| blob | 59bb98ebdd1cb3544e652a691c6f53efaac222ec |
1 ext4: add optimization for the lazytime mount option
3 Add an optimization for the MS_LAZYTIME mount option so that we will
4 opportunistically write out any inodes with the I_DIRTY_TIME flag set
5 in a particular inode table block when we need to update some inode in
6 that inode table block anyway.
8 Also add some temporary code so that we can set the lazytime mount
9 option without needing a modified /sbin/mount program which can set
10 MS_LAZYTIME. We can eventually make this go away once util-linux has
11 added support.
13 Google-Bug-Id: 18297052
15 Signed-off-by: Theodore Ts'o <tytso@mit.edu>
16 ---
17 fs/ext4/inode.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++---
18 fs/ext4/super.c | 9 ++++++++
19 include/trace/events/ext4.h | 30 +++++++++++++++++++++++++++
20 3 files changed, 102 insertions(+), 3 deletions(-)
22 diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
23 index 5653fa4..0e60d90 100644
24 --- a/fs/ext4/inode.c
25 +++ b/fs/ext4/inode.c
26 @@ -4139,6 +4139,66 @@ static int ext4_inode_blocks_set(handle_t *handle,
27 return 0;
28 }
30 +struct other_inode {
31 + unsigned long orig_ino;
32 + struct ext4_inode *raw_inode;
33 +};
34 +
35 +static int other_inode_match(struct inode * inode, unsigned long ino,
36 + void *data)
37 +{
38 + struct other_inode *oi = (struct other_inode *) data;
39 +
40 + if ((inode->i_ino != ino) ||
41 + (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
42 + I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
43 + ((inode->i_state & I_DIRTY_TIME) == 0))
44 + return 0;
45 + spin_lock(&inode->i_lock);
46 + if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
47 + I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
48 + (inode->i_state & I_DIRTY_TIME)) {
49 + struct ext4_inode_info *ei = EXT4_I(inode);
50 +
51 + inode->i_state &= ~I_DIRTY_TIME;
52 + inode->i_ts_dirty_day = 0;
53 + spin_unlock(&inode->i_lock);
54 +
55 + spin_lock(&ei->i_raw_lock);
56 + EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
57 + EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
58 + EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
59 + ext4_inode_csum_set(inode, oi->raw_inode, ei);
60 + spin_unlock(&ei->i_raw_lock);
61 + trace_ext4_other_inode_update_time(inode, oi->orig_ino);
62 + return -1;
63 + }
64 + spin_unlock(&inode->i_lock);
65 + return -1;
66 +}
67 +
68 +/*
69 + * Opportunistically update the other time fields for other inodes in
70 + * the same inode table block.
71 + */
72 +static void ext4_update_other_inodes_time(struct super_block *sb,
73 + unsigned long orig_ino, char *buf)
74 +{
75 + struct other_inode oi;
76 + unsigned long ino;
77 + int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
78 + int inode_size = EXT4_INODE_SIZE(sb);
79 +
80 + oi.orig_ino = orig_ino;
81 + ino = orig_ino & ~(inodes_per_block - 1);
82 + for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
83 + if (ino == orig_ino)
84 + continue;
85 + oi.raw_inode = (struct ext4_inode *) buf;
86 + (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
87 + }
88 +}
89 +
90 /*
91 * Post the struct inode info into an on-disk inode location in the
92 * buffer-cache. This gobbles the caller's reference to the
93 @@ -4237,7 +4297,6 @@ static int ext4_do_update_inode(handle_t *handle,
94 for (block = 0; block < EXT4_N_BLOCKS; block++)
95 raw_inode->i_block[block] = ei->i_data[block];
96 }
97 -
98 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
99 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
100 if (ei->i_extra_isize) {
101 @@ -4248,10 +4307,11 @@ static int ext4_do_update_inode(handle_t *handle,
102 cpu_to_le16(ei->i_extra_isize);
103 }
104 }
105 -
106 ext4_inode_csum_set(inode, raw_inode, ei);
107 -
108 spin_unlock(&ei->i_raw_lock);
109 + if (inode->i_sb->s_flags & MS_LAZYTIME)
110 + ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
111 + bh->b_data);
113 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
114 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
115 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
116 index 58859bc..93a2b7a 100644
117 --- a/fs/ext4/super.c
118 +++ b/fs/ext4/super.c
119 @@ -1132,6 +1132,7 @@ enum {
120 Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
121 Opt_usrquota, Opt_grpquota, Opt_i_version,
122 Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
123 + Opt_lazytime, Opt_nolazytime,
124 Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
125 Opt_inode_readahead_blks, Opt_journal_ioprio,
126 Opt_dioread_nolock, Opt_dioread_lock,
127 @@ -1195,6 +1196,8 @@ static const match_table_t tokens = {
128 {Opt_i_version, "i_version"},
129 {Opt_stripe, "stripe=%u"},
130 {Opt_delalloc, "delalloc"},
131 + {Opt_lazytime, "lazytime"},
132 + {Opt_nolazytime, "nolazytime"},
133 {Opt_nodelalloc, "nodelalloc"},
134 {Opt_removed, "mblk_io_submit"},
135 {Opt_removed, "nomblk_io_submit"},
136 @@ -1452,6 +1455,12 @@ static int handle_mount_opt(struct super_block *sb, char *opt, int token,
137 case Opt_i_version:
138 sb->s_flags |= MS_I_VERSION;
139 return 1;
140 + case Opt_lazytime:
141 + sb->s_flags |= MS_LAZYTIME;
142 + return 1;
143 + case Opt_nolazytime:
144 + sb->s_flags &= ~MS_LAZYTIME;
145 + return 1;
146 }
148 for (m = ext4_mount_opts; m->token != Opt_err; m++)
149 diff --git a/include/trace/events/ext4.h b/include/trace/events/ext4.h
150 index 6cfb841..6e5abd6 100644
151 --- a/include/trace/events/ext4.h
152 +++ b/include/trace/events/ext4.h
153 @@ -73,6 +73,36 @@ struct extent_status;
154 { FALLOC_FL_ZERO_RANGE, "ZERO_RANGE"})
157 +TRACE_EVENT(ext4_other_inode_update_time,
158 + TP_PROTO(struct inode *inode, ino_t orig_ino),
159 +
160 + TP_ARGS(inode, orig_ino),
161 +
162 + TP_STRUCT__entry(
163 + __field( dev_t, dev )
164 + __field( ino_t, ino )
165 + __field( ino_t, orig_ino )
166 + __field( uid_t, uid )
167 + __field( gid_t, gid )
168 + __field( __u16, mode )
169 + ),
170 +
171 + TP_fast_assign(
172 + __entry->orig_ino = orig_ino;
173 + __entry->dev = inode->i_sb->s_dev;
174 + __entry->ino = inode->i_ino;
175 + __entry->uid = i_uid_read(inode);
176 + __entry->gid = i_gid_read(inode);
177 + __entry->mode = inode->i_mode;
178 + ),
179 +
180 + TP_printk("dev %d,%d orig_ino %lu ino %lu mode 0%o uid %u gid %u",
181 + MAJOR(__entry->dev), MINOR(__entry->dev),
182 + (unsigned long) __entry->orig_ino,
183 + (unsigned long) __entry->ino, __entry->mode,
184 + __entry->uid, __entry->gid)
185 +);
186 +
187 TRACE_EVENT(ext4_free_inode,
188 TP_PROTO(struct inode *inode),