Good catch but I don't think that's quite correct. To be accurate, the function list_first_entry(...) returns the first entry in the list or a pointer to the list head if empty.
The return is LUN0 only because LUN0 is created automatically when the target is created.
The function list_first_entry(...) is called at 10 places in all the source and only ONE place to find a device.
So let me augment my patch and insert in target_cmd_queue(...) a test for this. I believe this will cause requests to a server with a LUN-less target to simply fail until a LUN gets defined - just like an absent target.
BTW - this begs the question about when the sockets become available.
===============================================================================
diff -Naurp -X diff.exclude old/usr/mgmt.c new/usr/mgmt.c
--- old/usr/mgmt.c 2012-10-30 21:49:30.000000000 -0400
+++ new/usr/mgmt.c 2012-11-28 19:18:51.000000000 -0500
@@ -247,7 +247,7 @@ static tgtadm_err device_mgmt(int lld_no
params, 1);
break;
case OP_DELETE:
- adm_err = tgt_device_destroy(req->tid, req->lun, 0);
+ adm_err = tgt_device_destroy(req->tid, req->lun, req->op);
break;
case OP_UPDATE:
adm_err = tgt_device_update(req->tid, req->lun, params);
@@ -777,7 +777,7 @@ int ipc_init(void)
err = bind(fd, (struct sockaddr *) &addr, sizeof(addr));
if (err) {
- eprintf("can't bind a socket, %m\n");
+ eprintf("can't bind to socket \"%s\", %m\n", mgmt_path);
goto out;
}
diff -Naurp -X diff.exclude old/usr/target.c new/usr/target.c
--- old/usr/target.c 2012-10-30 21:49:30.000000000 -0400
+++ new/usr/target.c 2012-12-10 14:30:09.000000000 -0500
@@ -702,7 +702,7 @@ tgtadm_err tgt_device_destroy(int tid, u
struct registration *reg, *reg_next;
int ret;
- dprintf("%u %" PRIu64 "\n", tid, lun);
+ dprintf("%u %" PRIu64 "%s\n", tid, lun, force ? " Forced" : "");
/* lun0 is special */
if (!lun && !force)
@@ -1093,7 +1093,15 @@ int target_cmd_queue(int tid, struct scs
return -ENOENT;
}
- cmd->c_target = target = itn->nexus_target;
+ target = itn->nexus_target;
+
+ if (list_empty(&target->device_list))
+ {
+ eprintf("No devices for nexus %d %" PRIx64 "\n", tid, itn_id);
+ return -ENOENT;
+ }
+
+ cmd->c_target = target;
cmd->it_nexus = itn;
dev_id = scsi_get_devid(target->lid, cmd->lun);
diff -Naurp -X diff.exclude old/usr/tgtadm.c new/usr/tgtadm.c
--- old/usr/tgtadm.c 2012-10-30 21:49:30.000000000 -0400
+++ new/usr/tgtadm.c 2012-11-28 19:06:06.000000000 -0500
@@ -818,7 +818,7 @@ int main(int argc, char **argv)
break;
case OP_DELETE:
case OP_STATS:
- rc = verify_mode_params(argc, argv, "LmotlC");
+ rc = verify_mode_params(argc, argv, "LmotlCF");
if (rc) {
eprintf("target mode: option '-%c' is not "
"allowed/supported\n", rc);
===================================================================
-----Original Message-----
From: Arne Redlich [mailto:arne.redlich@xxxxxxxxxxxxxx]
Sent: Monday, December 10, 2012 1:24 PM
To: Braun, David
Cc: Boaz Harrosh; ronnie sahlberg; stgt@xxxxxxxxxxxxxxx
Subject: Re: Why LUN0?
2012/12/10 Braun, David <David.Braun@xxxxxxx>:
> By "like today the dummy-LUN0 will automatically be replaced" you mean with your patch - right?
>
> My --force code was very simple - basically a 2 line change. The other changes are just prints. Here's what I did.
David,
It's not as simple as that, as the dummy device mapped to LUN 0 is also abused to handle SCSI requests sent to LUNs that are not backed by a device:
int target_cmd_queue(int tid, struct scsi_cmd *cmd) {
/* ... snip ... */
cmd->dev = device_lookup(target, dev_id);
/* use LUN0 */
if (!cmd->dev)
cmd->dev = list_first_entry(&target->device_list,
struct scsi_lu,
device_siblings);
/* ... snip ... */
/*
* Call struct scsi_lu->cmd_perform() that will either be setup for
* internal or passthrough CDB processing using 2 functions below.
*/
return cmd->dev->cmd_perform(tid, cmd); }
This is addressed by the patch Boaz is carrying in his tree.
HTH,
Arne
��.n��������+%������w��{.n����� ��{ay� ʇڙ���f���h������_�(�階�ݢj"��������G����?���&��
