blob: 574c6dda95d879ad37478f9b7caf60df40de2a8d [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 *
Lukas Czerner556615d2014-04-20 23:45:47 -0400492 * On success, it returns the number of blocks being mapped or allocated.
493 * if create==0 and the blocks are pre-allocated and unwritten block,
Mingming Caof5ab0d12008-02-25 15:29:55 -0500494 * 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 /*
Lukas Czerner556615d2014-04-20 23:45:47 -0400625 * New blocks allocate and/or writing to unwritten extent
Mingming Caof5ab0d12008-02-25 15:29:55 -0500626 * 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;
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001849 bool keep_towrite = false;
Alex Tomas64769242008-07-11 19:27:31 -04001850
Lukas Czernera9c667f2011-06-06 09:51:52 -04001851 trace_ext4_writepage(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001852 size = i_size_read(inode);
1853 if (page->index == size >> PAGE_CACHE_SHIFT)
1854 len = size & ~PAGE_CACHE_MASK;
1855 else
1856 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04001857
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001858 page_bufs = page_buffers(page);
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001859 /*
Jan Karafe386132013-01-28 21:06:42 -05001860 * We cannot do block allocation or other extent handling in this
1861 * function. If there are buffers needing that, we have to redirty
1862 * the page. But we may reach here when we do a journal commit via
1863 * journal_submit_inode_data_buffers() and in that case we must write
1864 * allocated buffers to achieve data=ordered mode guarantees.
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04001865 */
Tao Maf19d5872012-12-10 14:05:51 -05001866 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1867 ext4_bh_delay_or_unwritten)) {
Jan Karaf8bec372013-01-28 12:55:08 -05001868 redirty_page_for_writepage(wbc, page);
Jan Karafe386132013-01-28 21:06:42 -05001869 if (current->flags & PF_MEMALLOC) {
1870 /*
1871 * For memory cleaning there's no point in writing only
1872 * some buffers. So just bail out. Warn if we came here
1873 * from direct reclaim.
1874 */
1875 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1876 == PF_MEMALLOC);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001877 unlock_page(page);
1878 return 0;
1879 }
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001880 keep_towrite = true;
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001881 }
Alex Tomas64769242008-07-11 19:27:31 -04001882
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001883 if (PageChecked(page) && ext4_should_journal_data(inode))
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001884 /*
1885 * It's mmapped pagecache. Add buffers and journal it. There
1886 * doesn't seem much point in redirtying the page here.
1887 */
Wu Fengguang3f0ca302009-11-24 11:15:44 -05001888 return __ext4_journalled_writepage(page, len);
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001889
Jan Kara97a851e2013-06-04 11:58:58 -04001890 ext4_io_submit_init(&io_submit, wbc);
1891 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1892 if (!io_submit.io_end) {
1893 redirty_page_for_writepage(wbc, page);
1894 unlock_page(page);
1895 return -ENOMEM;
1896 }
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001897 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
Jan Kara36ade452013-01-28 09:30:52 -05001898 ext4_io_submit(&io_submit);
Jan Kara97a851e2013-06-04 11:58:58 -04001899 /* Drop io_end reference we got from init */
1900 ext4_put_io_end_defer(io_submit.io_end);
Alex Tomas64769242008-07-11 19:27:31 -04001901 return ret;
1902}
1903
Jan Kara5f1132b2013-08-17 10:02:33 -04001904static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1905{
1906 int len;
1907 loff_t size = i_size_read(mpd->inode);
1908 int err;
1909
1910 BUG_ON(page->index != mpd->first_page);
1911 if (page->index == size >> PAGE_CACHE_SHIFT)
1912 len = size & ~PAGE_CACHE_MASK;
1913 else
1914 len = PAGE_CACHE_SIZE;
1915 clear_page_dirty_for_io(page);
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001916 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
Jan Kara5f1132b2013-08-17 10:02:33 -04001917 if (!err)
1918 mpd->wbc->nr_to_write--;
1919 mpd->first_page++;
1920
1921 return err;
1922}
1923
Jan Kara4e7ea812013-06-04 13:17:40 -04001924#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1925
Mingming Cao61628a32008-07-11 19:27:31 -04001926/*
Jan Karafffb2732013-06-04 13:01:11 -04001927 * mballoc gives us at most this number of blocks...
1928 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
Anatol Pomozov70261f52013-08-28 14:40:12 -04001929 * The rest of mballoc seems to handle chunks up to full group size.
Mingming Cao61628a32008-07-11 19:27:31 -04001930 */
Jan Karafffb2732013-06-04 13:01:11 -04001931#define MAX_WRITEPAGES_EXTENT_LEN 2048
Mingming Cao525f4ed2008-08-19 22:15:58 -04001932
Jan Karafffb2732013-06-04 13:01:11 -04001933/*
Jan Kara4e7ea812013-06-04 13:17:40 -04001934 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1935 *
1936 * @mpd - extent of blocks
1937 * @lblk - logical number of the block in the file
Jan Kara09930042013-08-17 09:57:56 -04001938 * @bh - buffer head we want to add to the extent
Jan Kara4e7ea812013-06-04 13:17:40 -04001939 *
Jan Kara09930042013-08-17 09:57:56 -04001940 * The function is used to collect contig. blocks in the same state. If the
1941 * buffer doesn't require mapping for writeback and we haven't started the
1942 * extent of buffers to map yet, the function returns 'true' immediately - the
1943 * caller can write the buffer right away. Otherwise the function returns true
1944 * if the block has been added to the extent, false if the block couldn't be
1945 * added.
Jan Kara4e7ea812013-06-04 13:17:40 -04001946 */
Jan Kara09930042013-08-17 09:57:56 -04001947static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1948 struct buffer_head *bh)
Jan Kara4e7ea812013-06-04 13:17:40 -04001949{
1950 struct ext4_map_blocks *map = &mpd->map;
1951
Jan Kara09930042013-08-17 09:57:56 -04001952 /* Buffer that doesn't need mapping for writeback? */
1953 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1954 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1955 /* So far no extent to map => we write the buffer right away */
1956 if (map->m_len == 0)
1957 return true;
1958 return false;
1959 }
Jan Kara4e7ea812013-06-04 13:17:40 -04001960
1961 /* First block in the extent? */
1962 if (map->m_len == 0) {
1963 map->m_lblk = lblk;
1964 map->m_len = 1;
Jan Kara09930042013-08-17 09:57:56 -04001965 map->m_flags = bh->b_state & BH_FLAGS;
1966 return true;
Jan Kara4e7ea812013-06-04 13:17:40 -04001967 }
1968
Jan Kara09930042013-08-17 09:57:56 -04001969 /* Don't go larger than mballoc is willing to allocate */
1970 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1971 return false;
1972
Jan Kara4e7ea812013-06-04 13:17:40 -04001973 /* Can we merge the block to our big extent? */
1974 if (lblk == map->m_lblk + map->m_len &&
Jan Kara09930042013-08-17 09:57:56 -04001975 (bh->b_state & BH_FLAGS) == map->m_flags) {
Jan Kara4e7ea812013-06-04 13:17:40 -04001976 map->m_len++;
Jan Kara09930042013-08-17 09:57:56 -04001977 return true;
Jan Kara4e7ea812013-06-04 13:17:40 -04001978 }
Jan Kara09930042013-08-17 09:57:56 -04001979 return false;
Jan Kara4e7ea812013-06-04 13:17:40 -04001980}
1981
Jan Kara5f1132b2013-08-17 10:02:33 -04001982/*
1983 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1984 *
1985 * @mpd - extent of blocks for mapping
1986 * @head - the first buffer in the page
1987 * @bh - buffer we should start processing from
1988 * @lblk - logical number of the block in the file corresponding to @bh
1989 *
1990 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1991 * the page for IO if all buffers in this page were mapped and there's no
1992 * accumulated extent of buffers to map or add buffers in the page to the
1993 * extent of buffers to map. The function returns 1 if the caller can continue
1994 * by processing the next page, 0 if it should stop adding buffers to the
1995 * extent to map because we cannot extend it anymore. It can also return value
1996 * < 0 in case of error during IO submission.
1997 */
1998static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1999 struct buffer_head *head,
2000 struct buffer_head *bh,
2001 ext4_lblk_t lblk)
Jan Kara4e7ea812013-06-04 13:17:40 -04002002{
2003 struct inode *inode = mpd->inode;
Jan Kara5f1132b2013-08-17 10:02:33 -04002004 int err;
Jan Kara4e7ea812013-06-04 13:17:40 -04002005 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
2006 >> inode->i_blkbits;
2007
2008 do {
2009 BUG_ON(buffer_locked(bh));
2010
Jan Kara09930042013-08-17 09:57:56 -04002011 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
Jan Kara4e7ea812013-06-04 13:17:40 -04002012 /* Found extent to map? */
2013 if (mpd->map.m_len)
Jan Kara5f1132b2013-08-17 10:02:33 -04002014 return 0;
Jan Kara09930042013-08-17 09:57:56 -04002015 /* Everything mapped so far and we hit EOF */
Jan Kara5f1132b2013-08-17 10:02:33 -04002016 break;
Jan Kara4e7ea812013-06-04 13:17:40 -04002017 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002018 } while (lblk++, (bh = bh->b_this_page) != head);
Jan Kara5f1132b2013-08-17 10:02:33 -04002019 /* So far everything mapped? Submit the page for IO. */
2020 if (mpd->map.m_len == 0) {
2021 err = mpage_submit_page(mpd, head->b_page);
2022 if (err < 0)
2023 return err;
2024 }
2025 return lblk < blocks;
Jan Kara4e7ea812013-06-04 13:17:40 -04002026}
2027
2028/*
2029 * mpage_map_buffers - update buffers corresponding to changed extent and
2030 * submit fully mapped pages for IO
2031 *
2032 * @mpd - description of extent to map, on return next extent to map
2033 *
2034 * Scan buffers corresponding to changed extent (we expect corresponding pages
2035 * to be already locked) and update buffer state according to new extent state.
2036 * We map delalloc buffers to their physical location, clear unwritten bits,
Lukas Czerner556615d2014-04-20 23:45:47 -04002037 * and mark buffers as uninit when we perform writes to unwritten extents
Jan Kara4e7ea812013-06-04 13:17:40 -04002038 * and do extent conversion after IO is finished. If the last page is not fully
2039 * mapped, we update @map to the next extent in the last page that needs
2040 * mapping. Otherwise we submit the page for IO.
2041 */
2042static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2043{
2044 struct pagevec pvec;
2045 int nr_pages, i;
2046 struct inode *inode = mpd->inode;
2047 struct buffer_head *head, *bh;
2048 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
Jan Kara4e7ea812013-06-04 13:17:40 -04002049 pgoff_t start, end;
2050 ext4_lblk_t lblk;
2051 sector_t pblock;
2052 int err;
2053
2054 start = mpd->map.m_lblk >> bpp_bits;
2055 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2056 lblk = start << bpp_bits;
2057 pblock = mpd->map.m_pblk;
2058
2059 pagevec_init(&pvec, 0);
2060 while (start <= end) {
2061 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2062 PAGEVEC_SIZE);
2063 if (nr_pages == 0)
2064 break;
2065 for (i = 0; i < nr_pages; i++) {
2066 struct page *page = pvec.pages[i];
2067
2068 if (page->index > end)
2069 break;
Anatol Pomozov70261f52013-08-28 14:40:12 -04002070 /* Up to 'end' pages must be contiguous */
Jan Kara4e7ea812013-06-04 13:17:40 -04002071 BUG_ON(page->index != start);
2072 bh = head = page_buffers(page);
2073 do {
2074 if (lblk < mpd->map.m_lblk)
2075 continue;
2076 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2077 /*
2078 * Buffer after end of mapped extent.
2079 * Find next buffer in the page to map.
2080 */
2081 mpd->map.m_len = 0;
2082 mpd->map.m_flags = 0;
Jan Kara5f1132b2013-08-17 10:02:33 -04002083 /*
2084 * FIXME: If dioread_nolock supports
2085 * blocksize < pagesize, we need to make
2086 * sure we add size mapped so far to
2087 * io_end->size as the following call
2088 * can submit the page for IO.
2089 */
2090 err = mpage_process_page_bufs(mpd, head,
2091 bh, lblk);
Jan Kara4e7ea812013-06-04 13:17:40 -04002092 pagevec_release(&pvec);
Jan Kara5f1132b2013-08-17 10:02:33 -04002093 if (err > 0)
2094 err = 0;
2095 return err;
Jan Kara4e7ea812013-06-04 13:17:40 -04002096 }
2097 if (buffer_delay(bh)) {
2098 clear_buffer_delay(bh);
2099 bh->b_blocknr = pblock++;
2100 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002101 clear_buffer_unwritten(bh);
Jan Kara5f1132b2013-08-17 10:02:33 -04002102 } while (lblk++, (bh = bh->b_this_page) != head);
Jan Kara4e7ea812013-06-04 13:17:40 -04002103
2104 /*
2105 * FIXME: This is going to break if dioread_nolock
2106 * supports blocksize < pagesize as we will try to
2107 * convert potentially unmapped parts of inode.
2108 */
2109 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2110 /* Page fully mapped - let IO run! */
2111 err = mpage_submit_page(mpd, page);
2112 if (err < 0) {
2113 pagevec_release(&pvec);
2114 return err;
2115 }
2116 start++;
2117 }
2118 pagevec_release(&pvec);
2119 }
2120 /* Extent fully mapped and matches with page boundary. We are done. */
2121 mpd->map.m_len = 0;
2122 mpd->map.m_flags = 0;
2123 return 0;
2124}
2125
2126static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2127{
2128 struct inode *inode = mpd->inode;
2129 struct ext4_map_blocks *map = &mpd->map;
2130 int get_blocks_flags;
Lukas Czerner090f32e2014-04-20 23:44:47 -04002131 int err, dioread_nolock;
Jan Kara4e7ea812013-06-04 13:17:40 -04002132
2133 trace_ext4_da_write_pages_extent(inode, map);
2134 /*
2135 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
Lukas Czerner556615d2014-04-20 23:45:47 -04002136 * to convert an unwritten extent to be initialized (in the case
Jan Kara4e7ea812013-06-04 13:17:40 -04002137 * where we have written into one or more preallocated blocks). It is
2138 * possible that we're going to need more metadata blocks than
2139 * previously reserved. However we must not fail because we're in
2140 * writeback and there is nothing we can do about it so it might result
2141 * in data loss. So use reserved blocks to allocate metadata if
2142 * possible.
2143 *
2144 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2145 * in question are delalloc blocks. This affects functions in many
2146 * different parts of the allocation call path. This flag exists
2147 * primarily because we don't want to change *many* call functions, so
2148 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2149 * once the inode's allocation semaphore is taken.
2150 */
2151 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2152 EXT4_GET_BLOCKS_METADATA_NOFAIL;
Lukas Czerner090f32e2014-04-20 23:44:47 -04002153 dioread_nolock = ext4_should_dioread_nolock(inode);
2154 if (dioread_nolock)
Jan Kara4e7ea812013-06-04 13:17:40 -04002155 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2156 if (map->m_flags & (1 << BH_Delay))
2157 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2158
2159 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2160 if (err < 0)
2161 return err;
Lukas Czerner090f32e2014-04-20 23:44:47 -04002162 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
Jan Kara6b523df2013-06-04 13:21:11 -04002163 if (!mpd->io_submit.io_end->handle &&
2164 ext4_handle_valid(handle)) {
2165 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2166 handle->h_rsv_handle = NULL;
2167 }
Jan Kara3613d222013-06-04 13:19:34 -04002168 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
Jan Kara6b523df2013-06-04 13:21:11 -04002169 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002170
2171 BUG_ON(map->m_len == 0);
2172 if (map->m_flags & EXT4_MAP_NEW) {
2173 struct block_device *bdev = inode->i_sb->s_bdev;
2174 int i;
2175
2176 for (i = 0; i < map->m_len; i++)
2177 unmap_underlying_metadata(bdev, map->m_pblk + i);
2178 }
2179 return 0;
2180}
2181
2182/*
2183 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2184 * mpd->len and submit pages underlying it for IO
2185 *
2186 * @handle - handle for journal operations
2187 * @mpd - extent to map
Jan Kara7534e852013-10-16 08:26:08 -04002188 * @give_up_on_write - we set this to true iff there is a fatal error and there
2189 * is no hope of writing the data. The caller should discard
2190 * dirty pages to avoid infinite loops.
