -04001121 /*
1122 * A previous I/O error may have been due to temporary
1123 * failures, eg. multipath errors.
1124 * PG_error will be set again if readpage fails.
1125 */
1126 ClearPageError(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 /* Start the actual read. The read will unlock the page. */
1128 error = mapping->a_ops->readpage(filp, page);
1129
Zach Brown994fc28c2005-12-15 14:28:17 -08001130 if (unlikely(error)) {
1131 if (error == AOP_TRUNCATED_PAGE) {
1132 page_cache_release(page);
1133 goto find_page;
1134 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 goto readpage_error;
Zach Brown994fc28c2005-12-15 14:28:17 -08001136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
1138 if (!PageUptodate(page)) {
Oleg Nesterov85462322008-06-08 21:20:43 +04001139 error = lock_page_killable(page);
1140 if (unlikely(error))
1141 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 if (!PageUptodate(page)) {
1143 if (page->mapping == NULL) {
1144 /*
Christoph Hellwig2ecdc822010-01-26 17:27:20 +01001145 * invalidate_mapping_pages got it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 */
1147 unlock_page(page);
1148 page_cache_release(page);
1149 goto find_page;
1150 }
1151 unlock_page(page);
Fengguang Wu7ff81072007-10-16 01:24:35 -07001152 shrink_readahead_size_eio(filp, ra);
Oleg Nesterov85462322008-06-08 21:20:43 +04001153 error = -EIO;
1154 goto readpage_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
1156 unlock_page(page);
1157 }
1158
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 goto page_ok;
1160
1161readpage_error:
1162 /* UHHUH! A synchronous read error occurred. Report it */
1163 desc->error = error;
1164 page_cache_release(page);
1165 goto out;
1166
1167no_cached_page:
1168 /*
1169 * Ok, it wasn't cached, so we need to create a new
1170 * page..
1171 */
Nick Piggineb2be182007-10-16 01:24:57 -07001172 page = page_cache_alloc_cold(mapping);
1173 if (!page) {
1174 desc->error = -ENOMEM;
1175 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
Nick Piggineb2be182007-10-16 01:24:57 -07001177 error = add_to_page_cache_lru(page, mapping,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 index, GFP_KERNEL);
1179 if (error) {
Nick Piggineb2be182007-10-16 01:24:57 -07001180 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 if (error == -EEXIST)
1182 goto find_page;
1183 desc->error = error;
1184 goto out;
1185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 goto readpage;
1187 }
1188
1189out:
Fengguang Wu7ff81072007-10-16 01:24:35 -07001190 ra->prev_pos = prev_index;
1191 ra->prev_pos <<= PAGE_CACHE_SHIFT;
1192 ra->prev_pos |= prev_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Fengguang Wuf4e6b492007-10-16 01:24:33 -07001194 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
Krishna Kumar0c6aa262008-10-15 22:01:13 -07001195 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
1198int file_read_actor(read_descriptor_t *desc, struct page *page,
1199 unsigned long offset, unsigned long size)
1200{
1201 char *kaddr;
1202 unsigned long left, count = desc->count;
1203
1204 if (size > count)
1205 size = count;
1206
1207 /*
1208 * Faults on the destination of a read are common, so do it before
1209 * taking the kmap.
1210 */
1211 if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1212 kaddr = kmap_atomic(page, KM_USER0);
1213 left = __copy_to_user_inatomic(desc->arg.buf,
1214 kaddr + offset, size);
1215 kunmap_atomic(kaddr, KM_USER0);
1216 if (left == 0)
1217 goto success;
1218 }
1219
1220 /* Do it the slow way */
1221 kaddr = kmap(page);
1222 left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1223 kunmap(page);
1224
1225 if (left) {
1226 size -= left;
1227 desc->error = -EFAULT;
1228 }
1229success:
1230 desc->count = count - size;
1231 desc->written += size;
1232 desc->arg.buf += size;
1233 return size;
1234}
1235
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001236/*
1237 * Performs necessary checks before doing a write
1238 * @iov: io vector request
1239 * @nr_segs: number of segments in the iovec
1240 * @count: number of bytes to write
1241 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1242 *
1243 * Adjust number of segments and amount of bytes to write (nr_segs should be
1244 * properly initialized first). Returns appropriate error code that caller
1245 * should return or zero in case that write should be allowed.
1246 */
1247int generic_segment_checks(const struct iovec *iov,
1248 unsigned long *nr_segs, size_t *count, int access_flags)
1249{
1250 unsigned long seg;
1251 size_t cnt = 0;
1252 for (seg = 0; seg < *nr_segs; seg++) {
1253 const struct iovec *iv = &iov[seg];
1254
1255 /*
1256 * If any segment has a negative length, or the cumulative
1257 * length ever wraps negative then return -EINVAL.
1258 */
1259 cnt += iv->iov_len;
1260 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1261 return -EINVAL;
1262 if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1263 continue;
1264 if (seg == 0)
1265 return -EFAULT;
1266 *nr_segs = seg;
1267 cnt -= iv->iov_len; /* This segment is no good */
1268 break;
1269 }
1270 *count = cnt;
1271 return 0;
1272}
1273EXPORT_SYMBOL(generic_segment_checks);
1274
Randy Dunlap485bb992006-06-23 02:03:49 -07001275/**
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001276 * generic_file_aio_read - generic filesystem read routine
Randy Dunlap485bb992006-06-23 02:03:49 -07001277 * @iocb: kernel I/O control block
1278 * @iov: io vector request
1279 * @nr_segs: number of segments in the iovec
Henrik Kretzschmarb2abacf2006-10-04 02:15:22 -07001280 * @pos: current file position
Randy Dunlap485bb992006-06-23 02:03:49 -07001281 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 * This is the "read()" routine for all filesystems
1283 * that can use the page cache directly.
1284 */
1285ssize_t
Badari Pulavarty543ade12006-09-30 23:28:48 -07001286generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1287 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288{
1289 struct file *filp = iocb->ki_filp;
1290 ssize_t retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001291 unsigned long seg = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 size_t count;
Badari Pulavarty543ade12006-09-30 23:28:48 -07001293 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
1295 count = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07001296 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1297 if (retval)
1298 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
1300 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1301 if (filp->f_flags & O_DIRECT) {
Badari Pulavarty543ade12006-09-30 23:28:48 -07001302 loff_t size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 struct address_space *mapping;
1304 struct inode *inode;
1305
1306 mapping = filp->f_mapping;
1307 inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 if (!count)
1309 goto out; /* skip atime */
1310 size = i_size_read(inode);
1311 if (pos < size) {
Nick Piggin48b47c52009-01-06 14:40:22 -08001312 retval = filemap_write_and_wait_range(mapping, pos,
1313 pos + iov_length(iov, nr_segs) - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07001314 if (!retval) {
1315 retval = mapping->a_ops->direct_IO(READ, iocb,
1316 iov, pos, nr_segs);
1317 }
Josef Bacik66f998f2010-05-23 11:00:54 -04001318 if (retval > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 *ppos = pos + retval;
Josef Bacik66f998f2010-05-23 11:00:54 -04001320 count -= retval;
1321 }
1322
1323 /*
1324 * Btrfs can have a short DIO read if we encounter
1325 * compressed extents, so if there was an error, or if
1326 * we've already read everything we wanted to, or if
1327 * there was a short read because we hit EOF, go ahead
1328 * and return. Otherwise fallthrough to buffered io for
1329 * the rest of the read.
1330 */
1331 if (retval < 0 || !count || *ppos >= size) {
Hugh Dickins11fa9772008-07-23 21:27:34 -07001332 file_accessed(filp);
1333 goto out;
1334 }
Steven Whitehouse0e0bcae2006-09-27 14:45:07 -04001335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 }
1337
Josef Bacik66f998f2010-05-23 11:00:54 -04001338 count = retval;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001339 for (seg = 0; seg < nr_segs; seg++) {
1340 read_descriptor_t desc;
Josef Bacik66f998f2010-05-23 11:00:54 -04001341 loff_t offset = 0;
1342
1343 /*
1344 * If we did a short DIO read we need to skip the section of the
1345 * iov that we've already read data into.
1346 */
1347 if (count) {
1348 if (count > iov[seg].iov_len) {
1349 count -= iov[seg].iov_len;
1350 continue;
1351 }
1352 offset = count;
1353 count = 0;
1354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Hugh Dickins11fa9772008-07-23 21:27:34 -07001356 desc.written = 0;
Josef Bacik66f998f2010-05-23 11:00:54 -04001357 desc.arg.buf = iov[seg].iov_base + offset;
1358 desc.count = iov[seg].iov_len - offset;
Hugh Dickins11fa9772008-07-23 21:27:34 -07001359 if (desc.count == 0)
1360 continue;
1361 desc.error = 0;
1362 do_generic_file_read(filp, ppos, &desc, file_read_actor);
1363 retval += desc.written;
1364 if (desc.error) {
1365 retval = retval ?: desc.error;
1366 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 }
Hugh Dickins11fa9772008-07-23 21:27:34 -07001368 if (desc.count > 0)
1369 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 }
1371out:
1372 return retval;
1373}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374EXPORT_SYMBOL(generic_file_aio_read);
1375
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376static ssize_t
1377do_readahead(struct address_space *mapping, struct file *filp,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001378 pgoff_t index, unsigned long nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
1380 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage)
1381 return -EINVAL;
1382
Wu Fengguangf7e839d2009-06-16 15:31:20 -07001383 force_page_cache_readahead(mapping, filp, index, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 return 0;
1385}
1386
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001387SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 ssize_t ret;
1390 struct file *file;
1391
1392 ret = -EBADF;
1393 file = fget(fd);
1394 if (file) {
1395 if (file->f_mode & FMODE_READ) {
1396 struct address_space *mapping = file->f_mapping;
Fengguang Wu57f6b962007-10-16 01:24:37 -07001397 pgoff_t start = offset >> PAGE_CACHE_SHIFT;
1398 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 unsigned long len = end - start + 1;
1400 ret = do_readahead(mapping, file, start, len);
1401 }
1402 fput(file);
1403 }
1404 return ret;
1405}
Heiko Carstens6673e0c2009-01-14 14:14:02 +01001406#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1407asmlinkage long SyS_readahead(long fd, loff_t offset, long count)
1408{
1409 return SYSC_readahead((int) fd, offset, (size_t) count);
1410}
1411SYSCALL_ALIAS(sys_readahead, SyS_readahead);
1412#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414#ifdef CONFIG_MMU
Randy Dunlap485bb992006-06-23 02:03:49 -07001415/**
1416 * page_cache_read - adds requested page to the page cache if not already there
1417 * @file: file to read
1418 * @offset: page index
1419 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 * This adds the requested page to the page cache if it isn't already there,
1421 * and schedules an I/O to read in its contents from disk.
1422 */
Harvey Harrison920c7a52008-02-04 22:29:26 -08001423static int page_cache_read(struct file *file, pgoff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
1425 struct address_space *mapping = file->f_mapping;
1426 struct page *page;
Zach Brown994fc28c2005-12-15 14:28:17 -08001427 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Zach Brown994fc28c2005-12-15 14:28:17 -08001429 do {
1430 page = page_cache_alloc_cold(mapping);
1431 if (!page)
1432 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Zach Brown994fc28c2005-12-15 14:28:17 -08001434 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1435 if (ret == 0)
1436 ret = mapping->a_ops->readpage(file, page);
1437 else if (ret == -EEXIST)
1438 ret = 0; /* losing race to add is OK */
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
Zach Brown994fc28c2005-12-15 14:28:17 -08001442 } while (ret == AOP_TRUNCATED_PAGE);
1443
1444 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445}
1446
1447#define MMAP_LOTSAMISS (100)
1448
Linus Torvaldsef00e082009-06-16 15:31:25 -07001449/*
1450 * Synchronous readahead happens when we don't even find
1451 * a page in the page cache at all.
1452 */
1453static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1454 struct file_ra_state *ra,
1455 struct file *file,
1456 pgoff_t offset)
1457{
1458 unsigned long ra_pages;
1459 struct address_space *mapping = file->f_mapping;
1460
1461 /* If we don't want any read-ahead, don't bother */
1462 if (VM_RandomReadHint(vma))
1463 return;
1464
Wu Fengguang70ac23c2009-06-16 15:31:28 -07001465 if (VM_SequentialReadHint(vma) ||
1466 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) {
Wu Fengguang7ffc59b2009-06-16 15:31:38 -07001467 page_cache_sync_readahead(mapping, ra, file, offset,
1468 ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001469 return;
1470 }
1471
1472 if (ra->mmap_miss < INT_MAX)
1473 ra->mmap_miss++;
1474
1475 /*
1476 * Do we miss much more than hit in this file? If so,
1477 * stop bothering with read-ahead. It will only hurt.
1478 */
1479 if (ra->mmap_miss > MMAP_LOTSAMISS)
1480 return;
1481
Wu Fengguangd30a1102009-06-16 15:31:30 -07001482 /*
1483 * mmap read-around
1484 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001485 ra_pages = max_sane_readahead(ra->ra_pages);
1486 if (ra_pages) {
Wu Fengguangd30a1102009-06-16 15:31:30 -07001487 ra->start = max_t(long, 0, offset - ra_pages/2);
1488 ra->size = ra_pages;
1489 ra->async_size = 0;
1490 ra_submit(ra, mapping, file);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001491 }
1492}
1493
1494/*
1495 * Asynchronous readahead happens when we find the page and PG_readahead,
1496 * so we want to possibly extend the readahead further..
