First off - thanks a bunch for creating an open, software based iSCSI target
that supports SCSI-3 PGR.
Second, I am running snv_93 for the target and open-iscsi on RHEL 4.4 for the
initiator.
I am having some issues with iscsitgt not correctly honoring a write exclusive,
registrants only reservation. Writes are being allowed from an unregistered
host - I believe they should be conflicting. It seems to work with an
exclusive access reservation.
I was able to get /tmp/target-log going so I could see the calls taking place.
spc_pgr_check() is reporting the correct reservation (write exclusive, ro) but
it returns an allow.
The interesting thing is that the writes were coming in as WRITE_G1. If I look
at the code for spc_pgr_is_conflicting(), I don't see this command being
checked. Based on my reading, it would fall to the default case and be allowed:
95 /*
96 * []----
97 * | spc_pgr_is_conflicting
98 * | PGR reservation conflict checking.
99 * | SPC-3, Revision 23, Table 31
100 * []----
101 */
102 static int
103 spc_pgr_is_conflicting(uint8_t *cdb, uint_t type)
104 {
105 Boolean_t conflict = False;
106
107 switch (cdb[0]) {
108 case SCMD_FORMAT:
109 case SCMD_EXTENDED_COPY:
110 case SCMD_LOG_SELECT_G1:
111 case SCMD_MODE_SELECT:
112 case SCMD_MODE_SELECT_G1:
113 case SCMD_MODE_SENSE:
114 case SCMD_MODE_SENSE_G1:
115 case SCMD_READ_ATTRIBUTE:
116 case SCMD_READ_BUFFER:
117 case SCMD_GDIAG: /*
SCMD_RECEIVE_DIAGNOSTIC_RESULTS */
118 case SCMD_SDIAG: /* SCMD_SEND_DIAGNOSTIC_RESULTS
*/
119 case SCMD_WRITE_ATTRIBUTE:
120 case SCMD_WRITE_BUFFER:
121 conflict = True;
122 break;
123
124 case SCMD_DOORLOCK: /*
SCMD_PREVENT_ALLOW_MEDIA_REMOVAL */
125 /*
126 * As per SPC-3, Revision 23, Table 31
127 * (prevent <> 0)
128 */
129 conflict = (cdb[4] & 0x1) ? True: False;
130 break;
131
132 case SCMD_MAINTENANCE_IN: /* SCMD_REPORT_ */
133 /*
134 * As per SPC-3, Revision 23, Section 6.23
135 */
136 switch ((cdb[1] & 0x1f)) {
137 /*
SCMD_REPORT_SUPPORTED_OPERATION_CODES */
138 case 0x0c:
139 /*
SCMD_REPORT_SUPPORTED_MANAGEMENT_FUNCTIONS */
140 case 0x0d:
141
142 conflict = True;
143 break;
144 }
145 break;
146
147 case SCMD_MAINTENANCE_OUT:
148 /*
149 * SPC-3, Revision 23, Section 6.29
150 */
151 switch ((cdb[1] & 0x1F)) {
152 case SCMD_SET_DEVICE_IDENTIFIER:
153 case SCMD_SET_PRIORITY:
154 case SCMD_SET_TARGET_PORT_GROUPS:
155 case SCMD_SET_TIMESTAMP:
156 conflict = True;
157 break;
158 }
159 break;
160
161 case SCMD_READ:
162 case SCMD_READ_G1:
163 case SCMD_READ_G4:
164 /*
165 * Exclusive Access, and EA Registrants Only
166 */
167 if (type == PGR_TYPE_EX_AC || type ==
PGR_TYPE_EX_AC_RO)
168 conflict = True;
169 break;
170 }
171
172 return (conflict);
173 }
I don't know nearly enough about the SCSI protocol to say that SCMD_WRITE,
SCMD_WRITE_G1, and SCMD_WRITE_G4 should be added to the case statement at line
108, but it sure seems that way.
Anybody care to comment?
Thanks,
Dave
This message posted from opensolaris.org
_______________________________________________
storage-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/storage-discuss
