| Message ID | 20241004221556.19222-1-jack@suse.cz |
|---|---|
| State | New |
| Headers | show |
| Series | ext4: Avoid remount errors with 'abort' mount option | expand |
On Sat, Oct 5, 2024 at 12:16 AM Jan Kara <jack@suse.cz> wrote: > > When we remount filesystem with 'abort' mount option while changing > other mount options as well (as is LTP test doing), we can return error > from the system call after commit d3476f3dad4a ("ext4: don't set > SB_RDONLY after filesystem errors") because the application of mount > option changes detects shutdown filesystem and refuses to do anything. > The behavior of application of other mount options in presence of > 'abort' mount option is currently rather arbitary as some mount option > changes are handled before 'abort' and some after it. > > Move aborting of the filesystem to the end of remount handling so all > requested changes are properly applied before the filesystem is shutdown > to have a reasonably consistent behavior. > > Fixes: d3476f3dad4a ("ext4: don't set SB_RDONLY after filesystem errors") > Reported-by: Jan Stancek <jstancek@redhat.com> > Link: https://lore.kernel.org/all/Zvp6L+oFnfASaoHl@t14s > Signed-off-by: Jan Kara <jack@suse.cz> In case the mount option isn't getting deprecated right away: Tested-by: Jan Stancek <jstancek@redhat.com> > --- > fs/ext4/super.c | 11 ++++++++--- > 1 file changed, 8 insertions(+), 3 deletions(-) > > diff --git a/fs/ext4/super.c b/fs/ext4/super.c > index 16a4ce704460..4645f1629673 100644 > --- a/fs/ext4/super.c > +++ b/fs/ext4/super.c > @@ -6518,9 +6518,6 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb) > goto restore_opts; > } > > - if (test_opt2(sb, ABORT)) > - ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); > - > sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | > (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); > > @@ -6689,6 +6686,14 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb) > if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) > ext4_stop_mmpd(sbi); > > + /* > + * Handle aborting the filesystem as the last thing during remount to > + * avoid obsure errors during remount when some option changes fail to > + * apply due to shutdown filesystem. > + */ > + if (test_opt2(sb, ABORT)) > + ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); > + > return 0; > > restore_opts: > -- > 2.35.3 >
On Mon 07-10-24 11:13:38, Jan Stancek wrote: > On Sat, Oct 5, 2024 at 12:16 AM Jan Kara <jack@suse.cz> wrote: > > > > When we remount filesystem with 'abort' mount option while changing > > other mount options as well (as is LTP test doing), we can return error > > from the system call after commit d3476f3dad4a ("ext4: don't set > > SB_RDONLY after filesystem errors") because the application of mount > > option changes detects shutdown filesystem and refuses to do anything. > > The behavior of application of other mount options in presence of > > 'abort' mount option is currently rather arbitary as some mount option > > changes are handled before 'abort' and some after it. > > > > Move aborting of the filesystem to the end of remount handling so all > > requested changes are properly applied before the filesystem is shutdown > > to have a reasonably consistent behavior. > > > > Fixes: d3476f3dad4a ("ext4: don't set SB_RDONLY after filesystem errors") > > Reported-by: Jan Stancek <jstancek@redhat.com> > > Link: https://lore.kernel.org/all/Zvp6L+oFnfASaoHl@t14s > > Signed-off-by: Jan Kara <jack@suse.cz> > > In case the mount option isn't getting deprecated right away: > Tested-by: Jan Stancek <jstancek@redhat.com> Well, even if we decided to deprecate it, it would mean that now we'd just start warning about deprecation and it would get removed in an year or so. So fixing this still makes sense. Thanks for testing! Honza > > --- > > fs/ext4/super.c | 11 ++++++++--- > > 1 file changed, 8 insertions(+), 3 deletions(-) > > > > diff --git a/fs/ext4/super.c b/fs/ext4/super.c > > index 16a4ce704460..4645f1629673 100644 > > --- a/fs/ext4/super.c > > +++ b/fs/ext4/super.c > > @@ -6518,9 +6518,6 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb) > > goto restore_opts; > > } > > > > - if (test_opt2(sb, ABORT)) > > - ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); > > - > > sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | > > (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); > > > > @@ -6689,6 +6686,14 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb) > > if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) > > ext4_stop_mmpd(sbi); > > > > + /* > > + * Handle aborting the filesystem as the last thing during remount to > > + * avoid obsure errors during remount when some option changes fail to > > + * apply due to shutdown filesystem. > > + */ > > + if (test_opt2(sb, ABORT)) > > + ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); > > + > > return 0; > > > > restore_opts: > > -- > > 2.35.3 > > >
diff --git a/fs/ext4/super.c b/fs/ext4/super.c index 16a4ce704460..4645f1629673 100644 --- a/fs/ext4/super.c +++ b/fs/ext4/super.c @@ -6518,9 +6518,6 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb) goto restore_opts; } - if (test_opt2(sb, ABORT)) - ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); - sb->s_flags = (sb->s_flags & ~SB_POSIXACL) | (test_opt(sb, POSIX_ACL) ? SB_POSIXACL : 0); @@ -6689,6 +6686,14 @@ static int __ext4_remount(struct fs_context *fc, struct super_block *sb) if (!ext4_has_feature_mmp(sb) || sb_rdonly(sb)) ext4_stop_mmpd(sbi); + /* + * Handle aborting the filesystem as the last thing during remount to + * avoid obsure errors during remount when some option changes fail to + * apply due to shutdown filesystem. + */ + if (test_opt2(sb, ABORT)) + ext4_abort(sb, ESHUTDOWN, "Abort forced by user"); + return 0; restore_opts:
When we remount filesystem with 'abort' mount option while changing other mount options as well (as is LTP test doing), we can return error from the system call after commit d3476f3dad4a ("ext4: don't set SB_RDONLY after filesystem errors") because the application of mount option changes detects shutdown filesystem and refuses to do anything. The behavior of application of other mount options in presence of 'abort' mount option is currently rather arbitary as some mount option changes are handled before 'abort' and some after it. Move aborting of the filesystem to the end of remount handling so all requested changes are properly applied before the filesystem is shutdown to have a reasonably consistent behavior. Fixes: d3476f3dad4a ("ext4: don't set SB_RDONLY after filesystem errors") Reported-by: Jan Stancek <jstancek@redhat.com> Link: https://lore.kernel.org/all/Zvp6L+oFnfASaoHl@t14s Signed-off-by: Jan Kara <jack@suse.cz> --- fs/ext4/super.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)