blob: f51db730da3900bbec98102cecfaf4184252b04f [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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070040#include <linux/aio.h>
Theodore Ts'o00a1a052014-03-30 10:20:01 -040041#include <linux/bitops.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040042
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040043#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070044#include "xattr.h"
45#include "acl.h"
Theodore Ts'o9f125d62011-06-27 19:16:04 -040046#include "truncate.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070047
Theodore Ts'o9bffad12009-06-17 11:48:11 -040048#include <trace/events/ext4.h>
49
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040050#define MPAGE_DA_EXTENT_TAIL 0x01
51
Darrick J. Wong814525f2012-04-29 18:31:10 -040052static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
54{
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 __u16 csum_lo;
57 __u16 csum_hi = 0;
58 __u32 csum;
59
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040060 csum_lo = le16_to_cpu(raw->i_checksum_lo);
Darrick J. Wong814525f2012-04-29 18:31:10 -040061 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040064 csum_hi = le16_to_cpu(raw->i_checksum_hi);
Darrick J. Wong814525f2012-04-29 18:31:10 -040065 raw->i_checksum_hi = 0;
66 }
67
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
70
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040071 raw->i_checksum_lo = cpu_to_le16(csum_lo);
Darrick J. Wong814525f2012-04-29 18:31:10 -040072 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040074 raw->i_checksum_hi = cpu_to_le16(csum_hi);
Darrick J. Wong814525f2012-04-29 18:31:10 -040075
76 return csum;
77}
78
79static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
81{
82 __u32 provided, calculated;
83
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 return 1;
89
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
95 else
96 calculated &= 0xFFFF;
97
98 return provided == calculated;
99}
100
101static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
103{
104 __u32 csum;
105
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 return;
111
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117}
118
Jan Kara678aaf42008-07-11 19:27:31 -0400119static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 loff_t new_size)
121{
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500122 trace_ext4_begin_ordered_truncate(inode, new_size);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500123 /*
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
128 */
129 if (!EXT4_I(inode)->jinode)
130 return 0;
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
133 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -0400134}
135
Lukas Czernerd47992f2013-05-21 23:17:23 -0400136static void ext4_invalidatepage(struct page *page, unsigned int offset,
137 unsigned int length);
Theodore Ts'ocb20d512010-10-27 21:30:09 -0400138static int __ext4_journalled_writepage(struct page *page, unsigned int len);
139static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
Jan Karafffb2732013-06-04 13:01:11 -0400140static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
141 int pextents);
Alex Tomas64769242008-07-11 19:27:31 -0400142
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700143/*
144 * Test whether an inode is a fast symlink.
145 */
Mingming Cao617ba132006-10-11 01:20:53 -0700146static int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700147{
Yongqiang Yang65eddb52014-01-06 14:06:18 -0500148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700150
151 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
152}
153
154/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
157 * this transaction.
158 */
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500159int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
Jan Kara487caee2009-08-17 22:17:20 -0400160 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700161{
Jan Kara487caee2009-08-17 22:17:20 -0400162 int ret;
163
164 /*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
Jan Kara487caee2009-08-17 22:17:20 -0400166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
169 */
Frank Mayhar03901312009-01-07 00:06:22 -0500170 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700171 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400172 up_write(&EXT4_I(inode)->i_data_sem);
Amir Goldstein8e8eaab2011-02-27 23:32:12 -0500173 ret = ext4_journal_restart(handle, nblocks);
Jan Kara487caee2009-08-17 22:17:20 -0400174 down_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500175 ext4_discard_preallocations(inode);
Jan Kara487caee2009-08-17 22:17:20 -0400176
177 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700178}
179
180/*
181 * Called at the last iput() if i_nlink is zero.
182 */
Al Viro0930fcc2010-06-07 13:16:22 -0400183void ext4_evict_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700184{
185 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400186 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700187
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500188 trace_ext4_evict_inode(inode);
Jiaying Zhang2581fdc2011-08-13 12:17:13 -0400189
Al Viro0930fcc2010-06-07 13:16:22 -0400190 if (inode->i_nlink) {
Jan Kara2d859db2011-07-26 09:07:11 -0400191 /*
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
205 *
206 * Note that directories do not have this problem because they
207 * don't use page cache.
208 */
209 if (ext4_should_journal_data(inode) &&
Theodore Ts'o2b405bf2013-03-20 09:42:11 -0400210 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
211 inode->i_ino != EXT4_JOURNAL_INO) {
Jan Kara2d859db2011-07-26 09:07:11 -0400212 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
213 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
214
Theodore Ts'od76a3a772013-04-03 22:02:52 -0400215 jbd2_complete_transaction(journal, commit_tid);
Jan Kara2d859db2011-07-26 09:07:11 -0400216 filemap_write_and_wait(&inode->i_data);
217 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700218 truncate_inode_pages_final(&inode->i_data);
Jan Kara5dc23bd2013-06-04 14:46:12 -0400219
220 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
Al Viro0930fcc2010-06-07 13:16:22 -0400221 goto no_delete;
222 }
223
Christoph Hellwig907f4552010-03-03 09:05:06 -0500224 if (!is_bad_inode(inode))
Christoph Hellwig871a2932010-03-03 09:05:07 -0500225 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500226
Jan Kara678aaf42008-07-11 19:27:31 -0400227 if (ext4_should_order_data(inode))
228 ext4_begin_ordered_truncate(inode, 0);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700229 truncate_inode_pages_final(&inode->i_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700230
Jan Kara5dc23bd2013-06-04 14:46:12 -0400231 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232 if (is_bad_inode(inode))
233 goto no_delete;
234
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200235 /*
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
238 */
239 sb_start_intwrite(inode->i_sb);
Theodore Ts'o9924a922013-02-08 21:59:22 -0500240 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
241 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)) {
Theodore Ts'o8de5c322013-02-14 15:11:41 -0500349 ext4_warning(inode->i_sb, "%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)) {
Theodore Ts'o01a523e2013-02-14 15:51:58 -0500358 ext4_warning(inode->i_sb, "ino %lu, allocated %d "
359 "with only %d reserved metadata blocks "
360 "(releasing %d blocks with reserved %d data blocks)",
361 inode->i_ino, ei->i_allocated_meta_blocks,
362 ei->i_reserved_meta_blocks, used,
363 ei->i_reserved_data_blocks);
Brian Foster97795d22012-07-22 23:59:40 -0400364 WARN_ON(1);
365 ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks;
366 }
367
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500368 /* Update per-inode reservations */
369 ei->i_reserved_data_blocks -= used;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500370 ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks;
Theodore Ts'o57042652011-09-09 18:56:51 -0400371 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400372 used + ei->i_allocated_meta_blocks);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500373 ei->i_allocated_meta_blocks = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500374
375 if (ei->i_reserved_data_blocks == 0) {
376 /*
377 * We can release all of the reserved metadata blocks
378 * only when we have written all of the delayed
379 * allocation blocks.
380 */
Theodore Ts'o57042652011-09-09 18:56:51 -0400381 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400382 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -0500383 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500384 ei->i_da_metadata_calc_len = 0;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500385 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400386 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +0100387
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400388 /* Update quota subsystem for data blocks */
389 if (quota_claim)
Aditya Kali7b415bf2011-09-09 19:04:51 -0400390 dquot_claim_block(inode, EXT4_C2B(sbi, used));
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400391 else {
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500392 /*
393 * We did fallocate with an offset that is already delayed
394 * allocated. So on delayed allocated writeback we should
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400395 * not re-claim the quota for fallocated blocks.
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500396 */
Aditya Kali7b415bf2011-09-09 19:04:51 -0400397 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500398 }
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400399
400 /*
401 * If we have done all the pending block allocations and if
402 * there aren't any writers on the inode, we can discard the
403 * inode's preallocations.
404 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500405 if ((ei->i_reserved_data_blocks == 0) &&
406 (atomic_read(&inode->i_writecount) == 0))
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400407 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400408}
409
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400410static int __check_block_validity(struct inode *inode, const char *func,
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400411 unsigned int line,
412 struct ext4_map_blocks *map)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400413{
Theodore Ts'o24676da2010-05-16 21:00:00 -0400414 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
415 map->m_len)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400416 ext4_error_inode(inode, func, line, map->m_pblk,
417 "lblock %lu mapped to illegal pblock "
418 "(length %d)", (unsigned long) map->m_lblk,
419 map->m_len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400420 return -EIO;
421 }
422 return 0;
423}
424
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400425#define check_block_validity(inode, map) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400426 __check_block_validity((inode), __func__, __LINE__, (map))
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400427
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400428#ifdef ES_AGGRESSIVE_TEST
429static void ext4_map_blocks_es_recheck(handle_t *handle,
430 struct inode *inode,
431 struct ext4_map_blocks *es_map,
432 struct ext4_map_blocks *map,
433 int flags)
434{
435 int retval;
436
437 map->m_flags = 0;
438 /*
439 * There is a race window that the result is not the same.
440 * e.g. xfstests #223 when dioread_nolock enables. The reason
441 * is that we lookup a block mapping in extent status tree with
442 * out taking i_data_sem. So at the time the unwritten extent
443 * could be converted.
444 */
445 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
446 down_read((&EXT4_I(inode)->i_data_sem));
447 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
448 retval = ext4_ext_map_blocks(handle, inode, map, flags &
449 EXT4_GET_BLOCKS_KEEP_SIZE);
450 } else {
451 retval = ext4_ind_map_blocks(handle, inode, map, flags &
452 EXT4_GET_BLOCKS_KEEP_SIZE);
453 }
454 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
455 up_read((&EXT4_I(inode)->i_data_sem));
456 /*
457 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
458 * because it shouldn't be marked in es_map->m_flags.
459 */
460 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
461
462 /*
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
465 */
466 if (es_map->m_lblk != map->m_lblk ||
467 es_map->m_flags != map->m_flags ||
468 es_map->m_pblk != map->m_pblk) {
Theodore Ts'obdafe422013-07-13 00:40:31 -0400469 printk("ES cache assertion failed for inode: %lu "
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode->i_ino, es_map->m_lblk, es_map->m_len,
473 es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 map->m_len, map->m_pblk, map->m_flags,
475 retval, flags);
476 }
477}
478#endif /* ES_AGGRESSIVE_TEST */
479
Theodore Ts'o55138e02009-09-29 13:31:31 -0400480/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400481 * The ext4_map_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400482 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -0500483 *
Mingming Caof5ab0d12008-02-25 15:29:55 -0500484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
486 * mapped.
487 *
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -0500490 * based files
491 *
492 * On success, it returns the number of blocks being mapped or allocate.
493 * if create==0 and the blocks are pre-allocated and uninitialized block,
494 * the result buffer head is unmapped. If the create ==1, it will make sure
495 * the buffer head is mapped.
496 *
497 * It returns 0 if plain look up failed (blocks have not been allocated), in
Tao Madf3ab172011-10-08 15:53:49 -0400498 * that case, buffer head is unmapped
Mingming Caof5ab0d12008-02-25 15:29:55 -0500499 *
500 * It returns the error in case of allocation failure.
501 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400502int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 struct ext4_map_blocks *map, int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500504{
Zheng Liud100eef2013-02-18 00:29:59 -0500505 struct extent_status es;
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500506 int retval;
Lukas Czernerb8a86842014-03-18 18:05:35 -0400507 int ret = 0;
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400508#ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map;
510
511 memcpy(&orig_map, map, sizeof(*map));
512#endif
Mingming Caof5ab0d12008-02-25 15:29:55 -0500513
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400514 map->m_flags = 0;
515 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516 "logical block %lu\n", inode->i_ino, flags, map->m_len,
517 (unsigned long) map->m_lblk);
Zheng Liud100eef2013-02-18 00:29:59 -0500518
Theodore Ts'oe861b5e2014-02-20 12:54:05 -0500519 /*
520 * ext4_map_blocks returns an int, and m_len is an unsigned int
521 */
522 if (unlikely(map->m_len > INT_MAX))
523 map->m_len = INT_MAX;
524
Kazuya Mio4adb6ab2014-04-07 10:53:28 -0400525 /* We can handle the block number less than EXT_MAX_BLOCKS */
526 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
527 return -EIO;
528
Zheng Liud100eef2013-02-18 00:29:59 -0500529 /* Lookup extent status tree firstly */
530 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
Theodore Ts'o63b99962013-07-16 10:28:47 -0400531 ext4_es_lru_add(inode);
Zheng Liud100eef2013-02-18 00:29:59 -0500532 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
533 map->m_pblk = ext4_es_pblock(&es) +
534 map->m_lblk - es.es_lblk;
535 map->m_flags |= ext4_es_is_written(&es) ?
536 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
537 retval = es.es_len - (map->m_lblk - es.es_lblk);
538 if (retval > map->m_len)
539 retval = map->m_len;
540 map->m_len = retval;
541 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
542 retval = 0;
543 } else {
544 BUG_ON(1);
545 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400546#ifdef ES_AGGRESSIVE_TEST
547 ext4_map_blocks_es_recheck(handle, inode, map,
548 &orig_map, flags);
549#endif
Zheng Liud100eef2013-02-18 00:29:59 -0500550 goto found;
551 }
552
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500553 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -0400554 * Try to see if we can get the block without requesting a new
555 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500556 */
Zheng Liu729f52c2012-07-09 16:29:29 -0400557 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
558 down_read((&EXT4_I(inode)->i_data_sem));
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400559 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400560 retval = ext4_ext_map_blocks(handle, inode, map, flags &
561 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500562 } else {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400563 retval = ext4_ind_map_blocks(handle, inode, map, flags &
564 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500565 }
Zheng Liuf7fec032013-02-18 00:28:47 -0500566 if (retval > 0) {
Theodore Ts'o3be78c72013-08-16 21:22:41 -0400567 unsigned int status;
Zheng Liuf7fec032013-02-18 00:28:47 -0500568
Zheng Liu44fb851d2013-07-29 12:51:42 -0400569 if (unlikely(retval != map->m_len)) {
570 ext4_warning(inode->i_sb,
571 "ES len assertion failed for inode "
572 "%lu: retval %d != map->m_len %d",
573 inode->i_ino, retval, map->m_len);
574 WARN_ON(1);
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400575 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400576
Zheng Liuf7fec032013-02-18 00:28:47 -0500577 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
578 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
579 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
580 ext4_find_delalloc_range(inode, map->m_lblk,
581 map->m_lblk + map->m_len - 1))
582 status |= EXTENT_STATUS_DELAYED;
583 ret = ext4_es_insert_extent(inode, map->m_lblk,
584 map->m_len, map->m_pblk, status);
585 if (ret < 0)
586 retval = ret;
587 }
Zheng Liu729f52c2012-07-09 16:29:29 -0400588 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
589 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -0500590
Zheng Liud100eef2013-02-18 00:29:59 -0500591found:
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400592 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Lukas Czernerb8a86842014-03-18 18:05:35 -0400593 ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400594 if (ret != 0)
595 return ret;
596 }
597
Mingming Caof5ab0d12008-02-25 15:29:55 -0500598 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400599 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500600 return retval;
601
602 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500603 * Returns if the blocks have already allocated
604 *
605 * Note that if blocks have been preallocated
Tao Madf3ab172011-10-08 15:53:49 -0400606 * ext4_ext_get_block() returns the create = 0
Mingming Caof5ab0d12008-02-25 15:29:55 -0500607 * with buffer head unmapped.
608 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400609 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
Lukas Czernerb8a86842014-03-18 18:05:35 -0400610 /*
611 * If we need to convert extent to unwritten
612 * we continue and do the actual work in
613 * ext4_ext_map_blocks()
614 */
615 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
616 return retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -0500617
618 /*
Zheng Liua25a4e12013-02-18 00:28:04 -0500619 * Here we clear m_flags because after allocating an new extent,
620 * it will be set again.
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400621 */
Zheng Liua25a4e12013-02-18 00:28:04 -0500622 map->m_flags &= ~EXT4_MAP_FLAGS;
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400623
624 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500625 * New blocks allocate and/or writing to uninitialized extent
626 * will possibly result in updating i_data, so we take
627 * the write lock of i_data_sem, and call get_blocks()
628 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500629 */
630 down_write((&EXT4_I(inode)->i_data_sem));
Mingming Caod2a17632008-07-14 17:52:37 -0400631
632 /*
633 * if the caller is from delayed allocation writeout path
634 * we have already reserved fs blocks for allocation
635 * let the underlying get_block() function know to
636 * avoid double accounting
637 */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400638 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Theodore Ts'of2321092011-01-10 12:12:36 -0500639 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500640 /*
641 * We need to check for EXT4 here because migrate
642 * could have changed the inode type in between
643 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400644 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400645 retval = ext4_ext_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500646 } else {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400647 retval = ext4_ind_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400648
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400649 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400650 /*
651 * We allocated new blocks which will result in
652 * i_data's format changing. Force the migrate
653 * to fail by clearing migrate flags
654 */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -0500655 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400656 }
Mingming Caod2a17632008-07-14 17:52:37 -0400657
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500658 /*
659 * Update reserved blocks/metadata blocks after successful
660 * block allocation which had been deferred till now. We don't
661 * support fallocate for non extent files. So we can update
662 * reserve space here.
663 */
664 if ((retval > 0) &&
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -0500665 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500666 ext4_da_update_reserve_space(inode, retval, 1);
667 }
Zheng Liuf7fec032013-02-18 00:28:47 -0500668 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
Theodore Ts'of2321092011-01-10 12:12:36 -0500669 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -0400670
Zheng Liuf7fec032013-02-18 00:28:47 -0500671 if (retval > 0) {
Theodore Ts'o3be78c72013-08-16 21:22:41 -0400672 unsigned int status;
Zheng Liuf7fec032013-02-18 00:28:47 -0500673
Zheng Liu44fb851d2013-07-29 12:51:42 -0400674 if (unlikely(retval != map->m_len)) {
675 ext4_warning(inode->i_sb,
676 "ES len assertion failed for inode "
677 "%lu: retval %d != map->m_len %d",
678 inode->i_ino, retval, map->m_len);
679 WARN_ON(1);
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400680 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400681
Zheng Liuadb23552013-03-10 21:13:05 -0400682 /*
683 * If the extent has been zeroed out, we don't need to update
684 * extent status tree.
685 */
686 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
687 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
688 if (ext4_es_is_written(&es))
689 goto has_zeroout;
690 }
Zheng Liuf7fec032013-02-18 00:28:47 -0500691 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
692 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
693 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
694 ext4_find_delalloc_range(inode, map->m_lblk,
695 map->m_lblk + map->m_len - 1))
696 status |= EXTENT_STATUS_DELAYED;
697 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
698 map->m_pblk, status);
699 if (ret < 0)
700 retval = ret;
Aditya Kali5356f262011-09-09 19:20:51 -0400701 }
702
Zheng Liuadb23552013-03-10 21:13:05 -0400703has_zeroout:
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500704 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400705 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Lukas Czernerb8a86842014-03-18 18:05:35 -0400706 ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400707 if (ret != 0)
708 return ret;
709 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500710 return retval;
711}
712
Mingming Caof3bd1f32008-08-19 22:16:03 -0400713/* Maximum number of blocks we map for direct IO at once. */
714#define DIO_MAX_BLOCKS 4096
715
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400716static int _ext4_get_block(struct inode *inode, sector_t iblock,
717 struct buffer_head *bh, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700718{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800719 handle_t *handle = ext4_journal_current_handle();
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400720 struct ext4_map_blocks map;
Jan Kara7fb54092008-02-10 01:08:38 -0500721 int ret = 0, started = 0;
Mingming Caof3bd1f32008-08-19 22:16:03 -0400722 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723
Tao Ma46c7f252012-12-10 14:04:52 -0500724 if (ext4_has_inline_data(inode))
725 return -ERANGE;
726
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400727 map.m_lblk = iblock;
728 map.m_len = bh->b_size >> inode->i_blkbits;
729
Anatol Pomozov8b0f1652012-11-08 15:07:16 -0500730 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
Jan Kara7fb54092008-02-10 01:08:38 -0500731 /* Direct IO write... */
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400732 if (map.m_len > DIO_MAX_BLOCKS)
733 map.m_len = DIO_MAX_BLOCKS;
734 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
Theodore Ts'o9924a922013-02-08 21:59:22 -0500735 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
736 dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -0500737 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700738 ret = PTR_ERR(handle);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400739 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700740 }
Jan Kara7fb54092008-02-10 01:08:38 -0500741 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700742 }
743
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400744 ret = ext4_map_blocks(handle, inode, &map, flags);
Jan Kara7fb54092008-02-10 01:08:38 -0500745 if (ret > 0) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200746 ext4_io_end_t *io_end = ext4_inode_aio(inode);
747
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400748 map_bh(bh, inode->i_sb, map.m_pblk);
749 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200750 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
751 set_buffer_defer_completion(bh);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400752 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
Jan Kara7fb54092008-02-10 01:08:38 -0500753 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700754 }
Jan Kara7fb54092008-02-10 01:08:38 -0500755 if (started)
756 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700757 return ret;
758}
759
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400760int ext4_get_block(struct inode *inode, sector_t iblock,
761 struct buffer_head *bh, int create)
762{
763 return _ext4_get_block(inode, iblock, bh,
764 create ? EXT4_GET_BLOCKS_CREATE : 0);
765}
766
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700767/*
768 * `handle' can be NULL if create is zero
769 */
Mingming Cao617ba132006-10-11 01:20:53 -0700770struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500771 ext4_lblk_t block, int create, int *errp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700772{
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400773 struct ext4_map_blocks map;
774 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700775 int fatal = 0, err;
776
777 J_ASSERT(handle != NULL || create == 0);
778
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400779 map.m_lblk = block;
780 map.m_len = 1;
781 err = ext4_map_blocks(handle, inode, &map,
782 create ? EXT4_GET_BLOCKS_CREATE : 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700783
Carlos Maiolino90b0a972012-09-17 23:39:12 -0400784 /* ensure we send some value back into *errp */
785 *errp = 0;
786
Theodore Ts'o0f70b402013-02-15 03:35:57 -0500787 if (create && err == 0)
788 err = -ENOSPC; /* should never happen */
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400789 if (err < 0)
790 *errp = err;
791 if (err <= 0)
792 return NULL;
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400793
794 bh = sb_getblk(inode->i_sb, map.m_pblk);
Wang Shilongaebf0242013-01-12 16:28:47 -0500795 if (unlikely(!bh)) {
Theodore Ts'o860d21e2013-01-12 16:19:36 -0500796 *errp = -ENOMEM;
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400797 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700798 }
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400799 if (map.m_flags & EXT4_MAP_NEW) {
800 J_ASSERT(create != 0);
801 J_ASSERT(handle != NULL);
802
803 /*
804 * Now that we do not always journal data, we should
805 * keep in mind whether this should always journal the
806 * new buffer as metadata. For now, regular file
807 * writes use ext4_get_block instead, so it's not a
808 * problem.
809 */
810 lock_buffer(bh);
811 BUFFER_TRACE(bh, "call get_create_access");
812 fatal = ext4_journal_get_create_access(handle, bh);
813 if (!fatal && !buffer_uptodate(bh)) {
814 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
815 set_buffer_uptodate(bh);
816 }
817 unlock_buffer(bh);
818 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
819 err = ext4_handle_dirty_metadata(handle, inode, bh);
820 if (!fatal)
821 fatal = err;
822 } else {
823 BUFFER_TRACE(bh, "not a new buffer");
824 }
825 if (fatal) {
826 *errp = fatal;
827 brelse(bh);
828 bh = NULL;
829 }
830 return bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831}
832
Mingming Cao617ba132006-10-11 01:20:53 -0700833struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500834 ext4_lblk_t block, int create, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700835{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400836 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837
Mingming Cao617ba132006-10-11 01:20:53 -0700838 bh = ext4_getblk(handle, inode, block, create, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700839 if (!bh)
840 return bh;
841 if (buffer_uptodate(bh))
842 return bh;
Christoph Hellwig65299a32011-08-23 14:50:29 +0200843 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700844 wait_on_buffer(bh);
845 if (buffer_uptodate(bh))
846 return bh;
847 put_bh(bh);
848 *err = -EIO;
849 return NULL;
850}
851
Tao Maf19d5872012-12-10 14:05:51 -0500852int ext4_walk_page_buffers(handle_t *handle,
853 struct buffer_head *head,
854 unsigned from,
855 unsigned to,
856 int *partial,
857 int (*fn)(handle_t *handle,
858 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700859{
860 struct buffer_head *bh;
861 unsigned block_start, block_end;
862 unsigned blocksize = head->b_size;
863 int err, ret = 0;
864 struct buffer_head *next;
865
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400866 for (bh = head, block_start = 0;
867 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400868 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 next = bh->b_this_page;
870 block_end = block_start + blocksize;
871 if (block_end <= from || block_start >= to) {
872 if (partial && !buffer_uptodate(bh))
873 *partial = 1;
874 continue;
875 }
876 err = (*fn)(handle, bh);
877 if (!ret)
878 ret = err;
879 }
880 return ret;
881}
882
883/*
884 * To preserve ordering, it is essential that the hole instantiation and
885 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -0700886 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -0700887 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888 * prepare_write() is the right place.
