blob: cbfe13bf5b2aa3f39b4845fe4fce4f45a02b5f43 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/inode.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -070015 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
17 *
Mingming Cao617ba132006-10-11 01:20:53 -070018 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
Dave Kleikampac27a0e2006-10-11 01:20:50 -070019 */
20
Dave Kleikampac27a0e2006-10-11 01:20:50 -070021#include <linux/fs.h>
22#include <linux/time.h>
Mingming Caodab291a2006-10-11 01:21:01 -070023#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070024#include <linux/highuid.h>
25#include <linux/pagemap.h>
26#include <linux/quotaops.h>
27#include <linux/string.h>
28#include <linux/buffer_head.h>
29#include <linux/writeback.h>
Alex Tomas64769242008-07-11 19:27:31 -040030#include <linux/pagevec.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031#include <linux/mpage.h>
Duane Griffine83c1392008-12-19 20:47:15 +000032#include <linux/namei.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070033#include <linux/uio.h>
34#include <linux/bio.h>
Mingming Cao4c0425f2009-09-28 15:48:41 -040035#include <linux/workqueue.h>
Jiaying Zhang744692d2010-03-04 16:14:02 -050036#include <linux/kernel.h>
Andrew Morton6db26ff2011-01-12 16:59:13 -080037#include <linux/printk.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090038#include <linux/slab.h>
Theodore Ts'oa8901d32010-12-17 10:40:47 -050039#include <linux/ratelimit.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040040
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040041#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070042#include "xattr.h"
43#include "acl.h"
Theodore Ts'o9f125d62011-06-27 19:16:04 -040044#include "truncate.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070045
Theodore Ts'o9bffad12009-06-17 11:48:11 -040046#include <trace/events/ext4.h>
47
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040048#define MPAGE_DA_EXTENT_TAIL 0x01
49
Darrick J. Wong814525f2012-04-29 18:31:10 -040050static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
51 struct ext4_inode_info *ei)
52{
53 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
54 __u16 csum_lo;
55 __u16 csum_hi = 0;
56 __u32 csum;
57
58 csum_lo = raw->i_checksum_lo;
59 raw->i_checksum_lo = 0;
60 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
61 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
62 csum_hi = raw->i_checksum_hi;
63 raw->i_checksum_hi = 0;
64 }
65
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
67 EXT4_INODE_SIZE(inode->i_sb));
68
69 raw->i_checksum_lo = csum_lo;
70 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
71 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
72 raw->i_checksum_hi = csum_hi;
73
74 return csum;
75}
76
77static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
78 struct ext4_inode_info *ei)
79{
80 __u32 provided, calculated;
81
82 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
83 cpu_to_le32(EXT4_OS_LINUX) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
86 return 1;
87
88 provided = le16_to_cpu(raw->i_checksum_lo);
89 calculated = ext4_inode_csum(inode, raw, ei);
90 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
91 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
92 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
93 else
94 calculated &= 0xFFFF;
95
96 return provided == calculated;
97}
98
99static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
100 struct ext4_inode_info *ei)
101{
102 __u32 csum;
103
104 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
105 cpu_to_le32(EXT4_OS_LINUX) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
108 return;
109
110 csum = ext4_inode_csum(inode, raw, ei);
111 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
112 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
113 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
114 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
115}
116
Jan Kara678aaf42008-07-11 19:27:31 -0400117static inline int ext4_begin_ordered_truncate(struct inode *inode,
118 loff_t new_size)
119{
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500120 trace_ext4_begin_ordered_truncate(inode, new_size);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500121 /*
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
126 */
127 if (!EXT4_I(inode)->jinode)
128 return 0;
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
130 EXT4_I(inode)->jinode,
131 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -0400132}
133
Alex Tomas64769242008-07-11 19:27:31 -0400134static void ext4_invalidatepage(struct page *page, unsigned long offset);
Theodore Ts'ocb20d512010-10-27 21:30:09 -0400135static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
136 struct buffer_head *bh_result, int create);
137static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode);
138static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate);
139static int __ext4_journalled_writepage(struct page *page, unsigned int len);
140static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
Eric Sandeen5f163cc2012-01-04 22:33:28 -0500141static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
142 struct inode *inode, struct page *page, loff_t from,
143 loff_t length, int flags);
Alex Tomas64769242008-07-11 19:27:31 -0400144
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700145/*
146 * Test whether an inode is a fast symlink.
147 */
Mingming Cao617ba132006-10-11 01:20:53 -0700148static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700149{
Mingming Cao617ba132006-10-11 01:20:53 -0700150 int ea_blocks = EXT4_I(inode)->i_file_acl ?
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700151 (inode->i_sb->s_blocksize >> 9) : 0;
152
153 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
154}
155
156/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
159 * this transaction.
160 */
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500161int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
Jan Kara487caee2009-08-17 22:17:20 -0400162 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700163{
Jan Kara487caee2009-08-17 22:17:20 -0400164 int ret;
165
166 /*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
Jan Kara487caee2009-08-17 22:17:20 -0400168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
171 */
Frank Mayhar03901312009-01-07 00:06:22 -0500172 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700173 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400174 up_write(&EXT4_I(inode)->i_data_sem);
Amir Goldstein8e8eaab2011-02-27 23:32:12 -0500175 ret = ext4_journal_restart(handle, nblocks);
Jan Kara487caee2009-08-17 22:17:20 -0400176 down_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500177 ext4_discard_preallocations(inode);
Jan Kara487caee2009-08-17 22:17:20 -0400178
179 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700180}
181
182/*
183 * Called at the last iput() if i_nlink is zero.
184 */
Al Viro0930fcc2010-06-07 13:16:22 -0400185void ext4_evict_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700186{
187 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400188 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700189
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500190 trace_ext4_evict_inode(inode);
Jiaying Zhang2581fdc2011-08-13 12:17:13 -0400191
Jiaying Zhang2581fdc2011-08-13 12:17:13 -0400192 ext4_ioend_wait(inode);
193
Al Viro0930fcc2010-06-07 13:16:22 -0400194 if (inode->i_nlink) {
Jan Kara2d859db2011-07-26 09:07:11 -0400195 /*
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
209 *
210 * Note that directories do not have this problem because they
211 * don't use page cache.
212 */
213 if (ext4_should_journal_data(inode) &&
214 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) {
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
217
218 jbd2_log_start_commit(journal, commit_tid);
219 jbd2_log_wait_commit(journal, commit_tid);
220 filemap_write_and_wait(&inode->i_data);
221 }
Al Viro0930fcc2010-06-07 13:16:22 -0400222 truncate_inode_pages(&inode->i_data, 0);
223 goto no_delete;
224 }
225
Christoph Hellwig907f4552010-03-03 09:05:06 -0500226 if (!is_bad_inode(inode))
Christoph Hellwig871a2932010-03-03 09:05:07 -0500227 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500228
Jan Kara678aaf42008-07-11 19:27:31 -0400229 if (ext4_should_order_data(inode))
230 ext4_begin_ordered_truncate(inode, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231 truncate_inode_pages(&inode->i_data, 0);
232
233 if (is_bad_inode(inode))
234 goto no_delete;
235
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200236 /*
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
239 */
240 sb_start_intwrite(inode->i_sb);
Theodore Ts'o9f125d62011-06-27 19:16:04 -0400241 handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700242 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400243 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700244 /*
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
247 * cleaned up.
248 */
Mingming Cao617ba132006-10-11 01:20:53 -0700249 ext4_orphan_del(NULL, inode);
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200250 sb_end_intwrite(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251 goto no_delete;
252 }
253
254 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500255 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700256 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400257 err = ext4_mark_inode_dirty(handle, inode);
258 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500259 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400260 "couldn't mark inode dirty (err %d)", err);
261 goto stop_handle;
262 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700264 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400265
266 /*
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
271 */
Frank Mayhar03901312009-01-07 00:06:22 -0500272 if (!ext4_handle_has_enough_credits(handle, 3)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400273 err = ext4_journal_extend(handle, 3);
274 if (err > 0)
275 err = ext4_journal_restart(handle, 3);
276 if (err != 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500277 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400278 "couldn't extend journal (err %d)", err);
279 stop_handle:
280 ext4_journal_stop(handle);
Theodore Ts'o45388212010-07-29 15:06:10 -0400281 ext4_orphan_del(NULL, inode);
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200282 sb_end_intwrite(inode->i_sb);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400283 goto no_delete;
284 }
285 }
286
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700287 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700288 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700289 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700290 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700292 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700293 * (Well, we could do this if we need to, but heck - it works)
294 */
Mingming Cao617ba132006-10-11 01:20:53 -0700295 ext4_orphan_del(handle, inode);
296 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700297
298 /*
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
303 * fails.
304 */
Mingming Cao617ba132006-10-11 01:20:53 -0700305 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700306 /* If that failed, just do the required in-core inode clear. */
Al Viro0930fcc2010-06-07 13:16:22 -0400307 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700308 else
Mingming Cao617ba132006-10-11 01:20:53 -0700309 ext4_free_inode(handle, inode);
310 ext4_journal_stop(handle);
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200311 sb_end_intwrite(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700312 return;
313no_delete:
Al Viro0930fcc2010-06-07 13:16:22 -0400314 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315}
316
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300317#ifdef CONFIG_QUOTA
318qsize_t *ext4_get_reserved_space(struct inode *inode)
Mingming Cao60e58e02009-01-22 18:13:05 +0100319{
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300320 return &EXT4_I(inode)->i_reserved_quota;
Mingming Cao60e58e02009-01-22 18:13:05 +0100321}
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300322#endif
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500323
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400324/*
325 * Calculate the number of metadata blocks need to reserve
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500326 * to allocate a block located at @lblock
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400327 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -0500328static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400329{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400330 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500331 return ext4_ext_calc_metadata_amount(inode, lblock);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400332
Amir Goldstein8bb2b242011-06-27 17:10:28 -0400333 return ext4_ind_calc_metadata_amount(inode, lblock);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400334}
335
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500336/*
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
339 */
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500340void ext4_da_update_reserve_space(struct inode *inode,
341 int used, int quota_claim)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400342{
343 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500344 struct ext4_inode_info *ei = EXT4_I(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400345
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500346 spin_lock(&ei->i_block_reservation_lock);
Aditya Kalid8990242011-09-09 19:18:51 -0400347 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500348 if (unlikely(used > ei->i_reserved_data_blocks)) {
349 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d "
Theodore Ts'o1084f252012-03-19 23:13:43 -0400350 "with only %d reserved data blocks",
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500351 __func__, inode->i_ino, used,
352 ei->i_reserved_data_blocks);
353 WARN_ON(1);
354 used = ei->i_reserved_data_blocks;
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -0400355 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400356
Brian Foster97795d22012-07-22 23:59:40 -0400357 if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) {
358 ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d "
359 "with only %d reserved metadata blocks\n", __func__,
360 inode->i_ino, ei->i_allocated_meta_blocks,
361 ei->i_reserved_meta_blocks);
362 WARN_ON(1);
363 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
364 }
365
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500366 /* Update per-inode reservations */
367 ei->i_reserved_data_blocks -= used;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500368 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
Theodore Ts'o57042652011-09-09 18:56:51 -0400369 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400370 used + ei->i_allocated_meta_blocks);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500371 ei->i_allocated_meta_blocks = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500372
373 if (ei->i_reserved_data_blocks == 0) {
374 /*
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
377 * allocation blocks.
378 */
Theodore Ts'o57042652011-09-09 18:56:51 -0400379 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400380 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -0500381 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500382 ei->i_da_metadata_calc_len = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500383 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400384 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +0100385
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400386 /* Update quota subsystem for data blocks */
387 if (quota_claim)
Aditya Kali7b415bf2011-09-09 19:04:51 -0400388 dquot_claim_block(inode, EXT4_C2B(sbi, used));
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400389 else {
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500390 /*
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400393 * not re-claim the quota for fallocated blocks.
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500394 */
Aditya Kali7b415bf2011-09-09 19:04:51 -0400395 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500396 }
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400397
398 /*
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
402 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500403 if ((ei->i_reserved_data_blocks == 0) &&
404 (atomic_read(&inode->i_writecount) == 0))
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400405 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400406}
407
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400408static int __check_block_validity(struct inode *inode, const char *func,
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400409 unsigned int line,
410 struct ext4_map_blocks *map)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400411{
Theodore Ts'o24676da2010-05-16 21:00:00 -0400412 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
413 map->m_len)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400414 ext4_error_inode(inode, func, line, map->m_pblk,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map->m_lblk,
417 map->m_len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400418 return -EIO;
419 }
420 return 0;
421}
422
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400423#define check_block_validity(inode, map) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400424 __check_block_validity((inode), __func__, __LINE__, (map))
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400425
Mingming Caof5ab0d12008-02-25 15:29:55 -0500426/*
Theodore Ts'o1f945332009-09-30 22:57:41 -0400427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
Theodore Ts'o55138e02009-09-29 13:31:31 -0400429 */
430static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx,
431 unsigned int max_pages)
432{
433 struct address_space *mapping = inode->i_mapping;
434 pgoff_t index;
435 struct pagevec pvec;
436 pgoff_t num = 0;
437 int i, nr_pages, done = 0;
438
439 if (max_pages == 0)
440 return 0;
441 pagevec_init(&pvec, 0);
442 while (!done) {
443 index = idx;
444 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
445 PAGECACHE_TAG_DIRTY,
446 (pgoff_t)PAGEVEC_SIZE);
447 if (nr_pages == 0)
448 break;
449 for (i = 0; i < nr_pages; i++) {
450 struct page *page = pvec.pages[i];
451 struct buffer_head *bh, *head;
452
453 lock_page(page);
454 if (unlikely(page->mapping != mapping) ||
455 !PageDirty(page) ||
456 PageWriteback(page) ||
457 page->index != idx) {
458 done = 1;
459 unlock_page(page);
460 break;
461 }
Theodore Ts'o1f945332009-09-30 22:57:41 -0400462 if (page_has_buffers(page)) {
463 bh = head = page_buffers(page);
464 do {
465 if (!buffer_delay(bh) &&
466 !buffer_unwritten(bh))
467 done = 1;
468 bh = bh->b_this_page;
469 } while (!done && (bh != head));
470 }
Theodore Ts'o55138e02009-09-29 13:31:31 -0400471 unlock_page(page);
472 if (done)
473 break;
474 idx++;
475 num++;
Eric Sandeen659c6002010-10-27 21:30:03 -0400476 if (num >= max_pages) {
477 done = 1;
Theodore Ts'o55138e02009-09-29 13:31:31 -0400478 break;
Eric Sandeen659c6002010-10-27 21:30:03 -0400479 }
Theodore Ts'o55138e02009-09-29 13:31:31 -0400480 }
481 pagevec_release(&pvec);
482 }
483 return num;
484}
485
486/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400487 * The ext4_map_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400488 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -0500489 *
Mingming Caof5ab0d12008-02-25 15:29:55 -0500490 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
491 * and store the allocated blocks in the result buffer head and mark it
492 * mapped.
493 *
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400494 * If file type is extents based, it will call ext4_ext_map_blocks(),
495 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -0500496 * based files
497 *
498 * On success, it returns the number of blocks being mapped or allocate.
499 * if create==0 and the blocks are pre-allocated and uninitialized block,
500 * the result buffer head is unmapped. If the create ==1, it will make sure
501 * the buffer head is mapped.
502 *
503 * It returns 0 if plain look up failed (blocks have not been allocated), in
Tao Madf3ab172011-10-08 15:53:49 -0400504 * that case, buffer head is unmapped
Mingming Caof5ab0d12008-02-25 15:29:55 -0500505 *
506 * It returns the error in case of allocation failure.
507 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400508int ext4_map_blocks(handle_t *handle, struct inode *inode,
509 struct ext4_map_blocks *map, int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500510{
511 int retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -0500512
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400513 map->m_flags = 0;
514 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
515 "logical block %lu\n", inode->i_ino, flags, map->m_len,
516 (unsigned long) map->m_lblk);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500517 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -0400518 * Try to see if we can get the block without requesting a new
519 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500520 */
Zheng Liu729f52c2012-07-09 16:29:29 -0400521 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
522 down_read((&EXT4_I(inode)->i_data_sem));
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400523 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400524 retval = ext4_ext_map_blocks(handle, inode, map, flags &
525 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500526 } else {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400527 retval = ext4_ind_map_blocks(handle, inode, map, flags &
528 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500529 }
Zheng Liu729f52c2012-07-09 16:29:29 -0400530 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
531 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -0500532
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400533 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Zheng Liu51865fd2012-11-08 21:57:32 -0500534 int ret;
535 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
536 /* delayed alloc may be allocated by fallocate and
537 * coverted to initialized by directIO.
538 * we need to handle delayed extent here.
539 */
540 down_write((&EXT4_I(inode)->i_data_sem));
541 goto delayed_mapped;
542 }
543 ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400544 if (ret != 0)
545 return ret;
546 }
547
Mingming Caof5ab0d12008-02-25 15:29:55 -0500548 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400549 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500550 return retval;
551
552 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500553 * Returns if the blocks have already allocated
554 *
555 * Note that if blocks have been preallocated
Tao Madf3ab172011-10-08 15:53:49 -0400556 * ext4_ext_get_block() returns the create = 0
Mingming Caof5ab0d12008-02-25 15:29:55 -0500557 * with buffer head unmapped.
558 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400559 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
Mingming Caof5ab0d12008-02-25 15:29:55 -0500560 return retval;
561
562 /*
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400563 * When we call get_blocks without the create flag, the
564 * BH_Unwritten flag could have gotten set if the blocks
565 * requested were part of a uninitialized extent. We need to
566 * clear this flag now that we are committed to convert all or
567 * part of the uninitialized extent to be an initialized
568 * extent. This is because we need to avoid the combination
569 * of BH_Unwritten and BH_Mapped flags being simultaneously
570 * set on the buffer_head.
571 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400572 map->m_flags &= ~EXT4_MAP_UNWRITTEN;
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400573
574 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500575 * New blocks allocate and/or writing to uninitialized extent
576 * will possibly result in updating i_data, so we take
577 * the write lock of i_data_sem, and call get_blocks()
578 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500579 */
580 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -0400581
582 /*
583 * if the caller is from delayed allocation writeout path
584 * we have already reserved fs blocks for allocation
585 * let the underlying get_block() function know to
586 * avoid double accounting
587 */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400588 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Theodore Ts'of2321092011-01-10 12:12:36 -0500589 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500590 /*
591 * We need to check for EXT4 here because migrate
592 * could have changed the inode type in between
593 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400594 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400595 retval = ext4_ext_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500596 } else {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400597 retval = ext4_ind_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400598
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400599 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400600 /*
601 * We allocated new blocks which will result in
602 * i_data's format changing. Force the migrate
603 * to fail by clearing migrate flags
604 */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -0500605 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400606 }
Mingming Caod2a17632008-07-14 17:52:37 -0400607
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500608 /*
609 * Update reserved blocks/metadata blocks after successful
610 * block allocation which had been deferred till now. We don't
611 * support fallocate for non extent files. So we can update
612 * reserve space here.
613 */
614 if ((retval > 0) &&
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -0500615 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500616 ext4_da_update_reserve_space(inode, retval, 1);
617 }
Aditya Kali5356f2612011-09-09 19:20:51 -0400618 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
Theodore Ts'of2321092011-01-10 12:12:36 -0500619 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -0400620
Zheng Liu51865fd2012-11-08 21:57:32 -0500621 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
622 int ret;
Zheng Liu51865fd2012-11-08 21:57:32 -0500623delayed_mapped:
624 /* delayed allocation blocks has been allocated */
625 ret = ext4_es_remove_extent(inode, map->m_lblk,
626 map->m_len);
627 if (ret < 0)
628 retval = ret;
629 }
Aditya Kali5356f2612011-09-09 19:20:51 -0400630 }
631
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500632 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400633 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400634 int ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400635 if (ret != 0)
636 return ret;
637 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500638 return retval;
639}
640
Mingming Caof3bd1f32008-08-19 22:16:03 -0400641/* Maximum number of blocks we map for direct IO at once. */
642#define DIO_MAX_BLOCKS 4096
643
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400644static int _ext4_get_block(struct inode *inode, sector_t iblock,
645 struct buffer_head *bh, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700646{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800647 handle_t *handle = ext4_journal_current_handle();
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400648 struct ext4_map_blocks map;
Jan Kara7fb54092008-02-10 01:08:38 -0500649 int ret = 0, started = 0;
Mingming Caof3bd1f32008-08-19 22:16:03 -0400650 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700651
Tao Ma46c7f252012-12-10 14:04:52 -0500652 if (ext4_has_inline_data(inode))
653 return -ERANGE;
654
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400655 map.m_lblk = iblock;
656 map.m_len = bh->b_size >> inode->i_blkbits;
657
Anatol Pomozov8b0f1652012-11-08 15:07:16 -0500658 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
Jan Kara7fb54092008-02-10 01:08:38 -0500659 /* Direct IO write... */
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400660 if (map.m_len > DIO_MAX_BLOCKS)
661 map.m_len = DIO_MAX_BLOCKS;
662 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
Mingming Caof3bd1f32008-08-19 22:16:03 -0400663 handle = ext4_journal_start(inode, dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -0500664 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700665 ret = PTR_ERR(handle);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400666 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700667 }
Jan Kara7fb54092008-02-10 01:08:38 -0500668 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700669 }
670
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400671 ret = ext4_map_blocks(handle, inode, &map, flags);
Jan Kara7fb54092008-02-10 01:08:38 -0500672 if (ret > 0) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400673 map_bh(bh, inode->i_sb, map.m_pblk);
674 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
675 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
Jan Kara7fb54092008-02-10 01:08:38 -0500676 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700677 }
Jan Kara7fb54092008-02-10 01:08:38 -0500678 if (started)
679 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700680 return ret;
681}
682
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400683int ext4_get_block(struct inode *inode, sector_t iblock,
684 struct buffer_head *bh, int create)
685{
686 return _ext4_get_block(inode, iblock, bh,
687 create ? EXT4_GET_BLOCKS_CREATE : 0);
688}
689
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700690/*
691 * `handle' can be NULL if create is zero
692 */
Mingming Cao617ba132006-10-11 01:20:53 -0700693struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500694 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700695{
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400696 struct ext4_map_blocks map;
697 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700698 int fatal = 0, err;
699
700 J_ASSERT(handle != NULL || create == 0);
701
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400702 map.m_lblk = block;
703 map.m_len = 1;
704 err = ext4_map_blocks(handle, inode, &map,
705 create ? EXT4_GET_BLOCKS_CREATE : 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700706
Carlos Maiolino90b0a972012-09-17 23:39:12 -0400707 /* ensure we send some value back into *errp */
708 *errp = 0;
709
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400710 if (err < 0)
711 *errp = err;
712 if (err <= 0)
713 return NULL;
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400714
715 bh = sb_getblk(inode->i_sb, map.m_pblk);
716 if (!bh) {
717 *errp = -EIO;
718 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400720 if (map.m_flags & EXT4_MAP_NEW) {
721 J_ASSERT(create != 0);
722 J_ASSERT(handle != NULL);
723
724 /*
725 * Now that we do not always journal data, we should
726 * keep in mind whether this should always journal the
727 * new buffer as metadata. For now, regular file
728 * writes use ext4_get_block instead, so it's not a
729 * problem.