Jan Kara4e7ea812013-06-04 13:17:40 -04002191 *
2192 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2193 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2194 * them to initialized or split the described range from larger unwritten
2195 * extent. Note that we need not map all the described range since allocation
2196 * can return less blocks or the range is covered by more unwritten extents. We
2197 * cannot map more because we are limited by reserved transaction credits. On
2198 * the other hand we always make sure that the last touched page is fully
2199 * mapped so that it can be written out (and thus forward progress is
2200 * guaranteed). After mapping we submit all mapped pages for IO.
2201 */
2202static int mpage_map_and_submit_extent(handle_t *handle,
Theodore Ts'ocb530542013-07-01 08:12:40 -04002203 struct mpage_da_data *mpd,
2204 bool *give_up_on_write)
Jan Kara4e7ea812013-06-04 13:17:40 -04002205{
2206 struct inode *inode = mpd->inode;
2207 struct ext4_map_blocks *map = &mpd->map;
2208 int err;
2209 loff_t disksize;
2210
2211 mpd->io_submit.io_end->offset =
2212 ((loff_t)map->m_lblk) << inode->i_blkbits;
Jan Kara27d7c4e2013-07-05 21:57:22 -04002213 do {
Jan Kara4e7ea812013-06-04 13:17:40 -04002214 err = mpage_map_one_extent(handle, mpd);
2215 if (err < 0) {
2216 struct super_block *sb = inode->i_sb;
2217
Theodore Ts'ocb530542013-07-01 08:12:40 -04002218 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2219 goto invalidate_dirty_pages;
Jan Kara4e7ea812013-06-04 13:17:40 -04002220 /*
Theodore Ts'ocb530542013-07-01 08:12:40 -04002221 * Let the uper layers retry transient errors.
2222 * In the case of ENOSPC, if ext4_count_free_blocks()
2223 * is non-zero, a commit should free up blocks.
Jan Kara4e7ea812013-06-04 13:17:40 -04002224 */
Theodore Ts'ocb530542013-07-01 08:12:40 -04002225 if ((err == -ENOMEM) ||
2226 (err == -ENOSPC && ext4_count_free_clusters(sb)))
2227 return err;
2228 ext4_msg(sb, KERN_CRIT,
2229 "Delayed block allocation failed for "
2230 "inode %lu at logical offset %llu with"
2231 " max blocks %u with error %d",
2232 inode->i_ino,
2233 (unsigned long long)map->m_lblk,
2234 (unsigned)map->m_len, -err);
2235 ext4_msg(sb, KERN_CRIT,
2236 "This should not happen!! Data will "
2237 "be lost\n");
2238 if (err == -ENOSPC)
2239 ext4_print_free_blocks(inode);
2240 invalidate_dirty_pages:
2241 *give_up_on_write = true;
Jan Kara4e7ea812013-06-04 13:17:40 -04002242 return err;
2243 }
2244 /*
2245 * Update buffer state, submit mapped pages, and get us new
2246 * extent to map
2247 */
2248 err = mpage_map_and_submit_buffers(mpd);
2249 if (err < 0)
2250 return err;
Jan Kara27d7c4e2013-07-05 21:57:22 -04002251 } while (map->m_len);
Jan Kara4e7ea812013-06-04 13:17:40 -04002252
Theodore Ts'o622cad12014-04-11 10:35:17 -04002253 /*
2254 * Update on-disk size after IO is submitted. Races with
2255 * truncate are avoided by checking i_size under i_data_sem.
2256 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002257 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
Jan Kara4e7ea812013-06-04 13:17:40 -04002258 if (disksize > EXT4_I(inode)->i_disksize) {
2259 int err2;
Theodore Ts'o622cad12014-04-11 10:35:17 -04002260 loff_t i_size;
Jan Kara4e7ea812013-06-04 13:17:40 -04002261
Theodore Ts'o622cad12014-04-11 10:35:17 -04002262 down_write(&EXT4_I(inode)->i_data_sem);
2263 i_size = i_size_read(inode);
2264 if (disksize > i_size)
2265 disksize = i_size;
2266 if (disksize > EXT4_I(inode)->i_disksize)
2267 EXT4_I(inode)->i_disksize = disksize;
Jan Kara4e7ea812013-06-04 13:17:40 -04002268 err2 = ext4_mark_inode_dirty(handle, inode);
Theodore Ts'o622cad12014-04-11 10:35:17 -04002269 up_write(&EXT4_I(inode)->i_data_sem);
Jan Kara4e7ea812013-06-04 13:17:40 -04002270 if (err2)
2271 ext4_error(inode->i_sb,
2272 "Failed to mark inode %lu dirty",
2273 inode->i_ino);
2274 if (!err)
2275 err = err2;
2276 }
2277 return err;
2278}
2279
2280/*
Jan Karafffb2732013-06-04 13:01:11 -04002281 * Calculate the total number of credits to reserve for one writepages
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002282 * iteration. This is called from ext4_writepages(). We map an extent of
Anatol Pomozov70261f52013-08-28 14:40:12 -04002283 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
Jan Karafffb2732013-06-04 13:01:11 -04002284 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2285 * bpp - 1 blocks in bpp different extents.
2286 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002287static int ext4_da_writepages_trans_blocks(struct inode *inode)
2288{
Jan Karafffb2732013-06-04 13:01:11 -04002289 int bpp = ext4_journal_blocks_per_page(inode);
Mingming Cao525f4ed2008-08-19 22:15:58 -04002290
Jan Karafffb2732013-06-04 13:01:11 -04002291 return ext4_meta_trans_blocks(inode,
2292 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
Mingming Cao525f4ed2008-08-19 22:15:58 -04002293}
Mingming Cao61628a32008-07-11 19:27:31 -04002294
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002295/*
Jan Kara4e7ea812013-06-04 13:17:40 -04002296 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2297 * and underlying extent to map
2298 *
2299 * @mpd - where to look for pages
2300 *
2301 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2302 * IO immediately. When we find a page which isn't mapped we start accumulating
2303 * extent of buffers underlying these pages that needs mapping (formed by
2304 * either delayed or unwritten buffers). We also lock the pages containing
2305 * these buffers. The extent found is returned in @mpd structure (starting at
2306 * mpd->lblk with length mpd->len blocks).
2307 *
2308 * Note that this function can attach bios to one io_end structure which are
2309 * neither logically nor physically contiguous. Although it may seem as an
2310 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2311 * 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 -04002312 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002313static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002314{
Jan Kara4e7ea812013-06-04 13:17:40 -04002315 struct address_space *mapping = mpd->inode->i_mapping;
2316 struct pagevec pvec;
2317 unsigned int nr_pages;
Ming Leiaeac5892013-10-17 18:56:16 -04002318 long left = mpd->wbc->nr_to_write;
Jan Kara4e7ea812013-06-04 13:17:40 -04002319 pgoff_t index = mpd->first_page;
2320 pgoff_t end = mpd->last_page;
2321 int tag;
2322 int i, err = 0;
2323 int blkbits = mpd->inode->i_blkbits;
2324 ext4_lblk_t lblk;
2325 struct buffer_head *head;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002326
Jan Kara4e7ea812013-06-04 13:17:40 -04002327 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002328 tag = PAGECACHE_TAG_TOWRITE;
2329 else
2330 tag = PAGECACHE_TAG_DIRTY;
2331
Jan Kara4e7ea812013-06-04 13:17:40 -04002332 pagevec_init(&pvec, 0);
2333 mpd->map.m_len = 0;
2334 mpd->next_page = index;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002335 while (index <= end) {
Eric Sandeen5b41d922010-10-27 21:30:13 -04002336 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002337 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2338 if (nr_pages == 0)
Jan Kara4e7ea812013-06-04 13:17:40 -04002339 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002340
2341 for (i = 0; i < nr_pages; i++) {
2342 struct page *page = pvec.pages[i];
2343
2344 /*
2345 * At this point, the page may be truncated or
2346 * invalidated (changing page->mapping to NULL), or
2347 * even swizzled back from swapper_space to tmpfs file
2348 * mapping. However, page->index will not change
2349 * because we have a reference on the page.
2350 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002351 if (page->index > end)
2352 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002353
Ming Leiaeac5892013-10-17 18:56:16 -04002354 /*
2355 * Accumulated enough dirty pages? This doesn't apply
2356 * to WB_SYNC_ALL mode. For integrity sync we have to
2357 * keep going because someone may be concurrently
2358 * dirtying pages, and we might have synced a lot of
2359 * newly appeared dirty pages, but have not synced all
2360 * of the old dirty pages.
2361 */
2362 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2363 goto out;
2364
Jan Kara4e7ea812013-06-04 13:17:40 -04002365 /* If we can't merge this page, we are done. */
2366 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2367 goto out;
Theodore Ts'o78aaced2011-02-26 14:09:14 -05002368
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002369 lock_page(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002370 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002371 * If the page is no longer dirty, or its mapping no
2372 * longer corresponds to inode we are writing (which
2373 * means it has been truncated or invalidated), or the
2374 * page is already under writeback and we are not doing
2375 * a data integrity writeback, skip the page
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002376 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002377 if (!PageDirty(page) ||
2378 (PageWriteback(page) &&
Jan Kara4e7ea812013-06-04 13:17:40 -04002379 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002380 unlikely(page->mapping != mapping)) {
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002381 unlock_page(page);
2382 continue;
2383 }
2384
Darrick J. Wong7cb1a532011-05-18 13:53:20 -04002385 wait_on_page_writeback(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002386 BUG_ON(PageWriteback(page));
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002387
Jan Kara4e7ea812013-06-04 13:17:40 -04002388 if (mpd->map.m_len == 0)
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002389 mpd->first_page = page->index;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002390 mpd->next_page = page->index + 1;
Jan Karaf8bec372013-01-28 12:55:08 -05002391 /* Add all dirty buffers to mpd */
Jan Kara4e7ea812013-06-04 13:17:40 -04002392 lblk = ((ext4_lblk_t)page->index) <<
2393 (PAGE_CACHE_SHIFT - blkbits);
Jan Karaf8bec372013-01-28 12:55:08 -05002394 head = page_buffers(page);
Jan Kara5f1132b2013-08-17 10:02:33 -04002395 err = mpage_process_page_bufs(mpd, head, head, lblk);
2396 if (err <= 0)
Jan Kara4e7ea812013-06-04 13:17:40 -04002397 goto out;
Jan Kara5f1132b2013-08-17 10:02:33 -04002398 err = 0;
Ming Leiaeac5892013-10-17 18:56:16 -04002399 left--;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002400 }
2401 pagevec_release(&pvec);
2402 cond_resched();
2403 }
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002404 return 0;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002405out:
2406 pagevec_release(&pvec);
Jan Kara4e7ea812013-06-04 13:17:40 -04002407 return err;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002408}
2409
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002410static int __writepage(struct page *page, struct writeback_control *wbc,
2411 void *data)
2412{
2413 struct address_space *mapping = data;
2414 int ret = ext4_writepage(page, wbc);
2415 mapping_set_error(mapping, ret);
2416 return ret;
2417}
2418
2419static int ext4_writepages(struct address_space *mapping,
2420 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002421{
Jan Kara4e7ea812013-06-04 13:17:40 -04002422 pgoff_t writeback_index = 0;
2423 long nr_to_write = wbc->nr_to_write;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002424 int range_whole = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002425 int cycled = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002426 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002427 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002428 struct inode *inode = mapping->host;
Jan Kara6b523df2013-06-04 13:21:11 -04002429 int needed_blocks, rsv_blocks = 0, ret = 0;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002430 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Jan Kara4e7ea812013-06-04 13:17:40 -04002431 bool done;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002432 struct blk_plug plug;
Theodore Ts'ocb530542013-07-01 08:12:40 -04002433 bool give_up_on_write = false;
Mingming Cao61628a32008-07-11 19:27:31 -04002434
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002435 trace_ext4_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002436
Mingming Cao61628a32008-07-11 19:27:31 -04002437 /*
2438 * No pages to write? This is mainly a kludge to avoid starting
2439 * a transaction for special inodes like journal inode on last iput()
2440 * because that could violate lock ordering on umount
2441 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002442 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Ming Leibbf023c2013-10-30 07:27:16 -04002443 goto out_writepages;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002444
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002445 if (ext4_should_journal_data(inode)) {
2446 struct blk_plug plug;
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002447
2448 blk_start_plug(&plug);
2449 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2450 blk_finish_plug(&plug);
Ming Leibbf023c2013-10-30 07:27:16 -04002451 goto out_writepages;
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002452 }
2453
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002454 /*
2455 * If the filesystem has aborted, it is read-only, so return
2456 * right away instead of dumping stack traces later on that
2457 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002458 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002459 * the latter could be true if the filesystem is mounted
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002460 * read-only, and in that case, ext4_writepages should
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002461 * *never* be called, so if that ever happens, we would want
2462 * the stack trace.
2463 */
Ming Leibbf023c2013-10-30 07:27:16 -04002464 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2465 ret = -EROFS;
2466 goto out_writepages;
2467 }
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002468
Jan Kara6b523df2013-06-04 13:21:11 -04002469 if (ext4_should_dioread_nolock(inode)) {
2470 /*
Anatol Pomozov70261f52013-08-28 14:40:12 -04002471 * We may need to convert up to one extent per block in
Jan Kara6b523df2013-06-04 13:21:11 -04002472 * the page and we may dirty the inode.
2473 */
2474 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2475 }
2476
Jan Kara4e7ea812013-06-04 13:17:40 -04002477 /*
2478 * If we have inline data and arrive here, it means that
2479 * we will soon create the block for the 1st page, so
2480 * we'd better clear the inline data here.
2481 */
2482 if (ext4_has_inline_data(inode)) {
2483 /* Just inode will be modified... */
2484 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2485 if (IS_ERR(handle)) {
2486 ret = PTR_ERR(handle);
2487 goto out_writepages;
2488 }
2489 BUG_ON(ext4_test_inode_state(inode,
2490 EXT4_STATE_MAY_INLINE_DATA));
2491 ext4_destroy_inline_data(handle, inode);
2492 ext4_journal_stop(handle);
2493 }
2494
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002495 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2496 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002497
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002498 if (wbc->range_cyclic) {
Jan Kara4e7ea812013-06-04 13:17:40 -04002499 writeback_index = mapping->writeback_index;
2500 if (writeback_index)
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002501 cycled = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002502 mpd.first_page = writeback_index;
2503 mpd.last_page = -1;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002504 } else {
Jan Kara4e7ea812013-06-04 13:17:40 -04002505 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2506 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002507 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002508
Jan Kara4e7ea812013-06-04 13:17:40 -04002509 mpd.inode = inode;
2510 mpd.wbc = wbc;
2511 ext4_io_submit_init(&mpd.io_submit, wbc);
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002512retry:
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002513 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Jan Kara4e7ea812013-06-04 13:17:40 -04002514 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2515 done = false;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002516 blk_start_plug(&plug);
Jan Kara4e7ea812013-06-04 13:17:40 -04002517 while (!done && mpd.first_page <= mpd.last_page) {
2518 /* For each extent of pages we use new io_end */
2519 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2520 if (!mpd.io_submit.io_end) {
2521 ret = -ENOMEM;
2522 break;
2523 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002524
2525 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002526 * We have two constraints: We find one extent to map and we
2527 * must always write out whole page (makes a difference when
2528 * blocksize < pagesize) so that we don't block on IO when we
2529 * try to write out the rest of the page. Journalled mode is
2530 * not supported by delalloc.
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002531 */
2532 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002533 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002534
Jan Kara4e7ea812013-06-04 13:17:40 -04002535 /* start a new transaction */
Jan Kara6b523df2013-06-04 13:21:11 -04002536 handle = ext4_journal_start_with_reserve(inode,
2537 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
Mingming Cao61628a32008-07-11 19:27:31 -04002538 if (IS_ERR(handle)) {
2539 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002540 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04002541 "%ld pages, ino %lu; err %d", __func__,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002542 wbc->nr_to_write, inode->i_ino, ret);
Jan Kara4e7ea812013-06-04 13:17:40 -04002543 /* Release allocated io_end */
2544 ext4_put_io_end(mpd.io_submit.io_end);
2545 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002546 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002547
Jan Kara4e7ea812013-06-04 13:17:40 -04002548 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2549 ret = mpage_prepare_extent_to_map(&mpd);
2550 if (!ret) {
2551 if (mpd.map.m_len)
Theodore Ts'ocb530542013-07-01 08:12:40 -04002552 ret = mpage_map_and_submit_extent(handle, &mpd,
2553 &give_up_on_write);
Jan Kara4e7ea812013-06-04 13:17:40 -04002554 else {
2555 /*
2556 * We scanned the whole range (or exhausted
2557 * nr_to_write), submitted what was mapped and
2558 * didn't find anything needing mapping. We are
2559 * done.