889 *
Jan Kara36ade452013-01-28 09:30:52 -0500890 * Also, this function can nest inside ext4_writepage(). In that case, we
891 * *know* that ext4_writepage() has generated enough buffer credits to do the
892 * whole page. So we won't block on the journal in that case, which is good,
893 * because the caller may be PF_MEMALLOC.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700894 *
Mingming Cao617ba132006-10-11 01:20:53 -0700895 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896 * quota file writes. If we were to commit the transaction while thus
897 * reentered, there can be a deadlock - we would be holding a quota
898 * lock, and the commit would never complete if another thread had a
899 * transaction open and was blocking on the quota lock - a ranking
900 * violation.
901 *
Mingming Caodab291a2006-10-11 01:21:01 -0700902 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700903 * will _not_ run commit under these circumstances because handle->h_ref
904 * is elevated. We'll still have enough credits for the tiny quotafile
905 * write.
906 */
Tao Maf19d5872012-12-10 14:05:51 -0500907int do_journal_get_write_access(handle_t *handle,
908 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909{
Jan Kara56d35a42010-08-05 14:41:42 -0400910 int dirty = buffer_dirty(bh);
911 int ret;
912
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700913 if (!buffer_mapped(bh) || buffer_freed(bh))
914 return 0;
Jan Kara56d35a42010-08-05 14:41:42 -0400915 /*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200916 * __block_write_begin() could have dirtied some buffers. Clean
Jan Kara56d35a42010-08-05 14:41:42 -0400917 * the dirty bit as jbd2_journal_get_write_access() could complain
918 * otherwise about fs integrity issues. Setting of the dirty bit
Christoph Hellwigebdec242010-10-06 10:47:23 +0200919 * by __block_write_begin() isn't a real problem here as we clear
Jan Kara56d35a42010-08-05 14:41:42 -0400920 * the bit before releasing a page lock and thus writeback cannot
921 * ever write the buffer.
922 */
923 if (dirty)
924 clear_buffer_dirty(bh);
925 ret = ext4_journal_get_write_access(handle, bh);
926 if (!ret && dirty)
927 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
928 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700929}
930
Anatol Pomozov8b0f1652012-11-08 15:07:16 -0500931static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
932 struct buffer_head *bh_result, int create);
Nick Pigginbfc1af62007-10-16 01:25:05 -0700933static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400934 loff_t pos, unsigned len, unsigned flags,
935 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700936{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400937 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400938 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700939 handle_t *handle;
940 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400941 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400942 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400943 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700944
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400945 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400946 /*
947 * Reserve one block more for addition to orphan list in case
948 * we allocate blocks but write fails for some reason
949 */
950 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400951 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400952 from = pos & (PAGE_CACHE_SIZE - 1);
953 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700954
Tao Maf19d5872012-12-10 14:05:51 -0500955 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
956 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
957 flags, pagep);
958 if (ret < 0)
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500959 return ret;
960 if (ret == 1)
961 return 0;
Tao Maf19d5872012-12-10 14:05:51 -0500962 }
963
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500964 /*
965 * grab_cache_page_write_begin() can take a long time if the
966 * system is thrashing due to memory pressure, or if the page
967 * is being written back. So grab it first before we start
968 * the transaction handle. This also allows us to allocate
969 * the page (if needed) without using GFP_NOFS.
970 */
971retry_grab:
Nick Piggin54566b22009-01-04 12:00:53 -0800972 page = grab_cache_page_write_begin(mapping, index, flags);
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500973 if (!page)
974 return -ENOMEM;
975 unlock_page(page);
976
977retry_journal:
Theodore Ts'o9924a922013-02-08 21:59:22 -0500978 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700979 if (IS_ERR(handle)) {
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500980 page_cache_release(page);
981 return PTR_ERR(handle);
Jan Karacf108bc2008-07-11 19:27:31 -0400982 }
Tao Maf19d5872012-12-10 14:05:51 -0500983
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500984 lock_page(page);
985 if (page->mapping != mapping) {
986 /* The page got truncated from under us */
987 unlock_page(page);
988 page_cache_release(page);
Jan Karacf108bc2008-07-11 19:27:31 -0400989 ext4_journal_stop(handle);
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500990 goto retry_grab;
Jan Karacf108bc2008-07-11 19:27:31 -0400991 }
Dmitry Monakhov7afe5aa2013-08-28 14:30:47 -0400992 /* In case writeback began while the page was unlocked */
993 wait_for_stable_page(page);
Jan Karacf108bc2008-07-11 19:27:31 -0400994
Jiaying Zhang744692d2010-03-04 16:14:02 -0500995 if (ext4_should_dioread_nolock(inode))
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200996 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
Jiaying Zhang744692d2010-03-04 16:14:02 -0500997 else
Christoph Hellwig6e1db882010-06-04 11:29:57 +0200998 ret = __block_write_begin(page, pos, len, ext4_get_block);
Nick Pigginbfc1af62007-10-16 01:25:05 -0700999
1000 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -05001001 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1002 from, to, NULL,
1003 do_journal_get_write_access);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001004 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001005
1006 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001007 unlock_page(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001008 /*
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001009 * __block_write_begin may have instantiated a few blocks
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001010 * outside i_size. Trim these off again. Don't need
1011 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001012 *
1013 * Add inode to orphan list in case we crash before
1014 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001015 */
Jan Karaffacfa72009-07-13 16:22:22 -04001016 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001017 ext4_orphan_add(handle, inode);
1018
1019 ext4_journal_stop(handle);
1020 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001021 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001022 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001023 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001024 * still be on the orphan list; we need to
1025 * make sure the inode is removed from the
1026 * orphan list in that case.
1027 */
1028 if (inode->i_nlink)
1029 ext4_orphan_del(NULL, inode);
1030 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001031
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001032 if (ret == -ENOSPC &&
1033 ext4_should_retry_alloc(inode->i_sb, &retries))
1034 goto retry_journal;
1035 page_cache_release(page);
1036 return ret;
1037 }
1038 *pagep = page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 return ret;
1040}
1041
Nick Pigginbfc1af62007-10-16 01:25:05 -07001042/* For write_end() in data=journal mode */
1043static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001044{
Theodore Ts'o13fca322013-04-21 16:45:54 -04001045 int ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001046 if (!buffer_mapped(bh) || buffer_freed(bh))
1047 return 0;
1048 set_buffer_uptodate(bh);
Theodore Ts'o13fca322013-04-21 16:45:54 -04001049 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1050 clear_buffer_meta(bh);
1051 clear_buffer_prio(bh);
1052 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053}
1054
Zheng Liueed43332013-04-03 12:41:17 -04001055/*
1056 * We need to pick up the new inode size which generic_commit_write gave us
1057 * `file' can be NULL - eg, when called from page_symlink().
1058 *
1059 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1060 * buffers are managed internally.
1061 */
1062static int ext4_write_end(struct file *file,
1063 struct address_space *mapping,
1064 loff_t pos, unsigned len, unsigned copied,
1065 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001066{
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001067 handle_t *handle = ext4_journal_current_handle();
Zheng Liueed43332013-04-03 12:41:17 -04001068 struct inode *inode = mapping->host;
1069 int ret = 0, ret2;
1070 int i_size_changed = 0;
1071
1072 trace_ext4_write_end(inode, pos, len, copied);
1073 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1074 ret = ext4_jbd2_file_inode(handle, inode);
1075 if (ret) {
1076 unlock_page(page);
1077 page_cache_release(page);
1078 goto errout;
1079 }
1080 }
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001081
Theodore Ts'o42c832d2013-07-01 08:12:39 -04001082 if (ext4_has_inline_data(inode)) {
1083 ret = ext4_write_inline_data_end(inode, pos, len,
1084 copied, page);
1085 if (ret < 0)
1086 goto errout;
1087 copied = ret;
1088 } else
Tao Maf19d5872012-12-10 14:05:51 -05001089 copied = block_write_end(file, mapping, pos,
1090 len, copied, page, fsdata);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001091
1092 /*
1093 * No need to use i_size_read() here, the i_size
Zheng Liueed43332013-04-03 12:41:17 -04001094 * cannot change under us because we hole i_mutex.
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001095 *
1096 * But it's important to update i_size while still holding page lock:
1097 * page writeout could otherwise come in and zero beyond i_size.
1098 */
1099 if (pos + copied > inode->i_size) {
1100 i_size_write(inode, pos + copied);
1101 i_size_changed = 1;
1102 }
1103
Zheng Liueed43332013-04-03 12:41:17 -04001104 if (pos + copied > EXT4_I(inode)->i_disksize) {
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001105 /* We need to mark inode dirty even if
1106 * new_i_size is less that inode->i_size
Zheng Liueed43332013-04-03 12:41:17 -04001107 * but greater than i_disksize. (hint delalloc)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001108 */
1109 ext4_update_i_disksize(inode, (pos + copied));
1110 i_size_changed = 1;
1111 }
1112 unlock_page(page);
1113 page_cache_release(page);
1114
1115 /*
1116 * Don't mark the inode dirty under page lock. First, it unnecessarily
1117 * makes the holding time of page lock longer. Second, it forces lock
1118 * ordering of page lock and transaction start for journaling
1119 * filesystems.
1120 */
1121 if (i_size_changed)
1122 ext4_mark_inode_dirty(handle, inode);
1123
Jan Karaffacfa72009-07-13 16:22:22 -04001124 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001125 /* if we have allocated more blocks and copied
1126 * less. We will have blocks allocated outside
1127 * inode->i_size. So truncate them
1128 */
1129 ext4_orphan_add(handle, inode);
Theodore Ts'o74d553a2013-04-03 12:39:17 -04001130errout:
Mingming Cao617ba132006-10-11 01:20:53 -07001131 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001132 if (!ret)
1133 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001134
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001135 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001136 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001137 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001138 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001139 * on the orphan list; we need to make sure the inode
1140 * is removed from the orphan list in that case.
1141 */
1142 if (inode->i_nlink)
1143 ext4_orphan_del(NULL, inode);
1144 }
1145
Nick Pigginbfc1af62007-10-16 01:25:05 -07001146 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001147}
1148
Nick Pigginbfc1af62007-10-16 01:25:05 -07001149static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001150 struct address_space *mapping,
1151 loff_t pos, unsigned len, unsigned copied,
1152 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001153{
Mingming Cao617ba132006-10-11 01:20:53 -07001154 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001155 struct inode *inode = mapping->host;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001156 int ret = 0, ret2;
1157 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001158 unsigned from, to;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001159 loff_t new_i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001160
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001161 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001162 from = pos & (PAGE_CACHE_SIZE - 1);
1163 to = from + len;
1164
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001165 BUG_ON(!ext4_handle_valid(handle));
1166
Tao Ma3fdcfb62012-12-10 14:05:57 -05001167 if (ext4_has_inline_data(inode))
1168 copied = ext4_write_inline_data_end(inode, pos, len,
1169 copied, page);
1170 else {
1171 if (copied < len) {
1172 if (!PageUptodate(page))
1173 copied = 0;
1174 page_zero_new_buffers(page, from+copied, to);
1175 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176
Tao Ma3fdcfb62012-12-10 14:05:57 -05001177 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1178 to, &partial, write_end_fn);
1179 if (!partial)
1180 SetPageUptodate(page);
1181 }
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001182 new_i_size = pos + copied;
1183 if (new_i_size > inode->i_size)
Nick Pigginbfc1af62007-10-16 01:25:05 -07001184 i_size_write(inode, pos+copied);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001185 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Jan Kara2d859db2011-07-26 09:07:11 -04001186 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04001187 if (new_i_size > EXT4_I(inode)->i_disksize) {
1188 ext4_update_i_disksize(inode, new_i_size);
Mingming Cao617ba132006-10-11 01:20:53 -07001189 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001190 if (!ret)
1191 ret = ret2;
1192 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001193
Jan Karacf108bc2008-07-11 19:27:31 -04001194 unlock_page(page);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001195 page_cache_release(page);
Jan Karaffacfa72009-07-13 16:22:22 -04001196 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001197 /* if we have allocated more blocks and copied
1198 * less. We will have blocks allocated outside
1199 * inode->i_size. So truncate them
1200 */
1201 ext4_orphan_add(handle, inode);
1202
Mingming Cao617ba132006-10-11 01:20:53 -07001203 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001204 if (!ret)
1205 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001206 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001207 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001208 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001209 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001210 * on the orphan list; we need to make sure the inode
1211 * is removed from the orphan list in that case.
1212 */
1213 if (inode->i_nlink)
1214 ext4_orphan_del(NULL, inode);
1215 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001216
1217 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001218}
Mingming Caod2a17632008-07-14 17:52:37 -04001219
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001220/*
Lukas Czerner386ad672013-03-10 22:50:00 -04001221 * Reserve a metadata for a single block located at lblock
1222 */
1223static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock)
1224{
Lukas Czerner386ad672013-03-10 22:50:00 -04001225 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1226 struct ext4_inode_info *ei = EXT4_I(inode);
1227 unsigned int md_needed;
1228 ext4_lblk_t save_last_lblock;
1229 int save_len;
1230
1231 /*
1232 * recalculate the amount of metadata blocks to reserve
1233 * in order to allocate nrblocks
1234 * worse case is one extent per block
1235 */
Lukas Czerner386ad672013-03-10 22:50:00 -04001236 spin_lock(&ei->i_block_reservation_lock);
1237 /*
1238 * ext4_calc_metadata_amount() has side effects, which we have
1239 * to be prepared undo if we fail to claim space.
1240 */
1241 save_len = ei->i_da_metadata_calc_len;
1242 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1243 md_needed = EXT4_NUM_B2C(sbi,
1244 ext4_calc_metadata_amount(inode, lblock));
1245 trace_ext4_da_reserve_space(inode, md_needed);
1246
1247 /*
1248 * We do still charge estimated metadata to the sb though;
1249 * we cannot afford to run out of free blocks.
1250 */
1251 if (ext4_claim_free_clusters(sbi, md_needed, 0)) {
1252 ei->i_da_metadata_calc_len = save_len;
1253 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1254 spin_unlock(&ei->i_block_reservation_lock);
Lukas Czerner386ad672013-03-10 22:50:00 -04001255 return -ENOSPC;
1256 }
1257 ei->i_reserved_meta_blocks += md_needed;
1258 spin_unlock(&ei->i_block_reservation_lock);
1259
1260 return 0; /* success */
1261}
1262
1263/*
Aditya Kali7b415bf2011-09-09 19:04:51 -04001264 * Reserve a single cluster located at lblock
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001265 */
Theodore Ts'o01f49d02011-01-10 12:13:03 -05001266static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
Mingming Caod2a17632008-07-14 17:52:37 -04001267{
Mingming Cao60e58e02009-01-22 18:13:05 +01001268 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001269 struct ext4_inode_info *ei = EXT4_I(inode);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001270 unsigned int md_needed;
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001271 int ret;
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001272 ext4_lblk_t save_last_lblock;
1273 int save_len;
Mingming Caod2a17632008-07-14 17:52:37 -04001274
Mingming Cao60e58e02009-01-22 18:13:05 +01001275 /*
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001276 * We will charge metadata quota at writeout time; this saves
1277 * us from metadata over-estimation, though we may go over by
1278 * a small amount in the end. Here we just reserve for data.
Mingming Cao60e58e02009-01-22 18:13:05 +01001279 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04001280 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001281 if (ret)
1282 return ret;
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001283
1284 /*
1285 * recalculate the amount of metadata blocks to reserve
1286 * in order to allocate nrblocks
1287 * worse case is one extent per block
1288 */
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001289 spin_lock(&ei->i_block_reservation_lock);
1290 /*
1291 * ext4_calc_metadata_amount() has side effects, which we have
1292 * to be prepared undo if we fail to claim space.
1293 */
1294 save_len = ei->i_da_metadata_calc_len;
1295 save_last_lblock = ei->i_da_metadata_calc_last_lblock;
1296 md_needed = EXT4_NUM_B2C(sbi,
1297 ext4_calc_metadata_amount(inode, lblock));
1298 trace_ext4_da_reserve_space(inode, md_needed);
1299
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001300 /*
1301 * We do still charge estimated metadata to the sb though;
1302 * we cannot afford to run out of free blocks.
1303 */
Theodore Ts'oe7d5f312011-09-09 19:14:51 -04001304 if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) {
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001305 ei->i_da_metadata_calc_len = save_len;
1306 ei->i_da_metadata_calc_last_lblock = save_last_lblock;
1307 spin_unlock(&ei->i_block_reservation_lock);
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001308 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
Mingming Caod2a17632008-07-14 17:52:37 -04001309 return -ENOSPC;
1310 }
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001311 ei->i_reserved_data_blocks++;
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001312 ei->i_reserved_meta_blocks += md_needed;
1313 spin_unlock(&ei->i_block_reservation_lock);
Dmitry Monakhov39bc6802009-12-10 16:36:27 +00001314
Mingming Caod2a17632008-07-14 17:52:37 -04001315 return 0; /* success */
1316}
1317
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001318static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001319{
1320 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001321 struct ext4_inode_info *ei = EXT4_I(inode);
Mingming Caod2a17632008-07-14 17:52:37 -04001322
Mingming Caocd213222008-08-19 22:16:59 -04001323 if (!to_free)
1324 return; /* Nothing to release, exit */
1325
Mingming Caod2a17632008-07-14 17:52:37 -04001326 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001327
Li Zefan5a58ec82010-05-17 02:00:00 -04001328 trace_ext4_da_release_space(inode, to_free);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001329 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
Mingming Caocd213222008-08-19 22:16:59 -04001330 /*
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001331 * if there aren't enough reserved blocks, then the
1332 * counter is messed up somewhere. Since this
1333 * function is called from invalidate page, it's
1334 * harmless to return without any action.
Mingming Caocd213222008-08-19 22:16:59 -04001335 */
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001336 ext4_warning(inode->i_sb, "ext4_da_release_space: "
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001337 "ino %lu, to_free %d with only %d reserved "
Theodore Ts'o1084f252012-03-19 23:13:43 -04001338 "data blocks", inode->i_ino, to_free,
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001339 ei->i_reserved_data_blocks);
1340 WARN_ON(1);
1341 to_free = ei->i_reserved_data_blocks;
1342 }
1343 ei->i_reserved_data_blocks -= to_free;
1344
1345 if (ei->i_reserved_data_blocks == 0) {
1346 /*
1347 * We can release all of the reserved metadata blocks
1348 * only when we have written all of the delayed
1349 * allocation blocks.
Aditya Kali7b415bf2011-09-09 19:04:51 -04001350 * Note that in case of bigalloc, i_reserved_meta_blocks,
1351 * i_reserved_data_blocks, etc. refer to number of clusters.
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001352 */
Theodore Ts'o57042652011-09-09 18:56:51 -04001353 percpu_counter_sub(&sbi->s_dirtyclusters_counter,
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001354 ei->i_reserved_meta_blocks);
Theodore Ts'oee5f4d92010-01-01 02:36:15 -05001355 ei->i_reserved_meta_blocks = 0;
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001356 ei->i_da_metadata_calc_len = 0;
Mingming Caocd213222008-08-19 22:16:59 -04001357 }
1358
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001359 /* update fs dirty data blocks counter */
Theodore Ts'o57042652011-09-09 18:56:51 -04001360 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
Mingming Caod2a17632008-07-14 17:52:37 -04001361
Mingming Caod2a17632008-07-14 17:52:37 -04001362 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001363
Aditya Kali7b415bf2011-09-09 19:04:51 -04001364 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
Mingming Caod2a17632008-07-14 17:52:37 -04001365}
1366
1367static void ext4_da_page_release_reservation(struct page *page,
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001368 unsigned int offset,
1369 unsigned int length)
Mingming Caod2a17632008-07-14 17:52:37 -04001370{
1371 int to_release = 0;
1372 struct buffer_head *head, *bh;
1373 unsigned int curr_off = 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04001374 struct inode *inode = page->mapping->host;
1375 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001376 unsigned int stop = offset + length;
Aditya Kali7b415bf2011-09-09 19:04:51 -04001377 int num_clusters;
Zheng Liu51865fd2012-11-08 21:57:32 -05001378 ext4_fsblk_t lblk;
Mingming Caod2a17632008-07-14 17:52:37 -04001379
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001380 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1381
Mingming Caod2a17632008-07-14 17:52:37 -04001382 head = page_buffers(page);
1383 bh = head;
1384 do {
1385 unsigned int next_off = curr_off + bh->b_size;
1386
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001387 if (next_off > stop)
1388 break;
1389
Mingming Caod2a17632008-07-14 17:52:37 -04001390 if ((offset <= curr_off) && (buffer_delay(bh))) {
1391 to_release++;
1392 clear_buffer_delay(bh);
1393 }
1394 curr_off = next_off;
1395 } while ((bh = bh->b_this_page) != head);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001396
Zheng Liu51865fd2012-11-08 21:57:32 -05001397 if (to_release) {
1398 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1399 ext4_es_remove_extent(inode, lblk, to_release);
1400 }
1401
Aditya Kali7b415bf2011-09-09 19:04:51 -04001402 /* If we have released all the blocks belonging to a cluster, then we
1403 * need to release the reserved space for that cluster. */
1404 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1405 while (num_clusters > 0) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04001406 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1407 ((num_clusters - 1) << sbi->s_cluster_bits);
1408 if (sbi->s_cluster_ratio == 1 ||
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05001409 !ext4_find_delalloc_cluster(inode, lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04001410 ext4_da_release_space(inode, 1);
1411
1412 num_clusters--;
1413 }
Mingming Caod2a17632008-07-14 17:52:37 -04001414}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001415
1416/*
Alex Tomas64769242008-07-11 19:27:31 -04001417 * Delayed allocation stuff
1418 */
1419
Jan Kara4e7ea812013-06-04 13:17:40 -04001420struct mpage_da_data {
1421 struct inode *inode;
1422 struct writeback_control *wbc;
Jan Kara6b523df2013-06-04 13:21:11 -04001423
Jan Kara4e7ea812013-06-04 13:17:40 -04001424 pgoff_t first_page; /* The first page to write */
1425 pgoff_t next_page; /* Current page to examine */
1426 pgoff_t last_page; /* Last page to examine */
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001427 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04001428 * Extent to map - this can be after first_page because that can be
1429 * fully mapped. We somewhat abuse m_flags to store whether the extent
1430 * is delalloc or unwritten.
