linux-kernel - [PATCH v3 0/2] fix quota subdir mounts

lists.openwall.net		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 linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 21 Mar 2019 10:20:08 +0000
From:   Luis Henriques <lhenriques@...e.com>
To:     "Yan, Zheng" <zyan@...hat.com>, Sage Weil <sage@...hat.com>,
        Ilya Dryomov <idryomov@...il.com>
Cc:     ceph-devel@...r.kernel.org, linux-kernel@...r.kernel.org,
        Luis Henriques <lhenriques@...e.com>
Subject: [PATCH v3 0/2] fix quota subdir mounts

Hi,

As recently reported in the ceph-users mailing-list[1], the kernel client
behaves differently from the fuse client regarding mounting subdirs where
quotas are in effect.  I've also created a bug to track this issue[2].

The following patches are a possible way of fixing this issue.  The
performance impact should be close to zero if the mount is done in the
CephFS root inode.  When we're mounting subdirs, we may have extra
queries to the MDSs, depending on how many extra realms we'll need to
loop through.

Changes since v2:

- Replaced inodes list in mdsc by an rbtree.  This is because we need to
  able to keep track errors in lookupino so that we don't keep sending the
  same useless request for inodes that have failed.  This also resulted in
  reworking the locking in lookup_quotarealm_inode, so that 2 threads can
  lookup the same inode at the same time

- No need to set realm->inode in lookup_quotarealm_inode() as the
  lookupino has set it already.

Changes since v1:

- Loop to free mdsc->quotarealms_inodes_list list was moved further down
  where it's not possible to race with insertions.  This way there's no need
  to hold the spinlock anymore.

- Clarified comments regarding the get_quota_realm function 'retry'
  parameter, both in the function itself and in function
  ceph_quota_is_same_realm, where that param is set to 'false'

- Distinguish between 'realm->inode is NULL' and igrab failures, both in
  get_quota_realm and check_quota_exceeded

Changes since RFC:

The 1st patch hasn't been changed since the initial RFC.  The 2nd patch
has been refactored to include the following changes:

- Zheng Yan's suggestions, i.e, move inode references from the realms to
  ceph_mds_client instance

- It now also handles other cases where an MDS lookup may need to be
  performed:
  * statfs when there are quotas
  * renames, to forbid cross-quota renames

[1] http://lists.ceph.com/pipermail/ceph-users-ceph.com/2019-February/033357.html
[2] https://tracker.ceph.com/issues/38482

Cheers,
--
Luís

Luis Henriques (2):
  ceph: factor out ceph_lookup_inode()
  ceph: quota: fix quota subdir mounts

 fs/ceph/export.c     |  14 +++-
 fs/ceph/mds_client.c |  19 ++++++
 fs/ceph/mds_client.h |  18 +++++
 fs/ceph/quota.c      | 158 ++++++++++++++++++++++++++++++++++++++++---
 fs/ceph/super.h      |   1 +
 5 files changed, 199 insertions(+), 11 deletions(-)