2560 */
2561 done = true;
2562 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002563 }
Mingming Cao61628a32008-07-11 19:27:31 -04002564 ext4_journal_stop(handle);
Jan Kara4e7ea812013-06-04 13:17:40 -04002565 /* Submit prepared bio */
2566 ext4_io_submit(&mpd.io_submit);
2567 /* Unlock pages we didn't use */
Theodore Ts'ocb530542013-07-01 08:12:40 -04002568 mpage_release_unused_pages(&mpd, give_up_on_write);
Jan Kara4e7ea812013-06-04 13:17:40 -04002569 /* Drop our io_end reference we got from init */
2570 ext4_put_io_end(mpd.io_submit.io_end);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002571
Jan Kara4e7ea812013-06-04 13:17:40 -04002572 if (ret == -ENOSPC && sbi->s_journal) {
2573 /*
2574 * Commit the transaction which would
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002575 * free blocks released in the transaction
2576 * and try again
2577 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002578 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002579 ret = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002580 continue;
2581 }
2582 /* Fatal error - ENOMEM, EIO... */
2583 if (ret)
Mingming Cao61628a32008-07-11 19:27:31 -04002584 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002585 }
Shaohua Li1bce63d2011-10-18 10:55:51 -04002586 blk_finish_plug(&plug);
Jan Kara9c12a832013-09-16 08:24:26 -04002587 if (!ret && !cycled && wbc->nr_to_write > 0) {
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002588 cycled = 1;
Jan Kara4e7ea812013-06-04 13:17:40 -04002589 mpd.last_page = writeback_index - 1;
2590 mpd.first_page = 0;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002591 goto retry;
2592 }
Mingming Cao61628a32008-07-11 19:27:31 -04002593
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002594 /* Update index */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002595 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2596 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002597 * Set the writeback_index so that range_cyclic
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002598 * mode will write it back later
2599 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002600 mapping->writeback_index = mpd.first_page;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002601
Mingming Cao61628a32008-07-11 19:27:31 -04002602out_writepages:
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002603 trace_ext4_writepages_result(inode, wbc, ret,
2604 nr_to_write - wbc->nr_to_write);
Mingming Cao61628a32008-07-11 19:27:31 -04002605 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002606}
2607
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002608static int ext4_nonda_switch(struct super_block *sb)
2609{
Eric Whitney5c1ff332013-04-09 09:27:31 -04002610 s64 free_clusters, dirty_clusters;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002611 struct ext4_sb_info *sbi = EXT4_SB(sb);
2612
2613 /*
2614 * switch to non delalloc mode if we are running low
2615 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08002616 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002617 * accumulated on each CPU without updating global counters
2618 * Delalloc need an accurate free block accounting. So switch
2619 * to non delalloc when we are near to error range.
2620 */
Eric Whitney5c1ff332013-04-09 09:27:31 -04002621 free_clusters =
2622 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2623 dirty_clusters =
2624 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002625 /*
2626 * Start pushing delalloc when 1/2 of free blocks are dirty.
2627 */
Eric Whitney5c1ff332013-04-09 09:27:31 -04002628 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
Miao Xie10ee27a2013-01-10 13:47:57 +08002629 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002630
Eric Whitney5c1ff332013-04-09 09:27:31 -04002631 if (2 * free_clusters < 3 * dirty_clusters ||
2632 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002633 /*
Eric Sandeenc8afb442009-12-23 07:58:12 -05002634 * free block count is less than 150% of dirty blocks
2635 * or free blocks is less than watermark
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002636 */
2637 return 1;
2638 }
2639 return 0;
2640}
2641
Alex Tomas64769242008-07-11 19:27:31 -04002642static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002643 loff_t pos, unsigned len, unsigned flags,
2644 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002645{
Eric Sandeen72b8ab92010-05-16 11:00:00 -04002646 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002647 struct page *page;
2648 pgoff_t index;
Alex Tomas64769242008-07-11 19:27:31 -04002649 struct inode *inode = mapping->host;
2650 handle_t *handle;
2651
2652 index = pos >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002653
2654 if (ext4_nonda_switch(inode->i_sb)) {
2655 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2656 return ext4_write_begin(file, mapping, pos,
2657 len, flags, pagep, fsdata);
2658 }
2659 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002660 trace_ext4_da_write_begin(inode, pos, len, flags);
Tao Ma9c3569b2012-12-10 14:05:57 -05002661
2662 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2663 ret = ext4_da_write_inline_data_begin(mapping, inode,
2664 pos, len, flags,
2665 pagep, fsdata);
2666 if (ret < 0)
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002667 return ret;
2668 if (ret == 1)
2669 return 0;
Tao Ma9c3569b2012-12-10 14:05:57 -05002670 }
2671
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002672 /*
2673 * grab_cache_page_write_begin() can take a long time if the
2674 * system is thrashing due to memory pressure, or if the page
2675 * is being written back. So grab it first before we start
2676 * the transaction handle. This also allows us to allocate
2677 * the page (if needed) without using GFP_NOFS.
2678 */
2679retry_grab:
2680 page = grab_cache_page_write_begin(mapping, index, flags);
2681 if (!page)
2682 return -ENOMEM;
2683 unlock_page(page);
2684
Alex Tomas64769242008-07-11 19:27:31 -04002685 /*
2686 * With delayed allocation, we don't log the i_disksize update
2687 * if there is delayed block allocation. But we still need
2688 * to journalling the i_disksize update if writes to the end
2689 * of file which has an already mapped buffer.
2690 */
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002691retry_journal:
Theodore Ts'o9924a922013-02-08 21:59:22 -05002692 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
Alex Tomas64769242008-07-11 19:27:31 -04002693 if (IS_ERR(handle)) {
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002694 page_cache_release(page);
2695 return PTR_ERR(handle);
Alex Tomas64769242008-07-11 19:27:31 -04002696 }
2697
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002698 lock_page(page);
2699 if (page->mapping != mapping) {
2700 /* The page got truncated from under us */
2701 unlock_page(page);
2702 page_cache_release(page);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002703 ext4_journal_stop(handle);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002704 goto retry_grab;
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002705 }
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002706 /* In case writeback began while the page was unlocked */
Dmitry Monakhov7afe5aa2013-08-28 14:30:47 -04002707 wait_for_stable_page(page);
Alex Tomas64769242008-07-11 19:27:31 -04002708
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002709 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
Alex Tomas64769242008-07-11 19:27:31 -04002710 if (ret < 0) {
2711 unlock_page(page);
2712 ext4_journal_stop(handle);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04002713 /*
2714 * block_write_begin may have instantiated a few blocks
2715 * outside i_size. Trim these off again. Don't need
2716 * i_size_read because we hold i_mutex.
2717 */
2718 if (pos + len > inode->i_size)
Jan Karab9a42072009-12-08 21:24:33 -05002719 ext4_truncate_failed_write(inode);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002720
2721 if (ret == -ENOSPC &&
2722 ext4_should_retry_alloc(inode->i_sb, &retries))
2723 goto retry_journal;
2724
2725 page_cache_release(page);
2726 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002727 }
2728
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002729 *pagep = page;
Alex Tomas64769242008-07-11 19:27:31 -04002730 return ret;
2731}
2732
Mingming Cao632eaea2008-07-11 19:27:31 -04002733/*
2734 * Check if we should update i_disksize
2735 * when write to the end of file but not require block allocation
2736 */
2737static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002738 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04002739{
2740 struct buffer_head *bh;
2741 struct inode *inode = page->mapping->host;
2742 unsigned int idx;
2743 int i;
2744
2745 bh = page_buffers(page);
2746 idx = offset >> inode->i_blkbits;
2747
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002748 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04002749 bh = bh->b_this_page;
2750
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002751 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04002752 return 0;
2753 return 1;
2754}
2755
Alex Tomas64769242008-07-11 19:27:31 -04002756static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002757 struct address_space *mapping,
2758 loff_t pos, unsigned len, unsigned copied,
2759 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002760{
2761 struct inode *inode = mapping->host;
2762 int ret = 0, ret2;
2763 handle_t *handle = ext4_journal_current_handle();
2764 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002765 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002766 int write_mode = (int)(unsigned long)fsdata;
2767
Theodore Ts'o74d553a2013-04-03 12:39:17 -04002768 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2769 return ext4_write_end(file, mapping, pos,
2770 len, copied, page, fsdata);
Mingming Cao632eaea2008-07-11 19:27:31 -04002771
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002772 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04002773 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002774 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04002775
2776 /*
2777 * generic_write_end() will run mark_inode_dirty() if i_size
2778 * changes. So let's piggyback the i_disksize mark_inode_dirty
2779 * into that.
2780 */
Alex Tomas64769242008-07-11 19:27:31 -04002781 new_i_size = pos + copied;
Andrea Arcangeliea51d132011-12-13 21:41:15 -05002782 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
Tao Ma9c3569b2012-12-10 14:05:57 -05002783 if (ext4_has_inline_data(inode) ||
2784 ext4_da_should_update_i_disksize(page, end)) {
Mingming Cao632eaea2008-07-11 19:27:31 -04002785 down_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'of3b59292012-11-15 23:08:57 -05002786 if (new_i_size > EXT4_I(inode)->i_disksize)
Mingming Cao632eaea2008-07-11 19:27:31 -04002787 EXT4_I(inode)->i_disksize = new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002788 up_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002789 /* We need to mark inode dirty even if
2790 * new_i_size is less that inode->i_size
2791 * bu greater than i_disksize.(hint delalloc)
2792 */
2793 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04002794 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002795 }
Tao Ma9c3569b2012-12-10 14:05:57 -05002796
2797 if (write_mode != CONVERT_INLINE_DATA &&
2798 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2799 ext4_has_inline_data(inode))
2800 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2801 page);
2802 else
2803 ret2 = generic_write_end(file, mapping, pos, len, copied,
Alex Tomas64769242008-07-11 19:27:31 -04002804 page, fsdata);
Tao Ma9c3569b2012-12-10 14:05:57 -05002805
Alex Tomas64769242008-07-11 19:27:31 -04002806 copied = ret2;
2807 if (ret2 < 0)
2808 ret = ret2;
2809 ret2 = ext4_journal_stop(handle);
2810 if (!ret)
2811 ret = ret2;
2812
2813 return ret ? ret : copied;
2814}
2815
Lukas Czernerd47992f2013-05-21 23:17:23 -04002816static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2817 unsigned int length)
Alex Tomas64769242008-07-11 19:27:31 -04002818{
Alex Tomas64769242008-07-11 19:27:31 -04002819 /*
2820 * Drop reserved blocks
2821 */
2822 BUG_ON(!PageLocked(page));
2823 if (!page_has_buffers(page))
2824 goto out;
2825
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002826 ext4_da_page_release_reservation(page, offset, length);
Alex Tomas64769242008-07-11 19:27:31 -04002827
2828out:
Lukas Czernerd47992f2013-05-21 23:17:23 -04002829 ext4_invalidatepage(page, offset, length);
Alex Tomas64769242008-07-11 19:27:31 -04002830
2831 return;
2832}
2833
Theodore Ts'occd25062009-02-26 01:04:07 -05002834/*
2835 * Force all delayed allocation blocks to be allocated for a given inode.
2836 */
2837int ext4_alloc_da_blocks(struct inode *inode)
2838{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04002839 trace_ext4_alloc_da_blocks(inode);
2840
Theodore Ts'occd25062009-02-26 01:04:07 -05002841 if (!EXT4_I(inode)->i_reserved_data_blocks &&
2842 !EXT4_I(inode)->i_reserved_meta_blocks)
2843 return 0;
2844
2845 /*
2846 * We do something simple for now. The filemap_flush() will
2847 * also start triggering a write of the data blocks, which is
2848 * not strictly speaking necessary (and for users of
2849 * laptop_mode, not even desirable). However, to do otherwise
2850 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002851 *
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002852 * ext4_writepages() ->
Theodore Ts'occd25062009-02-26 01:04:07 -05002853 * write_cache_pages() ---> (via passed in callback function)
2854 * __mpage_da_writepage() -->
2855 * mpage_add_bh_to_extent()
2856 * mpage_da_map_blocks()
2857 *
2858 * The problem is that write_cache_pages(), located in
2859 * mm/page-writeback.c, marks pages clean in preparation for
2860 * doing I/O, which is not desirable if we're not planning on
2861 * doing I/O at all.
2862 *
2863 * We could call write_cache_pages(), and then redirty all of
Wu Fengguang380cf092010-11-11 19:23:29 +08002864 * the pages by calling redirty_page_for_writepage() but that
Theodore Ts'occd25062009-02-26 01:04:07 -05002865 * would be ugly in the extreme. So instead we would need to
2866 * replicate parts of the code in the above functions,
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002867 * simplifying them because we wouldn't actually intend to
Theodore Ts'occd25062009-02-26 01:04:07 -05002868 * write out the pages, but rather only collect contiguous
2869 * logical block extents, call the multi-block allocator, and
2870 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002871 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002872 * For now, though, we'll cheat by calling filemap_flush(),
2873 * which will map the blocks, and start the I/O, but not
2874 * actually wait for the I/O to complete.
2875 */
2876 return filemap_flush(inode->i_mapping);
2877}
Alex Tomas64769242008-07-11 19:27:31 -04002878
2879/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002880 * bmap() is special. It gets used by applications such as lilo and by
2881 * the swapper to find the on-disk block of a specific piece of data.
2882 *
2883 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07002884 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002885 * filesystem and enables swap, then they may get a nasty shock when the
2886 * data getting swapped to that swapfile suddenly gets overwritten by
2887 * the original zero's written out previously to the journal and
2888 * awaiting writeback in the kernel's buffer cache.
2889 *
2890 * So, if we see any bmap calls here on a modified, data-journaled file,
2891 * take extra steps to flush any blocks which might be in the cache.
2892 */
Mingming Cao617ba132006-10-11 01:20:53 -07002893static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002894{
2895 struct inode *inode = mapping->host;
2896 journal_t *journal;
2897 int err;
2898
Tao Ma46c7f252012-12-10 14:04:52 -05002899 /*
2900 * We can get here for an inline file via the FIBMAP ioctl
2901 */
2902 if (ext4_has_inline_data(inode))
2903 return 0;
2904
Alex Tomas64769242008-07-11 19:27:31 -04002905 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2906 test_opt(inode->i_sb, DELALLOC)) {
2907 /*
2908 * With delalloc we want to sync the file
2909 * so that we can make sure we allocate
2910 * blocks for file
2911 */
2912 filemap_write_and_wait(mapping);
2913 }
2914
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002915 if (EXT4_JOURNAL(inode) &&
2916 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002917 /*
2918 * This is a REALLY heavyweight approach, but the use of
2919 * bmap on dirty files is expected to be extremely rare:
2920 * only if we run lilo or swapon on a freshly made file
2921 * do we expect this to happen.
2922 *
2923 * (bmap requires CAP_SYS_RAWIO so this does not
2924 * represent an unprivileged user DOS attack --- we'd be
2925 * in trouble if mortal users could trigger this path at
2926 * will.)
2927 *
Mingming Cao617ba132006-10-11 01:20:53 -07002928 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002929 * regular files. If somebody wants to bmap a directory
2930 * or symlink and gets confused because the buffer
2931 * hasn't yet been flushed to disk, they deserve
2932 * everything they get.