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001431 */
Jan Kara4e7ea812013-06-04 13:17:40 -04001432 struct ext4_map_blocks map;
1433 struct ext4_io_submit io_submit; /* IO submission data */
1434};
Alex Tomas64769242008-07-11 19:27:31 -04001435
Jan Kara4e7ea812013-06-04 13:17:40 -04001436static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1437 bool invalidate)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001438{
1439 int nr_pages, i;
1440 pgoff_t index, end;
1441 struct pagevec pvec;
1442 struct inode *inode = mpd->inode;
1443 struct address_space *mapping = inode->i_mapping;
Jan Kara4e7ea812013-06-04 13:17:40 -04001444
1445 /* This is necessary when next_page == 0. */
1446 if (mpd->first_page >= mpd->next_page)
1447 return;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001448
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001449 index = mpd->first_page;
1450 end = mpd->next_page - 1;
Jan Kara4e7ea812013-06-04 13:17:40 -04001451 if (invalidate) {
1452 ext4_lblk_t start, last;
1453 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1454 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1455 ext4_es_remove_extent(inode, start, last - start + 1);
1456 }
Zheng Liu51865fd2012-11-08 21:57:32 -05001457
Eric Sandeen66bea922012-11-14 22:22:05 -05001458 pagevec_init(&pvec, 0);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001459 while (index <= end) {
1460 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1461 if (nr_pages == 0)
1462 break;
1463 for (i = 0; i < nr_pages; i++) {
1464 struct page *page = pvec.pages[i];
Jan Kara9b1d09982010-03-03 16:19:32 -05001465 if (page->index > end)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001466 break;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001467 BUG_ON(!PageLocked(page));
1468 BUG_ON(PageWriteback(page));
Jan Kara4e7ea812013-06-04 13:17:40 -04001469 if (invalidate) {
1470 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1471 ClearPageUptodate(page);
1472 }
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001473 unlock_page(page);
1474 }
Jan Kara9b1d09982010-03-03 16:19:32 -05001475 index = pvec.pages[nr_pages - 1]->index + 1;
1476 pagevec_release(&pvec);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001477 }
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001478}
1479
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001480static void ext4_print_free_blocks(struct inode *inode)
1481{
1482 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o92b97812012-03-19 23:41:49 -04001483 struct super_block *sb = inode->i_sb;
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001484 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'o92b97812012-03-19 23:41:49 -04001485
1486 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
Theodore Ts'o5dee5432011-09-09 19:10:51 -04001487 EXT4_C2B(EXT4_SB(inode->i_sb),
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001488 ext4_count_free_clusters(sb)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001489 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1490 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001491 (long long) EXT4_C2B(EXT4_SB(sb),
Theodore Ts'o57042652011-09-09 18:56:51 -04001492 percpu_counter_sum(&sbi->s_freeclusters_counter)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001493 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001494 (long long) EXT4_C2B(EXT4_SB(sb),
Aditya Kali7b415bf2011-09-09 19:04:51 -04001495 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001496 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1497 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001498 ei->i_reserved_data_blocks);
Theodore Ts'o92b97812012-03-19 23:41:49 -04001499 ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001500 ei->i_reserved_meta_blocks);
1501 ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u",
1502 ei->i_allocated_meta_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001503 return;
1504}
1505
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001506static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001507{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001508 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001509}
1510
Alex Tomas64769242008-07-11 19:27:31 -04001511/*
Aditya Kali5356f262011-09-09 19:20:51 -04001512 * This function is grabs code from the very beginning of
1513 * ext4_map_blocks, but assumes that the caller is from delayed write
1514 * time. This function looks up the requested blocks and sets the
1515 * buffer delay bit under the protection of i_data_sem.
1516 */
1517static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1518 struct ext4_map_blocks *map,
1519 struct buffer_head *bh)
1520{
Zheng Liud100eef2013-02-18 00:29:59 -05001521 struct extent_status es;
Aditya Kali5356f262011-09-09 19:20:51 -04001522 int retval;
1523 sector_t invalid_block = ~((sector_t) 0xffff);
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001524#ifdef ES_AGGRESSIVE_TEST
1525 struct ext4_map_blocks orig_map;
1526
1527 memcpy(&orig_map, map, sizeof(*map));
1528#endif
Aditya Kali5356f262011-09-09 19:20:51 -04001529
1530 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1531 invalid_block = ~0;
1532
1533 map->m_flags = 0;
1534 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1535 "logical block %lu\n", inode->i_ino, map->m_len,
1536 (unsigned long) map->m_lblk);
Zheng Liud100eef2013-02-18 00:29:59 -05001537
1538 /* Lookup extent status tree firstly */
1539 if (ext4_es_lookup_extent(inode, iblock, &es)) {
Theodore Ts'o63b99962013-07-16 10:28:47 -04001540 ext4_es_lru_add(inode);
Zheng Liud100eef2013-02-18 00:29:59 -05001541 if (ext4_es_is_hole(&es)) {
1542 retval = 0;
1543 down_read((&EXT4_I(inode)->i_data_sem));
1544 goto add_delayed;
1545 }
1546
1547 /*
1548 * Delayed extent could be allocated by fallocate.
1549 * So we need to check it.
1550 */
1551 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1552 map_bh(bh, inode->i_sb, invalid_block);
1553 set_buffer_new(bh);
1554 set_buffer_delay(bh);
1555 return 0;
1556 }
1557
1558 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1559 retval = es.es_len - (iblock - es.es_lblk);
1560 if (retval > map->m_len)
1561 retval = map->m_len;
1562 map->m_len = retval;
1563 if (ext4_es_is_written(&es))
1564 map->m_flags |= EXT4_MAP_MAPPED;
1565 else if (ext4_es_is_unwritten(&es))
1566 map->m_flags |= EXT4_MAP_UNWRITTEN;
1567 else
1568 BUG_ON(1);
1569
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001570#ifdef ES_AGGRESSIVE_TEST
1571 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1572#endif
Zheng Liud100eef2013-02-18 00:29:59 -05001573 return retval;
1574 }
1575
Aditya Kali5356f262011-09-09 19:20:51 -04001576 /*
1577 * Try to see if we can get the block without requesting a new
1578 * file system block.
1579 */
1580 down_read((&EXT4_I(inode)->i_data_sem));
Tao Ma9c3569b2012-12-10 14:05:57 -05001581 if (ext4_has_inline_data(inode)) {
1582 /*
1583 * We will soon create blocks for this page, and let
1584 * us pretend as if the blocks aren't allocated yet.
1585 * In case of clusters, we have to handle the work
1586 * of mapping from cluster so that the reserved space
1587 * is calculated properly.
1588 */
1589 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1590 ext4_find_delalloc_cluster(inode, map->m_lblk))
1591 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1592 retval = 0;
1593 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Zheng Liud100eef2013-02-18 00:29:59 -05001594 retval = ext4_ext_map_blocks(NULL, inode, map,
1595 EXT4_GET_BLOCKS_NO_PUT_HOLE);
Aditya Kali5356f262011-09-09 19:20:51 -04001596 else
Zheng Liud100eef2013-02-18 00:29:59 -05001597 retval = ext4_ind_map_blocks(NULL, inode, map,
1598 EXT4_GET_BLOCKS_NO_PUT_HOLE);
Aditya Kali5356f262011-09-09 19:20:51 -04001599
Zheng Liud100eef2013-02-18 00:29:59 -05001600add_delayed:
Aditya Kali5356f262011-09-09 19:20:51 -04001601 if (retval == 0) {
Zheng Liuf7fec032013-02-18 00:28:47 -05001602 int ret;
Aditya Kali5356f262011-09-09 19:20:51 -04001603 /*
1604 * XXX: __block_prepare_write() unmaps passed block,
1605 * is it OK?
1606 */
Lukas Czerner386ad672013-03-10 22:50:00 -04001607 /*
1608 * If the block was allocated from previously allocated cluster,
1609 * then we don't need to reserve it again. However we still need
1610 * to reserve metadata for every block we're going to write.
1611 */
Aditya Kali5356f262011-09-09 19:20:51 -04001612 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
Zheng Liuf7fec032013-02-18 00:28:47 -05001613 ret = ext4_da_reserve_space(inode, iblock);
1614 if (ret) {
Aditya Kali5356f262011-09-09 19:20:51 -04001615 /* not enough space to reserve */
Zheng Liuf7fec032013-02-18 00:28:47 -05001616 retval = ret;
Aditya Kali5356f262011-09-09 19:20:51 -04001617 goto out_unlock;
Zheng Liuf7fec032013-02-18 00:28:47 -05001618 }
Lukas Czerner386ad672013-03-10 22:50:00 -04001619 } else {
1620 ret = ext4_da_reserve_metadata(inode, iblock);
1621 if (ret) {
1622 /* not enough space to reserve */
1623 retval = ret;
1624 goto out_unlock;
1625 }
Aditya Kali5356f262011-09-09 19:20:51 -04001626 }
1627
Zheng Liuf7fec032013-02-18 00:28:47 -05001628 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1629 ~0, EXTENT_STATUS_DELAYED);
1630 if (ret) {
1631 retval = ret;
Zheng Liu51865fd2012-11-08 21:57:32 -05001632 goto out_unlock;
Zheng Liuf7fec032013-02-18 00:28:47 -05001633 }
Zheng Liu51865fd2012-11-08 21:57:32 -05001634
Aditya Kali5356f262011-09-09 19:20:51 -04001635 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1636 * and it should not appear on the bh->b_state.
1637 */
1638 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1639
1640 map_bh(bh, inode->i_sb, invalid_block);
1641 set_buffer_new(bh);
1642 set_buffer_delay(bh);
Zheng Liuf7fec032013-02-18 00:28:47 -05001643 } else if (retval > 0) {
1644 int ret;
Theodore Ts'o3be78c72013-08-16 21:22:41 -04001645 unsigned int status;
Zheng Liuf7fec032013-02-18 00:28:47 -05001646
Zheng Liu44fb851d2013-07-29 12:51:42 -04001647 if (unlikely(retval != map->m_len)) {
1648 ext4_warning(inode->i_sb,
1649 "ES len assertion failed for inode "
1650 "%lu: retval %d != map->m_len %d",
1651 inode->i_ino, retval, map->m_len);
1652 WARN_ON(1);
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001653 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001654
Zheng Liuf7fec032013-02-18 00:28:47 -05001655 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1656 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1657 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1658 map->m_pblk, status);
1659 if (ret != 0)
1660 retval = ret;
Aditya Kali5356f262011-09-09 19:20:51 -04001661 }
1662
1663out_unlock:
1664 up_read((&EXT4_I(inode)->i_data_sem));
1665
1666 return retval;
1667}
1668
1669/*
Theodore Ts'ob920c752009-05-14 00:54:29 -04001670 * This is a special get_blocks_t callback which is used by
1671 * ext4_da_write_begin(). It will either return mapped block or
1672 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001673 *
1674 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1675 * We also have b_blocknr = -1 and b_bdev initialized properly
1676 *
1677 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1678 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1679 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04001680 */
Tao Ma9c3569b2012-12-10 14:05:57 -05001681int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1682 struct buffer_head *bh, int create)
Alex Tomas64769242008-07-11 19:27:31 -04001683{
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001684 struct ext4_map_blocks map;
Alex Tomas64769242008-07-11 19:27:31 -04001685 int ret = 0;
1686
1687 BUG_ON(create == 0);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001688 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1689
1690 map.m_lblk = iblock;
1691 map.m_len = 1;
Alex Tomas64769242008-07-11 19:27:31 -04001692
1693 /*
1694 * first, we need to know whether the block is allocated already
1695 * preallocated blocks are unmapped but should treated
1696 * the same as allocated blocks.
1697 */
Aditya Kali5356f262011-09-09 19:20:51 -04001698 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1699 if (ret <= 0)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001700 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04001701
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001702 map_bh(bh, inode->i_sb, map.m_pblk);
1703 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1704
1705 if (buffer_unwritten(bh)) {
1706 /* A delayed write to unwritten bh should be marked
1707 * new and mapped. Mapped ensures that we don't do
1708 * get_block multiple times when we write to the same
1709 * offset and new ensures that we do proper zero out
1710 * for partial write.
1711 */
1712 set_buffer_new(bh);
Theodore Ts'oc8205632011-04-10 22:30:07 -04001713 set_buffer_mapped(bh);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001714 }
1715 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04001716}
Mingming Cao61628a32008-07-11 19:27:31 -04001717
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001718static int bget_one(handle_t *handle, struct buffer_head *bh)
1719{
1720 get_bh(bh);
1721 return 0;
1722}
1723
1724static int bput_one(handle_t *handle, struct buffer_head *bh)
1725{
1726 put_bh(bh);
1727 return 0;
1728}
1729
1730static int __ext4_journalled_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001731 unsigned int len)
1732{
1733 struct address_space *mapping = page->mapping;
1734 struct inode *inode = mapping->host;
Tao Ma3fdcfb62012-12-10 14:05:57 -05001735 struct buffer_head *page_bufs = NULL;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001736 handle_t *handle = NULL;
Tao Ma3fdcfb62012-12-10 14:05:57 -05001737 int ret = 0, err = 0;
1738 int inline_data = ext4_has_inline_data(inode);
1739 struct buffer_head *inode_bh = NULL;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001740
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001741 ClearPageChecked(page);
Tao Ma3fdcfb62012-12-10 14:05:57 -05001742
1743 if (inline_data) {
1744 BUG_ON(page->index != 0);
1745 BUG_ON(len > ext4_get_max_inline_size(inode));
1746 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1747 if (inode_bh == NULL)
1748 goto out;
1749 } else {
1750 page_bufs = page_buffers(page);
1751 if (!page_bufs) {
1752 BUG();
1753 goto out;
1754 }
1755 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1756 NULL, bget_one);
1757 }
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001758 /* As soon as we unlock the page, it can go away, but we have
1759 * references to buffers so we are safe */
1760 unlock_page(page);
1761
Theodore Ts'o9924a922013-02-08 21:59:22 -05001762 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1763 ext4_writepage_trans_blocks(inode));
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001764 if (IS_ERR(handle)) {
1765 ret = PTR_ERR(handle);
1766 goto out;
1767 }
1768
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001769 BUG_ON(!ext4_handle_valid(handle));
1770
Tao Ma3fdcfb62012-12-10 14:05:57 -05001771 if (inline_data) {
1772 ret = ext4_journal_get_write_access(handle, inode_bh);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001773
Tao Ma3fdcfb62012-12-10 14:05:57 -05001774 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1775
1776 } else {
1777 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1778 do_journal_get_write_access);
1779
1780 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1781 write_end_fn);
1782 }
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001783 if (ret == 0)
1784 ret = err;
Jan Kara2d859db2011-07-26 09:07:11 -04001785 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001786 err = ext4_journal_stop(handle);
1787 if (!ret)
1788 ret = err;
1789
Tao Ma3fdcfb62012-12-10 14:05:57 -05001790 if (!ext4_has_inline_data(inode))
Theodore Ts'o8c9367f2014-01-07 13:08:03 -05001791 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
Tao Ma3fdcfb62012-12-10 14:05:57 -05001792 NULL, bput_one);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001793 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001794out:
Tao Ma3fdcfb62012-12-10 14:05:57 -05001795 brelse(inode_bh);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001796 return ret;
1797}
1798
Mingming Cao61628a32008-07-11 19:27:31 -04001799/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001800 * Note that we don't need to start a transaction unless we're journaling data
1801 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1802 * need to file the inode to the transaction's list in ordered mode because if
1803 * we are writing back data added by write(), the inode is already there and if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001804 * we are writing back data modified via mmap(), no one guarantees in which
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001805 * transaction the data will hit the disk. In case we are journaling data, we
1806 * cannot start transaction directly because transaction start ranks above page
1807 * lock so we have to do some magic.
1808 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04001809 * This function can get called via...
Theodore Ts'o20970ba2013-06-06 14:00:46 -04001810 * - ext4_writepages after taking page lock (have journal handle)
Theodore Ts'ob920c752009-05-14 00:54:29 -04001811 * - journal_submit_inode_data_buffers (no journal handle)
Artem Bityutskiyf6463b02012-07-25 18:12:04 +03001812 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
Theodore Ts'ob920c752009-05-14 00:54:29 -04001813 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001814 *
1815 * We don't do any block allocation in this function. If we have page with
1816 * multiple blocks we need to write those buffer_heads that are mapped. This
1817 * is important for mmaped based write. So if we do with blocksize 1K
1818 * truncate(f, 1024);
1819 * a = mmap(f, 0, 4096);
1820 * a[0] = 'a';
1821 * truncate(f, 4096);
1822 * we have in the page first buffer_head mapped via page_mkwrite call back
Paul Bolle90802ed2011-12-05 13:00:34 +01001823 * but other buffer_heads would be unmapped but dirty (dirty done via the
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001824 * do_wp_page). So writepage should write the first block. If we modify
1825 * the mmap area beyond 1024 we will again get a page_fault and the
1826 * page_mkwrite callback will do the block allocation and mark the
1827 * buffer_heads mapped.
1828 *
1829 * We redirty the page if we have any buffer_heads that is either delay or
1830 * unwritten in the page.
1831 *
1832 * We can get recursively called as show below.
1833 *
1834 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1835 * ext4_writepage()
1836 *
1837 * But since we don't do any block allocation we should not deadlock.
1838 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04001839 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001840static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001841 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04001842{
Jan Karaf8bec372013-01-28 12:55:08 -05001843 int ret = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04001844 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001845 unsigned int len;
Jiaying Zhang744692d2010-03-04 16:14:02 -05001846 struct buffer_head *page_bufs = NULL;
Mingming Cao61628a32008-07-11 19:27:31 -04001847 struct inode *inode = page->mapping->host;
Jan Kara36ade452013-01-28 09:30:52 -05001848 struct ext4_io_submit io_submit;
Alex Tomas64769242008-07-11 19:27:31 -04001849
Lukas Czernera9c667f2011-06-06 09:51:52 -04001850 trace_ext4_writepage(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001851 size = i_size_read(inode);
1852 if (page->index == size >> PAGE_CACHE_SHIFT)
1853 len = size & ~PAGE_CACHE_MASK;
1854 else
1855 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04001856
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001857 page_bufs = page_buffers(page);
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001858 /*
Jan Karafe386132013-01-28 21:06:42 -05001859 * We cannot do block allocation or other extent handling in this
1860 * function. If there are buffers needing that, we have to redirty
1861 * the page. But we may reach here when we do a journal commit via
1862 * journal_submit_inode_data_buffers() and in that case we must write
1863 * allocated buffers to achieve data=ordered mode guarantees.
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04001864 */
Tao Maf19d5872012-12-10 14:05:51 -05001865 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1866 ext4_bh_delay_or_unwritten)) {
Jan Karaf8bec372013-01-28 12:55:08 -05001867 redirty_page_for_writepage(wbc, page);
Jan Karafe386132013-01-28 21:06:42 -05001868 if (current->flags & PF_MEMALLOC) {
1869 /*
1870 * For memory cleaning there's no point in writing only
1871 * some buffers. So just bail out. Warn if we came here
1872 * from direct reclaim.
1873 */
1874 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1875 == PF_MEMALLOC);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001876 unlock_page(page);
1877 return 0;
1878 }
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001879 }
Alex Tomas64769242008-07-11 19:27:31 -04001880
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001881 if (PageChecked(page) && ext4_should_journal_data(inode))
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001882 /*
1883 * It's mmapped pagecache. Add buffers and journal it. There
1884 * doesn't seem much point in redirtying the page here.
1885 */
Wu Fengguang3f0ca302009-11-24 11:15:44 -05001886 return __ext4_journalled_writepage(page, len);
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001887
Jan Kara97a851e2013-06-04 11:58:58 -04001888 ext4_io_submit_init(&io_submit, wbc);
1889 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1890 if (!io_submit.io_end) {
1891 redirty_page_for_writepage(wbc, page);
1892 unlock_page(page);
1893 return -ENOMEM;
1894 }
Jan Kara36ade452013-01-28 09:30:52 -05001895 ret = ext4_bio_write_page(&io_submit, page, len, wbc);
1896 ext4_io_submit(&io_submit);
Jan Kara97a851e2013-06-04 11:58:58 -04001897 /* Drop io_end reference we got from init */
1898 ext4_put_io_end_defer(io_submit.io_end);
Alex Tomas64769242008-07-11 19:27:31 -04001899 return ret;
1900}
1901
Jan Kara5f1132b2013-08-17 10:02:33 -04001902static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1903{
1904 int len;
1905 loff_t size = i_size_read(mpd->inode);
1906 int err;
1907
1908 BUG_ON(page->index != mpd->first_page);
1909 if (page->index == size >> PAGE_CACHE_SHIFT)
1910 len = size & ~PAGE_CACHE_MASK;
1911 else
1912 len = PAGE_CACHE_SIZE;
1913 clear_page_dirty_for_io(page);
1914 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc);
1915 if (!err)
1916 mpd->wbc->nr_to_write--;
1917 mpd->first_page++;
1918
1919 return err;
1920}
1921
Jan Kara4e7ea812013-06-04 13:17:40 -04001922#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1923
Mingming Cao61628a32008-07-11 19:27:31 -04001924/*
Jan Karafffb2732013-06-04 13:01:11 -04001925 * mballoc gives us at most this number of blocks...
1926 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
Anatol Pomozov70261f52013-08-28 14:40:12 -04001927 * The rest of mballoc seems to handle chunks up to full group size.
Mingming Cao61628a32008-07-11 19:27:31 -04001928 */
Jan Karafffb2732013-06-04 13:01:11 -04001929#define MAX_WRITEPAGES_EXTENT_LEN 2048
Mingming Cao525f4ed2008-08-19 22:15:58 -04001930
Jan Karafffb2732013-06-04 13:01:11 -04001931/*
Jan Kara4e7ea812013-06-04 13:17:40 -04001932 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1933 *
1934 * @mpd - extent of blocks
1935 * @lblk - logical number of the block in the file
Jan Kara09930042013-08-17 09:57:56 -04001936 * @bh - buffer head we want to add to the extent
Jan Kara4e7ea812013-06-04 13:17:40 -04001937 *
Jan Kara09930042013-08-17 09:57:56 -04001938 * The function is used to collect contig. blocks in the same state. If the
1939 * buffer doesn't require mapping for writeback and we haven't started the
1940 * extent of buffers to map yet, the function returns 'true' immediately - the
1941 * caller can write the buffer right away. Otherwise the function returns true
1942 * if the block has been added to the extent, false if the block couldn't be
1943 * added.
Jan Kara4e7ea812013-06-04 13:17:40 -04001944 */
Jan Kara09930042013-08-17 09:57:56 -04001945static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1946 struct buffer_head *bh)
Jan Kara4e7ea812013-06-04 13:17:40 -04001947{
1948 struct ext4_map_blocks *map = &mpd->map;
1949
Jan Kara09930042013-08-17 09:57:56 -04001950 /* Buffer that doesn't need mapping for writeback? */
1951 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1952 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1953 /* So far no extent to map => we write the buffer right away */
1954 if (map->m_len == 0)
1955 return true;
1956 return false;
1957 }
Jan Kara4e7ea812013-06-04 13:17:40 -04001958
1959 /* First block in the extent? */
1960 if (map->m_len == 0) {
1961 map->m_lblk = lblk;
1962 map->m_len = 1;
Jan Kara09930042013-08-17 09:57:56 -04001963 map->m_flags = bh->b_state & BH_FLAGS;
1964 return true;
Jan Kara4e7ea812013-06-04 13:17:40 -04001965 }
1966
Jan Kara09930042013-08-17 09:57:56 -04001967 /* Don't go larger than mballoc is willing to allocate */
1968 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1969 return false;
1970
Jan Kara4e7ea812013-06-04 13:17:40 -04001971 /* Can we merge the block to our big extent? */
1972 if (lblk == map->m_lblk + map->m_len &&
Jan Kara09930042013-08-17 09:57:56 -04001973 (bh->b_state & BH_FLAGS) == map->m_flags) {
Jan Kara4e7ea812013-06-04 13:17:40 -04001974 map->m_len++;
Jan Kara09930042013-08-17 09:57:56 -04001975 return true;
Jan Kara4e7ea812013-06-04 13:17:40 -04001976 }
Jan Kara09930042013-08-17 09:57:56 -04001977 return false;
Jan Kara4e7ea812013-06-04 13:17:40 -04001978}
1979
Jan Kara5f1132b2013-08-17 10:02:33 -04001980/*
1981 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1982 *
1983 * @mpd - extent of blocks for mapping
1984 * @head - the first buffer in the page
1985 * @bh - buffer we should start processing from
1986 * @lblk - logical number of the block in the file corresponding to @bh
1987 *
1988 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1989 * the page for IO if all buffers in this page were mapped and there's no
1990 * accumulated extent of buffers to map or add buffers in the page to the
1991 * extent of buffers to map. The function returns 1 if the caller can continue
1992 * by processing the next page, 0 if it should stop adding buffers to the
1993 * extent to map because we cannot extend it anymore. It can also return value
1994 * < 0 in case of error during IO submission.
