Re: [Qemu-devel] [PATCH] scsi-generic: Check sense key before request snooping and patching

[Dmitry Fomichev]

Paolo,

I've tested this version of the patch and it works fine.
Indeed, with this cleanup, the code is more straightforward and robust.

Dmitry

-----Original Message-----
From: Qemu-devel <qemu-devel-bounces+dmitry.fomichev=address@hidden> On Behalf 
Of Paolo Bonzini
Sent: Friday, July 19, 2019 4:17 AM
To: address@hidden
Cc: Shinichiro Kawasaki <address@hidden>
Subject: [Qemu-devel] [PATCH] scsi-generic: Check sense key before request 
snooping and patching

From: Shin'ichiro Kawasaki <address@hidden>

When READ CAPACITY command completes, scsi_read_complete() function
snoops the command result and updates SCSIDevice members blocksize and
max_lba . However, this update is executed even when READ CAPACITY
command indicates an error in sense data. This causes unexpected
blocksize update with zero value for SCSI devices without
READ CAPACITY(10) command support and eventually results in a divide
by zero. An emulated device by TCMU-runner is an example of a device
that doesn't support READ CAPACITY(10) command.

To avoid the unexpected update, add sense key check in
scsi_read_complete() function. The function already checks the sense key
for VPD Block Limits emulation. Do the scsi_parse_sense_buf() call for
all requests rather than just for VPD Block Limits emulation, so that
blocksize and max_lba are only updated if READ CAPACITY returns zero
sense key.

Signed-off-by: Shin'ichiro Kawasaki <address@hidden>
[Extend the check to all requests, not just READ CAPACITY]
Signed-off-by: Paolo Bonzini <address@hidden>
---
 hw/scsi/scsi-generic.c | 28 ++++++++++++++++------------
 1 file changed, 16 insertions(+), 12 deletions(-)

diff --git a/hw/scsi/scsi-generic.c b/hw/scsi/scsi-generic.c
index f07891b3f6..c11a0c9a84 100644
--- a/hw/scsi/scsi-generic.c
+++ b/hw/scsi/scsi-generic.c
@@ -254,24 +254,28 @@ static void scsi_read_complete(void * opaque, int ret)
 
     r->len = -1;
 
-    /*
-     * Check if this is a VPD Block Limits request that
-     * resulted in sense error but would need emulation.
-     * In this case, emulate a valid VPD response.
-     */
-    if (s->needs_vpd_bl_emulation && ret == 0 &&
-        (r->io_header.driver_status & SG_ERR_DRIVER_SENSE) &&
-        r->req.cmd.buf[0] == INQUIRY &&
-        (r->req.cmd.buf[1] & 0x01) &&
-        r->req.cmd.buf[2] == 0xb0) {
+    if (r->io_header.driver_status & SG_ERR_DRIVER_SENSE) {
         SCSISense sense =
             scsi_parse_sense_buf(r->req.sense, r->io_header.sb_len_wr);
-        if (sense.key == ILLEGAL_REQUEST) {
+
+        /*
+         * Check if this is a VPD Block Limits request that
+         * resulted in sense error but would need emulation.
+         * In this case, emulate a valid VPD response.
+         */
+        if (sense.key == ILLEGAL_REQUEST &&
+            s->needs_vpd_bl_emulation &&
+            r->req.cmd.buf[0] == INQUIRY &&
+            (r->req.cmd.buf[1] & 0x01) &&
+            r->req.cmd.buf[2] == 0xb0) {
             len = scsi_generic_emulate_block_limits(r, s);
             /*
-             * No need to let scsi_read_complete go on and handle an
+             * It's okay to jup to req_complete: no need to
+             * let scsi_handle_inquiry_reply handle an
              * INQUIRY VPD BL request we created manually.
              */
+        }
+        if (sense.key) {
             goto req_complete;
         }
     }
-- 
2.21.0