2933 */
2934
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002935 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
Mingming Cao617ba132006-10-11 01:20:53 -07002936 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07002937 jbd2_journal_lock_updates(journal);
2938 err = jbd2_journal_flush(journal);
2939 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002940
2941 if (err)
2942 return 0;
2943 }
2944
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002945 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002946}
2947
Mingming Cao617ba132006-10-11 01:20:53 -07002948static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002949{
Tao Ma46c7f252012-12-10 14:04:52 -05002950 int ret = -EAGAIN;
2951 struct inode *inode = page->mapping->host;
2952
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002953 trace_ext4_readpage(page);
Tao Ma46c7f252012-12-10 14:04:52 -05002954
2955 if (ext4_has_inline_data(inode))
2956 ret = ext4_readpage_inline(inode, page);
2957
2958 if (ret == -EAGAIN)
2959 return mpage_readpage(page, ext4_get_block);
2960
2961 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002962}
2963
2964static int
Mingming Cao617ba132006-10-11 01:20:53 -07002965ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002966 struct list_head *pages, unsigned nr_pages)
2967{
Tao Ma46c7f252012-12-10 14:04:52 -05002968 struct inode *inode = mapping->host;
2969
2970 /* If the file has inline data, no need to do readpages. */
2971 if (ext4_has_inline_data(inode))
2972 return 0;
2973
Mingming Cao617ba132006-10-11 01:20:53 -07002974 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002975}
2976
Lukas Czernerd47992f2013-05-21 23:17:23 -04002977static void ext4_invalidatepage(struct page *page, unsigned int offset,
2978 unsigned int length)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002979{
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002980 trace_ext4_invalidatepage(page, offset, length);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002981
Jan Kara4520fb32012-12-25 13:28:54 -05002982 /* No journalling happens on data buffers when this function is used */
2983 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2984
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002985 block_invalidatepage(page, offset, length);
Jan Kara4520fb32012-12-25 13:28:54 -05002986}
2987
Jan Kara53e87262012-12-25 13:29:52 -05002988static int __ext4_journalled_invalidatepage(struct page *page,
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002989 unsigned int offset,
2990 unsigned int length)
Jan Kara4520fb32012-12-25 13:28:54 -05002991{
2992 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2993
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002994 trace_ext4_journalled_invalidatepage(page, offset, length);
Jan Kara4520fb32012-12-25 13:28:54 -05002995
Jiaying Zhang744692d2010-03-04 16:14:02 -05002996 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002997 * If it's a full truncate we just forget about the pending dirtying
2998 */
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002999 if (offset == 0 && length == PAGE_CACHE_SIZE)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003000 ClearPageChecked(page);
3001
Lukas Czernerca99fdd2013-05-21 23:25:01 -04003002 return jbd2_journal_invalidatepage(journal, page, offset, length);
Jan Kara53e87262012-12-25 13:29:52 -05003003}
3004
3005/* Wrapper for aops... */
3006static void ext4_journalled_invalidatepage(struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -04003007 unsigned int offset,
3008 unsigned int length)
Jan Kara53e87262012-12-25 13:29:52 -05003009{
Lukas Czernerca99fdd2013-05-21 23:25:01 -04003010 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003011}
3012
Mingming Cao617ba132006-10-11 01:20:53 -07003013static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003014{
Mingming Cao617ba132006-10-11 01:20:53 -07003015 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003016
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003017 trace_ext4_releasepage(page);
3018
Jan Karae1c36592013-03-10 22:19:00 -04003019 /* Page has dirty journalled data -> cannot release */
3020 if (PageChecked(page))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003021 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05003022 if (journal)
3023 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3024 else
3025 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003026}
3027
3028/*
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003029 * ext4_get_block used when preparing for a DIO write or buffer write.
3030 * We allocate an uinitialized extent if blocks haven't been allocated.
3031 * The extent will be converted to initialized after the IO is complete.
3032 */
Tao Maf19d5872012-12-10 14:05:51 -05003033int ext4_get_block_write(struct inode *inode, sector_t iblock,
Mingming Cao4c0425f2009-09-28 15:48:41 -04003034 struct buffer_head *bh_result, int create)
3035{
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003036 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003037 inode->i_ino, create);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003038 return _ext4_get_block(inode, iblock, bh_result,
3039 EXT4_GET_BLOCKS_IO_CREATE_EXT);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003040}
3041
Zheng Liu729f52c2012-07-09 16:29:29 -04003042static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05003043 struct buffer_head *bh_result, int create)
Zheng Liu729f52c2012-07-09 16:29:29 -04003044{
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05003045 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3046 inode->i_ino, create);
3047 return _ext4_get_block(inode, iblock, bh_result,
3048 EXT4_GET_BLOCKS_NO_LOCK);
Zheng Liu729f52c2012-07-09 16:29:29 -04003049}
3050
Mingming Cao4c0425f2009-09-28 15:48:41 -04003051static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003052 ssize_t size, void *private)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003053{
3054 ext4_io_end_t *io_end = iocb->private;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003055
Jan Kara97a851e2013-06-04 11:58:58 -04003056 /* if not async direct IO just return */
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003057 if (!io_end)
Jan Kara97a851e2013-06-04 11:58:58 -04003058 return;
Mingming4b70df12009-11-03 14:44:54 -05003059
Zheng Liu88635ca2011-12-28 19:00:25 -05003060 ext_debug("ext4_end_io_dio(): io_end 0x%p "
Joe Perchesace36ad2012-03-19 23:11:43 -04003061 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003062 iocb->private, io_end->inode->i_ino, iocb, offset,
3063 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003064
Theodore Ts'ob5a7e972011-12-12 10:53:02 -05003065 iocb->private = NULL;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003066 io_end->offset = offset;
3067 io_end->size = size;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003068 ext4_put_io_end(io_end);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003069}
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003070
Mingming Cao4c0425f2009-09-28 15:48:41 -04003071/*
3072 * For ext4 extent files, ext4 will do direct-io write to holes,
3073 * preallocated extents, and those write extend the file, no need to
3074 * fall back to buffered IO.
3075 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003076 * For holes, we fallocate those blocks, mark them as unwritten
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003077 * If those blocks were preallocated, we mark sure they are split, but
Lukas Czerner556615d2014-04-20 23:45:47 -04003078 * still keep the range to write as unwritten.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003079 *
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003080 * The unwritten extents will be converted to written when DIO is completed.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003081 * For async direct IO, since the IO may still pending when return, we
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003082 * set up an end_io call back function, which will do the conversion
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003083 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003084 *
3085 * If the O_DIRECT write will extend the file then add this inode to the
3086 * orphan list. So recovery will truncate it back to the original size
3087 * if the machine crashes during the write.
3088 *
3089 */
3090static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
3091 const struct iovec *iov, loff_t offset,
3092 unsigned long nr_segs)
3093{
3094 struct file *file = iocb->ki_filp;
3095 struct inode *inode = file->f_mapping->host;
3096 ssize_t ret;
3097 size_t count = iov_length(iov, nr_segs);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003098 int overwrite = 0;
3099 get_block_t *get_block_func = NULL;
3100 int dio_flags = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003101 loff_t final_size = offset + count;
Jan Kara97a851e2013-06-04 11:58:58 -04003102 ext4_io_end_t *io_end = NULL;
Zheng Liu729f52c2012-07-09 16:29:29 -04003103
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003104 /* Use the old path for reads and writes beyond i_size. */
3105 if (rw != WRITE || final_size > inode->i_size)
3106 return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003107
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003108 BUG_ON(iocb->private == NULL);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003109
Jan Karae8340392013-06-04 14:27:38 -04003110 /*
3111 * Make all waiters for direct IO properly wait also for extent
3112 * conversion. This also disallows race between truncate() and
3113 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3114 */
3115 if (rw == WRITE)
3116 atomic_inc(&inode->i_dio_count);
3117
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003118 /* If we do a overwrite dio, i_mutex locking can be released */
3119 overwrite = *((int *)iocb->private);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003120
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003121 if (overwrite) {
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003122 down_read(&EXT4_I(inode)->i_data_sem);
3123 mutex_unlock(&inode->i_mutex);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003124 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003125
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003126 /*
3127 * We could direct write to holes and fallocate.
3128 *
3129 * Allocated blocks to fill the hole are marked as
Lukas Czerner556615d2014-04-20 23:45:47 -04003130 * unwritten to prevent parallel buffered read to expose
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003131 * the stale data before DIO complete the data IO.
3132 *
3133 * As to previously fallocated extents, ext4 get_block will
3134 * just simply mark the buffer mapped but still keep the
Lukas Czerner556615d2014-04-20 23:45:47 -04003135 * extents unwritten.
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003136 *
3137 * For non AIO case, we will convert those unwritten extents
3138 * to written after return back from blockdev_direct_IO.
3139 *
3140 * For async DIO, the conversion needs to be deferred when the
3141 * IO is completed. The ext4 end_io callback function will be
3142 * called to take care of the conversion work. Here for async
3143 * case, we allocate an io_end structure to hook to the iocb.
3144 */
3145 iocb->private = NULL;
3146 ext4_inode_aio_set(inode, NULL);
3147 if (!is_sync_kiocb(iocb)) {
Jan Kara97a851e2013-06-04 11:58:58 -04003148 io_end = ext4_init_io_end(inode, GFP_NOFS);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003149 if (!io_end) {
3150 ret = -ENOMEM;
3151 goto retake_lock;
3152 }
Jan Kara97a851e2013-06-04 11:58:58 -04003153 /*
3154 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3155 */
3156 iocb->private = ext4_get_io_end(io_end);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003157 /*
3158 * we save the io structure for current async direct
3159 * IO, so that later ext4_map_blocks() could flag the
3160 * io structure whether there is a unwritten extents
3161 * needs to be converted when IO is completed.
3162 */
3163 ext4_inode_aio_set(inode, io_end);
3164 }
3165
3166 if (overwrite) {
3167 get_block_func = ext4_get_block_write_nolock;
3168 } else {
3169 get_block_func = ext4_get_block_write;
3170 dio_flags = DIO_LOCKING;
3171 }
3172 ret = __blockdev_direct_IO(rw, iocb, inode,
3173 inode->i_sb->s_bdev, iov,
3174 offset, nr_segs,
3175 get_block_func,
3176 ext4_end_io_dio,
3177 NULL,
3178 dio_flags);
3179
Theodore Ts'oa5499842013-05-11 19:07:42 -04003180 /*
Jan Kara97a851e2013-06-04 11:58:58 -04003181 * Put our reference to io_end. This can free the io_end structure e.g.
3182 * in sync IO case or in case of error. It can even perform extent
3183 * conversion if all bios we submitted finished before we got here.
3184 * Note that in that case iocb->private can be already set to NULL
3185 * here.
Theodore Ts'oa5499842013-05-11 19:07:42 -04003186 */
Jan Kara97a851e2013-06-04 11:58:58 -04003187 if (io_end) {
3188 ext4_inode_aio_set(inode, NULL);
3189 ext4_put_io_end(io_end);
3190 /*
3191 * When no IO was submitted ext4_end_io_dio() was not
3192 * called so we have to put iocb's reference.
3193 */
3194 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3195 WARN_ON(iocb->private != io_end);
3196 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
Jan Kara97a851e2013-06-04 11:58:58 -04003197 ext4_put_io_end(io_end);
3198 iocb->private = NULL;
3199 }
3200 }
3201 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003202 EXT4_STATE_DIO_UNWRITTEN)) {
3203 int err;
3204 /*
3205 * for non AIO case, since the IO is already
3206 * completed, we could do the conversion right here
3207 */
Jan Kara6b523df2013-06-04 13:21:11 -04003208 err = ext4_convert_unwritten_extents(NULL, inode,
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003209 offset, ret);
3210 if (err < 0)
3211 ret = err;
3212 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3213 }
3214
3215retake_lock:
Jan Karae8340392013-06-04 14:27:38 -04003216 if (rw == WRITE)
3217 inode_dio_done(inode);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003218 /* take i_mutex locking again if we do a ovewrite dio */
3219 if (overwrite) {
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003220 up_read(&EXT4_I(inode)->i_data_sem);
3221 mutex_lock(&inode->i_mutex);
3222 }
3223
3224 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003225}
3226
3227static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
3228 const struct iovec *iov, loff_t offset,
3229 unsigned long nr_segs)
3230{
3231 struct file *file = iocb->ki_filp;
3232 struct inode *inode = file->f_mapping->host;
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003233 ssize_t ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003234
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003235 /*
3236 * If we are doing data journalling we don't support O_DIRECT
3237 */
3238 if (ext4_should_journal_data(inode))
3239 return 0;
3240
Tao Ma46c7f252012-12-10 14:04:52 -05003241 /* Let buffer I/O handle the inline data case. */
3242 if (ext4_has_inline_data(inode))
3243 return 0;
3244
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003245 trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003246 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003247 ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs);
3248 else
3249 ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs);
3250 trace_ext4_direct_IO_exit(inode, offset,
3251 iov_length(iov, nr_segs), rw, ret);
3252 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003253}
3254
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003255/*
Mingming Cao617ba132006-10-11 01:20:53 -07003256 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003257 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3258 * much here because ->set_page_dirty is called under VFS locks. The page is
3259 * not necessarily locked.
3260 *
3261 * We cannot just dirty the page and leave attached buffers clean, because the
3262 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3263 * or jbddirty because all the journalling code will explode.
3264 *
3265 * So what we do is to mark the page "pending dirty" and next time writepage
3266 * is called, propagate that into the buffers appropriately.
3267 */
Mingming Cao617ba132006-10-11 01:20:53 -07003268static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003269{
3270 SetPageChecked(page);
3271 return __set_page_dirty_nobuffers(page);
3272}
3273
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003274static const struct address_space_operations ext4_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003275 .readpage = ext4_readpage,
3276 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003277 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003278 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003279 .write_begin = ext4_write_begin,
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003280 .write_end = ext4_write_end,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003281 .bmap = ext4_bmap,
3282 .invalidatepage = ext4_invalidatepage,
3283 .releasepage = ext4_releasepage,
3284 .direct_IO = ext4_direct_IO,
3285 .migratepage = buffer_migrate_page,
3286 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003287 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003288};
3289
Mingming Cao617ba132006-10-11 01:20:53 -07003290static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003291 .readpage = ext4_readpage,
3292 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003293 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003294 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003295 .write_begin = ext4_write_begin,
3296 .write_end = ext4_journalled_write_end,
3297 .set_page_dirty = ext4_journalled_set_page_dirty,
3298 .bmap = ext4_bmap,
Jan Kara4520fb32012-12-25 13:28:54 -05003299 .invalidatepage = ext4_journalled_invalidatepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003300 .releasepage = ext4_releasepage,
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003301 .direct_IO = ext4_direct_IO,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003302 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003303 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003304};
3305
Alex Tomas64769242008-07-11 19:27:31 -04003306static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003307 .readpage = ext4_readpage,
3308 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003309 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003310 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003311 .write_begin = ext4_da_write_begin,
3312 .write_end = ext4_da_write_end,
3313 .bmap = ext4_bmap,
3314 .invalidatepage = ext4_da_invalidatepage,
3315 .releasepage = ext4_releasepage,
3316 .direct_IO = ext4_direct_IO,
3317 .migratepage = buffer_migrate_page,
3318 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003319 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003320};
3321
Mingming Cao617ba132006-10-11 01:20:53 -07003322void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003323{
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003324 switch (ext4_inode_journal_mode(inode)) {
3325 case EXT4_INODE_ORDERED_DATA_MODE:
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003326 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003327 break;
3328 case EXT4_INODE_WRITEBACK_DATA_MODE:
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003329 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003330 break;
3331 case EXT4_INODE_JOURNAL_DATA_MODE:
Mingming Cao617ba132006-10-11 01:20:53 -07003332 inode->i_mapping->a_ops = &ext4_journalled_aops;
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003333 return;
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003334 default:
3335 BUG();
3336 }
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003337 if (test_opt(inode->i_sb, DELALLOC))
3338 inode->i_mapping->a_ops = &ext4_da_aops;
3339 else
3340 inode->i_mapping->a_ops = &ext4_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003341}
3342
Lukas Czernerd863dc32013-05-27 23:32:35 -04003343/*
Lukas Czernerd863dc32013-05-27 23:32:35 -04003344 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3345 * starting from file offset 'from'. The range to be zero'd must
3346 * be contained with in one block. If the specified range exceeds
3347 * the end of the block it will be shortened to end of the block
3348 * that cooresponds to 'from'
3349 */
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003350static int ext4_block_zero_page_range(handle_t *handle,
Lukas Czernerd863dc32013-05-27 23:32:35 -04003351 struct address_space *mapping, loff_t from, loff_t length)
3352{
3353 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3354 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3355 unsigned blocksize, max, pos;
3356 ext4_lblk_t iblock;
3357 struct inode *inode = mapping->host;
3358 struct buffer_head *bh;
3359 struct page *page;
3360 int err = 0;
3361
3362 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3363 mapping_gfp_mask(mapping) & ~__GFP_FS);
3364 if (!page)
3365 return -ENOMEM;
3366
3367 blocksize = inode->i_sb->s_blocksize;
3368 max = blocksize - (offset & (blocksize - 1));
3369
3370 /*
3371 * correct length if it does not fall between
3372 * 'from' and the end of the block
3373 */
3374 if (length > max || length < 0)
3375 length = max;
3376
3377 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3378
3379 if (!page_has_buffers(page))
3380 create_empty_buffers(page, blocksize, 0);
3381
3382 /* Find the buffer that contains "offset" */
3383 bh = page_buffers(page);
3384 pos = blocksize;
3385 while (offset >= pos) {
3386 bh = bh->b_this_page;
3387 iblock++;
3388 pos += blocksize;
3389 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003390 if (buffer_freed(bh)) {
3391 BUFFER_TRACE(bh, "freed: skip");
3392 goto unlock;
3393 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003394 if (!buffer_mapped(bh)) {
3395 BUFFER_TRACE(bh, "unmapped");
3396 ext4_get_block(inode, iblock, bh, 0);
3397 /* unmapped? It's a hole - nothing to do */
3398 if (!buffer_mapped(bh)) {
3399 BUFFER_TRACE(bh, "still unmapped");
3400 goto unlock;
3401 }
3402 }
3403
3404 /* Ok, it's mapped. Make sure it's up-to-date */
3405 if (PageUptodate(page))
3406 set_buffer_uptodate(bh);
3407
3408 if (!buffer_uptodate(bh)) {
3409 err = -EIO;
3410 ll_rw_block(READ, 1, &bh);
3411 wait_on_buffer(bh);
3412 /* Uhhuh. Read error. Complain and punt. */
3413 if (!buffer_uptodate(bh))
3414 goto unlock;
3415 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003416 if (ext4_should_journal_data(inode)) {
3417 BUFFER_TRACE(bh, "get write access");
3418 err = ext4_journal_get_write_access(handle, bh);
3419 if (err)
3420 goto unlock;
3421 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003422 zero_user(page, offset, length);
Lukas Czernerd863dc32013-05-27 23:32:35 -04003423 BUFFER_TRACE(bh, "zeroed end of block");
3424
Lukas Czernerd863dc32013-05-27 23:32:35 -04003425 if (ext4_should_journal_data(inode)) {
3426 err = ext4_handle_dirty_metadata(handle, inode, bh);
Lukas Czerner0713ed02013-05-27 23:32:35 -04003427 } else {
jon ernst353eefd2013-07-01 08:12:39 -04003428 err = 0;
Lukas Czernerd863dc32013-05-27 23:32:35 -04003429 mark_buffer_dirty(bh);
Lukas Czerner0713ed02013-05-27 23:32:35 -04003430 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3431 err = ext4_jbd2_file_inode(handle, inode);
3432 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003433
3434unlock:
3435 unlock_page(page);
3436 page_cache_release(page);
3437 return err;
3438}
3439
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003440/*
3441 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3442 * up to the end of the block which corresponds to `from'.