730 */
731 lock_buffer(bh);
732 BUFFER_TRACE(bh, "call get_create_access");
733 fatal = ext4_journal_get_create_access(handle, bh);
734 if (!fatal && !buffer_uptodate(bh)) {
735 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
736 set_buffer_uptodate(bh);
737 }
738 unlock_buffer(bh);
739 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
740 err = ext4_handle_dirty_metadata(handle, inode, bh);
741 if (!fatal)
742 fatal = err;
743 } else {
744 BUFFER_TRACE(bh, "not a new buffer");
745 }
746 if (fatal) {
747 *errp = fatal;
748 brelse(bh);
749 bh = NULL;
750 }
751 return bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700752}
753
Mingming Cao617ba132006-10-11 01:20:53 -0700754struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500755 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700756{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400757 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700758
Mingming Cao617ba132006-10-11 01:20:53 -0700759 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700760 if (!bh)
761 return bh;
762 if (buffer_uptodate(bh))
763 return bh;
Christoph Hellwig65299a32011-08-23 14:50:29 +0200764 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700765 wait_on_buffer(bh);
766 if (buffer_uptodate(bh))
767 return bh;
768 put_bh(bh);
769 *err = -EIO;
770 return NULL;
771}
772
Tao Maf19d5872012-12-10 14:05:51 -0500773int ext4_walk_page_buffers(handle_t *handle,
774 struct buffer_head *head,
775 unsigned from,
776 unsigned to,
777 int *partial,
778 int (*fn)(handle_t *handle,
779 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700780{
781 struct buffer_head *bh;
782 unsigned block_start, block_end;
783 unsigned blocksize = head->b_size;
784 int err, ret = 0;
785 struct buffer_head *next;
786
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400787 for (bh = head, block_start = 0;
788 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400789 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700790 next = bh->b_this_page;
791 block_end = block_start + blocksize;
792 if (block_end <= from || block_start >= to) {
793 if (partial && !buffer_uptodate(bh))
794 *partial = 1;
795 continue;
796 }
797 err = (*fn)(handle, bh);
798 if (!ret)
799 ret = err;
800 }
801 return ret;
802}
803
804/*
805 * To preserve ordering, it is essential that the hole instantiation and
806 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -0700807 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -0700808 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700809 * prepare_write() is the right place.
810 *
Mingming Cao617ba132006-10-11 01:20:53 -0700811 * Also, this function can nest inside ext4_writepage() ->
812 * block_write_full_page(). In that case, we *know* that ext4_writepage()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700813 * has generated enough buffer credits to do the whole page. So we won't
814 * block on the journal in that case, which is good, because the caller may
815 * be PF_MEMALLOC.
816 *
Mingming Cao617ba132006-10-11 01:20:53 -0700817 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700818 * quota file writes. If we were to commit the transaction while thus
819 * reentered, there can be a deadlock - we would be holding a quota
820 * lock, and the commit would never complete if another thread had a
821 * transaction open and was blocking on the quota lock - a ranking
822 * violation.
823 *
Mingming Caodab291a2006-10-11 01:21:01 -0700824 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700825 * will _not_ run commit under these circumstances because handle->h_ref
826 * is elevated. We'll still have enough credits for the tiny quotafile
827 * write.
828 */
Tao Maf19d5872012-12-10 14:05:51 -0500829int do_journal_get_write_access(handle_t *handle,
830 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831{
Jan Kara56d35a42010-08-05 14:41:42 -0400832 int dirty = buffer_dirty(bh);
833 int ret;
834
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700835 if (!buffer_mapped(bh) || buffer_freed(bh))
836 return 0;
Jan Kara56d35a42010-08-05 14:41:42 -0400837 /*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200838 * __block_write_begin() could have dirtied some buffers. Clean
Jan Kara56d35a42010-08-05 14:41:42 -0400839 * the dirty bit as jbd2_journal_get_write_access() could complain
840 * otherwise about fs integrity issues. Setting of the dirty bit
Christoph Hellwigebdec242010-10-06 10:47:23 +0200841 * by __block_write_begin() isn't a real problem here as we clear
Jan Kara56d35a42010-08-05 14:41:42 -0400842 * the bit before releasing a page lock and thus writeback cannot
843 * ever write the buffer.
844 */
845 if (dirty)
846 clear_buffer_dirty(bh);
847 ret = ext4_journal_get_write_access(handle, bh);
848 if (!ret && dirty)
849 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
850 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700851}
852
Anatol Pomozov8b0f1652012-11-08 15:07:16 -0500853static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
854 struct buffer_head *bh_result, int create);
Nick Pigginbfc1af62007-10-16 01:25:05 -0700855static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400856 loff_t pos, unsigned len, unsigned flags,
857 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700858{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400859 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400860 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861 handle_t *handle;
862 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400863 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400864 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400865 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700866
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400867 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400868 /*
869 * Reserve one block more for addition to orphan list in case
870 * we allocate blocks but write fails for some reason
871 */
872 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400873 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400874 from = pos & (PAGE_CACHE_SIZE - 1);
875 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700876
Tao Maf19d5872012-12-10 14:05:51 -0500877 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
878 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
879 flags, pagep);
880 if (ret < 0)
881 goto out;
882 if (ret == 1) {
883 ret = 0;
884 goto out;
885 }
886 }
887
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888retry:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400889 handle = ext4_journal_start(inode, needed_blocks);
890 if (IS_ERR(handle)) {
891 ret = PTR_ERR(handle);
892 goto out;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700893 }
894
Jan Karaebd36102009-02-22 21:09:59 -0500895 /* We cannot recurse into the filesystem as the transaction is already
896 * started */
897 flags |= AOP_FLAG_NOFS;
898
Nick Piggin54566b22009-01-04 12:00:53 -0800899 page = grab_cache_page_write_begin(mapping, index, flags);
Jan Karacf108bc2008-07-11 19:27:31 -0400900 if (!page) {
901 ext4_journal_stop(handle);
902 ret = -ENOMEM;
903 goto out;
904 }
Tao Maf19d5872012-12-10 14:05:51 -0500905
Jan Karacf108bc2008-07-11 19:27:31 -0400906 *pagep = page;
907
Jiaying Zhang744692d2010-03-04 16:14:02 -0500908 if (ext4_should_dioread_nolock(inode))
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200909 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
Jiaying Zhang744692d2010-03-04 16:14:02 -0500910 else
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200911 ret = __block_write_begin(page, pos, len, ext4_get_block);
Nick Pigginbfc1af62007-10-16 01:25:05 -0700912
913 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -0500914 ret = ext4_walk_page_buffers(handle, page_buffers(page),
915 from, to, NULL,
916 do_journal_get_write_access);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700917 }
Nick Pigginbfc1af62007-10-16 01:25:05 -0700918
919 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400920 unlock_page(page);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400921 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -0400922 /*
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200923 * __block_write_begin may have instantiated a few blocks
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -0400924 * outside i_size. Trim these off again. Don't need
925 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400926 *
927 * Add inode to orphan list in case we crash before
928 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -0400929 */
Jan Karaffacfa72009-07-13 16:22:22 -0400930 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400931 ext4_orphan_add(handle, inode);
932
933 ext4_journal_stop(handle);
934 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -0500935 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400936 /*
Jan Karaffacfa72009-07-13 16:22:22 -0400937 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400938 * still be on the orphan list; we need to
939 * make sure the inode is removed from the
940 * orphan list in that case.
941 */
942 if (inode->i_nlink)
943 ext4_orphan_del(NULL, inode);
944 }
Nick Pigginbfc1af62007-10-16 01:25:05 -0700945 }
946
Mingming Cao617ba132006-10-11 01:20:53 -0700947 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948 goto retry;
Andrew Morton7479d2b2007-04-01 23:49:44 -0700949out:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700950 return ret;
951}
952
Nick Pigginbfc1af62007-10-16 01:25:05 -0700953/* For write_end() in data=journal mode */
954static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700955{
956 if (!buffer_mapped(bh) || buffer_freed(bh))
957 return 0;
958 set_buffer_uptodate(bh);
Frank Mayhar03901312009-01-07 00:06:22 -0500959 return ext4_handle_dirty_metadata(handle, NULL, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700960}
961
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400962static int ext4_generic_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400963 struct address_space *mapping,
964 loff_t pos, unsigned len, unsigned copied,
965 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400966{
967 int i_size_changed = 0;
968 struct inode *inode = mapping->host;
969 handle_t *handle = ext4_journal_current_handle();
970
Tao Maf19d5872012-12-10 14:05:51 -0500971 if (ext4_has_inline_data(inode))
972 copied = ext4_write_inline_data_end(inode, pos, len,
973 copied, page);
974 else
975 copied = block_write_end(file, mapping, pos,
976 len, copied, page, fsdata);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -0400977
978 /*
979 * No need to use i_size_read() here, the i_size
980 * cannot change under us because we hold i_mutex.
981 *
982 * But it's important to update i_size while still holding page lock:
983 * page writeout could otherwise come in and zero beyond i_size.
984 */
985 if (pos + copied > inode->i_size) {
986 i_size_write(inode, pos + copied);
987 i_size_changed = 1;
988 }
989
990 if (pos + copied > EXT4_I(inode)->i_disksize) {
991 /* We need to mark inode dirty even if
992 * new_i_size is less that inode->i_size
993 * bu greater than i_disksize.(hint delalloc)
994 */
995 ext4_update_i_disksize(inode, (pos + copied));
996 i_size_changed = 1;
997 }
998 unlock_page(page);
999 page_cache_release(page);
1000
1001 /*
1002 * Don't mark the inode dirty under page lock. First, it unnecessarily
1003 * makes the holding time of page lock longer. Second, it forces lock
1004 * ordering of page lock and transaction start for journaling
1005 * filesystems.
1006 */
1007 if (i_size_changed)
1008 ext4_mark_inode_dirty(handle, inode);
1009
1010 return copied;
1011}
1012
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001013/*
1014 * We need to pick up the new inode size which generic_commit_write gave us
1015 * `file' can be NULL - eg, when called from page_symlink().
1016 *
Mingming Cao617ba132006-10-11 01:20:53 -07001017 * ext4 never places buffers on inode->i_mapping->private_list. metadata
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001018 * buffers are managed internally.
1019 */
Nick Pigginbfc1af62007-10-16 01:25:05 -07001020static int ext4_ordered_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001021 struct address_space *mapping,
1022 loff_t pos, unsigned len, unsigned copied,
1023 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001024{
Mingming Cao617ba132006-10-11 01:20:53 -07001025 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001026 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001027 int ret = 0, ret2;
1028
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001029 trace_ext4_ordered_write_end(inode, pos, len, copied);
Jan Kara678aaf42008-07-11 19:27:31 -04001030 ret = ext4_jbd2_file_inode(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001031
1032 if (ret == 0) {
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001033 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001034 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001035 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001036 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001037 /* if we have allocated more blocks and copied
1038 * less. We will have blocks allocated outside
1039 * inode->i_size. So truncate them
1040 */
1041 ext4_orphan_add(handle, inode);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001042 if (ret2 < 0)
1043 ret = ret2;
Akira Fujita09e08342011-10-20 18:56:10 -04001044 } else {
1045 unlock_page(page);
1046 page_cache_release(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001047 }
Akira Fujita09e08342011-10-20 18:56:10 -04001048
Mingming Cao617ba132006-10-11 01:20:53 -07001049 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001050 if (!ret)
1051 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001052
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001053 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001054 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001055 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001056 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001057 * on the orphan list; we need to make sure the inode
1058 * is removed from the orphan list in that case.
1059 */
1060 if (inode->i_nlink)
1061 ext4_orphan_del(NULL, inode);
1062 }
1063
1064
Nick Pigginbfc1af62007-10-16 01:25:05 -07001065 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066}
1067
Nick Pigginbfc1af62007-10-16 01:25:05 -07001068static int ext4_writeback_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001069 struct address_space *mapping,
1070 loff_t pos, unsigned len, unsigned copied,
1071 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001072{
Mingming Cao617ba132006-10-11 01:20:53 -07001073 handle_t *handle = ext4_journal_current_handle();
Jan Karacf108bc2008-07-11 19:27:31 -04001074 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001075 int ret = 0, ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001076
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001077 trace_ext4_writeback_write_end(inode, pos, len, copied);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001078 ret2 = ext4_generic_write_end(file, mapping, pos, len, copied,
Nick Pigginbfc1af62007-10-16 01:25:05 -07001079 page, fsdata);
Roel Kluinf8a87d82008-04-29 22:01:18 -04001080 copied = ret2;
Jan Karaffacfa72009-07-13 16:22:22 -04001081 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001082 /* if we have allocated more blocks and copied
1083 * less. We will have blocks allocated outside
1084 * inode->i_size. So truncate them
1085 */
1086 ext4_orphan_add(handle, inode);
1087
Roel Kluinf8a87d82008-04-29 22:01:18 -04001088 if (ret2 < 0)
1089 ret = ret2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001090
Mingming Cao617ba132006-10-11 01:20:53 -07001091 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001092 if (!ret)
1093 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001094
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001095 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001096 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001097 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001098 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001099 * on the orphan list; we need to make sure the inode
1100 * is removed from the orphan list in that case.
1101 */
1102 if (inode->i_nlink)
1103 ext4_orphan_del(NULL, inode);
1104 }
1105
Nick Pigginbfc1af62007-10-16 01:25:05 -07001106 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001107}
1108
Nick Pigginbfc1af62007-10-16 01:25:05 -07001109static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001110 struct address_space *mapping,
1111 loff_t pos, unsigned len, unsigned copied,
1112 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001113{
Mingming Cao617ba132006-10-11 01:20:53 -07001114 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001115 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001116 int ret = 0, ret2;
1117 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001118 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001119 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001120
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001121 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001122 from = pos & (PAGE_CACHE_SIZE - 1);
1123 to = from + len;
1124
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001125 BUG_ON(!ext4_handle_valid(handle));
1126
Tao Ma3fdcfb62012-12-10 14:05:57 -05001127 if (ext4_has_inline_data(inode))
1128 copied = ext4_write_inline_data_end(inode, pos, len,
1129 copied, page);
1130 else {
1131 if (copied < len) {
1132 if (!PageUptodate(page))
1133 copied = 0;
1134 page_zero_new_buffers(page, from+copied, to);
1135 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001136
Tao Ma3fdcfb62012-12-10 14:05:57 -05001137 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1138 to, &partial, write_end_fn);
1139 if (!partial)
1140 SetPageUptodate(page);
1141 }
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001142 new_i_size = pos + copied;
1143 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001144 i_size_write(inode, pos+copied);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001145 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Jan Kara2d859db2011-07-26 09:07:11 -04001146 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001147 if (new_i_size > EXT4_I(inode)->i_disksize) {
1148 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001149 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001150 if (!ret)
1151 ret = ret2;
1152 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001153
Jan Karacf108bc2008-07-11 19:27:31 -04001154 unlock_page(page);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001155 page_cache_release(page);
Jan Karaffacfa72009-07-13 16:22:22 -04001156 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001157 /* if we have allocated more blocks and copied
1158 * less. We will have blocks allocated outside
1159 * inode->i_size. So truncate them
1160 */
1161 ext4_orphan_add(handle, inode);
1162
Mingming Cao617ba132006-10-11 01:20:53 -07001163 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164 if (!ret)
1165 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001166 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001167 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001168 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001169 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001170 * on the orphan list; we need to make sure the inode
1171 * is removed from the orphan list in that case.
1172 */
1173 if (inode->i_nlink)
1174 ext4_orphan_del(NULL, inode);
1175 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001176
1177 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001178}
Mingming Caod2a17632008-07-14 17:52:37 -04001179
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001180/*
Aditya Kali7b415bf2011-09-09 19:04:51 -04001181 * Reserve a single cluster located at lblock
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001182 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -05001183static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -04001184{
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001185 int retries = 0;
Mingming Cao60e58e02009-01-22 18:13:05 +01001186 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001187 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001188 unsigned int md_needed;
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001189 int ret;
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001190 ext4_lblk_t save_last_lblock;
1191 int save_len;
Mingming Caod2a17632008-07-14 17:52:37 -04001192
Mingming Cao60e58e02009-01-22 18:13:05 +01001193 /*
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001194 * We will charge metadata quota at writeout time; this saves
1195 * us from metadata over-estimation, though we may go over by
1196 * a small amount in the end. Here we just reserve for data.
Mingming Cao60e58e02009-01-22 18:13:05 +01001197 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04001198 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001199 if (ret)
1200 return ret;
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001201
1202 /*
1203 * recalculate the amount of metadata blocks to reserve
1204 * in order to allocate nrblocks
1205 * worse case is one extent per block
1206 */
1207repeat:
1208 spin_lock(&ei->i_block_reservation_lock);
1209 /*
1210 * ext4_calc_metadata_amount() has side effects, which we have
1211 * to be prepared undo if we fail to claim space.
1212 */
1213 save_len = ei->i_da_metadata_calc_len;
1214 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1215 md_needed = EXT4_NUM_B2C(sbi,
1216 ext4_calc_metadata_amount(inode, lblock));
1217 trace_ext4_da_reserve_space(inode, md_needed);
1218
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001219 /*
1220 * We do still charge estimated metadata to the sb though;
1221 * we cannot afford to run out of free blocks.
1222 */
Theodore Ts'oe7d5f312011-09-09 19:14:51 -04001223 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001224 ei->i_da_metadata_calc_len = save_len;
1225 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1226 spin_unlock(&ei->i_block_reservation_lock);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001227 if (ext4_should_retry_alloc(inode->i_sb, &retries)) {
1228 yield();
1229 goto repeat;
1230 }
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001231 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
Mingming Caod2a17632008-07-14 17:52:37 -04001232 return -ENOSPC;
1233 }
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001234 ei->i_reserved_data_blocks++;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001235 ei->i_reserved_meta_blocks += md_needed;
1236 spin_unlock(&ei->i_block_reservation_lock);
Dmitry Monakhov39bc6802009-12-10 16:36:27 +00001237
Mingming Caod2a17632008-07-14 17:52:37 -04001238 return 0; /* success */
1239}
1240
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001241static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001242{
1243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001244 struct ext4_inode_info *ei = EXT4_I(inode);
Mingming Caod2a17632008-07-14 17:52:37 -04001245
Mingming Caocd213222008-08-19 22:16:59 -04001246 if (!to_free)
1247 return; /* Nothing to release, exit */
1248
Mingming Caod2a17632008-07-14 17:52:37 -04001249 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001250
Li Zefan5a58ec82010-05-17 02:00:00 -04001251 trace_ext4_da_release_space(inode, to_free);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001252 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
Mingming Caocd213222008-08-19 22:16:59 -04001253 /*
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001254 * if there aren't enough reserved blocks, then the
1255 * counter is messed up somewhere. Since this
1256 * function is called from invalidate page, it's
1257 * harmless to return without any action.
Mingming Caocd213222008-08-19 22:16:59 -04001258 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001259 ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: "
1260 "ino %lu, to_free %d with only %d reserved "
Theodore Ts'o1084f252012-03-19 23:13:43 -04001261 "data blocks", inode->i_ino, to_free,
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001262 ei->i_reserved_data_blocks);
1263 WARN_ON(1);
1264 to_free = ei->i_reserved_data_blocks;
1265 }
1266 ei->i_reserved_data_blocks -= to_free;
1267
1268 if (ei->i_reserved_data_blocks == 0) {
1269 /*
1270 * We can release all of the reserved metadata blocks
1271 * only when we have written all of the delayed
1272 * allocation blocks.
Aditya Kali7b415bf2011-09-09 19:04:51 -04001273 * Note that in case of bigalloc, i_reserved_meta_blocks,
1274 * i_reserved_data_blocks, etc. refer to number of clusters.
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001275 */
Theodore Ts'o57042652011-09-09 18:56:51 -04001276 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001277 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -05001278 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001279 ei->i_da_metadata_calc_len = 0;
Mingming Caocd213222008-08-19 22:16:59 -04001280 }
1281
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001282 /* update fs dirty data blocks counter */
Theodore Ts'o57042652011-09-09 18:56:51 -04001283 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
Mingming Caod2a17632008-07-14 17:52:37 -04001284
Mingming Caod2a17632008-07-14 17:52:37 -04001285 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001286
Aditya Kali7b415bf2011-09-09 19:04:51 -04001287 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
Mingming Caod2a17632008-07-14 17:52:37 -04001288}
1289
1290static void ext4_da_page_release_reservation(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001291 unsigned long offset)
Mingming Caod2a17632008-07-14 17:52:37 -04001292{
1293 int to_release = 0;
1294 struct buffer_head *head, *bh;
1295 unsigned int curr_off = 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04001296 struct inode *inode = page->mapping->host;
1297 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1298 int num_clusters;
Zheng Liu51865fd2012-11-08 21:57:32 -05001299 ext4_fsblk_t lblk;
Mingming Caod2a17632008-07-14 17:52:37 -04001300
1301 head = page_buffers(page);
1302 bh = head;
1303 do {
1304 unsigned int next_off = curr_off + bh->b_size;
1305
1306 if ((offset <= curr_off) && (buffer_delay(bh))) {
1307 to_release++;
1308 clear_buffer_delay(bh);
1309 }
1310 curr_off = next_off;
1311 } while ((bh = bh->b_this_page) != head);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001312
Zheng Liu51865fd2012-11-08 21:57:32 -05001313 if (to_release) {
1314 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1315 ext4_es_remove_extent(inode, lblk, to_release);
1316 }
1317
Aditya Kali7b415bf2011-09-09 19:04:51 -04001318 /* If we have released all the blocks belonging to a cluster, then we
1319 * need to release the reserved space for that cluster. */
1320 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1321 while (num_clusters > 0) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04001322 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1323 ((num_clusters - 1) << sbi->s_cluster_bits);
1324 if (sbi->s_cluster_ratio == 1 ||
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05001325 !ext4_find_delalloc_cluster(inode, lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04001326 ext4_da_release_space(inode, 1);
1327
1328 num_clusters--;
1329 }
Mingming Caod2a17632008-07-14 17:52:37 -04001330}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001331
1332/*
Alex Tomas64769242008-07-11 19:27:31 -04001333 * Delayed allocation stuff
1334 */
1335
Alex Tomas64769242008-07-11 19:27:31 -04001336/*
1337 * mpage_da_submit_io - walks through extent of pages and try to write
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001338 * them with writepage() call back
Alex Tomas64769242008-07-11 19:27:31 -04001339 *
1340 * @mpd->inode: inode
1341 * @mpd->first_page: first page of the extent
1342 * @mpd->next_page: page after the last page of the extent
Alex Tomas64769242008-07-11 19:27:31 -04001343 *
1344 * By the time mpage_da_submit_io() is called we expect all blocks
1345 * to be allocated. this may be wrong if allocation failed.
1346 *
1347 * As pages are already locked by write_cache_pages(), we can't use it
1348 */
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001349static int mpage_da_submit_io(struct mpage_da_data *mpd,
1350 struct ext4_map_blocks *map)
Alex Tomas64769242008-07-11 19:27:31 -04001351{
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001352 struct pagevec pvec;
1353 unsigned long index, end;
1354 int ret = 0, err, nr_pages, i;
1355 struct inode *inode = mpd->inode;
1356 struct address_space *mapping = inode->i_mapping;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001357 loff_t size = i_size_read(inode);
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04001358 unsigned int len, block_start;
1359 struct buffer_head *bh, *page_bufs = NULL;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001360 int journal_data = ext4_should_journal_data(inode);
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001361 sector_t pblock = 0, cur_logical = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001362 struct ext4_io_submit io_submit;
Alex Tomas64769242008-07-11 19:27:31 -04001363
1364 BUG_ON(mpd->next_page <= mpd->first_page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001365 memset(&io_submit, 0, sizeof(io_submit));
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001366 /*
1367 * We need to start from the first_page to the next_page - 1
1368 * to make sure we also write the mapped dirty buffer_heads.
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001369 * If we look at mpd->b_blocknr we would only be looking
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001370 * at the currently mapped buffer_heads.
1371 */
Alex Tomas64769242008-07-11 19:27:31 -04001372 index = mpd->first_page;
1373 end = mpd->next_page - 1;
1374
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001375 pagevec_init(&pvec, 0);
Alex Tomas64769242008-07-11 19:27:31 -04001376 while (index <= end) {
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001377 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
Alex Tomas64769242008-07-11 19:27:31 -04001378 if (nr_pages == 0)
1379 break;
1380 for (i = 0; i < nr_pages; i++) {
Theodore Ts'o97498952011-02-26 14:08:01 -05001381 int commit_write = 0, skip_page = 0;
Alex Tomas64769242008-07-11 19:27:31 -04001382 struct page *page = pvec.pages[i];
1383
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001384 index = page->index;
1385 if (index > end)
1386 break;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001387
1388 if (index == size >> PAGE_CACHE_SHIFT)
1389 len = size & ~PAGE_CACHE_MASK;
1390 else
1391 len = PAGE_CACHE_SIZE;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001392 if (map) {
1393 cur_logical = index << (PAGE_CACHE_SHIFT -
1394 inode->i_blkbits);
1395 pblock = map->m_pblk + (cur_logical -
1396 map->m_lblk);
1397 }
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001398 index++;
1399
1400 BUG_ON(!PageLocked(page));
1401 BUG_ON(PageWriteback(page));
1402
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001403 /*
1404 * If the page does not have buffers (for
1405 * whatever reason), try to create them using
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001406 * __block_write_begin. If this fails,
Theodore Ts'o97498952011-02-26 14:08:01 -05001407 * skip the page and move on.
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001408 */
1409 if (!page_has_buffers(page)) {
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04001410 if (__block_write_begin(page, 0, len,
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001411 noalloc_get_block_write)) {
Theodore Ts'o97498952011-02-26 14:08:01 -05001412 skip_page:
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001413 unlock_page(page);
1414 continue;
1415 }
1416 commit_write = 1;
1417 }
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04001418
1419 bh = page_bufs = page_buffers(page);
1420 block_start = 0;
1421 do {
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001422 if (!bh)
Theodore Ts'o97498952011-02-26 14:08:01 -05001423 goto skip_page;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001424 if (map && (cur_logical >= map->m_lblk) &&
1425 (cur_logical <= (map->m_lblk +
1426 (map->m_len - 1)))) {
1427 if (buffer_delay(bh)) {
1428 clear_buffer_delay(bh);
1429 bh->b_blocknr = pblock;
1430 }
1431 if (buffer_unwritten(bh) ||
1432 buffer_mapped(bh))
1433 BUG_ON(bh->b_blocknr != pblock);
1434 if (map->m_flags & EXT4_MAP_UNINIT)
1435 set_buffer_uninit(bh);
1436 clear_buffer_unwritten(bh);
1437 }
1438
Yongqiang Yang13a79a42011-12-13 21:51:55 -05001439 /*
1440 * skip page if block allocation undone and
1441 * block is dirty
1442 */
1443 if (ext4_bh_delay_or_unwritten(NULL, bh))
Theodore Ts'o97498952011-02-26 14:08:01 -05001444 skip_page = 1;
Theodore Ts'o3ecdb3a2010-10-27 21:30:10 -04001445 bh = bh->b_this_page;
1446 block_start += bh->b_size;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001447 cur_logical++;
1448 pblock++;
1449 } while (bh != page_bufs);
1450
Theodore Ts'o97498952011-02-26 14:08:01 -05001451 if (skip_page)
1452 goto skip_page;
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001453
1454 if (commit_write)
1455 /* mark the buffer_heads as dirty & uptodate */
1456 block_commit_write(page, 0, len);
1457
Theodore Ts'o97498952011-02-26 14:08:01 -05001458 clear_page_dirty_for_io(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001459 /*
1460 * Delalloc doesn't support data journalling,
1461 * but eventually maybe we'll lift this
1462 * restriction.
1463 */
1464 if (unlikely(journal_data && PageChecked(page)))
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001465 err = __ext4_journalled_writepage(page, len);
Theodore Ts'o14490322010-12-14 15:27:50 -05001466 else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT))
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001467 err = ext4_bio_write_page(&io_submit, page,
1468 len, mpd->wbc);
Theodore Ts'o9dd75f12011-08-13 12:58:21 -04001469 else if (buffer_uninit(page_bufs)) {
1470 ext4_set_bh_endio(page_bufs, inode);
1471 err = block_write_full_page_endio(page,
1472 noalloc_get_block_write,
1473 mpd->wbc, ext4_end_io_buffer_write);
1474 } else
Theodore Ts'o14490322010-12-14 15:27:50 -05001475 err = block_write_full_page(page,
1476 noalloc_get_block_write, mpd->wbc);
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001477
1478 if (!err)
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001479 mpd->pages_written++;
Alex Tomas64769242008-07-11 19:27:31 -04001480 /*
1481 * In error case, we have to continue because
1482 * remaining pages are still locked
Alex Tomas64769242008-07-11 19:27:31 -04001483 */
1484 if (ret == 0)
1485 ret = err;
1486 }
1487 pagevec_release(&pvec);
1488 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001489 ext4_io_submit(&io_submit);
Alex Tomas64769242008-07-11 19:27:31 -04001490 return ret;
1491}
1492
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001493static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001494{
1495 int nr_pages, i;
1496 pgoff_t index, end;
1497 struct pagevec pvec;
1498 struct inode *inode = mpd->inode;
1499 struct address_space *mapping = inode->i_mapping;
Zheng Liu51865fd2012-11-08 21:57:32 -05001500 ext4_lblk_t start, last;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001501
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001502 index = mpd->first_page;
1503 end = mpd->next_page - 1;
Zheng Liu51865fd2012-11-08 21:57:32 -05001504
1505 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1506 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1507 ext4_es_remove_extent(inode, start, last - start + 1);
1508
Eric Sandeen66bea922012-11-14 22:22:05 -05001509 pagevec_init(&pvec, 0);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001510 while (index <= end) {
1511 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1512 if (nr_pages == 0)
1513 break;
1514 for (i = 0; i < nr_pages; i++) {
1515 struct page *page = pvec.pages[i];
Jan Kara9b1d09982010-03-03 16:19:32 -05001516 if (page->index > end)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001517 break;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001518 BUG_ON(!PageLocked(page));
1519 BUG_ON(PageWriteback(page));
1520 block_invalidatepage(page, 0);
1521 ClearPageUptodate(page);
1522 unlock_page(page);
1523 }
Jan Kara9b1d09982010-03-03 16:19:32 -05001524 index = pvec.pages[nr_pages - 1]->index + 1;
1525 pagevec_release(&pvec);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001526 }
1527 return;
1528}
1529
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001530static void ext4_print_free_blocks(struct inode *inode)
1531{
1532 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o92b97812012-03-19 23:41:49 -04001533 struct super_block *sb = inode->i_sb;
1534
1535 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
Theodore Ts'o5dee5432011-09-09 19:10:51 -04001536 EXT4_C2B(EXT4_SB(inode->i_sb),
1537 ext4_count_free_clusters(inode->i_sb)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001538 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1539 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
Theodore Ts'o57042652011-09-09 18:56:51 -04001540 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1541 percpu_counter_sum(&sbi->s_freeclusters_counter)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001542 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
Aditya Kali7b415bf2011-09-09 19:04:51 -04001543 (long long) EXT4_C2B(EXT4_SB(inode->i_sb),
1544 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001545 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1546 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1547 EXT4_I(inode)->i_reserved_data_blocks);
1548 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
Theodore Ts'o16939182009-09-26 17:43:59 -04001549 EXT4_I(inode)->i_reserved_meta_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001550 return;
1551}
1552
Theodore Ts'ob920c752009-05-14 00:54:29 -04001553/*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001554 * mpage_da_map_and_submit - go through given space, map them
1555 * if necessary, and then submit them for I/O
Alex Tomas64769242008-07-11 19:27:31 -04001556 *
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001557 * @mpd - bh describing space
Alex Tomas64769242008-07-11 19:27:31 -04001558 *
1559 * The function skips space we know is already mapped to disk blocks.
1560 *
Alex Tomas64769242008-07-11 19:27:31 -04001561 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001562static void mpage_da_map_and_submit(struct mpage_da_data *mpd)
Alex Tomas64769242008-07-11 19:27:31 -04001563{
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001564 int err, blks, get_blocks_flags;
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001565 struct ext4_map_blocks map, *mapp = NULL;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001566 sector_t next = mpd->b_blocknr;
1567 unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits;
1568 loff_t disksize = EXT4_I(mpd->inode)->i_disksize;
1569 handle_t *handle = NULL;
Alex Tomas64769242008-07-11 19:27:31 -04001570
1571 /*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001572 * If the blocks are mapped already, or we couldn't accumulate
1573 * any blocks, then proceed immediately to the submission stage.
Alex Tomas64769242008-07-11 19:27:31 -04001574 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001575 if ((mpd->b_size == 0) ||
1576 ((mpd->b_state & (1 << BH_Mapped)) &&
1577 !(mpd->b_state & (1 << BH_Delay)) &&
1578 !(mpd->b_state & (1 << BH_Unwritten))))
1579 goto submit_io;
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001580
1581 handle = ext4_journal_current_handle();
1582 BUG_ON(!handle);
1583
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04001584 /*
Eric Sandeen79e83032010-07-27 11:56:07 -04001585 * Call ext4_map_blocks() to allocate any delayed allocation
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001586 * blocks, or to convert an uninitialized extent to be
1587 * initialized (in the case where we have written into
1588 * one or more preallocated blocks).
1589 *
1590 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1591 * indicate that we are on the delayed allocation path. This
1592 * affects functions in many different parts of the allocation
1593 * call path. This flag exists primarily because we don't
Eric Sandeen79e83032010-07-27 11:56:07 -04001594 * want to change *many* call functions, so ext4_map_blocks()
Theodore Ts'of2321092011-01-10 12:12:36 -05001595 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001596 * inode's allocation semaphore is taken.
1597 *
1598 * If the blocks in questions were delalloc blocks, set
1599 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1600 * variables are updated after the blocks have been allocated.
Aneesh Kumar K.V79ffab32009-05-13 15:13:42 -04001601 */
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001602 map.m_lblk = next;
1603 map.m_len = max_blocks;
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05001604 get_blocks_flags = EXT4_GET_BLOCKS_CREATE;
Jiaying Zhang744692d2010-03-04 16:14:02 -05001605 if (ext4_should_dioread_nolock(mpd->inode))
1606 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -04001607 if (mpd->b_state & (1 << BH_Delay))
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -05001608 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
1609
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001610 blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001611 if (blks < 0) {
Eric Sandeene3570632010-07-27 11:56:08 -04001612 struct super_block *sb = mpd->inode->i_sb;
1613
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001614 err = blks;
Theodore Ts'oed5bde02009-02-23 10:48:07 -05001615 /*
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001616 * If get block returns EAGAIN or ENOSPC and there
Theodore Ts'o97498952011-02-26 14:08:01 -05001617 * appears to be free blocks we will just let
1618 * mpage_da_submit_io() unlock all of the pages.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001619 */
1620 if (err == -EAGAIN)
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001621 goto submit_io;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001622
Theodore Ts'o5dee5432011-09-09 19:10:51 -04001623 if (err == -ENOSPC && ext4_count_free_clusters(sb)) {
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001624 mpd->retval = err;
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001625 goto submit_io;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001626 }
1627
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001628 /*
Theodore Ts'oed5bde02009-02-23 10:48:07 -05001629 * get block failure will cause us to loop in
1630 * writepages, because a_ops->writepage won't be able
1631 * to make progress. The page will be redirtied by
1632 * writepage and writepages will again try to write
1633 * the same.
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001634 */
Eric Sandeene3570632010-07-27 11:56:08 -04001635 if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) {
1636 ext4_msg(sb, KERN_CRIT,
1637 "delayed block allocation failed for inode %lu "
1638 "at logical offset %llu with max blocks %zd "
1639 "with error %d", mpd->inode->i_ino,
1640 (unsigned long long) next,
1641 mpd->b_size >> mpd->inode->i_blkbits, err);
1642 ext4_msg(sb, KERN_CRIT,
1643 "This should not happen!! Data will be lost\n");
1644 if (err == -ENOSPC)
1645 ext4_print_free_blocks(mpd->inode);
Aneesh Kumar K.V030ba6b2008-09-08 23:14:50 -04001646 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001647 /* invalidate all the pages */
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001648 ext4_da_block_invalidatepages(mpd);
Curt Wohlgemuthe0fd9b92011-02-26 12:25:52 -05001649
1650 /* Mark this page range as having been completed */
1651 mpd->io_done = 1;
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001652 return;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001653 }
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001654 BUG_ON(blks == 0);
1655
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001656 mapp = &map;
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001657 if (map.m_flags & EXT4_MAP_NEW) {
1658 struct block_device *bdev = mpd->inode->i_sb->s_bdev;
1659 int i;
Alex Tomas64769242008-07-11 19:27:31 -04001660
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001661 for (i = 0; i < map.m_len; i++)
1662 unmap_underlying_metadata(bdev, map.m_pblk + i);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001663 }
1664
1665 /*
Jan Kara03f5d8b2009-06-09 00:17:05 -04001666 * Update on-disk size along with block allocation.
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001667 */
1668 disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits;
1669 if (disksize > i_size_read(mpd->inode))
1670 disksize = i_size_read(mpd->inode);
1671 if (disksize > EXT4_I(mpd->inode)->i_disksize) {
1672 ext4_update_i_disksize(mpd->inode, disksize);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001673 err = ext4_mark_inode_dirty(handle, mpd->inode);
1674 if (err)
1675 ext4_error(mpd->inode->i_sb,
1676 "Failed to mark inode %lu dirty",
1677 mpd->inode->i_ino);
Theodore Ts'o2fa3cdf2009-05-14 09:29:45 -04001678 }
1679
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001680submit_io:
Theodore Ts'o1de3e3d2010-10-27 21:30:10 -04001681 mpage_da_submit_io(mpd, mapp);
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001682 mpd->io_done = 1;
Alex Tomas64769242008-07-11 19:27:31 -04001683}
1684
Aneesh Kumar K.Vbf068ee2008-08-19 22:16:43 -04001685#define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1686 (1 << BH_Delay) | (1 << BH_Unwritten))
Alex Tomas64769242008-07-11 19:27:31 -04001687
1688/*
1689 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1690 *
1691 * @mpd->lbh - extent of blocks
1692 * @logical - logical number of the block in the file
1693 * @bh - bh of the block (used to access block's state)
1694 *
1695 * the function is used to collect contig. blocks in same state
1696 */
1697static void mpage_add_bh_to_extent(struct mpage_da_data *mpd,
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001698 sector_t logical, size_t b_size,
1699 unsigned long b_state)
Alex Tomas64769242008-07-11 19:27:31 -04001700{
Alex Tomas64769242008-07-11 19:27:31 -04001701 sector_t next;
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001702 int nrblocks = mpd->b_size >> mpd->inode->i_blkbits;
Alex Tomas64769242008-07-11 19:27:31 -04001703
Eric Sandeenc445e3e2010-05-16 04:00:00 -04001704 /*
1705 * XXX Don't go larger than mballoc is willing to allocate
1706 * This is a stopgap solution. We eventually need to fold
1707 * mpage_da_submit_io() into this function and then call
Eric Sandeen79e83032010-07-27 11:56:07 -04001708 * ext4_map_blocks() multiple times in a loop
Eric Sandeenc445e3e2010-05-16 04:00:00 -04001709 */
1710 if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize)
1711 goto flush_it;
1712
Mingming Cao525f4ed2008-08-19 22:15:58 -04001713 /* check if thereserved journal credits might overflow */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001714 if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) {
Mingming Cao525f4ed2008-08-19 22:15:58 -04001715 if (nrblocks >= EXT4_MAX_TRANS_DATA) {
1716 /*
1717 * With non-extent format we are limited by the journal
1718 * credit available. Total credit needed to insert
1719 * nrblocks contiguous blocks is dependent on the
1720 * nrblocks. So limit nrblocks.
1721 */
1722 goto flush_it;
1723 } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) >
1724 EXT4_MAX_TRANS_DATA) {
1725 /*
1726 * Adding the new buffer_head would make it cross the
1727 * allowed limit for which we have journal credit
1728 * reserved. So limit the new bh->b_size
1729 */
1730 b_size = (EXT4_MAX_TRANS_DATA - nrblocks) <<
1731 mpd->inode->i_blkbits;
1732 /* we will do mpage_da_submit_io in the next loop */
1733 }
1734 }
Alex Tomas64769242008-07-11 19:27:31 -04001735 /*
1736 * First block in the extent
1737 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001738 if (mpd->b_size == 0) {
1739 mpd->b_blocknr = logical;
1740 mpd->b_size = b_size;
1741 mpd->b_state = b_state & BH_FLAGS;
Alex Tomas64769242008-07-11 19:27:31 -04001742 return;
1743 }
1744
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001745 next = mpd->b_blocknr + nrblocks;
Alex Tomas64769242008-07-11 19:27:31 -04001746 /*
1747 * Can we merge the block to our big extent?
1748 */
Theodore Ts'o8dc207c2009-02-23 06:46:01 -05001749 if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) {
1750 mpd->b_size += b_size;
Alex Tomas64769242008-07-11 19:27:31 -04001751 return;
1752 }
1753
Mingming Cao525f4ed2008-08-19 22:15:58 -04001754flush_it:
Alex Tomas64769242008-07-11 19:27:31 -04001755 /*
1756 * We couldn't merge the block to our extent, so we
1757 * need to flush current extent and start new one
1758 */
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04001759 mpage_da_map_and_submit(mpd);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04001760 return;
Alex Tomas64769242008-07-11 19:27:31 -04001761}
1762
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001763static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001764{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001765 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001766}
1767
Alex Tomas64769242008-07-11 19:27:31 -04001768/*
Aditya Kali5356f2612011-09-09 19:20:51 -04001769 * This function is grabs code from the very beginning of
1770 * ext4_map_blocks, but assumes that the caller is from delayed write
1771 * time. This function looks up the requested blocks and sets the
1772 * buffer delay bit under the protection of i_data_sem.
1773 */
1774static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1775 struct ext4_map_blocks *map,
1776 struct buffer_head *bh)
1777{
1778 int retval;
1779 sector_t invalid_block = ~((sector_t) 0xffff);
1780
1781 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1782 invalid_block = ~0;
1783
1784 map->m_flags = 0;
1785 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1786 "logical block %lu\n", inode->i_ino, map->m_len,
1787 (unsigned long) map->m_lblk);
1788 /*
1789 * Try to see if we can get the block without requesting a new
1790 * file system block.
1791 */
1792 down_read((&EXT4_I(inode)->i_data_sem));
Tao Ma9c3569b2012-12-10 14:05:57 -05001793 if (ext4_has_inline_data(inode)) {
1794 /*
1795 * We will soon create blocks for this page, and let
1796 * us pretend as if the blocks aren't allocated yet.
1797 * In case of clusters, we have to handle the work
1798 * of mapping from cluster so that the reserved space
1799 * is calculated properly.
1800 */
1801 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1802 ext4_find_delalloc_cluster(inode, map->m_lblk))
1803 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1804 retval = 0;
1805 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Aditya Kali5356f2612011-09-09 19:20:51 -04001806 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1807 else
1808 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1809
1810 if (retval == 0) {
1811 /*
1812 * XXX: __block_prepare_write() unmaps passed block,
1813 * is it OK?
1814 */
1815 /* If the block was allocated from previously allocated cluster,
1816 * then we dont need to reserve it again. */
1817 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
1818 retval = ext4_da_reserve_space(inode, iblock);
1819 if (retval)
1820 /* not enough space to reserve */
1821 goto out_unlock;
1822 }
1823
Zheng Liu51865fd2012-11-08 21:57:32 -05001824 retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len);
1825 if (retval)
1826 goto out_unlock;
1827
Aditya Kali5356f2612011-09-09 19:20:51 -04001828 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1829 * and it should not appear on the bh->b_state.
1830 */
1831 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1832
1833 map_bh(bh, inode->i_sb, invalid_block);
1834 set_buffer_new(bh);
1835 set_buffer_delay(bh);
1836 }
1837
1838out_unlock:
1839 up_read((&EXT4_I(inode)->i_data_sem));
1840
1841 return retval;
1842}
1843
1844/*
Theodore Ts'ob920c752009-05-14 00:54:29 -04001845 * This is a special get_blocks_t callback which is used by
1846 * ext4_da_write_begin(). It will either return mapped block or
1847 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001848 *
1849 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1850 * We also have b_blocknr = -1 and b_bdev initialized properly
1851 *
1852 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1853 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1854 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04001855 */
Tao Ma9c3569b2012-12-10 14:05:57 -05001856int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1857 struct buffer_head *bh, int create)
Alex Tomas64769242008-07-11 19:27:31 -04001858{
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001859 struct ext4_map_blocks map;
Alex Tomas64769242008-07-11 19:27:31 -04001860 int ret = 0;
1861
1862 BUG_ON(create == 0);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001863 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1864
1865 map.m_lblk = iblock;
1866 map.m_len = 1;
Alex Tomas64769242008-07-11 19:27:31 -04001867
1868 /*
1869 * first, we need to know whether the block is allocated already
1870 * preallocated blocks are unmapped but should treated
1871 * the same as allocated blocks.
1872 */
Aditya Kali5356f2612011-09-09 19:20:51 -04001873 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1874 if (ret <= 0)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001875 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04001876
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001877 map_bh(bh, inode->i_sb, map.m_pblk);
1878 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1879
1880 if (buffer_unwritten(bh)) {
1881 /* A delayed write to unwritten bh should be marked
1882 * new and mapped. Mapped ensures that we don't do
1883 * get_block multiple times when we write to the same
1884 * offset and new ensures that we do proper zero out
1885 * for partial write.
1886 */
1887 set_buffer_new(bh);
Theodore Ts'oc8205632011-04-10 22:30:07 -04001888 set_buffer_mapped(bh);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001889 }
1890 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04001891}
Mingming Cao61628a32008-07-11 19:27:31 -04001892
Theodore Ts'ob920c752009-05-14 00:54:29 -04001893/*
1894 * This function is used as a standard get_block_t calback function
1895 * when there is no desire to allocate any blocks. It is used as a
Christoph Hellwigebdec242010-10-06 10:47:23 +02001896 * callback function for block_write_begin() and block_write_full_page().
Christoph Hellwig206f7ab2010-06-14 14:42:49 -04001897 * These functions should only try to map a single block at a time.
Theodore Ts'ob920c752009-05-14 00:54:29 -04001898 *
1899 * Since this function doesn't do block allocations even if the caller
1900 * requests it by passing in create=1, it is critically important that
1901 * any caller checks to make sure that any buffer heads are returned
1902 * by this function are either all already mapped or marked for
Christoph Hellwig206f7ab2010-06-14 14:42:49 -04001903 * delayed allocation before calling block_write_full_page(). Otherwise,
1904 * b_blocknr could be left unitialized, and the page write functions will
1905 * be taken by surprise.
Theodore Ts'ob920c752009-05-14 00:54:29 -04001906 */
1907static int noalloc_get_block_write(struct inode *inode, sector_t iblock,
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001908 struct buffer_head *bh_result, int create)
1909{
Theodore Ts'oa2dc52b2009-05-12 13:51:29 -04001910 BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001911 return _ext4_get_block(inode, iblock, bh_result, 0);
Mingming Cao61628a32008-07-11 19:27:31 -04001912}
1913
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001914static int bget_one(handle_t *handle, struct buffer_head *bh)
1915{
1916 get_bh(bh);
1917 return 0;
1918}
1919
1920static int bput_one(handle_t *handle, struct buffer_head *bh)
1921{
1922 put_bh(bh);
1923 return 0;
1924}
1925
1926static int __ext4_journalled_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001927 unsigned int len)
1928{
1929 struct address_space *mapping = page->mapping;
1930 struct inode *inode = mapping->host;
Tao Ma3fdcfb62012-12-10 14:05:57 -05001931 struct buffer_head *page_bufs = NULL;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001932 handle_t *handle = NULL;
Tao Ma3fdcfb62012-12-10 14:05:57 -05001933 int ret = 0, err = 0;
1934 int inline_data = ext4_has_inline_data(inode);
1935 struct buffer_head *inode_bh = NULL;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001936
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001937 ClearPageChecked(page);
Tao Ma3fdcfb62012-12-10 14:05:57 -05001938
1939 if (inline_data) {
1940 BUG_ON(page->index != 0);
1941 BUG_ON(len > ext4_get_max_inline_size(inode));
1942 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1943 if (inode_bh == NULL)
1944 goto out;
1945 } else {
1946 page_bufs = page_buffers(page);
1947 if (!page_bufs) {
1948 BUG();
1949 goto out;
1950 }
1951 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1952 NULL, bget_one);
1953 }
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001954 /* As soon as we unlock the page, it can go away, but we have
1955 * references to buffers so we are safe */
1956 unlock_page(page);
1957
1958 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
1959 if (IS_ERR(handle)) {
1960 ret = PTR_ERR(handle);
1961 goto out;
1962 }
1963
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001964 BUG_ON(!ext4_handle_valid(handle));
1965
Tao Ma3fdcfb62012-12-10 14:05:57 -05001966 if (inline_data) {
1967 ret = ext4_journal_get_write_access(handle, inode_bh);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001968
Tao Ma3fdcfb62012-12-10 14:05:57 -05001969 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1970
1971 } else {
1972 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1973 do_journal_get_write_access);
1974
1975 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1976 write_end_fn);
1977 }
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001978 if (ret == 0)
1979 ret = err;
Jan Kara2d859db2011-07-26 09:07:11 -04001980 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001981 err = ext4_journal_stop(handle);
1982 if (!ret)
1983 ret = err;
1984
Tao Ma3fdcfb62012-12-10 14:05:57 -05001985 if (!ext4_has_inline_data(inode))
1986 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1987 NULL, bput_one);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001988 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001989out:
Tao Ma3fdcfb62012-12-10 14:05:57 -05001990 brelse(inode_bh);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001991 return ret;
1992}
1993
Mingming Cao61628a32008-07-11 19:27:31 -04001994/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001995 * Note that we don't need to start a transaction unless we're journaling data
1996 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1997 * need to file the inode to the transaction's list in ordered mode because if
1998 * we are writing back data added by write(), the inode is already there and if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001999 * we are writing back data modified via mmap(), no one guarantees in which
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002000 * transaction the data will hit the disk. In case we are journaling data, we
2001 * cannot start transaction directly because transaction start ranks above page
2002 * lock so we have to do some magic.
2003 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04002004 * This function can get called via...
2005 * - ext4_da_writepages after taking page lock (have journal handle)
2006 * - journal_submit_inode_data_buffers (no journal handle)
Artem Bityutskiyf6463b02012-07-25 18:12:04 +03002007 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
Theodore Ts'ob920c752009-05-14 00:54:29 -04002008 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002009 *
2010 * We don't do any block allocation in this function. If we have page with
2011 * multiple blocks we need to write those buffer_heads that are mapped. This
2012 * is important for mmaped based write. So if we do with blocksize 1K
2013 * truncate(f, 1024);
2014 * a = mmap(f, 0, 4096);
2015 * a[0] = 'a';
2016 * truncate(f, 4096);
2017 * we have in the page first buffer_head mapped via page_mkwrite call back
Paul Bolle90802ed2011-12-05 13:00:34 +01002018 * but other buffer_heads would be unmapped but dirty (dirty done via the
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002019 * do_wp_page). So writepage should write the first block. If we modify
2020 * the mmap area beyond 1024 we will again get a page_fault and the
2021 * page_mkwrite callback will do the block allocation and mark the
2022 * buffer_heads mapped.
2023 *
2024 * We redirty the page if we have any buffer_heads that is either delay or
2025 * unwritten in the page.
2026 *
2027 * We can get recursively called as show below.
2028 *
2029 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2030 * ext4_writepage()
2031 *
2032 * But since we don't do any block allocation we should not deadlock.
2033 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04002034 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002035static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04002036 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002037{
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002038 int ret = 0, commit_write = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002039 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002040 unsigned int len;
Jiaying Zhang744692d2010-03-04 16:14:02 -05002041 struct buffer_head *page_bufs = NULL;
Mingming Cao61628a32008-07-11 19:27:31 -04002042 struct inode *inode = page->mapping->host;
Alex Tomas64769242008-07-11 19:27:31 -04002043
Lukas Czernera9c667f2011-06-06 09:51:52 -04002044 trace_ext4_writepage(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002045 size = i_size_read(inode);
2046 if (page->index == size >> PAGE_CACHE_SHIFT)
2047 len = size & ~PAGE_CACHE_MASK;
2048 else
2049 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04002050
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002051 /*
2052 * If the page does not have buffers (for whatever reason),
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002053 * try to create them using __block_write_begin. If this
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002054 * fails, redirty the page and move on.
2055 */
Theodore Ts'ob1142e82010-10-28 17:33:57 -04002056 if (!page_has_buffers(page)) {
Theodore Ts'oa107e5a2010-10-27 23:44:47 -04002057 if (__block_write_begin(page, 0, len,
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002058 noalloc_get_block_write)) {
2059 redirty_page:
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002060 redirty_page_for_writepage(wbc, page);
2061 unlock_page(page);
2062 return 0;
2063 }
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002064 commit_write = 1;
2065 }
2066 page_bufs = page_buffers(page);
Tao Maf19d5872012-12-10 14:05:51 -05002067 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2068 ext4_bh_delay_or_unwritten)) {
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002069 /*
Theodore Ts'ob1142e82010-10-28 17:33:57 -04002070 * We don't want to do block allocation, so redirty
2071 * the page and return. We may reach here when we do
2072 * a journal commit via journal_submit_inode_data_buffers.
Mel Gorman966dbde2011-10-31 17:07:48 -07002073 * We can also reach here via shrink_page_list but it
2074 * should never be for direct reclaim so warn if that
2075 * happens
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04002076 */
Mel Gorman966dbde2011-10-31 17:07:48 -07002077 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
2078 PF_MEMALLOC);
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002079 goto redirty_page;
2080 }
2081 if (commit_write)
Aneesh Kumar K.Ved9b3e32008-11-07 09:06:45 -05002082 /* now mark the buffer_heads as dirty and uptodate */
Aneesh Kumar K.Vb767e782009-06-04 08:06:06 -04002083 block_commit_write(page, 0, len);
Alex Tomas64769242008-07-11 19:27:31 -04002084
Theodore Ts'ocb20d512010-10-27 21:30:09 -04002085 if (PageChecked(page) && ext4_should_journal_data(inode))
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002086 /*
2087 * It's mmapped pagecache. Add buffers and journal it. There
2088 * doesn't seem much point in redirtying the page here.
2089 */
Wu Fengguang3f0ca302009-11-24 11:15:44 -05002090 return __ext4_journalled_writepage(page, len);
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04002091
Theodore Ts'oa42afc52010-10-27 21:30:09 -04002092 if (buffer_uninit(page_bufs)) {
Jiaying Zhang744692d2010-03-04 16:14:02 -05002093 ext4_set_bh_endio(page_bufs, inode);
2094 ret = block_write_full_page_endio(page, noalloc_get_block_write,
2095 wbc, ext4_end_io_buffer_write);
2096 } else
Theodore Ts'ob920c752009-05-14 00:54:29 -04002097 ret = block_write_full_page(page, noalloc_get_block_write,
2098 wbc);
Alex Tomas64769242008-07-11 19:27:31 -04002099
Alex Tomas64769242008-07-11 19:27:31 -04002100 return ret;
2101}
2102
Mingming Cao61628a32008-07-11 19:27:31 -04002103/*
Mingming Cao525f4ed2008-08-19 22:15:58 -04002104 * This is called via ext4_da_writepages() to
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002105 * calculate the total number of credits to reserve to fit
Mingming Cao525f4ed2008-08-19 22:15:58 -04002106 * a single extent allocation into a single transaction,
2107 * ext4_da_writpeages() will loop calling this before
2108 * the block allocation.
Mingming Cao61628a32008-07-11 19:27:31 -04002109 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002110
2111static int ext4_da_writepages_trans_blocks(struct inode *inode)
2112{
2113 int max_blocks = EXT4_I(inode)->i_reserved_data_blocks;
2114
2115 /*
2116 * With non-extent format the journal credit needed to
2117 * insert nrblocks contiguous block is dependent on
2118 * number of contiguous block. So we will limit
2119 * number of contiguous block to a sane value
2120 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002121 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
Mingming Cao525f4ed2008-08-19 22:15:58 -04002122 (max_blocks > EXT4_MAX_TRANS_DATA))
2123 max_blocks = EXT4_MAX_TRANS_DATA;
2124
2125 return ext4_chunk_trans_blocks(inode, max_blocks);
2126}
Mingming Cao61628a32008-07-11 19:27:31 -04002127
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002128/*
2129 * write_cache_pages_da - walk the list of dirty pages of the given
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002130 * address space and accumulate pages that need writing, and call
Theodore Ts'o168fc022011-02-26 14:09:20 -05002131 * mpage_da_map_and_submit to map a single contiguous memory region
2132 * and then write them.
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002133 */
Tao Ma9c3569b2012-12-10 14:05:57 -05002134static int write_cache_pages_da(handle_t *handle,
2135 struct address_space *mapping,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002136 struct writeback_control *wbc,
Eric Sandeen72f84e62010-10-27 21:30:13 -04002137 struct mpage_da_data *mpd,
2138 pgoff_t *done_index)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002139{
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002140 struct buffer_head *bh, *head;
Theodore Ts'o168fc022011-02-26 14:09:20 -05002141 struct inode *inode = mapping->host;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002142 struct pagevec pvec;
2143 unsigned int nr_pages;
2144 sector_t logical;
2145 pgoff_t index, end;
2146 long nr_to_write = wbc->nr_to_write;
2147 int i, tag, ret = 0;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002148
Theodore Ts'o168fc022011-02-26 14:09:20 -05002149 memset(mpd, 0, sizeof(struct mpage_da_data));
2150 mpd->wbc = wbc;
2151 mpd->inode = inode;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002152 pagevec_init(&pvec, 0);
2153 index = wbc->range_start >> PAGE_CACHE_SHIFT;
2154 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2155
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002156 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002157 tag = PAGECACHE_TAG_TOWRITE;
2158 else
2159 tag = PAGECACHE_TAG_DIRTY;
2160
Eric Sandeen72f84e62010-10-27 21:30:13 -04002161 *done_index = index;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002162 while (index <= end) {
Eric Sandeen5b41d922010-10-27 21:30:13 -04002163 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002164 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2165 if (nr_pages == 0)
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002166 return 0;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002167
2168 for (i = 0; i < nr_pages; i++) {
2169 struct page *page = pvec.pages[i];
2170
2171 /*
2172 * At this point, the page may be truncated or
2173 * invalidated (changing page->mapping to NULL), or
2174 * even swizzled back from swapper_space to tmpfs file
2175 * mapping. However, page->index will not change
2176 * because we have a reference on the page.
2177 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002178 if (page->index > end)
2179 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002180
Eric Sandeen72f84e62010-10-27 21:30:13 -04002181 *done_index = page->index + 1;
2182
Theodore Ts'o78aaced2011-02-26 14:09:14 -05002183 /*
2184 * If we can't merge this page, and we have
2185 * accumulated an contiguous region, write it
2186 */
2187 if ((mpd->next_page != page->index) &&
2188 (mpd->next_page != mpd->first_page)) {
2189 mpage_da_map_and_submit(mpd);
2190 goto ret_extent_tail;
2191 }
2192
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002193 lock_page(page);
2194
2195 /*
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002196 * If the page is no longer dirty, or its
2197 * mapping no longer corresponds to inode we
2198 * are writing (which means it has been
2199 * truncated or invalidated), or the page is
2200 * already under writeback and we are not
2201 * doing a data integrity writeback, skip the page
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002202 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002203 if (!PageDirty(page) ||
2204 (PageWriteback(page) &&
2205 (wbc->sync_mode == WB_SYNC_NONE)) ||
2206 unlikely(page->mapping != mapping)) {
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002207 unlock_page(page);
2208 continue;
2209 }
2210
Darrick J. Wong7cb1a532011-05-18 13:53:20 -04002211 wait_on_page_writeback(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002212 BUG_ON(PageWriteback(page));
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002213
Tao Ma9c3569b2012-12-10 14:05:57 -05002214 /*
2215 * If we have inline data and arrive here, it means that
2216 * we will soon create the block for the 1st page, so
2217 * we'd better clear the inline data here.
2218 */
2219 if (ext4_has_inline_data(inode)) {
2220 BUG_ON(ext4_test_inode_state(inode,
2221 EXT4_STATE_MAY_INLINE_DATA));
2222 ext4_destroy_inline_data(handle, inode);
2223 }
2224
Theodore Ts'o168fc022011-02-26 14:09:20 -05002225 if (mpd->next_page != page->index)
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002226 mpd->first_page = page->index;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002227 mpd->next_page = page->index + 1;
2228 logical = (sector_t) page->index <<
2229 (PAGE_CACHE_SHIFT - inode->i_blkbits);
2230
2231 if (!page_has_buffers(page)) {
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002232 mpage_add_bh_to_extent(mpd, logical,
2233 PAGE_CACHE_SIZE,
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002234 (1 << BH_Dirty) | (1 << BH_Uptodate));
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002235 if (mpd->io_done)
2236 goto ret_extent_tail;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002237 } else {
2238 /*
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002239 * Page with regular buffer heads,
2240 * just add all dirty ones
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002241 */
2242 head = page_buffers(page);
2243 bh = head;
2244 do {
2245 BUG_ON(buffer_locked(bh));
2246 /*
2247 * We need to try to allocate
2248 * unmapped blocks in the same page.
2249 * Otherwise we won't make progress
2250 * with the page in ext4_writepage
2251 */
2252 if (ext4_bh_delay_or_unwritten(NULL, bh)) {
2253 mpage_add_bh_to_extent(mpd, logical,
2254 bh->b_size,
2255 bh->b_state);
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002256 if (mpd->io_done)
2257 goto ret_extent_tail;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002258 } else if (buffer_dirty(bh) && (buffer_mapped(bh))) {
2259 /*
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002260 * mapped dirty buffer. We need
2261 * to update the b_state
2262 * because we look at b_state
2263 * in mpage_da_map_blocks. We
2264 * don't update b_size because
2265 * if we find an unmapped
2266 * buffer_head later we need to
2267 * use the b_state flag of that
2268 * buffer_head.
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002269 */
2270 if (mpd->b_size == 0)
2271 mpd->b_state = bh->b_state & BH_FLAGS;
2272 }
2273 logical++;
2274 } while ((bh = bh->b_this_page) != head);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002275 }
2276
2277 if (nr_to_write > 0) {
2278 nr_to_write--;
2279 if (nr_to_write == 0 &&
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002280 wbc->sync_mode == WB_SYNC_NONE)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002281 /*
2282 * We stop writing back only if we are
2283 * not doing integrity sync. In case of
2284 * integrity sync we have to keep going
2285 * because someone may be concurrently
2286 * dirtying pages, and we might have
2287 * synced a lot of newly appeared dirty
2288 * pages, but have not synced all of the
2289 * old dirty pages.
2290 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002291 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002292 }
2293 }
2294 pagevec_release(&pvec);
2295 cond_resched();
2296 }
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002297 return 0;
2298ret_extent_tail:
2299 ret = MPAGE_DA_EXTENT_TAIL;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002300out:
2301 pagevec_release(&pvec);
2302 cond_resched();
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002303 return ret;
2304}
2305
2306
Alex Tomas64769242008-07-11 19:27:31 -04002307static int ext4_da_writepages(struct address_space *mapping,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002308 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002309{
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002310 pgoff_t index;
2311 int range_whole = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04002312 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002313 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002314 struct inode *inode = mapping->host;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002315 int pages_written = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002316 unsigned int max_pages;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002317 int range_cyclic, cycled = 1, io_done = 0;
Theodore Ts'o55138e02009-09-29 13:31:31 -04002318 int needed_blocks, ret = 0;
2319 long desired_nr_to_write, nr_to_writebump = 0;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002320 loff_t range_start = wbc->range_start;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002321 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Eric Sandeen72f84e62010-10-27 21:30:13 -04002322 pgoff_t done_index = 0;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002323 pgoff_t end;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002324 struct blk_plug plug;
Mingming Cao61628a32008-07-11 19:27:31 -04002325
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002326 trace_ext4_da_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002327
Mingming Cao61628a32008-07-11 19:27:31 -04002328 /*
2329 * No pages to write? This is mainly a kludge to avoid starting
2330 * a transaction for special inodes like journal inode on last iput()
2331 * because that could violate lock ordering on umount
2332 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002333 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Mingming Cao61628a32008-07-11 19:27:31 -04002334 return 0;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002335
2336 /*
2337 * If the filesystem has aborted, it is read-only, so return
2338 * right away instead of dumping stack traces later on that
2339 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002340 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002341 * the latter could be true if the filesystem is mounted
2342 * read-only, and in that case, ext4_da_writepages should
2343 * *never* be called, so if that ever happens, we would want
2344 * the stack trace.
2345 */
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002346 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED))
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002347 return -EROFS;
2348
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002349 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2350 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002351
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002352 range_cyclic = wbc->range_cyclic;
2353 if (wbc->range_cyclic) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002354 index = mapping->writeback_index;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002355 if (index)
2356 cycled = 0;
2357 wbc->range_start = index << PAGE_CACHE_SHIFT;
2358 wbc->range_end = LLONG_MAX;
2359 wbc->range_cyclic = 0;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002360 end = -1;
2361 } else {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002362 index = wbc->range_start >> PAGE_CACHE_SHIFT;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002363 end = wbc->range_end >> PAGE_CACHE_SHIFT;
2364 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002365
Theodore Ts'o55138e02009-09-29 13:31:31 -04002366 /*
2367 * This works around two forms of stupidity. The first is in
2368 * the writeback code, which caps the maximum number of pages
2369 * written to be 1024 pages. This is wrong on multiple
2370 * levels; different architectues have a different page size,
2371 * which changes the maximum amount of data which gets
2372 * written. Secondly, 4 megabytes is way too small. XFS
2373 * forces this value to be 16 megabytes by multiplying
2374 * nr_to_write parameter by four, and then relies on its
2375 * allocator to allocate larger extents to make them
2376 * contiguous. Unfortunately this brings us to the second
2377 * stupidity, which is that ext4's mballoc code only allocates
2378 * at most 2048 blocks. So we force contiguous writes up to
2379 * the number of dirty blocks in the inode, or
2380 * sbi->max_writeback_mb_bump whichever is smaller.
2381 */
2382 max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT);
Eric Sandeenb443e732010-10-27 21:30:03 -04002383 if (!range_cyclic && range_whole) {
2384 if (wbc->nr_to_write == LONG_MAX)
2385 desired_nr_to_write = wbc->nr_to_write;
2386 else
2387 desired_nr_to_write = wbc->nr_to_write * 8;
2388 } else
Theodore Ts'o55138e02009-09-29 13:31:31 -04002389 desired_nr_to_write = ext4_num_dirty_pages(inode, index,
2390 max_pages);
2391 if (desired_nr_to_write > max_pages)
2392 desired_nr_to_write = max_pages;
2393
2394 if (wbc->nr_to_write < desired_nr_to_write) {
2395 nr_to_writebump = desired_nr_to_write - wbc->nr_to_write;
2396 wbc->nr_to_write = desired_nr_to_write;
2397 }
2398
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002399retry:
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002400 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002401 tag_pages_for_writeback(mapping, index, end);
2402
Shaohua Li1bce63d2011-10-18 10:55:51 -04002403 blk_start_plug(&plug);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002404 while (!ret && wbc->nr_to_write > 0) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002405
2406 /*
2407 * we insert one extent at a time. So we need
2408 * credit needed for single extent allocation.
2409 * journalled mode is currently not supported
2410 * by delalloc
2411 */
2412 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002413 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002414
Mingming Cao61628a32008-07-11 19:27:31 -04002415 /* start a new transaction*/
2416 handle = ext4_journal_start(inode, needed_blocks);
2417 if (IS_ERR(handle)) {
2418 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002419 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04002420 "%ld pages, ino %lu; err %d", __func__,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002421 wbc->nr_to_write, inode->i_ino, ret);
Namjae Jeon3c1fcb22011-11-07 11:01:13 -05002422 blk_finish_plug(&plug);
Mingming Cao61628a32008-07-11 19:27:31 -04002423 goto out_writepages;
2424 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002425
2426 /*
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002427 * Now call write_cache_pages_da() to find the next
Theodore Ts'of63e6002009-02-23 16:42:39 -05002428 * contiguous region of logical blocks that need
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002429 * blocks to be allocated by ext4 and submit them.
Theodore Ts'of63e6002009-02-23 16:42:39 -05002430 */
Tao Ma9c3569b2012-12-10 14:05:57 -05002431 ret = write_cache_pages_da(handle, mapping,
2432 wbc, &mpd, &done_index);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002433 /*
André Goddard Rosaaf901ca2009-11-14 13:09:05 -02002434 * If we have a contiguous extent of pages and we
Theodore Ts'of63e6002009-02-23 16:42:39 -05002435 * haven't done the I/O yet, map the blocks and submit
2436 * them for I/O.
2437 */
2438 if (!mpd.io_done && mpd.next_page != mpd.first_page) {
Theodore Ts'o5a87b7a2010-10-27 21:30:09 -04002439 mpage_da_map_and_submit(&mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002440 ret = MPAGE_DA_EXTENT_TAIL;
2441 }
Theodore Ts'ob3a3ca82009-08-31 23:13:11 -04002442 trace_ext4_da_write_pages(inode, &mpd);
Theodore Ts'of63e6002009-02-23 16:42:39 -05002443 wbc->nr_to_write -= mpd.pages_written;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002444
Mingming Cao61628a32008-07-11 19:27:31 -04002445 ext4_journal_stop(handle);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002446
Eric Sandeen8f64b322009-02-26 00:57:35 -05002447 if ((mpd.retval == -ENOSPC) && sbi->s_journal) {
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002448 /* commit the transaction which would
2449 * free blocks released in the transaction
2450 * and try again
2451 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002452 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002453 ret = 0;
2454 } else if (ret == MPAGE_DA_EXTENT_TAIL) {
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002455 /*
Kazuya Mio8de49e62011-10-20 19:23:08 -04002456 * Got one extent now try with rest of the pages.
2457 * If mpd.retval is set -EIO, journal is aborted.
2458 * So we don't need to write any more.
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002459 */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002460 pages_written += mpd.pages_written;
Kazuya Mio8de49e62011-10-20 19:23:08 -04002461 ret = mpd.retval;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002462 io_done = 1;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002463 } else if (wbc->nr_to_write)
Mingming Cao61628a32008-07-11 19:27:31 -04002464 /*
2465 * There is no more writeout needed
2466 * or we requested for a noblocking writeout
2467 * and we found the device congested
2468 */
Mingming Cao61628a32008-07-11 19:27:31 -04002469 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002470 }
Shaohua Li1bce63d2011-10-18 10:55:51 -04002471 blk_finish_plug(&plug);
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002472 if (!io_done && !cycled) {
2473 cycled = 1;
2474 index = 0;
2475 wbc->range_start = index << PAGE_CACHE_SHIFT;
2476 wbc->range_end = mapping->writeback_index - 1;
2477 goto retry;
2478 }
Mingming Cao61628a32008-07-11 19:27:31 -04002479
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002480 /* Update index */
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002481 wbc->range_cyclic = range_cyclic;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002482 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2483 /*
2484 * set the writeback_index so that range_cyclic
2485 * mode will write it back later
2486 */
Eric Sandeen72f84e62010-10-27 21:30:13 -04002487 mapping->writeback_index = done_index;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002488
Mingming Cao61628a32008-07-11 19:27:31 -04002489out_writepages:
Richard Kennedy2faf2e12009-12-25 15:46:07 -05002490 wbc->nr_to_write -= nr_to_writebump;
Theodore Ts'ode89de62009-08-31 17:00:59 -04002491 wbc->range_start = range_start;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002492 trace_ext4_da_writepages_result(inode, wbc, ret, pages_written);
Mingming Cao61628a32008-07-11 19:27:31 -04002493 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002494}
2495
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002496static int ext4_nonda_switch(struct super_block *sb)
2497{
2498 s64 free_blocks, dirty_blocks;
2499 struct ext4_sb_info *sbi = EXT4_SB(sb);
2500
2501 /*
2502 * switch to non delalloc mode if we are running low
2503 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08002504 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002505 * accumulated on each CPU without updating global counters
2506 * Delalloc need an accurate free block accounting. So switch
2507 * to non delalloc when we are near to error range.
2508 */
Theodore Ts'o57042652011-09-09 18:56:51 -04002509 free_blocks = EXT4_C2B(sbi,
2510 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
2511 dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002512 /*
2513 * Start pushing delalloc when 1/2 of free blocks are dirty.
2514 */
2515 if (dirty_blocks && (free_blocks < 2 * dirty_blocks) &&
2516 !writeback_in_progress(sb->s_bdi) &&
2517 down_read_trylock(&sb->s_umount)) {
2518 writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2519 up_read(&sb->s_umount);
2520 }
2521
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002522 if (2 * free_blocks < 3 * dirty_blocks ||
Theodore Ts'odf55c992011-09-09 19:16:51 -04002523 free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) {
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002524 /*
Eric Sandeenc8afb442009-12-23 07:58:12 -05002525 * free block count is less than 150% of dirty blocks
2526 * or free blocks is less than watermark
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002527 */
2528 return 1;
2529 }
2530 return 0;
2531}
2532
Alex Tomas64769242008-07-11 19:27:31 -04002533static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002534 loff_t pos, unsigned len, unsigned flags,
2535 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002536{
Eric Sandeen72b8ab92010-05-16 11:00:00 -04002537 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002538 struct page *page;
2539 pgoff_t index;
Alex Tomas64769242008-07-11 19:27:31 -04002540 struct inode *inode = mapping->host;
2541 handle_t *handle;
2542
2543 index = pos >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002544
2545 if (ext4_nonda_switch(inode->i_sb)) {
2546 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2547 return ext4_write_begin(file, mapping, pos,
2548 len, flags, pagep, fsdata);
2549 }
2550 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002551 trace_ext4_da_write_begin(inode, pos, len, flags);
Tao Ma9c3569b2012-12-10 14:05:57 -05002552
2553 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2554 ret = ext4_da_write_inline_data_begin(mapping, inode,
2555 pos, len, flags,
2556 pagep, fsdata);
2557 if (ret < 0)
2558 goto out;
2559 if (ret == 1) {
2560 ret = 0;
2561 goto out;
2562 }
2563 }
2564
Mingming Caod2a17632008-07-14 17:52:37 -04002565retry:
Alex Tomas64769242008-07-11 19:27:31 -04002566 /*
2567 * With delayed allocation, we don't log the i_disksize update
2568 * if there is delayed block allocation. But we still need
2569 * to journalling the i_disksize update if writes to the end
2570 * of file which has an already mapped buffer.
2571 */
2572 handle = ext4_journal_start(inode, 1);
2573 if (IS_ERR(handle)) {
2574 ret = PTR_ERR(handle);
2575 goto out;
2576 }
Jan Karaebd36102009-02-22 21:09:59 -05002577 /* We cannot recurse into the filesystem as the transaction is already
2578 * started */
2579 flags |= AOP_FLAG_NOFS;
Alex Tomas64769242008-07-11 19:27:31 -04002580
Nick Piggin54566b22009-01-04 12:00:53 -08002581 page = grab_cache_page_write_begin(mapping, index, flags);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002582 if (!page) {
2583 ext4_journal_stop(handle);
2584 ret = -ENOMEM;
2585 goto out;
2586 }
Alex Tomas64769242008-07-11 19:27:31 -04002587 *pagep = page;
2588
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002589 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04002590 if (ret < 0) {
2591 unlock_page(page);
2592 ext4_journal_stop(handle);
2593 page_cache_release(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04002594 /*
2595 * block_write_begin may have instantiated a few blocks
2596 * outside i_size. Trim these off again. Don't need
2597 * i_size_read because we hold i_mutex.
2598 */
2599 if (pos + len > inode->i_size)
Jan Karab9a42072009-12-08 21:24:33 -05002600 ext4_truncate_failed_write(inode);
Alex Tomas64769242008-07-11 19:27:31 -04002601 }
2602
Mingming Caod2a17632008-07-14 17:52:37 -04002603 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
2604 goto retry;
Alex Tomas64769242008-07-11 19:27:31 -04002605out:
2606 return ret;
2607}
2608
Mingming Cao632eaea2008-07-11 19:27:31 -04002609/*
2610 * Check if we should update i_disksize
2611 * when write to the end of file but not require block allocation
2612 */
2613static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002614 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04002615{
2616 struct buffer_head *bh;
2617 struct inode *inode = page->mapping->host;
2618 unsigned int idx;
2619 int i;
2620
2621 bh = page_buffers(page);
2622 idx = offset >> inode->i_blkbits;
2623
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002624 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04002625 bh = bh->b_this_page;
2626
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002627 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04002628 return 0;
2629 return 1;
2630}
2631
Alex Tomas64769242008-07-11 19:27:31 -04002632static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002633 struct address_space *mapping,
2634 loff_t pos, unsigned len, unsigned copied,
2635 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002636{
2637 struct inode *inode = mapping->host;
2638 int ret = 0, ret2;
2639 handle_t *handle = ext4_journal_current_handle();
2640 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002641 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002642 int write_mode = (int)(unsigned long)fsdata;
2643
2644 if (write_mode == FALL_BACK_TO_NONDELALLOC) {
Lukas Czerner3d2b1582012-02-20 17:53:00 -05002645 switch (ext4_inode_journal_mode(inode)) {
2646 case EXT4_INODE_ORDERED_DATA_MODE:
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002647 return ext4_ordered_write_end(file, mapping, pos,
2648 len, copied, page, fsdata);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05002649 case EXT4_INODE_WRITEBACK_DATA_MODE:
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002650 return ext4_writeback_write_end(file, mapping, pos,
2651 len, copied, page, fsdata);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05002652 default:
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002653 BUG();
2654 }
2655 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002656
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002657 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04002658 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002659 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04002660
2661 /*
2662 * generic_write_end() will run mark_inode_dirty() if i_size
2663 * changes. So let's piggyback the i_disksize mark_inode_dirty
2664 * into that.
2665 */
Alex Tomas64769242008-07-11 19:27:31 -04002666 new_i_size = pos + copied;
Andrea Arcangeliea51d132011-12-13 21:41:15 -05002667 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
Tao Ma9c3569b2012-12-10 14:05:57 -05002668 if (ext4_has_inline_data(inode) ||
2669 ext4_da_should_update_i_disksize(page, end)) {
Mingming Cao632eaea2008-07-11 19:27:31 -04002670 down_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'of3b59292012-11-15 23:08:57 -05002671 if (new_i_size > EXT4_I(inode)->i_disksize)
Mingming Cao632eaea2008-07-11 19:27:31 -04002672 EXT4_I(inode)->i_disksize = new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002673 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002674 /* We need to mark inode dirty even if
2675 * new_i_size is less that inode->i_size
2676 * bu greater than i_disksize.(hint delalloc)
2677 */
2678 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04002679 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002680 }
Tao Ma9c3569b2012-12-10 14:05:57 -05002681
2682 if (write_mode != CONVERT_INLINE_DATA &&
2683 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2684 ext4_has_inline_data(inode))
2685 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2686 page);
2687 else
2688 ret2 = generic_write_end(file, mapping, pos, len, copied,
Alex Tomas64769242008-07-11 19:27:31 -04002689 page, fsdata);
Tao Ma9c3569b2012-12-10 14:05:57 -05002690
Alex Tomas64769242008-07-11 19:27:31 -04002691 copied = ret2;
2692 if (ret2 < 0)
2693 ret = ret2;
2694 ret2 = ext4_journal_stop(handle);
2695 if (!ret)
2696 ret = ret2;
2697
2698 return ret ? ret : copied;
2699}
2700
2701static void ext4_da_invalidatepage(struct page *page, unsigned long offset)
2702{
Alex Tomas64769242008-07-11 19:27:31 -04002703 /*
2704 * Drop reserved blocks
2705 */
2706 BUG_ON(!PageLocked(page));
2707 if (!page_has_buffers(page))
2708 goto out;
2709
Mingming Caod2a17632008-07-14 17:52:37 -04002710 ext4_da_page_release_reservation(page, offset);
Alex Tomas64769242008-07-11 19:27:31 -04002711
2712out:
2713 ext4_invalidatepage(page, offset);
2714
2715 return;
2716}
2717
Theodore Ts'occd25062009-02-26 01:04:07 -05002718/*
2719 * Force all delayed allocation blocks to be allocated for a given inode.
2720 */
2721int ext4_alloc_da_blocks(struct inode *inode)
2722{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04002723 trace_ext4_alloc_da_blocks(inode);
2724
Theodore Ts'occd25062009-02-26 01:04:07 -05002725 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2726 !EXT4_I(inode)->i_reserved_meta_blocks)
2727 return 0;
2728
2729 /*
2730 * We do something simple for now. The filemap_flush() will
2731 * also start triggering a write of the data blocks, which is
2732 * not strictly speaking necessary (and for users of
2733 * laptop_mode, not even desirable). However, to do otherwise
2734 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002735 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002736 * ext4_da_writepages() ->
2737 * write_cache_pages() ---> (via passed in callback function)
2738 * __mpage_da_writepage() -->
2739 * mpage_add_bh_to_extent()
2740 * mpage_da_map_blocks()
2741 *
2742 * The problem is that write_cache_pages(), located in
2743 * mm/page-writeback.c, marks pages clean in preparation for
2744 * doing I/O, which is not desirable if we're not planning on
2745 * doing I/O at all.
2746 *
2747 * We could call write_cache_pages(), and then redirty all of
Wu Fengguang380cf092010-11-11 19:23:29 +08002748 * the pages by calling redirty_page_for_writepage() but that
Theodore Ts'occd25062009-02-26 01:04:07 -05002749 * would be ugly in the extreme. So instead we would need to
2750 * replicate parts of the code in the above functions,
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002751 * simplifying them because we wouldn't actually intend to
Theodore Ts'occd25062009-02-26 01:04:07 -05002752 * write out the pages, but rather only collect contiguous
2753 * logical block extents, call the multi-block allocator, and
2754 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002755 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002756 * For now, though, we'll cheat by calling filemap_flush(),
2757 * which will map the blocks, and start the I/O, but not
2758 * actually wait for the I/O to complete.
2759 */
2760 return filemap_flush(inode->i_mapping);
2761}
Alex Tomas64769242008-07-11 19:27:31 -04002762
2763/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002764 * bmap() is special. It gets used by applications such as lilo and by
2765 * the swapper to find the on-disk block of a specific piece of data.
2766 *
2767 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07002768 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002769 * filesystem and enables swap, then they may get a nasty shock when the
2770 * data getting swapped to that swapfile suddenly gets overwritten by
2771 * the original zero's written out previously to the journal and
2772 * awaiting writeback in the kernel's buffer cache.
2773 *
2774 * So, if we see any bmap calls here on a modified, data-journaled file,
2775 * take extra steps to flush any blocks which might be in the cache.
2776 */
Mingming Cao617ba132006-10-11 01:20:53 -07002777static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002778{
2779 struct inode *inode = mapping->host;
2780 journal_t *journal;
2781 int err;
2782
Tao Ma46c7f252012-12-10 14:04:52 -05002783 /*
2784 * We can get here for an inline file via the FIBMAP ioctl
2785 */
2786 if (ext4_has_inline_data(inode))
2787 return 0;
2788
Alex Tomas64769242008-07-11 19:27:31 -04002789 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2790 test_opt(inode->i_sb, DELALLOC)) {
2791 /*
2792 * With delalloc we want to sync the file
2793 * so that we can make sure we allocate
2794 * blocks for file
2795 */
2796 filemap_write_and_wait(mapping);
2797 }
2798
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002799 if (EXT4_JOURNAL(inode) &&
2800 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002801 /*
2802 * This is a REALLY heavyweight approach, but the use of
2803 * bmap on dirty files is expected to be extremely rare:
2804 * only if we run lilo or swapon on a freshly made file
2805 * do we expect this to happen.
2806 *
2807 * (bmap requires CAP_SYS_RAWIO so this does not
2808 * represent an unprivileged user DOS attack --- we'd be
2809 * in trouble if mortal users could trigger this path at
2810 * will.)
2811 *
Mingming Cao617ba132006-10-11 01:20:53 -07002812 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002813 * regular files. If somebody wants to bmap a directory
2814 * or symlink and gets confused because the buffer
2815 * hasn't yet been flushed to disk, they deserve
2816 * everything they get.
2817 */
2818
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002819 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
Mingming Cao617ba132006-10-11 01:20:53 -07002820 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07002821 jbd2_journal_lock_updates(journal);
2822 err = jbd2_journal_flush(journal);
2823 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002824
2825 if (err)
2826 return 0;
2827 }
2828
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002829 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002830}
2831
Mingming Cao617ba132006-10-11 01:20:53 -07002832static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002833{
Tao Ma46c7f252012-12-10 14:04:52 -05002834 int ret = -EAGAIN;
2835 struct inode *inode = page->mapping->host;
2836
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002837 trace_ext4_readpage(page);
Tao Ma46c7f252012-12-10 14:04:52 -05002838
2839 if (ext4_has_inline_data(inode))
2840 ret = ext4_readpage_inline(inode, page);
2841
2842 if (ret == -EAGAIN)
2843 return mpage_readpage(page, ext4_get_block);
2844
2845 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002846}
2847
2848static int
Mingming Cao617ba132006-10-11 01:20:53 -07002849ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002850 struct list_head *pages, unsigned nr_pages)
2851{
Tao Ma46c7f252012-12-10 14:04:52 -05002852 struct inode *inode = mapping->host;
2853
2854 /* If the file has inline data, no need to do readpages. */
2855 if (ext4_has_inline_data(inode))
2856 return 0;
2857
Mingming Cao617ba132006-10-11 01:20:53 -07002858 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002859}
2860
Jiaying Zhang744692d2010-03-04 16:14:02 -05002861static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset)
2862{
2863 struct buffer_head *head, *bh;
2864 unsigned int curr_off = 0;
2865
2866 if (!page_has_buffers(page))
2867 return;
2868 head = bh = page_buffers(page);
2869 do {
2870 if (offset <= curr_off && test_clear_buffer_uninit(bh)
2871 && bh->b_private) {
2872 ext4_free_io_end(bh->b_private);
2873 bh->b_private = NULL;
2874 bh->b_end_io = NULL;
2875 }
2876 curr_off = curr_off + bh->b_size;
2877 bh = bh->b_this_page;
2878 } while (bh != head);
2879}
2880
Mingming Cao617ba132006-10-11 01:20:53 -07002881static void ext4_invalidatepage(struct page *page, unsigned long offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002882{
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002883 trace_ext4_invalidatepage(page, offset);
2884
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002885 /*
Jiaying Zhang744692d2010-03-04 16:14:02 -05002886 * free any io_end structure allocated for buffers to be discarded
2887 */
2888 if (ext4_should_dioread_nolock(page->mapping->host))
2889 ext4_invalidatepage_free_endio(page, offset);
Jan Kara4520fb32012-12-25 13:28:54 -05002890
2891 /* No journalling happens on data buffers when this function is used */
2892 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2893
2894 block_invalidatepage(page, offset);
2895}
2896
Jan Kara53e87262012-12-25 13:29:52 -05002897static int __ext4_journalled_invalidatepage(struct page *page,
2898 unsigned long offset)
Jan Kara4520fb32012-12-25 13:28:54 -05002899{
2900 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2901
2902 trace_ext4_journalled_invalidatepage(page, offset);
2903
Jiaying Zhang744692d2010-03-04 16:14:02 -05002904 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002905 * If it's a full truncate we just forget about the pending dirtying
2906 */
2907 if (offset == 0)
2908 ClearPageChecked(page);
2909
Jan Kara53e87262012-12-25 13:29:52 -05002910 return jbd2_journal_invalidatepage(journal, page, offset);
2911}
2912
2913/* Wrapper for aops... */
2914static void ext4_journalled_invalidatepage(struct page *page,
2915 unsigned long offset)
2916{
2917 WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002918}
2919
Mingming Cao617ba132006-10-11 01:20:53 -07002920static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002921{
Mingming Cao617ba132006-10-11 01:20:53 -07002922 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002923
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002924 trace_ext4_releasepage(page);
2925
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002926 WARN_ON(PageChecked(page));
2927 if (!page_has_buffers(page))
2928 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05002929 if (journal)
2930 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2931 else
2932 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002933}
2934
2935/*
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002936 * ext4_get_block used when preparing for a DIO write or buffer write.
2937 * We allocate an uinitialized extent if blocks haven't been allocated.
2938 * The extent will be converted to initialized after the IO is complete.
2939 */
Tao Maf19d5872012-12-10 14:05:51 -05002940int ext4_get_block_write(struct inode *inode, sector_t iblock,
Mingming Cao4c0425f2009-09-28 15:48:41 -04002941 struct buffer_head *bh_result, int create)
2942{
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002943 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002944 inode->i_ino, create);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002945 return _ext4_get_block(inode, iblock, bh_result,
2946 EXT4_GET_BLOCKS_IO_CREATE_EXT);
Mingming Cao4c0425f2009-09-28 15:48:41 -04002947}
2948
Zheng Liu729f52c2012-07-09 16:29:29 -04002949static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05002950 struct buffer_head *bh_result, int create)
Zheng Liu729f52c2012-07-09 16:29:29 -04002951{
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05002952 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2953 inode->i_ino, create);
2954 return _ext4_get_block(inode, iblock, bh_result,
2955 EXT4_GET_BLOCKS_NO_LOCK);
Zheng Liu729f52c2012-07-09 16:29:29 -04002956}
2957
Mingming Cao4c0425f2009-09-28 15:48:41 -04002958static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
Christoph Hellwig552ef8022010-07-27 11:56:06 -04002959 ssize_t size, void *private, int ret,
2960 bool is_async)
Mingming Cao4c0425f2009-09-28 15:48:41 -04002961{
Christoph Hellwig72c50522011-06-24 14:29:48 -04002962 struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
Mingming Cao4c0425f2009-09-28 15:48:41 -04002963 ext4_io_end_t *io_end = iocb->private;
Mingming Cao4c0425f2009-09-28 15:48:41 -04002964
Mingming4b70df12009-11-03 14:44:54 -05002965 /* if not async direct IO or dio with 0 bytes write, just return */
2966 if (!io_end || !size)
Christoph Hellwig552ef8022010-07-27 11:56:06 -04002967 goto out;
Mingming4b70df12009-11-03 14:44:54 -05002968
Zheng Liu88635ca2011-12-28 19:00:25 -05002969 ext_debug("ext4_end_io_dio(): io_end 0x%p "
Joe Perchesace36ad2012-03-19 23:11:43 -04002970 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002971 iocb->private, io_end->inode->i_ino, iocb, offset,
2972 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002973
Theodore Ts'ob5a7e972011-12-12 10:53:02 -05002974 iocb->private = NULL;
2975
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002976 /* if not aio dio with unwritten extents, just free io and return */
Theodore Ts'obd2d0212010-10-27 21:30:10 -04002977 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002978 ext4_free_io_end(io_end);
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04002979out:
2980 if (is_async)
2981 aio_complete(iocb, ret, 0);
Christoph Hellwig72c50522011-06-24 14:29:48 -04002982 inode_dio_done(inode);
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04002983 return;
Mingming Cao8d5d02e2009-09-28 15:48:29 -04002984 }
2985
Mingming Cao4c0425f2009-09-28 15:48:41 -04002986 io_end->offset = offset;
2987 io_end->size = size;
jiayingz@google.com (Jiaying Zhang)5b3ff232010-07-27 11:56:06 -04002988 if (is_async) {
2989 io_end->iocb = iocb;
2990 io_end->result = ret;
2991 }
Mingming Cao4c0425f2009-09-28 15:48:41 -04002992
Dmitry Monakhov28a535f2012-09-29 00:14:55 -04002993 ext4_add_complete_io(io_end);
Mingming Cao4c0425f2009-09-28 15:48:41 -04002994}
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002995
Jiaying Zhang744692d2010-03-04 16:14:02 -05002996static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
2997{
2998 ext4_io_end_t *io_end = bh->b_private;
Jiaying Zhang744692d2010-03-04 16:14:02 -05002999 struct inode *inode;
Jiaying Zhang744692d2010-03-04 16:14:02 -05003000
3001 if (!test_clear_buffer_uninit(bh) || !io_end)
3002 goto out;
3003
3004 if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) {
Theodore Ts'o92b97812012-03-19 23:41:49 -04003005 ext4_msg(io_end->inode->i_sb, KERN_INFO,
3006 "sb umounted, discard end_io request for inode %lu",
3007 io_end->inode->i_ino);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003008 ext4_free_io_end(io_end);
3009 goto out;
3010 }
3011
Tao Ma32c80b32011-08-13 12:30:59 -04003012 /*
3013 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
3014 * but being more careful is always safe for the future change.
3015 */
Jiaying Zhang744692d2010-03-04 16:14:02 -05003016 inode = io_end->inode;
Tao Ma0edeb712011-10-31 17:30:44 -04003017 ext4_set_io_unwritten_flag(inode, io_end);
Dmitry Monakhov28a535f2012-09-29 00:14:55 -04003018 ext4_add_complete_io(io_end);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003019out:
3020 bh->b_private = NULL;
3021 bh->b_end_io = NULL;
3022 clear_buffer_uninit(bh);
3023 end_buffer_async_write(bh, uptodate);
3024}
3025
3026static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode)
3027{
3028 ext4_io_end_t *io_end;
3029 struct page *page = bh->b_page;
3030 loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT;
3031 size_t size = bh->b_size;
3032
3033retry:
3034 io_end = ext4_init_io_end(inode, GFP_ATOMIC);
3035 if (!io_end) {
Andrew Morton6db26ff2011-01-12 16:59:13 -08003036 pr_warn_ratelimited("%s: allocation fail\n", __func__);
Jiaying Zhang744692d2010-03-04 16:14:02 -05003037 schedule();
3038 goto retry;
3039 }
3040 io_end->offset = offset;
3041 io_end->size = size;
3042 /*
3043 * We need to hold a reference to the page to make sure it
3044 * doesn't get evicted before ext4_end_io_work() has a chance
3045 * to convert the extent from written to unwritten.
3046 */
3047 io_end->page = page;
3048 get_page(io_end->page);
3049
3050 bh->b_private = io_end;
3051 bh->b_end_io = ext4_end_io_buffer_write;
3052 return 0;
3053}
3054
Mingming Cao4c0425f2009-09-28 15:48:41 -04003055/*
3056 * For ext4 extent files, ext4 will do direct-io write to holes,
3057 * preallocated extents, and those write extend the file, no need to
3058 * fall back to buffered IO.
3059 *
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003060 * For holes, we fallocate those blocks, mark them as uninitialized
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003061 * If those blocks were preallocated, we mark sure they are split, but
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003062 * still keep the range to write as uninitialized.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003063 *
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003064 * The unwritten extents will be converted to written when DIO is completed.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003065 * For async direct IO, since the IO may still pending when return, we
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003066 * set up an end_io call back function, which will do the conversion
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003067 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003068 *
3069 * If the O_DIRECT write will extend the file then add this inode to the
3070 * orphan list. So recovery will truncate it back to the original size
3071 * if the machine crashes during the write.
3072 *
3073 */
3074static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3075 const struct iovec *iov, loff_t offset,
3076 unsigned long nr_segs)
3077{
3078 struct file *file = iocb->ki_filp;
3079 struct inode *inode = file->f_mapping->host;
3080 ssize_t ret;
3081 size_t count = iov_length(iov, nr_segs);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003082 int overwrite = 0;
3083 get_block_t *get_block_func = NULL;
3084 int dio_flags = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003085 loff_t final_size = offset + count;
Zheng Liu729f52c2012-07-09 16:29:29 -04003086
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003087 /* Use the old path for reads and writes beyond i_size. */
3088 if (rw != WRITE || final_size > inode->i_size)
3089 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003090
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003091 BUG_ON(iocb->private == NULL);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003092
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003093 /* If we do a overwrite dio, i_mutex locking can be released */
3094 overwrite = *((int *)iocb->private);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003095
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003096 if (overwrite) {
3097 atomic_inc(&inode->i_dio_count);
3098 down_read(&EXT4_I(inode)->i_data_sem);
3099 mutex_unlock(&inode->i_mutex);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003100 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003101
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003102 /*
3103 * We could direct write to holes and fallocate.
3104 *
3105 * Allocated blocks to fill the hole are marked as
3106 * uninitialized to prevent parallel buffered read to expose
3107 * the stale data before DIO complete the data IO.
3108 *
3109 * As to previously fallocated extents, ext4 get_block will
3110 * just simply mark the buffer mapped but still keep the
3111 * extents uninitialized.
3112 *
3113 * For non AIO case, we will convert those unwritten extents
3114 * to written after return back from blockdev_direct_IO.
3115 *
3116 * For async DIO, the conversion needs to be deferred when the
3117 * IO is completed. The ext4 end_io callback function will be
3118 * called to take care of the conversion work. Here for async
3119 * case, we allocate an io_end structure to hook to the iocb.
3120 */
3121 iocb->private = NULL;
3122 ext4_inode_aio_set(inode, NULL);
3123 if (!is_sync_kiocb(iocb)) {
3124 ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS);
3125 if (!io_end) {
3126 ret = -ENOMEM;
3127 goto retake_lock;
3128 }
3129 io_end->flag |= EXT4_IO_END_DIRECT;
3130 iocb->private = io_end;
3131 /*
3132 * we save the io structure for current async direct
3133 * IO, so that later ext4_map_blocks() could flag the
3134 * io structure whether there is a unwritten extents
3135 * needs to be converted when IO is completed.
3136 */
3137 ext4_inode_aio_set(inode, io_end);
3138 }
3139
3140 if (overwrite) {
3141 get_block_func = ext4_get_block_write_nolock;
3142 } else {
3143 get_block_func = ext4_get_block_write;
3144 dio_flags = DIO_LOCKING;
3145 }
3146 ret = __blockdev_direct_IO(rw, iocb, inode,
3147 inode->i_sb->s_bdev, iov,
3148 offset, nr_segs,
3149 get_block_func,
3150 ext4_end_io_dio,
3151 NULL,
3152 dio_flags);
3153
3154 if (iocb->private)
3155 ext4_inode_aio_set(inode, NULL);
3156 /*
3157 * The io_end structure takes a reference to the inode, that
3158 * structure needs to be destroyed and the reference to the
3159 * inode need to be dropped, when IO is complete, even with 0
3160 * byte write, or failed.
3161 *
3162 * In the successful AIO DIO case, the io_end structure will
3163 * be destroyed and the reference to the inode will be dropped
3164 * after the end_io call back function is called.
3165 *
3166 * In the case there is 0 byte write, or error case, since VFS
3167 * direct IO won't invoke the end_io call back function, we
3168 * need to free the end_io structure here.
3169 */
3170 if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) {
3171 ext4_free_io_end(iocb->private);
3172 iocb->private = NULL;
3173 } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3174 EXT4_STATE_DIO_UNWRITTEN)) {
3175 int err;
3176 /*
3177 * for non AIO case, since the IO is already
3178 * completed, we could do the conversion right here
3179 */
3180 err = ext4_convert_unwritten_extents(inode,
3181 offset, ret);
3182 if (err < 0)
3183 ret = err;
3184 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3185 }
3186
3187retake_lock:
3188 /* take i_mutex locking again if we do a ovewrite dio */
3189 if (overwrite) {
3190 inode_dio_done(inode);
3191 up_read(&EXT4_I(inode)->i_data_sem);
3192 mutex_lock(&inode->i_mutex);
3193 }
3194
3195 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003196}
3197
3198static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3199 const struct iovec *iov, loff_t offset,
3200 unsigned long nr_segs)
3201{
3202 struct file *file = iocb->ki_filp;
3203 struct inode *inode = file->f_mapping->host;
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003204 ssize_t ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003205
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003206 /*
3207 * If we are doing data journalling we don't support O_DIRECT
3208 */
3209 if (ext4_should_journal_data(inode))
3210 return 0;
3211
Tao Ma46c7f252012-12-10 14:04:52 -05003212 /* Let buffer I/O handle the inline data case. */
3213 if (ext4_has_inline_data(inode))
3214 return 0;
3215
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003216 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003217 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003218 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3219 else
3220 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3221 trace_ext4_direct_IO_exit(inode, offset,
3222 iov_length(iov, nr_segs), rw, ret);
3223 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003224}
3225
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003226/*
Mingming Cao617ba132006-10-11 01:20:53 -07003227 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003228 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3229 * much here because ->set_page_dirty is called under VFS locks. The page is
3230 * not necessarily locked.
3231 *
3232 * We cannot just dirty the page and leave attached buffers clean, because the
3233 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3234 * or jbddirty because all the journalling code will explode.
3235 *
3236 * So what we do is to mark the page "pending dirty" and next time writepage
3237 * is called, propagate that into the buffers appropriately.
3238 */
Mingming Cao617ba132006-10-11 01:20:53 -07003239static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003240{
3241 SetPageChecked(page);
3242 return __set_page_dirty_nobuffers(page);
3243}
3244
Mingming Cao617ba132006-10-11 01:20:53 -07003245static const struct address_space_operations ext4_ordered_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003246 .readpage = ext4_readpage,
3247 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003248 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003249 .write_begin = ext4_write_begin,
3250 .write_end = ext4_ordered_write_end,
3251 .bmap = ext4_bmap,
3252 .invalidatepage = ext4_invalidatepage,
3253 .releasepage = ext4_releasepage,
3254 .direct_IO = ext4_direct_IO,
3255 .migratepage = buffer_migrate_page,
3256 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003257 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003258};
3259
Mingming Cao617ba132006-10-11 01:20:53 -07003260static const struct address_space_operations ext4_writeback_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003261 .readpage = ext4_readpage,
3262 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003263 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003264 .write_begin = ext4_write_begin,
3265 .write_end = ext4_writeback_write_end,
3266 .bmap = ext4_bmap,
3267 .invalidatepage = ext4_invalidatepage,
3268 .releasepage = ext4_releasepage,
3269 .direct_IO = ext4_direct_IO,
3270 .migratepage = buffer_migrate_page,
3271 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003272 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003273};
3274
Mingming Cao617ba132006-10-11 01:20:53 -07003275static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003276 .readpage = ext4_readpage,
3277 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003278 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003279 .write_begin = ext4_write_begin,
3280 .write_end = ext4_journalled_write_end,
3281 .set_page_dirty = ext4_journalled_set_page_dirty,
3282 .bmap = ext4_bmap,
Jan Kara4520fb32012-12-25 13:28:54 -05003283 .invalidatepage = ext4_journalled_invalidatepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003284 .releasepage = ext4_releasepage,
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003285 .direct_IO = ext4_direct_IO,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003286 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003287 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003288};
3289
Alex Tomas64769242008-07-11 19:27:31 -04003290static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003291 .readpage = ext4_readpage,
3292 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003293 .writepage = ext4_writepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003294 .writepages = ext4_da_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003295 .write_begin = ext4_da_write_begin,
3296 .write_end = ext4_da_write_end,
3297 .bmap = ext4_bmap,
3298 .invalidatepage = ext4_da_invalidatepage,
3299 .releasepage = ext4_releasepage,
3300 .direct_IO = ext4_direct_IO,
3301 .migratepage = buffer_migrate_page,
3302 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003303 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003304};
3305
Mingming Cao617ba132006-10-11 01:20:53 -07003306void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003307{
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003308 switch (ext4_inode_journal_mode(inode)) {
3309 case EXT4_INODE_ORDERED_DATA_MODE:
3310 if (test_opt(inode->i_sb, DELALLOC))
3311 inode->i_mapping->a_ops = &ext4_da_aops;
3312 else
3313 inode->i_mapping->a_ops = &ext4_ordered_aops;
3314 break;
3315 case EXT4_INODE_WRITEBACK_DATA_MODE:
3316 if (test_opt(inode->i_sb, DELALLOC))
3317 inode->i_mapping->a_ops = &ext4_da_aops;
3318 else
3319 inode->i_mapping->a_ops = &ext4_writeback_aops;
3320 break;
3321 case EXT4_INODE_JOURNAL_DATA_MODE:
Mingming Cao617ba132006-10-11 01:20:53 -07003322 inode->i_mapping->a_ops = &ext4_journalled_aops;
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003323 break;
3324 default:
3325 BUG();
3326 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003327}
3328
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003329
3330/*
3331 * ext4_discard_partial_page_buffers()
3332 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3333 * This function finds and locks the page containing the offset
3334 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3335 * Calling functions that already have the page locked should call
3336 * ext4_discard_partial_page_buffers_no_lock directly.
3337 */
3338int ext4_discard_partial_page_buffers(handle_t *handle,
3339 struct address_space *mapping, loff_t from,
3340 loff_t length, int flags)
3341{
3342 struct inode *inode = mapping->host;
3343 struct page *page;
3344 int err = 0;
3345
3346 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3347 mapping_gfp_mask(mapping) & ~__GFP_FS);
3348 if (!page)
Yongqiang Yang5129d052011-10-31 17:56:10 -04003349 return -ENOMEM;
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003350
3351 err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page,
3352 from, length, flags);
3353
3354 unlock_page(page);
3355 page_cache_release(page);
3356 return err;
3357}
3358
3359/*
3360 * ext4_discard_partial_page_buffers_no_lock()
3361 * Zeros a page range of length 'length' starting from offset 'from'.
3362 * Buffer heads that correspond to the block aligned regions of the
3363 * zeroed range will be unmapped. Unblock aligned regions
3364 * will have the corresponding buffer head mapped if needed so that
3365 * that region of the page can be updated with the partial zero out.
3366 *
3367 * This function assumes that the page has already been locked. The
3368 * The range to be discarded must be contained with in the given page.
3369 * If the specified range exceeds the end of the page it will be shortened
3370 * to the end of the page that corresponds to 'from'. This function is
3371 * appropriate for updating a page and it buffer heads to be unmapped and
3372 * zeroed for blocks that have been either released, or are going to be
3373 * released.
3374 *
3375 * handle: The journal handle
3376 * inode: The files inode
3377 * page: A locked page that contains the offset "from"
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003378 * from: The starting byte offset (from the beginning of the file)
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003379 * to begin discarding
3380 * len: The length of bytes to discard
3381 * flags: Optional flags that may be used:
3382 *
3383 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3384 * Only zero the regions of the page whose buffer heads
3385 * have already been unmapped. This flag is appropriate
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003386 * for updating the contents of a page whose blocks may
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003387 * have already been released, and we only want to zero
3388 * out the regions that correspond to those released blocks.
3389 *
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003390 * Returns zero on success or negative on failure.
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003391 */
Eric Sandeen5f163cc2012-01-04 22:33:28 -05003392static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle,
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003393 struct inode *inode, struct page *page, loff_t from,
3394 loff_t length, int flags)
3395{
3396 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3397 unsigned int offset = from & (PAGE_CACHE_SIZE-1);
3398 unsigned int blocksize, max, pos;
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003399 ext4_lblk_t iblock;
3400 struct buffer_head *bh;
3401 int err = 0;
3402
3403 blocksize = inode->i_sb->s_blocksize;
3404 max = PAGE_CACHE_SIZE - offset;
3405
3406 if (index != page->index)
3407 return -EINVAL;
3408
3409 /*
3410 * correct length if it does not fall between
3411 * 'from' and the end of the page
3412 */
3413 if (length > max || length < 0)
3414 length = max;
3415
3416 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3417
Yongqiang Yang093e6e32011-12-13 22:05:05 -05003418 if (!page_has_buffers(page))
3419 create_empty_buffers(page, blocksize, 0);
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003420
3421 /* Find the buffer that contains "offset" */
3422 bh = page_buffers(page);
3423 pos = blocksize;
3424 while (offset >= pos) {
3425 bh = bh->b_this_page;
3426 iblock++;
3427 pos += blocksize;
3428 }
3429
3430 pos = offset;
3431 while (pos < offset + length) {
Yongqiang Yange260daf2011-10-31 17:54:36 -04003432 unsigned int end_of_block, range_to_discard;
3433
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003434 err = 0;
3435
3436 /* The length of space left to zero and unmap */
3437 range_to_discard = offset + length - pos;
3438
3439 /* The length of space until the end of the block */
3440 end_of_block = blocksize - (pos & (blocksize-1));
3441
3442 /*
3443 * Do not unmap or zero past end of block
3444 * for this buffer head
3445 */
3446 if (range_to_discard > end_of_block)
3447 range_to_discard = end_of_block;
3448
3449
3450 /*
3451 * Skip this buffer head if we are only zeroing unampped
3452 * regions of the page
3453 */
3454 if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED &&
3455 buffer_mapped(bh))
3456 goto next;
3457
3458 /* If the range is block aligned, unmap */
3459 if (range_to_discard == blocksize) {
3460 clear_buffer_dirty(bh);
3461 bh->b_bdev = NULL;
3462 clear_buffer_mapped(bh);
3463 clear_buffer_req(bh);
3464 clear_buffer_new(bh);
3465 clear_buffer_delay(bh);
3466 clear_buffer_unwritten(bh);
3467 clear_buffer_uptodate(bh);
3468 zero_user(page, pos, range_to_discard);
3469 BUFFER_TRACE(bh, "Buffer discarded");
3470 goto next;
3471 }
3472
3473 /*
3474 * If this block is not completely contained in the range
3475 * to be discarded, then it is not going to be released. Because
3476 * we need to keep this block, we need to make sure this part
3477 * of the page is uptodate before we modify it by writeing
3478 * partial zeros on it.
3479 */
3480 if (!buffer_mapped(bh)) {
3481 /*
3482 * Buffer head must be mapped before we can read
3483 * from the block
3484 */
3485 BUFFER_TRACE(bh, "unmapped");
3486 ext4_get_block(inode, iblock, bh, 0);
3487 /* unmapped? It's a hole - nothing to do */
3488 if (!buffer_mapped(bh)) {
3489 BUFFER_TRACE(bh, "still unmapped");
3490 goto next;
3491 }
3492 }
3493
3494 /* Ok, it's mapped. Make sure it's up-to-date */
3495 if (PageUptodate(page))
3496 set_buffer_uptodate(bh);
3497
3498 if (!buffer_uptodate(bh)) {
3499 err = -EIO;
3500 ll_rw_block(READ, 1, &bh);
3501 wait_on_buffer(bh);
3502 /* Uhhuh. Read error. Complain and punt.*/
3503 if (!buffer_uptodate(bh))
3504 goto next;
3505 }
3506
3507 if (ext4_should_journal_data(inode)) {
3508 BUFFER_TRACE(bh, "get write access");
3509 err = ext4_journal_get_write_access(handle, bh);
3510 if (err)
3511 goto next;
3512 }
3513
3514 zero_user(page, pos, range_to_discard);
3515
3516 err = 0;
3517 if (ext4_should_journal_data(inode)) {
3518 err = ext4_handle_dirty_metadata(handle, inode, bh);
Theodore Ts'odecbd912011-09-06 02:37:06 -04003519 } else
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003520 mark_buffer_dirty(bh);
Allison Henderson4e96b2d2011-09-03 11:51:09 -04003521
3522 BUFFER_TRACE(bh, "Partial buffer zeroed");
3523next:
3524 bh = bh->b_this_page;
3525 iblock++;
3526 pos += range_to_discard;
3527 }
3528
3529 return err;
3530}
3531
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003532int ext4_can_truncate(struct inode *inode)
3533{
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003534 if (S_ISREG(inode->i_mode))
3535 return 1;
3536 if (S_ISDIR(inode->i_mode))
3537 return 1;
3538 if (S_ISLNK(inode->i_mode))
3539 return !ext4_inode_is_fast_symlink(inode);
3540 return 0;
3541}
3542
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003543/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003544 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3545 * associated with the given offset and length
3546 *
3547 * @inode: File inode
3548 * @offset: The offset where the hole will begin
3549 * @len: The length of the hole
3550 *
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003551 * Returns: 0 on success or negative on failure
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003552 */
3553
3554int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3555{
3556 struct inode *inode = file->f_path.dentry->d_inode;
3557 if (!S_ISREG(inode->i_mode))
Allison Henderson73355192012-03-21 22:23:31 -04003558 return -EOPNOTSUPP;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003559
3560 if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
3561 /* TODO: Add support for non extent hole punching */
Allison Henderson73355192012-03-21 22:23:31 -04003562 return -EOPNOTSUPP;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003563 }
3564
Theodore Ts'obab08ab2011-09-09 18:36:51 -04003565 if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) {
3566 /* TODO: Add support for bigalloc file systems */
Allison Henderson73355192012-03-21 22:23:31 -04003567 return -EOPNOTSUPP;
Theodore Ts'obab08ab2011-09-09 18:36:51 -04003568 }
3569
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003570 return ext4_ext_punch_hole(file, offset, length);
3571}
3572
3573/*
Mingming Cao617ba132006-10-11 01:20:53 -07003574 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003575 *
Mingming Cao617ba132006-10-11 01:20:53 -07003576 * We block out ext4_get_block() block instantiations across the entire
3577 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003578 * simultaneously on behalf of the same inode.
3579 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003580 * As we work through the truncate and commit bits of it to the journal there
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003581 * is one core, guiding principle: the file's tree must always be consistent on
3582 * disk. We must be able to restart the truncate after a crash.
3583 *
3584 * The file's tree may be transiently inconsistent in memory (although it
3585 * probably isn't), but whenever we close off and commit a journal transaction,
3586 * the contents of (the filesystem + the journal) must be consistent and
3587 * restartable. It's pretty simple, really: bottom up, right to left (although
3588 * left-to-right works OK too).
3589 *
3590 * Note that at recovery time, journal replay occurs *before* the restart of
3591 * truncate against the orphan inode list.
3592 *
3593 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07003594 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003595 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07003596 * ext4_truncate() to have another go. So there will be instantiated blocks
3597 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003598 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07003599 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003600 */
Mingming Cao617ba132006-10-11 01:20:53 -07003601void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003602{
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003603 trace_ext4_truncate_enter(inode);
3604
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003605 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003606 return;
3607
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003608 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003609
Theodore Ts'o5534fb52009-09-17 09:34:16 -04003610 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003611 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05003612
Tao Maaef1c852012-12-10 14:06:02 -05003613 if (ext4_has_inline_data(inode)) {
3614 int has_inline = 1;
3615
3616 ext4_inline_data_truncate(inode, &has_inline);
3617 if (has_inline)
3618 return;
3619 }
3620
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003621 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Jan Karacf108bc2008-07-11 19:27:31 -04003622 ext4_ext_truncate(inode);
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003623 else
3624 ext4_ind_truncate(inode);
Alex Tomasa86c6182006-10-11 01:21:03 -07003625
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003626 trace_ext4_truncate_exit(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003627}
3628
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003629/*
Mingming Cao617ba132006-10-11 01:20:53 -07003630 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003631 * underlying buffer_head on success. If 'in_mem' is true, we have all
3632 * data in memory that is needed to recreate the on-disk version of this
3633 * inode.
3634 */
Mingming Cao617ba132006-10-11 01:20:53 -07003635static int __ext4_get_inode_loc(struct inode *inode,
3636 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003637{
Theodore Ts'o240799c2008-10-09 23:53:47 -04003638 struct ext4_group_desc *gdp;
3639 struct buffer_head *bh;
3640 struct super_block *sb = inode->i_sb;
3641 ext4_fsblk_t block;
3642 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003643
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003644 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003645 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003646 return -EIO;
3647
Theodore Ts'o240799c2008-10-09 23:53:47 -04003648 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3649 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3650 if (!gdp)
3651 return -EIO;
3652
3653 /*
3654 * Figure out the offset within the block group inode table
3655 */
Tao Ma00d09882011-05-09 10:26:41 -04003656 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003657 inode_offset = ((inode->i_ino - 1) %
3658 EXT4_INODES_PER_GROUP(sb));
3659 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3660 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3661
3662 bh = sb_getblk(sb, block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003663 if (!bh) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003664 EXT4_ERROR_INODE_BLOCK(inode, block,
3665 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003666 return -EIO;
3667 }
3668 if (!buffer_uptodate(bh)) {
3669 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04003670
3671 /*
3672 * If the buffer has the write error flag, we have failed
3673 * to write out another inode in the same block. In this
3674 * case, we don't have to read the block because we may
3675 * read the old inode data successfully.