1497 */
1498static void do_async_mmap_readahead(struct vm_area_struct *vma,
1499 struct file_ra_state *ra,
1500 struct file *file,
1501 struct page *page,
1502 pgoff_t offset)
1503{
1504 struct address_space *mapping = file->f_mapping;
1505
1506 /* If we don't want any read-ahead, don't bother */
1507 if (VM_RandomReadHint(vma))
1508 return;
1509 if (ra->mmap_miss > 0)
1510 ra->mmap_miss--;
1511 if (PageReadahead(page))
Wu Fengguang2fad6f52009-06-16 15:31:29 -07001512 page_cache_async_readahead(mapping, ra, file,
1513 page, offset, ra->ra_pages);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001514}
1515
Randy Dunlap485bb992006-06-23 02:03:49 -07001516/**
Nick Piggin54cb8822007-07-19 01:46:59 -07001517 * filemap_fault - read in file data for page fault handling
Nick Piggind0217ac2007-07-19 01:47:03 -07001518 * @vma: vma in which the fault was taken
1519 * @vmf: struct vm_fault containing details of the fault
Randy Dunlap485bb992006-06-23 02:03:49 -07001520 *
Nick Piggin54cb8822007-07-19 01:46:59 -07001521 * filemap_fault() is invoked via the vma operations vector for a
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 * mapped memory region to read in file data during a page fault.
1523 *
1524 * The goto's are kind of ugly, but this streamlines the normal case of having
1525 * it in the page cache, and handles the special cases reasonably without
1526 * having a lot of duplicated code.
1527 */
Nick Piggind0217ac2007-07-19 01:47:03 -07001528int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
1530 int error;
Nick Piggin54cb8822007-07-19 01:46:59 -07001531 struct file *file = vma->vm_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 struct address_space *mapping = file->f_mapping;
1533 struct file_ra_state *ra = &file->f_ra;
1534 struct inode *inode = mapping->host;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001535 pgoff_t offset = vmf->pgoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 struct page *page;
Jan Kara2004dc82008-02-08 04:20:11 -08001537 pgoff_t size;
Nick Piggin83c54072007-07-19 01:47:05 -07001538 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001541 if (offset >= size)
Linus Torvalds5307cc12007-10-31 09:19:46 -07001542 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 * Do we have something in the page cache already?
1546 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001547 page = find_get_page(mapping, offset);
1548 if (likely(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 /*
Linus Torvaldsef00e082009-06-16 15:31:25 -07001550 * We found the page, so try async readahead before
1551 * waiting for the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001553 do_async_mmap_readahead(vma, ra, file, page, offset);
Linus Torvaldsef00e082009-06-16 15:31:25 -07001554 } else {
1555 /* No page in the page cache at all */
1556 do_sync_mmap_readahead(vma, ra, file, offset);
1557 count_vm_event(PGMAJFAULT);
1558 ret = VM_FAULT_MAJOR;
1559retry_find:
Michel Lespinasseb522c942010-10-26 14:21:56 -07001560 page = find_get_page(mapping, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (!page)
1562 goto no_cached_page;
1563 }
1564
Michel Lespinassed88c0922010-11-02 13:05:18 -07001565 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1566 page_cache_release(page);
Michel Lespinassed065bd82010-10-26 14:21:57 -07001567 return ret | VM_FAULT_RETRY;
Michel Lespinassed88c0922010-11-02 13:05:18 -07001568 }
Michel Lespinasseb522c942010-10-26 14:21:56 -07001569
1570 /* Did it get truncated? */
1571 if (unlikely(page->mapping != mapping)) {
1572 unlock_page(page);
1573 put_page(page);
1574 goto retry_find;
1575 }
1576 VM_BUG_ON(page->index != offset);
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 /*
Nick Piggind00806b2007-07-19 01:46:57 -07001579 * We have a locked page in the page cache, now we need to check
1580 * that it's up-to-date. If not, it is going to be due to an error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 */
Nick Piggind00806b2007-07-19 01:46:57 -07001582 if (unlikely(!PageUptodate(page)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 goto page_not_uptodate;
1584
Linus Torvaldsef00e082009-06-16 15:31:25 -07001585 /*
1586 * Found the page and have a reference on it.
1587 * We must recheck i_size under page lock.
1588 */
Nick Piggind00806b2007-07-19 01:46:57 -07001589 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvaldsef00e082009-06-16 15:31:25 -07001590 if (unlikely(offset >= size)) {
Nick Piggind00806b2007-07-19 01:46:57 -07001591 unlock_page(page);
Yan Zheng745ad482007-10-08 10:08:37 -07001592 page_cache_release(page);
Linus Torvalds5307cc12007-10-31 09:19:46 -07001593 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001594 }
1595
Linus Torvaldsef00e082009-06-16 15:31:25 -07001596 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT;
Nick Piggind0217ac2007-07-19 01:47:03 -07001597 vmf->page = page;
Nick Piggin83c54072007-07-19 01:47:05 -07001598 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600no_cached_page:
1601 /*
1602 * We're only likely to ever get here if MADV_RANDOM is in
1603 * effect.
1604 */
Linus Torvaldsef00e082009-06-16 15:31:25 -07001605 error = page_cache_read(file, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
1607 /*
1608 * The page we want has now been added to the page cache.
1609 * In the unlikely event that someone removed it in the
1610 * meantime, we'll just come back here and read it again.
1611 */
1612 if (error >= 0)
1613 goto retry_find;
1614
1615 /*
1616 * An error return from page_cache_read can result if the
1617 * system is low on memory, or a problem occurs while trying
1618 * to schedule I/O.
1619 */
1620 if (error == -ENOMEM)
Nick Piggind0217ac2007-07-19 01:47:03 -07001621 return VM_FAULT_OOM;
1622 return VM_FAULT_SIGBUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624page_not_uptodate:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 /*
1626 * Umm, take care of errors if the page isn't up-to-date.
1627 * Try to re-read it _once_. We do this synchronously,
1628 * because there really aren't any performance issues here
1629 * and we need to check for errors.
1630 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 ClearPageError(page);
Zach Brown994fc28c2005-12-15 14:28:17 -08001632 error = mapping->a_ops->readpage(file, page);
Miklos Szeredi3ef0f722008-05-14 16:05:37 -07001633 if (!error) {
1634 wait_on_page_locked(page);
1635 if (!PageUptodate(page))
1636 error = -EIO;
1637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 page_cache_release(page);
Nick Piggind00806b2007-07-19 01:46:57 -07001639
1640 if (!error || error == AOP_TRUNCATED_PAGE)
1641 goto retry_find;
1642
1643 /* Things didn't work out. Return zero to tell the mm layer so. */
1644 shrink_readahead_size_eio(file, ra);
Nick Piggind0217ac2007-07-19 01:47:03 -07001645 return VM_FAULT_SIGBUS;
Nick Piggin54cb8822007-07-19 01:46:59 -07001646}
1647EXPORT_SYMBOL(filemap_fault);
1648
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001649const struct vm_operations_struct generic_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001650 .fault = filemap_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651};
1652
1653/* This is used for a general mmap of a disk file */
1654
1655int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1656{
1657 struct address_space *mapping = file->f_mapping;
1658
1659 if (!mapping->a_ops->readpage)
1660 return -ENOEXEC;
1661 file_accessed(file);
1662 vma->vm_ops = &generic_file_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001663 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 return 0;
1665}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667/*
1668 * This is for filesystems which do not implement ->writepage.