1995 */
1996static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1997 struct buffer_head *head,
1998 struct buffer_head *bh,
1999 ext4_lblk_t lblk)
Jan Kara4e7ea812013-06-04 13:17:40 -04002000{
2001 struct inode *inode = mpd->inode;
Jan Kara5f1132b2013-08-17 10:02:33 -04002002 int err;
Jan Kara4e7ea812013-06-04 13:17:40 -04002003 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2004 >> inode->i_blkbits;
2005
2006 do {
2007 BUG_ON(buffer_locked(bh));
2008
Jan Kara09930042013-08-17 09:57:56 -04002009 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
Jan Kara4e7ea812013-06-04 13:17:40 -04002010 /* Found extent to map? */
2011 if (mpd->map.m_len)
Jan Kara5f1132b2013-08-17 10:02:33 -04002012 return 0;
Jan Kara09930042013-08-17 09:57:56 -04002013 /* Everything mapped so far and we hit EOF */
Jan Kara5f1132b2013-08-17 10:02:33 -04002014 break;
Jan Kara4e7ea812013-06-04 13:17:40 -04002015 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002016 } while (lblk++, (bh = bh->b_this_page) != head);
Jan Kara5f1132b2013-08-17 10:02:33 -04002017 /* So far everything mapped? Submit the page for IO. */
2018 if (mpd->map.m_len == 0) {
2019 err = mpage_submit_page(mpd, head->b_page);
2020 if (err < 0)
2021 return err;
2022 }
2023 return lblk < blocks;
Jan Kara4e7ea812013-06-04 13:17:40 -04002024}
2025
2026/*
2027 * mpage_map_buffers - update buffers corresponding to changed extent and
2028 * submit fully mapped pages for IO
2029 *
2030 * @mpd - description of extent to map, on return next extent to map
2031 *
2032 * Scan buffers corresponding to changed extent (we expect corresponding pages
2033 * to be already locked) and update buffer state according to new extent state.
2034 * We map delalloc buffers to their physical location, clear unwritten bits,
2035 * and mark buffers as uninit when we perform writes to uninitialized extents
2036 * and do extent conversion after IO is finished. If the last page is not fully
2037 * mapped, we update @map to the next extent in the last page that needs
2038 * mapping. Otherwise we submit the page for IO.
2039 */
2040static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2041{
2042 struct pagevec pvec;
2043 int nr_pages, i;
2044 struct inode *inode = mpd->inode;
2045 struct buffer_head *head, *bh;
2046 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
Jan Kara4e7ea812013-06-04 13:17:40 -04002047 pgoff_t start, end;
2048 ext4_lblk_t lblk;
2049 sector_t pblock;
2050 int err;
2051
2052 start = mpd->map.m_lblk >> bpp_bits;
2053 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2054 lblk = start << bpp_bits;
2055 pblock = mpd->map.m_pblk;
2056
2057 pagevec_init(&pvec, 0);
2058 while (start <= end) {
2059 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2060 PAGEVEC_SIZE);
2061 if (nr_pages == 0)
2062 break;
2063 for (i = 0; i < nr_pages; i++) {
2064 struct page *page = pvec.pages[i];
2065
2066 if (page->index > end)
2067 break;
Anatol Pomozov70261f52013-08-28 14:40:12 -04002068 /* Up to 'end' pages must be contiguous */
Jan Kara4e7ea812013-06-04 13:17:40 -04002069 BUG_ON(page->index != start);
2070 bh = head = page_buffers(page);
2071 do {
2072 if (lblk < mpd->map.m_lblk)
2073 continue;
2074 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2075 /*
2076 * Buffer after end of mapped extent.
2077 * Find next buffer in the page to map.
2078 */
2079 mpd->map.m_len = 0;
2080 mpd->map.m_flags = 0;
Jan Kara5f1132b2013-08-17 10:02:33 -04002081 /*
2082 * FIXME: If dioread_nolock supports
2083 * blocksize < pagesize, we need to make
2084 * sure we add size mapped so far to
2085 * io_end->size as the following call
2086 * can submit the page for IO.
2087 */
2088 err = mpage_process_page_bufs(mpd, head,
2089 bh, lblk);
Jan Kara4e7ea812013-06-04 13:17:40 -04002090 pagevec_release(&pvec);
Jan Kara5f1132b2013-08-17 10:02:33 -04002091 if (err > 0)
2092 err = 0;
2093 return err;
Jan Kara4e7ea812013-06-04 13:17:40 -04002094 }
2095 if (buffer_delay(bh)) {
2096 clear_buffer_delay(bh);
2097 bh->b_blocknr = pblock++;
2098 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002099 clear_buffer_unwritten(bh);
Jan Kara5f1132b2013-08-17 10:02:33 -04002100 } while (lblk++, (bh = bh->b_this_page) != head);
Jan Kara4e7ea812013-06-04 13:17:40 -04002101
2102 /*
2103 * FIXME: This is going to break if dioread_nolock
2104 * supports blocksize < pagesize as we will try to
2105 * convert potentially unmapped parts of inode.
2106 */
2107 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2108 /* Page fully mapped - let IO run! */
2109 err = mpage_submit_page(mpd, page);
2110 if (err < 0) {
2111 pagevec_release(&pvec);
2112 return err;
2113 }
2114 start++;
2115 }
2116 pagevec_release(&pvec);
2117 }
2118 /* Extent fully mapped and matches with page boundary. We are done. */
2119 mpd->map.m_len = 0;
2120 mpd->map.m_flags = 0;
2121 return 0;
2122}
2123
2124static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2125{
2126 struct inode *inode = mpd->inode;
2127 struct ext4_map_blocks *map = &mpd->map;
2128 int get_blocks_flags;
2129 int err;
2130
2131 trace_ext4_da_write_pages_extent(inode, map);
2132 /*
2133 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2134 * to convert an uninitialized extent to be initialized (in the case
2135 * where we have written into one or more preallocated blocks). It is
2136 * possible that we're going to need more metadata blocks than
2137 * previously reserved. However we must not fail because we're in
2138 * writeback and there is nothing we can do about it so it might result
2139 * in data loss. So use reserved blocks to allocate metadata if
2140 * possible.
2141 *
2142 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2143 * in question are delalloc blocks. This affects functions in many
2144 * different parts of the allocation call path. This flag exists
2145 * primarily because we don't want to change *many* call functions, so
2146 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2147 * once the inode's allocation semaphore is taken.
2148 */
2149 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2150 EXT4_GET_BLOCKS_METADATA_NOFAIL;
2151 if (ext4_should_dioread_nolock(inode))
2152 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2153 if (map->m_flags & (1 << BH_Delay))
2154 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2155
2156 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2157 if (err < 0)
2158 return err;
Jan Kara6b523df2013-06-04 13:21:11 -04002159 if (map->m_flags & EXT4_MAP_UNINIT) {
2160 if (!mpd->io_submit.io_end->handle &&
2161 ext4_handle_valid(handle)) {
2162 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2163 handle->h_rsv_handle = NULL;
2164 }
Jan Kara3613d222013-06-04 13:19:34 -04002165 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
Jan Kara6b523df2013-06-04 13:21:11 -04002166 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002167
2168 BUG_ON(map->m_len == 0);
2169 if (map->m_flags & EXT4_MAP_NEW) {
2170 struct block_device *bdev = inode->i_sb->s_bdev;
2171 int i;
2172
2173 for (i = 0; i < map->m_len; i++)
2174 unmap_underlying_metadata(bdev, map->m_pblk + i);
2175 }
2176 return 0;
2177}
2178
2179/*
2180 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2181 * mpd->len and submit pages underlying it for IO
2182 *
2183 * @handle - handle for journal operations
2184 * @mpd - extent to map
Jan Kara7534e852013-10-16 08:26:08 -04002185 * @give_up_on_write - we set this to true iff there is a fatal error and there
2186 * is no hope of writing the data. The caller should discard
2187 * dirty pages to avoid infinite loops.
Jan Kara4e7ea812013-06-04 13:17:40 -04002188 *
2189 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2190 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2191 * them to initialized or split the described range from larger unwritten
2192 * extent. Note that we need not map all the described range since allocation
2193 * can return less blocks or the range is covered by more unwritten extents. We
2194 * cannot map more because we are limited by reserved transaction credits. On
2195 * the other hand we always make sure that the last touched page is fully
2196 * mapped so that it can be written out (and thus forward progress is
2197 * guaranteed). After mapping we submit all mapped pages for IO.
2198 */
2199static int mpage_map_and_submit_extent(handle_t *handle,
Theodore Ts'ocb530542013-07-01 08:12:40 -04002200 struct mpage_da_data *mpd,
2201 bool *give_up_on_write)
Jan Kara4e7ea812013-06-04 13:17:40 -04002202{
2203 struct inode *inode = mpd->inode;
2204 struct ext4_map_blocks *map = &mpd->map;
2205 int err;
2206 loff_t disksize;
2207
2208 mpd->io_submit.io_end->offset =
2209 ((loff_t)map->m_lblk) << inode->i_blkbits;
Jan Kara27d7c4e2013-07-05 21:57:22 -04002210 do {
Jan Kara4e7ea812013-06-04 13:17:40 -04002211 err = mpage_map_one_extent(handle, mpd);
2212 if (err < 0) {
2213 struct super_block *sb = inode->i_sb;
2214
Theodore Ts'ocb530542013-07-01 08:12:40 -04002215 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2216 goto invalidate_dirty_pages;
Jan Kara4e7ea812013-06-04 13:17:40 -04002217 /*
Theodore Ts'ocb530542013-07-01 08:12:40 -04002218 * Let the uper layers retry transient errors.
2219 * In the case of ENOSPC, if ext4_count_free_blocks()
2220 * is non-zero, a commit should free up blocks.
Jan Kara4e7ea812013-06-04 13:17:40 -04002221 */
Theodore Ts'ocb530542013-07-01 08:12:40 -04002222 if ((err == -ENOMEM) ||
2223 (err == -ENOSPC && ext4_count_free_clusters(sb)))
2224 return err;
2225 ext4_msg(sb, KERN_CRIT,
2226 "Delayed block allocation failed for "
2227 "inode %lu at logical offset %llu with"
2228 " max blocks %u with error %d",
2229 inode->i_ino,
2230 (unsigned long long)map->m_lblk,
2231 (unsigned)map->m_len, -err);
2232 ext4_msg(sb, KERN_CRIT,
2233 "This should not happen!! Data will "
2234 "be lost\n");
2235 if (err == -ENOSPC)
2236 ext4_print_free_blocks(inode);
2237 invalidate_dirty_pages:
2238 *give_up_on_write = true;
Jan Kara4e7ea812013-06-04 13:17:40 -04002239 return err;
2240 }
2241 /*
2242 * Update buffer state, submit mapped pages, and get us new
2243 * extent to map
2244 */
2245 err = mpage_map_and_submit_buffers(mpd);
2246 if (err < 0)
2247 return err;
Jan Kara27d7c4e2013-07-05 21:57:22 -04002248 } while (map->m_len);
Jan Kara4e7ea812013-06-04 13:17:40 -04002249
Theodore Ts'o622cad12014-04-11 10:35:17 -04002250 /*
2251 * Update on-disk size after IO is submitted. Races with
2252 * truncate are avoided by checking i_size under i_data_sem.
2253 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002254 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
Jan Kara4e7ea812013-06-04 13:17:40 -04002255 if (disksize > EXT4_I(inode)->i_disksize) {
2256 int err2;
Theodore Ts'o622cad12014-04-11 10:35:17 -04002257 loff_t i_size;
Jan Kara4e7ea812013-06-04 13:17:40 -04002258
Theodore Ts'o622cad12014-04-11 10:35:17 -04002259 down_write(&EXT4_I(inode)->i_data_sem);
2260 i_size = i_size_read(inode);
2261 if (disksize > i_size)
2262 disksize = i_size;
2263 if (disksize > EXT4_I(inode)->i_disksize)
2264 EXT4_I(inode)->i_disksize = disksize;
Jan Kara4e7ea812013-06-04 13:17:40 -04002265 err2 = ext4_mark_inode_dirty(handle, inode);
Theodore Ts'o622cad12014-04-11 10:35:17 -04002266 up_write(&EXT4_I(inode)->i_data_sem);
Jan Kara4e7ea812013-06-04 13:17:40 -04002267 if (err2)
2268 ext4_error(inode->i_sb,
2269 "Failed to mark inode %lu dirty",
2270 inode->i_ino);
2271 if (!err)
2272 err = err2;
2273 }
2274 return err;
2275}
2276
2277/*
Jan Karafffb2732013-06-04 13:01:11 -04002278 * Calculate the total number of credits to reserve for one writepages
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002279 * iteration. This is called from ext4_writepages(). We map an extent of
Anatol Pomozov70261f52013-08-28 14:40:12 -04002280 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
Jan Karafffb2732013-06-04 13:01:11 -04002281 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2282 * bpp - 1 blocks in bpp different extents.
2283 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002284static int ext4_da_writepages_trans_blocks(struct inode *inode)
2285{
Jan Karafffb2732013-06-04 13:01:11 -04002286 int bpp = ext4_journal_blocks_per_page(inode);
Mingming Cao525f4ed2008-08-19 22:15:58 -04002287
Jan Karafffb2732013-06-04 13:01:11 -04002288 return ext4_meta_trans_blocks(inode,
2289 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
Mingming Cao525f4ed2008-08-19 22:15:58 -04002290}
Mingming Cao61628a32008-07-11 19:27:31 -04002291
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002292/*
Jan Kara4e7ea812013-06-04 13:17:40 -04002293 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2294 * and underlying extent to map
2295 *
2296 * @mpd - where to look for pages
2297 *
2298 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2299 * IO immediately. When we find a page which isn't mapped we start accumulating
2300 * extent of buffers underlying these pages that needs mapping (formed by
2301 * either delayed or unwritten buffers). We also lock the pages containing
2302 * these buffers. The extent found is returned in @mpd structure (starting at
2303 * mpd->lblk with length mpd->len blocks).
2304 *
2305 * Note that this function can attach bios to one io_end structure which are
2306 * neither logically nor physically contiguous. Although it may seem as an
2307 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2308 * case as we need to track IO to all buffers underlying a page in one io_end.
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002309 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002310static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002311{
Jan Kara4e7ea812013-06-04 13:17:40 -04002312 struct address_space *mapping = mpd->inode->i_mapping;
2313 struct pagevec pvec;
2314 unsigned int nr_pages;
Ming Leiaeac5892013-10-17 18:56:16 -04002315 long left = mpd->wbc->nr_to_write;
Jan Kara4e7ea812013-06-04 13:17:40 -04002316 pgoff_t index = mpd->first_page;
2317 pgoff_t end = mpd->last_page;
2318 int tag;
2319 int i, err = 0;
2320 int blkbits = mpd->inode->i_blkbits;
2321 ext4_lblk_t lblk;
2322 struct buffer_head *head;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002323
Jan Kara4e7ea812013-06-04 13:17:40 -04002324 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002325 tag = PAGECACHE_TAG_TOWRITE;
2326 else
2327 tag = PAGECACHE_TAG_DIRTY;
2328
Jan Kara4e7ea812013-06-04 13:17:40 -04002329 pagevec_init(&pvec, 0);
2330 mpd->map.m_len = 0;
2331 mpd->next_page = index;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002332 while (index <= end) {
Eric Sandeen5b41d922010-10-27 21:30:13 -04002333 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002334 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2335 if (nr_pages == 0)
Jan Kara4e7ea812013-06-04 13:17:40 -04002336 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002337
2338 for (i = 0; i < nr_pages; i++) {
2339 struct page *page = pvec.pages[i];
2340
2341 /*
2342 * At this point, the page may be truncated or
2343 * invalidated (changing page->mapping to NULL), or
2344 * even swizzled back from swapper_space to tmpfs file
2345 * mapping. However, page->index will not change
2346 * because we have a reference on the page.
2347 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002348 if (page->index > end)
2349 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002350
Ming Leiaeac5892013-10-17 18:56:16 -04002351 /*
2352 * Accumulated enough dirty pages? This doesn't apply
2353 * to WB_SYNC_ALL mode. For integrity sync we have to
2354 * keep going because someone may be concurrently
2355 * dirtying pages, and we might have synced a lot of
2356 * newly appeared dirty pages, but have not synced all
2357 * of the old dirty pages.
2358 */
2359 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2360 goto out;
2361
Jan Kara4e7ea812013-06-04 13:17:40 -04002362 /* If we can't merge this page, we are done. */
2363 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2364 goto out;
Theodore Ts'o78aaced2011-02-26 14:09:14 -05002365
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002366 lock_page(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002367 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002368 * If the page is no longer dirty, or its mapping no
2369 * longer corresponds to inode we are writing (which
2370 * means it has been truncated or invalidated), or the
2371 * page is already under writeback and we are not doing
2372 * a data integrity writeback, skip the page
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002373 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002374 if (!PageDirty(page) ||
2375 (PageWriteback(page) &&
Jan Kara4e7ea812013-06-04 13:17:40 -04002376 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002377 unlikely(page->mapping != mapping)) {
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002378 unlock_page(page);
2379 continue;
2380 }
2381
Darrick J. Wong7cb1a532011-05-18 13:53:20 -04002382 wait_on_page_writeback(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002383 BUG_ON(PageWriteback(page));
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002384
Jan Kara4e7ea812013-06-04 13:17:40 -04002385 if (mpd->map.m_len == 0)
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002386 mpd->first_page = page->index;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002387 mpd->next_page = page->index + 1;
Jan Karaf8bec372013-01-28 12:55:08 -05002388 /* Add all dirty buffers to mpd */
Jan Kara4e7ea812013-06-04 13:17:40 -04002389 lblk = ((ext4_lblk_t)page->index) <<
2390 (PAGE_CACHE_SHIFT - blkbits);
Jan Karaf8bec372013-01-28 12:55:08 -05002391 head = page_buffers(page);
Jan Kara5f1132b2013-08-17 10:02:33 -04002392 err = mpage_process_page_bufs(mpd, head, head, lblk);
2393 if (err <= 0)
Jan Kara4e7ea812013-06-04 13:17:40 -04002394 goto out;
Jan Kara5f1132b2013-08-17 10:02:33 -04002395 err = 0;
Ming Leiaeac5892013-10-17 18:56:16 -04002396 left--;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002397 }
2398 pagevec_release(&pvec);
2399 cond_resched();
2400 }
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002401 return 0;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002402out:
2403 pagevec_release(&pvec);
Jan Kara4e7ea812013-06-04 13:17:40 -04002404 return err;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002405}
2406
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002407static int __writepage(struct page *page, struct writeback_control *wbc,
2408 void *data)
2409{
2410 struct address_space *mapping = data;
2411 int ret = ext4_writepage(page, wbc);
2412 mapping_set_error(mapping, ret);
2413 return ret;
2414}
2415
2416static int ext4_writepages(struct address_space *mapping,
2417 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002418{
Jan Kara4e7ea812013-06-04 13:17:40 -04002419 pgoff_t writeback_index = 0;
2420 long nr_to_write = wbc->nr_to_write;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002421 int range_whole = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002422 int cycled = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002423 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002424 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002425 struct inode *inode = mapping->host;
Jan Kara6b523df2013-06-04 13:21:11 -04002426 int needed_blocks, rsv_blocks = 0, ret = 0;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002427 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Jan Kara4e7ea812013-06-04 13:17:40 -04002428 bool done;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002429 struct blk_plug plug;
Theodore Ts'ocb530542013-07-01 08:12:40 -04002430 bool give_up_on_write = false;
Mingming Cao61628a32008-07-11 19:27:31 -04002431
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002432 trace_ext4_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002433
Mingming Cao61628a32008-07-11 19:27:31 -04002434 /*
2435 * No pages to write? This is mainly a kludge to avoid starting
2436 * a transaction for special inodes like journal inode on last iput()
2437 * because that could violate lock ordering on umount
2438 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002439 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Ming Leibbf023c2013-10-30 07:27:16 -04002440 goto out_writepages;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002441
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002442 if (ext4_should_journal_data(inode)) {
2443 struct blk_plug plug;
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002444
2445 blk_start_plug(&plug);
2446 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2447 blk_finish_plug(&plug);
Ming Leibbf023c2013-10-30 07:27:16 -04002448 goto out_writepages;
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002449 }
2450
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002451 /*
2452 * If the filesystem has aborted, it is read-only, so return
2453 * right away instead of dumping stack traces later on that
2454 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002455 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002456 * the latter could be true if the filesystem is mounted
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002457 * read-only, and in that case, ext4_writepages should
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002458 * *never* be called, so if that ever happens, we would want
2459 * the stack trace.
2460 */
Ming Leibbf023c2013-10-30 07:27:16 -04002461 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2462 ret = -EROFS;
2463 goto out_writepages;
2464 }
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002465
Jan Kara6b523df2013-06-04 13:21:11 -04002466 if (ext4_should_dioread_nolock(inode)) {
2467 /*
Anatol Pomozov70261f52013-08-28 14:40:12 -04002468 * We may need to convert up to one extent per block in
Jan Kara6b523df2013-06-04 13:21:11 -04002469 * the page and we may dirty the inode.
2470 */
2471 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2472 }
2473
Jan Kara4e7ea812013-06-04 13:17:40 -04002474 /*
2475 * If we have inline data and arrive here, it means that
2476 * we will soon create the block for the 1st page, so
2477 * we'd better clear the inline data here.
2478 */
2479 if (ext4_has_inline_data(inode)) {
2480 /* Just inode will be modified... */
2481 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2482 if (IS_ERR(handle)) {
2483 ret = PTR_ERR(handle);
2484 goto out_writepages;
2485 }
2486 BUG_ON(ext4_test_inode_state(inode,
2487 EXT4_STATE_MAY_INLINE_DATA));
2488 ext4_destroy_inline_data(handle, inode);
2489 ext4_journal_stop(handle);
2490 }
2491
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002492 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2493 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002494
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002495 if (wbc->range_cyclic) {
Jan Kara4e7ea812013-06-04 13:17:40 -04002496 writeback_index = mapping->writeback_index;
2497 if (writeback_index)
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002498 cycled = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002499 mpd.first_page = writeback_index;
2500 mpd.last_page = -1;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002501 } else {
Jan Kara4e7ea812013-06-04 13:17:40 -04002502 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2503 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002504 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002505
Jan Kara4e7ea812013-06-04 13:17:40 -04002506 mpd.inode = inode;
2507 mpd.wbc = wbc;
2508 ext4_io_submit_init(&mpd.io_submit, wbc);
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002509retry:
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002510 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Jan Kara4e7ea812013-06-04 13:17:40 -04002511 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2512 done = false;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002513 blk_start_plug(&plug);
Jan Kara4e7ea812013-06-04 13:17:40 -04002514 while (!done && mpd.first_page <= mpd.last_page) {
2515 /* For each extent of pages we use new io_end */
2516 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2517 if (!mpd.io_submit.io_end) {
2518 ret = -ENOMEM;
2519 break;
2520 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002521
2522 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002523 * We have two constraints: We find one extent to map and we
2524 * must always write out whole page (makes a difference when
2525 * blocksize < pagesize) so that we don't block on IO when we
2526 * try to write out the rest of the page. Journalled mode is
2527 * not supported by delalloc.
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002528 */
2529 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002530 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002531
Jan Kara4e7ea812013-06-04 13:17:40 -04002532 /* start a new transaction */
Jan Kara6b523df2013-06-04 13:21:11 -04002533 handle = ext4_journal_start_with_reserve(inode,
2534 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
Mingming Cao61628a32008-07-11 19:27:31 -04002535 if (IS_ERR(handle)) {
2536 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002537 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04002538 "%ld pages, ino %lu; err %d", __func__,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002539 wbc->nr_to_write, inode->i_ino, ret);
Jan Kara4e7ea812013-06-04 13:17:40 -04002540 /* Release allocated io_end */
2541 ext4_put_io_end(mpd.io_submit.io_end);
2542 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002543 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002544
Jan Kara4e7ea812013-06-04 13:17:40 -04002545 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2546 ret = mpage_prepare_extent_to_map(&mpd);
2547 if (!ret) {
2548 if (mpd.map.m_len)
Theodore Ts'ocb530542013-07-01 08:12:40 -04002549 ret = mpage_map_and_submit_extent(handle, &mpd,
2550 &give_up_on_write);
Jan Kara4e7ea812013-06-04 13:17:40 -04002551 else {
2552 /*
2553 * We scanned the whole range (or exhausted
2554 * nr_to_write), submitted what was mapped and
2555 * didn't find anything needing mapping. We are
2556 * done.
