Re: mounting with lazytime doesn't work on ext4
From: Lukas Czerner <hidden>
Date: 2017-09-19 14:46:22
Also in:
linux-fsdevel
Subsystem:
filesystems (vfs and infrastructure), the rest · Maintainers:
Alexander Viro, Christian Brauner, Linus Torvalds
On Tue, Sep 19, 2017 at 10:35:06AM +0200, Markus Trippelsdorf wrote:
quoted hunk ↗ jump to hunk
On 2017.09.18 at 21:26 +0200, Markus Trippelsdorf wrote:quoted
I switched back to ext4 yesterday, because my btrfs fs got corrupted. However mounting with lazytime doesn't work, neither specifying it in /etc/fstab nor a manual remount. It looks like the option is simply ignored. Strace shows, e.g.: # mount -o lazytime /boot mount("/dev/sdc2", "/boot", "ext4", MS_LAZYTIME, NULL) = 0 EXT4-fs (sdc2): mounted filesystem with ordered data mode. Opts: (null) /dev/sdc2 on /boot type ext4 (rw,relatime,data=ordered) # mount -o remount,lazytime /var mount("/dev/sdb2", "/var", 0x12c4460, MS_REMOUNT|MS_NOATIME|MS_LAZYTIME, NULL) = 0 EXT4-fs (sdb2): re-mounted. Opts: (null) /dev/sdb2 on /var type ext4 (rw,noatime,data=ordered) When I set "sb->s_flags |= MS_LAZYTIME;" unconditionally in fs/ext4/super.c:5057 (just deleting the if statement), then lazytime gets used when I remount. I'm running the latest git tree (4.14.0-rc1).The following patch seems to fix the issue for remounts:diff --git a/fs/namespace.c b/fs/namespace.c index 54059b142d6b..d0b386706c5b 100644 --- a/fs/namespace.c +++ b/fs/namespace.c@@ -2322,6 +2322,9 @@ static int do_remount(struct path *path, int ms_flags, int sb_flags, if (err) return err; + if (mnt_flags & MS_LAZYTIME) + sb_flags |= MS_LAZYTIME; + down_write(&sb->s_umount); if (ms_flags & MS_BIND) err = change_mount_flags(path->mnt, ms_flags);@@ -2809,6 +2812,8 @@ long do_mount(const char *dev_name, const char __user *dir_name, mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); if (flags & SB_RDONLY) mnt_flags |= MNT_READONLY; + if (flags & MS_LAZYTIME) + mnt_flags |= MS_LAZYTIME;
Hi, this is definitely not right. Currently it seems that MS_LAZYTIME is supposed to be per sb option, even though I do not see a reason why it should not be per mnt options, in fact I'd prefer that. However the problem happened when e462ec50cb5fad19f6003a3d8087f4a0945dd2b1 switched from masking out per mnt flags to masking in per sb flags. However David missed SB_LAZYFLAG. The fix should be easy like this:
diff --git a/fs/namespace.c b/fs/namespace.c
index 54059b1..3b48ef1 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c@@ -2823,7 +2823,8 @@ long do_mount(const char *dev_name, const char __user *dir_name, SB_MANDLOCK | SB_DIRSYNC | SB_SILENT | - SB_POSIXACL); + SB_POSIXACL | + SB_LAZYTIME); if (flags & MS_REMOUNT) retval = do_remount(&path, flags, sb_flags, mnt_flags,
I can send a proper patch, however maybe it would be better to convert it to per-mountpoint flags instead ? Any thoughts Ted ? -Lukas