1669 */
1670int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1671{
1672 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1673 return -EINVAL;
1674 return generic_file_mmap(file, vma);
1675}
1676#else
1677int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1678{
1679 return -ENOSYS;
1680}
1681int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1682{
1683 return -ENOSYS;
1684}
1685#endif /* CONFIG_MMU */
1686
1687EXPORT_SYMBOL(generic_file_mmap);
1688EXPORT_SYMBOL(generic_file_readonly_mmap);
1689
Nick Piggin6fe69002007-05-06 14:49:04 -07001690static struct page *__read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001691 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001693 void *data,
1694 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695{
Nick Piggineb2be182007-10-16 01:24:57 -07001696 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 int err;
1698repeat:
1699 page = find_get_page(mapping, index);
1700 if (!page) {
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001701 page = __page_cache_alloc(gfp | __GFP_COLD);
Nick Piggineb2be182007-10-16 01:24:57 -07001702 if (!page)
1703 return ERR_PTR(-ENOMEM);
1704 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL);
1705 if (unlikely(err)) {
1706 page_cache_release(page);
1707 if (err == -EEXIST)
1708 goto repeat;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 /* Presumably ENOMEM for radix tree node */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 return ERR_PTR(err);
1711 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 err = filler(data, page);
1713 if (err < 0) {
1714 page_cache_release(page);
1715 page = ERR_PTR(err);
1716 }
1717 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 return page;
1719}
1720
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001721static struct page *do_read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001722 pgoff_t index,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 int (*filler)(void *,struct page*),
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001724 void *data,
1725 gfp_t gfp)
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727{
1728 struct page *page;
1729 int err;
1730
1731retry:
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001732 page = __read_cache_page(mapping, index, filler, data, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 if (IS_ERR(page))
David Howellsc855ff32007-05-09 13:42:20 +01001734 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 if (PageUptodate(page))
1736 goto out;
1737
1738 lock_page(page);
1739 if (!page->mapping) {
1740 unlock_page(page);
1741 page_cache_release(page);
1742 goto retry;
1743 }
1744 if (PageUptodate(page)) {
1745 unlock_page(page);
1746 goto out;
1747 }
1748 err = filler(data, page);
1749 if (err < 0) {
1750 page_cache_release(page);
David Howellsc855ff32007-05-09 13:42:20 +01001751 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 }
David Howellsc855ff32007-05-09 13:42:20 +01001753out:
Nick Piggin6fe69002007-05-06 14:49:04 -07001754 mark_page_accessed(page);
1755 return page;
1756}
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001757
1758/**
1759 * read_cache_page_async - read into page cache, fill it if needed
1760 * @mapping: the page's address_space
1761 * @index: the page index
1762 * @filler: function to perform the read
1763 * @data: destination for read data
1764 *
1765 * Same as read_cache_page, but don't wait for page to become unlocked
1766 * after submitting it to the filler.
1767 *
1768 * Read into the page cache. If a page already exists, and PageUptodate() is
1769 * not set, try to fill the page but don't wait for it to become unlocked.
1770 *
1771 * If the page does not get brought uptodate, return -EIO.
1772 */
1773struct page *read_cache_page_async(struct address_space *mapping,
1774 pgoff_t index,
1775 int (*filler)(void *,struct page*),
1776 void *data)
1777{
1778 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1779}
Nick Piggin6fe69002007-05-06 14:49:04 -07001780EXPORT_SYMBOL(read_cache_page_async);
1781
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001782static struct page *wait_on_page_read(struct page *page)
1783{
1784 if (!IS_ERR(page)) {
1785 wait_on_page_locked(page);
1786 if (!PageUptodate(page)) {
1787 page_cache_release(page);
1788 page = ERR_PTR(-EIO);
1789 }
1790 }
1791 return page;
1792}
1793
1794/**
1795 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1796 * @mapping: the page's address_space
1797 * @index: the page index
1798 * @gfp: the page allocator flags to use if allocating
1799 *
1800 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1801 * any new page allocations done using the specified allocation flags. Note
1802 * that the Radix tree operations will still use GFP_KERNEL, so you can't
1803 * expect to do this atomically or anything like that - but you can pass in
1804 * other page requirements.
1805 *
1806 * If the page does not get brought uptodate, return -EIO.
1807 */
1808struct page *read_cache_page_gfp(struct address_space *mapping,
1809 pgoff_t index,
1810 gfp_t gfp)
1811{
1812 filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1813
1814 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1815}
1816EXPORT_SYMBOL(read_cache_page_gfp);
1817
Nick Piggin6fe69002007-05-06 14:49:04 -07001818/**
1819 * read_cache_page - read into page cache, fill it if needed
1820 * @mapping: the page's address_space
1821 * @index: the page index
1822 * @filler: function to perform the read
1823 * @data: destination for read data
1824 *
1825 * Read into the page cache. If a page already exists, and PageUptodate() is
1826 * not set, try to fill the page then wait for it to become unlocked.
1827 *
1828 * If the page does not get brought uptodate, return -EIO.
1829 */
1830struct page *read_cache_page(struct address_space *mapping,
Fengguang Wu57f6b962007-10-16 01:24:37 -07001831 pgoff_t index,
Nick Piggin6fe69002007-05-06 14:49:04 -07001832 int (*filler)(void *,struct page*),
1833 void *data)
1834{
Linus Torvalds0531b2a2010-01-27 09:20:03 -08001835 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837EXPORT_SYMBOL(read_cache_page);
1838
1839/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 * The logic we want is
1841 *
1842 * if suid or (sgid and xgrp)
1843 * remove privs
1844 */
Jens Axboe01de85e2006-10-17 19:50:36 +02001845int should_remove_suid(struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846{
1847 mode_t mode = dentry->d_inode->i_mode;
1848 int kill = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
1850 /* suid always must be killed */
1851 if (unlikely(mode & S_ISUID))
1852 kill = ATTR_KILL_SUID;
1853
1854 /*
1855 * sgid without any exec bits is just a mandatory locking mark; leave
1856 * it alone. If some exec bits are set, it's a real sgid; kill it.
