Re: [PATCH v4] RDMA/srpt: Reject an srp_sq_size that does not fit in a CQ

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] ib_srpt: UAF in configfs store due to hot-unplug race
--

commit 1f57db75193afbb14a2edb4aa5126369171bf0a9
Author: Serhat Kumral <serhatkumral1@gmail.com>

RDMA/srpt: Reject an srp_sq_size that does not fit in a CQ

This commit modifies the configfs store function for srp_sq_size to validate
the requested size against the device's max_cqe limit, preventing continuous
CQ allocation failures that can lead to memory exhaustion.

> diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
> --- a/drivers/infiniband/ulp/srpt/ib_srpt.c
> +++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
> @@ -3603,6 +3603,8 @@ static ssize_t srpt_tpg_attrib_srp_sq_size_store(struct config_item *item,
>  {
>  	struct se_portal_group *se_tpg = attrib_to_tpg(item);
>  	struct srpt_port *sport = srpt_tpg_to_sport(se_tpg);
> +	const struct ib_device_attr *attrs = &sport->sdev->device->attrs;
> +	u32 rq_size = min(attrs->max_qp_wr, MAX_SRPT_RQ_SIZE);

[Severity: High]
Is it safe to unconditionally dereference sport->sdev->device here?

If an InfiniBand device is hot-removed, srpt_remove_one() drops the driver's
reference to sdev, but configfs items created via srpt_make_tport() can hold
additional references, keeping the sport structure alive.

Once srpt_remove_one() returns, the IB core drops its reference to the
ib_device, freeing it. If a user writes to the srp_sq_size configfs
attribute during or after this process, couldn't this access the freed
ib_device and trigger a use-after-free?

[ ... ]

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260925213629.8039-1-serhatkumral1@gmail.com?part=1