3676 */
3677 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3678 set_buffer_uptodate(bh);
3679
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003680 if (buffer_uptodate(bh)) {
3681 /* someone brought it uptodate while we waited */
3682 unlock_buffer(bh);
3683 goto has_buffer;
3684 }
3685
3686 /*
3687 * If we have all information of the inode in memory and this
3688 * is the only valid inode in the block, we need not read the
3689 * block.
3690 */
3691 if (in_mem) {
3692 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003693 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003694
Theodore Ts'o240799c2008-10-09 23:53:47 -04003695 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003696
3697 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003698 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003699 if (!bitmap_bh)
3700 goto make_io;
3701
3702 /*
3703 * If the inode bitmap isn't in cache then the
3704 * optimisation may end up performing two reads instead
3705 * of one, so skip it.
3706 */
3707 if (!buffer_uptodate(bitmap_bh)) {
3708 brelse(bitmap_bh);
3709 goto make_io;
3710 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04003711 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003712 if (i == inode_offset)
3713 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07003714 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003715 break;
3716 }
3717 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003718 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003719 /* all other inodes are free, so skip I/O */
3720 memset(bh->b_data, 0, bh->b_size);
3721 set_buffer_uptodate(bh);
3722 unlock_buffer(bh);
3723 goto has_buffer;
3724 }
3725 }
3726
3727make_io:
3728 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04003729 * If we need to do any I/O, try to pre-readahead extra
3730 * blocks from the inode table.
3731 */
3732 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3733 ext4_fsblk_t b, end, table;
3734 unsigned num;
3735
3736 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04003737 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003738 b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1);
3739 if (table > b)
3740 b = table;
3741 end = b + EXT4_SB(sb)->s_inode_readahead_blks;
3742 num = EXT4_INODES_PER_GROUP(sb);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04003743 if (ext4_has_group_desc_csum(sb))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05003744 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003745 table += num / inodes_per_block;
3746 if (end > table)
3747 end = table;
3748 while (b <= end)
3749 sb_breadahead(sb, b++);
3750 }
3751
3752 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003753 * There are other valid inodes in the buffer, this inode
3754 * has in-inode xattrs, or we don't have this inode in memory.
3755 * Read the block from disk.
3756 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003757 trace_ext4_load_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003758 get_bh(bh);
3759 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02003760 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003761 wait_on_buffer(bh);
3762 if (!buffer_uptodate(bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003763 EXT4_ERROR_INODE_BLOCK(inode, block,
3764 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003765 brelse(bh);
3766 return -EIO;
3767 }
3768 }
3769has_buffer:
3770 iloc->bh = bh;
3771 return 0;
3772}
3773
Mingming Cao617ba132006-10-11 01:20:53 -07003774int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003775{
3776 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07003777 return __ext4_get_inode_loc(inode, iloc,
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003778 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003779}
3780
Mingming Cao617ba132006-10-11 01:20:53 -07003781void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003782{
Mingming Cao617ba132006-10-11 01:20:53 -07003783 unsigned int flags = EXT4_I(inode)->i_flags;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003784
3785 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Mingming Cao617ba132006-10-11 01:20:53 -07003786 if (flags & EXT4_SYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003787 inode->i_flags |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07003788 if (flags & EXT4_APPEND_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003789 inode->i_flags |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07003790 if (flags & EXT4_IMMUTABLE_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003791 inode->i_flags |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07003792 if (flags & EXT4_NOATIME_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003793 inode->i_flags |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07003794 if (flags & EXT4_DIRSYNC_FL)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003795 inode->i_flags |= S_DIRSYNC;
3796}
3797
Jan Karaff9ddf72007-07-18 09:24:20 -04003798/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3799void ext4_get_inode_flags(struct ext4_inode_info *ei)
3800{
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003801 unsigned int vfs_fl;
3802 unsigned long old_fl, new_fl;
Jan Karaff9ddf72007-07-18 09:24:20 -04003803
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003804 do {
3805 vfs_fl = ei->vfs_inode.i_flags;
3806 old_fl = ei->i_flags;
3807 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3808 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3809 EXT4_DIRSYNC_FL);
3810 if (vfs_fl & S_SYNC)
3811 new_fl |= EXT4_SYNC_FL;
3812 if (vfs_fl & S_APPEND)
3813 new_fl |= EXT4_APPEND_FL;
3814 if (vfs_fl & S_IMMUTABLE)
3815 new_fl |= EXT4_IMMUTABLE_FL;
3816 if (vfs_fl & S_NOATIME)
3817 new_fl |= EXT4_NOATIME_FL;
3818 if (vfs_fl & S_DIRSYNC)
3819 new_fl |= EXT4_DIRSYNC_FL;
3820 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
Jan Karaff9ddf72007-07-18 09:24:20 -04003821}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003822
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003823static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003824 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003825{
3826 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003827 struct inode *inode = &(ei->vfs_inode);
3828 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003829
3830 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3831 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3832 /* we are using combined 48 bit field */
3833 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3834 le32_to_cpu(raw_inode->i_blocks_lo);
Theodore Ts'o07a03822010-06-14 09:54:48 -04003835 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003836 /* i_blocks represent file system block size */
3837 return i_blocks << (inode->i_blkbits - 9);
3838 } else {
3839 return i_blocks;
3840 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003841 } else {
3842 return le32_to_cpu(raw_inode->i_blocks_lo);
3843 }
3844}
Jan Karaff9ddf72007-07-18 09:24:20 -04003845
Tao Ma152a7b02012-12-02 11:13:24 -05003846static inline void ext4_iget_extra_inode(struct inode *inode,
3847 struct ext4_inode *raw_inode,
3848 struct ext4_inode_info *ei)
3849{
3850 __le32 *magic = (void *)raw_inode +
3851 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
Tao Ma67cf5b02012-12-10 14:04:46 -05003852 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
Tao Ma152a7b02012-12-02 11:13:24 -05003853 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
Tao Ma67cf5b02012-12-10 14:04:46 -05003854 ext4_find_inline_data_nolock(inode);
Tao Maf19d5872012-12-10 14:05:51 -05003855 } else
3856 EXT4_I(inode)->i_inline_off = 0;
Tao Ma152a7b02012-12-02 11:13:24 -05003857}
3858
David Howells1d1fe1e2008-02-07 00:15:37 -08003859struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003860{
Mingming Cao617ba132006-10-11 01:20:53 -07003861 struct ext4_iloc iloc;
3862 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08003863 struct ext4_inode_info *ei;
David Howells1d1fe1e2008-02-07 00:15:37 -08003864 struct inode *inode;
Jan Karab436b9b2009-12-08 23:51:10 -05003865 journal_t *journal = EXT4_SB(sb)->s_journal;
David Howells1d1fe1e2008-02-07 00:15:37 -08003866 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003867 int block;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08003868 uid_t i_uid;
3869 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003870
David Howells1d1fe1e2008-02-07 00:15:37 -08003871 inode = iget_locked(sb, ino);
3872 if (!inode)
3873 return ERR_PTR(-ENOMEM);
3874 if (!(inode->i_state & I_NEW))
3875 return inode;
3876
3877 ei = EXT4_I(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05003878 iloc.bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003879
David Howells1d1fe1e2008-02-07 00:15:37 -08003880 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3881 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003882 goto bad_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07003883 raw_inode = ext4_raw_inode(&iloc);
Darrick J. Wong814525f2012-04-29 18:31:10 -04003884
3885 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3886 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3887 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3888 EXT4_INODE_SIZE(inode->i_sb)) {
3889 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3890 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3891 EXT4_INODE_SIZE(inode->i_sb));
3892 ret = -EIO;
3893 goto bad_inode;
3894 }
3895 } else
3896 ei->i_extra_isize = 0;
3897
3898 /* Precompute checksum seed for inode metadata */
3899 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3900 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3901 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3902 __u32 csum;
3903 __le32 inum = cpu_to_le32(inode->i_ino);
3904 __le32 gen = raw_inode->i_generation;
3905 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3906 sizeof(inum));
3907 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3908 sizeof(gen));
3909 }
3910
3911 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3912 EXT4_ERROR_INODE(inode, "checksum invalid");
3913 ret = -EIO;
3914 goto bad_inode;
3915 }
3916
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003917 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08003918 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3919 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04003920 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08003921 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3922 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003923 }
Eric W. Biederman08cefc72012-02-07 15:41:49 -08003924 i_uid_write(inode, i_uid);
3925 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02003926 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003927
Theodore Ts'o353eb832011-01-10 12:18:25 -05003928 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Tao Ma67cf5b02012-12-10 14:04:46 -05003929 ei->i_inline_off = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003930 ei->i_dir_start_lookup = 0;
3931 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3932 /* We now have enough fields to check if the inode was active or not.
3933 * This is needed because nfsd might try to access dead inodes
3934 * the test is that same one that e2fsck uses
3935 * NeilBrown 1999oct15
3936 */
3937 if (inode->i_nlink == 0) {
3938 if (inode->i_mode == 0 ||
Mingming Cao617ba132006-10-11 01:20:53 -07003939 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003940 /* this inode is deleted */
David Howells1d1fe1e2008-02-07 00:15:37 -08003941 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003942 goto bad_inode;
3943 }
3944 /* The only unlinked inodes we let through here have
3945 * valid i_mode and are being read by the orphan
3946 * recovery code: that's fine, we're about to complete
3947 * the process of deleting those. */
3948 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003949 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003950 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05003951 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04003952 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07003953 ei->i_file_acl |=
3954 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05003955 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003956 ei->i_disksize = inode->i_size;
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03003957#ifdef CONFIG_QUOTA
3958 ei->i_reserved_quota = 0;
3959#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003960 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3961 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04003962 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003963 /*
3964 * NOTE! The in-memory inode i_data array is in little-endian order
3965 * even on big-endian machines: we do NOT byteswap the block numbers!
3966 */
Mingming Cao617ba132006-10-11 01:20:53 -07003967 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003968 ei->i_data[block] = raw_inode->i_block[block];
3969 INIT_LIST_HEAD(&ei->i_orphan);
3970
Jan Karab436b9b2009-12-08 23:51:10 -05003971 /*
3972 * Set transaction id's of transactions that have to be committed
3973 * to finish f[data]sync. We set them to currently running transaction
3974 * as we cannot be sure that the inode or some of its metadata isn't
3975 * part of the transaction - the inode could have been reclaimed and
3976 * now it is reread from disk.
3977 */
3978 if (journal) {
3979 transaction_t *transaction;
3980 tid_t tid;
3981
Theodore Ts'oa931da62010-08-03 21:35:12 -04003982 read_lock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05003983 if (journal->j_running_transaction)
3984 transaction = journal->j_running_transaction;
3985 else
3986 transaction = journal->j_committing_transaction;
3987 if (transaction)
3988 tid = transaction->t_tid;
3989 else
3990 tid = journal->j_commit_sequence;
Theodore Ts'oa931da62010-08-03 21:35:12 -04003991 read_unlock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05003992 ei->i_sync_tid = tid;
3993 ei->i_datasync_tid = tid;
3994 }
3995
Eric Sandeen0040d982008-02-05 22:36:43 -05003996 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003997 if (ei->i_extra_isize == 0) {
3998 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07003999 ei->i_extra_isize = sizeof(struct ext4_inode) -
4000 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004001 } else {
Tao Ma152a7b02012-12-02 11:13:24 -05004002 ext4_iget_extra_inode(inode, raw_inode, ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004003 }
Darrick J. Wong814525f2012-04-29 18:31:10 -04004004 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004005
Kalpak Shahef7f3832007-07-18 09:15:20 -04004006 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4007 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4008 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4009 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4010
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004011 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4012 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4013 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4014 inode->i_version |=
4015 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4016 }
4017
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004018 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004019 if (ei->i_file_acl &&
Theodore Ts'o10329882009-11-15 15:29:56 -05004020 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04004021 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4022 ei->i_file_acl);
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004023 ret = -EIO;
4024 goto bad_inode;
Tao Maf19d5872012-12-10 14:05:51 -05004025 } else if (!ext4_has_inline_data(inode)) {
4026 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4027 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4028 (S_ISLNK(inode->i_mode) &&
4029 !ext4_inode_is_fast_symlink(inode))))
4030 /* Validate extent which is part of inode */
4031 ret = ext4_ext_check_inode(inode);
4032 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4033 (S_ISLNK(inode->i_mode) &&
4034 !ext4_inode_is_fast_symlink(inode))) {
4035 /* Validate block references which are part of inode */
4036 ret = ext4_ind_check_inode(inode);
4037 }
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004038 }
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004039 if (ret)
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004040 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04004041
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004042 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004043 inode->i_op = &ext4_file_inode_operations;
4044 inode->i_fop = &ext4_file_operations;
4045 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004046 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004047 inode->i_op = &ext4_dir_inode_operations;
4048 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004049 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00004050 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004051 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00004052 nd_terminate_link(ei->i_data, inode->i_size,
4053 sizeof(ei->i_data) - 1);
4054 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004055 inode->i_op = &ext4_symlink_inode_operations;
4056 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004057 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004058 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4059 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004060 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004061 if (raw_inode->i_block[0])
4062 init_special_inode(inode, inode->i_mode,
4063 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4064 else
4065 init_special_inode(inode, inode->i_mode,
4066 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004067 } else {
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004068 ret = -EIO;
Theodore Ts'o24676da2010-05-16 21:00:00 -04004069 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004070 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004071 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004072 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004073 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004074 unlock_new_inode(inode);
4075 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004076
4077bad_inode:
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004078 brelse(iloc.bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004079 iget_failed(inode);
4080 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004081}
4082
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004083static int ext4_inode_blocks_set(handle_t *handle,
4084 struct ext4_inode *raw_inode,
4085 struct ext4_inode_info *ei)
4086{
4087 struct inode *inode = &(ei->vfs_inode);
4088 u64 i_blocks = inode->i_blocks;
4089 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004090
4091 if (i_blocks <= ~0U) {
4092 /*
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004093 * i_blocks can be represented in a 32 bit variable
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004094 * as multiple of 512 bytes
4095 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004096 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004097 raw_inode->i_blocks_high = 0;
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004098 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004099 return 0;
4100 }
4101 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4102 return -EFBIG;
4103
4104 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004105 /*
4106 * i_blocks can be represented in a 48 bit variable
4107 * as multiple of 512 bytes
4108 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004109 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004110 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004111 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004112 } else {
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004113 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004114 /* i_block is stored in file system block size */
4115 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4116 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4117 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004118 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004119 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004120}
4121
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004122/*
4123 * Post the struct inode info into an on-disk inode location in the
4124 * buffer-cache. This gobbles the caller's reference to the
4125 * buffer_head in the inode location struct.
4126 *
4127 * The caller must have write access to iloc->bh.
4128 */
Mingming Cao617ba132006-10-11 01:20:53 -07004129static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004130 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04004131 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004132{
Mingming Cao617ba132006-10-11 01:20:53 -07004133 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4134 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004135 struct buffer_head *bh = iloc->bh;
4136 int err = 0, rc, block;
Jan Karab71fc072012-09-26 21:52:20 -04004137 int need_datasync = 0;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004138 uid_t i_uid;
4139 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004140
4141 /* For fields not not tracking in the in-memory inode,
4142 * initialise them to zero for new inodes. */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004143 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
Mingming Cao617ba132006-10-11 01:20:53 -07004144 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004145
Jan Karaff9ddf72007-07-18 09:24:20 -04004146 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004147 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004148 i_uid = i_uid_read(inode);
4149 i_gid = i_gid_read(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004150 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004151 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4152 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004153/*
4154 * Fix up interoperability with old kernels. Otherwise, old inodes get
4155 * re-used with the upper 16 bits of the uid/gid intact
4156 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004157 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004158 raw_inode->i_uid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004159 cpu_to_le16(high_16_bits(i_uid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004160 raw_inode->i_gid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004161 cpu_to_le16(high_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004162 } else {
4163 raw_inode->i_uid_high = 0;
4164 raw_inode->i_gid_high = 0;
4165 }
4166 } else {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004167 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4168 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004169 raw_inode->i_uid_high = 0;
4170 raw_inode->i_gid_high = 0;
4171 }
4172 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004173
4174 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4175 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4176 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4177 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4178
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004179 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4180 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004181 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o353eb832011-01-10 12:18:25 -05004182 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
Mingming Cao9b8f1f02006-10-11 01:21:13 -07004183 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
4184 cpu_to_le32(EXT4_OS_HURD))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004185 raw_inode->i_file_acl_high =
4186 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004187 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Jan Karab71fc072012-09-26 21:52:20 -04004188 if (ei->i_disksize != ext4_isize(raw_inode)) {
4189 ext4_isize_set(raw_inode, ei->i_disksize);
4190 need_datasync = 1;
4191 }
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004192 if (ei->i_disksize > 0x7fffffffULL) {
4193 struct super_block *sb = inode->i_sb;
4194 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4195 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4196 EXT4_SB(sb)->s_es->s_rev_level ==
4197 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4198 /* If this is the first large file
4199 * created, add a flag to the superblock.
4200 */
4201 err = ext4_journal_get_write_access(handle,
4202 EXT4_SB(sb)->s_sbh);
4203 if (err)
4204 goto out_brelse;
4205 ext4_update_dynamic_rev(sb);
4206 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07004207 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Frank Mayhar03901312009-01-07 00:06:22 -05004208 ext4_handle_sync(handle);
Artem Bityutskiyb50924c2012-07-22 20:37:31 -04004209 err = ext4_handle_dirty_super(handle, sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004210 }
4211 }
4212 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4213 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4214 if (old_valid_dev(inode->i_rdev)) {
4215 raw_inode->i_block[0] =
4216 cpu_to_le32(old_encode_dev(inode->i_rdev));
4217 raw_inode->i_block[1] = 0;
4218 } else {
4219 raw_inode->i_block[0] = 0;
4220 raw_inode->i_block[1] =
4221 cpu_to_le32(new_encode_dev(inode->i_rdev));
4222 raw_inode->i_block[2] = 0;
4223 }
Tao Maf19d5872012-12-10 14:05:51 -05004224 } else if (!ext4_has_inline_data(inode)) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004225 for (block = 0; block < EXT4_N_BLOCKS; block++)
4226 raw_inode->i_block[block] = ei->i_data[block];
Tao Maf19d5872012-12-10 14:05:51 -05004227 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004228
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004229 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4230 if (ei->i_extra_isize) {
4231 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4232 raw_inode->i_version_hi =
4233 cpu_to_le32(inode->i_version >> 32);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004234 raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize);
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004235 }
4236
Darrick J. Wong814525f2012-04-29 18:31:10 -04004237 ext4_inode_csum_set(inode, raw_inode, ei);
4238
Frank Mayhar830156c2009-09-29 10:07:47 -04004239 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05004240 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
Frank Mayhar830156c2009-09-29 10:07:47 -04004241 if (!err)
4242 err = rc;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004243 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004244
Jan Karab71fc072012-09-26 21:52:20 -04004245 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004246out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004247 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004248 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004249 return err;
4250}
4251
4252/*
Mingming Cao617ba132006-10-11 01:20:53 -07004253 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004254 *
4255 * We are called from a few places:
4256 *
4257 * - Within generic_file_write() for O_SYNC files.
4258 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004259 * transaction to commit.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004260 *
4261 * - Within sys_sync(), kupdate and such.
4262 * We wait on commit, if tol to.
4263 *
4264 * - Within prune_icache() (PF_MEMALLOC == true)
4265 * Here we simply return. We can't afford to block kswapd on the
4266 * journal commit.
4267 *
4268 * In all cases it is actually safe for us to return without doing anything,
4269 * because the inode has been copied into a raw inode buffer in
Mingming Cao617ba132006-10-11 01:20:53 -07004270 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004271 * knfsd.
4272 *
4273 * Note that we are absolutely dependent upon all inode dirtiers doing the
4274 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4275 * which we are interested.
4276 *
4277 * It would be a bug for them to not do this. The code:
4278 *
4279 * mark_inode_dirty(inode)
4280 * stuff();
4281 * inode->i_size = expr;
4282 *
4283 * is in error because a kswapd-driven write_inode() could occur while
4284 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4285 * will no longer be on the superblock's dirty inode list.
4286 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004287int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004288{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004289 int err;
4290
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004291 if (current->flags & PF_MEMALLOC)
4292 return 0;
4293
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004294 if (EXT4_SB(inode->i_sb)->s_journal) {
4295 if (ext4_journal_current_handle()) {
4296 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4297 dump_stack();
4298 return -EIO;
4299 }
4300
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004301 if (wbc->sync_mode != WB_SYNC_ALL)
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004302 return 0;
4303
4304 err = ext4_force_commit(inode->i_sb);
4305 } else {
4306 struct ext4_iloc iloc;
4307
Curt Wohlgemuth8b472d72010-04-03 16:45:06 -04004308 err = __ext4_get_inode_loc(inode, &iloc, 0);
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004309 if (err)
4310 return err;
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004311 if (wbc->sync_mode == WB_SYNC_ALL)
Frank Mayhar830156c2009-09-29 10:07:47 -04004312 sync_dirty_buffer(iloc.bh);
4313 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004314 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4315 "IO error syncing inode");
Frank Mayhar830156c2009-09-29 10:07:47 -04004316 err = -EIO;
4317 }
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004318 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004319 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004320 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004321}
4322
4323/*
Jan Kara53e87262012-12-25 13:29:52 -05004324 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4325 * buffers that are attached to a page stradding i_size and are undergoing
4326 * commit. In that case we have to wait for commit to finish and try again.
4327 */
4328static void ext4_wait_for_tail_page_commit(struct inode *inode)
4329{
4330 struct page *page;
4331 unsigned offset;
4332 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4333 tid_t commit_tid = 0;
4334 int ret;
4335
4336 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4337 /*
4338 * All buffers in the last page remain valid? Then there's nothing to
4339 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4340 * blocksize case
4341 */
4342 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4343 return;
4344 while (1) {
4345 page = find_lock_page(inode->i_mapping,
4346 inode->i_size >> PAGE_CACHE_SHIFT);
4347 if (!page)
4348 return;
4349 ret = __ext4_journalled_invalidatepage(page, offset);
4350 unlock_page(page);
4351 page_cache_release(page);
4352 if (ret != -EBUSY)
4353 return;
4354 commit_tid = 0;
4355 read_lock(&journal->j_state_lock);
4356 if (journal->j_committing_transaction)
4357 commit_tid = journal->j_committing_transaction->t_tid;
4358 read_unlock(&journal->j_state_lock);
4359 if (commit_tid)
4360 jbd2_log_wait_commit(journal, commit_tid);
4361 }
4362}
4363
4364/*
Mingming Cao617ba132006-10-11 01:20:53 -07004365 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004366 *
4367 * Called from notify_change.