3443 * This required during truncate. We need to physically zero the tail end
3444 * of that block so it doesn't yield old data if the file is later grown.
3445 */
Stephen Hemmingerc1978552014-05-12 10:50:23 -04003446static int ext4_block_truncate_page(handle_t *handle,
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003447 struct address_space *mapping, loff_t from)
3448{
3449 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3450 unsigned length;
3451 unsigned blocksize;
3452 struct inode *inode = mapping->host;
3453
3454 blocksize = inode->i_sb->s_blocksize;
3455 length = blocksize - (offset & (blocksize - 1));
3456
3457 return ext4_block_zero_page_range(handle, mapping, from, length);
3458}
3459
Lukas Czernera87dd182013-05-27 23:32:35 -04003460int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3461 loff_t lstart, loff_t length)
3462{
3463 struct super_block *sb = inode->i_sb;
3464 struct address_space *mapping = inode->i_mapping;
Lukas Czernere1be3a92013-07-01 08:12:39 -04003465 unsigned partial_start, partial_end;
Lukas Czernera87dd182013-05-27 23:32:35 -04003466 ext4_fsblk_t start, end;
3467 loff_t byte_end = (lstart + length - 1);
3468 int err = 0;
3469
Lukas Czernere1be3a92013-07-01 08:12:39 -04003470 partial_start = lstart & (sb->s_blocksize - 1);
3471 partial_end = byte_end & (sb->s_blocksize - 1);
3472
Lukas Czernera87dd182013-05-27 23:32:35 -04003473 start = lstart >> sb->s_blocksize_bits;
3474 end = byte_end >> sb->s_blocksize_bits;
3475
3476 /* Handle partial zero within the single block */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003477 if (start == end &&
3478 (partial_start || (partial_end != sb->s_blocksize - 1))) {
Lukas Czernera87dd182013-05-27 23:32:35 -04003479 err = ext4_block_zero_page_range(handle, mapping,
3480 lstart, length);
3481 return err;
3482 }
3483 /* Handle partial zero out on the start of the range */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003484 if (partial_start) {
Lukas Czernera87dd182013-05-27 23:32:35 -04003485 err = ext4_block_zero_page_range(handle, mapping,
3486 lstart, sb->s_blocksize);
3487 if (err)
3488 return err;
3489 }
3490 /* Handle partial zero out on the end of the range */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003491 if (partial_end != sb->s_blocksize - 1)
Lukas Czernera87dd182013-05-27 23:32:35 -04003492 err = ext4_block_zero_page_range(handle, mapping,
Lukas Czernere1be3a92013-07-01 08:12:39 -04003493 byte_end - partial_end,
3494 partial_end + 1);
Lukas Czernera87dd182013-05-27 23:32:35 -04003495 return err;
3496}
3497
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003498int ext4_can_truncate(struct inode *inode)
3499{
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003500 if (S_ISREG(inode->i_mode))
3501 return 1;
3502 if (S_ISDIR(inode->i_mode))
3503 return 1;
3504 if (S_ISLNK(inode->i_mode))
3505 return !ext4_inode_is_fast_symlink(inode);
3506 return 0;
3507}
3508
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003509/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003510 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3511 * associated with the given offset and length
3512 *
3513 * @inode: File inode
3514 * @offset: The offset where the hole will begin
3515 * @len: The length of the hole
3516 *
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003517 * Returns: 0 on success or negative on failure
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003518 */
3519
Ashish Sangwanaeb28172013-07-01 08:12:38 -04003520int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003521{
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003522 struct super_block *sb = inode->i_sb;
3523 ext4_lblk_t first_block, stop_block;
3524 struct address_space *mapping = inode->i_mapping;
Lukas Czernera87dd182013-05-27 23:32:35 -04003525 loff_t first_block_offset, last_block_offset;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003526 handle_t *handle;
3527 unsigned int credits;
3528 int ret = 0;
3529
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003530 if (!S_ISREG(inode->i_mode))
Allison Henderson73355192012-03-21 22:23:31 -04003531 return -EOPNOTSUPP;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003532
Lukas Czernerb8a86842014-03-18 18:05:35 -04003533 trace_ext4_punch_hole(inode, offset, length, 0);
Zheng Liuaaddea82013-01-16 20:21:26 -05003534
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003535 /*
3536 * Write out all dirty pages to avoid race conditions
3537 * Then release them.
3538 */
3539 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3540 ret = filemap_write_and_wait_range(mapping, offset,
3541 offset + length - 1);
3542 if (ret)
3543 return ret;
3544 }
3545
3546 mutex_lock(&inode->i_mutex);
Lukas Czerner9ef06ce2014-04-12 09:47:00 -04003547
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003548 /* No need to punch hole beyond i_size */
3549 if (offset >= inode->i_size)
3550 goto out_mutex;
3551
3552 /*
3553 * If the hole extends beyond i_size, set the hole
3554 * to end after the page that contains i_size
3555 */
3556 if (offset + length > inode->i_size) {
3557 length = inode->i_size +
3558 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3559 offset;
3560 }
3561
Jan Karaa3612932013-08-16 21:19:41 -04003562 if (offset & (sb->s_blocksize - 1) ||
3563 (offset + length) & (sb->s_blocksize - 1)) {
3564 /*
3565 * Attach jinode to inode for jbd2 if we do any zeroing of
3566 * partial block
3567 */
3568 ret = ext4_inode_attach_jinode(inode);
3569 if (ret < 0)
3570 goto out_mutex;
3571
3572 }
3573
Lukas Czernera87dd182013-05-27 23:32:35 -04003574 first_block_offset = round_up(offset, sb->s_blocksize);
3575 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003576
Lukas Czernera87dd182013-05-27 23:32:35 -04003577 /* Now release the pages and zero block aligned part of pages*/
3578 if (last_block_offset > first_block_offset)
3579 truncate_pagecache_range(inode, first_block_offset,
3580 last_block_offset);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003581
3582 /* Wait all existing dio workers, newcomers will block on i_mutex */
3583 ext4_inode_block_unlocked_dio(inode);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003584 inode_dio_wait(inode);
3585
3586 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3587 credits = ext4_writepage_trans_blocks(inode);
3588 else
3589 credits = ext4_blocks_for_truncate(inode);
3590 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3591 if (IS_ERR(handle)) {
3592 ret = PTR_ERR(handle);
3593 ext4_std_error(sb, ret);
3594 goto out_dio;
3595 }
3596
Lukas Czernera87dd182013-05-27 23:32:35 -04003597 ret = ext4_zero_partial_blocks(handle, inode, offset,
3598 length);
3599 if (ret)
3600 goto out_stop;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003601
3602 first_block = (offset + sb->s_blocksize - 1) >>
3603 EXT4_BLOCK_SIZE_BITS(sb);
3604 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3605
3606 /* If there are no blocks to remove, return now */
3607 if (first_block >= stop_block)
3608 goto out_stop;
3609
3610 down_write(&EXT4_I(inode)->i_data_sem);
3611 ext4_discard_preallocations(inode);
3612
3613 ret = ext4_es_remove_extent(inode, first_block,
3614 stop_block - first_block);
3615 if (ret) {
3616 up_write(&EXT4_I(inode)->i_data_sem);
3617 goto out_stop;
3618 }
3619
3620 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3621 ret = ext4_ext_remove_space(inode, first_block,
3622 stop_block - 1);
3623 else
3624 ret = ext4_free_hole_blocks(handle, inode, first_block,
3625 stop_block);
3626
Theodore Ts'o819c4922013-04-03 12:47:17 -04003627 up_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003628 if (IS_SYNC(inode))
3629 ext4_handle_sync(handle);
Maxim Patlasove251f9b2014-02-20 16:58:05 -05003630
3631 /* Now release the pages again to reduce race window */
3632 if (last_block_offset > first_block_offset)
3633 truncate_pagecache_range(inode, first_block_offset,
3634 last_block_offset);
3635
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003636 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3637 ext4_mark_inode_dirty(handle, inode);
3638out_stop:
3639 ext4_journal_stop(handle);
3640out_dio:
3641 ext4_inode_resume_unlocked_dio(inode);
3642out_mutex:
3643 mutex_unlock(&inode->i_mutex);
3644 return ret;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003645}
3646
Jan Karaa3612932013-08-16 21:19:41 -04003647int ext4_inode_attach_jinode(struct inode *inode)
3648{
3649 struct ext4_inode_info *ei = EXT4_I(inode);
3650 struct jbd2_inode *jinode;
3651
3652 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3653 return 0;
3654
3655 jinode = jbd2_alloc_inode(GFP_KERNEL);
3656 spin_lock(&inode->i_lock);
3657 if (!ei->jinode) {
3658 if (!jinode) {
3659 spin_unlock(&inode->i_lock);
3660 return -ENOMEM;
3661 }
3662 ei->jinode = jinode;
3663 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3664 jinode = NULL;
3665 }
3666 spin_unlock(&inode->i_lock);
3667 if (unlikely(jinode != NULL))
3668 jbd2_free_inode(jinode);
3669 return 0;
3670}
3671
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003672/*
Mingming Cao617ba132006-10-11 01:20:53 -07003673 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003674 *
Mingming Cao617ba132006-10-11 01:20:53 -07003675 * We block out ext4_get_block() block instantiations across the entire
3676 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003677 * simultaneously on behalf of the same inode.
3678 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003679 * As we work through the truncate and commit bits of it to the journal there
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003680 * is one core, guiding principle: the file's tree must always be consistent on
3681 * disk. We must be able to restart the truncate after a crash.
3682 *
3683 * The file's tree may be transiently inconsistent in memory (although it
3684 * probably isn't), but whenever we close off and commit a journal transaction,
3685 * the contents of (the filesystem + the journal) must be consistent and
3686 * restartable. It's pretty simple, really: bottom up, right to left (although
3687 * left-to-right works OK too).
3688 *
3689 * Note that at recovery time, journal replay occurs *before* the restart of
3690 * truncate against the orphan inode list.
3691 *
3692 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07003693 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003694 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07003695 * ext4_truncate() to have another go. So there will be instantiated blocks
3696 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003697 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07003698 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003699 */
Mingming Cao617ba132006-10-11 01:20:53 -07003700void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003701{
Theodore Ts'o819c4922013-04-03 12:47:17 -04003702 struct ext4_inode_info *ei = EXT4_I(inode);
3703 unsigned int credits;
3704 handle_t *handle;
3705 struct address_space *mapping = inode->i_mapping;
Theodore Ts'o819c4922013-04-03 12:47:17 -04003706
Theodore Ts'o19b5ef62013-04-03 21:58:52 -04003707 /*
3708 * There is a possibility that we're either freeing the inode
Matthew Wilcoxe04027e2014-03-24 15:15:07 -04003709 * or it's a completely new inode. In those cases we might not
Theodore Ts'o19b5ef62013-04-03 21:58:52 -04003710 * have i_mutex locked because it's not necessary.
3711 */
3712 if (!(inode->i_state & (I_NEW|I_FREEING)))
3713 WARN_ON(!mutex_is_locked(&inode->i_mutex));
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003714 trace_ext4_truncate_enter(inode);
3715
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003716 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003717 return;
3718
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003719 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003720
Theodore Ts'o5534fb52009-09-17 09:34:16 -04003721 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003722 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05003723
Tao Maaef1c852012-12-10 14:06:02 -05003724 if (ext4_has_inline_data(inode)) {
3725 int has_inline = 1;
3726
3727 ext4_inline_data_truncate(inode, &has_inline);
3728 if (has_inline)
3729 return;
3730 }
3731
Jan Karaa3612932013-08-16 21:19:41 -04003732 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3733 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3734 if (ext4_inode_attach_jinode(inode) < 0)
3735 return;
3736 }
3737
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003738 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Theodore Ts'o819c4922013-04-03 12:47:17 -04003739 credits = ext4_writepage_trans_blocks(inode);
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003740 else
Theodore Ts'o819c4922013-04-03 12:47:17 -04003741 credits = ext4_blocks_for_truncate(inode);
3742
3743 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3744 if (IS_ERR(handle)) {
3745 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3746 return;
3747 }
3748
Lukas Czernereb3544c2013-05-27 23:32:35 -04003749 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3750 ext4_block_truncate_page(handle, mapping, inode->i_size);
Theodore Ts'o819c4922013-04-03 12:47:17 -04003751
3752 /*
3753 * We add the inode to the orphan list, so that if this
3754 * truncate spans multiple transactions, and we crash, we will
3755 * resume the truncate when the filesystem recovers. It also
3756 * marks the inode dirty, to catch the new size.
3757 *
3758 * Implication: the file must always be in a sane, consistent
3759 * truncatable state while each transaction commits.
3760 */
3761 if (ext4_orphan_add(handle, inode))
3762 goto out_stop;
3763
3764 down_write(&EXT4_I(inode)->i_data_sem);
3765
3766 ext4_discard_preallocations(inode);
3767
3768 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3769 ext4_ext_truncate(handle, inode);
3770 else
3771 ext4_ind_truncate(handle, inode);
3772
3773 up_write(&ei->i_data_sem);
3774
3775 if (IS_SYNC(inode))
3776 ext4_handle_sync(handle);
3777
3778out_stop:
3779 /*
3780 * If this was a simple ftruncate() and the file will remain alive,
3781 * then we need to clear up the orphan record which we created above.
3782 * However, if this was a real unlink then we were called by
3783 * ext4_delete_inode(), and we allow that function to clean up the
3784 * orphan info for us.
3785 */
3786 if (inode->i_nlink)
3787 ext4_orphan_del(handle, inode);
3788
3789 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3790 ext4_mark_inode_dirty(handle, inode);
3791 ext4_journal_stop(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07003792
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003793 trace_ext4_truncate_exit(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003794}
3795
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003796/*
Mingming Cao617ba132006-10-11 01:20:53 -07003797 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003798 * underlying buffer_head on success. If 'in_mem' is true, we have all
3799 * data in memory that is needed to recreate the on-disk version of this
3800 * inode.
