Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Ojaswin Mujoo <ojaswin@linux.ibm.com>
To:     linux-ext4@vger.kernel.org, "Theodore Ts'o" <tytso@mit.edu>
Cc:     Ritesh Harjani <riteshh@linux.ibm.com>,
        linux-kernel@vger.kernel.org, Jan Kara <jack@suse.cz>,
        Naresh Kamboju <naresh.kamboju@linaro.org>
Subject: [PATCH 0/1] ext4: Fix regression in mballoc due to deleted inode PAs in rbtree
Date:   Sat, 22 Jul 2023 22:45:23 +0530
Message-Id: <cover.1690045963.git.ojaswin@linux.ibm.com>
Content-Transfer-Encoding: 8bit
MIME-Version: 1.0
Precedence: bulk

Hello,

Recently there was a regression found in the mballoc [1] due the existence
of deleted preallocations(PAs) in the per inode preallocation rbtree.
Such deleted PAs can occur because ext4_mb_discard_group_preallocations 
traverses the grp->bb_prealloc_list and marks the PAs as deleted without
taking any inode specific locks like i_data_sem. 

Due to presence of such PAs, we were sometimes missing some of the pa
entries when traversing the per node rbtree in ext4_mb_use_preallocated.
Due to this, in some rare conditions we ended up missing a PA that did
overlap with our original request start. when this happens, we exit 
ext4_mb_use_preallocated and proceed with the allocation. However,
during ext4_mb_normalize_request() we were hitting a bug on as
a PA that could satisfy our request already existed. Since normalize
request was already fixed earlier to account for the deleted PAs we
actually able to catch it there.

This patch aims to fix this regression by using a different appraoch to
traverse the rbtree in ext4_mb_use_preallocated(). More comments can be
found in the commit message.

We've run several overnight runs of xfstests generic/269 as well as
several hours of ltp test on both x86 machines with 4k blocks size and
Power machines with 64k and 4k block size and have yet to hit the
regression. Further we added some debug prints in our testing to make
sure we were correctly handling the conditions that were triggering the
bug ons previously.

[1]
https://lore.kernel.org/linux-ext4/CA+G9fYv2FRpLqBZf34ZinR8bU2_ZRAUOjKAD3+tKRFaEQHtt8Q@mail.gmail.com/

Regards,
ojaswin

Ojaswin Mujoo (1):
  ext4: Fix rbtree traversal bug in ext4_mb_use_preallocated

 fs/ext4/mballoc.c | 158 ++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 131 insertions(+), 27 deletions(-)

-- 
2.31.1