Please disregard. It looks like the original patch b68710a8094f reduced
the amount of data copied off so much that port logins are broken even
though with my patch the firmware now processes GET_RNID_PARAMS. The
breakage has been introduced during 6.5 (and propagated into 5.x stable
branches) so there is some urgency to getting a proper fix.
I will send a short series to revert b68710a8094f and replace it with
what I think is the correct overrun fix. AFAICT the overrun is only a
single dword.
--Steve
On 2023-08-09 09:10, Steve Magnani wrote:
At least some firmware (for example, QLE2692 FW 8.08.231 (d0d5))
returns
MBS_COMMAND_PARAMETER_ERROR when
MBC_GET_RNID_PARAMS(RNID_TYPE_PORT_LOGIN)
is issued with a transfer length of 4 (DWords). Correct the overrun fix
to
issue a "normal" transfer length and instead limit copy-out of desired
PLOGI data to the space available for it.
Fixes: b68710a8094f ("scsi: qla2xxx: Fix buffer overrun")
Signed-off-by: "Steven J. Magnani" <magnani@xxxxxxxx>
---
--- a/drivers/scsi/qla2xxx/qla_init.c 2023-08-01 03:46:21.727114453
-0500
+++ b/drivers/scsi/qla2xxx/qla_init.c 2023-08-08 23:18:46.475286995
-0500
@@ -5549,7 +5549,8 @@ static void qla_get_login_template(scsi_
__be32 *q;
memset(ha->init_cb, 0, ha->init_cb_size);
- sz = min_t(int, sizeof(struct fc_els_csp), ha->init_cb_size);
+ /* At least some firmware requires sz > sizeof(fc_els_csp) */
+ sz = min_t(int, sizeof(struct fc_els_flogi), ha->init_cb_size);
rval = qla24xx_get_port_login_templ(vha, ha->init_cb_dma,
ha->init_cb, sz);
if (rval != QLA_SUCCESS) {
@@ -5560,6 +5561,7 @@ static void qla_get_login_template(scsi_
q = (__be32 *)&ha->plogi_els_payld.fl_csp;
bp = (uint32_t *)ha->init_cb;
+ sz = min_t(u32, sizeof(struct fc_els_csp), sz);
cpu_to_be32_array(q, bp, sz / 4);
ha->flags.plogi_template_valid = 1;
}