4368 *
4369 * We want to trap VFS attempts to truncate the file as soon as
4370 * possible. In particular, we want to make sure that when the VFS
4371 * shrinks i_size, we put the inode on the orphan list and modify
4372 * i_disksize immediately, so that during the subsequent flushing of
4373 * dirty pages and freeing of disk blocks, we can guarantee that any
4374 * commit will leave the blocks being flushed in an unused state on
4375 * disk. (On recovery, the inode will get truncated and the blocks will
4376 * be freed, so we have a strong guarantee that no future commit will
4377 * leave these blocks visible to the user.)
4378 *
Jan Kara678aaf42008-07-11 19:27:31 -04004379 * Another thing we have to assure is that if we are in ordered mode
4380 * and inode is still attached to the committing transaction, we must
4381 * we start writeout of all the dirty pages which are being truncated.
4382 * This way we are sure that all the data written in the previous
4383 * transaction are already on disk (truncate waits for pages under
4384 * writeback).
4385 *
4386 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004387 */
Mingming Cao617ba132006-10-11 01:20:53 -07004388int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004389{
4390 struct inode *inode = dentry->d_inode;
4391 int error, rc = 0;
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004392 int orphan = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004393 const unsigned int ia_valid = attr->ia_valid;
4394
4395 error = inode_change_ok(inode, attr);
4396 if (error)
4397 return error;
4398
Dmitry Monakhov12755622010-04-08 22:04:20 +04004399 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05004400 dquot_initialize(inode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004401 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4402 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004403 handle_t *handle;
4404
4405 /* (user+group)*(old+new) structure, inode write (sb,
4406 * inode block, ? - but truncate inode update has it) */
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05004407 handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+
Dmitry Monakhov194074a2009-12-08 22:42:28 -05004408 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004409 if (IS_ERR(handle)) {
4410 error = PTR_ERR(handle);
4411 goto err_out;
4412 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05004413 error = dquot_transfer(inode, attr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004414 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07004415 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004416 return error;
4417 }
4418 /* Update corresponding info in inode so that everything is in
4419 * one transaction */
4420 if (attr->ia_valid & ATTR_UID)
4421 inode->i_uid = attr->ia_uid;
4422 if (attr->ia_valid & ATTR_GID)
4423 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07004424 error = ext4_mark_inode_dirty(handle, inode);
4425 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004426 }
4427
Eric Sandeene2b46572008-01-28 23:58:27 -05004428 if (attr->ia_valid & ATTR_SIZE) {
Christoph Hellwig562c72a2011-06-24 14:29:45 -04004429
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004430 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -05004431 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4432
Theodore Ts'o0c095c72010-07-27 11:56:06 -04004433 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4434 return -EFBIG;
Eric Sandeene2b46572008-01-28 23:58:27 -05004435 }
4436 }
4437
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004438 if (S_ISREG(inode->i_mode) &&
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05004439 attr->ia_valid & ATTR_SIZE &&
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004440 (attr->ia_size < inode->i_size)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004441 handle_t *handle;
4442
Mingming Cao617ba132006-10-11 01:20:53 -07004443 handle = ext4_journal_start(inode, 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004444 if (IS_ERR(handle)) {
4445 error = PTR_ERR(handle);
4446 goto err_out;
4447 }
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004448 if (ext4_handle_valid(handle)) {
4449 error = ext4_orphan_add(handle, inode);
4450 orphan = 1;
4451 }
Mingming Cao617ba132006-10-11 01:20:53 -07004452 EXT4_I(inode)->i_disksize = attr->ia_size;
4453 rc = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004454 if (!error)
4455 error = rc;
Mingming Cao617ba132006-10-11 01:20:53 -07004456 ext4_journal_stop(handle);
Jan Kara678aaf42008-07-11 19:27:31 -04004457
4458 if (ext4_should_order_data(inode)) {
4459 error = ext4_begin_ordered_truncate(inode,
4460 attr->ia_size);
4461 if (error) {
4462 /* Do as much error cleanup as possible */
4463 handle = ext4_journal_start(inode, 3);
4464 if (IS_ERR(handle)) {
4465 ext4_orphan_del(NULL, inode);
4466 goto err_out;
4467 }
4468 ext4_orphan_del(handle, inode);
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004469 orphan = 0;
Jan Kara678aaf42008-07-11 19:27:31 -04004470 ext4_journal_stop(handle);
4471 goto err_out;
4472 }
4473 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004474 }
4475
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004476 if (attr->ia_valid & ATTR_SIZE) {
Jan Kara53e87262012-12-25 13:29:52 -05004477 if (attr->ia_size != inode->i_size) {
4478 loff_t oldsize = inode->i_size;
4479
4480 i_size_write(inode, attr->ia_size);
4481 /*
4482 * Blocks are going to be removed from the inode. Wait
4483 * for dio in flight. Temporarily disable
4484 * dioread_nolock to prevent livelock.
4485 */
Dmitry Monakhov1b650072012-09-29 00:56:15 -04004486 if (orphan) {
Jan Kara53e87262012-12-25 13:29:52 -05004487 if (!ext4_should_journal_data(inode)) {
4488 ext4_inode_block_unlocked_dio(inode);
4489 inode_dio_wait(inode);
4490 ext4_inode_resume_unlocked_dio(inode);
4491 } else
4492 ext4_wait_for_tail_page_commit(inode);
Dmitry Monakhov1b650072012-09-29 00:56:15 -04004493 }
Jan Kara53e87262012-12-25 13:29:52 -05004494 /*
4495 * Truncate pagecache after we've waited for commit
4496 * in data=journal mode to make pages freeable.
4497 */
4498 truncate_pagecache(inode, oldsize, inode->i_size);
Dmitry Monakhov1c9114f2012-09-29 00:55:23 -04004499 }
Lukas Czernerafcff5d2012-03-21 21:47:55 -04004500 ext4_truncate(inode);
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004501 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004502
Christoph Hellwig10257742010-06-04 11:30:02 +02004503 if (!rc) {
4504 setattr_copy(inode, attr);
4505 mark_inode_dirty(inode);
4506 }
4507
4508 /*
4509 * If the call to ext4_truncate failed to get a transaction handle at
4510 * all, we need to clean up the in-core orphan list manually.
4511 */
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004512 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004513 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004514
4515 if (!rc && (ia_valid & ATTR_MODE))
Mingming Cao617ba132006-10-11 01:20:53 -07004516 rc = ext4_acl_chmod(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004517
4518err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07004519 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004520 if (!error)
4521 error = rc;
4522 return error;
4523}
4524
Mingming Cao3e3398a2008-07-11 19:27:31 -04004525int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4526 struct kstat *stat)
4527{
4528 struct inode *inode;
4529 unsigned long delalloc_blocks;
4530
4531 inode = dentry->d_inode;
4532 generic_fillattr(inode, stat);
4533
4534 /*
4535 * We can't update i_blocks if the block allocation is delayed
4536 * otherwise in the case of system crash before the real block
4537 * allocation is done, we will have i_blocks inconsistent with
4538 * on-disk file blocks.
4539 * We always keep i_blocks updated together with real
4540 * allocation. But to not confuse with user, stat
4541 * will return the blocks that include the delayed allocation
4542 * blocks for this file.
4543 */
Tao Ma96607552012-05-31 22:54:16 -04004544 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
4545 EXT4_I(inode)->i_reserved_data_blocks);
Mingming Cao3e3398a2008-07-11 19:27:31 -04004546
4547 stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9;
4548 return 0;
4549}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004550
Mingming Caoa02908f2008-08-19 22:16:07 -04004551static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk)
4552{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004553 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Amir Goldstein8bb2b242011-06-27 17:10:28 -04004554 return ext4_ind_trans_blocks(inode, nrblocks, chunk);
Theodore Ts'oac51d832008-11-06 16:49:36 -05004555 return ext4_ext_index_trans_blocks(inode, nrblocks, chunk);
Mingming Caoa02908f2008-08-19 22:16:07 -04004556}
Theodore Ts'oac51d832008-11-06 16:49:36 -05004557
Mingming Caoa02908f2008-08-19 22:16:07 -04004558/*
4559 * Account for index blocks, block groups bitmaps and block group
4560 * descriptor blocks if modify datablocks and index blocks
4561 * worse case, the indexs blocks spread over different block groups
4562 *
4563 * If datablocks are discontiguous, they are possible to spread over
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004564 * different block groups too. If they are contiguous, with flexbg,
Mingming Caoa02908f2008-08-19 22:16:07 -04004565 * they could still across block group boundary.
4566 *
4567 * Also account for superblock, inode, quota and xattr blocks
4568 */
Theodore Ts'o1f109d52010-10-27 21:30:14 -04004569static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk)
Mingming Caoa02908f2008-08-19 22:16:07 -04004570{
Theodore Ts'o8df96752009-05-01 08:50:38 -04004571 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4572 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04004573 int idxblocks;
4574 int ret = 0;
4575
4576 /*
4577 * How many index blocks need to touch to modify nrblocks?
4578 * The "Chunk" flag indicating whether the nrblocks is
4579 * physically contiguous on disk
4580 *
4581 * For Direct IO and fallocate, they calls get_block to allocate
4582 * one single extent at a time, so they could set the "Chunk" flag
4583 */
4584 idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk);
4585
4586 ret = idxblocks;
4587
4588 /*
4589 * Now let's see how many group bitmaps and group descriptors need
4590 * to account
4591 */
4592 groups = idxblocks;
4593 if (chunk)
4594 groups += 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004595 else
Mingming Caoa02908f2008-08-19 22:16:07 -04004596 groups += nrblocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004597
Mingming Caoa02908f2008-08-19 22:16:07 -04004598 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04004599 if (groups > ngroups)
4600 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04004601 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4602 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4603
4604 /* bitmaps and block group descriptor blocks */
4605 ret += groups + gdpblocks;
4606
4607 /* Blocks for super block, inode, quota and xattr blocks */
4608 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004609
4610 return ret;
4611}
4612
4613/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004614 * Calculate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04004615 * the modification of a single pages into a single transaction,
4616 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04004617 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004618 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04004619 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004620 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04004621 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04004622 */
4623int ext4_writepage_trans_blocks(struct inode *inode)
4624{
4625 int bpp = ext4_journal_blocks_per_page(inode);
4626 int ret;
4627
4628 ret = ext4_meta_trans_blocks(inode, bpp, 0);
4629
4630 /* Account for data blocks for journalled mode */
4631 if (ext4_should_journal_data(inode))
4632 ret += bpp;
4633 return ret;
4634}
Mingming Caof3bd1f32008-08-19 22:16:03 -04004635
4636/*
4637 * Calculate the journal credits for a chunk of data modification.
4638 *
4639 * This is called from DIO, fallocate or whoever calling
Eric Sandeen79e83032010-07-27 11:56:07 -04004640 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04004641 *
4642 * journal buffers for data blocks are not included here, as DIO
4643 * and fallocate do no need to journal data buffers.
4644 */
4645int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4646{
4647 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4648}
4649
Mingming Caoa02908f2008-08-19 22:16:07 -04004650/*
Mingming Cao617ba132006-10-11 01:20:53 -07004651 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004652 * Give this, we know that the caller already has write access to iloc->bh.
4653 */
Mingming Cao617ba132006-10-11 01:20:53 -07004654int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004655 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004656{
4657 int err = 0;
4658
Theodore Ts'oc64db502012-03-02 12:23:11 -05004659 if (IS_I_VERSION(inode))
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004660 inode_inc_iversion(inode);
4661
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004662 /* the do_update_inode consumes one bh->b_count */
4663 get_bh(iloc->bh);
4664
Mingming Caodab291a2006-10-11 01:21:01 -07004665 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04004666 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004667 put_bh(iloc->bh);
4668 return err;
4669}
4670
4671/*
4672 * On success, We end up with an outstanding reference count against
4673 * iloc->bh. This _must_ be cleaned up later.
4674 */
4675
4676int
Mingming Cao617ba132006-10-11 01:20:53 -07004677ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4678 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004679{
Frank Mayhar03901312009-01-07 00:06:22 -05004680 int err;
4681
4682 err = ext4_get_inode_loc(inode, iloc);
4683 if (!err) {
4684 BUFFER_TRACE(iloc->bh, "get_write_access");
4685 err = ext4_journal_get_write_access(handle, iloc->bh);
4686 if (err) {
4687 brelse(iloc->bh);
4688 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004689 }
4690 }
Mingming Cao617ba132006-10-11 01:20:53 -07004691 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004692 return err;
4693}
4694
4695/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004696 * Expand an inode by new_extra_isize bytes.
4697 * Returns 0 on success or negative error number on failure.
4698 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004699static int ext4_expand_extra_isize(struct inode *inode,
4700 unsigned int new_extra_isize,
4701 struct ext4_iloc iloc,
4702 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004703{
4704 struct ext4_inode *raw_inode;
4705 struct ext4_xattr_ibody_header *header;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004706
4707 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4708 return 0;
4709
4710 raw_inode = ext4_raw_inode(&iloc);
4711
4712 header = IHDR(inode, raw_inode);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004713
4714 /* No extended attributes present */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004715 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4716 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004717 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4718 new_extra_isize);
4719 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4720 return 0;
4721 }
4722
4723 /* try to expand with EAs present */
4724 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4725 raw_inode, handle);
4726}
4727
4728/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004729 * What we do here is to mark the in-core inode as clean with respect to inode
4730 * dirtiness (it may still be data-dirty).
4731 * This means that the in-core inode may be reaped by prune_icache
4732 * without having to perform any I/O. This is a very good thing,
4733 * because *any* task may call prune_icache - even ones which
4734 * have a transaction open against a different journal.
4735 *
4736 * Is this cheating? Not really. Sure, we haven't written the
4737 * inode out, but prune_icache isn't a user-visible syncing function.
4738 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4739 * we start and wait on commits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004740 */
Mingming Cao617ba132006-10-11 01:20:53 -07004741int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004742{
Mingming Cao617ba132006-10-11 01:20:53 -07004743 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004744 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4745 static unsigned int mnt_count;
4746 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004747
4748 might_sleep();
Theodore Ts'o7ff9c072010-11-08 13:51:33 -05004749 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
Mingming Cao617ba132006-10-11 01:20:53 -07004750 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05004751 if (ext4_handle_valid(handle) &&
4752 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004753 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004754 /*
4755 * We need extra buffer credits since we may write into EA block
4756 * with this same handle. If journal_extend fails, then it will
4757 * only result in a minor loss of functionality for that inode.
4758 * If this is felt to be critical, then e2fsck should be run to
4759 * force a large enough s_min_extra_isize.
4760 */
4761 if ((jbd2_journal_extend(handle,
4762 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4763 ret = ext4_expand_extra_isize(inode,
4764 sbi->s_want_extra_isize,
4765 iloc, handle);
4766 if (ret) {
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004767 ext4_set_inode_state(inode,
4768 EXT4_STATE_NO_EXPAND);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004769 if (mnt_count !=
4770 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05004771 ext4_warning(inode->i_sb,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004772 "Unable to expand inode %lu. Delete"
4773 " some EAs or run e2fsck.",
4774 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004775 mnt_count =
4776 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004777 }
4778 }
4779 }
4780 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004781 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07004782 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004783 return err;
4784}
4785
4786/*
Mingming Cao617ba132006-10-11 01:20:53 -07004787 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004788 *
4789 * We're really interested in the case where a file is being extended.
4790 * i_size has been changed by generic_commit_write() and we thus need
4791 * to include the updated inode in the current transaction.
4792 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05004793 * Also, dquot_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004794 * are allocated to the file.
4795 *
4796 * If the inode is marked synchronous, we don't honour that here - doing
4797 * so would cause a commit on atime updates, which we don't bother doing.
4798 * We handle synchronous inodes at the highest possible level.
4799 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04004800void ext4_dirty_inode(struct inode *inode, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004801{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004802 handle_t *handle;
4803
Mingming Cao617ba132006-10-11 01:20:53 -07004804 handle = ext4_journal_start(inode, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004805 if (IS_ERR(handle))
4806 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004807
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004808 ext4_mark_inode_dirty(handle, inode);
4809
Mingming Cao617ba132006-10-11 01:20:53 -07004810 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004811out:
4812 return;
4813}
4814
4815#if 0
4816/*
4817 * Bind an inode's backing buffer_head into this transaction, to prevent
4818 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07004819 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004820 * returns no iloc structure, so the caller needs to repeat the iloc
4821 * lookup to mark the inode dirty later.
4822 */
Mingming Cao617ba132006-10-11 01:20:53 -07004823static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004824{
Mingming Cao617ba132006-10-11 01:20:53 -07004825 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004826
4827 int err = 0;
4828 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07004829 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004830 if (!err) {
4831 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07004832 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004833 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05004834 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05004835 NULL,
Frank Mayhar03901312009-01-07 00:06:22 -05004836 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004837 brelse(iloc.bh);
4838 }
4839 }
Mingming Cao617ba132006-10-11 01:20:53 -07004840 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004841 return err;
4842}
4843#endif
4844
Mingming Cao617ba132006-10-11 01:20:53 -07004845int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004846{
4847 journal_t *journal;
4848 handle_t *handle;
4849 int err;
4850
4851 /*
4852 * We have to be very careful here: changing a data block's
4853 * journaling status dynamically is dangerous. If we write a
4854 * data block to the journal, change the status and then delete
4855 * that block, we risk forgetting to revoke the old log record
4856 * from the journal and so a subsequent replay can corrupt data.
4857 * So, first we make sure that the journal is empty and that
4858 * nobody is changing anything.
4859 */
4860
Mingming Cao617ba132006-10-11 01:20:53 -07004861 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05004862 if (!journal)
4863 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04004864 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004865 return -EROFS;
Yongqiang Yang2aff57b2011-12-28 12:02:13 -05004866 /* We have to allocate physical blocks for delalloc blocks
4867 * before flushing journal. otherwise delalloc blocks can not
4868 * be allocated any more. even more truncate on delalloc blocks
4869 * could trigger BUG by flushing delalloc blocks in journal.
4870 * There is no delalloc block in non-journal data mode.
4871 */
4872 if (val && test_opt(inode->i_sb, DELALLOC)) {
4873 err = ext4_alloc_da_blocks(inode);
4874 if (err < 0)
4875 return err;
4876 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004877
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04004878 /* Wait for all existing dio workers */
4879 ext4_inode_block_unlocked_dio(inode);
4880 inode_dio_wait(inode);
4881
Mingming Caodab291a2006-10-11 01:21:01 -07004882 jbd2_journal_lock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004883
4884 /*
4885 * OK, there are no updates running now, and all cached data is
4886 * synced to disk. We are now in a completely consistent state
4887 * which doesn't have anything in the journal, and we know that
4888 * no filesystem updates are running, so it is safe to modify
4889 * the inode's in-core data-journaling state flag now.
4890 */
4891
4892 if (val)
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004893 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05004894 else {
4895 jbd2_journal_flush(journal);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004896 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05004897 }
Mingming Cao617ba132006-10-11 01:20:53 -07004898 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004899
Mingming Caodab291a2006-10-11 01:21:01 -07004900 jbd2_journal_unlock_updates(journal);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04004901 ext4_inode_resume_unlocked_dio(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004902
4903 /* Finally we can mark the inode as dirty. */
4904
Mingming Cao617ba132006-10-11 01:20:53 -07004905 handle = ext4_journal_start(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004906 if (IS_ERR(handle))
4907 return PTR_ERR(handle);
4908
Mingming Cao617ba132006-10-11 01:20:53 -07004909 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05004910 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07004911 ext4_journal_stop(handle);
4912 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004913
4914 return err;
4915}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004916
4917static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4918{
4919 return !buffer_mapped(bh);
4920}
4921
Nick Pigginc2ec1752009-03-31 15:23:21 -07004922int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004923{
Nick Pigginc2ec1752009-03-31 15:23:21 -07004924 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004925 loff_t size;
4926 unsigned long len;
Jan Kara9ea7df52011-06-24 14:29:41 -04004927 int ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004928 struct file *file = vma->vm_file;
4929 struct inode *inode = file->f_path.dentry->d_inode;
4930 struct address_space *mapping = inode->i_mapping;
Jan Kara9ea7df52011-06-24 14:29:41 -04004931 handle_t *handle;
4932 get_block_t *get_block;
4933 int retries = 0;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004934
Jan Kara8e8ad8a2012-06-12 16:20:38 +02004935 sb_start_pagefault(inode->i_sb);
Theodore Ts'o041bbb6d2012-09-30 23:04:56 -04004936 file_update_time(vma->vm_file);
Jan Kara9ea7df52011-06-24 14:29:41 -04004937 /* Delalloc case is easy... */
4938 if (test_opt(inode->i_sb, DELALLOC) &&
4939 !ext4_should_journal_data(inode) &&
4940 !ext4_nonda_switch(inode->i_sb)) {
4941 do {
4942 ret = __block_page_mkwrite(vma, vmf,
4943 ext4_da_get_block_prep);
4944 } while (ret == -ENOSPC &&
4945 ext4_should_retry_alloc(inode->i_sb, &retries));
4946 goto out_ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004947 }
Darrick J. Wong0e499892011-05-18 13:55:20 -04004948
4949 lock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04004950 size = i_size_read(inode);
4951 /* Page got truncated from under us? */
4952 if (page->mapping != mapping || page_offset(page) > size) {
4953 unlock_page(page);
4954 ret = VM_FAULT_NOPAGE;
4955 goto out;
Darrick J. Wong0e499892011-05-18 13:55:20 -04004956 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004957
4958 if (page->index == size >> PAGE_CACHE_SHIFT)
4959 len = size & ~PAGE_CACHE_MASK;
4960 else
4961 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004962 /*
Jan Kara9ea7df52011-06-24 14:29:41 -04004963 * Return if we have all the buffers mapped. This avoids the need to do
4964 * journal_start/journal_stop which can block and take a long time
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004965 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004966 if (page_has_buffers(page)) {
Tao Maf19d5872012-12-10 14:05:51 -05004967 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
4968 0, len, NULL,
4969 ext4_bh_unmapped)) {
Jan Kara9ea7df52011-06-24 14:29:41 -04004970 /* Wait so that we don't change page under IO */
4971 wait_on_page_writeback(page);
4972 ret = VM_FAULT_LOCKED;
4973 goto out;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004974 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04004975 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04004976 unlock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04004977 /* OK, we need to fill the hole... */
4978 if (ext4_should_dioread_nolock(inode))
4979 get_block = ext4_get_block_write;
4980 else
4981 get_block = ext4_get_block;
4982retry_alloc:
4983 handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode));
4984 if (IS_ERR(handle)) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07004985 ret = VM_FAULT_SIGBUS;
Jan Kara9ea7df52011-06-24 14:29:41 -04004986 goto out;
4987 }
4988 ret = __block_page_mkwrite(vma, vmf, get_block);
4989 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -05004990 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
Jan Kara9ea7df52011-06-24 14:29:41 -04004991 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
4992 unlock_page(page);
4993 ret = VM_FAULT_SIGBUS;
Yongqiang Yangfcbb5512011-10-26 05:00:19 -04004994 ext4_journal_stop(handle);
Jan Kara9ea7df52011-06-24 14:29:41 -04004995 goto out;
4996 }
4997 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
4998 }
4999 ext4_journal_stop(handle);
5000 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5001 goto retry_alloc;
5002out_ret:
5003 ret = block_page_mkwrite_return(ret);
5004out:
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005005 sb_end_pagefault(inode->i_sb);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005006 return ret;
5007}