in current implementaion init_net is always used. But in most cases, if a user does a rbd map or ceph mount in a container, it's expected to use the container network namespace.
This patch saves the container's netns in ceph_options on a rbd map or ceph mount. And use the netns other than init_net when creating socket. Ref count of the netns is only taken by the ceph_options in ceph_client since lifetime of osds and mon is within that of ceph_client. I've tested this patch in docker container with below operations: - rbd map - write/read on the rbd - rbd unmap Signed-off-by: Hong Zhiguo <[email protected]> --- fs/ceph/mds_client.c | 3 ++- include/linux/ceph/libceph.h | 3 +++ include/linux/ceph/messenger.h | 4 +++- net/ceph/ceph_common.c | 7 ++++--- net/ceph/messenger.c | 12 +++++++++--- net/ceph/mon_client.c | 2 +- net/ceph/osd_client.c | 3 ++- 7 files changed, 24 insertions(+), 10 deletions(-) diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 8080d48..3fb0976 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -440,7 +440,8 @@ static struct ceph_mds_session *register_session(struct ceph_mds_client *mdsc, s->s_seq = 0; mutex_init(&s->s_mutex); - ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr); + ceph_con_init(&s->s_con, s, &mds_con_ops, &mdsc->fsc->client->msgr, + mdsc->fsc->client->options->netns); spin_lock_init(&s->s_gen_ttl_lock); s->s_cap_gen = 0; diff --git a/include/linux/ceph/libceph.h b/include/linux/ceph/libceph.h index d73a569..442d9f3 100644 --- a/include/linux/ceph/libceph.h +++ b/include/linux/ceph/libceph.h @@ -22,6 +22,8 @@ #include <linux/ceph/osd_client.h> #include <linux/ceph/ceph_fs.h> +struct net; + /* * mount options */ @@ -46,6 +48,7 @@ struct ceph_options { unsigned long mount_timeout; /* jiffies */ unsigned long osd_idle_ttl; /* jiffies */ unsigned long osd_keepalive_timeout; /* jiffies */ + struct net *netns; /* * any type that can't be simply compared or doesn't need need diff --git a/include/linux/ceph/messenger.h b/include/linux/ceph/messenger.h index e154994..3b0a314 100644 --- a/include/linux/ceph/messenger.h +++ b/include/linux/ceph/messenger.h @@ -14,6 +14,7 @@ struct ceph_msg; struct ceph_connection; +struct net; /* * Ceph defines these callbacks for handling connection events. @@ -189,6 +190,7 @@ struct ceph_connection { struct ceph_messenger *msgr; atomic_t sock_state; + struct net *netns; struct socket *sock; struct ceph_entity_addr peer_addr; /* peer address */ struct ceph_entity_addr peer_addr_for_me; @@ -270,7 +272,7 @@ extern void ceph_messenger_init(struct ceph_messenger *msgr, extern void ceph_con_init(struct ceph_connection *con, void *private, const struct ceph_connection_operations *ops, - struct ceph_messenger *msgr); + struct ceph_messenger *msgr, struct net *netns); extern void ceph_con_open(struct ceph_connection *con, __u8 entity_type, __u64 entity_num, struct ceph_entity_addr *addr); diff --git a/net/ceph/ceph_common.c b/net/ceph/ceph_common.c index 925d0c8..1c42d96 100644 --- a/net/ceph/ceph_common.c +++ b/net/ceph/ceph_common.c @@ -269,6 +269,9 @@ static match_table_t opt_tokens = { void ceph_destroy_options(struct ceph_options *opt) { dout("destroy_options %p\n", opt); + if (opt->netns) { + put_net(opt->netns); + } kfree(opt->name); if (opt->key) { ceph_crypto_key_destroy(opt->key); @@ -335,9 +338,6 @@ ceph_parse_options(char *options, const char *dev_name, int err = -ENOMEM; substring_t argstr[MAX_OPT_ARGS]; - if (current->nsproxy->net_ns != &init_net) - return ERR_PTR(-EINVAL); - opt = kzalloc(sizeof(*opt), GFP_KERNEL); if (!opt) return ERR_PTR(-ENOMEM); @@ -501,6 +501,7 @@ ceph_parse_options(char *options, const char *dev_name, } /* success */ + opt->netns = get_net(current->nsproxy->net_ns); return opt; out: diff --git a/net/ceph/messenger.c b/net/ceph/messenger.c index 967080a..0a62905 100644 --- a/net/ceph/messenger.c +++ b/net/ceph/messenger.c @@ -480,8 +480,8 @@ static int ceph_tcp_connect(struct ceph_connection *con) int ret; BUG_ON(con->sock); - ret = sock_create_kern(con->peer_addr.in_addr.ss_family, SOCK_STREAM, - IPPROTO_TCP, &sock); + ret = __sock_create(con->netns, con->peer_addr.in_addr.ss_family, SOCK_STREAM, + IPPROTO_TCP, &sock, 0); if (ret) return ret; sock->sk->sk_allocation = GFP_NOFS; @@ -736,7 +736,7 @@ bool ceph_con_opened(struct ceph_connection *con) */ void ceph_con_init(struct ceph_connection *con, void *private, const struct ceph_connection_operations *ops, - struct ceph_messenger *msgr) + struct ceph_messenger *msgr, struct net *netns) { dout("con_init %p\n", con); memset(con, 0, sizeof(*con)); @@ -744,6 +744,12 @@ void ceph_con_init(struct ceph_connection *con, void *private, con->ops = ops; con->msgr = msgr; + /* + * don't take extra refcnt of netns here since both mon and osds + * have lifetime within that of ceph_client + */ + con->netns = netns; + con_sock_state_init(con); mutex_init(&con->mutex); diff --git a/net/ceph/mon_client.c b/net/ceph/mon_client.c index 9d6ff12..04128af 100644 --- a/net/ceph/mon_client.c +++ b/net/ceph/mon_client.c @@ -832,7 +832,7 @@ int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) goto out_auth_reply; ceph_con_init(&monc->con, monc, &mon_con_ops, - &monc->client->msgr); + &monc->client->msgr, monc->client->options->netns); monc->cur_mon = -1; monc->hunting = true; diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c index 5003367..32d9fa9 100644 --- a/net/ceph/osd_client.c +++ b/net/ceph/osd_client.c @@ -1022,7 +1022,8 @@ static struct ceph_osd *create_osd(struct ceph_osd_client *osdc, int onum) INIT_LIST_HEAD(&osd->o_osd_lru); osd->o_incarnation = 1; - ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr); + ceph_con_init(&osd->o_con, osd, &osd_con_ops, &osdc->client->msgr, + osdc->client->options->netns); INIT_LIST_HEAD(&osd->o_keepalive_item); return osd; -- 1.9.1 -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