2557 */
2558 done = true;
2559 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002560 }
Mingming Cao61628a32008-07-11 19:27:31 -04002561 ext4_journal_stop(handle);
Jan Kara4e7ea812013-06-04 13:17:40 -04002562 /* Submit prepared bio */
2563 ext4_io_submit(&mpd.io_submit);
2564 /* Unlock pages we didn't use */
Theodore Ts'ocb530542013-07-01 08:12:40 -04002565 mpage_release_unused_pages(&mpd, give_up_on_write);
Jan Kara4e7ea812013-06-04 13:17:40 -04002566 /* Drop our io_end reference we got from init */
2567 ext4_put_io_end(mpd.io_submit.io_end);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002568
Jan Kara4e7ea812013-06-04 13:17:40 -04002569 if (ret == -ENOSPC && sbi->s_journal) {
2570 /*
2571 * Commit the transaction which would
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002572 * free blocks released in the transaction
2573 * and try again
2574 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002575 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002576 ret = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002577 continue;
2578 }
2579 /* Fatal error - ENOMEM, EIO... */
2580 if (ret)
Mingming Cao61628a32008-07-11 19:27:31 -04002581 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002582 }
Shaohua Li1bce63d2011-10-18 10:55:51 -04002583 blk_finish_plug(&plug);
Jan Kara9c12a832013-09-16 08:24:26 -04002584 if (!ret && !cycled && wbc->nr_to_write > 0) {
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002585 cycled = 1;
Jan Kara4e7ea812013-06-04 13:17:40 -04002586 mpd.last_page = writeback_index - 1;
2587 mpd.first_page = 0;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002588 goto retry;
2589 }
Mingming Cao61628a32008-07-11 19:27:31 -04002590
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002591 /* Update index */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002592 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2593 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002594 * Set the writeback_index so that range_cyclic
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002595 * mode will write it back later
2596 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002597 mapping->writeback_index = mpd.first_page;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002598
Mingming Cao61628a32008-07-11 19:27:31 -04002599out_writepages:
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002600 trace_ext4_writepages_result(inode, wbc, ret,
2601 nr_to_write - wbc->nr_to_write);
Mingming Cao61628a32008-07-11 19:27:31 -04002602 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002603}
2604
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002605static int ext4_nonda_switch(struct super_block *sb)
2606{
Eric Whitney5c1ff332013-04-09 09:27:31 -04002607 s64 free_clusters, dirty_clusters;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002608 struct ext4_sb_info *sbi = EXT4_SB(sb);
2609
2610 /*
2611 * switch to non delalloc mode if we are running low
2612 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08002613 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002614 * accumulated on each CPU without updating global counters
2615 * Delalloc need an accurate free block accounting. So switch
2616 * to non delalloc when we are near to error range.
2617 */
Eric Whitney5c1ff332013-04-09 09:27:31 -04002618 free_clusters =
2619 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2620 dirty_clusters =
2621 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002622 /*
2623 * Start pushing delalloc when 1/2 of free blocks are dirty.
2624 */
Eric Whitney5c1ff332013-04-09 09:27:31 -04002625 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
Miao Xie10ee27a2013-01-10 13:47:57 +08002626 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002627
Eric Whitney5c1ff332013-04-09 09:27:31 -04002628 if (2 * free_clusters < 3 * dirty_clusters ||
2629 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002630 /*
Eric Sandeenc8afb442009-12-23 07:58:12 -05002631 * free block count is less than 150% of dirty blocks
2632 * or free blocks is less than watermark
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002633 */
2634 return 1;
2635 }
2636 return 0;
2637}
2638
Alex Tomas64769242008-07-11 19:27:31 -04002639static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002640 loff_t pos, unsigned len, unsigned flags,
2641 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002642{
Eric Sandeen72b8ab92010-05-16 11:00:00 -04002643 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002644 struct page *page;
2645 pgoff_t index;
Alex Tomas64769242008-07-11 19:27:31 -04002646 struct inode *inode = mapping->host;
2647 handle_t *handle;
2648
2649 index = pos >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002650
2651 if (ext4_nonda_switch(inode->i_sb)) {
2652 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2653 return ext4_write_begin(file, mapping, pos,
2654 len, flags, pagep, fsdata);
2655 }
2656 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002657 trace_ext4_da_write_begin(inode, pos, len, flags);
Tao Ma9c3569b2012-12-10 14:05:57 -05002658
2659 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2660 ret = ext4_da_write_inline_data_begin(mapping, inode,
2661 pos, len, flags,
2662 pagep, fsdata);
2663 if (ret < 0)
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002664 return ret;
2665 if (ret == 1)
2666 return 0;
Tao Ma9c3569b2012-12-10 14:05:57 -05002667 }
2668
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002669 /*
2670 * grab_cache_page_write_begin() can take a long time if the
2671 * system is thrashing due to memory pressure, or if the page
2672 * is being written back. So grab it first before we start
2673 * the transaction handle. This also allows us to allocate
2674 * the page (if needed) without using GFP_NOFS.
2675 */
2676retry_grab:
2677 page = grab_cache_page_write_begin(mapping, index, flags);
2678 if (!page)
2679 return -ENOMEM;
2680 unlock_page(page);
2681
Alex Tomas64769242008-07-11 19:27:31 -04002682 /*
2683 * With delayed allocation, we don't log the i_disksize update
2684 * if there is delayed block allocation. But we still need
2685 * to journalling the i_disksize update if writes to the end
2686 * of file which has an already mapped buffer.
2687 */
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002688retry_journal:
Theodore Ts'o9924a922013-02-08 21:59:22 -05002689 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
Alex Tomas64769242008-07-11 19:27:31 -04002690 if (IS_ERR(handle)) {
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002691 page_cache_release(page);
2692 return PTR_ERR(handle);
Alex Tomas64769242008-07-11 19:27:31 -04002693 }
2694
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002695 lock_page(page);
2696 if (page->mapping != mapping) {
2697 /* The page got truncated from under us */
2698 unlock_page(page);
2699 page_cache_release(page);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002700 ext4_journal_stop(handle);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002701 goto retry_grab;
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002702 }
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002703 /* In case writeback began while the page was unlocked */
Dmitry Monakhov7afe5aa2013-08-28 14:30:47 -04002704 wait_for_stable_page(page);
Alex Tomas64769242008-07-11 19:27:31 -04002705
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002706 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04002707 if (ret < 0) {
2708 unlock_page(page);
2709 ext4_journal_stop(handle);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04002710 /*
2711 * block_write_begin may have instantiated a few blocks
2712 * outside i_size. Trim these off again. Don't need
2713 * i_size_read because we hold i_mutex.
2714 */
2715 if (pos + len > inode->i_size)
Jan Karab9a42072009-12-08 21:24:33 -05002716 ext4_truncate_failed_write(inode);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002717
2718 if (ret == -ENOSPC &&
2719 ext4_should_retry_alloc(inode->i_sb, &retries))
2720 goto retry_journal;
2721
2722 page_cache_release(page);
2723 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002724 }
2725
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002726 *pagep = page;
Alex Tomas64769242008-07-11 19:27:31 -04002727 return ret;
2728}
2729
Mingming Cao632eaea2008-07-11 19:27:31 -04002730/*
2731 * Check if we should update i_disksize
2732 * when write to the end of file but not require block allocation
2733 */
2734static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002735 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04002736{
2737 struct buffer_head *bh;
2738 struct inode *inode = page->mapping->host;
2739 unsigned int idx;
2740 int i;
2741
2742 bh = page_buffers(page);
2743 idx = offset >> inode->i_blkbits;
2744
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002745 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04002746 bh = bh->b_this_page;
2747
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002748 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04002749 return 0;
2750 return 1;
2751}
2752
Alex Tomas64769242008-07-11 19:27:31 -04002753static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002754 struct address_space *mapping,
2755 loff_t pos, unsigned len, unsigned copied,
2756 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002757{
2758 struct inode *inode = mapping->host;
2759 int ret = 0, ret2;
2760 handle_t *handle = ext4_journal_current_handle();
2761 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002762 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002763 int write_mode = (int)(unsigned long)fsdata;
2764
Theodore Ts'o74d553a2013-04-03 12:39:17 -04002765 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2766 return ext4_write_end(file, mapping, pos,
2767 len, copied, page, fsdata);
Mingming Cao632eaea2008-07-11 19:27:31 -04002768
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002769 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04002770 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002771 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04002772
2773 /*
2774 * generic_write_end() will run mark_inode_dirty() if i_size
2775 * changes. So let's piggyback the i_disksize mark_inode_dirty
2776 * into that.
2777 */
Alex Tomas64769242008-07-11 19:27:31 -04002778 new_i_size = pos + copied;
Andrea Arcangeliea51d132011-12-13 21:41:15 -05002779 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
Tao Ma9c3569b2012-12-10 14:05:57 -05002780 if (ext4_has_inline_data(inode) ||
2781 ext4_da_should_update_i_disksize(page, end)) {
Mingming Cao632eaea2008-07-11 19:27:31 -04002782 down_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'of3b59292012-11-15 23:08:57 -05002783 if (new_i_size > EXT4_I(inode)->i_disksize)
Mingming Cao632eaea2008-07-11 19:27:31 -04002784 EXT4_I(inode)->i_disksize = new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002785 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002786 /* We need to mark inode dirty even if
2787 * new_i_size is less that inode->i_size
2788 * bu greater than i_disksize.(hint delalloc)
2789 */
2790 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04002791 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002792 }
Tao Ma9c3569b2012-12-10 14:05:57 -05002793
2794 if (write_mode != CONVERT_INLINE_DATA &&
2795 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2796 ext4_has_inline_data(inode))
2797 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2798 page);
2799 else
2800 ret2 = generic_write_end(file, mapping, pos, len, copied,
Alex Tomas64769242008-07-11 19:27:31 -04002801 page, fsdata);
Tao Ma9c3569b2012-12-10 14:05:57 -05002802
Alex Tomas64769242008-07-11 19:27:31 -04002803 copied = ret2;
2804 if (ret2 < 0)
2805 ret = ret2;
2806 ret2 = ext4_journal_stop(handle);
2807 if (!ret)
2808 ret = ret2;
2809
2810 return ret ? ret : copied;
2811}
2812
Lukas Czernerd47992f2013-05-21 23:17:23 -04002813static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2814 unsigned int length)
Alex Tomas64769242008-07-11 19:27:31 -04002815{
Alex Tomas64769242008-07-11 19:27:31 -04002816 /*
2817 * Drop reserved blocks
2818 */
2819 BUG_ON(!PageLocked(page));
2820 if (!page_has_buffers(page))
2821 goto out;
2822
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002823 ext4_da_page_release_reservation(page, offset, length);
Alex Tomas64769242008-07-11 19:27:31 -04002824
2825out:
Lukas Czernerd47992f2013-05-21 23:17:23 -04002826 ext4_invalidatepage(page, offset, length);
Alex Tomas64769242008-07-11 19:27:31 -04002827
2828 return;
2829}
2830
Theodore Ts'occd25062009-02-26 01:04:07 -05002831/*
2832 * Force all delayed allocation blocks to be allocated for a given inode.
2833 */
2834int ext4_alloc_da_blocks(struct inode *inode)
2835{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04002836 trace_ext4_alloc_da_blocks(inode);
2837
Theodore Ts'occd25062009-02-26 01:04:07 -05002838 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2839 !EXT4_I(inode)->i_reserved_meta_blocks)
2840 return 0;
2841
2842 /*
2843 * We do something simple for now. The filemap_flush() will
2844 * also start triggering a write of the data blocks, which is
2845 * not strictly speaking necessary (and for users of
2846 * laptop_mode, not even desirable). However, to do otherwise
2847 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002848 *
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002849 * ext4_writepages() ->
Theodore Ts'occd25062009-02-26 01:04:07 -05002850 * write_cache_pages() ---> (via passed in callback function)
2851 * __mpage_da_writepage() -->
2852 * mpage_add_bh_to_extent()
2853 * mpage_da_map_blocks()
2854 *
2855 * The problem is that write_cache_pages(), located in
2856 * mm/page-writeback.c, marks pages clean in preparation for
2857 * doing I/O, which is not desirable if we're not planning on
2858 * doing I/O at all.
2859 *
2860 * We could call write_cache_pages(), and then redirty all of
Wu Fengguang380cf092010-11-11 19:23:29 +08002861 * the pages by calling redirty_page_for_writepage() but that
Theodore Ts'occd25062009-02-26 01:04:07 -05002862 * would be ugly in the extreme. So instead we would need to
2863 * replicate parts of the code in the above functions,
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002864 * simplifying them because we wouldn't actually intend to
Theodore Ts'occd25062009-02-26 01:04:07 -05002865 * write out the pages, but rather only collect contiguous
2866 * logical block extents, call the multi-block allocator, and
2867 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002868 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002869 * For now, though, we'll cheat by calling filemap_flush(),
2870 * which will map the blocks, and start the I/O, but not
2871 * actually wait for the I/O to complete.
2872 */
2873 return filemap_flush(inode->i_mapping);
2874}
Alex Tomas64769242008-07-11 19:27:31 -04002875
2876/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002877 * bmap() is special. It gets used by applications such as lilo and by
2878 * the swapper to find the on-disk block of a specific piece of data.
2879 *
2880 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07002881 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002882 * filesystem and enables swap, then they may get a nasty shock when the
2883 * data getting swapped to that swapfile suddenly gets overwritten by
2884 * the original zero's written out previously to the journal and
2885 * awaiting writeback in the kernel's buffer cache.
2886 *
2887 * So, if we see any bmap calls here on a modified, data-journaled file,
2888 * take extra steps to flush any blocks which might be in the cache.
2889 */
Mingming Cao617ba132006-10-11 01:20:53 -07002890static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002891{
2892 struct inode *inode = mapping->host;
2893 journal_t *journal;
2894 int err;
2895
Tao Ma46c7f252012-12-10 14:04:52 -05002896 /*
2897 * We can get here for an inline file via the FIBMAP ioctl
2898 */
2899 if (ext4_has_inline_data(inode))
2900 return 0;
2901
Alex Tomas64769242008-07-11 19:27:31 -04002902 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2903 test_opt(inode->i_sb, DELALLOC)) {
2904 /*
2905 * With delalloc we want to sync the file
2906 * so that we can make sure we allocate
2907 * blocks for file
2908 */
2909 filemap_write_and_wait(mapping);
2910 }
2911
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002912 if (EXT4_JOURNAL(inode) &&
2913 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002914 /*
2915 * This is a REALLY heavyweight approach, but the use of
2916 * bmap on dirty files is expected to be extremely rare:
2917 * only if we run lilo or swapon on a freshly made file
2918 * do we expect this to happen.
2919 *
2920 * (bmap requires CAP_SYS_RAWIO so this does not
2921 * represent an unprivileged user DOS attack --- we'd be
2922 * in trouble if mortal users could trigger this path at
2923 * will.)
2924 *
Mingming Cao617ba132006-10-11 01:20:53 -07002925 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002926 * regular files. If somebody wants to bmap a directory
2927 * or symlink and gets confused because the buffer
2928 * hasn't yet been flushed to disk, they deserve
2929 * everything they get.
2930 */
2931
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002932 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
Mingming Cao617ba132006-10-11 01:20:53 -07002933 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07002934 jbd2_journal_lock_updates(journal);
2935 err = jbd2_journal_flush(journal);
2936 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002937
2938 if (err)
2939 return 0;
2940 }
2941
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002942 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002943}
2944
Mingming Cao617ba132006-10-11 01:20:53 -07002945static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002946{
Tao Ma46c7f252012-12-10 14:04:52 -05002947 int ret = -EAGAIN;
2948 struct inode *inode = page->mapping->host;
2949
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002950 trace_ext4_readpage(page);
Tao Ma46c7f252012-12-10 14:04:52 -05002951
2952 if (ext4_has_inline_data(inode))
2953 ret = ext4_readpage_inline(inode, page);
2954
2955 if (ret == -EAGAIN)
2956 return mpage_readpage(page, ext4_get_block);
2957
2958 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002959}
2960
2961static int
Mingming Cao617ba132006-10-11 01:20:53 -07002962ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002963 struct list_head *pages, unsigned nr_pages)
2964{
Tao Ma46c7f252012-12-10 14:04:52 -05002965 struct inode *inode = mapping->host;
2966
2967 /* If the file has inline data, no need to do readpages. */
2968 if (ext4_has_inline_data(inode))
2969 return 0;
2970
Mingming Cao617ba132006-10-11 01:20:53 -07002971 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002972}
2973
Lukas Czernerd47992f2013-05-21 23:17:23 -04002974static void ext4_invalidatepage(struct page *page, unsigned int offset,
2975 unsigned int length)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002976{
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002977 trace_ext4_invalidatepage(page, offset, length);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002978
Jan Kara4520fb32012-12-25 13:28:54 -05002979 /* No journalling happens on data buffers when this function is used */
2980 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2981
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002982 block_invalidatepage(page, offset, length);
Jan Kara4520fb32012-12-25 13:28:54 -05002983}
2984
Jan Kara53e87262012-12-25 13:29:52 -05002985static int __ext4_journalled_invalidatepage(struct page *page,
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002986 unsigned int offset,
2987 unsigned int length)
Jan Kara4520fb32012-12-25 13:28:54 -05002988{
2989 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2990
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002991 trace_ext4_journalled_invalidatepage(page, offset, length);
Jan Kara4520fb32012-12-25 13:28:54 -05002992
Jiaying Zhang744692d2010-03-04 16:14:02 -05002993 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002994 * If it's a full truncate we just forget about the pending dirtying
2995 */
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002996 if (offset == 0 && length == PAGE_CACHE_SIZE)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002997 ClearPageChecked(page);
2998
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002999 return jbd2_journal_invalidatepage(journal, page, offset, length);
Jan Kara53e87262012-12-25 13:29:52 -05003000}
3001
3002/* Wrapper for aops... */
3003static void ext4_journalled_invalidatepage(struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -04003004 unsigned int offset,
3005 unsigned int length)
Jan Kara53e87262012-12-25 13:29:52 -05003006{
Lukas Czernerca99fdd2013-05-21 23:25:01 -04003007 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003008}
3009
Mingming Cao617ba132006-10-11 01:20:53 -07003010static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003011{
Mingming Cao617ba132006-10-11 01:20:53 -07003012 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003013
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003014 trace_ext4_releasepage(page);
3015
Jan Karae1c36592013-03-10 22:19:00 -04003016 /* Page has dirty journalled data -> cannot release */
3017 if (PageChecked(page))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003018 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05003019 if (journal)
3020 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3021 else
3022 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003023}
3024
3025/*
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003026 * ext4_get_block used when preparing for a DIO write or buffer write.
3027 * We allocate an uinitialized extent if blocks haven't been allocated.
3028 * The extent will be converted to initialized after the IO is complete.
3029 */
Tao Maf19d5872012-12-10 14:05:51 -05003030int ext4_get_block_write(struct inode *inode, sector_t iblock,
Mingming Cao4c0425f2009-09-28 15:48:41 -04003031 struct buffer_head *bh_result, int create)
3032{
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003033 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003034 inode->i_ino, create);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003035 return _ext4_get_block(inode, iblock, bh_result,
3036 EXT4_GET_BLOCKS_IO_CREATE_EXT);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003037}
3038
Zheng Liu729f52c2012-07-09 16:29:29 -04003039static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05003040 struct buffer_head *bh_result, int create)
Zheng Liu729f52c2012-07-09 16:29:29 -04003041{
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05003042 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3043 inode->i_ino, create);
3044 return _ext4_get_block(inode, iblock, bh_result,
3045 EXT4_GET_BLOCKS_NO_LOCK);
Zheng Liu729f52c2012-07-09 16:29:29 -04003046}
3047
Mingming Cao4c0425f2009-09-28 15:48:41 -04003048static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003049 ssize_t size, void *private)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003050{
3051 ext4_io_end_t *io_end = iocb->private;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003052
Jan Kara97a851e2013-06-04 11:58:58 -04003053 /* if not async direct IO just return */
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003054 if (!io_end)
Jan Kara97a851e2013-06-04 11:58:58 -04003055 return;
Mingming4b70df12009-11-03 14:44:54 -05003056
Zheng Liu88635ca2011-12-28 19:00:25 -05003057 ext_debug("ext4_end_io_dio(): io_end 0x%p "
Joe Perchesace36ad2012-03-19 23:11:43 -04003058 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003059 iocb->private, io_end->inode->i_ino, iocb, offset,
3060 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003061
Theodore Ts'ob5a7e972011-12-12 10:53:02 -05003062 iocb->private = NULL;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003063 io_end->offset = offset;
3064 io_end->size = size;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003065 ext4_put_io_end(io_end);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003066}
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003067
Mingming Cao4c0425f2009-09-28 15:48:41 -04003068/*
3069 * For ext4 extent files, ext4 will do direct-io write to holes,
3070 * preallocated extents, and those write extend the file, no need to
3071 * fall back to buffered IO.
3072 *
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003073 * For holes, we fallocate those blocks, mark them as uninitialized
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003074 * If those blocks were preallocated, we mark sure they are split, but
Uwe Kleine-Königb5950762010-11-01 15:38:34 -04003075 * still keep the range to write as uninitialized.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003076 *
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003077 * The unwritten extents will be converted to written when DIO is completed.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003078 * For async direct IO, since the IO may still pending when return, we
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003079 * set up an end_io call back function, which will do the conversion
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003080 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003081 *
3082 * If the O_DIRECT write will extend the file then add this inode to the
3083 * orphan list. So recovery will truncate it back to the original size
3084 * if the machine crashes during the write.
3085 *
3086 */
3087static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3088 const struct iovec *iov, loff_t offset,
3089 unsigned long nr_segs)
3090{
3091 struct file *file = iocb->ki_filp;
3092 struct inode *inode = file->f_mapping->host;
3093 ssize_t ret;
3094 size_t count = iov_length(iov, nr_segs);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003095 int overwrite = 0;
3096 get_block_t *get_block_func = NULL;
3097 int dio_flags = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003098 loff_t final_size = offset + count;
Jan Kara97a851e2013-06-04 11:58:58 -04003099 ext4_io_end_t *io_end = NULL;
Zheng Liu729f52c2012-07-09 16:29:29 -04003100
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003101 /* Use the old path for reads and writes beyond i_size. */
3102 if (rw != WRITE || final_size > inode->i_size)
3103 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003104
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003105 BUG_ON(iocb->private == NULL);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003106
Jan Karae8340392013-06-04 14:27:38 -04003107 /*
3108 * Make all waiters for direct IO properly wait also for extent
3109 * conversion. This also disallows race between truncate() and
3110 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3111 */
3112 if (rw == WRITE)
3113 atomic_inc(&inode->i_dio_count);
3114
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003115 /* If we do a overwrite dio, i_mutex locking can be released */
3116 overwrite = *((int *)iocb->private);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003117
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003118 if (overwrite) {
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003119 down_read(&EXT4_I(inode)->i_data_sem);
3120 mutex_unlock(&inode->i_mutex);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003121 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003122
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003123 /*
3124 * We could direct write to holes and fallocate.
3125 *
3126 * Allocated blocks to fill the hole are marked as
3127 * uninitialized to prevent parallel buffered read to expose
3128 * the stale data before DIO complete the data IO.
3129 *
3130 * As to previously fallocated extents, ext4 get_block will
3131 * just simply mark the buffer mapped but still keep the
3132 * extents uninitialized.
3133 *
3134 * For non AIO case, we will convert those unwritten extents
3135 * to written after return back from blockdev_direct_IO.
3136 *
3137 * For async DIO, the conversion needs to be deferred when the
3138 * IO is completed. The ext4 end_io callback function will be
3139 * called to take care of the conversion work. Here for async
3140 * case, we allocate an io_end structure to hook to the iocb.
3141 */
3142 iocb->private = NULL;
3143 ext4_inode_aio_set(inode, NULL);
3144 if (!is_sync_kiocb(iocb)) {
Jan Kara97a851e2013-06-04 11:58:58 -04003145 io_end = ext4_init_io_end(inode, GFP_NOFS);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003146 if (!io_end) {
3147 ret = -ENOMEM;
3148 goto retake_lock;
3149 }
Jan Kara97a851e2013-06-04 11:58:58 -04003150 /*
3151 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3152 */
3153 iocb->private = ext4_get_io_end(io_end);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003154 /*
3155 * we save the io structure for current async direct
3156 * IO, so that later ext4_map_blocks() could flag the
3157 * io structure whether there is a unwritten extents
3158 * needs to be converted when IO is completed.
3159 */
3160 ext4_inode_aio_set(inode, io_end);
3161 }
3162
3163 if (overwrite) {
3164 get_block_func = ext4_get_block_write_nolock;
3165 } else {
3166 get_block_func = ext4_get_block_write;
3167 dio_flags = DIO_LOCKING;
3168 }
3169 ret = __blockdev_direct_IO(rw, iocb, inode,
3170 inode->i_sb->s_bdev, iov,
3171 offset, nr_segs,
3172 get_block_func,
3173 ext4_end_io_dio,
3174 NULL,
3175 dio_flags);
3176
Theodore Ts'oa5499842013-05-11 19:07:42 -04003177 /*
Jan Kara97a851e2013-06-04 11:58:58 -04003178 * Put our reference to io_end. This can free the io_end structure e.g.