1857 */
1858 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1859 kill |= ATTR_KILL_SGID;
1860
Dmitri Monakhov7f5ff762008-12-01 14:34:56 -08001861 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode)))
Jens Axboe01de85e2006-10-17 19:50:36 +02001862 return kill;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863
Jens Axboe01de85e2006-10-17 19:50:36 +02001864 return 0;
1865}
Mark Fashehd23a1472006-10-17 17:05:18 -07001866EXPORT_SYMBOL(should_remove_suid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001867
Miklos Szeredi7f3d4ee2008-05-07 09:22:39 +02001868static int __remove_suid(struct dentry *dentry, int kill)
Jens Axboe01de85e2006-10-17 19:50:36 +02001869{
1870 struct iattr newattrs;
1871
1872 newattrs.ia_valid = ATTR_FORCE | kill;
1873 return notify_change(dentry, &newattrs);
1874}
1875
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001876int file_remove_suid(struct file *file)
Jens Axboe01de85e2006-10-17 19:50:36 +02001877{
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001878 struct dentry *dentry = file->f_path.dentry;
Serge E. Hallynb5376772007-10-16 23:31:36 -07001879 int killsuid = should_remove_suid(dentry);
1880 int killpriv = security_inode_need_killpriv(dentry);
1881 int error = 0;
Jens Axboe01de85e2006-10-17 19:50:36 +02001882
Serge E. Hallynb5376772007-10-16 23:31:36 -07001883 if (killpriv < 0)
1884 return killpriv;
1885 if (killpriv)
1886 error = security_inode_killpriv(dentry);
1887 if (!error && killsuid)
1888 error = __remove_suid(dentry, killsuid);
Jens Axboe01de85e2006-10-17 19:50:36 +02001889
Serge E. Hallynb5376772007-10-16 23:31:36 -07001890 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891}
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02001892EXPORT_SYMBOL(file_remove_suid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Nick Piggin2f718ff2007-10-16 01:24:59 -07001894static size_t __iovec_copy_from_user_inatomic(char *vaddr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 const struct iovec *iov, size_t base, size_t bytes)
1896{
Ingo Molnarf1800532009-03-02 11:00:57 +01001897 size_t copied = 0, left = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898
1899 while (bytes) {
1900 char __user *buf = iov->iov_base + base;
1901 int copy = min(bytes, iov->iov_len - base);
1902
1903 base = 0;
Ingo Molnarf1800532009-03-02 11:00:57 +01001904 left = __copy_from_user_inatomic(vaddr, buf, copy);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 copied += copy;
1906 bytes -= copy;
1907 vaddr += copy;
1908 iov++;
1909
NeilBrown01408c42006-06-25 05:47:58 -07001910 if (unlikely(left))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 }
1913 return copied - left;
1914}
1915
1916/*
Nick Piggin2f718ff2007-10-16 01:24:59 -07001917 * Copy as much as we can into the page and return the number of bytes which
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02001918 * were successfully copied. If a fault is encountered then return the number of
Nick Piggin2f718ff2007-10-16 01:24:59 -07001919 * bytes which were copied.
1920 */
1921size_t iov_iter_copy_from_user_atomic(struct page *page,
1922 struct iov_iter *i, unsigned long offset, size_t bytes)
1923{
1924 char *kaddr;
1925 size_t copied;
1926
1927 BUG_ON(!in_atomic());
1928 kaddr = kmap_atomic(page, KM_USER0);
1929 if (likely(i->nr_segs == 1)) {
1930 int left;
1931 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001932 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001933 copied = bytes - left;
1934 } else {
1935 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1936 i->iov, i->iov_offset, bytes);
1937 }
1938 kunmap_atomic(kaddr, KM_USER0);
1939
1940 return copied;
1941}
Nick Piggin89e10782007-10-16 01:25:07 -07001942EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001943
1944/*
1945 * This has the same sideeffects and return value as
1946 * iov_iter_copy_from_user_atomic().
1947 * The difference is that it attempts to resolve faults.
1948 * Page must not be locked.
1949 */
1950size_t iov_iter_copy_from_user(struct page *page,
1951 struct iov_iter *i, unsigned long offset, size_t bytes)
1952{
1953 char *kaddr;
1954 size_t copied;
1955
1956 kaddr = kmap(page);
1957 if (likely(i->nr_segs == 1)) {
1958 int left;
1959 char __user *buf = i->iov->iov_base + i->iov_offset;
Ingo Molnarf1800532009-03-02 11:00:57 +01001960 left = __copy_from_user(kaddr + offset, buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001961 copied = bytes - left;
1962 } else {
1963 copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1964 i->iov, i->iov_offset, bytes);
1965 }
1966 kunmap(page);
1967 return copied;
1968}
Nick Piggin89e10782007-10-16 01:25:07 -07001969EXPORT_SYMBOL(iov_iter_copy_from_user);
Nick Piggin2f718ff2007-10-16 01:24:59 -07001970
Nick Pigginf7009262008-03-10 11:43:59 -07001971void iov_iter_advance(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07001972{
Nick Pigginf7009262008-03-10 11:43:59 -07001973 BUG_ON(i->count < bytes);
1974
Nick Piggin2f718ff2007-10-16 01:24:59 -07001975 if (likely(i->nr_segs == 1)) {
1976 i->iov_offset += bytes;
Nick Pigginf7009262008-03-10 11:43:59 -07001977 i->count -= bytes;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001978 } else {
1979 const struct iovec *iov = i->iov;
1980 size_t base = i->iov_offset;
1981
Nick Piggin124d3b72008-02-02 15:01:17 +01001982 /*
1983 * The !iov->iov_len check ensures we skip over unlikely
Nick Pigginf7009262008-03-10 11:43:59 -07001984 * zero-length segments (without overruning the iovec).
Nick Piggin124d3b72008-02-02 15:01:17 +01001985 */
Linus Torvalds94ad3742008-07-30 14:45:12 -07001986 while (bytes || unlikely(i->count && !iov->iov_len)) {
Nick Pigginf7009262008-03-10 11:43:59 -07001987 int copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001988
Nick Pigginf7009262008-03-10 11:43:59 -07001989 copy = min(bytes, iov->iov_len - base);
1990 BUG_ON(!i->count || i->count < copy);
1991 i->count -= copy;
Nick Piggin2f718ff2007-10-16 01:24:59 -07001992 bytes -= copy;
1993 base += copy;
1994 if (iov->iov_len == base) {
1995 iov++;
1996 base = 0;
1997 }
1998 }
1999 i->iov = iov;
2000 i->iov_offset = base;
2001 }
2002}
Nick Piggin89e10782007-10-16 01:25:07 -07002003EXPORT_SYMBOL(iov_iter_advance);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002004
Nick Pigginafddba42007-10-16 01:25:01 -07002005/*
2006 * Fault in the first iovec of the given iov_iter, to a maximum length
2007 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2008 * accessed (ie. because it is an invalid address).
2009 *
2010 * writev-intensive code may want this to prefault several iovecs -- that
2011 * would be possible (callers must not rely on the fact that _only_ the
2012 * first iovec will be faulted with the current implementation).
