Re: HEADS UP: SUJ Going in to head today

Hi Jeff,
thank you for your effort in implementing the soft update journaling.
I tried to test SUJ on a provider with 4 kB block size. My system runs
9-CURRENT r207195 (i386).
Unfortunately, tunefs is unable to cope with the device. It can easily
reproduced with these steps:

# mdconfig -s 128M -S 4096
0
#  newfs -U /dev/md0
/dev/md0: 128.0MB (262144 sectors) block size 16384, fragment size 4096
        using 4 cylinder groups of 32.02MB, 2049 blks, 2112 inodes.
        with soft updates
# tunefs -j enable /dev/md0
Using inode 4 in cg 0 for 4194304 byte journal
tunefs: Failed to read dir block: Invalid argument
tunefs: soft updates journaling can not be enabled

The bread() in tunefs.c:701 fails as the requested block size (512) is
smaller than the provider's block size (4096 bytes).

As a simply attempt to fix it, I changed tunefs.c:760 to "    if
(dir_extend(blk, nblk, size, ino) == -1)", as I thought that this made
more sense. Then, tunefs succeeded, but mounting the file system
resulted in a panic:
panic: ufs_dirbad: /mnt/md-test: bad dir ino 2 at offset 512: mangled entry

db:0:kdb.enter.default>  bt
Tracing pid 2714 tid 100262 td 0xc7ea6480
kdb_enter(c0a21226,c0a21226,c0a49886,eb1e6714,0,...) at kdb_enter+0x3a
panic(c0a49886,c688f468,2,200,c0a498df,...) at panic+0x136
ufs_dirbad(c81bb000,200,c0a498df,0,eb1e67b0,...) at ufs_dirbad+0x46
ufs_lookup_ino(c81d5990,0,eb1e67d8,eb1e6800,0,...) at ufs_lookup_ino+0x367
softdep_journal_lookup(c688f288,eb1e68c4,c0a45eca,750,eb1e6834,...) at
softdep_journal_lookup+0xb0
softdep_mount(c7e3fbb0,c688f288,c8165000,c7bdf900,c7bdf900,...) at
softdep_mount+0xdb
ffs_mount(c688f288,0,c0a2df89,3d6,0,...) at ffs_mount+0x23e1
vfs_donmount(c7ea6480,0,c7bc6100,c7bc6100,c8031000,...) at vfs_donmount+0x1000
nmount(c7ea6480,eb1e6cf8,c,c,207,...) at nmount+0x64
syscall(eb1e6d38) at syscall+0x1da
Xint0x80_syscall() at Xint0x80_syscall+0x20
--- syscall (378, FreeBSD ELF32, nmount), eip = 0x280f205b, esp =
0xbfbfdcec, ebp = 0xbfbfe248 ---

... so this attempt did not succeed, but was worth a try ;-)

But it would be nice to use SUJ even on such a unusual configuration.

Lucius