3179 * in sync IO case or in case of error. It can even perform extent
3180 * conversion if all bios we submitted finished before we got here.
3181 * Note that in that case iocb->private can be already set to NULL
3182 * here.
Theodore Ts'oa5499842013-05-11 19:07:42 -04003183 */
Jan Kara97a851e2013-06-04 11:58:58 -04003184 if (io_end) {
3185 ext4_inode_aio_set(inode, NULL);
3186 ext4_put_io_end(io_end);
3187 /*
3188 * When no IO was submitted ext4_end_io_dio() was not
3189 * called so we have to put iocb's reference.
3190 */
3191 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3192 WARN_ON(iocb->private != io_end);
3193 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
Jan Kara97a851e2013-06-04 11:58:58 -04003194 ext4_put_io_end(io_end);
3195 iocb->private = NULL;
3196 }
3197 }
3198 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003199 EXT4_STATE_DIO_UNWRITTEN)) {
3200 int err;
3201 /*
3202 * for non AIO case, since the IO is already
3203 * completed, we could do the conversion right here
3204 */
Jan Kara6b523df2013-06-04 13:21:11 -04003205 err = ext4_convert_unwritten_extents(NULL, inode,
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003206 offset, ret);
3207 if (err < 0)
3208 ret = err;
3209 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3210 }
3211
3212retake_lock:
Jan Karae8340392013-06-04 14:27:38 -04003213 if (rw == WRITE)
3214 inode_dio_done(inode);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003215 /* take i_mutex locking again if we do a ovewrite dio */
3216 if (overwrite) {
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003217 up_read(&EXT4_I(inode)->i_data_sem);
3218 mutex_lock(&inode->i_mutex);
3219 }
3220
3221 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003222}
3223
3224static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05003225 struct iov_iter *iter, loff_t offset)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003226{
3227 struct file *file = iocb->ki_filp;
3228 struct inode *inode = file->f_mapping->host;
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003229 ssize_t ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003230
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003231 /*
3232 * If we are doing data journalling we don't support O_DIRECT
3233 */
3234 if (ext4_should_journal_data(inode))
3235 return 0;
3236
Tao Ma46c7f252012-12-10 14:04:52 -05003237 /* Let buffer I/O handle the inline data case. */
3238 if (ext4_has_inline_data(inode))
3239 return 0;
3240
Al Virod8d3d942014-03-04 21:27:34 -05003241 trace_ext4_direct_IO_enter(inode, offset, iov_length(iter->iov, iter->nr_segs), rw);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003242 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Al Virod8d3d942014-03-04 21:27:34 -05003243 ret = ext4_ext_direct_IO(rw, iocb, iter->iov, offset, iter->nr_segs);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003244 else
Al Virod8d3d942014-03-04 21:27:34 -05003245 ret = ext4_ind_direct_IO(rw, iocb, iter->iov, offset, iter->nr_segs);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003246 trace_ext4_direct_IO_exit(inode, offset,
Al Virod8d3d942014-03-04 21:27:34 -05003247 iov_length(iter->iov, iter->nr_segs), rw, ret);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003248 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003249}
3250
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003251/*
Mingming Cao617ba132006-10-11 01:20:53 -07003252 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003253 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3254 * much here because ->set_page_dirty is called under VFS locks. The page is
3255 * not necessarily locked.
3256 *
3257 * We cannot just dirty the page and leave attached buffers clean, because the
3258 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3259 * or jbddirty because all the journalling code will explode.
3260 *
3261 * So what we do is to mark the page "pending dirty" and next time writepage
3262 * is called, propagate that into the buffers appropriately.
3263 */
Mingming Cao617ba132006-10-11 01:20:53 -07003264static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003265{
3266 SetPageChecked(page);
3267 return __set_page_dirty_nobuffers(page);
3268}
3269
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003270static const struct address_space_operations ext4_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003271 .readpage = ext4_readpage,
3272 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003273 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003274 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003275 .write_begin = ext4_write_begin,
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003276 .write_end = ext4_write_end,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003277 .bmap = ext4_bmap,
3278 .invalidatepage = ext4_invalidatepage,
3279 .releasepage = ext4_releasepage,
3280 .direct_IO = ext4_direct_IO,
3281 .migratepage = buffer_migrate_page,
3282 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003283 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003284};
3285
Mingming Cao617ba132006-10-11 01:20:53 -07003286static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003287 .readpage = ext4_readpage,
3288 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003289 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003290 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003291 .write_begin = ext4_write_begin,
3292 .write_end = ext4_journalled_write_end,
3293 .set_page_dirty = ext4_journalled_set_page_dirty,
3294 .bmap = ext4_bmap,
Jan Kara4520fb32012-12-25 13:28:54 -05003295 .invalidatepage = ext4_journalled_invalidatepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003296 .releasepage = ext4_releasepage,
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003297 .direct_IO = ext4_direct_IO,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003298 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003299 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003300};
3301
Alex Tomas64769242008-07-11 19:27:31 -04003302static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003303 .readpage = ext4_readpage,
3304 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003305 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003306 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003307 .write_begin = ext4_da_write_begin,
3308 .write_end = ext4_da_write_end,
3309 .bmap = ext4_bmap,
3310 .invalidatepage = ext4_da_invalidatepage,
3311 .releasepage = ext4_releasepage,
3312 .direct_IO = ext4_direct_IO,
3313 .migratepage = buffer_migrate_page,
3314 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003315 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003316};
3317
Mingming Cao617ba132006-10-11 01:20:53 -07003318void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003319{
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003320 switch (ext4_inode_journal_mode(inode)) {
3321 case EXT4_INODE_ORDERED_DATA_MODE:
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003322 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003323 break;
3324 case EXT4_INODE_WRITEBACK_DATA_MODE:
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003325 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003326 break;
3327 case EXT4_INODE_JOURNAL_DATA_MODE:
Mingming Cao617ba132006-10-11 01:20:53 -07003328 inode->i_mapping->a_ops = &ext4_journalled_aops;
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003329 return;
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003330 default:
3331 BUG();
3332 }
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003333 if (test_opt(inode->i_sb, DELALLOC))
3334 inode->i_mapping->a_ops = &ext4_da_aops;
3335 else
3336 inode->i_mapping->a_ops = &ext4_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003337}
3338
Lukas Czernerd863dc32013-05-27 23:32:35 -04003339/*
Lukas Czernerd863dc32013-05-27 23:32:35 -04003340 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3341 * starting from file offset 'from'. The range to be zero'd must
3342 * be contained with in one block. If the specified range exceeds
3343 * the end of the block it will be shortened to end of the block
3344 * that cooresponds to 'from'
3345 */
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003346static int ext4_block_zero_page_range(handle_t *handle,
Lukas Czernerd863dc32013-05-27 23:32:35 -04003347 struct address_space *mapping, loff_t from, loff_t length)
3348{
3349 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3350 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3351 unsigned blocksize, max, pos;
3352 ext4_lblk_t iblock;
3353 struct inode *inode = mapping->host;
3354 struct buffer_head *bh;
3355 struct page *page;
3356 int err = 0;
3357
3358 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3359 mapping_gfp_mask(mapping) & ~__GFP_FS);
3360 if (!page)
3361 return -ENOMEM;
3362
3363 blocksize = inode->i_sb->s_blocksize;
3364 max = blocksize - (offset & (blocksize - 1));
3365
3366 /*
3367 * correct length if it does not fall between
3368 * 'from' and the end of the block
3369 */
3370 if (length > max || length < 0)
3371 length = max;
3372
3373 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3374
3375 if (!page_has_buffers(page))
3376 create_empty_buffers(page, blocksize, 0);
3377
3378 /* Find the buffer that contains "offset" */
3379 bh = page_buffers(page);
3380 pos = blocksize;
3381 while (offset >= pos) {
3382 bh = bh->b_this_page;
3383 iblock++;
3384 pos += blocksize;
3385 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003386 if (buffer_freed(bh)) {
3387 BUFFER_TRACE(bh, "freed: skip");
3388 goto unlock;
3389 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003390 if (!buffer_mapped(bh)) {
3391 BUFFER_TRACE(bh, "unmapped");
3392 ext4_get_block(inode, iblock, bh, 0);
3393 /* unmapped? It's a hole - nothing to do */
3394 if (!buffer_mapped(bh)) {
3395 BUFFER_TRACE(bh, "still unmapped");
3396 goto unlock;
3397 }
3398 }
3399
3400 /* Ok, it's mapped. Make sure it's up-to-date */
3401 if (PageUptodate(page))
3402 set_buffer_uptodate(bh);
3403
3404 if (!buffer_uptodate(bh)) {
3405 err = -EIO;
3406 ll_rw_block(READ, 1, &bh);
3407 wait_on_buffer(bh);
3408 /* Uhhuh. Read error. Complain and punt. */
3409 if (!buffer_uptodate(bh))
3410 goto unlock;
3411 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003412 if (ext4_should_journal_data(inode)) {
3413 BUFFER_TRACE(bh, "get write access");
3414 err = ext4_journal_get_write_access(handle, bh);
3415 if (err)
3416 goto unlock;
3417 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003418 zero_user(page, offset, length);
Lukas Czernerd863dc32013-05-27 23:32:35 -04003419 BUFFER_TRACE(bh, "zeroed end of block");
3420
Lukas Czernerd863dc32013-05-27 23:32:35 -04003421 if (ext4_should_journal_data(inode)) {
3422 err = ext4_handle_dirty_metadata(handle, inode, bh);
Lukas Czerner0713ed02013-05-27 23:32:35 -04003423 } else {
jon ernst353eefd2013-07-01 08:12:39 -04003424 err = 0;
Lukas Czernerd863dc32013-05-27 23:32:35 -04003425 mark_buffer_dirty(bh);
Lukas Czerner0713ed02013-05-27 23:32:35 -04003426 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3427 err = ext4_jbd2_file_inode(handle, inode);
3428 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003429
3430unlock:
3431 unlock_page(page);
3432 page_cache_release(page);
3433 return err;
3434}
3435
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003436/*
3437 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3438 * up to the end of the block which corresponds to `from'.
3439 * This required during truncate. We need to physically zero the tail end
3440 * of that block so it doesn't yield old data if the file is later grown.
3441 */
3442int ext4_block_truncate_page(handle_t *handle,
3443 struct address_space *mapping, loff_t from)
3444{
3445 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3446 unsigned length;
3447 unsigned blocksize;
3448 struct inode *inode = mapping->host;
3449
3450 blocksize = inode->i_sb->s_blocksize;
3451 length = blocksize - (offset & (blocksize - 1));
3452
3453 return ext4_block_zero_page_range(handle, mapping, from, length);
3454}
3455
Lukas Czernera87dd182013-05-27 23:32:35 -04003456int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3457 loff_t lstart, loff_t length)
3458{
3459 struct super_block *sb = inode->i_sb;
3460 struct address_space *mapping = inode->i_mapping;
Lukas Czernere1be3a92013-07-01 08:12:39 -04003461 unsigned partial_start, partial_end;
Lukas Czernera87dd182013-05-27 23:32:35 -04003462 ext4_fsblk_t start, end;
3463 loff_t byte_end = (lstart + length - 1);
3464 int err = 0;
3465
Lukas Czernere1be3a92013-07-01 08:12:39 -04003466 partial_start = lstart & (sb->s_blocksize - 1);
3467 partial_end = byte_end & (sb->s_blocksize - 1);
3468
Lukas Czernera87dd182013-05-27 23:32:35 -04003469 start = lstart >> sb->s_blocksize_bits;
3470 end = byte_end >> sb->s_blocksize_bits;
3471
3472 /* Handle partial zero within the single block */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003473 if (start == end &&
3474 (partial_start || (partial_end != sb->s_blocksize - 1))) {
Lukas Czernera87dd182013-05-27 23:32:35 -04003475 err = ext4_block_zero_page_range(handle, mapping,
3476 lstart, length);
3477 return err;
3478 }
3479 /* Handle partial zero out on the start of the range */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003480 if (partial_start) {
Lukas Czernera87dd182013-05-27 23:32:35 -04003481 err = ext4_block_zero_page_range(handle, mapping,
3482 lstart, sb->s_blocksize);
3483 if (err)
3484 return err;
3485 }
3486 /* Handle partial zero out on the end of the range */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003487 if (partial_end != sb->s_blocksize - 1)
Lukas Czernera87dd182013-05-27 23:32:35 -04003488 err = ext4_block_zero_page_range(handle, mapping,
Lukas Czernere1be3a92013-07-01 08:12:39 -04003489 byte_end - partial_end,
3490 partial_end + 1);
Lukas Czernera87dd182013-05-27 23:32:35 -04003491 return err;
3492}
3493
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003494int ext4_can_truncate(struct inode *inode)
3495{
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003496 if (S_ISREG(inode->i_mode))
3497 return 1;
3498 if (S_ISDIR(inode->i_mode))
3499 return 1;
3500 if (S_ISLNK(inode->i_mode))
3501 return !ext4_inode_is_fast_symlink(inode);
3502 return 0;
3503}
3504
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003505/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003506 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3507 * associated with the given offset and length
3508 *
3509 * @inode: File inode
3510 * @offset: The offset where the hole will begin
3511 * @len: The length of the hole
3512 *
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003513 * Returns: 0 on success or negative on failure
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003514 */
3515
Ashish Sangwanaeb28172013-07-01 08:12:38 -04003516int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003517{
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003518 struct super_block *sb = inode->i_sb;
3519 ext4_lblk_t first_block, stop_block;
3520 struct address_space *mapping = inode->i_mapping;
Lukas Czernera87dd182013-05-27 23:32:35 -04003521 loff_t first_block_offset, last_block_offset;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003522 handle_t *handle;
3523 unsigned int credits;
3524 int ret = 0;
3525
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003526 if (!S_ISREG(inode->i_mode))
Allison Henderson73355192012-03-21 22:23:31 -04003527 return -EOPNOTSUPP;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003528
Lukas Czernerb8a86842014-03-18 18:05:35 -04003529 trace_ext4_punch_hole(inode, offset, length, 0);
Zheng Liuaaddea82013-01-16 20:21:26 -05003530
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003531 /*
3532 * Write out all dirty pages to avoid race conditions
3533 * Then release them.
3534 */
3535 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3536 ret = filemap_write_and_wait_range(mapping, offset,
3537 offset + length - 1);
3538 if (ret)
3539 return ret;
3540 }
3541
3542 mutex_lock(&inode->i_mutex);
Lukas Czerner9ef06ce2014-04-12 09:47:00 -04003543
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003544 /* No need to punch hole beyond i_size */
3545 if (offset >= inode->i_size)
3546 goto out_mutex;
3547
3548 /*
3549 * If the hole extends beyond i_size, set the hole
3550 * to end after the page that contains i_size
3551 */
3552 if (offset + length > inode->i_size) {
3553 length = inode->i_size +
3554 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3555 offset;
3556 }
3557
Jan Karaa3612932013-08-16 21:19:41 -04003558 if (offset & (sb->s_blocksize - 1) ||
3559 (offset + length) & (sb->s_blocksize - 1)) {
3560 /*
3561 * Attach jinode to inode for jbd2 if we do any zeroing of
3562 * partial block
3563 */
3564 ret = ext4_inode_attach_jinode(inode);
3565 if (ret < 0)
3566 goto out_mutex;
3567
3568 }
3569
Lukas Czernera87dd182013-05-27 23:32:35 -04003570 first_block_offset = round_up(offset, sb->s_blocksize);
3571 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003572
Lukas Czernera87dd182013-05-27 23:32:35 -04003573 /* Now release the pages and zero block aligned part of pages*/
3574 if (last_block_offset > first_block_offset)
3575 truncate_pagecache_range(inode, first_block_offset,
3576 last_block_offset);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003577
3578 /* Wait all existing dio workers, newcomers will block on i_mutex */
3579 ext4_inode_block_unlocked_dio(inode);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003580 inode_dio_wait(inode);
3581
3582 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3583 credits = ext4_writepage_trans_blocks(inode);
3584 else
3585 credits = ext4_blocks_for_truncate(inode);
3586 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3587 if (IS_ERR(handle)) {
3588 ret = PTR_ERR(handle);
3589 ext4_std_error(sb, ret);
3590 goto out_dio;
3591 }
3592
Lukas Czernera87dd182013-05-27 23:32:35 -04003593 ret = ext4_zero_partial_blocks(handle, inode, offset,
3594 length);
3595 if (ret)
3596 goto out_stop;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003597
3598 first_block = (offset + sb->s_blocksize - 1) >>
3599 EXT4_BLOCK_SIZE_BITS(sb);
3600 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3601
3602 /* If there are no blocks to remove, return now */
3603 if (first_block >= stop_block)
3604 goto out_stop;
3605
3606 down_write(&EXT4_I(inode)->i_data_sem);
3607 ext4_discard_preallocations(inode);
3608
3609 ret = ext4_es_remove_extent(inode, first_block,
3610 stop_block - first_block);
3611 if (ret) {
3612 up_write(&EXT4_I(inode)->i_data_sem);
3613 goto out_stop;
3614 }
3615
3616 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3617 ret = ext4_ext_remove_space(inode, first_block,
3618 stop_block - 1);
3619 else
3620 ret = ext4_free_hole_blocks(handle, inode, first_block,
3621 stop_block);
3622
Theodore Ts'o819c4922013-04-03 12:47:17 -04003623 up_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003624 if (IS_SYNC(inode))
3625 ext4_handle_sync(handle);
Maxim Patlasove251f9b2014-02-20 16:58:05 -05003626
3627 /* Now release the pages again to reduce race window */
3628 if (last_block_offset > first_block_offset)
3629 truncate_pagecache_range(inode, first_block_offset,
3630 last_block_offset);
3631
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003632 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3633 ext4_mark_inode_dirty(handle, inode);
3634out_stop:
3635 ext4_journal_stop(handle);
3636out_dio:
3637 ext4_inode_resume_unlocked_dio(inode);
3638out_mutex:
3639 mutex_unlock(&inode->i_mutex);
3640 return ret;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003641}
3642
Jan Karaa3612932013-08-16 21:19:41 -04003643int ext4_inode_attach_jinode(struct inode *inode)
3644{
3645 struct ext4_inode_info *ei = EXT4_I(inode);
3646 struct jbd2_inode *jinode;
3647
3648 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3649 return 0;
3650
3651 jinode = jbd2_alloc_inode(GFP_KERNEL);
3652 spin_lock(&inode->i_lock);
3653 if (!ei->jinode) {
3654 if (!jinode) {
3655 spin_unlock(&inode->i_lock);
3656 return -ENOMEM;
3657 }
3658 ei->jinode = jinode;
3659 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3660 jinode = NULL;
3661 }
3662 spin_unlock(&inode->i_lock);
3663 if (unlikely(jinode != NULL))
3664 jbd2_free_inode(jinode);
3665 return 0;
3666}
3667
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003668/*
Mingming Cao617ba132006-10-11 01:20:53 -07003669 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003670 *
Mingming Cao617ba132006-10-11 01:20:53 -07003671 * We block out ext4_get_block() block instantiations across the entire
3672 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003673 * simultaneously on behalf of the same inode.
3674 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003675 * As we work through the truncate and commit bits of it to the journal there
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003676 * is one core, guiding principle: the file's tree must always be consistent on
3677 * disk. We must be able to restart the truncate after a crash.
3678 *
3679 * The file's tree may be transiently inconsistent in memory (although it
3680 * probably isn't), but whenever we close off and commit a journal transaction,
3681 * the contents of (the filesystem + the journal) must be consistent and
3682 * restartable. It's pretty simple, really: bottom up, right to left (although
3683 * left-to-right works OK too).
3684 *
3685 * Note that at recovery time, journal replay occurs *before* the restart of
3686 * truncate against the orphan inode list.
3687 *
3688 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07003689 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003690 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07003691 * ext4_truncate() to have another go. So there will be instantiated blocks
3692 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003693 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07003694 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003695 */
Mingming Cao617ba132006-10-11 01:20:53 -07003696void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003697{
Theodore Ts'o819c4922013-04-03 12:47:17 -04003698 struct ext4_inode_info *ei = EXT4_I(inode);
3699 unsigned int credits;
3700 handle_t *handle;
3701 struct address_space *mapping = inode->i_mapping;
Theodore Ts'o819c4922013-04-03 12:47:17 -04003702
Theodore Ts'o19b5ef62013-04-03 21:58:52 -04003703 /*
3704 * There is a possibility that we're either freeing the inode
Matthew Wilcoxe04027e2014-03-24 15:15:07 -04003705 * or it's a completely new inode. In those cases we might not
Theodore Ts'o19b5ef62013-04-03 21:58:52 -04003706 * have i_mutex locked because it's not necessary.
3707 */
3708 if (!(inode->i_state & (I_NEW|I_FREEING)))
3709 WARN_ON(!mutex_is_locked(&inode->i_mutex));
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003710 trace_ext4_truncate_enter(inode);
3711
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003712 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003713 return;
3714
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003715 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003716
Theodore Ts'o5534fb52009-09-17 09:34:16 -04003717 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003718 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05003719
Tao Maaef1c852012-12-10 14:06:02 -05003720 if (ext4_has_inline_data(inode)) {
3721 int has_inline = 1;
3722
3723 ext4_inline_data_truncate(inode, &has_inline);
3724 if (has_inline)
3725 return;
3726 }
3727
Jan Karaa3612932013-08-16 21:19:41 -04003728 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3729 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3730 if (ext4_inode_attach_jinode(inode) < 0)
3731 return;
3732 }
3733
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003734 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Theodore Ts'o819c4922013-04-03 12:47:17 -04003735 credits = ext4_writepage_trans_blocks(inode);
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003736 else
Theodore Ts'o819c4922013-04-03 12:47:17 -04003737 credits = ext4_blocks_for_truncate(inode);
3738
3739 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3740 if (IS_ERR(handle)) {
3741 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3742 return;
3743 }
3744
Lukas Czernereb3544c2013-05-27 23:32:35 -04003745 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3746 ext4_block_truncate_page(handle, mapping, inode->i_size);
Theodore Ts'o819c4922013-04-03 12:47:17 -04003747
3748 /*
3749 * We add the inode to the orphan list, so that if this
3750 * truncate spans multiple transactions, and we crash, we will
3751 * resume the truncate when the filesystem recovers. It also
3752 * marks the inode dirty, to catch the new size.
3753 *
3754 * Implication: the file must always be in a sane, consistent
3755 * truncatable state while each transaction commits.
3756 */
3757 if (ext4_orphan_add(handle, inode))
3758 goto out_stop;
3759
3760 down_write(&EXT4_I(inode)->i_data_sem);
3761
3762 ext4_discard_preallocations(inode);
3763
3764 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3765 ext4_ext_truncate(handle, inode);
3766 else
3767 ext4_ind_truncate(handle, inode);
3768
3769 up_write(&ei->i_data_sem);
3770
3771 if (IS_SYNC(inode))
3772 ext4_handle_sync(handle);
3773
3774out_stop:
3775 /*
3776 * If this was a simple ftruncate() and the file will remain alive,
3777 * then we need to clear up the orphan record which we created above.
3778 * However, if this was a real unlink then we were called by
3779 * ext4_delete_inode(), and we allow that function to clean up the
3780 * orphan info for us.
3781 */
3782 if (inode->i_nlink)
3783 ext4_orphan_del(handle, inode);
3784
3785 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3786 ext4_mark_inode_dirty(handle, inode);
3787 ext4_journal_stop(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07003788
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003789 trace_ext4_truncate_exit(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003790}
3791
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003792/*
Mingming Cao617ba132006-10-11 01:20:53 -07003793 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003794 * underlying buffer_head on success. If 'in_mem' is true, we have all
3795 * data in memory that is needed to recreate the on-disk version of this
3796 * inode.
