net/smc/smc_core.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
smc_switch_conns() takes a reference on the socket with sock_hold() before
dropping conns_lock to prefetch a tx slot. On success the reference is
released by the sock_put() after smc_switch_cursor(). When
smc_cdc_get_free_slot() fails, however, the function jumps straight to
err_out without a sock_put(), leaking the reference and leaving the socket
unreclaimable. During SMC-R link failover with an exhausted backup link
this leaks one reference per failing connection switch.
Release the reference before jumping to err_out. The err_out label is also
reached from the smc_switch_cursor() failure path, which has already done
its own sock_put(), so the put must stay at the get_free_slot failure site
rather than at the label.
Fixes: c6f02ebeea3a ("net/smc: switch connections to alternate link")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Yifei Gao <gyf161023@gmail.com>
---
net/smc/smc_core.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/net/smc/smc_core.c b/net/smc/smc_core.c
index cf6b620fef05..04c6ce608ec7 100644
--- a/net/smc/smc_core.c
+++ b/net/smc/smc_core.c
@@ -1148,8 +1148,10 @@ struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
read_unlock_bh(&lgr->conns_lock);
/* pre-fetch buffer outside of send_lock, might sleep */
rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
- if (rc)
+ if (rc) {
+ sock_put(&smc->sk);
goto err_out;
+ }
/* avoid race with smcr_tx_sndbuf_nonempty() */
spin_lock_bh(&conn->send_lock);
smc_switch_link_and_count(conn, to_lnk);
--
2.43.0
Hi Yifei,
Thanks for looking into this.
Yes I have submitted a patch for the same issue.
https://lore.kernel.org/netdev/20260804082800.498672-1-hidayath@linux.ibm.com/
Hidayath
On 05/08/26 3:09 am, Yifei Gao wrote:
> smc_switch_conns() takes a reference on the socket with sock_hold() before
> dropping conns_lock to prefetch a tx slot. On success the reference is
> released by the sock_put() after smc_switch_cursor(). When
> smc_cdc_get_free_slot() fails, however, the function jumps straight to
> err_out without a sock_put(), leaking the reference and leaving the socket
> unreclaimable. During SMC-R link failover with an exhausted backup link
> this leaks one reference per failing connection switch.
>
> Release the reference before jumping to err_out. The err_out label is also
> reached from the smc_switch_cursor() failure path, which has already done
> its own sock_put(), so the put must stay at the get_free_slot failure site
> rather than at the label.
>
> Fixes: c6f02ebeea3a ("net/smc: switch connections to alternate link")
> Cc: stable@vger.kernel.org
> Assisted-by: Claude:claude-opus-4-8
> Signed-off-by: Yifei Gao <gyf161023@gmail.com>
> ---
> net/smc/smc_core.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/net/smc/smc_core.c b/net/smc/smc_core.c
> index cf6b620fef05..04c6ce608ec7 100644
> --- a/net/smc/smc_core.c
> +++ b/net/smc/smc_core.c
> @@ -1148,8 +1148,10 @@ struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
> read_unlock_bh(&lgr->conns_lock);
> /* pre-fetch buffer outside of send_lock, might sleep */
> rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
> - if (rc)
> + if (rc) {
> + sock_put(&smc->sk);
> goto err_out;
> + }
> /* avoid race with smcr_tx_sndbuf_nonempty() */
> spin_lock_bh(&conn->send_lock);
> smc_switch_link_and_count(conn, to_lnk);
Hi Hidayath,
Thanks for the fix — I see yours covers the same issue and was posted
first. Please disregard my duplicate.
Hidayathulla Khan I <hidayath@linux.ibm.com> 于2026年8月5日周三 02:26写道:
>
> Hi Yifei,
>
> Thanks for looking into this.
>
> Yes I have submitted a patch for the same issue.
> https://lore.kernel.org/netdev/20260804082800.498672-1-hidayath@linux.ibm.com/
>
> Hidayath
>
> On 05/08/26 3:09 am, Yifei Gao wrote:
> > smc_switch_conns() takes a reference on the socket with sock_hold() before
> > dropping conns_lock to prefetch a tx slot. On success the reference is
> > released by the sock_put() after smc_switch_cursor(). When
> > smc_cdc_get_free_slot() fails, however, the function jumps straight to
> > err_out without a sock_put(), leaking the reference and leaving the socket
> > unreclaimable. During SMC-R link failover with an exhausted backup link
> > this leaks one reference per failing connection switch.
> >
> > Release the reference before jumping to err_out. The err_out label is also
> > reached from the smc_switch_cursor() failure path, which has already done
> > its own sock_put(), so the put must stay at the get_free_slot failure site
> > rather than at the label.
