1 Subject: [PATCH-v7 0/3] add support for a lazytime mount option
3 This is an updated version of what had originally been an
4 ext4-specific patch which significantly improves performance by lazily
5 writing timestamp updates (and in particular, mtime updates) to disk.
6 The in-memory timestamps are always correct, but they are only written
7 to disk when required for correctness.
9 This provides a huge performance boost for ext4 due to how it handles
10 journalling, but it's valuable for all file systems running on flash
11 storage or drive-managed SMR disks by reducing the metadata write
12 load. So upon request, I've moved the functionality to the VFS layer.
13 Once the /sbin/mount program adds support for MS_LAZYTIME, all file
14 systems should be able to benefit from this optimization.
16 There is still an ext4-specific optimization, which may be applicable
17 for other file systems which store more than one inode in a block, but
18 it will require file system specific code. It is purely optional,
21 For people interested seeing how timestamp updates are held back, the
22 following example commands to enable the tracepoints debugging may be
25 mount -o remount,lazytime /
26 cd /sys/kernel/debug/tracing
27 echo 1 > events/writeback/writeback_lazytime/enable
28 echo 1 > events/writeback/writeback_lazytime_iput/enable
29 echo "state & 2048" > events/writeback/writeback_dirty_inode_enqueue/filter
30 echo 1 > events/writeback/writeback_dirty_inode_enqueue/enable
31 echo 1 > events/ext4/ext4_other_inode_update_time/enable
34 You can also see how many lazytime inodes are in memory by looking in
35 /sys/kernel/debug/bdi/<bdi>/stats
37 Changes since -v7: (not yet released)
39 - Clear the I_DIRTY_TIME flag if I_DIRTY_INODE gets added
42 - Add a new tracepoint writeback_dirty_inode_enqueue
43 - Move generic handling of update_time() to generic_update_time(),
44 so filesystems can more easily hook or modify update_time()
45 - The file system's dirty_inode() will now always get called with
46 I_DIRTY_TIME when the inode time is updated. (I_DIRTY_SYNC will
47 also be set if the inode should be updated right away.) This allows
48 file systems such as XFS to update its on-disk copy of the inode if
52 - Tweak move_expired_inodes to handle sync() and syncfs(), and drop
53 flush_sb_dirty_time().
54 - Move logic for handling the b_dirty_time list into
56 - Move I_DIRTY back to its original definition, and use I_DIRTY_ALL
57 for I_DIRTY plus I_DIRTY_TIME.
58 - Fold some patches together to make the first patch easier to
59 review (and modify/update).
60 - Use the pre-existing writeback tracepoints instead of creating a new
64 - Fix ext4 optimization so it does not need to increment (and more
65 problematically, decrement) the inode reference count
66 - Per Christoph's suggestion, drop support for btrfs and xfs for now,
67 issues with how btrfs and xfs handle dirty inode tracking. We can add
68 btrfs and xfs support back later or at the end of this series if we
69 want to revisit this decision.
70 - Miscellaneous cleanups
73 - inodes with I_DIRTY_TIME set are placed on a new bdi list,
74 b_dirty_time. This allows filesystem-level syncs to more
75 easily iterate over those inodes that need to have their
76 timestamps written to disk.
77 - dirty timestamps will be written out asynchronously on the final
78 iput, instead of when the inode gets evicted.
79 - separate the definition of the new function
80 find_active_inode_nowait() to a separate patch
81 - create separate flag masks: I_DIRTY_WB and I_DIRTY_INODE, which
82 indicate whether the inode needs to be on the write back lists,
83 or whether the inode itself is dirty, while I_DIRTY means any one
84 of the inode dirty flags are set. This simplifies the fs
85 writeback logic which needs to test for different combinations of
86 the inode dirty flags in different places.
89 - If update_time() updates i_version, it will not use lazytime (i..e,
90 the inode will be marked dirty so the change will be persisted on to
91 disk sooner rather than later). Yes, this eliminates the
92 benefits of lazytime if the user is experting the file system via
93 NFSv4. Sad, but NFS's requirements seem to mandate this.
94 - Fix time wrapping bug 49 days after the system boots (on a system
95 with a 32-bit jiffies). Use get_monotonic_boottime() instead.
96 - Clean up type warning in include/tracing/ext4.h
97 - Added explicit parenthesis for stylistic reasons
98 - Added an is_readonly() inode operations method so btrfs doesn't
99 have to duplicate code in update_time().
102 - Added explanatory comments in update_time() regarding i_ts_dirty_days
103 - Fix type used for days_since_boot
104 - Improve SMP scalability in update_time and ext4_update_other_inodes_time
105 - Added tracepoints to help test and characterize how often and under
106 what circumstances inodes have their timestamps lazily updated
109 vfs: add support for a lazytime mount option
110 vfs: add find_inode_nowait() function
111 ext4: add optimization for the lazytime mount option
113 fs/ext4/inode.c | 71 ++++++++++++++++++++++++--
114 fs/ext4/super.c | 10 ++++
115 fs/fs-writeback.c | 64 ++++++++++++++++++-----
116 fs/gfs2/file.c | 4 +-
117 fs/inode.c | 106 +++++++++++++++++++++++++++++++++------
120 fs/proc_namespace.c | 1 +
122 include/linux/backing-dev.h | 1 +
123 include/linux/fs.h | 10 ++++
124 include/trace/events/ext4.h | 30 +++++++++++
125 include/trace/events/writeback.h | 60 +++++++++++++++++++++-
126 include/uapi/linux/fs.h | 4 +-
127 mm/backing-dev.c | 10 +++-
128 15 files changed, 344 insertions(+), 39 deletions(-)