2013 */
2014int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
Nick Piggin2f718ff2007-10-16 01:24:59 -07002015{
Nick Piggin2f718ff2007-10-16 01:24:59 -07002016 char __user *buf = i->iov->iov_base + i->iov_offset;
Nick Pigginafddba42007-10-16 01:25:01 -07002017 bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2018 return fault_in_pages_readable(buf, bytes);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002019}
Nick Piggin89e10782007-10-16 01:25:07 -07002020EXPORT_SYMBOL(iov_iter_fault_in_readable);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002021
2022/*
2023 * Return the count of just the current iov_iter segment.
2024 */
2025size_t iov_iter_single_seg_count(struct iov_iter *i)
2026{
2027 const struct iovec *iov = i->iov;
2028 if (i->nr_segs == 1)
2029 return i->count;
2030 else
2031 return min(i->count, iov->iov_len - i->iov_offset);
2032}
Nick Piggin89e10782007-10-16 01:25:07 -07002033EXPORT_SYMBOL(iov_iter_single_seg_count);
Nick Piggin2f718ff2007-10-16 01:24:59 -07002034
2035/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 * Performs necessary checks before doing a write
2037 *
Randy Dunlap485bb992006-06-23 02:03:49 -07002038 * Can adjust writing position or amount of bytes to write.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 * Returns appropriate error code that caller should return or
2040 * zero in case that write should be allowed.
2041 */
2042inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2043{
2044 struct inode *inode = file->f_mapping->host;
Jiri Slaby59e99e52010-03-05 13:41:44 -08002045 unsigned long limit = rlimit(RLIMIT_FSIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 if (unlikely(*pos < 0))
2048 return -EINVAL;
2049
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (!isblk) {
2051 /* FIXME: this is for backwards compatibility with 2.4 */
2052 if (file->f_flags & O_APPEND)
2053 *pos = i_size_read(inode);
2054
2055 if (limit != RLIM_INFINITY) {
2056 if (*pos >= limit) {
2057 send_sig(SIGXFSZ, current, 0);
2058 return -EFBIG;
2059 }
2060 if (*count > limit - (typeof(limit))*pos) {
2061 *count = limit - (typeof(limit))*pos;
2062 }
2063 }
2064 }
2065
2066 /*
2067 * LFS rule
2068 */
2069 if (unlikely(*pos + *count > MAX_NON_LFS &&
2070 !(file->f_flags & O_LARGEFILE))) {
2071 if (*pos >= MAX_NON_LFS) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return -EFBIG;
2073 }
2074 if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2075 *count = MAX_NON_LFS - (unsigned long)*pos;
2076 }
2077 }
2078
2079 /*
2080 * Are we about to exceed the fs block limit ?
2081 *
2082 * If we have written data it becomes a short write. If we have
2083 * exceeded without writing data we send a signal and return EFBIG.
2084 * Linus frestrict idea will clean these up nicely..
2085 */
2086 if (likely(!isblk)) {
2087 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2088 if (*count || *pos > inode->i_sb->s_maxbytes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 return -EFBIG;
2090 }
2091 /* zero-length writes at ->s_maxbytes are OK */
2092 }
2093
2094 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2095 *count = inode->i_sb->s_maxbytes - *pos;
2096 } else {
David Howells93614012006-09-30 20:45:40 +02002097#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 loff_t isize;
2099 if (bdev_read_only(I_BDEV(inode)))
2100 return -EPERM;
2101 isize = i_size_read(inode);
2102 if (*pos >= isize) {
2103 if (*count || *pos > isize)
2104 return -ENOSPC;
2105 }
2106
2107 if (*pos + *count > isize)
2108 *count = isize - *pos;
David Howells93614012006-09-30 20:45:40 +02002109#else
2110 return -EPERM;
2111#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
2113 return 0;
2114}
2115EXPORT_SYMBOL(generic_write_checks);
2116
Nick Pigginafddba42007-10-16 01:25:01 -07002117int pagecache_write_begin(struct file *file, struct address_space *mapping,
2118 loff_t pos, unsigned len, unsigned flags,
2119 struct page **pagep, void **fsdata)
2120{
2121 const struct address_space_operations *aops = mapping->a_ops;
2122
Nick Piggin4e02ed42008-10-29 14:00:55 -07002123 return aops->write_begin(file, mapping, pos, len, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002124 pagep, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002125}
2126EXPORT_SYMBOL(pagecache_write_begin);
2127
2128int pagecache_write_end(struct file *file, struct address_space *mapping,
2129 loff_t pos, unsigned len, unsigned copied,
2130 struct page *page, void *fsdata)
2131{
2132 const struct address_space_operations *aops = mapping->a_ops;
Nick Pigginafddba42007-10-16 01:25:01 -07002133
Nick Piggin4e02ed42008-10-29 14:00:55 -07002134 mark_page_accessed(page);
2135 return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
Nick Pigginafddba42007-10-16 01:25:01 -07002136}
2137EXPORT_SYMBOL(pagecache_write_end);
2138
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139ssize_t
2140generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2141 unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2142 size_t count, size_t ocount)
2143{
2144 struct file *file = iocb->ki_filp;
2145 struct address_space *mapping = file->f_mapping;
2146 struct inode *inode = mapping->host;
2147 ssize_t written;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002148 size_t write_len;
2149 pgoff_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150
2151 if (count != ocount)
2152 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2153
Christoph Hellwiga969e902008-07-23 21:27:04 -07002154 write_len = iov_length(iov, *nr_segs);
2155 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002156
Nick Piggin48b47c52009-01-06 14:40:22 -08002157 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
Christoph Hellwiga969e902008-07-23 21:27:04 -07002158 if (written)
2159 goto out;
2160
2161 /*
2162 * After a write we want buffered reads to be sure to go to disk to get
2163 * the new data. We invalidate clean cached page from the region we're
2164 * about to write. We do this *before* the write so that we can return
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002165 * without clobbering -EIOCBQUEUED from ->direct_IO().
Christoph Hellwiga969e902008-07-23 21:27:04 -07002166 */
2167 if (mapping->nrpages) {
2168 written = invalidate_inode_pages2_range(mapping,
2169 pos >> PAGE_CACHE_SHIFT, end);
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002170 /*
2171 * If a page can not be invalidated, return 0 to fall back
2172 * to buffered write.
2173 */
2174 if (written) {
2175 if (written == -EBUSY)
2176 return 0;
Christoph Hellwiga969e902008-07-23 21:27:04 -07002177 goto out;
Hisashi Hifumi6ccfa802008-09-02 14:35:40 -07002178 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002179 }
2180
2181 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2182
2183 /*
2184 * Finally, try again to invalidate clean pages which might have been
2185 * cached by non-direct readahead, or faulted in by get_user_pages()
2186 * if the source of the write was an mmap'ed region of the file
2187 * we're writing. Either one is a pretty crazy thing to do,
2188 * so we don't support it 100%. If this invalidation
2189 * fails, tough, the write still worked...