> >
> > Fixes: c6f02ebeea3a ("net/smc: switch connections to alternate link")
> > Cc: stable@vger.kernel.org
> > Assisted-by: Claude:claude-opus-4-8
> > Signed-off-by: Yifei Gao <gyf161023@gmail.com>
> > ---
> > net/smc/smc_core.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/net/smc/smc_core.c b/net/smc/smc_core.c
> > index cf6b620fef05..04c6ce608ec7 100644
> > --- a/net/smc/smc_core.c
> > +++ b/net/smc/smc_core.c
> > @@ -1148,8 +1148,10 @@ struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
> > read_unlock_bh(&lgr->conns_lock);
> > /* pre-fetch buffer outside of send_lock, might sleep */
> > rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
> > - if (rc)
> > + if (rc) {
> > + sock_put(&smc->sk);
> > goto err_out;
> > + }
> > /* avoid race with smcr_tx_sndbuf_nonempty() */
> > spin_lock_bh(&conn->send_lock);
> > smc_switch_link_and_count(conn, to_lnk);
On 05/08/26 3:09 am, Yifei Gao wrote:
> smc_switch_conns() takes a reference on the socket with sock_hold() before
> dropping conns_lock to prefetch a tx slot. On success the reference is
> released by the sock_put() after smc_switch_cursor(). When
> smc_cdc_get_free_slot() fails, however, the function jumps straight to
> err_out without a sock_put(), leaking the reference and leaving the socket
> unreclaimable. During SMC-R link failover with an exhausted backup link
> this leaks one reference per failing connection switch.
>
> Release the reference before jumping to err_out. The err_out label is also
> reached from the smc_switch_cursor() failure path, which has already done
> its own sock_put(), so the put must stay at the get_free_slot failure site
> rather than at the label.
>
> Fixes: c6f02ebeea3a ("net/smc: switch connections to alternate link")
> Cc: stable@vger.kernel.org
> Assisted-by: Claude:claude-opus-4-8
> Signed-off-by: Yifei Gao <gyf161023@gmail.com>
> ---
> net/smc/smc_core.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/net/smc/smc_core.c b/net/smc/smc_core.c
> index cf6b620fef05..04c6ce608ec7 100644
> --- a/net/smc/smc_core.c
> +++ b/net/smc/smc_core.c
> @@ -1148,8 +1148,10 @@ struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
> read_unlock_bh(&lgr->conns_lock);
> /* pre-fetch buffer outside of send_lock, might sleep */
> rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
> - if (rc)
> + if (rc) {
> + sock_put(&smc->sk);
> goto err_out;
> + }
This patch has already been posted by Hidayath:
https://lore.kernel.org/netdev/20260804082800.498672-1-hidayath@linux.ibm.com/
Adding Hidayath to confirm, but this appears to be a duplicate submission.
> Subject: [PATCH] net/smc: fix sock refcount leak in smc_switch_conns() error path
Please use [PATCH net] as the subject prefix when sending v2, since
this fix targets the net tree.
On Tue, Aug 04, 2026 at 09:39:17PM +0000, Yifei Gao wrote:
> smc_switch_conns() takes a reference on the socket with sock_hold() before
> dropping conns_lock to prefetch a tx slot. On success the reference is
> released by the sock_put() after smc_switch_cursor(). When
> smc_cdc_get_free_slot() fails, however, the function jumps straight to
> err_out without a sock_put(), leaking the reference and leaving the socket
> unreclaimable. During SMC-R link failover with an exhausted backup link
> this leaks one reference per failing connection switch.
>
> Release the reference before jumping to err_out. The err_out label is also
> reached from the smc_switch_cursor() failure path, which has already done
> its own sock_put(), so the put must stay at the get_free_slot failure site
> rather than at the label.
>
> Fixes: c6f02ebeea3a ("net/smc: switch connections to alternate link")
> Cc: stable@vger.kernel.org
> Assisted-by: Claude:claude-opus-4-8
> Signed-off-by: Yifei Gao <gyf161023@gmail.com>
> ---
> net/smc/smc_core.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/net/smc/smc_core.c b/net/smc/smc_core.c
> index cf6b620fef05..04c6ce608ec7 100644
> --- a/net/smc/smc_core.c
> +++ b/net/smc/smc_core.c
> @@ -1148,8 +1148,10 @@ struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
> read_unlock_bh(&lgr->conns_lock);
> /* pre-fetch buffer outside of send_lock, might sleep */
> rc = smc_cdc_get_free_slot(conn, to_lnk, &wr_buf, NULL, &pend);
> - if (rc)
> + if (rc) {
> + sock_put(&smc->sk);
> goto err_out;
> + }
> /* avoid race with smcr_tx_sndbuf_nonempty() */
> spin_lock_bh(&conn->send_lock);
> smc_switch_link_and_count(conn, to_lnk);
> --
> 2.43.0
© 2016 - 2026 Red Hat, Inc.