On 2022/10/24 15:46, Greg KH wrote:
On Mon, Oct 24, 2022 at 03:08:21PM +0800, Qu Wenruo wrote:
On 2022/10/16 20:24, gregkh@xxxxxxxxxxxxxxxxxxx wrote:
The patch below does not apply to the 5.10-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable@xxxxxxxxxxxxxxx>.
Possible dependencies:
81d5d61454c3 ("btrfs: enhance unsupported compat RO flags handling")
dfe8aec4520b ("btrfs: add a btrfs_block_group_root() helper")
b6e9f16c5fda ("btrfs: replace open coded while loop with proper construct")
42437a6386ff ("btrfs: introduce mount option rescue=ignorebadroots")
Hi Greg and btrfs guys,
I'm not confident enough to continue backporting this patch (and its
dependency) for older branches.
This patch itself relies on the new function fill_dummy_bgs() to skip the
full block group items search.
That function is the core functionality to allow us to mount fses read-only
with unsupported compat_ro flags.
But if we backport 42437a6386ff ("btrfs: introduce mount option
rescue=ignorebadroots"), we need the "rescue=" mount option series, which
can be super large, especially for older and older branches.
The other solution is to still implement that fill_dummy_bgs(), but not
introducing the "rescue=" mount option.
This means, we will have some moderate changes to 42437a6386ff ("btrfs:
introduce mount option rescue=ignorebadroots"), but at least no full
dependency on the "rescue=" mount options.
I'm wondering which solution would be more acceptable from the POV of stable
kernels?
First off, is this really needed in older kernels? It was marked as
such, but that's up to you all to decide if it's true or not.
This is mostly related to a corner case that, certain compat_ro flags
should be allowed to be mounted RO for older kernels.
There is already one such feature, called block-group-tree, and with
that patch, we should be able to mount a fs with such feature read-only.
(RW mount is still not supported until v6.1 kernel).
The root cause is from our uncertain understanding on which features
should be counted as compat_ro.
Until recently we don't really bothered incompat/compat_ro flags, and
tends to caused quite some incompatible features which should be compat_ro.
Thus I want to mark this patch for all stable kernels, and push for the
more accurate incompat/compat_ro flags splitting.
Without this patch, the new block-group-tree feature will just be
rejected (no matter RO or not) due to older kernel can not skip the
unnecessary block group items search.
Maybe it's not that important and time can solve it?
Thanks,
Qu
So I'll defer to you and the btrfs maintainers on what to do, just let
us know.
thanks,
greg k-h
