lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Wed, 25 Aug 2021 13:48:18 -0400
From:   "Theodore Ts'o" <>
To:     Jan Kara <>
Subject: Re: [PATCH 0/5 v7] ext4: Speedup orphan file handling

On Wed, Aug 25, 2021 at 06:13:31PM +0200, Jan Kara wrote:
> So I had a look into the other failures... So ext4/044 works for me after
> fixing e2fsck (both in 1k and 4k cases). ext4/033, ext4/045, generic/273
> fail for me in the 1k case even without orphan file patches so I don't
> think they are a regression caused by my changes (specifically ext4/045 is
> a buggy test - I think the directory h-tree is not able to hold that many
> directories for 1k block size). Interestingly, I couldn't make generic/476
> fail for me either with or without my patches so that may be some random
> failure. I'm now running that test in a loop to see whether the failure
> will reproduce to investigate.

Oh, you're right.  I had forgotten I had the following in my
1k.exclude file, and I hadn't copied them over when I set up the
orphan_file_1k config.

# The test fails due to too many block group descriptors when the
# block size is 1k

# This test uses dioread_nolock which currently isn't supported when
# block_size != PAGE_SIZE.

# This test tries to create 65536 directories, and with 1k blocks,
# and long names, we run out of htree depth

# This test creates too many inodes on when the block size is 1k
# without using special mkfs.ext4 options to change the inode size.
# This test is a bit bogus anyway, and uses a bunch of magic calculations
# where it's not clear what it was originally trying to test in the
# first place.  So let's just skip it for now.

# This test creates too many extended attributes to fit in a 1k block

# Normal configurations don't support dax
-g dax

(We can drop ext4/034 from the exclude list since we now *do* support
dioread_nolock when block_size < page_size.)

Thanks for the investigating, and apologies for not the reason why I
hadn't seen the failures in the 1k case was because they had been

						- Ted

Powered by blists - more mailing lists