RDMA/rxe: Use validated num_sge in local buffer
Subsystems (2)
PATCH 1/1 RDMA/rxe: Use validated num_sge in local buffer
For both SRQ and non-SRQ receive paths, the WQE is copied into a local buffer to provide a kernel-owned, validated copy. While calculating the memcpy size from the validated num_sge prevents overflow during the copy, memcpy() itself still copies num_sge from shared memory.
A concurrent userspace modification before or during memcpy() leaves an unvalidated num_sge in the local buffer, leading to potential out-of-bounds reads in rxe_resp_check_length() and copy_data().
Explicitly assign the validated num_sge to the local buffer after the copy to prevent this race.
--- drivers/infiniband/sw/rxe/rxe_resp.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/infiniband/sw/rxe/rxe_resp.c +2 -0
@@ -288,6 +288,7 @@ static enum resp_states get_srq_wqe(struct rxe_qp *qp) } size =sizeof (*wqe) + num_sge *sizeof (struct rxe_sge );memcpy (&qp->resp.srq_wqe, wqe, size); + qp->resp.srq_wqe.wqe.dma.num_sge = num_sge; qp->resp.wqe = &qp->resp.srq_wqe.wqe;queue_advance_consumer (q,QUEUE_TYPE_FROM_CLIENT );
@@ -328,6 +329,7 @@ static enum resp_states rxe_get_recv_wqe(struct rxe_qp *qp) } size =sizeof (*wqe) + num_sge *sizeof (struct rxe_sge );memcpy (&qp->resp.srq_wqe, wqe, size); + qp->resp.srq_wqe.wqe.dma.num_sge = num_sge; qp->resp.wqe = &qp->resp.srq_wqe.wqe;return RESPST_CHK_LENGTH ;
Signature · 1 line
2.54.0
Quoted 9 lines · Nicolas Morey
For both SRQ and non-SRQ receive paths, the WQE is copied into a local buffer to provide a kernel-owned, validated copy. While calculating the memcpy size from the validated
num_sgeprevents overflow during the copy,memcpy()itself still copiesnum_sgefrom shared memory.A concurrent userspace modification before or during
memcpy()leaves an unvalidatednum_sgein the local buffer, leading to potential out-of-bounds reads inrxe_resp_check_length()andcopy_data().
Hi Nicolas,
Thanks for the patch. The logic makes total sense to prevent the TOCTOU race condition after memcpy.
Just out of curiosity, do you happen to have a reproducer or a POC script that demonstrates this race in practice?
It would be great to know if this can be reliably reproduced or integrated into testing setups (like rdma-core tests, or tools/testing/selftests/rdma) to catch similar double-read issues in the future.
Thanks a lot for your effort.
Zhu Yanjun
Quoted 30 lines
Explicitly assign the validated
num_sgeto the local buffer after the copy to prevent this race.Fixes 22b8fbded65b ("RDMA/rxe: Fix TOCTOU heap overflow in get_srq_wqe")Fixes d6ab440240a0 ("RDMA/rxe: Copy WQE to local buffer in non-SRQ receive path")Signed-off-by Nicolas Morey <[email protected]>--- drivers/infiniband/sw/rxe/rxe_resp.c | 2 ++ 1 file changed, 2 insertions(+)drivers/infiniband/sw/rxe/rxe_resp.c +2
-0@@ -288,6 +288,7 @@ static enum resp_states get_srq_wqe(struct rxe_qp *qp) } size =sizeof (*wqe) + num_sge *sizeof (struct rxe_sge );memcpy (&qp->resp.srq_wqe, wqe, size); + qp->resp.srq_wqe.wqe.dma.num_sge = num_sge; qp->resp.wqe = &qp->resp.srq_wqe.wqe;queue_advance_consumer (q,QUEUE_TYPE_FROM_CLIENT );@@ -328,6 +329,7 @@ static enum resp_states rxe_get_recv_wqe(struct rxe_qp *qp) } size =sizeof (*wqe) + num_sge *sizeof (struct rxe_sge );memcpy (&qp->resp.srq_wqe, wqe, size); + qp->resp.srq_wqe.wqe.dma.num_sge = num_sge; qp->resp.wqe = &qp->resp.srq_wqe.wqe;return RESPST_CHK_LENGTH ;
Quoted 11 lines · Zhu Yanjun
Quoted 9 lines · Nicolas Morey
For both SRQ and non-SRQ receive paths, the WQE is copied into a local buffer to provide a kernel-owned, validated copy. While calculating the memcpy size from the validated
num_sgeprevents overflow during the copy,memcpy()itself still copiesnum_sgefrom shared memory.A concurrent userspace modification before or during
memcpy()leaves an unvalidatednum_sgein the local buffer, leading to potential out-of-bounds reads inrxe_resp_check_length()andcopy_data().