3801 */
Mingming Cao617ba132006-10-11 01:20:53 -07003802static int __ext4_get_inode_loc(struct inode *inode,
3803 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003804{
Theodore Ts'o240799c2008-10-09 23:53:47 -04003805 struct ext4_group_desc *gdp;
3806 struct buffer_head *bh;
3807 struct super_block *sb = inode->i_sb;
3808 ext4_fsblk_t block;
3809 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003810
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003811 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003812 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003813 return -EIO;
3814
Theodore Ts'o240799c2008-10-09 23:53:47 -04003815 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3816 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3817 if (!gdp)
3818 return -EIO;
3819
3820 /*
3821 * Figure out the offset within the block group inode table
3822 */
Tao Ma00d09882011-05-09 10:26:41 -04003823 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003824 inode_offset = ((inode->i_ino - 1) %
3825 EXT4_INODES_PER_GROUP(sb));
3826 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3827 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3828
3829 bh = sb_getblk(sb, block);
Wang Shilongaebf0242013-01-12 16:28:47 -05003830 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05003831 return -ENOMEM;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003832 if (!buffer_uptodate(bh)) {
3833 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04003834
3835 /*
3836 * If the buffer has the write error flag, we have failed
3837 * to write out another inode in the same block. In this
3838 * case, we don't have to read the block because we may
3839 * read the old inode data successfully.
3840 */
3841 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3842 set_buffer_uptodate(bh);
3843
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003844 if (buffer_uptodate(bh)) {
3845 /* someone brought it uptodate while we waited */
3846 unlock_buffer(bh);
3847 goto has_buffer;
3848 }
3849
3850 /*
3851 * If we have all information of the inode in memory and this
3852 * is the only valid inode in the block, we need not read the
3853 * block.
3854 */
3855 if (in_mem) {
3856 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003857 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003858
Theodore Ts'o240799c2008-10-09 23:53:47 -04003859 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003860
3861 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003862 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Wang Shilongaebf0242013-01-12 16:28:47 -05003863 if (unlikely(!bitmap_bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003864 goto make_io;
3865
3866 /*
3867 * If the inode bitmap isn't in cache then the
3868 * optimisation may end up performing two reads instead
3869 * of one, so skip it.
3870 */
3871 if (!buffer_uptodate(bitmap_bh)) {
3872 brelse(bitmap_bh);
3873 goto make_io;
3874 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04003875 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003876 if (i == inode_offset)
3877 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07003878 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003879 break;
3880 }
3881 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003882 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003883 /* all other inodes are free, so skip I/O */
3884 memset(bh->b_data, 0, bh->b_size);
3885 set_buffer_uptodate(bh);
3886 unlock_buffer(bh);
3887 goto has_buffer;
3888 }
3889 }
3890
3891make_io:
3892 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04003893 * If we need to do any I/O, try to pre-readahead extra
3894 * blocks from the inode table.
3895 */
3896 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3897 ext4_fsblk_t b, end, table;
3898 unsigned num;
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003899 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003900
3901 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04003902 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003903 b = block & ~((ext4_fsblk_t) ra_blks - 1);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003904 if (table > b)
3905 b = table;
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003906 end = b + ra_blks;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003907 num = EXT4_INODES_PER_GROUP(sb);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04003908 if (ext4_has_group_desc_csum(sb))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05003909 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003910 table += num / inodes_per_block;
3911 if (end > table)
3912 end = table;
3913 while (b <= end)
3914 sb_breadahead(sb, b++);
3915 }
3916
3917 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003918 * There are other valid inodes in the buffer, this inode
3919 * has in-inode xattrs, or we don't have this inode in memory.
3920 * Read the block from disk.
3921 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003922 trace_ext4_load_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003923 get_bh(bh);
3924 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02003925 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003926 wait_on_buffer(bh);
3927 if (!buffer_uptodate(bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003928 EXT4_ERROR_INODE_BLOCK(inode, block,
3929 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003930 brelse(bh);
3931 return -EIO;
3932 }
3933 }
3934has_buffer:
3935 iloc->bh = bh;
3936 return 0;
3937}
3938
Mingming Cao617ba132006-10-11 01:20:53 -07003939int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003940{
3941 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07003942 return __ext4_get_inode_loc(inode, iloc,
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003943 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003944}
3945
Mingming Cao617ba132006-10-11 01:20:53 -07003946void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003947{
Mingming Cao617ba132006-10-11 01:20:53 -07003948 unsigned int flags = EXT4_I(inode)->i_flags;
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003949 unsigned int new_fl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003950
Mingming Cao617ba132006-10-11 01:20:53 -07003951 if (flags & EXT4_SYNC_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003952 new_fl |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07003953 if (flags & EXT4_APPEND_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003954 new_fl |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07003955 if (flags & EXT4_IMMUTABLE_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003956 new_fl |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07003957 if (flags & EXT4_NOATIME_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003958 new_fl |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07003959 if (flags & EXT4_DIRSYNC_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003960 new_fl |= S_DIRSYNC;
Theodore Ts'o5f16f322014-03-24 14:43:12 -04003961 inode_set_flags(inode, new_fl,
3962 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003963}
3964
Jan Karaff9ddf72007-07-18 09:24:20 -04003965/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3966void ext4_get_inode_flags(struct ext4_inode_info *ei)
3967{
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003968 unsigned int vfs_fl;
3969 unsigned long old_fl, new_fl;
Jan Karaff9ddf72007-07-18 09:24:20 -04003970
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003971 do {
3972 vfs_fl = ei->vfs_inode.i_flags;
3973 old_fl = ei->i_flags;
3974 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3975 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3976 EXT4_DIRSYNC_FL);
3977 if (vfs_fl & S_SYNC)
3978 new_fl |= EXT4_SYNC_FL;
3979 if (vfs_fl & S_APPEND)
3980 new_fl |= EXT4_APPEND_FL;
3981 if (vfs_fl & S_IMMUTABLE)
3982 new_fl |= EXT4_IMMUTABLE_FL;
3983 if (vfs_fl & S_NOATIME)
3984 new_fl |= EXT4_NOATIME_FL;
3985 if (vfs_fl & S_DIRSYNC)
3986 new_fl |= EXT4_DIRSYNC_FL;
3987 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
Jan Karaff9ddf72007-07-18 09:24:20 -04003988}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003989
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003990static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003991 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003992{
3993 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003994 struct inode *inode = &(ei->vfs_inode);
3995 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003996
3997 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3998 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3999 /* we are using combined 48 bit field */
4000 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4001 le32_to_cpu(raw_inode->i_blocks_lo);
Theodore Ts'o07a03822010-06-14 09:54:48 -04004002 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004003 /* i_blocks represent file system block size */
4004 return i_blocks << (inode->i_blkbits - 9);
4005 } else {
4006 return i_blocks;
4007 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004008 } else {
4009 return le32_to_cpu(raw_inode->i_blocks_lo);
4010 }
4011}
Jan Karaff9ddf72007-07-18 09:24:20 -04004012
Tao Ma152a7b02012-12-02 11:13:24 -05004013static inline void ext4_iget_extra_inode(struct inode *inode,
4014 struct ext4_inode *raw_inode,
4015 struct ext4_inode_info *ei)
4016{
4017 __le32 *magic = (void *)raw_inode +
4018 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
Tao Ma67cf5b02012-12-10 14:04:46 -05004019 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
Tao Ma152a7b02012-12-02 11:13:24 -05004020 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
Tao Ma67cf5b02012-12-10 14:04:46 -05004021 ext4_find_inline_data_nolock(inode);
Tao Maf19d5872012-12-10 14:05:51 -05004022 } else
4023 EXT4_I(inode)->i_inline_off = 0;
Tao Ma152a7b02012-12-02 11:13:24 -05004024}
4025
David Howells1d1fe1e2008-02-07 00:15:37 -08004026struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004027{
Mingming Cao617ba132006-10-11 01:20:53 -07004028 struct ext4_iloc iloc;
4029 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004030 struct ext4_inode_info *ei;
David Howells1d1fe1e2008-02-07 00:15:37 -08004031 struct inode *inode;
Jan Karab436b9b2009-12-08 23:51:10 -05004032 journal_t *journal = EXT4_SB(sb)->s_journal;
David Howells1d1fe1e2008-02-07 00:15:37 -08004033 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004034 int block;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004035 uid_t i_uid;
4036 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004037
David Howells1d1fe1e2008-02-07 00:15:37 -08004038 inode = iget_locked(sb, ino);
4039 if (!inode)
4040 return ERR_PTR(-ENOMEM);
4041 if (!(inode->i_state & I_NEW))
4042 return inode;
4043
4044 ei = EXT4_I(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004045 iloc.bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004046
David Howells1d1fe1e2008-02-07 00:15:37 -08004047 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4048 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004049 goto bad_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07004050 raw_inode = ext4_raw_inode(&iloc);
Darrick J. Wong814525f2012-04-29 18:31:10 -04004051
4052 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4053 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4054 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4055 EXT4_INODE_SIZE(inode->i_sb)) {
4056 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4057 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4058 EXT4_INODE_SIZE(inode->i_sb));
4059 ret = -EIO;
4060 goto bad_inode;
4061 }
4062 } else
4063 ei->i_extra_isize = 0;
4064
4065 /* Precompute checksum seed for inode metadata */
4066 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4067 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
4068 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4069 __u32 csum;
4070 __le32 inum = cpu_to_le32(inode->i_ino);
4071 __le32 gen = raw_inode->i_generation;
4072 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4073 sizeof(inum));
4074 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4075 sizeof(gen));
4076 }
4077
4078 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4079 EXT4_ERROR_INODE(inode, "checksum invalid");
4080 ret = -EIO;
4081 goto bad_inode;
4082 }
4083
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004084 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004085 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4086 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004087 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004088 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4089 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004090 }
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004091 i_uid_write(inode, i_uid);
4092 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02004093 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004094
Theodore Ts'o353eb832011-01-10 12:18:25 -05004095 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Tao Ma67cf5b02012-12-10 14:04:46 -05004096 ei->i_inline_off = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004097 ei->i_dir_start_lookup = 0;
4098 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4099 /* We now have enough fields to check if the inode was active or not.
4100 * This is needed because nfsd might try to access dead inodes
4101 * the test is that same one that e2fsck uses
4102 * NeilBrown 1999oct15
4103 */
4104 if (inode->i_nlink == 0) {
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004105 if ((inode->i_mode == 0 ||
4106 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4107 ino != EXT4_BOOT_LOADER_INO) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004108 /* this inode is deleted */
David Howells1d1fe1e2008-02-07 00:15:37 -08004109 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004110 goto bad_inode;
4111 }
4112 /* The only unlinked inodes we let through here have
4113 * valid i_mode and are being read by the orphan
4114 * recovery code: that's fine, we're about to complete
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004115 * the process of deleting those.
4116 * OR it is the EXT4_BOOT_LOADER_INO which is
4117 * not initialized on a new filesystem. */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004118 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004119 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004120 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004121 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04004122 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004123 ei->i_file_acl |=
4124 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004125 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004126 ei->i_disksize = inode->i_size;
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03004127#ifdef CONFIG_QUOTA
4128 ei->i_reserved_quota = 0;
4129#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004130 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4131 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04004132 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004133 /*
4134 * NOTE! The in-memory inode i_data array is in little-endian order
4135 * even on big-endian machines: we do NOT byteswap the block numbers!
4136 */
Mingming Cao617ba132006-10-11 01:20:53 -07004137 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004138 ei->i_data[block] = raw_inode->i_block[block];
4139 INIT_LIST_HEAD(&ei->i_orphan);
4140
Jan Karab436b9b2009-12-08 23:51:10 -05004141 /*
4142 * Set transaction id's of transactions that have to be committed
4143 * to finish f[data]sync. We set them to currently running transaction
4144 * as we cannot be sure that the inode or some of its metadata isn't
4145 * part of the transaction - the inode could have been reclaimed and
4146 * now it is reread from disk.
4147 */
4148 if (journal) {
4149 transaction_t *transaction;
4150 tid_t tid;
4151
Theodore Ts'oa931da62010-08-03 21:35:12 -04004152 read_lock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004153 if (journal->j_running_transaction)
4154 transaction = journal->j_running_transaction;
4155 else
4156 transaction = journal->j_committing_transaction;
4157 if (transaction)
4158 tid = transaction->t_tid;
4159 else
4160 tid = journal->j_commit_sequence;
Theodore Ts'oa931da62010-08-03 21:35:12 -04004161 read_unlock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004162 ei->i_sync_tid = tid;
4163 ei->i_datasync_tid = tid;
4164 }
4165
Eric Sandeen0040d982008-02-05 22:36:43 -05004166 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004167 if (ei->i_extra_isize == 0) {
4168 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07004169 ei->i_extra_isize = sizeof(struct ext4_inode) -
4170 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004171 } else {
Tao Ma152a7b02012-12-02 11:13:24 -05004172 ext4_iget_extra_inode(inode, raw_inode, ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004173 }
Darrick J. Wong814525f2012-04-29 18:31:10 -04004174 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004175
Kalpak Shahef7f3832007-07-18 09:15:20 -04004176 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4177 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4178 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4179 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4180
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004181 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
Theodore Ts'oc4f65702014-03-20 00:32:57 -04004182 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4183 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4184 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4185 inode->i_version |=
4186 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4187 }
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004188 }
4189
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004190 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004191 if (ei->i_file_acl &&
Theodore Ts'o10329882009-11-15 15:29:56 -05004192 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04004193 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4194 ei->i_file_acl);
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004195 ret = -EIO;
4196 goto bad_inode;
Tao Maf19d5872012-12-10 14:05:51 -05004197 } else if (!ext4_has_inline_data(inode)) {
4198 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4199 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4200 (S_ISLNK(inode->i_mode) &&
4201 !ext4_inode_is_fast_symlink(inode))))
4202 /* Validate extent which is part of inode */
4203 ret = ext4_ext_check_inode(inode);
4204 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4205 (S_ISLNK(inode->i_mode) &&
4206 !ext4_inode_is_fast_symlink(inode))) {
4207 /* Validate block references which are part of inode */
4208 ret = ext4_ind_check_inode(inode);
4209 }
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004210 }
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004211 if (ret)
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004212 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04004213
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004214 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004215 inode->i_op = &ext4_file_inode_operations;
4216 inode->i_fop = &ext4_file_operations;
4217 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004218 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004219 inode->i_op = &ext4_dir_inode_operations;
4220 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004221 } else if (S_ISLNK(inode->i_mode)) {
Duane Griffine83c1392008-12-19 20:47:15 +00004222 if (ext4_inode_is_fast_symlink(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004223 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00004224 nd_terminate_link(ei->i_data, inode->i_size,
4225 sizeof(ei->i_data) - 1);
4226 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004227 inode->i_op = &ext4_symlink_inode_operations;
4228 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004229 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004230 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4231 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004232 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004233 if (raw_inode->i_block[0])
4234 init_special_inode(inode, inode->i_mode,
4235 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4236 else
4237 init_special_inode(inode, inode->i_mode,
4238 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004239 } else if (ino == EXT4_BOOT_LOADER_INO) {
4240 make_bad_inode(inode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004241 } else {
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004242 ret = -EIO;
Theodore Ts'o24676da2010-05-16 21:00:00 -04004243 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004244 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004245 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004246 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004247 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004248 unlock_new_inode(inode);
4249 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004250
4251bad_inode:
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004252 brelse(iloc.bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004253 iget_failed(inode);
4254 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004255}
4256
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004257static int ext4_inode_blocks_set(handle_t *handle,
4258 struct ext4_inode *raw_inode,
4259 struct ext4_inode_info *ei)
4260{
4261 struct inode *inode = &(ei->vfs_inode);
4262 u64 i_blocks = inode->i_blocks;
4263 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004264
4265 if (i_blocks <= ~0U) {
4266 /*
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004267 * i_blocks can be represented in a 32 bit variable
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004268 * as multiple of 512 bytes
4269 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004270 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004271 raw_inode->i_blocks_high = 0;
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004272 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004273 return 0;
4274 }
4275 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4276 return -EFBIG;
4277
4278 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004279 /*
4280 * i_blocks can be represented in a 48 bit variable
4281 * as multiple of 512 bytes
4282 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004283 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004284 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004285 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004286 } else {
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004287 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004288 /* i_block is stored in file system block size */
4289 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4290 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4291 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004292 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004293 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004294}
4295
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004296/*
4297 * Post the struct inode info into an on-disk inode location in the
4298 * buffer-cache. This gobbles the caller's reference to the
4299 * buffer_head in the inode location struct.
4300 *
4301 * The caller must have write access to iloc->bh.