3797 */
Mingming Cao617ba132006-10-11 01:20:53 -07003798static int __ext4_get_inode_loc(struct inode *inode,
3799 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003800{
Theodore Ts'o240799c2008-10-09 23:53:47 -04003801 struct ext4_group_desc *gdp;
3802 struct buffer_head *bh;
3803 struct super_block *sb = inode->i_sb;
3804 ext4_fsblk_t block;
3805 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003806
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003807 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003808 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003809 return -EIO;
3810
Theodore Ts'o240799c2008-10-09 23:53:47 -04003811 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3812 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3813 if (!gdp)
3814 return -EIO;
3815
3816 /*
3817 * Figure out the offset within the block group inode table
3818 */
Tao Ma00d09882011-05-09 10:26:41 -04003819 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003820 inode_offset = ((inode->i_ino - 1) %
3821 EXT4_INODES_PER_GROUP(sb));
3822 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3823 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3824
3825 bh = sb_getblk(sb, block);
Wang Shilongaebf0242013-01-12 16:28:47 -05003826 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05003827 return -ENOMEM;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003828 if (!buffer_uptodate(bh)) {
3829 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04003830
3831 /*
3832 * If the buffer has the write error flag, we have failed
3833 * to write out another inode in the same block. In this
3834 * case, we don't have to read the block because we may
3835 * read the old inode data successfully.
3836 */
3837 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3838 set_buffer_uptodate(bh);
3839
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003840 if (buffer_uptodate(bh)) {
3841 /* someone brought it uptodate while we waited */
3842 unlock_buffer(bh);
3843 goto has_buffer;
3844 }
3845
3846 /*
3847 * If we have all information of the inode in memory and this
3848 * is the only valid inode in the block, we need not read the
3849 * block.
3850 */
3851 if (in_mem) {
3852 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003853 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003854
Theodore Ts'o240799c2008-10-09 23:53:47 -04003855 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003856
3857 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003858 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Wang Shilongaebf0242013-01-12 16:28:47 -05003859 if (unlikely(!bitmap_bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003860 goto make_io;
3861
3862 /*
3863 * If the inode bitmap isn't in cache then the
3864 * optimisation may end up performing two reads instead
3865 * of one, so skip it.
3866 */
3867 if (!buffer_uptodate(bitmap_bh)) {
3868 brelse(bitmap_bh);
3869 goto make_io;
3870 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04003871 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003872 if (i == inode_offset)
3873 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07003874 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003875 break;
3876 }
3877 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003878 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003879 /* all other inodes are free, so skip I/O */
3880 memset(bh->b_data, 0, bh->b_size);
3881 set_buffer_uptodate(bh);
3882 unlock_buffer(bh);
3883 goto has_buffer;
3884 }
3885 }
3886
3887make_io:
3888 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04003889 * If we need to do any I/O, try to pre-readahead extra
3890 * blocks from the inode table.
3891 */
3892 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3893 ext4_fsblk_t b, end, table;
3894 unsigned num;
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003895 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003896
3897 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04003898 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003899 b = block & ~((ext4_fsblk_t) ra_blks - 1);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003900 if (table > b)
3901 b = table;
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003902 end = b + ra_blks;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003903 num = EXT4_INODES_PER_GROUP(sb);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04003904 if (ext4_has_group_desc_csum(sb))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05003905 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003906 table += num / inodes_per_block;
3907 if (end > table)
3908 end = table;
3909 while (b <= end)
3910 sb_breadahead(sb, b++);
3911 }
3912
3913 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003914 * There are other valid inodes in the buffer, this inode
3915 * has in-inode xattrs, or we don't have this inode in memory.
3916 * Read the block from disk.
3917 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003918 trace_ext4_load_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003919 get_bh(bh);
3920 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02003921 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003922 wait_on_buffer(bh);
3923 if (!buffer_uptodate(bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003924 EXT4_ERROR_INODE_BLOCK(inode, block,
3925 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003926 brelse(bh);
3927 return -EIO;
3928 }
3929 }
3930has_buffer:
3931 iloc->bh = bh;
3932 return 0;
3933}
3934
Mingming Cao617ba132006-10-11 01:20:53 -07003935int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003936{
3937 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07003938 return __ext4_get_inode_loc(inode, iloc,
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003939 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003940}
3941
Mingming Cao617ba132006-10-11 01:20:53 -07003942void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003943{
Mingming Cao617ba132006-10-11 01:20:53 -07003944 unsigned int flags = EXT4_I(inode)->i_flags;
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003945 unsigned int new_fl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003946
Mingming Cao617ba132006-10-11 01:20:53 -07003947 if (flags & EXT4_SYNC_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003948 new_fl |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07003949 if (flags & EXT4_APPEND_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003950 new_fl |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07003951 if (flags & EXT4_IMMUTABLE_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003952 new_fl |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07003953 if (flags & EXT4_NOATIME_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003954 new_fl |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07003955 if (flags & EXT4_DIRSYNC_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003956 new_fl |= S_DIRSYNC;
Theodore Ts'o5f16f322014-03-24 14:43:12 -04003957 inode_set_flags(inode, new_fl,
3958 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003959}
3960
Jan Karaff9ddf72007-07-18 09:24:20 -04003961/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3962void ext4_get_inode_flags(struct ext4_inode_info *ei)
3963{
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003964 unsigned int vfs_fl;
3965 unsigned long old_fl, new_fl;
Jan Karaff9ddf72007-07-18 09:24:20 -04003966
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003967 do {
3968 vfs_fl = ei->vfs_inode.i_flags;
3969 old_fl = ei->i_flags;
3970 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3971 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3972 EXT4_DIRSYNC_FL);
3973 if (vfs_fl & S_SYNC)
3974 new_fl |= EXT4_SYNC_FL;
3975 if (vfs_fl & S_APPEND)
3976 new_fl |= EXT4_APPEND_FL;
3977 if (vfs_fl & S_IMMUTABLE)
3978 new_fl |= EXT4_IMMUTABLE_FL;
3979 if (vfs_fl & S_NOATIME)
3980 new_fl |= EXT4_NOATIME_FL;
3981 if (vfs_fl & S_DIRSYNC)
3982 new_fl |= EXT4_DIRSYNC_FL;
3983 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
Jan Karaff9ddf72007-07-18 09:24:20 -04003984}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003985
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003986static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003987 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003988{
3989 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003990 struct inode *inode = &(ei->vfs_inode);
3991 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003992
3993 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3994 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3995 /* we are using combined 48 bit field */
3996 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3997 le32_to_cpu(raw_inode->i_blocks_lo);
Theodore Ts'o07a03822010-06-14 09:54:48 -04003998 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003999 /* i_blocks represent file system block size */
4000 return i_blocks << (inode->i_blkbits - 9);
4001 } else {
4002 return i_blocks;
4003 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004004 } else {
4005 return le32_to_cpu(raw_inode->i_blocks_lo);
4006 }
4007}
Jan Karaff9ddf72007-07-18 09:24:20 -04004008
Tao Ma152a7b02012-12-02 11:13:24 -05004009static inline void ext4_iget_extra_inode(struct inode *inode,
4010 struct ext4_inode *raw_inode,
4011 struct ext4_inode_info *ei)
4012{
4013 __le32 *magic = (void *)raw_inode +
4014 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
Tao Ma67cf5b02012-12-10 14:04:46 -05004015 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
Tao Ma152a7b02012-12-02 11:13:24 -05004016 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
Tao Ma67cf5b02012-12-10 14:04:46 -05004017 ext4_find_inline_data_nolock(inode);
Tao Maf19d5872012-12-10 14:05:51 -05004018 } else
4019 EXT4_I(inode)->i_inline_off = 0;
Tao Ma152a7b02012-12-02 11:13:24 -05004020}
4021
David Howells1d1fe1e2008-02-07 00:15:37 -08004022struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004023{
Mingming Cao617ba132006-10-11 01:20:53 -07004024 struct ext4_iloc iloc;
4025 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004026 struct ext4_inode_info *ei;
David Howells1d1fe1e2008-02-07 00:15:37 -08004027 struct inode *inode;
Jan Karab436b9b2009-12-08 23:51:10 -05004028 journal_t *journal = EXT4_SB(sb)->s_journal;
David Howells1d1fe1e2008-02-07 00:15:37 -08004029 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004030 int block;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004031 uid_t i_uid;
4032 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004033
David Howells1d1fe1e2008-02-07 00:15:37 -08004034 inode = iget_locked(sb, ino);
4035 if (!inode)
4036 return ERR_PTR(-ENOMEM);
4037 if (!(inode->i_state & I_NEW))
4038 return inode;
4039
4040 ei = EXT4_I(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004041 iloc.bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004042
David Howells1d1fe1e2008-02-07 00:15:37 -08004043 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4044 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004045 goto bad_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07004046 raw_inode = ext4_raw_inode(&iloc);
Darrick J. Wong814525f2012-04-29 18:31:10 -04004047
4048 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4049 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4050 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4051 EXT4_INODE_SIZE(inode->i_sb)) {
4052 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4053 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4054 EXT4_INODE_SIZE(inode->i_sb));
4055 ret = -EIO;
4056 goto bad_inode;
4057 }
4058 } else
4059 ei->i_extra_isize = 0;
4060
4061 /* Precompute checksum seed for inode metadata */
4062 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4063 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4064 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4065 __u32 csum;
4066 __le32 inum = cpu_to_le32(inode->i_ino);
4067 __le32 gen = raw_inode->i_generation;
4068 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4069 sizeof(inum));
4070 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4071 sizeof(gen));
4072 }
4073
4074 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4075 EXT4_ERROR_INODE(inode, "checksum invalid");
4076 ret = -EIO;
4077 goto bad_inode;
4078 }
4079
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004080 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004081 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4082 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004083 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004084 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4085 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004086 }
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004087 i_uid_write(inode, i_uid);
4088 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02004089 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004090
Theodore Ts'o353eb832011-01-10 12:18:25 -05004091 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Tao Ma67cf5b02012-12-10 14:04:46 -05004092 ei->i_inline_off = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004093 ei->i_dir_start_lookup = 0;
4094 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4095 /* We now have enough fields to check if the inode was active or not.
4096 * This is needed because nfsd might try to access dead inodes
4097 * the test is that same one that e2fsck uses
4098 * NeilBrown 1999oct15
4099 */
4100 if (inode->i_nlink == 0) {
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004101 if ((inode->i_mode == 0 ||
4102 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4103 ino != EXT4_BOOT_LOADER_INO) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004104 /* this inode is deleted */
David Howells1d1fe1e2008-02-07 00:15:37 -08004105 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004106 goto bad_inode;
4107 }
4108 /* The only unlinked inodes we let through here have
4109 * valid i_mode and are being read by the orphan
4110 * recovery code: that's fine, we're about to complete
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004111 * the process of deleting those.
4112 * OR it is the EXT4_BOOT_LOADER_INO which is
4113 * not initialized on a new filesystem. */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004114 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004115 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004116 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004117 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04004118 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004119 ei->i_file_acl |=
4120 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004121 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004122 ei->i_disksize = inode->i_size;
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03004123#ifdef CONFIG_QUOTA
4124 ei->i_reserved_quota = 0;
4125#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004126 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4127 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04004128 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004129 /*
4130 * NOTE! The in-memory inode i_data array is in little-endian order
4131 * even on big-endian machines: we do NOT byteswap the block numbers!
4132 */
Mingming Cao617ba132006-10-11 01:20:53 -07004133 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004134 ei->i_data[block] = raw_inode->i_block[block];
4135 INIT_LIST_HEAD(&ei->i_orphan);
4136
Jan Karab436b9b2009-12-08 23:51:10 -05004137 /*
4138 * Set transaction id's of transactions that have to be committed
4139 * to finish f[data]sync. We set them to currently running transaction
4140 * as we cannot be sure that the inode or some of its metadata isn't
4141 * part of the transaction - the inode could have been reclaimed and
4142 * now it is reread from disk.
4143 */
4144 if (journal) {
4145 transaction_t *transaction;
4146 tid_t tid;
4147
Theodore Ts'oa931da62010-08-03 21:35:12 -04004148 read_lock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004149 if (journal->j_running_transaction)
4150 transaction = journal->j_running_transaction;
4151 else
4152 transaction = journal->j_committing_transaction;
4153 if (transaction)
4154 tid = transaction->t_tid;
4155 else
4156 tid = journal->j_commit_sequence;
Theodore Ts'oa931da62010-08-03 21:35:12 -04004157 read_unlock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004158 ei->i_sync_tid = tid;
4159 ei->i_datasync_tid = tid;
4160 }
4161
Eric Sandeen0040d982008-02-05 22:36:43 -05004162 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004163 if (ei->i_extra_isize == 0) {
4164 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07004165 ei->i_extra_isize = sizeof(struct ext4_inode) -
4166 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004167 } else {
Tao Ma152a7b02012-12-02 11:13:24 -05004168 ext4_iget_extra_inode(inode, raw_inode, ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004169 }
Darrick J. Wong814525f2012-04-29 18:31:10 -04004170 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004171
Kalpak Shahef7f3832007-07-18 09:15:20 -04004172 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4173 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4174 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4175 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4176
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004177 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
Theodore Ts'oc4f65702014-03-20 00:32:57 -04004178 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4179 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4180 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4181 inode->i_version |=
4182 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4183 }
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004184 }
4185
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004186 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004187 if (ei->i_file_acl &&
Theodore Ts'o10329882009-11-15 15:29:56 -05004188 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04004189 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4190 ei->i_file_acl);
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004191 ret = -EIO;
4192 goto bad_inode;
Tao Maf19d5872012-12-10 14:05:51 -05004193 } else if (!ext4_has_inline_data(inode)) {
4194 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4195 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4196 (S_ISLNK(inode->i_mode) &&
4197 !ext4_inode_is_fast_symlink(inode))))
4198 /* Validate extent which is part of inode */
4199 ret = ext4_ext_check_inode(inode);
4200 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4201 (S_ISLNK(inode->i_mode) &&
4202 !ext4_inode_is_fast_symlink(inode))) {
4203 /* Validate block references which are part of inode */
4204 ret = ext4_ind_check_inode(inode);
4205 }
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004206 }
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004207 if (ret)
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004208 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04004209
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004210 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004211 inode->i_op = &ext4_file_inode_operations;
4212 inode->i_fop = &ext4_file_operations;
4213 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004214 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004215 inode->i_op = &ext4_dir_inode_operations;
4216 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004217 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00004218 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004219 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00004220 nd_terminate_link(ei->i_data, inode->i_size,
4221 sizeof(ei->i_data) - 1);
4222 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004223 inode->i_op = &ext4_symlink_inode_operations;
4224 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004225 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004226 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4227 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004228 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004229 if (raw_inode->i_block[0])
4230 init_special_inode(inode, inode->i_mode,
4231 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4232 else
4233 init_special_inode(inode, inode->i_mode,
4234 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004235 } else if (ino == EXT4_BOOT_LOADER_INO) {
4236 make_bad_inode(inode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004237 } else {
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004238 ret = -EIO;
Theodore Ts'o24676da2010-05-16 21:00:00 -04004239 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004240 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004241 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004242 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004243 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004244 unlock_new_inode(inode);
4245 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004246
4247bad_inode:
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004248 brelse(iloc.bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004249 iget_failed(inode);
4250 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004251}
4252
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004253static int ext4_inode_blocks_set(handle_t *handle,
4254 struct ext4_inode *raw_inode,
4255 struct ext4_inode_info *ei)
4256{
4257 struct inode *inode = &(ei->vfs_inode);
4258 u64 i_blocks = inode->i_blocks;
4259 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004260
4261 if (i_blocks <= ~0U) {
4262 /*
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004263 * i_blocks can be represented in a 32 bit variable
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004264 * as multiple of 512 bytes
4265 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004266 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004267 raw_inode->i_blocks_high = 0;
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004268 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004269 return 0;
4270 }
4271 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4272 return -EFBIG;
4273
4274 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004275 /*
4276 * i_blocks can be represented in a 48 bit variable
4277 * as multiple of 512 bytes
4278 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004279 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004280 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004281 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004282 } else {
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004283 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004284 /* i_block is stored in file system block size */
4285 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4286 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4287 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004288 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004289 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004290}
4291
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004292/*
4293 * Post the struct inode info into an on-disk inode location in the
4294 * buffer-cache. This gobbles the caller's reference to the
4295 * buffer_head in the inode location struct.
4296 *
4297 * The caller must have write access to iloc->bh.
4298 */
Mingming Cao617ba132006-10-11 01:20:53 -07004299static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004300 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04004301 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004302{
Mingming Cao617ba132006-10-11 01:20:53 -07004303 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4304 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004305 struct buffer_head *bh = iloc->bh;
4306 int err = 0, rc, block;
Jan Karab71fc072012-09-26 21:52:20 -04004307 int need_datasync = 0;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004308 uid_t i_uid;
4309 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004310
4311 /* For fields not not tracking in the in-memory inode,
4312 * initialise them to zero for new inodes. */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004313 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
Mingming Cao617ba132006-10-11 01:20:53 -07004314 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004315
Jan Karaff9ddf72007-07-18 09:24:20 -04004316 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004317 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004318 i_uid = i_uid_read(inode);
4319 i_gid = i_gid_read(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004320 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004321 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4322 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004323/*
4324 * Fix up interoperability with old kernels. Otherwise, old inodes get
4325 * re-used with the upper 16 bits of the uid/gid intact
4326 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004327 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004328 raw_inode->i_uid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004329 cpu_to_le16(high_16_bits(i_uid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004330 raw_inode->i_gid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004331 cpu_to_le16(high_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004332 } else {
4333 raw_inode->i_uid_high = 0;
4334 raw_inode->i_gid_high = 0;
4335 }
4336 } else {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004337 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4338 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004339 raw_inode->i_uid_high = 0;
4340 raw_inode->i_gid_high = 0;
4341 }
4342 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004343
4344 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4345 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4346 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4347 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4348
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004349 if (ext4_inode_blocks_set(handle, raw_inode, ei))
4350 goto out_brelse;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004351 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o353eb832011-01-10 12:18:25 -05004352 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004353 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004354 raw_inode->i_file_acl_high =
4355 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004356 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Jan Karab71fc072012-09-26 21:52:20 -04004357 if (ei->i_disksize != ext4_isize(raw_inode)) {
4358 ext4_isize_set(raw_inode, ei->i_disksize);
4359 need_datasync = 1;
4360 }
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004361 if (ei->i_disksize > 0x7fffffffULL) {
4362 struct super_block *sb = inode->i_sb;
4363 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4364 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4365 EXT4_SB(sb)->s_es->s_rev_level ==
4366 cpu_to_le32(EXT4_GOOD_OLD_REV)) {
4367 /* If this is the first large file
4368 * created, add a flag to the superblock.
4369 */
4370 err = ext4_journal_get_write_access(handle,
4371 EXT4_SB(sb)->s_sbh);
4372 if (err)
4373 goto out_brelse;
4374 ext4_update_dynamic_rev(sb);
4375 EXT4_SET_RO_COMPAT_FEATURE(sb,
Mingming Cao617ba132006-10-11 01:20:53 -07004376 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
Frank Mayhar03901312009-01-07 00:06:22 -05004377 ext4_handle_sync(handle);
Artem Bityutskiyb50924c2012-07-22 20:37:31 -04004378 err = ext4_handle_dirty_super(handle, sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004379 }
4380 }
4381 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4382 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4383 if (old_valid_dev(inode->i_rdev)) {
4384 raw_inode->i_block[0] =
4385 cpu_to_le32(old_encode_dev(inode->i_rdev));
4386 raw_inode->i_block[1] = 0;
4387 } else {
4388 raw_inode->i_block[0] = 0;
4389 raw_inode->i_block[1] =
4390 cpu_to_le32(new_encode_dev(inode->i_rdev));
4391 raw_inode->i_block[2] = 0;
4392 }
Tao Maf19d5872012-12-10 14:05:51 -05004393 } else if (!ext4_has_inline_data(inode)) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004394 for (block = 0; block < EXT4_N_BLOCKS; block++)
4395 raw_inode->i_block[block] = ei->i_data[block];
Tao Maf19d5872012-12-10 14:05:51 -05004396 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004397
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004398 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
Theodore Ts'oc4f65702014-03-20 00:32:57 -04004399 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4400 if (ei->i_extra_isize) {
4401 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4402 raw_inode->i_version_hi =
4403 cpu_to_le32(inode->i_version >> 32);
4404 raw_inode->i_extra_isize =
4405 cpu_to_le16(ei->i_extra_isize);
4406 }
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004407 }
4408
Darrick J. Wong814525f2012-04-29 18:31:10 -04004409 ext4_inode_csum_set(inode, raw_inode, ei);
4410
Frank Mayhar830156c2009-09-29 10:07:47 -04004411 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05004412 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
Frank Mayhar830156c2009-09-29 10:07:47 -04004413 if (!err)
4414 err = rc;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004415 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004416
Jan Karab71fc072012-09-26 21:52:20 -04004417 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004418out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004419 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004420 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004421 return err;
4422}
4423
4424/*
Mingming Cao617ba132006-10-11 01:20:53 -07004425 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004426 *
4427 * We are called from a few places:
4428 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004429 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004430 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004431 * transaction to commit.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004432 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004433 * - Within flush work (sys_sync(), kupdate and such).
4434 * We wait on commit, if told to.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004435 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004436 * - Within iput_final() -> write_inode_now()
4437 * We wait on commit, if told to.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004438 *
4439 * In all cases it is actually safe for us to return without doing anything,
4440 * because the inode has been copied into a raw inode buffer in
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004441 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4442 * writeback.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004443 *
4444 * Note that we are absolutely dependent upon all inode dirtiers doing the
4445 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4446 * which we are interested.
4447 *
4448 * It would be a bug for them to not do this. The code:
4449 *
4450 * mark_inode_dirty(inode)
4451 * stuff();
4452 * inode->i_size = expr;
4453 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004454 * is in error because write_inode() could occur while `stuff()' is running,
4455 * and the new i_size will be lost. Plus the inode will no longer be on the
4456 * superblock's dirty inode list.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004457 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004458int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004459{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004460 int err;
4461
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004462 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004463 return 0;
4464
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004465 if (EXT4_SB(inode->i_sb)->s_journal) {
4466 if (ext4_journal_current_handle()) {
4467 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4468 dump_stack();
4469 return -EIO;
4470 }
4471
Jan Kara10542c22014-03-04 10:50:50 -05004472 /*
4473 * No need to force transaction in WB_SYNC_NONE mode. Also
4474 * ext4_sync_fs() will force the commit after everything is
4475 * written.
4476 */
4477 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004478 return 0;
4479
4480 err = ext4_force_commit(inode->i_sb);
4481 } else {
4482 struct ext4_iloc iloc;
4483
Curt Wohlgemuth8b472d72010-04-03 16:45:06 -04004484 err = __ext4_get_inode_loc(inode, &iloc, 0);
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004485 if (err)
4486 return err;
Jan Kara10542c22014-03-04 10:50:50 -05004487 /*
4488 * sync(2) will flush the whole buffer cache. No need to do
4489 * it here separately for each inode.
4490 */
4491 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
Frank Mayhar830156c2009-09-29 10:07:47 -04004492 sync_dirty_buffer(iloc.bh);
4493 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004494 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4495 "IO error syncing inode");
Frank Mayhar830156c2009-09-29 10:07:47 -04004496 err = -EIO;
4497 }
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004498 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004499 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004500 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004501}
4502
4503/*
Jan Kara53e87262012-12-25 13:29:52 -05004504 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4505 * buffers that are attached to a page stradding i_size and are undergoing
4506 * commit. In that case we have to wait for commit to finish and try again.
4507 */
4508static void ext4_wait_for_tail_page_commit(struct inode *inode)
4509{
4510 struct page *page;
4511 unsigned offset;
4512 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4513 tid_t commit_tid = 0;
4514 int ret;
4515
4516 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4517 /*
4518 * All buffers in the last page remain valid? Then there's nothing to
4519 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4520 * blocksize case
4521 */
4522 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4523 return;
4524 while (1) {
4525 page = find_lock_page(inode->i_mapping,
4526 inode->i_size >> PAGE_CACHE_SHIFT);
4527 if (!page)
4528 return;
Lukas Czernerca99fdd2013-05-21 23:25:01 -04004529 ret = __ext4_journalled_invalidatepage(page, offset,
4530 PAGE_CACHE_SIZE - offset);
Jan Kara53e87262012-12-25 13:29:52 -05004531 unlock_page(page);
4532 page_cache_release(page);
4533 if (ret != -EBUSY)
4534 return;
4535 commit_tid = 0;
4536 read_lock(&journal->j_state_lock);
4537 if (journal->j_committing_transaction)
4538 commit_tid = journal->j_committing_transaction->t_tid;
4539 read_unlock(&journal->j_state_lock);
4540 if (commit_tid)
4541 jbd2_log_wait_commit(journal, commit_tid);
4542 }
4543}
4544
4545/*
Mingming Cao617ba132006-10-11 01:20:53 -07004546 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004547 *
4548 * Called from notify_change.
