RDMA/erdma: Fix exception handling in erdma_accept_newconn()
Subsystems (3)
PATCH v2 1/1 RDMA/erdma: Fix exception handling in erdma_accept_newconn()
From: Markus Elfring <[email protected]>
Date: Thu, 13 Mar 2025 11:44:50 +0100
The label “error” was used to jump to another pointer check despite of the detail in the implementation of the function “erdma_accept_newconn” that it was determined already that corresponding variables contained still null pointers.
1. Thus return directly if
* the cep state is not the value “ERDMA_EPSTATE_LISTENING” or
* a call of the function “erdma_cep_alloc” failed.
2. Use more appropriate labels instead.
3. Delete two questionable checks.
4. Omit extra initialisations (for the variables “new_cep”, “new_s” and “ret”) which became unnecessary with this refactoring.
This issue was detected by using the Coccinelle software.
---
See also:
* https://lore.kernel.org/cocci/[email protected]/
* https://lkml.org/lkml/2023/3/19/191
V2:
The change suggestion was rebased on source files of the software “Linux next-20250313”. A corresponding implementation detail was improved by the commit 83437689249e6a17b25e27712fbee292e42e7855 ("RDMA/erdma: Prevent use-after-free in erdma_accept_newconn()") on 2025-03-06.
drivers/infiniband/hw/erdma/erdma_cm.c | 37 +++++++++++--------------- 1 file changed, 16 insertions(+), 21 deletions(-)
drivers/infiniband/hw/erdma/erdma_cm.c +16 -21
@@ -642,16 +642,16 @@ static int erdma_proc_mpareply(struct erdma_cep *cep)static void erdma_accept_newconn (struct erdma_cep *cep) {struct socket *s = cep->sock; -struct socket *new_s =NULL ; -struct erdma_cep *new_cep =NULL ; -int ret =0 ; +struct socket *new_s; +struct erdma_cep *new_cep; +int ret;if (cep->state !=ERDMA_EPSTATE_LISTENING ) - goto error; +return ; new_cep =erdma_cep_alloc (cep->dev);if (!new_cep) - goto error; +return ;/* * 4: Allocate a sufficient number of work elements
@@ -659,7 +659,7 @@ static void erdma_accept_newconn(struct erdma_cep *cep)* events, MPA header processing + MPA timeout. */ if (erdma_cm_alloc_work (new_cep,4 ) !=0 ) - goto error; + goto put_cep;/* * Copy saved socket callbacks from listening CEP
@@ -671,7 +671,7 @@ static void erdma_accept_newconn(struct erdma_cep *cep) ret =kernel_accept (s, &new_s,O_NONBLOCK );if (ret !=0 ) - goto error; + goto put_cep; new_cep->sock = new_s;erdma_cep_get (new_cep);
@@ -682,7 +682,7 @@ static void erdma_accept_newconn(struct erdma_cep *cep) ret =erdma_cm_queue_work (new_cep,ERDMA_CM_WORK_MPATIMEOUT );if (ret) - goto error; + goto disassoc_socket; new_cep->listen_cep = cep;erdma_cep_get (cep);
@@ -696,25 +696,20 @@ static void erdma_accept_newconn(struct erdma_cep *cep) new_cep->listen_cep =NULL ;if (ret) {erdma_cep_set_free (new_cep); - goto error; + goto disassoc_socket; } }erdma_cep_set_free (new_cep); }return ; -error: -if (new_cep) { - new_cep->state =ERDMA_EPSTATE_CLOSED ; -erdma_cancel_mpatimer (new_cep); - -erdma_cep_put (new_cep); - } - -if (new_s) { -erdma_socket_disassoc (new_s); -sock_release (new_s); - } +disassoc_socket: +erdma_socket_disassoc (new_s); +sock_release (new_s); + new_cep->state =ERDMA_EPSTATE_CLOSED ; +erdma_cancel_mpatimer (new_cep); +put_cep: +erdma_cep_put (new_cep); }static int erdma_newconn_connected (struct erdma_cep *cep)
--
2.48.1
Cheng Xu
maintainerQuoted 17 lines · Markus Elfring
From: Markus Elfring <[email protected]>
Date: Thu, 13 Mar 2025 11:44:50 +0100The label “error” was used to jump to another pointer check despite of the detail in the implementation of the function “erdma_accept_newconn” that it was determined already that corresponding variables contained still null pointers.
1. Thus return directly if
* the cep state is not the value “ERDMA_EPSTATE_LISTENING” or
* a call of the function “erdma_cep_alloc” failed.2. Use more appropriate labels instead.
3. Delete two questionable checks.
Quoted 7 lines
4. Omit extra initialisations (for the variables “new_cep”, “new_s” and “ret”) which became unnecessary with this refactoring.
This issue was detected by using the Coccinelle software.
Fixes 920d93eac8b9 ("RDMA/erdma: Add connection management (CM) support")
I think this patch does not fix issues, so fix line is not needed.
Thanks,
Cheng Xu
Quoted 3 lines
Signed-off-by Markus Elfring <[email protected]>---
<...>