Quoted 8 lines
Hi Nicolas,
Thanks for the patch. The logic makes total sense to prevent the TOCTOU race condition after memcpy.
Just out of curiosity, do you happen to have a reproducer or a POC script that demonstrates this race in practice?
It would be great to know if this can be reliably reproduced or integrated into testing setups (like rdma-core tests, or
tools/testing/selftests/rdma) to catch similar double-read issues in the future.
No reproducer or PoC sadly. I haven't tried to make one though.
This got caught by one of our AI tools when checking the backport of CVE-2026-74377.
I assumed from the 2 existing fixes (ending up as CVEs) that this could happened.
Nicolas
Quoted 16 lines · Nicolas Morey
Quoted 15 lines · Zhu Yanjun
Quoted 9 lines · Nicolas Morey
For both SRQ and non-SRQ receive paths, the WQE is copied into a local buffer to provide a kernel-owned, validated copy. While calculating the memcpy size from the validated
num_sgeprevents overflow during the copy,memcpy()itself still copiesnum_sgefrom shared memory.A concurrent userspace modification before or during
memcpy()leaves an unvalidatednum_sgein the local buffer, leading to potential out-of-bounds reads inrxe_resp_check_length()andcopy_data().Hi Nicolas,
Thanks for the patch. The logic makes total sense to prevent the TOCTOU race condition after memcpy.
Quoted 7 lines
Quoted 4 lines
Just out of curiosity, do you happen to have a reproducer or a POC script that demonstrates this race in practice?
It would be great to know if this can be reliably reproduced or integrated into testing setups (like rdma-core tests, or
tools/testing/selftests/rdma) to catch similar double-read issues in the future.No reproducer or PoC sadly. I haven't tried to make one though.
This got caught by one of our AI tools when checking the backport of CVE-2026-74377.
Hi, Nicolas
Thanks for sharing this. Could you clarify whether your AI tool detected this purely through static analysis (and provided a suggested fix), or if it actually triggered a runtime bug with a backtrace/calltrace?
If you have a calltrace, please share it—that would help a lot in understanding the issue. Also, if this was flagged and fixed by a specific AI tool, adding an Assisted-by: tag (or referencing the tool in the commit description) would be appropriate.
Thanks,
Zhu Yanjun
Quoted 4 lines
I assumed from the 2 existing fixes (ending up as CVEs) that this could happened.
Nicolas
Quoted 27 lines · Zhu Yanjun
Quoted 16 lines · Nicolas Morey
Quoted 15 lines · Zhu Yanjun
Quoted 9 lines · Nicolas Morey
For both SRQ and non-SRQ receive paths, the WQE is copied into a local buffer to provide a kernel-owned, validated copy. While calculating the memcpy size from the validated
num_sgeprevents overflow during the copy,memcpy()itself still copiesnum_sgefrom shared memory.A concurrent userspace modification before or during
memcpy()leaves an unvalidatednum_sgein the local buffer, leading to potential out-of-bounds reads inrxe_resp_check_length()andcopy_data().Hi Nicolas,
Thanks for the patch. The logic makes total sense to prevent the TOCTOU race condition after memcpy.