4549 *
4550 * We want to trap VFS attempts to truncate the file as soon as
4551 * possible. In particular, we want to make sure that when the VFS
4552 * shrinks i_size, we put the inode on the orphan list and modify
4553 * i_disksize immediately, so that during the subsequent flushing of
4554 * dirty pages and freeing of disk blocks, we can guarantee that any
4555 * commit will leave the blocks being flushed in an unused state on
4556 * disk. (On recovery, the inode will get truncated and the blocks will
4557 * be freed, so we have a strong guarantee that no future commit will
4558 * leave these blocks visible to the user.)
4559 *
Jan Kara678aaf42008-07-11 19:27:31 -04004560 * Another thing we have to assure is that if we are in ordered mode
4561 * and inode is still attached to the committing transaction, we must
4562 * we start writeout of all the dirty pages which are being truncated.
4563 * This way we are sure that all the data written in the previous
4564 * transaction are already on disk (truncate waits for pages under
4565 * writeback).
4566 *
4567 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004568 */
Mingming Cao617ba132006-10-11 01:20:53 -07004569int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004570{
4571 struct inode *inode = dentry->d_inode;
4572 int error, rc = 0;
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004573 int orphan = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004574 const unsigned int ia_valid = attr->ia_valid;
4575
4576 error = inode_change_ok(inode, attr);
4577 if (error)
4578 return error;
4579
Dmitry Monakhov12755622010-04-08 22:04:20 +04004580 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05004581 dquot_initialize(inode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004582 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4583 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004584 handle_t *handle;
4585
4586 /* (user+group)*(old+new) structure, inode write (sb,
4587 * inode block, ? - but truncate inode update has it) */
Theodore Ts'o9924a922013-02-08 21:59:22 -05004588 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4589 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4590 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004591 if (IS_ERR(handle)) {
4592 error = PTR_ERR(handle);
4593 goto err_out;
4594 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05004595 error = dquot_transfer(inode, attr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004596 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07004597 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004598 return error;
4599 }
4600 /* Update corresponding info in inode so that everything is in
4601 * one transaction */
4602 if (attr->ia_valid & ATTR_UID)
4603 inode->i_uid = attr->ia_uid;
4604 if (attr->ia_valid & ATTR_GID)
4605 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07004606 error = ext4_mark_inode_dirty(handle, inode);
4607 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004608 }
4609
Jan Kara52083862013-08-17 10:07:17 -04004610 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4611 handle_t *handle;
Christoph Hellwig562c72a2011-06-24 14:29:45 -04004612
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004613 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -05004614 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4615
Theodore Ts'o0c095c72010-07-27 11:56:06 -04004616 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4617 return -EFBIG;
Eric Sandeene2b46572008-01-28 23:58:27 -05004618 }
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08004619
4620 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4621 inode_inc_iversion(inode);
4622
Jan Kara52083862013-08-17 10:07:17 -04004623 if (S_ISREG(inode->i_mode) &&
4624 (attr->ia_size < inode->i_size)) {
4625 if (ext4_should_order_data(inode)) {
4626 error = ext4_begin_ordered_truncate(inode,
Jan Kara678aaf42008-07-11 19:27:31 -04004627 attr->ia_size);
Jan Kara52083862013-08-17 10:07:17 -04004628 if (error)
Jan Kara678aaf42008-07-11 19:27:31 -04004629 goto err_out;
Jan Kara52083862013-08-17 10:07:17 -04004630 }
4631 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4632 if (IS_ERR(handle)) {
4633 error = PTR_ERR(handle);
4634 goto err_out;
4635 }
4636 if (ext4_handle_valid(handle)) {
4637 error = ext4_orphan_add(handle, inode);
4638 orphan = 1;
4639 }
Jan Kara90e775b2013-08-17 10:09:31 -04004640 down_write(&EXT4_I(inode)->i_data_sem);
Jan Kara52083862013-08-17 10:07:17 -04004641 EXT4_I(inode)->i_disksize = attr->ia_size;
4642 rc = ext4_mark_inode_dirty(handle, inode);
4643 if (!error)
4644 error = rc;
Jan Kara90e775b2013-08-17 10:09:31 -04004645 /*
4646 * We have to update i_size under i_data_sem together
4647 * with i_disksize to avoid races with writeback code
4648 * running ext4_wb_update_i_disksize().
4649 */
4650 if (!error)
4651 i_size_write(inode, attr->ia_size);
4652 up_write(&EXT4_I(inode)->i_data_sem);
Jan Kara52083862013-08-17 10:07:17 -04004653 ext4_journal_stop(handle);
4654 if (error) {
4655 ext4_orphan_del(NULL, inode);
Jan Kara678aaf42008-07-11 19:27:31 -04004656 goto err_out;
4657 }
Jan Kara90e775b2013-08-17 10:09:31 -04004658 } else
4659 i_size_write(inode, attr->ia_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004660
Jan Kara52083862013-08-17 10:07:17 -04004661 /*
4662 * Blocks are going to be removed from the inode. Wait
4663 * for dio in flight. Temporarily disable
4664 * dioread_nolock to prevent livelock.
4665 */
4666 if (orphan) {
4667 if (!ext4_should_journal_data(inode)) {
4668 ext4_inode_block_unlocked_dio(inode);
4669 inode_dio_wait(inode);
4670 ext4_inode_resume_unlocked_dio(inode);
4671 } else
4672 ext4_wait_for_tail_page_commit(inode);
Dmitry Monakhov1c9114f2012-09-29 00:55:23 -04004673 }
Jan Kara52083862013-08-17 10:07:17 -04004674 /*
4675 * Truncate pagecache after we've waited for commit
4676 * in data=journal mode to make pages freeable.
4677 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -07004678 truncate_pagecache(inode, inode->i_size);
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004679 }
Jan Kara52083862013-08-17 10:07:17 -04004680 /*
4681 * We want to call ext4_truncate() even if attr->ia_size ==
4682 * inode->i_size for cases like truncation of fallocated space
4683 */
4684 if (attr->ia_valid & ATTR_SIZE)
4685 ext4_truncate(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004686
Christoph Hellwig10257742010-06-04 11:30:02 +02004687 if (!rc) {
4688 setattr_copy(inode, attr);
4689 mark_inode_dirty(inode);
4690 }
4691
4692 /*
4693 * If the call to ext4_truncate failed to get a transaction handle at
4694 * all, we need to clean up the in-core orphan list manually.
4695 */
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004696 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004697 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004698
4699 if (!rc && (ia_valid & ATTR_MODE))
Christoph Hellwig64e178a2013-12-20 05:16:44 -08004700 rc = posix_acl_chmod(inode, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004701
4702err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07004703 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004704 if (!error)
4705 error = rc;
4706 return error;
4707}
4708
Mingming Cao3e3398a2008-07-11 19:27:31 -04004709int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4710 struct kstat *stat)
4711{
4712 struct inode *inode;
Jan Kara8af8eec2013-05-31 19:39:56 -04004713 unsigned long long delalloc_blocks;
Mingming Cao3e3398a2008-07-11 19:27:31 -04004714
4715 inode = dentry->d_inode;
4716 generic_fillattr(inode, stat);
4717
4718 /*
Andreas Dilger9206c562013-11-11 22:38:12 -05004719 * If there is inline data in the inode, the inode will normally not
4720 * have data blocks allocated (it may have an external xattr block).
4721 * Report at least one sector for such files, so tools like tar, rsync,
4722 * others doen't incorrectly think the file is completely sparse.
4723 */
4724 if (unlikely(ext4_has_inline_data(inode)))
4725 stat->blocks += (stat->size + 511) >> 9;
4726
4727 /*
Mingming Cao3e3398a2008-07-11 19:27:31 -04004728 * We can't update i_blocks if the block allocation is delayed
4729 * otherwise in the case of system crash before the real block
4730 * allocation is done, we will have i_blocks inconsistent with
4731 * on-disk file blocks.
4732 * We always keep i_blocks updated together with real
4733 * allocation. But to not confuse with user, stat
4734 * will return the blocks that include the delayed allocation
4735 * blocks for this file.
4736 */
Tao Ma96607552012-05-31 22:54:16 -04004737 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
Andreas Dilger9206c562013-11-11 22:38:12 -05004738 EXT4_I(inode)->i_reserved_data_blocks);
4739 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
Mingming Cao3e3398a2008-07-11 19:27:31 -04004740 return 0;
4741}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004742
Jan Karafffb2732013-06-04 13:01:11 -04004743static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4744 int pextents)
Mingming Caoa02908f2008-08-19 22:16:07 -04004745{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004746 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Jan Karafffb2732013-06-04 13:01:11 -04004747 return ext4_ind_trans_blocks(inode, lblocks);
4748 return ext4_ext_index_trans_blocks(inode, pextents);
Mingming Caoa02908f2008-08-19 22:16:07 -04004749}
Theodore Ts'oac51d832008-11-06 16:49:36 -05004750
Mingming Caoa02908f2008-08-19 22:16:07 -04004751/*
4752 * Account for index blocks, block groups bitmaps and block group
4753 * descriptor blocks if modify datablocks and index blocks
4754 * worse case, the indexs blocks spread over different block groups
4755 *
4756 * If datablocks are discontiguous, they are possible to spread over
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004757 * different block groups too. If they are contiguous, with flexbg,
Mingming Caoa02908f2008-08-19 22:16:07 -04004758 * they could still across block group boundary.
4759 *
4760 * Also account for superblock, inode, quota and xattr blocks
4761 */
Jan Karafffb2732013-06-04 13:01:11 -04004762static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4763 int pextents)
Mingming Caoa02908f2008-08-19 22:16:07 -04004764{
Theodore Ts'o8df96752009-05-01 08:50:38 -04004765 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4766 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04004767 int idxblocks;
4768 int ret = 0;
4769
4770 /*
Jan Karafffb2732013-06-04 13:01:11 -04004771 * How many index blocks need to touch to map @lblocks logical blocks
4772 * to @pextents physical extents?
Mingming Caoa02908f2008-08-19 22:16:07 -04004773 */
Jan Karafffb2732013-06-04 13:01:11 -04004774 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
Mingming Caoa02908f2008-08-19 22:16:07 -04004775
4776 ret = idxblocks;
4777
4778 /*
4779 * Now let's see how many group bitmaps and group descriptors need
4780 * to account
4781 */
Jan Karafffb2732013-06-04 13:01:11 -04004782 groups = idxblocks + pextents;
Mingming Caoa02908f2008-08-19 22:16:07 -04004783 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04004784 if (groups > ngroups)
4785 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04004786 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4787 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4788
4789 /* bitmaps and block group descriptor blocks */
4790 ret += groups + gdpblocks;
4791
4792 /* Blocks for super block, inode, quota and xattr blocks */
4793 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004794
4795 return ret;
4796}
4797
4798/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004799 * Calculate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04004800 * the modification of a single pages into a single transaction,
4801 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04004802 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004803 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04004804 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004805 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04004806 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04004807 */
4808int ext4_writepage_trans_blocks(struct inode *inode)
4809{
4810 int bpp = ext4_journal_blocks_per_page(inode);
4811 int ret;
4812
Jan Karafffb2732013-06-04 13:01:11 -04004813 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
Mingming Caoa02908f2008-08-19 22:16:07 -04004814
4815 /* Account for data blocks for journalled mode */
4816 if (ext4_should_journal_data(inode))
4817 ret += bpp;
4818 return ret;
4819}
Mingming Caof3bd1f32008-08-19 22:16:03 -04004820
4821/*
4822 * Calculate the journal credits for a chunk of data modification.
4823 *
4824 * This is called from DIO, fallocate or whoever calling
Eric Sandeen79e83032010-07-27 11:56:07 -04004825 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04004826 *
4827 * journal buffers for data blocks are not included here, as DIO
4828 * and fallocate do no need to journal data buffers.
4829 */
4830int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4831{
4832 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4833}
4834
Mingming Caoa02908f2008-08-19 22:16:07 -04004835/*
Mingming Cao617ba132006-10-11 01:20:53 -07004836 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004837 * Give this, we know that the caller already has write access to iloc->bh.
4838 */
Mingming Cao617ba132006-10-11 01:20:53 -07004839int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004840 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004841{
4842 int err = 0;
4843
Theodore Ts'oc64db502012-03-02 12:23:11 -05004844 if (IS_I_VERSION(inode))
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004845 inode_inc_iversion(inode);
4846
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004847 /* the do_update_inode consumes one bh->b_count */
4848 get_bh(iloc->bh);
4849
Mingming Caodab291a2006-10-11 01:21:01 -07004850 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04004851 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004852 put_bh(iloc->bh);
4853 return err;
4854}
4855
4856/*
4857 * On success, We end up with an outstanding reference count against
4858 * iloc->bh. This _must_ be cleaned up later.
4859 */
4860
4861int
Mingming Cao617ba132006-10-11 01:20:53 -07004862ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4863 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004864{
Frank Mayhar03901312009-01-07 00:06:22 -05004865 int err;
4866
4867 err = ext4_get_inode_loc(inode, iloc);
4868 if (!err) {
4869 BUFFER_TRACE(iloc->bh, "get_write_access");
4870 err = ext4_journal_get_write_access(handle, iloc->bh);
4871 if (err) {
4872 brelse(iloc->bh);
4873 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004874 }
4875 }
Mingming Cao617ba132006-10-11 01:20:53 -07004876 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004877 return err;
4878}
4879
4880/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004881 * Expand an inode by new_extra_isize bytes.
4882 * Returns 0 on success or negative error number on failure.
4883 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004884static int ext4_expand_extra_isize(struct inode *inode,
4885 unsigned int new_extra_isize,
4886 struct ext4_iloc iloc,
4887 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004888{
4889 struct ext4_inode *raw_inode;
4890 struct ext4_xattr_ibody_header *header;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004891
4892 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4893 return 0;
4894
4895 raw_inode = ext4_raw_inode(&iloc);
4896
4897 header = IHDR(inode, raw_inode);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004898
4899 /* No extended attributes present */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004900 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4901 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004902 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4903 new_extra_isize);
4904 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4905 return 0;
4906 }
4907
4908 /* try to expand with EAs present */
4909 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4910 raw_inode, handle);
4911}
4912
4913/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004914 * What we do here is to mark the in-core inode as clean with respect to inode
4915 * dirtiness (it may still be data-dirty).
4916 * This means that the in-core inode may be reaped by prune_icache
4917 * without having to perform any I/O. This is a very good thing,
4918 * because *any* task may call prune_icache - even ones which
4919 * have a transaction open against a different journal.
4920 *
4921 * Is this cheating? Not really. Sure, we haven't written the
4922 * inode out, but prune_icache isn't a user-visible syncing function.
4923 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4924 * we start and wait on commits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004925 */
Mingming Cao617ba132006-10-11 01:20:53 -07004926int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004927{
Mingming Cao617ba132006-10-11 01:20:53 -07004928 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004929 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4930 static unsigned int mnt_count;
4931 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004932
4933 might_sleep();
Theodore Ts'o7ff9c072010-11-08 13:51:33 -05004934 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
Mingming Cao617ba132006-10-11 01:20:53 -07004935 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05004936 if (ext4_handle_valid(handle) &&
4937 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004938 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004939 /*
4940 * We need extra buffer credits since we may write into EA block
4941 * with this same handle. If journal_extend fails, then it will
4942 * only result in a minor loss of functionality for that inode.
4943 * If this is felt to be critical, then e2fsck should be run to
4944 * force a large enough s_min_extra_isize.
4945 */
4946 if ((jbd2_journal_extend(handle,
4947 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4948 ret = ext4_expand_extra_isize(inode,
4949 sbi->s_want_extra_isize,
4950 iloc, handle);
4951 if (ret) {
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004952 ext4_set_inode_state(inode,
4953 EXT4_STATE_NO_EXPAND);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004954 if (mnt_count !=
4955 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05004956 ext4_warning(inode->i_sb,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004957 "Unable to expand inode %lu. Delete"
4958 " some EAs or run e2fsck.",
4959 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004960 mnt_count =
4961 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004962 }
4963 }
4964 }
4965 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004966 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07004967 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004968 return err;
4969}
4970
4971/*
Mingming Cao617ba132006-10-11 01:20:53 -07004972 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004973 *
4974 * We're really interested in the case where a file is being extended.
4975 * i_size has been changed by generic_commit_write() and we thus need
4976 * to include the updated inode in the current transaction.
4977 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05004978 * Also, dquot_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004979 * are allocated to the file.
4980 *
4981 * If the inode is marked synchronous, we don't honour that here - doing
4982 * so would cause a commit on atime updates, which we don't bother doing.
4983 * We handle synchronous inodes at the highest possible level.
4984 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04004985void ext4_dirty_inode(struct inode *inode, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004986{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004987 handle_t *handle;
4988
Theodore Ts'o9924a922013-02-08 21:59:22 -05004989 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004990 if (IS_ERR(handle))
4991 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004992
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004993 ext4_mark_inode_dirty(handle, inode);
4994
Mingming Cao617ba132006-10-11 01:20:53 -07004995 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004996out:
4997 return;
4998}
4999
5000#if 0
5001/*
5002 * Bind an inode's backing buffer_head into this transaction, to prevent
5003 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07005004 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005005 * returns no iloc structure, so the caller needs to repeat the iloc
5006 * lookup to mark the inode dirty later.
5007 */
Mingming Cao617ba132006-10-11 01:20:53 -07005008static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005009{
Mingming Cao617ba132006-10-11 01:20:53 -07005010 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005011
5012 int err = 0;
5013 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07005014 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005015 if (!err) {
5016 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07005017 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005018 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05005019 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05005020 NULL,
Frank Mayhar03901312009-01-07 00:06:22 -05005021 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005022 brelse(iloc.bh);
5023 }
5024 }
Mingming Cao617ba132006-10-11 01:20:53 -07005025 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005026 return err;
5027}
5028#endif
5029
Mingming Cao617ba132006-10-11 01:20:53 -07005030int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005031{
5032 journal_t *journal;
5033 handle_t *handle;
5034 int err;
5035
5036 /*
5037 * We have to be very careful here: changing a data block's
5038 * journaling status dynamically is dangerous. If we write a
5039 * data block to the journal, change the status and then delete
5040 * that block, we risk forgetting to revoke the old log record
5041 * from the journal and so a subsequent replay can corrupt data.
5042 * So, first we make sure that the journal is empty and that
5043 * nobody is changing anything.
5044 */
5045
Mingming Cao617ba132006-10-11 01:20:53 -07005046 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005047 if (!journal)
5048 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04005049 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005050 return -EROFS;
Yongqiang Yang2aff57b2011-12-28 12:02:13 -05005051 /* We have to allocate physical blocks for delalloc blocks
5052 * before flushing journal. otherwise delalloc blocks can not
5053 * be allocated any more. even more truncate on delalloc blocks
5054 * could trigger BUG by flushing delalloc blocks in journal.
5055 * There is no delalloc block in non-journal data mode.
5056 */
5057 if (val && test_opt(inode->i_sb, DELALLOC)) {
5058 err = ext4_alloc_da_blocks(inode);
5059 if (err < 0)
5060 return err;
5061 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005062
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04005063 /* Wait for all existing dio workers */
5064 ext4_inode_block_unlocked_dio(inode);
5065 inode_dio_wait(inode);
5066
Mingming Caodab291a2006-10-11 01:21:01 -07005067 jbd2_journal_lock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005068
5069 /*
5070 * OK, there are no updates running now, and all cached data is
5071 * synced to disk. We are now in a completely consistent state
5072 * which doesn't have anything in the journal, and we know that
5073 * no filesystem updates are running, so it is safe to modify
5074 * the inode's in-core data-journaling state flag now.
5075 */
5076
5077 if (val)
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005078 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05005079 else {
5080 jbd2_journal_flush(journal);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005081 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05005082 }
Mingming Cao617ba132006-10-11 01:20:53 -07005083 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005084
Mingming Caodab291a2006-10-11 01:21:01 -07005085 jbd2_journal_unlock_updates(journal);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04005086 ext4_inode_resume_unlocked_dio(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005087
5088 /* Finally we can mark the inode as dirty. */
5089
Theodore Ts'o9924a922013-02-08 21:59:22 -05005090 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005091 if (IS_ERR(handle))
5092 return PTR_ERR(handle);
5093
Mingming Cao617ba132006-10-11 01:20:53 -07005094 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005095 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07005096 ext4_journal_stop(handle);
5097 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005098
5099 return err;
5100}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005101
5102static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5103{
5104 return !buffer_mapped(bh);
5105}
5106
Nick Pigginc2ec1752009-03-31 15:23:21 -07005107int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005108{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005109 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005110 loff_t size;
5111 unsigned long len;
Jan Kara9ea7df52011-06-24 14:29:41 -04005112 int ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005113 struct file *file = vma->vm_file;
Al Viro496ad9a2013-01-23 17:07:38 -05005114 struct inode *inode = file_inode(file);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005115 struct address_space *mapping = inode->i_mapping;
Jan Kara9ea7df52011-06-24 14:29:41 -04005116 handle_t *handle;
5117 get_block_t *get_block;
5118 int retries = 0;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005119
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005120 sb_start_pagefault(inode->i_sb);
Theodore Ts'o041bbb6d2012-09-30 23:04:56 -04005121 file_update_time(vma->vm_file);
Jan Kara9ea7df52011-06-24 14:29:41 -04005122 /* Delalloc case is easy... */
5123 if (test_opt(inode->i_sb, DELALLOC) &&
5124 !ext4_should_journal_data(inode) &&
5125 !ext4_nonda_switch(inode->i_sb)) {
5126 do {
5127 ret = __block_page_mkwrite(vma, vmf,
5128 ext4_da_get_block_prep);
5129 } while (ret == -ENOSPC &&
5130 ext4_should_retry_alloc(inode->i_sb, &retries));
5131 goto out_ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005132 }
Darrick J. Wong0e499892011-05-18 13:55:20 -04005133
5134 lock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005135 size = i_size_read(inode);
5136 /* Page got truncated from under us? */
5137 if (page->mapping != mapping || page_offset(page) > size) {
5138 unlock_page(page);
5139 ret = VM_FAULT_NOPAGE;
5140 goto out;
Darrick J. Wong0e499892011-05-18 13:55:20 -04005141 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005142
5143 if (page->index == size >> PAGE_CACHE_SHIFT)
5144 len = size & ~PAGE_CACHE_MASK;
5145 else
5146 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005147 /*
Jan Kara9ea7df52011-06-24 14:29:41 -04005148 * Return if we have all the buffers mapped. This avoids the need to do
5149 * journal_start/journal_stop which can block and take a long time
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005150 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005151 if (page_has_buffers(page)) {
Tao Maf19d5872012-12-10 14:05:51 -05005152 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5153 0, len, NULL,
5154 ext4_bh_unmapped)) {
Jan Kara9ea7df52011-06-24 14:29:41 -04005155 /* Wait so that we don't change page under IO */
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08005156 wait_for_stable_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005157 ret = VM_FAULT_LOCKED;
5158 goto out;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005159 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005160 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005161 unlock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005162 /* OK, we need to fill the hole... */
5163 if (ext4_should_dioread_nolock(inode))
5164 get_block = ext4_get_block_write;
5165 else
5166 get_block = ext4_get_block;
5167retry_alloc:
Theodore Ts'o9924a922013-02-08 21:59:22 -05005168 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5169 ext4_writepage_trans_blocks(inode));
Jan Kara9ea7df52011-06-24 14:29:41 -04005170 if (IS_ERR(handle)) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07005171 ret = VM_FAULT_SIGBUS;
Jan Kara9ea7df52011-06-24 14:29:41 -04005172 goto out;
5173 }
5174 ret = __block_page_mkwrite(vma, vmf, get_block);
5175 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -05005176 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
Jan Kara9ea7df52011-06-24 14:29:41 -04005177 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5178 unlock_page(page);
5179 ret = VM_FAULT_SIGBUS;
Yongqiang Yangfcbb5512011-10-26 05:00:19 -04005180 ext4_journal_stop(handle);
Jan Kara9ea7df52011-06-24 14:29:41 -04005181 goto out;
5182 }
5183 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5184 }
5185 ext4_journal_stop(handle);
5186 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5187 goto retry_alloc;
5188out_ret:
5189 ret = block_page_mkwrite_return(ret);
5190out:
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005191 sb_end_pagefault(inode->i_sb);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005192 return ret;
5193}