4302 */
Mingming Cao617ba132006-10-11 01:20:53 -07004303static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004304 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04004305 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004306{
Mingming Cao617ba132006-10-11 01:20:53 -07004307 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4308 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004309 struct buffer_head *bh = iloc->bh;
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004310 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004311 int err = 0, rc, block;
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004312 int need_datasync = 0, set_large_file = 0;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004313 uid_t i_uid;
4314 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004315
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004316 spin_lock(&ei->i_raw_lock);
4317
4318 /* For fields not tracked in the in-memory inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004319 * initialise them to zero for new inodes. */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004320 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
Mingming Cao617ba132006-10-11 01:20:53 -07004321 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004322
Jan Karaff9ddf72007-07-18 09:24:20 -04004323 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004324 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004325 i_uid = i_uid_read(inode);
4326 i_gid = i_gid_read(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004327 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004328 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4329 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004330/*
4331 * Fix up interoperability with old kernels. Otherwise, old inodes get
4332 * re-used with the upper 16 bits of the uid/gid intact
4333 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004334 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004335 raw_inode->i_uid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004336 cpu_to_le16(high_16_bits(i_uid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004337 raw_inode->i_gid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004338 cpu_to_le16(high_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004339 } else {
4340 raw_inode->i_uid_high = 0;
4341 raw_inode->i_gid_high = 0;
4342 }
4343 } else {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004344 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4345 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004346 raw_inode->i_uid_high = 0;
4347 raw_inode->i_gid_high = 0;
4348 }
4349 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004350
4351 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4352 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4353 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4354 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4355
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004356 if (ext4_inode_blocks_set(handle, raw_inode, ei)) {
4357 spin_unlock(&ei->i_raw_lock);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004358 goto out_brelse;
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004359 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004360 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o353eb832011-01-10 12:18:25 -05004361 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004362 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004363 raw_inode->i_file_acl_high =
4364 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004365 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Jan Karab71fc072012-09-26 21:52:20 -04004366 if (ei->i_disksize != ext4_isize(raw_inode)) {
4367 ext4_isize_set(raw_inode, ei->i_disksize);
4368 need_datasync = 1;
4369 }
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004370 if (ei->i_disksize > 0x7fffffffULL) {
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004371 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4372 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4373 EXT4_SB(sb)->s_es->s_rev_level ==
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004374 cpu_to_le32(EXT4_GOOD_OLD_REV))
4375 set_large_file = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004376 }
4377 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4378 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4379 if (old_valid_dev(inode->i_rdev)) {
4380 raw_inode->i_block[0] =
4381 cpu_to_le32(old_encode_dev(inode->i_rdev));
4382 raw_inode->i_block[1] = 0;
4383 } else {
4384 raw_inode->i_block[0] = 0;
4385 raw_inode->i_block[1] =
4386 cpu_to_le32(new_encode_dev(inode->i_rdev));
4387 raw_inode->i_block[2] = 0;
4388 }
Tao Maf19d5872012-12-10 14:05:51 -05004389 } else if (!ext4_has_inline_data(inode)) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004390 for (block = 0; block < EXT4_N_BLOCKS; block++)
4391 raw_inode->i_block[block] = ei->i_data[block];
Tao Maf19d5872012-12-10 14:05:51 -05004392 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004393
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004394 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
Theodore Ts'oc4f65702014-03-20 00:32:57 -04004395 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4396 if (ei->i_extra_isize) {
4397 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4398 raw_inode->i_version_hi =
4399 cpu_to_le32(inode->i_version >> 32);
4400 raw_inode->i_extra_isize =
4401 cpu_to_le16(ei->i_extra_isize);
4402 }
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004403 }
4404
Darrick J. Wong814525f2012-04-29 18:31:10 -04004405 ext4_inode_csum_set(inode, raw_inode, ei);
4406
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004407 spin_unlock(&ei->i_raw_lock);
4408
Frank Mayhar830156c2009-09-29 10:07:47 -04004409 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05004410 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
Frank Mayhar830156c2009-09-29 10:07:47 -04004411 if (!err)
4412 err = rc;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004413 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004414 if (set_large_file) {
4415 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4416 if (err)
4417 goto out_brelse;
4418 ext4_update_dynamic_rev(sb);
4419 EXT4_SET_RO_COMPAT_FEATURE(sb,
4420 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4421 ext4_handle_sync(handle);
4422 err = ext4_handle_dirty_super(handle, sb);
4423 }
Jan Karab71fc072012-09-26 21:52:20 -04004424 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004425out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004426 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004427 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004428 return err;
4429}
4430
4431/*
Mingming Cao617ba132006-10-11 01:20:53 -07004432 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004433 *
4434 * We are called from a few places:
4435 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004436 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004437 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004438 * transaction to commit.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004439 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004440 * - Within flush work (sys_sync(), kupdate and such).
4441 * We wait on commit, if told to.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004442 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004443 * - Within iput_final() -> write_inode_now()
4444 * We wait on commit, if told to.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004445 *
4446 * In all cases it is actually safe for us to return without doing anything,
4447 * because the inode has been copied into a raw inode buffer in
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004448 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4449 * writeback.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004450 *
4451 * Note that we are absolutely dependent upon all inode dirtiers doing the
4452 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4453 * which we are interested.
4454 *
4455 * It would be a bug for them to not do this. The code:
4456 *
4457 * mark_inode_dirty(inode)
4458 * stuff();
4459 * inode->i_size = expr;
4460 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004461 * is in error because write_inode() could occur while `stuff()' is running,
4462 * and the new i_size will be lost. Plus the inode will no longer be on the
4463 * superblock's dirty inode list.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004464 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004465int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004466{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004467 int err;
4468
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004469 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004470 return 0;
4471
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004472 if (EXT4_SB(inode->i_sb)->s_journal) {
4473 if (ext4_journal_current_handle()) {
4474 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4475 dump_stack();
4476 return -EIO;
4477 }
4478
Jan Kara10542c22014-03-04 10:50:50 -05004479 /*
4480 * No need to force transaction in WB_SYNC_NONE mode. Also
4481 * ext4_sync_fs() will force the commit after everything is
4482 * written.
4483 */
4484 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004485 return 0;
4486
4487 err = ext4_force_commit(inode->i_sb);
4488 } else {
4489 struct ext4_iloc iloc;
4490
Curt Wohlgemuth8b472d72010-04-03 16:45:06 -04004491 err = __ext4_get_inode_loc(inode, &iloc, 0);
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004492 if (err)
4493 return err;
Jan Kara10542c22014-03-04 10:50:50 -05004494 /*
4495 * sync(2) will flush the whole buffer cache. No need to do
4496 * it here separately for each inode.
4497 */
4498 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
Frank Mayhar830156c2009-09-29 10:07:47 -04004499 sync_dirty_buffer(iloc.bh);
4500 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004501 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4502 "IO error syncing inode");
Frank Mayhar830156c2009-09-29 10:07:47 -04004503 err = -EIO;
4504 }
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004505 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004506 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004507 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004508}
4509
4510/*
Jan Kara53e87262012-12-25 13:29:52 -05004511 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4512 * buffers that are attached to a page stradding i_size and are undergoing
4513 * commit. In that case we have to wait for commit to finish and try again.
4514 */
4515static void ext4_wait_for_tail_page_commit(struct inode *inode)
4516{
4517 struct page *page;
4518 unsigned offset;
4519 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4520 tid_t commit_tid = 0;
4521 int ret;
4522
4523 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4524 /*
4525 * All buffers in the last page remain valid? Then there's nothing to
4526 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4527 * blocksize case
4528 */
4529 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4530 return;
4531 while (1) {
4532 page = find_lock_page(inode->i_mapping,
4533 inode->i_size >> PAGE_CACHE_SHIFT);
4534 if (!page)
4535 return;
Lukas Czernerca99fdd2013-05-21 23:25:01 -04004536 ret = __ext4_journalled_invalidatepage(page, offset,
4537 PAGE_CACHE_SIZE - offset);
Jan Kara53e87262012-12-25 13:29:52 -05004538 unlock_page(page);
4539 page_cache_release(page);
4540 if (ret != -EBUSY)
4541 return;
4542 commit_tid = 0;
4543 read_lock(&journal->j_state_lock);
4544 if (journal->j_committing_transaction)
4545 commit_tid = journal->j_committing_transaction->t_tid;
4546 read_unlock(&journal->j_state_lock);
4547 if (commit_tid)
4548 jbd2_log_wait_commit(journal, commit_tid);
4549 }
4550}
4551
4552/*
Mingming Cao617ba132006-10-11 01:20:53 -07004553 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004554 *
4555 * Called from notify_change.
4556 *
4557 * We want to trap VFS attempts to truncate the file as soon as
4558 * possible. In particular, we want to make sure that when the VFS
4559 * shrinks i_size, we put the inode on the orphan list and modify
4560 * i_disksize immediately, so that during the subsequent flushing of
4561 * dirty pages and freeing of disk blocks, we can guarantee that any
4562 * commit will leave the blocks being flushed in an unused state on
4563 * disk. (On recovery, the inode will get truncated and the blocks will
4564 * be freed, so we have a strong guarantee that no future commit will
4565 * leave these blocks visible to the user.)
4566 *
Jan Kara678aaf42008-07-11 19:27:31 -04004567 * Another thing we have to assure is that if we are in ordered mode
4568 * and inode is still attached to the committing transaction, we must
4569 * we start writeout of all the dirty pages which are being truncated.
4570 * This way we are sure that all the data written in the previous
4571 * transaction are already on disk (truncate waits for pages under
4572 * writeback).
4573 *
4574 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004575 */
Mingming Cao617ba132006-10-11 01:20:53 -07004576int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004577{
4578 struct inode *inode = dentry->d_inode;
4579 int error, rc = 0;
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004580 int orphan = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004581 const unsigned int ia_valid = attr->ia_valid;
4582
4583 error = inode_change_ok(inode, attr);
4584 if (error)
4585 return error;
4586
Dmitry Monakhov12755622010-04-08 22:04:20 +04004587 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05004588 dquot_initialize(inode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004589 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4590 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004591 handle_t *handle;
4592
4593 /* (user+group)*(old+new) structure, inode write (sb,
4594 * inode block, ? - but truncate inode update has it) */
Theodore Ts'o9924a922013-02-08 21:59:22 -05004595 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4596 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4597 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004598 if (IS_ERR(handle)) {
4599 error = PTR_ERR(handle);
4600 goto err_out;
4601 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05004602 error = dquot_transfer(inode, attr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004603 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07004604 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004605 return error;
4606 }
4607 /* Update corresponding info in inode so that everything is in
4608 * one transaction */
4609 if (attr->ia_valid & ATTR_UID)
4610 inode->i_uid = attr->ia_uid;
4611 if (attr->ia_valid & ATTR_GID)
4612 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07004613 error = ext4_mark_inode_dirty(handle, inode);
4614 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004615 }
4616
Jan Kara52083862013-08-17 10:07:17 -04004617 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4618 handle_t *handle;
Christoph Hellwig562c72a2011-06-24 14:29:45 -04004619
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004620 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -05004621 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4622
Theodore Ts'o0c095c72010-07-27 11:56:06 -04004623 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4624 return -EFBIG;
Eric Sandeene2b46572008-01-28 23:58:27 -05004625 }
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08004626
4627 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4628 inode_inc_iversion(inode);
4629
Jan Kara52083862013-08-17 10:07:17 -04004630 if (S_ISREG(inode->i_mode) &&
4631 (attr->ia_size < inode->i_size)) {
4632 if (ext4_should_order_data(inode)) {
4633 error = ext4_begin_ordered_truncate(inode,
Jan Kara678aaf42008-07-11 19:27:31 -04004634 attr->ia_size);
Jan Kara52083862013-08-17 10:07:17 -04004635 if (error)
Jan Kara678aaf42008-07-11 19:27:31 -04004636 goto err_out;
Jan Kara52083862013-08-17 10:07:17 -04004637 }
4638 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4639 if (IS_ERR(handle)) {
4640 error = PTR_ERR(handle);
4641 goto err_out;
4642 }
4643 if (ext4_handle_valid(handle)) {
4644 error = ext4_orphan_add(handle, inode);
4645 orphan = 1;
4646 }
Jan Kara90e775b2013-08-17 10:09:31 -04004647 down_write(&EXT4_I(inode)->i_data_sem);
Jan Kara52083862013-08-17 10:07:17 -04004648 EXT4_I(inode)->i_disksize = attr->ia_size;
4649 rc = ext4_mark_inode_dirty(handle, inode);
4650 if (!error)
4651 error = rc;
Jan Kara90e775b2013-08-17 10:09:31 -04004652 /*
4653 * We have to update i_size under i_data_sem together
4654 * with i_disksize to avoid races with writeback code
4655 * running ext4_wb_update_i_disksize().
4656 */
4657 if (!error)
4658 i_size_write(inode, attr->ia_size);
4659 up_write(&EXT4_I(inode)->i_data_sem);
Jan Kara52083862013-08-17 10:07:17 -04004660 ext4_journal_stop(handle);
4661 if (error) {
4662 ext4_orphan_del(NULL, inode);
Jan Kara678aaf42008-07-11 19:27:31 -04004663 goto err_out;
4664 }
Jan Kara90e775b2013-08-17 10:09:31 -04004665 } else
4666 i_size_write(inode, attr->ia_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004667
Jan Kara52083862013-08-17 10:07:17 -04004668 /*
4669 * Blocks are going to be removed from the inode. Wait
4670 * for dio in flight. Temporarily disable
4671 * dioread_nolock to prevent livelock.
4672 */
4673 if (orphan) {
4674 if (!ext4_should_journal_data(inode)) {
4675 ext4_inode_block_unlocked_dio(inode);
4676 inode_dio_wait(inode);
4677 ext4_inode_resume_unlocked_dio(inode);
4678 } else
4679 ext4_wait_for_tail_page_commit(inode);
Dmitry Monakhov1c9114f2012-09-29 00:55:23 -04004680 }
Jan Kara52083862013-08-17 10:07:17 -04004681 /*
4682 * Truncate pagecache after we've waited for commit
4683 * in data=journal mode to make pages freeable.
4684 */
Kirill A. Shutemov7caef262013-09-12 15:13:56 -07004685 truncate_pagecache(inode, inode->i_size);
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004686 }
Jan Kara52083862013-08-17 10:07:17 -04004687 /*
4688 * We want to call ext4_truncate() even if attr->ia_size ==
4689 * inode->i_size for cases like truncation of fallocated space
4690 */
4691 if (attr->ia_valid & ATTR_SIZE)
4692 ext4_truncate(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004693
Christoph Hellwig10257742010-06-04 11:30:02 +02004694 if (!rc) {
4695 setattr_copy(inode, attr);
4696 mark_inode_dirty(inode);
4697 }
4698
4699 /*
4700 * If the call to ext4_truncate failed to get a transaction handle at
4701 * all, we need to clean up the in-core orphan list manually.
4702 */
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004703 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004704 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004705
4706 if (!rc && (ia_valid & ATTR_MODE))
Christoph Hellwig64e178a2013-12-20 05:16:44 -08004707 rc = posix_acl_chmod(inode, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004708
4709err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07004710 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004711 if (!error)
4712 error = rc;
4713 return error;
4714}
4715
Mingming Cao3e3398a2008-07-11 19:27:31 -04004716int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4717 struct kstat *stat)
4718{
4719 struct inode *inode;
Jan Kara8af8eec2013-05-31 19:39:56 -04004720 unsigned long long delalloc_blocks;
Mingming Cao3e3398a2008-07-11 19:27:31 -04004721
4722 inode = dentry->d_inode;
4723 generic_fillattr(inode, stat);
4724
4725 /*
Andreas Dilger9206c562013-11-11 22:38:12 -05004726 * If there is inline data in the inode, the inode will normally not
4727 * have data blocks allocated (it may have an external xattr block).
4728 * Report at least one sector for such files, so tools like tar, rsync,
4729 * others doen't incorrectly think the file is completely sparse.
4730 */
4731 if (unlikely(ext4_has_inline_data(inode)))
4732 stat->blocks += (stat->size + 511) >> 9;
4733
4734 /*
Mingming Cao3e3398a2008-07-11 19:27:31 -04004735 * We can't update i_blocks if the block allocation is delayed
4736 * otherwise in the case of system crash before the real block
4737 * allocation is done, we will have i_blocks inconsistent with
4738 * on-disk file blocks.
4739 * We always keep i_blocks updated together with real
4740 * allocation. But to not confuse with user, stat
4741 * will return the blocks that include the delayed allocation
4742 * blocks for this file.
