Hi, as I haven't found discussion by searching, I thought I bring up a gotcha I ran into today, on a host recently updated to tumbleweed 20230801 level. With a somewhat recent update to e2fsprogs, now at 1.47.0 compared to May '23 at 1.46.5, the set of mkfs flags has changed: --- /root/mke2fs.conf.20230512 2023-08-09 12:00:14.379476597 +0200 +++ /etc/mke2fs.conf 2023-06-14 19:20:22.000000000 +0200 @@ -14 +14 @@ - features = has_journal,extent,huge_file,flex_bg,metadata_csum,64bit,dir_nlink,extra_isize + features = has_journal,extent,huge_file,flex_bg,metadata_csum,metadata_csum_seed,64bit,dir_nlink,extra_isize,orphan_file For me, this resulted in an otherwise fine virtual machine with 20230512 userlevel, not booting after an "rsync copy migration" into a fresh root filesystem mkfs'ed with these new flags. Reason being that 3 months older fsck - the sytemd boot time rootfs service actually - goes "failed". I guess I'll need just use flag override options going forward, to cope with that, but thought I'd bring it to attention here at least. best regards Patrick
On Wed, Aug 9, 2023 at 1:11 PM Patrick Schaaf <patrick.schaaf@yalwa.com> wrote:
Hi,
as I haven't found discussion by searching, I thought I bring up a gotcha I ran into today, on a host recently updated to tumbleweed 20230801 level.
With a somewhat recent update to e2fsprogs, now at 1.47.0 compared to May '23 at 1.46.5, the set of mkfs flags has changed:
--- /root/mke2fs.conf.20230512 2023-08-09 12:00:14.379476597 +0200 +++ /etc/mke2fs.conf 2023-06-14 19:20:22.000000000 +0200 @@ -14 +14 @@ - features = has_journal,extent,huge_file,flex_bg,metadata_csum,64bit,dir_nlink,extra_isize + features = has_journal,extent,huge_file,flex_bg,metadata_csum,metadata_csum_seed,64bit,dir_nlink,extra_isize,orphan_file
This is the upstream change. The mke2fs program (via the mke2fs.conf file) now enables the metadata_csum_seed and orphan_file features by default.
On Wed, 9 Aug 2023 at 13:01, Andrei Borzenkov <arvidjaar@gmail.com> wrote:
This is the upstream change.
The mke2fs program (via the mke2fs.conf file) now enables the metadata_csum_seed and orphan_file features by default.
Yeah, I get that. What I find rather unintuitive and potentially problematic for some more people, is that the exact previous version (in tumbleweed) of fsck, fails with the same flags. Here is what it says, on a block device I made with the newer mkfs flags: hostA:~ # fsck.ext4 /dev/system/test-blk e2fsck 1.46.5 (30-Dec-2021) /dev/system/test-blk has unsupported feature(s): FEATURE_C12 e2fsck: Get a newer version of e2fsck! With the updated package, of course it works hostB:~ # fsck.ext4 /dev/system/test-blk e2fsck 1.47.0 (5-Feb-2023) /dev/system/test-blk: clean, 12/16384 files, 9545/65536 blocks I would have hoped that fsck would at least maybe for one or two major releases, would already learn about new flags before they become enabled for mkfs. But obviously that's not the case. A suggestion for tumbleweed, maybe, for the future, would be to leave out new flags in the /etc/mke2fs.conf file for one or two such (e2fsprogs) release cycles, to smooth the transition. For my own setup, I guess that's what I will do with some "sed -i" reverting the change for a year or so... best regards Patrick
participants (3)
-
Andrei Borzenkov -
deeplymonth7822＠gmail.com -
Patrick Schaaf