On 8/26/21 5:08 PM, Bill O'Donnell wrote:
On Thu, Aug 26, 2021 at 01:16:22PM -0500, Eric Sandeen wrote:
On 8/26/21 1:09 PM, Darrick J. Wong wrote:
On Thu, Aug 26, 2021 at 12:30:12PM -0500, Bill O'Donnell wrote:
@@ -1584,7 +1586,7 @@ xfs_fs_fill_super(
if (xfs_has_crc(mp))
sb->s_flags |= SB_I_VERSION;
- if (xfs_has_dax_always(mp)) {
+ if (xfs_has_dax_always(mp) || xfs_has_dax_inode(mp)) {
Er... can't this be done without burning another feature bit by:
if (xfs_has_dax_always(mp) || (!xfs_has_dax_always(mp) &&
!xfs_has_dax_never(mp))) {
...
xfs_warn(mp, "DAX IS EXPERIMENTAL");
}
changing this conditional in this way will also fail dax=inode mounts on
reflink-capable (i.e. default) filesystems, no?
Correct. My original patch tests fine, and still handles the reflink and dax
incompatibility. The new suggested logic is problematic.
-Bill
I think that both your proposed patch and Darrick's suggestion have this problem.
"mount -o dax=inode" makes your new xfs_has_dax_inode(mp) true, and in that
conditional, if the filesystem has reflink enabled, mount fails:
# mkfs.xfs -f /dev/pmem0p1
meta-data=/dev/pmem0p1 isize=512 agcount=4, agsize=4194304 blks
= sectsz=4096 attr=2, projid32bit=1
= crc=1 finobt=1, sparse=1, rmapbt=0
= reflink=1 bigtime=0 inobtcount=0
data = bsize=4096 blocks=16777216, imaxpct=25
= sunit=0 swidth=0 blks
naming =version 2 bsize=4096 ascii-ci=0, ftype=1
log =internal log bsize=4096 blocks=8192, version=2
= sectsz=4096 sunit=1 blks, lazy-count=1
realtime =none extsz=4096 blocks=0, rtextents=0
# mount -o dax=inode /dev/pmem0p1 /mnt/test
mount: wrong fs type, bad option, bad superblock on /dev/pmem0p1,
missing codepage or helper program, or other error
In some cases useful info is found in syslog - try
dmesg | tail or so.
# dmesg | tail -n 2
[ 192.691733] XFS (pmem0p1): DAX enabled. Warning: EXPERIMENTAL, use at your own risk
[ 192.700300] XFS (pmem0p1): DAX and reflink cannot be used together!