2190 */
2191 if (mapping->nrpages) {
2192 invalidate_inode_pages2_range(mapping,
2193 pos >> PAGE_CACHE_SHIFT, end);
2194 }
2195
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 if (written > 0) {
Namhyung Kim01166512010-10-26 14:21:58 -07002197 pos += written;
2198 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2199 i_size_write(inode, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 mark_inode_dirty(inode);
2201 }
Namhyung Kim01166512010-10-26 14:21:58 -07002202 *ppos = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 }
Christoph Hellwiga969e902008-07-23 21:27:04 -07002204out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 return written;
2206}
2207EXPORT_SYMBOL(generic_file_direct_write);
2208
Nick Piggineb2be182007-10-16 01:24:57 -07002209/*
2210 * Find or create a page at the given pagecache position. Return the locked
2211 * page. This function is specifically for buffered writes.
2212 */
Nick Piggin54566b22009-01-04 12:00:53 -08002213struct page *grab_cache_page_write_begin(struct address_space *mapping,
2214 pgoff_t index, unsigned flags)
Nick Piggineb2be182007-10-16 01:24:57 -07002215{
2216 int status;
2217 struct page *page;
Nick Piggin54566b22009-01-04 12:00:53 -08002218 gfp_t gfp_notmask = 0;
2219 if (flags & AOP_FLAG_NOFS)
2220 gfp_notmask = __GFP_FS;
Nick Piggineb2be182007-10-16 01:24:57 -07002221repeat:
2222 page = find_lock_page(mapping, index);
2223 if (likely(page))
2224 return page;
2225
Nick Piggin54566b22009-01-04 12:00:53 -08002226 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002227 if (!page)
2228 return NULL;
Nick Piggin54566b22009-01-04 12:00:53 -08002229 status = add_to_page_cache_lru(page, mapping, index,
2230 GFP_KERNEL & ~gfp_notmask);
Nick Piggineb2be182007-10-16 01:24:57 -07002231 if (unlikely(status)) {
2232 page_cache_release(page);
2233 if (status == -EEXIST)
2234 goto repeat;
2235 return NULL;
2236 }
2237 return page;
2238}
Nick Piggin54566b22009-01-04 12:00:53 -08002239EXPORT_SYMBOL(grab_cache_page_write_begin);
Nick Piggineb2be182007-10-16 01:24:57 -07002240
Nick Pigginafddba42007-10-16 01:25:01 -07002241static ssize_t generic_perform_write(struct file *file,
2242 struct iov_iter *i, loff_t pos)
2243{
2244 struct address_space *mapping = file->f_mapping;
2245 const struct address_space_operations *a_ops = mapping->a_ops;
2246 long status = 0;
2247 ssize_t written = 0;
Nick Piggin674b8922007-10-16 01:25:03 -07002248 unsigned int flags = 0;
2249
2250 /*
2251 * Copies from kernel address space cannot fail (NFSD is a big user).
2252 */
2253 if (segment_eq(get_fs(), KERNEL_DS))
2254 flags |= AOP_FLAG_UNINTERRUPTIBLE;
Nick Pigginafddba42007-10-16 01:25:01 -07002255
2256 do {
2257 struct page *page;
Nick Pigginafddba42007-10-16 01:25:01 -07002258 unsigned long offset; /* Offset into pagecache page */
2259 unsigned long bytes; /* Bytes to write to page */
2260 size_t copied; /* Bytes copied from user */
2261 void *fsdata;
2262
2263 offset = (pos & (PAGE_CACHE_SIZE - 1));
Nick Pigginafddba42007-10-16 01:25:01 -07002264 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2265 iov_iter_count(i));
2266
2267again:
2268
2269 /*
2270 * Bring in the user page that we will copy from _first_.
2271 * Otherwise there's a nasty deadlock on copying from the
2272 * same page as we're writing to, without it being marked
2273 * up-to-date.
2274 *
2275 * Not only is this an optimisation, but it is also required
2276 * to check that the address is actually valid, when atomic
2277 * usercopies are used, below.
2278 */
2279 if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2280 status = -EFAULT;
2281 break;
2282 }
2283
Nick Piggin674b8922007-10-16 01:25:03 -07002284 status = a_ops->write_begin(file, mapping, pos, bytes, flags,
Nick Pigginafddba42007-10-16 01:25:01 -07002285 &page, &fsdata);
2286 if (unlikely(status))
2287 break;
2288
anfei zhou931e80e2010-02-02 13:44:02 -08002289 if (mapping_writably_mapped(mapping))
2290 flush_dcache_page(page);
2291
Nick Pigginafddba42007-10-16 01:25:01 -07002292 pagefault_disable();
2293 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2294 pagefault_enable();
2295 flush_dcache_page(page);
2296
Josef Bacikc8236db2009-07-05 12:08:18 -07002297 mark_page_accessed(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002298 status = a_ops->write_end(file, mapping, pos, bytes, copied,
2299 page, fsdata);
2300 if (unlikely(status < 0))
2301 break;
2302 copied = status;
2303
2304 cond_resched();
2305
Nick Piggin124d3b72008-02-02 15:01:17 +01002306 iov_iter_advance(i, copied);
Nick Pigginafddba42007-10-16 01:25:01 -07002307 if (unlikely(copied == 0)) {
2308 /*
2309 * If we were unable to copy any data at all, we must
2310 * fall back to a single segment length write.
2311 *
2312 * If we didn't fallback here, we could livelock
2313 * because not all segments in the iov can be copied at
2314 * once without a pagefault.
2315 */
2316 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2317 iov_iter_single_seg_count(i));
2318 goto again;
2319 }
Nick Pigginafddba42007-10-16 01:25:01 -07002320 pos += copied;
2321 written += copied;
2322
2323 balance_dirty_pages_ratelimited(mapping);
2324
2325 } while (iov_iter_count(i));
2326
2327 return written ? written : status;
2328}
2329
2330ssize_t
2331generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2332 unsigned long nr_segs, loff_t pos, loff_t *ppos,
2333 size_t count, ssize_t written)
2334{
2335 struct file *file = iocb->ki_filp;
Nick Pigginafddba42007-10-16 01:25:01 -07002336 ssize_t status;
2337 struct iov_iter i;
2338
2339 iov_iter_init(&i, iov, nr_segs, count, written);
Nick Piggin4e02ed42008-10-29 14:00:55 -07002340 status = generic_perform_write(file, &i, pos);
Nick Pigginafddba42007-10-16 01:25:01 -07002341
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 if (likely(status >= 0)) {
Nick Pigginafddba42007-10-16 01:25:01 -07002343 written += status;
2344 *ppos = pos + status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 }
2346
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 return written ? written : status;
2348}
2349EXPORT_SYMBOL(generic_file_buffered_write);
2350
Jan Karae4dd9de2009-08-17 18:10:06 +02002351/**
2352 * __generic_file_aio_write - write data to a file
2353 * @iocb: IO state structure (file, offset, etc.)
