On 2020/2/12 19:17, Ilya Dryomov wrote:
On Wed, Feb 12, 2020 at 9:55 AM <xiubli@xxxxxxxxxx> wrote:
From: Xiubo Li <xiubli@xxxxxxxxxx>
This will enable the remount and reconfiguring superblock params
for the fs. Currently some mount options are not allowed to be
reconfigured.
It will working like:
$ mount.ceph :/ /mnt/cephfs -o remount,mount_timeout=100
URL:https://tracker.ceph.com/issues/44071
Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
---
block/bfq-cgroup.c | 1 +
drivers/block/rbd.c | 2 +-
fs/ceph/caps.c | 2 +
fs/ceph/mds_client.c | 5 +-
fs/ceph/super.c | 126 +++++++++++++++++++++++++++++------
fs/ceph/super.h | 2 +
include/linux/ceph/libceph.h | 4 +-
net/ceph/ceph_common.c | 83 ++++++++++++++++++++---
8 files changed, 192 insertions(+), 33 deletions(-)
diff --git a/block/bfq-cgroup.c b/block/bfq-cgroup.c
index e1419edde2ec..b3d42200182e 100644
--- a/block/bfq-cgroup.c
+++ b/block/bfq-cgroup.c
@@ -12,6 +12,7 @@
#include <linux/ioprio.h>
#include <linux/sbitmap.h>
#include <linux/delay.h>
+#include <linux/rbtree.h>
Hi Xiubo,
This hunk touches the block layer.
Ah, okay, I thought I have removed this.
[...]
+ if (memcmp(&new_opts->fsid, &opts->fsid, sizeof(opts->fsid)))
+ return invalf(fc, "ceph: reconfiguration of fsid not allowed");
+
+ if (strcmp_null(new_opts->name, opts->name))
+ return invalf(fc, "ceph: reconfiguration of name not allowed");
+
+ if (new_opts->key && (!opts->key ||
+ new_opts->key->type != opts->key->type ||
+ new_opts->key->created.tv_sec != opts->key->created.tv_sec ||
+ new_opts->key->created.tv_nsec != opts->key->created.tv_nsec ||
+ new_opts->key->len != opts->key->len ||
+ memcmp(new_opts->key->key, opts->key->key, opts->key->len)))
+ return invalf(fc, "ceph: reconfiguration of secret not allowed");
+
+ opts->osd_keepalive_timeout = new_opts->osd_keepalive_timeout;
+ opts->osd_idle_ttl = new_opts->osd_idle_ttl;
+ opts->mount_timeout = new_opts->mount_timeout;
+ opts->osd_request_timeout = new_opts->osd_request_timeout;
What is the use case for reconfiguring any of these options? They are
all low level and aren't really meant to be reconfigured.
Okay, actually now these are not needed to reconfigured.
+ opts->flags = new_opts->flags;
This is particularly true for flags. Reconfiguring tcp_nodelay will
not have effect on sockets that are already open, reconfiguring noshare
will not unshare the client instance, etc. Messing with crc or
cephx_sign_messages is very likely to hang the client.
Okay, will check this more.
Thanks,
Ilya