4743 */
Tao Ma96607552012-05-31 22:54:16 -04004744 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
Andreas Dilger9206c562013-11-11 22:38:12 -05004745 EXT4_I(inode)->i_reserved_data_blocks);
4746 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
Mingming Cao3e3398a2008-07-11 19:27:31 -04004747 return 0;
4748}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004749
Jan Karafffb2732013-06-04 13:01:11 -04004750static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4751 int pextents)
Mingming Caoa02908f2008-08-19 22:16:07 -04004752{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004753 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Jan Karafffb2732013-06-04 13:01:11 -04004754 return ext4_ind_trans_blocks(inode, lblocks);
4755 return ext4_ext_index_trans_blocks(inode, pextents);
Mingming Caoa02908f2008-08-19 22:16:07 -04004756}
Theodore Ts'oac51d832008-11-06 16:49:36 -05004757
Mingming Caoa02908f2008-08-19 22:16:07 -04004758/*
4759 * Account for index blocks, block groups bitmaps and block group
4760 * descriptor blocks if modify datablocks and index blocks
4761 * worse case, the indexs blocks spread over different block groups
4762 *
4763 * If datablocks are discontiguous, they are possible to spread over
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004764 * different block groups too. If they are contiguous, with flexbg,
Mingming Caoa02908f2008-08-19 22:16:07 -04004765 * they could still across block group boundary.
4766 *
4767 * Also account for superblock, inode, quota and xattr blocks
4768 */
Jan Karafffb2732013-06-04 13:01:11 -04004769static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4770 int pextents)
Mingming Caoa02908f2008-08-19 22:16:07 -04004771{
Theodore Ts'o8df96752009-05-01 08:50:38 -04004772 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4773 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04004774 int idxblocks;
4775 int ret = 0;
4776
4777 /*
Jan Karafffb2732013-06-04 13:01:11 -04004778 * How many index blocks need to touch to map @lblocks logical blocks
4779 * to @pextents physical extents?
Mingming Caoa02908f2008-08-19 22:16:07 -04004780 */
Jan Karafffb2732013-06-04 13:01:11 -04004781 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
Mingming Caoa02908f2008-08-19 22:16:07 -04004782
4783 ret = idxblocks;
4784
4785 /*
4786 * Now let's see how many group bitmaps and group descriptors need
4787 * to account
4788 */
Jan Karafffb2732013-06-04 13:01:11 -04004789 groups = idxblocks + pextents;
Mingming Caoa02908f2008-08-19 22:16:07 -04004790 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04004791 if (groups > ngroups)
4792 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04004793 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4794 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4795
4796 /* bitmaps and block group descriptor blocks */
4797 ret += groups + gdpblocks;
4798
4799 /* Blocks for super block, inode, quota and xattr blocks */
4800 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004801
4802 return ret;
4803}
4804
4805/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004806 * Calculate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04004807 * the modification of a single pages into a single transaction,
4808 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04004809 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004810 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04004811 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004812 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04004813 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04004814 */
4815int ext4_writepage_trans_blocks(struct inode *inode)
4816{
4817 int bpp = ext4_journal_blocks_per_page(inode);
4818 int ret;
4819
Jan Karafffb2732013-06-04 13:01:11 -04004820 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
Mingming Caoa02908f2008-08-19 22:16:07 -04004821
4822 /* Account for data blocks for journalled mode */
4823 if (ext4_should_journal_data(inode))
4824 ret += bpp;
4825 return ret;
4826}
Mingming Caof3bd1f32008-08-19 22:16:03 -04004827
4828/*
4829 * Calculate the journal credits for a chunk of data modification.
4830 *
4831 * This is called from DIO, fallocate or whoever calling
Eric Sandeen79e83032010-07-27 11:56:07 -04004832 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04004833 *
4834 * journal buffers for data blocks are not included here, as DIO
4835 * and fallocate do no need to journal data buffers.
4836 */
4837int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4838{
4839 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4840}
4841
Mingming Caoa02908f2008-08-19 22:16:07 -04004842/*
Mingming Cao617ba132006-10-11 01:20:53 -07004843 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004844 * Give this, we know that the caller already has write access to iloc->bh.
4845 */
Mingming Cao617ba132006-10-11 01:20:53 -07004846int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004847 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004848{
4849 int err = 0;
4850
Theodore Ts'oc64db502012-03-02 12:23:11 -05004851 if (IS_I_VERSION(inode))
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004852 inode_inc_iversion(inode);
4853
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004854 /* the do_update_inode consumes one bh->b_count */
4855 get_bh(iloc->bh);
4856
Mingming Caodab291a2006-10-11 01:21:01 -07004857 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04004858 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004859 put_bh(iloc->bh);
4860 return err;
4861}
4862
4863/*
4864 * On success, We end up with an outstanding reference count against
4865 * iloc->bh. This _must_ be cleaned up later.
4866 */
4867
4868int
Mingming Cao617ba132006-10-11 01:20:53 -07004869ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4870 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004871{
Frank Mayhar03901312009-01-07 00:06:22 -05004872 int err;
4873
4874 err = ext4_get_inode_loc(inode, iloc);
4875 if (!err) {
4876 BUFFER_TRACE(iloc->bh, "get_write_access");
4877 err = ext4_journal_get_write_access(handle, iloc->bh);
4878 if (err) {
4879 brelse(iloc->bh);
4880 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004881 }
4882 }
Mingming Cao617ba132006-10-11 01:20:53 -07004883 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004884 return err;
4885}
4886
4887/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004888 * Expand an inode by new_extra_isize bytes.
4889 * Returns 0 on success or negative error number on failure.
4890 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004891static int ext4_expand_extra_isize(struct inode *inode,
4892 unsigned int new_extra_isize,
4893 struct ext4_iloc iloc,
4894 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004895{
4896 struct ext4_inode *raw_inode;
4897 struct ext4_xattr_ibody_header *header;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004898
4899 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4900 return 0;
4901
4902 raw_inode = ext4_raw_inode(&iloc);
4903
4904 header = IHDR(inode, raw_inode);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004905
4906 /* No extended attributes present */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004907 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4908 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004909 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4910 new_extra_isize);
4911 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4912 return 0;
4913 }
4914
4915 /* try to expand with EAs present */
4916 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4917 raw_inode, handle);
4918}
4919
4920/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004921 * What we do here is to mark the in-core inode as clean with respect to inode
4922 * dirtiness (it may still be data-dirty).
4923 * This means that the in-core inode may be reaped by prune_icache
4924 * without having to perform any I/O. This is a very good thing,
4925 * because *any* task may call prune_icache - even ones which
4926 * have a transaction open against a different journal.
4927 *
4928 * Is this cheating? Not really. Sure, we haven't written the
4929 * inode out, but prune_icache isn't a user-visible syncing function.
4930 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4931 * we start and wait on commits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004932 */
Mingming Cao617ba132006-10-11 01:20:53 -07004933int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004934{
Mingming Cao617ba132006-10-11 01:20:53 -07004935 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004936 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4937 static unsigned int mnt_count;
4938 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004939
4940 might_sleep();
Theodore Ts'o7ff9c072010-11-08 13:51:33 -05004941 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
Mingming Cao617ba132006-10-11 01:20:53 -07004942 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05004943 if (ext4_handle_valid(handle) &&
4944 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004945 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004946 /*
4947 * We need extra buffer credits since we may write into EA block
4948 * with this same handle. If journal_extend fails, then it will
4949 * only result in a minor loss of functionality for that inode.
4950 * If this is felt to be critical, then e2fsck should be run to
4951 * force a large enough s_min_extra_isize.
4952 */
4953 if ((jbd2_journal_extend(handle,
4954 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4955 ret = ext4_expand_extra_isize(inode,
4956 sbi->s_want_extra_isize,
4957 iloc, handle);
4958 if (ret) {
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004959 ext4_set_inode_state(inode,
4960 EXT4_STATE_NO_EXPAND);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004961 if (mnt_count !=
4962 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05004963 ext4_warning(inode->i_sb,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004964 "Unable to expand inode %lu. Delete"
4965 " some EAs or run e2fsck.",
4966 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04004967 mnt_count =
4968 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004969 }
4970 }
4971 }
4972 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004973 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07004974 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004975 return err;
4976}
4977
4978/*
Mingming Cao617ba132006-10-11 01:20:53 -07004979 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004980 *
4981 * We're really interested in the case where a file is being extended.
4982 * i_size has been changed by generic_commit_write() and we thus need
4983 * to include the updated inode in the current transaction.
4984 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05004985 * Also, dquot_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004986 * are allocated to the file.
4987 *
4988 * If the inode is marked synchronous, we don't honour that here - doing
4989 * so would cause a commit on atime updates, which we don't bother doing.
4990 * We handle synchronous inodes at the highest possible level.
4991 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04004992void ext4_dirty_inode(struct inode *inode, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004993{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004994 handle_t *handle;
4995
Theodore Ts'o9924a922013-02-08 21:59:22 -05004996 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004997 if (IS_ERR(handle))
4998 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04004999
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005000 ext4_mark_inode_dirty(handle, inode);
5001
Mingming Cao617ba132006-10-11 01:20:53 -07005002 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005003out:
5004 return;
5005}
5006
5007#if 0
5008/*
5009 * Bind an inode's backing buffer_head into this transaction, to prevent
5010 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07005011 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005012 * returns no iloc structure, so the caller needs to repeat the iloc
5013 * lookup to mark the inode dirty later.
5014 */
Mingming Cao617ba132006-10-11 01:20:53 -07005015static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005016{
Mingming Cao617ba132006-10-11 01:20:53 -07005017 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005018
5019 int err = 0;
5020 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07005021 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005022 if (!err) {
5023 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07005024 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005025 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05005026 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05005027 NULL,
Frank Mayhar03901312009-01-07 00:06:22 -05005028 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005029 brelse(iloc.bh);
5030 }
5031 }
Mingming Cao617ba132006-10-11 01:20:53 -07005032 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005033 return err;
5034}
5035#endif
5036
Mingming Cao617ba132006-10-11 01:20:53 -07005037int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005038{
5039 journal_t *journal;
5040 handle_t *handle;
5041 int err;
5042
5043 /*
5044 * We have to be very careful here: changing a data block's
5045 * journaling status dynamically is dangerous. If we write a
5046 * data block to the journal, change the status and then delete
5047 * that block, we risk forgetting to revoke the old log record
5048 * from the journal and so a subsequent replay can corrupt data.
5049 * So, first we make sure that the journal is empty and that
5050 * nobody is changing anything.
5051 */
5052
Mingming Cao617ba132006-10-11 01:20:53 -07005053 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005054 if (!journal)
5055 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04005056 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005057 return -EROFS;
Yongqiang Yang2aff57b2011-12-28 12:02:13 -05005058 /* We have to allocate physical blocks for delalloc blocks
5059 * before flushing journal. otherwise delalloc blocks can not
5060 * be allocated any more. even more truncate on delalloc blocks
5061 * could trigger BUG by flushing delalloc blocks in journal.
5062 * There is no delalloc block in non-journal data mode.
5063 */
5064 if (val && test_opt(inode->i_sb, DELALLOC)) {
5065 err = ext4_alloc_da_blocks(inode);
5066 if (err < 0)
5067 return err;
5068 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005069
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04005070 /* Wait for all existing dio workers */
5071 ext4_inode_block_unlocked_dio(inode);
5072 inode_dio_wait(inode);
5073
Mingming Caodab291a2006-10-11 01:21:01 -07005074 jbd2_journal_lock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005075
5076 /*
5077 * OK, there are no updates running now, and all cached data is
5078 * synced to disk. We are now in a completely consistent state
5079 * which doesn't have anything in the journal, and we know that
5080 * no filesystem updates are running, so it is safe to modify
5081 * the inode's in-core data-journaling state flag now.
5082 */
5083
5084 if (val)
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005085 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05005086 else {
5087 jbd2_journal_flush(journal);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005088 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05005089 }
Mingming Cao617ba132006-10-11 01:20:53 -07005090 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005091
Mingming Caodab291a2006-10-11 01:21:01 -07005092 jbd2_journal_unlock_updates(journal);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04005093 ext4_inode_resume_unlocked_dio(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005094
5095 /* Finally we can mark the inode as dirty. */
5096
Theodore Ts'o9924a922013-02-08 21:59:22 -05005097 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005098 if (IS_ERR(handle))
5099 return PTR_ERR(handle);
5100
Mingming Cao617ba132006-10-11 01:20:53 -07005101 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005102 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07005103 ext4_journal_stop(handle);
5104 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005105
5106 return err;
5107}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005108
5109static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5110{
5111 return !buffer_mapped(bh);
5112}
5113
Nick Pigginc2ec1752009-03-31 15:23:21 -07005114int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005115{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005116 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005117 loff_t size;
5118 unsigned long len;
Jan Kara9ea7df52011-06-24 14:29:41 -04005119 int ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005120 struct file *file = vma->vm_file;
Al Viro496ad9a2013-01-23 17:07:38 -05005121 struct inode *inode = file_inode(file);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005122 struct address_space *mapping = inode->i_mapping;
Jan Kara9ea7df52011-06-24 14:29:41 -04005123 handle_t *handle;
5124 get_block_t *get_block;
5125 int retries = 0;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005126
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005127 sb_start_pagefault(inode->i_sb);
Theodore Ts'o041bbb6d2012-09-30 23:04:56 -04005128 file_update_time(vma->vm_file);
Jan Kara9ea7df52011-06-24 14:29:41 -04005129 /* Delalloc case is easy... */
5130 if (test_opt(inode->i_sb, DELALLOC) &&
5131 !ext4_should_journal_data(inode) &&
5132 !ext4_nonda_switch(inode->i_sb)) {
5133 do {
5134 ret = __block_page_mkwrite(vma, vmf,
5135 ext4_da_get_block_prep);
5136 } while (ret == -ENOSPC &&
5137 ext4_should_retry_alloc(inode->i_sb, &retries));
5138 goto out_ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005139 }
Darrick J. Wong0e499892011-05-18 13:55:20 -04005140
5141 lock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005142 size = i_size_read(inode);
5143 /* Page got truncated from under us? */
5144 if (page->mapping != mapping || page_offset(page) > size) {
5145 unlock_page(page);
5146 ret = VM_FAULT_NOPAGE;
5147 goto out;
Darrick J. Wong0e499892011-05-18 13:55:20 -04005148 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005149
5150 if (page->index == size >> PAGE_CACHE_SHIFT)
5151 len = size & ~PAGE_CACHE_MASK;
5152 else
5153 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005154 /*
Jan Kara9ea7df52011-06-24 14:29:41 -04005155 * Return if we have all the buffers mapped. This avoids the need to do
5156 * journal_start/journal_stop which can block and take a long time
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005157 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005158 if (page_has_buffers(page)) {
Tao Maf19d5872012-12-10 14:05:51 -05005159 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5160 0, len, NULL,
5161 ext4_bh_unmapped)) {
Jan Kara9ea7df52011-06-24 14:29:41 -04005162 /* Wait so that we don't change page under IO */
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08005163 wait_for_stable_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005164 ret = VM_FAULT_LOCKED;
5165 goto out;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005166 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005167 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005168 unlock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005169 /* OK, we need to fill the hole... */
5170 if (ext4_should_dioread_nolock(inode))
5171 get_block = ext4_get_block_write;
5172 else
5173 get_block = ext4_get_block;
5174retry_alloc:
Theodore Ts'o9924a922013-02-08 21:59:22 -05005175 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5176 ext4_writepage_trans_blocks(inode));
Jan Kara9ea7df52011-06-24 14:29:41 -04005177 if (IS_ERR(handle)) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07005178 ret = VM_FAULT_SIGBUS;
Jan Kara9ea7df52011-06-24 14:29:41 -04005179 goto out;
5180 }
5181 ret = __block_page_mkwrite(vma, vmf, get_block);
5182 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -05005183 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
Jan Kara9ea7df52011-06-24 14:29:41 -04005184 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5185 unlock_page(page);
5186 ret = VM_FAULT_SIGBUS;
Yongqiang Yangfcbb5512011-10-26 05:00:19 -04005187 ext4_journal_stop(handle);
Jan Kara9ea7df52011-06-24 14:29:41 -04005188 goto out;
5189 }
5190 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5191 }
5192 ext4_journal_stop(handle);
5193 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5194 goto retry_alloc;
5195out_ret:
5196 ret = block_page_mkwrite_return(ret);
5197out:
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005198 sb_end_pagefault(inode->i_sb);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005199 return ret;
5200}