Quoted 7 lines
Quoted 4 lines
Just out of curiosity, do you happen to have a reproducer or a POC script that demonstrates this race in practice?
It would be great to know if this can be reliably reproduced or integrated into testing setups (like rdma-core tests, or
tools/testing/selftests/rdma) to catch similar double-read issues in the future.No reproducer or PoC sadly. I haven't tried to make one though.
This got caught by one of our AI tools when checking the backport of CVE-2026-74377.Hi, Nicolas
Quoted 8 lines
Thanks for sharing this. Could you clarify whether your AI tool detected this purely through static analysis (and provided a suggested fix), or if it actually triggered a runtime bug with a backtrace/calltrace?
If you have a calltrace, please share it—that would help a lot in understanding the issue. Also, if this was flagged and fixed by a specific AI tool, adding an Assisted-by: tag (or referencing the tool in the commit description) would be appropriate.
Thanks,
Zhu Yanjun
Hi Zhu,
The tool is a LLM static analyser that simply detected that d6ab440240a0 ("RDMA/rxe: Copy WQE to local buffer in non-SRQ receive path") was a valid patch but did not fix the whole issue. It did not suggest any breaker PoC nor any fix. All the rest is human (me) made.
Does that warrant an Assisted-by tag ?
It seems sashiko picked the same issue for 22b8fbded65b ("RDMA/rxe: Fix TOCTOU heap overflow in get_srq_wqe")
https://sashiko.dev/#/patchset/20260518215040.1598586-1-tristan%40talencesecurity.com
Quoted 3 lines
If a concurrent userspace thread modifies
wqe->dma.num_sgeafter the bounds check but before the memcpy completes, doesn't this copy the unvalidated value into the kernel heap?
By tweaking the reproducer from Tristan, I have been able to reproduce it once out of sheer luck I think:
==================================================================
BUG: KASAN: slab-out-of-bounds in rxe_receiver+0x8109/0x9ec0 [rdma_rxe] Read of size 4 at addr ffff88812c4867f8 by task kworker/u9:6/361 CPU: 0 UID: 0 PID: 361 Comm: kworker/u9:6 Tainted: G E 7.3.0-rc2-00006-g28924df2a08f #3 PREEMPT(full) 318a88bba3045f81bad31a7694727561b4bd4965 Tainted: [E]=UNSIGNED_MODULE Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.3-2-gc13ff2cd-prebuilt.qemu.org 04/01/2014 Workqueue: rxe_wq do_work [rdma_rxe] Call Trace: <TASK> dump_stack_lvl+0x4b/0x70 print_report+0x153/0x4b5 ? __pfx__raw_spin_lock_irqsave+0x10/0x10 ? stack_trace_save+0x93/0xd0 kasan_report+0xbc/0xf0 ? rxe_receiver+0x8109/0x9ec0 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] ? rxe_receiver+0x8109/0x9ec0 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] rxe_receiver+0x8109/0x9ec0 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] ? pick_task_fair+0x12e/0x1b50 ? __pfx_rxe_receiver+0x10/0x10 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] ? __pfx__raw_spin_lock_irqsave+0x10/0x10 ? hrtimer_start_range_ns+0xe7/0x320 ? ktime_get+0xe3/0x170 ? _raw_spin_unlock+0xe/0x30 ? _raw_spin_lock_irqsave+0x8a/0xf0 ? __pfx__raw_spin_lock_irqsave+0x10/0x10 ? __pfx_rxe_receiver+0x10/0x10 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] do_work+0x149/0x610 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] process_one_work+0x726/0x10a0 ? __pfx___schedule+0x10/0x10 ? __pfx_process_one_work+0x10/0x10 ? _raw_spin_lock_irq+0x85/0xe0 ? __pfx__raw_spin_lock_irq+0x10/0x10 worker_thread+0x500/0xd70 ? __kthread_parkme+0x8d/0x170 ? __pfx_worker_thread+0x10/0x10 ? __pfx_worker_thread+0x10/0x10 kthread+0x329/0x410 ? recalc_sigpending+0x15c/0x200 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x4cf/0x760 ? __pfx_ret_from_fork+0x10/0x10 ? __switch_to+0x575/0x11c0 ? __pfx_kthread+0x10/0x10 ret_from_fork_asm+0x1a/0x30 </TASK> Allocated by task 1698: kasan_save_stack+0x20/0x40 kasan_save_track+0x14/0x30 __kasan_kmalloc+0x9a/0xb0 __kmalloc_noprof+0x209/0x560 create_qp.part.0+0x760/0x9d0 [ib_core] ib_create_qp_user+0xa2/0x500 [ib_core] ib_uverbs_handler_UVERBS_METHOD_QP_CREATE+0x99b/0x12af [ib_uverbs] ib_uverbs_cmd_verbs+0x214b/0x32d0 [ib_uverbs] ib_uverbs_ioctl+0x131/0x200 [ib_uverbs] __x64_sys_ioctl+0x13c/0x1c0 do_syscall_64+0xba/0x530 entry_SYSCALL_64_after_hwframe+0x76/0x7e Last potentially related work creation: kasan_save_stack+0x20/0x40 kasan_record_aux_stack+0xb0/0xc0 __queue_work+0x8c5/0x11f0 queue_work_on+0x60/0x70 rxe_sched_task+0x1d2/0x250 [rdma_rxe] rxe_rcv+0x847/0x1830 [rdma_rxe] rxe_xmit_packet+0x408/0x950 [rdma_rxe] rxe_requester+0x1878/0x5460 [rdma_rxe] rxe_sender+0x17/0x40 [rdma_rxe] do_work+0x149/0x610 [rdma_rxe] process_one_work+0x726/0x10a0 worker_thread+0x500/0xd70 kthread+0x329/0x410 ret_from_fork+0x4cf/0x760 ret_from_fork_asm+0x1a/0x30 The buggy address belongs to the object at ffff88812c486000 which belongs to the cache kmalloc-part-13-2k of size 2048
The buggy address is located 0 bytes to the right of
allocated 2040-byte region [ffff88812c486000, ffff88812c4867f8)
The buggy address belongs to the physical page: page: refcount:0 mapcount:0 mapping:0000000000000000 index:0xffff88812c480000 pfn:0x12c480 head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 flags: 0x17ffffc0000240(workingset|head|node=0|zone=2|lastcpupid=0x1fffff) page_type: f5(slab) raw: 0017ffffc0000240 ffff888100059140 ffffea0004499010 ffffea00049a7a10 raw: ffff88812c480000 0000000200080007 00000000f5000000 0000000000000000
head: 0017ffffc0000240 ffff888100059140 ffffea0004499010 ffffea00049a7a10
head: ffff88812c480000 0000000200080007 00000000f5000000 0000000000000000
head: 0017ffffc0000003 fffffffffffffe01 00000000ffffffff 00000000ffffffff
head: ffffffffffffffff 0000000000000000 00000000ffffffff 0000000000000008
page dumped because: kasan: bad access detected
Memory state around the buggy address:
ffff88812c486680: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
ffff88812c486700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
>ffff88812c486780: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fc
^
ffff88812c486800: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
ffff88812c486880: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
==================================================================Zhu Yanjun
Reviewed-byQuoted 45 lines · Nicolas Morey
Quoted 27 lines · Zhu Yanjun
Quoted 16 lines · Nicolas Morey
Quoted 15 lines · Zhu Yanjun
Quoted 9 lines · Nicolas Morey
For both SRQ and non-SRQ receive paths, the WQE is copied into a local buffer to provide a kernel-owned, validated copy. While calculating the memcpy size from the validated
num_sgeprevents overflow during the copy,memcpy()itself still copiesnum_sgefrom shared memory.A concurrent userspace modification before or during
memcpy()leaves an unvalidatednum_sgein the local buffer, leading to potential out-of-bounds reads inrxe_resp_check_length()andcopy_data().Hi Nicolas,
Thanks for the patch. The logic makes total sense to prevent the TOCTOU race condition after memcpy.