2354 * @iov: vector with data to write
2355 * @nr_segs: number of segments in the vector
2356 * @ppos: position where to write
2357 *
2358 * This function does all the work needed for actually writing data to a
2359 * file. It does all basic checks, removes SUID from the file, updates
2360 * modification times and calls proper subroutines depending on whether we
2361 * do direct IO or a standard buffered write.
2362 *
2363 * It expects i_mutex to be grabbed unless we work on a block device or similar
2364 * object which does not need locking at all.
2365 *
2366 * This function does *not* take care of syncing data in case of O_SYNC write.
2367 * A caller has to handle it. This is mainly due to the fact that we want to
2368 * avoid syncing under i_mutex.
2369 */
2370ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2371 unsigned long nr_segs, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372{
2373 struct file *file = iocb->ki_filp;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002374 struct address_space * mapping = file->f_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 size_t ocount; /* original count */
2376 size_t count; /* after file limit checks */
2377 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 loff_t pos;
2379 ssize_t written;
2380 ssize_t err;
2381
2382 ocount = 0;
Dmitriy Monakhov0ceb3312007-05-08 00:23:02 -07002383 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2384 if (err)
2385 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386
2387 count = ocount;
2388 pos = *ppos;
2389
2390 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2391
2392 /* We can write back this queue in page reclaim */
2393 current->backing_dev_info = mapping->backing_dev_info;
2394 written = 0;
2395
2396 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2397 if (err)
2398 goto out;
2399
2400 if (count == 0)
2401 goto out;
2402
Miklos Szeredi2f1936b2008-06-24 16:50:14 +02002403 err = file_remove_suid(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (err)
2405 goto out;
2406
Christoph Hellwig870f4812006-01-09 20:52:01 -08002407 file_update_time(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408
2409 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2410 if (unlikely(file->f_flags & O_DIRECT)) {
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002411 loff_t endbyte;
2412 ssize_t written_buffered;
2413
2414 written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2415 ppos, count, ocount);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 if (written < 0 || written == count)
2417 goto out;
2418 /*
2419 * direct-io write to a hole: fall through to buffered I/O
2420 * for completing the rest of the request.
2421 */
2422 pos += written;
2423 count -= written;
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002424 written_buffered = generic_file_buffered_write(iocb, iov,
2425 nr_segs, pos, ppos, count,
2426 written);
2427 /*
2428 * If generic_file_buffered_write() retuned a synchronous error
2429 * then we want to return the number of bytes which were
2430 * direct-written, or the error code if that was zero. Note
2431 * that this differs from normal direct-io semantics, which
2432 * will return -EFOO even if some bytes were written.
2433 */
2434 if (written_buffered < 0) {
2435 err = written_buffered;
2436 goto out;
2437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002439 /*
2440 * We need to ensure that the page cache pages are written to
2441 * disk and invalidated to preserve the expected O_DIRECT
2442 * semantics.
2443 */
2444 endbyte = pos + written_buffered - written - 1;
Christoph Hellwigc05c4ed2009-09-23 15:07:30 +02002445 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
Jeff Moyerfb5527e2006-10-19 23:28:13 -07002446 if (err == 0) {
2447 written = written_buffered;
2448 invalidate_mapping_pages(mapping,
2449 pos >> PAGE_CACHE_SHIFT,
2450 endbyte >> PAGE_CACHE_SHIFT);
2451 } else {
2452 /*
2453 * We don't know how much we wrote, so just return
2454 * the number of bytes which were direct-written
2455 */
2456 }
2457 } else {
2458 written = generic_file_buffered_write(iocb, iov, nr_segs,
2459 pos, ppos, count, written);
2460 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461out:
2462 current->backing_dev_info = NULL;
2463 return written ? written : err;
2464}
Jan Karae4dd9de2009-08-17 18:10:06 +02002465EXPORT_SYMBOL(__generic_file_aio_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Jan Karae4dd9de2009-08-17 18:10:06 +02002467/**
2468 * generic_file_aio_write - write data to a file
2469 * @iocb: IO state structure
2470 * @iov: vector with data to write
2471 * @nr_segs: number of segments in the vector
2472 * @pos: position in file where to write
2473 *
2474 * This is a wrapper around __generic_file_aio_write() to be used by most
2475 * filesystems. It takes care of syncing the file in case of O_SYNC file
2476 * and acquires i_mutex as needed.
2477 */
Badari Pulavarty027445c2006-09-30 23:28:46 -07002478ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2479 unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480{
2481 struct file *file = iocb->ki_filp;
Jan Kara148f9482009-08-17 19:52:36 +02002482 struct inode *inode = file->f_mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
2485 BUG_ON(iocb->ki_pos != pos);
2486
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002487 mutex_lock(&inode->i_mutex);
Jan Karae4dd9de2009-08-17 18:10:06 +02002488 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002489 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490
Jan Kara148f9482009-08-17 19:52:36 +02002491 if (ret > 0 || ret == -EIOCBQUEUED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 ssize_t err;
2493
Jan Kara148f9482009-08-17 19:52:36 +02002494 err = generic_write_sync(file, pos, ret);
Jan Karac7b50db2009-08-18 16:18:20 +02002495 if (err < 0 && ret > 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 ret = err;
2497 }
2498 return ret;
2499}
2500EXPORT_SYMBOL(generic_file_aio_write);
2501
David Howellscf9a2ae2006-08-29 19:05:54 +01002502/**
2503 * try_to_release_page() - release old fs-specific metadata on a page
2504 *
2505 * @page: the page which the kernel is trying to free
2506 * @gfp_mask: memory allocation flags (and I/O mode)
2507 *
2508 * The address_space is to try to release any data against the page
2509 * (presumably at page->private). If the release was successful, return `1'.
2510 * Otherwise return zero.
2511 *
David Howells266cf652009-04-03 16:42:36 +01002512 * This may also be called if PG_fscache is set on a page, indicating that the
2513 * page is known to the local caching routines.
2514 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002515 * The @gfp_mask argument specifies whether I/O may be performed to release
Mingming Cao3f31fdd2008-07-25 01:46:22 -07002516 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
David Howellscf9a2ae2006-08-29 19:05:54 +01002517 *
David Howellscf9a2ae2006-08-29 19:05:54 +01002518 */
2519int try_to_release_page(struct page *page, gfp_t gfp_mask)
2520{
2521 struct address_space * const mapping = page->mapping;
2522
2523 BUG_ON(!PageLocked(page));
2524 if (PageWriteback(page))
2525 return 0;
2526
2527 if (mapping && mapping->a_ops->releasepage)
2528 return mapping->a_ops->releasepage(page, gfp_mask);
2529 return try_to_free_buffers(page);
2530}
2531
2532EXPORT_SYMBOL(try_to_release_page);