Quoted 7 lines
Quoted 4 lines
Just out of curiosity, do you happen to have a reproducer or a POC script that demonstrates this race in practice?
It would be great to know if this can be reliably reproduced or integrated into testing setups (like rdma-core tests, or
tools/testing/selftests/rdma) to catch similar double-read issues in the future.No reproducer or PoC sadly. I haven't tried to make one though.
This got caught by one of our AI tools when checking the backport of CVE-2026-74377.Hi, Nicolas
Quoted 8 lines
Thanks for sharing this. Could you clarify whether your AI tool detected this purely through static analysis (and provided a suggested fix), or if it actually triggered a runtime bug with a backtrace/calltrace?
If you have a calltrace, please share it—that would help a lot in understanding the issue. Also, if this was flagged and fixed by a specific AI tool, adding an Assisted-by: tag (or referencing the tool in the commit description) would be appropriate.
Thanks,
Zhu Yanjun
Hi Zhu,
The tool is a LLM static analyser that simply detected that
d6ab440240a0("RDMA/rxe: Copy WQE to local buffer in non-SRQ receive path") was a valid patch but did not fix the whole issue. It did not suggest any breaker PoC nor any fix. All the rest is human (me) made.
Does that warrant an Assisted-by tag ?
It seems sashiko picked the same issue for22b8fbded65b("RDMA/rxe: Fix TOCTOU heap overflow inget_srq_wqe")
https://sashiko.dev/#/patchset/20260518215040.1598586-1-tristan%40talencesecurity.com
Quoted 5 lines
Quoted 3 lines
If a concurrent userspace thread modifies
wqe->dma.num_sgeafter the bounds check but before the memcpy completes, doesn't this copy the unvalidated value into the kernel heap?By tweaking the reproducer from Tristan, I have been able to reproduce it once out of sheer luck I think:
Thanks a lot. Please add the following call trace into the commit log and send out a new commit.
Except the above, I am fine with this commit.
Zhu Yanjun
Quoted 104 lines
==================================================================
BUG: KASAN: slab-out-of-bounds in rxe_receiver+0x8109/0x9ec0 [rdma_rxe] Read of size 4 at addr ffff88812c4867f8 by task kworker/u9:6/361 CPU: 0 UID: 0 PID: 361 Comm: kworker/u9:6 Tainted: G E 7.3.0-rc2-00006-g28924df2a08f #3 PREEMPT(full) 318a88bba3045f81bad31a7694727561b4bd4965 Tainted: [E]=UNSIGNED_MODULE Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.16.3-2-gc13ff2cd-prebuilt.qemu.org 04/01/2014 Workqueue: rxe_wq do_work [rdma_rxe] Call Trace: <TASK> dump_stack_lvl+0x4b/0x70 print_report+0x153/0x4b5 ? __pfx__raw_spin_lock_irqsave+0x10/0x10 ? stack_trace_save+0x93/0xd0 kasan_report+0xbc/0xf0 ? rxe_receiver+0x8109/0x9ec0 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] ? rxe_receiver+0x8109/0x9ec0 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] rxe_receiver+0x8109/0x9ec0 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] ? pick_task_fair+0x12e/0x1b50 ? __pfx_rxe_receiver+0x10/0x10 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] ? __pfx__raw_spin_lock_irqsave+0x10/0x10 ? hrtimer_start_range_ns+0xe7/0x320 ? ktime_get+0xe3/0x170 ? _raw_spin_unlock+0xe/0x30 ? _raw_spin_lock_irqsave+0x8a/0xf0 ? __pfx__raw_spin_lock_irqsave+0x10/0x10 ? __pfx_rxe_receiver+0x10/0x10 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] do_work+0x149/0x610 [rdma_rxe b8d153a5fa9911e345612beff278ac3241bac327] process_one_work+0x726/0x10a0 ? __pfx___schedule+0x10/0x10 ? __pfx_process_one_work+0x10/0x10 ? _raw_spin_lock_irq+0x85/0xe0 ? __pfx__raw_spin_lock_irq+0x10/0x10 worker_thread+0x500/0xd70 ? __kthread_parkme+0x8d/0x170 ? __pfx_worker_thread+0x10/0x10 ? __pfx_worker_thread+0x10/0x10 kthread+0x329/0x410 ? recalc_sigpending+0x15c/0x200 ? __pfx_kthread+0x10/0x10 ret_from_fork+0x4cf/0x760 ? __pfx_ret_from_fork+0x10/0x10 ? __switch_to+0x575/0x11c0 ? __pfx_kthread+0x10/0x10 ret_from_fork_asm+0x1a/0x30 </TASK> Allocated by task 1698: kasan_save_stack+0x20/0x40 kasan_save_track+0x14/0x30 __kasan_kmalloc+0x9a/0xb0 __kmalloc_noprof+0x209/0x560 create_qp.part.0+0x760/0x9d0 [ib_core] ib_create_qp_user+0xa2/0x500 [ib_core] ib_uverbs_handler_UVERBS_METHOD_QP_CREATE+0x99b/0x12af [ib_uverbs] ib_uverbs_cmd_verbs+0x214b/0x32d0 [ib_uverbs] ib_uverbs_ioctl+0x131/0x200 [ib_uverbs] __x64_sys_ioctl+0x13c/0x1c0 do_syscall_64+0xba/0x530 entry_SYSCALL_64_after_hwframe+0x76/0x7e Last potentially related work creation: kasan_save_stack+0x20/0x40 kasan_record_aux_stack+0xb0/0xc0 __queue_work+0x8c5/0x11f0 queue_work_on+0x60/0x70 rxe_sched_task+0x1d2/0x250 [rdma_rxe] rxe_rcv+0x847/0x1830 [rdma_rxe] rxe_xmit_packet+0x408/0x950 [rdma_rxe] rxe_requester+0x1878/0x5460 [rdma_rxe] rxe_sender+0x17/0x40 [rdma_rxe] do_work+0x149/0x610 [rdma_rxe] process_one_work+0x726/0x10a0 worker_thread+0x500/0xd70 kthread+0x329/0x410 ret_from_fork+0x4cf/0x760 ret_from_fork_asm+0x1a/0x30 The buggy address belongs to the object at ffff88812c486000 which belongs to the cache kmalloc-part-13-2k of size 2048The buggy address is located 0 bytes to the right of
allocated 2040-byte region [ffff88812c486000,ffff88812c4867f8)The buggy address belongs to the physical page: page: refcount:0 mapcount:0 mapping:0000000000000000 index:0xffff88812c480000 pfn:0x12c480 head: order:3 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0 flags: 0x17ffffc0000240(workingset|head|node=0|zone=2|lastcpupid=0x1fffff) page_type: f5(slab) raw: 0017ffffc0000240 ffff888100059140 ffffea0004499010 ffffea00049a7a10 raw: ffff88812c480000 0000000200080007 00000000f5000000 0000000000000000head:
0017ffffc0000240ffff888100059140ffffea0004499010ffffea00049a7a10
head:ffff88812c480000000000020008000700000000f50000000000000000000000
head:0017ffffc0000003fffffffffffffe0100000000ffffffff00000000ffffffff
head: ffffffffffffffff 000000000000000000000000ffffffff0000000000000008page dumped because: kasan: bad access detected Memory state around the buggy address: ffff88812c486680: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ffff88812c486700: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00Quoted 1 line
ffff88812c486780: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fc^
ffff88812c486800: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
ffff88812c486880: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
==================================================================